CUDA, float16 and bfloat16 support

The module table below counts entries in the Linux CPU known-failure manifests (cpu_float16.txt, cpu_bfloat16.txt), pinned to Python 3.11 and PyTorch 2.14.0. There are 243 float16 and 256 bfloat16 recorded failures. These are parametrized test cases, not counts of unsupported functions. The remaining work is tracked in #4153. No CI job covers CUDA or MPS half precision; CPU results do not establish support on those backends.

Known failures by module (Linux CPU)

Module	float16	bfloat16	Known failures	Notes
kornia.color	✅	✅	0 / 0	Lab/Luv conversions and rgb255_to_normals compute half inputs in float32 and return the input dtype
kornia.filters	✅	✅	0 / 0	Canny, Gaussian kernels/blur and CPU FFT convolution compute half inputs in float32 and return the input dtype
kornia.enhance	✅	✅	0 / 0	ZCA keeps half-input fitted state in float32; whitening and JPEG return the input dtype after float32 computation
kornia.morphology	✅	✅	0 / 0
kornia.augmentation	⚠️	⚠️	59 / 48	float16: container autocast/runtime errors and accuracy; both dtypes: 3D backward and random-generator accuracy
kornia.geometry.transform	⚠️	⚠️	31 / 31	Rotation matrices, affine/perspective warps and homography-warper accuracy
kornia.geometry.camera	⚠️	⚠️	13 / 23	Projection, distortion round trips and stereo reprojection; includes test-side dtype assertions
kornia.geometry.calibration	⚠️	⚠️	13 / 12	solve_pnp_dlt rejects half inputs; undistort_points accuracy
kornia.geometry.epipolar	⚠️	⚠️	36 / 36	Essential-matrix decomposition/motion and KRt_from_projection lack CPU half kernels; fundamental-matrix accuracy
kornia.geometry.homography	⚠️	⚠️	10 / 16	DLT solver accuracy, including iterated and line-based variants
kornia.geometry.liegroup	✅	✅	0 / 0	So2/Se2 reject bfloat16 rotation construction: PyTorch has no complex bfloat16 dtype; these expected errors/skips are outside the manifests
kornia.geometry.solvers	✅	✅	0 / 0
kornia.geometry.subpix	⚠️	⚠️	6 / 3	SpatialSoftArgmax2d and ConvQuadInterp3d accuracy; bfloat16: render_gaussian2d accuracy
kornia.geometry.conversions	⚠️	⚠️	65 / 65	Angle-axis, quaternion and rotation-matrix accuracy; includes tests/integration/test_conversions.py
kornia.geometry.ransac	⚠️	⚠️	2 / 2	Essential-model fitting raises in epipolar solvers
kornia.geometry (other)	⚠️	⚠️	6 / 13	Box-transform and depth accuracy
kornia.image	⚠️	⚠️	2 / 2	Circle/ellipse polygon-fill comparisons use vertices and reference masks computed in half precision
kornia.losses	✅	⚠️	0 / 2	bfloat16: total-variation accuracy
kornia.feature	✅	✅	0 / 0	Matching uses a manual cdist fallback for half dtypes; LightGlue’s float16 tests are skipped
kornia.metrics	✅	✅	0 / 0
kornia.models	✅	⚠️	0 / 1	bfloat16: RT-DETR RepVGG fusion accuracy
contrib, core, io, onnx, sensors, tracking, utils	✅	⚠️	0 / 2	bfloat16: histogram matching and _torch_svd_cast accuracy

✅ No entries in the CPU known-failure manifest ⚠️ Has recorded CPU failures. Neither symbol promises complete dtype support: skips, expected exceptions and source-level xfails are outside these counts. Some operators compute internally in float32 or float64 and return the input dtype.

Observed CPU CI outcomes:

The scheduled run at 93f1376e, 2026-10-21 used Linux x86_64, Python 3.11 and PyTorch 2.14.0, without slow or compile tests:

Run	Passed	Failed	Skipped	XFAIL (all)	XPASS	Strict manifest entries
CPU float16	23905	0	5132	284	4	250
CPU bfloat16	23726	0	5286	309	4	274

Each job also deselected 3 ONNX conformance tests. XFAIL totals include both the strict manifest entries and source-level expected failures. The 4 XPASSes in each successful job are non-strict expected failures that passed; a passing manifest entry makes verification fail, as do a changed exception, a skipped entry or a stale node ID. A green job therefore means the recorded baseline reproduced, even though known failures and skips remain. These totals describe the linked revision; the module table describes the checked-in manifests.

Verify the complete CPU profiles in the pinned environment with:

KORNIA_TEST_OPTIMIZER= pixi run test-module tests/ --verify-known-failures --known-failure-profile=cpu-float16
KORNIA_TEST_OPTIMIZER= pixi run test-module tests/ --verify-known-failures --known-failure-profile=cpu-bfloat16

pixi run test-half is an unseeded CPU sweep, without the strict profile baseline. See TESTING.md for focused checks and baseline maintenance.

CUDA support and measurement limits:

Kornia accepts CUDA tensors for GPU acceleration, subject to each operator’s dtype and backend limitations. There is no blocking CUDA CI job or checked-in CUDA known-failure manifest, so the CPU xfail counts above do not provide a current CUDA pass rate. The previous local measurements are retained here as historical evidence only:

Historical run (f8449854, 2026-09-23)	Passed	Failed	Skipped
CUDA float32	12193	22	3957
CUDA float16	11738	451	3983
CUDA bfloat16	11736	500	3936

These used an RTX 4090, Python 3.11 and PyTorch 2.14.0+cu130. Source-marked xfails were excluded. Half tests ran in-process with KORNIA_TEST_IN_SUBPROCESS=1, followed by isolated retries of failures; two homography tests were deselected. They are not a measurement of the current branch. Those two tests (TestFindHomographyDLT::test_nocrash and test_nocrash_lu) now pass locally in all three CUDA dtypes at 649b3f86d; a fresh full CUDA sweep is needed before quoting new totals.

For CUDA half tests, use --isolate-half-precision so each test gets a fresh process. They are skipped by default without it; selecting a half dtype alone does not exercise CUDA half support:

pixi run test-module tests/color/ --device=cuda --dtype=float16,bfloat16 --isolate-half-precision