Integrate C++ Co-arrays #2732
Integrate C++ Co-arrays #2732
Conversation
atrantan
force-pushed
the
from_view_to_coarray
branch
2 times, most recently
from
d631b60
to
cbca594
Compare
atrantan
force-pushed
the
from_view_to_coarray
branch
6 times, most recently
from
35b7caf
to
43e8412
Compare
| /////////////////////////////////////////////////////////////////////////////// | ||
| /// \cond NOINTERNAL | ||
|
|
||
| #define HPX_REGISTER_COARRAY(T) HPX_REGISTER_PARTITIONED_VECTOR(T) |
There was a problem hiding this comment.
HPX_REGISTER_PARTITIONED_VECTOR is a variadic macro. Should HPX_REGISTER_COARRAY be variadic as well?
| } | ||
|
|
||
| // Used for "automatic" coarray subscript and "automatic" coarray size | ||
| constexpr hpx::detail::auto_subscript _; |
There was a problem hiding this comment.
Should we move _ into its own namespace? hpx::coarray::placeholders::_ perhaps?
There was a problem hiding this comment.
hpx::coarray is already use for the class. I momentarily put it in hpx::container namespace
| bools<true, std::is_integral<T>::value ...>, | ||
| bools<std::is_integral<T>::value ...,true > >::value > | ||
| {}; | ||
|
|
There was a problem hiding this comment.
Wouldn't it be simpler to just use hpx::util::detail::all_of<std::is_integral<T>::value...> instead (see https://github.com/STEllAR-GROUP/hpx/blob/master/hpx/util/detail/pack.hpp#L62-L81)?
| { | ||
| using type = typename last_element<R...>::type; | ||
| }; | ||
|
|
There was a problem hiding this comment.
What's the rationale to use an unrolled implementation instead of a simple variadic one:
template <typename ... Ts>
struct last_element
{
using type = typename util::detail::at_index<sizeof...(Ts)-1, Ts...>::type;
};
There was a problem hiding this comment.
... or even:
template <typename ... Ts>
using last_element<Ts...> = util::detail::at_index<sizeof...(Ts)-1, Ts...> {};
a4a2de2
to
8ba579d
Compare
| iterator end() | ||
| { | ||
| return sizes_.end(); | ||
| } |
There was a problem hiding this comment.
Should there be const versions of begin and end as well?
| } | ||
|
|
||
| template<typename View, typename Iterator, std::size_t ... I> | ||
| void update_view( |
There was a problem hiding this comment.
Shouldn't this return the updated view instead?
| std::size_t num_segments, | ||
| std::size_t unroll) | ||
| { | ||
| using iterator = typename |
| using iterator = typename | ||
| std::vector<hpx::naming::id_type>::iterator; | ||
|
|
||
| using const_iterator = typename |
| this->set_data(hpx::launch::sync, Data(other) ); | ||
| } | ||
|
|
||
| return std::move(*this); |
There was a problem hiding this comment.
This looks dangerous... Could you remind me what's the rationale for this?
| @@ -182,7 +176,7 @@ namespace hpx | |||
| iterator end() | |||
There was a problem hiding this comment.
Same question as above: should this expose const versions of begin and end as well?
There was a problem hiding this comment.
It implies to duplicate some code, I think similarly to partitioned_vector segmented iterators.
atrantan
force-pushed
the
from_view_to_coarray
branch
5 times, most recently
from
bb5750e
to
0c3ef18
Compare
atrantan
force-pushed
the
from_view_to_coarray
branch
from
a2c638a
to
79dba18
Compare
atrantan
force-pushed
the
from_view_to_coarray
branch
from
6e2d65d
to
97bebac
Compare
atrantan
force-pushed
the
from_view_to_coarray
branch
from
97bebac
to
7c9f374
Compare
This PR proposes to integrate C++ Co-arrays into hpx. It is a combination of the recently merged spmd multidimensionnal views and
partitioned_vector.A Co-array:
partitioned_vector_viewand owns apartitioned_vectoras a variable memberpartitioned_vector_viewsince its last dimension size cannot be defined by the user (equal to the current number of images)partitioned_vectorlayout since partitions that share the same last dimension index lives in the same runtime locality.Features of Co-arrays (with
_aconstexprvariable) :a(i,j,k) = b(i,j,k) or a(i,j,k) = some_std_vectorstd::vector<T> v = ( std::vector<T> ) a(i,j,k)a(i,j,_)(_means the rank of the current image)