From dbeaed6cf049a0be97631ab74afa1f4ab9a800bf Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Marek=20Ol=C5=A1=C3=A1k?= Date: Thu, 16 Apr 2015 19:41:33 +0200 Subject: [PATCH 09/29] winsys/amdgpu: add addrlib - texture addressing and alignment calculator This is an internal project that Catalyst uses and now open source will do too. Signed-off-by: Arindam Nath --- src/gallium/winsys/radeon/amdgpu/Makefile.am | 7 +- src/gallium/winsys/radeon/amdgpu/Makefile.sources | 23 + .../winsys/radeon/amdgpu/addrlib/addrinterface.cpp | 1008 +++++ .../winsys/radeon/amdgpu/addrlib/addrinterface.h | 2166 +++++++++ .../winsys/radeon/amdgpu/addrlib/addrtypes.h | 590 +++ .../winsys/radeon/amdgpu/addrlib/core/addrcommon.h | 558 +++ .../radeon/amdgpu/addrlib/core/addrelemlib.cpp | 1678 +++++++ .../radeon/amdgpu/addrlib/core/addrelemlib.h | 270 ++ .../winsys/radeon/amdgpu/addrlib/core/addrlib.cpp | 4028 +++++++++++++++++ .../winsys/radeon/amdgpu/addrlib/core/addrlib.h | 695 +++ .../radeon/amdgpu/addrlib/core/addrobject.cpp | 246 ++ .../winsys/radeon/amdgpu/addrlib/core/addrobject.h | 89 + .../amdgpu/addrlib/inc/chip/r800/si_gb_reg.h | 155 + .../radeon/amdgpu/addrlib/inc/lnx_common_defs.h | 129 + .../addrlib/r800/chip/si_ci_vi_merged_enum.h | 40 + .../radeon/amdgpu/addrlib/r800/ciaddrlib.cpp | 1777 ++++++++ .../winsys/radeon/amdgpu/addrlib/r800/ciaddrlib.h | 197 + .../radeon/amdgpu/addrlib/r800/egbaddrlib.cpp | 4578 ++++++++++++++++++++ .../winsys/radeon/amdgpu/addrlib/r800/egbaddrlib.h | 411 ++ .../radeon/amdgpu/addrlib/r800/siaddrlib.cpp | 2818 ++++++++++++ .../winsys/radeon/amdgpu/addrlib/r800/siaddrlib.h | 262 ++ src/gallium/winsys/radeon/amdgpu/amdgpu_id.h | 157 + src/gallium/winsys/radeon/amdgpu/amdgpu_surface.c | 436 ++ src/gallium/winsys/radeon/amdgpu/amdgpu_winsys.c | 50 + src/gallium/winsys/radeon/amdgpu/amdgpu_winsys.h | 6 + 25 files changed, 22373 insertions(+), 1 deletion(-) create mode 100644 src/gallium/winsys/radeon/amdgpu/addrlib/addrinterface.cpp create mode 100644 src/gallium/winsys/radeon/amdgpu/addrlib/addrinterface.h create mode 100644 src/gallium/winsys/radeon/amdgpu/addrlib/addrtypes.h create mode 100644 src/gallium/winsys/radeon/amdgpu/addrlib/core/addrcommon.h create mode 100644 src/gallium/winsys/radeon/amdgpu/addrlib/core/addrelemlib.cpp create mode 100644 src/gallium/winsys/radeon/amdgpu/addrlib/core/addrelemlib.h create mode 100644 src/gallium/winsys/radeon/amdgpu/addrlib/core/addrlib.cpp create mode 100644 src/gallium/winsys/radeon/amdgpu/addrlib/core/addrlib.h create mode 100644 src/gallium/winsys/radeon/amdgpu/addrlib/core/addrobject.cpp create mode 100644 src/gallium/winsys/radeon/amdgpu/addrlib/core/addrobject.h create mode 100644 src/gallium/winsys/radeon/amdgpu/addrlib/inc/chip/r800/si_gb_reg.h create mode 100644 src/gallium/winsys/radeon/amdgpu/addrlib/inc/lnx_common_defs.h create mode 100644 src/gallium/winsys/radeon/amdgpu/addrlib/r800/chip/si_ci_vi_merged_enum.h create mode 100644 src/gallium/winsys/radeon/amdgpu/addrlib/r800/ciaddrlib.cpp create mode 100644 src/gallium/winsys/radeon/amdgpu/addrlib/r800/ciaddrlib.h create mode 100644 src/gallium/winsys/radeon/amdgpu/addrlib/r800/egbaddrlib.cpp create mode 100644 src/gallium/winsys/radeon/amdgpu/addrlib/r800/egbaddrlib.h create mode 100644 src/gallium/winsys/radeon/amdgpu/addrlib/r800/siaddrlib.cpp create mode 100644 src/gallium/winsys/radeon/amdgpu/addrlib/r800/siaddrlib.h create mode 100644 src/gallium/winsys/radeon/amdgpu/amdgpu_id.h create mode 100644 src/gallium/winsys/radeon/amdgpu/amdgpu_surface.c diff --git a/src/gallium/winsys/radeon/amdgpu/Makefile.am b/src/gallium/winsys/radeon/amdgpu/Makefile.am index 80ecb75..a719913 100644 --- a/src/gallium/winsys/radeon/amdgpu/Makefile.am +++ b/src/gallium/winsys/radeon/amdgpu/Makefile.am @@ -3,7 +3,12 @@ include $(top_srcdir)/src/gallium/Automake.inc AM_CFLAGS = \ $(GALLIUM_WINSYS_CFLAGS) \ - $(AMDGPU_CFLAGS) + $(AMDGPU_CFLAGS) \ + -I$(srcdir)/addrlib \ + -I$(srcdir)/addrlib/core \ + -I$(srcdir)/addrlib/inc/chip/r800 \ + -I$(srcdir)/addrlib/r800/chip \ + -DBRAHMA_BUILD=1 AM_CXXFLAGS = $(AM_CFLAGS) diff --git a/src/gallium/winsys/radeon/amdgpu/Makefile.sources b/src/gallium/winsys/radeon/amdgpu/Makefile.sources index 0f55010..6b33841 100644 --- a/src/gallium/winsys/radeon/amdgpu/Makefile.sources +++ b/src/gallium/winsys/radeon/amdgpu/Makefile.sources @@ -1,8 +1,31 @@ C_SOURCES := \ + addrlib/addrinterface.cpp \ + addrlib/addrinterface.h \ + addrlib/addrtypes.h \ + addrlib/core/addrcommon.h \ + addrlib/core/addrelemlib.cpp \ + addrlib/core/addrelemlib.h \ + addrlib/core/addrlib.cpp \ + addrlib/core/addrlib.h \ + addrlib/core/addrobject.cpp \ + addrlib/core/addrobject.h \ + addrlib/inc/chip/r800/si_gb_reg.h \ + addrlib/inc/lnx_common_defs.h \ + addrlib/r800/chip/si_ci_merged_enum.h \ + addrlib/r800/chip/si_ci_vi_merged_enum.h \ + addrlib/r800/chip/si_enum.h \ + addrlib/r800/ciaddrlib.cpp \ + addrlib/r800/ciaddrlib.h \ + addrlib/r800/egbaddrlib.cpp \ + addrlib/r800/egbaddrlib.h \ + addrlib/r800/siaddrlib.cpp \ + addrlib/r800/siaddrlib.h \ amdgpu_bo.c \ amdgpu_bo.h \ amdgpu_cs.c \ amdgpu_cs.h \ + amdgpu_id.h \ amdgpu_public.h \ + amdgpu_surface.c \ amdgpu_winsys.c \ amdgpu_winsys.h diff --git a/src/gallium/winsys/radeon/amdgpu/addrlib/addrinterface.cpp b/src/gallium/winsys/radeon/amdgpu/addrlib/addrinterface.cpp new file mode 100644 index 0000000..6556927 --- /dev/null +++ b/src/gallium/winsys/radeon/amdgpu/addrlib/addrinterface.cpp @@ -0,0 +1,1008 @@ +/* + * Copyright © 2014 Advanced Micro Devices, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES + * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NON-INFRINGEMENT. IN NO EVENT SHALL THE COPYRIGHT HOLDERS, AUTHORS + * AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + */ + +/** +*************************************************************************************************** +* @file addrinterface.cpp +* @brief Contains the addrlib interface functions +*************************************************************************************************** +*/ +#include "addrinterface.h" +#include "addrlib.h" + +#include "addrcommon.h" + +/////////////////////////////////////////////////////////////////////////////////////////////////// +// Create/Destroy/Config functions +/////////////////////////////////////////////////////////////////////////////////////////////////// + +/** +*************************************************************************************************** +* AddrCreate +* +* @brief +* Create address lib object +* +* @return +* ADDR_OK if successful, otherwise an error code of ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrCreate( + const ADDR_CREATE_INPUT* pAddrCreateIn, ///< [in] infomation for creating address lib object + ADDR_CREATE_OUTPUT* pAddrCreateOut) ///< [out] address lib handle +{ + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + returnCode = AddrLib::Create(pAddrCreateIn, pAddrCreateOut); + + return returnCode; +} + + + +/** +*************************************************************************************************** +* AddrDestroy +* +* @brief +* Destroy address lib object +* +* @return +* ADDR_OK if successful, otherwise an error code of ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrDestroy( + ADDR_HANDLE hLib) ///< [in] address lib handle +{ + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + if (hLib) + { + AddrLib* pLib = AddrLib::GetAddrLib(hLib); + pLib->Destroy(); + } + else + { + returnCode = ADDR_ERROR; + } + + return returnCode; +} + + + +/////////////////////////////////////////////////////////////////////////////////////////////////// +// Surface functions +/////////////////////////////////////////////////////////////////////////////////////////////////// + +/** +*************************************************************************************************** +* AddrComputeSurfaceInfo +* +* @brief +* Calculate surface width/height/depth/alignments and suitable tiling mode +* +* @return +* ADDR_OK if successful, otherwise an error code of ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrComputeSurfaceInfo( + ADDR_HANDLE hLib, ///< [in] address lib handle + const ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn, ///< [in] surface information + ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pOut) ///< [out] surface parameters and alignments +{ + AddrLib* pLib = AddrLib::GetAddrLib(hLib); + + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + if (pLib != NULL) + { + returnCode = pLib->ComputeSurfaceInfo(pIn, pOut); + } + else + { + returnCode = ADDR_ERROR; + } + + return returnCode; +} + + + +/** +*************************************************************************************************** +* AddrComputeSurfaceAddrFromCoord +* +* @brief +* Compute surface address according to coordinates +* +* @return +* ADDR_OK if successful, otherwise an error code of ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrComputeSurfaceAddrFromCoord( + ADDR_HANDLE hLib, ///< [in] address lib handle + const ADDR_COMPUTE_SURFACE_ADDRFROMCOORD_INPUT* pIn, ///< [in] surface info and coordinates + ADDR_COMPUTE_SURFACE_ADDRFROMCOORD_OUTPUT* pOut) ///< [out] surface address +{ + AddrLib* pLib = AddrLib::GetAddrLib(hLib); + + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + if (pLib != NULL) + { + returnCode = pLib->ComputeSurfaceAddrFromCoord(pIn, pOut); + } + else + { + returnCode = ADDR_ERROR; + } + + return returnCode; +} + +/** +*************************************************************************************************** +* AddrComputeSurfaceCoordFromAddr +* +* @brief +* Compute coordinates according to surface address +* +* @return +* ADDR_OK if successful, otherwise an error code of ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrComputeSurfaceCoordFromAddr( + ADDR_HANDLE hLib, ///< [in] address lib handle + const ADDR_COMPUTE_SURFACE_COORDFROMADDR_INPUT* pIn, ///< [in] surface info and address + ADDR_COMPUTE_SURFACE_COORDFROMADDR_OUTPUT* pOut) ///< [out] coordinates +{ + AddrLib* pLib = AddrLib::GetAddrLib(hLib); + + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + if (pLib != NULL) + { + returnCode = pLib->ComputeSurfaceCoordFromAddr(pIn, pOut); + } + else + { + returnCode = ADDR_ERROR; + } + + return returnCode; +} + + + +/////////////////////////////////////////////////////////////////////////////////////////////////// +// HTile functions +/////////////////////////////////////////////////////////////////////////////////////////////////// + +/** +*************************************************************************************************** +* AddrComputeHtileInfo +* +* @brief +* Compute Htile pitch, height, base alignment and size in bytes +* +* @return +* ADDR_OK if successful, otherwise an error code of ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrComputeHtileInfo( + ADDR_HANDLE hLib, ///< [in] address lib handle + const ADDR_COMPUTE_HTILE_INFO_INPUT* pIn, ///< [in] Htile information + ADDR_COMPUTE_HTILE_INFO_OUTPUT* pOut) ///< [out] Htile pitch, height and size in bytes +{ + AddrLib* pLib = AddrLib::GetAddrLib(hLib); + + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + if (pLib != NULL) + { + returnCode = pLib->ComputeHtileInfo(pIn, pOut); + } + else + { + returnCode = ADDR_ERROR; + } + + return returnCode; +} + +/** +*************************************************************************************************** +* AddrComputeHtileAddrFromCoord +* +* @brief +* Compute Htile address according to coordinates (of depth buffer) +* +* @return +* ADDR_OK if successful, otherwise an error code of ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrComputeHtileAddrFromCoord( + ADDR_HANDLE hLib, ///< [in] address lib handle + const ADDR_COMPUTE_HTILE_ADDRFROMCOORD_INPUT* pIn, ///< [in] Htile info and coordinates + ADDR_COMPUTE_HTILE_ADDRFROMCOORD_OUTPUT* pOut) ///< [out] Htile address +{ + AddrLib* pLib = AddrLib::GetAddrLib(hLib); + + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + if (pLib != NULL) + { + returnCode = pLib->ComputeHtileAddrFromCoord(pIn, pOut); + } + else + { + returnCode = ADDR_ERROR; + } + + return returnCode; +} + +/** +*************************************************************************************************** +* AddrComputeHtileCoordFromAddr +* +* @brief +* Compute coordinates within depth buffer (1st pixel of a micro tile) according to +* Htile address +* +* @return +* ADDR_OK if successful, otherwise an error code of ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrComputeHtileCoordFromAddr( + ADDR_HANDLE hLib, ///< [in] address lib handle + const ADDR_COMPUTE_HTILE_COORDFROMADDR_INPUT* pIn, ///< [in] Htile info and address + ADDR_COMPUTE_HTILE_COORDFROMADDR_OUTPUT* pOut) ///< [out] Htile coordinates +{ + AddrLib* pLib = AddrLib::GetAddrLib(hLib); + + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + if (pLib != NULL) + { + returnCode = pLib->ComputeHtileCoordFromAddr(pIn, pOut); + } + else + { + returnCode = ADDR_ERROR; + } + + return returnCode; +} + + + +/////////////////////////////////////////////////////////////////////////////////////////////////// +// C-mask functions +/////////////////////////////////////////////////////////////////////////////////////////////////// + +/** +*************************************************************************************************** +* AddrComputeCmaskInfo +* +* @brief +* Compute Cmask pitch, height, base alignment and size in bytes from color buffer +* info +* +* @return +* ADDR_OK if successful, otherwise an error code of ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrComputeCmaskInfo( + ADDR_HANDLE hLib, ///< [in] address lib handle + const ADDR_COMPUTE_CMASK_INFO_INPUT* pIn, ///< [in] Cmask pitch and height + ADDR_COMPUTE_CMASK_INFO_OUTPUT* pOut) ///< [out] Cmask pitch, height and size in bytes +{ + AddrLib* pLib = AddrLib::GetAddrLib(hLib); + + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + if (pLib != NULL) + { + returnCode = pLib->ComputeCmaskInfo(pIn, pOut); + } + else + { + returnCode = ADDR_ERROR; + } + + return returnCode; +} + +/** +*************************************************************************************************** +* AddrComputeCmaskAddrFromCoord +* +* @brief +* Compute Cmask address according to coordinates (of MSAA color buffer) +* +* @return +* ADDR_OK if successful, otherwise an error code of ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrComputeCmaskAddrFromCoord( + ADDR_HANDLE hLib, ///< [in] address lib handle + const ADDR_COMPUTE_CMASK_ADDRFROMCOORD_INPUT* pIn, ///< [in] Cmask info and coordinates + ADDR_COMPUTE_CMASK_ADDRFROMCOORD_OUTPUT* pOut) ///< [out] Cmask address +{ + AddrLib* pLib = AddrLib::GetAddrLib(hLib); + + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + if (pLib != NULL) + { + returnCode = pLib->ComputeCmaskAddrFromCoord(pIn, pOut); + } + else + { + returnCode = ADDR_ERROR; + } + + return returnCode; +} + +/** +*************************************************************************************************** +* AddrComputeCmaskCoordFromAddr +* +* @brief +* Compute coordinates within color buffer (1st pixel of a micro tile) according to +* Cmask address +* +* @return +* ADDR_OK if successful, otherwise an error code of ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrComputeCmaskCoordFromAddr( + ADDR_HANDLE hLib, ///< [in] address lib handle + const ADDR_COMPUTE_CMASK_COORDFROMADDR_INPUT* pIn, ///< [in] Cmask info and address + ADDR_COMPUTE_CMASK_COORDFROMADDR_OUTPUT* pOut) ///< [out] Cmask coordinates +{ + AddrLib* pLib = AddrLib::GetAddrLib(hLib); + + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + if (pLib != NULL) + { + returnCode = pLib->ComputeCmaskCoordFromAddr(pIn, pOut); + } + else + { + returnCode = ADDR_ERROR; + } + + return returnCode; +} + + + +/////////////////////////////////////////////////////////////////////////////////////////////////// +// F-mask functions +/////////////////////////////////////////////////////////////////////////////////////////////////// + +/** +*************************************************************************************************** +* AddrComputeFmaskInfo +* +* @brief +* Compute Fmask pitch/height/depth/alignments and size in bytes +* +* @return +* ADDR_OK if successful, otherwise an error code of ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrComputeFmaskInfo( + ADDR_HANDLE hLib, ///< [in] address lib handle + const ADDR_COMPUTE_FMASK_INFO_INPUT* pIn, ///< [in] Fmask information + ADDR_COMPUTE_FMASK_INFO_OUTPUT* pOut) ///< [out] Fmask pitch and height +{ + AddrLib* pLib = AddrLib::GetAddrLib(hLib); + + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + if (pLib != NULL) + { + returnCode = pLib->ComputeFmaskInfo(pIn, pOut); + } + else + { + returnCode = ADDR_ERROR; + } + + return returnCode; +} + +/** +*************************************************************************************************** +* AddrComputeFmaskAddrFromCoord +* +* @brief +* Compute Fmask address according to coordinates (x,y,slice,sample,plane) +* +* @return +* ADDR_OK if successful, otherwise an error code of ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrComputeFmaskAddrFromCoord( + ADDR_HANDLE hLib, ///< [in] address lib handle + const ADDR_COMPUTE_FMASK_ADDRFROMCOORD_INPUT* pIn, ///< [in] Fmask info and coordinates + ADDR_COMPUTE_FMASK_ADDRFROMCOORD_OUTPUT* pOut) ///< [out] Fmask address +{ + AddrLib* pLib = AddrLib::GetAddrLib(hLib); + + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + if (pLib != NULL) + { + returnCode = pLib->ComputeFmaskAddrFromCoord(pIn, pOut); + } + else + { + returnCode = ADDR_ERROR; + } + + return returnCode; +} + +/** +*************************************************************************************************** +* AddrComputeFmaskCoordFromAddr +* +* @brief +* Compute coordinates (x,y,slice,sample,plane) according to Fmask address +* +* @return +* ADDR_OK if successful, otherwise an error code of ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrComputeFmaskCoordFromAddr( + ADDR_HANDLE hLib, ///< [in] address lib handle + const ADDR_COMPUTE_FMASK_COORDFROMADDR_INPUT* pIn, ///< [in] Fmask info and address + ADDR_COMPUTE_FMASK_COORDFROMADDR_OUTPUT* pOut) ///< [out] Fmask coordinates +{ + AddrLib* pLib = AddrLib::GetAddrLib(hLib); + + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + if (pLib != NULL) + { + returnCode = pLib->ComputeFmaskCoordFromAddr(pIn, pOut); + } + else + { + returnCode = ADDR_ERROR; + } + + return returnCode; +} + + + +/////////////////////////////////////////////////////////////////////////////////////////////////// +// DCC key functions +/////////////////////////////////////////////////////////////////////////////////////////////////// + +/** +*************************************************************************************************** +* AddrComputeDccInfo +* +* @brief +* Compute DCC key size, base alignment based on color surface size, tile info or tile index +* +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrComputeDccInfo( + ADDR_HANDLE hLib, ///< [in] handle of addrlib + const ADDR_COMPUTE_DCCINFO_INPUT* pIn, ///< [in] input + ADDR_COMPUTE_DCCINFO_OUTPUT* pOut) ///< [out] output +{ + ADDR_E_RETURNCODE returnCode; + + AddrLib* pLib = AddrLib::GetAddrLib(hLib); + + if (pLib != NULL) + { + returnCode = pLib->ComputeDccInfo(pIn, pOut); + } + else + { + returnCode = ADDR_ERROR; + } + + return returnCode; +} + + + +/////////////////////////////////////////////////////////////////////////////// +// Below functions are element related or helper functions +/////////////////////////////////////////////////////////////////////////////// + +/** +*************************************************************************************************** +* AddrGetVersion +* +* @brief +* Get AddrLib version number. Client may check this return value against ADDRLIB_VERSION +* defined in addrinterface.h to see if there is a mismatch. +*************************************************************************************************** +*/ +UINT_32 ADDR_API AddrGetVersion(ADDR_HANDLE hLib) +{ + UINT_32 version = 0; + + AddrLib* pLib = AddrLib::GetAddrLib(hLib); + + ADDR_ASSERT(pLib != NULL); + + if (pLib) + { + version = pLib->GetVersion(); + } + + return version; +} + +/** +*************************************************************************************************** +* AddrUseTileIndex +* +* @brief +* Return TRUE if tileIndex is enabled in this address library +*************************************************************************************************** +*/ +BOOL_32 ADDR_API AddrUseTileIndex(ADDR_HANDLE hLib) +{ + BOOL_32 useTileIndex = FALSE; + + AddrLib* pLib = AddrLib::GetAddrLib(hLib); + + ADDR_ASSERT(pLib != NULL); + + if (pLib) + { + useTileIndex = pLib->UseTileIndex(0); + } + + return useTileIndex; +} + +/** +*************************************************************************************************** +* AddrUseCombinedSwizzle +* +* @brief +* Return TRUE if combined swizzle is enabled in this address library +*************************************************************************************************** +*/ +BOOL_32 ADDR_API AddrUseCombinedSwizzle(ADDR_HANDLE hLib) +{ + BOOL_32 useCombinedSwizzle = FALSE; + + AddrLib* pLib = AddrLib::GetAddrLib(hLib); + + ADDR_ASSERT(pLib != NULL); + + if (pLib) + { + useCombinedSwizzle = pLib->UseCombinedSwizzle(); + } + + return useCombinedSwizzle; +} + +/** +*************************************************************************************************** +* AddrExtractBankPipeSwizzle +* +* @brief +* Extract Bank and Pipe swizzle from base256b +* @return +* ADDR_OK if successful, otherwise an error code of ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrExtractBankPipeSwizzle( + ADDR_HANDLE hLib, ///< [in] addrlib handle + const ADDR_EXTRACT_BANKPIPE_SWIZZLE_INPUT* pIn, ///< [in] input structure + ADDR_EXTRACT_BANKPIPE_SWIZZLE_OUTPUT* pOut) ///< [out] output structure +{ + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + AddrLib* pLib = AddrLib::GetAddrLib(hLib); + + if (pLib != NULL) + { + returnCode = pLib->ExtractBankPipeSwizzle(pIn, pOut); + } + else + { + returnCode = ADDR_ERROR; + } + + return returnCode; +} + +/** +*************************************************************************************************** +* AddrCombineBankPipeSwizzle +* +* @brief +* Combine Bank and Pipe swizzle +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrCombineBankPipeSwizzle( + ADDR_HANDLE hLib, + const ADDR_COMBINE_BANKPIPE_SWIZZLE_INPUT* pIn, + ADDR_COMBINE_BANKPIPE_SWIZZLE_OUTPUT* pOut) +{ + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + AddrLib* pLib = AddrLib::GetAddrLib(hLib); + + if (pLib != NULL) + { + returnCode = pLib->CombineBankPipeSwizzle(pIn, pOut); + } + else + { + returnCode = ADDR_ERROR; + } + + return returnCode; +} + +/** +*************************************************************************************************** +* AddrComputeSliceSwizzle +* +* @brief +* Compute a swizzle for slice from a base swizzle +* @return +* ADDR_OK if no error +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrComputeSliceSwizzle( + ADDR_HANDLE hLib, + const ADDR_COMPUTE_SLICESWIZZLE_INPUT* pIn, + ADDR_COMPUTE_SLICESWIZZLE_OUTPUT* pOut) +{ + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + AddrLib* pLib = AddrLib::GetAddrLib(hLib); + + if (pLib != NULL) + { + returnCode = pLib->ComputeSliceTileSwizzle(pIn, pOut); + } + else + { + returnCode = ADDR_ERROR; + } + + return returnCode; +} + +/** +*************************************************************************************************** +* AddrComputeBaseSwizzle +* +* @brief +* Return a Combined Bank and Pipe swizzle base on surface based on surface type/index +* @return +* ADDR_OK if no error +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrComputeBaseSwizzle( + ADDR_HANDLE hLib, + const ADDR_COMPUTE_BASE_SWIZZLE_INPUT* pIn, + ADDR_COMPUTE_BASE_SWIZZLE_OUTPUT* pOut) +{ + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + AddrLib* pLib = AddrLib::GetAddrLib(hLib); + + if (pLib != NULL) + { + returnCode = pLib->ComputeBaseSwizzle(pIn, pOut); + } + else + { + returnCode = ADDR_ERROR; + } + + return returnCode; +} + +/** +*************************************************************************************************** +* ElemFlt32ToDepthPixel +* +* @brief +* Convert a FLT_32 value to a depth/stencil pixel value +* +* @return +* ADDR_OK if successful, otherwise an error code of ADDR_E_RETURNCODE +* +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API ElemFlt32ToDepthPixel( + ADDR_HANDLE hLib, ///< [in] addrlib handle + const ELEM_FLT32TODEPTHPIXEL_INPUT* pIn, ///< [in] per-component value + ELEM_FLT32TODEPTHPIXEL_OUTPUT* pOut) ///< [out] final pixel value +{ + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + AddrLib* pLib = AddrLib::GetAddrLib(hLib); + + if (pLib != NULL) + { + pLib->Flt32ToDepthPixel(pIn, pOut); + } + else + { + returnCode = ADDR_ERROR; + } + + return returnCode; +} + +/** +*************************************************************************************************** +* ElemFlt32ToColorPixel +* +* @brief +* Convert a FLT_32 value to a red/green/blue/alpha pixel value +* +* @return +* ADDR_OK if successful, otherwise an error code of ADDR_E_RETURNCODE +* +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API ElemFlt32ToColorPixel( + ADDR_HANDLE hLib, ///< [in] addrlib handle + const ELEM_FLT32TOCOLORPIXEL_INPUT* pIn, ///< [in] format, surface number and swap value + ELEM_FLT32TOCOLORPIXEL_OUTPUT* pOut) ///< [out] final pixel value +{ + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + AddrLib* pLib = AddrLib::GetAddrLib(hLib); + + if (pLib != NULL) + { + pLib->Flt32ToColorPixel(pIn, pOut); + } + else + { + returnCode = ADDR_ERROR; + } + + return returnCode; +} + +/** +*************************************************************************************************** +* ElemGetExportNorm +* +* @brief +* Helper function to check one format can be EXPORT_NUM, +* which is a register CB_COLOR_INFO.SURFACE_FORMAT. +* FP16 can be reported as EXPORT_NORM for rv770 in r600 +* family +* +*************************************************************************************************** +*/ +BOOL_32 ADDR_API ElemGetExportNorm( + ADDR_HANDLE hLib, ///< [in] addrlib handle + const ELEM_GETEXPORTNORM_INPUT* pIn) ///< [in] input structure +{ + AddrLib* pLib = AddrLib::GetAddrLib(hLib); + BOOL_32 enabled = FALSE; + + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + if (pLib != NULL) + { + enabled = pLib->GetExportNorm(pIn); + } + else + { + returnCode = ADDR_ERROR; + } + + ADDR_ASSERT(returnCode == ADDR_OK); + + return enabled; +} + +/** +*************************************************************************************************** +* AddrConvertTileInfoToHW +* +* @brief +* Convert tile info from real value to hardware register value +* +* @return +* ADDR_OK if successful, otherwise an error code of ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrConvertTileInfoToHW( + ADDR_HANDLE hLib, ///< [in] address lib handle + const ADDR_CONVERT_TILEINFOTOHW_INPUT* pIn, ///< [in] tile info with real value + ADDR_CONVERT_TILEINFOTOHW_OUTPUT* pOut) ///< [out] tile info with HW register value +{ + AddrLib* pLib = AddrLib::GetAddrLib(hLib); + + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + if (pLib != NULL) + { + returnCode = pLib->ConvertTileInfoToHW(pIn, pOut); + } + else + { + returnCode = ADDR_ERROR; + } + + return returnCode; +} + +/** +*************************************************************************************************** +* AddrConvertTileIndex +* +* @brief +* Convert tile index to tile mode/type/info +* +* @return +* ADDR_OK if successful, otherwise an error code of ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrConvertTileIndex( + ADDR_HANDLE hLib, ///< [in] address lib handle + const ADDR_CONVERT_TILEINDEX_INPUT* pIn, ///< [in] input - tile index + ADDR_CONVERT_TILEINDEX_OUTPUT* pOut) ///< [out] tile mode/type/info +{ + AddrLib* pLib = AddrLib::GetAddrLib(hLib); + + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + if (pLib != NULL) + { + returnCode = pLib->ConvertTileIndex(pIn, pOut); + } + else + { + returnCode = ADDR_ERROR; + } + + return returnCode; +} + +/** +*************************************************************************************************** +* AddrConvertTileIndex1 +* +* @brief +* Convert tile index to tile mode/type/info +* +* @return +* ADDR_OK if successful, otherwise an error code of ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrConvertTileIndex1( + ADDR_HANDLE hLib, ///< [in] address lib handle + const ADDR_CONVERT_TILEINDEX1_INPUT* pIn, ///< [in] input - tile index + ADDR_CONVERT_TILEINDEX_OUTPUT* pOut) ///< [out] tile mode/type/info +{ + AddrLib* pLib = AddrLib::GetAddrLib(hLib); + + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + if (pLib != NULL) + { + returnCode = pLib->ConvertTileIndex1(pIn, pOut); + } + else + { + returnCode = ADDR_ERROR; + } + + return returnCode; +} + +/** +*************************************************************************************************** +* AddrGetTileIndex +* +* @brief +* Get tile index from tile mode/type/info +* +* @return +* ADDR_OK if successful, otherwise an error code of ADDR_E_RETURNCODE +* +* @note +* Only meaningful for SI (and above) +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrGetTileIndex( + ADDR_HANDLE hLib, + const ADDR_GET_TILEINDEX_INPUT* pIn, + ADDR_GET_TILEINDEX_OUTPUT* pOut) +{ + AddrLib* pLib = AddrLib::GetAddrLib(hLib); + + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + if (pLib != NULL) + { + returnCode = pLib->GetTileIndex(pIn, pOut); + } + else + { + returnCode = ADDR_ERROR; + } + + return returnCode; +} + +/** +*************************************************************************************************** +* AddrComputePrtInfo +* +* @brief +* Interface function for ComputePrtInfo +* +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrComputePrtInfo( + ADDR_HANDLE hLib, + const ADDR_PRT_INFO_INPUT* pIn, + ADDR_PRT_INFO_OUTPUT* pOut) +{ + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + AddrLib* pLib = AddrLib::GetAddrLib(hLib); + + if (pLib != NULL) + { + returnCode = pLib->ComputePrtInfo(pIn, pOut); + } + else + { + returnCode = ADDR_ERROR; + } + + return returnCode; +} + diff --git a/src/gallium/winsys/radeon/amdgpu/addrlib/addrinterface.h b/src/gallium/winsys/radeon/amdgpu/addrlib/addrinterface.h new file mode 100644 index 0000000..03fbf2b --- /dev/null +++ b/src/gallium/winsys/radeon/amdgpu/addrlib/addrinterface.h @@ -0,0 +1,2166 @@ +/* + * Copyright © 2014 Advanced Micro Devices, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES + * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NON-INFRINGEMENT. IN NO EVENT SHALL THE COPYRIGHT HOLDERS, AUTHORS + * AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + */ + +/** +*************************************************************************************************** +* @file addrinterface.h +* @brief Contains the addrlib interfaces declaration and parameter defines +*************************************************************************************************** +*/ +#ifndef __ADDR_INTERFACE_H__ +#define __ADDR_INTERFACE_H__ + +#if defined(__cplusplus) +extern "C" +{ +#endif + +#include "addrtypes.h" + +#define ADDRLIB_VERSION_MAJOR 5 +#define ADDRLIB_VERSION_MINOR 25 +#define ADDRLIB_VERSION ((ADDRLIB_VERSION_MAJOR << 16) | ADDRLIB_VERSION_MINOR) + +/// Virtually all interface functions need ADDR_HANDLE as first parameter +typedef VOID* ADDR_HANDLE; + +/// Client handle used in callbacks +typedef VOID* ADDR_CLIENT_HANDLE; + +/** +* ///////////////////////////////////////////////////////////////////////////////////////////////// +* // Callback functions +* ///////////////////////////////////////////////////////////////////////////////////////////////// +* typedef VOID* (ADDR_API* ADDR_ALLOCSYSMEM)( +* const ADDR_ALLOCSYSMEM_INPUT* pInput); +* typedef ADDR_E_RETURNCODE (ADDR_API* ADDR_FREESYSMEM)( +* VOID* pVirtAddr); +* typedef ADDR_E_RETURNCODE (ADDR_API* ADDR_DEBUGPRINT)( +* const ADDR_DEBUGPRINT_INPUT* pInput); +* +* ///////////////////////////////////////////////////////////////////////////////////////////////// +* // Create/Destroy/Config functions +* ///////////////////////////////////////////////////////////////////////////////////////////////// +* AddrCreate() +* AddrDestroy() +* +* ///////////////////////////////////////////////////////////////////////////////////////////////// +* // Surface functions +* ///////////////////////////////////////////////////////////////////////////////////////////////// +* AddrComputeSurfaceInfo() +* AddrComputeSurfaceAddrFromCoord() +* AddrComputeSurfaceCoordFromAddr() +* +* ///////////////////////////////////////////////////////////////////////////////////////////////// +* // HTile functions +* ///////////////////////////////////////////////////////////////////////////////////////////////// +* AddrComputeHtileInfo() +* AddrComputeHtileAddrFromCoord() +* AddrComputeHtileCoordFromAddr() +* +* ///////////////////////////////////////////////////////////////////////////////////////////////// +* // C-mask functions +* ///////////////////////////////////////////////////////////////////////////////////////////////// +* AddrComputeCmaskInfo() +* AddrComputeCmaskAddrFromCoord() +* AddrComputeCmaskCoordFromAddr() +* +* ///////////////////////////////////////////////////////////////////////////////////////////////// +* // F-mask functions +* ///////////////////////////////////////////////////////////////////////////////////////////////// +* AddrComputeFmaskInfo() +* AddrComputeFmaskAddrFromCoord() +* AddrComputeFmaskCoordFromAddr() +* +* ///////////////////////////////////////////////////////////////////////////////////////////////// +* // Element/Utility functions +* ///////////////////////////////////////////////////////////////////////////////////////////////// +* ElemFlt32ToDepthPixel() +* ElemFlt32ToColorPixel() +* AddrExtractBankPipeSwizzle() +* AddrCombineBankPipeSwizzle() +* AddrComputeSliceSwizzle() +* AddrConvertTileInfoToHW() +* AddrConvertTileIndex() +* AddrConvertTileIndex1() +* AddrGetTileIndex() +* AddrComputeBaseSwizzle() +* AddrUseTileIndex() +* AddrUseCombinedSwizzle() +* +* ///////////////////////////////////////////////////////////////////////////////////////////////// +* // Dump functions +* ///////////////////////////////////////////////////////////////////////////////////////////////// +* AddrDumpSurfaceInfo() +* AddrDumpFmaskInfo() +* AddrDumpCmaskInfo() +* AddrDumpHtileInfo() +* +**/ + +/////////////////////////////////////////////////////////////////////////////////////////////////// +// Callback functions +/////////////////////////////////////////////////////////////////////////////////////////////////// + +/** +*************************************************************************************************** +* @brief Alloc system memory flags. +* @note These flags are reserved for future use and if flags are added will minimize the impact +* of the client. +*************************************************************************************************** +*/ +typedef union _ADDR_ALLOCSYSMEM_FLAGS +{ + struct + { + UINT_32 reserved : 32; ///< Reserved for future use. + } fields; + UINT_32 value; + +} ADDR_ALLOCSYSMEM_FLAGS; + +/** +*************************************************************************************************** +* @brief Alloc system memory input structure +*************************************************************************************************** +*/ +typedef struct _ADDR_ALLOCSYSMEM_INPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + ADDR_ALLOCSYSMEM_FLAGS flags; ///< System memory flags. + UINT_32 sizeInBytes; ///< System memory allocation size in bytes. + ADDR_CLIENT_HANDLE hClient; ///< Client handle +} ADDR_ALLOCSYSMEM_INPUT; + +/** +*************************************************************************************************** +* ADDR_ALLOCSYSMEM +* @brief +* Allocate system memory callback function. Returns valid pointer on success. +*************************************************************************************************** +*/ +typedef VOID* (ADDR_API* ADDR_ALLOCSYSMEM)( + const ADDR_ALLOCSYSMEM_INPUT* pInput); + +/** +*************************************************************************************************** +* @brief Free system memory input structure +*************************************************************************************************** +*/ +typedef struct _ADDR_FREESYSMEM_INPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + VOID* pVirtAddr; ///< Virtual address + ADDR_CLIENT_HANDLE hClient; ///< Client handle +} ADDR_FREESYSMEM_INPUT; + +/** +*************************************************************************************************** +* ADDR_FREESYSMEM +* @brief +* Free system memory callback function. +* Returns ADDR_OK on success. +*************************************************************************************************** +*/ +typedef ADDR_E_RETURNCODE (ADDR_API* ADDR_FREESYSMEM)( + const ADDR_FREESYSMEM_INPUT* pInput); + +/** +*************************************************************************************************** +* @brief Print debug message input structure +*************************************************************************************************** +*/ +typedef struct _ADDR_DEBUGPRINT_INPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + CHAR* pDebugString; ///< Debug print string + va_list ap; ///< Variable argument list + ADDR_CLIENT_HANDLE hClient; ///< Client handle +} ADDR_DEBUGPRINT_INPUT; + +/** +*************************************************************************************************** +* ADDR_DEBUGPRINT +* @brief +* Print debug message callback function. +* Returns ADDR_OK on success. +*************************************************************************************************** +*/ +typedef ADDR_E_RETURNCODE (ADDR_API* ADDR_DEBUGPRINT)( + const ADDR_DEBUGPRINT_INPUT* pInput); + +/** +*************************************************************************************************** +* ADDR_CALLBACKS +* +* @brief +* Address Library needs client to provide system memory alloc/free routines. +*************************************************************************************************** +*/ +typedef struct _ADDR_CALLBACKS +{ + ADDR_ALLOCSYSMEM allocSysMem; ///< Routine to allocate system memory + ADDR_FREESYSMEM freeSysMem; ///< Routine to free system memory + ADDR_DEBUGPRINT debugPrint; ///< Routine to print debug message +} ADDR_CALLBACKS; + +/////////////////////////////////////////////////////////////////////////////////////////////////// +// Create/Destroy functions +/////////////////////////////////////////////////////////////////////////////////////////////////// + +/** +*************************************************************************************************** +* ADDR_CREATE_FLAGS +* +* @brief +* This structure is used to pass some setup in creation of AddrLib +* @note +*************************************************************************************************** +*/ +typedef union _ADDR_CREATE_FLAGS +{ + struct + { + UINT_32 noCubeMipSlicesPad : 1; ///< Turn cubemap faces padding off + UINT_32 fillSizeFields : 1; ///< If clients fill size fields in all input and + /// output structure + UINT_32 useTileIndex : 1; ///< Make tileIndex field in input valid + UINT_32 useCombinedSwizzle : 1; ///< Use combined tile swizzle + UINT_32 checkLast2DLevel : 1; ///< Check the last 2D mip sub level + UINT_32 useHtileSliceAlign : 1; ///< Do htile single slice alignment + UINT_32 degradeBaseLevel : 1; ///< Degrade to 1D modes automatically for base level + UINT_32 allowLargeThickTile : 1; ///< Allow 64*thickness*bytesPerPixel > rowSize + UINT_32 reserved : 24; ///< Reserved bits for future use + }; + + UINT_32 value; +} ADDR_CREATE_FLAGS; + +/** +*************************************************************************************************** +* ADDR_REGISTER_VALUE +* +* @brief +* Data from registers to setup AddrLib global data, used in AddrCreate +*************************************************************************************************** +*/ +typedef struct _ADDR_REGISTER_VALUE +{ + UINT_32 gbAddrConfig; ///< For R8xx, use GB_ADDR_CONFIG register value. + /// For R6xx/R7xx, use GB_TILING_CONFIG. + /// But they can be treated as the same. + /// if this value is 0, use chip to set default value + UINT_32 backendDisables; ///< 1 bit per backend, starting with LSB. 1=disabled,0=enabled. + /// Register value of CC_RB_BACKEND_DISABLE.BACKEND_DISABLE + + /// R800 registers----------------------------------------------- + UINT_32 noOfBanks; ///< Number of h/w ram banks - For r800: MC_ARB_RAMCFG.NOOFBANK + /// No enums for this value in h/w header files + /// 0: 4 + /// 1: 8 + /// 2: 16 + UINT_32 noOfRanks; /// MC_ARB_RAMCFG.NOOFRANK + /// 0: 1 + /// 1: 2 + /// SI (R1000) registers----------------------------------------- + const UINT_32* pTileConfig; ///< Global tile setting tables + UINT_32 noOfEntries; ///< Number of entries in pTileConfig + + ///< CI registers------------------------------------------------- + const UINT_32* pMacroTileConfig; ///< Global macro tile mode table + UINT_32 noOfMacroEntries; ///< Number of entries in pMacroTileConfig + +} ADDR_REGISTER_VALUE; + +/** +*************************************************************************************************** +* ADDR_CREATE_INPUT +* +* @brief +* Parameters use to create an AddrLib Object. Caller must provide all fields. +* +*************************************************************************************************** +*/ +typedef struct _ADDR_CREATE_INPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + UINT_32 chipEngine; ///< Chip Engine + UINT_32 chipFamily; ///< Chip Family + UINT_32 chipRevision; ///< Chip Revision + ADDR_CALLBACKS callbacks; ///< Callbacks for sysmem alloc/free/print + ADDR_CREATE_FLAGS createFlags; ///< Flags to setup AddrLib + ADDR_REGISTER_VALUE regValue; ///< Data from registers to setup AddrLib global data + ADDR_CLIENT_HANDLE hClient; ///< Client handle + UINT_32 minPitchAlignPixels; ///< Minimum pitch alignment in pixels +} ADDR_CREATE_INPUT; + +/** +*************************************************************************************************** +* ADDR_CREATEINFO_OUTPUT +* +* @brief +* Return AddrLib handle to client driver +* +*************************************************************************************************** +*/ +typedef struct _ADDR_CREATE_OUTPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + ADDR_HANDLE hLib; ///< Address lib handle +} ADDR_CREATE_OUTPUT; + +/** +*************************************************************************************************** +* AddrCreate +* +* @brief +* Create AddrLib object, must be called before any interface calls +* +* @return +* ADDR_OK if successful +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrCreate( + const ADDR_CREATE_INPUT* pAddrCreateIn, + ADDR_CREATE_OUTPUT* pAddrCreateOut); + + + +/** +*************************************************************************************************** +* AddrDestroy +* +* @brief +* Destroy AddrLib object, must be called to free internally allocated resources. +* +* @return +* ADDR_OK if successful +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrDestroy( + ADDR_HANDLE hLib); + + + +/////////////////////////////////////////////////////////////////////////////////////////////////// +// Surface functions +/////////////////////////////////////////////////////////////////////////////////////////////////// + +/** +*************************************************************************************************** +* @brief +* Bank/tiling parameters. On function input, these can be set as desired or +* left 0 for AddrLib to calculate/default. On function output, these are the actual +* parameters used. +* @note +* Valid bankWidth/bankHeight value: +* 1,2,4,8. They are factors instead of pixels or bytes. +* +* The bank number remains constant across each row of the +* macro tile as each pipe is selected, so the number of +* tiles in the x direction with the same bank number will +* be bank_width * num_pipes. +*************************************************************************************************** +*/ +typedef struct _ADDR_TILEINFO +{ + /// Any of these parameters can be set to 0 to use the HW default. + UINT_32 banks; ///< Number of banks, numerical value + UINT_32 bankWidth; ///< Number of tiles in the X direction in the same bank + UINT_32 bankHeight; ///< Number of tiles in the Y direction in the same bank + UINT_32 macroAspectRatio; ///< Macro tile aspect ratio. 1-1:1, 2-4:1, 4-16:1, 8-64:1 + UINT_32 tileSplitBytes; ///< Tile split size, in bytes + AddrPipeCfg pipeConfig; ///< Pipe Config = HW enum + 1 +} ADDR_TILEINFO; + +// Create a define to avoid client change. The removal of R800 is because we plan to implement SI +// within 800 HWL - An AddrPipeCfg is added in above data structure +typedef ADDR_TILEINFO ADDR_R800_TILEINFO; + +/** +*************************************************************************************************** +* @brief +* Information needed by quad buffer stereo support +*************************************************************************************************** +*/ +typedef struct _ADDR_QBSTEREOINFO +{ + UINT_32 eyeHeight; ///< Height (in pixel rows) to right eye + UINT_32 rightOffset; ///< Offset (in bytes) to right eye + UINT_32 rightSwizzle; ///< TileSwizzle for right eyes +} ADDR_QBSTEREOINFO; + +/** +*************************************************************************************************** +* ADDR_SURFACE_FLAGS +* +* @brief +* Surface flags +*************************************************************************************************** +*/ +typedef union _ADDR_SURFACE_FLAGS +{ + struct + { + UINT_32 color : 1; ///< Flag indicates this is a color buffer + UINT_32 depth : 1; ///< Flag indicates this is a depth/stencil buffer + UINT_32 stencil : 1; ///< Flag indicates this is a stencil buffer + UINT_32 texture : 1; ///< Flag indicates this is a texture + UINT_32 cube : 1; ///< Flag indicates this is a cubemap + + UINT_32 volume : 1; ///< Flag indicates this is a volume texture + UINT_32 fmask : 1; ///< Flag indicates this is an fmask + UINT_32 cubeAsArray : 1; ///< Flag indicates if treat cubemap as arrays + UINT_32 compressZ : 1; ///< Flag indicates z buffer is compressed + UINT_32 overlay : 1; ///< Flag indicates this is an overlay surface + UINT_32 noStencil : 1; ///< Flag indicates this depth has no separate stencil + UINT_32 display : 1; ///< Flag indicates this should match display controller req. + UINT_32 opt4Space : 1; ///< Flag indicates this surface should be optimized for space + /// i.e. save some memory but may lose performance + UINT_32 prt : 1; ///< Flag for partially resident texture + UINT_32 qbStereo : 1; ///< Quad buffer stereo surface + UINT_32 pow2Pad : 1; ///< SI: Pad to pow2, must set for mipmap (include level0) + UINT_32 interleaved : 1; ///< Special flag for interleaved YUV surface padding + UINT_32 degrade4Space : 1; ///< Degrade base level's tile mode to save memory + UINT_32 tcCompatible : 1; ///< Flag indicates surface needs to be shader readable + UINT_32 dispTileType : 1; ///< NI: force display Tiling for 128 bit shared resoruce + UINT_32 dccCompatible : 1; ///< VI: whether to support dcc fast clear + UINT_32 czDispCompatible: 1; ///< SI+: CZ family (Carrizo) has a HW bug needs special alignment. + ///< This flag indicates we need to follow the alignment with + ///< CZ families or other ASICs under PX configuration + CZ. + UINT_32 reserved :10; ///< Reserved bits + }; + + UINT_32 value; +} ADDR_SURFACE_FLAGS; + +/** +*************************************************************************************************** +* ADDR_COMPUTE_SURFACE_INFO_INPUT +* +* @brief +* Input structure for AddrComputeSurfaceInfo +*************************************************************************************************** +*/ +typedef struct _ADDR_COMPUTE_SURFACE_INFO_INPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + AddrTileMode tileMode; ///< Tile mode + AddrFormat format; ///< If format is set to valid one, bpp/width/height + /// might be overwritten + UINT_32 bpp; ///< Bits per pixel + UINT_32 numSamples; ///< Number of samples + UINT_32 width; ///< Width, in pixels + UINT_32 height; ///< Height, in pixels + UINT_32 numSlices; ///< Number surface slice/depth, + /// Note: + /// For cubemap, driver clients usually set numSlices + /// to 1 in per-face calc. + /// For 7xx and above, we need pad faces as slices. + /// In this case, clients should set numSlices to 6 and + /// this is also can be turned off by createFlags when + /// calling AddrCreate + UINT_32 slice; ///< Slice index + UINT_32 mipLevel; ///< Current mipmap level. + /// Padding/tiling have different rules for level0 and + /// sublevels + ADDR_SURFACE_FLAGS flags; ///< Surface type flags + UINT_32 numFrags; ///< Number of fragments, leave it zero or the same as + /// number of samples for normal AA; Set it to the + /// number of fragments for EQAA + /// r800 and later HWL parameters + // Needed by 2D tiling, for linear and 1D tiling, just keep them 0's + ADDR_TILEINFO* pTileInfo; ///< 2D tile parameters. Set to 0 to default/calculate + AddrTileType tileType; ///< Micro tiling type, not needed when tileIndex != -1 + INT_32 tileIndex; ///< Tile index, MUST be -1 if you don't want to use it + /// while the global useTileIndex is set to 1 + UINT_32 basePitch; ///< Base level pitch in pixels, 0 means ignored, is a + /// must for mip levels from SI+. + /// Don't use pitch in blocks for compressed formats! +} ADDR_COMPUTE_SURFACE_INFO_INPUT; + +/** +*************************************************************************************************** +* ADDR_COMPUTE_SURFACE_INFO_OUTPUT +* +* @brief +* Output structure for AddrComputeSurfInfo +* @note + Element: AddrLib unit for computing. e.g. BCn: 4x4 blocks; R32B32B32: 32bit with 3x pitch + Pixel: Original pixel +*************************************************************************************************** +*/ +typedef struct _ADDR_COMPUTE_SURFACE_INFO_OUTPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + UINT_32 pitch; ///< Pitch in elements (in blocks for compressed formats) + UINT_32 height; ///< Height in elements (in blocks for compressed formats) + UINT_32 depth; ///< Number of slice/depth + UINT_64 surfSize; ///< Surface size in bytes + AddrTileMode tileMode; ///< Actual tile mode. May differ from that in input + UINT_32 baseAlign; ///< Base address alignment + UINT_32 pitchAlign; ///< Pitch alignment, in elements + UINT_32 heightAlign; ///< Height alignment, in elements + UINT_32 depthAlign; ///< Depth alignment, aligned to thickness, for 3d texture + UINT_32 bpp; ///< Bits per elements (e.g. blocks for BCn, 1/3 for 96bit) + UINT_32 pixelPitch; ///< Pitch in original pixels + UINT_32 pixelHeight; ///< Height in original pixels + UINT_32 pixelBits; ///< Original bits per pixel, passed from input + UINT_64 sliceSize; ///< Size of slice specified by input's slice + /// The result is controlled by surface flags & createFlags + /// By default this value equals to surfSize for volume + UINT_32 pitchTileMax; ///< PITCH_TILE_MAX value for h/w register + UINT_32 heightTileMax; ///< HEIGHT_TILE_MAX value for h/w register + UINT_32 sliceTileMax; ///< SLICE_TILE_MAX value for h/w register + + UINT_32 numSamples; ///< Pass the effective numSamples processed in this call + + /// r800 and later HWL parameters + ADDR_TILEINFO* pTileInfo; ///< Tile parameters used. Filled in if 0 on input + AddrTileType tileType; ///< Micro tiling type, only valid when tileIndex != -1 + INT_32 tileIndex; ///< Tile index, MAY be "downgraded" + + INT_32 macroModeIndex; ///< Index in macro tile mode table if there is one (CI) + /// Special information to work around SI mipmap swizzle bug UBTS #317508 + BOOL_32 last2DLevel; ///< TRUE if this is the last 2D(3D) tiled + ///< Only meaningful when create flag checkLast2DLevel is set + /// Stereo info + ADDR_QBSTEREOINFO* pStereoInfo;///< Stereo information, needed when .qbStereo flag is TRUE +} ADDR_COMPUTE_SURFACE_INFO_OUTPUT; + +/** +*************************************************************************************************** +* AddrComputeSurfaceInfo +* +* @brief +* Compute surface width/height/depth/alignments and suitable tiling mode +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrComputeSurfaceInfo( + ADDR_HANDLE hLib, + const ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn, + ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pOut); + + + +/** +*************************************************************************************************** +* ADDR_COMPUTE_SURFACE_ADDRFROMCOORD_INPUT +* +* @brief +* Input structure for AddrComputeSurfaceAddrFromCoord +*************************************************************************************************** +*/ +typedef struct _ADDR_COMPUTE_SURFACE_ADDRFROMCOORD_INPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + UINT_32 x; ///< X coordinate + UINT_32 y; ///< Y coordinate + UINT_32 slice; ///< Slice index + UINT_32 sample; ///< Sample index, use fragment index for EQAA + + UINT_32 bpp; ///< Bits per pixel + UINT_32 pitch; ///< Surface pitch, in pixels + UINT_32 height; ///< Surface height, in pixels + UINT_32 numSlices; ///< Surface depth + UINT_32 numSamples; ///< Number of samples + + AddrTileMode tileMode; ///< Tile mode + BOOL_32 isDepth; ///< TRUE if the surface uses depth sample ordering within + /// micro tile. Textures can also choose depth sample order + UINT_32 tileBase; ///< Base offset (in bits) inside micro tile which handles + /// the case that components are stored separately + UINT_32 compBits; ///< The component bits actually needed(for planar surface) + + UINT_32 numFrags; ///< Number of fragments, leave it zero or the same as + /// number of samples for normal AA; Set it to the + /// number of fragments for EQAA + /// r800 and later HWL parameters + // Used for 1D tiling above + AddrTileType tileType; ///< See defintion of AddrTileType + struct + { + UINT_32 ignoreSE : 1; ///< TRUE if shader engines are ignored. This is texture + /// only flag. Only non-RT texture can set this to TRUE + UINT_32 reserved :31; ///< Reserved for future use. + }; + // 2D tiling needs following structure + ADDR_TILEINFO* pTileInfo; ///< 2D tile parameters. Client must provide all data + INT_32 tileIndex; ///< Tile index, MUST be -1 if you don't want to use it + /// while the global useTileIndex is set to 1 + union + { + struct + { + UINT_32 bankSwizzle; ///< Bank swizzle + UINT_32 pipeSwizzle; ///< Pipe swizzle + }; + UINT_32 tileSwizzle; ///< Combined swizzle, if useCombinedSwizzle is TRUE + }; + +#if ADDR_AM_BUILD // These two fields are not valid in SW blt since no HTILE access + UINT_32 addr5Swizzle; ///< ADDR5_SWIZZLE_MASK of DB_DEPTH_INFO + BOOL_32 is32ByteTile; ///< Caller must have access to HTILE buffer and know if + /// this tile is compressed to 32B +#endif +} ADDR_COMPUTE_SURFACE_ADDRFROMCOORD_INPUT; + +/** +*************************************************************************************************** +* ADDR_COMPUTE_SURFACE_ADDRFROMCOORD_OUTPUT +* +* @brief +* Output structure for AddrComputeSurfaceAddrFromCoord +*************************************************************************************************** +*/ +typedef struct _ADDR_COMPUTE_SURFACE_ADDRFROMCOORD_OUTPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + UINT_64 addr; ///< Byte address + UINT_32 bitPosition; ///< Bit position within surfaceAddr, 0-7. + /// For surface bpp < 8, e.g. FMT_1. + UINT_32 prtBlockIndex; ///< Index of a PRT tile (64K block) +} ADDR_COMPUTE_SURFACE_ADDRFROMCOORD_OUTPUT; + +/** +*************************************************************************************************** +* AddrComputeSurfaceAddrFromCoord +* +* @brief +* Compute surface address from a given coordinate. +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrComputeSurfaceAddrFromCoord( + ADDR_HANDLE hLib, + const ADDR_COMPUTE_SURFACE_ADDRFROMCOORD_INPUT* pIn, + ADDR_COMPUTE_SURFACE_ADDRFROMCOORD_OUTPUT* pOut); + + + +/** +*************************************************************************************************** +* ADDR_COMPUTE_SURFACE_COORDFROMADDR_INPUT +* +* @brief +* Input structure for AddrComputeSurfaceCoordFromAddr +*************************************************************************************************** +*/ +typedef struct _ADDR_COMPUTE_SURFACE_COORDFROMADDR_INPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + UINT_64 addr; ///< Address in bytes + UINT_32 bitPosition; ///< Bit position in addr. 0-7. for surface bpp < 8, + /// e.g. FMT_1; + UINT_32 bpp; ///< Bits per pixel + UINT_32 pitch; ///< Pitch, in pixels + UINT_32 height; ///< Height in pixels + UINT_32 numSlices; ///< Surface depth + UINT_32 numSamples; ///< Number of samples + + AddrTileMode tileMode; ///< Tile mode + BOOL_32 isDepth; ///< Surface uses depth sample ordering within micro tile. + /// Note: Textures can choose depth sample order as well. + UINT_32 tileBase; ///< Base offset (in bits) inside micro tile which handles + /// the case that components are stored separately + UINT_32 compBits; ///< The component bits actually needed(for planar surface) + + UINT_32 numFrags; ///< Number of fragments, leave it zero or the same as + /// number of samples for normal AA; Set it to the + /// number of fragments for EQAA + /// r800 and later HWL parameters + // Used for 1D tiling above + AddrTileType tileType; ///< See defintion of AddrTileType + struct + { + UINT_32 ignoreSE : 1; ///< TRUE if shader engines are ignored. This is texture + /// only flag. Only non-RT texture can set this to TRUE + UINT_32 reserved :31; ///< Reserved for future use. + }; + // 2D tiling needs following structure + ADDR_TILEINFO* pTileInfo; ///< 2D tile parameters. Client must provide all data + INT_32 tileIndex; ///< Tile index, MUST be -1 if you don't want to use it + /// while the global useTileIndex is set to 1 + union + { + struct + { + UINT_32 bankSwizzle; ///< Bank swizzle + UINT_32 pipeSwizzle; ///< Pipe swizzle + }; + UINT_32 tileSwizzle; ///< Combined swizzle, if useCombinedSwizzle is TRUE + }; +} ADDR_COMPUTE_SURFACE_COORDFROMADDR_INPUT; + +/** +*************************************************************************************************** +* ADDR_COMPUTE_SURFACE_COORDFROMADDR_OUTPUT +* +* @brief +* Output structure for AddrComputeSurfaceCoordFromAddr +*************************************************************************************************** +*/ +typedef struct _ADDR_COMPUTE_SURFACE_COORDFROMADDR_OUTPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + UINT_32 x; ///< X coordinate + UINT_32 y; ///< Y coordinate + UINT_32 slice; ///< Index of slices + UINT_32 sample; ///< Index of samples, means fragment index for EQAA +} ADDR_COMPUTE_SURFACE_COORDFROMADDR_OUTPUT; + +/** +*************************************************************************************************** +* AddrComputeSurfaceCoordFromAddr +* +* @brief +* Compute coordinate from a given surface address +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrComputeSurfaceCoordFromAddr( + ADDR_HANDLE hLib, + const ADDR_COMPUTE_SURFACE_COORDFROMADDR_INPUT* pIn, + ADDR_COMPUTE_SURFACE_COORDFROMADDR_OUTPUT* pOut); + +/////////////////////////////////////////////////////////////////////////////////////////////////// +// HTile functions +/////////////////////////////////////////////////////////////////////////////////////////////////// + +/** +*************************************************************************************************** +* ADDR_HTILE_FLAGS +* +* @brief +* HTILE flags +*************************************************************************************************** +*/ +typedef union _ADDR_HTILE_FLAGS +{ + struct + { + UINT_32 tcCompatible : 1; ///< Flag indicates surface needs to be shader readable + UINT_32 reserved :31; ///< Reserved bits + }; + + UINT_32 value; +} ADDR_HTILE_FLAGS; + +/** +*************************************************************************************************** +* ADDR_COMPUTE_HTILE_INFO_INPUT +* +* @brief +* Input structure of AddrComputeHtileInfo +*************************************************************************************************** +*/ +typedef struct _ADDR_COMPUTE_HTILE_INFO_INPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + ADDR_HTILE_FLAGS flags; ///< HTILE flags + UINT_32 pitch; ///< Surface pitch, in pixels + UINT_32 height; ///< Surface height, in pixels + UINT_32 numSlices; ///< Number of slices + BOOL_32 isLinear; ///< Linear or tiled HTILE layout + AddrHtileBlockSize blockWidth; ///< 4 or 8. EG above only support 8 + AddrHtileBlockSize blockHeight; ///< 4 or 8. EG above only support 8 + ADDR_TILEINFO* pTileInfo; ///< Tile info + + INT_32 tileIndex; ///< Tile index, MUST be -1 if you don't want to use it + /// while the global useTileIndex is set to 1 + INT_32 macroModeIndex; ///< Index in macro tile mode table if there is one (CI) + ///< README: When tileIndex is not -1, this must be valid +} ADDR_COMPUTE_HTILE_INFO_INPUT; + +/** +*************************************************************************************************** +* ADDR_COMPUTE_HTILE_INFO_OUTPUT +* +* @brief +* Output structure of AddrComputeHtileInfo +*************************************************************************************************** +*/ +typedef struct _ADDR_COMPUTE_HTILE_INFO_OUTPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + UINT_32 pitch; ///< Pitch in pixels of depth buffer represented in this + /// HTile buffer. This might be larger than original depth + /// buffer pitch when called with an unaligned pitch. + UINT_32 height; ///< Height in pixels, as above + UINT_64 htileBytes; ///< Size of HTILE buffer, in bytes + UINT_32 baseAlign; ///< Base alignment + UINT_32 bpp; ///< Bits per pixel for HTILE is how many bits for an 8x8 block! + UINT_32 macroWidth; ///< Macro width in pixels, actually squared cache shape + UINT_32 macroHeight; ///< Macro height in pixels + UINT_64 sliceSize; ///< Slice size, in bytes. +} ADDR_COMPUTE_HTILE_INFO_OUTPUT; + +/** +*************************************************************************************************** +* AddrComputeHtileInfo +* +* @brief +* Compute Htile pitch, height, base alignment and size in bytes +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrComputeHtileInfo( + ADDR_HANDLE hLib, + const ADDR_COMPUTE_HTILE_INFO_INPUT* pIn, + ADDR_COMPUTE_HTILE_INFO_OUTPUT* pOut); + + + +/** +*************************************************************************************************** +* ADDR_COMPUTE_HTILE_ADDRFROMCOORD_INPUT +* +* @brief +* Input structure for AddrComputeHtileAddrFromCoord +*************************************************************************************************** +*/ +typedef struct _ADDR_COMPUTE_HTILE_ADDRFROMCOORD_INPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + UINT_32 pitch; ///< Pitch, in pixels + UINT_32 height; ///< Height in pixels + UINT_32 x; ///< X coordinate + UINT_32 y; ///< Y coordinate + UINT_32 slice; ///< Index of slice + UINT_32 numSlices; ///< Number of slices + BOOL_32 isLinear; ///< Linear or tiled HTILE layout + AddrHtileBlockSize blockWidth; ///< 4 or 8. 1 means 8, 0 means 4. EG above only support 8 + AddrHtileBlockSize blockHeight; ///< 4 or 8. 1 means 8, 0 means 4. EG above only support 8 + ADDR_TILEINFO* pTileInfo; ///< Tile info + + INT_32 tileIndex; ///< Tile index, MUST be -1 if you don't want to use it + /// while the global useTileIndex is set to 1 + INT_32 macroModeIndex; ///< Index in macro tile mode table if there is one (CI) + ///< README: When tileIndex is not -1, this must be valid +} ADDR_COMPUTE_HTILE_ADDRFROMCOORD_INPUT; + +/** +*************************************************************************************************** +* ADDR_COMPUTE_HTILE_ADDRFROMCOORD_OUTPUT +* +* @brief +* Output structure for AddrComputeHtileAddrFromCoord +*************************************************************************************************** +*/ +typedef struct _ADDR_COMPUTE_HTILE_ADDRFROMCOORD_OUTPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + UINT_64 addr; ///< Address in bytes + UINT_32 bitPosition; ///< Bit position, 0 or 4. CMASK and HTILE shares some lib method. + /// So we keep bitPosition for HTILE as well +} ADDR_COMPUTE_HTILE_ADDRFROMCOORD_OUTPUT; + +/** +*************************************************************************************************** +* AddrComputeHtileAddrFromCoord +* +* @brief +* Compute Htile address according to coordinates (of depth buffer) +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrComputeHtileAddrFromCoord( + ADDR_HANDLE hLib, + const ADDR_COMPUTE_HTILE_ADDRFROMCOORD_INPUT* pIn, + ADDR_COMPUTE_HTILE_ADDRFROMCOORD_OUTPUT* pOut); + + + +/** +*************************************************************************************************** +* ADDR_COMPUTE_HTILE_COORDFROMADDR_INPUT +* +* @brief +* Input structure for AddrComputeHtileCoordFromAddr +*************************************************************************************************** +*/ +typedef struct _ADDR_COMPUTE_HTILE_COORDFROMADDR_INPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + UINT_64 addr; ///< Address + UINT_32 bitPosition; ///< Bit position 0 or 4. CMASK and HTILE share some methods + /// so we keep bitPosition for HTILE as well + UINT_32 pitch; ///< Pitch, in pixels + UINT_32 height; ///< Height, in pixels + UINT_32 numSlices; ///< Number of slices + BOOL_32 isLinear; ///< Linear or tiled HTILE layout + AddrHtileBlockSize blockWidth; ///< 4 or 8. 1 means 8, 0 means 4. R8xx/R9xx only support 8 + AddrHtileBlockSize blockHeight; ///< 4 or 8. 