Update BackwardRefsWithLocalCache to do in-place update of backward
references w.r.t local color cache index.
No impact on the compression density or compression speed.
Change-Id: Ie066251464c3928c044e037b43df3af28b48ca30
histogram.c:
- Verified (earlier) that there's low correlation between Red & Blue colors
(particularly after applying Cross-color transform). The Bin based histogram
merge, bins on three entropies viz literal, red & blue symbols. Removing
either of blue or red increases the compression density. So keeping the bins
for red & blue sybmols.
- Keeping the compact bins method as-is. This way it's simpler to read.
huffman_encode.h: Added field comments for struct HuffmanTree and removed the TODO.
Change-Id: Ia76f7bc730079d1b3b644038c5d9931db3797f0e
Use the refs_lz77 computed (with cache_bits=0) in the method 'CalculateBestCacheSize'
to regenerate the LZ77 references corresponding to the optimum cache_bits and avoid
calling costly 'BackwardReferencesLz77' one extra time.
This change leaves the compression density unchanged and speeds up compression
by 10-15%.
Change-Id: I5a92e11788d3c3f656aa7e1fba54fb5d96ee0027
This wasn't working for this specific scenario:
- Encode an RGBA 'pic' (with trivial alpha) using lossy encoding.
(so that pic->a == NULL after import happens).
- Modify the 'pic->argb' so that it has non-trivial alpha.
- Encode the same 'pic' again.
This used to fail to encode alpha data as pic->a == NULL.
Change-Id: Ieaaa7bd09825c42f54fbd99e6781d98f0b19cc0c
- The optimal cache bits is evaluated inside the method 'VP8LGetBackwardReferences'.
- The input cache_bits to 'VP8LGetBackwardReferences' sets the maximum cache
bits to use (passing 0 implies disabling the local color cache).
- The local color cache is disabled for lowerf (<= 25) quality levels (as before).
- Enabled local color cache for palette images as well. This saves additional
0.017% bytes with a slight (2-3%) improvement in the compression speed.
- Removed 'use_2d_locality' parameter from method VP8LGetBackwardReferences, as
this option is not an option now (after we freeze the lossless bit-stream).
Change-Id: I33430401e465474fa1be899f330387cd2b466280
Updated BackwardReferencesRle method by utilizing the local color cache.
Also changed the name of method BackwardReferencesHashChain to
BackwardReferencesLz77 to reflect the LZ77 coding.
For the 1000 image corpus, this change saves 0.2% bytes
(at default settings) and is 2-5% faster to encode.
Change-Id: Ic3f288253b3bbb101a69945a80994c3fd0917f8b
Optimize backwardreferences (about 0.1% byte savings) with almost same
compression speed (3% faster on defaut compression settings).
1.) Simplified iteration logic for HashChainFindCopy.
- Remapped the iter_max constant.
2.) Simplified main for loop for BackwardReferencesHashChain
- Removed 'if' conditions for corner cases in the main loop.
- Refactored the method(AddSingleLiteral) for adding one pixel.
Change-Id: I1bc44832fd81f11e714868a13e606c8f83157e64
Speed up BackwardReferencesHashChainDistanceOnly method by:
1.) Remove for loop for shortmax code path.
2.) Execute the shortmax code path after regular call to
HashChainFindCopy, only if HashChainFindCopy() returns length > 2 (MIN_LENGTH).
3.) Also for shortmax, call method HashChainFindOffset (for length = 2),
instead of expensive method HashChainFindCopy().
4.) Handling first pixel (i==0) outside main loop and removing one if
condition (i > 0) per pixel.
5.) Handle the last pixel outside the main 'for' loop.
Overall compression speedup observed is around 5% (+/- noise).
Change-Id: Ifa30c4035f8d26e6e43e3c4881244d777961c22b
Enable bin-partition entropy based heuristic for merging histograms
for higher (q >= 90) qualities as well. Keep the old behavior at the
maximum quality level (q==100).
This speeds up the compression between Q=90-99 (method=4) by factor 5-7X
and with loss of 0.5-0.8% in the compression density.
Change-Id: I011182cb8ae5403c565a150362bc302630b3f330
The Maximum allowed limit is 11.
The Q=25 and below is not impacted as cache bits are forced to 0.
This saves 0.05% - 0.1% bytes for other quality with almost same compression
speed (+/- 2-3%, that's more of a noise).
Change-Id: Icf972a98f298c89e140e37a627baf709134be9a0
Updated the logic to limit the Histogram size to a constant, instead of
computing the same based on the Histogram size (that's variable size based on
the cache bits) for the maximum possible cache bits. The actual cache bits may
be lower than the maximum.
Note: The constant 2600 is 16MB/Sizeof(HistogramSize(MAX_COLOR_CACHE_BITS)).
The compression density remains the same with this change, with little faster
compression speed.
Change-Id: I3149894962852e9dad2501b9aa16bb847a20fd86
The method VP8LCalculateEstimateForCacheSize is not evaluating the all possible
range for cache_bits.
Also added a small penality for choosing the larger cache-size. This is done to
strike a balance between additional memory/CPU cost (with larger cache-size) and
byte savings from smaller WebP lossless files.
This change saves about 0.07% bytes and speeds up compression by 8% (default
settings). There's small speedup at Q=50 along with byte savings as well.
Compression at Quality=25 is not effected by this change.
Change-Id: Id8f87dee6b5bccb2baa6dbdee479ee9cda8f4f77
Instead of calling HashChainFindMethod, call a new (subset) method
HashChainFindOffset to get the offset/distance for a given length.
The encoding is tad faster at default compression
Before After
bpp/rate bpp/rate
442 Palette 0.2720/5.270 MP/s 0.2720/5.790 MP/s
558 non-palette 3.7607/0.797 MP/s 3.7607/0.816 MP/s
Change-Id: If4041a9c18f7e972f49fcbab8c3e2f013d8bf1cf
Updated VP8LGetBackwardReferences and HashChainFindCopy method with following:
- Remove the recursive CostModelBuild.
- Reuse the lz77 backward refs in CostModelBuild, instead of evaluating it
again (as it was done for recursion_level=0).
- Consolidated the Match-length logic inside FindMatchLength method.
- Removed the logic for altering best_length/val based on the 2D distance.
The additional 162 value (+= 9 * 9 + 9 * 9 - y * y - x * x) can't change the
best_val eval computation to choose a different curr_length, as best_val was
set to 'curr_length << 16'.
Following is the impact on the compression speed/density at default & max
quality, overall this speeds up compression by 5-15% (q=100 -> 75) with a tad
drop (0.02-0.03%) in compression density for the non-palette images.
Before After
bpp/Rate(MP/s) bpp/Rate(MP/s)
q=75 (def)
All 1000 2.4492/1.049 MP/s 2.4498/1.230 MP/s
Palette 0.2719/5.060 MP/s 0.2719/6.110 MP/s
non-Palette 3.7597/0.732 MP/s 3.7607/0.840 MP/s
q=100
All 1000 2.4134/0.125 MP/s 2.4142/0.131 MP/s
Palette 0.2692/2.585 MP/s 0.2692/2.885 MP/s
non-Palette 3.7040/0.079 MP/s 3.7053/0.083 MP/s
Change-Id: I27a5eff3356d876c3e949fd32262244b25678b7a
Compared to previous mode it gives another 10-30% improvement in compression keeping comparable PSNR on corresponding quality settings.
