0.5% faster loop
Reviewed-by: Paul B Mahol <onemda@gmail.com>
Reviewed-by: Steven Liu <lingjiujianke@gmail.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

Reviewed-by: Paul B Mahol <onemda@gmail.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

inner reader loop becomes 16% faster
Reviewed-by: Paul B Mahol <onemda@gmail.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

2.5% faster vlc decoding
Reviewed-by: Paul B Mahol <onemda@gmail.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

This is not needed when the buffer is large enough for the worst case of a line

2% faster vlc reading
Reviewed-by: Paul B Mahol <onemda@gmail.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

☭ tests/checkasm/checkasm --bench --test=aacpsdsp
checkasm: using random seed 3318985180
MMX implied by specified flags
MMX implied by specified flags
NEON:
 - aacpsdsp.add_squares        [OK]
 - aacpsdsp.mul_pair_single    [OK]
 - aacpsdsp.hybrid_analysis    [OK]
 - aacpsdsp.stereo_interpolate [OK]
checkasm: all 5 tests passed
nop: 10.0
ps_add_squares_c: 63221.2
ps_add_squares_neon: 22311.7
ps_hybrid_analysis_c: 2466.6
ps_hybrid_analysis_neon: 1521.9
ps_mul_pair_single_c: 68592.0
ps_mul_pair_single_neon: 17426.6
ps_stereo_interpolate_c: 72344.3
ps_stereo_interpolate_neon: 72308.8
ps_stereo_interpolate_ipdopd_c: 117415.2
ps_stereo_interpolate_ipdopd_neon: 113386.3

This includes various fixes and improvements from James Almer.
Signed-off-by: James Almer <jamrial@gmail.com>

The code originally pre-multiply by 2 the steps, causing the running sum
of the h factors to drift away due to the lack of precision. It quickly
causes an inaccuracy > 0.01.

I tried diverse approaches such as multiply by 2.0 (instead of adding
the value itself) without success.

I'm unable to bench the impact of this change, feel free to compare.

This commit fixes the incoming aacpsdsp tests.

Following is an alternative simplified function (matching the incoming
AArch64 code) that may be used:

function ff_ps_stereo_interpolate_neon, export=1
        vld1.32         {q0}, [r2]
        vld1.32         {q1}, [r3]
        ldr             r12, [sp]
        vmov.f32        q8, q0
        vmov.f32        q9, q1
        vzip.32         q8, q0
        vzip.32         q9, q1
1:
        vld1.32         {d4}, [r0,:64]
        vld1.32         {d6}, [r1,:64]
        vadd.f32        q8, q8, q9
        vadd.f32        q0, q0, q1
        vmov.f32        d5, d4
        vmov.f32        d7, d6
        vmul.f32        q2, q2, q8
        vmla.f32        q2, q3, q0
        vst1.32         {d4}, [r0,:64]!
        vst1.32         {d5}, [r1,:64]!
        subs            r12, r12, #1
        bgt             1b
        bx              lr
endfunc

Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

Fixes: runtime error: signed integer overflow: 2147483520 + 128 cannot be represented in type 'int'
Fixes: 2385/clusterfuzz-testcase-minimized-6594333576790016

Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpegSigned-off-by: Michael Niedermayer <michael@niedermayer.cc>

vp9_diag_downright_32x32_12bpp_c: 429.7
vp9_diag_downright_32x32_12bpp_sse2: 158.9
vp9_diag_downright_32x32_12bpp_ssse3: 144.6
vp9_diag_downright_32x32_12bpp_avx: 141.0
vp9_diag_downright_32x32_12bpp_avx2: 73.8

Almost 50% faster than avx implementation
Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>

Process more pixels per loop.
Reviewed-by: Paul B Mahol <onemda@gmail.com>
Signed-off-by: James Almer <jamrial@gmail.com>

Reviewed-by: Paul B Mahol <onemda@gmail.com>
Signed-off-by: James Almer <jamrial@gmail.com>

Hardware accelerated decoding generally uses AVHWFramesContext for pool
allocation of hardware surfaces. These are setup to allocate surfaces
aligned to hardware and hwaccel API requirements. Due to the
architecture, av_hwframe_get_buffer() will return AVFrames with
the dimensions set to the aligned sizes.

