Skip to content

Conversation

@9prady9
Copy link
Member

@9prady9 9prady9 commented May 28, 2020

There are few locations where initializing the flags or buffers were
earlier using synchronous copy to GPU memory which is not needed since
the kernel execution in-order. Hence, changed them to be asynchronous
copies.

Fixes #2909

There are few locations where initializing the flags or buffers were
earlier using synchronous copy to GPU memory which is not needed since
the kernel execution in-order. Hence, changed them to be asynchronous
copies.
@9prady9 9prady9 added this to the 3.7.2 milestone May 28, 2020
@9prady9 9prady9 requested a review from umar456 May 28, 2020 04:42
@9prady9 9prady9 merged commit bb76c89 into arrayfire:master May 28, 2020
@9prady9 9prady9 deleted the fix_async_write_vision branch May 28, 2020 07:21
umar456 pushed a commit to umar456/arrayfire that referenced this pull request Jun 27, 2020
* enqueueWriteBuffer asynchronously in vision kernels

There are few locations where initializing the flags or buffers were
earlier using synchronous copy to GPU memory which is not needed since
the kernel execution in-order. Hence, changed them to be asynchronous
copies.

* Fix formatting

* Correct the scope of h_desc_lvl on orb
9prady9 added a commit that referenced this pull request Jun 27, 2020
* enqueueWriteBuffer asynchronously in vision kernels

There are few locations where initializing the flags or buffers were
earlier using synchronous copy to GPU memory which is not needed since
the kernel execution in-order. Hence, changed them to be asynchronous
copies.

* Fix formatting

* Correct the scope of h_desc_lvl on orb
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

af_fast / af::fast performs unnecessary synchronization with clEnqueueWriteBuffer.

2 participants