[oneDPL] oneDPL '__future' class code cleanup and API fix #1962
Conversation
| auto | ||
| get() | ||
| get(std::size_t __idx = 0) |
There was a problem hiding this comment.
Everybody who will take a look at this function will ask - what is template param and run-time param here,
Explanation required, I think.
There was a problem hiding this comment.
The bullets 1 and 2 #1962 (comment) explain it.
Yes, probably it make sense to write a comment in the code.
There was a problem hiding this comment.
Better to understand this from code...
| @@ -701,31 +701,42 @@ struct __deferrable_mode | |||
| { | |||
| }; | |||
|
|
|||
| //A contract for future class: <sycl::event or other event, a value, sycl::buffers..., or __usm_host_or_buffer_storage> | |||
| //Impl details: inheritance (private) instead of aggregation for enabling the empty base optimization. | |||
| // An overload of __wait_and_get_value for 'sycl::buffer' | |||
There was a problem hiding this comment.
Can we modify these function without move them outside?
There was a problem hiding this comment.
The change also solves a code coupling issue.
There was a code coupling between following types: __future, 'sycl::buffer', '__result_and_scratch_storage'.
There was a problem hiding this comment.
But why should move them outside of class __future ?
Let's still use them as class-member functions, why not?
MikeDvorskiy
changed the title
MikeDvorskiy
force-pushed
the
dev/mdvorski/future_code_cleanup
branch
2 times, most recently
from
56968d6 to
065b75d
Compare
MikeDvorskiy
force-pushed
the
dev/mdvorski/future_code_cleanup
branch
from
065b75d to
8232c73
Compare
MikeDvorskiy
force-pushed
the
dev/mdvorski/future_code_cleanup
branch
from
e1e50e3 to
07fd3f8
Compare
|
In general, the the change doesn't solve a use case when we need to return/get a couple of values from a future objects. |
A little bit refactoring and class API fix for oneDPL
futuretype.__futureinstance via__future::getmethod.__futureinstance via__future::getmethod.code couplingbetween following types:__future, 'sycl::buffer', '__result_and_scratch_storage'The PR fixes the 1,2,3 mentioned issues.