Skip to content

[2025-06 LWG 19] P3481R5 std::execution::bulk() issues #8020

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

[2025-06 LWG 19] P3481R5 std::execution::bulk() issues #8020

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
206 changes: 184 additions & 22 deletions source/exec.tex
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -629,6 +629,8 @@
struct @\libglobal{let_error_t}@ { @\unspec@ };
struct @\libglobal{let_stopped_t}@ { @\unspec@ };
struct @\libglobal{bulk_t}@ { @\unspec@ };
struct @\libglobal{bulk_chunked_t}@ { @\unspec@ };
struct @\libglobal{bulk_unchunked_t}@ { @\unspec@ };
struct @\libglobal{split_t}@ { @\unspec@ };
struct @\libglobal{when_all_t}@ { @\unspec@ };
struct @\libglobal{when_all_with_variant_t}@ { @\unspec@ };
Expand All @@ -647,6 +649,8 @@
inline constexpr let_error_t @\libglobal{let_error}@{};
inline constexpr let_stopped_t @\libglobal{let_stopped}@{};
inline constexpr bulk_t @\libglobal{bulk}@{};
inline constexpr bulk_chunked_t @\libglobal{bulk_chunked}@{};
inline constexpr bulk_unchunked_t @\libglobal{bulk_unchunked}@{};
inline constexpr split_t @\libglobal{split}@{};
inline constexpr when_all_t @\libglobal{when_all}@{};
inline constexpr when_all_with_variant_t @\libglobal{when_all_with_variant}@{};
Expand Down Expand Up @@ -3608,52 +3612,133 @@
propagates the other completion operations sent by \tcode{sndr}.
\end{itemize}

\rSec3[exec.bulk]{\tcode{execution::bulk}}
\rSec3[exec.bulk]{\tcode{execution::bulk}, \tcode{execution::bulk_chunked}, and \tcode{execution::bulk_unchunked}}

\pnum
\tcode{bulk} runs a task repeatedly for every index in an index space.
\tcode{bulk}, \tcode{bulk_chunked}, and \tcode{bulk_unchunked}
run a task repeatedly for every index in an index space.

The name \tcode{bulk} denotes a pipeable sender adaptor object.
For subexpressions \tcode{sndr}, \tcode{shape}, and \tcode{f},
let \tcode{Shape} be \tcode{decltype(auto(shape))}.
\pnum
The names \tcode{bulk}, \tcode{bulk_chunked}, and \tcode{bulk_unchunked}
denote pipeable sender adaptor objects.
Let \placeholder{bulk-algo} be either
\tcode{bulk}, \tcode{bulk_chunked}, or \tcode{bulk_unchunked}.
For subexpressions \tcode{sndr}, \tcode{policy}, \tcode{shape}, and \tcode{f},
let
\tcode{Policy} be \tcode{remove_cvref_t<decltype(policy)>},
\tcode{Shape} be \tcode{decltype(auto(shape))}, and
\tcode{Func} be \tcode{decay_t<decltype((f))>}.
If
\begin{itemize}
\item
\tcode{decltype((sndr))} does not satisfy \libconcept{sender}, or
\item
\tcode{is_execution_policy_v<Policy>} is \tcode{false}, or
\item
\tcode{Shape} does not satisfy \libconcept{integral}, or
\item
\tcode{decltype((f))} does not satisfy \exposconcept{movable-value},
\tcode{Func} does not model \libconcept{copy_constructible},
\end{itemize}
\tcode{bulk(sndr, shape, f)} is ill-formed.
\tcode{\placeholder{bulk-algo}(sndr, policy, shape, f)} is ill-formed.

\pnum
Otherwise,
the expression \tcode{bulk(sndr, shape, f)} is expression-equivalent to:
the expression \tcode{\placeholder{bulk-algo}(sndr, policy, shape, f)}
is expression-equivalent to:

\begin{codeblock}
transform_sender(@\exposid{get-domain-early}@(sndr), @\exposid{make-sender}@(bulk, @\exposid{product-type}@{shape, f}, sndr))
transform_sender(@\exposid{get-domain-early}@(sndr),
@\exposid{make-sender}@(@\placeholder{bulk-algo}@,
@\exposid{product-type}@<@\seebelow@, Shape, Func>{policy, shape, f},
sndr))
\end{codeblock}
except that \tcode{sndr} is evaluated only once.

