[X86] combineConcatVectorOps - attempt to recursively call combineConcatVectorOps from inside ConcatSubOperand. #131303
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…catVectorOps from inside ConcatSubOperand. Before falling back to creating a generic ISD::CONCAT_VECTORS node, see if we can directly concat the subvectors if we peek through any bitcasts.
|
@llvm/pr-subscribers-backend-x86 Author: Simon Pilgrim (RKSimon) ChangesBefore falling back to creating a generic ISD::CONCAT_VECTORS node, see if we can directly concat the subvectors if we peek through any bitcasts. Patch is 293.59 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/131303.diff 6 Files Affected:
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 3360f199ccfc4..ac39c8fc0cba3 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -57937,11 +57937,14 @@ static SDValue combineConcatVectorOps(const SDLoc &DL, MVT VT,
// Attempt to peek through bitcasts and concat the original subvectors.
EVT SubVT = peekThroughBitcasts(Subs[0]).getValueType();
if (SubVT.isSimple() && SubVT.isVector()) {
- EVT ConcatVT =
- EVT::getVectorVT(Ctx, SubVT.getScalarType(),
+ MVT ConcatVT =
+ MVT::getVectorVT(SubVT.getSimpleVT().getScalarType(),
SubVT.getVectorElementCount() * Subs.size());
for (SDValue &Sub : Subs)
Sub = DAG.getBitcast(SubVT, Sub);
+ if (SDValue ConcatSrc = combineConcatVectorOps(DL, ConcatVT, Subs, DAG,
+ Subtarget, Depth + 1))
+ return DAG.getBitcast(VT, ConcatSrc);
return DAG.getBitcast(
VT, DAG.getNode(ISD::CONCAT_VECTORS, DL, ConcatVT, Subs));
}
diff --git a/llvm/test/CodeGen/X86/vector-interleaved-store-i16-stride-3.ll b/llvm/test/CodeGen/X86/vector-interleaved-store-i16-stride-3.ll
index 77e7397e2fc96..44bd9c03c7ce7 100644
--- a/llvm/test/CodeGen/X86/vector-interleaved-store-i16-stride-3.ll
+++ b/llvm/test/CodeGen/X86/vector-interleaved-store-i16-stride-3.ll
@@ -1653,305 +1653,305 @@ define void @store_i16_stride3_vf32(ptr %in.vecptr0, ptr %in.vecptr1, ptr %in.ve
;
; AVX512-LABEL: store_i16_stride3_vf32:
; AVX512: # %bb.0:
-; AVX512-NEXT: vmovdqa (%rsi), %xmm0
-; AVX512-NEXT: vmovdqa 16(%rsi), %xmm3
-; AVX512-NEXT: vmovdqa 32(%rsi), %xmm2
-; AVX512-NEXT: vprold $16, %xmm2, %xmm4
-; AVX512-NEXT: vmovdqa (%rdi), %xmm1
-; AVX512-NEXT: vmovdqa 16(%rdi), %xmm5
-; AVX512-NEXT: vmovdqa 32(%rdi), %xmm6
-; AVX512-NEXT: vpshufd {{.*#+}} xmm7 = xmm6[1,1,2,2]
-; AVX512-NEXT: vpblendw {{.*#+}} xmm4 = xmm7[0,1],xmm4[2],xmm7[3,4],xmm4[5],xmm7[6,7]
-; AVX512-NEXT: vpunpcklwd {{.*#+}} xmm6 = xmm6[0],xmm2[0],xmm6[1],xmm2[1],xmm6[2],xmm2[2],xmm6[3],xmm2[3]
