Repository navigation
opencl: reduce moments in local memory without float atomics - #3726
Draft
melonakos wants to merge 2 commits into
Draft
melonakos wants to merge 2 commits into
melonakos wants to merge 2 commits into
Conversation
Copy link
Copy Markdown
Member
Sorry, something went wrong.
On Intel GPUs af::moments returned 0 for every moment slot after the first whenever more than one moment was requested, so AF_MOMENT_FIRST_ORDER came back as (M00, M01, 0, 0). The kernel accumulated into local memory through a hand-rolled float compare-and-swap loop, which Intel's GPU compiler mishandles for the second and later slots while the same code is fine on the Intel CPU OpenCL device. Each work-item now accumulates its rows privately and the group reduces with a tree in local memory, which needs no local atomics and is cheaper. The cross-group global atomic add is unchanged.
…EADS to the launch
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
On Intel GPUs af::moments returned 0 for every moment slot after the first whenever more than one moment was requested, so AF_MOMENT_FIRST_ORDER came back as (M00, M01, 0, 0) and six of the seven moments tests failed on an Arc B580 (driver 32.0.101.8991) and a UHD 770 (32.0.101.7088); the Intel CPU OpenCL device was correct. The kernel accumulated into local memory through a hand-rolled float compare-and-swap loop, and on those two devices that loop returns zeros for the second and later slots; I have no reduced repro or upstream report for why. Each work-item now accumulates its rows privately and the group reduces with a tree in local memory, so the kernel needs no local atomics; the cross-group global atomic add is unchanged. The moments suite passes on both Intel GPUs and the CPU device with this. Not run on AMD or NVIDIA OpenCL.
sift_nonfree.cl carries the same local float compare-and-swap and is left alone here; it needs its own issue.