[ctx_prof] Fix ProfileAnnotator::allTakenPathsExit
#109183
Merged
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
This stack of pull requests is managed by Graphite. Learn more about stacking. |
2 similar comments
This stack of pull requests is managed by Graphite. Learn more about stacking. |
This stack of pull requests is managed by Graphite. Learn more about stacking. |
|
@llvm/pr-subscribers-llvm-transforms @llvm/pr-subscribers-llvm-analysis Author: Mircea Trofin (mtrofin) ChangesAdded tests to the validator and fixed issues stemming from the previous skipping over BBs with single successors - which is incorrect. That would be now picked by added tests where the assertions are expected to be triggered. Full diff: https://github.com/llvm/llvm-project/pull/109183.diff 2 Files Affected:
diff --git a/llvm/lib/Transforms/Instrumentation/PGOCtxProfFlattening.cpp b/llvm/lib/Transforms/Instrumentation/PGOCtxProfFlattening.cpp
index e76689e2f5f0a5..91f950e2ba4c3e 100644
--- a/llvm/lib/Transforms/Instrumentation/PGOCtxProfFlattening.cpp
+++ b/llvm/lib/Transforms/Instrumentation/PGOCtxProfFlattening.cpp
@@ -233,28 +233,37 @@ class ProfileAnnotator final {
std::deque<const BasicBlock *> Worklist;
DenseSet<const BasicBlock *> Visited;
Worklist.push_back(&F.getEntryBlock());
- Visited.insert(&F.getEntryBlock());
+ bool HitExit = false;
while (!Worklist.empty()) {
const auto *BB = Worklist.front();
Worklist.pop_front();
- if (succ_size(BB) <= 1)
+ if (!Visited.insert(BB).second)
continue;
+ if (succ_size(BB) == 0) {
+ if (isa<UnreachableInst>(BB->getTerminator()))
+ return false;
+ HitExit = true;
+ continue;
+ }
+ if (succ_size(BB) == 1) {
+ llvm::append_range(Worklist, successors(BB));
+ continue;
+ }
const auto &BBInfo = getBBInfo(*BB);
- bool Inserted = false;
+ bool HasAWayOut = false;
for (auto I = 0U; I < BB->getTerminator()->getNumSuccessors(); ++I) {
const auto *Succ = BB->getTerminator()->getSuccessor(I);
if (!shouldExcludeEdge(*BB, *Succ)) {
- if (BBInfo.getEdgeCount(I) > 0)
- if (Visited.insert(Succ).second) {
- Worklist.push_back(Succ);
- Inserted = true;
- }
+ if (BBInfo.getEdgeCount(I) > 0) {
+ HasAWayOut = true;
+ Worklist.push_back(Succ);
+ }
}
}
- if (!Inserted)
+ if (!HasAWayOut)
return false;
}
- return true;
+ return HitExit;
}
public:
diff --git a/llvm/test/Analysis/CtxProfAnalysis/flatten-check-path.ll b/llvm/test/Analysis/CtxProfAnalysis/flatten-check-path.ll
new file mode 100644
index 00000000000000..dbe14c3522a8a9
--- /dev/null
+++ b/llvm/test/Analysis/CtxProfAnalysis/flatten-check-path.ll
@@ -0,0 +1,85 @@
+; REQUIRES: asserts
+; Check that the profile annotator works: we hit an exit and non-zero paths to
+; already visited blocks are not disconsidered.
+;
+; RUN: split-file %s %t
+; RUN: llvm-ctxprof-util fromJSON --input=%t/profile_ok.json --output=%t/profile_ok.ctxprofdata
+; RUN: llvm-ctxprof-util fromJSON --input=%t/profile_pump.json --output=%t/profile_pump.ctxprofdata
+; RUN: llvm-ctxprof-util fromJSON --input=%t/profile_unreachable.json --output=%t/profile_unreachable.ctxprofdata
+;
+; RUN: opt -passes=ctx-prof-flatten %t/example_ok.ll -use-ctx-profile=%t/profile_ok.ctxprofdata -S -o - | FileCheck %s
+; RUN: not --crash opt -passes=ctx-prof-flatten %t/message_pump.ll -use-ctx-profile=%t/profile_pump.ctxprofdata -S 2>&1 | FileCheck %s --check-prefix=ASSERTION
+; RUN: not --crash opt -passes=ctx-prof-flatten %t/unreachable.ll -use-ctx-profile=%t/profile_unreachable.ctxprofdata -S 2>&1 | FileCheck %s --check-prefix=ASSERTION
+
+; CHECK: br i1 %x, label %b1, label %exit, !prof ![[PROF1:[0-9]+]]
+; CHECK: br i1 %y, label %blk, label %exit, !prof ![[PROF2:[0-9]+]]
+; CHECK: ![[PROF1]] = !{!"branch_weights", i32 1, i32 1}
+; CHECK: ![[PROF2]] = !{!"branch_weights", i32 0, i32 1}
+; ASSERTION: Assertion `allTakenPathsExit()
+
+; b1->exit is the only way out from b1, but the exit block would have been
+; already visited from blk. That should not result in an assertion, though.
