cs_parall.h

Go to the documentation of this file.

#ifndef CS_PARALL_H
#define CS_PARALL_H
 
/*============================================================================
 * Functions dealing with parallelism
 *============================================================================*/
 
/*
  This file is part of code_saturne, a general-purpose CFD tool.
 
  Copyright (C) 1998-2026 EDF S.A.
 
  This program is free software; you can redistribute it and/or modify it under
  the terms of the GNU General Public License as published by the Free Software
  Foundation; either version 2 of the License, or (at your option) any later
  version.
 
  This program is distributed in the hope that it will be useful, but WITHOUT
  ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
  FOR A PARTICULAR PURPOSE.  See the GNU General Public License for more
  details.
 
  You should have received a copy of the GNU General Public License along with
  this program; if not, write to the Free Software Foundation, Inc., 51 Franklin
  Street, Fifth Floor, Boston, MA 02110-1301, USA.
*/
 
/*----------------------------------------------------------------------------*/
 
/*----------------------------------------------------------------------------
 *  Local headers
 *----------------------------------------------------------------------------*/
 
#include "base/cs_defs.h"
#include "base/cs_execution_context.h"
 
/*----------------------------------------------------------------------------*/
 
BEGIN_C_DECLS
 
/*============================================================================
 * General types and macros used throughout code_saturne
 *============================================================================*/
 
/*----------------------------------------------------------------------------
 * Variable value type.
 *----------------------------------------------------------------------------*/
 
typedef enum {
 
  CS_E2N_SUM_SCATTER,  
  CS_E2N_SUM_GATHER    
} cs_e2n_sum_t;
 
/*============================================================================
 * Global variables
 *============================================================================*/
 
/* Preferred indexed sum option, adapted to shared-memory parallelism */
 
extern cs_e2n_sum_t cs_glob_e2n_sum_type;
 
/*=============================================================================
 * Public function prototypes
 *============================================================================*/
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
inline static void
cs_parall_counter(cs_gnum_t   cpt[],
                  const int   n)
{
#if defined(HAVE_MPI)
 
  if (cs_glob_n_ranks > 1) {
    MPI_Allreduce(MPI_IN_PLACE, cpt, n, CS_MPI_GNUM, MPI_SUM,
                  cs_glob_mpi_comm);
  }
 
#else
 
  CS_UNUSED(cpt);
  CS_UNUSED(n);
 
#endif
}
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
inline static void
cs_parall_counter_max(cs_lnum_t   cpt[],
                      const int   n)
{
#if defined(HAVE_MPI)
 
  if (cs_glob_n_ranks > 1) {
    MPI_Allreduce(MPI_IN_PLACE, cpt, n, CS_MPI_LNUM, MPI_MAX,
                  cs_glob_mpi_comm);
  }
 
#else
 
  CS_UNUSED(cpt);
  CS_UNUSED(n);
 
#endif
}
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
inline static void
cs_parall_sum(int             n,
              cs_datatype_t   datatype,
              void           *val)
{
#if defined(HAVE_MPI)
 
  if (cs_glob_n_ranks > 1) {
    MPI_Allreduce(MPI_IN_PLACE, val, n, cs_datatype_to_mpi[datatype], MPI_SUM,
                  cs_glob_mpi_comm);
  }
 
#else
 
  CS_UNUSED(n);
  CS_UNUSED(datatype);
  CS_UNUSED(val);
 
#endif
}
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
inline static void
cs_parall_max(int             n,
              cs_datatype_t   datatype,
              void           *val)
{
#if defined(HAVE_MPI)
 
  if (cs_glob_n_ranks > 1) {
    MPI_Allreduce(MPI_IN_PLACE, val, n, cs_datatype_to_mpi[datatype], MPI_MAX,
                  cs_glob_mpi_comm);
  }
 
#else
 
  CS_UNUSED(n);
  CS_UNUSED(datatype);
  CS_UNUSED(val);
 
#endif
}
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
inline static void
cs_parall_min(int             n,
              cs_datatype_t   datatype,
              void           *val)
{
#if defined(HAVE_MPI)
 
  if (cs_glob_n_ranks > 1) {
    MPI_Allreduce(MPI_IN_PLACE, val, n, cs_datatype_to_mpi[datatype], MPI_MIN,
                  cs_glob_mpi_comm);
  }
 
#else
 
  CS_UNUSED(n);
  CS_UNUSED(datatype);
  CS_UNUSED(val);
 
#endif
}
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
inline static void
cs_parall_bcast(int             root_rank,
                int             n,
                cs_datatype_t   datatype,
                void           *val)
{
#if defined(HAVE_MPI)
 
  if (cs_glob_n_ranks > 1)
    MPI_Bcast(val, n, cs_datatype_to_mpi[datatype], root_rank,
              cs_glob_mpi_comm);
 
