[PATCH] D63016: LoopVersioning: Respect convergent
Matt Arsenault via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Fri Jun 7 11:24:59 PDT 2019
arsenm created this revision.
arsenm added reviewers: anemet, fhahn, Meinersbur, kbarton, jdoerfert.
Herald added a subscriber: wdng.
This changes the standalone pass only. Arguably the utility class
itself should assert there are no convergent calls. However, a target
pass with additional context may still be able to version a loop if
all of the dynamic conditions are sufficiently uniform.
https://reviews.llvm.org/D63016
Files:
lib/Transforms/Utils/LoopVersioning.cpp
test/Transforms/LoopVersioning/convergent.ll
Index: test/Transforms/LoopVersioning/convergent.ll
===================================================================
--- /dev/null
+++ test/Transforms/LoopVersioning/convergent.ll
@@ -0,0 +1,40 @@
+; RUN: opt -basicaa -loop-versioning -S < %s | FileCheck %s
+
+target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
+
+; Do not version this loop because of a convergent operation
+
+; CHECK-LABEL: @f(
+; CHECK: call i32 @llvm.convergent(
+; CHECK-NOT: call i32 @llvm.convergent(
+define void @f(i32* %a, i32* %b, i32* %c) #0 {
+entry:
+ br label %for.body
+
+for.body: ; preds = %for.body, %entry
+ %ind = phi i64 [ 0, %entry ], [ %add, %for.body ]
+
+ %arrayidxA = getelementptr inbounds i32, i32* %a, i64 %ind
+ %loadA = load i32, i32* %arrayidxA, align 4
+
+ %arrayidxB = getelementptr inbounds i32, i32* %b, i64 %ind
+ %loadB = load i32, i32* %arrayidxB, align 4
+ %convergentB = call i32 @llvm.convergent(i32 %loadB)
+
+ %mulC = mul i32 %loadA, %convergentB
+
+ %arrayidxC = getelementptr inbounds i32, i32* %c, i64 %ind
+ store i32 %mulC, i32* %arrayidxC, align 4
+
+ %add = add nuw nsw i64 %ind, 1
+ %exitcond = icmp eq i64 %add, 20
+ br i1 %exitcond, label %for.end, label %for.body
+
+for.end: ; preds = %for.body
+ ret void
+}
+
+declare i32 @llvm.convergent(i32) #1
+
+attributes #0 = { nounwind convergent }
+attributes #1 = { nounwind readnone convergent }
Index: lib/Transforms/Utils/LoopVersioning.cpp
===================================================================
--- lib/Transforms/Utils/LoopVersioning.cpp
+++ lib/Transforms/Utils/LoopVersioning.cpp
@@ -280,8 +280,9 @@
bool Changed = false;
for (Loop *L : Worklist) {
const LoopAccessInfo &LAI = LAA->getInfo(L);
- if (L->isLoopSimplifyForm() && (LAI.getNumRuntimePointerChecks() ||
- !LAI.getPSE().getUnionPredicate().isAlwaysTrue())) {
+ if (L->isLoopSimplifyForm() && !LAI.hasConvergentOp() &&
+ (LAI.getNumRuntimePointerChecks() ||
+ !LAI.getPSE().getUnionPredicate().isAlwaysTrue())) {
LoopVersioning LVer(LAI, L, LI, DT, SE);
LVer.versionLoop();
LVer.annotateLoopWithNoAlias();
-------------- next part --------------
A non-text attachment was scrubbed...
Name: D63016.203587.patch
Type: text/x-patch
Size: 2244 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20190607/adb62dd0/attachment.bin>
More information about the llvm-commits
mailing list