[CUB] Reduce BlockAdjacentDifference shared memory usage by 50% #6901
+18
−19
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
BlockAdjacentDifference::_TempStorage previously allocated two separate arrays (first_items and last_items), but SubtractLeft* operations only use last_items while SubtractRight* operations only use first_items. Since these operations are never used together, a single array suffices. This change halves the shared memory footprint, which can improve occupancy and performance for kernels using BlockAdjacentDifference. For a 256-thread block with 4-byte elements: - Before: 256 × 2 × 4 = 2,048 bytes - After: 256 × 1 × 4 = 1,024 bytes closes NVIDIA#3711
Contributor
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Outdated suggestions cannot be applied.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.
Description
closes #3711
BlockAdjacentDifference::_TempStoragepreviously allocated two separate arrays (first_itemsandlast_items), each of sizeBLOCK_THREADS. However, after analyzing all member functions:first_itemslast_itemsSubtractLeftSubtractLeft(with predecessor)SubtractLeftPartialTileSubtractLeftPartialTile(with predecessor)SubtractRightSubtractRight(with successor)SubtractRightPartialTileThe arrays are never used together. This PR unifies them into a single
itemsarray, reducing shared memory usage by 50%.Impact
For a 256-thread block with 4-byte elements:
This reduction in shared memory can improve occupancy and performance for kernels using
BlockAdjacentDifference.