Skip to content

Introduce debuginfo to statements in MIR #142771

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 8 commits into
base: master
Choose a base branch
from

Conversation

dianqk
Copy link
Member

@dianqk dianqk commented Jun 20, 2025

The PR introduces support for debug information within dead statements. Currently, only the reference statement is supported, which is sufficient to fix #128081.

I don't modify Stable MIR, as I don't think we need debug information when using it.

This PR represents the debug information for the dead reference statement via #dbg_value. For example, let _foo_b = &foo.b becomes #dbg_value(ptr %foo, !22, !DIExpression(DW_OP_plus_uconst, 4, DW_OP_stack_value), !26). You can see this here: https://rust.godbolt.org/z/d43js6adv.

The general principle for handling debug information is to never provide less debug information than the optimized LLVM IR.

The current rules for dropping debug information in this PR are:

I haven't drop debuginfos in MatchBranchSimplification, because LLVM also pick one branch for it.

For the perf result:

I expected this to introduce some regressions; however, the results mixed the effects of inlining. Looking at the doc profile, this is a clear optimization. One potential regression I'm investigating is serde-1.0.219-debug-full.

@rustbot rustbot added A-LLVM Area: Code generation parts specific to LLVM. Both correctness bugs and optimization-related issues. T-compiler Relevant to the compiler team, which will review and decide on the PR/issue. labels Jun 20, 2025
@rust-log-analyzer

This comment has been minimized.

@dianqk dianqk force-pushed the mir-stmt-debuginfo branch from 15c968a to 6b013d4 Compare June 20, 2025 08:01
@rust-log-analyzer

This comment has been minimized.

@dianqk dianqk force-pushed the mir-stmt-debuginfo branch from 6b013d4 to 51576e7 Compare June 20, 2025 09:23
@rust-log-analyzer

This comment has been minimized.

@dianqk
Copy link
Member Author

dianqk commented Jun 20, 2025

@bors2 try @rust-timer queue

@rust-timer

This comment has been minimized.

@rust-bors
Copy link

rust-bors bot commented Jun 20, 2025

⌛ Trying commit 51576e7 with merge eb83156

To cancel the try build, run the command @bors2 try cancel.

rust-bors bot added a commit that referenced this pull request Jun 20, 2025
Introduce debuginfo to statements in MIR

Not ready for reviewing. Something known:

- [ ] Retain debuginfo when concatenating bbs
- [ ] Document about when to drop debuginfos (don't be worse than the optimized LLVM IR)
- [ ] Missing tests

r? ghost
@rustbot rustbot added the S-waiting-on-perf Status: Waiting on a perf run to be completed. label Jun 20, 2025
@rust-bors
Copy link

rust-bors bot commented Jun 20, 2025

☀️ Try build successful (CI)
Build commit: eb83156 (eb83156169ae3bbdd1385d498455dbc44283f5ff, parent: 18491d5be00eb3ed2f1ccee2ac5b792694f2a7a0)

@rust-timer

This comment has been minimized.

@rust-timer
Copy link
Collaborator

Finished benchmarking commit (eb83156): comparison URL.

Overall result: ❌✅ regressions and improvements - please read the text below

Benchmarking this pull request means it may be perf-sensitive – we'll automatically label it not fit for rolling up. You can override this, but we strongly advise not to, due to possible changes in compiler perf.

Next Steps: If you can justify the regressions found in this try perf run, please do so in sufficient writing along with @rustbot label: +perf-regression-triaged. If not, please fix the regressions and do another perf run. If its results are neutral or positive, the label will be automatically removed.

@bors rollup=never
@rustbot label: -S-waiting-on-perf +perf-regression

Instruction count

Our most reliable metric. Used to determine the overall result above. However, even this metric can be noisy.

mean range count
Regressions ❌
(primary)
0.9% [0.2%, 14.5%] 51
Regressions ❌
(secondary)
0.8% [0.1%, 1.6%] 52
Improvements ✅
(primary)
-0.4% [-2.1%, -0.2%] 35
Improvements ✅
(secondary)
-0.3% [-0.5%, -0.2%] 39
All ❌✅ (primary) 0.4% [-2.1%, 14.5%] 86

Max RSS (memory usage)

Results (primary 2.3%, secondary 3.4%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

mean range count
Regressions ❌
(primary)
2.3% [0.8%, 4.7%] 54
Regressions ❌
(secondary)
3.6% [0.7%, 7.4%] 38
Improvements ✅
(primary)
- - 0
Improvements ✅
(secondary)
-3.3% [-3.3%, -3.3%] 1
All ❌✅ (primary) 2.3% [0.8%, 4.7%] 54

Cycles

Results (primary 3.2%, secondary 2.5%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

mean range count
Regressions ❌
(primary)
3.9% [1.1%, 14.4%] 7
Regressions ❌
(secondary)
2.5% [2.0%, 3.3%] 3
Improvements ✅
(primary)
-2.0% [-2.0%, -2.0%] 1
Improvements ✅
(secondary)
- - 0
All ❌✅ (primary) 3.2% [-2.0%, 14.4%] 8

Binary size

Results (primary 0.5%, secondary 0.3%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

mean range count
Regressions ❌
(primary)
0.7% [0.0%, 2.2%] 123
Regressions ❌
(secondary)
1.1% [0.0%, 3.5%] 59
Improvements ✅
(primary)
-0.9% [-6.5%, -0.1%] 11
Improvements ✅
(secondary)
-0.5% [-13.0%, -0.1%] 55
All ❌✅ (primary) 0.5% [-6.5%, 2.2%] 134

Bootstrap: 691.482s -> 692.445s (0.14%)
Artifact size: 371.94 MiB -> 372.12 MiB (0.05%)

@rustbot rustbot added perf-regression Performance regression. and removed S-waiting-on-perf Status: Waiting on a perf run to be completed. labels Jun 20, 2025
@dianqk dianqk force-pushed the mir-stmt-debuginfo branch from 51576e7 to e72c3ae Compare June 21, 2025 02:31
@dianqk
Copy link
Member Author

dianqk commented Jun 21, 2025

@bors2 try @rust-timer queue

@rust-timer

This comment has been minimized.

@rust-bors
Copy link

rust-bors bot commented Jun 21, 2025

⌛ Trying commit e72c3ae with merge 77d5c6a

To cancel the try build, run the command @bors2 try cancel.

rust-bors bot added a commit that referenced this pull request Jun 21, 2025
Introduce debuginfo to statements in MIR

Not ready for reviewing. Something known:

- [ ] Retain debuginfo when concatenating bbs
- [ ] Document about when to drop debuginfos (don't be worse than the optimized LLVM IR)
- [ ] Missing tests

r? ghost
@rustbot rustbot added the S-waiting-on-perf Status: Waiting on a perf run to be completed. label Jun 21, 2025
@rust-log-analyzer

This comment has been minimized.

@rust-bors
Copy link

rust-bors bot commented Jun 21, 2025

☀️ Try build successful (CI)
Build commit: 77d5c6a (77d5c6a20a77fabdee3790a412618b82178e9ab4, parent: 15c701fbc995eb6c5b3a86021c18185f8eee020d)

@rust-timer

This comment has been minimized.

@rust-timer
Copy link
Collaborator

Finished benchmarking commit (77d5c6a): comparison URL.

Overall result: ❌✅ regressions and improvements - please read the text below

Benchmarking this pull request means it may be perf-sensitive – we'll automatically label it not fit for rolling up. You can override this, but we strongly advise not to, due to possible changes in compiler perf.

Next Steps: If you can justify the regressions found in this try perf run, please do so in sufficient writing along with @rustbot label: +perf-regression-triaged. If not, please fix the regressions and do another perf run. If its results are neutral or positive, the label will be automatically removed.

@bors rollup=never
@rustbot label: -S-waiting-on-perf +perf-regression

Instruction count

Our most reliable metric. Used to determine the overall result above. However, even this metric can be noisy.

mean range count
Regressions ❌
(primary)
0.9% [0.1%, 14.5%] 53
Regressions ❌
(secondary)
0.8% [0.1%, 1.6%] 50
Improvements ✅
(primary)
-0.4% [-3.0%, -0.2%] 46
Improvements ✅
(secondary)
-0.3% [-0.5%, -0.2%] 41
All ❌✅ (primary) 0.3% [-3.0%, 14.5%] 99

Max RSS (memory usage)

Results (primary 2.4%, secondary 3.4%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

mean range count
Regressions ❌
(primary)
2.4% [0.5%, 5.7%] 38
Regressions ❌
(secondary)
4.2% [1.3%, 6.2%] 25
Improvements ✅
(primary)
- - 0
Improvements ✅
(secondary)
-1.7% [-3.0%, -1.1%] 4
All ❌✅ (primary) 2.4% [0.5%, 5.7%] 38

Cycles

Results (primary 2.1%, secondary 0.5%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

mean range count
Regressions ❌
(primary)
3.0% [0.8%, 14.1%] 10
Regressions ❌
(secondary)
3.3% [2.3%, 4.3%] 2
Improvements ✅
(primary)
-2.3% [-2.9%, -1.7%] 2
Improvements ✅
(secondary)
-2.2% [-2.6%, -1.9%] 2
All ❌✅ (primary) 2.1% [-2.9%, 14.1%] 12

