Given that there are actually ASM functions of this type,
one can't simply remove the ff_ prefix from the definitions
and declare them as static. Yet one can do nearly that
if one keeps the ff_ prefix and removes the declarations
for the (now static) functions defined in dsp_init.c
from hevcdsp.h and if one defines the functions in the correct
order (smaller width first) so that no forward declarations
are necessary (which was already true).
The new declarations avoid nested macros to simplify things.
It nevertheless turned out to be beneficial line-wise.
(It would be possible to avoid most of these declarations:
It is legal to repeat a function declaration without static
if the first declaration declared a function as static.
So if the macros simply declared all the functions that
they call, one could avoid declarations for the functions
that are called. While this is legal C, it unfortuntaly
clashes with GCC's -Wredundant-decls (which configure enables)
and it is also ugly, as these macro definitions would
provide declarations used in ff_hevc_dsp_init_x86().)
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
This also means that if a plane*slice has only 1 color nothing
is stored after the remap table
This also corrects the RCT offset to the exact value after remap
not a fixed 65536
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
The Reference-Only feature in DirectX 12 is a memory optimization
technique designed for video decoding scenarios.
This feature requires that reference resources must be allocated with
the D3D12_RESOURCE_FLAG_VIDEO_DECODE_REFERENCE_ONLY resource flag.
Reference textures must also be separated from output textures.
This feature is not supported in the current version of ffmpeg.
Since AMD GPU uses this feature in Direct 12 decoder,
ffmpeg does not support AMD GPU Direct 12 decoding.
Signed-off-by: Tong Wu <wutong1208@outlook.com>
This reduces needed memory and also removes the 65536 maximum for remap
on the decoder side
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
That is instead of a fixed 65536, we now allocate only as many as there
are pixels.
We also allocate only for the encoder and only when remapping is enabled
and only for 32bit per sample
This should reduce memory consumption, the 2nd array will be
dealt with in a future commit
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
BSF can update extradata, e.g., vvc_mp4toannexb. If we don't copy
bsf->par_out back to avcodec context, decoder can get extradata in
mp4 format, while packets are in annexb format.
mpegvideodec.h includes mpegvideodata.h which in turn contains
a "static const uint8_t *const ff_mpeg1_dc_scale_table =
ff_mpeg12_dc_scale_table[0]"; yet if mpegvideo is disabled,
ff_mpeg12_dc_scale_table is not available, yet a non-optimizing
compiler (like GCC with -O0) may emit ff_mpeg1_dc_scale_table
and therefore a reference to ff_mpeg12_dc_scale_table nevertheless.
Fix this by only including mpegvideodec.h if it is needed.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
The ff_h2656_put_{uni_,}8tap_hv{32,64,128}_8_avx2
and ff_h2656_put_{uni_,}4tap_hv{64,128}_8_avx2
functions were unused and have been removed.
This saved 3712B of .text here.
(ff_h2656_put_{uni_,}4tap_hv32_8_avx2 are now only
called from exactly one callsite (in
ff_hevc_put_{uni_,}epel_hv32_8_avx2) and could be inlined.)
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
big secret, reading a float as the corresponding integer is a good
approximation of log2() for numbers not too close to 1.0. At the same
time it maintains strict monotonicity
this reduces run time from 19sec to 17sec for the slowest of my testcases
(with default remap_optimizer 3)
Compression is about the same:
-rw-r----- 1 michael michael 497603370 Mar 30 15:23 float-303503-fixed-40-optim-3-1log2F.nut
-rw-r----- 1 michael michael 497603374 Mar 28 11:27 float-303503-fixed-40-optim-3b.nut
-rw-r----- 1 michael michael 549938852 Mar 28 11:27 float-303503-float16-40-optim-3b.nut
-rw-r----- 1 michael michael 549938857 Mar 30 15:23 float-303503-float16-40-optim-3-1log2F.nut
-rw-r----- 1 michael michael 1150827841 Mar 28 11:28 float-303503-float-40-optim-3b.nut
-rw-r----- 1 michael michael 1150832913 Mar 30 15:22 float-303503-float-40-optim-3ref-log2F2.nut
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
This data is only used by sbcdsp.c, so delete sbcdsp_data.h,
make a header out of sbcdsp_data.c and make the data contained
therein static.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
It is simpler, avoids several loops and also makes GCC no longer
emit bogus -Wstringop-overflow= warnings.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Doing so gets rid of a stupid GCC -Wstringop-overflow= warning
(GCC somehow believes that fbw_channels can be 7 with the old
form of the code, so that channel_uses_spx[7] would be written
to, but now it no longer believes so).
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
GCC considers declarations using a parameter of pointer
type (or equivalently a parameter using an array of unspecified
dimensions) to be inconsistent with a declaration using
a known-length array type and emits a -Warray-parameter warning
for several ff_j_rev_dct* functions for this.
This patch makes the declarations match the actual definitions
to suppress these (IMO nonsensical) warnings.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Previously, AV1 used filler data with CBR by default while H.264
and HEVC may or may not depending on driver version. Make this
consistent by using not filler data in CBR mode for all codecs.
Since there are valid reasons to use CBR with or without filler,
also add a cbr_padding option to allow users to override this.
Signed-off-by: Cameron Gutman <aicommander@gmail.com>
Signed-off-by: Timo Rothenpieler <timo@rothenpieler.org>
