Use GPU registers more effectively #277
Open
Conversation
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
Release builds still crash
Still crashes in release builds
Yields 4x improvement
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Outdated suggestions cannot be applied.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.
All of our algorithms on GPU extensively leverage shared memory and warp-level synchronization. That might be suboptimal for very small input sizes. There, we should keep everything in the registers.
So I'm suggesting a new set of kernels processing the DP matrix row-by-row, but keep only one row and one scalar in GPU registers. Moreover, it process the matrix of
uint8_tcells in slices of 4 continuous entries forminguint32_t-s in each row. That minimizes the number of loads & stores, conversions between 8-bit and 32-bit representations - assuming GPUs don't have much custom logic for 8-bit times and upcast practically every time.