diff options
| author | Owen Anderson <resistor@mac.com> | 2012-05-07 20:51:25 +0000 |
|---|---|---|
| committer | Owen Anderson <resistor@mac.com> | 2012-05-07 20:51:25 +0000 |
| commit | 713e953118175d693a3fddbf6a61dc73aa69ad87 (patch) | |
| tree | ef945f02d87468cf3a09338e3fda0a36da75128f | |
| parent | 423f19f2dadee9325766008b63c1faf3c644043b (diff) | |
| download | llvm-713e953118175d693a3fddbf6a61dc73aa69ad87.tar.gz llvm-713e953118175d693a3fddbf6a61dc73aa69ad87.tar.bz2 llvm-713e953118175d693a3fddbf6a61dc73aa69ad87.tar.xz | |
Teach DAG combine to fold x-x to 0.0 when unsafe FP math is enabled.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@156324 91177308-0d34-0410-b5e6-96231b3b80d8
| -rw-r--r-- | lib/CodeGen/SelectionDAG/DAGCombiner.cpp | 4 | ||||
| -rw-r--r-- | test/CodeGen/ARM/unsafe-fsub.ll | 18 |
2 files changed, 22 insertions, 0 deletions
diff --git a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp index c2b7f6c513..753ba937c0 100644 --- a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp +++ b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp @@ -5670,9 +5670,13 @@ SDValue DAGCombiner::visitFSUB(SDNode *N) { GetNegatedExpression(N1, DAG, LegalOperations)); // If 'unsafe math' is enabled, fold + // (fsub x, x) -> 0.0 & // (fsub x, (fadd x, y)) -> (fneg y) & // (fsub x, (fadd y, x)) -> (fneg y) if (DAG.getTarget().Options.UnsafeFPMath) { + if (N0 == N1) + return DAG.getConstantFP(0.0f, VT); + if (N1.getOpcode() == ISD::FADD) { SDValue N10 = N1->getOperand(0); SDValue N11 = N1->getOperand(1); diff --git a/test/CodeGen/ARM/unsafe-fsub.ll b/test/CodeGen/ARM/unsafe-fsub.ll new file mode 100644 index 0000000000..3a4477d315 --- /dev/null +++ b/test/CodeGen/ARM/unsafe-fsub.ll @@ -0,0 +1,18 @@ +; RUN: llc -march=arm -mcpu=cortex-a9 < %s | FileCheck -check-prefix=SAFE %s +; RUN: llc -march=arm -mcpu=cortex-a9 -enable-unsafe-fp-math < %s | FileCheck -check-prefix=FAST %s + +target triple = "armv7-apple-ios" + +; SAFE: test +; FAST: test +define float @test(float %x, float %y) { +entry: +; SAFE: vmul.f32 +; SAFE: vsub.f32 +; FAST: mov r0, #0 + %0 = fmul float %x, %y + %1 = fsub float %0, %0 + ret float %1 +} + + |