Skip to content

Latest commit

 

History

History
 
 

gesv

cuSOLVER iterative refinement solver example

Description

This code demonstrates usage of cuSOLVER gesv functions introduced in CUDA 10.2 that provides interface to linear system solver with multiple right hand sides using factorization of initial system in specified precision. cuSOLVER provides two sets of APIs for Iterative Refinement Solver functionality - one is similar to LAPACK's GESV and another 'expert' API which gives more configurable options that the user can set through solver parameters. Examples perform following steps for both APIs:

  • Generating random diagonal dominant matrix of provided type on the host
  • Generating random right hand side vectors for the linear system on the host
  • Initializing required CUDA and cuSOLVER miscelaneous variables
  • Allocating required device memory for input data and workbuffer for the solver
  • Copying input data to the device
  • Solving the system of equations
  • Checking return errors and information
  • Releasing used resources

Key Concepts

Linear Solver, Factorization, Mixed Precision, Tensor Cores

Supported SM Architectures

SM 7.0
SM 7.2
SM 7.5
SM 8.0
SM 8.6

Supported OSes

Linux
Windows

Supported CPU Architecture

x86_64
ppc64le
arm64-sbsa

CUDA APIs involved

Building (make)

Prerequisites

  • A Linux/Windows system with recent NVIDIA drivers.
  • CMake version 3.18 minimum
  • Minimum CUDA 10.2 toolkit is required.

Build command on Linux

$ mkdir build
$ cd build
$ cmake ..
$ make

Make sure that CMake finds expected CUDA Toolkit. If that is not the case you can add argument -DCMAKE_CUDA_COMPILER=/path/to/cuda-10.2/bin/nvcc to cmake command.

Build command on Windows

$ mkdir build
$ cd build
$ cmake -DCMAKE_GENERATOR_PLATFORM=x64 ..
$ Open cusolver_examples.sln project in Visual Studio and build

Usage

Produced are two binaries - one uses expert API for gesv() function, and another uses lapack style API, with interface similar to LAPACK GESV function.

Lapack style API

Usage:

$  ./cusolver_irs_lapack

Sample example output:

Generating matrix A on host...
make A diagonal dominant...
Generating matrix B on host...
Generating matrix X on host...
Initializing CUDA...
Allocating memory on device...
Workspace is 12591744 bytes
Solving matrix on device...
Solve info is: 0, iter is: 2
Releasing resources...
Done!

Expert API

Usage:

$  ./cusolver_irs_expert

Sample example output:

Generating matrix A on host...
make A diagonal dominant...
Generating matrix B on host...
Generating matrix X on host...
Initializing CUDA...
Setting up gesv() parameters...
Allocating memory on device...
Workspace is 12591744 bytes
Solving matrix on device...
Solve info is: 0, iter is: 2
Solved matrix 1024x1024 with 1 right hand sides in 19.6782ms
Releasing resources...
Done!