#else
 
  CS_UNUSED(root_rank);
  CS_UNUSED(n);
  CS_UNUSED(datatype);
  CS_UNUSED(val);
 
#endif
}
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
void
cs_parall_allgather_r(int        n_elts,
                      int        n_g_elts,
                      cs_real_t  array[],
                      cs_real_t  g_array[]);
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
void
cs_parall_allgather_ordered_r(int        n_elts,
                              int        n_g_elts,
                              int        stride,
                              cs_real_t  o_key[],
                              cs_real_t  array[],
                              cs_real_t  g_array[]);
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
void
cs_parall_gather_r(int               root_rank,
                   int               n_elts,
                   int               n_g_elts,
                   const cs_real_t   array[],
                   cs_real_t         g_array[]);
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
void
cs_parall_gather_ordered_r(int        root_rank,
                           int        n_elts,
                           int        n_g_elts,
                           int        stride,
                           cs_real_t  o_key[],
                           cs_real_t  array[],
                           cs_real_t  g_array[]);
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
void
cs_parall_scatter_r(int               root_rank,
                    int               n_elts,
                    int               n_g_elts,
                    const cs_real_t   g_array[],
                    cs_real_t         array[]);
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
void
cs_parall_gather_f(int             root_rank,
                   int             n_elts,
                   int             n_g_elts,
                   const float     array[],
                   float           g_array[]);
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
void
cs_parall_scatter_f(int           root_rank,
                    int           n_elts,
                    int           n_g_elts,
                    const float   g_array[],
                    float         array[]);
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
void
cs_parall_max_loc_vals(int         n,
                       cs_real_t  *max,
                       cs_real_t   max_loc_vals[]);
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
void
cs_parall_min_loc_vals(int         n,
                       cs_real_t  *min,
                       cs_real_t   min_loc_vals[]);
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
void
cs_parall_min_id_rank_r(cs_lnum_t  *elt_id,
                        int        *rank_id,
                        cs_real_t   val);
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
size_t
cs_parall_get_min_coll_buf_size(void);
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
void
cs_parall_set_min_coll_buf_size(size_t buffer_size);
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
inline static int
cs_parall_n_threads(cs_lnum_t  n_elements,
                    cs_lnum_t  min_thread_elements)
{
#if defined(HAVE_OPENMP)
  int n_t = omp_get_max_threads();
  int n_t_l = n_elements / min_thread_elements;
  if (n_t_l < n_t)
    n_t = n_t_l;
  if (n_t < 1)
    n_t = 1;
  return n_t;
#else
  CS_UNUSED(n_elements);         /* avoid compiler warning */
  CS_UNUSED(min_thread_elements);
  return 1;
#endif
}
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
inline static void
cs_parall_thread_range(cs_lnum_t    n,
                       size_t       type_size,
                       cs_lnum_t   *s_id,
                       cs_lnum_t   *e_id)
{
#if defined(HAVE_OPENMP)
  const int t_id = omp_get_thread_num();
  const int n_t = omp_get_num_threads();
  const cs_lnum_t t_n = (n + n_t - 1) / n_t;
  const cs_lnum_t cl_m = CS_CL_SIZE / type_size;  /* Cache line multiple */
 
  *s_id =  t_id    * t_n;
  *e_id = (t_id+1) * t_n;
  *s_id = cs_align(*s_id, cl_m);
  *e_id = cs_align(*e_id, cl_m);
  if (*e_id > n) *e_id = n;
#else
  CS_UNUSED(type_size);         /* avoid compiler warning */
  *s_id = 0;
  *e_id = n;
#endif
}
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
inline static void
cs_parall_thread_range_upper(cs_lnum_t    n,
                             size_t       type_size,
                             cs_lnum_t   *s_id,
                             cs_lnum_t   *e_id)
{
#if defined(HAVE_OPENMP)
  const int t_id = omp_get_thread_num();
  const double n_t = omp_get_num_threads();
  const cs_lnum_t cl_m = CS_CL_SIZE / type_size;  /* Cache line multiple */
 
  double r0 = (double)t_id / (double)n_t;
  double r1 = (double)(t_id+1) / (double)n_t;
 
  r0 = r0*r0;
  r1 = r1*r1;
 
  const cs_lnum_t t_0 = r0*n;
  const cs_lnum_t t_1 = r1*n;
 
  *s_id = t_0 * n;
  *e_id = t_1 * n;
  *s_id = cs_align(*s_id, cl_m);
  *e_id = cs_align(*e_id, cl_m);
  if (*e_id > n) *e_id = n;
#else
  CS_UNUSED(type_size);         /* avoid compiler warning */
  *s_id = 0;
  *e_id = n;
#endif
}
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
static inline size_t
cs_parall_block_count(size_t  n,
                      size_t  block_size)
{
  return (n % block_size) ?  n/block_size + 1 : n/block_size;
}
 
