Faster ptt

[36000]

No description provided.

[Copilot]

Pull request overview

This PR optimizes the PTT (Parallel Transport Tractography) direction getter by switching from global device memory to CUDA texture memory for ODF data, which enables hardware-accelerated trilinear interpolation. It also adds caching for FA and CSD ODF computations, and introduces a seed for reproducible seed generation.

Changes:

• Replaced global memory reads with CUDA 3D texture memory lookups for PTT, adding texture allocation/deallocation in the GPU tracker and a new ptt_init.cu kernel for texture-based initial direction finding.
• Added a DATA_T template parameter to genStreamlinesMergeProb_k and tracker_d to support both raw pointer and texture object data access paths.
• Added FA and CSD ODF disk caching (--cache-dir) and a --seed-seed argument for reproducible seed generation in the run script.

Reviewed changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated 7 comments.

Show a summary per file

File	Description
run_gpu_streamlines.py	Added --cache-dir and --seed-seed CLI args; FA/ODF caching logic
cuslines/cuda_python/cu_tractography.py	Texture memory allocation/deallocation for PTT; removed unused MEGABYTE import
cuslines/cuda_python/cu_direction_getters.py	Updated kernel template names with DATA_T; handle texture objects in launch calls
cuslines/cuda_c/ptt_init.cu	New kernel for PTT initial direction finding using texture memory
cuslines/cuda_c/ptt.cu	Changed pmf parameter type to cudaTextureObject_t; replaced trilinear interp with tex3D
cuslines/cuda_c/generate_streamlines_cuda.cu	Added DATA_T template parameter to tracker_d and genStreamlinesMergeProb_k; included ptt_init.cu
cuslines/cuda_python/cu_propagate_seeds.py	Removed a TODO comment

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

This PR aims to speed up PTT (Parallel Transport Tractography) by moving more parameters to compile-time CUDA macros and introducing CUDA texture-memory paths, while also modernizing Python packaging/lint tooling and adding caching to the example runner.

Changes:

• Refactor CUDA kernels + Python launch code to use compile-time macros and texture-backed data/metric maps (notably for PTT).
• Add optional FA/CSD-ODF caching and deterministic seeding controls to run_gpu_streamlines.py.
• Migrate packaging metadata into pyproject.toml and add Ruff + pre-commit configuration.

Reviewed changes

Copilot reviewed 20 out of 20 changed files in this pull request and generated 7 comments.

Show a summary per file

File	Description
setup.py	Removed legacy setuptools build hook (previously used for build-time codegen).
pyproject.toml	Adds Ruff config, moves URLs/package-data into pyproject, adds scipy + dev extras.
.pre-commit-config.yaml	Adds Ruff + codespell hooks.
run_gpu_streamlines.py	Reformats imports/args; adds --cache-dir and caching for FA/CSD ODF; adds --seed-seed.
cuslines/cuda_python/cutils.py	Adds allocate_texture() helper and runtime imports for texture creation/copies.
cuslines/cuda_python/cu_tractography.py	Allocates dataf/metric_map via textures (PTT + stopping map), compiles kernels via direction getter.
cuslines/cuda_python/cu_direction_getters.py	Introduces macro-driven compilation and adds PTT direction getter implementation details.
cuslines/cuda_python/cu_propagate_seeds.py	Minor import/format cleanup.
cuslines/cuda_python/_globals.py	Updates shared constants module content/comments.
cuslines/cuda_python/init.py	Reorders exports/imports.
cuslines/init.py	Import ordering tweaks for backend-specific classes.
cuslines/boot_utils.py	Formatting-only refactor.
cuslines/cuda_c/* (multiple)	Kernel signature refactors, macro usage, PTT init kernel added, texture-based stopping checks, and related updates.
CLAUDE.md	Updates repository documentation around constants/codegen conventions.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.