1473 Commits

Author SHA1 Message Date
Angie Chiang
dfa532cc2a Let txfm's constant bit be the same for each stage
Change-Id: I763f2924afca526db371231bca18b38879bdf793
2016-04-14 15:46:54 -07:00
Angie Chiang
02d23fbbf4 Fit adst/dct's stage range into 32-bit in bd12
Change-Id: Ie428c6f0655873de3e77e844a2f2e4203cf47dff
2016-04-14 15:44:05 -07:00
Jingning Han
019683e963 Merge "Clean up motion vector precision check in the encoding process" into nextgenv2 2016-04-14 20:55:51 +00:00
Jingning Han
79bef030f2 Merge "Apply motion vector precision check to candidate mv" into nextgenv2 2016-04-14 20:55:45 +00:00
Jingning Han
03a468f9ac Merge "Enable mode conversion in sub8x8 block" into nextgenv2 2016-04-14 19:01:15 +00:00
Jingning Han
6af8f63d96 Clean up motion vector precision check in the encoding process
Remove unnecessary motion vector precision check in the encoding
process.

Change-Id: Ica32933c7d138f499f36b1dedec14c894b27d85a
2016-04-14 11:37:19 -07:00
Jingning Han
525995a3d9 Apply motion vector precision check to candidate mv
This avoids repeatedly checking the candidate motion vector
precision level at the decoder end. The compression performance
varies at 0.01% level.

Change-Id: I4a88e95decd900d0cac9a0c2e70ba43ef7ecac38
2016-04-14 09:44:41 -07:00
Jingning Han
cd39224cff Merge "Speed up dynamic motion vector referencing system" into nextgenv2 2016-04-14 16:16:43 +00:00
Hui Su
436a6cc4e7 Merge "ext-tx: use raster scan order for identity transform" into nextgenv2 2016-04-13 23:52:35 +00:00
Jingning Han
885a81f468 Merge "Fix a few mis-use cases of MAX_MV_REF_CANDIDATES" into nextgenv2 2016-04-13 23:44:25 +00:00
Angie Chiang
716f0ea3cf Merge changes I92819356,I50b5a313,I807e60c6,I8a8df9fd into nextgenv2
* changes:
  Branch dct to new implementation for bd12
  Change dct32x32's range
  Fit dct's stage range into 32-bit when bitdepth is 12
  Pass tx_type into get_tx_scale
2016-04-13 23:24:41 +00:00
Alex Converse
91c985fc28 Merge "Convert some vpx boolcoder calls back to vp10 generic calls." into nextgenv2 2016-04-13 23:04:17 +00:00
Hui Su
85a3f5b740 Merge "Speed-up in tx_size search" into nextgenv2 2016-04-13 23:02:21 +00:00
Jingning Han
9a1a8f1d8e Speed up dynamic motion vector referencing system
Skip transform type search in modes with ref_mv_idx > 0. This
brings down the additional encoding time cost due to the DMR system
from 32% to 17%, at minimal coding performance regression.

Change-Id: Ie82e1d2831a313c6f1e47f7da221b51345023eb3
2016-04-13 15:51:36 -07:00
Jingning Han
f33a0a8215 Fix a few mis-use cases of MAX_MV_REF_CANDIDATES
Fix several use cases where MAX_MV_REF_CANDIDATES is mixed up with
is_compound flag to avoid potential coding interruption.

Change-Id: Ifdee1ef8a81ef6d1c155315c6c6a3074aa7a8b5e
2016-04-13 15:16:55 -07:00
Alex Converse
5d2b0f93b9 Use an exponential growth approach for the ANS reversal buffer.
Memory constrained hardware can window the data via our standard windowing
mechanism, tiles.

Change-Id: Ib1cfd157604a8c9d9f9a9f2b0ba3bc2fd0643082
2016-04-13 15:16:29 -07:00
Jingning Han
e07dbaa2f5 Enable mode conversion in sub8x8 block
Convert the newmv mode into reference motion vector modes.

