diff options
Diffstat (limited to 'llvm/lib/Target/AMDGPU/SIFoldOperands.cpp')
| -rw-r--r-- | llvm/lib/Target/AMDGPU/SIFoldOperands.cpp | 13 |
1 files changed, 12 insertions, 1 deletions
diff --git a/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp b/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp index 5c411a095587..7bf6a635158e 100644 --- a/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp +++ b/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp @@ -1519,6 +1519,9 @@ const MachineOperand *SIFoldOperands::isClamp(const MachineInstr &MI) const { case AMDGPU::V_MAX_F64_e64: case AMDGPU::V_MAX_NUM_F64_e64: case AMDGPU::V_PK_MAX_F16: { + if (MI.mayRaiseFPException()) + return nullptr; + if (!TII->getNamedOperand(MI, AMDGPU::OpName::clamp)->getImm()) return nullptr; @@ -1565,6 +1568,9 @@ bool SIFoldOperands::tryFoldClamp(MachineInstr &MI) { if (TII->getClampMask(*Def) != TII->getClampMask(MI)) return false; + if (Def->mayRaiseFPException()) + return false; + MachineOperand *DefClamp = TII->getNamedOperand(*Def, AMDGPU::OpName::clamp); if (!DefClamp) return false; @@ -1650,7 +1656,9 @@ SIFoldOperands::isOMod(const MachineInstr &MI) const { ((Op == AMDGPU::V_MUL_F64_e64 || Op == AMDGPU::V_MUL_F64_pseudo_e64 || Op == AMDGPU::V_MUL_F16_e64 || Op == AMDGPU::V_MUL_F16_t16_e64 || Op == AMDGPU::V_MUL_F16_fake16_e64) && - MFI->getMode().FP64FP16Denormals.Output != DenormalMode::PreserveSign)) + MFI->getMode().FP64FP16Denormals.Output != + DenormalMode::PreserveSign) || + MI.mayRaiseFPException()) return std::pair(nullptr, SIOutMods::NONE); const MachineOperand *RegOp = nullptr; @@ -1725,6 +1733,9 @@ bool SIFoldOperands::tryFoldOMod(MachineInstr &MI) { if (!DefOMod || DefOMod->getImm() != SIOutMods::NONE) return false; + if (Def->mayRaiseFPException()) + return false; + // Clamp is applied after omod. If the source already has clamp set, don't // fold it. if (TII->hasModifiersSet(*Def, AMDGPU::OpName::clamp)) |