- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi,
I can't get default device queue using clGetCommandQueueInfo(.., CL_QUEUE_DEVICE_DEFAULT, ..) if before that command I increase and decrease reference counter of that default device queue. I get CL_INVALID_VALUE error. I don't get any errors from clRetainCommandQueue() and clReleaseCommandQueue().
clGetCommandQueueInfo(.., CL_QUEUE_DEVICE_DEFAULT, ..) works just fine if I don't change reference counter of the default device queue.
Test case in the attachment (just adjust intel platform index INTEL_PLATFORM_ID).
Link Copied
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Can anyone confirm that it is a bug? (I don't need a fix now) I haven't check the spec if there is some special case for device queues.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
I think I've reproduced this issue. Can you confirm that the behavior is limited to the CPU OpenCL device, and that you get no CL_INVALID_VALUE with the GPU OpenCL device? Thanks!
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Ben Ashbaugh (Intel) wrote:
I think I've reproduced this issue. Can you confirm that the behavior is limited to the CPU OpenCL device, and that you get no CL_INVALID_VALUE with the GPU OpenCL device? Thanks!
Unfortunately, I can't confirm that, because my iGPU does not support OpenCL 2.1. Currently, I only have access to Experimental OpenCL 2.1 CPU Only Platform (Intel Core i5-6600K CPU), and, IIRC, I would need Intel Kaby Lake processor to run OpenCL 2.1 on the GPU OpenCL device.
I understand that you tested it on Kaby Lake processor (non-experimental Intel OpenCL 2.1 platform) and the bug does not occur on GPU, but it does on CPU, yes?
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Jakub S. wrote:
I understand that you tested it on Kaby Lake processor (non-experimental Intel OpenCL 2.1 platform) and the bug does not occur on GPU, but it does on CPU, yes?
Yes, that is correct.
I will pass this information along to our CPU folks
Thanks for the excellent reproducer!
- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page