/*----------------------------------------------------------------------------*/
 
END_C_DECLS
 
#if defined(__cplusplus)
 
/*=============================================================================
 * Public C++ functions
 *============================================================================*/
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
inline static void
cs_parall_counter
(
  [[maybe_unused]] const cs_mpi_wrapper& mpi_w,
  [[maybe_unused]] cs_gnum_t             cpt[],
  [[maybe_unused]] const int             n)
{
#if defined(HAVE_MPI)
 
  if (mpi_w.active()) {
    MPI_Allreduce(MPI_IN_PLACE, cpt, n, CS_MPI_GNUM, MPI_SUM,
                  mpi_w.comm());
  }
 
#endif
}
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
inline static void
cs_parall_counter_max
(
  [[maybe_unused]] const cs_mpi_wrapper& mpi_w,
  [[maybe_unused]] cs_lnum_t             cpt[],
  [[maybe_unused]] const int             n
)
{
#if defined(HAVE_MPI)
 
  if (mpi_w.active()) {
    MPI_Allreduce(MPI_IN_PLACE, cpt, n, CS_MPI_LNUM, MPI_MAX,
                  mpi_w.comm());
  }
 
#endif
}
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
inline static void
cs_parall_sum
(
  [[maybe_unused]] const cs_mpi_wrapper&  mpi_w,
  [[maybe_unused]] int                    n,
  [[maybe_unused]] cs_datatype_t          datatype,
  [[maybe_unused]] void                  *val
)
{
#if defined(HAVE_MPI)
 
  if (mpi_w.active()) {
    MPI_Allreduce(MPI_IN_PLACE, val, n,
                  cs_datatype_to_mpi[datatype], MPI_SUM,
                  mpi_w.comm());
  }
 
#endif
}
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
inline static void
cs_parall_max
(
  [[maybe_unused]] const cs_mpi_wrapper&  mpi_w,
  [[maybe_unused]] int                    n,
  [[maybe_unused]] cs_datatype_t          datatype,
  [[maybe_unused]] void                  *val
)
{
#if defined(HAVE_MPI)
 
  if (mpi_w.active()) {
    MPI_Allreduce(MPI_IN_PLACE, val, n,
                  cs_datatype_to_mpi[datatype], MPI_MAX,
                  mpi_w.comm());
  }
 
#endif
}
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
inline static void
cs_parall_min
(
  [[maybe_unused]] const cs_mpi_wrapper&  mpi_w,
  [[maybe_unused]] int                    n,
  [[maybe_unused]] cs_datatype_t          datatype,
  [[maybe_unused]] void                  *val
)
{
#if defined(HAVE_MPI)
 
  if (mpi_w.active()) {
    MPI_Allreduce(MPI_IN_PLACE, val, n,
                  cs_datatype_to_mpi[datatype], MPI_MIN,
                  mpi_w.comm());
  }
 
#endif
}
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
inline static void
cs_parall_thread_range(cs_lnum_t    n,
                       size_t       type_size,
                       int          t_id,
                       int          n_t,
                       cs_lnum_t   *s_id,
                       cs_lnum_t   *e_id)
{
  const cs_lnum_t t_n = (n + n_t - 1) / n_t;
  const cs_lnum_t cl_m = CS_CL_SIZE / type_size;  /* Cache line multiple */
 
  *s_id =  t_id    * t_n;
  *e_id = (t_id+1) * t_n;
  *s_id = cs_align(*s_id, cl_m);
  *e_id = cs_align(*e_id, cl_m);
  if (*s_id > n) *s_id = n;
  if (*e_id > n) *e_id = n;
}
 
/*=============================================================================
 * Public C++ templates
 *============================================================================*/
 
namespace cs {
namespace parall {
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
template <typename T,
          typename... Vals,
          typename std::enable_if<!std::is_array<T>::value, int>::type = 0>
static void
sum
(
  const cs_mpi_wrapper& mpi_w, 
  T&                    first, 
  Vals&...              values 
)
{
#if defined(HAVE_MPI)
  /* If no parallel ranks exit the function */
  if (!mpi_w.active())
    return;
 
  /* Count number of values */
  constexpr int n_vals = sizeof...(Vals);
 
  /* Set datatype for global communication */
  cs_datatype_t datatype = cs_datatype_from_type<T>();
 
  /* Temporary work array and parallel sum */
  if (n_vals == 0)
    cs_parall_sum(mpi_w, 1, datatype, &first);
  else {
    /* Unpack values */
    T *_values[] = {&values ...};
 
    T w[n_vals + 1];
    w[0] = first;
    for (int i = 0; i < n_vals; i++)
      w[i+1] = *(_values[i]);
 
    cs_parall_sum(mpi_w, n_vals + 1, datatype, w);
 
    first = w[0];
    for (int i = 0; i < n_vals; i++)
      *(_values[i]) = w[i+1];
  }
#else
  return;
#endif
}
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
template <typename T,
          typename... Vals,
          typename std::enable_if<!std::is_array<T>::value, int>::type = 0>
static void
sum(T &first,       
    Vals &...values 
)
{
#if defined(HAVE_MPI)
  const auto& mpi_w = cs::execution::default_mpi();
  sum<T>(mpi_w, first, values...);
#else
  return;
#endif
}
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
template <int Stride, typename T, typename... Vals>
static void
sum
(
  const cs_mpi_wrapper& mpi_w,   
  T                     first[], 
  Vals&&...             values   
)
{
#if defined(HAVE_MPI)
 
  if (!mpi_w.active())
    return;
 
  /* Count number of values */
  constexpr int n_vals = sizeof...(Vals);
 
  /* Set datatype for global communication */
  cs_datatype_t datatype = cs_datatype_from_type<T>();
 
  /* Temporary work array and parallel sum */
  if (n_vals == 0)
    cs_parall_sum(mpi_w, Stride, datatype, first);
  else {
    /* Unpack values */
    T *_values[] = {values ...};
 
    constexpr int work_size = (n_vals + 1) * Stride;
 
    T w[work_size];
    for (int i = 0; i < Stride; i++)
      w[i] = first[i];
 
    for (int i = 0; i < n_vals; i++) {
      for (int j = 0; j < Stride; j++)
        w[(i+1)*Stride + j] = _values[i][j];
    }
 
    cs_parall_sum(mpi_w, work_size, datatype, w);
 
    for (int i = 0; i < Stride; i++)
      first[i] = w[i];
 
    for (int i = 0; i < n_vals; i++) {
      for (int j = 0; j < Stride; j++)
        _values[i][j] = w[(i+1)*Stride + j];
    }
  }
 
#endif
}
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
template <int Stride, typename T, typename... Vals>
static void
sum
(
  T         first[], 
  Vals&&... values   
)
{
#if defined(HAVE_MPI)
  const auto& mpi_w = cs::execution::default_mpi();
  sum<Stride, T>(mpi_w, first, values...);
#else
  return;
#endif
}
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
template <typename T, int N, typename... Vals>
static void
sum
(
  T (&first)[N],   
  Vals &&...values 
)
{
  // Appel à la version existante avec Stride = N
  sum<N>(first, std::forward<Vals>(values)...);
}
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
template <typename T, int N, typename... Vals>
static void
sum
(
  const cs_mpi_wrapper& mpi_w,       
  T                     (&first)[N], 
  Vals &&...            values       
)
{
  sum<N>(mpi_w, first, std::forward<Vals>(values)...);
}
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
template <typename T, typename... Vals>
static void
max
(
  const cs_mpi_wrapper& mpi_w, 
  T&                    first, 
  Vals&...              values 
)
{
#if defined(HAVE_MPI)
 
  /* Count number of values */
  constexpr int n_vals = sizeof...(Vals);
 
  /* Set datatype for global communication */
  cs_datatype_t datatype = cs_datatype_from_type<T>();
 
  /* Temporary work array and parallel sum */
  if (n_vals == 0)
    cs_parall_max(mpi_w, 1, datatype, &first);
  else {
 
    /* Unpack values */
    T *_values[] = {&values ...};
 
    T w[n_vals + 1];
    w[0] = first;
    for (int i = 0; i < n_vals; i++)
      w[i+1] = *(_values[i]);
 
    cs_parall_max(mpi_w, n_vals + 1, datatype, w);
 
    first = w[0];
    for (int i = 0; i < n_vals; i++)
      *(_values[i]) = w[i+1];
  }
 
#endif // defined(HAVE_MPI)
}
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
template <typename T, typename... Vals>
static void
max
(
  T&       first, 
  Vals&... values 
)
{
#if defined(HAVE_MPI)
  const auto& mpi_w = cs::execution::default_mpi();
  max(mpi_w, first, values...);
#else
  return;
#endif
}
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
template <int Stride, typename T, typename... Vals>
static void
max
(
  const cs_mpi_wrapper& mpi_w,   
  T                     first[], 
  Vals&&...             values   
)
{
#if defined(HAVE_MPI)
 
  if (!mpi_w.active())
    return;
 
  /* Count number of values */
  constexpr int n_vals = sizeof...(Vals);
 
  /* Set datatype for global communication */
  cs_datatype_t datatype = cs_datatype_from_type<T>();
 
  /* Temporary work array and parallel sum */
  if (n_vals == 0)
    cs_parall_max(mpi_w, Stride, datatype, first);
  else {
    /* Unpack values */
    T *_values[] = {values ...};
 
    constexpr int work_size = (n_vals + 1) * Stride;
 