Change-Id: I51bd2543dafb70345c1340fba700b44f67f20853
2016-04-13 14:35:54 -07:00
Zoe Liu
1d043d56da Merge "Make ext-refs respect encoding flags." into nextgenv2 2016-04-13 19:31:30 +00:00
hui su
6a7ddd84bb Speed-up in tx_size search
Do not consider 4x4 transform when the maximum possible transform
size is 32x32.

Overall encoding speed is increased by more than 10%. Compression
performance is neutral on lowres, midres, and hdres.

Change-Id: Ifac61c3c9f4b0ab392bffd4d1faa373d91014cf1
2016-04-13 10:19:00 -07:00
hui su
b72aa72a90 ext-tx: use raster scan order for identity transform
coding gain of ext-tx:
screen_content 12.73% -> 13.05%

Change-Id: I5fc8cf0db84c3e56dd3cb7675e1d81c9c575bc57
2016-04-13 09:42:43 -07:00
Geza Lore
c50aaf3049 Make ext-refs respect encoding flags.
The VP8_EFLAG_NO_UPD_LAST and VP8_EFLAG_NO_REF_LAST flags can be
passed to the encoder to signal that it should not update/reference
the LAST ref frame when encoding the current frame. With
--enable-ext-refs turned on, the new LAST2 LAST3 and LAST4 ref frames
could still be used or updated, which causes the
  VP10/ErrorResilienceTestLarge.DropFramesWithoutRecovery/{0,1,2}
tests to fail.

With this patch, if --enable-ext-refs is used, then
VP8_EFLAG_NO_UPD_LAST and VP8_EFLAG_NO_REF_LAST also applies to the
new LAST2 LAST3 and LAST4 ref frames, as well as the LAST ref frame.

Change-Id: If482b1c09bbaf914eca8e0348a2367bff261661d
2016-04-13 12:03:58 +01:00
Geza Lore
c6cf7a6111 Fix 2 warning when building with GCC 5.
These caused the following warning with GCC 5:
     warning: logical not is only applied to the left hand side of
     comparison [-Wlogical-not-parentheses]
     assert(!is_compound == (cm->reference_mode == SINGLE_REFERENCE));

Change-Id: If296aabb2311ceb7d903b395c1549ef81c2cbf9b
2016-04-13 10:49:52 +01:00
Alex Converse
c1729d12b8 Merge "ANS: Remove extra buffer size checks causing a false decode error." into nextgenv2 2016-04-13 01:37:05 +00:00
Hui Su
9e8cad3be7 Merge "Add vp10_ prefix to full_to_model_counts and fill_token_costs" into nextgenv2 2016-04-12 23:38:47 +00:00
Alex Converse
493a585273 ANS: Remove extra buffer size checks causing a false decode error.
The minimal ans partition size is now one byte. This is checked in
ans_read_init().

The read_is_valid() condition is handled by setup_token_decoder().

Change-Id: I7b202b896630bc4285532208bf7cf84567afe158
2016-04-12 15:19:30 -07:00
Debargha Mukherjee
ec1365a0c9 Merge "Extend variance based partitioning to 128x128 superblocks" into nextgenv2 2016-04-12 19:42:35 +00:00
hui su
0792748646 Add vp10_ prefix to full_to_model_counts and fill_token_costs
Change-Id: I5e6c644fb09f7a80c88142dfdfa05cf5be260241
2016-04-12 11:06:47 -07:00
Angie Chiang
027d12b7d6 Merge changes I359aa49c,Ic8ca5afb into nextgenv2
* changes:
  Generalize txfm scale in highbd quantizer
  Parameterize transform scale for quantizer
2016-04-12 18:02:05 +00:00
Alex Converse
e7224b7866 Convert some vpx boolcoder calls back to vp10 generic calls.
Change-Id: I362f753ff42d4c4fb94df2419cdaad423d7a4229
2016-04-12 11:00:52 -07:00
Geza Lore
61af8981b0 Extend variance based partitioning to 128x128 superblocks
Change-Id: I41edf266d5540a9b070a5e65bc397dd3da210507
2016-04-12 09:40:11 +01:00
Debargha Mukherjee
648538959d Merge "Use reduced transform set for 16x16" into nextgenv2 2016-04-11 23:32:29 +00:00
Debargha Mukherjee
c4da5d500e Use reduced transform set for 16x16
Speed increase for ext-tx by 20% for a BDRATE drop of 0.26%.
The ext-tx expt becomes -2.66% BDRATE (reduced from -2.92%) for
the lowres set.

It turns out that reducing the set of transforms for intra from
12 to 5 makes very little difference in coding performance (~0.04%).
Most of the performance drop comes from the reduction is transform
set for inter. Currently there is a provision to control that with
a macro.

Change-Id: I7de05527bf72f96acc1e0ab8a74a849da0a141e5
2016-04-11 13:04:41 -07:00
Sarah Parker
33ccd0f85e Merge "Fix prune one and two to make compatible with new transforms" into nextgenv2 2016-04-11 17:12:28 +00:00
Yi Luo
fd367c243e Merge "Some cosmetic improvements since HBD variance 4x4 optimization" into nextgenv2 2016-04-11 16:04:02 +00:00
Debargha Mukherjee
9930a00ed7 Merge "Refactor PC_TREE root handling." into nextgenv2 2016-04-09 13:33:53 +00:00
Debargha Mukherjee
38b26b0dc3 Merge "Make subpel masked motion work with upsampled refs" into nextgenv2 2016-04-09 13:30:09 +00:00
Hui Su
5e558121fe Merge "Changes to scan order neighbors" into nextgenv2 2016-04-08 21:18:16 +00:00
Hui Su
7fcf94eb23 Merge "Reformat scan order neighbors" into nextgenv2 2016-04-08 20:57:50 +00:00
Sarah Parker
19e3c6415c Fix prune one and two to make compatible with new transforms
Update svm parameters with training data using new transforms
and remove DST from pruning functions.

Change-Id: I7bd1c4744455d571c1ecfb4cea14c25ac291f002
2016-04-08 11:47:48 -07:00
hui su
f94d699c09 Changes to scan order neighbors
-Fix some bugs in row_scan and col_scan. In some cases, the above
or left neighbor was not considered even though it is available.

-When above or left neighbor is not available, try using the
top-left, top-right or bottom-left neighbor.

Compression improvement:
lowres   0.20%
midres   0.16%
hdres    0.20%

Change-Id: If521665589c7f29277b8e9223f21f4a8bf3fef39
2016-04-08 11:08:57 -07:00
hui su
b76118b736 Reformat scan order neighbors
Change-Id: Iafcd080612012b08f3cbff45335c12f434543f38
2016-04-08 10:50:13 -07:00
Yi Luo
e5f4e8eab9 Some cosmetic improvements since HBD variance 4x4 optimization
Change-Id: I414c1fabd2e3a9b1d9daa8a90f85a0bace8bd3cd
2016-04-08 10:32:13 -07:00
Geza Lore
f2be4f6058 Refactor PC_TREE root handling.
Change-Id: Id8b16c1b18bd6f909e72aae3fd582dd3503c88c6
2016-04-08 17:01:00 +01:00
Debargha Mukherjee
c485b10416 Make subpel masked motion work with upsampled refs
Change-Id: Id483354e73e983793370b55a1a6a1f2dcd137dc9
2016-04-08 08:54:58 -07:00
Alex Converse
bb0e692151 Convert palette from double to float.
About 20% less time spent coding in vp10_k_means().

Change-Id: I5cf7605cde869a269776197bace70de353b07d83
2016-04-07 15:17:30 -07:00
hui su
69c7ad3407 Correct comments for scan order neighbors
Change-Id: I5e2dc39bf0ee8e501e4dd358be2e92ae50934593
2016-04-07 11:07:21 -07:00
Geza Lore
454989ff32 Make superblock size variable at the frame level.
The uncompressed frame header contains a bit to signal whether the
frame is encoded using 64x64 or 128x128 superblocks. This can vary
between any 2 frames.

vpxenc gained the --sb-size={64,128,dynamic} option, which allows the
configuration of the superblock size used (default is dynamic). 64/128
will force the encoder to always use the specified superblock size.
Dynamic would enable the encoder to choose the sb size for each
frame, but this is not implemented yet (dynamic does the same as 128
for now).

Constraints on tile sizes depend on the superblock size, the following
is a summary of the current bitstream syntax and semantics:

If both --enable-ext-tile is OFF and --enable-ext-partition is OFF:
     The tile coding in this case is the same as VP9. In particular,
     tiles have a minimum width of 256 pixels and a maximum width of
     4096 pixels. The tile width must be multiples of 64 pixels
     (except for the rightmost tile column). There can be a maximum
     of 64 tile columns and 4 tile rows.

If --enable-ext-tile is OFF and --enable-ext-partition is ON:
     Same constraints as above, except that tile width must be
     multiples of 128 pixels (except for the rightmost tile column).

There is no change in the bitstream syntax used for coding the tile
configuration if --enable-ext-tile is OFF.

If --enable-ext-tile is ON and --enable-ext-partition is ON:
     This is the new large scale tile coding configuration. The
     minimum/maximum tile width and height are 64/4096 pixels. Tile
     width and height must be multiples of 64 pixels. The uncompressed
     header contains two 6 bit fields that hold the tile width/heigh
     in units of 64 pixels. The maximum number of tile rows/columns
     is only limited by the maximum frame size of 65536x65536 pixels
     that can be coded in the bitstream. This yields a maximum of
     1024x1024 tile rows and columns (of 64x64 tiles in a 65536x65536
     frame).

If both --enable-ext-tile is ON and --enable-ext-partition is ON:
     Same applies as above, except that in the bitstream the 2 fields
     containing the tile width/height are in units of the superblock
     size, and the superblock size itself is also coded in the bitstream.
     If the uncompressed header signals the use of 64x64 superblocks,
     then the tile width/height fields are 6 bits wide and are in units
     of 64 pixels. If the uncompressed header signals the use of 128x128
     superblocks, then the tile width/height fields are 5 bits wide and
     are in units of 128 pixels.

The above is a summary of the bitstream. The user interface to vpxenc
(and the equivalent encoder API) behaves a follows:

If --enable-ext-tile is OFF:
     No change in the user interface. --tile-columns and --tile-rows
     specify the base 2 logarithm of the desired number of tile columns
     and tile rows. The actual number of tile rows and tile columns,
     and the particular tile width and tile height are computed by the
     codec ensuring all of the above constraints are respected.

If --enable-ext-tile is ON, but --enable-ext-partition is OFF:
     No change in the user interface. --tile-columns and --tile-rows
     specify the WIDTH and HEIGHT of the tiles in unit of 64 pixels.
     The valid values are in the range [1, 64] (which corresponds to
     [64, 4096] pixels in increments of 64.

If both --enable-ext-tile is ON and --enable-ext-partition is ON:
     If --sb-size=64 (default):
         The user interface is the same as in the previous point.
         --tile-columns and --tile-rows specify tile WIDTH and HEIGHT,
         in units of 64 pixels, in the range [1, 64] (which corresponds
         to [64, 4096] pixels in increments of 64).
     If --sb-size=128 or --sb-size=dynamic:
         --tile-columns and --tile-rows specify tile WIDTH and HEIGHT,
         in units of 128 pixels in the range [1, 32] (which corresponds
         to [128, 4096] pixels in increments of 128).

Change-Id: Idc9beee1ad12ff1634e83671985d14c680f9179a
2016-04-07 10:34:25 +01:00
Julia Robson
4300e50cce Fixing assertion in *Large unit tests
In certain cases the code was subtracting the obmc cost
despite it not having been added previously.
For example with ref_mv, supertx, ext_inter, obmc & ext_refs
enabled the following test was failing but now passes:
"VP10/ArfFreqTestLarge.MinArfFreqTest/33"

Change-Id: I966853f34c18d5a1d4c7a56fa201c1b02973fc88
2016-04-06 11:22:10 +01:00
Julia Robson
0f05873c89 Fixing error when building with ref_mv experiment
Code was using variable that was only defined when
var_tx was also enabled

Change-Id: Ide02ff99b433bfc5c95b71e700c66562020cedae
2016-04-06 11:22:10 +01:00
Debargha Mukherjee
de3d15bb2c Merge "Refactoring and cosmetic changes to ext-inter expt" into nextgenv2 2016-04-06 01:19:06 +00:00