Closed Bug 1663139 Opened 5 years ago Closed 5 years ago

2.16 - 10.84% build times / compiler_metrics num_static_constructors (linux64-shippable, windows2012-64) regression on push 57b1de9d90dce7d2050d3894aa5b8b03f3e1460d (Mon August 31 2020)

Categories

(Firefox Build System :: Toolchains, defect)

Firefox 82
defect

Tracking

(firefox82 affected)

RESOLVED WONTFIX
Tracking Status
firefox82 --- affected

People

(Reporter: Bebe, Unassigned)

References

(Regression)

Details

(Keywords: perf-alert, regression)

Perfherder has detected a build_metrics performance regression from push 57b1de9d90dce7d2050d3894aa5b8b03f3e1460d. As author of one of the patches included in that push, we need your help to address this regression.

Regressions:

11% build times linux64-shippable opt nightly taskcluster-c5.4xlarge 2,567.81 -> 2,846.12
10% build times linux64-shippable opt nightly taskcluster-c5d.4xlarge 2,521.52 -> 2,763.16
10% build times linux64-shippable opt nightly taskcluster-m5.4xlarge 2,697.41 -> 2,955.80
9% build times linux64-shippable opt nightly taskcluster-c5d.4xlarge 2,535.43 -> 2,762.42
5% build times windows2012-64 debug plain taskcluster-c4.4xlarge 2,471.75 -> 2,603.16
2% compiler_metrics num_static_constructors linux64-shippable opt instrumented 139.00 -> 142.00

Details of the alert can be found in the alert summary, including links to graphs and comparisons for each of the affected tests. Please follow our guide to handling regression bugs and let us know your plans within 3 business days, or the offending patch(es) will be backed out in accordance with our regression policy.

For more information on performance sheriffing please see our FAQ.

Component: Performance → Toolchains
Flags: needinfo?(dmajor)
Product: Testing → Firefox Build System
See Also: → 1663138

All things considered this is not as bad as I was expecting. In the previous failed attempt to upgrade to clang 10, the regressions such as bug 1638302 were on all platforms. It seems that LLVM in general has recovered from some of those regressions. We're still seeing an increase on Linux PGO, but I suspect that is largely from bug 1641674, where the extra time is spent optimizing code that had previously been accidentally overlooked -- so at least we're getting something in return for the extra processing. I think this time the scope of the regressions is limited enough to be acceptable.

Status: NEW → RESOLVED
Closed: 5 years ago
Flags: needinfo?(dmajor)
Resolution: --- → WONTFIX
Has Regression Range: --- → yes
You need to log in before you can comment on or make changes to this bug.