Fix a bug that caused globalopt to miscompile tramp3d: don't miss

unruly indices for arrays that are members of structs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79337 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Transforms/IPO/GlobalOpt.cpp b/lib/Transforms/IPO/GlobalOpt.cpp
index 46ff307..6ec2012 100644 (file)
--- a/lib/Transforms/IPO/GlobalOpt.cpp
+++ b/lib/Transforms/IPO/GlobalOpt.cpp
@@ -426,13 +426,18 @@ static bool IsUserOfGlobalSafeForSRA(User *U, GlobalValue *GV) {
     // Scalar replacing *just* the outer index of the array is probably not
     // going to be a win anyway, so just give up.
     for (++GEPI; // Skip array index.
-         GEPI != E && (isa<ArrayType>(*GEPI) || isa<VectorType>(*GEPI));
+         GEPI != E;
          ++GEPI) {
       uint64_t NumElements;
       if (const ArrayType *SubArrayTy = dyn_cast<ArrayType>(*GEPI))
         NumElements = SubArrayTy->getNumElements();
-      else
-        NumElements = cast<VectorType>(*GEPI)->getNumElements();
+      else if (const VectorType *SubVectorTy = dyn_cast<VectorType>(*GEPI))
+        NumElements = SubVectorTy->getNumElements();
+      else {
+        assert(isa<StructType>(*GEPI) &&
+               "Indexed GEP type is not array, vector, or struct!");
+        continue;
+      }
       
       ConstantInt *IdxVal = dyn_cast<ConstantInt>(GEPI.getOperand());
       if (!IdxVal || IdxVal->getZExtValue() >= NumElements)

diff --git a/test/Transforms/GlobalOpt/globalsra-unknown-index.ll b/test/Transforms/GlobalOpt/globalsra-unknown-index.ll
new file mode 100644 (file)
index 0000000..7a5d5b7
--- /dev/null
+++ b/test/Transforms/GlobalOpt/globalsra-unknown-index.ll
@@ -0,0 +1,41 @@
+; RUN: llvm-as < %s | opt -globalopt | llvm-dis > %t
+; RUN: grep {@Y = internal global \\\[3 x \[%\]struct.X\\\] zeroinitializer} %t
+; RUN: grep load %t | count 6
+; RUN: grep {add i32 \[%\]a, \[%\]b} %t | count 3
+
+; globalopt should not sra the global, because it can't see the index.
+
+%struct.X = type { [3 x i32], [3 x i32] }
+
+@Y = internal global [3 x %struct.X] zeroinitializer
+
+@addr = external global i8
+
+define void @frob() {
+  store i32 1, i32* getelementptr inbounds ([3 x %struct.X]* @Y, i64 0, i64 0, i32 0, i64 ptrtoint (i8* @addr to i64)), align 4
+  ret void
+}
+define i32 @borf(i64 %i, i64 %j) {
+  %p = getelementptr inbounds [3 x %struct.X]* @Y, i64 0, i64 0, i32 0, i64 0
+  %a = load i32* %p
+  %q = getelementptr inbounds [3 x %struct.X]* @Y, i64 0, i64 0, i32 1, i64 0
+  %b = load i32* %q
+  %c = add i32 %a, %b
+  ret i32 %c
+}
+define i32 @borg(i64 %i, i64 %j) {
+  %p = getelementptr inbounds [3 x %struct.X]* @Y, i64 0, i64 1, i32 0, i64 1
+  %a = load i32* %p
+  %q = getelementptr inbounds [3 x %struct.X]* @Y, i64 0, i64 1, i32 1, i64 1
+  %b = load i32* %q
+  %c = add i32 %a, %b
+  ret i32 %c
+}
+define i32 @borh(i64 %i, i64 %j) {
+  %p = getelementptr inbounds [3 x %struct.X]* @Y, i64 0, i64 2, i32 0, i64 2
+  %a = load i32* %p
+  %q = getelementptr inbounds [3 x %struct.X]* @Y, i64 0, i64 2, i32 1, i64 2
+  %b = load i32* %q
+  %c = add i32 %a, %b
+  ret i32 %c
+}