    T w[work_size];
    for (int i = 0; i < Stride; i++)
      w[i] = first[i];
 
    for (int i = 0; i < n_vals; i++)
      for (int j = 0; j < Stride; j++)
        w[(i+1)*Stride + j] = _values[i][j];
 
    cs_parall_max(mpi_w, work_size, datatype, w);
 
    for (int i = 0; i < Stride; i++)
      first[i] = w[i];
 
    for (int i = 0; i < n_vals; i++)
      for (int j = 0; j < Stride; j++)
        _values[i][j] = w[(i+1)*Stride + j];
  }
 
#endif // defined(HAVE_MPI)
}
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
template <int Stride, typename T, typename... Vals>
static void
max
(
  T          first[], 
  Vals&&...  values   
)
{
#if defined(HAVE_MPI)
  const auto& mpi_w = cs::execution::default_mpi();
  max<Stride>(mpi_w, first, values...);
#else
  return;
#endif
}
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
template <typename T, typename... Vals>
static void
min
(
  const cs_mpi_wrapper& mpi_w, 
  T&                    first, 
  Vals&...              values 
)
{
#if defined(HAVE_MPI)
 
  if (!mpi_w.active())
    return;
 
  /* Count number of values */
  constexpr int n_vals = sizeof...(Vals);
 
  /* Set datatype for global communication */
  cs_datatype_t datatype = cs_datatype_from_type<T>();
 
  if (n_vals == 0)
    cs_parall_min(mpi_w, 1, datatype, &first);
 
  else {
    /* Temporary work array and parallel sum */
 
    /* Unpack values */
    T *_values[] = {&values ...};
 
    T w[n_vals + 1];
    w[0] = first;
    for (int i = 0; i < n_vals; i++)
      w[i + 1] = *(_values[i]);
 
    cs_parall_min(mpi_w, n_vals + 1, datatype, w);
 
    first = w[0];
    for (int i = 0; i < n_vals; i++)
      *(_values[i]) = w[i + 1];
  }
 
#endif // defined(HAVE_MPI)
}
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
template <typename T, typename... Vals>
static void
min
(
  T&       first, 
  Vals&... values 
)
{
#if defined(HAVE_MPI)
  const auto& mpi_w = cs::execution::default_mpi();
  min(mpi_w, first, values...);
#else
  return;
#endif
}
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
template <int Stride, typename T, typename... Vals>
static void
min
(
  const cs_mpi_wrapper& mpi_w,   
  T                     first[], 
  Vals&&...             values   
)
{
#if defined(HAVE_MPI)
 
  if (!mpi_w.active())
    return;
 
  /* Count number of values */
  constexpr int n_vals = sizeof...(Vals);
 
  /* Set datatype for global communication */
  cs_datatype_t datatype = cs_datatype_from_type<T>();
 
  /* Temporary work array and parallel sum */
  if (n_vals == 0)
    cs_parall_min(mpi_w, Stride, datatype, first);
  else {
    /* Unpack values */
    T *_values[] = {values ...};
 
    constexpr int work_size = (n_vals + 1) * Stride;
 
    T w[work_size];
    for (int i = 0; i < Stride; i++)
      w[i] = first[i];
 
    for (int i = 0; i < n_vals; i++)
      for (int j = 0; j < Stride; j++)
        w[(i+1)*Stride + j] = _values[i][j];
 
    cs_parall_min(mpi_w, work_size, datatype, w);
 
    for (int i = 0; i < Stride; i++)
      first[i] = w[i];
 
    for (int i = 0; i < n_vals; i++)
      for (int j = 0; j < Stride; j++)
        _values[i][j] = w[(i+1)*Stride + j];
  }
 
#endif
}
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
template <int Stride, typename T, typename... Vals>
static void
min
(
  T         first[], 
  Vals&&... values   
)
{
#if defined(HAVE_MPI)
  const auto& mpi_w = cs::execution::default_mpi();
  min<Stride>(mpi_w, first, values...);
#else
  return;
#endif
}
 
} /* namespace parall */
} /* namespace cs */
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
template <typename T, typename... Vals>
static void
cs_parall_sum_scalars
(
  T&       first, 
  Vals&... values 
)
{
#if defined(HAVE_MPI)
 
  if (cs_glob_n_ranks == 1)
    return;
 
  /* Count number of values */
  constexpr size_t n_vals = sizeof...(Vals);
 
  /* Set datatype for global communication */
  cs_datatype_t datatype = cs_datatype_from_type<T>();
 
  /* Temporary work array and parallel sum */
  if (n_vals == 0)
    cs_parall_sum(1, datatype, &first);
  else {
    /* Unpack values */
    T *_values[] = {&values ...};
 
