Commit Graph

13731 Commits

Author SHA1 Message Date
Alex Converse
d8426d6f12 Add an SSE2 version of vp9_iwht4x4_16_add
Roughly half as many cycles as plain C.

Change-Id: I8c16c29940b76d54ee7e4fb874c328ce90bff5d4
2015-07-14 14:23:32 -07:00
Yaowu Xu
0cdc85d8cf Merge "Revert "Add an SSE2 version of vp9_iwht4x4_16_add."" 2015-07-13 16:27:10 +00:00
Yaowu Xu
ae5394b9e2 Revert "Add an SSE2 version of vp9_iwht4x4_16_add."
This reverts commit f8d3501640.

Change-Id: If8c7af403c091b7fb447a6f0c73fecdbccbc51b3
2015-07-13 16:26:27 +00:00
Jim Bankoski
c243835303 Merge "Revert "Fill buffer speed up"" 2015-07-13 14:39:01 +00:00
Jim Bankoski
da9db83270 Revert "Fill buffer speed up"
This reverts commit 9b4f9f45ee.

Change-Id: I23545ac8c7464127f7466fc6a58de517874fe0cf
2015-07-13 13:47:46 +00:00
Parag Salasakar
e4fb53ea6e Merge "mips msa vp8 loop filter optimization" 2015-07-13 01:16:45 +00:00
Jim Bankoski
db50037ece Merge "Fill buffer speed up" 2015-07-09 20:26:23 +00:00
Jim Bankoski
9b4f9f45ee Fill buffer speed up
Eliminates the byte by byte read from bool decoder,  by reading
in a size_t and then shifting it into place.

Change-Id: Id89241977103fc3b973e4ed172a5cbf246998e5d
2015-07-09 11:41:30 -07:00
paulwilkins
4b44e46de0 Merge "Changes to use of rectangular partitions." 2015-07-09 18:34:41 +00:00
Yaowu Xu
49fa5276fe Merge "Remove clamp operations." 2015-07-09 17:49:18 +00:00
Yaowu Xu
f70c80289c Merge "Clean out more MSVC warnings" 2015-07-09 17:49:08 +00:00
Scott LaVarnway
e8103f3676 Merge "Eliminate num_8x8 and num_4x4 width/height lookups" 2015-07-09 17:16:22 +00:00
Alex Converse
74f869b962 Merge "Add an SSE2 version of vp9_iwht4x4_16_add." 2015-07-09 16:57:03 +00:00
paulwilkins
2d637ca36d Merge "Change speed and rd features for formatting bars." 2015-07-09 16:38:38 +00:00
Scott LaVarnway
13a4f14710 Eliminate num_8x8 and num_4x4 width/height lookups
Also some log2 lookups.

Pass in 8x8 block width/height and log2 num4x4s instead.

Change-Id: I8ea9a1ec1e0bbab23f8ba556954a1b5433f4d613
2015-07-09 05:30:46 -07:00
Yaowu Xu
b58c99eb71 Remove clamp operations.
The clamp calls with INT32_MIN and INT32_MAX have no effect at all on
int values passed in, therefore this commit removes those effectless
clamps and also adds more const intermediate results to make the code
more readable.

Change-Id: I66d8811f58bb74ec31cbec9a6c441983a662352e
2015-07-08 17:44:19 -07:00
Jingning Han
535cc6d87f Format fixes in vp9_encodeframe.c and vp9_encodemb.c
Change-Id: Ib1303dac9043ab1b1f8fce54611cf4ea8a208038
2015-07-09 00:04:28 +00:00
Jingning Han
8783a8a97c Refactor transform block loop for inter mode decoding
Rework the inter mode transform block decoding loop. Replace the
block index with the row and col index as the input argument. It
saves function call to compute the row and col index according to
the block index and overall block size, and many if statements
associated with the transform block position relative to the coding
block. For the test bit-stream pedestrian_area 1080p at 5 Mbps,
the decoding speed goes up from 81.13 fps to 81.92 fps.

Note that the intra coded block decoding needs more refactoring
work than the inter ones. So keep it using foreach_transforme_block
as for now.

Change-Id: I5622bdae7be28ed5af96693274057f55ba9b4fb4
2015-07-08 22:55:16 +00:00
Yaowu Xu
c369daf3ea Clean out more MSVC warnings
Change-Id: I1bab0c104df2ec4825d050cd516e26ab635a7b3e
2015-07-08 15:09:20 -07:00
Alex Converse
f8d3501640 Add an SSE2 version of vp9_iwht4x4_16_add.
80% fewer cycles than C

Change-Id: I841bde1e268ddd33ae2ee75eee94737a400e2cde
2015-07-08 15:00:51 -07:00
Alex Converse
8bf791e7ef Merge "Don't allocate dqcoeff in MACROBLOCKD." 2015-07-08 20:42:36 +00:00
Alex Converse
89090d8046 Don't allocate dqcoeff in MACROBLOCKD.
The encoder gets its dqcoeff from the context tree. In the decoder move
it to directly after MACROBLOCKD.

Change-Id: I46c9b76f26956a360d17de0b26ecb994dae34ecb
2015-07-08 12:37:55 -07:00
Jingning Han
66da771040 Merge "Refactor inverse_transform_block argument list" 2015-07-08 19:28:25 +00:00
Jingning Han
0497d3a827 Merge "Reset dqcoeff[0] only if eob is 1" 2015-07-08 19:27:22 +00:00
Frank Galligan
b770def572 Merge "VP9_LPF_VERTICAL_16_DUAL_SSE2 optimization" 2015-07-08 18:15:39 +00:00
Johann
340d82efb9 Merge "Use 'unsigned long' for _BitScanReverse parameter" 2015-07-08 17:50:33 +00:00
paulwilkins
8dd466edc8 Changes to use of rectangular partitions.
Changes to allow more use of rectangular partitions at
speeds 1 and 2 for content classed by the first pass as
animation and for blocks near the active image edge.

This has quite a big impact in quality for the animated
test sequence but also hurts encode speed for speed 2.

For other content types the impact on both speed and
quality is small.

Added some plumbing for detection of internal vertical
image edges.

Change-Id: I3fc48de2349f8cb87946caaf0b06dbb0ea261a9a
2015-07-08 18:14:12 +01:00
paulwilkins
a126b6ce7d Change speed and rd features for formatting bars.
Change speed features / behavior for split mode when there
is an internal active edge (e.g. formatting bars).

Remove some threshold constraints in rd code near the active
edge of the image.

Add some plumbing for left and right active edge detection.

Patch set 5. Limit rd pass through for sub 8x8 to internal active edges.
This takes away any speed penalty for most clips but keeps the enhanced
edge coding for the more critical case of internal image edges

Change-Id: If644e4762874de4fe9cbb0a66211953fa74c13a5
2015-07-08 17:51:42 +01:00
Jingning Han
7e0d0de211 Refactor inverse_transform_block argument list
Replace block index with transform type in the argument list. This
allows to save an extra fetch to the prediction mode. For pedestrian
area 1080p coded at 5 Mbps with single tile, the average decoding
speed goes up from 80.55 fps (before the refactoring series) to
81.13 fps.

Change-Id: Icbebf84ce63c19c0c92f3690ed201f6c3eab7881
2015-07-08 09:26:02 -07:00
Parag Salasakar
0ea2684c2c mips msa vp8 loop filter optimization
average improvement ~2x-4x

Change-Id: I20c4f900ef95d99b18f9cf4db592cd352c2212eb
2015-07-08 12:41:00 +05:30
James Zern
892128f6ca Merge "vp9_entropymv: remove vp9_get_mv_mag()" 2015-07-08 01:27:13 +00:00
Johann
0f96939ceb Use 'unsigned long' for _BitScanReverse parameter
New clang warnings in chromium reveal an incompatible pointer being
passed.

