platform/upstream/libvpx.git
10 years agoMerge "Add neon optimize vp9_short_idct16x16_add."
hkuang [Wed, 14 Aug 2013 21:16:20 +0000 (14:16 -0700)]
Merge "Add neon optimize vp9_short_idct16x16_add."

10 years agoAdd neon optimize vp9_short_idct16x16_add.
hkuang [Tue, 6 Aug 2013 23:05:14 +0000 (16:05 -0700)]
Add neon optimize vp9_short_idct16x16_add.

Change-Id: I27134b9a5cace2bdad53534562c91d829b48838d

10 years agoforeach_transformed_block_in_plane cleanup, explicit tx_size var.
Dmitry Kovalev [Wed, 14 Aug 2013 18:39:31 +0000 (11:39 -0700)]
foreach_transformed_block_in_plane cleanup, explicit tx_size var.

Making foreach_transformed_block_in_plane more clear (it's not finished
yet). Using explicit tx_size variable consistently instead of
(ss_txfrm_size / 2) or (ss_txfrm_size >> 1) expression.

Change-Id: I1b9bba2c0a9f817fca72c88324bbe6004766fb7d

10 years agoAdding const to arguments of intra prediction functions.
Dmitry Kovalev [Wed, 14 Aug 2013 17:35:56 +0000 (10:35 -0700)]
Adding const to arguments of intra prediction functions.

Adding const to above and left pointers. Cleanup.

Change-Id: I51e195fa2e2923048043fe68b4e38a47ee82cda1

10 years agoMerge "Honor min_partition_size properly for non-square splits"
Paul Wilkins [Wed, 14 Aug 2013 11:45:18 +0000 (04:45 -0700)]
Merge "Honor min_partition_size properly for non-square splits"

10 years agoHonor min_partition_size properly for non-square splits
Guillaume Martres [Tue, 13 Aug 2013 19:34:32 +0000 (12:34 -0700)]
Honor min_partition_size properly for non-square splits

Don't do vertical or horizontal splits if subsize < min_partition_size,
except for edge blocks where it makes sense.

Change-Id: I479aa66ba1838d227b5de8312d46be184a8d6401

10 years agoMerge "Little cleanup inside decode_tile() function."
Dmitry Kovalev [Tue, 13 Aug 2013 21:43:10 +0000 (14:43 -0700)]
Merge "Little cleanup inside decode_tile() function."

10 years agoMerge "Trivial clean up."
Guillaume Martres [Tue, 13 Aug 2013 19:40:37 +0000 (12:40 -0700)]
Merge "Trivial clean up."

10 years agoUse lookup table to find largest txfm size
Jingning Han [Mon, 12 Aug 2013 23:56:32 +0000 (16:56 -0700)]
Use lookup table to find largest txfm size

Refactor choose_largest_txfm_size_ and make it find the largest
transform size via lookup table.

Change-Id: I685e0396d71111b599d5367ab1b9c934bd5490c8

10 years agoMerge "Using is_inter_block() instead of repetitive code."
Dmitry Kovalev [Tue, 13 Aug 2013 17:00:01 +0000 (10:00 -0700)]
Merge "Using is_inter_block() instead of repetitive code."

10 years agoMerge "Refactor model based tx search in super_block_yrd"
Jingning Han [Tue, 13 Aug 2013 15:48:49 +0000 (08:48 -0700)]
Merge "Refactor model based tx search in super_block_yrd"

10 years agoTrivial clean up.
Paul Wilkins [Tue, 13 Aug 2013 12:22:13 +0000 (13:22 +0100)]
Trivial clean up.

Delete unused / commented out  variable references.

Change-Id: Iaf20c0c3744f89adb296d153b516b5ea41b4f3b4

10 years agoMerge "Honor min_partition_size properly"
Paul Wilkins [Tue, 13 Aug 2013 12:19:51 +0000 (05:19 -0700)]
Merge "Honor min_partition_size properly"

10 years agoMerge "Broken loop filter case."
Paul Wilkins [Tue, 13 Aug 2013 08:56:29 +0000 (01:56 -0700)]
Merge "Broken loop filter case."

