dividiti/gemmbench

Benchmark for generic matrix-matrix multiplication (GEMM)

C++BSD-3-Clause

Benchmark for generic matrix-matrix multiplication (GEMM)

Prerequisites

Collective Knowledge Framework: http://github.com/ctuning/ck
OpenCL library

Installation

ck pull repo --url=https://github.com/dividiti/gemmbench

or

ck pull repo --url=git@github.com:dividiti/gemmbench.git

TBD: register OpenCL with CK

Usage

Compile benchmark

ck compile program:gemmbench-cl-launcher-1.0

Run benchmark (select "default" CMD when asked)

ck run program:gemmbench-cl-launcher-1.0

Run benchmark with custom CMD (for example, change matrix order, platform ID and device ID):

ck run program:gemmbench-cl-launcher-1.0 --extra_run_cmd="-n 512 -p 0 -d 1"

Publications

http://arxiv.org/abs/1511.03742