Implement opt-bisect-limit for MIR

[sgasho]

View all comments

closes: #150910

Enable bisecting MIR optimization passes to enhance debuggability.

discussions on zulip: https://rust-lang.zulipchat.com/#narrow/channel/182449-t-compiler.2Fhelp/topic/MIR.20dump.20the.20pass.20names/with/573219207

Check it works

Sample code

fn abs(num: isize) -> usize {
    if num < 0 { -num as usize } else { num as usize }
}

fn main() {
    println!("{}", abs(-10));
}

Output

rustc +mir -Zmir-opt-bisect-limit=30 src/main.rs
BISECT: running pass (1) CheckAlignment on main[89d5]::main
BISECT: running pass (2) CheckNull on main[89d5]::main
BISECT: running pass (3) CheckEnums on main[89d5]::main
BISECT: running pass (4) LowerSliceLenCalls on main[89d5]::main
BISECT: running pass (5) InstSimplify-before-inline on main[89d5]::main
BISECT: running pass (6) ForceInline on main[89d5]::main
BISECT: running pass (7) RemoveStorageMarkers on main[89d5]::main
BISECT: running pass (8) RemoveZsts on main[89d5]::main
BISECT: running pass (9) RemoveUnneededDrops on main[89d5]::main
BISECT: running pass (10) UnreachableEnumBranching on main[89d5]::main
BISECT: running pass (11) SimplifyCfg-after-unreachable-enum-branching on main[89d5]::main
BISECT: running pass (12) InstSimplify-after-simplifycfg on main[89d5]::main
BISECT: running pass (13) SimplifyConstCondition-after-inst-simplify on main[89d5]::main
BISECT: running pass (14) SimplifyLocals-before-const-prop on main[89d5]::main
BISECT: running pass (15) SimplifyLocals-after-value-numbering on main[89d5]::main
BISECT: running pass (16) MatchBranchSimplification on main[89d5]::main
BISECT: running pass (17) SingleUseConsts on main[89d5]::main
BISECT: running pass (18) SimplifyConstCondition-after-const-prop on main[89d5]::main
BISECT: running pass (19) SimplifyConstCondition-final on main[89d5]::main
BISECT: running pass (20) RemoveNoopLandingPads on main[89d5]::main
BISECT: running pass (21) SimplifyCfg-final on main[89d5]::main
BISECT: running pass (22) CopyProp on main[89d5]::main
BISECT: running pass (23) SimplifyLocals-final on main[89d5]::main
BISECT: running pass (24) AddCallGuards on main[89d5]::main
BISECT: running pass (25) PreCodegen on main[89d5]::main
BISECT: running pass (26) CheckAlignment on main[89d5]::abs
BISECT: running pass (27) CheckNull on main[89d5]::abs
BISECT: running pass (28) CheckEnums on main[89d5]::abs
BISECT: running pass (29) LowerSliceLenCalls on main[89d5]::abs
BISECT: running pass (30) InstSimplify-before-inline on main[89d5]::abs
BISECT: NOT running pass (31) ForceInline on main[89d5]::abs
BISECT: NOT running pass (32) RemoveStorageMarkers on main[89d5]::abs
BISECT: NOT running pass (33) RemoveZsts on main[89d5]::abs
BISECT: NOT running pass (34) RemoveUnneededDrops on main[89d5]::abs
BISECT: NOT running pass (35) UnreachableEnumBranching on main[89d5]::abs
BISECT: NOT running pass (36) SimplifyCfg-after-unreachable-enum-branching on main[89d5]::abs
BISECT: NOT running pass (37) InstSimplify-after-simplifycfg on main[89d5]::abs
BISECT: NOT running pass (38) SimplifyConstCondition-after-inst-simplify on main[89d5]::abs
BISECT: NOT running pass (39) SimplifyLocals-before-const-prop on main[89d5]::abs
BISECT: NOT running pass (40) SimplifyLocals-after-value-numbering on main[89d5]::abs
BISECT: NOT running pass (41) MatchBranchSimplification on main[89d5]::abs
BISECT: NOT running pass (42) SingleUseConsts on main[89d5]::abs
BISECT: NOT running pass (43) SimplifyConstCondition-after-const-prop on main[89d5]::abs
BISECT: NOT running pass (44) SimplifyConstCondition-final on main[89d5]::abs
BISECT: NOT running pass (45) RemoveNoopLandingPads on main[89d5]::abs
BISECT: NOT running pass (46) SimplifyCfg-final on main[89d5]::abs
BISECT: NOT running pass (47) CopyProp on main[89d5]::abs
BISECT: NOT running pass (48) SimplifyLocals-final on main[89d5]::abs
BISECT: NOT running pass (49) AddCallGuards on main[89d5]::abs
BISECT: NOT running pass (50) PreCodegen on main[89d5]::abs