10 years agoMerge "SSE2 high precision 32x32 forward DCT"
Jingning Han [Tue, 13 Aug 2013 06:03:47 +0000 (23:03 -0700)]
Merge "SSE2 high precision 32x32 forward DCT"

10 years agoLittle cleanup inside decode_tile() function.
Dmitry Kovalev [Tue, 13 Aug 2013 01:28:13 +0000 (18:28 -0700)]
Little cleanup inside decode_tile() function.

Change-Id: I3ed4beb59371fe21ca3e82253aa98e0cbd5e0630

10 years agoMerge "vp9: neon: optimise convolve8_vert functions"
Johann [Tue, 13 Aug 2013 00:54:47 +0000 (17:54 -0700)]
Merge "vp9: neon: optimise convolve8_vert functions"

10 years agoMerge "vp9: neon: optimise convolve8_horiz functions"
Johann [Tue, 13 Aug 2013 00:54:42 +0000 (17:54 -0700)]
Merge "vp9: neon: optimise convolve8_horiz functions"

10 years agoUsing is_inter_block() instead of repetitive code.
Dmitry Kovalev [Tue, 13 Aug 2013 00:42:14 +0000 (17:42 -0700)]
Using is_inter_block() instead of repetitive code.

Change-Id: If0b04c476c34fb8c102c9f750d7fe5669a86a532

10 years agoSSE2 high precision 32x32 forward DCT
Jingning Han [Wed, 7 Aug 2013 21:45:37 +0000 (14:45 -0700)]
SSE2 high precision 32x32 forward DCT

Enable SSE2 implementation of high precision 32x32 forward DCT. The
intermediate stacks are of 32-bits. The run-time goes down from
32126 cycles to 13442 cycles.

Change-Id: Ib5ccafe3176c65bd6f2dbdef790bd47bbc880e56

10 years agoRefactor model based tx search in super_block_yrd
Jingning Han [Mon, 12 Aug 2013 23:34:48 +0000 (16:34 -0700)]
Refactor model based tx search in super_block_yrd

Remove unnecessary conditional branches in model-based transform
size search.

Change-Id: Ic862dc33ed6710a186f6248239dd5f09b5c19981

10 years agoMerge "Simplifying vp9_mvref_common.c."
Dmitry Kovalev [Mon, 12 Aug 2013 23:24:22 +0000 (16:24 -0700)]
Merge "Simplifying vp9_mvref_common.c."

10 years agoMerge "Removing foreach_predicted_block_uv function."
Dmitry Kovalev [Mon, 12 Aug 2013 22:54:01 +0000 (15:54 -0700)]
Merge "Removing foreach_predicted_block_uv function."

10 years agoMerge "Using MV* instead of int_mv* as argument of vp9_clamp_mv_min_max."
Dmitry Kovalev [Mon, 12 Aug 2013 22:53:25 +0000 (15:53 -0700)]
Merge "Using MV* instead of int_mv* as argument of vp9_clamp_mv_min_max."

10 years agoSimplifying vp9_mvref_common.c.
Dmitry Kovalev [Mon, 12 Aug 2013 22:52:08 +0000 (15:52 -0700)]
Simplifying vp9_mvref_common.c.

Change-Id: I272df2e33fa05310466acf06c179728514dd7494

10 years agoMerge "Entropy context related cleanups."
Dmitry Kovalev [Mon, 12 Aug 2013 22:18:24 +0000 (15:18 -0700)]
Merge "Entropy context related cleanups."

10 years agoMerge "Making scaling code more clear."
Dmitry Kovalev [Mon, 12 Aug 2013 22:17:26 +0000 (15:17 -0700)]
Merge "Making scaling code more clear."

10 years agoRemoving unused convolve_avg_c function + cleanup.
Dmitry Kovalev [Mon, 12 Aug 2013 21:28:00 +0000 (14:28 -0700)]
Removing unused convolve_avg_c function + cleanup.

Change-Id: Id2b126c6456627c25e4041a82e304d0151d951ba

10 years agoUsing MV* instead of int_mv* as argument of vp9_clamp_mv_min_max.
Dmitry Kovalev [Mon, 12 Aug 2013 18:18:28 +0000 (11:18 -0700)]
Using MV* instead of int_mv* as argument of vp9_clamp_mv_min_max.

Change-Id: I3c45916a9059f11b41e9d798e34ffee052969a44

10 years agoRemoving foreach_predicted_block_uv function.
Dmitry Kovalev [Mon, 12 Aug 2013 20:54:13 +0000 (13:54 -0700)]
Removing foreach_predicted_block_uv function.

Adding function build_inter_predictors_for_planes to build inter
predictors for specified planes. This function allows to remove
condition "#if CONFIG_ALPHA" and use MAX_MB_PLANE for general case.
Renaming 'which_mv' local var to 'ref', and 'weight' argument to 'ref'.

Change-Id: I1a97160c9263006929d38953f266bc68e9c56c7d

10 years agoMaking scaling code more clear.
Dmitry Kovalev [Mon, 12 Aug 2013 20:30:26 +0000 (13:30 -0700)]
Making scaling code more clear.

Reusing existing functions, using constants instead of magic numbers.

Change-Id: Idc689ffba52c9a8b203fcf26bd67110ecb5635f9

10 years agoBroken loop filter case.
Paul Wilkins [Mon, 12 Aug 2013 18:46:47 +0000 (19:46 +0100)]
Broken loop filter case.

Loop filter level moved to common but this case missed.

Change-Id: I7fcb557e46ef4ed8e2b5e9c3e82cb042b55bbd7f

10 years agoFix a compile failure in vp9_get_compressed_data
Jingning Han [Mon, 12 Aug 2013 18:41:09 +0000 (11:41 -0700)]
Fix a compile failure in vp9_get_compressed_data

The lf struct is now with VP9_COMMON, instead of MACROBLOCKD.

Change-Id: Idfdd4f91f78f486078a138322d58bb61e93e1bc9

10 years agoEntropy context related cleanups.
Dmitry Kovalev [Mon, 12 Aug 2013 18:24:24 +0000 (11:24 -0700)]
Entropy context related cleanups.

Adding set_skip_context() function used from both encoder and decoder.

Change-Id: Ia22cfad3211a00a63eb294f64f857b78f4aa9b85

10 years agovp9: neon: optimise convolve8_vert functions
Mans Rullgard [Mon, 12 Aug 2013 14:37:48 +0000 (15:37 +0100)]
vp9: neon: optimise convolve8_vert functions

Invert loops to operate vertically in the inner loop.  This allows
removing redundant loads.

Also add preloading of data.

Change-Id: I4fa85c0ab1735bcb1dd6ea58937efac949172bdc

10 years agoMerge "Removing redundant code and function arguments."
Dmitry Kovalev [Sun, 11 Aug 2013 19:20:58 +0000 (12:20 -0700)]
Merge "Removing redundant code and function arguments."

10 years agovp9: neon: optimise convolve8_horiz functions
Mans Rullgard [Sun, 11 Aug 2013 14:34:24 +0000 (15:34 +0100)]
vp9: neon: optimise convolve8_horiz functions

Each iteration of the horizontal loop reuses 7 of the 11 source
values.  Loading only the 4 new values saves some time.

Also add preload for source data.

Overall 4% faster on Chromebook.

Change-Id: I8f69e749f2b7f79e9734620dcee51dbfcd716b44

10 years agoRenaming BLOCK_SIZE_TYPES constant to BLOCK_SIZES.
Dmitry Kovalev [Sat, 10 Aug 2013 00:47:32 +0000 (17:47 -0700)]
Renaming BLOCK_SIZE_TYPES constant to BLOCK_SIZES.

There will be another change set to rename BLOCK_SIZE_TYPE enum to
BLOCK_SIZE.

Change-Id: I8d1dfc873d6186fa5e554262f5169e929978085e

10 years agoHonor min_partition_size properly
Guillaume Martres [Sat, 10 Aug 2013 00:28:33 +0000 (17:28 -0700)]
Honor min_partition_size properly

It represents the minimum partition size, so don't split if
bsize == min_partition_size .

Change-Id: Id77c32d6afef7d2ddec0368eaae18fb13227d30e

10 years agoRemoving redundant code and function arguments.
Dmitry Kovalev [Sat, 10 Aug 2013 00:24:40 +0000 (17:24 -0700)]
Removing redundant code and function arguments.

Change-Id: Ia5cdda0f755befcd1e64397452c42cb7031ca574

10 years agoMerge "Inlining 16 as a stride for BLOCK_OFFSET macro."
Dmitry Kovalev [Sat, 10 Aug 2013 00:22:46 +0000 (17:22 -0700)]
Merge "Inlining 16 as a stride for BLOCK_OFFSET macro."

10 years agoMerge "vp9_rd_pick_inter_mode_sb: fix uninitialized value"
James Zern [Sat, 10 Aug 2013 00:13:32 +0000 (17:13 -0700)]
Merge "vp9_rd_pick_inter_mode_sb: fix uninitialized value"

10 years agoInlining 16 as a stride for BLOCK_OFFSET macro.
Dmitry Kovalev [Fri, 9 Aug 2013 23:40:05 +0000 (16:40 -0700)]
Inlining 16 as a stride for BLOCK_OFFSET macro.

Change-Id: I7f23d174eb089e5500f268a10db09648634c1b82

10 years agovp9_rd_pick_inter_mode_sb: fix uninitialized value
James Zern [Fri, 9 Aug 2013 23:15:08 +0000 (16:15 -0700)]
vp9_rd_pick_inter_mode_sb: fix uninitialized value

'skippable' can remain unset and negatively affect later decisions

address one aspect of issue #599

Change-Id: Iffdf0ac2e49ac481c27dc27c87fa546d4167bb28

10 years agoMerge "Using MV struct instead of int[2] array."
Dmitry Kovalev [Fri, 9 Aug 2013 22:33:08 +0000 (15:33 -0700)]
Merge "Using MV struct instead of int[2] array."

10 years agoMerge "Removing plane_block_{width, height}_log2by4 functions."
Dmitry Kovalev [Fri, 9 Aug 2013 22:26:51 +0000 (15:26 -0700)]
Merge "Removing plane_block_{width, height}_log2by4 functions."

10 years agoMerge "Moving loopfilter struct to VP9_COMMON."
Dmitry Kovalev [Fri, 9 Aug 2013 22:23:00 +0000 (15:23 -0700)]
Merge "Moving loopfilter struct to VP9_COMMON."

10 years agoMoving loopfilter struct to VP9_COMMON.
Dmitry Kovalev [Fri, 9 Aug 2013 21:41:51 +0000 (14:41 -0700)]
Moving loopfilter struct to VP9_COMMON.

Loop filter configuration doesn't belong to macroblock, so moving it from
MACROBLOCKD to VP9_COMMON. Also moving the declaration of loopfilter struct
from vp9_blockd.h to vp9_loopfilter.h.

Change-Id: I4b3e34be9623b47cda35f9b1f9951f8c5b1d5d28

10 years agoMoving scale_factors and related code to separate files.
Dmitry Kovalev [Fri, 9 Aug 2013 21:07:09 +0000 (14:07 -0700)]
Moving scale_factors and related code to separate files.

Change-Id: I531829e5aee2a4a7a112d528ecccbddf052d0e74

10 years agoMerge "Bug fix: call set_offsets before rd_auto_partition_range"
Scott LaVarnway [Fri, 9 Aug 2013 19:30:52 +0000 (12:30 -0700)]
Merge "Bug fix: call set_offsets before rd_auto_partition_range"

10 years agoMerge "Using buf_2d struct instead of separate buffer and stride vars."
Dmitry Kovalev [Fri, 9 Aug 2013 18:50:58 +0000 (11:50 -0700)]
Merge "Using buf_2d struct instead of separate buffer and stride vars."

10 years agoBug fix: call set_offsets before rd_auto_partition_range
Scott LaVarnway [Fri, 9 Aug 2013 18:09:49 +0000 (14:09 -0400)]
Bug fix: call set_offsets before rd_auto_partition_range

The set_offsets call is necessary inorder to set the
mode_info_context ptr correctly.

Change-Id: I644910cc5bacc50ee9cd78458843274ad8ee636d

10 years agoMerge "Correct bug in loopfilter initialization"
Adrian Grange [Fri, 9 Aug 2013 16:51:58 +0000 (09:51 -0700)]
Merge "Correct bug in loopfilter initialization"

10 years agoCorrect bug in loopfilter initialization
Adrian Grange [Fri, 9 Aug 2013 16:21:15 +0000 (09:21 -0700)]
Correct bug in loopfilter initialization

The memset sets 16 bytes rather than the correct size of the
final array dimension (MAX_MODE_LF_DELTAS).

(In response to bug posted by Manjit Hota to codec-devel
and webm-discuss lists)

Change-Id: I8980f5aa71ddc9d7ef57c5b4700bc28ddf8651c7

10 years agoAdded lpf level picking using partial frame
Yaowu Xu [Wed, 7 Aug 2013 19:29:45 +0000 (12:29 -0700)]
Added lpf level picking using partial frame

Change-Id: I599ab1bd22b5f3f10d5962c609952abdef8ff67a

10 years agorenamed vp8_yv12_copy_y to vpx_yv12_copy_y
Yaowu Xu [Wed, 7 Aug 2013 17:54:20 +0000 (10:54 -0700)]
renamed vp8_yv12_copy_y to vpx_yv12_copy_y

Becuase the routine is used by both vp8 and vp9

Change-Id: I2d35b287b5bc2394865d931a27da61f4ce7edeeb

10 years agoadded a speed feature on lpf level picking
Yaowu Xu [Tue, 6 Aug 2013 22:46:26 +0000 (15:46 -0700)]
added a speed feature on lpf level picking

Change-Id: Id578f8afdeab3702fc8386969f2d832d8f1b5420

10 years agoMerge "fix unit test failure on win32 vs2008 build"
Yaowu Xu [Fri, 9 Aug 2013 14:19:59 +0000 (07:19 -0700)]
Merge "fix unit test failure on win32 vs2008 build"

10 years agoUsing buf_2d struct instead of separate buffer and stride vars.
Dmitry Kovalev [Fri, 9 Aug 2013 04:25:48 +0000 (21:25 -0700)]
Using buf_2d struct instead of separate buffer and stride vars.

Change-Id: Id5cc3566cc16d1e3030ddb4d1c58459320321dca

10 years agoGeneral code cleanup.
Dmitry Kovalev [Thu, 8 Aug 2013 21:52:39 +0000 (14:52 -0700)]
General code cleanup.

Removing redundant parenthesis and curly braces. Combining declarations
with initializations. Adding useful intermediate variables instead of
recalculating expressions every time.

Change-Id: I00106f404afd60bfc189905b0fded881684f941a

10 years agofix unit test failure on win32 vs2008 build
Yaowu Xu [Fri, 9 Aug 2013 01:25:03 +0000 (18:25 -0700)]
fix unit test failure on win32 vs2008 build

The mix use of double type and simd code caused invalid values stored
in double variables, further caused unit tests to fail. The failures
were only observed on x86-win32-vs9 build with vs2008.

Change-Id: If0131754a3bf217a5ace303b7963e8f5162c34b5

10 years agoMerge "Cleanup inside vp9_reconinter.c."
Dmitry Kovalev [Thu, 8 Aug 2013 21:59:38 +0000 (14:59 -0700)]
Merge "Cleanup inside vp9_reconinter.c."

10 years agoMerge "Adds a new subpel motion function"
Deb Mukherjee [Thu, 8 Aug 2013 19:26:55 +0000 (12:26 -0700)]
Merge "Adds a new subpel motion function"

10 years agoMerge "Removing unneeded intermediate entropy_nodes_adapt var."
Dmitry Kovalev [Thu, 8 Aug 2013 19:16:57 +0000 (12:16 -0700)]
Merge "Removing unneeded intermediate entropy_nodes_adapt var."

10 years agoUsing MV struct instead of int[2] array.
Dmitry Kovalev [Thu, 8 Aug 2013 19:01:56 +0000 (12:01 -0700)]
Using MV struct instead of int[2] array.

Change-Id: Iab951c555037e36b154f319f351c5e67f9abb931

10 years agoRemoving unneeded intermediate entropy_nodes_adapt var.
Dmitry Kovalev [Thu, 8 Aug 2013 18:52:02 +0000 (11:52 -0700)]
Removing unneeded intermediate entropy_nodes_adapt var.

