TorqueGameEngines/Torque3D
1
0
Fork 0
mirror of https://github.com/TorqueGameEngines/Torque3D.git synced 2026-07-16 00:54:54 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Engine directory for ticket #1

Browse source
This commit is contained in:
DavidWyand-GG 2012-09-19 11:15:01 -04:00
parent 352279af7a
commit 7dbfe6994d
3795 changed files with 1363358 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

169
Engine/lib/libtheora/CHANGES Normal file
View file

@ -0,0 +1,169 @@
libtheora 1.0 (2008 November 3)
- Merge x86 assembly for forward DCT from Thusnelda branch.
- Update 32 bit MMX with loop filter fix.
- Check for an uninitialized state before dereferencing in propagating
decode calls.
- Remove all TH_DEBUG statements.
- Rename the bitpacker source files copied from libogg to avoid
confusing simple build systems using both libraries.
- Declare bitfield entries to be explicitly signed for Solaris cc.
- Set quantization parameters to default values when an empty buffer is
passed with TH_ENCCTL_SET_QUANT_PARAMS.
- Split encoder and decoder tests depending on configure settings.
- Return lstylex.sty to the distribution.
- Disable inline assembly on gcc versions prior to 3.1.
- Remove extern references for OC_*_QUANT_MIN.
- Make various data tables static const so they can be read-only.
- Remove ENCCTL codes from the old encoder API.
- Implement TH_ENCCTL_SET_KEYFRAME_FREQUENCY_FORCE ctl.
- Fix segfault when exactly one of the width or height is not a multiple
of 16, but the other is.
- Compute the correct vertical offset for chroma.
- cpuid assembly fix for MSVC.
- Add VS2008 project files.
- Build updates for 64-bit platforms, Mingw32, VS and XCode.
- Do not clobber the cropping rectangle.
- Declare ourselves 1.0final to pkg-config to sort after beta releases.
- Fix the scons build to include asm in libtheoradec/enc.
libtheora 1.0beta3 (2008 April 16)
- Build new libtheoradec and libtheoraenc libraries
supporting the new API from theora-exp. This API should
not be considered stable yet.
- Change granule_frame() to return an index as documented.
This is a change of behaviour from 1.0beta1.
- Document that granule_time() returns the end of the
presentation interval.
- Use a custom copy of the libogg bitpacker in the decoder
to avoid function call overhead.
- MMX code improved and ported to MSVC.
- Fix a problem with the MMX code on SELinux.
- Fix a problem with decoder quantizer initialization.
- Fix a page queue problem with png2theora.
- Improved robustness.
- Updated VS2005 project files.
- Dropped build support for Microsoft VS2003.
- Dropped build support for the unreleased libogg2.
- Added the specification to the autotools build.
- Specification corrections.
libtheora 1.0beta2 (2007 October 12)
- Fix a crash bug on char-is-unsigned architectures (PowerPC)
- Fix a buffer sizing issue that caused rare encoder crashes
- Fix a buffer alignment issue
- Build fixes for MingW32, MSVC
- Improved format documentation.
libtheora 1.0beta1 (2007 September 22)
- Granulepos scheme modified to match other codecs. This bumps
the bitstream revision to 3.2.1. Bitstreams marked 3.2.0 are
handled correctly by this decoder. Older decoders will show
a one frame sync error in the less noticable direction.
libtheora 1.0alpha8 (2007 September 18)
- Switch to new spec compliant decoder from theora-exp branch.
Written by Dr. Timothy Terriberry.
- Add support to the encoder for using quantization settings
provided by the application.
- more assembly optimizations
libtheora 1.0alpha7 (2006 June 20)
- Enable mmx assembly by default
- Avoid some relocations that caused problems on SELinux
- Other build fixes
- time testing mode (-f) for the dump_video example
libtheora 1.0alpha6 (2006 May 30)
* Merge theora-mmx simd acceleration (x86_32 and x86_64)
* Major RTP payload specification update
* Minor format specification updates
* Fix some spurious calls to free() instead of _ogg_free()
* Fix invalid array indexing in PixelLineSearch()
* Improve robustness against invalid input
* General warning cleanup
* The offset_y member now means what every application thought it meant
(offset from the top). This will mean some old files (those with a
non-centered image created with a buggy encoder) will display differently.
libtheora 1.0alpha5 (2005 August 20)
* Fixed bitrate management bugs that caused popping and encode
errors
* Fixed a crash problem with the theora_state internals not
being intialized properly.
* new utility function:
- theora_granule_shift()
* dump_video example now makes YUV4MPEG files by default, so
the results can be fed back to encoder_example and similar
tools. The old behavior is restored through the '-r' switch.
* ./configure now prints a summary
* simple unit test of the comment api under 'make check'
* misc code cleanup, warning and leak fixes
libtheora 1.0alpha4 (2004 December 15)
* first draft of the Theora I Format Specification
* API documentation generated from theora.h with Doxygen
* fix a double-update bug in the motion analysis
* apply the loop filter before filling motion vector border
in the reference frame
* new utility functions:
- theora_packet_isheader(),
- theora_packet_iskeyframe()
- theora_granule_frame()
* optional support for building without floating point
* optional support for building without encode support
* various build and packaging fixes
* pkg-config support
* SymbianOS build support
libtheora 1.0alpha3 (2004 March 20)
UPDATE: on 2004 July 1 the Theora I bitstream format was frozen. Files
produced by the libtheora 1.0alpha3 reference encoder will always be
decodable by the Theora I spec.
* Bitstream info header FORMAT CHANGES:
- move the granulepos shift field to maintain byte alignment longer.
- reserve 5 additional bits for subsampling and interlace flags.
* Bitstream setup header FORMAT CHANGES:
- support for a range of interpolated quant matricies.
- include the in-loop block filter coeff.
* Bitsteam data packet FORMAT CHANGES:
- Reserve a bit for per-block Q index selection.
- Flip the coded image orientation for compatibility with VP3.
This allows lossless transcoding of VP3 content, but files
encoded with earlier theora releases would play upside down.
* example VP3 lossless transcoder
* optional support for libogg2
* timing improvements in the example player
* packaging and build system updates and fixes
libtheora 1.0alpha2 (2003 June 9)
* bitstream FORMAT CHANGES:
- store the quant tables in a third setup header for
future encoder flexibility
- store the huffman tables in the third setup header
- add a field for marking the colorspace to the info header
- add crop parameters for non-multiple-of-16 frame sizes
- add a second vorbiscomment-style metadata header
* API changes to handle multiple headers with a single
theora_decode_header() call, like libvorbis
* code cleanup and minor fixes
* new dump_video code example/utility
* experimental win32 code examples
libtheora 1.0alpha1 (2002 September 25)
* First release of the theora reference implementation
* Port of the newly opened VP3 code to the Ogg container
* Rewrite of the code for portability and to use the libogg bitpacker

28
Engine/lib/libtheora/COPYING Normal file
View file

@ -0,0 +1,28 @@
Copyright (C) 2002-2008 Xiph.Org Foundation and contributors.
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions
are met:
- Redistributions of source code must retain the above copyright
notice, this list of conditions and the following disclaimer.
- Redistributions in binary form must reproduce the above copyright
notice, this list of conditions and the following disclaimer in the
documentation and/or other materials provided with the distribution.
- Neither the name of the Xiph.Org Foundation nor the names of its
contributors may be used to endorse or promote products derived from
this software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION
OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

594
Engine/lib/libtheora/include/theora/codec.h Normal file
View file

@ -0,0 +1,594 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: theora.h,v 1.8 2004/03/15 22:17:32 derf Exp $
********************************************************************/
/**\mainpage
*
* \section intro Introduction
*
* This is the documentation for <tt>libtheora</tt> C API.
* The current reference
* implementation for <a href="http://www.theora.org/">Theora</a>, a free,
* patent-unencumbered video codec.
* Theora is derived from On2's VP3 codec with additional features and
* integration for Ogg multimedia formats by
* <a href="http://www.xiph.org/">the Xiph.Org Foundation</a>.
* Complete documentation of the format itself is available in
* <a href="http://www.theora.org/doc/Theora_I_spec.pdf">the Theora
* specification</a>.
*
* \subsection Organization
*
* The functions documented here are actually subdivided into three
* separate libraries:
* - <tt>libtheoraenc</tt> contains the encoder interface,
* described in \ref encfuncs.
* - <tt>libtheoradec</tt> contains the decoder interface and
* routines shared with the encoder.
* You must also link to this if you link to <tt>libtheoraenc</tt>.
* The routines in this library are described in \ref decfuncs and
* \ref basefuncs.
* - <tt>libtheora</tt> contains the \ref oldfuncs.
*
* New code should link to <tt>libtheoradec</tt> and, if using encoder
* features, <tt>libtheoraenc</tt>. Together these two export both
* the standard and the legacy API, so this is all that is needed by
* any code. The older <tt>libtheora</tt> library is provided just for
* compatibility with older build configurations.
*
* In general the recommended 1.x API symbols can be distinguished
* by their <tt>th_</tt> or <tt>TH_</tt> namespace prefix.
* The older, legacy API uses <tt>theora_</tt> or <tt>OC_</tt>
* prefixes instead.
*/
/**\file
* The shared <tt>libtheoradec</tt> and <tt>libtheoraenc</tt> C API.
* You don't need to include this directly.*/
#if !defined(_O_THEORA_CODEC_H_)
# define _O_THEORA_CODEC_H_ (1)
# include <ogg/ogg.h>
#if defined(__cplusplus)
extern "C" {
#endif
/**\name Return codes*/
/*@{*/
/**An invalid pointer was provided.*/
#define TH_EFAULT (-1)
/**An invalid argument was provided.*/
#define TH_EINVAL (-10)
/**The contents of the header were incomplete, invalid, or unexpected.*/
#define TH_EBADHEADER (-20)
/**The header does not belong to a Theora stream.*/
#define TH_ENOTFORMAT (-21)
/**The bitstream version is too high.*/
#define TH_EVERSION (-22)
/**The specified function is not implemented.*/
#define TH_EIMPL (-23)
/**There were errors in the video data packet.*/
#define TH_EBADPACKET (-24)
/**The decoded packet represented a dropped frame.
The player can continue to display the current frame, as the contents of the
decoded frame buffer have not changed.*/
#define TH_DUPFRAME (1)
/*@}*/
/**The currently defined color space tags.
* See <a href="http://www.theora.org/doc/Theora_I_spec.pdf">the Theora
* specification</a>, Chapter 4, for exact details on the meaning of each of
* these color spaces.*/
typedef enum{
/**The color space was not specified at the encoder.
It may be conveyed by an external means.*/
TH_CS_UNSPECIFIED,
/**A color space designed for NTSC content.*/
TH_CS_ITU_REC_470M,
/**A color space designed for PAL/SECAM content.*/
TH_CS_ITU_REC_470BG,
/**The total number of currently defined color spaces.*/
TH_CS_NSPACES
}th_colorspace;
/**The currently defined pixel format tags.
* See <a href="http://www.theora.org/doc/Theora_I_spec.pdf">the Theora
* specification</a>, Section 4.4, for details on the precise sample
* locations.*/
typedef enum{
/**Chroma decimation by 2 in both the X and Y directions (4:2:0).
The Cb and Cr chroma planes are half the width and half the height of the
luma plane.*/
TH_PF_420,
/**Currently reserved.*/
TH_PF_RSVD,
/**Chroma decimation by 2 in the X direction (4:2:2).
The Cb and Cr chroma planes are half the width of the luma plane, but full
height.*/
TH_PF_422,
/**No chroma decimation (4:4:4).
The Cb and Cr chroma planes are full width and full height.*/
TH_PF_444,
/**The total number of currently defined pixel formats.*/
TH_PF_NFORMATS
}th_pixel_fmt;
/**A buffer for a single color plane in an uncompressed image.
* This contains the image data in a left-to-right, top-down format.
* Each row of pixels is stored contiguously in memory, but successive rows
* need not be.
* Use \a stride to compute the offset of the next row.
* The encoder accepts both positive \a stride values (top-down in memory) and
* negative (bottom-up in memory).
* The decoder currently always generates images with positive strides.*/
typedef struct{
/**The width of this plane.*/
int width;
/**The height of this plane.*/
int height;
/**The offset in bytes between successive rows.*/
int stride;
/**A pointer to the beginning of the first row.*/
unsigned char *data;
}th_img_plane;
/**A complete image buffer for an uncompressed frame.
* The chroma planes may be decimated by a factor of two in either direction,
* as indicated by th_info#pixel_fmt.
* The width and height of the Y' plane must be multiples of 16.
* They may need to be cropped for display, using the rectangle specified by
* th_info#pic_x, th_info#pic_y, th_info#pic_width, and
* th_info#pic_height.
* All samples are 8 bits.
* \note The term YUV often used to describe a colorspace is ambiguous.
* The exact parameters of the RGB to YUV conversion process aside, in many
* contexts the U and V channels actually have opposite meanings.
* To avoid this confusion, we are explicit: the name of the color channels are
* Y'CbCr, and they appear in that order, always.
* The prime symbol denotes that the Y channel is non-linear.
* Cb and Cr stand for "Chroma blue" and "Chroma red", respectively.*/
typedef th_img_plane th_ycbcr_buffer[3];
/**Theora bitstream information.
* This contains the basic playback parameters for a stream, and corresponds to
* the initial 'info' header packet.
* To initialize an encoder, the application fills in this structure and
* passes it to th_encode_alloc().
* A default encoding mode is chosen based on the values of the #quality and
* #target_bitrate fields.
* On decode, it is filled in by th_decode_headerin(), and then passed to
* th_decode_alloc().
*
* Encoded Theora frames must be a multiple of 16 in size;
* this is what the #frame_width and #frame_height members represent.
* To handle arbitrary picture sizes, a crop rectangle is specified in the
* #pic_x, #pic_y, #pic_width and #pic_height members.
*
* All frame buffers contain pointers to the full, padded frame.
* However, the current encoder <em>will not</em> reference pixels outside of
* the cropped picture region, and the application does not need to fill them
* in.
* The decoder <em>will</em> allocate storage for a full frame, but the
* application <em>should not</em> rely on the padding containing sensible
* data.
*
* It is also generally recommended that the offsets and sizes should still be
* multiples of 2 to avoid chroma sampling shifts when chroma is sub-sampled.
* See <a href="http://www.theora.org/doc/Theora_I_spec.pdf">the Theora
* specification</a>, Section 4.4, for more details.
*
* Frame rate, in frames per second, is stored as a rational fraction, as is
* the pixel aspect ratio.
* Note that this refers to the aspect ratio of the individual pixels, not of
* the overall frame itself.
* The frame aspect ratio can be computed from pixel aspect ratio using the
* image dimensions.*/
typedef struct{
/**\name Theora version
* Bitstream version information.*/
/*@{*/
unsigned char version_major;
unsigned char version_minor;
unsigned char version_subminor;
/*@}*/
/**The encoded frame width.
* This must be a multiple of 16, and less than 1048576.*/
ogg_uint32_t frame_width;
/**The encoded frame height.
* This must be a multiple of 16, and less than 1048576.*/
ogg_uint32_t frame_height;
/**The displayed picture width.
* This must be no larger than width.*/
ogg_uint32_t pic_width;
/**The displayed picture height.
* This must be no larger than height.*/
ogg_uint32_t pic_height;
/**The X offset of the displayed picture.
* This must be no larger than #frame_width-#pic_width or 255, whichever is
* smaller.*/
ogg_uint32_t pic_x;
/**The Y offset of the displayed picture.
* This must be no larger than #frame_height-#pic_height, and
* #frame_height-#pic_height-#pic_y must be no larger than 255.
* This slightly funny restriction is due to the fact that the offset is
* specified from the top of the image for consistency with the standard
* graphics left-handed coordinate system used throughout this API, while it
* is stored in the encoded stream as an offset from the bottom.*/
ogg_uint32_t pic_y;
/**\name Frame rate
* The frame rate, as a fraction.
* If either is 0, the frame rate is undefined.*/
/*@{*/
ogg_uint32_t fps_numerator;
ogg_uint32_t fps_denominator;
/*@}*/
/**\name Aspect ratio
* The aspect ratio of the pixels.
* If either value is zero, the aspect ratio is undefined.
* If not specified by any external means, 1:1 should be assumed.
* The aspect ratio of the full picture can be computed as
* \code
* aspect_numerator*pic_width/(aspect_denominator*pic_height).
* \endcode */
/*@{*/
ogg_uint32_t aspect_numerator;
ogg_uint32_t aspect_denominator;
/*@}*/
/**The color space.*/
th_colorspace colorspace;
/**The pixel format.*/
th_pixel_fmt pixel_fmt;
/**The target bit-rate in bits per second.
If initializing an encoder with this struct, set this field to a non-zero
value to activate CBR encoding by default.*/
/*TODO: Current encoder does not support CBR mode, or anything like it.
We also don't really know what nominal rate each quality level
corresponds to yet.*/
int target_bitrate;
/**The target quality level.
Valid values range from 0 to 63, inclusive, with higher values giving
higher quality.
If initializing an encoder with this struct, and #target_bitrate is set
to zero, VBR encoding at this quality will be activated by default.*/
/*Currently this is set so that a qi of 0 corresponds to distortions of 24
times the JND, and each increase by 16 halves that value.
This gives us fine discrimination at low qualities, yet effective rate
control at high qualities.
The qi value 63 is special, however.
For this, the highest quality, we use one half of a JND for our threshold.
Due to the lower bounds placed on allowable quantizers in Theora, we will
not actually be able to achieve quality this good, but this should
provide as close to visually lossless quality as Theora is capable of.
We could lift the quantizer restrictions without breaking VP3.1
compatibility, but this would result in quantized coefficients that are
too large for the current bitstream to be able to store.
We'd have to redesign the token syntax to store these large coefficients,
which would make transcoding complex.*/
int quality;
/**The amount to shift to extract the last keyframe number from the granule
* position.
* This can be at most 31.
* th_info_init() will set this to a default value (currently <tt>6</tt>,
* which is good for streaming applications), but you can set it to 0 to
* make every frame a keyframe.
* The maximum distance between key frames is
* <tt>1<<#keyframe_granule_shift</tt>.
* The keyframe frequency can be more finely controlled with
* #TH_ENCCTL_SET_KEYFRAME_FREQUENCY_FORCE, which can also be adjusted
* during encoding (for example, to force the next frame to be a keyframe),
* but it cannot be set larger than the amount permitted by this field after
* the headers have been output.*/
int keyframe_granule_shift;
}th_info;
/**The comment information.
*
* This structure holds the in-stream metadata corresponding to
* the 'comment' header packet.
* The comment header is meant to be used much like someone jotting a quick
* note on the label of a video.
* It should be a short, to the point text note that can be more than a couple
* words, but not more than a short paragraph.
*
* The metadata is stored as a series of (tag, value) pairs, in
* length-encoded string vectors.
* The first occurrence of the '=' character delimits the tag and value.
* A particular tag may occur more than once, and order is significant.
* The character set encoding for the strings is always UTF-8, but the tag
* names are limited to ASCII, and treated as case-insensitive.
* See <a href="http://www.theora.org/doc/Theora_I_spec.pdf">the Theora
* specification</a>, Section 6.3.3 for details.
*
* In filling in this structure, th_decode_headerin() will null-terminate
* the user_comment strings for safety.
* However, the bitstream format itself treats them as 8-bit clean vectors,
* possibly containing null characters, and so the length array should be
* treated as their authoritative length.
*/
typedef struct th_comment{
/**The array of comment string vectors.*/
char **user_comments;
/**An array of the corresponding length of each vector, in bytes.*/
int *comment_lengths;
/**The total number of comment strings.*/
int comments;
/**The null-terminated vendor string.
This identifies the software used to encode the stream.*/
char *vendor;
}th_comment;
/**A single base matrix.*/
typedef unsigned char th_quant_base[64];
/**A set of \a qi ranges.*/
typedef struct{
/**The number of ranges in the set.*/
int nranges;
/**The size of each of the #nranges ranges.
These must sum to 63.*/
const int *sizes;
/**#nranges <tt>+1</tt> base matrices.
Matrices \a i and <tt>i+1</tt> form the endpoints of range \a i.*/
const th_quant_base *base_matrices;
}th_quant_ranges;
/**A complete set of quantization parameters.
The quantizer for each coefficient is calculated as:
\code
Q=MAX(MIN(qmin[qti][ci!=0],scale[ci!=0][qi]*base[qti][pli][qi][ci]/100),
1024).
\endcode
\a qti is the quantization type index: 0 for intra, 1 for inter.
<tt>ci!=0</tt> is 0 for the DC coefficient and 1 for AC coefficients.
\a qi is the quality index, ranging between 0 (low quality) and 63 (high
quality).
\a pli is the color plane index: 0 for Y', 1 for Cb, 2 for Cr.
\a ci is the DCT coefficient index.
Coefficient indices correspond to the normal 2D DCT block
ordering--row-major with low frequencies first--\em not zig-zag order.
Minimum quantizers are constant, and are given by:
\code
qmin[2][2]={{4,2},{8,4}}.
\endcode
Parameters that can be stored in the bitstream are as follows:
- The two scale matrices ac_scale and dc_scale.
\code
scale[2][64]={dc_scale,ac_scale}.
\endcode
- The base matrices for each \a qi, \a qti and \a pli (up to 384 in all).
In order to avoid storing a full 384 base matrices, only a sparse set of
matrices are stored, and the rest are linearly interpolated.
This is done as follows.
For each \a qti and \a pli, a series of \a n \a qi ranges is defined.
The size of each \a qi range can vary arbitrarily, but they must sum to
63.
Then, <tt>n+1</tt> matrices are specified, one for each endpoint of the
ranges.
For interpolation purposes, each range's endpoints are the first \a qi
value it contains and one past the last \a qi value it contains.
Fractional values are rounded to the nearest integer, with ties rounded
away from zero.
Base matrices are stored by reference, so if the same matrices are used
multiple times, they will only appear once in the bitstream.
The bitstream is also capable of omitting an entire set of ranges and
its associated matrices if they are the same as either the previous
set (indexed in row-major order) or if the inter set is the same as the
intra set.
- Loop filter limit values.
The same limits are used for the loop filter in all color planes, despite
potentially differing levels of quantization in each.
For the current encoder, <tt>scale[ci!=0][qi]</tt> must be no greater
than <tt>scale[ci!=0][qi-1]</tt> and <tt>base[qti][pli][qi][ci]</tt> must
be no greater than <tt>base[qti][pli][qi-1][ci]</tt>.
These two conditions ensure that the actual quantizer for a given \a qti,
\a pli, and \a ci does not increase as \a qi increases.
This is not required by the decoder.*/
typedef struct{
/**The DC scaling factors.*/
ogg_uint16_t dc_scale[64];
/**The AC scaling factors.*/
ogg_uint16_t ac_scale[64];
/**The loop filter limit values.*/
unsigned char loop_filter_limits[64];
/**The \a qi ranges for each \a ci and \a pli.*/
th_quant_ranges qi_ranges[2][3];
}th_quant_info;
/**The number of Huffman tables used by Theora.*/
#define TH_NHUFFMAN_TABLES (80)
/**The number of DCT token values in each table.*/
#define TH_NDCT_TOKENS (32)
/**A Huffman code for a Theora DCT token.
* Each set of Huffman codes in a given table must form a complete, prefix-free
* code.
* There is no requirement that all the tokens in a table have a valid code,
* but the current encoder is not optimized to take advantage of this.
* If each of the five grouops of 16 tables does not contain at least one table
* with a code for every token, then the encoder may fail to encode certain
* frames.
* The complete table in the first group of 16 does not have to be in the same
* place as the complete table in the other groups, but the complete tables in
* the remaining four groups must all be in the same place.*/
typedef struct{
/**The bit pattern for the code, with the LSbit of the pattern aligned in
* the LSbit of the word.*/
ogg_uint32_t pattern;
/**The number of bits in the code.
* This must be between 0 and 32, inclusive.*/
int nbits;
}th_huff_code;
/**\defgroup basefuncs Functions Shared by Encode and Decode*/
/*@{*/
/**\name Basic shared functions*/
/*@{*/
/**Retrieves a human-readable string to identify the library vendor and
* version.
* \return the version string.*/
extern const char *th_version_string(void);
/**Retrieves the library version number.
* This is the highest bitstream version that the encoder library will produce,
* or that the decoder library can decode.
* This number is composed of a 16-bit major version, 8-bit minor version
* and 8 bit sub-version, composed as follows:
* \code
* (VERSION_MAJOR<<16)+(VERSION_MINOR<<8)+(VERSION_SUBMINOR)
* \endcode
* \return the version number.*/
extern ogg_uint32_t th_version_number(void);
/**Converts a granule position to an absolute frame index, starting at
* <tt>0</tt>.
* The granule position is interpreted in the context of a given
* #th_enc_ctx or #th_dec_ctx handle (either will suffice).
* \param _encdec A previously allocated #th_enc_ctx or #th_dec_ctx
* handle.
* \param _granpos The granule position to convert.
* \returns The absolute frame index corresponding to \a _granpos.
* \retval -1 The given granule position was invalid (i.e. negative).*/
extern ogg_int64_t th_granule_frame(void *_encdec,ogg_int64_t _granpos);
/**Converts a granule position to an absolute time in seconds.
* The granule position is interpreted in the context of a given
* #th_enc_ctx or #th_dec_ctx handle (either will suffice).
* \param _encdec A previously allocated #th_enc_ctx or #th_dec_ctx
* handle.
* \param _granpos The granule position to convert.
* \return The absolute time in seconds corresponding to \a _granpos.
* This is the "end time" for the frame, or the latest time it should
* be displayed.
* It is not the presentation time.
* \retval -1 The given granule position was invalid (i.e. negative).*/
extern double th_granule_time(void *_encdec,ogg_int64_t _granpos);
/**Determines whether a Theora packet is a header or not.
* This function does no verification beyond checking the packet type bit, so
* it should not be used for bitstream identification; use
* th_decode_headerin() for that.
* As per the Theora specification, an empty (0-byte) packet is treated as a
* data packet (a delta frame with no coded blocks).
* \param _op An <tt>ogg_packet</tt> containing encoded Theora data.
* \retval 1 The packet is a header packet
* \retval 0 The packet is a video data packet.*/
extern int th_packet_isheader(ogg_packet *_op);
/**Determines whether a theora packet is a key frame or not.
* This function does no verification beyond checking the packet type and
* key frame bits, so it should not be used for bitstream identification; use
* th_decode_headerin() for that.
* As per the Theora specification, an empty (0-byte) packet is treated as a
* delta frame (with no coded blocks).
* \param _op An <tt>ogg_packet</tt> containing encoded Theora data.
* \retval 1 The packet contains a key frame.
* \retval 0 The packet contains a delta frame.
* \retval -1 The packet is not a video data packet.*/
extern int th_packet_iskeyframe(ogg_packet *_op);
/*@}*/
/**\name Functions for manipulating header data*/
/*@{*/
/**Initializes a th_info structure.
* This should be called on a freshly allocated #th_info structure before
* attempting to use it.
* \param _info The #th_info struct to initialize.*/
extern void th_info_init(th_info *_info);
/**Clears a #th_info structure.
* This should be called on a #th_info structure after it is no longer
* needed.
* \param _info The #th_info struct to clear.*/
extern void th_info_clear(th_info *_info);
/**Initialize a #th_comment structure.
* This should be called on a freshly allocated #th_comment structure
* before attempting to use it.
* \param _tc The #th_comment struct to initialize.*/
extern void th_comment_init(th_comment *_tc);
/**Add a comment to an initialized #th_comment structure.
* \note Neither th_comment_add() nor th_comment_add_tag() support
* comments containing null values, although the bitstream format does
* support them.
* To add such comments you will need to manipulate the #th_comment
* structure directly.
* \param _tc The #th_comment struct to add the comment to.
* \param _comment Must be a null-terminated UTF-8 string containing the
* comment in "TAG=the value" form.*/
extern void th_comment_add(th_comment *_tc, char *_comment);
/**Add a comment to an initialized #th_comment structure.
* \note Neither th_comment_add() nor th_comment_add_tag() support
* comments containing null values, although the bitstream format does
* support them.
* To add such comments you will need to manipulate the #th_comment
* structure directly.
* \param _tc The #th_comment struct to add the comment to.
* \param _tag A null-terminated string containing the tag associated with
* the comment.
* \param _val The corresponding value as a null-terminated string.*/
extern void th_comment_add_tag(th_comment *_tc,char *_tag,char *_val);
/**Look up a comment value by its tag.
* \param _tc An initialized #th_comment structure.
* \param _tag The tag to look up.
* \param _count The instance of the tag.
* The same tag can appear multiple times, each with a distinct
* value, so an index is required to retrieve them all.
* The order in which these values appear is significant and
* should be preserved.
* Use th_comment_query_count() to get the legal range for
* the \a _count parameter.
* \return A pointer to the queried tag's value.
* This points directly to data in the #th_comment structure.
* It should not be modified or freed by the application, and
* modifications to the structure may invalidate the pointer.
* \retval NULL If no matching tag is found.*/
extern char *th_comment_query(th_comment *_tc,char *_tag,int _count);
/**Look up the number of instances of a tag.
* Call this first when querying for a specific tag and then iterate over the
* number of instances with separate calls to th_comment_query() to
* retrieve all the values for that tag in order.
* \param _tc An initialized #th_comment structure.
* \param _tag The tag to look up.
* \return The number on instances of this particular tag.*/
extern int th_comment_query_count(th_comment *_tc,char *_tag);
/**Clears a #th_comment structure.
* This should be called on a #th_comment structure after it is no longer
* needed.
* It will free all memory used by the structure members.
* \param _tc The #th_comment struct to clear.*/
extern void th_comment_clear(th_comment *_tc);
/*@}*/
/*@}*/
#if defined(__cplusplus)
}
#endif
#endif

794
Engine/lib/libtheora/include/theora/theora.h Normal file
View file

@ -0,0 +1,794 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: theora.h,v 1.17 2003/12/06 18:06:19 arc Exp $
********************************************************************/
#ifndef _O_THEORA_H_
#define _O_THEORA_H_
#ifdef __cplusplus
extern "C"
{
#endif /* __cplusplus */
#include <stddef.h> /* for size_t */
#include <ogg/ogg.h>
/** \defgroup oldfuncs Legacy pre-1.0 C API */
/* @{ */
/** \mainpage
*
* \section intro Introduction
*
* This is the documentation for the libtheora legacy C API, declared in
* the theora.h header, which describes the old interface used before
* the 1.0 release. This API was widely deployed for several years and
* remains supported, but for new code we recommend the cleaner API
* declared in theoradec.h and theoraenc.h.
*
* libtheora is the reference implementation for
* <a href="http://www.theora.org/">Theora</a>, a free video codec.
* Theora is derived from On2's VP3 codec with improved integration for
* Ogg multimedia formats by <a href="http://www.xiph.org/">Xiph.Org</a>.
*
* \section overview Overview
*
* This library will both decode and encode theora packets to/from raw YUV
* frames. In either case, the packets will most likely either come from or
* need to be embedded in an Ogg stream. Use
* <a href="http://xiph.org/ogg/">libogg</a> or
* <a href="http://www.annodex.net/software/liboggz/index.html">liboggz</a>
* to extract/package these packets.
*
* \section decoding Decoding Process
*
* Decoding can be separated into the following steps:
* -# initialise theora_info and theora_comment structures using
* theora_info_init() and theora_comment_init():
\verbatim
theora_info info;
theora_comment comment;
theora_info_init(&info);
theora_comment_init(&comment);
\endverbatim
* -# retrieve header packets from Ogg stream (there should be 3) and decode
* into theora_info and theora_comment structures using
* theora_decode_header(). See \ref identification for more information on
* identifying which packets are theora packets.
\verbatim
int i;
for (i = 0; i < 3; i++)
{
(get a theora packet "op" from the Ogg stream)
theora_decode_header(&info, &comment, op);
}
\endverbatim
* -# initialise the decoder based on the information retrieved into the
* theora_info struct by theora_decode_header(). You will need a
* theora_state struct.
\verbatim
theora_state state;
theora_decode_init(&state, &info);
\endverbatim
* -# pass in packets and retrieve decoded frames! See the yuv_buffer
* documentation for information on how to retrieve raw YUV data.
\verbatim
yuf_buffer buffer;
while (last packet was not e_o_s) {
(get a theora packet "op" from the Ogg stream)
theora_decode_packetin(&state, op);
theora_decode_YUVout(&state, &buffer);
}
\endverbatim
*
*
* \subsection identification Identifying Theora Packets
*
* All streams inside an Ogg file have a unique serial_no attached to the
* stream. Typically, you will want to
* - retrieve the serial_no for each b_o_s (beginning of stream) page
* encountered within the Ogg file;
* - test the first (only) packet on that page to determine if it is a theora
* packet;
* - once you have found a theora b_o_s page then use the retrieved serial_no
* to identify future packets belonging to the same theora stream.
*
* Note that you \e cannot use theora_packet_isheader() to determine if a
* packet is a theora packet or not, as this function does not perform any
* checking beyond whether a header bit is present. Instead, use the
* theora_decode_header() function and check the return value; or examine the
* header bytes at the beginning of the Ogg page.
*
* \subsection example Example Decoder
*
* See <a href="http://svn.xiph.org/trunk/theora/examples/dump_video.c">
* examples/dump_video.c</a> for a simple decoder implementation.
*
* \section encoding Encoding Process
*
* See <a href="http://svn.xiph.org/trunk/theora/examples/encoder_example.c">
* examples/encoder_example.c</a> for a simple encoder implementation.
*/
/** \file
* The libtheora pre-1.0 legacy C API.
*/
/**
* A YUV buffer for passing uncompressed frames to and from the codec.
* This holds a Y'CbCr frame in planar format. The CbCr planes can be
* subsampled and have their own separate dimensions and row stride
* offsets. Note that the strides may be negative in some
* configurations. For theora the width and height of the largest plane
* must be a multiple of 16. The actual meaningful picture size and
* offset are stored in the theora_info structure; frames returned by
* the decoder may need to be cropped for display.
*
* All samples are 8 bits. Within each plane samples are ordered by
* row from the top of the frame to the bottom. Within each row samples
* are ordered from left to right.
*
* During decode, the yuv_buffer struct is allocated by the user, but all
* fields (including luma and chroma pointers) are filled by the library.
* These pointers address library-internal memory and their contents should
* not be modified.
*
* Conversely, during encode the user allocates the struct and fills out all
* fields. The user also manages the data addressed by the luma and chroma
* pointers. See the encoder_example.c and dump_video.c example files in
* theora/examples/ for more information.
*/
typedef struct {
int y_width; /**< Width of the Y' luminance plane */
int y_height; /**< Height of the luminance plane */
int y_stride; /**< Offset in bytes between successive rows */
int uv_width; /**< Width of the Cb and Cr chroma planes */
int uv_height; /**< Height of the chroma planes */
int uv_stride; /**< Offset between successive chroma rows */
unsigned char *y; /**< Pointer to start of luminance data */
unsigned char *u; /**< Pointer to start of Cb data */
unsigned char *v; /**< Pointer to start of Cr data */
} yuv_buffer;
/**
* A Colorspace.
*/
typedef enum {
OC_CS_UNSPECIFIED, /**< The colorspace is unknown or unspecified */
OC_CS_ITU_REC_470M, /**< This is the best option for 'NTSC' content */
OC_CS_ITU_REC_470BG, /**< This is the best option for 'PAL' content */
OC_CS_NSPACES /**< This marks the end of the defined colorspaces */
} theora_colorspace;
/**
* A Chroma subsampling
*
* These enumerate the available chroma subsampling options supported
* by the theora format. See Section 4.4 of the specification for
* exact definitions.
*/
typedef enum {
OC_PF_420, /**< Chroma subsampling by 2 in each direction (4:2:0) */
OC_PF_RSVD, /**< Reserved value */
OC_PF_422, /**< Horizonatal chroma subsampling by 2 (4:2:2) */
OC_PF_444, /**< No chroma subsampling at all (4:4:4) */
} theora_pixelformat;
/**
* Theora bitstream info.
* Contains the basic playback parameters for a stream,
* corresponding to the initial 'info' header packet.
*
* Encoded theora frames must be a multiple of 16 in width and height.
* To handle other frame sizes, a crop rectangle is specified in
* frame_height and frame_width, offset_x and * offset_y. The offset
* and size should still be a multiple of 2 to avoid chroma sampling
* shifts. Offset values in this structure are measured from the
* upper left of the image.
*
* Frame rate, in frames per second, is stored as a rational
* fraction. Aspect ratio is also stored as a rational fraction, and
* refers to the aspect ratio of the frame pixels, not of the
* overall frame itself.
*
* See <a href="http://svn.xiph.org/trunk/theora/examples/encoder_example.c">
* examples/encoder_example.c</a> for usage examples of the
* other paramters and good default settings for the encoder parameters.
*/
typedef struct {
ogg_uint32_t width; /**< encoded frame width */
ogg_uint32_t height; /**< encoded frame height */
ogg_uint32_t frame_width; /**< display frame width */
ogg_uint32_t frame_height; /**< display frame height */
ogg_uint32_t offset_x; /**< horizontal offset of the displayed frame */
ogg_uint32_t offset_y; /**< vertical offset of the displayed frame */
ogg_uint32_t fps_numerator; /**< frame rate numerator **/
ogg_uint32_t fps_denominator; /**< frame rate denominator **/
ogg_uint32_t aspect_numerator; /**< pixel aspect ratio numerator */
ogg_uint32_t aspect_denominator; /**< pixel aspect ratio denominator */
theora_colorspace colorspace; /**< colorspace */
int target_bitrate; /**< nominal bitrate in bits per second */
int quality; /**< Nominal quality setting, 0-63 */
int quick_p; /**< Quick encode/decode */
/* decode only */
unsigned char version_major;
unsigned char version_minor;
unsigned char version_subminor;
void *codec_setup;
/* encode only */
int dropframes_p;
int keyframe_auto_p;
ogg_uint32_t keyframe_frequency;
ogg_uint32_t keyframe_frequency_force; /* also used for decode init to
get granpos shift correct */
ogg_uint32_t keyframe_data_target_bitrate;
ogg_int32_t keyframe_auto_threshold;
ogg_uint32_t keyframe_mindistance;
ogg_int32_t noise_sensitivity;
ogg_int32_t sharpness;
theora_pixelformat pixelformat; /**< chroma subsampling mode to expect */
} theora_info;
/** Codec internal state and context.
*/
typedef struct{
theora_info *i;
ogg_int64_t granulepos;
void *internal_encode;
void *internal_decode;
} theora_state;
/**
* Comment header metadata.
*
* This structure holds the in-stream metadata corresponding to
* the 'comment' header packet.
*
* Meta data is stored as a series of (tag, value) pairs, in
* length-encoded string vectors. The first occurence of the
* '=' character delimits the tag and value. A particular tag
* may occur more than once. The character set encoding for
* the strings is always UTF-8, but the tag names are limited
* to case-insensitive ASCII. See the spec for details.
*
* In filling in this structure, theora_decode_header() will
* null-terminate the user_comment strings for safety. However,
* the bitstream format itself treats them as 8-bit clean,
* and so the length array should be treated as authoritative
* for their length.
*/
typedef struct theora_comment{
char **user_comments; /**< An array of comment string vectors */
int *comment_lengths; /**< An array of corresponding string vector lengths in bytes */
int comments; /**< The total number of comment string vectors */
char *vendor; /**< The vendor string identifying the encoder, null terminated */
} theora_comment;
/**\name theora_control() codes */
/**\anchor decctlcodes
* These are the available request codes for theora_control()
* when called with a decoder instance.
* By convention, these are odd, to distinguish them from the
* \ref encctlcodes "encoder control codes".
* Keep any experimental or vendor-specific values above \c 0x8000.*/
/**Get the maximum post-processing level.
* The decoder supports a post-processing filter that can improve
* the appearance of the decoded images. This returns the highest
* level setting for this post-processor, corresponding to maximum
* improvement and computational expense.
*/
#define TH_DECCTL_GET_PPLEVEL_MAX (1)
/**Set the post-processing level.
* Sets the level of post-processing to use when decoding the
* compressed stream. This must be a value between zero (off)
* and the maximum returned by TH_DECCTL_GET_PPLEVEL_MAX.
*/
#define TH_DECCTL_SET_PPLEVEL (3)
/**Sets the maximum distance between key frames.
* This can be changed during an encode, but will be bounded by
* <tt>1<<th_info#keyframe_granule_shift</tt>.
* If it is set before encoding begins, th_info#keyframe_granule_shift will
* be enlarged appropriately.
*
* \param[in] buf <tt>ogg_uint32_t</tt>: The maximum distance between key
* frames.
* \param[out] buf <tt>ogg_uint32_t</tt>: The actual maximum distance set.
* \retval TH_FAULT \a theora_state or \a buf is <tt>NULL</tt>.
* \retval TH_EINVAL \a buf_sz is not <tt>sizeof(ogg_uint32_t)</tt>.
* \retval TH_IMPL Not supported by this implementation.*/
#define TH_ENCCTL_SET_KEYFRAME_FREQUENCY_FORCE (4)
/**Set the granule position.
* Call this after a seek, to update the internal granulepos
* in the decoder, to insure that subsequent frames are marked
* properly. If you track timestamps yourself and do not use
* the granule postion returned by the decoder, then you do
* not need to use this control.
*/
#define TH_DECCTL_SET_GRANPOS (5)
/**\anchor encctlcodes
* These are the available request codes for theora_control()
* when called with an encoder instance.
* By convention, these are even, to distinguish them from the
* \ref decctlcodes "decoder control codes".
* Keep any experimental or vendor-specific values above \c 0x8000.*/
/*@{*/
/**Sets the quantization parameters to use.
* The parameters are copied, not stored by reference, so they can be freed
* after this call.
* <tt>NULL</tt> may be specified to revert to the default parameters.
* For the current encoder, <tt>scale[ci!=0][qi]</tt> must be no greater than
* <tt>scale[ci!=0][qi-1]</tt> and <tt>base[qti][pli][qi][ci]</tt> must be no
* greater than <tt>base[qti][pli][qi-1][ci]</tt>.
* These two conditions ensure that the actual quantizer for a given \a qti,
* \a pli, and \a ci does not increase as \a qi increases.
*
* \param[in] buf #th_quant_info
* \retval TH_FAULT \a theora_state is <tt>NULL</tt>.
* \retval TH_EINVAL Encoding has already begun, the quantization parameters
* do not meet one of the above stated conditions, \a buf
* is <tt>NULL</tt> and \a buf_sz is not zero, or \a buf
* is non-<tt>NULL</tt> and \a buf_sz is not
* <tt>sizeof(#th_quant_info)</tt>.
* \retval TH_IMPL Not supported by this implementation.*/
#define TH_ENCCTL_SET_QUANT_PARAMS (2)
/**Disables any encoder features that would prevent lossless transcoding back
* to VP3.
* This primarily means disabling block-level QI values and not using 4MV mode
* when any of the luma blocks in a macro block are not coded.
* It also includes using the VP3 quantization tables and Huffman codes; if you
* set them explicitly after calling this function, the resulting stream will
* not be VP3-compatible.
* If you enable VP3-compatibility when encoding 4:2:2 or 4:4:4 source
* material, or when using a picture region smaller than the full frame (e.g.
* a non-multiple-of-16 width or height), then non-VP3 bitstream features will
* still be disabled, but the stream will still not be VP3-compatible, as VP3
* was not capable of encoding such formats.
* If you call this after encoding has already begun, then the quantization
* tables and codebooks cannot be changed, but the frame-level features will
* be enabled or disabled as requested.
*
* \param[in] buf <tt>int</tt>: a non-zero value to enable VP3 compatibility,
* or 0 to disable it (the default).
* \param[out] buf <tt>int</tt>: 1 if all bitstream features required for
* VP3-compatibility could be set, and 0 otherwise.
* The latter will be returned if the pixel format is not
* 4:2:0, the picture region is smaller than the full frame,
* or if encoding has begun, preventing the quantization
* tables and codebooks from being set.
* \retval TH_FAULT \a theora_state or \a buf is <tt>NULL</tt>.
* \retval TH_EINVAL \a buf_sz is not <tt>sizeof(int)</tt>.
* \retval TH_IMPL Not supported by this implementation.*/
#define TH_ENCCTL_SET_VP3_COMPATIBLE (10)
/**Gets the maximum speed level.
* Higher speed levels favor quicker encoding over better quality per bit.
* Depending on the encoding mode, and the internal algorithms used, quality
* may actually improve, but in this case bitrate will also likely increase.
* In any case, overall rate/distortion performance will probably decrease.
* The maximum value, and the meaning of each value, may change depending on
* the current encoding mode (VBR vs. CQI, etc.).
*
* \param[out] buf int: The maximum encoding speed level.
* \retval TH_FAULT \a theora_state or \a buf is <tt>NULL</tt>.
* \retval TH_EINVAL \a buf_sz is not <tt>sizeof(int)</tt>.
* \retval TH_IMPL Not supported by this implementation in the current
* encoding mode.*/
#define TH_ENCCTL_GET_SPLEVEL_MAX (12)
/**Sets the speed level.
* By default a speed value of 1 is used.
*
* \param[in] buf int: The new encoding speed level.
* 0 is slowest, larger values use less CPU.
* \retval TH_FAULT \a theora_state or \a buf is <tt>NULL</tt>.
* \retval TH_EINVAL \a buf_sz is not <tt>sizeof(int)</tt>, or the
* encoding speed level is out of bounds.
* The maximum encoding speed level may be
* implementation- and encoding mode-specific, and can be
* obtained via #TH_ENCCTL_GET_SPLEVEL_MAX.
* \retval TH_IMPL Not supported by this implementation in the current
* encoding mode.*/
#define TH_ENCCTL_SET_SPLEVEL (14)
/*@}*/
#define OC_FAULT -1 /**< General failure */
#define OC_EINVAL -10 /**< Library encountered invalid internal data */
#define OC_DISABLED -11 /**< Requested action is disabled */
#define OC_BADHEADER -20 /**< Header packet was corrupt/invalid */
#define OC_NOTFORMAT -21 /**< Packet is not a theora packet */
#define OC_VERSION -22 /**< Bitstream version is not handled */
#define OC_IMPL -23 /**< Feature or action not implemented */
#define OC_BADPACKET -24 /**< Packet is corrupt */
#define OC_NEWPACKET -25 /**< Packet is an (ignorable) unhandled extension */
#define OC_DUPFRAME 1 /**< Packet is a dropped frame */
/**
* Retrieve a human-readable string to identify the encoder vendor and version.
* \returns A version string.
*/
extern const char *theora_version_string(void);
/**
* Retrieve a 32-bit version number.
* This number is composed of a 16-bit major version, 8-bit minor version
* and 8 bit sub-version, composed as follows:
<pre>
(VERSION_MAJOR<<16) + (VERSION_MINOR<<8) + (VERSION_SUB)
</pre>
* \returns The version number.
*/
extern ogg_uint32_t theora_version_number(void);
/**
* Initialize the theora encoder.
* \param th The theora_state handle to initialize for encoding.
* \param ti A theora_info struct filled with the desired encoding parameters.
* \retval 0 Success
*/
extern int theora_encode_init(theora_state *th, theora_info *ti);
/**
* Submit a YUV buffer to the theora encoder.
* \param t A theora_state handle previously initialized for encoding.
* \param yuv A buffer of YUV data to encode. Note that both the yuv_buffer
* struct and the luma/chroma buffers within should be allocated by
* the user.
* \retval OC_EINVAL Encoder is not ready, or is finished.
* \retval -1 The size of the given frame differs from those previously input
* \retval 0 Success
*/
extern int theora_encode_YUVin(theora_state *t, yuv_buffer *yuv);
/**
* Request the next packet of encoded video.
* The encoded data is placed in a user-provided ogg_packet structure.
* \param t A theora_state handle previously initialized for encoding.
* \param last_p whether this is the last packet the encoder should produce.
* \param op An ogg_packet structure to fill. libtheora will set all
* elements of this structure, including a pointer to encoded
* data. The memory for the encoded data is owned by libtheora.
* \retval 0 No internal storage exists OR no packet is ready
* \retval -1 The encoding process has completed
* \retval 1 Success
*/
extern int theora_encode_packetout( theora_state *t, int last_p,
ogg_packet *op);
/**
* Request a packet containing the initial header.
* A pointer to the header data is placed in a user-provided ogg_packet
* structure.
* \param t A theora_state handle previously initialized for encoding.
* \param op An ogg_packet structure to fill. libtheora will set all
* elements of this structure, including a pointer to the header
* data. The memory for the header data is owned by libtheora.
* \retval 0 Success
*/
extern int theora_encode_header(theora_state *t, ogg_packet *op);
/**
* Request a comment header packet from provided metadata.
* A pointer to the comment data is placed in a user-provided ogg_packet
* structure.
* \param tc A theora_comment structure filled with the desired metadata
* \param op An ogg_packet structure to fill. libtheora will set all
* elements of this structure, including a pointer to the encoded
* comment data. The memory for the comment data is owned by
* libtheora.
* \retval 0 Success
*/
extern int theora_encode_comment(theora_comment *tc, ogg_packet *op);
/**
* Request a packet containing the codebook tables for the stream.
* A pointer to the codebook data is placed in a user-provided ogg_packet
* structure.
* \param t A theora_state handle previously initialized for encoding.
* \param op An ogg_packet structure to fill. libtheora will set all
* elements of this structure, including a pointer to the codebook
* data. The memory for the header data is owned by libtheora.
* \retval 0 Success
*/
extern int theora_encode_tables(theora_state *t, ogg_packet *op);
/**
* Decode an Ogg packet, with the expectation that the packet contains
* an initial header, comment data or codebook tables.
*
* \param ci A theora_info structure to fill. This must have been previously
* initialized with theora_info_init(). If \a op contains an initial
* header, theora_decode_header() will fill \a ci with the
* parsed header values. If \a op contains codebook tables,
* theora_decode_header() will parse these and attach an internal
* representation to \a ci->codec_setup.
* \param cc A theora_comment structure to fill. If \a op contains comment
* data, theora_decode_header() will fill \a cc with the parsed
* comments.
* \param op An ogg_packet structure which you expect contains an initial
* header, comment data or codebook tables.
*
* \retval OC_BADHEADER \a op is NULL; OR the first byte of \a op->packet
* has the signature of an initial packet, but op is
* not a b_o_s packet; OR this packet has the signature
* of an initial header packet, but an initial header
* packet has already been seen; OR this packet has the
* signature of a comment packet, but the initial header
* has not yet been seen; OR this packet has the signature
* of a comment packet, but contains invalid data; OR
* this packet has the signature of codebook tables,
* but the initial header or comments have not yet
* been seen; OR this packet has the signature of codebook
* tables, but contains invalid data;
* OR the stream being decoded has a compatible version
* but this packet does not have the signature of a
* theora initial header, comments, or codebook packet
* \retval OC_VERSION The packet data of \a op is an initial header with
* a version which is incompatible with this version of
* libtheora.
* \retval OC_NEWPACKET the stream being decoded has an incompatible (future)
* version and contains an unknown signature.
* \retval 0 Success
*
* \note The normal usage is that theora_decode_header() be called on the
* first three packets of a theora logical bitstream in succession.
*/
extern int theora_decode_header(theora_info *ci, theora_comment *cc,
ogg_packet *op);
/**
* Initialize a theora_state handle for decoding.
* \param th The theora_state handle to initialize.
* \param c A theora_info struct filled with the desired decoding parameters.
* This is of course usually obtained from a previous call to
* theora_decode_header().
* \retval 0 Success
*/
extern int theora_decode_init(theora_state *th, theora_info *c);
/**
* Input a packet containing encoded data into the theora decoder.
* \param th A theora_state handle previously initialized for decoding.
* \param op An ogg_packet containing encoded theora data.
* \retval 0 Success
* \retval OC_BADPACKET \a op does not contain encoded video data
*/
extern int theora_decode_packetin(theora_state *th,ogg_packet *op);
/**
* Output the next available frame of decoded YUV data.
* \param th A theora_state handle previously initialized for decoding.
* \param yuv A yuv_buffer in which libtheora should place the decoded data.
* Note that the buffer struct itself is allocated by the user, but
* that the luma and chroma pointers will be filled in by the
* library. Also note that these luma and chroma regions should be
* considered read-only by the user.
* \retval 0 Success
*/
extern int theora_decode_YUVout(theora_state *th,yuv_buffer *yuv);
/**
* Report whether a theora packet is a header or not
* This function does no verification beyond checking the header
* flag bit so it should not be used for bitstream identification;
* use theora_decode_header() for that.
*
* \param op An ogg_packet containing encoded theora data.
* \retval 1 The packet is a header packet
* \retval 0 The packet is not a header packet (and so contains frame data)
*
* Thus function was added in the 1.0alpha4 release.
*/
extern int theora_packet_isheader(ogg_packet *op);
/**
* Report whether a theora packet is a keyframe or not
*
* \param op An ogg_packet containing encoded theora data.
* \retval 1 The packet contains a keyframe image
* \retval 0 The packet is contains an interframe delta
* \retval -1 The packet is not an image data packet at all
*
* Thus function was added in the 1.0alpha4 release.
*/
extern int theora_packet_iskeyframe(ogg_packet *op);
/**
* Report the granulepos shift radix
*
* When embedded in Ogg, Theora uses a two-part granulepos,
* splitting the 64-bit field into two pieces. The more-significant
* section represents the frame count at the last keyframe,
* and the less-significant section represents the count of
* frames since the last keyframe. In this way the overall
* field is still non-decreasing with time, but usefully encodes
* a pointer to the last keyframe, which is necessary for
* correctly restarting decode after a seek.
*
* This function reports the number of bits used to represent
* the distance to the last keyframe, and thus how the granulepos
* field must be shifted or masked to obtain the two parts.
*
* Since libtheora returns compressed data in an ogg_packet
* structure, this may be generally useful even if the Theora
* packets are not being used in an Ogg container.
*
* \param ti A previously initialized theora_info struct
* \returns The bit shift dividing the two granulepos fields
*
* This function was added in the 1.0alpha5 release.
*/
int theora_granule_shift(theora_info *ti);
/**
* Convert a granulepos to an absolute frame index, starting at 0.
* The granulepos is interpreted in the context of a given theora_state handle.
*
* Note that while the granulepos encodes the frame count (i.e. starting
* from 1) this call returns the frame index, starting from zero. Thus
* One can calculate the presentation time by multiplying the index by
* the rate.
*
* \param th A previously initialized theora_state handle (encode or decode)
* \param granulepos The granulepos to convert.
* \returns The frame index corresponding to \a granulepos.
* \retval -1 The given granulepos is undefined (i.e. negative)
*
* Thus function was added in the 1.0alpha4 release.
*/
extern ogg_int64_t theora_granule_frame(theora_state *th,ogg_int64_t granulepos);
/**
* Convert a granulepos to absolute time in seconds. The granulepos is
* interpreted in the context of a given theora_state handle, and gives
* the end time of a frame's presentation as used in Ogg mux ordering.
*
* \param th A previously initialized theora_state handle (encode or decode)
* \param granulepos The granulepos to convert.
* \returns The absolute time in seconds corresponding to \a granulepos.
* This is the "end time" for the frame, or the latest time it should
* be displayed.
* It is not the presentation time.
* \retval -1. The given granulepos is undefined (i.e. negative), or
* \retval -1. The function has been disabled because floating
* point support is not available.
*/
extern double theora_granule_time(theora_state *th,ogg_int64_t granulepos);
/**
* Initialize a theora_info structure. All values within the given theora_info
* structure are initialized, and space is allocated within libtheora for
* internal codec setup data.
* \param c A theora_info struct to initialize.
*/
extern void theora_info_init(theora_info *c);
/**
* Clear a theora_info structure. All values within the given theora_info
* structure are cleared, and associated internal codec setup data is freed.
* \param c A theora_info struct to initialize.
*/
extern void theora_info_clear(theora_info *c);
/**
* Free all internal data associated with a theora_state handle.
* \param t A theora_state handle.
*/
extern void theora_clear(theora_state *t);
/**
* Initialize an allocated theora_comment structure
* \param tc An allocated theora_comment structure
**/
extern void theora_comment_init(theora_comment *tc);
/**
* Add a comment to an initialized theora_comment structure
* \param tc A previously initialized theora comment structure
* \param comment A null-terminated string encoding the comment in the form
* "TAG=the value"
*
* Neither theora_comment_add() nor theora_comment_add_tag() support
* comments containing null values, although the bitstream format
* supports this. To add such comments you will need to manipulate
* the theora_comment structure directly.
**/
extern void theora_comment_add(theora_comment *tc, char *comment);
/**
* Add a comment to an initialized theora_comment structure.
* \param tc A previously initialized theora comment structure
* \param tag A null-terminated string containing the tag
* associated with the comment.
* \param value The corresponding value as a null-terminated string
*
* Neither theora_comment_add() nor theora_comment_add_tag() support
* comments containing null values, although the bitstream format
* supports this. To add such comments you will need to manipulate
* the theora_comment structure directly.
**/
extern void theora_comment_add_tag(theora_comment *tc,
char *tag, char *value);
/**
* Look up a comment value by tag.
* \param tc Tn initialized theora_comment structure
* \param tag The tag to look up
* \param count The instance of the tag. The same tag can appear multiple
* times, each with a distinct and ordered value, so an index
* is required to retrieve them all.
* \returns A pointer to the queried tag's value
* \retval NULL No matching tag is found
*
* \note Use theora_comment_query_count() to get the legal range for the
* count parameter.
**/
extern char *theora_comment_query(theora_comment *tc, char *tag, int count);
/** Look up the number of instances of a tag.
* \param tc An initialized theora_comment structure
* \param tag The tag to look up
* \returns The number on instances of a particular tag.
*
* Call this first when querying for a specific tag and then interate
* over the number of instances with separate calls to
* theora_comment_query() to retrieve all instances in order.
**/
extern int theora_comment_query_count(theora_comment *tc, char *tag);
/**
* Clear an allocated theora_comment struct so that it can be freed.
* \param tc An allocated theora_comment structure.
**/
extern void theora_comment_clear(theora_comment *tc);
/**Encoder control function.
* This is used to provide advanced control the encoding process.
* \param th A #theora_state handle.
* \param req The control code to process.
* See \ref encctlcodes "the list of available control codes"
* for details.
* \param buf The parameters for this control code.
* \param buf_sz The size of the parameter buffer.*/
extern int theora_control(theora_state *th,int req,void *buf,size_t buf_sz);
/* @} */ /* end oldfuncs doxygen group */
#ifdef __cplusplus
}
#endif /* __cplusplus */
#endif /* _O_THEORA_H_ */

