Fix the atomic management of the bcast and reduce freelist

API consistent with other collective modules
Add comments
Other minor cleanups.

Signed-off-by: George Bosilca <bosilca@icl.utk.edu>

Author: bosilca

ompi/communicator/communicator.h

@@ -3,7 +3,7 @@
  * Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
  *                         University Research and Technology
  *                         Corporation.  All rights reserved.
- * Copyright (c) 2004-2017 The University of Tennessee and The University
+ * Copyright (c) 2004-2020 The University of Tennessee and The University
  *                         of Tennessee Research Foundation.  All rights
  *                         reserved.
  * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
@@ -187,10 +187,14 @@ struct ompi_communicator_t {
 
     /* Collectives module interface and data */
     mca_coll_base_comm_coll_t *c_coll;
-    
-    /* Non-blocking collective tag */
-    _Atomic int32_t c_ibcast_tag;
-    _Atomic int32_t c_ireduce_tag;
+
+    /* Non-blocking collective tag. These are added here as they should be
+     * shared between all non-blocking collective modules (to avoid message
+     * collisions between them in the case where multiple outstanding
+     * non-blocking collective coexists using multiple backends).
+     */
+    opal_atomic_int32_t c_ibcast_tag;
+    opal_atomic_int32_t c_ireduce_tag;
 };
 typedef struct ompi_communicator_t ompi_communicator_t;
 

ompi/mca/coll/adapt/Makefile.am

@@ -1,5 +1,5 @@
 #
-# Copyright (c) 2014      The University of Tennessee and The University
+# Copyright (c) 2014-2020 The University of Tennessee and The University
 #                         of Tennessee Research Foundation.  All rights
 #                         reserved.
 # $COPYRIGHT$
@@ -13,11 +13,11 @@
 sources = \
 	coll_adapt_component.c \
 	coll_adapt_module.c \
-    coll_adapt_bcast.c \
+	coll_adapt_bcast.c \
 	coll_adapt_ibcast.c \
 	coll_adapt_reduce.c \
 	coll_adapt_ireduce.c \
-    coll_adapt.h \
+	coll_adapt.h \
 	coll_adapt_algorithms.h \
 	coll_adapt_context.h \
 	coll_adapt_context.c \

ompi/mca/coll/adapt/coll_adapt.h

@@ -38,8 +38,9 @@ typedef struct mca_coll_adapt_component_t {
     /* MCA parameter: Priority of this component */
     int adapt_priority;
 
-    /* MCA parameter: Output verbose level */
+    /* MCA parameter: Output stream and verbose level */
     int adapt_output;
+    int adapt_verbose;
 
     /* MCA parameter: Maximum number of segment in context free list */
     int adapt_context_free_list_max;
@@ -57,7 +58,6 @@ typedef struct mca_coll_adapt_component_t {
     int adapt_ibcast_max_recv_requests;
     /* Bcast free list */
     opal_free_list_t *adapt_ibcast_context_free_list;
-    opal_atomic_int32_t adapt_ibcast_context_free_list_enabled;
 
     /* Reduce MCA parameter */
     int adapt_ireduce_algorithm;
@@ -70,7 +70,6 @@ typedef struct mca_coll_adapt_component_t {
 
     /* Reduce free list */
     opal_free_list_t *adapt_ireduce_context_free_list;
-    opal_atomic_int32_t adapt_ireduce_context_free_list_enabled;
 
 } mca_coll_adapt_component_t;
 
@@ -91,7 +90,7 @@ OMPI_MODULE_DECLSPEC extern mca_coll_adapt_component_t mca_coll_adapt_component;
 int ompi_coll_adapt_init_query(bool enable_progress_threads, bool enable_mpi_threads);
 mca_coll_base_module_t * ompi_coll_adapt_comm_query(struct ompi_communicator_t *comm, int *priority);
 
-/* Free ADAPT quest */
+/* ADAPT request free */
 int ompi_coll_adapt_request_free(ompi_request_t **request);
 
