perf: CUDAExecutionProviderにおける畳み込みアルゴリズム検索をEXHAUSTIVE (0)からHEURISTIC (1)に変更(CUDA EP使用時のWarmupにかかる時間を短縮)
#1162ClosedComparing
Sanzentyo:perf/use-cudnn-heuristic-algo-search
(5e2c945
) withmain
(868289d
)0%
- Untouched
- 14
- Skipped
- 8
- Archived
- 2
Unknown Walltime execution environment detected
Using the Walltime instrument on standard Hosted Runners will lead to inconsistent data. For the most accurate results, we recommend using CodSpeed Macro Runners: bare-metal machines fine-tuned for performance measurement consistency.Benchmarks
24 total
replace_mora_pitch[SHORT]
crates/voicevox_core/benches/benches.rs::blocking+5%
640.5 µs607.2 µs
replace_mora_pitch[SHORT]
crates/voicevox_core/benches/benches.rs::nonblocking+3%
682.6 µs661 µs
replace_phoneme_length[SHORT]
crates/voicevox_core/benches/benches.rs::blocking+1%
105.1 µs104.4 µs
analyze_text[SHORT]
crates/voicevox_core/benches/benches.rs::blocking0%
315 µs313.6 µs
unload_and_load_vvm
crates/voicevox_core/benches/benches.rs::nonblocking0%
1.9 s1.9 s
analyze_text[SHORT]
crates/voicevox_core/benches/benches.rs::nonblocking0%
358.2 µs357 µs
synthesis[SHORT]
crates/voicevox_core/benches/benches.rs::nonblocking0%
2.3 s2.3 s
test_blocking_tts
crates/voicevox_core_python_api/python/benches/test_tts.py0%
2.3 s2.3 s
replace_phoneme_length[SHORT]
crates/voicevox_core/benches/benches.rs::nonblocking0%
132.1 µs132.1 µs
test_asyncio_tts
crates/voicevox_core_python_api/python/benches/test_tts.py0%
2.3 s2.3 s
synthesis[SHORT]
crates/voicevox_core/benches/benches.rs::blocking0%
2.3 s2.3 s
construct_open_jtalk
crates/voicevox_core/benches/benches.rs::blocking0%
87.7 µs88.1 µs
unload_and_load_vvm
crates/voicevox_core/benches/benches.rs::blocking-1%
1.7 s1.7 s
construct_open_jtalk
crates/voicevox_core/benches/benches.rs::nonblocking-1%
136.1 µs138 µs
The benchmarks below were skipped, so their baseline results are used instead. If they were deleted in your codebase, archive them to remove them from the performance reports.
Learn more about archiving benchmarks
Learn more about archiving benchmarks
replace_phoneme_length[LONG]
crates/voicevox_core/benches/benches.rs::blockingSkipped
296.1 µs*
analyze_text[LONG]
crates/voicevox_core/benches/benches.rs::blockingSkipped
1.7 ms*
replace_phoneme_length[LONG]
crates/voicevox_core/benches/benches.rs::nonblockingSkipped
324.8 µs*
analyze_text[LONG]
crates/voicevox_core/benches/benches.rs::nonblockingSkipped
1.6 ms*
replace_mora_pitch[LONG]
crates/voicevox_core/benches/benches.rs::nonblockingSkipped
2.3 ms*
synthesis[LONG]
crates/voicevox_core/benches/benches.rs::blockingSkipped
7.8 s*
replace_mora_pitch[LONG]
crates/voicevox_core/benches/benches.rs::blockingSkipped
2.2 ms*
synthesis[LONG]
crates/voicevox_core/benches/benches.rs::nonblockingSkipped
7.8 s*
The benchmarks below were run, but are now archived.
If they were deleted in another branch, consider rebasing to remove them from the report.
If they were added back in this branch, you can restore them to include them in future performance reports.
Learn more about archiving benchmarks
If they were deleted in another branch, consider rebasing to remove them from the report.
If they were added back in this branch, you can restore them to include them in future performance reports.
Learn more about archiving benchmarks
open_and_close_vvm
crates/voicevox_core/benches/benches.rs::blockingArchived
-5%
381.9 µs402 µs
open_and_close_vvm
crates/voicevox_core/benches/benches.rs::nonblockingArchived
-1%
1.3 ms1.3 ms
Commits
Click on a commit to change the comparison rangeBase
main
868289d
+0.09%
perf: CUDA実行プロバイダーにおける畳み込みアルゴリズム検索をEXHAUSTIVE (0)からHEURISTIC (1)に変更(CUDA EP使用時のWarmupにかかる時間を短縮)
5e2c945
5 months ago
by Sanzentyo