[X86] combineX86ShufflesRecursively - peek through one use bitcasts to find additional (free) extract_subvector nodes

Author: RKSimon

llvm/lib/Target/X86/X86ISelLowering.cpp

@@ -41084,12 +41084,17 @@ static SDValue combineX86ShufflesRecursively(
     }
   }
 
-  // Peek through any free extract_subvector nodes back to root size.
-  for (SDValue &Op : Ops)
-    while (Op.getOpcode() == ISD::EXTRACT_SUBVECTOR &&
-           (RootSizeInBits % Op.getOperand(0).getValueSizeInBits()) == 0 &&
-           isNullConstant(Op.getOperand(1)))
-      Op = Op.getOperand(0);
+  // Peek through any free bitcasts/extract_subvector nodes back to root size.
+  for (SDValue &Op : Ops){
+    SDValue BC = Op;
+    if (BC.getOpcode() == ISD::BITCAST && BC.hasOneUse())
+      BC = peekThroughOneUseBitcasts(BC);
+    while (BC.getOpcode() == ISD::EXTRACT_SUBVECTOR &&
+           (RootSizeInBits % BC.getOperand(0).getValueSizeInBits()) == 0 &&
+           isNullConstant(BC.getOperand(1))) {
+      Op = BC = BC.getOperand(0);
+    }
+  }
 
   // Remove unused/repeated shuffle source ops.
   resolveTargetShuffleInputsAndMask(Ops, Mask);

llvm/test/CodeGen/X86/vector-replicaton-i1-mask.ll

@@ -12867,46 +12867,25 @@ define void @mask_replication_factor8_vf8(ptr %in.maskvec, ptr %in.vec, ptr %out
 ; AVX512DQ-NEXT:    vzeroupper
 ; AVX512DQ-NEXT:    retq
 ;
-; AVX512BW-ONLY-LABEL: mask_replication_factor8_vf8:
-; AVX512BW-ONLY:       # %bb.0:
-; AVX512BW-ONLY-NEXT:    kmovq (%rdi), %k0
-; AVX512BW-ONLY-NEXT:    vpmovm2b %k0, %zmm0
-; AVX512BW-ONLY-NEXT:    vpbroadcastq %xmm0, %zmm0
-; AVX512BW-ONLY-NEXT:    vpshufb {{.*#+}} zmm0 = zmm0[0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,18,18,18,18,18,18,18,18,19,19,19,19,19,19,19,19,36,36,36,36,36,36,36,36,37,37,37,37,37,37,37,37,54,54,54,54,54,54,54,54,55,55,55,55,55,55,55,55]
-; AVX512BW-ONLY-NEXT:    vpmovb2m %zmm0, %k1
-; AVX512BW-ONLY-NEXT:    kshiftrq $16, %k1, %k2
-; AVX512BW-ONLY-NEXT:    vmovdqa32 64(%rsi), %zmm0 {%k2} {z}
-; AVX512BW-ONLY-NEXT:    vmovdqa32 (%rsi), %zmm1 {%k1} {z}
-; AVX512BW-ONLY-NEXT:    kshiftrq $48, %k1, %k2
-; AVX512BW-ONLY-NEXT:    vmovdqa32 192(%rsi), %zmm2 {%k2} {z}
-; AVX512BW-ONLY-NEXT:    kshiftrq $32, %k1, %k1
-; AVX512BW-ONLY-NEXT:    vmovdqa32 128(%rsi), %zmm3 {%k1} {z}
-; AVX512BW-ONLY-NEXT:    vmovdqa64 %zmm3, 128(%rdx)
-; AVX512BW-ONLY-NEXT:    vmovdqa64 %zmm2, 192(%rdx)
-; AVX512BW-ONLY-NEXT:    vmovdqa64 %zmm1, (%rdx)
-; AVX512BW-ONLY-NEXT:    vmovdqa64 %zmm0, 64(%rdx)
-; AVX512BW-ONLY-NEXT:    vzeroupper
-; AVX512BW-ONLY-NEXT:    retq
-;
-; AVX512VBMI-ONLY-LABEL: mask_replication_factor8_vf8:
-; AVX512VBMI-ONLY:       # %bb.0:
-; AVX512VBMI-ONLY-NEXT:    kmovq (%rdi), %k0
-; AVX512VBMI-ONLY-NEXT:    vpmovm2b %k0, %zmm0
-; AVX512VBMI-ONLY-NEXT:    vpmovsxbq %xmm0, %zmm0
-; AVX512VBMI-ONLY-NEXT:    vpmovb2m %zmm0, %k1
-; AVX512VBMI-ONLY-NEXT:    kshiftrq $16, %k1, %k2
-; AVX512VBMI-ONLY-NEXT:    vmovdqa32 64(%rsi), %zmm0 {%k2} {z}
-; AVX512VBMI-ONLY-NEXT:    vmovdqa32 (%rsi), %zmm1 {%k1} {z}
-; AVX512VBMI-ONLY-NEXT:    kshiftrq $48, %k1, %k2
-; AVX512VBMI-ONLY-NEXT:    vmovdqa32 192(%rsi), %zmm2 {%k2} {z}
-; AVX512VBMI-ONLY-NEXT:    kshiftrq $32, %k1, %k1
-; AVX512VBMI-ONLY-NEXT:    vmovdqa32 128(%rsi), %zmm3 {%k1} {z}
-; AVX512VBMI-ONLY-NEXT:    vmovdqa64 %zmm3, 128(%rdx)
-; AVX512VBMI-ONLY-NEXT:    vmovdqa64 %zmm2, 192(%rdx)
-; AVX512VBMI-ONLY-NEXT:    vmovdqa64 %zmm1, (%rdx)
-; AVX512VBMI-ONLY-NEXT:    vmovdqa64 %zmm0, 64(%rdx)
-; AVX512VBMI-ONLY-NEXT:    vzeroupper
-; AVX512VBMI-ONLY-NEXT:    retq
+; AVX512BW-LABEL: mask_replication_factor8_vf8:
+; AVX512BW:       # %bb.0:
+; AVX512BW-NEXT:    kmovq (%rdi), %k0
+; AVX512BW-NEXT:    vpmovm2b %k0, %zmm0
+; AVX512BW-NEXT:    vpmovsxbq %xmm0, %zmm0
+; AVX512BW-NEXT:    vpmovb2m %zmm0, %k1
+; AVX512BW-NEXT:    kshiftrq $16, %k1, %k2
+; AVX512BW-NEXT:    vmovdqa32 64(%rsi), %zmm0 {%k2} {z}
+; AVX512BW-NEXT:    vmovdqa32 (%rsi), %zmm1 {%k1} {z}
+; AVX512BW-NEXT:    kshiftrq $48, %k1, %k2
+; AVX512BW-NEXT:    vmovdqa32 192(%rsi), %zmm2 {%k2} {z}
+; AVX512BW-NEXT:    kshiftrq $32, %k1, %k1
+; AVX512BW-NEXT:    vmovdqa32 128(%rsi), %zmm3 {%k1} {z}
+; AVX512BW-NEXT:    vmovdqa64 %zmm3, 128(%rdx)
+; AVX512BW-NEXT:    vmovdqa64 %zmm2, 192(%rdx)
+; AVX512BW-NEXT:    vmovdqa64 %zmm1, (%rdx)
+; AVX512BW-NEXT:    vmovdqa64 %zmm0, 64(%rdx)
+; AVX512BW-NEXT:    vzeroupper
+; AVX512BW-NEXT:    retq
   %src.mask.padded = load <64 x i1>, ptr %in.maskvec, align 64
   %src.mask = shufflevector <64 x i1> %src.mask.padded, <64 x i1> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
   %tgt.mask = shufflevector <8 x i1> %src.mask, <8 x i1> poison, <64 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>

