ompi/request: Support non-PML persistent requests

kawashima-fj · kawashima-fj · commit 0cbdbe32f7b1 · 2017-06-02T13:08:17.000+09:00
This commit adds the `req_start` member to the `ompi_request_t` struct.
The `MPI_START` and `MPI_STARTALL` routines call this callback function
instead of `MCA_PML_CALL(start(...))`. So components that return
persistent request must set this member to their request objects.

`mca_pml_base_module_t::pml_start` is not deleted because
`MCA_PML_CALL(start(...))` is still used elsewhere across OMPI.

Signed-off-by: KAWASHIMA Takahiro &lt;t-kawashima@jp.fujitsu.com&gt;
diff --git a/ompi/mca/pml/bfo/pml_bfo_recvreq.c b/ompi/mca/pml/bfo/pml_bfo_recvreq.c
@@ -154,6 +154,7 @@ static int mca_pml_bfo_recv_request_cancel(struct ompi_request_t* ompi_request,
 static void mca_pml_bfo_recv_request_construct(mca_pml_bfo_recv_request_t* request)
 {
     request->req_recv.req_base.req_type = MCA_PML_REQUEST_RECV;
+    request->req_recv.req_base.req_ompi.req_start = mca_pml_bfo_start;
     request->req_recv.req_base.req_ompi.req_free = mca_pml_bfo_recv_request_free;
     request->req_recv.req_base.req_ompi.req_cancel = mca_pml_bfo_recv_request_cancel;
     request->req_rdma_cnt = 0;
diff --git a/ompi/mca/pml/bfo/pml_bfo_sendreq.c b/ompi/mca/pml/bfo/pml_bfo_sendreq.c
@@ -131,6 +131,7 @@ static int mca_pml_bfo_send_request_cancel(struct ompi_request_t* request, int c
 static void mca_pml_bfo_send_request_construct(mca_pml_bfo_send_request_t* req)
 {
     req->req_send.req_base.req_type = MCA_PML_REQUEST_SEND;
+    req->req_send.req_base.req_ompi.req_start = mca_pml_bfo_start;
     req->req_send.req_base.req_ompi.req_free = mca_pml_bfo_send_request_free;
     req->req_send.req_base.req_ompi.req_cancel = mca_pml_bfo_send_request_cancel;
     req->req_rdma_cnt = 0;
diff --git a/ompi/mca/pml/cm/pml_cm_recvreq.c b/ompi/mca/pml/cm/pml_cm_recvreq.c
@@ -56,6 +56,7 @@ void mca_pml_cm_recv_request_completion(struct mca_mtl_request_t *mtl_request)
 static void
 mca_pml_cm_recv_request_construct(mca_pml_cm_thin_recv_request_t* recvreq)
 {
+    recvreq->req_base.req_ompi.req_start = mca_pml_cm_start;
     recvreq->req_base.req_ompi.req_free = mca_pml_cm_recv_request_free;
     recvreq->req_base.req_ompi.req_cancel = mca_pml_cm_cancel;
     OBJ_CONSTRUCT( &(recvreq->req_base.req_convertor), opal_convertor_t );
diff --git a/ompi/mca/pml/cm/pml_cm_sendreq.c b/ompi/mca/pml/cm/pml_cm_sendreq.c
@@ -63,6 +63,7 @@ mca_pml_cm_send_request_completion(struct mca_mtl_request_t *mtl_request)
 static void mca_pml_cm_send_request_construct(mca_pml_cm_hvy_send_request_t* sendreq)
 {
     /* no need to reinit for every send -- never changes */
+    sendreq->req_send.req_base.req_ompi.req_start = mca_pml_cm_start;
     sendreq->req_send.req_base.req_ompi.req_free = mca_pml_cm_send_request_free;
     sendreq->req_send.req_base.req_ompi.req_cancel = mca_pml_cm_cancel;
 }
diff --git a/ompi/mca/pml/ob1/pml_ob1_recvreq.c b/ompi/mca/pml/ob1/pml_ob1_recvreq.c
@@ -143,6 +143,7 @@ static int mca_pml_ob1_recv_request_cancel(struct ompi_request_t* ompi_request,
 static void mca_pml_ob1_recv_request_construct(mca_pml_ob1_recv_request_t* request)
 {
     /* the request type is set by the superclass */
+    request->req_recv.req_base.req_ompi.req_start = mca_pml_ob1_start;
     request->req_recv.req_base.req_ompi.req_free = mca_pml_ob1_recv_request_free;
     request->req_recv.req_base.req_ompi.req_cancel = mca_pml_ob1_recv_request_cancel;
     request->req_rdma_cnt = 0;
diff --git a/ompi/mca/pml/ob1/pml_ob1_sendreq.c b/ompi/mca/pml/ob1/pml_ob1_sendreq.c
@@ -132,6 +132,7 @@ static int mca_pml_ob1_send_request_cancel(struct ompi_request_t* request, int c
 static void mca_pml_ob1_send_request_construct(mca_pml_ob1_send_request_t* req)
 {
     req->req_send.req_base.req_type = MCA_PML_REQUEST_SEND;
+    req->req_send.req_base.req_ompi.req_start = mca_pml_ob1_start;
     req->req_send.req_base.req_ompi.req_free = mca_pml_ob1_send_request_free;
     req->req_send.req_base.req_ompi.req_cancel = mca_pml_ob1_send_request_cancel;
     req->req_rdma_cnt = 0;
diff --git a/ompi/mca/pml/pml.h b/ompi/mca/pml/pml.h
@@ -69,6 +69,7 @@
 #include "ompi/mca/mca.h"
 #include "mpi.h" /* needed for MPI_ANY_TAG */
 #include "ompi/mca/pml/pml_constants.h"
+#include "ompi/request/request.h"
 
