Merge pull request #1474 from alex-mikheev/topic/ucp_tm_offload_prep

UCP: preparations for the tag matching offload

src/ucp/core/ucp_context.c

@@ -7,6 +7,7 @@
 
 #include "ucp_context.h"
 #include "ucp_request.h"
+#include <ucp/proto/proto.h>
 
 #include <ucs/config/parser.h>
 #include <ucs/algorithm/crc.h>
@@ -125,6 +126,11 @@ static ucs_config_field_t ucp_config_table[] = {
    "y      - Use mutex for multithreading support in UCP.\n",
    ucs_offsetof(ucp_config_t, ctx.use_mt_mutex), UCS_CONFIG_TYPE_BOOL},
 
+  {"TM_THRESH", "1024",
+   "Threshold for using tag matching offload capabilities.\n"
+   "Smaller buffers will not be posted to the transport.",
+   ucs_offsetof(ucp_config_t, ctx.tm_thresh), UCS_CONFIG_TYPE_MEMUNITS},
+
   {NULL}
 };
 
@@ -705,6 +711,12 @@ static ucs_status_t ucp_fill_config(ucp_context_h context,
                                               : UCP_MT_TYPE_SPINLOCK);
     context->config.ext = config->ctx;
 
+    /* Post threshold for tag offload should not be less than ucp_request_hdr_t
+     * size, because this header may be scattered to user buffer in case of
+     * expected SW RNDV protocol. */
+    context->tm.post_thresh = ucs_max(context->config.ext.tm_thresh,
+                                      sizeof(ucp_request_hdr_t));
+
     /* always init MT lock in context even though it is disabled by user,
      * because we need to use context lock to protect ucp_mm_ and ucp_rkey_
      * routines */
@@ -917,3 +929,18 @@ void ucp_context_print_info(ucp_context_h context, FILE *stream)
 
     fprintf(stream, "#\n");
 }
+
+
+void ucp_context_tag_offload_enable(ucp_context_h context)
+{
+    if (ucs_queue_length(&context->tm.offload_ifaces) == 1) {
+        /* Enable offload, because just one tag offload capable interface is present */
+        context->tm.post_thresh = ucs_max(context->config.ext.tm_thresh,
+                                          sizeof(ucp_request_hdr_t));
+    } else {
+        /* Some offload interface/s already configured. Disable TM receive offload,
+         * because multiple offload ifaces are not supported yet. */
+        context->tm.post_thresh = SIZE_MAX;
+    }
+}
+

src/ucp/core/ucp_context.h

@@ -35,6 +35,9 @@ typedef struct ucp_context_config {
     size_t                                 zcopy_thresh;
     /** Estimation of bcopy bandwidth */
     size_t                                 bcopy_bw;
+    /** Threshold for using tag matching offload capabilities. Smaller buffers
+     *  will not be posted to the transport. */
+    size_t                                 tm_thresh;
     /** Maximal size of worker name for debugging */
     unsigned                               max_worker_name;
     /** Atomic mode */
@@ -97,7 +100,7 @@ typedef struct ucp_context {
     ucp_tl_resource_desc_t        *tl_rscs;   /* Array of communication resources */
     ucp_rsc_index_t               num_tls;    /* Number of resources in the array*/
 
-    ucp_tag_match_t               tm;         /* Tag-matching queues */
+    ucp_tag_match_t               tm;         /* Tag-matching queues and offload info */
 
     struct {
 
@@ -161,6 +164,8 @@ extern ucp_am_handler_t ucp_am_handlers[];
 void ucp_dump_payload(ucp_context_h context, char *buffer, size_t max,
                       const void *data, size_t length);
 
+void ucp_context_tag_offload_enable(ucp_context_h context);
+
 uint64_t ucp_context_uct_atomic_iface_flags(ucp_context_h context);
 
 const char * ucp_find_tl_name_by_csum(ucp_context_t *context, uint16_t tl_name_csum);