nvptx

How to: Run Rust code on your NVIDIA GPU

The missing puzzle piece for NVPTX experience with Rust

Convenient `build.rs` helper for NVPTX crates

Pure-Zig bindings for the NVIDIA CUDA Driver API. No @cImport, no toolkit at build time. Includes a working vector_add example with a Zig kernel compiled to PTX.

A Triton-inspired tile DSL in Zig, for parallel programming

nvptx-tools: a collection of tools for use with nvptx-none GCC toolchains.

Just another build script for gcc an nvptx

LLVM-free rustc codegen backend in pure Rust — MIR to aarch64/NVPTX machine code, compiles arrow-rs; testbed for backward profile-guided optimization

Self-hosted systems language and compiler for machine-learning workloads, forked from KernRift. Generates native AMD GPU (AMDGCN) and NVIDIA PTX code itself, with no ROCm or CUDA toolkit, and runs LLM inference (Llama, Qwen, Mistral, Gemma) matching reference outputs bit-exactly.

PTX interpreter (C++) + differential fuzzing, test-case reduction and Z3 translation validation of LLVM's NVPTX backend

A compiler for a small C-like language, in C++17 on LLVM 18: lexer, recursive-descent parser, type checker, LLVM IR codegen, -O1 to -O3, native executables, and GPU code generation (PTX).

Convenient helper to build NVPTX crates

Static analysis and lints for NVIDIA PTX, no GPU needed. 不需要显卡的 NVIDIA PTX 静态分析工具