AIPS BOOM

NVIDIA-Stdpar

2 min read

NVIDIA Stdpar

Type: Programming Model Tags: NVIDIA, stdpar, C++ parallel algorithms, Fortran DO CONCURRENT, HPC SDK, GPU offload, nvc++, nvfortran Related: NVIDIA-HPC-SDK, NVIDIA-HPC-Compilers, NVIDIA-OpenACC, CUDA-Fortran, NVIDIA-CUDA, CUDA-Unified-Memory, Thrust, NVPL, Nsight-Systems, Nsight-Compute Sources: https://docs.nvidia.com/hpc-sdk/compilers/hpc-compilers-user-guide/index.html, https://docs.nvidia.com/hpc-sdk/index.html Last Updated: 2026-04-29

Summary

NVIDIA Stdpar is the standard-language parallelism path in NVIDIA-HPC-SDK for GPU acceleration through nvc++ and nvfortran. It covers C++17 parallel algorithms with execution policies and Fortran standard parallel constructs such as DO CONCURRENT, letting developers express parallelism in standard language syntax while the NVIDIA compilers generate CPU or NVIDIA GPU code.

Detail

Purpose

Stdpar is useful when teams want GPU acceleration without adding CUDA kernels or directive regions. The NVIDIA compiler analyzes standard C++/Fortran parallel constructs and generates GPU code, with memory-mode rules and limitations documented in the HPC Compilers User’s Guide.

Current scope

  • nvc++ -stdpar support for C++17 parallel algorithms using execution policies.
  • GPU and multicore CPU compilation modes, including GPU/multicore fallback behavior.
  • OpenACC-backed implementation path for selected C++ parallel algorithms through -stdpar=gpu:acc.
  • Fortran standard parallelism through nvfortran, including DO CONCURRENT GPU execution and interoperability with OpenACC and CUDA Fortran.
  • Managed Memory and Unified Memory guidance for data accessed inside parallel algorithms.
  • Device-function inference for functions called from parallel algorithm regions, with external function/device annotation caveats.

NVIDIA context

NVIDIA Stdpar is distinct from CUDA-Fortran and NVIDIA-OpenACC. CUDA Fortran is explicit CUDA programming for Fortran, OpenACC is directive-based offload, and stdpar tries to keep the source closer to standard C++/Fortran while still generating NVIDIA GPU code.

Connections

  • NVIDIA-HPC-SDK - distribution and documentation hub.
  • NVIDIA-HPC-Compilers - nvc++ and nvfortran implement the stdpar path.
  • NVIDIA-OpenACC - selected C++ parallel algorithms can use an OpenACC implementation path, and stdpar can interoperate with OpenACC data directives.
  • CUDA-Fortran - Fortran stdpar guidance sits beside CUDA Fortran and interoperability examples.
  • CUDA-Unified-Memory - central memory-mode concept for stdpar GPU execution.
  • Thrust - C++ parallel algorithms and GPU execution are conceptually adjacent to Thrust-style parallel primitives.
  • NVPL - CPU-side performance libraries can complement mixed CPU/GPU standard parallel workflows.
  • Nsight-Systems and Nsight-Compute - profiler context for generated GPU kernels.

Resources

Graph View

Backlinks