[LoopUnroll] Do not copy !llvm.loop from latch to non-latch #165635
+115
−23
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
When LoopUnroll copies the original loop's latch to the corresponding non-latch branch in an unrolled iteration, any `!llvm.loop` is copied along with it, but `!llvm.loop` is useless and misleading there. This patch discards it. e06831a did the same for LoopPeel.
jdenny-ornl
requested review from
MatzeB,
arsenm,
fhahn,
jdoerfert,
mtrofin,
nikic and
vzakhari
Member
|
@llvm/pr-subscribers-llvm-transforms Author: Joel E. Denny (jdenny-ornl) ChangesWhen LoopUnroll copies the original loop's latch to the corresponding non-latch branch in an unrolled iteration, any e06831a did the same for LoopPeel. Full diff: https://github.com/llvm/llvm-project/pull/165635.diff 6 Files Affected:
diff --git a/llvm/lib/Transforms/Utils/LoopUnroll.cpp b/llvm/lib/Transforms/Utils/LoopUnroll.cpp
index 4fe736ac29b0a..8bbc857715d27 100644
--- a/llvm/lib/Transforms/Utils/LoopUnroll.cpp
+++ b/llvm/lib/Transforms/Utils/LoopUnroll.cpp
@@ -919,6 +919,12 @@ llvm::UnrollLoop(Loop *L, UnrollLoopOptions ULO, LoopInfo *LI,
Latches[i]->getTerminator()->replaceSuccessorWith(Headers[i], Headers[j]);
}
+ // Remove loop metadata copied from the original loop latch to branches that
+ // are no longer latches.
+ for (unsigned I = 0, E = Latches.size() - (CompletelyUnroll ? 0 : 1); I < E;
+ ++I)
+ Latches[I]->getTerminator()->setMetadata(LLVMContext::MD_loop, nullptr);
+
// Update dominators of blocks we might reach through exits.
// Immediate dominator of such block might change, because we add more
// routes which can lead to the exit: we can now reach it from the copied
diff --git a/llvm/test/Transforms/LoopUnroll/convergent.controlled.ll b/llvm/test/Transforms/LoopUnroll/convergent.controlled.ll
index 6e600d2a659da..5dc613e733f00 100644
--- a/llvm/test/Transforms/LoopUnroll/convergent.controlled.ll
+++ b/llvm/test/Transforms/LoopUnroll/convergent.controlled.ll
@@ -491,41 +491,41 @@ define i32 @unroll_nest(i32 %n, i1 %cond) {
; CHECK-NEXT: entry:
; CHECK-NEXT: br label [[L3:%.*]]
; CHECK: l3:
-; CHECK-NEXT: br label [[L2:%.*]], !llvm.loop [[LOOP4]]
+; CHECK-NEXT: br label [[L2:%.*]]
; CHECK: l2:
; CHECK-NEXT: [[TOK_L2:%.*]] = call token @llvm.experimental.convergence.anchor()
; CHECK-NEXT: call void @f() [ "convergencectrl"(token [[TOK_L2]]) ]
-; CHECK-NEXT: br i1 [[COND:%.*]], label [[L2_1:%.*]], label [[LATCH:%.*]], !llvm.loop [[LOOP4]]
+; CHECK-NEXT: br i1 [[COND:%.*]], label [[L2_1:%.*]], label [[LATCH:%.*]]
; CHECK: l2.1:
; CHECK-NEXT: [[TOK_L2_1:%.*]] = call token @llvm.experimental.convergence.anchor()
; CHECK-NEXT: call void @f() [ "convergencectrl"(token [[TOK_L2_1]]) ]
