Commit Graph

390 Commits

Author SHA1 Message Date
James Zern
ea59a8e980 Merge "Merge tag 'v0.4.0'" 2014-01-10 17:09:19 -08:00
skal
7574bed43d fix comments related to array sizes
thanks to foivos_g4 at hotmail dot com for spotting this.

Change-Id: I8cd301bae58a6edbc3ef47607654bc21321721ca
2014-01-09 17:30:25 +01:00
James Zern
8c524db84c bump version to 0.4.0
libwebp{,decoder} - 0.4.0
libwebp libtool - 5.0.0
libwebpdecoder libtool - 1.0.0

mux/demux - 0.2.0
libtool - 1.0.0

Change-Id: Idbd067f95a6af2f0057d6a63ab43176fcdbb767d
2013-12-18 19:20:00 -08:00
Pascal Massimino
495bef413d fix bug in TrellisQuantize
the *quantized* level should be clipped to 2047, not the
original coeff.
(similar problem was fixed in the regular quantize function
quite some time ago)

Change-Id: I2fd2f8d94561ff0204e60535321ab41a565e8f85
2013-12-17 11:08:01 -08:00
James Zern
605a712701 simplify __cplusplus ifdef
drop c_plusplus which is from a quite ancient pre-standard compiler

Change-Id: I9e357b3292a6b52b14c2641ba11f4f872c04b7fb
2013-12-16 20:16:02 -08:00
James Zern
5227d99146 drop: ifdef __cplusplus checks from C files
the prototypes are already marked in the headers

Change-Id: I172fe742200c939ca32a70a2299809b8baf9b094
2013-12-13 11:42:13 -08:00
skal
73b731fb42 introduce a special quantization function for WHT
WHT is somewhat a special case: no sharpen[] bias, etc.
Will be useful in a later CL when precision of input is changed.

Change-Id: I851b06deb94abdfc1ef00acafb8aa731801b4299
2013-12-10 14:21:47 +01:00
skal
a3359f5d2c Only compute quantization params once
(all quantization params #1..#15 are the same)

Change-Id: If04058bd89fe2677b5b118ee4e1bcce88f0e4bf5
2013-12-10 05:36:23 +01:00
skal
d513bb62bc * fix off-by-one zthresh calculation
* remove the sharpening for non luma-AC coeffs
* adjust the bias a little bit to compensate for this

Using the multiply-by-reciprocal doesn't always give the same result
as the exact divide, given the QFIX fixed-point precision we use.
-> removed few now-unneeded SSE2 instructions (and checked for
bit-exactness using -noasm)

Change-Id: Ib68057cbdd69c4e589af56a01a8e7085db762c24
2013-12-09 13:56:04 +01:00
skal
6f104034a1 fix bug due to overzealous check in WebPPictureYUVAToARGB()
This tests prevented views to be converted to ARGB

https://code.google.com/p/webp/issues/detail?id=178

Change-Id: I5ba66da2791e6f1d2bfd8c55b5fffe6955263374
2013-12-02 15:23:12 +01:00
James Zern
cc55790e37 Merge changes I8bb7a4dc,I2c180051,I021a014f,I8a224a62
* changes:
  mux: add some missing casts
  enc/vp8l: add a missing cast
  idec: add some missing casts
  ErrorStatusLossless: correct return type
2013-11-27 17:06:35 -08:00
James Zern
4931c3294b cosmetics: fix some typos
Change-Id: I0d6efebd817815139db5ae87236fd8911df4d53c
2013-11-26 19:21:14 -08:00
James Zern
617d93480e enc/vp8l: add a missing cast
Change-Id: I2c1800516eb4573ae2599866ace10017b865f23a
2013-11-25 20:48:13 -08:00
skal
cb261f790f fix a descaling bug for vertical/horizontal U/V interpolation
RGBToU/V calls expects two extra precision bits, they were only
given one by SUM2H and SUM2H macros.

For rounding coherency, also changed SUM1 macro.

Change-Id: I05f96a46f5d4f17b830d0420eaf79b066cdf78d4
2013-11-19 11:24:09 +01:00
skal
ff379db317 few % speedup of lossless encoding
mostly visible for method 4 and up

Change-Id: I1561d871bc055ec5f7998eb193d927927d3f2add
2013-11-12 00:09:45 +01:00
James Zern
7708e60908 Merge "detect flatness in blocks and favor DC prediction" 2013-11-01 11:24:09 -07:00
James Zern
06b1503eff Merge "add comment about the kLevelsFromDelta[][] LUT generation" 2013-10-31 21:37:55 -07:00
skal
5935259cd0 add comment about the kLevelsFromDelta[][] LUT generation
Change-Id: Id1a91932bf05aeee421761667af351ef2200c334
2013-10-31 21:35:55 -07:00
skal
e3312ea681 detect flatness in blocks and favor DC prediction
this avoids local-minima that look bad, even if the distortion
looks low (e.g. gradients, sky,...). Mostly visible in the q=50-80 range.
Output size is mostly unchanged.