307
Engine/lib/libtheora/include/theora/theoradec.h Normal file
View file

@ -0,0 +1,307 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: theora.h,v 1.8 2004/03/15 22:17:32 derf Exp $
********************************************************************/
/**\file
* The <tt>libtheoradec</tt> C decoding API.*/
#if !defined(_O_THEORA_THEORADEC_H_)
# define _O_THEORA_THEORADEC_H_ (1)
# include <stddef.h>
# include <ogg/ogg.h>
# include "codec.h"
#if defined(__cplusplus)
extern "C" {
#endif
/**\name th_decode_ctl() codes
* \anchor decctlcodes
* These are the available request codes for th_decode_ctl().
* By convention, these are odd, to distinguish them from the
* \ref encctlcodes "encoder control codes".
* Keep any experimental or vendor-specific values above \c 0x8000.*/
/*@{*/
/**Gets the maximum post-processing level.
*
* \param[out] _buf int: The maximum post-processing level.
* \retval TH_EFAULT \a _dec_ctx or \a _buf is <tt>NULL</tt>.
* \retval TH_EINVAL \a _buf_sz is not <tt>sizeof(int)</tt>.
* \retval TH_EIMPL Not supported by this implementation.*/
#define TH_DECCTL_GET_PPLEVEL_MAX (1)
/**Sets the post-processing level.
* By default, post-processing is disabled.
*
* \param[in] _buf int: The new post-processing level.
* 0 to disable; larger values use more CPU.
* \retval TH_EFAULT \a _dec_ctx or \a _buf is <tt>NULL</tt>.
* \retval TH_EINVAL \a _buf_sz is not <tt>sizeof(int)</tt>, or the
* post-processing level is out of bounds.
* The maximum post-processing level may be
* implementation-specific, and can be obtained via
* #TH_DECCTL_GET_PPLEVEL_MAX.
* \retval TH_EIMPL Not supported by this implementation.*/
#define TH_DECCTL_SET_PPLEVEL (3)
/**Sets the granule position.
* Call this after a seek, before decoding the first frame, to ensure that the
* proper granule position is returned for all subsequent frames.
* If you track timestamps yourself and do not use the granule position
* returned by the decoder, then you need not call this function.
*
* \param[in] _buf <tt>ogg_int64_t</tt>: The granule position of the next
* frame.
* \retval TH_EFAULT \a _dec_ctx or \a _buf is <tt>NULL</tt>.
* \retval TH_EINVAL \a _buf_sz is not <tt>sizeof(ogg_int64_t)</tt>, or the
* granule position is negative.*/
#define TH_DECCTL_SET_GRANPOS (5)
/**Sets the striped decode callback function.
* If set, this function will be called as each piece of a frame is fully
* decoded in th_decode_packetin().
* You can pass in a #th_stripe_callback with
* th_stripe_callback#stripe_decoded set to <tt>NULL</tt> to disable the
* callbacks at any point.
* Enabling striped decode does not prevent you from calling
* th_decode_ycbcr_out() after the frame is fully decoded.
*
* \param[in] _buf #th_stripe_callback: The callback parameters.
* \retval TH_EFAULT \a _dec_ctx or \a _buf is <tt>NULL</tt>.
* \retval TH_EINVAL \a _buf_sz is not
* <tt>sizeof(th_stripe_callback)</tt>.*/
#define TH_DECCTL_SET_STRIPE_CB (7)
/*@}*/
/**A callback function for striped decode.
* This is a function pointer to an application-provided function that will be
* called each time a section of the image is fully decoded in
* th_decode_packetin().
* This allows the application to process the section immediately, while it is
* still in cache.
* Note that the frame is decoded bottom to top, so \a _yfrag0 will steadily
* decrease with each call until it reaches 0, at which point the full frame
* is decoded.
* The number of fragment rows made available in each call depends on the pixel
* format and the number of post-processing filters enabled, and may not even
* be constant for the entire frame.
* If a non-<tt>NULL</tt> \a _granpos pointer is passed to
* th_decode_packetin(), the granule position for the frame will be stored
* in it before the first callback is made.
* If an entire frame is dropped (a 0-byte packet), then no callbacks will be
* made at all for that frame.
* \param _ctx An application-provided context pointer.
* \param _buf The image buffer for the decoded frame.
* \param _yfrag0 The Y coordinate of the first row of 8x8 fragments
* decoded.
* Multiply this by 8 to obtain the pixel row number in the
* luma plane.
* If the chroma planes are subsampled in the Y direction,
* this will always be divisible by two.
* \param _yfrag_end The Y coordinate of the first row of 8x8 fragments past
* the newly decoded section.
* If the chroma planes are subsampled in the Y direction,
* this will always be divisible by two.
* I.e., this section contains fragment rows
* <tt>\a _yfrag0 ...\a _yfrag_end -1</tt>.*/
typedef void (*th_stripe_decoded_func)(void *_ctx,th_ycbcr_buffer _buf,
int _yfrag0,int _yfrag_end);
/**The striped decode callback data to pass to #TH_DECCTL_SET_STRIPE_CB.*/
typedef struct{
/**An application-provided context pointer.
* This will be passed back verbatim to the application.*/
void *ctx;
/**The callback function pointer.*/
th_stripe_decoded_func stripe_decoded;
}th_stripe_callback;
/**\name Decoder state
The following data structures are opaque, and their contents are not
publicly defined by this API.
Referring to their internals directly is unsupported, and may break without
warning.*/
/*@{*/
/**The decoder context.*/
typedef struct th_dec_ctx th_dec_ctx;
/**Setup information.
This contains auxiliary information (Huffman tables and quantization
parameters) decoded from the setup header by th_decode_headerin() to be
passed to th_decode_alloc().
It can be re-used to initialize any number of decoders, and can be freed
via th_setup_free() at any time.*/
typedef struct th_setup_info th_setup_info;
/*@}*/
/**\defgroup decfuncs Functions for Decoding*/
/*@{*/
/**\name Functions for decoding
* You must link to <tt>libtheoradec</tt> if you use any of the
* functions in this section.
*
* The functions are listed in the order they are used in a typical decode.
* The basic steps are:
* - Parse the header packets by repeatedly calling th_decode_headerin().
* - Allocate a #th_dec_ctx handle with th_decode_alloc().
* - Call th_setup_free() to free any memory used for codec setup
* information.
* - Perform any additional decoder configuration with th_decode_ctl().
* - For each video data packet:
* - Submit the packet to the decoder via th_decode_packetin().
* - Retrieve the uncompressed video data via th_decode_ycbcr_out().
* - Call th_decode_free() to release all decoder memory.*/
/*@{*/
/**Decodes the header packets of a Theora stream.
* This should be called on the initial packets of the stream, in succession,
* until it returns <tt>0</tt>, indicating that all headers have been
* processed, or an error is encountered.
* At least three header packets are required, and additional optional header
* packets may follow.
* This can be used on the first packet of any logical stream to determine if
* that stream is a Theora stream.
* \param _info A #th_info structure to fill in.
* This must have been previously initialized with
* th_info_init().
* The application may immediately begin using the contents of
* this structure after the first header is decoded, though it
* must continue to be passed in on all subsequent calls.
* \param _tc A #th_comment structure to fill in.
* The application may immediately begin using the contents of
* this structure after the second header is decoded, though it
* must continue to be passed in on all subsequent calls.
* \param _setup Returns a pointer to additional, private setup information
* needed by the decoder.
* The contents of this pointer must be initialized to
* <tt>NULL</tt> on the first call, and the returned value must
* continue to be passed in on all subsequent calls.
* \param _op An <tt>ogg_packet</tt> structure which contains one of the
* initial packets of an Ogg logical stream.
* \return A positive value indicates that a Theora header was successfully
* processed.
* \retval 0 The first video data packet was encountered after all
* required header packets were parsed.
* The packet just passed in on this call should be saved
* and fed to th_decode_packetin() to begin decoding
* video data.
* \retval TH_EFAULT One of \a _info, \a _tc, or \a _setup was
* <tt>NULL</tt>.
* \retval TH_EBADHEADER \a _op was <tt>NULL</tt>, the packet was not the next
* header packet in the expected sequence, or the format
* of the header data was invalid.
* \retval TH_EVERSION The packet data was a Theora info header, but for a
* bitstream version not decodable with this version of
* <tt>libtheoradec</tt>.
* \retval TH_ENOTFORMAT The packet was not a Theora header.
*/
extern int th_decode_headerin(th_info *_info,th_comment *_tc,
th_setup_info **_setup,ogg_packet *_op);
/**Allocates a decoder instance.
*
* <b>Security Warning:</b> The Theora format supports very large frame sizes,
* potentially even larger than the address space of a 32-bit machine, and
* creating a decoder context allocates the space for several frames of data.
* If the allocation fails here, your program will crash, possibly at some
* future point because the OS kernel returned a valid memory range and will
* only fail when it tries to map the pages in it the first time they are
* used.
* Even if it succeeds, you may experience a denial of service if the frame
* size is large enough to cause excessive paging.
* If you are integrating libtheora in a larger application where such things
* are undesirable, it is highly recommended that you check the frame size in
* \a _info before calling this function and refuse to decode streams where it
* is larger than some reasonable maximum.
* libtheora will not check this for you, because there may be machines that
* can handle such streams and applications that wish to.
* \param _info A #th_info struct filled via th_decode_headerin().
* \param _setup A #th_setup_info handle returned via
* th_decode_headerin().
* \return The initialized #th_dec_ctx handle.
* \retval NULL If the decoding parameters were invalid.*/
extern th_dec_ctx *th_decode_alloc(const th_info *_info,
const th_setup_info *_setup);
/**Releases all storage used for the decoder setup information.
* This should be called after you no longer want to create any decoders for
* a stream whose headers you have parsed with th_decode_headerin().
* \param _setup The setup information to free.
* This can safely be <tt>NULL</tt>.*/
extern void th_setup_free(th_setup_info *_setup);
/**Decoder control function.
* This is used to provide advanced control of the decoding process.
* \param _dec A #th_dec_ctx handle.
* \param _req The control code to process.
* See \ref decctlcodes "the list of available control codes"
* for details.
* \param _buf The parameters for this control code.
* \param _buf_sz The size of the parameter buffer.*/
extern int th_decode_ctl(th_dec_ctx *_dec,int _req,void *_buf,
size_t _buf_sz);
/**Submits a packet containing encoded video data to the decoder.
* \param _dec A #th_dec_ctx handle.
* \param _op An <tt>ogg_packet</tt> containing encoded video data.
* \param _granpos Returns the granule position of the decoded packet.
* If non-<tt>NULL</tt>, the granule position for this specific
* packet is stored in this location.
* This is computed incrementally from previously decoded
* packets.
* After a seek, the correct granule position must be set via
* #TH_DECCTL_SET_GRANPOS for this to work properly.
* \retval 0 Success.
* A new decoded frame can be retrieved by calling
* th_decode_ycbcr_out().
* \retval TH_DUPFRAME The packet represented a dropped (0-byte) frame.
* The player can skip the call to th_decode_ycbcr_out(),
* as the contents of the decoded frame buffer have not
* changed.
* \retval TH_EFAULT \a _dec or \a _op was <tt>NULL</tt>.
* \retval TH_EBADPACKET \a _op does not contain encoded video data.
* \retval TH_EIMPL The video data uses bitstream features which this
* library does not support.*/
extern int th_decode_packetin(th_dec_ctx *_dec,const ogg_packet *_op,
ogg_int64_t *_granpos);
/**Outputs the next available frame of decoded Y'CbCr data.
* If a striped decode callback has been set with #TH_DECCTL_SET_STRIPE_CB,
* then the application does not need to call this function.
* \param _dec A #th_dec_ctx handle.
* \param _ycbcr A video buffer structure to fill in.
* <tt>libtheoradec</tt> will fill in all the members of this
* structure, including the pointers to the uncompressed video
* data.
* The memory for this video data is owned by
* <tt>libtheoradec</tt>.
* It may be freed or overwritten without notification when
* subsequent frames are decoded.
* \retval 0 Success
*/
extern int th_decode_ycbcr_out(th_dec_ctx *_dec,
th_ycbcr_buffer _ycbcr);
/**Frees an allocated decoder instance.
* \param _dec A #th_dec_ctx handle.*/
extern void th_decode_free(th_dec_ctx *_dec);
/*@}*/
/*@}*/
#if defined(__cplusplus)
}
#endif
#endif

266
Engine/lib/libtheora/include/theora/theoraenc.h Normal file
View file

@ -0,0 +1,266 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2003 *
* by the Xiph.Org Foundation http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: theora.h,v 1.8 2004/03/15 22:17:32 derf Exp $
********************************************************************/
/**\file
* The <tt>libtheoraenc</tt> C encoding API.*/
#if !defined(_O_THEORA_THEORAENC_H_)
# define _O_THEORA_THEORAENC_H_ (1)
# include <stddef.h>
# include <ogg/ogg.h>
# include "codec.h"
#if defined(__cplusplus)
extern "C" {
#endif
/**\name th_encode_ctl() codes
* \anchor encctlcodes
* These are the available request codes for th_encode_ctl().
* By convention, these are even, to distinguish them from the
* \ref decctlcodes "decoder control codes".
* Keep any experimental or vendor-specific values above \c 0x8000.*/
/*@{*/
/**Sets the Huffman tables to use.
* The tables are copied, not stored by reference, so they can be freed after
* this call.
* <tt>NULL</tt> may be specified to revert to the default tables.
*
* \param[in] _buf <tt>#th_huff_code[#TH_NHUFFMAN_TABLES][#TH_NDCT_TOKENS]</tt>
* \retval TH_EFAULT \a _enc_ctx is <tt>NULL</tt>.
* \retval TH_EINVAL Encoding has already begun or one or more of the given
* tables is not full or prefix-free, \a _buf is
* <tt>NULL</tt> and \a _buf_sz is not zero, or \a _buf is
* non-<tt>NULL</tt> and \a _buf_sz is not
* <tt>sizeof(#th_huff_code)*#TH_NHUFFMAN_TABLES*#TH_NDCT_TOKENS</tt>.
* \retval TH_IMPL Not supported by this implementation.*/
#define TH_ENCCTL_SET_HUFFMAN_CODES (0)
/**Sets the quantization parameters to use.
* The parameters are copied, not stored by reference, so they can be freed
* after this call.
* <tt>NULL</tt> may be specified to revert to the default parameters.
* For the current encoder, <tt>scale[ci!=0][qi]</tt> must be no greater than
* <tt>scale[ci!=0][qi-1]</tt> and <tt>base[qti][pli][qi][ci]</tt> must be no
* greater than <tt>base[qti][pli][qi-1][ci]</tt>.
* These two conditions ensure that the actual quantizer for a given \a qti,
* \a pli, and \a ci does not increase as \a qi increases.
*
* \param[in] _buf #th_quant_info
* \retval TH_EFAULT \a _enc_ctx is <tt>NULL</tt>.
* \retval TH_EINVAL Encoding has already begun, the quantization parameters
* do not meet one of the above stated conditions, \a _buf
* is <tt>NULL</tt> and \a _buf_sz is not zero, or \a _buf
* is non-<tt>NULL</tt> and \a _buf_sz is not
* <tt>sizeof(#th_quant_info)</tt>.
* \retval TH_IMPL Not supported by this implementation.*/
#define TH_ENCCTL_SET_QUANT_PARAMS (2)
/**Sets the maximum distance between key frames.
* This can be changed during an encode, but will be bounded by
* <tt>1<<th_info#keyframe_granule_shift</tt>.
* If it is set before encoding begins, th_info#keyframe_granule_shift will
* be enlarged appropriately.
*
* \param[in] _buf <tt>ogg_uint32_t</tt>: The maximum distance between key
* frames.
* \param[out] _buf <tt>ogg_uint32_t</tt>: The actual maximum distance set.
* \retval TH_EFAULT \a _enc_ctx or \a _buf is <tt>NULL</tt>.
* \retval TH_EINVAL \a _buf_sz is not <tt>sizeof(ogg_uint32_t)</tt>.
* \retval TH_IMPL Not supported by this implementation.*/
#define TH_ENCCTL_SET_KEYFRAME_FREQUENCY_FORCE (4)
/**Disables any encoder features that would prevent lossless transcoding back
* to VP3.
* This primarily means disabling block-level QI values and not using 4MV mode
* when any of the luma blocks in a macro block are not coded.
* It also includes using the VP3 quantization tables and Huffman codes; if you
* set them explicitly after calling this function, the resulting stream will
* not be VP3-compatible.
* If you enable VP3-compatibility when encoding 4:2:2 or 4:4:4 source
* material, or when using a picture region smaller than the full frame (e.g.
* a non-multiple-of-16 width or height), then non-VP3 bitstream features will
* still be disabled, but the stream will still not be VP3-compatible, as VP3
* was not capable of encoding such formats.
* If you call this after encoding has already begun, then the quantization
* tables and codebooks cannot be changed, but the frame-level features will
* be enabled or disabled as requested.
*
* \param[in] _buf <tt>int</tt>: a non-zero value to enable VP3 compatibility,
* or 0 to disable it (the default).
* \param[out] _buf <tt>int</tt>: 1 if all bitstream features required for
* VP3-compatibility could be set, and 0 otherwise.
* The latter will be returned if the pixel format is not
* 4:2:0, the picture region is smaller than the full frame,
* or if encoding has begun, preventing the quantization
* tables and codebooks from being set.
* \retval TH_EFAULT \a _enc_ctx or \a _buf is <tt>NULL</tt>.
* \retval TH_EINVAL \a _buf_sz is not <tt>sizeof(int)</tt>.
* \retval TH_IMPL Not supported by this implementation.*/
#define TH_ENCCTL_SET_VP3_COMPATIBLE (10)
/**Gets the maximum speed level.
* Higher speed levels favor quicker encoding over better quality per bit.
* Depending on the encoding mode, and the internal algorithms used, quality
* may actually improve, but in this case bitrate will also likely increase.
* In any case, overall rate/distortion performance will probably decrease.
* The maximum value, and the meaning of each value, may change depending on
* the current encoding mode (VBR vs. CQI, etc.).
*
* \param[out] _buf int: The maximum encoding speed level.
* \retval TH_EFAULT \a _enc_ctx or \a _buf is <tt>NULL</tt>.
* \retval TH_EINVAL \a _buf_sz is not <tt>sizeof(int)</tt>.
* \retval TH_IMPL Not supported by this implementation in the current
* encoding mode.*/
#define TH_ENCCTL_GET_SPLEVEL_MAX (12)
/**Sets the speed level.
* By default, the slowest speed (0) is used.
*
* \param[in] _buf int: The new encoding speed level.
* 0 is slowest, larger values use less CPU.
* \retval TH_EFAULT \a _enc_ctx or \a _buf is <tt>NULL</tt>.
* \retval TH_EINVAL \a _buf_sz is not <tt>sizeof(int)</tt>, or the
* encoding speed level is out of bounds.
* The maximum encoding speed level may be
* implementation- and encoding mode-specific, and can be
* obtained via #TH_ENCCTL_GET_SPLEVEL_MAX.
* \retval TH_IMPL Not supported by this implementation in the current
* encoding mode.*/
#define TH_ENCCTL_SET_SPLEVEL (14)
/*@}*/
/**The quantization parameters used by VP3.*/
extern const th_quant_info TH_VP31_QUANT_INFO;
/**The Huffman tables used by VP3.*/
extern const th_huff_code
TH_VP31_HUFF_CODES[TH_NHUFFMAN_TABLES][TH_NDCT_TOKENS];
/**\name Encoder state
The following data structure is opaque, and its contents are not publicly
defined by this API.
Referring to its internals directly is unsupported, and may break without
warning.*/
/*@{*/
/**The encoder context.*/
typedef struct th_enc_ctx th_enc_ctx;
/*@}*/
/**\defgroup encfuncs Functions for Encoding*/
/*@{*/
/**\name Functions for encoding
* You must link to <tt>libtheoraenc</tt> and <tt>libtheoradec</tt>
* if you use any of the functions in this section.
*
* The functions are listed in the order they are used in a typical encode.
* The basic steps are:
* - Fill in a #th_info structure with details on the format of the video you
* wish to encode.
* - Allocate a #th_enc_ctx handle with th_encode_alloc().
* - Perform any additional encoder configuration required with
* th_encode_ctl().
* - Repeatedly call th_encode_flushheader() to retrieve all the header
* packets.
* - For each uncompressed frame:
* - Submit the uncompressed frame via th_encode_ycbcr_in()
* - Repeatedly call th_encode_packetout() to retrieve any video data packets
* that are ready.
* - Call th_encode_free() to release all encoder memory.*/
/*@{*/
/**Allocates an encoder instance.
* \param _info A #th_info struct filled with the desired encoding parameters.
* \return The initialized #th_enc_ctx handle.
* \retval NULL If the encoding parameters were invalid.*/
extern th_enc_ctx *th_encode_alloc(const th_info *_info);
/**Encoder control function.
* This is used to provide advanced control the encoding process.
* \param _enc A #th_enc_ctx handle.
* \param _req The control code to process.
* See \ref encctlcodes "the list of available control codes"
* for details.
* \param _buf The parameters for this control code.
* \param _buf_sz The size of the parameter buffer.*/
extern int th_encode_ctl(th_enc_ctx *_enc,int _req,void *_buf,size_t _buf_sz);
/**Outputs the next header packet.
* This should be called repeatedly after encoder initialization until it
* returns 0 in order to get all of the header packets, in order, before
* encoding actual video data.
* \param _enc A #th_enc_ctx handle.
* \param _comments The metadata to place in the comment header, when it is
* encoded.
* \param _op An <tt>ogg_packet</tt> structure to fill.
* All of the elements of this structure will be set,
* including a pointer to the header data.
* The memory for the header data is owned by
* <tt>libtheoraenc</tt>, and may be invalidated when the
* next encoder function is called.
* \return A positive value indicates that a header packet was successfully
* produced.
* \retval 0 No packet was produced, and no more header packets remain.
* \retval TH_EFAULT \a _enc, \a _comments, or \a _op was <tt>NULL</tt>.*/
extern int th_encode_flushheader(th_enc_ctx *_enc,
th_comment *_comments,ogg_packet *_op);
/**Submits an uncompressed frame to the encoder.
* \param _enc A #th_enc_ctx handle.
* \param _ycbcr A buffer of Y'CbCr data to encode.
* \retval 0 Success.
* \retval TH_EFAULT \a _enc or \a _ycbcr is <tt>NULL</tt>.
* \retval TH_EINVAL The buffer size does not match the frame size the encoder
* was initialized with, or encoding has already
* completed.*/
extern int th_encode_ycbcr_in(th_enc_ctx *_enc,th_ycbcr_buffer _ycbcr);
/**Retrieves encoded video data packets.
* This should be called repeatedly after each frame is submitted to flush any
* encoded packets, until it returns 0.
* The encoder will not buffer these packets as subsequent frames are
* compressed, so a failure to do so will result in lost video data.
* \note Currently the encoder operates in a one-frame-in, one-packet-out
* manner.
* However, this may be changed in the future.
* \param _enc A #th_enc_ctx handle.
* \param _last Set this flag to a non-zero value if no more uncompressed
* frames will be submitted.
* This ensures that a proper EOS flag is set on the last packet.
* \param _op An <tt>ogg_packet</tt> structure to fill.
* All of the elements of this structure will be set, including a
* pointer to the video data.
* The memory for the video data is owned by
* <tt>libtheoraenc</tt>, and may be invalidated when the next
* encoder function is called.
* \return A positive value indicates that a video data packet was successfully
* produced.
* \retval 0 No packet was produced, and no more encoded video data
* remains.
* \retval TH_EFAULT \a _enc or \a _op was <tt>NULL</tt>.*/
extern int th_encode_packetout(th_enc_ctx *_enc,int _last,ogg_packet *_op);
/**Frees an allocated encoder instance.
* \param _enc A #th_enc_ctx handle.*/
extern void th_encode_free(th_enc_ctx *_enc);
/*@}*/
/*@}*/
#if defined(__cplusplus)
}
#endif
#endif

227
Engine/lib/libtheora/lib/cpu.c Normal file
View file

@ -0,0 +1,227 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2008 *
* by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
* *
********************************************************************
CPU capability detection for x86 processors.
Originally written by Rudolf Marek.
function:
last mod: $Id: cpu.c 15427 2008-10-21 02:36:19Z xiphmont $
********************************************************************/
#include "cpu.h"
#if !defined(USE_ASM)
static ogg_uint32_t oc_cpu_flags_get(void){
return 0;
}
#else
# if !defined(_MSC_VER)
# if defined(__amd64__)||defined(__x86_64__)
/*On x86-64, gcc seems to be able to figure out how to save %rbx for us when
compiling with -fPIC.*/
# define cpuid(_op,_eax,_ebx,_ecx,_edx) \
__asm__ __volatile__( \
"cpuid\n\t" \
:[eax]"=a"(_eax),[ebx]"=b"(_ebx),[ecx]"=c"(_ecx),[edx]"=d"(_edx) \
:"a"(_op) \
:"cc" \
)
# else
/*On x86-32, not so much.*/
# define cpuid(_op,_eax,_ebx,_ecx,_edx) \
__asm__ __volatile__( \
"xchgl %%ebx,%[ebx]\n\t" \
"cpuid\n\t" \
"xchgl %%ebx,%[ebx]\n\t" \
:[eax]"=a"(_eax),[ebx]"=r"(_ebx),[ecx]"=c"(_ecx),[edx]"=d"(_edx) \
:"a"(_op) \
:"cc" \
)
# endif
# else
/*Why does MSVC need this complicated rigamarole?
At this point I honestly do not care.*/
/*Visual C cpuid helper function.
For VS2005 we could as well use the _cpuid builtin, but that wouldn't work
for VS2003 users, so we do it in inline assembler.*/
static void oc_cpuid_helper(ogg_uint32_t _cpu_info[4],ogg_uint32_t _op){
_asm{
mov eax,[_op]
mov esi,_cpu_info
cpuid
mov [esi+0],eax
mov [esi+4],ebx
mov [esi+8],ecx
mov [esi+12],edx
}
}
# define cpuid(_op,_eax,_ebx,_ecx,_edx) \
do{ \
ogg_uint32_t cpu_info[4]; \
oc_cpuid_helper(cpu_info,_op); \
(_eax)=cpu_info[0]; \
(_ebx)=cpu_info[1]; \
(_ecx)=cpu_info[2]; \
(_edx)=cpu_info[3]; \
}while(0)
static void oc_detect_cpuid_helper(ogg_uint32_t *_eax,ogg_uint32_t *_ebx){
_asm{
pushfd
pushfd
pop eax
mov ebx,eax
xor eax,200000h
push eax
popfd
pushfd
pop eax
popfd
mov ecx,_eax
mov [ecx],eax
mov ecx,_ebx
mov [ecx],ebx
}
}
# endif
static ogg_uint32_t oc_parse_intel_flags(ogg_uint32_t _edx,ogg_uint32_t _ecx){
ogg_uint32_t flags;
/*If there isn't even MMX, give up.*/
if(!(_edx&0x00800000))return 0;
flags=OC_CPU_X86_MMX;
if(_edx&0x02000000)flags|=OC_CPU_X86_MMXEXT|OC_CPU_X86_SSE;
if(_edx&0x04000000)flags|=OC_CPU_X86_SSE2;
if(_ecx&0x00000001)flags|=OC_CPU_X86_PNI;
if(_ecx&0x00000100)flags|=OC_CPU_X86_SSSE3;
if(_ecx&0x00080000)flags|=OC_CPU_X86_SSE4_1;
if(_ecx&0x00100000)flags|=OC_CPU_X86_SSE4_2;
return flags;
}
static ogg_uint32_t oc_parse_amd_flags(ogg_uint32_t _edx,ogg_uint32_t _ecx){
ogg_uint32_t flags;
/*If there isn't even MMX, give up.*/
if(!(_edx&0x00800000))return 0;
flags=OC_CPU_X86_MMX;
if(_edx&0x00400000)flags|=OC_CPU_X86_MMXEXT;
if(_edx&0x80000000)flags|=OC_CPU_X86_3DNOW;
if(_edx&0x40000000)flags|=OC_CPU_X86_3DNOWEXT;
if(_ecx&0x00000040)flags|=OC_CPU_X86_SSE4A;
if(_ecx&0x00000800)flags|=OC_CPU_X86_SSE5;
return flags;
}
static ogg_uint32_t oc_cpu_flags_get(void){
ogg_uint32_t flags;
ogg_uint32_t eax;
ogg_uint32_t ebx;
ogg_uint32_t ecx;
ogg_uint32_t edx;
# if !defined(__amd64__)&&!defined(__x86_64__)
/*Not all x86-32 chips support cpuid, so we have to check.*/
# if !defined(_MSC_VER)
__asm__ __volatile__(
"pushfl\n\t"
"pushfl\n\t"
"popl %[a]\n\t"
"movl %[a],%[b]\n\t"
"xorl $0x200000,%[a]\n\t"
"pushl %[a]\n\t"
"popfl\n\t"
"pushfl\n\t"
"popl %[a]\n\t"
"popfl\n\t"
:[a]"=r"(eax),[b]"=r"(ebx)
:
:"cc"
);
# else
oc_detect_cpuid_helper(&eax,&ebx);
# endif
/*No cpuid.*/
if(eax==ebx)return 0;
# endif
cpuid(0,eax,ebx,ecx,edx);
/* l e t n I e n i u n e G*/
if(ecx==0x6C65746E&&edx==0x49656E69&&ebx==0x756E6547||
/* 6 8 x M T e n i u n e G*/
ecx==0x3638784D&&edx==0x54656E69&&ebx==0x756E6547){
/*Intel, Transmeta (tested with Crusoe TM5800):*/
cpuid(1,eax,ebx,ecx,edx);
flags=oc_parse_intel_flags(edx,ecx);
}
/* D M A c i t n e h t u A*/
else if(ecx==0x444D4163&&edx==0x69746E65&&ebx==0x68747541||
/* C S N y b e d o e G*/
ecx==0x43534E20&&edx==0x79622065&&ebx==0x646F6547){
/*AMD, Geode:*/
cpuid(0x80000000,eax,ebx,ecx,edx);
if(eax<0x80000001)flags=0;
else{
cpuid(0x80000001,eax,ebx,ecx,edx);
flags=oc_parse_amd_flags(edx,ecx);
}
/*Also check for SSE.*/
cpuid(1,eax,ebx,ecx,edx);
flags|=oc_parse_intel_flags(edx,ecx);
}
/*Technically some VIA chips can be configured in the BIOS to return any
string here the user wants.
There is a special detection method that can be used to identify such
processors, but in my opinion, if the user really wants to change it, they
deserve what they get.*/
/* s l u a H r u a t n e C*/
else if(ecx==0x736C7561&&edx==0x48727561&&ebx==0x746E6543){
/*VIA:*/
/*I only have documentation for the C7 (Esther) and Isaiah (forthcoming)
chips (thanks to the engineers from Centaur Technology who provided it).
These chips support Intel-like cpuid info.
The C3-2 (Nehemiah) cores appear to, as well.*/
cpuid(1,eax,ebx,ecx,edx);
flags=oc_parse_intel_flags(edx,ecx);
cpuid(0x80000000,eax,ebx,ecx,edx);
if(eax>=0x80000001){
/*The (non-Nehemiah) C3 processors support AMD-like cpuid info.
We need to check this even if the Intel test succeeds to pick up 3DNow!
support on these processors.
Unlike actual AMD processors, we cannot _rely_ on this info, since
some cores (e.g., the 693 stepping of the Nehemiah) claim to support
this function, yet return edx=0, despite the Intel test indicating
MMX support.
Therefore the features detected here are strictly added to those
detected by the Intel test.*/
/*TODO: How about earlier chips?*/
cpuid(0x80000001,eax,ebx,ecx,edx);
/*Note: As of the C7, this function returns Intel-style extended feature
flags, not AMD-style.
Currently, this only defines bits 11, 20, and 29 (0x20100800), which
do not conflict with any of the AMD flags we inspect.
For the remaining bits, Intel tells us, "Do not count on their value",
but VIA assures us that they will all be zero (at least on the C7 and
Isaiah chips).
In the (unlikely) event a future processor uses bits 18, 19, 30, or 31
(0xC0C00000) for something else, we will have to add code to detect
the model to decide when it is appropriate to inspect them.*/
flags|=oc_parse_amd_flags(edx,ecx);
}
}
else{
/*Implement me.*/
flags=0;
}
return flags;
}
#endif

34
Engine/lib/libtheora/lib/cpu.h Normal file
View file

@ -0,0 +1,34 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: cpu.h 15430 2008-10-21 05:03:55Z giles $
********************************************************************/
#if !defined(_x86_cpu_H)
# define _x86_cpu_H (1)
#include "internal.h"
#define OC_CPU_X86_MMX (1<<0)
#define OC_CPU_X86_3DNOW (1<<1)
#define OC_CPU_X86_3DNOWEXT (1<<2)
#define OC_CPU_X86_MMXEXT (1<<3)
#define OC_CPU_X86_SSE (1<<4)
#define OC_CPU_X86_SSE2 (1<<5)
#define OC_CPU_X86_PNI (1<<6)
#define OC_CPU_X86_SSSE3 (1<<7)
#define OC_CPU_X86_SSE4_1 (1<<8)
#define OC_CPU_X86_SSE4_2 (1<<9)
#define OC_CPU_X86_SSE4A (1<<10)
#define OC_CPU_X86_SSE5 (1<<11)
#endif

166
Engine/lib/libtheora/lib/dec/apiwrapper.c Normal file
View file

@ -0,0 +1,166 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: apiwrapper.c 15400 2008-10-15 12:10:58Z tterribe $
********************************************************************/
#include <stdlib.h>
#include <string.h>
#include <limits.h>
#include "apiwrapper.h"
const char *theora_version_string(void){
return th_version_string();
}
ogg_uint32_t theora_version_number(void){
return th_version_number();
}
void theora_info_init(theora_info *_ci){
memset(_ci,0,sizeof(*_ci));
}
void theora_info_clear(theora_info *_ci){
th_api_wrapper *api;
api=(th_api_wrapper *)_ci->codec_setup;
memset(_ci,0,sizeof(*_ci));
if(api!=NULL){
if(api->clear!=NULL)(*api->clear)(api);
_ogg_free(api);
}
}
void theora_clear(theora_state *_th){
/*Provide compatibility with mixed encoder and decoder shared lib versions.*/
if(_th->internal_decode!=NULL){
(*((oc_state_dispatch_vtbl *)_th->internal_decode)->clear)(_th);
}
if(_th->internal_encode!=NULL){
(*((oc_state_dispatch_vtbl *)_th->internal_encode)->clear)(_th);
}
if(_th->i!=NULL)theora_info_clear(_th->i);
memset(_th,0,sizeof(*_th));
}
int theora_control(theora_state *_th,int _req,void *_buf,size_t _buf_sz){
/*Provide compatibility with mixed encoder and decoder shared lib versions.*/
if(_th->internal_decode!=NULL){
return (*((oc_state_dispatch_vtbl *)_th->internal_decode)->control)(_th,
_req,_buf,_buf_sz);
}
else if(_th->internal_encode!=NULL){
return (*((oc_state_dispatch_vtbl *)_th->internal_encode)->control)(_th,
_req,_buf,_buf_sz);
}
else return TH_EINVAL;
}
ogg_int64_t theora_granule_frame(theora_state *_th,ogg_int64_t _gp){
/*Provide compatibility with mixed encoder and decoder shared lib versions.*/
if(_th->internal_decode!=NULL){
return (*((oc_state_dispatch_vtbl *)_th->internal_decode)->granule_frame)(
_th,_gp);
}
else if(_th->internal_encode!=NULL){
return (*((oc_state_dispatch_vtbl *)_th->internal_encode)->granule_frame)(
_th,_gp);
}
else return -1;
}
double theora_granule_time(theora_state *_th, ogg_int64_t _gp){
/*Provide compatibility with mixed encoder and decoder shared lib versions.*/
if(_th->internal_decode!=NULL){
return (*((oc_state_dispatch_vtbl *)_th->internal_decode)->granule_time)(
_th,_gp);
}
else if(_th->internal_encode!=NULL){
return (*((oc_state_dispatch_vtbl *)_th->internal_encode)->granule_time)(
_th,_gp);
}
else return -1;
}
void oc_theora_info2th_info(th_info *_info,const theora_info *_ci){
_info->version_major=_ci->version_major;
_info->version_minor=_ci->version_minor;
_info->version_subminor=_ci->version_subminor;
_info->frame_width=_ci->width;
_info->frame_height=_ci->height;
_info->pic_width=_ci->frame_width;
_info->pic_height=_ci->frame_height;
_info->pic_x=_ci->offset_x;
_info->pic_y=_ci->offset_y;
_info->fps_numerator=_ci->fps_numerator;
_info->fps_denominator=_ci->fps_denominator;
_info->aspect_numerator=_ci->aspect_numerator;
_info->aspect_denominator=_ci->aspect_denominator;
switch(_ci->colorspace){
case OC_CS_ITU_REC_470M:_info->colorspace=TH_CS_ITU_REC_470M;break;
case OC_CS_ITU_REC_470BG:_info->colorspace=TH_CS_ITU_REC_470BG;break;
default:_info->colorspace=TH_CS_UNSPECIFIED;break;
}
switch(_ci->pixelformat){
case OC_PF_420:_info->pixel_fmt=TH_PF_420;break;
case OC_PF_422:_info->pixel_fmt=TH_PF_422;break;
case OC_PF_444:_info->pixel_fmt=TH_PF_444;break;
default:_info->pixel_fmt=TH_PF_RSVD;
}
_info->target_bitrate=_ci->target_bitrate;
_info->quality=_ci->quality;
_info->keyframe_granule_shift=_ci->keyframe_frequency_force>0?
OC_MINI(31,oc_ilog(_ci->keyframe_frequency_force-1)):0;
}
int theora_packet_isheader(ogg_packet *_op){
return th_packet_isheader(_op);
}
int theora_packet_iskeyframe(ogg_packet *_op){
return th_packet_iskeyframe(_op);
}
int theora_granule_shift(theora_info *_ci){
/*This breaks when keyframe_frequency_force is not positive or is larger than
2**31 (if your int is more than 32 bits), but that's what the original
function does.*/
return oc_ilog(_ci->keyframe_frequency_force-1);
}
void theora_comment_init(theora_comment *_tc){
th_comment_init((th_comment *)_tc);
}
char *theora_comment_query(theora_comment *_tc,char *_tag,int _count){
return th_comment_query((th_comment *)_tc,_tag,_count);
}
int theora_comment_query_count(theora_comment *_tc,char *_tag){
return th_comment_query_count((th_comment *)_tc,_tag);
}
void theora_comment_clear(theora_comment *_tc){
th_comment_clear((th_comment *)_tc);
}
void theora_comment_add(theora_comment *_tc,char *_comment){
th_comment_add((th_comment *)_tc,_comment);
}
void theora_comment_add_tag(theora_comment *_tc, char *_tag, char *_value){
th_comment_add_tag((th_comment *)_tc,_tag,_value);
}

55
Engine/lib/libtheora/lib/dec/apiwrapper.h Normal file
View file

@ -0,0 +1,55 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: apiwrapper.h 13596 2007-08-23 20:05:38Z tterribe $
********************************************************************/
#if !defined(_apiwrapper_H)
# define _apiwrapper_H (1)
# include <ogg/ogg.h>
# include <theora/theora.h>
# include "theora/theoradec.h"
/*# include "theora/theoraenc.h"*/
typedef struct th_enc_ctx th_enc_ctx;
# include "../internal.h"
typedef struct th_api_wrapper th_api_wrapper;
typedef struct th_api_info th_api_info;
/*Provide an entry point for the codec setup to clear itself in case we ever
want to break pieces off into a common base library shared by encoder and
decoder.
In addition, this makes several other pieces of the API wrapper cleaner.*/
typedef void (*oc_setup_clear_func)(void *_ts);
/*Generally only one of these pointers will be non-NULL in any given instance.
Technically we do not even really need this struct, since we should be able
to figure out which one from "context", but doing it this way makes sure we
don't flub it up.*/
struct th_api_wrapper{
oc_setup_clear_func clear;
th_setup_info *setup;
th_dec_ctx *decode;
th_enc_ctx *encode;
};
struct th_api_info{
th_api_wrapper api;
theora_info info;
};
void oc_theora_info2th_info(th_info *_info,const theora_info *_ci);
#endif

121
Engine/lib/libtheora/lib/dec/bitpack.c Normal file
View file

@ -0,0 +1,121 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE OggTheora SOURCE CODE IS (C) COPYRIGHT 1994-2008 *
* by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
* *
********************************************************************
function: packing variable sized words into an octet stream
last mod: $Id: bitpack.c 15400 2008-10-15 12:10:58Z tterribe $
********************************************************************/
/*We're 'MSb' endian; if we write a word but read individual bits,
then we'll read the MSb first.*/
#include <string.h>
#include <stdlib.h>
#include "bitpack.h"
void theorapackB_readinit(oggpack_buffer *_b,unsigned char *_buf,int _bytes){
memset(_b,0,sizeof(*_b));
_b->buffer=_b->ptr=_buf;
_b->storage=_bytes;
}
int theorapackB_look1(oggpack_buffer *_b,long *_ret){
if(_b->endbyte>=_b->storage){
*_ret=0L;
return -1;
}
*_ret=(_b->ptr[0]>>7-_b->endbit)&1;
return 0;
}
void theorapackB_adv1(oggpack_buffer *_b){
if(++(_b->endbit)>7){
_b->endbit=0;
_b->ptr++;
_b->endbyte++;
}
}
/*Here we assume that 0<=_bits&&_bits<=32.*/
int theorapackB_read(oggpack_buffer *_b,int _bits,long *_ret){
long ret;
long m;
long d;
int fail;
m=32-_bits;
_bits+=_b->endbit;
d=_b->storage-_b->endbyte;
if(d<=4){
/*Not the main path.*/
if(d*8<_bits){
*_ret=0L;
fail=-1;
goto overflow;
}
/*Special case to avoid reading _b->ptr[0], which might be past the end of
the buffer; also skips some useless accounting.*/
else if(!_bits){
*_ret=0L;
return 0;
}
}
ret=_b->ptr[0]<<24+_b->endbit;
if(_bits>8){
ret|=_b->ptr[1]<<16+_b->endbit;
if(_bits>16){
ret|=_b->ptr[2]<<8+_b->endbit;
if(_bits>24){
ret|=_b->ptr[3]<<_b->endbit;
if(_bits>32)ret|=_b->ptr[4]>>8-_b->endbit;
}
}
}
*_ret=((ret&0xFFFFFFFFUL)>>(m>>1))>>(m+1>>1);
fail=0;
overflow:
_b->ptr+=_bits>>3;
_b->endbyte+=_bits>>3;
_b->endbit=_bits&7;
return fail;
}
int theorapackB_read1(oggpack_buffer *_b,long *_ret){
int fail;
if(_b->endbyte>=_b->storage){
/*Not the main path.*/
*_ret=0L;
fail=-1;
}
else{
*_ret=(_b->ptr[0]>>7-_b->endbit)&1;
fail=0;
}
_b->endbit++;
if(_b->endbit>7){
_b->endbit=0;
_b->ptr++;
_b->endbyte++;
}
return fail;
}
long theorapackB_bytes(oggpack_buffer *_b){
return _b->endbyte+(_b->endbit+7>>3);
}
long theorapackB_bits(oggpack_buffer *_b){
return _b->endbyte*8+_b->endbit;
}
unsigned char *theorapackB_get_buffer(oggpack_buffer *_b){
return _b->buffer;
}

38
Engine/lib/libtheora/lib/dec/bitpack.h Normal file
View file

@ -0,0 +1,38 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE OggTheora SOURCE CODE IS (C) COPYRIGHT 1994-2008 *
* by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
* *
********************************************************************
function: packing variable sized words into an octet stream
last mod: $Id: bitwise.c 7675 2004-09-01 00:34:39Z xiphmont $
********************************************************************/
#if !defined(_bitpack_H)
# define _bitpack_H (1)
# include <ogg/ogg.h>
void theorapackB_readinit(oggpack_buffer *_b,unsigned char *_buf,int _bytes);
int theorapackB_look1(oggpack_buffer *_b,long *_ret);
void theorapackB_adv1(oggpack_buffer *_b);
/*Here we assume 0<=_bits&&_bits<=32.*/
int theorapackB_read(oggpack_buffer *_b,int _bits,long *_ret);
int theorapackB_read1(oggpack_buffer *_b,long *_ret);
long theorapackB_bytes(oggpack_buffer *_b);
long theorapackB_bits(oggpack_buffer *_b);
unsigned char *theorapackB_get_buffer(oggpack_buffer *_b);
/*These two functions are implemented locally in huffdec.c*/
/*Read in bits without advancing the bitptr.
Here we assume 0<=_bits&&_bits<=32.*/
/*static int theorapackB_look(oggpack_buffer *_b,int _bits,long *_ret);*/
/*static void theorapackB_adv(oggpack_buffer *_b,int _bits);*/
#endif

31
Engine/lib/libtheora/lib/dec/dct.h Normal file
View file

@ -0,0 +1,31 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: dct.h 15400 2008-10-15 12:10:58Z tterribe $
********************************************************************/
/*Definitions shared by the forward and inverse DCT transforms.*/
#if !defined(_dct_H)
# define _dct_H (1)
/*cos(n*pi/16) (resp. sin(m*pi/16)) scaled by 65536.*/
#define OC_C1S7 ((ogg_int32_t)64277)
#define OC_C2S6 ((ogg_int32_t)60547)
#define OC_C3S5 ((ogg_int32_t)54491)
#define OC_C4S4 ((ogg_int32_t)46341)
#define OC_C5S3 ((ogg_int32_t)36410)
#define OC_C6S2 ((ogg_int32_t)25080)
#define OC_C7S1 ((ogg_int32_t)12785)
#endif

