[llvm] [Transforms] Let amdgcn take advantage of sin(-x) --> -sin(x) (PR #79700)

[Matt Arsenault via llvm-commits]

================
@@ -1023,6 +1023,47 @@ define float @cos_fabs_unary_fneg_f32(float %x) {
   ret float %cos
 }
 
+
+; --------------------------------------------------------------------
+; llvm.amdgcn.sin
+; --------------------------------------------------------------------
+declare float @llvm.amdgcn.sin.f32(float) nounwind readnone
+
+define float @sin_fneg_f32(float %x) {
+; CHECK-LABEL: @sin_fneg_f32(
+; CHECK-NEXT:    [[X_FNEG:%.*]] = fsub float 0.000000e+00, [[X:%.*]]
+; CHECK-NEXT:    [[SIN:%.*]] = call float @llvm.amdgcn.sin.f32(float [[X_FNEG]])
+; CHECK-NEXT:    ret float [[SIN]]
+;
+  %x.fneg = fsub float 0.0, %x
+  %sin = call float @llvm.amdgcn.sin.f32(float %x.fneg)
+  ret float %sin
+}
+
+define float @sin_fabs_f32(float %x) {
+; CHECK-LABEL: @sin_fabs_f32(
+; CHECK-NEXT:    [[X_FABS:%.*]] = call float @llvm.fabs.f32(float [[X:%.*]])
+; CHECK-NEXT:    [[SIN:%.*]] = call float @llvm.amdgcn.sin.f32(float [[X_FABS]])
+; CHECK-NEXT:    ret float [[SIN]]
+;
+  %x.fabs = call float @llvm.fabs.f32(float %x)
+  %sin = call float @llvm.amdgcn.sin.f32(float %x.fabs)
+  ret float %sin
+}
+
+define float @sin_fabs_fneg_f32(float %x) {
+; CHECK-LABEL: @sin_fabs_fneg_f32(
+; CHECK-NEXT:    [[X_FABS:%.*]] = call float @llvm.fabs.f32(float [[X:%.*]])
+; CHECK-NEXT:    [[TMP1:%.*]] = call float @llvm.amdgcn.sin.f32(float [[X_FABS]])
+; CHECK-NEXT:    [[SIN:%.*]] = fneg float [[TMP1]]
+; CHECK-NEXT:    ret float [[SIN]]
+;
+  %x.fabs = call float @llvm.fabs.f32(float %x)
+  %x.fabs.fneg = fsub float -0.0, %x.fabs
----------------
arsenm wrote:

```suggestion
  %x.fabs.fneg = fneg float %x.fabs
```

https://github.com/llvm/llvm-project/pull/79700