From: Xavi Zhang Date: Mon, 29 Feb 2016 06:36:08 +0000 (-0500) Subject: amdgpu/addrlib: Check prt flag for PRT_THIN1 extra padding for DCC. X-Git-Url: https://git.libre-soc.org/?a=commitdiff_plain;h=cb8844392cd34613db2f79edcc5ae2ac68fe5ec0;p=mesa.git amdgpu/addrlib: Check prt flag for PRT_THIN1 extra padding for DCC. --- diff --git a/src/amd/addrlib/r800/ciaddrlib.cpp b/src/amd/addrlib/r800/ciaddrlib.cpp index 7272c49affe..cf08566f869 100644 --- a/src/amd/addrlib/r800/ciaddrlib.cpp +++ b/src/amd/addrlib/r800/ciaddrlib.cpp @@ -896,7 +896,7 @@ VOID CiLib::HwlOptimizeTileMode( { UINT_32 thickness = Thickness(tileMode); - if (pInOut->maxBaseAlign < Block64K) + if ((pInOut->maxBaseAlign != 0) && (pInOut->maxBaseAlign < Block64K)) { tileMode = (thickness == 1) ? ADDR_TM_1D_TILED_THIN1 : ADDR_TM_1D_TILED_THICK; } @@ -1213,7 +1213,7 @@ VOID CiLib::HwlSetupTileInfo( INT macroModeIndex = TileIndexInvalid; // Fail-safe code - if (!IsLinear(tileMode)) + if (IsLinear(tileMode) == FALSE) { // Thick tile modes must use thick micro tile mode but Bonaire does not support due to // old derived netlists (UBTS 404321) @@ -1825,9 +1825,6 @@ INT_32 CiLib::HwlComputeMacroModeIndex( 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]; } @@ -2027,25 +2024,27 @@ UINT_64 CiLib::HwlComputeMetadataNibbleAddress( **************************************************************************************************** */ VOID CiLib::HwlComputeSurfaceAlignmentsMacroTiled( - 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 - UINT_32* pMacroTileWidth, ///< [out] macro tile width in pixels - UINT_32* pMacroTileHeight ///< [out] macro tile height in pixels + 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_COMPUTE_SURFACE_INFO_OUTPUT* pOut ///< [in,out] Surface output ) const { + // This is to workaround a H/W limitation that DCC doesn't work when pipe config is switched to + // P4. In theory, all asics that have such switching should be patched but we now only know what + // to pad for Fiji. if ((m_settings.isFiji == TRUE) && (flags.dccCompatible == TRUE) && + (flags.prt == FALSE) && (mipLevel == 0) && - (tileMode == ADDR_TM_PRT_TILED_THIN1)) + (tileMode == ADDR_TM_PRT_TILED_THIN1) && + (pOut->dccUnsupport == TRUE)) { - *pPitchAlign = PowTwoAlign(*pPitchAlign, 256); + pOut->pitchAlign = PowTwoAlign(pOut->pitchAlign, 256); + // In case the client still requests DCC usage. + pOut->dccUnsupport = FALSE; } } diff --git a/src/amd/addrlib/r800/ciaddrlib.h b/src/amd/addrlib/r800/ciaddrlib.h index 7fa6b75e864..93b2853d066 100644 --- a/src/amd/addrlib/r800/ciaddrlib.h +++ b/src/amd/addrlib/r800/ciaddrlib.h @@ -177,9 +177,7 @@ protected: virtual VOID HwlComputeSurfaceAlignmentsMacroTiled( 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, - UINT_32* pMacroTileWidth, UINT_32* pMacroTileHeight) const; + UINT_32 mipLevel, UINT_32 numSamples, ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pOut) const; private: VOID ReadGbTileMode( @@ -207,6 +205,8 @@ private: UINT_32 numOfSamplesPerSplit) const; static const UINT_32 MacroTileTableSize = 16; + static const UINT_32 PrtMacroModeOffset = MacroTileTableSize / 2; + ADDR_TILEINFO m_macroTileTable[MacroTileTableSize]; UINT_32 m_noOfMacroEntries; BOOL_32 m_allowNonDispThickModes; diff --git