Commit Graph

2247 Commits

Author SHA1 Message Date
James Zern
de693f2502 lossless_neon: disable VP8LConvert* functions
due to breakage with NDK/gcc-4.6 builds

Change-Id: Id96258e710ee33e08a023354b3227f27da986620
2014-04-04 20:38:29 -07:00
skal
4143332b22 NEON intrinsics for encoding
* inverse transform is actually slower with intrinsics + gcc-4.6,
  so is left disabled for now.
  With gcc-4.8, it's a bit faster than inlined assembly.

* Sum of Square error function provide a 2-3% speed up
  There's enabled by default (since there's no inlined-asm equivalent)

Change-Id: I361b3f0497bc935da4cf5b35e330e379e71f498a
2014-04-04 15:02:56 -07:00
Djordje Pesut
0ca2914b23 MIPS: MIPS32r1: Add optimization for ITransform
Change-Id: Ie4c8b9bc3a7826bd443cdebf05386786fafe8c56
2014-04-04 10:50:35 +02:00
James Zern
71bca5ecf3 dec_neon: use vst_lane instead of vget_lane
results in fewer instructions, small speed improvement

Change-Id: I98de632d09ff09f295368c0d744cb4397b585084
2014-04-03 14:56:26 -07:00
skal
bf06105293 Intrinsics NEON version of TransformOne
+ misc cosmetics

* seems 4% slower than inlined-asm with gcc-4.6
* is a tad faster (<1%) with gcc-4.8
(disabled for now)

Change-Id: Iea6cd00053a2e9c1b1ccfdad1378be26584f1095
2014-04-03 14:41:56 -07:00
pascal massimino
19c6f1ba74 Merge "dec_neon: use vld?_lane instead of vset?_lane" 2014-04-03 01:16:29 -07:00
James Zern
7a94c0cf75 upsampling_neon: drop NEON suffix from local functions
Change-Id: I6583ad74aacf78dcbeb5a0ff0218a39bc3460e5a
2014-04-02 23:24:39 -07:00
James Zern
d14669c83c upsampling_sse2: drop SSE2 suffix from local functions
Change-Id: I2349c1a8e5e15e1d204642096f84f3202721c297
2014-04-02 23:24:39 -07:00
James Zern
2ca42a4fb7 enc_sse2: drop SSE2 suffix from local functions
Change-Id: I5d61605a9d410761d50b689b046114f0ab3ba24e
2014-04-02 23:24:36 -07:00
James Zern
d038e6193b dec_sse2: drop SSE2 suffix from local functions
Change-Id: Ie171778b84038d5b04c5dc6972f6015caf555882
2014-04-02 23:10:39 -07:00
James Zern
fa52d7525f dec_neon: use vld?_lane instead of vset?_lane
results in fewer instructions, small speed improvement

Change-Id: I61ab48d09a5ce7c5158eac8244d28287457edc7a
2014-04-02 23:03:18 -07:00
Pascal Massimino
c520e77d94 cosmetic: fix long line
Change-Id: Id04b368aea5784a98c705f323b32d35b362742ea
2014-04-02 23:00:50 -07:00
James Zern
4b0f2dae6f Merge "add intrinsics NEON code for chroma strong-filtering" 2014-04-02 22:57:44 -07:00
skal
e351ec0759 add intrinsics NEON code for chroma strong-filtering
The nice trick is to pack 8 u + 8 v samples into a single uint8x16x_t
register, and re-use the previous (luma) functions

Change-Id: Idf50ed2d6b7137ea080d603062bc9e0c66d79f38
2014-04-03 06:58:21 +02:00
pascal massimino
aaf734b8b0 Merge "Add SSE2 version of forward cross-color transform" 2014-04-02 14:18:59 -07:00
Urvang Joshi
c90a902eff Add SSE2 version of forward cross-color transform
Encoding speed is roughly the same.

Change-Id: I6b976d0eb24e1847714e719762cb8403768da66c
2014-04-02 12:21:20 -07:00
Vikas Arora
bc374ff39e Use histogram_bits to initalize transform_bits.
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
2014-04-02 11:46:40 -07:00
James Zern
2132992d47 Merge "Add strong filtering intrinsics (inner and outer edges)" 2014-04-02 00:10:01 -07:00
skal
5fbff3a646 Add strong filtering intrinsics (inner and outer edges)
+ added some work-around gcc-4.6 to make it compile (except one function).
+ lots of revamping

All variants tested ok.
Speed-up is ~5-7%

Change-Id: I5ceda2ee5debfada090907fe3696889eb66269c3
2014-04-02 08:28:55 +02:00
Urvang Joshi
d4813f0cb2 Add SSE2 function for Inverse Cross-color Transform
Lossless decoding is now ~3% faster.

