SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2605.00837 · 2026

Fast Log-Domain Sinkhorn Optimal Transport with Warp-Level GPU Reductions

Hao Xiao

arXiv · PDF · Open in the Atlas

Code that ran

We lifted 12 functions out of this paper's own repositories and ran 12 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.

RepositoryRoleRan
xiao98/Fast-Sinkhorn-CUDA canonical 12 of 12
FunctionStatusWhere it lives
compute_ot_matching Ran xiao98/Fast-Sinkhorn-CUDA/experiments/applications/point_cloud_matching.py
code served (permissive licence) · get_code("f785f92416c75d1d")
compute_transport_cost Ran xiao98/Fast-Sinkhorn-CUDA/experiments/baselines/bench_pytorch_sinkhorn.py
code served (permissive licence) · get_code("59b5efe0d457c5fa")
generate_bunny_like_points Ran xiao98/Fast-Sinkhorn-CUDA/experiments/applications/point_cloud_matching.py
code served (permissive licence) · get_code("f9768a4f1c357138")
generate_synthetic_image Ran xiao98/Fast-Sinkhorn-CUDA/experiments/applications/color_transfer.py
code served (permissive licence) · get_code("95368d0df733a6f4")
load_data Ran xiao98/Fast-Sinkhorn-CUDA/experiments/baselines/plot_baselines.py
code served (permissive licence) · get_code("6829ee1542b9660c")
make_distributions Ran xiao98/Fast-Sinkhorn-CUDA/experiments/baselines/bench_pytorch_sinkhorn.py
code served (permissive licence) · get_code("ac42230eccda1f28")
make_distributions Ran xiao98/Fast-Sinkhorn-CUDA/experiments/baselines/bench_geomloss.py
code served (permissive licence) · get_code("c1fab6a120849fc3")
ot_color_transfer Ran xiao98/Fast-Sinkhorn-CUDA/experiments/applications/color_transfer.py
code served (permissive licence) · get_code("cc03a4060e69a92d")
rotation_matrix Ran xiao98/Fast-Sinkhorn-CUDA/experiments/applications/point_cloud_matching.py
code served (permissive licence) · get_code("9b6a364baf249a7b")
run_sinkhorn Ran xiao98/Fast-Sinkhorn-CUDA/experiments/baselines/bench_pot.py
code served (permissive licence) · get_code("ee2f8afaec88b236")
sample_pixels Ran xiao98/Fast-Sinkhorn-CUDA/experiments/applications/color_transfer.py
code served (permissive licence) · get_code("ef6edc9926cd2d12")
sinkhorn_log_pytorch Ran xiao98/Fast-Sinkhorn-CUDA/experiments/baselines/bench_pytorch_sinkhorn.py
code served (permissive licence) · get_code("f44585a13a67239f")

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Entropic regularized optimal transport (OT) via the Sinkhorn algorithm has become a fundamental tool in machine learning, yet existing implementations either suffer from numerical instability for small regularization parameters or incur significant overhead from deep learning frameworks. We present FASTSINKHORN, a lightweight, native CUDA implementation of the log-domain Sinkhorn algorithm that combines warp-level shuffle reductions with shared-memory tiling to achieve high GPU utilization without sacrificing numerical stability. Our solver operates entirely in the log-domain, enabling robust computation for regularization parameters as small as ε = 10 -4 where standard-domain methods fail. On dense OT problems with n = m = 8192, our implementation achieves 12× speedup over the widely-used POT library and 5.9× speedup over GPU-accelerated Py-Torch baselines, while consuming only 256 MB of GPU memory. We validate our solver on image color transfer, 3D point cloud matching, and convergence analysis, demonstrating that native CUDA kernels with careful numerical treatment provide a practical and efficient foundation for large-scale optimal transport computation.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2605.00837")
get_code_for_paper("2605.00837")
have("2605.00837")

Connect an agent — have() is free.