]> granicus.if.org Git - libvpx/log
libvpx
10 years agoMerge "Prevent accidental changes to the previous frame mode_infos"
Guillaume Martres [Wed, 16 Oct 2013 23:07:05 +0000 (16:07 -0700)]
Merge "Prevent accidental changes to the previous frame mode_infos"

10 years agoMerge "change to use vp9_idct_32x32_add"
Yaowu Xu [Wed, 16 Oct 2013 22:53:37 +0000 (15:53 -0700)]
Merge "change to use vp9_idct_32x32_add"

10 years agoInlining and removing fwd_txm16x16 and fwd_txm8x8 pointers.
Dmitry Kovalev [Wed, 16 Oct 2013 22:00:48 +0000 (15:00 -0700)]
Inlining and removing fwd_txm16x16 and fwd_txm8x8 pointers.

Change-Id: I3528ba1c3fee761918509f9d9dc2d842c69f5a44

10 years agoMerge "Cleanup of decode_coefs() function."
Dmitry Kovalev [Wed, 16 Oct 2013 21:47:17 +0000 (14:47 -0700)]
Merge "Cleanup of decode_coefs() function."

10 years agochange to use vp9_idct_32x32_add
Yaowu Xu [Wed, 16 Oct 2013 21:39:14 +0000 (14:39 -0700)]
change to use vp9_idct_32x32_add

instead of vp9_idct32x32_1024_add by making use the eob positions

Change-Id: Iafcad20f0c75d8e00536c6a20dda76fe40a973fb

10 years agoMerge "Removing print_prob_tree function and vp9_coeff_probs typedef."
Dmitry Kovalev [Wed, 16 Oct 2013 20:13:25 +0000 (13:13 -0700)]
Merge "Removing print_prob_tree function and vp9_coeff_probs typedef."

10 years agoMerge "Moving FILTER_BITS constant from vp9_convolve.h to vp9_filter.h."
Dmitry Kovalev [Wed, 16 Oct 2013 20:13:14 +0000 (13:13 -0700)]
Merge "Moving FILTER_BITS constant from vp9_convolve.h to vp9_filter.h."

10 years agoMerge "Using constants instead of plain numbers."
Dmitry Kovalev [Wed, 16 Oct 2013 20:13:04 +0000 (13:13 -0700)]
Merge "Using constants instead of plain numbers."

10 years agoInitial 1-pass.
Marco Paniconi [Thu, 3 Oct 2013 00:13:59 +0000 (17:13 -0700)]
Initial 1-pass.

Change-Id: I58c5436f5c95f6012fb2891cd2a02f76e4870b6a

10 years agoImplement variance-based adaptive quantization
Guillaume Martres [Mon, 7 Oct 2013 18:20:10 +0000 (19:20 +0100)]
Implement variance-based adaptive quantization

This should be similar to what x264 does with --aq-mode 1.
It works well with clips like parkjoy and touhou
(http://x264.nl/developers/Dark_Shikari/LosslessTouhou.mkv).
At low bitrates, the segmentation signaling overhead may negate the
benefits of this feature.

(PGW) Default changed to feature OFF to allow provisional merge.
Change-Id: I938abf9bb487e1d4ad3b0264ea03d9826275c70b

10 years agoMerge "lint-hunks: remove runtime/int filter"
James Zern [Wed, 16 Oct 2013 08:44:24 +0000 (01:44 -0700)]
Merge "lint-hunks: remove runtime/int filter"

10 years agoMerge "Updated encoder to handle intra-only frames"
Adrian Grange [Wed, 16 Oct 2013 00:19:28 +0000 (17:19 -0700)]
Merge "Updated encoder to handle intra-only frames"

10 years agoMerge "Remove Windows-style newlines using dos2unix"
Johann [Wed, 16 Oct 2013 00:05:32 +0000 (17:05 -0700)]
Merge "Remove Windows-style newlines using dos2unix"

10 years agoMerge "Make vp9_zero use cases of consistent format"
Jingning Han [Tue, 15 Oct 2013 23:49:05 +0000 (16:49 -0700)]
Merge "Make vp9_zero use cases of consistent format"

10 years agoUpdated encoder to handle intra-only frames
Alexander Voronov [Thu, 3 Oct 2013 16:07:24 +0000 (20:07 +0400)]
Updated encoder to handle intra-only frames

Updated the encoder to handle frames that are coded
intra-only. Intra-only frames must be non-showable,
that is, the "show frame" flag must be set to 0 in
the frame header.

