[cudapoa] improving cudaPOA performance #552
Merged
Conversation
This file contains 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
…ith 1 predecessor and band_start > 4
…ecessor - rev3" rev3 change did not optimize much and even in some cases could a bit slow down, therefore reverted. This reverts commit 833ac3ad
…aGenomicsAnalysis into cudapoa_optimization
…uce long_score_board-rev 4
… matrix allocation
…score matrix allocation
…s where POA groups have the same number of reads- rev 4b
…op iterations- rev 5
… adding __align__ to Seq4T fails though
…ptive, static and full alignments are separate, reg count for static down to 71 from 83; rev 7
… by changing banded_score_matrix_size from int64_t to float, reduced 1 register! :) (from 79 to 78)
…ngle-thread work in updating vertical scores, removed set_and_get_first_column_score().
… loop. This reduced register count from 78 to 75.
…umn == 0 to -1, to get rid of it is a better solution! also replaced get_score() with get_score_adaptive() in nw_adaptive, better solution is to uify similar kernels
…onvince compiler finding a way to minimize register usage down to 72. It worked without any register spills. rev 8
…some cases (random) it can result in misaligned address error
…n while loop in cudapoa-full alignment - rev9
…aGenomicsAnalysis into cudapoa_optimization_v2
…aGenomicsAnalysis into cudapoa_optimization_v2
r-mafi
added
enhancement
…is helps compiler perform more aggressive optimizations. When using unsigned type compiler can't perform such optimizations due to overflow check semantics.
…nce_to_head, although the latter did not waste any memory.
tijyojwad
suggested changes
Sep 10, 2020
… ScoreT to int32_t, reduced reg count down to 64
… in nw-full to be consisten with the changes in banded versions, although register count remained the same, 64
…(to see the actual number of registers, launchbounds() should be commented)
…NELS_MAX_THREADS_PER_BLOCK to 1024, to enforce 64 register count
…aGenomicsAnalysis into cudapoa_optimization_v2
tijyojwad
approved these changes
Sep 14, 2020
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
a round of optimization consisting of reducing back register usage, hiding global memory access where applicable and reducing NW inner while loop iterations in an effort to improve compute-time and SOL metrics.
in cudapoa binary API, added a new option
-sto allow managing allocated memory for adaptive score matrix