 BEGIN_C_DECLS
 
@@ -350,14 +351,11 @@ typedef int (*mca_pml_base_module_send_fn_t)(
 /**
  * Initiate one or more persistent requests.
  *
- * @param count    Number of requests
- * @param request  Array of persistent requests
- * @return         OMPI_SUCCESS or failure status.
+ * @param count (IN)        Number of requests
+ * @param requests (IN/OUT) Array of persistent requests
+ * @return                  OMPI_SUCCESS or failure status.
  */
-typedef int (*mca_pml_base_module_start_fn_t)(
-    size_t count,
-    struct ompi_request_t** requests
-);
+typedef ompi_request_start_fn_t mca_pml_base_module_start_fn_t;
 
 /**
  * Probe to poll for pending recv.
diff --git a/ompi/mca/pml/ucx/pml_ucx_request.c b/ompi/mca/pml/ucx/pml_ucx_request.c
@@ -136,6 +136,7 @@ static void mca_pml_ucx_request_init_common(ompi_request_t* ompi_req,
     OMPI_REQUEST_INIT(ompi_req, req_persistent);
     ompi_req->req_type             = OMPI_REQUEST_PML;
     ompi_req->req_state            = state;
+    ompi_req->req_start            = mca_pml_ucx_start;
     ompi_req->req_free             = req_free;
     ompi_req->req_cancel           = req_cancel;
     /* This field is used to attach persistant request to a temporary req.
diff --git a/ompi/mca/pml/yalla/pml_yalla_request.c b/ompi/mca/pml/yalla/pml_yalla_request.c
@@ -149,6 +149,7 @@ static void init_base_req(mca_pml_yalla_base_request_t *req)
 {
     OMPI_REQUEST_INIT(&req->ompi, false);
     req->ompi.req_type             = OMPI_REQUEST_PML;
+    req->ompi.req_start            = mca_pml_yalla_start;
     req->ompi.req_cancel           = NULL;
     req->ompi.req_complete_cb      = NULL;
     req->ompi.req_complete_cb_data = NULL;
diff --git a/ompi/mpi/c/start.c b/ompi/mpi/c/start.c
@@ -68,7 +68,7 @@ int MPI_Start(MPI_Request *request)
     case OMPI_REQUEST_PML:
         OPAL_CR_ENTER_LIBRARY();
 
-        ret = MCA_PML_CALL(start(1, request));
+        ret = (*request)->req_start(1, request);
 
         OPAL_CR_EXIT_LIBRARY();
         return ret;
diff --git a/ompi/mpi/c/startall.c b/ompi/mpi/c/startall.c
@@ -44,11 +44,11 @@ static const char FUNC_NAME[] = "MPI_Startall";
 
 int MPI_Startall(int count, MPI_Request requests[])
 {
-    int i;
+    int i, j;
     int ret = OMPI_SUCCESS;
+    ompi_request_start_fn_t start_fn = NULL;
 
     MEMCHECKER(
-        int j;
         for (j = 0; j < count; j++){
             memchecker_request(&requests[j]);
         }
@@ -76,7 +76,7 @@ int MPI_Startall(int count, MPI_Request requests[])
 
     OPAL_CR_ENTER_LIBRARY();
 
-    for (i = 0; i < count; ++i) {
+    for (i = 0, j = -1; i < count; ++i) {
         /* Per MPI it is invalid to start an active request */
         if (OMPI_REQUEST_INACTIVE != requests[i]->req_state) {
             return OMPI_ERRHANDLER_INVOKE(MPI_COMM_WORLD, MPI_ERR_REQUEST, FUNC_NAME);
@@ -91,9 +91,21 @@ int MPI_Startall(int count, MPI_Request requests[])
              */
             requests[i]->req_state = OMPI_REQUEST_ACTIVE;
         }
+
+        /* Call a req_start callback function per requests which have the
+         * same req_start value. */
+        if (requests[i]->req_start != start_fn) {
+            if (NULL != start_fn && i != 0) {
+                start_fn(i - j, requests + j);
+            }
+            start_fn = requests[i]->req_start;
+            j = i;
+        }
     }
 
