[PATCH v6] graph: mcore: optimize graph search

[David Marchand]

Salut Dodji,

On Mon, Dec 16, 2024 at 2:44 AM Huichao Cai <chcchc88 at 163.com> wrote:
> diff --git a/devtools/libabigail.abignore b/devtools/libabigail.abignore
> index 21b8cd6113..a92ee29512 100644
> --- a/devtools/libabigail.abignore
> +++ b/devtools/libabigail.abignore
> @@ -33,3 +33,8 @@
>  ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
>  ; Temporary exceptions till next major ABI version ;
>  ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
> +[suppress_type]
> +       name = rte_node
> +       has_size_change = no
> +       has_data_member_inserted_between =
> +{offset_of(total_sched_fail), offset_of(xstat_off)}

Here is a suppression rule I suggested but does not have the intended effect.

For the context:

Before the change (that you can find below with the next hunk), we
made sure to zero the whole rte_node object at runtime in the library
allocator.
And the offset of the field next to 'dispatch' is fixed with an
explicit alignas() statement.

        /** Fast schedule area for mcore dispatch model. */
        union {
                alignas(RTE_CACHE_LINE_MIN_SIZE) struct {
                        unsigned int lcore_id;  /**< Node running
lcore. */
                        uint64_t total_sched_objs; /**< Number of
objects scheduled. */
                        uint64_t total_sched_fail; /**< Number of
scheduled failure. */
                } dispatch;
        };

        /** Fast path area cache line 1. */
        alignas(RTE_CACHE_LINE_MIN_SIZE)
        rte_graph_off_t xstat_off; /**< Offset to xstat counters. */

If you want the whole definition, you can have a look at:
https://git.dpdk.org/dpdk/tree/lib/graph/rte_graph_worker_common.h#n87

[...]

> diff --git a/lib/graph/rte_graph_worker_common.h b/lib/graph/rte_graph_worker_common.h
> index d3ec88519d..aef0f65673 100644
> --- a/lib/graph/rte_graph_worker_common.h
> +++ b/lib/graph/rte_graph_worker_common.h
> @@ -110,6 +110,7 @@ struct __rte_cache_aligned rte_node {
>                         unsigned int lcore_id;  /**< Node running lcore. */
>                         uint64_t total_sched_objs; /**< Number of objects scheduled. */
>                         uint64_t total_sched_fail; /**< Number of scheduled failure. */
> +                       struct rte_graph *graph;  /**< Graph corresponding to lcore_id. */
>                 } dispatch;
>         };

Now, the patch adds a new field in the struct {} dispatch field.

Here is what abidiff reports:

$ abidiff --version
abidiff: 2.6.0

$ abidiff --suppr
/home/dmarchan/git/pub/dpdk.org/main/devtools/libabigail.abignore
--no-added-syms --headers-dir1
/home/dmarchan/abi/v24.11/build-gcc-shared/usr/local/include
--headers-dir2 /home/dmarchan/builds/main/build-gcc-shared/install/usr/local/include
/home/dmarchan/abi/v24.11/build-gcc-shared/usr/local/lib64/librte_graph.so.25.0
/home/dmarchan/builds/main/build-gcc-shared/install/usr/local/lib64/librte_graph.so.25.1
Functions changes summary: 0 Removed, 1 Changed (9 filtered out), 0
Added functions
Variables changes summary: 0 Removed, 0 Changed, 0 Added variable

1 function with some indirect sub-type change:

  [C] 'function bool
__rte_graph_mcore_dispatch_sched_node_enqueue(rte_node*,
rte_graph_rq_head*)' at rte_graph_model_mcore_dispatch.c:117:1 has
some indirect sub-type changes:
    parameter 1 of type 'rte_node*' has sub-type changes:
      in pointed to type 'struct rte_node' at rte_graph_worker_common.h:92:1:
        type size hasn't changed
        1 data member changes (2 filtered):
          anonymous data member at offset 1536 (in bits) changed from:
            union {struct {unsigned int lcore_id; uint64_t
total_sched_objs; uint64_t total_sched_fail;} dispatch;}
          to:
            union {struct {unsigned int lcore_id; uint64_t
total_sched_objs; uint64_t total_sched_fail; rte_graph* graph;}
dispatch;}


What would be the best way to suppress this warning?
I tried the following which seems to work, but I prefer to ask for your advice.

[suppress_type]
    name = rte_node
    has_data_member_at = offset_of(total_sched_fail)


Thanks.


-- 
David Marchand