Don't combine splats with other shuffles.

[efriedma]

Summary

We often end up creating shuffles which are worse than the obvious translation of the IR.

Depends on https://reviews.llvm.org/D27787 .

Fixes https://llvm.org/bugs/show_bug.cgi?id=31301 .

[RKSimon]

This seems sensible to me, without a concept of "shuffle complexity" to drive the combine probably the best thing to do is wait until target specific combining.

[mkuper]

Looks sensible to me too.

llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp

SDValue DAGCombiner::visitVECTOR_SHUFFLE(SDNode *N) {
   //   shuffle(shuffle(A, B, M0), C, M1) -> shuffle(A, C, M2)
   //   shuffle(shuffle(A, B, M0), C, M1) -> shuffle(B, C, M2)
   // Don't try to fold shuffles with illegal type.
   // Only fold if this shuffle is the only user of the other shuffle.
   if (N0.getOpcode() == ISD::VECTOR_SHUFFLE && N->isOnlyUserOf(N0.getNode()) &&
       Level < AfterLegalizeDAG && TLI.isTypeLegal(VT)) {
     ShuffleVectorSDNode *OtherSV = cast<ShuffleVectorSDNode>(N0);
 
+    // Don't try to fold splats; they're likely to simplify somehow, or they
+    // might be free.
+    if (OtherSV->isSplat())
+      return SDValue();
+
     // The incoming shuffle must be of the same type as the result of the
     // current shuffle.
     assert(OtherSV->getOperand(0).getValueType() == VT &&
            "Shuffle types don't match");
 
     SDValue SV0, SV1;
     SmallVector<int, 4> Mask;
     // Compute the combined shuffle mask for a shuffle with SV0 as the first

llvm/trunk/test/CodeGen/ARM/vzip.ll

entry:
   %0 = shufflevector <8 x i16> %tmp1, <8 x i16> undef, <4 x i32> <i32 4, i32 4, i32 5, i32 3>
   store <4 x i16> %0, <4 x i16>* %B
   ret void
 }
 
 define <8 x i8> @vdup_zip(i8* nocapture readonly %x, i8* nocapture readonly %y)  {
 entry:
   ; CHECK-LABEL: vdup_zip:
-  ; CHECK: ldrb
-  ; CHECK-NEXT: ldrb
-  ; CHECK-NEXT: vmov.i16 d{{.*}}, #0x800
-  ; CHECK-NEXT: vmov.8
-  ; CHECK-NEXT: vmov.8
-  ; CHECK-NEXT: vtbl.8
-  ; CHECK-NEXT: vmov    r0, r1
+  ; CHECK: vld1.8
+  ; CHECK-NEXT: vld1.8
+  ; CHECK-NEXT: vzip.8
+  ; CHECK-NEXT: vmov r0, r1
   %0 = load i8, i8* %x, align 1
   %1 = insertelement <8 x i8> undef, i8 %0, i32 0
   %lane = shufflevector <8 x i8> %1, <8 x i8> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 undef, i32 undef, i32 undef, i32 undef>
   %2 = load i8, i8* %y, align 1
   %3 = insertelement <8 x i8> undef, i8 %2, i32 0
   %lane3 = shufflevector <8 x i8> %3, <8 x i8> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 undef, i32 undef, i32 undef, i32 undef>
   %vzip.i = shufflevector <8 x i8> %lane, <8 x i8> %lane3, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 2, i32 10, i32 3, i32 11>
   ret <8 x i8> %vzip.i
 }