1 means 8, 0 means 4. R8xx/R9xx only support 8 + ADDR_TILEINFO* pTileInfo; ///< Tile info + + INT_32 tileIndex; ///< Tile index, MUST be -1 if you don't want to use it + /// while the global useTileIndex is set to 1 + INT_32 macroModeIndex; ///< Index in macro tile mode table if there is one (CI) + ///< README: When tileIndex is not -1, this must be valid +} ADDR_COMPUTE_HTILE_COORDFROMADDR_INPUT; + +/** +*************************************************************************************************** +* ADDR_COMPUTE_HTILE_COORDFROMADDR_OUTPUT +* +* @brief +* Output structure for AddrComputeHtileCoordFromAddr +*************************************************************************************************** +*/ +typedef struct _ADDR_COMPUTE_HTILE_COORDFROMADDR_OUTPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + UINT_32 x; ///< X coordinate + UINT_32 y; ///< Y coordinate + UINT_32 slice; ///< Slice index +} ADDR_COMPUTE_HTILE_COORDFROMADDR_OUTPUT; + +/** +*************************************************************************************************** +* AddrComputeHtileCoordFromAddr +* +* @brief +* Compute coordinates within depth buffer (1st pixel of a micro tile) according to +* Htile address +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrComputeHtileCoordFromAddr( + ADDR_HANDLE hLib, + const ADDR_COMPUTE_HTILE_COORDFROMADDR_INPUT* pIn, + ADDR_COMPUTE_HTILE_COORDFROMADDR_OUTPUT* pOut); + + + +/////////////////////////////////////////////////////////////////////////////////////////////////// +// C-mask functions +/////////////////////////////////////////////////////////////////////////////////////////////////// + +/** +*************************************************************************************************** +* ADDR_CMASK_FLAGS +* +* @brief +* CMASK flags +*************************************************************************************************** +*/ +typedef union _ADDR_CMASK_FLAGS +{ + struct + { + UINT_32 tcCompatible : 1; ///< Flag indicates surface needs to be shader readable + UINT_32 reserved :31; ///< Reserved bits + }; + + UINT_32 value; +} ADDR_CMASK_FLAGS; + +/** +*************************************************************************************************** +* ADDR_COMPUTE_CMASK_INFO_INPUT +* +* @brief +* Input structure of AddrComputeCmaskInfo +*************************************************************************************************** +*/ +typedef struct _ADDR_COMPUTE_CMASKINFO_INPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + ADDR_CMASK_FLAGS flags; ///< CMASK flags + UINT_32 pitch; ///< Pitch, in pixels, of color buffer + UINT_32 height; ///< Height, in pixels, of color buffer + UINT_32 numSlices; ///< Number of slices, of color buffer + BOOL_32 isLinear; ///< Linear or tiled layout, Only SI can be linear + ADDR_TILEINFO* pTileInfo; ///< Tile info + + INT_32 tileIndex; ///< Tile index, MUST be -1 if you don't want to use it + /// while the global useTileIndex is set to 1 + INT_32 macroModeIndex; ///< Index in macro tile mode table if there is one (CI) + ///< README: When tileIndex is not -1, this must be valid +} ADDR_COMPUTE_CMASK_INFO_INPUT; + +/** +*************************************************************************************************** +* ADDR_COMPUTE_CMASK_INFO_OUTPUT +* +* @brief +* Output structure of AddrComputeCmaskInfo +*************************************************************************************************** +*/ +typedef struct _ADDR_COMPUTE_CMASK_INFO_OUTPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + UINT_32 pitch; ///< Pitch in pixels of color buffer which + /// this Cmask matches. The size might be larger than + /// original color buffer pitch when called with + /// an unaligned pitch. + UINT_32 height; ///< Height in pixels, as above + UINT_64 cmaskBytes; ///< Size in bytes of CMask buffer + UINT_32 baseAlign; ///< Base alignment + UINT_32 blockMax; ///< Cmask block size. Need this to set CB_COLORn_MASK register + UINT_32 macroWidth; ///< Macro width in pixels, actually squared cache shape + UINT_32 macroHeight; ///< Macro height in pixels + UINT_64 sliceSize; ///< Slice size, in bytes. +} ADDR_COMPUTE_CMASK_INFO_OUTPUT; + +/** +*************************************************************************************************** +* AddrComputeCmaskInfo +* +* @brief +* Compute Cmask pitch, height, base alignment and size in bytes from color buffer +* info +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrComputeCmaskInfo( + ADDR_HANDLE hLib, + const ADDR_COMPUTE_CMASK_INFO_INPUT* pIn, + ADDR_COMPUTE_CMASK_INFO_OUTPUT* pOut); + + + +/** +*************************************************************************************************** +* ADDR_COMPUTE_CMASK_ADDRFROMCOORD_INPUT +* +* @brief +* Input structure for AddrComputeCmaskAddrFromCoord +* +*************************************************************************************************** +*/ +typedef struct _ADDR_COMPUTE_CMASK_ADDRFROMCOORD_INPUT +{ + UINT_32 size; ///< Size of this structure in bytes + UINT_32 x; ///< X coordinate + UINT_32 y; ///< Y coordinate + UINT_64 fmaskAddr; ///< Fmask addr for tc compatible Cmask + UINT_32 slice; ///< Slice index + UINT_32 pitch; ///< Pitch in pixels, of color buffer + UINT_32 height; ///< Height in pixels, of color buffer + UINT_32 numSlices; ///< Number of slices + UINT_32 bpp; + BOOL_32 isLinear; ///< Linear or tiled layout, Only SI can be linear + ADDR_CMASK_FLAGS flags; ///< CMASK flags + ADDR_TILEINFO* pTileInfo; ///< Tile info + + INT_32 tileIndex; ///< Tile index, MUST be -1 if you don't want to use it + ///< while the global useTileIndex is set to 1 + INT_32 macroModeIndex; ///< Index in macro tile mode table if there is one (CI) + ///< README: When tileIndex is not -1, this must be valid +} ADDR_COMPUTE_CMASK_ADDRFROMCOORD_INPUT; + +/** +*************************************************************************************************** +* ADDR_COMPUTE_CMASK_ADDRFROMCOORD_OUTPUT +* +* @brief +* Output structure for AddrComputeCmaskAddrFromCoord +*************************************************************************************************** +*/ +typedef struct _ADDR_COMPUTE_CMASK_ADDRFROMCOORD_OUTPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + UINT_64 addr; ///< CMASK address in bytes + UINT_32 bitPosition; ///< Bit position within addr, 0-7. CMASK is 4 bpp, + /// so the address may be located in bit 0 (0) or 4 (4) +} ADDR_COMPUTE_CMASK_ADDRFROMCOORD_OUTPUT; + +/** +*************************************************************************************************** +* AddrComputeCmaskAddrFromCoord +* +* @brief +* Compute Cmask address according to coordinates (of MSAA color buffer) +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrComputeCmaskAddrFromCoord( + ADDR_HANDLE hLib, + const ADDR_COMPUTE_CMASK_ADDRFROMCOORD_INPUT* pIn, + ADDR_COMPUTE_CMASK_ADDRFROMCOORD_OUTPUT* pOut); + + + +/** +*************************************************************************************************** +* ADDR_COMPUTE_CMASK_COORDFROMADDR_INPUT +* +* @brief +* Input structure for AddrComputeCmaskCoordFromAddr +*************************************************************************************************** +*/ +typedef struct _ADDR_COMPUTE_CMASK_COORDFROMADDR_INPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + UINT_64 addr; ///< CMASK address in bytes + UINT_32 bitPosition; ///< Bit position within addr, 0-7. CMASK is 4 bpp, + /// so the address may be located in bit 0 (0) or 4 (4) + UINT_32 pitch; ///< Pitch, in pixels + UINT_32 height; ///< Height in pixels + UINT_32 numSlices; ///< Number of slices + BOOL_32 isLinear; ///< Linear or tiled layout, Only SI can be linear + ADDR_TILEINFO* pTileInfo; ///< Tile info + + INT_32 tileIndex; ///< Tile index, MUST be -1 if you don't want to use it + /// while the global useTileIndex is set to 1 + INT_32 macroModeIndex; ///< Index in macro tile mode table if there is one (CI) + ///< README: When tileIndex is not -1, this must be valid +} ADDR_COMPUTE_CMASK_COORDFROMADDR_INPUT; + +/** +*************************************************************************************************** +* ADDR_COMPUTE_CMASK_COORDFROMADDR_OUTPUT +* +* @brief +* Output structure for AddrComputeCmaskCoordFromAddr +*************************************************************************************************** +*/ +typedef struct _ADDR_COMPUTE_CMASK_COORDFROMADDR_OUTPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + UINT_32 x; ///< X coordinate + UINT_32 y; ///< Y coordinate + UINT_32 slice; ///< Slice index +} ADDR_COMPUTE_CMASK_COORDFROMADDR_OUTPUT; + +/** +*************************************************************************************************** +* AddrComputeCmaskCoordFromAddr +* +* @brief +* Compute coordinates within color buffer (1st pixel of a micro tile) according to +* Cmask address +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrComputeCmaskCoordFromAddr( + ADDR_HANDLE hLib, + const ADDR_COMPUTE_CMASK_COORDFROMADDR_INPUT* pIn, + ADDR_COMPUTE_CMASK_COORDFROMADDR_OUTPUT* pOut); + + + +/////////////////////////////////////////////////////////////////////////////////////////////////// +// F-mask functions +/////////////////////////////////////////////////////////////////////////////////////////////////// + +/** +*************************************************************************************************** +* ADDR_COMPUTE_FMASK_INFO_INPUT +* +* @brief +* Input structure for AddrComputeFmaskInfo +*************************************************************************************************** +*/ +typedef struct _ADDR_COMPUTE_FMASK_INFO_INPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + AddrTileMode tileMode; ///< Tile mode + UINT_32 pitch; ///< Surface pitch, in pixels + UINT_32 height; ///< Surface height, in pixels + UINT_32 numSlices; ///< Number of slice/depth + UINT_32 numSamples; ///< Number of samples + UINT_32 numFrags; ///< Number of fragments, leave it zero or the same as + /// number of samples for normal AA; Set it to the + /// number of fragments for EQAA + /// r800 and later HWL parameters + struct + { + UINT_32 resolved: 1; ///< TRUE if the surface is for resolved fmask, only used + /// by H/W clients. S/W should always set it to FALSE. + UINT_32 reserved: 31; ///< Reserved for future use. + }; + ADDR_TILEINFO* pTileInfo; ///< 2D tiling parameters. Clients must give valid data + INT_32 tileIndex; ///< Tile index, MUST be -1 if you don't want to use it + /// while the global useTileIndex is set to 1 +} ADDR_COMPUTE_FMASK_INFO_INPUT; + +/** +*************************************************************************************************** +* ADDR_COMPUTE_FMASK_INFO_OUTPUT +* +* @brief +* Output structure for AddrComputeFmaskInfo +*************************************************************************************************** +*/ +typedef struct _ADDR_COMPUTE_FMASK_INFO_OUTPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + UINT_32 pitch; ///< Pitch of fmask in pixels + UINT_32 height; ///< Height of fmask in pixels + UINT_32 numSlices; ///< Slices of fmask + UINT_64 fmaskBytes; ///< Size of fmask in bytes + UINT_32 baseAlign; ///< Base address alignment + UINT_32 pitchAlign; ///< Pitch alignment + UINT_32 heightAlign; ///< Height alignment + UINT_32 bpp; ///< Bits per pixel of FMASK is: number of bit planes + UINT_32 numSamples; ///< Number of samples, used for dump, export this since input + /// may be changed in 9xx and above + /// r800 and later HWL parameters + ADDR_TILEINFO* pTileInfo; ///< Tile parameters used. Fmask can have different + /// bank_height from color buffer + INT_32 tileIndex; ///< Tile index, MUST be -1 if you don't want to use it + /// while the global useTileIndex is set to 1 + INT_32 macroModeIndex; ///< Index in macro tile mode table if there is one (CI) + UINT_64 sliceSize; ///< Size of slice in bytes +} ADDR_COMPUTE_FMASK_INFO_OUTPUT; + +/** +*************************************************************************************************** +* AddrComputeFmaskInfo +* +* @brief +* Compute Fmask pitch/height/depth/alignments and size in bytes +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrComputeFmaskInfo( + ADDR_HANDLE hLib, + const ADDR_COMPUTE_FMASK_INFO_INPUT* pIn, + ADDR_COMPUTE_FMASK_INFO_OUTPUT* pOut); + + + +/** +*************************************************************************************************** +* ADDR_COMPUTE_FMASK_ADDRFROMCOORD_INPUT +* +* @brief +* Input structure for AddrComputeFmaskAddrFromCoord +*************************************************************************************************** +*/ +typedef struct _ADDR_COMPUTE_FMASK_ADDRFROMCOORD_INPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + UINT_32 x; ///< X coordinate + UINT_32 y; ///< Y coordinate + UINT_32 slice; ///< Slice index + UINT_32 plane; ///< Plane number + UINT_32 sample; ///< Sample index (fragment index for EQAA) + + UINT_32 pitch; ///< Surface pitch, in pixels + UINT_32 height; ///< Surface height, in pixels + UINT_32 numSamples; ///< Number of samples + UINT_32 numFrags; ///< Number of fragments, leave it zero or the same as + /// number of samples for normal AA; Set it to the + /// number of fragments for EQAA + + AddrTileMode tileMode; ///< Tile mode + union + { + struct + { + UINT_32 bankSwizzle; ///< Bank swizzle + UINT_32 pipeSwizzle; ///< Pipe swizzle + }; + UINT_32 tileSwizzle; ///< Combined swizzle, if useCombinedSwizzle is TRUE + }; + + /// r800 and later HWL parameters + struct + { + UINT_32 resolved: 1; ///< TRUE if this is a resolved fmask, used by H/W clients + UINT_32 ignoreSE: 1; ///< TRUE if shader engines are ignored. + UINT_32 reserved: 30; ///< Reserved for future use. + }; + ADDR_TILEINFO* pTileInfo; ///< 2D tiling parameters. Client must provide all data + +} ADDR_COMPUTE_FMASK_ADDRFROMCOORD_INPUT; + +/** +*************************************************************************************************** +* ADDR_COMPUTE_FMASK_ADDRFROMCOORD_OUTPUT +* +* @brief +* Output structure for AddrComputeFmaskAddrFromCoord +*************************************************************************************************** +*/ +typedef struct _ADDR_COMPUTE_FMASK_ADDRFROMCOORD_OUTPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + UINT_64 addr; ///< Fmask address + UINT_32 bitPosition; ///< Bit position within fmaskAddr, 0-7. +} ADDR_COMPUTE_FMASK_ADDRFROMCOORD_OUTPUT; + +/** +*************************************************************************************************** +* AddrComputeFmaskAddrFromCoord +* +* @brief +* Compute Fmask address according to coordinates (x,y,slice,sample,plane) +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrComputeFmaskAddrFromCoord( + ADDR_HANDLE hLib, + const ADDR_COMPUTE_FMASK_ADDRFROMCOORD_INPUT* pIn, + ADDR_COMPUTE_FMASK_ADDRFROMCOORD_OUTPUT* pOut); + + + +/** +*************************************************************************************************** +* ADDR_COMPUTE_FMASK_COORDFROMADDR_INPUT +* +* @brief +* Input structure for AddrComputeFmaskCoordFromAddr +*************************************************************************************************** +*/ +typedef struct _ADDR_COMPUTE_FMASK_COORDFROMADDR_INPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + UINT_64 addr; ///< Address + UINT_32 bitPosition; ///< Bit position within addr, 0-7. + + UINT_32 pitch; ///< Pitch, in pixels + UINT_32 height; ///< Height in pixels + UINT_32 numSamples; ///< Number of samples + UINT_32 numFrags; ///< Number of fragments + AddrTileMode tileMode; ///< Tile mode + union + { + struct + { + UINT_32 bankSwizzle; ///< Bank swizzle + UINT_32 pipeSwizzle; ///< Pipe swizzle + }; + UINT_32 tileSwizzle; ///< Combined swizzle, if useCombinedSwizzle is TRUE + }; + + /// r800 and later HWL parameters + struct + { + UINT_32 resolved: 1; ///< TRUE if this is a resolved fmask, used by HW components + UINT_32 ignoreSE: 1; ///< TRUE if shader engines are ignored. + UINT_32 reserved: 30; ///< Reserved for future use. + }; + ADDR_TILEINFO* pTileInfo; ///< 2D tile parameters. Client must provide all data + +} ADDR_COMPUTE_FMASK_COORDFROMADDR_INPUT; + +/** +*************************************************************************************************** +* ADDR_COMPUTE_FMASK_COORDFROMADDR_OUTPUT +* +* @brief +* Output structure for AddrComputeFmaskCoordFromAddr +*************************************************************************************************** +*/ +typedef struct _ADDR_COMPUTE_FMASK_COORDFROMADDR_OUTPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + UINT_32 x; ///< X coordinate + UINT_32 y; ///< Y coordinate + UINT_32 slice; ///< Slice index + UINT_32 plane; ///< Plane number + UINT_32 sample; ///< Sample index (fragment index for EQAA) +} ADDR_COMPUTE_FMASK_COORDFROMADDR_OUTPUT; + +/** +*************************************************************************************************** +* AddrComputeFmaskCoordFromAddr +* +* @brief +* Compute FMASK coordinate from an given address +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrComputeFmaskCoordFromAddr( + ADDR_HANDLE hLib, + const ADDR_COMPUTE_FMASK_COORDFROMADDR_INPUT* pIn, + ADDR_COMPUTE_FMASK_COORDFROMADDR_OUTPUT* pOut); + + + +/////////////////////////////////////////////////////////////////////////////////////////////////// +// Element/utility functions +/////////////////////////////////////////////////////////////////////////////////////////////////// + +/** +*************************************************************************************************** +* AddrGetVersion +* +* @brief +* Get AddrLib version number +*************************************************************************************************** +*/ +UINT_32 ADDR_API AddrGetVersion(ADDR_HANDLE hLib); + +/** +*************************************************************************************************** +* AddrUseTileIndex +* +* @brief +* Return TRUE if tileIndex is enabled in this address library +*************************************************************************************************** +*/ +BOOL_32 ADDR_API AddrUseTileIndex(ADDR_HANDLE hLib); + +/** +*************************************************************************************************** +* AddrUseCombinedSwizzle +* +* @brief +* Return TRUE if combined swizzle is enabled in this address library +*************************************************************************************************** +*/ +BOOL_32 ADDR_API AddrUseCombinedSwizzle(ADDR_HANDLE hLib); + +/** +*************************************************************************************************** +* ADDR_EXTRACT_BANKPIPE_SWIZZLE_INPUT +* +* @brief +* Input structure of AddrExtractBankPipeSwizzle +*************************************************************************************************** +*/ +typedef struct _ADDR_EXTRACT_BANKPIPE_SWIZZLE_INPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + UINT_32 base256b; ///< Base256b value + + /// r800 and later HWL parameters + ADDR_TILEINFO* pTileInfo; ///< 2D tile parameters. Client must provide all data + + INT_32 tileIndex; ///< Tile index, MUST be -1 if you don't want to use it + /// while the global useTileIndex is set to 1 + INT_32 macroModeIndex; ///< Index in macro tile mode table if there is one (CI) + ///< README: When tileIndex is not -1, this must be valid +} ADDR_EXTRACT_BANKPIPE_SWIZZLE_INPUT; + +/** +*************************************************************************************************** +* ADDR_EXTRACT_BANKPIPE_SWIZZLE_OUTPUT +* +* @brief +* Output structure of AddrExtractBankPipeSwizzle +*************************************************************************************************** +*/ +typedef struct _ADDR_EXTRACT_BANKPIPE_SWIZZLE_OUTPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + UINT_32 bankSwizzle; ///< Bank swizzle + UINT_32 pipeSwizzle; ///< Pipe swizzle +} ADDR_EXTRACT_BANKPIPE_SWIZZLE_OUTPUT; + +/** +*************************************************************************************************** +* AddrExtractBankPipeSwizzle +* +* @brief +* Extract Bank and Pipe swizzle from base256b +* @return +* ADDR_OK if no error +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrExtractBankPipeSwizzle( + ADDR_HANDLE hLib, + const ADDR_EXTRACT_BANKPIPE_SWIZZLE_INPUT* pIn, + ADDR_EXTRACT_BANKPIPE_SWIZZLE_OUTPUT* pOut); + + +/** +*************************************************************************************************** +* ADDR_COMBINE_BANKPIPE_SWIZZLE_INPUT +* +* @brief +* Input structure of AddrCombineBankPipeSwizzle +*************************************************************************************************** +*/ +typedef struct _ADDR_COMBINE_BANKPIPE_SWIZZLE_INPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + UINT_32 bankSwizzle; ///< Bank swizzle + UINT_32 pipeSwizzle; ///< Pipe swizzle + UINT_64 baseAddr; ///< Base address (leave it zero for driver clients) + + /// r800 and later HWL parameters + ADDR_TILEINFO* pTileInfo; ///< 2D tile parameters. Client must provide all data + + INT_32 tileIndex; ///< Tile index, MUST be -1 if you don't want to use it + /// while the global useTileIndex is set to 1 + INT_32 macroModeIndex; ///< Index in macro tile mode table if there is one (CI) + ///< README: When tileIndex is not -1, this must be valid +} ADDR_COMBINE_BANKPIPE_SWIZZLE_INPUT; + +/** +*************************************************************************************************** +* ADDR_COMBINE_BANKPIPE_SWIZZLE_OUTPUT +* +* @brief +* Output structure of AddrCombineBankPipeSwizzle +*************************************************************************************************** +*/ +typedef struct _ADDR_COMBINE_BANKPIPE_SWIZZLE_OUTPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + UINT_32 tileSwizzle; ///< Combined swizzle +} ADDR_COMBINE_BANKPIPE_SWIZZLE_OUTPUT; + +/** +*************************************************************************************************** +* AddrCombineBankPipeSwizzle +* +* @brief +* Combine Bank and Pipe swizzle +* @return +* ADDR_OK if no error +* @note +* baseAddr here is full MCAddress instead of base256b +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrCombineBankPipeSwizzle( + ADDR_HANDLE hLib, + const ADDR_COMBINE_BANKPIPE_SWIZZLE_INPUT* pIn, + ADDR_COMBINE_BANKPIPE_SWIZZLE_OUTPUT* pOut); + + + +/** +*************************************************************************************************** +* ADDR_COMPUTE_SLICESWIZZLE_INPUT +* +* @brief +* Input structure of AddrComputeSliceSwizzle +*************************************************************************************************** +*/ +typedef struct _ADDR_COMPUTE_SLICESWIZZLE_INPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + AddrTileMode tileMode; ///< Tile Mode + UINT_32 baseSwizzle; ///< Base tile swizzle + UINT_32 slice; ///< Slice index + UINT_64 baseAddr; ///< Base address, driver should leave it 0 in most cases + + /// r800 and later HWL parameters + ADDR_TILEINFO* pTileInfo; ///< 2D tile parameters. Actually banks needed here! + + INT_32 tileIndex; ///< Tile index, MUST be -1 if you don't want to use it + /// while the global useTileIndex is set to 1 + INT_32 macroModeIndex; ///< Index in macro tile mode table if there is one (CI) + ///< README: When tileIndex is not -1, this must be valid +} ADDR_COMPUTE_SLICESWIZZLE_INPUT; + + + +/** +*************************************************************************************************** +* ADDR_COMPUTE_SLICESWIZZLE_OUTPUT +* +* @brief +* Output structure of AddrComputeSliceSwizzle +*************************************************************************************************** +*/ +typedef struct _ADDR_COMPUTE_SLICESWIZZLE_OUTPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + UINT_32 tileSwizzle; ///< Recalculated tileSwizzle value +} ADDR_COMPUTE_SLICESWIZZLE_OUTPUT; + +/** +*************************************************************************************************** +* AddrComputeSliceSwizzle +* +* @brief +* Extract Bank and Pipe swizzle from base256b +* @return +* ADDR_OK if no error +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrComputeSliceSwizzle( + ADDR_HANDLE hLib, + const ADDR_COMPUTE_SLICESWIZZLE_INPUT* pIn, + ADDR_COMPUTE_SLICESWIZZLE_OUTPUT* pOut); + + +/** +*************************************************************************************************** +* AddrSwizzleGenOption +* +* @brief +* Which swizzle generating options: legacy or linear +*************************************************************************************************** +*/ +typedef enum _AddrSwizzleGenOption +{ + ADDR_SWIZZLE_GEN_DEFAULT = 0, ///< As is in client driver implemention for swizzle + ADDR_SWIZZLE_GEN_LINEAR = 1, ///< Using a linear increment of swizzle +} AddrSwizzleGenOption; + +/** +*************************************************************************************************** +* AddrSwizzleOption +* +* @brief +* Controls how swizzle is generated +*************************************************************************************************** +*/ +typedef union _ADDR_SWIZZLE_OPTION +{ + struct + { + UINT_32 genOption : 1; ///< The way swizzle is generated, see AddrSwizzleGenOption + UINT_32 reduceBankBit : 1; ///< TRUE if we need reduce swizzle bits + UINT_32 reserved :30; ///< Reserved bits + }; + + UINT_32 value; + +} ADDR_SWIZZLE_OPTION; + +/** +*************************************************************************************************** +* ADDR_COMPUTE_BASE_SWIZZLE_INPUT +* +* @brief +* Input structure of AddrComputeBaseSwizzle +*************************************************************************************************** +*/ +typedef struct _ADDR_COMPUTE_BASE_SWIZZLE_INPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + ADDR_SWIZZLE_OPTION option; ///< Swizzle option + UINT_32 surfIndex; ///< Index of this surface type + AddrTileMode tileMode; ///< Tile Mode + + /// r800 and later HWL parameters + ADDR_TILEINFO* pTileInfo; ///< 2D tile parameters. Actually banks needed here! + + INT_32 tileIndex; ///< Tile index, MUST be -1 if you don't want to use it + /// while the global useTileIndex is set to 1 + INT_32 macroModeIndex; ///< Index in macro tile mode table if there is one (CI) + ///< README: When tileIndex is not -1, this must be valid +} ADDR_COMPUTE_BASE_SWIZZLE_INPUT; + +/** +*************************************************************************************************** +* ADDR_COMPUTE_BASE_SWIZZLE_OUTPUT +* +* @brief +* Output structure of AddrComputeBaseSwizzle +*************************************************************************************************** +*/ +typedef struct _ADDR_COMPUTE_BASE_SWIZZLE_OUTPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + UINT_32 tileSwizzle; ///< Combined swizzle +} ADDR_COMPUTE_BASE_SWIZZLE_OUTPUT; + +/** +*************************************************************************************************** +* AddrComputeBaseSwizzle +* +* @brief +* Return a Combined Bank and Pipe swizzle base on surface based on surface type/index +* @return +* ADDR_OK if no error +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrComputeBaseSwizzle( + ADDR_HANDLE hLib, + const ADDR_COMPUTE_BASE_SWIZZLE_INPUT* pIn, + ADDR_COMPUTE_BASE_SWIZZLE_OUTPUT* pOut); + + + +/** +*************************************************************************************************** +* ELEM_GETEXPORTNORM_INPUT +* +* @brief +* Input structure for ElemGetExportNorm +* +*************************************************************************************************** +*/ +typedef struct _ELEM_GETEXPORTNORM_INPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + AddrColorFormat format; ///< Color buffer format; Client should use ColorFormat + AddrSurfaceNumber num; ///< Surface number type; Client should use NumberType + AddrSurfaceSwap swap; ///< Surface swap byte swap; Client should use SurfaceSwap + UINT_32 numSamples; ///< Number of samples +} ELEM_GETEXPORTNORM_INPUT; + +/** +*************************************************************************************************** +* ElemGetExportNorm +* +* @brief +* Helper function to check one format can be EXPORT_NUM, which is a register +* CB_COLOR_INFO.SURFACE_FORMAT. FP16 can be reported as EXPORT_NORM for rv770 in r600 +* family +* @note +* The implementation is only for r600. +* 00 - EXPORT_FULL: PS exports are 4 pixels with 4 components with 32-bits-per-component. (two +* clocks per export) +* 01 - EXPORT_NORM: PS exports are 4 pixels with 4 components with 16-bits-per-component. (one +* clock per export) +* +*************************************************************************************************** +*/ +BOOL_32 ADDR_API ElemGetExportNorm( + ADDR_HANDLE hLib, + const ELEM_GETEXPORTNORM_INPUT* pIn); + + + +/** +*************************************************************************************************** +* ELEM_FLT32TODEPTHPIXEL_INPUT +* +* @brief +* Input structure for addrFlt32ToDepthPixel +* +*************************************************************************************************** +*/ +typedef struct _ELEM_FLT32TODEPTHPIXEL_INPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + AddrDepthFormat format; ///< Depth buffer format + ADDR_FLT_32 comps[2]; ///< Component values (Z/stencil) +} ELEM_FLT32TODEPTHPIXEL_INPUT; + +/** +*************************************************************************************************** +* ELEM_FLT32TODEPTHPIXEL_INPUT +* +* @brief +* Output structure for ElemFlt32ToDepthPixel +* +*************************************************************************************************** +*/ +typedef struct _ELEM_FLT32TODEPTHPIXEL_OUTPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + UINT_8* pPixel; ///< Real depth value. Same data type as depth buffer. + /// Client must provide enough storage for this type. + UINT_32 depthBase; ///< Tile base in bits for depth bits + UINT_32 stencilBase; ///< Tile base in bits for stencil bits + UINT_32 depthBits; ///< Bits for depth + UINT_32 stencilBits; ///< Bits for stencil +} ELEM_FLT32TODEPTHPIXEL_OUTPUT; + +/** +*************************************************************************************************** +* ElemFlt32ToDepthPixel +* +* @brief +* Convert a FLT_32 value to a depth/stencil pixel value +* +* @return +* Return code +* +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API ElemFlt32ToDepthPixel( + ADDR_HANDLE hLib, + const ELEM_FLT32TODEPTHPIXEL_INPUT* pIn, + ELEM_FLT32TODEPTHPIXEL_OUTPUT* pOut); + + + +/** +*************************************************************************************************** +* ELEM_FLT32TOCOLORPIXEL_INPUT +* +* @brief +* Input structure for addrFlt32ToColorPixel +* +*************************************************************************************************** +*/ +typedef struct _ELEM_FLT32TOCOLORPIXEL_INPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + AddrColorFormat format; ///< Color buffer format + AddrSurfaceNumber surfNum; ///< Surface number + AddrSurfaceSwap surfSwap; ///< Surface swap + ADDR_FLT_32 comps[4]; ///< Component values (r/g/b/a) +} ELEM_FLT32TOCOLORPIXEL_INPUT; + +/** +*************************************************************************************************** +* ELEM_FLT32TOCOLORPIXEL_INPUT +* +* @brief +* Output structure for ElemFlt32ToColorPixel +* +*************************************************************************************************** +*/ +typedef struct _ELEM_FLT32TOCOLORPIXEL_OUTPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + UINT_8* pPixel; ///< Real color value. Same data type as color buffer. + /// Client must provide enough storage for this type. +} ELEM_FLT32TOCOLORPIXEL_OUTPUT; + +/** +*************************************************************************************************** +* ElemFlt32ToColorPixel +* +* @brief +* Convert a FLT_32 value to a red/green/blue/alpha pixel value +* +* @return +* Return code +* +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API ElemFlt32ToColorPixel( + ADDR_HANDLE hLib, + const ELEM_FLT32TOCOLORPIXEL_INPUT* pIn, + ELEM_FLT32TOCOLORPIXEL_OUTPUT* pOut); + + +/** +*************************************************************************************************** +* ADDR_CONVERT_TILEINFOTOHW_INPUT +* +* @brief +* Input structure for AddrConvertTileInfoToHW +* @note +* When reverse is TRUE, indices are igonred +*************************************************************************************************** +*/ +typedef struct _ADDR_CONVERT_TILEINFOTOHW_INPUT +{ + UINT_32 size; ///< Size of this structure in bytes + BOOL_32 reverse; ///< Convert control flag. + /// FALSE: convert from real value to HW value; + /// TRUE: convert from HW value to real value. + + /// r800 and later HWL parameters + ADDR_TILEINFO* pTileInfo; ///< Tile parameters with real value + + INT_32 tileIndex; ///< Tile index, MUST be -1 if you don't want to use it + /// while the global useTileIndex is set to 1 + INT_32 macroModeIndex; ///< Index in macro tile mode table if there is one (CI) + ///< README: When tileIndex is not -1, this must be valid +} ADDR_CONVERT_TILEINFOTOHW_INPUT; + +/** +*************************************************************************************************** +* ADDR_CONVERT_TILEINFOTOHW_OUTPUT +* +* @brief +* Output structure for AddrConvertTileInfoToHW +*************************************************************************************************** +*/ +typedef struct _ADDR_CONVERT_TILEINFOTOHW_OUTPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + /// r800 and later HWL parameters + ADDR_TILEINFO* pTileInfo; ///< Tile parameters with hardware register value + +} ADDR_CONVERT_TILEINFOTOHW_OUTPUT; + +/** +*************************************************************************************************** +* AddrConvertTileInfoToHW +* +* @brief +* Convert tile info from real value to hardware register value +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrConvertTileInfoToHW( + ADDR_HANDLE hLib, + const ADDR_CONVERT_TILEINFOTOHW_INPUT* pIn, + ADDR_CONVERT_TILEINFOTOHW_OUTPUT* pOut); + + + +/** +*************************************************************************************************** +* ADDR_CONVERT_TILEINDEX_INPUT +* +* @brief +* Input structure for AddrConvertTileIndex +*************************************************************************************************** +*/ +typedef struct _ADDR_CONVERT_TILEINDEX_INPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + INT_32 tileIndex; ///< Tile index + INT_32 macroModeIndex; ///< Index in macro tile mode table if there is one (CI) + BOOL_32 tileInfoHw; ///< Set to TRUE if client wants HW enum, otherwise actual +} ADDR_CONVERT_TILEINDEX_INPUT; + +/** +*************************************************************************************************** +* ADDR_CONVERT_TILEINDEX_OUTPUT +* +* @brief +* Output structure for AddrConvertTileIndex +*************************************************************************************************** +*/ +typedef struct _ADDR_CONVERT_TILEINDEX_OUTPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + AddrTileMode tileMode; ///< Tile mode + AddrTileType tileType; ///< Tile type + ADDR_TILEINFO* pTileInfo; ///< Tile info + +} ADDR_CONVERT_TILEINDEX_OUTPUT; + +/** +*************************************************************************************************** +* AddrConvertTileIndex +* +* @brief +* Convert tile index to tile mode/type/info +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrConvertTileIndex( + ADDR_HANDLE hLib, + const ADDR_CONVERT_TILEINDEX_INPUT* pIn, + ADDR_CONVERT_TILEINDEX_OUTPUT* pOut); + + + +/** +*************************************************************************************************** +* ADDR_CONVERT_TILEINDEX1_INPUT +* +* @brief +* Input structure for AddrConvertTileIndex1 (without macro mode index) +*************************************************************************************************** +*/ +typedef struct _ADDR_CONVERT_TILEINDEX1_INPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + INT_32 tileIndex; ///< Tile index + UINT_32 bpp; ///< Bits per pixel + UINT_32 numSamples; ///< Number of samples + BOOL_32 tileInfoHw; ///< Set to TRUE if client wants HW enum, otherwise actual +} ADDR_CONVERT_TILEINDEX1_INPUT; + +/** +*************************************************************************************************** +* AddrConvertTileIndex1 +* +* @brief +* Convert tile index to tile mode/type/info +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrConvertTileIndex1( + ADDR_HANDLE hLib, + const ADDR_CONVERT_TILEINDEX1_INPUT* pIn, + ADDR_CONVERT_TILEINDEX_OUTPUT* pOut); + + + +/** +*************************************************************************************************** +* ADDR_GET_TILEINDEX_INPUT +* +* @brief +* Input structure for AddrGetTileIndex +*************************************************************************************************** +*/ +typedef struct _ADDR_GET_TILEINDEX_INPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + AddrTileMode tileMode; ///< Tile mode + AddrTileType tileType; ///< Tile-type: disp/non-disp/... + ADDR_TILEINFO* pTileInfo; ///< Pointer to tile-info structure, can be NULL for linear/1D +} ADDR_GET_TILEINDEX_INPUT; + +/** +*************************************************************************************************** +* ADDR_GET_TILEINDEX_OUTPUT +* +* @brief +* Output structure for AddrGetTileIndex +*************************************************************************************************** +*/ +typedef struct _ADDR_GET_TILEINDEX_OUTPUT +{ + UINT_32 size; ///< Size of this structure in bytes + + INT_32 index; ///< index in table +} ADDR_GET_TILEINDEX_OUTPUT; + +/** +*************************************************************************************************** +* AddrGetTileIndex +* +* @brief +* Get the tiling mode index in table +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrGetTileIndex( + ADDR_HANDLE hLib, + const ADDR_GET_TILEINDEX_INPUT* pIn, + ADDR_GET_TILEINDEX_OUTPUT* pOut); + + + + +/** +*************************************************************************************************** +* ADDR_PRT_INFO_INPUT +* +* @brief +* Input structure for AddrComputePrtInfo +*************************************************************************************************** +*/ +typedef struct _ADDR_PRT_INFO_INPUT +{ + AddrFormat format; ///< Surface format + UINT_32 baseMipWidth; ///< Base mipmap width + UINT_32 baseMipHeight; ///< Base mipmap height + UINT_32 baseMipDepth; ///< Base mipmap depth + UINT_32 numFrags; ///< Number of fragments, +} ADDR_PRT_INFO_INPUT; + +/** +*************************************************************************************************** +* ADDR_PRT_INFO_OUTPUT +* +* @brief +* Input structure for AddrComputePrtInfo +*************************************************************************************************** +*/ +typedef struct _ADDR_PRT_INFO_OUTPUT +{ + UINT_32 prtTileWidth; + UINT_32 prtTileHeight; +} ADDR_PRT_INFO_OUTPUT; + +/** +*************************************************************************************************** +* AddrComputePrtInfo +* +* @brief +* Compute prt surface related information +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrComputePrtInfo( + ADDR_HANDLE hLib, + const ADDR_PRT_INFO_INPUT* pIn, + ADDR_PRT_INFO_OUTPUT* pOut); + +/////////////////////////////////////////////////////////////////////////////////////////////////// +// DCC key functions +/////////////////////////////////////////////////////////////////////////////////////////////////// + +/** +*************************************************************************************************** +* _ADDR_COMPUTE_DCCINFO_INPUT +* +* @brief +* Input structure of AddrComputeDccInfo +*************************************************************************************************** +*/ +typedef struct _ADDR_COMPUTE_DCCINFO_INPUT +{ + UINT_32 size; ///< Size of this structure in bytes + UINT_32 bpp; ///< BitPP of color surface + UINT_32 numSamples; ///< Sample number of color surface + UINT_64 colorSurfSize; ///< Size of color surface to which dcc key is bound + AddrTileMode tileMode; ///< Tile mode of color surface + ADDR_TILEINFO tileInfo; ///< Tile info of color surface + UINT_32 tileSwizzle; ///< Tile swizzle + INT_32 tileIndex; ///< Tile index of color surface, + ///< MUST be -1 if you don't want to use it + ///< while the global useTileIndex is set to 1 + INT_32 macroModeIndex; ///< Index in macro tile mode table if there is one (CI) + ///< README: When tileIndex is not -1, this must be valid +} ADDR_COMPUTE_DCCINFO_INPUT; + +/** +*************************************************************************************************** +* ADDR_COMPUTE_DCCINFO_OUTPUT +* +* @brief +* Output structure of AddrComputeDccInfo +*************************************************************************************************** +*/ +typedef struct _ADDR_COMPUTE_DCCINFO_OUTPUT +{ + UINT_32 size; ///< Size of this structure in bytes + UINT_64 dccRamBaseAlign; ///< Base alignment of dcc key + UINT_64 dccRamSize; ///< Size of dcc key + UINT_64 dccFastClearSize; ///< Size of dcc key portion that can be fast cleared + BOOL_32 subLvlCompressible; ///< whether sub resource is compressiable +} ADDR_COMPUTE_DCCINFO_OUTPUT; + +/** +*************************************************************************************************** +* AddrComputeDccInfo +* +* @brief +* Compute DCC key size, base alignment +* info +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE ADDR_API AddrComputeDccInfo( + ADDR_HANDLE hLib, + const ADDR_COMPUTE_DCCINFO_INPUT* pIn, + ADDR_COMPUTE_DCCINFO_OUTPUT* pOut); + +#if defined(__cplusplus) +} +#endif + +#endif // __ADDR_INTERFACE_H__ + + diff --git a/src/gallium/winsys/radeon/amdgpu/addrlib/addrtypes.h b/src/gallium/winsys/radeon/amdgpu/addrlib/addrtypes.h new file mode 100644 index 0000000..4c68ac5 --- /dev/null +++ b/src/gallium/winsys/radeon/amdgpu/addrlib/addrtypes.h @@ -0,0 +1,590 @@ +/* + * Copyright © 2014 Advanced Micro Devices, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES + * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NON-INFRINGEMENT. IN NO EVENT SHALL THE COPYRIGHT HOLDERS, AUTHORS + * AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + */ + +/** +*************************************************************************************************** +* @file addrtypes.h +* @brief Contains the helper function and constants +*************************************************************************************************** +*/ +#ifndef __ADDR_TYPES_H__ +#define __ADDR_TYPES_H__ + +#if defined(__APPLE__) || defined(TCORE_BUILD) +// External definitions header maintained by Mac driver team (and TCORE team) +// Helps address compilation issues & reduces code covered by NDA +#include "addrExtDef.h" + +#else + +// Windows and/or Linux +#if !defined(VOID) +typedef void VOID; +#endif + +#if !defined(FLOAT) +typedef float FLOAT; +#endif + +#if !defined(CHAR) +typedef char CHAR; +#endif + +#if !defined(INT) +typedef int INT; +#endif + +#include // va_list...etc need this header + +#endif // defined (__APPLE__) + +/** +*************************************************************************************************** +* Calling conventions +*************************************************************************************************** +*/ +#ifndef ADDR_CDECL + #if defined(__GNUC__) + #define ADDR_CDECL __attribute__((cdecl)) + #else + #define ADDR_CDECL __cdecl + #endif +#endif + +#ifndef ADDR_STDCALL + #if defined(__GNUC__) + #if defined(__AMD64__) + #define ADDR_STDCALL + #else + #define ADDR_STDCALL __attribute__((stdcall)) + #endif + #else + #define ADDR_STDCALL __stdcall + #endif +#endif + +#ifndef ADDR_FASTCALL + #if defined(__GNUC__) + #define ADDR_FASTCALL __attribute__((regparm(0))) + #else + #define ADDR_FASTCALL __fastcall + #endif +#endif + +#ifndef GC_CDECL + #define GC_CDECL ADDR_CDECL +#endif + +#ifndef GC_STDCALL + #define GC_STDCALL ADDR_STDCALL +#endif + +#ifndef GC_FASTCALL + #define GC_FASTCALL ADDR_FASTCALL +#endif + + +#if defined(__GNUC__) + #define ADDR_INLINE static inline // inline needs to be static to link +#else + // win32, win64, other platforms + #define ADDR_INLINE __inline +#endif // #if defined(__GNUC__) + +#define ADDR_API ADDR_FASTCALL //default call convention is fast call + +/** +*************************************************************************************************** +* Global defines used by other modules +*************************************************************************************************** +*/ +#if !defined(TILEINDEX_INVALID) +#define TILEINDEX_INVALID -1 +#endif + +#if !defined(TILEINDEX_LINEAR_GENERAL) +#define TILEINDEX_LINEAR_GENERAL -2 +#endif + +#if !defined(TILEINDEX_LINEAR_ALIGNED) +#define TILEINDEX_LINEAR_ALIGNED 8 +#endif + +/** +*************************************************************************************************** +* Return codes +*************************************************************************************************** +*/ +typedef enum _ADDR_E_RETURNCODE +{ + // General Return + ADDR_OK = 0, + ADDR_ERROR = 1, + + // Specific Errors + ADDR_OUTOFMEMORY, + ADDR_INVALIDPARAMS, + ADDR_NOTSUPPORTED, + ADDR_NOTIMPLEMENTED, + ADDR_PARAMSIZEMISMATCH, + ADDR_INVALIDGBREGVALUES, + +} ADDR_E_RETURNCODE; + +/** +*************************************************************************************************** +* @brief +* Neutral enums that define tile modes for all H/W +* @note +* R600/R800 tiling mode can be cast to hw enums directly but never cast into HW enum from +* ADDR_TM_2D_TILED_XTHICK +* +*************************************************************************************************** +*/ +typedef enum _AddrTileMode +{ + ADDR_TM_LINEAR_GENERAL = 0, ///< Least restrictions, pitch: multiple of 8 if not buffer + ADDR_TM_LINEAR_ALIGNED = 1, ///< Requests pitch or slice to be multiple of 64 pixels + ADDR_TM_1D_TILED_THIN1 = 2, ///< Linear array of 8x8 tiles + ADDR_TM_1D_TILED_THICK = 3, ///< Linear array of 8x8x4 tiles + ADDR_TM_2D_TILED_THIN1 = 4, ///< A set of macro tiles consist of 8x8 tiles + ADDR_TM_2D_TILED_THIN2 = 5, ///< 600 HWL only, macro tile ratio is 1:4 + ADDR_TM_2D_TILED_THIN4 = 6, ///< 600 HWL only, macro tile ratio is 1:16 + ADDR_TM_2D_TILED_THICK = 7, ///< A set of macro tiles consist of 8x8x4 tiles + ADDR_TM_2B_TILED_THIN1 = 8, ///< 600 HWL only, with bank swap + ADDR_TM_2B_TILED_THIN2 = 9, ///< 600 HWL only, with bank swap and ratio is 1:4 + ADDR_TM_2B_TILED_THIN4 = 10, ///< 600 HWL only, with bank swap and ratio is 1:16 + ADDR_TM_2B_TILED_THICK = 11, ///< 600 HWL only, with bank swap, consists of 8x8x4 tiles + ADDR_TM_3D_TILED_THIN1 = 12, ///< Macro tiling w/ pipe rotation between slices + ADDR_TM_3D_TILED_THICK = 13, ///< Macro tiling w/ pipe rotation bwtween slices, thick + ADDR_TM_3B_TILED_THIN1 = 14, ///< 600 HWL only, with bank swap + ADDR_TM_3B_TILED_THICK = 15, ///< 600 HWL only, with bank swap, thick + ADDR_TM_2D_TILED_XTHICK = 16, ///< Tile is 8x8x8, valid from NI + ADDR_TM_3D_TILED_XTHICK = 17, ///< Tile is 8x8x8, valid from NI + ADDR_TM_POWER_SAVE = 18, ///< Power save mode, only used by KMD on NI + ADDR_TM_PRT_TILED_THIN1 = 19, ///< No bank/pipe rotation or hashing beyond macrotile size + ADDR_TM_PRT_2D_TILED_THIN1 = 20, ///< Same as 2D_TILED_THIN1, PRT only + ADDR_TM_PRT_3D_TILED_THIN1 = 21, ///< Same as 3D_TILED_THIN1, PRT only + ADDR_TM_PRT_TILED_THICK = 22, ///< No bank/pipe rotation or hashing beyond macrotile size + ADDR_TM_PRT_2D_TILED_THICK = 23, ///< Same as 2D_TILED_THICK, PRT only + ADDR_TM_PRT_3D_TILED_THICK = 24, ///< Same as 3D_TILED_THICK, PRT only + ADDR_TM_COUNT = 25, ///< Must be the value of the last tile mode +} AddrTileMode; + +/** +*************************************************************************************************** +* AddrFormat +* +* @brief +* Neutral enum for SurfaceFormat +* +*************************************************************************************************** +*/ +typedef enum _AddrFormat { + ADDR_FMT_INVALID = 0x00000000, + ADDR_FMT_8 = 0x00000001, + ADDR_FMT_4_4 = 0x00000002, + ADDR_FMT_3_3_2 = 0x00000003, + ADDR_FMT_RESERVED_4 = 0x00000004, + ADDR_FMT_16 = 0x00000005, + ADDR_FMT_16_FLOAT = 0x00000006, + ADDR_FMT_8_8 = 0x00000007, + ADDR_FMT_5_6_5 = 0x00000008, + ADDR_FMT_6_5_5 = 0x00000009, + ADDR_FMT_1_5_5_5 = 0x0000000a, + ADDR_FMT_4_4_4_4 = 0x0000000b, + ADDR_FMT_5_5_5_1 = 0x0000000c, + ADDR_FMT_32 = 0x0000000d, + ADDR_FMT_32_FLOAT = 0x0000000e, + ADDR_FMT_16_16 = 0x0000000f, + ADDR_FMT_16_16_FLOAT = 0x00000010, + ADDR_FMT_8_24 = 0x00000011, + ADDR_FMT_8_24_FLOAT = 0x00000012, + ADDR_FMT_24_8 = 0x00000013, + ADDR_FMT_24_8_FLOAT = 0x00000014, + ADDR_FMT_10_11_11 = 0x00000015, + ADDR_FMT_10_11_11_FLOAT = 0x00000016, + ADDR_FMT_11_11_10 = 0x00000017, + ADDR_FMT_11_11_10_FLOAT = 0x00000018, + ADDR_FMT_2_10_10_10 = 0x00000019, + ADDR_FMT_8_8_8_8 = 0x0000001a, + ADDR_FMT_10_10_10_2 = 0x0000001b, + ADDR_FMT_X24_8_32_FLOAT = 0x0000001c, + ADDR_FMT_32_32 = 0x0000001d, + ADDR_FMT_32_32_FLOAT = 0x0000001e, + ADDR_FMT_16_16_16_16 = 0x0000001f, + ADDR_FMT_16_16_16_16_FLOAT = 0x00000020, + ADDR_FMT_RESERVED_33 = 0x00000021, + ADDR_FMT_32_32_32_32 = 0x00000022, + ADDR_FMT_32_32_32_32_FLOAT = 0x00000023, + ADDR_FMT_RESERVED_36 = 0x00000024, + ADDR_FMT_1 = 0x00000025, + ADDR_FMT_1_REVERSED = 0x00000026, + ADDR_FMT_GB_GR = 0x00000027, + ADDR_FMT_BG_RG = 0x00000028, + ADDR_FMT_32_AS_8 = 0x00000029, + ADDR_FMT_32_AS_8_8 = 0x0000002a, + ADDR_FMT_5_9_9_9_SHAREDEXP = 0x0000002b, + ADDR_FMT_8_8_8 = 0x0000002c, + ADDR_FMT_16_16_16 = 0x0000002d, + ADDR_FMT_16_16_16_FLOAT = 0x0000002e, + ADDR_FMT_32_32_32 = 0x0000002f, + ADDR_FMT_32_32_32_FLOAT = 0x00000030, + ADDR_FMT_BC1 = 0x00000031, + ADDR_FMT_BC2 = 0x00000032, + ADDR_FMT_BC3 = 0x00000033, + ADDR_FMT_BC4 = 0x00000034, + ADDR_FMT_BC5 = 0x00000035, + ADDR_FMT_BC6 = 0x00000036, + ADDR_FMT_BC7 = 0x00000037, + ADDR_FMT_32_AS_32_32_32_32 = 0x00000038, + ADDR_FMT_APC3 = 0x00000039, + ADDR_FMT_APC4 = 0x0000003a, + ADDR_FMT_APC5 = 0x0000003b, + ADDR_FMT_APC6 = 0x0000003c, + ADDR_FMT_APC7 = 0x0000003d, + ADDR_FMT_CTX1 = 0x0000003e, + ADDR_FMT_RESERVED_63 = 0x0000003f, +} AddrFormat; + +/** +*************************************************************************************************** +* AddrDepthFormat +* +* @brief +* Neutral enum for addrFlt32ToDepthPixel +* +*************************************************************************************************** +*/ +typedef enum _AddrDepthFormat +{ + ADDR_DEPTH_INVALID = 0x00000000, + ADDR_DEPTH_16 = 0x00000001, + ADDR_DEPTH_X8_24 = 0x00000002, + ADDR_DEPTH_8_24 = 0x00000003, + ADDR_DEPTH_X8_24_FLOAT = 0x00000004, + ADDR_DEPTH_8_24_FLOAT = 0x00000005, + ADDR_DEPTH_32_FLOAT = 0x00000006, + ADDR_DEPTH_X24_8_32_FLOAT = 0x00000007, + +} AddrDepthFormat; + +/** +*************************************************************************************************** +* AddrColorFormat +* +* @brief +* Neutral enum for ColorFormat +* +*************************************************************************************************** +*/ +typedef enum _AddrColorFormat +{ + ADDR_COLOR_INVALID = 0x00000000, + ADDR_COLOR_8 = 0x00000001, + ADDR_COLOR_4_4 = 0x00000002, + ADDR_COLOR_3_3_2 = 0x00000003, + ADDR_COLOR_RESERVED_4 = 0x00000004, + ADDR_COLOR_16 = 0x00000005, + ADDR_COLOR_16_FLOAT = 0x00000006, + ADDR_COLOR_8_8 = 0x00000007, + ADDR_COLOR_5_6_5 = 0x00000008, + ADDR_COLOR_6_5_5 = 0x00000009, + ADDR_COLOR_1_5_5_5 = 0x0000000a, + ADDR_COLOR_4_4_4_4 = 0x0000000b, + ADDR_COLOR_5_5_5_1 = 0x0000000c, + ADDR_COLOR_32 = 0x0000000d, + ADDR_COLOR_32_FLOAT = 0x0000000e, + ADDR_COLOR_16_16 = 0x0000000f, + ADDR_COLOR_16_16_FLOAT = 0x00000010, + ADDR_COLOR_8_24 = 0x00000011, + ADDR_COLOR_8_24_FLOAT = 0x00000012, + ADDR_COLOR_24_8 = 0x00000013, + ADDR_COLOR_24_8_FLOAT = 0x00000014, + ADDR_COLOR_10_11_11 = 0x00000015, + ADDR_COLOR_10_11_11_FLOAT = 0x00000016, + ADDR_COLOR_11_11_10 = 0x00000017, + ADDR_COLOR_11_11_10_FLOAT = 0x00000018, + ADDR_COLOR_2_10_10_10 = 0x00000019, + ADDR_COLOR_8_8_8_8 = 0x0000001a, + ADDR_COLOR_10_10_10_2 = 0x0000001b, + ADDR_COLOR_X24_8_32_FLOAT = 0x0000001c, + ADDR_COLOR_32_32 = 0x0000001d, + ADDR_COLOR_32_32_FLOAT = 0x0000001e, + ADDR_COLOR_16_16_16_16 = 0x0000001f, + ADDR_COLOR_16_16_16_16_FLOAT = 0x00000020, + ADDR_COLOR_RESERVED_33 = 0x00000021, + ADDR_COLOR_32_32_32_32 = 0x00000022, + ADDR_COLOR_32_32_32_32_FLOAT = 0x00000023, +} AddrColorFormat; + +/** +*************************************************************************************************** +* AddrSurfaceNumber +* +* @brief +* Neutral enum for SurfaceNumber +* +*************************************************************************************************** +*/ +typedef enum _AddrSurfaceNumber { + ADDR_NUMBER_UNORM = 0x00000000, + ADDR_NUMBER_SNORM = 0x00000001, + ADDR_NUMBER_USCALED = 0x00000002, + ADDR_NUMBER_SSCALED = 0x00000003, + ADDR_NUMBER_UINT = 0x00000004, + ADDR_NUMBER_SINT = 0x00000005, + ADDR_NUMBER_SRGB = 0x00000006, + ADDR_NUMBER_FLOAT = 0x00000007, +} AddrSurfaceNumber; + +/** +*************************************************************************************************** +* AddrSurfaceSwap +* +* @brief +* Neutral enum for SurfaceSwap +* +*************************************************************************************************** +*/ +typedef enum _AddrSurfaceSwap { + ADDR_SWAP_STD = 0x00000000, + ADDR_SWAP_ALT = 0x00000001, + ADDR_SWAP_STD_REV = 0x00000002, + ADDR_SWAP_ALT_REV = 0x00000003, +} AddrSurfaceSwap; + +/** +*************************************************************************************************** +* AddrHtileBlockSize +* +* @brief +* Size of HTILE blocks, valid values are 4 or 8 for now +*************************************************************************************************** +*/ +typedef enum _AddrHtileBlockSize +{ + ADDR_HTILE_BLOCKSIZE_4 = 4, + ADDR_HTILE_BLOCKSIZE_8 = 8, +} AddrHtileBlockSize; + + +/** +*************************************************************************************************** +* AddrPipeCfg +* +* @brief +* The pipe configuration field specifies both the number of pipes and +* how pipes are interleaved on the surface. +* The expression of number of pipes, the shader engine tile size, and packer tile size +* is encoded in a PIPE_CONFIG register field. +* In general the number of pipes usually matches the number of memory channels of the +* hardware configuration. +* For hw configurations w/ non-pow2 memory number of memory channels, it usually matches +* the number of ROP units(? TODO: which registers??) +* The enum value = hw enum + 1 which is to reserve 0 for requesting default. +*************************************************************************************************** +*/ +typedef enum _AddrPipeCfg +{ + ADDR_PIPECFG_INVALID = 0, + ADDR_PIPECFG_P2 = 1, /// 2 pipes, + ADDR_PIPECFG_P4_8x16 = 5, /// 4 pipes, + ADDR_PIPECFG_P4_16x16 = 6, + ADDR_PIPECFG_P4_16x32 = 7, + ADDR_PIPECFG_P4_32x32 = 8, + ADDR_PIPECFG_P8_16x16_8x16 = 9, /// 8 pipes + ADDR_PIPECFG_P8_16x32_8x16 = 10, + ADDR_PIPECFG_P8_32x32_8x16 = 11, + ADDR_PIPECFG_P8_16x32_16x16 = 12, + ADDR_PIPECFG_P8_32x32_16x16 = 13, + ADDR_PIPECFG_P8_32x32_16x32 = 14, + ADDR_PIPECFG_P8_32x64_32x32 = 15, + ADDR_PIPECFG_P16_32x32_8x16 = 17, /// 16 pipes + ADDR_PIPECFG_P16_32x32_16x16 = 18, + ADDR_PIPECFG_MAX = 19, +} AddrPipeCfg; + +/** +*************************************************************************************************** +* AddrTileType +* +* @brief +* Neutral enums that specifies micro tile type (MICRO_TILE_MODE) +*************************************************************************************************** +*/ +typedef enum _AddrTileType +{ + ADDR_DISPLAYABLE = 0, ///< Displayable tiling + ADDR_NON_DISPLAYABLE = 1, ///< Non-displayable tiling, a.k.a thin micro tiling + ADDR_DEPTH_SAMPLE_ORDER = 2, ///< Same as non-displayable plus depth-sample-order + ADDR_ROTATED = 3, ///< Rotated displayable tiling + ADDR_THICK = 4, ///< Thick micro-tiling, only valid for THICK and XTHICK +} AddrTileType; + +/////////////////////////////////////////////////////////////////////////////////////////////////// +// +// Type definitions: short system-independent names for address library types +// +/////////////////////////////////////////////////////////////////////////////////////////////////// + +#if !defined(__APPLE__) + +#ifndef BOOL_32 // no bool type in C +/// @brief Boolean type, since none is defined in C +/// @ingroup type +#define BOOL_32 int +#endif + +#ifndef INT_32 +#define INT_32 int +#endif + +#ifndef UINT_32 +#define UINT_32 unsigned int +#endif + +#ifndef INT_16 +#define INT_16 short +#endif + +#ifndef UINT_16 +#define UINT_16 unsigned short +#endif + +#ifndef INT_8 +#define INT_8 char +#endif + +#ifndef UINT_8 +#define UINT_8 unsigned char +#endif + +#ifndef NULL +#define NULL 0 +#endif + +#ifndef TRUE +#define TRUE 1 +#endif + +#ifndef FALSE +#define FALSE 0 +#endif + +// +// 64-bit integer types depend on the compiler +// +#if defined( __GNUC__ ) || defined( __WATCOMC__ ) +#define INT_64 long long +#define UINT_64 unsigned long long + +#elif defined( _WIN32 ) +#define INT_64 __int64 +#define UINT_64 unsigned __int64 + +#else +#error Unsupported compiler and/or operating system for 64-bit integers + +/// @brief 64-bit signed integer type (compiler dependent) +/// @ingroup type +/// +/// The addrlib defines a 64-bit signed integer type for either +/// Gnu/Watcom compilers (which use the first syntax) or for +/// the Windows VCC compiler (which uses the second syntax). +#define INT_64 long long OR __int64 + +/// @brief 64-bit unsigned integer type (compiler dependent) +/// @ingroup type +/// +/// The addrlib defines a 64-bit unsigned integer type for either +/// Gnu/Watcom compilers (which use the first syntax) or for +/// the Windows VCC compiler (which uses the second syntax). +/// +#define UINT_64 unsigned long long OR unsigned __int64 +#endif + +#endif // #if !defined(__APPLE__) + +// ADDR64X is used to print addresses in hex form on both Windows and Linux +// +#if defined( __GNUC__ ) || defined( __WATCOMC__ ) +#define ADDR64X "llx" +#define ADDR64D "lld" + +#elif defined( _WIN32 ) +#define ADDR64X "I64x" +#define ADDR64D "I64d" + +#else +#error Unsupported compiler and/or operating system for 64-bit integers + +/// @brief Addrlib device address 64-bit printf tag (compiler dependent) +/// @ingroup type +/// +/// This allows printf to display an ADDR_64 for either the Windows VCC compiler +/// (which used this value) or the Gnu/Watcom compilers (which use "llx". +/// An example of use is printf("addr 0x%"ADDR64X"\n", address); +/// +#define ADDR64X "llx" OR "I64x" +#define ADDR64D "lld" OR "I64d" +#endif + + +/// @brief Union for storing a 32-bit float or 32-bit integer +/// @ingroup type +/// +/// This union provides a simple way to convert between a 32-bit float +/// and a 32-bit integer. It also prevents the compiler from producing +/// code that alters NaN values when assiging or coying floats. +/// Therefore, all address library routines that pass or return 32-bit +/// floating point data do so by passing or returning a FLT_32. +/// +typedef union { + INT_32 i; + UINT_32 u; + float f; +} ADDR_FLT_32; + + +/////////////////////////////////////////////////////////////////////////////////////////////////// +// +// Macros for controlling linking and building on multiple systems +// +/////////////////////////////////////////////////////////////////////////////////////////////////// +#if defined(_MSC_VER) +#if defined(va_copy) +#undef va_copy //redefine va_copy to support VC2013 +#endif +#endif + +#if !defined(va_copy) +#define va_copy(dst, src) \ + ((void) memcpy(&(dst), &(src), sizeof(va_list))) +#endif + +#endif // __ADDR_TYPES_H__ + diff --git a/src/gallium/winsys/radeon/amdgpu/addrlib/core/addrcommon.h b/src/gallium/winsys/radeon/amdgpu/addrlib/core/addrcommon.h new file mode 100644 index 0000000..f996c9a --- /dev/null +++ b/src/gallium/winsys/radeon/amdgpu/addrlib/core/addrcommon.h @@ -0,0 +1,558 @@ +/* + * Copyright © 2014 Advanced Micro Devices, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES + * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NON-INFRINGEMENT. IN NO EVENT SHALL THE COPYRIGHT HOLDERS, AUTHORS + * AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + */ + +/** +*************************************************************************************************** +* @file addrcommon.h +* @brief Contains the helper function and constants +*************************************************************************************************** +*/ + +#ifndef __ADDR_COMMON_H__ +#define __ADDR_COMMON_H__ + +#include "addrinterface.h" + + +// ADDR_LNX_KERNEL_BUILD is for internal build +// Moved from addrinterface.h so __KERNEL__ is not needed any more +#if ADDR_LNX_KERNEL_BUILD // || (defined(__GNUC__) && defined(__KERNEL__)) + #include "lnx_common_defs.h" // ported from cmmqs +#elif !defined(__APPLE__) + #include + #include +#endif + +/////////////////////////////////////////////////////////////////////////////////////////////////// +// Common constants +/////////////////////////////////////////////////////////////////////////////////////////////////// +static const UINT_32 MicroTileWidth = 8; ///< Micro tile width, for 1D and 2D tiling +static const UINT_32 MicroTileHeight = 8; ///< Micro tile height, for 1D and 2D tiling +static const UINT_32 ThickTileThickness = 4; ///< Micro tile thickness, for THICK modes +static const UINT_32 XThickTileThickness = 8; ///< Extra thick tiling thickness +static const UINT_32 PowerSaveTileBytes = 64; ///< Nuber of bytes per tile for power save 64 +static const UINT_32 CmaskCacheBits = 1024; ///< Number of bits for CMASK cache +static const UINT_32 CmaskElemBits = 4; ///< Number of bits for CMASK element +static const UINT_32 HtileCacheBits = 16384; ///< Number of bits for HTILE cache 512*32 + +static const UINT_32 MicroTilePixels = MicroTileWidth * MicroTileHeight; + +static const INT_32 TileIndexInvalid = TILEINDEX_INVALID; +static const INT_32 TileIndexLinearGeneral = TILEINDEX_LINEAR_GENERAL; +static const INT_32 TileIndexNoMacroIndex = -3; + +/////////////////////////////////////////////////////////////////////////////////////////////////// +// Common macros +/////////////////////////////////////////////////////////////////////////////////////////////////// +#define BITS_PER_BYTE 8 +#define BITS_TO_BYTES(x) ( ((x) + (BITS_PER_BYTE-1)) / BITS_PER_BYTE ) +#define BYTES_TO_BITS(x) ( (x) * BITS_PER_BYTE ) + +/// Helper macros to select a single bit from an int (undefined later in section) +#define _BIT(v,b) (((v) >> (b) ) & 1) + +/** +*************************************************************************************************** +* @brief Enums to identify AddrLib type +*************************************************************************************************** +*/ +enum AddrLibClass +{ + BASE_ADDRLIB = 0x0, + R600_ADDRLIB = 0x6, + R800_ADDRLIB = 0x8, + SI_ADDRLIB = 0xa, + CI_ADDRLIB = 0xb, +}; + +/** +*************************************************************************************************** +* AddrChipFamily +* +* @brief +* Neutral enums that specifies chip family. +* +*************************************************************************************************** +*/ +enum AddrChipFamily +{ + ADDR_CHIP_FAMILY_IVLD, ///< Invalid family + ADDR_CHIP_FAMILY_R6XX, + ADDR_CHIP_FAMILY_R7XX, + ADDR_CHIP_FAMILY_R8XX, + ADDR_CHIP_FAMILY_NI, + ADDR_CHIP_FAMILY_SI, + ADDR_CHIP_FAMILY_CI, + ADDR_CHIP_FAMILY_VI, +}; + +/** +*************************************************************************************************** +* ADDR_CONFIG_FLAGS +* +* @brief +* This structure is used to set addr configuration flags. +*************************************************************************************************** +*/ +union ADDR_CONFIG_FLAGS +{ + struct + { + /// Clients do not need to set these flags except forceLinearAligned. + /// There flags are set up by AddrLib inside thru AddrInitGlobalParamsFromRegister + UINT_32 optimalBankSwap : 1; ///< New bank tiling for RV770 only + UINT_32 noCubeMipSlicesPad : 1; ///< Disables faces padding for cubemap mipmaps + UINT_32 fillSizeFields : 1; ///< If clients fill size fields in all input and + /// output structure + UINT_32 ignoreTileInfo : 1; ///< Don't use tile info structure + UINT_32 useTileIndex : 1; ///< Make tileIndex field in input valid + UINT_32 useCombinedSwizzle : 1; ///< Use combined swizzle + UINT_32 checkLast2DLevel : 1; ///< Check the last 2D mip sub level + UINT_32 useHtileSliceAlign : 1; ///< Do htile single slice alignment + UINT_32 degradeBaseLevel : 1; ///< Degrade to 1D modes automatically for base level + UINT_32 allowLargeThickTile : 1; ///< Allow 64*thickness*bytesPerPixel > rowSize + UINT_32 reserved : 22; ///< Reserved bits for future use + }; + + UINT_32 value; +}; + +/////////////////////////////////////////////////////////////////////////////////////////////////// +// Platform specific debug break defines +/////////////////////////////////////////////////////////////////////////////////////////////////// +#if DEBUG + #if defined(__GNUC__) + #define ADDR_DBG_BREAK() + #elif defined(__APPLE__) + #define ADDR_DBG_BREAK() { IOPanic("");} + #else + #define ADDR_DBG_BREAK() { __debugbreak(); } + #endif +#else + #define ADDR_DBG_BREAK() +#endif +/////////////////////////////////////////////////////////////////////////////////////////////////// + +/////////////////////////////////////////////////////////////////////////////////////////////////// +// Debug assertions used in AddrLib +/////////////////////////////////////////////////////////////////////////////////////////////////// +#if DEBUG +#define ADDR_ASSERT(__e) if ( !((__e) ? TRUE : FALSE)) { ADDR_DBG_BREAK(); } +#define ADDR_ASSERT_ALWAYS() ADDR_DBG_BREAK() +#define ADDR_UNHANDLED_CASE() ADDR_ASSERT(!"Unhandled case") +#define ADDR_NOT_IMPLEMENTED() ADDR_ASSERT(!"Not implemented"); +#else //DEBUG +#define ADDR_ASSERT(__e) +#define ADDR_ASSERT_ALWAYS() +#define ADDR_UNHANDLED_CASE() +#define ADDR_NOT_IMPLEMENTED() +#endif //DEBUG +/////////////////////////////////////////////////////////////////////////////////////////////////// + +/////////////////////////////////////////////////////////////////////////////////////////////////// +// Debug print macro from legacy address library +/////////////////////////////////////////////////////////////////////////////////////////////////// +#if DEBUG + +#define ADDR_PRNT(a) AddrObject::DebugPrint a + +/// @brief Macro for reporting informational messages +/// @ingroup util +/// +/// This macro optionally prints an informational message to stdout. +/// The first parameter is a condition -- if it is true, nothing is done. +/// The second pararmeter MUST be a parenthesis-enclosed list of arguments, +/// starting with a string. This is passed to printf() or an equivalent +/// in order to format the informational message. For example, +/// ADDR_INFO(0, ("test %d",3) ); prints out "test 3". +/// +#define ADDR_INFO(cond, a) \ +{ if (!(cond)) { ADDR_PRNT(a); } } + + +/// @brief Macro for reporting error warning messages +/// @ingroup util +/// +/// This macro optionally prints an error warning message to stdout, +/// followed by the file name and line number where the macro was called. +/// The first parameter is a condition -- if it is true, nothing is done. +/// The second pararmeter MUST be a parenthesis-enclosed list of arguments, +/// starting with a string. This is passed to printf() or an equivalent +/// in order to format the informational message. For example, +/// ADDR_WARN(0, ("test %d",3) ); prints out "test 3" followed by +/// a second line with the file name and line number. +/// +#define ADDR_WARN(cond, a) \ +{ if (!(cond)) \ + { ADDR_PRNT(a); \ + ADDR_PRNT((" WARNING in file %s, line %d\n", __FILE__, __LINE__)); \ +} } + + +/// @brief Macro for reporting fatal error conditions +/// @ingroup util +/// +/// This macro optionally stops execution of the current routine +/// after printing an error warning message to stdout, +/// followed by the file name and line number where the macro was called. +/// The first parameter is a condition -- if it is true, nothing is done. +/// The second pararmeter MUST be a parenthesis-enclosed list of arguments, +/// starting with a string. This is passed to printf() or an equivalent +/// in order to format the informational message. For example, +/// ADDR_EXIT(0, ("test %d",3) ); prints out "test 3" followed by +/// a second line with the file name and line number, then stops execution. +/// +#define ADDR_EXIT(cond, a) \ +{ if (!(cond)) \ + { ADDR_PRNT(a); ADDR_DBG_BREAK();\ +} } + +#else // DEBUG + +#define ADDRDPF 1 ? (void)0 : (void) + +#define ADDR_PRNT(a) + +#define ADDR_DBG_BREAK() + +#define ADDR_INFO(cond, a) + +#define ADDR_WARN(cond, a) + +#define ADDR_EXIT(cond, a) + +#endif // DEBUG +/////////////////////////////////////////////////////////////////////////////////////////////////// + +/////////////////////////////////////////////////////////////////////////////////////////////////// +// Misc helper functions +//////////////////////////////////////////////////////////////////////////////////////////////////// + +/** +*************************************************************************************************** +* AddrXorReduce +* +* @brief +* Xor the right-side numberOfBits bits of x. +*************************************************************************************************** +*/ +static inline UINT_32 XorReduce( + UINT_32 x, + UINT_32 numberOfBits) +{ + UINT_32 i; + UINT_32 result = x & 1; + + for (i=1; i>i) & 1); + } + + return result; +} + +/** +*************************************************************************************************** +* IsPow2 +* +* @brief +* Check if the size (UINT_32) is pow 2 +*************************************************************************************************** +*/ +static inline UINT_32 IsPow2( + UINT_32 dim) ///< [in] dimension of miplevel +{ + ADDR_ASSERT(dim > 0); + return !(dim & (dim - 1)); +} + +/** +*************************************************************************************************** +* IsPow2 +* +* @brief +* Check if the size (UINT_64) is pow 2 +*************************************************************************************************** +*/ +static inline UINT_64 IsPow2( + UINT_64 dim) ///< [in] dimension of miplevel +{ + ADDR_ASSERT(dim > 0); + return !(dim & (dim - 1)); +} + +/** +*************************************************************************************************** +* ByteAlign +* +* @brief +* Align UINT_32 "x" to "align" alignment, "align" should be power of 2 +*************************************************************************************************** +*/ +static inline UINT_32 PowTwoAlign( + UINT_32 x, + UINT_32 align) +{ + // + // Assert that x is a power of two. + // + ADDR_ASSERT(IsPow2(align)); + return (x + (align - 1)) & (~(align - 1)); +} + +/** +*************************************************************************************************** +* ByteAlign +* +* @brief +* Align UINT_64 "x" to "align" alignment, "align" should be power of 2 +*************************************************************************************************** +*/ +static inline UINT_64 PowTwoAlign( + UINT_64 x, + UINT_64 align) +{ + // + // Assert that x is a power of two. + // + ADDR_ASSERT(IsPow2(align)); + return (x + (align - 1)) & (~(align - 1)); +} + +/** +*************************************************************************************************** +* Min +* +* @brief +* Get the min value between two unsigned values +*************************************************************************************************** +*/ +static inline UINT_32 Min( + UINT_32 value1, + UINT_32 value2) +{ + return ((value1 < (value2)) ? (value1) : value2); +} + +/** +*************************************************************************************************** +* Min +* +* @brief +* Get the min value between two signed values +*************************************************************************************************** +*/ +static inline INT_32 Min( + INT_32 value1, + INT_32 value2) +{ + return ((value1 < (value2)) ? (value1) : value2); +} + +/** +*************************************************************************************************** +* Max +* +* @brief +* Get the max value between two unsigned values +*************************************************************************************************** +*/ +static inline UINT_32 Max( + UINT_32 value1, + UINT_32 value2) +{ + return ((value1 > (value2)) ? (value1) : value2); +} + +/** +*************************************************************************************************** +* Max +* +* @brief +* Get the max value between two signed values +*************************************************************************************************** +*/ +static inline INT_32 Max( + INT_32 value1, + INT_32 value2) +{ + return ((value1 > (value2)) ? (value1) : value2); +} + +/** +*************************************************************************************************** +* NextPow2 +* +* @brief +* Compute the mipmap's next level dim size +*************************************************************************************************** +*/ +static inline UINT_32 NextPow2( + UINT_32 dim) ///< [in] dimension of miplevel +{ + UINT_32 newDim; + + newDim = 1; + + if (dim > 0x7fffffff) + { + ADDR_ASSERT_ALWAYS(); + newDim = 0x80000000; + } + else + { + while (newDim < dim) + { + newDim <<= 1; + } + } + + return newDim; +} + +/** +*************************************************************************************************** +* Log2 +* +* @brief +* Compute log of base 2 +*************************************************************************************************** +*/ +static inline UINT_32 Log2( + UINT_32 x) ///< [in] the value should calculate log based 2 +{ + UINT_32 y; + + // + // Assert that x is a power of two. + // + ADDR_ASSERT(IsPow2(x)); + + y = 0; + while (x > 1) + { + x >>= 1; + y++; + } + + return y; +} + +/** +*************************************************************************************************** +* QLog2 +* +* @brief +* Compute log of base 2 quickly (<= 16) +*************************************************************************************************** +*/ +static inline UINT_32 QLog2( + UINT_32 x) ///< [in] the value should calculate log based 2 +{ + ADDR_ASSERT(x <= 16); + + UINT_32 y = 0; + + switch (x) + { + case 1: + y = 0; + break; + case 2: + y = 1; + break; + case 4: + y = 2; + break; + case 8: + y = 3; + break; + case 16: + y = 4; + break; + default: + ADDR_ASSERT_ALWAYS(); + } + + return y; +} + +/** +*************************************************************************************************** +* SafeAssign +* +* @brief +* NULL pointer safe assignment +*************************************************************************************************** +*/ +static inline VOID SafeAssign( + UINT_32* pLVal, ///< [in] Pointer to left val + UINT_32 rVal) ///< [in] Right value +{ + if (pLVal) + { + *pLVal = rVal; + } +} + +/** +*************************************************************************************************** +* SafeAssign +* +* @brief +* NULL pointer safe assignment for 64bit values +*************************************************************************************************** +*/ +static inline VOID SafeAssign( + UINT_64* pLVal, ///< [in] Pointer to left val + UINT_64 rVal) ///< [in] Right value +{ + if (pLVal) + { + *pLVal = rVal; + } +} + +/** +*************************************************************************************************** +* SafeAssign +* +* @brief +* NULL pointer safe assignment for AddrTileMode +*************************************************************************************************** +*/ +static inline VOID SafeAssign( + AddrTileMode* pLVal, ///< [in] Pointer to left val + AddrTileMode rVal) ///< [in] Right value +{ + if (pLVal) + { + *pLVal = rVal; + } +} + +#endif // __ADDR_COMMON_H__ + diff --git a/src/gallium/winsys/radeon/amdgpu/addrlib/core/addrelemlib.cpp b/src/gallium/winsys/radeon/amdgpu/addrlib/core/addrelemlib.cpp new file mode 100644 index 0000000..eb1b7de --- /dev/null +++ b/src/gallium/winsys/radeon/amdgpu/addrlib/core/addrelemlib.cpp @@ -0,0 +1,1678 @@ +/* + * Copyright © 2014 Advanced Micro Devices, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES + * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NON-INFRINGEMENT. IN NO EVENT SHALL THE COPYRIGHT HOLDERS, AUTHORS + * AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + */ + +/** +*************************************************************************************************** +* @file addrelemlib.cpp +* @brief Contains the class implementation for element/pixel related functions +*************************************************************************************************** +*/ + +#include "addrelemlib.h" +#include "addrlib.h" + + +/** +*************************************************************************************************** +* AddrElemLib::AddrElemLib +* +* @brief +* constructor +* +* @return +* N/A +*************************************************************************************************** +*/ +AddrElemLib::AddrElemLib( + AddrLib* const pAddrLib) : ///< [in] Parent addrlib instance pointer + AddrObject(pAddrLib->GetClient()), + m_pAddrLib(pAddrLib) +{ + switch (m_pAddrLib->GetAddrChipFamily()) + { + case ADDR_CHIP_FAMILY_R6XX: + m_depthPlanarType = ADDR_DEPTH_PLANAR_R600; + m_fp16ExportNorm = 0; + break; + case ADDR_CHIP_FAMILY_R7XX: + m_depthPlanarType = ADDR_DEPTH_PLANAR_R600; + m_fp16ExportNorm = 1; + break; + case ADDR_CHIP_FAMILY_R8XX: + case ADDR_CHIP_FAMILY_NI: // Same as 8xx + m_depthPlanarType = ADDR_DEPTH_PLANAR_R800; + m_fp16ExportNorm = 1; + break; + default: + m_fp16ExportNorm = 1; + m_depthPlanarType = ADDR_DEPTH_PLANAR_R800; + } + + m_configFlags.value = 0; +} + +/** +*************************************************************************************************** +* AddrElemLib::~AddrElemLib +* +* @brief +* destructor +* +* @return +* N/A +*************************************************************************************************** +*/ +AddrElemLib::~AddrElemLib() +{ +} + +/** +*************************************************************************************************** +* AddrElemLib::Create +* +* @brief +* Creates and initializes AddrLib object. +* +* @return +* Returns point to ADDR_CREATEINFO if successful. +*************************************************************************************************** +*/ +AddrElemLib* AddrElemLib::Create( + const AddrLib* const pAddrLib) ///< [in] Pointer of parent AddrLib instance +{ + AddrElemLib* pElemLib = NULL; + + if (pAddrLib) + { + pElemLib = new(pAddrLib->GetClient()) AddrElemLib(const_cast(pAddrLib)); + } + + return pElemLib; +} + +/************************************************************************************************** +* AddrElemLib::Flt32sToInt32s +* +* @brief +* Convert a ADDR_FLT_32 value to Int32 value +* +* @return +* N/A +*************************************************************************************************** +*/ +VOID AddrElemLib::Flt32sToInt32s( + ADDR_FLT_32 value, ///< [in] ADDR_FLT_32 value + UINT_32 bits, ///< [in] nubmer of bits in value + AddrNumberType numberType, ///< [in] the type of number + UINT_32* pResult) ///< [out] Int32 value +{ + UINT_8 round = 128; //ADDR_ROUND_BY_HALF + UINT_32 uscale; + UINT_32 sign; + + //convert each component to an INT_32 + switch ( numberType ) + { + case ADDR_NO_NUMBER: //fall through + case ADDR_ZERO: //fall through + case ADDR_ONE: //fall through + case ADDR_EPSILON: //fall through + return; // these are zero-bit components, so don't set result + + case ADDR_UINT_BITS: // unsigned integer bit field, clamped to range + uscale = (1< uscale)) + { + *pResult = uscale; + } + else + { + *pResult = value.i; + } + return; + } + + // The algorithm used in the DB and TX differs at one value for 24-bit unorms + case ADDR_UNORM_R6XXDB: // unsigned repeating fraction + if ((bits==24) && (value.i == 0x33000000)) + { + *pResult = 1; + return; + } // Else treat like ADDR_UNORM_R6XX + + case ADDR_UNORM_R6XX: // unsigned repeating fraction + if (value.f <= 0) + { + *pResult = 0; // first clamp to [0..1] + } + else + { + if (value.f >= 1) + { + *pResult = (1<(f + (round/256.0f)); + } + #endif + else + { + ADDR_FLT_32 scaled; + ADDR_FLT_32 shifted; + UINT_64 truncated, rounded; + UINT_32 altShift; + UINT_32 mask = (1 << bits) - 1; + UINT_32 half = 1 << (bits - 1); + UINT_32 mant24 = (value.i & 0x7FFFFF) + 0x800000; + UINT_64 temp = mant24 - (mant24>>bits) - + static_cast((mant24 & mask) > half); + UINT_32 exp8 = value.i >> 23; + UINT_32 shift = 126 - exp8 + 24 - bits; + UINT_64 final; + + if (shift >= 32) // This is zero, even with maximum dither add + { + final = 0; + } + else + { + final = ((temp<<8) + (static_cast(round)<> (shift+8); + } + //ADDR_EXIT( *pResult == final, + // ("Float %x converted to %d-bit Unorm %x != bitwise %x", + // value.u, bits, (UINT_32)*pResult, (UINT_32)final) ); + if (final > mask) + { + final = mask; + } + + scaled.f = value.f * ((1<>23)&0xFF); + truncated = (altShift > 60) ? 0 : truncated >> altShift; + rounded = static_cast((round + truncated) >> 8); + //if (rounded > ((1<(rounded); //(INT_32)final; + } + } + } + + return; + + case ADDR_S8FLOAT32: // 32-bit IEEE float, passes through NaN values + *pResult = value.i; + return; + + // @@ FIX ROUNDING in this code, fix the denorm case + case ADDR_U4FLOATC: // Unsigned float, 4-bit exponent. bias 15, clamped [0..1] + sign = (value.i >> 31) & 1; + if ((value.i&0x7F800000) == 0x7F800000) // If NaN or INF: + { + if ((value.i&0x007FFFFF) != 0) // then if NaN + { + *pResult = 0; // return 0 + } + else + { + *pResult = (sign)?0:0xF00000; // else +INF->+1, -INF->0 + } + return; + } + if (value.f <= 0) + { + *pResult = 0; + } + else + { + if (value.f>=1) + { + *pResult = 0xF << (bits-4); + } + else + { + if ((value.i>>23) > 112 ) + { + // 24-bit float: normalized + // value.i += 1 << (22-bits+4); + // round the IEEE mantissa to mantissa size + // @@ NOTE: add code to support rounding + value.u &= 0x7FFFFFF; // mask off high 4 exponent bits + *pResult = value.i >> (23-bits+4);// shift off unused mantissa bits + } + else + { + // 24-bit float: denormalized + value.f = value.f / (1<<28) / (1<<28); + value.f = value.f / (1<<28) / (1<<28); // convert to IEEE denorm + // value.i += 1 << (22-bits+4); + // round the IEEE mantissa to mantissa size + // @@ NOTE: add code to support rounding + *pResult = value.i >> (23-bits+4); // shift off unused mantissa bits + } + } + } + + return; + + default: // invalid number mode + //ADDR_EXIT(0, ("Invalid AddrNumber %d", numberType) ); + break; + + } +} + +/** +*************************************************************************************************** +* AddrElemLib::Int32sToPixel +* +* @brief +* Pack 32-bit integer values into an uncompressed pixel, +* in the proper order +* +* @return +* N/A +* +* @note +* This entry point packes four 32-bit integer values into +* an uncompressed pixel. The pixel values are specifies in +* standard order, e.g. depth/stencil. This routine asserts +* if called on compressed pixel. +*************************************************************************************************** +*/ +VOID AddrElemLib::Int32sToPixel( + UINT_32 numComps, ///< [in] number of components + UINT_32* pComps, ///< [in] compnents + UINT_32* pCompBits, ///< [in] total bits in each component + UINT_32* pCompStart, ///< [in] the first bit position of each component + ADDR_COMPONENT_FLAGS properties, ///< [in] properties about byteAligned, exportNorm + UINT_32 resultBits, ///< [in] result bits: total bpp after decompression + UINT_8* pPixel) ///< [out] a depth/stencil pixel value +{ + UINT_32 i; + UINT_32 j; + UINT_32 start; + UINT_32 size; + UINT_32 byte; + UINT_32 value = 0; + UINT_32 compMask; + UINT_32 elemMask=0; + UINT_32 elementXor = 0; // address xor when reading bytes from elements + + + // @@ NOTE: assert if called on a compressed format! + + if (properties.byteAligned) // Components are all byte-sized + { + for (i = 0; i < numComps; i++) // Then for each component + { + // Copy the bytes of the component into the element + start = pCompStart[i] / 8; + size = pCompBits[i] / 8; + for (j = 0; j < size; j++) + { + pPixel[(j+start)^elementXor] = static_cast(pComps[i] >> (8*j)); + } + } + } + else // Element is 32-bits or less, components are bit fields + { + // First, extract each component in turn and combine it into a 32-bit value + for (i = 0; i < numComps; i++) + { + compMask = (1 << pCompBits[i]) - 1; + elemMask |= compMask << pCompStart[i]; + value |= (pComps[i] & compMask) << pCompStart[i]; + } + + // Mext, copy the masked value into the element + size = (resultBits + 7) / 8; + for (i = 0; i < size; i++) + { + byte = pPixel[i^elementXor] & ~(elemMask >> (8*i)); + pPixel[i^elementXor] = static_cast(byte | ((elemMask & value) >> (8*i))); + } + } +} + +/** +*************************************************************************************************** +* Flt32ToDepthPixel +* +* @brief +* Convert a FLT_32 value to a depth/stencil pixel value +* +* @return +* N/A +*************************************************************************************************** +*/ +VOID AddrElemLib::Flt32ToDepthPixel( + AddrDepthFormat format, ///< [in] Depth format + const ADDR_FLT_32 comps[2], ///< [in] two components of depth + UINT_8* pPixel ///< [out] depth pixel value + ) const +{ + UINT_32 i; + UINT_32 values[2]; + ADDR_COMPONENT_FLAGS properties; // byteAligned, exportNorm + UINT_32 resultBits = 0; // result bits: total bits per pixel after decompression + + ADDR_PIXEL_FORMATINFO fmt; + + // get type for each component + PixGetDepthCompInfo(format, &fmt); + + //initialize properties + properties.byteAligned = TRUE; + properties.exportNorm = TRUE; + properties.floatComp = FALSE; + + //set properties and result bits + for (i = 0; i < 2; i++) + { + if ((fmt.compBit[i] & 7) || (fmt.compStart[i] & 7)) + { + properties.byteAligned = FALSE; + } + + if (resultBits < fmt.compStart[i] + fmt.compBit[i]) + { + resultBits = fmt.compStart[i] + fmt.compBit[i]; + } + + // Clear ADDR_EXPORT_NORM if can't be represented as 11-bit or smaller [-1..