Remove MIR assignments to ZST types #80493
Closed
Conversation
This file contains 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
rust-highfive
added
the
S-waiting-on-review
oli-obk
reviewed
Dec 29, 2020
|
The job Click to see the possible cause of the failure (guessed by this bot) |
|
@oli-obk I would need some help in what to do with those test errors. |
|
The job Click to see the possible cause of the failure (guessed by this bot) |
Jup, exactly what I had in mind. I just thought we already had some code for this and just forgot turning it on again, but apparently not? You can bless most tests that changed. We do need to fix |
oli-obk
reviewed
Dec 30, 2020
|
The job Click to see the possible cause of the failure (guessed by this bot) |
|
You can fix codegen tests by looking at |
|
When the MIR does not have |
|
@bors r+ weird... but I guess that is an independent codegen problem |
|
📌 Commit 22f94499cc9c397420c4cf8ba0bbcabf4018bae7 has been approved by |
bors
added
S-waiting-on-bors
|
@bors r- I thought we had already did this via a combination of the |
bors
added
S-waiting-on-author
|
Ah, ok. I see there was discussion in #80475 about this. I'm willing to accept the logic that we never need writes of ZST values even in the return place but in that case, I think the better thing to do would be to change the logic in |
|
I'll see how I can modify SimplifyLocals 👍 |
simonvandel
force-pushed
the
remove-zst-assign
branch
from
22f9449 to
d1a87e0
Compare
|
Force pushed a new approach that is an extension of SimplifyLocals instead. |
wesleywiser
reviewed
Dec 30, 2020
| let mut is_zst = false; | ||
|
|
||
| // ZST locals can be removed | ||
| if used && !is_static { |
There was a problem hiding this comment.
We refactored this loop a few months ago to push most of this kind of logic into UsedLocals. i think that's probably the place we need to make this change.
It might just be a matter of adjusting this logic:
rust/compiler/rustc_mir/src/transform/simplify.rs
Lines 392 to 398 in e226704
There was a problem hiding this comment.
Took a stab at moving the logic into UsedLocals with the lastest commits
nagisa
reviewed
Dec 31, 2020
src/test/codegen/naked-functions.rs
Outdated
| @@ -9,6 +9,7 @@ | |||
| #[naked] | |||
| pub fn naked_empty() { | |||
| // CHECK-NEXT: {{.+}}: | |||
| // CHECK-NEXT: %0 = alloca {}, align 1 | |||
There was a problem hiding this comment.
I feel like this is probably invalid – for #[naked] functions to be well-formed, they usually should only contain an optional inline assembly call and ret. alloca here will likely make any use of #[naked] plain unsound.
If #[naked] wasn't as unstable as they are, this would be blocking, and IMHO it needing change now warrants at least a folllow-up issue.
There was a problem hiding this comment.
FWIW I'm pretty sure we skip codegenning ZST assignments/allocas/etc in the llvm backend which is why with _0 = const () in MIR it never generated any LLVM. Now that there are no assignments to the _0, its probably taking a different path in the backend and ends up generating the alloca anyway.
If I had to guess this local is failing to be marked as a SSA value?
There was a problem hiding this comment.
So this pr should be blocked until the codegen is fixed to not generate ZST allocas for _0 if it is not assigned in MIR?
I haven't looked at codegen before, so any pointers are appreciated.
There was a problem hiding this comment.
Yeah, that looks like a use of an uninitialized ZST local marking it as non-SSA (implicit use in return terminator?). That part would be in non_ssa_locals.
There was a problem hiding this comment.
The latest commit tries to eliminate the alloca
|
☔ The latest upstream changes (presumably #79328) made this pull request unmergeable. Please resolve the merge conflicts. |
nagisa
reviewed
Jan 14, 2021
|
|
||
| let is_zst = self.fx.cx.layout_of(ty).is_zst(); | ||
| debug!("is_zst: {}", is_zst); | ||
| is_zst |
There was a problem hiding this comment.
I wonder if anything breaks if we do in fact just go ahead and unconditionally mark any variable without assignments as "ssa".
There was a problem hiding this comment.
I tried replacing is_zst with true which at least didn't break test/{mir-opt,codegen}. Should i push that instead of the current zst handling?
There was a problem hiding this comment.
Lets not do that in the current PR, especially because apparently this was added as part of a ICE fix, but without tests. I will experiment with it separately.
cjgillot
reviewed
Jan 21, 2021
| @@ -56,7 +56,6 @@ | |||
| StorageLive(_6); // scope 3 at $DIR/cycle.rs:14:10: 14:11 | |||
| - _6 = _1; // scope 3 at $DIR/cycle.rs:14:10: 14:11 | |||
| + _6 = _4; // scope 3 at $DIR/cycle.rs:14:10: 14:11 | |||
| _5 = const (); // scope 4 at $DIR/cycle.rs:14:5: 14:12 | |||
| StorageDead(_6); // scope 3 at $DIR/cycle.rs:14:11: 14:12 | |||
| StorageDead(_5); // scope 3 at $DIR/cycle.rs:14:12: 14:13 | |||
| _0 = const (); // scope 0 at $DIR/cycle.rs:8:11: 15:2 | |||
There was a problem hiding this comment.
Some assignments are not removed compared to earlier versions of this PR. Why are those not optimized now?
simonvandel
force-pushed
the
remove-zst-assign
branch
from
f1ae27c to
60b0bb3
Compare
|
The job Click to see the possible cause of the failure (guessed by this bot) |
simonvandel
force-pushed
the
remove-zst-assign
branch
from
60b0bb3 to
bab2e56
Compare
|
@bors r+ rollup- |
|
📌 Commit bab2e56 has been approved by |
bors
added
S-waiting-on-bors
|
I think you wanted to do @bors rollup=never
|
|
ooops, thanks. I had a feeling something was wrong when I typed that, but couldn't figure out what |
|
⌛ Testing commit bab2e56 with merge ef6fe77536ea1d8c722aad9fc949e031f01123c9... |
|
The job Click to see the possible cause of the failure (guessed by this bot) |
|
💔 Test failed - checks-actions |
bors
added
S-waiting-on-review
oli-obk
added
S-waiting-on-author
|
☔ The latest upstream changes (presumably #79100) made this pull request unmergeable. Please resolve the merge conflicts. |
|
ping from triage: |
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.
Based on my understanding of #80475 (comment)
r? @oli-obk