llvm/trunk/test/CodeGen/X86/vector-shuffle-128-v16.ll

; AVX512VL-NEXT:    retq
   ret <16 x i8> %result
 }
 
 define <16 x i8> @PR31301(i8* nocapture readonly %x, i8* nocapture readonly %y)  {
 ; SSE2-LABEL: PR31301:
 ; SSE2:       # BB#0: # %entry
 ; SSE2-NEXT:    movzbl (%rdi), %eax
 ; SSE2-NEXT:    movd %eax, %xmm0
+; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,1,1]
 ; SSE2-NEXT:    movzbl (%rsi), %eax
 ; SSE2-NEXT:    movd %eax, %xmm1
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[0,0,1,1]
-; SSE2-NEXT:    movdqa {{.*#+}} xmm0 = [255,0,255,0,255,0,255,0,255,0,255,0,255,0,255,0]
-; SSE2-NEXT:    pand %xmm0, %xmm2
 ; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
 ; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,0,0,0,4,5,6,7]
 ; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,0,1,1]
-; SSE2-NEXT:    pandn %xmm1, %xmm0
-; SSE2-NEXT:    por %xmm2, %xmm0
+; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
 ; SSE2-NEXT:    retq
 ;
 ; SSSE3-LABEL: PR31301:
 ; SSSE3:       # BB#0: # %entry
 ; SSSE3-NEXT:    movzbl (%rdi), %eax
 ; SSSE3-NEXT:    movd %eax, %xmm0
+; SSSE3-NEXT:    pxor %xmm1, %xmm1
+; SSSE3-NEXT:    pshufb %xmm1, %xmm0
 ; SSSE3-NEXT:    movzbl (%rsi), %eax
-; SSSE3-NEXT:    movd %eax, %xmm1
-; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
-; SSSE3-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
-; SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,1,1]
+; SSSE3-NEXT:    movd %eax, %xmm2
+; SSSE3-NEXT:    pshufb %xmm1, %xmm2
+; SSSE3-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3],xmm0[4],xmm2[4],xmm0[5],xmm2[5],xmm0[6],xmm2[6],xmm0[7],xmm2[7]
 ; SSSE3-NEXT:    retq
 ;
 ; SSE41-LABEL: PR31301:
 ; SSE41:       # BB#0: # %entry
 ; SSE41-NEXT:    movzbl (%rdi), %eax
 ; SSE41-NEXT:    movd %eax, %xmm0
+; SSE41-NEXT:    pxor %xmm1, %xmm1
+; SSE41-NEXT:    pshufb %xmm1, %xmm0
 ; SSE41-NEXT:    movzbl (%rsi), %eax
-; SSE41-NEXT:    movd %eax, %xmm1
-; SSE41-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
-; SSE41-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
-; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,0,1,1]
+; SSE41-NEXT:    movd %eax, %xmm2
+; SSE41-NEXT:    pshufb %xmm1, %xmm2
+; SSE41-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3],xmm0[4],xmm2[4],xmm0[5],xmm2[5],xmm0[6],xmm2[6],xmm0[7],xmm2[7]
 ; SSE41-NEXT:    retq
 ;
 ; AVX1-LABEL: PR31301:
 ; AVX1:       # BB#0: # %entry
 ; AVX1-NEXT:    movzbl (%rdi), %eax
 ; AVX1-NEXT:    vmovd %eax, %xmm0
+; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX1-NEXT:    vpshufb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    movzbl (%rsi), %eax
-; AVX1-NEXT:    vmovd %eax, %xmm1
+; AVX1-NEXT:    vmovd %eax, %xmm2
+; AVX1-NEXT:    vpshufb %xmm1, %xmm2, %xmm1
 ; AVX1-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
-; AVX1-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
-; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,0,1,1]
 ; AVX1-NEXT:    retq
 ;
 ; AVX2OR512VL-LABEL: PR31301:
 ; AVX2OR512VL:       # BB#0: # %entry
-; AVX2OR512VL-NEXT:    movzbl (%rdi), %eax
-; AVX2OR512VL-NEXT:    vmovd %eax, %xmm0
-; AVX2OR512VL-NEXT:    movzbl (%rsi), %eax
-; AVX2OR512VL-NEXT:    vmovd %eax, %xmm1
+; AVX2OR512VL-NEXT:    vpbroadcastb (%rdi), %xmm0
+; AVX2OR512VL-NEXT:    vpbroadcastb (%rsi), %xmm1
 ; AVX2OR512VL-NEXT:    vpunpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
-; AVX2OR512VL-NEXT:    vpbroadcastw %xmm0, %xmm0
 ; AVX2OR512VL-NEXT:    retq
 entry:
   %0 = load i8, i8* %x, align 1
   %1 = insertelement <16 x i8> undef, i8 %0, i32 0
   %lane = shufflevector <16 x i8> %1, <16 x i8> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %2 = load i8, i8* %y, align 1
   %3 = insertelement <16 x i8> undef, i8 %2, i32 0
   %lane3 = shufflevector <16 x i8> %3, <16 x i8> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
   %vzip.i = shufflevector <16 x i8> %lane, <16 x i8> %lane3, <16 x i32> <i32 0, i32 16, i32 1, i32 17, i32 2, i32 18, i32 3, i32 19, i32 4, i32 20, i32 5, i32 21, i32 6, i32 22, i32 7, i32 23>
   ret <16 x i8> %vzip.i
 }