\par
The first template argument of \exposid{product-type} is \tcode{Policy}
if \tcode{Policy} models \libconcept{copy_constructible}, and
\tcode{const Policy\&} otherwise.

\pnum
Let \tcode{sndr} and \tcode{env} be subexpressions such that
\tcode{Sndr} is \tcode{decltype((sndr))}.
If \tcode{\exposconcept{sender-for}<Sndr, bulk_t>} is \tcode{false}, then
the expression \tcode{bulk.transform_sender(sndr, env)} is ill-formed;
otherwise, it is equivalent to:
\begin{codeblock}
auto [_, data, child] = sndr;
auto& [policy, shape, f] = data;
auto new_f = [func=std::move(f)](Shape begin, Shape end, auto&&... vs)
noexcept(noexcept(f(begin, vs...))) {
while (begin != end) func(begin++, vs...);
}
return bulk_chunked(std::move(child), policy, shape, std::move(new_f));
\end{codeblock}

\begin{note}
This causes the \tcode{bulk(sndr, policy, shape, f)} sender to be
expressed in terms of \tcode{bulk_chunked(sndr, policy, shape, f)} when
it is connected to a receiver whose
execution domain does not customize \tcode{bulk}.
\end{note}

\pnum
The exposition-only class template \exposid{impls-for}\iref{exec.snd.general}
is specialized for \tcode{bulk_t} as follows:
is specialized for \tcode{bulk_chunked_t} as follows:
\begin{codeblock}
namespace std::execution {
template<>
struct @\exposid{impls-for}@<bulk_t> : @\exposid{default-impls}@ {
struct @\exposid{impls-for}@<bulk_chunked_t> : @\exposid{default-impls}@ {
static constexpr auto @\exposid{complete}@ = @\seebelow@;
};
}
\end{codeblock}

\par
The member \tcode{\exposid{impls-for}<bulk_chunked_t>::\exposid{complete}}
is initialized with a callable object equivalent to the following lambda:
\begin{codeblock}
[]<class Index, class State, class Rcvr, class Tag, class... Args>
(Index, State& state, Rcvr& rcvr, Tag, Args&&... args) noexcept
-> void requires @\seebelow@ {
if constexpr (@\libconcept{same_as}@<Tag, set_value_t>) {
auto& [policy, shape, f] = state;
constexpr bool nothrow = noexcept(f(auto(shape), auto(shape), args...));
@\exposid{TRY-EVAL}@(rcvr, [&]() noexcept(nothrow) {
f(static_cast<decltype(auto(shape))>(0), auto(shape), args...);
Tag()(std::move(rcvr), std::forward<Args>(args)...);
}());
} else {
Tag()(std::move(rcvr), std::forward<Args>(args)...);
}
}
\end{codeblock}

\par
The expression in the \grammarterm{requires-clause} of the lambda above is
\tcode{true} if and only
if \tcode{Tag} denotes a type other than \tcode{set_value_t} or
if the expression \tcode{f(auto(shape), auto(shape), args...)} is well-formed.

\pnum
The member \tcode{\exposid{impls-for}<bulk_t>::\exposid{complete}}
The exposition-only class template \exposid{impls-for}\iref{exec.snd.general}
is specialized for \tcode{bulk_unchunked_t} as follows:
\begin{codeblock}
namespace std::execution {
template<>
struct @\exposid{impls-for}@<bulk_unchunked_t> : @\exposid{default-impls}@ {
static constexpr auto @\exposid{complete}@ = @\seebelow@;
};
}
\end{codeblock}

\par
The member \tcode{\exposid{impls-for}<bulk_unchunked_t>::\exposid{complete}}
is initialized with a callable object equivalent to the following lambda:
\begin{codeblock}
[]<class Index, class State, class Rcvr, class Tag, class... Args>
(Index, State& state, Rcvr& rcvr, Tag, Args&&... args) noexcept -> void requires @\seebelow@ {
(Index, State& state, Rcvr& rcvr, Tag, Args&&... args) noexcept
-> void requires @\seebelow@ {
if constexpr (@\libconcept{same_as}@<Tag, set_value_t>) {
auto& [shape, f] = state;
constexpr bool nothrow = noexcept(f(auto(shape), args...));
Expand All @@ -3669,32 +3754,109 @@
}
\end{codeblock}

\pnum
\par
The expression in the \grammarterm{requires-clause} of the lambda above
is \tcode{true} if and only
if \tcode{Tag} denotes a type other than \tcode{set_value_t} or
if the expression \tcode{f(auto(shape), args...)} is well-formed.