 #endif /* MCA_COLL_ADAPT_EXPORT_H */

ompi/mca/coll/adapt/coll_adapt_algorithms.h

@@ -20,82 +20,45 @@ typedef struct ompi_coll_adapt_algorithm_index_s {
 } ompi_coll_adapt_algorithm_index_t;
 
 /* Bcast */
-int ompi_coll_adapt_ibcast_init(void);
+int ompi_coll_adapt_ibcast_register(void);
 int ompi_coll_adapt_ibcast_fini(void);
-int ompi_coll_adapt_bcast(void *buff, int count, struct ompi_datatype_t *datatype, int root,
-                         struct ompi_communicator_t *comm, mca_coll_base_module_t * module);
-int ompi_coll_adapt_ibcast(void *buff, int count, struct ompi_datatype_t *datatype, int root,
-                          struct ompi_communicator_t *comm, ompi_request_t ** request,
-                          mca_coll_base_module_t * module);
-int ompi_coll_adapt_ibcast_generic(void *buff, int count, struct ompi_datatype_t *datatype, int root,
-                                  struct ompi_communicator_t *comm, ompi_request_t ** request,
-                                  mca_coll_base_module_t * module, ompi_coll_tree_t * tree,
-                                  size_t seg_size, int ibcast_tag);
-int ompi_coll_adapt_ibcast_binomial(void *buff, int count, struct ompi_datatype_t *datatype,
-                                   int root, struct ompi_communicator_t *comm,
-                                   ompi_request_t ** request, mca_coll_base_module_t * module,
-                                   int ibcast_tag);
-int ompi_coll_adapt_ibcast_in_order_binomial(void *buff, int count, struct ompi_datatype_t *datatype,
-                                            int root, struct ompi_communicator_t *comm,
-                                            ompi_request_t ** request,
-                                            mca_coll_base_module_t * module, int ibcast_tag);
-int ompi_coll_adapt_ibcast_binary(void *buff, int count, struct ompi_datatype_t *datatype, int root,
-                                 struct ompi_communicator_t *comm, ompi_request_t ** request,
-                                 mca_coll_base_module_t * module, int ibcast_tag);
-int ompi_coll_adapt_ibcast_pipeline(void *buff, int count, struct ompi_datatype_t *datatype,
-                                   int root, struct ompi_communicator_t *comm,
-                                   ompi_request_t ** request, mca_coll_base_module_t * module,
-                                   int ibcast_tag);
-int ompi_coll_adapt_ibcast_chain(void *buff, int count, struct ompi_datatype_t *datatype, int root,
-                                struct ompi_communicator_t *comm, ompi_request_t ** request,
-                                mca_coll_base_module_t * module, int ibcast_tag);
-int ompi_coll_adapt_ibcast_linear(void *buff, int count, struct ompi_datatype_t *datatype, int root,
-                                 struct ompi_communicator_t *comm, ompi_request_t ** request,
-                                 mca_coll_base_module_t * module, int ibcast_tag);
-int ompi_coll_adapt_ibcast_tuned(void *buff, int count, struct ompi_datatype_t *datatype, int root,
-				struct ompi_communicator_t *comm, ompi_request_t ** request,
-				mca_coll_base_module_t *module, int ibcast_tag); 
+int ompi_coll_adapt_bcast(BCAST_ARGS);
+int ompi_coll_adapt_ibcast(IBCAST_ARGS);
+int ompi_coll_adapt_ibcast_generic(IBCAST_ARGS,
+                                   ompi_coll_tree_t * tree, size_t seg_size, int ibcast_tag);
+int ompi_coll_adapt_ibcast_binomial(IBCAST_ARGS,
+                                    int ibcast_tag);
+int ompi_coll_adapt_ibcast_in_order_binomial(IBCAST_ARGS,
+                                             int ibcast_tag);
+int ompi_coll_adapt_ibcast_binary(IBCAST_ARGS,
+                                  int ibcast_tag);
+int ompi_coll_adapt_ibcast_pipeline(IBCAST_ARGS,
+                                    int ibcast_tag);
+int ompi_coll_adapt_ibcast_chain(IBCAST_ARGS,
+                                 int ibcast_tag);
+int ompi_coll_adapt_ibcast_linear(IBCAST_ARGS,
+                                  int ibcast_tag);
+int ompi_coll_adapt_ibcast_tuned(IBCAST_ARGS,
+                                 int ibcast_tag);
 
