| View previous topic :: View next topic |
| Author |
Message |
Praveen B K
Joined: 04 Jan 2012 Posts: 5
|
Posted: Thu May 16, 2013 11:34 pm Post subject: |
|
|
Dear Mat,
I have sent a note to customer service for follow-up.
Since this is bit critical for us, can you please let us know if there is a workaround or quick fix for this problem.
Keep us posted on this thread. |
|
| Back to top |
|
 |
mkcolg
Joined: 30 Jun 2004 Posts: 4995 Location: The Portland Group Inc.
|
Posted: Fri May 17, 2013 10:24 am Post subject: |
|
|
Hi Praveen,
I added a note and increased the priority. I'll try and keep an eye on it for any updates. Note, I made a typo above. I logged this as TPR#19346, not 19033.
- Mat |
|
| Back to top |
|
 |
mkcolg
Joined: 30 Jun 2004 Posts: 4995 Location: The Portland Group Inc.
|
Posted: Fri May 17, 2013 12:43 pm Post subject: |
|
|
I've had some exchanges with engineering and we found a workaround. Put your kernel launch in a critical section.
| Code: |
!$OMP CRITICAL
call ProcessArray_Kernel_1<<<numOfBlocks, threadsPerBlock>>>(VAR_1_D)
!$OMP END CRITICAL
ierr=cudaThreadSynchronize()
|
Hope this helps,
Mat |
|
| Back to top |
|
 |
Praveen B K
Joined: 04 Jan 2012 Posts: 5
|
Posted: Sun May 19, 2013 10:31 pm Post subject: |
|
|
Thanks Mat,
Workaround works. But as you know it slows down our performance. Will be waiting for the fix. |
|
| Back to top |
|
 |
|