Still protected by the WEBP_EXPERIMENTAL_FEATURES flag.
Change-Id: I4821815b9a508f4f38c98821acaddb74c73c60ac
Evaluate if for Palette images (num_colors <= 256), non-palette
compression path (Subtract green, predictor transform etc) yield an
optimal compression density.
This change reduces the WebP file (for palette images) size by 0.4% with
drop of 3-5% in compression speed.
Change-Id: I1ad66fa94db4fd7ba7bc215763791ef662cd4f42
the number of segments are previously validated, but an explicit check
is needed to avoid a warning under gcc-4.9
Change-Id: Ifa7c0dd7f3f075b3860fa8ec176d2c98ff54fcea
We compact the palette by weighted distance, favoring the green channel.
Average gain on paletted file is ~0.5%, with gain up to 6-7% on some favorable cases.
Encoding speed is unaffected.
Disabled for alpha (or any single-channel input)
Also: always use quality=20 for EncodePalette() since it
doesn't make any real difference.
Change-Id: I19fb14316a366f139a941b45aef5663a33c905e1
SSE2 version is 2.1x faster
This is used to transfer the alpha plane to green channel before lossless compression.
Change-Id: I01d9df0051c183b1ff5d6eb69961d4f43e33141a
Don't combine the Histograms that have trivial (single valued A, R & B)
symbols.
Following is the compression savings data along with compression time (before
& after) per image.
Before After
bpp, rate(MP/s) bpp, rate(MP/s)
Q=25, method = 4 2.508, 1.807 2.499, 1.916
Q=50, method = 4 2.460, 1.488 2.456, 1.512
Q=75, method = 4 2.452, 1.078 2.450, 1.092
Q=25, method = 5 2.505, 1.398 2.496, 1.383
Q=50, method = 5 2.458, 1.170 2.453, 1.143
Q=75, method = 5 2.453, 0.886 2.450, 0.855
This change provides 0.1-0.4% compression gains and speeds up the lossless
compression for the default method=4 (the drop in compression speed is between 1-3.5% for method=5).
Change-Id: Idfd88c2092f37afacd26a97097b3053f8183953a
Tested on 1000 pngs corpus with quality 90-100 it gives ~0.15% improvement
in compression density and ~7% speed up.
Change-Id: I460f56c96707edb3c1f0b51a024e5122e10458df
if ALPHA_LOSSLESS_COMPRESSION produces a too big file (very rare!),
we fall-back to no-compression automatically.
Change-Id: I5f3f509c635ce43a5e7c23f5d0f0c8329a5f24b7
rightmost pixel was missing a copy, which could lead to invalid read.
Also added a lower dimension of 4, below which we use the regular conversion.
This is to prevent corner cases, in addition to not being overkill.
Change-Id: Iac12e7a3d74590f12fe8eeb1830b9891e61439f6
with a special case for dithering==0., it gets somewhat faster on x86
thanks to inlining.
Also, less macros.
Change-Id: Ic2f2bf6718310743bb40cef2104fa759a073e6d5
New function: WebPPictureSmartARGBToYUVA()
This implement smart RGB->YUV conversion.
This is rather undocumented for now, and is triggered using '-pre 4'
preprocessing option.
This is slow-ish and use quite some memory, but should be improvable.
This is somehow a usable beta version.
Change-Id: Ia50a8c30134e4cab8a7d3eb70aef13ce1f6187a1
This compresses the uimage using lossless compression and controlable
decimating pre-process.
Code is under WEBP_EXPERIMENTAL_FEATURE while it's being experimented with.
Change-Id: I8b7f4cfcc3c6afc52a556102842bdbb045ed5ee8
(typecast uint32 pointer to uint64).
The proposed change is little (0.05%) slower but avoids uint32 to uint64
pointer conversion.
Change-Id: I6b8828077ea1324fabd04bfa7e7439e324776250
Sometimes, the error-code was not set correctly.
We now return OUT_OF_MEMORY everytimes it's appropriate
(tested using MALLOC_FAIL_AT mechanism)
Took the opportunity to clean-up the code and dust the error
code returned (some were erroneously set to INVALID_CONFIGURATION)
Change-Id: I56f7331e2447557b3dd038e245daace4fc82214c
only 1 of <lib>_CPPFLAGS and AM_CPPFLAGS is used, with the former
getting precedence when it's defined. configure's DEFAULT_INCLUDES is
covering what's necessary given the include paths are all source
relative.
Change-Id: I7d14076acd266b28a88a3d92bcc3d7165284d5f3
this change has the side-effect of using directory names in the
include, silencing a lint warning.
Change-Id: Ib91cf63a90534e32fadfa5c2372bfdb29f854d02
if res->first = 1, coeffs[0]=0 because of quant.c:749 and line
added at quant.c:744
So, no need for the extra case.
Going forward, TrellisQuantizeBlock() should also be calling
a variant of VP8SetResidualCoeffs() to set the 'last' field.
also: fixes a warning for win64
+ slight speed-up
Change-Id: Ib24b611f7396d24aeb5b56dc74d5c39160f048f0
The luminance needs to be pre- and post- multiplied by
the alpha value in case of rescaling, for proper averaging.
Also:
- removed util/alpha_processing and moved it to dsp/
- removed WebPInitPremultiply() which was mostly useless
and merged it with the new function WebPInitAlphaProcessing()
Change-Id: If089cefd4ec53f6880a791c476fb1c7f7c5a8e60
also changed the token-page layout a little bit to remove
a not-needed field.
This reduces the number of malloc()/free() calls substantially
with minimal increase in memory consumption (~2%).
For the tail of large sources, the number of malloc calls goes
typically from ~10000 to ~100 (e.g.: bryce_big.jpg: 22711 -> 105)
Change-Id: Ib847f41e618ed8c303d26b76da982fbc48de45b9
Non-photo source produce far less literal reference and their
buffer is usually much smaller than the picture size if its compresses
well. Hence, use a block-base allocation (and recycling) to avoid
pre-allocating a buffer with maximal size.
This can reduce memory consumption up to 50% for non-photographic
content. Encode speed is also a little better (1-2%)
Change-Id: Icbc229e1e5a08976348e600c8906beaa26954a11
the unique instance of VP8LHashChain (1MB size corresponding to hash_to_first_index_)
is now wholy part of VP8LEncoder, instead of maintaining the pointer to VP8LHashChain
in the encoder.
Change-Id: Ib6fe52019fdd211fbbc78dc0ba731a4af0728677
We use automatic int->uint64_t promotion where applicable.
(uint64_t should be kept only for overflow checking and memory alloc).
Change-Id: I1f41b0f73e2e6380e7d65cc15c1f730696862125
* merged the two HistogramAdd/AddEval() into a single call
(with detection of special case when b==out)
* added a SSE2 variant
* harmonize the histogram type to 'uint32_t' instead
of just 'int'. This has a lot of ripples on signatures.
* 1-2% faster
Change-Id: I10299ff300f36cdbca5a560df1ae4d4df149d306
Reduce calls to Malloc (WebPSafeMalloc/WebPSafeCalloc) for:
- Building HashChain data-structure used in creating the backward references.
- Creating Backward references for LZ77 or RLE coding.
- Creating Huffman tree for encoding the image.
For the above mentioned code-paths, allocate memory once and re-use it
subsequently.
