Change the app_idx type to uint32_t to support users who use large numbers of app_contexts. Set it up as a new typedef so we can change it later without as much effort.

This commit was SVN r22727.

orte/include/orte/types.h

@@ -52,6 +52,10 @@ typedef uint16_t orte_node_rank_t;
 #define ORTE_LOCAL_RANK_INVALID UINT16_MAX
 #define ORTE_NODE_RANK_INVALID  UINT16_MAX
 
+/* index for app_contexts */
+typedef uint32_t orte_app_idx_t;
+#define ORTE_APP_IDX        OPAL_UINT32
+#define ORTE_APP_IDX_MAX    UINT32_MAX
 
 /*
  * general typedefs & structures

orte/mca/odls/base/odls_base_default_fns.c

@@ -89,9 +89,9 @@ int orte_odls_base_default_get_add_procs_data(opal_buffer_t *data,
     orte_job_map_t *map;
     opal_buffer_t *wireup;
     opal_byte_object_t bo, *boptr;
-    int32_t numbytes, *tmp32;
+    int32_t numbytes, *restarts;
     int8_t flag;
-    int8_t *tmp;
+    orte_app_idx_t *app_idx;
     orte_vpid_t i;
     int j;
     orte_daemon_cmd_flag_t command;
@@ -358,7 +358,7 @@ int orte_odls_base_default_get_add_procs_data(opal_buffer_t *data,
     }
     
     /* pack the number of app_contexts for this job */
-    if (ORTE_SUCCESS != (rc = opal_dss.pack(data, &jdata->num_apps, 1, ORTE_STD_CNTR))) {
+    if (ORTE_SUCCESS != (rc = opal_dss.pack(data, &jdata->num_apps, 1, ORTE_APP_IDX))) {
         ORTE_ERROR_LOG(rc);
         return rc;
     }
@@ -386,25 +386,25 @@ int orte_odls_base_default_get_add_procs_data(opal_buffer_t *data,
     free(bo.bytes);
     
     /* transfer and pack the app_idx and restart arrays for this job */
-    tmp = (int8_t*)malloc(jdata->num_procs);
+    app_idx = (orte_app_idx_t*)malloc(jdata->num_procs * sizeof(orte_app_idx_t));
-    tmp32 = (int32_t*)malloc(jdata->num_procs * sizeof(int32_t));
+    restarts = (int32_t*)malloc(jdata->num_procs * sizeof(int32_t));
     for (j=0, i=0; i < jdata->num_procs && j < jdata->procs->size; j++) {
         if (NULL == (proc = (orte_proc_t*)opal_pointer_array_get_item(jdata->procs, j))) {
             continue;
         }
-        tmp[i] = proc->app_idx;
+        app_idx[i] = proc->app_idx;
-        tmp32[i++] = proc->restarts;
+        restarts[i++] = proc->restarts;
     }
-    if (ORTE_SUCCESS != (rc = opal_dss.pack(data, tmp, jdata->num_procs, OPAL_INT8))) {
+    if (ORTE_SUCCESS != (rc = opal_dss.pack(data, app_idx, jdata->num_procs, ORTE_APP_IDX))) {
         ORTE_ERROR_LOG(rc);
         return rc;
     }
-    free(tmp);
+    free(app_idx);
-    if (ORTE_SUCCESS != (rc = opal_dss.pack(data, tmp32, jdata->num_procs, OPAL_INT32))) {
+    if (ORTE_SUCCESS != (rc = opal_dss.pack(data, restarts, jdata->num_procs, OPAL_INT32))) {
         ORTE_ERROR_LOG(rc);
         return rc;
     }
-    free(tmp32);
+    free(restarts);
     
