Solo: a shared memory collective module

ompi/mca/coll/solo/Makefile.am

@@ -0,0 +1,43 @@
+#
+# Copyright (c) 2019      The University of Tennessee and The University
+#                         of Tennessee Research Foundation.  All rights
+#                         reserved.
+# $COPYRIGHT$
+#
+# Additional copyrights may follow
+#
+# $HEADER$
+#
+
+sources = \
+        coll_solo.h \
+        coll_solo_mpool.h \
+        coll_solo_barrier.c \
+        coll_solo_reduce.c \
+        coll_solo_bcast.c \
+        coll_solo_allreduce.c \
+        coll_solo_component.c \
+        coll_solo_module.c \
+        coll_solo_mpool.c
+
+# Make the output library in this directory, and name it either
+# mca_<type>_<name>.la (for DSO builds) or libmca_<type>_<name>.la
+# (for static builds).
+
+component_noinst =
+component_install =
+if MCA_BUILD_ompi_coll_solo_DSO
+component_install += mca_coll_solo.la
+else
+component_noinst += libmca_coll_solo.la
+endif
+
+mcacomponentdir = $(ompilibdir)
+mcacomponent_LTLIBRARIES = $(component_install)
+mca_coll_solo_la_SOURCES = $(sources)
+mca_coll_solo_la_LDFLAGS = -module -avoid-version
+mca_coll_solo_la_LIBADD = 
+
+noinst_LTLIBRARIES = $(component_noinst)
+libmca_coll_solo_la_SOURCES =$(sources)
+libmca_coll_solo_la_LDFLAGS = -module -avoid-version

