|
LLVM 24.0.0git
|
Enumerations | |
| enum | { FIXED_NUM_SGPRS_FOR_INIT_BUG = AMDGPU::FIXED_NUM_SGPRS_FOR_INIT_BUG , TRAP_NUM_SGPRS = 16 } |
Variables | |
| static constexpr unsigned | MaxDynamicVGPRBlocks = 8 |
| Maximum number of VGPR blocks that can be allocated in dynamic VGPR mode. | |
| anonymous enum |
| Enumerator | |
|---|---|
| FIXED_NUM_SGPRS_FOR_INIT_BUG | |
| TRAP_NUM_SGPRS | |
Definition at line 170 of file AMDGPUBaseInfo.h.
| unsigned llvm::AMDGPU::IsaInfo::getAddressableLocalMemorySize | ( | const MCSubtargetInfo & | STI | ) |
STI. Definition at line 1150 of file AMDGPUBaseInfo.cpp.
References getLocalMemorySize(), and getMaxHWAddressableLocalMemorySize().
| unsigned llvm::AMDGPU::IsaInfo::getAddressableNumArchVGPRs | ( | const MCSubtargetInfo & | STI | ) |
STI. Definition at line 1325 of file AMDGPUBaseInfo.cpp.
References llvm::MCSubtargetInfo::getFeatureBits(), and llvm::FeatureBitset::test().
Referenced by llvm::GCNSubtarget::getAddressableNumArchVGPRs(), and getAddressableNumVGPRs().
| unsigned llvm::AMDGPU::IsaInfo::getAddressableNumVGPRs | ( | const MCSubtargetInfo & | STI, |
| unsigned | DynamicVGPRBlockSize ) |
STI. Definition at line 1332 of file AMDGPUBaseInfo.cpp.
References getAddressableNumArchVGPRs(), llvm::MCSubtargetInfo::getFeatureBits(), getVGPRAllocGranule(), and MaxDynamicVGPRBlocks.
Referenced by llvm::GCNSubtarget::getAddressableNumVGPRs(), getMaxNumVGPRs(), getMinNumVGPRs(), and llvm::GCNSchedStrategy::initialize().
| unsigned llvm::AMDGPU::IsaInfo::getAllocatedNumVGPRBlocks | ( | const MCSubtargetInfo & | STI, |
| unsigned | NumVGPRs, | ||
| unsigned | DynamicVGPRBlockSize, | ||
| std::optional< bool > | EnableWavefrontSize32 = std::nullopt ) |
STI when NumVGPRs are used. Definition at line 1444 of file AMDGPUBaseInfo.cpp.
References getGranulatedNumRegisterBlocks(), and getVGPRAllocGranule().
Referenced by llvm::GCNSchedStage::shouldRevertScheduling().
| unsigned llvm::AMDGPU::IsaInfo::getArchVGPRAllocGranule | ( | ) |
For subtargets with a unified VGPR file and mixed ArchVGPR/AGPR usage, returns the allocation granule for ArchVGPRs.
Definition at line 1323 of file AMDGPUBaseInfo.cpp.
Referenced by llvm::GCNRegPressure::getUnifiedVGPRNum().
|
static |
Definition at line 1275 of file AMDGPUBaseInfo.cpp.
References llvm::divideCeil().
Referenced by getAllocatedNumVGPRBlocks(), and getNumSGPRBlocks().
| unsigned llvm::AMDGPU::IsaInfo::getInstCacheLineSize | ( | const MCSubtargetInfo & | STI | ) |
STI. Definition at line 1087 of file AMDGPUBaseInfo.cpp.
References llvm::MCSubtargetInfo::getFeatureBits(), and llvm::FeatureBitset::test().
| unsigned llvm::AMDGPU::IsaInfo::getLocalMemorySize | ( | const MCSubtargetInfo & | STI | ) |
STI. Definition at line 1141 of file AMDGPUBaseInfo.cpp.
References getPhysicalLocalMemorySize(), llvm::AMDGPU::isFullSIMDMode(), and Size.
Referenced by getAddressableLocalMemorySize().
|
static |
Definition at line 1106 of file AMDGPUBaseInfo.cpp.
References llvm::MCSubtargetInfo::getFeatureBits(), and llvm::FeatureBitset::test().
Referenced by getAddressableLocalMemorySize(), and getPhysicalLocalMemorySize().
| unsigned llvm::AMDGPU::IsaInfo::getMaxNumSGPRs | ( | const MCSubtargetInfo & | STI, |
| unsigned | WavesPerEU, | ||
| bool | Addressable ) |
STI. Definition at line 1227 of file AMDGPUBaseInfo.cpp.
