-
-
Notifications
You must be signed in to change notification settings - Fork 204
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
AMD SI GPU's - Weird result. #301
Comments
OK, that looks like a weird problem, not really sure I like the solution, but I've done something similar here: |
Ok, As I had pointed out to Denis, Hainan is just one of the card names. The actual name should be SI (southern island). I will check the code and get back to you. |
Well this is a first #endif // End of the C 11 raw string literal __kernel void null_kernel() {} Segmentation fault (core dumped) So it would be interesting to know what have you done with your .cl files that make it work in the first place! |
OK, I have just pushed new commits to the branch. Could you try again?
I could reproduce and I've fixed that
OK, renamed it.
Didn't change anything else, so the bug should still be there if you don't enable this workaround. |
Ok, code builds and I tried a few examples naths@naths-HP-Laptop-15-bs1xx:~/build/clblast$ ./clblast_tuner_xaxpy
| ID | total | param | compiles | time | GB/s | status |
There is a bug that I Have reported here also |
Results of clinfo Platform Name: AMD Accelerated Parallel Processing Platform Name: Intel(R) OpenCL HD Graphics |
OK, so what are you saying exactly? I see an error |
naths@naths-HP-Laptop-15-bs1xx:~/build/clblast$ ./clblast_test_xherk
I think this is a test where the error is seen. |
OK, you are right, that is indeed an issue. I fixed it now, but now I'm not sure it still fixes your original issue, since the kernel is enqueued after but not based on events. Could you try again? |
As I had mentioned, your code even without the AMD_SI hack seems to work! I wonder why :)
|
Ok, now when I use your updated code with Remora, I get the following error: |
OK, now I'm confused. So if the original code did work, why did you open this issue? Do all tests ( About the other thing, that is a non CLBlast program, so you'll have to debug that yourself I'm afraid. The OpenCL error -5 indicates |
Yes you can close this issue and merge the code. |
OK, I've just made a PR, will merge as soon as the tests finish. Added a note in the readme as well. |
I have raised an issue in the vexcl repo
ddemidov/vexcl#254
and a solution was provided that solved the issue. Can a similar workaround be provided for CLBlast
also given that this library is used by Caffe?
The text was updated successfully, but these errors were encountered: