[llvm] [X86] LowerCTTZ - prefer legal CTPOP expansion vs GFNI to avoid constant pool load (PR #191623)

[via llvm-commits]

================
@@ -146,6 +170,14 @@ define <32 x i8> @testv32i8u(<32 x i8> %in) nounwind {
 ; GFNIAVX512-NEXT:    vpand %ymm1, %ymm0, %ymm0
 ; GFNIAVX512-NEXT:    vgf2p8affineqb $8, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to4}, %ymm0, %ymm0
 ; GFNIAVX512-NEXT:    retq
+;
+; GFNIAVX512BITALG-LABEL: testv32i8u:
+; GFNIAVX512BITALG:       # %bb.0:
+; GFNIAVX512BITALG-NEXT:    vpcmpeqd %ymm1, %ymm1, %ymm1
----------------
Andarwinux wrote:

#191776

https://github.com/llvm/llvm-project/pull/191623