     /* are there cpu_list strings? */
     if (jdata->map->cpu_lists) {
@@ -579,7 +579,7 @@ int orte_odls_base_default_construct_child_list(opal_buffer_t *data,
     opal_buffer_t alert;
     opal_list_item_t *item;
     int8_t flag;
-    int8_t *app_idx=NULL;
+    orte_app_idx_t *app_idx=NULL;
     int32_t *restarts=NULL;
     char **slot_str=NULL;
     orte_jobid_t debugger;
@@ -643,8 +643,8 @@ int orte_odls_base_default_construct_child_list(opal_buffer_t *data,
             }
         }
         /* fake an app_idx array */
-        app_idx = (int8_t*)malloc(jobdat->num_procs * sizeof(int8_t));
+        app_idx = (orte_app_idx_t*)malloc(jobdat->num_procs * sizeof(orte_app_idx_t));
-        memset(app_idx, 0, jobdat->num_procs * sizeof(int8_t));
+        memset(app_idx, 0, jobdat->num_procs * sizeof(orte_app_idx_t));
         /* if we are doing a timing test, store the time the msg was recvd */
         if (orte_timing) {
             jobdat->launch_msg_recvd.tv_sec = orte_daemon_msg_recvd.tv_sec;
@@ -684,8 +684,9 @@ int orte_odls_base_default_construct_child_list(opal_buffer_t *data,
             ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
             goto REPORT_ERROR;
         }
+        cnt = 1;
         if (ORTE_SUCCESS != (rc = opal_dss.unpack(data, orte_odls_globals.debugger->apps,
-                                                  &(orte_odls_globals.debugger->num_apps), ORTE_APP_CONTEXT))) {
+                                                  &cnt, ORTE_APP_CONTEXT))) {
             ORTE_ERROR_LOG(rc);
             goto REPORT_ERROR;
         }
@@ -803,7 +804,7 @@ int orte_odls_base_default_construct_child_list(opal_buffer_t *data,
     }
     /* unpack the number of app_contexts for this job */
     cnt=1;
-    if (ORTE_SUCCESS != (rc = opal_dss.unpack(data, &jobdat->num_apps, &cnt, ORTE_STD_CNTR))) {
+    if (ORTE_SUCCESS != (rc = opal_dss.unpack(data, &jobdat->num_apps, &cnt, ORTE_APP_IDX))) {
         ORTE_ERROR_LOG(rc);
         goto REPORT_ERROR;
     }
@@ -822,7 +823,8 @@ int orte_odls_base_default_construct_child_list(opal_buffer_t *data,
         ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
         goto REPORT_ERROR;
     }
-    if (ORTE_SUCCESS != (rc = opal_dss.unpack(data, jobdat->apps, &jobdat->num_apps, ORTE_APP_CONTEXT))) {
+    cnt = jobdat->num_apps;
+    if (ORTE_SUCCESS != (rc = opal_dss.unpack(data, jobdat->apps, &cnt, ORTE_APP_CONTEXT))) {
         ORTE_ERROR_LOG(rc);
         goto REPORT_ERROR;
     }
@@ -846,10 +848,10 @@ int orte_odls_base_default_construct_child_list(opal_buffer_t *data,
     }
    
     /* allocate memory for app_idx */
-    app_idx = (int8_t*)malloc(jobdat->num_procs);
+    app_idx = (orte_app_idx_t*)malloc(jobdat->num_procs * sizeof(orte_app_idx_t));
     /* unpack app_idx in one shot */
     cnt=jobdat->num_procs;
-    if (ORTE_SUCCESS != (rc = opal_dss.unpack(data, app_idx, &cnt, OPAL_INT8))) {
+    if (ORTE_SUCCESS != (rc = opal_dss.unpack(data, app_idx, &cnt, ORTE_APP_IDX))) {
         ORTE_ERROR_LOG(rc);
         goto REPORT_ERROR;
     }
@@ -1269,9 +1271,9 @@ int orte_odls_base_default_launch_local(orte_jobid_t job,
     char *job_str, *vpid_str, *param, *value;
     opal_list_item_t *item;
     orte_app_context_t *app, **apps;
-    orte_std_cntr_t num_apps;
+    orte_app_idx_t i, num_apps;
     orte_odls_child_t *child=NULL;
-    int i, num_processors;
+    int num_processors;
     bool oversubscribed;
     int rc=ORTE_SUCCESS, ret;
     bool launch_failed=true;

orte/mca/odls/base/odls_base_open.c

@@ -79,7 +79,7 @@ static void orte_odls_child_constructor(orte_odls_child_t *ptr)
     ptr->name = NULL;
     ptr->restarts = 0;
     ptr->pid = 0;
-    ptr->app_idx = -1;
+    ptr->app_idx = 0;
     ptr->alive = false;
     ptr->coll_recvd = false;
     /* set the default state to "failed to start" so
@@ -135,7 +135,7 @@ static void orte_odls_job_constructor(orte_odls_job_t *ptr)
 }
 static void orte_odls_job_destructor(orte_odls_job_t *ptr)
 {
-    orte_std_cntr_t i;
+    orte_app_idx_t i;
     
     OBJ_DESTRUCT(&ptr->lock);
     OBJ_DESTRUCT(&ptr->cond);

orte/mca/odls/odls_types.h

@@ -93,7 +93,7 @@ typedef struct {
     orte_process_name_t *name;   /* the OmpiRTE name of the proc */
     int32_t restarts;            /* number of times this proc has been restarted */
     pid_t pid;                   /* local pid of the proc */
-    orte_std_cntr_t app_idx;     /* index of the app_context for this proc */
+    orte_app_idx_t app_idx;      /* index of the app_context for this proc */
     bool alive;                  /* is this proc alive? */
     bool coll_recvd;             /* collective operation recvd */
     orte_proc_state_t state;     /* the state of the process */
@@ -122,7 +122,7 @@ typedef struct orte_odls_job_t {
     orte_jobid_t            jobid;                  /* jobid for this data */
     bool                    launch_msg_processed;   /* launch msg has been fully processed */
     orte_app_context_t      **apps;                 /* app_contexts for this job */
-    orte_std_cntr_t         num_apps;               /* number of app_contexts */
+    orte_app_idx_t          num_apps;               /* number of app_contexts */
     orte_mapping_policy_t   policy;                 /* mapping policy */
     int16_t                 cpus_per_rank;          /* number of cpus/rank */
     int16_t                 stride;                 /* step size between cores of multi-core/rank procs */