    T w[n_vals + 1];
    w[0] = first;
    for (size_t i = 0; i < n_vals; i++)
      w[i+1] = *(_values[i]);
 
    cs_parall_sum(n_vals + 1, datatype, w);
 
    first = w[0];
    for (size_t i = 0; i < n_vals; i++)
      *(_values[i]) = w[i+1];
  }
 
#endif // defined(HAVE_MPI)
}
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
template <typename T, typename... Vals>
static void
cs_parall_sum_scalars
(
  const cs_mpi_wrapper& mpi_w,    
  T&                    first, 
  Vals&...              values 
)
{
#if defined(HAVE_MPI)
 
  /* Count number of values */
  constexpr size_t n_vals = sizeof...(Vals);
 
  /* Set datatype for global communication */
  cs_datatype_t datatype = cs_datatype_from_type<T>();
 
  /* Temporary work array and parallel sum */
  if (n_vals == 0)
    cs_parall_sum(mpi_w, 1, datatype, &first);
  else {
    /* Unpack values */
    T *_values[] = {&values ...};
 
    T w[n_vals + 1];
    w[0] = first;
    for (size_t i = 0; i < n_vals; i++)
      w[i+1] = *(_values[i]);
 
    cs_parall_sum(mpi_w, n_vals + 1, datatype, w);
 
    first = w[0];
    for (size_t i = 0; i < n_vals; i++)
      *(_values[i]) = w[i+1];
  }
 
#endif // defined(HAVE_MPI)
}
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
template <int Stride, typename T, typename... Vals>
static void
cs_parall_sum_strided
(
  const cs_mpi_wrapper& mpi_w,      
  T                     first[], 
  Vals&&...             values   
)
{
#if defined(HAVE_MPI)
  if (!mpi_w.active())
    return;
 
  /* Count number of values */
  constexpr size_t n_vals = sizeof...(Vals);
 
  /* Set datatype for global communication */
  cs_datatype_t datatype = cs_datatype_from_type<T>();
 
  /* Temporary work array and parallel sum */
  if (n_vals == 0)
    cs_parall_sum(mpi_w, Stride, datatype, first);
  else {
    /* Unpack values */
    T *_values[] = {values ...};
 
    constexpr size_t work_size = (n_vals + 1) * Stride;
 
    T w[work_size];
    for (int i = 0; i < Stride; i++)
      w[i] = first[i];
 
    for (size_t i = 0; i < n_vals; i++)
      for (int j = 0; j < Stride; j++)
        w[(i+1)*Stride + j] = _values[i][j];
 
    cs_parall_sum(mpi_w, work_size, datatype, w);
 
    for (int i = 0; i < Stride; i++)
      first[i] = w[i];
 
    for (size_t i = 0; i < n_vals; i++) {
      for (int j = 0; j < Stride; j++)
        _values[i][j] = w[(i+1)*Stride + j];
    }
  }
 
#endif // defined(HAVE_MPI)
}
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
template <int Stride, typename T, typename... Vals>
static void
cs_parall_sum_strided
(
  T         first[], 
  Vals&&... values   
)
{
#if defined(HAVE_MPI)
 
  if (cs_glob_n_ranks == 1)
    return;
 
  /* Count number of values */
  constexpr size_t n_vals = sizeof...(Vals);
 
  /* Set datatype for global communication */
  cs_datatype_t datatype = cs_datatype_from_type<T>();
 
  /* Temporary work array and parallel sum */
  if (n_vals == 0)
    cs_parall_sum(Stride, datatype, first);
  else {
    /* Unpack values */
    T *_values[] = {values ...};
 
    constexpr size_t work_size = (n_vals + 1) * Stride;
 
    T w[work_size];
    for (int i = 0; i < Stride; i++)
      w[i] = first[i];
 
    for (int i = 0; i < n_vals; i++) {
      for (int j = 0; j < Stride; j++)
        w[(i+1)*Stride + j] = _values[i][j];
    }
 
    cs_parall_sum(work_size, datatype, w);
 
    for (int i = 0; i < Stride; i++)
      first[i] = w[i];
 
    for (size_t i = 0; i < n_vals; i++) {
      for (int j = 0; j < Stride; j++)
        _values[i][j] = w[(i+1)*Stride + j];
    }
  }
 
#endif // defined(HAVE_MPI)
}
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
template <typename T, typename... Vals>
static void
cs_parall_max_scalars
(
  T&       first, 
  Vals&... values 
)
{
#if defined(HAVE_MPI)
 
  if (cs_glob_n_ranks == 1)
    return;
 
  /* Count number of values */
  constexpr size_t n_vals = sizeof...(Vals);
 
  /* Set datatype for global communication */
  cs_datatype_t datatype = cs_datatype_from_type<T>();
 
