33 Execution control library [exec]

parallel_scheduler models scheduler.

Let sch be an object of type parallel_scheduler, and let BACKEND-OF(sch) be *ptr, where sch is associated with ptr.

The expression get_forward_progress_guarantee(sch) has the value forward_progress_guarantee​::​​parallel.

Let sch2 be an object of type parallel_scheduler.

Two objects sch and sch2 compare equal if and only if BACKEND-OF(sch) and BACKEND-OF(sch2) refer to the same object.

Let rcvr be a receiver.

A proxy for rcvr with base B is an lvalue r of type B such that:

• (5.1)
  r.set_value() has effects equivalent to set_value(std​::​move(rcvr)).
• (5.2)
  r.set_error(e), where e is an exception_ptr object, has effects equivalent to set_error(std​::​move(​rcvr), std​::​move(e)).
• (5.3)
  r.set_stopped() has effects equivalent to set_stopped(std​::​move(rcvr)).

A preallocated backend storage for a proxy r is an object s of type span<byte> such that the range s remains valid and may be overwritten until one of set_value, set_error, or set_stopped is called on r.

Let b be BACKEND-OF(sch), let sndr be the object returned by schedule(sch), and let rcvr be a receiver.

If rcvr is connected to sndr and the resulting operation state is started, then:

• (7.1)
  If sndr completes successfully, then b.schedule(r, s) is called, where

  • (7.1.1)
    r is a proxy for rcvr with base parallel_scheduler_replacement​::​receiver_proxy and
  • (7.1.2)
    s is a preallocated backend storage for r.
• (7.2)
  All other completion operations are forwarded unchanged.

The expression get_domain(sch) returns an expression of exposition-only type parallel-scheduler-domain, that is equivalent with:

For argument sndr of the above transform_sender, let child, pol, shape and f be defined by the following declarations: auto& [_, data, child] = sndr; auto& [pol, shape, f] = data;

Let p be

• (10.1)
  true, if the type of the expression pol is cv parallel_policy or cv parallel_unsequenced_policy;
• (10.2)
  implementation-defined, if pol is an implementation-defined execution policy;
• (10.3)
  false, otherwise.

The transform_sender overload that accepts senders with tag bulk_chunked_t returns a sender such that if it is connected to a receiver rcvr and the resulting operation state is started, then:

• (11.1)
  If child completes with values vals, let args be a pack of lvalue subexpressions designating vals, then b.schedule_bulk_chunked(p ? shape : 1, r, s) is called, where

  • (11.1.1)
    r is a proxy for rcvr with base parallel_scheduler_replacement​::​bulk_item_receiver_proxy such that r.execute(i, j) for indices i and j has effects equivalent to f(i, j, args...) if p is true and f(0, shape, args...) otherwise; and
  • (11.1.2)
    s is a preallocated backend storage for r.
• (11.2)
  All other completion operations are forwarded unchanged.

The transform_sender overload that accepts senders with tag bulk_unchunked_t returns a sender such that if it is connected to a receiver rcvr and the resulting operation state is started, then:

• (12.1)
  If child completes with values vals, let args be a pack of lvalue subexpressions designating vals, then b.schedule_bulk_unchunked(p ? shape : 1, r, s) is called, where

  • (12.1.1)
    r is a proxy for rcvr with base parallel_scheduler_replacement​::​bulk_item_receiver_proxy such that r.execute(i, i + 1) for index i has effects equivalent to f(i, args...) if p is true and for (decltype(shape) i = 0; i < shape; i++) { f(i, args...); } otherwise; and
  • (12.1.2)
    s is a preallocated backend storage for r.
• (12.2)
  All other completion operations are forwarded unchanged.