OpenCL

[cozy]

Hello.

I'm trying 'max_constant_size attribute' to optimize OpenCL performance.

But compiler(clBuildProgram) ignored this attribute. Can I have any workaround?

Regards,

source code

kernel void mykernel(global int* a,
constant int* b __attribute__((max_constant_size (65536)))
){
size_t idx = get_global_id(0);
a[idx] = b[idx & 0x3FFF];
}

compiler message

error: unknown attribute 'max_constant_size' ignored [-Werror,-Wunknown-attributes]

My platform

Ubuntu 22.04.1 LTS

ROCm 5.3.0

Radeon Pro VII

[dipak]

Hi @cozy ,

Thank you for the query. I have whitelisted you and moved the post to the AMD OpenCL forum. 

Thanks.

[dipak]

I have forwarded the query to the OpenCL team. I will let you know once I get any feedback from them.

Please note, ROCm related support is provided at the ROCm GitHub site itself. Here is the link to post any  query/issue related to ROCm: https://github.com/RadeonOpenCompute/ROCm/issues

Thanks.

[cozy]

Thank you for your response.

Sorry, I didn't know that support site.

Thanks.

[dipak · Accepted solution]

Here is the OpenCL team's feedback on this: 

"The "max_constant_size" attribute is not a standard OpenCL attribute. Maybe it was supported by the earlier compiler tool-chains like AMDIL or HSAIL, however the current OpenCL compiler does not support this attribute. 

It is to note that it would have little benefit for current AMD GPUs."

[cozy]

I understand. Thank you for your reply.

Please give my regards to OpenCL team.