Fix #428

Use the avifContainerDump() function in the info() function in avifdec.c
to dump the container info instead of the first frame's info.

I noticed this issue while trying to debug monochrome images encoded
with libaom before commit 2172ed0.
The container yuvFormat was YUV400 but the first frame's yuvFormat was
YUV420. avifdec -i printed YUV420, so I could not see the container
yuvFormat was wrong. Obviously it would be best to dump both the
container info and every frame's info, but that could be confusing to
average users, or at least too verbose.

* Ext: Update to dav1d 0.8.0

Release notes: https://code.videolan.org/videolan/dav1d/-/tags/0.8.0

* Changelog: Add entry for dav1d 0.8.0 update

It was released on 2020-11-28:
https://github.com/AOMediaCodec/SVT-AV1/releases/tag/v0.8.6

* Export symbols in shared lib

* Minor clean on cmake script

* Use header macros only

… iprp dedup

…code

Fixes: #443

(Mentioned in #443 conversation)

Add 'static' to function signatures of avifImageSplitGrid() and
avifEncoderFindExistingChunk(). This fix was suggested by Pascal
Massimino.

Change the type of the cellImages parameter of avifEncoderAddImageGrid()
from:
    const avifImage ** cellImages
to
    const avifImage * const * cellImages
so that we can cast a avifImage ** pointer to the new type without the
-Wcast-qual warning. This fix was suggested by Yuan Tong.

Fix https://crbug.com/oss-fuzz/28622.

* Better handling for export headers

* More refinement on export headers.

* Add comment to explain export macros.

* Minor change on comment

Fix some clang warnings

- Disable `-Wused-but-marked-unused` on Clang on MINGW. This solves #435.
  I created an issue but didn't get response from MINGW maintainers for a week, so for now just disable the warning in this case.

- Remove extra semicolon in `codec_svt.c`. This fixes `-Wextra-semi-stmt` warning.

Set the gridCols and gridRows fields of gridAlphaItem.

Pass cellIndex, not 0, as the last argument to the
avifEncoderDataCreateItem(encoder->data, "av01", "Alpha", 6, 0) call.

Change "MS CL" to "MSVC" in the status message and comment.

Merge two target_compile_definitions() calls on the avif target.

The gridX and gridY parameters of avifImageDump() are renamed gridCols
and gridRows, respectively.

Declare the cellIndex parameter of avifEncoderDataCreateItem() as
uint32_t to be consistent with the type of cellIndex elsewhere.

Add parentheses around the expressions to be cast to uint8_t.

Use a simpler equivalent expression to calculate mdatEndSearchOffset.

Declare imageWidth and imageHeight right before first use.

Remove an unneeded uint32_t cast when assigning the return value of
avifRWStreamOffset() to chunkOffset, which is no longer of the uint32_t
type.

Simplify some conditionals in avifEncoderAddImageInternal().

Since we require all cellImages have the same width, height, and depth,
move the checks for invalid values of width, height, and depth outside
the for loop and perform those checks on just firstCell.

Check gridCount > 1 instead of whether gridColorID or gridAlphaID is
nonzero. gridCount > 1 is an equivalent condition and is used elsewhere
in the function.

In the single-image case, we only need to check if the alpha planes of
cellImages have fully opaque values if firstCell has an alpha plane. (If
firstCell does not have an alpha plane, then no cellImages have an alpha
plane.)

…guity

`avifImageSplitGrid` split image into grid at wrong place on x-axis on high-depth images.

avifImageSplitGrid() does not need to call avifImageAllocatePlanes() on
the cell images. It can just point the cell images' planes to the pixel
data of gridSplitImage. This issue was reported by Nicholas Hayes
(0xC0000054) in #331.

Fix a bug in aomCodecEncodeImage() when copying pixel data from
avifImage to aom_image_t. The number of bytes per row should be the
width times bytesPerPixel, not image->yuvRowBytes[yuvPlane].

Declare gridCols and gridRows as uint32_t (instead of uint8_t) so that
256 (the maximum legal value) can be represented.

Also make some cleanup changes.

Change-Id: Ic5a2a781d08f00d878a160d8a7aae4f0a563a62f

Update nasm to 2.15.05
Update aom to v2.0.1
Update dav1d to 0.8.0
Update libgav1 to commit b712ad2
Update rav1e to the 0.3 maintenance branch

…il-cpp

…format files

Film grain is a post-decode step in AV1 which adds synthetic noise on the image.

When encoding an image using 8 bit, limited range (this is the best choice If performance and compatibility are both the case), color banding artifact can usually appear on color gradient areas. Adding noise (dithering) can reduce such artifact, but doing such will significantly increase the file size (random data is hard to compress). Film grain uses several parameters to model the noise, so it solves banding without inflating file size.

Fixes: #440

(I adjusted the wrong defaults.)

Fixes: #440

Speed 6 is the fastest speed in libaom's good-quality usage mode. In one
testing Frank Galligan performed on 20 images, the average encoding
times for speeds 3, 4, 6, 8 are roughly:
  - Speed 3: 125 seconds
  - Speed 4:  63 seconds
  - Speed 6:  42 seconds
  - Speed 8:  11 seconds

So speed 6 is likely to save 1/3 of the encoding time compared with
speed 4.

#440

Otherwise AOM_OPTION_xxx looks like a macro or enum constant defined by
libaom.

Add comment so that developers know that avifIO's `write` member is currently unused. They can simply set it to a null pointer.

Old `avifReformatStateYToUNorm` and `avifReformatStateUVToUNorm`
functions quantize YUV value at very first without extra precision, and
can introduce severer than normal banding. This PR changes to follow the
procedure specified in H.273, and can give better result on limited
range images.

Updated formulas passed avifyuv test with maxDrift=1 (only tested 8bit
and 10bit. 12bit is too slow).

dav1d 0.8.1, libgav1 4a89dc3

update abseil-cpp -b lts_2020_09_23
update libgav1 to 4a89dc3

now in sync with docker/build.sh