Make the MemCpyOptimizer a bit more aggressive. I can't think of a scenerio

where this would be bad as the backend shouldn't have a problem inlining small
memcpys.
rdar://10510150

llvm-svn: 145865

2 files changed, 18 insertions, 2 deletions

diff --git a/llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp b/llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp
index 9e4f51f4549..6c6db60cc3e 100644
--- a/llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp
+++ b/llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp
@@ -148,7 +148,7 @@ struct MemsetRange {
 
 bool MemsetRange::isProfitableToUseMemset(const TargetData &TD) const {
   // If we found more than 8 stores to merge or 64 bytes, use memset.
-  if (TheStores.size() >= 8 || End-Start >= 64) return true;
+  if (TheStores.size() >= 4 || End-Start >= 16) return true;
 
   // If there is nothing to merge, don't do anything.
   if (TheStores.size() < 2) return false;
diff --git a/llvm/test/Transforms/MemCpyOpt/form-memset.ll b/llvm/test/Transforms/MemCpyOpt/form-memset.ll
index 1ac97e9e6b9..e5ace332702 100644
--- a/llvm/test/Transforms/MemCpyOpt/form-memset.ll
+++ b/llvm/test/Transforms/MemCpyOpt/form-memset.ll
@@ -57,8 +57,8 @@ entry:
 
 declare i32 @bar(...)
 
+%struct.MV = type { i16, i16 }
 
-	%struct.MV = type { i16, i16 }
 
 define void @test2() nounwind  {
 entry:
@@ -220,3 +220,19 @@ entry:
 ; CHECK: call void @llvm.memset.p0i8.i64(i8* %2, i8 0, i64 24, i32 1, i1 false)
 }
 
+; More aggressive heuristic
+; rdar://9892684
+define void @test7(i32* nocapture %c) nounwind optsize {
+  store i32 -1, i32* %c, align 4
+  %1 = getelementptr inbounds i32* %c, i32 1
+  store i32 -1, i32* %1, align 4
+  %2 = getelementptr inbounds i32* %c, i32 2
+  store i32 -1, i32* %2, align 4
+  %3 = getelementptr inbounds i32* %c, i32 3
+  store i32 -1, i32* %3, align 4
+  %4 = getelementptr inbounds i32* %c, i32 4
+  store i32 -1, i32* %4, align 4
+; CHECK: @test7
+; CHECK: call void @llvm.memset.p0i8.i64(i8* %5, i8 -1, i64 20, i32 4, i1 false)
+  ret void
+}