Check sign to detect overflow before changing compare stride.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@54710 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Transforms/Scalar/LoopStrengthReduce.cpp b/lib/Transforms/Scalar/LoopStrengthReduce.cpp
index 3d82fd279c5432b085ab9fe9e10e06556261cf2b..f47a5afcfc55af82ed48e7eda290621a0ad1769b 100644 (file)
--- a/lib/Transforms/Scalar/LoopStrengthReduce.cpp
+++ b/lib/Transforms/Scalar/LoopStrengthReduce.cpp
@@ -1542,6 +1542,12 @@ ICmpInst *LoopStrengthReduce::ChangeCompareStride(Loop *L, ICmpInst *Cond,
   Value *NewIncV = NULL;
   int64_t Scale = 1;
 
+  // Check stride constant and the comparision constant signs to detect
+  // overflow.
+  if (ICmpInst::isSignedPredicate(Predicate) &&
+      (CmpVal & SignBit) != (CmpSSInt & SignBit))
+    return Cond;
+
   // Look for a suitable stride / iv as replacement.
   std::stable_sort(StrideOrder.begin(), StrideOrder.end(), StrideCompare());
   for (unsigned i = 0, e = StrideOrder.size(); i != e; ++i) {
@@ -1640,11 +1646,12 @@ ICmpInst *LoopStrengthReduce::ChangeCompareStride(Loop *L, ICmpInst *Cond,
   // before the branch. See
   // test/Transforms/LoopStrengthReduce/change-compare-stride-trickiness-*.ll
   // for an example of this situation.
-  if (!Cond->hasOneUse())
+  if (!Cond->hasOneUse()) {
     for (BasicBlock::iterator I = Cond, E = Cond->getParent()->end();
          I != E; ++I)
       if (I == NewIncV)
         return Cond;
+  }
 
   if (NewCmpVal != CmpVal) {
     // Create a new compare instruction using new stride / iv.

diff --git a/test/Transforms/LoopStrengthReduce/2008-08-13-CmpStride.ll b/test/Transforms/LoopStrengthReduce/2008-08-13-CmpStride.ll
new file mode 100644 (file)
index 0000000..fbd3c1e
--- /dev/null
+++ b/test/Transforms/LoopStrengthReduce/2008-08-13-CmpStride.ll
@@ -0,0 +1,31 @@
+; RUN: llvm-as < %s | opt -loop-reduce | llvm-dis | grep add | count 2
+; PR 2662
+@g_3 = common global i16 0             ; <i16*> [#uses=2]
+@"\01LC" = internal constant [4 x i8] c"%d\0A\00"              ; <[4 x i8]*> [#uses=1]
+
+define void @func_1() nounwind {
+entry:
+       br label %bb
+
+bb:            ; preds = %bb, %entry
+       %l_2.0.reg2mem.0 = phi i16 [ 0, %entry ], [ %t1, %bb ]          ; <i16> [#uses=2]
+       %t0 = shl i16 %l_2.0.reg2mem.0, 1               ; <i16>:0 [#uses=1]
+       volatile store i16 %t0, i16* @g_3, align 2
+       %t1 = add i16 %l_2.0.reg2mem.0, -3              ; <i16>:1 [#uses=2]
+       %t2 = icmp slt i16 %t1, 1               ; <i1>:2 [#uses=1]
+       br i1 %t2, label %bb, label %return
+
+return:                ; preds = %bb
+       ret void
+}
+
+define i32 @main() nounwind {
+entry:
+       tail call void @func_1( ) nounwind
+       volatile load i16* @g_3, align 2                ; <i16>:0 [#uses=1]
+       zext i16 %0 to i32              ; <i32>:1 [#uses=1]
+       tail call i32 (i8*, ...)* @printf( i8* getelementptr ([4 x i8]* @"\01LC", i32 0, i32 0), i32 %1 ) nounwind              ; <i32>:2 [#uses=0]
+       ret i32 0
+}
+
+declare i32 @printf(i8*, ...) nounwind