Change-Id: I0d893993726ecf899f467c16175bfb8c76329731
https://code.google.com/p/chromium/issues/detail?id=504696
2015-07-07 17:45:06 -07:00
Frank Galligan
5327fcf857 Merge "Add vp9_int_pro_row_neon." 2015-07-08 00:16:03 +00:00
Johann
ac7f403cbe Merge "Move sub pixel variance to vpx_dsp" 2015-07-07 23:57:18 +00:00
Jingning Han
55c2646666 Merge "Rework scan order fetch logic for decoder" 2015-07-07 23:09:39 +00:00
Johann
6a82f0d7fb Move sub pixel variance to vpx_dsp
Change-Id: I66bf6720c396c89aa2d1fd26d5d52bf5d5e3dff1
2015-07-07 15:51:04 -07:00
Marco
155b9416b3 Merge "Update to speed 5 non-rd mode partition search." 2015-07-07 22:47:47 +00:00
Jingning Han
c2d0f9ddeb Merge "Add vp9_ prefix to init_macroblockd" 2015-07-07 22:35:45 +00:00
Jingning Han
6e6c57da9a Merge "Reduce dqcoeff array size in decoder" 2015-07-07 22:35:31 +00:00
Jingning Han
76ccba9ec8 Reset dqcoeff[0] only if eob is 1
If only the first dequantized coefficient is non-zero, reset
dqcoeff[0] to zero directly.

Change-Id: I0197ba72028a8ec436f0b1b9abcc1c0ae5d70abe
2015-07-07 15:20:34 -07:00
Jingning Han
97d1f1aaae Rework scan order fetch logic for decoder
Save redundant call for getting prediction mode to obtain scan
order for detokenization.

Change-Id: I0683ef119f1579d1261ed5d59052a1745b68ef6f
2015-07-07 15:03:21 -07:00
Marco
1acdd04d2d Merge "Update to vpx_temporal_svc_encoder:" 2015-07-07 21:34:09 +00:00
Jingning Han
9d251f9510 Merge "Unify subtract function used in VP8/9" 2015-07-07 20:42:19 +00:00
Yaowu Xu
ea5450b280 Merge "Allows using optimzed version vp9_fdct8x8" 2015-07-07 19:28:57 +00:00
Johann
3caf2bb606 Merge "Remove comment about disabled tests" 2015-07-07 19:14:35 +00:00
Jingning Han
a652048efd Add vp9_ prefix to init_macroblockd
Change-Id: I202d4924e627eec94838741df004ed9259d38b88
2015-07-07 12:00:01 -07:00
Marco
478fbc8f23 Update to speed 5 non-rd mode partition search.
If the pre-selected partition size (from variance partition) is
32x32, also apply nonrd partition search for 32x32 and 16x16 size.

Overall small positive gain in metrics, average ~1%.
Some visual improvement, for lower resolutions.

Change-Id: I69cb425bda94f7d13d34c451ab30e9276335a30e
2015-07-07 11:52:01 -07:00
Jingning Han
cccad1c5de Reduce dqcoeff array size in decoder
The decoding process handles detokenization and reconstruction per
transform block sequentially. There is no need to offset the dqcoeff
buffer according to the transform block index. This allows to
reduce the memory spill and improve cache performance.

Change-Id: Ibb8bfe532a7a08fcabaf6d42cbec1e986901d32d
2015-07-07 11:36:05 -07:00
Parag Salasakar
45b7e160d8 Merge "mips msa vp8 idct optimization" 2015-07-07 18:14:20 +00:00
Marco
83f8ab5001 Update to vpx_temporal_svc_encoder:
Add set_tune_content control setting.

Change-Id: I8f7af8e8e9bc35a2d1546c5a0b2dd14be3cce750
2015-07-07 11:08:50 -07:00