[0/6] net/mlx5: reduce Tx datapath compile time

  The mlx5_rxtx.c file contains a lot of Tx burst functions, each of those is performance-optimized for the specific set of requested offloads.
These ones are generated on the basis of the template function and it takes significant time to compile, just due to a large number of giant functions generated in the same file and this compilation is not being done in parallel with using multithreading.

Therefore, in this series we split the mlx5_rxtx.c file into several separate files to allow different functions to be compiled simultaneously.

Michael Baum (6):
  net/mlx5: separate Rx function declarations to another file
  net/mlx5: separate Rx function implementations to new file
  net/mlx5: separate Tx function declarations to another file
  net/mlx5: separate Tx burst template to header file
  net/mlx5: separate Tx function implementations to new file
  net/mlx5: separate Tx burst functions to different files

 drivers/net/mlx5/linux/mlx5_mp_os.c |    2 +
 drivers/net/mlx5/linux/mlx5_os.c    |    2 +
 drivers/net/mlx5/linux/mlx5_verbs.c |    3 +-
 drivers/net/mlx5/meson.build        |    6 +
 drivers/net/mlx5/mlx5.c             |    2 +
 drivers/net/mlx5/mlx5_devx.c        |    3 +-
 drivers/net/mlx5/mlx5_ethdev.c      |    2 +
 drivers/net/mlx5/mlx5_flow.c        |    3 +-
 drivers/net/mlx5/mlx5_flow_dv.c     |    3 +-
 drivers/net/mlx5/mlx5_flow_verbs.c  |    2 +-
 drivers/net/mlx5/mlx5_mr.c          |    2 +
 drivers/net/mlx5/mlx5_rss.c         |    1 +
 drivers/net/mlx5/mlx5_rx.c          | 1203 ++++++++
 drivers/net/mlx5/mlx5_rx.h          |  598 ++++
 drivers/net/mlx5/mlx5_rxmode.c      |    1 -
 drivers/net/mlx5/mlx5_rxq.c         |    3 +-
 drivers/net/mlx5/mlx5_rxtx.c        | 5468 +----------------------------------
 drivers/net/mlx5/mlx5_rxtx.h        |  915 +-----
 drivers/net/mlx5/mlx5_rxtx_vec.c    |    1 +
 drivers/net/mlx5/mlx5_stats.c       |    3 +-
 drivers/net/mlx5/mlx5_trigger.c     |    3 +-
 drivers/net/mlx5/mlx5_tx.c          |  780 +++++
 drivers/net/mlx5/mlx5_tx.h          | 3734 ++++++++++++++++++++++++
 drivers/net/mlx5/mlx5_tx_empw.c     |   71 +
 drivers/net/mlx5/mlx5_tx_mpw.c      |   34 +
 drivers/net/mlx5/mlx5_tx_nompw.c    |   71 +
 drivers/net/mlx5/mlx5_tx_txpp.c     |   45 +
 drivers/net/mlx5/mlx5_txpp.c        |    3 +-
 drivers/net/mlx5/mlx5_txq.c         |    3 +-
 drivers/net/mlx5/mlx5_vlan.c        |    1 +
 drivers/net/mlx5/windows/mlx5_os.c  |    2 +
 31 files changed, 6581 insertions(+), 6389 deletions(-)
 create mode 100644 drivers/net/mlx5/mlx5_rx.c
 create mode 100644 drivers/net/mlx5/mlx5_rx.h
 create mode 100644 drivers/net/mlx5/mlx5_tx.c
 create mode 100644 drivers/net/mlx5/mlx5_tx.h
 create mode 100644 drivers/net/mlx5/mlx5_tx_empw.c
 create mode 100644 drivers/net/mlx5/mlx5_tx_mpw.c
 create mode 100644 drivers/net/mlx5/mlx5_tx_nompw.c
 create mode 100644 drivers/net/mlx5/mlx5_tx_txpp.c
  

On Mon, Apr 5, 2021 at 4:01 PM Michael Baum <michaelba@nvidia.com> wrote:
>
> The mlx5_rxtx.c file contains a lot of Tx burst functions, each of those is performance-optimized for the specific set of requested offloads.
> These ones are generated on the basis of the template function and it takes significant time to compile, just due to a large number of giant functions generated in the same file and this compilation is not being done in parallel with using multithreading.
>
> Therefore, in this series we split the mlx5_rxtx.c file into several separate files to allow different functions to be compiled simultaneously.
>
> Michael Baum (6):
>   net/mlx5: separate Rx function declarations to another file
>   net/mlx5: separate Rx function implementations to new file
>   net/mlx5: separate Tx function declarations to another file
>   net/mlx5: separate Tx burst template to header file
>   net/mlx5: separate Tx function implementations to new file
>   net/mlx5: separate Tx burst functions to different files

Please fix the indentation in meson.build updates.
Out of curiosity, what is the gain in compilation time?
  

> -----Original Message-----
> From: David Marchand <david.marchand@redhat.com>
> Sent: Tuesday, April 6, 2021 12:34
> To: Michael Baum <michaelba@nvidia.com>
> Cc: dev <dev@dpdk.org>; Matan Azrad <matan@nvidia.com>; Raslan
> Darawsheh <rasland@nvidia.com>; Slava Ovsiienko
> <viacheslavo@nvidia.com>
> Subject: Re: [dpdk-dev] [PATCH 0/6] net/mlx5: reduce Tx datapath compile
> time
> 
> On Mon, Apr 5, 2021 at 4:01 PM Michael Baum <michaelba@nvidia.com>
> wrote:
> >
> > The mlx5_rxtx.c file contains a lot of Tx burst functions, each of those is
> performance-optimized for the specific set of requested offloads.
> > These ones are generated on the basis of the template function and it
> takes significant time to compile, just due to a large number of giant
> functions generated in the same file and this compilation is not being done in
> parallel with using multithreading.
> >
> > Therefore, in this series we split the mlx5_rxtx.c file into several separate
> files to allow different functions to be compiled simultaneously.
> >
> > Michael Baum (6):
> >   net/mlx5: separate Rx function declarations to another file
> >   net/mlx5: separate Rx function implementations to new file
> >   net/mlx5: separate Tx function declarations to another file
> >   net/mlx5: separate Tx burst template to header file
> >   net/mlx5: separate Tx function implementations to new file
> >   net/mlx5: separate Tx burst functions to different files
> 
> Please fix the indentation in meson.build updates.
> Out of curiosity, what is the gain in compilation time?
Om my development setup, with this patch the partial compilation
(for example if we changed the mlx5.h header) takes just several seconds,
before one (especially with full debug option)  - up to 1 minute - all this
this time was eaten by mlx5_rxtx.c.

Also this patch is the preparation step for mlx5 Rx datapath consolidation,
we aare thing about implementing the rx_burst template routine as we
done for the tx_burst, so - it would be nice to resolve the compilation time
issue at early stage.

With best regards,
Slava


> 
> 
> --
> David Marchand