root/ompi/mca/coll/cuda/coll_cuda.h

INCLUDED FROM

   1 /*
   2  * Copyright (c) 2014      The University of Tennessee and The University
   3  *                         of Tennessee Research Foundation.  All rights
   4  *                         reserved.
   5  * Copyright (c) 2014-2015 NVIDIA Corporation.  All rights reserved.
   6  * $COPYRIGHT$
   7  *
   8  * Additional copyrights may follow
   9  *
  10  * $HEADER$
  11  */
  12 
  13 #ifndef MCA_COLL_CUDA_EXPORT_H
  14 #define MCA_COLL_CUDA_EXPORT_H
  15 
  16 #include "ompi_config.h"
  17 
  18 #include "mpi.h"
  19 
  20 #include "opal/class/opal_object.h"
  21 #include "ompi/mca/mca.h"
  22 
  23 #include "ompi/constants.h"
  24 #include "ompi/mca/coll/coll.h"
  25 #include "ompi/mca/coll/base/base.h"
  26 #include "ompi/communicator/communicator.h"
  27 
  28 BEGIN_C_DECLS
  29 
  30 /* API functions */
  31 
  32 int mca_coll_cuda_init_query(bool enable_progress_threads,
  33                              bool enable_mpi_threads);
  34 mca_coll_base_module_t
  35 *mca_coll_cuda_comm_query(struct ompi_communicator_t *comm,
  36                           int *priority);
  37 
  38 int mca_coll_cuda_module_enable(mca_coll_base_module_t *module,
  39                                 struct ompi_communicator_t *comm);
  40 
  41 int
  42 mca_coll_cuda_allreduce(const void *sbuf, void *rbuf, int count,
  43                         struct ompi_datatype_t *dtype,
  44                         struct ompi_op_t *op,
  45                         struct ompi_communicator_t *comm,
  46                         mca_coll_base_module_t *module);
  47 
  48 int mca_coll_cuda_reduce(const void *sbuf, void *rbuf, int count,
  49                          struct ompi_datatype_t *dtype,
  50                          struct ompi_op_t *op,
  51                          int root,
  52                          struct ompi_communicator_t *comm,
  53                          mca_coll_base_module_t *module);
  54 
  55 int mca_coll_cuda_exscan(const void *sbuf, void *rbuf, int count,
  56                          struct ompi_datatype_t *dtype,
  57                          struct ompi_op_t *op,
  58                          struct ompi_communicator_t *comm,
  59                          mca_coll_base_module_t *module);
  60 
  61 int mca_coll_cuda_scan(const void *sbuf, void *rbuf, int count,
  62                        struct ompi_datatype_t *dtype,
  63                        struct ompi_op_t *op,
  64                        struct ompi_communicator_t *comm,
  65                        mca_coll_base_module_t *module);
  66 
  67 int
  68 mca_coll_cuda_reduce_scatter_block(const void *sbuf, void *rbuf, int rcount,
  69                                    struct ompi_datatype_t *dtype,
  70                                    struct ompi_op_t *op,
  71                                    struct ompi_communicator_t *comm,
  72                                    mca_coll_base_module_t *module);
  73 
  74 /* Types */
  75 /* Module */
  76 
  77 typedef struct mca_coll_cuda_module_t {
  78     mca_coll_base_module_t super;
  79 
  80     /* Pointers to all the "real" collective functions */
  81     mca_coll_base_comm_coll_t c_coll;
  82 } mca_coll_cuda_module_t;
  83 
  84 OBJ_CLASS_DECLARATION(mca_coll_cuda_module_t);
  85 
  86 /* Component */
  87 
  88 typedef struct mca_coll_cuda_component_t {
  89     mca_coll_base_component_2_0_0_t super;
  90 
  91     int priority; /* Priority of this component */
  92     int disable_cuda_coll;  /* Force disable of the CUDA collective component */
  93 } mca_coll_cuda_component_t;
  94 
  95 /* Globally exported variables */
  96 
  97 OMPI_MODULE_DECLSPEC extern mca_coll_cuda_component_t mca_coll_cuda_component;
  98 
  99 END_C_DECLS
 100 
 101 #endif /* MCA_COLL_CUDA_EXPORT_H */