Use thread local dep graph encoding

[Zoxc]

This adds thread local encoding of dep graph nodes. Each thread has a MemEncoder that gets flushed to the global FileEncoder when it exceeds 64 kB. Each thread also has a local cache of dep indices. This means there can now be empty gaps in SerializedDepGraph.

Indices are marked green and also allocated by the new atomic operation DepNodeColorMap::try_mark_green as the encoder lock is removed.

This is based on #139756.

[rustbot]

r? @lcnr

rustbot has assigned @lcnr.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

[Zoxc]

@bors try @rust-timer queue

[bors]

⌛ Trying commit 8167fdb with merge 5545f9c...

[Zoxc]

Local results, 1 thread, including #139756, excluding #139636:

Benchmark	Before	After		Before	After		Before	After
	Time	Time	%	Physical Memory	Physical Memory	%	Committed Memory	Committed Memory	%
🟣 clap:check:unchanged	0.3046s	0.3076s	0.98%	96.94 MiB	98.15 MiB	💔 1.24%	168.25 MiB	169.79 MiB	0.92%
🟣 hyper:check:unchanged	0.1211s	0.1221s	0.84%	61.84 MiB	62.02 MiB	0.29%	125.11 MiB	125.54 MiB	0.35%
🟣 regex:check:unchanged	0.2250s	0.2271s	0.92%	78.13 MiB	78.98 MiB	💔 1.08%	145.56 MiB	146.65 MiB	0.75%
🟣 syn:check:unchanged	0.4928s	0.4971s	0.88%	118.76 MiB	120.38 MiB	💔 1.36%	193.57 MiB	195.51 MiB	💔 1.00%
Total	1.1436s	1.1540s	0.91%	355.67 MiB	359.52 MiB	💔 1.08%	632.49 MiB	637.50 MiB	0.79%
Summary	1.0000s	1.0090s	0.90%	1 byte	1.01 bytes	0.99%	1 byte	1.01 bytes	0.75%

Local results, 7 threads, including #139756, excluding #139636:

Benchmark	Before	After		Before	After		Before	After
	Time	Time	%	Physical Memory	Physical Memory	%	Committed Memory	Committed Memory	%
🟣 clap:check:unchanged	0.3710s	0.3140s	💚 -15.37%	133.25 MiB	134.96 MiB	💔 1.29%	235.08 MiB	237.04 MiB	0.83%
🟣 hyper:check:unchanged	0.1466s	0.1303s	💚 -11.11%	85.72 MiB	86.31 MiB	0.69%	191.98 MiB	192.88 MiB	0.47%
🟣 regex:check:unchanged	0.2831s	0.2353s	💚 -16.86%	113.02 MiB	114.25 MiB	💔 1.09%	212.53 MiB	214.18 MiB	0.78%
🟣 syn:check:unchanged	0.5756s	0.5078s	💚 -11.78%	157.85 MiB	160.31 MiB	💔 1.56%	260.80 MiB	263.61 MiB	💔 1.08%
Total	1.3763s	1.1875s	💚 -13.72%	489.84 MiB	495.83 MiB	💔 1.22%	900.39 MiB	907.70 MiB	0.81%
Summary	1.0000s	0.8622s	💚 -13.78%	1 byte	1.01 bytes	💔 1.16%	1 byte	1.01 bytes	0.79%

[bors]

☀️ Try build successful - checks-actions
Build commit: 5545f9c (5545f9c456a47860eda2f034f2d7636b31cff81c)

[rust-timer]

Finished benchmarking commit (5545f9c): comparison URL.

Overall result: ❌✅ regressions and improvements - please read the text below

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 may lead 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-perf +perf-regression

Instruction count

This 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.

	mean	range	count
Regressions ❌ (primary)	0.3%	[0.2%, 0.5%]	16
Regressions ❌ (secondary)	0.5%	[0.4%, 0.8%]	11
Improvements ✅ (primary)	-0.6%	[-0.9%, -0.4%]	103
Improvements ✅ (secondary)	-0.6%	[-1.0%, -0.3%]	24
All ❌✅ (primary)	-0.4%	[-0.9%, 0.5%]	119

Max RSS (memory usage)

Results (primary 1.8%, secondary 0.3%)

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.

	mean	range	count
Regressions ❌ (primary)	1.8%	[1.1%, 3.0%]	35
Regressions ❌ (secondary)	3.8%	[2.2%, 9.2%]	7
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-3.3%	[-5.1%, -2.3%]	7
All ❌✅ (primary)	1.8%	[1.1%, 3.0%]	35

Cycles

This benchmark run did not return any relevant results for this metric.

Binary size

This benchmark run did not return any relevant results for this metric.

Bootstrap: 779.607s -> 782.154s (0.33%)
Artifact size: 365.51 MiB -> 365.50 MiB (-0.00%)

[lcnr]

r? compiler

[Zoxc]

I'd probably disregard the instruction improvements here and in #139756 as they don't translate into wall time improvements nor are they expected. Maybe MemEncoder uses less instructions?

I think the trade off towards the parallel compiler is reasonable here as incremental builds tends to have more free cores. Do note the 8.6% regression in dep graph size due to the added indices too.

#139756 is split out mostly to get a separate perf run and doesn't make sense without also merging this PR.

r? @oli-obk

[bors]

☔ The latest upstream changes (presumably #139938) made this pull request unmergeable. Please resolve the merge conflicts.

[cjgillot]

Is this batch allocation really necessary? Could we have a single atomic counter to assign the index and avoid the complexity?

[Zoxc]

We need at least 1 thread local allocation to avoid races when marking nodes green. A single counter would probably result in some congestion as promoting nodes is fairly hot too.