-; AVX512-NEXT: vmovdqa {{.*#+}} xmm2 = [0,1,2,3,4,5,4,5,6,7,10,11,8,9,10,11]
-; AVX512-NEXT: vpshufb %xmm2, %xmm6, %xmm6
-; AVX512-NEXT: vinserti128 $1, %xmm4, %ymm6, %ymm4
-; AVX512-NEXT: vpunpckhwd {{.*#+}} xmm6 = xmm3[4],xmm5[4],xmm3[5],xmm5[5],xmm3[6],xmm5[6],xmm3[7],xmm5[7]
-; AVX512-NEXT: vmovdqa {{.*#+}} xmm7 = [4,5,10,11,10,11,8,9,8,9,14,15,12,13,14,15]
-; AVX512-NEXT: vpshufb %xmm7, %xmm6, %xmm6
-; AVX512-NEXT: vprold $16, %xmm3, %xmm3
-; AVX512-NEXT: vpshufd {{.*#+}} xmm5 = xmm5[1,1,2,2]
-; AVX512-NEXT: vpblendw {{.*#+}} xmm3 = xmm5[0,1],xmm3[2],xmm5[3,4],xmm3[5],xmm5[6,7]
-; AVX512-NEXT: vinserti128 $1, %xmm6, %ymm3, %ymm3
-; AVX512-NEXT: vshufi64x2 {{.*#+}} zmm4 = zmm3[0,1,2,3],zmm4[0,1,2,3]
-; AVX512-NEXT: vmovdqa (%rdx), %ymm3
-; AVX512-NEXT: vmovdqa 32(%rdx), %ymm5
-; AVX512-NEXT: vpmovsxbd {{.*#+}} zmm6 = [21,21,0,22,22,0,23,23,0,0,0,0,1,1,0,2]
-; AVX512-NEXT: vpermi2d (%rdx), %zmm5, %zmm6
-; AVX512-NEXT: vpternlogq {{.*#+}} zmm6 = zmm6 ^ (mem & (zmm6 ^ zmm4))
-; AVX512-NEXT: vmovdqa 32(%rdi), %ymm4
-; AVX512-NEXT: vmovdqa {{.*#+}} ymm8 = [128,128,128,128,12,13,128,128,128,128,14,15,128,128,128,128,16,17,128,128,128,128,18,19,128,128,128,128,20,21,128,128]
-; AVX512-NEXT: vpshufb %ymm8, %ymm4, %ymm4
-; AVX512-NEXT: vmovdqa 32(%rsi), %ymm9
-; AVX512-NEXT: vmovdqa {{.*#+}} ymm10 = [10,11,0,1,128,128,12,13,2,3,128,128,14,15,4,5,128,128,16,17,28,29,128,128,18,19,18,19,128,128,20,21]
-; AVX512-NEXT: vpshufb %ymm10, %ymm9, %ymm9
-; AVX512-NEXT: vpor %ymm4, %ymm9, %ymm4
-; AVX512-NEXT: vmovdqa 48(%rdi), %xmm9
-; AVX512-NEXT: vmovdqa 48(%rsi), %xmm11
-; AVX512-NEXT: vpunpckhwd {{.*#+}} xmm12 = xmm11[4],xmm9[4],xmm11[5],xmm9[5],xmm11[6],xmm9[6],xmm11[7],xmm9[7]
-; AVX512-NEXT: vpshufb %xmm7, %xmm12, %xmm7
-; AVX512-NEXT: vprold $16, %xmm11, %xmm11
-; AVX512-NEXT: vpshufd {{.*#+}} xmm9 = xmm9[1,1,2,2]
-; AVX512-NEXT: vpblendw {{.*#+}} xmm9 = xmm9[0,1],xmm11[2],xmm9[3,4],xmm11[5],xmm9[6,7]
-; AVX512-NEXT: vinserti128 $1, %xmm7, %ymm9, %ymm7
-; AVX512-NEXT: vshufi64x2 {{.*#+}} zmm4 = zmm4[0,1,2,3],zmm7[0,1,2,3]
-; AVX512-NEXT: vmovdqa {{.*#+}} ymm7 = [128,128,10,11,128,128,128,128,12,13,128,128,128,128,14,15,128,128,128,128,16,17,128,128,128,128,18,19,128,128,128,128]
-; AVX512-NEXT: vpshufb %ymm7, %ymm5, %ymm9
-; AVX512-NEXT: vpmovsxbd {{.*#+}} ymm11 = [5,5,0,6,6,0,7,7]
-; AVX512-NEXT: vpermd %ymm5, %ymm11, %ymm5
-; AVX512-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm5, %ymm5
-; AVX512-NEXT: vinserti64x4 $1, %ymm5, %zmm9, %zmm5
-; AVX512-NEXT: vpternlogq {{.*#+}} zmm5 = zmm5 | (zmm4 & mem)
