start to implement the send side.

This commit was SVN r2724.

src/mca/ptl/sm/src/ptl_sm.c

@@ -6,6 +6,7 @@
 
 #include <string.h>
 #include <stdlib.h>
+#include <sys/types.h>
 
 #include "util/output.h"
 #include "util/if.h"
@@ -25,6 +26,9 @@
 #include "util/proc_info.h"
 #include "util/printf.h"
 #include "mca/ptl/sm/src/ptl_sm_sendreq.h"
+#include "class/ompi_fifo.h"
+#include "threads/mutex.h"
+#include "datatype/datatype.h"
 
 mca_ptl_sm_t mca_ptl_sm = {
     {
@@ -44,9 +48,9 @@ mca_ptl_sm_t mca_ptl_sm = {
     mca_ptl_sm_add_procs,
     mca_ptl_sm_del_procs,
     mca_ptl_sm_finalize,
-    mca_ptl_sm_send,
+    mca_ptl_sm_send,  /* first fragment send function */
-    mca_ptl_sm_send, /* function */
+    mca_ptl_sm_send,  /* second and subsequent send function */
-    NULL,
+    NULL,  /* get function */
     mca_ptl_sm_matched, /* function called after match is made */
     mca_ptl_sm_send_request_init, /* initialization routine */
     mca_ptl_sm_request_return
@@ -232,11 +236,6 @@ int mca_ptl_sm_add_procs(
                 return_code=OMPI_ERR_OUT_OF_RESOURCE;
                 goto CLEANUP;
             }
-            mca_ptl_sm_component.sm_ctl_header->fifo=
-                (volatile ompi_fifo_t **)
-                ( (char *)(mca_ptl_sm_component.sm_ctl_header->fifo)-
-                  (char *)(mca_ptl_sm_component.sm_mpool->mpool_base()) );
-
             /* allocate vectors of ompi_fifo_t - one per
              * process - offsets will be stored */
             size=n_to_allocate*sizeof(ompi_fifo_t);
@@ -264,26 +263,34 @@ int mca_ptl_sm_add_procs(
                 }
             }
 
+            /* set the fifo address to be a relative address, so that
+             * it can be used by other procs */
+            mca_ptl_sm_component.sm_ctl_header->fifo=
+                (volatile ompi_fifo_t **)
+                ( (char *)(mca_ptl_sm_component.sm_ctl_header->fifo)-
+                  (char *)(mca_ptl_sm_component.sm_mpool->mpool_base()) );
+
+
             /* allow other procs to use this shared memory map */
             mca_ptl_sm_component.mmap_file->map_seg->seg_inited=true;
         }
    
         /* Note:  Need to make sure that proc 0 initializes control
          * structures before any of the other procs can progress */
-        if( 0 != mca_ptl_sm_component.my_smp_rank ) {
+        if( 0 != mca_ptl_sm_component.my_smp_rank ) 
-
+        {
             /* spin unitl local proc 0 initializes the segment */
             while(!mca_ptl_sm_component.mmap_file->map_seg->seg_inited)
-                {
+            { ; }
-                    ;
-                }
         }
 
-        /* Initizlize queue data structures 
+        /* cache the pointer to the 2d fifo array.  This is a virtual
-         *   - proc with lowest local rank does this
+         * address, whereas the address in the virtual memory segment
-         *   - all the rest of the procs block until the queues are
+         * is a relative address */
-         *     initialized
+        mca_ptl_sm_component.fifo=(ompi_fifo_t **)
-         *   - initial queue size is zero */
+            ( (char *)(mca_ptl_sm_component.sm_ctl_header->fifo) +
+              (size_t)(mca_ptl_sm_component.sm_mpool->mpool_base()) );
+
     }
 