Change-Id: I541a178d997b4541e0e2d4d5b854e2ed6b113c3a

10 years agoAdds a new subpel motion function
Deb Mukherjee [Thu, 8 Aug 2013 00:01:43 +0000 (17:01 -0700)]
Adds a new subpel motion function

Adds a new subpel motion estimation function that uses a 2-level
tree-structured decision tree to eliminate redundant computations.
It searches fewer points than iterative search (which can search
the same point multiple times) but has the same quality roughly.

This is made the default setting at speeds 0 and 1, while at
speed 2 and above only a 1-level search is used.

Also includes various cleanups for consistency and redundancy removal.

Results:
derf: +0.012% psnr
stdhd: +0.09% psnr
Speedup of about 2-3%

Change-Id: Iedde4866f5475586dea0f0ba4cb7428fba24eee9

10 years agoMoved fast motion search level decision to function
Adrian Grange [Sat, 3 Aug 2013 00:19:29 +0000 (17:19 -0700)]
Moved fast motion search level decision to function

Moving this block of code into a function makes the
code easier to read and change.

Change-Id: If4ede570cce1eab1982b188c4d3e4fd3d4db236e

10 years agoSimplify & fix potential bug in rd_pick_partition
Adrian Grange [Thu, 1 Aug 2013 21:57:46 +0000 (14:57 -0700)]
Simplify & fix potential bug in rd_pick_partition

Different partitionings were not being evaluated against
best_rd and there were unnecessary calls to RDCOST. This
could have resulted in a non-optimal partioning being
selected.

I simplified the variables used to track the rate,
distortion and RD values throughout the function.

Change-Id: Ifa7085ee80d824e86791432a5bc6d8fea5a3e313

10 years agoMerge "Use low precision 32x32fdct for encodemb in speed1"
Jingning Han [Thu, 8 Aug 2013 02:05:14 +0000 (19:05 -0700)]
Merge "Use low precision 32x32fdct for encodemb in speed1"

10 years agoRemoving plane_block_{width, height}_log2by4 functions.
Dmitry Kovalev [Thu, 8 Aug 2013 00:06:33 +0000 (17:06 -0700)]
Removing plane_block_{width, height}_log2by4 functions.

Change-Id: I040b82b8e32aee272d10cbb021c7ba1c76343d7a

10 years agoCleanup inside vp9_reconinter.c.
Dmitry Kovalev [Thu, 8 Aug 2013 00:02:28 +0000 (17:02 -0700)]
Cleanup inside vp9_reconinter.c.

Using block width and block height instead of their logarithms. Using
SUBPEL_BITS and SUBPEL_SHIFTS constants instead of magic numbers.

Change-Id: I4e10e93c907c8a5e1cb27dfe74d1fcdcc4995448

10 years agoMerge "Using only one scale function in scale_factors struct."
Dmitry Kovalev [Wed, 7 Aug 2013 23:32:09 +0000 (16:32 -0700)]
Merge "Using only one scale function in scale_factors struct."

10 years agoMerge "Adding ss_size_lookup table."
Dmitry Kovalev [Wed, 7 Aug 2013 23:08:24 +0000 (16:08 -0700)]
Merge "Adding ss_size_lookup table."

10 years agoUse low precision 32x32fdct for encodemb in speed1
Jingning Han [Wed, 7 Aug 2013 22:22:51 +0000 (15:22 -0700)]
Use low precision 32x32fdct for encodemb in speed1

The low precision 32x32 fdct has all the intermediate steps within
16-bit depth, hence allowing faster SSE2 implementation, at the
expense of larger round-trip error. It was used in the rate-distortion
optimization search loop only.

Using the low precision version, in replace of the high precision one,
affects the compression performance by about 0.7% (derf, stdhd) at
speed 0. For speed 1, it makes derf set down by only 0.017%.

Change-Id: I4e7d18fac5bea5317b91c8e7dabae143bc6b5c8b

10 years agoAdding ss_size_lookup table.
Dmitry Kovalev [Wed, 7 Aug 2013 22:33:17 +0000 (15:33 -0700)]
Adding ss_size_lookup table.

Removing the old one bsize_from_dim_lookup. Now we have a way to determine
block size for plane using its subsampling values (ss_size_lookup). And
then we can find the number of pixels in the block (num_pels_log2_lookup).