; CHECK-NEXT: br i1 [[COND]], label [[L2]], label [[LATCH]], !llvm.loop [[LOOP9:![0-9]+]]
; CHECK: latch:
-; CHECK-NEXT: br label [[L2_12:%.*]], !llvm.loop [[LOOP4]]
+; CHECK-NEXT: br label [[L2_12:%.*]]
; CHECK: l2.12:
; CHECK-NEXT: [[TOK_L2_11:%.*]] = call token @llvm.experimental.convergence.anchor()
; CHECK-NEXT: call void @f() [ "convergencectrl"(token [[TOK_L2_11]]) ]
-; CHECK-NEXT: br i1 [[COND]], label [[L2_1_1:%.*]], label [[LATCH_1:%.*]], !llvm.loop [[LOOP4]]
+; CHECK-NEXT: br i1 [[COND]], label [[L2_1_1:%.*]], label [[LATCH_1:%.*]]
; CHECK: l2.1.1:
; CHECK-NEXT: [[TOK_L2_1_1:%.*]] = call token @llvm.experimental.convergence.anchor()
; CHECK-NEXT: call void @f() [ "convergencectrl"(token [[TOK_L2_1_1]]) ]
; CHECK-NEXT: br i1 [[COND]], label [[L2_12]], label [[LATCH_1]], !llvm.loop [[LOOP9]]
; CHECK: latch.1:
-; CHECK-NEXT: br label [[L2_2:%.*]], !llvm.loop [[LOOP4]]
+; CHECK-NEXT: br label [[L2_2:%.*]]
; CHECK: l2.2:
; CHECK-NEXT: [[TOK_L2_2:%.*]] = call token @llvm.experimental.convergence.anchor()
; CHECK-NEXT: call void @f() [ "convergencectrl"(token [[TOK_L2_2]]) ]
-; CHECK-NEXT: br i1 [[COND]], label [[L2_1_2:%.*]], label [[LATCH_2:%.*]], !llvm.loop [[LOOP4]]
+; CHECK-NEXT: br i1 [[COND]], label [[L2_1_2:%.*]], label [[LATCH_2:%.*]]
; CHECK: l2.1.2:
; CHECK-NEXT: [[TOK_L2_1_2:%.*]] = call token @llvm.experimental.convergence.anchor()
; CHECK-NEXT: call void @f() [ "convergencectrl"(token [[TOK_L2_1_2]]) ]
; CHECK-NEXT: br i1 [[COND]], label [[L2_2]], label [[LATCH_2]], !llvm.loop [[LOOP9]]
; CHECK: latch.2:
-; CHECK-NEXT: br label [[L2_3:%.*]], !llvm.loop [[LOOP4]]
+; CHECK-NEXT: br label [[L2_3:%.*]]
; CHECK: l2.3:
; CHECK-NEXT: [[TOK_L2_3:%.*]] = call token @llvm.experimental.convergence.anchor()
; CHECK-NEXT: call void @f() [ "convergencectrl"(token [[TOK_L2_3]]) ]
-; CHECK-NEXT: br i1 [[COND]], label [[L2_1_3:%.*]], label [[LATCH_3:%.*]], !llvm.loop [[LOOP4]]
+; CHECK-NEXT: br i1 [[COND]], label [[L2_1_3:%.*]], label [[LATCH_3:%.*]]
; CHECK: l2.1.3:
; CHECK-NEXT: [[TOK_L2_1_3:%.*]] = call token @llvm.experimental.convergence.anchor()
; CHECK-NEXT: call void @f() [ "convergencectrl"(token [[TOK_L2_1_3]]) ]
@@ -541,7 +541,7 @@ l3:
%tok.loop = call token @llvm.experimental.convergence.anchor()
%inc = add nsw i32 %x.0, 1
%exitcond = icmp eq i32 %inc, 4
- br label %l2, !llvm.loop !1
+ br label %l2
l2:
%tok.l2 = call token @llvm.experimental.convergence.anchor()
diff --git a/llvm/test/Transforms/LoopUnroll/followup.ll b/llvm/test/Transforms/LoopUnroll/followup.ll
index 051e43d52b3be..b6561781d29ca 100644
--- a/llvm/test/Transforms/LoopUnroll/followup.ll
+++ b/llvm/test/Transforms/LoopUnroll/followup.ll
@@ -36,11 +36,11 @@ for.end: ; preds = %for.body, %entry
