[llvm] [X86] X86DAGToDAGISel - attempt to merge XMM/YMM loads with YMM/ZMM loads of the same ptr (PR #73126)

[Simon Pilgrim via llvm-commits]

================
@@ -209,8 +209,9 @@ define <4 x i64> @ext_i4_4i64(i4 %a0) {
 ; AVX1-NEXT:    vmovd %edi, %xmm0
 ; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,1,0,1]
 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
-; AVX1-NEXT:    vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
-; AVX1-NEXT:    vpcmpeqq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm1
+; AVX1-NEXT:    vmovaps {{.*#+}} ymm1 = [1,2,4,8]
+; AVX1-NEXT:    vandps %ymm1, %ymm0, %ymm0
+; AVX1-NEXT:    vpcmpeqq %xmm1, %xmm0, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
 ; AVX1-NEXT:    vpcmpeqq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
----------------
RKSimon wrote:

This is (re)loading the upper bits [4,8] of the ymm data, technically we could extract the upper subvector, but that would create an extra instruction - is that worth it do you think?

https://github.com/llvm/llvm-project/pull/73126