platform/upstream/libvpx.git
11 years agoAdd unit test for x4 multi-SAD functions
John Koleszar [Fri, 1 Mar 2013 20:43:41 +0000 (12:43 -0800)]
Add unit test for x4 multi-SAD functions

Update the function prototypes to match between VP9 and VP8.

Change-Id: If58965073989e87df3b62b67a030ec6ce23ca04f

11 years agoAdd VP9 1 block SAD functions to unit test
John Koleszar [Fri, 1 Mar 2013 01:03:02 +0000 (17:03 -0800)]
Add VP9 1 block SAD functions to unit test

Change-Id: I06b5ba5c457944cfa4cd9f53c3bd8cda132439c2

11 years agoMerge master branch into experimental
John Koleszar [Fri, 1 Mar 2013 00:25:38 +0000 (16:25 -0800)]
Merge master branch into experimental

Picks up some build system changes, compiler warning fixes, etc.

Change-Id: I2712f99e653502818a101a72696ad54018152d4e

11 years agoMerge "Adjust the max_gf_interval initialization" into experimental
Yaowu Xu [Fri, 1 Mar 2013 19:02:23 +0000 (11:02 -0800)]
Merge "Adjust the max_gf_interval initialization" into experimental

11 years agoMerge "Add eob<=10 case in idct32x32" into experimental
Yunqing Wang [Fri, 1 Mar 2013 16:58:19 +0000 (08:58 -0800)]
Merge "Add eob<=10 case in idct32x32" into experimental

11 years agoAdjust the max_gf_interval initialization
Yaowu Xu [Thu, 28 Feb 2013 17:14:47 +0000 (09:14 -0800)]
Adjust the max_gf_interval initialization

to be a fixed value of 15.

Test results:
cif:  .124%, .068%, .081%
std-hd: 2.809%, 3.174%, 2.705%

Change-Id: I380c8152c973506094da15eab59e3aa22b75a983

11 years agoMerge "Code cleanup." into experimental
Dmitry Kovalev [Fri, 1 Mar 2013 01:22:50 +0000 (17:22 -0800)]
Merge "Code cleanup." into experimental

11 years agoAdd eob<=10 case in idct32x32
Yunqing Wang [Fri, 1 Mar 2013 00:32:05 +0000 (16:32 -0800)]
Add eob<=10 case in idct32x32

Simplified idct32x32 calculation when there are only 10 or less
non-zero coefficients in 32x32 block. This helps the decoder
performance.

Change-Id: If7f8893d27b64a9892b4b2621a37fdf4ac0c2a6d

11 years agoMerge changes I9be9c990,Ic3b97339 into experimental
Dmitry Kovalev [Fri, 1 Mar 2013 00:07:45 +0000 (16:07 -0800)]
Merge changes I9be9c990,Ic3b97339 into experimental

* changes:
  Ignoring test video sequences in the source tree.
  Code cleanup.

11 years agofirstpass.c: correct casting around gf_group_bits
James Zern [Thu, 28 Feb 2013 23:45:29 +0000 (15:45 -0800)]
firstpass.c: correct casting around gf_group_bits

gf_group_bits is int64_t remove casts to int.

Change-Id: I3b4225905041fac9af9fdfcbcb6f1c357ea4b593

11 years agoMerge "Fix use of uninitialized memory in CONFIG_ABOVESPREFMV" into experimental
John Koleszar [Thu, 28 Feb 2013 23:18:50 +0000 (15:18 -0800)]
Merge "Fix use of uninitialized memory in CONFIG_ABOVESPREFMV" into experimental

11 years agoMerge "mv dct_sse2.c dct_sse2_intrinsics.c to avoid collision" into experimental
Jim Bankoski [Thu, 28 Feb 2013 23:16:43 +0000 (15:16 -0800)]
Merge "mv dct_sse2.c dct_sse2_intrinsics.c to avoid collision" into experimental

11 years agoIgnoring test video sequences in the source tree.
Dmitry Kovalev [Thu, 28 Feb 2013 23:10:48 +0000 (15:10 -0800)]
Ignoring test video sequences in the source tree.

Adding /*.ivf and /*.ivf.md5 filename patterns to .gitignore.

Change-Id: I9be9c99023e6b59bcc9cfbb6c9d6cdc8e500fc4f

11 years agoCode cleanup.
Dmitry Kovalev [Thu, 28 Feb 2013 22:40:20 +0000 (14:40 -0800)]
Code cleanup.

Lower case variable names, converting while loops to for loops.

Change-Id: Ic3b973391eef7472a99d18d02fe79cfef5e04e62

11 years agoMerge "Refactor vp9_dequant_idct_add function" into experimental
Yunqing Wang [Thu, 28 Feb 2013 22:34:27 +0000 (14:34 -0800)]
Merge "Refactor vp9_dequant_idct_add function" into experimental

11 years agoRefactor vp9_dequant_idct_add function
Yunqing Wang [Thu, 28 Feb 2013 21:01:41 +0000 (13:01 -0800)]
Refactor vp9_dequant_idct_add function

Provided a wrapper and removed duplicate code.

Change-Id: Iaef842226ec348422e459202793b001d0983ea30

11 years agoRemoved vp9_dequantize_b
Scott LaVarnway [Thu, 28 Feb 2013 19:05:00 +0000 (11:05 -0800)]
Removed vp9_dequantize_b

Change-Id: Ie89bd00d58e30bf4094cb748a282f1dfa81a31d8

11 years agomv dct_sse2.c dct_sse2_intrinsics.c to avoid collision
Jim Bankoski [Thu, 28 Feb 2013 21:58:15 +0000 (13:58 -0800)]
mv dct_sse2.c dct_sse2_intrinsics.c to avoid collision

Change-Id: Id786be31da3c91d95d2955aa569ecdc6e66650df

11 years agoFix use of uninitialized memory in CONFIG_ABOVESPREFMV
John Koleszar [Thu, 28 Feb 2013 21:45:47 +0000 (13:45 -0800)]
Fix use of uninitialized memory in CONFIG_ABOVESPREFMV

The ABOVESPREFMV experiment uses four pixels to the left of the
current block, which don't exist for the left-most column.

Change-Id: I4cf0b42ae8f54c0b3e7b1ed8755704b74fafc39c

11 years agoMerge "Dequantization code cleanup." into experimental
Dmitry Kovalev [Thu, 28 Feb 2013 21:46:43 +0000 (13:46 -0800)]
Merge "Dequantization code cleanup." into experimental

11 years agoDequantization code cleanup.
Dmitry Kovalev [Thu, 28 Feb 2013 21:28:05 +0000 (13:28 -0800)]
Dequantization code cleanup.

Removing redundant variables, using x *= y instead x = x * y, moving
variable declarations into inner blocks.

Change-Id: I884f95c755f55d51b7c1c6585f10296919063e41

11 years agoCode cleanup.
Dmitry Kovalev [Thu, 28 Feb 2013 21:18:02 +0000 (13:18 -0800)]
Code cleanup.

Removing redundant 'extern' keyword, better formatting, code
simplification.

Change-Id: I132fea14f08c706ee9ea147d19464d03f833f25b

11 years agoFix incorrect comparison of frame size
John Koleszar [Thu, 28 Feb 2013 18:52:04 +0000 (10:52 -0800)]
Fix incorrect comparison of frame size

The width and height stored in the reference frames are padded out to
a multiple of 16. The Width and Height variables in common are the
displayed size, which may be smaller. The incorrect comparison was
causing scaling related code to be called when it shouldn't have
been. A notable case where this happens is 1080p, since 1088 != 1080.

Change-Id: I55f743eeeeaefbf2e777e193bc9a77ff726e16b5

11 years agoMerge "vp8 fast quantizer with intrinsics"
Johann [Thu, 28 Feb 2013 19:32:03 +0000 (11:32 -0800)]
Merge "vp8 fast quantizer with intrinsics"

11 years agothis commit converts all sad ptrs to uint32
Jim Bankoski [Thu, 28 Feb 2013 16:32:14 +0000 (08:32 -0800)]
this commit converts all sad ptrs to uint32

sse4_1 code used uint16_t for returning sad, but that
won't work for 32x32 or 64x64.   This code fixes the
assembly for those and also reenables sse4_1 on linux

Change-Id: I5ce7288d581db870a148e5f7c5092826f59edd81

11 years agofix to parameters to match rtcd
Jim Bankoski [Thu, 28 Feb 2013 16:10:08 +0000 (08:10 -0800)]
fix to parameters to match rtcd

Change-Id: I919e2dd72292fe44f2e53ada56bd42287d50cdeb
Signed-off-by: Jim Bankoski <jimbankoski@google.com>
11 years agoMerge "Faster vp9_short_fdct8x8." into experimental
Christian Duvivier [Thu, 28 Feb 2013 01:50:27 +0000 (17:50 -0800)]
Merge "Faster vp9_short_fdct8x8." into experimental

11 years agoMerge "Fix rollover and pass 1 time estimate" into experimental
John Koleszar [Thu, 28 Feb 2013 01:43:03 +0000 (17:43 -0800)]
Merge "Fix rollover and pass 1 time estimate" into experimental

11 years agoFaster vp9_short_fdct8x8.
Christian Duvivier [Wed, 27 Feb 2013 20:29:06 +0000 (12:29 -0800)]
Faster vp9_short_fdct8x8.