188
Engine/lib/libtheora/lib/dec/decapiwrapper.c Normal file
View file

@ -0,0 +1,188 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: decapiwrapper.c 13596 2007-08-23 20:05:38Z tterribe $
********************************************************************/
#include <stdlib.h>
#include <string.h>
#include <limits.h>
#include "apiwrapper.h"
#include "theora/theoradec.h"
static void th_dec_api_clear(th_api_wrapper *_api){
if(_api->setup)th_setup_free(_api->setup);
if(_api->decode)th_decode_free(_api->decode);
memset(_api,0,sizeof(*_api));
}
static void theora_decode_clear(theora_state *_td){
if(_td->i!=NULL)theora_info_clear(_td->i);
memset(_td,0,sizeof(*_td));
}
static int theora_decode_control(theora_state *_td,int _req,
void *_buf,size_t _buf_sz){
return th_decode_ctl(((th_api_wrapper *)_td->i->codec_setup)->decode,
_req,_buf,_buf_sz);
}
static ogg_int64_t theora_decode_granule_frame(theora_state *_td,
ogg_int64_t _gp){
return th_granule_frame(((th_api_wrapper *)_td->i->codec_setup)->decode,_gp);
}
static double theora_decode_granule_time(theora_state *_td,ogg_int64_t _gp){
return th_granule_time(((th_api_wrapper *)_td->i->codec_setup)->decode,_gp);
}
static const oc_state_dispatch_vtbl OC_DEC_DISPATCH_VTBL={
(oc_state_clear_func)theora_decode_clear,
(oc_state_control_func)theora_decode_control,
(oc_state_granule_frame_func)theora_decode_granule_frame,
(oc_state_granule_time_func)theora_decode_granule_time,
};
static void th_info2theora_info(theora_info *_ci,const th_info *_info){
_ci->version_major=_info->version_major;
_ci->version_minor=_info->version_minor;
_ci->version_subminor=_info->version_subminor;
_ci->width=_info->frame_width;
_ci->height=_info->frame_height;
_ci->frame_width=_info->pic_width;
_ci->frame_height=_info->pic_height;
_ci->offset_x=_info->pic_x;
_ci->offset_y=_info->pic_y;
_ci->fps_numerator=_info->fps_numerator;
_ci->fps_denominator=_info->fps_denominator;
_ci->aspect_numerator=_info->aspect_numerator;
_ci->aspect_denominator=_info->aspect_denominator;
switch(_info->colorspace){
case TH_CS_ITU_REC_470M:_ci->colorspace=OC_CS_ITU_REC_470M;break;
case TH_CS_ITU_REC_470BG:_ci->colorspace=OC_CS_ITU_REC_470BG;break;
default:_ci->colorspace=OC_CS_UNSPECIFIED;break;
}
switch(_info->pixel_fmt){
case TH_PF_420:_ci->pixelformat=OC_PF_420;break;
case TH_PF_422:_ci->pixelformat=OC_PF_422;break;
case TH_PF_444:_ci->pixelformat=OC_PF_444;break;
default:_ci->pixelformat=OC_PF_RSVD;
}
_ci->target_bitrate=_info->target_bitrate;
_ci->quality=_info->quality;
_ci->keyframe_frequency_force=1<<_info->keyframe_granule_shift;
}
int theora_decode_init(theora_state *_td,theora_info *_ci){
th_api_info *apiinfo;
th_api_wrapper *api;
th_info info;
api=(th_api_wrapper *)_ci->codec_setup;
/*Allocate our own combined API wrapper/theora_info struct.
We put them both in one malloc'd block so that when the API wrapper is
freed, the info struct goes with it.
This avoids having to figure out whether or not we need to free the info
struct in either theora_info_clear() or theora_clear().*/
apiinfo=(th_api_info *)_ogg_calloc(1,sizeof(*apiinfo));
/*Make our own copy of the info struct, since its lifetime should be
independent of the one we were passed in.*/
*&apiinfo->info=*_ci;
/*Convert the info struct now instead of saving the the one we decoded with
theora_decode_header(), since the user might have modified values (i.e.,
color space, aspect ratio, etc. can be specified from a higher level).
The user also might be doing something "clever" with the header packets if
they are not using an Ogg encapsulation.*/
oc_theora_info2th_info(&info,_ci);
/*Don't bother to copy the setup info; th_decode_alloc() makes its own copy
of the stuff it needs.*/
apiinfo->api.decode=th_decode_alloc(&info,api->setup);
if(apiinfo->api.decode==NULL){
_ogg_free(apiinfo);
return OC_EINVAL;
}
apiinfo->api.clear=(oc_setup_clear_func)th_dec_api_clear;
_td->internal_encode=NULL;
/*Provide entry points for ABI compatibility with old decoder shared libs.*/
_td->internal_decode=(void *)&OC_DEC_DISPATCH_VTBL;
_td->granulepos=0;
_td->i=&apiinfo->info;
_td->i->codec_setup=&apiinfo->api;
return 0;
}
int theora_decode_header(theora_info *_ci,theora_comment *_cc,ogg_packet *_op){
th_api_wrapper *api;
th_info info;
int ret;
api=(th_api_wrapper *)_ci->codec_setup;
/*Allocate an API wrapper struct on demand, since it will not also include a
theora_info struct like the ones that are used in a theora_state struct.*/
if(api==NULL){
_ci->codec_setup=_ogg_calloc(1,sizeof(*api));
api=(th_api_wrapper *)_ci->codec_setup;
api->clear=(oc_setup_clear_func)th_dec_api_clear;
}
/*Convert from the theora_info struct instead of saving our own th_info
struct between calls.
The user might be doing something "clever" with the header packets if they
are not using an Ogg encapsulation, and we don't want to break this.*/
oc_theora_info2th_info(&info,_ci);
/*We rely on the fact that theora_comment and th_comment structures are
actually identical.
Take care not to change this fact unless you change the code here as
well!*/
ret=th_decode_headerin(&info,(th_comment *)_cc,&api->setup,_op);
/*We also rely on the fact that the error return code values are the same,
and that the implementations of these two functions return the same set of
them.
Note that theora_decode_header() really can return OC_NOTFORMAT, even
though it is not currently documented to do so.*/
if(ret<0)return ret;
th_info2theora_info(_ci,&info);
return 0;
}
int theora_decode_packetin(theora_state *_td,ogg_packet *_op){
th_api_wrapper *api;
ogg_int64_t gp;
int ret;
if(!_td||!_td->i||!_td->i->codec_setup)return OC_FAULT;
api=(th_api_wrapper *)_td->i->codec_setup;
ret=th_decode_packetin(api->decode,_op,&gp);
if(ret<0)return OC_BADPACKET;
_td->granulepos=gp;
return 0;
}
int theora_decode_YUVout(theora_state *_td,yuv_buffer *_yuv){
th_api_wrapper *api;
th_ycbcr_buffer buf;
int ret;
if(!_td||!_td->i||!_td->i->codec_setup)return OC_FAULT;
api=(th_api_wrapper *)_td->i->codec_setup;
if(!api->decode)return OC_FAULT;
ret=th_decode_ycbcr_out(api->decode,buf);
if(ret>=0){
_yuv->y_width=buf[0].width;
_yuv->y_height=buf[0].height;
_yuv->y_stride=buf[0].stride;
_yuv->uv_width=buf[1].width;
_yuv->uv_height=buf[1].height;
_yuv->uv_stride=buf[1].stride;
_yuv->y=buf[0].data;
_yuv->u=buf[1].data;
_yuv->v=buf[2].data;
}
return ret;
}

239
Engine/lib/libtheora/lib/dec/decinfo.c Normal file
View file

@ -0,0 +1,239 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: decinfo.c 15400 2008-10-15 12:10:58Z tterribe $
********************************************************************/
#include <stdlib.h>
#include <string.h>
#include <limits.h>
#include "decint.h"
/*Unpacks a series of octets from a given byte array into the pack buffer.
No checking is done to ensure the buffer contains enough data.
_opb: The pack buffer to read the octets from.
_buf: The byte array to store the unpacked bytes in.
_len: The number of octets to unpack.*/
static void oc_unpack_octets(oggpack_buffer *_opb,char *_buf,size_t _len){
while(_len-->0){
long val;
theorapackB_read(_opb,8,&val);
*_buf++=(char)val;
}
}
/*Unpacks a 32-bit integer encoded by octets in little-endian form.*/
static long oc_unpack_length(oggpack_buffer *_opb){
long ret[4];
int i;
for(i=0;i<4;i++)theorapackB_read(_opb,8,ret+i);
return ret[0]|ret[1]<<8|ret[2]<<16|ret[3]<<24;
}
static int oc_info_unpack(oggpack_buffer *_opb,th_info *_info){
long val;
/*Check the codec bitstream version.*/
theorapackB_read(_opb,8,&val);
_info->version_major=(unsigned char)val;
theorapackB_read(_opb,8,&val);
_info->version_minor=(unsigned char)val;
theorapackB_read(_opb,8,&val);
_info->version_subminor=(unsigned char)val;
/*verify we can parse this bitstream version.
We accept earlier minors and all subminors, by spec*/
if(_info->version_major>TH_VERSION_MAJOR||
_info->version_major==TH_VERSION_MAJOR&&
_info->version_minor>TH_VERSION_MINOR){
return TH_EVERSION;
}
/*Read the encoded frame description.*/
theorapackB_read(_opb,16,&val);
_info->frame_width=(ogg_uint32_t)val<<4;
theorapackB_read(_opb,16,&val);
_info->frame_height=(ogg_uint32_t)val<<4;
theorapackB_read(_opb,24,&val);
_info->pic_width=(ogg_uint32_t)val;
theorapackB_read(_opb,24,&val);
_info->pic_height=(ogg_uint32_t)val;
theorapackB_read(_opb,8,&val);
_info->pic_x=(ogg_uint32_t)val;
/*Note: The sense of pic_y is inverted in what we pass back to the
application compared to how it is stored in the bitstream.
This is because the bitstream uses a right-handed coordinate system, while
applications expect a left-handed one.*/
theorapackB_read(_opb,8,&val);
_info->pic_y=_info->frame_height-_info->pic_height-(ogg_uint32_t)val;
theorapackB_read(_opb,32,&val);
_info->fps_numerator=(ogg_uint32_t)val;
theorapackB_read(_opb,32,&val);
_info->fps_denominator=(ogg_uint32_t)val;
if(_info->frame_width==0||_info->frame_height==0||
_info->pic_width+_info->pic_x>_info->frame_width||
_info->pic_height+_info->pic_y>_info->frame_height||
_info->fps_numerator==0||_info->fps_denominator==0){
return TH_EBADHEADER;
}
theorapackB_read(_opb,24,&val);
_info->aspect_numerator=(ogg_uint32_t)val;
theorapackB_read(_opb,24,&val);
_info->aspect_denominator=(ogg_uint32_t)val;
theorapackB_read(_opb,8,&val);
_info->colorspace=(th_colorspace)val;
theorapackB_read(_opb,24,&val);
_info->target_bitrate=(int)val;
theorapackB_read(_opb,6,&val);
_info->quality=(int)val;
theorapackB_read(_opb,5,&val);
_info->keyframe_granule_shift=(int)val;
theorapackB_read(_opb,2,&val);
_info->pixel_fmt=(th_pixel_fmt)val;
if(_info->pixel_fmt==TH_PF_RSVD)return TH_EBADHEADER;
if(theorapackB_read(_opb,3,&val)<0||val!=0)return TH_EBADHEADER;
return 0;
}
static int oc_comment_unpack(oggpack_buffer *_opb,th_comment *_tc){
long len;
int i;
/*Read the vendor string.*/
len=oc_unpack_length(_opb);
if(len<0||theorapackB_bytes(_opb)+len>_opb->storage)return TH_EBADHEADER;
_tc->vendor=_ogg_malloc((size_t)len+1);
oc_unpack_octets(_opb,_tc->vendor,len);
_tc->vendor[len]='\0';
/*Read the user comments.*/
_tc->comments=(int)oc_unpack_length(_opb);
if(_tc->comments<0||_tc->comments>(LONG_MAX>>2)||
theorapackB_bytes(_opb)+((long)_tc->comments<<2)>_opb->storage){
return TH_EBADHEADER;
}
_tc->comment_lengths=(int *)_ogg_malloc(
_tc->comments*sizeof(_tc->comment_lengths[0]));
_tc->user_comments=(char **)_ogg_malloc(
_tc->comments*sizeof(_tc->user_comments[0]));
for(i=0;i<_tc->comments;i++){
len=oc_unpack_length(_opb);
if(len<0||theorapackB_bytes(_opb)+len>_opb->storage){
_tc->comments=i;
return TH_EBADHEADER;
}
_tc->comment_lengths[i]=len;
_tc->user_comments[i]=_ogg_malloc((size_t)len+1);
oc_unpack_octets(_opb,_tc->user_comments[i],len);
_tc->user_comments[i][len]='\0';
}
return theorapackB_read(_opb,0,&len)<0?TH_EBADHEADER:0;
}
static int oc_setup_unpack(oggpack_buffer *_opb,th_setup_info *_setup){
int ret;
/*Read the quantizer tables.*/
ret=oc_quant_params_unpack(_opb,&_setup->qinfo);
if(ret<0)return ret;
/*Read the Huffman trees.*/
return oc_huff_trees_unpack(_opb,_setup->huff_tables);
}
static void oc_setup_clear(th_setup_info *_setup){
oc_quant_params_clear(&_setup->qinfo);
oc_huff_trees_clear(_setup->huff_tables);
}
static int oc_dec_headerin(oggpack_buffer *_opb,th_info *_info,
th_comment *_tc,th_setup_info **_setup,ogg_packet *_op){
char buffer[6];
long val;
int packtype;
int ret;
theorapackB_read(_opb,8,&val);
packtype=(int)val;
/*If we're at a data packet and we have received all three headers, we're
done.*/
if(!(packtype&0x80)&&_info->frame_width>0&&_tc->vendor!=NULL&&*_setup!=NULL){
return 0;
}
/*Check the codec string.*/
oc_unpack_octets(_opb,buffer,6);
if(memcmp(buffer,"theora",6)!=0)return TH_ENOTFORMAT;
switch(packtype){
/*Codec info header.*/
case 0x80:{
/*This should be the first packet, and we should not already be
initialized.*/
if(!_op->b_o_s||_info->frame_width>0)return TH_EBADHEADER;
ret=oc_info_unpack(_opb,_info);
if(ret<0)th_info_clear(_info);
else ret=3;
}break;
/*Comment header.*/
case 0x81:{
if(_tc==NULL)return TH_EFAULT;
/*We shoud have already decoded the info header, and should not yet have
decoded the comment header.*/
if(_info->frame_width==0||_tc->vendor!=NULL)return TH_EBADHEADER;
ret=oc_comment_unpack(_opb,_tc);
if(ret<0)th_comment_clear(_tc);
else ret=2;
}break;
/*Codec setup header.*/
case 0x82:{
oc_setup_info *setup;
if(_tc==NULL||_setup==NULL)return TH_EFAULT;
/*We should have already decoded the info header and the comment header,
and should not yet have decoded the setup header.*/
if(_info->frame_width==0||_tc->vendor==NULL||*_setup!=NULL){
return TH_EBADHEADER;
}
setup=(oc_setup_info *)_ogg_calloc(1,sizeof(*setup));
ret=oc_setup_unpack(_opb,setup);
if(ret<0){
oc_setup_clear(setup);
_ogg_free(setup);
}
else{
*_setup=setup;
ret=1;
}
}break;
default:{
/*We don't know what this header is.*/
return TH_EBADHEADER;
}break;
}
return ret;
}
/*Decodes one header packet.
This should be called repeatedly with the packets at the beginning of the
stream until it returns 0.*/
int th_decode_headerin(th_info *_info,th_comment *_tc,
th_setup_info **_setup,ogg_packet *_op){
oggpack_buffer opb;
int ret;
if(_op==NULL)return TH_EBADHEADER;
if(_info==NULL)return TH_EFAULT;
theorapackB_readinit(&opb,_op->packet,_op->bytes);
ret=oc_dec_headerin(&opb,_info,_tc,_setup,_op);
return ret;
}
void th_setup_free(th_setup_info *_setup){
if(_setup!=NULL){
oc_setup_clear(_setup);
_ogg_free(_setup);
}
}

95
Engine/lib/libtheora/lib/dec/decint.h Normal file
View file

@ -0,0 +1,95 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: decint.h 15400 2008-10-15 12:10:58Z tterribe $
********************************************************************/
#include <limits.h>
#if !defined(_decint_H)
# define _decint_H (1)
# include "theora/theoradec.h"
# include "../internal.h"
# include "bitpack.h"
typedef struct th_setup_info oc_setup_info;
typedef struct th_dec_ctx oc_dec_ctx;
# include "idct.h"
# include "huffdec.h"
# include "dequant.h"
/*Constants for the packet-in state machine specific to the decoder.*/
/*Next packet to read: Data packet.*/
#define OC_PACKET_DATA (0)
struct th_setup_info{
/*The Huffman codes.*/
oc_huff_node *huff_tables[TH_NHUFFMAN_TABLES];
/*The quantization parameters.*/
th_quant_info qinfo;
};
struct th_dec_ctx{
/*Shared encoder/decoder state.*/
oc_theora_state state;
/*Whether or not packets are ready to be emitted.
This takes on negative values while there are remaining header packets to
be emitted, reaches 0 when the codec is ready for input, and goes to 1
when a frame has been processed and a data packet is ready.*/
int packet_state;
/*Buffer in which to assemble packets.*/
oggpack_buffer opb;
/*Huffman decode trees.*/
oc_huff_node *huff_tables[TH_NHUFFMAN_TABLES];
/*The index of one past the last token in each plane for each coefficient.
The final entries are the total number of tokens for each coefficient.*/
int ti0[3][64];
/*The index of one past the last extra bits entry in each plane for each
coefficient.
The final entries are the total number of extra bits entries for each
coefficient.*/
int ebi0[3][64];
/*The number of outstanding EOB runs at the start of each coefficient in each
plane.*/
int eob_runs[3][64];
/*The DCT token lists.*/
unsigned char **dct_tokens;
/*The extra bits associated with DCT tokens.*/
ogg_uint16_t **extra_bits;
/*The out-of-loop post-processing level.*/
int pp_level;
/*The DC scale used for out-of-loop deblocking.*/
int pp_dc_scale[64];
/*The sharpen modifier used for out-of-loop deringing.*/
int pp_sharp_mod[64];
/*The DC quantization index of each block.*/
unsigned char *dc_qis;
/*The variance of each block.*/
int *variances;
/*The storage for the post-processed frame buffer.*/
unsigned char *pp_frame_data;
/*Whether or not the post-processsed frame buffer has space for chroma.*/
int pp_frame_has_chroma;
/*The buffer used for the post-processed frame.*/
th_ycbcr_buffer pp_frame_buf;
/*The striped decode callback function.*/
th_stripe_callback stripe_cb;
};
#endif

2057
Engine/lib/libtheora/lib/dec/decode.c Normal file
View file

File diff suppressed because it is too large Load diff

170
Engine/lib/libtheora/lib/dec/dequant.c Normal file
View file

@ -0,0 +1,170 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: dequant.c 15400 2008-10-15 12:10:58Z tterribe $
********************************************************************/
#include <stdlib.h>
#include <string.h>
#include <ogg/ogg.h>
#include "dequant.h"
#include "decint.h"
int oc_quant_params_unpack(oggpack_buffer *_opb,
th_quant_info *_qinfo){
th_quant_base *base_mats;
long val;
int nbase_mats;
int sizes[64];
int indices[64];
int nbits;
int bmi;
int ci;
int qti;
int pli;
int qri;
int qi;
int i;
theorapackB_read(_opb,3,&val);
nbits=(int)val;
for(qi=0;qi<64;qi++){
theorapackB_read(_opb,nbits,&val);
_qinfo->loop_filter_limits[qi]=(unsigned char)val;
}
theorapackB_read(_opb,4,&val);
nbits=(int)val+1;
for(qi=0;qi<64;qi++){
theorapackB_read(_opb,nbits,&val);
_qinfo->ac_scale[qi]=(ogg_uint16_t)val;
}
theorapackB_read(_opb,4,&val);
nbits=(int)val+1;
for(qi=0;qi<64;qi++){
theorapackB_read(_opb,nbits,&val);
_qinfo->dc_scale[qi]=(ogg_uint16_t)val;
}
theorapackB_read(_opb,9,&val);
nbase_mats=(int)val+1;
base_mats=_ogg_malloc(nbase_mats*sizeof(base_mats[0]));
for(bmi=0;bmi<nbase_mats;bmi++){
for(ci=0;ci<64;ci++){
theorapackB_read(_opb,8,&val);
base_mats[bmi][ci]=(unsigned char)val;
}
}
nbits=oc_ilog(nbase_mats-1);
for(i=0;i<6;i++){
th_quant_ranges *qranges;
th_quant_base *qrbms;
int *qrsizes;
qti=i/3;
pli=i%3;
qranges=_qinfo->qi_ranges[qti]+pli;
if(i>0){
theorapackB_read1(_opb,&val);
if(!val){
int qtj;
int plj;
if(qti>0){
theorapackB_read1(_opb,&val);
if(val){
qtj=qti-1;
plj=pli;
}
else{
qtj=(i-1)/3;
plj=(i-1)%3;
}
}
else{
qtj=(i-1)/3;
plj=(i-1)%3;
}
*qranges=*(_qinfo->qi_ranges[qtj]+plj);
continue;
}
}
theorapackB_read(_opb,nbits,&val);
indices[0]=(int)val;
for(qi=qri=0;qi<63;){
theorapackB_read(_opb,oc_ilog(62-qi),&val);
sizes[qri]=(int)val+1;
qi+=(int)val+1;
theorapackB_read(_opb,nbits,&val);
indices[++qri]=(int)val;
}
/*Note: The caller is responsible for cleaning up any partially
constructed qinfo.*/
if(qi>63){
_ogg_free(base_mats);
return TH_EBADHEADER;
}
qranges->nranges=qri;
qranges->sizes=qrsizes=(int *)_ogg_malloc(qri*sizeof(qrsizes[0]));
memcpy(qrsizes,sizes,qri*sizeof(qrsizes[0]));
qrbms=(th_quant_base *)_ogg_malloc((qri+1)*sizeof(qrbms[0]));
qranges->base_matrices=(const th_quant_base *)qrbms;
do{
bmi=indices[qri];
/*Note: The caller is responsible for cleaning up any partially
constructed qinfo.*/
if(bmi>=nbase_mats){
_ogg_free(base_mats);
return TH_EBADHEADER;
}
memcpy(qrbms[qri],base_mats[bmi],sizeof(qrbms[qri]));
}
while(qri-->0);
}
_ogg_free(base_mats);
return 0;
}
void oc_quant_params_clear(th_quant_info *_qinfo){
int i;
for(i=6;i-->0;){
int qti;
int pli;
qti=i/3;
pli=i%3;
/*Clear any duplicate pointer references.*/
if(i>0){
int qtj;
int plj;
qtj=(i-1)/3;
plj=(i-1)%3;
if(_qinfo->qi_ranges[qti][pli].sizes==
_qinfo->qi_ranges[qtj][plj].sizes){
_qinfo->qi_ranges[qti][pli].sizes=NULL;
}
if(_qinfo->qi_ranges[qti][pli].base_matrices==
_qinfo->qi_ranges[qtj][plj].base_matrices){
_qinfo->qi_ranges[qti][pli].base_matrices=NULL;
}
}
if(qti>0){
if(_qinfo->qi_ranges[1][pli].sizes==
_qinfo->qi_ranges[0][pli].sizes){
_qinfo->qi_ranges[1][pli].sizes=NULL;
}
if(_qinfo->qi_ranges[1][pli].base_matrices==
_qinfo->qi_ranges[0][pli].base_matrices){
_qinfo->qi_ranges[1][pli].base_matrices=NULL;
}
}
/*Now free all the non-duplicate storage.*/
_ogg_free((void *)_qinfo->qi_ranges[qti][pli].sizes);
_ogg_free((void *)_qinfo->qi_ranges[qti][pli].base_matrices);
}
}

26
Engine/lib/libtheora/lib/dec/dequant.h Normal file
View file

@ -0,0 +1,26 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: dequant.h 15400 2008-10-15 12:10:58Z tterribe $
********************************************************************/
#if !defined(_dequant_H)
# define _dequant_H (1)
# include "quant.h"
int oc_quant_params_unpack(oggpack_buffer *_opb,
th_quant_info *_qinfo);
void oc_quant_params_clear(th_quant_info *_qinfo);
#endif

199
Engine/lib/libtheora/lib/dec/fragment.c Normal file
View file

@ -0,0 +1,199 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: fragment.c 15469 2008-10-30 12:49:42Z tterribe $
********************************************************************/
#include "../internal.h"
void oc_frag_recon_intra(const oc_theora_state *_state,unsigned char *_dst,
int _dst_ystride,const ogg_int16_t *_residue){
_state->opt_vtable.frag_recon_intra(_dst,_dst_ystride,_residue);
}
void oc_frag_recon_intra_c(unsigned char *_dst,int _dst_ystride,
const ogg_int16_t *_residue){
int i;
for(i=0;i<8;i++){
int j;
for(j=0;j<8;j++){
int res;
res=*_residue++;
_dst[j]=OC_CLAMP255(res+128);
}
_dst+=_dst_ystride;
}
}
void oc_frag_recon_inter(const oc_theora_state *_state,unsigned char *_dst,
int _dst_ystride,const unsigned char *_src,int _src_ystride,
const ogg_int16_t *_residue){
_state->opt_vtable.frag_recon_inter(_dst,_dst_ystride,_src,_src_ystride,
_residue);
}
void oc_frag_recon_inter_c(unsigned char *_dst,int _dst_ystride,
const unsigned char *_src,int _src_ystride,const ogg_int16_t *_residue){
int i;
for(i=0;i<8;i++){
int j;
for(j=0;j<8;j++){
int res;
res=*_residue++;
_dst[j]=OC_CLAMP255(res+_src[j]);
}
_dst+=_dst_ystride;
_src+=_src_ystride;
}
}
void oc_frag_recon_inter2(const oc_theora_state *_state,unsigned char *_dst,
int _dst_ystride,const unsigned char *_src1,int _src1_ystride,
const unsigned char *_src2,int _src2_ystride,const ogg_int16_t *_residue){
_state->opt_vtable.frag_recon_inter2(_dst,_dst_ystride,_src1,_src1_ystride,
_src2,_src2_ystride,_residue);
}
void oc_frag_recon_inter2_c(unsigned char *_dst,int _dst_ystride,
const unsigned char *_src1,int _src1_ystride,const unsigned char *_src2,
int _src2_ystride,const ogg_int16_t *_residue){
int i;
for(i=0;i<8;i++){
int j;
for(j=0;j<8;j++){
int res;
res=*_residue++;
_dst[j]=OC_CLAMP255(res+((int)_src1[j]+_src2[j]>>1));
}
_dst+=_dst_ystride;
_src1+=_src1_ystride;
_src2+=_src2_ystride;
}
}
/*Computes the predicted DC value for the given fragment.
This requires that the fully decoded DC values be available for the left,
upper-left, upper, and upper-right fragments (if they exist).
_frag: The fragment to predict the DC value for.
_fplane: The fragment plane the fragment belongs to.
_x: The x-coordinate of the fragment.
_y: The y-coordinate of the fragment.
_pred_last: The last fully-decoded DC value for each predictor frame
(OC_FRAME_GOLD, OC_FRAME_PREV and OC_FRAME_SELF).
This should be initialized to 0's for the first fragment in each
color plane.
Return: The predicted DC value for this fragment.*/
int oc_frag_pred_dc(const oc_fragment *_frag,
const oc_fragment_plane *_fplane,int _x,int _y,int _pred_last[3]){
static const int PRED_SCALE[16][4]={
/*0*/
{0,0,0,0},
/*OC_PL*/
{1,0,0,0},
/*OC_PUL*/
{1,0,0,0},
/*OC_PL|OC_PUL*/
{1,0,0,0},
/*OC_PU*/
{1,0,0,0},
/*OC_PL|OC_PU*/
{1,1,0,0},
/*OC_PUL|OC_PU*/
{0,1,0,0},
/*OC_PL|OC_PUL|PC_PU*/
{29,-26,29,0},
/*OC_PUR*/
{1,0,0,0},
/*OC_PL|OC_PUR*/
{75,53,0,0},
/*OC_PUL|OC_PUR*/
{1,1,0,0},
/*OC_PL|OC_PUL|OC_PUR*/
{75,0,53,0},
/*OC_PU|OC_PUR*/
{1,0,0,0},
/*OC_PL|OC_PU|OC_PUR*/
{75,0,53,0},
/*OC_PUL|OC_PU|OC_PUR*/
{3,10,3,0},
/*OC_PL|OC_PUL|OC_PU|OC_PUR*/
{29,-26,29,0}
};
static const int PRED_SHIFT[16]={0,0,0,0,0,1,0,5,0,7,1,7,0,7,4,5};
static const int PRED_RMASK[16]={0,0,0,0,0,1,0,31,0,127,1,127,0,127,15,31};
static const int BC_MASK[8]={
/*No boundary condition.*/
OC_PL|OC_PUL|OC_PU|OC_PUR,
/*Left column.*/
OC_PU|OC_PUR,
/*Top row.*/
OC_PL,
/*Top row, left column.*/
0,
/*Right column.*/
OC_PL|OC_PUL|OC_PU,
/*Right and left column.*/
OC_PU,
/*Top row, right column.*/
OC_PL,
/*Top row, right and left column.*/
0
};
/*Predictor fragments, left, up-left, up, up-right.*/
const oc_fragment *predfr[4];
/*The frame used for prediction for this fragment.*/
int pred_frame;
/*The boundary condition flags.*/
int bc;
/*DC predictor values: left, up-left, up, up-right, missing values
skipped.*/
int p[4];
/*Predictor count.*/
int np;
/*Which predictor constants to use.*/
int pflags;
/*The predicted DC value.*/
int ret;
int i;
pred_frame=OC_FRAME_FOR_MODE[_frag->mbmode];
bc=(_x==0)+((_y==0)<<1)+((_x+1==_fplane->nhfrags)<<2);
predfr[0]=_frag-1;
predfr[1]=_frag-_fplane->nhfrags-1;
predfr[2]=predfr[1]+1;
predfr[3]=predfr[2]+1;
np=0;
pflags=0;
for(i=0;i<4;i++){
int pflag;
pflag=1<<i;
if((BC_MASK[bc]&pflag)&&predfr[i]->coded&&
OC_FRAME_FOR_MODE[predfr[i]->mbmode]==pred_frame){
p[np++]=predfr[i]->dc;
pflags|=pflag;
}
}
if(pflags==0)return _pred_last[pred_frame];
else{
ret=PRED_SCALE[pflags][0]*p[0];
/*LOOP VECTORIZES.*/
for(i=1;i<np;i++)ret+=PRED_SCALE[pflags][i]*p[i];
ret=OC_DIV_POW2(ret,PRED_SHIFT[pflags],PRED_RMASK[pflags]);
}
if((pflags&(OC_PL|OC_PUL|OC_PU))==(OC_PL|OC_PUL|OC_PU)){
if(abs(ret-p[2])>128)ret=p[2];
else if(abs(ret-p[0])>128)ret=p[0];
else if(abs(ret-p[1])>128)ret=p[1];
}
return ret;
}

325
Engine/lib/libtheora/lib/dec/huffdec.c Normal file
View file

@ -0,0 +1,325 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: huffdec.c 15431 2008-10-21 05:04:02Z giles $
********************************************************************/
#include <stdlib.h>
#include <ogg/ogg.h>
#include "huffdec.h"
#include "decint.h"
/*The ANSI offsetof macro is broken on some platforms (e.g., older DECs).*/
#define _ogg_offsetof(_type,_field)\
((size_t)((char *)&((_type *)0)->_field-(char *)0))
/*These two functions are really part of the bitpack.c module, but
they are only used here. Declaring local static versions so they
can be inlined saves considerable function call overhead.*/
/*Read in bits without advancing the bitptr.
Here we assume 0<=_bits&&_bits<=32.*/
static int theorapackB_look(oggpack_buffer *_b,int _bits,long *_ret){
long ret;
long m;
long d;
m=32-_bits;
_bits+=_b->endbit;
d=_b->storage-_b->endbyte;
if(d<=4){
/*Not the main path.*/
if(d<=0){
*_ret=0L;
return -(_bits>d*8);
}
/*If we have some bits left, but not enough, return the ones we have.*/
if(d*8<_bits)_bits=d*8;
}
ret=_b->ptr[0]<<24+_b->endbit;
if(_bits>8){
ret|=_b->ptr[1]<<16+_b->endbit;
if(_bits>16){
ret|=_b->ptr[2]<<8+_b->endbit;
if(_bits>24){
ret|=_b->ptr[3]<<_b->endbit;
if(_bits>32)ret|=_b->ptr[4]>>8-_b->endbit;
}
}
}
*_ret=((ret&0xFFFFFFFF)>>(m>>1))>>(m+1>>1);
return 0;
}
/*advance the bitptr*/
static void theorapackB_adv(oggpack_buffer *_b,int _bits){
_bits+=_b->endbit;
_b->ptr+=_bits>>3;
_b->endbyte+=_bits>>3;
_b->endbit=_bits&7;
}
/*The log_2 of the size of a lookup table is allowed to grow to relative to
the number of unique nodes it contains.
E.g., if OC_HUFF_SLUSH is 2, then at most 75% of the space in the tree is
wasted (each node will have an amortized cost of at most 20 bytes when using
4-byte pointers).
Larger numbers can decode tokens with fewer read operations, while smaller
numbers may save more space (requiring as little as 8 bytes amortized per
node, though there will be more nodes).
With a sample file:
32233473 read calls are required when no tree collapsing is done (100.0%).
19269269 read calls are required when OC_HUFF_SLUSH is 0 (59.8%).
11144969 read calls are required when OC_HUFF_SLUSH is 1 (34.6%).
10538563 read calls are required when OC_HUFF_SLUSH is 2 (32.7%).
10192578 read calls are required when OC_HUFF_SLUSH is 3 (31.6%).
Since a value of 1 gets us the vast majority of the speed-up with only a
small amount of wasted memory, this is what we use.*/
#define OC_HUFF_SLUSH (1)
/*Allocates a Huffman tree node that represents a subtree of depth _nbits.
_nbits: The depth of the subtree.
If this is 0, the node is a leaf node.
Otherwise 1<<_nbits pointers are allocated for children.
Return: The newly allocated and fully initialized Huffman tree node.*/
static oc_huff_node *oc_huff_node_alloc(int _nbits){
oc_huff_node *ret;
size_t size;
size=_ogg_offsetof(oc_huff_node,nodes);
if(_nbits>0)size+=sizeof(oc_huff_node *)*(1<<_nbits);
ret=_ogg_calloc(1,size);
ret->nbits=(unsigned char)_nbits;
return ret;
}
/*Frees a Huffman tree node allocated with oc_huf_node_alloc.
_node: The node to free.
This may be NULL.*/
static void oc_huff_node_free(oc_huff_node *_node){
_ogg_free(_node);
}
/*Frees the memory used by a Huffman tree.
_node: The Huffman tree to free.
This may be NULL.*/
static void oc_huff_tree_free(oc_huff_node *_node){
if(_node==NULL)return;
if(_node->nbits){
int nchildren;
int i;
int inext;
nchildren=1<<_node->nbits;
for(i=0;i<nchildren;i=inext){
inext=i+(_node->nodes[i]!=NULL?1<<_node->nbits-_node->nodes[i]->depth:1);
oc_huff_tree_free(_node->nodes[i]);
}
}
oc_huff_node_free(_node);
}
/*Unpacks a sub-tree from the given buffer.
_opb: The buffer to unpack from.
_binode: The location to store a pointer to the sub-tree in.
_depth: The current depth of the tree.
This is used to prevent infinite recursion.
Return: 0 on success, or a negative value on error.*/
static int oc_huff_tree_unpack(oggpack_buffer *_opb,
oc_huff_node **_binode,int _depth){
oc_huff_node *binode;
long bits;
/*Prevent infinite recursion.*/
if(++_depth>32)return TH_EBADHEADER;
if(theorapackB_read1(_opb,&bits)<0)return TH_EBADHEADER;
/*Read an internal node:*/
if(!bits){
int ret;
binode=oc_huff_node_alloc(1);
binode->depth=(unsigned char)(_depth>1);
ret=oc_huff_tree_unpack(_opb,binode->nodes,_depth);
if(ret>=0)ret=oc_huff_tree_unpack(_opb,binode->nodes+1,_depth);
if(ret<0){
oc_huff_tree_free(binode);
*_binode=NULL;
return ret;
}
}
/*Read a leaf node:*/
else{
if(theorapackB_read(_opb,OC_NDCT_TOKEN_BITS,&bits)<0)return TH_EBADHEADER;
binode=oc_huff_node_alloc(0);
binode->depth=(unsigned char)(_depth>1);
binode->token=(unsigned char)bits;
}
*_binode=binode;
return 0;
}
/*Finds the depth of shortest branch of the given sub-tree.
The tree must be binary.
_binode: The root of the given sub-tree.
_binode->nbits must be 0 or 1.
Return: The smallest depth of a leaf node in this sub-tree.
0 indicates this sub-tree is a leaf node.*/
static int oc_huff_tree_mindepth(oc_huff_node *_binode){
int depth0;
int depth1;
if(_binode->nbits==0)return 0;
depth0=oc_huff_tree_mindepth(_binode->nodes[0]);
depth1=oc_huff_tree_mindepth(_binode->nodes[1]);
return OC_MINI(depth0,depth1)+1;
}
/*Finds the number of internal nodes at a given depth, plus the number of
leaves at that depth or shallower.
The tree must be binary.
_binode: The root of the given sub-tree.
_binode->nbits must be 0 or 1.
Return: The number of entries that would be contained in a jump table of the
given depth.*/
static int oc_huff_tree_occupancy(oc_huff_node *_binode,int _depth){
if(_binode->nbits==0||_depth<=0)return 1;
else{
return oc_huff_tree_occupancy(_binode->nodes[0],_depth-1)+
oc_huff_tree_occupancy(_binode->nodes[1],_depth-1);
}
}
static oc_huff_node *oc_huff_tree_collapse(oc_huff_node *_binode);
/*Fills the given nodes table with all the children in the sub-tree at the
given depth.
The nodes in the sub-tree with a depth less than that stored in the table
are freed.
The sub-tree must be binary and complete up until the given depth.
_nodes: The nodes table to fill.
_binode: The root of the sub-tree to fill it with.
_binode->nbits must be 0 or 1.
_level: The current level in the table.
0 indicates that the current node should be stored, regardless of
whether it is a leaf node or an internal node.
_depth: The depth of the nodes to fill the table with, relative to their
parent.*/
static void oc_huff_node_fill(oc_huff_node **_nodes,
oc_huff_node *_binode,int _level,int _depth){
if(_level<=0||_binode->nbits==0){
int i;
_binode->depth=(unsigned char)(_depth-_level);
_nodes[0]=oc_huff_tree_collapse(_binode);
for(i=1;i<1<<_level;i++)_nodes[i]=_nodes[0];
}
else{
_level--;
oc_huff_node_fill(_nodes,_binode->nodes[0],_level,_depth);
oc_huff_node_fill(_nodes+(1<<_level),_binode->nodes[1],_level,_depth);
oc_huff_node_free(_binode);
}
}
/*Finds the largest complete sub-tree rooted at the current node and collapses
it into a single node.
This procedure is then applied recursively to all the children of that node.
_binode: The root of the sub-tree to collapse.
_binode->nbits must be 0 or 1.
Return: The new root of the collapsed sub-tree.*/
static oc_huff_node *oc_huff_tree_collapse(oc_huff_node *_binode){
oc_huff_node *root;
int mindepth;
int depth;
int loccupancy;
int occupancy;
depth=mindepth=oc_huff_tree_mindepth(_binode);
occupancy=1<<mindepth;
do{
loccupancy=occupancy;
occupancy=oc_huff_tree_occupancy(_binode,++depth);
}
while(occupancy>loccupancy&&occupancy>=1<<OC_MAXI(depth-OC_HUFF_SLUSH,0));
depth--;
if(depth<=1)return _binode;
root=oc_huff_node_alloc(depth);
root->depth=_binode->depth;
oc_huff_node_fill(root->nodes,_binode,depth,depth);
return root;
}
/*Makes a copy of the given Huffman tree.
_node: The Huffman tree to copy.
Return: The copy of the Huffman tree.*/
static oc_huff_node *oc_huff_tree_copy(const oc_huff_node *_node){
oc_huff_node *ret;
ret=oc_huff_node_alloc(_node->nbits);
ret->depth=_node->depth;
if(_node->nbits){
int nchildren;
int i;
int inext;
nchildren=1<<_node->nbits;
for(i=0;i<nchildren;){
ret->nodes[i]=oc_huff_tree_copy(_node->nodes[i]);
inext=i+(1<<_node->nbits-ret->nodes[i]->depth);
while(++i<inext)ret->nodes[i]=ret->nodes[i-1];
}
}
else ret->token=_node->token;
return ret;
}
/*Unpacks a set of Huffman trees, and reduces them to a collapsed
representation.
_opb: The buffer to unpack the trees from.
_nodes: The table to fill with the Huffman trees.
Return: 0 on success, or a negative value on error.*/
int oc_huff_trees_unpack(oggpack_buffer *_opb,
oc_huff_node *_nodes[TH_NHUFFMAN_TABLES]){
int i;
for(i=0;i<TH_NHUFFMAN_TABLES;i++){
int ret;
ret=oc_huff_tree_unpack(_opb,_nodes+i,0);
if(ret<0)return ret;
_nodes[i]=oc_huff_tree_collapse(_nodes[i]);
}
return 0;
}
/*Makes a copy of the given set of Huffman trees.
_dst: The array to store the copy in.
_src: The array of trees to copy.*/
void oc_huff_trees_copy(oc_huff_node *_dst[TH_NHUFFMAN_TABLES],
const oc_huff_node *const _src[TH_NHUFFMAN_TABLES]){
int i;
for(i=0;i<TH_NHUFFMAN_TABLES;i++)_dst[i]=oc_huff_tree_copy(_src[i]);
}
/*Frees the memory used by a set of Huffman trees.
_nodes: The array of trees to free.*/
void oc_huff_trees_clear(oc_huff_node *_nodes[TH_NHUFFMAN_TABLES]){
int i;
for(i=0;i<TH_NHUFFMAN_TABLES;i++)oc_huff_tree_free(_nodes[i]);
}
/*Unpacks a single token using the given Huffman tree.
_opb: The buffer to unpack the token from.
_node: The tree to unpack the token with.
Return: The token value.*/
int oc_huff_token_decode(oggpack_buffer *_opb,const oc_huff_node *_node){
long bits;
while(_node->nbits!=0){
theorapackB_look(_opb,_node->nbits,&bits);
_node=_node->nodes[bits];
theorapackB_adv(_opb,_node->depth);
}
return _node->token;
}

91
Engine/lib/libtheora/lib/dec/huffdec.h Normal file
View file

@ -0,0 +1,91 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: huffdec.h 15400 2008-10-15 12:10:58Z tterribe $
********************************************************************/
#if !defined(_huffdec_H)
# define _huffdec_H (1)
# include "huffman.h"
typedef struct oc_huff_node oc_huff_node;
/*A node in the Huffman tree.
Instead of storing every branching in the tree, subtrees can be collapsed
into one node, with a table of size 1<<nbits pointing directly to its
descedents nbits levels down.
This allows more than one bit to be read at a time, and avoids following all
the intermediate branches with next to no increased code complexity once
the collapsed tree has been built.
We do _not_ require that a subtree be complete to be collapsed, but instead
store duplicate pointers in the table, and record the actual depth of the
node below its parent.
This tells us the number of bits to advance the stream after reaching it.
This turns out to be equivalent to the method described in \cite{Hash95},
without the requirement that codewords be sorted by length.
If the codewords were sorted by length (so-called ``canonical-codes''), they
could be decoded much faster via either Lindell and Moffat's approach or
Hashemian's Condensed Huffman Code approach, the latter of which has an
extremely small memory footprint.
We can't use Choueka et al.'s finite state machine approach, which is
extremely fast, because we can't allow multiple symbols to be output at a
time; the codebook can and does change between symbols.
It also has very large memory requirements, which impairs cache coherency.
@ARTICLE{Hash95,
author="Reza Hashemian",
title="Memory Efficient and High-Speed Search {Huffman} Coding",
journal="{IEEE} Transactions on Communications",
volume=43,
number=10,
pages="2576--2581",
month=Oct,
year=1995
}*/
struct oc_huff_node{
/*The number of bits of the code needed to descend through this node.
0 indicates a leaf node.
Otherwise there are 1<<nbits nodes in the nodes table, which can be
indexed by reading nbits bits from the stream.*/
unsigned char nbits;
/*The value of a token stored in a leaf node.
The value in non-leaf nodes is undefined.*/
unsigned char token;
/*The depth of the current node, relative to its parent in the collapsed
tree.
This can be less than its parent's nbits value, in which case there are
1<<nbits-depth copies of this node in the table, and the bitstream should
only be advanced depth bits after reaching this node.*/
unsigned char depth;
/*The table of child nodes.
The ACTUAL size of this array is 1<<nbits, despite what the declaration
below claims.
The exception is that for leaf nodes the size is 0.*/
oc_huff_node *nodes[1];
};
int oc_huff_trees_unpack(oggpack_buffer *_opb,
oc_huff_node *_nodes[TH_NHUFFMAN_TABLES]);
void oc_huff_trees_copy(oc_huff_node *_dst[TH_NHUFFMAN_TABLES],
const oc_huff_node *const _src[TH_NHUFFMAN_TABLES]);
void oc_huff_trees_clear(oc_huff_node *_nodes[TH_NHUFFMAN_TABLES]);
int oc_huff_token_decode(oggpack_buffer *_opb,const oc_huff_node *_node);
#endif

70
Engine/lib/libtheora/lib/dec/huffman.h Normal file
View file

@ -0,0 +1,70 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: huffman.h 15400 2008-10-15 12:10:58Z tterribe $
********************************************************************/
#if !defined(_huffman_H)
# define _hufffman_H (1)
# include "theora/codec.h"
# include "ocintrin.h"
/*The range of valid quantized DCT coefficient values.
VP3 used 511 in the encoder, but the bitstream is capable of 580.*/
#define OC_DCT_VAL_RANGE (580)
#define OC_NDCT_TOKEN_BITS (5)
#define OC_DCT_EOB1_TOKEN (0)
#define OC_DCT_EOB2_TOKEN (1)
#define OC_DCT_EOB3_TOKEN (2)
#define OC_DCT_REPEAT_RUN0_TOKEN (3)
#define OC_DCT_REPEAT_RUN1_TOKEN (4)
#define OC_DCT_REPEAT_RUN2_TOKEN (5)
#define OC_DCT_REPEAT_RUN3_TOKEN (6)
#define OC_DCT_SHORT_ZRL_TOKEN (7)
#define OC_DCT_ZRL_TOKEN (8)
#define OC_ONE_TOKEN (9)
#define OC_MINUS_ONE_TOKEN (10)
#define OC_TWO_TOKEN (11)
#define OC_MINUS_TWO_TOKEN (12)
#define OC_DCT_VAL_CAT2 (13)
#define OC_DCT_VAL_CAT3 (17)
#define OC_DCT_VAL_CAT4 (18)
#define OC_DCT_VAL_CAT5 (19)
#define OC_DCT_VAL_CAT6 (20)
#define OC_DCT_VAL_CAT7 (21)
#define OC_DCT_VAL_CAT8 (22)
#define OC_DCT_RUN_CAT1A (23)
#define OC_DCT_RUN_CAT1B (28)
#define OC_DCT_RUN_CAT1C (29)
#define OC_DCT_RUN_CAT2A (30)
#define OC_DCT_RUN_CAT2B (31)
#define OC_NDCT_EOB_TOKEN_MAX (7)
#define OC_NDCT_ZRL_TOKEN_MAX (9)
#define OC_NDCT_VAL_MAX (23)
#define OC_NDCT_VAL_CAT1_MAX (13)
#define OC_NDCT_VAL_CAT2_MAX (17)
#define OC_NDCT_VAL_CAT2_SIZE (OC_NDCT_VAL_CAT2_MAX-OC_DCT_VAL_CAT2)
#define OC_NDCT_RUN_MAX (32)
#define OC_NDCT_RUN_CAT1A_MAX (28)
extern const int OC_DCT_TOKEN_EXTRA_BITS[TH_NDCT_TOKENS];
#endif