Binary size

Results (primary 0.5%, secondary 0.3%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

mean range count
Regressions ❌
(primary)
0.7% [0.0%, 3.0%] 122
Regressions ❌
(secondary)
1.1% [0.0%, 3.5%] 59
Improvements ✅
(primary)
-0.7% [-4.9%, -0.0%] 15
Improvements ✅
(secondary)
-0.5% [-12.7%, -0.0%] 55
All ❌✅ (primary) 0.5% [-4.9%, 3.0%] 137

Bootstrap: 690.617s -> 691.47s (0.12%)
Artifact size: 371.83 MiB -> 372.01 MiB (0.05%)

@rustbot rustbot removed the S-waiting-on-perf Status: Waiting on a perf run to be completed. label Jun 21, 2025
@bors
Copy link
Collaborator

bors commented Jun 25, 2025

☔ The latest upstream changes (presumably #142870) made this pull request unmergeable. Please resolve the merge conflicts.

@dianqk
Copy link
Member Author

dianqk commented Jul 20, 2025

@bors2 try parent=3014e79f9c8d5510ea7b3a3b70d171d0948b1e96 @rust-timer queue

@rust-timer

This comment has been minimized.

@rust-bors
Copy link

rust-bors bot commented Jul 20, 2025

⌛ Trying commit 0218e19 with merge 5fdaef0

To cancel the try build, run the command @bors2 try cancel.

rust-bors bot added a commit that referenced this pull request Jul 20, 2025
Introduce debuginfo to statements in MIR

The PR introduces support for debug information within dead statements. Currently, only the reference statement is supported, which is sufficient to fix #128081.

I don't modify Stable MIR, as I don't think we need debug information when using it.

This PR represents the debug information for the dead reference statement via `#dbg_value`. For example, `let _foo_b = &foo.b` becomes `#dbg_value(ptr %foo, !22, !DIExpression(DW_OP_plus_uconst, 4, DW_OP_stack_value), !26)`. You can see this here: https://rust.godbolt.org/z/d43js6adv.

The general principle for handling debug information is to never provide less debug information than the optimized LLVM IR.

The current rules for dropping debug information in this PR are:

- If the LLVM IR cannot represent a reference address, it's replaced with poison or simply dropped. For example, see: https://rust.godbolt.org/z/shGqPec8W. I'm using poison in all such cases now.
- All debuginfos is dropped when merging multiple successor BBs. An example is available here: https://rust.godbolt.org/z/TE1q3Wq6M.

> I haven't drop debuginfos in `MatchBranchSimplification`, because LLVM also pick one branch for it.

For [the perf result](#142771 (comment)):

I expected this to introduce some regressions;  however, the results mixed the effects of inlining. Looking at the doc profile, this is a clear optimization. One potential regression I'm investigating is `serde-1.0.219-debug-full`.
@rustbot rustbot added the S-waiting-on-perf Status: Waiting on a perf run to be completed. label Jul 20, 2025
@rust-bors
Copy link

rust-bors bot commented Jul 20, 2025

☀️ Try build successful (CI)
Build commit: 5fdaef0 (5fdaef0b905cae04c067bad86d20cd228389afad, parent: 3014e79f9c8d5510ea7b3a3b70d171d0948b1e96)

@rust-timer

This comment has been minimized.

@rust-timer
Copy link
Collaborator

Finished benchmarking commit (5fdaef0): comparison URL.

Overall result: ❌✅ regressions and improvements - please read the text below

Benchmarking this pull request means it may be perf-sensitive – we'll automatically label it not fit for rolling up. You can override this, but we strongly advise not to, due to possible changes in compiler perf.

Next Steps: If you can justify the regressions found in this try perf run, please do so in sufficient writing along with @rustbot label: +perf-regression-triaged. If not, please fix the regressions and do another perf run. If its results are neutral or positive, the label will be automatically removed.