+1] format + if (fmt.compBit[i] > 11 || fmt.numType[i] >= ADDR_USCALED) + { + properties.exportNorm = FALSE; + } + + // Mark if there are any floating point components + if ((fmt.numType[i] == ADDR_U4FLOATC) || (fmt.numType[i] >= ADDR_S8FLOAT) ) + { + properties.floatComp = TRUE; + } + } + + // Convert the two input floats to integer values + for (i = 0; i < 2; i++) + { + Flt32sToInt32s(comps[i], fmt.compBit[i], fmt.numType[i], &values[i]); + } + + // Then pack the two integer components, in the proper order + Int32sToPixel(2, values, fmt.compBit, fmt.compStart, properties, resultBits, pPixel ); + +} + +/** +*************************************************************************************************** +* Flt32ToColorPixel +* +* @brief +* Convert a FLT_32 value to a red/green/blue/alpha pixel value +* +* @return +* N/A +*************************************************************************************************** +*/ +VOID AddrElemLib::Flt32ToColorPixel( + AddrColorFormat format, ///< [in] Color format + AddrSurfaceNumber surfNum, ///< [in] Surface number + AddrSurfaceSwap surfSwap, ///< [in] Surface swap + const ADDR_FLT_32 comps[4], ///< [in] four components of color + UINT_8* pPixel ///< [out] a red/green/blue/alpha pixel value + ) const +{ + ADDR_PIXEL_FORMATINFO pixelInfo; + + UINT_32 i; + UINT_32 values[4]; + ADDR_COMPONENT_FLAGS properties; // byteAligned, exportNorm + UINT_32 resultBits = 0; // result bits: total bits per pixel after decompression + + memset(&pixelInfo, 0, sizeof(ADDR_PIXEL_FORMATINFO)); + + PixGetColorCompInfo(format, surfNum, surfSwap, &pixelInfo); + + //initialize properties + properties.byteAligned = TRUE; + properties.exportNorm = TRUE; + properties.floatComp = FALSE; + + //set properties and result bits + for (i = 0; i < 4; i++) + { + if ( (pixelInfo.compBit[i] & 7) || (pixelInfo.compStart[i] & 7) ) + { + properties.byteAligned = FALSE; + } + + if (resultBits < pixelInfo.compStart[i] + pixelInfo.compBit[i]) + { + resultBits = pixelInfo.compStart[i] + pixelInfo.compBit[i]; + } + + if (m_fp16ExportNorm) + { + // Clear ADDR_EXPORT_NORM if can't be represented as 11-bit or smaller [-1..+1] format + // or if it's not FP and <=16 bits + if (((pixelInfo.compBit[i] > 11) || (pixelInfo.numType[i] >= ADDR_USCALED)) + && (pixelInfo.numType[i] !=ADDR_U4FLOATC)) + { + properties.exportNorm = FALSE; + } + } + else + { + // Clear ADDR_EXPORT_NORM if can't be represented as 11-bit or smaller [-1..+1] format + if (pixelInfo.compBit[i] > 11 || pixelInfo.numType[i] >= ADDR_USCALED) + { + properties.exportNorm = FALSE; + } + } + + // Mark if there are any floating point components + if ( (pixelInfo.numType[i] == ADDR_U4FLOATC) || + (pixelInfo.numType[i] >= ADDR_S8FLOAT) ) + { + properties.floatComp = TRUE; + } + } + + // Convert the four input floats to integer values + for (i = 0; i < 4; i++) + { + Flt32sToInt32s(comps[i], pixelInfo.compBit[i], pixelInfo.numType[i], &values[i]); + } + + // Then pack the four integer components, in the proper order + Int32sToPixel(4, values, &pixelInfo.compBit[0], &pixelInfo.compStart[0], + properties, resultBits, pPixel); +} + +/** +*************************************************************************************************** +* AddrElemLib::GetCompType +* +* @brief +* Fill per component info +* +* @return +* N/A +* +*************************************************************************************************** +*/ +VOID AddrElemLib::GetCompType( + AddrColorFormat format, ///< [in] surface format + AddrSurfaceNumber numType, ///< [in] number type + ADDR_PIXEL_FORMATINFO* pInfo) ///< [in][out] per component info out +{ + BOOL_32 handled = FALSE; + + // Floating point formats override the number format + switch (format) + { + case ADDR_COLOR_16_FLOAT: // fall through for all pure floating point format + case ADDR_COLOR_16_16_FLOAT: + case ADDR_COLOR_16_16_16_16_FLOAT: + case ADDR_COLOR_32_FLOAT: + case ADDR_COLOR_32_32_FLOAT: + case ADDR_COLOR_32_32_32_32_FLOAT: + case ADDR_COLOR_10_11_11_FLOAT: + case ADDR_COLOR_11_11_10_FLOAT: + numType = ADDR_NUMBER_FLOAT; + break; + // Special handling for the depth formats + case ADDR_COLOR_8_24: // fall through for these 2 similar format + case ADDR_COLOR_24_8: + for (UINT_32 c = 0; c < 4; c++) + { + if (pInfo->compBit[c] == 8) + { + pInfo->numType[c] = ADDR_UINT_BITS; + } + else if (pInfo->compBit[c] == 24) + { + pInfo->numType[c] = ADDR_UNORM_R6XX; + } + else + { + pInfo->numType[c] = ADDR_NO_NUMBER; + } + } + handled = TRUE; + break; + case ADDR_COLOR_8_24_FLOAT: // fall through for these 3 similar format + case ADDR_COLOR_24_8_FLOAT: + case ADDR_COLOR_X24_8_32_FLOAT: + for (UINT_32 c = 0; c < 4; c++) + { + if (pInfo->compBit[c] == 8) + { + pInfo->numType[c] = ADDR_UINT_BITS; + } + else if (pInfo->compBit[c] == 24) + { + pInfo->numType[c] = ADDR_U4FLOATC; + } + else if (pInfo->compBit[c] == 32) + { + pInfo->numType[c] = ADDR_S8FLOAT32; + } + else + { + pInfo->numType[c] = ADDR_NO_NUMBER; + } + } + handled = TRUE; + break; + default: + break; + } + + if (!handled) + { + for (UINT_32 c = 0; c < 4; c++) + { + // Assign a number type for each component + AddrSurfaceNumber cnum; + + // First handle default component values + if (pInfo->compBit[c] == 0) + { + if (c < 3) + { + pInfo->numType[c] = ADDR_ZERO; // Default is zero for RGB + } + else if (numType == ADDR_NUMBER_UINT || numType == ADDR_NUMBER_SINT) + { + pInfo->numType[c] = ADDR_EPSILON; // Alpha INT_32 bits default is 0x01 + } + else + { + pInfo->numType[c] = ADDR_ONE; // Alpha normal default is float 1.0 + } + continue; + } + // Now handle small components + else if (pInfo->compBit[c] == 1) + { + if (numType == ADDR_NUMBER_UINT || numType == ADDR_NUMBER_SINT) + { + cnum = ADDR_NUMBER_UINT; + } + else + { + cnum = ADDR_NUMBER_UNORM; + } + } + else + { + cnum = numType; + } + + // If no default, set the number type fom num, compbits, and architecture + switch (cnum) + { + case ADDR_NUMBER_SRGB: + pInfo->numType[c] = (c < 3) ? ADDR_GAMMA8_R6XX : ADDR_UNORM_R6XX; + break; + case ADDR_NUMBER_UNORM: + pInfo->numType[c] = ADDR_UNORM_R6XX; + break; + case ADDR_NUMBER_SNORM: + pInfo->numType[c] = ADDR_SNORM_R6XX; + break; + case ADDR_NUMBER_USCALED: + pInfo->numType[c] = ADDR_USCALED; // @@ Do we need separate Pele routine? + break; + case ADDR_NUMBER_SSCALED: + pInfo->numType[c] = ADDR_SSCALED; // @@ Do we need separate Pele routine? + break; + case ADDR_NUMBER_FLOAT: + if (pInfo->compBit[c] == 32) + { + pInfo->numType[c] = ADDR_S8FLOAT32; + } + else if (pInfo->compBit[c] == 16) + { + pInfo->numType[c] = ADDR_S5FLOAT; + } + else if (pInfo->compBit[c] >= 10) + { + pInfo->numType[c] = ADDR_U5FLOAT; + } + else + { + ADDR_ASSERT_ALWAYS(); + } + break; + case ADDR_NUMBER_SINT: + pInfo->numType[c] = ADDR_SINT_BITS; + break; + case ADDR_NUMBER_UINT: + pInfo->numType[c] = ADDR_UINT_BITS; + break; + + default: + ADDR_ASSERT(!"Invalid number type"); + pInfo->numType[c] = ADDR_NO_NUMBER; + break; + } + } + } +} + +/** +*************************************************************************************************** +* AddrElemLib::GetCompSwap +* +* @brief +* Get components swapped for color surface +* +* @return +* N/A +* +*************************************************************************************************** +*/ +VOID AddrElemLib::GetCompSwap( + AddrSurfaceSwap swap, ///< [in] swap mode + ADDR_PIXEL_FORMATINFO* pInfo) ///< [in/out] output per component info +{ + switch (pInfo->comps) + { + case 4: + switch (swap) + { + case ADDR_SWAP_ALT: + SwapComps( 0, 2, pInfo ); + break; // BGRA + case ADDR_SWAP_STD_REV: + SwapComps( 0, 3, pInfo ); + SwapComps( 1, 2, pInfo ); + break; // ABGR + case ADDR_SWAP_ALT_REV: + SwapComps( 0, 3, pInfo ); + SwapComps( 0, 2, pInfo ); + SwapComps( 0, 1, pInfo ); + break; // ARGB + default: + break; + } + break; + case 3: + switch (swap) + { + case ADDR_SWAP_ALT_REV: + SwapComps( 0, 3, pInfo ); + SwapComps( 0, 2, pInfo ); + break; // AGR + case ADDR_SWAP_STD_REV: + SwapComps( 0, 2, pInfo ); + break; // BGR + case ADDR_SWAP_ALT: + SwapComps( 2, 3, pInfo ); + break; // RGA + default: + break; // RGB + } + break; + case 2: + switch (swap) + { + case ADDR_SWAP_ALT_REV: + SwapComps( 0, 1, pInfo ); + SwapComps( 1, 3, pInfo ); + break; // AR + case ADDR_SWAP_STD_REV: + SwapComps( 0, 1, pInfo ); + break; // GR + case ADDR_SWAP_ALT: + SwapComps( 1, 3, pInfo ); + break; // RA + default: + break; // RG + } + break; + case 1: + switch (swap) + { + case ADDR_SWAP_ALT_REV: + SwapComps( 0, 3, pInfo ); + break; // A + case ADDR_SWAP_STD_REV: + SwapComps( 0, 2, pInfo ); + break; // B + case ADDR_SWAP_ALT: + SwapComps( 0, 1, pInfo ); + break; // G + default: + break; // R + } + break; + } +} + +/** +*************************************************************************************************** +* AddrElemLib::GetCompSwap +* +* @brief +* Get components swapped for color surface +* +* @return +* N/A +* +*************************************************************************************************** +*/ +VOID AddrElemLib::SwapComps( + UINT_32 c0, ///< [in] component index 0 + UINT_32 c1, ///< [in] component index 1 + ADDR_PIXEL_FORMATINFO* pInfo) ///< [in/out] output per component info +{ + UINT_32 start; + UINT_32 bits; + + start = pInfo->compStart[c0]; + pInfo->compStart[c0] = pInfo->compStart[c1]; + pInfo->compStart[c1] = start; + + bits = pInfo->compBit[c0]; + pInfo->compBit[c0] = pInfo->compBit[c1]; + pInfo->compBit[c1] = bits; +} + +/** +*************************************************************************************************** +* AddrElemLib::PixGetColorCompInfo +* +* @brief +* Get per component info for color surface +* +* @return +* N/A +* +*************************************************************************************************** +*/ +VOID AddrElemLib::PixGetColorCompInfo( + AddrColorFormat format, ///< [in] surface format, read from register + AddrSurfaceNumber number, ///< [in] pixel number type + AddrSurfaceSwap swap, ///< [in] component swap mode + ADDR_PIXEL_FORMATINFO* pInfo ///< [out] output per component info + ) const +{ + // 1. Get componet bits + switch (format) + { + case ADDR_COLOR_8: + GetCompBits(8, 0, 0, 0, pInfo); + break; + case ADDR_COLOR_1_5_5_5: + GetCompBits(5, 5, 5, 1, pInfo); + break; + case ADDR_COLOR_5_6_5: + GetCompBits(8, 6, 5, 0, pInfo); + break; + case ADDR_COLOR_6_5_5: + GetCompBits(5, 5, 6, 0, pInfo); + break; + case ADDR_COLOR_8_8: + GetCompBits(8, 8, 0, 0, pInfo); + break; + case ADDR_COLOR_4_4_4_4: + GetCompBits(4, 4, 4, 4, pInfo); + break; + case ADDR_COLOR_16: + GetCompBits(16, 0, 0, 0, pInfo); + break; + case ADDR_COLOR_8_8_8_8: + GetCompBits(8, 8, 8, 8, pInfo); + break; + case ADDR_COLOR_2_10_10_10: + GetCompBits(10, 10, 10, 2, pInfo); + break; + case ADDR_COLOR_10_11_11: + GetCompBits(11, 11, 10, 0, pInfo); + break; + case ADDR_COLOR_11_11_10: + GetCompBits(10, 11, 11, 0, pInfo); + break; + case ADDR_COLOR_16_16: + GetCompBits(16, 16, 0, 0, pInfo); + break; + case ADDR_COLOR_16_16_16_16: + GetCompBits(16, 16, 16, 16, pInfo); + break; + case ADDR_COLOR_16_FLOAT: + GetCompBits(16, 0, 0, 0, pInfo); + break; + case ADDR_COLOR_16_16_FLOAT: + GetCompBits(16, 16, 0, 0, pInfo); + break; + case ADDR_COLOR_32_FLOAT: + GetCompBits(32, 0, 0, 0, pInfo); + break; + case ADDR_COLOR_32_32_FLOAT: + GetCompBits(32, 32, 0, 0, pInfo); + break; + case ADDR_COLOR_16_16_16_16_FLOAT: + GetCompBits(16, 16, 16, 16, pInfo); + break; + case ADDR_COLOR_32_32_32_32_FLOAT: + GetCompBits(32, 32, 32, 32, pInfo); + break; + + case ADDR_COLOR_32: + GetCompBits(32, 0, 0, 0, pInfo); + break; + case ADDR_COLOR_32_32: + GetCompBits(32, 32, 0, 0, pInfo); + break; + case ADDR_COLOR_32_32_32_32: + GetCompBits(32, 32, 32, 32, pInfo); + break; + case ADDR_COLOR_10_10_10_2: + GetCompBits(2, 10, 10, 10, pInfo); + break; + case ADDR_COLOR_10_11_11_FLOAT: + GetCompBits(11, 11, 10, 0, pInfo); + break; + case ADDR_COLOR_11_11_10_FLOAT: + GetCompBits(10, 11, 11, 0, pInfo); + break; + case ADDR_COLOR_5_5_5_1: + GetCompBits(1, 5, 5, 5, pInfo); + break; + case ADDR_COLOR_3_3_2: + GetCompBits(2, 3, 3, 0, pInfo); + break; + case ADDR_COLOR_4_4: + GetCompBits(4, 4, 0, 0, pInfo); + break; + case ADDR_COLOR_8_24: + case ADDR_COLOR_8_24_FLOAT: // same bit count, fall through + GetCompBits(24, 8, 0, 0, pInfo); + break; + case ADDR_COLOR_24_8: + case ADDR_COLOR_24_8_FLOAT: // same bit count, fall through + GetCompBits(8, 24, 0, 0, pInfo); + break; + case ADDR_COLOR_X24_8_32_FLOAT: + GetCompBits(32, 8, 0, 0, pInfo); + break; + + case ADDR_COLOR_INVALID: + GetCompBits(0, 0, 0, 0, pInfo); + break; + default: + ADDR_ASSERT(0); + GetCompBits(0, 0, 0, 0, pInfo); + break; + } + + // 2. Get component number type + + GetCompType(format, number, pInfo); + + // 3. Swap components if needed + + GetCompSwap(swap, pInfo); +} + +/** +*************************************************************************************************** +* AddrElemLib::PixGetDepthCompInfo +* +* @brief +* Get per component info for depth surface +* +* @return +* N/A +* +*************************************************************************************************** +*/ +VOID AddrElemLib::PixGetDepthCompInfo( + AddrDepthFormat format, ///< [in] surface format, read from register + ADDR_PIXEL_FORMATINFO* pInfo ///< [out] output per component bits and type + ) const +{ + if (m_depthPlanarType == ADDR_DEPTH_PLANAR_R800) + { + if (format == ADDR_DEPTH_8_24_FLOAT) + { + format = ADDR_DEPTH_X24_8_32_FLOAT; // Use this format to represent R800's D24FS8 + } + + if (format == ADDR_DEPTH_X8_24_FLOAT) + { + format = ADDR_DEPTH_32_FLOAT; + } + } + + switch (format) + { + case ADDR_DEPTH_16: + GetCompBits(16, 0, 0, 0, pInfo); + break; + case ADDR_DEPTH_8_24: + case ADDR_DEPTH_8_24_FLOAT: // similar format, fall through + GetCompBits(24, 8, 0, 0, pInfo); + break; + case ADDR_DEPTH_X8_24: + case ADDR_DEPTH_X8_24_FLOAT: // similar format, fall through + GetCompBits(24, 0, 0, 0, pInfo); + break; + case ADDR_DEPTH_32_FLOAT: + GetCompBits(32, 0, 0, 0, pInfo); + break; + case ADDR_DEPTH_X24_8_32_FLOAT: + GetCompBits(32, 8, 0, 0, pInfo); + break; + case ADDR_DEPTH_INVALID: + GetCompBits(0, 0, 0, 0, pInfo); + break; + default: + ADDR_ASSERT(0); + GetCompBits(0, 0, 0, 0, pInfo); + break; + } + + switch (format) + { + case ADDR_DEPTH_16: + pInfo->numType [0] = ADDR_UNORM_R6XX; + pInfo->numType [1] = ADDR_ZERO; + break; + case ADDR_DEPTH_8_24: + pInfo->numType [0] = ADDR_UNORM_R6XXDB; + pInfo->numType [1] = ADDR_UINT_BITS; + break; + case ADDR_DEPTH_8_24_FLOAT: + pInfo->numType [0] = ADDR_U4FLOATC; + pInfo->numType [1] = ADDR_UINT_BITS; + break; + case ADDR_DEPTH_X8_24: + pInfo->numType [0] = ADDR_UNORM_R6XXDB; + pInfo->numType [1] = ADDR_ZERO; + break; + case ADDR_DEPTH_X8_24_FLOAT: + pInfo->numType [0] = ADDR_U4FLOATC; + pInfo->numType [1] = ADDR_ZERO; + break; + case ADDR_DEPTH_32_FLOAT: + pInfo->numType [0] = ADDR_S8FLOAT32; + pInfo->numType [1] = ADDR_ZERO; + break; + case ADDR_DEPTH_X24_8_32_FLOAT: + pInfo->numType [0] = ADDR_S8FLOAT32; + pInfo->numType [1] = ADDR_UINT_BITS; + break; + default: + pInfo->numType [0] = ADDR_NO_NUMBER; + pInfo->numType [1] = ADDR_NO_NUMBER; + break; + } + + pInfo->numType [2] = ADDR_NO_NUMBER; + pInfo->numType [3] = ADDR_NO_NUMBER; +} + +/** +*************************************************************************************************** +* AddrElemLib::PixGetExportNorm +* +* @brief +* Check if fp16 export norm can be enabled. +* +* @return +* TRUE if this can be enabled. +* +*************************************************************************************************** +*/ +BOOL_32 AddrElemLib::PixGetExportNorm( + AddrColorFormat colorFmt, ///< [in] surface format, read from register + AddrSurfaceNumber numberFmt, ///< [in] pixel number type + AddrSurfaceSwap swap ///< [in] components swap type + ) const +{ + BOOL_32 enabled = TRUE; + + ADDR_PIXEL_FORMATINFO formatInfo; + + PixGetColorCompInfo(colorFmt, numberFmt, swap, &formatInfo); + + for (UINT_32 c = 0; c < 4; c++) + { + if (m_fp16ExportNorm) + { + if (((formatInfo.compBit[c] > 11) || (formatInfo.numType[c] > ADDR_USCALED)) && + (formatInfo.numType[c] != ADDR_U4FLOATC) && + (formatInfo.numType[c] != ADDR_S5FLOAT) && + (formatInfo.numType[c] != ADDR_S5FLOATM) && + (formatInfo.numType[c] != ADDR_U5FLOAT) && + (formatInfo.numType[c] != ADDR_U3FLOATM)) + { + enabled = FALSE; + break; + } + } + else + { + if ((formatInfo.compBit[c] > 11) || (formatInfo.numType[c] > ADDR_USCALED)) + { + enabled = FALSE; + break; + } + } + } + + return enabled; +} + +/** +*************************************************************************************************** +* AddrElemLib::AdjustSurfaceInfo +* +* @brief +* Adjust bpp/base pitch/width/height according to elemMode and expandX/Y +* +* @return +* N/A +*************************************************************************************************** +*/ +VOID AddrElemLib::AdjustSurfaceInfo( + AddrElemMode elemMode, ///< [in] element mode + UINT_32 expandX, ///< [in] decompression expansion factor in X + UINT_32 expandY, ///< [in] decompression expansion factor in Y + UINT_32* pBpp, ///< [in/out] bpp + UINT_32* pBasePitch, ///< [in/out] base pitch + UINT_32* pWidth, ///< [in/out] width + UINT_32* pHeight) ///< [in/out] height +{ + UINT_32 packedBits; + UINT_32 basePitch; + UINT_32 width; + UINT_32 height; + UINT_32 bpp; + BOOL_32 bBCnFormat = FALSE; + + ADDR_ASSERT(pBpp != NULL); + ADDR_ASSERT(pWidth != NULL && pHeight != NULL && pBasePitch != NULL); + + if (pBpp) + { + bpp = *pBpp; + + switch (elemMode) + { + case ADDR_EXPANDED: + packedBits = bpp / expandX / expandY; + break; + case ADDR_PACKED_STD: // Different bit order + case ADDR_PACKED_REV: + packedBits = bpp * expandX * expandY; + break; + case ADDR_PACKED_GBGR: + case ADDR_PACKED_BGRG: + packedBits = bpp; // 32-bit packed ==> 2 32-bit result + break; + case ADDR_PACKED_BC1: // Fall through + case ADDR_PACKED_BC4: + packedBits = 64; + bBCnFormat = TRUE; + break; + case ADDR_PACKED_BC2: // Fall through + case ADDR_PACKED_BC3: // Fall through + case ADDR_PACKED_BC5: // Fall through + bBCnFormat = TRUE; + packedBits = 128; + break; + case ADDR_ROUND_BY_HALF: // Fall through + case ADDR_ROUND_TRUNCATE: // Fall through + case ADDR_ROUND_DITHER: // Fall through + case ADDR_UNCOMPRESSED: + packedBits = bpp; + break; + default: + packedBits = bpp; + ADDR_ASSERT_ALWAYS(); + break; + } + + *pBpp = packedBits; + } + + if (pWidth && pHeight && pBasePitch) + { + basePitch = *pBasePitch; + width = *pWidth; + height = *pHeight; + + if ((expandX > 1) || (expandY > 1)) + { + if (elemMode == ADDR_EXPANDED) + { + basePitch *= expandX; + width *= expandX; + height *= expandY; + } + else + { + // Evergreen family workaround + if (bBCnFormat && (m_pAddrLib->GetAddrChipFamily() == ADDR_CHIP_FAMILY_R8XX)) + { + // For BCn we now pad it to POW2 at the beginning so it is safe to + // divide by 4 directly + basePitch = basePitch / expandX; + width = width / expandX; + height = height / expandY; +#if DEBUG + width = (width == 0) ? 1 : width; + height = (height == 0) ? 1 : height; + + if ((*pWidth > PowTwoAlign(width, 8) * expandX) || + (*pHeight > PowTwoAlign(height, 8) * expandY)) // 8 is 1D tiling alignment + { + // if this assertion is hit we may have issues if app samples + // rightmost/bottommost pixels + ADDR_ASSERT_ALWAYS(); + } +#endif + } + else // Not BCn format we still keep old way (FMT_1? No real test yet) + { + basePitch = (basePitch + expandX - 1) / expandX; + width = (width + expandX - 1) / expandX; + height = (height + expandY - 1) / expandY; + } + } + + *pBasePitch = basePitch; // 0 is legal value for base pitch. + *pWidth = (width == 0) ? 1 : width; + *pHeight = (height == 0) ? 1 : height; + } //if (pWidth && pHeight && pBasePitch) + } +} + +/** +*************************************************************************************************** +* AddrElemLib::RestoreSurfaceInfo +* +* @brief +* Reverse operation of AdjustSurfaceInfo +* +* @return +* N/A +*************************************************************************************************** +*/ +VOID AddrElemLib::RestoreSurfaceInfo( + AddrElemMode elemMode, ///< [in] element mode + UINT_32 expandX, ///< [in] decompression expansion factor in X + UINT_32 expandY, ///< [out] decompression expansion factor in Y + UINT_32* pBpp, ///< [in/out] bpp + UINT_32* pWidth, ///< [in/out] width + UINT_32* pHeight) ///< [in/out] height +{ + UINT_32 originalBits; + UINT_32 width; + UINT_32 height; + UINT_32 bpp; + + BOOL_32 bBCnFormat = FALSE; + + ADDR_ASSERT(pBpp != NULL); + ADDR_ASSERT(pWidth != NULL && pHeight != NULL); + + if (pBpp) + { + bpp = *pBpp; + + switch (elemMode) + { + case ADDR_EXPANDED: + originalBits = bpp * expandX * expandY; + break; + case ADDR_PACKED_STD: // Different bit order + case ADDR_PACKED_REV: + originalBits = bpp / expandX / expandY; + break; + case ADDR_PACKED_GBGR: + case ADDR_PACKED_BGRG: + originalBits = bpp; // 32-bit packed ==> 2 32-bit result + break; + case ADDR_PACKED_BC1: // Fall through + case ADDR_PACKED_BC4: + originalBits = 64; + bBCnFormat = TRUE; + break; + case ADDR_PACKED_BC2: // Fall through + case ADDR_PACKED_BC3: // Fall through + case ADDR_PACKED_BC5: + bBCnFormat = TRUE; + originalBits = 128; + break; + case ADDR_ROUND_BY_HALF: // Fall through + case ADDR_ROUND_TRUNCATE: // Fall through + case ADDR_ROUND_DITHER: // Fall through + case ADDR_UNCOMPRESSED: + originalBits = bpp; + break; + default: + originalBits = bpp; + ADDR_ASSERT_ALWAYS(); + break; + } + + *pBpp = originalBits; + } + + if (pWidth && pHeight) + { + width = *pWidth; + height = *pHeight; + + if ((expandX > 1) || (expandY > 1)) + { + if (elemMode == ADDR_EXPANDED) + { + width /= expandX; + height /= expandY; + } + else + { + width *= expandX; + height *= expandY; + } + } + + *pWidth = (width == 0) ? 1 : width; + *pHeight = (height == 0) ? 1 : height; + } +} + +/** +*************************************************************************************************** +* AddrElemLib::GetBitsPerPixel +* +* @brief +* Compute the total bits per element according to a format +* code. For compressed formats, this is not the same as +* the number of bits per decompressed element. +* +* @return +* Bits per pixel +*************************************************************************************************** +*/ +UINT_32 AddrElemLib::GetBitsPerPixel( + AddrFormat format, ///< [in] surface format code + AddrElemMode* pElemMode, ///< [out] element mode + UINT_32* pExpandX, ///< [out] decompression expansion factor in X + UINT_32* pExpandY, ///< [out] decompression expansion factor in Y + UINT_32* pUnusedBits) ///< [out] bits unused +{ + UINT_32 bpp; + UINT_32 expandX = 1; + UINT_32 expandY = 1; + UINT_32 bitUnused = 0; + AddrElemMode elemMode = ADDR_UNCOMPRESSED; // default value + + switch (format) + { + case ADDR_FMT_8: + bpp = 8; + break; + case ADDR_FMT_1_5_5_5: + case ADDR_FMT_5_6_5: + case ADDR_FMT_6_5_5: + case ADDR_FMT_8_8: + case ADDR_FMT_4_4_4_4: + case ADDR_FMT_16: + case ADDR_FMT_16_FLOAT: + bpp = 16; + break; + case ADDR_FMT_GB_GR: // treat as FMT_8_8 + elemMode = ADDR_PACKED_GBGR; + bpp = 16; + break; + case ADDR_FMT_BG_RG: // treat as FMT_8_8 + elemMode = ADDR_PACKED_BGRG; + bpp = 16; + break; + case ADDR_FMT_8_8_8_8: + case ADDR_FMT_2_10_10_10: + case ADDR_FMT_10_11_11: + case ADDR_FMT_11_11_10: + case ADDR_FMT_16_16: + case ADDR_FMT_16_16_FLOAT: + case ADDR_FMT_32: + case ADDR_FMT_32_FLOAT: + case ADDR_FMT_24_8: + case ADDR_FMT_24_8_FLOAT: + bpp = 32; + break; + case ADDR_FMT_16_16_16_16: + case ADDR_FMT_16_16_16_16_FLOAT: + case ADDR_FMT_32_32: + case ADDR_FMT_32_32_FLOAT: + case ADDR_FMT_CTX1: + bpp = 64; + break; + case ADDR_FMT_32_32_32_32: + case ADDR_FMT_32_32_32_32_FLOAT: + bpp = 128; + break; + case ADDR_FMT_INVALID: + bpp = 0; + break; + case ADDR_FMT_1_REVERSED: + elemMode = ADDR_PACKED_REV; + expandX = 8; + bpp = 1; + break; + case ADDR_FMT_1: + elemMode = ADDR_PACKED_STD; + expandX = 8; + bpp = 1; + break; + case ADDR_FMT_4_4: + case ADDR_FMT_3_3_2: + bpp = 8; + break; + case ADDR_FMT_5_5_5_1: + bpp = 16; + break; + case ADDR_FMT_32_AS_8: + case ADDR_FMT_32_AS_8_8: + case ADDR_FMT_8_24: + case ADDR_FMT_8_24_FLOAT: + case ADDR_FMT_10_10_10_2: + case ADDR_FMT_10_11_11_FLOAT: + case ADDR_FMT_11_11_10_FLOAT: + case ADDR_FMT_5_9_9_9_SHAREDEXP: + bpp = 32; + break; + case ADDR_FMT_X24_8_32_FLOAT: + bpp = 64; + bitUnused = 24; + break; + case ADDR_FMT_8_8_8: + elemMode = ADDR_EXPANDED; + bpp = 24;//@@ 8; // read 3 elements per pixel + expandX = 3; + break; + case ADDR_FMT_16_16_16: + case ADDR_FMT_16_16_16_FLOAT: + elemMode = ADDR_EXPANDED; + bpp = 48;//@@ 16; // read 3 elements per pixel + expandX = 3; + break; + case ADDR_FMT_32_32_32_FLOAT: + case ADDR_FMT_32_32_32: + elemMode = ADDR_EXPANDED; + expandX = 3; + bpp = 96;//@@ 32; // read 3 elements per pixel + break; + case ADDR_FMT_BC1: + elemMode = ADDR_PACKED_BC1; + expandX = 4; + expandY = 4; + bpp = 64; + break; + case ADDR_FMT_BC4: + elemMode = ADDR_PACKED_BC4; + expandX = 4; + expandY = 4; + bpp = 64; + break; + case ADDR_FMT_BC2: + elemMode = ADDR_PACKED_BC2; + expandX = 4; + expandY = 4; + bpp = 128; + break; + case ADDR_FMT_BC3: + elemMode = ADDR_PACKED_BC3; + expandX = 4; + expandY = 4; + bpp = 128; + break; + case ADDR_FMT_BC5: + case ADDR_FMT_BC6: // reuse ADDR_PACKED_BC5 + case ADDR_FMT_BC7: // reuse ADDR_PACKED_BC5 + elemMode = ADDR_PACKED_BC5; + expandX = 4; + expandY = 4; + bpp = 128; + break; + default: + bpp = 0; + ADDR_ASSERT_ALWAYS(); + break; + // @@ or should this be an error? + } + + SafeAssign(pExpandX, expandX); + SafeAssign(pExpandY, expandY); + SafeAssign(pUnusedBits, bitUnused); + SafeAssign(reinterpret_cast(pElemMode), elemMode); + + return bpp; +} + +/** +*************************************************************************************************** +* AddrElemLib::GetCompBits +* +* @brief +* Set each component's bit size and bit start. And set element mode and number type +* +* @return +* N/A +*************************************************************************************************** +*/ +VOID AddrElemLib::GetCompBits( + UINT_32 c0, ///< [in] bits of component 0 + UINT_32 c1, ///< [in] bits of component 1 + UINT_32 c2, ///< [in] bits of component 2 + UINT_32 c3, ///< [in] bits of component 3 + ADDR_PIXEL_FORMATINFO* pInfo, ///< [out] per component info out + AddrElemMode elemMode) ///< [in] element mode +{ + pInfo->comps = 0; + + pInfo->compBit[0] = c0; + pInfo->compBit[1] = c1; + pInfo->compBit[2] = c2; + pInfo->compBit[3] = c3; + + pInfo->compStart[0] = 0; + pInfo->compStart[1] = c0; + pInfo->compStart[2] = c0+c1; + pInfo->compStart[3] = c0+c1+c2; + + pInfo->elemMode = elemMode; + // still needed since component swap may depend on number of components + for (INT i=0; i<4; i++) + { + if (pInfo->compBit[i] == 0) + { + pInfo->compStart[i] = 0; // all null components start at bit 0 + pInfo->numType[i] = ADDR_NO_NUMBER; // and have no number type + } + else + { + pInfo->comps++; + } + } +} + +/** +*************************************************************************************************** +* AddrElemLib::GetCompBits +* +* @brief +* Set the clear color (or clear depth/stencil) for a surface +* +* @note +* If clearColor is zero, a default clear value is used in place of comps[4]. +* If float32 is set, full precision is used, else the mantissa is reduced to 12-bits +* +* @return +* N/A +*************************************************************************************************** +*/ +VOID AddrElemLib::SetClearComps( + ADDR_FLT_32 comps[4], ///< [in/out] components + BOOL_32 clearColor, ///< [in] TRUE if clear color is set (CLEAR_COLOR) + BOOL_32 float32) ///< [in] TRUE if float32 component (BLEND_FLOAT32) +{ + INT_32 i; + + // Use default clearvalues if clearColor is disabled + if (clearColor == FALSE) + { + for (i=0; i<3; i++) + { + comps[i].f = 0.0; + } + comps[3].f = 1.0; + } + + // Otherwise use the (modified) clear value + else + { + for (i=0; i<4; i++) + { // If full precision, use clear value unchanged + if (float32) + { + // Do nothing + //comps[i] = comps[i]; + } + // Else if it is a NaN, use the standard NaN value + else if ((comps[i].u & 0x7FFFFFFF) > 0x7F800000) + { + comps[i].u = 0xFFC00000; + } + // Else reduce the mantissa precision + else + { + comps[i].u = comps[i].u & 0xFFFFF000; + } + } + } +} + +/** +*************************************************************************************************** +* AddrElemLib::IsBlockCompressed +* +* @brief +* TRUE if this is block compressed format +* +* @note +* +* @return +* BOOL_32 +*************************************************************************************************** +*/ +BOOL_32 AddrElemLib::IsBlockCompressed( + AddrFormat format) ///< [in] Format +{ + return format >= ADDR_FMT_BC1 && format <= ADDR_FMT_BC7; +} + + +/** +*************************************************************************************************** +* AddrElemLib::IsCompressed +* +* @brief +* TRUE if this is block compressed format or 1 bit format +* +* @note +* +* @return +* BOOL_32 +*************************************************************************************************** +*/ +BOOL_32 AddrElemLib::IsCompressed( + AddrFormat format) ///< [in] Format +{ + return IsBlockCompressed(format) || format == ADDR_FMT_BC1 || format == ADDR_FMT_BC7; +} + +/** +*************************************************************************************************** +* AddrElemLib::IsExpand3x +* +* @brief +* TRUE if this is 3x expand format +* +* @note +* +* @return +* BOOL_32 +*************************************************************************************************** +*/ +BOOL_32 AddrElemLib::IsExpand3x( + AddrFormat format) ///< [in] Format +{ + BOOL_32 is3x = FALSE; + + switch (format) + { + case ADDR_FMT_8_8_8: + case ADDR_FMT_16_16_16: + case ADDR_FMT_16_16_16_FLOAT: + case ADDR_FMT_32_32_32: + case ADDR_FMT_32_32_32_FLOAT: + is3x = TRUE; + break; + default: + break; + } + + return is3x; +} + + diff --git a/src/gallium/winsys/radeon/amdgpu/addrlib/core/addrelemlib.h b/src/gallium/winsys/radeon/amdgpu/addrlib/core/addrelemlib.h new file mode 100644 index 0000000..c302b3b --- /dev/null +++ b/src/gallium/winsys/radeon/amdgpu/addrlib/core/addrelemlib.h @@ -0,0 +1,270 @@ +/* + * Copyright © 2014 Advanced Micro Devices, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES + * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NON-INFRINGEMENT. IN NO EVENT SHALL THE COPYRIGHT HOLDERS, AUTHORS + * AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + */ + +/** +*************************************************************************************************** +* @file addrelemlib.h +* @brief Contains the class for element/pixel related functions +*************************************************************************************************** +*/ + +#ifndef __ELEM_LIB_H__ +#define __ELEM_LIB_H__ + +#include "addrinterface.h" +#include "addrobject.h" +#include "addrcommon.h" + +class AddrLib; + +// The masks for property bits within the Properties INT_32 +union ADDR_COMPONENT_FLAGS +{ + struct + { + UINT_32 byteAligned : 1; ///< all components are byte aligned + UINT_32 exportNorm : 1; ///< components support R6xx NORM compression + UINT_32 floatComp : 1; ///< there is at least one floating point component + }; + + UINT_32 value; +}; + +// Copy from legacy lib's AddrNumberType +enum AddrNumberType +{ + // The following number types have the range [-1..1] + ADDR_NO_NUMBER, // This component doesn't exist and has no default value + ADDR_EPSILON, // Force component value to integer 0x00000001 + ADDR_ZERO, // Force component value to integer 0x00000000 + ADDR_ONE, // Force component value to floating point 1.0 + // Above values don't have any bits per component (keep ADDR_ONE the last of these) + + ADDR_UNORM, // Unsigned normalized (repeating fraction) full precision + ADDR_SNORM, // Signed normalized (repeating fraction) full precision + ADDR_GAMMA, // Gamma-corrected, full precision + + ADDR_UNORM_R5XXRB, // Unsigned normalized (repeating fraction) for r5xx RB + ADDR_SNORM_R5XXRB, // Signed normalized (repeating fraction) for r5xx RB + ADDR_GAMMA_R5XXRB, // Gamma-corrected for r5xx RB (note: unnormalized value) + ADDR_UNORM_R5XXBC, // Unsigned normalized (repeating fraction) for r5xx BC + ADDR_SNORM_R5XXBC, // Signed normalized (repeating fraction) for r5xx BC + ADDR_GAMMA_R5XXBC, // Gamma-corrected for r5xx BC (note: unnormalized value) + + ADDR_UNORM_R6XX, // Unsigned normalized (repeating fraction) for R6xx + ADDR_UNORM_R6XXDB, // Unorms for 24-bit depth: one value differs from ADDR_UNORM_R6XX + ADDR_SNORM_R6XX, // Signed normalized (repeating fraction) for R6xx + ADDR_GAMMA8_R6XX, // Gamma-corrected for r6xx + ADDR_GAMMA8_R7XX_TP, // Gamma-corrected for r7xx TP 12bit unorm 8.4. + + ADDR_U4FLOATC, // Unsigned float: 4-bit exponent, bias=15, no NaN, clamp [0..1] + ADDR_GAMMA_4SEG, // Gamma-corrected, four segment approximation + ADDR_U0FIXED, // Unsigned 0.N-bit fixed point + + // The following number types have large ranges (LEAVE ADDR_USCALED first or fix Finish routine) + ADDR_USCALED, // Unsigned integer converted to/from floating point + ADDR_SSCALED, // Signed integer converted to/from floating point + ADDR_USCALED_R5XXRB, // Unsigned integer to/from floating point for r5xx RB + ADDR_SSCALED_R5XXRB, // Signed integer to/from floating point for r5xx RB + ADDR_UINT_BITS, // Keep in unsigned integer form, clamped to specified range + ADDR_SINT_BITS, // Keep in signed integer form, clamped to specified range + ADDR_UINTBITS, // @@ remove Keep in unsigned integer form, use modulus to reduce bits + ADDR_SINTBITS, // @@ remove Keep in signed integer form, use modulus to reduce bits + + // The following number types and ADDR_U4FLOATC have exponents + // (LEAVE ADDR_S8FLOAT first or fix Finish routine) + ADDR_S8FLOAT, // Signed floating point with 8-bit exponent, bias=127 + ADDR_S8FLOAT32, // 32-bit IEEE float, passes through NaN values + ADDR_S5FLOAT, // Signed floating point with 5-bit exponent, bias=15 + ADDR_S5FLOATM, // Signed floating point with 5-bit exponent, bias=15, no NaN/Inf + ADDR_U5FLOAT, // Signed floating point with 5-bit exponent, bias=15 + ADDR_U3FLOATM, // Unsigned floating point with 3-bit exponent, bias=3 + + ADDR_S5FIXED, // Signed 5.N-bit fixed point, with rounding + + ADDR_END_NUMBER // Used for range comparisons +}; + +// Copy from legacy lib's AddrElement +enum AddrElemMode +{ + // These formats allow both packing an unpacking + ADDR_ROUND_BY_HALF, // add 1/2 and truncate when packing this element + ADDR_ROUND_TRUNCATE, // truncate toward 0 for sign/mag, else toward neg + ADDR_ROUND_DITHER, // Pack by dithering -- requires (x,y) position + + // These formats only allow unpacking, no packing + ADDR_UNCOMPRESSED, // Elements are not compressed: one data element per pixel/texel + ADDR_EXPANDED, // Elements are split up and stored in multiple data elements + ADDR_PACKED_STD, // Elements are compressed into ExpandX by ExpandY data elements + ADDR_PACKED_REV, // Like ADDR_PACKED, but X order of pixels is reverved + ADDR_PACKED_GBGR, // Elements are compressed 4:2:2 in G1B_G0R order (high to low) + ADDR_PACKED_BGRG, // Elements are compressed 4:2:2 in BG1_RG0 order (high to low) + ADDR_PACKED_BC1, // Each data element is uncompressed to a 4x4 pixel/texel array + ADDR_PACKED_BC2, // Each data element is uncompressed to a 4x4 pixel/texel array + ADDR_PACKED_BC3, // Each data element is uncompressed to a 4x4 pixel/texel array + ADDR_PACKED_BC4, // Each data element is uncompressed to a 4x4 pixel/texel array + ADDR_PACKED_BC5, // Each data element is uncompressed to a 4x4 pixel/texel array + + // These formats provide various kinds of compression + ADDR_ZPLANE_R5XX, // Compressed Zplane using r5xx architecture format + ADDR_ZPLANE_R6XX, // Compressed Zplane using r6xx architecture format + //@@ Fill in the compression modes + + ADDR_END_ELEMENT // Used for range comparisons +}; + +enum AddrDepthPlanarType +{ + ADDR_DEPTH_PLANAR_NONE = 0, // No plane z/stencl + ADDR_DEPTH_PLANAR_R600 = 1, // R600 z and stencil planes are store within a tile + ADDR_DEPTH_PLANAR_R800 = 2, // R800 has separate z and stencil planes +}; + +/** +*************************************************************************************************** +* ADDR_PIXEL_FORMATINFO +* +* @brief +* Per component info +* +*************************************************************************************************** +*/ +struct ADDR_PIXEL_FORMATINFO +{ + UINT_32 compBit[4]; + AddrNumberType numType[4]; + UINT_32 compStart[4]; + AddrElemMode elemMode; + UINT_32 comps; ///< Number of components +}; + +/** +*************************************************************************************************** +* @brief This class contains asic indepentent element related attributes and operations +*************************************************************************************************** +*/ +class AddrElemLib : public AddrObject +{ +protected: + AddrElemLib(AddrLib* const pAddrLib); + +public: + + /// Makes this class virtual + virtual ~AddrElemLib(); + + static AddrElemLib *Create( + const AddrLib* const pAddrLib); + + /// The implementation is only for R6xx/R7xx, so make it virtual in case we need for R8xx + BOOL_32 PixGetExportNorm( + AddrColorFormat colorFmt, + AddrSurfaceNumber numberFmt, AddrSurfaceSwap swap) const; + + /// Below method are asic independent, so make them just static. + /// Remove static if we need different operation in hwl. + + VOID Flt32ToDepthPixel( + AddrDepthFormat format, const ADDR_FLT_32 comps[2], UINT_8 *pPixel) const; + + VOID Flt32ToColorPixel( + AddrColorFormat format, AddrSurfaceNumber surfNum, AddrSurfaceSwap surfSwap, + const ADDR_FLT_32 comps[4], UINT_8 *pPixel) const; + + static VOID Flt32sToInt32s( + ADDR_FLT_32 value, UINT_32 bits, AddrNumberType numberType, UINT_32* pResult); + + static VOID Int32sToPixel( + UINT_32 numComps, UINT_32* pComps, UINT_32* pCompBits, UINT_32* pCompStart, + ADDR_COMPONENT_FLAGS properties, UINT_32 resultBits, UINT_8* pPixel); + + VOID PixGetColorCompInfo( + AddrColorFormat format, AddrSurfaceNumber number, AddrSurfaceSwap swap, + ADDR_PIXEL_FORMATINFO* pInfo) const; + + VOID PixGetDepthCompInfo( + AddrDepthFormat format, ADDR_PIXEL_FORMATINFO* pInfo) const; + + UINT_32 GetBitsPerPixel( + AddrFormat format, AddrElemMode* pElemMode, + UINT_32* pExpandX = NULL, UINT_32* pExpandY = NULL, UINT_32* pBitsUnused = NULL); + + static VOID SetClearComps( + ADDR_FLT_32 comps[4], BOOL_32 clearColor, BOOL_32 float32); + + VOID AdjustSurfaceInfo( + AddrElemMode elemMode, UINT_32 expandX, UINT_32 expandY, + UINT_32* pBpp, UINT_32* pBasePitch, UINT_32* pWidth, UINT_32* pHeight); + + VOID RestoreSurfaceInfo( + AddrElemMode elemMode, UINT_32 expandX, UINT_32 expandY, + UINT_32* pBpp, UINT_32* pWidth, UINT_32* pHeight); + + /// Checks if depth and stencil are planar inside a tile + BOOL_32 IsDepthStencilTilePlanar() + { + return (m_depthPlanarType == ADDR_DEPTH_PLANAR_R600) ? TRUE : FALSE; + } + + /// Sets m_configFlags, copied from AddrLib + VOID SetConfigFlags(ADDR_CONFIG_FLAGS flags) + { + m_configFlags = flags; + } + + static BOOL_32 IsCompressed(AddrFormat format); + static BOOL_32 IsBlockCompressed(AddrFormat format); + static BOOL_32 IsExpand3x(AddrFormat format); + +protected: + + static VOID GetCompBits( + UINT_32 c0, UINT_32 c1, UINT_32 c2, UINT_32 c3, + ADDR_PIXEL_FORMATINFO* pInfo, + AddrElemMode elemMode = ADDR_ROUND_BY_HALF); + + static VOID GetCompType( + AddrColorFormat format, AddrSurfaceNumber numType, + ADDR_PIXEL_FORMATINFO* pInfo); + + static VOID GetCompSwap( + AddrSurfaceSwap swap, ADDR_PIXEL_FORMATINFO* pInfo); + + static VOID SwapComps( + UINT_32 c0, UINT_32 c1, ADDR_PIXEL_FORMATINFO* pInfo); + +private: + + UINT_32 m_fp16ExportNorm; ///< If allow FP16 to be reported as EXPORT_NORM + AddrDepthPlanarType m_depthPlanarType; + + ADDR_CONFIG_FLAGS m_configFlags; ///< Copy of AddrLib's configFlags + AddrLib* const m_pAddrLib; ///< Pointer to parent addrlib instance +}; + +#endif + diff --git a/src/gallium/winsys/radeon/amdgpu/addrlib/core/addrlib.cpp b/src/gallium/winsys/radeon/amdgpu/addrlib/core/addrlib.cpp new file mode 100644 index 0000000..51b1eab --- /dev/null +++ b/src/gallium/winsys/radeon/amdgpu/addrlib/core/addrlib.cpp @@ -0,0 +1,4028 @@ +/* + * Copyright © 2014 Advanced Micro Devices, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES + * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NON-INFRINGEMENT. IN NO EVENT SHALL THE COPYRIGHT HOLDERS, AUTHORS + * AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + */ + +/** +*************************************************************************************************** +* @file addrlib.cpp +* @brief Contains the implementation for the AddrLib base class.. +*************************************************************************************************** +*/ + +#include "addrinterface.h" +#include "addrlib.h" +#include "addrcommon.h" + +#if defined(__APPLE__) + +UINT_32 div64_32(UINT_64 n, UINT_32 base) +{ + UINT_64 rem = n; + UINT_64 b = base; + UINT_64 res, d = 1; + UINT_32 high = rem >> 32; + + res = 0; + if (high >= base) + { + high /= base; + res = (UINT_64) high << 32; + rem -= (UINT_64) (high*base) << 32; + } + + while ((INT_64)b > 0 && b < rem) + { + b = b+b; + d = d+d; + } + + do + { + if (rem >= b) + { + rem -= b; + res += d; + } + b >>= 1; + d >>= 1; + } while (d); + + n = res; + return rem; +} + +extern "C" +UINT_32 __umoddi3(UINT_64 n, UINT_32 base) +{ + return div64_32(n, base); +} + +#endif // __APPLE__ + +/////////////////////////////////////////////////////////////////////////////////////////////////// +// Static Const Member +/////////////////////////////////////////////////////////////////////////////////////////////////// + +const AddrTileModeFlags AddrLib::m_modeFlags[ADDR_TM_COUNT] = +{// T L 1 2 3 P Pr B + {1, 1, 0, 0, 0, 0, 0, 0}, // ADDR_TM_LINEAR_GENERAL + {1, 1, 0, 0, 0, 0, 0, 0}, // ADDR_TM_LINEAR_ALIGNED + {1, 0, 1, 0, 0, 0, 0, 0}, // ADDR_TM_1D_TILED_THIN1 + {4, 0, 1, 0, 0, 0, 0, 0}, // ADDR_TM_1D_TILED_THICK + {1, 0, 0, 1, 0, 0, 0, 0}, // ADDR_TM_2D_TILED_THIN1 + {1, 0, 0, 1, 0, 0, 0, 0}, // ADDR_TM_2D_TILED_THIN2 + {1, 0, 0, 1, 0, 0, 0, 0}, // ADDR_TM_2D_TILED_THIN4 + {4, 0, 0, 1, 0, 0, 0, 0}, // ADDR_TM_2D_TILED_THICK + {1, 0, 0, 1, 0, 0, 0, 1}, // ADDR_TM_2B_TILED_THIN1 + {1, 0, 0, 1, 0, 0, 0, 1}, // ADDR_TM_2B_TILED_THIN2 + {1, 0, 0, 1, 0, 0, 0, 1}, // ADDR_TM_2B_TILED_THIN4 + {4, 0, 0, 1, 0, 0, 0, 1}, // ADDR_TM_2B_TILED_THICK + {1, 0, 0, 1, 1, 0, 0, 0}, // ADDR_TM_3D_TILED_THIN1 + {4, 0, 0, 1, 1, 0, 0, 0}, // ADDR_TM_3D_TILED_THICK + {1, 0, 0, 1, 1, 0, 0, 1}, // ADDR_TM_3B_TILED_THIN1 + {4, 0, 0, 1, 1, 0, 0, 1}, // ADDR_TM_3B_TILED_THICK + {8, 0, 0, 1, 0, 0, 0, 0}, // ADDR_TM_2D_TILED_XTHICK + {8, 0, 0, 1, 1, 0, 0, 0}, // ADDR_TM_3D_TILED_XTHICK + {1, 0, 0, 0, 0, 0, 0, 0}, // ADDR_TM_POWER_SAVE + {1, 0, 0, 1, 0, 1, 1, 0}, // ADDR_TM_PRT_TILED_THIN1 + {1, 0, 0, 1, 0, 1, 0, 0}, // ADDR_TM_PRT_2D_TILED_THIN1 + {1, 0, 0, 1, 1, 1, 0, 0}, // ADDR_TM_PRT_3D_TILED_THIN1 + {4, 0, 0, 1, 0, 1, 1, 0}, // ADDR_TM_PRT_TILED_THICK + {4, 0, 0, 1, 0, 1, 0, 0}, // ADDR_TM_PRT_2D_TILED_THICK + {4, 0, 0, 1, 1, 1, 0, 0}, // ADDR_TM_PRT_3D_TILED_THICK +}; + +/////////////////////////////////////////////////////////////////////////////////////////////////// +// Constructor/Destructor +/////////////////////////////////////////////////////////////////////////////////////////////////// + +/** +*************************************************************************************************** +* AddrLib::AddrLib +* +* @brief +* Constructor for the AddrLib class +* +*************************************************************************************************** +*/ +AddrLib::AddrLib() : + m_class(BASE_ADDRLIB), + m_chipFamily(ADDR_CHIP_FAMILY_IVLD), + m_chipRevision(0), + m_version(ADDRLIB_VERSION), + m_pipes(0), + m_banks(0), + m_pipeInterleaveBytes(0), + m_rowSize(0), + m_minPitchAlignPixels(1), + m_maxSamples(8), + m_pElemLib(NULL) +{ + m_configFlags.value = 0; +} + +/** +*************************************************************************************************** +* AddrLib::AddrLib +* +* @brief +* Constructor for the AddrLib class with hClient as parameter +* +*************************************************************************************************** +*/ +AddrLib::AddrLib(const AddrClient* pClient) : + AddrObject(pClient), + m_class(BASE_ADDRLIB), + m_chipFamily(ADDR_CHIP_FAMILY_IVLD), + m_chipRevision(0), + m_version(ADDRLIB_VERSION), + m_pipes(0), + m_banks(0), + m_pipeInterleaveBytes(0), + m_rowSize(0), + m_minPitchAlignPixels(1), + m_maxSamples(8), + m_pElemLib(NULL) +{ + m_configFlags.value = 0; +} + +/** +*************************************************************************************************** +* AddrLib::~AddrLib +* +* @brief +* Destructor for the AddrLib class +* +*************************************************************************************************** +*/ +AddrLib::~AddrLib() +{ + if (m_pElemLib) + { + delete m_pElemLib; + } +} + + + +/////////////////////////////////////////////////////////////////////////////////////////////////// +// Initialization/Helper +/////////////////////////////////////////////////////////////////////////////////////////////////// + +/** +*************************************************************************************************** +* AddrLib::Create +* +* @brief +* Creates and initializes AddrLib object. +* +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE AddrLib::Create( + const ADDR_CREATE_INPUT* pCreateIn, ///< [in] pointer to ADDR_CREATE_INPUT + ADDR_CREATE_OUTPUT* pCreateOut) ///< [out] pointer to ADDR_CREATE_OUTPUT +{ + AddrLib* pLib = NULL; + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + if (pCreateIn->createFlags.fillSizeFields == TRUE) + { + if ((pCreateIn->size != sizeof(ADDR_CREATE_INPUT)) || + (pCreateOut->size != sizeof(ADDR_CREATE_OUTPUT))) + { + returnCode = ADDR_PARAMSIZEMISMATCH; + } + } + + if ((returnCode == ADDR_OK) && + (pCreateIn->callbacks.allocSysMem != NULL) && + (pCreateIn->callbacks.freeSysMem != NULL)) + { + AddrClient client = { + pCreateIn->hClient, + pCreateIn->callbacks + }; + + switch (pCreateIn->chipEngine) + { + case CIASICIDGFXENGINE_SOUTHERNISLAND: + switch (pCreateIn->chipFamily) + { + case FAMILY_SI: + pLib = AddrSIHwlInit(&client); + break; + case FAMILY_VI: + case FAMILY_CZ: // VI based fusion(carrizo) + case FAMILY_CI: + case FAMILY_KV: // CI based fusion + pLib = AddrCIHwlInit(&client); + break; + default: + ADDR_ASSERT_ALWAYS(); + break; + } + break; + default: + ADDR_ASSERT_ALWAYS(); + break; + } + } + + if ((pLib != NULL)) + { + BOOL_32 initValid; + + // Pass createFlags to configFlags first since these flags may be overwritten + pLib->m_configFlags.noCubeMipSlicesPad = pCreateIn->createFlags.noCubeMipSlicesPad; + pLib->m_configFlags.fillSizeFields = pCreateIn->createFlags.fillSizeFields; + pLib->m_configFlags.useTileIndex = pCreateIn->createFlags.useTileIndex; + pLib->m_configFlags.useCombinedSwizzle = pCreateIn->createFlags.useCombinedSwizzle; + pLib->m_configFlags.checkLast2DLevel = pCreateIn->createFlags.checkLast2DLevel; + pLib->m_configFlags.useHtileSliceAlign = pCreateIn->createFlags.useHtileSliceAlign; + pLib->m_configFlags.degradeBaseLevel = pCreateIn->createFlags.degradeBaseLevel; + pLib->m_configFlags.allowLargeThickTile = pCreateIn->createFlags.allowLargeThickTile; + + pLib->SetAddrChipFamily(pCreateIn->chipFamily, pCreateIn->chipRevision); + + pLib->SetMinPitchAlignPixels(pCreateIn->minPitchAlignPixels); + + // Global parameters initialized and remaining configFlags bits are set as well + initValid = pLib->HwlInitGlobalParams(pCreateIn); + + if (initValid) + { + pLib->m_pElemLib = AddrElemLib::Create(pLib); + } + else + { + pLib->m_pElemLib = NULL; // Don't go on allocating element lib + returnCode = ADDR_INVALIDGBREGVALUES; + } + + if (pLib->m_pElemLib == NULL) + { + delete pLib; + pLib = NULL; + ADDR_ASSERT_ALWAYS(); + } + else + { + pLib->m_pElemLib->SetConfigFlags(pLib->m_configFlags); + } + } + + pCreateOut->hLib = pLib; + + if ((pLib == NULL) && + (returnCode == ADDR_OK)) + { + // Unknown failures, we return the general error code + returnCode = ADDR_ERROR; + } + + return returnCode; +} + +/** +*************************************************************************************************** +* AddrLib::SetAddrChipFamily +* +* @brief +* Convert familyID defined in atiid.h to AddrChipFamily and set m_chipFamily/m_chipRevision +* @return +* N/A +*************************************************************************************************** +*/ +VOID AddrLib::SetAddrChipFamily( + UINT_32 uChipFamily, ///< [in] chip family defined in atiih.h + UINT_32 uChipRevision) ///< [in] chip revision defined in "asic_family"_id.h +{ + AddrChipFamily family = ADDR_CHIP_FAMILY_IVLD; + + family = HwlConvertChipFamily(uChipFamily, uChipRevision); + + ADDR_ASSERT(family != ADDR_CHIP_FAMILY_IVLD); + + m_chipFamily = family; + m_chipRevision = uChipRevision; +} + +/** +*************************************************************************************************** +* AddrLib::SetMinPitchAlignPixels +* +* @brief +* Set m_minPitchAlignPixels with input param +* +* @return +* N/A +*************************************************************************************************** +*/ +VOID AddrLib::SetMinPitchAlignPixels( + UINT_32 minPitchAlignPixels) ///< [in] minmum pitch alignment in pixels +{ + m_minPitchAlignPixels = (minPitchAlignPixels == 0)? 1 : minPitchAlignPixels; +} + +/** +*************************************************************************************************** +* AddrLib::GetAddrLib +* +* @brief +* Get AddrLib pointer +* +* @return +* An AddrLib class pointer +*************************************************************************************************** +*/ +AddrLib * AddrLib::GetAddrLib( + ADDR_HANDLE hLib) ///< [in] handle of ADDR_HANDLE +{ + return static_cast(hLib); +} + + + +/////////////////////////////////////////////////////////////////////////////////////////////////// +// Surface Methods +/////////////////////////////////////////////////////////////////////////////////////////////////// + + +/** +*************************************************************************************************** +* AddrLib::ComputeSurfaceInfo +* +* @brief +* Interface function stub of AddrComputeSurfaceInfo. +* +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE AddrLib::ComputeSurfaceInfo( + const ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn, ///< [in] input structure + ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pOut ///< [out] output structure + ) const +{ + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + if (GetFillSizeFieldsFlags() == TRUE) + { + if ((pIn->size != sizeof(ADDR_COMPUTE_SURFACE_INFO_INPUT)) || + (pOut->size != sizeof(ADDR_COMPUTE_SURFACE_INFO_OUTPUT))) + { + returnCode = ADDR_PARAMSIZEMISMATCH; + } + } + + // We suggest client do sanity check but a check here is also good + if (pIn->bpp > 128) + { + returnCode = ADDR_INVALIDPARAMS; + } + + // Thick modes don't support multisample + if (ComputeSurfaceThickness(pIn->tileMode) > 1 && pIn->numSamples > 1) + { + returnCode = ADDR_INVALIDPARAMS; + } + + if (returnCode == ADDR_OK) + { + // Get a local copy of input structure and only reference pIn for unadjusted values + ADDR_COMPUTE_SURFACE_INFO_INPUT localIn = *pIn; + ADDR_TILEINFO tileInfoNull = {0}; + + if (UseTileInfo()) + { + // If the original input has a valid ADDR_TILEINFO pointer then copy its contents. + // Otherwise the default 0's in tileInfoNull are used. + if (pIn->pTileInfo) + { + tileInfoNull = *pIn->pTileInfo; + } + localIn.pTileInfo = &tileInfoNull; + } + + localIn.numSamples = pIn->numSamples == 0 ? 1 : pIn->numSamples; + + // Do mipmap check first + // If format is BCn, pre-pad dimension to power-of-two according to HWL + ComputeMipLevel(&localIn); + + if (m_configFlags.checkLast2DLevel) + { + // Save this level's original height in pixels + pOut->height = pIn->height; + } + + UINT_32 expandX = 1; + UINT_32 expandY = 1; + AddrElemMode elemMode; + + // Save outputs that may not go through HWL + pOut->pixelBits = localIn.bpp; + pOut->numSamples = localIn.numSamples; + pOut->last2DLevel = FALSE; + +#if !ALT_TEST + if (localIn.numSamples > 1) + { + ADDR_ASSERT(localIn.mipLevel == 0); + } +#endif + + if (localIn.format != ADDR_FMT_INVALID) // Set format to INVALID will skip this conversion + { + // Get compression/expansion factors and element mode + // (which indicates compression/expansion + localIn.bpp = GetElemLib()->GetBitsPerPixel(localIn.format, + &elemMode, + &expandX, + &expandY); + + // Special flag for 96 bit surface. 96 (or 48 if we support) bit surface's width is + // pre-multiplied by 3 and bpp is divided by 3. So pitch alignment for linear- + // aligned does not meet 64-pixel in real. We keep special handling in hwl since hw + // restrictions are different. + // Also Mip 1+ needs an element pitch of 32 bits so we do not need this workaround + // but we use this flag to skip RestoreSurfaceInfo below + + if ((elemMode == ADDR_EXPANDED) && + (expandX > 1)) + { + ADDR_ASSERT(localIn.tileMode == ADDR_TM_LINEAR_ALIGNED || localIn.height == 1); + } + + GetElemLib()->AdjustSurfaceInfo(elemMode, + expandX, + expandY, + &localIn.bpp, + &localIn.basePitch, + &localIn.width, + &localIn.height); + + // Overwrite these parameters if we have a valid format + } + else if (localIn.bpp != 0) + { + localIn.width = (localIn.width != 0) ? localIn.width : 1; + localIn.height = (localIn.height != 0) ? localIn.height : 1; + } + else // Rule out some invalid parameters + { + ADDR_ASSERT_ALWAYS(); + + returnCode = ADDR_INVALIDPARAMS; + } + + // Check mipmap after surface expansion + if (returnCode == ADDR_OK) + { + returnCode = PostComputeMipLevel(&localIn, pOut); + } + + if (returnCode == ADDR_OK) + { + if (UseTileIndex(localIn.tileIndex)) + { + // Make sure pTileInfo is not NULL + ADDR_ASSERT(localIn.pTileInfo); + + UINT_32 numSamples = GetNumFragments(localIn.numSamples, localIn.numFrags); + + INT_32 macroModeIndex = TileIndexNoMacroIndex; + + if (localIn.tileIndex != TileIndexLinearGeneral) + { + // Try finding a macroModeIndex + macroModeIndex = HwlComputeMacroModeIndex(localIn.tileIndex, + localIn.flags, + localIn.bpp, + numSamples, + localIn.pTileInfo, + &localIn.tileMode, + &localIn.tileType); + } + + // If macroModeIndex is not needed, then call HwlSetupTileCfg to get tile info + if (macroModeIndex == TileIndexNoMacroIndex) + { + returnCode = HwlSetupTileCfg(localIn.tileIndex, macroModeIndex, + localIn.pTileInfo, + &localIn.tileMode, &localIn.tileType); + } + // If macroModeIndex is invalid, then assert this is not macro tiled + else if (macroModeIndex == TileIndexInvalid) + { + ADDR_ASSERT(!IsMacroTiled(localIn.tileMode)); + } + } + } + + if (returnCode == ADDR_OK) + { + AddrTileMode tileMode = localIn.tileMode; + AddrTileType tileType = localIn.tileType; + + // HWL layer may override tile mode if necessary + if (HwlOverrideTileMode(&localIn, &tileMode, &tileType)) + { + localIn.tileMode = tileMode; + localIn.tileType = tileType; + } + // Degrade base level if applicable + if (DegradeBaseLevel(&localIn, &tileMode)) + { + localIn.tileMode = tileMode; + } + } + + // Call main function to compute surface info + if (returnCode == ADDR_OK) + { + returnCode = HwlComputeSurfaceInfo(&localIn, pOut); + } + + if (returnCode == ADDR_OK) + { + // Since bpp might be changed we just pass it through + pOut->bpp = localIn.bpp; + + // Also original width/height/bpp + pOut->pixelPitch = pOut->pitch; + pOut->pixelHeight = pOut->height; + +#if DEBUG + if (localIn.flags.display) + { + ADDR_ASSERT((pOut->pitchAlign % 32) == 0); + } +#endif //DEBUG + + if (localIn.format != ADDR_FMT_INVALID) + { + // + // 96 bits surface of level 1+ requires element pitch of 32 bits instead + // In hwl function we skip multiplication of 3 then we should skip division of 3 + // We keep pitch that represents 32 bit element instead of 96 bits since we + // will get an odd number if divided by 3. + // + if (!((expandX == 3) && (localIn.mipLevel > 0))) + { + + GetElemLib()->RestoreSurfaceInfo(elemMode, + expandX, + expandY, + &localIn.bpp, + &pOut->pixelPitch, + &pOut->pixelHeight); + } + } + + if (localIn.flags.qbStereo) + { + if (pOut->pStereoInfo) + { + ComputeQbStereoInfo(pOut); + } + } + + if (localIn.flags.volume) // For volume sliceSize equals to all z-slices + { + pOut->sliceSize = pOut->surfSize; + } + else // For array: sliceSize is likely to have slice-padding (the last one) + { + pOut->sliceSize = pOut->surfSize / pOut->depth; + + // array or cubemap + if (pIn->numSlices > 1) + { + // If this is the last slice then add the padding size to this slice + if (pIn->slice == (pIn->numSlices - 1)) + { + pOut->sliceSize += pOut->sliceSize * (pOut->depth - pIn->numSlices); + } + else if (m_configFlags.checkLast2DLevel) + { + // Reset last2DLevel flag if this is not the last array slice + pOut->last2DLevel = FALSE; + } + } + } + + pOut->pitchTileMax = pOut->pitch / 8 - 1; + pOut->heightTileMax = pOut->height / 8 - 1; + pOut->sliceTileMax = pOut->pitch * pOut->height / 64 - 1; + } + } + + return returnCode; +} + +/** +*************************************************************************************************** +* AddrLib::ComputeSurfaceInfo +* +* @brief +* Interface function stub of AddrComputeSurfaceInfo. +* +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE AddrLib::ComputeSurfaceAddrFromCoord( + const ADDR_COMPUTE_SURFACE_ADDRFROMCOORD_INPUT* pIn, ///< [in] input structure + ADDR_COMPUTE_SURFACE_ADDRFROMCOORD_OUTPUT* pOut ///< [out] output structure + ) const +{ + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + if (GetFillSizeFieldsFlags() == TRUE) + { + if ((pIn->size != sizeof(ADDR_COMPUTE_SURFACE_ADDRFROMCOORD_INPUT)) || + (pOut->size != sizeof(ADDR_COMPUTE_SURFACE_ADDRFROMCOORD_OUTPUT))) + { + returnCode = ADDR_PARAMSIZEMISMATCH; + } + } + + if (returnCode == ADDR_OK) + { + ADDR_TILEINFO tileInfoNull; + ADDR_COMPUTE_SURFACE_ADDRFROMCOORD_INPUT input; + + if (UseTileIndex(pIn->tileIndex)) + { + input = *pIn; + // Use temp tile info for calcalation + input.pTileInfo = &tileInfoNull; + + const ADDR_SURFACE_FLAGS flags = {{0}}; + UINT_32 numSamples = GetNumFragments(pIn->numSamples, pIn->numFrags); + + // Try finding a macroModeIndex + INT_32 macroModeIndex = HwlComputeMacroModeIndex(input.tileIndex, + flags, + input.bpp, + numSamples, + input.pTileInfo, + &input.tileMode, + &input.tileType); + + // If macroModeIndex is not needed, then call HwlSetupTileCfg to get tile info + if (macroModeIndex == TileIndexNoMacroIndex) + { + returnCode = HwlSetupTileCfg(input.tileIndex, macroModeIndex, + input.pTileInfo, &input.tileMode, &input.tileType); + } + // If macroModeIndex is invalid, then assert this is not macro tiled + else if (macroModeIndex == TileIndexInvalid) + { + ADDR_ASSERT(!IsMacroTiled(input.tileMode)); + } + + // Change the input structure + pIn = &input; + } + + if (returnCode == ADDR_OK) + { + returnCode = HwlComputeSurfaceAddrFromCoord(pIn, pOut); + + if (returnCode == ADDR_OK) + { + pOut->prtBlockIndex = static_cast(pOut->addr / (64 * 1024)); + } + } + } + + return returnCode; +} + +/** +*************************************************************************************************** +* AddrLib::ComputeSurfaceCoordFromAddr +* +* @brief +* Interface function stub of ComputeSurfaceCoordFromAddr. +* +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE AddrLib::ComputeSurfaceCoordFromAddr( + const ADDR_COMPUTE_SURFACE_COORDFROMADDR_INPUT* pIn, ///< [in] input structure + ADDR_COMPUTE_SURFACE_COORDFROMADDR_OUTPUT* pOut ///< [out] output structure + ) const +{ + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + if (GetFillSizeFieldsFlags() == TRUE) + { + if ((pIn->size != sizeof(ADDR_COMPUTE_SURFACE_COORDFROMADDR_INPUT)) || + (pOut->size != sizeof(ADDR_COMPUTE_SURFACE_COORDFROMADDR_OUTPUT))) + { + returnCode = ADDR_PARAMSIZEMISMATCH; + } + } + + if (returnCode == ADDR_OK) + { + ADDR_TILEINFO tileInfoNull; + ADDR_COMPUTE_SURFACE_COORDFROMADDR_INPUT input; + + if (UseTileIndex(pIn->tileIndex)) + { + input = *pIn; + // Use temp tile info for calcalation + input.pTileInfo = &tileInfoNull; + + const ADDR_SURFACE_FLAGS flags = {{0}}; + UINT_32 numSamples = GetNumFragments(pIn->numSamples, pIn->numFrags); + + // Try finding a macroModeIndex + INT_32 macroModeIndex = HwlComputeMacroModeIndex(input.tileIndex, + flags, + input.bpp, + numSamples, + input.pTileInfo, + &input.tileMode, + &input.tileType); + + // If macroModeIndex is not needed, then call HwlSetupTileCfg to get tile info + if (macroModeIndex == TileIndexNoMacroIndex) + { + returnCode = HwlSetupTileCfg(input.tileIndex, macroModeIndex, + input.pTileInfo, &input.tileMode, &input.tileType); + } + // If macroModeIndex is invalid, then assert this is not macro tiled + else if (macroModeIndex == TileIndexInvalid) + { + ADDR_ASSERT(!IsMacroTiled(input.tileMode)); + } + + // Change the input structure + pIn = &input; + } + + if (returnCode == ADDR_OK) + { + returnCode = HwlComputeSurfaceCoordFromAddr(pIn, pOut); + } + } + + return returnCode; +} + +/** +*************************************************************************************************** +* AddrLib::ComputeSliceTileSwizzle +* +* @brief +* Interface function stub of ComputeSliceTileSwizzle. +* +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE AddrLib::ComputeSliceTileSwizzle( + const ADDR_COMPUTE_SLICESWIZZLE_INPUT* pIn, ///< [in] input structure + ADDR_COMPUTE_SLICESWIZZLE_OUTPUT* pOut ///< [out] output structure + ) const +{ + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + if (GetFillSizeFieldsFlags() == TRUE) + { + if ((pIn->size != sizeof(ADDR_COMPUTE_SLICESWIZZLE_INPUT)) || + (pOut->size != sizeof(ADDR_COMPUTE_SLICESWIZZLE_OUTPUT))) + { + returnCode = ADDR_PARAMSIZEMISMATCH; + } + } + + if (returnCode == ADDR_OK) + { + ADDR_TILEINFO tileInfoNull; + ADDR_COMPUTE_SLICESWIZZLE_INPUT input; + + if (UseTileIndex(pIn->tileIndex)) + { + input = *pIn; + // Use temp tile info for calcalation + input.pTileInfo = &tileInfoNull; + + returnCode = HwlSetupTileCfg(input.tileIndex, input.macroModeIndex, + input.pTileInfo, &input.tileMode); + // Change the input structure + pIn = &input; + } + + if (returnCode == ADDR_OK) + { + returnCode = HwlComputeSliceTileSwizzle(pIn, pOut); + } + } + + return returnCode; +} + +/** +*************************************************************************************************** +* AddrLib::ExtractBankPipeSwizzle +* +* @brief +* Interface function stub of AddrExtractBankPipeSwizzle. +* +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE AddrLib::ExtractBankPipeSwizzle( + const ADDR_EXTRACT_BANKPIPE_SWIZZLE_INPUT* pIn, ///< [in] input structure + ADDR_EXTRACT_BANKPIPE_SWIZZLE_OUTPUT* pOut ///< [out] output structure + ) const +{ + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + if (GetFillSizeFieldsFlags() == TRUE) + { + if ((pIn->size != sizeof(ADDR_EXTRACT_BANKPIPE_SWIZZLE_INPUT)) || + (pOut->size != sizeof(ADDR_EXTRACT_BANKPIPE_SWIZZLE_OUTPUT))) + { + returnCode = ADDR_PARAMSIZEMISMATCH; + } + } + + if (returnCode == ADDR_OK) + { + ADDR_TILEINFO tileInfoNull; + ADDR_EXTRACT_BANKPIPE_SWIZZLE_INPUT input; + + if (UseTileIndex(pIn->tileIndex)) + { + input = *pIn; + // Use temp tile info for calcalation + input.pTileInfo = &tileInfoNull; + + returnCode = HwlSetupTileCfg(input.tileIndex, input.macroModeIndex, input.pTileInfo); + // Change the input structure + pIn = &input; + } + + if (returnCode == ADDR_OK) + { + returnCode = HwlExtractBankPipeSwizzle(pIn, pOut); + } + } + + return returnCode; +} + +/** +*************************************************************************************************** +* AddrLib::CombineBankPipeSwizzle +* +* @brief +* Interface function stub of AddrCombineBankPipeSwizzle. +* +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE AddrLib::CombineBankPipeSwizzle( + const ADDR_COMBINE_BANKPIPE_SWIZZLE_INPUT* pIn, ///< [in] input structure + ADDR_COMBINE_BANKPIPE_SWIZZLE_OUTPUT* pOut ///< [out] output structure + ) const +{ + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + if (GetFillSizeFieldsFlags() == TRUE) + { + if ((pIn->size != sizeof(ADDR_COMPUTE_FMASK_INFO_INPUT)) || + (pOut->size != sizeof(ADDR_COMPUTE_FMASK_INFO_OUTPUT))) + { + returnCode = ADDR_PARAMSIZEMISMATCH; + } + } + + if (returnCode == ADDR_OK) + { + ADDR_TILEINFO tileInfoNull; + ADDR_COMBINE_BANKPIPE_SWIZZLE_INPUT input; + + if (UseTileIndex(pIn->tileIndex)) + { + input = *pIn; + // Use temp tile info for calcalation + input.pTileInfo = &tileInfoNull; + + returnCode = HwlSetupTileCfg(input.tileIndex, input.macroModeIndex, input.pTileInfo); + // Change the input structure + pIn = &input; + } + + if (returnCode == ADDR_OK) + { + returnCode = HwlCombineBankPipeSwizzle(pIn->bankSwizzle, + pIn->pipeSwizzle, + pIn->pTileInfo, + pIn->baseAddr, + &pOut->tileSwizzle); + } + } + + return returnCode; +} + +/** +*************************************************************************************************** +* AddrLib::ComputeBaseSwizzle +* +* @brief +* Interface function stub of AddrCompueBaseSwizzle. +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE AddrLib::ComputeBaseSwizzle( + const ADDR_COMPUTE_BASE_SWIZZLE_INPUT* pIn, + ADDR_COMPUTE_BASE_SWIZZLE_OUTPUT* pOut) const +{ + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + if (GetFillSizeFieldsFlags() == TRUE) + { + if ((pIn->size != sizeof(ADDR_COMPUTE_BASE_SWIZZLE_INPUT)) || + (pOut->size != sizeof(ADDR_COMPUTE_BASE_SWIZZLE_OUTPUT))) + { + returnCode = ADDR_PARAMSIZEMISMATCH; + } + } + + if (returnCode == ADDR_OK) + { + ADDR_TILEINFO tileInfoNull; + ADDR_COMPUTE_BASE_SWIZZLE_INPUT input; + + if (UseTileIndex(pIn->tileIndex)) + { + input = *pIn; + // Use temp tile info for calcalation + input.pTileInfo = &tileInfoNull; + + returnCode = HwlSetupTileCfg(input.tileIndex, input.macroModeIndex, input.pTileInfo); + // Change the input structure + pIn = &input; + } + + if (returnCode == ADDR_OK) + { + if (IsMacroTiled(pIn->tileMode)) + { + returnCode = HwlComputeBaseSwizzle(pIn, pOut); + } + else + { + pOut->tileSwizzle = 0; + } + } + } + + return returnCode; +} + +/** +*************************************************************************************************** +* AddrLib::ComputeFmaskInfo +* +* @brief +* Interface function stub of ComputeFmaskInfo. +* +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE AddrLib::ComputeFmaskInfo( + const ADDR_COMPUTE_FMASK_INFO_INPUT* pIn, ///< [in] input structure + ADDR_COMPUTE_FMASK_INFO_OUTPUT* pOut ///< [out] output structure + ) +{ + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + if (GetFillSizeFieldsFlags() == TRUE) + { + if ((pIn->size != sizeof(ADDR_COMPUTE_FMASK_INFO_INPUT)) || + (pOut->size != sizeof(ADDR_COMPUTE_FMASK_INFO_OUTPUT))) + { + returnCode = ADDR_PARAMSIZEMISMATCH; + } + } + + // No thick MSAA + if (ComputeSurfaceThickness(pIn->tileMode) > 1) + { + returnCode = ADDR_INVALIDPARAMS; + } + + if (returnCode == ADDR_OK) + { + ADDR_TILEINFO tileInfoNull; + ADDR_COMPUTE_FMASK_INFO_INPUT input; + + if (UseTileIndex(pIn->tileIndex)) + { + input = *pIn; + + if (pOut->pTileInfo) + { + // Use temp tile info for calcalation + input.pTileInfo = pOut->pTileInfo; + } + else + { + input.pTileInfo = &tileInfoNull; + } + + ADDR_SURFACE_FLAGS flags = {{0}}; + flags.fmask = 1; + + // Try finding a macroModeIndex + INT_32 macroModeIndex = HwlComputeMacroModeIndex(pIn->tileIndex, + flags, + HwlComputeFmaskBits(pIn, NULL), + pIn->numSamples, + input.pTileInfo, + &input.tileMode); + + // If macroModeIndex is not needed, then call HwlSetupTileCfg to get tile info + if (macroModeIndex == TileIndexNoMacroIndex) + { + returnCode = HwlSetupTileCfg(input.tileIndex, macroModeIndex, + input.pTileInfo, &input.tileMode); + } + + ADDR_ASSERT(macroModeIndex != TileIndexInvalid); + + // Change the input structure + pIn = &input; + } + + if (returnCode == ADDR_OK) + { + if (pIn->numSamples > 1) + { + returnCode = HwlComputeFmaskInfo(pIn, pOut); + } + else + { + memset(pOut, 0, sizeof(ADDR_COMPUTE_FMASK_INFO_OUTPUT)); + + returnCode = ADDR_INVALIDPARAMS; + } + } + } + + return returnCode; +} + +/** +*************************************************************************************************** +* AddrLib::ComputeFmaskAddrFromCoord +* +* @brief +* Interface function stub of ComputeFmaskAddrFromCoord. +* +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE AddrLib::ComputeFmaskAddrFromCoord( + const ADDR_COMPUTE_FMASK_ADDRFROMCOORD_INPUT* pIn, ///< [in] input structure + ADDR_COMPUTE_FMASK_ADDRFROMCOORD_OUTPUT* pOut ///< [out] output structure + ) const +{ + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + if (GetFillSizeFieldsFlags() == TRUE) + { + if ((pIn->size != sizeof(ADDR_COMPUTE_FMASK_ADDRFROMCOORD_INPUT)) || + (pOut->size != sizeof(ADDR_COMPUTE_FMASK_ADDRFROMCOORD_OUTPUT))) + { + returnCode = ADDR_PARAMSIZEMISMATCH; + } + } + + if (returnCode == ADDR_OK) + { + ADDR_ASSERT(pIn->numSamples > 1); + + if (pIn->numSamples > 1) + { + returnCode = HwlComputeFmaskAddrFromCoord(pIn, pOut); + } + else + { + returnCode = ADDR_INVALIDPARAMS; + } + } + + return returnCode; +} + +/** +*************************************************************************************************** +* AddrLib::ComputeFmaskCoordFromAddr +* +* @brief +* Interface function stub of ComputeFmaskAddrFromCoord. +* +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE AddrLib::ComputeFmaskCoordFromAddr( + const ADDR_COMPUTE_FMASK_COORDFROMADDR_INPUT* pIn, ///< [in] input structure + ADDR_COMPUTE_FMASK_COORDFROMADDR_OUTPUT* pOut ///< [out] output structure + ) const +{ + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + if (GetFillSizeFieldsFlags() == TRUE) + { + if ((pIn->size != sizeof(ADDR_COMPUTE_FMASK_COORDFROMADDR_INPUT)) || + (pOut->size != sizeof(ADDR_COMPUTE_FMASK_COORDFROMADDR_OUTPUT))) + { + returnCode = ADDR_PARAMSIZEMISMATCH; + } + } + + if (returnCode == ADDR_OK) + { + ADDR_ASSERT(pIn->numSamples > 1); + + if (pIn->numSamples > 1) + { + returnCode = HwlComputeFmaskCoordFromAddr(pIn, pOut); + } + else + { + returnCode = ADDR_INVALIDPARAMS; + } + } + + return returnCode; +} + +/** +*************************************************************************************************** +* AddrLib::ConvertTileInfoToHW +* +* @brief +* Convert tile info from real value to HW register value in HW layer +* +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE AddrLib::ConvertTileInfoToHW( + const ADDR_CONVERT_TILEINFOTOHW_INPUT* pIn, ///< [in] input structure + ADDR_CONVERT_TILEINFOTOHW_OUTPUT* pOut ///< [out] output structure + ) const +{ + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + if (GetFillSizeFieldsFlags() == TRUE) + { + if ((pIn->size != sizeof(ADDR_CONVERT_TILEINFOTOHW_INPUT)) || + (pOut->size != sizeof(ADDR_CONVERT_TILEINFOTOHW_OUTPUT))) + { + returnCode = ADDR_PARAMSIZEMISMATCH; + } + } + + if (returnCode == ADDR_OK) + { + ADDR_TILEINFO tileInfoNull; + ADDR_CONVERT_TILEINFOTOHW_INPUT input; + // if pIn->reverse is TRUE, indices are ignored + if (pIn->reverse == FALSE && UseTileIndex(pIn->tileIndex)) + { + input = *pIn; + input.pTileInfo = &tileInfoNull; + + returnCode = HwlSetupTileCfg(input.tileIndex, input.macroModeIndex, input.pTileInfo); + + pIn = &input; + } + + if (returnCode == ADDR_OK) + { + returnCode = HwlConvertTileInfoToHW(pIn, pOut); + } + } + + return returnCode; +} + +/** +*************************************************************************************************** +* AddrLib::ConvertTileIndex +* +* @brief +* Convert tile index to tile mode/type/info +* +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE AddrLib::ConvertTileIndex( + const ADDR_CONVERT_TILEINDEX_INPUT* pIn, ///< [in] input structure + ADDR_CONVERT_TILEINDEX_OUTPUT* pOut ///< [out] output structure + ) const +{ + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + if (GetFillSizeFieldsFlags() == TRUE) + { + if ((pIn->size != sizeof(ADDR_CONVERT_TILEINDEX_INPUT)) || + (pOut->size != sizeof(ADDR_CONVERT_TILEINDEX_OUTPUT))) + { + returnCode = ADDR_PARAMSIZEMISMATCH; + } + } + + if (returnCode == ADDR_OK) + { + + returnCode = HwlSetupTileCfg(pIn->tileIndex, pIn->macroModeIndex, + pOut->pTileInfo, &pOut->tileMode, &pOut->tileType); + + if (returnCode == ADDR_OK && pIn->tileInfoHw) + { + ADDR_CONVERT_TILEINFOTOHW_INPUT hwInput = {0}; + ADDR_CONVERT_TILEINFOTOHW_OUTPUT hwOutput = {0}; + + hwInput.pTileInfo = pOut->pTileInfo; + hwInput.tileIndex = -1; + hwOutput.pTileInfo = pOut->pTileInfo; + + returnCode = HwlConvertTileInfoToHW(&hwInput, &hwOutput); + } + } + + return returnCode; +} + +/** +*************************************************************************************************** +* AddrLib::ConvertTileIndex1 +* +* @brief +* Convert tile index to tile mode/type/info +* +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE AddrLib::ConvertTileIndex1( + const ADDR_CONVERT_TILEINDEX1_INPUT* pIn, ///< [in] input structure + ADDR_CONVERT_TILEINDEX_OUTPUT* pOut ///< [out] output structure + ) const +{ + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + if (GetFillSizeFieldsFlags() == TRUE) + { + if ((pIn->size != sizeof(ADDR_CONVERT_TILEINDEX1_INPUT)) || + (pOut->size != sizeof(ADDR_CONVERT_TILEINDEX_OUTPUT))) + { + returnCode = ADDR_PARAMSIZEMISMATCH; + } + } + + if (returnCode == ADDR_OK) + { + ADDR_SURFACE_FLAGS flags = {{0}}; + + HwlComputeMacroModeIndex(pIn->tileIndex, flags, pIn->bpp, pIn->numSamples, + pOut->pTileInfo, &pOut->tileMode, &pOut->tileType); + + if (pIn->tileInfoHw) + { + ADDR_CONVERT_TILEINFOTOHW_INPUT hwInput = {0}; + ADDR_CONVERT_TILEINFOTOHW_OUTPUT hwOutput = {0}; + + hwInput.pTileInfo = pOut->pTileInfo; + hwInput.tileIndex = -1; + hwOutput.pTileInfo = pOut->pTileInfo; + + returnCode = HwlConvertTileInfoToHW(&hwInput, &hwOutput); + } + } + + return returnCode; +} + +/** +*************************************************************************************************** +* AddrLib::GetTileIndex +* +* @brief +* Get tile index from tile mode/type/info +* +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE AddrLib::GetTileIndex( + const ADDR_GET_TILEINDEX_INPUT* pIn, ///< [in] input structure + ADDR_GET_TILEINDEX_OUTPUT* pOut ///< [out] output structure + ) const +{ + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + if (GetFillSizeFieldsFlags() == TRUE) + { + if ((pIn->size != sizeof(ADDR_GET_TILEINDEX_INPUT)) || + (pOut->size != sizeof(ADDR_GET_TILEINDEX_OUTPUT))) + { + returnCode = ADDR_PARAMSIZEMISMATCH; + } + } + + if (returnCode == ADDR_OK) + { + returnCode = HwlGetTileIndex(pIn, pOut); + } + + return returnCode; +} + +/** +*************************************************************************************************** +* AddrLib::ComputeSurfaceThickness +* +* @brief +* Compute surface thickness +* +* @return +* Surface thickness +*************************************************************************************************** +*/ +UINT_32 AddrLib::ComputeSurfaceThickness( + AddrTileMode tileMode) ///< [in] tile mode +{ + return m_modeFlags[tileMode].thickness; +} + + + +/////////////////////////////////////////////////////////////////////////////////////////////////// +// CMASK/HTILE +/////////////////////////////////////////////////////////////////////////////////////////////////// + +/** +*************************************************************************************************** +* AddrLib::ComputeHtileInfo +* +* @brief +* Interface function stub of AddrComputeHtilenfo +* +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE AddrLib::ComputeHtileInfo( + const ADDR_COMPUTE_HTILE_INFO_INPUT* pIn, ///< [in] input structure + ADDR_COMPUTE_HTILE_INFO_OUTPUT* pOut ///< [out] output structure + ) const +{ + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + BOOL_32 isWidth8 = (pIn->blockWidth == 8) ? TRUE : FALSE; + BOOL_32 isHeight8 = (pIn->blockHeight == 8) ? TRUE : FALSE; + + if (GetFillSizeFieldsFlags() == TRUE) + { + if ((pIn->size != sizeof(ADDR_COMPUTE_HTILE_INFO_INPUT)) || + (pOut->size != sizeof(ADDR_COMPUTE_HTILE_INFO_OUTPUT))) + { + returnCode = ADDR_PARAMSIZEMISMATCH; + } + } + + if (returnCode == ADDR_OK) + { + ADDR_TILEINFO tileInfoNull; + ADDR_COMPUTE_HTILE_INFO_INPUT input; + + if (UseTileIndex(pIn->tileIndex)) + { + input = *pIn; + // Use temp tile info for calcalation + input.pTileInfo = &tileInfoNull; + + returnCode = HwlSetupTileCfg(input.tileIndex, input.macroModeIndex, input.pTileInfo); + + // Change the input structure + pIn = &input; + } + + if (returnCode == ADDR_OK) + { + pOut->bpp = ComputeHtileInfo(pIn->flags, + pIn->pitch, + pIn->height, + pIn->numSlices, + pIn->isLinear, + isWidth8, + isHeight8, + pIn->pTileInfo, + &pOut->pitch, + &pOut->height, + &pOut->htileBytes, + &pOut->macroWidth, + &pOut->macroHeight, + &pOut->sliceSize, + &pOut->baseAlign); + } + } + + return returnCode; +} + +/** +*************************************************************************************************** +* AddrLib::ComputeCmaskInfo +* +* @brief +* Interface function stub of AddrComputeCmaskInfo +* +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE AddrLib::ComputeCmaskInfo( + const ADDR_COMPUTE_CMASK_INFO_INPUT* pIn, ///< [in] input structure + ADDR_COMPUTE_CMASK_INFO_OUTPUT* pOut ///< [out] output structure + ) const +{ + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + if (GetFillSizeFieldsFlags() == TRUE) + { + if ((pIn->size != sizeof(ADDR_COMPUTE_CMASK_INFO_INPUT)) || + (pOut->size != sizeof(ADDR_COMPUTE_CMASK_INFO_OUTPUT))) + { + returnCode = ADDR_PARAMSIZEMISMATCH; + } + } + + if (returnCode == ADDR_OK) + { + ADDR_TILEINFO tileInfoNull; + ADDR_COMPUTE_CMASK_INFO_INPUT input; + + if (UseTileIndex(pIn->tileIndex)) + { + input = *pIn; + // Use temp tile info for calcalation + input.pTileInfo = &tileInfoNull; + + returnCode = HwlSetupTileCfg(input.tileIndex, input.macroModeIndex, input.pTileInfo); + + // Change the input structure + pIn = &input; + } + + if (returnCode == ADDR_OK) + { + returnCode = ComputeCmaskInfo(pIn->flags, + pIn->pitch, + pIn->height, + pIn->numSlices, + pIn->isLinear, + pIn->pTileInfo, + &pOut->pitch, + &pOut->height, + &pOut->cmaskBytes, + &pOut->macroWidth, + &pOut->macroHeight, + &pOut->sliceSize, + &pOut->baseAlign, + &pOut->blockMax); + } + } + + return returnCode; +} + +/** +*************************************************************************************************** +* AddrLib::ComputeDccInfo +* +* @brief +* Interface function to compute DCC key info +* +* @return +* return code of HwlComputeDccInfo +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE AddrLib::ComputeDccInfo( + const ADDR_COMPUTE_DCCINFO_INPUT* pIn, ///< [in] input structure + ADDR_COMPUTE_DCCINFO_OUTPUT* pOut ///< [out] output structure + ) const +{ + ADDR_E_RETURNCODE ret = ADDR_OK; + + if (GetFillSizeFieldsFlags() == TRUE) + { + if ((pIn->size != sizeof(ADDR_COMPUTE_DCCINFO_INPUT)) || + (pOut->size != sizeof(ADDR_COMPUTE_DCCINFO_OUTPUT))) + { + ret = ADDR_PARAMSIZEMISMATCH; + } + } + + if (ret == ADDR_OK) + { + ADDR_COMPUTE_DCCINFO_INPUT input; + + if (UseTileIndex(pIn->tileIndex)) + { + input = *pIn; + + ret = HwlSetupTileCfg(input.tileIndex, input.macroModeIndex, + &input.tileInfo, &input.tileMode); + + pIn = &input; + } + + if (ADDR_OK == ret) + { + ret = HwlComputeDccInfo(pIn, pOut); + } + } + + return ret; +} + +/** +*************************************************************************************************** +* AddrLib::ComputeHtileAddrFromCoord +* +* @brief +* Interface function stub of AddrComputeHtileAddrFromCoord +* +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE AddrLib::ComputeHtileAddrFromCoord( + const ADDR_COMPUTE_HTILE_ADDRFROMCOORD_INPUT* pIn, ///< [in] input structure + ADDR_COMPUTE_HTILE_ADDRFROMCOORD_OUTPUT* pOut ///< [out] output structure + ) const +{ + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + BOOL_32 isWidth8 = (pIn->blockWidth == 8) ? TRUE : FALSE; + BOOL_32 isHeight8 = (pIn->blockHeight == 8) ? TRUE : FALSE; + + if (GetFillSizeFieldsFlags() == TRUE) + { + if ((pIn->size != sizeof(ADDR_COMPUTE_HTILE_ADDRFROMCOORD_INPUT)) || + (pOut->size != sizeof(ADDR_COMPUTE_HTILE_ADDRFROMCOORD_OUTPUT))) + { + returnCode = ADDR_PARAMSIZEMISMATCH; + } + } + + if (returnCode == ADDR_OK) + { + ADDR_TILEINFO tileInfoNull; + ADDR_COMPUTE_HTILE_ADDRFROMCOORD_INPUT input; + + if (UseTileIndex(pIn->tileIndex)) + { + input = *pIn; + // Use temp tile info for calcalation + input.pTileInfo = &tileInfoNull; + + returnCode = HwlSetupTileCfg(input.tileIndex, input.macroModeIndex, input.pTileInfo); + + // Change the input structure + pIn = &input; + } + + if (returnCode == ADDR_OK) + { + pOut->addr = HwlComputeXmaskAddrFromCoord(pIn->pitch, + pIn->height, + pIn->x, + pIn->y, + pIn->slice, + pIn->numSlices, + 1, + pIn->isLinear, + isWidth8, + isHeight8, + pIn->pTileInfo, + &pOut->bitPosition); + } + } + + return returnCode; + +} + +/** +*************************************************************************************************** +* AddrLib::ComputeHtileCoordFromAddr +* +* @brief +* Interface function stub of AddrComputeHtileCoordFromAddr +* +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE AddrLib::ComputeHtileCoordFromAddr( + const ADDR_COMPUTE_HTILE_COORDFROMADDR_INPUT* pIn, ///< [in] input structure + ADDR_COMPUTE_HTILE_COORDFROMADDR_OUTPUT* pOut ///< [out] output structure + ) const +{ + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + BOOL_32 isWidth8 = (pIn->blockWidth == 8) ? TRUE : FALSE; + BOOL_32 isHeight8 = (pIn->blockHeight == 8) ? TRUE : FALSE; + + if (GetFillSizeFieldsFlags() == TRUE) + { + if ((pIn->size != sizeof(ADDR_COMPUTE_HTILE_COORDFROMADDR_INPUT)) || + (pOut->size != sizeof(ADDR_COMPUTE_HTILE_COORDFROMADDR_OUTPUT))) + { + returnCode = ADDR_PARAMSIZEMISMATCH; + } + } + + if (returnCode == ADDR_OK) + { + ADDR_TILEINFO tileInfoNull; + ADDR_COMPUTE_HTILE_COORDFROMADDR_INPUT input; + + if (UseTileIndex(pIn->tileIndex)) + { + input = *pIn; + // Use temp tile info for calcalation + input.pTileInfo = &tileInfoNull; + + returnCode = HwlSetupTileCfg(input.tileIndex, input.macroModeIndex, input.pTileInfo); + + // Change the input structure + pIn = &input; + } + + if (returnCode == ADDR_OK) + { + HwlComputeXmaskCoordFromAddr(pIn->addr, + pIn->bitPosition, + pIn->pitch, + pIn->height, + pIn->numSlices, + 1, + pIn->isLinear, + isWidth8, + isHeight8, + pIn->pTileInfo, + &pOut->x, + &pOut->y, + &pOut->slice); + } + } + + return returnCode; +} + +/** +*************************************************************************************************** +* AddrLib::ComputeCmaskAddrFromCoord +* +* @brief +* Interface function stub of AddrComputeCmaskAddrFromCoord +* +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE AddrLib::ComputeCmaskAddrFromCoord( + const ADDR_COMPUTE_CMASK_ADDRFROMCOORD_INPUT* pIn, ///< [in] input structure + ADDR_COMPUTE_CMASK_ADDRFROMCOORD_OUTPUT* pOut ///< [out] output structure + ) const +{ + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + if (GetFillSizeFieldsFlags() == TRUE) + { + if ((pIn->size != sizeof(ADDR_COMPUTE_CMASK_ADDRFROMCOORD_INPUT)) || + (pOut->size != sizeof(ADDR_COMPUTE_CMASK_ADDRFROMCOORD_OUTPUT))) + { + returnCode = ADDR_PARAMSIZEMISMATCH; + } + } + + if (returnCode == ADDR_OK) + { + ADDR_TILEINFO tileInfoNull; + ADDR_COMPUTE_CMASK_ADDRFROMCOORD_INPUT input; + + if (UseTileIndex(pIn->tileIndex)) + { + input = *pIn; + // Use temp tile info for calcalation + input.pTileInfo = &tileInfoNull; + + returnCode = HwlSetupTileCfg(input.tileIndex, input.macroModeIndex, input.pTileInfo); + + // Change the input structure + pIn = &input; + } + + if (returnCode == ADDR_OK) + { + if (pIn->flags.tcCompatible == TRUE) + { + returnCode = HwlComputeCmaskAddrFromCoord(pIn, pOut); + } + else + { + pOut->addr = HwlComputeXmaskAddrFromCoord(pIn->pitch, + pIn->height, + pIn->x, + pIn->y, + pIn->slice, + pIn->numSlices, + 2, + pIn->isLinear, + FALSE, //this is cmask, isWidth8 is not needed + FALSE, //this is cmask, isHeight8 is not needed + pIn->pTileInfo, + &pOut->bitPosition); + } + + } + } + + return returnCode; +} + +/** +*************************************************************************************************** +* AddrLib::ComputeCmaskCoordFromAddr +* +* @brief +* Interface function stub of AddrComputeCmaskCoordFromAddr +* +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE AddrLib::ComputeCmaskCoordFromAddr( + const ADDR_COMPUTE_CMASK_COORDFROMADDR_INPUT* pIn, ///< [in] input structure + ADDR_COMPUTE_CMASK_COORDFROMADDR_OUTPUT* pOut ///< [out] output structure + ) const +{ + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + if (GetFillSizeFieldsFlags() == TRUE) + { + if ((pIn->size != sizeof(ADDR_COMPUTE_CMASK_COORDFROMADDR_INPUT)) || + (pOut->size != sizeof(ADDR_COMPUTE_CMASK_COORDFROMADDR_OUTPUT))) + { + returnCode = ADDR_PARAMSIZEMISMATCH; + } + } + + if (returnCode == ADDR_OK) + { + ADDR_TILEINFO tileInfoNull; + ADDR_COMPUTE_CMASK_COORDFROMADDR_INPUT input; + + if (UseTileIndex(pIn->tileIndex)) + { + input = *pIn; + // Use temp tile info for calcalation + input.pTileInfo = &tileInfoNull; + + returnCode = HwlSetupTileCfg(input.tileIndex, input.macroModeIndex, input.pTileInfo); + + // Change the input structure + pIn = &input; + } + + if (returnCode == ADDR_OK) + { + HwlComputeXmaskCoordFromAddr(pIn->addr, + pIn->bitPosition, + pIn->pitch, + pIn->height, + pIn->numSlices, + 2, + pIn->isLinear, + FALSE, + FALSE, + pIn->pTileInfo, + &pOut->x, + &pOut->y, + &pOut->slice); + } + } + + return returnCode; +} + +/** +*************************************************************************************************** +* AddrLib::ComputeTileDataWidthAndHeight +* +* @brief +* Compute the squared cache shape for per-tile data (CMASK and HTILE) +* +* @return +* N/A +* +* @note +* MacroWidth and macroHeight are measured in pixels +*************************************************************************************************** +*/ +VOID AddrLib::ComputeTileDataWidthAndHeight( + UINT_32 bpp, ///< [in] bits per pixel + UINT_32 cacheBits, ///< [in] bits of cache + ADDR_TILEINFO* pTileInfo, ///< [in] Tile info + UINT_32* pMacroWidth, ///< [out] macro tile width + UINT_32* pMacroHeight ///< [out] macro tile height + ) const +{ + UINT_32 height = 1; + UINT_32 width = cacheBits / bpp; + UINT_32 pipes = HwlGetPipes(pTileInfo); + + // Double height until the macro-tile is close to square + // Height can only be doubled if width is even + + while ((width > height * 2 * pipes) && !