diff options
| author | Stanislav Mekhanoshin <Stanislav.Mekhanoshin@amd.com> | 2019-07-09 22:42:24 +0000 |
|---|---|---|
| committer | Stanislav Mekhanoshin <Stanislav.Mekhanoshin@amd.com> | 2019-07-09 22:42:24 +0000 |
| commit | 1e9eae95afab696c7eefa04b102ca4422612c96b (patch) | |
| tree | 06236114aaee48bc22564aad3b55885377e206d0 /llvm/lib | |
| parent | 67a06d9495479882c89c7447b398d8eb6f48573c (diff) | |
| download | bcm5719-llvm-1e9eae95afab696c7eefa04b102ca4422612c96b.tar.gz bcm5719-llvm-1e9eae95afab696c7eefa04b102ca4422612c96b.zip | |
[AMDGPU] gfx908 v_pk_fmac_f16 support
Differential Revision: https://reviews.llvm.org/D64433
llvm-svn: 365573
Diffstat (limited to 'llvm/lib')
| -rw-r--r-- | llvm/lib/Target/AMDGPU/AMDGPU.td | 4 | ||||
| -rw-r--r-- | llvm/lib/Target/AMDGPU/VOP2Instructions.td | 10 |
2 files changed, 10 insertions, 4 deletions
diff --git a/llvm/lib/Target/AMDGPU/AMDGPU.td b/llvm/lib/Target/AMDGPU/AMDGPU.td index 47e64218ac6..baeba534012 100644 --- a/llvm/lib/Target/AMDGPU/AMDGPU.td +++ b/llvm/lib/Target/AMDGPU/AMDGPU.td @@ -666,8 +666,8 @@ def FeatureGFX10 : GCNSubtargetFeatureGeneration<"GFX10", FeatureFlatInstOffsets, FeatureFlatGlobalInsts, FeatureFlatScratchInsts, FeatureAddNoCarryInsts, FeatureFmaMixInsts, FeatureGFX8Insts, FeatureNoSdstCMPX, FeatureVscnt, FeatureRegisterBanking, - FeatureVOP3Literal, FeatureDPP8, FeatureNoDataDepHazard, - FeatureDoesNotSupportSRAMECC + FeatureVOP3Literal, FeatureDPP8, + FeatureNoDataDepHazard, FeaturePkFmacF16Inst, FeatureDoesNotSupportSRAMECC ] >; diff --git a/llvm/lib/Target/AMDGPU/VOP2Instructions.td b/llvm/lib/Target/AMDGPU/VOP2Instructions.td index 1465f6ba209..e4f38229a3b 100644 --- a/llvm/lib/Target/AMDGPU/VOP2Instructions.td +++ b/llvm/lib/Target/AMDGPU/VOP2Instructions.td @@ -694,10 +694,12 @@ let Constraints = "$vdst = $src2", defm V_FMAC_F16 : VOP2Inst <"v_fmac_f16", VOP_MAC_F16>; } -defm V_PK_FMAC_F16 : VOP2Inst<"v_pk_fmac_f16", VOP_V2F16_V2F16_V2F16>; - } // End SubtargetPredicate = isGFX10Plus +let SubtargetPredicate = HasPkFmacF16Inst in { +defm V_PK_FMAC_F16 : VOP2Inst<"v_pk_fmac_f16", VOP_V2F16_V2F16_V2F16>; +} // End SubtargetPredicate = HasPkFmacF16Inst + // Note: 16-bit instructions produce a 0 result in the high 16-bits // on GFX8 and GFX9 and preserve high 16 bits on GFX10+ def ClearHI16 : OutPatFrag<(ops node:$op), @@ -1548,3 +1550,7 @@ let SubtargetPredicate = HasDot5Insts in { let SubtargetPredicate = HasDot6Insts in { defm V_DOT4C_I32_I8 : VOP2_Real_DOT_ACC_gfx10<0x0d>; } + +let SubtargetPredicate = HasPkFmacF16Inst in { +defm V_PK_FMAC_F16 : VOP2_Real_e32_vi<0x3c>; +} // End SubtargetPredicate = HasPkFmacF16Inst |