     /* free local memory */
@@ -341,6 +348,9 @@ void mca_ptl_sm_request_return(struct mca_ptl_base_module_t* ptl, struct mca_pml
  *  descriptor allocated with the send requests, otherwise obtain
  *  one from the free list. Initialize the fragment and foward
  *  on to the peer.
+ *
+ *  NOTE: this routine assumes that only one sending thread will be accessing
+ *        the send descriptor at a time.
  */
 
 int mca_ptl_sm_send(
@@ -351,7 +361,101 @@ int mca_ptl_sm_send(
     size_t size,
     int flags)
 {
-    return OMPI_SUCCESS;
+    mca_ptl_sm_send_request_t *sm_request;
+    int my_local_smp_rank, peer_local_smp_rank;
+    int return_status=OMPI_SUCCESS;
+    ompi_fifo_t *send_fifo;
+    mca_ptl_base_header_t* hdr;
+    void *sm_data_ptr, *user_data_ptr;
+
+    /* cast to shared memory send descriptor */
+    sm_request=(mca_ptl_sm_send_request_t *)sendreq;
+
+    /* determine if send descriptor is obtained from the cache.  If
+     * so, all the memory resource needed have been obtained */
+    if( !sm_request->super.req_cached) {
+    }
+
+    /* if needed, pack data in payload buffer */
+    if( 0 <= size ) {
+        ompi_convertor_t *convertor;
+        struct iovec address;
+
+        sm_data_ptr=sm_request->req_frag->buff;
+        user_data_ptr=sendreq->req_base.req_addr;
+
+        /* set up the shared memory iovec */
+        address.iov_base=sm_data_ptr;
+        address.iov_len=sm_request->req_frag->buff_length;
+
+        convertor = &sendreq->req_convertor;
+        return_status=ompi_convertor_pack(convertor,&address,1);
+        if( 0 > return_status ) {
+            return OMPI_ERROR;
+        }
+    }
+
+    /* fill in the fragment descriptor */
+    /* get pointer to the fragment header */
+    hdr = &(sm_request->req_frag->super.frag_base.frag_header);
+
+    hdr->hdr_common.hdr_type = MCA_PTL_HDR_TYPE_MATCH;
+    hdr->hdr_common.hdr_flags = flags;
+    hdr->hdr_common.hdr_size = sizeof(mca_ptl_base_match_header_t);
+    hdr->hdr_frag.hdr_frag_seq = 0;
+    hdr->hdr_frag.hdr_src_ptr.lval = 0; /* for VALGRIND/PURIFY - 
+                                           REPLACE WITH MACRO */
+    hdr->hdr_frag.hdr_src_ptr.pval = sendreq;
+    hdr->hdr_frag.hdr_dst_ptr.lval = 0;
+    hdr->hdr_match.hdr_contextid = sendreq->req_base.req_comm->c_contextid;
+    hdr->hdr_match.hdr_src = sendreq->req_base.req_comm->c_my_rank;
+    hdr->hdr_match.hdr_dst = sendreq->req_base.req_peer;
+    hdr->hdr_match.hdr_tag = sendreq->req_base.req_tag;
+    hdr->hdr_match.hdr_msg_length = sendreq->req_bytes_packed;
+
+    /* update the offset within the payload */
+    sendreq->req_offset += size;
+
+    /* 
+     * post the descriptor in the queue - post with the relative
+     * address 
+     */
+    /* see if queues are allocated */
+    my_local_smp_rank=ptl_peer->my_smp_rank;
+    peer_local_smp_rank=ptl_peer->peer_smp_rank;
+    send_fifo=&(mca_ptl_sm_component.fifo
+            [my_local_smp_rank][peer_local_smp_rank]);
+    if(OMPI_CB_FREE == send_fifo->head){
+        /* no queues have been allocated - allocate now */
+        return_status=ompi_fifo_init(mca_ptl_sm_component.size_of_cb_queue,
+                mca_ptl_sm_component.cb_lazy_free_freq,
+                /* at this stage we are not doing anything with memory
+                 * locality */
+                0,0,0,
+                send_fifo, mca_ptl_sm_component.sm_mpool);
+        if( return_status != OMPI_SUCCESS ) {
+            return return_status;
+        }
+    }
+
+    /* post descriptor */
+    /* lock for thread safety - using atomic lock, not mutex, since
+     * we need shared memory access to these lock, and in some pthread
+     * implementation, such mutex's don't work correctly */
+    if( ompi_using_threads() ) {
+        ompi_atomic_lock(&(send_fifo->head_lock));
+    }
+
+    return_status=ompi_fifo_write_to_head(sm_request->req_frag_offset_from_base,
+            send_fifo, mca_ptl_sm_component.sm_mpool,
+            mca_ptl_sm_component.sm_offset);
+
+    /* release threa lock */
+    if( ompi_using_threads() ) {
+        ompi_atomic_unlock(&(send_fifo->head_lock));
+    }
+    /* return */
+    return return_status;
 }
 
