commit 3c42f1c3c90c5c20f3539048f1ed3f685cc914c6
author Yaxun Liu <Yaxun.Liu@amd.com> Fri Mar 02 16:22:32 2018 +0000
committer Yaxun Liu <Yaxun.Liu@amd.com> Fri Mar 02 16:22:32 2018 +0000
tree 0d8f309bd855bb5a5075af127ab1b97cc3c3f4d0
parent 505614bb4f695cb2d028726c8f37951f4ed831cf

LoopUnroll: respect pragma unroll when AllowRemainder is disabled

Currently when AllowRemainder is disabled, pragma unroll count is not
respected even though there is no remainder. This bug causes a loop
fully unrolled in many cases even though the user specifies a unroll
count. Especially it affects OpenCL/CUDA since in many cases a loop
contains convergent instructions and currently AllowRemainder is
disabled for such loops.

Differential Revision: https://reviews.llvm.org/D43826

llvm-svn: 326585

llvm/lib/Transforms/Scalar/LoopUnrollPass.cpp

diff --git a/llvm/lib/Transforms/Scalar/LoopUnrollPass.cpp b/llvm/lib/Transforms/Scalar/LoopUnrollPass.cpp
index 15e7da5..1e3bd25 100644
--- a/llvm/lib/Transforms/Scalar/LoopUnrollPass.cpp
+++ b/llvm/lib/Transforms/Scalar/LoopUnrollPass.cpp
@@ -729,7 +729,7 @@
     UP.Runtime = true;
     UP.AllowExpensiveTripCount = true;
     UP.Force = true;
-    if (UP.AllowRemainder &&
+    if ((UP.AllowRemainder || (TripMultiple % PragmaCount == 0)) &&
         getUnrolledLoopSize(LoopSize, UP) < PragmaUnrollThreshold)
       return true;
   }