268
Engine/lib/libtheora/lib/dec/idct.c Normal file
View file

@ -0,0 +1,268 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: idct.c 15400 2008-10-15 12:10:58Z tterribe $
********************************************************************/
#include <string.h>
#include <ogg/ogg.h>
#include "dct.h"
#include "idct.h"
/*Performs an inverse 8 point Type-II DCT transform.
The output is scaled by a factor of 2 relative to the orthonormal version of
the transform.
_y: The buffer to store the result in.
Data will be placed in every 8th entry (e.g., in a column of an 8x8
block).
_x: The input coefficients.
The first 8 entries are used (e.g., from a row of an 8x8 block).*/
static void idct8(ogg_int16_t *_y,const ogg_int16_t _x[8]){
ogg_int32_t t[8];
ogg_int32_t r;
/*Stage 1:*/
/*0-1 butterfly.*/
t[0]=OC_C4S4*(ogg_int16_t)(_x[0]+_x[4])>>16;
t[1]=OC_C4S4*(ogg_int16_t)(_x[0]-_x[4])>>16;
/*2-3 rotation by 6pi/16.*/
t[2]=(OC_C6S2*_x[2]>>16)-(OC_C2S6*_x[6]>>16);
t[3]=(OC_C2S6*_x[2]>>16)+(OC_C6S2*_x[6]>>16);
/*4-7 rotation by 7pi/16.*/
t[4]=(OC_C7S1*_x[1]>>16)-(OC_C1S7*_x[7]>>16);
/*5-6 rotation by 3pi/16.*/
t[5]=(OC_C3S5*_x[5]>>16)-(OC_C5S3*_x[3]>>16);
t[6]=(OC_C5S3*_x[5]>>16)+(OC_C3S5*_x[3]>>16);
t[7]=(OC_C1S7*_x[1]>>16)+(OC_C7S1*_x[7]>>16);
/*Stage 2:*/
/*4-5 butterfly.*/
r=t[4]+t[5];
t[5]=OC_C4S4*(ogg_int16_t)(t[4]-t[5])>>16;
t[4]=r;
/*7-6 butterfly.*/
r=t[7]+t[6];
t[6]=OC_C4S4*(ogg_int16_t)(t[7]-t[6])>>16;
t[7]=r;
/*Stage 3:*/
/*0-3 butterfly.*/
r=t[0]+t[3];
t[3]=t[0]-t[3];
t[0]=r;
/*1-2 butterfly.*/
r=t[1]+t[2];
t[2]=t[1]-t[2];
t[1]=r;
/*6-5 butterfly.*/
r=t[6]+t[5];
t[5]=t[6]-t[5];
t[6]=r;
/*Stage 4:*/
/*0-7 butterfly.*/
_y[0<<3]=(ogg_int16_t)(t[0]+t[7]);
/*1-6 butterfly.*/
_y[1<<3]=(ogg_int16_t)(t[1]+t[6]);
/*2-5 butterfly.*/
_y[2<<3]=(ogg_int16_t)(t[2]+t[5]);
/*3-4 butterfly.*/
_y[3<<3]=(ogg_int16_t)(t[3]+t[4]);
_y[4<<3]=(ogg_int16_t)(t[3]-t[4]);
_y[5<<3]=(ogg_int16_t)(t[2]-t[5]);
_y[6<<3]=(ogg_int16_t)(t[1]-t[6]);
_y[7<<3]=(ogg_int16_t)(t[0]-t[7]);
}
/*Performs an inverse 8 point Type-II DCT transform.
The output is scaled by a factor of 2 relative to the orthonormal version of
the transform.
_y: The buffer to store the result in.
Data will be placed in every 8th entry (e.g., in a column of an 8x8
block).
_x: The input coefficients.
Only the first 4 entries are used.
The other 4 are assumed to be 0.*/
static void idct8_4(ogg_int16_t *_y,const ogg_int16_t _x[8]){
ogg_int32_t t[8];
ogg_int32_t r;
/*Stage 1:*/
t[0]=OC_C4S4*_x[0]>>16;
t[2]=OC_C6S2*_x[2]>>16;
t[3]=OC_C2S6*_x[2]>>16;
t[4]=OC_C7S1*_x[1]>>16;
t[5]=-(OC_C5S3*_x[3]>>16);
t[6]=OC_C3S5*_x[3]>>16;
t[7]=OC_C1S7*_x[1]>>16;
/*Stage 2:*/
r=t[4]+t[5];
t[5]=OC_C4S4*(ogg_int16_t)(t[4]-t[5])>>16;
t[4]=r;
r=t[7]+t[6];
t[6]=OC_C4S4*(ogg_int16_t)(t[7]-t[6])>>16;
t[7]=r;
/*Stage 3:*/
t[1]=t[0]+t[2];
t[2]=t[0]-t[2];
r=t[0]+t[3];
t[3]=t[0]-t[3];
t[0]=r;
r=t[6]+t[5];
t[5]=t[6]-t[5];
t[6]=r;
/*Stage 4:*/
_y[0<<3]=(ogg_int16_t)(t[0]+t[7]);
_y[1<<3]=(ogg_int16_t)(t[1]+t[6]);
_y[2<<3]=(ogg_int16_t)(t[2]+t[5]);
_y[3<<3]=(ogg_int16_t)(t[3]+t[4]);
_y[4<<3]=(ogg_int16_t)(t[3]-t[4]);
_y[5<<3]=(ogg_int16_t)(t[2]-t[5]);
_y[6<<3]=(ogg_int16_t)(t[1]-t[6]);
_y[7<<3]=(ogg_int16_t)(t[0]-t[7]);
}
/*Performs an inverse 8 point Type-II DCT transform.
The output is scaled by a factor of 2 relative to the orthonormal version of
the transform.
_y: The buffer to store the result in.
Data will be placed in every 8th entry (e.g., in a column of an 8x8
block).
_x: The input coefficients.
Only the first 3 entries are used.
The other 5 are assumed to be 0.*/
static void idct8_3(ogg_int16_t *_y,const ogg_int16_t _x[8]){
ogg_int32_t t[8];
ogg_int32_t r;
/*Stage 1:*/
t[0]=OC_C4S4*_x[0]>>16;
t[2]=OC_C6S2*_x[2]>>16;
t[3]=OC_C2S6*_x[2]>>16;
t[4]=OC_C7S1*_x[1]>>16;
t[7]=OC_C1S7*_x[1]>>16;
/*Stage 2:*/
t[5]=OC_C4S4*t[4]>>16;
t[6]=OC_C4S4*t[7]>>16;
/*Stage 3:*/
t[1]=t[0]+t[2];
t[2]=t[0]-t[2];
r=t[0]+t[3];
t[3]=t[0]-t[3];
t[0]=r;
r=t[6]+t[5];
t[5]=t[6]-t[5];
t[6]=r;
/*Stage 4:*/
_y[0<<3]=(ogg_int16_t)(t[0]+t[7]);
_y[1<<3]=(ogg_int16_t)(t[1]+t[6]);
_y[2<<3]=(ogg_int16_t)(t[2]+t[5]);
_y[3<<3]=(ogg_int16_t)(t[3]+t[4]);
_y[4<<3]=(ogg_int16_t)(t[3]-t[4]);
_y[5<<3]=(ogg_int16_t)(t[2]-t[5]);
_y[6<<3]=(ogg_int16_t)(t[1]-t[6]);
_y[7<<3]=(ogg_int16_t)(t[0]-t[7]);
}
/*Performs an inverse 8 point Type-II DCT transform.
The output is scaled by a factor of 2 relative to the orthonormal version of
the transform.
_y: The buffer to store the result in.
Data will be placed in every 8th entry (e.g., in a column of an 8x8
block).
_x: The input coefficients.
Only the first 2 entries are used.
The other 6 are assumed to be 0.*/
static void idct8_2(ogg_int16_t *_y,const ogg_int16_t _x[8]){
ogg_int32_t t[8];
ogg_int32_t r;
/*Stage 1:*/
t[0]=OC_C4S4*_x[0]>>16;
t[4]=OC_C7S1*_x[1]>>16;
t[7]=OC_C1S7*_x[1]>>16;
/*Stage 2:*/
t[5]=OC_C4S4*t[4]>>16;
t[6]=OC_C4S4*t[7]>>16;
/*Stage 3:*/
r=t[6]+t[5];
t[5]=t[6]-t[5];
t[6]=r;
/*Stage 4:*/
_y[0<<3]=(ogg_int16_t)(t[0]+t[7]);
_y[1<<3]=(ogg_int16_t)(t[0]+t[6]);
_y[2<<3]=(ogg_int16_t)(t[0]+t[5]);
_y[3<<3]=(ogg_int16_t)(t[0]+t[4]);
_y[4<<3]=(ogg_int16_t)(t[0]-t[4]);
_y[5<<3]=(ogg_int16_t)(t[0]-t[5]);
_y[6<<3]=(ogg_int16_t)(t[0]-t[6]);
_y[7<<3]=(ogg_int16_t)(t[0]-t[7]);
}
/*Performs an inverse 8 point Type-II DCT transform.
The output is scaled by a factor of 2 relative to the orthonormal version of
the transform.
_y: The buffer to store the result in.
Data will be placed in every 8th entry (e.g., in a column of an 8x8
block).
_x: The input coefficients.
Only the first entry is used.
The other 7 are assumed to be 0.*/
static void idct8_1(ogg_int16_t *_y,const ogg_int16_t _x[1]){
_y[0<<3]=_y[1<<3]=_y[2<<3]=_y[3<<3]=
_y[4<<3]=_y[5<<3]=_y[6<<3]=_y[7<<3]=(ogg_int16_t)(OC_C4S4*_x[0]>>16);
}
/*Performs an inverse 8x8 Type-II DCT transform.
The input is assumed to be scaled by a factor of 4 relative to orthonormal
version of the transform.
_y: The buffer to store the result in.
This may be the same as _x.
_x: The input coefficients. */
void oc_idct8x8_c(ogg_int16_t _y[64],const ogg_int16_t _x[64]){
const ogg_int16_t *in;
ogg_int16_t *end;
ogg_int16_t *out;
ogg_int16_t w[64];
/*Transform rows of x into columns of w.*/
for(in=_x,out=w,end=out+8;out<end;in+=8,out++)idct8(out,in);
/*Transform rows of w into columns of y.*/
for(in=w,out=_y,end=out+8;out<end;in+=8,out++)idct8(out,in);
/*Adjust for scale factor.*/
for(out=_y,end=out+64;out<end;out++)*out=(ogg_int16_t)(*out+8>>4);
}
/*Performs an inverse 8x8 Type-II DCT transform.
The input is assumed to be scaled by a factor of 4 relative to orthonormal
version of the transform.
All coefficients but the first 10 in zig-zag scan order are assumed to be 0:
x x x x 0 0 0 0
x x x 0 0 0 0 0
x x 0 0 0 0 0 0
x 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
_y: The buffer to store the result in.
This may be the same as _x.
_x: The input coefficients. */
void oc_idct8x8_10_c(ogg_int16_t _y[64],const ogg_int16_t _x[64]){
const ogg_int16_t *in;
ogg_int16_t *end;
ogg_int16_t *out;
ogg_int16_t w[64];
/*Transform rows of x into columns of w.*/
idct8_4(w,_x);
idct8_3(w+1,_x+8);
idct8_2(w+2,_x+16);
idct8_1(w+3,_x+24);
/*Transform rows of w into columns of y.*/
for(in=w,out=_y,end=out+8;out<end;in+=8,out++)idct8_4(out,in);
/*Adjust for scale factor.*/
for(out=_y,end=out+64;out<end;out++)*out=(ogg_int16_t)(*out+8>>4);
}

26
Engine/lib/libtheora/lib/dec/idct.h Normal file
View file

@ -0,0 +1,26 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: idct.h 15400 2008-10-15 12:10:58Z tterribe $
********************************************************************/
/*Inverse DCT transforms.*/
#include <ogg/ogg.h>
#if !defined(_idct_H)
# define _idct_H (1)
void oc_idct8x8_c(ogg_int16_t _y[64],const ogg_int16_t _x[64]);
void oc_idct8x8_10_c(ogg_int16_t _y[64],const ogg_int16_t _x[64]);
#endif

123
Engine/lib/libtheora/lib/dec/info.c Normal file
View file

@ -0,0 +1,123 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: info.c 15400 2008-10-15 12:10:58Z tterribe $
********************************************************************/
#include <stdlib.h>
#include <ctype.h>
#include <string.h>
#include "../internal.h"
/*This is more or less the same as strncasecmp, but that doesn't exist
everywhere, and this is a fairly trivial function, so we include it.
Note: We take advantage of the fact that we know _n is less than or equal to
the length of at least one of the strings.*/
static int oc_tagcompare(const char *_s1,const char *_s2,int _n){
int c;
for(c=0;c<_n;c++){
if(toupper(_s1[c])!=toupper(_s2[c]))return !0;
}
return _s1[c]!='=';
}
void th_info_init(th_info *_info){
memset(_info,0,sizeof(*_info));
_info->version_major=TH_VERSION_MAJOR;
_info->version_minor=TH_VERSION_MINOR;
_info->version_subminor=TH_VERSION_SUB;
_info->keyframe_granule_shift=6;
}
void th_info_clear(th_info *_info){
memset(_info,0,sizeof(*_info));
}
void th_comment_init(th_comment *_tc){
memset(_tc,0,sizeof(*_tc));
}
void th_comment_add(th_comment *_tc,char *_comment){
int comment_len;
_tc->user_comments=_ogg_realloc(_tc->user_comments,
(_tc->comments+2)*sizeof(*_tc->user_comments));
_tc->comment_lengths=_ogg_realloc(_tc->comment_lengths,
(_tc->comments+2)*sizeof(*_tc->comment_lengths));
comment_len=strlen(_comment);
_tc->comment_lengths[_tc->comments]=comment_len;
_tc->user_comments[_tc->comments]=_ogg_malloc(comment_len+1);
memcpy(_tc->user_comments[_tc->comments],_comment,comment_len+1);
_tc->comments++;
_tc->user_comments[_tc->comments]=NULL;
}
void th_comment_add_tag(th_comment *_tc,char *_tag,char *_val){
char *comment;
int tag_len;
int val_len;
tag_len=strlen(_tag);
val_len=strlen(_val);
/*+2 for '=' and '\0'.*/
comment=_ogg_malloc(tag_len+val_len+2);
memcpy(comment,_tag,tag_len);
comment[tag_len]='=';
memcpy(comment+tag_len+1,_val,val_len+1);
th_comment_add(_tc,comment);
_ogg_free(comment);
}
char *th_comment_query(th_comment *_tc,char *_tag,int _count){
long i;
int found;
int tag_len;
tag_len=strlen(_tag);
found=0;
for(i=0;i<_tc->comments;i++){
if(!oc_tagcompare(_tc->user_comments[i],_tag,tag_len)){
/*We return a pointer to the data, not a copy.*/
if(_count==found++)return _tc->user_comments[i]+tag_len+1;
}
}
/*Didn't find anything.*/
return NULL;
}
int th_comment_query_count(th_comment *_tc,char *_tag){
long i;
int tag_len;
int count;
tag_len=strlen(_tag);
count=0;
for(i=0;i<_tc->comments;i++){
if(!oc_tagcompare(_tc->user_comments[i],_tag,tag_len))count++;
}
return count;
}
void th_comment_clear(th_comment *_tc){
if(_tc!=NULL){
long i;
for(i=0;i<_tc->comments;i++)_ogg_free(_tc->user_comments[i]);
_ogg_free(_tc->user_comments);
_ogg_free(_tc->comment_lengths);
_ogg_free(_tc->vendor);
memset(_tc,0,sizeof(*_tc));
}
}

383
Engine/lib/libtheora/lib/dec/internal.c Normal file
View file

@ -0,0 +1,383 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: internal.c 15400 2008-10-15 12:10:58Z tterribe $
********************************************************************/
#include <stdlib.h>
#include <limits.h>
#include <string.h>
#include "../internal.h"
#include "idct.h"
/*A map from the index in the zig zag scan to the coefficient number in a
block.
All zig zag indices beyond 63 are sent to coefficient 64, so that zero runs
past the end of a block in bogus streams get mapped to a known location.*/
const int OC_FZIG_ZAG[128]={
0, 1, 8,16, 9, 2, 3,10,
17,24,32,25,18,11, 4, 5,
12,19,26,33,40,48,41,34,
27,20,13, 6, 7,14,21,28,
35,42,49,56,57,50,43,36,
29,22,15,23,30,37,44,51,
58,59,52,45,38,31,39,46,
53,60,61,54,47,55,62,63,
64,64,64,64,64,64,64,64,
64,64,64,64,64,64,64,64,
64,64,64,64,64,64,64,64,
64,64,64,64,64,64,64,64,
64,64,64,64,64,64,64,64,
64,64,64,64,64,64,64,64,
64,64,64,64,64,64,64,64,
64,64,64,64,64,64,64,64
};
/*A map from the coefficient number in a block to its index in the zig zag
scan.*/
const int OC_IZIG_ZAG[64]={
0, 1, 5, 6,14,15,27,28,
2, 4, 7,13,16,26,29,42,
3, 8,12,17,25,30,41,43,
9,11,18,24,31,40,44,53,
10,19,23,32,39,45,52,54,
20,22,33,38,46,51,55,60,
21,34,37,47,50,56,59,61,
35,36,48,49,57,58,62,63
};
/*The predictor frame to use for each macro block mode.*/
const int OC_FRAME_FOR_MODE[8]={
/*OC_MODE_INTER_NOMV*/
OC_FRAME_PREV,
/*OC_MODE_INTRA*/
OC_FRAME_SELF,
/*OC_MODE_INTER_MV*/
OC_FRAME_PREV,
/*OC_MODE_INTER_MV_LAST*/
OC_FRAME_PREV,
/*OC_MODE_INTER_MV_LAST2*/
OC_FRAME_PREV,
/*OC_MODE_GOLDEN*/
OC_FRAME_GOLD,
/*OC_MODE_GOLDEN_MV*/
OC_FRAME_GOLD,
/*OC_MODE_INTER_MV_FOUR*/
OC_FRAME_PREV,
};
/*A map from physical macro block ordering to bitstream macro block
ordering within a super block.*/
const int OC_MB_MAP[2][2]={{0,3},{1,2}};
/*A list of the indices in the oc_mb.map array that can be valid for each of
the various chroma decimation types.*/
const int OC_MB_MAP_IDXS[TH_PF_NFORMATS][12]={
{0,1,2,3,4,8},
{0,1,2,3,4,5,8,9},
{0,1,2,3,4,6,8,10},
{0,1,2,3,4,5,6,7,8,9,10,11}
};
/*The number of indices in the oc_mb.map array that can be valid for each of
the various chroma decimation types.*/
const int OC_MB_MAP_NIDXS[TH_PF_NFORMATS]={6,8,8,12};
/*The number of extra bits that are coded with each of the DCT tokens.
Each DCT token has some fixed number of additional bits (possibly 0) stored
after the token itself, containing, for example, coefficient magnitude,
sign bits, etc.*/
const int OC_DCT_TOKEN_EXTRA_BITS[TH_NDCT_TOKENS]={
0,0,0,2,3,4,12,3,6,
0,0,0,0,
1,1,1,1,2,3,4,5,6,10,
1,1,1,1,1,3,4,
2,3
};
int oc_ilog(unsigned _v){
int ret;
for(ret=0;_v;ret++)_v>>=1;
return ret;
}
/*Determines the number of blocks or coefficients to be skipped for a given
token value.
_token: The token value to skip.
_extra_bits: The extra bits attached to this token.
Return: A positive value indicates that number of coefficients are to be
skipped in the current block.
Otherwise, the negative of the return value indicates that number of
blocks are to be ended.*/
typedef int (*oc_token_skip_func)(int _token,int _extra_bits);
/*Handles the simple end of block tokens.*/
static int oc_token_skip_eob(int _token,int _extra_bits){
static const int NBLOCKS_ADJUST[OC_NDCT_EOB_TOKEN_MAX]={1,2,3,4,8,16,0};
return -_extra_bits-NBLOCKS_ADJUST[_token];
}
/*The last EOB token has a special case, where an EOB run of size zero ends all
the remaining blocks in the frame.*/
static int oc_token_skip_eob6(int _token,int _extra_bits){
if(!_extra_bits)return -INT_MAX;
return -_extra_bits;
}
/*Handles the pure zero run tokens.*/
static int oc_token_skip_zrl(int _token,int _extra_bits){
return _extra_bits+1;
}
/*Handles a normal coefficient value token.*/
static int oc_token_skip_val(void){
return 1;
}
/*Handles a category 1A zero run/coefficient value combo token.*/
static int oc_token_skip_run_cat1a(int _token){
return _token-OC_DCT_RUN_CAT1A+2;
}
/*Handles category 1b and 2 zero run/coefficient value combo tokens.*/
static int oc_token_skip_run(int _token,int _extra_bits){
static const int NCOEFFS_ADJUST[OC_NDCT_RUN_MAX-OC_DCT_RUN_CAT1B]={
7,11,2,3
};
static const int NCOEFFS_MASK[OC_NDCT_RUN_MAX-OC_DCT_RUN_CAT1B]={
3,7,0,1
};
_token-=OC_DCT_RUN_CAT1B;
return (_extra_bits&NCOEFFS_MASK[_token])+NCOEFFS_ADJUST[_token];
}
/*A jump table for computing the number of coefficients or blocks to skip for
a given token value.
This reduces all the conditional branches, etc., needed to parse these token
values down to one indirect jump.*/
static const oc_token_skip_func OC_TOKEN_SKIP_TABLE[TH_NDCT_TOKENS]={
oc_token_skip_eob,
oc_token_skip_eob,
oc_token_skip_eob,
oc_token_skip_eob,
oc_token_skip_eob,
oc_token_skip_eob,
oc_token_skip_eob6,
oc_token_skip_zrl,
oc_token_skip_zrl,
(oc_token_skip_func)oc_token_skip_val,
(oc_token_skip_func)oc_token_skip_val,
(oc_token_skip_func)oc_token_skip_val,
(oc_token_skip_func)oc_token_skip_val,
(oc_token_skip_func)oc_token_skip_val,
(oc_token_skip_func)oc_token_skip_val,
(oc_token_skip_func)oc_token_skip_val,
(oc_token_skip_func)oc_token_skip_val,
(oc_token_skip_func)oc_token_skip_val,
(oc_token_skip_func)oc_token_skip_val,
(oc_token_skip_func)oc_token_skip_val,
(oc_token_skip_func)oc_token_skip_val,
(oc_token_skip_func)oc_token_skip_val,
(oc_token_skip_func)oc_token_skip_val,
(oc_token_skip_func)oc_token_skip_run_cat1a,
(oc_token_skip_func)oc_token_skip_run_cat1a,
(oc_token_skip_func)oc_token_skip_run_cat1a,
(oc_token_skip_func)oc_token_skip_run_cat1a,
(oc_token_skip_func)oc_token_skip_run_cat1a,
oc_token_skip_run,
oc_token_skip_run,
oc_token_skip_run,
oc_token_skip_run
};
/*Determines the number of blocks or coefficients to be skipped for a given
token value.
_token: The token value to skip.
_extra_bits: The extra bits attached to this token.
Return: A positive value indicates that number of coefficients are to be
skipped in the current block.
Otherwise, the negative of the return value indicates that number of
blocks are to be ended.
0 will never be returned, so that at least one coefficient in one
block will always be decoded for every token.*/
int oc_dct_token_skip(int _token,int _extra_bits){
return (*OC_TOKEN_SKIP_TABLE[_token])(_token,_extra_bits);
}
/*The function used to fill in the chroma plane motion vectors for a macro
block when 4 different motion vectors are specified in the luma plane.
This version is for use with chroma decimated in the X and Y directions.
_cbmvs: The chroma block-level motion vectors to fill in.
_lbmvs: The luma block-level motion vectors.*/
static void oc_set_chroma_mvs00(oc_mv _cbmvs[4],const oc_mv _lbmvs[4]){
int dx;
int dy;
dx=_lbmvs[0][0]+_lbmvs[1][0]+_lbmvs[2][0]+_lbmvs[3][0];
dy=_lbmvs[0][1]+_lbmvs[1][1]+_lbmvs[2][1]+_lbmvs[3][1];
_cbmvs[0][0]=(signed char)OC_DIV_ROUND_POW2(dx,2,2);
_cbmvs[0][1]=(signed char)OC_DIV_ROUND_POW2(dy,2,2);
}
/*The function used to fill in the chroma plane motion vectors for a macro
block when 4 different motion vectors are specified in the luma plane.
This version is for use with chroma decimated in the Y direction.
_cbmvs: The chroma block-level motion vectors to fill in.
_lbmvs: The luma block-level motion vectors.*/
static void oc_set_chroma_mvs01(oc_mv _cbmvs[4],const oc_mv _lbmvs[4]){
int dx;
int dy;
dx=_lbmvs[0][0]+_lbmvs[2][0];
dy=_lbmvs[0][1]+_lbmvs[2][1];
_cbmvs[0][0]=(signed char)OC_DIV_ROUND_POW2(dx,1,1);
_cbmvs[0][1]=(signed char)OC_DIV_ROUND_POW2(dy,1,1);
dx=_lbmvs[1][0]+_lbmvs[3][0];
dy=_lbmvs[1][1]+_lbmvs[3][1];
_cbmvs[1][0]=(signed char)OC_DIV_ROUND_POW2(dx,1,1);
_cbmvs[1][1]=(signed char)OC_DIV_ROUND_POW2(dy,1,1);
}
/*The function used to fill in the chroma plane motion vectors for a macro
block when 4 different motion vectors are specified in the luma plane.
This version is for use with chroma decimated in the X direction.
_cbmvs: The chroma block-level motion vectors to fill in.
_lbmvs: The luma block-level motion vectors.*/
static void oc_set_chroma_mvs10(oc_mv _cbmvs[4],const oc_mv _lbmvs[4]){
int dx;
int dy;
dx=_lbmvs[0][0]+_lbmvs[1][0];
dy=_lbmvs[0][1]+_lbmvs[1][1];
_cbmvs[0][0]=(signed char)OC_DIV_ROUND_POW2(dx,1,1);
_cbmvs[0][1]=(signed char)OC_DIV_ROUND_POW2(dy,1,1);
dx=_lbmvs[2][0]+_lbmvs[3][0];
dy=_lbmvs[2][1]+_lbmvs[3][1];
_cbmvs[2][0]=(signed char)OC_DIV_ROUND_POW2(dx,1,1);
_cbmvs[2][1]=(signed char)OC_DIV_ROUND_POW2(dy,1,1);
}
/*The function used to fill in the chroma plane motion vectors for a macro
block when 4 different motion vectors are specified in the luma plane.
This version is for use with no chroma decimation.
_cbmvs: The chroma block-level motion vectors to fill in.
_lmbmv: The luma macro-block level motion vector to fill in for use in
prediction.
_lbmvs: The luma block-level motion vectors.*/
static void oc_set_chroma_mvs11(oc_mv _cbmvs[4],const oc_mv _lbmvs[4]){
memcpy(_cbmvs,_lbmvs,4*sizeof(_lbmvs[0]));
}
/*A table of functions used to fill in the chroma plane motion vectors for a
macro block when 4 different motion vectors are specified in the luma
plane.*/
const oc_set_chroma_mvs_func OC_SET_CHROMA_MVS_TABLE[TH_PF_NFORMATS]={
(oc_set_chroma_mvs_func)oc_set_chroma_mvs00,
(oc_set_chroma_mvs_func)oc_set_chroma_mvs01,
(oc_set_chroma_mvs_func)oc_set_chroma_mvs10,
(oc_set_chroma_mvs_func)oc_set_chroma_mvs11
};
void **oc_malloc_2d(size_t _height,size_t _width,size_t _sz){
size_t rowsz;
size_t colsz;
size_t datsz;
char *ret;
colsz=_height*sizeof(void *);
rowsz=_sz*_width;
datsz=rowsz*_height;
/*Alloc array and row pointers.*/
ret=(char *)_ogg_malloc(datsz+colsz);
/*Initialize the array.*/
if(ret!=NULL){
size_t i;
void **p;
char *datptr;
p=(void **)ret;
i=_height;
for(datptr=ret+colsz;i-->0;p++,datptr+=rowsz)*p=(void *)datptr;
}
return (void **)ret;
}
void **oc_calloc_2d(size_t _height,size_t _width,size_t _sz){
size_t colsz;
size_t rowsz;
size_t datsz;
char *ret;
colsz=_height*sizeof(void *);
rowsz=_sz*_width;
datsz=rowsz*_height;
/*Alloc array and row pointers.*/
ret=(char *)_ogg_calloc(datsz+colsz,1);
/*Initialize the array.*/
if(ret!=NULL){
size_t i;
void **p;
char *datptr;
p=(void **)ret;
i=_height;
for(datptr=ret+colsz;i-->0;p++,datptr+=rowsz)*p=(void *)datptr;
}
return (void **)ret;
}
void oc_free_2d(void *_ptr){
_ogg_free(_ptr);
}
/*Fills in a Y'CbCr buffer with a pointer to the image data in the first
buffer, but with the opposite vertical orientation.
_dst: The destination buffer.
This can be the same as _src.
_src: The source buffer.*/
void oc_ycbcr_buffer_flip(th_ycbcr_buffer _dst,
const th_ycbcr_buffer _src){
int pli;
for(pli=0;pli<3;pli++){
_dst[pli].width=_src[pli].width;
_dst[pli].height=_src[pli].height;
_dst[pli].stride=-_src[pli].stride;
_dst[pli].data=_src[pli].data+(1-_dst[pli].height)*_dst[pli].stride;
}
}
const char *th_version_string(void){
return OC_VENDOR_STRING;
}
ogg_uint32_t th_version_number(void){
return (TH_VERSION_MAJOR<<16)+(TH_VERSION_MINOR<<8)+(TH_VERSION_SUB);
}
/*Determines the packet type.
Note that this correctly interprets a 0-byte packet as a video data packet.
Return: 1 for a header packet, 0 for a data packet.*/
int th_packet_isheader(ogg_packet *_op){
return _op->bytes>0?_op->packet[0]>>7:0;
}
/*Determines the frame type of a video data packet.
Note that this correctly interprets a 0-byte packet as a delta frame.
Return: 1 for a key frame, 0 for a delta frame, and -1 for a header
packet.*/
int th_packet_iskeyframe(ogg_packet *_op){
return _op->bytes<=0?0:_op->packet[0]&0x80?-1:!(_op->packet[0]&0x40);
}

88
Engine/lib/libtheora/lib/dec/ocintrin.h Normal file
View file

@ -0,0 +1,88 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: ocintrin.h 15400 2008-10-15 12:10:58Z tterribe $
********************************************************************/
/*Some common macros for potential platform-specific optimization.*/
#include <math.h>
#if !defined(_ocintrin_H)
# define _ocintrin_H (1)
/*Some specific platforms may have optimized intrinsic or inline assembly
versions of these functions which can substantially improve performance.
We define macros for them to allow easy incorporation of these non-ANSI
features.*/
/*Branchless, but not correct for differences larger than INT_MAX.
static int oc_mini(int _a,int _b){
int ambsign;
ambsign=_a-_b>>sizeof(int)*8-1;
return (_a&~ambsign)+(_b&ambsign);
}*/
#define OC_MAXI(_a,_b) ((_a)<(_b)?(_b):(_a))
#define OC_MINI(_a,_b) ((_a)>(_b)?(_b):(_a))
/*Clamps an integer into the given range.
If _a>_c, then the lower bound _a is respected over the upper bound _c (this
behavior is required to meet our documented API behavior).
_a: The lower bound.
_b: The value to clamp.
_c: The upper boud.*/
#define OC_CLAMPI(_a,_b,_c) (OC_MAXI(_a,OC_MINI(_b,_c)))
#define OC_CLAMP255(_x) ((unsigned char)((((_x)<0)-1)&((_x)|-((_x)>255))))
/*Divides an integer by a power of two, truncating towards 0.
_dividend: The integer to divide.
_shift: The non-negative power of two to divide by.
_rmask: (1<<_shift)-1*/
#define OC_DIV_POW2(_dividend,_shift,_rmask)\
((_dividend)+(((_dividend)>>sizeof(_dividend)*8-1)&(_rmask))>>(_shift))
/*Divides _x by 65536, truncating towards 0.*/
#define OC_DIV2_16(_x) OC_DIV_POW2(_x,16,0xFFFF)
/*Divides _x by 2, truncating towards 0.*/
#define OC_DIV2(_x) OC_DIV_POW2(_x,1,0x1)
/*Divides _x by 8, truncating towards 0.*/
#define OC_DIV8(_x) OC_DIV_POW2(_x,3,0x7)
/*Divides _x by 16, truncating towards 0.*/
#define OC_DIV16(_x) OC_DIV_POW2(_x,4,0xF)
/*Right shifts _dividend by _shift, adding _rval, and subtracting one for
negative dividends first..
When _rval is (1<<_shift-1), this is equivalent to division with rounding
ties towards positive infinity.*/
#define OC_DIV_ROUND_POW2(_dividend,_shift,_rval)\
((_dividend)+((_dividend)>>sizeof(_dividend)*8-1)+(_rval)>>(_shift))
/*Swaps two integers _a and _b if _a>_b.*/
#define OC_SORT2I(_a,_b)\
if((_a)>(_b)){\
int t__;\
t__=(_a);\
(_a)=(_b);\
(_b)=t__;\
}
/*All of these macros should expect floats as arguments.*/
#define OC_MAXF(_a,_b) ((_a)<(_b)?(_b):(_a))
#define OC_MINF(_a,_b) ((_a)>(_b)?(_b):(_a))
#define OC_CLAMPF(_a,_b,_c) (OC_MINF(_a,OC_MAXF(_b,_c)))
#define OC_FABSF(_f) ((float)fabs(_f))
#define OC_SQRTF(_f) ((float)sqrt(_f))
#define OC_POWF(_b,_e) ((float)pow(_b,_e))
#define OC_LOGF(_f) ((float)log(_f))
#define OC_IFLOORF(_f) ((int)floor(_f))
#define OC_ICEILF(_f) ((int)ceil(_f))
#endif

122
Engine/lib/libtheora/lib/dec/quant.c Normal file
View file

@ -0,0 +1,122 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: quant.c 15400 2008-10-15 12:10:58Z tterribe $
********************************************************************/
#include <stdlib.h>
#include <string.h>
#include <ogg/ogg.h>
#include "quant.h"
#include "decint.h"
static const unsigned OC_DC_QUANT_MIN[2]={4<<2,8<<2};
static const unsigned OC_AC_QUANT_MIN[2]={2<<2,4<<2};
/*Initializes the dequantization tables from a set of quantizer info.
Currently the dequantizer (and elsewhere enquantizer) tables are expected to
be initialized as pointing to the storage reserved for them in the
oc_theora_state (resp. oc_enc_ctx) structure.
If some tables are duplicates of others, the pointers will be adjusted to
point to a single copy of the tables, but the storage for them will not be
freed.
If you're concerned about the memory footprint, the obvious thing to do is
to move the storage out of its fixed place in the structures and allocate
it on demand.
However, a much, much better option is to only store the quantization
matrices being used for the current frame, and to recalculate these as the
qi values change between frames (this is what VP3 did).*/
void oc_dequant_tables_init(oc_quant_table *_dequant[2][3],
int _pp_dc_scale[64],const th_quant_info *_qinfo){
/*coding mode: intra or inter.*/
int qti;
/*Y', C_b, C_r*/
int pli;
for(qti=0;qti<2;qti++){
for(pli=0;pli<3;pli++){
oc_quant_tables stage;
/*Quality index.*/
int qi;
/*Range iterator.*/
int qri;
for(qi=0,qri=0; qri<=_qinfo->qi_ranges[qti][pli].nranges; qri++){
th_quant_base base;
ogg_uint32_t q;
int qi_start;
int qi_end;
int ci;
memcpy(base,_qinfo->qi_ranges[qti][pli].base_matrices[qri],
sizeof(base));
qi_start=qi;
if(qri==_qinfo->qi_ranges[qti][pli].nranges)qi_end=qi+1;
else qi_end=qi+_qinfo->qi_ranges[qti][pli].sizes[qri];
/*Iterate over quality indicies in this range.*/
for(;;){
ogg_uint32_t qfac;
/*In the original VP3.2 code, the rounding offset and the size of the
dead zone around 0 were controlled by a "sharpness" parameter.
The size of our dead zone is now controlled by the per-coefficient
quality thresholds returned by our HVS module.
We round down from a more accurate value when the quality of the
reconstruction does not fall below our threshold and it saves bits.
Hence, all of that VP3.2 code is gone from here, and the remaining
floating point code has been implemented as equivalent integer code
with exact precision.*/
qfac=(ogg_uint32_t)_qinfo->dc_scale[qi]*base[0];
/*For postprocessing, not dequantization.*/
if(_pp_dc_scale!=NULL)_pp_dc_scale[qi]=(int)(qfac/160);
/*Scale DC the coefficient from the proper table.*/
q=(qfac/100)<<2;
q=OC_CLAMPI(OC_DC_QUANT_MIN[qti],q,OC_QUANT_MAX);
stage[qi][0]=(ogg_uint16_t)q;
/*Now scale AC coefficients from the proper table.*/
for(ci=1;ci<64;ci++){
q=((ogg_uint32_t)_qinfo->ac_scale[qi]*base[ci]/100)<<2;
q=OC_CLAMPI(OC_AC_QUANT_MIN[qti],q,OC_QUANT_MAX);
stage[qi][ci]=(ogg_uint16_t)q;
}
if(++qi>=qi_end)break;
/*Interpolate the next base matrix.*/
for(ci=0;ci<64;ci++){
base[ci]=(unsigned char)(
(2*((qi_end-qi)*_qinfo->qi_ranges[qti][pli].base_matrices[qri][ci]+
(qi-qi_start)*_qinfo->qi_ranges[qti][pli].base_matrices[qri+1][ci])
+_qinfo->qi_ranges[qti][pli].sizes[qri])/
(2*_qinfo->qi_ranges[qti][pli].sizes[qri]));
}
}
}
/*Staging matrices complete; commit to memory only if this isn't a
duplicate of a preceeding plane.
This simple check helps us improve cache coherency later.*/
{
int dupe;
int qtj;
int plj;
dupe=0;
for(qtj=0;qtj<=qti;qtj++){
for(plj=0;plj<(qtj<qti?3:pli);plj++){
if(!memcmp(stage,_dequant[qtj][plj],sizeof(stage))){
dupe=1;
break;
}
}
if(dupe)break;
}
if(dupe)_dequant[qti][pli]=_dequant[qtj][plj];
else memcpy(_dequant[qti][pli],stage,sizeof(stage));
}
}
}
}

34
Engine/lib/libtheora/lib/dec/quant.h Normal file
View file

@ -0,0 +1,34 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: quant.h 15400 2008-10-15 12:10:58Z tterribe $
********************************************************************/
#if !defined(_quant_H)
# define _quant_H (1)
# include "theora/codec.h"
# include "ocintrin.h"
typedef ogg_uint16_t oc_quant_table[64];
typedef oc_quant_table oc_quant_tables[64];
/*Maximum scaled quantizer value.*/
#define OC_QUANT_MAX (1024<<2)
void oc_dequant_tables_init(oc_quant_table *_dequant[2][3],
int _pp_dc_scale[64],const th_quant_info *_qinfo);
#endif

1192
Engine/lib/libtheora/lib/dec/state.c Normal file
View file

File diff suppressed because it is too large Load diff

291
Engine/lib/libtheora/lib/dec/x86/mmxfrag.c Normal file
View file

@ -0,0 +1,291 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2003 *
* by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: mmxfrag.c 15400 2008-10-15 12:10:58Z tterribe $
********************************************************************/
/*MMX acceleration of fragment reconstruction for motion compensation.
Originally written by Rudolf Marek.
Additional optimization by Nils Pipenbrinck.
Note: Loops are unrolled for best performance.
The iteration each instruction belongs to is marked in the comments as #i.*/
#include "x86int.h"
#include <stddef.h>
#if defined(USE_ASM)
void oc_frag_recon_intra_mmx(unsigned char *_dst,int _dst_ystride,
const ogg_int16_t *_residue){
__asm__ __volatile__(
/*Set mm0 to 0xFFFFFFFFFFFFFFFF.*/
"pcmpeqw %%mm0,%%mm0\n\t"
/*#0 Load low residue.*/
"movq 0*8(%[residue]),%%mm1\n\t"
/*#0 Load high residue.*/
"movq 1*8(%[residue]),%%mm2\n\t"
/*Set mm0 to 0x8000800080008000.*/
"psllw $15,%%mm0\n\t"
/*#1 Load low residue.*/
"movq 2*8(%[residue]),%%mm3\n\t"
/*#1 Load high residue.*/
"movq 3*8(%[residue]),%%mm4\n\t"
/*Set mm0 to 0x0080008000800080.*/
"psrlw $8,%%mm0\n\t"
/*#2 Load low residue.*/
"movq 4*8(%[residue]),%%mm5\n\t"
/*#2 Load high residue.*/
"movq 5*8(%[residue]),%%mm6\n\t"
/*#0 Bias low residue.*/
"paddsw %%mm0,%%mm1\n\t"
/*#0 Bias high residue.*/
"paddsw %%mm0,%%mm2\n\t"
/*#0 Pack to byte.*/
"packuswb %%mm2,%%mm1\n\t"
/*#1 Bias low residue.*/
"paddsw %%mm0,%%mm3\n\t"
/*#1 Bias high residue.*/
"paddsw %%mm0,%%mm4\n\t"
/*#1 Pack to byte.*/
"packuswb %%mm4,%%mm3\n\t"
/*#2 Bias low residue.*/
"paddsw %%mm0,%%mm5\n\t"
/*#2 Bias high residue.*/
"paddsw %%mm0,%%mm6\n\t"
/*#2 Pack to byte.*/
"packuswb %%mm6,%%mm5\n\t"
/*#0 Write row.*/
"movq %%mm1,(%[dst])\n\t"
/*#1 Write row.*/
"movq %%mm3,(%[dst],%[dst_ystride])\n\t"
/*#2 Write row.*/
"movq %%mm5,(%[dst],%[dst_ystride],2)\n\t"
/*#3 Load low residue.*/
"movq 6*8(%[residue]),%%mm1\n\t"
/*#3 Load high residue.*/
"movq 7*8(%[residue]),%%mm2\n\t"
/*#4 Load high residue.*/
"movq 8*8(%[residue]),%%mm3\n\t"
/*#4 Load high residue.*/
"movq 9*8(%[residue]),%%mm4\n\t"
/*#5 Load high residue.*/
"movq 10*8(%[residue]),%%mm5\n\t"
/*#5 Load high residue.*/
"movq 11*8(%[residue]),%%mm6\n\t"
/*#3 Bias low residue.*/
"paddsw %%mm0,%%mm1\n\t"
/*#3 Bias high residue.*/
"paddsw %%mm0,%%mm2\n\t"
/*#3 Pack to byte.*/
"packuswb %%mm2,%%mm1\n\t"
/*#4 Bias low residue.*/
"paddsw %%mm0,%%mm3\n\t"
/*#4 Bias high residue.*/
"paddsw %%mm0,%%mm4\n\t"
/*#4 Pack to byte.*/
"packuswb %%mm4,%%mm3\n\t"
/*#5 Bias low residue.*/
"paddsw %%mm0,%%mm5\n\t"
/*#5 Bias high residue.*/
"paddsw %%mm0,%%mm6\n\t"
/*#5 Pack to byte.*/
"packuswb %%mm6,%%mm5\n\t"
/*#3 Write row.*/
"movq %%mm1,(%[dst],%[dst_ystride3])\n\t"
/*#4 Write row.*/
"movq %%mm3,(%[dst4])\n\t"
/*#5 Write row.*/
"movq %%mm5,(%[dst4],%[dst_ystride])\n\t"
/*#6 Load low residue.*/
"movq 12*8(%[residue]),%%mm1\n\t"
/*#6 Load high residue.*/
"movq 13*8(%[residue]),%%mm2\n\t"
/*#7 Load low residue.*/
"movq 14*8(%[residue]),%%mm3\n\t"
/*#7 Load high residue.*/
"movq 15*8(%[residue]),%%mm4\n\t"
/*#6 Bias low residue.*/
"paddsw %%mm0,%%mm1\n\t"
/*#6 Bias high residue.*/
"paddsw %%mm0,%%mm2\n\t"
/*#6 Pack to byte.*/
"packuswb %%mm2,%%mm1\n\t"
/*#7 Bias low residue.*/
"paddsw %%mm0,%%mm3\n\t"
/*#7 Bias high residue.*/
"paddsw %%mm0,%%mm4\n\t"
/*#7 Pack to byte.*/
"packuswb %%mm4,%%mm3\n\t"
/*#6 Write row.*/
"movq %%mm1,(%[dst4],%[dst_ystride],2)\n\t"
/*#7 Write row.*/
"movq %%mm3,(%[dst4],%[dst_ystride3])\n\t"
:
:[residue]"r"(_residue),
[dst]"r"(_dst),
[dst4]"r"(_dst+(_dst_ystride<<2)),
[dst_ystride]"r"((ptrdiff_t)_dst_ystride),
[dst_ystride3]"r"((ptrdiff_t)_dst_ystride*3)
:"memory"
);
}
void oc_frag_recon_inter_mmx(unsigned char *_dst,int _dst_ystride,
const unsigned char *_src,int _src_ystride,const ogg_int16_t *_residue){
int i;
/*Zero mm0.*/
__asm__ __volatile__("pxor %%mm0,%%mm0\n\t"::);
for(i=4;i-->0;){
__asm__ __volatile__(
/*#0 Load source.*/
"movq (%[src]),%%mm3\n\t"
/*#1 Load source.*/
"movq (%[src],%[src_ystride]),%%mm7\n\t"
/*#0 Get copy of src.*/
"movq %%mm3,%%mm4\n\t"
/*#0 Expand high source.*/
"punpckhbw %%mm0,%%mm4\n\t"
/*#0 Expand low source.*/
"punpcklbw %%mm0,%%mm3\n\t"
/*#0 Add residue high.*/
"paddsw 8(%[residue]),%%mm4\n\t"
/*#1 Get copy of src.*/
"movq %%mm7,%%mm2\n\t"
/*#0 Add residue low.*/
"paddsw (%[residue]), %%mm3\n\t"
/*#1 Expand high source.*/
"punpckhbw %%mm0,%%mm2\n\t"
/*#0 Pack final row pixels.*/
"packuswb %%mm4,%%mm3\n\t"
/*#1 Expand low source.*/
"punpcklbw %%mm0,%%mm7\n\t"
/*#1 Add residue low.*/
"paddsw 16(%[residue]),%%mm7\n\t"
/*#1 Add residue high.*/
"paddsw 24(%[residue]),%%mm2\n\t"
/*Advance residue.*/
"lea 32(%[residue]),%[residue]\n\t"
/*#1 Pack final row pixels.*/
"packuswb %%mm2,%%mm7\n\t"
/*Advance src.*/
"lea (%[src],%[src_ystride],2),%[src]\n\t"
/*#0 Write row.*/
"movq %%mm3,(%[dst])\n\t"
/*#1 Write row.*/
"movq %%mm7,(%[dst],%[dst_ystride])\n\t"
/*Advance dst.*/
"lea (%[dst],%[dst_ystride],2),%[dst]\n\t"
:[residue]"+r"(_residue),[dst]"+r"(_dst),[src]"+r"(_src)
:[dst_ystride]"r"((ptrdiff_t)_dst_ystride),
[src_ystride]"r"((ptrdiff_t)_src_ystride)
:"memory"
);
}
}
void oc_frag_recon_inter2_mmx(unsigned char *_dst,int _dst_ystride,
const unsigned char *_src1,int _src1_ystride,const unsigned char *_src2,
int _src2_ystride,const ogg_int16_t *_residue){
int i;
/*NOTE: This assumes that
_dst_ystride==_src1_ystride&&_dst_ystride==_src2_ystride.
This is currently always the case, but a slower fallback version will need
to be written if it ever is not.*/
/*Zero mm7.*/
__asm__ __volatile__("pxor %%mm7,%%mm7\n\t"::);
for(i=4;i-->0;){
__asm__ __volatile__(
/*#0 Load src1.*/
"movq (%[src1]),%%mm0\n\t"
/*#0 Load src2.*/
"movq (%[src2]),%%mm2\n\t"
/*#0 Copy src1.*/
"movq %%mm0,%%mm1\n\t"
/*#0 Copy src2.*/
"movq %%mm2,%%mm3\n\t"
/*#1 Load src1.*/
"movq (%[src1],%[ystride]),%%mm4\n\t"
/*#0 Unpack lower src1.*/
"punpcklbw %%mm7,%%mm0\n\t"
/*#1 Load src2.*/
"movq (%[src2],%[ystride]),%%mm5\n\t"
/*#0 Unpack higher src1.*/
"punpckhbw %%mm7,%%mm1\n\t"
/*#0 Unpack lower src2.*/
"punpcklbw %%mm7,%%mm2\n\t"
/*#0 Unpack higher src2.*/
"punpckhbw %%mm7,%%mm3\n\t"
/*Advance src1 ptr.*/
"lea (%[src1],%[ystride],2),%[src1]\n\t"
/*Advance src2 ptr.*/
"lea (%[src2],%[ystride],2),%[src2]\n\t"
/*#0 Lower src1+src2.*/
"paddsw %%mm2,%%mm0\n\t"
/*#0 Higher src1+src2.*/
"paddsw %%mm3,%%mm1\n\t"
/*#1 Copy src1.*/
"movq %%mm4,%%mm2\n\t"
/*#0 Build lo average.*/
"psraw $1,%%mm0\n\t"
/*#1 Copy src2.*/
"movq %%mm5,%%mm3\n\t"
/*#1 Unpack lower src1.*/
"punpcklbw %%mm7,%%mm4\n\t"
/*#0 Build hi average.*/
"psraw $1,%%mm1\n\t"
/*#1 Unpack higher src1.*/
"punpckhbw %%mm7,%%mm2\n\t"
/*#0 low+=residue.*/
"paddsw (%[residue]),%%mm0\n\t"
/*#1 Unpack lower src2.*/
"punpcklbw %%mm7,%%mm5\n\t"
/*#0 high+=residue.*/
"paddsw 8(%[residue]),%%mm1\n\t"
/*#1 Unpack higher src2.*/
"punpckhbw %%mm7,%%mm3\n\t"
/*#1 Lower src1+src2.*/
"paddsw %%mm4,%%mm5\n\t"
/*#0 Pack and saturate.*/
"packuswb %%mm1,%%mm0\n\t"
/*#1 Higher src1+src2.*/
"paddsw %%mm2,%%mm3\n\t"
/*#0 Write row.*/
"movq %%mm0,(%[dst])\n\t"
/*#1 Build lo average.*/
"psraw $1,%%mm5\n\t"
/*#1 Build hi average.*/
"psraw $1,%%mm3\n\t"
/*#1 low+=residue.*/
"paddsw 16(%[residue]),%%mm5\n\t"
/*#1 high+=residue.*/
"paddsw 24(%[residue]),%%mm3\n\t"
/*#1 Pack and saturate.*/
"packuswb %%mm3,%%mm5\n\t"
/*#1 Write row ptr.*/
"movq %%mm5,(%[dst],%[ystride])\n\t"
/*Advance residue ptr.*/
"add $32,%[residue]\n\t"
/*Advance dest ptr.*/
"lea (%[dst],%[ystride],2),%[dst]\n\t"
:[dst]"+r"(_dst),[residue]"+r"(_residue),
[src1]"+r"(_src1),[src2]"+r"(_src2)
:[ystride]"r"((ptrdiff_t)_dst_ystride)
:"memory"
);
}
}
void oc_restore_fpu_mmx(void){
__asm__ __volatile__("emms\n\t");
}
#endif