Change-Id: Idafb5c73e5cfb272cc3661d841f79971f9da0743
2014-04-01 15:52:25 -07:00
James Zern
26029568b7 dec_neon: add strong loopfilter intrinsics
vertical only currently, 2.5-3% faster
placed under USE_INTRINSICS as this change depends on the simple
loopfilter
improves the simple loopfilter slightly thanks to some reorganization

Change-Id: I6611441fa54228549b21ea74c013cb78d53c7155
2014-04-01 01:13:50 -07:00
James Zern
cca7d7ef0f Merge "add intrinsics version of SimpleHFilter16NEON()" 2014-04-01 00:57:11 -07:00
James Zern
1a05dfa7f5 windows: fix dll builds
WebPSafe* need to be marked external to allow mux/demux to access them
through libwebp.dll

Change-Id: Ib6620e00d376f7aa5a0550e1e244f759977f97a0
2014-03-31 17:46:12 -07:00
skal
d6c50d8ac2 Merge "add some colorspace conversion functions in NEON" 2014-03-31 13:15:18 -07:00
Urvang Joshi
4fd7c82e6a SSE2 variants of Subtract-Green: Rectify loop condition
When 4 pixels are left, they should be processed with SSE2.

Decoding is marginally faster (~0.4%).
Encoding speed: No observable difference.

Change-Id: I3cf21c07145a560ff795451e65e64faf148d5c3e
2014-03-31 10:51:45 -07:00
skal
97e5fac389 add some colorspace conversion functions in NEON
new file: lossless_neon.c
speedup is ~5%

gcc 4.6.3 seems to be doing some sub-optimal things here,
storing register on stack using 'vstmia' and such.
Looks similar to gcc.gnu.org/bugzilla/show_bug.cgi?id=51509

I've tried adding  -fno-split-wide-types and it does help
the generated assembly. But the overall speed gets worse with
this flag. We should only compile lossless_neon.c with it -> urk.

Change-Id: I2ccc0929f5ef9dfb0105960e65c0b79b5f18d3b0
2014-03-31 17:47:46 +02:00
skal
b9a7a45f1f add intrinsics version of SimpleHFilter16NEON()
It's disable for now, because it crashes gcc-4.6.3 during compilation
with -O2 or -O3. It's been tested OK with -O1.

Code is still globally disabled with USE_INTRINSICS, though.

Change-Id: I3ca6cf83f3b9545ad8909556f700758b3cefa61c
2014-03-31 16:31:31 +02:00
Pascal Massimino
daccbf400d add light filtering NEON intrinsics
disabled for now (but tested OK), thanks to the USE_INTRINSICS #define
We'll activate the code when we're on par with non-intrinsics

Change-Id: Idbfb9cb01f4c7c9f5131b270f8c11b70d0d485ff
2014-03-30 22:15:55 -07:00
Pascal Massimino
af44460880 fix typo in STORE_WHT
was working ok because dst == out

Change-Id: I27095129a11f468422250dd2b8fad8b3bd4e5bbd
2014-03-28 10:34:44 -07:00
Vikas Arora
6af6b8e1b6 Tune HistogramCombineBin for large images.
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
2014-03-28 07:09:59 -07:00
skal
af93bdd6bc use WebPSafe[CM]alloc/WebPSafeFree instead of [cm]alloc/free
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
2014-03-27 15:50:59 -07:00
James Zern
51f406a5d7 lossless_sse2: relocate VP8LDspInitSSE2 proto
this is in line with the other dsp files and silences a build warning.

