[llvm] [X86] Use GFNI for vXi8 shifts/rotates (PR #89115)

[Simon Pilgrim via llvm-commits]

================
@@ -134,34 +99,25 @@ define <32 x i8> @splatconstant_shl_v32i8(<32 x i8> %a) nounwind {
 define <32 x i8> @splatconstant_lshr_v32i8(<32 x i8> %a) nounwind {
 ; GFNISSE-LABEL: splatconstant_lshr_v32i8:
 ; GFNISSE:       # %bb.0:
-; GFNISSE-NEXT:    psrlw $1, %xmm0
-; GFNISSE-NEXT:    movdqa {{.*#+}} xmm2 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; GFNISSE-NEXT:    pand %xmm2, %xmm0
-; GFNISSE-NEXT:    psrlw $1, %xmm1
-; GFNISSE-NEXT:    pand %xmm2, %xmm1
+; GFNISSE-NEXT:    movdqa {{.*#+}} xmm2 = [145249953336295424,145249953336295424]
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm2, %xmm0
+; GFNISSE-NEXT:    gf2p8affineqb $0, %xmm2, %xmm1
 ; GFNISSE-NEXT:    retq
 ;
 ; GFNIAVX1-LABEL: splatconstant_lshr_v32i8:
 ; GFNIAVX1:       # %bb.0:
-; GFNIAVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
-; GFNIAVX1-NEXT:    vpsrlw $1, %xmm1, %xmm1
-; GFNIAVX1-NEXT:    vbroadcastss {{.*#+}} xmm2 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
-; GFNIAVX1-NEXT:    vpand %xmm2, %xmm1, %xmm1
-; GFNIAVX1-NEXT:    vpsrlw $1, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vpand %xmm2, %xmm0, %xmm0
-; GFNIAVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; GFNIAVX1-NEXT:    vgf2p8affineqb $0, {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
 ; GFNIAVX1-NEXT:    retq
 ;
 ; GFNIAVX2-LABEL: splatconstant_lshr_v32i8:
 ; GFNIAVX2:       # %bb.0:
-; GFNIAVX2-NEXT:    vpsrlw $1, %ymm0, %ymm0
-; GFNIAVX2-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
----------------
RKSimon wrote:

gfni + broadcast codegen will be faster

https://github.com/llvm/llvm-project/pull/89115