Scalar path is about 1.4x faster (4% overall encoder speedup).
SSE2 path is about 7x faster (13% overall encoder speedup).

Change-Id: I7e85d8225a914a74c61ea370210414696560094d

11 years agoMerge "Code cleanup." into experimental
Dmitry Kovalev [Thu, 28 Feb 2013 00:43:04 +0000 (16:43 -0800)]
Merge "Code cleanup." into experimental

11 years agoCode cleanup.
Dmitry Kovalev [Wed, 27 Feb 2013 19:17:38 +0000 (11:17 -0800)]
Code cleanup.

Fixing code style, using array lookup instead of switch statements for
forward hybrid transforms (in the same way as for their inverses).
Consistent usage of ROUND_POWER_OF_TWO macro in appropriate places.

Change-Id: I0d3822ae11f928905fdbfbe4158f91d97c71015f

11 years agoMerge changes Idc1c490f,I6b5fe1a4 into experimental
John Koleszar [Wed, 27 Feb 2013 21:45:42 +0000 (13:45 -0800)]
Merge changes Idc1c490f,I6b5fe1a4 into experimental

* changes:
  convolve test: validate 1D filters are 1D
  Run all filters through convolve test

11 years agoMerge "Motion vectors code cleanup." into experimental
Dmitry Kovalev [Wed, 27 Feb 2013 21:34:56 +0000 (13:34 -0800)]
Merge "Motion vectors code cleanup." into experimental

11 years agoMerge "Remove unused file" into experimental
Yunqing Wang [Wed, 27 Feb 2013 21:00:10 +0000 (13:00 -0800)]
Merge "Remove unused file" into experimental

11 years agoFix rollover and pass 1 time estimate
John Koleszar [Wed, 27 Feb 2013 19:14:23 +0000 (11:14 -0800)]
Fix rollover and pass 1 time estimate

Fixes a rollover of the cx_time variable for encodes that take
over ~4200 seconds. Also enable the time estimate in first pass.

Change-Id: Ib5a98ee71bccd79a804d709cec7260651d0b7141

11 years agoMerge "Remove unused vp9_copy32xn" into experimental
John Koleszar [Wed, 27 Feb 2013 20:23:45 +0000 (12:23 -0800)]
Merge "Remove unused vp9_copy32xn" into experimental

11 years agoMerge "Remove unused file" into experimental
Yunqing Wang [Wed, 27 Feb 2013 19:58:28 +0000 (11:58 -0800)]
Merge "Remove unused file" into experimental

11 years agoMotion vectors code cleanup.
Dmitry Kovalev [Wed, 27 Feb 2013 19:48:13 +0000 (11:48 -0800)]
Motion vectors code cleanup.

Fixing indentation, removing redundant parenthesis, deciphering single
letter variable names, better spacing.

Change-Id: I1d447a7d69eddbf1e94e0820423615f40ea2d591

11 years agoMerge "Move eob from BLOCKD to MACROBLOCKD." into experimental
Ronald S. Bultje [Wed, 27 Feb 2013 19:39:16 +0000 (11:39 -0800)]
Merge "Move eob from BLOCKD to MACROBLOCKD." into experimental

11 years agoMerge "Optimize vp9_dc_only_idct_add_c function" into experimental
Yunqing Wang [Wed, 27 Feb 2013 19:38:45 +0000 (11:38 -0800)]
Merge "Optimize vp9_dc_only_idct_add_c function" into experimental

11 years agoMerge "vp8/encoder/mcomp.c: remove an unused variable"
James Zern [Wed, 27 Feb 2013 19:33:18 +0000 (11:33 -0800)]
Merge "vp8/encoder/mcomp.c: remove an unused variable"

11 years agoMerge "add vp8 variance test"
John Koleszar [Wed, 27 Feb 2013 19:22:46 +0000 (11:22 -0800)]
Merge "add vp8 variance test"

11 years agoMerge "give vp9 variance struct a unique name"
John Koleszar [Wed, 27 Feb 2013 19:22:36 +0000 (11:22 -0800)]
Merge "give vp9 variance struct a unique name"

