From: Pavan Nikhilesh Bhagavatula <pbhagavatula@marvell.com>
To: Pavan Nikhilesh Bhagavatula <pbhagavatula@marvell.com>,
Jerin Jacob <jerinj@marvell.com>
Cc: "dev@dpdk.org" <dev@dpdk.org>,
"pravin.pathak@intel.com" <pravin.pathak@intel.com>,
"hemant.agrawal@nxp.com" <hemant.agrawal@nxp.com>,
"sachin.saxena@nxp.com" <sachin.saxena@nxp.com>,
"mattias.ronnblom@ericsson.com" <mattias.ronnblom@ericsson.com>,
Jerin Jacob <jerinj@marvell.com>,
"liangma@liangbit.com" <liangma@liangbit.com>,
"peter.mccarthy@intel.com" <peter.mccarthy@intel.com>,
"harry.van.haaren@intel.com" <harry.van.haaren@intel.com>,
"erik.g.carrillo@intel.com" <erik.g.carrillo@intel.com>,
"abhinandan.gujjar@intel.com" <abhinandan.gujjar@intel.com>,
Amit Prakash Shukla <amitprakashs@marvell.com>,
"s.v.naga.harish.k@intel.com" <s.v.naga.harish.k@intel.com>,
"anatoly.burakov@intel.com" <anatoly.burakov@intel.com>
Subject: RE: [RFC 0/2] introduce event vector adapter
Date: Wed, 26 Mar 2025 17:06:57 +0000 [thread overview]
Message-ID: <CO6PR18MB4084EBB953C30015BCB7183CDEA62@CO6PR18MB4084.namprd18.prod.outlook.com> (raw)
In-Reply-To: <20250326131441.5965-1-pbhagavatula@marvell.com>
++
> -----Original Message-----
> From: pbhagavatula@marvell.com <pbhagavatula@marvell.com>
> Sent: Wednesday, March 26, 2025 6:45 PM
> To: Jerin Jacob <jerinj@marvell.com>
> Cc: dev@dpdk.org; Pavan Nikhilesh Bhagavatula
> <pbhagavatula@marvell.com>
> Subject: [RFC 0/2] introduce event vector adapter
>
> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
>
> The event vector adapter supports offloading the creation of event vectors
> by vectorizing objects (mbufs/ptrs/u64s).
>
> An event vector adapter has the following working model:
>
> ┌──────────┐
> │ Vector ├─┐
> │ adapter0 │ │
> └──────────┘ │
> ┌──────────┐ │ ┌──────────┐
> │ Vector ├─┼──►│ Event │
> │ adapter1 │ │ │ Queue0 │
> └──────────┘ │ └──────────┘
> ┌──────────┐ │
> │ Vector ├─┘
> │ adapter2 │
> └──────────┘
>
> ┌──────────┐
> │ Vector ├─┐
> │ adapter0 │ │ ┌──────────┐
> └──────────┘ ├──►│ Event │
> ┌──────────┐ │ │ Queue1 │
> │ Vector ├─┘ └──────────┘
> │ adapter1 │
> └──────────┘
>
> - A vector adapter can be seen as an extension to event queue. It helps in
> aggregating objects and generating a vector event which is enqueued to the
> event queue.
>
> - Multiple vector adapters can be created on an event queue, each with its
> own unique properties such as event properties, vector size, and timeout.
> Note: If the target event queue doesn't support
> RTE_EVENT_QUEUE_CFG_ALL_TYPES,
> then the vector adapter should use the same schedule type as the event
> queue.
>
> - Each vector adapter aggregates objects, generates a vector event and
> enqueues it to the event queue with the event properties mentioned in
> rte_event_vector_adapter_conf::ev.
>
> - After configuring the vector adapter, the Application needs to use the
> rte_event_vector_adapter_enqueue() function to enqueue objects i.e.,
> mbufs/ptrs/u64s to the vector adapter.
> On reaching the configured vector size or timeout, the vector adapter
> enqueues the event vector to the event queue.
> Note: Application should use the event_type and sub_event_type properly
> identifying the contents of vector event on dequeue.
>
> - If the vector adapter advertises the
> RTE_EVENT_VECTOR_ADAPTER_CAP_SOV_EOV
> capability, application can use the RTE_EVENT_VECTOR_ENQ_[S|E]OV flags
> to indicate the start and end of a vector event.
> * When RTE_EVENT_VECTOR_ENQ_SOV is set, the vector adapter will flush
> any
> aggregation in progress as a vector event and start aggregating a new
> vector event with the enqueued ptr.
