Repository navigation
Fold aggregates with masked and lane-parallel kernels - #10947
robert3005 wants to merge 7 commits into
8 benchmarks regressed
3 warnings
⚠️ Walltime results may be unreliableSome benchmarks used the Walltime instrument outside CodSpeed Macro Runners, which can make the results inconsistent.
⚠️ 1 benchmark is dominated by system callsSystem calls cannot be simulated consistently and are left out of the measure. The reported value understates the real cost. Switch to the Walltime instrument to measure them.
⚠️ Base and head ran on different runner typesSome benchmarks with significant performance changes were compared across different runner types, so the changes might not come from the code. Run base and head on the same runner type, or use CodSpeed Macro Runners for more stable results.
⚡ 35 improved benchmarks
❌ 8 regressed benchmarks
✅ 2263 untouched benchmarks
🆕 61 new benchmarks
Warning
Please fix the performance issues or acknowledge them on CodSpeed.
Performance Changes
| Mode | Benchmark | BASE |
HEAD |
Efficiency | |
|---|---|---|---|---|---|
| ❌ | CPU Simulation | filter_powerlaw_by_random[10000] |
142.2 ns | 247.5 ns | -42.55% |
| ❌ | CPU Simulation | sum[msp_i64_p18_canonical] |
129 µs | 217.9 µs | -40.8% |
| ❌ | CPU Simulation | sum[msp_i64_p38] |
143.2 µs | 232.4 µs | -38.37% |
| ❌ | CPU Simulation | min_max_decimal_i256[ScatteredNulls] |
672.5 µs | 1,001.7 µs | -32.86% |
| ❌ | CPU Simulation | missing_stats[16] |
66.8 µs | 85 µs | -21.45% |
| ❌ | CPU Simulation | compare_bool_nullable |
48.1 µs | 55.2 µs | -12.86% |
| ❌ | CPU Simulation | take_map[(0.005, 0.5)] |
87.5 µs | 98 µs | -10.75% |
| ❌ | CPU Simulation | take_map[(0.005, 1.0)] |
147 µs | 163.6 µs | -10.15% |
| ⚡ | CPU Simulation | sum_i32_nulls_scattered |
690.4 µs | 99.4 µs | ×6.9 |
| ⚡ | CPU Simulation | sum_v2_i32_nulls_scattered |
733 µs | 124.8 µs | ×5.9 |
| ⚡ | CPU Simulation | max_i32_nulls_scattered |
766.6 µs | 138.7 µs | ×5.5 |
| ⚡ | CPU Simulation | integer_stats_all_nullable |
2,432.7 µs | 517.6 µs | ×4.7 |
| ⚡ | CPU Simulation | cast_i32_to_u32[8192] |
389.2 µs | 122.5 µs | ×3.2 |
| ⚡ | CPU Simulation | integer_stats_nullable |
1,789.6 µs | 563.9 µs | ×3.2 |
| ⚡ | CPU Simulation | min_max[msp_i32_p18_nullable] |
226.3 µs | 115.3 µs | +96.36% |
| ⚡ | CPU Simulation | sum[msp_i32_p18_nullable] |
156.5 µs | 87.5 µs | +79% |
| ⚡ | CPU Simulation | max_f64 |
313.6 µs | 189.5 µs | +65.48% |
| ⚡ | Walltime | threshold_neon[64] |
4.4 µs | 2.7 µs | +64.15% |
| ⚡ | Walltime | filter_neon[64] |
2.7 µs | 1.8 µs | +51.02% |
| ⚡ | CPU Simulation | sum_f64 |
204.8 µs | 135.7 µs | +50.92% |
| ... | ... | ... | ... | ... | ... |
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/fold-masked (1df80b0) with develop (d4d9234)
