[X86] Prefer AND over PSHUFB for v64i8 when possible

[craig.topper]

Summary

If the shuffle is a blend and one input is a 0 vector, we should prefer AND over PSHUFB since its available on more execution ports.

[RKSimon]

LGTM

llvm/lib/Target/X86/X86ISelLowering.cpp

if (SDValue Rotate = lowerShuffleAsByteRotate(DL, MVT::v64i8, V1, V2, Mask,
     return Rotate;
 
   // Try to use bit rotation instructions.
   if (V2.isUndef())
     if (SDValue Rotate =
             lowerShuffleAsBitRotate(DL, MVT::v64i8, V1, Mask, Subtarget, DAG))
       return Rotate;
 
+  // Lower as AND if possible.
+  if (SDValue Masked = lowerShuffleAsBitMask(DL, MVT::v64i8, V1, V2, Mask,
+                                             Zeroable, Subtarget, DAG))
+    return Masked;
+
   if (SDValue PSHUFB = lowerShuffleWithPSHUFB(DL, MVT::v64i8, Mask, V1, V2,
                                               Zeroable, Subtarget, DAG))
     return PSHUFB;
 
   // VBMI can use VPERMV/VPERMV3 byte shuffles.
   if (Subtarget.hasVBMI())
     return lowerShuffleWithPERMV(DL, MVT::v64i8, Mask, V1, V2, DAG);
 

llvm/test/CodeGen/X86/avx512-mask-op.ll

 ; KNL-NEXT:    vandps {{.*}}(%rip), %ymm0, %ymm1
 ; KNL-NEXT:    vextractf64x4 $1, %zmm0, %ymm0
 ; KNL-NEXT:    vandps {{.*}}(%rip), %ymm0, %ymm0
 ; KNL-NEXT:    vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
 ; KNL-NEXT:    retq
 ;
 ; SKX-LABEL: test_build_vec_v64i1:
 ; SKX:       ## %bb.0:
-; SKX-NEXT:    vpshufb {{.*#+}} zmm0 = zero,zero,zmm0[2],zero,zero,zero,zmm0[6],zero,zmm0[8],zero,zmm0[10],zero,zmm0[12],zero,zero,zmm0[15],zero,zero,zmm0[18],zero,zmm0[20],zero,zmm0[22],zero,zmm0[24],zero,zero,zmm0[27],zero,zero,zmm0[30],zero,zmm0[32],zero,zmm0[34],zero,zero,zero,zmm0[38],zero,zmm0[40],zero,zero,zmm0[43,44],zero,zmm0[46],zero,zmm0[48],zero,zmm0[50],zero,zero,zero,zmm0[54],zero,zmm0[56],zero,zero,zmm0[59,60],zero,zmm0[62],zero
+; SKX-NEXT:    vandps {{.*}}(%rip), %zmm0, %zmm0
 ; SKX-NEXT:    retq
 ;
 ; AVX512BW-LABEL: test_build_vec_v64i1:
 ; AVX512BW:       ## %bb.0:
-; AVX512BW-NEXT:    vpshufb {{.*#+}} zmm0 = zero,zero,zmm0[2],zero,zero,zero,zmm0[6],zero,zmm0[8],zero,zmm0[10],zero,zmm0[12],zero,zero,zmm0[15],zero,zero,zmm0[18],zero,zmm0[20],zero,zmm0[22],zero,zmm0[24],zero,zero,zmm0[27],zero,zero,zmm0[30],zero,zmm0[32],zero,zmm0[34],zero,zero,zero,zmm0[38],zero,zmm0[40],zero,zero,zmm0[43,44],zero,zmm0[46],zero,zmm0[48],zero,zmm0[50],zero,zero,zero,zmm0[54],zero,zmm0[56],zero,zero,zmm0[59,60],zero,zmm0[62],zero
+; AVX512BW-NEXT:    vpandq {{.*}}(%rip), %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: test_build_vec_v64i1:
 ; AVX512DQ:       ## %bb.0:
 ; AVX512DQ-NEXT:    vandps {{.*}}(%rip), %ymm0, %ymm1
 ; AVX512DQ-NEXT:    vextractf64x4 $1, %zmm0, %ymm0
 ; AVX512DQ-NEXT:    vandps {{.*}}(%rip), %ymm0, %ymm0
 ; AVX512DQ-NEXT:    vinsertf64x4 $1, %ymm0, %zmm1, %zmm0
 ; AVX512DQ-NEXT:    retq
 ;
 ; X86-LABEL: test_build_vec_v64i1:
 ; X86:       ## %bb.0:
-; X86-NEXT:    vpshufb {{.*#+}} zmm0 = zero,zero,zmm0[2],zero,zero,zero,zmm0[6],zero,zmm0[8],zero,zmm0[10],zero,zmm0[12],zero,zero,zmm0[15],zero,zero,zmm0[18],zero,zmm0[20],zero,zmm0[22],zero,zmm0[24],zero,zero,zmm0[27],zero,zero,zmm0[30],zero,zmm0[32],zero,zmm0[34],zero,zero,zero,zmm0[38],zero,zmm0[40],zero,zero,zmm0[43,44],zero,zmm0[46],zero,zmm0[48],zero,zmm0[50],zero,zero,zero,zmm0[54],zero,zmm0[56],zero,zero,zmm0[59,60],zero,zmm0[62],zero
+; X86-NEXT:    vandps LCPI43_0, %zmm0, %zmm0
 ; X86-NEXT:    retl
   %ret = select <64 x i1> <i1 false, i1 false, i1 true, i1 false, i1 false, i1 false, i1 true, i1 false, i1 true, i1 false, i1 true, i1 false, i1 true, i1 false, i1 false, i1 true, i1 false, i1 false, i1 true, i1 false, i1 true, i1 false, i1 true, i1 false, i1 true, i1 false, i1 false, i1 true, i1 false, i1 false, i1 true, i1 false, i1 true, i1 false, i1 true, i1 false, i1 false, i1 false, i1 true, i1 false, i1 true, i1 false, i1 false, i1 true, i1 true, i1 false, i1 true, i1 false, i1 true, i1 false, i1 true, i1 false, i1 false, i1 false, i1 true, i1 false, i1 true, i1 false, i1 false, i1 true, i1 true, i1 false, i1 true, i1 false>, <64 x i8> %x, <64 x i8> zeroinitializer
   ret <64 x i8> %ret
 }
 
