Add local DSE optimization pass

[JakobDegen]

This PR adds an optimization pass that eliminates stores that can be proven dead within the basic block in which they appear. The full behavior and arguments for correctness of the optimization are documented in the source.

Making this a pass of its own was almost an afterthought; I originally needed this functionality for another opt I was working. If it turns out that this is bad for perf or something, then the pass can definitely be disabled, and we can just leave the functionality in for other opts to use.

Marking as a draft because the src/test/mir-opt/simplify-locals.rs test is still broken under this. I could not find a way to modify that test in a way that cause simplify locals but not this optimization to fire. Does anyone have more creative ideas? Should the test just be removed? Does simplify locals need to be adjusted somehow?

[rust-highfive]

r? @nagisa

(rust-highfive has picked a reviewer for you, use r? to override)

[rust-log-analyzer]

The job x86_64-gnu-llvm-12 failed! Check out the build log: (web) (plain)

Click to see the possible cause of the failure (guessed by this bot)

failures:

---- [mir-opt] mir-opt/simplify-locals.rs stdout ----
15   
16       bb0: {
17           StorageLive(_1);                 // scope 0 at $DIR/simplify-locals.rs:14:9: 14:14
-           _1 = [const 0_u8; 10];           // scope 0 at $DIR/simplify-locals.rs:14:17: 14:26
19 -         StorageLive(_2);                 // scope 1 at $DIR/simplify-locals.rs:16:20: 16:26
20 -         StorageLive(_3);                 // scope 1 at $DIR/simplify-locals.rs:16:20: 16:26
21 -         StorageLive(_4);                 // scope 1 at $DIR/simplify-locals.rs:16:20: 16:26

- -         _4 = &_1;                        // scope 1 at $DIR/simplify-locals.rs:16:20: 16:26
- -         _3 = _4;                         // scope 1 at $DIR/simplify-locals.rs:16:20: 16:26
- -         _2 = move _3 as &[u8] (Pointer(Unsize)); // scope 1 at $DIR/simplify-locals.rs:16:20: 16:26
25 -         StorageDead(_3);                 // scope 1 at $DIR/simplify-locals.rs:16:25: 16:26
26 -         StorageDead(_4);                 // scope 1 at $DIR/simplify-locals.rs:16:26: 16:27
27 -         StorageDead(_2);                 // scope 1 at $DIR/simplify-locals.rs:16:26: 16:27

thread '[mir-opt] mir-opt/simplify-locals.rs' panicked at 'Actual MIR output differs from expected MIR output /checkout/src/test/mir-opt/simplify_locals.c.SimplifyLocals.diff', src/tools/compiletest/src/runtest.rs:3375:25


failures:
    [mir-opt] mir-opt/simplify-locals.rs

[jackh726]

@bors try @rust-timer queue

[rust-timer]

Awaiting bors try build completion.

@rustbot label: +S-waiting-on-perf

[bors]

⌛ Trying commit 7a30d3e with merge 60bf4a09dcb49ff672dce0a99473714e8a327a74...

[bors]

☀️ Try build successful - checks-actions
Build commit: 60bf4a09dcb49ff672dce0a99473714e8a327a74 (60bf4a09dcb49ff672dce0a99473714e8a327a74)

[rust-timer]

Queued 60bf4a09dcb49ff672dce0a99473714e8a327a74 with parent b8c56fa, future comparison URL.

[rust-timer]

Finished benchmarking commit (60bf4a09dcb49ff672dce0a99473714e8a327a74): comparison url.

Summary: This benchmark run shows 2 relevant improvements 🎉 but 23 relevant regressions 😿 to instruction counts.

• Average relevant regression: 0.9%
• Average relevant improvement: -0.9%
• Largest improvement in instruction counts: -1.1% on full builds of ripgrep opt
• Largest regression in instruction counts: 3.8% on incr-patched: b9b3e592dd cherry picked builds of style-servo debug

If you disagree with this performance assessment, please file an issue in rust-lang/rustc-perf.

Benchmarking 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 led to changes in compiler perf.

Next Steps: If you can justify the regressions found in this try perf run, please indicate this with @rustbot label: +perf-regression-triaged along with sufficient written justification. If you cannot justify the regressions please fix the regressions and do another perf run. If the next run shows neutral or positive results, the label will be automatically removed.

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

[JakobDegen]

Yeah, this is probably just duplicating work that either simplify locals/LLVM are also trying to do, so the mild regression isn't terribly surprising. I'll let review take a look, and can bump this to mir_opt_level >= 3 if there's nothing that stands out

[JakobDegen]

I'm actually going to close this for now, I think I have a more promising idea than this. May re-open if that doesn't pan out