perf: Reduce per-call overhead in representation converters#94
Draft
mbrobbel wants to merge 1 commit intoNVIDIA:mainfrom
Draft
perf: Reduce per-call overhead in representation converters#94mbrobbel wants to merge 1 commit intoNVIDIA:mainfrom
mbrobbel wants to merge 1 commit intoNVIDIA:mainfrom
Conversation
- Use shared_mutex in converter registry so concurrent convert/has_converter calls no longer serialize on an exclusive lock - Remove redundant stream.synchronize() before cudf::pack in cross-GPU converter - Eliminate unnecessary metadata vector copy in packed HOST→GPU converter by reading source metadata directly (source outlives the unpack call) - Remove redundant stream.synchronize() before cudf::unpack in packed HOST→GPU converter — stream ordering between H2D copies and table construction is guaranteed since both use the same stream - Skip cudaGetDevice/cudaSetDevice round-trip in single-GPU scenarios for both packed and fast HOST→GPU converters - Move BatchCopyAccumulator definition earlier in the file for reuse Benchmarked improvement on packed HOST→GPU path (back-to-back A/B, 5 reps): 256K/1T: 0.056ms → 0.044ms (21% faster) 1M/1T: 0.095ms → 0.083ms (13% faster) 64K/4T: 0.092ms → 0.077ms (16% faster) 4M/4T: 0.726ms → 0.682ms (6% faster) Large transfers (512M+) are PCIe bandwidth-limited and unchanged. Created by Claude. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
|
Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually. Contributors can view more details about this message here. |
Member
Author
|
/ok to test b42aba2 |
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.
Summary
shared_mutexin converter registry so concurrentconvert/has_convertercalls no longer serialize on an exclusive lockstream.synchronize()calls in cross-GPU and packed HOST→GPU converters where stream ordering already guarantees correctnesscudaGetDevice/cudaSetDeviceround-trip in single-GPU scenarios for both packed and fast HOST→GPU convertersBenchmark results
Packed HOST→GPU path (back-to-back A/B, 5 repetitions):
All other paths (GpuToHost packed, GpuToHostFast, HostFastToGpu) verified with reversed-order A/B tests showing no regressions.
Test plan
ctest)🤖 Created by Claude
Co-Authored-By: Claude Opus 4.6 (1M context) noreply@anthropic.com