GitHub - NVIDIA/NVPLSamples: NVIDIA Performance Libraries: Sample code (original) (raw)

NVPL Samples

The NVIDIA Performance Libraries (NVPL) are a collection of high performance mathematical libraries optimized for the NVIDIA Grace Armv9.0-A Neoverse-V2 architecture.

These CPU-only libraries have no dependencies on CUDA or CTK, and are drop in replacements for standard C and Fortran mathematical APIs allowing HPC applications to achieve maximum performance on the Grace platform.

The provided sample codes show how to call and link to NVPL Libraries in Fortran, C, and C++ applications and libraries. Most examples use CMake, but are easily modified for use in custom build environments.

NVPL Documentation

Installation

NVPL Downloads
Latest release: NVPL-25.1

Library Samples

Samples are compatible with the latest nvpl release. Compatibility with older releases is not guaranteed.

Support

Systems

Architecture: aarch64-linux
Platform: Arm SBSA
CPUs Supported
- NVIDIA Grace (Armv9.0-A Neoverse-V2)
- AWS Graviton 4 (Armv9.00-A Neoverse-V2)
- AWS Graviton 3/3e (Armv8.4-A Neoverse-V1)
- AWS Graviton 2 (Arm-8.2-A Neoverse-N1)
- Ampere Altra (Armv8.2-A Neoverse-N1)
- Any CPU with Armv8.1-A or later micro Architecture
OS (Linux)
- Ubuntu: 20.04, 22.04, 24.04, 24.10
- Debian: 12
- RHEL: RHEL8, RHEL9
- Fedora: 39, 40, 41
- SLES: SLES15 (15.6)
- OpenSUSE/leap: 15.6
- AmazonLinux: 2, 2023
- Generally any Linux OS with support for aarch64

Compilers

GCC-8 - GCC-14+
Clang-14 - Clang-19+
Clang for NVIDIA Grace: 16.x, 17.x, 18.x, 19.x
NVIDA HPC Compilers: 23.9 - 24.11

Languages

C: All libraries
C++: All libraries via C interfaces
Fortran: Selected libraries
- GFortran ABI
- NVPL BLAS, LAPACK, and ScaLAPACK provide lp64 and ilp64 integer ABIs
- NVPL FFT provides FFTW Fortran '77 and '03 compatible interfaces
- See individual libraries samples documentation for further details

OpenMP

All libraries support the following OpenMP runtime libraries. See individual libraries documentation for details and API extensions supporting nested parallelism.

GCC: libgomp.so
Clang: libomp.so
NVHPC: libnvomp.so

MPI

NVPL provides standard BLACS interfaces for the following MPI distributions. See the NVPL ScaLAPACK Samples Documentation for details.

MPICH: Runtime support for >=mpich-4.0
OpenMPI-3.x
OpenMPI-4.x
OpenMPI-5.x
NVIDIA HPC-X: Use openmpi4 BLACS interface

CMake Usage

NVPL provides CMake Package Configfiles for the each component library.

Finding NVPL Packages

If NVPL was installed via the OS package manager under the /usrdirectory, the NVPL packages will already be on the defaultCMAKE_PREFIX_PATH. The nvpl_ROOT environment can be used to override the default search path and force finding nvpl under a specific prefix.

Thefind_package()command is used to find nvpl and any component libraries:

Each NVPL component library found will print a brief status message with important locations.

Variable nvpl_FOUND will be true if nvpl is successfully found
Variable nvpl_VERSION will contain the found version
Pass the REQUIRED keyword to raise an error if nvpl package is not found.
Regardless of the COMPONENTS keyword, all available nvpl component libraries installed in the same prefix will be found.
To raise an error if a particular component is not found, useREQUIRED COMPONENTS ...
Set QUIET to avoid printing status messages, or reporting an error if nvpl is not found
find_package(nvpl) can safely be called multiple times from different locations in a project.

Linking to NVPL Packages

The NVPL component libraries provide Imported Interface Targetsunder the common nvpl:: namespace. To add all the necessary flags to compile and link against NVPL libraries, use thetarget_link_libraries()command:

target_link_libraries(my_target PUBLIC nvpl::_)

Here <lib> is the lowercase shorthand for the library and and <opts>are defined by the library.

NVPL Targets

NVPL component and target names use all-lowercase naming schema. See individual libraries documentation for details on available options.

Component	Targets	Options / Notes
blas	nvpl::blas__	: lp64, ilp64: seq, omp
fft	nvpl::fftw	FFTW API interface
lapack	nvpl::lapack__	: lp64, ilp64: seq, omp
rand	nvpl::randnvpl::rand_mt	Single-threadedMulti-threaded (OpenMP)
scalapack	nvpl::blacs__nvpl::scalapack_	: lp64, ilp64: mpich, openmpi3,openmpi4, openmpi5
sparse	nvpl::sparse
tensor	nvpl::tensor

NVPL Variables

Each nvpl component library also exports variables

nvpl_<comp>_VERSION - Version of component library
nvpl_<comp>_INCLUDE_DIR - Full path to component headers directory
nvpl_<comp>_LIBRARY_DIR - Full path to component libraries directory

LICENSE

These Sample codes are provided under the NVIDIA Software license for NVPL SDK.