This is getting back to the old behavior which is actually better for
compression and speed with the latest patches.
Change-Id: I35884bab02589297c25d6e1e66dc5f13e05f7aa7
This was defined (slightly differently) at two places. Created a common
method and moved to utils/utils.[hc].
Change-Id: I66c3ac6dea24e0cd2c0eaa5440f3142b4dbbe23b
we don't need to store the resulting histogram, so no need to
call HistogramAddEval().
Allows some signature simplifications...
Change-Id: I3fff6c45f4a7c6179499c6078ff159df4ca0ac53
In case where the same offset is found in consecutive pixels,
the cost computation from one pixel can be re-used for the next.
Change-Id: Ic03c7d4ab95f3612eafc703349cfefd75273c3d7
and also recycle the malloc'd intervals
This avoids quite some malloc/free cycles during interval managment.
Change-Id: Ic2892e7c0260d0fca0e455d4728f261fb4c3800e
In a lot of cases, only one interval is used. This can cause
a lot of malloc/free cycles for only 56 bytes. By caching this
single interval and re-using it, we remove this cycle in most
frequent cases.
Change-Id: Ia22d583f60ae438c216612062316b20ecb34f029
In some cases, the hash chain for a function is filled several
times:
- GetBackwardReferences -> CalculateBestCacheSize ->
BackwardReferencesLz77 that computes the hash chain
- GetBackwardReferences ->
(not always) BackwardReferencesTraceBackwards ->
BackwardReferencesHashChainDistanceOnly that computes the hash
chain in a slightly different way
Speed and compression performance are slightly changed (+ or -)
but will be homogneized in a later patch.
Change-Id: I43f0ecc7a9312c2ed6cdba1c0fabc6c5ad91c953
Instead of comparing all the following pixels over len (which can
frequently reach the maximum MAX_LENGTH=4096 for some images),
intervals are stored and compared.
Change-Id: I0dafef6cc988dde3c1c03ae07305ac48901d60ee
The old implementation in enc/near_lossless.c performing a separate
preprocessing step is used only when a prediction filter is not used,
otherwise a new implementation integrated into lossless_enc.c is used.
It retains the same logic for converting near lossless quality into max
number of bits dropped, and for adjusting the number of bits based on
the smoothness of the image at a given pixel. As before, borders are not
changed.
Then, instead of quantizing raw component values, the residual after
subtract green and after prediction is quantized according to the
resulting number of bits, taking care to not cross the boundary between
255 and 0 after decoding. Ties are resolved by moving closer to the
prediction instead of by bankers’ rounding.
This results in about 15% size decrease for the same quality.
Change-Id: If3e9c388158c2e3e75ef88876703f40b932f671f
the number of segments are previously validated, but an explicit check
is needed to avoid a warning under gcc-4.9
this is similar to the changes made in:
c8a87bb AssignSegments: quiet -Warray-bounds warning
3e7f34a AssignSegments: quiet array-bounds warning
Change-Id: Iec7d470be424390c66f769a19576021d0cd9a2fd
This avoids generating file that would trigger a decoding bug
found in 0.4.0 -> 0.4.3 libwebp versions.
This reverts commit 6ecd72f845.
Change-Id: I4667cc8f7b851ba44479e3fe2b9d844b2c56fcf4
The mode's bits were not taken into account, which is ok for most of cases.
But in case of super large image, with 'easy' content, their overhead starts
mattering a lot and we were omitting to optimize for these.
Now, these mode bits have their own lambda values associated, limiting
the jerkiness. We also limit (for -m 2 only) the individual number of bits
to something that will prevent the partition 0 overflow.
removed the I4_PENALTY constant, which was a rather crude approximation.
Replaced by some q-dependent expression.
fixes issue #289
Change-Id: I956ae2d2308c339adc4706d52722f0bb61ccf18c
This is in preparation for some SSE2 code.
And generally speaking, the whole SSIM code needs some
revamp: we're not averaging the SSIM value at each pixels
but just computing the overall SSIM value once, for the whole
plane. The former might be better than the latter.
Change-Id: I935784a917f84a18ef08dc5ec9a7b528abea46a5
- The result is now indeed closest among possible results for all inputs, which
was not the case for bits>4, where the mapping was not even monotonic because
GetValAndDistance was correct only if the significant part of initial fit in
a byte at most twice.
- The set of results for a larger number of bits dropped is a subset of values
for a smaller number of bits dropped. This implies that subsequent
discretizations for a smaller number of bits dropped do not change already
discretized pixels, which improves the quality (changes do not accumulate)
and compression density (values tend to repeat more often).
- Errors are more fairly distributed between upwards and downwards thanks to
bankers’ rounding, which avoids images getting darker or lighter in overall.
- Deltas between discretized values are more repetitive. This improves
compression density if delta encoding is used.
Also, the implementation is much shorter now.
Change-Id: I0a98e7d5255e91a7b9c193a156cf5405d9701f16
We were not updating the current_width_, which is usually
not a problem, unless we use Delta Palette with small number
of colors
-> Addressed this re-entrancy problem by checking we have
enough capacity for transform buffer.
The problem is not currently visible, until we restrict
the number of gradient used in delta-palette to less than 16.
Then the buffers have different current_width_ and the problem
surfaces.
Change-Id: Icd84b919905d7789014bb6668bfb6813c93fb36e
The code and logic is unified when computing bit entropy + Huffman cost.
Speed-wise, we gain 8% for lossless encoding.
Logic-wise, the beginning/end of the distributions are handled properly
and the compression ratio does not change much.
Change-Id: Ifa91d7d3e667c9a9a421faec4e845ecb6479a633
setting all transparent pixels to black rather than the "flatten" method.
0.3% smaller filesize on the 1000 PNGs if alpha cleanup is used (before: 18685774, after: 18622472)
Change-Id: Ib0db9e7ccde55b36e82de07855f2dbb630fe62b1
The functions containing magic constants are moved out of ./dsp .
VP8LPopulationCost got put back in ./enc
VP8LGetCombinedEntropy is now unrefined (refinement happening in ./enc)
VP8LBitsEntropy is now unrefined (refinement happening in ./enc)
VP8LHistogramEstimateBits got put back in ./enc
VP8LHistogramEstimateBitsBulk got deleted.
Change-Id: I09c4101eebbc6f174403157026fe4a23a5316beb
The previous priority system used a heap which was too heavy to
maintain (what was gained from insertions / deletions was lost
due to a linear that still happened on the heap for invalidation).
The new structure is a priority queue where only the head is
ordered.
Change-Id: Id13f8694885a934fe2b2f115f8f84ada061b9016
SimpleQuantize()
it's now a single function, that reconstructs the intra4x4 block during the scan
The I4_PENALTY had to be adjusted.
Overall, result is better quality-wise (esp. at q < 50), and a tad faster too.
method #0, #1 and #3+ are unchanged
Change-Id: If262aeb552397860b3dd532df8df6b1357779222
Gives 0.9% smaller (2.4% compared to before alpha cleanup) size on the 1000 PNGs dataset:
Alpha cleanup before: 18856614
Alpha cleanup after: 18685802
For reference, with no alpha cleanup: 19159992
Note: WebPCleanupTransparentArea is still also called in WebPEncode. This cleanup still helps
preprocessing in the encoder, and the cases when the prediction transform is not used.
Change-Id: I63e69f48af6ddeb9804e2e603c59dde2718c6c28
The 32-bit buffers are actually rarely 64-bit aligned.
The new solution uses memcmp and is alignment agnostic.
It is also slightly faster.
Change-Id: I863003e9ee4ee8a3eed25b7b2478cb82a0ddbb20
Arrays were compared 32 bits at a time, it is now done 64 bits at a time.
Overall encoding speed-up is only of 0.2% on @skal's small PNG corpus.
It is of 3% on my initial 1.3 Mp desktop screenshot image.
Change-Id: I1acb32b437397a7bf3dcffbecbcd4b06d29c05e1
instead of per block. This prepares for a next CL that can make the
predictors alter RGB value behind transparent pixels for denser
encoding. Some predictors depend on the top-right pixel, and it must
have been already processed to know its new RGB value, so requires per
scanline instead of per block.
Running the encode speed test on 1000 PNGs 10 times with default
settings:
Before:
Compression (output/input): 2.3745/3.2667 bpp, Encode rate (raw data): 1.497 MP/s
After:
Compression (output/input): 2.3745/3.2667 bpp, Encode rate (raw data): 1.501 MP/s
Same but with quality 0, method 0 and 30 iterations:
Before:
Compression (output/input): 2.9120/3.2667 bpp, Encode rate (raw data): 36.379 MP/s
After:
Compression (output/input): 2.9120/3.2667 bpp, Encode rate (raw data): 36.462 MP/s
No effect on compressed size, this produces exactly same files. No
significant measured effect on speed. Expected faster speed from better
memory layout with scanline processing but slower speed due to needing
to get predictor mode per pixel, may compensate each other.
Change-Id: I40f766f1c1c19f87b62c1e2a1c4cd7627a2c3334
Rename the flag to exact instead of the opposite cleanup_alpha. Add the flag to
WebPConfig. Do the cleanup in the webp encoder library rather than the cwebp
binary, this will be needed for the next stage: smarter alpha cleanup for
better compression which cannot be done as a preprocessing due to depending on
predictor choices in the encoder.
Change-Id: I2fbf57f918a35f2da6186ef0b5d85e5fd0020eef
global effect is ~2% faster encoding from JPG source
and ~8% faster lossless-webp source decoding to PGM (e.g.)
Also revamped the YUVA case to first accumulate R/G/B value into 16b
temporary buffer, and then doing the UV conversion.
-> New function: WebPConvertRGBA32ToUV
Change-Id: I1d7d0c4003aa02966ad33490ce0fcdc7925cf9f5
Just for RGB24/BGR24 for now, which are the hard-to-optimize ones.
SSE2 implementation coming next.
ConvertRowToY() should go into dsp/ too, at some point.
Change-Id: Ibc705ede5cbf674deefd0d9332cd82f618bc2425
Note that ALIGN_CST is still kept different in dec/frame.c for now,
because the values is 31 there, not 15. We might re-unite these two
later.
Change-Id: Ibbee607fac4eef02f175b56f0bb0ba359fda3b87
same functionality, but better code layout.
What changed:
* don't trash the palette_[] in EncodePalette(), so it can be re-used
* split generation of image from bit-stream coding
* move all the delta-palette code to delta_palettization.c, and only have 1 entry point there WebPSearchOptimalDeltaPalette()
* minimize the number of "#ifdef WEBP_EXPERIMENTAL_FEATURES" in vp8l.c
* clarify the TransformBuffer stuff. more clean-up to come here...
This should make experimenting with delta-palettization easier and more compartimentalized.
Change-Id: Iadaa90e6c5b9dabc7791aec2530e18c973a94610
New palette compresses more than 20% better with minimum quality loss.
Tested on set of wikipedia images with command line:
cwebp -delta_palettization
Change-Id: I82ec7d513136599cd70386f607f634502eb9095d
* vertical expansion now uses bilinear interpolation
* heavily assumes that the alpha plane is decoded in full, not row-by-row
* split the RescalerExportRow and RescalerImportRow methods into Shrink
and Expand variants.
* MIPS implementation of ExportRowExpand is missing.
There's room for extra speed optim and code re-org, but let's keep that for later patches.
addresses https://code.google.com/p/webp/issues/detail?id=254
Change-Id: I8f12b855342bf07dd467fe85e4fde5fd814effdb
This makes the chains more efficient and a larger variety of data is tested.
0.02 % compression gain at q 100, 0.05 % at default quality. 0.8 % speedup by
callgrind.
0.16 % compression gain for lossy alpha ?!
Change-Id: I888120133352799eb14f5f602c7f40ab404bd665
using a *tmp_plane buffer to split a/r/g/b planes up appeared to
be the easiest route, compared to copy-pasting the whole code and
making it x_stride aware...
Change-Id: I0898ef1df62bd3e1713b77187b31b5eeef3832fe
Slightly faster on -m 0 -q 0, particularly for small images (50 x 75
image was 0.1 % faster on callgrind measurement).
Increases compression density by 0.005 % for the 1000 images, but small
images can improve even 0.5 % (about 4 bytes, depending on the
characteristics of the palette).
Change-Id: I94f568d396ac62a054a829abeeef3eb0af6b3f94
the x_add/x_sub increments were wrong for u/v in the upscaling case.
They shouldn't be left to the caller's discretion, but set up by
WebPRescalerInit to their exact necessary values.
-> Cleaned-up WebPRescalerInit() param list.
-> added safety asserts
-> removed the mips32/mips_r2 variant of "ImportRow" which were buggy prior
Change-Id: I347c75804d835811e7025de92a0758d7929dfc09
a total impact of 1 % on encoding speed
This allows for performance neutral removal of the binary search
in cache bits selection. This will give a small improvement in
compression density.
Change-Id: If5d4d59460fa1924ce71af977320834a47c2054a
0.21 % compression density improvement for 1000 png corpus in
lossless mode
0.50 % compression density improvement for 1000 png corpus in
lossy mode
Change-Id: I14ee8c427ae5d3e116b0ee6695fcdea3321a319d
do not do length 2 matches far away
speedup for non compressible data by inserting two literals at a time
when no matches are found
Change-Id: Ia8e033071f4186bb8148bb2bf13ca37586734aa3
Increases compression density by 0.03 % for lossy.
Speeds up at least one of the lossy alpha images by 20 %.
Palette entropy 'kludge' seems to save 1-2 % on alpha images.
Change-Id: I2116b8d81593ac8173bfba54a7c833997fca0804
share the computation between different modes
3-5 % speedup for lossless alpha
1 % for lossy alpha
no change in compression density
Change-Id: I5e31413b3efcd4319121587da8320ac4f14550b2
introduced in:
"lossless: 0.37 % compression density improvement"
Uses the statistics of red and blue histograms to decide if to run
cross color correction at all.
Improves compression density by 0.02 % or so.
Change-Id: I47429557e9cdbd9fa90c584696f241b17427d73f
No significant size degradation (+0.001 %) for 1000 image corpus
Fixes the 8 ms vs 2 ms degradation from:
"lossless: 0.37 % compression density improvement"
Change-Id: Id540169a305d9d5c6213a82b46c879761b3ca608
counting the entropy expectation for five different configurations:
palette
non-predicted
non-predicted with subtract green
predicted
predicted with subtract green
and choose the strategy with the smallest expected entropy
Change-Id: Iaaf209c0d565660a54a4f9b3959067afb9951960
Speed-wise equivalent on x86 and ARM (maybe a tad faster, hard to tell).
Note that the two 32-bit multiples are not strictly equivalent
to the 64-bit one, since we're missing one carry propagation.
In practice, no observable difference was seen because of this
slightly different hashing result.
Change-Id: I8f2381175eae1cb20dabf149e6b27e1768fba6ab
had to rename few structs.
-> we can now include both vp8i.h and vp8enci.h without naming
conflicts.
Change-Id: Ib41b498f1b57aab3d6b796361afc45210ec75174
we look at average global improvement and stop when things are
moving slow, or when we had a quite good first iteration already
(means: the picture is "not difficult")
Change-Id: I8ab7d100353039b5b32bb5fac3fe03c8440c78d5
Speedup method StoreImageToBitMask by replacing the code to find histogram
index and Huffman tree codes at every iteration to a more optimal code that
updates these only when the current pixel (to write) crosses the histogram
tile-row boundary.
This change speeds up the StoreImageToBitMask method by 5%.
Change-Id: If01a1ccd7820f9a3a3e5bc449d070defa51be14b
The MIPS code for cost is not updated yet, that's why i keep Residual::*cost
around for now. Should be removed in favor of *costs later.
Change-Id: Id1d09a8c37ea8c5b34ad5eb8811d6a3ec6c4d89f
removes circular dependency between dsp and enc.
since:
a987fae MIPS: dspr2: added optimization for function GetResidualCost
Change-Id: Ifeb8fc02de89e2ba982ed7ffacd925d649bfec3c
kGammaFix is now only defined with USE_GAMMA_COMPRESSION;
fixes:
use of undeclared identifier 'kGammaFix'
Change-Id: Ib1e2f410eff9b83be065894f88181f91dd2776e1
set/get residual C functions moved to new file in src/dsp
mips32 version of GetResidualCost moved to new file
Change-Id: I7cebb7933a89820ff28c187249a9181f281081d2
the input to the function is non-const and the pointer being operated is
being free'd; removes an unnecessary cast in the process
Change-Id: Ic515ed672ddf7f8e4e36eeac696ff7aa8a3652f7
Updated the near-lossless level mapping and make it correlated to lossy
quality i.e 100 => minimum loss (in-fact no-loss) and the visual-quality loss
increases with decrease in near-lossless level (quality) till value 0.
The new mapping implies following (PSNR) loss-metric:
-near_lossless 100: No-loss (bit-stream same as -lossless).
-near_lossless 80: Very very high PSNR (around 54dB).
-near_lossless 60: Very high PSNR (around 48dB).
-near_lossless 40: High PSNR (around 42dB).
-near_lossless 20: Moderate PSNR (around 36dB).
-near_lossless 0: Low PSNR (around 30dB).
Change-Id: I930de4b18950faf2868c97d42e9e49ba0b642960
AnalyzeSubtractGreen constitutes about 8-10% of the comression CPU cycles.
Statistically, subtract-green is proved to be useful for most of the
non-palette compression. So instead of evaluating the entropy (by calling
AnalyzeSubtractGreen) apply subtract-green transform for the low-effort
compression.
This changes speeds up the compression at m=0 by 8-10% (with very slight loss
of 0.07% in the compression density).
Change-Id: I9797dc39437ae089716acb14631bbc77d367acf4
Speed up AnalyzeSubtractGreen by looping through the image pixel once to
compute the two histograms.
AnalyzeEntropy code cleanup.
Removed some 'if' conditions and pointer indirections inside pixel iterate loop.
Change-Id: Ia65e3033988ff67df8e3ecce19d6e34cfc76358e
Enable the WebP near-lossless feature by pre-processing the image to smoothen
the pixels.
On a 1000 PNG image corpus, for which WebP lossless (default settings) gets
25% compression gains, following is the performance of near-lossless feature
at various '-near_lossless' levels:
-near_lossless 90: 30% (very very high PSNR 54-60dB)
-near_lossless 75: 38% (very high PSNR 48-54dB)
-near_lossless 50: 45% (high PSNR 42-48dB)
-near_lossless 25: 48% (moderate PSNR 36-42dB)
-near_lossless 10: 50% (PSNR 30-36dB)
WebP near-lossless is specifically useful for discrete-tone images like
line-art, icons etc.
Change-Id: I7d12a2c9362ccd076d09710ea05c85fa64664c38
Simplify and speedup backward references for low-effort settings by evaluating
LZ77 references only. This change speeds up compression by 10-25% at lower
(q <= 25) quality range with a slight drop (0.2%) in the compression density.
Change-Id: Ibd6f03b1a062d8ab9191786c2a425e9132e4779f
Cleaup Near-lossless code
- Simplified and refactored the code.
- Removed the requirement (TODO) to allocate the buffer of size WxH and work
with buffer of size 3xW.
- Disabled the Near-lossless prr-processing for small icon images (W < 64 and H < 64).
Change-Id: Id7ee90c90622368d5528de4dd14fd5ead593bb1b
* use the same TFIX == YFIX precision (2bits)
* use int instead of float in LinearToGammaF()
output is visually equivalent. Code is a little faster.
Change-Id: Ie3cfebca351dbcbd924b3d00801d6523dca6981f
check enc->argb_ to quiet an msvs /analyze warning:
C6387: 'enc->argb_+y*width' could be '0': this does not adhere to the
specification for the function 'memcpy'.
Change-Id: I87544e92ee0d3ea38942a475c30c6d552f9877b7
Disable costly TraceBackwards heuristic for computing the backward references
for low_effort (method=0) compression.
The TraceBackwards heuristic is already disabled for lower (q < 25) quality
range. Following is the compression data for 1000 image corpus for q >= 25.
This speeds up compression (q >= 25) by a factor of 2.5-3X with slight loss of
compression density (0.7% for lower quality range and 1.2% for higher qualities).
Change-Id: I256c9e2137c7de4083f423ea32ee12d3b0f46253
- Lower the threshold parameters for HashChainFindCopy.
For 1000 image PNG corpus (m=0), this change yields speedup of 15-20% at
lower quality range (0.25% drop in compression density) and about 10%
for higher quality range without any drop in the compression density.
Following is the compression stats (before/after) for method = 0:
Before After
bpp/MPs bpp/MPs
q=0 2.8615/18.000 2.8651/18.631
q=5 2.8615/18.216 2.8650/20.517
q=10 2.8572/18.070 2.8650/21.992
q=15 2.8519/18.371 2.8584/21.747
q=20 2.8454/18.975 2.8515/20.448
q=25 2.8230/8.531 2.8253/9.585
// Compression density remains same for q-range [30-100]
q=30 2.7310/7.706 2.7310/8.028
q=35 2.7253/6.855 2.7253/7.184
q=40 2.7231/6.364 2.7231/6.604
q=45 2.7216/5.844 2.7216/6.223
q=50 2.7196/5.210 2.7196/5.731
q=55 2.7208/4.766 2.7208/4.970
q=60 2.7195/4.495 2.7195/4.602
q=65 2.7185/4.024 2.7185/4.236
q=70 2.7174/3.699 2.7174/3.861
q=75 2.7164/3.449 2.7164/3.605
q=80 2.7161/3.222 2.7161/3.038
q=85 2.7153/2.919 2.7153/2.946
q=90 2.7145/2.766 2.7145/2.771
q=95 2.7124/2.548 2.7124/2.575
q=100 2.6873/2.253 2.6873/2.335
Change-Id: I0e17581fb71f6094032ad06c6203350bd502f9a1
- Do light weight entropy based histogram combine and leave out CPU
intensive stochastic and greedy heuristics for combining the
histograms.
For 1000 image PNG corpus (m=0), this change yields speedup of 10% at
lower quality range (1% drop in compression density) and about 5% for
higher quality range (1% drop in compression density). Following is the
compression stats (before/after) for method = 0:
Before After
bpp/MPs bpp/MPs
q=0 2.8336/16.577 2.8615/18.000
q=5 2.8336/16.504 2.8615/18.216
q=10 2.8293/16.419 2.8572/18.070
q=15 2.8242/17.582 2.8519/18.371
q=20 2.8182/16.131 2.8454/18.975
q=25 2.7924/7.670 2.8230/8.531
q=30 2.7078/6.635 2.7310/7.706
q=35 2.7028/6.203 2.7253/6.855
q=40 2.7005/6.198 2.7231/6.364
q=45 2.6989/5.570 2.7216/5.844
q=50 2.6970/5.087 2.7196/5.210
q=55 2.6963/4.589 2.7208/4.766
q=60 2.6949/4.292 2.7195/4.495
q=65 2.6940/3.970 2.7185/4.024
q=70 2.6929/3.698 2.7174/3.699
q=75 2.6919/3.427 2.7164/3.449
q=80 2.6918/3.106 2.7161/3.222
q=85 2.6909/2.856 2.7153/2.919
q=90 2.6902/2.695 2.7145/2.766
q=95 2.6881/2.499 2.7124/2.548
q=100 2.6873/2.253 2.6873/2.285
Change-Id: I0567945068f8dc7888041e93d872f9def91f50ba
inline function MakeARGB32 calls changed to call
via pointers to functions which make (a)rgb for
entire row
Change-Id: Ia4bd4be171a46c1e1821e408b073ff5791c587a9
most of the time, we don't need to actually move the
data.
Compression is randomly slightly different, because HistogramCompactBins() changed.
Timing is about the same.
Change-Id: Ia6af8e9780581014d6860f2b546189ac817cfad1
we don't need to store the whole distribution in order to compute the alpha
Later, we can incorporate the max_value / last_non_zero bookkeeping
in SSE2 directly.
Change-Id: I748ccea4ac17965d7afcab91845ef01be3aa3e15
this is a first step to unifying encoding/decoding cache stride
and possibly sharing the prediction functions in dsp/
With this layout, there's a little (~7%) space lost with unused samples.
But no speed change was observed.
Change-Id: I016df8cad41bde5088df3579e6ad65d884ee711e
Move all the Entropy evaluation methods to lossless.c (from histogram.c).
There's slight difference in the way entropy is computed for evaluating
entropy in prediction methods and histogram (literal) for huffman trees.
Plan (later) to merge few (static) methods and reduce the code size.
This change has no impact on the compression speed/density.
Change-Id: Ife3d96a3c4a8d78a91723d9e0a8d1b78c0256a15
Remove handling for WEBP_HINT_GRAPH w.r.t use_palette flag.
The WEBP_HINT_GRAPH is now used at one place, to set the initial size of the
Bit Writer as bpp for photo images are generally larger than the graphical
images.
Change-Id: I1b9c4436c85a8f69da74c0dbcd292397323f2696
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
