NVIDIA Fortran CUDA Interfaces

Type: Library Tags: NVIDIA, Fortran, CUDA libraries, HPC SDK, CUDA Fortran, OpenACC, OpenMP, cuBLAS, cuFFT, cuRAND, cuSPARSE Related: CUDA-Fortran, NVIDIA-HPC-SDK, NVIDIA-HPC-Compilers, NVIDIA-OpenACC, NVIDIA-Stdpar, NVIDIA-CUDA, cuBLAS, cuFFT, cuRAND, cuSPARSE, cuSOLVER, cuTENSOR, NCCL, NVSHMEM, NVTX Sources: https://docs.nvidia.com/hpc-sdk/compilers/fortran-cuda-interfaces/frontmatter.html, https://docs.nvidia.com/hpc-sdk/compilers/fortran-cuda-interfaces/iface-introduction.html, https://docs.nvidia.com/hpc-sdk/index.html Last Updated: 2026-04-29

Summary

NVIDIA Fortran CUDA Interfaces document the NVIDIA-provided Fortran modules and interfaces for calling CUDA libraries from Fortran code. The current docs cover pre-built interface modules for cuBLAS, cuFFT, cuRAND, cuSPARSE, cuSOLVER, cuTENSOR, NCCL, NVSHMEM, NVTX, and related CUDA library calls from CUDA-Fortran, NVIDIA-OpenACC, or OpenMP host code.

Detail

Purpose

Fortran HPC applications often need CUDA library performance without rewriting call sites in C or C++. These interfaces let NVIDIA Fortran code call CUDA library functions directly, with module definitions, derived types, enumerations, wrappers, and examples that bridge Fortran language rules to CUDA C library APIs.

Current scope

• Fortran module and interface reference for CUDA library functions.
• BLAS interfaces for cuBLAS.
• FFT interfaces for cuFFT.
• Random-number interfaces for cuRAND.
• Sparse-matrix interfaces for cuSPARSE.
• Solver interfaces for cuSOLVER.
• Tensor primitive interfaces for cuTENSOR.
• Communication/profiling interfaces for NCCL, NVSHMEM, and NVTX.
• Usage from OpenACC/OpenMP host code and from CUDA-Fortran host/device contexts.
• Guidance for writing custom CUDA interfaces using iso_c_binding, explicit interfaces, and wrapper patterns.

NVIDIA context

This page complements CUDA-Fortran. CUDA Fortran explains the programming model, while NVIDIA Fortran CUDA Interfaces explain how Fortran applications call the CUDA library ecosystem once code and data are on the GPU.

Connections

• CUDA-Fortran - programming model that uses these library interfaces.
• NVIDIA-HPC-SDK - documentation and distribution context.
• NVIDIA-HPC-Compilers - nvfortran compiler and module context.
• NVIDIA-OpenACC - host-data and CUDA stream interoperability path for OpenACC code calling CUDA libraries.
• NVIDIA-Stdpar - adjacent Fortran/C++ standard parallelism path in the HPC compilers.
• cuBLAS, cuFFT, cuRAND, cuSPARSE, cuSOLVER, and cuTENSOR - CUDA libraries with Fortran interface coverage.
• NCCL, NVSHMEM, and NVTX - communication/profiling APIs with Fortran interface coverage.

Resources

• Fortran CUDA Interfaces Preface
• Fortran CUDA Interfaces Introduction