  /* Temporary work array and parallel sum */
  if (n_vals == 0)
    cs_parall_max(1, datatype, &first);
  else {
 
    /* Unpack values */
    T *_values[] = {&values ...};
 
    T w[n_vals + 1];
    w[0] = first;
    for (size_t i = 0; i < n_vals; i++)
      w[i+1] = *(_values[i]);
 
    cs_parall_max(n_vals + 1, datatype, w);
 
    first = w[0];
    for (size_t i = 0; i < n_vals; i++)
      *(_values[i]) = w[i+1];
  }
 
#endif // defined(HAVE_MPI)
}
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
template <typename T, typename... Vals>
static void
cs_parall_max_scalars
(
  const cs_mpi_wrapper& mpi_w,    
  T&                    first, 
  Vals&...              values 
)
{
#if defined(HAVE_MPI)
 
  /* Count number of values */
  constexpr size_t n_vals = sizeof...(Vals);
 
  /* Set datatype for global communication */
  cs_datatype_t datatype = cs_datatype_from_type<T>();
 
  /* Temporary work array and parallel sum */
  if (n_vals == 0)
    cs_parall_max(mpi_w, 1, datatype, &first);
  else {
 
    /* Unpack values */
    T *_values[] = {&values ...};
 
    T w[n_vals + 1];
    w[0] = first;
    for (size_t i = 0; i < n_vals; i++)
      w[i+1] = *(_values[i]);
 
    cs_parall_max(mpi_w, n_vals + 1, datatype, w);
 
    first = w[0];
    for (size_t i = 0; i < n_vals; i++)
      *(_values[i]) = w[i+1];
  }
 
#endif // defined(HAVE_MPI)
}
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
template <int Stride, typename T, typename... Vals>
static void
cs_parall_max_strided
(
  T         first[], 
  Vals&&... values   
)
{
#if defined(HAVE_MPI)
 
  if (cs_glob_n_ranks == 1)
    return;
 
  /* Count number of values */
  constexpr size_t n_vals = sizeof...(Vals);
 
  /* Set datatype for global communication */
  cs_datatype_t datatype = cs_datatype_from_type<T>();
 
  /* Temporary work array and parallel sum */
  if (n_vals == 0)
    cs_parall_max(Stride, datatype, first);
  else {
    /* Unpack values */
    T *_values[] = {values ...};
 
    constexpr size_t work_size = (n_vals + 1) * Stride;
 
    T w[work_size];
    for (int i = 0; i < Stride; i++)
      w[i] = first[i];
 
    for (size_t i = 0; i < n_vals; i++)
      for (int j = 0; j < Stride; j++)
        w[(i+1)*Stride + j] = _values[i][j];
 
    cs_parall_max(work_size, datatype, w);
 
    for (int i = 0; i < Stride; i++)
      first[i] = w[i];
 
    for (size_t i = 0; i < n_vals; i++)
      for (int j = 0; j < Stride; j++)
        _values[i][j] = w[(i+1)*Stride + j];
  }
 
#endif // defined(HAVE_MPI)
}
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
template <int Stride, typename T, typename... Vals>
static void
cs_parall_max_strided
(
  const cs_mpi_wrapper& mpi_w,      
  T                     first[], 
  Vals&&...             values   
)
{
#if defined(HAVE_MPI)
 
  /* Count number of values */
  constexpr size_t n_vals = sizeof...(Vals);
 
  /* Set datatype for global communication */
  cs_datatype_t datatype = cs_datatype_from_type<T>();
 
  /* Temporary work array and parallel sum */
  if (n_vals == 0)
    cs_parall_max(mpi_w, Stride, datatype, first);
  else {
    /* Unpack values */
    T *_values[] = {values ...};
 
    constexpr size_t work_size = (n_vals + 1) * Stride;
 
    T w[work_size];
    for (int i = 0; i < Stride; i++)
      w[i] = first[i];
 
    for (size_t i = 0; i < n_vals; i++)
      for (int j = 0; j < Stride; j++)
        w[(i+1)*Stride + j] = _values[i][j];
 
    cs_parall_max(mpi_w, work_size, datatype, w);
 
    for (int i = 0; i < Stride; i++)
      first[i] = w[i];
 
    for (size_t i = 0; i < n_vals; i++)
      for (int j = 0; j < Stride; j++)
        _values[i][j] = w[(i+1)*Stride + j];
  }
 
#endif // defined(HAVE_MPI)
}
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
template <typename T, typename... Vals>
static void
cs_parall_min_scalars
(
  T&       first, 
  Vals&... values 
)
{
#if defined(HAVE_MPI)
 
  if (cs_glob_n_ranks == 1)
    return;
 
  /* Count number of values */
  constexpr size_t n_vals = sizeof...(Vals);
 
