[PR #400] [CLOSED] feat(native): add standalone CUDA core #420

New issue

Closed

opened 2026-10-26 21:17:34 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-26 21:17:34 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/deepseek-ai/DeepGEMM/pull/400
Author: @yogurfrul
Created: 8/7/2026
Status: ❌ Closed

Base: main ← Head: dp_native_core

📝 Commits (1)

c3f13ca feat(native): add standalone CUDA core

📊 Changes

4 files changed (+311 additions, -9 deletions)

View changed files

📝 CMakeLists.txt (+41 -9)
➕ csrc/native_api.cpp (+209 -0)
➕ csrc/native_api.h (+38 -0)
📝 setup.py (+23 -0)

📄 Description

Edge deployment needs to link a runtime library without loading the Python wheel frontend, PyTorch, ATen, or pybind symbols.

- split CMake configuration so the Python extension and native core build independently
- expose a stable C ABI over opaque CUDA streams and row-major tensor views
- implement the NN GEMM path with CUDA Runtime and cuBLASLt only
- package the shared library and public header inside the wheel for build-time discovery and artifact copying
- retain the existing Python extension path as the default build

_{🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.}

## 📋 Pull Request Information **Original PR:** https://github.com/deepseek-ai/DeepGEMM/pull/400 **Author:** [@yogurfrul](https://github.com/yogurfrul) **Created:** 8/7/2026 **Status:** ❌ Closed **Base:** `main` ← **Head:** `dp_native_core` --- ### 📝 Commits (1) - [`c3f13ca`](https://github.com/deepseek-ai/DeepGEMM/commit/c3f13ca641bd33236d4a3eb867e40e53a811ecb1) feat(native): add standalone CUDA core ### 📊 Changes **4 files changed** (+311 additions, -9 deletions) <details> <summary>View changed files</summary> 📝 `CMakeLists.txt` (+41 -9) ➕ `csrc/native_api.cpp` (+209 -0) ➕ `csrc/native_api.h` (+38 -0) 📝 `setup.py` (+23 -0) </details> ### 📄 Description Edge deployment needs to link a runtime library without loading the Python wheel frontend, PyTorch, ATen, or pybind symbols. - split CMake configuration so the Python extension and native core build independently - expose a stable C ABI over opaque CUDA streams and row-major tensor views - implement the NN GEMM path with CUDA Runtime and cuBLASLt only - package the shared library and public header inside the wheel for build-time discovery and artifact copying - retain the existing Python extension path as the default build --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi

2026-10-26 21:17:34 +00:00