Initial SYCL backend rebased for backend refactoring #3447
Conversation
masterleinad
force-pushed
the
sycl_refactor_rebase
branch
from
32d2187 to
c85f574
Compare
|
Retest this please. |
masterleinad
force-pushed
the
sycl_refactor_rebase
branch
from
c85f574 to
dd88c22
Compare
masterleinad
force-pushed
the
sycl_refactor_rebase
branch
from
dd88c22 to
cac31c0
Compare
|
Rebased. |
|
The relevant CI check ( |
| using execution_space = SYCL; | ||
| using memory_space = SYCLDeviceUSMSpace; | ||
| using device_type = Kokkos::Device<execution_space, memory_space>; | ||
| using size_type = unsigned int; |
There was a problem hiding this comment.
@nliber did you put much thought into this or just copied it? Not that I see an issue with it, just curious.
There was a problem hiding this comment.
make the size_type a signed integer. I actually would like to change that on CUDA. Generally as much as possible lets try and avoid unsigned,
There was a problem hiding this comment.
I just copied it, as and the general size_type is expected to be unsigned.
There was a problem hiding this comment.
Why do you want this to be signed? Note that this is an unsigned type in most of the memory spaces I could find. Of course, I am fine creating a pull request to make this consistent (preferrably after merging this one).
masterleinad
force-pushed
the
sycl_refactor_rebase
branch
from
108cf1b to
fea8b6b
Compare
masterleinad
force-pushed
the
sycl_refactor_rebase
branch
2 times, most recently
from
a39e064 to
96e22b8
Compare
Co-authored-by: Nevin Liber <[email protected]>
masterleinad
force-pushed
the
sycl_refactor_rebase
branch
from
96e22b8 to
d3b5600
Compare
Superseding #3444:
This is meant as a stepping stone to integrate #3231 just providing the most basic functionalities to define the backend and pass the Serial tests with Kokkos_ENABLE_SYCL. It turns out that this is enough to go up to feature level 2 in the incremental tests.