535
Engine/lib/libtheora/lib/dec/x86/mmxidct.c Normal file
View file

@ -0,0 +1,535 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: mmxidct.c 15400 2008-10-15 12:10:58Z tterribe $
********************************************************************/
/*MMX acceleration of Theora's iDCT.
Originally written by Rudolf Marek, based on code from On2's VP3.*/
#include <ogg/ogg.h>
#include "../dct.h"
#include "../idct.h"
#include "x86int.h"
#if defined(USE_ASM)
/*These are offsets into the table of constants below.*/
/*7 rows of cosines, in order: pi/16 * (1 ... 7).*/
#define OC_COSINE_OFFSET (0)
/*A row of 8's.*/
#define OC_EIGHT_OFFSET (56)
/*A table of constants used by the MMX routines.*/
static const ogg_uint16_t __attribute__((aligned(8),used))
OC_IDCT_CONSTS[(7+1)*4]={
(ogg_uint16_t)OC_C1S7,(ogg_uint16_t)OC_C1S7,
(ogg_uint16_t)OC_C1S7,(ogg_uint16_t)OC_C1S7,
(ogg_uint16_t)OC_C2S6,(ogg_uint16_t)OC_C2S6,
(ogg_uint16_t)OC_C2S6,(ogg_uint16_t)OC_C2S6,
(ogg_uint16_t)OC_C3S5,(ogg_uint16_t)OC_C3S5,
(ogg_uint16_t)OC_C3S5,(ogg_uint16_t)OC_C3S5,
(ogg_uint16_t)OC_C4S4,(ogg_uint16_t)OC_C4S4,
(ogg_uint16_t)OC_C4S4,(ogg_uint16_t)OC_C4S4,
(ogg_uint16_t)OC_C5S3,(ogg_uint16_t)OC_C5S3,
(ogg_uint16_t)OC_C5S3,(ogg_uint16_t)OC_C5S3,
(ogg_uint16_t)OC_C6S2,(ogg_uint16_t)OC_C6S2,
(ogg_uint16_t)OC_C6S2,(ogg_uint16_t)OC_C6S2,
(ogg_uint16_t)OC_C7S1,(ogg_uint16_t)OC_C7S1,
(ogg_uint16_t)OC_C7S1,(ogg_uint16_t)OC_C7S1,
8, 8, 8, 8
};
/*Converts the expression in the argument to a string.*/
#define OC_M2STR(_s) #_s
/*38 cycles*/
#define OC_IDCT_BEGIN \
"#OC_IDCT_BEGIN\n\t" \
"movq "OC_I(3)",%%mm2\n\t" \
"movq "OC_C(3)",%%mm6\n\t" \
"movq %%mm2,%%mm4\n\t" \
"movq "OC_J(5)",%%mm7\n\t" \
"pmulhw %%mm6,%%mm4\n\t" \
"movq "OC_C(5)",%%mm1\n\t" \
"pmulhw %%mm7,%%mm6\n\t" \
"movq %%mm1,%%mm5\n\t" \
"pmulhw %%mm2,%%mm1\n\t" \
"movq "OC_I(1)",%%mm3\n\t" \
"pmulhw %%mm7,%%mm5\n\t" \
"movq "OC_C(1)",%%mm0\n\t" \
"paddw %%mm2,%%mm4\n\t" \
"paddw %%mm7,%%mm6\n\t" \
"paddw %%mm1,%%mm2\n\t" \
"movq "OC_J(7)",%%mm1\n\t" \
"paddw %%mm5,%%mm7\n\t" \
"movq %%mm0,%%mm5\n\t" \
"pmulhw %%mm3,%%mm0\n\t" \
"paddw %%mm7,%%mm4\n\t" \
"pmulhw %%mm1,%%mm5\n\t" \
"movq "OC_C(7)",%%mm7\n\t" \
"psubw %%mm2,%%mm6\n\t" \
"paddw %%mm3,%%mm0\n\t" \
"pmulhw %%mm7,%%mm3\n\t" \
"movq "OC_I(2)",%%mm2\n\t" \
"pmulhw %%mm1,%%mm7\n\t" \
"paddw %%mm1,%%mm5\n\t" \
"movq %%mm2,%%mm1\n\t" \
"pmulhw "OC_C(2)",%%mm2\n\t" \
"psubw %%mm5,%%mm3\n\t" \
"movq "OC_J(6)",%%mm5\n\t" \
"paddw %%mm7,%%mm0\n\t" \
"movq %%mm5,%%mm7\n\t" \
"psubw %%mm4,%%mm0\n\t" \
"pmulhw "OC_C(2)",%%mm5\n\t" \
"paddw %%mm1,%%mm2\n\t" \
"pmulhw "OC_C(6)",%%mm1\n\t" \
"paddw %%mm4,%%mm4\n\t" \
"paddw %%mm0,%%mm4\n\t" \
"psubw %%mm6,%%mm3\n\t" \
"paddw %%mm7,%%mm5\n\t" \
"paddw %%mm6,%%mm6\n\t" \
"pmulhw "OC_C(6)",%%mm7\n\t" \
"paddw %%mm3,%%mm6\n\t" \
"movq %%mm4,"OC_I(1)"\n\t" \
"psubw %%mm5,%%mm1\n\t" \
"movq "OC_C(4)",%%mm4\n\t" \
"movq %%mm3,%%mm5\n\t" \
"pmulhw %%mm4,%%mm3\n\t" \
"paddw %%mm2,%%mm7\n\t" \
"movq %%mm6,"OC_I(2)"\n\t" \
"movq %%mm0,%%mm2\n\t" \
"movq "OC_I(0)",%%mm6\n\t" \
"pmulhw %%mm4,%%mm0\n\t" \
"paddw %%mm3,%%mm5\n\t" \
"movq "OC_J(4)",%%mm3\n\t" \
"psubw %%mm1,%%mm5\n\t" \
"paddw %%mm0,%%mm2\n\t" \
"psubw %%mm3,%%mm6\n\t" \
"movq %%mm6,%%mm0\n\t" \
"pmulhw %%mm4,%%mm6\n\t" \
"paddw %%mm3,%%mm3\n\t" \
"paddw %%mm1,%%mm1\n\t" \
"paddw %%mm0,%%mm3\n\t" \
"paddw %%mm5,%%mm1\n\t" \
"pmulhw %%mm3,%%mm4\n\t" \
"paddw %%mm0,%%mm6\n\t" \
"psubw %%mm2,%%mm6\n\t" \
"paddw %%mm2,%%mm2\n\t" \
"movq "OC_I(1)",%%mm0\n\t" \
"paddw %%mm6,%%mm2\n\t" \
"paddw %%mm3,%%mm4\n\t" \
"psubw %%mm1,%%mm2\n\t" \
"#end OC_IDCT_BEGIN\n\t" \
/*38+8=46 cycles.*/
#define OC_ROW_IDCT \
"#OC_ROW_IDCT\n" \
OC_IDCT_BEGIN \
/*r3=D'*/ \
"movq "OC_I(2)",%%mm3\n\t" \
/*r4=E'=E-G*/ \
"psubw %%mm7,%%mm4\n\t" \
/*r1=H'+H'*/ \
"paddw %%mm1,%%mm1\n\t" \
/*r7=G+G*/ \
"paddw %%mm7,%%mm7\n\t" \
/*r1=R1=A''+H'*/ \
"paddw %%mm2,%%mm1\n\t" \
/*r7=G'=E+G*/ \
"paddw %%mm4,%%mm7\n\t" \
/*r4=R4=E'-D'*/ \
"psubw %%mm3,%%mm4\n\t" \
"paddw %%mm3,%%mm3\n\t" \
/*r6=R6=F'-B''*/ \
"psubw %%mm5,%%mm6\n\t" \
"paddw %%mm5,%%mm5\n\t" \
/*r3=R3=E'+D'*/ \
"paddw %%mm4,%%mm3\n\t" \
/*r5=R5=F'+B''*/ \
"paddw %%mm6,%%mm5\n\t" \
/*r7=R7=G'-C'*/ \
"psubw %%mm0,%%mm7\n\t" \
"paddw %%mm0,%%mm0\n\t" \
/*Save R1.*/ \
"movq %%mm1,"OC_I(1)"\n\t" \
/*r0=R0=G.+C.*/ \
"paddw %%mm7,%%mm0\n\t" \
"#end OC_ROW_IDCT\n\t" \
/*The following macro does two 4x4 transposes in place.
At entry, we assume:
r0 = a3 a2 a1 a0
I(1) = b3 b2 b1 b0
r2 = c3 c2 c1 c0
r3 = d3 d2 d1 d0
r4 = e3 e2 e1 e0
r5 = f3 f2 f1 f0
r6 = g3 g2 g1 g0
r7 = h3 h2 h1 h0
At exit, we have:
I(0) = d0 c0 b0 a0
I(1) = d1 c1 b1 a1
I(2) = d2 c2 b2 a2
I(3) = d3 c3 b3 a3
J(4) = h0 g0 f0 e0
J(5) = h1 g1 f1 e1
J(6) = h2 g2 f2 e2
J(7) = h3 g3 f3 e3
I(0) I(1) I(2) I(3) is the transpose of r0 I(1) r2 r3.
J(4) J(5) J(6) J(7) is the transpose of r4 r5 r6 r7.
Since r1 is free at entry, we calculate the Js first.*/
/*19 cycles.*/
#define OC_TRANSPOSE \
"#OC_TRANSPOSE\n\t" \
"movq %%mm4,%%mm1\n\t" \
"punpcklwd %%mm5,%%mm4\n\t" \
"movq %%mm0,"OC_I(0)"\n\t" \
"punpckhwd %%mm5,%%mm1\n\t" \
"movq %%mm6,%%mm0\n\t" \
"punpcklwd %%mm7,%%mm6\n\t" \
"movq %%mm4,%%mm5\n\t" \
"punpckldq %%mm6,%%mm4\n\t" \
"punpckhdq %%mm6,%%mm5\n\t" \
"movq %%mm1,%%mm6\n\t" \
"movq %%mm4,"OC_J(4)"\n\t" \
"punpckhwd %%mm7,%%mm0\n\t" \
"movq %%mm5,"OC_J(5)"\n\t" \
"punpckhdq %%mm0,%%mm6\n\t" \
"movq "OC_I(0)",%%mm4\n\t" \
"punpckldq %%mm0,%%mm1\n\t" \
"movq "OC_I(1)",%%mm5\n\t" \
"movq %%mm4,%%mm0\n\t" \
"movq %%mm6,"OC_J(7)"\n\t" \
"punpcklwd %%mm5,%%mm0\n\t" \
"movq %%mm1,"OC_J(6)"\n\t" \
"punpckhwd %%mm5,%%mm4\n\t" \
"movq %%mm2,%%mm5\n\t" \
"punpcklwd %%mm3,%%mm2\n\t" \
"movq %%mm0,%%mm1\n\t" \
"punpckldq %%mm2,%%mm0\n\t" \
"punpckhdq %%mm2,%%mm1\n\t" \
"movq %%mm4,%%mm2\n\t" \
"movq %%mm0,"OC_I(0)"\n\t" \
"punpckhwd %%mm3,%%mm5\n\t" \
"movq %%mm1,"OC_I(1)"\n\t" \
"punpckhdq %%mm5,%%mm4\n\t" \
"punpckldq %%mm5,%%mm2\n\t" \
"movq %%mm4,"OC_I(3)"\n\t" \
"movq %%mm2,"OC_I(2)"\n\t" \
"#end OC_TRANSPOSE\n\t" \
/*38+19=57 cycles.*/
#define OC_COLUMN_IDCT \
"#OC_COLUMN_IDCT\n" \
OC_IDCT_BEGIN \
"paddw "OC_8",%%mm2\n\t" \
/*r1=H'+H'*/ \
"paddw %%mm1,%%mm1\n\t" \
/*r1=R1=A''+H'*/ \
"paddw %%mm2,%%mm1\n\t" \
/*r2=NR2*/ \
"psraw $4,%%mm2\n\t" \
/*r4=E'=E-G*/ \
"psubw %%mm7,%%mm4\n\t" \
/*r1=NR1*/ \
"psraw $4,%%mm1\n\t" \
/*r3=D'*/ \
"movq "OC_I(2)",%%mm3\n\t" \
/*r7=G+G*/ \
"paddw %%mm7,%%mm7\n\t" \
/*Store NR2 at I(2).*/ \
"movq %%mm2,"OC_I(2)"\n\t" \
/*r7=G'=E+G*/ \
"paddw %%mm4,%%mm7\n\t" \
/*Store NR1 at I(1).*/ \
"movq %%mm1,"OC_I(1)"\n\t" \
/*r4=R4=E'-D'*/ \
"psubw %%mm3,%%mm4\n\t" \
"paddw "OC_8",%%mm4\n\t" \
/*r3=D'+D'*/ \
"paddw %%mm3,%%mm3\n\t" \
/*r3=R3=E'+D'*/ \
"paddw %%mm4,%%mm3\n\t" \
/*r4=NR4*/ \
"psraw $4,%%mm4\n\t" \
/*r6=R6=F'-B''*/ \
"psubw %%mm5,%%mm6\n\t" \
/*r3=NR3*/ \
"psraw $4,%%mm3\n\t" \
"paddw "OC_8",%%mm6\n\t" \
/*r5=B''+B''*/ \
"paddw %%mm5,%%mm5\n\t" \
/*r5=R5=F'+B''*/ \
"paddw %%mm6,%%mm5\n\t" \
/*r6=NR6*/ \
"psraw $4,%%mm6\n\t" \
/*Store NR4 at J(4).*/ \
"movq %%mm4,"OC_J(4)"\n\t" \
/*r5=NR5*/ \
"psraw $4,%%mm5\n\t" \
/*Store NR3 at I(3).*/ \
"movq %%mm3,"OC_I(3)"\n\t" \
/*r7=R7=G'-C'*/ \
"psubw %%mm0,%%mm7\n\t" \
"paddw "OC_8",%%mm7\n\t" \
/*r0=C'+C'*/ \
"paddw %%mm0,%%mm0\n\t" \
/*r0=R0=G'+C'*/ \
"paddw %%mm7,%%mm0\n\t" \
/*r7=NR7*/ \
"psraw $4,%%mm7\n\t" \
/*Store NR6 at J(6).*/ \
"movq %%mm6,"OC_J(6)"\n\t" \
/*r0=NR0*/ \
"psraw $4,%%mm0\n\t" \
/*Store NR5 at J(5).*/ \
"movq %%mm5,"OC_J(5)"\n\t" \
/*Store NR7 at J(7).*/ \
"movq %%mm7,"OC_J(7)"\n\t" \
/*Store NR0 at I(0).*/ \
"movq %%mm0,"OC_I(0)"\n\t" \
"#end OC_COLUMN_IDCT\n\t" \
#define OC_MID(_m,_i) OC_M2STR(_m+(_i)*8)"(%[c])"
#define OC_C(_i) OC_MID(OC_COSINE_OFFSET,_i-1)
#define OC_8 OC_MID(OC_EIGHT_OFFSET,0)
void oc_idct8x8_mmx(ogg_int16_t _y[64]){
/*This routine accepts an 8x8 matrix, but in transposed form.
Every 4x4 submatrix is transposed.*/
__asm__ __volatile__(
#define OC_I(_k) OC_M2STR((_k*16))"(%[y])"
#define OC_J(_k) OC_M2STR(((_k-4)*16)+8)"(%[y])"
OC_ROW_IDCT
OC_TRANSPOSE
#undef OC_I
#undef OC_J
#define OC_I(_k) OC_M2STR((_k*16)+64)"(%[y])"
#define OC_J(_k) OC_M2STR(((_k-4)*16)+72)"(%[y])"
OC_ROW_IDCT
OC_TRANSPOSE
#undef OC_I
#undef OC_J
#define OC_I(_k) OC_M2STR((_k*16))"(%[y])"
#define OC_J(_k) OC_I(_k)
OC_COLUMN_IDCT
#undef OC_I
#undef OC_J
#define OC_I(_k) OC_M2STR((_k*16)+8)"(%[y])"
#define OC_J(_k) OC_I(_k)
OC_COLUMN_IDCT
#undef OC_I
#undef OC_J
"emms\n\t"
:
:[y]"r"(_y),[c]"r"(OC_IDCT_CONSTS)
);
}
/*25 cycles.*/
#define OC_IDCT_BEGIN_10 \
"#OC_IDCT_BEGIN_10\n\t" \
"movq "OC_I(3)",%%mm2\n\t" \
"nop\n\t" \
"movq "OC_C(3)",%%mm6\n\t" \
"movq %%mm2,%%mm4\n\t" \
"movq "OC_C(5)",%%mm1\n\t" \
"pmulhw %%mm6,%%mm4\n\t" \
"movq "OC_I(1)",%%mm3\n\t" \
"pmulhw %%mm2,%%mm1\n\t" \
"movq "OC_C(1)",%%mm0\n\t" \
"paddw %%mm2,%%mm4\n\t" \
"pxor %%mm6,%%mm6\n\t" \
"paddw %%mm1,%%mm2\n\t" \
"movq "OC_I(2)",%%mm5\n\t" \
"pmulhw %%mm3,%%mm0\n\t" \
"movq %%mm5,%%mm1\n\t" \
"paddw %%mm3,%%mm0\n\t" \
"pmulhw "OC_C(7)",%%mm3\n\t" \
"psubw %%mm2,%%mm6\n\t" \
"pmulhw "OC_C(2)",%%mm5\n\t" \
"psubw %%mm4,%%mm0\n\t" \
"movq "OC_I(2)",%%mm7\n\t" \
"paddw %%mm4,%%mm4\n\t" \
"paddw %%mm5,%%mm7\n\t" \
"paddw %%mm0,%%mm4\n\t" \
"pmulhw "OC_C(6)",%%mm1\n\t" \
"psubw %%mm6,%%mm3\n\t" \
"movq %%mm4,"OC_I(1)"\n\t" \
"paddw %%mm6,%%mm6\n\t" \
"movq "OC_C(4)",%%mm4\n\t" \
"paddw %%mm3,%%mm6\n\t" \
"movq %%mm3,%%mm5\n\t" \
"pmulhw %%mm4,%%mm3\n\t" \
"movq %%mm6,"OC_I(2)"\n\t" \
"movq %%mm0,%%mm2\n\t" \
"movq "OC_I(0)",%%mm6\n\t" \
"pmulhw %%mm4,%%mm0\n\t" \
"paddw %%mm3,%%mm5\n\t" \
"paddw %%mm0,%%mm2\n\t" \
"psubw %%mm1,%%mm5\n\t" \
"pmulhw %%mm4,%%mm6\n\t" \
"paddw "OC_I(0)",%%mm6\n\t" \
"paddw %%mm1,%%mm1\n\t" \
"movq %%mm6,%%mm4\n\t" \
"paddw %%mm5,%%mm1\n\t" \
"psubw %%mm2,%%mm6\n\t" \
"paddw %%mm2,%%mm2\n\t" \
"movq "OC_I(1)",%%mm0\n\t" \
"paddw %%mm6,%%mm2\n\t" \
"psubw %%mm1,%%mm2\n\t" \
"nop\n\t" \
"#end OC_IDCT_BEGIN_10\n\t" \
/*25+8=33 cycles.*/
#define OC_ROW_IDCT_10 \
"#OC_ROW_IDCT_10\n\t" \
OC_IDCT_BEGIN_10 \
/*r3=D'*/ \
"movq "OC_I(2)",%%mm3\n\t" \
/*r4=E'=E-G*/ \
"psubw %%mm7,%%mm4\n\t" \
/*r1=H'+H'*/ \
"paddw %%mm1,%%mm1\n\t" \
/*r7=G+G*/ \
"paddw %%mm7,%%mm7\n\t" \
/*r1=R1=A''+H'*/ \
"paddw %%mm2,%%mm1\n\t" \
/*r7=G'=E+G*/ \
"paddw %%mm4,%%mm7\n\t" \
/*r4=R4=E'-D'*/ \
"psubw %%mm3,%%mm4\n\t" \
"paddw %%mm3,%%mm3\n\t" \
/*r6=R6=F'-B''*/ \
"psubw %%mm5,%%mm6\n\t" \
"paddw %%mm5,%%mm5\n\t" \
/*r3=R3=E'+D'*/ \
"paddw %%mm4,%%mm3\n\t" \
/*r5=R5=F'+B''*/ \
"paddw %%mm6,%%mm5\n\t" \
/*r7=R7=G'-C'*/ \
"psubw %%mm0,%%mm7\n\t" \
"paddw %%mm0,%%mm0\n\t" \
/*Save R1.*/ \
"movq %%mm1,"OC_I(1)"\n\t" \
/*r0=R0=G'+C'*/ \
"paddw %%mm7,%%mm0\n\t" \
"#end OC_ROW_IDCT_10\n\t" \
/*25+19=44 cycles'*/
#define OC_COLUMN_IDCT_10 \
"#OC_COLUMN_IDCT_10\n\t" \
OC_IDCT_BEGIN_10 \
"paddw "OC_8",%%mm2\n\t" \
/*r1=H'+H'*/ \
"paddw %%mm1,%%mm1\n\t" \
/*r1=R1=A''+H'*/ \
"paddw %%mm2,%%mm1\n\t" \
/*r2=NR2*/ \
"psraw $4,%%mm2\n\t" \
/*r4=E'=E-G*/ \
"psubw %%mm7,%%mm4\n\t" \
/*r1=NR1*/ \
"psraw $4,%%mm1\n\t" \
/*r3=D'*/ \
"movq "OC_I(2)",%%mm3\n\t" \
/*r7=G+G*/ \
"paddw %%mm7,%%mm7\n\t" \
/*Store NR2 at I(2).*/ \
"movq %%mm2,"OC_I(2)"\n\t" \
/*r7=G'=E+G*/ \
"paddw %%mm4,%%mm7\n\t" \
/*Store NR1 at I(1).*/ \
"movq %%mm1,"OC_I(1)"\n\t" \
/*r4=R4=E'-D'*/ \
"psubw %%mm3,%%mm4\n\t" \
"paddw "OC_8",%%mm4\n\t" \
/*r3=D'+D'*/ \
"paddw %%mm3,%%mm3\n\t" \
/*r3=R3=E'+D'*/ \
"paddw %%mm4,%%mm3\n\t" \
/*r4=NR4*/ \
"psraw $4,%%mm4\n\t" \
/*r6=R6=F'-B''*/ \
"psubw %%mm5,%%mm6\n\t" \
/*r3=NR3*/ \
"psraw $4,%%mm3\n\t" \
"paddw "OC_8",%%mm6\n\t" \
/*r5=B''+B''*/ \
"paddw %%mm5,%%mm5\n\t" \
/*r5=R5=F'+B''*/ \
"paddw %%mm6,%%mm5\n\t" \
/*r6=NR6*/ \
"psraw $4,%%mm6\n\t" \
/*Store NR4 at J(4).*/ \
"movq %%mm4,"OC_J(4)"\n\t" \
/*r5=NR5*/ \
"psraw $4,%%mm5\n\t" \
/*Store NR3 at I(3).*/ \
"movq %%mm3,"OC_I(3)"\n\t" \
/*r7=R7=G'-C'*/ \
"psubw %%mm0,%%mm7\n\t" \
"paddw "OC_8",%%mm7\n\t" \
/*r0=C'+C'*/ \
"paddw %%mm0,%%mm0\n\t" \
/*r0=R0=G'+C'*/ \
"paddw %%mm7,%%mm0\n\t" \
/*r7=NR7*/ \
"psraw $4,%%mm7\n\t" \
/*Store NR6 at J(6).*/ \
"movq %%mm6,"OC_J(6)"\n\t" \
/*r0=NR0*/ \
"psraw $4,%%mm0\n\t" \
/*Store NR5 at J(5).*/ \
"movq %%mm5,"OC_J(5)"\n\t" \
/*Store NR7 at J(7).*/ \
"movq %%mm7,"OC_J(7)"\n\t" \
/*Store NR0 at I(0).*/ \
"movq %%mm0,"OC_I(0)"\n\t" \
"#end OC_COLUMN_IDCT_10\n\t" \
void oc_idct8x8_10_mmx(ogg_int16_t _y[64]){
__asm__ __volatile__(
#define OC_I(_k) OC_M2STR((_k*16))"(%[y])"
#define OC_J(_k) OC_M2STR(((_k-4)*16)+8)"(%[y])"
/*Done with dequant, descramble, and partial transpose.
Now do the iDCT itself.*/
OC_ROW_IDCT_10
OC_TRANSPOSE
#undef OC_I
#undef OC_J
#define OC_I(_k) OC_M2STR((_k*16))"(%[y])"
#define OC_J(_k) OC_I(_k)
OC_COLUMN_IDCT_10
#undef OC_I
#undef OC_J
#define OC_I(_k) OC_M2STR((_k*16)+8)"(%[y])"
#define OC_J(_k) OC_I(_k)
OC_COLUMN_IDCT_10
#undef OC_I
#undef OC_J
"emms\n\t"
:
:[y]"r"(_y),[c]"r"(OC_IDCT_CONSTS)
);
}
#endif

653
Engine/lib/libtheora/lib/dec/x86/mmxstate.c Normal file
View file

@ -0,0 +1,653 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: mmxstate.c 15400 2008-10-15 12:10:58Z tterribe $
********************************************************************/
/*MMX acceleration of complete fragment reconstruction algorithm.
Originally written by Rudolf Marek.*/
#include "x86int.h"
#include "../../internal.h"
#include <stddef.h>
#if defined(USE_ASM)
static const __attribute__((aligned(8),used)) int OC_FZIG_ZAGMMX[64]={
0, 8, 1, 2, 9,16,24,17,
10, 3,32,11,18,25, 4,12,
5,26,19,40,33,34,41,48,
27, 6,13,20,28,21,14, 7,
56,49,42,35,43,50,57,36,
15,22,29,30,23,44,37,58,
51,59,38,45,52,31,60,53,
46,39,47,54,61,62,55,63
};
void oc_state_frag_recon_mmx(oc_theora_state *_state,oc_fragment *_frag,
int _pli,ogg_int16_t _dct_coeffs[128],int _last_zzi,int _ncoefs,
ogg_uint16_t _dc_iquant,const ogg_uint16_t _ac_iquant[64]){
ogg_int16_t __attribute__((aligned(8))) res_buf[64];
int dst_framei;
int dst_ystride;
int zzi;
/*_last_zzi is subtly different from an actual count of the number of
coefficients we decoded for this block.
It contains the value of zzi BEFORE the final token in the block was
decoded.
In most cases this is an EOB token (the continuation of an EOB run from a
previous block counts), and so this is the same as the coefficient count.
However, in the case that the last token was NOT an EOB token, but filled
the block up with exactly 64 coefficients, _last_zzi will be less than 64.
Provided the last token was not a pure zero run, the minimum value it can
be is 46, and so that doesn't affect any of the cases in this routine.
However, if the last token WAS a pure zero run of length 63, then _last_zzi
will be 1 while the number of coefficients decoded is 64.
Thus, we will trigger the following special case, where the real
coefficient count would not.
Note also that a zero run of length 64 will give _last_zzi a value of 0,
but we still process the DC coefficient, which might have a non-zero value
due to DC prediction.
Although convoluted, this is arguably the correct behavior: it allows us to
dequantize fewer coefficients and use a smaller transform when the block
ends with a long zero run instead of a normal EOB token.
It could be smarter... multiple separate zero runs at the end of a block
will fool it, but an encoder that generates these really deserves what it
gets.
Needless to say we inherited this approach from VP3.*/
/*Special case only having a DC component.*/
if(_last_zzi<2){
ogg_uint16_t p;
/*Why is the iquant product rounded in this case and no others?
Who knows.*/
p=(ogg_int16_t)((ogg_int32_t)_frag->dc*_dc_iquant+15>>5);
/*Fill res_buf with p.*/
__asm__ __volatile__(
/*mm0=0000 0000 0000 AAAA*/
"movd %[p],%%mm0\n\t"
/*mm1=0000 0000 0000 AAAA*/
"movd %[p],%%mm1\n\t"
/*mm0=0000 0000 AAAA 0000*/
"pslld $16,%%mm0\n\t"
/*mm0=0000 0000 AAAA AAAA*/
"por %%mm1,%%mm0\n\t"
/*mm0=AAAA AAAA AAAA AAAA*/
"punpcklwd %%mm0,%%mm0\n\t"
"movq %%mm0,(%[res_buf])\n\t"
"movq %%mm0,8(%[res_buf])\n\t"
"movq %%mm0,16(%[res_buf])\n\t"
"movq %%mm0,24(%[res_buf])\n\t"
"movq %%mm0,32(%[res_buf])\n\t"
"movq %%mm0,40(%[res_buf])\n\t"
"movq %%mm0,48(%[res_buf])\n\t"
"movq %%mm0,56(%[res_buf])\n\t"
"movq %%mm0,64(%[res_buf])\n\t"
"movq %%mm0,72(%[res_buf])\n\t"
"movq %%mm0,80(%[res_buf])\n\t"
"movq %%mm0,88(%[res_buf])\n\t"
"movq %%mm0,96(%[res_buf])\n\t"
"movq %%mm0,104(%[res_buf])\n\t"
"movq %%mm0,112(%[res_buf])\n\t"
"movq %%mm0,120(%[res_buf])\n\t"
:
:[res_buf]"r"(res_buf),[p]"r"((unsigned)p)
:"memory"
);
}
else{
/*Then, fill in the remainder of the coefficients with 0's, and perform
the iDCT.*/
/*First zero the buffer.*/
/*On K7, etc., this could be replaced with movntq and sfence.*/
__asm__ __volatile__(
"pxor %%mm0,%%mm0\n\t"
"movq %%mm0,(%[res_buf])\n\t"
"movq %%mm0,8(%[res_buf])\n\t"
"movq %%mm0,16(%[res_buf])\n\t"
"movq %%mm0,24(%[res_buf])\n\t"
"movq %%mm0,32(%[res_buf])\n\t"
"movq %%mm0,40(%[res_buf])\n\t"
"movq %%mm0,48(%[res_buf])\n\t"
"movq %%mm0,56(%[res_buf])\n\t"
"movq %%mm0,64(%[res_buf])\n\t"
"movq %%mm0,72(%[res_buf])\n\t"
"movq %%mm0,80(%[res_buf])\n\t"
"movq %%mm0,88(%[res_buf])\n\t"
"movq %%mm0,96(%[res_buf])\n\t"
"movq %%mm0,104(%[res_buf])\n\t"
"movq %%mm0,112(%[res_buf])\n\t"
"movq %%mm0,120(%[res_buf])\n\t"
:
:[res_buf]"r"(res_buf)
:"memory"
);
res_buf[0]=(ogg_int16_t)((ogg_int32_t)_frag->dc*_dc_iquant);
/*This is planned to be rewritten in MMX.*/
for(zzi=1;zzi<_ncoefs;zzi++){
int ci;
ci=OC_FZIG_ZAG[zzi];
res_buf[OC_FZIG_ZAGMMX[zzi]]=(ogg_int16_t)((ogg_int32_t)_dct_coeffs[zzi]*
_ac_iquant[ci]);
}
if(_last_zzi<10)oc_idct8x8_10_mmx(res_buf);
else oc_idct8x8_mmx(res_buf);
}
/*Fill in the target buffer.*/
dst_framei=_state->ref_frame_idx[OC_FRAME_SELF];
dst_ystride=_state->ref_frame_bufs[dst_framei][_pli].stride;
/*For now ystride values in all ref frames assumed to be equal.*/
if(_frag->mbmode==OC_MODE_INTRA){
oc_frag_recon_intra_mmx(_frag->buffer[dst_framei],dst_ystride,res_buf);
}
else{
int ref_framei;
int ref_ystride;
int mvoffsets[2];
ref_framei=_state->ref_frame_idx[OC_FRAME_FOR_MODE[_frag->mbmode]];
ref_ystride=_state->ref_frame_bufs[ref_framei][_pli].stride;
if(oc_state_get_mv_offsets(_state,mvoffsets,_frag->mv[0],_frag->mv[1],
ref_ystride,_pli)>1){
oc_frag_recon_inter2_mmx(_frag->buffer[dst_framei],dst_ystride,
_frag->buffer[ref_framei]+mvoffsets[0],ref_ystride,
_frag->buffer[ref_framei]+mvoffsets[1],ref_ystride,res_buf);
}
else{
oc_frag_recon_inter_mmx(_frag->buffer[dst_framei],dst_ystride,
_frag->buffer[ref_framei]+mvoffsets[0],ref_ystride,res_buf);
}
}
oc_restore_fpu(_state);
}
/*Copies the fragments specified by the lists of fragment indices from one
frame to another.
_fragis: A pointer to a list of fragment indices.
_nfragis: The number of fragment indices to copy.
_dst_frame: The reference frame to copy to.
_src_frame: The reference frame to copy from.
_pli: The color plane the fragments lie in.*/
void oc_state_frag_copy_mmx(const oc_theora_state *_state,const int *_fragis,
int _nfragis,int _dst_frame,int _src_frame,int _pli){
const int *fragi;
const int *fragi_end;
int dst_framei;
ptrdiff_t dst_ystride;
int src_framei;
ptrdiff_t src_ystride;
dst_framei=_state->ref_frame_idx[_dst_frame];
src_framei=_state->ref_frame_idx[_src_frame];
dst_ystride=_state->ref_frame_bufs[dst_framei][_pli].stride;
src_ystride=_state->ref_frame_bufs[src_framei][_pli].stride;
fragi_end=_fragis+_nfragis;
for(fragi=_fragis;fragi<fragi_end;fragi++){
oc_fragment *frag;
unsigned char *dst;
unsigned char *src;
ptrdiff_t s;
frag=_state->frags+*fragi;
dst=frag->buffer[dst_framei];
src=frag->buffer[src_framei];
__asm__ __volatile__(
/*src+0*src_ystride*/
"movq (%[src]),%%mm0\n\t"
/*s=src_ystride*3*/
"lea (%[src_ystride],%[src_ystride],2),%[s]\n\t"
/*src+1*src_ystride*/
"movq (%[src],%[src_ystride]),%%mm1\n\t"
/*src+2*src_ystride*/
"movq (%[src],%[src_ystride],2),%%mm2\n\t"
/*src+3*src_ystride*/
"movq (%[src],%[s]),%%mm3\n\t"
/*dst+0*dst_ystride*/
"movq %%mm0,(%[dst])\n\t"
/*s=dst_ystride*3*/
"lea (%[dst_ystride],%[dst_ystride],2),%[s]\n\t"
/*dst+1*dst_ystride*/
"movq %%mm1,(%[dst],%[dst_ystride])\n\t"
/*Pointer to next 4.*/
"lea (%[src],%[src_ystride],4),%[src]\n\t"
/*dst+2*dst_ystride*/
"movq %%mm2,(%[dst],%[dst_ystride],2)\n\t"
/*dst+3*dst_ystride*/
"movq %%mm3,(%[dst],%[s])\n\t"
/*Pointer to next 4.*/
"lea (%[dst],%[dst_ystride],4),%[dst]\n\t"
/*src+0*src_ystride*/
"movq (%[src]),%%mm0\n\t"
/*s=src_ystride*3*/
"lea (%[src_ystride],%[src_ystride],2),%[s]\n\t"
/*src+1*src_ystride*/
"movq (%[src],%[src_ystride]),%%mm1\n\t"
/*src+2*src_ystride*/
"movq (%[src],%[src_ystride],2),%%mm2\n\t"
/*src+3*src_ystride*/
"movq (%[src],%[s]),%%mm3\n\t"
/*dst+0*dst_ystride*/
"movq %%mm0,(%[dst])\n\t"
/*s=dst_ystride*3*/
"lea (%[dst_ystride],%[dst_ystride],2),%[s]\n\t"
/*dst+1*dst_ystride*/
"movq %%mm1,(%[dst],%[dst_ystride])\n\t"
/*dst+2*dst_ystride*/
"movq %%mm2,(%[dst],%[dst_ystride],2)\n\t"
/*dst+3*dst_ystride*/
"movq %%mm3,(%[dst],%[s])\n\t"
:[s]"=&r"(s)
:[dst]"r"(dst),[src]"r"(src),[dst_ystride]"r"(dst_ystride),
[src_ystride]"r"(src_ystride)
:"memory"
);
}
/*This needs to be removed when decode specific functions are implemented:*/
__asm__ __volatile__("emms\n\t");
}
static void loop_filter_v(unsigned char *_pix,int _ystride,
const ogg_int16_t *_ll){
ptrdiff_t s;
_pix-=_ystride*2;
__asm__ __volatile__(
/*mm0=0*/
"pxor %%mm0,%%mm0\n\t"
/*s=_ystride*3*/
"lea (%[ystride],%[ystride],2),%[s]\n\t"
/*mm7=_pix[0...8]*/
"movq (%[pix]),%%mm7\n\t"
/*mm4=_pix[0...8+_ystride*3]*/
"movq (%[pix],%[s]),%%mm4\n\t"
/*mm6=_pix[0...8]*/
"movq %%mm7,%%mm6\n\t"
/*Expand unsigned _pix[0...3] to 16 bits.*/
"punpcklbw %%mm0,%%mm6\n\t"
"movq %%mm4,%%mm5\n\t"
/*Expand unsigned _pix[4...8] to 16 bits.*/
"punpckhbw %%mm0,%%mm7\n\t"
/*Expand other arrays too.*/
"punpcklbw %%mm0,%%mm4\n\t"
"punpckhbw %%mm0,%%mm5\n\t"
/*mm7:mm6=_p[0...8]-_p[0...8+_ystride*3]:*/
"psubw %%mm4,%%mm6\n\t"
"psubw %%mm5,%%mm7\n\t"
/*mm5=mm4=_pix[0...8+_ystride]*/
"movq (%[pix],%[ystride]),%%mm4\n\t"
/*mm1=mm3=mm2=_pix[0..8]+_ystride*2]*/
"movq (%[pix],%[ystride],2),%%mm2\n\t"
"movq %%mm4,%%mm5\n\t"
"movq %%mm2,%%mm3\n\t"
"movq %%mm2,%%mm1\n\t"
/*Expand these arrays.*/
"punpckhbw %%mm0,%%mm5\n\t"
"punpcklbw %%mm0,%%mm4\n\t"
"punpckhbw %%mm0,%%mm3\n\t"
"punpcklbw %%mm0,%%mm2\n\t"
/*mm0=3 3 3 3
mm3:mm2=_pix[0...8+_ystride*2]-_pix[0...8+_ystride]*/
"pcmpeqw %%mm0,%%mm0\n\t"
"psubw %%mm5,%%mm3\n\t"
"psrlw $14,%%mm0\n\t"
"psubw %%mm4,%%mm2\n\t"
/*Scale by 3.*/
"pmullw %%mm0,%%mm3\n\t"
"pmullw %%mm0,%%mm2\n\t"
/*mm0=4 4 4 4
f=mm3:mm2==_pix[0...8]-_pix[0...8+_ystride*3]+
3*(_pix[0...8+_ystride*2]-_pix[0...8+_ystride])*/
"psrlw $1,%%mm0\n\t"
"paddw %%mm7,%%mm3\n\t"
"psllw $2,%%mm0\n\t"
"paddw %%mm6,%%mm2\n\t"
/*Add 4.*/
"paddw %%mm0,%%mm3\n\t"
"paddw %%mm0,%%mm2\n\t"
/*"Divide" by 8.*/
"psraw $3,%%mm3\n\t"
"psraw $3,%%mm2\n\t"
/*Now compute lflim of mm3:mm2 cf. Section 7.10 of the sepc.*/
/*Free up mm5.*/
"packuswb %%mm5,%%mm4\n\t"
/*mm0=L L L L*/
"movq (%[ll]),%%mm0\n\t"
/*if(R_i<-2L||R_i>2L)R_i=0:*/
"movq %%mm2,%%mm5\n\t"
"pxor %%mm6,%%mm6\n\t"
"movq %%mm0,%%mm7\n\t"
"psubw %%mm0,%%mm6\n\t"
"psllw $1,%%mm7\n\t"
"psllw $1,%%mm6\n\t"
/*mm2==R_3 R_2 R_1 R_0*/
/*mm5==R_3 R_2 R_1 R_0*/
/*mm6==-2L -2L -2L -2L*/
/*mm7==2L 2L 2L 2L*/
"pcmpgtw %%mm2,%%mm7\n\t"
"pcmpgtw %%mm6,%%mm5\n\t"
"pand %%mm7,%%mm2\n\t"
"movq %%mm0,%%mm7\n\t"
"pand %%mm5,%%mm2\n\t"
"psllw $1,%%mm7\n\t"
"movq %%mm3,%%mm5\n\t"
/*mm3==R_7 R_6 R_5 R_4*/
/*mm5==R_7 R_6 R_5 R_4*/
/*mm6==-2L -2L -2L -2L*/
/*mm7==2L 2L 2L 2L*/
"pcmpgtw %%mm3,%%mm7\n\t"
"pcmpgtw %%mm6,%%mm5\n\t"
"pand %%mm7,%%mm3\n\t"
"movq %%mm0,%%mm7\n\t"
"pand %%mm5,%%mm3\n\t"
/*if(R_i<-L)R_i'=R_i+2L;
if(R_i>L)R_i'=R_i-2L;
if(R_i<-L||R_i>L)R_i=-R_i':*/
"psraw $1,%%mm6\n\t"
"movq %%mm2,%%mm5\n\t"
"psllw $1,%%mm7\n\t"
/*mm2==R_3 R_2 R_1 R_0*/
/*mm5==R_3 R_2 R_1 R_0*/
/*mm6==-L -L -L -L*/
/*mm0==L L L L*/
/*mm5=R_i>L?FF:00*/
"pcmpgtw %%mm0,%%mm5\n\t"
/*mm6=-L>R_i?FF:00*/
"pcmpgtw %%mm2,%%mm6\n\t"
/*mm7=R_i>L?2L:0*/
"pand %%mm5,%%mm7\n\t"
/*mm2=R_i>L?R_i-2L:R_i*/
"psubw %%mm7,%%mm2\n\t"
"movq %%mm0,%%mm7\n\t"
/*mm5=-L>R_i||R_i>L*/
"por %%mm6,%%mm5\n\t"
"psllw $1,%%mm7\n\t"
/*mm7=-L>R_i?2L:0*/
"pand %%mm6,%%mm7\n\t"
"pxor %%mm6,%%mm6\n\t"
/*mm2=-L>R_i?R_i+2L:R_i*/
"paddw %%mm7,%%mm2\n\t"
"psubw %%mm0,%%mm6\n\t"
/*mm5=-L>R_i||R_i>L?-R_i':0*/
"pand %%mm2,%%mm5\n\t"
"movq %%mm0,%%mm7\n\t"
/*mm2=-L>R_i||R_i>L?0:R_i*/
"psubw %%mm5,%%mm2\n\t"
"psllw $1,%%mm7\n\t"
/*mm2=-L>R_i||R_i>L?-R_i':R_i*/
"psubw %%mm5,%%mm2\n\t"
"movq %%mm3,%%mm5\n\t"
/*mm3==R_7 R_6 R_5 R_4*/
/*mm5==R_7 R_6 R_5 R_4*/
/*mm6==-L -L -L -L*/
/*mm0==L L L L*/
/*mm6=-L>R_i?FF:00*/
"pcmpgtw %%mm3,%%mm6\n\t"
/*mm5=R_i>L?FF:00*/
"pcmpgtw %%mm0,%%mm5\n\t"
/*mm7=R_i>L?2L:0*/
"pand %%mm5,%%mm7\n\t"
/*mm2=R_i>L?R_i-2L:R_i*/
"psubw %%mm7,%%mm3\n\t"
"psllw $1,%%mm0\n\t"
/*mm5=-L>R_i||R_i>L*/
"por %%mm6,%%mm5\n\t"
/*mm0=-L>R_i?2L:0*/
"pand %%mm6,%%mm0\n\t"
/*mm3=-L>R_i?R_i+2L:R_i*/
"paddw %%mm0,%%mm3\n\t"
/*mm5=-L>R_i||R_i>L?-R_i':0*/
"pand %%mm3,%%mm5\n\t"
/*mm2=-L>R_i||R_i>L?0:R_i*/
"psubw %%mm5,%%mm3\n\t"
/*mm2=-L>R_i||R_i>L?-R_i':R_i*/
"psubw %%mm5,%%mm3\n\t"
/*Unfortunately, there's no unsigned byte+signed byte with unsigned
saturation op code, so we have to promote things back 16 bits.*/
"pxor %%mm0,%%mm0\n\t"
"movq %%mm4,%%mm5\n\t"
"punpcklbw %%mm0,%%mm4\n\t"
"punpckhbw %%mm0,%%mm5\n\t"
"movq %%mm1,%%mm6\n\t"
"punpcklbw %%mm0,%%mm1\n\t"
"punpckhbw %%mm0,%%mm6\n\t"
/*_pix[0...8+_ystride]+=R_i*/
"paddw %%mm2,%%mm4\n\t"
"paddw %%mm3,%%mm5\n\t"
/*_pix[0...8+_ystride*2]-=R_i*/
"psubw %%mm2,%%mm1\n\t"
"psubw %%mm3,%%mm6\n\t"
"packuswb %%mm5,%%mm4\n\t"
"packuswb %%mm6,%%mm1\n\t"
/*Write it back out.*/
"movq %%mm4,(%[pix],%[ystride])\n\t"
"movq %%mm1,(%[pix],%[ystride],2)\n\t"
:[s]"=&r"(s)
:[pix]"r"(_pix),[ystride]"r"((ptrdiff_t)_ystride),[ll]"r"(_ll)
:"memory"
);
}
/*This code implements the bulk of loop_filter_h().
Data are striped p0 p1 p2 p3 ... p0 p1 p2 p3 ..., so in order to load all
four p0's to one register we must transpose the values in four mmx regs.
When half is done we repeat this for the rest.*/
static void loop_filter_h4(unsigned char *_pix,ptrdiff_t _ystride,
const ogg_int16_t *_ll){
ptrdiff_t s;
/*d doesn't technically need to be 64-bit on x86-64, but making it so will
help avoid partial register stalls.*/
ptrdiff_t d;
__asm__ __volatile__(
/*x x x x 3 2 1 0*/
"movd (%[pix]),%%mm0\n\t"
/*s=_ystride*3*/
"lea (%[ystride],%[ystride],2),%[s]\n\t"
/*x x x x 7 6 5 4*/
"movd (%[pix],%[ystride]),%%mm1\n\t"
/*x x x x B A 9 8*/
"movd (%[pix],%[ystride],2),%%mm2\n\t"
/*x x x x F E D C*/
"movd (%[pix],%[s]),%%mm3\n\t"
/*mm0=7 3 6 2 5 1 4 0*/
"punpcklbw %%mm1,%%mm0\n\t"
/*mm2=F B E A D 9 C 8*/
"punpcklbw %%mm3,%%mm2\n\t"
/*mm1=7 3 6 2 5 1 4 0*/
"movq %%mm0,%%mm1\n\t"
/*mm0=F B 7 3 E A 6 2*/
"punpckhwd %%mm2,%%mm0\n\t"
/*mm1=D 9 5 1 C 8 4 0*/
"punpcklwd %%mm2,%%mm1\n\t"
"pxor %%mm7,%%mm7\n\t"
/*mm5=D 9 5 1 C 8 4 0*/
"movq %%mm1,%%mm5\n\t"
/*mm1=x C x 8 x 4 x 0==pix[0]*/
"punpcklbw %%mm7,%%mm1\n\t"
/*mm5=x D x 9 x 5 x 1==pix[1]*/
"punpckhbw %%mm7,%%mm5\n\t"
/*mm3=F B 7 3 E A 6 2*/
"movq %%mm0,%%mm3\n\t"
/*mm0=x E x A x 6 x 2==pix[2]*/
"punpcklbw %%mm7,%%mm0\n\t"
/*mm3=x F x B x 7 x 3==pix[3]*/
"punpckhbw %%mm7,%%mm3\n\t"
/*mm1=mm1-mm3==pix[0]-pix[3]*/
"psubw %%mm3,%%mm1\n\t"
/*Save a copy of pix[2] for later.*/
"movq %%mm0,%%mm4\n\t"
/*mm2=3 3 3 3
mm0=mm0-mm5==pix[2]-pix[1]*/
"pcmpeqw %%mm2,%%mm2\n\t"
"psubw %%mm5,%%mm0\n\t"
"psrlw $14,%%mm2\n\t"
/*Scale by 3.*/
"pmullw %%mm2,%%mm0\n\t"
/*mm2=4 4 4 4
f=mm1==_pix[0]-_pix[3]+ 3*(_pix[2]-_pix[1])*/
"psrlw $1,%%mm2\n\t"
"paddw %%mm1,%%mm0\n\t"
"psllw $2,%%mm2\n\t"
/*Add 4.*/
"paddw %%mm2,%%mm0\n\t"
/*"Divide" by 8, producing the residuals R_i.*/
"psraw $3,%%mm0\n\t"
/*Now compute lflim of mm0 cf. Section 7.10 of the sepc.*/
/*mm6=L L L L*/
"movq (%[ll]),%%mm6\n\t"
/*if(R_i<-2L||R_i>2L)R_i=0:*/
"movq %%mm0,%%mm1\n\t"
"pxor %%mm2,%%mm2\n\t"
"movq %%mm6,%%mm3\n\t"
"psubw %%mm6,%%mm2\n\t"
"psllw $1,%%mm3\n\t"
"psllw $1,%%mm2\n\t"
/*mm0==R_3 R_2 R_1 R_0*/
/*mm1==R_3 R_2 R_1 R_0*/
/*mm2==-2L -2L -2L -2L*/
/*mm3==2L 2L 2L 2L*/
"pcmpgtw %%mm0,%%mm3\n\t"
"pcmpgtw %%mm2,%%mm1\n\t"
"pand %%mm3,%%mm0\n\t"
"pand %%mm1,%%mm0\n\t"
/*if(R_i<-L)R_i'=R_i+2L;
if(R_i>L)R_i'=R_i-2L;
if(R_i<-L||R_i>L)R_i=-R_i':*/
"psraw $1,%%mm2\n\t"
"movq %%mm0,%%mm1\n\t"
"movq %%mm6,%%mm3\n\t"
/*mm0==R_3 R_2 R_1 R_0*/
/*mm1==R_3 R_2 R_1 R_0*/
/*mm2==-L -L -L -L*/
/*mm6==L L L L*/
/*mm2=-L>R_i?FF:00*/
"pcmpgtw %%mm0,%%mm2\n\t"
/*mm1=R_i>L?FF:00*/
"pcmpgtw %%mm6,%%mm1\n\t"
/*mm3=2L 2L 2L 2L*/
"psllw $1,%%mm3\n\t"
/*mm6=2L 2L 2L 2L*/
"psllw $1,%%mm6\n\t"
/*mm3=R_i>L?2L:0*/
"pand %%mm1,%%mm3\n\t"
/*mm6=-L>R_i?2L:0*/
"pand %%mm2,%%mm6\n\t"
/*mm0=R_i>L?R_i-2L:R_i*/
"psubw %%mm3,%%mm0\n\t"
/*mm1=-L>R_i||R_i>L*/
"por %%mm2,%%mm1\n\t"
/*mm0=-L>R_i?R_i+2L:R_i*/
"paddw %%mm6,%%mm0\n\t"
/*mm1=-L>R_i||R_i>L?R_i':0*/
"pand %%mm0,%%mm1\n\t"
/*mm0=-L>R_i||R_i>L?0:R_i*/
"psubw %%mm1,%%mm0\n\t"
/*mm0=-L>R_i||R_i>L?-R_i':R_i*/
"psubw %%mm1,%%mm0\n\t"
/*_pix[1]+=R_i;*/
"paddw %%mm0,%%mm5\n\t"
/*_pix[2]-=R_i;*/
"psubw %%mm0,%%mm4\n\t"
/*mm5=x x x x D 9 5 1*/
"packuswb %%mm7,%%mm5\n\t"
/*mm4=x x x x E A 6 2*/
"packuswb %%mm7,%%mm4\n\t"
/*mm5=E D A 9 6 5 2 1*/
"punpcklbw %%mm4,%%mm5\n\t"
/*d=6 5 2 1*/
"movd %%mm5,%[d]\n\t"
"movw %w[d],1(%[pix])\n\t"
/*Why is there such a big stall here?*/
"psrlq $32,%%mm5\n\t"
"shr $16,%[d]\n\t"
"movw %w[d],1(%[pix],%[ystride])\n\t"
/*d=E D A 9*/
"movd %%mm5,%[d]\n\t"
"movw %w[d],1(%[pix],%[ystride],2)\n\t"
"shr $16,%[d]\n\t"
"movw %w[d],1(%[pix],%[s])\n\t"
:[s]"=&r"(s),[d]"=&r"(d),
[pix]"+r"(_pix),[ystride]"+r"(_ystride),[ll]"+r"(_ll)
:
:"memory"
);
}
static void loop_filter_h(unsigned char *_pix,int _ystride,
const ogg_int16_t *_ll){
_pix-=2;
loop_filter_h4(_pix,_ystride,_ll);
loop_filter_h4(_pix+(_ystride<<2),_ystride,_ll);
}
/*We copy the whole function because the MMX routines will be inlined 4 times,
and we can do just a single emms call at the end this way.
We also do not use the _bv lookup table, instead computing the values that
would lie in it on the fly.*/
/*Apply the loop filter to a given set of fragment rows in the given plane.
The filter may be run on the bottom edge, affecting pixels in the next row of
fragments, so this row also needs to be available.
_bv: The bounding values array.
_refi: The index of the frame buffer to filter.
_pli: The color plane to filter.
_fragy0: The Y coordinate of the first fragment row to filter.
_fragy_end: The Y coordinate of the fragment row to stop filtering at.*/
void oc_state_loop_filter_frag_rows_mmx(oc_theora_state *_state,int *_bv,
int _refi,int _pli,int _fragy0,int _fragy_end){
ogg_int16_t __attribute__((aligned(8))) ll[4];
th_img_plane *iplane;
oc_fragment_plane *fplane;
oc_fragment *frag_top;
oc_fragment *frag0;
oc_fragment *frag;
oc_fragment *frag_end;
oc_fragment *frag0_end;
oc_fragment *frag_bot;
ll[0]=ll[1]=ll[2]=ll[3]=
(ogg_int16_t)_state->loop_filter_limits[_state->qis[0]];
iplane=_state->ref_frame_bufs[_refi]+_pli;
fplane=_state->fplanes+_pli;
/*The following loops are constructed somewhat non-intuitively on purpose.
The main idea is: if a block boundary has at least one coded fragment on
it, the filter is applied to it.
However, the order that the filters are applied in matters, and VP3 chose
the somewhat strange ordering used below.*/
frag_top=_state->frags+fplane->froffset;
frag0=frag_top+_fragy0*fplane->nhfrags;
frag0_end=frag0+(_fragy_end-_fragy0)*fplane->nhfrags;
frag_bot=_state->frags+fplane->froffset+fplane->nfrags;
while(frag0<frag0_end){
frag=frag0;
frag_end=frag+fplane->nhfrags;
while(frag<frag_end){
if(frag->coded){
if(frag>frag0){
loop_filter_h(frag->buffer[_refi],iplane->stride,ll);
}
if(frag0>frag_top){
loop_filter_v(frag->buffer[_refi],iplane->stride,ll);
}
if(frag+1<frag_end&&!(frag+1)->coded){
loop_filter_h(frag->buffer[_refi]+8,iplane->stride,ll);
}
if(frag+fplane->nhfrags<frag_bot&&!(frag+fplane->nhfrags)->coded){
loop_filter_v((frag+fplane->nhfrags)->buffer[_refi],
iplane->stride,ll);
}
}
frag++;
}
frag0+=fplane->nhfrags;
}
/*This needs to be removed when decode specific functions are implemented:*/
__asm__ __volatile__("emms\n\t");
}
#endif