-; AVX512-NEXT: vmovdqa (%rdi), %ymm4
-; AVX512-NEXT: vpshufb %ymm8, %ymm4, %ymm4
-; AVX512-NEXT: vmovdqa (%rsi), %ymm8
-; AVX512-NEXT: vpshufb %ymm10, %ymm8, %ymm8
-; AVX512-NEXT: vpor %ymm4, %ymm8, %ymm4
-; AVX512-NEXT: vprold $16, %xmm0, %xmm8
-; AVX512-NEXT: vpshufd {{.*#+}} xmm9 = xmm1[1,1,2,2]
-; AVX512-NEXT: vpblendw {{.*#+}} xmm8 = xmm9[0,1],xmm8[2],xmm9[3,4],xmm8[5],xmm9[6,7]
-; AVX512-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; AVX512-NEXT: vpshufb %xmm2, %xmm0, %xmm0
-; AVX512-NEXT: vinserti128 $1, %xmm8, %ymm0, %ymm0
-; AVX512-NEXT: vshufi64x2 {{.*#+}} zmm0 = zmm0[0,1,2,3],zmm4[0,1,2,3]
-; AVX512-NEXT: vpshufb %ymm7, %ymm3, %ymm1
-; AVX512-NEXT: vpmovsxbd {{.*#+}} ymm2 = [0,0,0,0,1,1,0,2]
-; AVX512-NEXT: vpermd %ymm3, %ymm2, %ymm2
-; AVX512-NEXT: vmovdqa64 {{.*#+}} zmm3 = [65535,65535,0,65535,65535,0,65535,65535,0,65535,65535,0,65535,65535,0,65535,65535,0,65535,65535,0,65535,65535,0,65535,65535,0,65535,65535,0,65535,65535]
-; AVX512-NEXT: vpandn %ymm2, %ymm3, %ymm2
-; AVX512-NEXT: vinserti64x4 $1, %ymm1, %zmm2, %zmm1
-; AVX512-NEXT: vpternlogq {{.*#+}} zmm1 = zmm1 | (zmm0 & zmm3)
-; AVX512-NEXT: vmovdqa64 %zmm1, (%rcx)
-; AVX512-NEXT: vmovdqa64 %zmm5, 128(%rcx)
-; AVX512-NEXT: vmovdqa64 %zmm6, 64(%rcx)
+; AVX512-NEXT: vmovdqa64 (%rdx), %zmm0
+; AVX512-NEXT: vmovdqa (%rsi), %xmm1
+; AVX512-NEXT: vmovdqa 16(%rsi), %xmm4
+; AVX512-NEXT: vmovdqa 32(%rsi), %xmm3
+; AVX512-NEXT: vprold $16, %xmm3, %xmm5
+; AVX512-NEXT: vmovdqa (%rdi), %xmm2
+; AVX512-NEXT: vmovdqa 16(%rdi), %xmm6
+; AVX512-NEXT: vmovdqa 32(%rdi), %xmm7
+; AVX512-NEXT: vpshufd {{.*#+}} xmm8 = xmm7[1,1,2,2]
+; AVX512-NEXT: vpblendw {{.*#+}} xmm5 = xmm8[0,1],xmm5[2],xmm8[3,4],xmm5[5],xmm8[6,7]
+; AVX512-NEXT: vpunpcklwd {{.*#+}} xmm7 = xmm7[0],xmm3[0],xmm7[1],xmm3[1],xmm7[2],xmm3[2],xmm7[3],xmm3[3]
+; AVX512-NEXT: vmovdqa {{.*#+}} xmm3 = [0,1,2,3,4,5,4,5,6,7,10,11,8,9,10,11]
+; AVX512-NEXT: vpshufb %xmm3, %xmm7, %xmm7
+; AVX512-NEXT: vinserti128 $1, %xmm5, %ymm7, %ymm5
+; AVX512-NEXT: vpunpckhwd {{.*#+}} xmm7 = xmm4[4],xmm6[4],xmm4[5],xmm6[5],xmm4[6],xmm6[6],xmm4[7],xmm6[7]
+; AVX512-NEXT: vmovdqa {{.*#+}} xmm8 = [4,5,10,11,10,11,8,9,8,9,14,15,12,13,14,15]
+; AVX512-NEXT: vpshufb %xmm8, %xmm7, %xmm7
+; AVX512-NEXT: vprold $16, %xmm4, %xmm4
+; AVX512-NEXT: vpshufd {{.*#+}} xmm6 = xmm6[1,1,2,2]
+; AVX512-NEXT: vpblendw {{.*#+}} xmm4 = xmm6[0,1],xmm4[2],xmm6[3,4],xmm4[5],xmm6[6,7]
+; AVX512-NEXT: vinserti128 $1, %xmm7, %ymm4, %ymm4
+; AVX512-NEXT: vshufi64x2 {{.*#+}} zmm5 = zmm4[0,1,2,3],zmm5[0,1,2,3]
+; AVX512-NEXT: vmovdqa (%rdx), %ymm4