; COMMON-LABEL: @test(
-; COUNT: br i1 %exitcond.1, label %for.end.loopexit, label %for.body, !llvm.loop ![[LOOP:[0-9]+]]
+; COUNT: br i1 %exitcond.1, label %for.end.loopexit, label %for.body, !llvm.loop ![[#LOOP:]]
-; COUNT: ![[FOLLOWUP_ALL:[0-9]+]] = !{!"FollowupAll"}
-; COUNT: ![[FOLLOWUP_UNROLLED:[0-9]+]] = !{!"FollowupUnrolled"}
-; COUNT: ![[LOOP]] = distinct !{![[LOOP]], ![[FOLLOWUP_ALL]], ![[FOLLOWUP_UNROLLED]]}
+; COUNT: ![[#LOOP]] = distinct !{![[#LOOP]], ![[#FOLLOWUP_ALL:]], ![[#FOLLOWUP_UNROLLED:]]}
+; COUNT: ![[#FOLLOWUP_ALL]] = !{!"FollowupAll"}
+; COUNT: ![[#FOLLOWUP_UNROLLED]] = !{!"FollowupUnrolled"}
; EPILOG: br i1 %niter.ncmp.7, label %for.end.loopexit.unr-lcssa, label %for.body, !llvm.loop ![[LOOP_0:[0-9]+]]
diff --git a/llvm/test/Transforms/LoopUnroll/pr131465.ll b/llvm/test/Transforms/LoopUnroll/pr131465.ll
index 643b020c6c110..60bea6a4d9043 100644
--- a/llvm/test/Transforms/LoopUnroll/pr131465.ll
+++ b/llvm/test/Transforms/LoopUnroll/pr131465.ll
@@ -11,11 +11,11 @@ define i32 @pr131465(i1 %x) mustprogress {
; CHECK-NEXT: [[INDVAR:%.*]] = phi i32 [ 2, %[[ENTRY]] ], [ [[NEXT_1:%.*]], %[[FOR_BODY_1:.*]] ]
; CHECK-NEXT: [[NEXT:%.*]] = add nsw i32 [[INDVAR]], [[INC]]
; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[NEXT]], 2
-; CHECK-NEXT: br i1 [[EXITCOND]], label %[[FOR_END:.*]], label %[[FOR_BODY_1]], !llvm.loop [[LOOP0:![0-9]+]]
+; CHECK-NEXT: br i1 [[EXITCOND]], label %[[FOR_END:.*]], label %[[FOR_BODY_1]]
; CHECK: [[FOR_BODY_1]]:
; CHECK-NEXT: [[NEXT_1]] = add nsw i32 [[NEXT]], [[INC]]
; CHECK-NEXT: [[EXITCOND_1:%.*]] = icmp eq i32 [[NEXT_1]], 2
-; CHECK-NEXT: br i1 [[EXITCOND_1]], label %[[FOR_END]], label %[[FOR_BODY]], !llvm.loop [[LOOP2:![0-9]+]]
+; CHECK-NEXT: br i1 [[EXITCOND_1]], label %[[FOR_END]], label %[[FOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
; CHECK: [[FOR_END]]:
; CHECK-NEXT: ret i32 0
;
@@ -37,7 +37,5 @@ for.end:
!0 = !{!0, !{!"llvm.loop.unroll.count", i32 2}}
;.
; CHECK: [[LOOP0]] = distinct !{[[LOOP0]], [[META1:![0-9]+]]}
-; CHECK: [[META1]] = !{!"llvm.loop.unroll.count", i32 2}
-; CHECK: [[LOOP2]] = distinct !{[[LOOP2]], [[META3:![0-9]+]]}
-; CHECK: [[META3]] = !{!"llvm.loop.unroll.disable"}
+; CHECK: [[META1]] = !{!"llvm.loop.unroll.disable"}
;.
diff --git a/llvm/test/Transforms/LoopUnroll/unroll-dont-copy-latch-loop-id.ll b/llvm/test/Transforms/LoopUnroll/unroll-dont-copy-latch-loop-id.ll
new file mode 100644
index 0000000000000..191ebb023b7c1
--- /dev/null
+++ b/llvm/test/Transforms/LoopUnroll/unroll-dont-copy-latch-loop-id.ll
@@ -0,0 +1,90 @@
+; Check that !llvm.loop is not copied from the original loop's latch to the
+; corresponding non-latch branch in any unrolled iteration.