Reduce the foorprint of VP8LHistogram struct by changing the Struct
field 'literal_' from an array of constant size to dynamically allocated
buffer based on the input parameter cache_bits.
Initialize BitWriter buffer corresponding to 16bpp (2*W*H).
There are some hard-files that are compressed at 12 bpp or more. The
realloc is costly and can be avoided for most of the WebP lossless
images by allocating some extra memory at the encoder initializaiton.
Change-Id: I1ea8cf60df727b8eb41547901f376c9a585e6095
This is to help further optimizations.
(like in https://gerrit.chromium.org/gerrit/#/c/69787/)
There's a small slowdown (~0.5% at -z 9 quality) due to
function pointer usage. Note that, for speed, it's important
to return VP8LStreaks by value, and not pass a pointer.
Change-Id: Id4167366765fb7fc5dff89c1fd75dee456737000
+ reorganize the cost-evaluation code by moving some functions
to cost.h/cost.c and exposing VP8Residual
Change-Id: Id976299b5d4484e65da8bed31b3d2eb9cb4c1f7d
This change gains back 1% in compression density for method=3 and 0.5% for
method=4, at the expense of 10% slower compression speed.
Change-Id: I491aa1c726def934161d4a4377e009737fbeff82
Tune HistogramCombineBin for hard images that are larger than 1-2 Mega
pixel and represent photographic images.
This speeds up lossless encoding on 1000 image corpus by 10-12% and compression
penalty of 0.1-0.2%.
Change-Id: Ifd03b75c503b9e886098e5fe6f86be0391ca8e81
there's still some malloc/free in the external example
This is an encoder API change because of the introduction
of WebPMemoryWriterClear() for symmetry reasons.
The MemoryWriter object should probably go in examples/ instead
of being in the main lib, though.
mux_types.h stil contain some inlined free()/malloc() that are
harder to remove (we need to put them in the libwebputils lib
and make sure link is ok). Left as a TODO for now.
Also: WebPDecodeRGB*() function are still returning a pointer
that needs to be free()'d. We should call WebPSafeFree() on
these, but it means exposing the whole mechanism. TODO(later).
Change-Id: Iad2c9060f7fa6040e3ba489c8b07f4caadfab77b
(and ~2-3% on ARM)
We don't need to store cost/score for each node, but only for
the current and previous one -> simplify code and save some memory.
Also made the 'Node' structure tighter.
Change-Id: Ie3ad7d3b678992b396242f56e2ac387fe43852e6
all the functions involved return double and later these locals are used
in double calculations. fixes a vs build warning
Change-Id: Idb547104ef00b48c71c124a774ef6f2ec5f30f14
Optimize and re-structured VP8LGetHistoImageSymbols method, by using the bin-hash
for merging the Histograms more efficiently, instead of the randomized
heuristic of existing method HistogramCombine.
This change speeds up the Lossless encoding by 40-50% (for method=4 and Q > 50)
with 0.8% penalty in compression density. For lower method, the speed up is 25-30%,
with 0.4% penalty in the compression density.
Change-Id: If61adadb1a041b95def6405aa1fe3b83c3cb25ce
These are presets for lossless coding, similar to zlib.
The shortcut for lossless coding is now, e.g.:
cwebp -z 5 in.png -o out_lossless.webp
There are 10 possible values for -z parameter:
0 (fastest, lowest compression)
to 9 (slowest, best compression)
A reasonable tradeoff is -z 6, e.g.
-z 9 can be quite slow, so use with care.
This -z option is just a shortcut for some pre-defined
'-lossless -m xx -q yy' combinations.
Change-Id: I6ae716456456aea065469c916c2d5ca4d6c6cf04
(We didn't need the exact value of the max_error properly.
We can work with relative values instead of absolute)
Output is bitwise the same as before.
Change-Id: I67aeaaea5f81bfd9ca8e1158387a5083a2b6c649
Refactor code for HistogramCombine and optimize the code by calculating
the combined entropy and avoid un-necessary Histogram merges.
This speeds up lossless encoding by 1-2% and almost no impact on compression
density.
Change-Id: Iedfcf4c1f3e88077bc77fc7b8c780c4cd5d6362b
mostly by:
- storing a single rd-score instead of cost / distortion separately
- evaluating terminal cost only once
- getting some invariants out of the loops
- more consts behind fewer variables
Change-Id: I79451f3fd1143d6537200fb8b90d0ba252809f8c
incorporate non-last cost in per-level cost table
also: correct trellis-quant cost evaluation at nodes
(output a little bit different now). Method 6 is ~4% faster.
Change-Id: Ic48bd6d33f9193838216e7dc3a9f9c5508a1fbe8
Speedup lossless encoder by 20-25% by optimizing:
- GetBestColorTransformForTile: Use techniques like binary search and
local minima search to reduce the search space.
- VP8LFastSLog2Slow & VP8LFastLog2Slow: Adding the correction factor for
log(1 + x) and increase the threshold for calling the approximate
version of log_2 (compared to costly call to log()).
Change-Id: Ia2444c914521ac298492aafa458e617028fc2f9d
Increase the initial buffer size for VP8L Bit Writer from 4bpp to 8bpp.
The resize buffer is expensive (requires realloc and copy) and this additional
memory (0.5 * W * H) doesn't add much overhead on the lossless encoder.
Change-Id: Ic1fe55cd7bc3d1afadc799e4c2c8786ec848ee66
Optimize 'VP8LCalculateEstimateForCacheSize' for lower quality ranges (Q < 50).
The entropy is generally lower for higher cache_bits, so start searching from
higher cache_bits and settle for a local minima, instead of evaluating all
values.
This speeds up the lossless encoding at lower qualities by 10-15%.
Change-Id: I33c1e958515a2549f2e6f64b1aab3f128660dcec
This makes the segmentation overall less prone to
local-optimum or boundary effect.
(and overall, encoding is a little faster)
Change-Id: I35688098b0f43c28b5cb81c4a92e1575bb0eddb9
the -alpha_cleanup flag was ineffective since we switched cwebp
to using ARGB input always.
Original idea by David Eckel (dvdckl at gmail dot com)
Change-Id: I0917a8b91ce15a43199728ff4ee2a163be443bab
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
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
* 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
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
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
* 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
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
-> 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
"src\enc\frame.c(88) : warning C4244: '=' : conversion from 'const double' to 'float', possible loss of data"
Change-Id: I143cb0bb6b69e1b8befe9b4f24b71adbc28095c2
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
* 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
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
-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
* 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
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
Also created variant VP8LPrefixEncodeBits that returns the
code & extra_bits only.
There's no impact on compression density and compression speed.
Change-Id: I2cafdd3438ac9270cd72ad9d57b383cdddfdfa4c
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
Disable costly 'BackwardReferencesTraceBackwards' for encoding Alpha plane.
Increase the threshold for triggering 'BackwardReferencesTraceBackwards' to
quality 25 and above. Also lower the Alpha quality (at method 3) to be
lesser than this threshold (25).
Change-Id: Ic29fb2e6943472c564223df9fe099b19ccda0f31
This speeds up WebP lossless decoding by 20%. In particular, the
photographic images get 35% speedup.
