PGO benchmarks for different serialization/deserialization libraries with https://github.com/llogiq/serdebench benchsuite
- Macbook M1 Pro (6+2 CPU, 16 Gib RAM, 512 SSD)
- macOS 13.6 Ventura
- Rustc version:
rustc 1.77.0-nightly (d78329b92 2024-01-13) - serdebench version:
4cd6b3404758dff633615029b790f32859bf4e0ccommit from themasterbranch
Release results are collected with cargo +nightly bench. PGO optimization is done with cargo-pgo. PGO instrumentation is done with cargo +nightly pgo bench. PGO optimization is done with cargo +nightly pgo optimize bench.
I got the following results:
- Release: https://gist.github.com/zamazan4ik/b030a926ce164aedc46fdf4d5ba0739a
- PGO optimized (compared to Release): https://gist.github.com/zamazan4ik/a9d4123ed9d1b6820d0591431be9dbfc
- (just for reference) PGO instrumented: https://gist.github.com/zamazan4ik/1258945a185d0d72299cd1913332d0fa
According to the tests, PGO helps with optimizing performance in many cases. However, in the benchmarks, there are regressions with PGO in some rare scenarios.