Skip to content

[2026-03 LWG Motion 25] P3804R2 Iterating on parallel_scheduler #8912

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
eisenwave wants to merge 2 commits into
base: main
Choose a base branch
from
Open

[2026-03 LWG Motion 25] P3804R2 Iterating on parallel_scheduler #8912

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
6fe6862
P3804R2 Iterating on parallel_scheduler
eisenwave Apr 3, 2026
1dd037c
Fixup: grammar
eisenwave Apr 3, 2026
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
133 changes: 88 additions & 45 deletions source/exec.tex
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8494,24 +8494,6 @@
for operations launched via calls to \tcode{parallel_scheduler_backend}.
\end{note}

\pnum
A \defnx{bulk chunked proxy for \tcode{rcvr}
with callable \tcode{f} and arguments \tcode{args}}{proxy!bulk chunked}
is a proxy \tcode{r} for \tcode{rcvr}
with base \tcode{system_context_replaceability::bulk_item_receiver_proxy}
such that
\tcode{r.execute(i, j)} for indices \tcode{i} and \tcode{j}
has effects equivalent to \tcode{f(i, j, args...)}.

\pnum
A \defnx{bulk unchunked proxy for \tcode{rcvr}
with callable \tcode{f} and arguments \tcode{args}}{proxy!bulk unchunked}
is a proxy \tcode{r} for \tcode{rcvr}
with base \tcode{system_context_replaceability::bulk_item_receiver_proxy}
such that
\tcode{r.execute(i, i + 1)} for index \tcode{i}
has effects equivalent to \tcode{f(i, args...)}.

\pnum
Let \tcode{b} be \tcode{\exposid{BACKEND-OF}(sch)},
let \tcode{sndr} be the object returned by \tcode{schedule(sch)}, and
Expand All @@ -8534,52 +8516,100 @@
\end{itemize}

\pnum
\tcode{parallel_scheduler} provides a customized implementation of
the \tcode{bulk_chunked} algorithm\iref{exec.bulk}.
If a receiver \tcode{rcvr} is connected to the sender
returned by \tcode{bulk_chunked(sndr, pol, shape, f)} and
The expression \tcode{get_domain(sch)}
returns an expression of exposition-only type
\exposid{parallel-scheduler-do\-main}, that is equivalent to:

\begin{itemdecl}
struct @\exposid{parallel-scheduler-domain}@ {
template<@\exposconcept{sender-for}@<bulk_chunked_t> Sndr, @\exposconcept{queryable}@ Env>
static constexpr decltype(auto)
transform_sender(set_value_t, Sndr&& sndr, const Env& env) const noexcept {
return @\seebelow@;
}
template<@\exposconcept{sender-for}@<bulk_unchunked_t> Sndr, @\exposconcept{queryable}@ Env>
static constexpr decltype(auto)
transform_sender(set_value_t, Sndr&& sndr, const Env& env) const noexcept {
return @\seebelow@;
}
};
\end{itemdecl}

\begin{itemdescr}
\pnum
For argument \tcode{sndr} of the above \tcode{transform_sender},
let \tcode{child}, \tcode{pol}, \tcode{shape} and \tcode{f}
be defined by the following declarations:
\begin{codeblock}
auto& [_, data, child] = sndr;
auto& [pol, shape, f] = data;
\end{codeblock}

\pnum
Let \tcode{p} be:
\begin{itemize}
\item
\tcode{true}, if the type of the expression \tcode{pol} is
\cv{}~\tcode{parallel_policy} or \cv{}~\tcode{parallel_unsequenced_policy};
\item
implementation-defined, if \tcode{pol} is an implementation-defined execution policy;
\item
\tcode{false}, otherwise.
\end{itemize}

\pnum
The \tcode{transform_sender} overload
that accepts senders with tag \tcode{bulk_chunked_t}
returns a sender such that if it is connected to a receiver \tcode{rcvr} and
the resulting operation state is started, then:
\begin{itemize}
\item
If \tcode{sndr} completes with values \tcode{vals},
If \tcode{child} completes with values \tcode{vals},
let \tcode{args} be a pack of lvalue subexpressions designating \tcode{vals},
then \tcode{b.schedule_bulk_chunked(shape, r, s)} is called, where
then \tcode{b.schedule_bulk_chunked(p ? shape : 1, r, s)} is called, where
\begin{itemize}
\item
\tcode{r} is a bulk chunked proxy for \tcode{rcvr}
with callable \tcode{f} and arguments \tcode{args} and
\tcode{r} is a proxy for \tcode{rcvr}
with base \tcode{system_context_replaceability::bulk_item_receiver_proxy}
such that \tcode{r.execute(i, j)} for indices \tcode{i} and \tcode{j}
has effects equivalent to
\tcode{f(i, j, args...)} if \tcode{p} is true and
\tcode{f(0, shape, args...)} otherwise; and
\item
\tcode{s} is a preallocated backend storage for \tcode{r}.
\end{itemize}
\item
All other completion operations are forwarded unchanged.
\end{itemize}
\begin{note}
Customizing the behavior of \tcode{bulk_chunked}
affects the default implementation of \tcode{bulk}.
\end{note}

