cancel
Showing results for 
Search instead for 
Did you mean: 

Archives Discussions

nou
Exemplar

Segfault in clEnqueueNDRange() when is device memory full Catalyst 12.8

I noticed bug in new Catalyst 12.8 OpenCL runtime. Program just segafault when i enqueue kernel with buffer which doesn't fit into device memory.

for(int i=0;i<count;i++)

    {

        buffers.push_back(new cl::Buffer(context, CL_MEM_READ_WRITE, size*sizeof(float), 0, &err));

        if(err)break;

        kernel.setArg(0, *buffers);

        err = queue.enqueueNDRangeKernel(kernel, cl::NullRange, cl::NDRange(size));//crash at 5th loop

        cout << "enqueueu kernel" << err << endl;

        queue.flush();

    }

with runtime from original SDK i got normal -5 CL_OUT_OF_RESOURCE error.

OS: Linux Ubuntu 11.10

GPU: Radeon 5850 1GB

attached full source code test case.

4 Replies
kcarney
Staff

Hello!

Thank you very much for taking the time to report this bug. The catalyst QA team has asked me to direct all Catalyst bug reports to this form: http://www.amdsurveys.com/se.ashx?s=5A1E27D20B2F3EBF This way, they can follow up with you directly.

Thanks!

Kristen

0 Likes

this is bug in AMD APP runtime. not the catalyst driver itself. it most likely inside of libamdocl64.so as when I revert it segfault goes away.

0 Likes

Hmmm...the APP SDK runtime is in the Catalyst driver. So I think a Catalyst bug report would still be appropriate, no?

0 Likes
nou
Exemplar

installed catalyst 12.11 beta and it seem like solved.

0 Likes