llvm/test/CodeGen/X86/zero_extend_vector_inreg_of_broadcast.ll

@@ -5117,15 +5117,11 @@ define void @vec384_i32_widen_to_i128_factor4_broadcast_to_v3i128_factor3(ptr %i
 ; AVX512BW-FAST-LABEL: vec384_i32_widen_to_i128_factor4_broadcast_to_v3i128_factor3:
 ; AVX512BW-FAST:       # %bb.0:
 ; AVX512BW-FAST-NEXT:    vmovdqa64 (%rdi), %zmm0
-; AVX512BW-FAST-NEXT:    vpmovsxbd {{.*#+}} ymm1 = [0,13,14,15,0,1,2,3]
 ; AVX512BW-FAST-NEXT:    vpaddb (%rsi), %zmm0, %zmm0
-; AVX512BW-FAST-NEXT:    vpermd %zmm0, %zmm1, %zmm1
-; AVX512BW-FAST-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX512BW-FAST-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
-; AVX512BW-FAST-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX512BW-FAST-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm2[2,3,4,5,6,7]
-; AVX512BW-FAST-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
-; AVX512BW-FAST-NEXT:    vpaddb (%rdx), %zmm0, %zmm0
+; AVX512BW-FAST-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX512BW-FAST-NEXT:    vpmovsxbd {{.*#+}} zmm2 = [0,13,14,15,0,21,22,23,0,25,26,27,28,29,30,31]
+; AVX512BW-FAST-NEXT:    vpermi2d %zmm1, %zmm0, %zmm2
+; AVX512BW-FAST-NEXT:    vpaddb (%rdx), %zmm2, %zmm0
 ; AVX512BW-FAST-NEXT:    vmovdqa64 %zmm0, (%rcx)
 ; AVX512BW-FAST-NEXT:    vzeroupper
 ; AVX512BW-FAST-NEXT:    retq