 

src/mca/ptl/sm/src/ptl_sm.h

@@ -48,11 +48,6 @@ struct mca_ptl_sm_component_t {
     char* sm_mpool_name;                  /**< name of shared memory pool module */
     mca_mpool_base_module_t* sm_mpool; /**< shared memory pool */
     void* sm_mpool_base;                  /**< base address of shared memory pool */
-    ompi_free_list_t sm_send_requests;    /**< free list of sm send requests -- sendreq + sendfrag */
-    ompi_free_list_t sm_first_frags;       /**< free list of sm first
-                                             fragments */
-    ompi_free_list_t sm_second_frags;       /**< free list of sm second
-                                              and above fragments */
     size_t first_fragment_size;            /**< first fragment size */
     size_t max_fragment_size;            /**< maximum (second and
                                              beyone) fragment size */
@@ -64,6 +59,13 @@ struct mca_ptl_sm_component_t {
                                            file */
     mca_ptl_sm_module_resource_t *sm_ctl_header;  /* control header in
                                                      shared memory */
+    ompi_fifo_t **fifo;    /**< cached copy of the pointer to the 2D
+                             fifo array.  The address in the shared
+                             memory segment sm_ctl_header is a relative,
+                             but this one, in process private memory, is
+                             a real virtual address */
+    size_t size_of_cb_queue; /**< size of each circular buffer queue array */
+    size_t cb_lazy_free_freq; /**< frequency of lazy free */
     size_t sm_offset;        /**< offset to be applied to shared memory
                               addresses */
     int num_smp_procs;      /**< current number of smp procs on this
@@ -119,6 +121,12 @@ extern int mca_ptl_sm_component_progress(
  */
 struct mca_ptl_sm_t {
     mca_ptl_base_module_t  super;       /**< base PTL interface */
+
+    ompi_free_list_t sm_first_frags;    /**< free list of sm first
+                                             fragments */
+    ompi_free_list_t sm_second_frags;   /**< free list of sm second
+                                              and above fragments */
+    ompi_free_list_t sm_send_requests;    /**< free list of sm send requests -- sendreq + sendfrag */
 };
 typedef struct mca_ptl_sm_t mca_ptl_sm_t;
 

src/mca/ptl/sm/src/ptl_sm_address.h

@@ -0,0 +1,20 @@
+/*
+ * $HEADER$
+ */
+/**
+ * @file
+ */
+#ifndef MCA_PTL_ADDRESS_H
+#define MCA_PTL_ADDRESS_H
+
+
+/*
+ * macro to convert virtual address, to address relative to a base
+ * offset
+ */
+#define  RELATIVE_ADDRESS(A,B)  (void *) ( (char *)(A) -  \
+        (size_t)(B) )
+
+
+#endif /* !ADDRESS */
+

