// e64 version if available, -1 otherwise
int OpcodeE64 = AMDGPU::getVOPe64(Opcode);
const MCInstrDesc *DescE64 = OpcodeE64 == -1 ? nullptr : &TII->get(OpcodeE64);
+ int InputModifiers[3] = {0};
assert(!DescE64 || DescE64->getNumDefs() == NumDefs);
}
}
- if (DescE64 && !Immediate) {
+ if (Immediate)
+ continue;
+
+ if (DescE64) {
// Test if it makes sense to switch to e64 encoding
unsigned OtherRegClass = DescE64->OpInfo[Op].RegClass;
DescE64 = nullptr;
}
}
+
+ if (!DescE64 && !Promote2e64)
+ continue;
+ if (!Operand.isMachineOpcode())
+ continue;
+ if (Operand.getMachineOpcode() == AMDGPU::FNEG_SI) {
+ Ops.pop_back();
+ Ops.push_back(Operand.getOperand(0));
+ InputModifiers[i] = 1;
+ Promote2e64 = true;
+ if (!DescE64)
+ continue;
+ Desc = DescE64;
+ DescE64 = 0;
+ }
+ else if (Operand.getMachineOpcode() == AMDGPU::FABS_SI) {
+ Ops.pop_back();
+ Ops.push_back(Operand.getOperand(0));
+ InputModifiers[i] = 2;
+ Promote2e64 = true;
+ if (!DescE64)
+ continue;
+ Desc = DescE64;
+ DescE64 = 0;
+ }
}
if (Promote2e64) {
Ops.clear();
for (unsigned i = 0; i < OldOps.size(); ++i) {
// src_modifier
- Ops.push_back(DAG.getTargetConstant(0, MVT::i32));
+ Ops.push_back(DAG.getTargetConstant(InputModifiers[i], MVT::i32));
Ops.push_back(OldOps[i]);
}
// Add the modifier flags while promoting
ret void
}
+; SI-CHECK-LABEL: @fabs_fold
+; SI-CHECK-NOT: V_AND_B32_e32
+; SI-CHECK: V_MUL_F32_e64 v{{[0-9]+}}, s{{[0-9]+}}, |v{{[0-9]+}}|
+define void @fabs_fold(float addrspace(1)* %out, float %in0, float %in1) {
+entry:
+ %0 = call float @fabs(float %in0)
+ %1 = fmul float %0, %in1
+ store float %1, float addrspace(1)* %out
+ ret void
+}
+
declare float @fabs(float ) readnone
declare <2 x float> @llvm.fabs.v2f32(<2 x float> ) readnone
declare <4 x float> @llvm.fabs.v4f32(<4 x float> ) readnone
store float %1, float addrspace(1)* %out
ret void
}
+
+; SI-CHECK-LABEL: @fneg_fold
+; SI-CHECK-NOT: V_XOR_B32
+; SI-CHECK: V_MUL_F32_e64 v{{[0-9]+}}, s{{[0-9]+}}, -v{{[0-9]+}}
+define void @fneg_fold(float addrspace(1)* %out, float %in) {
+entry:
+ %0 = fsub float -0.0, %in
+ %1 = fmul float %0, %in
+ store float %1, float addrspace(1)* %out
+ ret void
+}