performance optimization: process completions in the batch manner

This commit was SVN r27559.

ompi/mca/btl/openib/btl_openib.h

@@ -57,6 +57,7 @@ BEGIN_C_DECLS
 #define MCA_BTL_IB_LEAVE_PINNED 1
 #define IB_DEFAULT_GID_PREFIX 0xfe80000000000000ll
 #define MCA_BTL_IB_PKEY_MASK 0x7fff
+#define MCA_BTL_OPENIB_CQ_POLL_BATCH_DEFAULT (256)
 
 
 /*--------------------------------------------------------------------*/
@@ -278,6 +279,7 @@ struct mca_btl_openib_component_t {
     bool use_message_coalescing;
     uint32_t cq_poll_ratio;
     uint32_t cq_poll_progress;
+    uint32_t cq_poll_batch;
     uint32_t eager_rdma_poll_ratio;
 #ifdef HAVE_IBV_FORK_INIT
     /** Whether we want fork support or not */

ompi/mca/btl/openib/btl_openib_component.c

@@ -3461,12 +3461,13 @@ static int poll_device(mca_btl_openib_device_t* device, int count)
 {
     int ne = 0, cq;
     uint32_t hp_iter = 0;
-    struct ibv_wc wc;
+    struct ibv_wc wc[MCA_BTL_OPENIB_CQ_POLL_BATCH_DEFAULT];
+    int i;
 
     device->pollme = false;
     for(cq = 0; cq < 2 && hp_iter < mca_btl_openib_component.cq_poll_progress;)
     {
-        ne = ibv_poll_cq(device->ib_cq[cq], 1, &wc);
+        ne = ibv_poll_cq(device->ib_cq[cq], mca_btl_openib_component.cq_poll_batch, wc);
         if(0 == ne) {
             /* don't check low prio cq if there was something in high prio cq,
              * but for each cq_poll_ratio hp cq polls poll lp cq once */
@@ -3488,7 +3489,8 @@ static int poll_device(mca_btl_openib_device_t* device, int count)
             device->hp_cq_polls--;
         }
 
-        handle_wc(device, cq, &wc);
+        for (i = 0; i < ne; i++)
+            handle_wc(device, cq, &wc[i]);
     }
 
     return count;

ompi/mca/btl/openib/btl_openib_mca.c

@@ -174,6 +174,12 @@ int btl_openib_register_mca_params(void)
                   0, &ival, 0));
     mca_btl_openib_component.abort_not_enough_reg_mem = (0 != ival);
 
+    CHECK(reg_int("poll_cq_batch", NULL,
+                  "Retrieve up to poll_cq_batch completions from CQ",
+                  MCA_BTL_OPENIB_CQ_POLL_BATCH_DEFAULT, &ival, REGINT_GE_ONE));
+
+    mca_btl_openib_component.cq_poll_batch = (ival > MCA_BTL_OPENIB_CQ_POLL_BATCH_DEFAULT)? MCA_BTL_OPENIB_CQ_POLL_BATCH_DEFAULT : ival;
+
     if (OMPI_HAVE_IBV_FORK_INIT) {
         ival2 = -1;
     } else {