References assert(), llvm::AMDGPU::getAddressableNumSGPRs(), llvm::MCSubtargetInfo::getCPU(), llvm::AMDGPU::getIsaVersion(), llvm::AMDGPU::getSGPRAllocGranule(), getSGPRBudgetPerWave(), getSGPRTrapHandlerReserve(), llvm::AMDGPU::getTotalNumSGPRs(), llvm::AMDGPU::parseArchAMDGCN(), and llvm::Version.
Referenced by llvm::GCNSubtarget::getMaxNumSGPRs().
| unsigned llvm::AMDGPU::IsaInfo::getMaxNumVGPRs | ( | const MCSubtargetInfo & | STI, |
| unsigned | WavesPerEU, | ||
| unsigned | DynamicVGPRBlockSize ) |
STI. Definition at line 1424 of file AMDGPUBaseInfo.cpp.
References llvm::alignDown(), assert(), getAddressableNumVGPRs(), llvm::MCSubtargetInfo::getCPU(), llvm::MCSubtargetInfo::getFeatureBits(), llvm::AMDGPU::getTotalNumVGPRs(), getVGPRAllocGranule(), llvm::AMDGPU::parseArchAMDGCN(), and llvm::FeatureBitset::test().
Referenced by llvm::GCNSubtarget::getMaxNumVGPRs().
| unsigned llvm::AMDGPU::IsaInfo::getMaxWorkGroupsPerCU | ( | const MCSubtargetInfo & | STI, |
| unsigned | FlatWorkGroupSize ) |
STI and limited by given FlatWorkGroupSize. Definition at line 1155 of file AMDGPUBaseInfo.cpp.
References assert(), llvm::MCSubtargetInfo::getCPU(), llvm::MCSubtargetInfo::getFeatureBits(), llvm::AMDGPU::getMaxWavesPerEU(), llvm::AMDGPU::getNumWorkGroupSIMDs(), llvm::MCSubtargetInfo::getTargetTriple(), getWavesPerWorkGroup(), llvm::Triple::isAMDGCN(), llvm::AMDGPU::isFullSIMDMode(), llvm::AMDGPU::isGFX10Plus(), N, llvm::AMDGPU::parseArchAMDGCN(), and llvm::FeatureBitset::test().
Referenced by llvm::GCNSubtarget::getMaxWorkGroupsPerCU(), and llvm::R600Subtarget::getMaxWorkGroupsPerCU().
| unsigned llvm::AMDGPU::IsaInfo::getMinNumSGPRs | ( | const MCSubtargetInfo & | STI, |
| unsigned | WavesPerEU ) |
STI. Definition at line 1208 of file AMDGPUBaseInfo.cpp.
References assert(), llvm::AMDGPU::getAddressableNumSGPRs(), llvm::MCSubtargetInfo::getCPU(), llvm::AMDGPU::getIsaVersion(), llvm::AMDGPU::getMaxWavesPerEU(), llvm::AMDGPU::getSGPRAllocGranule(), getSGPRBudgetPerWave(), getSGPRTrapHandlerReserve(), llvm::AMDGPU::getTotalNumSGPRs(), llvm::AMDGPU::parseArchAMDGCN(), and llvm::Version.
Referenced by llvm::GCNSubtarget::getMinNumSGPRs().
| unsigned llvm::AMDGPU::IsaInfo::getMinNumVGPRs | ( | const MCSubtargetInfo & | STI, |
| unsigned | WavesPerEU, | ||
| unsigned | DynamicVGPRBlockSize ) |
STI. Definition at line 1387 of file AMDGPUBaseInfo.cpp.
References llvm::alignDown(), assert(), getAddressableNumVGPRs(), llvm::MCSubtargetInfo::getCPU(), llvm::MCSubtargetInfo::getFeatureBits(), llvm::AMDGPU::getMaxWavesPerEU(), getMinNumVGPRs(), getNumWavesPerEUWithNumVGPRs(), llvm::AMDGPU::getTotalNumVGPRs(), getVGPRAllocGranule(), llvm::AMDGPU::parseArchAMDGCN(), and llvm::FeatureBitset::test().
Referenced by getMinNumVGPRs(), and llvm::GCNSubtarget::getMinNumVGPRs().
| unsigned llvm::AMDGPU::IsaInfo::getNumExtraSGPRs | ( | const MCSubtargetInfo & | STI, |
| bool | VCCUsed, | ||
| bool | FlatScrUsed, | ||
| bool | XNACKUsed ) |
STI when the given special registers are used. Definition at line 1250 of file AMDGPUBaseInfo.cpp.
