summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorNick Lewycky <nicholas@mxc.ca>2014-03-26 23:45:15 +0000
committerNick Lewycky <nicholas@mxc.ca>2014-03-26 23:45:15 +0000
commitce49ab2b0571074fefd093160f68e5889578e21b (patch)
treeb1d0cd3f615291c848290417ad26cf88e7ca263c
parent1e0751bd4c515b3d36f051e3aa4ececf2c056201 (diff)
downloadllvm-ce49ab2b0571074fefd093160f68e5889578e21b.tar.gz
llvm-ce49ab2b0571074fefd093160f68e5889578e21b.tar.bz2
llvm-ce49ab2b0571074fefd093160f68e5889578e21b.tar.xz
Treat lifetime.start'd memory like we treat freshly alloca'd memory. Patch by Björn Steinbrink!
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204876 91177308-0d34-0410-b5e6-96231b3b80d8
-rw-r--r--lib/Transforms/Scalar/MemCpyOptimizer.cpp20
-rw-r--r--test/Transforms/MemCpyOpt/memcpy-undef.ll21
2 files changed, 37 insertions, 4 deletions
diff --git a/lib/Transforms/Scalar/MemCpyOptimizer.cpp b/lib/Transforms/Scalar/MemCpyOptimizer.cpp
index 143ba38ec9..2603c969c5 100644
--- a/lib/Transforms/Scalar/MemCpyOptimizer.cpp
+++ b/lib/Transforms/Scalar/MemCpyOptimizer.cpp
@@ -851,9 +851,9 @@ bool MemCpyOpt::processMemCpy(MemCpyInst *M) {
// The are three possible optimizations we can do for memcpy:
// a) memcpy-memcpy xform which exposes redundance for DSE.
// b) call-memcpy xform for return slot optimization.
- // c) memcpy from freshly alloca'd space copies undefined data, and we can
- // therefore eliminate the memcpy in favor of the data that was already
- // at the destination.
+ // c) memcpy from freshly alloca'd space or space that has just started its
+ // lifetime copies undefined data, and we can therefore eliminate the
+ // memcpy in favor of the data that was already at the destination.
MemDepResult DepInfo = MD->getDependency(M);
if (DepInfo.isClobber()) {
if (CallInst *C = dyn_cast<CallInst>(DepInfo.getInst())) {
@@ -874,7 +874,19 @@ bool MemCpyOpt::processMemCpy(MemCpyInst *M) {
if (MemCpyInst *MDep = dyn_cast<MemCpyInst>(SrcDepInfo.getInst()))
return processMemCpyMemCpyDependence(M, MDep, CopySize->getZExtValue());
} else if (SrcDepInfo.isDef()) {
- if (isa<AllocaInst>(SrcDepInfo.getInst())) {
+ Instruction *I = SrcDepInfo.getInst();
+ bool hasUndefContents = false;
+
+ if (isa<AllocaInst>(I)) {
+ hasUndefContents = true;
+ } else if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(I)) {
+ if (II->getIntrinsicID() == Intrinsic::lifetime_start)
+ if (ConstantInt *LTSize = dyn_cast<ConstantInt>(II->getArgOperand(0)))
+ if (LTSize->getZExtValue() >= CopySize->getZExtValue())
+ hasUndefContents = true;
+ }
+
+ if (hasUndefContents) {
MD->removeInstruction(M);
M->eraseFromParent();
++NumMemCpyInstr;
diff --git a/test/Transforms/MemCpyOpt/memcpy-undef.ll b/test/Transforms/MemCpyOpt/memcpy-undef.ll
index fd4965f4dc..663b8dcfd3 100644
--- a/test/Transforms/MemCpyOpt/memcpy-undef.ll
+++ b/test/Transforms/MemCpyOpt/memcpy-undef.ll
@@ -21,5 +21,26 @@ define i32 @test1(%struct.foo* nocapture %foobie) nounwind noinline ssp uwtable
; CHECK-NOT: call void @llvm.memcpy
}
+define void @test2(i8* sret noalias nocapture %out, i8* %in) nounwind noinline ssp uwtable {
+ call void @llvm.lifetime.start(i64 8, i8* %in)
+ call void @llvm.memcpy.p0i8.p0i8.i64(i8* %out, i8* %in, i64 8, i32 1, i1 false)
+ ret void
+
+; Check that the memcpy is removed.
+; CHECK-LABEL: @test2(
+; CHECK-NOT: call void @llvm.memcpy
+}
+
+define void @test3(i8* sret noalias nocapture %out, i8* %in) nounwind noinline ssp uwtable {
+ call void @llvm.lifetime.start(i64 4, i8* %in)
+ call void @llvm.memcpy.p0i8.p0i8.i64(i8* %out, i8* %in, i64 8, i32 1, i1 false)
+ ret void
+
+; Check that the memcpy is not removed.
+; CHECK-LABEL: @test3(
+; CHECK: call void @llvm.memcpy
+}
+
declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture, i8* nocapture, i64, i32, i1) nounwind
+declare void @llvm.lifetime.start(i64, i8* nocapture) nounwind