From: Wolfgang Bangerth Date: Thu, 25 Sep 2014 22:51:52 +0000 (-0500) Subject: Improve documentation of TimerOutput. X-Git-Tag: v8.2.0-rc1~128^2 X-Git-Url: https://gitweb.dealii.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=refs%2Fpull%2F173%2Fhead;p=dealii.git Improve documentation of TimerOutput. --- diff --git a/include/deal.II/base/timer.h b/include/deal.II/base/timer.h index 6a7df44962..c41877d221 100644 --- a/include/deal.II/base/timer.h +++ b/include/deal.II/base/timer.h @@ -1,6 +1,6 @@ // --------------------------------------------------------------------- // -// Copyright (C) 1998 - 2013 by the deal.II authors +// Copyright (C) 1998 - 2014 by the deal.II authors // // This file is part of the deal.II library. // @@ -106,7 +106,7 @@ public: * compiler. */ Timer (MPI_Comm mpi_communicator, - bool sync_wall_time = false); + const bool sync_wall_time = false); /** * Returns a reference to the data @@ -364,7 +364,56 @@ private: * @endcode * * - * See the step-32 tutorial program for usage of this class. + *

Usage in parallel programs using MPI

+ * + * In a parallel program built on MPI, using the class in a way such + * as the one shown above would result in a situation where each + * process times the corresponding sections and then outputs the resulting + * timing information at the end. This is annoying since you'd get a lot + * of output -- one set of timing information from each processor. + * + * This can be avoided by only letting one processor generate screen output, + * typically by using an object of type ConditionalOStream instead of + * std::cout to write to screen (see, for example, step-17, + * step-18, step-32 and step-40, all of which use this method). + * + * This way, only a single processor outputs timing information, typically + * the first process in the MPI universe. However, + * if you take the above code snippet as an example, imagine what would happen + * if setup_dofs() is fast on processor zero and slow on at + * least one of the other processors; and if the first thing + * assemble_system_1() does is something that requires all + * processors to communicate. In this case, on processor zero, the timing + * section with name "Setup dof system" will yield a short + * run time on processor zero, whereas the section "Assemble" + * will take a long time: not because assemble_system_1() + * takes a particularly long time, but because on the processor on which + * we time (or, rather, the one on which we generate output) happens to + * have to wait for a long time till the other processor is finally done + * with setup_dofs() and starts to participate in + * assemble_system_1(). In other words, the timing that is + * reported is unreliable because it reflects run times from other + * processors. Furthermore, the run time of this section on processor zero + * has nothing to do with the run time of the section on other processors + * but instead with the run time of the previous section on another + * processor. + * + * The usual way to avoid this is to introduce a barrier into the parallel + * code just before we start and stop timing sections. This ensures that + * all processes are at the same place and the timing information then + * reflects the maximal run time across all processors. To achieve this, + * you need to initialize the TimerOutput object with an MPI communicator + * object, for example as in the following code: + * @code + * TimerOutput timer (MPI_COMM_WORLD, + * pcout, + * TimerOutput::summary, + * TimerOutput::wall_times); + * @endcode + * Here, pcout is an object of type ConditionalOStream that + * makes sure that we only generate output on a single processor. + * See the step-32 and step-40 tutorial programs for this kind of usage + * of this class. * * @ingroup utilities * @author M. Kronbichler, 2009. @@ -459,12 +508,12 @@ public: #ifdef DEAL_II_WITH_MPI /** - * Constructor that takes an MPI - * communicator as input. A timer - * constructed this way will sum up the - * CPU times over all processors in the - * MPI network for calculating the CPU - * time. + * Constructor that takes an MPI communicator as input. A timer + * constructed this way will sum up the CPU times over all + * processors in the MPI network for calculating the CPU time, or + * take the maximum over all processors, depending on the value of + * @p output_type . See the documentation of this class for the + * rationale for this constructor and an example. * * @param mpi_comm An MPI communicator across which we should accumulate or * otherwise synchronize the timing information we produce on every MPI process. @@ -472,7 +521,13 @@ public: * @param output_frequency A variable indicating when output is to be written * to the given stream. * @param output_type A variable indicating what kind of timing the output - * should represent (CPU or wall time). + * should represent (CPU or wall time). In this parallel context, when this + * argument selects CPU time, then times are accumulated over all processes + * participating in the MPI communicator. If this argument selects wall + * time, then reported times are the maximum over all processors' run times + * for this section. (The latter is computed by placing an + * MPI_Barrier call before starting and stopping the timer + * for each section. */ TimerOutput (MPI_Comm mpi_comm, std::ostream &stream, @@ -480,12 +535,12 @@ public: const enum OutputType output_type); /** - * Constructor that takes an MPI - * communicator as input. A timer - * constructed this way will sum up the - * CPU times over all processors in the - * MPI network for calculating the CPU - * time. + * Constructor that takes an MPI communicator as input. A timer + * constructed this way will sum up the CPU times over all + * processors in the MPI network for calculating the CPU time, or + * take the maximum over all processors, depending on the value of + * @p output_type . See the documentation of this class for the + * rationale for this constructor and an example. * * @param mpi_comm An MPI communicator across which we should accumulate or * otherwise synchronize the timing information we produce on every MPI process. @@ -493,7 +548,13 @@ public: * @param output_frequency A variable indicating when output is to be written * to the given stream. * @param output_type A variable indicating what kind of timing the output - * should represent (CPU or wall time). + * should represent (CPU or wall time). In this parallel context, when this + * argument selects CPU time, then times are accumulated over all processes + * participating in the MPI communicator. If this argument selects wall + * time, then reported times are the maximum over all processors' run times + * for this section. (The latter is computed by placing an + * MPI_Barrier call before starting and stopping the timer + * for each section.) */ TimerOutput (MPI_Comm mpi_comm, ConditionalOStream &stream, diff --git a/source/base/timer.cc b/source/base/timer.cc index d62d48b631..7a24d3aabe 100644 --- a/source/base/timer.cc +++ b/source/base/timer.cc @@ -1,6 +1,6 @@ // --------------------------------------------------------------------- // -// Copyright (C) 1998 - 2013 by the deal.II authors +// Copyright (C) 1998 - 2014 by the deal.II authors // // This file is part of the deal.II library. // @@ -354,6 +354,10 @@ TimerOutput::enter_subsection (const std::string §ion_name) #ifdef DEAL_II_WITH_MPI if (mpi_communicator != MPI_COMM_SELF) { + // create a new timer for this section. the second argument + // will ensure that we have an MPI barrier before starting + // and stopping a timer, and this ensures that we get the + // maximum run time for this section over all processors sections[section_name].timer = Timer(mpi_communicator, true); } #endif @@ -400,11 +404,9 @@ TimerOutput::leave_subsection (const std::string §ion_name) sections[actual_section_name].total_wall_time += sections[actual_section_name].timer.wall_time(); - // get cpu time. on MPI systems, add - // the local contributions. we could - // do that also in the Timer class - // itself, but we didn't initialize - // the Timers here according to that + // get cpu time. on MPI systems, add the local contributions. we + // could do that also in the Timer class itself, but we didn't + // initialize the Timers here according to that double cpu_time = sections[actual_section_name].timer(); sections[actual_section_name].total_cpu_time += (Utilities::System::job_supports_mpi() @@ -413,8 +415,7 @@ TimerOutput::leave_subsection (const std::string §ion_name) : cpu_time); - // in case we have to print out - // something, do that here... + // in case we have to print out something, do that here... if ((output_frequency == every_call || output_frequency == every_call_and_summary) && output_is_enabled == true) {