[X86][AVX] Add X86ISD::VALIGN target shuffle decode support

Allows us to combine VALIGN instructions with other shuffles - the combiner doesn't create VALIGN yet though.

llvm/lib/Target/X86/X86ISelLowering.cpp

@@ -4752,6 +4752,7 @@ static bool isTargetShuffle(unsigned Opcode) {
   case X86ISD::INSERTPS:
   case X86ISD::EXTRQI:
   case X86ISD::INSERTQI:
+  case X86ISD::VALIGN:
   case X86ISD::PALIGNR:
   case X86ISD::VSHLDQ:
   case X86ISD::VSRLDQ:
@@ -6732,6 +6733,17 @@ static bool getTargetShuffleMask(SDNode *N, MVT VT, bool AllowSentinelZero,
     DecodeMOVLHPSMask(NumElems, Mask);
     IsUnary = IsFakeUnary = N->getOperand(0) == N->getOperand(1);
     break;
+  case X86ISD::VALIGN:
+    assert((VT.getScalarType() == MVT::i32 || VT.getScalarType() == MVT::i64) &&
+           "Only 32-bit and 64-bit elements are supported!");
+    assert(N->getOperand(0).getValueType() == VT && "Unexpected value type");
+    assert(N->getOperand(1).getValueType() == VT && "Unexpected value type");
+    ImmN = N->getConstantOperandVal(N->getNumOperands() - 1);
+    DecodeVALIGNMask(NumElems, ImmN, Mask);
+    IsUnary = IsFakeUnary = N->getOperand(0) == N->getOperand(1);
+    Ops.push_back(N->getOperand(1));
+    Ops.push_back(N->getOperand(0));
+    break;
   case X86ISD::PALIGNR:
     assert(VT.getScalarType() == MVT::i8 && "Byte vector expected");
     assert(N->getOperand(0).getValueType() == VT && "Unexpected value type");
@@ -47251,6 +47263,7 @@ SDValue X86TargetLowering::PerformDAGCombine(SDNode *N,
   case X86ISD::INSERTPS:
   case X86ISD::EXTRQI:
   case X86ISD::INSERTQI:
+  case X86ISD::VALIGN:
   case X86ISD::PALIGNR:
   case X86ISD::VSHLDQ:
   case X86ISD::VSRLDQ:

llvm/test/CodeGen/X86/vector-shuffle-128-v4.ll

@@ -1583,11 +1583,18 @@ define <4 x i32> @shuffle_v4i32_2456(<4 x i32> %a, <4 x i32> %b) {
 ; SSE41-NEXT:    movdqa %xmm1, %xmm0
 ; SSE41-NEXT:    retq
 ;
-; AVX-LABEL: shuffle_v4i32_2456:
+; AVX1OR2-LABEL: shuffle_v4i32_2456:
-; AVX:       # %bb.0:
+; AVX1OR2:       # %bb.0:
-; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,1,2,2]
+; AVX1OR2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,1,2,2]
-; AVX-NEXT:    vpalignr {{.*#+}} xmm0 = xmm0[12,13,14,15],xmm1[0,1,2,3,4,5,6,7,8,9,10,11]
+; AVX1OR2-NEXT:    vpalignr {{.*#+}} xmm0 = xmm0[12,13,14,15],xmm1[0,1,2,3,4,5,6,7,8,9,10,11]
-; AVX-NEXT:    retq
+; AVX1OR2-NEXT:    retq
+;
+; AVX512VL-LABEL: shuffle_v4i32_2456:
+; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm2 = [6,0,1,2]
+; AVX512VL-NEXT:    vpermi2d %xmm0, %xmm1, %xmm2
+; AVX512VL-NEXT:    vmovdqa %xmm2, %xmm0
+; AVX512VL-NEXT:    retq
   %s1 = shufflevector <4 x i32> %a, <4 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 2>
   %s2 = shufflevector <4 x i32> %s1, <4 x i32> %b, <4 x i32> <i32 3, i32 4, i32 5, i32 6>
   ret <4 x i32> %s2