+;--- example_ok.ll
+define void @foo(i32 %t) !guid !0 {
+entry:
+ call void @llvm.instrprof.increment(ptr @foo, i64 42, i32 42, i32 0)
+ br label %blk
+blk:
+ call void @llvm.instrprof.increment(ptr @foo, i64 42, i32 42, i32 1)
+ %x = icmp eq i32 %t, 0
+ br i1 %x, label %b1, label %exit
+b1:
+ call void @llvm.instrprof.increment(ptr @foo, i64 42, i32 42, i32 2)
+ %y = icmp eq i32 %t, 0
+ br i1 %y, label %blk, label %exit
+exit:
+ call void @llvm.instrprof.increment(ptr @foo, i64 42, i32 42, i32 3)
+ ret void
+}
+!0 = !{i64 1234}
+
+;--- profile_ok.json
+[{"Guid":1234, "Counters":[2, 2, 1, 2]}]
+
+;--- message_pump.ll
+; This is a message pump: the loop never exits. This should result in an
+; assertion because we can't reach an exit BB
+
+define void @foo(i32 %t) !guid !0 {
+entry:
+ call void @llvm.instrprof.increment(ptr @foo, i64 42, i32 42, i32 0)
+ br label %blk
+blk:
+ call void @llvm.instrprof.increment(ptr @foo, i64 42, i32 42, i32 1)
+ %x = icmp eq i32 %t, 0
+ br i1 %x, label %blk, label %exit
+exit:
+ call void @llvm.instrprof.increment(ptr @foo, i64 42, i32 42, i32 2)
+ ret void
+}
+!0 = !{i64 1234}
+
+;--- profile_pump.json
+[{"Guid":1234, "Counters":[2, 10, 0]}]
+
+;--- unreachable.ll
+; An unreachable block is reached, that's an error
+define void @foo(i32 %t) !guid !0 {
+entry:
+ call void @llvm.instrprof.increment(ptr @foo, i64 42, i32 42, i32 0)
+ br label %blk
+blk:
+ call void @llvm.instrprof.increment(ptr @foo, i64 42, i32 42, i32 1)
+ %x = icmp eq i32 %t, 0
+ br i1 %x, label %b1, label %exit
+b1:
+ call void @llvm.instrprof.increment(ptr @foo, i64 42, i32 42, i32 2)
+ unreachable
+exit:
+ call void @llvm.instrprof.increment(ptr @foo, i64 42, i32 42, i32 3)
+ ret void
+}
+!0 = !{i64 1234}
+
+;--- profile_unreachable.json
+[{"Guid":1234, "Counters":[2, 1, 1, 2]}]
\ No newline at end of file
|
|
@llvm/pr-subscribers-pgo Author: Mircea Trofin (mtrofin) ChangesAdded tests to the validator and fixed issues stemming from the previous skipping over BBs with single successors - which is incorrect. That would be now picked by added tests where the assertions are expected to be triggered. Full diff: https://github.com/llvm/llvm-project/pull/109183.diff 2 Files Affected:
diff --git a/llvm/lib/Transforms/Instrumentation/PGOCtxProfFlattening.cpp b/llvm/lib/Transforms/Instrumentation/PGOCtxProfFlattening.cpp
index e76689e2f5f0a5..91f950e2ba4c3e 100644
--- a/llvm/lib/Transforms/Instrumentation/PGOCtxProfFlattening.cpp
+++ b/llvm/lib/Transforms/Instrumentation/PGOCtxProfFlattening.cpp
@@ -233,28 +233,37 @@ class ProfileAnnotator final {
std::deque<const BasicBlock *> Worklist;
DenseSet<const BasicBlock *> Visited;
Worklist.push_back(&F.getEntryBlock());