+
+; The -implicit-check-not options make sure that no additional label or
+; !llvm.loop shows up.
+; DEFINE: %{unroll} = opt < %s -passes=loop-unroll -S
+; DEFINE: %{fc} = FileCheck %s \
+; DEFINE: -implicit-check-not='{{^[^ ;]*:}}' \
+; DEFINE: -implicit-check-not='!llvm.loop' \
+; DEFINE: -check-prefixes
+
+; Check partial unroll: only the unrolled loop's latch has !llvm.loop.
+; RUN: %{unroll} -unroll-count=3 | %{fc} ALL,UR,UR3
+
+; Check complete unroll: no !llvm.loop remains because no loop remains.
+; RUN: %{unroll} -unroll-count=4 | %{fc} ALL,UR,UR4
+
+; Check remainder: both loops have a !llvm.loop.
+; DEFINE: %{rt} = %{unroll} -unroll-count=3 -unroll-runtime
+; RUN: %{rt} -unroll-runtime-epilog=true | %{fc} ALL,RT,EPILOG
+; RUN: %{rt} -unroll-runtime-epilog=false | %{fc} ALL,RT,PROLOG
+
+; ALL: define void @test(i32 %n) {
+; ALL: entry:
+; UR: br label %body
+; EPILOG: br i1 %{{.*}}, label %body.epil.preheader, label %entry.new
+; PROLOG: br i1 %{{.*}}, label %body.prol.preheader, label %body.prol.loopexit
+; PROLOG: body.prol.preheader:
+; PROLOG: br label %body.prol
+; PROLOG: body.prol:
+; PROLOG: br i1 %{{.*}}, label %body.prol, label %body.prol.loopexit.unr-lcssa, !llvm.loop !0
+; PROLOG: body.prol.loopexit.unr-lcssa:
+; PROLOG: br label %body.prol.loopexit
+; PROLOG: body.prol.loopexit:
+; PROLOG: br i1 %{{.*}}, label %exit, label %entry.new
+; RT: entry.new:
+; RT: br label %body
+; ALL: body:
+; UR: br i1 %c, label %body.1, label %exit
+; EPILOG: br i1 %{{.*}}, label %body, label %exit.unr-lcssa, !llvm.loop !0
+; PROLOG: br i1 %{{.*}}, label %body, label %exit.unr-lcssa, !llvm.loop !2
+; UR: body.1:
+; UR: br i1 %c.1, label %body.2, label %exit
+; UR: body.2:
+; UR3: br i1 %c.2, label %body, label %exit, !llvm.loop !0
+; UR4: br i1 %c.2, label %body.3, label %exit
+; UR4: body.3:
+; UR4: br label %exit
+; RT: exit.unr-lcssa:
+; EPILOG: br i1 {{.*}}, label %body.epil.preheader, label %exit
+; PROLOG: br label %exit
+; EPILOG: body.epil.preheader:
+; EPILOG: br label %body.epil
+; EPILOG: body.epil:
+; EPILOG: br i1 {{.*}}, label %body.epil, label %exit.epilog-lcssa, !llvm.loop !3
+; EPILOG: exit.epilog-lcssa:
+; EPILOG: br label %exit
+; ALL: exit:
+; ALL: ret void
+; ALL: }
+define void @test(i32 %n) {
+entry:
+ %max = call i32 @llvm.umin.i32(i32 %n, i32 3)
+ br label %body
+
+body:
+ %i = phi i32 [ 0, %entry ], [ %inc, %body ]
+ %inc = add i32 %i, 1
+ %c = icmp ult i32 %i, %max
+ br i1 %c, label %body, label %exit, !llvm.loop !0
+
+exit:
+ ret void
+}
+
+; UR3: !0 = distinct !{!0, !1, !2}
+; UR3: !1 = !{!"copied"}
+; UR3: !2 = !{!"llvm.loop.unroll.disable"}
+;
+; EPILOG: !0 = distinct !{!0, !1, !2}