Tested by forcing the ARF frames to be coded intra-
only.

Note: The rate control code will need to be modified
to account for intra-only frames better than they
are currently handled.

Change-Id: I6a9dd5337deddcecc599d3a44a7431909ed21079

10 years agoMerge "Remove unused variable vp9_64x64_zeros"
Jingning Han [Tue, 15 Oct 2013 23:24:34 +0000 (16:24 -0700)]
Merge "Remove unused variable vp9_64x64_zeros"

10 years agoMerge "Fix a few indent format issues in buffer defs"
Jingning Han [Tue, 15 Oct 2013 23:23:50 +0000 (16:23 -0700)]
Merge "Fix a few indent format issues in buffer defs"

10 years agoMerge "Remove unused comment"
Jingning Han [Tue, 15 Oct 2013 23:23:44 +0000 (16:23 -0700)]
Merge "Remove unused comment"

10 years agoMerge "Re-design all-zero-coeff block index buffer use"
Jingning Han [Tue, 15 Oct 2013 23:23:38 +0000 (16:23 -0700)]
Merge "Re-design all-zero-coeff block index buffer use"

10 years agoMake vp9_zero use cases of consistent format
Jingning Han [Tue, 15 Oct 2013 23:09:28 +0000 (16:09 -0700)]
Make vp9_zero use cases of consistent format

Remove the semicolon in the definition of vp9_zero macro. Make all
the use cases of vp9_zero of consistent format.

Change-Id: Ibaf9751e8595872b12766381a93d185a4d90df8f

10 years agoRemove Windows-style newlines using dos2unix
Guillaume Martres [Tue, 15 Oct 2013 22:49:32 +0000 (15:49 -0700)]
Remove Windows-style newlines using dos2unix

Change-Id: I0a0f9c07e774450896abc9455728b97fd38ef00c

10 years agolint-hunks: remove runtime/int filter
Guillaume Martres [Tue, 15 Oct 2013 22:43:42 +0000 (15:43 -0700)]
lint-hunks: remove runtime/int filter

There's no good reason to use types like long or short in new code.

Change-Id: Ic6de6259d5a99b7af478e9c6ab74e65d81b022e2

10 years agoMerge "Converted assertion to returning error"
Yaowu Xu [Tue, 15 Oct 2013 22:27:48 +0000 (15:27 -0700)]
Merge "Converted assertion to returning error"

10 years agoMerge "Added trap for invalid key frame"
Yaowu Xu [Tue, 15 Oct 2013 21:28:21 +0000 (14:28 -0700)]
Merge "Added trap for invalid key frame"

10 years agoMerge "Prevent access to invalid pointer"
Yaowu Xu [Tue, 15 Oct 2013 21:28:15 +0000 (14:28 -0700)]
Merge "Prevent access to invalid pointer"

10 years agoMerge "Add clamp to prevent out of bound access"
Yaowu Xu [Tue, 15 Oct 2013 21:28:08 +0000 (14:28 -0700)]
Merge "Add clamp to prevent out of bound access"

10 years agoConverted assertion to returning error
Yaowu Xu [Tue, 15 Oct 2013 21:06:00 +0000 (14:06 -0700)]
Converted assertion to returning error

Assertion happens for invalid input data, the commit replace the
assertion with returning error.

Change-Id: I1b73ae752d64882d984cd23936efe75a757c2b41

10 years agoAdded trap for invalid key frame
Yaowu Xu [Tue, 15 Oct 2013 18:05:06 +0000 (11:05 -0700)]
Added trap for invalid key frame

Change-Id: I698e8df9b336d38bffe01e656acba00d4003695f

10 years agoPrevent access to invalid pointer
Yaowu Xu [Tue, 15 Oct 2013 20:46:29 +0000 (13:46 -0700)]
Prevent access to invalid pointer

The commit added check to make sure no invalid memory access even when
the decoder instance is never initialized.

