mpi/coll: add TSP-based scatter/ring-based-allgather algo for Ibcast

pmodels · Feb 9, 2022 · a0f1e29 · a0f1e29
1 parent 038496d
commit a0f1e29
Show file tree

Hide file tree

Showing 7 changed files with 58 additions and 9 deletions.
diff --git a/src/mpi/coll/coll_algorithms.txt b/src/mpi/coll/coll_algorithms.txt
@@ -81,8 +81,11 @@ ibcast-intra:
         cvar_params: TREE_TYPE, TREE_KVAL, TREE_PIPELINE_CHUNK_SIZE
     tsp_scatterv_recexch_allgatherv
         func_name: scatterv_allgatherv
-        extra_params: scatterv_k, allgatherv_k
-        cvar_params: SCATTERV_KVAL, ALLGATHERV_RECEXCH_KVAL
+        extra_params: allgatherv_algo=MPIR_CVAR_IALLGATHERV_INTRA_ALGORITHM_tsp_recexch_doubling, scatterv_k, allgatherv_k
+        cvar_params: -, SCATTERV_KVAL, ALLGATHERV_RECEXCH_KVAL
+    tsp_scatterv_ring_allgatherv
+        extra_params: scatterv_k
+        cvar_params:  SCATTERV_KVAL
     tsp_ring
         func_name: tree
         extra_params: tree_type=MPIR_TREE_TYPE_KARY, k=1, chunk_size

diff --git a/src/mpi/coll/cvars.txt b/src/mpi/coll/cvars.txt
@@ -299,6 +299,7 @@ cvars:
         sched_scatter_ring_allgather               - Force Scatter Ring Allgather algorithm
         tsp_tree                               - Force Generic Transport Tree algorithm
         tsp_scatterv_recexch_allgatherv        - Force Generic Transport Scatterv followed by Recursive Exchange Allgatherv algorithm
+        tsp_scatterv_ring_allgatherv           - Force Generic Transport Scatterv followed by Ring Allgatherv algorithm
         tsp_ring                               - Force Generic Transport Ring algorithm
 
     - name        : MPIR_CVAR_IBCAST_SCATTERV_KVAL

diff --git a/src/mpi/coll/ibcast/Makefile.mk b/src/mpi/coll/ibcast/Makefile.mk
@@ -14,6 +14,7 @@ mpi_core_sources += \
     src/mpi/coll/ibcast/ibcast_intra_sched_smp.c                                  \
     src/mpi/coll/ibcast/ibcast_inter_sched_flat.c                                 \
     src/mpi/coll/ibcast/ibcast_tsp_scatterv_allgatherv.c \
+    src/mpi/coll/ibcast/ibcast_tsp_scatterv_ring_allgatherv.c \
     src/mpi/coll/ibcast/ibcast_tsp_tree.c 					  \
     src/mpi/coll/ibcast/ibcast_utils.c
 

diff --git a/src/mpi/coll/ibcast/ibcast_tsp_scatterv_allgatherv.c b/src/mpi/coll/ibcast/ibcast_tsp_scatterv_allgatherv.c
@@ -9,8 +9,9 @@
 /* Routine to schedule a scatter followed by recursive exchange based broadcast */
 int MPIR_TSP_Ibcast_sched_intra_scatterv_allgatherv(void *buffer, MPI_Aint count,
                                                     MPI_Datatype datatype, int root,
-                                                    MPIR_Comm * comm, int scatterv_k,
-                                                    int allgatherv_k, MPIR_TSP_sched_t sched)
+                                                    MPIR_Comm * comm, int allgatherv_algo,
+                                                    int scatterv_k, int allgatherv_k,
+                                                    MPIR_TSP_sched_t sched)
 {
     int mpi_errno = MPI_SUCCESS;
     int mpi_errno_ret ATTRIBUTE((unused)) = MPI_SUCCESS;
@@ -168,13 +169,19 @@ int MPIR_TSP_Ibcast_sched_intra_scatterv_allgatherv(void *buffer, MPI_Aint count
     mpi_errno = MPIR_TSP_sched_fence(sched);    /* wait for scatter to complete */
     MPIR_ERR_COLL_CHECKANDCONT(mpi_errno, errflag);
 