(width & 1)) + { + width /= 2; + height *= 2; + } + + *pMacroWidth = 8 * width; + *pMacroHeight = 8 * height * pipes; + + // Note: The above iterative comptuation is equivalent to the following + // + //int log2_height = ((log2(cacheBits)-log2(bpp)-log2(pipes))/2); + //int macroHeight = pow2( 3+log2(pipes)+log2_height ); +} + +/** +*************************************************************************************************** +* AddrLib::HwlComputeTileDataWidthAndHeightLinear +* +* @brief +* Compute the squared cache shape for per-tile data (CMASK and HTILE) for linear layout +* +* @return +* N/A +* +* @note +* MacroWidth and macroHeight are measured in pixels +*************************************************************************************************** +*/ +VOID AddrLib::HwlComputeTileDataWidthAndHeightLinear( + UINT_32* pMacroWidth, ///< [out] macro tile width + UINT_32* pMacroHeight, ///< [out] macro tile height + UINT_32 bpp, ///< [in] bits per pixel + ADDR_TILEINFO* pTileInfo ///< [in] tile info + ) const +{ + ADDR_ASSERT(bpp != 4); // Cmask does not support linear layout prior to SI + *pMacroWidth = 8 * 512 / bpp; // Align width to 512-bit memory accesses + *pMacroHeight = 8 * m_pipes; // Align height to number of pipes +} + +/** +*************************************************************************************************** +* AddrLib::ComputeHtileInfo +* +* @brief +* Compute htile pitch,width, bytes per 2D slice +* +* @return +* Htile bpp i.e. How many bits for an 8x8 tile +* Also returns by output parameters: +* *Htile pitch, height, total size in bytes, macro-tile dimensions and slice size* +*************************************************************************************************** +*/ +UINT_32 AddrLib::ComputeHtileInfo( + ADDR_HTILE_FLAGS flags, ///< [in] htile flags + UINT_32 pitchIn, ///< [in] pitch input + UINT_32 heightIn, ///< [in] height input + UINT_32 numSlices, ///< [in] number of slices + BOOL_32 isLinear, ///< [in] if it is linear mode + BOOL_32 isWidth8, ///< [in] if htile block width is 8 + BOOL_32 isHeight8, ///< [in] if htile block height is 8 + ADDR_TILEINFO* pTileInfo, ///< [in] Tile info + UINT_32* pPitchOut, ///< [out] pitch output + UINT_32* pHeightOut, ///< [out] height output + UINT_64* pHtileBytes, ///< [out] bytes per 2D slice + UINT_32* pMacroWidth, ///< [out] macro-tile width in pixels + UINT_32* pMacroHeight, ///< [out] macro-tile width in pixels + UINT_64* pSliceSize, ///< [out] slice size in bytes + UINT_32* pBaseAlign ///< [out] base alignment + ) const +{ + + UINT_32 macroWidth; + UINT_32 macroHeight; + UINT_32 baseAlign; + UINT_64 surfBytes; + UINT_64 sliceBytes; + + numSlices = Max(1u, numSlices); + + const UINT_32 bpp = HwlComputeHtileBpp(isWidth8, isHeight8); + const UINT_32 cacheBits = HtileCacheBits; + + if (isLinear) + { + HwlComputeTileDataWidthAndHeightLinear(¯oWidth, + ¯oHeight, + bpp, + pTileInfo); + } + else + { + ComputeTileDataWidthAndHeight(bpp, + cacheBits, + pTileInfo, + ¯oWidth, + ¯oHeight); + } + + *pPitchOut = PowTwoAlign(pitchIn, macroWidth); + *pHeightOut = PowTwoAlign(heightIn, macroHeight); + + baseAlign = HwlComputeHtileBaseAlign(flags.tcCompatible, isLinear, pTileInfo); + + surfBytes = HwlComputeHtileBytes(*pPitchOut, + *pHeightOut, + bpp, + isLinear, + numSlices, + &sliceBytes, + baseAlign); + + *pHtileBytes = surfBytes; + + // + // Use SafeAssign since they are optional + // + SafeAssign(pMacroWidth, macroWidth); + + SafeAssign(pMacroHeight, macroHeight); + + SafeAssign(pSliceSize, sliceBytes); + + SafeAssign(pBaseAlign, baseAlign); + + return bpp; +} + +/** +*************************************************************************************************** +* AddrLib::ComputeCmaskBaseAlign +* +* @brief +* Compute cmask base alignment +* +* @return +* Cmask base alignment +*************************************************************************************************** +*/ +UINT_32 AddrLib::ComputeCmaskBaseAlign( + ADDR_CMASK_FLAGS flags, ///< [in] Cmask flags + ADDR_TILEINFO* pTileInfo ///< [in] Tile info + ) const +{ + UINT_32 baseAlign = m_pipeInterleaveBytes * HwlGetPipes(pTileInfo); + + if (flags.tcCompatible) + { + ADDR_ASSERT(pTileInfo != NULL); + if (pTileInfo) + { + baseAlign *= pTileInfo->banks; + } + } + + return baseAlign; +} + +/** +*************************************************************************************************** +* AddrLib::ComputeCmaskBytes +* +* @brief +* Compute cmask size in bytes +* +* @return +* Cmask size in bytes +*************************************************************************************************** +*/ +UINT_64 AddrLib::ComputeCmaskBytes( + UINT_32 pitch, ///< [in] pitch + UINT_32 height, ///< [in] height + UINT_32 numSlices ///< [in] number of slices + ) const +{ + return BITS_TO_BYTES(static_cast(pitch) * height * numSlices * CmaskElemBits) / + MicroTilePixels; +} + +/** +*************************************************************************************************** +* AddrLib::ComputeCmaskInfo +* +* @brief +* Compute cmask pitch,width, bytes per 2D slice +* +* @return +* BlockMax. Also by output parameters: Cmask pitch,height, total size in bytes, +* macro-tile dimensions +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE AddrLib::ComputeCmaskInfo( + ADDR_CMASK_FLAGS flags, ///< [in] cmask flags + UINT_32 pitchIn, ///< [in] pitch input + UINT_32 heightIn, ///< [in] height input + UINT_32 numSlices, ///< [in] number of slices + BOOL_32 isLinear, ///< [in] is linear mode + ADDR_TILEINFO* pTileInfo, ///< [in] Tile info + UINT_32* pPitchOut, ///< [out] pitch output + UINT_32* pHeightOut, ///< [out] height output + UINT_64* pCmaskBytes, ///< [out] bytes per 2D slice + UINT_32* pMacroWidth, ///< [out] macro-tile width in pixels + UINT_32* pMacroHeight, ///< [out] macro-tile width in pixels + UINT_64* pSliceSize, ///< [out] slice size in bytes + UINT_32* pBaseAlign, ///< [out] base alignment + UINT_32* pBlockMax ///< [out] block max == slice / 128 / 128 - 1 + ) const +{ + UINT_32 macroWidth; + UINT_32 macroHeight; + UINT_32 baseAlign; + UINT_64 surfBytes; + UINT_64 sliceBytes; + + numSlices = Max(1u, numSlices); + + const UINT_32 bpp = CmaskElemBits; + const UINT_32 cacheBits = CmaskCacheBits; + + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + if (isLinear) + { + HwlComputeTileDataWidthAndHeightLinear(¯oWidth, + ¯oHeight, + bpp, + pTileInfo); + } + else + { + ComputeTileDataWidthAndHeight(bpp, + cacheBits, + pTileInfo, + ¯oWidth, + ¯oHeight); + } + + *pPitchOut = (pitchIn + macroWidth - 1) & ~(macroWidth - 1); + *pHeightOut = (heightIn + macroHeight - 1) & ~(macroHeight - 1); + + + sliceBytes = ComputeCmaskBytes(*pPitchOut, + *pHeightOut, + 1); + + baseAlign = ComputeCmaskBaseAlign(flags, pTileInfo); + + while (sliceBytes % baseAlign) + { + *pHeightOut += macroHeight; + + sliceBytes = ComputeCmaskBytes(*pPitchOut, + *pHeightOut, + 1); + } + + surfBytes = sliceBytes * numSlices; + + *pCmaskBytes = surfBytes; + + // + // Use SafeAssign since they are optional + // + SafeAssign(pMacroWidth, macroWidth); + + SafeAssign(pMacroHeight, macroHeight); + + SafeAssign(pBaseAlign, baseAlign); + + SafeAssign(pSliceSize, sliceBytes); + + UINT_32 slice = (*pPitchOut) * (*pHeightOut); + UINT_32 blockMax = slice / 128 / 128 - 1; + +#if DEBUG + if (slice % (64*256) != 0) + { + ADDR_ASSERT_ALWAYS(); + } +#endif //DEBUG + + UINT_32 maxBlockMax = HwlGetMaxCmaskBlockMax(); + + if (blockMax > maxBlockMax) + { + blockMax = maxBlockMax; + returnCode = ADDR_INVALIDPARAMS; + } + + SafeAssign(pBlockMax, blockMax); + + return returnCode; +} + +/** +*************************************************************************************************** +* AddrLib::ComputeXmaskCoordYFromPipe +* +* @brief +* Compute the Y coord from pipe number for cmask/htile +* +* @return +* Y coordinate +* +*************************************************************************************************** +*/ +UINT_32 AddrLib::ComputeXmaskCoordYFromPipe( + UINT_32 pipe, ///< [in] pipe number + UINT_32 x ///< [in] x coordinate + ) const +{ + UINT_32 pipeBit0; + UINT_32 pipeBit1; + UINT_32 xBit0; + UINT_32 xBit1; + UINT_32 yBit0; + UINT_32 yBit1; + + UINT_32 y = 0; + + UINT_32 numPipes = m_pipes; // SI has its implementation + // + // Convert pipe + x to y coordinate. + // + switch (numPipes) + { + case 1: + // + // 1 pipe + // + // p0 = 0 + // + y = 0; + break; + case 2: + // + // 2 pipes + // + // p0 = x0 ^ y0 + // + // y0 = p0 ^ x0 + // + pipeBit0 = pipe & 0x1; + + xBit0 = x & 0x1; + + yBit0 = pipeBit0 ^ xBit0; + + y = yBit0; + break; + case 4: + // + // 4 pipes + // + // p0 = x1 ^ y0 + // p1 = x0 ^ y1 + // + // y0 = p0 ^ x1 + // y1 = p1 ^ x0 + // + pipeBit0 = pipe & 0x1; + pipeBit1 = (pipe & 0x2) >> 1; + + xBit0 = x & 0x1; + xBit1 = (x & 0x2) >> 1; + + yBit0 = pipeBit0 ^ xBit1; + yBit1 = pipeBit1 ^ xBit0; + + y = (yBit0 | + (yBit1 << 1)); + break; + case 8: + // + // 8 pipes + // + // r600 and r800 have different method + // + y = HwlComputeXmaskCoordYFrom8Pipe(pipe, x); + break; + default: + break; + } + return y; +} + +/** +*************************************************************************************************** +* AddrLib::HwlComputeXmaskCoordFromAddr +* +* @brief +* Compute the coord from an address of a cmask/htile +* +* @return +* N/A +* +* @note +* This method is reused by htile, so rename to Xmask +*************************************************************************************************** +*/ +VOID AddrLib::HwlComputeXmaskCoordFromAddr( + UINT_64 addr, ///< [in] address + UINT_32 bitPosition, ///< [in] bitPosition in a byte + UINT_32 pitch, ///< [in] pitch + UINT_32 height, ///< [in] height + UINT_32 numSlices, ///< [in] number of slices + UINT_32 factor, ///< [in] factor that indicates cmask or htile + BOOL_32 isLinear, ///< [in] linear or tiled HTILE layout + BOOL_32 isWidth8, ///< [in] TRUE if width is 8, FALSE means 4. It's register value + BOOL_32 isHeight8, ///< [in] TRUE if width is 8, FALSE means 4. It's register value + ADDR_TILEINFO* pTileInfo, ///< [in] Tile info + UINT_32* pX, ///< [out] x coord + UINT_32* pY, ///< [out] y coord + UINT_32* pSlice ///< [out] slice index + ) const +{ + UINT_32 pipe; + UINT_32 numPipes; + UINT_32 numGroupBits; + UINT_32 numPipeBits; + UINT_32 macroTilePitch; + UINT_32 macroTileHeight; + + UINT_64 bitAddr; + + UINT_32 microTileCoordY; + + UINT_32 elemBits; + + UINT_32 pitchAligned = pitch; + UINT_32 heightAligned = height; + UINT_64 totalBytes; + + UINT_64 elemOffset; + + UINT_64 macroIndex; + UINT_32 microIndex; + + UINT_64 macroNumber; + UINT_32 microNumber; + + UINT_32 macroX; + UINT_32 macroY; + UINT_32 macroZ; + + UINT_32 microX; + UINT_32 microY; + + UINT_32 tilesPerMacro; + UINT_32 macrosPerPitch; + UINT_32 macrosPerSlice; + + // + // Extract pipe. + // + numPipes = HwlGetPipes(pTileInfo); + pipe = ComputePipeFromAddr(addr, numPipes); + + // + // Compute the number of group and pipe bits. + // + numGroupBits = Log2(m_pipeInterleaveBytes); + numPipeBits = Log2(numPipes); + + UINT_32 groupBits = 8 * m_pipeInterleaveBytes; + UINT_32 pipes = numPipes; + + + // + // Compute the micro tile size, in bits. And macro tile pitch and height. + // + if (factor == 2) //CMASK + { + ADDR_CMASK_FLAGS flags = {{0}}; + + elemBits = CmaskElemBits; + + ComputeCmaskInfo(flags, + pitch, + height, + numSlices, + isLinear, + pTileInfo, + &pitchAligned, + &heightAligned, + &totalBytes, + ¯oTilePitch, + ¯oTileHeight); + } + else //HTILE + { + ADDR_HTILE_FLAGS flags = {{0}}; + + if (factor != 1) + { + factor = 1; + } + + elemBits = HwlComputeHtileBpp(isWidth8, isHeight8); + + ComputeHtileInfo(flags, + pitch, + height, + numSlices, + isLinear, + isWidth8, + isHeight8, + pTileInfo, + &pitchAligned, + &heightAligned, + &totalBytes, + ¯oTilePitch, + ¯oTileHeight); + } + + // Should use aligned dims + // + pitch = pitchAligned; + height = heightAligned; + + + // + // Convert byte address to bit address. + // + bitAddr = BYTES_TO_BITS(addr) + bitPosition; + + + // + // Remove pipe bits from address. + // + + bitAddr = (bitAddr % groupBits) + ((bitAddr/groupBits/pipes)*groupBits); + + + elemOffset = bitAddr / elemBits; + + tilesPerMacro = (macroTilePitch/factor) * macroTileHeight / MicroTilePixels >> numPipeBits; + + macrosPerPitch = pitch / (macroTilePitch/factor); + macrosPerSlice = macrosPerPitch * height / macroTileHeight; + + macroIndex = elemOffset / factor / tilesPerMacro; + microIndex = static_cast(elemOffset % (tilesPerMacro * factor)); + + macroNumber = macroIndex * factor + microIndex % factor; + microNumber = microIndex / factor; + + macroX = static_cast((macroNumber % macrosPerPitch)); + macroY = static_cast((macroNumber % macrosPerSlice) / macrosPerPitch); + macroZ = static_cast((macroNumber / macrosPerSlice)); + + + microX = microNumber % (macroTilePitch / factor / MicroTileWidth); + microY = (microNumber / (macroTilePitch / factor / MicroTileHeight)); + + *pX = macroX * (macroTilePitch/factor) + microX * MicroTileWidth; + *pY = macroY * macroTileHeight + (microY * MicroTileHeight << numPipeBits); + *pSlice = macroZ; + + microTileCoordY = ComputeXmaskCoordYFromPipe(pipe, + *pX/MicroTileWidth); + + + // + // Assemble final coordinates. + // + *pY += microTileCoordY * MicroTileHeight; + +} + +/** +*************************************************************************************************** +* AddrLib::HwlComputeXmaskAddrFromCoord +* +* @brief +* Compute the address from an address of cmask (prior to si) +* +* @return +* Address in bytes +* +*************************************************************************************************** +*/ +UINT_64 AddrLib::HwlComputeXmaskAddrFromCoord( + UINT_32 pitch, ///< [in] pitch + UINT_32 height, ///< [in] height + UINT_32 x, ///< [in] x coord + UINT_32 y, ///< [in] y coord + UINT_32 slice, ///< [in] slice/depth index + UINT_32 numSlices, ///< [in] number of slices + UINT_32 factor, ///< [in] factor that indicates cmask(2) or htile(1) + BOOL_32 isLinear, ///< [in] linear or tiled HTILE layout + BOOL_32 isWidth8, ///< [in] TRUE if width is 8, FALSE means 4. It's register value + BOOL_32 isHeight8, ///< [in] TRUE if width is 8, FALSE means 4. It's register value + ADDR_TILEINFO* pTileInfo, ///< [in] Tile info + UINT_32* pBitPosition ///< [out] bit position inside a byte + ) const +{ + UINT_64 addr; + UINT_32 numGroupBits; + UINT_32 numPipeBits; + UINT_32 newPitch = 0; + UINT_32 newHeight = 0; + UINT_64 sliceBytes = 0; + UINT_64 totalBytes = 0; + UINT_64 sliceOffset; + UINT_32 pipe; + UINT_32 macroTileWidth; + UINT_32 macroTileHeight; + UINT_32 macroTilesPerRow; + UINT_32 macroTileBytes; + UINT_32 macroTileIndexX; + UINT_32 macroTileIndexY; + UINT_64 macroTileOffset; + UINT_32 pixelBytesPerRow; + UINT_32 pixelOffsetX; + UINT_32 pixelOffsetY; + UINT_32 pixelOffset; + UINT_64 totalOffset; + UINT_64 offsetLo; + UINT_64 offsetHi; + UINT_64 groupMask; + + + UINT_32 elemBits = 0; + + UINT_32 numPipes = m_pipes; // This function is accessed prior to si only + + if (factor == 2) //CMASK + { + elemBits = CmaskElemBits; + + // For asics before SI, cmask is always tiled + isLinear = FALSE; + } + else //HTILE + { + if (factor != 1) // Fix compile warning + { + factor = 1; + } + + elemBits = HwlComputeHtileBpp(isWidth8, isHeight8); + } + + // + // Compute the number of group bits and pipe bits. + // + numGroupBits = Log2(m_pipeInterleaveBytes); + numPipeBits = Log2(numPipes); + + // + // Compute macro tile dimensions. + // + if (factor == 2) // CMASK + { + ADDR_CMASK_FLAGS flags = {{0}}; + + ComputeCmaskInfo(flags, + pitch, + height, + numSlices, + isLinear, + pTileInfo, + &newPitch, + &newHeight, + &totalBytes, + ¯oTileWidth, + ¯oTileHeight); + + sliceBytes = totalBytes / numSlices; + } + else // HTILE + { + ADDR_HTILE_FLAGS flags = {{0}}; + + ComputeHtileInfo(flags, + pitch, + height, + numSlices, + isLinear, + isWidth8, + isHeight8, + pTileInfo, + &newPitch, + &newHeight, + &totalBytes, + ¯oTileWidth, + ¯oTileHeight, + &sliceBytes); + } + + sliceOffset = slice * sliceBytes; + + // + // Get the pipe. Note that neither slice rotation nor pipe swizzling apply for CMASK. + // + pipe = ComputePipeFromCoord(x, + y, + 0, + ADDR_TM_2D_TILED_THIN1, + 0, + FALSE, + pTileInfo); + + // + // Compute the number of macro tiles per row. + // + macroTilesPerRow = newPitch / macroTileWidth; + + // + // Compute the number of bytes per macro tile. + // + macroTileBytes = BITS_TO_BYTES((macroTileWidth * macroTileHeight * elemBits) / MicroTilePixels); + + // + // Compute the offset to the macro tile containing the specified coordinate. + // + macroTileIndexX = x / macroTileWidth; + macroTileIndexY = y / macroTileHeight; + macroTileOffset = ((macroTileIndexY * macroTilesPerRow) + macroTileIndexX) * macroTileBytes; + + // + // Compute the pixel offset within the macro tile. + // + pixelBytesPerRow = BITS_TO_BYTES(macroTileWidth * elemBits) / MicroTileWidth; + + // + // The nibbles are interleaved (see below), so the part of the offset relative to the x + // coordinate repeats halfway across the row. (Not for HTILE) + // + if (factor == 2) + { + pixelOffsetX = (x % (macroTileWidth / 2)) / MicroTileWidth; + } + else + { + pixelOffsetX = (x % (macroTileWidth)) / MicroTileWidth * BITS_TO_BYTES(elemBits); + } + + // + // Compute the y offset within the macro tile. + // + pixelOffsetY = (((y % macroTileHeight) / MicroTileHeight) / numPipes) * pixelBytesPerRow; + + pixelOffset = pixelOffsetX + pixelOffsetY; + + // + // Combine the slice offset and macro tile offset with the pixel offset, accounting for the + // pipe bits in the middle of the address. + // + totalOffset = ((sliceOffset + macroTileOffset) >> numPipeBits) + pixelOffset; + + // + // Split the offset to put some bits below the pipe bits and some above. + // + groupMask = (1 << numGroupBits) - 1; + offsetLo = totalOffset & groupMask; + offsetHi = (totalOffset & ~groupMask) << numPipeBits; + + // + // Assemble the address from its components. + // + addr = offsetLo; + addr |= offsetHi; + // This is to remove warning with /analyze option + UINT_32 pipeBits = pipe << numGroupBits; + addr |= pipeBits; + + // + // Compute the bit position. The lower nibble is used when the x coordinate within the macro + // tile is less than half of the macro tile width, and the upper nibble is used when the x + // coordinate within the macro tile is greater than or equal to half the macro tile width. + // + *pBitPosition = ((x % macroTileWidth) < (macroTileWidth / factor)) ? 0 : 4; + + return addr; +} + +/////////////////////////////////////////////////////////////////////////////////////////////////// +// Surface Addressing Shared +/////////////////////////////////////////////////////////////////////////////////////////////////// + +/** +*************************************************************************************************** +* AddrLib::ComputeSurfaceAddrFromCoordLinear +* +* @brief +* Compute address from coord for linear surface +* +* @return +* Address in bytes +* +*************************************************************************************************** +*/ +UINT_64 AddrLib::ComputeSurfaceAddrFromCoordLinear( + UINT_32 x, ///< [in] x coord + UINT_32 y, ///< [in] y coord + UINT_32 slice, ///< [in] slice/depth index + UINT_32 sample, ///< [in] sample index + UINT_32 bpp, ///< [in] bits per pixel + UINT_32 pitch, ///< [in] pitch + UINT_32 height, ///< [in] height + UINT_32 numSlices, ///< [in] number of slices + UINT_32* pBitPosition ///< [out] bit position inside a byte + ) const +{ + const UINT_64 sliceSize = static_cast(pitch) * height; + + UINT_64 sliceOffset = (slice + sample * numSlices)* sliceSize; + UINT_64 rowOffset = static_cast(y) * pitch; + UINT_64 pixOffset = x; + + UINT_64 addr = (sliceOffset + rowOffset + pixOffset) * bpp; + + *pBitPosition = static_cast(addr % 8); + addr /= 8; + + return addr; +} + +/** +*************************************************************************************************** +* AddrLib::ComputeSurfaceCoordFromAddrLinear +* +* @brief +* Compute the coord from an address of a linear surface +* +* @return +* N/A +*************************************************************************************************** +*/ +VOID AddrLib::ComputeSurfaceCoordFromAddrLinear( + UINT_64 addr, ///< [in] address + UINT_32 bitPosition, ///< [in] bitPosition in a byte + UINT_32 bpp, ///< [in] bits per pixel + UINT_32 pitch, ///< [in] pitch + UINT_32 height, ///< [in] height + UINT_32 numSlices, ///< [in] number of slices + UINT_32* pX, ///< [out] x coord + UINT_32* pY, ///< [out] y coord + UINT_32* pSlice, ///< [out] slice/depth index + UINT_32* pSample ///< [out] sample index + ) const +{ + const UINT_64 sliceSize = static_cast(pitch) * height; + const UINT_64 linearOffset = (BYTES_TO_BITS(addr) + bitPosition) / bpp; + + *pX = static_cast((linearOffset % sliceSize) % pitch); + *pY = static_cast((linearOffset % sliceSize) / pitch % height); + *pSlice = static_cast((linearOffset / sliceSize) % numSlices); + *pSample = static_cast((linearOffset / sliceSize) / numSlices); +} + +/** +*************************************************************************************************** +* AddrLib::ComputeSurfaceCoordFromAddrMicroTiled +* +* @brief +* Compute the coord from an address of a micro tiled surface +* +* @return +* N/A +*************************************************************************************************** +*/ +VOID AddrLib::ComputeSurfaceCoordFromAddrMicroTiled( + UINT_64 addr, ///< [in] address + UINT_32 bitPosition, ///< [in] bitPosition in a byte + UINT_32 bpp, ///< [in] bits per pixel + UINT_32 pitch, ///< [in] pitch + UINT_32 height, ///< [in] height + UINT_32 numSamples, ///< [in] number of samples + AddrTileMode tileMode, ///< [in] tile mode + UINT_32 tileBase, ///< [in] base offset within a tile + UINT_32 compBits, ///< [in] component bits actually needed(for planar surface) + UINT_32* pX, ///< [out] x coord + UINT_32* pY, ///< [out] y coord + UINT_32* pSlice, ///< [out] slice/depth index + UINT_32* pSample, ///< [out] sample index, + AddrTileType microTileType, ///< [in] micro tiling order + BOOL_32 isDepthSampleOrder ///< [in] TRUE if in depth sample order + ) const +{ + UINT_64 bitAddr; + UINT_32 microTileThickness; + UINT_32 microTileBits; + UINT_64 sliceBits; + UINT_64 rowBits; + UINT_32 sliceIndex; + UINT_32 microTileCoordX; + UINT_32 microTileCoordY; + UINT_32 pixelOffset; + UINT_32 pixelCoordX = 0; + UINT_32 pixelCoordY = 0; + UINT_32 pixelCoordZ = 0; + UINT_32 pixelCoordS = 0; + + // + // Convert byte address to bit address. + // + bitAddr = BYTES_TO_BITS(addr) + bitPosition; + + // + // Compute the micro tile size, in bits. + // + switch (tileMode) + { + case ADDR_TM_1D_TILED_THICK: + microTileThickness = ThickTileThickness; + break; + default: + microTileThickness = 1; + break; + } + + microTileBits = MicroTilePixels * microTileThickness * bpp * numSamples; + + // + // Compute number of bits per slice and number of bits per row of micro tiles. + // + sliceBits = static_cast(pitch) * height * microTileThickness * bpp * numSamples; + + rowBits = (pitch / MicroTileWidth) * microTileBits; + + // + // Extract the slice index. + // + sliceIndex = static_cast(bitAddr / sliceBits); + bitAddr -= sliceIndex * sliceBits; + + // + // Extract the y coordinate of the micro tile. + // + microTileCoordY = static_cast(bitAddr / rowBits) * MicroTileHeight; + bitAddr -= (microTileCoordY / MicroTileHeight) * rowBits; + + // + // Extract the x coordinate of the micro tile. + // + microTileCoordX = static_cast(bitAddr / microTileBits) * MicroTileWidth; + + // + // Compute the pixel offset within the micro tile. + // + pixelOffset = static_cast(bitAddr % microTileBits); + + // + // Extract pixel coordinates from the offset. + // + HwlComputePixelCoordFromOffset(pixelOffset, + bpp, + numSamples, + tileMode, + tileBase, + compBits, + &pixelCoordX, + &pixelCoordY, + &pixelCoordZ, + &pixelCoordS, + microTileType, + isDepthSampleOrder); + + // + // Assemble final coordinates. + // + *pX = microTileCoordX + pixelCoordX; + *pY = microTileCoordY + pixelCoordY; + *pSlice = (sliceIndex * microTileThickness) + pixelCoordZ; + *pSample = pixelCoordS; + + if (microTileThickness > 1) + { + *pSample = 0; + } +} + +/** +*************************************************************************************************** +* AddrLib::ComputePipeFromAddr +* +* @brief +* Compute the pipe number from an address +* +* @return +* Pipe number +* +*************************************************************************************************** +*/ +UINT_32 AddrLib::ComputePipeFromAddr( + UINT_64 addr, ///< [in] address + UINT_32 numPipes ///< [in] number of banks + ) const +{ + UINT_32 pipe; + + UINT_32 groupBytes = m_pipeInterleaveBytes; //just different terms + + // R600 + // The LSBs of the address are arranged as follows: + // bank | pipe | group + // + // To get the pipe number, shift off the group bits and mask the pipe bits. + // + + // R800 + // The LSBs of the address are arranged as follows: + // bank | bankInterleave | pipe | pipeInterleave + // + // To get the pipe number, shift off the pipe interleave bits and mask the pipe bits. + // + + pipe = static_cast(addr >> Log2(groupBytes)) & (numPipes - 1); + + return pipe; +} + +/** +*************************************************************************************************** +* AddrLib::ComputePixelIndexWithinMicroTile +* +* @brief +* Compute the pixel index inside a micro tile of surface +* +* @return +* Pixel index +* +*************************************************************************************************** +*/ +UINT_32 AddrLib::ComputePixelIndexWithinMicroTile( + UINT_32 x, ///< [in] x coord + UINT_32 y, ///< [in] y coord + UINT_32 z, ///< [in] slice/depth index + UINT_32 bpp, ///< [in] bits per pixel + AddrTileMode tileMode, ///< [in] tile mode + AddrTileType microTileType ///< [in] pixel order in display/non-display mode + ) const +{ + UINT_32 pixelBit0 = 0; + UINT_32 pixelBit1 = 0; + UINT_32 pixelBit2 = 0; + UINT_32 pixelBit3 = 0; + UINT_32 pixelBit4 = 0; + UINT_32 pixelBit5 = 0; + UINT_32 pixelBit6 = 0; + UINT_32 pixelBit7 = 0; + UINT_32 pixelBit8 = 0; + UINT_32 pixelNumber; + + UINT_32 x0 = _BIT(x, 0); + UINT_32 x1 = _BIT(x, 1); + UINT_32 x2 = _BIT(x, 2); + UINT_32 y0 = _BIT(y, 0); + UINT_32 y1 = _BIT(y, 1); + UINT_32 y2 = _BIT(y, 2); + UINT_32 z0 = _BIT(z, 0); + UINT_32 z1 = _BIT(z, 1); + UINT_32 z2 = _BIT(z, 2); + + UINT_32 thickness = ComputeSurfaceThickness(tileMode); + + // Compute the pixel number within the micro tile. + + if (microTileType != ADDR_THICK) + { + if (microTileType == ADDR_DISPLAYABLE) + { + switch (bpp) + { + case 8: + pixelBit0 = x0; + pixelBit1 = x1; + pixelBit2 = x2; + pixelBit3 = y1; + pixelBit4 = y0; + pixelBit5 = y2; + break; + case 16: + pixelBit0 = x0; + pixelBit1 = x1; + pixelBit2 = x2; + pixelBit3 = y0; + pixelBit4 = y1; + pixelBit5 = y2; + break; + case 32: + pixelBit0 = x0; + pixelBit1 = x1; + pixelBit2 = y0; + pixelBit3 = x2; + pixelBit4 = y1; + pixelBit5 = y2; + break; + case 64: + pixelBit0 = x0; + pixelBit1 = y0; + pixelBit2 = x1; + pixelBit3 = x2; + pixelBit4 = y1; + pixelBit5 = y2; + break; + case 128: + pixelBit0 = y0; + pixelBit1 = x0; + pixelBit2 = x1; + pixelBit3 = x2; + pixelBit4 = y1; + pixelBit5 = y2; + break; + default: + ADDR_ASSERT_ALWAYS(); + break; + } + } + else if (microTileType == ADDR_NON_DISPLAYABLE || microTileType == ADDR_DEPTH_SAMPLE_ORDER) + { + pixelBit0 = x0; + pixelBit1 = y0; + pixelBit2 = x1; + pixelBit3 = y1; + pixelBit4 = x2; + pixelBit5 = y2; + } + else if (microTileType == ADDR_ROTATED) + { + ADDR_ASSERT(thickness == 1); + + switch (bpp) + { + case 8: + pixelBit0 = y0; + pixelBit1 = y1; + pixelBit2 = y2; + pixelBit3 = x1; + pixelBit4 = x0; + pixelBit5 = x2; + break; + case 16: + pixelBit0 = y0; + pixelBit1 = y1; + pixelBit2 = y2; + pixelBit3 = x0; + pixelBit4 = x1; + pixelBit5 = x2; + break; + case 32: + pixelBit0 = y0; + pixelBit1 = y1; + pixelBit2 = x0; + pixelBit3 = y2; + pixelBit4 = x1; + pixelBit5 = x2; + break; + case 64: + pixelBit0 = y0; + pixelBit1 = x0; + pixelBit2 = y1; + pixelBit3 = x1; + pixelBit4 = x2; + pixelBit5 = y2; + break; + default: + ADDR_ASSERT_ALWAYS(); + break; + } + } + + if (thickness > 1) + { + pixelBit6 = z0; + pixelBit7 = z1; + } + } + else // ADDR_THICK + { + ADDR_ASSERT(thickness > 1); + + switch (bpp) + { + case 8: + case 16: + pixelBit0 = x0; + pixelBit1 = y0; + pixelBit2 = x1; + pixelBit3 = y1; + pixelBit4 = z0; + pixelBit5 = z1; + break; + case 32: + pixelBit0 = x0; + pixelBit1 = y0; + pixelBit2 = x1; + pixelBit3 = z0; + pixelBit4 = y1; + pixelBit5 = z1; + break; + case 64: + case 128: + pixelBit0 = y0; + pixelBit1 = x0; + pixelBit2 = z0; + pixelBit3 = x1; + pixelBit4 = y1; + pixelBit5 = z1; + break; + default: + ADDR_ASSERT_ALWAYS(); + break; + } + + pixelBit6 = x2; + pixelBit7 = y2; + } + + if (thickness == 8) + { + pixelBit8 = z2; + } + + pixelNumber = ((pixelBit0 ) | + (pixelBit1 << 1) | + (pixelBit2 << 2) | + (pixelBit3 << 3) | + (pixelBit4 << 4) | + (pixelBit5 << 5) | + (pixelBit6 << 6) | + (pixelBit7 << 7) | + (pixelBit8 << 8)); + + return pixelNumber; +} + +/** +*************************************************************************************************** +* AddrLib::AdjustPitchAlignment +* +* @brief +* Adjusts pitch alignment for flipping surface +* +* @return +* N/A +* +*************************************************************************************************** +*/ +VOID AddrLib::AdjustPitchAlignment( + ADDR_SURFACE_FLAGS flags, ///< [in] Surface flags + UINT_32* pPitchAlign ///< [out] Pointer to pitch alignment + ) const +{ + // Display engine hardwires lower 5 bit of GRPH_PITCH to ZERO which means 32 pixel alignment + // Maybe it will be fixed in future but let's make it general for now. + if (flags.display || flags.overlay) + { + *pPitchAlign = PowTwoAlign(*pPitchAlign, 32); + + if(flags.display) + { + *pPitchAlign = Max(m_minPitchAlignPixels, *pPitchAlign); + } + } +} + +/** +*************************************************************************************************** +* AddrLib::PadDimensions +* +* @brief +* Helper function to pad dimensions +* +* @return +* N/A +* +*************************************************************************************************** +*/ +VOID AddrLib::PadDimensions( + AddrTileMode tileMode, ///< [in] tile mode + UINT_32 bpp, ///< [in] bits per pixel + ADDR_SURFACE_FLAGS flags, ///< [in] surface flags + UINT_32 numSamples, ///< [in] number of samples + ADDR_TILEINFO* pTileInfo, ///< [in/out] bank structure. + UINT_32 padDims, ///< [in] Dimensions to pad valid value 1,2,3 + UINT_32 mipLevel, ///< [in] MipLevel + UINT_32* pPitch, ///< [in/out] pitch in pixels + UINT_32 pitchAlign, ///< [in] pitch alignment + UINT_32* pHeight, ///< [in/out] height in pixels + UINT_32 heightAlign, ///< [in] height alignment + UINT_32* pSlices, ///< [in/out] number of slices + UINT_32 sliceAlign ///< [in] number of slice alignment + ) const +{ + UINT_32 thickness = ComputeSurfaceThickness(tileMode); + + ADDR_ASSERT(padDims <= 3); + + // + // Override padding for mip levels + // + if (mipLevel > 0) + { + if (flags.cube) + { + // for cubemap, we only pad when client call with 6 faces as an identity + if (*pSlices > 1) + { + padDims = 3; // we should pad cubemap sub levels when we treat it as 3d texture + } + else + { + padDims = 2; + } + } + } + + // Any possibilities that padDims is 0? + if (padDims == 0) + { + padDims = 3; + } + + if (IsPow2(pitchAlign)) + { + *pPitch = PowTwoAlign((*pPitch), pitchAlign); + } + else // add this code to pass unit test, r600 linear mode is not align bpp to pow2 for linear + { + *pPitch += pitchAlign - 1; + *pPitch /= pitchAlign; + *pPitch *= pitchAlign; + } + + if (padDims > 1) + { + *pHeight = PowTwoAlign((*pHeight), heightAlign); + } + + if (padDims > 2 || thickness > 1) + { + // for cubemap single face, we do not pad slices. + // if we pad it, the slice number should be set to 6 and current mip level > 1 + if (flags.cube && (!m_configFlags.noCubeMipSlicesPad || flags.cubeAsArray)) + { + *pSlices = NextPow2(*pSlices); + } + + // normal 3D texture or arrays or cubemap has a thick mode? (Just pass unit test) + if (thickness > 1) + { + *pSlices = PowTwoAlign((*pSlices), sliceAlign); + } + + } + + HwlPadDimensions(tileMode, + bpp, + flags, + numSamples, + pTileInfo, + padDims, + mipLevel, + pPitch, + pitchAlign, + pHeight, + heightAlign, + pSlices, + sliceAlign); +} + + +/** +*************************************************************************************************** +* AddrLib::HwlPreHandleBaseLvl3xPitch +* +* @brief +* Pre-handler of 3x pitch (96 bit) adjustment +* +* @return +* Expected pitch +*************************************************************************************************** +*/ +UINT_32 AddrLib::HwlPreHandleBaseLvl3xPitch( + const ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn, ///< [in] input + UINT_32 expPitch ///< [in] pitch + ) const +{ + ADDR_ASSERT(pIn->width == expPitch); + // + // If pitch is pre-multiplied by 3, we retrieve original one here to get correct miplevel size + // + if (AddrElemLib::IsExpand3x(pIn->format) && + pIn->mipLevel == 0 && + pIn->tileMode == ADDR_TM_LINEAR_ALIGNED) + { + expPitch /= 3; + expPitch = NextPow2(expPitch); + } + + return expPitch; +} + +/** +*************************************************************************************************** +* AddrLib::HwlPostHandleBaseLvl3xPitch +* +* @brief +* Post-handler of 3x pitch adjustment +* +* @return +* Expected pitch +*************************************************************************************************** +*/ +UINT_32 AddrLib::HwlPostHandleBaseLvl3xPitch( + const ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn, ///< [in] input + UINT_32 expPitch ///< [in] pitch + ) const +{ + // + // 96 bits surface of sub levels require element pitch of 32 bits instead + // So we just return pitch in 32 bit pixels without timing 3 + // + if (AddrElemLib::IsExpand3x(pIn->format) && + pIn->mipLevel == 0 && + pIn->tileMode == ADDR_TM_LINEAR_ALIGNED) + { + expPitch *= 3; + } + + return expPitch; +} + + +/** +*************************************************************************************************** +* AddrLib::IsMacroTiled +* +* @brief +* Check if the tile mode is macro tiled +* +* @return +* TRUE if it is macro tiled (2D/2B/3D/3B) +*************************************************************************************************** +*/ +BOOL_32 AddrLib::IsMacroTiled( + AddrTileMode tileMode) ///< [in] tile mode +{ + return m_modeFlags[tileMode].isMacro; +} + +/** +*************************************************************************************************** +* AddrLib::IsMacro3dTiled +* +* @brief +* Check if the tile mode is 3D macro tiled +* +* @return +* TRUE if it is 3D macro tiled +*************************************************************************************************** +*/ +BOOL_32 AddrLib::IsMacro3dTiled( + AddrTileMode tileMode) ///< [in] tile mode +{ + return m_modeFlags[tileMode].isMacro3d; +} + +/** +*************************************************************************************************** +* AddrLib::IsMicroTiled +* +* @brief +* Check if the tile mode is micro tiled +* +* @return +* TRUE if micro tiled +*************************************************************************************************** +*/ +BOOL_32 AddrLib::IsMicroTiled( + AddrTileMode tileMode) ///< [in] tile mode +{ + return m_modeFlags[tileMode].isMicro; +} + +/** +*************************************************************************************************** +* AddrLib::IsLinear +* +* @brief +* Check if the tile mode is linear +* +* @return +* TRUE if linear +*************************************************************************************************** +*/ +BOOL_32 AddrLib::IsLinear( + AddrTileMode tileMode) ///< [in] tile mode +{ + return m_modeFlags[tileMode].isLinear; +} + +/** +*************************************************************************************************** +* AddrLib::IsPrtNoRotationTileMode +* +* @brief +* Return TRUE if it is prt tile without rotation +* @note +* This function just used by CI +*************************************************************************************************** +*/ +BOOL_32 AddrLib::IsPrtNoRotationTileMode( + AddrTileMode tileMode) +{ + return m_modeFlags[tileMode].isPrtNoRotation; +} + +/** +*************************************************************************************************** +* AddrLib::IsPrtTileMode +* +* @brief +* Return TRUE if it is prt tile +* @note +* This function just used by CI +*************************************************************************************************** +*/ +BOOL_32 AddrLib::IsPrtTileMode( + AddrTileMode tileMode) +{ + return m_modeFlags[tileMode].isPrt; +} + +/** +*************************************************************************************************** +* AddrLib::Bits2Number +* +* @brief +* Cat a array of binary bit to a number +* +* @return +* The number combined with the array of bits +*************************************************************************************************** +*/ +UINT_32 AddrLib::Bits2Number( + UINT_32 bitNum, ///< [in] how many bits + ...) ///< [in] varaible bits value starting from MSB +{ + UINT_32 number = 0; + UINT_32 i; + va_list bits_ptr; + + va_start(bits_ptr, bitNum); + + for(i = 0; i < bitNum; i++) + { + number |= va_arg(bits_ptr, UINT_32); + number <<= 1; + } + + number>>=1; + + va_end(bits_ptr); + + return number; +} + +/** +*************************************************************************************************** +* AddrLib::ComputeMipLevel +* +* @brief +* Compute mipmap level width/height/slices +* @return +* N/A +*************************************************************************************************** +*/ +VOID AddrLib::ComputeMipLevel( + ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn ///< [in/out] Input structure + ) const +{ + // Check if HWL has handled + BOOL_32 hwlHandled = FALSE; + + if (AddrElemLib::IsBlockCompressed(pIn->format)) + { + if (pIn->mipLevel == 0) + { + // DXTn's level 0 must be multiple of 4 + // But there are exceptions: + // 1. Internal surface creation in hostblt/vsblt/etc... + // 2. Runtime doesn't reject ATI1/ATI2 whose width/height are not multiple of 4 + pIn->width = PowTwoAlign(pIn->width, 4); + pIn->height = PowTwoAlign(pIn->height, 4); + } + } + + hwlHandled = HwlComputeMipLevel(pIn); +} + +/** +*************************************************************************************************** +* AddrLib::DegradeBaseLevel +* +* @brief +* Check if base level's tile mode can be degraded +* @return +* TRUE if degraded, also returns degraded tile mode (unchanged if not degraded) +*************************************************************************************************** +*/ +BOOL_32 AddrLib::DegradeBaseLevel( + const ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn, ///< [in] Input structure for surface info + AddrTileMode* pTileMode ///< [out] Degraded tile mode + ) const +{ + BOOL_32 degraded = FALSE; + AddrTileMode tileMode = pIn->tileMode; + UINT_32 thickness = ComputeSurfaceThickness(tileMode); + + if (m_configFlags.degradeBaseLevel) // This is a global setting + { + if (pIn->flags.degrade4Space && // Degradation per surface + pIn->mipLevel == 0 && + pIn->numSamples == 1 && + IsMacroTiled(tileMode)) + { + if (HwlDegradeBaseLevel(pIn)) + { + *pTileMode = thickness == 1 ? ADDR_TM_1D_TILED_THIN1 : ADDR_TM_1D_TILED_THICK; + degraded = TRUE; + } + else if (thickness > 1) + { + // As in the following HwlComputeSurfaceInfo, thick modes may be degraded to + // thinner modes, we should re-evaluate whether the corresponding thinner modes + // need to be degraded. If so, we choose 1D thick mode instead. + tileMode = DegradeLargeThickTile(pIn->tileMode, pIn->bpp); + if (tileMode != pIn->tileMode) + { + ADDR_COMPUTE_SURFACE_INFO_INPUT input = *pIn; + input.tileMode = tileMode; + if (HwlDegradeBaseLevel(&input)) + { + *pTileMode = ADDR_TM_1D_TILED_THICK; + degraded = TRUE; + } + } + } + } + } + + return degraded; +} + +/** +*************************************************************************************************** +* AddrLib::DegradeLargeThickTile +* +* @brief +* Check if the thickness needs to be reduced if a tile is too large +* @return +* The degraded tile mode (unchanged if not degraded) +*************************************************************************************************** +*/ +AddrTileMode AddrLib::DegradeLargeThickTile( + AddrTileMode tileMode, + UINT_32 bpp) const +{ + // Override tilemode + // When tile_width (8) * tile_height (8) * thickness * element_bytes is > row_size, + // it is better to just use THIN mode in this case + UINT_32 thickness = ComputeSurfaceThickness(tileMode); + + if (thickness > 1 && m_configFlags.allowLargeThickTile == 0) + { + UINT_32 tileSize = MicroTilePixels * thickness * (bpp >> 3); + + if (tileSize > m_rowSize) + { + switch (tileMode) + { + case ADDR_TM_2D_TILED_XTHICK: + if ((tileSize >> 1) <= m_rowSize) + { + tileMode = ADDR_TM_2D_TILED_THICK; + break; + } + // else fall through + case ADDR_TM_2D_TILED_THICK: + tileMode = ADDR_TM_2D_TILED_THIN1; + break; + + case ADDR_TM_3D_TILED_XTHICK: + if ((tileSize >> 1) <= m_rowSize) + { + tileMode = ADDR_TM_3D_TILED_THICK; + break; + } + // else fall through + case ADDR_TM_3D_TILED_THICK: + tileMode = ADDR_TM_3D_TILED_THIN1; + break; + + case ADDR_TM_PRT_TILED_THICK: + tileMode = ADDR_TM_PRT_TILED_THIN1; + break; + + case ADDR_TM_PRT_2D_TILED_THICK: + tileMode = ADDR_TM_PRT_2D_TILED_THIN1; + break; + + case ADDR_TM_PRT_3D_TILED_THICK: + tileMode = ADDR_TM_PRT_3D_TILED_THIN1; + break; + + default: + break; + } + } + } + + return tileMode; +} + +/** +*************************************************************************************************** +* AddrLib::PostComputeMipLevel +* @brief +* Compute MipLevel info (including level 0) after surface adjustment +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE AddrLib::PostComputeMipLevel( + ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn, ///< [in/out] Input structure + ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pOut ///< [out] Output structure + ) const +{ + // Mipmap including level 0 must be pow2 padded since either SI hw expects so or it is + // required by CFX for Hw Compatibility between NI and SI. Otherwise it is only needed for + // mipLevel > 0. Any h/w has different requirement should implement its own virtual function + + if (pIn->flags.pow2Pad) + { + pIn->width = NextPow2(pIn->width); + pIn->height = NextPow2(pIn->height); + pIn->numSlices = NextPow2(pIn->numSlices); + } + else if (pIn->mipLevel > 0) + { + pIn->width = NextPow2(pIn->width); + pIn->height = NextPow2(pIn->height); + + if (!pIn->flags.cube) + { + pIn->numSlices = NextPow2(pIn->numSlices); + } + + // for cubemap, we keep its value at first + } + + return ADDR_OK; +} + +/** +*************************************************************************************************** +* AddrLib::HwlSetupTileCfg +* +* @brief +* Map tile index to tile setting. +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE AddrLib::HwlSetupTileCfg( + INT_32 index, ///< [in] Tile index + INT_32 macroModeIndex, ///< [in] Index in macro tile mode table(CI) + ADDR_TILEINFO* pInfo, ///< [out] Tile Info + AddrTileMode* pMode, ///< [out] Tile mode + AddrTileType* pType ///< [out] Tile type + ) const +{ + return ADDR_NOTSUPPORTED; +} + +/** +*************************************************************************************************** +* AddrLib::HwlGetPipes +* +* @brief +* Get number pipes +* @return +* num pipes +*************************************************************************************************** +*/ +UINT_32 AddrLib::HwlGetPipes( + const ADDR_TILEINFO* pTileInfo ///< [in] Tile info + ) const +{ + //pTileInfo can be NULL when asic is 6xx and 8xx. + return m_pipes; +} + +/** +*************************************************************************************************** +* AddrLib::ComputeQbStereoInfo +* +* @brief +* Get quad buffer stereo information +* @return +* TRUE if no error +*************************************************************************************************** +*/ +BOOL_32 AddrLib::ComputeQbStereoInfo( + ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pOut ///< [in/out] updated pOut+pStereoInfo + ) const +{ + BOOL_32 success = FALSE; + + if (pOut->pStereoInfo) + { + ADDR_ASSERT(pOut->bpp >= 8); + ADDR_ASSERT((pOut->surfSize % pOut->baseAlign) == 0); + + // Save original height + pOut->pStereoInfo->eyeHeight = pOut->height; + + // Right offset + pOut->pStereoInfo->rightOffset = static_cast(pOut->surfSize); + + pOut->pStereoInfo->rightSwizzle = HwlComputeQbStereoRightSwizzle(pOut); + // Double height + pOut->height <<= 1; + pOut->pixelHeight <<= 1; + + // Double size + pOut->surfSize <<= 1; + + // Right start address meets the base align since it is guaranteed by AddrLib + + // 1D surface on SI may break this rule, but we can force it to meet by checking .qbStereo. + success = TRUE; + } + + return success; +} + +/////////////////////////////////////////////////////////////////////////////////////////////////// +// Element lib +/////////////////////////////////////////////////////////////////////////////////////////////////// + + +/** +*************************************************************************************************** +* AddrLib::Flt32ToColorPixel +* +* @brief +* Convert a FLT_32 value to a depth/stencil pixel value +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE AddrLib::Flt32ToDepthPixel( + const ELEM_FLT32TODEPTHPIXEL_INPUT* pIn, + ELEM_FLT32TODEPTHPIXEL_OUTPUT* pOut) const +{ + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + if (GetFillSizeFieldsFlags() == TRUE) + { + if ((pIn->size != sizeof(ELEM_FLT32TODEPTHPIXEL_INPUT)) || + (pOut->size != sizeof(ELEM_FLT32TODEPTHPIXEL_OUTPUT))) + { + returnCode = ADDR_PARAMSIZEMISMATCH; + } + } + + if (returnCode == ADDR_OK) + { + GetElemLib()->Flt32ToDepthPixel(pIn->format, + pIn->comps, + pOut->pPixel); + UINT_32 depthBase = 0; + UINT_32 stencilBase = 0; + UINT_32 depthBits = 0; + UINT_32 stencilBits = 0; + + switch (pIn->format) + { + case ADDR_DEPTH_16: + depthBits = 16; + break; + case ADDR_DEPTH_X8_24: + case ADDR_DEPTH_8_24: + case ADDR_DEPTH_X8_24_FLOAT: + case ADDR_DEPTH_8_24_FLOAT: + depthBase = 8; + depthBits = 24; + stencilBits = 8; + break; + case ADDR_DEPTH_32_FLOAT: + depthBits = 32; + break; + case ADDR_DEPTH_X24_8_32_FLOAT: + depthBase = 8; + depthBits = 32; + stencilBits = 8; + break; + default: + break; + } + + // Overwrite base since R800 has no "tileBase" + if (GetElemLib()->IsDepthStencilTilePlanar() == FALSE) + { + depthBase = 0; + stencilBase = 0; + } + + depthBase *= 64; + stencilBase *= 64; + + pOut->stencilBase = stencilBase; + pOut->depthBase = depthBase; + pOut->depthBits = depthBits; + pOut->stencilBits = stencilBits; + } + + return returnCode; +} + +/** +*************************************************************************************************** +* AddrLib::Flt32ToColorPixel +* +* @brief +* Convert a FLT_32 value to a red/green/blue/alpha pixel value +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE AddrLib::Flt32ToColorPixel( + const ELEM_FLT32TOCOLORPIXEL_INPUT* pIn, + ELEM_FLT32TOCOLORPIXEL_OUTPUT* pOut) const +{ + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + if (GetFillSizeFieldsFlags() == TRUE) + { + if ((pIn->size != sizeof(ELEM_FLT32TOCOLORPIXEL_INPUT)) || + (pOut->size != sizeof(ELEM_FLT32TOCOLORPIXEL_OUTPUT))) + { + returnCode = ADDR_PARAMSIZEMISMATCH; + } + } + + if (returnCode == ADDR_OK) + { + GetElemLib()->Flt32ToColorPixel(pIn->format, + pIn->surfNum, + pIn->surfSwap, + pIn->comps, + pOut->pPixel); + } + + return returnCode; +} + + +/** +*************************************************************************************************** +* AddrLib::GetExportNorm +* +* @brief +* Check one format can be EXPORT_NUM +* @return +* TRUE if EXPORT_NORM can be used +*************************************************************************************************** +*/ +BOOL_32 AddrLib::GetExportNorm( + const ELEM_GETEXPORTNORM_INPUT* pIn) const +{ + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + BOOL_32 enabled = FALSE; + + if (GetFillSizeFieldsFlags() == TRUE) + { + if (pIn->size != sizeof(ELEM_GETEXPORTNORM_INPUT)) + { + returnCode = ADDR_PARAMSIZEMISMATCH; + } + } + + if (returnCode == ADDR_OK) + { + enabled = GetElemLib()->PixGetExportNorm(pIn->format, + pIn->num, + pIn->swap); + } + + return enabled; +} + +/** +*************************************************************************************************** +* AddrLib::ComputePrtInfo +* +* @brief +* Compute prt surface related info +* +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE AddrLib::ComputePrtInfo( + const ADDR_PRT_INFO_INPUT* pIn, + ADDR_PRT_INFO_OUTPUT* pOut) const +{ + ADDR_ASSERT(pOut != NULL); + + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + UINT_32 expandX = 1; + UINT_32 expandY = 1; + AddrElemMode elemMode; + + UINT_32 bpp = GetElemLib()->GetBitsPerPixel(pIn->format, + &elemMode, + &expandX, + &expandY); + + if (bpp <8 || bpp == 24 || bpp == 48 || bpp == 96 ) + { + returnCode = ADDR_INVALIDPARAMS; + } + + UINT_32 numFrags = pIn->numFrags; + ADDR_ASSERT(numFrags <= 8); + + UINT_32 tileWidth = 0; + UINT_32 tileHeight = 0; + if (returnCode == ADDR_OK) + { + // 3D texture without depth or 2d texture + if (pIn->baseMipDepth > 1 || pIn->baseMipHeight > 1) + { + if (bpp == 8) + { + tileWidth = 256; + tileHeight = 256; + } + else if (bpp == 16) + { + tileWidth = 256; + tileHeight = 128; + } + else if (bpp == 32) + { + tileWidth = 128; + tileHeight = 128; + } + else if (bpp == 64) + { + // assume it is BC1/4 + tileWidth = 512; + tileHeight = 256; + + if (elemMode == ADDR_UNCOMPRESSED) + { + tileWidth = 128; + tileHeight = 64; + } + } + else if (bpp == 128) + { + // assume it is BC2/3/5/6H/7 + tileWidth = 256; + tileHeight = 256; + + if (elemMode == ADDR_UNCOMPRESSED) + { + tileWidth = 64; + tileHeight = 64; + } + } + + if (numFrags == 2) + { + tileWidth = tileWidth / 2; + } + else if (numFrags == 4) + { + tileWidth = tileWidth / 2; + tileHeight = tileHeight / 2; + } + else if (numFrags == 8) + { + tileWidth = tileWidth / 4; + tileHeight = tileHeight / 2; + } + } + else // 1d + { + tileHeight = 1; + if (bpp == 8) + { + tileWidth = 65536; + } + else if (bpp == 16) + { + tileWidth = 32768; + } + else if (bpp == 32) + { + tileWidth = 16384; + } + else if (bpp == 64) + { + tileWidth = 8192; + } + else if (bpp == 128) + { + tileWidth = 4096; + } + } + } + + pOut->prtTileWidth = tileWidth; + pOut->prtTileHeight = tileHeight; + + return returnCode; +} diff --git a/src/gallium/winsys/radeon/amdgpu/addrlib/core/addrlib.h b/src/gallium/winsys/radeon/amdgpu/addrlib/core/addrlib.h new file mode 100644 index 0000000..43c55ff --- /dev/null +++ b/src/gallium/winsys/radeon/amdgpu/addrlib/core/addrlib.h @@ -0,0 +1,695 @@ +/* + * Copyright © 2014 Advanced Micro Devices, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES + * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NON-INFRINGEMENT. IN NO EVENT SHALL THE COPYRIGHT HOLDERS, AUTHORS + * AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + */ + +/** +*************************************************************************************************** +* @file addrlib.h +* @brief Contains the AddrLib base class definition. +*************************************************************************************************** +*/ + +#ifndef __ADDR_LIB_H__ +#define __ADDR_LIB_H__ + + +#include "addrinterface.h" +#include "addrobject.h" +#include "addrelemlib.h" + +#if BRAHMA_BUILD +#include "amdgpu_id.h" +#else +#include "atiid.h" +#endif + +#ifndef CIASICIDGFXENGINE_R600 +#define CIASICIDGFXENGINE_R600 0x00000006 +#endif + +#ifndef CIASICIDGFXENGINE_R800 +#define CIASICIDGFXENGINE_R800 0x00000008 +#endif + +#ifndef CIASICIDGFXENGINE_SOUTHERNISLAND +#define CIASICIDGFXENGINE_SOUTHERNISLAND 0x0000000A +#endif + +#ifndef CIASICIDGFXENGINE_SEAISLAND +#define CIASICIDGFXENGINE_SEAISLAND 0x0000000B +#endif +/** +*************************************************************************************************** +* @brief Neutral enums that define pipeinterleave +*************************************************************************************************** +*/ +enum AddrPipeInterleave +{ + ADDR_PIPEINTERLEAVE_256B = 256, + ADDR_PIPEINTERLEAVE_512B = 512, +}; + +/** +*************************************************************************************************** +* @brief Neutral enums that define DRAM row size +*************************************************************************************************** +*/ +enum AddrRowSize +{ + ADDR_ROWSIZE_1KB = 1024, + ADDR_ROWSIZE_2KB = 2048, + ADDR_ROWSIZE_4KB = 4096, + ADDR_ROWSIZE_8KB = 8192, +}; + +/** +*************************************************************************************************** +* @brief Neutral enums that define bank interleave +*************************************************************************************************** +*/ +enum AddrBankInterleave +{ + ADDR_BANKINTERLEAVE_1 = 1, + ADDR_BANKINTERLEAVE_2 = 2, + ADDR_BANKINTERLEAVE_4 = 4, + ADDR_BANKINTERLEAVE_8 = 8, +}; + +/** +*************************************************************************************************** +* @brief Neutral enums that define MGPU chip tile size +*************************************************************************************************** +*/ +enum AddrChipTileSize +{ + ADDR_CHIPTILESIZE_16 = 16, + ADDR_CHIPTILESIZE_32 = 32, + ADDR_CHIPTILESIZE_64 = 64, + ADDR_CHIPTILESIZE_128 = 128, +}; + +/** +*************************************************************************************************** +* @brief Neutral enums that define shader engine tile size +*************************************************************************************************** +*/ +enum AddrEngTileSize +{ + ADDR_SE_TILESIZE_16 = 16, + ADDR_SE_TILESIZE_32 = 32, +}; + +/** +*************************************************************************************************** +* @brief Neutral enums that define bank swap size +*************************************************************************************************** +*/ +enum AddrBankSwapSize +{ + ADDR_BANKSWAP_128B = 128, + ADDR_BANKSWAP_256B = 256, + ADDR_BANKSWAP_512B = 512, + ADDR_BANKSWAP_1KB = 1024, +}; + +/** +*************************************************************************************************** +* @brief Neutral enums that define bank swap size +*************************************************************************************************** +*/ +enum AddrSampleSplitSize +{ + ADDR_SAMPLESPLIT_1KB = 1024, + ADDR_SAMPLESPLIT_2KB = 2048, + ADDR_SAMPLESPLIT_4KB = 4096, + ADDR_SAMPLESPLIT_8KB = 8192, +}; + +/** +*************************************************************************************************** +* @brief Flags for AddrTileMode +*************************************************************************************************** +*/ +struct AddrTileModeFlags +{ + UINT_32 thickness : 4; + UINT_32 isLinear : 1; + UINT_32 isMicro : 1; + UINT_32 isMacro : 1; + UINT_32 isMacro3d : 1; + UINT_32 isPrt : 1; + UINT_32 isPrtNoRotation : 1; + UINT_32 isBankSwapped : 1; +}; + +/** +*************************************************************************************************** +* @brief This class contains asic independent address lib functionalities +*************************************************************************************************** +*/ +class AddrLib : public AddrObject +{ +public: + virtual ~AddrLib(); + + static ADDR_E_RETURNCODE Create( + const ADDR_CREATE_INPUT* pCreateInfo, ADDR_CREATE_OUTPUT* pCreateOut); + + /// Pair of Create + VOID Destroy() + { + delete this; + } + + static AddrLib* GetAddrLib( + ADDR_HANDLE hLib); + + /// Returns AddrLib version (from compiled binary instead include file) + UINT_32 GetVersion() + { + return m_version; + } + + /// Returns asic chip family name defined by AddrLib + AddrChipFamily GetAddrChipFamily() + { + return m_chipFamily; + } + + /// Returns tileIndex support + BOOL_32 UseTileIndex(INT_32 index) const + { + return m_configFlags.useTileIndex && (index != TileIndexInvalid); + } + + /// Returns combined swizzle support + BOOL_32 UseCombinedSwizzle() const + { + return m_configFlags.useCombinedSwizzle; + } + + // + // Interface stubs + // + ADDR_E_RETURNCODE ComputeSurfaceInfo( + const ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn, + ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pOut) const; + + ADDR_E_RETURNCODE ComputeSurfaceAddrFromCoord( + const ADDR_COMPUTE_SURFACE_ADDRFROMCOORD_INPUT* pIn, + ADDR_COMPUTE_SURFACE_ADDRFROMCOORD_OUTPUT* pOut) const; + + ADDR_E_RETURNCODE ComputeSurfaceCoordFromAddr( + const ADDR_COMPUTE_SURFACE_COORDFROMADDR_INPUT* pIn, + ADDR_COMPUTE_SURFACE_COORDFROMADDR_OUTPUT* pOut) const; + + ADDR_E_RETURNCODE ComputeSliceTileSwizzle( + const ADDR_COMPUTE_SLICESWIZZLE_INPUT* pIn, + ADDR_COMPUTE_SLICESWIZZLE_OUTPUT* pOut) const; + + ADDR_E_RETURNCODE ExtractBankPipeSwizzle( + const ADDR_EXTRACT_BANKPIPE_SWIZZLE_INPUT* pIn, + ADDR_EXTRACT_BANKPIPE_SWIZZLE_OUTPUT* pOut) const; + + ADDR_E_RETURNCODE CombineBankPipeSwizzle( + const ADDR_COMBINE_BANKPIPE_SWIZZLE_INPUT* pIn, + ADDR_COMBINE_BANKPIPE_SWIZZLE_OUTPUT* pOut) const; + + ADDR_E_RETURNCODE ComputeBaseSwizzle( + const ADDR_COMPUTE_BASE_SWIZZLE_INPUT* pIn, + ADDR_COMPUTE_BASE_SWIZZLE_OUTPUT* pOut) const; + + ADDR_E_RETURNCODE ComputeFmaskInfo( + const ADDR_COMPUTE_FMASK_INFO_INPUT* pIn, + ADDR_COMPUTE_FMASK_INFO_OUTPUT* pOut); + + ADDR_E_RETURNCODE ComputeFmaskAddrFromCoord( + const ADDR_COMPUTE_FMASK_ADDRFROMCOORD_INPUT* pIn, + ADDR_COMPUTE_FMASK_ADDRFROMCOORD_OUTPUT* pOut) const; + + ADDR_E_RETURNCODE ComputeFmaskCoordFromAddr( + const ADDR_COMPUTE_FMASK_COORDFROMADDR_INPUT* pIn, + ADDR_COMPUTE_FMASK_COORDFROMADDR_OUTPUT* pOut) const; + + ADDR_E_RETURNCODE ConvertTileInfoToHW( + const ADDR_CONVERT_TILEINFOTOHW_INPUT* pIn, + ADDR_CONVERT_TILEINFOTOHW_OUTPUT* pOut) const; + + ADDR_E_RETURNCODE ConvertTileIndex( + const ADDR_CONVERT_TILEINDEX_INPUT* pIn, + ADDR_CONVERT_TILEINDEX_OUTPUT* pOut) const; + + ADDR_E_RETURNCODE ConvertTileIndex1( + const ADDR_CONVERT_TILEINDEX1_INPUT* pIn, + ADDR_CONVERT_TILEINDEX_OUTPUT* pOut) const; + + ADDR_E_RETURNCODE GetTileIndex( + const ADDR_GET_TILEINDEX_INPUT* pIn, + ADDR_GET_TILEINDEX_OUTPUT* pOut) const; + + ADDR_E_RETURNCODE ComputeHtileInfo( + const ADDR_COMPUTE_HTILE_INFO_INPUT* pIn, + ADDR_COMPUTE_HTILE_INFO_OUTPUT* pOut) const; + + ADDR_E_RETURNCODE ComputeCmaskInfo( + const ADDR_COMPUTE_CMASK_INFO_INPUT* pIn, + ADDR_COMPUTE_CMASK_INFO_OUTPUT* pOut) const; + + ADDR_E_RETURNCODE ComputeDccInfo( + const ADDR_COMPUTE_DCCINFO_INPUT* pIn, + ADDR_COMPUTE_DCCINFO_OUTPUT* pOut) const; + + ADDR_E_RETURNCODE ComputeHtileAddrFromCoord( + const ADDR_COMPUTE_HTILE_ADDRFROMCOORD_INPUT* pIn, + ADDR_COMPUTE_HTILE_ADDRFROMCOORD_OUTPUT* pOut) const; + + ADDR_E_RETURNCODE ComputeCmaskAddrFromCoord( + const ADDR_COMPUTE_CMASK_ADDRFROMCOORD_INPUT* pIn, + ADDR_COMPUTE_CMASK_ADDRFROMCOORD_OUTPUT* pOut) const; + + ADDR_E_RETURNCODE ComputeHtileCoordFromAddr( + const ADDR_COMPUTE_HTILE_COORDFROMADDR_INPUT* pIn, + ADDR_COMPUTE_HTILE_COORDFROMADDR_OUTPUT* pOut) const; + + ADDR_E_RETURNCODE ComputeCmaskCoordFromAddr( + const ADDR_COMPUTE_CMASK_COORDFROMADDR_INPUT* pIn, + ADDR_COMPUTE_CMASK_COORDFROMADDR_OUTPUT* pOut) const; + + ADDR_E_RETURNCODE ComputePrtInfo( + const ADDR_PRT_INFO_INPUT* pIn, + ADDR_PRT_INFO_OUTPUT* pOut) const; + + ADDR_E_RETURNCODE Flt32ToDepthPixel( + const ELEM_FLT32TODEPTHPIXEL_INPUT* pIn, + ELEM_FLT32TODEPTHPIXEL_OUTPUT* pOut) const; + + ADDR_E_RETURNCODE Flt32ToColorPixel( + const ELEM_FLT32TOCOLORPIXEL_INPUT* pIn, + ELEM_FLT32TOCOLORPIXEL_OUTPUT* pOut) const; + + BOOL_32 GetExportNorm( + const ELEM_GETEXPORTNORM_INPUT* pIn) const; + +protected: + AddrLib(); // Constructor is protected + AddrLib(const AddrClient* pClient); + + /// Pure Virtual function for Hwl computing surface info + virtual ADDR_E_RETURNCODE HwlComputeSurfaceInfo( + const ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn, + ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pOut) const = 0; + + /// Pure Virtual function for Hwl computing surface address from coord + virtual ADDR_E_RETURNCODE HwlComputeSurfaceAddrFromCoord( + const ADDR_COMPUTE_SURFACE_ADDRFROMCOORD_INPUT* pIn, + ADDR_COMPUTE_SURFACE_ADDRFROMCOORD_OUTPUT* pOut) const = 0; + + /// Pure Virtual function for Hwl computing surface coord from address + virtual ADDR_E_RETURNCODE HwlComputeSurfaceCoordFromAddr( + const ADDR_COMPUTE_SURFACE_COORDFROMADDR_INPUT* pIn, + ADDR_COMPUTE_SURFACE_COORDFROMADDR_OUTPUT* pOut) const = 0; + + /// Pure Virtual function for Hwl computing surface tile swizzle + virtual ADDR_E_RETURNCODE HwlComputeSliceTileSwizzle( + const ADDR_COMPUTE_SLICESWIZZLE_INPUT* pIn, + ADDR_COMPUTE_SLICESWIZZLE_OUTPUT* pOut) const = 0; + + /// Pure Virtual function for Hwl extracting bank/pipe swizzle from base256b + virtual ADDR_E_RETURNCODE HwlExtractBankPipeSwizzle( + const ADDR_EXTRACT_BANKPIPE_SWIZZLE_INPUT* pIn, + ADDR_EXTRACT_BANKPIPE_SWIZZLE_OUTPUT* pOut) const = 0; + + /// Pure Virtual function for Hwl combining bank/pipe swizzle + virtual ADDR_E_RETURNCODE HwlCombineBankPipeSwizzle( + UINT_32 bankSwizzle, UINT_32 pipeSwizzle, ADDR_TILEINFO* pTileInfo, + UINT_64 baseAddr, UINT_32* pTileSwizzle) const = 0; + + /// Pure Virtual function for Hwl computing base swizzle + virtual ADDR_E_RETURNCODE HwlComputeBaseSwizzle( + const ADDR_COMPUTE_BASE_SWIZZLE_INPUT* pIn, + ADDR_COMPUTE_BASE_SWIZZLE_OUTPUT* pOut) const = 0; + + /// Pure Virtual function for Hwl computing HTILE base align + virtual UINT_32 HwlComputeHtileBaseAlign( + BOOL_32 isTcCompatible, BOOL_32 isLinear, ADDR_TILEINFO* pTileInfo) const = 0; + + /// Pure Virtual function for Hwl computing HTILE bpp + virtual UINT_32 HwlComputeHtileBpp( + BOOL_32 isWidth8, BOOL_32 isHeight8) const = 0; + + /// Pure Virtual function for Hwl computing HTILE bytes + virtual UINT_64 HwlComputeHtileBytes( + UINT_32 pitch, UINT_32 height, UINT_32 bpp, + BOOL_32 isLinear, UINT_32 numSlices, UINT_64* pSliceBytes, UINT_32 baseAlign) const = 0; + + /// Pure Virtual function for Hwl computing FMASK info + virtual ADDR_E_RETURNCODE HwlComputeFmaskInfo( + const ADDR_COMPUTE_FMASK_INFO_INPUT* pIn, + ADDR_COMPUTE_FMASK_INFO_OUTPUT* pOut) = 0; + + /// Pure Virtual function for Hwl FMASK address from coord + virtual ADDR_E_RETURNCODE HwlComputeFmaskAddrFromCoord( + const ADDR_COMPUTE_FMASK_ADDRFROMCOORD_INPUT* pIn, + ADDR_COMPUTE_FMASK_ADDRFROMCOORD_OUTPUT* pOut) const = 0; + + /// Pure Virtual function for Hwl FMASK coord from address + virtual ADDR_E_RETURNCODE HwlComputeFmaskCoordFromAddr( + const ADDR_COMPUTE_FMASK_COORDFROMADDR_INPUT* pIn, + ADDR_COMPUTE_FMASK_COORDFROMADDR_OUTPUT* pOut) const = 0; + + /// Pure Virtual function for Hwl convert tile info from real value to HW value + virtual ADDR_E_RETURNCODE HwlConvertTileInfoToHW( + const ADDR_CONVERT_TILEINFOTOHW_INPUT* pIn, + ADDR_CONVERT_TILEINFOTOHW_OUTPUT* pOut) const = 0; + + /// Pure Virtual function for Hwl compute mipmap info + virtual BOOL_32 HwlComputeMipLevel( + ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn) const = 0; + + /// Pure Virtual function for Hwl compute max cmask blockMax value + virtual BOOL_32 HwlGetMaxCmaskBlockMax() const = 0; + + /// Pure Virtual function for Hwl compute fmask bits + virtual UINT_32 HwlComputeFmaskBits( + const ADDR_COMPUTE_FMASK_INFO_INPUT* pIn, + UINT_32* pNumSamples) const = 0; + + /// Virtual function to get index (not pure then no need to implement this in all hwls + virtual ADDR_E_RETURNCODE HwlGetTileIndex( + const ADDR_GET_TILEINDEX_INPUT* pIn, + ADDR_GET_TILEINDEX_OUTPUT* pOut) const + { + return ADDR_NOTSUPPORTED; + } + + /// Virtual function for Hwl to compute Dcc info + virtual ADDR_E_RETURNCODE HwlComputeDccInfo( + const ADDR_COMPUTE_DCCINFO_INPUT* pIn, + ADDR_COMPUTE_DCCINFO_OUTPUT* pOut) const + { + return ADDR_NOTSUPPORTED; + } + + /// Virtual function to get cmask address for tc compatible cmask + virtual ADDR_E_RETURNCODE HwlComputeCmaskAddrFromCoord( + const ADDR_COMPUTE_CMASK_ADDRFROMCOORD_INPUT* pIn, + ADDR_COMPUTE_CMASK_ADDRFROMCOORD_OUTPUT* pOut) const + { + return ADDR_NOTSUPPORTED; + } + // Compute attributes + + // HTILE + UINT_32 ComputeHtileInfo( + ADDR_HTILE_FLAGS flags, + UINT_32 pitchIn, UINT_32 heightIn, UINT_32 numSlices, + BOOL_32 isLinear, BOOL_32 isWidth8, BOOL_32 isHeight8, + ADDR_TILEINFO* pTileInfo, + UINT_32* pPitchOut, UINT_32* pHeightOut, UINT_64* pHtileBytes, + UINT_32* pMacroWidth = NULL, UINT_32* pMacroHeight = NULL, + UINT_64* pSliceSize = NULL, UINT_32* pBaseAlign = NULL) const; + + // CMASK + ADDR_E_RETURNCODE ComputeCmaskInfo( + ADDR_CMASK_FLAGS flags, + UINT_32 pitchIn, UINT_32 heightIn, UINT_32 numSlices, BOOL_32 isLinear, + ADDR_TILEINFO* pTileInfo, UINT_32* pPitchOut, UINT_32* pHeightOut, UINT_64* pCmaskBytes, + UINT_32* pMacroWidth, UINT_32* pMacroHeight, UINT_64* pSliceSize = NULL, + UINT_32* pBaseAlign = NULL, UINT_32* pBlockMax = NULL) const; + + virtual VOID HwlComputeTileDataWidthAndHeightLinear( + UINT_32* pMacroWidth, UINT_32* pMacroHeight, + UINT_32 bpp, ADDR_TILEINFO* pTileInfo) const; + + // CMASK & HTILE addressing + virtual UINT_64 HwlComputeXmaskAddrFromCoord( + UINT_32 pitch, UINT_32 height, UINT_32 x, UINT_32 y, UINT_32 slice, + UINT_32 numSlices, UINT_32 factor, BOOL_32 isLinear, BOOL_32 isWidth8, + BOOL_32 isHeight8, ADDR_TILEINFO* pTileInfo, + UINT_32* bitPosition) const; + + virtual VOID HwlComputeXmaskCoordFromAddr( + UINT_64 addr, UINT_32 bitPosition, UINT_32 pitch, UINT_32 height, UINT_32 numSlices, + UINT_32 factor, BOOL_32 isLinear, BOOL_32 isWidth8, BOOL_32 isHeight8, + ADDR_TILEINFO* pTileInfo, UINT_32* pX, UINT_32* pY, UINT_32* pSlice) const; + + // Surface mipmap + VOID ComputeMipLevel( + ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn) const; + + /// Pure Virtual function for Hwl checking degrade for base level + virtual BOOL_32 HwlDegradeBaseLevel( + const ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn) const = 0; + + virtual BOOL_32 HwlOverrideTileMode( + const ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn, + AddrTileMode* pTileMode, + AddrTileType* pTileType) const + { + // not supported in hwl layer, FALSE for not-overrided + return FALSE; + } + + AddrTileMode DegradeLargeThickTile(AddrTileMode tileMode, UINT_32 bpp) const; + + VOID PadDimensions( + AddrTileMode tileMode, UINT_32 bpp, ADDR_SURFACE_FLAGS flags, + UINT_32 numSamples, ADDR_TILEINFO* pTileInfo, UINT_32 padDims, UINT_32 mipLevel, + UINT_32* pPitch, UINT_32 