References llvm::MCSubtargetInfo::getCPU(), llvm::MCSubtargetInfo::getFeatureBits(), llvm::AMDGPU::getIsaVersion(), llvm::FeatureBitset::test(), and llvm::Version.
| unsigned llvm::AMDGPU::IsaInfo::getNumSGPRBlocks | ( | const MCSubtargetInfo & | STI, |
| unsigned | NumSGPRs ) |
STI when NumSGPRs are used. NumSGPRs should already include any special register counts. Definition at line 1280 of file AMDGPUBaseInfo.cpp.
References getGranulatedNumRegisterBlocks(), and getSGPREncodingGranule().
| unsigned llvm::AMDGPU::IsaInfo::getNumWavesPerEUWithNumVGPRs | ( | const MCSubtargetInfo & | STI, |
| unsigned | NumVGPRs, | ||
| unsigned | DynamicVGPRBlockSize ) |
NumVGPRs usage for given subtarget STI. Definition at line 1346 of file AMDGPUBaseInfo.cpp.
References llvm::MCSubtargetInfo::getCPU(), llvm::MCSubtargetInfo::getFeatureBits(), llvm::AMDGPU::getMaxWavesPerEU(), getNumWavesPerEUWithNumVGPRs(), llvm::AMDGPU::getTotalNumVGPRs(), getVGPRAllocGranule(), llvm::AMDGPU::parseArchAMDGCN(), and llvm::FeatureBitset::test().
Referenced by getMinNumVGPRs(), getNumWavesPerEUWithNumVGPRs(), and llvm::GCNSubtarget::getOccupancyWithNumVGPRs().
| unsigned llvm::AMDGPU::IsaInfo::getNumWavesPerEUWithNumVGPRs | ( | unsigned | NumVGPRs, |
| unsigned | Granule, | ||
| unsigned | MaxWaves, | ||
| unsigned | TotalNumVGPRs ) |
NumVGPRs usage, Granule size, MaxWaves possible, and TotalNumVGPRs available. Definition at line 1356 of file AMDGPUBaseInfo.cpp.
References llvm::alignTo().
| unsigned llvm::AMDGPU::IsaInfo::getOccupancyWithNumSGPRs | ( | const MCSubtargetInfo & | STI, |
| unsigned | SGPRs ) |
STI: the inverse of getMaxNumSGPRs(). Unlike getMaxNumSGPRs() the budget is not clamped to the addressable count, since the allocated count callers pass in can exceed it. Definition at line 1375 of file AMDGPUBaseInfo.cpp.
References llvm::MCSubtargetInfo::getCPU(), llvm::AMDGPU::getMaxWavesPerEU(), getOccupancyWithNumSGPRs(), llvm::AMDGPU::getSGPRAllocGranule(), getSGPRTrapHandlerReserve(), llvm::AMDGPU::getTotalNumSGPRs(), isSGPROccupancyLimited(), and llvm::AMDGPU::parseArchAMDGCN().
| unsigned llvm::AMDGPU::IsaInfo::getOccupancyWithNumSGPRs | ( | unsigned | SGPRs, |
| unsigned | MaxWaves, | ||
| unsigned | TotalNumSGPRs, | ||
| unsigned | Granule, | ||
| unsigned | TrapReserve ) |
MaxWaves, TotalNumSGPRs, Granule, TrapReserve). Subtarget-free core shared by the overload above and the occupancy MCExpr. Callers must check isSGPROccupancyLimited() first. Definition at line 1365 of file AMDGPUBaseInfo.cpp.
References llvm::alignTo().
Referenced by getOccupancyWithNumSGPRs(), and llvm::GCNSubtarget::getOccupancyWithNumSGPRs().
|
static |
Definition at line 1125 of file AMDGPUBaseInfo.cpp.
References llvm::MCSubtargetInfo::getFeatureBits(), getMaxHWAddressableLocalMemorySize(), and llvm::FeatureBitset::test().
Referenced by getLocalMemorySize().
|
static |
Definition at line 1199 of file AMDGPUBaseInfo.cpp.
References llvm::alignDown(), and assert().
Referenced by getMaxNumSGPRs(), and getMinNumSGPRs().
| unsigned llvm::AMDGPU::IsaInfo::getSGPREncodingGranule | ( | const MCSubtargetInfo & | STI | ) |
STI. Definition at line 1187 of file AMDGPUBaseInfo.cpp.
Referenced by llvm::AMDGPUDisassembler::decodeCOMPUTE_PGM_RSRC1(), getNumSGPRBlocks(), and llvm::GCNSubtarget::getSGPREncodingGranule().
|
static |
Definition at line 1190 of file AMDGPUBaseInfo.cpp.
