[InstCombine] Don't replace allocas with smaller globals

Summary:
InstCombine replaces large allocas with small globals consts causing buffer overflows
on valid code, see PR33372.

This fix permits this optimization only if the global is dereference for alloca size.

Fixes PR33372

Reviewers: eugenis, majnemer, chandlerc

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D34311

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@306194 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp b/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp
index a4d84ae81aa02489d3ffaf96808252d2cc96f722..ee30aa44674304401988dacf58eadce241ee757e 100644 (file)
--- a/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp
+++ b/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp
@@ -169,6 +169,18 @@ isOnlyCopiedFromConstantGlobal(AllocaInst *AI,
   return nullptr;
 }
 
+/// Returns true if V is dereferenceable for size of alloca.
+static bool isDereferenceableForAllocaSize(const Value *V, const AllocaInst *AI,
+                                           const DataLayout &DL) {
+  if (AI->isArrayAllocation())
+    return false;
+  uint64_t AllocaSize = DL.getTypeStoreSize(AI->getAllocatedType());
+  if (!AllocaSize)
+    return false;
+  return isDereferenceableAndAlignedPointer(V, AI->getAlignment(),
+                                            APInt(64, AllocaSize), DL);
+}
+
 static Instruction *simplifyAllocaArraySize(InstCombiner &IC, AllocaInst &AI) {
   // Check for array size of 1 (scalar allocation).
   if (!AI.isArrayAllocation()) {
@@ -390,7 +402,8 @@ Instruction *InstCombiner::visitAllocaInst(AllocaInst &AI) {
     if (MemTransferInst *Copy = isOnlyCopiedFromConstantGlobal(&AI, ToDelete)) {
       unsigned SourceAlign = getOrEnforceKnownAlignment(
           Copy->getSource(), AI.getAlignment(), DL, &AI, &AC, &DT);
-      if (AI.getAlignment() <= SourceAlign) {
+      if (AI.getAlignment() <= SourceAlign &&
+          isDereferenceableForAllocaSize(Copy->getSource(), &AI, DL)) {
         DEBUG(dbgs() << "Found alloca equal to global: " << AI << '\n');
         DEBUG(dbgs() << "  memcpy = " << *Copy << '\n');
         for (unsigned i = 0, e = ToDelete.size(); i != e; ++i)

diff --git a/test/Transforms/InstCombine/memcpy-from-global.ll b/test/Transforms/InstCombine/memcpy-from-global.ll
index 7c9384d89ba340a0281cfbf1fc5d7aae02143910..c8b329e94b06e8a78cf60bd3a29a66c5b997f25a 100644 (file)
--- a/test/Transforms/InstCombine/memcpy-from-global.ll
+++ b/test/Transforms/InstCombine/memcpy-from-global.ll
@@ -204,3 +204,34 @@ define void @test9_addrspacecast() {
 ; CHECK-NEXT: call void @bar(i8* bitcast (%U* getelementptr inbounds ([2 x %U], [2 x %U]* @H, i64 0, i64 1) to i8*))
   ret void
 }
+
+@bbb = local_unnamed_addr global [1000000 x i8] zeroinitializer, align 16
+@_ZL3KKK = internal unnamed_addr constant [3 x i8] c"\01\01\02", align 1
+
+; Should not replace alloca with global because of size mismatch.
+define void @test9_small_global() {
+; CHECK-LABEL: @test9_small_global(
+; CHECK-NOT: call void @llvm.memcpy.p0i8.p0i8.i64({{.*}}@bbb,{{.*}}@_ZL3KKK, 
+; CHECK: alloca [1000000 x i8]
+entry:
+  %cc = alloca [1000000 x i8], align 16
+  %cc.0..sroa_idx = getelementptr inbounds [1000000 x i8], [1000000 x i8]* %cc, i64 0, i64 0
+  %arraydecay = getelementptr inbounds [1000000 x i8], [1000000 x i8]* %cc, i32 0, i32 0
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* %arraydecay, i8* getelementptr inbounds ([3 x i8], [3 x i8]* @_ZL3KKK, i32 0, i32 0), i64 3, i32 1, i1 false)
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* getelementptr inbounds ([1000000 x i8], [1000000 x i8]* @bbb, i32 0, i32 0), i8* %arraydecay, i64 1000000, i32 16, i1 false)
+  ret void
+}
+
+; Should replace alloca with global as they have exactly the same size.
+define void @test10_same_global() {
+; CHECK-LABEL: @test10_same_global(
+; CHECK-NOT: alloca
+; CHECK: call void @llvm.memcpy.p0i8.p0i8.i64({{.*}}@bbb,{{.*}}@_ZL3KKK,{{.*}}, i64 3,
+entry:
+  %cc = alloca [3 x i8], align 1
+  %cc.0..sroa_idx = getelementptr inbounds [3 x i8], [3 x i8]* %cc, i64 0, i64 0
+  %arraydecay = getelementptr inbounds [3 x i8], [3 x i8]* %cc, i32 0, i32 0
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* %arraydecay, i8* getelementptr inbounds ([3 x i8], [3 x i8]* @_ZL3KKK, i32 0, i32 0), i64 3, i32 1, i1 false)
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* getelementptr inbounds ([1000000 x i8], [1000000 x i8]* @bbb, i32 0, i32 0), i8* %arraydecay, i64 3, i32 1, i1 false)
+  ret void
+}