pitchAlign, UINT_32* pHeight, UINT_32 heightAlign, + UINT_32* pSlices, UINT_32 sliceAlign) const; + + virtual VOID HwlPadDimensions( + AddrTileMode tileMode, UINT_32 bpp, ADDR_SURFACE_FLAGS flags, + UINT_32 numSamples, ADDR_TILEINFO* pTileInfo, UINT_32 padDims, UINT_32 mipLevel, + UINT_32* pPitch, UINT_32 pitchAlign, UINT_32* pHeight, UINT_32 heightAlign, + UINT_32* pSlices, UINT_32 sliceAlign) const + { + } + + // + // Addressing shared for linear/1D tiling + // + UINT_64 ComputeSurfaceAddrFromCoordLinear( + UINT_32 x, UINT_32 y, UINT_32 slice, UINT_32 sample, + UINT_32 bpp, UINT_32 pitch, UINT_32 height, UINT_32 numSlices, + UINT_32* pBitPosition) const; + + VOID ComputeSurfaceCoordFromAddrLinear( + UINT_64 addr, UINT_32 bitPosition, UINT_32 bpp, + UINT_32 pitch, UINT_32 height, UINT_32 numSlices, + UINT_32* pX, UINT_32* pY, UINT_32* pSlice, UINT_32* pSample) const; + + VOID ComputeSurfaceCoordFromAddrMicroTiled( + UINT_64 addr, UINT_32 bitPosition, + UINT_32 bpp, UINT_32 pitch, UINT_32 height, UINT_32 numSamples, + AddrTileMode tileMode, UINT_32 tileBase, UINT_32 compBits, + UINT_32* pX, UINT_32* pY, UINT_32* pSlice, UINT_32* pSample, + AddrTileType microTileType, BOOL_32 isDepthSampleOrder) const; + + UINT_32 ComputePixelIndexWithinMicroTile( + UINT_32 x, UINT_32 y, UINT_32 z, + UINT_32 bpp, AddrTileMode tileMode, AddrTileType microTileType) const; + + /// Pure Virtual function for Hwl computing coord from offset inside micro tile + virtual VOID HwlComputePixelCoordFromOffset( + UINT_32 offset, UINT_32 bpp, UINT_32 numSamples, + AddrTileMode tileMode, UINT_32 tileBase, UINT_32 compBits, + UINT_32* pX, UINT_32* pY, UINT_32* pSlice, UINT_32* pSample, + AddrTileType microTileType, BOOL_32 isDepthSampleOrder) const = 0; + + // + // Addressing shared by all + // + virtual UINT_32 HwlGetPipes( + const ADDR_TILEINFO* pTileInfo) const; + + UINT_32 ComputePipeFromAddr( + UINT_64 addr, UINT_32 numPipes) const; + + /// Pure Virtual function for Hwl computing pipe from coord + virtual UINT_32 ComputePipeFromCoord( + UINT_32 x, UINT_32 y, UINT_32 slice, AddrTileMode tileMode, + UINT_32 pipeSwizzle, BOOL_32 flags, ADDR_TILEINFO* pTileInfo) const = 0; + + /// Pure Virtual function for Hwl computing coord Y for 8 pipe cmask/htile + virtual UINT_32 HwlComputeXmaskCoordYFrom8Pipe( + UINT_32 pipe, UINT_32 x) const = 0; + + // + // Initialization + // + /// Pure Virtual function for Hwl computing internal global parameters from h/w registers + virtual BOOL_32 HwlInitGlobalParams( + const ADDR_CREATE_INPUT* pCreateIn) = 0; + + /// Pure Virtual function for Hwl converting chip family + virtual AddrChipFamily HwlConvertChipFamily(UINT_32 uChipFamily, UINT_32 uChipRevision) = 0; + + // + // Misc helper + // + static const AddrTileModeFlags m_modeFlags[ADDR_TM_COUNT]; + + static UINT_32 ComputeSurfaceThickness( + AddrTileMode tileMode); + + // Checking tile mode + static BOOL_32 IsMacroTiled(AddrTileMode tileMode); + static BOOL_32 IsMacro3dTiled(AddrTileMode tileMode); + static BOOL_32 IsLinear(AddrTileMode tileMode); + static BOOL_32 IsMicroTiled(AddrTileMode tileMode); + static BOOL_32 IsPrtTileMode(AddrTileMode tileMode); + static BOOL_32 IsPrtNoRotationTileMode(AddrTileMode tileMode); + + static UINT_32 Bits2Number(UINT_32 bitNum,...); + + static UINT_32 GetNumFragments(UINT_32 numSamples, UINT_32 numFrags) + { + return numFrags != 0 ? numFrags : Max(1u, numSamples); + } + + /// Returns pointer of AddrElemLib + AddrElemLib* GetElemLib() const + { + return m_pElemLib; + } + + /// Return TRUE if tile info is needed + BOOL_32 UseTileInfo() const + { + return !m_configFlags.ignoreTileInfo; + } + + /// Returns fillSizeFields flag + UINT_32 GetFillSizeFieldsFlags() const + { + return m_configFlags.fillSizeFields; + } + + /// Adjusts pitch alignment for flipping surface + VOID AdjustPitchAlignment( + ADDR_SURFACE_FLAGS flags, UINT_32* pPitchAlign) const; + + /// Overwrite tile config according to tile index + virtual ADDR_E_RETURNCODE HwlSetupTileCfg( + INT_32 index, INT_32 macroModeIndex, + ADDR_TILEINFO* pInfo, AddrTileMode* mode = NULL, AddrTileType* type = NULL) const; + + /// Overwrite macro tile config according to tile index + virtual INT_32 HwlComputeMacroModeIndex( + INT_32 index, ADDR_SURFACE_FLAGS flags, UINT_32 bpp, UINT_32 numSamples, + ADDR_TILEINFO* pTileInfo, AddrTileMode *pTileMode = NULL, AddrTileType *pTileType = NULL + ) const + { + return TileIndexNoMacroIndex; + } + + /// Pre-handler of 3x pitch (96 bit) adjustment + virtual UINT_32 HwlPreHandleBaseLvl3xPitch( + const ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn, UINT_32 expPitch) const; + /// Post-handler of 3x pitch adjustment + virtual UINT_32 HwlPostHandleBaseLvl3xPitch( + const ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn, UINT_32 expPitch) const; + /// Check miplevel after surface adjustment + ADDR_E_RETURNCODE PostComputeMipLevel( + ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn, + ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pOut) const; + + /// Quad buffer stereo support, has its implementation in ind. layer + virtual BOOL_32 ComputeQbStereoInfo( + ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pOut) const; + + /// Pure virutual function to compute stereo bank swizzle for right eye + virtual UINT_32 HwlComputeQbStereoRightSwizzle( + ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pOut) const = 0; + +private: + // Disallow the copy constructor + AddrLib(const AddrLib& a); + + // Disallow the assignment operator + AddrLib& operator=(const AddrLib& a); + + VOID SetAddrChipFamily(UINT_32 uChipFamily, UINT_32 uChipRevision); + + UINT_32 ComputeCmaskBaseAlign( + ADDR_CMASK_FLAGS flags, ADDR_TILEINFO* pTileInfo) const; + + UINT_64 ComputeCmaskBytes( + UINT_32 pitch, UINT_32 height, UINT_32 numSlices) const; + + // + // CMASK/HTILE shared methods + // + VOID ComputeTileDataWidthAndHeight( + UINT_32 bpp, UINT_32 cacheBits, ADDR_TILEINFO* pTileInfo, + UINT_32* pMacroWidth, UINT_32* pMacroHeight) const; + + UINT_32 ComputeXmaskCoordYFromPipe( + UINT_32 pipe, UINT_32 x) const; + + VOID SetMinPitchAlignPixels(UINT_32 minPitchAlignPixels); + + BOOL_32 DegradeBaseLevel( + const ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn, AddrTileMode* pTileMode) const; + +protected: + AddrLibClass m_class; ///< Store class type (HWL type) + + AddrChipFamily m_chipFamily; ///< Chip family translated from the one in atiid.h + + UINT_32 m_chipRevision; ///< Revision id from xxx_id.h + + UINT_32 m_version; ///< Current version + + // + // Global parameters + // + ADDR_CONFIG_FLAGS m_configFlags; ///< Global configuration flags. Note this is setup by + /// AddrLib instead of Client except forceLinearAligned + + UINT_32 m_pipes; ///< Number of pipes + UINT_32 m_banks; ///< Number of banks + /// For r800 this is MC_ARB_RAMCFG.NOOFBANK + /// Keep it here to do default parameter calculation + + UINT_32 m_pipeInterleaveBytes; + ///< Specifies the size of contiguous address space + /// within each tiling pipe when making linear + /// accesses. (Formerly Group Size) + + UINT_32 m_rowSize; ///< DRAM row size, in bytes + + UINT_32 m_minPitchAlignPixels; ///< Minimum pitch alignment in pixels + UINT_32 m_maxSamples; ///< Max numSamples +private: + AddrElemLib* m_pElemLib; ///< Element Lib pointer +}; + +AddrLib* AddrSIHwlInit (const AddrClient* pClient); +AddrLib* AddrCIHwlInit (const AddrClient* pClient); + +#endif + diff --git a/src/gallium/winsys/radeon/amdgpu/addrlib/core/addrobject.cpp b/src/gallium/winsys/radeon/amdgpu/addrlib/core/addrobject.cpp new file mode 100644 index 0000000..863a252 --- /dev/null +++ b/src/gallium/winsys/radeon/amdgpu/addrlib/core/addrobject.cpp @@ -0,0 +1,246 @@ +/* + * Copyright © 2014 Advanced Micro Devices, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES + * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NON-INFRINGEMENT. IN NO EVENT SHALL THE COPYRIGHT HOLDERS, AUTHORS + * AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + */ + +/** +*************************************************************************************************** +* @file addrobject.cpp +* @brief Contains the AddrObject base class implementation. +*************************************************************************************************** +*/ + +#include "addrinterface.h" +#include "addrobject.h" + +/** +*************************************************************************************************** +* AddrObject::AddrObject +* +* @brief +* Constructor for the AddrObject class. +*************************************************************************************************** +*/ +AddrObject::AddrObject() +{ + m_client.handle = NULL; + m_client.callbacks.allocSysMem = NULL; + m_client.callbacks.freeSysMem = NULL; + m_client.callbacks.debugPrint = NULL; +} + +/** +*************************************************************************************************** +* AddrObject::AddrObject +* +* @brief +* Constructor for the AddrObject class. +*************************************************************************************************** +*/ +AddrObject::AddrObject(const AddrClient* pClient) +{ + m_client = *pClient; +} + +/** +*************************************************************************************************** +* AddrObject::~AddrObject +* +* @brief +* Destructor for the AddrObject class. +*************************************************************************************************** +*/ +AddrObject::~AddrObject() +{ +} + +/** +*************************************************************************************************** +* AddrObject::ClientAlloc +* +* @brief +* Calls instanced allocSysMem inside AddrClient +*************************************************************************************************** +*/ +VOID* AddrObject::ClientAlloc( + size_t objSize, ///< [in] Size to allocate + const AddrClient* pClient) ///< [in] Client pointer +{ + VOID* pObjMem = NULL; + + if (pClient->callbacks.allocSysMem != NULL) + { + ADDR_ALLOCSYSMEM_INPUT allocInput = {0}; + + allocInput.size = sizeof(ADDR_ALLOCSYSMEM_INPUT); + allocInput.flags.value = 0; + allocInput.sizeInBytes = static_cast(objSize); + allocInput.hClient = pClient->handle; + + pObjMem = pClient->callbacks.allocSysMem(&allocInput); + } + + return pObjMem; +} + +/** +*************************************************************************************************** +* AddrObject::AddrMalloc +* +* @brief +* A wrapper of ClientAlloc +*************************************************************************************************** +*/ +VOID* AddrObject::AddrMalloc( + size_t objSize) const ///< [in] Size to allocate +{ + return ClientAlloc(objSize, &m_client);; +} + +/** +*************************************************************************************************** +* AddrObject::ClientFree +* +* @brief +* Calls freeSysMem inside AddrClient +*************************************************************************************************** +*/ +VOID AddrObject::ClientFree( + VOID* pObjMem, ///< [in] User virtual address to free. + const AddrClient* pClient) ///< [in] Client pointer +{ + if (pClient->callbacks.freeSysMem != NULL) + { + if (pObjMem != NULL) + { + ADDR_FREESYSMEM_INPUT freeInput = {0}; + + freeInput.size = sizeof(ADDR_FREESYSMEM_INPUT); + freeInput.hClient = pClient->handle; + freeInput.pVirtAddr = pObjMem; + + pClient->callbacks.freeSysMem(&freeInput); + } + } +} + +/** +*************************************************************************************************** +* AddrObject::AddrFree +* +* @brief +* A wrapper of ClientFree +*************************************************************************************************** +*/ +VOID AddrObject::AddrFree( + VOID* pObjMem) const ///< [in] User virtual address to free. +{ + ClientFree(pObjMem, &m_client); +} + +/** +*************************************************************************************************** +* AddrObject::operator new +* +* @brief +* Allocates memory needed for AddrObject object. (with ADDR_CLIENT_HANDLE) +* +* @return +* Returns NULL if unsuccessful. +*************************************************************************************************** +*/ +VOID* AddrObject::operator new( + size_t objSize, ///< [in] Size to allocate + const AddrClient* pClient) ///< [in] Client pointer +{ + return ClientAlloc(objSize, pClient); +} + + +/** +*************************************************************************************************** +* AddrObject::operator delete +* +* @brief +* Frees AddrObject object memory. +*************************************************************************************************** +*/ +VOID AddrObject::operator delete( + VOID* pObjMem, ///< [in] User virtual address to free. + const AddrClient* pClient) ///< [in] Client handle +{ + ClientFree(pObjMem, pClient); +} + +/** +*************************************************************************************************** +* AddrObject::operator delete +* +* @brief +* Frees AddrObject object memory. +*************************************************************************************************** +*/ +VOID AddrObject::operator delete( + VOID* pObjMem) ///< [in] User virtual address to free. +{ + AddrObject* pObj = static_cast(pObjMem); + ClientFree(pObjMem, &pObj->m_client); +} + +/** +*************************************************************************************************** +* AddrObject::DebugPrint +* +* @brief +* Print debug message +* +* @return +* N/A +*************************************************************************************************** +*/ +VOID AddrObject::DebugPrint( + const CHAR* pDebugString, ///< [in] Debug string + ...) const +{ +#if DEBUG + if (m_client.callbacks.debugPrint != NULL) + { + va_list ap; + + va_start(ap, pDebugString); + + ADDR_DEBUGPRINT_INPUT debugPrintInput = {0}; + + debugPrintInput.size = sizeof(ADDR_DEBUGPRINT_INPUT); + debugPrintInput.pDebugString = const_cast(pDebugString); + debugPrintInput.hClient = m_client.handle; + va_copy(debugPrintInput.ap, ap); + + m_client.callbacks.debugPrint(&debugPrintInput); + + va_end(ap); + } +#endif +} + diff --git a/src/gallium/winsys/radeon/amdgpu/addrlib/core/addrobject.h b/src/gallium/winsys/radeon/amdgpu/addrlib/core/addrobject.h new file mode 100644 index 0000000..3540088 --- /dev/null +++ b/src/gallium/winsys/radeon/amdgpu/addrlib/core/addrobject.h @@ -0,0 +1,89 @@ +/* + * Copyright © 2014 Advanced Micro Devices, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES + * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NON-INFRINGEMENT. IN NO EVENT SHALL THE COPYRIGHT HOLDERS, AUTHORS + * AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + */ + +/** +*************************************************************************************************** +* @file addrobject.h +* @brief Contains the AddrObject base class definition. +*************************************************************************************************** +*/ + +#ifndef __ADDR_OBJECT_H__ +#define __ADDR_OBJECT_H__ + +#include "addrtypes.h" +#include "addrcommon.h" + +/** +*************************************************************************************************** +* @brief This structure contains client specific data +*************************************************************************************************** +*/ +struct AddrClient +{ + ADDR_CLIENT_HANDLE handle; + ADDR_CALLBACKS callbacks; +}; +/** +*************************************************************************************************** +* @brief This class is the base class for all ADDR class objects. +*************************************************************************************************** +*/ +class AddrObject +{ +public: + AddrObject(); + AddrObject(const AddrClient* pClient); + virtual ~AddrObject(); + + VOID* operator new(size_t size, const AddrClient* pClient); + VOID operator delete(VOID* pObj, const AddrClient* pClient); + VOID operator delete(VOID* pObj); + VOID* AddrMalloc(size_t size) const; + VOID AddrFree(VOID* pObj) const; + + VOID DebugPrint( + const CHAR* pDebugString, + ...) const; + + const AddrClient* GetClient() const {return &m_client;} + +protected: + AddrClient m_client; + +private: + static VOID* ClientAlloc(size_t size, const AddrClient* pClient); + static VOID ClientFree(VOID* pObj, const AddrClient* pClient); + + // disallow the copy constructor + AddrObject(const AddrObject& a); + + // disallow the assignment operator + AddrObject& operator=(const AddrObject& a); +}; + +#endif + diff --git a/src/gallium/winsys/radeon/amdgpu/addrlib/inc/chip/r800/si_gb_reg.h b/src/gallium/winsys/radeon/amdgpu/addrlib/inc/chip/r800/si_gb_reg.h new file mode 100644 index 0000000..cf67f60 --- /dev/null +++ b/src/gallium/winsys/radeon/amdgpu/addrlib/inc/chip/r800/si_gb_reg.h @@ -0,0 +1,155 @@ +#if !defined (__SI_GB_REG_H__) +#define __SI_GB_REG_H__ + +/* + * Copyright © 2014 Advanced Micro Devices, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES + * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NON-INFRINGEMENT. IN NO EVENT SHALL THE COPYRIGHT HOLDERS, AUTHORS + * AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + */ + +// +// Make sure the necessary endian defines are there. +// +#if defined(LITTLEENDIAN_CPU) +#elif defined(BIGENDIAN_CPU) +#else +#error "BIGENDIAN_CPU or LITTLEENDIAN_CPU must be defined" +#endif + +/* + * GB_ADDR_CONFIG struct + */ + +#if defined(LITTLEENDIAN_CPU) + + typedef struct _GB_ADDR_CONFIG_T { + unsigned int num_pipes : 3; + unsigned int : 1; + unsigned int pipe_interleave_size : 3; + unsigned int : 1; + unsigned int bank_interleave_size : 3; + unsigned int : 1; + unsigned int num_shader_engines : 2; + unsigned int : 2; + unsigned int shader_engine_tile_size : 3; + unsigned int : 1; + unsigned int num_gpus : 3; + unsigned int : 1; + unsigned int multi_gpu_tile_size : 2; + unsigned int : 2; + unsigned int row_size : 2; + unsigned int num_lower_pipes : 1; + unsigned int : 1; + } GB_ADDR_CONFIG_T; + +#elif defined(BIGENDIAN_CPU) + + typedef struct _GB_ADDR_CONFIG_T { + unsigned int : 1; + unsigned int num_lower_pipes : 1; + unsigned int row_size : 2; + unsigned int : 2; + unsigned int multi_gpu_tile_size : 2; + unsigned int : 1; + unsigned int num_gpus : 3; + unsigned int : 1; + unsigned int shader_engine_tile_size : 3; + unsigned int : 2; + unsigned int num_shader_engines : 2; + unsigned int : 1; + unsigned int bank_interleave_size : 3; + unsigned int : 1; + unsigned int pipe_interleave_size : 3; + unsigned int : 1; + unsigned int num_pipes : 3; + } GB_ADDR_CONFIG_T; + +#endif + +typedef union { + unsigned int val : 32; + GB_ADDR_CONFIG_T f; +} GB_ADDR_CONFIG; + +#if defined(LITTLEENDIAN_CPU) + + typedef struct _GB_TILE_MODE_T { + unsigned int micro_tile_mode : 2; + unsigned int array_mode : 4; + unsigned int pipe_config : 5; + unsigned int tile_split : 3; + unsigned int bank_width : 2; + unsigned int bank_height : 2; + unsigned int macro_tile_aspect : 2; + unsigned int num_banks : 2; + unsigned int micro_tile_mode_new : 3; + unsigned int sample_split : 2; + unsigned int : 5; + } GB_TILE_MODE_T; + + typedef struct _GB_MACROTILE_MODE_T { + unsigned int bank_width : 2; + unsigned int bank_height : 2; + unsigned int macro_tile_aspect : 2; + unsigned int num_banks : 2; + unsigned int : 24; + } GB_MACROTILE_MODE_T; + +#elif defined(BIGENDIAN_CPU) + + typedef struct _GB_TILE_MODE_T { + unsigned int : 5; + unsigned int sample_split : 2; + unsigned int micro_tile_mode_new : 3; + unsigned int num_banks : 2; + unsigned int macro_tile_aspect : 2; + unsigned int bank_height : 2; + unsigned int bank_width : 2; + unsigned int tile_split : 3; + unsigned int pipe_config : 5; + unsigned int array_mode : 4; + unsigned int micro_tile_mode : 2; + } GB_TILE_MODE_T; + + typedef struct _GB_MACROTILE_MODE_T { + unsigned int : 24; + unsigned int num_banks : 2; + unsigned int macro_tile_aspect : 2; + unsigned int bank_height : 2; + unsigned int bank_width : 2; + } GB_MACROTILE_MODE_T; + +#endif + +typedef union { + unsigned int val : 32; + GB_TILE_MODE_T f; +} GB_TILE_MODE; + +typedef union { + unsigned int val : 32; + GB_MACROTILE_MODE_T f; +} GB_MACROTILE_MODE; + +#endif + diff --git a/src/gallium/winsys/radeon/amdgpu/addrlib/inc/lnx_common_defs.h b/src/gallium/winsys/radeon/amdgpu/addrlib/inc/lnx_common_defs.h new file mode 100644 index 0000000..61540f4 --- /dev/null +++ b/src/gallium/winsys/radeon/amdgpu/addrlib/inc/lnx_common_defs.h @@ -0,0 +1,129 @@ +/* + * Copyright © 2014 Advanced Micro Devices, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES + * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NON-INFRINGEMENT. IN NO EVENT SHALL THE COPYRIGHT HOLDERS, AUTHORS + * AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + */ +#ifndef _lnx_common_defs_h_ +#define _lnx_common_defs_h_ + +#if DBG +#include // We do not have any choice: need variable + // number of parameters support for debug + // build. +#endif // #if DBG + +// +// -------------- External functions from Linux kernel driver ---------------- +// +// Note: The definitions/declararions below must match the original ones. + +#ifdef __cplusplus +extern "C" { +#endif + +typedef unsigned long __ke_size_t; // as it is defined in firegl_public.h +typedef int __kernel_ptrdiff_t; // as it is defined in posix_types.h + + +#if !defined(ATI_API_CALL) +#define ATI_API_CALL __attribute__((regparm(0))) +#endif + +extern void * ATI_API_CALL __ke_memset(void* s, int c, __ke_size_t count); +extern void * ATI_API_CALL __ke_memcpy(void* d, const void* s, __ke_size_t count); +extern ATI_API_CALL __ke_size_t __ke_strlen(const char *s); +extern char* ATI_API_CALL __ke_strcpy(char* d, const char* s); +extern char* ATI_API_CALL __ke_strncpy(char* d, const char* s, __ke_size_t count); +extern void __ke_printk(const char* fmt, ...); + +extern int ATI_API_CALL __ke_snprintf(char* buf, __ke_size_t size, const char* fmt, ...); +extern int ATI_API_CALL KCL_CopyFromUserSpace(void* to, const void* from, __ke_size_t size); +extern int ATI_API_CALL KCL_CopyToUserSpace(void* to, const void* from, __ke_size_t size); +#define __ke_copy_from_user KCL_CopyFromUserSpace +#define __ke_copy_to_user KCL_CopyToUserSpace +extern int ATI_API_CALL __ke_verify_area(int type, const void * addr, unsigned long size); + +extern unsigned long ATI_API_CALL KAS_GetTickCounter(void); +extern unsigned long ATI_API_CALL KAS_GetTicksPerSecond(void); + + +#if DBG +extern int ATI_API_CALL __ke_vsnprintf(char *buf, __ke_size_t size, const char *fmt, va_list ap); +#define vsnprintf(_dst, _size, _fmt, varg) __ke_snprintf(_dst, _size, _fmt, varg) +#endif // #if DBG + + +// Note: This function is not defined in firegl_public.h. +void firegl_hardwareHangRecovery(void); + +#ifdef __cplusplus +} +#endif + +// +// -------------------------- C/C++ standard typedefs ---------------------------- +// +#ifdef __SIZE_TYPE__ +typedef __SIZE_TYPE__ size_t; +#else // #ifdef __SIZE_TYPE__ +typedef unsigned int size_t; +#endif // #ifdef __SIZE_TYPE__ + +#ifdef __PTRDIFF_TYPE__ +typedef __PTRDIFF_TYPE__ ptrdiff_t; +#else // #ifdef __PTRDIFF_TYPE__ +typedef int ptrdiff_t; +#endif // #ifdef __PTRDIFF_TYPE__ + +#ifndef NULL +#ifdef __cplusplus +#define NULL __null +#else +#define NULL ((void *)0) +#endif +#endif + + +// +// ------------------------- C/C++ standard macros --------------------------- +// + +#define offsetof(TYPE, MEMBER) ((size_t) &((TYPE *)0)->MEMBER) // as it is defined in stddef.h +#define CHAR_BIT 8 // as it is defined in limits.h + +// +// --------------------------------- C RTL ----------------------------------- +// + +#define memset(_p, _v, _n) __ke_memset(_p, _v, _n) +#define memcpy(_d, _s, _n) __ke_memcpy(_d, _s, _n) +#define strlen(_s) __ke_strlen(_s) +#define strcpy(_d, _s) __ke_strcpy(_d, _s) +#define strncpy(_d, _s, _n) __ke_strncpy(_d, _s, _n) +// Note: C99 supports macros with variable number of arguments. GCC also supports this C99 feature as +// C++ extension. +#define snprintf(_dst, _size, _fmt, arg...) __ke_snprintf(_dst, _size, _fmt, ##arg) + + +#endif // #ifdef _lnx_common_defs_h_ + diff --git a/src/gallium/winsys/radeon/amdgpu/addrlib/r800/chip/si_ci_vi_merged_enum.h b/src/gallium/winsys/radeon/amdgpu/addrlib/r800/chip/si_ci_vi_merged_enum.h new file mode 100644 index 0000000..5ed81ad --- /dev/null +++ b/src/gallium/winsys/radeon/amdgpu/addrlib/r800/chip/si_ci_vi_merged_enum.h @@ -0,0 +1,40 @@ +/* + * Copyright © 2014 Advanced Micro Devices, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES + * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NON-INFRINGEMENT. IN NO EVENT SHALL THE COPYRIGHT HOLDERS, AUTHORS + * AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + */ +#if !defined (SI_CI_VI_MERGED_ENUM_HEADER) +#define SI_CI_VI_MERGED_ENUM_HEADER + +typedef enum PipeInterleaveSize { +ADDR_CONFIG_PIPE_INTERLEAVE_256B = 0x00000000, +ADDR_CONFIG_PIPE_INTERLEAVE_512B = 0x00000001, +} PipeInterleaveSize; + +typedef enum RowSize { +ADDR_CONFIG_1KB_ROW = 0x00000000, +ADDR_CONFIG_2KB_ROW = 0x00000001, +ADDR_CONFIG_4KB_ROW = 0x00000002, +} RowSize; + +#endif diff --git a/src/gallium/winsys/radeon/amdgpu/addrlib/r800/ciaddrlib.cpp b/src/gallium/winsys/radeon/amdgpu/addrlib/r800/ciaddrlib.cpp new file mode 100644 index 0000000..264e2ef --- /dev/null +++ b/src/gallium/winsys/radeon/amdgpu/addrlib/r800/ciaddrlib.cpp @@ -0,0 +1,1777 @@ +/* + * Copyright © 2014 Advanced Micro Devices, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES + * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NON-INFRINGEMENT. IN NO EVENT SHALL THE COPYRIGHT HOLDERS, AUTHORS + * AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + */ + +/** +*************************************************************************************************** +* @file ciaddrlib.cpp +* @brief Contains the implementation for the CIAddrLib class. +*************************************************************************************************** +*/ + +#include "ciaddrlib.h" + +#include "si_gb_reg.h" + +#include "si_ci_vi_merged_enum.h" + +#if BRAHMA_BUILD +#include "amdgpu_id.h" +#else +#include "ci_id.h" +#include "kv_id.h" +#include "vi_id.h" +#endif + +/////////////////////////////////////////////////////////////////////////////////////////////////// +/////////////////////////////////////////////////////////////////////////////////////////////////// + +/** +*************************************************************************************************** +* AddrMask +* +* @brief +* Gets a mask of "width" +* @return +* Bit mask +*************************************************************************************************** +*/ +static UINT_64 AddrMask( + UINT_32 width) ///< Width of bits +{ + UINT_64 ret; + + if (width >= sizeof(UINT_64)*8) + { + ret = ~((UINT_64) 0); + } + else + { + return (((UINT_64) 1) << width) - 1; + } + return ret; +} + +/** +*************************************************************************************************** +* AddrGetBits +* +* @brief +* Gets bits within a range of [msb, lsb] +* @return +* Bits of this range +*************************************************************************************************** +*/ +static UINT_64 AddrGetBits( + UINT_64 bits, ///< Source bits + UINT_32 msb, ///< Most signicant bit + UINT_32 lsb) ///< Least signicant bit +{ + UINT_64 ret = 0; + + if (msb >= lsb) + { + ret = (bits >> lsb) & (AddrMask(1 + msb - lsb)); + } + return ret; +} + +/** +*************************************************************************************************** +* AddrRemoveBits +* +* @brief +* Removes bits within the range of [msb, lsb] +* @return +* Modified bits +*************************************************************************************************** +*/ +static UINT_64 AddrRemoveBits( + UINT_64 bits, ///< Source bits + UINT_32 msb, ///< Most signicant bit + UINT_32 lsb) ///< Least signicant bit +{ + UINT_64 ret = bits; + + if (msb >= lsb) + { + ret = AddrGetBits(bits, lsb - 1, 0) // low bits + | (AddrGetBits(bits, 8 * sizeof(bits) - 1, msb + 1) << lsb); //high bits + } + return ret; +} + +/** +*************************************************************************************************** +* AddrInsertBits +* +* @brief +* Inserts new bits into the range of [msb, lsb] +* @return +* Modified bits +*************************************************************************************************** +*/ +static UINT_64 AddrInsertBits( + UINT_64 bits, ///< Source bits + UINT_64 newBits, ///< New bits to be inserted + UINT_32 msb, ///< Most signicant bit + UINT_32 lsb) ///< Least signicant bit +{ + UINT_64 ret = bits; + + if (msb >= lsb) + { + ret = AddrGetBits(bits, lsb - 1, 0) // old low bitss + | (AddrGetBits(newBits, msb - lsb, 0) << lsb) //new bits + | (AddrGetBits(bits, 8 * sizeof(bits) - 1, lsb) << (msb + 1)); //old high bits + } + return ret; +} + + +/** +*************************************************************************************************** +* AddrCIHwlInit +* +* @brief +* Creates an CIAddrLib object. +* +* @return +* Returns an CIAddrLib object pointer. +*************************************************************************************************** +*/ +AddrLib* AddrCIHwlInit(const AddrClient* pClient) +{ + return CIAddrLib::CreateObj(pClient); +} + +/** +*************************************************************************************************** +* CIAddrLib::CIAddrLib +* +* @brief +* Constructor +* +*************************************************************************************************** +*/ +CIAddrLib::CIAddrLib(const AddrClient* pClient) : + SIAddrLib(pClient), + m_noOfMacroEntries(0), + m_allowNonDispThickModes(FALSE) +{ + m_class = CI_ADDRLIB; + memset(&m_settings, 0, sizeof(m_settings)); +} + +/** +*************************************************************************************************** +* CIAddrLib::~CIAddrLib +* +* @brief +* Destructor +*************************************************************************************************** +*/ +CIAddrLib::~CIAddrLib() +{ +} + +/** +*************************************************************************************************** +* CIAddrLib::HwlComputeDccInfo +* +* @brief +* Compute DCC key size, base alignment +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE CIAddrLib::HwlComputeDccInfo( + const ADDR_COMPUTE_DCCINFO_INPUT* pIn, + ADDR_COMPUTE_DCCINFO_OUTPUT* pOut) const +{ + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + if (m_settings.isVolcanicIslands && IsMacroTiled(pIn->tileMode)) + { + UINT_64 dccFastClearSize = pIn->colorSurfSize >> 8; + + ADDR_ASSERT(0 == (pIn->colorSurfSize & 0xff)); + + if (pIn->numSamples > 1) + { + UINT_32 tileSizePerSample = BITS_TO_BYTES(pIn->bpp * MicroTileWidth * MicroTileHeight); + UINT_32 samplesPerSplit = pIn->tileInfo.tileSplitBytes / tileSizePerSample; + + if (samplesPerSplit < pIn->numSamples) + { + UINT_32 numSplits = pIn->numSamples / samplesPerSplit; + UINT_32 fastClearBaseAlign = HwlGetPipes(&pIn->tileInfo) * m_pipeInterleaveBytes; + + ADDR_ASSERT(IsPow2(fastClearBaseAlign)); + + dccFastClearSize /= numSplits; + + if (0 != (dccFastClearSize & (fastClearBaseAlign - 1))) + { + // Disable dcc fast clear + // if key size of fisrt sample split is not pipe*interleave aligned + dccFastClearSize = 0; + } + } + } + + pOut->dccRamSize = pIn->colorSurfSize >> 8; + pOut->dccRamBaseAlign = pIn->tileInfo.banks * + HwlGetPipes(&pIn->tileInfo) * + m_pipeInterleaveBytes; + pOut->dccFastClearSize = dccFastClearSize; + + ADDR_ASSERT(IsPow2(pOut->dccRamBaseAlign)); + + if (0 == (pOut->dccRamSize & (pOut->dccRamBaseAlign - 1))) + { + pOut->subLvlCompressible = TRUE; + } + else + { + UINT_64 dccRamSizeAlign = HwlGetPipes(&pIn->tileInfo) * m_pipeInterleaveBytes; + + if (pOut->dccRamSize == pOut->dccFastClearSize) + { + pOut->dccFastClearSize = PowTwoAlign(pOut->dccRamSize, dccRamSizeAlign); + } + pOut->dccRamSize = PowTwoAlign(pOut->dccRamSize, dccRamSizeAlign); + pOut->subLvlCompressible = FALSE; + } + } + else + { + returnCode = ADDR_NOTSUPPORTED; + } + + return returnCode; +} + +/** +*************************************************************************************************** +* CIAddrLib::HwlComputeCmaskAddrFromCoord +* +* @brief +* Compute tc compatible Cmask address from fmask ram address +* +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE CIAddrLib::HwlComputeCmaskAddrFromCoord( + const ADDR_COMPUTE_CMASK_ADDRFROMCOORD_INPUT* pIn, ///< [in] fmask addr/bpp/tile input + ADDR_COMPUTE_CMASK_ADDRFROMCOORD_OUTPUT* pOut ///< [out] cmask address + ) const +{ + ADDR_E_RETURNCODE returnCode = ADDR_NOTSUPPORTED; + + if ((m_settings.isVolcanicIslands == TRUE) && + (pIn->flags.tcCompatible == TRUE)) + { + UINT_32 numOfPipes = HwlGetPipes(pIn->pTileInfo); + UINT_32 numOfBanks = pIn->pTileInfo->banks; + UINT_64 fmaskAddress = pIn->fmaskAddr; + UINT_32 elemBits = pIn->bpp; + UINT_32 blockByte = 64 * elemBits / 8; + UINT_64 metaNibbleAddress = HwlComputeMetadataNibbleAddress(fmaskAddress, + 0, + 0, + 4, + elemBits, + blockByte, + m_pipeInterleaveBytes, + numOfPipes, + numOfBanks, + 1); + pOut->addr = (metaNibbleAddress >> 1); + pOut->bitPosition = (metaNibbleAddress % 2) ? 4 : 0; + returnCode = ADDR_OK; + } + + return returnCode; +} +/** +*************************************************************************************************** +* CIAddrLib::HwlConvertChipFamily +* +* @brief +* Convert familyID defined in atiid.h to AddrChipFamily and set m_chipFamily/m_chipRevision +* @return +* AddrChipFamily +*************************************************************************************************** +*/ +AddrChipFamily CIAddrLib::HwlConvertChipFamily( + UINT_32 uChipFamily, ///< [in] chip family defined in atiih.h + UINT_32 uChipRevision) ///< [in] chip revision defined in "asic_family"_id.h +{ + AddrChipFamily family = ADDR_CHIP_FAMILY_CI; + + switch (uChipFamily) + { + case FAMILY_CI: + m_settings.isSeaIsland = 1; + m_settings.isBonaire = ASICREV_IS_BONAIRE_M(uChipRevision); + m_settings.isHawaii = ASICREV_IS_HAWAII_P(uChipRevision); + break; + case FAMILY_KV: + m_settings.isKaveri = 1; + m_settings.isSpectre = ASICREV_IS_SPECTRE(uChipRevision); + m_settings.isSpooky = ASICREV_IS_SPOOKY(uChipRevision); + m_settings.isKalindi = ASICREV_IS_KALINDI(uChipRevision); + break; + case FAMILY_VI: + m_settings.isVolcanicIslands = 1; + m_settings.isIceland = ASICREV_IS_ICELAND_M(uChipRevision); + m_settings.isTonga = ASICREV_IS_TONGA_P(uChipRevision); + break; + case FAMILY_CZ: + m_settings.isCarrizo = 1; + m_settings.isVolcanicIslands = 1; + break; + default: + ADDR_ASSERT(!"This should be a unexpected Fusion"); + break; + } + + return family; +} + +/** +*************************************************************************************************** +* CIAddrLib::HwlInitGlobalParams +* +* @brief +* Initializes global parameters +* +* @return +* TRUE if all settings are valid +* +*************************************************************************************************** +*/ +BOOL_32 CIAddrLib::HwlInitGlobalParams( + const ADDR_CREATE_INPUT* pCreateIn) ///< [in] create input +{ + BOOL_32 valid = TRUE; + + const ADDR_REGISTER_VALUE* pRegValue = &pCreateIn->regValue; + + valid = DecodeGbRegs(pRegValue); + + // The following assignments for m_pipes is only for fail-safe, InitTileSettingTable should + // read the correct pipes from tile mode table + if (m_settings.isHawaii) + { + // Hawaii has 16-pipe, see GFXIP_Config_Summary.xls + m_pipes = 16; + } + else if (m_settings.isBonaire || m_settings.isSpectre) + { + m_pipes = 4; + } + else // Treat other KV asics to be 2-pipe + { + m_pipes = 2; + } + + // @todo: VI + // Move this to VI code path once created + if (m_settings.isTonga) + { + m_pipes = 8; + } + else if (m_settings.isIceland) + { + m_pipes = 2; + } + + if (valid) + { + valid = InitTileSettingTable(pRegValue->pTileConfig, pRegValue->noOfEntries); + } + if (valid) + { + valid = InitMacroTileCfgTable(pRegValue->pMacroTileConfig, pRegValue->noOfMacroEntries); + } + + return valid; +} + +/** +*************************************************************************************************** +* CIAddrLib::HwlPostCheckTileIndex +* +* @brief +* Map a tile setting to index if curIndex is invalid, otherwise check if curIndex matches +* tile mode/type/info and change the index if needed +* @return +* Tile index. +*************************************************************************************************** +*/ +INT_32 CIAddrLib::HwlPostCheckTileIndex( + const ADDR_TILEINFO* pInfo, ///< [in] Tile Info + AddrTileMode mode, ///< [in] Tile mode + AddrTileType type, ///< [in] Tile type + INT curIndex ///< [in] Current index assigned in HwlSetupTileInfo + ) const +{ + INT_32 index = curIndex; + + if (mode == ADDR_TM_LINEAR_GENERAL) + { + index = TileIndexLinearGeneral; + } + else + { + BOOL_32 macroTiled = IsMacroTiled(mode); + + // We need to find a new index if either of them is true + // 1. curIndex is invalid + // 2. tile mode is changed + // 3. tile info does not match for macro tiled + if ((index == TileIndexInvalid) || + (mode != m_tileTable[index].mode) || + (macroTiled && pInfo->pipeConfig != m_tileTable[index].info.pipeConfig)) + { + for (index = 0; index < static_cast(m_noOfEntries); index++) + { + if (macroTiled) + { + // macro tile modes need all to match + if ((pInfo->pipeConfig == m_tileTable[index].info.pipeConfig) && + (mode == m_tileTable[index].mode) && + (type == m_tileTable[index].type)) + { + // tileSplitBytes stored in m_tileTable is only valid for depth entries + if (type == ADDR_DEPTH_SAMPLE_ORDER) + { + if (pInfo->tileSplitBytes == m_tileTable[index].info.tileSplitBytes) + { + break; + } + } + else // other entries are determined by other 3 fields + { + break; + } + } + } + else if (mode == ADDR_TM_LINEAR_ALIGNED) + { + // linear mode only needs tile mode to match + if (mode == m_tileTable[index].mode) + { + break; + } + } + else + { + // micro tile modes only need tile mode and tile type to match + if (mode == m_tileTable[index].mode && + type == m_tileTable[index].type) + { + break; + } + } + } + } + } + + ADDR_ASSERT(index < static_cast(m_noOfEntries)); + + if (index >= static_cast(m_noOfEntries)) + { + index = TileIndexInvalid; + } + + return index; +} + +/** +*************************************************************************************************** +* CIAddrLib::HwlSetupTileCfg +* +* @brief +* Map tile index to tile setting. +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE CIAddrLib::HwlSetupTileCfg( + INT_32 index, ///< [in] Tile index + INT_32 macroModeIndex, ///< [in] Index in macro tile mode table(CI) + ADDR_TILEINFO* pInfo, ///< [out] Tile Info + AddrTileMode* pMode, ///< [out] Tile mode + AddrTileType* pType ///< [out] Tile type + ) const +{ + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + // Global flag to control usage of tileIndex + if (UseTileIndex(index)) + { + if (static_cast(index) >= m_noOfEntries) + { + returnCode = ADDR_INVALIDPARAMS; + } + else + { + const ADDR_TILECONFIG* pCfgTable = GetTileSetting(index); + + if (pInfo != NULL) + { + if (IsMacroTiled(pCfgTable->mode)) + { + ADDR_ASSERT(((macroModeIndex != TileIndexInvalid) + && (macroModeIndex != TileIndexNoMacroIndex))); + // Here we used tile_bytes to replace of tile_split + // According info as below: + // "tile_split_c = MIN(ROW_SIZE, tile_split) + // "tile_bytes = MIN(tile_split_c, num_samples * tile_bytes_1x) + // when using tile_bytes replacing of tile_split, the result of + // alignment and others(such as slicesPerTile) are unaffected - + // since if tile_split_c is larger, split won't happen, otherwise + // (num_samples * tile_bytes_1x is larger), a correct tile_split is + // returned. + *pInfo = m_macroTileTable[macroModeIndex]; + + if (pCfgTable->type == ADDR_DEPTH_SAMPLE_ORDER) + { + pInfo->tileSplitBytes = pCfgTable->info.tileSplitBytes; + } + pInfo->pipeConfig = pCfgTable->info.pipeConfig; + } + else // 1D and linear modes, we return default value stored in table + { + *pInfo = pCfgTable->info; + } + } + + if (pMode != NULL) + { + *pMode = pCfgTable->mode; + } + + if (pType != NULL) + { + *pType = pCfgTable->type; + } + } + } + + return returnCode; +} + +/** +*************************************************************************************************** +* CIAddrLib::HwlComputeSurfaceInfo +* +* @brief +* Entry of ci's ComputeSurfaceInfo +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE CIAddrLib::HwlComputeSurfaceInfo( + const ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn, ///< [in] input structure + ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pOut ///< [out] output structure + ) const +{ + // If tileIndex is invalid, force macroModeIndex to be invalid, too + if (pIn->tileIndex == TileIndexInvalid) + { + pOut->macroModeIndex = TileIndexInvalid; + } + + ADDR_E_RETURNCODE retCode = SIAddrLib::HwlComputeSurfaceInfo(pIn,pOut); + + if (pOut->macroModeIndex == TileIndexNoMacroIndex) + { + pOut->macroModeIndex = TileIndexInvalid; + } + + return retCode; +} + +/** +*************************************************************************************************** +* CIAddrLib::HwlFmaskSurfaceInfo +* @brief +* Entry of r800's ComputeFmaskInfo +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE CIAddrLib::HwlComputeFmaskInfo( + const ADDR_COMPUTE_FMASK_INFO_INPUT* pIn, ///< [in] input structure + ADDR_COMPUTE_FMASK_INFO_OUTPUT* pOut ///< [out] output structure + ) +{ + ADDR_E_RETURNCODE retCode = ADDR_OK; + + ADDR_TILEINFO tileInfo = {0}; + ADDR_COMPUTE_FMASK_INFO_INPUT fmaskIn; + fmaskIn = *pIn; + + AddrTileMode tileMode = pIn->tileMode; + + // Use internal tile info if pOut does not have a valid pTileInfo + if (pOut->pTileInfo == NULL) + { + pOut->pTileInfo = &tileInfo; + } + + ADDR_ASSERT(tileMode == ADDR_TM_2D_TILED_THIN1 || + tileMode == ADDR_TM_3D_TILED_THIN1 || + tileMode == ADDR_TM_PRT_TILED_THIN1 || + tileMode == ADDR_TM_PRT_2D_TILED_THIN1 || + tileMode == ADDR_TM_PRT_3D_TILED_THIN1); + + ADDR_ASSERT(m_tileTable[14].mode == ADDR_TM_2D_TILED_THIN1); + ADDR_ASSERT(m_tileTable[15].mode == ADDR_TM_3D_TILED_THIN1); + + // The only valid tile modes for fmask are 2D_THIN1 and 3D_THIN1 plus non-displayable + INT_32 tileIndex = tileMode == ADDR_TM_2D_TILED_THIN1 ? 14 : 15; + ADDR_SURFACE_FLAGS flags = {{0}}; + flags.fmask = 1; + + INT_32 macroModeIndex = TileIndexInvalid; + + UINT_32 numSamples = pIn->numSamples; + UINT_32 numFrags = pIn->numFrags == 0 ? numSamples : pIn->numFrags; + + UINT_32 bpp = QLog2(numFrags); + + // EQAA needs one more bit + if (numSamples > numFrags) + { + bpp++; + } + + if (bpp == 3) + { + bpp = 4; + } + + bpp = Max(8u, bpp * numSamples); + + macroModeIndex = HwlComputeMacroModeIndex(tileIndex, flags, bpp, numSamples, pOut->pTileInfo); + + fmaskIn.tileIndex = tileIndex; + fmaskIn.pTileInfo = pOut->pTileInfo; + pOut->macroModeIndex = macroModeIndex; + pOut->tileIndex = tileIndex; + + retCode = DispatchComputeFmaskInfo(&fmaskIn, pOut); + + if (retCode == ADDR_OK) + { + pOut->tileIndex = + HwlPostCheckTileIndex(pOut->pTileInfo, pIn->tileMode, ADDR_NON_DISPLAYABLE, + pOut->tileIndex); + } + + // Resets pTileInfo to NULL if the internal tile info is used + if (pOut->pTileInfo == &tileInfo) + { + pOut->pTileInfo = NULL; + } + + return retCode; +} + +/** +*************************************************************************************************** +* CIAddrLib::HwlFmaskPreThunkSurfInfo +* +* @brief +* Some preparation before thunking a ComputeSurfaceInfo call for Fmask +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +VOID CIAddrLib::HwlFmaskPreThunkSurfInfo( + const ADDR_COMPUTE_FMASK_INFO_INPUT* pFmaskIn, ///< [in] Input of fmask info + const ADDR_COMPUTE_FMASK_INFO_OUTPUT* pFmaskOut, ///< [in] Output of fmask info + ADDR_COMPUTE_SURFACE_INFO_INPUT* pSurfIn, ///< [out] Input of thunked surface info + ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pSurfOut ///< [out] Output of thunked surface info + ) const +{ + pSurfIn->tileIndex = pFmaskIn->tileIndex; + pSurfOut->macroModeIndex = pFmaskOut->macroModeIndex; +} + +/** +*************************************************************************************************** +* CIAddrLib::HwlFmaskPostThunkSurfInfo +* +* @brief +* Copy hwl extra field after calling thunked ComputeSurfaceInfo +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +VOID CIAddrLib::HwlFmaskPostThunkSurfInfo( + const ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pSurfOut, ///< [in] Output of surface info + ADDR_COMPUTE_FMASK_INFO_OUTPUT* pFmaskOut ///< [out] Output of fmask info + ) const +{ + pFmaskOut->tileIndex = pSurfOut->tileIndex; + pFmaskOut->macroModeIndex = pSurfOut->macroModeIndex; +} + +/** +*************************************************************************************************** +* CIAddrLib::HwlDegradeThickTileMode +* +* @brief +* Degrades valid tile mode for thick modes if needed +* +* @return +* Suitable tile mode +*************************************************************************************************** +*/ +AddrTileMode CIAddrLib::HwlDegradeThickTileMode( + AddrTileMode baseTileMode, ///< [in] base tile mode + UINT_32 numSlices, ///< [in] current number of slices + UINT_32* pBytesPerTile ///< [in/out] pointer to bytes per slice + ) const +{ + return baseTileMode; +} + +/** +*************************************************************************************************** +* CIAddrLib::HwlOverrideTileMode +* +* @brief +* Override THICK to THIN, for specific formats on CI +* +* @return +* Suitable tile mode +* +*************************************************************************************************** +*/ +BOOL_32 CIAddrLib::HwlOverrideTileMode( + const ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn, ///< [in] input structure + AddrTileMode* pTileMode, ///< [in/out] pointer to the tile mode + AddrTileType* pTileType ///< [in/out] pointer to the tile type + ) const +{ + BOOL_32 bOverrided = FALSE; + AddrTileMode tileMode = *pTileMode; + + // currently, all CI/VI family do not + // support ADDR_TM_PRT_2D_TILED_THICK,ADDR_TM_PRT_3D_TILED_THICK and + // ADDR_TM_PRT_2D_TILED_THIN1, ADDR_TM_PRT_3D_TILED_THIN1 + switch (tileMode) + { + case ADDR_TM_PRT_2D_TILED_THICK: + case ADDR_TM_PRT_3D_TILED_THICK: + tileMode = ADDR_TM_PRT_TILED_THICK; + break; + case ADDR_TM_PRT_2D_TILED_THIN1: + case ADDR_TM_PRT_3D_TILED_THIN1: + tileMode = ADDR_TM_PRT_TILED_THIN1; + break; + default: + break; + } + + // UBTS#404321, we do not need such overriding, as THICK+THICK entries removed from the tile-mode table + if (!m_settings.isBonaire) + { + UINT_32 thickness = ComputeSurfaceThickness(tileMode); + + // tile_thickness = (array_mode == XTHICK) ? 8 : ((array_mode == THICK) ? 4 : 1) + if (thickness > 1) + { + switch (pIn->format) + { + // see //gfxip/gcB/devel/cds/src/verif/tc/models/csim/tcp.cpp + // tcpError("Thick micro tiling is not supported for format... + case ADDR_FMT_X24_8_32_FLOAT: + case ADDR_FMT_32_AS_8: + case ADDR_FMT_32_AS_8_8: + case ADDR_FMT_32_AS_32_32_32_32: + + // packed formats + case ADDR_FMT_GB_GR: + case ADDR_FMT_BG_RG: + case ADDR_FMT_1_REVERSED: + case ADDR_FMT_1: + case ADDR_FMT_BC1: + case ADDR_FMT_BC2: + case ADDR_FMT_BC3: + case ADDR_FMT_BC4: + case ADDR_FMT_BC5: + case ADDR_FMT_BC6: + case ADDR_FMT_BC7: + switch (tileMode) + { + case ADDR_TM_1D_TILED_THICK: + tileMode = ADDR_TM_1D_TILED_THIN1; + break; + + case ADDR_TM_2D_TILED_XTHICK: + case ADDR_TM_2D_TILED_THICK: + tileMode = ADDR_TM_2D_TILED_THIN1; + break; + + case ADDR_TM_3D_TILED_XTHICK: + case ADDR_TM_3D_TILED_THICK: + tileMode = ADDR_TM_3D_TILED_THIN1; + break; + + case ADDR_TM_PRT_TILED_THICK: + tileMode = ADDR_TM_PRT_TILED_THIN1; + break; + + case ADDR_TM_PRT_2D_TILED_THICK: + tileMode = ADDR_TM_PRT_2D_TILED_THIN1; + break; + + case ADDR_TM_PRT_3D_TILED_THICK: + tileMode = ADDR_TM_PRT_3D_TILED_THIN1; + break; + + default: + break; + + } + + // Switch tile type from thick to thin + if (tileMode != *pTileMode) + { + // see tileIndex: 13-18 + *pTileType = ADDR_NON_DISPLAYABLE; + } + + break; + default: + break; + } + } + } + + if (tileMode != *pTileMode) + { + *pTileMode = tileMode; + bOverrided = TRUE; + } + + return bOverrided; +} + +/** +*************************************************************************************************** +* CIAddrLib::HwlSetupTileInfo +* +* @brief +* Setup default value of tile info for SI +*************************************************************************************************** +*/ +VOID CIAddrLib::HwlSetupTileInfo( + AddrTileMode tileMode, ///< [in] Tile mode + ADDR_SURFACE_FLAGS flags, ///< [in] Surface type flags + UINT_32 bpp, ///< [in] Bits per pixel + UINT_32 pitch, ///< [in] Pitch in pixels + UINT_32 height, ///< [in] Height in pixels + UINT_32 numSamples, ///< [in] Number of samples + ADDR_TILEINFO* pTileInfoIn, ///< [in] Tile info input: NULL for default + ADDR_TILEINFO* pTileInfoOut, ///< [out] Tile info output + AddrTileType inTileType, ///< [in] Tile type + ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pOut ///< [out] Output + ) const +{ + UINT_32 thickness = ComputeSurfaceThickness(tileMode); + ADDR_TILEINFO* pTileInfo = pTileInfoOut; + INT index = TileIndexInvalid; + INT macroModeIndex = TileIndexInvalid; + + // Fail-safe code + if (!IsLinear(tileMode)) + { + // Thick tile modes must use thick micro tile mode but Bonaire does not support due to + // old derived netlists (UBTS 404321) + if (thickness > 1) + { + if (m_settings.isBonaire) + { + inTileType = ADDR_NON_DISPLAYABLE; + } + else if ((m_allowNonDispThickModes == FALSE) || (inTileType != ADDR_NON_DISPLAYABLE)) + { + inTileType = ADDR_THICK; + } + } + // 128 bpp tiling must be non-displayable. + // Fmask reuse color buffer's entry but bank-height field can be from another entry + // To simplify the logic, fmask entry should be picked from non-displayable ones + else if (bpp == 128 || flags.fmask) + { + inTileType = ADDR_NON_DISPLAYABLE; + } + // These two modes only have non-disp entries though they can be other micro tile modes + else if (tileMode == ADDR_TM_3D_TILED_THIN1 || tileMode == ADDR_TM_PRT_3D_TILED_THIN1) + { + inTileType = ADDR_NON_DISPLAYABLE; + } + + if (flags.depth || flags.stencil) + { + inTileType = ADDR_DEPTH_SAMPLE_ORDER; + } + } + + if (IsTileInfoAllZero(pTileInfo)) + { + // See table entries 0-4 + if (flags.depth || flags.stencil) + { + if (flags.depth && flags.tcCompatible) + { + // tileSize = bpp * numSamples * 8 * 8 / 8 + UINT_32 tileSize = bpp * numSamples * 8; + + // Texure readable depth surface should not be split + switch (tileSize) + { + case 128: + index = 1; + break; + case 256: + index = 2; + break; + case 512: + index = 3; + break; + default: + index = 4; + break; + } + } + else + { + // Depth and stencil need to use the same index, thus the pre-defined tile_split + // can meet the requirement to choose the same macro mode index + // uncompressed depth/stencil are not supported for now + switch (numSamples) + { + case 1: + index = 0; + break; + case 2: + case 4: + index = 1; + break; + case 8: + index = 2; + break; + default: + break; + } + } + } + + // See table entries 5-6 + if (inTileType == ADDR_DEPTH_SAMPLE_ORDER) + { + switch (tileMode) + { + case ADDR_TM_1D_TILED_THIN1: + index = 5; + break; + case ADDR_TM_PRT_TILED_THIN1: + index = 6; + break; + default: + break; + } + } + + // See table entries 8-12 + if (inTileType == ADDR_DISPLAYABLE) + { + switch (tileMode) + { + case ADDR_TM_1D_TILED_THIN1: + index = 9; + break; + case ADDR_TM_2D_TILED_THIN1: + index = 10; + break; + case ADDR_TM_PRT_TILED_THIN1: + index = 11; + break; + default: + break; + } + } + + // See table entries 13-18 + if (inTileType == ADDR_NON_DISPLAYABLE) + { + switch (tileMode) + { + case ADDR_TM_1D_TILED_THIN1: + index = 13; + break; + case ADDR_TM_2D_TILED_THIN1: + index = 14; + break; + case ADDR_TM_3D_TILED_THIN1: + index = 15; + break; + case ADDR_TM_PRT_TILED_THIN1: + index = 16; + break; + default: + break; + } + } + + // See table entries 19-26 + if (thickness > 1) + { + switch (tileMode) + { + case ADDR_TM_1D_TILED_THICK: + //special check for bonaire, for the compatablity between old KMD and new UMD for bonaire + index = ((inTileType == ADDR_THICK) || m_settings.isBonaire) ? 19 : 18; + break; + case ADDR_TM_2D_TILED_THICK: + // special check for bonaire, for the compatablity between old KMD and new UMD for bonaire + index = ((inTileType == ADDR_THICK) || m_settings.isBonaire) ? 20 : 24; + break; + case ADDR_TM_3D_TILED_THICK: + index = 21; + break; + case ADDR_TM_PRT_TILED_THICK: + index = 22; + break; + case ADDR_TM_2D_TILED_XTHICK: + index = 25; + break; + case ADDR_TM_3D_TILED_XTHICK: + index = 26; + break; + default: + break; + } + } + + // See table entries 27-30 + if (inTileType == ADDR_ROTATED) + { + switch (tileMode) + { + case ADDR_TM_1D_TILED_THIN1: + index = 27; + break; + case ADDR_TM_2D_TILED_THIN1: + index = 28; + break; + case ADDR_TM_PRT_TILED_THIN1: + index = 29; + break; + case ADDR_TM_PRT_2D_TILED_THIN1: + index = 30; + break; + default: + break; + } + } + + if (m_pipes >= 8) + { + ADDR_ASSERT((index + 1) < static_cast(m_noOfEntries)); + // Only do this when tile mode table is updated. + if (((tileMode == ADDR_TM_PRT_TILED_THIN1) || (tileMode == ADDR_TM_PRT_TILED_THICK)) && + (m_tileTable[index+1].mode == tileMode)) + { + UINT_32 bytesXSamples = bpp * numSamples / 8; + UINT_32 bytesXThickness = bpp * thickness / 8; + UINT_32 switchP4Threshold = (m_pipes == 16) ? 8 : 32; + + if ((bytesXSamples > switchP4Threshold) || (bytesXThickness > switchP4Threshold)) + { + // Pick next 4 pipe entry + index += 1; + } + } + } + } + else + { + // A pre-filled tile info is ready + index = pOut->tileIndex; + macroModeIndex = pOut->macroModeIndex; + + // pass tile type back for post tile index compute + pOut->tileType = inTileType; + } + + // We only need to set up tile info if there is a valid index but macroModeIndex is invalid + if (index != TileIndexInvalid && macroModeIndex == TileIndexInvalid) + { + macroModeIndex = HwlComputeMacroModeIndex(index, flags, bpp, numSamples, pTileInfo); + + /// Copy to pOut->tileType/tileIndex/macroModeIndex + pOut->tileIndex = index; + pOut->tileType = m_tileTable[index].type; // Or inTileType, the samea + pOut->macroModeIndex = macroModeIndex; + } + else if (tileMode == ADDR_TM_LINEAR_GENERAL) + { + pOut->tileIndex = TileIndexLinearGeneral; + + // Copy linear-aligned entry?? + *pTileInfo = m_tileTable[8].info; + } + else if (tileMode == ADDR_TM_LINEAR_ALIGNED) + { + pOut->tileIndex = 8; + *pTileInfo = m_tileTable[8].info; + } +} + +/** +*************************************************************************************************** +* CIAddrLib::ReadGbTileMode +* +* @brief +* Convert GB_TILE_MODE HW value to ADDR_TILE_CONFIG. +* @return +* NA. +*************************************************************************************************** +*/ +VOID CIAddrLib::ReadGbTileMode( + UINT_32 regValue, ///< [in] GB_TILE_MODE register + ADDR_TILECONFIG* pCfg ///< [out] output structure + ) const +{ + GB_TILE_MODE gbTileMode; + gbTileMode.val = regValue; + + pCfg->type = static_cast(gbTileMode.f.micro_tile_mode_new); + pCfg->info.pipeConfig = static_cast(gbTileMode.f.pipe_config + 1); + + if (pCfg->type == ADDR_DEPTH_SAMPLE_ORDER) + { + pCfg->info.tileSplitBytes = 64 << gbTileMode.f.tile_split; + } + else + { + pCfg->info.tileSplitBytes = 1 << gbTileMode.f.sample_split; + } + + UINT_32 regArrayMode = gbTileMode.f.array_mode; + + pCfg->mode = static_cast(regArrayMode); + + switch (regArrayMode) + { + case 5: + pCfg->mode = ADDR_TM_PRT_TILED_THIN1; + break; + case 6: + pCfg->mode = ADDR_TM_PRT_2D_TILED_THIN1; + break; + case 8: + pCfg->mode = ADDR_TM_2D_TILED_XTHICK; + break; + case 9: + pCfg->mode = ADDR_TM_PRT_TILED_THICK; + break; + case 0xa: + pCfg->mode = ADDR_TM_PRT_2D_TILED_THICK; + break; + case 0xb: + pCfg->mode = ADDR_TM_PRT_3D_TILED_THIN1; + break; + case 0xe: + pCfg->mode = ADDR_TM_3D_TILED_XTHICK; + break; + case 0xf: + pCfg->mode = ADDR_TM_PRT_3D_TILED_THICK; + break; + default: + break; + } + + // Fail-safe code for these always convert tile info, as the non-macro modes + // return the entry of tile mode table directly without looking up macro mode table + if (!IsMacroTiled(pCfg->mode)) + { + pCfg->info.banks = 2; + pCfg->info.bankWidth = 1; + pCfg->info.bankHeight = 1; + pCfg->info.macroAspectRatio = 1; + pCfg->info.tileSplitBytes = 64; + } +} + +/** +*************************************************************************************************** +* CIAddrLib::InitTileSettingTable +* +* @brief +* Initialize the ADDR_TILE_CONFIG table. +* @return +* TRUE if tile table is correctly initialized +*************************************************************************************************** +*/ +BOOL_32 CIAddrLib::InitTileSettingTable( + const UINT_32* pCfg, ///< [in] Pointer to table of tile configs + UINT_32 noOfEntries ///< [in] Numbe of entries in the table above + ) +{ + BOOL_32 initOk = TRUE; + + ADDR_ASSERT(noOfEntries <= TileTableSize); + + memset(m_tileTable, 0, sizeof(m_tileTable)); + + if (noOfEntries != 0) + { + m_noOfEntries = noOfEntries; + } + else + { + m_noOfEntries = TileTableSize; + } + + if (pCfg) // From Client + { + for (UINT_32 i = 0; i < m_noOfEntries; i++) + { + ReadGbTileMode(*(pCfg + i), &m_tileTable[i]); + } + } + else + { + ADDR_ASSERT_ALWAYS(); + initOk = FALSE; + } + + if (initOk) + { + ADDR_ASSERT(m_tileTable[TILEINDEX_LINEAR_ALIGNED].mode == ADDR_TM_LINEAR_ALIGNED); + + if (m_settings.isBonaire == FALSE) + { + // Check if entry 18 is "thick+thin" combination + if ((m_tileTable[18].mode == ADDR_TM_1D_TILED_THICK) && + (m_tileTable[18].type == ADDR_NON_DISPLAYABLE)) + { + m_allowNonDispThickModes = TRUE; + ADDR_ASSERT(m_tileTable[24].mode == ADDR_TM_2D_TILED_THICK); + } + } + else + { + m_allowNonDispThickModes = TRUE; + } + + // Assume the first entry is always programmed with full pipes + m_pipes = HwlGetPipes(&m_tileTable[0].info); + } + + return initOk; +} + +/** +*************************************************************************************************** +* CIAddrLib::ReadGbMacroTileCfg +* +* @brief +* Convert GB_MACRO_TILE_CFG HW value to ADDR_TILE_CONFIG. +* @return +* NA. +*************************************************************************************************** +*/ +VOID CIAddrLib::ReadGbMacroTileCfg( + UINT_32 regValue, ///< [in] GB_MACRO_TILE_MODE register + ADDR_TILEINFO* pCfg ///< [out] output structure + ) const +{ + GB_MACROTILE_MODE gbTileMode; + gbTileMode.val = regValue; + + pCfg->bankHeight = 1 << gbTileMode.f.bank_height; + pCfg->bankWidth = 1 << gbTileMode.f.bank_width; + pCfg->banks = 1 << (gbTileMode.f.num_banks + 1); + pCfg->macroAspectRatio = 1 << gbTileMode.f.macro_tile_aspect; +} + +/** +*************************************************************************************************** +* CIAddrLib::InitMacroTileCfgTable +* +* @brief +* Initialize the ADDR_MACRO_TILE_CONFIG table. +* @return +* TRUE if macro tile table is correctly initialized +*************************************************************************************************** +*/ +BOOL_32 CIAddrLib::InitMacroTileCfgTable( + const UINT_32* pCfg, ///< [in] Pointer to table of tile configs + UINT_32 noOfMacroEntries ///< [in] Numbe of entries in the table above + ) +{ + BOOL_32 initOk = TRUE; + + ADDR_ASSERT(noOfMacroEntries <= MacroTileTableSize); + + memset(m_macroTileTable, 0, sizeof(m_macroTileTable)); + + if (noOfMacroEntries != 0) + { + m_noOfMacroEntries = noOfMacroEntries; + } + else + { + m_noOfMacroEntries = MacroTileTableSize; + } + + if (pCfg) // From Client + { + for (UINT_32 i = 0; i < m_noOfMacroEntries; i++) + { + ReadGbMacroTileCfg(*(pCfg + i), &m_macroTileTable[i]); + + m_macroTileTable[i].tileSplitBytes = 64 << (i % 8); + } + } + else + { + ADDR_ASSERT_ALWAYS(); + initOk = FALSE; + } + return initOk; +} + +/** +*************************************************************************************************** +* CIAddrLib::HwlComputeMacroModeIndex +* +* @brief +* Computes macro tile mode index +* @return +* TRUE if macro tile table is correctly initialized +*************************************************************************************************** +*/ +INT_32 CIAddrLib::HwlComputeMacroModeIndex( + INT_32 tileIndex, ///< [in] Tile mode index + ADDR_SURFACE_FLAGS flags, ///< [in] Surface flags + UINT_32 bpp, ///< [in] Bit per pixel + UINT_32 numSamples, ///< [in] Number of samples + ADDR_TILEINFO* pTileInfo, ///< [out] Pointer to ADDR_TILEINFO + AddrTileMode* pTileMode, ///< [out] Pointer to AddrTileMode + AddrTileType* pTileType ///< [out] Pointer to AddrTileType + ) const +{ + INT_32 macroModeIndex = TileIndexInvalid; + + if (flags.tcCompatible && flags.stencil) + { + // Don't compute macroModeIndex for tc compatible stencil surface + macroModeIndex = TileIndexNoMacroIndex; + } + else + { + AddrTileMode tileMode = m_tileTable[tileIndex].mode; + AddrTileType tileType = m_tileTable[tileIndex].type; + UINT_32 thickness = ComputeSurfaceThickness(tileMode); + + if (!IsMacroTiled(tileMode)) + { + *pTileInfo = m_tileTable[tileIndex].info; + macroModeIndex = TileIndexNoMacroIndex; + } + else + { + UINT_32 tileBytes1x = BITS_TO_BYTES(bpp * MicroTilePixels * thickness); + UINT_32 tileSplit; + + if (m_tileTable[tileIndex].type == ADDR_DEPTH_SAMPLE_ORDER) + { + // Depth entries store real tileSplitBytes + tileSplit = m_tileTable[tileIndex].info.tileSplitBytes; + } + else + { + // Non-depth entries store a split factor + UINT_32 sampleSplit = m_tileTable[tileIndex].info.tileSplitBytes; + UINT_32 colorTileSplit = Max(256u, sampleSplit * tileBytes1x); + + tileSplit = colorTileSplit; + } + + UINT_32 tileSplitC = Min(m_rowSize, tileSplit); + UINT_32 tileBytes; + + if (flags.fmask) + { + tileBytes = Min(tileSplitC, tileBytes1x); + } + else + { + tileBytes = Min(tileSplitC, numSamples * tileBytes1x); + } + + if (tileBytes < 64) + { + tileBytes = 64; + } + + macroModeIndex = Log2(tileBytes / 64); + + if (flags.prt || IsPrtTileMode(tileMode)) + { + // Unknown - assume it is 1/2 of table size + const UINT_32 PrtMacroModeOffset = MacroTileTableSize / 2; + + macroModeIndex += PrtMacroModeOffset; + *pTileInfo = m_macroTileTable[macroModeIndex]; + } + else + { + *pTileInfo = m_macroTileTable[macroModeIndex]; + } + + pTileInfo->pipeConfig = m_tileTable[tileIndex].info.pipeConfig; + + if (m_tileTable[tileIndex].type != ADDR_DEPTH_SAMPLE_ORDER) + { + pTileInfo->tileSplitBytes = tileSplitC; + } + else + { + pTileInfo->tileSplitBytes = m_tileTable[tileIndex].info.tileSplitBytes; + } + } + + if (NULL != pTileMode) + { + *pTileMode = tileMode; + } + + if (NULL != pTileType) + { + *pTileType = tileType; + } + } + + return macroModeIndex; +} + +/** +*************************************************************************************************** +* CIAddrLib::HwlComputeTileDataWidthAndHeightLinear +* +* @brief +* Compute the squared cache shape for per-tile data (CMASK and HTILE) for linear layout +* +* @return +* N/A +* +* @note +* MacroWidth and macroHeight are measured in pixels +*************************************************************************************************** +*/ +VOID CIAddrLib::HwlComputeTileDataWidthAndHeightLinear( + UINT_32* pMacroWidth, ///< [out] macro tile width + UINT_32* pMacroHeight, ///< [out] macro tile height + UINT_32 bpp, ///< [in] bits per pixel + ADDR_TILEINFO* pTileInfo ///< [in] tile info + ) const +{ + ADDR_ASSERT(pTileInfo != NULL); + + UINT_32 numTiles; + + switch (pTileInfo->pipeConfig) + { + case ADDR_PIPECFG_P16_32x32_8x16: + case ADDR_PIPECFG_P16_32x32_16x16: + case ADDR_PIPECFG_P8_32x64_32x32: + case ADDR_PIPECFG_P8_32x32_16x32: + case ADDR_PIPECFG_P8_32x32_16x16: + case ADDR_PIPECFG_P8_32x32_8x16: + case ADDR_PIPECFG_P4_32x32: + numTiles = 8; + break; + default: + numTiles = 4; + break; + } + + *pMacroWidth = numTiles * MicroTileWidth; + *pMacroHeight = numTiles * MicroTileHeight; +} + +/** +*************************************************************************************************** +* CIAddrLib::HwlStereoCheckRightOffsetPadding +* +* @brief +* check if the height needs extra padding for stereo right eye offset, to avoid swizzling +* +* @return +* TRUE is the extra padding is needed +* +* @note +* Kalindi (Kabini) is the only one that needs this padding as there is a uncertain +* possible HW issue where the right eye displays incorrectly with some type of swizzles, if +* the right eye offset is not 64KB aligned - EPR#366461 +* Other Kaveri APUs also need the padding according to DXX team's report otherwise +* corruption observed. - EPR#374788 +*************************************************************************************************** +*/ +BOOL_32 CIAddrLib::HwlStereoCheckRightOffsetPadding() const +{ + BOOL_32 bNeedPadding = FALSE; + + if (m_settings.isKaveri) + { + bNeedPadding = TRUE; + } + + return bNeedPadding; +} + +/** +*************************************************************************************************** +* CIAddrLib::HwlComputeMetadataNibbleAddress +* +* @brief +* calculate meta data address based on input information +* +* ¶meter +* uncompressedDataByteAddress - address of a pixel in color surface +* dataBaseByteAddress - base address of color surface +* metadataBaseByteAddress - base address of meta ram +* metadataBitSize - meta key size, 8 for DCC, 4 for cmask +* elementBitSize - element size of color surface +* blockByteSize - compression block size, 256 for DCC +* pipeInterleaveBytes - pipe interleave size +* numOfPipes - number of pipes +* numOfBanks - number of banks +* numOfSamplesPerSplit - number of samples per tile split +* @return +* meta data nibble address (nibble address is used to support DCC compatible cmask) +* +*************************************************************************************************** +*/ +UINT_64 CIAddrLib::HwlComputeMetadataNibbleAddress( + UINT_64 uncompressedDataByteAddress, + UINT_64 dataBaseByteAddress, + UINT_64 metadataBaseByteAddress, + UINT_32 metadataBitSize, + UINT_32 elementBitSize, + UINT_32 blockByteSize, + UINT_32 pipeInterleaveBytes, + UINT_32 numOfPipes, + UINT_32 numOfBanks, + UINT_32 numOfSamplesPerSplit) const +{ + ///-------------------------------------------------------------------------------------------- + /// Get pipe interleave, bank and pipe bits + ///-------------------------------------------------------------------------------------------- + UINT_32 pipeInterleaveBits = Log2(pipeInterleaveBytes); + UINT_32 pipeBits = Log2(numOfPipes); + UINT_32 bankBits = Log2(numOfBanks); + + ///-------------------------------------------------------------------------------------------- + /// Clear pipe and bank swizzles + ///-------------------------------------------------------------------------------------------- + UINT_32 dataMacrotileBits = pipeInterleaveBits + pipeBits + bankBits; + UINT_32 metadataMacrotileBits = pipeInterleaveBits + pipeBits + bankBits; + + UINT_64 dataMacrotileClearMask = ~((1L << dataMacrotileBits) - 1); + UINT_64 metadataMacrotileClearMask = ~((1L << metadataMacrotileBits) - 1); + + UINT_64 dataBaseByteAddressNoSwizzle = dataBaseByteAddress & dataMacrotileClearMask; + UINT_64 metadataBaseByteAddressNoSwizzle = metadataBaseByteAddress & metadataMacrotileClearMask; + + ///-------------------------------------------------------------------------------------------- + /// Modify metadata base before adding in so that when final address is divided by data ratio, + /// the base address returns to where it should be + ///-------------------------------------------------------------------------------------------- + ADDR_ASSERT((0 != metadataBitSize)); + UINT_64 metadataBaseShifted = metadataBaseByteAddressNoSwizzle * blockByteSize * 8 / + metadataBitSize; + UINT_64 offset = uncompressedDataByteAddress - + dataBaseByteAddressNoSwizzle + + metadataBaseShifted; + + ///-------------------------------------------------------------------------------------------- + /// Save bank data bits + ///-------------------------------------------------------------------------------------------- + UINT_32 lsb = pipeBits + pipeInterleaveBits; + UINT_32 msb = bankBits - 1 + lsb; + + UINT_64 bankDataBits = AddrGetBits(offset, msb, lsb); + + ///-------------------------------------------------------------------------------------------- + /// Save pipe data bits + ///-------------------------------------------------------------------------------------------- + lsb = pipeInterleaveBits; + msb = pipeBits - 1 + lsb; + + UINT_64 pipeDataBits = AddrGetBits(offset, msb, lsb); + + ///-------------------------------------------------------------------------------------------- + /// Remove pipe and bank bits + ///-------------------------------------------------------------------------------------------- + lsb = pipeInterleaveBits; + msb = dataMacrotileBits - 1; + + UINT_64 offsetWithoutPipeBankBits = AddrRemoveBits(offset, msb, lsb); + + ADDR_ASSERT((0 != blockByteSize)); + UINT_64 blockInBankpipe = offsetWithoutPipeBankBits / blockByteSize; + + UINT_32 tileSize = 8 * 8 * elementBitSize/8 * numOfSamplesPerSplit; + UINT_32 blocksInTile = tileSize / blockByteSize; + + if (0 == blocksInTile) + { + lsb = 0; + } + else + { + lsb = Log2(blocksInTile); + } + msb = bankBits - 1 + lsb; + + UINT_64 blockInBankpipeWithBankBits = AddrInsertBits(blockInBankpipe, bankDataBits, msb, lsb); + + /// NOTE *2 because we are converting to Nibble address in this step + UINT_64 metaAddressInPipe = blockInBankpipeWithBankBits * 2 * metadataBitSize / 8; + + + ///-------------------------------------------------------------------------------------------- + /// Reinsert pipe bits back into the final address + ///-------------------------------------------------------------------------------------------- + lsb = pipeInterleaveBits + 1; ///<+1 due to Nibble address now gives interleave bits extra lsb. + msb = pipeBits - 1 + lsb; + UINT_64 metadataAddress = AddrInsertBits(metaAddressInPipe, pipeDataBits, msb, lsb); + + return metadataAddress; +} + +/** +*************************************************************************************************** +* CIAddrLib::HwlPadDimensions +* +* @brief +* Helper function to pad dimensions +* +* @return +* N/A +* +*************************************************************************************************** +*/ +VOID CIAddrLib::HwlPadDimensions( + AddrTileMode tileMode, ///< [in] tile mode + UINT_32 bpp, ///< [in] bits per pixel + ADDR_SURFACE_FLAGS flags, ///< [in] surface flags + UINT_32 numSamples, ///< [in] number of samples + ADDR_TILEINFO* pTileInfo, ///< [in/out] bank structure. + UINT_32 padDims, ///< [in] Dimensions to pad valid value 1,2,3 + UINT_32 mipLevel, ///< [in] MipLevel + UINT_32* pPitch, ///< [in/out] pitch in pixels + UINT_32 pitchAlign, ///< [in] pitch alignment + UINT_32* pHeight, ///< [in/out] height in pixels + UINT_32 heightAlign, ///< [in] height alignment + UINT_32* pSlices, ///< [in/out] number of slices + UINT_32 sliceAlign ///< [in] number of slice alignment + ) const +{ + if (m_settings.isVolcanicIslands && + flags.dccCompatible && + (numSamples > 1) && + (mipLevel == 0) && + IsMacroTiled(tileMode)) + { + UINT_32 tileSizePerSample = BITS_TO_BYTES(bpp * MicroTileWidth * MicroTileHeight); + UINT_32 samplesPerSplit = pTileInfo->tileSplitBytes / tileSizePerSample; + + if (samplesPerSplit < numSamples) + { + UINT_32 dccFastClearByteAlign = HwlGetPipes(pTileInfo) * m_pipeInterleaveBytes * 256; + UINT_32 bytesPerSplit = BITS_TO_BYTES((*pPitch) * (*pHeight) * bpp * samplesPerSplit); + + ADDR_ASSERT(IsPow2(dccFastClearByteAlign)); + + if (0 != (bytesPerSplit & (dccFastClearByteAlign - 1))) + { + UINT_32 dccFastClearPixelAlign = dccFastClearByteAlign / + BITS_TO_BYTES(bpp) / + samplesPerSplit; + UINT_32 macroTilePixelAlign = pitchAlign * heightAlign; + + if ((dccFastClearPixelAlign >= macroTilePixelAlign) && + ((dccFastClearPixelAlign % macroTilePixelAlign) == 0)) + { + UINT_32 dccFastClearPitchAlignInMacroTile = + dccFastClearPixelAlign / macroTilePixelAlign; + UINT_32 heightInMacroTile = *pHeight / heightAlign; + UINT_32 dccFastClearPitchAlignInPixels; + + while ((heightInMacroTile > 1) && + ((heightInMacroTile % 2) == 0) && + (dccFastClearPitchAlignInMacroTile > 1) && + ((dccFastClearPitchAlignInMacroTile % 2) == 0)) + { + heightInMacroTile >>= 1; + dccFastClearPitchAlignInMacroTile >>= 1; + } + + dccFastClearPitchAlignInPixels = pitchAlign * dccFastClearPitchAlignInMacroTile; + + if (IsPow2(dccFastClearPitchAlignInPixels)) + { + *pPitch = PowTwoAlign((*pPitch), dccFastClearPitchAlignInPixels); + } + else + { + *pPitch += (dccFastClearPitchAlignInPixels - 1); + *pPitch /= dccFastClearPitchAlignInPixels; + *pPitch *= dccFastClearPitchAlignInPixels; + } + } + } + } + } +} + diff --git a/src/gallium/winsys/radeon/amdgpu/addrlib/r800/ciaddrlib.h b/src/gallium/winsys/radeon/amdgpu/addrlib/r800/ciaddrlib.h new file mode 100644 index 0000000..0220736 --- /dev/null +++ b/src/gallium/winsys/radeon/amdgpu/addrlib/r800/ciaddrlib.h @@ -0,0 +1,197 @@ +/* + * Copyright © 2014 Advanced Micro Devices, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES + * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NON-INFRINGEMENT. IN NO EVENT SHALL THE COPYRIGHT HOLDERS, AUTHORS + * AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + */ + +/** +*************************************************************************************************** +* @file ciaddrlib.h +* @brief Contains the CIAddrLib class definition. +*************************************************************************************************** +*/ + +#ifndef __CI_ADDR_LIB_H__ +#define __CI_ADDR_LIB_H__ + +#include "addrlib.h" +#include "siaddrlib.h" + +/** +*************************************************************************************************** +* @brief CI specific settings structure. +*************************************************************************************************** +*/ +struct CIChipSettings +{ + struct + { + UINT_32 isSeaIsland : 1; + UINT_32 isBonaire : 1; + UINT_32 isKaveri : 1; + UINT_32 isSpectre : 1; + UINT_32 isSpooky : 1; + UINT_32 isKalindi : 1; + // Hawaii is GFXIP 7.2, similar with CI (Bonaire) + UINT_32 isHawaii : 1; + + // VI + UINT_32 isVolcanicIslands : 1; + UINT_32 isIceland : 1; + UINT_32 isTonga : 1; + // VI fusion (Carrizo) + UINT_32 isCarrizo : 1; + }; +}; + +/** +*************************************************************************************************** +* @brief This class is the CI specific address library +* function set. +*************************************************************************************************** +*/ +class CIAddrLib : public SIAddrLib +{ +public: + /// Creates CIAddrLib object + static AddrLib* CreateObj(const AddrClient* pClient) + { + return new(pClient) CIAddrLib(pClient); + } + +private: + CIAddrLib(const AddrClient* pClient); + virtual ~CIAddrLib(); + +protected: + + // Hwl interface - defined in AddrLib + virtual ADDR_E_RETURNCODE HwlComputeSurfaceInfo( + const ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn, + ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pOut) const; + + virtual ADDR_E_RETURNCODE HwlComputeFmaskInfo( + const ADDR_COMPUTE_FMASK_INFO_INPUT* pIn, + ADDR_COMPUTE_FMASK_INFO_OUTPUT* pOut); + + virtual AddrChipFamily HwlConvertChipFamily( + UINT_32 uChipFamily, UINT_32 uChipRevision); + + virtual BOOL_32 HwlInitGlobalParams( + const ADDR_CREATE_INPUT* pCreateIn); + + virtual ADDR_E_RETURNCODE HwlSetupTileCfg( + INT_32 index, INT_32 macroModeIndex, ADDR_TILEINFO* pInfo, + AddrTileMode* pMode = 0, AddrTileType* pType = 0) const; + + virtual VOID HwlComputeTileDataWidthAndHeightLinear( + UINT_32* pMacroWidth, UINT_32* pMacroHeight, + UINT_32 bpp, ADDR_TILEINFO* pTileInfo) const; + + virtual INT_32 HwlComputeMacroModeIndex( + INT_32 tileIndex, ADDR_SURFACE_FLAGS flags, UINT_32 bpp, UINT_32 numSamples, + ADDR_TILEINFO* pTileInfo, AddrTileMode* pTileMode = NULL, AddrTileType* pTileType = NULL + ) const; + + // Sub-hwl interface - defined in EgBasedAddrLib + virtual VOID HwlSetupTileInfo( + AddrTileMode tileMode, ADDR_SURFACE_FLAGS flags, + UINT_32 bpp, UINT_32 pitch, UINT_32 height, UINT_32 numSamples, + ADDR_TILEINFO* inputTileInfo, ADDR_TILEINFO* outputTileInfo, + AddrTileType inTileType, ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pOut) const; + + virtual INT_32 HwlPostCheckTileIndex( + const ADDR_TILEINFO* pInfo, AddrTileMode mode, AddrTileType type, + INT curIndex = TileIndexInvalid) const; + + virtual