feat[gpu]: arrow device array stream support#8483
Draft
0ax1 wants to merge 23 commits into
Draft
Conversation
Signed-off-by: Alexander Droste <alexander.droste@protonmail.com>
Signed-off-by: Alexander Droste <alexander.droste@protonmail.com>
Signed-off-by: Alexander Droste <alexander.droste@protonmail.com>
Signed-off-by: Alexander Droste <alexander.droste@protonmail.com>
Signed-off-by: Alexander Droste <alexander.droste@protonmail.com>
Signed-off-by: Alexander Droste <alexander.droste@protonmail.com>
Signed-off-by: Alexander Droste <alexander.droste@protonmail.com>
This reverts commit 52952d2. Signed-off-by: Alexander Droste <alexander.droste@protonmail.com>
Signed-off-by: Alexander Droste <alexander.droste@protonmail.com>
Merging this PR will not alter performance
|
| Mode | Benchmark | BASE |
HEAD |
Efficiency | |
|---|---|---|---|---|---|
| ❌ | Simulation | take_10k_random |
197.9 µs | 255.8 µs | -22.61% |
| ❌ | Simulation | take_10k_contiguous |
218.5 µs | 276.3 µs | -20.93% |
| ❌ | Simulation | patched_take_10k_contiguous_patches |
232.2 µs | 290.9 µs | -20.18% |
| ❌ | Simulation | patched_take_10k_random |
244.2 µs | 303 µs | -19.4% |
| ❌ | Simulation | chunked_varbinview_opt_canonical_into[(1000, 10)] |
178 µs | 213.8 µs | -16.78% |
| ❌ | Simulation | chunked_varbinview_opt_into_canonical[(1000, 10)] |
193.4 µs | 229.6 µs | -15.78% |
| ⚡ | Simulation | chunked_bool_canonical_into[(1000, 10)] |
34.9 µs | 20.3 µs | +71.95% |
| ⚡ | Simulation | chunked_varbinview_canonical_into[(1000, 10)] |
198.5 µs | 162.2 µs | +22.35% |
| ⚡ | Simulation | chunked_varbinview_into_canonical[(1000, 10)] |
214.4 µs | 178 µs | +20.48% |
| ⚡ | WallTime | cuda/bitpacked_u8/unpack/3bw[100M] |
352.6 µs | 298.7 µs | +18.04% |
| ⚡ | Simulation | chunked_varbinview_canonical_into[(100, 100)] |
308.7 µs | 273.1 µs | +13.02% |
| ⚡ | Simulation | chunked_varbinview_into_canonical[(100, 100)] |
367.7 µs | 332.8 µs | +10.48% |
| ⚡ | Simulation | eq_i64_constant |
322.6 µs | 292.8 µs | +10.21% |
Tip
Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.
Comparing ad/arrow-device-array-stream (b89a5c9) with develop (d020924)
Signed-off-by: Alexander Droste <alexander.droste@protonmail.com>
Signed-off-by: Alexander Droste <alexander.droste@protonmail.com>
Signed-off-by: Alexander Droste <alexander.droste@protonmail.com>
Signed-off-by: Alexander Droste <alexander.droste@protonmail.com>
0ax1
force-pushed
the
ad/arrow-device-array-stream
branch
from
3ae0d00 to
07926a0
Compare
Signed-off-by: Alexander Droste <alexander.droste@protonmail.com>
Signed-off-by: Alexander Droste <alexander.droste@protonmail.com>
Signed-off-by: Alexander Droste <alexander.droste@protonmail.com>
The Arrow C device array stream export drove the Vortex stream on a private CurrentThreadRuntime, but a partition scan spawns its decode work onto the session's runtime (vortex-ffi's RUNTIME). Nothing ever drove that runtime during streaming, so the first get_next on a real partition deadlocked waiting on tasks that never ran. The existing tests only exercise an inert in-memory stream, so they never hit it. Thread the session's runtime through export_device_array_stream and drive the stream and per-array exports on it, removing the private runtime and worker pool. Expose vortex_ffi::runtime() so layered FFI crates can pass the same runtime the partition's scan spawns onto. Signed-off-by: Alexander Droste <alexander.droste@protonmail.com>
The device stream derives its schema from the first array and rejects any later array whose Arrow schema differs, which is required by the Arrow C stream contract but means a stream whose chunks vary their encoding (a dictionary-encoded chunk among plain chunks) fails mid-stream. Document this on the trait, note that an empty stream reports a dtype-derived schema that can differ from a non-empty run, and sharpen the mismatch error to name the cause. Signed-off-by: Alexander Droste <alexander.droste@protonmail.com>
Add a shared ArrowDeviceArray::empty() constructor and build the end-of-stream marker from it, replacing the hand-rolled struct literal. The stream tests now call the module-level release_schema/release_device_array helpers instead of redefining byte-for-byte copies, and drop the duplicate empty_device_array placeholder in favor of ArrowDeviceArray::empty(). Signed-off-by: Alexander Droste <alexander.droste@protonmail.com>
Several doc and line comments added for the Arrow device array stream exceeded the 100-column limit. Wrap them; no behavior change. Signed-off-by: Alexander Droste <alexander.droste@protonmail.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.