Change-Id: I4da343d0b3c78c27777ac7f5ce7688562c69f0c5

10 years agoAdd clamp to prevent out of bound access
Yaowu Xu [Tue, 15 Oct 2013 16:17:11 +0000 (09:17 -0700)]
Add clamp to prevent out of bound access

For bad input data, the decoder may access the array out of bounds. The
commit added clamp to prevent such out of bound access

Change-Id: I0a1cfd9b8786ea7113a998053c76605c963b077a

10 years agoRemove unused variable vp9_64x64_zeros
Jingning Han [Tue, 15 Oct 2013 18:52:54 +0000 (11:52 -0700)]
Remove unused variable vp9_64x64_zeros

Remove the unused variable vp9_64x64_zeros from vp9_encodeframe_.

Change-Id: I34bfdcab9a9105440ad05154c1e0516e70258785

10 years agoFix a few indent format issues in buffer defs
Jingning Han [Tue, 15 Oct 2013 18:50:25 +0000 (11:50 -0700)]
Fix a few indent format issues in buffer defs

Change-Id: Iac55891ac9e6f13718c9f822aa099b5ca491832a

10 years agoRemove unused comment
Jingning Han [Tue, 15 Oct 2013 18:45:14 +0000 (11:45 -0700)]
Remove unused comment

Change-Id: I2d96940fae4c7a16661a43c2bf6907d8b1c1a127

10 years agoRemoving unused 8x4 transform from the encoder.
Dmitry Kovalev [Tue, 15 Oct 2013 18:27:28 +0000 (11:27 -0700)]
Removing unused 8x4 transform from the encoder.

Change-Id: Icbcf68b5b685a56f255ebc3859c9692accdadf9e

10 years agoRe-design all-zero-coeff block index buffer use
Jingning Han [Mon, 14 Oct 2013 23:03:23 +0000 (16:03 -0700)]
Re-design all-zero-coeff block index buffer use

Use the zcoeff_blk buffer of PICK_MODE_CONTEXT to store the indexes
of all-zero-coeff block of the current best mode. Remove the temporary
buffer best_zcoeff_blk defined in the rate-distortion optimization
loop. This improves the speed performance by about 0.5% in all speed
settings.

Change-Id: Ie3e15988ddfa581eafa2e19a8228d3fe4a46095c

10 years agoMerge "add a test vector with frame parallel mode enabled"
James Zern [Tue, 15 Oct 2013 09:02:16 +0000 (02:02 -0700)]
Merge "add a test vector with frame parallel mode enabled"

10 years agoMoving FILTER_BITS constant from vp9_convolve.h to vp9_filter.h.
Dmitry Kovalev [Tue, 15 Oct 2013 04:15:40 +0000 (21:15 -0700)]
Moving FILTER_BITS constant from vp9_convolve.h to vp9_filter.h.

Change-Id: Idd7bdb0c364d94c5a0d24c87bb8574292e4c840c

10 years agoRemoving print_prob_tree function and vp9_coeff_probs typedef.
Dmitry Kovalev [Tue, 15 Oct 2013 04:08:21 +0000 (21:08 -0700)]
Removing print_prob_tree function and vp9_coeff_probs typedef.

Change-Id: If14265084e9b4c85c75b43e8d33a6fafad468cbc

10 years agoUsing constants instead of plain numbers.
Dmitry Kovalev [Tue, 15 Oct 2013 03:33:37 +0000 (20:33 -0700)]
Using constants instead of plain numbers.

Replacing
  22 with TREE_SIZE(MAX_ENTROPY_TOKENS)
  12 with MAX_ENTROPY_TOKENS

Change-Id: If24919336e8ace9cf64991bd5ae33fa6656f7b93

10 years agoCleanup of decode_coefs() function.
Dmitry Kovalev [Sat, 12 Oct 2013 01:01:30 +0000 (18:01 -0700)]
Cleanup of decode_coefs() function.

Moving entropy context calculation to outer scope and adding consts to
pointers.

