[llvm] Ensure `collectTransitivePredecessors` returns Pred only from the Loop. (PR #113831)
Manish Kausik H via llvm-commits
llvm-commits at lists.llvm.org
Tue Oct 29 09:30:03 PDT 2024
https://github.com/Nirhar updated https://github.com/llvm/llvm-project/pull/113831
>From 3ab467b300185debd54bf6ed2e8da4875d233a8d Mon Sep 17 00:00:00 2001
From: Manish Kausik H <hmanishkausik at gmail.com>
Date: Mon, 28 Oct 2024 00:24:55 +0530
Subject: [PATCH] Ensure `collectTransitivePredecessors` returns Pred only from
the Loop.
It's possible that we encounter Irreducible control flow, due to
which, we may find that a few predecessors of BB are not a part of the
CurLoop. Currently we crash in the function for such cases. This patch
ensures that we only return Predecessors that are a part of CurLoop and
gracefully ignore other Predecessors.
For example, consider Irreducible IR of this form:
```
define i64 @baz() {
bb:
br label %bb1
bb1: ; preds = %bb3, %bb
br label %bb3
bb2: ; No predecessors!
br label %bb3
bb3: ; preds = %bb2, %bb1
%load = load ptr addrspace(1), ptr addrspace(1) null, align 8
br label %bb1
}
```
This crashes when `collectTransitivePredecessors` is called on the
`%bb1<Header>, %bb3<latch>` loop, because the loop body has a predecessor
`%bb2` which is not a part of the loop.
---
llvm/lib/Analysis/MustExecute.cpp | 7 ++++-
.../Analysis/MustExecute/irreducible-cfg.ll | 29 +++++++++++++++++++
2 files changed, 35 insertions(+), 1 deletion(-)
create mode 100644 llvm/test/Analysis/MustExecute/irreducible-cfg.ll
diff --git a/llvm/lib/Analysis/MustExecute.cpp b/llvm/lib/Analysis/MustExecute.cpp
index caed62679a683c..8e233a5c8d7a54 100644
--- a/llvm/lib/Analysis/MustExecute.cpp
+++ b/llvm/lib/Analysis/MustExecute.cpp
@@ -162,6 +162,9 @@ static bool CanProveNotTakenFirstIteration(const BasicBlock *ExitBlock,
/// Collect all blocks from \p CurLoop which lie on all possible paths from
/// the header of \p CurLoop (inclusive) to BB (exclusive) into the set
/// \p Predecessors. If \p BB is the header, \p Predecessors will be empty.
+/// Note: It's possible that we encounter Irreducible control flow, due to
+/// which, we may find that a few predecessors of \p BB are not a part of the
+/// \p CurLoop. We only return Predecessors that are a part of \p CurLoop.
static void collectTransitivePredecessors(
const Loop *CurLoop, const BasicBlock *BB,
SmallPtrSetImpl<const BasicBlock *> &Predecessors) {
@@ -171,6 +174,8 @@ static void collectTransitivePredecessors(
return;
SmallVector<const BasicBlock *, 4> WorkList;
for (const auto *Pred : predecessors(BB)) {
+ if (!CurLoop->contains(Pred))
+ continue;
Predecessors.insert(Pred);
WorkList.push_back(Pred);
}
@@ -187,7 +192,7 @@ static void collectTransitivePredecessors(
// We can ignore backedge of all loops containing BB to get a sligtly more
// optimistic result.
for (const auto *PredPred : predecessors(Pred))
- if (Predecessors.insert(PredPred).second)
+ if (CurLoop->contains(PredPred) && Predecessors.insert(PredPred).second)
WorkList.push_back(PredPred);
}
}
diff --git a/llvm/test/Analysis/MustExecute/irreducible-cfg.ll b/llvm/test/Analysis/MustExecute/irreducible-cfg.ll
new file mode 100644
index 00000000000000..a452761ab33566
--- /dev/null
+++ b/llvm/test/Analysis/MustExecute/irreducible-cfg.ll
@@ -0,0 +1,29 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -disable-output -passes=print-mustexecute %s 2>&1 | FileCheck %s
+
+; The loop body has two predecessors, %header and %side-entry. This leads to irreducible-cfg
+define i64 @baz() {
+; CHECK-LABEL: define i64 @baz() {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: br label %[[HEADER:.*]]
+; CHECK: [[HEADER]]:
+; CHECK-NEXT: br label %[[BODY:.*]] ; (mustexec in: header)
+; CHECK: [[SIDE_ENTRY:.*:]]
+; CHECK-NEXT: br label %[[BODY]]
+; CHECK: [[BODY]]:
+; CHECK-NEXT: [[LOAD:%.*]] = load ptr addrspace(1), ptr addrspace(1) null, align 8 ; (mustexec in: header)
+; CHECK-NEXT: br label %[[HEADER]] ; (mustexec in: header)
+;
+entry:
+ br label %header
+
+header:
+ br label %body
+
+side-entry:
+ br label %body
+
+body:
+ %load = load ptr addrspace(1), ptr addrspace(1) null, align 8
+ br label %header
+}
More information about the llvm-commits
mailing list