llvm/test/CodeGen/X86/vector-shuffle-256-v8.ll

@@ -2524,11 +2524,18 @@ define <8 x i32> @shuffle_v8i32_089abcde(<8 x i32> %a, <8 x i32> %b) {
 ; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3,4,5,6,7]
 ; AVX2-NEXT:    retq
 ;
-; AVX512VL-LABEL: shuffle_v8i32_089abcde:
+; AVX512VL-SLOW-LABEL: shuffle_v8i32_089abcde:
-; AVX512VL:       # %bb.0:
+; AVX512VL-SLOW:       # %bb.0:
-; AVX512VL-NEXT:    valignd {{.*#+}} ymm1 = ymm1[7,0,1,2,3,4,5,6]
+; AVX512VL-SLOW-NEXT:    valignd {{.*#+}} ymm1 = ymm1[7,0,1,2,3,4,5,6]
-; AVX512VL-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3,4,5,6,7]
+; AVX512VL-SLOW-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3,4,5,6,7]
-; AVX512VL-NEXT:    retq
+; AVX512VL-SLOW-NEXT:    retq
+;
+; AVX512VL-FAST-LABEL: shuffle_v8i32_089abcde:
+; AVX512VL-FAST:       # %bb.0:
+; AVX512VL-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [8,0,1,2,3,4,5,6]
+; AVX512VL-FAST-NEXT:    vpermi2d %ymm0, %ymm1, %ymm2
+; AVX512VL-FAST-NEXT:    vmovdqa %ymm2, %ymm0
+; AVX512VL-FAST-NEXT:    retq
   %shuffle = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14>
   ret <8 x i32> %shuffle
 }

llvm/test/CodeGen/X86/vector-shuffle-v1.ll

@@ -62,8 +62,9 @@ define <2 x i1> @shuf2i1_1_2(<2 x i1> %a) {
 ; AVX512VL-NEXT:    vmovdqa64 %xmm0, %xmm1 {%k1} {z}
 ; AVX512VL-NEXT:    movq $-1, %rax
 ; AVX512VL-NEXT:    vmovq %rax, %xmm2
-; AVX512VL-NEXT:    vpalignr {{.*#+}} xmm1 = xmm1[8,9,10,11,12,13,14,15],xmm2[0,1,2,3,4,5,6,7]
+; AVX512VL-NEXT:    vmovdqa {{.*#+}} xmm3 = [1,2]
-; AVX512VL-NEXT:    vptestmq %xmm1, %xmm1, %k1
+; AVX512VL-NEXT:    vpermi2q %xmm2, %xmm1, %xmm3
+; AVX512VL-NEXT:    vptestmq %xmm3, %xmm3, %k1
 ; AVX512VL-NEXT:    vmovdqa64 %xmm0, %xmm0 {%k1} {z}
 ; AVX512VL-NEXT:    retq
 ;
@@ -74,8 +75,9 @@ define <2 x i1> @shuf2i1_1_2(<2 x i1> %a) {
 ; VL_BW_DQ-NEXT:    movq $-1, %rax
 ; VL_BW_DQ-NEXT:    vmovq %rax, %xmm0
 ; VL_BW_DQ-NEXT:    vpmovm2q %k0, %xmm1
-; VL_BW_DQ-NEXT:    vpalignr {{.*#+}} xmm0 = xmm1[8,9,10,11,12,13,14,15],xmm0[0,1,2,3,4,5,6,7]
+; VL_BW_DQ-NEXT:    vmovdqa {{.*#+}} xmm2 = [1,2]
-; VL_BW_DQ-NEXT:    vpmovq2m %xmm0, %k0
+; VL_BW_DQ-NEXT:    vpermi2q %xmm0, %xmm1, %xmm2
+; VL_BW_DQ-NEXT:    vpmovq2m %xmm2, %k0
 ; VL_BW_DQ-NEXT:    vpmovm2q %k0, %xmm0
 ; VL_BW_DQ-NEXT:    retq
   %b = shufflevector <2 x i1> %a, <2 x i1> <i1 1, i1 0>, <2 x i32> <i32 1, i32 2>