OpenCL SGEMM implemented using AMD GPU

This is a simple implementation of OpenCL SGEMM using AMD GPU based on various web materials, written only for 64-bit Igor Pro 7, other version not tested.

*It works only for matrices which are divisible by 16 for now.
It is not well optimized, but might serve as a starting point for users who would like to explore computing using GPU with Igor Pro.
With AMD HD 7950, it provides around a factor of 3 speed up over Intel Core i7-4790.

To use it, simply paste 3 files, OpenCLSGEMM64.xop, OpenCLSGEMM Help.ihf and OpenCLSGEMM.cl into "C:\Program Files\WaveMetrics\Igor Pro 7 Folder\Igor Extensions (64-bit)".
You will also need to have AMD cards supporting OpenCL 1.2 or above, and the latest available driver for the card installed.

If you wish to use it with Nvidia card, or you want to keep the .cl kernel file somewhere else, you will need to recompile the XOP. All the source files are also attached for your convenience.

Project Details

Current Project Release

OpenCL SGEMM implemented using AMD GPU IGOR.7.00.x-0.0-2

Release File: OpenCLSGEMM_0.zip
Version: IGOR.7.00.x-0.0-2
Version Date: Thu, 09/21/2017 - 14:08
Version Major: 0
Version Extra: 2
Version Patch Level: 0
OS Compatibility: Windows
Release Notes: -Fixed the GPU memory checking issue for input data size
View All Releases

Forum

Support

Gallery

Igor Pro 8

Learn More

Igor XOP Toolkit

Learn More

Igor NIDAQ Tools MX

Learn More