X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/360d1e6510190d279551d9ca0510badccb1f8b37..ea74f5d95928a521a588737e81f1de94eef25d19:/include/simgrid/s4u/Exec.hpp

diff --git a/include/simgrid/s4u/Exec.hpp b/include/simgrid/s4u/Exec.hpp
index b140bdbc61..754b4bd488 100644
--- a/include/simgrid/s4u/Exec.hpp
+++ b/include/simgrid/s4u/Exec.hpp
@@ -1,4 +1,4 @@
-/* Copyright (c) 2017. The SimGrid Team. All rights reserved.          */
+/* Copyright (c) 2017-2022. The SimGrid Team. All rights reserved.          */
 
 /* This program is free software; you can redistribute it and/or modify it
  * under the terms of the license (GNU LGPL) which comes with this package. */
@@ -8,42 +8,91 @@
 
 #include <simgrid/forward.h>
 #include <simgrid/s4u/Activity.hpp>
-#include <simgrid/s4u/forward.hpp>
-
-#include <atomic>
+#include <simgrid/s4u/Actor.hpp>
+#include <xbt/ex.h>
 
 namespace simgrid {
 namespace s4u {
 
-XBT_PUBLIC_CLASS Exec : public Activity
-{
-  Exec() : Activity() {}
+/** Computation Activity, representing the asynchronous executions.
+ *
+ * @beginrst
+ * Most of them are created with :cpp:func:`simgrid::s4u::this_actor::exec_init()` or
+ * :cpp:func:`simgrid::s4u::Host::execute()`, and represent a classical (sequential) execution. This can be used to
+ * simulate some computation occurring in another thread when the calling actor is not blocked during the execution.
+ *
+ * You can also use :cpp:func:`simgrid::s4u::this_actor::parallel_execute()` to create *parallel* executions. These
+ * objects represent distributed computations involving computations on several hosts and communications between them.
+ * Such objects can for example represent a matrix multiplication done with ScaLAPACK on a real system. Once created,
+ * parallel Exec are very similar to the sequential ones. The only difference is that you cannot migrate them, and their
+ * remaining amount of work can only be defined as a ratio. See the doc of :cpp:func:`simgrid::s4u::Exec::get_remaining`
+ * and :cpp:func:`simgrid::s4u::Exec::get_remaining_ratio` for more info.
+ * @endrst
+ */
+class XBT_PUBLIC Exec : public Activity_T<Exec> {
+#ifndef DOXYGEN
+  friend kernel::activity::ExecImpl;
+  friend kernel::EngineImpl; // Auto-completes the execs of maestro (in simDAG)
+#endif
+
+  bool parallel_ = false;
+
+protected:
+  explicit Exec(kernel::activity::ExecImplPtr pimpl);
+
+  void reset() const;
+
+  static xbt::signal<void(Exec const&)> on_start;
+
 public:
-  friend void intrusive_ptr_release(simgrid::s4u::Exec * e);
-  friend void intrusive_ptr_add_ref(simgrid::s4u::Exec * e);
-  friend ExecPtr this_actor::exec_init(double flops_amount);
+#ifndef DOXYGEN
+  Exec(Exec const&) = delete;
+  Exec& operator=(Exec const&) = delete;
+#endif
+  /*! Signal fired each time that an execution actually starts (no veto) */
+  static void on_start_cb(const std::function<void(Exec const&)>& cb) { on_start.connect(cb); }
+
+  static ExecPtr init();
+  Exec* start() override;
+
+  /*! take a vector of s4u::ExecPtr and return when one of them is finished.
+   * The return value is the rank of the first finished ExecPtr. */
+  static ssize_t wait_any(const std::vector<ExecPtr>& execs) { return wait_any_for(execs, -1); }
+  /*! Same as wait_any, but with a timeout. If the timeout occurs, parameter last is returned.*/
+  static ssize_t wait_any_for(const std::vector<ExecPtr>& execs, double timeout);
 
-  ~Exec() = default;
+#ifndef DOXYGEN
+  XBT_ATTRIB_DEPRECATED_v332("Please use a plain vector for parameter")
+  static int wait_any(std::vector<ExecPtr>* execs) { return static_cast<int>(wait_any_for(*execs, -1)); }
+  XBT_ATTRIB_DEPRECATED_v332("Please use a plain vector for first parameter")
+  static int wait_any_for(std::vector<ExecPtr>* execs, double timeout) { return static_cast<int>(wait_any_for(*execs, timeout)); }
+#endif
 
-  void start() override;
-  void wait() override;
-  void wait(double timeout) override;
-  bool test();
+  /** @brief On sequential executions, returns the amount of flops that remain to be done; This cannot be used on
+   * parallel executions. */
+  double get_remaining() const override;
+  double get_remaining_ratio() const;
+  ExecPtr set_host(Host* host);
+  ExecPtr set_hosts(const std::vector<Host*>& hosts);
+  ExecPtr unset_host();
+  ExecPtr unset_hosts() { return unset_host(); }
 
-  ExecPtr setPriority(double priority);
-  ExecPtr setHost(Host * host);
+  ExecPtr set_flops_amount(double flops_amount);
+  ExecPtr set_flops_amounts(const std::vector<double>& flops_amounts);
+  ExecPtr set_bytes_amounts(const std::vector<double>& bytes_amounts);
 
-  double getRemains() override;
-  double getRemainingRatio();
+  ExecPtr set_bound(double bound);
+  ExecPtr set_priority(double priority);
+  ExecPtr update_priority(double priority);
 
-private:
-  Host* host_          = nullptr;
-  double flops_amount_ = 0.0;
-  double priority_     = 1.0;
+  Host* get_host() const;
+  unsigned int get_host_number() const;
+  double get_cost() const;
+  bool is_parallel() const { return parallel_; }
+  bool is_assigned() const override;
+};
 
-  std::atomic_int_fast32_t refcount_{0};
-}; // class
-}
-}; // Namespace simgrid::s4u
+} // namespace s4u
+} // namespace simgrid
 
 #endif /* SIMGRID_S4U_EXEC_HPP */