+; AVX512-NEXT: vmovdqa 32(%rdx), %ymm6
+; AVX512-NEXT: vpmovsxbd {{.*#+}} zmm7 = [5,5,0,6,6,0,7,7,0,16,16,0,17,17,0,18]
+; AVX512-NEXT: vpermt2d %zmm6, %zmm7, %zmm0
+; AVX512-NEXT: vpternlogq {{.*#+}} zmm0 = zmm0 ^ (mem & (zmm0 ^ zmm5))
+; AVX512-NEXT: vmovdqa 32(%rdi), %ymm5
+; AVX512-NEXT: vmovdqa {{.*#+}} ymm9 = [128,128,128,128,12,13,128,128,128,128,14,15,128,128,128,128,16,17,128,128,128,128,18,19,128,128,128,128,20,21,128,128]
+; AVX512-NEXT: vpshufb %ymm9, %ymm5, %ymm5
+; AVX512-NEXT: vmovdqa 32(%rsi), %ymm10
+; AVX512-NEXT: vmovdqa {{.*#+}} ymm11 = [10,11,0,1,128,128,12,13,2,3,128,128,14,15,4,5,128,128,16,17,28,29,128,128,18,19,18,19,128,128,20,21]
+; AVX512-NEXT: vpshufb %ymm11, %ymm10, %ymm10
+; AVX512-NEXT: vpor %ymm5, %ymm10, %ymm5
+; AVX512-NEXT: vmovdqa 48(%rdi), %xmm10
+; AVX512-NEXT: vmovdqa 48(%rsi), %xmm12
+; AVX512-NEXT: vpunpckhwd {{.*#+}} xmm13 = xmm12[4],xmm10[4],xmm12[5],xmm10[5],xmm12[6],xmm10[6],xmm12[7],xmm10[7]
+; AVX512-NEXT: vpshufb %xmm8, %xmm13, %xmm8
+; AVX512-NEXT: vprold $16, %xmm12, %xmm12
+; AVX512-NEXT: vpshufd {{.*#+}} xmm10 = xmm10[1,1,2,2]
+; AVX512-NEXT: vpblendw {{.*#+}} xmm10 = xmm10[0,1],xmm12[2],xmm10[3,4],xmm12[5],xmm10[6,7]
+; AVX512-NEXT: vinserti128 $1, %xmm8, %ymm10, %ymm8
+; AVX512-NEXT: vshufi64x2 {{.*#+}} zmm5 = zmm5[0,1,2,3],zmm8[0,1,2,3]
+; AVX512-NEXT: vmovdqa {{.*#+}} ymm8 = [128,128,10,11,128,128,128,128,12,13,128,128,128,128,14,15,128,128,128,128,16,17,128,128,128,128,18,19,128,128,128,128]
+; AVX512-NEXT: vpshufb %ymm8, %ymm6, %ymm10
+; AVX512-NEXT: vpermd %ymm6, %ymm7, %ymm6
+; AVX512-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm6, %ymm6
+; AVX512-NEXT: vinserti64x4 $1, %ymm6, %zmm10, %zmm6
+; AVX512-NEXT: vpternlogq {{.*#+}} zmm6 = zmm6 | (zmm5 & mem)
+; AVX512-NEXT: vmovdqa (%rdi), %ymm5
+; AVX512-NEXT: vpshufb %ymm9, %ymm5, %ymm5
+; AVX512-NEXT: vmovdqa (%rsi), %ymm7
+; AVX512-NEXT: vpshufb %ymm11, %ymm7, %ymm7
+; AVX512-NEXT: vpor %ymm5, %ymm7, %ymm5
+; AVX512-NEXT: vprold $16, %xmm1, %xmm7
+; AVX512-NEXT: vpshufd {{.*#+}} xmm9 = xmm2[1,1,2,2]
+; AVX512-NEXT: vpblendw {{.*#+}} xmm7 = xmm9[0,1],xmm7[2],xmm9[3,4],xmm7[5],xmm9[6,7]
+; AVX512-NEXT: vpunpcklwd {{.*#+}} xmm1 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3]
+; AVX512-NEXT: vpshufb %xmm3, %xmm1, %xmm1
+; AVX512-NEXT: vinserti128 $1, %xmm7, %ymm1, %ymm1
+; AVX512-NEXT: vshufi64x2 {{.*#+}} zmm1 = zmm1[0,1,2,3],zmm5[0,1,2,3]
+; AVX512-NEXT: vpshufb %ymm8, %ymm4, %ymm2
+; AVX512-NEXT: vpmovsxbd {{.*#+}} ymm3 = [0,0,0,0,1,1,0,2]
+; AVX512-NEXT: vpermd %ymm4, %ymm3, %ymm3
