[FFmpeg-devel] [PATCH 10/17] swscale/ops_backend: add reference backend basend on C templates

[Niklas Haas]

On Fri, 02 May 2025 17:06:27 +0200 Michael Niedermayer <michael at niedermayer.cc> wrote:
> On Sat, Apr 26, 2025 at 07:41:14PM +0200, Niklas Haas wrote:
> > From: Niklas Haas <git at haasn.dev>
> >
> > This will serve as a reference for the SIMD backends to come. That said,
> > with auto-vectorization enabled, the performance of this is not atrocious, and
> > can often beat even the old SIMD.
> >
> > In theory, we can dramatically speed it up by using GCC vectors instead of
> > arrays, but the performance gains from this are too dependent on exact GCC
> > versions and flags, so it practice it's not a substitute for a SIMD
> > implementation.
> > ---
> >  libswscale/Makefile          |   6 +
> >  libswscale/ops.c             |   3 +
> >  libswscale/ops.h             |   2 -
> >  libswscale/ops_backend.c     | 101 ++++++
> >  libswscale/ops_backend.h     | 181 +++++++++++
> >  libswscale/ops_tmpl_common.c | 176 ++++++++++
> >  libswscale/ops_tmpl_float.c  | 255 +++++++++++++++
> >  libswscale/ops_tmpl_int.c    | 609 +++++++++++++++++++++++++++++++++++
> >  8 files changed, 1331 insertions(+), 2 deletions(-)
> >  create mode 100644 libswscale/ops_backend.c
> >  create mode 100644 libswscale/ops_backend.h
> >  create mode 100644 libswscale/ops_tmpl_common.c
> >  create mode 100644 libswscale/ops_tmpl_float.c
> >  create mode 100644 libswscale/ops_tmpl_int.c
>
> arm breaker
>
> CC      libswscale/ops_backend.o
> In file included from src/libswscale/ops_backend.c:21:0:
> src/libswscale/ops_tmpl_int.c:492:12: error: initializer element is not constant
>          fn(op_read_planar1),
>             ^
> src/libswscale/ops_backend.h:78:27: note: in definition of macro ‘bitfn2’
>  #define bitfn2(name, ext) name ## _ ## ext
>                            ^~~~
> src/libswscale/ops_backend.h:82:19: note: in expansion of macro ‘bitfn’
>  #define fn(name)  bitfn(name, FN_SUFFIX)
>                    ^~~~~
> src/libswscale/ops_tmpl_int.c:492:9: note: in expansion of macro ‘fn’
>          fn(op_read_planar1),
>          ^~
> src/libswscale/ops_tmpl_int.c:492:12: note: (near initialization for ‘op_table_int_u8.entries[0]’)
>          fn(op_read_planar1),
>             ^
> src/libswscale/ops_backend.h:78:27: note: in definition of macro ‘bitfn2’
>  #define bitfn2(name, ext) name ## _ ## ext
>                            ^~~~
> src/libswscale/ops_backend.h:82:19: note: in expansion of macro ‘bitfn’
>  #define fn(name)  bitfn(name, FN_SUFFIX)
>                    ^~~~~
> src/libswscale/ops_tmpl_int.c:492:9: note: in expansion of macro ‘fn’
>          fn(op_read_planar1),
>          ^~
> src/libswscale/ops_tmpl_int.c:493:12: error: initializer element is not constant
>          fn(op_read_planar2),
>             ^
> src/libswscale/ops_backend.h:78:27: note: in definition of macro ‘bitfn2’
>  #define bitfn2(name, ext) name ## _ ## ext
>                            ^~~~
> src/libswscale/ops_backend.h:82:19: note: in expansion of macro ‘bitfn’
>  #define fn(name)  bitfn(name, FN_SUFFIX)
>                    ^~~~~
> src/libswscale/ops_tmpl_int.c:493:9: note: in expansion of macro ‘fn’
>          fn(op_read_planar2),
>          ^~
> src/libswscale/ops_tmpl_int.c:493:12: note: (near initialization for ‘op_table_int_u8.entries[1]’)

Fixed (hopefully) by making the op table entries indirect.

>
> ................
>
> [...]
> --
> Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB
>
> In a rich man's house there is no place to spit but his face.
> -- Diogenes of Sinope
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".