Michael Niedermayer:
> On Wed, Mar 19, 2025 at 10:18:38PM +0100, Andreas Rheinhardt wrote:
>> First part of a patchset; the second part will be sent separately
>> because the complete set crosses the ML thresholds ("Message body is too
>> big: 1731572 bytes with a limit of 1000 KB"). A complete branch can be
> 
>> found here: https://github.com/mkver/FFmpeg/tree/mpvenc
> 
> tested the branch on mips:
> 
> make -k
> CC	libavcodec/mpegvideo_unquantize.o
> src/libavcodec/mpegvideo_unquantize.c: In function ‘ff_mpv_unquantize_init’:
> src/libavcodec/mpegvideo_unquantize.c:272:34: error: ‘s’ is a pointer; did you mean to use ‘->’?
>      ff_mpv_unquantize_init_mips(s. avctx, q_scale_type);
>                                   ^
>                                   ->
> src/libavcodec/mpegvideo_unquantize.c:272:43: warning: passing argument 2 of ‘ff_mpv_unquantize_init_mips’ makes pointer from integer without a cast [-Wint-conversion]
>      ff_mpv_unquantize_init_mips(s. avctx, q_scale_type);
>                                            ^~~~~~~~~~~~
> In file included from src/libavcodec/mpegvideo_unquantize.c:34:0:
> src/libavcodec/mpegvideo_unquantize.h:60:6: note: expected ‘AVCodecContext * {aka struct AVCodecContext *}’ but argument is of type ‘int’
>  void ff_mpv_unquantize_init_mips(MPVUnquantDSPContext *s, AVCodecContext *avctx,
>       ^~~~~~~~~~~~~~~~~~~~~~~~~~~
> src/libavcodec/mpegvideo_unquantize.c:272:5: error: too few arguments to function ‘ff_mpv_unquantize_init_mips’
>      ff_mpv_unquantize_init_mips(s. avctx, q_scale_type);
>      ^~~~~~~~~~~~~~~~~~~~~~~~~~~
> In file included from src/libavcodec/mpegvideo_unquantize.c:34:0:
> src/libavcodec/mpegvideo_unquantize.h:60:6: note: declared here
>  void ff_mpv_unquantize_init_mips(MPVUnquantDSPContext *s, AVCodecContext *avctx,
>       ^~~~~~~~~~~~~~~~~~~~~~~~~~~
> make: *** [src/ffbuild/common.mak:81: libavcodec/mpegvideo_unquantize.o] Error 1
> make: Target 'all' not remade because of errors.
> 
> [...]

Thanks. I fixed this as well as a bug in the ppc/altivec unquantize
functions and force pushed my branch. Please replace the last two
patches with the ones attached here (or just look at the github branch).

- Andreas

PS: Both mips as well as arm (neon) always use the inter scantable. This
works because the derived raster_end for inter is always >= the
raster_end for intra. I didn't fix MIPS, because I can't seem to compile
MMI; and I didn't change ARM, because it would entail tinkering with
external assembly.