ompi/mca/coll/solo/coll_solo.h

@@ -0,0 +1,191 @@
+/**
+ * Copyright (c) 2019      The University of Tennessee and The University
+ *                         of Tennessee Research Foundation.  All rights
+ *                         reserved.
+ * $COPYRIGHT$
+ *
+ * Additional copyrights may follow
+ *
+ * $HEADER$
+ */
+
+#ifndef MCA_COLL_SOLO_EXPORT_H
+#define MCA_COLL_SOLO_EXPORT_H
+
+#include "ompi_config.h"
+
+#include "mpi.h"
+#include "ompi/mca/mca.h"
+#include "ompi/mca/coll/coll.h"
+#include "ompi/communicator/communicator.h"
+#include "ompi/win/win.h"
+#include "ompi/include/mpi.h"
+#include "ompi/mca/coll/base/coll_base_functions.h"
+#include "opal/util/info.h"
+#include "ompi/op/op.h"
+#include "opal/runtime/opal_progress.h"
+#include "ompi/mca/pml/pml.h"
+#include "ompi/mca/coll/base/coll_tags.h"
+#include "coll_solo_mpool.h"
+
+BEGIN_C_DECLS
+/**
+ * Structure to hold the solo coll component.  First it holds the base coll component, and then 
+ * holds a bunch of solo-coll-component-specific stuff (e.g., current MCA param values).
+ */
+    typedef struct mca_coll_solo_component_t {
+    /* Base coll component */
+    mca_coll_base_component_2_0_0_t super;
+
+    /* MCA parameters */
+    /* Priority of the solo module */
+    int solo_priority;
+    /* The size of data_bufs in the static_win */
+    uint32_t static_block_size;
+    uint32_t mpool_small_block_size;
+    uint32_t mpool_small_block_num;
+    uint32_t mpool_large_block_size;
+    uint32_t mpool_large_block_num;
+
+    /* Shared memory pool */
+    mca_coll_solo_mpool_t *solo_mpool;
+} mca_coll_solo_component_t;
+
+/* Coll solo module */
+typedef struct mca_coll_solo_module_t {
+    /* Base module */
+    mca_coll_base_module_t super;
+
+    /* Whether this module has been lazily initialized or not yet */
+    bool enabled;
+
+    /**
+     * This window is created by ompi_win_allocate_shared such that each process contains a shared 
+     * memory data buffer, and this data buffer is divided into two parts - ctrl_bufs and data_bufs.
+     */
+    MPI_Win static_win;
+    /** 
+     * The first 4 * opal_cache_line_size bytes in the shared memory data buffer in static_win, used
+     * to store control messages.
+     */
+    char **ctrl_bufs;
+    /** 
+     * The rest of the shared memory data buffer in static_win, which is intent to be used to 
+     * tranfer very small messages. Its size is set by static_block_size.
+     */
+    char **data_bufs;
+
+    /* Identify which ctrl_buf is currently used in mac_coll_solo_barrier_intra. */
+    int barrier_tag;
+} mca_coll_solo_module_t;
+OBJ_CLASS_DECLARATION(mca_coll_solo_module_t);
+
+/**
+ * Global component instance
+ */
+OMPI_MODULE_DECLSPEC extern mca_coll_solo_component_t mca_coll_solo_component;
+
+/**
+ * coll module functions
+ */
+int mca_coll_solo_init_query(bool enable_progress_threads, bool enable_mpi_threads);
+
+mca_coll_base_module_t *mca_coll_solo_comm_query(struct ompi_communicator_t *comm, int *priority);
+
+/* Lazily enable a module (since it involves expensive memory allocation, etc.) */
+int mca_coll_solo_lazy_enable(mca_coll_base_module_t * module, struct ompi_communicator_t *comm);
+
+/* Setup and initialize the static_win of a communicator */
+void mca_coll_solo_setup_static_win(mca_coll_solo_module_t *solo_module,
+                                    struct ompi_communicator_t *comm, 
+                                    size_t data_buf_size);
+
+/* MPI_Barrier algorithms */
+int mac_coll_solo_barrier_intra(struct ompi_communicator_t *comm, 
+                                mca_coll_base_module_t * module);
+
+/* MPI_Bcast algorithms */
+int mca_coll_solo_bcast_intra(void *buff, int count,
+                              struct ompi_datatype_t *dtype, 
+                              int root,
+                              struct ompi_communicator_t *comm, 
+                              mca_coll_base_module_t * module);
+
+int mca_coll_solo_bcast_linear_intra_memcpy(void *buff, int count, 
+                                            struct ompi_datatype_t *dtype, 
+                                            int root, 
+                                            struct ompi_communicator_t *comm, 
+                                            mca_coll_base_module_t * module);
+
+int mca_coll_solo_bcast_pipeline_intra_memcpy(void *buff, int count,
+                                              struct ompi_datatype_t *dtype, 
+                                              int root,
+                                              struct ompi_communicator_t *comm, 
+                                              mca_coll_base_module_t * module,
+                                              size_t seg_size);
+
+/* MPI_Reduce algorithms */
+int mca_coll_solo_reduce_intra(const void *sbuf, void *rbuf, int count,
+                               struct ompi_datatype_t *dtype,
+                               struct ompi_op_t *op,
+                               int root,
+                               struct ompi_communicator_t *comm, 
+                               mca_coll_base_module_t * module);
+
+int mca_coll_solo_reduce_ring_intra_memcpy(const void *sbuf, void *rbuf, int count,
+                                           struct ompi_datatype_t *dtype,
+                                           struct ompi_op_t *op,
+                                           int root, 
+                                           struct ompi_communicator_t
+                                           *comm, mca_coll_base_module_t * module);
+
+
+/* MPI_Allreduce algorithms */
+int mca_coll_solo_allreduce_intra(const void *sbuf, void *rbuf, int count,
+                                  struct ompi_datatype_t *dtype,
+                                  struct ompi_op_t *op,
+                                  struct ompi_communicator_t *comm,
+                                  mca_coll_base_module_t * module);
+
+int mca_coll_solo_allreduce_ring_intra_memcpy(const void *sbuf, void *rbuf, int count, 
+                                              struct ompi_datatype_t *dtype, 
+                                              struct ompi_op_t *op, 
+                                              struct ompi_communicator_t *comm, 
+                                              mca_coll_base_module_t * module);
+
+
+/* Solo pack to shared memory */
+static inline void mca_coll_solo_pack_to_shared(void *local_buf, void *shared_buf, struct ompi_datatype_t *dtype, int count, ptrdiff_t extent) {
+    if (ompi_datatype_is_predefined(dtype)) {
+        memcpy((char *) shared_buf, (char *) local_buf, count * extent);
+    }
+    else {
+        int pos = 0;        
+        MPI_Pack(local_buf, count, dtype, shared_buf, count * extent, &pos, MPI_COMM_SELF);
+    }
+}
+
+/* Solo unpack from shared memory */
+static inline void mca_coll_solo_unpack_from_shared(void *local_buf, void *shared_buf, struct ompi_datatype_t *dtype, int count, ptrdiff_t extent) {
+    if (ompi_datatype_is_predefined(dtype)) {
+        memcpy((char *) local_buf, (char *) shared_buf, count * extent);
+    }
+    else {
+        int pos = 0;
+        MPI_Unpack(shared_buf, count * extent, &pos, local_buf, count, dtype, MPI_COMM_SELF);
+    }
+}
+
+/* Solo copy from source to target */
+static inline void mca_coll_solo_copy(void *source, void *target, struct ompi_datatype_t *dtype, int count, ptrdiff_t extent) {
+    if (ompi_datatype_is_predefined(dtype)) {
+        memcpy(target, source, count * extent);
+    }
+    else {
+        ompi_datatype_copy_content_same_ddt(dtype, count, target, source);
+    }
+    return;
+} 
+
+END_C_DECLS
+#endif                          /* MCA_COLL_SOLO_EXPORT_H */