dany/libwebp
1
0
Fork 0
mirror of https://github.com/webmproject/libwebp.git synced 2024-12-29 15:08:20 +01:00
Code Issues Actions Packages Projects Releases Wiki Activity
3,517 Commits 35 Branches 66 Tags 15 MiB
693bf74ec0
Commit Graph

718 Commits

Author SHA1 Message Date
Pascal Massimino
693bf74ec0 move the SSIM calculation code in ssim.c / ssim_sse2.c 
Change-Id: I63a63fa7f44f257f2e17e45358b206c23069c448
 2017-02-21 12:53:35 +01:00
Pascal Massimino
4105d565d3 disable WEBP_USE_XXX optimisations when EMSCRIPTEN is defined 
Currently, none are available. If WEBP_HAVE_SSE2 eventually works,
we'll have to refine this conditionals.

BUG=webp:261

Change-Id: Ibc63ee1c013f2a4169eeb85cc8b6317b6420c2ad
 2017-02-08 15:44:20 +00:00
Parag Salasakar
aa893914fc Add clang build fix for MSA 
Change-Id: If139f4ecbdce756c69ba4ae032a70f81179683f8
 2017-02-01 17:45:17 +05:30
Pascal Massimino
4f3e3bbd44 disable GradientUnfilter_NEON 
Compile with XCode, it appears quite slower than the C-version,
especially for arm64.

Change-Id: Ic46dba184a36be454fef674129d2f909003788fc
 2017-01-25 16:33:26 -08:00
Pascal Massimino
79bf46f120 rename the pretentious SmartYUV into SharpYUV 
Change-Id: Ifeeb9cb85896c5f3ba0cc1c2c821f8d00295f69e
 2017-01-20 14:36:21 +01:00
James Zern
668e1dd44f src/{dec,enc,utils}: give filenames a unique suffix 
this avoids duplicates between these trees and dsp/, e.g., enc/tree.c,
dec/tree.c, making pulling the whole library source tree into one target
possible

BUG=webp:279

Change-Id: I060a614833c7c24ddd37bf641702ae6a5eef1775
 2017-01-19 19:09:48 -08:00
Pascal Massimino
71c53f1aeb NEON: speed-up strong filtering 
The sub-expression trick removes two constants and
two vmlal_s8 instructions.

Change-Id: I200022573b4880871b528b13a11a8f3d95def113
 2017-01-19 20:46:48 +00:00
Pascal Massimino
749a45a520 Merge "NEON: implement alpha-filters (horizontal/vertical/gradient)" 2017-01-17 15:13:08 +00:00
Pascal Massimino
74c053b57d Merge "NEON: fix overflow in SSE NxN calculation" 2017-01-17 15:10:54 +00:00
Pascal Massimino
1de931c669 NEON: implement alpha-filters (horizontal/vertical/gradient) 
gradient-filter code is not much faster, but maybe improvable in the future.

Change-Id: Ia16070e409fe8703b02276166f19526917df6b35
 2017-01-17 15:44:46 +01:00
Pascal Massimino
9b3aca404d NEON: fix overflow in SSE NxN calculation 
vmlal_u8() is prone to overflow during the accumulation.
There was a mismatch happening at low q mostly. Because in this
case the distortion is important and the accumulated sum was
later than 16bit-unsigned.

Change-Id: I1a08a2f744bcdf0b26647e61b9ee92a0c2e28fe8
 2017-01-17 11:47:36 +01:00
Pascal Massimino
1c07a3c639 dsp: WebPExtractGreen function for alpha decompression 
+ NEON implementation

Change-Id: I67204f99d6e4c5974718bdf21dad30381978f72c
 2017-01-17 09:33:25 +00:00
Pascal Massimino
8fda56126e Merge "add a kSlowSSSE3 feature for CPUInfo" 2017-01-13 07:01:48 +00:00
Pascal Massimino
86bbd24552 add a kSlowSSSE3 feature for CPUInfo 
This is meant to be used for run-time detection of slow platforms
regarding instructions like pshufb and bsr.

