-
Notifications
You must be signed in to change notification settings - Fork 140
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
Strange behaviour in radius_kernel.cu #11
Comments
Thanks, I will check this. |
I have identified the error and fixed it. The error occurred in the GPU version in case there were more potential neighbors than |
Yeah, Thanks. |
Since the last update this test fails again in my environment. I checked the diff and could not find something suspicous (I attached the file). I get the following error: Can you check please? Thanks in advance! |
Try increasing the |
Hi,
I face a very strange behaviour with the radius_kernel.cu on MASTER causing different results on CPU and GPU.
My Test-Scenario is as follows:
I searched a lot and it seems that the inner loop of
radius_kernel.cu
runs too often.This results in too many comparisons and some points are retrieved multiple times.
Do you have an idea what causes the following debug-output?
This
printf
should never occur or what am I overseeing?EDIT
My failure was that I tried to
printf
with %d specifier which is wrong. %ld is the right forptrdiff_t
datatype (assuming a long cast e.g. (long)start_idx_x
).The correct
printf
is:printf("WTF? nx %ld: batch_idx: %ld \n idx: %ld \n start_idx_x: %ld \n end_idx_x: %ld \n start_idx_y: %ld \n end_idx_y: %ld \n", (long) n_x, (long) batch_idx, (long) idx, (long) start_idx_x, (long) end_idx_x, (long) start_idx_y, (long) end_idx_y);
Anyway, the function still returns different results for CPU and GPU. I updated the test-case.
Best regards,
Matthias
The text was updated successfully, but these errors were encountered: