[TOY] Extend GVN to perform local value numbering.

[cjgillot]

This PR is more a toy than anything else, but I still think the implementation is sound.

Current GVN MIR opt can be easily extended to track values that change inside a basic block. This PR attempts that.

r? @ghost for perf

[cjgillot]

@bors try @rust-timer queue

[bors]

⌛ Trying commit c3be10b with merge faceb52...

[bors]

☀️ Try build successful - checks-actions
Build commit: faceb52 (faceb52e72c06f94883ba351a73c83544ab4154b)

[rust-timer]

Finished benchmarking commit (faceb52): 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)	15.4%	[0.2%, 937.4%]	227
Regressions ❌ (secondary)	10.3%	[0.2%, 569.6%]	237
Improvements ✅ (primary)	-1.1%	[-1.8%, -0.4%]	2
Improvements ✅ (secondary)	-1.3%	[-4.1%, -0.3%]	16
All ❌✅ (primary)	15.3%	[-1.8%, 937.4%]	229

Max RSS (memory usage)

Results (primary 9.1%, secondary 2.4%)

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

	mean	range	count
Regressions ❌ (primary)	71.9%	[1.6%, 377.4%]	14
Regressions ❌ (secondary)	28.4%	[2.2%, 134.5%]	15
Improvements ✅ (primary)	-2.0%	[-7.5%, -0.7%]	79
Improvements ✅ (secondary)	-2.5%	[-3.5%, -0.8%]	80
All ❌✅ (primary)	9.1%	[-7.5%, 377.4%]	93

Cycles

Results (primary 44.8%, secondary 29.7%)

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

	mean	range	count
Regressions ❌ (primary)	46.5%	[2.3%, 299.4%]	27
Regressions ❌ (secondary)	40.6%	[2.4%, 263.2%]	28
Improvements ✅ (primary)	-2.1%	[-2.1%, -2.1%]	1
Improvements ✅ (secondary)	-4.2%	[-6.7%, -2.4%]	9
All ❌✅ (primary)	44.8%	[-2.1%, 299.4%]	28

Binary size

Results (primary -0.1%, secondary 0.1%)

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

	mean	range	count
Regressions ❌ (primary)	0.2%	[0.0%, 0.6%]	37
Regressions ❌ (secondary)	0.5%	[0.0%, 1.7%]	70
Improvements ✅ (primary)	-0.3%	[-1.3%, -0.0%]	54
Improvements ✅ (secondary)	-1.0%	[-12.8%, -0.0%]	24
All ❌✅ (primary)	-0.1%	[-1.3%, 0.6%]	91

Bootstrap: 458.797s -> 582.037s (26.86%)
Artifact size: 372.11 MiB -> 361.76 MiB (-2.78%)

[cjgillot]

@bors try @rust-timer queue

[bors]

⌛ Trying commit e521268 with merge dc4aa90...

[cjgillot]

@bors try @rust-timer queue

[bors]

⌛ Trying commit 0293254 with merge 4d16c54...

[rust-log-analyzer]

The job pr-check-2 failed! Check out the build log: (web) (plain enhanced) (plain)

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

    Checking rustc_mir_transform v0.0.0 (/checkout/compiler/rustc_mir_transform)
[RUSTC-TIMING] rustc_codegen_llvm test:false 3.796
    Checking rustc_passes v0.0.0 (/checkout/compiler/rustc_passes)
[RUSTC-TIMING] rustc_passes test:false 1.696
error[E0609]: no field `non_ssa` on type `(&VnIndex, &SmallVec<[(rustc_middle::mir::Local, rustc_middle::mir::Location); 1]>)`
   --> compiler/rustc_mir_transform/src/gvn.rs:398:45
    |
398 |                 for (other, _) in local_vec.non_ssa {
    |                                             ^^^^^^^ unknown field

For more information about this error, try `rustc --explain E0609`.
[RUSTC-TIMING] rustc_mir_transform test:false 3.904
error: could not compile `rustc_mir_transform` (lib) due to 1 previous error

[bors]

☀️ Try build successful - checks-actions
Build commit: 4d16c54 (4d16c544e50b0f84cc5347467c108e93106aa8f9)

[Kobzol]

Note: we had some network issues on the collector and this benchmark was restarted multiple times. I hope that it didn't affect the results, but in case it comes out weirdly, I'd rerun it (this PR would probably have "weird" results anyway though 😆).

[rust-timer]

Finished benchmarking commit (4d16c54): 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)	3.1%	[0.1%, 24.5%]	33
Regressions ❌ (secondary)	3.2%	[0.0%, 21.9%]	27
Improvements ✅ (primary)	-0.4%	[-2.1%, -0.2%]	12
Improvements ✅ (secondary)	-0.4%	[-0.9%, -0.1%]	55
All ❌✅ (primary)	2.2%	[-2.1%, 24.5%]	45

Max RSS (memory usage)

Results (primary 48.7%, secondary 33.4%)

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

	mean	range	count
Regressions ❌ (primary)	57.1%	[1.0%, 382.4%]	18
Regressions ❌ (secondary)	51.3%	[2.8%, 136.0%]	8
Improvements ✅ (primary)	-1.8%	[-2.5%, -1.4%]	3
Improvements ✅ (secondary)	-2.3%	[-2.6%, -2.1%]	4
All ❌✅ (primary)	48.7%	[-2.5%, 382.4%]	21

Cycles

Results (primary 8.6%, secondary 14.5%)

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

	mean	range	count
Regressions ❌ (primary)	10.9%	[1.4%, 36.5%]	10
Regressions ❌ (secondary)	14.5%	[2.8%, 34.4%]	7
Improvements ✅ (primary)	-2.5%	[-2.7%, -2.3%]	2
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	8.6%	[-2.7%, 36.5%]	12

Binary size

Results (primary -0.1%, secondary -0.7%)

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

	mean	range	count
Regressions ❌ (primary)	0.4%	[0.0%, 1.0%]	3
Regressions ❌ (secondary)	0.2%	[0.1%, 0.5%]	4
Improvements ✅ (primary)	-0.1%	[-0.7%, -0.0%]	51
Improvements ✅ (secondary)	-0.9%	[-13.0%, -0.0%]	17
All ❌✅ (primary)	-0.1%	[-0.7%, 1.0%]	54

Bootstrap: 460.756s -> 480.789s (4.35%)
Artifact size: 372.14 MiB -> 371.68 MiB (-0.12%)