Make LSR's OptimizeShadowIV ignore induction variables with negative

strides for now, because it doesn't handle them correctly. This fixes a
miscompile of SingleSource/Benchmarks/Misc-C++/ray.

This problem was usually hidden because indvars transforms such induction
variables into negations of canonical induction variables.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85118 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Transforms/Scalar/LoopStrengthReduce.cpp b/lib/Transforms/Scalar/LoopStrengthReduce.cpp
index 4c305aae832f726822cd7890106c0b4fec49126f..e20fb16fb168ef358ed36afa404ab7aac1ab0775 100644 (file)
--- a/lib/Transforms/Scalar/LoopStrengthReduce.cpp
+++ b/lib/Transforms/Scalar/LoopStrengthReduce.cpp
@@ -2262,6 +2262,10 @@ void LoopStrengthReduce::OptimizeShadowIV(Loop *L) {
 
       if (!C) continue;
 
+      // Ignore negative constants, as the code below doesn't handle them
+      // correctly. TODO: Remove this restriction.
+      if (!C->getValue().isStrictlyPositive()) continue;
+
       /* Add new PHINode. */
       PHINode *NewPH = PHINode::Create(DestTy, "IV.S.", PH);
 

diff --git a/test/CodeGen/X86/negative-stride-fptosi-user.ll b/test/CodeGen/X86/negative-stride-fptosi-user.ll
new file mode 100644 (file)
index 0000000..332e0b9
--- /dev/null
+++ b/test/CodeGen/X86/negative-stride-fptosi-user.ll
@@ -0,0 +1,25 @@
+; RUN: llc < %s -march=x86-64 | grep cvtsi2sd
+
+; LSR previously eliminated the sitofp by introducing an induction
+; variable which stepped by a bogus ((double)UINT32_C(-1)). It's theoretically
+; possible to eliminate the sitofp using a proper -1.0 step though; this
+; test should be changed if that is done.
+
+define void @foo(i32 %N) nounwind {
+entry:
+  %0 = icmp slt i32 %N, 0                         ; <i1> [#uses=1]
+  br i1 %0, label %bb, label %return
+
+bb:                                               ; preds = %bb, %entry
+  %i.03 = phi i32 [ 0, %entry ], [ %2, %bb ]      ; <i32> [#uses=2]
+  %1 = sitofp i32 %i.03 to double                  ; <double> [#uses=1]
+  tail call void @bar(double %1) nounwind
+  %2 = add nsw i32 %i.03, -1                       ; <i32> [#uses=2]
+  %exitcond = icmp eq i32 %2, %N                  ; <i1> [#uses=1]
+  br i1 %exitcond, label %return, label %bb
+
+return:                                           ; preds = %bb, %entry
+  ret void
+}
+
+declare void @bar(double)