[llvm] r372886 - [DAGCombiner] add one-use restriction to vector transform with cheap extract
Sanjay Patel via llvm-commits
llvm-commits at lists.llvm.org
Wed Sep 25 08:08:33 PDT 2019
Author: spatel
Date: Wed Sep 25 08:08:33 2019
New Revision: 372886
URL: http://llvm.org/viewvc/llvm-project?rev=372886&view=rev
Log:
[DAGCombiner] add one-use restriction to vector transform with cheap extract
We might be able to do better on the example in the test,
but in general, we should not scalarize a splatted vector
binop if there are other uses of the binop. Otherwise, we
can end up with code as we had - a scalar op that is
redundant with a vector op.
Modified:
llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
llvm/trunk/test/CodeGen/X86/scalarize-fp.ll
Modified: llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp?rev=372886&r1=372885&r2=372886&view=diff
==============================================================================
--- llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp (original)
+++ llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp Wed Sep 25 08:08:33 2019
@@ -18990,7 +18990,7 @@ SDValue DAGCombiner::visitVECTOR_SHUFFLE
// build_vector.
if (SVN->isSplat() && SVN->getSplatIndex() < (int)NumElts) {
int SplatIndex = SVN->getSplatIndex();
- if (TLI.isExtractVecEltCheap(VT, SplatIndex) &&
+ if (N0.hasOneUse() && TLI.isExtractVecEltCheap(VT, SplatIndex) &&
TLI.isBinOp(N0.getOpcode()) && N0.getNode()->getNumValues() == 1) {
// splat (vector_bo L, R), Index -->
// splat (scalar_bo (extelt L, Index), (extelt R, Index))
Modified: llvm/trunk/test/CodeGen/X86/scalarize-fp.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/scalarize-fp.ll?rev=372886&r1=372885&r2=372886&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/scalarize-fp.ll (original)
+++ llvm/trunk/test/CodeGen/X86/scalarize-fp.ll Wed Sep 25 08:08:33 2019
@@ -777,21 +777,18 @@ define <8 x float> @splat0_fdiv_const_op
define <4 x float> @multi_use_binop(<4 x float> %x, <4 x float> %y) {
; SSE-LABEL: multi_use_binop:
; SSE: # %bb.0:
-; SSE-NEXT: movaps %xmm0, %xmm2
-; SSE-NEXT: mulps %xmm1, %xmm2
-; SSE-NEXT: mulss %xmm1, %xmm0
-; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,0,0,0]
-; SSE-NEXT: movlhps {{.*#+}} xmm1 = xmm1[0],xmm2[0]
+; SSE-NEXT: mulps %xmm1, %xmm0
+; SSE-NEXT: movlhps {{.*#+}} xmm1 = xmm1[0],xmm0[0]
+; SSE-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,1,2,0]
; SSE-NEXT: addps %xmm1, %xmm0
; SSE-NEXT: retq
;
; AVX-LABEL: multi_use_binop:
; AVX: # %bb.0:
-; AVX-NEXT: vmulps %xmm1, %xmm0, %xmm2
-; AVX-NEXT: vmulss %xmm1, %xmm0, %xmm0
-; AVX-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[0,0,0,0]
-; AVX-NEXT: vmovddup {{.*#+}} xmm1 = xmm2[0,0]
-; AVX-NEXT: vaddps %xmm1, %xmm0, %xmm0
+; AVX-NEXT: vmulps %xmm1, %xmm0, %xmm0
+; AVX-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[0,1,2,0]
+; AVX-NEXT: vmovddup {{.*#+}} xmm0 = xmm0[0,0]
+; AVX-NEXT: vaddps %xmm0, %xmm1, %xmm0
; AVX-NEXT: retq
%mul = fmul <4 x float> %x, %y
%mul0 = shufflevector <4 x float> %mul, <4 x float> undef, <4 x i32> <i32 undef, i32 undef, i32 undef, i32 0>
More information about the llvm-commits
mailing list