diff options
| author | Stanislav Mekhanoshin <Stanislav.Mekhanoshin@amd.com> | 2025-08-07 12:59:50 -0700 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2025-08-07 12:59:50 -0700 |
| commit | d09dbdabb93ffdd6df25ae487c95a552f13e5e16 (patch) | |
| tree | 1d74ca8b8acc501386a3aae07d1ab15f5e8e713f /llvm/lib | |
| parent | c43c1c0c45fc1ec3fab7abd6e19b318f6468bf28 (diff) | |
| download | llvm-d09dbdabb93ffdd6df25ae487c95a552f13e5e16.zip llvm-d09dbdabb93ffdd6df25ae487c95a552f13e5e16.tar.gz llvm-d09dbdabb93ffdd6df25ae487c95a552f13e5e16.tar.bz2 | |
[AMDGPU] bf16 clamp folding (#152573)
Diffstat (limited to 'llvm/lib')
| -rw-r--r-- | llvm/lib/Target/AMDGPU/SIFoldOperands.cpp | 10 |
1 files changed, 7 insertions, 3 deletions
diff --git a/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp b/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp index 0c653b1..962c276 100644 --- a/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp +++ b/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp @@ -2081,7 +2081,9 @@ SIFoldOperandsImpl::isClamp(const MachineInstr &MI) const { case AMDGPU::V_MAX_F16_fake16_e64: case AMDGPU::V_MAX_F64_e64: case AMDGPU::V_MAX_NUM_F64_e64: - case AMDGPU::V_PK_MAX_F16: { + case AMDGPU::V_PK_MAX_F16: + case AMDGPU::V_MAX_BF16_PSEUDO_e64: + case AMDGPU::V_PK_MAX_NUM_BF16: { if (MI.mayRaiseFPException()) return nullptr; @@ -2108,8 +2110,10 @@ SIFoldOperandsImpl::isClamp(const MachineInstr &MI) const { // Having a 0 op_sel_hi would require swizzling the output in the source // instruction, which we can't do. - unsigned UnsetMods = (Op == AMDGPU::V_PK_MAX_F16) ? SISrcMods::OP_SEL_1 - : 0u; + unsigned UnsetMods = + (Op == AMDGPU::V_PK_MAX_F16 || Op == AMDGPU::V_PK_MAX_NUM_BF16) + ? SISrcMods::OP_SEL_1 + : 0u; if (Src0Mods != UnsetMods && Src1Mods != UnsetMods) return nullptr; return Src0; |