Repository navigation
Initialise ArrayParts directly with provided slots - #10072
Conversation
ee1ddbb to
e4baedb
Compare
Merging this PR will regress 5 benchmarks
|
| Mode | Benchmark | BASE |
HEAD |
Efficiency | |
|---|---|---|---|---|---|
| ❌ | Simulation | new_raw_prim_test_between[i32, 2048] |
67.3 µs | 94 µs | -28.43% |
| ❌ | WallTime | mul_u64_nonnull_neon |
29.1 µs | 40.6 µs | -28.25% |
| ❌ | Simulation | new_bp_prim_test_between[i64, 2048] |
78.5 µs | 97.1 µs | -19.13% |
| ❌ | WallTime | multiply_shapes_neon[(32768, PerRowPerRow)] |
32.9 µs | 38.8 µs | -15.34% |
| ❌ | WallTime | mul_i64_nonnull_neon |
33 µs | 38.9 µs | -15.22% |
| ⚡ | Simulation | compare[1] |
135.5 µs | 114.2 µs | +18.62% |
| ⚡ | Simulation | new_alp_prim_test_between[f32, 2048] |
102.2 µs | 86.4 µs | +18.22% |
| ⚡ | Simulation | filter_random_by_power_law[1000] |
27.1 µs | 24.2 µs | +12.34% |
| ⚡ | Simulation | filter_powerlaw_by_power_law[1000] |
27 µs | 24.1 µs | +12.15% |
| ⚡ | Simulation | filter_random_by_random[1000] |
28 µs | 25 µs | +11.91% |
| ⚡ | Simulation | filter_random_by_correlated_runs[1000] |
28.2 µs | 25.2 µs | +11.83% |
| ⚡ | Simulation | filter_random_by_mostly_true[1000] |
28.3 µs | 25.3 µs | +11.78% |
| ⚡ | Simulation | filter_all_true[100000] |
14.6 µs | 13.1 µs | +11.75% |
| ⚡ | Simulation | filter_all_true[10000] |
14.6 µs | 13.1 µs | +11.74% |
| ⚡ | Simulation | filter_all_true[250000] |
14.6 µs | 13.1 µs | +11.74% |
| ⚡ | Simulation | filter_powerlaw_by_random[1000] |
28 µs | 25.1 µs | +11.66% |
| ⚡ | Simulation | filter_powerlaw_by_mostly_true[1000] |
28.2 µs | 25.2 µs | +11.6% |
| ⚡ | Simulation | filter_random_by_mostly_false[1000] |
28.9 µs | 25.9 µs | +11.51% |
| ⚡ | Simulation | filter_powerlaw_by_mostly_false[1000] |
28.7 µs | 25.8 µs | +11.34% |
| ⚡ | Simulation | filter_powerlaw_by_mostly_true[10000] |
31.3 µs | 28.1 µs | +11.18% |
| ... | ... | ... | ... | ... | ... |
ℹ️ Only the first 20 benchmarks are displayed. Go to the app to view all benchmarks.
Tip
Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.
Comparing rk/nodrops (d067deb) with develop (5079a20)
Footnotes
-
518 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩
Signed-off-by: Robert Kruszewski <github@robertk.io>
`ArrayParts::new` now takes the slots directly (vortex-data#10072), and blocked bit-packing (vortex-data#10203) replaced `BitPacked::bit_width` with `bit_widths`. The low-bits width check still applies to a globally packed child, which is what the encoder writes; a blocked child keeps its widths in an array rather than metadata, so it is accepted like any other integer child. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: rapour <reza.apour@gmail.com>
I think this is something that's different between Vec and SmallVec since
smallvec is stack allocated. Confirmed that SmallVec creation and drop survive
in optimised builds.