gvn: bail out unavoidable non-ssa locals in repeat #141252
Conversation
We cannot transform `*elem` to `array[idx1]` in the following code,
as `idx1` has already been modified.
```rust
mir! {
let array;
let elem;
{
array = [*val; 5];
elem = &array[idx1];
idx1 = idx2;
RET = *elem;
Return()
}
}
```
rustbot
added
S-waiting-on-review
| _4 = [copy (*_3); 5]; | ||
| _5 = &_4[_1]; | ||
| _1 = copy _2; | ||
| _0 = copy (*_5); |
There was a problem hiding this comment.
|
Your change does work, but I feel like we have a systematic problem that this patch doesn't fix. Isn't GVN supposed to notice that |
| @@ -682,6 +683,7 @@ impl<'body, 'tcx> VnState<'body, 'tcx> { | |||
| } | |||
| ProjectionElem::Index(idx) => { | |||
| if let Value::Repeat(inner, _) = self.get(value) { | |||
| *from_non_ssa_index |= self.locals[idx].is_none(); | |||
| return Some(*inner); | |||
| } | |||
| let idx = self.locals[idx]?; | |||
There was a problem hiding this comment.
@saethlin Yes, right here. This is an exception for the repeat array, because whichever value we access, it's all the same. However, we have overlooked the out-of-bounds case here.
|
Sorry it took me a while to convince myself this is a good strategy for fixing this. I now agree. @bors r+ |
bors
added
S-waiting-on-bors
Rollup of 8 pull requests Successful merges: - #140367 (add `asm_cfg`: `#[cfg(...)]` within `asm!`) - #140894 (Make check-cfg diagnostics work in `#[doc(cfg(..))]`) - #141252 (gvn: bail out unavoidable non-ssa locals in repeat) - #141517 (rustdoc: use descriptive tooltip if doctest is conditionally ignored) - #141551 (Make two transmute-related MIR lints into HIR lint) - #141591 (ci: fix llvm test coverage) - #141647 (Bump master `stage0` compiler) - #141659 (Add `Result::map_or_default` and `Option::map_or_default`) r? `@ghost` `@rustbot` modify labels: rollup
Rollup merge of #141252 - dianqk:gvn-repeat-index, r=saethlin gvn: bail out unavoidable non-ssa locals in repeat Fixes #141251. We cannot transform `*elem` to `array[idx1]` in the following code, as `idx1` has already been modified. ```rust mir! { let array; let elem; { array = [*val; 5]; elem = &array[idx1]; idx1 = idx2; RET = *elem; Return() } } ``` Perhaps I could transform it to `array[0]`, but I prefer the conservative approach. r? mir-opt
Rollup of 8 pull requests Successful merges: - rust-lang/rust#140367 (add `asm_cfg`: `#[cfg(...)]` within `asm!`) - rust-lang/rust#140894 (Make check-cfg diagnostics work in `#[doc(cfg(..))]`) - rust-lang/rust#141252 (gvn: bail out unavoidable non-ssa locals in repeat) - rust-lang/rust#141517 (rustdoc: use descriptive tooltip if doctest is conditionally ignored) - rust-lang/rust#141551 (Make two transmute-related MIR lints into HIR lint) - rust-lang/rust#141591 (ci: fix llvm test coverage) - rust-lang/rust#141647 (Bump master `stage0` compiler) - rust-lang/rust#141659 (Add `Result::map_or_default` and `Option::map_or_default`) r? `@ghost` `@rustbot` modify labels: rollup
Rollup of 8 pull requests Successful merges: - rust-lang#140367 (add `asm_cfg`: `#[cfg(...)]` within `asm!`) - rust-lang#140894 (Make check-cfg diagnostics work in `#[doc(cfg(..))]`) - rust-lang#141252 (gvn: bail out unavoidable non-ssa locals in repeat) - rust-lang#141517 (rustdoc: use descriptive tooltip if doctest is conditionally ignored) - rust-lang#141551 (Make two transmute-related MIR lints into HIR lint) - rust-lang#141591 (ci: fix llvm test coverage) - rust-lang#141647 (Bump master `stage0` compiler) - rust-lang#141659 (Add `Result::map_or_default` and `Option::map_or_default`) r? `@ghost` `@rustbot` modify labels: rollup
|
@rust-timer build 11adb2b |
This comment has been minimized.
This comment has been minimized.
|
Finished benchmarking commit (11adb2b): comparison URL. Overall result: ❌ regressions - no action neededBenchmarking this pull request likely means that it is perf-sensitive, so we're automatically marking it as not fit for rolling up. While you can manually mark this PR as fit for rollup, we strongly recommend not doing so since this PR may lead to changes in compiler perf. @bors rollup=never Instruction countThis is the most reliable metric that we have; it was used to determine the overall result at the top of this comment. However, even this metric can sometimes exhibit noise.
Max RSS (memory usage)Results (primary 6.1%, secondary -0.5%)This is a less reliable metric that may be of interest but was not used to determine the overall result at the top of this comment.
CyclesThis benchmark run did not return any relevant results for this metric. Binary sizeThis benchmark run did not return any relevant results for this metric. Bootstrap: 779.377s -> 780.041s (0.09%) |
Fixes #141251.
We cannot transform
*elemtoarray[idx1]in the following code, asidx1has already been modified.Perhaps I could transform it to
array[0], but I prefer the conservative approach.r? mir-opt