diff options
| author | Michael Zolotukhin <mzolotukhin@apple.com> | 2014-11-20 20:19:55 +0000 |
|---|---|---|
| committer | Michael Zolotukhin <mzolotukhin@apple.com> | 2014-11-20 20:19:55 +0000 |
| commit | 0dcae714493752ed441ee827a7d3abc95c98e499 (patch) | |
| tree | eb6cbc4aec42108f046b857911f82d0b3ba5c6a2 /llvm/test/Transforms/LoopUnroll/tripcount-overflow.ll | |
| parent | 365eb05c705800f952ec9d6847478e600fffe4e7 (diff) | |
| download | bcm5719-llvm-0dcae714493752ed441ee827a7d3abc95c98e499.tar.gz bcm5719-llvm-0dcae714493752ed441ee827a7d3abc95c98e499.zip | |
Fix a trip-count overflow issue in LoopUnroll.
Currently LoopUnroll generates a prologue loop before the main loop
body to execute first N%UnrollFactor iterations. Also, this loop is
used if trip-count can overflow - it's determined by a runtime check.
However, we've been mistakenly optimizing this loop to a linear code for
UnrollFactor = 2, not taking into account that it also serves as a safe
version of the loop if its trip-count overflows.
llvm-svn: 222451
Diffstat (limited to 'llvm/test/Transforms/LoopUnroll/tripcount-overflow.ll')
| -rw-r--r-- | llvm/test/Transforms/LoopUnroll/tripcount-overflow.ll | 30 |
1 files changed, 30 insertions, 0 deletions
diff --git a/llvm/test/Transforms/LoopUnroll/tripcount-overflow.ll b/llvm/test/Transforms/LoopUnroll/tripcount-overflow.ll new file mode 100644 index 00000000000..d59368578ec --- /dev/null +++ b/llvm/test/Transforms/LoopUnroll/tripcount-overflow.ll @@ -0,0 +1,30 @@ +; RUN: opt < %s -S -unroll-runtime -unroll-count=2 -loop-unroll | FileCheck %s +target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128" + +; When prologue is fully unrolled, the branch on its end is unconditional. +; Unrolling it is illegal if we can't prove that trip-count+1 doesn't overflow, +; like in this example, where it comes from an argument. +; +; This test is based on an example from here: +; http://stackoverflow.com/questions/23838661/why-is-clang-optimizing-this-code-out +; +; CHECK: while.body.prol: +; CHECK: br i1 +; CHECK: entry.split: + +; Function Attrs: nounwind readnone ssp uwtable +define i32 @foo(i32 %N) #0 { +entry: + br label %while.body + +while.body: ; preds = %while.body, %entry + %i = phi i32 [ 0, %entry ], [ %inc, %while.body ] + %cmp = icmp eq i32 %i, %N + %inc = add i32 %i, 1 + br i1 %cmp, label %while.end, label %while.body + +while.end: ; preds = %while.body + ret i32 %i +} + +attributes #0 = { nounwind readnone 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" } |