Adapted from libvpx patch: https://chromium-review.googlesource.com/#/c/367731

Change-Id: I2c22fbb9aae699d87a041393ba1ad5f1f21ff640
 2017-01-13 06:19:27 +00:00
Vincent Rabaud
7c2779e95a Get code to fully compile in C++. 
Change-Id: I6d8490c8c9b955d90dcc89ee8a9cf29ca0f93b08
 2017-01-12 18:03:55 +01:00
Vincent Rabaud
250c358662 Merge "When compiling as C++, avoid narrowing warnings." 2017-01-12 13:00:56 +00:00
Vincent Rabaud
c0648ac2ae When compiling as C++, avoid narrowing warnings. 
The gcc compilation warning was: narrowing conversion from ‘int’ to ‘int8_t’

Change-Id: I4803dd60ad04060cdb5d61a1aa98b25215b9d4eb
 2017-01-12 13:39:22 +01:00
Pascal Massimino
0d55f60c91 40% faster ApplyAlphaMultiply_SSE2 
process four pixels at a time

Change-Id: I1dee7f70772be4915654fc6638ef4729a1a239d4
 2017-01-12 02:33:09 -08:00
Pascal Massimino
49d0280df1 NEON: implement several alpha-processing functions 
- ApplyAlphaMultiply
 - DispatchAlpha
 - DispatchAlphaToGreen
 - ExtractAlpha

Decoding to Argb / rgbA / ... is 10-15% faster (measured on N4)

new file: alpha_processing_neon.c

Change-Id: I40f1a809e9885d1031ff0bc886d8d001efa66bca
 2017-01-11 17:39:29 +01:00
Pascal Massimino
48b1e85fbe SSE2: 15% faster alpha-processing functions 
ApplyAlphaMultiply / MultARGBRow / MultRow

we use now: x/255 = (x * 0x8081) >> (16 + 7)
and x/255 + .5 = ((x + 128) * 0x0101) >> 16

Change-Id: I8931091316ffc8bbf65aa3402f2e7d2b800e1971
 2017-01-11 15:35:16 +01:00
Pascal Massimino
28fe054e73 SSE2: 30% faster ApplyAlphaMultiply() 
and 15% faster MultARGBRow()

by switching to formulae:
    X / 255 = (X + 1 + (X >> 8)) >> 8 for any 16bit value X.
   (X / 255 + .5) = (XX + (XX >> 8)) >> 8, with XX = X + 128

Change-Id: Ia4a7408aee74d7f61b58f5dff304d05546c04e81
 2017-01-10 23:34:22 +01:00
Pascal Massimino
be0ef6395f fix a comment typo 
Change-Id: I0fabd08cd8abd3cea7ddfd2e498507adb0d3c67e
 2017-01-10 21:17:13 +01:00
Pascal Massimino
00b08c88c0 Merge "NEON: 5% faster conversion to RGB565 and RGBA4444" 2016-12-22 08:39:01 +00:00
Pascal Massimino
0e7f444702 Merge "NEON: faster fancy upsampling" 2016-12-21 14:53:24 +00:00
Pascal Massimino
b016cb91c5 NEON: faster fancy upsampling 
2-3% faster decoding overall

Change-Id: I2c53e50dc7e0ade5245cff8cc5d7b96a14062955
 2016-12-21 15:23:54 +01:00
Vincent Rabaud
1cb638010c Call the C function to finish off lossless SSE loops only when necessary. 
Change-Id: I4e221d80879dc9c90c24d69a40bc5811d73787ad
 2016-12-21 14:25:54 +01:00
Vincent Rabaud
875fafc191 Implement BundleColorMap in SSE2. 
Change-Id: I44cd23647bd0a49330b6b2b3ed08050a5500e58e
 2016-12-21 10:44:31 +01:00
Pascal Massimino
341d711c43 NEON: 5% faster conversion to RGB565 and RGBA4444 
We use the magic 'shift and insert' instruction instead of
the multiple shifts and or's.

Change-Id: I48df0320668b502a91792defc0423a9441669d19
 2016-12-20 17:01:48 +01:00
Pascal Massimino
a4bbe4b38b fix indentation 
Change-Id: I5593fb2441f253c6b8cc43949c11909f19184b55
 2016-12-13 22:50:29 -08:00
Pascal Massimino
58fc507842 Merge "PredictorSub: implement fully-SSE2 version" 2016-12-13 11:03:13 +00:00
Pascal Massimino
9cc421675b PredictorSub: implement fully-SSE2 version 
and inline the C-version too.

Predictor #13 is still a hard one.

Change-Id: Iedecfb5cbf216da4e28ccfdd0810286133f42331
 2016-12-13 02:19:35 -08:00
James Zern
2423017a28 dsp/lossless.c,cosmetics: fix indent 
after:
fbba5bc optimize predictor #1 in plain-C For some reason, gcc has hard
time inlining this one...