> * When RTE_EVENT_VECTOR_ENQ_EOV is set, the vector adapter will add the
> current ptr enqueued to the aggregated event and enqueue the vector event
> to the event queue.
> * If both flags are set, the vector adapter will flush the current aggregation
> as a vector event and enqueue the current ptr as a single event to the event
> queue.
>
> - If the vector adapter reaches the configured vector size, it will enqueue
> the aggregated vector event to the event queue.
>
> - If the vector adapter reaches the configured vector timeout, it will flush
> the current aggregation as a vector event if the minimum vector size is
> reached, if not it will enqueue the objects as single events to the event
> queue.
>
> - If the vector adapter is unable to aggregate the objects into a vector event,
> it will enqueue the objects as single events to the event queue with the event
> properties mentioned in rte_event_vector_adapter_conf::ev_fallback.
>
> Before using the vector adapter, the application has to create and configure
> an event device and based on the event device capability it might require
> creating an additional event port.
>
> When the application creates the vector adapter using the
> ``rte_event_vector_adapter_create()`` function, the event device driver
> capabilities are checked. If an in-built port is absent, the application
> uses the default function to create a new event port.
> For finer control over event port creation, the application should use
> the ``rte_event_vector_adapter_create_ext()`` function.
>
> The application can enqueue one or more objects to the vector adapter using
> the
> ``rte_event_vector_adapter_enqueue()`` function and control the aggregation
> using the flags.
>
> Vector adapters report stats using the
> ``rte_event_vector_adapter_stats_get()``
> function and reset the stats using the
> ``rte_event_vector_adapter_stats_reset()``.
>
> The application can destroy the vector adapter using the
> ``rte_event_vector_adapter_destroy()`` function.
>
> Pavan Nikhilesh (2):
> eventdev: introduce event vector adapter
> eventdev: add default software vector adapter
>
> config/rte_config.h | 1 +
> lib/eventdev/event_vector_adapter_pmd.h | 87 +++
> lib/eventdev/eventdev_pmd.h | 38 ++
> lib/eventdev/meson.build | 3 +
> lib/eventdev/rte_event_vector_adapter.c | 762
> ++++++++++++++++++++++++
> lib/eventdev/rte_event_vector_adapter.h | 469 +++++++++++++++
> lib/eventdev/rte_eventdev.c | 23 +
> lib/eventdev/rte_eventdev.h | 8 +
> lib/eventdev/version.map | 13 +
> 9 files changed, 1404 insertions(+)
> create mode 100644 lib/eventdev/event_vector_adapter_pmd.h
> create mode 100644 lib/eventdev/rte_event_vector_adapter.c
> create mode 100644 lib/eventdev/rte_event_vector_adapter.h
>
> --
> 2.43.0
prev parent reply other threads:[~2025-03-26 17:07 UTC|newest]
Thread overview: 8+ messages / expand[flat|nested] mbox.gz Atom feed top
2025-03-26 13:14 pbhagavatula
2025-03-26 13:14 ` [RFC 1/2] eventdev: " pbhagavatula
2025-03-26 13:14 ` [RFC 2/2] eventdev: add default software " pbhagavatula
2025-03-26 14:18 ` Stephen Hemminger
2025-03-26 17:25 ` [EXTERNAL] " Pavan Nikhilesh Bhagavatula
2025-03-26 20:25 ` Stephen Hemminger
2025-03-26 14:22 ` Stephen Hemminger
2025-03-26 17:06 ` Pavan Nikhilesh Bhagavatula [this message]
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=CO6PR18MB4084EBB953C30015BCB7183CDEA62@CO6PR18MB4084.namprd18.prod.outlook.com \
--to=pbhagavatula@marvell.com \
--cc=abhinandan.gujjar@intel.com \
--cc=amitprakashs@marvell.com \
--cc=anatoly.burakov@intel.com \
--cc=dev@dpdk.org \
--cc=erik.g.carrillo@intel.com \
--cc=harry.van.haaren@intel.com \
--cc=hemant.agrawal@nxp.com \
--cc=jerinj@marvell.com \
--cc=liangma@liangbit.com \
--cc=mattias.ronnblom@ericsson.com \
--cc=peter.mccarthy@intel.com \
--cc=pravin.pathak@intel.com \
--cc=s.v.naga.harish.k@intel.com \
--cc=sachin.saxena@nxp.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).