[llvm] [SimplifyLibCalls] Shrink llvm.sincos.f64 to llvm.sincos.f32 (PR #211218)

[Matt Arsenault via llvm-commits]

================
@@ -134,6 +134,64 @@ define { float, float } @sin_cos_pair_shrink(float %f) {
   ret { float, float } %r1
 }
 
+define { <2 x float>, <2 x float> } @sincos_shrink_vector(<2 x float> %f) {
+; CHECK-LABEL: define { <2 x float>, <2 x float> } @sincos_shrink_vector(
+; CHECK-SAME: <2 x float> [[F:%.*]]) {
+; CHECK-NEXT:    [[TMP1:%.*]] = call fast { <2 x float>, <2 x float> } @llvm.sincos.v2f32(<2 x float> [[F]])
+; CHECK-NEXT:    ret { <2 x float>, <2 x float> } [[TMP1]]
+;
+  %d = fpext fast <2 x float> %f to <2 x double>
----------------
arsenm wrote:

Restrict to minimal flags needed for the transform (which is none, in which case just have a specific flag preserve test or use an innocuous example)

https://github.com/llvm/llvm-project/pull/211218