\pnum
\tcode{parallel_scheduler} provides a customized implementation of
the \tcode{bulk_unchunked} algorithm\iref{exec.bulk}.
If a receiver \tcode{rcvr} is connected to the sender
returned by \tcode{bulk_unchunked(sndr, pol, shape, f)} and
The \tcode{transform_sender} overload
that accepts senders with tag \tcode{bulk_unchunked_t}
returns a sender such that if it is connected to a receiver \tcode{rcvr} and
the resulting operation state is started, then:
\begin{itemize}
\item
If \tcode{sndr} completes with values \tcode{vals},
If \tcode{child} completes with values \tcode{vals},
let \tcode{args} be a pack of lvalue subexpressions designating \tcode{vals},
then \tcode{b.schedule_bulk_unchunked(shape, r, s)} is called, where
then \tcode{b.schedule_bulk_unchunked(p ? shape : 1, r, s)} is called, where
\begin{itemize}
\item
\tcode{r} is a bulk unchunked proxy for \tcode{rcvr}
with callable \tcode{f} and arguments \tcode{args} and
\tcode{r} is a proxy for \tcode{rcvr}
with base \tcode{system_context_replaceability::bulk_item_receiver_proxy}
such that \tcode{r.execute(i, i + 1)} for index \tcode{i}
has effects equivalent to
\tcode{f(i, args...)} if \tcode{p} is true and
\begin{codeblock}
for (decltype(shape) i = 0; i < shape; i++) { f(i, args...); }
\end{codeblock}
otherwise; and
\item
\tcode{s} is a preallocated backend storage for \tcode{r}.
\end{itemize}
\item
All other completion operations are forwarded unchanged.
\end{itemize}
\end{itemdescr}

\indexlibraryglobal{get_parallel_scheduler}%
\begin{itemdecl}
Expand Down Expand Up @@ -8609,8 +8639,6 @@
\begin{codeblock}
namespace std::execution::system_context_replaceability {
struct @\libglobal{receiver_proxy}@ {
virtual ~receiver_proxy() = default;

protected:
virtual bool @\exposidnc{query-env}@(@\unspecnc@) noexcept = 0; // \expos

Expand All @@ -8620,7 +8648,7 @@
virtual void set_stopped() noexcept = 0;

template<class P, @\exposconcept{class-type}@ Query>
optional<P> try_query(Query q) noexcept;
optional<P> try_query(Query q) const noexcept;
};

struct @\libglobal{bulk_item_receiver_proxy}@ : receiver_proxy {
Expand All @@ -8642,7 +8670,7 @@

\begin{itemdecl}
template<class P, @\exposconcept{class-type}@ Query>
optional<P> @\libglobal{try_query}@(Query q) noexcept;
optional<P> @\libglobal{try_query}@(Query q) const noexcept;
\end{itemdecl}

\begin{itemdescr}
Expand All @@ -8662,11 +8690,13 @@
\tcode{P} is not a member of an implementation-defined set
of supported result types for \tcode{Query}; or
\item
the expression \tcode{q(env)} is not well-formed or
does not have type \cv{} \tcode{P},
the expression \tcode{q(env)} is not well-formed,
\end{itemize}
then returns \tcode{nullopt}.
Otherwise, returns \tcode{q(env)}.
Otherwise, if \tcode{q(env)} has type \cv{}~\tcode{P},
then returns \tcode{q(env)}.
Otherwise, returns an implementation-defined value
of type \tcode{optional<P>}.

\pnum
\remarks
Expand All @@ -8675,6 +8705,19 @@
\tcode{inplace_stop_token} is a member
of the implementation-defined set of supported result types
for \tcode{get_stop_token_t}.

\pnum
\recommended
If \tcode{P} is \tcode{inplace_stop_token} and
\tcode{T} is a type other than \tcode{inplace_stop_token}
that models \libconcept{stoppable_token},
\tcode{try_query} should return an object of type \tcode{inplace_stop_token}
such that until one of
\tcode{set_value},
\tcode{set_error} or
\tcode{set_stopped}
is called on \tcode{*this},
all calls to \tcode{try_query} return equivalent \tcode{inplace_stop_token} objects.
\end{itemdescr}

\rSec2[exec.sysctxrepl.query]{\tcode{query_parallel_scheduler_backend}}
Expand Down
Loading