tree 86c1257ea7f67113ee2dd769071b846194fc2315
parent 651e16f083bb30729561be1e8bb994893fc2bb53
author Takuto Ikuta <tikuta@google.com> 1771315990 +0900
committer GN LUCI <gn-scoped@luci-project-accounts.iam.gserviceaccount.com> 1771339382 -0800

[perf] Share ResolvedTargetData across worker threads

This commit modifies ResolvedTargetData to be thread-safe using a
combination of std::shared_mutex for the target-to-info map and
per-target mutexes with atomic flags for individual property
computations (double-checked locking).

By sharing a single ResolvedTargetData instance across all worker
threads in command_gen.cc, we eliminate redundant computations of
transitive dependencies (libs, hard_deps, etc.) that were previously
performed independently by each thread.

Hyperfine benchmark results on a large Chromium build graph (gn gen):
- main: 3.538 s ± 0.041 s [User: 18.883 s, System: 2.922 s]
- shared: 2.894 s ± 0.041 s [User: 13.647 s, System: 3.072 s]
- Improvement: ~22% faster wall-clock time, ~28% reduction in user CPU time.

Bug: 484863025
Change-Id: Ib9c31feee6a118423b1bb6cf4e5552be660cb7c4
Reviewed-on: https://gn-review.googlesource.com/c/gn/+/21000
Commit-Queue: Takuto Ikuta <tikuta@google.com>
Reviewed-by: David Turner <digit@google.com>