Getting Started with pyOpenCL in WSL

By -

In the past, I experimented with GPU programming, using CUDA on Nvidia GPUs. Back then, the bandwidth of the PCIe interface between the CPU and the GPU was deemed to be one of it's Achilles heel. It was about the same time when AMD was putting out APUs post-ATI acquisition, but at that time, the company was just struggling compared to it's peer, Nvidia. Having highly capable and complex CPU cores with massively parallel  GPU cores share the same memory space and working together in a heterogeneous compute environment sounded like a dream.

I recently had to look into buying a Windows PC for my wife and I found some nice used mini PCs with a  35 W AMD Ryzen 5 PRO 2400GE on ebay. I found them very interesting and decided to purchase one. 

I'll attempt to use this GPU for general purpose compute and see how things go. I'll just install the necessary libraries, and run some sample code. This is also an opportunity for me to test this out in WSL and see if it works.


I used the guide in https://homepages.math.uic.edu/~jan/mcs572f16/mcs572notes/lec29.html to get started.

Installed pyopencl by running

pip install pyopencl
and then jumped into the python console 
import pyopencl
from pyopencl.tools import get_test_platforms_and_devices
get_test_platforms_and_devices()
gave me the error 
LogicError: clGetPlatformIDs failed: PLATFORM_NOT_FOUND_KHR
It looks like there isn't opencl intsalled. I followed the instructions here to install amd radeon software - unsure what this does https://rocm.docs.amd.com/projects/radeon/en/latest/docs/install/wsl/install-radeon.html 

That seemed to have worked, and now when I run get_test_platforms_and_devices() , I get 
ROCR: unsupported GPU
[(pyopencl.Platform 'AMD Accelerated Parallel Processing' at 0x7f1ed0437fd0,
  [])]
That led me to find out if my device is supported. Apparently not: https://rocm.docs.amd.com/projects/radeon/en/latest/docs/compatibility/wsl/wsl_compatibility.html 

That is it for part 1. I'll see if I can get this working in Windows before installing a full Linux distro on this mini PC. 

Popular posts from this blog

Dissecting OpenCL code

By -
Image
In my previous post , I got sample OpenCL matrix multiplication kernel to run inside Python on Windows. That was one of the first time I got to work on openCL. The code itself is pretty self explanatory for the most part, but it is interesting to see the comparisons with CUDA, and also how the kernel gets invoked though python. import pyopencl as cl import numpy as np import os os.environ['PYOPENCL_CTX']='0' (n, m, p) = (3, 4, 5) a = np.random.randn(n, m).astype(np.float32) b = np.random.randn(m, p).astype(np.float32) c = np.zeros((n*p), dtype=np.float32) context = cl.create_some_context() queue = cl.CommandQueue(context) mf = cl.mem_flags a_buf = cl.Buffer(context, mf.READ_ONLY | mf.COPY_HOST_PTR, hostbuf=a) b_buf = cl.Buffer(context, mf.READ_ONLY | mf.COPY_HOST_PTR, hostbuf=b) c_buf = cl.Buffer(context, mf.WRITE_ONLY, c.nbytes) prg = cl.Program(context, """ __kernel void multiply(ushort n, ushort m, ushort p, __global float *a, __global fl...
Read more

Starting and using GT.M

By -
To set UTF-8 and start gt.m. $ export gtm_chset="UTF-8" $ export LC_CTYPE="en_US.utf8" $ export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:<path_to_gtm_installation_directory> $ export PATH=$PATH:<path_to_gtm_installation_directory> $ export localWorkingDir="/home/harish/mBackend/" $ source /opt/gtm/gtmprofile $ export gtmgbldir=$localWorkingDir/database $ export export gtmroutines="$localWorkingDir/o($localWorkingDir/r) /opt/gtm/libgtmutil" Launch with $ gtm If encryption is desired, set the gtm_passwd and gtm_dbkeys environment variables first and then run the gtm script
Read more