11 years agoMerge "rtcd: make include guard unique"
John Koleszar [Wed, 27 Feb 2013 19:22:26 +0000 (11:22 -0800)]
Merge "rtcd: make include guard unique"

11 years agoconvolve test: validate 1D filters are 1D
John Koleszar [Thu, 21 Feb 2013 00:32:02 +0000 (16:32 -0800)]
convolve test: validate 1D filters are 1D

Since the 8-tap lowpass filter is non-interpolating, the results are
different between applying it at whole-pel values and not. This
means that 1D-only versions are requried to be implemented, as
opposed to being an optimization of the 2D case. Calling the 2D
filter instead of the horizontal-only filter is not equivalent
in this case. Update the test to pass invalid filters to the
unused stage of the 1D-only calls, to verify they're unused.

Change-Id: Idc1c490f059adadd4cc80dbe770c1ccefe628b0a

11 years agoRun all filters through convolve test
John Koleszar [Thu, 21 Feb 2013 00:13:01 +0000 (16:13 -0800)]
Run all filters through convolve test

Updates the convolve test to verify that all filters match the
reference implementation. This verifies commit 30f866f, which
fixed some problems with the SSE3 version of the filters for
the vp9_sub_pel_filters_8s and vp9_sub_pel_filters_8lp banks
due to overflow and order of operations.

Change-Id: I6b5fe1a41bc20062e2e64633b1355ae58c9c592c

11 years agoRemove unused file
Yunqing Wang [Wed, 27 Feb 2013 19:13:19 +0000 (11:13 -0800)]
Remove unused file

Removed vp9/decoder/x86/vp9_idct_blk_mmx.c

Change-Id: I07ab06382a394cf556fa5a8e3c98b91f6e4f9ce8

11 years agoMerge "vpxdec: support scaling output" into experimental
John Koleszar [Wed, 27 Feb 2013 19:09:56 +0000 (11:09 -0800)]
Merge "vpxdec: support scaling output" into experimental

11 years agoRemove unused file
Yunqing Wang [Wed, 27 Feb 2013 19:00:58 +0000 (11:00 -0800)]
Remove unused file

Removed vp9_idctllm_mmx.asm

Change-Id: I7152756f23a5a09ed69e8fb40edb2ab3237290fe

11 years agoMove eob from BLOCKD to MACROBLOCKD.
Ronald S. Bultje [Wed, 27 Feb 2013 18:00:24 +0000 (10:00 -0800)]
Move eob from BLOCKD to MACROBLOCKD.

Consistent with VP8.

Change-Id: I8c316ee49f072e15abbb033a80e9c36617891f07

11 years agoMerge "vpxenc: support scaling prior to encoding" into experimental
John Koleszar [Wed, 27 Feb 2013 18:59:21 +0000 (10:59 -0800)]
Merge "vpxenc: support scaling prior to encoding" into experimental

11 years agoMerge "Use ref_frame_map vice active_ref_idx on the encoder" into experimental
John Koleszar [Wed, 27 Feb 2013 18:59:08 +0000 (10:59 -0800)]
Merge "Use ref_frame_map vice active_ref_idx on the encoder" into experimental

11 years agoMerge "Test upscaling as well as downscaling" into experimental
John Koleszar [Wed, 27 Feb 2013 18:25:51 +0000 (10:25 -0800)]
Merge "Test upscaling as well as downscaling" into experimental

11 years agoRemove unused vp9_copy32xn
John Koleszar [Wed, 27 Feb 2013 18:23:06 +0000 (10:23 -0800)]
Remove unused vp9_copy32xn

This function was part of an optimization used in VP8 that required
caching two macroblocks. This is unused in VP9, and might not
survive refactoring to support superblocks, so removing it for now.

Change-Id: I744e585206ccc1ef9a402665c33863fc9fb46f0d

11 years agoMerge changes I922f8602,I0ac3343d into experimental
John Koleszar [Wed, 27 Feb 2013 18:08:53 +0000 (10:08 -0800)]
Merge changes I922f8602,I0ac3343d into experimental

* changes:
  Use 256-byte aligned filter tables
  Set scale factors consistently for SPLITMV

11 years agoFix --as=nasm compatibility for new asm code.
Jan Kratochvil [Wed, 27 Feb 2013 17:51:59 +0000 (09:51 -0800)]
Fix --as=nasm compatibility for new asm code.

s/movd/movq/

Change-Id: Id1a56de91551f8dc796f14f1056c565dfc1ba626

11 years agoMerge "Combined motion compensation with scaled predictors" into experimental
John Koleszar [Wed, 27 Feb 2013 17:46:12 +0000 (09:46 -0800)]
Merge "Combined motion compensation with scaled predictors" into experimental

11 years agovpxdec: support scaling output
John Koleszar [Tue, 26 Feb 2013 00:58:47 +0000 (16:58 -0800)]
vpxdec: support scaling output

Adds an option, --scale, that will rescale any frames produced by
the decoder that don't match the resolution of the first frame to
that resolution. This is useful for playback of files that use
spatial resampling.

Change-Id: I867adc650d535df7ec5b107549359712ea3aaaa0

11 years agovpxenc: support scaling prior to encoding
John Koleszar [Thu, 1 Mar 2012 20:50:40 +0000 (12:50 -0800)]
vpxenc: support scaling prior to encoding

Scales the input of the encoder using libyuv's "box filter". Each stream
may have a different width and height specified. If the width (or
height) parameter is missing (or is explicitly set to 0) then the value
will be calculated based on the specified height (or width) and the
input file's dimensions, preserving its aspect ratio. Leaving the height
unspecified behaves similarly.

Change-Id: Ic7026810b13be030826be80dc6f7fc4aaf0c35d0

11 years agoUse ref_frame_map vice active_ref_idx on the encoder
John Koleszar [Fri, 22 Feb 2013 19:22:03 +0000 (11:22 -0800)]
Use ref_frame_map vice active_ref_idx on the encoder

This patch makes the encoder's use of ref_frame_map and active_ref_idx
consistent with the decoder. ref_frame_map[] maps a reference buffer
index to its actual location in the yv12_fb array, since many
references may share an underlying buffer. active_ref_idx[] mirrors
cpi->{lst,gld,alt}_fb_idx, holding the active references in each
slot.

This also fixes a bug in setup_buffer_inter() where the incorrect
reference was used to populate the scaling factors.

Change-Id: Id3728f6d77cffcd27c248903bf51f9c3e594287e

11 years agoTest upscaling as well as downscaling
John Koleszar [Thu, 21 Feb 2013 18:38:27 +0000 (10:38 -0800)]
Test upscaling as well as downscaling

Fixes a bug in vp9_set_internal_size() that prevented returning to
the unscaled state. Updated the ResizeInternalTest to scale both
down and up. Added a check that all frames are within 2.5% of the
quality of the initial keyframe.

Change-Id: I3b7ef17cdac144ed05b9148dce6badfa75cff5c8

11 years agoUse 256-byte aligned filter tables
John Koleszar [Wed, 20 Feb 2013 23:59:20 +0000 (15:59 -0800)]
Use 256-byte aligned filter tables

This avoids duplicating all the filters twice. Includes fixups to the
convolve routines and associated tests to make this work.

Change-Id: I922f86021594e55072ddb63b42b2313605db6e00

11 years agoCombined motion compensation with scaled predictors
John Koleszar [Wed, 20 Feb 2013 20:34:31 +0000 (12:34 -0800)]
Combined motion compensation with scaled predictors

This patch extends the previous support for using references of a
different resolution in ZEROMV mode to all inter prediction modes.
Subpixel based best-mv scoring is disabled when the reference frame
differs in resolution from the current frame.

Change-Id: Id4dc3e5e6692de98d9857fd56bfad3ac57e944ac

11 years agoSet scale factors consistently for SPLITMV
John Koleszar [Wed, 20 Feb 2013 21:46:55 +0000 (13:46 -0800)]
Set scale factors consistently for SPLITMV

This commit updates the 4x4 prediction to consistently use the
build_2x1_inter_predictor() method. That function is updated to
calculate the scale offset, rather than relying on the caller
to calculate it. In the case that the 2x1 prediction can not
be used, the scale offset is recalculated for each 1x1 block.
The idea here is that the offsets are calculated before each
call to vp9_build_scaled_inter_predictor().

Change-Id: I0ac3343dd54e2846efa3c4195fcd328b709ca04d

11 years agoMerge "Enable 32x32 dct tests" into experimental
Yaowu Xu [Wed, 27 Feb 2013 15:57:07 +0000 (07:57 -0800)]
Merge "Enable 32x32 dct tests" into experimental

11 years agoMerge "Improve 32x32 forward dct" into experimental
Yaowu Xu [Wed, 27 Feb 2013 15:56:42 +0000 (07:56 -0800)]
Merge "Improve 32x32 forward dct" into experimental

11 years agoSpatial resamping of ZEROMV predictors
John Koleszar [Mon, 25 Feb 2013 04:55:14 +0000 (20:55 -0800)]
Spatial resamping of ZEROMV predictors

This patch allows coding frames using references of different
resolution, in ZEROMV mode. For compound prediction, either
reference may be scaled.

To test, I use the resize_test and enable WRITE_RECON_BUFFER
in vp9_onyxd_if.c. It's also useful to apply this patch to
test/i420_video_source.h:

  --- a/test/i420_video_source.h
  +++ b/test/i420_video_source.h
  @@ -93,6 +93,7 @@ class I420VideoSource : public VideoSource {

     virtual void FillFrame() {
       // Read a frame from input_file.
  +    if (frame_ != 3)
       if (fread(img_->img_data, raw_sz_, 1, input_file_) == 0) {
         limit_ = frame_;
       }

This forces the frame that the resolution changes on to be coded
with no motion, only scaling, and improves the quality of the
result.

Change-Id: I1ee75d19a437ff801192f767fd02a36bcbd1d496

11 years agoMerge "Removing redundant 'extern' keyword from function declarations." into experimental
Dmitry Kovalev [Wed, 27 Feb 2013 04:56:32 +0000 (20:56 -0800)]
Merge "Removing redundant 'extern' keyword from function declarations." into experimental

11 years agoMerge "Merge cnvcontext experiment." into experimental
Ronald S. Bultje [Wed, 27 Feb 2013 03:39:39 +0000 (19:39 -0800)]
Merge "Merge cnvcontext experiment." into experimental

11 years agoMerge "Fix modes.stt output printf format string." into experimental
Ronald S. Bultje [Wed, 27 Feb 2013 03:39:33 +0000 (19:39 -0800)]
Merge "Fix modes.stt output printf format string." into experimental

11 years agoMerge "Minor cosmetics in rdopt." into experimental
Ronald S. Bultje [Wed, 27 Feb 2013 03:39:28 +0000 (19:39 -0800)]
Merge "Minor cosmetics in rdopt." into experimental

11 years agoOptimize vp9_dc_only_idct_add_c function
Yunqing Wang [Wed, 27 Feb 2013 00:27:41 +0000 (16:27 -0800)]
Optimize vp9_dc_only_idct_add_c function

Wrote SSE2 version of vp9_dc_only_idct_add_c function. In order to
improve performance, clipped the absolute diff values to [0, 255].
This allowed us to keep the additions/subtractions in 8 bits.
Test showed an over 2% decoder performance increase.

Change-Id: Ie1a236d23d207e4ffcd1fc9f3d77462a9c7fe09d

11 years agoMerge "vp9: promote gf_group_bits calculation to 64-bit" into experimental
James Zern [Wed, 27 Feb 2013 00:27:45 +0000 (16:27 -0800)]
Merge "vp9: promote gf_group_bits calculation to 64-bit" into experimental

11 years agoRemoving redundant 'extern' keyword from function declarations.
Dmitry Kovalev [Tue, 26 Feb 2013 23:52:05 +0000 (15:52 -0800)]
Removing redundant 'extern' keyword from function declarations.

Change-Id: I893fa36297b9bd9cff93d082f1736f6860b15c0d

11 years agoMerge "Refactor inter recon functions to support scaling" into experimental
John Koleszar [Tue, 26 Feb 2013 19:45:28 +0000 (11:45 -0800)]
Merge "Refactor inter recon functions to support scaling" into experimental

11 years agovp8 fast quantizer with intrinsics
Johann [Sat, 2 Feb 2013 00:14:38 +0000 (16:14 -0800)]
vp8 fast quantizer with intrinsics

Reduce dependency on offsets file by using intrinsics. Disassembly shows
improvements over previous assembly specifically in register management,
preloading, and {pro,epi}log. Speed change is within margin of error.

Change-Id: I8131b4b4d62bc092407fe847bfaa8f2c0e1384ff

11 years agoMerge "Changing pitch value meaning for fht and iht transforms." into experimental
Dmitry Kovalev [Tue, 26 Feb 2013 18:44:15 +0000 (10:44 -0800)]
Merge "Changing pitch value meaning for fht and iht transforms." into experimental

11 years agoMerge cnvcontext experiment.
Ronald S. Bultje [Tue, 26 Feb 2013 18:37:49 +0000 (10:37 -0800)]
Merge cnvcontext experiment.

Change-Id: I35e64998b25694a3bb4a62164bba3c03c1db4bc7

11 years agoFix modes.stt output printf format string.
Ronald S. Bultje [Tue, 26 Feb 2013 17:38:29 +0000 (09:38 -0800)]
Fix modes.stt output printf format string.