VOID HwlFmaskPreThunkSurfInfo( + const ADDR_COMPUTE_FMASK_INFO_INPUT* pFmaskIn, + const ADDR_COMPUTE_FMASK_INFO_OUTPUT* pFmaskOut, + ADDR_COMPUTE_SURFACE_INFO_INPUT* pSurfIn, + ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pSurfOut) const; + + virtual VOID HwlFmaskPostThunkSurfInfo( + const ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pSurfOut, + ADDR_COMPUTE_FMASK_INFO_OUTPUT* pFmaskOut) const; + + virtual AddrTileMode HwlDegradeThickTileMode( + AddrTileMode baseTileMode, UINT_32 numSlices, UINT_32* pBytesPerTile) const; + + virtual BOOL_32 HwlOverrideTileMode( + const ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn, + AddrTileMode* pTileMode, + AddrTileType* pTileType) const; + + virtual BOOL_32 HwlStereoCheckRightOffsetPadding() const; + + virtual ADDR_E_RETURNCODE HwlComputeDccInfo( + const ADDR_COMPUTE_DCCINFO_INPUT* pIn, + ADDR_COMPUTE_DCCINFO_OUTPUT* pOut) const; + + virtual ADDR_E_RETURNCODE HwlComputeCmaskAddrFromCoord( + const ADDR_COMPUTE_CMASK_ADDRFROMCOORD_INPUT* pIn, + ADDR_COMPUTE_CMASK_ADDRFROMCOORD_OUTPUT* pOut) const; + +protected: + virtual VOID HwlPadDimensions( + AddrTileMode tileMode, UINT_32 bpp, ADDR_SURFACE_FLAGS flags, + UINT_32 numSamples, ADDR_TILEINFO* pTileInfo, UINT_32 padDims, UINT_32 mipLevel, + UINT_32* pPitch, UINT_32 pitchAlign, UINT_32* pHeight, UINT_32 heightAlign, + UINT_32* pSlices, UINT_32 sliceAlign) const; + +private: + VOID ReadGbTileMode( + UINT_32 regValue, ADDR_TILECONFIG* pCfg) const; + + VOID ReadGbMacroTileCfg( + UINT_32 regValue, ADDR_TILEINFO* pCfg) const; + + BOOL_32 InitTileSettingTable( + const UINT_32 *pSetting, UINT_32 noOfEntries); + + BOOL_32 InitMacroTileCfgTable( + const UINT_32 *pSetting, UINT_32 noOfEntries); + + UINT_64 HwlComputeMetadataNibbleAddress( + UINT_64 uncompressedDataByteAddress, + UINT_64 dataBaseByteAddress, + UINT_64 metadataBaseByteAddress, + UINT_32 metadataBitSize, + UINT_32 elementBitSize, + UINT_32 blockByteSize, + UINT_32 pipeInterleaveBytes, + UINT_32 numOfPipes, + UINT_32 numOfBanks, + UINT_32 numOfSamplesPerSplit) const; + + static const UINT_32 MacroTileTableSize = 16; + ADDR_TILEINFO m_macroTileTable[MacroTileTableSize]; + UINT_32 m_noOfMacroEntries; + BOOL_32 m_allowNonDispThickModes; + + CIChipSettings m_settings; +}; + +#endif + + diff --git a/src/gallium/winsys/radeon/amdgpu/addrlib/r800/egbaddrlib.cpp b/src/gallium/winsys/radeon/amdgpu/addrlib/r800/egbaddrlib.cpp new file mode 100644 index 0000000..2e16cb3 --- /dev/null +++ b/src/gallium/winsys/radeon/amdgpu/addrlib/r800/egbaddrlib.cpp @@ -0,0 +1,4578 @@ +/* + * Copyright © 2014 Advanced Micro Devices, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES + * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NON-INFRINGEMENT. IN NO EVENT SHALL THE COPYRIGHT HOLDERS, AUTHORS + * AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + */ + +/** +*************************************************************************************************** +* @file egbaddrlib.cpp +* @brief Contains the EgBasedAddrLib class implementation +*************************************************************************************************** +*/ + +#include "egbaddrlib.h" + +/** +*************************************************************************************************** +* EgBasedAddrLib::EgBasedAddrLib +* +* @brief +* Constructor +* +* @note +* +*************************************************************************************************** +*/ +EgBasedAddrLib::EgBasedAddrLib(const AddrClient* pClient) : + AddrLib(pClient), + m_ranks(0), + m_logicalBanks(0), + m_bankInterleave(1) +{ +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::~EgBasedAddrLib +* +* @brief +* Destructor +*************************************************************************************************** +*/ +EgBasedAddrLib::~EgBasedAddrLib() +{ +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::DispatchComputeSurfaceInfo +* +* @brief +* Compute surface sizes include padded pitch,height,slices,total size in bytes, +* meanwhile output suitable tile mode and base alignment might be changed in this +* call as well. Results are returned through output parameters. +* +* @return +* TRUE if no error occurs +*************************************************************************************************** +*/ +BOOL_32 EgBasedAddrLib::DispatchComputeSurfaceInfo( + const ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn, ///< [in] input structure + ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pOut ///< [out] output structure + ) const +{ + AddrTileMode tileMode = pIn->tileMode; + UINT_32 bpp = pIn->bpp; + UINT_32 numSamples = pIn->numSamples; + UINT_32 numFrags = ((pIn->numFrags == 0) ? numSamples : pIn->numFrags); + UINT_32 pitch = pIn->width; + UINT_32 height = pIn->height; + UINT_32 numSlices = pIn->numSlices; + UINT_32 mipLevel = pIn->mipLevel; + ADDR_SURFACE_FLAGS flags = pIn->flags; + + ADDR_TILEINFO tileInfoDef = {0}; + ADDR_TILEINFO* pTileInfo = &tileInfoDef; + + UINT_32 padDims = 0; + BOOL_32 valid; + + tileMode = DegradeLargeThickTile(tileMode, bpp); + + // Only override numSamples for NI above + if (m_chipFamily >= ADDR_CHIP_FAMILY_NI) + { + if (numFrags != numSamples) // This means EQAA + { + // The real surface size needed is determined by number of fragments + numSamples = numFrags; + } + + // Save altered numSamples in pOut + pOut->numSamples = numSamples; + } + + // Caller makes sure pOut->pTileInfo is not NULL, see HwlComputeSurfaceInfo + ADDR_ASSERT(pOut->pTileInfo); + + if (pOut->pTileInfo != NULL) + { + pTileInfo = pOut->pTileInfo; + } + + // Set default values + if (pIn->pTileInfo != NULL) + { + if (pTileInfo != pIn->pTileInfo) + { + *pTileInfo = *pIn->pTileInfo; + } + } + else + { + memset(pTileInfo, 0, sizeof(ADDR_TILEINFO)); + } + + // For macro tile mode, we should calculate default tiling parameters + HwlSetupTileInfo(tileMode, + flags, + bpp, + pitch, + height, + numSamples, + pIn->pTileInfo, + pTileInfo, + pIn->tileType, + pOut); + + if (flags.cube) + { + if (mipLevel == 0) + { + padDims = 2; + } + + if (numSlices == 1) + { + // This is calculating one face, remove cube flag + flags.cube = 0; + } + } + + switch (tileMode) + { + case ADDR_TM_LINEAR_GENERAL://fall through + case ADDR_TM_LINEAR_ALIGNED: + valid = ComputeSurfaceInfoLinear(pIn, pOut, padDims); + break; + + case ADDR_TM_1D_TILED_THIN1://fall through + case ADDR_TM_1D_TILED_THICK: + valid = ComputeSurfaceInfoMicroTiled(pIn, pOut, padDims, tileMode); + break; + + case ADDR_TM_2D_TILED_THIN1: //fall through + case ADDR_TM_2D_TILED_THICK: //fall through + case ADDR_TM_3D_TILED_THIN1: //fall through + case ADDR_TM_3D_TILED_THICK: //fall through + case ADDR_TM_2D_TILED_XTHICK: //fall through + case ADDR_TM_3D_TILED_XTHICK: //fall through + case ADDR_TM_PRT_TILED_THIN1: //fall through + case ADDR_TM_PRT_2D_TILED_THIN1://fall through + case ADDR_TM_PRT_3D_TILED_THIN1://fall through + case ADDR_TM_PRT_TILED_THICK: //fall through + case ADDR_TM_PRT_2D_TILED_THICK://fall through + case ADDR_TM_PRT_3D_TILED_THICK: + valid = ComputeSurfaceInfoMacroTiled(pIn, pOut, padDims, tileMode); + break; + + default: + valid = FALSE; + ADDR_ASSERT_ALWAYS(); + break; + } + + return valid; +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::ComputeSurfaceInfoLinear +* +* @brief +* Compute linear surface sizes include padded pitch, height, slices, total size in +* bytes, meanwhile alignments as well. Since it is linear mode, so output tile mode +* will not be changed here. Results are returned through output parameters. +* +* @return +* TRUE if no error occurs +*************************************************************************************************** +*/ +BOOL_32 EgBasedAddrLib::ComputeSurfaceInfoLinear( + const ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn, ///< [in] Input structure + ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pOut, ///< [out] Output structure + UINT_32 padDims ///< [in] Dimensions to padd + ) const +{ + UINT_32 expPitch = pIn->width; + UINT_32 expHeight = pIn->height; + UINT_32 expNumSlices = pIn->numSlices; + + // No linear MSAA on real H/W, keep this for TGL + UINT_32 numSamples = pOut->numSamples; + + const UINT_32 microTileThickness = 1; + + // + // Compute the surface alignments. + // + ComputeSurfaceAlignmentsLinear(pIn->tileMode, + pIn->bpp, + pIn->flags, + &pOut->baseAlign, + &pOut->pitchAlign, + &pOut->heightAlign); + + if ((pIn->tileMode == ADDR_TM_LINEAR_GENERAL) && pIn->flags.color && (pIn->height > 1)) + { +#if !ALT_TEST + // When linear_general surface is accessed in multiple lines, it requires 8 pixels in pitch + // alignment since PITCH_TILE_MAX is in unit of 8 pixels. + // It is OK if it is accessed per line. + ADDR_ASSERT((pIn->width % 8) == 0); +#endif + } + + pOut->depthAlign = microTileThickness; + + expPitch = HwlPreHandleBaseLvl3xPitch(pIn, expPitch); + + // + // Pad pitch and height to the required granularities. + // + PadDimensions(pIn->tileMode, + pIn->bpp, + pIn->flags, + numSamples, + pOut->pTileInfo, + padDims, + pIn->mipLevel, + &expPitch, pOut->pitchAlign, + &expHeight, pOut->heightAlign, + &expNumSlices, microTileThickness); + + expPitch = HwlPostHandleBaseLvl3xPitch(pIn, expPitch); + + // + // Adjust per HWL + // + + UINT_64 logicalSliceSize; + + logicalSliceSize = HwlGetSizeAdjustmentLinear(pIn->tileMode, + pIn->bpp, + numSamples, + pOut->baseAlign, + pOut->pitchAlign, + &expPitch, + &expHeight, + &pOut->heightAlign); + + + pOut->pitch = expPitch; + pOut->height = expHeight; + pOut->depth = expNumSlices; + + pOut->surfSize = logicalSliceSize * expNumSlices; + + pOut->tileMode = pIn->tileMode; + + return TRUE; +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::ComputeSurfaceInfoMicroTiled +* +* @brief +* Compute 1D/Micro Tiled surface sizes include padded pitch, height, slices, total +* size in bytes, meanwhile alignments as well. Results are returned through output +* parameters. +* +* @return +* TRUE if no error occurs +*************************************************************************************************** +*/ +BOOL_32 EgBasedAddrLib::ComputeSurfaceInfoMicroTiled( + const ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn, ///< [in] Input structure + ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pOut, ///< [out] Output structure + UINT_32 padDims, ///< [in] Dimensions to padd + AddrTileMode expTileMode ///< [in] Expected tile mode + ) const +{ + BOOL_32 valid = TRUE; + + UINT_32 microTileThickness; + UINT_32 expPitch = pIn->width; + UINT_32 expHeight = pIn->height; + UINT_32 expNumSlices = pIn->numSlices; + + // No 1D MSAA on real H/W, keep this for TGL + UINT_32 numSamples = pOut->numSamples; + + // + // Compute the micro tile thickness. + // + microTileThickness = ComputeSurfaceThickness(expTileMode); + + // + // Extra override for mip levels + // + if (pIn->mipLevel > 0) + { + // + // Reduce tiling mode from thick to thin if the number of slices is less than the + // micro tile thickness. + // + if ((expTileMode == ADDR_TM_1D_TILED_THICK) && + (expNumSlices < ThickTileThickness)) + { + expTileMode = HwlDegradeThickTileMode(ADDR_TM_1D_TILED_THICK, expNumSlices, NULL); + if (expTileMode != ADDR_TM_1D_TILED_THICK) + { + microTileThickness = 1; + } + } + } + + // + // Compute the surface restrictions. + // + ComputeSurfaceAlignmentsMicroTiled(expTileMode, + pIn->bpp, + pIn->flags, + numSamples, + &pOut->baseAlign, + &pOut->pitchAlign, + &pOut->heightAlign); + + pOut->depthAlign = microTileThickness; + + // + // Pad pitch and height to the required granularities. + // Compute surface size. + // Return parameters. + // + PadDimensions(expTileMode, + pIn->bpp, + pIn->flags, + numSamples, + pOut->pTileInfo, + padDims, + pIn->mipLevel, + &expPitch, pOut->pitchAlign, + &expHeight, pOut->heightAlign, + &expNumSlices, microTileThickness); + + // + // Get HWL specific pitch adjustment + // + UINT_64 logicalSliceSize = HwlGetSizeAdjustmentMicroTiled(microTileThickness, + pIn->bpp, + pIn->flags, + numSamples, + pOut->baseAlign, + pOut->pitchAlign, + &expPitch, + &expHeight); + + + pOut->pitch = expPitch; + pOut->height = expHeight; + pOut->depth = expNumSlices; + + pOut->surfSize = logicalSliceSize * expNumSlices; + + pOut->tileMode = expTileMode; + + return valid; +} + + +/** +*************************************************************************************************** +* EgBasedAddrLib::ComputeSurfaceInfoMacroTiled +* +* @brief +* Compute 2D/macro tiled surface sizes include padded pitch, height, slices, total +* size in bytes, meanwhile output suitable tile mode and alignments might be changed +* in this call as well. Results are returned through output parameters. +* +* @return +* TRUE if no error occurs +*************************************************************************************************** +*/ +BOOL_32 EgBasedAddrLib::ComputeSurfaceInfoMacroTiled( + const ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn, ///< [in] Input structure + ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pOut, ///< [out] Output structure + UINT_32 padDims, ///< [in] Dimensions to padd + AddrTileMode expTileMode ///< [in] Expected tile mode + ) const +{ + BOOL_32 valid = TRUE; + + AddrTileMode origTileMode = expTileMode; + UINT_32 microTileThickness; + + UINT_32 paddedPitch; + UINT_32 paddedHeight; + UINT_64 bytesPerSlice; + + UINT_32 expPitch = pIn->width; + UINT_32 expHeight = pIn->height; + UINT_32 expNumSlices = pIn->numSlices; + + UINT_32 numSamples = pOut->numSamples; + + // + // Compute the surface restrictions as base + // SanityCheckMacroTiled is called in ComputeSurfaceAlignmentsMacroTiled + // + valid = ComputeSurfaceAlignmentsMacroTiled(expTileMode, + pIn->bpp, + pIn->flags, + pIn->mipLevel, + numSamples, + pOut->pTileInfo, + &pOut->baseAlign, + &pOut->pitchAlign, + &pOut->heightAlign); + + if (valid) + { + // + // Compute the micro tile thickness. + // + microTileThickness = ComputeSurfaceThickness(expTileMode); + + // + // Find the correct tiling mode for mip levels + // + if (pIn->mipLevel > 0) + { + // + // Try valid tile mode + // + expTileMode = ComputeSurfaceMipLevelTileMode(expTileMode, + pIn->bpp, + expPitch, + expHeight, + expNumSlices, + numSamples, + pOut->pitchAlign, + pOut->heightAlign, + pOut->pTileInfo); + + if (!IsMacroTiled(expTileMode)) // Downgraded to micro-tiled + { + return ComputeSurfaceInfoMicroTiled(pIn, pOut, padDims, expTileMode); + } + else + { + if (microTileThickness != ComputeSurfaceThickness(expTileMode)) + { + // + // Re-compute if thickness changed since bank-height may be changed! + // + return ComputeSurfaceInfoMacroTiled(pIn, pOut, padDims, expTileMode); + } + } + } + + paddedPitch = expPitch; + paddedHeight = expHeight; + + // + // Re-cal alignment + // + if (expTileMode != origTileMode) // Tile mode is changed but still macro-tiled + { + valid = ComputeSurfaceAlignmentsMacroTiled(expTileMode, + pIn->bpp, + pIn->flags, + pIn->mipLevel, + numSamples, + pOut->pTileInfo, + &pOut->baseAlign, + &pOut->pitchAlign, + &pOut->heightAlign); + } + + // + // Do padding + // + PadDimensions(expTileMode, + pIn->bpp, + pIn->flags, + numSamples, + pOut->pTileInfo, + padDims, + pIn->mipLevel, + &paddedPitch, pOut->pitchAlign, + &paddedHeight, pOut->heightAlign, + &expNumSlices, microTileThickness); + + if (pIn->flags.qbStereo && + (pOut->pStereoInfo != NULL) && + HwlStereoCheckRightOffsetPadding()) + { + // Eye height's bank bits are different from y == 0? + // Since 3D rendering treats right eye buffer starting from y == "eye height" while + // display engine treats it to be 0, so the bank bits may be different, we pad + // more in height to make sure y == "eye height" has the same bank bits as y == 0. + UINT_32 checkMask = pOut->pTileInfo->banks - 1; + UINT_32 bankBits = 0; + do + { + bankBits = (paddedHeight / 8 / pOut->pTileInfo->bankHeight) & checkMask; + + if (bankBits) + { + paddedHeight += pOut->heightAlign; + } + } while (bankBits); + } + + // + // Compute the size of a slice. + // + bytesPerSlice = BITS_TO_BYTES(static_cast(paddedPitch) * + paddedHeight * NextPow2(pIn->bpp) * numSamples); + + pOut->pitch = paddedPitch; + // Put this check right here to workaround special mipmap cases which the original height + // is needed. + // The original height is pre-stored in pOut->height in PostComputeMipLevel and + // pOut->pitch is needed in HwlCheckLastMacroTiledLvl, too. + if (m_configFlags.checkLast2DLevel && numSamples == 1) // Don't check MSAA + { + // Set a TRUE in pOut if next Level is the first 1D sub level + HwlCheckLastMacroTiledLvl(pIn, pOut); + } + pOut->height = paddedHeight; + + pOut->depth = expNumSlices; + + pOut->surfSize = bytesPerSlice * expNumSlices; + + pOut->tileMode = expTileMode; + + pOut->depthAlign = microTileThickness; + + } // if (valid) + + return valid; +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::ComputeSurfaceAlignmentsLinear +* +* @brief +* Compute linear surface alignment, calculation results are returned through +* output parameters. +* +* @return +* TRUE if no error occurs +*************************************************************************************************** +*/ +BOOL_32 EgBasedAddrLib::ComputeSurfaceAlignmentsLinear( + AddrTileMode tileMode, ///< [in] tile mode + UINT_32 bpp, ///< [in] bits per pixel + ADDR_SURFACE_FLAGS flags, ///< [in] surface flags + UINT_32* pBaseAlign, ///< [out] base address alignment in bytes + UINT_32* pPitchAlign, ///< [out] pitch alignment in pixels + UINT_32* pHeightAlign ///< [out] height alignment in pixels + ) const +{ + BOOL_32 valid = TRUE; + + switch (tileMode) + { + case ADDR_TM_LINEAR_GENERAL: + // + // The required base alignment and pitch and height granularities is to 1 element. + // + *pBaseAlign = (bpp > 8) ? bpp / 8 : 1; + *pPitchAlign = 1; + *pHeightAlign = 1; + break; + case ADDR_TM_LINEAR_ALIGNED: + // + // The required alignment for base is the pipe interleave size. + // The required granularity for pitch is hwl dependent. + // The required granularity for height is one row. + // + *pBaseAlign = m_pipeInterleaveBytes; + *pPitchAlign = HwlGetPitchAlignmentLinear(bpp, flags); + *pHeightAlign = 1; + break; + default: + *pBaseAlign = 1; + *pPitchAlign = 1; + *pHeightAlign = 1; + ADDR_UNHANDLED_CASE(); + break; + } + + AdjustPitchAlignment(flags, pPitchAlign); + + return valid; +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::ComputeSurfaceAlignmentsMicroTiled +* +* @brief +* Compute 1D tiled surface alignment, calculation results are returned through +* output parameters. +* +* @return +* TRUE if no error occurs +*************************************************************************************************** +*/ +BOOL_32 EgBasedAddrLib::ComputeSurfaceAlignmentsMicroTiled( + AddrTileMode tileMode, ///< [in] tile mode + UINT_32 bpp, ///< [in] bits per pixel + ADDR_SURFACE_FLAGS flags, ///< [in] surface flags + UINT_32 numSamples, ///< [in] number of samples + UINT_32* pBaseAlign, ///< [out] base address alignment in bytes + UINT_32* pPitchAlign, ///< [out] pitch alignment in pixels + UINT_32* pHeightAlign ///< [out] height alignment in pixels + ) const +{ + BOOL_32 valid = TRUE; + + // + // The required alignment for base is the pipe interleave size. + // + *pBaseAlign = m_pipeInterleaveBytes; + + *pPitchAlign = HwlGetPitchAlignmentMicroTiled(tileMode, bpp, flags, numSamples); + + *pHeightAlign = MicroTileHeight; + + AdjustPitchAlignment(flags, pPitchAlign); + + // ECR#393489 + // Workaround 2 for 1D tiling - There is HW bug for Carrizo + // where it requires the following alignments for 1D tiling. + if (flags.czDispCompatible) + { + *pBaseAlign = PowTwoAlign(*pBaseAlign, 4096); //Base address MOD 4096 = 0 + *pPitchAlign = PowTwoAlign(*pPitchAlign, 512 >> (BITS_TO_BYTES(bpp))); //(8 lines * pitch * bytes per pixel) MOD 4096 = 0 + } + // end Carrizo workaround for 1D tilling + + return valid; +} + + +/** +*************************************************************************************************** +* EgBasedAddrLib::HwlReduceBankWidthHeight +* +* @brief +* Additional checks, reduce bankHeight/bankWidth if needed and possible +* tileSize*BANK_WIDTH*BANK_HEIGHT <= ROW_SIZE +* +* @return +* TRUE if no error occurs +*************************************************************************************************** +*/ +BOOL_32 EgBasedAddrLib::HwlReduceBankWidthHeight( + UINT_32 tileSize, ///< [in] tile size + UINT_32 bpp, ///< [in] bits per pixel + ADDR_SURFACE_FLAGS flags, ///< [in] surface flags + UINT_32 numSamples, ///< [in] number of samples + UINT_32 bankHeightAlign, ///< [in] bank height alignment + UINT_32 pipes, ///< [in] pipes + ADDR_TILEINFO* pTileInfo ///< [in/out] bank structure. + ) const +{ + UINT_32 macroAspectAlign; + BOOL_32 valid = TRUE; + + if (tileSize * pTileInfo->bankWidth * pTileInfo->bankHeight > m_rowSize) + { + BOOL_32 stillGreater = TRUE; + + // Try reducing bankWidth first + if (stillGreater && pTileInfo->bankWidth > 1) + { + while (stillGreater && pTileInfo->bankWidth > 0) + { + pTileInfo->bankWidth >>= 1; + + if (pTileInfo->bankWidth == 0) + { + pTileInfo->bankWidth = 1; + break; + } + + stillGreater = + tileSize * pTileInfo->bankWidth * pTileInfo->bankHeight > m_rowSize; + } + + // bankWidth is reduced above, so we need to recalculate bankHeight and ratio + bankHeightAlign = Max(1u, + m_pipeInterleaveBytes * m_bankInterleave / + (tileSize * pTileInfo->bankWidth) + ); + + // We cannot increase bankHeight so just assert this case. + ADDR_ASSERT((pTileInfo->bankHeight % bankHeightAlign) == 0); + + if (numSamples == 1) + { + macroAspectAlign = Max(1u, + m_pipeInterleaveBytes * m_bankInterleave / + (tileSize * pipes * pTileInfo->bankWidth) + ); + pTileInfo->macroAspectRatio = PowTwoAlign(pTileInfo->macroAspectRatio, + macroAspectAlign); + } + } + + // Early quit bank_height degradation for "64" bit z buffer + if (flags.depth && bpp >= 64) + { + stillGreater = FALSE; + } + + // Then try reducing bankHeight + if (stillGreater && pTileInfo->bankHeight > bankHeightAlign) + { + while (stillGreater && pTileInfo->bankHeight > bankHeightAlign) + { + pTileInfo->bankHeight >>= 1; + + if (pTileInfo->bankHeight < bankHeightAlign) + { + pTileInfo->bankHeight = bankHeightAlign; + break; + } + + stillGreater = + tileSize * pTileInfo->bankWidth * pTileInfo->bankHeight > m_rowSize; + } + } + + valid = !stillGreater; + + // Generate a warning if we still fail to meet this constraint + if (!valid) + { + ADDR_WARN( + 0, ("TILE_SIZE(%d)*BANK_WIDTH(%d)*BANK_HEIGHT(%d) <= ROW_SIZE(%d)", + tileSize, pTileInfo->bankWidth, pTileInfo->bankHeight, m_rowSize)); + } + } + + return valid; +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::ComputeSurfaceAlignmentsMacroTiled +* +* @brief +* Compute 2D tiled surface alignment, calculation results are returned through +* output parameters. +* +* @return +* TRUE if no error occurs +*************************************************************************************************** +*/ +BOOL_32 EgBasedAddrLib::ComputeSurfaceAlignmentsMacroTiled( + AddrTileMode tileMode, ///< [in] tile mode + UINT_32 bpp, ///< [in] bits per pixel + ADDR_SURFACE_FLAGS flags, ///< [in] surface flags + UINT_32 mipLevel, ///< [in] mip level + UINT_32 numSamples, ///< [in] number of samples + ADDR_TILEINFO* pTileInfo, ///< [in/out] bank structure. + UINT_32* pBaseAlign, ///< [out] base address alignment in bytes + UINT_32* pPitchAlign, ///< [out] pitch alignment in pixels + UINT_32* pHeightAlign ///< [out] height alignment in pixels + ) const +{ + BOOL_32 valid = SanityCheckMacroTiled(pTileInfo); + + if (valid) + { + UINT_32 macroTileWidth; + UINT_32 macroTileHeight; + + UINT_32 tileSize; + UINT_32 bankHeightAlign; + UINT_32 macroAspectAlign; + + UINT_32 thickness = ComputeSurfaceThickness(tileMode); + UINT_32 pipes = HwlGetPipes(pTileInfo); + + // + // Align bank height first according to latest h/w spec + // + + // tile_size = MIN(tile_split, 64 * tile_thickness * element_bytes * num_samples) + tileSize = Min(pTileInfo->tileSplitBytes, + BITS_TO_BYTES(64 * thickness * bpp * numSamples)); + + // bank_height_align = + // MAX(1, (pipe_interleave_bytes * bank_interleave)/(tile_size*bank_width)) + bankHeightAlign = Max(1u, + m_pipeInterleaveBytes * m_bankInterleave / + (tileSize * pTileInfo->bankWidth) + ); + + pTileInfo->bankHeight = PowTwoAlign(pTileInfo->bankHeight, bankHeightAlign); + + // num_pipes * bank_width * macro_tile_aspect >= + // (pipe_interleave_size * bank_interleave) / tile_size + if (numSamples == 1) + { + // this restriction is only for mipmap (mipmap's numSamples must be 1) + macroAspectAlign = Max(1u, + m_pipeInterleaveBytes * m_bankInterleave / + (tileSize * pipes * pTileInfo->bankWidth) + ); + pTileInfo->macroAspectRatio = PowTwoAlign(pTileInfo->macroAspectRatio, macroAspectAlign); + } + + valid = HwlReduceBankWidthHeight(tileSize, + bpp, + flags, + numSamples, + bankHeightAlign, + pipes, + pTileInfo); + + // + // The required granularity for pitch is the macro tile width. + // + macroTileWidth = MicroTileWidth * pTileInfo->bankWidth * pipes * + pTileInfo->macroAspectRatio; + + *pPitchAlign = macroTileWidth; + + AdjustPitchAlignment(flags, pPitchAlign); + + // + // The required granularity for height is the macro tile height. + // + macroTileHeight = MicroTileHeight * pTileInfo->bankHeight * pTileInfo->banks / + pTileInfo->macroAspectRatio; + + *pHeightAlign = macroTileHeight; + + // + // Compute base alignment + // + *pBaseAlign = pipes * + pTileInfo->bankWidth * pTileInfo->banks * pTileInfo->bankHeight * tileSize; + + if ((mipLevel == 0) && (flags.prt) && (m_chipFamily == ADDR_CHIP_FAMILY_SI)) + { + static const UINT_32 PrtTileSize = 0x10000; + + UINT_32 macroTileSize = macroTileWidth * macroTileHeight * numSamples * bpp / 8; + + if (macroTileSize < PrtTileSize) + { + UINT_32 numMacroTiles = PrtTileSize / macroTileSize; + + ADDR_ASSERT((PrtTileSize % macroTileSize) == 0); + + *pPitchAlign *= numMacroTiles; + *pBaseAlign *= numMacroTiles; + } + } + } + + return valid; +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::SanityCheckMacroTiled +* +* @brief +* Check if macro-tiled parameters are valid +* @return +* TRUE if valid +*************************************************************************************************** +*/ +BOOL_32 EgBasedAddrLib::SanityCheckMacroTiled( + ADDR_TILEINFO* pTileInfo ///< [in] macro-tiled parameters + ) const +{ + BOOL_32 valid = TRUE; + UINT_32 numPipes = HwlGetPipes(pTileInfo); + + switch (pTileInfo->banks) + { + case 2: //fall through + case 4: //fall through + case 8: //fall through + case 16: + break; + default: + valid = FALSE; + break; + + } + + if (valid) + { + switch (pTileInfo->bankWidth) + { + case 1: //fall through + case 2: //fall through + case 4: //fall through + case 8: + break; + default: + valid = FALSE; + break; + } + } + + if (valid) + { + switch (pTileInfo->bankHeight) + { + case 1: //fall through + case 2: //fall through + case 4: //fall through + case 8: + break; + default: + valid = FALSE; + break; + } + } + + if (valid) + { + switch (pTileInfo->macroAspectRatio) + { + case 1: //fall through + case 2: //fall through + case 4: //fall through + case 8: + break; + default: + valid = FALSE; + break; + } + } + + if (valid) + { + if (pTileInfo->banks < pTileInfo->macroAspectRatio) + { + // This will generate macro tile height <= 1 + valid = FALSE; + } + } + + if (valid) + { + if (pTileInfo->tileSplitBytes > m_rowSize) + { + valid = FALSE; + } + } + + if (valid) + { + valid = HwlSanityCheckMacroTiled(pTileInfo); + } + + ADDR_ASSERT(valid == TRUE); + + // Add this assert for guidance + ADDR_ASSERT(numPipes * pTileInfo->banks >= 4); + + return valid; +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::ComputeSurfaceMipLevelTileMode +* +* @brief +* Compute valid tile mode for surface mipmap sub-levels +* +* @return +* Suitable tile mode +*************************************************************************************************** +*/ +AddrTileMode EgBasedAddrLib::ComputeSurfaceMipLevelTileMode( + AddrTileMode baseTileMode, ///< [in] base tile mode + UINT_32 bpp, ///< [in] bits per pixels + UINT_32 pitch, ///< [in] current level pitch + UINT_32 height, ///< [in] current level height + UINT_32 numSlices, ///< [in] current number of slices + UINT_32 numSamples, ///< [in] number of samples + UINT_32 pitchAlign, ///< [in] pitch alignment + UINT_32 heightAlign, ///< [in] height alignment + ADDR_TILEINFO* pTileInfo ///< [in] ptr to bank structure + ) const +{ + UINT_64 bytesPerSlice; + UINT_32 bytesPerTile; + + AddrTileMode expTileMode = baseTileMode; + UINT_32 microTileThickness = ComputeSurfaceThickness(expTileMode); + UINT_32 interleaveSize = m_pipeInterleaveBytes * m_bankInterleave; + + // + // Compute the size of a slice. + // + bytesPerSlice = BITS_TO_BYTES(static_cast(pitch) * height * bpp * numSamples); + bytesPerTile = BITS_TO_BYTES(MicroTilePixels * microTileThickness * NextPow2(bpp) * numSamples); + + // + // Reduce tiling mode from thick to thin if the number of slices is less than the + // micro tile thickness. + // + if (numSlices < microTileThickness) + { + expTileMode = HwlDegradeThickTileMode(expTileMode, numSlices, &bytesPerTile); + } + + if (bytesPerTile > pTileInfo->tileSplitBytes) + { + bytesPerTile = pTileInfo->tileSplitBytes; + } + + UINT_32 threshold1 = + bytesPerTile * HwlGetPipes(pTileInfo) * pTileInfo->bankWidth * pTileInfo->macroAspectRatio; + + UINT_32 threshold2 = + bytesPerTile * pTileInfo->bankWidth * pTileInfo->bankHeight; + + // + // Reduce the tile mode from 2D/3D to 1D in following conditions + // + switch (expTileMode) + { + case ADDR_TM_2D_TILED_THIN1: //fall through + case ADDR_TM_3D_TILED_THIN1: + case ADDR_TM_PRT_TILED_THIN1: + case ADDR_TM_PRT_2D_TILED_THIN1: + case ADDR_TM_PRT_3D_TILED_THIN1: + if ((pitch < pitchAlign) || + (height < heightAlign) || + (interleaveSize > threshold1) || + (interleaveSize > threshold2)) + { + expTileMode = ADDR_TM_1D_TILED_THIN1; + } + break; + case ADDR_TM_2D_TILED_THICK: //fall through + case ADDR_TM_3D_TILED_THICK: + case ADDR_TM_2D_TILED_XTHICK: + case ADDR_TM_3D_TILED_XTHICK: + case ADDR_TM_PRT_TILED_THICK: + case ADDR_TM_PRT_2D_TILED_THICK: + case ADDR_TM_PRT_3D_TILED_THICK: + if ((pitch < pitchAlign) || + (height < heightAlign)) + { + expTileMode = ADDR_TM_1D_TILED_THICK; + } + break; + default: + break; + } + + return expTileMode; +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::HwlDegradeBaseLevel +* @brief +* Check if degrade is needed for base level +* @return +* TRUE if degrade is suggested +*************************************************************************************************** +*/ +BOOL_32 EgBasedAddrLib::HwlDegradeBaseLevel( + const ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn) const +{ + BOOL_32 degrade = FALSE; + BOOL_32 valid = TRUE; + + ADDR_ASSERT(IsMacroTiled(pIn->tileMode)); + + UINT_32 baseAlign; + UINT_32 pitchAlign; + UINT_32 heightAlign; + + ADDR_ASSERT(pIn->pTileInfo); + ADDR_TILEINFO tileInfo = *pIn->pTileInfo; + ADDR_COMPUTE_SURFACE_INFO_OUTPUT out = {0}; + + if (UseTileIndex(pIn->tileIndex)) + { + out.tileIndex = pIn->tileIndex; + out.macroModeIndex = TileIndexInvalid; + } + + HwlSetupTileInfo(pIn->tileMode, + pIn->flags, + pIn->bpp, + pIn->width, + pIn->height, + pIn->numSamples, + &tileInfo, + &tileInfo, + pIn->tileType, + &out); + + valid = ComputeSurfaceAlignmentsMacroTiled(pIn->tileMode, + pIn->bpp, + pIn->flags, + pIn->mipLevel, + pIn->numSamples, + &tileInfo, + &baseAlign, + &pitchAlign, + &heightAlign); + + if (valid) + { + degrade = (pIn->width < pitchAlign || pIn->height < heightAlign); + } + else + { + degrade = TRUE; + } + + return degrade; +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::HwlDegradeThickTileMode +* +* @brief +* Degrades valid tile mode for thick modes if needed +* +* @return +* Suitable tile mode +*************************************************************************************************** +*/ +AddrTileMode EgBasedAddrLib::HwlDegradeThickTileMode( + AddrTileMode baseTileMode, ///< [in] base tile mode + UINT_32 numSlices, ///< [in] current number of slices + UINT_32* pBytesPerTile ///< [in/out] pointer to bytes per slice + ) const +{ + ADDR_ASSERT(numSlices < ComputeSurfaceThickness(baseTileMode)); + // if pBytesPerTile is NULL, this is a don't-care.... + UINT_32 bytesPerTile = pBytesPerTile != NULL ? *pBytesPerTile : 64; + + AddrTileMode expTileMode = baseTileMode; + switch (baseTileMode) + { + case ADDR_TM_1D_TILED_THICK: + expTileMode = ADDR_TM_1D_TILED_THIN1; + bytesPerTile >>= 2; + break; + case ADDR_TM_2D_TILED_THICK: + expTileMode = ADDR_TM_2D_TILED_THIN1; + bytesPerTile >>= 2; + break; + case ADDR_TM_3D_TILED_THICK: + expTileMode = ADDR_TM_3D_TILED_THIN1; + bytesPerTile >>= 2; + break; + case ADDR_TM_2D_TILED_XTHICK: + if (numSlices < ThickTileThickness) + { + expTileMode = ADDR_TM_2D_TILED_THIN1; + bytesPerTile >>= 3; + } + else + { + expTileMode = ADDR_TM_2D_TILED_THICK; + bytesPerTile >>= 1; + } + break; + case ADDR_TM_3D_TILED_XTHICK: + if (numSlices < ThickTileThickness) + { + expTileMode = ADDR_TM_3D_TILED_THIN1; + bytesPerTile >>= 3; + } + else + { + expTileMode = ADDR_TM_3D_TILED_THICK; + bytesPerTile >>= 1; + } + break; + default: + ADDR_ASSERT_ALWAYS(); + break; + } + + if (pBytesPerTile != NULL) + { + *pBytesPerTile = bytesPerTile; + } + + return expTileMode; +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::DispatchComputeSurfaceAddrFromCoord +* +* @brief +* Compute surface address from given coord (x, y, slice,sample) +* +* @return +* Address in bytes +*************************************************************************************************** +*/ +UINT_64 EgBasedAddrLib::DispatchComputeSurfaceAddrFromCoord( + const ADDR_COMPUTE_SURFACE_ADDRFROMCOORD_INPUT* pIn, ///< [in] input structure + ADDR_COMPUTE_SURFACE_ADDRFROMCOORD_OUTPUT* pOut ///< [out] output structure + ) const +{ + UINT_32 x = pIn->x; + UINT_32 y = pIn->y; + UINT_32 slice = pIn->slice; + UINT_32 sample = pIn->sample; + UINT_32 bpp = pIn->bpp; + UINT_32 pitch = pIn->pitch; + UINT_32 height = pIn->height; + UINT_32 numSlices = pIn->numSlices; + UINT_32 numSamples = ((pIn->numSamples == 0) ? 1 : pIn->numSamples); + UINT_32 numFrags = ((pIn->numFrags == 0) ? numSamples : pIn->numFrags); + AddrTileMode tileMode = pIn->tileMode; + AddrTileType microTileType = pIn->tileType; + BOOL_32 ignoreSE = pIn->ignoreSE; + BOOL_32 isDepthSampleOrder = pIn->isDepth; + ADDR_TILEINFO* pTileInfo = pIn->pTileInfo; + + UINT_32* pBitPosition = &pOut->bitPosition; + UINT_64 addr; + +#if ADDR_AM_BUILD + UINT_32 addr5Bit = 0; + UINT_32 addr5Swizzle = pIn->addr5Swizzle; + BOOL_32 is32ByteTile = pIn->is32ByteTile; +#endif + + // ADDR_DEPTH_SAMPLE_ORDER = non-disp + depth-sample-order + if (microTileType == ADDR_DEPTH_SAMPLE_ORDER) + { + isDepthSampleOrder = TRUE; + } + + if (m_chipFamily >= ADDR_CHIP_FAMILY_NI) + { + if (numFrags != numSamples) + { + numSamples = numFrags; + ADDR_ASSERT(sample < numSamples); + } + + /// @note + /// 128 bit/thick tiled surface doesn't support display tiling and + /// mipmap chain must have the same tileType, so please fill tileType correctly + if (!IsLinear(pIn->tileMode)) + { + if (bpp >= 128 || ComputeSurfaceThickness(tileMode) > 1) + { + ADDR_ASSERT(microTileType != ADDR_DISPLAYABLE); + } + } + } + + switch (tileMode) + { + case ADDR_TM_LINEAR_GENERAL://fall through + case ADDR_TM_LINEAR_ALIGNED: + addr = ComputeSurfaceAddrFromCoordLinear(x, + y, + slice, + sample, + bpp, + pitch, + height, + numSlices, + pBitPosition); + break; + case ADDR_TM_1D_TILED_THIN1://fall through + case ADDR_TM_1D_TILED_THICK: + addr = ComputeSurfaceAddrFromCoordMicroTiled(x, + y, + slice, + sample, + bpp, + pitch, + height, + numSamples, + tileMode, + microTileType, + isDepthSampleOrder, + pBitPosition); + break; + case ADDR_TM_2D_TILED_THIN1: //fall through + case ADDR_TM_2D_TILED_THICK: //fall through + case ADDR_TM_3D_TILED_THIN1: //fall through + case ADDR_TM_3D_TILED_THICK: //fall through + case ADDR_TM_2D_TILED_XTHICK: //fall through + case ADDR_TM_3D_TILED_XTHICK: //fall through + case ADDR_TM_PRT_TILED_THIN1: //fall through + case ADDR_TM_PRT_2D_TILED_THIN1://fall through + case ADDR_TM_PRT_3D_TILED_THIN1://fall through + case ADDR_TM_PRT_TILED_THICK: //fall through + case ADDR_TM_PRT_2D_TILED_THICK://fall through + case ADDR_TM_PRT_3D_TILED_THICK: + UINT_32 pipeSwizzle; + UINT_32 bankSwizzle; + + if (m_configFlags.useCombinedSwizzle) + { + ExtractBankPipeSwizzle(pIn->tileSwizzle, pIn->pTileInfo, + &bankSwizzle, &pipeSwizzle); + } + else + { + pipeSwizzle = pIn->pipeSwizzle; + bankSwizzle = pIn->bankSwizzle; + } + + addr = ComputeSurfaceAddrFromCoordMacroTiled(x, + y, + slice, + sample, + bpp, + pitch, + height, + numSamples, + tileMode, + microTileType, + ignoreSE, + isDepthSampleOrder, + pipeSwizzle, + bankSwizzle, + pTileInfo, + pBitPosition); + break; + default: + addr = 0; + ADDR_ASSERT_ALWAYS(); + break; + } + +#if ADDR_AM_BUILD + if (m_chipFamily >= ADDR_CHIP_FAMILY_NI) + { + if (addr5Swizzle && isDepthSampleOrder && is32ByteTile) + { + UINT_32 tx = x >> 3; + UINT_32 ty = y >> 3; + UINT_32 tileBits = ((ty&0x3) << 2) | (tx&0x3); + + tileBits = tileBits & addr5Swizzle; + addr5Bit = XorReduce(tileBits, 4); + + addr = addr | static_cast(addr5Bit << 5); + } + } +#endif + + return addr; +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::ComputeSurfaceAddrFromCoordMicroTiled +* +* @brief +* Computes the surface address and bit position from a +* coordinate for 2D tilied (macro tiled) +* @return +* The byte address +*************************************************************************************************** +*/ +UINT_64 EgBasedAddrLib::ComputeSurfaceAddrFromCoordMacroTiled( + UINT_32 x, ///< [in] x coordinate + UINT_32 y, ///< [in] y coordinate + UINT_32 slice, ///< [in] slice index + UINT_32 sample, ///< [in] sample index + UINT_32 bpp, ///< [in] bits per pixel + UINT_32 pitch, ///< [in] surface pitch, in pixels + UINT_32 height, ///< [in] surface height, in pixels + UINT_32 numSamples, ///< [in] number of samples + AddrTileMode tileMode, ///< [in] tile mode + AddrTileType microTileType, ///< [in] micro tiling type + BOOL_32 ignoreSE, ///< [in] TRUE if shader enginers can be ignored + BOOL_32 isDepthSampleOrder, ///< [in] TRUE if it depth sample ordering is used + UINT_32 pipeSwizzle, ///< [in] pipe swizzle + UINT_32 bankSwizzle, ///< [in] bank swizzle + ADDR_TILEINFO* pTileInfo, ///< [in] bank structure + /// **All fields to be valid on entry** + UINT_32* pBitPosition ///< [out] bit position, e.g. FMT_1 will use this + ) const +{ + UINT_64 addr; + + UINT_32 microTileBytes; + UINT_32 microTileBits; + UINT_32 sampleOffset; + UINT_32 pixelIndex; + UINT_32 pixelOffset; + UINT_32 elementOffset; + UINT_32 tileSplitSlice; + UINT_32 pipe; + UINT_32 bank; + UINT_64 sliceBytes; + UINT_64 sliceOffset; + UINT_32 macroTilePitch; + UINT_32 macroTileHeight; + UINT_32 macroTilesPerRow; + UINT_32 macroTilesPerSlice; + UINT_64 macroTileBytes; + UINT_32 macroTileIndexX; + UINT_32 macroTileIndexY; + UINT_64 macroTileOffset; + UINT_64 totalOffset; + UINT_64 pipeInterleaveMask; + UINT_64 bankInterleaveMask; + UINT_64 pipeInterleaveOffset; + UINT_32 bankInterleaveOffset; + UINT_64 offset; + UINT_32 tileRowIndex; + UINT_32 tileColumnIndex; + UINT_32 tileIndex; + UINT_32 tileOffset; + + UINT_32 microTileThickness = ComputeSurfaceThickness(tileMode); + + // + // Compute the number of group, pipe, and bank bits. + // + UINT_32 numPipes = HwlGetPipes(pTileInfo); + UINT_32 numPipeInterleaveBits = Log2(m_pipeInterleaveBytes); + UINT_32 numPipeBits = Log2(numPipes); + UINT_32 numBankInterleaveBits = Log2(m_bankInterleave); + UINT_32 numBankBits = Log2(pTileInfo->banks); + + // + // Compute the micro tile size. + // + microTileBits = MicroTilePixels * microTileThickness * bpp * numSamples; + + microTileBytes = microTileBits / 8; + // + // Compute the pixel index within the micro tile. + // + pixelIndex = ComputePixelIndexWithinMicroTile(x, + y, + slice, + bpp, + tileMode, + microTileType); + + // + // Compute the sample offset and pixel offset. + // + if (isDepthSampleOrder) + { + // + // For depth surfaces, samples are stored contiguously for each element, so the sample + // offset is the sample number times the element size. + // + sampleOffset = sample * bpp; + pixelOffset = pixelIndex * bpp * numSamples; + } + else + { + // + // For color surfaces, all elements for a particular sample are stored contiguously, so + // the sample offset is the sample number times the micro tile size divided yBit the number + // of samples. + // + sampleOffset = sample * (microTileBits / numSamples); + pixelOffset = pixelIndex * bpp; + } + + // + // Compute the element offset. + // + elementOffset = pixelOffset + sampleOffset; + + *pBitPosition = static_cast(elementOffset % 8); + + elementOffset /= 8; //bit-to-byte + + // + // Determine if tiles need to be split across slices. + // + // If the size of the micro tile is larger than the tile split size, then the tile will be + // split across multiple slices. + // + UINT_32 slicesPerTile = 1; + + if ((microTileBytes > pTileInfo->tileSplitBytes) && (microTileThickness == 1)) + { //don't support for thick mode + + // + // Compute the number of slices per tile. + // + slicesPerTile = microTileBytes / pTileInfo->tileSplitBytes; + + // + // Compute the tile split slice number for use in rotating the bank. + // + tileSplitSlice = elementOffset / pTileInfo->tileSplitBytes; + + // + // Adjust the element offset to account for the portion of the tile that is being moved to + // a new slice.. + // + elementOffset %= pTileInfo->tileSplitBytes; + + // + // Adjust the microTileBytes size to tileSplitBytes size since + // a new slice.. + // + microTileBytes = pTileInfo->tileSplitBytes; + } + else + { + tileSplitSlice = 0; + } + + // + // Compute macro tile pitch and height. + // + macroTilePitch = + (MicroTileWidth * pTileInfo->bankWidth * numPipes) * pTileInfo->macroAspectRatio; + macroTileHeight = + (MicroTileHeight * pTileInfo->bankHeight * pTileInfo->banks) / pTileInfo->macroAspectRatio; + + // + // Compute the number of bytes per macro tile. Note: bytes of the same bank/pipe actually + // + macroTileBytes = + static_cast(microTileBytes) * + (macroTilePitch / MicroTileWidth) * (macroTileHeight / MicroTileHeight) / + (numPipes * pTileInfo->banks); + + // + // Compute the number of macro tiles per row. + // + macroTilesPerRow = pitch / macroTilePitch; + + // + // Compute the offset to the macro tile containing the specified coordinate. + // + macroTileIndexX = x / macroTilePitch; + macroTileIndexY = y / macroTileHeight; + macroTileOffset = ((macroTileIndexY * macroTilesPerRow) + macroTileIndexX) * macroTileBytes; + + // + // Compute the number of macro tiles per slice. + // + macroTilesPerSlice = macroTilesPerRow * (height / macroTileHeight); + + // + // Compute the slice size. + // + sliceBytes = macroTilesPerSlice * macroTileBytes; + + // + // Compute the slice offset. + // + sliceOffset = sliceBytes * (tileSplitSlice + slicesPerTile * (slice / microTileThickness)); + + // + // Compute tile offest + // + tileRowIndex = (y / MicroTileHeight) % pTileInfo->bankHeight; + tileColumnIndex = ((x / MicroTileWidth) / numPipes) % pTileInfo->bankWidth; + tileIndex = (tileRowIndex * pTileInfo->bankWidth) + tileColumnIndex; + tileOffset = tileIndex * microTileBytes; + + // + // Combine the slice offset and macro tile offset with the pixel and sample offsets, accounting + // for the pipe and bank bits in the middle of the address. + // + totalOffset = sliceOffset + macroTileOffset + elementOffset + tileOffset; + + // + // Get the pipe and bank. + // + + // when the tileMode is PRT type, then adjust x and y coordinates + if (IsPrtNoRotationTileMode(tileMode)) + { + x = x % macroTilePitch; + y = y % macroTileHeight; + } + + pipe = ComputePipeFromCoord(x, + y, + slice, + tileMode, + pipeSwizzle, + ignoreSE, + pTileInfo); + + bank = ComputeBankFromCoord(x, + y, + slice, + tileMode, + bankSwizzle, + tileSplitSlice, + pTileInfo); + + + // + // Split the offset to put some bits below the pipe+bank bits and some above. + // + pipeInterleaveMask = (1 << numPipeInterleaveBits) - 1; + bankInterleaveMask = (1 << numBankInterleaveBits) - 1; + pipeInterleaveOffset = totalOffset & pipeInterleaveMask; + bankInterleaveOffset = static_cast((totalOffset >> numPipeInterleaveBits) & + bankInterleaveMask); + offset = totalOffset >> (numPipeInterleaveBits + numBankInterleaveBits); + + // + // Assemble the address from its components. + // + addr = pipeInterleaveOffset; + // This is to remove /analyze warnings + UINT_32 pipeBits = pipe << numPipeInterleaveBits; + UINT_32 bankInterleaveBits = bankInterleaveOffset << (numPipeInterleaveBits + numPipeBits); + UINT_32 bankBits = bank << (numPipeInterleaveBits + numPipeBits + + numBankInterleaveBits); + UINT_64 offsetBits = offset << (numPipeInterleaveBits + numPipeBits + + numBankInterleaveBits + numBankBits); + + addr |= pipeBits; + addr |= bankInterleaveBits; + addr |= bankBits; + addr |= offsetBits; + + return addr; +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::ComputeSurfaceAddrFromCoordMicroTiled +* +* @brief +* Computes the surface address and bit position from a coordinate for 1D tilied +* (micro tiled) +* @return +* The byte address +*************************************************************************************************** +*/ +UINT_64 EgBasedAddrLib::ComputeSurfaceAddrFromCoordMicroTiled( + UINT_32 x, ///< [in] x coordinate + UINT_32 y, ///< [in] y coordinate + UINT_32 slice, ///< [in] slice index + UINT_32 sample, ///< [in] sample index + UINT_32 bpp, ///< [in] bits per pixel + UINT_32 pitch, ///< [in] pitch, in pixels + UINT_32 height, ///< [in] height, in pixels + UINT_32 numSamples, ///< [in] number of samples + AddrTileMode tileMode, ///< [in] tile mode + AddrTileType microTileType, ///< [in] micro tiling type + BOOL_32 isDepthSampleOrder, ///< [in] TRUE if depth sample ordering is used + UINT_32* pBitPosition ///< [out] bit position, e.g. FMT_1 will use this + ) const +{ + UINT_64 addr = 0; + + UINT_32 microTileBytes; + UINT_64 sliceBytes; + UINT_32 microTilesPerRow; + UINT_32 microTileIndexX; + UINT_32 microTileIndexY; + UINT_32 microTileIndexZ; + UINT_64 sliceOffset; + UINT_64 microTileOffset; + UINT_32 sampleOffset; + UINT_32 pixelIndex; + UINT_32 pixelOffset; + + UINT_32 microTileThickness = ComputeSurfaceThickness(tileMode); + + // + // Compute the micro tile size. + // + microTileBytes = BITS_TO_BYTES(MicroTilePixels * microTileThickness * bpp * numSamples); + + // + // Compute the slice size. + // + sliceBytes = + BITS_TO_BYTES(static_cast(pitch) * height * microTileThickness * bpp * numSamples); + + // + // Compute the number of micro tiles per row. + // + microTilesPerRow = pitch / MicroTileWidth; + + // + // Compute the micro tile index. + // + microTileIndexX = x / MicroTileWidth; + microTileIndexY = y / MicroTileHeight; + microTileIndexZ = slice / microTileThickness; + + // + // Compute the slice offset. + // + sliceOffset = static_cast(microTileIndexZ) * sliceBytes; + + // + // Compute the offset to the micro tile containing the specified coordinate. + // + microTileOffset = (static_cast(microTileIndexY) * microTilesPerRow + microTileIndexX) * + microTileBytes; + + // + // Compute the pixel index within the micro tile. + // + pixelIndex = ComputePixelIndexWithinMicroTile(x, + y, + slice, + bpp, + tileMode, + microTileType); + + // Compute the sample offset. + // + if (isDepthSampleOrder) + { + // + // For depth surfaces, samples are stored contiguously for each element, so the sample + // offset is the sample number times the element size. + // + sampleOffset = sample * bpp; + pixelOffset = pixelIndex * bpp * numSamples; + } + else + { + // + // For color surfaces, all elements for a particular sample are stored contiguously, so + // the sample offset is the sample number times the micro tile size divided yBit the number + // of samples. + // + sampleOffset = sample * (microTileBytes*8 / numSamples); + pixelOffset = pixelIndex * bpp; + } + + // + // Compute the bit position of the pixel. Each element is stored with one bit per sample. + // + + UINT_32 elemOffset = sampleOffset + pixelOffset; + + *pBitPosition = elemOffset % 8; + elemOffset /= 8; + + // + // Combine the slice offset, micro tile offset, sample offset, and pixel offsets. + // + addr = sliceOffset + microTileOffset + elemOffset; + + return addr; +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::HwlComputePixelCoordFromOffset +* +* @brief +* Compute pixel coordinate from offset inside a micro tile +* @return +* N/A +*************************************************************************************************** +*/ +VOID EgBasedAddrLib::HwlComputePixelCoordFromOffset( + UINT_32 offset, ///< [in] offset inside micro tile in bits + UINT_32 bpp, ///< [in] bits per pixel + UINT_32 numSamples, ///< [in] number of samples + AddrTileMode tileMode, ///< [in] tile mode + UINT_32 tileBase, ///< [in] base offset within a tile + UINT_32 compBits, ///< [in] component bits actually needed(for planar surface) + UINT_32* pX, ///< [out] x coordinate + UINT_32* pY, ///< [out] y coordinate + UINT_32* pSlice, ///< [out] slice index + UINT_32* pSample, ///< [out] sample index + AddrTileType microTileType, ///< [in] micro tiling type + BOOL_32 isDepthSampleOrder ///< [in] TRUE if depth sample order in microtile is used + ) const +{ + UINT_32 x = 0; + UINT_32 y = 0; + UINT_32 z = 0; + UINT_32 thickness = ComputeSurfaceThickness(tileMode); + + // For planar surface, we adjust offset acoording to tile base + if ((bpp != compBits) && (compBits != 0) && isDepthSampleOrder) + { + offset -= tileBase; + + ADDR_ASSERT(microTileType == ADDR_NON_DISPLAYABLE || + microTileType == ADDR_DEPTH_SAMPLE_ORDER); + + bpp = compBits; + } + + UINT_32 sampleTileBits; + UINT_32 samplePixelBits; + UINT_32 pixelIndex; + + if (isDepthSampleOrder) + { + samplePixelBits = bpp * numSamples; + pixelIndex = offset / samplePixelBits; + *pSample = (offset % samplePixelBits) / bpp; + } + else + { + sampleTileBits = MicroTilePixels * bpp * thickness; + *pSample = offset / sampleTileBits; + pixelIndex = (offset % sampleTileBits) / bpp; + } + + if (microTileType != ADDR_THICK) + { + if (microTileType == ADDR_DISPLAYABLE) // displayable + { + switch (bpp) + { + case 8: + x = pixelIndex & 0x7; + y = Bits2Number(3, _BIT(pixelIndex,5),_BIT(pixelIndex,3),_BIT(pixelIndex,4)); + break; + case 16: + x = pixelIndex & 0x7; + y = Bits2Number(3, _BIT(pixelIndex,5),_BIT(pixelIndex,4),_BIT(pixelIndex,3)); + break; + case 32: + x = Bits2Number(3, _BIT(pixelIndex,3),_BIT(pixelIndex,1),_BIT(pixelIndex,0)); + y = Bits2Number(3, _BIT(pixelIndex,5),_BIT(pixelIndex,4),_BIT(pixelIndex,2)); + break; + case 64: + x = Bits2Number(3, _BIT(pixelIndex,3),_BIT(pixelIndex,2),_BIT(pixelIndex,0)); + y = Bits2Number(3, _BIT(pixelIndex,5),_BIT(pixelIndex,4),_BIT(pixelIndex,1)); + break; + case 128: + x = Bits2Number(3, _BIT(pixelIndex,3),_BIT(pixelIndex,2),_BIT(pixelIndex,1)); + y = Bits2Number(3, _BIT(pixelIndex,5),_BIT(pixelIndex,4),_BIT(pixelIndex,0)); + break; + default: + break; + } + } + else if (microTileType == ADDR_NON_DISPLAYABLE || microTileType == ADDR_DEPTH_SAMPLE_ORDER) + { + x = Bits2Number(3, _BIT(pixelIndex,4),_BIT(pixelIndex,2),_BIT(pixelIndex,0)); + y = Bits2Number(3, _BIT(pixelIndex,5),_BIT(pixelIndex,3),_BIT(pixelIndex,1)); + } + else if (microTileType == ADDR_ROTATED) + { + /* + 8-Bit Elements + element_index[5:0] = { x[2], x[0], x[1], y[2], y[1], y[0] } + + 16-Bit Elements + element_index[5:0] = { x[2], x[1], x[0], y[2], y[1], y[0] } + + 32-Bit Elements + element_index[5:0] = { x[2], x[1], y[2], x[0], y[1], y[0] } + + 64-Bit Elements + element_index[5:0] = { y[2], x[2], x[1], y[1], x[0], y[0] } + */ + switch(bpp) + { + case 8: + x = Bits2Number(3, _BIT(pixelIndex,5),_BIT(pixelIndex,3),_BIT(pixelIndex,4)); + y = pixelIndex & 0x7; + break; + case 16: + x = Bits2Number(3, _BIT(pixelIndex,5),_BIT(pixelIndex,4),_BIT(pixelIndex,3)); + y = pixelIndex & 0x7; + break; + case 32: + x = Bits2Number(3, _BIT(pixelIndex,5),_BIT(pixelIndex,4),_BIT(pixelIndex,2)); + y = Bits2Number(3, _BIT(pixelIndex,3),_BIT(pixelIndex,1),_BIT(pixelIndex,0)); + break; + case 64: + x = Bits2Number(3, _BIT(pixelIndex,4),_BIT(pixelIndex,3),_BIT(pixelIndex,1)); + y = Bits2Number(3, _BIT(pixelIndex,5),_BIT(pixelIndex,2),_BIT(pixelIndex,0)); + break; + default: + ADDR_ASSERT_ALWAYS(); + break; + } + } + + if (thickness > 1) // thick + { + z = Bits2Number(3, _BIT(pixelIndex,8),_BIT(pixelIndex,7),_BIT(pixelIndex,6)); + } + } + else + { + ADDR_ASSERT((m_chipFamily >= ADDR_CHIP_FAMILY_CI) && (thickness > 1)); + /* + 8-Bit Elements and 16-Bit Elements + element_index[7:0] = { y[2], x[2], z[1], z[0], y[1], x[1], y[0], x[0] } + + 32-Bit Elements + element_index[7:0] = { y[2], x[2], z[1], y[1], z[0], x[1], y[0], x[0] } + + 64-Bit Elements and 128-Bit Elements + element_index[7:0] = { y[2], x[2], z[1], y[1], x[1], z[0], y[0], x[0] } + + The equation to compute the element index for the extra thick tile: + element_index[8] = z[2] + */ + switch (bpp) + { + case 8: + case 16: // fall-through + x = Bits2Number(3, _BIT(pixelIndex,6),_BIT(pixelIndex,2),_BIT(pixelIndex,0)); + y = Bits2Number(3, _BIT(pixelIndex,7),_BIT(pixelIndex,3),_BIT(pixelIndex,1)); + z = Bits2Number(2, _BIT(pixelIndex,5),_BIT(pixelIndex,4)); + break; + case 32: + x = Bits2Number(3, _BIT(pixelIndex,6),_BIT(pixelIndex,2),_BIT(pixelIndex,0)); + y = Bits2Number(3, _BIT(pixelIndex,7),_BIT(pixelIndex,4),_BIT(pixelIndex,1)); + z = Bits2Number(2, _BIT(pixelIndex,5),_BIT(pixelIndex,3)); + break; + case 64: + case 128: // fall-through + x = Bits2Number(3, _BIT(pixelIndex,6),_BIT(pixelIndex,3),_BIT(pixelIndex,0)); + y = Bits2Number(3, _BIT(pixelIndex,7),_BIT(pixelIndex,4),_BIT(pixelIndex,1)); + z = Bits2Number(2, _BIT(pixelIndex,5),_BIT(pixelIndex,2)); + break; + default: + ADDR_ASSERT_ALWAYS(); + break; + } + + if (thickness == 8) + { + z += Bits2Number(3,_BIT(pixelIndex,8),0,0); + } + } + + *pX = x; + *pY = y; + *pSlice += z; +} + + +/** +*************************************************************************************************** +* EgBasedAddrLib::DispatchComputeSurfaceCoordFromAddrDispatch +* +* @brief +* Compute (x,y,slice,sample) coordinates from surface address +* @return +* N/A +*************************************************************************************************** +*/ +VOID EgBasedAddrLib::DispatchComputeSurfaceCoordFromAddr( + const ADDR_COMPUTE_SURFACE_COORDFROMADDR_INPUT* pIn, ///< [in] input structure + ADDR_COMPUTE_SURFACE_COORDFROMADDR_OUTPUT* pOut ///< [out] output structure + ) const +{ + UINT_64 addr = pIn->addr; + UINT_32 bitPosition = pIn->bitPosition; + UINT_32 bpp = pIn->bpp; + UINT_32 pitch = pIn->pitch; + UINT_32 height = pIn->height; + UINT_32 numSlices = pIn->numSlices; + UINT_32 numSamples = ((pIn->numSamples == 0) ? 1 : pIn->numSamples); + UINT_32 numFrags = ((pIn->numFrags == 0) ? numSamples : pIn->numFrags); + AddrTileMode tileMode = pIn->tileMode; + UINT_32 tileBase = pIn->tileBase; + UINT_32 compBits = pIn->compBits; + AddrTileType microTileType = pIn->tileType; + BOOL_32 ignoreSE = pIn->ignoreSE; + BOOL_32 isDepthSampleOrder = pIn->isDepth; + ADDR_TILEINFO* pTileInfo = pIn->pTileInfo; + + UINT_32* pX = &pOut->x; + UINT_32* pY = &pOut->y; + UINT_32* pSlice = &pOut->slice; + UINT_32* pSample = &pOut->sample; + + if (microTileType == ADDR_DEPTH_SAMPLE_ORDER) + { + isDepthSampleOrder = TRUE; + } + + if (m_chipFamily >= ADDR_CHIP_FAMILY_NI) + { + if (numFrags != numSamples) + { + numSamples = numFrags; + } + + /// @note + /// 128 bit/thick tiled surface doesn't support display tiling and + /// mipmap chain must have the same tileType, so please fill tileType correctly + if (!IsLinear(pIn->tileMode)) + { + if (bpp >= 128 || ComputeSurfaceThickness(tileMode) > 1) + { + ADDR_ASSERT(microTileType != ADDR_DISPLAYABLE); + } + } + } + + switch (tileMode) + { + case ADDR_TM_LINEAR_GENERAL://fall through + case ADDR_TM_LINEAR_ALIGNED: + ComputeSurfaceCoordFromAddrLinear(addr, + bitPosition, + bpp, + pitch, + height, + numSlices, + pX, + pY, + pSlice, + pSample); + break; + case ADDR_TM_1D_TILED_THIN1://fall through + case ADDR_TM_1D_TILED_THICK: + ComputeSurfaceCoordFromAddrMicroTiled(addr, + bitPosition, + bpp, + pitch, + height, + numSamples, + tileMode, + tileBase, + compBits, + pX, + pY, + pSlice, + pSample, + microTileType, + isDepthSampleOrder); + break; + case ADDR_TM_2D_TILED_THIN1: //fall through + case ADDR_TM_2D_TILED_THICK: //fall through + case ADDR_TM_3D_TILED_THIN1: //fall through + case ADDR_TM_3D_TILED_THICK: //fall through + case ADDR_TM_2D_TILED_XTHICK: //fall through + case ADDR_TM_3D_TILED_XTHICK: //fall through + case ADDR_TM_PRT_TILED_THIN1: //fall through + case ADDR_TM_PRT_2D_TILED_THIN1://fall through + case ADDR_TM_PRT_3D_TILED_THIN1://fall through + case ADDR_TM_PRT_TILED_THICK: //fall through + case ADDR_TM_PRT_2D_TILED_THICK://fall through + case ADDR_TM_PRT_3D_TILED_THICK: + UINT_32 pipeSwizzle; + UINT_32 bankSwizzle; + + if (m_configFlags.useCombinedSwizzle) + { + ExtractBankPipeSwizzle(pIn->tileSwizzle, pIn->pTileInfo, + &bankSwizzle, &pipeSwizzle); + } + else + { + pipeSwizzle = pIn->pipeSwizzle; + bankSwizzle = pIn->bankSwizzle; + } + + ComputeSurfaceCoordFromAddrMacroTiled(addr, + bitPosition, + bpp, + pitch, + height, + numSamples, + tileMode, + tileBase, + compBits, + microTileType, + ignoreSE, + isDepthSampleOrder, + pipeSwizzle, + bankSwizzle, + pTileInfo, + pX, + pY, + pSlice, + pSample); + break; + default: + ADDR_ASSERT_ALWAYS(); + } +} + + +/** +*************************************************************************************************** +* EgBasedAddrLib::ComputeSurfaceCoordFromAddrMacroTiled +* +* @brief +* Compute surface coordinates from address for macro tiled surface +* @return +* N/A +*************************************************************************************************** +*/ +VOID EgBasedAddrLib::ComputeSurfaceCoordFromAddrMacroTiled( + UINT_64 addr, ///< [in] byte address + UINT_32 bitPosition, ///< [in] bit position + UINT_32 bpp, ///< [in] bits per pixel + UINT_32 pitch, ///< [in] pitch in pixels + UINT_32 height, ///< [in] height in pixels + UINT_32 numSamples, ///< [in] number of samples + AddrTileMode tileMode, ///< [in] tile mode + UINT_32 tileBase, ///< [in] tile base offset + UINT_32 compBits, ///< [in] component bits (for planar surface) + AddrTileType microTileType, ///< [in] micro tiling type + BOOL_32 ignoreSE, ///< [in] TRUE if shader engines can be ignored + BOOL_32 isDepthSampleOrder, ///< [in] TRUE if depth sample order is used + UINT_32 pipeSwizzle, ///< [in] pipe swizzle + UINT_32 bankSwizzle, ///< [in] bank swizzle + ADDR_TILEINFO* pTileInfo, ///< [in] bank structure. + /// **All fields to be valid on entry** + UINT_32* pX, ///< [out] X coord + UINT_32* pY, ///< [out] Y coord + UINT_32* pSlice, ///< [out] slice index + UINT_32* pSample ///< [out] sample index + ) const +{ + UINT_32 mx; + UINT_32 my; + UINT_64 tileBits; + UINT_64 macroTileBits; + UINT_32 slices; + UINT_32 tileSlices; + UINT_64 elementOffset; + UINT_64 macroTileIndex; + UINT_32 tileIndex; + UINT_64 totalOffset; + + + UINT_32 bank; + UINT_32 pipe; + UINT_32 groupBits = m_pipeInterleaveBytes << 3; + UINT_32 pipes = HwlGetPipes(pTileInfo); + UINT_32 banks = pTileInfo->banks; + + UINT_32 bankInterleave = m_bankInterleave; + + UINT_64 addrBits = BYTES_TO_BITS(addr) + bitPosition; + + // + // remove bits for bank and pipe + // + totalOffset = (addrBits % groupBits) + + (((addrBits / groupBits / pipes) % bankInterleave) * groupBits) + + (((addrBits / groupBits / pipes) / bankInterleave) / banks) * groupBits * bankInterleave; + + UINT_32 microTileThickness = ComputeSurfaceThickness(tileMode); + + UINT_32 microTileBits = bpp * microTileThickness * MicroTilePixels * numSamples; + + UINT_32 microTileBytes = BITS_TO_BYTES(microTileBits); + // + // Determine if tiles need to be split across slices. + // + // If the size of the micro tile is larger than the tile split size, then the tile will be + // split across multiple slices. + // + UINT_32 slicesPerTile = 1; //_State->TileSlices + + if ((microTileBytes > pTileInfo->tileSplitBytes) && (microTileThickness == 1)) + { //don't support for thick mode + + // + // Compute the number of slices per tile. + // + slicesPerTile = microTileBytes / pTileInfo->tileSplitBytes; + } + + tileBits = microTileBits / slicesPerTile; // micro tile bits + + // in micro tiles because not MicroTileWidth timed. + UINT_32 macroWidth = pTileInfo->bankWidth * pipes * pTileInfo->macroAspectRatio; + // in micro tiles as well + UINT_32 macroHeight = pTileInfo->bankHeight * banks / pTileInfo->macroAspectRatio; + + UINT_32 pitchInMacroTiles = pitch / MicroTileWidth / macroWidth; + + macroTileBits = (macroWidth * macroHeight) * tileBits / (banks * pipes); + + macroTileIndex = totalOffset / macroTileBits; + + // pitchMacros * height / heightMacros; macroTilesPerSlice == _State->SliceMacros + UINT_32 macroTilesPerSlice = (pitch / (macroWidth * MicroTileWidth)) * height / + (macroHeight * MicroTileWidth); + + slices = static_cast(macroTileIndex / macroTilesPerSlice); + + *pSlice = static_cast(slices / slicesPerTile * microTileThickness); + + // + // calculate element offset and x[2:0], y[2:0], z[1:0] for thick + // + tileSlices = slices % slicesPerTile; + + elementOffset = tileSlices * tileBits; + elementOffset += totalOffset % tileBits; + + UINT_32 coordZ = 0; + + HwlComputePixelCoordFromOffset(static_cast(elementOffset), + bpp, + numSamples, + tileMode, + tileBase, + compBits, + pX, + pY, + &coordZ, + pSample, + microTileType, + isDepthSampleOrder); + + macroTileIndex = macroTileIndex % macroTilesPerSlice; + *pY += static_cast(macroTileIndex / pitchInMacroTiles * macroHeight * MicroTileHeight); + *pX += static_cast(macroTileIndex % pitchInMacroTiles * macroWidth * MicroTileWidth); + + *pSlice += coordZ; + + tileIndex = static_cast((totalOffset % macroTileBits) / tileBits); + + my = (tileIndex / pTileInfo->bankWidth) % pTileInfo->bankHeight * MicroTileHeight; + mx = (tileIndex % pTileInfo->bankWidth) * pipes * MicroTileWidth; + + *pY += my; + *pX += mx; + + bank = ComputeBankFromAddr(addr, banks, pipes); + pipe = ComputePipeFromAddr(addr, pipes); + + HwlComputeSurfaceCoord2DFromBankPipe(tileMode, + pX, + pY, + *pSlice, + bank, + pipe, + bankSwizzle, + pipeSwizzle, + tileSlices, + ignoreSE, + pTileInfo); +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::ComputeSurfaceCoord2DFromBankPipe +* +* @brief +* Compute surface x,y coordinates from bank/pipe info +* @return +* N/A +*************************************************************************************************** +*/ +VOID EgBasedAddrLib::ComputeSurfaceCoord2DFromBankPipe( + AddrTileMode tileMode, ///< [in] tile mode + UINT_32 x, ///< [in] x coordinate + UINT_32 y, ///< [in] y coordinate + UINT_32 slice, ///< [in] slice index + UINT_32 bank, ///< [in] bank number + UINT_32 pipe, ///< [in] pipe number + UINT_32 bankSwizzle,///< [in] bank swizzle + UINT_32 pipeSwizzle,///< [in] pipe swizzle + UINT_32 tileSlices, ///< [in] slices in a micro tile + ADDR_TILEINFO* pTileInfo, ///< [in] bank structure. **All fields to be valid on entry** + CoordFromBankPipe* pOutput ///< [out] pointer to extracted x/y bits + ) const +{ + UINT_32 yBit3 = 0; + UINT_32 yBit4 = 0; + UINT_32 yBit5 = 0; + UINT_32 yBit6 = 0; + + UINT_32 xBit3 = 0; + UINT_32 xBit4 = 0; + UINT_32 xBit5 = 0; + + UINT_32 tileSplitRotation; + + UINT_32 numPipes = HwlGetPipes(pTileInfo); + + UINT_32 bankRotation = ComputeBankRotation(tileMode, + pTileInfo->banks, numPipes); + + UINT_32 pipeRotation = ComputePipeRotation(tileMode, numPipes); + + UINT_32 xBit = x / (MicroTileWidth * pTileInfo->bankWidth * numPipes); + UINT_32 yBit = y / (MicroTileHeight * pTileInfo->bankHeight); + + //calculate the bank and pipe before rotation and swizzle + + switch (tileMode) + { + case ADDR_TM_2D_TILED_THIN1: //fall through + case ADDR_TM_2D_TILED_THICK: //fall through + case ADDR_TM_2D_TILED_XTHICK: //fall through + case ADDR_TM_3D_TILED_THIN1: //fall through + case ADDR_TM_3D_TILED_THICK: //fall through + case ADDR_TM_3D_TILED_XTHICK: + tileSplitRotation = ((pTileInfo->banks / 2) + 1); + break; + default: + tileSplitRotation = 0; + break; + } + + UINT_32 microTileThickness = ComputeSurfaceThickness(tileMode); + + bank ^= tileSplitRotation * tileSlices; + if (pipeRotation == 0) + { + bank ^= bankRotation * (slice / microTileThickness) + bankSwizzle; + bank %= pTileInfo->banks; + pipe ^= pipeSwizzle; + } + else + { + bank ^= bankRotation * (slice / microTileThickness) / numPipes + bankSwizzle; + bank %= pTileInfo->banks; + pipe ^= pipeRotation * (slice / microTileThickness) + pipeSwizzle; + } + + if (pTileInfo->macroAspectRatio == 1) + { + switch (pTileInfo->banks) + { + case 2: + yBit3 = _BIT(bank, 0) ^ _BIT(xBit,0); + break; + case 4: + yBit4 = _BIT(bank, 0) ^ _BIT(xBit,0); + yBit3 = _BIT(bank, 1) ^ _BIT(xBit,1); + break; + case 8: + yBit3 = _BIT(bank, 2) ^ _BIT(xBit,2); + yBit5 = _BIT(bank, 0) ^ _BIT(xBit,0); + yBit4 = _BIT(bank, 1) ^ _BIT(xBit,1) ^ yBit5; + break; + case 16: + yBit3 = _BIT(bank, 3) ^ _BIT(xBit, 3); + yBit4 = _BIT(bank, 2) ^ _BIT(xBit, 2); + yBit6 = _BIT(bank, 0) ^ _BIT(xBit, 0); + yBit5 = _BIT(bank, 1) ^ _BIT(xBit, 1) ^ yBit6; + break; + default: + break; + } + + } + else if (pTileInfo->macroAspectRatio == 2) + { + switch (pTileInfo->banks) + { + case 2: //xBit3 = yBit3^b0 + xBit3 = _BIT(bank, 0) ^ _BIT(yBit,0); + break; + case 4: //xBit3=yBit4^b0; yBit3=xBit4^b1 + xBit3 = _BIT(bank, 0) ^ _BIT(yBit,1); + yBit3 = _BIT(bank, 1) ^ _BIT(xBit,1); + break; + case 8: //xBit4, xBit5, yBit5 are known + xBit3 = _BIT(bank, 0) ^ _BIT(yBit,2); + yBit3 = _BIT(bank, 2) ^ _BIT(xBit,2); + yBit4 = _BIT(bank, 1) ^ _BIT(xBit,1) ^ _BIT(yBit, 2); + break; + case 16://x4,x5,x6,y6 are known + xBit3 = _BIT(bank, 0) ^ _BIT(yBit, 3); //x3 = y6 ^ b0 + yBit3 = _BIT(bank, 3) ^ _BIT(xBit, 3); //y3 = x6 ^ b3 + yBit4 = _BIT(bank, 2) ^ _BIT(xBit, 2); //y4 = x5 ^ b2 + yBit5 = _BIT(bank, 1) ^ _BIT(xBit, 1) ^ _BIT(yBit, 3); //y5=x4^y6^b1 + break; + default: + break; + } + } + else if (pTileInfo->macroAspectRatio == 4) + { + switch (pTileInfo->banks) + { + case 4: //yBit3, yBit4 + xBit3 = _BIT(bank, 0) ^ _BIT(yBit,1); + xBit4 = _BIT(bank, 1) ^ _BIT(yBit,0); + break; + case 8: //xBit5, yBit4, yBit5 + xBit3 = _BIT(bank, 0) ^ _BIT(yBit,2); + yBit3 = _BIT(bank, 2) ^ _BIT(xBit,2); + xBit4 = _BIT(bank, 1) ^ _BIT(yBit,1) ^ _BIT(yBit,2); + break; + case 16: //xBit5, xBit6, yBit5, yBit6 + xBit3 = _BIT(bank, 0) ^ _BIT(yBit, 3);//x3 = b0 ^ y6 + xBit4 = _BIT(bank, 1) ^ _BIT(yBit, 2) ^ _BIT(yBit, 3);//x4 = b1 ^ y5 ^ y6; + yBit3 = _BIT(bank, 3) ^ _BIT(xBit, 3); //y3 = b3 ^ x6; + yBit4 = _BIT(bank, 2) ^ _BIT(xBit, 2); //y4 = b2 ^ x5; + break; + default: + break; + } + } + else if (pTileInfo->macroAspectRatio == 8) + { + switch (pTileInfo->banks) + { + case 8: //yBit3, yBit4, yBit5 + xBit3 = _BIT(bank, 0) ^ _BIT(yBit,2); //x3 = b0 ^ y5; + xBit4 = _BIT(bank, 1) ^ _BIT(yBit,1) ^ _BIT(yBit, 2);//x4 = b1 ^ y4 ^ y5; + xBit5 = _BIT(bank, 2) ^ _BIT(yBit,0); + break; + case 16: //xBit6, yBit4, yBit5, yBit6 + xBit3 = _BIT(bank, 0) ^ _BIT(yBit, 3);//x3 = y6 ^ b0 + xBit4 = _BIT(bank, 1) ^ _BIT(yBit, 2) ^ _BIT(yBit, 3);//x4 = y5 ^ y6 ^ b1 + xBit5 = _BIT(bank, 2) ^ _BIT(yBit, 1);//x5 = y4 ^ b2 + yBit3 = _BIT(bank, 3) ^ _BIT(xBit, 3); //y3 = x6 ^ b3 + break; + default: + break; + } + } + + pOutput->xBits = xBit; + pOutput->yBits = yBit; + + pOutput->xBit3 = xBit3; + pOutput->xBit4 = xBit4; + pOutput->xBit5 = xBit5; + pOutput->yBit3 = yBit3; + pOutput->yBit4 = yBit4; + pOutput->yBit5 = yBit5; + pOutput->yBit6 = yBit6; +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::HwlExtractBankPipeSwizzle +* @brief +* Entry of EgBasedAddrLib ExtractBankPipeSwizzle +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE EgBasedAddrLib::HwlExtractBankPipeSwizzle( + const ADDR_EXTRACT_BANKPIPE_SWIZZLE_INPUT* pIn, ///< [in] input structure + ADDR_EXTRACT_BANKPIPE_SWIZZLE_OUTPUT* pOut ///< [out] output structure + ) const +{ + ExtractBankPipeSwizzle(pIn->base256b, + pIn->pTileInfo, + &pOut->bankSwizzle, + &pOut->pipeSwizzle); + + return ADDR_OK; +} + + +/** +*************************************************************************************************** +* EgBasedAddrLib::HwlCombineBankPipeSwizzle +* @brief +* Combine bank/pipe swizzle +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE EgBasedAddrLib::HwlCombineBankPipeSwizzle( + UINT_32 bankSwizzle, ///< [in] bank swizzle + UINT_32 pipeSwizzle, ///< [in] pipe swizzle + ADDR_TILEINFO* pTileInfo, ///< [in] tile info + UINT_64 baseAddr, ///< [in] base address + UINT_32* pTileSwizzle ///< [out] combined swizzle + ) const +{ + ADDR_E_RETURNCODE retCode = ADDR_OK; + + if (pTileSwizzle) + { + *pTileSwizzle = GetBankPipeSwizzle(bankSwizzle, pipeSwizzle, baseAddr, pTileInfo); + } + else + { + retCode = ADDR_INVALIDPARAMS; + } + + return retCode; +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::HwlComputeBaseSwizzle +* @brief +* Compute base swizzle +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE EgBasedAddrLib::HwlComputeBaseSwizzle( + const ADDR_COMPUTE_BASE_SWIZZLE_INPUT* pIn, + ADDR_COMPUTE_BASE_SWIZZLE_OUTPUT* pOut + ) const +{ + UINT_32 bankSwizzle = 0; + UINT_32 pipeSwizzle = 0; + ADDR_TILEINFO* pTileInfo = pIn->pTileInfo; + + ADDR_ASSERT(IsMacroTiled(pIn->tileMode)); + ADDR_ASSERT(pIn->pTileInfo); + + /// This is a legacy misreading of h/w doc, use it as it doesn't hurt. + static const UINT_8 bankRotationArray[4][16] = { + { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }, // ADDR_SURF_2_BANK + { 0, 1, 2, 3, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }, // ADDR_SURF_4_BANK + { 0, 3, 6, 1, 4, 7, 2, 5, 0, 0, 0, 0, 0, 0, 0, 0 }, // ADDR_SURF_8_BANK + { 0, 7, 14, 5, 12, 3, 10, 1, 8, 15, 6, 13, 4, 11, 2, 9 }, // ADDR_SURF_16_BANK + }; + + UINT_32 pipes = HwlGetPipes(pTileInfo); + UINT_32 banks = pTileInfo ? pTileInfo->banks : 2; + UINT_32 hwNumBanks; + + // Uses less bank swizzle bits + if (pIn->option.reduceBankBit && banks > 2) + { + banks >>= 1; + } + + switch (banks) + { + case 2: + hwNumBanks = 0; + break; + case 4: + hwNumBanks = 1; + break; + case 8: + hwNumBanks = 2; + break; + case 16: + hwNumBanks = 3; + break; + default: + ADDR_ASSERT_ALWAYS(); + hwNumBanks = 0; + break; + } + + if (pIn->option.genOption == ADDR_SWIZZLE_GEN_LINEAR) + { + bankSwizzle = pIn->surfIndex & (banks - 1); + } + else // (pIn->option.genOption == ADDR_SWIZZLE_GEN_DEFAULT) + { + bankSwizzle = bankRotationArray[hwNumBanks][pIn->surfIndex & (banks - 1)]; + } + + if (IsMacro3dTiled(pIn->tileMode)) + { + pipeSwizzle = pIn->surfIndex & (HwlGetPipes(pTileInfo) - 1); + } + + return HwlCombineBankPipeSwizzle(bankSwizzle, pipeSwizzle, pTileInfo, 0, &pOut->tileSwizzle); +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::ExtractBankPipeSwizzle +* @brief +* Extract bank/pipe swizzle from base256b +* @return +* N/A +*************************************************************************************************** +*/ +VOID EgBasedAddrLib::ExtractBankPipeSwizzle( + UINT_32 base256b, ///< [in] input base256b register value + ADDR_TILEINFO* pTileInfo, ///< [in] 2D tile parameters. Client must provide all data + UINT_32* pBankSwizzle, ///< [out] bank swizzle + UINT_32* pPipeSwizzle ///< [out] pipe swizzle + ) const +{ + UINT_32 bankSwizzle = 0; + UINT_32 pipeSwizzle = 0; + + if (base256b != 0) + { + UINT_32 numPipes = HwlGetPipes(pTileInfo); + UINT_32 bankBits = QLog2(pTileInfo->banks); + UINT_32 pipeBits = QLog2(numPipes); + UINT_32 groupBytes = m_pipeInterleaveBytes; + UINT_32 bankInterleave = m_bankInterleave; + + pipeSwizzle = + (base256b / (groupBytes >> 8)) & ((1<> 8) / numPipes / bankInterleave) & ((1 << bankBits) - 1); + } + + *pPipeSwizzle = pipeSwizzle; + *pBankSwizzle = bankSwizzle; +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::GetBankPipeSwizzle +* @brief +* Combine bank/pipe swizzle +* @return +* Base256b bits (only filled bank/pipe bits) +*************************************************************************************************** +*/ +UINT_32 EgBasedAddrLib::GetBankPipeSwizzle( + UINT_32 bankSwizzle, ///< [in] bank swizzle + UINT_32 pipeSwizzle, ///< [in] pipe swizzle + UINT_64 baseAddr, ///< [in] base address + ADDR_TILEINFO* pTileInfo ///< [in] tile info + ) const +{ + UINT_32 pipeBits = QLog2(HwlGetPipes(pTileInfo)); + UINT_32 bankInterleaveBits = QLog2(m_bankInterleave); + UINT_32 tileSwizzle = pipeSwizzle + ((bankSwizzle << bankInterleaveBits) << pipeBits); + + baseAddr ^= tileSwizzle * m_pipeInterleaveBytes; + baseAddr >>= 8; + + return static_cast(baseAddr); +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::ComputeSliceTileSwizzle +* @brief +* Compute cubemap/3d texture faces/slices tile swizzle +* @return +* Tile swizzle +*************************************************************************************************** +*/ +UINT_32 EgBasedAddrLib::ComputeSliceTileSwizzle( + AddrTileMode tileMode, ///< [in] Tile mode + UINT_32 baseSwizzle, ///< [in] Base swizzle + UINT_32 slice, ///< [in] Slice index, Cubemap face index, 0 means +X + UINT_64 baseAddr, ///< [in] Base address + ADDR_TILEINFO* pTileInfo ///< [in] Bank structure + ) const +{ + UINT_32 tileSwizzle = 0; + + if (IsMacroTiled(tileMode)) // Swizzle only for macro tile mode + { + UINT_32 firstSlice = slice / ComputeSurfaceThickness(tileMode); + + UINT_32 numPipes = HwlGetPipes(pTileInfo); + UINT_32 numBanks = pTileInfo->banks; + + UINT_32 pipeRotation; + UINT_32 bankRotation; + + UINT_32 bankSwizzle = 0; + UINT_32 pipeSwizzle = 0; + + pipeRotation = ComputePipeRotation(tileMode, numPipes); + bankRotation = ComputeBankRotation(tileMode, numBanks, numPipes); + + if (baseSwizzle != 0) + { + ExtractBankPipeSwizzle(baseSwizzle, + pTileInfo, + &bankSwizzle, + &pipeSwizzle); + } + + if (pipeRotation == 0) //2D mode + { + bankSwizzle += firstSlice * bankRotation; + bankSwizzle %= numBanks; + } + else //3D mode + { + pipeSwizzle += firstSlice * pipeRotation; + pipeSwizzle %= numPipes; + bankSwizzle += firstSlice * bankRotation / numPipes; + bankSwizzle %= numBanks; + } + + tileSwizzle = GetBankPipeSwizzle(bankSwizzle, + pipeSwizzle, + baseAddr, + pTileInfo); + } + + return tileSwizzle; +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::HwlComputeQbStereoRightSwizzle +* +* @brief +* Compute right eye swizzle +* @return +* swizzle +*************************************************************************************************** +*/ +UINT_32 EgBasedAddrLib::HwlComputeQbStereoRightSwizzle( + ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pInfo ///< [in] Surface info, must be valid + ) const +{ + UINT_32 bankBits = 0; + UINT_32 swizzle = 0; + + // The assumption is default swizzle for left eye is 0 + if (IsMacroTiled(pInfo->tileMode) && pInfo->pStereoInfo && pInfo->pTileInfo) + { + bankBits = ComputeBankFromCoord(0, pInfo->height, 0, + pInfo->tileMode, 0, 0, pInfo->pTileInfo); + + if (bankBits) + { + HwlCombineBankPipeSwizzle(bankBits, 0, pInfo->pTileInfo, 0, &swizzle); + } + } + + return swizzle; +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::ComputeBankFromCoord +* +* @brief +* Compute bank number from coordinates +* @return +* Bank number +*************************************************************************************************** +*/ +UINT_32 EgBasedAddrLib::ComputeBankFromCoord( + UINT_32 x, ///< [in] x coordinate + UINT_32 y, ///< [in] y coordinate + UINT_32 slice, ///< [in] slice index + AddrTileMode tileMode, ///< [in] tile mode + UINT_32 bankSwizzle, ///< [in] bank swizzle + UINT_32 tileSplitSlice, ///< [in] If the size of the pixel offset is larger than the + /// tile split size, then the pixel will be moved to a separate + /// slice. This value equals pixelOffset / tileSplitBytes + /// in this case. Otherwise this is 0. + ADDR_TILEINFO* pTileInfo ///< [in] tile info + ) const +{ + UINT_32 pipes = HwlGetPipes(pTileInfo); + UINT_32 bankBit0 = 0; + UINT_32 bankBit1 = 0; + UINT_32 bankBit2 = 0; + UINT_32 bankBit3 = 0; + UINT_32 sliceRotation; + UINT_32 tileSplitRotation; + UINT_32 bank; + UINT_32 numBanks = pTileInfo->banks; + UINT_32 bankWidth = pTileInfo->bankWidth; + UINT_32 bankHeight = pTileInfo->bankHeight; + + UINT_32 tx = x / MicroTileWidth / (bankWidth * pipes); + UINT_32 ty = y / MicroTileHeight / bankHeight; + + UINT_32 x3 = _BIT(tx,0); + UINT_32 x4 = _BIT(tx,1); + UINT_32 x5 = _BIT(tx,2); + UINT_32 x6 = _BIT(tx,3); + UINT_32 y3 = _BIT(ty,0); + UINT_32 y4 = _BIT(ty,1); + UINT_32 y5 = _BIT(ty,2); + UINT_32 y6 = _BIT(ty,3); + + switch (numBanks) + { + case 16: + bankBit0 = x3 ^ y6; + bankBit1 = x4 ^ y5 ^ y6; + bankBit2 = x5 ^ y4; + bankBit3 = x6 ^ y3; + break; + case 8: + bankBit0 = x3 ^ y5; + bankBit1 = x4 ^ y4 ^ y5; + bankBit2 = x5 ^ y3; + break; + case 4: + bankBit0 = x3 ^ y4; + bankBit1 = x4 ^ y3; + break; + case 2: + bankBit0 = x3 ^ y3; + break; + default: + ADDR_ASSERT_ALWAYS(); + break; + } + + bank = bankBit0 | (bankBit1 << 1) | (bankBit2 << 2) | (bankBit3 << 3); + + //Bits2Number(4, bankBit3, bankBit2, bankBit1, bankBit0); + + bank = HwlPreAdjustBank((x / MicroTileWidth), bank, pTileInfo); + // + // Compute bank rotation for the slice. + // + UINT_32 microTileThickness = ComputeSurfaceThickness(tileMode); + + switch (tileMode) + { + case ADDR_TM_2D_TILED_THIN1: // fall through + case ADDR_TM_2D_TILED_THICK: // fall through + case ADDR_TM_2D_TILED_XTHICK: + sliceRotation = ((numBanks / 2) - 1) * (slice / microTileThickness); + break; + case ADDR_TM_3D_TILED_THIN1: // fall through + case ADDR_TM_3D_TILED_THICK: // fall through + case ADDR_TM_3D_TILED_XTHICK: + sliceRotation = + Max(1u, (pipes / 2) - 1) * (slice / microTileThickness) / pipes; + break; + default: + sliceRotation = 0; + break; + } + + + // + // Compute bank rotation for the tile split slice. + // + // The sample slice will be non-zero if samples must be split across multiple slices. + // This situation arises when the micro tile size multiplied yBit the number of samples exceeds + // the split size (set in GB_ADDR_CONFIG). + // + switch (tileMode) + { + case ADDR_TM_2D_TILED_THIN1: //fall through + case ADDR_TM_3D_TILED_THIN1: //fall through + case ADDR_TM_PRT_2D_TILED_THIN1: //fall through + case ADDR_TM_PRT_3D_TILED_THIN1: //fall through + tileSplitRotation = ((numBanks / 2) + 1) * tileSplitSlice; + break; + default: + tileSplitRotation = 0; + break; + } + + // + // Apply bank rotation for the slice and tile split slice. + // + bank ^= bankSwizzle + sliceRotation; + bank ^= tileSplitRotation; + + bank &= (numBanks - 1); + + return bank; +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::ComputeBankFromAddr +* +* @brief +* Compute the bank number from an address +* @return +* Bank number +*************************************************************************************************** +*/ +UINT_32 EgBasedAddrLib::ComputeBankFromAddr( + UINT_64 addr, ///< [in] address + UINT_32 numBanks, ///< [in] number of banks + UINT_32 numPipes ///< [in] number of pipes + ) const +{ + UINT_32 bank; + + // + // The LSBs of the address are arranged as follows: + // bank | bankInterleave | pipe | pipeInterleave + // + // To get the bank number, shift off the pipe interleave, pipe, and bank interlave bits and + // mask the bank bits. + // + bank = static_cast( + (addr >> Log2(m_pipeInterleaveBytes * numPipes * m_bankInterleave)) & + (numBanks - 1) + ); + + return bank; +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::ComputePipeRotation +* +* @brief +* Compute pipe rotation value +* @return +* Pipe rotation +*************************************************************************************************** +*/ +UINT_32 EgBasedAddrLib::ComputePipeRotation( + AddrTileMode tileMode, ///< [in] tile mode + UINT_32 numPipes ///< [in] number of pipes + ) const +{ + UINT_32 rotation; + + switch (tileMode) + { + case ADDR_TM_3D_TILED_THIN1: //fall through + case ADDR_TM_3D_TILED_THICK: //fall through + case ADDR_TM_3D_TILED_XTHICK: //fall through + case ADDR_TM_PRT_3D_TILED_THIN1: //fall through + case ADDR_TM_PRT_3D_TILED_THICK: + rotation = (numPipes < 4) ? 