Add tensor row inputs for RowFn - #9776
Conversation
Merging this PR will regress 2 benchmarks
|
| Mode | Benchmark | BASE |
HEAD |
Efficiency | |
|---|---|---|---|---|---|
| ❌ | Simulation | random_i16[0.95] |
77.4 µs | 95.9 µs | -19.22% |
| ❌ | Simulation | decompress[u64, (4000, 1024)] |
70.4 µs | 85.7 µs | -17.82% |
| ⚡ | WallTime | arrow_checked_add_u32_neon[16384] |
20.4 µs | 12.2 µs | +67.46% |
| ⚡ | Simulation | random_i8[0.5] |
91.4 µs | 68 µs | +34.4% |
| ⚡ | Simulation | column_x_extension_constant[256] |
540.2 µs | 425.3 µs | +27% |
| ⚡ | Simulation | allocate_drop_arrow[0] |
456.9 ns | 402.7 ns | +13.45% |
| ⚡ | Simulation | column_x_extension_constant[32] |
192.1 µs | 169.5 µs | +13.36% |
| ⚡ | Simulation | allocate_drop_bytes[0] |
575.7 ns | 521.6 ns | +10.39% |
Tip
Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.
Comparing ct/row-fn-tensor-rows (6ca28c3) with develop (5c8c397)2
Footnotes
-
218 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. ↩
-
No successful run was found on
develop(803a3dc) during the generation of this report, so 5c8c397 was used instead as the comparison base. There might be some changes unrelated to this pull request in this report. ↩
Signed-off-by: Connor Tsui <connor.tsui20@gmail.com>
5eba5b2 to
6ca28c3
Compare
1 participant
Summary
Adds the tensor row types used by RowFn scalar functions without migrating a function in the same layer.
Changes
TensorRowdecodes ordinary tensor columns, literal constants, and extension-backed constants into flat typed rows. It is dense-safe because tensor float kernels can consume the stored payload of null rows while RowFn owns output validity.