Change-Id: I03ee3032c11d4c731cc10bfa0a2dcb6866756ba2
2014-03-27 15:07:43 -07:00
skal
0f4f721b12 separate SSE2 lossless functions into its own file
expose the predictor array as function pointers instead
of each individual sub-function

+ merged Average2() into ClampedAddSubtractHalf directly
+ unified the signature as "VP8LProcessBlueAndRedFunc"

no speed diff observed

Change-Id: Ic3c45dff11884a8330a9ad38c2c8e82491c6e044
2014-03-27 21:43:55 +01:00
skal
514fc251df VP8LConvertFromBGRA: use conversion function pointers
Change-Id: I863b97119d7487e4eef337e5df69e1ae2a911d4c
2014-03-27 09:00:35 +01:00
James Zern
6d2f35273d dsp/dec: TransformDCUV: use VP8TransformDC
rather than forcing the C version; this is similar to TransformUV

Change-Id: I2778194f05fca33e9b2b71323e92947c0b395e9a
2014-03-26 16:43:47 -07:00
skal
defc8e1b01 Merge "fix out-of-bound read during alpha-plane decoding" 2014-03-26 15:22:42 -07:00
James Zern
fbed36433d Merge "dsp: reuse wht transform from dec in encoder" 2014-03-26 15:13:07 -07:00
skal
d846708400 Merge "Add SSE2 version of ARGB -> BGR/RGB/... conversion functions" 2014-03-26 15:01:46 -07:00
skal
207d03b484 fix out-of-bound read during alpha-plane decoding
With -bypass_filter switched on, the lossless-compressed
data is decoded ahead of time (before being transformed and
display). Hence, the last row was called twice.

http://code.google.com/p/webp/issues/detail?id=193

Change-Id: I9e13f495f6bd6f75fa84c4a21911f14c402d4b10
2014-03-26 22:45:03 +01:00
skal
d1b33ad58b 2-5% faster trellis with clang/MacOS
(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
2014-03-26 22:33:01 +01:00
skal
369c26dd3f Add SSE2 version of ARGB -> BGR/RGB/... conversion functions
~4-6% faster lossless decoding

Change-Id: I3ed1131ff2b2a0217da315fac143cd0d58293361
2014-03-26 22:19:00 +01:00
James Zern
df230f2723 dsp: reuse wht transform from dec in encoder
Change-Id: Ide663db9eaecb7a37fe0e6ad4cd5f37de190c717
2014-03-22 13:25:08 -07:00
James Zern
80e218d43a Android.mk: fix build with APP_ABI=armeabi-v7a-hard
added in r9d; relax the check to build neon code

Change-Id: Ic52b3fbd3bf53617ee52b07a55b0ed05f6f9b26f
2014-03-20 23:24:39 -07:00
Pascal Massimino
59daf08362 Merge "cosmetics:" 2014-03-18 04:02:33 -07:00
Pascal Massimino
536220084c cosmetics:
- use VP8ScanUV, separate from VP8Scan[] (for luma)
 - fix indentation
 - few missing consts
 - change TrellisQuantizeBlock() signature

Change-Id: I94b437d791cbf887015772b5923feb83dd145530
2014-03-18 03:34:56 -07:00
James Zern
3e7f34a3fb AssignSegments: quiet array-bounds warning
nb (enc->segment_hdr_.num_segments_) will be in the range
[1, NUM_MB_SEGMENTS].

Change-Id: I5c2bd0bb82b17c99aff39c98b6b1747fc040dc16
2014-03-14 18:47:52 -07:00
James Zern
3c2ebf58a4 Merge "UpdateHistogramCost: avoid implicit double->float" 2014-03-14 15:50:57 -07:00
James Zern
cf821c821f UpdateHistogramCost: avoid implicit double->float
all the functions involved return double and later these locals are used
in double calculations. fixes a vs build warning

Change-Id: Idb547104ef00b48c71c124a774ef6f2ec5f30f14
2014-03-14 11:18:52 -07:00
Vikas Arora
312e638f30 Extend the search space for GetBestGreenRedToBlue
Get back some of the compression gains by extending the search space for
GetBestGreenRedToBlue. Also removed the SkipRepeatedPixels call, as it was not
helping much in yielding better compression density.

Before:
 1000 files, 63530337 pixels, 1 loops => 45.0s (45.0 ms/file/iterations)
 Compression (output/input): 2.463/3.268 bpp, Encode rate (raw data): 1.347 MP/s

After:
1000 files, 63530337 pixels, 1 loops => 45.9s (45.9 ms/file/iterations)
 Compression (output/input): 2.461/3.268 bpp, Encode rate (raw data): 1.321 MP/s

Change-Id: I044ba9d3f5bec088305e94a7c40c053ca237fd9d
2014-03-14 09:56:00 -07:00
Vikas Arora
1c58526fe1 Fix few nits
Add/remove few casts, fixed indentation.

Change-Id: Icd141694201843c04e476f09142ce4be6e502dff
2014-03-13 13:57:39 -07:00