OpenCL

[timchist]

Scenario:

1. Allocate N buffers, 1 GB each

2. Schedule a kernel call for each of the N buffers

3. Call clFlush

4. Call clSynchronize

5. Call clReleaseMemObject for each of the created buffers

If N is chosen slightly above the amount of Dedicated + Shared GPU in GB on Windows, then all calls return no errors, however, one of the clReleaseMemObject calls causes a crash in amdocl64.dll. For example:

Exception thrown at 0x00007FF941C96BBA (amdocl64.dll) in test.exe: 0xC0000005: Access violation reading location 0x0000000000000000

For example, on my computer with 128 GB RAM and AMD RX 7900 XTX with 24 GB VRAM the test code crashes with N = 90.

[timchist]

When I try to paste the full test code, I am getting this message:

The message body contains "<brace character>", which is not permitted in this community. Please remove this content before sending your post.

How can I provide the sample code?

[timchist]

Here is a link to source code (valid for 30 days): https://we.tl/t-PVjI0QMPj6