[PATCH] D38696: [DAGCombine] Permit combining of shuffle of equivalent splat BUILD_VECTORs

[Simon Pilgrim via Phabricator via llvm-commits]

RKSimon created this revision.

combineShuffleOfScalars is very conservative about shuffled BUILD_VECTOsR that can be combined together.

I'd like to add one additional case - if both BUILD_VECTORs represent splats of the same scalar value but with different UNDEF elements, then we should create a single splat BUILD_VECTOR, sharing only the common UNDEF elements.


Repository:
  rL LLVM

https://reviews.llvm.org/D38696

Files:
  lib/CodeGen/SelectionDAG/DAGCombiner.cpp
  test/CodeGen/X86/avx-vbroadcast.ll


Index: test/CodeGen/X86/avx-vbroadcast.ll
===================================================================
--- test/CodeGen/X86/avx-vbroadcast.ll
+++ test/CodeGen/X86/avx-vbroadcast.ll
@@ -853,14 +853,12 @@
 ; X32-LABEL: broadcast_shuffle1032:
 ; X32:       ## BB#0:
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    vmovddup {{.*#+}} xmm0 = mem[0,0]
-; X32-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
+; X32-NEXT:    vbroadcastsd (%eax), %ymm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: broadcast_shuffle1032:
 ; X64:       ## BB#0:
-; X64-NEXT:    vmovddup {{.*#+}} xmm0 = mem[0,0]
-; X64-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
+; X64-NEXT:    vbroadcastsd (%rdi), %ymm0
 ; X64-NEXT:    retq
   %1 = load double, double* %p
   %2 = insertelement <2 x double> undef, double %1, i32 1
Index: lib/CodeGen/SelectionDAG/DAGCombiner.cpp
===================================================================
--- lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -15489,6 +15489,24 @@
       return SDValue();
   }
 
+  // If both inputs are splats of the same value then we can safely merge this
+  // to a single BUILD_VECTOR with only the common undef elements.
+  auto *BV0 = dyn_cast<BuildVectorSDNode>(N0);
+  auto *BV1 = dyn_cast<BuildVectorSDNode>(N1);
+  if (BV0 && BV1) {
+    BitVector Undefs0, Undefs1;
+    SDValue Splat0 = BV0->getSplatValue(&Undefs0);
+    SDValue Splat1 = BV1->getSplatValue(&Undefs1);
+    if (Splat0 && Splat0 == Splat1) {
+      Undefs0 &= Undefs1;
+      SmallVector<SDValue, 8> Ops(NumElts, Splat0);
+      for (unsigned i = 0; i != NumElts; ++i)
+        if (Undefs0[0])
+          Ops[i] = DAG.getUNDEF(Splat0.getValueType());
+      return DAG.getBuildVector(VT, SDLoc(SVN), Ops);
+    }
+  }
+
   SmallVector<SDValue, 8> Ops;
   SmallSet<SDValue, 16> DuplicateOps;
   for (int M : SVN->getMask()) {


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D38696.118239.patch
Type: text/x-patch
Size: 1927 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20171009/be100788/attachment.bin>