Change-Id: I2e2416593acd4c9d14958d8757bfd284d999100b
 2016-12-12 12:53:23 -08:00
Pascal Massimino
fbba5bc2c1 optimize predictor #1 in plain-C 
For some reason, gcc has hard time inlining this one...

Also optimize predictor #0 and #1 for encoding, so we don't have to
call the generic pointers VP8LPredictors[...]

Change-Id: I1ff31e3b83874b53f84fe23487f644619fd61db9
 2016-12-12 17:41:36 +01:00
Pascal Massimino
9ae0b3f65a Merge "SSE2: slightly (~2%) faster Predictor #1" 2016-12-12 14:46:21 +00:00
Pascal Massimino
c1f97bd758 SSE2: slightly (~2%) faster Predictor #1 
by removing a load from memory

Change-Id: If6c4aa7fb99309d09f943393ec772891449971f0
 2016-12-12 02:24:38 -08:00
Pascal Massimino
ea664b8995 SSE2: 10% faster Predictor #11 
Change-Id: I14ae5f6603071b86dfdbe8e6f7dfdbe5d8510185
 2016-12-12 02:20:41 -08:00
Pascal Massimino
b3fb8bb602 slightly faster Predictor #11 in NEON 
(+some slight modifications on Predictor #12)

Change-Id: Ic2132dcd83d961cd069fa01ca1670e35e35274e2
 2016-12-08 07:32:51 -08:00
Pascal Massimino
76ebbfff28 NEON: implement predictor #13 
~5-7% faster

Change-Id: I3361b0bbc978f3721168db15778a67337309c18a
 2016-12-07 14:58:49 -08:00
Vincent Rabaud
95b12a08ae Merge "Revert Average3 and Average4" 2016-12-07 15:38:56 +00:00
Vincent Rabaud
54ab2e758f Revert Average3 and Average4 
Average3 created a slowdown of 1-2% in lossless decoding.
Average4 created a slowdown of 2-3% in lossless decoding.

Change-Id: Ic2e62cdd83fc897887ec2bf41ea7cadbada84fe5
 2016-12-07 15:32:33 +01:00
Pascal Massimino
fe12330c81 3-5% faster Predictor #5, #6, #7 and #10 for NEON 
Change-Id: Ica48c7088d4384f0888dd171a47e68ebd25729b2
 2016-12-07 15:25:33 +01:00
Pascal Massimino
fbfb3bef7b ~2% faster predictor #10 for NEON 
Change-Id: Icd9cff90c227d702c3ba319131996c5475094520
 2016-12-06 13:47:35 +00:00
Pascal Massimino
d4b7d801db lossless_sse2: use the local functions 
...instead of the pointers stored in the array.
Should be faster (inlined) and safer.

Also: suffix explicitly the functions with _SSE2

Change-Id: Ie7de4b8876caea15067fdbe44abfedd72b299a90
 2016-12-06 14:20:41 +01:00
Vincent Rabaud
a5e3b22574 Lossless decoder SSE2 improvements. 
Change-Id: Ia901014ac63156a2e278b81e035256c30bdf8706
 2016-12-06 13:45:09 +01:00
Pascal Massimino
58a1f124c2 ~2% faster predictor #12 in NEON. 
Change-Id: I6772bb865d0f72720a65561eb55028e538df236d
 2016-12-06 10:24:27 +01:00
Pascal Massimino
906c3b6392 Merge "Implement lossless transforms in NEON." 2016-12-03 16:55:14 +00:00
Vincent Rabaud
d23abe4e9f Implement lossless transforms in NEON. 
Change-Id: I2172b1a763eb9dfe25d2b9bf1fb6501d7e192e55
 2016-12-03 11:20:22 +00:00
Vincent Rabaud
2e6cb6f34e Give more flexibility to the predictor generating macro. 
Change-Id: Ia651afa8322cb5c5ae87128340d05245c0f6a900
 2016-12-02 12:33:12 -08:00
Vincent Rabaud
28e0bb7088 Merge "Fix race condition in multi-threading initialization." 2016-12-02 17:45:10 +00:00
Vincent Rabaud
647045305a Fix race condition in multi-threading initialization. 
Before, a first thread could enter VP8LDspInitSSE2, set
VP8LPredictorsAdd to an SSE2 version BEFORE another thread
would do the memcpy from VP8LPredictorsAdd to VP8LPredictorsAdd_C
thus leading to a C version actually being the SSE2 one (which
would then create an infinite recursion in the SSE2 predictors
at execution).

Change-Id: I224f4ceab31d38f77a1375a7e2636a6014080e3a
 2016-12-02 18:28:57 +01:00