Change-Id: Idb94750342a140ec05df52c07e12be4bba335adc
+ some revamp and cleanup of the alpha-filter trial loop
+ EncodeAlphaInternal() now just takes a FilterTrial param
Change-Id: Ief84385083b1cba02678bbcd3dbf707245ee962f
* 0.3.0: (57 commits)
update ChangeLog
Regression fix for alpha channels using color cache:
wicdec: silence a format warning
muxedit: silence some uninitialized warnings
update ChangeLog
update NEWS
bump version to 0.3.1
Revert "add WebPBlendAlpha() function to blend colors against background"
Simplify forward-WHT + SSE2 version
probe input file and quick-check for WebP format.
configure: improve gl/glut library test
update copyright text
configure: remove use of AS_VAR_APPEND
fix EXIF parsing in PNG
add doc precision for WebPPictureCopy() and WebPPictureView()
remove datatype qualifier for vmnv
fix a memory leak in gif2webp
fix two minor memory leaks in webpmux
remove some cruft from swig/libwebp.jar
README: update swig notes
...
Conflicts:
NEWS
examples/gif2webp.c
src/dec/alpha.c
src/dec/idec.c
src/dec/vp8l.c
src/enc/alpha.c
src/enc/vp8l.c
Change-Id: Ib202fad7825a090c3b3a5169acd171369cface47
rather than symlink the webm/vpx terms, use the same header as libvpx to
reference in-tree files
based on the discussion in:
https://codereview.chromium.org/12771026/
Change-Id: Ia3067ecddefaa7ee01550136e00f7b3f086d4af4
(cherry picked from commit d640614d54)
The auto-infer logic of detecting the 'Alpha' use case
(via check '(palette[i] & 0x00ff00ffu) != 0' is failing
for this corner case image with all black pixels (rgb = 0)
and different Alpha values.
-> switch generic use-LUT detection
Change-Id: I982a8b28c8bcc43e3dc68ac358f978a4bcc14c36
(cherry picked from commit afa3450c11)
Added 1 pixel cache for palette colors for faster lookup.
This will speedup images that require ApplyPalette by 6.5% for lossless
compression.
Change-Id: Id0c5174d797ffabdb09905c2ba76e60601b686f8
(cherry picked from commit 742110ccce)
Start VP8EncLoop/VP8EncTokenLoop only if VP8EncStartAlpha succeeded.
Change-Id: Id1faca3e6def88102329ae2b4974bd4d6d4c4a7a
(cherry picked from commit 67708d6701)
new option: -blend_alpha 0xrrggbb
also: don't force picture.use_argb value for lossless. Instead,
delay the YUVA<->ARGB conversion till WebPEncode() is called.
This make the blending more accurate when source is ARGB
and lossy compression is used (YUVA).
This has an effect on cropping/rescaling. E.g. for PNG, these
are now done in ARGB colorspace instead of YUV when lossy compression
is used.
Change-Id: I18571f1b1179881737a8dbd23ad0aa8cddae3c6b
(cherry picked from commit e7d9548c9b)
Tuned the cross_color transform parameter (step) for lower quality
levels. This change gives speedup of 20% at lower qualities (25) and 10% at
moderate quality level (50) with a loss of 0.25% in compression density.
Also removed TODO for cross_color transform. Observed good correlation of
this with the predict transform.
Change-Id: I8a1044e9f24e6a5f84295c030fd444d0eec7d154
rather than symlink the webm/vpx terms, use the same header as libvpx to
reference in-tree files
based on the discussion in:
https://codereview.chromium.org/12771026/
Change-Id: Ia3067ecddefaa7ee01550136e00f7b3f086d4af4
The auto-infer logic of detecting the 'Alpha' use case
(via check '(palette[i] & 0x00ff00ffu) != 0' is failing
for this corner case image with all black pixels (rgb = 0)
and different Alpha values.
-> switch generic use-LUT detection
Change-Id: I982a8b28c8bcc43e3dc68ac358f978a4bcc14c36
Added 1 pixel cache for palette colors for faster lookup.
This will speedup images that require ApplyPalette by 6.5% for lossless
compression.
Change-Id: Id0c5174d797ffabdb09905c2ba76e60601b686f8
'mem' was being offset once by DO_ALIGN() then shifted 'nz_size' which
would end up accounting for more than ALIGN_CST and exceed the allocation.
broken since:
9bf3129 align VP8Encoder::nz_ allocation
Change-Id: I04a4e0bbf80d909253ce057f8550ed98e0cf1054
new option: -blend_alpha 0xrrggbb
also: don't force picture.use_argb value for lossless. Instead,
delay the YUVA<->ARGB conversion till WebPEncode() is called.
This make the blending more accurate when source is ARGB
and lossy compression is used (YUVA).
This has an effect on cropping/rescaling. E.g. for PNG, these
are now done in ARGB colorspace instead of YUV when lossy compression
is used.
Change-Id: I18571f1b1179881737a8dbd23ad0aa8cddae3c6b
user can now call WebPEncode() with any YUVA or ARGB format, for
lossy or lossless compression
also: simplified error reporting, which is done in WebPPictureARGBToYUVA()
and WebPPictureYUVAToARGB()
Change-Id: Ifb68909217175bcf5a050e5c68d06de9849468f7
(cherry picked from commit 07d87bda1b)
Saturation was done on input coeff, not quantized one.
This saturation is not absolutely needed: output of FTransformWHT
is in range [-16320, 16321]. At quality 100, max quantization steps is 8,
so the maximal range used by QuantizeBlock() is [-2040, 2040].
But there's some extra bias (mtx->bias_[] and mtx->sharpen_[]) so
it's better to leave this saturation check for now.
addresses issue #145
Change-Id: I4b14f71cdc80c46f9eaadb2a4e8e03d396879d28
* merge cost calculation functions (BitsEntropy() and HuffmanCost())
* have HistogramAdd() specialized into separate functions
* use threshold to bail-out early
* revamp code a bit
* also: save memory by freeing free(histogram_image)
Change-Id: I8ee5d2cfa1462d5d6ea6361f5c89925a3720ef55
This is required for WebP lossy+Alpha images, where Alpha channel is taking
60-70% of the compression (CPU) cycles.
Also evaluated on 1000 PNG corpus and overall compression speed
is 15-40% better for lossy (PNG+Alpha) compression.
The pure lossless compression numbers are almost same (or little
better) with this change.
Change-Id: I9e5ae7372ed6227a9a5b64cd9cff84c747195a57
using token-buffer (that is: slightly more memory. O(output_size))
This change is ON by default. To return to previous behaviour, use
'cwebp -low_memory' or set config.low_memory to true.
Side-effect of this new mode: it forces 1 partition only (which was
default anyway), and makes some statistics about the bitstream
no longer available. cwebp will no longer report 'intra4-coeffs', etc.
This mode also doesn't work (yet) with multi-pass, and -low_memory
is currently forced for multi-pass.
also: reversed the flag: USE_TOKEN_BUFFER -> DISABLE_TOKEN_BUFFER
also: fixed the kAverageBytesPerMB estimate
Change-Id: I4ea80382038d6df4309663e0cb7bd88d9bca9cf1
broken since:
ad25032 Merge "multi-threaded alpha encoding for lossy"
this produced an error due to an empty VP8TBuffer struct.
Change-Id: I640809d07d20092c1d660e2b59b58a62a12e4371
new option: 'cwebp -mt ...'
new config flag: config.thread_level
(allowed thread_level are 0 or 1 for now. Maybe more later...)
If -mt is activated (and WEBP_USE_THREAD is used for compile), the alpha-compression
will be done in parallel to RGB coding for lossy. Can save quite a bit of latency...