src/mca/ptl/sm/src/ptl_sm_component.c

@@ -126,6 +126,23 @@ int mca_ptl_sm_component_open(void)
     mca_ptl_sm_component.fragment_alignment =
         mca_ptl_sm_param_register_int("fragment_alignment",
                 CACHE_LINE_SIZE);
+    mca_ptl_sm_component.size_of_cb_queue =
+        mca_ptl_sm_param_register_int("size_of_cb_queue", 128);
+    mca_ptl_sm_component.cb_lazy_free_freq =
+        mca_ptl_sm_param_register_int("cb_lazy_free_freq", 128);
+    /* make sure that queue size and lazy free frequency are consistent -
+     * want to make sure that slots are freed at a rate they can be
+     * reused, w/o allocating extra new circular buffer fifo arrays */
+    if( (float)(mca_ptl_sm_component.cb_lazy_free_freq) >=
+            0.95*(float)(mca_ptl_sm_component.size_of_cb_queue) ) {
+        /* upper limit */
+        mca_ptl_sm_component.cb_lazy_free_freq=
+            (int)(0.95*(float)(mca_ptl_sm_component.size_of_cb_queue));
+        /* lower limit */
+        if( 0>= mca_ptl_sm_component.cb_lazy_free_freq ) {
+            mca_ptl_sm_component.cb_lazy_free_freq=1;
+        }
+    }
 
     /* default number of extra procs to allow for future growth */
     mca_ptl_sm_component.sm_extra_procs =
@@ -133,8 +150,8 @@ int mca_ptl_sm_component_open(void)
 
     /* initialize objects */
     OBJ_CONSTRUCT(&mca_ptl_sm_component.sm_lock, ompi_mutex_t);
-    OBJ_CONSTRUCT(&mca_ptl_sm_component.sm_send_requests, ompi_free_list_t);
+    OBJ_CONSTRUCT(&mca_ptl_sm.sm_send_requests, ompi_free_list_t);
-    OBJ_CONSTRUCT(&mca_ptl_sm_component.sm_first_frags, ompi_free_list_t);
+    OBJ_CONSTRUCT(&mca_ptl_sm.sm_first_frags, ompi_free_list_t);
    
     return OMPI_SUCCESS;
 }
@@ -147,8 +164,8 @@ int mca_ptl_sm_component_open(void)
 int mca_ptl_sm_component_close(void)
 {
     OBJ_DESTRUCT(&mca_ptl_sm_component.sm_lock);
-    OBJ_DESTRUCT(&mca_ptl_sm_component.sm_send_requests);
+    OBJ_DESTRUCT(&mca_ptl_sm.sm_send_requests);
-    OBJ_DESTRUCT(&mca_ptl_sm_component.sm_first_frags);
+    OBJ_DESTRUCT(&mca_ptl_sm.sm_first_frags);
     return OMPI_SUCCESS;
 }
 
@@ -185,7 +202,7 @@ mca_ptl_base_module_t** mca_ptl_sm_component_init(
     length=sizeof(mca_ptl_sm_frag_t)+mca_ptl_sm_component.fragment_alignment+
         mca_ptl_sm_component.first_fragment_size;
 
-    ompi_free_list_init(&mca_ptl_sm_component.sm_first_frags, length,
+    ompi_free_list_init(&mca_ptl_sm.sm_first_frags, length,
         OBJ_CLASS(mca_ptl_sm_frag_t),
         mca_ptl_sm_component.sm_first_frag_free_list_num,
         mca_ptl_sm_component.sm_first_frag_free_list_max,
@@ -201,7 +218,7 @@ mca_ptl_base_module_t** mca_ptl_sm_component_init(
     length=sizeof(mca_ptl_sm_frag_t)+mca_ptl_sm_component.fragment_alignment+
         mca_ptl_sm_component.max_fragment_size;
 