\pnum
Let the subexpression \tcode{out_sndr} denote
the result of the invocation \tcode{bulk(sndr, shape, f)} or
the result of the invocation
\tcode{\placeholder{bulk-algo}(sndr, policy, shape, f)} or
an object equal to such, and
let the subexpression \tcode{rcvr} denote a receiver
such that the expression \tcode{connect(out_sndr, rcvr)} is well-formed.
The expression \tcode{connect(out_sndr, rcvr)} has undefined behavior
unless it creates an asynchronous operation\iref{exec.async.ops} that,
when started,
when started:

\begin{itemize}
\item
on a value completion operation,
invokes \tcode{f(i, args...)}
for every \tcode{i} of type \tcode{Shape} in \range{\tcode{0}}{\tcode{shape}},
where \tcode{args} is a pack of lvalue subexpressions
referring to the value completion result datums of the input sender, and
If \tcode{sndr} has a successful completion, where
\tcode{args} is a pack of lvalue subexpressions
referring to the value completion result datums of \tcode{sndr}, or
decayed copies of those values if they model copy_constructible, then:

\begin{itemize}
\item
If \tcode{out_sndr} also completes successfully, then:

\begin{itemize}
\item
for \tcode{bulk},
invokes \tcode{f($i$, args...)} for every $i$ of type \tcode{Shape}
from \tcode{0} to \tcode{shape};

\item
for \tcode{bulk_unchunked},
invokes \tcode{f($i$, args...)} for every $i$ of type \tcode{Shape}
from \tcode{0} to \tcode{shape};

\recommended
The underlying scheduler should execute each iteration
on a distinct execution agent.

\item
for \tcode{bulk_chunked},
invokes \tcode{f($b$, $e$, args...)} zero or more times
with pairs of $b$ and $e$ of type \tcode{Shape}
in range \range{\tcode{0}}{\tcode{shape}},
such that $b < e$ and
for every $i$ of type \tcode{Shape} from \tcode{0} to \tcode{shape},
there is exactly one invocation with a pair $b$ and $e$,
such that $i$ is in the range \range{$b$}{$e$}.
\end{itemize}

\item
If \tcode{out_sndr} completes with \tcode{set_error(rcvr, eptr)}, then
the asynchronous operation may invoke a subset of
the invocations of \tcode{f}
before the error completion handler is called, and
\tcode{eptr} is an \tcode{exception_ptr} containing either:
\begin{itemize}
\item
an exception thrown by an invocation of \tcode{f}, or
\item
a \tcode{bad_alloc} exception if
the implementation fails to allocate required resources, or
\item
an exception derived from \tcode{runtime_error}.
\end{itemize}

\item
If \tcode{out_sndr} completes with \tcode{set_stopped(rcvr)}, then
the asynchronous operation may invoke a subset of
the invocations of \tcode{f}
before the stopped completion handler.
\end{itemize}

\item
propagates all completion operations sent by \tcode{sndr}.
If \tcode{sndr} does not complete with \tcode{set_value}, then
the completion is forwarded to \tcode{recv}.

\item
For \placeholder{bulk-algo},
the parameter \tcode{policy} describes
the manner in which
the execution of the asynchronous operations corresponding to these algorithms
may be parallelized and
the manner in which
%%FIXME: Should this be "apply to f"?
they apply \tcode{f}.
Permissions and requirements
on parallel algorithm element access functions\iref{algorithms.parallel.exec}
apply to \tcode{f}.
\end{itemize}

\pnum
\begin{note}
The asynchronous operation corresponding to
\tcode{\placeholder{bulk-algo}(sndr, policy, shape, f)}
can complete with \tcode{set_stopped} if cancellation is requested or
ignore cancellation requests.
\end{note}

\rSec3[exec.split]{\tcode{execution::split}}

\pnum
Expand Down