+; EPILOG: !1 = !{!"copied"}
+; EPILOG: !2 = !{!"llvm.loop.unroll.disable"}
+; EPILOG: !3 = distinct !{!3, !2}
+;
+; PROLOG: !0 = distinct !{!0, !1}
+; PROLOG: !1 = !{!"llvm.loop.unroll.disable"}
+; PROLOG: !2 = distinct !{!2, !3, !1}
+; PROLOG: !3 = !{!"copied"}
+!0 = distinct !{!0, !1}
+!1 = !{!"copied"}
diff --git a/llvm/test/Transforms/LoopUnroll/unroll-loads-cse.ll b/llvm/test/Transforms/LoopUnroll/unroll-loads-cse.ll
index f85aac75539c6..77dd3429b684e 100644
--- a/llvm/test/Transforms/LoopUnroll/unroll-loads-cse.ll
+++ b/llvm/test/Transforms/LoopUnroll/unroll-loads-cse.ll
@@ -419,7 +419,7 @@ define void @loop_body_with_dead_blocks(ptr %src) {
; CHECK-NEXT: call void @foo()
; CHECK-NEXT: [[L_2:%.*]] = load i32, ptr [[SRC]], align 8
; CHECK-NEXT: [[C_2:%.*]] = icmp eq i32 [[L_2]], 1
-; CHECK-NEXT: br i1 [[C_2]], label [[EXIT:%.*]], label [[LOOP_HEADER_1:%.*]], !llvm.loop [[LOOP7:![0-9]+]]
+; CHECK-NEXT: br i1 [[C_2]], label [[EXIT:%.*]], label [[LOOP_HEADER_1:%.*]]
; CHECK: loop.header.1:
; CHECK-NEXT: br label [[LOOP_BB_1:%.*]]
; CHECK: loop.bb.1:
@@ -429,7 +429,7 @@ define void @loop_body_with_dead_blocks(ptr %src) {
; CHECK-NEXT: call void @foo()
; CHECK-NEXT: [[L_2_1:%.*]] = load i32, ptr [[SRC]], align 8
; CHECK-NEXT: [[C_2_1:%.*]] = icmp eq i32 [[L_2_1]], 1
-; CHECK-NEXT: br i1 [[C_2_1]], label [[EXIT]], label [[LOOP_HEADER]], !llvm.loop [[LOOP9:![0-9]+]]
+; CHECK-NEXT: br i1 [[C_2_1]], label [[EXIT]], label [[LOOP_HEADER]], !llvm.loop [[LOOP7:![0-9]+]]
; CHECK: exit:
; CHECK-NEXT: ret void
;
@@ -471,7 +471,5 @@ exit:
; CHECK: [[LOOP4]] = distinct !{[[LOOP4]], [[META1]], [[META2]]}
; CHECK: [[LOOP5]] = distinct !{[[LOOP5]], [[META1]], [[META2]]}
; CHECK: [[LOOP6]] = distinct !{[[LOOP6]], [[META1]], [[META2]]}
-; CHECK: [[LOOP7]] = distinct !{[[LOOP7]], [[META1]], [[META8:![0-9]+]]}
-; CHECK: [[META8]] = !{!"llvm.loop.unroll.count", i32 2}
-; CHECK: [[LOOP9]] = distinct !{[[LOOP9]], [[META1]], [[META2]]}
+; CHECK: [[LOOP7]] = distinct !{[[LOOP7]], [[META1]], [[META2]]}
;.
|
nikic
reviewed
Oct 30, 2025
Contributor
There was a problem hiding this comment.
This looks like the kind of test that is going to really piss me off if I ever have to update it. Either de-magic and reduce the scope of this test, or add a UTC flag that adds {{$}} to the end of lines so absence of metadata can be verified.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
When LoopUnroll copies the original loop's latch to the corresponding non-latch branch in an unrolled iteration, any
!llvm.loopis copied along with it, but!llvm.loopis useless and misleading there. This patch discards it.e06831a did the same for LoopPeel.