orte/runtime/data_type_support/orte_dt_packing_fns.c

@@ -165,7 +165,7 @@ int orte_dt_pack_job(opal_buffer_t *buffer, const void *src,
 
         /* pack the number of apps */
         if (ORTE_SUCCESS != (rc = opal_dss_pack_buffer(buffer,
-                         (void*)(&(jobs[i]->num_apps)), 1, ORTE_STD_CNTR))) {
+                         (void*)(&(jobs[i]->num_apps)), 1, ORTE_APP_IDX))) {
             ORTE_ERROR_LOG(rc);
             return rc;
         }

orte/runtime/data_type_support/orte_dt_size_fns.c

@@ -106,7 +106,8 @@ int orte_dt_size_job(size_t *size, orte_job_t *src, opal_data_type_t type)
 {
     size_t sz;
     int32_t i;
-    orte_app_context_t **apps;
+    orte_app_context_t *app;
+    orte_proc_t *proc;
     
     /* account for the object itself */
     *size = sizeof(orte_job_t);
@@ -114,9 +115,11 @@ int orte_dt_size_job(size_t *size, orte_job_t *src, opal_data_type_t type)
     /* if src is NULL, then that's all we wanted */
     if (NULL == src) return ORTE_SUCCESS;
 
-    apps = (orte_app_context_t**)src->apps->addr;
+    for (i=0; i < src->apps->size; i++) {
-    for (i=0; i < src->num_apps; i++) {
+        if (NULL == (app = (orte_app_context_t*)opal_pointer_array_get_item(src->apps, i))) {
-        opal_dss.size(&sz, apps[i], ORTE_APP_CONTEXT);
+            continue;
+        }
+        opal_dss.size(&sz, app, ORTE_APP_CONTEXT);
         *size += sz;
     }
     
@@ -124,10 +127,11 @@ int orte_dt_size_job(size_t *size, orte_job_t *src, opal_data_type_t type)
     *size += sz;
     
     for (i=0; i < src->procs->size; i++) {
-        if (NULL != src->procs->addr[i]) {
+        if (NULL == (proc = (orte_proc_t*)opal_pointer_array_get_item(src->procs, i))) {
-            orte_dt_size_proc(&sz, (orte_proc_t *) src->procs->addr[i], ORTE_PROC);
+            continue;
-            *size += sz;
         }
+        orte_dt_size_proc(&sz, proc, ORTE_PROC);
+        *size += sz;
     }
 
 #if OPAL_ENABLE_FT == 1

orte/runtime/data_type_support/orte_dt_unpacking_fns.c

@@ -150,9 +150,10 @@ int orte_dt_unpack_job(opal_buffer_t *buffer, void *dest,
                        int32_t *num_vals, opal_data_type_t type)
 {
     int rc;
-    int32_t i, j, n, np, nprocs;
+    int32_t i, n, np, nprocs;
     orte_job_t **jobs;
     orte_proc_t *proc;
+    orte_app_idx_t j;
     
     /* unpack into array of orte_job_t objects */
     jobs = (orte_job_t**) dest;
@@ -176,7 +177,7 @@ int orte_dt_unpack_job(opal_buffer_t *buffer, void *dest,
         /* unpack the num apps */
         n = 1;
         if (ORTE_SUCCESS != (rc = opal_dss_unpack_buffer(buffer,
-                                (&(jobs[i]->num_apps)), &n, ORTE_STD_CNTR))) {
+                                (&(jobs[i]->num_apps)), &n, ORTE_APP_IDX))) {
             ORTE_ERROR_LOG(rc);
             return rc;
         }

orte/runtime/orte_globals.c

@@ -782,7 +782,7 @@ static void orte_proc_construct(orte_proc_t* proc)
     proc->local_rank = ORTE_LOCAL_RANK_INVALID;
     proc->node_rank = ORTE_NODE_RANK_INVALID;
     proc->state = ORTE_PROC_STATE_UNDEF;
-    proc->app_idx = -1;
+    proc->app_idx = 0;
     proc->slot_list = NULL;
     proc->node = NULL;
     proc->nodename = NULL;

