[X86] combineExtractWithShuffle - handle extract(truncate(x), 0)

Eventually we need to generalize combineExtractWithShuffle to handle all faux shuffles and handle truncate (and X86ISD::VTRUNC etc.) there, but we're not ready yet (still creates nodes on the fly, incomplete DemandedElts support, bad use of recursive Depth limit).

llvm-svn: 369134

2 files changed, 15 insertions, 15 deletions

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 297fa942c7e..a6131f9e4b5 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -35202,6 +35202,7 @@ static SDValue combineExtractWithShuffle(SDNode *N, SelectionDAG &DAG,
   if (DCI.isBeforeLegalizeOps())
     return SDValue();
 
+  SDLoc dl(N);
   SDValue Src = N->getOperand(0);
   SDValue Idx = N->getOperand(1);
 
@@ -35223,6 +35224,16 @@ static SDValue combineExtractWithShuffle(SDNode *N, SelectionDAG &DAG,
       return DAG.getBitcast(VT, SrcOp);
   }
 
+  // Handle extract(truncate(x)) for 0'th index.
+  // TODO: Treat this as a faux shuffle?
+  // TODO: When can we use this for general indices?
+  if (ISD::TRUNCATE == Src.getOpcode() && SrcVT.is128BitVector() &&
+      isNullConstant(Idx)) {
+    Src = extract128BitVector(Src.getOperand(0), 0, DAG, dl);
+    Src = DAG.getBitcast(SrcVT, Src);
+    return DAG.getNode(N->getOpcode(), dl, VT, Src, Idx);
+  }
+
   // Resolve the target shuffle inputs and mask.
   SmallVector<int, 16> Mask;
   SmallVector<SDValue, 2> Ops;
@@ -35260,7 +35271,6 @@ static SDValue combineExtractWithShuffle(SDNode *N, SelectionDAG &DAG,
     return SDValue();
 
   int SrcIdx = Mask[N->getConstantOperandVal(1)];
-  SDLoc dl(N);
 
   // If the shuffle source element is undef/zero then we can just accept it.
   if (SrcIdx == SM_SentinelUndef)
diff --git a/llvm/test/CodeGen/X86/vector-reduce-mul.ll b/llvm/test/CodeGen/X86/vector-reduce-mul.ll
index b4cca0015bd..b70f964a542 100644
--- a/llvm/test/CodeGen/X86/vector-reduce-mul.ll
+++ b/llvm/test/CodeGen/X86/vector-reduce-mul.ll
@@ -1828,9 +1828,7 @@ define i8 @test_v16i8(<16 x i8> %a0) {
 ; AVX512BW-NEXT:    vpmullw %ymm1, %ymm0, %ymm0
 ; AVX512BW-NEXT:    vpmovwb %zmm0, %ymm1
 ; AVX512BW-NEXT:    vpsrlw $8, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
-; AVX512BW-NEXT:    vpmullw %ymm1, %ymm0, %ymm0
-; AVX512BW-NEXT:    vpmovwb %zmm0, %ymm0
+; AVX512BW-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512BW-NEXT:    vpextrb $0, %xmm0, %eax
 ; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512BW-NEXT:    vzeroupper
@@ -1852,9 +1850,7 @@ define i8 @test_v16i8(<16 x i8> %a0) {
 ; AVX512BWVL-NEXT:    vpmullw %ymm1, %ymm0, %ymm0
 ; AVX512BWVL-NEXT:    vpmovwb %ymm0, %xmm1
 ; AVX512BWVL-NEXT:    vpsrlw $8, %xmm1, %xmm1
-; AVX512BWVL-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
-; AVX512BWVL-NEXT:    vpmullw %ymm1, %ymm0, %ymm0
-; AVX512BWVL-NEXT:    vpmovwb %ymm0, %xmm0
+; AVX512BWVL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512BWVL-NEXT:    vpextrb $0, %xmm0, %eax
 ; AVX512BWVL-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512BWVL-NEXT:    vzeroupper
@@ -1879,10 +1875,7 @@ define i8 @test_v16i8(<16 x i8> %a0) {
 ; AVX512DQ-NEXT:    vpmovzxwd {{.*#+}} zmm1 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
 ; AVX512DQ-NEXT:    vpmovdb %zmm1, %xmm1
 ; AVX512DQ-NEXT:    vpsrlw $8, %xmm1, %xmm1
-; AVX512DQ-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
-; AVX512DQ-NEXT:    vpmullw %ymm1, %ymm0, %ymm0
-; AVX512DQ-NEXT:    vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
-; AVX512DQ-NEXT:    vpmovdb %zmm0, %xmm0
+; AVX512DQ-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512DQ-NEXT:    vpextrb $0, %xmm0, %eax
 ; AVX512DQ-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512DQ-NEXT:    vzeroupper
@@ -1907,10 +1900,7 @@ define i8 @test_v16i8(<16 x i8> %a0) {
 ; AVX512DQVL-NEXT:    vpmovzxwd {{.*#+}} zmm1 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
 ; AVX512DQVL-NEXT:    vpmovdb %zmm1, %xmm1
 ; AVX512DQVL-NEXT:    vpsrlw $8, %xmm1, %xmm1
-; AVX512DQVL-NEXT:    vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero
-; AVX512DQVL-NEXT:    vpmullw %ymm1, %ymm0, %ymm0
-; AVX512DQVL-NEXT:    vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
-; AVX512DQVL-NEXT:    vpmovdb %zmm0, %xmm0
+; AVX512DQVL-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX512DQVL-NEXT:    vpextrb $0, %xmm0, %eax
 ; AVX512DQVL-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX512DQVL-NEXT:    vzeroupper