-    ret = MCA_PML_CALL(start(count, requests));
+    if (NULL != start_fn) {
+        start_fn(i - j, requests + j);
+    }
 
     OPAL_CR_EXIT_LIBRARY();
     return ret;
diff --git a/ompi/request/request.c b/ompi/request/request.c
@@ -55,6 +55,7 @@ static void ompi_request_construct(ompi_request_t* req)
     req->req_state        = OMPI_REQUEST_INVALID;
     req->req_complete     = false;
     req->req_persistent   = false;
+    req->req_start        = NULL;
     req->req_free         = NULL;
     req->req_cancel       = NULL;
     req->req_complete_cb  = NULL;
@@ -123,6 +124,7 @@ int ompi_request_init(void)
     ompi_request_null.request.req_persistent = false;
     ompi_request_null.request.req_f_to_c_index =
         opal_pointer_array_add(&ompi_request_f_to_c_table, &ompi_request_null);
+    ompi_request_null.request.req_start = NULL; /* should not be called */
     ompi_request_null.request.req_free = ompi_request_null_free;
     ompi_request_null.request.req_cancel = ompi_request_null_cancel;
     ompi_request_null.request.req_mpi_object.comm = &ompi_mpi_comm_world.comm;
@@ -155,6 +157,7 @@ int ompi_request_init(void)
     ompi_request_empty.req_persistent = false;
     ompi_request_empty.req_f_to_c_index =
         opal_pointer_array_add(&ompi_request_f_to_c_table, &ompi_request_empty);
+    ompi_request_empty.req_start = NULL; /* should not be called */
     ompi_request_empty.req_free = ompi_request_empty_free;
     ompi_request_empty.req_cancel = ompi_request_null_cancel;
     ompi_request_empty.req_mpi_object.comm = &ompi_mpi_comm_world.comm;
diff --git a/ompi/request/request.h b/ompi/request/request.h
@@ -55,6 +55,26 @@ OMPI_DECLSPEC OBJ_CLASS_DECLARATION(ompi_request_t);
 
 struct ompi_request_t;
 
+/**
+ * Initiate one or more persistent requests.
+ *
+ * This function is called by MPI_START and MPI_STARTALL.
+ *
+ * When called by MPI_START, count is 1.
+ *
+ * When called by MPI_STARTALL, multiple requests which have the same
+ * req_start value are passed. This may help scheduling optimization
+ * of multiple communications.
+ *
+ * @param count (IN)        Number of requests
+ * @param requests (IN/OUT) Array of persistent requests
+ * @return                  OMPI_SUCCESS or failure status.
+ */
+typedef int (*ompi_request_start_fn_t)(
+    size_t count,
+    struct ompi_request_t ** requests
+);
+
 /*
  * Required function to free the request and any associated resources.
  */
@@ -109,6 +129,7 @@ struct ompi_request_t {
     volatile ompi_request_state_t req_state;    /**< enum indicate state of the request */
     bool req_persistent;                        /**< flag indicating if the this is a persistent request */
     int req_f_to_c_index;                       /**< Index in Fortran <-> C translation array */
+    ompi_request_start_fn_t req_start;          /**< Called by MPI_START and MPI_STARTALL */
     ompi_request_free_fn_t req_free;            /**< Called by free */
     ompi_request_cancel_fn_t req_cancel;        /**< Optional function to cancel the request */
     ompi_request_complete_fn_t req_complete_cb; /**< Called when the request is MPI completed */