orte/runtime/orte_globals.h

@@ -167,7 +167,7 @@ typedef struct {
     /** Parent object */
     opal_object_t super;
     /** Unique index when multiple apps per job */
-    int8_t idx;
+    orte_app_idx_t idx;
     /** Absolute pathname of argv[0] */
     char   *app;
     /** Number of copies of this process that are to be launched */
@@ -341,7 +341,7 @@ typedef struct {
     /* app_context array for this job */
     opal_pointer_array_t *apps;
     /* number of app_contexts in the array */
-    orte_std_cntr_t num_apps;
+    orte_app_idx_t num_apps;
     /* flags to control the launch of this job - see above
      * for description of supported flags
      */
@@ -423,7 +423,7 @@ struct orte_proc_t {
     /* exit code */
     orte_exit_code_t exit_code;
     /* the app_context that generated this proc */
-    int8_t app_idx;
+    orte_app_idx_t app_idx;
     /* a cpu list, if specified by the user */
     char *slot_list;
     /* pointer to the node where this proc is executing */

orte/tools/orterun/debuggers.c

@@ -560,8 +560,8 @@ static void check_debugger(int fd, short event, void *arg)
 void orte_debugger_init_before_spawn(orte_job_t *jdata)
 {
     char *env_name;
-    orte_app_context_t **apps, *app;
+    orte_app_context_t *app;
-    orte_std_cntr_t i;
+    int i;
     int32_t ljob;
 
     if (!MPIR_being_debugged && !orte_in_parallel_debugger) {
@@ -582,12 +582,14 @@ void orte_debugger_init_before_spawn(orte_job_t *jdata)
     }
     
     /* tell the procs they are being debugged */
-    apps = (orte_app_context_t**)jdata->apps->addr;
     env_name = mca_base_param_environ_variable("orte", 
                                                "in_parallel_debugger", NULL);
     
-    for (i=0; i < jdata->num_apps; i++) {
+    for (i=0; i < jdata->apps->size; i++) {
-        opal_setenv(env_name, "1", true, &apps[i]->env);
+        if (NULL == (app = (orte_app_context_t*)opal_pointer_array_get_item(jdata->apps, i))) {
+            continue;
+        }
+        opal_setenv(env_name, "1", true, &app->env);
     }
     free(env_name);
     

orte/util/pre_condition_transports.c

@@ -67,8 +67,8 @@ int orte_pre_condition_transports(orte_job_t *jdata)
     char *cs_env, *string_key = NULL, *format = NULL;
     uint64_t unique_key[2];
     unsigned int *int_ptr;
-    orte_std_cntr_t n;
+    int n;
-    orte_app_context_t **apps;
+    orte_app_context_t *app;
 
 #if !defined(__WINDOWS__)
     int fd_rand;
@@ -152,9 +152,11 @@ int orte_pre_condition_transports(orte_job_t *jdata)
         return ORTE_ERR_OUT_OF_RESOURCE;
     }
     
-    apps = (orte_app_context_t**)jdata->apps->addr;
+    for (n=0; n < jdata->apps->size; n++) {
-    for (n=0; n < jdata->num_apps; n++) {
+        if (NULL == (app = (orte_app_context_t*)opal_pointer_array_get_item(jdata->apps, n))) {
-        opal_setenv(cs_env, string_key, true, &apps[n]->env);
+            continue;
+        }
+        opal_setenv(cs_env, string_key, true, &app->env);
     }
 
     free(cs_env);

orte/util/proc_info.c

@@ -43,7 +43,7 @@ ORTE_DECLSPEC orte_proc_info_t orte_process_info = {
     /*  .my_hnp =               */   {ORTE_JOBID_INVALID, ORTE_VPID_INVALID},
     /*  .my_hnp_uri =           */   NULL,
     /*  .hnp_pid =              */   0,
-    /*  .app_num =              */   -1,
+    /*  .app_num =              */   0,
     /*  .num_procs =            */   1,
     /*  .num_nodes =            */   1,
     /*  .nodename =             */   NULL,
@@ -116,7 +116,7 @@ int orte_proc_info(void)
     
     mca_base_param_reg_int_name("orte", "app_num",
                                 "Index of the app_context that defines this proc",
-                                true, false, -1, &tmp);
+                                true, false, 0, &tmp);
     orte_process_info.app_num = tmp;
     
     /* get the process id */

orte/util/proc_info.h

@@ -82,7 +82,7 @@ struct orte_proc_info_t {
     orte_process_name_t my_hnp;         /**< Name of my hnp */
     char *my_hnp_uri;                   /**< Contact info for my hnp */
     pid_t hnp_pid;                      /**< hnp pid - used if singleton */
-    orte_std_cntr_t app_num;            /**< our index into the app_context array */
+    orte_app_idx_t app_num;             /**< our index into the app_context array */
     orte_vpid_t num_procs;              /**< number of processes in this job */
     int num_nodes;                      /**< number of nodes in the job */
     char *nodename;                     /**< string name for this node */