Add grid-stride + ROCm cap to linearize_cache_indices kernels - #6098
Open
q10 wants to merge 2 commits into
Open
Conversation
Contributor
|
@q10 has exported this pull request. If you are a Meta employee, you can view the originating Diff in D113351691. |
Summary: lxu_cache_lookup_kernel (the sibling of the already-fixed direct_mapped_lxu_cache_lookup_kernel) launched grid = div_round_up(N, kMaxThreads) with block = dim3(kWarpSize, kMaxThreads / kWarpSize) (N = linear_cache_indices.numel()) and did not grid-stride, so on ROCm total threads ~= N exceeds the HIP 2^32 threads-per-launch limit for large unique-index sets. Cap the launch with utils::cuda::cap_grid_dim_x_from_workload(N, kMaxThreads, stream) (mirroring the direct_mapped sibling) and add a ROCm grid-stride loop over n0 (stride = gridDim.x * blockDim.y * blockDim.x). No-op on CUDA. Reviewed By: cthi Differential Revision: D113351696
Summary: linearize_cache_indices_kernel and linearize_cache_indices_from_row_idx_kernel launch grid = div_round_up(num_indices, kMaxThreads) with block = kMaxThreads and process one index per thread with an early-return guard, so on ROCm total threads ~= num_indices can exceed the HIP 2^32 threads-per-launch limit for very large index counts. Cap both launches with utils::cuda::cap_grid_dim_x_from_workload(num_indices, kMaxThreads, stream) and add a ROCm grid-stride loop over index to both kernels. No-op on CUDA. Reviewed By: cthi Differential Revision: D113351691
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary:
linearize_cache_indices_kernel and linearize_cache_indices_from_row_idx_kernel
launch grid = div_round_up(num_indices, kMaxThreads) with block = kMaxThreads
and process one index per thread with an early-return guard, so on ROCm total
threads ~= num_indices can exceed the HIP 2^32 threads-per-launch limit for very
large index counts.
Cap both launches with utils::cuda::cap_grid_dim_x_from_workload(num_indices,
kMaxThreads, stream) and add a ROCm grid-stride loop over index to both kernels.
No-op on CUDA.
Reviewed By: cthi
Differential Revision: D113351691