Change-Id: I17e2d2f6a4da86d9e4af7bebdea0bf5d154da084

11 years agoMinor cosmetics in rdopt.
Ronald S. Bultje [Tue, 26 Feb 2013 17:37:51 +0000 (09:37 -0800)]
Minor cosmetics in rdopt.

Change-Id: I62497dcf2074b4bb4787bf660e727e5cf1bf3472

11 years agoMerge "make cost_coeffs to use combined context" into experimental
Ronald S. Bultje [Tue, 26 Feb 2013 18:32:01 +0000 (10:32 -0800)]
Merge "make cost_coeffs to use combined context" into experimental

11 years agoRefactor inter recon functions to support scaling
John Koleszar [Sat, 9 Feb 2013 01:49:44 +0000 (17:49 -0800)]
Refactor inter recon functions to support scaling

Ensure that all inter prediction goes through a common code path
that takes scaling into account. Removes a bunch of duplicate
1st/2nd predictor code. Also introduces a 16x8 mode for 8x8
MVs, similar to the 8x4 trick we were doing before. This has an
unexpected effect with EIGHTTAP_SMOOTH, so it's disabled in that
case for now.

Change-Id: Ia053e823a8bc616a988a0af30452e1e75a739cba

11 years agoEnable 32x32 dct tests
Yaowu Xu [Tue, 26 Feb 2013 17:07:49 +0000 (09:07 -0800)]
Enable 32x32 dct tests

Also
1. Removed the test code for fDCT from the iDCT test.
2. changed the criteria of round trip error to be below 1/block, this
is quite strict comparing to smaller transforms when size differences
are accounted for.

Change-Id: Idb46a6380b04c93fc8e2845c75f5a850366b0090

11 years agoImprove 32x32 forward dct
Yaowu Xu [Mon, 25 Feb 2013 23:21:01 +0000 (15:21 -0800)]
Improve 32x32 forward dct

The commit improves the 32x32 forward dct implementation:
1. change to use same constants and rounding as other forward dcts
2. select rounding to specifically minimize the roundtrip error, which
improved average 19/block to .77/block using 100000 random input.

Test showed a small but consistent gain on all test sets, about .15%

Change-Id: If0afd6a71880a522f60c1c234be0462092c2eb53

11 years agoChanging pitch value meaning for fht and iht transforms.
Dmitry Kovalev [Tue, 26 Feb 2013 02:19:55 +0000 (18:19 -0800)]
Changing pitch value meaning for fht and iht transforms.

Pitch now means the number of elements, not the number of bytes.

Change-Id: Idb9f2f012e39b09d596a3cc1802305a80b7c13af

11 years agomake cost_coeffs to use combined context
Yaowu Xu [Tue, 26 Feb 2013 01:01:30 +0000 (17:01 -0800)]
make cost_coeffs to use combined context

Change-Id: Ia15f4244595fab49bffda0c651a750a8a9481d28

11 years agoCode cleanup.
Dmitry Kovalev [Mon, 25 Feb 2013 23:14:01 +0000 (15:14 -0800)]
Code cleanup.

Removing switch statements for inverse hybrid transforms. Making code style
consistent for all similar transform implementations. Renaming shortpitch
and short_pitch variables to half_pitch.

Change-Id: I875f7a82aae4e8063a58777bf1cc3f1e67b48582

11 years agoMerge "Code cleanup." into experimental
Dmitry Kovalev [Mon, 25 Feb 2013 22:14:22 +0000 (14:14 -0800)]
Merge "Code cleanup." into experimental

11 years agoMerge "Code cleanup." into experimental
Dmitry Kovalev [Mon, 25 Feb 2013 21:58:06 +0000 (13:58 -0800)]
Merge "Code cleanup." into experimental

11 years agoMerge "Improving the forward 16x16 ADST/DCT accuracy" into experimental
Jingning Han [Mon, 25 Feb 2013 21:38:33 +0000 (13:38 -0800)]
Merge "Improving the forward 16x16 ADST/DCT accuracy" into experimental

11 years agoCode cleanup.
Dmitry Kovalev [Mon, 25 Feb 2013 21:38:18 +0000 (13:38 -0800)]
Code cleanup.

Removing redundant parentheses, better code formatting, introducing
ROUND_POWER_OF_TWO macro to replace repeated expression.

Change-Id: I91aad7a53ed03482428b2419de4bb99fd92c6771

11 years agoCode cleanup.
Dmitry Kovalev [Sat, 23 Feb 2013 01:27:34 +0000 (17:27 -0800)]
Code cleanup.