-    ompi_free_list_init(&mca_ptl_sm_component.sm_second_frags, length,
+    ompi_free_list_init(&mca_ptl_sm.sm_second_frags, length,
         OBJ_CLASS(mca_ptl_sm_frag_t),
         mca_ptl_sm_component.sm_second_frag_free_list_num,
         mca_ptl_sm_component.sm_second_frag_free_list_max,

src/mca/ptl/sm/src/ptl_sm_frag.c

@@ -6,6 +6,7 @@
 #include <sys/errno.h>
 #include "ptl_sm.h"
 #include "ptl_sm_frag.h"
+#include "ptl_sm_address.h"
 
 
 static void mca_ptl_sm_first_frag_construct(mca_ptl_sm_frag_t* frag);
@@ -44,6 +45,9 @@ static void mca_ptl_sm_first_frag_construct(mca_ptl_sm_frag_t* frag)
         mca_ptl_sm_component.fragment_alignment;
     /* align */
     ptr=ptr-(((size_t)ptr)%(mca_ptl_sm_component.fragment_alignment));
+    frag->buff=ptr;
+    frag->buff_offset_from_segment_base=RELATIVE_ADDRESS(
+            ptr,mca_ptl_sm_component.sm_mpool_base);
 }
 
 
@@ -71,6 +75,9 @@ static void mca_ptl_sm_second_frag_construct(mca_ptl_sm_frag_t* frag)
         mca_ptl_sm_component.fragment_alignment;
     /* align */
     ptr=ptr-(((size_t)ptr)%(mca_ptl_sm_component.fragment_alignment));
+    frag->buff=ptr;
+    frag->buff_offset_from_segment_base=RELATIVE_ADDRESS(
+            ptr,mca_ptl_sm_component.sm_mpool_base);
 }
 
 

src/mca/ptl/sm/src/ptl_sm_frag.h

@@ -30,6 +30,9 @@ struct mca_ptl_sm_frag_t {
     mca_ptl_base_recv_frag_t super; /**< base receive fragment descriptor */
     size_t buff_length;   /**< size of buffer */
     void *buff;           /**< pointer to buffer */
+    void *buff_offset_from_segment_base;   /**< pointer to buffer,
+                                             relative to base of the
+                                             shared memory segment */
 };
 typedef struct mca_ptl_sm_frag_t mca_ptl_sm_frag_t;
 

src/mca/ptl/sm/src/ptl_sm_sendreq.c

@@ -8,6 +8,7 @@
 #include "mca/pml/base/pml_base_sendreq.h"
 #include "ptl_sm.h"
 #include "ptl_sm_sendreq.h"
+#include "ptl_sm_address.h"
 
 
 static void mca_ptl_sm_send_request_construct(mca_ptl_sm_send_request_t*);
@@ -43,8 +44,29 @@ int mca_ptl_sm_send_request_init(struct mca_ptl_base_module_t* ptl,
         struct mca_pml_base_send_request_t* request)
 {
 
+    mca_ptl_sm_send_request_t *sm_request;
+    mca_ptl_sm_t *ptl_sm;
+
     int return_value=OMPI_SUCCESS;
 
+    /* cast to shared memory send descriptor */
+    sm_request=(mca_ptl_sm_send_request_t *)request;
+
+    /* cast to shared memory ptl */
+    ptl_sm=(mca_ptl_sm_t *)ptl;
+
+    /* get first fragment descritor from free list - the pointer
+     * returned is valid only in this process, since different
+     * processes may have different base addresses 
+     */
+    sm_request->req_frag=(mca_ptl_sm_frag_t *)ompi_list_get_first(
+            (void *)&(ptl_sm->sm_first_frags));
+    if(NULL == sm_request->req_frag){
+        return_value=OMPI_ERR_OUT_OF_RESOURCE;
+    }
+    sm_request->req_frag_offset_from_base=RELATIVE_ADDRESS(
+            sm_request->req_frag,mca_ptl_sm_component.sm_mpool_base);
+
     return return_value;
 }
 

src/mca/ptl/sm/src/ptl_sm_sendreq.h

@@ -27,6 +27,9 @@ struct mca_ptl_sm_send_request_t {
   
     /* pointer to first fragment descriptor */
     mca_ptl_sm_frag_t *req_frag;
+    /* same pointer, but relative to the base of the shared memory
+       segment */
+    mca_ptl_sm_frag_t *req_frag_offset_from_base;
 
 };
 typedef struct mca_ptl_sm_send_request_t mca_ptl_sm_send_request_t;