1 : (numPipes / 2 - 1); + break; + default: + rotation = 0; + } + + return rotation; +} + + + +/** +*************************************************************************************************** +* EgBasedAddrLib::ComputeBankRotation +* +* @brief +* Compute bank rotation value +* @return +* Bank rotation +*************************************************************************************************** +*/ +UINT_32 EgBasedAddrLib::ComputeBankRotation( + AddrTileMode tileMode, ///< [in] tile mode + UINT_32 numBanks, ///< [in] number of banks + UINT_32 numPipes ///< [in] number of pipes + ) const +{ + UINT_32 rotation; + + switch (tileMode) + { + case ADDR_TM_2D_TILED_THIN1: // fall through + case ADDR_TM_2D_TILED_THICK: // fall through + case ADDR_TM_2D_TILED_XTHICK: + case ADDR_TM_PRT_2D_TILED_THIN1: + case ADDR_TM_PRT_2D_TILED_THICK: + // Rotate banks per Z-slice yBit 1 for 4-bank or 3 for 8-bank + rotation = numBanks / 2 - 1; + break; + case ADDR_TM_3D_TILED_THIN1: // fall through + case ADDR_TM_3D_TILED_THICK: // fall through + case ADDR_TM_3D_TILED_XTHICK: + case ADDR_TM_PRT_3D_TILED_THIN1: + case ADDR_TM_PRT_3D_TILED_THICK: + rotation = (numPipes < 4) ? 1 : (numPipes / 2 - 1); // rotate pipes & banks + break; + default: + rotation = 0; + } + + return rotation; +} + + +/** +*************************************************************************************************** +* EgBasedAddrLib::ComputeHtileBytes +* +* @brief +* Compute htile size in bytes +* +* @return +* Htile size in bytes +*************************************************************************************************** +*/ +UINT_64 EgBasedAddrLib::ComputeHtileBytes( + UINT_32 pitch, ///< [in] pitch + UINT_32 height, ///< [in] height + UINT_32 bpp, ///< [in] bits per pixel + BOOL_32 isLinear, ///< [in] if it is linear mode + UINT_32 numSlices, ///< [in] number of slices + UINT_64* sliceBytes, ///< [out] bytes per slice + UINT_32 baseAlign ///< [in] base alignments + ) const +{ + UINT_64 surfBytes; + + const UINT_64 HtileCacheLineSize = BITS_TO_BYTES(HtileCacheBits); + + *sliceBytes = BITS_TO_BYTES(static_cast(pitch) * height * bpp / 64); + + if (m_configFlags.useHtileSliceAlign) + { + // Align the sliceSize to htilecachelinesize * pipes at first + *sliceBytes = PowTwoAlign(*sliceBytes, HtileCacheLineSize * m_pipes); + surfBytes = *sliceBytes * numSlices; + } + else + { + // Align the surfSize to htilecachelinesize * pipes at last + surfBytes = *sliceBytes * numSlices; + surfBytes = PowTwoAlign(surfBytes, HtileCacheLineSize * m_pipes); + } + + return surfBytes; +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::DispatchComputeFmaskInfo +* +* @brief +* Compute fmask sizes include padded pitch, height, slices, total size in bytes, +* meanwhile output suitable tile mode and alignments as well. Results are returned +* through output parameters. +* +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE EgBasedAddrLib::DispatchComputeFmaskInfo( + const ADDR_COMPUTE_FMASK_INFO_INPUT* pIn, ///< [in] input structure + ADDR_COMPUTE_FMASK_INFO_OUTPUT* pOut) ///< [out] output structure +{ + ADDR_E_RETURNCODE retCode = ADDR_OK; + + ADDR_COMPUTE_SURFACE_INFO_INPUT surfIn = {0}; + ADDR_COMPUTE_SURFACE_INFO_OUTPUT surfOut = {0}; + + // Setup input structure + surfIn.tileMode = pIn->tileMode; + surfIn.width = pIn->pitch; + surfIn.height = pIn->height; + surfIn.numSlices = pIn->numSlices; + surfIn.pTileInfo = pIn->pTileInfo; + surfIn.tileType = ADDR_NON_DISPLAYABLE; + surfIn.flags.fmask = 1; + + // Setup output structure + surfOut.pTileInfo = pOut->pTileInfo; + + // Setup hwl specific fields + HwlFmaskPreThunkSurfInfo(pIn, pOut, &surfIn, &surfOut); + + surfIn.bpp = HwlComputeFmaskBits(pIn, &surfIn.numSamples); + + // ComputeSurfaceInfo needs numSamples in surfOut as surface routines need adjusted numSamples + surfOut.numSamples = surfIn.numSamples; + + retCode = HwlComputeSurfaceInfo(&surfIn, &surfOut); + + // Save bpp field for surface dump support + surfOut.bpp = surfIn.bpp; + + if (retCode == ADDR_OK) + { + pOut->bpp = surfOut.bpp; + pOut->pitch = surfOut.pitch; + pOut->height = surfOut.height; + pOut->numSlices = surfOut.depth; + pOut->fmaskBytes = surfOut.surfSize; + pOut->baseAlign = surfOut.baseAlign; + pOut->pitchAlign = surfOut.pitchAlign; + pOut->heightAlign = surfOut.heightAlign; + + if (surfOut.depth > 1) + { + // For fmask, expNumSlices is stored in depth. + pOut->sliceSize = surfOut.surfSize / surfOut.depth; + } + else + { + pOut->sliceSize = surfOut.surfSize; + } + + // Save numSamples field for surface dump support + pOut->numSamples = surfOut.numSamples; + + HwlFmaskPostThunkSurfInfo(&surfOut, pOut); + } + + return retCode; +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::HwlFmaskSurfaceInfo +* @brief +* Entry of EgBasedAddrLib ComputeFmaskInfo +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE EgBasedAddrLib::HwlComputeFmaskInfo( + const ADDR_COMPUTE_FMASK_INFO_INPUT* pIn, ///< [in] input structure + ADDR_COMPUTE_FMASK_INFO_OUTPUT* pOut ///< [out] output structure + ) +{ + ADDR_E_RETURNCODE retCode = ADDR_OK; + + ADDR_TILEINFO tileInfo = {0}; + + // Use internal tile info if pOut does not have a valid pTileInfo + if (pOut->pTileInfo == NULL) + { + pOut->pTileInfo = &tileInfo; + } + + retCode = DispatchComputeFmaskInfo(pIn, pOut); + + if (retCode == ADDR_OK) + { + pOut->tileIndex = + HwlPostCheckTileIndex(pOut->pTileInfo, pIn->tileMode, ADDR_NON_DISPLAYABLE, + pOut->tileIndex); + } + + // Resets pTileInfo to NULL if the internal tile info is used + if (pOut->pTileInfo == &tileInfo) + { + pOut->pTileInfo = NULL; + } + + return retCode; +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::HwlComputeFmaskAddrFromCoord +* @brief +* Entry of EgBasedAddrLib ComputeFmaskAddrFromCoord +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE EgBasedAddrLib::HwlComputeFmaskAddrFromCoord( + const ADDR_COMPUTE_FMASK_ADDRFROMCOORD_INPUT* pIn, ///< [in] input structure + ADDR_COMPUTE_FMASK_ADDRFROMCOORD_OUTPUT* pOut ///< [out] output structure + ) const +{ + ADDR_E_RETURNCODE retCode = ADDR_OK; + +#if ADDR_AM_BUILD + if ((pIn->x > pIn->pitch) || + (pIn->y > pIn->height) || + (pIn->numSamples > m_maxSamples) || + (pIn->sample >= m_maxSamples)) + { + retCode = ADDR_INVALIDPARAMS; + } + else + { + pOut->addr = DispatchComputeFmaskAddrFromCoord(pIn, pOut); + } +#endif + + return retCode; +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::HwlComputeFmaskCoordFromAddr +* @brief +* Entry of EgBasedAddrLib ComputeFmaskCoordFromAddr +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE EgBasedAddrLib::HwlComputeFmaskCoordFromAddr( + const ADDR_COMPUTE_FMASK_COORDFROMADDR_INPUT* pIn, ///< [in] input structure + ADDR_COMPUTE_FMASK_COORDFROMADDR_OUTPUT* pOut ///< [out] output structure + ) const +{ + ADDR_E_RETURNCODE retCode = ADDR_OK; + +#if ADDR_AM_BUILD + if ((pIn->bitPosition >= 8) || + (pIn->numSamples > m_maxSamples)) + { + retCode = ADDR_INVALIDPARAMS; + } + else + { + DispatchComputeFmaskCoordFromAddr(pIn, pOut); + } +#endif + + return retCode; +} + +#if ADDR_AM_BUILD +/** +*************************************************************************************************** +* EgBasedAddrLib::DispatchComputeFmaskAddrFromCoord +* +* @brief +* Computes the FMASK address and bit position from a coordinate. +* @return +* The byte address +*************************************************************************************************** +*/ +UINT_64 EgBasedAddrLib::DispatchComputeFmaskAddrFromCoord( + const ADDR_COMPUTE_FMASK_ADDRFROMCOORD_INPUT* pIn, ///< [in] input structure + ADDR_COMPUTE_FMASK_ADDRFROMCOORD_OUTPUT* pOut ///< [out] output structure + ) const +{ + UINT_32 x = pIn->x; + UINT_32 y = pIn->y; + UINT_32 slice = pIn->slice; + UINT_32 sample = pIn->sample; + UINT_32 plane = pIn->plane; + UINT_32 pitch = pIn->pitch; + UINT_32 height = pIn->height; + UINT_32 numSamples = pIn->numSamples; + AddrTileMode tileMode = pIn->tileMode; + BOOL_32 ignoreSE = pIn->ignoreSE; + ADDR_TILEINFO* pTileInfo = pIn->pTileInfo; + BOOL_32 resolved = pIn->resolved; + + UINT_32* pBitPosition = &pOut->bitPosition; + UINT_64 addr = 0; + + ADDR_ASSERT(numSamples > 1); + ADDR_ASSERT(ComputeSurfaceThickness(tileMode) == 1); + + switch (tileMode) + { + case ADDR_TM_1D_TILED_THIN1: + addr = ComputeFmaskAddrFromCoordMicroTiled(x, + y, + slice, + sample, + plane, + pitch, + height, + numSamples, + tileMode, + resolved, + pBitPosition); + break; + case ADDR_TM_2D_TILED_THIN1: //fall through + case ADDR_TM_3D_TILED_THIN1: + UINT_32 pipeSwizzle; + UINT_32 bankSwizzle; + + if (m_configFlags.useCombinedSwizzle) + { + ExtractBankPipeSwizzle(pIn->tileSwizzle, pIn->pTileInfo, + &bankSwizzle, &pipeSwizzle); + } + else + { + pipeSwizzle = pIn->pipeSwizzle; + bankSwizzle = pIn->bankSwizzle; + } + + addr = ComputeFmaskAddrFromCoordMacroTiled(x, + y, + slice, + sample, + plane, + pitch, + height, + numSamples, + tileMode, + pipeSwizzle, + bankSwizzle, + ignoreSE, + pTileInfo, + resolved, + pBitPosition); + break; + default: + *pBitPosition = 0; + break; + } + + return addr; +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::ComputeFmaskAddrFromCoordMicroTiled +* +* @brief +* Computes the FMASK address and bit position from a coordinate for 1D tilied (micro +* tiled) +* @return +* The byte address +*************************************************************************************************** +*/ +UINT_64 EgBasedAddrLib::ComputeFmaskAddrFromCoordMicroTiled( + UINT_32 x, ///< [in] x coordinate + UINT_32 y, ///< [in] y coordinate + UINT_32 slice, ///< [in] slice index + UINT_32 sample, ///< [in] sample number + UINT_32 plane, ///< [in] plane number + UINT_32 pitch, ///< [in] surface pitch in pixels + UINT_32 height, ///< [in] surface height in pixels + UINT_32 numSamples, ///< [in] number of samples + AddrTileMode tileMode, ///< [in] tile mode + BOOL_32 resolved, ///< [in] TRUE if this is for resolved fmask + UINT_32* pBitPosition ///< [out] pointer to returned bit position + ) const +{ + UINT_64 addr = 0; + UINT_32 effectiveBpp; + UINT_32 effectiveSamples; + + // + // 2xAA use the same layout as 4xAA + // + if (numSamples == 2) + { + numSamples = 4; + } + + // + // Compute the number of planes. + // + if (!resolved) + { + effectiveSamples = ComputeFmaskNumPlanesFromNumSamples(numSamples);; + effectiveBpp = numSamples; + + // + // Compute the address just like a color surface with numSamples bits per element and + // numPlanes samples. + // + addr = ComputeSurfaceAddrFromCoordMicroTiled(x, + y, + slice, + plane, // sample + effectiveBpp, + pitch, + height, + effectiveSamples, + tileMode, + ADDR_NON_DISPLAYABLE, + FALSE, + pBitPosition); + + // + // Compute the real bit position. Each (sample, plane) is stored with one bit per sample. + // + + // + // Compute the pixel index with in the micro tile + // + UINT_32 pixelIndex = ComputePixelIndexWithinMicroTile(x % 8, + y % 8, + slice, + 1, + tileMode, + ADDR_NON_DISPLAYABLE); + + *pBitPosition = ((pixelIndex * numSamples) + sample) & (BITS_PER_BYTE-1); + + UINT_64 bitAddr = BYTES_TO_BITS(addr) + *pBitPosition; + + addr = bitAddr / 8; + } + else + { + effectiveBpp = ComputeFmaskResolvedBppFromNumSamples(numSamples); + effectiveSamples = 1; + + // + // Compute the address just like a color surface with numSamples bits per element and + // numPlanes samples. + // + addr = ComputeSurfaceAddrFromCoordMicroTiled(x, + y, + slice, + sample, + effectiveBpp, + pitch, + height, + effectiveSamples, + tileMode, + ADDR_NON_DISPLAYABLE, + TRUE, + pBitPosition); + } + + return addr; +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::ComputeFmaskAddrFromCoordMacroTiled +* +* @brief +* Computes the FMASK address and bit position from a coordinate for 2D tilied (macro +* tiled) +* @return +* The byte address +*************************************************************************************************** +*/ +UINT_64 EgBasedAddrLib::ComputeFmaskAddrFromCoordMacroTiled( + UINT_32 x, ///< [in] x coordinate + UINT_32 y, ///< [in] y coordinate + UINT_32 slice, ///< [in] slice index + UINT_32 sample, ///< [in] sample number + UINT_32 plane, ///< [in] plane number + UINT_32 pitch, ///< [in] surface pitch in pixels + UINT_32 height, ///< [in] surface height in pixels + UINT_32 numSamples, ///< [in] number of samples + AddrTileMode tileMode, ///< [in] tile mode + UINT_32 pipeSwizzle, ///< [in] pipe swizzle + UINT_32 bankSwizzle, ///< [in] bank swizzle + BOOL_32 ignoreSE, ///< [in] TRUE if ignore shader engine + ADDR_TILEINFO* pTileInfo, ///< [in] bank structure.**All fields to be valid on entry** + BOOL_32 resolved, ///< [in] TRUE if this is for resolved fmask + UINT_32* pBitPosition ///< [out] pointer to returned bit position + ) const +{ + UINT_64 addr = 0; + UINT_32 effectiveBpp; + UINT_32 effectiveSamples; + + // + // 2xAA use the same layout as 4xAA + // + if (numSamples == 2) + { + numSamples = 4; + } + + // + // Compute the number of planes. + // + if (!resolved) + { + effectiveSamples = ComputeFmaskNumPlanesFromNumSamples(numSamples); + effectiveBpp = numSamples; + + // + // Compute the address just like a color surface with numSamples bits per element and + // numPlanes samples. + // + addr = ComputeSurfaceAddrFromCoordMacroTiled(x, + y, + slice, + plane, // sample + effectiveBpp, + pitch, + height, + effectiveSamples, + tileMode, + ADDR_NON_DISPLAYABLE,// isdisp + ignoreSE,// ignore_shader + FALSE,// depth_sample_order + pipeSwizzle, + bankSwizzle, + pTileInfo, + pBitPosition); + + // + // Compute the real bit position. Each (sample, plane) is stored with one bit per sample. + // + + + // + // Compute the pixel index with in the micro tile + // + UINT_32 pixelIndex = ComputePixelIndexWithinMicroTile(x , + y , + slice, + effectiveBpp, + tileMode, + ADDR_NON_DISPLAYABLE); + + *pBitPosition = ((pixelIndex * numSamples) + sample) & (BITS_PER_BYTE-1); + + UINT_64 bitAddr = BYTES_TO_BITS(addr) + *pBitPosition; + + addr = bitAddr / 8; + + } + else + { + effectiveBpp = ComputeFmaskResolvedBppFromNumSamples(numSamples); + effectiveSamples = 1; + + // + // Compute the address just like a color surface with numSamples bits per element and + // numPlanes samples. + // + addr = ComputeSurfaceAddrFromCoordMacroTiled(x, + y, + slice, + sample, + effectiveBpp, + pitch, + height, + effectiveSamples, + tileMode, + ADDR_NON_DISPLAYABLE, + ignoreSE, + TRUE, + pipeSwizzle, + bankSwizzle, + pTileInfo, + pBitPosition); + } + + return addr; +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::ComputeFmaskCoordFromAddrMicroTiled +* +* @brief +* Compute (x,y,slice,sample,plane) coordinates from fmask address +* @return +* N/A +* +*************************************************************************************************** +*/ +VOID EgBasedAddrLib::ComputeFmaskCoordFromAddrMicroTiled( + UINT_64 addr, ///< [in] byte address + UINT_32 bitPosition,///< [in] bit position + UINT_32 pitch, ///< [in] pitch in pixels + UINT_32 height, ///< [in] height in pixels + UINT_32 numSamples, ///< [in] number of samples (of color buffer) + AddrTileMode tileMode, ///< [in] tile mode + BOOL_32 resolved, ///< [in] TRUE if it is resolved fmask + UINT_32* pX, ///< [out] X coord + UINT_32* pY, ///< [out] Y coord + UINT_32* pSlice, ///< [out] slice index + UINT_32* pSample, ///< [out] sample index + UINT_32* pPlane ///< [out] plane index + ) const +{ + UINT_32 effectiveBpp; + UINT_32 effectiveSamples; + + // 2xAA use the same layout as 4xAA + if (numSamples == 2) + { + numSamples = 4; + } + + if (!resolved) + { + effectiveSamples = ComputeFmaskNumPlanesFromNumSamples(numSamples); + effectiveBpp = numSamples; + + ComputeSurfaceCoordFromAddrMicroTiled(addr, + bitPosition, + effectiveBpp, + pitch, + height, + effectiveSamples, + tileMode, + 0, // tileBase + 0, // compBits + pX, + pY, + pSlice, + pPlane, + ADDR_NON_DISPLAYABLE, // microTileType + FALSE // isDepthSampleOrder + ); + + + if ( pSample ) + { + *pSample = bitPosition % numSamples; + } + } + else + { + effectiveBpp = ComputeFmaskResolvedBppFromNumSamples(numSamples); + effectiveSamples = 1; + + ComputeSurfaceCoordFromAddrMicroTiled(addr, + bitPosition, + effectiveBpp, + pitch, + height, + effectiveSamples, + tileMode, + 0, // tileBase + 0, // compBits + pX, + pY, + pSlice, + pSample, + ADDR_NON_DISPLAYABLE, // microTileType + TRUE // isDepthSampleOrder + ); + } +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::ComputeFmaskCoordFromAddrMacroTiled +* +* @brief +* Compute (x,y,slice,sample,plane) coordinates from +* fmask address +* @return +* N/A +* +*************************************************************************************************** +*/ +VOID EgBasedAddrLib::ComputeFmaskCoordFromAddrMacroTiled( + UINT_64 addr, ///< [in] byte address + UINT_32 bitPosition,///< [in] bit position + UINT_32 pitch, ///< [in] pitch in pixels + UINT_32 height, ///< [in] height in pixels + UINT_32 numSamples, ///< [in] number of samples (of color buffer) + AddrTileMode tileMode, ///< [in] tile mode + UINT_32 pipeSwizzle,///< [in] pipe swizzle + UINT_32 bankSwizzle,///< [in] bank swizzle + BOOL_32 ignoreSE, ///< [in] TRUE if ignore shader engine + ADDR_TILEINFO* pTileInfo, ///< [in] bank structure. **All fields to be valid on entry** + BOOL_32 resolved, ///< [in] TRUE if it is resolved fmask + UINT_32* pX, ///< [out] X coord + UINT_32* pY, ///< [out] Y coord + UINT_32* pSlice, ///< [out] slice index + UINT_32* pSample, ///< [out] sample index + UINT_32* pPlane ///< [out] plane index + ) const +{ + UINT_32 effectiveBpp; + UINT_32 effectiveSamples; + + // 2xAA use the same layout as 4xAA + if (numSamples == 2) + { + numSamples = 4; + } + + // + // Compute the number of planes. + // + if (!resolved) + { + effectiveSamples = ComputeFmaskNumPlanesFromNumSamples(numSamples); + effectiveBpp = numSamples; + + ComputeSurfaceCoordFromAddrMacroTiled(addr, + bitPosition, + effectiveBpp, + pitch, + height, + effectiveSamples, + tileMode, + 0, // No tileBase + 0, // No compBits + ADDR_NON_DISPLAYABLE, + ignoreSE, + FALSE, + pipeSwizzle, + bankSwizzle, + pTileInfo, + pX, + pY, + pSlice, + pPlane); + + if (pSample) + { + *pSample = bitPosition % numSamples; + } + } + else + { + effectiveBpp = ComputeFmaskResolvedBppFromNumSamples(numSamples); + effectiveSamples = 1; + + ComputeSurfaceCoordFromAddrMacroTiled(addr, + bitPosition, + effectiveBpp, + pitch, + height, + effectiveSamples, + tileMode, + 0, // No tileBase + 0, // No compBits + ADDR_NON_DISPLAYABLE, + ignoreSE, + TRUE, + pipeSwizzle, + bankSwizzle, + pTileInfo, + pX, + pY, + pSlice, + pSample); + } +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::DispatchComputeFmaskCoordFromAddr +* +* @brief +* Compute (x,y,slice,sample,plane) coordinates from +* fmask address +* @return +* N/A +* +*************************************************************************************************** +*/ +VOID EgBasedAddrLib::DispatchComputeFmaskCoordFromAddr( + const ADDR_COMPUTE_FMASK_COORDFROMADDR_INPUT* pIn, ///< [in] input structure + ADDR_COMPUTE_FMASK_COORDFROMADDR_OUTPUT* pOut ///< [out] output structure + ) const +{ + UINT_64 addr = pIn->addr; + UINT_32 bitPosition = pIn->bitPosition; + UINT_32 pitch = pIn->pitch; + UINT_32 height = pIn->height; + UINT_32 numSamples = pIn->numSamples; + AddrTileMode tileMode = pIn->tileMode; + BOOL_32 ignoreSE = pIn->ignoreSE; + ADDR_TILEINFO* pTileInfo = pIn->pTileInfo; + BOOL_32 resolved = pIn->resolved; + + UINT_32* pX = &pOut->x; + UINT_32* pY = &pOut->y; + UINT_32* pSlice = &pOut->slice; + UINT_32* pSample = &pOut->sample; + UINT_32* pPlane = &pOut->plane; + + switch (tileMode) + { + case ADDR_TM_1D_TILED_THIN1: + ComputeFmaskCoordFromAddrMicroTiled(addr, + bitPosition, + pitch, + height, + numSamples, + tileMode, + resolved, + pX, + pY, + pSlice, + pSample, + pPlane); + break; + case ADDR_TM_2D_TILED_THIN1://fall through + case ADDR_TM_3D_TILED_THIN1: + UINT_32 pipeSwizzle; + UINT_32 bankSwizzle; + + if (m_configFlags.useCombinedSwizzle) + { + ExtractBankPipeSwizzle(pIn->tileSwizzle, pIn->pTileInfo, + &bankSwizzle, &pipeSwizzle); + } + else + { + pipeSwizzle = pIn->pipeSwizzle; + bankSwizzle = pIn->bankSwizzle; + } + + ComputeFmaskCoordFromAddrMacroTiled(addr, + bitPosition, + pitch, + height, + numSamples, + tileMode, + pipeSwizzle, + bankSwizzle, + ignoreSE, + pTileInfo, + resolved, + pX, + pY, + pSlice, + pSample, + pPlane); + break; + default: + ADDR_ASSERT_ALWAYS(); + break; + + } +} +#endif + +/** +*************************************************************************************************** +* EgBasedAddrLib::ComputeFmaskNumPlanesFromNumSamples +* +* @brief +* Compute fmask number of planes from number of samples +* +* @return +* Number of planes +*************************************************************************************************** +*/ +UINT_32 EgBasedAddrLib::ComputeFmaskNumPlanesFromNumSamples( + UINT_32 numSamples) ///< [in] number of samples +{ + UINT_32 numPlanes; + + // + // FMASK is stored such that each micro tile is composed of elements containing N bits, where + // N is the number of samples. There is a micro tile for each bit in the FMASK address, and + // micro tiles for each address bit, sometimes referred to as a plane, are stored sequentially. + // The FMASK for a 2-sample surface looks like a general surface with 2 bits per element. + // The FMASK for a 4-sample surface looks like a general surface with 4 bits per element and + // 2 samples. The FMASK for an 8-sample surface looks like a general surface with 8 bits per + // element and 4 samples. R6xx and R7xx only stored 3 planes for 8-sample FMASK surfaces. + // This was changed for R8xx to simplify the logic in the CB. + // + switch (numSamples) + { + case 2: + numPlanes = 1; + break; + case 4: + numPlanes = 2; + break; + case 8: + numPlanes = 4; + break; + default: + ADDR_UNHANDLED_CASE(); + numPlanes = 0; + break; + } + return numPlanes; +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::ComputeFmaskResolvedBppFromNumSamples +* +* @brief +* Compute resolved fmask effective bpp based on number of samples +* +* @return +* bpp +*************************************************************************************************** +*/ +UINT_32 EgBasedAddrLib::ComputeFmaskResolvedBppFromNumSamples( + UINT_32 numSamples) ///< number of samples +{ + UINT_32 bpp; + + // + // Resolved FMASK surfaces are generated yBit the CB and read yBit the texture unit + // so that the texture unit can read compressed multi-sample color data. + // These surfaces store each index value packed per element. + // Each element contains at least num_samples * log2(num_samples) bits. + // Resolved FMASK surfaces are addressed as follows: + // 2-sample Addressed similarly to a color surface with 8 bits per element and 1 sample. + // 4-sample Addressed similarly to a color surface with 8 bits per element and 1 sample. + // 8-sample Addressed similarly to a color surface with 32 bits per element and 1 sample. + + switch (numSamples) + { + case 2: + bpp = 8; + break; + case 4: + bpp = 8; + break; + case 8: + bpp = 32; + break; + default: + ADDR_UNHANDLED_CASE(); + bpp = 0; + break; + } + return bpp; +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::IsTileInfoAllZero +* +* @brief +* Return TRUE if all field are zero +* @note +* Since NULL input is consider to be all zero +*************************************************************************************************** +*/ +BOOL_32 EgBasedAddrLib::IsTileInfoAllZero( + ADDR_TILEINFO* pTileInfo) +{ + BOOL_32 allZero = TRUE; + + if (pTileInfo) + { + if ((pTileInfo->banks != 0) || + (pTileInfo->bankWidth != 0) || + (pTileInfo->bankHeight != 0) || + (pTileInfo->macroAspectRatio != 0) || + (pTileInfo->tileSplitBytes != 0) || + (pTileInfo->pipeConfig != 0) + ) + { + allZero = FALSE; + } + } + + return allZero; +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::HwlTileInfoEqual +* +* @brief +* Return TRUE if all field are equal +* @note +* Only takes care of current HWL's data +*************************************************************************************************** +*/ +BOOL_32 EgBasedAddrLib::HwlTileInfoEqual( + const ADDR_TILEINFO* pLeft, ///<[in] Left compare operand + const ADDR_TILEINFO* pRight ///<[in] Right compare operand + ) const +{ + BOOL_32 equal = FALSE; + + if (pLeft->banks == pRight->banks && + pLeft->bankWidth == pRight->bankWidth && + pLeft->bankHeight == pRight->bankHeight && + pLeft->macroAspectRatio == pRight->macroAspectRatio && + pLeft->tileSplitBytes == pRight->tileSplitBytes) + { + equal = TRUE; + } + + return equal; +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::HwlConvertTileInfoToHW +* @brief +* Entry of EgBasedAddrLib ConvertTileInfoToHW +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE EgBasedAddrLib::HwlConvertTileInfoToHW( + const ADDR_CONVERT_TILEINFOTOHW_INPUT* pIn, ///< [in] input structure + ADDR_CONVERT_TILEINFOTOHW_OUTPUT* pOut ///< [out] output structure + ) const +{ + ADDR_E_RETURNCODE retCode = ADDR_OK; + + ADDR_TILEINFO *pTileInfoIn = pIn->pTileInfo; + ADDR_TILEINFO *pTileInfoOut = pOut->pTileInfo; + + if ((pTileInfoIn != NULL) && (pTileInfoOut != NULL)) + { + if (pIn->reverse == FALSE) + { + switch (pTileInfoIn->banks) + { + case 2: + pTileInfoOut->banks = 0; + break; + case 4: + pTileInfoOut->banks = 1; + break; + case 8: + pTileInfoOut->banks = 2; + break; + case 16: + pTileInfoOut->banks = 3; + break; + default: + ADDR_ASSERT_ALWAYS(); + retCode = ADDR_INVALIDPARAMS; + pTileInfoOut->banks = 0; + break; + } + + switch (pTileInfoIn->bankWidth) + { + case 1: + pTileInfoOut->bankWidth = 0; + break; + case 2: + pTileInfoOut->bankWidth = 1; + break; + case 4: + pTileInfoOut->bankWidth = 2; + break; + case 8: + pTileInfoOut->bankWidth = 3; + break; + default: + ADDR_ASSERT_ALWAYS(); + retCode = ADDR_INVALIDPARAMS; + pTileInfoOut->bankWidth = 0; + break; + } + + switch (pTileInfoIn->bankHeight) + { + case 1: + pTileInfoOut->bankHeight = 0; + break; + case 2: + pTileInfoOut->bankHeight = 1; + break; + case 4: + pTileInfoOut->bankHeight = 2; + break; + case 8: + pTileInfoOut->bankHeight = 3; + break; + default: + ADDR_ASSERT_ALWAYS(); + retCode = ADDR_INVALIDPARAMS; + pTileInfoOut->bankHeight = 0; + break; + } + + switch (pTileInfoIn->macroAspectRatio) + { + case 1: + pTileInfoOut->macroAspectRatio = 0; + break; + case 2: + pTileInfoOut->macroAspectRatio = 1; + break; + case 4: + pTileInfoOut->macroAspectRatio = 2; + break; + case 8: + pTileInfoOut->macroAspectRatio = 3; + break; + default: + ADDR_ASSERT_ALWAYS(); + retCode = ADDR_INVALIDPARAMS; + pTileInfoOut->macroAspectRatio = 0; + break; + } + + switch (pTileInfoIn->tileSplitBytes) + { + case 64: + pTileInfoOut->tileSplitBytes = 0; + break; + case 128: + pTileInfoOut->tileSplitBytes = 1; + break; + case 256: + pTileInfoOut->tileSplitBytes = 2; + break; + case 512: + pTileInfoOut->tileSplitBytes = 3; + break; + case 1024: + pTileInfoOut->tileSplitBytes = 4; + break; + case 2048: + pTileInfoOut->tileSplitBytes = 5; + break; + case 4096: + pTileInfoOut->tileSplitBytes = 6; + break; + default: + ADDR_ASSERT_ALWAYS(); + retCode = ADDR_INVALIDPARAMS; + pTileInfoOut->tileSplitBytes = 0; + break; + } + } + else + { + switch (pTileInfoIn->banks) + { + case 0: + pTileInfoOut->banks = 2; + break; + case 1: + pTileInfoOut->banks = 4; + break; + case 2: + pTileInfoOut->banks = 8; + break; + case 3: + pTileInfoOut->banks = 16; + break; + default: + ADDR_ASSERT_ALWAYS(); + retCode = ADDR_INVALIDPARAMS; + pTileInfoOut->banks = 2; + break; + } + + switch (pTileInfoIn->bankWidth) + { + case 0: + pTileInfoOut->bankWidth = 1; + break; + case 1: + pTileInfoOut->bankWidth = 2; + break; + case 2: + pTileInfoOut->bankWidth = 4; + break; + case 3: + pTileInfoOut->bankWidth = 8; + break; + default: + ADDR_ASSERT_ALWAYS(); + retCode = ADDR_INVALIDPARAMS; + pTileInfoOut->bankWidth = 1; + break; + } + + switch (pTileInfoIn->bankHeight) + { + case 0: + pTileInfoOut->bankHeight = 1; + break; + case 1: + pTileInfoOut->bankHeight = 2; + break; + case 2: + pTileInfoOut->bankHeight = 4; + break; + case 3: + pTileInfoOut->bankHeight = 8; + break; + default: + ADDR_ASSERT_ALWAYS(); + retCode = ADDR_INVALIDPARAMS; + pTileInfoOut->bankHeight = 1; + break; + } + + switch (pTileInfoIn->macroAspectRatio) + { + case 0: + pTileInfoOut->macroAspectRatio = 1; + break; + case 1: + pTileInfoOut->macroAspectRatio = 2; + break; + case 2: + pTileInfoOut->macroAspectRatio = 4; + break; + case 3: + pTileInfoOut->macroAspectRatio = 8; + break; + default: + ADDR_ASSERT_ALWAYS(); + retCode = ADDR_INVALIDPARAMS; + pTileInfoOut->macroAspectRatio = 1; + break; + } + + switch (pTileInfoIn->tileSplitBytes) + { + case 0: + pTileInfoOut->tileSplitBytes = 64; + break; + case 1: + pTileInfoOut->tileSplitBytes = 128; + break; + case 2: + pTileInfoOut->tileSplitBytes = 256; + break; + case 3: + pTileInfoOut->tileSplitBytes = 512; + break; + case 4: + pTileInfoOut->tileSplitBytes = 1024; + break; + case 5: + pTileInfoOut->tileSplitBytes = 2048; + break; + case 6: + pTileInfoOut->tileSplitBytes = 4096; + break; + default: + ADDR_ASSERT_ALWAYS(); + retCode = ADDR_INVALIDPARAMS; + pTileInfoOut->tileSplitBytes = 64; + break; + } + } + + if (pTileInfoIn != pTileInfoOut) + { + pTileInfoOut->pipeConfig = pTileInfoIn->pipeConfig; + } + } + else + { + ADDR_ASSERT_ALWAYS(); + retCode = ADDR_INVALIDPARAMS; + } + + return retCode; +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::HwlComputeSurfaceInfo +* @brief +* Entry of EgBasedAddrLib ComputeSurfaceInfo +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE EgBasedAddrLib::HwlComputeSurfaceInfo( + const ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn, ///< [in] input structure + ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pOut ///< [out] output structure + ) const +{ + ADDR_E_RETURNCODE retCode = ADDR_OK; + + if (pIn->numSamples < pIn->numFrags) + { + retCode = ADDR_INVALIDPARAMS; + } + + ADDR_TILEINFO tileInfo = {0}; + + if (retCode == ADDR_OK) + { + // Uses internal tile info if pOut does not have a valid pTileInfo + if (pOut->pTileInfo == NULL) + { + pOut->pTileInfo = &tileInfo; + } + + if (!DispatchComputeSurfaceInfo(pIn, pOut)) + { + retCode = ADDR_INVALIDPARAMS; + } + + // Returns an index + pOut->tileIndex = HwlPostCheckTileIndex(pOut->pTileInfo, + pOut->tileMode, + pOut->tileType, + pOut->tileIndex); + + if (IsMacroTiled(pOut->tileMode) && (pOut->macroModeIndex == TileIndexInvalid)) + { + pOut->macroModeIndex = HwlComputeMacroModeIndex(pOut->tileIndex, + pIn->flags, + pIn->bpp, + pIn->numSamples, + pOut->pTileInfo); + } + + // Resets pTileInfo to NULL if the internal tile info is used + if (pOut->pTileInfo == &tileInfo) + { +#if DEBUG + // Client does not pass in a valid pTileInfo + if (IsMacroTiled(pOut->tileMode)) + { + // If a valid index is returned, then no pTileInfo is okay + ADDR_ASSERT(!m_configFlags.useTileIndex || pOut->tileIndex != TileIndexInvalid); + + if (!IsTileInfoAllZero(pIn->pTileInfo)) + { + // The initial value of pIn->pTileInfo is copied to tileInfo + // We do not expect any of these value to be changed nor any 0 of inputs + ADDR_ASSERT(tileInfo.banks == pIn->pTileInfo->banks); + ADDR_ASSERT(tileInfo.bankWidth == pIn->pTileInfo->bankWidth); + ADDR_ASSERT(tileInfo.bankHeight == pIn->pTileInfo->bankHeight); + ADDR_ASSERT(tileInfo.macroAspectRatio == pIn->pTileInfo->macroAspectRatio); + ADDR_ASSERT(tileInfo.tileSplitBytes == pIn->pTileInfo->tileSplitBytes); + } + } +#endif + pOut->pTileInfo = NULL; + } + } + + return retCode; +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::HwlComputeSurfaceAddrFromCoord +* @brief +* Entry of EgBasedAddrLib ComputeSurfaceAddrFromCoord +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE EgBasedAddrLib::HwlComputeSurfaceAddrFromCoord( + const ADDR_COMPUTE_SURFACE_ADDRFROMCOORD_INPUT* pIn, ///< [in] input structure + ADDR_COMPUTE_SURFACE_ADDRFROMCOORD_OUTPUT* pOut ///< [out] output structure + ) const +{ + ADDR_E_RETURNCODE retCode = ADDR_OK; + + if ( +#if !ALT_TEST // Overflow test needs this out-of-boundary coord + (pIn->x > pIn->pitch) || + (pIn->y > pIn->height) || +#endif + (pIn->numSamples > m_maxSamples)) + { + retCode = ADDR_INVALIDPARAMS; + } + else + { + pOut->addr = DispatchComputeSurfaceAddrFromCoord(pIn, pOut); + } + + return retCode; +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::HwlComputeSurfaceCoordFromAddr +* @brief +* Entry of EgBasedAddrLib ComputeSurfaceCoordFromAddr +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE EgBasedAddrLib::HwlComputeSurfaceCoordFromAddr( + const ADDR_COMPUTE_SURFACE_COORDFROMADDR_INPUT* pIn, ///< [in] input structure + ADDR_COMPUTE_SURFACE_COORDFROMADDR_OUTPUT* pOut ///< [out] output structure + ) const +{ + ADDR_E_RETURNCODE retCode = ADDR_OK; + + if ((pIn->bitPosition >= 8) || + (pIn->numSamples > m_maxSamples)) + { + retCode = ADDR_INVALIDPARAMS; + } + else + { + DispatchComputeSurfaceCoordFromAddr(pIn, pOut); + } + return retCode; +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::HwlComputeSliceTileSwizzle +* @brief +* Entry of EgBasedAddrLib ComputeSurfaceCoordFromAddr +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE EgBasedAddrLib::HwlComputeSliceTileSwizzle( + const ADDR_COMPUTE_SLICESWIZZLE_INPUT* pIn, ///< [in] input structure + ADDR_COMPUTE_SLICESWIZZLE_OUTPUT* pOut ///< [out] output structure + ) const +{ + ADDR_E_RETURNCODE retCode = ADDR_OK; + + if (pIn->pTileInfo && (pIn->pTileInfo->banks > 0)) + { + + pOut->tileSwizzle = ComputeSliceTileSwizzle(pIn->tileMode, + pIn->baseSwizzle, + pIn->slice, + pIn->baseAddr, + pIn->pTileInfo); + } + else + { + retCode = ADDR_INVALIDPARAMS; + } + + return retCode; +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::HwlComputeHtileBpp +* +* @brief +* Compute htile bpp +* +* @return +* Htile bpp +*************************************************************************************************** +*/ +UINT_32 EgBasedAddrLib::HwlComputeHtileBpp( + BOOL_32 isWidth8, ///< [in] TRUE if block width is 8 + BOOL_32 isHeight8 ///< [in] TRUE if block height is 8 + ) const +{ + // only support 8x8 mode + ADDR_ASSERT(isWidth8 && isHeight8); + return 32; +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::HwlComputeHtileBaseAlign +* +* @brief +* Compute htile base alignment +* +* @return +* Htile base alignment +*************************************************************************************************** +*/ +UINT_32 EgBasedAddrLib::HwlComputeHtileBaseAlign( + BOOL_32 isTcCompatible, ///< [in] if TC compatible + BOOL_32 isLinear, ///< [in] if it is linear mode + ADDR_TILEINFO* pTileInfo ///< [in] Tile info + ) const +{ + UINT_32 baseAlign = m_pipeInterleaveBytes * HwlGetPipes(pTileInfo); + + if (isTcCompatible) + { + ADDR_ASSERT(pTileInfo != NULL); + if (pTileInfo) + { + baseAlign *= pTileInfo->banks; + } + } + + return baseAlign; +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::HwlGetPitchAlignmentMicroTiled +* +* @brief +* Compute 1D tiled surface pitch alignment, calculation results are returned through +* output parameters. +* +* @return +* pitch alignment +*************************************************************************************************** +*/ +UINT_32 EgBasedAddrLib::HwlGetPitchAlignmentMicroTiled( + AddrTileMode tileMode, ///< [in] tile mode + UINT_32 bpp, ///< [in] bits per pixel + ADDR_SURFACE_FLAGS flags, ///< [in] surface flags + UINT_32 numSamples ///< [in] number of samples + ) const +{ + UINT_32 pitchAlign; + + UINT_32 microTileThickness = ComputeSurfaceThickness(tileMode); + + UINT_32 pixelsPerMicroTile; + UINT_32 pixelsPerPipeInterleave; + UINT_32 microTilesPerPipeInterleave; + + // + // Special workaround for depth/stencil buffer, use 8 bpp to meet larger requirement for + // stencil buffer since pitch alignment is related to bpp. + // For a depth only buffer do not set this. + // + // Note: this actually does not work for mipmap but mipmap depth texture is not really + // sampled with mipmap. + // + if (flags.depth && !flags.noStencil) + { + bpp = 8; + } + + pixelsPerMicroTile = MicroTilePixels * microTileThickness; + pixelsPerPipeInterleave = BYTES_TO_BITS(m_pipeInterleaveBytes) / (bpp * numSamples); + microTilesPerPipeInterleave = pixelsPerPipeInterleave / pixelsPerMicroTile; + + pitchAlign = Max(MicroTileWidth, microTilesPerPipeInterleave * MicroTileWidth); + + return pitchAlign; +} + +/** +*************************************************************************************************** +* EgBasedAddrLib::HwlGetSizeAdjustmentMicroTiled +* +* @brief +* Adjust 1D tiled surface pitch and slice size +* +* @return +* Logical slice size in bytes +*************************************************************************************************** +*/ +UINT_64 EgBasedAddrLib::HwlGetSizeAdjustmentMicroTiled( + UINT_32 thickness, ///< [in] thickness + UINT_32 bpp, ///< [in] bits per pixel + ADDR_SURFACE_FLAGS flags, ///< [in] surface flags + UINT_32 numSamples, ///< [in] number of samples + UINT_32 baseAlign, ///< [in] base alignment + UINT_32 pitchAlign, ///< [in] pitch alignment + UINT_32* pPitch, ///< [in/out] pointer to pitch + UINT_32* pHeight ///< [in/out] pointer to height + ) const +{ + UINT_64 logicalSliceSize; + UINT_64 physicalSliceSize; + + UINT_32 pitch = *pPitch; + UINT_32 height = *pHeight; + + // Logical slice: pitch * height * bpp * numSamples (no 1D MSAA so actually numSamples == 1) + logicalSliceSize = BITS_TO_BYTES(static_cast(pitch) * height * bpp * numSamples); + + // Physical slice: multiplied by thickness + physicalSliceSize = logicalSliceSize * thickness; + + // + // R800 will always pad physical slice size to baseAlign which is pipe_interleave_bytes + // + ADDR_ASSERT((physicalSliceSize % baseAlign) == 0) + + return logicalSliceSize; +} + diff --git a/src/gallium/winsys/radeon/amdgpu/addrlib/r800/egbaddrlib.h b/src/gallium/winsys/radeon/amdgpu/addrlib/r800/egbaddrlib.h new file mode 100644 index 0000000..84adb66 --- /dev/null +++ b/src/gallium/winsys/radeon/amdgpu/addrlib/r800/egbaddrlib.h @@ -0,0 +1,411 @@ +/* + * Copyright © 2014 Advanced Micro Devices, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES + * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NON-INFRINGEMENT. IN NO EVENT SHALL THE COPYRIGHT HOLDERS, AUTHORS + * AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + */ + +/** +*************************************************************************************************** +* @file egbaddrlib.h +* @brief Contains the EgBasedAddrLib class definition. +*************************************************************************************************** +*/ + +#ifndef __EG_BASED_ADDR_LIB_H__ +#define __EG_BASED_ADDR_LIB_H__ + +#include "addrlib.h" + + +/// Structures for functions +struct CoordFromBankPipe +{ + UINT_32 xBits : 3; + UINT_32 yBits : 4; + + UINT_32 xBit3 : 1; + UINT_32 xBit4 : 1; + UINT_32 xBit5 : 1; + UINT_32 yBit3 : 1; + UINT_32 yBit4 : 1; + UINT_32 yBit5 : 1; + UINT_32 yBit6 : 1; +}; + +/** +*************************************************************************************************** +* @brief This class is the Evergreen based address library +* @note Abstract class +*************************************************************************************************** +*/ +class EgBasedAddrLib : public AddrLib +{ +protected: + EgBasedAddrLib(const AddrClient* pClient); + virtual ~EgBasedAddrLib(); + +public: + + /// Surface info functions + + // NOTE: DispatchComputeSurfaceInfo using TileInfo takes both an input and an output. + // On input: + // One or more fields may be 0 to be calculated/defaulted - pre-SI h/w. + // H/W using tile mode index only accepts none or all 0's - SI and newer h/w. + // It then returns the actual tiling configuration used. + // Other methods' TileInfo must be valid on entry + BOOL_32 DispatchComputeSurfaceInfo( + const ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn, + ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pOut) const; + + ADDR_E_RETURNCODE DispatchComputeFmaskInfo( + const ADDR_COMPUTE_FMASK_INFO_INPUT* pIn, + ADDR_COMPUTE_FMASK_INFO_OUTPUT* pOut); + +protected: + // Hwl interface + virtual ADDR_E_RETURNCODE HwlComputeSurfaceInfo( + const ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn, + ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pOut) const; + + virtual ADDR_E_RETURNCODE HwlComputeSurfaceAddrFromCoord( + const ADDR_COMPUTE_SURFACE_ADDRFROMCOORD_INPUT* pIn, + ADDR_COMPUTE_SURFACE_ADDRFROMCOORD_OUTPUT* pOut) const; + + virtual ADDR_E_RETURNCODE HwlComputeSurfaceCoordFromAddr( + const ADDR_COMPUTE_SURFACE_COORDFROMADDR_INPUT* pIn, + ADDR_COMPUTE_SURFACE_COORDFROMADDR_OUTPUT* pOut) const; + + virtual ADDR_E_RETURNCODE HwlComputeSliceTileSwizzle( + const ADDR_COMPUTE_SLICESWIZZLE_INPUT* pIn, + ADDR_COMPUTE_SLICESWIZZLE_OUTPUT* pOut) const; + + virtual ADDR_E_RETURNCODE HwlExtractBankPipeSwizzle( + const ADDR_EXTRACT_BANKPIPE_SWIZZLE_INPUT* pIn, + ADDR_EXTRACT_BANKPIPE_SWIZZLE_OUTPUT* pOut) const; + + virtual ADDR_E_RETURNCODE HwlCombineBankPipeSwizzle( + UINT_32 bankSwizzle, UINT_32 pipeSwizzle, ADDR_TILEINFO* pTileInfo, + UINT_64 baseAddr, UINT_32* pTileSwizzle) const; + + virtual ADDR_E_RETURNCODE HwlComputeBaseSwizzle( + const ADDR_COMPUTE_BASE_SWIZZLE_INPUT* pIn, + ADDR_COMPUTE_BASE_SWIZZLE_OUTPUT* pOut) const; + + virtual ADDR_E_RETURNCODE HwlConvertTileInfoToHW( + const ADDR_CONVERT_TILEINFOTOHW_INPUT* pIn, + ADDR_CONVERT_TILEINFOTOHW_OUTPUT* pOut) const; + + virtual UINT_32 HwlComputeHtileBpp( + BOOL_32 isWidth8, BOOL_32 isHeight8) const; + + virtual UINT_32 HwlComputeHtileBaseAlign( + BOOL_32 isTcCompatible, BOOL_32 isLinear, ADDR_TILEINFO* pTileInfo) const; + + virtual ADDR_E_RETURNCODE HwlComputeFmaskInfo( + const ADDR_COMPUTE_FMASK_INFO_INPUT* pIn, + ADDR_COMPUTE_FMASK_INFO_OUTPUT* pOut); + + virtual ADDR_E_RETURNCODE HwlComputeFmaskAddrFromCoord( + const ADDR_COMPUTE_FMASK_ADDRFROMCOORD_INPUT* pIn, + ADDR_COMPUTE_FMASK_ADDRFROMCOORD_OUTPUT* pOut) const; + + virtual ADDR_E_RETURNCODE HwlComputeFmaskCoordFromAddr( + const ADDR_COMPUTE_FMASK_COORDFROMADDR_INPUT* pIn, + ADDR_COMPUTE_FMASK_COORDFROMADDR_OUTPUT* pOut) const; + + virtual BOOL_32 HwlDegradeBaseLevel( + const ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn) const; + + virtual UINT_32 HwlComputeQbStereoRightSwizzle( + ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pInfo) const; + + virtual VOID HwlComputePixelCoordFromOffset( + UINT_32 offset, UINT_32 bpp, UINT_32 numSamples, + AddrTileMode tileMode, UINT_32 tileBase, UINT_32 compBits, + UINT_32* pX, UINT_32* pY, UINT_32* pSlice, UINT_32* pSample, + AddrTileType microTileType, BOOL_32 isDepthSampleOrder) const; + + /// Return Cmask block max + virtual BOOL_32 HwlGetMaxCmaskBlockMax() const + { + return 16383; // 14 bits + } + + // Sub-hwl interface + /// Pure virtual function to setup tile info (indices) if client requests to do so + virtual VOID HwlSetupTileInfo( + AddrTileMode tileMode, ADDR_SURFACE_FLAGS flags, + UINT_32 bpp, UINT_32 pitch, UINT_32 height, UINT_32 numSamples, + ADDR_TILEINFO* inputTileInfo, ADDR_TILEINFO* outputTileInfo, + AddrTileType inTileType, ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pOut) const = 0; + + /// Pure virtual function to get pitch alignment for linear modes + virtual UINT_32 HwlGetPitchAlignmentLinear(UINT_32 bpp, ADDR_SURFACE_FLAGS flags) const = 0; + + /// Pure virtual function to get size adjustment for linear modes + virtual UINT_64 HwlGetSizeAdjustmentLinear( + AddrTileMode tileMode, + UINT_32 bpp, UINT_32 numSamples, UINT_32 baseAlign, UINT_32 pitchAlign, + UINT_32 *pPitch, UINT_32 *pHeight, UINT_32 *pHeightAlign) const = 0; + + virtual UINT_32 HwlGetPitchAlignmentMicroTiled( + AddrTileMode tileMode, UINT_32 bpp, ADDR_SURFACE_FLAGS flags, UINT_32 numSamples) const; + + virtual UINT_64 HwlGetSizeAdjustmentMicroTiled( + UINT_32 thickness, UINT_32 bpp, ADDR_SURFACE_FLAGS flags, UINT_32 numSamples, + UINT_32 baseAlign, UINT_32 pitchAlign, + UINT_32 *pPitch, UINT_32 *pHeight) const; + + /// Pure virtual function to do extra sanity check + virtual BOOL_32 HwlSanityCheckMacroTiled( + ADDR_TILEINFO* pTileInfo) const = 0; + + /// Pure virtual function to check current level to be the last macro tiled one + virtual VOID HwlCheckLastMacroTiledLvl( + const ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn, + ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pOut) const = 0; + + /// Adjusts bank before bank is modified by rotation + virtual UINT_32 HwlPreAdjustBank( + UINT_32 tileX, UINT_32 bank, ADDR_TILEINFO* pTileInfo) const = 0; + + virtual VOID HwlComputeSurfaceCoord2DFromBankPipe( + AddrTileMode tileMode, UINT_32* pX, UINT_32* pY, UINT_32 slice, + UINT_32 bank, UINT_32 pipe, + UINT_32 bankSwizzle, UINT_32 pipeSwizzle, UINT_32 tileSlices, + BOOL_32 ignoreSE, + ADDR_TILEINFO* pTileInfo) const = 0; + + virtual BOOL_32 HwlTileInfoEqual( + const ADDR_TILEINFO* pLeft, const ADDR_TILEINFO* pRight) const; + + virtual AddrTileMode HwlDegradeThickTileMode( + AddrTileMode baseTileMode, UINT_32 numSlices, UINT_32* pBytesPerTile) const; + + virtual INT_32 HwlPostCheckTileIndex( + const ADDR_TILEINFO* pInfo, AddrTileMode mode, AddrTileType type, + INT curIndex = TileIndexInvalid) const + { + return TileIndexInvalid; + } + + virtual VOID HwlFmaskPreThunkSurfInfo( + const ADDR_COMPUTE_FMASK_INFO_INPUT* pFmaskIn, + const ADDR_COMPUTE_FMASK_INFO_OUTPUT* pFmaskOut, + ADDR_COMPUTE_SURFACE_INFO_INPUT* pSurfIn, + ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pSurfOut) const + { + } + + virtual VOID HwlFmaskPostThunkSurfInfo( + const ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pSurfOut, + ADDR_COMPUTE_FMASK_INFO_OUTPUT* pFmaskOut) const + { + } + + /// Virtual function to check if the height needs extra padding + /// for stereo right eye offset, to avoid bank pipe swizzle + virtual BOOL_32 HwlStereoCheckRightOffsetPadding() const + { + return FALSE; + } + + virtual BOOL_32 HwlReduceBankWidthHeight( + UINT_32 tileSize, UINT_32 bpp, ADDR_SURFACE_FLAGS flags, UINT_32 numSamples, + UINT_32 bankHeightAlign, UINT_32 pipes, + ADDR_TILEINFO* pTileInfo) const; + + // Protected non-virtual functions + + /// Mip level functions + AddrTileMode ComputeSurfaceMipLevelTileMode( + AddrTileMode baseTileMode, UINT_32 bpp, + UINT_32 pitch, UINT_32 height, UINT_32 numSlices, UINT_32 numSamples, + UINT_32 pitchAlign, UINT_32 heightAlign, + ADDR_TILEINFO* pTileInfo) const; + + /// Swizzle functions + VOID ExtractBankPipeSwizzle( + UINT_32 base256b, ADDR_TILEINFO* pTileInfo, + UINT_32* pBankSwizzle, UINT_32* pPipeSwizzle) const; + + UINT_32 GetBankPipeSwizzle( + UINT_32 bankSwizzle, UINT_32 pipeSwizzle, + UINT_64 baseAddr, ADDR_TILEINFO* pTileInfo) const; + + UINT_32 ComputeSliceTileSwizzle( + AddrTileMode tileMode, UINT_32 baseSwizzle, UINT_32 slice, UINT_64 baseAddr, + ADDR_TILEINFO* pTileInfo) const; + + /// Addressing functions + UINT_32 ComputeBankFromCoord( + UINT_32 x, UINT_32 y, UINT_32 slice, + AddrTileMode tileMode, UINT_32 bankSwizzle, UINT_32 tileSpitSlice, + ADDR_TILEINFO* pTileInfo) const; + + UINT_32 ComputeBankFromAddr( + UINT_64 addr, UINT_32 numBanks, UINT_32 numPipes) const; + + UINT_32 ComputePipeRotation( + AddrTileMode tileMode, UINT_32 numPipes) const; + + UINT_32 ComputeBankRotation( + AddrTileMode tileMode, UINT_32 numBanks, + UINT_32 numPipes) const; + + VOID ComputeSurfaceCoord2DFromBankPipe( + AddrTileMode tileMode, UINT_32 x, UINT_32 y, UINT_32 slice, + UINT_32 bank, UINT_32 pipe, + UINT_32 bankSwizzle, UINT_32 pipeSwizzle, UINT_32 tileSlices, + ADDR_TILEINFO* pTileInfo, + CoordFromBankPipe *pOutput) const; + + /// Htile/Cmask functions + UINT_64 ComputeHtileBytes( + UINT_32 pitch, UINT_32 height, UINT_32 bpp, + BOOL_32 isLinear, UINT_32 numSlices, UINT_64* sliceBytes, UINT_32 baseAlign) const; + + // Static functions + static BOOL_32 IsTileInfoAllZero(ADDR_TILEINFO* pTileInfo); + static UINT_32 ComputeFmaskNumPlanesFromNumSamples(UINT_32 numSamples); + static UINT_32 ComputeFmaskResolvedBppFromNumSamples(UINT_32 numSamples); + +private: + + BOOL_32 ComputeSurfaceInfoLinear( + const ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn, + ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pOut, + UINT_32 padDims) const; + + BOOL_32 ComputeSurfaceInfoMicroTiled( + const ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn, + ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pOut, + UINT_32 padDims, + AddrTileMode expTileMode) const; + + BOOL_32 ComputeSurfaceInfoMacroTiled( + const ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn, + ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pOut, + UINT_32 padDims, + AddrTileMode expTileMode) const; + + BOOL_32 ComputeSurfaceAlignmentsLinear( + AddrTileMode tileMode, UINT_32 bpp, ADDR_SURFACE_FLAGS flags, + UINT_32* pBaseAlign, UINT_32* pPitchAlign, UINT_32* pHeightAlign) const; + + BOOL_32 ComputeSurfaceAlignmentsMicroTiled( + AddrTileMode tileMode, UINT_32 bpp, ADDR_SURFACE_FLAGS flags, UINT_32 numSamples, + UINT_32* pBaseAlign, UINT_32* pPitchAlign, UINT_32* pHeightAlign) const; + + BOOL_32 ComputeSurfaceAlignmentsMacroTiled( + AddrTileMode tileMode, UINT_32 bpp, ADDR_SURFACE_FLAGS flags, + UINT_32 mipLevel, UINT_32 numSamples, + ADDR_TILEINFO* pTileInfo, + UINT_32* pBaseAlign, UINT_32* pPitchAlign, UINT_32* pHeightAlign) const; + + /// Surface addressing functions + UINT_64 DispatchComputeSurfaceAddrFromCoord( + const ADDR_COMPUTE_SURFACE_ADDRFROMCOORD_INPUT* pIn, + ADDR_COMPUTE_SURFACE_ADDRFROMCOORD_OUTPUT* pOut) const; + + VOID DispatchComputeSurfaceCoordFromAddr( + const ADDR_COMPUTE_SURFACE_COORDFROMADDR_INPUT* pIn, + ADDR_COMPUTE_SURFACE_COORDFROMADDR_OUTPUT* pOut) const; + + UINT_64 ComputeSurfaceAddrFromCoordMicroTiled( + UINT_32 x, UINT_32 y, UINT_32 slice, UINT_32 sample, + UINT_32 bpp, UINT_32 pitch, UINT_32 height, UINT_32 numSamples, + AddrTileMode tileMode, + AddrTileType microTileType, BOOL_32 isDepthSampleOrder, + UINT_32* pBitPosition) const; + + UINT_64 ComputeSurfaceAddrFromCoordMacroTiled( + UINT_32 x, UINT_32 y, UINT_32 slice, UINT_32 sample, + UINT_32 bpp, UINT_32 pitch, UINT_32 height, UINT_32 numSamples, + AddrTileMode tileMode, + AddrTileType microTileType, BOOL_32 ignoreSE, BOOL_32 isDepthSampleOrder, + UINT_32 pipeSwizzle, UINT_32 bankSwizzle, + ADDR_TILEINFO* pTileInfo, + UINT_32* pBitPosition) const; + + VOID ComputeSurfaceCoordFromAddrMacroTiled( + UINT_64 addr, UINT_32 bitPosition, + UINT_32 bpp, UINT_32 pitch, UINT_32 height, UINT_32 numSamples, + AddrTileMode tileMode, UINT_32 tileBase, UINT_32 compBits, + AddrTileType microTileType, BOOL_32 ignoreSE, BOOL_32 isDepthSampleOrder, + UINT_32 pipeSwizzle, UINT_32 bankSwizzle, + ADDR_TILEINFO* pTileInfo, + UINT_32* pX, UINT_32* pY, UINT_32* pSlice, UINT_32* pSample) const; + + /// Fmask functions + UINT_64 DispatchComputeFmaskAddrFromCoord( + const ADDR_COMPUTE_FMASK_ADDRFROMCOORD_INPUT* pIn, + ADDR_COMPUTE_FMASK_ADDRFROMCOORD_OUTPUT* pOut) const; + + VOID DispatchComputeFmaskCoordFromAddr( + const ADDR_COMPUTE_FMASK_COORDFROMADDR_INPUT* pIn, + ADDR_COMPUTE_FMASK_COORDFROMADDR_OUTPUT* pOut) const; + + // FMASK related methods - private + UINT_64 ComputeFmaskAddrFromCoordMicroTiled( + UINT_32 x, UINT_32 y, UINT_32 slice, UINT_32 sample, UINT_32 plane, + UINT_32 pitch, UINT_32 height, UINT_32 numSamples, AddrTileMode tileMode, + BOOL_32 resolved, UINT_32* pBitPosition) const; + + VOID ComputeFmaskCoordFromAddrMicroTiled( + UINT_64 addr, UINT_32 bitPosition, + UINT_32 pitch, UINT_32 height, UINT_32 numSamples, + AddrTileMode tileMode, BOOL_32 resolved, + UINT_32* pX, UINT_32* pY, UINT_32* pSlice, UINT_32* pSample, UINT_32* pPlane) const; + + VOID ComputeFmaskCoordFromAddrMacroTiled( + UINT_64 addr, UINT_32 bitPosition, + UINT_32 pitch, UINT_32 height, UINT_32 numSamples, AddrTileMode tileMode, + UINT_32 pipeSwizzle, UINT_32 bankSwizzle, + BOOL_32 ignoreSE, + ADDR_TILEINFO* pTileInfo, + BOOL_32 resolved, + UINT_32* pX, UINT_32* pY, UINT_32* pSlice, UINT_32* pSample, UINT_32* pPlane) const; + + UINT_64 ComputeFmaskAddrFromCoordMacroTiled( + UINT_32 x, UINT_32 y, UINT_32 slice, UINT_32 sample, UINT_32 plane, + UINT_32 pitch, UINT_32 height, UINT_32 numSamples, + AddrTileMode tileMode, UINT_32 pipeSwizzle, UINT_32 bankSwizzle, + BOOL_32 ignoreSE, + ADDR_TILEINFO* pTileInfo, + BOOL_32 resolved, + UINT_32* pBitPosition) const; + + /// Sanity check functions + BOOL_32 SanityCheckMacroTiled( + ADDR_TILEINFO* pTileInfo) const; + +protected: + UINT_32 m_ranks; ///< Number of ranks - MC_ARB_RAMCFG.NOOFRANK + UINT_32 m_logicalBanks; ///< Logical banks = m_banks * m_ranks if m_banks != 16 + UINT_32 m_bankInterleave; ///< Bank interleave, as a multiple of pipe interleave size +}; + +#endif + diff --git a/src/gallium/winsys/radeon/amdgpu/addrlib/r800/siaddrlib.cpp b/src/gallium/winsys/radeon/amdgpu/addrlib/r800/siaddrlib.cpp new file mode 100644 index 0000000..a858b55 --- /dev/null +++ b/src/gallium/winsys/radeon/amdgpu/addrlib/r800/siaddrlib.cpp @@ -0,0 +1,2818 @@ +/* + * Copyright © 2014 Advanced Micro Devices, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES + * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NON-INFRINGEMENT. IN NO EVENT SHALL THE COPYRIGHT HOLDERS, AUTHORS + * AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + */ + +/** +*************************************************************************************************** +* @file siaddrlib.cpp +* @brief Contains the implementation for the SIAddrLib class. +*************************************************************************************************** +*/ + +#include "siaddrlib.h" + +#include "si_gb_reg.h" + +#include "si_ci_vi_merged_enum.h" + +#if BRAHMA_BUILD +#include "amdgpu_id.h" +#else +#include "si_id.h" +#endif + +/////////////////////////////////////////////////////////////////////////////////////////////////// +/////////////////////////////////////////////////////////////////////////////////////////////////// + +/** +*************************************************************************************************** +* AddrSIHwlInit +* +* @brief +* Creates an SIAddrLib object. +* +* @return +* Returns an SIAddrLib object pointer. +*************************************************************************************************** +*/ +AddrLib* AddrSIHwlInit(const AddrClient* pClient) +{ + return SIAddrLib::CreateObj(pClient); +} + +/** +*************************************************************************************************** +* SIAddrLib::SIAddrLib +* +* @brief +* Constructor +* +*************************************************************************************************** +*/ +SIAddrLib::SIAddrLib(const AddrClient* pClient) : + EgBasedAddrLib(pClient), + m_noOfEntries(0) +{ + m_class = SI_ADDRLIB; + memset(&m_settings, 0, sizeof(m_settings)); +} + +/** +*************************************************************************************************** +* SIAddrLib::~SIAddrLib +* +* @brief +* Destructor +*************************************************************************************************** +*/ +SIAddrLib::~SIAddrLib() +{ +} + +/** +*************************************************************************************************** +* SIAddrLib::HwlGetPipes +* +* @brief +* Get number pipes +* @return +* num pipes +*************************************************************************************************** +*/ +UINT_32 SIAddrLib::HwlGetPipes( + const ADDR_TILEINFO* pTileInfo ///< [in] Tile info + ) const +{ + UINT_32 numPipes; + + if (pTileInfo) + { + numPipes = GetPipePerSurf(pTileInfo->pipeConfig); + } + else + { + ADDR_ASSERT_ALWAYS(); + numPipes = m_pipes; // Suppose we should still have a global pipes + } + + return numPipes; +} + +/** +*************************************************************************************************** +* SIAddrLib::GetPipePerSurf +* @brief +* get pipe num base on inputing tileinfo->pipeconfig +* @return +* pipe number +*************************************************************************************************** +*/ +UINT_32 SIAddrLib::GetPipePerSurf( + AddrPipeCfg pipeConfig ///< [in] pipe config + ) const +{ + UINT_32 numPipes = 0; + + switch (pipeConfig) + { + case ADDR_PIPECFG_P2: + numPipes = 2; + break; + case ADDR_PIPECFG_P4_8x16: + case ADDR_PIPECFG_P4_16x16: + case ADDR_PIPECFG_P4_16x32: + case ADDR_PIPECFG_P4_32x32: + numPipes = 4; + break; + case ADDR_PIPECFG_P8_16x16_8x16: + case ADDR_PIPECFG_P8_16x32_8x16: + case ADDR_PIPECFG_P8_32x32_8x16: + case ADDR_PIPECFG_P8_16x32_16x16: + case ADDR_PIPECFG_P8_32x32_16x16: + case ADDR_PIPECFG_P8_32x32_16x32: + case ADDR_PIPECFG_P8_32x64_32x32: + numPipes = 8; + break; + case ADDR_PIPECFG_P16_32x32_8x16: + case ADDR_PIPECFG_P16_32x32_16x16: + numPipes = 16; + break; + default: + ADDR_ASSERT(!"Invalid pipe config"); + numPipes = m_pipes; + } + return numPipes; +} + +/** +*************************************************************************************************** +* SIAddrLib::ComputePipeFromCoord +* +* @brief +* Compute pipe number from coordinates +* @return +* Pipe number +*************************************************************************************************** +*/ +UINT_32 SIAddrLib::ComputePipeFromCoord( + UINT_32 x, ///< [in] x coordinate + UINT_32 y, ///< [in] y coordinate + UINT_32 slice, ///< [in] slice index + AddrTileMode tileMode, ///< [in] tile mode + UINT_32 pipeSwizzle, ///< [in] pipe swizzle + BOOL_32 ignoreSE, ///< [in] TRUE if shader engines are ignored + ADDR_TILEINFO* pTileInfo ///< [in] Tile info + ) const +{ + UINT_32 pipe; + UINT_32 pipeBit0 = 0; + UINT_32 pipeBit1 = 0; + UINT_32 pipeBit2 = 0; + UINT_32 pipeBit3 = 0; + UINT_32 sliceRotation; + UINT_32 numPipes = 0; + + UINT_32 tx = x / MicroTileWidth; + UINT_32 ty = y / MicroTileHeight; + UINT_32 x3 = _BIT(tx,0); + UINT_32 x4 = _BIT(tx,1); + UINT_32 x5 = _BIT(tx,2); + UINT_32 x6 = _BIT(tx,3); + UINT_32 y3 = _BIT(ty,0); + UINT_32 y4 = _BIT(ty,1); + UINT_32 y5 = _BIT(ty,2); + UINT_32 y6 = _BIT(ty,3); + + switch (pTileInfo->pipeConfig) + { + case ADDR_PIPECFG_P2: + pipeBit0 = x3 ^ y3; + numPipes = 2; + break; + case ADDR_PIPECFG_P4_8x16: + pipeBit0 = x4 ^ y3; + pipeBit1 = x3 ^ y4; + numPipes = 4; + break; + case ADDR_PIPECFG_P4_16x16: + pipeBit0 = x3 ^ y3 ^ x4; + pipeBit1 = x4 ^ y4; + numPipes = 4; + break; + case ADDR_PIPECFG_P4_16x32: + pipeBit0 = x3 ^ y3 ^ x4; + pipeBit1 = x4 ^ y5; + numPipes = 4; + break; + case ADDR_PIPECFG_P4_32x32: + pipeBit0 = x3 ^ y3 ^ x5; + pipeBit1 = x5 ^ y5; + numPipes = 4; + break; + case ADDR_PIPECFG_P8_16x16_8x16: + pipeBit0 = x4 ^ y3 ^ x5; + pipeBit1 = x3 ^ y5; + numPipes = 8; + break; + case ADDR_PIPECFG_P8_16x32_8x16: + pipeBit0 = x4 ^ y3 ^ x5; + pipeBit1 = x3 ^ y4; + pipeBit2 = x4 ^ y5; + numPipes = 8; + break; + case ADDR_PIPECFG_P8_16x32_16x16: + pipeBit0 = x3 ^ y3 ^ x4; + pipeBit1 = x5 ^ y4; + pipeBit2 = x4 ^ y5; + numPipes = 8; + break; + case ADDR_PIPECFG_P8_32x32_8x16: + pipeBit0 = x4 ^ y3 ^ x5; + pipeBit1 = x3 ^ y4; + pipeBit2 = x5 ^ y5; + numPipes = 8; + break; + case ADDR_PIPECFG_P8_32x32_16x16: + pipeBit0 = x3 ^ y3 ^ x4; + pipeBit1 = x4 ^ y4; + pipeBit2 = x5 ^ y5; + numPipes = 8; + break; + case ADDR_PIPECFG_P8_32x32_16x32: + pipeBit0 = x3 ^ y3 ^ x4; + pipeBit1 = x4 ^ y6; + pipeBit2 = x5 ^ y5; + numPipes = 8; + break; + case ADDR_PIPECFG_P8_32x64_32x32: + pipeBit0 = x3 ^ y3 ^ x5; + pipeBit1 = x6 ^ y5; + pipeBit2 = x5 ^ y6; + numPipes = 8; + break; + case ADDR_PIPECFG_P16_32x32_8x16: + pipeBit0 = x4 ^ y3; + pipeBit1 = x3 ^ y4; + pipeBit2 = x5 ^ y6; + pipeBit3 = x6 ^ y5; + numPipes = 16; + break; + case ADDR_PIPECFG_P16_32x32_16x16: + pipeBit0 = x3 ^ y3 ^ x4; + pipeBit1 = x4 ^ y4; + pipeBit2 = x5 ^ y6; + pipeBit3 = x6 ^ y5; + numPipes = 16; + break; + default: + ADDR_UNHANDLED_CASE(); + break; + } + pipe = pipeBit0 | (pipeBit1 << 1) | (pipeBit2 << 2) | (pipeBit3 << 3); + + UINT_32 microTileThickness = ComputeSurfaceThickness(tileMode); + + // + // Apply pipe rotation for the slice. + // + switch (tileMode) + { + case ADDR_TM_3D_TILED_THIN1: //fall through thin + case ADDR_TM_3D_TILED_THICK: //fall through thick + case ADDR_TM_3D_TILED_XTHICK: + sliceRotation = + Max(1, static_cast(numPipes / 2) - 1) * (slice / microTileThickness); + break; + default: + sliceRotation = 0; + break; + } + pipeSwizzle += sliceRotation; + pipeSwizzle &= (numPipes - 1); + + pipe = pipe ^ pipeSwizzle; + + return pipe; +} + +/** +*************************************************************************************************** +* SIAddrLib::ComputeTileCoordFromPipeAndElemIdx +* +* @brief +* Compute (x,y) of a tile within a macro tile from address +* @return +* Pipe number +*************************************************************************************************** +*/ +VOID SIAddrLib::ComputeTileCoordFromPipeAndElemIdx( + UINT_32 elemIdx, ///< [in] per pipe element index within a macro tile + UINT_32 pipe, ///< [in] pipe index + AddrPipeCfg pipeCfg, ///< [in] pipe config + UINT_32 pitchInMacroTile, ///< [in] surface pitch in macro tile + UINT_32 x, ///< [in] x coordinate of the (0,0) tile in a macro tile + UINT_32 y, ///< [in] y coordinate of the (0,0) tile in a macro tile + UINT_32* pX, ///< [out] x coordinate + UINT_32* pY ///< [out] y coordinate + ) const +{ + UINT_32 pipebit0 = _BIT(pipe,0); + UINT_32 pipebit1 = _BIT(pipe,1); + UINT_32 pipebit2 = _BIT(pipe,2); + UINT_32 pipebit3 = _BIT(pipe,3); + UINT_32 elemIdx0 = _BIT(elemIdx,0); + UINT_32 elemIdx1 = _BIT(elemIdx,1); + UINT_32 elemIdx2 = _BIT(elemIdx,2); + UINT_32 x3 = 0; + UINT_32 x4 = 0; + UINT_32 x5 = 0; + UINT_32 x6 = 0; + UINT_32 y3 = 0; + UINT_32 y4 = 0; + UINT_32 y5 = 0; + UINT_32 y6 = 0; + + switch(pipeCfg) + { + case ADDR_PIPECFG_P2: + x4 = elemIdx2; + y4 = elemIdx1 ^ x4; + y3 = elemIdx0 ^ x4; + x3 = pipebit0 ^ y3; + *pY = Bits2Number(2, y4, y3); + *pX = Bits2Number(2, x4, x3); + break; + case ADDR_PIPECFG_P4_8x16: + x4 = elemIdx1; + y4 = elemIdx0 ^ x4; + x3 = pipebit1 ^ y4; + y3 = pipebit0 ^ x4; + *pY = Bits2Number(2, y4, y3); + *pX = Bits2Number(2, x4, x3); + break; + case ADDR_PIPECFG_P4_16x16: + x4 = elemIdx1; + y3 = elemIdx0 ^ x4; + y4 = pipebit1 ^ x4; + x3 = pipebit0 ^ y3 ^ x4; + *pY = Bits2Number(2, y4, y3); + *pX = Bits2Number(2, x4, x3); + break; + case ADDR_PIPECFG_P4_16x32: + x3 = elemIdx0 ^ pipebit0; + y5 = _BIT(y,5); + x4 = pipebit1 ^ y5; + y3 = pipebit0 ^ x3 ^ x4; + y4 = elemIdx1 ^ x4; + *pY = Bits2Number(2, y4, y3); + *pX = Bits2Number(2, x4, x3); + break; + case ADDR_PIPECFG_P4_32x32: + x4 = elemIdx2; + y3 = elemIdx0 ^ x4; + y4 = elemIdx1 ^ x4; + if((pitchInMacroTile % 2) == 0) + { //even + y5 = _BIT(y,5); + x5 = pipebit1 ^ y5; + x3 = pipebit0 ^ y3 ^ x5; + *pY = Bits2Number(2, y4, y3); + *pX = Bits2Number(3, x5, x4, x3); + } + else + { //odd + x5 = _BIT(x,5); + x3 = pipebit0 ^ y3 ^ x5; + *pY = Bits2Number(2, y4, y3); + *pX = Bits2Number(2, x4, x3); + } + break; + case ADDR_PIPECFG_P8_16x16_8x16: + x4 = elemIdx0; + y5 = _BIT(y,5); + x5 = _BIT(x,5); + x3 = pipebit1 ^ y5; + y4 = pipebit2 ^ x4; + y3 = pipebit0 ^ x5 ^ x4; + *pY = Bits2Number(2, y4, y3); + *pX = Bits2Number(2, x4, x3); + break; + case ADDR_PIPECFG_P8_16x32_8x16: + x3 = elemIdx0; + y4 = pipebit1 ^ x3; + y5 = _BIT(y,5); + x5 = _BIT(x,5); + x4 = pipebit2 ^ y5; + y3 = pipebit0 ^ x4 ^ x5; + *pY = Bits2Number(2, y4, y3); + *pX = Bits2Number(2, x4, x3); + break; + case ADDR_PIPECFG_P8_32x32_8x16: + x4 = elemIdx1; + y4 = elemIdx0 ^ x4; + x3 = pipebit1 ^ y4; + if((pitchInMacroTile % 2) == 0) + { //even + y5 = _BIT(y,5); + x5 = _BIT(x,5); + x5 = pipebit2 ^ y5; + y3 = pipebit0 ^ x4 ^ x5; + *pY = Bits2Number(2, y4, y3); + *pX = Bits2Number(3, x5, x4, x3); + } + else + { //odd + x5 = _BIT(x,5); + y3 = pipebit0 ^ x4 ^ x5; + *pY = Bits2Number(2, y4, y3); + *pX = Bits2Number(2, x4, x3); + } + break; + case ADDR_PIPECFG_P8_16x32_16x16: + x3 = elemIdx0; + x5 = _BIT(x,5); + y5 = _BIT(y,5); + x4 = pipebit2 ^ y5; + y4 = pipebit1 ^ x5; + y3 = pipebit0 ^ x3 ^ x4; + *pY = Bits2Number(2, y4, y3); + *pX = Bits2Number(2, x4, x3); + break; + case ADDR_PIPECFG_P8_32x32_16x16: + x4 = elemIdx1; + y3 = elemIdx0 ^ x4; + x3 = y3^x4^pipebit0; + y4 = pipebit1 ^ x4; + if((pitchInMacroTile % 2) == 0) + { //even + y5 = _BIT(y,5); + x5 = pipebit2 ^ y5; + *pY = Bits2Number(2, y4, y3); + *pX = Bits2Number(3, x5, x4, x3); + } + else + { //odd + *pY = Bits2Number(2, y4, y3); + *pX = Bits2Number(2, x4, x3); + } + break; + case ADDR_PIPECFG_P8_32x32_16x32: + if((pitchInMacroTile % 2) == 0) + { //even + y5 = _BIT(y,5); + y6 = _BIT(y,6); + x4 = pipebit1 ^ y6; + y3 = elemIdx0 ^ x4; + y4 = elemIdx1 ^ x4; + x3 = pipebit0 ^ y3 ^ x4; + x5 = pipebit2 ^ y5; + *pY = Bits2Number(2, y4, y3); + *pX = Bits2Number(3, x5, x4, x3); + } + else + { //odd + y6 = _BIT(y,6); + x4 = pipebit1 ^ y6; + y3 = elemIdx0 ^ x4; + y4 = elemIdx1 ^ x4; + x3 = pipebit0 ^ y3 ^ x4; + *pY = Bits2Number(2, y4, y3); + *pX = Bits2Number(2, x4, x3); + } + break; + case ADDR_PIPECFG_P8_32x64_32x32: + x4 = elemIdx2; + y3 = elemIdx0 ^ x4; + y4 = elemIdx1 ^ x4; + if((pitchInMacroTile % 4) == 0) + { //multiple of 4 + y5 = _BIT(y,5); + y6 = _BIT(y,6); + x5 = pipebit2 ^ y6; + x6 = pipebit1 ^ y5; + x3 = pipebit0 ^ y3 ^ x5; + *pY = Bits2Number(2, y4, y3); + *pX = Bits2Number(4, x6, x5, x4, x3); + } + else + { + y6 = _BIT(y,6); + x5 = pipebit2 ^ y6; + x3 = pipebit0 ^ y3 ^ x5; + *pY = Bits2Number(2, y4, y3); + *pX = Bits2Number(3, x5, x4, x3); + } + break; + case ADDR_PIPECFG_P16_32x32_8x16: + x4 = elemIdx1; + y4 = elemIdx0 ^ x4; + y3 = pipebit0 ^ x4; + x3 = pipebit1 ^ y4; + if((pitchInMacroTile % 4) == 0) + { //multiple of 4 + y5 = _BIT(y,5); + y6 = _BIT(y,6); + x5 = pipebit2 ^ y6; + x6 = pipebit3 ^ y5; + *pY = Bits2Number(2, y4, y3); + *pX = Bits2Number(4, x6, x5,x4, x3); + } + else + { + y6 = _BIT(y,6); + x5 = pipebit2 ^ y6; + *pY = Bits2Number(2, y4, y3); + *pX = Bits2Number(3, x5, x4, x3); + } + break; + case ADDR_PIPECFG_P16_32x32_16x16: + x4 = elemIdx1; + y3 = elemIdx0 ^ x4; + y4 = pipebit1 ^ x4; + x3 = pipebit0 ^ y3 ^ x4; + if((pitchInMacroTile % 4) == 0) + { //multiple of 4 + y5 = _BIT(y,5); + y6 = _BIT(y,6); + x5 = pipebit2 ^ y6; + x6 = pipebit3 ^ y5; + *pY = Bits2Number(2, y4, y3); + *pX = Bits2Number(4, x6, x5, x4, x3); + } + else + { + y6 = _BIT(y,6); + x5 = pipebit2 ^ y6; + *pY = Bits2Number(2, y4, y3); + *pX = Bits2Number(3, x5, x4, x3); + } + break; + default: + ADDR_UNHANDLED_CASE(); + } +} + +/** +*************************************************************************************************** +* SIAddrLib::TileCoordToMaskElementIndex +* +* @brief +* Compute element index from coordinates in tiles +* @return +* Element index +*************************************************************************************************** +*/ +UINT_32 SIAddrLib::TileCoordToMaskElementIndex( + UINT_32 tx, ///< [in] x coord, in Tiles + UINT_32 ty, ///< [in] y coord, in Tiles + AddrPipeCfg pipeConfig, ///< [in] pipe config + UINT_32* macroShift, ///< [out] macro shift + UINT_32* elemIdxBits ///< [out] tile offset bits + ) const +{ + UINT_32 elemIdx = 0; + UINT_32 elemIdx0, elemIdx1, elemIdx2; + UINT_32 tx0, tx1; + UINT_32 ty0, ty1; + + tx0 = _BIT(tx,0); + tx1 = _BIT(tx,1); + ty0 = _BIT(ty,0); + ty1 = _BIT(ty,1); + + switch(pipeConfig) + { + case ADDR_PIPECFG_P2: + *macroShift = 3; + *elemIdxBits =3; + elemIdx2 = tx1; + elemIdx1 = tx1 ^ ty1; + elemIdx0 = tx1 ^ ty0; + elemIdx = Bits2Number(3,elemIdx2,elemIdx1,elemIdx0); + break; + case ADDR_PIPECFG_P4_8x16: + *macroShift = 2; + *elemIdxBits =2; + elemIdx1 = tx1; + elemIdx0 = tx1 ^ ty1; + elemIdx = Bits2Number(2,elemIdx1,elemIdx0); + break; + case ADDR_PIPECFG_P4_16x16: + *macroShift = 2; + *elemIdxBits =2; + elemIdx0 = tx1^ty0; + elemIdx1 = tx1; + elemIdx = Bits2Number(2, elemIdx1, elemIdx0); + break; + case ADDR_PIPECFG_P4_16x32: + *macroShift = 2; + *elemIdxBits =2; + elemIdx0 = tx1^ty0; + elemIdx1 = tx1^ty1; + elemIdx = Bits2Number(2, elemIdx1, elemIdx0); + break; + case ADDR_PIPECFG_P4_32x32: + *macroShift = 2; + *elemIdxBits =3; + elemIdx0 = tx1^ty0; + elemIdx1 = tx1^ty1; + elemIdx2 = tx1; + elemIdx = Bits2Number(3, elemIdx2, elemIdx1, elemIdx0); + break; + case ADDR_PIPECFG_P8_16x16_8x16: + *macroShift = 1; + *elemIdxBits =1; + elemIdx0 = tx1; + elemIdx = elemIdx0; + break; + case ADDR_PIPECFG_P8_16x32_8x16: + *macroShift = 1; + *elemIdxBits =1; + elemIdx0 = tx0; + elemIdx = elemIdx0; + break; + case ADDR_PIPECFG_P8_32x32_8x16: + *macroShift = 1; + *elemIdxBits =2; + elemIdx1 = tx1; + elemIdx0 = tx1^ty1; + elemIdx = Bits2Number(2, elemIdx1, elemIdx0); + break; + case ADDR_PIPECFG_P8_16x32_16x16: + *macroShift = 1; + *elemIdxBits =1; + elemIdx0 = tx0; + elemIdx = elemIdx0; + break; + case ADDR_PIPECFG_P8_32x32_16x16: + *macroShift = 1; + *elemIdxBits =2; + elemIdx0 = tx1^ty0; + elemIdx1 = tx1; + elemIdx = Bits2Number(2, elemIdx1, elemIdx0); + break; + case ADDR_PIPECFG_P8_32x32_16x32: + *macroShift = 1; + *elemIdxBits =2; + elemIdx0 = tx1^ty0; + elemIdx1 = tx1^ty1; + elemIdx = Bits2Number(2, elemIdx1, elemIdx0); + break; + case ADDR_PIPECFG_P8_32x64_32x32: + *macroShift = 1; + *elemIdxBits =3; + elemIdx0 = tx1^ty0; + elemIdx1 = tx1^ty1; + elemIdx2 = tx1; + elemIdx = Bits2Number(3, elemIdx2, elemIdx1, elemIdx0); + break; + case ADDR_PIPECFG_P16_32x32_8x16: + *macroShift = 0; + *elemIdxBits =2; + elemIdx0 = tx1^ty1; + elemIdx1 = tx1; + elemIdx = Bits2Number(2, elemIdx1, elemIdx0); + break; + case ADDR_PIPECFG_P16_32x32_16x16: + *macroShift = 0; + *elemIdxBits =2; + elemIdx0 = tx1^ty0; + elemIdx1 = tx1; + elemIdx = Bits2Number(2, elemIdx1, elemIdx0); + break; + default: + ADDR_UNHANDLED_CASE(); + break; + } + + return elemIdx; +} + +/** +*************************************************************************************************** +* SIAddrLib::HwlComputeTileDataWidthAndHeightLinear +* +* @brief +* Compute the squared cache shape for per-tile data (CMASK and HTILE) for linear layout +* +* @return +* N/A +* +* @note +* MacroWidth and macroHeight are measured in pixels +*************************************************************************************************** +*/ +VOID SIAddrLib::HwlComputeTileDataWidthAndHeightLinear( + UINT_32* pMacroWidth, ///< [out] macro tile width + UINT_32* pMacroHeight, ///< [out] macro tile height + UINT_32 bpp, ///< [in] bits per pixel + ADDR_TILEINFO* pTileInfo ///< [in] tile info + ) const +{ + ADDR_ASSERT(pTileInfo != NULL); + UINT_32 macroWidth; + UINT_32 macroHeight; + + /// In linear mode, the htile or cmask buffer must be padded out to 4 tiles + /// but for P8_32x64_32x32, it must be padded out to 8 tiles + /// Actually there are more pipe configs which need 8-tile padding but SI family + /// has a bug which is fixed in CI family + if ((pTileInfo->pipeConfig == ADDR_PIPECFG_P8_32x64_32x32) || + (pTileInfo->pipeConfig == ADDR_PIPECFG_P16_32x32_8x16) || + (pTileInfo->pipeConfig == ADDR_PIPECFG_P8_32x32_16x16)) + { + macroWidth = 8*MicroTileWidth; + macroHeight = 8*MicroTileHeight; + } + else + { + macroWidth = 4*MicroTileWidth; + macroHeight = 4*MicroTileHeight; + } + + *pMacroWidth = macroWidth; + *pMacroHeight = macroHeight; +} + +/** +*************************************************************************************************** +* SIAddrLib::HwlComputeHtileBytes +* +* @brief +* Compute htile size in bytes +* +* @return +* Htile size in bytes +*************************************************************************************************** +*/ +UINT_64 SIAddrLib::HwlComputeHtileBytes( + UINT_32 pitch, ///< [in] pitch + UINT_32 height, ///< [in] height + UINT_32 bpp, ///< [in] bits per pixel + BOOL_32 isLinear, ///< [in] if it is linear mode + UINT_32 numSlices, ///< [in] number of slices + UINT_64* pSliceBytes, ///< [out] bytes per slice + UINT_32 baseAlign ///< [in] base alignments + ) const +{ + return ComputeHtileBytes(pitch, height, bpp, isLinear, numSlices, pSliceBytes, baseAlign); +} + +/** +*************************************************************************************************** +* SIAddrLib::HwlComputeXmaskAddrFromCoord +* +* @brief +* Compute address from coordinates for htile/cmask +* @return +* Byte address +*************************************************************************************************** +*/ +UINT_64 SIAddrLib::HwlComputeXmaskAddrFromCoord( + UINT_32 pitch, ///< [in] pitch + UINT_32 height, ///< [in] height + UINT_32 x, ///< [in] x coord + UINT_32 y, ///< [in] y coord + UINT_32 slice, ///< [in] slice/depth index + UINT_32 numSlices, ///< [in] number of slices + UINT_32 factor, ///< [in] factor that indicates cmask(2) or htile(1) + BOOL_32 isLinear, ///< [in] linear or tiled HTILE layout + BOOL_32 isWidth8, ///< [in] TRUE if width is 8, FALSE means 4. It's register value + BOOL_32 isHeight8, ///< [in] TRUE if width is 8, FALSE means 4. It's register value + ADDR_TILEINFO* pTileInfo, ///< [in] Tile info + UINT_32* pBitPosition ///< [out] bit position inside a byte + ) const +{ + UINT_32 tx = x / MicroTileWidth; + UINT_32 ty = y / MicroTileHeight; + UINT_32 newPitch; + UINT_32 newHeight; + UINT_64 totalBytes; + UINT_32 macroWidth; + UINT_32 macroHeight; + UINT_64 pSliceBytes; + UINT_32 pBaseAlign; + UINT_32 tileNumPerPipe; + UINT_32 elemBits; + + if (factor == 2) //CMASK + { + ADDR_CMASK_FLAGS flags = {{0}}; + + tileNumPerPipe = 256; + + ComputeCmaskInfo(flags, + pitch, + height, + numSlices, + isLinear, + pTileInfo, + &newPitch, + &newHeight, + &totalBytes, + ¯oWidth, + ¯oHeight); + elemBits = CmaskElemBits; + } + else //HTile + { + ADDR_HTILE_FLAGS flags = {{0}}; + + tileNumPerPipe = 512; + + ComputeHtileInfo(flags, + pitch, + height, + numSlices, + isLinear, + TRUE, + TRUE, + pTileInfo, + &newPitch, + &newHeight, + &totalBytes, + ¯oWidth, + ¯oHeight, + &pSliceBytes, + &pBaseAlign); + elemBits = 32; + } + + const UINT_32 pitchInTile = newPitch / MicroTileWidth; + const UINT_32 heightInTile = newHeight / MicroTileWidth; + UINT_64 macroOffset; // Per pipe starting offset of the macro tile in which this tile lies. + UINT_64 microNumber; // Per pipe starting offset of the macro tile in which this tile lies. + UINT_32 microX; + UINT_32 microY; + UINT_64 microOffset; + UINT_32 microShift; + UINT_64 totalOffset; + UINT_32 elemIdxBits; + UINT_32 elemIdx = + TileCoordToMaskElementIndex(tx, ty, pTileInfo->pipeConfig, µShift, &elemIdxBits); + + UINT_32 numPipes = HwlGetPipes(pTileInfo); + + if (isLinear) + { //linear addressing + // Linear addressing is extremelly wasting memory if slice > 1, since each pipe has the full + // slice memory foot print instead of divided by numPipes. + microX = tx / 4; // Macro Tile is 4x4 + microY = ty / 4 ; + microNumber = static_cast(microX + microY * (pitchInTile / 4)) << microShift; + + UINT_32 sliceBits = pitchInTile * heightInTile; + + // do htile single slice alignment if the flag is true + if (m_configFlags.useHtileSliceAlign && (factor == 1)) //Htile + { + sliceBits = PowTwoAlign(sliceBits, BITS_TO_BYTES(HtileCacheBits) * numPipes / elemBits); + } + macroOffset = slice * (sliceBits / numPipes) * elemBits ; + } + else + { //tiled addressing + const UINT_32 macroWidthInTile = macroWidth / MicroTileWidth; // Now in unit of Tiles + const UINT_32 macroHeightInTile = macroHeight / MicroTileHeight; + const UINT_32 pitchInCL = pitchInTile / macroWidthInTile; + const UINT_32 heightInCL = heightInTile / macroHeightInTile; + + const UINT_32 macroX = x / macroWidth; + const UINT_32 macroY = y / macroHeight; + const UINT_32 macroNumber = macroX + macroY * pitchInCL + slice * pitchInCL * heightInCL; + + // Per pipe starting offset of the cache line in which this tile lies. + microX = (x % macroWidth) / MicroTileWidth / 4; // Macro Tile is 4x4 + microY = (y % macroHeight) / MicroTileHeight / 4 ; + microNumber = static_cast(microX + microY * (macroWidth / MicroTileWidth / 4)) << microShift; + + macroOffset = macroNumber * tileNumPerPipe * elemBits; + } + + if(elemIdxBits == microShift) + { + microNumber += elemIdx; + } + else + { + microNumber >>= elemIdxBits; + microNumber <<= elemIdxBits; + microNumber += elemIdx; + } + + microOffset = elemBits * microNumber; + totalOffset = microOffset + macroOffset; + + UINT_32 pipe = ComputePipeFromCoord(x, y, 0, ADDR_TM_2D_TILED_THIN1, 0, FALSE, pTileInfo); + UINT_64 addrInBits = totalOffset % (m_pipeInterleaveBytes * 8) + + pipe * (m_pipeInterleaveBytes * 8) + + totalOffset / (m_pipeInterleaveBytes * 8) * (m_pipeInterleaveBytes * 8) * numPipes; + *pBitPosition = static_cast(addrInBits) % 8; + UINT_64 addr = addrInBits / 8; + + return addr; +} + +/** +*************************************************************************************************** +* SIAddrLib::HwlComputeXmaskCoordFromAddr +* +* @brief +* Compute the coord from an address of a cmask/htile +* +* @return +* N/A +* +* @note +* This method is reused by htile, so rename to Xmask +*************************************************************************************************** +*/ +VOID SIAddrLib::HwlComputeXmaskCoordFromAddr( + UINT_64 addr, ///< [in] address + UINT_32 bitPosition, ///< [in] bitPosition in a byte + UINT_32 pitch, ///< [in] pitch + UINT_32 height, ///< [in] height + UINT_32 numSlices, ///< [in] number of slices + UINT_32 factor, ///< [in] factor that indicates cmask or htile + BOOL_32 isLinear, ///< [in] linear or tiled HTILE layout + BOOL_32 isWidth8, ///< [in] Not used by SI + BOOL_32 isHeight8, ///< [in] Not used by SI + ADDR_TILEINFO* pTileInfo, ///< [in] Tile info + UINT_32* pX, ///< [out] x coord + UINT_32* pY, ///< [out] y coord + UINT_32* pSlice ///< [out] slice index + ) const +{ + UINT_32 newPitch; + UINT_32 newHeight; + UINT_64 totalBytes; + UINT_32 clWidth; + UINT_32 clHeight; + UINT_32 tileNumPerPipe; + UINT_64 sliceBytes; + + *pX = 0; + *pY = 0; + *pSlice = 0; + + if (factor == 2) //CMASK + { + ADDR_CMASK_FLAGS flags = {{0}}; + + tileNumPerPipe = 256; + + ComputeCmaskInfo(flags, + pitch, + height, + numSlices, + isLinear, + pTileInfo, + &newPitch, + &newHeight, + &totalBytes, + &clWidth, + &clHeight); + } + else //HTile + { + ADDR_HTILE_FLAGS flags = {{0}}; + + tileNumPerPipe = 512; + + ComputeHtileInfo(flags, + pitch, + height, + numSlices, + isLinear, + TRUE, + TRUE, + pTileInfo, + &newPitch, + &newHeight, + &totalBytes, + &clWidth, + &clHeight, + &sliceBytes); + } + + const UINT_32 pitchInTile = newPitch / MicroTileWidth; + const UINT_32 heightInTile = newHeight / MicroTileWidth; + const UINT_32 pitchInMacroTile = pitchInTile / 4; + UINT_32 macroShift; + UINT_32 elemIdxBits; + // get macroShift and elemIdxBits + TileCoordToMaskElementIndex(0, 0, pTileInfo->pipeConfig, ¯oShift, &elemIdxBits); + + const UINT_32 numPipes = HwlGetPipes(pTileInfo); + const UINT_32 pipe = (UINT_32)((addr / m_pipeInterleaveBytes) % numPipes); + // per pipe + UINT_64 localOffset = (addr % m_pipeInterleaveBytes) + + (addr / m_pipeInterleaveBytes / numPipes)* m_pipeInterleaveBytes; + + UINT_32 tileIndex; + if (factor == 2) //CMASK + { + tileIndex = (UINT_32)(localOffset * 2 + (bitPosition != 0)); + } + else + { + tileIndex = (UINT_32)(localOffset / 4); + } + + UINT_32 macroOffset; + if (isLinear) + { + UINT_32 sliceSizeInTile = pitchInTile * heightInTile; + + // do htile single slice alignment if the flag is true + if (m_configFlags.useHtileSliceAlign && (factor == 1)) //Htile + { + sliceSizeInTile = PowTwoAlign(sliceSizeInTile, static_cast(sliceBytes) / 64); + } + *pSlice = tileIndex / (sliceSizeInTile / numPipes); + macroOffset = tileIndex % (sliceSizeInTile / numPipes); + } + else + { + const UINT_32 clWidthInTile = clWidth / MicroTileWidth; // Now in unit of Tiles + const UINT_32 clHeightInTile = clHeight / MicroTileHeight; + const UINT_32 pitchInCL = pitchInTile / clWidthInTile; + const UINT_32 heightInCL = heightInTile / clHeightInTile; + const UINT_32 clIndex = tileIndex / tileNumPerPipe; + + UINT_32 clX = clIndex % pitchInCL; + UINT_32 clY = (clIndex % (heightInCL * pitchInCL)) / pitchInCL; + + *pX = clX * clWidthInTile * MicroTileWidth; + *pY = clY * clHeightInTile * MicroTileHeight; + *pSlice = clIndex / (heightInCL * pitchInCL); + + macroOffset = tileIndex % tileNumPerPipe; + } + + UINT_32 elemIdx = macroOffset & 7; + macroOffset >>= elemIdxBits; + + if (elemIdxBits != macroShift) + { + macroOffset <<= (elemIdxBits - macroShift); + + UINT_32 pipebit1 = _BIT(pipe,1); + UINT_32 pipebit2 = _BIT(pipe,2); + UINT_32 pipebit3 = _BIT(pipe,3); + if (pitchInMacroTile % 2) + { //odd + switch (pTileInfo->pipeConfig) + { + case ADDR_PIPECFG_P4_32x32: + macroOffset |= pipebit1; + break; + case ADDR_PIPECFG_P8_32x32_8x16: + case ADDR_PIPECFG_P8_32x32_16x16: + case ADDR_PIPECFG_P8_32x32_16x32: + macroOffset |= pipebit2; + break; + default: + break; + } + + } + + if (pitchInMacroTile % 4) + { + if (pTileInfo->pipeConfig == ADDR_PIPECFG_P8_32x64_32x32) + { + macroOffset |= (pipebit1<<1); + } + if((pTileInfo->pipeConfig == ADDR_PIPECFG_P16_32x32_8x16) || + (pTileInfo->pipeConfig == ADDR_PIPECFG_P16_32x32_16x16)) + { + macroOffset |= (pipebit3<<1); + } + } + } + + UINT_32 macroX; + UINT_32 macroY; + + if (isLinear) + { + macroX = macroOffset % pitchInMacroTile; + macroY = macroOffset / pitchInMacroTile; + } + else + { + const UINT_32 clWidthInMacroTile = clWidth / (MicroTileWidth * 4); + macroX = macroOffset % clWidthInMacroTile; + macroY = macroOffset / clWidthInMacroTile; + } + + *pX += macroX * 4 * MicroTileWidth; + *pY += macroY * 4 * MicroTileHeight; + + UINT_32 microX; + UINT_32 microY; + ComputeTileCoordFromPipeAndElemIdx(elemIdx, pipe, pTileInfo->pipeConfig, pitchInMacroTile, + *pX, *pY, µX, µY); + + *pX += microX * MicroTileWidth; + *pY += microY * MicroTileWidth; +} + +/** +*************************************************************************************************** +* SIAddrLib::HwlGetPitchAlignmentLinear +* @brief +* Get pitch alignment +* @return +* pitch alignment +*************************************************************************************************** +*/ +UINT_32 SIAddrLib::HwlGetPitchAlignmentLinear( + UINT_32 bpp, ///< [in] bits per pixel + ADDR_SURFACE_FLAGS flags ///< [in] surface flags + ) const +{ + UINT_32 pitchAlign; + + // Interleaved access requires a 256B aligned pitch, so fall back to pre-SI alignment + if (flags.interleaved) + { + pitchAlign = Max(64u, m_pipeInterleaveBytes / BITS_TO_BYTES(bpp)); + + } + else + { + pitchAlign = Max(8u, 64 / BITS_TO_BYTES(bpp)); + } + + return pitchAlign; +} + +/** +*************************************************************************************************** +* SIAddrLib::HwlGetSizeAdjustmentLinear +* +* @brief +* Adjust linear surface pitch and slice size +* +* @return +* Logical slice size in bytes +*************************************************************************************************** +*/ +UINT_64 SIAddrLib::HwlGetSizeAdjustmentLinear( + AddrTileMode tileMode, ///< [in] tile mode + UINT_32 bpp, ///< [in] bits per pixel + UINT_32 numSamples, ///< [in] number of samples + UINT_32 baseAlign, ///< [in] base alignment + UINT_32 pitchAlign, ///< [in] pitch alignment + UINT_32* pPitch, ///< [in/out] pointer to pitch + UINT_32* pHeight, ///< [in/out] pointer to height + UINT_32* pHeightAlign ///< [in/out] pointer to height align + ) const +{ + UINT_64 sliceSize; + if (tileMode == ADDR_TM_LINEAR_GENERAL) + { + sliceSize = BITS_TO_BYTES(static_cast(*pPitch) * (*pHeight) * bpp * numSamples); + } + else + { + UINT_32 pitch = *pPitch; + UINT_32 height = *pHeight; + + UINT_32 pixelsPerPipeInterleave = m_pipeInterleaveBytes / BITS_TO_BYTES(bpp); + UINT_32 sliceAlignInPixel = pixelsPerPipeInterleave < 64 ? 