Change-Id: I25d0987e97495326ab1e8046e3fa3082c0584485

10 years agoMerge "vp9_receive_compressed_data: remove unnecessary indent"
James Zern [Mon, 14 Oct 2013 20:22:08 +0000 (13:22 -0700)]
Merge "vp9_receive_compressed_data: remove unnecessary indent"

10 years agoMerge "Move token_cache from cost_coeffs to MACROBLOCK"
Jingning Han [Mon, 14 Oct 2013 20:05:32 +0000 (13:05 -0700)]
Merge "Move token_cache from cost_coeffs to MACROBLOCK"

10 years agoMove token_cache from cost_coeffs to MACROBLOCK
Jingning Han [Fri, 11 Oct 2013 18:26:32 +0000 (11:26 -0700)]
Move token_cache from cost_coeffs to MACROBLOCK

This commit moves token_cache buffer into macroblock struct, instead
of defining as a local variable in cost_coeffs. This avoids repeatedly
re-allocating memory space in the rate-distortion optimization loop.

The runtime at speed 0 reduces:
bus 2000kbps, 161692ms to 159951ms
football 600kbps, 229505ms to 225821ms

Change-Id: If7da6b0b6d8c5138a16271a33c4548fba33d8840

10 years agoadd a test vector with frame parallel mode enabled
James Zern [Mon, 14 Oct 2013 14:20:08 +0000 (16:20 +0200)]
add a test vector with frame parallel mode enabled

vp90-2-07-frame_parallel.webm:
vpxenc stefan_sif.y4m \
  --codec=vp9 -p 2 \
  --frame-parallel=1 \
  --limit=10 \
  --auto-alt-ref=1 \
  --lag-in-frames=5

Change-Id: I7381a69aaaec238b309169a51b34cb6bf29a9c50

10 years agoMerge "Moving libmkv library to third_party folder."
Dmitry Kovalev [Sun, 13 Oct 2013 19:24:56 +0000 (12:24 -0700)]
Merge "Moving libmkv library to third_party folder."

10 years agoMerge "Making input pointer of any inverse transform constant."
Dmitry Kovalev [Sun, 13 Oct 2013 19:22:55 +0000 (12:22 -0700)]
Merge "Making input pointer of any inverse transform constant."

10 years agoMerge "Adding TREE_SIZE macro + cleanup."
Dmitry Kovalev [Sun, 13 Oct 2013 19:21:09 +0000 (12:21 -0700)]
Merge "Adding TREE_SIZE macro + cleanup."

10 years agoMerge "Adjust icc compiler options"
Erik Niemeyer [Sun, 13 Oct 2013 16:20:18 +0000 (09:20 -0700)]
Merge "Adjust icc compiler options"

10 years agoAdjust icc compiler options
Yunqing Wang [Sat, 12 Oct 2013 01:57:22 +0000 (18:57 -0700)]
Adjust icc compiler options

"-no-prec-div" option helps codec performance, so it was added back.
"-no-intel-extensions" was added to suppress link warning #10237.
option '-use-asm' is deprecated and removed.

Tested icc 32bit build and 64bit build.

Change-Id: I736ec2619857efd425ef76338dc52f8fbc0bcc7e

10 years agoMaking input pointer of any inverse transform constant.
Dmitry Kovalev [Sat, 12 Oct 2013 01:27:12 +0000 (18:27 -0700)]
Making input pointer of any inverse transform constant.

Also renaming dest_stride to stride in some places.

Change-Id: I75f602b623a5a7071d4922b747c45fa0b7d7a940

10 years agoMerge "Get libvpx to compile on VS2013."
Johann [Sat, 12 Oct 2013 00:26:29 +0000 (17:26 -0700)]
Merge "Get libvpx to compile on VS2013."

10 years agoAdding TREE_SIZE macro + cleanup.
Dmitry Kovalev [Fri, 11 Oct 2013 23:25:50 +0000 (16:25 -0700)]
Adding TREE_SIZE macro + cleanup.

Using TREE_SIZE for the following trees:
  vp9_intra_mode_tree
  vp9_inter_mode_tree
  vp9_partition_tree
  vp9_switchable_interp_tree
  vp9_mv_joint_tree
  vp9_mv_class_tree
  vp9_mv_class0_tree
  vp9_mv_fp_tree

Change-Id: I0212bb4c1ee6648249f68517e28a67a56591ee1b

10 years agoConsistent names for inverse hybrid transforms (2 of 2).
Dmitry Kovalev [Fri, 11 Oct 2013 22:49:05 +0000 (15:49 -0700)]
Consistent names for inverse hybrid transforms (2 of 2).

Renames:
  vp9_iht_add       -> vp9_iht4x4_add
  vp9_iht_add_8x8   -> vp9_iht8x8_add
  vp9_iht_add_16x16 -> vp9_iht16x16_add

Change-Id: I8f1a2913e02d90d41f174f27e4ee2fad0dbd4a21

10 years agoMerge "Consistent names for inverse hybrid transforms (1 of 2)."
Dmitry Kovalev [Fri, 11 Oct 2013 22:33:00 +0000 (15:33 -0700)]
Merge "Consistent names for inverse hybrid transforms (1 of 2)."

10 years agoGet libvpx to compile on VS2013.
Scott Graham [Fri, 11 Oct 2013 21:27:00 +0000 (14:27 -0700)]
Get libvpx to compile on VS2013.

`round` is defined in the runtime library now.
https://codereview.chromium.org/23922008/

Change-Id: I3852740058d32f63ce283579acbe284865e32dba

10 years agoMerge "Replacing {VP9_COEF, MODE}_UPDATE_PROB with DIFF_UPDATE_PROB."
Dmitry Kovalev [Fri, 11 Oct 2013 21:15:46 +0000 (14:15 -0700)]
Merge "Replacing {VP9_COEF, MODE}_UPDATE_PROB with DIFF_UPDATE_PROB."

10 years agoMerge "Change in rddiv parameter to make it a power of 2"
Deb Mukherjee [Fri, 11 Oct 2013 20:53:59 +0000 (13:53 -0700)]
Merge "Change in rddiv parameter to make it a power of 2"

10 years agoConsistent names for inverse hybrid transforms (1 of 2).
Dmitry Kovalev [Fri, 11 Oct 2013 20:31:32 +0000 (13:31 -0700)]
Consistent names for inverse hybrid transforms (1 of 2).

Renames:
  vp9_short_iht4x4_add     -> vp9_iht4x4_16_add
  vp9_short_iht8x8_add     -> vp9_iht8x8_64_add
  vp9_short_iht16x16_add_c -> vp9_iht16x16_256_add

Change-Id: Ibca7a188fd062b196787ac5efc1ea545e7f166c0

10 years agoMerge "Adding const to the input argument of all 1D transforms."
Dmitry Kovalev [Fri, 11 Oct 2013 20:20:57 +0000 (13:20 -0700)]
Merge "Adding const to the input argument of all 1D transforms."

10 years agoMerge "Masking intra mode choice adaptively"
Yaowu Xu [Fri, 11 Oct 2013 18:25:52 +0000 (11:25 -0700)]
Merge "Masking intra mode choice adaptively"

10 years agoAdding const to the input argument of all 1D transforms.
Dmitry Kovalev [Fri, 11 Oct 2013 18:19:58 +0000 (11:19 -0700)]
Adding const to the input argument of all 1D transforms.

Also adding static to iadst16_1d and fadst16 functions.

Change-Id: I13c7df3b776f0f8efc6e80099bdb0a2f6d29edaf

10 years agoReplacing {VP9_COEF, MODE}_UPDATE_PROB with DIFF_UPDATE_PROB.
Dmitry Kovalev [Fri, 11 Oct 2013 17:47:22 +0000 (10:47 -0700)]
Replacing {VP9_COEF, MODE}_UPDATE_PROB with DIFF_UPDATE_PROB.

Values of MODE_UPDATE_PROB and VP9_COEF_UPDATE_PROB are equal, so replacing
them with one constant. Inlining appropriate arguments for functions:
  vp9_cond_prob_diff_update (encoder)
  vp9_diff_update_prob (decoder)

Change-Id: I1255a1cb477743b799b3bfbbcd8de6b32b067338

10 years agoMerge "Removing vp9_tree_p typedef."
Dmitry Kovalev [Fri, 11 Oct 2013 17:44:04 +0000 (10:44 -0700)]
Merge "Removing vp9_tree_p typedef."

10 years agoMerge "Removing vp9_idct4_1d_sse2 function."
Dmitry Kovalev [Fri, 11 Oct 2013 17:43:56 +0000 (10:43 -0700)]
Merge "Removing vp9_idct4_1d_sse2 function."

10 years agoChange in rddiv parameter to make it a power of 2
Deb Mukherjee [Fri, 11 Oct 2013 00:24:55 +0000 (17:24 -0700)]
Change in rddiv parameter to make it a power of 2

Converts the constant rddiv parameter to 128 (from 100) and
implements RDCOST with bit-shift rather than multiplication.
Other parameters are also adjusted to roughly keep the same
balance between Rate and Distortion.

