IVF-PQ: Add a (faster) direct conversion fp8->half #1644
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
achirkin
added
3 - Ready for Review
improvement
achirkin
added a commit
to achirkin/raft
that referenced
this pull request
Jul 12, 2023
|
/merge |
achirkin
added a commit
that referenced
this pull request
Jul 19, 2023
#1644 introduced a direct conversion from the custom ivf-pq fp8 type to half and a bug alongside. The exponent bias value wrong by one bit.
rapids-bot bot
pushed a commit
that referenced
this pull request
Jul 19, 2023
#1644 introduced a direct conversion from the custom ivf-pq fp8 type to half and a bug alongside. The exponent bias value wrong by one bit. The result is the loss of precision in some datasets. This hotfix just changes this constant. Authors: - Artem M. Chirkin (https://github.com/achirkin) Approvers: - Tamas Bela Feher (https://github.com/tfeher) - Corey J. Nolet (https://github.com/cjnolet) URL: #1654
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Outdated suggestions cannot be applied.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.
Add a missing direct conversion from the custom ivf-pq
fp_8bittype tohalf. This conversion is used in a tight ALU-bound loop that computes the distances between the query and the encoded cluster vectors.This change improves QPS by 0-20% on the
deep-1Bdataset wheninternal_distance_dtype = CUDA_R_16Fandlut_dtype = CUDA_R_8U. Note, however, on this dataset the <CUDA_R_16F, CUDA_R_8U> combination is often still slower than the <CUDA_R_16F, CUDA_R_16F> combination due to the couple extra ALU instructions.lut_dtype = CUDA_R_8Uis the most beneficial when there's not enough shared memory for the lookup table, which is not the case for the best-performing parameter combinations on thedeep-1Bdataset.