transport.samples.kernels
module torchmatch.transport.samples.kernelsTriton kernels for streaming optimal transport.
Members
module
- apply_fused_sqeuclidFused Schur complement matvec kernel for HVP CG acceleration.
- apply_rawraw-form apply kernels (mat5 for HVP, deprecated vec/mat wrappers).
- apply_shiftedshifted-form apply kernels (shifted potentials, s_I cancellation).
- apply_sqeuclidBackward-compatible re-exports for apply kernels.
- c_transform_sqeuclidC-Transform (hard argmin) kernel for squared Euclidean cost.
- cg_denseDense CG solver with cached transport plan matrix.
- cg_python_batchedPython-level batched CG for HVP acceleration.
- grad_sqeuclid
- streaming_sqeuclidSinkhorn OT with streaming softmax and shifted potentials.
Source: torchmatch/transport/samples/kernels/__init__.py:1