Commit Graph

57 Commits

Author SHA1 Message Date
Scott LaVarnway
98165ec074 Neon version of vp9_sub_pixel_variance8x8(),
vp9_variance8x8(), and vp9_get8x8var().

On a Nexus 7, vpxenc (in realtime mode, speed -12)
reported a performance improvement of ~1.2%.

Change-Id: I8a66ac2a0f550b407caa27816833bdc563395102
2014-08-01 11:35:55 -07:00
Frank Galligan
5487b6067c Merge "Neon version of vp9_sub_pixel_variance32x32()," 2014-08-01 09:46:37 -07:00
Scott LaVarnway
6f4b8dcdc2 Neon version of vp9_subtract_block()
On a Nexus 7, vpxenc (in realtime mode, speed -12)
reported a performance improvement of ~3.2%

Change-Id: I8862497264142171b7efc32df1a67714a23539f4
2014-07-31 09:28:06 -07:00
Scott LaVarnway
d39448e2d4 Neon version of vp9_sub_pixel_variance32x32(),
vp9_variance32x32(), and vp9_get32x32var().

Change-Id: I8137e2540e50984744da59ae3a41e94f8af4a548
2014-07-31 08:00:36 -07:00
Scott LaVarnway
d4a37db5b8 Neon version of vp9_quantize_fp()
On a Nexus 7, vpxenc (in realtime mode, speed -12)
reported a performance improvement of ~12.4%

Change-Id: Id29d215acf58bb108489e218a259adf74b4768d7
2014-07-30 09:33:46 -07:00
Scott LaVarnway
521cf7e879 Neon version of vp9_sub_pixel_variance16x16(),
vp9_variance16x16(), and vp9_get16x16var().

On a Nexus 7, vpxenc (in realtime mode, speed -12)
reported a performance improvement of ~16.7%.

Change-Id: Ib163aa99f56e680194aabe00dacdd7f0899a4ecb
2014-07-30 08:17:32 -07:00
Scott LaVarnway
d19d222db6 Added vp9_fdct8x8_neon(), vp9_fdct8x8_1_neon()
On a Nexus 7, vpxenc (in realtime mode, speed -12)
reported a performance improvement of ~3.7%.

Change-Id: I428c72c40df82c6d537955e320a8debf99343004
2014-07-29 08:56:05 -07:00
levytamar82
4ba92dc5ab Fix bug 805
Remove all the redundant dct functions (dct4x4, dct8x8)
in avx2 except dct32x32 those functions were copied originally from dct_sse2

Change-Id: I742576fbf5175f3ac09f2076976a9247b259323e
2014-07-28 15:46:01 -07:00
Scott LaVarnway
ba0652e83a Merge "Added vp9_sad64x64_neon(), vp9_sad32x32_neon()" 2014-07-17 11:42:16 -07:00
Scott LaVarnway
696fa52eaa Added vp9_sad64x64_neon(), vp9_sad32x32_neon()
and vp9_sad16x16_neon()

On a Nexus 7, vpxenc (in realtime mode, speed -6)
reported a performance improvement of ~17%.

Change-Id: I91e070cde2973451083d3f3d63b49b7886de9a85
2014-07-16 12:54:46 -07:00
Yunqing Wang
75cd57503d Refactor vp9_diamond_search_sad function
Currently, vp9_diamond_search_sadx4() is only called when sse3 is
enabled, which is improper since sse2 optimization of sdx4df
functions are available. Changed to always use
vp9_diamond_search_sadx4().

Change-Id: I4b95d6b7a3c6c645783c373f0ba8d645ece24717
2014-07-10 09:19:03 -07:00
Yunqing Wang
30117a576d Refactor refining_search_sad code
There are sse2 optimization of sdx4df functions. Instead of calling
vp9_refining_search_sadx4 only when sse3 is enabled, call it always.

Change-Id: I24f93818f7d4209d1425039e0eb099ff9ff08fe9
2014-07-09 16:50:11 -07:00
Jingning Han
9ad1b9fc67 Re-design quantization process for 32x32 transform block
This commit enables a new quantization process for 32x32 2D-DCT
transform coefficient blocks. It improves the compression
performance of speed 5 by 1.4%. The overall compression gains of
speed 5 due to the new quantization scheme is 4.7%. It also includes
the SSSE3 implementation of the 32x32 quantization process.

Change-Id: I0855b124fd6462418683f783f5bcb44255c9993b
2014-07-08 16:55:28 -07:00
Jingning Han
9ac2f66320 Re-design quantization process
This commit re-designs the quantization process for transform
coefficient blocks of size 4x4 to 16x16. It improves compression
performance for speed 7 by 3.85%. The SSSE3 version for the
new quantization process is included.

The average runtime of the 8x8 block quantization is reduced
from 285 cycles -> 255 cycles, i.e., over 10% faster.

Change-Id: I61278aa02efc70599b962d3314671db5b0446a50
2014-07-01 17:00:07 -07:00
James Zern
88df435d6b Merge "vp9_rtcd: correct avx2 references" 2014-06-16 17:39:13 -07:00
Jingning Han
d5ae43318e Merge "Fast computation path for forward transform and quantization" 2014-06-12 11:59:52 -07:00
Jingning Han
ccba289f8d Fast computation path for forward transform and quantization
This commit enables a fast path computational flow for forward
transformation. It checks the sse and variance of prediction
residuals and decides if the quantized coefficients are all
zero, dc only, or more. It then selects the corresponding coding
path in the forward transformation and quantization stage.

It is currently enabled in rtc coding mode. Will do it for rd
coding mode next.

In speed -6, the runtime for pedestrian_area 1080p at 1000 kbps
goes down from 14234 ms to 13704 ms, i.e., about 4% speed-up.
Overall coding performance for rtc set is changed by -0.18%.

Change-Id: I0452da1786d59bc8bcbe0a35fdae9f623d1d44e1
2014-06-12 11:10:54 -07:00
James Zern
9f3a0dbb5e vp9_rtcd: correct avx2 references
s/"\$avx2_x86inc"/"avx2"/

avx2 code is all intrinsics and as a result doesn't rely on x86inc.asm

Change-Id: I76ad39474d8a00658f3e43131830ef0f4f34772a
2014-06-10 16:26:36 -07:00
James Zern
520cb3f39f vp9_sub_pixel_*variance*: disable avx2 variants
tests failing under Win32/Win64

+ variance_test: add missing avx2 functions (partially disabled)

Change-Id: I6abc0657ea076379ab9ca65c12678b9ea199849d
2014-06-10 16:11:15 -07:00
James Zern
d3ff009d84 vp9_sad*x4d: disable avx2 variants
tests failing under Win32/Win64

+ sad_test: add missing avx2 functions (disabled)

Change-Id: I8224fba2b270f6039ab1877d71e1e512f0081856
2014-06-10 16:10:12 -07:00
James Zern
dd9f502933 vp9_f(dct|ht): disable avx2 variants
tests failing under Win32/Win64

+ dct16x16_test: add missing avx2 functions (partially disabled)

exercises the forward transforms
no idct/iht implementations, so the c-code is used

Change-Id: I04f64a457fa0828a00f32b5c9fe4f55294f21f61
2014-06-09 18:48:11 -07:00
James Zern
5704578f5f convolve: disable avx2 variants
tests failing under Win32/Win64

Change-Id: I5d49d11911bcda3a832b14efe5500d22597bedcf
2014-06-09 18:42:03 -07:00
Jingning Han
0c4a4225ec Merge "Enable SSSE3 inverse 2D-DCT with 10 non-zero coeffs" 2014-06-03 16:51:39 -07:00
Dmitry Kovalev
f7ff24cdd0 Reusing existing vp9_get{8x8, 16x16}var() instead of new ones.
Change-Id: I87b7c657d8813d7fb383ab519d150c0ffb1dd377
2014-05-29 11:14:06 -07:00
Jingning Han
6d21cbd20b Enable SSSE3 inverse 2D-DCT with 10 non-zero coeffs
This commit enables SSSE3 implementation of the inverse 2D-DCT
with only first 10 coefficients non-zero. It reduces the runtime
of SSE2 version from 745 cycles to 538 cycles, i.e., 27% speed-up.

Change-Id: I18ba4128859b09c704a6ee361d69a86c09fe8dfe
2014-05-28 10:53:33 -07:00
Yunqing Wang
1f2200080b Revert "Making vp9_get_sse_sum_{8x8, 16x16} static."
This reverts commit e8bbb3d9db.

Change-Id: Ie368d36fd249d323d859d208609c711f04537bbc
2014-05-27 13:37:08 -07:00
Deb Mukherjee
444f93945b Merge "Remove Wextra warnings from vp9_sad.c" 2014-05-27 11:54:05 -07:00
Jingning Han
48b0891370 Inverse 16x16 2D-DCT SSSE3 implementation
This commit enables the SSSE3 implementation of full inverse 16x16
2D-DCT. The unit runtime goes down from 1642 cycles to 1519 cycles,
about 7% speed-up.

Change-Id: I14d2fdf9da1fb4ed1e5db7ce24f77a1bfc8ea90d
2014-05-23 15:09:35 -07:00
Deb Mukherjee
916550428d Remove Wextra warnings from vp9_sad.c
As a side-effect, the sad unit tests for VP8 and VP9
had to be separated.

Change-Id: I068cc2391eed51e9b140ea6aba78338c5fec8d71
2014-05-22 22:21:16 -07:00
Deb Mukherjee
a185bc3350 Extends temporal filtering to work for 422 data
This is needed for profiles 1 and 2.

Change-Id: I5dd7644c2932d055ab89e050d4be7d4117cd1028
2014-05-20 15:19:40 -07:00
Jim Bankoski
ec82d2dfec Merge "Revert "Remove Wextra warnings from vp9_sad.c"" 2014-05-15 11:54:23 -07:00
Yunqing Wang
c661cf0dad Merge "AVX2 To VP9 Block Error Optimization" 2014-05-15 11:29:29 -07:00
Jim Bankoski
a16794dd31 Revert "Remove Wextra warnings from vp9_sad.c"
This reverts commit 7ab9a9587b

Nightly test http://build.webmproject.org/jenkins/view/libvpx-nightly-tests/job/libvpx%20unit%20tests%20(valgrind-2)/arch=x86_64-linux-gcc,filter=-*VP8*:*Large.*/276/console

Failed 

This patch did not address all the assembly issues 
some of the vp8 assembly counts on 5 arguments being passed in to this function:   

one example : vp8_sad8x16_wmt

Please address or split this into vp9 and vp8 patches.

Change-Id: I78afcc171649894f887bb8ee3c66de24aaddc7ca
2014-05-15 08:31:20 -07:00
levytamar82
1fbab853c8 AVX2 To VP9 Block Error Optimization
vp9_block_error_sse2 can only handle 16 bytes at a time but
the function requires to handle a sequence of 32 bytes at a time
so each 16 bytes is handled in a different register.
With AVX2 optimization the 32 bytes can be handled in one register instead
of two in the SSE2
The vp9_block_error was optimized by 85%.
The user level was optimized by 1.2%

Change-Id: Ia8fffe60e61eff7432a5fbd538757894f6c319fd
2014-05-14 11:51:07 -07:00
Deb Mukherjee
7ab9a9587b Remove Wextra warnings from vp9_sad.c
As a side-effect, the max_sad check is removed from the
C-implementation of VP8, for consistency with VP9, and to
ensure that the SAD tests common to VP8/VP9 pass.
That will make the VP8 C implementation of sad a little slower
but given that is rarely used in practice, the impact will be
minimal.