This causes some decoders (like hevc) return these aligned size as
final frame size, instead of cropping them to the video's actual
dimensions. To make sure this doesn't happen, crop the frame to the
size the decoder expects when ff_get_buffer() is called.

Merges Libav commit 3fdf50f9.
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

Helpful for debugging.

Merges Libav commit 068eaa53.
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

Basically copied from VLC (LGPL):

http://git.videolan.org/?p=vlc.git;a=blob;f=modules/video_output/win32/direct3d11.c;h=e9fcb83dcabfe778f26e63d19f218caf06a7c3ae;hb=HEAD#l1482
http://git.videolan.org/?p=vlc.git;a=blob;f=modules/codec/avcodec/d3d11va.c;h=85e7d25caebc059a9770da2ef4bb8fe90816d76d;hb=HEAD#l599

Merges Libav commit cfc9e7c9.
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

Some devices (some phones, apparently) will support only this opaque
format. Of course this won't work with CLI, because copying data
directly is not supported.

Automatic frame allocation (setting AVCodecContext.hw_device_ctx) does
not support this mode, even if it's the only supported mode. But since
opaque surfaces are generally less useful, that's probably ok.

Merges Libav commit 5030e385.
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

Makes dealing with formats that can not be used for staging textures
easier (DXGI_FORMAT_420_OPAQUE). It also saves memory if the staging
texture is never needed, so this is a good thing.

Merges Libav commit 98d73e41.
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

It appears in this case, frames_ininit is called twice (once by
av_hwframe_ctx_init(), and again by unreffing the frames ctx ref).

Merges Libav commit 086321c6.
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

Some existed since forever, some are new.

The cast in get_surface() is silly, but unless we change the av_log
function signature, or all callers of ff_dxva2_get_surface_index(), it's
needed to remove the const warning.

Merges Libav commit 752ddb45.
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

Make supported codec profiles part of each dxva_modes entry. Every DXVA2
mode is representative for a codec with a subset of supported profiles,
so reflecting that in dxva_modes seems appropriate.

In practice, this will more strictly check MPEG2 profiles, will stop
relying on the surface format checks for selecting the correct HEVC
profile, and remove the verbose messages for mismatching H264/HEVC
profiles. Instead of the latter, it will now print the more nebulous "No
decoder device for codec found" verbose message.

This also respects AV_HWACCEL_FLAG_ALLOW_PROFILE_MISMATCH. Move the
Main10 HEVC entry before the normal one to make this work better.

Originally inspired by VLC's code.

Merges Libav commit 70e5e7c0.
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

When targeting the UWP API subset, the LoadLibrary function is not
available (and the fallback, LoadPackagedLibrary, can't be used to
load system DLLs). In these cases, link directly to the functions
in the DLLs instead of trying to load them dynamically at runtime.

Merges Libav commit fd1ffa1f.
Signed-off-by: Martin Storsjö <martin@martin.st>

This also adds support to avconv (which is trivial due to the new
hwaccel API being generic enough).

The new decoder setup code in dxva2.c is significantly based on work by
Steve Lhomme <robux4@gmail.com>, but with heavy changes/rewrites.

Merges Libav commit f9e7a2f9.
Also adds untested VP9 support.
The check for DXVA2 COBJs is removed. Just update your MinGW to
something newer than a 5 year old release.
Signed-off-by: Diego Biurrun <diego@biurrun.de>

I want to make it non-mandatory to set a mutex in the D3D11 device
context, and replacing it with user callbacks seems like the best
solution. This is preparation for it. Also makes the code slightly more
readable.

Merges Libav commit 831cfe10.
Signed-off-by: Diego Biurrun <diego@biurrun.de>

The actual hwaccel code will need to access an internal context instead
of avctx->hwaccel_context, so add a new DXVA_CONTEXT() macro, that will
dispatch between the "old" external and the new internal context.

Also, the new API requires a new D3D11 pixfmt, so all places which check
for the pixfmt need to be adjusted. Introduce a ff_dxva2_is_d3d11()
function, which does the check.