+; AVX512-NEXT: vmovdqa64 {{.*#+}} zmm4 = [65535,65535,0,65535,65535,0,65535,65535,0,65535,65535,0,65535,65535,0,65535,65535,0,65535,65535,0,65535,65535,0,65535,65535,0,65535,65535,0,65535,65535]
+; AVX512-NEXT: vpandn %ymm3, %ymm4, %ymm3
+; AVX512-NEXT: vinserti64x4 $1, %ymm2, %zmm3, %zmm2
+; AVX512-NEXT: vpternlogq {{.*#+}} zmm2 = zmm2 | (zmm1 & zmm4)
+; AVX512-NEXT: vmovdqa64 %zmm2, (%rcx)
+; AVX512-NEXT: vmovdqa64 %zmm6, 128(%rcx)
+; AVX512-NEXT: vmovdqa64 %zmm0, 64(%rcx)
; AVX512-NEXT: vzeroupper
; AVX512-NEXT: retq
;
; AVX512-FCP-LABEL: store_i16_stride3_vf32:
; AVX512-FCP: # %bb.0:
-; AVX512-FCP-NEXT: vmovdqa (%rsi), %xmm0
-; AVX512-FCP-NEXT: vmovdqa 16(%rsi), %xmm3
-; AVX512-FCP-NEXT: vmovdqa 32(%rsi), %xmm2
-; AVX512-FCP-NEXT: vprold $16, %xmm2, %xmm4
-; AVX512-FCP-NEXT: vmovdqa (%rdi), %xmm1
-; AVX512-FCP-NEXT: vmovdqa 16(%rdi), %xmm5
-; AVX512-FCP-NEXT: vmovdqa 32(%rdi), %xmm6
-; AVX512-FCP-NEXT: vpshufd {{.*#+}} xmm7 = xmm6[1,1,2,2]
-; AVX512-FCP-NEXT: vpblendw {{.*#+}} xmm4 = xmm7[0,1],xmm4[2],xmm7[3,4],xmm4[5],xmm7[6,7]
-; AVX512-FCP-NEXT: vpunpcklwd {{.*#+}} xmm6 = xmm6[0],xmm2[0],xmm6[1],xmm2[1],xmm6[2],xmm2[2],xmm6[3],xmm2[3]
-; AVX512-FCP-NEXT: vmovdqa {{.*#+}} xmm2 = [0,1,2,3,4,5,4,5,6,7,10,11,8,9,10,11]
-; AVX512-FCP-NEXT: vpshufb %xmm2, %xmm6, %xmm6
-; AVX512-FCP-NEXT: vinserti128 $1, %xmm4, %ymm6, %ymm4
-; AVX512-FCP-NEXT: vpunpckhwd {{.*#+}} xmm6 = xmm3[4],xmm5[4],xmm3[5],xmm5[5],xmm3[6],xmm5[6],xmm3[7],xmm5[7]
-; AVX512-FCP-NEXT: vmovdqa {{.*#+}} xmm7 = [4,5,10,11,10,11,8,9,8,9,14,15,12,13,14,15]
-; AVX512-FCP-NEXT: vpshufb %xmm7, %xmm6, %xmm6
-; AVX512-FCP-NEXT: vprold $16, %xmm3, %xmm3
-; AVX512-FCP-NEXT: vpshufd {{.*#+}} xmm5 = xmm5[1,1,2,2]
-; AVX512-FCP-NEXT: vpblendw {{.*#+}} xmm3 = xmm5[0,1],xmm3[2],xmm5[3,4],xmm3[5],xmm5[6,7]
-; AVX512-FCP-NEXT: vinserti128 $1, %xmm6, %ymm3, %ymm3
-; AVX512-FCP-NEXT: vshufi64x2 {{.*#+}} zmm4 = zmm3[0,1,2,3],zmm4[0,1,2,3]
-; AVX512-FCP-NEXT: vmovdqa (%rdx), %ymm3
-; AVX512-FCP-NEXT: vmovdqa 32(%rdx), %ymm5
-; AVX512-FCP-NEXT: vpmovsxbd {{.*#+}} zmm6 = [21,21,0,22,22,0,23,23,0,0,0,0,1,1,0,2]
-; AVX512-FCP-NEXT: vpermi2d (%rdx), %zmm5, %zmm6
-; AVX512-FCP-NEXT: vpternlogq {{.*#+}} zmm6 = zmm6 ^ (mem & (zmm6 ^ zmm4))
-; AVX512-FCP-NEXT: vmovdqa 32(%rdi), %ymm4
-; AVX512-FCP-NEXT: vmovdqa {{.*#+}} ymm8 = [128,128,128,128,12,13,128,128,128,128,14,15,128,128,128,128,16,17,128,128,128,128,18,19,128,128,128,128,20,21,128,128]
-; AVX512-FCP-NEXT: vpshufb %ymm8, %ymm4, %ymm4
-; AVX512-FCP-NEXT: vmovdqa 32(%rsi), %ymm9
