diff options
Diffstat (limited to 'llvm/lib/Target/AMDGPU/AMDGPU.td')
| -rw-r--r-- | llvm/lib/Target/AMDGPU/AMDGPU.td | 105 |
1 files changed, 101 insertions, 4 deletions
diff --git a/llvm/lib/Target/AMDGPU/AMDGPU.td b/llvm/lib/Target/AMDGPU/AMDGPU.td index b2b2b37..071c940 100644 --- a/llvm/lib/Target/AMDGPU/AMDGPU.td +++ b/llvm/lib/Target/AMDGPU/AMDGPU.td @@ -89,6 +89,12 @@ def FeatureEnableFlatScratch : SubtargetFeature<"enable-flat-scratch", "Use scratch_* flat memory instructions to access scratch" >; +def FeatureFlatGVSMode : SubtargetFeature<"flat-gvs-mode", + "FlatGVSMode", + "true", + "Have GVS addressing mode with flat_* instructions" +>; + def FeatureAddNoCarryInsts : SubtargetFeature<"add-no-carry-insts", "AddNoCarryInsts", "true", @@ -143,6 +149,12 @@ def FeatureFmaMixInsts : SubtargetFeature<"fma-mix-insts", "Has v_fma_mix_f32, v_fma_mixlo_f16, v_fma_mixhi_f16 instructions" >; +def FeatureFmaMixBF16Insts : SubtargetFeature<"fma-mix-bf16-insts", + "HasFmaMixBF16Insts", + "true", + "Has v_fma_mix_f32_bf16, v_fma_mixlo_bf16, v_fma_mixhi_bf16 instructions" +>; + def FeatureIEEEMinimumMaximumInsts : SubtargetFeature<"ieee-minimum-maximum-insts", "HasIEEEMinimumMaximumInsts", "true", @@ -161,6 +173,12 @@ def FeatureMinimum3Maximum3F16 : SubtargetFeature<"minimum3-maximum3-f16", "Has v_minimum3_f16 and v_maximum3_f16 instructions" >; +def FeatureMin3Max3PKF16 : SubtargetFeature<"min3-max3-pkf16", + "HasMin3Max3PKF16", + "true", + "Has v_pk_min3_num_f16 and v_pk_max3_num_f16 instructions" +>; + def FeatureMinimum3Maximum3PKF16 : SubtargetFeature<"minimum3-maximum3-pkf16", "HasMinimum3Maximum3PKF16", "true", @@ -250,12 +268,30 @@ def FeatureInstFwdPrefetchBug : SubtargetFeature<"inst-fwd-prefetch-bug", "S_INST_PREFETCH instruction causes shader to hang" >; +def FeatureVmemPrefInsts : SubtargetFeature<"vmem-pref-insts", + "HasVmemPrefInsts", + "true", + "Has flat_prefect_b8 and global_prefetch_b8 instructions" +>; + def FeatureSafeSmemPrefetch : SubtargetFeature<"safe-smem-prefetch", "HasSafeSmemPrefetch", "true", "SMEM prefetches do not fail on illegal address" >; +def FeatureSafeCUPrefetch : SubtargetFeature<"safe-cu-prefetch", + "HasSafeCUPrefetch", + "true", + "VMEM CU scope prefetches do not fail on illegal address" +>; + +def FeatureCUStores : SubtargetFeature<"cu-stores", + "HasCUStores", + "true", + "Whether SCOPE_CU stores can be used on GFX12.5" +>; + def FeatureVcmpxExecWARHazard : SubtargetFeature<"vcmpx-exec-war-hazard", "HasVcmpxExecWARHazard", "true", @@ -553,6 +589,12 @@ def FeatureBF16ConversionInsts : SubtargetFeature<"bf16-cvt-insts", "Has bf16 conversion instructions" >; +def FeatureBF16PackedInsts : SubtargetFeature<"bf16-pk-insts", + "HasBF16PackedInsts", + "true", + "Has bf16 packed instructions (fma, add, mul, max, min)" +>; + def FeatureVOP3P : SubtargetFeature<"vop3p", "HasVOP3PInsts", "true", @@ -1112,6 +1154,12 @@ def FeatureBitOp3Insts : SubtargetFeature<"bitop3-insts", "Has v_bitop3_b32/v_bitop3_b16 instructions" >; +def FeatureTanhInsts : SubtargetFeature<"tanh-insts", + "HasTanhInsts", + "true", + "Has v_tanh_f32/f16 instructions" +>; + def FeatureTransposeLoadF4F6Insts : SubtargetFeature<"transpose-load-f4f6-insts", "HasTransposeLoadF4F6Insts", "true", @@ -1337,6 +1385,13 @@ def FeatureLshlAddU64Inst : SubtargetFeature<"lshl-add-u64-inst", "HasLshlAddU64Inst", "true", "Has v_lshl_add_u64 instruction">; +def FeatureAddSubU64Insts + : SubtargetFeature<"add-sub-u64-insts", "HasAddSubU64Insts", "true", + "Has v_add_u64 and v_sub_u64 instructions">; + +def FeatureMadU32Inst : SubtargetFeature<"mad-u32-inst", "HasMadU32Inst", + "true", "Has v_mad_u32 instruction">; + def FeatureMemToLDSLoad : SubtargetFeature<"vmem-to-lds-load-insts", "HasVMemToLDSLoad", "true", @@ -1836,7 +1891,8 @@ def FeatureISAVersion11_Common : FeatureSet< FeatureImageInsts, FeaturePackedTID, FeatureVcmpxPermlaneHazard, - FeatureMemoryAtomicFAddF32DenormalSupport]>; + FeatureMemoryAtomicFAddF32DenormalSupport, + FeatureRealTrue16Insts]>; // There are few workarounds that need to be // added to all targets. This pessimizes codegen @@ -1856,8 +1912,7 @@ def FeatureISAVersion11_0_Common : FeatureSet< [FeatureMSAALoadDstSelBug, FeatureVALUTransUseHazard, FeatureMADIntraFwdBug, - FeaturePrivEnabledTrap2NopBug, - FeatureRealTrue16Insts])>; + FeaturePrivEnabledTrap2NopBug])>; def FeatureISAVersion11_0_0 : FeatureSet< !listconcat(FeatureISAVersion11_0_Common.Features, @@ -1942,6 +1997,7 @@ def FeatureISAVersion12 : FeatureSet< def FeatureISAVersion12_50 : FeatureSet< [FeatureGFX12, FeatureGFX1250Insts, + FeatureCUStores, FeatureCuMode, Feature64BitLiterals, FeatureLDSBankCount32, @@ -1954,6 +2010,7 @@ def FeatureISAVersion12_50 : FeatureSet< FeatureShaderCyclesHiLoRegisters, FeatureArchitectedFlatScratch, FeatureArchitectedSGPRs, + FeatureFlatGVSMode, FeatureAtomicFaddRtnInsts, FeatureAtomicFaddNoRtnInsts, FeatureAtomicDsPkAdd16Insts, @@ -1972,10 +2029,14 @@ def FeatureISAVersion12_50 : FeatureSet< FeatureScalarDwordx3Loads, FeatureDPPSrc1SGPR, FeatureBitOp3Insts, + FeatureTanhInsts, FeatureTransposeLoadF4F6Insts, FeatureBF16TransInsts, FeatureBF16ConversionInsts, + FeatureBF16PackedInsts, FeatureCvtPkF16F32Inst, + FeatureFmaMixBF16Insts, + FeatureMin3Max3PKF16, FeatureMinimum3Maximum3PKF16, FeaturePrngInst, FeaturePermlane16Swap, @@ -1988,7 +2049,10 @@ def FeatureISAVersion12_50 : FeatureSet< FeatureFlatBufferGlobalAtomicFaddF64Inst, FeatureMemoryAtomicFAddF32DenormalSupport, FeatureKernargPreload, + FeatureVmemPrefInsts, FeatureLshlAddU64Inst, + FeatureAddSubU64Insts, + FeatureMadU32Inst, FeatureLdsBarrierArriveAtomic, FeatureSetPrioIncWgInst, ]>; @@ -2335,6 +2399,10 @@ def HasMinimum3Maximum3F16 : Predicate<"Subtarget->hasMinimum3Maximum3F16()">, AssemblerPredicate<(all_of FeatureMinimum3Maximum3F16)>; +def HasMin3Max3PKF16 : + Predicate<"Subtarget->hasMin3Max3PKF16()">, + AssemblerPredicate<(all_of FeatureMin3Max3PKF16)>; + def HasMinimum3Maximum3PKF16 : Predicate<"Subtarget->hasMinimum3Maximum3PKF16()">, AssemblerPredicate<(all_of FeatureMinimum3Maximum3PKF16)>; @@ -2365,7 +2433,7 @@ def HasAtomicFMinFMaxF64FlatInsts : def HasLdsAtomicAddF64 : Predicate<"Subtarget->hasLdsAtomicAddF64()">, - AssemblerPredicate<(any_of FeatureGFX90AInsts)>; + AssemblerPredicate<(any_of FeatureGFX90AInsts, FeatureGFX1250Insts)>; def HasFlatGlobalInsts : Predicate<"Subtarget->hasFlatGlobalInsts()">, AssemblerPredicate<(all_of FeatureFlatGlobalInsts)>; @@ -2381,6 +2449,9 @@ def HasFlatScratchSTMode : Predicate<"Subtarget->hasFlatScratchSTMode()">, def HasFlatScratchSVSMode : Predicate<"Subtarget->hasFlatScratchSVSMode()">, AssemblerPredicate<(any_of FeatureGFX940Insts, FeatureGFX11Insts)>; +def HasFlatGVSMode : Predicate<"Subtarget->hasFlatGVSMode()">, + AssemblerPredicate<(all_of FeatureFlatGVSMode)>; + def HasGFX10_AEncoding : Predicate<"Subtarget->hasGFX10_AEncoding()">, AssemblerPredicate<(all_of FeatureGFX10_AEncoding)>; @@ -2455,6 +2526,9 @@ def HasBF16TransInsts : Predicate<"Subtarget->hasBF16TransInsts()">, def HasBF16ConversionInsts : Predicate<"Subtarget->hasBF16ConversionInsts()">, AssemblerPredicate<(all_of FeatureBF16ConversionInsts)>; +def HasBF16PackedInsts : Predicate<"Subtarget->hasBF16PackedInsts()">, + AssemblerPredicate<(all_of FeatureBF16PackedInsts)>; + def HasVOP3PInsts : Predicate<"Subtarget->hasVOP3PInsts()">, AssemblerPredicate<(all_of FeatureVOP3P)>; @@ -2502,6 +2576,14 @@ def HasFmaakFmamkF64Insts : Predicate<"Subtarget->hasFmaakFmamkF64Insts()">, AssemblerPredicate<(any_of FeatureGFX1250Insts)>; +def HasPkAddMinMaxInsts : + Predicate<"Subtarget->hasPkAddMinMaxInsts()">, + AssemblerPredicate<(any_of FeatureGFX1250Insts)>; + +def HasPkMinMax3Insts : + Predicate<"Subtarget->hasPkMinMax3Insts()">, + AssemblerPredicate<(any_of FeatureGFX1250Insts)>; + def HasImageInsts : Predicate<"Subtarget->hasImageInsts()">, AssemblerPredicate<(all_of FeatureImageInsts)>; @@ -2548,6 +2630,9 @@ def HasMovrel : Predicate<"Subtarget->hasMovrel()">, def HasFmaMixInsts : Predicate<"Subtarget->hasFmaMixInsts()">, AssemblerPredicate<(all_of FeatureFmaMixInsts)>; +def HasFmaMixBF16Insts : Predicate<"Subtarget->hasFmaMixBF16Insts()">, + AssemblerPredicate<(all_of FeatureFmaMixBF16Insts)>; + def HasDLInsts : Predicate<"Subtarget->hasDLInsts()">, AssemblerPredicate<(all_of FeatureDLInsts)>; @@ -2693,6 +2778,9 @@ def HasPseudoScalarTrans : Predicate<"Subtarget->hasPseudoScalarTrans()">, def HasBitOp3Insts : Predicate<"Subtarget->hasBitOp3Insts()">, AssemblerPredicate<(all_of FeatureBitOp3Insts)>; +def HasTanhInsts : Predicate<"Subtarget->hasTanhInsts()">, + AssemblerPredicate<(all_of FeatureTanhInsts)>; + def HasTransposeLoadF4F6Insts : Predicate<"Subtarget->hasTransposeLoadF4F6Insts()">, AssemblerPredicate<(all_of FeatureTransposeLoadF4F6Insts)>; @@ -2743,12 +2831,21 @@ def HasScalarDwordx3Loads : Predicate<"Subtarget->hasScalarDwordx3Loads()">; def HasXF32Insts : Predicate<"Subtarget->hasXF32Insts()">, AssemblerPredicate<(all_of FeatureXF32Insts)>; +def HasVmemPrefInsts : Predicate<"Subtarget->hasVmemPrefInsts()">, + AssemblerPredicate<(all_of FeatureVmemPrefInsts)>; + def HasAshrPkInsts : Predicate<"Subtarget->hasAshrPkInsts()">, AssemblerPredicate<(all_of FeatureAshrPkInsts)>; def HasLshlAddU64Inst : Predicate<"Subtarget->hasLshlAddU64Inst()">, AssemblerPredicate<(all_of FeatureLshlAddU64Inst)>; +def HasAddSubU64Insts : Predicate<"Subtarget->hasAddSubU64Insts()">, + AssemblerPredicate<(all_of FeatureAddSubU64Insts)>; + +def HasMadU32Inst : Predicate<"Subtarget->hasMadU32Inst()">, + AssemblerPredicate<(all_of FeatureMadU32Inst)>; + def HasLdsBarrierArriveAtomic : Predicate<"Subtarget->hasLdsBarrierArriveAtomic()">, AssemblerPredicate<(all_of FeatureLdsBarrierArriveAtomic)>; |