- Visited.insert(&F.getEntryBlock());
+ bool HitExit = false;
while (!Worklist.empty()) {
const auto *BB = Worklist.front();
Worklist.pop_front();
- if (succ_size(BB) <= 1)
+ if (!Visited.insert(BB).second)
continue;
+ if (succ_size(BB) == 0) {
+ if (isa<UnreachableInst>(BB->getTerminator()))
+ return false;
+ HitExit = true;
+ continue;
+ }
+ if (succ_size(BB) == 1) {
+ llvm::append_range(Worklist, successors(BB));
+ continue;
+ }
const auto &BBInfo = getBBInfo(*BB);
- bool Inserted = false;
+ bool HasAWayOut = false;
for (auto I = 0U; I < BB->getTerminator()->getNumSuccessors(); ++I) {
const auto *Succ = BB->getTerminator()->getSuccessor(I);
if (!shouldExcludeEdge(*BB, *Succ)) {
- if (BBInfo.getEdgeCount(I) > 0)
- if (Visited.insert(Succ).second) {
- Worklist.push_back(Succ);
- Inserted = true;
- }
+ if (BBInfo.getEdgeCount(I) > 0) {
+ HasAWayOut = true;
+ Worklist.push_back(Succ);
+ }
}
}
- if (!Inserted)
+ if (!HasAWayOut)
return false;
}
- return true;
+ return HitExit;
}
public:
diff --git a/llvm/test/Analysis/CtxProfAnalysis/flatten-check-path.ll b/llvm/test/Analysis/CtxProfAnalysis/flatten-check-path.ll
new file mode 100644
index 00000000000000..dbe14c3522a8a9
--- /dev/null
+++ b/llvm/test/Analysis/CtxProfAnalysis/flatten-check-path.ll
@@ -0,0 +1,85 @@
+; REQUIRES: asserts
+; Check that the profile annotator works: we hit an exit and non-zero paths to
+; already visited blocks are not disconsidered.
+;
+; RUN: split-file %s %t
+; RUN: llvm-ctxprof-util fromJSON --input=%t/profile_ok.json --output=%t/profile_ok.ctxprofdata
+; RUN: llvm-ctxprof-util fromJSON --input=%t/profile_pump.json --output=%t/profile_pump.ctxprofdata
+; RUN: llvm-ctxprof-util fromJSON --input=%t/profile_unreachable.json --output=%t/profile_unreachable.ctxprofdata
+;
+; RUN: opt -passes=ctx-prof-flatten %t/example_ok.ll -use-ctx-profile=%t/profile_ok.ctxprofdata -S -o - | FileCheck %s
+; RUN: not --crash opt -passes=ctx-prof-flatten %t/message_pump.ll -use-ctx-profile=%t/profile_pump.ctxprofdata -S 2>&1 | FileCheck %s --check-prefix=ASSERTION
+; RUN: not --crash opt -passes=ctx-prof-flatten %t/unreachable.ll -use-ctx-profile=%t/profile_unreachable.ctxprofdata -S 2>&1 | FileCheck %s --check-prefix=ASSERTION
+
+; CHECK: br i1 %x, label %b1, label %exit, !prof ![[PROF1:[0-9]+]]
+; CHECK: br i1 %y, label %blk, label %exit, !prof ![[PROF2:[0-9]+]]
+; CHECK: ![[PROF1]] = !{!"branch_weights", i32 1, i32 1}
+; CHECK: ![[PROF2]] = !{!"branch_weights", i32 0, i32 1}
+; ASSERTION: Assertion `allTakenPathsExit()
+
+; b1->exit is the only way out from b1, but the exit block would have been
+; already visited from blk. That should not result in an assertion, though.