-; AVX512-FCP-NEXT: vmovdqa {{.*#+}} ymm10 = [10,11,0,1,128,128,12,13,2,3,128,128,14,15,4,5,128,128,16,17,28,29,128,128,18,19,18,19,128,128,20,21]
-; AVX512-FCP-NEXT: vpshufb %ymm10, %ymm9, %ymm9
-; AVX512-FCP-NEXT: vpor %ymm4, %ymm9, %ymm4
-; AVX512-FCP-NEXT: vmovdqa 48(%rdi), %xmm9
-; AVX512-FCP-NEXT: vmovdqa 48(%rsi), %xmm11
-; AVX512-FCP-NEXT: vpunpckhwd {{.*#+}} xmm12 = xmm11[4],xmm9[4],xmm11[5],xmm9[5],xmm11[6],xmm9[6],xmm11[7],xmm9[7]
-; AVX512-FCP-NEXT: vpshufb %xmm7, %xmm12, %xmm7
-; AVX512-FCP-NEXT: vprold $16, %xmm11, %xmm11
-; AVX512-FCP-NEXT: vpshufd {{.*#+}} xmm9 = xmm9[1,1,2,2]
-; AVX512-FCP-NEXT: vpblendw {{.*#+}} xmm9 = xmm9[0,1],xmm11[2],xmm9[3,4],xmm11[5],xmm9[6,7]
-; AVX512-FCP-NEXT: vinserti128 $1, %xmm7, %ymm9, %ymm7
-; AVX512-FCP-NEXT: vshufi64x2 {{.*#+}} zmm4 = zmm4[0,1,2,3],zmm7[0,1,2,3]
-; AVX512-FCP-NEXT: vmovdqa {{.*#+}} ymm7 = [128,128,10,11,128,128,128,128,12,13,128,128,128,128,14,15,128,128,128,128,16,17,128,128,128,128,18,19,128,128,128,128]
-; AVX512-FCP-NEXT: vpshufb %ymm7, %ymm5, %ymm9
-; AVX512-FCP-NEXT: vpmovsxbd {{.*#+}} ymm11 = [5,5,0,6,6,0,7,7]
-; AVX512-FCP-NEXT: vpermd %ymm5, %ymm11, %ymm5
-; AVX512-FCP-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm5, %ymm5
-; AVX512-FCP-NEXT: vinserti64x4 $1, %ymm5, %zmm9, %zmm5
-; AVX512-FCP-NEXT: vpternlogq {{.*#+}} zmm5 = zmm5 | (zmm4 & mem)
-; AVX512-FCP-NEXT: vmovdqa (%rdi), %ymm4
-; AVX512-FCP-NEXT: vpshufb %ymm8, %ymm4, %ymm4
-; AVX512-FCP-NEXT: vmovdqa (%rsi), %ymm8
-; AVX512-FCP-NEXT: vpshufb %ymm10, %ymm8, %ymm8
-; AVX512-FCP-NEXT: vpor %ymm4, %ymm8, %ymm4
-; AVX512-FCP-NEXT: vprold $16, %xmm0, %xmm8
-; AVX512-FCP-NEXT: vpshufd {{.*#+}} xmm9 = xmm1[1,1,2,2]
-; AVX512-FCP-NEXT: vpblendw {{.*#+}} xmm8 = xmm9[0,1],xmm8[2],xmm9[3,4],xmm8[5],xmm9[6,7]
-; AVX512-FCP-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
-; AVX512-FCP-NEXT: vpshufb %xmm2, %xmm0, %xmm0
-; AVX512-FCP-NEXT: vinserti128 $1, %xmm8, %ymm0, %ymm0
-; AVX512-FCP-NEXT: vshufi64x2 {{.*#+}} zmm0 = zmm0[0,1,2,3],zmm4[0,1,2,3]
-; AVX512-FCP-NEXT: vpshufb %ymm7, %ymm3, %ymm1
-; AVX512-FCP-NEXT: vpmovsxbd {{.*#+}} ymm2 = [0,0,0,0,1,1,0,2]
-; AVX512-FCP-NEXT: vpermd %ymm3, %ymm2, %ymm2
-; AVX512-FCP-NEXT: vmovdqa64 {{.*#+}} zmm3 = [65535,65535,0,65535,65535,0,65535,65535,0,65535,65535,0,65535,65535,0,65535,65535,0,65535,65535,0,65535,65535,0,65535,65535,0,65535,65535,0,65535,65535]
-; AVX512-FCP-NEXT: vpandn %ymm2, %ymm3, %ymm2
-; AVX512-FCP-NEXT: vinserti64x4 $1, %ymm1, %zmm2, %zmm1
-; AVX512-FCP-NEXT: vpternlogq {{.*#+}} zmm1 = zmm1 | (zmm0 & zmm3)
-; AVX512-FCP-NEXT: vmovdqa64 %zmm1, (%rcx)