Change-Id: I425b600ec45420db409911367cda375870bc2c63
2013-11-01 00:47:04 +01:00
skal
a014e9c9cd tune quantization biases toward higher precision
* raise U/V quantization bias to more neutral values
* also raise the non-zero AC bias for Y1/Y2 matrices

(we need all the precision we can for U/V leves, which are often empty)
This will increase quality in the higher range (q >= 90) mostly.
Files size is exacted to raise a little (5-7%). and SSIM accordingly of course.

Change-Id: I8a9ffdb6d8fb6dadb959e3fd392e66dc5aaed64e
2013-10-30 23:57:23 +01:00
skal
1e898619cb add helpful PrintBlockInfo() function
(protected under a DEBUG_BLOCK compile flag)

Change-Id: Icb8da02dbd00e5cf856c314943c212f1c9578d9b
2013-10-30 19:25:27 +01:00
Pascal Massimino
98aa33cf1e extract random utils to their own file util/random.[ch]
they'll be used for decoding too, probably.

Change-Id: Id9cbb250c74fc0e876d4ea46b1b3dbf8356d6725
2013-10-30 02:00:33 -07:00
James Zern
d3408720d8 Merge "fast auto-determined filtering strength" 2013-10-29 12:47:58 -07:00
skal
f8bfd5cd1e fast auto-determined filtering strength
kLevelsFromDelta[sharpness][delta] is an inverse look-up table
that tells the minimum filtering strength needed to trigger the
filtering of a step with amplitude 'delta'. We use this table
in various situations:

a) when computing the initial (/global) filtering
strength for each segment. We look at the quantization
step and deduce the proper filtering strength needed
to result this quantization noise (talking the -f option
into account).

b) during intra16 calculation, when a block ends up
very empty (only DC coeffs are non-zero, all ACs have
vanished). We'll rely on the in-loop filtering to
restore the smoothness (if the source was gradient-like
smooth. That's why we look at the distortion too before
triggering the filtering).

Step b) goes _in addition_ to a), potentially raising
the filtering strength if blockiness is likely.

Change-Id: Icaeca93ef21da195b079e6587a44d9edfc8e9efa
2013-10-29 20:13:29 +01:00
Pascal Massimino
ac0bf951ca small clean-up in ExpandMatrix()
Change-Id: Ib06cb1658a6548f06bb7320310b3864881b606a7
2013-10-29 19:58:57 +01:00
skal
7c098d1814 Use some gamma-curve range compression when computing U/V average
This helps for discolorated chroma-subsampled edges.

Change-Id: I1d8ce87b66cb7e8b3572e6722905beabf0f50554
2013-10-18 21:28:05 +02:00
skal
0b2b05049f Use deterministic random-dithering during RGB->YUV conversion
-> helps debanding (sky, gradients, etc.)

This dithering can only be triggered when using -preset photo
or -pre 2 (as a preprocessing). Everything is unchanged otherwise.

Note that this change is likely to make the perceived PSNR/SSIM drop
since we're altering the input internally.

Change-Id: Id8d4326245d9b828141de162c94ba381b1fa5813
2013-10-17 22:36:49 +02:00
Pascal Massimino
f7146bc1e6 small optimization in segment-smoothing loop
probably not much of a speed difference

Change-Id: I08c41d82c3c2eb5ff9ec9ca9d81af2bb09b362de
2013-10-07 07:44:51 -07:00
skal
2b0a759335 Merge "fix some warnings from static analysis" 2013-09-13 13:00:12 -07:00
skal
d51f45f047 fix some warnings from static analysis
http://code.google.com/p/webp/issues/detail?id=138

Change-Id: I21470e965357cc14eab356e2c477c7846ff76ef2
2013-09-13 11:33:30 +02:00
Pascal Massimino
d134307b7f fix conversion warning on MSVC
"src\enc\frame.c(88) : warning C4244: '=' : conversion from 'const double' to 'float', possible loss of data"

Change-Id: I143cb0bb6b69e1b8befe9b4f24b71adbc28095c2
2013-09-12 18:57:51 -07:00
skal
80b54e1c69 allow search with token buffer loop and fix PARTITION0 problem
The convergence algo is noticeably faster and more accurate.

Try it with: 'cwebp -size xxxxx -pass 8 ...' or 'cwebp -psnr 39 -pass 8 ...'
for instance

Allow full-looping with TokenBuffer case, and make the non-TokenBuffer
case match too.

In case Partition0 is likely to overflow, retry encoding with harder
limits on max_i4_header_bits_.

This CL should make -partition_limit option somewhat useless,
since the fix made automatically (albeit in a non-optimal way yet).

Change-Id: I46fde3564188b13b89d4cb69f847a5f24b8c735b
2013-09-11 21:15:28 +02:00
skal
b7d4e04255 add VP8EstimateTokenSize()
estimates final size of coded tokens given a set of probabilities

Change-Id: Ia5a459422557d98b78b3cd8e1a88cb30835825b6
2013-09-11 10:08:49 +02:00
James Zern
10fddf53bb enc/quant.c: silence a warning
score_t -> int: rd_i4.H contains a value from a uint16_t
lookup

Change-Id: I7227de2dfab74b4f796abbc47955197ffa0e6110
2013-09-11 00:04:11 -07:00
Pascal Massimino
9f24519e82 encoder: misc rate-related fixes
* fix VP8FixedCostsI4ÆÅ table
   (the constant cost '211' was erronenously included)
* use the rd-score for '211' correctly (calling SetRDScore() for good)
* count partition0 bits separately during rd-opt

No meaningful difference in rd-curve.