64 : pixelsPerPipeInterleave; + + // numSamples should be 1 in real cases (no MSAA for linear but TGL may pass non 1 value) + UINT_64 pixelPerSlice = static_cast(pitch) * height * numSamples; + + while (pixelPerSlice % sliceAlignInPixel) + { + pitch += pitchAlign; + pixelPerSlice = static_cast(pitch) * height * numSamples; + } + + *pPitch = pitch; + + UINT_32 heightAlign = 1; + + while ((pitch * heightAlign) % sliceAlignInPixel) + { + heightAlign++; + } + + *pHeightAlign = heightAlign; + + sliceSize = BITS_TO_BYTES(pixelPerSlice * bpp); + } + + return sliceSize; +} + +/** +*************************************************************************************************** +* SIAddrLib::HwlPreHandleBaseLvl3xPitch +* +* @brief +* Pre-handler of 3x pitch (96 bit) adjustment +* +* @return +* Expected pitch +*************************************************************************************************** +*/ +UINT_32 SIAddrLib::HwlPreHandleBaseLvl3xPitch( + const ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn, ///< [in] input + UINT_32 expPitch ///< [in] pitch + ) const +{ + ADDR_ASSERT(pIn->width == expPitch); + + // From SI, if pow2Pad is 1 the pitch is expanded 3x first, then padded to pow2, so nothing to + // do here + if (!pIn->flags.pow2Pad) + { + AddrLib::HwlPreHandleBaseLvl3xPitch(pIn, expPitch); + } + else + { + ADDR_ASSERT(IsPow2(expPitch)); + } + + return expPitch; +} + +/** +*************************************************************************************************** +* SIAddrLib::HwlPostHandleBaseLvl3xPitch +* +* @brief +* Post-handler of 3x pitch adjustment +* +* @return +* Expected pitch +*************************************************************************************************** +*/ +UINT_32 SIAddrLib::HwlPostHandleBaseLvl3xPitch( + const ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn, ///< [in] input + UINT_32 expPitch ///< [in] pitch + ) const +{ + /** + * @note The pitch will be divided by 3 in the end so the value will look odd but h/w should + * be able to compute a correct pitch from it as h/w address library is doing the job. + */ + // From SI, the pitch is expanded 3x first, then padded to pow2, so no special handler here + if (!pIn->flags.pow2Pad) + { + AddrLib::HwlPostHandleBaseLvl3xPitch(pIn, expPitch); + } + + return expPitch; +} + +/** +*************************************************************************************************** +* SIAddrLib::HwlGetPitchAlignmentMicroTiled +* +* @brief +* Compute 1D tiled surface pitch alignment +* +* @return +* pitch alignment +*************************************************************************************************** +*/ +UINT_32 SIAddrLib::HwlGetPitchAlignmentMicroTiled( + AddrTileMode tileMode, ///< [in] tile mode + UINT_32 bpp, ///< [in] bits per pixel + ADDR_SURFACE_FLAGS flags, ///< [in] surface flags + UINT_32 numSamples ///< [in] number of samples + ) const +{ + UINT_32 pitchAlign; + + if (flags.qbStereo) + { + pitchAlign = EgBasedAddrLib::HwlGetPitchAlignmentMicroTiled(tileMode,bpp,flags,numSamples); + } + else + { + pitchAlign = 8; + } + + return pitchAlign; +} + +/** +*************************************************************************************************** +* SIAddrLib::HwlGetSizeAdjustmentMicroTiled +* +* @brief +* Adjust 1D tiled surface pitch and slice size +* +* @return +* Logical slice size in bytes +*************************************************************************************************** +*/ +UINT_64 SIAddrLib::HwlGetSizeAdjustmentMicroTiled( + UINT_32 thickness, ///< [in] thickness + UINT_32 bpp, ///< [in] bits per pixel + ADDR_SURFACE_FLAGS flags, ///< [in] surface flags + UINT_32 numSamples, ///< [in] number of samples + UINT_32 baseAlign, ///< [in] base alignment + UINT_32 pitchAlign, ///< [in] pitch alignment + UINT_32* pPitch, ///< [in/out] pointer to pitch + UINT_32* pHeight ///< [in/out] pointer to height + ) const +{ + UINT_64 logicalSliceSize; + UINT_64 physicalSliceSize; + + UINT_32 pitch = *pPitch; + UINT_32 height = *pHeight; + + // Logical slice: pitch * height * bpp * numSamples (no 1D MSAA so actually numSamples == 1) + logicalSliceSize = BITS_TO_BYTES(static_cast(pitch) * height * bpp * numSamples); + + // Physical slice: multiplied by thickness + physicalSliceSize = logicalSliceSize * thickness; + + // Pitch alignment is always 8, so if slice size is not padded to base alignment + // (pipe_interleave_size), we need to increase pitch + while ((physicalSliceSize % baseAlign) != 0) + { + pitch += pitchAlign; + + logicalSliceSize = BITS_TO_BYTES(static_cast(pitch) * height * bpp * numSamples); + + physicalSliceSize = logicalSliceSize * thickness; + } + +#if !ALT_TEST + // + // Special workaround for depth/stencil buffer, use 8 bpp to align depth buffer again since + // the stencil plane may have larger pitch if the slice size is smaller than base alignment. + // + // Note: this actually does not work for mipmap but mipmap depth texture is not really + // sampled with mipmap. + // + if (flags.depth && !flags.noStencil) + { + ADDR_ASSERT(numSamples == 1); + + UINT_64 logicalSiceSizeStencil = static_cast(pitch) * height; // 1 byte stencil + + while ((logicalSiceSizeStencil % baseAlign) != 0) + { + pitch += pitchAlign; // Stencil plane's pitch alignment is the same as depth plane's + + logicalSiceSizeStencil = static_cast(pitch) * height; + } + + if (pitch != *pPitch) + { + // If this is a mipmap, this padded one cannot be sampled as a whole mipmap! + logicalSliceSize = logicalSiceSizeStencil * BITS_TO_BYTES(bpp); + } + } +#endif + *pPitch = pitch; + + // No adjust for pHeight + + return logicalSliceSize; +} + +/** +*************************************************************************************************** +* SIAddrLib::HwlConvertChipFamily +* +* @brief +* Convert familyID defined in atiid.h to AddrChipFamily and set m_chipFamily/m_chipRevision +* @return +* AddrChipFamily +*************************************************************************************************** +*/ +AddrChipFamily SIAddrLib::HwlConvertChipFamily( + UINT_32 uChipFamily, ///< [in] chip family defined in atiih.h + UINT_32 uChipRevision) ///< [in] chip revision defined in "asic_family"_id.h +{ + AddrChipFamily family = ADDR_CHIP_FAMILY_SI; + + switch (uChipFamily) + { + case FAMILY_SI: + m_settings.isSouthernIsland = 1; + m_settings.isTahiti = ASICREV_IS_TAHITI_P(uChipRevision); + m_settings.isPitCairn = ASICREV_IS_PITCAIRN_PM(uChipRevision); + m_settings.isCapeVerde = ASICREV_IS_CAPEVERDE_M(uChipRevision); + m_settings.isOland = ASICREV_IS_OLAND_M(uChipRevision); + m_settings.isHainan = ASICREV_IS_HAINAN_V(uChipRevision); + break; + default: + ADDR_ASSERT(!"This should be a Fusion"); + break; + } + + return family; +} + +/** +*************************************************************************************************** +* SIAddrLib::HwlSetupTileInfo +* +* @brief +* Setup default value of tile info for SI +*************************************************************************************************** +*/ +VOID SIAddrLib::HwlSetupTileInfo( + AddrTileMode tileMode, ///< [in] Tile mode + ADDR_SURFACE_FLAGS flags, ///< [in] Surface type flags + UINT_32 bpp, ///< [in] Bits per pixel + UINT_32 pitch, ///< [in] Pitch in pixels + UINT_32 height, ///< [in] Height in pixels + UINT_32 numSamples, ///< [in] Number of samples + ADDR_TILEINFO* pTileInfoIn, ///< [in] Tile info input: NULL for default + ADDR_TILEINFO* pTileInfoOut, ///< [out] Tile info output + AddrTileType inTileType, ///< [in] Tile type + ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pOut ///< [out] Output + ) const +{ + UINT_32 thickness = ComputeSurfaceThickness(tileMode); + ADDR_TILEINFO* pTileInfo = pTileInfoOut; + INT index = TileIndexInvalid; + + // Fail-safe code + if (!IsLinear(tileMode)) + { + // 128 bpp/thick tiling must be non-displayable. + // Fmask reuse color buffer's entry but bank-height field can be from another entry + // To simplify the logic, fmask entry should be picked from non-displayable ones + if (bpp == 128 || thickness > 1 || flags.fmask || flags.prt) + { + inTileType = ADDR_NON_DISPLAYABLE; + } + + if (flags.depth || flags.stencil) + { + inTileType = ADDR_DEPTH_SAMPLE_ORDER; + } + } + + // Partial valid fields are not allowed for SI. + if (IsTileInfoAllZero(pTileInfo)) + { + if (IsMacroTiled(tileMode)) + { + if (flags.prt) + { + if (numSamples == 1) + { + if (flags.depth) + { + switch (bpp) + { + case 16: + index = 3; + break; + case 32: + index = 6; + break; + default: + ADDR_ASSERT_ALWAYS(); + break; + } + } + else + { + switch (bpp) + { + case 8: + index = 21; + break; + case 16: + index = 22; + break; + case 32: + index = 23; + break; + case 64: + index = 24; + break; + case 128: + index = 25; + break; + default: + break; + } + + if (thickness > 1) + { + ADDR_ASSERT(bpp != 128); + index += 5; + } + } + } + else + { + ADDR_ASSERT(numSamples == 4); + + if (flags.depth) + { + switch (bpp) + { + case 16: + index = 5; + break; + case 32: + index = 7; + break; + default: + ADDR_ASSERT_ALWAYS(); + break; + } + } + else + { + switch (bpp) + { + case 8: + index = 23; + break; + case 16: + index = 24; + break; + case 32: + index = 25; + break; + case 64: + index = 30; + break; + default: + ADDR_ASSERT_ALWAYS(); + break; + } + } + } + }//end of PRT part + // See table entries 0-7 + else if (flags.depth || flags.stencil) + { + if (flags.compressZ) + { + if (flags.stencil) + { + index = 0; + } + else + { + // optimal tile index for compressed depth/stencil. + switch (numSamples) + { + case 1: + index = 0; + break; + case 2: + case 4: + index = 1; + break; + case 8: + index = 2; + break; + default: + break; + } + } + } + else // unCompressZ + { + index = 3; + } + } + else //non PRT & non Depth & non Stencil + { + // See table entries 9-12 + if (inTileType == ADDR_DISPLAYABLE) + { + switch (bpp) + { + case 8: + index = 10; + break; + case 16: + index = 11; + break; + case 32: + index = 12; + break; + case 64: + index = 12; + break; + default: + break; + } + } + else + { + // See table entries 13-17 + if (thickness == 1) + { + if (flags.fmask) + { + UINT_32 fmaskPixelSize = bpp * numSamples; + + switch (fmaskPixelSize) + { + case 8: + index = 14; + break; + case 16: + index = 15; + break; + case 32: + index = 16; + break; + case 64: + index = 17; + break; + default: + ADDR_ASSERT_ALWAYS(); + } + } + else + { + switch (bpp) + { + case 8: + index = 14; + break; + case 16: + index = 15; + break; + case 32: + index = 16; + break; + case 64: + index = 17; + break; + case 128: + index = 17; + break; + default: + break; + } + } + } + else // thick tiling - entries 18-20 + { + switch (thickness) + { + case 4: + index = 20; + break; + case 8: + index = 19; + break; + default: + break; + } + } + } + } + } + else + { + if (tileMode == ADDR_TM_LINEAR_ALIGNED) + { + index = 8; + } + else if (tileMode == ADDR_TM_LINEAR_GENERAL) + { + index = TileIndexLinearGeneral; + } + else + { + if (flags.depth || flags.stencil) + { + index = 4; + } + else if (inTileType == ADDR_DISPLAYABLE) + { + index = 9; + } + else if (thickness == 1) + { + index = 13; + } + else + { + index = 18; + } + } + } + + if (index >= 0 && index <= 31) + { + *pTileInfo = m_tileTable[index].info; + pOut->tileType = m_tileTable[index].type; + } + + if (index == TileIndexLinearGeneral) + { + *pTileInfo = m_tileTable[8].info; + pOut->tileType = m_tileTable[8].type; + } + } + else + { + if (pTileInfoIn) + { + if (flags.stencil && pTileInfoIn->tileSplitBytes == 0) + { + // Stencil always uses index 0 + *pTileInfo = m_tileTable[0].info; + } + } + // Pass through tile type + pOut->tileType = inTileType; + } + + pOut->tileIndex = index; +} + +/** +*************************************************************************************************** +* SIAddrLib::DecodeGbRegs +* +* @brief +* Decodes GB_ADDR_CONFIG and noOfBanks/noOfRanks +* +* @return +* TRUE if all settings are valid +* +*************************************************************************************************** +*/ +BOOL_32 SIAddrLib::DecodeGbRegs( + const ADDR_REGISTER_VALUE* pRegValue) ///< [in] create input +{ + GB_ADDR_CONFIG reg; + BOOL_32 valid = TRUE; + + reg.val = pRegValue->gbAddrConfig; + + switch (reg.f.pipe_interleave_size) + { + case ADDR_CONFIG_PIPE_INTERLEAVE_256B: + m_pipeInterleaveBytes = ADDR_PIPEINTERLEAVE_256B; + break; + case ADDR_CONFIG_PIPE_INTERLEAVE_512B: + m_pipeInterleaveBytes = ADDR_PIPEINTERLEAVE_512B; + break; + default: + valid = FALSE; + ADDR_UNHANDLED_CASE(); + break; + } + + switch (reg.f.row_size) + { + case ADDR_CONFIG_1KB_ROW: + m_rowSize = ADDR_ROWSIZE_1KB; + break; + case ADDR_CONFIG_2KB_ROW: + m_rowSize = ADDR_ROWSIZE_2KB; + break; + case ADDR_CONFIG_4KB_ROW: + m_rowSize = ADDR_ROWSIZE_4KB; + break; + default: + valid = FALSE; + ADDR_UNHANDLED_CASE(); + break; + } + + switch (pRegValue->noOfBanks) + { + case 0: + m_banks = 4; + break; + case 1: + m_banks = 8; + break; + case 2: + m_banks = 16; + break; + default: + valid = FALSE; + ADDR_UNHANDLED_CASE(); + break; + } + + switch (pRegValue->noOfRanks) + { + case 0: + m_ranks = 1; + break; + case 1: + m_ranks = 2; + break; + default: + valid = FALSE; + ADDR_UNHANDLED_CASE(); + break; + } + + m_logicalBanks = m_banks * m_ranks; + + ADDR_ASSERT(m_logicalBanks <= 16); + + return valid; +} + +/** +*************************************************************************************************** +* SIAddrLib::HwlInitGlobalParams +* +* @brief +* Initializes global parameters +* +* @return +* TRUE if all settings are valid +* +*************************************************************************************************** +*/ +BOOL_32 SIAddrLib::HwlInitGlobalParams( + const ADDR_CREATE_INPUT* pCreateIn) ///< [in] create input +{ + BOOL_32 valid = TRUE; + const ADDR_REGISTER_VALUE* pRegValue = &pCreateIn->regValue; + + valid = DecodeGbRegs(pRegValue); + + if (valid) + { + if (m_settings.isTahiti || m_settings.isPitCairn) + { + m_pipes = 8; + } + else if (m_settings.isCapeVerde || m_settings.isOland) + { + m_pipes = 4; + } + else + { + // Hainan is 2-pipe (m_settings.isHainan == 1) + m_pipes = 2; + } + + valid = InitTileSettingTable(pRegValue->pTileConfig, pRegValue->noOfEntries); + + m_maxSamples = 16; + } + + return valid; +} + +/** +*************************************************************************************************** +* SIAddrLib::HwlConvertTileInfoToHW +* @brief +* Entry of si's ConvertTileInfoToHW +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE SIAddrLib::HwlConvertTileInfoToHW( + const ADDR_CONVERT_TILEINFOTOHW_INPUT* pIn, ///< [in] input structure + ADDR_CONVERT_TILEINFOTOHW_OUTPUT* pOut ///< [out] output structure + ) const +{ + ADDR_E_RETURNCODE retCode = ADDR_OK; + + retCode = EgBasedAddrLib::HwlConvertTileInfoToHW(pIn, pOut); + + if (retCode == ADDR_OK) + { + if (pIn->reverse == FALSE) + { + if (pIn->pTileInfo->pipeConfig == ADDR_PIPECFG_INVALID) + { + retCode = ADDR_INVALIDPARAMS; + } + else + { + pOut->pTileInfo->pipeConfig = + static_cast(pIn->pTileInfo->pipeConfig - 1); + } + } + else + { + pOut->pTileInfo->pipeConfig = + static_cast(pIn->pTileInfo->pipeConfig + 1); + } + } + + return retCode; +} + +/** +*************************************************************************************************** +* SIAddrLib::HwlComputeXmaskCoordYFrom8Pipe +* +* @brief +* Compute the Y coord which will be added to Xmask Y +* coord. +* @return +* Y coord +*************************************************************************************************** +*/ +UINT_32 SIAddrLib::HwlComputeXmaskCoordYFrom8Pipe( + UINT_32 pipe, ///< [in] pipe id + UINT_32 x ///< [in] tile coord x, which is original x coord / 8 + ) const +{ + // This function should never be called since it is 6xx/8xx specfic. + // Keep this empty implementation to avoid any mis-use. + ADDR_ASSERT_ALWAYS(); + + return 0; +} + +/** +*************************************************************************************************** +* SIAddrLib::HwlComputeSurfaceCoord2DFromBankPipe +* +* @brief +* Compute surface x,y coordinates from bank/pipe info +* @return +* N/A +*************************************************************************************************** +*/ +VOID SIAddrLib::HwlComputeSurfaceCoord2DFromBankPipe( + AddrTileMode tileMode, ///< [in] tile mode + UINT_32* pX, ///< [in/out] x coordinate + UINT_32* pY, ///< [in/out] y coordinate + UINT_32 slice, ///< [in] slice index + UINT_32 bank, ///< [in] bank number + UINT_32 pipe, ///< [in] pipe number + UINT_32 bankSwizzle,///< [in] bank swizzle + UINT_32 pipeSwizzle,///< [in] pipe swizzle + UINT_32 tileSlices, ///< [in] slices in a micro tile + BOOL_32 ignoreSE, ///< [in] TRUE if shader engines are ignored + ADDR_TILEINFO* pTileInfo ///< [in] bank structure. **All fields to be valid on entry** + ) const +{ + UINT_32 xBit; + UINT_32 yBit; + UINT_32 yBit3 = 0; + UINT_32 yBit4 = 0; + UINT_32 yBit5 = 0; + UINT_32 yBit6 = 0; + + UINT_32 xBit3 = 0; + UINT_32 xBit4 = 0; + UINT_32 xBit5 = 0; + + UINT_32 numPipes = GetPipePerSurf(pTileInfo->pipeConfig); + + CoordFromBankPipe xyBits = {0}; + ComputeSurfaceCoord2DFromBankPipe(tileMode, *pX, *pY, slice, bank, pipe, + bankSwizzle, pipeSwizzle, tileSlices, pTileInfo, + &xyBits); + yBit3 = xyBits.yBit3; + yBit4 = xyBits.yBit4; + yBit5 = xyBits.yBit5; + yBit6 = xyBits.yBit6; + + xBit3 = xyBits.xBit3; + xBit4 = xyBits.xBit4; + xBit5 = xyBits.xBit5; + + yBit = xyBits.yBits; + + UINT_32 yBitTemp = 0; + + if ((pTileInfo->pipeConfig == ADDR_PIPECFG_P4_32x32) || + (pTileInfo->pipeConfig == ADDR_PIPECFG_P8_32x64_32x32)) + { + ADDR_ASSERT(pTileInfo->bankWidth == 1 && pTileInfo->macroAspectRatio > 1); + UINT_32 yBitToCheck = QLog2(pTileInfo->banks) - 1; + + ADDR_ASSERT(yBitToCheck <= 3); + + yBitTemp = _BIT(yBit, yBitToCheck); + + xBit3 = 0; + } + + yBit = Bits2Number(4, yBit6, yBit5, yBit4, yBit3); + xBit = Bits2Number(3, xBit5, xBit4, xBit3); + + *pY += yBit * pTileInfo->bankHeight * MicroTileHeight; + *pX += xBit * numPipes * pTileInfo->bankWidth * MicroTileWidth; + + //calculate the bank and pipe bits in x, y + UINT_32 xTile; //x in micro tile + UINT_32 x3 = 0; + UINT_32 x4 = 0; + UINT_32 x5 = 0; + UINT_32 x6 = 0; + UINT_32 y = *pY; + + UINT_32 pipeBit0 = _BIT(pipe,0); + UINT_32 pipeBit1 = _BIT(pipe,1); + UINT_32 pipeBit2 = _BIT(pipe,2); + + UINT_32 y3 = _BIT(y, 3); + UINT_32 y4 = _BIT(y, 4); + UINT_32 y5 = _BIT(y, 5); + UINT_32 y6 = _BIT(y, 6); + + // bankbit0 after ^x4^x5 + UINT_32 bankBit00 = _BIT(bank,0); + UINT_32 bankBit0 = 0; + + switch (pTileInfo->pipeConfig) + { + case ADDR_PIPECFG_P2: + x3 = pipeBit0 ^ y3; + break; + case ADDR_PIPECFG_P4_8x16: + x4 = pipeBit0 ^ y3; + x3 = pipeBit0 ^ y4; + break; + case ADDR_PIPECFG_P4_16x16: + x4 = pipeBit1 ^ y4; + x3 = pipeBit0 ^ y3 ^ x4; + break; + case ADDR_PIPECFG_P4_16x32: + x4 = pipeBit1 ^ y4; + x3 = pipeBit0 ^ y3 ^ x4; + break; + case ADDR_PIPECFG_P4_32x32: + x5 = pipeBit1 ^ y5; + x3 = pipeBit0 ^ y3 ^ x5; + bankBit0 = yBitTemp ^ x5; + x4 = bankBit00 ^ x5 ^ bankBit0; + *pX += x5 * 4 * 1 * 8; // x5 * num_pipes * bank_width * 8; + break; + case ADDR_PIPECFG_P8_16x16_8x16: + x3 = pipeBit1 ^ y5; + x4 = pipeBit2 ^ y4; + x5 = pipeBit0 ^ y3 ^ x4; + break; + case ADDR_PIPECFG_P8_16x32_8x16: + x3 = pipeBit1 ^ y4; + x4 = pipeBit2 ^ y5; + x5 = pipeBit0 ^ y3 ^ x4; + break; + case ADDR_PIPECFG_P8_32x32_8x16: + x3 = pipeBit1 ^ y4; + x5 = pipeBit2 ^ y5; + x4 = pipeBit0 ^ y3 ^ x5; + break; + case ADDR_PIPECFG_P8_16x32_16x16: + x4 = pipeBit2 ^ y5; + x5 = pipeBit1 ^ y4; + x3 = pipeBit0 ^ y3 ^ x4; + break; + case ADDR_PIPECFG_P8_32x32_16x16: + x5 = pipeBit2 ^ y5; + x4 = pipeBit1 ^ y4; + x3 = pipeBit0 ^ y3 ^ x4; + break; + case ADDR_PIPECFG_P8_32x32_16x32: + x5 = pipeBit2 ^ y5; + x4 = pipeBit1 ^ y6; + x3 = pipeBit0 ^ y3 ^ x4; + break; + case ADDR_PIPECFG_P8_32x64_32x32: + x6 = pipeBit1 ^ y5; + x5 = pipeBit2 ^ y6; + x3 = pipeBit0 ^ y3 ^ x5; + bankBit0 = yBitTemp ^ x6; + x4 = bankBit00 ^ x5 ^ bankBit0; + *pX += x6 * 8 * 1 * 8; // x6 * num_pipes * bank_width * 8; + break; + default: + ADDR_ASSERT_ALWAYS(); + } + + xTile = Bits2Number(3, x5, x4, x3); + + *pX += xTile << 3; +} + +/** +*************************************************************************************************** +* SIAddrLib::HwlPreAdjustBank +* +* @brief +* Adjust bank before calculating address acoording to bank/pipe +* @return +* Adjusted bank +*************************************************************************************************** +*/ +UINT_32 SIAddrLib::HwlPreAdjustBank( + UINT_32 tileX, ///< [in] x coordinate in unit of tile + UINT_32 bank, ///< [in] bank + ADDR_TILEINFO* pTileInfo ///< [in] tile info + ) const +{ + if (((pTileInfo->pipeConfig == ADDR_PIPECFG_P4_32x32) || + (pTileInfo->pipeConfig == ADDR_PIPECFG_P8_32x64_32x32)) && (pTileInfo->bankWidth == 1)) + { + UINT_32 bankBit0 = _BIT(bank, 0); + UINT_32 x4 = _BIT(tileX, 1); + UINT_32 x5 = _BIT(tileX, 2); + + bankBit0 = bankBit0 ^ x4 ^ x5; + bank |= bankBit0; + + ADDR_ASSERT(pTileInfo->macroAspectRatio > 1) + } + + return bank; +} + +/** +*************************************************************************************************** +* SIAddrLib::HwlComputeSurfaceInfo +* +* @brief +* Entry of si's ComputeSurfaceInfo +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE SIAddrLib::HwlComputeSurfaceInfo( + const ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn, ///< [in] input structure + ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pOut ///< [out] output structure + ) const +{ + pOut->tileIndex = pIn->tileIndex; + + return EgBasedAddrLib::HwlComputeSurfaceInfo(pIn,pOut); +} + +/** +*************************************************************************************************** +* SIAddrLib::HwlComputeMipLevel +* @brief +* Compute MipLevel info (including level 0) +* @return +* TRUE if HWL's handled +*************************************************************************************************** +*/ +BOOL_32 SIAddrLib::HwlComputeMipLevel( + ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn ///< [in/out] Input structure + ) const +{ + // basePitch is calculated from level 0 so we only check this for mipLevel > 0 + if (pIn->mipLevel > 0) + { + // Note: Don't check expand 3x formats(96 bit) as the basePitch is not pow2 even if + // we explicity set pow2Pad flag. The 3x base pitch is padded to pow2 but after being + // divided by expandX factor (3) - to program texture pitch, the basePitch is never pow2. + if (!AddrElemLib::IsExpand3x(pIn->format)) + { + // Sublevel pitches are generated from base level pitch instead of width on SI + // If pow2Pad is 0, we don't assert - as this is not really used for a mip chain + ADDR_ASSERT(!pIn->flags.pow2Pad || ((pIn->basePitch != 0) && IsPow2(pIn->basePitch))); + } + + if (pIn->basePitch != 0) + { + pIn->width = Max(1u, pIn->basePitch >> pIn->mipLevel); + } + } + + // pow2Pad is done in PostComputeMipLevel + + return TRUE; +} + +/** +*************************************************************************************************** +* SIAddrLib::HwlCheckLastMacroTiledLvl +* +* @brief +* Sets pOut->last2DLevel to TRUE if it is +* @note +* +*************************************************************************************************** +*/ +VOID SIAddrLib::HwlCheckLastMacroTiledLvl( + const ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn, ///< [in] Input structure + ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pOut ///< [in/out] Output structure (used as input, too) + ) const +{ + // pow2Pad covers all mipmap cases + if (pIn->flags.pow2Pad) + { + ADDR_ASSERT(IsMacroTiled(pIn->tileMode)); + + UINT_32 nextPitch; + UINT_32 nextHeight; + UINT_32 nextSlices; + + AddrTileMode nextTileMode; + + if (pIn->mipLevel == 0 || pIn->basePitch == 0) + { + // Base level or fail-safe case (basePitch == 0) + nextPitch = pOut->pitch >> 1; + } + else + { + // Sub levels + nextPitch = pIn->basePitch >> (pIn->mipLevel + 1); + } + + // nextHeight must be shifted from this level's original height rather than a pow2 padded + // one but this requires original height stored somewhere (pOut->height) + ADDR_ASSERT(pOut->height != 0); + + // next level's height is just current level's >> 1 in pixels + nextHeight = pOut->height >> 1; + // Special format such as FMT_1 and FMT_32_32_32 can be linear only so we consider block + // compressed foramts + if (AddrElemLib::IsBlockCompressed(pIn->format)) + { + nextHeight = (nextHeight + 3) / 4; + } + nextHeight = NextPow2(nextHeight); + + // nextSlices may be 0 if this level's is 1 + if (pIn->flags.volume) + { + nextSlices = Max(1u, pIn->numSlices >> 1); + } + else + { + nextSlices = pIn->numSlices; + } + + nextTileMode = ComputeSurfaceMipLevelTileMode(pIn->tileMode, + pIn->bpp, + nextPitch, + nextHeight, + nextSlices, + pIn->numSamples, + pOut->pitchAlign, + pOut->heightAlign, + pOut->pTileInfo); + + pOut->last2DLevel = IsMicroTiled(nextTileMode); + } +} + +/** +*************************************************************************************************** +* SIAddrLib::HwlDegradeThickTileMode +* +* @brief +* Degrades valid tile mode for thick modes if needed +* +* @return +* Suitable tile mode +*************************************************************************************************** +*/ +AddrTileMode SIAddrLib::HwlDegradeThickTileMode( + AddrTileMode baseTileMode, ///< [in] base tile mode + UINT_32 numSlices, ///< [in] current number of slices + UINT_32* pBytesPerTile ///< [in/out] pointer to bytes per slice + ) const +{ + return EgBasedAddrLib::HwlDegradeThickTileMode(baseTileMode, numSlices, pBytesPerTile); +} + +/** +*************************************************************************************************** +* SIAddrLib::HwlTileInfoEqual +* +* @brief +* Return TRUE if all field are equal +* @note +* Only takes care of current HWL's data +*************************************************************************************************** +*/ +BOOL_32 SIAddrLib::HwlTileInfoEqual( + const ADDR_TILEINFO* pLeft, ///<[in] Left compare operand + const ADDR_TILEINFO* pRight ///<[in] Right compare operand + ) const +{ + BOOL_32 equal = FALSE; + + if (pLeft->pipeConfig == pRight->pipeConfig) + { + equal = EgBasedAddrLib::HwlTileInfoEqual(pLeft, pRight); + } + + return equal; +} + +/** +*************************************************************************************************** +* SIAddrLib::GetTileSettings +* +* @brief +* Get tile setting infos by index. +* @return +* Tile setting info. +*************************************************************************************************** +*/ +const ADDR_TILECONFIG* SIAddrLib::GetTileSetting( + UINT_32 index ///< [in] Tile index + ) const +{ + ADDR_ASSERT(index < m_noOfEntries); + return &m_tileTable[index]; +} + +/** +*************************************************************************************************** +* SIAddrLib::HwlPostCheckTileIndex +* +* @brief +* Map a tile setting to index if curIndex is invalid, otherwise check if curIndex matches +* tile mode/type/info and change the index if needed +* @return +* Tile index. +*************************************************************************************************** +*/ +INT_32 SIAddrLib::HwlPostCheckTileIndex( + const ADDR_TILEINFO* pInfo, ///< [in] Tile Info + AddrTileMode mode, ///< [in] Tile mode + AddrTileType type, ///< [in] Tile type + INT curIndex ///< [in] Current index assigned in HwlSetupTileInfo + ) const +{ + INT_32 index = curIndex; + + if (mode == ADDR_TM_LINEAR_GENERAL) + { + index = TileIndexLinearGeneral; + } + else + { + BOOL_32 macroTiled = IsMacroTiled(mode); + + // We need to find a new index if either of them is true + // 1. curIndex is invalid + // 2. tile mode is changed + // 3. tile info does not match for macro tiled + if ((index == TileIndexInvalid || + (mode != m_tileTable[index].mode) || + (macroTiled && !HwlTileInfoEqual(pInfo, &m_tileTable[index].info)))) + { + for (index = 0; index < static_cast(m_noOfEntries); index++) + { + if (macroTiled) + { + // macro tile modes need all to match + if (HwlTileInfoEqual(pInfo, &m_tileTable[index].info) && + (mode == m_tileTable[index].mode) && + (type == m_tileTable[index].type)) + { + break; + } + } + else if (mode == ADDR_TM_LINEAR_ALIGNED) + { + // linear mode only needs tile mode to match + if (mode == m_tileTable[index].mode) + { + break; + } + } + else + { + // micro tile modes only need tile mode and tile type to match + if (mode == m_tileTable[index].mode && + type == m_tileTable[index].type) + { + break; + } + } + } + } + } + + ADDR_ASSERT(index < static_cast(m_noOfEntries)); + + if (index >= static_cast(m_noOfEntries)) + { + index = TileIndexInvalid; + } + + return index; +} + +/** +*************************************************************************************************** +* SIAddrLib::HwlSetupTileCfg +* +* @brief +* Map tile index to tile setting. +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE SIAddrLib::HwlSetupTileCfg( + INT_32 index, ///< [in] Tile index + INT_32 macroModeIndex, ///< [in] Index in macro tile mode table(CI) + ADDR_TILEINFO* pInfo, ///< [out] Tile Info + AddrTileMode* pMode, ///< [out] Tile mode + AddrTileType* pType ///< [out] Tile type + ) const +{ + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + // Global flag to control usage of tileIndex + if (UseTileIndex(index)) + { + if (index == TileIndexLinearGeneral) + { + if (pMode) + { + *pMode = ADDR_TM_LINEAR_GENERAL; + } + + if (pType) + { + *pType = ADDR_DISPLAYABLE; + } + + if (pInfo) + { + pInfo->banks = 2; + pInfo->bankWidth = 1; + pInfo->bankHeight = 1; + pInfo->macroAspectRatio = 1; + pInfo->tileSplitBytes = 64; + pInfo->pipeConfig = ADDR_PIPECFG_P2; + } + } + else if (static_cast(index) >= m_noOfEntries) + { + returnCode = ADDR_INVALIDPARAMS; + } + else + { + const ADDR_TILECONFIG* pCfgTable = GetTileSetting(index); + + if (pInfo) + { + *pInfo = pCfgTable->info; + } + else + { + if (IsMacroTiled(pCfgTable->mode)) + { + returnCode = ADDR_INVALIDPARAMS; + } + } + + if (pMode) + { + *pMode = pCfgTable->mode; + } + + if (pType) + { + *pType = pCfgTable->type; + } + } + } + + return returnCode; +} + +/** +*************************************************************************************************** +* SIAddrLib::ReadGbTileMode +* +* @brief +* Convert GB_TILE_MODE HW value to ADDR_TILE_CONFIG. +* @return +* NA. +*************************************************************************************************** +*/ +VOID SIAddrLib::ReadGbTileMode( + UINT_32 regValue, ///< [in] GB_TILE_MODE register + ADDR_TILECONFIG* pCfg ///< [out] output structure + ) const +{ + GB_TILE_MODE gbTileMode; + gbTileMode.val = regValue; + + pCfg->type = static_cast(gbTileMode.f.micro_tile_mode); + pCfg->info.bankHeight = 1 << gbTileMode.f.bank_height; + pCfg->info.bankWidth = 1 << gbTileMode.f.bank_width; + pCfg->info.banks = 1 << (gbTileMode.f.num_banks + 1); + pCfg->info.macroAspectRatio = 1 << gbTileMode.f.macro_tile_aspect; + pCfg->info.tileSplitBytes = 64 << gbTileMode.f.tile_split; + pCfg->info.pipeConfig = static_cast(gbTileMode.f.pipe_config + 1); + + UINT_32 regArrayMode = gbTileMode.f.array_mode; + + pCfg->mode = static_cast(regArrayMode); + + if (regArrayMode == 8) //ARRAY_2D_TILED_XTHICK + { + pCfg->mode = ADDR_TM_2D_TILED_XTHICK; + } + else if (regArrayMode >= 14) //ARRAY_3D_TILED_XTHICK + { + pCfg->mode = static_cast(pCfg->mode + 3); + } +} + +/** +*************************************************************************************************** +* SIAddrLib::InitTileSettingTable +* +* @brief +* Initialize the ADDR_TILE_CONFIG table. +* @return +* TRUE if tile table is correctly initialized +*************************************************************************************************** +*/ +BOOL_32 SIAddrLib::InitTileSettingTable( + const UINT_32* pCfg, ///< [in] Pointer to table of tile configs + UINT_32 noOfEntries ///< [in] Numbe of entries in the table above + ) +{ + BOOL_32 initOk = TRUE; + + ADDR_ASSERT(noOfEntries <= TileTableSize); + + memset(m_tileTable, 0, sizeof(m_tileTable)); + + if (noOfEntries != 0) + { + m_noOfEntries = noOfEntries; + } + else + { + m_noOfEntries = TileTableSize; + } + + if (pCfg) // From Client + { + for (UINT_32 i = 0; i < m_noOfEntries; i++) + { + ReadGbTileMode(*(pCfg + i), &m_tileTable[i]); + } + } + else + { + ADDR_ASSERT_ALWAYS(); + initOk = FALSE; + } + + if (initOk) + { + ADDR_ASSERT(m_tileTable[TILEINDEX_LINEAR_ALIGNED].mode == ADDR_TM_LINEAR_ALIGNED); + } + + return initOk; +} + +/** +*************************************************************************************************** +* SIAddrLib::HwlGetTileIndex +* +* @brief +* Return the virtual/real index for given mode/type/info +* @return +* ADDR_OK if successful. +*************************************************************************************************** +*/ +ADDR_E_RETURNCODE SIAddrLib::HwlGetTileIndex( + const ADDR_GET_TILEINDEX_INPUT* pIn, + ADDR_GET_TILEINDEX_OUTPUT* pOut) const +{ + ADDR_E_RETURNCODE returnCode = ADDR_OK; + + pOut->index = HwlPostCheckTileIndex(pIn->pTileInfo, pIn->tileMode, pIn->tileType); + + return returnCode; +} + +/** +*************************************************************************************************** +* SIAddrLib::HwlFmaskPreThunkSurfInfo +* +* @brief +* Some preparation before thunking a ComputeSurfaceInfo call for Fmask +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +VOID SIAddrLib::HwlFmaskPreThunkSurfInfo( + const ADDR_COMPUTE_FMASK_INFO_INPUT* pFmaskIn, ///< [in] Input of fmask info + const ADDR_COMPUTE_FMASK_INFO_OUTPUT* pFmaskOut, ///< [in] Output of fmask info + ADDR_COMPUTE_SURFACE_INFO_INPUT* pSurfIn, ///< [out] Input of thunked surface info + ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pSurfOut ///< [out] Output of thunked surface info + ) const +{ + pSurfIn->tileIndex = pFmaskIn->tileIndex; +} + +/** +*************************************************************************************************** +* SIAddrLib::HwlFmaskPostThunkSurfInfo +* +* @brief +* Copy hwl extra field after calling thunked ComputeSurfaceInfo +* @return +* ADDR_E_RETURNCODE +*************************************************************************************************** +*/ +VOID SIAddrLib::HwlFmaskPostThunkSurfInfo( + const ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pSurfOut, ///< [in] Output of surface info + ADDR_COMPUTE_FMASK_INFO_OUTPUT* pFmaskOut ///< [out] Output of fmask info + ) const +{ + pFmaskOut->macroModeIndex = TileIndexInvalid; + pFmaskOut->tileIndex = pSurfOut->tileIndex; +} + +/** +*************************************************************************************************** +* SIAddrLib::HwlComputeFmaskBits +* @brief +* Computes fmask bits +* @return +* Fmask bits +*************************************************************************************************** +*/ +UINT_32 SIAddrLib::HwlComputeFmaskBits( + const ADDR_COMPUTE_FMASK_INFO_INPUT* pIn, + UINT_32* pNumSamples + ) const +{ + UINT_32 numSamples = pIn->numSamples; + UINT_32 numFrags = GetNumFragments(numSamples, pIn->numFrags); + UINT_32 bpp; + + if (numFrags != numSamples) // EQAA + { + ADDR_ASSERT(numFrags <= 8); + + if (!pIn->resolved) + { + if (numFrags == 1) + { + bpp = 1; + numSamples = numSamples == 16 ? 16 : 8; + } + else if (numFrags == 2) + { + ADDR_ASSERT(numSamples >= 4); + + bpp = 2; + numSamples = numSamples; + } + else if (numFrags == 4) + { + ADDR_ASSERT(numSamples >= 4); + + bpp = 4; + numSamples = numSamples; + } + else // numFrags == 8 + { + ADDR_ASSERT(numSamples == 16); + + bpp = 4; + numSamples = numSamples; + } + } + else + { + if (numFrags == 1) + { + bpp = (numSamples == 16) ? 16 : 8; + numSamples = 1; + } + else if (numFrags == 2) + { + ADDR_ASSERT(numSamples >= 4); + + bpp = numSamples*2; + numSamples = 1; + } + else if (numFrags == 4) + { + ADDR_ASSERT(numSamples >= 4); + + bpp = numSamples*4; + numSamples = 1; + } + else // numFrags == 8 + { + ADDR_ASSERT(numSamples >= 16); + + bpp = 16*4; + numSamples = 1; + } + } + } + else // Normal AA + { + if (!pIn->resolved) + { + bpp = ComputeFmaskNumPlanesFromNumSamples(numSamples); + numSamples = numSamples == 2 ? 8 : numSamples; + } + else + { + // The same as 8XX + bpp = ComputeFmaskResolvedBppFromNumSamples(numSamples); + numSamples = 1; // 1x sample + } + } + + SafeAssign(pNumSamples, numSamples); + + return bpp; +} + +/** +*************************************************************************************************** +* SIAddrLib::HwlOverrideTileMode +* +* @brief +* Override tile modes (for PRT only, avoid client passes in an invalid PRT mode for SI. +* +* @return +* Suitable tile mode +* +*************************************************************************************************** +*/ +BOOL_32 SIAddrLib::HwlOverrideTileMode( + const ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn, ///< [in] input structure + AddrTileMode* pTileMode, ///< [in/out] pointer to the tile mode + AddrTileType* pTileType ///< [in/out] pointer to the tile type + ) const +{ + BOOL_32 bOverrided = FALSE; + AddrTileMode tileMode = *pTileMode; + + switch (tileMode) + { + case ADDR_TM_PRT_TILED_THIN1: + tileMode = ADDR_TM_2D_TILED_THIN1; + break; + + case ADDR_TM_PRT_TILED_THICK: + tileMode = ADDR_TM_2D_TILED_THICK; + break; + + case ADDR_TM_PRT_2D_TILED_THICK: + tileMode = ADDR_TM_2D_TILED_THICK; + break; + + case ADDR_TM_PRT_3D_TILED_THICK: + tileMode = ADDR_TM_3D_TILED_THICK; + break; + + default: + break; + } + + if (tileMode != *pTileMode) + { + *pTileMode = tileMode; + bOverrided = TRUE; + ADDR_ASSERT(pIn->flags.prt == TRUE); + } + + return bOverrided; +} + diff --git a/src/gallium/winsys/radeon/amdgpu/addrlib/r800/siaddrlib.h b/src/gallium/winsys/radeon/amdgpu/addrlib/r800/siaddrlib.h new file mode 100644 index 0000000..897beb1 --- /dev/null +++ b/src/gallium/winsys/radeon/amdgpu/addrlib/r800/siaddrlib.h @@ -0,0 +1,262 @@ +/* + * Copyright © 2014 Advanced Micro Devices, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES + * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NON-INFRINGEMENT. IN NO EVENT SHALL THE COPYRIGHT HOLDERS, AUTHORS + * AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + */ + +/** +*************************************************************************************************** +* @file siaddrlib.h +* @brief Contains the R800AddrLib class definition. +*************************************************************************************************** +*/ + +#ifndef __SI_ADDR_LIB_H__ +#define __SI_ADDR_LIB_H__ + +#include "addrlib.h" +#include "egbaddrlib.h" + +/** +*************************************************************************************************** +* @brief Describes the information in tile mode table +*************************************************************************************************** +*/ +struct ADDR_TILECONFIG +{ + AddrTileMode mode; + AddrTileType type; + ADDR_TILEINFO info; +}; + +/** +*************************************************************************************************** +* @brief SI specific settings structure. +*************************************************************************************************** +*/ +struct SIChipSettings +{ + struct + { + UINT_32 isSouthernIsland : 1; + UINT_32 isTahiti : 1; + UINT_32 isPitCairn : 1; + UINT_32 isCapeVerde : 1; + /// Oland/Hainan are of GFXIP 6.0, similar with SI + UINT_32 isOland : 1; + UINT_32 isHainan : 1; + }; +}; + +/** +*************************************************************************************************** +* @brief This class is the SI specific address library +* function set. +*************************************************************************************************** +*/ +class SIAddrLib : public EgBasedAddrLib +{ +public: + /// Creates SIAddrLib object + static AddrLib* CreateObj(const AddrClient* pClient) + { + return new(pClient) SIAddrLib(pClient); + } + +protected: + SIAddrLib(const AddrClient* pClient); + virtual ~SIAddrLib(); + + // Hwl interface - defined in AddrLib + virtual ADDR_E_RETURNCODE HwlComputeSurfaceInfo( + const ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn, + ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pOut) const; + + virtual ADDR_E_RETURNCODE HwlConvertTileInfoToHW( + const ADDR_CONVERT_TILEINFOTOHW_INPUT* pIn, + ADDR_CONVERT_TILEINFOTOHW_OUTPUT* pOut) const; + + virtual UINT_64 HwlComputeXmaskAddrFromCoord( + UINT_32 pitch, UINT_32 height, UINT_32 x, UINT_32 y, UINT_32 slice, UINT_32 numSlices, + UINT_32 factor, BOOL_32 isLinear, BOOL_32 isWidth8, BOOL_32 isHeight8, + ADDR_TILEINFO* pTileInfo, UINT_32* pBitPosition) const; + + virtual VOID HwlComputeXmaskCoordFromAddr( + UINT_64 addr, UINT_32 bitPosition, UINT_32 pitch, UINT_32 height, UINT_32 numSlices, + UINT_32 factor, BOOL_32 isLinear, BOOL_32 isWidth8, BOOL_32 isHeight8, + ADDR_TILEINFO* pTileInfo, UINT_32* pX, UINT_32* pY, UINT_32* pSlice) const; + + virtual ADDR_E_RETURNCODE HwlGetTileIndex( + const ADDR_GET_TILEINDEX_INPUT* pIn, + ADDR_GET_TILEINDEX_OUTPUT* pOut) const; + + virtual BOOL_32 HwlComputeMipLevel( + ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn) const; + + virtual AddrChipFamily HwlConvertChipFamily( + UINT_32 uChipFamily, UINT_32 uChipRevision); + + virtual BOOL_32 HwlInitGlobalParams( + const ADDR_CREATE_INPUT* pCreateIn); + + virtual ADDR_E_RETURNCODE HwlSetupTileCfg( + INT_32 index, INT_32 macroModeIndex, + ADDR_TILEINFO* pInfo, AddrTileMode* pMode = 0, AddrTileType* pType = 0) const; + + virtual VOID HwlComputeTileDataWidthAndHeightLinear( + UINT_32* pMacroWidth, UINT_32* pMacroHeight, + UINT_32 bpp, ADDR_TILEINFO* pTileInfo) const; + + virtual UINT_64 HwlComputeHtileBytes( + UINT_32 pitch, UINT_32 height, UINT_32 bpp, + BOOL_32 isLinear, UINT_32 numSlices, UINT_64* pSliceBytes, UINT_32 baseAlign) const; + + virtual UINT_32 ComputePipeFromCoord( + UINT_32 x, UINT_32 y, UINT_32 slice, + AddrTileMode tileMode, UINT_32 pipeSwizzle, BOOL_32 ignoreSE, + ADDR_TILEINFO* pTileInfo) const; + + virtual UINT_32 HwlGetPipes(const ADDR_TILEINFO* pTileInfo) const; + + /// Pre-handler of 3x pitch (96 bit) adjustment + virtual UINT_32 HwlPreHandleBaseLvl3xPitch( + const ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn, UINT_32 expPitch) const; + /// Post-handler of 3x pitch adjustment + virtual UINT_32 HwlPostHandleBaseLvl3xPitch( + const ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn, UINT_32 expPitch) const; + + /// Dummy function to finalize the inheritance + virtual UINT_32 HwlComputeXmaskCoordYFrom8Pipe( + UINT_32 pipe, UINT_32 x) const; + + // Sub-hwl interface - defined in EgBasedAddrLib + virtual VOID HwlSetupTileInfo( + AddrTileMode tileMode, ADDR_SURFACE_FLAGS flags, + UINT_32 bpp, UINT_32 pitch, UINT_32 height, UINT_32 numSamples, + ADDR_TILEINFO* inputTileInfo, ADDR_TILEINFO* outputTileInfo, + AddrTileType inTileType, ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pOut) const; + + virtual UINT_32 HwlGetPitchAlignmentMicroTiled( + AddrTileMode tileMode, UINT_32 bpp, ADDR_SURFACE_FLAGS flags, UINT_32 numSamples) const; + + virtual UINT_64 HwlGetSizeAdjustmentMicroTiled( + UINT_32 thickness, UINT_32 bpp, ADDR_SURFACE_FLAGS flags, UINT_32 numSamples, + UINT_32 baseAlign, UINT_32 pitchAlign, + UINT_32 *pPitch, UINT_32 *pHeight) const; + + virtual VOID HwlCheckLastMacroTiledLvl( + const ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn, ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pOut) const; + + virtual BOOL_32 HwlTileInfoEqual( + const ADDR_TILEINFO* pLeft, const ADDR_TILEINFO* pRight) const; + + virtual AddrTileMode HwlDegradeThickTileMode( + AddrTileMode baseTileMode, UINT_32 numSlices, UINT_32* pBytesPerTile) const; + + virtual BOOL_32 HwlOverrideTileMode( + const ADDR_COMPUTE_SURFACE_INFO_INPUT* pIn, + AddrTileMode* pTileMode, + AddrTileType* pTileType) const; + + virtual BOOL_32 HwlSanityCheckMacroTiled( + ADDR_TILEINFO* pTileInfo) const + { + return TRUE; + } + + virtual UINT_32 HwlGetPitchAlignmentLinear(UINT_32 bpp, ADDR_SURFACE_FLAGS flags) const; + + virtual UINT_64 HwlGetSizeAdjustmentLinear( + AddrTileMode tileMode, + UINT_32 bpp, UINT_32 numSamples, UINT_32 baseAlign, UINT_32 pitchAlign, + UINT_32 *pPitch, UINT_32 *pHeight, UINT_32 *pHeightAlign) const; + + virtual VOID HwlComputeSurfaceCoord2DFromBankPipe( + AddrTileMode tileMode, UINT_32* pX, UINT_32* pY, UINT_32 slice, + UINT_32 bank, UINT_32 pipe, + UINT_32 bankSwizzle, UINT_32 pipeSwizzle, UINT_32 tileSlices, + BOOL_32 ignoreSE, + ADDR_TILEINFO* pTileInfo) const; + + virtual UINT_32 HwlPreAdjustBank( + UINT_32 tileX, UINT_32 bank, ADDR_TILEINFO* pTileInfo) const; + + virtual INT_32 HwlPostCheckTileIndex( + const ADDR_TILEINFO* pInfo, AddrTileMode mode, AddrTileType type, + INT curIndex = TileIndexInvalid) const; + + virtual VOID HwlFmaskPreThunkSurfInfo( + const ADDR_COMPUTE_FMASK_INFO_INPUT* pFmaskIn, + const ADDR_COMPUTE_FMASK_INFO_OUTPUT* pFmaskOut, + ADDR_COMPUTE_SURFACE_INFO_INPUT* pSurfIn, + ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pSurfOut) const; + + virtual VOID HwlFmaskPostThunkSurfInfo( + const ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pSurfOut, + ADDR_COMPUTE_FMASK_INFO_OUTPUT* pFmaskOut) const; + + virtual UINT_32 HwlComputeFmaskBits( + const ADDR_COMPUTE_FMASK_INFO_INPUT* pIn, + UINT_32* pNumSamples) const; + + virtual BOOL_32 HwlReduceBankWidthHeight( + UINT_32 tileSize, UINT_32 bpp, ADDR_SURFACE_FLAGS flags, UINT_32 numSamples, + UINT_32 bankHeightAlign, UINT_32 pipes, + ADDR_TILEINFO* pTileInfo) const + { + return TRUE; + } + + // Protected non-virtual functions + VOID ComputeTileCoordFromPipeAndElemIdx( + UINT_32 elemIdx, UINT_32 pipe, AddrPipeCfg pipeCfg, UINT_32 pitchInMacroTile, + UINT_32 x, UINT_32 y, UINT_32* pX, UINT_32* pY) const; + + UINT_32 TileCoordToMaskElementIndex( + UINT_32 tx, UINT_32 ty, AddrPipeCfg pipeConfig, + UINT_32 *macroShift, UINT_32 *elemIdxBits) const; + + BOOL_32 DecodeGbRegs( + const ADDR_REGISTER_VALUE* pRegValue); + + const ADDR_TILECONFIG* GetTileSetting( + UINT_32 index) const; + + static const UINT_32 TileTableSize = 32; + ADDR_TILECONFIG m_tileTable[TileTableSize]; + UINT_32 m_noOfEntries; + +private: + + UINT_32 GetPipePerSurf(AddrPipeCfg pipeConfig) const; + + VOID ReadGbTileMode( + UINT_32 regValue, ADDR_TILECONFIG* pCfg) const; + BOOL_32 InitTileSettingTable( + const UINT_32 *pSetting, UINT_32 noOfEntries); + + SIChipSettings m_settings; +}; + +#endif + diff --git a/src/gallium/winsys/radeon/amdgpu/amdgpu_id.h b/src/gallium/winsys/radeon/amdgpu/amdgpu_id.h new file mode 100644 index 0000000..08a1591 --- /dev/null +++ b/src/gallium/winsys/radeon/amdgpu/amdgpu_id.h @@ -0,0 +1,157 @@ +/* + * Copyright © 2014 Advanced Micro Devices, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES + * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NON-INFRINGEMENT. IN NO EVENT SHALL THE COPYRIGHT HOLDERS, AUTHORS + * AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + */ + +/** + * This file is included by addrlib. It adds GPU family definitions and + * macros compatible with addrlib. + */ + +#ifndef AMDGPU_ID_H +#define AMDGPU_ID_H + +#include "pipe/p_config.h" + +#if defined(PIPE_ARCH_LITTLE_ENDIAN) +#define LITTLEENDIAN_CPU +#elif defined(PIPE_ARCH_BIG_ENDIAN) +#define BIGENDIAN_CPU +#endif + +enum { + FAMILY_UNKNOWN, + FAMILY_SI, + FAMILY_CI, + FAMILY_KV, + FAMILY_VI, + FAMILY_CZ, + FAMILY_PI, + FAMILY_LAST, +}; + +/* SI specific rev IDs */ +enum { + SI_TAHITI_P_A11 = 1, + SI_TAHITI_P_A0 = SI_TAHITI_P_A11, /*A0 is alias of A11*/ + SI_TAHITI_P_A21 = 5, + SI_TAHITI_P_B0 = SI_TAHITI_P_A21, /*B0 is alias of A21*/ + SI_TAHITI_P_A22 = 6, + SI_TAHITI_P_B1 = SI_TAHITI_P_A22, /*B1 is alias of A22*/ + + SI_PITCAIRN_PM_A11 = 20, + SI_PITCAIRN_PM_A0 = SI_PITCAIRN_PM_A11, /*A0 is alias of A11*/ + SI_PITCAIRN_PM_A12 = 21, + SI_PITCAIRN_PM_A1 = SI_PITCAIRN_PM_A12, /*A1 is alias of A12*/ + + SI_CAPEVERDE_M_A11 = 40, + SI_CAPEVERDE_M_A0 = SI_CAPEVERDE_M_A11, /*A0 is alias of A11*/ + SI_CAPEVERDE_M_A12 = 41, + SI_CAPEVERDE_M_A1 = SI_CAPEVERDE_M_A12, /*A1 is alias of A12*/ + + SI_OLAND_M_A0 = 60, + + SI_HAINAN_V_A0 = 70, + + SI_UNKNOWN = 0xFF +}; + + +#define ASICREV_IS_TAHITI_P(eChipRev) \ + (eChipRev < SI_PITCAIRN_PM_A11) +#define ASICREV_IS_PITCAIRN_PM(eChipRev) \ + ((eChipRev >= SI_PITCAIRN_PM_A11) && (eChipRev < SI_CAPEVERDE_M_A11)) +#define ASICREV_IS_CAPEVERDE_M(eChipRev) \ + ((eChipRev >= SI_CAPEVERDE_M_A11) && (eChipRev < SI_OLAND_M_A0)) +#define ASICREV_IS_OLAND_M(eChipRev) \ + ((eChipRev >= SI_OLAND_M_A0) && (eChipRev < SI_HAINAN_V_A0)) +#define ASICREV_IS_HAINAN_V(eChipRev) \ +(eChipRev >= SI_HAINAN_V_A0) + +/* CI specific revIDs */ +enum { + CI_BONAIRE_M_A0 = 20, + CI_BONAIRE_M_A1 = 21, + + CI_HAWAII_P_A0 = 40, + + CI_UNKNOWN = 0xFF +}; + +#define ASICREV_IS_BONAIRE_M(eChipRev) \ + ((eChipRev >= CI_BONAIRE_M_A0) && (eChipRev < CI_HAWAII_P_A0)) +#define ASICREV_IS_HAWAII_P(eChipRev) \ + (eChipRev >= CI_HAWAII_P_A0) + +/* KV specific rev IDs */ +enum { + KV_SPECTRE_A0 = 0x01, /* KV1 with Spectre GFX core, 8-8-1-2 (CU-Pix-Primitive-RB) */ + KV_SPOOKY_A0 = 0x41, /* KV2 with Spooky GFX core, including downgraded from Spectre core, 3-4-1-1 (CU-Pix-Primitive-RB) */ + KB_KALINDI_A0 = 0x81, /* KB with Kalindi GFX core, 2-4-1-1 (CU-Pix-Primitive-RB) */ + KB_KALINDI_A1 = 0x82, /* KB with Kalindi GFX core, 2-4-1-1 (CU-Pix-Primitive-RB) */ + BV_KALINDI_A2 = 0x85, /* BV with Kalindi GFX core, 2-4-1-1 (CU-Pix-Primitive-RB) */ + ML_GODAVARI_A0 = 0xa1, /* ML with Godavari GFX core, 2-4-1-1 (CU-Pix-Primitive-RB) */ + ML_GODAVARI_A1 = 0xa2, /* ML with Godavari GFX core, 2-4-1-1 (CU-Pix-Primitive-RB) */ + KV_UNKNOWN = 0xFF +}; + +#define ASICREV_IS_SPECTRE(eChipRev) \ + ((eChipRev >= KV_SPECTRE_A0) && (eChipRev < KV_SPOOKY_A0)) /* identify all versions of SPRECTRE and supported features set */ +#define ASICREV_IS_SPOOKY(eChipRev) \ + ((eChipRev >= KV_SPOOKY_A0) && (eChipRev < KB_KALINDI_A0)) /* identify all versions of SPOOKY and supported features set */ +#define ASICREV_IS_KALINDI(eChipRev) \ + ((eChipRev >= KB_KALINDI_A0) && (eChipRev < KV_UNKNOWN)) /* identify all versions of KALINDI and supported features set */ + +/* Following macros are subset of ASICREV_IS_KALINDI macro */ +#define ASICREV_IS_KALINDI_BHAVANI(eChipRev) \ + ((eChipRev >= BV_KALINDI_A2) && (eChipRev < ML_GODAVARI_A0)) /* identify all versions of BHAVANI and supported features set */ +#define ASICREV_IS_KALINDI_GODAVARI(eChipRev) \ + ((eChipRev >= ML_GODAVARI_A0) && (eChipRev < KV_UNKNOWN)) /* identify all versions of GODAVARI and supported features set */ + +/* VI specific rev IDs */ +enum { + VI_ICELAND_M_A0 = 1, + + VI_TONGA_P_A0 = 20, + VI_TONGA_P_A1 = 21, + + VI_UNKNOWN = 0xFF +}; + + +#define ASICREV_IS_ICELAND_M(eChipRev) \ + (eChipRev < VI_TONGA_P_A0) +#define ASICREV_IS_TONGA_P(eChipRev) \ + (eChipRev >= VI_TONGA_P_A0) + +/* CZ specific rev IDs */ +enum { + CZ_CARRIZO_A0 = 0x01, + CZ_UNKNOWN = 0xFF +}; + +#define ASICREV_IS_CARRIZO(eChipRev) \ + (eChipRev >= CARRIZO_A0) + +#endif /* AMDGPU_ID_H */ diff --git a/src/gallium/winsys/radeon/amdgpu/amdgpu_surface.c b/src/gallium/winsys/radeon/amdgpu/amdgpu_surface.c new file mode 100644 index 0000000..aa32413 --- /dev/null +++ b/src/gallium/winsys/radeon/amdgpu/amdgpu_surface.c @@ -0,0 +1,436 @@ +/* + * Copyright © 2011 Red Hat All Rights Reserved. + * Copyright © 2014 Advanced Micro Devices, Inc. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining + * a copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES + * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND + * NON-INFRINGEMENT. IN NO EVENT SHALL THE COPYRIGHT HOLDERS, AUTHORS + * AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + */ + +/* Contact: + * Marek Olšák + */ + +#include "amdgpu_winsys.h" + +#ifndef NO_ENTRIES +#define NO_ENTRIES 32 +#endif + +#ifndef NO_MACRO_ENTRIES +#define NO_MACRO_ENTRIES 16 +#endif + +#ifndef CIASICIDGFXENGINE_SOUTHERNISLAND +#define CIASICIDGFXENGINE_SOUTHERNISLAND 0x0000000A +#endif + + +static int amdgpu_surface_sanity(const struct radeon_surf *surf) +{ + unsigned type = RADEON_SURF_GET(surf->flags, TYPE); + + if (!(surf->flags & RADEON_SURF_HAS_TILE_MODE_INDEX)) + return -EINVAL; + + /* all dimension must be at least 1 ! */ + if (!surf->npix_x || !surf->npix_y || !surf->npix_z || + !surf->array_size) + return -EINVAL; + + if (!surf->blk_w || !surf->blk_h || !surf->blk_d) + return -EINVAL; + + switch (surf->nsamples) { + case 1: + case 2: + case 4: + case 8: + break; + default: + return -EINVAL; + } + + switch (type) { + case RADEON_SURF_TYPE_1D: + if (surf->npix_y > 1) + return -EINVAL; + /* fall through */ + case RADEON_SURF_TYPE_2D: + case RADEON_SURF_TYPE_CUBEMAP: + if (surf->npix_z > 1 || surf->array_size > 1) + return -EINVAL; + break; + case RADEON_SURF_TYPE_3D: + if (surf->array_size > 1) + return -EINVAL; + break; + case RADEON_SURF_TYPE_1D_ARRAY: + if (surf->npix_y > 1) + return -EINVAL; + /* fall through */ + case RADEON_SURF_TYPE_2D_ARRAY: + if (surf->npix_z > 1) + return -EINVAL; + break; + default: + return -EINVAL; + } + return 0; +} + +static void *ADDR_API allocSysMem(const ADDR_ALLOCSYSMEM_INPUT * pInput) +{ + return malloc(pInput->sizeInBytes); +} + +static ADDR_E_RETURNCODE ADDR_API freeSysMem(const ADDR_FREESYSMEM_INPUT * pInput) +{ + free(pInput->pVirtAddr); + return ADDR_OK; +} + +/** + * This returns the number of banks for the surface. + * Possible values: 2, 4, 8, 16. + */ +static uint32_t cik_num_banks(struct amdgpu_winsys *ws, + struct radeon_surf *surf) +{ + unsigned index, tileb; + + tileb = 8 * 8 * surf->bpe; + tileb = MIN2(surf->tile_split, tileb); + + for (index = 0; tileb > 64; index++) { + tileb >>= 1; + } + assert(index < 16); + + return 2 << ((ws->amdinfo.gb_macro_tile_mode[index] >> 6) & 0x3); +} + +ADDR_HANDLE amdgpu_addr_create(struct amdgpu_winsys *ws) +{ + ADDR_CREATE_INPUT addrCreateInput = {0}; + ADDR_CREATE_OUTPUT addrCreateOutput = {0}; + ADDR_REGISTER_VALUE regValue = {0}; + ADDR_CREATE_FLAGS createFlags = {{0}}; + ADDR_E_RETURNCODE addrRet; + + addrCreateInput.size = sizeof(ADDR_CREATE_INPUT); + addrCreateOutput.size = sizeof(ADDR_CREATE_OUTPUT); + + regValue.noOfBanks = ws->amdinfo.mc_arb_ramcfg & 0x3; + regValue.gbAddrConfig = ws->amdinfo.gb_addr_cfg; + regValue.noOfRanks = (ws->amdinfo.mc_arb_ramcfg & 0x4) >> 2; + + regValue.backendDisables = ws->amdinfo.backend_disable[0]; + regValue.pTileConfig = ws->amdinfo.gb_tile_mode; + regValue.noOfEntries = sizeof(ws->amdinfo.gb_tile_mode) / + sizeof(ws->amdinfo.gb_tile_mode[0]); + regValue.pMacroTileConfig = ws->amdinfo.gb_macro_tile_mode; + regValue.noOfMacroEntries = sizeof(ws->amdinfo.gb_macro_tile_mode) / + sizeof(ws->amdinfo.gb_macro_tile_mode[0]); + + createFlags.value = 0; + createFlags.useTileIndex = 1; + createFlags.degradeBaseLevel = 1; + + addrCreateInput.chipEngine = CIASICIDGFXENGINE_SOUTHERNISLAND; + addrCreateInput.chipFamily = ws->family; + addrCreateInput.chipRevision = ws->rev_id; + addrCreateInput.createFlags = createFlags; + addrCreateInput.callbacks.allocSysMem = allocSysMem; + addrCreateInput.callbacks.freeSysMem = freeSysMem; + addrCreateInput.callbacks.debugPrint = 0; + addrCreateInput.regValue = regValue; + + addrRet = AddrCreate(&addrCreateInput, &addrCreateOutput); + if (addrRet != ADDR_OK) + return NULL; + + return addrCreateOutput.hLib; +} + +static int compute_level(struct amdgpu_winsys *ws, + struct radeon_surf *surf, bool is_stencil, + unsigned level, unsigned type, bool compressed, + ADDR_COMPUTE_SURFACE_INFO_INPUT *AddrSurfInfoIn, + ADDR_COMPUTE_SURFACE_INFO_OUTPUT *AddrSurfInfoOut) +{ + struct radeon_surf_level *surf_level; + ADDR_E_RETURNCODE ret; + + AddrSurfInfoIn->mipLevel = level; + AddrSurfInfoIn->width = u_minify(surf->npix_x, level); + AddrSurfInfoIn->height = u_minify(surf->npix_y, level); + + if (type == RADEON_SURF_TYPE_3D) + AddrSurfInfoIn->numSlices = u_minify(surf->npix_z, level); + else if (type == RADEON_SURF_TYPE_CUBEMAP) + AddrSurfInfoIn->numSlices = 6; + else + AddrSurfInfoIn->numSlices = surf->array_size; + + if (level > 0) { + /* Set the base level pitch. This is needed for calculation + * of non-zero levels. */ + if (is_stencil) + AddrSurfInfoIn->basePitch = surf->stencil_level[0].nblk_x; + else + AddrSurfInfoIn->basePitch = surf->level[0].nblk_x; + + /* Convert blocks to pixels for compressed formats. */ + if (compressed) + AddrSurfInfoIn->basePitch *= surf->blk_w; + } + + ret = AddrComputeSurfaceInfo(ws->addrlib, + AddrSurfInfoIn, + AddrSurfInfoOut); + if (ret != ADDR_OK) { + return ret; + } + + surf_level = is_stencil ? &surf->stencil_level[level] : &surf->level[level]; + surf_level->offset = align(surf->bo_size, AddrSurfInfoOut->baseAlign); + surf_level->slice_size = AddrSurfInfoOut->sliceSize; + surf_level->pitch_bytes = AddrSurfInfoOut->pitch * (is_stencil ? 1 : surf->bpe); + surf_level->npix_x = u_minify(surf->npix_x, level); + surf_level->npix_y = u_minify(surf->npix_y, level); + surf_level->npix_z = u_minify(surf->npix_z, level); + surf_level->nblk_x = AddrSurfInfoOut->pitch; + surf_level->nblk_y = AddrSurfInfoOut->height; + if (type == RADEON_SURF_TYPE_3D) + surf_level->nblk_z = AddrSurfInfoOut->depth; + else + surf_level->nblk_z = 1; + + switch (AddrSurfInfoOut->tileMode) { + case ADDR_TM_LINEAR_GENERAL: + surf_level->mode = RADEON_SURF_MODE_LINEAR; + break; + case ADDR_TM_LINEAR_ALIGNED: + surf_level->mode = RADEON_SURF_MODE_LINEAR_ALIGNED; + break; + case ADDR_TM_1D_TILED_THIN1: + surf_level->mode = RADEON_SURF_MODE_1D; + break; + case ADDR_TM_2D_TILED_THIN1: + surf_level->mode = RADEON_SURF_MODE_2D; + break; + default: + assert(0); + } + + if (is_stencil) + surf->stencil_tiling_index[level] = AddrSurfInfoOut->tileIndex; + else + surf->tiling_index[level] = AddrSurfInfoOut->tileIndex; + + surf->bo_size = surf_level->offset + AddrSurfInfoOut->surfSize; + return 0; +} + +static int amdgpu_surface_init(struct radeon_winsys *rws, + struct radeon_surf *surf) +{ + struct amdgpu_winsys *ws = (struct amdgpu_winsys*)rws; + unsigned level, mode, type; + bool compressed; + ADDR_COMPUTE_SURFACE_INFO_INPUT AddrSurfInfoIn = {0}; + ADDR_COMPUTE_SURFACE_INFO_OUTPUT AddrSurfInfoOut = {0}; + ADDR_TILEINFO AddrTileInfoIn = {0}; + ADDR_TILEINFO AddrTileInfoOut = {0}; + int r; + + r = amdgpu_surface_sanity(surf); + if (r) + return r; + + AddrSurfInfoIn.size = sizeof(ADDR_COMPUTE_SURFACE_INFO_INPUT); + AddrSurfInfoOut.size = sizeof(ADDR_COMPUTE_SURFACE_INFO_OUTPUT); + AddrSurfInfoOut.pTileInfo = &AddrTileInfoOut; + + type = RADEON_SURF_GET(surf->flags, TYPE); + mode = RADEON_SURF_GET(surf->flags, MODE); + compressed = surf->blk_w == 4 && surf->blk_h == 4; + + /* MSAA and FMASK require 2D tiling. */ + if (surf->nsamples > 1 || + (surf->flags & RADEON_SURF_FMASK)) + mode = RADEON_SURF_MODE_2D; + + /* DB doesn't support linear layouts. */ + if (surf->flags & (RADEON_SURF_Z_OR_SBUFFER) && + mode < RADEON_SURF_MODE_1D) + mode = RADEON_SURF_MODE_1D; + + /* Set the requested tiling mode. */ + switch (mode) { + case RADEON_SURF_MODE_LINEAR: + AddrSurfInfoIn.tileMode = ADDR_TM_LINEAR_GENERAL; + break; + case RADEON_SURF_MODE_LINEAR_ALIGNED: + AddrSurfInfoIn.tileMode = ADDR_TM_LINEAR_ALIGNED; + break; + case RADEON_SURF_MODE_1D: + AddrSurfInfoIn.tileMode = ADDR_TM_1D_TILED_THIN1; + break; + case RADEON_SURF_MODE_2D: + AddrSurfInfoIn.tileMode = ADDR_TM_2D_TILED_THIN1; + break; + default: + assert(0); + } + + /* The format must be set correctly for the allocation of compressed + * textures to work. In other cases, setting the bpp is sufficient. */ + if (compressed) { + switch (surf->bpe) { + case 8: + AddrSurfInfoIn.format = ADDR_FMT_BC1; + break; + case 16: + AddrSurfInfoIn.format = ADDR_FMT_BC3; + break; + default: + assert(0); + } + } + else { + AddrSurfInfoIn.bpp = surf->bpe * 8; + } + + AddrSurfInfoIn.numSamples = surf->nsamples; + AddrSurfInfoIn.tileIndex = -1; + + /* Set the micro tile type. */ + if (surf->flags & RADEON_SURF_SCANOUT) + AddrSurfInfoIn.tileType = ADDR_DISPLAYABLE; + else if (surf->flags & RADEON_SURF_Z_OR_SBUFFER) + AddrSurfInfoIn.tileType = ADDR_DEPTH_SAMPLE_ORDER; + else + AddrSurfInfoIn.tileType = ADDR_NON_DISPLAYABLE; + + AddrSurfInfoIn.flags.color = !(surf->flags & RADEON_SURF_Z_OR_SBUFFER); + AddrSurfInfoIn.flags.depth = (surf->flags & RADEON_SURF_ZBUFFER) != 0; + AddrSurfInfoIn.flags.stencil = (surf->flags & RADEON_SURF_SBUFFER) != 0; + AddrSurfInfoIn.flags.cube = type == RADEON_SURF_TYPE_CUBEMAP; + AddrSurfInfoIn.flags.display = (surf->flags & RADEON_SURF_SCANOUT) != 0; + AddrSurfInfoIn.flags.pow2Pad = surf->last_level > 0; + AddrSurfInfoIn.flags.degrade4Space = 1; + + /* This disables incorrect calculations (hacks) in addrlib. */ + AddrSurfInfoIn.flags.noStencil = 1; + + /* Set preferred macrotile parameters. This is usually required + * for shared resources. This is for 2D tiling only. */ + if (AddrSurfInfoIn.tileMode >= ADDR_TM_2D_TILED_THIN1 && + surf->bankw && surf->bankh && surf->mtilea && surf->tile_split) { + /* If any of these parameters are incorrect, the calculation + * will fail. */ + AddrTileInfoIn.banks = cik_num_banks(ws, surf); + AddrTileInfoIn.bankWidth = surf->bankw; + AddrTileInfoIn.bankHeight = surf->bankh; + AddrTileInfoIn.macroAspectRatio = surf->mtilea; + AddrTileInfoIn.tileSplitBytes = surf->tile_split; + AddrSurfInfoIn.flags.degrade4Space = 0; + AddrSurfInfoIn.pTileInfo = &AddrTileInfoIn; + + /* If AddrSurfInfoIn.pTileInfo is set, Addrlib doesn't set + * the tile index, because we are expected to know it if + * we know the other parameters. + * + * This is something that can easily be fixed in Addrlib. + * For now, just figure it out here. + * Note that only 2D_TILE_THIN1 is handled here. + */ + assert(!(surf->flags & RADEON_SURF_Z_OR_SBUFFER)); + assert(AddrSurfInfoIn.tileMode == ADDR_TM_2D_TILED_THIN1); + + if (AddrSurfInfoIn.tileType == ADDR_DISPLAYABLE) + AddrSurfInfoIn.tileIndex = 10; /* 2D displayable */ + else + AddrSurfInfoIn.tileIndex = 14; /* 2D non-displayable */ + } + + surf->bo_size = 0; + + /* Calculate texture layout information. */ + for (level = 0; level <= surf->last_level; level++) { + r = compute_level(ws, surf, false, level, type, compressed, + &AddrSurfInfoIn, &AddrSurfInfoOut); + if (r) + return r; + + if (level == 0) { + surf->bo_alignment = AddrSurfInfoOut.baseAlign; + + /* For 2D modes only. */ + if (AddrSurfInfoOut.tileMode >= ADDR_TM_2D_TILED_THIN1) { + surf->bankw = AddrSurfInfoOut.pTileInfo->bankWidth; + surf->bankh = AddrSurfInfoOut.pTileInfo->bankHeight; + surf->mtilea = AddrSurfInfoOut.pTileInfo->macroAspectRatio; + surf->tile_split = AddrSurfInfoOut.pTileInfo->tileSplitBytes; + } + } + } + + /* Calculate texture layout information for stencil. */ + if (surf->flags & RADEON_SURF_SBUFFER) { + AddrSurfInfoIn.bpp = 8; + /* This will be ignored if AddrSurfInfoIn.pTileInfo is NULL. */ + AddrTileInfoIn.tileSplitBytes = surf->stencil_tile_split; + + for (level = 0; level <= surf->last_level; level++) { + r = compute_level(ws, surf, true, level, type, compressed, + &AddrSurfInfoIn, &AddrSurfInfoOut); + if (r) + return r; + + if (level == 0) { + surf->stencil_offset = surf->stencil_level[0].offset; + + /* For 2D modes only. */ + if (AddrSurfInfoOut.tileMode >= ADDR_TM_2D_TILED_THIN1) { + surf->stencil_tile_split = + AddrSurfInfoOut.pTileInfo->tileSplitBytes; + } + } + } + } + + return 0; +} + +static int amdgpu_surface_best(struct radeon_winsys *rws, + struct radeon_surf *surf) +{ + return 0; +} + +void amdgpu_surface_init_functions(struct amdgpu_winsys *ws) +{ + ws->base.surface_init = amdgpu_surface_init; + ws->base.surface_best = amdgpu_surface_best; +} diff --git a/src/gallium/winsys/radeon/amdgpu/amdgpu_winsys.c b/src/gallium/winsys/radeon/amdgpu/amdgpu_winsys.c index 0f3367a..d3992e9 100644 --- a/src/gallium/winsys/radeon/amdgpu/amdgpu_winsys.c +++ b/src/gallium/winsys/radeon/amdgpu/amdgpu_winsys.c @@ -39,6 +39,7 @@ #include #include #include +#include "amdgpu_id.h" #define CIK_TILE_MODE_COLOR_2D 14 @@ -198,6 +199,51 @@ static boolean do_winsys_init(struct amdgpu_winsys *ws) goto fail; } + /* family and rev_id are for addrlib */ + switch (ws->info.family) { + case CHIP_BONAIRE: + ws->family = FAMILY_CI; + ws->rev_id = CI_BONAIRE_M_A0; + break; + case CHIP_KAVERI: + ws->family = FAMILY_KV; + ws->rev_id = KV_SPECTRE_A0; + break; + case CHIP_KABINI: + ws->family = FAMILY_KV; + ws->rev_id = KB_KALINDI_A0; + break; + case CHIP_HAWAII: + ws->family = FAMILY_CI; + ws->rev_id = CI_HAWAII_P_A0; + break; + case CHIP_MULLINS: + ws->family = FAMILY_KV; + ws->rev_id = ML_GODAVARI_A0; + break; + case CHIP_TONGA: + ws->family = FAMILY_VI; + ws->rev_id = VI_TONGA_P_A0; + break; + case CHIP_ICELAND: + ws->family = FAMILY_VI; + ws->rev_id = VI_ICELAND_M_A0; + break; + case CHIP_CARRIZO: + ws->family = FAMILY_CZ; + ws->rev_id = CZ_CARRIZO_A0; + break; + default: + fprintf(stderr, "amdgpu: Unknown family.\n"); + goto fail; + } + + ws->addrlib = amdgpu_addr_create(ws); + if (!ws->addrlib) { + fprintf(stderr, "amdgpu: Cannot create addrlib.\n"); + goto fail; + } + /* Set hardware information. */ ws->info.gart_size = gtt.heap_size; ws->info.vram_size = vram.heap_size; @@ -232,6 +278,8 @@ fail: if (ws->ctx) { amdgpu_cs_ctx_free(ws->dev, ws->ctx); } + if (ws->addrlib) + AddrDestroy(ws->addrlib); amdgpu_device_deinitialize(ws->dev); ws->dev = NULL; return FALSE; @@ -251,6 +299,7 @@ static void amdgpu_winsys_destroy(struct radeon_winsys *rws) ws->cman->destroy(ws->cman); ws->kman->destroy(ws->kman); + AddrDestroy(ws->addrlib); amdgpu_cs_ctx_free(ws->dev, ws->ctx); amdgpu_device_deinitialize(ws->dev); @@ -451,6 +500,7 @@ struct radeon_winsys * amdgpu_bomgr_init_functions(ws); amdgpu_cs_init_functions(ws); + amdgpu_surface_init_functions(ws); pipe_mutex_init(ws->cs_stack_lock); diff --git a/src/gallium/winsys/radeon/amdgpu/amdgpu_winsys.h b/src/gallium/winsys/radeon/amdgpu/amdgpu_winsys.h index fc27f1c..8755108 100644 --- a/src/gallium/winsys/radeon/amdgpu/amdgpu_winsys.h +++ b/src/gallium/winsys/radeon/amdgpu/amdgpu_winsys.h @@ -31,6 +31,7 @@ #define AMDGPU_DRM_WINSYS_H #include "../radeon_winsys.h" +#include "addrlib/addrinterface.h" #include "os/os_thread.h" #include @@ -67,6 +68,9 @@ struct amdgpu_winsys { struct amdgpu_cs *cs_stack[RING_LAST]; struct amdgpu_gpu_info amdinfo; + ADDR_HANDLE addrlib; + uint32_t rev_id; + unsigned family; }; static INLINE struct amdgpu_winsys * @@ -76,5 +80,7 @@ amdgpu_winsys(struct radeon_winsys *base) } void amdgpu_ws_queue_cs(struct amdgpu_winsys *ws, struct amdgpu_cs *cs); +void amdgpu_surface_init_functions(struct amdgpu_winsys *ws); +ADDR_HANDLE amdgpu_addr_create(struct amdgpu_winsys *ws); #endif -- 1.9.1