-
Notifications
You must be signed in to change notification settings - Fork 217
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
CUDA Bug: volatile isParticle Flag necessary #538
Comments
I noticed that we have 128 cells in y-direction and we have 32 cells in y-direction per GPU. The black slices look like they have a length of 1/4 of the GPU length in y-direction. |
I added a workaround that solves the problem. please do not close this issue |
#539 is a workaround. The issue will stay open till a general solution for this problem is found. Do you agree @psychocoderHPC and @ax3l ? |
The issue stays open for now, #539 implemented a work around. We have to write a minimal example to report the bug to get it fixed in future versions of CUDA. |
Is this solved by other means than the workaround #539? |
no, that was an auto-close due to the merge to |
work-around was applied with #539 and does not cause problems right now that we know of. new scope of this issueSince the |
What is the status of this issue? |
It was solved with #539 but it is not clear if the workaround can be removed or not. We should keep it open. |
We could try if it still occurs, when removing the work-around, with CUDA 8+ and if not then it was fixed upstream in nvcc. |
This and other |
@sbastrakov we are setting |
While testing the
Bunch
example in 2D, I discovered the folowing bug, that also exists in 3D.Not all particles get the given momentum and thus stay at their initial possition.
This is clearly visable in 2D.
Some particles at the outer edge of the Gaussian blob do not move and stay as halo behind.
In 3D this is not directly visible.
However, in the
BinEnergyElectrons.dat
there are particles in the first bin (zero Energy) at the first time step for both 3D and 2D.With the help of @psychocoderHPC - we saw that if we set
isParticle
inParticles.kernel
totrue
, all particles are initialized correctly. SettingblockingKernel
toon
did not help. Same for settingtypedef uint16_t lcellId_t;
totypedef uint32_t lcellId_t;
inframe_types.hpp
.The error does not occure for the
KelvinHelmholtz
example.The text was updated successfully, but these errors were encountered: