Commit Graph

32 Commits

Author SHA1 Message Date
James Zern
93e0ce27f4 lossless_neon: harmonize function suffixes
BUG=webp:355

Change-Id: I4210081a39800b5c2589c443da237269908af666
2017-10-20 19:00:53 -07:00
James Zern
a439972175 WIP: list includes as descendants of the project dir
#include "(.|..)/..." -> #include "src/..."

Change-Id: I772880aa097a770722043c8a4393552ba38a89b6
2017-10-10 23:04:05 -07:00
skal
622242aaba Lossess dec: harmonize the function suffixes
BUG=webp:355

Change-Id: I445d64df6aa2e347f41e7af306be12a77e2ac6a5
2017-08-07 18:22:41 -07:00
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
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
58a1f124c2 ~2% faster predictor #12 in NEON.
Change-Id: I6772bb865d0f72720a65561eb55028e538df236d
2016-12-06 10:24:27 +01:00
Vincent Rabaud
d23abe4e9f Implement lossless transforms in NEON.
Change-Id: I2172b1a763eb9dfe25d2b9bf1fb6501d7e192e55
2016-12-03 11:20:22 +00:00
Vincent Rabaud
71e2f5cadf Remove memcpy in lossless decoding.
Change-Id: Iba694b306486d67764e2fc5576c98a974c9b886c
2016-11-24 17:45:24 +01:00
James Zern
a53c336919 lossless_neon: add VP8LTransformColorInverse
based on SSE2, only ~11% faster

Change-Id: I45434639d81e153f01f77c1f5d2da510b542170e
2015-08-04 23:22:36 -07:00
James Zern
2a010f992a lossless_neon: remove predictors 5-13
operating on single uint32's isn't helped by NEON.
this improves aarch64 performance by ~4%

Change-Id: I9fb25a8962de7b80e893e756ee7c76393cfd40c7
2015-07-28 19:44:58 -07:00
James Zern
f27f773576 lossless_neon: enable VP8LAddGreenToBlueAndRed
this moves the function outside the WEBP_USE_INTRINSICS check.
there's no alternative version and it's ~70% faster at the
function level and 1-2% faster overall

Change-Id: I59fb4918ec86b1ac3a47cbd5d05ce62f007461cb
2015-07-01 22:50:54 -07:00
James Zern
b44eda3f60 dsp: add DSP_INIT_STUB
generates a stub function when the specific architecture is not enabled,
exposing a symbol in the module, avoiding a compiler warning

Change-Id: Ia9336e57466a9b5241b85c1c95838e91c9283147
2015-04-02 23:55:35 -07:00
James Zern
553051f741 dsp/lossless: split enc/dec functions
adds lossless_enc*.c; reduces the size of the decode-only so: ~78K
w/gcc-4.8.2 on x86_64.

Change-Id: If5e4610b67d05eba5896bc64bab79e9df92b2092
2015-03-23 22:57:50 -07:00
James Zern
1d93ddec19 dsp/lossless*.c: rework WEBP_USE_<arch> ifdef
add a dummy init rather than repeating the '#ifdef WEBP_USE_...'
pattern.

Change-Id: If8b4459556e6bfaa36ef046f66520558b9444fc2
2015-03-20 19:19:46 -07:00
James Zern
602a00f93f fix iOS arm64 build with Xcode 6.3
the standard vtbl functions are available there [1][2].
based on a patch from: aaroncrespo
fixes issue #243.

[1]
http://adcdownload.apple.com//Developer_Tools/Xcode_6.3_beta/Xcode_6.3_beta_Release_Notes.pdf
[2] Apple LLVM Compiler Version 6.1
- Xcode 6.3 updates the Apple LLVM compiler to version 6.1.0.
[...]
Support for the arm64 architecture has been significantly revised to
align with ARM's implementation, where the most visible impact is that a
few of the vector intrinsics have changed to match ARM's specifications.