Change-Id: I6fc981da2ae093de81741d3d78eaefed11015db9

10 years agoMerge "Removing NMS_STATS defines."
Dmitry Kovalev [Wed, 7 Aug 2013 22:28:30 +0000 (15:28 -0700)]
Merge "Removing NMS_STATS defines."

10 years agoNeon version of vp9_short_idct4x4_add.
Christian Duvivier [Mon, 5 Aug 2013 00:37:05 +0000 (17:37 -0700)]
Neon version of vp9_short_idct4x4_add.

Change-Id: Idec4cae0cb9b3a29835fd2750d354c1393d47aa4

10 years agoMerge "Clean ups of the subpel search functions"
Deb Mukherjee [Wed, 7 Aug 2013 00:28:48 +0000 (17:28 -0700)]
Merge "Clean ups of the subpel search functions"

10 years agoClean ups of the subpel search functions
Deb Mukherjee [Tue, 6 Aug 2013 22:53:35 +0000 (15:53 -0700)]
Clean ups of the subpel search functions

Removes some unused code and speed features, and organizes the
interfaces for fractional mv step functions for use in new speed
features to come.

In the process a new speed feature - number of iterations per
step during the subpel search - is exposed.

No change when this parameter is set as the original value of 3.

Results:
subpel_iters_per_step = 3: baseline
subpel_iters_per_step = 2: psnr -0.067%, 1% speedup
subpel_iters_per_step = 1: psnr -0.331%, 3-4% speedup

Change-Id: I2eba8a21f6461be8caf56af04a5337257a5693a8

10 years agoMerge "Motion vector code cleanup."
Dmitry Kovalev [Tue, 6 Aug 2013 23:00:01 +0000 (16:00 -0700)]
Merge "Motion vector code cleanup."

10 years agoUsing only one scale function in scale_factors struct.
Dmitry Kovalev [Tue, 6 Aug 2013 22:43:56 +0000 (15:43 -0700)]
Using only one scale function in scale_factors struct.

Functions scale_mv_q4 and scale_mv_q3_to_q4 were almost identical except
q3->q4 conversion in scale_mv_q3_to_q4. Now q3->q4 conversion happens
directly in vp9_build_inter_predictor.

Also adding useful constants: SUBPEL_BITS and SUBPEL_MASK.

Change-Id: Ia0a6ad2ac07c45fdf95a5139ece6286c035e9639

10 years agoMerge "Place holder for high-precision 32x32 fdct"
Jingning Han [Tue, 6 Aug 2013 21:47:30 +0000 (14:47 -0700)]
Merge "Place holder for high-precision 32x32 fdct"

10 years agovariance x86inc guards
Jim Bankoski [Tue, 6 Aug 2013 15:25:10 +0000 (08:25 -0700)]
variance x86inc guards

also fixed bug in sad calcs

Change-Id: I6571fcbe37556c16ae32be66dc0fd879852aac1d

10 years agosse3 intrapred x86inc protected
Jim Bankoski [Tue, 6 Aug 2013 14:44:56 +0000 (07:44 -0700)]
sse3 intrapred x86inc protected

Change-Id: I4a3c83119cdf8a205920034c8019d855d5504605

10 years agoMerge "Flexible support for various pattern searches"
Deb Mukherjee [Tue, 6 Aug 2013 21:03:27 +0000 (14:03 -0700)]
Merge "Flexible support for various pattern searches"

10 years agosad + miscellaneous updates
Jim Bankoski [Tue, 6 Aug 2013 14:31:54 +0000 (07:31 -0700)]
sad + miscellaneous updates

Enable use_x86inc as a commandline option.  Fix Bug with sse2 when
x86inc is disabled. Adds Sad asm protection to x86inc protection

Change-Id: Iee0f9dd235ea10e8ace512eb362ba9bebe8c9df6

10 years agoMerge "Inlining vp9_get_pred_probs_switchable_interp function."
Dmitry Kovalev [Tue, 6 Aug 2013 18:57:45 +0000 (11:57 -0700)]
Merge "Inlining vp9_get_pred_probs_switchable_interp function."

10 years agoFlexible support for various pattern searches
Deb Mukherjee [Mon, 22 Jul 2013 21:47:57 +0000 (14:47 -0700)]
Flexible support for various pattern searches

Adds a few pattern searches to achieve various tradeoffs
between motion estimation complexity and performance.
The search framework is unified across these searches so that a
common pattern search function is used for all. Besides it will
be easier to experiment with various patterns or combinations
thereof at different scales in the future.

