[llvm] [VectorCombine] Add foldShuffleToIdentity (PR #88693)

[David Green via llvm-commits]

================
@@ -192,11 +164,9 @@ define <8 x half> @splatandidentity(<8 x half> %a, <8 x half> %b) {
 
 define <8 x half> @splattwice(<8 x half> %a, <8 x half> %b) {
 ; CHECK-LABEL: @splattwice(
-; CHECK-NEXT:    [[AS:%.*]] = shufflevector <8 x half> [[A:%.*]], <8 x half> poison, <4 x i32> zeroinitializer
-; CHECK-NEXT:    [[BS:%.*]] = shufflevector <8 x half> [[B:%.*]], <8 x half> poison, <4 x i32> zeroinitializer
-; CHECK-NEXT:    [[AB1:%.*]] = fadd <4 x half> [[AS]], [[BS]]
-; CHECK-NEXT:    [[AB2:%.*]] = fadd <4 x half> [[AS]], [[BS]]
-; CHECK-NEXT:    [[R:%.*]] = shufflevector <4 x half> [[AB1]], <4 x half> [[AB2]], <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <8 x half> [[B:%.*]], <8 x half> poison, <8 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <8 x half> [[A:%.*]], <8 x half> poison, <8 x i32> zeroinitializer
+; CHECK-NEXT:    [[R:%.*]] = fadd <8 x half> [[TMP2]], [[TMP1]]
----------------
davemgreen wrote:

All these tests were just added as examples for testing this combine, many of the smaller ones like this are already optimized by instcombine. They were only added to make sure we were testing the different cases that can come up when folding the shuffles to identities.

https://github.com/llvm/llvm-project/pull/88693