Enable CLBlast on i.MX8MQ and i.MX8QM/QXP for Yocto project

3 Kudos

CLBlast (https://github.com/CNugteren/CLBlast) is a modern, lightweight, performant and tunable OpenCL BLAS library written in C++11. CLBlast implements BLAS routines: basic linear algebra subprograms operating on vectors and matrices. I enable the library on i.MX8MQ EVK/ i.MX8QXP MEK based on Vivante GPU GC7000L and i.MX8QM MEK Vivante GPU GC7000XSVX. And I also tune its performance on i.MX8MQ/8QM/8QXP following https://github.com/CNugteren/CLBlast/blob/master/doc/tuning.md. The attached Yocto Recipe bb file base on L4.14.98 GA release.

pachedo · ‎06-06-2019

Hi could you tell me how to add this layer to yocto?

xiaodong_zhang · ‎06-09-2019

Hi Luis Neto

you need to downlaod attated file clblast.tgz and uncompress it. And then copy folder clblast to <i.MX Yocto Project folder>/sources/meta-fsl-bsp-release/imx/meta-sdk/recipes-support/. Later, you can run "bitbake clbast" to build it after you run the command of i.MX Linux Yocto Project Setup, like "DISTRO=fsl-imx-wayland MACHINE=imx8mqevk source fsl-setup-release.sh -b imx8mqevk".

pachedo · ‎06-13-2019

It worked!!

Could you share your tuning results?

xiaodong_zhang · ‎06-13-2019

Hi Luis Neto

My tuning results also be included in 0001-tuning-for-Vivante-OpenCL-Device-GPU.patch.