References llvm::MCSubtargetInfo::getFeatureBits(), llvm::FeatureBitset::test(), and TRAP_NUM_SGPRS.
Referenced by getMaxNumSGPRs(), getMinNumSGPRs(), and getOccupancyWithNumSGPRs().
| unsigned llvm::AMDGPU::IsaInfo::getVGPRAllocGranule | ( | const MCSubtargetInfo & | STI, |
| unsigned | DynamicVGPRBlockSize, | ||
| std::optional< bool > | EnableWavefrontSize32 = std::nullopt ) |
STI.For subtargets which support it, EnableWavefrontSize32 should match the ENABLE_WAVEFRONT_SIZE32 kernel descriptor field.
Definition at line 1286 of file AMDGPUBaseInfo.cpp.
References llvm::MCSubtargetInfo::getFeatureBits(), llvm::AMDGPU::hasGFX10_3Insts(), and llvm::FeatureBitset::test().
Referenced by createOccupancy(), llvm::SIFrameLowering::emitEntryFunctionPrologue(), getAddressableNumVGPRs(), getAllocatedNumVGPRBlocks(), getMaxNumVGPRs(), getMinNumVGPRs(), getNumWavesPerEUWithNumVGPRs(), llvm::GCNSubtarget::getVGPRAllocGranule(), and llvm::GCNSchedStrategy::initialize().
| unsigned llvm::AMDGPU::IsaInfo::getVGPREncodingGranule | ( | const MCSubtargetInfo & | STI, |
| std::optional< bool > | EnableWavefrontSize32 = std::nullopt ) |
STI.For subtargets which support it, EnableWavefrontSize32 should match the ENABLE_WAVEFRONT_SIZE32 kernel descriptor field.
Definition at line 1308 of file AMDGPUBaseInfo.cpp.
References llvm::MCSubtargetInfo::getFeatureBits(), and llvm::FeatureBitset::test().
Referenced by llvm::AMDGPUDisassembler::decodeCOMPUTE_PGM_RSRC1(), and llvm::GCNSubtarget::getVGPREncodingGranule().
| unsigned llvm::AMDGPU::IsaInfo::getWavefrontSize | ( | const MCSubtargetInfo & | STI | ) |
STI. Definition at line 1095 of file AMDGPUBaseInfo.cpp.
References llvm::MCSubtargetInfo::getFeatureBits(), and llvm::FeatureBitset::test().
Referenced by getWavesPerWorkGroup().
| unsigned llvm::AMDGPU::IsaInfo::getWavesPerEUForWorkGroup | ( | const MCSubtargetInfo & | STI, |
| unsigned | FlatWorkGroupSize ) |
FlatWorkGroupSize. Definition at line 1176 of file AMDGPUBaseInfo.cpp.
References llvm::divideCeil(), llvm::AMDGPU::getNumWorkGroupSIMDs(), getWavesPerWorkGroup(), and llvm::AMDGPU::isFullSIMDMode().
Referenced by llvm::GCNSubtarget::getWavesPerEUForWorkGroup(), and llvm::R600Subtarget::getWavesPerEUForWorkGroup().
| unsigned llvm::AMDGPU::IsaInfo::getWavesPerWorkGroup | ( | const MCSubtargetInfo & | STI, |
| unsigned | FlatWorkGroupSize ) |
STI and FlatWorkGroupSize. Definition at line 1182 of file AMDGPUBaseInfo.cpp.
References llvm::divideCeil(), and getWavefrontSize().
Referenced by getMaxWorkGroupsPerCU(), and getWavesPerEUForWorkGroup().
| bool llvm::AMDGPU::IsaInfo::isSGPROccupancyLimited | ( | const MCSubtargetInfo & | STI | ) |
STI (true pre-GFX10). One named capability so callers don't test the version. Definition at line 1244 of file AMDGPUBaseInfo.cpp.
References llvm::MCSubtargetInfo::getCPU(), llvm::AMDGPU::getIsaVersion(), and llvm::AMDGPU::IsaVersion::Major.
Referenced by createOccupancy(), and getOccupancyWithNumSGPRs().
|
inline |
Returns true if Lhs and Rhs are incompatible (both specific but different).
Definition at line 177 of file AMDGPUBaseInfo.h.
|
staticconstexpr |
Maximum number of VGPR blocks that can be allocated in dynamic VGPR mode.
Definition at line 253 of file AMDGPUBaseInfo.h.
Referenced by getAddressableNumVGPRs().