codegen: Add NewSink LLVM pass for instruction sinking #60879
+4,185
−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
gbaraldi
force-pushed
the
gb/newsink
branch
3 times, most recently
from
3699514 to
543d575
Compare
vchuravy
reviewed
Jan 30, 2026
Comment on lines
+63
to
+85
| entry: | ||
| %loc = alloca i64, align 8 | ||
| store i64 1, ptr %loc, align 8 | ||
| br i1 %c.1, label %if.then, label %if.end | ||
|
|
||
| if.then: | ||
| call void @throw(ptr %loc) | ||
| unreachable | ||
|
|
||
| if.end: | ||
| store i64 0, ptr %loc, align 8 | ||
| ret void | ||
| } |
Member
There was a problem hiding this comment.
Missing the actuall checks? Also in other tests.
KristofferC
added
the
needs pkgeval
Adds a new LLVM pass that sinks instructions to paths where they are actually needed. This is useful for Julia's bounds checking where error message data is computed eagerly but only needed when bounds checks fail. The pass uses MemorySSA to track memory dependencies and verify that sinking writes doesn't change observable behavior. It handles: - Sinking to single-predecessor successors (D136218-style) - Sinking to noreturn/error blocks - Memory intrinsics (memset, memcpy, memmove) - Pointer captures in dominated blocks Loop handling uses LoopInfo and PostDominatorTree: - Won't sink into a different loop - In loops, requires target to post-dominate source (ensures all iterations execute the store, avoiding stale values) - Noreturn targets exempt from loop checks (no future iterations) This addresses the limitation that MemorySSA doesn't model cross-iteration effects in loops. All transformations validated with alive-tv. Co-Authored-By: Claude Opus 4.5 <[email protected]>
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.
This is very vibe coded, but also has lots of tests and I have gone through it and haven't seen anything at that glance.
This fixes some of the issues we have in code like by sinking the stores into the error path
new
old
Adds a new LLVM pass that sinks instructions to paths where they are actually needed. This is useful for Julia's bounds checking where error message data is computed eagerly but only needed when bounds checks fail.
The pass uses MemorySSA to track memory dependencies and verify that sinking writes doesn't change observable behavior. It handles:
All transformations validated with alive-tv.