Change-Id: I6c49a150cf28928d9a92c32fff097600d7145ca4
2013-09-10 00:25:32 -07:00
James Zern
c663bb214a Merge "simplify VP8IteratorSaveBoundary() arg passing" 2013-09-06 14:21:42 -07:00
Pascal Massimino
f691f0e461 simplify VP8IteratorSaveBoundary() arg passing
we only need to save yuv_out_, so no need for the arg

Change-Id: I7bad5d910e81ed2eda5c9787821fd1cfe905bd92
2013-09-06 02:11:16 -07:00
Pascal Massimino
42542be855 up to 6% faster encoding with clang compiler
mostly by revamping the main loop of GetResidualCost() and avoiding some branches

Change-Id: Ib05763e18a6bf46c82dc3d5d1d8eb65e99474207
2013-09-05 10:38:22 -07:00
skal
93402f02db multi-threaded segment analysis
When -mt is used, the analysis pass will be split in two
and each halves performed in parallel. This gives a 5%-9% speed-up.

This was a good occasion to revamp the iterator and analysis-loop
code. As a result, the default (non-mt) behaviour is a tad (~1%) faster.

Change-Id: Id0828c2ebe2e968db8ca227da80af591d6a4055f
2013-09-05 09:13:36 +02:00
Pascal Massimino
2fd091c9ae Merge "use NULL for lf_stats_ testing, not bool" 2013-09-04 07:18:42 -07:00
skal
cfb56b1707 make -pass option work with token buffers
-pass 2 can be useful sometimes. More passes usually don't help more.
This change is a step toward being able to re-code the whole picture
with varying parameter (when token buffer is used).

Change-Id: Ia2538e2069a53c080e2ad248c18a1e04623a9304
2013-09-03 23:37:42 +02:00
Pascal Massimino
35dba337a3 use NULL for lf_stats_ testing, not bool
Change-Id: I14c6341bce1c745b7f2d1b790f2d4f8441d01be6
2013-09-03 08:45:59 -07:00
skal
733a7faae4 enc->Iterator memory cleanup
* move yuv_in_/out_* scratch buffers to iterator
* add y_top_/uv_top_ shortcuts in iterator

That's ~3k of stack size instead of heap.
But it allows having several iterators work in parallel.

Change-Id: I6a437c0f2ef1e5d398c1d6a2fd4974fa0869f0c1
2013-08-31 23:38:11 +02:00
skal
ad6ac32d7c simplify upsampler calls: only allow 'bottom' to be NULL
If 'top' was meant to be NULL, then bottom and top can be
swapped. Logic is simpler.

+ fix compilation in non-FANCY_UPSAMPLING mode

Change-Id: I7c62bbb59454017f072c0945d1ff2d24d89286ff
2013-08-19 16:47:51 -07:00
Pascal Massimino
f72fab7045 cosmetics
Change-Id: Ib6fa2a7e1db8c5c48607c7097520ffca34a6cb66
2013-08-15 04:16:37 -07:00
Vikas Arora
e081f2f359 Pack code & extra_bits to Struct (VP8LPrefixCode).
Also created variant VP8LPrefixEncodeBits that returns the
code & extra_bits only.
There's no impact on compression density and compression speed.

Change-Id: I2cafdd3438ac9270cd72ad9d57b383cdddfdfa4c
2013-08-12 11:56:42 -07:00
Vikas Arora
69257f70df Create LUT for PrefixEncode.
This speeds up lossless compression by 5%.
Change-Id: Ifd114b1d9850dc3aac74593809e7d48529d35e3d
2013-08-05 10:20:18 -07:00
skal
de4d4ad598 VP8EncIterator clean-up
- remove unused fields from iterator
- introduce VP8IteratorSetRow() too
- rename 'done_' to 'countdown_'
- bring y_left_/u_left_/v_left_ from VP8Encoder

Change-Id: Idc1c15743157936e4cbb7002ebb5cc3c90e7f92a
2013-08-01 23:05:54 -07:00
Pascal Massimino
8dcae8b3cf fix rescaling-with-alpha inaccuracy
(still missing YUVA decoding case for now)

https://code.google.com/p/webp/issues/detail?id=160

Change-Id: If723b4a5c0a303d0853ec9d839f995adce056095
2013-07-26 12:10:26 -07:00
Vikas Arora
7d60bbc6d9 Speed up HashChainFindCopy function.
Speed up HashChainFindCopy by optimizing on number of calls to
FindMatchLength method.
This change speeds up the lossless & lossy (Alpha) encoding by 20%
without loss of compression density.
At method=3, lossy (Alpha) compression speed (and density) remains
unchanged, as at that settings, the costly Backward Refs method is not
called

Change-Id: Ia1797148e9e4ee2787011837fa248afbae2242cb
2013-07-16 19:58:18 -07:00