 /* Reduce */
-int ompi_coll_adapt_ireduce_init(void);
+int ompi_coll_adapt_ireduce_register(void);
 int ompi_coll_adapt_ireduce_fini(void);
-int ompi_coll_adapt_reduce(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 ompi_coll_adapt_ireduce(const void *sbuf, void *rbuf, int count, struct ompi_datatype_t *dtype,
-                           struct ompi_op_t *op, int root, struct ompi_communicator_t *comm,
-                           ompi_request_t ** request, mca_coll_base_module_t * module);
-int ompi_coll_adapt_ireduce_generic(const void *sbuf, void *rbuf, int count,
-                                   struct ompi_datatype_t *dtype, struct ompi_op_t *op, int root,
-                                   struct ompi_communicator_t *comm, ompi_request_t ** request,
-                                   mca_coll_base_module_t * module, ompi_coll_tree_t * tree,
-                                   size_t seg_size, int ireduce_tag);
-int ompi_coll_adapt_ireduce_tuned(const void *sbuf, void *rbuf, int count,
-				 struct ompi_datatype_t *dtype, struct ompi_op_t *op, int root,
-				 struct ompi_communicator_t *comm, ompi_request_t ** request,
-				 mca_coll_base_module_t *module, int ireduce_tag);
-int ompi_coll_adapt_ireduce_binomial(const void *sbuf, void *rbuf, int count,
-                                    struct ompi_datatype_t *dtype, struct ompi_op_t *op, int root,
-                                    struct ompi_communicator_t *comm, ompi_request_t ** request,
-                                    mca_coll_base_module_t * module, int ireduce_tag);
-int ompi_coll_adapt_ireduce_in_order_binomial(const void *sbuf, void *rbuf, int count,
-                                             struct ompi_datatype_t *dtype, struct ompi_op_t *op,
-                                             int root, struct ompi_communicator_t *comm,
-                                             ompi_request_t ** request,
-                                             mca_coll_base_module_t * module, int ireduce_tag);
-int ompi_coll_adapt_ireduce_binary(const void *sbuf, void *rbuf, int count,
-                                  struct ompi_datatype_t *dtype, struct ompi_op_t *op, int root,
-                                  struct ompi_communicator_t *comm, ompi_request_t ** request,
-                                  mca_coll_base_module_t * module, int ireduce_tag);
-int ompi_coll_adapt_ireduce_pipeline(const void *sbuf, void *rbuf, int count,
-                                    struct ompi_datatype_t *dtype, struct ompi_op_t *op, int root,
-                                    struct ompi_communicator_t *comm, ompi_request_t ** request,
-                                    mca_coll_base_module_t * module, int ireduce_tag);
-int ompi_coll_adapt_ireduce_chain(const void *sbuf, void *rbuf, int count,
-                                 struct ompi_datatype_t *dtype, struct ompi_op_t *op, int root,
-                                 struct ompi_communicator_t *comm, ompi_request_t ** request,
-                                 mca_coll_base_module_t * module, int ireduce_tag);
-int ompi_coll_adapt_ireduce_linear(const void *sbuf, void *rbuf, int count,
-                                  struct ompi_datatype_t *dtype, struct ompi_op_t *op, int root,
-                                  struct ompi_communicator_t *comm, ompi_request_t ** request,
-                                  mca_coll_base_module_t * module, int ireduce_tag);
+int ompi_coll_adapt_reduce(REDUCE_ARGS);
+int ompi_coll_adapt_ireduce(IREDUCE_ARGS);
+int ompi_coll_adapt_ireduce_generic(IREDUCE_ARGS,
+                                    ompi_coll_tree_t * tree, size_t seg_size, int ireduce_tag);
+int ompi_coll_adapt_ireduce_tuned(IREDUCE_ARGS,
+                                  int ireduce_tag);
+int ompi_coll_adapt_ireduce_binomial(IREDUCE_ARGS,
+                                     int ireduce_tag);
+int ompi_coll_adapt_ireduce_in_order_binomial(IREDUCE_ARGS,
+                                              int ireduce_tag);
+int ompi_coll_adapt_ireduce_binary(IREDUCE_ARGS,
+                                   int ireduce_tag);
+int ompi_coll_adapt_ireduce_pipeline(IREDUCE_ARGS,
+                                     int ireduce_tag);
+int ompi_coll_adapt_ireduce_chain(IREDUCE_ARGS,
+                                  int ireduce_tag);
+int ompi_coll_adapt_ireduce_linear(IREDUCE_ARGS,
+                                   int ireduce_tag);

