AI/rocm-systems
2
0
Forka 0
Codice Problemi Pull Request Actions Pacchetti Progetti Rilasci Wiki Attività

Update addrlib.

Sfoglia il codice sorgente 
Corrects type errors and adds gfx1030 support.

Change-Id: I7fe7b5ded68b1e26fb899676712ad019089a45f3
This commit is contained in:
Sean Keely
2020-06-17 14:09:23 -05:00
parent 0dae86f686
commit 4081c564ee
6 ha cambiato i file con 96 aggiunte e 124 eliminazioni
Scarica il file Patch Scarica il file Diff Expand all files Collapse all files
runtime/hsa-runtime/image/addrlib/src/gfx10/gfx10addrlib.cpp
+55 -70
Vedi File
@@ -673,88 +673,67 @@ ADDR_E_RETURNCODE Gfx10Lib::HwlComputeDccAddrFromCoord(
}
else
{
ADDR2_COMPUTE_DCCINFO_INPUT input = {0};
input.size = sizeof(input);
input.dccKeyFlags = pIn->dccKeyFlags;
input.colorFlags = pIn->colorFlags;
input.swizzleMode = pIn->swizzleMode;
input.resourceType = pIn->resourceType;
input.bpp = pIn->bpp;
input.unalignedWidth = Max(pIn->unalignedWidth, 1u);
input.unalignedHeight = Max(pIn->unalignedHeight, 1u);
input.numSlices = Max(pIn->numSlices, 1u);
input.numFrags = Max(pIn->numFrags, 1u);
input.numMipLevels = Max(pIn->numMipLevels, 1u);
const UINT_32 elemLog2 = Log2(pIn->bpp >> 3);
const UINT_32 numPipeLog2 = m_pipesLog2;
const UINT_32 pipeMask = (1 << numPipeLog2) - 1;
UINT_32 index = m_dccBaseIndex + elemLog2;
const UINT_8* patIdxTable;
ADDR2_COMPUTE_DCCINFO_OUTPUT output = {0};
output.size = sizeof(output);
returnCode = ComputeDccInfo(&input, &output);
if (returnCode == ADDR_OK)
if (m_settings.supportRbPlus)
{
const UINT_32 elemLog2 = Log2(pIn->bpp >> 3);
const UINT_32 numPipeLog2 = m_pipesLog2;
const UINT_32 pipeMask = (1 << numPipeLog2) - 1;
UINT_32 index = m_dccBaseIndex + elemLog2;
const UINT_8* patIdxTable;
patIdxTable = DCC_64K_R_X_RBPLUS_PATIDX;
if (m_settings.supportRbPlus)
if (pIn->dccKeyFlags.pipeAligned)
{
patIdxTable = DCC_64K_R_X_RBPLUS_PATIDX;
index += MaxNumOfBpp;
if (pIn->dccKeyFlags.pipeAligned)
if (m_numPkrLog2 < 2)
{
index += MaxNumOfBpp;
if (m_numPkrLog2 < 2)
{
index += m_pipesLog2 * MaxNumOfBpp;
}
else
{
// 4 groups for "m_numPkrLog2 < 2" case
index += 4 * MaxNumOfBpp;
const UINT_32 dccPipePerPkr = 3;
index += (m_numPkrLog2 - 2) * dccPipePerPkr * MaxNumOfBpp +
(m_pipesLog2 - m_numPkrLog2) * MaxNumOfBpp;
}
}
}
else
{
patIdxTable = DCC_64K_R_X_PATIDX;
if (pIn->dccKeyFlags.pipeAligned)
{
index += (numPipeLog2 + UnalignedDccType) * MaxNumOfBpp;
index += m_pipesLog2 * MaxNumOfBpp;
}
else
{
index += Min(numPipeLog2, UnalignedDccType - 1) * MaxNumOfBpp;
// 4 groups for "m_numPkrLog2 < 2" case
index += 4 * MaxNumOfBpp;
const UINT_32 dccPipePerPkr = 3;
index += (m_numPkrLog2 - 2) * dccPipePerPkr * MaxNumOfBpp +
(m_pipesLog2 - m_numPkrLog2) * MaxNumOfBpp;
}
}
const UINT_32 blkSizeLog2 = Log2(output.metaBlkWidth) + Log2(output.metaBlkHeight) + elemLog2 - 8;
const UINT_32 blkMask = (1 << blkSizeLog2) - 1;
const UINT_32 blkOffset = ComputeOffsetFromSwizzlePattern(DCC_64K_R_X_SW_PATTERN[patIdxTable[index]],
blkSizeLog2 + 1, // +1 for nibble offset
pIn->x,
pIn->y,
pIn->slice,
0);
const UINT_32 xb = pIn->x / output.metaBlkWidth;
const UINT_32 yb = pIn->y / output.metaBlkHeight;
const UINT_32 pb = output.pitch / output.metaBlkWidth;
const UINT_32 blkIndex = (yb * pb) + xb;
const UINT_32 pipeXor = ((pIn->pipeXor & pipeMask) << m_pipeInterleaveLog2) & blkMask;
pOut->addr = (static_cast<UINT_64>(output.dccRamSliceSize) * pIn->slice) +
(blkIndex * (1 << blkSizeLog2)) +
((blkOffset >> 1) ^ pipeXor);
}
else
{
patIdxTable = DCC_64K_R_X_PATIDX;
if (pIn->dccKeyFlags.pipeAligned)
{
index += (numPipeLog2 + UnalignedDccType) * MaxNumOfBpp;
}
else
{
index += Min(numPipeLog2, UnalignedDccType - 1) * MaxNumOfBpp;
}
}
const UINT_32 blkSizeLog2 = Log2(pIn->metaBlkWidth) + Log2(pIn->metaBlkHeight) + elemLog2 - 8;
const UINT_32 blkMask = (1 << blkSizeLog2) - 1;
const UINT_32 blkOffset = ComputeOffsetFromSwizzlePattern(DCC_64K_R_X_SW_PATTERN[patIdxTable[index]],
blkSizeLog2 + 1, // +1 for nibble offset
pIn->x,
pIn->y,
pIn->slice,
0);
const UINT_32 xb = pIn->x / pIn->metaBlkWidth;
const UINT_32 yb = pIn->y / pIn->metaBlkHeight;
const UINT_32 pb = pIn->pitch / pIn->metaBlkWidth;
const UINT_32 blkIndex = (yb * pb) + xb;
const UINT_32 pipeXor = ((pIn->pipeXor & pipeMask) << m_pipeInterleaveLog2) & blkMask;
pOut->addr = (static_cast<UINT_64>(pIn->dccRamSliceSize) * pIn->slice) +
(blkIndex * (1 << blkSizeLog2)) +
((blkOffset >> 1) ^ pipeXor);
}
return returnCode;
@@ -943,6 +922,12 @@ ChipFamily Gfx10Lib::HwlConvertChipFamily(
{
case FAMILY_NV:
m_settings.isDcn2 = 1;
if (ASICREV_IS_SIENNA_M(chipRevision))
{
m_settings.supportRbPlus = 1;
m_settings.dccUnsup3DSwDis = 0;
}
break;
default:
ADDR_ASSERT(!"Unknown chip family");