Change-Id: I79a0016f44b9dbe36d0373f7f00a50ab3c2ca447
2015-02-19 12:16:58 -08:00
James Zern
b969f5dfac dsp: normalize WEBP_TSAN_IGNORE_FUNCTION usage
the attribute is only necessary in one location; remove it from the
prototypes.

Change-Id: I3820a3c34fbb029fd7ac69a1b0a9b76091bdbde2
2015-02-13 15:23:40 -08:00
James Zern
416e1cea9b lossless_neon: enable subtract green for aarch64
similar to:
1ba61b0 enable NEON intrinsics in aarch64 builds

vtbl1_u8 is available everywhere but Xcode-based iOS arm64 builds, use
vtbl1q_u8 there.

performance varies based on the input, 1-3% on encode was observed

Change-Id: Ifec35b37eb856acfcf69ed7f16fa078cd40b7034
2015-01-31 11:32:05 -08:00
James Zern
f8740f0d6c dsp: s/USE_INTRINSICS/WEBP_USE_INTRINSICS/
for consistency with other defines shared across modules

Change-Id: I30cdb9f892e9ea48265883f560500ffb1d6799ee
2015-01-12 14:27:36 -08:00
James Zern
a4c3a31b8f WEBP_TSAN_IGNORE_FUNCTION: fix gcc compat warning
move the attribute to the front of the function to quiet clang warning:
GCC does not allow no_sanitize_thread attribute in this position on a
function definition

Change-Id: Ie4cc6e35a07bd00eab67d9cd6801bd2be9cfe676
2014-10-16 18:06:43 +02:00
Pascal Massimino
80247291c6 mark some init function as being safe for thread_sanitizer.
introduces the macro WEBP_TSAN_IGNORE_FUNCTION

Change-Id: I3de2b6c1a2076fba4da7ae50322551e026b2082b
2014-10-16 16:34:07 +02:00
James Zern
bc03670f01 neon: add INIT_VECTOR4
used to initialize NxMx4 vector types
replaces initialization via '{{ }}' gnu-ism.

Change-Id: I0da7b3d321f3d48579b7863fb2e4d3f449ae7f5e
2014-07-01 00:18:23 -07:00
James Zern
bf0e003067 lossless_neon: use vcreate_*() where appropriate
this is more portable than {} initialization.
more involved cases are left for a follow-up.

Change-Id: If7e111864f287ea0a5de6311454aeda37afbb52a
2014-05-27 16:27:46 -07:00
James Zern
1ba61b09f9 enable NEON intrinsics in aarch64 builds
avoids functions that use vtbl? as in iOS builds these are marked
unavailable

Change-Id: I17aedc3c7dc8f1d5be0941205de0b22c3772ef1b
2014-05-03 12:37:42 -07:00
James Zern
b9d2bb67d6 dsp/neon.h: coalesce intrinsics-related defines
Change-Id: Ifadd41a5bbf7f99eeb6d75d2b67daa25e0544946
2014-05-03 11:34:07 -07:00
James Zern
f937e01261 move LOCAL_GCC_VERSION def to dsp.h
+ add LOCAL_GCC_PREREQ and use it in lossless_neon.c

Change-Id: Ic9fd99540bc3e19c027d1598e4530dfdc9b9de00
2014-04-26 19:09:04 -07:00
skal
41c6efbdc5 fix lossless_neon.c
* some extra {xx , 0 } in initializers
* replaced by vget_lane_u32() where appropriate

Change-Id: Iabcd8ec34d7c853920491fb147a10d4472280a36
2014-04-14 14:27:11 +02:00
skal
3d49871dbe NEON functions for lossless coding
Verified OK, but right now they don't seem faster.
So they are disabled behind a USE_INTRINSICS flag (off for now)

Change-Id: I72a1c4fa3798f98c1e034f7ca781914c36d3392c
2014-04-10 15:32:08 +02:00
skal
abe6f48709 fix the gcc-4.6.0 bug by implementing alternative method
previous functions are a bit faster with gcc-4.8, so we keep them
for now.

Change-Id: I4081e5af66fbf606295d8a83875c1b889729b4dc
2014-04-09 07:53:55 +02:00
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
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