Fix NaN handling in focal_stats CUDA kernels (#1092)#1093
Merged
Conversation
All focal_stats CUDA kernels (_focal_mean_cuda, _focal_sum_cuda, _focal_std_cuda, _focal_var_cuda, _focal_range_cuda, _focal_min_cuda, _focal_max_cuda) now skip NaN neighbors with `if v != v: continue`, matching the numpy path which uses np.nanmean/nansum/nanstd/etc. Previously, NaN propagated through arithmetic, giving different results on GPU vs CPU when input contained NaN.
- test_focal_stats_nan_handling_1092: verifies all 7 stats (mean, sum, min, max, std, var, range) skip NaN neighbors across all 4 backends. - test_focal_stats_all_nan_window_1092: all-NaN window gives NaN for mean/min/max and 0 for sum (matching numpy nansum behavior). - Fixed sum kernel to return 0 (not NaN) for all-NaN windows, matching numpy nansum semantics.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
Fixes #1092. The focal_stats CUDA kernels (
_focal_mean_cuda,_focal_sum_cuda,_focal_std_cuda,_focal_var_cuda,_focal_range_cuda,_focal_min_cuda,_focal_max_cuda) propagated NaN through arithmetic instead of skipping it. The numpy path usesnp.nanmean/nansum/nanstd/etc which skip NaN, so the same data gave different results on GPU vs CPU.The fix adds
if v != v: continueNaN checks to each CUDA kernel, matching the numpy nan-safe behavior. The non-focalmean()function already had these checks in its_mean_gpukernel.For min/max, there was also a subtler bug: if the first neighbor encountered was NaN,
m = NaNandfound = True, after which all subsequentv < m/v > mcomparisons returned False (NaN comparisons always return False), so NaN got stuck as the result.Test plan
test_focal_stats_nan_handling_1092: tests all 7 stats (mean, sum, min, max, std, var, range) with NaN in input, across all 4 backends (numpy, cupy, dask+numpy, dask+cupy)test_focal_stats_all_nan_window_1092: all-NaN window produces NaN for mean/min/max, 0 for sum (matching numpy nansum)test_focal.pysuite: 122 passed