  /* Set datatype for global communication */
  cs_datatype_t datatype = cs_datatype_from_type<T>();
 
  if (n_vals == 0)
    cs_parall_min(1, datatype, &first);
 
  else {
    /* Temporary work array and parallel sum */
 
    /* Unpack values */
    T *_values[] = {&values ...};
 
    T w[n_vals + 1];
    w[0] = first;
    for (size_t i = 0; i < n_vals; i++)
      w[i + 1] = *(_values[i]);
 
    cs_parall_min(n_vals + 1, datatype, w);
 
    first = w[0];
    for (size_t i = 0; i < n_vals; i++)
      *(_values[i]) = w[i + 1];
  }
 
#endif // defined(HAVE_MPI)
}
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
template <typename T, typename... Vals>
static void
cs_parall_min_scalars
(
  const cs_mpi_wrapper& mpi_w,     
  T&                    first,  
  Vals&...              values  
)
{
#if defined(HAVE_MPI)
 
  /* Count number of values */
  constexpr size_t n_vals = sizeof...(Vals);
 
  /* Set datatype for global communication */
  cs_datatype_t datatype = cs_datatype_from_type<T>();
 
  if (n_vals == 0)
    cs_parall_min(mpi_w, 1, datatype, &first);
 
  else {
    /* Temporary work array and parallel sum */
 
    /* Unpack values */
    T *_values[] = {&values ...};
 
    T w[n_vals + 1];
    w[0] = first;
    for (size_t i = 0; i < n_vals; i++)
      w[i + 1] = *(_values[i]);
 
    cs_parall_min(mpi_w, n_vals + 1, datatype, w);
 
    first = w[0];
    for (size_t i = 0; i < n_vals; i++)
      *(_values[i]) = w[i + 1];
  }
 
#endif // defined(HAVE_MPI)
}
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
template <int Stride, typename T, typename... Vals>
static void
cs_parall_min_strided
(
  T         first[],  
  Vals&&... values    
)
{
#if defined(HAVE_MPI)
 
  if (cs_glob_n_ranks == 1)
    return;
 
  /* Count number of values */
  constexpr size_t n_vals = sizeof...(Vals);
 
  /* Set datatype for global communication */
  cs_datatype_t datatype = cs_datatype_from_type<T>();
 
  /* Temporary work array and parallel sum */
  if (n_vals == 0)
    cs_parall_min(Stride, datatype, first);
  else {
    /* Unpack values */
    T *_values[] = {values ...};
 
    constexpr size_t work_size = (n_vals + 1) * Stride;
 
    T w[work_size];
    for (int i = 0; i < Stride; i++)
      w[i] = first[i];
 
    for (size_t i = 0; i < n_vals; i++)
      for (int j = 0; j < Stride; j++)
        w[(i+1)*Stride + j] = _values[i][j];
 
    cs_parall_min(work_size, datatype, w);
 
    for (int i = 0; i < Stride; i++)
      first[i] = w[i];
 
    for (size_t i = 0; i < n_vals; i++)
      for (int j = 0; j < Stride; j++)
        _values[i][j] = w[(i+1)*Stride + j];
  }
 
#endif
}
 
/*----------------------------------------------------------------------------*/
/*----------------------------------------------------------------------------*/
 
template <int Stride, typename T, typename... Vals>
static void
cs_parall_min_strided
(
  const cs_mpi_wrapper& mpi_w,      
  T                     first[], 
  Vals&&...             values   
)
{
#if defined(HAVE_MPI)
 
  /* Count number of values */
  constexpr size_t n_vals = sizeof...(Vals);
 
  /* Set datatype for global communication */
  cs_datatype_t datatype = cs_datatype_from_type<T>();
 
  /* Temporary work array and parallel sum */
  if (n_vals == 0)
    cs_parall_min(mpi_w, Stride, datatype, first);
 
  else {
    /* Unpack values */
    T *_values[] = {values ...};
 
    constexpr size_t work_size = (n_vals + 1) * Stride;
 
    T w[work_size];
    for (int i = 0; i < Stride; i++)
      w[i] = first[i];
 
    for (size_t i = 0; i < n_vals; i++)
      for (int j = 0; j < Stride; j++)
        w[(i+1)*Stride + j] = _values[i][j];
 
    cs_parall_min(mpi_w, work_size, datatype, w);
 
    for (int i = 0; i < Stride; i++)
      first[i] = w[i];
 
    for (size_t i = 0; i < n_vals; i++)
      for (int j = 0; j < Stride; j++)
        _values[i][j] = w[(i+1)*Stride + j];
  }
#endif // defined(HAVE_MPI)
}
 
#endif //__cplusplus
 
/*----------------------------------------------------------------------------*/
 
#endif /* CS_PARALL_H */