42
Engine/lib/libtheora/lib/dec/x86/x86int.h Normal file
View file

@ -0,0 +1,42 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: x86int.h 15400 2008-10-15 12:10:58Z tterribe $
********************************************************************/
#if !defined(_x86_x86int_H)
# define _x86_x86int_H (1)
# include "../../internal.h"
void oc_state_vtable_init_x86(oc_theora_state *_state);
void oc_frag_recon_intra_mmx(unsigned char *_dst,int _dst_ystride,
const ogg_int16_t *_residue);
void oc_frag_recon_inter_mmx(unsigned char *_dst,int _dst_ystride,
const unsigned char *_src,int _src_ystride,const ogg_int16_t *_residue);
void oc_frag_recon_inter2_mmx(unsigned char *_dst,int _dst_ystride,
const unsigned char *_src1,int _src1_ystride,const unsigned char *_src2,
int _src2_ystride,const ogg_int16_t *_residue);
void oc_state_frag_copy_mmx(const oc_theora_state *_state,const int *_fragis,
int _nfragis,int _dst_frame,int _src_frame,int _pli);
void oc_state_frag_recon_mmx(oc_theora_state *_state,oc_fragment *_frag,
int _pli,ogg_int16_t _dct_coeffs[128],int _last_zzi,int _ncoefs,
ogg_uint16_t _dc_iquant,const ogg_uint16_t _ac_iquant[64]);
void oc_restore_fpu_mmx(void);
void oc_idct8x8_mmx(ogg_int16_t _y[64]);
void oc_idct8x8_10_mmx(ogg_int16_t _y[64]);
void oc_fill_idct_constants_mmx(void);
void oc_state_loop_filter_frag_rows_mmx(oc_theora_state *_state,int *_bv,
int _refi,int _pli,int _fragy0,int _fragy_end);
#endif

38
Engine/lib/libtheora/lib/dec/x86/x86state.c Normal file
View file

@ -0,0 +1,38 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: x86state.c 15427 2008-10-21 02:36:19Z xiphmont $
********************************************************************/
#include "x86int.h"
#if defined(USE_ASM)
#include "../../cpu.c"
void oc_state_vtable_init_x86(oc_theora_state *_state){
_state->cpu_flags=oc_cpu_flags_get();
if(_state->cpu_flags&OC_CPU_X86_MMX){
_state->opt_vtable.frag_recon_intra=oc_frag_recon_intra_mmx;
_state->opt_vtable.frag_recon_inter=oc_frag_recon_inter_mmx;
_state->opt_vtable.frag_recon_inter2=oc_frag_recon_inter2_mmx;
_state->opt_vtable.state_frag_copy=oc_state_frag_copy_mmx;
_state->opt_vtable.state_frag_recon=oc_state_frag_recon_mmx;
_state->opt_vtable.state_loop_filter_frag_rows=
oc_state_loop_filter_frag_rows_mmx;
_state->opt_vtable.restore_fpu=oc_restore_fpu_mmx;
}
else oc_state_vtable_init_c(_state);
}
#endif

214
Engine/lib/libtheora/lib/dec/x86_vc/mmxfrag.c Normal file
View file

@ -0,0 +1,214 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id:
********************************************************************/
#include "../../internal.h"
/* ------------------------------------------------------------------------
MMX reconstruction fragment routines for Visual Studio.
Tested with VS2005. Should compile for VS2003 and VC6 as well.
Initial implementation 2007 by Nils Pipenbrinck.
---------------------------------------------------------------------*/
#if defined(USE_ASM)
void oc_frag_recon_intra_mmx(unsigned char *_dst,int _dst_ystride,
const ogg_int16_t *_residue){
/* ---------------------------------------------------------------------
This function does the inter reconstruction step with 8 iterations
unrolled. The iteration for each instruction is noted by the #id in the
comments (in case you want to reconstruct it)
--------------------------------------------------------------------- */
_asm{
mov edi, [_residue] /* load residue ptr */
mov eax, 0x00800080 /* generate constant */
mov ebx, [_dst_ystride] /* load dst-stride */
mov edx, [_dst] /* load dest pointer */
/* unrolled loop begins here */
movd mm0, eax /* load constant */
movq mm1, [edi+ 8*0] /* #1 load low residue */
movq mm2, [edi+ 8*1] /* #1 load high residue */
punpckldq mm0, mm0 /* build constant */
movq mm3, [edi+ 8*2] /* #2 load low residue */
movq mm4, [edi+ 8*3] /* #2 load high residue */
movq mm5, [edi+ 8*4] /* #3 load low residue */
movq mm6, [edi+ 8*5] /* #3 load high residue */
paddsw mm1, mm0 /* #1 bias low residue */
paddsw mm2, mm0 /* #1 bias high residue */
packuswb mm1, mm2 /* #1 pack to byte */
paddsw mm3, mm0 /* #2 bias low residue */
paddsw mm4, mm0 /* #2 bias high residue */
packuswb mm3, mm4 /* #2 pack to byte */
paddsw mm5, mm0 /* #3 bias low residue */
paddsw mm6, mm0 /* #3 bias high residue */
packuswb mm5, mm6 /* #3 pack to byte */
movq [edx], mm1 /* #1 write row */
movq [edx + ebx], mm3 /* #2 write row */
movq [edx + ebx*2], mm5 /* #3 write row */
movq mm1, [edi+ 8*6] /* #4 load low residue */
lea ecx, [ebx + ebx*2] /* make dst_ystride * 3 */
movq mm2, [edi+ 8*7] /* #4 load high residue */
movq mm3, [edi+ 8*8] /* #5 load low residue */
lea esi, [ebx*4 + ebx] /* make dst_ystride * 5 */
movq mm4, [edi+ 8*9] /* #5 load high residue */
movq mm5, [edi+ 8*10] /* #6 load low residue */
lea eax, [ecx*2 + ebx] /* make dst_ystride * 7 */
movq mm6, [edi+ 8*11] /* #6 load high residue */
paddsw mm1, mm0 /* #4 bias low residue */
paddsw mm2, mm0 /* #4 bias high residue */
packuswb mm1, mm2 /* #4 pack to byte */
paddsw mm3, mm0 /* #5 bias low residue */
paddsw mm4, mm0 /* #5 bias high residue */
packuswb mm3, mm4 /* #5 pack to byte */
paddsw mm5, mm0 /* #6 bias low residue */
paddsw mm6, mm0 /* #6 bias high residue */
packuswb mm5, mm6 /* #6 pack to byte */
movq [edx + ecx], mm1 /* #4 write row */
movq [edx + ebx*4], mm3 /* #5 write row */
movq [edx + esi], mm5 /* #6 write row */
movq mm1, [edi+ 8*12] /* #7 load low residue */
movq mm2, [edi+ 8*13] /* #7 load high residue */
movq mm3, [edi+ 8*14] /* #8 load low residue */
movq mm4, [edi+ 8*15] /* #8 load high residue */
paddsw mm1, mm0 /* #7 bias low residue */
paddsw mm2, mm0 /* #7 bias high residue */
packuswb mm1, mm2 /* #7 pack to byte */
paddsw mm3, mm0 /* #8 bias low residue */
paddsw mm4, mm0 /* #8 bias high residue */
packuswb mm3, mm4 /* #8 pack to byte */
movq [edx + ecx*2], mm1 /* #7 write row */
movq [edx + eax], mm3 /* #8 write row */
}
}
void oc_frag_recon_inter_mmx (unsigned char *_dst, int _dst_ystride,
const unsigned char *_src, int _src_ystride, const ogg_int16_t *_residue){
/* ---------------------------------------------------------------------
This function does the inter reconstruction step with two iterations
running in parallel to hide some load-latencies and break the dependency
chains. The iteration for each instruction is noted by the #id in the
comments (in case you want to reconstruct it)
--------------------------------------------------------------------- */
_asm{
pxor mm0, mm0 /* generate constant 0 */
mov esi, [_src]
mov edi, [_residue]
mov eax, [_src_ystride]
mov edx, [_dst]
mov ebx, [_dst_ystride]
mov ecx, 4
align 16
nextchunk:
movq mm3, [esi] /* #1 load source */
movq mm1, [edi+0] /* #1 load residium low */
movq mm2, [edi+8] /* #1 load residium high */
movq mm7, [esi+eax] /* #2 load source */
movq mm4, mm3 /* #1 get copy of src */
movq mm5, [edi+16] /* #2 load residium low */
punpckhbw mm4, mm0 /* #1 expand high source */
movq mm6, [edi+24] /* #2 load residium high */
punpcklbw mm3, mm0 /* #1 expand low source */
paddsw mm4, mm2 /* #1 add residium high */
movq mm2, mm7 /* #2 get copy of src */
paddsw mm3, mm1 /* #1 add residium low */
punpckhbw mm2, mm0 /* #2 expand high source */
packuswb mm3, mm4 /* #1 final row pixels */
punpcklbw mm7, mm0 /* #2 expand low source */
movq [edx], mm3 /* #1 write row */
paddsw mm2, mm6 /* #2 add residium high */
add edi, 32 /* residue += 4 */
paddsw mm7, mm5 /* #2 add residium low */
sub ecx, 1 /* update loop counter */
packuswb mm7, mm2 /* #2 final row */
lea esi, [esi+eax*2] /* src += stride * 2 */
movq [edx + ebx], mm7 /* #2 write row */
lea edx, [edx+ebx*2] /* dst += stride * 2 */
jne nextchunk
}
}
void oc_frag_recon_inter2_mmx(unsigned char *_dst, int _dst_ystride,
const unsigned char *_src1, int _src1_ystride, const unsigned char *_src2,
int _src2_ystride,const ogg_int16_t *_residue){
/* ---------------------------------------------------------------------
This function does the inter2 reconstruction step.The building of the
average is done with a bit-twiddeling trick to avoid excessive register
copy work during byte to word conversion.
average = (a & b) + (((a ^ b) & 0xfe) >> 1);
(shown for a single byte; it's done with 8 of them at a time)
Slightly faster than the obvious method using add and shift, but not
earthshaking improvement either.
If anyone comes up with a way that produces bit-identical outputs
using the pavgb instruction let me know and I'll do the 3dnow codepath.
--------------------------------------------------------------------- */
_asm{
mov eax, 0xfefefefe
mov esi, [_src1]
mov edi, [_src2]
movd mm1, eax
mov ebx, [_residue]
mov edx, [_dst]
mov eax, [_dst_ystride]
punpckldq mm1, mm1 /* replicate lsb32 */
mov ecx, 8 /* init loop counter */
pxor mm0, mm0 /* constant zero */
sub edx, eax /* dst -= dst_stride */
align 16
nextrow:
movq mm2, [esi] /* load source1 */
movq mm3, [edi] /* load source2 */
movq mm5, [ebx + 0] /* load lower residue */
movq mm6, [ebx + 8] /* load higer residue */
add esi, _src1_ystride /* src1 += src1_stride */
add edi, _src2_ystride /* src2 += src1_stride */
movq mm4, mm2 /* get copy of source1 */
pand mm2, mm3 /* s1 & s2 (avg part) */
pxor mm3, mm4 /* s1 ^ s2 (avg part) */
add ebx, 16 /* residue++ */
pand mm3, mm1 /* mask out low bits */
psrlq mm3, 1 /* shift xor avg-part */
paddd mm3, mm2 /* build final average */
add edx, eax /* dst += dst_stride */
movq mm2, mm3 /* get copy of average */
punpckhbw mm3, mm0 /* average high */
punpcklbw mm2, mm0 /* average low */
paddsw mm3, mm6 /* high + residue */
paddsw mm2, mm5 /* low + residue */
sub ecx, 1 /* update loop counter */
packuswb mm2, mm3 /* pack and saturate */
movq [edx], mm2 /* write row */
jne nextrow
}
}
void oc_restore_fpu_mmx(void){
_asm { emms }
}
#endif

1006
Engine/lib/libtheora/lib/dec/x86_vc/mmxidct.c Normal file
View file

File diff suppressed because it is too large Load diff

377
Engine/lib/libtheora/lib/dec/x86_vc/mmxloopfilter.c Normal file
View file

@ -0,0 +1,377 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id:
********************************************************************/
/* -------------------------------------------------------------------
MMX based loop filter for the theora codec.
Originally written by Rudolf Marek, based on code from On2's VP3.
Converted to Visual Studio inline assembly by Nils Pipenbrinck.
Note: I can't test these since my example files never get into the
loop filters, but the code has been converted semi-automatic from
the GCC sources, so it ought to work.
---------------------------------------------------------------------*/
#include "../../internal.h"
#include "x86int.h"
#include <mmintrin.h>
#if defined(USE_ASM)
static void loop_filter_v(unsigned char *_pix,int _ystride,
const ogg_int16_t *_ll){
_asm {
mov eax, [_pix]
mov edx, [_ystride]
mov ebx, [_ll]
/* _pix -= ystride */
sub eax, edx
/* mm0=0 */
pxor mm0, mm0
/* _pix -= ystride */
sub eax, edx
/* esi=_ystride*3 */
lea esi, [edx + edx*2]
/* mm7=_pix[0...8]*/
movq mm7, [eax]
/* mm4=_pix[0...8+_ystride*3]*/
movq mm4, [eax + esi]
/* mm6=_pix[0...8]*/
movq mm6, mm7
/* Expand unsigned _pix[0...3] to 16 bits.*/
punpcklbw mm6, mm0
movq mm5, mm4
/* Expand unsigned _pix[4...7] to 16 bits.*/
punpckhbw mm7, mm0
punpcklbw mm4, mm0
/* Expand other arrays too.*/
punpckhbw mm5, mm0
/*mm7:mm6=_p[0...7]-_p[0...7+_ystride*3]:*/
psubw mm6, mm4
psubw mm7, mm5
/*mm5=mm4=_pix[0...7+_ystride]*/
movq mm4, [eax + edx]
/*mm1=mm3=mm2=_pix[0..7]+_ystride*2]*/
movq mm2, [eax + edx*2]
movq mm5, mm4
movq mm3, mm2
movq mm1, mm2
/*Expand these arrays.*/
punpckhbw mm5, mm0
punpcklbw mm4, mm0
punpckhbw mm3, mm0
punpcklbw mm2, mm0
pcmpeqw mm0, mm0
/*mm0=3 3 3 3
mm3:mm2=_pix[0...8+_ystride*2]-_pix[0...8+_ystride]*/
psubw mm3, mm5
psrlw mm0, 14
psubw mm2, mm4
/*Scale by 3.*/
pmullw mm3, mm0
pmullw mm2, mm0
/*mm0=4 4 4 4
f=mm3:mm2==_pix[0...8]-_pix[0...8+_ystride*3]+
3*(_pix[0...8+_ystride*2]-_pix[0...8+_ystride])*/
psrlw mm0, 1
paddw mm3, mm7
psllw mm0, 2
paddw mm2, mm6
/*Add 4.*/
paddw mm3, mm0
paddw mm2, mm0
/*"Divide" by 8.*/
psraw mm3, 3
psraw mm2, 3
/*Now compute lflim of mm3:mm2 cf. Section 7.10 of the sepc.*/
/*Free up mm5.*/
packuswb mm4, mm5
/*mm0=L L L L*/
movq mm0, [ebx]
/*if(R_i<-2L||R_i>2L)R_i=0:*/
movq mm5, mm2
pxor mm6, mm6
movq mm7, mm0
psubw mm6, mm0
psllw mm7, 1
psllw mm6, 1
/*mm2==R_3 R_2 R_1 R_0*/
/*mm5==R_3 R_2 R_1 R_0*/
/*mm6==-2L -2L -2L -2L*/
/*mm7==2L 2L 2L 2L*/
pcmpgtw mm7, mm2
pcmpgtw mm5, mm6
pand mm2, mm7
movq mm7, mm0
pand mm2, mm5
psllw mm7, 1
movq mm5, mm3
/*mm3==R_7 R_6 R_5 R_4*/
/*mm5==R_7 R_6 R_5 R_4*/
/*mm6==-2L -2L -2L -2L*/
/*mm7==2L 2L 2L 2L*/
pcmpgtw mm7, mm3
pcmpgtw mm5, mm6
pand mm3, mm7
movq mm7, mm0
pand mm3, mm5
/*if(R_i<-L)R_i'=R_i+2L;
if(R_i>L)R_i'=R_i-2L;
if(R_i<-L||R_i>L)R_i=-R_i':*/
psraw mm6, 1
movq mm5, mm2
psllw mm7, 1
/*mm2==R_3 R_2 R_1 R_0*/
/*mm5==R_3 R_2 R_1 R_0*/
/*mm6==-L -L -L -L*/
/*mm0==L L L L*/
/*mm5=R_i>L?FF:00*/
pcmpgtw mm5, mm0
/*mm6=-L>R_i?FF:00*/
pcmpgtw mm6, mm2
/*mm7=R_i>L?2L:0*/
pand mm7, mm5
/*mm2=R_i>L?R_i-2L:R_i*/
psubw mm2, mm7
movq mm7, mm0
/*mm5=-L>R_i||R_i>L*/
por mm5, mm6
psllw mm7, 1
/*mm7=-L>R_i?2L:0*/
pand mm7, mm6
pxor mm6, mm6
/*mm2=-L>R_i?R_i+2L:R_i*/
paddw mm2, mm7
psubw mm6, mm0
/*mm5=-L>R_i||R_i>L?-R_i':0*/
pand mm5, mm2
movq mm7, mm0
/*mm2=-L>R_i||R_i>L?0:R_i*/
psubw mm2, mm5
psllw mm7, 1
/*mm2=-L>R_i||R_i>L?-R_i':R_i*/
psubw mm2, mm5
movq mm5, mm3
/*mm3==R_7 R_6 R_5 R_4*/
/*mm5==R_7 R_6 R_5 R_4*/
/*mm6==-L -L -L -L*/
/*mm0==L L L L*/
/*mm6=-L>R_i?FF:00*/
pcmpgtw mm6, mm3
/*mm5=R_i>L?FF:00*/
pcmpgtw mm5, mm0
/*mm7=R_i>L?2L:0*/
pand mm7, mm5
/*mm2=R_i>L?R_i-2L:R_i*/
psubw mm3, mm7
psllw mm0, 1
/*mm5=-L>R_i||R_i>L*/
por mm5, mm6
/*mm0=-L>R_i?2L:0*/
pand mm0, mm6
/*mm3=-L>R_i?R_i+2L:R_i*/
paddw mm3, mm0
/*mm5=-L>R_i||R_i>L?-R_i':0*/
pand mm5, mm3
/*mm2=-L>R_i||R_i>L?0:R_i*/
psubw mm3, mm5
/*mm3=-L>R_i||R_i>L?-R_i':R_i*/
psubw mm3, mm5
/*Unfortunately, there's no unsigned byte+signed byte with unsigned
saturation op code, so we have to promote things back 16 bits.*/
pxor mm0, mm0
movq mm5, mm4
punpcklbw mm4, mm0
punpckhbw mm5, mm0
movq mm6, mm1
punpcklbw mm1, mm0
punpckhbw mm6, mm0
/*_pix[0...8+_ystride]+=R_i*/
paddw mm4, mm2
paddw mm5, mm3
/*_pix[0...8+_ystride*2]-=R_i*/
psubw mm1, mm2
psubw mm6, mm3
packuswb mm4, mm5
packuswb mm1, mm6
/*Write it back out.*/
movq [eax + edx], mm4
movq [eax + edx*2], mm1
}
}
/*This code implements the bulk of loop_filter_h().
Data are striped p0 p1 p2 p3 ... p0 p1 p2 p3 ..., so in order to load all
four p0's to one register we must transpose the values in four mmx regs.
When half is done we repeat this for the rest.*/
static void loop_filter_h4(unsigned char *_pix,long _ystride,
const ogg_int16_t *_ll){
/* todo: merge the comments from the GCC sources */
_asm {
mov ecx, [_pix]
mov edx, [_ystride]
mov eax, [_ll]
/*esi=_ystride*3*/
lea esi, [edx + edx*2]
movd mm0, dword ptr [ecx]
movd mm1, dword ptr [ecx + edx]
movd mm2, dword ptr [ecx + edx*2]
movd mm3, dword ptr [ecx + esi]
punpcklbw mm0, mm1
punpcklbw mm2, mm3
movq mm1, mm0
punpckhwd mm0, mm2
punpcklwd mm1, mm2
pxor mm7, mm7
movq mm5, mm1
punpcklbw mm1, mm7
punpckhbw mm5, mm7
movq mm3, mm0
punpcklbw mm0, mm7
punpckhbw mm3, mm7
psubw mm1, mm3
movq mm4, mm0
pcmpeqw mm2, mm2
psubw mm0, mm5
psrlw mm2, 14
pmullw mm0, mm2
psrlw mm2, 1
paddw mm0, mm1
psllw mm2, 2
paddw mm0, mm2
psraw mm0, 3
movq mm6, qword ptr [eax]
movq mm1, mm0
pxor mm2, mm2
movq mm3, mm6
psubw mm2, mm6
psllw mm3, 1
psllw mm2, 1
pcmpgtw mm3, mm0
pcmpgtw mm1, mm2
pand mm0, mm3
pand mm0, mm1
psraw mm2, 1
movq mm1, mm0
movq mm3, mm6
pcmpgtw mm2, mm0
pcmpgtw mm1, mm6
psllw mm3, 1
psllw mm6, 1
pand mm3, mm1
pand mm6, mm2
psubw mm0, mm3
por mm1, mm2
paddw mm0, mm6
pand mm1, mm0
psubw mm0, mm1
psubw mm0, mm1
paddw mm5, mm0
psubw mm4, mm0
packuswb mm5, mm7
packuswb mm4, mm7
punpcklbw mm5, mm4
movd edi, mm5
mov word ptr [ecx + 01H], di
psrlq mm5, 32
shr edi, 16
mov word ptr [ecx + edx + 01H], di
movd edi, mm5
mov word ptr [ecx + edx*2 + 01H], di
shr edi, 16
mov word ptr [ecx + esi + 01H], di
}
}
static void loop_filter_h(unsigned char *_pix,int _ystride,
const ogg_int16_t *_ll){
_pix-=2;
loop_filter_h4(_pix,_ystride,_ll);
loop_filter_h4(_pix+(_ystride<<2),_ystride,_ll);
}
/*We copy the whole function because the MMX routines will be inlined 4 times,
and we can do just a single emms call at the end this way.
We also do not use the _bv lookup table, instead computing the values that
would lie in it on the fly.*/
/*Apply the loop filter to a given set of fragment rows in the given plane.
The filter may be run on the bottom edge, affecting pixels in the next row of
fragments, so this row also needs to be available.
_bv: The bounding values array.
_refi: The index of the frame buffer to filter.
_pli: The color plane to filter.
_fragy0: The Y coordinate of the first fragment row to filter.
_fragy_end: The Y coordinate of the fragment row to stop filtering at.*/
void oc_state_loop_filter_frag_rows_mmx(oc_theora_state *_state,int *_bv,
int _refi,int _pli,int _fragy0,int _fragy_end){
ogg_int16_t __declspec(align(8)) ll[4];
th_img_plane *iplane;
oc_fragment_plane *fplane;
oc_fragment *frag_top;
oc_fragment *frag0;
oc_fragment *frag;
oc_fragment *frag_end;
oc_fragment *frag0_end;
oc_fragment *frag_bot;
ll[0]=ll[1]=ll[2]=ll[3]=
(ogg_int16_t)_state->loop_filter_limits[_state->qis[0]];
iplane=_state->ref_frame_bufs[_refi]+_pli;
fplane=_state->fplanes+_pli;
/*The following loops are constructed somewhat non-intuitively on purpose.
The main idea is: if a block boundary has at least one coded fragment on
it, the filter is applied to it.
However, the order that the filters are applied in matters, and VP3 chose
the somewhat strange ordering used below.*/
frag_top=_state->frags+fplane->froffset;
frag0=frag_top+_fragy0*fplane->nhfrags;
frag0_end=frag0+(_fragy_end-_fragy0)*fplane->nhfrags;
frag_bot=_state->frags+fplane->froffset+fplane->nfrags;
while(frag0<frag0_end){
frag=frag0;
frag_end=frag+fplane->nhfrags;
while(frag<frag_end){
if(frag->coded){
if(frag>frag0){
loop_filter_h(frag->buffer[_refi],iplane->stride,ll);
}
if(frag0>frag_top){
loop_filter_v(frag->buffer[_refi],iplane->stride,ll);
}
if(frag+1<frag_end&&!(frag+1)->coded){
loop_filter_h(frag->buffer[_refi]+8,iplane->stride,ll);
}
if(frag+fplane->nhfrags<frag_bot&&!(frag+fplane->nhfrags)->coded){
loop_filter_v((frag+fplane->nhfrags)->buffer[_refi],
iplane->stride,ll);
}
}
frag++;
}
frag0+=fplane->nhfrags;
}
/*This needs to be removed when decode specific functions are implemented:*/
_mm_empty();
}
#endif

189
Engine/lib/libtheora/lib/dec/x86_vc/mmxstate.c Normal file
View file

@ -0,0 +1,189 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2008 *
* by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: mmxstate.c 15400 2008-10-15 12:10:58Z tterribe $
********************************************************************/
/* ------------------------------------------------------------------------
MMX acceleration of complete fragment reconstruction algorithm.
Originally written by Rudolf Marek.
Conversion to MSC intrinsics by Nils Pipenbrinck.
---------------------------------------------------------------------*/
#if defined(USE_ASM)
#include "../../internal.h"
#include "../idct.h"
#include "x86int.h"
#include <mmintrin.h>
static const unsigned char OC_FZIG_ZAGMMX[64]=
{
0, 8, 1, 2, 9,16,24,17,
10, 3,32,11,18,25, 4,12,
5,26,19,40,33,34,41,48,
27, 6,13,20,28,21,14, 7,
56,49,42,35,43,50,57,36,
15,22,29,30,23,44,37,58,
51,59,38,45,52,31,60,53,
46,39,47,54,61,62,55,63
};
/* Fill a block with value */
static __inline void loc_fill_mmx_value (__m64 * _dst, __m64 _value){
__m64 t = _value;
_dst[0] = t; _dst[1] = t; _dst[2] = t; _dst[3] = t;
_dst[4] = t; _dst[5] = t; _dst[6] = t; _dst[7] = t;
_dst[8] = t; _dst[9] = t; _dst[10] = t; _dst[11] = t;
_dst[12] = t; _dst[13] = t; _dst[14] = t; _dst[15] = t;
}
/* copy a block of 8 byte elements using different strides */
static __inline void loc_blockcopy_mmx (unsigned char * _dst, int _dst_ystride,
unsigned char * _src, int _src_ystride){
__m64 a,b,c,d,e,f,g,h;
a = *(__m64*)(_src + 0 * _src_ystride);
b = *(__m64*)(_src + 1 * _src_ystride);
c = *(__m64*)(_src + 2 * _src_ystride);
d = *(__m64*)(_src + 3 * _src_ystride);
e = *(__m64*)(_src + 4 * _src_ystride);
f = *(__m64*)(_src + 5 * _src_ystride);
g = *(__m64*)(_src + 6 * _src_ystride);
h = *(__m64*)(_src + 7 * _src_ystride);
*(__m64*)(_dst + 0 * _dst_ystride) = a;
*(__m64*)(_dst + 1 * _dst_ystride) = b;
*(__m64*)(_dst + 2 * _dst_ystride) = c;
*(__m64*)(_dst + 3 * _dst_ystride) = d;
*(__m64*)(_dst + 4 * _dst_ystride) = e;
*(__m64*)(_dst + 5 * _dst_ystride) = f;
*(__m64*)(_dst + 6 * _dst_ystride) = g;
*(__m64*)(_dst + 7 * _dst_ystride) = h;
}
void oc_state_frag_recon_mmx(oc_theora_state *_state,const oc_fragment *_frag,
int _pli,ogg_int16_t _dct_coeffs[128],int _last_zzi,int _ncoefs,
ogg_uint16_t _dc_iquant,const ogg_uint16_t _ac_iquant[64]){
ogg_int16_t __declspec(align(16)) res_buf[64];
int dst_framei;
int dst_ystride;
int zzi;
/*_last_zzi is subtly different from an actual count of the number of
coefficients we decoded for this block.
It contains the value of zzi BEFORE the final token in the block was
decoded.
In most cases this is an EOB token (the continuation of an EOB run from a
previous block counts), and so this is the same as the coefficient count.
However, in the case that the last token was NOT an EOB token, but filled
the block up with exactly 64 coefficients, _last_zzi will be less than 64.
Provided the last token was not a pure zero run, the minimum value it can
be is 46, and so that doesn't affect any of the cases in this routine.
However, if the last token WAS a pure zero run of length 63, then _last_zzi
will be 1 while the number of coefficients decoded is 64.
Thus, we will trigger the following special case, where the real
coefficient count would not.
Note also that a zero run of length 64 will give _last_zzi a value of 0,
but we still process the DC coefficient, which might have a non-zero value
due to DC prediction.
Although convoluted, this is arguably the correct behavior: it allows us to
dequantize fewer coefficients and use a smaller transform when the block
ends with a long zero run instead of a normal EOB token.
It could be smarter... multiple separate zero runs at the end of a block
will fool it, but an encoder that generates these really deserves what it
gets.
Needless to say we inherited this approach from VP3.*/
/*Special case only having a DC component.*/
if(_last_zzi<2){
__m64 p;
/*Why is the iquant product rounded in this case and no others? Who knows.*/
p = _m_from_int((ogg_int32_t)_frag->dc*_dc_iquant+15>>5);
/* broadcast 16 bits into all 4 mmx subregisters */
p = _m_punpcklwd (p,p);
p = _m_punpckldq (p,p);
loc_fill_mmx_value ((__m64 *)res_buf, p);
}
else{
/*Then, fill in the remainder of the coefficients with 0's, and perform
the iDCT.*/
/*First zero the buffer.*/
/*On K7, etc., this could be replaced with movntq and sfence.*/
loc_fill_mmx_value ((__m64 *)res_buf, _mm_setzero_si64());
res_buf[0]=(ogg_int16_t)((ogg_int32_t)_frag->dc*_dc_iquant);
/*This is planned to be rewritten in MMX.*/
for(zzi=1;zzi<_ncoefs;zzi++)
{
int ci;
ci=OC_FZIG_ZAG[zzi];
res_buf[OC_FZIG_ZAGMMX[zzi]]=(ogg_int16_t)((ogg_int32_t)_dct_coeffs[zzi]*
_ac_iquant[ci]);
}
if(_last_zzi<10){
oc_idct8x8_10_mmx(res_buf);
}
else {
oc_idct8x8_mmx(res_buf);
}
}
/*Fill in the target buffer.*/
dst_framei=_state->ref_frame_idx[OC_FRAME_SELF];
dst_ystride=_state->ref_frame_bufs[dst_framei][_pli].stride;
/*For now ystride values in all ref frames assumed to be equal.*/
if(_frag->mbmode==OC_MODE_INTRA){
oc_frag_recon_intra_mmx(_frag->buffer[dst_framei],dst_ystride,res_buf);
}
else{
int ref_framei;
int ref_ystride;
int mvoffsets[2];
ref_framei=_state->ref_frame_idx[OC_FRAME_FOR_MODE[_frag->mbmode]];
ref_ystride=_state->ref_frame_bufs[ref_framei][_pli].stride;
if(oc_state_get_mv_offsets(_state,mvoffsets,_frag->mv[0],
_frag->mv[1],ref_ystride,_pli)>1){
oc_frag_recon_inter2_mmx(_frag->buffer[dst_framei],dst_ystride,
_frag->buffer[ref_framei]+mvoffsets[0],ref_ystride,
_frag->buffer[ref_framei]+mvoffsets[1],ref_ystride,res_buf);
}
else{
oc_frag_recon_inter_mmx(_frag->buffer[dst_framei],dst_ystride,
_frag->buffer[ref_framei]+mvoffsets[0],ref_ystride,res_buf);
}
}
_mm_empty();
}
void oc_state_frag_copy_mmx(const oc_theora_state *_state,const int *_fragis,
int _nfragis,int _dst_frame,int _src_frame,int _pli){
const int *fragi;
const int *fragi_end;
int dst_framei;
int dst_ystride;
int src_framei;
int src_ystride;
dst_framei=_state->ref_frame_idx[_dst_frame];
src_framei=_state->ref_frame_idx[_src_frame];
dst_ystride=_state->ref_frame_bufs[dst_framei][_pli].stride;
src_ystride=_state->ref_frame_bufs[src_framei][_pli].stride;
fragi_end=_fragis+_nfragis;
for(fragi=_fragis;fragi<fragi_end;fragi++){
oc_fragment *frag = _state->frags+*fragi;
loc_blockcopy_mmx (frag->buffer[dst_framei], dst_ystride,
frag->buffer[src_framei], src_ystride);
}
_m_empty();
}
#endif

49
Engine/lib/libtheora/lib/dec/x86_vc/x86int.h Normal file
View file

@ -0,0 +1,49 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: x86int.h 15400 2008-10-15 12:10:58Z tterribe $
********************************************************************/
#if !defined(_x86_x86int_vc_H)
# define _x86_x86int_vc_H (1)
# include "../../internal.h"
void oc_state_vtable_init_x86(oc_theora_state *_state);
void oc_frag_recon_intra_mmx(unsigned char *_dst,int _dst_ystride,
const ogg_int16_t *_residue);
void oc_frag_recon_inter_mmx(unsigned char *_dst,int _dst_ystride,
const unsigned char *_src,int _src_ystride,const ogg_int16_t *_residue);
void oc_frag_recon_inter2_mmx(unsigned char *_dst,int _dst_ystride,
const unsigned char *_src1,int _src1_ystride,const unsigned char *_src2,
int _src2_ystride,const ogg_int16_t *_residue);
void oc_state_frag_copy_mmx(const oc_theora_state *_state,const int *_fragis,
int _nfragis,int _dst_frame,int _src_frame,int _pli);
void oc_restore_fpu_mmx(void);
void oc_state_frag_recon_mmx(oc_theora_state *_state,const oc_fragment *_frag,
int _pli,ogg_int16_t _dct_coeffs[128],int _last_zzi,int _ncoefs,
ogg_uint16_t _dc_iquant,const ogg_uint16_t _ac_iquant[64]);
void oc_idct8x8_mmx(ogg_int16_t _y[64]);
void oc_idct8x8_10_mmx(ogg_int16_t _y[64]);
void oc_state_loop_filter_frag_rows_mmx(oc_theora_state *_state,int *_bv,
int _refi,int _pli,int _fragy0,int _fragy_end);
#endif

41
Engine/lib/libtheora/lib/dec/x86_vc/x86state.c Normal file
View file

@ -0,0 +1,41 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2008 *
* by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: x86state.c 15427 2008-10-21 02:36:19Z xiphmont $
********************************************************************/
#if defined(USE_ASM)
#include "x86int.h"
#include "../../cpu.c"
void oc_state_vtable_init_x86(oc_theora_state *_state){
_state->cpu_flags=oc_cpu_flags_get();
/* fill with defaults */
oc_state_vtable_init_c(_state);
/* patch MMX functions */
if(_state->cpu_flags&OC_CPU_X86_MMX){
_state->opt_vtable.frag_recon_intra=oc_frag_recon_intra_mmx;
_state->opt_vtable.frag_recon_inter=oc_frag_recon_inter_mmx;
_state->opt_vtable.frag_recon_inter2=oc_frag_recon_inter2_mmx;
_state->opt_vtable.restore_fpu=oc_restore_fpu_mmx;
_state->opt_vtable.state_frag_copy=oc_state_frag_copy_mmx;
_state->opt_vtable.state_frag_recon=oc_state_frag_recon_mmx;
_state->opt_vtable.state_loop_filter_frag_rows=oc_state_loop_filter_frag_rows_mmx;
}
}
#endif

37
Engine/lib/libtheora/lib/enc/block_inline.h Normal file
View file

@ -0,0 +1,37 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: block_inline.h 14059 2007-10-28 23:43:27Z xiphmont $
********************************************************************/
#include "codec_internal.h"
static const ogg_int32_t MBOrderMap[4] = { 0, 2, 3, 1 };
static const ogg_int32_t BlockOrderMap1[4][4] = {
{ 0, 1, 3, 2 },
{ 0, 2, 3, 1 },
{ 0, 2, 3, 1 },
{ 3, 2, 0, 1 }
};
static ogg_int32_t QuadMapToIndex1( ogg_int32_t (*BlockMap)[4][4],
ogg_uint32_t SB, ogg_uint32_t MB,
ogg_uint32_t B ){
return BlockMap[SB][MBOrderMap[MB]][BlockOrderMap1[MB][B]];
}
static ogg_int32_t QuadMapToMBTopLeft( ogg_int32_t (*BlockMap)[4][4],
ogg_uint32_t SB, ogg_uint32_t MB ){
return BlockMap[SB][MBOrderMap[MB]][0];
}

99
Engine/lib/libtheora/lib/enc/blockmap.c Normal file
View file

@ -0,0 +1,99 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: blockmap.c 14059 2007-10-28 23:43:27Z xiphmont $
********************************************************************/
#include "codec_internal.h"
static void CreateMapping ( ogg_int32_t (*BlockMap)[4][4],
ogg_uint32_t FirstSB,
ogg_uint32_t FirstFrag, ogg_uint32_t HFrags,
ogg_uint32_t VFrags ){
ogg_uint32_t i, j = 0;
ogg_uint32_t xpos;
ogg_uint32_t ypos;
ogg_uint32_t SBrow, SBcol;
ogg_uint32_t SBRows, SBCols;
ogg_uint32_t MB, B;
ogg_uint32_t SB=FirstSB;
ogg_uint32_t FragIndex=FirstFrag;
/* Set Super-Block dimensions */
SBRows = VFrags/4 + ( VFrags%4 ? 1 : 0 );
SBCols = HFrags/4 + ( HFrags%4 ? 1 : 0 );
/* Map each Super-Block */
for ( SBrow=0; SBrow<SBRows; SBrow++ ){
for ( SBcol=0; SBcol<SBCols; SBcol++ ){
/* Y co-ordinate of Super-Block in Block units */
ypos = SBrow<<2;
/* Map Blocks within this Super-Block */
for ( i=0; (i<4) && (ypos<VFrags); i++, ypos++ ){
/* X co-ordinate of Super-Block in Block units */
xpos = SBcol<<2;
for ( j=0; (j<4) && (xpos<HFrags); j++, xpos++ ){
if ( i<2 ){
MB = ( j<2 ? 0 : 1 );
}else{
MB = ( j<2 ? 2 : 3 );
}
if ( i%2 ){
B = ( j%2 ? 3 : 2 );
}else{
B = ( j%2 ? 1 : 0 );
}
/* Set mapping and move to next fragment */
BlockMap[SB][MB][B] = FragIndex++;
}
/* Move to first fragment in next row in Super-Block */
FragIndex += HFrags-j;
}
/* Move on to next Super-Block */
SB++;
FragIndex -= i*HFrags-j;
}
/* Move to first Super-Block in next row */
FragIndex += 3*HFrags;
}
}
void CreateBlockMapping ( ogg_int32_t (*BlockMap)[4][4],
ogg_uint32_t YSuperBlocks,
ogg_uint32_t UVSuperBlocks,
ogg_uint32_t HFrags, ogg_uint32_t VFrags ) {
ogg_uint32_t i, j;
for ( i=0; i<YSuperBlocks + UVSuperBlocks * 2; i++ ){
for ( j=0; j<4; j++ ) {
BlockMap[i][j][0] = -1;
BlockMap[i][j][1] = -1;
BlockMap[i][j][2] = -1;
BlockMap[i][j][3] = -1;
}
}
CreateMapping ( BlockMap, 0, 0, HFrags, VFrags );
CreateMapping ( BlockMap, YSuperBlocks, HFrags*VFrags, HFrags/2, VFrags/2 );
CreateMapping ( BlockMap, YSuperBlocks + UVSuperBlocks, (HFrags*VFrags*5)/4,
HFrags/2, VFrags/2 );
}

