Commit Graph

2039 Commits

Author SHA1 Message Date
Paul Wilkins
372336d1e5 Merge "Fix corrupt arf groups due to low "lag_in_frames"" 2017-08-16 18:25:29 +00:00
Linfeng Zhang
f95686895b Merge changes I08b562b6,Ia275940a,I51106e90
* changes:
  Add vpx_highbd_idct32x32_{34, 135, 1024}_add_{sse2, sse4_1}
  Update highbd idct x86 optimizations.
  Update 32x32 idct sse2 and ssse3 optimizations.
2017-08-16 16:36:37 +00:00
paulwilkins
48110d0f79 Fix corrupt arf groups due to low "lag_in_frames"
Having a very small value for "lag_in_frames" can result in
corrupt arf groups including displayed frames that update
the arf buffer and fake overlay frames that are not in fact
overlays of real arfs but are nevertheless starved of bits.

Leaving lag_in_frames at the default of 25 for these 5 frame two
pass VBR tests should now give rise to a valid ARF coding pattern
as follows:-  K(ey), A(rf), N(ormal), N, N, O(verlay).

This change is part of a response to BUG=webm:1454 where broken
arf groups interacted badly with a change that corrects for large rate
misses. However, it may still in some cases increase encode time by
virtue of the fact that the unit test now codes a correct coding pattern
with "hidden" ARF frames.

Change-Id: Ifd0246a4c1d0be247247c754024d7a4ed5f66a6b
2017-08-16 14:07:24 +01:00
Johann Koenig
c59d1a4dc7 Merge changes I1f1edeaa,I89313cac
* changes:
  quantize: silence unsigned overflow warning
  quantize test: quiet overflow warning
2017-08-15 17:37:59 +00:00
Johann
08cb7b5c68 quantize test: quiet overflow warning
Promote the result of RandRange to signed

Change-Id: I89313cace3bcbe9af96946bef00b6857fc48b128
2017-08-15 08:28:09 -07:00
Linfeng Zhang
d72e20b123 Add vpx_highbd_idct32x32_{34, 135, 1024}_add_{sse2, sse4_1}
BUG=webm:1412

Change-Id: I08b562b60fa85fbc2fec1c15c323a3444b44618f
2017-08-14 17:05:22 -07:00
Scott LaVarnway
fa85cf131c vp9: strip temporal filter code
when CONFIG_REALTIME_ONLY is enabled.

BUG=webm:1446

Change-Id: Id547783ec75383966c40ab5cf6abb4a0f7984f52
2017-08-14 14:27:53 -07:00
Johann Koenig
ff184e482a Merge changes I4b4beab1,I02f74dec
* changes:
  quantize test: check skip_block
  quantize test: use negative input
2017-08-14 20:52:52 +00:00
Johann Koenig
45b39750d6 Merge "temporal filter test: adjust inputs and runtime" 2017-08-14 20:46:22 +00:00
Johann
c06d6649c5 temporal filter test: adjust inputs and runtime
Use input with a narrow range because the filter only applies when the
frames are similar.

Run CompareReferenceRandom more times. Especially before narrowing the
input range, the filter frequently did not apply.

Change-Id: Ie249bedf6d0d33dfa5884611cb1835788e418b38
2017-08-14 17:24:11 +00:00
James Zern
746c0eab3b disable SSSE3/VP9QuantizeTest* in hbd builds
this test fails with the configuration similar to the assembly prior to:
d52cb5972 quantize: copy ssse3 optimizations to intrinsics

BUG=webm:1458

Change-Id: Idc5c0b84c0598259fc49609a9f0756de531d3baf
2017-08-14 09:31:14 -07:00
Johann Koenig
9bb8ce5efb Merge "neon: vpx_quantize_b_32x32" 2017-08-10 15:42:49 +00:00
Johann Koenig
0b393ae505 Merge "quantize: copy ssse3 optimizations to intrinsics" 2017-08-10 15:42:20 +00:00
Johann
357adb68b2 quantize test: check skip_block
Not all sizes were tested previously. Only 4x4 and 32x32

Change-Id: I4b4beab1b92a810a097a7306de04cc9e0e260315
2017-08-08 14:21:58 -07:00
Johann
1092cc7f1a quantize test: use negative input
coeff contains signed values.

Change-Id: I02f74decf30379a28122169ab3e844d0f3bd7d23
2017-08-08 14:19:56 -07:00
Johann
93166c5e51 neon: vpx_quantize_b_32x32
With skip block the neon is about twice as fast as C.

The neon has no shortcut for coeff < zbin so it always takes the
same amount of time. Even if the C can take the shortcut, it is over
twice as fast in neon. If it can't, that gap increases to over 10x.

BUG=webm:1426

Change-Id: I400722146c1b5a5f6289f67d85fd642463d2bfc6
2017-08-08 14:05:18 -07:00
Johann
d52cb59729 quantize: copy ssse3 optimizations to intrinsics
Fairly minor differences from sse2. pabsw and psignw are the big gains.
Also re-uses some values in eob calculation to avoid an extra pcmp.

Fixes test failures in HBD and OS X builds.

Allows using it in 32bit builds, where it is about 40% faster than sse2.

Substantially faster than the assembly for skip_block. 10-20% faster the
rest of the time.

Change-Id: If783bb3567e561e47667e10133b9c84414a334e2
2017-08-08 12:22:14 -07:00
Linfeng Zhang
853165ba39 Update 32x32 idct sse2 funcs, add partial case 135
Change-Id: I2b9add83f6fd8f9138fed3bec04a59877a237a6a
2017-08-07 17:37:02 -07:00
Linfeng Zhang
7f20c3ac44 Add vpx_highbd_idct16x16_{10, 38, 256}_add_sse4_1
BUG=webm:1412

Change-Id: I8877c986b4042f7b8e33f5674c86700675a0e4ca
2017-08-04 15:31:17 -07:00
Johann Koenig
cbb83ba4aa Merge "quantize test: consolidate sizes" 2017-08-04 20:34:50 +00:00
Johann
9578a84205 quantize test: consolidate sizes
Pass a max txfm size parameter and combine the base quantize
test with the 32x32 test.

Change-Id: I72ddf020fe6888e864ea9f3642ee2d9a8e48a04b
2017-08-04 12:45:32 -07:00
Linfeng Zhang
563d58ab84 Rewrite vpx_idct16x16_{10,256}_add_sse2() and add case 38 function
BUG=webm:1412

Change-Id: I945f0fb6807b8948747243794dc7352b959221f7
2017-08-03 13:59:47 -07:00
Yunqing Wang
6843e7c7f3 Merge "Force the bit exactness in the first pass" 2017-08-03 00:03:10 +00:00
Yunqing Wang
bfd0f41f9b Force the bit exactness in the first pass
Originally, for the purpose of keeping a fast first pass, the first-pass
stats between row_mt_mode = 0 and row_mt_mode = 1 are not bit exact, but
that difference is very small that doesn't cause a mismatch between the
final bitstreams. However, if the encoder changes, this minor difference
may cause a mismatch. Thus, this patch always forces the first pass to
be bit exact.

BUG=webm:1453

Change-Id: I2b67cf529dee81f660f9d9e7fe9a60ea3c7b12b8
2017-08-02 15:58:39 -07:00
Johann
1059b5cc52 quantize test: add speed comparison
Test some possible scenarios.

Change-Id: I1a612e7153b31756be66390ceea55877856d5a33
2017-08-02 09:33:35 -07:00
Johann Koenig
847394fe77 Merge "neon: vpx_quantize_b" 2017-08-01 16:44:31 +00:00
Johann
2d6b5df657 neon: vpx_quantize_b
With skip block or coeff < zbin it is about twice as fast as C.

If most coeff values are > zbin it is about 10-15x as fast as C.

BUG=webm:1426

Change-Id: I5d3c007b014a372d5ef0882b39bb48983b4131c7
2017-07-31 10:38:46 -07:00
Linfeng Zhang
75653b7032 Merge changes Ia0e20f5f,I28150789,I35df041b,I221dff34
* changes:
  Update vpx_idct16x16_10_add_sse2()
  Add vpx_idct16x16_38_add_sse2()
  Rewrite vpx_highbd_idct8x8_{12,64}_add_sse2
  Refactor highbd idct 4x4 and 8x8 x86 functions
2017-07-28 22:43:00 +00:00
James Zern
3c73e587d1 Revert "quantize ssse3: declare all variables"
This reverts commit 03f5e300d6.

This causes test failures under OSX:
SSSE3/VP9QuantizeTest.EOBCheck/0
SSSE3/VP9QuantizeTest.OperationCheck/0

Change-Id: I122732717ead1f7af5b04c529a6948e382e5e59b
2017-07-28 01:22:16 -07:00
Linfeng Zhang
7f4acf8700 Add vpx_idct16x16_38_add_sse2()
Change-Id: I28150789feadc0b63d2fadc707e48971b41f9898
2017-07-27 18:02:43 -07:00
Linfeng Zhang
9c43d81bc2 Refactor highbd idct 4x4 and 8x8 x86 functions
BUG=webm:1412

Change-Id: I221dff34dd5f71b390b5e043d0a137ccb0a01dec
2017-07-27 18:01:03 -07:00
Johann Koenig
a83e1f1d53 Merge "quantize ssse3: declare all variables" 2017-07-27 21:18:35 +00:00
Johann
af08fbb444 quantize test: promote RandRange() result to signed
Avoid unsigned overflow warning:
unsigned integer overflow: 19974 - 32703 cannot be represented in type
'unsigned int'

Change-Id: Ifebee014342e4c6f3b53306c0cad6ae0b465ac12
2017-07-20 08:17:48 -07:00
Johann
c782f27ead quantize test: lowbd functions do not pass in highbd
qcoeff output looks OK but dqcoeff is no good.

BUG=webm:1448

Change-Id: I07211db8a8b74f1f45fdd059852e2de0e5ee18fd
2017-07-20 08:17:48 -07:00
Johann
bde2e4aa36 quantize test: eob is output
eob values are generated by the function.

Change-Id: I8ce92100e83022bff99888a5a7e6ef378c49fda3
2017-07-19 14:17:19 -07:00
Johann
03f5e300d6 quantize ssse3: declare all variables
Copy missing line from avx implementation.

Change-Id: I9755c5b4d4034867de6fa9f741c24bf49dce3a27
2017-07-18 12:32:57 -07:00
Johann
101981b736 quantize test: test sse2 and avx optimizations
ssse3 does not pass either of the tests.

avx 32x32 does not pass.

Change-Id: I62c2e31336fd2327327afaa0da896ad79a3def44
2017-07-18 12:08:16 -07:00
Johann
c7ebe82253 quantize test: extend arrays
Officially the quant structures are 8 elements, with one dc element and
7 repeated ac elements. The low bit depth optimizations take advantage
of this to fill the xmm registers. The high bit depth version manually
duplicates the values.

If all the optimizations were unified, the structure sizes could be
greatly reduced.

Change-Id: Ibd7a0337a7832ce2a1a05ee433c310077e1059ae
2017-07-18 09:55:47 -07:00
Johann
cb61ba02f4 quantize test: restrict and correct input
Use only valid values for quantize inputs. These were determined by
looping over vp9_init_quantizer and looking for max and min values.

This allows extending the test to the low bit depth functions which were
not designed to handle all possible inputs but only valid inputs.

Change-Id: I94e1d8863a49ac227845b65c6b50130e10e6319e
2017-07-18 09:40:45 -07:00
James Zern
9223b947ca Merge "fix 'make exampletest' w/CONFIG_REALTIME_ONLY" 2017-07-15 18:37:10 +00:00
Johann
e3fa4ae8e3 quantize test: use Buffer
Although the low bitdepth functions are identical (excepting the need
for larger intermediate values) they do not pass these tests. This
improves the error output to aid debugging.

Simplify buffer usage with Buffer and removing unnecessarily aligned
variables.

eob is a single element and never written using aligned instructions.

BUG=webm:1426

Change-Id: Ic95789a135cf1e8a3846d85270f2b818f6ec7e35
2017-07-13 15:54:48 -07:00
James Zern
960466939d fix 'make exampletest' w/CONFIG_REALTIME_ONLY
for tests that aren't explicitly testing 2-pass behavior use --passes=1
with this configuration

Change-Id: I6a1520ecc65d0f626486604310af29dacb9f197f
2017-07-13 10:47:20 -07:00
Johann
e381753926 sad4d neon: 64x[32,64]
Rewrite 64x64.

BUG=webm:1425

Change-Id: I336bf5a3aa4b783389c10b16a50f0f559346ecbf
2017-07-12 13:26:39 +00:00
Johann
e1bde306c8 sad4d neon: 32x[16,32,64]
Rewrite 32x32. Use half the accumulator registers.

BUG=webm:1425

Change-Id: Ibf5e61dc4ba15056102aef8495f4a02c668c5d13
2017-07-12 13:25:18 +00:00
Johann
807ce8fb1e sad4d neon: 16x[8,16,32]
Rewrite 16x16. Use half the accumulator registers.

BUG=webm:1425

Change-Id: I44b48512b1e3629505d83c2645e800f53878ccc2
2017-07-12 13:25:11 +00:00
Johann
8152b0904d sad4d neon: 8x[4,8,16]
BUG=webm:1425

Change-Id: I7de2500cca4b621f21478c4b0333c56d76dbc9a4
2017-07-12 13:25:03 +00:00
Johann
dd4347e9ec sad4d neon: 4x4, 4x8
BUG=webm:1425

Change-Id: I5081b5ce131821d590c53ac1206a94f50cb8b468
2017-07-12 03:38:03 +00:00
Johann Koenig
4e16f70703 Merge changes Id84d9780,Iaa6ea75b,I3362e0dd,I0020a49e,Ia42e4f36, ...
* changes:
  sad neon: avg for 64x[32,64]
  sad neon: macroize 64xN definitions
  sad neon: avg for 32x[16,32,64]
  sad neon: macroize 32xN definitions
  sad neon: avg for 16x[8,16,32]
  sad neon: macroize 16xN definitions
2017-07-07 21:01:23 +00:00
James Zern
5d6060b62f Merge "cosmetics,vp9/: normalize inv/fwd_txfm naming" 2017-07-07 19:15:02 +00:00
Johann Koenig
6c375b9cd0 Merge "fdct neon: 32x32_rd" 2017-07-07 14:05:51 +00:00