summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorTom Stellard <thomas.stellard@amd.com>2014-04-08 21:18:14 +0000
committerTom Stellard <thomas.stellard@amd.com>2014-04-08 21:18:14 +0000
commit76bf269759e5e18566e2cb3af119c1f509eba110 (patch)
treec7a530a4bf912dac9467e81d1d875d4fa5511bf9
parentdbb5c29d89d7d0e1d80a3863e436e4d2082e3c22 (diff)
downloadllvm-76bf269759e5e18566e2cb3af119c1f509eba110.tar.gz
llvm-76bf269759e5e18566e2cb3af119c1f509eba110.tar.bz2
llvm-76bf269759e5e18566e2cb3af119c1f509eba110.tar.xz
Merging r202273:
------------------------------------------------------------------------ r202273 | atrick | 2014-02-26 11:31:56 -0500 (Wed, 26 Feb 2014) | 4 lines Fix PR18165: LSR must avoid scaling factors that exceed the limit on truncated use. Patch by Michael Zolotukhin! ------------------------------------------------------------------------ git-svn-id: https://llvm.org/svn/llvm-project/llvm/branches/release_34@205796 91177308-0d34-0410-b5e6-96231b3b80d8
-rw-r--r--lib/Transforms/Scalar/LoopStrengthReduce.cpp12
-rw-r--r--test/Transforms/LoopStrengthReduce/pr18165.ll88
2 files changed, 100 insertions, 0 deletions
diff --git a/lib/Transforms/Scalar/LoopStrengthReduce.cpp b/lib/Transforms/Scalar/LoopStrengthReduce.cpp
index eff5268c44..6133962e42 100644
--- a/lib/Transforms/Scalar/LoopStrengthReduce.cpp
+++ b/lib/Transforms/Scalar/LoopStrengthReduce.cpp
@@ -3390,6 +3390,10 @@ void LSRInstance::GenerateICmpZeroScales(LSRUse &LU, unsigned LUIdx,
int64_t NewBaseOffset = (uint64_t)Base.BaseOffset * Factor;
if (NewBaseOffset / Factor != Base.BaseOffset)
continue;
+ // If the offset will be truncated at this use, check that it is in bounds.
+ if (!IntTy->isPointerTy() &&
+ !ConstantInt::isValueValidForType(IntTy, NewBaseOffset))
+ continue;
// Check that multiplying with the use offset doesn't overflow.
int64_t Offset = LU.MinOffset;
@@ -3398,6 +3402,10 @@ void LSRInstance::GenerateICmpZeroScales(LSRUse &LU, unsigned LUIdx,
Offset = (uint64_t)Offset * Factor;
if (Offset / Factor != LU.MinOffset)
continue;
+ // If the offset will be truncated at this use, check that it is in bounds.
+ if (!IntTy->isPointerTy() &&
+ !ConstantInt::isValueValidForType(IntTy, Offset))
+ continue;
Formula F = Base;
F.BaseOffset = NewBaseOffset;
@@ -3432,6 +3440,10 @@ void LSRInstance::GenerateICmpZeroScales(LSRUse &LU, unsigned LUIdx,
F.UnfoldedOffset = (uint64_t)F.UnfoldedOffset * Factor;
if (F.UnfoldedOffset / Factor != Base.UnfoldedOffset)
continue;
+ // If the offset will be truncated, check that it is in bounds.
+ if (!IntTy->isPointerTy() &&
+ !ConstantInt::isValueValidForType(IntTy, F.UnfoldedOffset))
+ continue;
}
// If we make it here and it's legal, add it.
diff --git a/test/Transforms/LoopStrengthReduce/pr18165.ll b/test/Transforms/LoopStrengthReduce/pr18165.ll
new file mode 100644
index 0000000000..89adef7bd4
--- /dev/null
+++ b/test/Transforms/LoopStrengthReduce/pr18165.ll
@@ -0,0 +1,88 @@
+; RUN: opt < %s -loop-reduce -S | FileCheck %s
+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-S128"
+target triple = "x86_64-apple-macosx10.9.0"
+
+; LSR shouldn't reuse IV if the resultant offset is not valid for the operand type.
