[llvm][gvn-sink] Don't try to sink inline asm #138414
Merged
+48
−1
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
ilovepi
added
the
llvm:GVN
Graphite App
|
@llvm/pr-subscribers-llvm-transforms Author: Paul Kirth (ilovepi) ChangesFixes #138345. Before this patch, gvn-sink would try to sink inline llvm-project/llvm/lib/Transforms/Scalar/GVN.cpp Line 2932 in b4fac94 Similarly, gvn-sink should skip these instructions, since they are not safe to move. The test added is reduced from a failure when compiling Fuchsia. There Full diff: https://github.com/llvm/llvm-project/pull/138414.diff 2 Files Affected:
diff --git a/llvm/lib/Transforms/Scalar/GVNSink.cpp b/llvm/lib/Transforms/Scalar/GVNSink.cpp
index 2058df33ea331..40f7a61cb5717 100644
--- a/llvm/lib/Transforms/Scalar/GVNSink.cpp
+++ b/llvm/lib/Transforms/Scalar/GVNSink.cpp
@@ -548,6 +548,9 @@ class GVNSink {
if (isa<PHINode>(I) || I->isEHPad() || isa<AllocaInst>(I) ||
I->getType()->isTokenTy())
return true;
+ // Inline asm can't be sunk either.
+ if (auto *CB = dyn_cast<CallBase>(I); CB->isInlineAsm())
+ return true;
return false;
}
diff --git a/llvm/test/Transforms/GVNSink/pr138345.ll b/llvm/test/Transforms/GVNSink/pr138345.ll
new file mode 100644
index 0000000000000..1c9ff5936625f
--- /dev/null
+++ b/llvm/test/Transforms/GVNSink/pr138345.ll
@@ -0,0 +1,51 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; REQUIRES: x86-registered-target
+; RUN: opt -passes="gvn-sink" -S %s | FileCheck %s
+; RUN: opt -passes="gvn-sink,correlated-propagation" -S %s | FileCheck %s
+
+;; See https://github.com/llvm/llvm-project/issues/138345 for details.
+;; The program below used to crash due to taking the address of the inline asm.
+;; gvn-sink shouldn't do anything in this case, so test that the pass no longer
+;; generates invalid IR and no longer crashes.
+
+target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+define void @c(i64 %num, ptr %ptr) {
+; CHECK-LABEL: define void @c(
+; CHECK-SAME: i64 [[NUM:%.*]], ptr [[PTR:%.*]]) {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: switch i64 [[NUM]], label %[[SW_EPILOG:.*]] [
+; CHECK-NEXT: i64 1, label %[[SW_BB:.*]]
+; CHECK-NEXT: i64 0, label %[[SW_BB1:.*]]
+; CHECK-NEXT: ]
+; CHECK: [[SW_BB]]:
+; CHECK-NEXT: [[TMP1:%.*]] = load i8, ptr [[PTR]], align 1
+; CHECK-NEXT: call void asm sideeffect "", "r,r,~{dirflag},~{fpsr},~{flags}"(i8 [[TMP1]], ptr @c)
+; CHECK-NEXT: br label %[[SW_EPILOG]]
+; CHECK: [[SW_BB1]]:
+; CHECK-NEXT: [[TMP2:%.*]] = load i8, ptr [[PTR]], align 1
+; CHECK-NEXT: call void asm sideeffect "movdqu 0 [[XMM0:%.*]] \0A\09", "r,r,~{dirflag},~{fpsr},~{flags}"(i8 [[TMP2]], ptr @c)
+; CHECK-NEXT: br label %[[SW_EPILOG]]
+; CHECK: [[SW_EPILOG]]:
+; CHECK-NEXT: ret void
+;
+entry:
+ switch i64 %num, label %sw.epilog [
+ i64 1, label %sw.bb
+ i64 0, label %sw.bb1
+ ]
+
+sw.bb: ; preds = %entry
+ %1 = load i8, ptr %ptr, align 1
+ call void asm sideeffect "", "r,r,~{dirflag},~{fpsr},~{flags}"(i8 %1, ptr @c)
+ br label %sw.epilog
+
+sw.bb1: ; preds = %entry
+ %2 = load i8, ptr %ptr, align 1
+ call void asm sideeffect "movdqu 0 %xmm0 \0A\09", "r,r,~{dirflag},~{fpsr},~{flags}"(i8 %2, ptr @c)
+ br label %sw.epilog
+
+sw.epilog: ; preds = %sw.bb1, %sw.bb, %entry
+ ret void
+}
|
nikic
reviewed
May 3, 2025
nikic
reviewed
May 3, 2025
|
Yes, that seems to be why. For GVNSink, I'd guess its still better to skip numbering those instructions, but maybe the check in |
ilovepi
force-pushed
the
users/ilovepi/gvn-sink-bug
branch
from
49c2f79 to
e680ddb
Compare
|
✅ With the latest revision this PR passed the C/C++ code formatter. |
|
I updated |
ilovepi
force-pushed
the
users/ilovepi/gvn-sink-bug
branch
from
e680ddb to
b3b25a5
Compare
nikic
reviewed
May 5, 2025
|
I'd probably only leave the canReplaceOperandWithVariable() check -- you're right that we can avoid numbering this way, but given how rare this case should be, it's probably not worth having a duplicate check. |
|
Sounds good. I'll do that and try out your suggestion. Any idea what's happening on Windows? I'm having trouble getting llvm to build on my windows workstation for some reason (Its been quite some time since I tried, and I think some configs may have changed). It looks like a memory error, but I don't see how my change would introduce that only on Windows. |
Fixes #138345. Before this patch, gvn-sink would try to sink inline assembly statements. Other GVN passes avoid them (see https://github.com/llvm/llvm-project/blob/b4fac94181c4cf17dbb7ecc2ae975712b0e4a6d1/llvm/lib/Transforms/Scalar/GVN.cpp#L2932 Similarly, gvn-sink should skip these instructions, since they are not safe to move. To do this, we update the early exit in canReplaceOperandWithVariable, since it should have caught this case. It's more efficient to also skip numbering in GVNSink if the instruction is InlineAsm, but that should be infrequent. The test added is reduced from a failure when compiling Fuchsia with gvn-sink.
ilovepi
force-pushed
the
users/ilovepi/gvn-sink-bug
branch
from
b3b25a5 to
170ab29
Compare
ilovepi
commented
May 5, 2025
|
Well I guess windows is passing now, so I guess safe to ignore the question about windows. I still don't see anything in the old patch set that makes me think we were reading an invalid pointer for some reason. |
ilovepi
added a commit
that referenced
this pull request
May 6, 2025
I had forgotten to upload the formatting change.
ilovepi
added a commit
that referenced
this pull request
May 6, 2025
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/138/builds/12689 Here is the relevant piece of the build log for the reference |
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.
Fixes #138345. Before this patch, gvn-sink would try to sink inline
assembly statements. Other GVN passes avoid them (see
llvm-project/llvm/lib/Transforms/Scalar/GVN.cpp
Line 2932 in b4fac94
Similarly, gvn-sink should skip these instructions, since they are not
safe to move. To do this, we update the early exit in
canReplaceOperandWithVariable, since it should have caught this case.
It's more efficient to also skip numbering in GVNSink if the instruction
is InlineAsm, but that should be infrequent.
The test added is reduced from a failure when compiling Fuchsia with
gvn-sink.