Try and improve some noise from cuda builds on macos - #8508
Conversation
Signed-off-by: Adam Gutglick <adam@spiraldb.com>
Merging this PR will not alter performance
|
| Mode | Benchmark | BASE |
HEAD |
Efficiency | |
|---|---|---|---|---|---|
| ❌ | Simulation | decompress_rd[f64, (10000, 0.01)] |
108.7 µs | 139.1 µs | -21.89% |
| ❌ | Simulation | decompress_rd[f64, (10000, 0.1)] |
109 µs | 139.5 µs | -21.85% |
| ❌ | Simulation | decompress_rd[f64, (10000, 0.0)] |
108.7 µs | 139.1 µs | -21.83% |
| ❌ | Simulation | decompress_rd[f32, (100000, 0.0)] |
496 µs | 583.8 µs | -15.05% |
| ❌ | Simulation | decompress_rd[f32, (10000, 0.1)] |
78.1 µs | 91.2 µs | -14.43% |
| ❌ | Simulation | decompress_rd[f32, (10000, 0.01)] |
78.1 µs | 91 µs | -14.2% |
| ❌ | Simulation | decompress_rd[f32, (10000, 0.0)] |
78.5 µs | 91.2 µs | -13.91% |
| ⚡ | Simulation | chunked_varbinview_opt_canonical_into[(1000, 10)] |
206.8 µs | 170.2 µs | +21.46% |
| ⚡ | Simulation | bitwise_not_vortex_buffer_mut[128] |
244.4 ns | 215.3 ns | +13.55% |
| ⚡ | Simulation | chunked_varbinview_into_canonical[(100, 100)] |
307.1 µs | 272.8 µs | +12.59% |
| ⚡ | Simulation | bitwise_not_vortex_buffer_mut[1024] |
304.7 ns | 275.6 ns | +10.58% |
Tip
Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.
Comparing adamg/reduce-cuda-build-noise (06c7e58) with develop (5d55fbc)
| - name: Locate built CUDA kernels | ||
| id: cuda_kernels | ||
| run: | | ||
| ptx_file="$(find target -type f -path '*/out/kernels/*/dynamic_dispatch.ptx' -print -quit)" |
There was a problem hiding this comment.
Does this work with debug & release configs? The gen ptx differs based on the build mode. I guess only one mode is relevant here only anyway.
2 participants
Summary
Closes: #000
Testing