diff options
author | Klement Sekera <ksekera@cisco.com> | 2020-07-13 13:59:33 +0000 |
---|---|---|
committer | Matthew Smith <mgsmith@netgate.com> | 2020-07-15 16:27:29 +0000 |
commit | 4d119a6f15333e107bd9766112410eaf4ce1a57f (patch) | |
tree | bc4d880bcbf4de552a694dc6673ec672f294d678 | |
parent | a7b963df2758d7c25de366db1999ca1024e12d30 (diff) |
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 <ksekera@cisco.com>
Change-Id: Ib559c74bf8168e3ddd764d51b7e5bcd2a557f591
-rw-r--r-- | src/plugins/nat/in2out_ed.c | 14 | ||||
-rw-r--r-- | src/plugins/nat/out2in_ed.c | 15 |
2 files changed, 27 insertions, 2 deletions
diff --git a/src/plugins/nat/in2out_ed.c b/src/plugins/nat/in2out_ed.c index c59c2d4536c..080f37e7c15 100644 --- 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 index 8286e66ffda..05fc75fac71 100644 --- 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, |