CUDA Libraries
Material by: Alan Gray, Rupert Nash
Computational Libraries
- There are many computational operations that are relevant for multiple problems
- It is not productive for each user to implement their own version from scratch
- It is also usually very complex to implement in a way that gets optimal performance
Solution: re-usable libraries
- User just integrates call to library function within code
- Obviously only works if desired library exists
- Library implementation is optimised for platform in use
- Obviously only if desired library is optimised!
Many CPU libraries have developed and in use for many years and an increasing number of GPU libraries are now available
Simple Example: Matrix Multiplication
- Basic Linear Algebra Subprograms (BLAS) has been around since 1979, and provides a range of basic linear algebra operations - optimised for modern CPUs
- cuBLAS, a GPU-accelerated implementation, is available as part of the CUDA distribution
- GPUs are well suited to these highly parallel operations and much faster that CPUs when the matrices are large
Simple Example: Matrix Multiplication
- Other more complex linear algebra operations (matrix inversion, eigenvalue determination, etc) can be built out of multiple BLAS operations)
- On CPU can use one of several LAPACK implementations
- On GPU can use MAGMA (open source)
- Curated list of libraries: http://www.netlib.org/utk/people/JackDongarra/la-sw.html
CUDA includes many libraries
Linear Algebra and Math Libraries
cuBLAS
GPU-accelerated standard BLAS library
CUDA Math Library
GPU-accelerated standard mathematical function library
cuSPARSE
GPU-accelerated BLAS for sparse matrices
Linear Algebra and Math Libraries
cuRAND
GPU-accelerated random number generation (RNG)
cuSOLVER
Dense and sparse direct solvers for Computer Vision, CFD, Computational Chemistry, and Linear Optimization applications
AmgX
GPU accelerated linear solvers for simulations and implicit unstructured methods
Signal, Image and Video Libraries
cuFFT
GPU-accelerated library for Fast Fourier Transforms
NVIDIA Performance Primitives
GPU-accelerated library for image and signal processing
NVIDIA Codec SDK
High-performance APIs and tools for hardware accelerated video encode and decode
Deep Learning Libraries
GPU-accelerated library of primitives for deep neural networks
GPU-accelerated neural network inference library for building deep learning applications
Advanced GPU-accelerated video inference library
Parallel Algorithm Libraries
NCCL
Collective Communications Library for scaling apps across multiple GPUs and nodes
nvGRAPH
GPU-accelerated library for graph analytics
Thrust
GPU-accelerated library of parallel algorithms and data structures
CUB
Cooperative primitives for CUDA C++, such as sort, reduce, or scan, across warps, blocks, and whole devices.
Partner Libraries
GPU-accelerated open-source library for computer vision, image processing and machine learning, now supporting real-time operation
Open-source multi-media framework with a library of plugins for audio and video processing
GPU-accelerated open source library for matrix, signal, and image processing
Partner Libraries
GPU-accelerated linear algebra routines for heterogeneous architectures, by Magma
GPU-accelerated open-source Fortran library with functions for math, signal and image processing, statistics, by RogueWave
Library for graph-processing designed specifically for the GPU
Partner Libraries
GPU-accelerated functions for sparse direct solvers, included in SuiteSparse linear algebra package authored by Prof.
GPU-accelerated linear algebra library by EM Photonics
GPU-accelerated linear algebra (LA) routines for the R platform for statistical computing supporting heterogeneous
Partner Libraries
GPU-accelerated computational geometry engine for advanced GIS, EDA, computer vision, and motion planning, by Fixstars
GPU-accelerated library for sparse iterative methods by Paralution
Real-time visual simulation of oceans, water bodies in games, simulation, and training applications, by Triton