+; CHECK-NOT: trunc i32 %.ph to i8
+
+%struct.anon = type { i32, i32, i32 }
+
+@c = global i32 1, align 4
+@.str = private unnamed_addr constant [4 x i8] c"%d\0A\00", align 1
+@b = common global i32 0, align 4
+@a = common global %struct.anon zeroinitializer, align 4
+@e = common global %struct.anon zeroinitializer, align 4
+@d = common global i32 0, align 4
+@f = common global i32 0, align 4
+@g = common global i32 0, align 4
+@h = common global i32 0, align 4
+
+; Function Attrs: nounwind optsize ssp uwtable
+define i32 @main() #0 {
+entry:
+ %0 = load i32* getelementptr inbounds (%struct.anon* @a, i64 0, i32 0), align 4, !tbaa !1
+ %tobool7.i = icmp eq i32 %0, 0
+ %.promoted.i = load i32* getelementptr inbounds (%struct.anon* @a, i64 0, i32 2), align 4, !tbaa !6
+ %f.promoted.i = load i32* @f, align 4, !tbaa !7
+ br label %for.body6.i.outer
+
+for.body6.i.outer: ; preds = %entry, %lor.end.i
+ %.ph = phi i32 [ %add.i, %lor.end.i ], [ 0, %entry ]
+ %or1512.i.ph = phi i32 [ %or15.i, %lor.end.i ], [ %f.promoted.i, %entry ]
+ %or1410.i.ph = phi i32 [ %or14.i, %lor.end.i ], [ %.promoted.i, %entry ]
+ %p.addr.16.i.ph = phi i8 [ %inc10.i, %lor.end.i ], [ -128, %entry ]
+ br i1 %tobool7.i, label %if.end9.i, label %lbl.loopexit.i
+
+lbl.loopexit.i: ; preds = %for.body6.i.outer, %lbl.loopexit.i
+ br label %lbl.loopexit.i
+
+if.end9.i: ; preds = %for.body6.i.outer
+ %inc10.i = add i8 %p.addr.16.i.ph, 1
+ %tobool12.i = icmp eq i8 %p.addr.16.i.ph, 0
+ br i1 %tobool12.i, label %lor.rhs.i, label %lor.end.i
+
+lor.rhs.i: ; preds = %if.end9.i
+ %1 = load i32* @b, align 4, !tbaa !7
+ %dec.i = add nsw i32 %1, -1
+ store i32 %dec.i, i32* @b, align 4, !tbaa !7
+ %tobool13.i = icmp ne i32 %1, 0
+ br label %lor.end.i
+
+lor.end.i: ; preds = %lor.rhs.i, %if.end9.i
+ %2 = phi i1 [ true, %if.end9.i ], [ %tobool13.i, %lor.rhs.i ]
+ %lor.ext.i = zext i1 %2 to i32
+ %or14.i = or i32 %lor.ext.i, %or1410.i.ph
+ %or15.i = or i32 %or14.i, %or1512.i.ph
+ %add.i = add nsw i32 %.ph, 2
+ %cmp.i = icmp slt i32 %add.i, 21
+ br i1 %cmp.i, label %for.body6.i.outer, label %fn1.exit
+
+fn1.exit: ; preds = %lor.end.i
+ store i32 0, i32* @g, align 4, !tbaa !7
+ store i32 %or14.i, i32* getelementptr inbounds (%struct.anon* @a, i64 0, i32 2), align 4, !tbaa !6
+ store i32 %or15.i, i32* @f, align 4, !tbaa !7
+ store i32 %add.i, i32* getelementptr inbounds (%struct.anon* @e, i64 0, i32 1), align 4, !tbaa !8
+ store i32 0, i32* @h, align 4, !tbaa !7
+ %3 = load i32* @b, align 4, !tbaa !7
+ %call1 = tail call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([4 x i8]* @.str, i64 0, i64 0), i32 %3) #2
+ ret i32 0
+}
+
+; Function Attrs: nounwind optsize
+declare i32 @printf(i8* nocapture readonly, ...) #1
+
+attributes #0 = { nounwind optsize ssp uwtable "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #1 = { nounwind optsize "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #2 = { nounwind optsize }
+
+!llvm.ident = !{!0}
+
+!0 = metadata !{metadata !"clang version 3.5 "}
+!1 = metadata !{metadata !2, metadata !3, i64 0}
+!2 = metadata !{metadata !"", metadata !3, i64 0, metadata !3, i64 4, metadata !3, i64 8}
+!3 = metadata !{metadata !"int", metadata !4, i64 0}
+!4 = metadata !{metadata !"omnipotent char", metadata !5, i64 0}
+!5 = metadata !{metadata !"Simple C/C++ TBAA"}
+!6 = metadata !{metadata !2, metadata !3, i64 8}
+!7 = metadata !{metadata !3, metadata !3, i64 0}
+!8 = metadata !{metadata !2, metadata !3, i64 4}