r? @saethlin

[rustbot]

Some changes occurred to MIR optimizations

cc @rust-lang/wg-mir-opt

[saethlin]

This might have weird interactions with -Zthreads, but I don't see a coherent way to resolve that, apart from mandating that the parallel compiler runs everything in a deterministic order. I think the current approach is good enough for a debugging tool.

squash/fixup the commits into one and then I'll approve :)

[hanna-kruppe]

This might have weird interactions with -Zthreads, but I don't see a coherent way to resolve that, apart from mandating that the parallel compiler runs everything in a deterministic order. I think the current approach is good enough for a debugging tool.

FWIW, an interesting alternative that avoids this problem is hash based bisect. It's harder to implement than a binary search over a single integer, both for the program being bisected (here: rustc) and for doing the bisection (really annoying without automation, maybe the Go tool can be reused), but in return it's order/threading-independent and enables minimizing more precisely than the "fuel" approach.

[saethlin]

🤦 I missed that you actually squashed this down until I got the notification from the merge conflict

[saethlin]

This should prevent it from happening again:

@bors delegate=sgasho

[rust-bors]

Invalid command: Invalid delegation type sgasho. Possible values are try/review. Run @bors help to see available commands.

[saethlin]

@bors delegate

[rust-bors]

✌️ @sgasho, you can now approve this pull request!

If @saethlin told you to "r=me" after making some further change, then please make that change and post @bors r=saethlin.

[rustbot]

This PR was rebased onto a different main commit. Here's a range-diff highlighting what actually changed.

Rebasing is a normal part of keeping PRs up to date, so no action is needed—this note is just to help reviewers.

[sgasho]

@bors r=saethlin rollup

[rust-bors]

📌 Commit 25c2191 has been approved by saethlin

It is now in the queue for this repository.

[jhpratt]

@bors r-

#152691 (comment)

[rust-bors]

Commit 25c2191 has been unapproved.

This PR was contained in a rollup (#152691), which was also unapproved.

[sgasho]

@bors try jobs=dist-various-1

[rust-bors]

☀️ Try build successful (CI)
Build commit: 77ac94f (77ac94f44327a9cbd1105ef1e08305fa768d3804, parent: fef627b1ebdc7369ddf8a4031a5d25733ac1fb34)

[sgasho]

dist-various-1 has been finished successfully.
Just in case, I would like to verify it works on some other targets

@bors try jobs=test-various,x86_64-gnu-llvm-20-2

[rust-bors]

☀️ Try build successful (CI)
Build commit: 601f65d (601f65d36b979a4f572a8f9ef641188a74dea109, parent: 71e00273c0921e1bc850ae8cc4161fbb44cfa848)

[sgasho]

I only added no_std at tests/run-make/mir-opt-bisect-limit/main.rs. so I'm going to bors r=saethlin rollup again when CI is green

[sgasho]

@bors r=saethlin rollup

[rust-bors]

📌 Commit 28c8d71 has been approved by saethlin

It is now in the queue for this repository.