842
Engine/lib/libtheora/lib/enc/codec_internal.h Normal file
View file

@ -0,0 +1,842 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2005 *
* by the Xiph.Org Foundation http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: codec_internal.h 15153 2008-08-04 18:37:55Z tterribe $
********************************************************************/
#ifndef ENCODER_INTERNAL_H
#define ENCODER_INTERNAL_H
#ifdef HAVE_CONFIG_H
# include "config.h"
#endif
typedef struct PB_INSTANCE PB_INSTANCE;
#include "dsp.h"
#include "theora/theora.h"
#include "encoder_huffman.h"
#define theora_read(x,y,z) ( *z = oggpackB_read(x,y) )
#define CURRENT_ENCODE_VERSION 1
#define HUGE_ERROR (1<<28) /* Out of range test value */
/* Baseline dct height and width. */
#define BLOCK_HEIGHT_WIDTH 8
#define HFRAGPIXELS 8
#define VFRAGPIXELS 8
/* Blocks on INTRA/INTER Y/U/V planes */
enum BlockMode {
BLOCK_Y,
BLOCK_U,
BLOCK_V,
BLOCK_INTER_Y,
BLOCK_INTER_U,
BLOCK_INTER_V
};
/* Baseline dct block size */
#define BLOCK_SIZE (BLOCK_HEIGHT_WIDTH * BLOCK_HEIGHT_WIDTH)
/* Border is for unrestricted mv's */
#define UMV_BORDER 16
#define STRIDE_EXTRA (UMV_BORDER * 2)
#define Q_TABLE_SIZE 64
#define KEY_FRAME 0
#define DELTA_FRAME 1
#define MAX_MODES 8
#define MODE_BITS 3
#define MODE_METHODS 8
#define MODE_METHOD_BITS 3
/* Different key frame types/methods */
#define DCT_KEY_FRAME 0
#define KEY_FRAME_CONTEXT 5
/* Preprocessor defines */
#define MAX_PREV_FRAMES 16
/* Number of search sites for a 4-step search (at pixel accuracy) */
#define MAX_SEARCH_SITES 33
#define VERY_BEST_Q 10
#define MIN_BPB_FACTOR 0.3
#define MAX_BPB_FACTOR 3.0
#define MAX_MV_EXTENT 31 /* Max search distance in half pixel increments */
typedef struct CONFIG_TYPE2{
double OutputFrameRate;
ogg_uint32_t TargetBandwidth;
ogg_uint32_t KeyFrameDataTarget ; /* Data rate target for key frames */
ogg_uint32_t FirstFrameQ;
ogg_uint32_t BaseQ;
ogg_uint32_t MaxQ; /* Absolute Max Q allowed. */
ogg_uint32_t ActiveMaxQ; /* Currently active Max Q */
} CONFIG_TYPE2;
typedef struct coeffNode{
int i;
struct coeffNode *next;
} COEFFNODE;
typedef struct{
unsigned char * Yuv0ptr;
unsigned char * Yuv1ptr;
unsigned char * SrfWorkSpcPtr;
unsigned char * disp_fragments;
ogg_uint32_t * RegionIndex; /* Gives pixel index for top left of
each block */
ogg_uint32_t VideoFrameHeight;
ogg_uint32_t VideoFrameWidth;
} SCAN_CONFIG_DATA;
typedef unsigned char YUV_BUFFER_ENTRY;
typedef struct{
ogg_int32_t x;
ogg_int32_t y;
} MOTION_VECTOR;
typedef MOTION_VECTOR COORDINATE;
/** Quantizer matrix entry */
typedef ogg_int16_t Q_LIST_ENTRY;
/** Decode Post-Processor instance */
typedef struct PP_INSTANCE {
ogg_uint32_t PrevFrameLimit;
ogg_uint32_t *ScanPixelIndexTable;
signed char *ScanDisplayFragments;
signed char *PrevFragments[MAX_PREV_FRAMES];
ogg_uint32_t *FragScores; /* The individual frame difference ratings. */
signed char *SameGreyDirPixels;
signed char *BarBlockMap;
/* Number of pixels changed by diff threshold in row of a fragment. */
unsigned char *FragDiffPixels;
unsigned char *PixelScores;
unsigned char *PixelChangedMap;
unsigned char *ChLocals;
ogg_int16_t *yuv_differences;
ogg_int32_t *RowChangedPixels;
signed char *TmpCodedMap;
/* Plane pointers and dimension variables */
unsigned char * YPlanePtr0;
unsigned char * YPlanePtr1;
unsigned char * UPlanePtr0;
unsigned char * UPlanePtr1;
unsigned char * VPlanePtr0;
unsigned char * VPlanePtr1;
ogg_uint32_t VideoYPlaneWidth;
ogg_uint32_t VideoYPlaneHeight;
ogg_uint32_t VideoUVPlaneWidth;
ogg_uint32_t VideoUVPlaneHeight;
ogg_uint32_t VideoYPlaneStride;
ogg_uint32_t VideoUPlaneStride;
ogg_uint32_t VideoVPlaneStride;
/* Scan control variables. */
unsigned char HFragPixels;
unsigned char VFragPixels;
ogg_uint32_t ScanFrameFragments;
ogg_uint32_t ScanYPlaneFragments;
ogg_uint32_t ScanUVPlaneFragments;
ogg_uint32_t ScanHFragments;
ogg_uint32_t ScanVFragments;
ogg_uint32_t YFramePixels;
ogg_uint32_t UVFramePixels;
ogg_uint32_t SgcThresh;
ogg_uint32_t OutputBlocksUpdated;
ogg_uint32_t KFIndicator;
/* The pre-processor scan configuration. */
SCAN_CONFIG_DATA ScanConfig;
ogg_int32_t SRFGreyThresh;
ogg_int32_t SRFColThresh;
ogg_int32_t SgcLevelThresh;
ogg_int32_t SuvcLevelThresh;
ogg_uint32_t NoiseSupLevel;
/* Block Thresholds. */
ogg_uint32_t PrimaryBlockThreshold;
unsigned char LineSearchTripTresh;
int PAKEnabled;
int LevelThresh;
int NegLevelThresh;
int SrfThresh;
int NegSrfThresh;
int HighChange;
int NegHighChange;
/* Threshold lookup tables */
unsigned char SrfPakThreshTable[512];
unsigned char SrfThreshTable[512];
unsigned char SgcThreshTable[512];
/* Variables controlling S.A.D. break outs. */
ogg_uint32_t GrpLowSadThresh;
ogg_uint32_t GrpHighSadThresh;
ogg_uint32_t ModifiedGrpLowSadThresh;
ogg_uint32_t ModifiedGrpHighSadThresh;
ogg_int32_t PlaneHFragments;
ogg_int32_t PlaneVFragments;
ogg_int32_t PlaneHeight;
ogg_int32_t PlaneWidth;
ogg_int32_t PlaneStride;
ogg_uint32_t BlockThreshold;
ogg_uint32_t BlockSgcThresh;
double UVBlockThreshCorrection;
double UVSgcCorrection;
double YUVPlaneCorrectionFactor;
double AbsDiff_ScoreMultiplierTable[256];
unsigned char NoiseScoreBoostTable[256];
unsigned char MaxLineSearchLen;
ogg_int32_t YuvDiffsCircularBufferSize;
ogg_int32_t ChLocalsCircularBufferSize;
ogg_int32_t PixelMapCircularBufferSize;
DspFunctions dsp; /* Selected functions for this platform */
} PP_INSTANCE;
/** block coding modes */
typedef enum{
CODE_INTER_NO_MV = 0x0, /* INTER prediction, (0,0) motion
vector implied. */
CODE_INTRA = 0x1, /* INTRA i.e. no prediction. */
CODE_INTER_PLUS_MV = 0x2, /* INTER prediction, non zero motion
vector. */
CODE_INTER_LAST_MV = 0x3, /* Use Last Motion vector */
CODE_INTER_PRIOR_LAST = 0x4, /* Prior last motion vector */
CODE_USING_GOLDEN = 0x5, /* 'Golden frame' prediction (no MV). */
CODE_GOLDEN_MV = 0x6, /* 'Golden frame' prediction plus MV. */
CODE_INTER_FOURMV = 0x7 /* Inter prediction 4MV per macro block. */
} CODING_MODE;
/** Huffman table entry */
typedef struct HUFF_ENTRY {
struct HUFF_ENTRY *ZeroChild;
struct HUFF_ENTRY *OneChild;
struct HUFF_ENTRY *Previous;
struct HUFF_ENTRY *Next;
ogg_int32_t Value;
ogg_uint32_t Frequency;
} HUFF_ENTRY;
typedef struct qmat_range_table {
int startq, startqi; /* index where this range starts */
Q_LIST_ENTRY *qmat; /* qmat at this range boundary */
} qmat_range_table;
/** codec setup data, maps to the third bitstream header */
typedef struct codec_setup_info {
ogg_uint32_t QThreshTable[Q_TABLE_SIZE];
Q_LIST_ENTRY DcScaleFactorTable[Q_TABLE_SIZE];
int MaxQMatrixIndex;
Q_LIST_ENTRY *qmats;
qmat_range_table *range_table[6];
HUFF_ENTRY *HuffRoot[NUM_HUFF_TABLES];
} codec_setup_info;
/** Decoder (Playback) instance -- installed in a theora_state */
struct PB_INSTANCE {
oggpack_buffer *opb;
theora_info info;
/* flag to indicate if the headers already have been written */
int HeadersWritten;
/* how far do we shift the granulepos to seperate out P frame counts? */
int keyframe_granule_shift;
/***********************************************************************/
/* Decoder and Frame Type Information */
int DecoderErrorCode;
int FramesHaveBeenSkipped;
int PostProcessEnabled;
ogg_uint32_t PostProcessingLevel; /* Perform post processing */
/* Frame Info */
CODING_MODE CodingMode;
unsigned char FrameType;
unsigned char KeyFrameType;
ogg_uint32_t QualitySetting;
ogg_uint32_t FrameQIndex; /* Quality specified as a
table index */
ogg_uint32_t ThisFrameQualityValue; /* Quality value for this frame */
ogg_uint32_t LastFrameQualityValue; /* Last Frame's Quality */
ogg_int32_t CodedBlockIndex; /* Number of Coded Blocks */
ogg_uint32_t CodedBlocksThisFrame; /* Index into coded blocks */
ogg_uint32_t FrameSize; /* The number of bytes in the frame. */
/**********************************************************************/
/* Frame Size & Index Information */
ogg_uint32_t YPlaneSize;
ogg_uint32_t UVPlaneSize;
ogg_uint32_t YStride;
ogg_uint32_t UVStride;
ogg_uint32_t VFragments;
ogg_uint32_t HFragments;
ogg_uint32_t UnitFragments;
ogg_uint32_t YPlaneFragments;
ogg_uint32_t UVPlaneFragments;
ogg_uint32_t ReconYPlaneSize;
ogg_uint32_t ReconUVPlaneSize;
ogg_uint32_t YDataOffset;
ogg_uint32_t UDataOffset;
ogg_uint32_t VDataOffset;
ogg_uint32_t ReconYDataOffset;
ogg_uint32_t ReconUDataOffset;
ogg_uint32_t ReconVDataOffset;
ogg_uint32_t YSuperBlocks; /* Number of SuperBlocks in a Y frame */
ogg_uint32_t UVSuperBlocks; /* Number of SuperBlocks in a U or V frame */
ogg_uint32_t SuperBlocks; /* Total number of SuperBlocks in a
Y,U,V frame */
ogg_uint32_t YSBRows; /* Number of rows of SuperBlocks in a
Y frame */
ogg_uint32_t YSBCols; /* Number of cols of SuperBlocks in a
Y frame */
ogg_uint32_t UVSBRows; /* Number of rows of SuperBlocks in a
U or V frame */
ogg_uint32_t UVSBCols; /* Number of cols of SuperBlocks in a
U or V frame */
ogg_uint32_t MacroBlocks; /* Total number of Macro-Blocks */
/**********************************************************************/
/* Frames */
YUV_BUFFER_ENTRY *ThisFrameRecon;
YUV_BUFFER_ENTRY *GoldenFrame;
YUV_BUFFER_ENTRY *LastFrameRecon;
YUV_BUFFER_ENTRY *PostProcessBuffer;
/**********************************************************************/
/* Fragment Information */
ogg_uint32_t *pixel_index_table; /* start address of first
pixel of fragment in
source */
ogg_uint32_t *recon_pixel_index_table; /* start address of first
pixel in recon buffer */
unsigned char *display_fragments; /* Fragment update map */
unsigned char *skipped_display_fragments;/* whether fragment YUV
Conversion and update is to be
skipped */
ogg_int32_t *CodedBlockList; /* A list of fragment indices for
coded blocks. */
MOTION_VECTOR *FragMVect; /* Frag motion vectors */
ogg_uint32_t *FragTokenCounts; /* Number of tokens per fragment */
ogg_uint32_t (*TokenList)[128]; /* Fragment Token Pointers */
ogg_int32_t *FragmentVariances;
ogg_uint32_t *FragQIndex; /* Fragment Quality used in
PostProcess */
Q_LIST_ENTRY (*PPCoefBuffer)[64]; /* PostProcess Buffer for
coefficients data */
unsigned char *FragCoeffs; /* # of coeffs decoded so far for
fragment */
unsigned char *FragCoefEOB; /* Position of last non 0 coef
within QFragData */
Q_LIST_ENTRY (*QFragData)[64]; /* Fragment Coefficients
Array Pointers */
CODING_MODE *FragCodingMethod; /* coding method for the
fragment */
/***********************************************************************/
/* pointers to addresses used for allocation and deallocation the
others are rounded up to the nearest 32 bytes */
COEFFNODE *_Nodes;
ogg_uint32_t *transIndex; /* ptr to table of
transposed indexes */
/***********************************************************************/
ogg_int32_t bumpLast;
/* Macro Block and SuperBlock Information */
ogg_int32_t (*BlockMap)[4][4]; /* super block + sub macro
block + sub frag ->
FragIndex */
/* Coded flag arrays and counters for them */
unsigned char *SBCodedFlags;
unsigned char *SBFullyFlags;
unsigned char *MBCodedFlags;
unsigned char *MBFullyFlags;
/**********************************************************************/
ogg_uint32_t EOB_Run;
COORDINATE *FragCoordinates;
MOTION_VECTOR MVector;
ogg_int32_t ReconPtr2Offset; /* Offset for second reconstruction
in half pixel MC */
Q_LIST_ENTRY *quantized_list;
ogg_int16_t *ReconDataBuffer;
Q_LIST_ENTRY InvLastIntraDC;
Q_LIST_ENTRY InvLastInterDC;
Q_LIST_ENTRY LastIntraDC;
Q_LIST_ENTRY LastInterDC;
ogg_uint32_t BlocksToDecode; /* Blocks to be decoded this frame */
ogg_uint32_t DcHuffChoice; /* Huffman table selection variables */
unsigned char ACHuffChoice;
ogg_uint32_t QuadMBListIndex;
ogg_int32_t ByteCount;
ogg_uint32_t bit_pattern;
unsigned char bits_so_far;
unsigned char NextBit;
ogg_int32_t BitsLeft;
ogg_int16_t *DequantBuffer;
ogg_int32_t fp_quant_InterUV_coeffs[64];
ogg_int32_t fp_quant_InterUV_round[64];
ogg_int32_t fp_ZeroBinSize_InterUV[64];
ogg_int16_t *TmpReconBuffer;
ogg_int16_t *TmpDataBuffer;
/* Loop filter bounding values */
ogg_int16_t FiltBoundingValue[256];
/* Naming convention for all quant matrices and related data structures:
* Fields containing "Inter" in their name are for Inter frames, the
* rest is Intra. */
/* Dequantiser and rounding tables */
ogg_uint16_t *QThreshTable;
Q_LIST_ENTRY dequant_Y_coeffs[64];
Q_LIST_ENTRY dequant_U_coeffs[64];
Q_LIST_ENTRY dequant_V_coeffs[64];
Q_LIST_ENTRY dequant_InterY_coeffs[64];
Q_LIST_ENTRY dequant_InterU_coeffs[64];
Q_LIST_ENTRY dequant_InterV_coeffs[64];
Q_LIST_ENTRY *dequant_coeffs; /* currently active quantizer */
unsigned int zigzag_index[64];
HUFF_ENTRY *HuffRoot_VP3x[NUM_HUFF_TABLES];
ogg_uint32_t *HuffCodeArray_VP3x[NUM_HUFF_TABLES];
unsigned char *HuffCodeLengthArray_VP3x[NUM_HUFF_TABLES];
const unsigned char *ExtraBitLengths_VP3x;
th_quant_info quant_info;
oc_quant_tables quant_tables[2][3];
/* Quantiser and rounding tables */
/* this is scheduled to be replaced a new mechanism
that will simply reuse the dequantizer information. */
ogg_int32_t fp_quant_Y_coeffs[64]; /* used in reiniting quantizers */
ogg_int32_t fp_quant_U_coeffs[64];
ogg_int32_t fp_quant_V_coeffs[64];
ogg_int32_t fp_quant_Inter_Y_coeffs[64];
ogg_int32_t fp_quant_Inter_U_coeffs[64];
ogg_int32_t fp_quant_Inter_V_coeffs[64];
ogg_int32_t fp_quant_Y_round[64];
ogg_int32_t fp_quant_U_round[64];
ogg_int32_t fp_quant_V_round[64];
ogg_int32_t fp_quant_Inter_Y_round[64];
ogg_int32_t fp_quant_Inter_U_round[64];
ogg_int32_t fp_quant_Inter_V_round[64];
ogg_int32_t fp_ZeroBinSize_Y[64];
ogg_int32_t fp_ZeroBinSize_U[64];
ogg_int32_t fp_ZeroBinSize_V[64];
ogg_int32_t fp_ZeroBinSize_Inter_Y[64];
ogg_int32_t fp_ZeroBinSize_Inter_U[64];
ogg_int32_t fp_ZeroBinSize_Inter_V[64];
ogg_int32_t *fquant_coeffs;
ogg_int32_t *fquant_round;
ogg_int32_t *fquant_ZbSize;
/* Predictor used in choosing entropy table for decoding block patterns. */
unsigned char BlockPatternPredictor;
short Modifier[4][512];
short *ModifierPointer[4];
unsigned char *DataOutputInPtr;
DspFunctions dsp; /* Selected functions for this platform */
};
/* Encoder (Compressor) instance -- installed in a theora_state */
typedef struct CP_INSTANCE {
/*This structure must be first.
It contains entry points accessed by the decoder library's API wrapper, and
is the only assumption that library makes about our internal format.*/
oc_state_dispatch_vtbl dispatch_vtbl;
/* Compressor Configuration */
SCAN_CONFIG_DATA ScanConfig;
CONFIG_TYPE2 Configuration;
int GoldenFrameEnabled;
int InterPrediction;
int MotionCompensation;
ogg_uint32_t LastKeyFrame ;
ogg_int32_t DropCount ;
ogg_int32_t MaxConsDroppedFrames ;
ogg_int32_t DropFrameTriggerBytes;
int DropFrameCandidate;
/* Compressor Statistics */
double TotErrScore;
ogg_int64_t KeyFrameCount; /* Count of key frames. */
ogg_int64_t TotKeyFrameBytes;
ogg_uint32_t LastKeyFrameSize;
ogg_uint32_t PriorKeyFrameSize[KEY_FRAME_CONTEXT];
ogg_uint32_t PriorKeyFrameDistance[KEY_FRAME_CONTEXT];
ogg_int32_t FrameQuality[6];
int DecoderErrorCode; /* Decoder error flag. */
ogg_int32_t ThreshMapThreshold;
ogg_int32_t TotalMotionScore;
ogg_int64_t TotalByteCount;
ogg_int32_t FixedQ;
/* Frame Statistics */
signed char InterCodeCount;
ogg_int64_t CurrentFrame;
ogg_int64_t CarryOver ;
ogg_uint32_t LastFrameSize;
ogg_uint32_t FrameBitCount;
int ThisIsFirstFrame;
int ThisIsKeyFrame;
ogg_int32_t MotionScore;
ogg_uint32_t RegulationBlocks;
ogg_int32_t RecoveryMotionScore;
int RecoveryBlocksAdded ;
double ProportionRecBlocks;
double MaxRecFactor ;
/* Rate Targeting variables. */
ogg_uint32_t ThisFrameTargetBytes;
double BpbCorrectionFactor;
/* Up regulation variables */
ogg_uint32_t FinalPassLastPos; /* Used to regulate a final
unrestricted high quality
pass. */
ogg_uint32_t LastEndSB; /* Where we were in the loop
last time. */
ogg_uint32_t ResidueLastEndSB; /* Where we were in the residue
update loop last time. */
/* Controlling Block Selection */
ogg_uint32_t MVChangeFactor;
ogg_uint32_t FourMvChangeFactor;
ogg_uint32_t MinImprovementForNewMV;
ogg_uint32_t ExhaustiveSearchThresh;
ogg_uint32_t MinImprovementForFourMV;
ogg_uint32_t FourMVThreshold;
/* Module shared data structures. */
ogg_int32_t frame_target_rate;
ogg_int32_t BaseLineFrameTargetRate;
ogg_int32_t min_blocks_per_frame;
ogg_uint32_t tot_bytes_old;
/*********************************************************************/
/* Frames Used in the selecetive convolution filtering of the Y plane. */
unsigned char *ConvDestBuffer;
YUV_BUFFER_ENTRY *yuv0ptr;
YUV_BUFFER_ENTRY *yuv1ptr;
/*********************************************************************/
/*********************************************************************/
/* Token Buffers */
ogg_uint32_t *OptimisedTokenListEb; /* Optimised token list extra bits */
unsigned char *OptimisedTokenList; /* Optimised token list. */
unsigned char *OptimisedTokenListHi; /* Optimised token list huffman
table index */
unsigned char *OptimisedTokenListPl; /* Plane to which the token
belongs Y = 0 or UV = 1 */
ogg_int32_t OptimisedTokenCount; /* Count of Optimized tokens */
ogg_uint32_t RunHuffIndex; /* Huffman table in force at
the start of a run */
ogg_uint32_t RunPlaneIndex; /* The plane (Y=0 UV=1) to
which the first token in
an EOB run belonged. */
ogg_uint32_t TotTokenCount;
ogg_int32_t TokensToBeCoded;
ogg_int32_t TokensCoded;
/********************************************************************/
/* SuperBlock, MacroBLock and Fragment Information */
/* Coded flag arrays and counters for them */
unsigned char *PartiallyCodedFlags;
unsigned char *PartiallyCodedMbPatterns;
unsigned char *UncodedMbFlags;
unsigned char *extra_fragments; /* extra updates not
recommended by pre-processor */
ogg_int16_t *OriginalDC;
ogg_uint32_t *FragmentLastQ; /* Array used to keep track of
quality at which each
fragment was last
updated. */
unsigned char *FragTokens;
ogg_uint32_t *FragTokenCounts; /* Number of tokens per fragment */
ogg_uint32_t *RunHuffIndices;
ogg_uint32_t *LastCodedErrorScore;
ogg_uint32_t *ModeList;
MOTION_VECTOR *MVList;
unsigned char *BlockCodedFlags;
ogg_uint32_t MvListCount;
ogg_uint32_t ModeListCount;
unsigned char *DataOutputBuffer;
/*********************************************************************/
ogg_uint32_t RunLength;
ogg_uint32_t MaxBitTarget; /* Cut off target for rate capping */
double BitRateCapFactor; /* Factor relating delta frame target
to cut off target. */
unsigned char MBCodingMode; /* Coding mode flags */
ogg_int32_t MVPixelOffsetY[MAX_SEARCH_SITES];
ogg_uint32_t InterTripOutThresh;
unsigned char MVEnabled;
ogg_uint32_t MotionVectorSearchCount;
ogg_uint32_t FrameMVSearcOunt;
ogg_int32_t MVSearchSteps;
ogg_int32_t MVOffsetX[MAX_SEARCH_SITES];
ogg_int32_t MVOffsetY[MAX_SEARCH_SITES];
ogg_int32_t HalfPixelRef2Offset[9]; /* Offsets for half pixel
compensation */
signed char HalfPixelXOffset[9]; /* Half pixel MV offsets for X */
signed char HalfPixelYOffset[9]; /* Half pixel MV offsets for Y */
ogg_uint32_t bit_pattern ;
unsigned char bits_so_far ;
ogg_uint32_t lastval ;
ogg_uint32_t lastrun ;
Q_LIST_ENTRY *quantized_list;
MOTION_VECTOR MVector;
ogg_uint32_t TempBitCount;
ogg_int16_t *DCT_codes; /* Buffer that stores the result of
Forward DCT */
ogg_int16_t *DCTDataBuffer; /* Input data buffer for Forward DCT */
/* Motion compensation related variables */
ogg_uint32_t MvMaxExtent;
double QTargetModifier[Q_TABLE_SIZE];
/* instances (used for reconstructing buffers and to hold tokens etc.) */
PP_INSTANCE pp; /* preprocessor */
PB_INSTANCE pb; /* playback */
/* ogg bitpacker for use in packet coding, other API state */
oggpack_buffer *oggbuffer;
int readyflag;
int packetflag;
int doneflag;
DspFunctions dsp; /* Selected functions for this platform */
} CP_INSTANCE;
#define clamp255(x) ((unsigned char)((((x)<0)-1) & ((x) | -((x)>255))))
extern void ConfigurePP( PP_INSTANCE *ppi, int Level ) ;
extern ogg_uint32_t YUVAnalyseFrame( PP_INSTANCE *ppi,
ogg_uint32_t * KFIndicator );
extern void ClearPPInstance(PP_INSTANCE *ppi);
extern void InitPPInstance(PP_INSTANCE *ppi, DspFunctions *funcs);
extern void InitPBInstance(PB_INSTANCE *pbi);
extern void ClearPBInstance(PB_INSTANCE *pbi);
extern void IDct1( Q_LIST_ENTRY * InputData,
ogg_int16_t *QuantMatrix,
ogg_int16_t * OutputData );
extern void ReconIntra( PB_INSTANCE *pbi, unsigned char * ReconPtr,
ogg_int16_t * ChangePtr, ogg_uint32_t LineStep );
extern void ReconInter( PB_INSTANCE *pbi, unsigned char * ReconPtr,
unsigned char * RefPtr, ogg_int16_t * ChangePtr,
ogg_uint32_t LineStep ) ;
extern void ReconInterHalfPixel2( PB_INSTANCE *pbi, unsigned char * ReconPtr,
unsigned char * RefPtr1,
unsigned char * RefPtr2,
ogg_int16_t * ChangePtr,
ogg_uint32_t LineStep ) ;
extern void SetupLoopFilter(PB_INSTANCE *pbi);
extern void CopyBlock(unsigned char *src,
unsigned char *dest,
unsigned int srcstride);
extern void LoopFilter(PB_INSTANCE *pbi);
extern void ReconRefFrames (PB_INSTANCE *pbi);
extern void ExpandToken( Q_LIST_ENTRY * ExpandedBlock,
unsigned char * CoeffIndex, ogg_uint32_t Token,
ogg_int32_t ExtraBits );
extern void ClearDownQFragData(PB_INSTANCE *pbi);
extern void select_quantiser (PB_INSTANCE *pbi, int type);
extern void quantize( PB_INSTANCE *pbi,
ogg_int16_t * DCT_block,
Q_LIST_ENTRY * quantized_list);
extern void UpdateQ( PB_INSTANCE *pbi, int NewQIndex );
extern void UpdateQC( CP_INSTANCE *cpi, ogg_uint32_t NewQ );
extern void fdct_short ( ogg_int16_t * InputData, ogg_int16_t * OutputData );
extern ogg_uint32_t DPCMTokenizeBlock (CP_INSTANCE *cpi,
ogg_int32_t FragIndex);
extern void TransformQuantizeBlock (CP_INSTANCE *cpi, ogg_int32_t FragIndex,
ogg_uint32_t PixelsPerLine ) ;
extern void ClearFragmentInfo(PB_INSTANCE * pbi);
extern void InitFragmentInfo(PB_INSTANCE * pbi);
extern void ClearFrameInfo(PB_INSTANCE * pbi);
extern void InitFrameInfo(PB_INSTANCE * pbi, unsigned int FrameSize);
extern void InitializeFragCoordinates(PB_INSTANCE *pbi);
extern void InitFrameDetails(PB_INSTANCE *pbi);
extern void WriteQTables(PB_INSTANCE *pbi,oggpack_buffer *opb);
extern void InitQTables( PB_INSTANCE *pbi );
extern void quant_tables_init( PB_INSTANCE *pbi, const th_quant_info *qinfo);
extern void InitHuffmanSet( PB_INSTANCE *pbi );
extern void ClearHuffmanSet( PB_INSTANCE *pbi );
extern int ReadHuffmanTrees(codec_setup_info *ci, oggpack_buffer *opb);
extern void WriteHuffmanTrees(HUFF_ENTRY *HuffRoot[NUM_HUFF_TABLES],
oggpack_buffer *opb);
extern void InitHuffmanTrees(PB_INSTANCE *pbi, const codec_setup_info *ci);
extern void ClearHuffmanTrees(HUFF_ENTRY *HuffRoot[NUM_HUFF_TABLES]);
extern int ReadFilterTables(codec_setup_info *ci, oggpack_buffer *opb);
extern void QuadDecodeDisplayFragments ( PB_INSTANCE *pbi );
extern void PackAndWriteDFArray( CP_INSTANCE *cpi );
extern void UpdateFragQIndex(PB_INSTANCE *pbi);
extern void PostProcess(PB_INSTANCE *pbi);
extern void InitMotionCompensation ( CP_INSTANCE *cpi );
extern ogg_uint32_t GetMBIntraError (CP_INSTANCE *cpi, ogg_uint32_t FragIndex,
ogg_uint32_t PixelsPerLine ) ;
extern ogg_uint32_t GetMBInterError (CP_INSTANCE *cpi,
unsigned char * SrcPtr,
unsigned char * RefPtr,
ogg_uint32_t FragIndex,
ogg_int32_t LastXMV,
ogg_int32_t LastYMV,
ogg_uint32_t PixelsPerLine ) ;
extern void WriteFrameHeader( CP_INSTANCE *cpi) ;
extern ogg_uint32_t GetMBMVInterError (CP_INSTANCE *cpi,
unsigned char * RefFramePtr,
ogg_uint32_t FragIndex,
ogg_uint32_t PixelsPerLine,
ogg_int32_t *MVPixelOffset,
MOTION_VECTOR *MV );
extern ogg_uint32_t GetMBMVExhaustiveSearch (CP_INSTANCE *cpi,
unsigned char * RefFramePtr,
ogg_uint32_t FragIndex,
ogg_uint32_t PixelsPerLine,
MOTION_VECTOR *MV );
extern ogg_uint32_t GetFOURMVExhaustiveSearch (CP_INSTANCE *cpi,
unsigned char * RefFramePtr,
ogg_uint32_t FragIndex,
ogg_uint32_t PixelsPerLine,
MOTION_VECTOR *MV ) ;
extern ogg_uint32_t EncodeData(CP_INSTANCE *cpi);
extern ogg_uint32_t PickIntra( CP_INSTANCE *cpi,
ogg_uint32_t SBRows,
ogg_uint32_t SBCols);
extern ogg_uint32_t PickModes(CP_INSTANCE *cpi,
ogg_uint32_t SBRows,
ogg_uint32_t SBCols,
ogg_uint32_t PixelsPerLine,
ogg_uint32_t *InterError,
ogg_uint32_t *IntraError);
extern CODING_MODE FrArrayUnpackMode(PB_INSTANCE *pbi);
extern void CreateBlockMapping ( ogg_int32_t (*BlockMap)[4][4],
ogg_uint32_t YSuperBlocks,
ogg_uint32_t UVSuperBlocks,
ogg_uint32_t HFrags, ogg_uint32_t VFrags );
extern void UpRegulateDataStream (CP_INSTANCE *cpi, ogg_uint32_t RegulationQ,
ogg_int32_t RecoveryBlocks ) ;
extern void RegulateQ( CP_INSTANCE *cpi, ogg_int32_t UpdateScore );
extern void CopyBackExtraFrags(CP_INSTANCE *cpi);
extern void UpdateUMVBorder( PB_INSTANCE *pbi,
unsigned char * DestReconPtr );
extern void PInitFrameInfo(PP_INSTANCE * ppi);
extern double GetEstimatedBpb( CP_INSTANCE *cpi, ogg_uint32_t TargetQ );
extern void ClearTmpBuffers(PB_INSTANCE * pbi);
extern void InitTmpBuffers(PB_INSTANCE * pbi);
extern void ScanYUVInit( PP_INSTANCE * ppi,
SCAN_CONFIG_DATA * ScanConfigPtr);
#endif /* ENCODER_INTERNAL_H */

268
Engine/lib/libtheora/lib/enc/dct.c Normal file
View file

@ -0,0 +1,268 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: dct.c 13884 2007-09-22 08:38:10Z giles $
********************************************************************/
#include "codec_internal.h"
#include "dsp.h"
#include "../cpu.h"
static ogg_int32_t xC1S7 = 64277;
static ogg_int32_t xC2S6 = 60547;
static ogg_int32_t xC3S5 = 54491;
static ogg_int32_t xC4S4 = 46341;
static ogg_int32_t xC5S3 = 36410;
static ogg_int32_t xC6S2 = 25080;
static ogg_int32_t xC7S1 = 12785;
#define SIGNBITDUPPED(X) ((signed )(((X) & 0x80000000)) >> 31)
#define DOROUND(X) ( (SIGNBITDUPPED(X) & (0xffff)) + (X) )
static void fdct_short__c ( ogg_int16_t * InputData, ogg_int16_t * OutputData ){
int loop;
ogg_int32_t is07, is12, is34, is56;
ogg_int32_t is0734, is1256;
ogg_int32_t id07, id12, id34, id56;
ogg_int32_t irot_input_x, irot_input_y;
ogg_int32_t icommon_product1; /* Re-used product (c4s4 * (s12 - s56)). */
ogg_int32_t icommon_product2; /* Re-used product (c4s4 * (d12 + d56)). */
ogg_int32_t temp1, temp2; /* intermediate variable for computation */
ogg_int32_t InterData[64];
ogg_int32_t *ip = InterData;
ogg_int16_t * op = OutputData;
for (loop = 0; loop < 8; loop++){
/* Pre calculate some common sums and differences. */
is07 = InputData[0] + InputData[7];
is12 = InputData[1] + InputData[2];
is34 = InputData[3] + InputData[4];
is56 = InputData[5] + InputData[6];
id07 = InputData[0] - InputData[7];
id12 = InputData[1] - InputData[2];
id34 = InputData[3] - InputData[4];
id56 = InputData[5] - InputData[6];
is0734 = is07 + is34;
is1256 = is12 + is56;
/* Pre-Calculate some common product terms. */
icommon_product1 = xC4S4*(is12 - is56);
icommon_product1 = DOROUND(icommon_product1);
icommon_product1>>=16;
icommon_product2 = xC4S4*(id12 + id56);
icommon_product2 = DOROUND(icommon_product2);
icommon_product2>>=16;
ip[0] = (xC4S4*(is0734 + is1256));
ip[0] = DOROUND(ip[0]);
ip[0] >>= 16;
ip[4] = (xC4S4*(is0734 - is1256));
ip[4] = DOROUND(ip[4]);
ip[4] >>= 16;
/* Define inputs to rotation for outputs 2 and 6 */
irot_input_x = id12 - id56;
irot_input_y = is07 - is34;
/* Apply rotation for outputs 2 and 6. */
temp1=xC6S2*irot_input_x;
temp1=DOROUND(temp1);
temp1>>=16;
temp2=xC2S6*irot_input_y;
temp2=DOROUND(temp2);
temp2>>=16;
ip[2] = temp1 + temp2;
temp1=xC6S2*irot_input_y;
temp1=DOROUND(temp1);
temp1>>=16;
temp2=xC2S6*irot_input_x ;
temp2=DOROUND(temp2);
temp2>>=16;
ip[6] = temp1 -temp2 ;
/* Define inputs to rotation for outputs 1 and 7 */
irot_input_x = icommon_product1 + id07;
irot_input_y = -( id34 + icommon_product2 );
/* Apply rotation for outputs 1 and 7. */
temp1=xC1S7*irot_input_x;
temp1=DOROUND(temp1);
temp1>>=16;
temp2=xC7S1*irot_input_y;
temp2=DOROUND(temp2);
temp2>>=16;
ip[1] = temp1 - temp2;
temp1=xC7S1*irot_input_x;
temp1=DOROUND(temp1);
temp1>>=16;
temp2=xC1S7*irot_input_y ;
temp2=DOROUND(temp2);
temp2>>=16;
ip[7] = temp1 + temp2 ;
/* Define inputs to rotation for outputs 3 and 5 */
irot_input_x = id07 - icommon_product1;
irot_input_y = id34 - icommon_product2;
/* Apply rotation for outputs 3 and 5. */
temp1=xC3S5*irot_input_x;
temp1=DOROUND(temp1);
temp1>>=16;
temp2=xC5S3*irot_input_y ;
temp2=DOROUND(temp2);
temp2>>=16;
ip[3] = temp1 - temp2 ;
temp1=xC5S3*irot_input_x;
temp1=DOROUND(temp1);
temp1>>=16;
temp2=xC3S5*irot_input_y;
temp2=DOROUND(temp2);
temp2>>=16;
ip[5] = temp1 + temp2;
/* Increment data pointer for next row. */
InputData += 8 ;
ip += 8; /* advance pointer to next row */
}
/* Performed DCT on rows, now transform the columns */
ip = InterData;
for (loop = 0; loop < 8; loop++){
/* Pre calculate some common sums and differences. */
is07 = ip[0 * 8] + ip[7 * 8];
is12 = ip[1 * 8] + ip[2 * 8];
is34 = ip[3 * 8] + ip[4 * 8];
is56 = ip[5 * 8] + ip[6 * 8];
id07 = ip[0 * 8] - ip[7 * 8];
id12 = ip[1 * 8] - ip[2 * 8];
id34 = ip[3 * 8] - ip[4 * 8];
id56 = ip[5 * 8] - ip[6 * 8];
is0734 = is07 + is34;
is1256 = is12 + is56;
/* Pre-Calculate some common product terms. */
icommon_product1 = xC4S4*(is12 - is56) ;
icommon_product2 = xC4S4*(id12 + id56) ;
icommon_product1 = DOROUND(icommon_product1);
icommon_product2 = DOROUND(icommon_product2);
icommon_product1>>=16;
icommon_product2>>=16;
temp1 = xC4S4*(is0734 + is1256) ;
temp2 = xC4S4*(is0734 - is1256) ;
temp1 = DOROUND(temp1);
temp2 = DOROUND(temp2);
temp1>>=16;
temp2>>=16;
op[0*8] = (ogg_int16_t) temp1;
op[4*8] = (ogg_int16_t) temp2;
/* Define inputs to rotation for outputs 2 and 6 */
irot_input_x = id12 - id56;
irot_input_y = is07 - is34;
/* Apply rotation for outputs 2 and 6. */
temp1=xC6S2*irot_input_x;
temp1=DOROUND(temp1);
temp1>>=16;
temp2=xC2S6*irot_input_y;
temp2=DOROUND(temp2);
temp2>>=16;
op[2*8] = (ogg_int16_t) (temp1 + temp2);
temp1=xC6S2*irot_input_y;
temp1=DOROUND(temp1);
temp1>>=16;
temp2=xC2S6*irot_input_x ;
temp2=DOROUND(temp2);
temp2>>=16;
op[6*8] = (ogg_int16_t) (temp1 -temp2) ;
/* Define inputs to rotation for outputs 1 and 7 */
irot_input_x = icommon_product1 + id07;
irot_input_y = -( id34 + icommon_product2 );
/* Apply rotation for outputs 1 and 7. */
temp1=xC1S7*irot_input_x;
temp1=DOROUND(temp1);
temp1>>=16;
temp2=xC7S1*irot_input_y;
temp2=DOROUND(temp2);
temp2>>=16;
op[1*8] = (ogg_int16_t) (temp1 - temp2);
temp1=xC7S1*irot_input_x;
temp1=DOROUND(temp1);
temp1>>=16;
temp2=xC1S7*irot_input_y ;
temp2=DOROUND(temp2);
temp2>>=16;
op[7*8] = (ogg_int16_t) (temp1 + temp2);
/* Define inputs to rotation for outputs 3 and 5 */
irot_input_x = id07 - icommon_product1;
irot_input_y = id34 - icommon_product2;
/* Apply rotation for outputs 3 and 5. */
temp1=xC3S5*irot_input_x;
temp1=DOROUND(temp1);
temp1>>=16;
temp2=xC5S3*irot_input_y ;
temp2=DOROUND(temp2);
temp2>>=16;
op[3*8] = (ogg_int16_t) (temp1 - temp2) ;
temp1=xC5S3*irot_input_x;
temp1=DOROUND(temp1);
temp1>>=16;
temp2=xC3S5*irot_input_y;
temp2=DOROUND(temp2);
temp2>>=16;
op[5*8] = (ogg_int16_t) (temp1 + temp2);
/* Increment data pointer for next column. */
ip ++;
op ++;
}
}
void dsp_dct_init (DspFunctions *funcs, ogg_uint32_t cpu_flags)
{
funcs->fdct_short = fdct_short__c;
dsp_dct_decode_init(funcs, cpu_flags);
dsp_idct_init(funcs, cpu_flags);
#if defined(USE_ASM)
if (cpu_flags & OC_CPU_X86_MMX) {
dsp_mmx_fdct_init(funcs);
}
#endif
}

941
Engine/lib/libtheora/lib/enc/dct_decode.c Normal file
View file

@ -0,0 +1,941 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: dct_decode.c 15400 2008-10-15 12:10:58Z tterribe $
********************************************************************/
#include <stdlib.h>
#include <string.h>
#include "codec_internal.h"
#include "quant_lookup.h"
#define GOLDEN_FRAME_THRESH_Q 50
#define PUR 8
#define PU 4
#define PUL 2
#define PL 1
#define HIGHBITDUPPED(X) (((signed short) X) >> 15)
static const int ModeUsesMC[MAX_MODES] = { 0, 0, 1, 1, 1, 0, 1, 1 };
static void SetupBoundingValueArray_Generic(ogg_int16_t *BoundingValuePtr,
ogg_int32_t FLimit){
ogg_int32_t i;
/* Set up the bounding value array. */
memset ( BoundingValuePtr, 0, (256*sizeof(*BoundingValuePtr)) );
for ( i = 0; i < FLimit; i++ ){
BoundingValuePtr[127-i-FLimit] = (-FLimit+i);
BoundingValuePtr[127-i] = -i;
BoundingValuePtr[127+i] = i;
BoundingValuePtr[127+i+FLimit] = FLimit-i;
}
}
static void ExpandKFBlock ( PB_INSTANCE *pbi, ogg_int32_t FragmentNumber ){
ogg_uint32_t ReconPixelsPerLine;
ogg_int32_t ReconPixelIndex;
/* Select the appropriate inverse Q matrix and line stride */
if ( FragmentNumber<(ogg_int32_t)pbi->YPlaneFragments ){
ReconPixelsPerLine = pbi->YStride;
pbi->dequant_coeffs = pbi->dequant_Y_coeffs;
}else if ( FragmentNumber<(ogg_int32_t)(pbi->YPlaneFragments + pbi->UVPlaneFragments) ){
ReconPixelsPerLine = pbi->UVStride;
pbi->dequant_coeffs = pbi->dequant_U_coeffs;
}else{
ReconPixelsPerLine = pbi->UVStride;
pbi->dequant_coeffs = pbi->dequant_V_coeffs;
}
/* Set up pointer into the quantisation buffer. */
pbi->quantized_list = &pbi->QFragData[FragmentNumber][0];
/* Invert quantisation and DCT to get pixel data. */
switch(pbi->FragCoefEOB[FragmentNumber]){
case 0:case 1:
IDct1( pbi->quantized_list, pbi->dequant_coeffs, pbi->ReconDataBuffer );
break;
case 2: case 3:
dsp_IDct3(pbi->dsp, pbi->quantized_list, pbi->dequant_coeffs, pbi->ReconDataBuffer );
break;
case 4:case 5:case 6:case 7:case 8: case 9:case 10:
dsp_IDct10(pbi->dsp, pbi->quantized_list, pbi->dequant_coeffs, pbi->ReconDataBuffer );
break;
default:
dsp_IDctSlow(pbi->dsp, pbi->quantized_list, pbi->dequant_coeffs, pbi->ReconDataBuffer );
}
/* Convert fragment number to a pixel offset in a reconstruction buffer. */
ReconPixelIndex = pbi->recon_pixel_index_table[FragmentNumber];
/* Get the pixel index for the first pixel in the fragment. */
dsp_recon_intra8x8 (pbi->dsp, (unsigned char *)(&pbi->ThisFrameRecon[ReconPixelIndex]),
(ogg_int16_t *)pbi->ReconDataBuffer, ReconPixelsPerLine);
}
static void ExpandBlock ( PB_INSTANCE *pbi, ogg_int32_t FragmentNumber){
unsigned char *LastFrameRecPtr; /* Pointer into previous frame
reconstruction. */
unsigned char *LastFrameRecPtr2; /* Pointer into previous frame
reconstruction for 1/2 pixel MC. */
ogg_uint32_t ReconPixelsPerLine; /* Pixels per line */
ogg_int32_t ReconPixelIndex; /* Offset for block into a
reconstruction buffer */
ogg_int32_t ReconPtr2Offset; /* Offset for second
reconstruction in half pixel
MC */
ogg_int32_t MVOffset; /* Baseline motion vector offset */
ogg_int32_t MvShift ; /* Shift to correct to 1/2 or 1/4 pixel */
ogg_int32_t MvModMask; /* Mask to determine whether 1/2
pixel is used */
/* Get coding mode for this block */
if ( pbi->FrameType == KEY_FRAME ){
pbi->CodingMode = CODE_INTRA;
}else{
/* Get Motion vector and mode for this block. */
pbi->CodingMode = pbi->FragCodingMethod[FragmentNumber];
}
/* Select the appropriate inverse Q matrix and line stride */
if ( FragmentNumber<(ogg_int32_t)pbi->YPlaneFragments ) {
ReconPixelsPerLine = pbi->YStride;
MvShift = 1;
MvModMask = 0x00000001;
/* Select appropriate dequantiser matrix. */
if ( pbi->CodingMode == CODE_INTRA )
pbi->dequant_coeffs = pbi->dequant_Y_coeffs;
else
pbi->dequant_coeffs = pbi->dequant_InterY_coeffs;
}else{
ReconPixelsPerLine = pbi->UVStride;
MvShift = 2;
MvModMask = 0x00000003;
/* Select appropriate dequantiser matrix. */
if ( pbi->CodingMode == CODE_INTRA )
if ( FragmentNumber <
(ogg_int32_t)(pbi->YPlaneFragments + pbi->UVPlaneFragments) )
pbi->dequant_coeffs = pbi->dequant_U_coeffs;
else
pbi->dequant_coeffs = pbi->dequant_V_coeffs;
else
if ( FragmentNumber <
(ogg_int32_t)(pbi->YPlaneFragments + pbi->UVPlaneFragments) )
pbi->dequant_coeffs = pbi->dequant_InterU_coeffs;
else
pbi->dequant_coeffs = pbi->dequant_InterV_coeffs;
}
/* Set up pointer into the quantisation buffer. */
pbi->quantized_list = &pbi->QFragData[FragmentNumber][0];
/* Invert quantisation and DCT to get pixel data. */
switch(pbi->FragCoefEOB[FragmentNumber]){
case 0:case 1:
IDct1( pbi->quantized_list, pbi->dequant_coeffs, pbi->ReconDataBuffer );
break;
case 2: case 3:
dsp_IDct3(pbi->dsp, pbi->quantized_list, pbi->dequant_coeffs, pbi->ReconDataBuffer );
break;
case 4:case 5:case 6:case 7:case 8: case 9:case 10:
dsp_IDct10(pbi->dsp, pbi->quantized_list, pbi->dequant_coeffs, pbi->ReconDataBuffer );
break;
default:
dsp_IDctSlow(pbi->dsp, pbi->quantized_list, pbi->dequant_coeffs, pbi->ReconDataBuffer );
}
/* Convert fragment number to a pixel offset in a reconstruction buffer. */
ReconPixelIndex = pbi->recon_pixel_index_table[FragmentNumber];
/* Action depends on decode mode. */
if ( pbi->CodingMode == CODE_INTER_NO_MV ){
/* Inter with no motion vector */
/* Reconstruct the pixel data using the last frame reconstruction
and change data when the motion vector is (0,0), the recon is
based on the lastframe without loop filtering---- for testing */
dsp_recon_inter8x8 (pbi->dsp, &pbi->ThisFrameRecon[ReconPixelIndex],
&pbi->LastFrameRecon[ReconPixelIndex],
pbi->ReconDataBuffer, ReconPixelsPerLine);
}else if ( ModeUsesMC[pbi->CodingMode] ) {
/* The mode uses a motion vector. */
/* Get vector from list */
pbi->MVector.x = pbi->FragMVect[FragmentNumber].x;
pbi->MVector.y = pbi->FragMVect[FragmentNumber].y;
/* Work out the base motion vector offset and the 1/2 pixel offset
if any. For the U and V planes the MV specifies 1/4 pixel
accuracy. This is adjusted to 1/2 pixel as follows ( 0->0,
1/4->1/2, 1/2->1/2, 3/4->1/2 ). */
MVOffset = 0;
ReconPtr2Offset = 0;
if ( pbi->MVector.x > 0 ){
MVOffset = pbi->MVector.x >> MvShift;
if ( pbi->MVector.x & MvModMask )
ReconPtr2Offset += 1;
} else if ( pbi->MVector.x < 0 ) {
MVOffset -= (-pbi->MVector.x) >> MvShift;
if ( (-pbi->MVector.x) & MvModMask )
ReconPtr2Offset -= 1;
}
if ( pbi->MVector.y > 0 ){
MVOffset += (pbi->MVector.y >> MvShift) * ReconPixelsPerLine;
if ( pbi->MVector.y & MvModMask )
ReconPtr2Offset += ReconPixelsPerLine;
} else if ( pbi->MVector.y < 0 ){
MVOffset -= ((-pbi->MVector.y) >> MvShift) * ReconPixelsPerLine;
if ( (-pbi->MVector.y) & MvModMask )
ReconPtr2Offset -= ReconPixelsPerLine;
}
/* Set up the first of the two reconstruction buffer pointers. */
if ( pbi->CodingMode==CODE_GOLDEN_MV ) {
LastFrameRecPtr = &pbi->GoldenFrame[ReconPixelIndex] + MVOffset;
}else{
LastFrameRecPtr = &pbi->LastFrameRecon[ReconPixelIndex] + MVOffset;
}
/* Set up the second of the two reconstruction pointers. */
LastFrameRecPtr2 = LastFrameRecPtr + ReconPtr2Offset;
/* Select the appropriate reconstruction function */
if ( (int)(LastFrameRecPtr - LastFrameRecPtr2) == 0 ) {
/* Reconstruct the pixel dats from the reference frame and change data
(no half pixel in this case as the two references were the same. */
dsp_recon_inter8x8 (pbi->dsp,
&pbi->ThisFrameRecon[ReconPixelIndex],
LastFrameRecPtr, pbi->ReconDataBuffer,
ReconPixelsPerLine);
}else{
/* Fractional pixel reconstruction. */
/* Note that we only use two pixels per reconstruction even for
the diagonal. */
dsp_recon_inter8x8_half(pbi->dsp, &pbi->ThisFrameRecon[ReconPixelIndex],
LastFrameRecPtr, LastFrameRecPtr2,
pbi->ReconDataBuffer, ReconPixelsPerLine);
}
} else if ( pbi->CodingMode == CODE_USING_GOLDEN ){
/* Golden frame with motion vector */
/* Reconstruct the pixel data using the golden frame
reconstruction and change data */
dsp_recon_inter8x8 (pbi->dsp, &pbi->ThisFrameRecon[ReconPixelIndex],
&pbi->GoldenFrame[ ReconPixelIndex ],
pbi->ReconDataBuffer, ReconPixelsPerLine);
} else {
/* Simple Intra coding */
/* Get the pixel index for the first pixel in the fragment. */
dsp_recon_intra8x8 (pbi->dsp, &pbi->ThisFrameRecon[ReconPixelIndex],
pbi->ReconDataBuffer, ReconPixelsPerLine);
}
}
static void UpdateUMV_HBorders( PB_INSTANCE *pbi,
unsigned char * DestReconPtr,
ogg_uint32_t PlaneFragOffset ) {
ogg_uint32_t i;
ogg_uint32_t PixelIndex;
ogg_uint32_t PlaneStride;
ogg_uint32_t BlockVStep;
ogg_uint32_t PlaneFragments;
ogg_uint32_t LineFragments;
ogg_uint32_t PlaneBorderWidth;
unsigned char *SrcPtr1;
unsigned char *SrcPtr2;
unsigned char *DestPtr1;
unsigned char *DestPtr2;
/* Work out various plane specific values */
if ( PlaneFragOffset == 0 ) {
/* Y Plane */
BlockVStep = (pbi->YStride *
(VFRAGPIXELS - 1));
PlaneStride = pbi->YStride;
PlaneBorderWidth = UMV_BORDER;
PlaneFragments = pbi->YPlaneFragments;
LineFragments = pbi->HFragments;
}else{
/* U or V plane. */
BlockVStep = (pbi->UVStride *
(VFRAGPIXELS - 1));
PlaneStride = pbi->UVStride;
PlaneBorderWidth = UMV_BORDER / 2;
PlaneFragments = pbi->UVPlaneFragments;
LineFragments = pbi->HFragments / 2;
}
/* Setup the source and destination pointers for the top and bottom
borders */
PixelIndex = pbi->recon_pixel_index_table[PlaneFragOffset];
SrcPtr1 = &DestReconPtr[ PixelIndex - PlaneBorderWidth ];
DestPtr1 = SrcPtr1 - (PlaneBorderWidth * PlaneStride);
PixelIndex = pbi->recon_pixel_index_table[PlaneFragOffset +
PlaneFragments - LineFragments] +
BlockVStep;
SrcPtr2 = &DestReconPtr[ PixelIndex - PlaneBorderWidth];
DestPtr2 = SrcPtr2 + PlaneStride;
/* Now copy the top and bottom source lines into each line of the
respective borders */
for ( i = 0; i < PlaneBorderWidth; i++ ) {
memcpy( DestPtr1, SrcPtr1, PlaneStride );
memcpy( DestPtr2, SrcPtr2, PlaneStride );
DestPtr1 += PlaneStride;
DestPtr2 += PlaneStride;
}
}
static void UpdateUMV_VBorders( PB_INSTANCE *pbi,
unsigned char * DestReconPtr,
ogg_uint32_t PlaneFragOffset ){
ogg_uint32_t i;
ogg_uint32_t PixelIndex;
ogg_uint32_t PlaneStride;
ogg_uint32_t LineFragments;
ogg_uint32_t PlaneBorderWidth;
ogg_uint32_t PlaneHeight;
unsigned char *SrcPtr1;
unsigned char *SrcPtr2;
unsigned char *DestPtr1;
unsigned char *DestPtr2;
/* Work out various plane specific values */
if ( PlaneFragOffset == 0 ) {
/* Y Plane */
PlaneStride = pbi->YStride;
PlaneBorderWidth = UMV_BORDER;
LineFragments = pbi->HFragments;
PlaneHeight = pbi->info.height;
}else{
/* U or V plane. */
PlaneStride = pbi->UVStride;
PlaneBorderWidth = UMV_BORDER / 2;
LineFragments = pbi->HFragments / 2;
PlaneHeight = pbi->info.height / 2;
}
/* Setup the source data values and destination pointers for the
left and right edge borders */
PixelIndex = pbi->recon_pixel_index_table[PlaneFragOffset];
SrcPtr1 = &DestReconPtr[ PixelIndex ];
DestPtr1 = &DestReconPtr[ PixelIndex - PlaneBorderWidth ];
PixelIndex = pbi->recon_pixel_index_table[PlaneFragOffset +
LineFragments - 1] +
(HFRAGPIXELS - 1);
SrcPtr2 = &DestReconPtr[ PixelIndex ];
DestPtr2 = &DestReconPtr[ PixelIndex + 1 ];
/* Now copy the top and bottom source lines into each line of the
respective borders */
for ( i = 0; i < PlaneHeight; i++ ) {
memset( DestPtr1, SrcPtr1[0], PlaneBorderWidth );
memset( DestPtr2, SrcPtr2[0], PlaneBorderWidth );
SrcPtr1 += PlaneStride;
SrcPtr2 += PlaneStride;
DestPtr1 += PlaneStride;
DestPtr2 += PlaneStride;
}
}
void UpdateUMVBorder( PB_INSTANCE *pbi,
unsigned char * DestReconPtr ) {
ogg_uint32_t PlaneFragOffset;
/* Y plane */
PlaneFragOffset = 0;
UpdateUMV_VBorders( pbi, DestReconPtr, PlaneFragOffset );
UpdateUMV_HBorders( pbi, DestReconPtr, PlaneFragOffset );
/* Then the U and V Planes */
PlaneFragOffset = pbi->YPlaneFragments;
UpdateUMV_VBorders( pbi, DestReconPtr, PlaneFragOffset );
UpdateUMV_HBorders( pbi, DestReconPtr, PlaneFragOffset );
PlaneFragOffset = pbi->YPlaneFragments + pbi->UVPlaneFragments;
UpdateUMV_VBorders( pbi, DestReconPtr, PlaneFragOffset );
UpdateUMV_HBorders( pbi, DestReconPtr, PlaneFragOffset );
}
static void CopyRecon( PB_INSTANCE *pbi, unsigned char * DestReconPtr,
unsigned char * SrcReconPtr ) {
ogg_uint32_t i;
ogg_uint32_t PlaneLineStep; /* Pixels per line */
ogg_uint32_t PixelIndex;
unsigned char *SrcPtr; /* Pointer to line of source image data */
unsigned char *DestPtr; /* Pointer to line of destination image data */
/* Copy over only updated blocks.*/
/* First Y plane */
PlaneLineStep = pbi->YStride;
for ( i = 0; i < pbi->YPlaneFragments; i++ ) {
if ( pbi->display_fragments[i] ) {
PixelIndex = pbi->recon_pixel_index_table[i];
SrcPtr = &SrcReconPtr[ PixelIndex ];
DestPtr = &DestReconPtr[ PixelIndex ];
dsp_copy8x8 (pbi->dsp, SrcPtr, DestPtr, PlaneLineStep);
}
}
/* Then U and V */
PlaneLineStep = pbi->UVStride;
for ( i = pbi->YPlaneFragments; i < pbi->UnitFragments; i++ ) {
if ( pbi->display_fragments[i] ) {
PixelIndex = pbi->recon_pixel_index_table[i];
SrcPtr = &SrcReconPtr[ PixelIndex ];
DestPtr = &DestReconPtr[ PixelIndex ];
dsp_copy8x8 (pbi->dsp, SrcPtr, DestPtr, PlaneLineStep);
}
}
}
static void CopyNotRecon( PB_INSTANCE *pbi, unsigned char * DestReconPtr,
unsigned char * SrcReconPtr ) {
ogg_uint32_t i;
ogg_uint32_t PlaneLineStep; /* Pixels per line */
ogg_uint32_t PixelIndex;
unsigned char *SrcPtr; /* Pointer to line of source image data */
unsigned char *DestPtr; /* Pointer to line of destination image data*/
/* Copy over only updated blocks. */
/* First Y plane */
PlaneLineStep = pbi->YStride;
for ( i = 0; i < pbi->YPlaneFragments; i++ ) {
if ( !pbi->display_fragments[i] ) {
PixelIndex = pbi->recon_pixel_index_table[i];
SrcPtr = &SrcReconPtr[ PixelIndex ];
DestPtr = &DestReconPtr[ PixelIndex ];
dsp_copy8x8 (pbi->dsp, SrcPtr, DestPtr, PlaneLineStep);
}
}
/* Then U and V */
PlaneLineStep = pbi->UVStride;
for ( i = pbi->YPlaneFragments; i < pbi->UnitFragments; i++ ) {
if ( !pbi->display_fragments[i] ) {
PixelIndex = pbi->recon_pixel_index_table[i];
SrcPtr = &SrcReconPtr[ PixelIndex ];
DestPtr = &DestReconPtr[ PixelIndex ];
dsp_copy8x8 (pbi->dsp, SrcPtr, DestPtr, PlaneLineStep);
}
}
}
void ExpandToken( Q_LIST_ENTRY * ExpandedBlock,
unsigned char * CoeffIndex, ogg_uint32_t Token,
ogg_int32_t ExtraBits ){
/* Is the token is a combination run and value token. */
if ( Token >= DCT_RUN_CATEGORY1 ){
/* Expand the token and additional bits to a zero run length and
data value. */
if ( Token < DCT_RUN_CATEGORY2 ) {
/* Decoding method depends on token */
if ( Token < DCT_RUN_CATEGORY1B ) {
/* Step on by the zero run length */
*CoeffIndex += (unsigned char)((Token - DCT_RUN_CATEGORY1) + 1);
/* The extra bit determines the sign. */
if ( ExtraBits & 0x01 )
ExpandedBlock[*CoeffIndex] = -1;
else
ExpandedBlock[*CoeffIndex] = 1;
} else if ( Token == DCT_RUN_CATEGORY1B ) {
/* Bits 0-1 determines the zero run length */
*CoeffIndex += (6 + (ExtraBits & 0x03));
/* Bit 2 determines the sign */
if ( ExtraBits & 0x04 )
ExpandedBlock[*CoeffIndex] = -1;
else
ExpandedBlock[*CoeffIndex] = 1;
}else{
/* Bits 0-2 determines the zero run length */
*CoeffIndex += (10 + (ExtraBits & 0x07));
/* Bit 3 determines the sign */
if ( ExtraBits & 0x08 )
ExpandedBlock[*CoeffIndex] = -1;
else
ExpandedBlock[*CoeffIndex] = 1;
}
}else{
/* If token == DCT_RUN_CATEGORY2 we have a single 0 followed by
a value */
if ( Token == DCT_RUN_CATEGORY2 ){
/* Step on by the zero run length */
*CoeffIndex += 1;
/* Bit 1 determines sign, bit 0 the value */
if ( ExtraBits & 0x02 )
ExpandedBlock[*CoeffIndex] = -(2 + (ExtraBits & 0x01));
else
ExpandedBlock[*CoeffIndex] = 2 + (ExtraBits & 0x01);
}else{
/* else we have 2->3 zeros followed by a value */
/* Bit 0 determines the zero run length */
*CoeffIndex += 2 + (ExtraBits & 0x01);
/* Bit 2 determines the sign, bit 1 the value */
if ( ExtraBits & 0x04 )
ExpandedBlock[*CoeffIndex] = -(2 + ((ExtraBits & 0x02) >> 1));
else
ExpandedBlock[*CoeffIndex] = 2 + ((ExtraBits & 0x02) >> 1);
}
}
/* Step on over value */
*CoeffIndex += 1;
} else if ( Token == DCT_SHORT_ZRL_TOKEN ) {
/* Token is a ZRL token so step on by the appropriate number of zeros */
*CoeffIndex += ExtraBits + 1;
} else if ( Token == DCT_ZRL_TOKEN ) {
/* Token is a ZRL token so step on by the appropriate number of zeros */
*CoeffIndex += ExtraBits + 1;
} else if ( Token < LOW_VAL_TOKENS ) {
/* Token is a small single value token. */
switch ( Token ) {
case ONE_TOKEN:
ExpandedBlock[*CoeffIndex] = 1;
break;
case MINUS_ONE_TOKEN:
ExpandedBlock[*CoeffIndex] = -1;
break;
case TWO_TOKEN:
ExpandedBlock[*CoeffIndex] = 2;
break;
case MINUS_TWO_TOKEN:
ExpandedBlock[*CoeffIndex] = -2;
break;
}
/* Step on the coefficient index. */
*CoeffIndex += 1;
}else{
/* Token is a larger single value token */
/* Expand the token and additional bits to a data value. */
if ( Token < DCT_VAL_CATEGORY3 ) {
/* Offset from LOW_VAL_TOKENS determines value */
Token = Token - LOW_VAL_TOKENS;
/* Extra bit determines sign */
if ( ExtraBits )
ExpandedBlock[*CoeffIndex] =
-((Q_LIST_ENTRY)(Token + DCT_VAL_CAT2_MIN));
else
ExpandedBlock[*CoeffIndex] =
(Q_LIST_ENTRY)(Token + DCT_VAL_CAT2_MIN);
} else if ( Token == DCT_VAL_CATEGORY3 ) {
/* Bit 1 determines sign, Bit 0 the value */
if ( ExtraBits & 0x02 )
ExpandedBlock[*CoeffIndex] = -(DCT_VAL_CAT3_MIN + (ExtraBits & 0x01));
else
ExpandedBlock[*CoeffIndex] = DCT_VAL_CAT3_MIN + (ExtraBits & 0x01);
} else if ( Token == DCT_VAL_CATEGORY4 ) {
/* Bit 2 determines sign, Bit 0-1 the value */
if ( ExtraBits & 0x04 )
ExpandedBlock[*CoeffIndex] = -(DCT_VAL_CAT4_MIN + (ExtraBits & 0x03));
else
ExpandedBlock[*CoeffIndex] = DCT_VAL_CAT4_MIN + (ExtraBits & 0x03);
} else if ( Token == DCT_VAL_CATEGORY5 ) {
/* Bit 3 determines sign, Bit 0-2 the value */
if ( ExtraBits & 0x08 )
ExpandedBlock[*CoeffIndex] = -(DCT_VAL_CAT5_MIN + (ExtraBits & 0x07));
else
ExpandedBlock[*CoeffIndex] = DCT_VAL_CAT5_MIN + (ExtraBits & 0x07);
} else if ( Token == DCT_VAL_CATEGORY6 ) {
/* Bit 4 determines sign, Bit 0-3 the value */
if ( ExtraBits & 0x10 )
ExpandedBlock[*CoeffIndex] = -(DCT_VAL_CAT6_MIN + (ExtraBits & 0x0F));
else
ExpandedBlock[*CoeffIndex] = DCT_VAL_CAT6_MIN + (ExtraBits & 0x0F);
} else if ( Token == DCT_VAL_CATEGORY7 ) {
/* Bit 5 determines sign, Bit 0-4 the value */
if ( ExtraBits & 0x20 )
ExpandedBlock[*CoeffIndex] = -(DCT_VAL_CAT7_MIN + (ExtraBits & 0x1F));
else
ExpandedBlock[*CoeffIndex] = DCT_VAL_CAT7_MIN + (ExtraBits & 0x1F);
} else if ( Token == DCT_VAL_CATEGORY8 ) {
/* Bit 9 determines sign, Bit 0-8 the value */
if ( ExtraBits & 0x200 )
ExpandedBlock[*CoeffIndex] = -(DCT_VAL_CAT8_MIN + (ExtraBits & 0x1FF));
else
ExpandedBlock[*CoeffIndex] = DCT_VAL_CAT8_MIN + (ExtraBits & 0x1FF);
}
/* Step on the coefficient index. */
*CoeffIndex += 1;
}
}
void ClearDownQFragData(PB_INSTANCE *pbi){
ogg_int32_t i;
Q_LIST_ENTRY * QFragPtr;
for ( i = 0; i < pbi->CodedBlockIndex; i++ ) {
/* Get the linear index for the current fragment. */
QFragPtr = pbi->QFragData[pbi->CodedBlockList[i]];
memset(QFragPtr, 0, 64*sizeof(Q_LIST_ENTRY));
}
}
static void loop_filter_h(unsigned char * PixelPtr,
ogg_int32_t LineLength,
ogg_int16_t *BoundingValuePtr){
ogg_int32_t j;
ogg_int32_t FiltVal;
PixelPtr-=2;
for ( j = 0; j < 8; j++ ){
FiltVal =
( PixelPtr[0] ) -
( PixelPtr[1] * 3 ) +
( PixelPtr[2] * 3 ) -
( PixelPtr[3] );
FiltVal = *(BoundingValuePtr+((FiltVal + 4) >> 3));
PixelPtr[1] = clamp255(PixelPtr[1] + FiltVal);
PixelPtr[2] = clamp255(PixelPtr[2] - FiltVal);
PixelPtr += LineLength;
}
}
static void loop_filter_v(unsigned char * PixelPtr,
ogg_int32_t LineLength,
ogg_int16_t *BoundingValuePtr){
ogg_int32_t j;
ogg_int32_t FiltVal;
PixelPtr -= 2*LineLength;
for ( j = 0; j < 8; j++ ) {
FiltVal = ( (ogg_int32_t)PixelPtr[0] ) -
( (ogg_int32_t)PixelPtr[LineLength] * 3 ) +
( (ogg_int32_t)PixelPtr[2 * LineLength] * 3 ) -
( (ogg_int32_t)PixelPtr[3 * LineLength] );
FiltVal = *(BoundingValuePtr+((FiltVal + 4) >> 3));
PixelPtr[LineLength] = clamp255(PixelPtr[LineLength] + FiltVal);
PixelPtr[2 * LineLength] = clamp255(PixelPtr[2*LineLength] - FiltVal);
PixelPtr ++;
}
}
static void LoopFilter__c(PB_INSTANCE *pbi, int FLimit){
int j;
ogg_int16_t BoundingValues[256];
ogg_int16_t *bvp = BoundingValues+127;
unsigned char *cp = pbi->display_fragments;
ogg_uint32_t *bp = pbi->recon_pixel_index_table;
if ( FLimit == 0 ) return;
SetupBoundingValueArray_Generic(BoundingValues, FLimit);
for ( j = 0; j < 3 ; j++){
ogg_uint32_t *bp_begin = bp;
ogg_uint32_t *bp_end;
int stride;
int h;
switch(j) {
case 0: /* y */
bp_end = bp + pbi->YPlaneFragments;
h = pbi->HFragments;
stride = pbi->YStride;
break;
default: /* u,v, 4:20 specific */
bp_end = bp + pbi->UVPlaneFragments;
h = pbi->HFragments >> 1;
stride = pbi->UVStride;
break;
}
while(bp<bp_end){
ogg_uint32_t *bp_left = bp;
ogg_uint32_t *bp_right = bp + h;
while(bp<bp_right){
if(cp[0]){
if(bp>bp_left)
loop_filter_h(&pbi->LastFrameRecon[bp[0]],stride,bvp);
if(bp_left>bp_begin)
loop_filter_v(&pbi->LastFrameRecon[bp[0]],stride,bvp);
if(bp+1<bp_right && !cp[1])
loop_filter_h(&pbi->LastFrameRecon[bp[0]]+8,stride,bvp);
if(bp+h<bp_end && !cp[h])
loop_filter_v(&pbi->LastFrameRecon[bp[h]],stride,bvp);
}
bp++;
cp++;
}
}
}
}
void ReconRefFrames (PB_INSTANCE *pbi){
ogg_int32_t i;
unsigned char *SwapReconBuffersTemp;
/* predictor multiplier up-left, up, up-right,left, shift
Entries are packed in the order L, UL, U, UR, with missing entries
moved to the end (before the shift parameters). */
static const ogg_int16_t pc[16][6]={
{0,0,0,0,0,0},
{1,0,0,0,0,0}, /* PL */
{1,0,0,0,0,0}, /* PUL */
{1,0,0,0,0,0}, /* PUL|PL */
{1,0,0,0,0,0}, /* PU */
{1,1,0,0,1,1}, /* PU|PL */
{0,1,0,0,0,0}, /* PU|PUL */
{29,-26,29,0,5,31}, /* PU|PUL|PL */
{1,0,0,0,0,0}, /* PUR */
{75,53,0,0,7,127}, /* PUR|PL */
{1,1,0,0,1,1}, /* PUR|PUL */
{75,0,53,0,7,127}, /* PUR|PUL|PL */
{1,0,0,0,0,0}, /* PUR|PU */
{75,0,53,0,7,127}, /* PUR|PU|PL */
{3,10,3,0,4,15}, /* PUR|PU|PUL */
{29,-26,29,0,5,31} /* PUR|PU|PUL|PL */
};
/* boundary case bit masks. */
static const int bc_mask[8]={
/* normal case no boundary condition */
PUR|PU|PUL|PL,
/* left column */
PUR|PU,
/* top row */
PL,
/* top row, left column */
0,
/* right column */
PU|PUL|PL,
/* right and left column */
PU,
/* top row, right column */
PL,
/* top row, right and left column */
0
};
/* value left value up-left, value up, value up-right, missing
values skipped. */
int v[4];
/* fragment number left, up-left, up, up-right */
int fn[4];
/* predictor count. */
int pcount;
short wpc;
static const short Mode2Frame[] = {
1, /* CODE_INTER_NO_MV 0 => Encoded diff from same MB last frame */
0, /* CODE_INTRA 1 => DCT Encoded Block */
1, /* CODE_INTER_PLUS_MV 2 => Encoded diff from included MV MB last frame */
1, /* CODE_INTER_LAST_MV 3 => Encoded diff from MRU MV MB last frame */
1, /* CODE_INTER_PRIOR_MV 4 => Encoded diff from included 4 separate MV blocks */
2, /* CODE_USING_GOLDEN 5 => Encoded diff from same MB golden frame */
2, /* CODE_GOLDEN_MV 6 => Encoded diff from included MV MB golden frame */
1 /* CODE_INTER_FOUR_MV 7 => Encoded diff from included 4 separate MV blocks */
};
short Last[3];
short PredictedDC;
int FragsAcross=pbi->HFragments;
int FromFragment,ToFragment;
int FragsDown = pbi->VFragments;
int WhichFrame;
int WhichCase;
int j,k,m,n;
void (*ExpandBlockA) ( PB_INSTANCE *pbi, ogg_int32_t FragmentNumber );
if ( pbi->FrameType == KEY_FRAME )
ExpandBlockA=ExpandKFBlock;
else
ExpandBlockA=ExpandBlock;
/* for y,u,v */
for ( j = 0; j < 3 ; j++) {
/* pick which fragments based on Y, U, V */
switch(j){
case 0: /* y */
FromFragment = 0;
ToFragment = pbi->YPlaneFragments;
FragsAcross = pbi->HFragments;
FragsDown = pbi->VFragments;
break;
case 1: /* u */
FromFragment = pbi->YPlaneFragments;
ToFragment = pbi->YPlaneFragments + pbi->UVPlaneFragments ;
FragsAcross = pbi->HFragments >> 1;
FragsDown = pbi->VFragments >> 1;
break;
/*case 2: v */
default:
FromFragment = pbi->YPlaneFragments + pbi->UVPlaneFragments;
ToFragment = pbi->YPlaneFragments + (2 * pbi->UVPlaneFragments) ;
FragsAcross = pbi->HFragments >> 1;
FragsDown = pbi->VFragments >> 1;
break;
}
/* initialize our array of last used DC Components */
for(k=0;k<3;k++)
Last[k]=0;
i=FromFragment;
/* do prediction on all of Y, U or V */
for ( m = 0 ; m < FragsDown ; m++) {
for ( n = 0 ; n < FragsAcross ; n++, i++){
/* only do 2 prediction if fragment coded and on non intra or
if all fragments are intra */
if( pbi->display_fragments[i] || (pbi->FrameType == KEY_FRAME) ){
/* Type of Fragment */
WhichFrame = Mode2Frame[pbi->FragCodingMethod[i]];
/* Check Borderline Cases */
WhichCase = (n==0) + ((m==0) << 1) + ((n+1 == FragsAcross) << 2);
fn[0]=i-1;
fn[1]=i-FragsAcross-1;
fn[2]=i-FragsAcross;
fn[3]=i-FragsAcross+1;
/* fragment valid for prediction use if coded and it comes
from same frame as the one we are predicting */
for(k=pcount=wpc=0; k<4; k++) {
int pflag;
pflag=1<<k;
if((bc_mask[WhichCase]&pflag) &&
pbi->display_fragments[fn[k]] &&
(Mode2Frame[pbi->FragCodingMethod[fn[k]]] == WhichFrame)){
v[pcount]=pbi->QFragData[fn[k]][0];
wpc|=pflag;
pcount++;
}
}
if(wpc==0){
/* fall back to the last coded fragment */
pbi->QFragData[i][0] += Last[WhichFrame];
}else{
/* don't do divide if divisor is 1 or 0 */
PredictedDC = pc[wpc][0]*v[0];
for(k=1; k<pcount; k++){
PredictedDC += pc[wpc][k]*v[k];
}
/* if we need to do a shift */
if(pc[wpc][4] != 0 ){
/* If negative add in the negative correction factor */
PredictedDC += (HIGHBITDUPPED(PredictedDC) & pc[wpc][5]);
/* Shift in lieu of a divide */
PredictedDC >>= pc[wpc][4];
}
/* check for outranging on the two predictors that can outrange */
if((wpc&(PU|PUL|PL)) == (PU|PUL|PL)){
if( abs(PredictedDC - v[2]) > 128) {
PredictedDC = v[2];
} else if( abs(PredictedDC - v[0]) > 128) {
PredictedDC = v[0];
} else if( abs(PredictedDC - v[1]) > 128) {
PredictedDC = v[1];
}
}
pbi->QFragData[i][0] += PredictedDC;
}
/* Save the last fragment coded for whatever frame we are
predicting from */
Last[WhichFrame] = pbi->QFragData[i][0];
/* Inverse DCT and reconstitute buffer in thisframe */
ExpandBlockA( pbi, i );
}
}
}
}
/* Copy the current reconstruction back to the last frame recon buffer. */
if(pbi->CodedBlockIndex > (ogg_int32_t) (pbi->UnitFragments >> 1)){
SwapReconBuffersTemp = pbi->ThisFrameRecon;
pbi->ThisFrameRecon = pbi->LastFrameRecon;
pbi->LastFrameRecon = SwapReconBuffersTemp;
CopyNotRecon( pbi, pbi->LastFrameRecon, pbi->ThisFrameRecon );
}else{
CopyRecon( pbi, pbi->LastFrameRecon, pbi->ThisFrameRecon );
}
/* Apply a loop filter to edge pixels of updated blocks */
dsp_LoopFilter(pbi->dsp, pbi, pbi->quant_info.loop_filter_limits[pbi->FrameQIndex]);
/* We may need to update the UMV border */
UpdateUMVBorder(pbi, pbi->LastFrameRecon);
/* Reconstruct the golden frame if necessary.
For VFW codec only on key frames */
if ( pbi->FrameType == KEY_FRAME ){
CopyRecon( pbi, pbi->GoldenFrame, pbi->LastFrameRecon );
/* We may need to update the UMV border */
UpdateUMVBorder(pbi, pbi->GoldenFrame);
}
}
void dsp_dct_decode_init (DspFunctions *funcs, ogg_uint32_t cpu_flags)
{
funcs->LoopFilter = LoopFilter__c;
#if defined(USE_ASM)
// Todo: Port the dct for MSC one day.
#if !defined (_MSC_VER)
if (cpu_flags & OC_CPU_X86_MMX) {
dsp_mmx_dct_decode_init(funcs);
}
#endif
#endif
}