Has no effect for lossless encoding.
Change-Id: I769d0bf90e7380cf99344ad62cd77277f4df5a46
signed integer overflow behavior is undefined, split PrefixEncode() to
two branches to avoid this.
Change-Id: I6e2761d0d77f0aaceafdc4e07232e089c22beb64
This option remaps internal parameters to better match
the expected compression curve of JPEG and produce output files
of similar size, but with better quality.
Change-Id: I96a1cbb480b1f6a0c6845a23c33dfd63f197b689
also change lossless encoder logic, which was relying on explicit
NULL return from WebPSafeMalloc(0)
renamed function to CheckSizeArgumentsOverflow() explicitly
addresses issue #138
Change-Id: Ibbd51cc0281e60e86dfd4c5496274399e4c0f7f3
* treat the last coeff as a special case
* re-arrange the inner code to be shorter
* replace some VP8EncBands[n] by n, for n = 0 or 1
Change-Id: I71e17b014cffad7b073e787fde06260905a6953f
- Separate out mux.h and demux.h
- muxtypes.h: new header for data types common to mux/demux
- Move some misc read/write utilities to utils/utils.h
- Remove some duplicate methods.
- Separate out mux/demux libraries
Change-Id: If9b9569b10d55d922ad9317ef51710544315d6de
10-15% faster encoding.
Almost same output, binary wise. The main difference is
that we can't compute uv_alpha susceptibility, means there
can be subtle differences with different -sns values.
Change-Id: Id1b1a50929bf125b6372212fee1ed75a3bed975f
Number of pairs selected are limited between 25% of histogram
images (at start) and number of histogram images left at any iteration.
Increase the range of iter_mult.
Removed min_cluster_size as parameter for tuning HistogramCombine.
Change-Id: Ia4068cd7af4d0f63c5af9001aceda8a40b9de740
* commit 'v0.2.1':
Update ChangeLog
update NEWS
bump version to 0.2.1
libwebp: validate chunk size in ParseOptionalChunks
cwebp (windows): fix alpha image import on XP
autoconf/libwebp: enable dll builds for mingw
[cd]webp: always output windows errors
fix double to float conversion warning
cwebp: fix jpg encodes on XP
VP8LAllocateHistogramSet: fix overflow in size calculation
GetHistoBits: fix integer overflow
EncodeImageInternal: fix uninitialized free
fix the -g/O3 discrepancy for 32bit compile
fix the BITS=8 case
Make *InitSSE2() functions be empty on non-SSE2 platform
make *InitSSE2() functions be empty on non-SSE2 platform
make VP8DspInitNEON() public
Conflicts:
src/Makefile.am
src/dsp/dec_neon.c
Change-Id: Iddc5152e4a6892db96c12d7c3f74adbc85fe6178
- Changed the dynamic range where more aggressive
(BackwardReferencesTraceBackward) heuristic is run from quality > 10
(instead of quality > 25).
- Limit the backward-ref Window size to 16*width & 256*width for lower
qualities ([0, 25[ & [25, 50[) respectively, instead of 1M window.
- Evaluate the params for HashChainFindCopy outside this function call
and pass it, instead of recomputing them for every call.
Change-Id: If9eedfc14b978e7632d7cf69c96186e2910b0554
the multiplications done for total_size would be done with integers,
possibly overflowing, before being promoted to 64-bit for the addition
Change-Id: I32c3a6400fc2ef120c38e01a8693f4cb1727234d
huff_image_size was a size_t (=32 bits with 32-bit builds) which could
rollover causing an incorrectly sized allocation and a crash in lossless
encoding.
fixes issue #128
Change-Id: I0f20cee98c29b2b40b02607930b6b7a7ca56996d
in debug mode, some float operations see their intermediate
values stored in memory rather than staying in the FPU (which
is 80bit precision).
Several fixes are possible (breaking long calculations into
atomic steps for instance), but simpler of all is just about
turning the cost[] array into float* instead of double*.
The code is a tad faster, and i didn't see any major output
size difference.
Change-Id: I053e6d340850f02761687e072b0782c6734d4bf8
LSIM stands for "local similarity": before matching
a compressed pixel to the source, we search around in the source
and minimise the squared error. So, this is close to PSNR calculation,
but mitigates some of its limitations (pure translation and noise for instance).
There's a new -print_lsim option to cwebp too.
Change-Id: Ia38561034c7a90e71d2ea0f55bb1de527eda245b
Make the heuristic for combining Histograms a function of compression
quality. This change will speed-up compression time for compression
quality less than 75. The compression time/density remains unchanged
for compression quality 75 and higher.
Change-Id: I94513d51078340fbc0737d459fab2cebdd2d6082
the multiplications done for total_size would be done with integers,
possibly overflowing, before being promoted to 64-bit for the addition
Change-Id: Id5c127c8a497ce5de89a276c17f36b59eeb67c21
huff_image_size was a size_t (=32 bits with 32-bit builds) which could
rollover causing an incorrectly sized allocation and a crash in lossless
encoding.
fixes issue #128
Change-Id: I175c8c6132ba9792034807c5c1028dfddfeb4ea5
in debug mode, some float operations see their intermediate
values stored in memory rather than staying in the FPU (which
is 80bit precision).
Several fixes are possible (breaking long calculations into
atomic steps for instance), but simpler of all is just about
turning the cost[] array into float* instead of double*.
The code is a tad faster, and i didn't see any major output
size difference.
Change-Id: Icf1f833e15f8ee4ecc7f9a521d07fdc96ef711aa
Returning 0 (equal) can lead to undefined behaviour.
And, in our cases we'll never have equal keys (added asserts for that)
Change-Id: Ifaf202df321d3f877ad2a03de42e0d6cdd1b2388
fixes the 'blocky sky problem' (saturation problem: when luma was flat,
chroma noise was taking over, resulting in random segment id assigned.
When just using a common uniform segment was better).
+ side clean-up and readibility/experimentability MACRO'ization
+ added '-map 7' option
Change-Id: I35982a9e43c0fecbfdd7b05e4813e8ba8c121d71
Added a threshold of MAX_COLORS_FOR_GRAPH for color-palettes, above
which the graph hint is ignored.
Change-Id: Ia5d7f45e52731b6eaf2806999d6be82861744fd3
spurious in this case, but addresses e.g.,
... potentially uninitialized local variable 'weighted_average' used
Change-Id: Ib99998bf49e4af7a82ee66f13fb850ca5b17dc71
Order-by-cost mostly unchanged (up to a scaling constant 1/log(2))
(except for few minor diff in < 2% of cases)
+ remove unused field cost_mode->cache_bits_
Change-Id: I714f8ab12f49a23f5d499a64c741382c9b489a3e
* ~1-4% faster
* if it's not used, don't use it
* remove the special handling of cache_bits = 0
* remove some tests in the loops
Change-Id: I19d87c3ca731052ff532ea8b2d8e89816507b75f
For low-color images, it may be better to not use color-palettes.
Users should treat this as one another hint (as with Photo &
Picture) and another parameter for tuning the compression density.
The optimum compression can still be obtained by running (outer loop)
compression with all possible tunable parameters.
Change-Id: Icb1a4face2a84774e16e801aee4a8ae97e232e8a
also relocate user_data from WebPAuxStats to the WebPPicture struct to
make clearing easier while placing it closer to the progress hook with
which it's used.
prior to this change some spurious lossless data could be reported in
the lossy (sans alpha) encoding case. additionally user_data could be
lost during lossless encoding.
