[PATCH] D31753: [LoopPeeling] Fix loop size check for non-forced invariant-based peeling

[Max Kazantsev via Phabricator via llvm-commits]

mkazantsev created this revision.

When peeling loops basing on phis becoming invariants, we make a wrong loop size check.
UP.Threshold should be compared against the total numbers of instructions after the transformation,
which is equal to 2 * LoopSize in case of peeling one iteration. Currently we have a wrong check
against just LoopSize.


https://reviews.llvm.org/D31753

Files:
  lib/Transforms/Utils/LoopUnrollPeel.cpp
  test/Transforms/LoopUnroll/peel-loop-not-forced.ll


Index: test/Transforms/LoopUnroll/peel-loop-not-forced.ll
===================================================================
--- test/Transforms/LoopUnroll/peel-loop-not-forced.ll
+++ test/Transforms/LoopUnroll/peel-loop-not-forced.ll
@@ -1,4 +1,4 @@
-; RUN: opt < %s -S -loop-unroll -unroll-threshold=4 | FileCheck %s
+; RUN: opt < %s -S -loop-unroll -unroll-threshold=8 | FileCheck %s
 
 define i32 @invariant_backedge_1(i32 %a, i32 %b) {
 ; CHECK-LABEL: @invariant_backedge_1
Index: lib/Transforms/Utils/LoopUnrollPeel.cpp
===================================================================
--- lib/Transforms/Utils/LoopUnrollPeel.cpp
+++ lib/Transforms/Utils/LoopUnrollPeel.cpp
@@ -82,7 +82,7 @@
   // its only back edge. If there is such Phi, peeling 1 iteration from the
   // loop is profitable, because starting from 2nd iteration we will have an
   // invariant instead of this Phi.
-  if (LoopSize <= UP.Threshold) {
+  if (2 * LoopSize <= UP.Threshold) {
     BasicBlock *BackEdge = L->getLoopLatch();
     assert(BackEdge && "Loop is not in simplified form?");
     BasicBlock *Header = L->getHeader();


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D31753.94347.patch
Type: text/x-patch
Size: 1116 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20170406/f15eefad/attachment.bin>