The new pattern search is multi-scale and is capable of using
different patterns at different scales.

The new hex search uses 8 points at the smallest scale
and 6 points at other scales.
Two other pattern searches - big-diamond and square are
also added. Big diamond uses 4 points at the smallest scale and
8 points in diamond shape at the larger scales.
Square is very similar conceptually to the default n-step search
but is somewhat faster since it keeps only one survivor across
all scales.

Psnr/speed-up results on derf300:

hex: -1.6% psnr%, 6-8% speed-up
big-diamond: -0.96% psnr, 4-5% speedup
square: -0.93% psnr, 4-5% speedup

Change-Id: I02a7ef5193f762601e0994e2c99399a3535a43d2

10 years agoPlace holder for high-precision 32x32 fdct
Jingning Han [Tue, 6 Aug 2013 18:10:12 +0000 (11:10 -0700)]
Place holder for high-precision 32x32 fdct

Resolve compile warnings on re-define FDCT32x32_2D template.

Change-Id: Idb3a54ef8d2710ce7245b726379a0e5c875f5cad

10 years agoInlining vp9_get_pred_probs_switchable_interp function.
Dmitry Kovalev [Tue, 6 Aug 2013 18:04:31 +0000 (11:04 -0700)]
Inlining vp9_get_pred_probs_switchable_interp function.

There was no benefit having this function. For example, inside
read_switchable_filter_type switchable filter context was calculated twice.

Change-Id: I79cd5bf95cbc0f6d8bf91a2e32289e01b18dcff1

10 years agoMerge "Move fdct32x32 SSE2 implementation in separate file."
Jingning Han [Tue, 6 Aug 2013 17:46:41 +0000 (10:46 -0700)]
Merge "Move fdct32x32 SSE2 implementation in separate file."

10 years agoMerge "intrapred x86inc guards"
Jim Bankoski [Tue, 6 Aug 2013 17:39:19 +0000 (10:39 -0700)]
Merge "intrapred x86inc guards"

10 years agoMotion vector code cleanup.
Dmitry Kovalev [Mon, 5 Aug 2013 23:27:51 +0000 (16:27 -0700)]
Motion vector code cleanup.

Converting arguments of two functions (clamp_mv_ref, lower_mv_precision)
from int_mv* to MV*. Rewriting is_inside function to make it much shorter.

Change-Id: Ie4c4cf3eccd46707c7df099ec21fb1b61c72fc7a

10 years agoMerge "Finally removing all old block size constants."
Dmitry Kovalev [Tue, 6 Aug 2013 17:30:37 +0000 (10:30 -0700)]
Merge "Finally removing all old block size constants."

10 years agoMerge "Changing the order switchable filter enum constants."
Dmitry Kovalev [Tue, 6 Aug 2013 17:30:26 +0000 (10:30 -0700)]
Merge "Changing the order switchable filter enum constants."

10 years agoMerge "Removing unused functions."
Dmitry Kovalev [Tue, 6 Aug 2013 17:29:57 +0000 (10:29 -0700)]
Merge "Removing unused functions."

10 years agoMerge "Add variance based mode/skipping"
Deb Mukherjee [Tue, 6 Aug 2013 17:19:15 +0000 (10:19 -0700)]
Merge "Add variance based mode/skipping"

10 years agoMove fdct32x32 SSE2 implementation in separate file.
Christian Duvivier [Mon, 5 Aug 2013 22:22:13 +0000 (15:22 -0700)]
Move fdct32x32 SSE2 implementation in separate file.

This is in preparation for the SSE2 version of the high-precision
32x32 forward DCT which will share a lot of code with the existing
low precision version used for rate-distortion search.

Change-Id: I7084b6bdfb480b1fabb8493fb14e3f7fcc7888c0

10 years agointrapred x86inc guards
Jim Bankoski [Tue, 6 Aug 2013 16:39:30 +0000 (09:39 -0700)]
intrapred x86inc guards

Change-Id: If0399d8e11f4ebe75a5c91abb8d6a52a7709065b