[llvm] r278211 - [DAGCombine] Avoid INSERT_SUBVECTOR reinsertions (PR28678)

[Simon Pilgrim via llvm-commits]

Author: rksimon
Date: Wed Aug 10 05:50:53 2016
New Revision: 278211

URL: http://llvm.org/viewvc/llvm-project?rev=278211&view=rev
Log:
[DAGCombine] Avoid INSERT_SUBVECTOR reinsertions (PR28678)

If the input vector to INSERT_SUBVECTOR is another INSERT_SUBVECTOR, and this inserted subvector replaces the last insertion, then insert into the common source vector.

i.e. 
INSERT_SUBVECTOR( INSERT_SUBVECTOR( Vec, SubOld, Idx ), SubNew, Idx ) --> INSERT_SUBVECTOR( Vec, SubNew, Idx )

Differential Revision: https://reviews.llvm.org/D23330

Modified:
    llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
    llvm/trunk/test/CodeGen/X86/avx-intrinsics-fast-isel.ll
    llvm/trunk/test/CodeGen/X86/insertelement-zero.ll

Modified: llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp?rev=278211&r1=278210&r2=278211&view=diff
==============================================================================
--- llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp (original)
+++ llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp Wed Aug 10 05:50:53 2016
@@ -13802,10 +13802,20 @@ SDValue DAGCombiner::visitSCALAR_TO_VECT
 }
 
 SDValue DAGCombiner::visitINSERT_SUBVECTOR(SDNode *N) {
+  EVT VT = N->getValueType(0);
   SDValue N0 = N->getOperand(0);
   SDValue N1 = N->getOperand(1);
   SDValue N2 = N->getOperand(2);
 
+  // Combine INSERT_SUBVECTORs where we are inserting to the same index.
+  // INSERT_SUBVECTOR( INSERT_SUBVECTOR( Vec, SubOld, Idx ), SubNew, Idx )
+  // --> INSERT_SUBVECTOR( Vec, SubNew, Idx )
+  if (N0.getOpcode() == ISD::INSERT_SUBVECTOR &&
+      N0.getOperand(1).getValueType() == N1.getValueType() &&
+      N0.getOperand(2) == N2)
+    return DAG.getNode(ISD::INSERT_SUBVECTOR, SDLoc(N), VT, N0.getOperand(0),
+                       N1, N2);
+
   if (N0.getValueType() != N1.getValueType())
     return SDValue();
 
@@ -13814,7 +13824,6 @@ SDValue DAGCombiner::visitINSERT_SUBVECT
   if (N0.getOpcode() == ISD::CONCAT_VECTORS && N0->getNumOperands() == 2 &&
       N2.getOpcode() == ISD::Constant) {
     APInt InsIdx = cast<ConstantSDNode>(N2)->getAPIntValue();
-    EVT VT = N->getValueType(0);
 
     // Lower half: fold (insert_subvector (concat_vectors X, Y), Z) ->
     // (concat_vectors Z, Y)

Modified: llvm/trunk/test/CodeGen/X86/avx-intrinsics-fast-isel.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/avx-intrinsics-fast-isel.ll?rev=278211&r1=278210&r2=278211&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/avx-intrinsics-fast-isel.ll (original)
+++ llvm/trunk/test/CodeGen/X86/avx-intrinsics-fast-isel.ll Wed Aug 10 05:50:53 2016
@@ -1026,9 +1026,8 @@ define <4 x i64> @test_mm256_insert_epi6
 ; X32:       # BB#0:
 ; X32-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; X32-NEXT:    vpinsrd $2, {{[0-9]+}}(%esp), %xmm1, %xmm1
-; X32-NEXT:    vpinsrd $3, {{[0-9]+}}(%esp), %xmm1, %xmm2
+; X32-NEXT:    vpinsrd $3, {{[0-9]+}}(%esp), %xmm1, %xmm1
 ; X32-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
-; X32-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: test_mm256_insert_epi64:

Modified: llvm/trunk/test/CodeGen/X86/insertelement-zero.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/insertelement-zero.ll?rev=278211&r1=278210&r2=278211&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/insertelement-zero.ll (original)
+++ llvm/trunk/test/CodeGen/X86/insertelement-zero.ll Wed Aug 10 05:50:53 2016
@@ -587,7 +587,6 @@ define <32 x i8> @insert_v32i8_z12345678
 ; AVX1-NEXT:    vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX1-NEXT:    vpinsrb $14, %eax, %xmm1, %xmm1
-; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    vpinsrb $15, %eax, %xmm1, %xmm1
 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
@@ -601,7 +600,6 @@ define <32 x i8> @insert_v32i8_z12345678
 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
 ; AVX2-NEXT:    vpinsrb $14, %eax, %xmm1, %xmm1
-; AVX2-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vpinsrb $15, %eax, %xmm1, %xmm1
 ; AVX2-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
 ; AVX2-NEXT:    retq