source_span query optimized to be inlined function in case of non-incremental build

[azhogin]

source_span query renamed to source_span_q.
source_span inlined function implemented to call source_span implementation in case of non-incremental build and call query otherwise.

Some other queries may be optimized in a such way.
Statistics prepared (queries with small 'avg_ns_norm' and big 'cache_hits' may be optimized):
Non-feedable only and cache_hits > 500:
queries_stat_non_feedable.csv

[petrochenkov]

@bors2 try @rust-timer queue

[rust-bors]

☀️ Try build successful (CI)
Build commit: 396339e (396339e6a2c138a7c89886391a30bf6ba178ed8a, parent: 40560823602064f4c726aea3e15e104449e1a392)

[rust-timer]

Finished benchmarking commit (396339e): comparison URL.

Overall result: ✅ improvements - no action needed

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.

@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)	-	-	0
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-0.5%	[-0.7%, -0.3%]	2
All ❌✅ (primary)	-	-	0

Max RSS (memory usage)

Results (primary -2.9%)

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

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	-	-	0
Improvements ✅ (primary)	-2.9%	[-2.9%, -2.9%]	1
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-2.9%	[-2.9%, -2.9%]	1

Cycles

Results (secondary -1.6%)

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

	mean	range	count
Regressions ❌ (primary)	-	-	0
Regressions ❌ (secondary)	2.8%	[2.3%, 3.4%]	3
Improvements ✅ (primary)	-	-	0
Improvements ✅ (secondary)	-3.5%	[-4.7%, -2.1%]	7
All ❌✅ (primary)	-	-	0

Binary size

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

Bootstrap: 472.868s -> 472.157s (-0.15%)
Artifact size: 389.97 MiB -> 389.93 MiB (-0.01%)

[zetanumbers]

queries_stat.csv

This stats were derived from collected data about individual query calls/executions during a compilation (or rather a check) on the serde crate. In case anyone finds this data useful.

Details here: #t-compiler > How does the option `query_dep_graph` work? @ 💬

[petrochenkov]

Could you change this to a version that checks the incremental flag every time instead of introducing a hook?
@rustbot author

[rustbot]

Reminder, once the PR becomes ready for a review, use @rustbot ready.

[azhogin]

Could you change this to a version that checks the incremental flag every time instead of introducing a hook?

Changed.
@rustbot ready

[azhogin]

@rustbot ready

[petrochenkov]

Ok, now this looks like a legitimate improvement, but why does it happen in incr-full mode, I don't think it should?

The PR title and description need to be updated before I send this to someone with better query system expertize.
The files with statistics in particular are irrelevant to the change itself.

[petrochenkov]

In the incr-full scenarios the old source_span is still called, under the new name source_span_q, and with the same number of executions and hits (218 and 19357).
It means the wg-grammar change is also noise, after all.

Let's rerun perf on the new master just in case.
@bors2 try @rust-timer queue

[petrochenkov]

If the difference is indeed noise, it means that either:

• Query system overhead in non-incremental case is not larger than a synchronized map lookup like tcx.untracked.source_span.get(...).
• Or the query is just not called often enough to see the difference.

[rust-bors]

☀️ Try build successful (CI)
Build commit: 13bb904 (13bb904abf88308cb050948b3af4c7e2363fffa3, parent: 21a13b8864a7dd614e9a96afd57b58c7fcf0bd6b)

[rust-timer]

Finished benchmarking commit (13bb904): comparison URL.

Overall result: no relevant changes - no action needed

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.

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

Instruction count

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

Max RSS (memory usage)

Results (primary -2.5%, secondary 3.1%)

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

	mean	range	count
Regressions ❌ (primary)	2.1%	[2.1%, 2.1%]	1
Regressions ❌ (secondary)	3.1%	[3.1%, 3.1%]	1
Improvements ✅ (primary)	-7.0%	[-7.0%, -7.0%]	1
Improvements ✅ (secondary)	-	-	0
All ❌✅ (primary)	-2.5%	[-7.0%, 2.1%]	2

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: 470.943s -> 470.808s (-0.03%)
Artifact size: 387.66 MiB -> 387.65 MiB (-0.00%)

[petrochenkov]

Yeah, it's noise.
(I'll leave this open since cjgillot self-assigned.)

[petrochenkov]

#147387 (comment)