-    /* Schedule Allgatherv */
-    mpi_errno =
-        MPIR_TSP_Iallgatherv_sched_intra_recexch(MPI_IN_PLACE, cnts[rank], MPI_BYTE, tmp_buf, cnts,
-                                                 displs, MPI_BYTE, comm, 0, allgatherv_k, sched);
+    if (allgatherv_algo == MPIR_CVAR_IALLGATHERV_INTRA_ALGORITHM_tsp_ring)
+        /* Schedule Allgatherv ring */
+        mpi_errno =
+            MPIR_TSP_Iallgatherv_sched_intra_ring(MPI_IN_PLACE, cnts[rank], MPI_BYTE, tmp_buf,
+                                                  cnts, displs, MPI_BYTE, comm, sched);
+    else
+        /* Schedule Allgatherv recexch */
+        mpi_errno =
+            MPIR_TSP_Iallgatherv_sched_intra_recexch(MPI_IN_PLACE, cnts[rank], MPI_BYTE, tmp_buf,
+                                                     cnts, displs, MPI_BYTE, comm, 0, allgatherv_k,
+                                                     sched);
     MPIR_ERR_CHECK(mpi_errno);
 
-
     if (!is_contig) {
         if (rank != root) {
             mpi_errno = MPIR_TSP_sched_sink(sched, &sink_id);   /* wait for allgather to complete */

diff --git a/src/mpi/coll/ibcast/ibcast_tsp_scatterv_ring_allgatherv.c b/src/mpi/coll/ibcast/ibcast_tsp_scatterv_ring_allgatherv.c
@@ -0,0 +1,31 @@
+/*
+ * Copyright (C) by Argonne National Laboratory
+ *     See COPYRIGHT in top-level directory
+ */
+
+#include "mpiimpl.h"
+#include "algo_common.h"
+
+/* Routine to schedule a scatter followed by ring based broadcast */
+int MPIR_TSP_Ibcast_sched_intra_scatterv_ring_allgatherv(void *buffer, MPI_Aint count,
+                                                         MPI_Datatype datatype, int root,
+                                                         MPIR_Comm * comm, int scatterv_k,
+                                                         MPIR_TSP_sched_t sched)
+{
+    int mpi_errno = MPI_SUCCESS;
+
+    MPIR_FUNC_ENTER;
+
+    mpi_errno = MPIR_TSP_Ibcast_sched_intra_scatterv_allgatherv(buffer, count, datatype, root,
+                                                                comm,
+                                                                MPIR_CVAR_IALLGATHERV_INTRA_ALGORITHM_tsp_ring,
+                                                                scatterv_k, 0, sched);
+    MPIR_ERR_CHECK(mpi_errno);
+
+  fn_exit:
+    MPIR_FUNC_EXIT;
+    return mpi_errno;
+  fn_fail:
+    goto fn_exit;
+
+}
diff --git a/src/mpi/coll/include/csel_container.h b/src/mpi/coll/include/csel_container.h
@@ -115,6 +115,7 @@ typedef enum {
     MPII_CSEL_CONTAINER_TYPE__ALGORITHM__MPIR_Ibarrier_inter_sched_bcast,
     MPII_CSEL_CONTAINER_TYPE__ALGORITHM__MPIR_Ibcast_intra_tsp_tree,
     MPII_CSEL_CONTAINER_TYPE__ALGORITHM__MPIR_Ibcast_intra_tsp_scatterv_recexch_allgatherv,
+    MPII_CSEL_CONTAINER_TYPE__ALGORITHM__MPIR_Ibcast_intra_tsp_scatterv_ring_allgatherv,
     MPII_CSEL_CONTAINER_TYPE__ALGORITHM__MPIR_Ibcast_intra_tsp_ring,
     MPII_CSEL_CONTAINER_TYPE__ALGORITHM__MPIR_Ibcast_intra_sched_binomial,
     MPII_CSEL_CONTAINER_TYPE__ALGORITHM__MPIR_Ibcast_intra_sched_scatter_recursive_doubling_allgather,
@@ -287,6 +288,9 @@ typedef struct {
                 int scatterv_k;
                 int allgatherv_k;
             } intra_tsp_scatterv_recexch_allgatherv;
+            struct {
+                int scatterv_k;
+            } intra_tsp_scatterv_ring_allgatherv;
         } ibcast;
         struct {
             struct {

diff --git a/test/mpi/maint/coll_cvars.txt b/test/mpi/maint/coll_cvars.txt
@@ -350,6 +350,8 @@ algorithms:
             tsp_scatterv_recexch_allgatherv
                 .MPIR_CVAR_IBCAST_SCATTERV_KVAL=3
                 .MPIR_CVAR_IBCAST_ALLGATHER_RECEXCH=3
+            tsp_scatterv_ring_allgatherv
+                .MPIR_CVAR_IBCAST_SCATTERV_KVAL=3
             tsp_ring
                 .MPIR_CVAR_IBCAST_RING_CHUNK_SIZE=4096
     exscan: