[llvm] [X86] Fold VPERMV3(WIDEN(X),M,WIDEN(Y)) -> VPERMV(CONCAT(X,Y),M') iff the CONCAT is free (PR #122750)

[Phoebe Wang via llvm-commits]

================
@@ -460,11 +458,9 @@ define <4 x double> @PR34175(ptr %p) {
 ;
 ; AVX512VBMI-LABEL: PR34175:
 ; AVX512VBMI:       # %bb.0:
-; AVX512VBMI-NEXT:    vmovq {{.*#+}} xmm0 = [0,8,32,40,0,0,0,0]
-; AVX512VBMI-NEXT:    vmovdqu (%rdi), %ymm1
-; AVX512VBMI-NEXT:    vmovdqu 32(%rdi), %ymm2
-; AVX512VBMI-NEXT:    vpermt2w %zmm2, %zmm0, %zmm1
-; AVX512VBMI-NEXT:    vpmovzxwd {{.*#+}} xmm0 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero
+; AVX512VBMI-NEXT:    vmovq {{.*#+}} xmm0 = [0,8,16,24,0,0,0,0]
+; AVX512VBMI-NEXT:    vpermw (%rdi), %zmm0, %zmm0
+; AVX512VBMI-NEXT:    vpmovzxwd {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
----------------
phoebewang wrote:

Seems AVX512VBMI and AVX512VBMIVL are identical in the file now.

https://github.com/llvm/llvm-project/pull/122750