ompi/mca/coll/adapt/coll_adapt_bcast.c

@@ -17,10 +17,13 @@ int ompi_coll_adapt_bcast(void *buff, int count, struct ompi_datatype_t *datatyp
 {
     if (count == 0) {
         return MPI_SUCCESS;
-    } else {
-        ompi_request_t *request;
-        int err = ompi_coll_adapt_ibcast(buff, count, datatype, root, comm, &request, module);
-        ompi_request_wait(&request, MPI_STATUS_IGNORE);
-        return err;
     }
+    ompi_request_t *request = NULL;
+    int err = ompi_coll_adapt_ibcast(buff, count, datatype, root, comm, &request, module);
+    if( MPI_SUCCESS != err ) {
+        if( NULL == request )
+            return err;
+    }
+    ompi_request_wait(&request, MPI_STATUS_IGNORE);
+    return err;
 }

ompi/mca/coll/adapt/coll_adapt_component.c

@@ -65,11 +65,10 @@ mca_coll_adapt_component_t mca_coll_adapt_component = {
 
     /* adapt-component specific information */
 
-    /* (default) priority */
-    0,
+    0, /* (default) priority */
 
-    /* (default) verbose level */
-    0,
+    0, /* (default) output stream */
+    0, /* (default) verbose level */
 
     /* default values for non-MCA parameters */
     /* Not specifying values here gives us all 0's */
@@ -78,25 +77,13 @@ mca_coll_adapt_component_t mca_coll_adapt_component = {
 /* Open the component */
 static int adapt_open(void)
 {
-    int param;
     mca_coll_adapt_component_t *cs = &mca_coll_adapt_component;
 
-    /*
-     * Get the global coll verbosity: it will be ours
-     */
-    param = mca_base_var_find("ompi", "coll", "base", "verbose");
-    if (param >= 0) {
-        const int *verbose = NULL;
-        mca_base_var_get_value(param, &verbose, NULL, NULL);
-        if (verbose && verbose[0] > 0) {
-            cs->adapt_output = opal_output_open(NULL);
-            opal_output_set_verbosity(cs->adapt_output, verbose[0]);
-        }
+    if (cs->adapt_verbose > 0) {
+        cs->adapt_output = opal_output_open(NULL);
+        opal_output_set_verbosity(cs->adapt_output, cs->adapt_verbose);
     }
 
-    opal_output_verbose(1, cs->adapt_output,
-                        "coll:adapt:component_open: done!");
-
     return OMPI_SUCCESS;
 }
 
@@ -131,40 +118,38 @@ static int adapt_register(void)
                                            OPAL_INFO_LVL_9,
                                            MCA_BASE_VAR_SCOPE_READONLY, &cs->adapt_priority);
 
-    int adapt_verbose = 0;
+    cs->adapt_verbose = ompi_coll_base_framework.framework_verbose;
     (void) mca_base_component_var_register(c, "verbose",
-                                           "Verbose level",
+                                           "Verbose level (default set to the collective framework verbosity)",
                                            MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
                                            OPAL_INFO_LVL_9,
-                                           MCA_BASE_VAR_SCOPE_READONLY, &adapt_verbose);
-    cs->adapt_output = opal_output_open(NULL);
-    opal_output_set_verbosity(cs->adapt_output, adapt_verbose);
+                                           MCA_BASE_VAR_SCOPE_READONLY, &cs->adapt_verbose);
 