+;--- example_ok.ll
+define void @foo(i32 %t) !guid !0 {
+entry:
+ call void @llvm.instrprof.increment(ptr @foo, i64 42, i32 42, i32 0)
+ br label %blk
+blk:
+ call void @llvm.instrprof.increment(ptr @foo, i64 42, i32 42, i32 1)
+ %x = icmp eq i32 %t, 0
+ br i1 %x, label %b1, label %exit
+b1:
+ call void @llvm.instrprof.increment(ptr @foo, i64 42, i32 42, i32 2)
+ %y = icmp eq i32 %t, 0
+ br i1 %y, label %blk, label %exit
+exit:
+ call void @llvm.instrprof.increment(ptr @foo, i64 42, i32 42, i32 3)
+ ret void
+}
+!0 = !{i64 1234}
+
+;--- profile_ok.json
+[{"Guid":1234, "Counters":[2, 2, 1, 2]}]
+
+;--- message_pump.ll
+; This is a message pump: the loop never exits. This should result in an
+; assertion because we can't reach an exit BB
+
+define void @foo(i32 %t) !guid !0 {
+entry:
+ call void @llvm.instrprof.increment(ptr @foo, i64 42, i32 42, i32 0)
+ br label %blk
+blk:
+ call void @llvm.instrprof.increment(ptr @foo, i64 42, i32 42, i32 1)
+ %x = icmp eq i32 %t, 0
+ br i1 %x, label %blk, label %exit
+exit:
+ call void @llvm.instrprof.increment(ptr @foo, i64 42, i32 42, i32 2)
+ ret void
+}
+!0 = !{i64 1234}
+
+;--- profile_pump.json
+[{"Guid":1234, "Counters":[2, 10, 0]}]
+
+;--- unreachable.ll
+; An unreachable block is reached, that's an error
+define void @foo(i32 %t) !guid !0 {
+entry:
+ call void @llvm.instrprof.increment(ptr @foo, i64 42, i32 42, i32 0)
+ br label %blk
+blk:
+ call void @llvm.instrprof.increment(ptr @foo, i64 42, i32 42, i32 1)
+ %x = icmp eq i32 %t, 0
+ br i1 %x, label %b1, label %exit
+b1:
+ call void @llvm.instrprof.increment(ptr @foo, i64 42, i32 42, i32 2)
+ unreachable
+exit:
+ call void @llvm.instrprof.increment(ptr @foo, i64 42, i32 42, i32 3)
+ ret void
+}
+!0 = !{i64 1234}
+
+;--- profile_unreachable.json
+[{"Guid":1234, "Counters":[2, 1, 1, 2]}]
\ No newline at end of file
|
teresajohnson
approved these changes
Sep 18, 2024
| const auto &BBInfo = getBBInfo(*BB); | ||
| bool Inserted = false; | ||
| bool HasAWayOut = false; | ||
| for (auto I = 0U; I < BB->getTerminator()->getNumSuccessors(); ++I) { |
There was a problem hiding this comment.
can this use succ_size as used above, for consistency? And maybe call that once and save the result.
There was a problem hiding this comment.
I'd prefer using this current API (getNumSuccessors) here, they should be the same, but this is "closer" to the API used below (BB->getTerminator()->getSuccessor).
| @@ -0,0 +1,85 @@ | |||
| ; REQUIRES: asserts | |||
| ; Check that the profile annotator works: we hit an exit and non-zero paths to | |||
| ; already visited blocks are not disconsidered. | |||
There was a problem hiding this comment.
Not sure what you are trying to say by "are not disconsidered" here, especially with the double-negative.
mtrofin
force-pushed
the
users/mtrofin/09-17-_ctx_prof_fix_profileannotator_alltakenpathsexit_
branch
from
6d23eef to
1c6e49a
Compare
3 similar comments
mtrofin
force-pushed
the
users/mtrofin/09-17-_ctx_prof_fix_profileannotator_alltakenpathsexit_
branch
from
1c6e49a to
bbb3ba0
Compare
This was referenced Sep 19, 2024
Merged
mtrofin
deleted the
users/mtrofin/09-17-_ctx_prof_fix_profileannotator_alltakenpathsexit_
branch
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/190/builds/6033 Here is the relevant piece of the build log for the reference |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/89/builds/6666 Here is the relevant piece of the build log for the reference |
tmsri
pushed a commit
to tmsri/llvm-project
that referenced
this pull request
Sep 19, 2024
Added tests to the validator and fixed issues stemming from the previous skipping over BBs with single successors - which is incorrect. That would be now picked by added tests where the assertions are expected to be triggered.
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.
Added tests to the validator and fixed issues stemming from the previous skipping over BBs with single successors - which is incorrect. That would be now picked by added tests where the assertions are expected to be triggered.