Clarify access::target::host_task #537
Conversation
hdelan
force-pushed
the
host-task-accessor
branch
from
c595226 to
4234c45
Compare
This reflects the restructure that @gmlueck suggested in KhronosGroup#537.
This reflects the restructure that @gmlueck suggested in KhronosGroup#537.
hdelan
force-pushed
the
host-task-accessor
branch
from
0a7190b to
c4df016
Compare
hdelan
force-pushed
the
host-task-accessor
branch
from
a011e40 to
1c02843
Compare
|
Thanks @gmlueck . Would you be able to merge this? |
|
I only have unilateral authority to merge editorial changes. I added the "agenda" label, so we can discuss this at the next WG meeting. |
|
Ok sure thanks |
An accessor with access::target::host_task is accessible on the host from within the host task lambda. It is necessary to specify that no get_native_mem method is needed and also that target::host_task means not only 'accessible within host_task' but 'accessible within host_task on the host'. This is necessary as host_task defines a scope, but it doesn't define where the memory is accessible from. It also adds reference to deprecated targets for accessors: target::local and target::constant_buffer for accessors.
hdelan
force-pushed
the
host-task-accessor
branch
from
b98911b to
75162eb
Compare
|
@gmlueck is this OK to merge now? |
|
Merge approvals normally happen during a WG call. Since there are now 2 organization approvals on this PR, I would expect it to be approved in next week's call. Is there a reason you want this merged sooner? If so, we could ask for an exception to the process (but no guarantees). |
Thanks for info. I didn't know this process. Just out of curiosity - how many organizations need to approve a patch in order for a SYCL spec change to be made?
No reason at all. Happy to wait for more approvals. |
For changes to the SYCL spec, two organizations must approve, which are both different from the organization that authored the patch. For changes to the CTS, we decided that only one organization (different from the author) is required. |
|
Thanks for info |
|
WG approved to merge. |
Clarify access::target::host_task
Clarify access::target::host_task (cherry picked from commit 01b97ec)
Clarify access::target::host_task (cherry picked from commit 01b97ec)
An accessor with
access::target::host_taskis accessible on the host from within the host task lambda.Specify that
access::target::host_taskmeans not only 'accessible within host_task' but 'accessible within host_task from the host'. This is necessary ashost_taskdefines a scope, but it doesn't define where the memory is accessible from.