nat: add prefetching to in2out_ed/out2in_ed

This saves about 20 clocks/packet in both code paths. Type: improvement Signed-off-by: Klement Sekera <[email protected]> Change-Id: Ib559c74bf8168e3ddd764d51b7e5bcd2a557f591
FDio · Jul 15, 2020 · 4d119a6 · 4d119a6
1 parent a7b963d
commit 4d119a6
Show file tree

Hide file tree

Showing 2 changed files with 27 additions and 2 deletions.
diff --git a/src/plugins/nat/in2out_ed.c b/src/plugins/nat/in2out_ed.c
@@ -943,6 +943,19 @@ nat44_ed_in2out_fast_path_node_fn_inline (vlib_main_t * vm,
       ip_csum_t sum0;
 
       b0 = *b;
+      b++;
+
+      /* Prefetch next iteration. */
+      if (PREDICT_TRUE (n_left_from >= 2))
+	{
+	  vlib_buffer_t *p2;
+
+	  p2 = *b;
+
+	  vlib_prefetch_buffer_header (p2, LOAD);
+
+	  CLIB_PREFETCH (p2->data, CLIB_CACHE_LINE_BYTES, LOAD);
+	}
 
       if (is_output_feature)
 	{
@@ -1158,7 +1171,6 @@ nat44_ed_in2out_fast_path_node_fn_inline (vlib_main_t * vm,
 
       n_left_from--;
       next++;
-      b++;
     }
 
   vlib_buffer_enqueue_to_next (vm, node, from, (u16 *) nexts,

diff --git a/src/plugins/nat/out2in_ed.c b/src/plugins/nat/out2in_ed.c
@@ -705,6 +705,20 @@ nat44_ed_out2in_fast_path_node_fn_inline (vlib_main_t * vm,
       ip_csum_t sum0;
 
       b0 = *b;
+      b++;
+
+      /* Prefetch next iteration. */
+      if (PREDICT_TRUE (n_left_from >= 2))
+	{
+	  vlib_buffer_t *p2;
+
+	  p2 = *b;
+
+	  vlib_prefetch_buffer_header (p2, LOAD);
+
+	  CLIB_PREFETCH (p2->data, CLIB_CACHE_LINE_BYTES, LOAD);
+	}
+
       next[0] = vnet_buffer2 (b0)->nat.arc_next;
 
       vnet_buffer (b0)->snat.flags = 0;
@@ -927,7 +941,6 @@ nat44_ed_out2in_fast_path_node_fn_inline (vlib_main_t * vm,
 
       n_left_from--;
       next++;
-      b++;
     }
 
   vlib_buffer_enqueue_to_next (vm, node, from, (u16 *) nexts,