Running code on cpu_selector

#1

Hi,
I am trying to run the matrix multiplication code on CPU_SELECTOR but i am getting this error on runtime

Allocate memory…
Running on Intel® Core™ i7-8750H CPU @ 2.20GHz - Intel® Corporation
The Device Max Work Group Size is : 8192
The order is : 16777216
The blockSize is : 64
Internal compiler error invalid llvm.linker.options
Please report the issue on Intel OpenCL forum
https://software.intel.com/en-us/forums/opencl for assistance.

Any pointers about what i am missing here?

Any help is appreciated.
Thank you,
Aatif

#2

Hi Aatif,
The error message is suggesting this is an error within the OpenCL driver provided by Intel. It would be best to go to their website to ask for help.
What Intel OpenCL drivers are you using including the version number?
What is the output of computecpp_info?
Thanks.

#3

Hi Rod,

Thank you for the response. So i was able to resolve this error. I had to uninstall the already installed drivers for my machine and reinstalled opencl drivers from intel then it worked. Thank you

1 Like
#4

Hi Rod,

I was able to run my code on CPU and GPU, but the results i am getting are not what we would expect. I am doing a matrix multiplication, using the code you guys provide in the sample. Here are the timings of this operation.

Please have a look and let me know if they are possible or we could be doing something wrong.

#5

The sample is purely for learning purposes and is not very optimized.
What Cuda code are you comparing this to?

We have some BLAS benchmarks that include matrix multiplication that I will share with you. The PR is in progress but will be available soon. That will offer a better way to compare.

#6

I used the already existing implementation of sgeam method in cublas.
Please do share the code that you are referring to.

#7

The benchmark code and README has now been pushed and is available here. Amongst other things the README explains how to run the GEMM benchmarks. Let me know how you get on.

1 Like