When looking for dependencies on the src pointer, scan the src pointer. Scanning

on the memcpy call will pull up other unrelated stuff. Fixes PR11142.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@142150 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Transforms/Scalar/MemCpyOptimizer.cpp b/lib/Transforms/Scalar/MemCpyOptimizer.cpp
index eeb8931446dcada7e84897c41016ecc25d02bbba..298d6927497ca86c7fee6c45909038489cecb334 100644 (file)
--- a/lib/Transforms/Scalar/MemCpyOptimizer.cpp
+++ b/lib/Transforms/Scalar/MemCpyOptimizer.cpp
@@ -806,21 +806,26 @@ bool MemCpyOpt::processMemCpy(MemCpyInst *M) {
   //   a) memcpy-memcpy xform which exposes redundance for DSE.
   //   b) call-memcpy xform for return slot optimization.
   MemDepResult DepInfo = MD->getDependency(M);
-  if (!DepInfo.isClobber())
-    return false;
-  
-  if (MemCpyInst *MDep = dyn_cast<MemCpyInst>(DepInfo.getInst()))
-    return processMemCpyMemCpyDependence(M, MDep, CopySize->getZExtValue());
-    
-  if (CallInst *C = dyn_cast<CallInst>(DepInfo.getInst())) {
-    if (performCallSlotOptzn(M, M->getDest(), M->getSource(),
-                             CopySize->getZExtValue(), C)) {
-      MD->removeInstruction(M);
-      M->eraseFromParent();
-      return true;
+  if (DepInfo.isClobber()) {
+    if (CallInst *C = dyn_cast<CallInst>(DepInfo.getInst())) {
+      if (performCallSlotOptzn(M, M->getDest(), M->getSource(),
+                               CopySize->getZExtValue(), C)) {
+        MD->removeInstruction(M);
+        M->eraseFromParent();
+        return true;
+      }
     }
   }
   
+  AliasAnalysis &AA = getAnalysis<AliasAnalysis>();
+  AliasAnalysis::Location SrcLoc = AA.getLocationForSource(M);
+  MemDepResult SrcDepInfo = MD->getPointerDependencyFrom(SrcLoc, true,
+                                                         M, M->getParent());
+  if (SrcDepInfo.isClobber()) {
+    if (MemCpyInst *MDep = dyn_cast<MemCpyInst>(SrcDepInfo.getInst()))
+      return processMemCpyMemCpyDependence(M, MDep, CopySize->getZExtValue());
+  }
+
   return false;
 }
 

diff --git a/test/Transforms/MemCpyOpt/memcpy.ll b/test/Transforms/MemCpyOpt/memcpy.ll
index 71d4d4e8a11f03f422ece78521b65f9c3140f7bf..e828e4419fe0b8acb37d91b37888f6264d6a9362 100644 (file)
--- a/test/Transforms/MemCpyOpt/memcpy.ll
+++ b/test/Transforms/MemCpyOpt/memcpy.ll
@@ -59,7 +59,7 @@ define void @test3(%0* noalias sret %agg.result) nounwind  {
   call void @llvm.memcpy.p0i8.p0i8.i32(i8* %agg.result2, i8* %x.01, i32 32, i32 16, i1 false)
   ret void
 ; CHECK: @test3
-; CHECK-NEXT: %agg.result2 = bitcast 
+; CHECK-NEXT: %agg.result1 = bitcast 
 ; CHECK-NEXT: call void @llvm.memcpy
 ; CHECK-NEXT: ret void
 }
@@ -130,3 +130,21 @@ declare i32 @g(%struct.p* byval align 8)
 
 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* nocapture, i8* nocapture, i32, i32, i1) nounwind
 
+; PR11142 - When looking for a memcpy-memcpy dependency, don't get stuck on
+; instructions between the memcpy's that only affect the destination pointer.
+@test8.str = internal constant [7 x i8] c"ABCDEF\00"
+
+define void @test8() {
+; CHECK: test8
+  %A = tail call i8* @malloc(i32 10)
+  %B = getelementptr inbounds i8* %A, i64 2
+  tail call void @llvm.memcpy.p0i8.p0i8.i32(i8* %B, i8* getelementptr inbounds ([7 x i8]* @test8.str, i64 0, i64 0), i32 7, i32 1, i1 false)
+; CHECK: tail call void @llvm.memcpy.p0i8.p0i8.i32(i8* %D, i8* getelementptr
+  %C = tail call i8* @malloc(i32 10)
+  %D = getelementptr inbounds i8* %C, i64 2
+  tail call void @llvm.memcpy.p0i8.p0i8.i32(i8* %D, i8* %B, i32 7, i32 1, i1 false)
+; CHECK: tail call void @llvm.memcpy.p0i8.p0i8.i32(i8* %D, i8* getelementptr
+  ret void
+}
+
+declare noalias i8* @malloc(i32)