@bors rollup=never
@rustbot label: -S-waiting-on-perf +perf-regression

Instruction count

Our most reliable metric. Used to determine the overall result above. However, even this metric can be noisy.

mean range count
Regressions ❌
(primary)
0.6% [0.2%, 3.3%] 38
Regressions ❌
(secondary)
0.7% [0.1%, 1.6%] 51
Improvements ✅
(primary)
-0.4% [-1.3%, -0.1%] 88
Improvements ✅
(secondary)
-0.6% [-2.0%, -0.0%] 73
All ❌✅ (primary) -0.1% [-1.3%, 3.3%] 126

Max RSS (memory usage)

Results (primary 2.5%, secondary 3.1%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

mean range count
Regressions ❌
(primary)
2.6% [0.7%, 8.8%] 60
Regressions ❌
(secondary)
3.8% [1.0%, 6.0%] 30
Improvements ✅
(primary)
-0.9% [-0.9%, -0.9%] 1
Improvements ✅
(secondary)
-2.5% [-3.3%, -2.2%] 4
All ❌✅ (primary) 2.5% [-0.9%, 8.8%] 61

Cycles

Results (primary -3.0%, secondary -2.2%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

mean range count
Regressions ❌
(primary)
- - 0
Regressions ❌
(secondary)
2.6% [2.6%, 2.6%] 1
Improvements ✅
(primary)
-3.0% [-3.0%, -3.0%] 1
Improvements ✅
(secondary)
-4.6% [-6.4%, -2.8%] 2
All ❌✅ (primary) -3.0% [-3.0%, -3.0%] 1

Binary size

Results (primary 0.5%, secondary 0.4%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

mean range count
Regressions ❌
(primary)
0.7% [0.1%, 3.2%] 107
Regressions ❌
(secondary)
1.1% [0.0%, 3.5%] 57
Improvements ✅
(primary)
-0.5% [-6.6%, -0.0%] 25
Improvements ✅
(secondary)
-0.3% [-6.6%, -0.0%] 56
All ❌✅ (primary) 0.5% [-6.6%, 3.2%] 132

Bootstrap: 465.928s -> 465.853s (-0.02%)
Artifact size: 374.79 MiB -> 374.42 MiB (-0.10%)

@rustbot rustbot removed the S-waiting-on-perf Status: Waiting on a perf run to be completed. label Jul 20, 2025
@dianqk dianqk force-pushed the mir-stmt-debuginfo branch from 0218e19 to dabec66 Compare July 21, 2025 15:12
@dianqk
Copy link
Member Author

dianqk commented Jul 21, 2025

@bors2 try @rust-timer queue

@rust-timer

This comment has been minimized.

@rust-bors
Copy link

rust-bors bot commented Jul 21, 2025

⌛ Trying commit dabec66 with merge 623f848

To cancel the try build, run the command @bors2 try cancel.

rust-bors bot added a commit that referenced this pull request Jul 21, 2025
Introduce debuginfo to statements in MIR

The PR introduces support for debug information within dead statements. Currently, only the reference statement is supported, which is sufficient to fix #128081.

I don't modify Stable MIR, as I don't think we need debug information when using it.

This PR represents the debug information for the dead reference statement via `#dbg_value`. For example, `let _foo_b = &foo.b` becomes `#dbg_value(ptr %foo, !22, !DIExpression(DW_OP_plus_uconst, 4, DW_OP_stack_value), !26)`. You can see this here: https://rust.godbolt.org/z/d43js6adv.

The general principle for handling debug information is to never provide less debug information than the optimized LLVM IR.

The current rules for dropping debug information in this PR are:

- If the LLVM IR cannot represent a reference address, it's replaced with poison or simply dropped. For example, see: https://rust.godbolt.org/z/shGqPec8W. I'm using poison in all such cases now.
- All debuginfos is dropped when merging multiple successor BBs. An example is available here: https://rust.godbolt.org/z/TE1q3Wq6M.

> I haven't drop debuginfos in `MatchBranchSimplification`, because LLVM also pick one branch for it.

For [the perf result](#142771 (comment)):

I expected this to introduce some regressions;  however, the results mixed the effects of inlining. Looking at the doc profile, this is a clear optimization. One potential regression I'm investigating is `serde-1.0.219-debug-full`.
@rustbot rustbot added the S-waiting-on-perf Status: Waiting on a perf run to be completed. label Jul 21, 2025
@rust-bors
Copy link

rust-bors bot commented Jul 21, 2025

☀️ Try build successful (CI)
Build commit: 623f848 (623f848d1cb53fd526ea3c096cb88e185942d48d, parent: 67819923ac8ea353aaa775303f4c3aacbf41d010)

@rust-timer

This comment has been minimized.

@rust-timer
Copy link
Collaborator

Finished benchmarking commit (623f848): comparison URL.

Overall result: ❌✅ regressions and improvements - please read the text below

Benchmarking this pull request means it may be perf-sensitive – we'll automatically label it not fit for rolling up. You can override this, but we strongly advise not to, due to possible changes in compiler perf.