a/src/amd/addrlib/r800/egbaddrlib.cpp b/src/amd/addrlib/r800/egbaddrlib.cpp index 2d1123a3ccd..7942d20d75a 100644 --- a/src/amd/addrlib/r800/egbaddrlib.cpp +++ b/src/amd/addrlib/r800/egbaddrlib.cpp @@ -463,12 +463,7 @@ BOOL_32 EgBasedLib::ComputeSurfaceInfoMacroTiled( pIn->flags, pIn->mipLevel, numSamples, - pOut->pTileInfo, - &pOut->baseAlign, - &pOut->pitchAlign, - &pOut->heightAlign, - &pOut->blockWidth, - &pOut->blockHeight); + pOut); if (valid) { @@ -521,12 +516,7 @@ BOOL_32 EgBasedLib::ComputeSurfaceInfoMacroTiled( pIn->flags, pIn->mipLevel, numSamples, - pOut->pTileInfo, - &pOut->baseAlign, - &pOut->pitchAlign, - &pOut->heightAlign, - &pOut->blockWidth, - &pOut->blockHeight); + pOut); } // @@ -854,19 +844,16 @@ BOOL_32 EgBasedLib::HwlReduceBankWidthHeight( **************************************************************************************************** */ BOOL_32 EgBasedLib::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 - UINT_32* pMacroTileWidth, ///< [out] macro tile width in pixels - UINT_32* pMacroTileHeight ///< [out] macro tile height in pixels + 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_COMPUTE_SURFACE_INFO_OUTPUT* pOut ///< [in,out] Surface output ) const { + ADDR_TILEINFO* pTileInfo = pOut->pTileInfo; + BOOL_32 valid = SanityCheckMacroTiled(pTileInfo); if (valid) @@ -924,10 +911,10 @@ BOOL_32 EgBasedLib::ComputeSurfaceAlignmentsMacroTiled( macroTileWidth = MicroTileWidth * pTileInfo->bankWidth * pipes * pTileInfo->macroAspectRatio; - *pPitchAlign = macroTileWidth; - *pMacroTileWidth = macroTileWidth; + pOut->pitchAlign = macroTileWidth; + pOut->blockWidth = macroTileWidth; - AdjustPitchAlignment(flags, pPitchAlign); + AdjustPitchAlignment(flags, &pOut->pitchAlign); // // The required granularity for height is the macro tile height. @@ -935,18 +922,16 @@ BOOL_32 EgBasedLib::ComputeSurfaceAlignmentsMacroTiled( macroTileHeight = MicroTileHeight * pTileInfo->bankHeight * pTileInfo->banks / pTileInfo->macroAspectRatio; - *pHeightAlign = macroTileHeight; - *pMacroTileHeight = macroTileHeight; + pOut->heightAlign = macroTileHeight; + pOut->blockHeight = macroTileHeight; // // Compute base alignment // - *pBaseAlign = pipes * - pTileInfo->bankWidth * pTileInfo->banks * pTileInfo->bankHeight * tileSize; + pOut->baseAlign = + pipes * pTileInfo->bankWidth * pTileInfo->banks * pTileInfo->bankHeight * tileSize; - HwlComputeSurfaceAlignmentsMacroTiled(tileMode, bpp, flags, mipLevel, numSamples, - pTileInfo, pBaseAlign, pPitchAlign, pHeightAlign, - pMacroTileWidth, pMacroTileHeight); + HwlComputeSurfaceAlignmentsMacroTiled(tileMode, bpp, flags, mipLevel, numSamples, pOut); } return valid; @@ -1169,16 +1154,12 @@ BOOL_32 EgBasedLib::HwlGetAlignmentInfoMacroTiled( ADDR_ASSERT(IsMacroTiled(pIn->tileMode)); - UINT_32 baseAlign; - UINT_32 pitchAlign; - UINT_32 heightAlign; - UINT_32 macroTileWidth; - UINT_32 macroTileHeight; UINT_32 numSamples = (pIn->numFrags == 0) ? pIn->numSamples : pIn->numFrags; ADDR_ASSERT(pIn->pTileInfo); ADDR_TILEINFO tileInfo = *pIn->pTileInfo; ADDR_COMPUTE_SURFACE_INFO_OUTPUT out = {0}; + out.pTileInfo = &tileInfo; if (UseTileIndex(pIn->tileIndex)) { @@ -1202,18 +1183,13 @@ BOOL_32 EgBasedLib::HwlGetAlignmentInfoMacroTiled( pIn->flags, pIn->mipLevel, numSamples, - &tileInfo, - &baseAlign, - &pitchAlign, - &heightAlign, - ¯oTileWidth, - ¯oTileHeight); + &out); if (valid) { - *pPitchAlign = pitchAlign; - *pHeightAlign = heightAlign; - *pSizeAlign = baseAlign; + *pPitchAlign = out.pitchAlign; + *pHeightAlign = out.heightAlign; + *pSizeAlign = out.baseAlign; } return valid; diff --git a/src/amd/addrlib/r800/egbaddrlib.h b/src/amd/addrlib/r800/egbaddrlib.h index 50fd3d927ad..dbe712a6714 100644 --- a/src/amd/addrlib/r800/egbaddrlib.h +++ b/src/amd/addrlib/r800/egbaddrlib.h @@ -303,9 +303,7 @@ protected: virtual VOID HwlComputeSurfaceAlignmentsMacroTiled( 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, - UINT_32* pMacroTileWidth, UINT_32* pMacroTileHeight) const + UINT_32 mipLevel, UINT_32 numSamples, ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pOut) const { } @@ -340,9 +338,7 @@ private: 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, - UINT_32* pMacroTileWidth, UINT_32* pMacroTileHeight) const; + ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pOut) const; /// Surface addressing functions UINT_64 DispatchComputeSurfaceAddrFromCoord( diff --git a/src/amd/addrlib/r800/siaddrlib.cpp b/src/amd/addrlib/r800/siaddrlib.cpp index 299951edc01..c1f6dac15a1 100644 --- a/src/amd/addrlib/r800/siaddrlib.cpp +++ b/src/amd/addrlib/r800/siaddrlib.cpp @@ -3381,22 +3381,17 @@ ADDR_E_RETURNCODE SiLib::HwlGetMaxAlignments( **************************************************************************************************** */ VOID SiLib::HwlComputeSurfaceAlignmentsMacroTiled( - 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 - UINT_32* pMacroTileWidth, ///< [out] macro tile width in pixels - UINT_32* pMacroTileHeight ///< [out] macro tile height in pixels + 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_COMPUTE_SURFACE_INFO_OUTPUT* pOut ///< [in,out] Surface output ) const { if ((mipLevel == 0) && (flags.prt)) { - UINT_32 macroTileSize = (*pMacroTileWidth) * (*pMacroTileHeight) * numSamples * bpp / 8; + UINT_32 macroTileSize = pOut->blockWidth * pOut->blockHeight * numSamples * bpp / 8; if (macroTileSize < PrtTileSize) { @@ -3404,8 +3399,8 @@ VOID SiLib::HwlComputeSurfaceAlignmentsMacroTiled( ADDR_ASSERT((PrtTileSize % macroTileSize) == 0); - *pPitchAlign *= numMacroTiles; - *pBaseAlign *= numMacroTiles; + pOut->pitchAlign *= numMacroTiles; + pOut->baseAlign *= numMacroTiles; } } } diff --git a/src/amd/addrlib/r800/siaddrlib.h b/src/amd/addrlib/r800/siaddrlib.h index 53ec68ba06a..6daaa2388a7 100644 --- a/src/amd/addrlib/r800/siaddrlib.h +++ b/src/amd/addrlib/r800/siaddrlib.h @@ -249,9 +249,7 @@ protected: virtual VOID HwlComputeSurfaceAlignmentsMacroTiled( 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, - UINT_32* pMacroTileWidth, UINT_32* pMacroTileHeight) const; + UINT_32 mipLevel, UINT_32 numSamples, ADDR_COMPUTE_SURFACE_INFO_OUTPUT* pOut) const; // Get equation table pointer and number of equations virtual UINT_32 HwlGetEquationTableInfo(const ADDR_EQUATION** ppEquationTable) const @@ -286,7 +284,7 @@ protected: UINT_32 GetPipePerSurf(AddrPipeCfg pipeConfig) const; static const UINT_32 TileTableSize = 32; - TileConfig m_tileTable[TileTableSize]; + TileConfig m_tileTable[TileTableSize]; UINT_32 m_noOfEntries; // Max number of bpp (8bpp/16bpp/32bpp/64bpp/128bpp)