-; AVX512-FCP-NEXT: vmovdqa64 %zmm5, 128(%rcx)
-; AVX512-FCP-NEXT: vmovdqa64 %zmm6, 64(%rcx)
+; AVX512-FCP-NEXT: vmovdqa64 (%rdx), %zmm0
+; AVX512-FCP-NEXT: vmovdqa (%rsi), %xmm1
+; AVX512-FCP-NEXT: vmovdqa 16(%rsi), %xmm4
+; AVX512-FCP-NEXT: vmovdqa 32(%rsi), %xmm3
+; AVX512-FCP-NEXT: vprold $16, %xmm3, %xmm5
+; AVX512-FCP-NEXT: vmovdqa (%rdi), %xmm2
+; AVX512-FCP-NEXT: vmovdqa 16(%rdi), %xmm6
+; AVX512-FCP-NEXT: vmovdqa 32(%rdi), %xmm7
+; AVX512-FCP-NEXT: vpshufd {{.*#+}} xmm8 = xmm7[1,1,2,2]
+; AVX512-FCP-NEXT: vpblendw {{.*#+}} xmm5 = xmm8[0,1],xmm5[2],xmm8[3,4],xmm5[5],xmm8[6,7]
+; AVX512-FCP-NEXT: vpunpcklwd {{.*#+}} xmm7 = xmm7[0],xmm3[0],xmm7[1],xmm3[1],xmm7[2],xmm3[2],xmm7[3],xmm3[3]
+; AVX512-FCP-NEXT: vmovdqa {{.*#+}} xmm3 = [0,1,2,3,4,5,4,5,6,7,10,11,8,9,10,11]
+; AVX512-FCP-NEXT: vpshufb %xmm3, %xmm7, %xmm7
+; AVX512-FCP-NEXT: vinserti128 $1, %xmm5, %ymm7, %ymm5
+; AVX512-FCP-NEXT: vpunpckhwd {{.*#+}} xmm7 = xmm4[4],xmm6[4],xmm4[5],xmm6[5],xmm4[6],xmm6[6],xmm4[7],xmm6[7]
+; AVX512-FCP-NEXT: vmovdqa {{.*#+}} xmm8 = [4,5,10,11,10,11,8,9,8,9,14,15,12,13,14,15]
+; AVX512-FCP-NEXT: vpshufb %xmm8, %xmm7, %xmm7
+; AVX512-FCP-NEXT: vprold $16, %xmm4, %xmm4
+; AVX512-FCP-NEXT: vpshufd {{.*#+}} xmm6 = xmm6[1,1,2,2]
+; AVX512-FCP-NEXT: vpblendw {{.*#+}} xmm4 = xmm6[0,1],xmm4[2],xmm6[3,4],xmm4[5],xmm6[6,7]
+; AVX512-FCP-NEXT: vinserti128 $1, %xmm7, %ymm4, %ymm4
+; AVX512-FCP-NEXT: vshufi64x2 {{.*#+}} zmm5 = zmm4[0,1,2,3],zmm5[0,1,2,3]
+; AVX512-FCP-NEXT: vmovdqa (%rdx), %ymm4
+; AVX512-FCP-NEXT: vmovdqa 32(%rdx), %ymm6
+; AVX512-FCP-NEXT: vpmovsxbd {{.*#+}} zmm7 = [5,5,0,6,6,0,7,7,0,16,16,0,17,17,0,18]
+; AVX512-FCP-NEXT: vpermt2d %zmm6, %zmm7, %zmm0
+; AVX512-FCP-NEXT: vpternlogq {{.*#+}} zmm0 = zmm0 ^ (mem & (zmm0 ^ zmm5))
+; AVX512-FCP-NEXT: vmovdqa 32(%rdi), %ymm5
+; AVX512-FCP-NEXT: vmovdqa {{.*#+}} ymm9 = [128,128,128,128,12,13,128,128,128,128,14,15,128,128,128,128,16,17,128,128,128,128,18,19,128,128,128,128,20,21,128,128]
+; AVX512-FCP-NEXT: vpshufb %ymm9, %ymm5, %ymm5
+; AVX512-FCP-NEXT: vmovdqa 32(%rsi), %ymm10
+; AVX512-FCP-NEXT: vmovdqa {{.*#+}} ymm11 = [10,11,0,1,128,128,12,13,2,3,128,128,14,15,4,5,128,128,16,17,28,29,128,128,18,19,18,19,128,128,20,21]
+; AVX512-FCP-NEXT: vpshufb %ymm11, %ymm10, %ymm10
+; AVX512-FCP-NEXT: vpor %ymm5, %ymm10, %ymm5
+; AVX512-FCP-NEXT: vmovdqa 48(%rdi), %xmm10
+; AVX512-FCP-NEXT: vmovdqa 48(%rsi), %xmm12
+; AVX512-FCP-NEXT: vpunpckhwd {{.*#+}} xmm13 = xmm12[4],xmm10[4],xmm12[5],xmm10[5],xmm12[6],xmm10[6],xmm12[7],xmm10[7]
+; AVX512-FCP-NEXT: vpshufb %xmm8, %xmm13, %xmm8
