[SYCL] Add missing reduction parallel_for shortcuts to queue #6070
Conversation
These changes add the missing parallel_for shortcuts with the variadic parameters containing reduction operations and a kernel functor or lambda. Signed-off-by: Larsen, Steffen <steffen.larsen@intel.com>
|
/verify with intel/llvm-test-suite#1010 |
|
This PR is similar to existing one: #4352 |
There was a problem hiding this comment.
I have couple comments, see below.
Also, this PR duplicates the work done in #4352, which I hoped to merge soon after getting review from @tovinkere
| /// \param Rest is a parameter pack argument of size N where N > 1 and the | ||
| /// first N-1 arguments are reduction operations and the last argument is a | ||
| /// kernel functor or lambda. | ||
| template <typename KernelName = detail::auto_name, int Dims, |
There was a problem hiding this comment.
Here the more general implementation is add, i.e. the one that accepts 'range' instead of having 3 different: 1 for range<1>, one for range<2>, and 3rd for range<3>.
Such split was done on purpose to allow conversions from int scalar: 4009b8b
| (sizeof...(RestT) > 1 && | ||
| ext::oneapi::detail::AreAllButLastReductions<RestT...>::value), | ||
| event> | ||
| parallel_for(range<Dims> NumWorkItems, RestT... Rest) { |
There was a problem hiding this comment.
SYCL 2020 SPEC says it should be 'RestT&&... Rest'
There was a problem hiding this comment.
You're absolutely right, though we do not have that in the handler parallel_for implementation so it made sense to me to keep the signatures the same and migrate at the same time. Either is probably fine.
These changes add the missing parallel_for shortcuts with the variadic parameters containing reduction operations and a kernel functor or lambda.