Merges Libav commit 4dec101a.
Adds changes to vp9 over the Libav patch.
Signed-off-by: Diego Biurrun <diego@biurrun.de>

So a hwaccel can access avctx->hwaccel in init for whatever reason. This
is for the new d3d hwaccel API. We could create separate entrypoints for
each of the 3 hwaccel types (dxva2, d3d11va, new d3d11va), but this
seems nicer.

Merges Libav commit bd747b92.
Signed-off-by: Diego Biurrun <diego@biurrun.de>

To be used with the new d3d11 hwaccel decode API.

With the new hwaccel API, we don't want surfaces to depend on the
decoder (other than the required dimension and format). The old D3D11VA
pixfmt uses ID3D11VideoDecoderOutputView pointers, which include the
decoder configuration, and thus is incompatible with the new hwaccel
API. This patch introduces AV_PIX_FMT_D3D11, which uses ID3D11Texture2D
and an index. It's simpler and compatible with the new hwaccel API.

The introduced hwcontext supports only the new pixfmt.

Frame upload code untested.

Significantly based on work by Steve Lhomme <robux4@gmail.com>, but with
heavy changes/rewrites.

Merges Libav commit fff90422.
Signed-off-by: Diego Biurrun <diego@biurrun.de>

Yasm:
src/libavfilter/x86/af_volume.asm:24: warning: Standard COFF does not support read-only data sections
src/libavfilter/x86/af_volume.asm:24: warning: Unrecognized qualifier `align'

Nasm:
src/libavfilter/x86/af_volume.asm:24: error: standard COFF does not support section alignment specification
src/libavutil/x86/x86inc.asm:92: ... from macro `SECTION_RODATA' defined here
Tested-by: Clément Bœsch <u@pkh.me>
Signed-off-by: James Almer <jamrial@gmail.com>

Signed-off-by: Paul B Mahol <onemda@gmail.com>

Signed-off-by: Hein-Pieter van Braam <hp@tmm.cx>

This implements the 0x10 frame format for Interplay MVE movies. The
format is a variation on the 0x06 format with some changes. In addition
to the decoding map there's also a skip map. This skip map is used to
determine what 8x8 blocks can change in a particular frame.

This format expects to be able to copy an 8x8 block from before the last
time it was changed. This can be an arbitrary time in the past. In order
to implement this this decoder allocates two additional AVFrames where
actual decoding happens. At the end of a frame decoding changed blocks
are copied to a finished frame based on the skip map.

The skip map's encoding is a little convulted, I'll refer to the code
for details.

Values in the decoding map are the same as in format 0x06.
Signed-off-by: Hein-Pieter van Braam <hp@tmm.cx>

This implements the 0x06 frame format for Interplay MVE movies. The
format is relatively simple. The video data consists of two parts:

16 bits per 8x8 block movement data
a number of 8x8 blocks of pixel data

For each 8x8 block of pixel data the movement data is consulted. There
are 3 possible meanings of the movement data:
* zero     : copy the 8x8 block from the pixel data
* negative : copy the 8x8 block from the previous frame from an offset
             determined by the actual value of the entry -0xC000.
* positive : copy the 8x8 block from the current frame from an offset
             determined by the actual value of the entry -0x4000

Decoding happens in two passes, in the fist pass only new pixeldata is
copied, during the second pass data is copied from the previous and
current frames.

The codec expects that the current frame being decoded to still has the
data from 2 frames ago on it when decoding starts.
Signed-off-by: Hein-Pieter van Braam <hp@tmm.cx>

Interplay MVE can contain up to three different frame formats. They
require different streams of information to render a frame. This patch
changes the IP packet format to prepare for the extra frame formats.
Signed-off-by: Hein-Pieter van Braam <hp@tmm.cx>

Interplay MVE movies have a SEND_BUFFER operation. Only after this
command does the current decoding buffer get displayed. This is required
for the other frame formats. They are fixed-size and can't always encode
a full frame worth of pixeldata.

This code prevents half-finished frames from being emitted.
Signed-off-by: Hein-Pieter van Braam <hp@tmm.cx>

Signed-off-by: Paul B Mahol <onemda@gmail.com>