Next Steps: If you can justify the regressions found in this try perf run, please do so in sufficient writing along with @rustbot label: +perf-regression-triaged. If not, please fix the regressions and do another perf run. If its results are neutral or positive, the label will be automatically removed.

@bors rollup=never
@rustbot label: -S-waiting-on-perf +perf-regression

Instruction count

Our most reliable metric. Used to determine the overall result above. However, even this metric can be noisy.

mean range count
Regressions ❌
(primary)
0.8% [0.2%, 2.8%] 9
Regressions ❌
(secondary)
0.6% [0.0%, 1.0%] 27
Improvements ✅
(primary)
-0.4% [-1.2%, -0.2%] 59
Improvements ✅
(secondary)
-0.4% [-1.2%, -0.2%] 47
All ❌✅ (primary) -0.2% [-1.2%, 2.8%] 68

Max RSS (memory usage)

Results (primary 2.1%, secondary 3.4%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

mean range count
Regressions ❌
(primary)
2.3% [0.4%, 6.7%] 50
Regressions ❌
(secondary)
3.7% [1.4%, 6.3%] 36
Improvements ✅
(primary)
-3.8% [-5.5%, -2.0%] 2
Improvements ✅
(secondary)
-3.0% [-3.5%, -2.4%] 2
All ❌✅ (primary) 2.1% [-5.5%, 6.7%] 52

Cycles

Results (primary -2.2%, secondary 0.2%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

mean range count
Regressions ❌
(primary)
- - 0
Regressions ❌
(secondary)
1.8% [1.7%, 1.9%] 2
Improvements ✅
(primary)
-2.2% [-2.2%, -2.2%] 1
Improvements ✅
(secondary)
-3.1% [-3.1%, -3.1%] 1
All ❌✅ (primary) -2.2% [-2.2%, -2.2%] 1

Binary size

Results (primary 0.5%, secondary 0.4%)

A less reliable metric. May be of interest, but not used to determine the overall result above.

mean range count
Regressions ❌
(primary)
0.7% [0.1%, 3.2%] 111
Regressions ❌
(secondary)
1.1% [0.0%, 3.6%] 57
Improvements ✅
(primary)
-0.5% [-6.6%, -0.0%] 22
Improvements ✅
(secondary)
-0.3% [-6.6%, -0.1%] 55
All ❌✅ (primary) 0.5% [-6.6%, 3.2%] 133

Bootstrap: 465.321s -> 466.6s (0.27%)
Artifact size: 374.50 MiB -> 374.73 MiB (0.06%)

@rustbot rustbot removed the S-waiting-on-perf Status: Waiting on a perf run to be completed. label Jul 21, 2025
@bors
Copy link
Collaborator

bors commented Jul 22, 2025

☔ The latest upstream changes (presumably #144249) made this pull request unmergeable. Please resolve the merge conflicts.

F: FnMut(&Statement<'tcx>) -> bool,
{
// Fast path without debugging information
if self.statements.iter().all(|stmt| stmt.debuginfos.is_empty()) {
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

is it really faster to iterate twice compared with having a few is_empty() checks within the retain_mut closure?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Emm, local perf says no.
I will remove this at the next rebase.

self.statements.retain_mut(|stmt| f(stmt));
return;
}
let mut debuginfos = Vec::new();
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

could make this a boolean and directly push to after_last_stmt_debuginfos

Comment on lines +28 to +34
let replaced_stmt = std::mem::replace(&mut self.kind, StatementKind::Nop);
if !drop_debuginfo {
let Some(debuginfo) = replaced_stmt.as_debuginfo() else {
bug!("debuginfo is not yet supported.")
};
self.debuginfos.push(debuginfo);
}
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

So... am I understanding it right that only StatementKind::Nop statements can have a non-empty debuginfos?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

After strip_nops, a non-empty debuginfos can be placed at any statement.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

hmm... maybe we could have a StatementKind::DebugInfo(debuginfos) that gets produced instead of a nop when a statement is erased that should have debuginfo. Then strip_nops could also collapse consecutive DebugInfo statements. This would also avoid having to have debuginfos in the basic block and handling them in block-merging optimizations.

Not sure that is better than what you have right now, but I'd like to hear your thoughts on it (maybe this was already dismissed in other discussions?).

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
A-LLVM Area: Code generation parts specific to LLVM. Both correctness bugs and optimization-related issues. perf-regression Performance regression. S-waiting-on-author Status: This is awaiting some action (such as code changes or more information) from the author. T-compiler Relevant to the compiler team, which will review and decide on the PR/issue.
Projects
None yet
Development

Successfully merging this pull request may close these issues.

Bad codegen for non-copy-derived struct with all Copy derived fields
8 participants