10000 Crash on win10, cuda 10.1 · Issue #70 · NVIDIA/gvdb-voxels · GitHub
[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
Skip to content

Crash on win10, cuda 10.1 #70

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.

Sign up for GitHub

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

Jump to bottom
Closed
chena1982 opened this issue Jun 4, 2019 · 8 comments

Comments

@chena1982
Copy link

win10, cuda 10.1, memory 32G.
C:\GitHub\gvdb-voxels\source\shared_cudpp\src\cudpp\app\reduce_app.cu line 176
call cudaMalloc, will run out of system memory, the used system memory is up to 32G, and the application will crash.

So how cloud i debug this problem?
If have some plane to test on win10, cuda 10.1?
Thanks.
@chena1982 chena1982 changed the title Crash on win10, cuda 10 Crash on win10, cuda 10.1 Jun 4, 2019
@nathanchrs
Copy link
Contributor

Hi @chena1982,

Can you give more details - what code are you trying to run? What GPU are you using?

@chena1982
Copy link
Author
chena1982 commented Jun 7, 2019 •
edited
Loading

Hi @nathanchrs
I just want to init the gvdb.

image

GPU is 980Ti, drive is 425.25.

@nathanchrs
Copy link
Contributor

It seems that this is an issue with CUDPP and CUDA 10: CUDPP issue #185. This might also be related to #34.

For the time being, maybe you can downgrade to CUDA 9 as a workaround?

@johoe
Copy link
johoe commented Jul 11, 2019

Hi, I postet a solution here.
The problem arises from the __shfl functions, you can either replace all __shfl function of cudpp with its corresponding sync version or use compute capability less than 6.0

@chena1982
Copy link
Author

@johoe Thanks for your share.
I replace the __shfl and __shfl_up in H:\gvdb-voxels-master\source\shared_cudpp\src\cudpp\kernel\multisplit_kernel.cuh and H:\gvdb-voxels-master\source\shared_cudpp\ext\moderngpu\include\device\intrinsics.cuh.
But it still don't work.

@johoe
Copy link
johoe commented Jul 16, 2019

Hi @chena1982, we had to modify these files:

  • gvdb-voxels/shared_cudpp/ext/cub/cub/util_ptx.cuh
  • gvdb-voxels/shared_cudpp/ext/cub/cub/warp/specializations/warp_scan_shfl.cuh
  • gvdb-voxels/shared_cudpp/ext/moderngpu/include/device/intrinsics.cuh
  • gvdb-voxels/shared_cudpp/src/cudpp/kernel/multisplit_kernel.cuh

I commited my changes to here. You may have a look.

@chena1982
Copy link
Author

@johoe it works now.
Thanks.

@jhwang7628
Copy link
jhwang7628 commented Aug 20, 2019 •
edited
Loading

Thanks for all the help. @johoe I am worried about the other deprecated functions used in multisplit_kernel function (cub and moderngpu seemed to be patched now). In particular, my MVCC compiler is issuing warnings about deprecated intrinsics about

  • __ballot()
  • instruction 'vote' (warning : Instruction 'vote' without '.sync' is deprecated since PTX ISA version 6.0 and will be discontinued in a future PTX ISA version)

I am curious if you ran into any problem with these? Should we patch them as well?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
5 participants
@johoe @ramakarl @chena1982 @jhwang7628 @nathanchrs
0