gesvdj

cuSOLVER Singular Value Decomposition with singular vector (via Jacobi method) example

Description

This code demonstrates a usage of cuSOLVER gesvdj function to perform singular value decomposition

A = U * Σ * V^H

A is a 3x2 dense matrices,

A = | 1.0 | 2.0 |
    | 4.0 | 5.0 |
    | 2.0 | 1.0 |

Supported SM Architectures

All GPUs supported by CUDA Toolkit (https://developer.nvidia.com/cuda-gpus)

Supported OSes

Linux
Windows

Supported CPU Architecture

x86_64
ppc64le
arm64-sbsa

CUDA APIs involved

• cusolverDnDgesvdj_bufferSize API
• cusolverDnDgesvdj API

Building (make)

Prerequisites

• A Linux/Windows system with recent NVIDIA drivers.
• CMake version 3.18 minimum
• Minimum CUDA 10.1 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/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

$  ./cusolver_gesvdj_example

Sample example output:

m = 3, n = 2
tol = 1.000000E-07, default value is machine zero
max. sweeps = 15, default value is 100
econ = 0
A = (matlab base-1)
1.00 2.00
4.00 5.00
2.00 1.00
=====
gesvdj converges
S = singular values (matlab base-1)
7.07
1.04
=====
U = left singular vectors (matlab base-1)
0.31 -0.49 0.82
0.91 -0.11 -0.41
0.29 0.87 0.41
=====
V = right singular vectors (matlab base-1)
0.64 0.77
0.77 -0.64
=====
|S - S_exact|_sup = 4.440892E-16
residual |A - U*S*V**H|_F = 3.511066E-16
number of executed sweeps = 1