Change-Id: I929fae3dfde4d445ff81bbaad51445ea586dd80b
* Extend AuxStats with new fields
it's slightly ABI-incompatible, but i guess it's ok for 0.1.99+
I expect to add more stats later, possibly (predictor stats, etc.)
* Have cwebp report the features used by lossless
compression (either for alpha or full lossless coding)
* Print the PSNR for alpha (useful in case of -alpha_q)
* clean-up alpha.c signatures
+ misc cleanup (added const '* const ptr', etc.)
Change-Id: I157a21581f1793cb0c6cc0882e7b0a2dde68a970
Evaluated the impact of this change over 1000 image corpus.
The compression density is up (on average) by 1.2% and encoding time has
gone down considerably from 716 ms (per file) to 146 ms (per file)
(4.9X improvement in encoding time).
Change-Id: Ida562cc0bfe18c9d6f5f00873c95f8396b480eab
Adds new methods WebPPictureARGBToYUVA() and WebPPictureYUVAToARGB()
Depending on the value of picture->use_argb_input,
the main call WebPEncode() will convert appropriately.
Note that both conversions are lossy, so it's recommended to:
* use YUVA input for lossy compression (picture->use_argb_input=0)
* use ARGB input for lossless compression (picture->use_argb_input=1)
Change-Id: I8269d607723ee8a1136b9f4999f7ff4e657bbb04
* add a real proper pointer for holding memory chunk pointer
(instead of using y and argb fields)
* polish the doc with details
* add a WebPPictureView() that extract a view from a picture
without any copy (kind of a fast-crop).
* properly snap the top-left corner for Crop/View. Previously,
the luma position was not snapped, and was off compared to the chroma.
Change-Id: I8a3620c7f5fc6f7d1f8dd89d9da167c91e237439
Adds a test of enc->mb_h_ to VP8IteratorProgress() to avoid a division
by zero. Fixes issue #121.
Original patch from even rouault (even dot rouault at gmail dot com).
Change-Id: Ie5fcc1821860c0a9366d5aa11f3aded4f5b98ed7
This limit corresponds to default histo_bits=3 for images upto sizes 400x400.
Any image higher than this dimension will bump up the histo_bits to 4 internally.
Change-Id: Ic8ba3dcd50e9c588cbbc4a0457289086498ff4ee
Fix inequality assertion on number of palette colors.
Fix inequality assertion test in BackwardReferencesHashChainFollowChosenPath.
Change-Id: Ie3242f1bbeaf96db91b839b6732ccce2634cebf3
Used when we don't code a Huffman Image.
-> Simplify the code quite some because we don't have to
deal with special cases of histo bits
Change-Id: I0c3f46cbf3b501e021c093e07253e7404c01ff4f
Updated histo_bits to 3 from 4 and changed the quality threshold for inner loop for HashChainFindCopy.
Impact: 0.5%-0.8% better bpp with 15%-20% hit on encoding throughput
at default encoding settings.
Change-Id: I316ef88403148b1e19036fa0817d944eb0301255
Ensure that the lossless bit-stream doesn't allow for such cases and
safe-gaurd decoder against indefinite recursion.
Change-Id: Ia6d7f519291de8739f79a977a5800982872aae71
When importing BGRA or RGBA data for encoding, provide variants of
the WEBPImportPicture API for RGBX and BRGX data meaning the alpha
channel should be ignored.
Author: noel@chromium.org
from Chromium patch: https://chromiumcodereview.appspot.com/10496016/
Change-Id: I15fcaa4160c69a2b5549394204b6e6d7a1c5d333
VP8-lossy will now avoid writing an ALPH chunk if the
alpha values are trivial.
+ changed DumpPicture() accordingly in cwebp
+ prevented the -d option to be active with lossless
(DumpPicture wouldn't work).
Change-Id: I34fdb108a2b6207e93fa6cd00b1d2509a8e1dc4b
Change the lossless signature to 0x2f
Add 1 bit indicator for 'droppable (or trivial) alpha)'.
Add 3 bit lossless version (for future extension like yuv support).
Change the sub-resolution information to 3 bits implying range [2 .. 9]
Change-Id: Ic7b8c069240bbcd326cf5d5d4cd2dde8667851e2
Take picture and percent value storage location instead of VP8Encoder.
This will allow reuse by the lossless encoder.
Change-Id: Ic49dbc800cc3e2df60d20f4ebac277f68ed6031b
CostModelBuild() and TrackBackwards() returns weren't checked
+ code clean-up
+ de-inline VP8LBackwardRefs non-critical methods
+ shuffle the .h around to group things together
+ extract some constants as #define's
+ fixed the "if (!(cc_init = ...)) {...}" constructs
+ removed some unneeded VP8L prefixes
Change-Id: Ic634cb87bc6b2033242d3e8e8731fab4c134f327
we vary linearly lossless-method between 0 and 6,
and lossless-quality between 50 and 100, so that encoding
speed can go from 'quite fast' to 'rather slow'.
Impact on size is moderate, but visible.
Change-Id: I0b7917e7170eb50258afb1a4e248028cd9e9207d
This saves ~26 bytes of headers.
* introduce new VP8LDecodeAlphaImageStream() for decoding
* use VP8LEncodeStream() for encoding
* refactor code a bit
still TODO: make the alpha-quality/enc-method user-configurable
Change-Id: I23e599bebe335cfb5868e746e076c3358ef12e71
* RIFF header is omitted
* rename NewVP8LEncoder and DeleteVP8Encoder
* change the signature to take a "const WebPPicture*"
(it was non-const just because we were setting some error potentially)
* made the pic_ field const in VP8LEncoder too.
* trap the bitwriter::error_ too
* simplify some signatures to take WebPPicture* instead
of unneeded VP8LEncoder*
VP8LEncodeStream() will be called directly to compress alpha channel
header-less.
Change-Id: Ibceef63d2b3fbc412f0dffc38dc05c2dee6b6bbf
This is a border-case situation: the picture is not const, because
we're change its error status. But taking it non-const forces
the caller to carry a non-const picture all around the code just
in case (0.00001% of the time?) something bad happen.
This pretty much the same as making all objects non-const because
we'll eventually call delete or free() on them, which is quite a
non-const operation. Well... Better allow constness enforcement for
the remaining 99.9999% of the code.
Change-Id: I9b93892a189a50feaec1a3a518ebf488eb6ff22f
now, we only use 2 bits for the filtering method, and 2 bits
for the compression method.
There's two additional bits which are INFORMATIVE, to specify
whether the source has been pre-processed (level reduction)
during compression. This can be used at decompression time
for some post-processing (see DequantizeLevels()).
New relevant spec excerpt:
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| ChunkHeader('ALPH') |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|Rsv| P | F | C | Alpha Bitstream... |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Compression method (C): 2 bits
: The compression method used:
* `0`: No compression.
* `1`: Backward reference counts encoded with arithmetic encoder.
Filtering method (F): 2 bits
: The filtering method used:
* `0`: None.
* `1`: Horizontal filter.
* `2`: Vertical filter.
* `3`: Gradient filter.
Pre-processing (P): 2 bits
: These INFORMATIVE bits are used to signal the pre-processing that has
been performed during compression. The decoder can use this information to
e.g. dither the values or smooth the gradients prior to display.
* `0`: no pre-processing
* `1`: level reduction
Decoders are not required to use this information in any specified way.
Reserved (Rsv): 2 bits
: SHOULD be `0`.
Alpha bitstream: _Chunk Size_ - `1` bytes
: Encoded alpha bitstream.
This optional chunk contains encoded alpha data for a single tile.
Either **ALL or NONE** of the tiles must contain this chunk.
The alpha channel data is losslessly stored as raw data (when
compression method is '0') or compressed using the lossless format
(when the compression method is '1').
Change-Id: Ied8f5fb922707a953e6a2b601c69c73e552dda6b
* Method #1 is now calling the lossless encoder on the alpha plane.
Format is not final, it's just a first draft. We need ad-hoc functions.
* removed now useless utils/alpha.*
* added utils/quant_levels.h instead
* removed the TCoder code altogether
Change-Id: I636840b6129a43171b74860e0a0fc5bb1bcffc6a
-> lot of simplifications ensue and we should be able to get rid of
ClearHuffmanTreeIfOnlyOneSymbol() too, in a subsequent patch.
Change-Id: Ic4c51d05e4b1970e37f94ffd85fae6a02e4a6422
Allocate big chunk of memory in GetHuffBitLengthsAndCodes, instead of allocating in a loop.
Also fixed the potential memleak.
Change-Id: Idc23ffa306f76100217304444191a8d2fef9c44a
The size written in VP8L header should be without padding.
(Also clarified this code using consts).
Change-Id: Ic6583d760c0f52ef61924ab0330c65c668a12fdc
if histogram_image_size is reduced in when writing the histogram_image
the bit arrays would leak any remaining elements. store their element
count separately.
Change-Id: I710142a11ebd4325faec7bd65c2d2572aae19307
The current implementation doesn't take care one byte
signature and associated one byte padding (for odd sized chunk).
Change-Id: I35b81d0644818cdba38189aa48c75db5f92e68f4
The new method VP8LGetBackwardReferences hides internal
heuristics used for choosing RLE or LZ77 based refs.
- Tuned VP8LHashChainFindCopy for better compression at higher Q.
- Refactored code.
- Removed the unused method VP8LVerifyBackwardReferences.
Change-Id: Ibb7bb072bab5a49a001577a20d88226f52e6c663
arrays can be passed directly as only their members are being modified.
this also reduces the allocation for bit_codes[] by taking the
sizeof(type)=2 rather than sizeof(ptr)=4/8 in one case.
Change-Id: Idad20cead58c218b58d90b71699374fefd01cad9
* lossless_encoder: (46 commits)
split StoreHuffmanCode() into smaller functions
more consolidation: introduce VP8LHistogramSet
big code clean-up and refactoring and optimization
Some cosmetics in histogram.c
Approximate FastLog between value range [256, 8192]
Forgot to update out_bit_costs to symbol_bit_costs at one instance.
Evaluate output cluster's bit_costs once in HistogramRefine.
Simple Huffman code changes.
Lossless decoder: remove an unneeded param in ReadHuffmanCodeLengths().
Reducing emerging palette size from 11 to 9 bits.
Move GetHistImageSymbols to histogram.c
Improve predict vs no-predict heuristic.
code-moving and clean-up
reduce memory usage by allocating only one histo
Restrict histo_bits to ensure histo_image size is under 32MB
further simplification for the meta-Huffman coding
A quick pass of cleanup in backward reference code
Make transform bits a function of encode method (-m).
introduce -lossless option, protected by USE_LOSSLESS_ENCODER
Run TraceBackwards for higher qualities.
...
Conflicts:
src/enc/webpenc.c
Change-Id: I9a5d98cba0889ea91d10699466939cc283da345a
VP8LHistogramSet is container for pointers to histograms that
we can shuffle around. Allocation is one big chunk of memory.
Downside is that we don't de-allocate memory on-the-go during
HistogramRefine().
+ renamed HistogramRefine() into HistogramRemap(), so we don't
confuse with "HistogramCombine"
+ made VP8LHistogramClear() static.
Change-Id: Idf1a748a871c3b942cca5c8050072ccd82c7511d
* de-inline some function
* make VP8LBackwardRefs be more like a vectorwith max capacity
* add bit_cost_ field to VP8LHistogram
* general code simplifications
* remove some memmov() from HistogramRefine
* simplify HistogramDistance()
...
Change-Id: I16904d9fa2380e1cf4a3fdddf56ed1fcadfa25dc
Avoid bit_costs evaluated every time in function HistogramDistance. Also moved
VP8LInitBackwardRefs and VP8LClearBackwardRefs to backward_references.h
Change-Id: Id507f164d0fc64480aebc4a3ea3e6950ed377a60
No empty trees are codified with the simple Huffman code. The simple Huffman
code is simplified to be either a 1-bit code or 8-bit code for symbols.
Change-Id: I3e2813027b5a643862729339303d80197c497aff
This improves compression density. For example, at quality 95 on 1000 PNGs:
bpp(before) = 2.447 and bpp(after) = 2.412
Change-Id: I19c343ba05cca48a6940293721066502a5c3d693
* removed use_lz77_ field, and added cache_bits_ one.
* use more BackwardRefs params
* move code around to organize more logically
* reduce memory use on histo
...
Change-Id: I833217a1b950189cf486704049e3fe28382ce335
instead of lz77+rle
* introduce VP8LBackwardRefs structure and simplify the code by not passing
around {PixOrCopy/int} pairs.
More functions should be turned into using this struct (TODO(later)).
Change-Id: I69c5c9fa61dddd61a2abc2824d70b8606a1c55b6
* don't transmit the number of Huffman tree group explicitly
* move color-cache information before the meta-Huffman block
* also add a check that color_cache_bits is in [1..11] range, as per spec.
Change-Id: I81d7711068653b509cdbc1151d93e229c4254580
(Essentially, there was no need of a separate 'argb_palette' array. And
argb_palette[0] was never being set).
Change-Id: Id0a8c7e063d3af41e39fc9b8661611b51ccc55cd
so that it uses original values of left, top etc for prediction rather than the
predicted values of the same. Also, do some renaming in the same to make it
more readable.
Change-Id: I2fe94e35a6700bd437f5c601e2af12323bf32445
- VP8LEncAnalyze, EvalAndApplySubtractGreen, ApplyPredictFilter,
ApplyCrossColorFilter
- Added palette handling and transform buffer management in VP8LEncodeImage()
- Add Transforms (subtract Green, Predict, cross_color) to dsp/lossless.c.
These are more-or-less copied from src/lossless code.
After this Change, will implement the EncodeImageInternal() method.
Change-Id: Idf71f803c24b3b5ae3b5079b15e019721784611d
src/dec/Makefile.am: add missing reference to vp8li.h
src/{dec,dsp,enc}/Makefile.am: move some headers to noinst_
Change-Id: I0e2bc69980bd8175d99ad0ab63f537ef9e425b77
- use common file organization across subdir makefiles
- append lib/source/header list variables and sort
Change-Id: I0653e1c73a4552b0c43d21f321b22b4972d6e87b
- remove some unused functions
- move global arrays from data to read only section
- explicitly cast malloc returns; not specifically necessary, but helps
show intent
- miscellaneous formatting
Change-Id: Ib15fe5b37fe6c29c369ad928bdc3a7290cd13c84
Add a dirty_ flag to keep track of updated probabilities and the need to
recompute the level costs.
This only makes a difference for "-m 2" method which was sub-optimal.
But it's overall cleaner to have this flag.
Change-Id: I21c71201e1d07a923d97a3adf2fbbd7d67d35433
This proved being ok, even for large pictures, provided one
takes care of overflow. When an overflow is bound to occur, the
counters are renormalized.
Overall, shaves ~12k of memory.
Change-Id: I2ba21a407964fe1a34c352371cba15166e0c4548
the p0 proba was incorrectly accumulated. Merging its contribution into
the LevelCost[] was creating more problems than anything (esp. with trellis)
so let's just not.
Change-Id: I4c07bfee471085df901228d97b20a4d9606ba44e
These will report the 7x7-averaged PSNR or SSIM, using the
new internal function WebPPictureDistortion().
This is for information only. These flags have no encoding impact.
+misc opportunistic cosmetics
Change-Id: I64c0a7eca679134d39062e438886274b22bb643f
to 'clean up' the fully-transparent area and make it more compressible
new cwebp flags: -alpha_cleanup (off by default, since gain is not 100% guaranteed)
Change-Id: I74d77e1915eee146584cd61c9c1132a41db922eb
.. where only 2 filtering modes are potentially
tried, instead of all of them. This is fast than the exhaustive 'best'
mode, and not much worse.
Options for cwebp are:
-alpha_filter none
-alpha_filter fast (<- default)
-alpha_filter best (<- slow)
Change-Id: I8cb90ee11b8f981811e013ea4ad5bf72ba3ea7d4
Add predictive filtering option for Alpha plane.
Valid range for filter option is [0, 5] corresponding to prediction
methods none, horizontal, vertical, gradient & paeth filter.
The prediction method 5 will try all the prediction methods (0 to 4)
and pick the prediction method that gives best compression.
Change-Id: I9244d4a9c5017501a9696c7cec5045f04c16d49b
Extend WebP Encode functionality to encode Alpha data and produce
bit-stream (RIFF+VP8X+ALPH+VP8) corresponding to WebP-Alpha.
Change-Id: I983b4cd97be94a86a8e6d03b3b9c728db851bf48
This has been pointed as a useful information to have in the header (for
the non VP8-specs savvy ones)
Change-Id: I494b1da41dfafce882a94e3677d1cd6206bc504b
Was getting compiler error when I included bit_writer.h from non libwebp
directory. bit_writer.h includes vp8enci.h and that uses VP8BitWriter without
having it's definition.
Change-Id: I1ca82594292979b9eb7e60e2fffb22c16768dd30
Gathers all DSP-related function (and SSE2 implementations).
Clean-up some unwanted symbolic dependencies so that webp_encode,
webp_decode and webp_dsp are truly independent libraries.
+ opportunistic clean-up:
* remove unneeded VP8DspInitTables(), now integrated in VP8DspInit()
* make consistent use of VP8GetCPUInfo() in the various DspInit() funcs
* change OUT macro to DST
uint8_t is a gcc extension which msvc similarly supports, but for
greater compatibility, and to match the change already made in
dec/vp8i.h, update the remaining bitfield to use unsigned int.
Change-Id: Id9dca470345871e00e82893255a306dfe5d3fa29
Although it degrades quality, this option is useful to avoid the 512k
limit for partition #0.
If not enough to reach the lower bound of 4bits per macroblock header,
one should also limit the number of segments used (down to -segments 1)
See the man file for extra details.
Change-Id: Ia59ffac13176c85b809ddd6340d37b54ee9487ea
was missing from the RD-computation of intra-4x4 score.
Doesn't change anything significantly, it's just More Correct.
Change-Id: I25c5b53a810d97e6fb7f98c549fd23bbe55e1bf4
~5-10% faster.
Heavy 8bit arithmetic trickery!
Patch by Somnath Banerjee (somnath at google dot com)
Change-Id: I9fd2c511d9f631e9cf4b008c46127b49fb527b47
Use _MSC_VER as the intrinsics compile without /arch:SSE2 on x86.
Also avoids applying the same flag to all files which defeated the
purpose of the runtime cpu-detection.
Thanks to Frank B. for the suggestion!
Change-Id: Iae9933a3cee704e663d9bbd53d0fa68e8c025425
picture->error_code can be looked up for finer error diagnose.
Added readable error messages to cwebp too.
Should close bug #75 (http://code.google.com/p/webp/issues/detail?id=75)
Change-Id: I8889d06642d90702f698cd5c27441a058ddb3636
sometimes, gcc insert sse2 storeu instructions (like in VP8InitFilter())
with aligment requirements.
Bug was visible 'sometimes' in non-debug mode, when trying to use -af.
Change-Id: If3ec282bbbb9f9d0d33ca4b2c4bed46cd26fe495
we were reading past the end of the dqs[] array.
reported by Mathias Schindler (on cygwin only)
http://code.google.com/p/webp/issues/detail?id=71
Change-Id: Ib38c4c139e3cac3e8915626d63e16b403d6bbd63
+ add a simple rescaling function: WebPPictureRescale() for encoding
+ clean-up the memory managment around the alpha plane
+ fix some includes path by using "../webp/xxx.h" instead of "webp/xxx.h"
New flags for 'cwebp':
-resize <width> <height>
-444 (no effect)
-422 (no effect)
-400
Change-Id: I25a95f901493f939c2dd789e658493b83bd1abfa
This is a (minor) bitstream change: if the 'color_space' bit is set to '1'
(which is normally an undefined/invalid behaviour), we add extra data at the
end of partition #0 (so-called 'extensions')
Namely, we add the size of the extension data as 3 bytes (little-endian),
followed by a set of bits telling which extensions we're incorporating.
The data then _preceeds_ this trailing tags.
This is all experimental, and you'll need to have
'#define WEBP_EXPERIMENTAL_FEATURES' in webp/types.h to enable this code
(at your own risk! :))
Still, this hack produces almost-valid WebP file for decoders that don't
check this color_space bit. In particular, previous 'dwebp' (and for instance
Chrome) will recognize this files and decode them, but without the alpha
of course. Other decoder will just see random extra stuff at the end of
partition #0.
To experiment with the alpha-channel, you need to compile on Unix platform
and use PNGs for input/output.
If 'alpha.png' is a source with alpha channel, then you can try (on Unix):
cwebp alpha.png -o alpha.webp
dwebp alpha.webp -o test.png
cwebp now has a '-noalpha' flag to ignore any alpha information from the
source, if present.
More hacking and experimenting welcome!
Change-Id: I3c7b1fd8411c9e7a9f77690e898479ad85c52f3e
For now, SSE2 functions are compiled a-minima: only on platforms
where __SSE2__ is defined. Let's later add some autoconf-based
config to enable/disable at will.
One can disable SSE2 at run-time by hooking-up VP8GetInfo.
There is a new option "-noasm" in cwebp for that.
Output should be binary the same between C and SSE2 version. If not,
that's a bug!
patch by Christian Duvivier (cduvivier at google dot com)
Change-Id: Iae006c3cdcb7e8280e846cedb94d239dab1e42ae
to return the sum directly.
output is bitwise the same, speed up 1-2%. This is preparatory to a
more efficient SSE2 implementation.
Change-Id: I0bcdf05808c93420fbe9dcb75e5e7e55a4ae5b89