Change-Id: I7f43089fdea047fbf1862e40c21e4715c30f07ca
2014-05-14 03:17:31 -07:00
Johann
ce23931a3f Only build neon assembly for armv7 targets
Allow selectively building just the intrinsics for armv8

Change-Id: I2f29b2e4508b8b8e5649c2906b3159ad1d4ec477
2014-05-12 08:52:02 -07:00
Alex Converse
ec8a3272fa Merge "Add an x86inc MMX fwht4x4." 2014-05-09 13:48:49 -07:00
Jingning Han
9412785b02 Merge changes I3edd4b95,I4514f974,Ie7fa4386
* changes:
  Turn on unit tests for SSSE3 8x8 forward and inverse 2D-DCT
  Change eob threshold for partial inverse 8x8 2D-DCT to 12
  SSSE3 8x8 inverse 2D-DCT with first 10 coeffs non-zero
2014-05-09 09:58:39 -07:00
Alex Converse
b5422fab46 Add an x86inc MMX fwht4x4.
Change-Id: Ib0a73d4863478f9b8a00976379d25d2f6ebbb197
2014-05-08 12:01:27 -07:00
Jingning Han
41a350a83d Change eob threshold for partial inverse 8x8 2D-DCT to 12
The scanning order has the first 12 coefficients of the 8x8 2D-DCT
sitting in the top left 4x4 block. Hence the partial inverse 8x8
2D-DCT allows to handle cases with eob below 12.

The overall runtime of the inverse 8x8 2D-DCT unit is reduced from
166 cycles (using SSE2) to 150 cycles (using SSSE3).

Change-Id: I4514f9748042809ac84df4c14382c00f313f1cd2
2014-05-08 09:48:58 -07:00
Jingning Han
9e7b09bc5d SSSE3 8x8 inverse 2D-DCT with first 10 coeffs non-zero
This commit enables ssse3 assembly implementation of the 8x8
inverse 2D-DCT with only first 10 coefficients non-zero. The
average runtime for this unit goes down from 198 cycles to 129
cycles (34.8% faster).

Change-Id: Ie7fa4386f6d3a2fe0d47a2eb26fc2a6bbc592ac7
2014-05-07 17:40:02 -07:00
Paul Wilkins
33b1c457ed Revert "Add an MMX fwht4x4"
Includes changes that are not compatible with VS windows builds.
Amongst other things stdint.h is not supported in VS.

This reverts commit 89fbf3de50.

Change-Id: Ifa86d7df250578d1ada9b539c9ff12ed0c523cdd
2014-05-07 12:53:27 +01:00
Alex Converse
75d05d5ed4 Merge "Add an MMX fwht4x4" 2014-05-06 11:12:27 -07:00
Jingning Han
d289deb04c Merge "SSSE3 implementation of full inverse 8x8 2D-DCT" 2014-05-06 09:17:22 -07:00
Dmitry Kovalev
e8bbb3d9db Making vp9_get_sse_sum_{8x8, 16x16} static.
Change-Id: Ifb7937c977308c682986f0ce9645a0807d2aa46a
2014-05-05 19:12:38 -07:00
Alex Converse
89fbf3de50 Add an MMX fwht4x4
7% faster encoding a desktop lossless at RT speed 4.

Change-Id: I41627f5b737752616b6512bb91a36ec45995bf64
2014-05-05 15:10:48 -07:00
Jingning Han
52ae97b6aa SSSE3 implementation of full inverse 8x8 2D-DCT
This commit enables SSSE3 version full inverse 8x8 2D-DCT and
reconstruction. It makes the runtime of vp9_idct8x8_64_add down
from 256 cycles (SSE2) to 246 cycles.

Change-Id: I0600feac894d6a443a3c9d18daf34156d4e225c3
2014-05-05 10:49:27 -07:00
Jingning Han
39761eb5d6 Merge "Enable SSSE3 implementation of 8x8 forward 2D-DCT" 2014-04-30 13:41:36 -07:00
Dmitry Kovalev
d2bc8816a1 Merge "Adding search_site_config struct." 2014-04-29 16:59:47 -07:00
Jingning Han
1eaa3a76dc Enable SSSE3 implementation of 8x8 forward 2D-DCT
Assembly implementation of ssse3 8x8 forward 2D-DCT. The current
version is turned on only for x86_64. The average unit runtime
goes from 157 cycles down to 136 cycles, i.e., about 12.8% faster.
This translates into about 1.5% speed-up for pedestrian_area 1080p
at speed 2.

Change-Id: I0f12435857e9425ed7ce12541344dfa16837f4f4
2014-04-29 15:49:18 -07:00