 define void @ktest_1(<8 x double> %in, double * %base) {
 ; KNL-LABEL: ktest_1:
 ; KNL:       ## %bb.0:

llvm/test/CodeGen/X86/vector-shuffle-512-v64.ll

 ; AVX512F-LABEL: shuffle_v64i8_0zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vmovaps {{.*#+}} xmm1 = [255,0,0,0]
 ; AVX512F-NEXT:    vandps %ymm1, %ymm0, %ymm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512BW-LABEL: shuffle_v64i8_0zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpshufb {{.*#+}} zmm0 = zmm0[0],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm1 = [255,0,0,0]
+; AVX512BW-NEXT:    vpandq %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: shuffle_v64i8_0zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    vmovaps {{.*#+}} xmm1 = [255,0,0,0]
 ; AVX512DQ-NEXT:    vandps %ymm1, %ymm0, %ymm0
 ; AVX512DQ-NEXT:    retq
 ;
 ; AVX512VBMI-LABEL: shuffle_v64i8_0zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz:
 ; AVX512VBMI:       # %bb.0:
-; AVX512VBMI-NEXT:    vpshufb {{.*#+}} zmm0 = zmm0[0],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
+; AVX512VBMI-NEXT:    vmovdqa {{.*#+}} xmm1 = [255,0,0,0]
+; AVX512VBMI-NEXT:    vpandq %zmm1, %zmm0, %zmm0
 ; AVX512VBMI-NEXT:    retq
   %shuffle = shufflevector <64 x i8> %a, <64 x i8> zeroinitializer, <64 x i32> <i32 0, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64>
   ret <64 x i8> %shuffle
 }
 
 define <64 x i8> @shuffle_v64i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00(<64 x i8> %a, <64 x i8> %b) {
 ; AVX512F-LABEL: shuffle_v64i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
 ; AVX512F:       # %bb.0: