Commit Graph

907 Commits

Author SHA1 Message Date
skal
24cc307ae3 ~20% faster lossless decoding
We use help from small LUTs for Huffman decoding.

Change-Id: I58db4ecd37282822094519f2aa14cf544beba975
2013-06-20 09:41:29 +02:00
skal
313d853fa9 Speedup for decoding lossless WebP photographs:
use bit_pos instead num_bit to update the bit-reader
37.69 MB/s -> 39.79 MB/s, 5.6 %

Change-Id: Ica476cc3e2fc2db084d6961bea8586b050bb221d
2013-06-18 17:26:52 +02:00
skal
24ee098a00 change the bytes_per_pixels_ field into more evocative use_8b_decode
+ split AllocateInternalBuffers() into two 32b/8b variants instead of
trying to do everything in one function.

Change-Id: I35cac9fcd990a2194c95da4b2a4046ca3a514343
2013-06-18 17:24:14 +02:00
skal
3307c16327 Don't set alpha-channel to 0xff for alpha->green uplift
usually saves ~4 bytes on average (but, up to 10 -or even 16- sometimes).

Change-Id: Ib500e1a35471a2f3da453ffc8c7e95d28b8d34fe
2013-06-18 00:03:32 +02:00
Urvang Joshi
a37eff47d6 Regression fix for alpha channels using color cache:
Considering the fact that insert to/lookup from the color cache is always 32
bit, use DecodeImageData() variant in that case.

Change-Id: I6c665a6cfbd9bd10651c1e82fa54e687cbd54a2b
2013-06-17 03:49:30 -07:00
pascal massimino
241cf99bc0 Merge "muxedit: silence some uninitialized warnings" 2013-06-16 23:23:20 -07:00
Urvang Joshi
c8f9c84d5e Regression fix for alpha unfiltering:
For vertical filter, correctly initialize 'preds' when not starting at
row 0.

Change-Id: I69697d73dde8a2f701f376ff3986d09fa6f7fee3
2013-06-16 02:00:44 -07:00
James Zern
14cd5c6c40 muxedit: silence some uninitialized warnings
src/mux/muxedit.c:490: warning: 'x_offset' may be used uninitialized in this function
src/mux/muxedit.c:490: warning: 'y_offset' may be used uninitialized in this function

Change-Id: I4fd27f717e59a556354d0560b633d0edafe7a4d8
2013-06-15 14:30:22 -07:00
James Zern
a368db81c6 dec/vp8l: quiet vs9 x64 type conversion warning
src\dec\vp8l.c(816) : warning C4244: '=' : conversion from '__int64' to
'int', possible loss of data
src\dec\vp8l.c(817) : warning C4244: '=' : conversion from '__int64' to
'int', possible loss of data

Change-Id: I1d376d5dea909395bff8741aba16e8eed83a6e8f
2013-06-14 19:24:54 -07:00
Urvang Joshi
8cf0701eb0 Alpha encoding: never filter in case of NO_COMPRESSION
This is because, filtering would never affect compression density for
this case.

Change-Id: I4bb14d3eb7da0a3805fda140eb1dfbf9ccc134f5
2013-06-14 13:34:25 -07:00
skal
86daf77c47 store top Y/U/V samples in packed fashion
More cache-line friendly than storing them split.

Change-Id: Ifb23cc3518ff1b5c37afe007558d4278868d75ea
2013-06-13 06:01:27 +02:00
Urvang Joshi
068db59e26 Intertwined decoding of alpha and RGB
This will reduce the time to first decoded pixel.

Change-Id: I07b900c0ed4af3aac806b2731e11cd18ec16d016
2013-06-12 14:10:59 -07:00
Urvang Joshi
3fa595a571 Support decoding upto given row in DECODE_DATA_FUNC
This is in preparation for supporting Intertwined decoding of RGB and
alpha.

Change-Id: Ie1905d1212f4a00718a64dca020be3f4b1b870ec
2013-06-11 16:00:58 -07:00
Urvang Joshi
520f005f85 DequantizeLevels(): Add 'row' and 'num_rows' args
Change-Id: I0032e25c30f34362105d942f2dbb9ac75200f911
2013-06-11 15:59:21 -07:00
Urvang Joshi
47374b8273 Alpha unfilter for given set of rows
Support reconstruction of small number of rows at a time.

Change-Id: Ief1bc78c7ad011ec6df856551f3beb5f907fd8e0
2013-06-11 15:57:42 -07:00
Vikas Arora
bf867bf296 Tuned cross_color parameter (step) for lower qual
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
2013-06-11 12:15:07 -07:00
James Zern
d640614d54 update copyright text
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
2013-06-06 23:09:14 -07:00
skal
ea63d61937 fix a type warning on VS9 x86
"warning C4244: 'function' : conversion from 'uint64_t' to 'size_t', possible loss of data"

Change-Id: Ibd9f6a24993518d658d08127d616a17d7b99e0e4
2013-06-05 10:14:04 +02:00
skal
f4710e3b89 collect macroblock reconstruction data in VP8MBData struct
This is to better separate bitstream parsing from reconstruction.

Change-Id: I872b58e9940c4b14f72ebee50fba545468ff754c
2013-05-31 22:38:12 +02:00
skal
23d28e216d add doc precision for WebPPictureCopy() and WebPPictureView()
output picture object is overwritten, not free'd or destroyed.

Change-Id: Ibb47ab444063e7ad90ff3d296260807ffe7ddbf9
2013-05-31 10:41:19 +02:00
skal
af358e68ed Merge "remove datatype qualifier for vmnv" 2013-05-23 06:12:06 -07:00
skal
3fe91635df remove datatype qualifier for vmnv
this fix is for clang (LLVM v4.2). gcc was fine.

Change-Id: Id4076cda84813f6f9548a01775b094cff22b4be9
2013-05-23 13:52:24 +02:00
skal
91413be2f9 reduce memory for VP8MB and remove bitfields use
~1% faster on ARMv7

Change-Id: I3e3524e0c25ebd31a04269aae0d304932f1a781a
2013-05-21 23:20:04 +02:00
Vikas Arora
7413394e7f Fix the memory leak in ApplyFilters.
Change-Id: Iba1b1adf3088ea9c43e4f602a93e77450f6c6170
2013-05-21 14:00:26 -07:00
skal
2053c2cff2 simplify the alpha-filter testing loop
Change-Id: Iacebae749c37edc87a3c94c76cd589a2565ee642
2013-05-21 10:29:27 +02:00
skal
0d25876bad use uint8_t for inv_palette[]
Change-Id: I5005ce68d89bfb657d46ad8acc4368c29fa0c4fd
2013-05-18 17:13:58 +02:00
skal
afa3450c11 Fix the bug in ApplyPalette.
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
2013-05-18 17:03:18 +02:00
pascal massimino
2d6ac422cf Merge "webp/lossless: fix big endian BGRA output" 2013-05-17 00:36:15 -07:00
James Zern
2ca83968ae webp/lossless: fix big endian BGRA output
Change-Id: I3d4b3d21f561cb526dbe7697a31ea847d3e8b2c1
2013-05-17 00:32:01 -07:00
Vikas Arora
742110ccce Speed up ApplyPalette for ARGB pixels.
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
2013-05-16 15:44:21 -07:00
skal
2451e47dca misc code cleanup
* remove dec->skip_
* fix some naming

(no speed diff observed)

Change-Id: I12545ef79d29dd6f893c344d8fb171b0a8c7cc46
2013-05-15 20:03:15 +02:00
pascal massimino
d5f9b8f383 Merge "libwebp: fix vp8 encoder mem alloc offsetting" 2013-05-15 00:41:14 -07:00
James Zern
d8edd83551 libwebp: fix vp8 encoder mem alloc offsetting
'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
2013-05-15 00:31:23 -07:00
skal
8983b83ee1 remove use of bit-fields in VP8FInfo
(in favor of just plain uint8_t's)

Change-Id: I6187587a4d8a9f5c304a132d98ec42ce24fd244a
2013-05-15 09:21:30 +02:00
skal
87a4fca25f remove some warnings:
* "declaration of ‘index’ shadows a global declaration [-Wshadow]"
* "signed and unsigned type in conditional expression [-Wsign-compare]"

Change-Id: I891182d919b18b6c84048486e0385027bd93b57d
2013-05-14 22:28:32 +02:00
Pascal Massimino
ba8f74e229 Merge "fix for big-endian" 2013-05-14 01:58:06 -07:00
pascal massimino
a65067fa77 Merge "Further reduce memory to decode lossy+alpha images" 2013-05-14 01:56:54 -07:00
Urvang Joshi
64c844863a Further reduce memory to decode lossy+alpha images
Earlier such images were using roughly 9 * width * height bytes for
decoding. Now, they take 6 * width * height memory.

Change-Id: Ie4a681ca5074d96d64f30b2597fafdca648dd8f7
2013-05-13 16:24:49 -07:00
Urvang Joshi
332130b9b3 Mux: make a few methods static
Change-Id: I8a8b0b403116c89933e84c93502a8230026f819e
2013-05-13 13:26:33 -07:00
Pascal Massimino
4437061735 fix for big-endian
(Issue #150: https://code.google.com/p/webp/issues/detail?id=150)

Change-Id: Iad46d375a8c5eabae37cde8f55b3e7448601f264
2013-05-13 10:19:29 -07:00
pascal massimino
5fbc734b90 Merge "GetFeatures: Detect invalid VP8X/VP8/VP8L data" 2013-05-11 00:51:39 -07:00
pascal massimino
d5060c873c Merge "mux.h: A comment fix + some consistency fixes" 2013-05-11 00:41:35 -07:00
Urvang Joshi
352d0dee99 GetFeatures: Detect invalid VP8X/VP8/VP8L data
This facilitates early error detection during decode/render.
Also, related refactoring.

Change-Id: Ia6c7cd91dec202a2a68dae2118f5981cf1eaa83d
2013-05-10 14:27:11 -07:00
Urvang Joshi
3ef79fefec Cosmetic: "width * height"
Change-Id: I567c0d95355160a9f6721f949b38e2b8b6270b7a
2013-05-10 13:39:58 -07:00
Urvang Joshi
5818cff770 mux.h: A comment fix + some consistency fixes
Change-Id: I0aee1090322bac3ae3dabf9a48661cbb6de3ca52
2013-05-10 13:35:16 -07:00
Vikas Arora
8eae188a62 WebP-Lossless encoding improvements.
Lossy (with Alpha) image compression gets 2.3X speedup.
Compressing lossless images is 20%-40% faster now.

Change-Id: I41f0225838b48ae5c60b1effd1b0de72fecb3ae6
2013-05-08 17:22:11 -07:00
pascal massimino
4cb234d5b5 Merge "Mux: make ValidateForSingleImage() method static" 2013-05-08 01:54:28 -07:00
pascal massimino
ed6f53086b Merge "Add GetCanvasSize() method to mux" 2013-05-08 01:53:22 -07:00
Urvang Joshi
1d530c9a7e Mux: make ValidateForSingleImage() method static
Change-Id: I96ac5e3be26b8e8ecd9f055501a5feb7710bc324
2013-05-07 12:57:51 -07:00
Urvang Joshi
fffefd18c3 Add GetCanvasSize() method to mux
Change-Id: If910f5024f4c301a92e6c2e8ee9c315a103c5df7
2013-05-07 12:47:48 -07:00