Lower case names of variables. Removing redundant spaces, parentheses,
casts, and variables.

Change-Id: I55b80c55b7d5adca44c1e8adb40a124c0680f229

11 years agovp9: promote gf_group_bits calculation to 64-bit
James Zern [Mon, 25 Feb 2013 21:00:18 +0000 (13:00 -0800)]
vp9: promote gf_group_bits calculation to 64-bit

avoids signed integer overflow

Change-Id: I9ffcdba90b21edb324d1b173fd11d613e0592931

11 years agoMerge "Minor rate control refactoring and experiments." into experimental
Paul Wilkins [Mon, 25 Feb 2013 20:49:54 +0000 (12:49 -0800)]
Merge "Minor rate control refactoring and experiments." into experimental

11 years agoImproving the forward 16x16 ADST/DCT accuracy
Jingning Han [Mon, 25 Feb 2013 20:08:29 +0000 (12:08 -0800)]
Improving the forward 16x16 ADST/DCT accuracy

Increase the first stage dynamic range by 4 times, and reduce it
back with proper rounding before applying the second stage. Hence
it still fits in the given dynamic range and slightly improves
the key frame coding performance.

Change-Id: Ia4c5907446f20a95dc3de079c314b3ad1221d8aa

11 years agoclean up forward and inverse hybrid transform
Jingning Han [Sat, 23 Feb 2013 00:29:28 +0000 (16:29 -0800)]
clean up forward and inverse hybrid transform

Rebased.

Remove the old matrix multiplication transform computation. The 16x16
ADST/DCT can be switched on/off and evaluated by setting ACTIVE_HT16
300/0 in vp9/common/vp9_blockd.h.

Change-Id: Icab2dbd18538987e1dc4e88c45abfc4cfc6e133f

11 years agoMinor rate control refactoring and experiments.
Paul Wilkins [Mon, 25 Feb 2013 12:36:38 +0000 (12:36 +0000)]
Minor rate control refactoring and experiments.

Some minor refactoring code relating to estimates of
bits per MB at a given Q and estimating the allowed Q range.

Most of the changes here were included in a previous commit.
This commit seeks to separate out the refactoring from more
the material changes.

Two #define control flags have been added for experimentation.

ONE_SHOT_Q_ESTIMATE force the two pass encoder to
use its initial Q range estimate for the whole clip even if this results
in a miss on the target data rate. In effect this tightens the Q range
seen at the expense of rate control accuracy.

DISABLE_RC_LONG_TERM_MEM is a related flag that disables the
long term memory in the rate control. Local adjustments are still
made to try and better hit the rate target on a per frame basis but
the impact of rate control misses is not propagated to the remainder
of the clip. This means that for example an overshoot early on will not
cause frames later in the clip to be starved of bits. Again the result
of this relaxation amy be less rate control accuracy especially on short
clips.

The flags are disabled by default for now.

Change-Id: I7482f980146d8ea033b5d50cc689f772e4bd119e

11 years agooptimize forward 16x16 DCT for accuracy
Yaowu Xu [Sat, 23 Feb 2013 00:18:08 +0000 (16:18 -0800)]
optimize forward 16x16 DCT for accuracy

This commit added pre/post scaling for first half of fDCT16x16 to
reduce error, by simulation of 100,000 blocks for random inputs,
the average sse reduced from 2.1/block to 0.0498/block.

also enabled tests for 16x16 fDCT and iDCT

Change-Id: Id2a95f0464c6dd4118797d456237ae90274c0f02

11 years agoSplit coefficient token tables intra vs. inter.
Ronald S. Bultje [Tue, 19 Feb 2013 21:36:38 +0000 (13:36 -0800)]
Split coefficient token tables intra vs. inter.

Change-Id: I5416455f8f129ca0f450d00e48358d2012605072

11 years agoFurther changes to coefficient contexts.
Paul Wilkins [Tue, 19 Feb 2013 18:12:00 +0000 (10:12 -0800)]
Further changes to coefficient contexts.

This patch alters the balance of context between the
coefficient bands (reflecting the position of coefficients
within a transform blocks) and the energy of the previous
token (or tokens) within a block.

In this case the number of coefficient bands is reduced
but more previous token energy bands are supported.

Some initial rebalancing of the default tables has been
by running multiple derf clips at multiple data rates using
the ENTOPY_STATS macro. Further balancing needs to be
done using larger image formatsd especially in regard to
the bigger transform sizes which are not as well represented
in encodings of smaller image formats.

Change-Id: If9736e95c391e711b04aef6393d26f60f36e1f8a