There is a slight speed-up of about 0.5-1% (at speed 0) as
testted on football_cif.

There is a slight change in performance due to small change
in the parameters.
derfraw300: +0.033%
stdhdraw250; +0.102%

Change-Id: I70ac69f58fa71c83108f68fe41796cd19d1fc760

10 years agoMasking intra mode choice adaptively
Yaowu Xu [Fri, 11 Oct 2013 00:29:46 +0000 (17:29 -0700)]
Masking intra mode choice adaptively

The commit changes to mask available intra prediction modes for test
based on prediction block size.

With this patch, encoding time of CpuUsed 2 reduces from 10% to 20% for
HD clips with a compression drop of 0.2%

Change-Id: I65f320f1237c0f5ae3a355bf7caf447f55625455

10 years agoMerge "Code cleanup"
Yunqing Wang [Fri, 11 Oct 2013 16:38:24 +0000 (09:38 -0700)]
Merge "Code cleanup"

10 years agoMerge "Restore mode skip feature in sub8x8 rd loop"
Jingning Han [Fri, 11 Oct 2013 16:21:06 +0000 (09:21 -0700)]
Merge "Restore mode skip feature in sub8x8 rd loop"

10 years agoCode cleanup
Yunqing Wang [Fri, 11 Oct 2013 16:08:41 +0000 (09:08 -0700)]
Code cleanup

Minor code cleanup.

Change-Id: I47c1f794842d4570bb39cfd23b80f54f5606bba6

10 years agoMerge "Experimental rate control change."
Paul Wilkins [Fri, 11 Oct 2013 15:45:13 +0000 (08:45 -0700)]
Merge "Experimental rate control change."

10 years agoMerge "Disable recode loop."
Paul Wilkins [Fri, 11 Oct 2013 15:45:00 +0000 (08:45 -0700)]
Merge "Disable recode loop."

10 years agoMerge "SSE2 8-tap sub-pixel filter optimization"
Yunqing Wang [Fri, 11 Oct 2013 15:44:56 +0000 (08:44 -0700)]
Merge "SSE2 8-tap sub-pixel filter optimization"

10 years agoExperimental rate control change.
Paul Wilkins [Mon, 7 Oct 2013 18:20:10 +0000 (19:20 +0100)]
Experimental rate control change.

When the codec in VBR (or cq) mode hits its max q limits and is
struggling to hit a target bandwidth, the bit target per frame collapses.

In the first instance normal frames cap out at the maximum allowed
Q and then the ARF and GFs do the same. This latter behavior is not
generally desirable as GFs and ARFs are only effective from a quality
and data rate perspective if they have at lease some level of -Q delta
compared to the surrounding frames.

In this patch I define a separate max Q for GFs and ARFs that is
derived from but somewhat lower than that defined for normal frames.
In effect there is a minimum Q delta that will always be available for
GFs and ARFs regardless of the target rate and MAXQ setting.

This may of course mean that the absolute lowest rate obtainable for
a given clip is somewhat higher.

Change-Id: I268868b28401900d0cd87e51e609cd3b784ab54a

10 years agoDisable recode loop.
Paul Wilkins [Wed, 9 Oct 2013 13:57:28 +0000 (14:57 +0100)]
Disable recode loop.

For VBR coding disable the recode loop for speeds > 0.

Results pending.

Change-Id: I2cd9a87c3fcbe39c05b954798d0671a4ca62c37f

10 years agoAdjustment to allowed range in resize unit test
Paul Wilkins [Fri, 11 Oct 2013 12:27:22 +0000 (13:27 +0100)]
Adjustment to allowed range in resize unit test

Change-Id: I5222e3db2627a3a9f7fc34f2ab4554aa5807ed51

10 years agoRemoving vp9_tree_p typedef.
Dmitry Kovalev [Fri, 11 Oct 2013 00:16:20 +0000 (17:16 -0700)]
Removing vp9_tree_p typedef.

It is used only two times and it is more clear to use real type instead
of typedef.

Change-Id: Idc25c16504c3da4d040e0cdb33a2987631bb6a5b

10 years agoRemoving vp9_idct4_1d_sse2 function.
Dmitry Kovalev [Thu, 10 Oct 2013 23:50:43 +0000 (16:50 -0700)]
Removing vp9_idct4_1d_sse2 function.

We have two SSE2-optimized functions for idct4_1d:
  vp9_idct4_1d_sse2 <-- removing this one
  idct4_1d_sse2

vp9_idct4_1d_sse2 was used only by the following functions which already
have SSE2 optimized variants:
  vp9_idct4x4_16_add_c   -> vp9_idct4x4_16_add_see2
  idct8_1d               -> vp9_idct8x8_{16, 10, 1}_see2
  vp9_short_iht4x4_add_c -> vp9_short_iht4x4_add_see2

Change-Id: Ib0a7f6d1373dbaf7a4a41208cd9d0671fdf15edb

10 years agod207 intra prediction ssse3 using bytes
Scott LaVarnway [Fri, 27 Sep 2013 17:56:30 +0000 (13:56 -0400)]
d207 intra prediction ssse3 using bytes

byte version of ronalds d207 ssse3 optimizations
(commit: f891f84d3ba9345b0074e682f0fea09b8ddf4f1e)

Change-Id: If15f71a589ea16f78ac86a501b0c5c6231dc9af1

10 years agoMerge "Giving consistent names to IDCT 32x32 functions."
Dmitry Kovalev [Thu, 10 Oct 2013 22:31:25 +0000 (15:31 -0700)]
Merge "Giving consistent names to IDCT 32x32 functions."

10 years agoMerge "Consistent names for FDCT functions."
Dmitry Kovalev [Thu, 10 Oct 2013 22:29:29 +0000 (15:29 -0700)]
Merge "Consistent names for FDCT functions."

10 years agoMerge "d153 intra prediction (32x32) ssse3 using bytes"
Yunqing Wang [Thu, 10 Oct 2013 22:16:45 +0000 (15:16 -0700)]
Merge "d153 intra prediction (32x32) ssse3 using bytes"

10 years agoSSE2 8-tap sub-pixel filter optimization
Yunqing Wang [Thu, 10 Oct 2013 20:51:35 +0000 (13:51 -0700)]
SSE2 8-tap sub-pixel filter optimization

To ensure fast encoding/decoding on devices without ssse3 support,
SSE2 optimization of sub-pixel filters was done. Test using 1080p
clip showed the decoder speeds were ~70fps with ssse3 filters, ~60fps
with sse2 filters, and ~15fps with c filters.

Change-Id: Ie2088f87d83a889fba80a613e4d0e287aadd785c

10 years agoMerge "Fix typo in comment message"
Adrian Grange [Thu, 10 Oct 2013 21:05:51 +0000 (14:05 -0700)]
Merge "Fix typo in comment message"

10 years agoMerge "change to avoid out-of-range computation"
Yaowu Xu [Thu, 10 Oct 2013 20:38:16 +0000 (13:38 -0700)]
Merge "change to avoid out-of-range computation"

10 years agoMerge "Re-design rate-distortion cost tracking buffers"
Jingning Han [Thu, 10 Oct 2013 19:57:31 +0000 (12:57 -0700)]
Merge "Re-design rate-distortion cost tracking buffers"

10 years agoPrevent accidental changes to the previous frame mode_infos
Guillaume Martres [Thu, 10 Oct 2013 17:38:34 +0000 (10:38 -0700)]
Prevent accidental changes to the previous frame mode_infos

This is needed to fix mbgraph but shouldn't affect anything else

Change-Id: I2f515052f62e348cd3794b7ff0c139802225ea95

10 years agoFix typo in comment message
Jingning Han [Mon, 7 Oct 2013 17:45:23 +0000 (10:45 -0700)]
Fix typo in comment message

Change-Id: Ifef756a3a91423bb9f5411f06fa092027be21ecf

10 years agoConsistent names for FDCT functions.
Dmitry Kovalev [Thu, 10 Oct 2013 18:53:55 +0000 (11:53 -0700)]
Consistent names for FDCT functions.

Renames:
  fdct4_1d   -> fdct4
  fadst4_1d  -> fadst4
  fdct8_1d   -> fdct8
  fadst8_1d  -> fadst8
  fdct16_1d  -> fdct16
  fadst16_1d -> fadst16

"_1d" suffix is redundant, so removing it. The same will happen with idct
in the next change sets.