+; AVX512-FCP-NEXT: vprold $16, %xmm12, %xmm12
+; AVX512-FCP-NEXT: vpshufd {{.*#+}} xmm10 = xmm10[1,1,2,2]
+; AVX512-FCP-NEXT: vpblendw {{.*#+}} xmm10 = xmm10[0,1],xmm12[2],xmm10[3,4],xmm12[5],xmm10[6,7]
+; AVX512-FCP-NEXT: vinserti128 $1, %xmm8, %ymm10, %ymm8
+; AVX512-FCP-NEXT: vshufi64x2 {{.*#+}} zmm5 = zmm5[0,1,2,3],zmm8[0,1,2,3]
+; AVX512-FCP-NEXT: vmovdqa {{.*#+}} ymm8 = [128,128,10,11,128,128,128,128,12,13,128,128,128,128,14,15,128,128,128,128,16,17,128,128,128,128,18,19,128,128,128,128]
+; AVX512-FCP-NEXT: vpshufb %ymm8, %ymm6, %ymm10
+; AVX512-FCP-NEXT: vpermd %ymm6, %ymm7, %ymm6
+; AVX512-FCP-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm6, %ymm6
+; AVX512-FCP-NEXT: vinserti64x4 $1, %ymm6, %zmm10, %zmm6
+; AVX512-FCP-NEXT: vpternlogq {{.*#+}} zmm6 = zmm6 | (zmm5 & mem)
+; AVX512-FCP-NEXT: vmovdqa (%rdi), %ymm5
+; AVX512-FCP-NEXT: vpshufb %ymm9, %ymm5, %ymm5
+; AVX512-FCP-NEXT: vmovdqa (%rsi), %ymm7
+; AVX512-FCP-NEXT: vpshufb %ymm11, %ymm7, %ymm7
+; AVX512-FCP-NEXT: vpor %ymm5, %ymm7, %ymm5
+; AVX512-FCP-NEXT: vprold $16, %xmm1, %xmm7
+; AVX512-FCP-NEXT: vpshufd {{.*#+}} xmm9 = xmm2[1,1,2,2]
+; AVX512-FCP-NEXT: vpblendw {{.*#+}} xmm7 = xmm9[0,1],xmm7[2],xmm9[3,4],xmm7[5],xmm9[6,7]
+; AVX512-FCP-NEXT: vpunpcklwd {{.*#+}} xmm1 = xmm2[0],xmm1[0],xmm2[1],xmm1[1],xmm2[2],xmm1[2],xmm2[3],xmm1[3]
+; AVX512-FCP-NEXT: vpshufb %xmm3, %xmm1, %xmm1
+; AVX512-FCP-NEXT: vinserti128 $1, %xmm7, %ymm1, %ymm1
+; AVX512-FCP-NEXT: vshufi64x2 {{.*#+}} zmm1 = zmm1[0,1,2,3],zmm5[0,1,2,3]
+; AVX512-FCP-NEXT: vpshufb %ymm8, %ymm4, %ymm2
+; AVX512-FCP-NEXT: vpmovsxbd {{.*#+}} ymm3 = [0,0,0,0,1,1,0,2]
+; AVX512-FCP-NEXT: vpermd %ymm4, %ymm3, %ymm3
+; AVX512-FCP-NEXT: vmovdqa64 {{.*#+}} zmm4 = [65535,65535,0,65535,65535,0,65535,65535,0,65535,65535,0,65535,65535,0,65535,65535,0,65535,65535,0,65535,65535,0,65535,65535,0,65535,65535,0,65535,65535]
+; AVX512-FCP-NEXT: vpandn %ymm3, %ymm4, %ymm3
+; AVX512-FCP-NEXT: vinserti64x4 $1, %ymm2, %zmm3, %zmm2
+; AVX512-FCP-NEXT: vpternlogq {{.*#+}} zmm2 = zmm2 | (zm...
[truncated]
|
frederik-h
pushed a commit
to frederik-h/llvm-project
that referenced
this pull request
Mar 18, 2025
…catVectorOps from inside ConcatSubOperand. (llvm#131303) Before falling back to creating a generic ISD::CONCAT_VECTORS node, see if we can directly concat the subvectors if we peek through any bitcasts.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Before falling back to creating a generic ISD::CONCAT_VECTORS node, see if we can directly concat the subvectors if we peek through any bitcasts.