From e9ea05106521cc146d86323d9e8858b2299557bf Mon Sep 17 00:00:00 2001
From: Simon Pilgrim <llvm-dev@redking.me.uk>
Date: Fri, 1 Feb 2019 15:31:01 +0000
Subject: [PATCH] [X86][AVX] Combine INSERT_SUBVECTOR(SRC0,
 BITCAST(SHUFFLE(EXTRACT_SUBVECTOR(SRC1)))

Enable peeking through one use bitcasts to the subvector shuffle.

This still depends on the subvector being the same scalar-size but D57514 has already helped with the more tricky patterns

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@352879 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Target/X86/X86ISelLowering.cpp        |  7 ++++---
 test/CodeGen/X86/vector-shuffle-256-v8.ll | 25 +++++++++++++++--------
 2 files changed, 20 insertions(+), 12 deletions(-)

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index cc00970569e..36d3fe1dd60 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -6592,7 +6592,6 @@ static bool getFauxShuffleMask(SDValue N, SmallVectorImpl<int> &Mask,
   case ISD::INSERT_SUBVECTOR: {
     // Handle INSERT_SUBVECTOR(SRC0, SHUFFLE(EXTRACT_SUBVECTOR(SRC1)) where
     // SRC0/SRC1 are both of the same valuetype VT.
-    // TODO - add peekThroughOneUseBitcasts support.
     SDValue Src = N.getOperand(0);
     SDValue Sub = N.getOperand(1);
     EVT SubVT = Sub.getValueType();
@@ -6602,8 +6601,10 @@ static bool getFauxShuffleMask(SDValue N, SmallVectorImpl<int> &Mask,
       return false;
     SmallVector<int, 64> SubMask;
     SmallVector<SDValue, 2> SubInputs;
-    if (!resolveTargetShuffleInputs(Sub, SubInputs, SubMask, DAG) ||
-        SubMask.size() != NumSubElts)
+    if (!resolveTargetShuffleInputs(peekThroughOneUseBitcasts(Sub), SubInputs,
+                                    SubMask, DAG))
+      return false;
+    if (SubMask.size() != NumSubElts)
       return false;
     Ops.push_back(Src);
     for (SDValue &SubInput : SubInputs) {
diff --git a/test/CodeGen/X86/vector-shuffle-256-v8.ll b/test/CodeGen/X86/vector-shuffle-256-v8.ll
index 63c65d7abda..2eb33ff0af8 100644
--- a/test/CodeGen/X86/vector-shuffle-256-v8.ll
+++ b/test/CodeGen/X86/vector-shuffle-256-v8.ll
@@ -1553,12 +1553,19 @@ define <8 x i32> @shuffle_v8i32_08991abb(<8 x i32> %a, <8 x i32> %b) {
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7]
 ; AVX2-NEXT:    retq
 ;
-; AVX512VL-LABEL: shuffle_v8i32_08991abb:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm0[0],zero,xmm0[1],zero
-; AVX512VL-NEXT:    vmovdqa {{.*#+}} ymm0 = [8,0,1,1,10,2,3,3]
-; AVX512VL-NEXT:    vpermi2d %ymm2, %ymm1, %ymm0
-; AVX512VL-NEXT:    retq
+; AVX512VL-SLOW-LABEL: shuffle_v8i32_08991abb:
+; AVX512VL-SLOW:       # %bb.0:
+; AVX512VL-SLOW-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm0[0],zero,xmm0[1],zero
+; AVX512VL-SLOW-NEXT:    vmovdqa {{.*#+}} ymm0 = [8,0,1,1,10,2,3,3]
+; AVX512VL-SLOW-NEXT:    vpermi2d %ymm2, %ymm1, %ymm0
+; AVX512VL-SLOW-NEXT:    retq
+;
+; AVX512VL-FAST-LABEL: shuffle_v8i32_08991abb:
+; AVX512VL-FAST:       # %bb.0:
+; AVX512VL-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [8,0,1,1,9,2,3,3]
+; 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 9, i32 1, i32 10, i32 11, i32 11>
   ret <8 x i32> %shuffle
 }
@@ -1605,9 +1612,9 @@ define <8 x i32> @shuffle_v8i32_09ab1def(<8 x i32> %a, <8 x i32> %b) {
 ;
 ; AVX512VL-FAST-LABEL: shuffle_v8i32_09ab1def:
 ; AVX512VL-FAST:       # %bb.0:
-; AVX512VL-FAST-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm0[0],zero,xmm0[1],zero
-; AVX512VL-FAST-NEXT:    vmovdqa {{.*#+}} ymm0 = [8,1,2,3,10,5,6,7]
-; AVX512VL-FAST-NEXT:    vpermi2d %ymm2, %ymm1, %ymm0
+; AVX512VL-FAST-NEXT:    vmovdqa {{.*#+}} ymm2 = [8,1,2,3,9,5,6,7]
+; 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 9, i32 10, i32 11, i32 1, i32 13, i32 14, i32 15>
   ret <8 x i32> %shuffle
-- 
2.40.0