Change-Id: Ibf421cd2f569146c6079269df7a31819c098265e

10 years agoGiving consistent names to IDCT 32x32 functions.
Dmitry Kovalev [Thu, 10 Oct 2013 18:27:39 +0000 (11:27 -0700)]
Giving consistent names to IDCT 32x32 functions.

Renames:
  vp9_short_idct32x32_add   -> vp9_idct32x32_1024_add
  vp9_short_idct32x32_1_add -> vp9_idct32x32_1_add
  vp9_idct_add_32x32        -> vp9_idct32x32_add

Change-Id: Id85306f5814bac6c47463a6b5901a93082510666

10 years agoRe-design rate-distortion cost tracking buffers
Jingning Han [Thu, 10 Oct 2013 17:37:40 +0000 (10:37 -0700)]
Re-design rate-distortion cost tracking buffers

This commit re-designs the per transformed block rate-distortion
costs tracking buffers. It removes redundant buffer usage, makes
the needed context memory allocation per VP9_COMP instance and
reuses the same buffer sets inside the rate-distortion optimization
search loop, thereby avoiding repeatedly requiring memory space.

It reduces speed 0 runtime:

bus at 2000 kbps from 166763ms to 158967ms,
football at 600 kbps from 246614ms to 234257ms.

Both about 5% speed-up. Local tests suggest about 2% to 5% speed-up
for speed 1 and 2 settings. This does not change compression
performance.

Change-Id: I363514c5276b5cf9a38c7251088ffc6ab7f9a4c3

10 years agochange to avoid out-of-range computation
Yaowu Xu [Thu, 10 Oct 2013 18:00:57 +0000 (11:00 -0700)]
change to avoid out-of-range computation

Change-Id: Id5e31833a0ef40de9f64c2f5674af7083233bf14

10 years agoMerge "Removing inv_txm4x4_1_add and inv_txm4x4_add function pointers."
Dmitry Kovalev [Thu, 10 Oct 2013 17:49:27 +0000 (10:49 -0700)]
Merge "Removing inv_txm4x4_1_add and inv_txm4x4_add function pointers."

10 years agoMerge "Moving all scan/iscan code into separate vp9_scan.{h, c} files."
Dmitry Kovalev [Thu, 10 Oct 2013 17:45:07 +0000 (10:45 -0700)]
Merge "Moving all scan/iscan code into separate vp9_scan.{h, c} files."

10 years agoMerge "Giving consistent names to IDCT 16x16 functions."
Dmitry Kovalev [Thu, 10 Oct 2013 17:43:14 +0000 (10:43 -0700)]
Merge "Giving consistent names to IDCT 16x16 functions."

10 years agoMerge "Adding const to several pointers."
Dmitry Kovalev [Thu, 10 Oct 2013 17:43:05 +0000 (10:43 -0700)]
Merge "Adding const to several pointers."

10 years agoMerge "Adjustment to mv cost parameters"
Deb Mukherjee [Thu, 10 Oct 2013 16:08:58 +0000 (09:08 -0700)]
Merge "Adjustment to mv cost parameters"

10 years agoMerge "Fix intra dist model of skip_encode feature"
Jingning Han [Thu, 10 Oct 2013 16:00:20 +0000 (09:00 -0700)]
Merge "Fix intra dist model of skip_encode feature"

10 years agoMerge "Allow sub8x8 intra modes test for alt frame coding"
Jingning Han [Thu, 10 Oct 2013 16:00:08 +0000 (09:00 -0700)]
Merge "Allow sub8x8 intra modes test for alt frame coding"

10 years agoMerge "Disable MODE_TEST_HIT_STATS"
Paul Wilkins [Thu, 10 Oct 2013 12:52:06 +0000 (05:52 -0700)]
Merge "Disable MODE_TEST_HIT_STATS"

10 years agoAdjustment to mv cost parameters
Deb Mukherjee [Wed, 9 Oct 2013 21:14:34 +0000 (14:14 -0700)]
Adjustment to mv cost parameters

Increases these parameters.
There is a small efficiency gain.

Change-Id: Ie5f0ddb39c907d335e0dafa5eb112365a81f4542
derfraw300: +0.091%
stdhdraw250: +0.238%