Original file line number	Diff line number	Diff line change
`@@ -154,6 +154,7 @@ static int mca_pml_bfo_recv_request_cancel(struct ompi_request_t* ompi_request,`
`154`	`154`	`static void mca_pml_bfo_recv_request_construct(mca_pml_bfo_recv_request_t* request)`
`155`	`155`	`{`
`156`	`156`	`request->req_recv.req_base.req_type = MCA_PML_REQUEST_RECV;`
	`157`	`+ request->req_recv.req_base.req_ompi.req_start = mca_pml_bfo_start;`
`157`	`158`	`request->req_recv.req_base.req_ompi.req_free = mca_pml_bfo_recv_request_free;`
`158`	`159`	`request->req_recv.req_base.req_ompi.req_cancel = mca_pml_bfo_recv_request_cancel;`
`159`	`160`	`request->req_rdma_cnt = 0;`
Original file line number	Diff line number	Diff line change
`@@ -131,6 +131,7 @@ static int mca_pml_bfo_send_request_cancel(struct ompi_request_t* request, int c`
`131`	`131`	`static void mca_pml_bfo_send_request_construct(mca_pml_bfo_send_request_t* req)`
`132`	`132`	`{`
`133`	`133`	`req->req_send.req_base.req_type = MCA_PML_REQUEST_SEND;`
	`134`	`+ req->req_send.req_base.req_ompi.req_start = mca_pml_bfo_start;`
`134`	`135`	`req->req_send.req_base.req_ompi.req_free = mca_pml_bfo_send_request_free;`
`135`	`136`	`req->req_send.req_base.req_ompi.req_cancel = mca_pml_bfo_send_request_cancel;`
`136`	`137`	`req->req_rdma_cnt = 0;`
Original file line number	Diff line number	Diff line change
`@@ -56,6 +56,7 @@ void mca_pml_cm_recv_request_completion(struct mca_mtl_request_t *mtl_request)`
`56`	`56`	`static void`
`57`	`57`	`mca_pml_cm_recv_request_construct(mca_pml_cm_thin_recv_request_t* recvreq)`
`58`	`58`	`{`
	`59`	`+ recvreq->req_base.req_ompi.req_start = mca_pml_cm_start;`
`59`	`60`	`recvreq->req_base.req_ompi.req_free = mca_pml_cm_recv_request_free;`
`60`	`61`	`recvreq->req_base.req_ompi.req_cancel = mca_pml_cm_cancel;`
`61`	`62`	`OBJ_CONSTRUCT( &(recvreq->req_base.req_convertor), opal_convertor_t );`
Original file line number	Diff line number	Diff line change
`@@ -63,6 +63,7 @@ mca_pml_cm_send_request_completion(struct mca_mtl_request_t *mtl_request)`
`63`	`63`	`static void mca_pml_cm_send_request_construct(mca_pml_cm_hvy_send_request_t* sendreq)`
`64`	`64`	`{`
`65`	`65`	`/* no need to reinit for every send -- never changes */`
	`66`	`+ sendreq->req_send.req_base.req_ompi.req_start = mca_pml_cm_start;`
`66`	`67`	`sendreq->req_send.req_base.req_ompi.req_free = mca_pml_cm_send_request_free;`
`67`	`68`	`sendreq->req_send.req_base.req_ompi.req_cancel = mca_pml_cm_cancel;`
`68`	`69`	`}`
Original file line number	Diff line number	Diff line change
`@@ -143,6 +143,7 @@ static int mca_pml_ob1_recv_request_cancel(struct ompi_request_t* ompi_request,`
`143`	`143`	`static void mca_pml_ob1_recv_request_construct(mca_pml_ob1_recv_request_t* request)`
`144`	`144`	`{`
`145`	`145`	`/* the request type is set by the superclass */`
	`146`	`+ request->req_recv.req_base.req_ompi.req_start = mca_pml_ob1_start;`
`146`	`147`	`request->req_recv.req_base.req_ompi.req_free = mca_pml_ob1_recv_request_free;`
`147`	`148`	`request->req_recv.req_base.req_ompi.req_cancel = mca_pml_ob1_recv_request_cancel;`
`148`	`149`	`request->req_rdma_cnt = 0;`
Original file line number	Diff line number	Diff line change
`@@ -132,6 +132,7 @@ static int mca_pml_ob1_send_request_cancel(struct ompi_request_t* request, int c`
`132`	`132`	`static void mca_pml_ob1_send_request_construct(mca_pml_ob1_send_request_t* req)`
`133`	`133`	`{`
`134`	`134`	`req->req_send.req_base.req_type = MCA_PML_REQUEST_SEND;`
	`135`	`+ req->req_send.req_base.req_ompi.req_start = mca_pml_ob1_start;`
`135`	`136`	`req->req_send.req_base.req_ompi.req_free = mca_pml_ob1_send_request_free;`
`136`	`137`	`req->req_send.req_base.req_ompi.req_cancel = mca_pml_ob1_send_request_cancel;`
`137`	`138`	`req->req_rdma_cnt = 0;`
Original file line number	Diff line number	Diff line change
`@@ -149,6 +149,7 @@ static void init_base_req(mca_pml_yalla_base_request_t *req)`
`149`	`149`	`{`
`150`	`150`	`OMPI_REQUEST_INIT(&req->ompi, false);`
`151`	`151`	`req->ompi.req_type = OMPI_REQUEST_PML;`
	`152`	`+ req->ompi.req_start = mca_pml_yalla_start;`
`152`	`153`	`req->ompi.req_cancel = NULL;`
`153`	`154`	`req->ompi.req_complete_cb = NULL;`
`154`	`155`	`req->ompi.req_complete_cb_data = NULL;`