-    cs->adapt_context_free_list_min = 10;
+    cs->adapt_context_free_list_min = 64;
     (void) mca_base_component_var_register(c, "context_free_list_min",
                                            "Minimum number of segments in context free list",
                                            MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
                                            OPAL_INFO_LVL_9,
                                            MCA_BASE_VAR_SCOPE_READONLY,
                                            &cs->adapt_context_free_list_min);
 
-    cs->adapt_context_free_list_max = 10000;
+    cs->adapt_context_free_list_max = 1024;
     (void) mca_base_component_var_register(c, "context_free_list_max",
                                            "Maximum number of segments in context free list",
                                            MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
                                            OPAL_INFO_LVL_9,
                                            MCA_BASE_VAR_SCOPE_READONLY,
                                            &cs->adapt_context_free_list_max);
 
-    cs->adapt_context_free_list_inc = 10;
+    cs->adapt_context_free_list_inc = 32;
     (void) mca_base_component_var_register(c, "context_free_list_inc",
                                            "Increasement number of segments in context free list",
                                            MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
                                            OPAL_INFO_LVL_9,
                                            MCA_BASE_VAR_SCOPE_READONLY,
                                            &cs->adapt_context_free_list_inc);
-    ompi_coll_adapt_ibcast_init();
-    ompi_coll_adapt_ireduce_init();
+    ompi_coll_adapt_ibcast_register();
+    ompi_coll_adapt_ireduce_register();
 
     return adapt_verify_mca_variables();
 }

ompi/mca/coll/adapt/coll_adapt_context.c

@@ -12,58 +12,15 @@
 #include "ompi/mca/coll/coll.h"
 #include "coll_adapt_context.h"
 
-static void ompi_coll_adapt_bcast_context_constructor(ompi_coll_adapt_bcast_context_t * bcast_context)
-{
-}
-
-static void ompi_coll_adapt_bcast_context_destructor(ompi_coll_adapt_bcast_context_t * bcast_context)
-{
-}
-
-static void
-ompi_coll_adapt_constant_bcast_context_constructor(ompi_coll_adapt_constant_bcast_context_t * con)
-{
-}
-
-static void ompi_coll_adapt_constant_bcast_context_destructor(ompi_coll_adapt_constant_bcast_context_t
-                                                             * con)
-{
-}
-
 
 OBJ_CLASS_INSTANCE(ompi_coll_adapt_bcast_context_t, opal_free_list_item_t,
-                   ompi_coll_adapt_bcast_context_constructor,
-                   ompi_coll_adapt_bcast_context_destructor);
+                   NULL, NULL);
 
 OBJ_CLASS_INSTANCE(ompi_coll_adapt_constant_bcast_context_t, opal_object_t,
-                   ompi_coll_adapt_constant_bcast_context_constructor,
-                   ompi_coll_adapt_constant_bcast_context_destructor);
-
-static void ompi_coll_adapt_reduce_context_constructor(ompi_coll_adapt_reduce_context_t *
-                                                      reduce_context)
-{
-}
-
-static void ompi_coll_adapt_reduce_context_destructor(ompi_coll_adapt_reduce_context_t *
-                                                     reduce_context)
-{
-}
-
-static void
-ompi_coll_adapt_constant_reduce_context_constructor(ompi_coll_adapt_constant_reduce_context_t * con)
-{
-}
-
-static void
-ompi_coll_adapt_constant_reduce_context_destructor(ompi_coll_adapt_constant_reduce_context_t * con)
-{
-}
-
+                   NULL, NULL);
 
 OBJ_CLASS_INSTANCE(ompi_coll_adapt_reduce_context_t, opal_free_list_item_t,
-                   ompi_coll_adapt_reduce_context_constructor,
-                   ompi_coll_adapt_reduce_context_destructor);
+                   NULL, NULL);
 
 OBJ_CLASS_INSTANCE(ompi_coll_adapt_constant_reduce_context_t, opal_object_t,
-                   ompi_coll_adapt_constant_reduce_context_constructor,
-                   ompi_coll_adapt_constant_reduce_context_destructor);
+                   NULL, NULL);