CUDA Toolkit Conan Packages

This repository contains Conan 2.0 packages for Nvidia CUDA Toolkit.

Status

• nvcc
  • package recipe
  • test package recipe
    • working under Windows (MSVC) / Linux
• cudart working
  • package recipe
  • test package recipe
    • working under Windows (MSVC) / Linux
  • installing only shared or static libs
• nvvm working correctly as separate package
  • package recipe
  • test package recipe

License

This repository is licensed under MIT license.

Disclaimer: Using any of the packages contained in the repository automatically means that user read & agree to NVidia CUDA Toolkit EULA.

Packages

Nvidia NVCC component

nvcrt

Nvidia Compiler Runtime containing headers include/crt/*, include/fatbin_section.h required by cudart & nvcc

nvcc

Nvidia Compiler - exposes nvcc as tool.

Nvidia Runtime component

cuda-runtime (CUDA::cudart_static & CUDA::cudart)

Nvidia CUDA Runtime creates:

• cudart library containing only shared or shatic library (depending on self.option.shared)
• cudart_deps link to CUDA::cudart_static_deps (no link)

Headers copied from cudart package:

• builtin_types.h
• channel_descriptor.h
• common_functions.h
• cuda_d3d10_interop.h
• cuda_d3d11_interop.h
• cuda_d3d9_interop.h
• cuda_egl_interop.h
• cuda_gl_interop.h
• cuda_occupancy.h
• cuda_runtime.h
• cuda_runtime_api.h
• cuda_texture_types.h
• cuda_surface_types.h
• cudart_platform.h
• device_atomic_functions.h
• device_atomic_functions.hpp
• device_double_functions.h
• device_types.h
• driver_functions.h
• driver_types.h
• host_config.h
• host_defines.h
• library_types.h
• math_constants.h
• math_functions.h
• device_functions.h
• device_launch_parameters.h
• texture_types.h
• texture_fetch_functions.h
• texture_indirect_functions.h
• vector_functions.h
• vector_functions.hpp
• surface_indirect_functions.h
• surface_functions.h
• sm_60_atomic_functions.h
• sm_60_atomic_functions.hpp
• sm_61_intrinsics.h
• sm_61_intrinsics.hpp
• sm_20_atomic_functions.h
• sm_20_atomic_functions.hpp
• sm_20_intrinsics.h
• sm_20_intrinsics.hpp
• sm_30_intrinsics.h
• sm_30_intrinsics.hpp
• sm_32_atomic_functions.h
• sm_32_atomic_functions.hpp
• sm_32_intrinsics.h
• sm_32_intrinsics.hpp
• sm_35_atomic_functions.h
• sm_35_intrinsics.h
• surface_types.h
• vector_types.h
• cuda_device_runtime_api.h

cudadevrt/cuda-device-runtime - CUDA device runtime

Always static library without headers - required by nvcc

Nvidia CUDA Device runtime - contains cudadevrt static library (no headers?) Create CUDA::cudadevrt - used only by nvcc during linking device-link

nvcuda/cuda-driver (CUDA::cuda_driver)

Nvidia cuda.lib dynamic library for driver (links driver nvcuda.dll in windows)

Headers nvcuda:

• cuda.h
• cuComplex.h
• cudaD3D10.h
• cudaD3D10Typedefs.h
• cudaD3D11.h
• cudaD3D11Typedefs.h
• cudaD3D9.h
• cudaD3D9Typedefs.h
• cudaGL.h
• cudaGLTypedefs.h
• cudaTypedefs.h

cudacrt (cuda common runtime files) - only headers

Math API headers are placed between multiple packages

Headers copied from cudart package:

• cuda_bf16.h
• cuda_bf16.hpp
• cuda_fp16.h
• cuda_fp16.hpp
• cuda_fp8.h
• cuda_fp8.hpp
• mma.h (mixed math arithmemetic)

See https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html?highlight=nvfunctional#polymorphic-function-wrappers

Headers copied from cudart package

• nvfunctional

coporative groups see. https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#cooperative-groups

• cooperative_groups.h
• cuda_awbarrier.h
• cuda_awbarrier_helpers.h
• cuda_awbarrier_primitives.h

cuda extensions see. https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#c-language-extensions

• cuda_pipeline_helpers.h
• cuda_pipeline_primitives.h
• cuda_pipeline.h

some profiler things (missing cudaProfiler header):

• cudaProfilerTypedefs.h