Apply nodata mask in read_geotiff_gpu (#1542)#1547
Merged
brendancol merged 3 commits intomainfrom May 11, 2026
Merged
Conversation
The GPU read backend silently differed from the CPU and dask paths when the file declared a nodata sentinel. open_geotiff(path, gpu=True) returned a DataArray whose attrs had no 'nodata' key and whose pixel data still carried the raw sentinel value. Integer rasters were not promoted to float64, and float rasters kept the sentinel rather than NaN. NaN-aware code that worked on the CPU and dask paths quietly produced wrong results on the GPU path. Add an _apply_nodata_mask_gpu helper that mirrors the CPU masking logic with cupy operations, and call it from both the tiled main path and the stripped fallback inside read_geotiff_gpu. Also set attrs['nodata'] from geo_info.nodata so callers can still see the sentinel value. Two existing tests had codified the old behaviour and needed updates: test_sparse_tile_gpu_round_trip checked dtype=uint16, but the GPU read now promotes to float64 to represent NaN, matching the CPU path. test_*_sentinel_nodata in test_lerc_valid_mask_gpu compared read_geotiff_gpu against read_to_array (low-level), so the test helpers now restore the sentinel on the GPU side before the bit-for-bit comparison so the LERC mask preservation check still holds.
Contributor
There was a problem hiding this comment.
Pull request overview
This PR fixes backend parity for GeoTIFF GPU reads by making read_geotiff_gpu apply the same nodata masking and attrs['nodata'] propagation as the CPU eager and dask paths (fixes #1542), preventing silent incorrect results in NaN-aware downstream code when open_geotiff(..., gpu=True) is used.
Changes:
- Added a CuPy-based
_apply_nodata_mask_gpuhelper and applied it in both the tiled GPU path and the stripped-file fallback insideread_geotiff_gpu. - Updated existing GPU tests that previously pinned the old “raw sentinel” behavior to now expect NaN-masked output and
attrs['nodata']. - Added a new regression test module covering multiple nodata/dtype/layout cases and a 4-backend agreement check (numpy, dask+numpy, cupy, dask+cupy).
Reviewed changes
Copilot reviewed 4 out of 4 changed files in this pull request and generated 2 comments.
| File | Description |
|---|---|
xrspatial/geotiff/__init__.py |
Adds _apply_nodata_mask_gpu and applies nodata masking + attrs['nodata'] propagation in read_geotiff_gpu for stripped and tiled paths. |
xrspatial/geotiff/tests/test_sparse_cog.py |
Updates sparse COG GPU test expectations to match new nodata→NaN masking and dtype promotion behavior. |
xrspatial/geotiff/tests/test_lerc_valid_mask_gpu.py |
Adjusts LERC mask preservation tests to account for GPU nodata masking by restoring sentinels for bit-exact comparisons. |
xrspatial/geotiff/tests/test_gpu_nodata_1542.py |
Adds new regression coverage for nodata parity across GPU/CPU and dask/non-dask paths. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Comment on lines
+89
to
+92
| The GPU reader (``read_geotiff_gpu``) applies the same nodata masking | ||
| that ``open_geotiff`` does (PR #1542), so its output uses NaN where | ||
| the sentinel was. Callers that want a bit-for-bit comparison should | ||
| pass ``raw_gpu=True`` to skip the high-level masking on the GPU side. |
Contributor
Author
There was a problem hiding this comment.
Docstring rewritten to describe what the file actually does. The bit-exact-comparison path is _restore_sentinel on the caller side (line 103 below); the raw_gpu=True mention was aspirational from a draft. 54606a2.
Conflict in xrspatial/geotiff/__init__.py: both #1546 (orientation handling on GPU read) and this PR (#1547 nodata mask) added a new block between the multi-band shape check and the dtype cast. They are independent, so kept both. Order matches the stripped early-return path above: orientation first (so the geo_info transform reflects the post- flip pixel layout), then nodata mask (so the value semantics agree with the CPU read).
- Drop unused os/tempfile imports in test_gpu_nodata_1542 (tmp_path fixture covers the file paths). - Rewrite the _read_cpu_gpu docstring: the raw_gpu=True hook does not exist; callers run the GPU result through _restore_sentinel for the bit-exact comparison.
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
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 #1542.
read_geotiff_gpu(used whenopen_geotiff(path, gpu=True)is called) silently differed from the CPU and dask paths on rasters with a declared nodata sentinel: integer rasters kept the sentinel literal in a uint16 array, float rasters kept the sentinel rather than NaN, andattrs['nodata']was never set. NaN-aware code that worked on the CPU and dask paths quietly produced wrong results on the GPU path.The fix adds an
_apply_nodata_mask_gpuhelper that mirrors the CPU eager masking logic using cupy operations. Both the tiled main path and the stripped fallback insideread_geotiff_gpunow promote integer rasters to float64 with NaN at sentinel positions, replace finite sentinels in float arrays with NaN, and setattrs['nodata']so the original value stays visible.Two existing GPU tests had pinned the old behaviour and needed adjustment.
test_sparse_tile_gpu_round_tripasserteddtype=uint16; it now expectsfloat64and NaN at sparse positions.test_*_sentinel_nodataintest_lerc_valid_mask_gpucomparedread_geotiff_gpuagainstread_to_array(low-level reader); the helpers now restore the sentinel on the GPU side before the bit-for-bit comparison so the LERC mask preservation check still holds.Test plan
test_gpu_nodata_1542.pycovers tiled + stripped paths, float32 and uint16 sentinels, signed int negative sentinel, NaN nodata, no-nodata-attr, and a four-backend agreement check across numpy, dask+numpy, cupy, and dask+cupy.balanced_allocationdask+cupy failure is unrelated to geotiff.