tpttr

cuBLAS Extension APIs - cublas<t>tpttr

Description

This code demonstrates a usage of cuBLAS tpttr function to perform the conversion from the triangular packed format to the triangular format

A = | 1.0 | 2.0 |
    | 3.0 | 4.0 |

See documentation for further details.

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

• cublas<t>tpttr() API

Building (make)

Prerequisites

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

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 cublas_examples.sln project in Visual Studio and build

Usage

$  ./cublas_tpttr_example

Sample example output:

AP
1.00 2.00 
3.00 4.00 
=====
A
1.00 3.00 
0.00 2.00 
=====