[llvm-commits] [llvm] r132485 - in /llvm/trunk: lib/Transforms/Scalar/MemCpyOptimizer.cpp test/Transforms/MemCpyOpt/2011-06-02-CallSlotOverwritten.ll

[Eli Friedman]

Author: efriedma
Date: Thu Jun  2 16:24:42 2011
New Revision: 132485

URL: http://llvm.org/viewvc/llvm-project?rev=132485&view=rev
Log:
PR10067: Add missing safety check to call return transformation in MemCpyOpt::processStore.  If something accesses the dest of the "copy" between the call and the copy, the performCallSlotOptzn transformation is not valid.


Added:
    llvm/trunk/test/Transforms/MemCpyOpt/2011-06-02-CallSlotOverwritten.ll
Modified:
    llvm/trunk/lib/Transforms/Scalar/MemCpyOptimizer.cpp

Modified: llvm/trunk/lib/Transforms/Scalar/MemCpyOptimizer.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/MemCpyOptimizer.cpp?rev=132485&r1=132484&r2=132485&view=diff
==============================================================================
--- llvm/trunk/lib/Transforms/Scalar/MemCpyOptimizer.cpp (original)
+++ llvm/trunk/lib/Transforms/Scalar/MemCpyOptimizer.cpp Thu Jun  2 16:24:42 2011
@@ -488,11 +488,28 @@
   // a memcpy.
   if (LoadInst *LI = dyn_cast<LoadInst>(SI->getOperand(0))) {
     if (!LI->isVolatile() && LI->hasOneUse()) {
-      MemDepResult dep = MD->getDependency(LI);
+      MemDepResult ldep = MD->getDependency(LI);
       CallInst *C = 0;
-      if (dep.isClobber() && !isa<MemCpyInst>(dep.getInst()))
-        C = dyn_cast<CallInst>(dep.getInst());
-      
+      if (ldep.isClobber() && !isa<MemCpyInst>(ldep.getInst()))
+        C = dyn_cast<CallInst>(ldep.getInst());
+
+      if (C) {
+        // Check that nothing touches the dest of the "copy" between
+        // the call and the store.
+        MemDepResult sdep = MD->getDependency(SI);
+        if (!sdep.isNonLocal()) {
+          bool FoundCall = false;
+          for (BasicBlock::iterator I = SI, E = sdep.getInst(); I != E; --I) {
+            if (&*I == C) {
+              FoundCall = true;
+              break;
+            }
+          }
+          if (!FoundCall)
+            C = 0;
+        }
+      }
+
       if (C) {
         bool changed = performCallSlotOptzn(LI,
                         SI->getPointerOperand()->stripPointerCasts(), 

Added: llvm/trunk/test/Transforms/MemCpyOpt/2011-06-02-CallSlotOverwritten.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/MemCpyOpt/2011-06-02-CallSlotOverwritten.ll?rev=132485&view=auto
==============================================================================
--- llvm/trunk/test/Transforms/MemCpyOpt/2011-06-02-CallSlotOverwritten.ll (added)
+++ llvm/trunk/test/Transforms/MemCpyOpt/2011-06-02-CallSlotOverwritten.ll Thu Jun  2 16:24:42 2011
@@ -0,0 +1,36 @@
+; RUN: opt < %s -basicaa -memcpyopt -S | FileCheck %s
+; PR10067
+; Make sure the call+copy isn't optimized in such a way that
+; %ret ends up with the wrong value.
+
+target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128-n8:16:32"
+target triple = "i386-apple-darwin10"
+
+%struct1 = type { i32, i32 }
+%struct2 = type { %struct1, i8* }
+
+declare void @bar(%struct1* nocapture sret %agg.result) nounwind
+
+define i32 @foo() nounwind {
+  %x = alloca %struct1, align 8
+  %y = alloca %struct2, align 8
+  call void @bar(%struct1* sret %x) nounwind
+; CHECK: call void @bar(%struct1* sret %x)
+
+  %gepn1 = getelementptr inbounds %struct2* %y, i32 0, i32 0, i32 0
+  store i32 0, i32* %gepn1, align 8
+  %gepn2 = getelementptr inbounds %struct2* %y, i32 0, i32 0, i32 1
+  store i32 0, i32* %gepn2, align 4
+
+  %bit1 = bitcast %struct1* %x to i64*
+  %bit2 = bitcast %struct2* %y to i64*
+  %load = load i64* %bit1, align 8
+  store i64 %load, i64* %bit2, align 8
+
+; CHECK: %load = load i64* %bit1, align 8
+; CHECK: store i64 %load, i64* %bit2, align 8
+
+  %gep1 = getelementptr %struct2* %y, i32 0, i32 0, i32 0
+  %ret = load i32* %gep1
+  ret i32 %ret
+}