469
Engine/lib/libtheora/lib/enc/dct_encode.c Normal file
View file

@ -0,0 +1,469 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: dct_encode.c 15153 2008-08-04 18:37:55Z tterribe $
********************************************************************/
#include <stdlib.h>
#include "codec_internal.h"
#include "dsp.h"
#include "quant_lookup.h"
static int ModeUsesMC[MAX_MODES] = { 0, 0, 1, 1, 1, 0, 1, 1 };
static unsigned char TokenizeDctValue (ogg_int16_t DataValue,
ogg_uint32_t * TokenListPtr ){
unsigned char tokens_added = 0;
ogg_uint32_t AbsDataVal = abs( (ogg_int32_t)DataValue );
/* Values are tokenised as category value and a number of additional
bits that define the position within the category. */
if ( DataValue == 0 ) return 0;
if ( AbsDataVal == 1 ){
if ( DataValue == 1 )
TokenListPtr[0] = ONE_TOKEN;
else
TokenListPtr[0] = MINUS_ONE_TOKEN;
tokens_added = 1;
} else if ( AbsDataVal == 2 ) {
if ( DataValue == 2 )
TokenListPtr[0] = TWO_TOKEN;
else
TokenListPtr[0] = MINUS_TWO_TOKEN;
tokens_added = 1;
} else if ( AbsDataVal <= MAX_SINGLE_TOKEN_VALUE ) {
TokenListPtr[0] = LOW_VAL_TOKENS + (AbsDataVal - DCT_VAL_CAT2_MIN);
if ( DataValue > 0 )
TokenListPtr[1] = 0;
else
TokenListPtr[1] = 1;
tokens_added = 2;
} else if ( AbsDataVal <= 8 ) {
/* Bit 1 determines sign, Bit 0 the value */
TokenListPtr[0] = DCT_VAL_CATEGORY3;
if ( DataValue > 0 )
TokenListPtr[1] = (AbsDataVal - DCT_VAL_CAT3_MIN);
else
TokenListPtr[1] = (0x02) + (AbsDataVal - DCT_VAL_CAT3_MIN);
tokens_added = 2;
} else if ( AbsDataVal <= 12 ) {
/* Bit 2 determines sign, Bit 0-2 the value */
TokenListPtr[0] = DCT_VAL_CATEGORY4;
if ( DataValue > 0 )
TokenListPtr[1] = (AbsDataVal - DCT_VAL_CAT4_MIN);
else
TokenListPtr[1] = (0x04) + (AbsDataVal - DCT_VAL_CAT4_MIN);
tokens_added = 2;
} else if ( AbsDataVal <= 20 ) {
/* Bit 3 determines sign, Bit 0-2 the value */
TokenListPtr[0] = DCT_VAL_CATEGORY5;
if ( DataValue > 0 )
TokenListPtr[1] = (AbsDataVal - DCT_VAL_CAT5_MIN);
else
TokenListPtr[1] = (0x08) + (AbsDataVal - DCT_VAL_CAT5_MIN);
tokens_added = 2;
} else if ( AbsDataVal <= 36 ) {
/* Bit 4 determines sign, Bit 0-3 the value */
TokenListPtr[0] = DCT_VAL_CATEGORY6;
if ( DataValue > 0 )
TokenListPtr[1] = (AbsDataVal - DCT_VAL_CAT6_MIN);
else
TokenListPtr[1] = (0x010) + (AbsDataVal - DCT_VAL_CAT6_MIN);
tokens_added = 2;
} else if ( AbsDataVal <= 68 ) {
/* Bit 5 determines sign, Bit 0-4 the value */
TokenListPtr[0] = DCT_VAL_CATEGORY7;
if ( DataValue > 0 )
TokenListPtr[1] = (AbsDataVal - DCT_VAL_CAT7_MIN);
else
TokenListPtr[1] = (0x20) + (AbsDataVal - DCT_VAL_CAT7_MIN);
tokens_added = 2;
} else if ( AbsDataVal <= 511 ) {
/* Bit 9 determines sign, Bit 0-8 the value */
TokenListPtr[0] = DCT_VAL_CATEGORY8;
if ( DataValue > 0 )
TokenListPtr[1] = (AbsDataVal - DCT_VAL_CAT8_MIN);
else
TokenListPtr[1] = (0x200) + (AbsDataVal - DCT_VAL_CAT8_MIN);
tokens_added = 2;
} else {
TokenListPtr[0] = DCT_VAL_CATEGORY8;
if ( DataValue > 0 )
TokenListPtr[1] = (511 - DCT_VAL_CAT8_MIN);
else
TokenListPtr[1] = (0x200) + (511 - DCT_VAL_CAT8_MIN);
tokens_added = 2;
}
/* Return the total number of tokens added */
return tokens_added;
}
static unsigned char TokenizeDctRunValue (unsigned char RunLength,
ogg_int16_t DataValue,
ogg_uint32_t * TokenListPtr ){
unsigned char tokens_added = 0;
ogg_uint32_t AbsDataVal = abs( (ogg_int32_t)DataValue );
/* Values are tokenised as category value and a number of additional
bits that define the category. */
if ( DataValue == 0 ) return 0;
if ( AbsDataVal == 1 ) {
/* Zero runs of 1-5 */
if ( RunLength <= 5 ) {
TokenListPtr[0] = DCT_RUN_CATEGORY1 + (RunLength - 1);
if ( DataValue > 0 )
TokenListPtr[1] = 0;
else
TokenListPtr[1] = 1;
} else if ( RunLength <= 9 ) {
/* Zero runs of 6-9 */
TokenListPtr[0] = DCT_RUN_CATEGORY1B;
if ( DataValue > 0 )
TokenListPtr[1] = (RunLength - 6);
else
TokenListPtr[1] = 0x04 + (RunLength - 6);
} else {
/* Zero runs of 10-17 */
TokenListPtr[0] = DCT_RUN_CATEGORY1C;
if ( DataValue > 0 )
TokenListPtr[1] = (RunLength - 10);
else
TokenListPtr[1] = 0x08 + (RunLength - 10);
}
tokens_added = 2;
} else if ( AbsDataVal <= 3 ) {
if ( RunLength == 1 ) {
TokenListPtr[0] = DCT_RUN_CATEGORY2;
/* Extra bits token bit 1 indicates sign, bit 0 indicates value */
if ( DataValue > 0 )
TokenListPtr[1] = (AbsDataVal - 2);
else
TokenListPtr[1] = (0x02) + (AbsDataVal - 2);
tokens_added = 2;
}else{
TokenListPtr[0] = DCT_RUN_CATEGORY2 + 1;
/* Extra bits token. */
/* bit 2 indicates sign, bit 1 indicates value, bit 0 indicates
run length */
if ( DataValue > 0 )
TokenListPtr[1] = ((AbsDataVal - 2) << 1) + (RunLength - 2);
else
TokenListPtr[1] = (0x04) + ((AbsDataVal - 2) << 1) + (RunLength - 2);
tokens_added = 2;
}
} else {
tokens_added = 2; /* ERROR */
/*IssueWarning( "Bad Input to TokenizeDctRunValue" );*/
}
/* Return the total number of tokens added */
return tokens_added;
}
static unsigned char TokenizeDctBlock (ogg_int16_t * RawData,
ogg_uint32_t * TokenListPtr ) {
ogg_uint32_t i;
unsigned char run_count;
unsigned char token_count = 0; /* Number of tokens crated. */
ogg_uint32_t AbsData;
/* Tokenize the block */
for( i = 0; i < BLOCK_SIZE; i++ ){
run_count = 0;
/* Look for a zero run. */
/* NOTE the use of & instead of && which is faster (and
equivalent) in this instance. */
/* NO, NO IT ISN'T --Monty */
while( (i < BLOCK_SIZE) && (!RawData[i]) ){
run_count++;
i++;
}
/* If we have reached the end of the block then code EOB */
if ( i == BLOCK_SIZE ){
TokenListPtr[token_count] = DCT_EOB_TOKEN;
token_count++;
}else{
/* If we have a short zero run followed by a low data value code
the two as a composite token. */
if ( run_count ){
AbsData = abs(RawData[i]);
if ( ((AbsData == 1) && (run_count <= 17)) ||
((AbsData <= 3) && (run_count <= 3)) ) {
/* Tokenise the run and subsequent value combination value */
token_count += TokenizeDctRunValue( run_count,
RawData[i],
&TokenListPtr[token_count] );
}else{
/* Else if we have a long non-EOB run or a run followed by a
value token > MAX_RUN_VAL then code the run and token
seperately */
if ( run_count <= 8 )
TokenListPtr[token_count] = DCT_SHORT_ZRL_TOKEN;
else
TokenListPtr[token_count] = DCT_ZRL_TOKEN;
token_count++;
TokenListPtr[token_count] = run_count - 1;
token_count++;
/* Now tokenize the value */
token_count += TokenizeDctValue( RawData[i],
&TokenListPtr[token_count] );
}
}else{
/* Else there was NO zero run. */
/* Tokenise the value */
token_count += TokenizeDctValue( RawData[i],
&TokenListPtr[token_count] );
}
}
}
/* Return the total number of tokens (including additional bits
tokens) used. */
return token_count;
}
ogg_uint32_t DPCMTokenizeBlock (CP_INSTANCE *cpi,
ogg_int32_t FragIndex){
ogg_uint32_t token_count;
if ( cpi->pb.FrameType == KEY_FRAME ){
/* Key frame so code block in INTRA mode. */
cpi->pb.CodingMode = CODE_INTRA;
}else{
/* Get Motion vector and mode for this block. */
cpi->pb.CodingMode = cpi->pb.FragCodingMethod[FragIndex];
}
/* Tokenise the dct data. */
token_count = TokenizeDctBlock( cpi->pb.QFragData[FragIndex],
cpi->pb.TokenList[FragIndex] );
cpi->FragTokenCounts[FragIndex] = token_count;
cpi->TotTokenCount += token_count;
/* Return number of pixels coded (i.e. 8x8). */
return BLOCK_SIZE;
}
static int AllZeroDctData( Q_LIST_ENTRY * QuantList ){
ogg_uint32_t i;
for ( i = 0; i < 64; i ++ )
if ( QuantList[i] != 0 )
return 0;
return 1;
}
static void MotionBlockDifference (CP_INSTANCE * cpi, unsigned char * FiltPtr,
ogg_int16_t *DctInputPtr, ogg_int32_t MvDevisor,
unsigned char* old_ptr1, unsigned char* new_ptr1,
ogg_uint32_t FragIndex,ogg_uint32_t PixelsPerLine,
ogg_uint32_t ReconPixelsPerLine) {
ogg_int32_t MvShift;
ogg_int32_t MvModMask;
ogg_int32_t AbsRefOffset;
ogg_int32_t AbsXOffset;
ogg_int32_t AbsYOffset;
ogg_int32_t MVOffset; /* Baseline motion vector offset */
ogg_int32_t ReconPtr2Offset; /* Offset for second reconstruction in
half pixel MC */
unsigned char *ReconPtr1; /* DCT reconstructed image pointers */
unsigned char *ReconPtr2; /* Pointer used in half pixel MC */
switch(MvDevisor) {
case 2:
MvShift = 1;
MvModMask = 1;
break;
case 4:
MvShift = 2;
MvModMask = 3;
break;
default:
break;
}
cpi->MVector.x = cpi->pb.FragMVect[FragIndex].x;
cpi->MVector.y = cpi->pb.FragMVect[FragIndex].y;
/* Set up the baseline offset for the motion vector. */
MVOffset = ((cpi->MVector.y / MvDevisor) * ReconPixelsPerLine) +
(cpi->MVector.x / MvDevisor);
/* Work out the offset of the second reference position for 1/2
pixel interpolation. For the U and V planes the MV specifies 1/4
pixel accuracy. This is adjusted to 1/2 pixel as follows ( 0->0,
1/4->1/2, 1/2->1/2, 3/4->1/2 ). */
ReconPtr2Offset = 0;
AbsXOffset = cpi->MVector.x % MvDevisor;
AbsYOffset = cpi->MVector.y % MvDevisor;
if ( AbsXOffset ) {
if ( cpi->MVector.x > 0 )
ReconPtr2Offset += 1;
else
ReconPtr2Offset -= 1;
}
if ( AbsYOffset ) {
if ( cpi->MVector.y > 0 )
ReconPtr2Offset += ReconPixelsPerLine;
else
ReconPtr2Offset -= ReconPixelsPerLine;
}
if ( cpi->pb.CodingMode==CODE_GOLDEN_MV ) {
ReconPtr1 = &cpi->
pb.GoldenFrame[cpi->pb.recon_pixel_index_table[FragIndex]];
} else {
ReconPtr1 = &cpi->
pb.LastFrameRecon[cpi->pb.recon_pixel_index_table[FragIndex]];
}
ReconPtr1 += MVOffset;
ReconPtr2 = ReconPtr1 + ReconPtr2Offset;
AbsRefOffset = abs((int)(ReconPtr1 - ReconPtr2));
/* Is the MV offset exactly pixel alligned */
if ( AbsRefOffset == 0 ){
dsp_sub8x8(cpi->dsp, FiltPtr, ReconPtr1, DctInputPtr,
PixelsPerLine, ReconPixelsPerLine);
dsp_copy8x8 (cpi->dsp, new_ptr1, old_ptr1, PixelsPerLine);
} else {
/* Fractional pixel MVs. */
/* Note that we only use two pixel values even for the diagonal */
dsp_sub8x8avg2(cpi->dsp, FiltPtr, ReconPtr1,ReconPtr2,DctInputPtr,
PixelsPerLine, ReconPixelsPerLine);
dsp_copy8x8 (cpi->dsp, new_ptr1, old_ptr1, PixelsPerLine);
}
}
void TransformQuantizeBlock (CP_INSTANCE *cpi, ogg_int32_t FragIndex,
ogg_uint32_t PixelsPerLine) {
unsigned char *new_ptr1; /* Pointers into current frame */
unsigned char *old_ptr1; /* Pointers into old frame */
unsigned char *FiltPtr; /* Pointers to srf filtered pixels */
ogg_int16_t *DctInputPtr; /* Pointer into buffer containing input to DCT */
int LeftEdge; /* Flag if block at left edge of component */
ogg_uint32_t ReconPixelsPerLine; /* Line length for recon buffers. */
unsigned char *ReconPtr1; /* DCT reconstructed image pointers */
ogg_int32_t MvDevisor; /* Defines MV resolution (2 = 1/2
pixel for Y or 4 = 1/4 for UV) */
new_ptr1 = &cpi->yuv1ptr[cpi->pb.pixel_index_table[FragIndex]];
old_ptr1 = &cpi->yuv0ptr[cpi->pb.pixel_index_table[FragIndex]];
DctInputPtr = cpi->DCTDataBuffer;
/* Set plane specific values */
if (FragIndex < (ogg_int32_t)cpi->pb.YPlaneFragments){
ReconPixelsPerLine = cpi->pb.YStride;
MvDevisor = 2; /* 1/2 pixel accuracy in Y */
}else{
ReconPixelsPerLine = cpi->pb.UVStride;
MvDevisor = 4; /* UV planes at 1/2 resolution of Y */
}
/* adjusted / filtered pointers */
FiltPtr = &cpi->ConvDestBuffer[cpi->pb.pixel_index_table[FragIndex]];
if ( cpi->pb.FrameType == KEY_FRAME ) {
/* Key frame so code block in INTRA mode. */
cpi->pb.CodingMode = CODE_INTRA;
}else{
/* Get Motion vector and mode for this block. */
cpi->pb.CodingMode = cpi->pb.FragCodingMethod[FragIndex];
}
/* Selection of Quantiser matrix and set other plane related values. */
if ( FragIndex < (ogg_int32_t)cpi->pb.YPlaneFragments ){
LeftEdge = !(FragIndex%cpi->pb.HFragments);
/* Select the appropriate Y quantiser matrix */
if ( cpi->pb.CodingMode == CODE_INTRA )
select_quantiser(&cpi->pb, BLOCK_Y);
else
select_quantiser(&cpi->pb, BLOCK_INTER_Y);
} else {
LeftEdge = !((FragIndex-cpi->pb.YPlaneFragments)%(cpi->pb.HFragments>>1));
if(FragIndex < (ogg_int32_t)cpi->pb.YPlaneFragments + (ogg_int32_t)cpi->pb.UVPlaneFragments) {
/* U plane */
if ( cpi->pb.CodingMode == CODE_INTRA )
select_quantiser(&cpi->pb, BLOCK_U);
else
select_quantiser(&cpi->pb, BLOCK_INTER_U);
} else {
/* V plane */
if ( cpi->pb.CodingMode == CODE_INTRA )
select_quantiser(&cpi->pb, BLOCK_V);
else
select_quantiser(&cpi->pb, BLOCK_INTER_V);
}
}
if ( ModeUsesMC[cpi->pb.CodingMode] ){
MotionBlockDifference(cpi, FiltPtr, DctInputPtr, MvDevisor,
old_ptr1, new_ptr1, FragIndex, PixelsPerLine,
ReconPixelsPerLine);
} else if ( (cpi->pb.CodingMode==CODE_INTER_NO_MV ) ||
( cpi->pb.CodingMode==CODE_USING_GOLDEN ) ) {
if ( cpi->pb.CodingMode==CODE_INTER_NO_MV ) {
ReconPtr1 = &cpi->
pb.LastFrameRecon[cpi->pb.recon_pixel_index_table[FragIndex]];
} else {
ReconPtr1 = &cpi->
pb.GoldenFrame[cpi->pb.recon_pixel_index_table[FragIndex]];
}
dsp_sub8x8(cpi->dsp, FiltPtr, ReconPtr1, DctInputPtr,
PixelsPerLine, ReconPixelsPerLine);
dsp_copy8x8 (cpi->dsp, new_ptr1, old_ptr1, PixelsPerLine);
} else if ( cpi->pb.CodingMode==CODE_INTRA ) {
dsp_sub8x8_128(cpi->dsp, FiltPtr, DctInputPtr, PixelsPerLine);
dsp_copy8x8 (cpi->dsp, new_ptr1, old_ptr1, PixelsPerLine);
}
/* Proceed to encode the data into the encode buffer if the encoder
is enabled. */
/* Perform a 2D DCT transform on the data. */
dsp_fdct_short(cpi->dsp, cpi->DCTDataBuffer, cpi->DCT_codes );
/* Quantize that transform data. */
quantize ( &cpi->pb, cpi->DCT_codes, cpi->pb.QFragData[FragIndex] );
if ( (cpi->pb.CodingMode == CODE_INTER_NO_MV) &&
( AllZeroDctData(cpi->pb.QFragData[FragIndex]) ) ) {
cpi->pb.display_fragments[FragIndex] = 0;
}
}

422
Engine/lib/libtheora/lib/enc/dsp.c Normal file
View file

@ -0,0 +1,422 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: dsp.c 15427 2008-10-21 02:36:19Z xiphmont $
********************************************************************/
#include <stdlib.h>
#include "codec_internal.h"
#include "../cpu.c"
#define DSP_OP_AVG(a,b) ((((int)(a)) + ((int)(b)))/2)
#define DSP_OP_DIFF(a,b) (((int)(a)) - ((int)(b)))
#define DSP_OP_ABS_DIFF(a,b) abs((((int)(a)) - ((int)(b))))
static void sub8x8__c (unsigned char *FiltPtr, unsigned char *ReconPtr,
ogg_int16_t *DctInputPtr, ogg_uint32_t PixelsPerLine,
ogg_uint32_t ReconPixelsPerLine) {
int i;
/* For each block row */
for (i=8; i; i--) {
DctInputPtr[0] = (ogg_int16_t) DSP_OP_DIFF (FiltPtr[0], ReconPtr[0]);
DctInputPtr[1] = (ogg_int16_t) DSP_OP_DIFF (FiltPtr[1], ReconPtr[1]);
DctInputPtr[2] = (ogg_int16_t) DSP_OP_DIFF (FiltPtr[2], ReconPtr[2]);
DctInputPtr[3] = (ogg_int16_t) DSP_OP_DIFF (FiltPtr[3], ReconPtr[3]);
DctInputPtr[4] = (ogg_int16_t) DSP_OP_DIFF (FiltPtr[4], ReconPtr[4]);
DctInputPtr[5] = (ogg_int16_t) DSP_OP_DIFF (FiltPtr[5], ReconPtr[5]);
DctInputPtr[6] = (ogg_int16_t) DSP_OP_DIFF (FiltPtr[6], ReconPtr[6]);
DctInputPtr[7] = (ogg_int16_t) DSP_OP_DIFF (FiltPtr[7], ReconPtr[7]);
/* Start next row */
FiltPtr += PixelsPerLine;
ReconPtr += ReconPixelsPerLine;
DctInputPtr += 8;
}
}
static void sub8x8_128__c (unsigned char *FiltPtr, ogg_int16_t *DctInputPtr,
ogg_uint32_t PixelsPerLine) {
int i;
/* For each block row */
for (i=8; i; i--) {
/* INTRA mode so code raw image data */
/* We convert the data to 8 bit signed (by subtracting 128) as
this reduces the internal precision requirments in the DCT
transform. */
DctInputPtr[0] = (ogg_int16_t) DSP_OP_DIFF (FiltPtr[0], 128);
DctInputPtr[1] = (ogg_int16_t) DSP_OP_DIFF (FiltPtr[1], 128);
DctInputPtr[2] = (ogg_int16_t) DSP_OP_DIFF (FiltPtr[2], 128);
DctInputPtr[3] = (ogg_int16_t) DSP_OP_DIFF (FiltPtr[3], 128);
DctInputPtr[4] = (ogg_int16_t) DSP_OP_DIFF (FiltPtr[4], 128);
DctInputPtr[5] = (ogg_int16_t) DSP_OP_DIFF (FiltPtr[5], 128);
DctInputPtr[6] = (ogg_int16_t) DSP_OP_DIFF (FiltPtr[6], 128);
DctInputPtr[7] = (ogg_int16_t) DSP_OP_DIFF (FiltPtr[7], 128);
/* Start next row */
FiltPtr += PixelsPerLine;
DctInputPtr += 8;
}
}
static void sub8x8avg2__c (unsigned char *FiltPtr, unsigned char *ReconPtr1,
unsigned char *ReconPtr2, ogg_int16_t *DctInputPtr,
ogg_uint32_t PixelsPerLine,
ogg_uint32_t ReconPixelsPerLine)
{
int i;
/* For each block row */
for (i=8; i; i--) {
DctInputPtr[0] = (ogg_int16_t) DSP_OP_DIFF (FiltPtr[0], DSP_OP_AVG (ReconPtr1[0], ReconPtr2[0]));
DctInputPtr[1] = (ogg_int16_t) DSP_OP_DIFF (FiltPtr[1], DSP_OP_AVG (ReconPtr1[1], ReconPtr2[1]));
DctInputPtr[2] = (ogg_int16_t) DSP_OP_DIFF (FiltPtr[2], DSP_OP_AVG (ReconPtr1[2], ReconPtr2[2]));
DctInputPtr[3] = (ogg_int16_t) DSP_OP_DIFF (FiltPtr[3], DSP_OP_AVG (ReconPtr1[3], ReconPtr2[3]));
DctInputPtr[4] = (ogg_int16_t) DSP_OP_DIFF (FiltPtr[4], DSP_OP_AVG (ReconPtr1[4], ReconPtr2[4]));
DctInputPtr[5] = (ogg_int16_t) DSP_OP_DIFF (FiltPtr[5], DSP_OP_AVG (ReconPtr1[5], ReconPtr2[5]));
DctInputPtr[6] = (ogg_int16_t) DSP_OP_DIFF (FiltPtr[6], DSP_OP_AVG (ReconPtr1[6], ReconPtr2[6]));
DctInputPtr[7] = (ogg_int16_t) DSP_OP_DIFF (FiltPtr[7], DSP_OP_AVG (ReconPtr1[7], ReconPtr2[7]));
/* Start next row */
FiltPtr += PixelsPerLine;
ReconPtr1 += ReconPixelsPerLine;
ReconPtr2 += ReconPixelsPerLine;
DctInputPtr += 8;
}
}
static ogg_uint32_t row_sad8__c (unsigned char *Src1, unsigned char *Src2)
{
ogg_uint32_t SadValue;
ogg_uint32_t SadValue1;
SadValue = DSP_OP_ABS_DIFF (Src1[0], Src2[0]) +
DSP_OP_ABS_DIFF (Src1[1], Src2[1]) +
DSP_OP_ABS_DIFF (Src1[2], Src2[2]) +
DSP_OP_ABS_DIFF (Src1[3], Src2[3]);
SadValue1 = DSP_OP_ABS_DIFF (Src1[4], Src2[4]) +
DSP_OP_ABS_DIFF (Src1[5], Src2[5]) +
DSP_OP_ABS_DIFF (Src1[6], Src2[6]) +
DSP_OP_ABS_DIFF (Src1[7], Src2[7]);
SadValue = ( SadValue > SadValue1 ) ? SadValue : SadValue1;
return SadValue;
}
static ogg_uint32_t col_sad8x8__c (unsigned char *Src1, unsigned char *Src2,
ogg_uint32_t stride)
{
ogg_uint32_t SadValue[8] = {0,0,0,0,0,0,0,0};
ogg_uint32_t SadValue2[8] = {0,0,0,0,0,0,0,0};
ogg_uint32_t MaxSad = 0;
ogg_uint32_t i;
for ( i = 0; i < 4; i++ ){
SadValue[0] += abs(Src1[0] - Src2[0]);
SadValue[1] += abs(Src1[1] - Src2[1]);
SadValue[2] += abs(Src1[2] - Src2[2]);
SadValue[3] += abs(Src1[3] - Src2[3]);
SadValue[4] += abs(Src1[4] - Src2[4]);
SadValue[5] += abs(Src1[5] - Src2[5]);
SadValue[6] += abs(Src1[6] - Src2[6]);
SadValue[7] += abs(Src1[7] - Src2[7]);
Src1 += stride;
Src2 += stride;
}
for ( i = 0; i < 4; i++ ){
SadValue2[0] += abs(Src1[0] - Src2[0]);
SadValue2[1] += abs(Src1[1] - Src2[1]);
SadValue2[2] += abs(Src1[2] - Src2[2]);
SadValue2[3] += abs(Src1[3] - Src2[3]);
SadValue2[4] += abs(Src1[4] - Src2[4]);
SadValue2[5] += abs(Src1[5] - Src2[5]);
SadValue2[6] += abs(Src1[6] - Src2[6]);
SadValue2[7] += abs(Src1[7] - Src2[7]);
Src1 += stride;
Src2 += stride;
}
for ( i = 0; i < 8; i++ ){
if ( SadValue[i] > MaxSad )
MaxSad = SadValue[i];
if ( SadValue2[i] > MaxSad )
MaxSad = SadValue2[i];
}
return MaxSad;
}
static ogg_uint32_t sad8x8__c (unsigned char *ptr1, ogg_uint32_t stride1,
unsigned char *ptr2, ogg_uint32_t stride2)
{
ogg_uint32_t i;
ogg_uint32_t sad = 0;
for (i=8; i; i--) {
sad += DSP_OP_ABS_DIFF(ptr1[0], ptr2[0]);
sad += DSP_OP_ABS_DIFF(ptr1[1], ptr2[1]);
sad += DSP_OP_ABS_DIFF(ptr1[2], ptr2[2]);
sad += DSP_OP_ABS_DIFF(ptr1[3], ptr2[3]);
sad += DSP_OP_ABS_DIFF(ptr1[4], ptr2[4]);
sad += DSP_OP_ABS_DIFF(ptr1[5], ptr2[5]);
sad += DSP_OP_ABS_DIFF(ptr1[6], ptr2[6]);
sad += DSP_OP_ABS_DIFF(ptr1[7], ptr2[7]);
/* Step to next row of block. */
ptr1 += stride1;
ptr2 += stride2;
}
return sad;
}
static ogg_uint32_t sad8x8_thres__c (unsigned char *ptr1, ogg_uint32_t stride1,
unsigned char *ptr2, ogg_uint32_t stride2,
ogg_uint32_t thres)
{
ogg_uint32_t i;
ogg_uint32_t sad = 0;
for (i=8; i; i--) {
sad += DSP_OP_ABS_DIFF(ptr1[0], ptr2[0]);
sad += DSP_OP_ABS_DIFF(ptr1[1], ptr2[1]);
sad += DSP_OP_ABS_DIFF(ptr1[2], ptr2[2]);
sad += DSP_OP_ABS_DIFF(ptr1[3], ptr2[3]);
sad += DSP_OP_ABS_DIFF(ptr1[4], ptr2[4]);
sad += DSP_OP_ABS_DIFF(ptr1[5], ptr2[5]);
sad += DSP_OP_ABS_DIFF(ptr1[6], ptr2[6]);
sad += DSP_OP_ABS_DIFF(ptr1[7], ptr2[7]);
if (sad > thres )
break;
/* Step to next row of block. */
ptr1 += stride1;
ptr2 += stride2;
}
return sad;
}
static ogg_uint32_t sad8x8_xy2_thres__c (unsigned char *SrcData, ogg_uint32_t SrcStride,
unsigned char *RefDataPtr1,
unsigned char *RefDataPtr2, ogg_uint32_t RefStride,
ogg_uint32_t thres)
{
ogg_uint32_t i;
ogg_uint32_t sad = 0;
for (i=8; i; i--) {
sad += DSP_OP_ABS_DIFF(SrcData[0], DSP_OP_AVG (RefDataPtr1[0], RefDataPtr2[0]));
sad += DSP_OP_ABS_DIFF(SrcData[1], DSP_OP_AVG (RefDataPtr1[1], RefDataPtr2[1]));
sad += DSP_OP_ABS_DIFF(SrcData[2], DSP_OP_AVG (RefDataPtr1[2], RefDataPtr2[2]));
sad += DSP_OP_ABS_DIFF(SrcData[3], DSP_OP_AVG (RefDataPtr1[3], RefDataPtr2[3]));
sad += DSP_OP_ABS_DIFF(SrcData[4], DSP_OP_AVG (RefDataPtr1[4], RefDataPtr2[4]));
sad += DSP_OP_ABS_DIFF(SrcData[5], DSP_OP_AVG (RefDataPtr1[5], RefDataPtr2[5]));
sad += DSP_OP_ABS_DIFF(SrcData[6], DSP_OP_AVG (RefDataPtr1[6], RefDataPtr2[6]));
sad += DSP_OP_ABS_DIFF(SrcData[7], DSP_OP_AVG (RefDataPtr1[7], RefDataPtr2[7]));
if ( sad > thres )
break;
/* Step to next row of block. */
SrcData += SrcStride;
RefDataPtr1 += RefStride;
RefDataPtr2 += RefStride;
}
return sad;
}
static ogg_uint32_t intra8x8_err__c (unsigned char *DataPtr, ogg_uint32_t Stride)
{
ogg_uint32_t i;
ogg_uint32_t XSum=0;
ogg_uint32_t XXSum=0;
for (i=8; i; i--) {
/* Examine alternate pixel locations. */
XSum += DataPtr[0];
XXSum += DataPtr[0]*DataPtr[0];
XSum += DataPtr[1];
XXSum += DataPtr[1]*DataPtr[1];
XSum += DataPtr[2];
XXSum += DataPtr[2]*DataPtr[2];
XSum += DataPtr[3];
XXSum += DataPtr[3]*DataPtr[3];
XSum += DataPtr[4];
XXSum += DataPtr[4]*DataPtr[4];
XSum += DataPtr[5];
XXSum += DataPtr[5]*DataPtr[5];
XSum += DataPtr[6];
XXSum += DataPtr[6]*DataPtr[6];
XSum += DataPtr[7];
XXSum += DataPtr[7]*DataPtr[7];
/* Step to next row of block. */
DataPtr += Stride;
}
/* Compute population variance as mis-match metric. */
return (( (XXSum<<6) - XSum*XSum ) );
}
static ogg_uint32_t inter8x8_err__c (unsigned char *SrcData, ogg_uint32_t SrcStride,
unsigned char *RefDataPtr, ogg_uint32_t RefStride)
{
ogg_uint32_t i;
ogg_uint32_t XSum=0;
ogg_uint32_t XXSum=0;
ogg_int32_t DiffVal;
for (i=8; i; i--) {
DiffVal = DSP_OP_DIFF (SrcData[0], RefDataPtr[0]);
XSum += DiffVal;
XXSum += DiffVal*DiffVal;
DiffVal = DSP_OP_DIFF (SrcData[1], RefDataPtr[1]);
XSum += DiffVal;
XXSum += DiffVal*DiffVal;
DiffVal = DSP_OP_DIFF (SrcData[2], RefDataPtr[2]);
XSum += DiffVal;
XXSum += DiffVal*DiffVal;
DiffVal = DSP_OP_DIFF (SrcData[3], RefDataPtr[3]);
XSum += DiffVal;
XXSum += DiffVal*DiffVal;
DiffVal = DSP_OP_DIFF (SrcData[4], RefDataPtr[4]);
XSum += DiffVal;
XXSum += DiffVal*DiffVal;
DiffVal = DSP_OP_DIFF (SrcData[5], RefDataPtr[5]);
XSum += DiffVal;
XXSum += DiffVal*DiffVal;
DiffVal = DSP_OP_DIFF (SrcData[6], RefDataPtr[6]);
XSum += DiffVal;
XXSum += DiffVal*DiffVal;
DiffVal = DSP_OP_DIFF (SrcData[7], RefDataPtr[7]);
XSum += DiffVal;
XXSum += DiffVal*DiffVal;
/* Step to next row of block. */
SrcData += SrcStride;
RefDataPtr += RefStride;
}
/* Compute and return population variance as mis-match metric. */
return (( (XXSum<<6) - XSum*XSum ));
}
static ogg_uint32_t inter8x8_err_xy2__c (unsigned char *SrcData, ogg_uint32_t SrcStride,
unsigned char *RefDataPtr1,
unsigned char *RefDataPtr2, ogg_uint32_t RefStride)
{
ogg_uint32_t i;
ogg_uint32_t XSum=0;
ogg_uint32_t XXSum=0;
ogg_int32_t DiffVal;
for (i=8; i; i--) {
DiffVal = DSP_OP_DIFF(SrcData[0], DSP_OP_AVG (RefDataPtr1[0], RefDataPtr2[0]));
XSum += DiffVal;
XXSum += DiffVal*DiffVal;
DiffVal = DSP_OP_DIFF(SrcData[1], DSP_OP_AVG (RefDataPtr1[1], RefDataPtr2[1]));
XSum += DiffVal;
XXSum += DiffVal*DiffVal;
DiffVal = DSP_OP_DIFF(SrcData[2], DSP_OP_AVG (RefDataPtr1[2], RefDataPtr2[2]));
XSum += DiffVal;
XXSum += DiffVal*DiffVal;
DiffVal = DSP_OP_DIFF(SrcData[3], DSP_OP_AVG (RefDataPtr1[3], RefDataPtr2[3]));
XSum += DiffVal;
XXSum += DiffVal*DiffVal;
DiffVal = DSP_OP_DIFF(SrcData[4], DSP_OP_AVG (RefDataPtr1[4], RefDataPtr2[4]));
XSum += DiffVal;
XXSum += DiffVal*DiffVal;
DiffVal = DSP_OP_DIFF(SrcData[5], DSP_OP_AVG (RefDataPtr1[5], RefDataPtr2[5]));
XSum += DiffVal;
XXSum += DiffVal*DiffVal;
DiffVal = DSP_OP_DIFF(SrcData[6], DSP_OP_AVG (RefDataPtr1[6], RefDataPtr2[6]));
XSum += DiffVal;
XXSum += DiffVal*DiffVal;
DiffVal = DSP_OP_DIFF(SrcData[7], DSP_OP_AVG (RefDataPtr1[7], RefDataPtr2[7]));
XSum += DiffVal;
XXSum += DiffVal*DiffVal;
/* Step to next row of block. */
SrcData += SrcStride;
RefDataPtr1 += RefStride;
RefDataPtr2 += RefStride;
}
/* Compute and return population variance as mis-match metric. */
return (( (XXSum<<6) - XSum*XSum ));
}
static void nop (void) { /* NOP */ }
void dsp_init(DspFunctions *funcs)
{
funcs->save_fpu = nop;
funcs->restore_fpu = nop;
funcs->sub8x8 = sub8x8__c;
funcs->sub8x8_128 = sub8x8_128__c;
funcs->sub8x8avg2 = sub8x8avg2__c;
funcs->row_sad8 = row_sad8__c;
funcs->col_sad8x8 = col_sad8x8__c;
funcs->sad8x8 = sad8x8__c;
funcs->sad8x8_thres = sad8x8_thres__c;
funcs->sad8x8_xy2_thres = sad8x8_xy2_thres__c;
funcs->intra8x8_err = intra8x8_err__c;
funcs->inter8x8_err = inter8x8_err__c;
funcs->inter8x8_err_xy2 = inter8x8_err_xy2__c;
}
void dsp_static_init(DspFunctions *funcs)
{
ogg_uint32_t cpuflags;
cpuflags = oc_cpu_flags_get ();
dsp_init (funcs);
dsp_recon_init (funcs, cpuflags);
dsp_dct_init (funcs, cpuflags);
#if defined(USE_ASM)
if (cpuflags & OC_CPU_X86_MMX) {
dsp_mmx_init(funcs);
}
# ifndef WIN32
/* This is implemented for win32 yet */
if (cpuflags & OC_CPU_X86_MMXEXT) {
dsp_mmxext_init(funcs);
}
# endif
#endif
}

166
Engine/lib/libtheora/lib/enc/dsp.h Normal file
View file

@ -0,0 +1,166 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: dsp.h 15153 2008-08-04 18:37:55Z tterribe $
********************************************************************/
#ifndef DSP_H
#define DSP_H
#include "theora/theora.h"
#include "../cpu.h"
typedef struct
{
void (*save_fpu) (void);
void (*restore_fpu) (void);
void (*sub8x8) (unsigned char *FiltPtr, unsigned char *ReconPtr,
ogg_int16_t *DctInputPtr, ogg_uint32_t PixelsPerLine,
ogg_uint32_t ReconPixelsPerLine);
void (*sub8x8_128) (unsigned char *FiltPtr, ogg_int16_t *DctInputPtr,
ogg_uint32_t PixelsPerLine);
void (*sub8x8avg2) (unsigned char *FiltPtr, unsigned char *ReconPtr1,
unsigned char *ReconPtr2, ogg_int16_t *DctInputPtr,
ogg_uint32_t PixelsPerLine,
ogg_uint32_t ReconPixelsPerLine);
void (*copy8x8) (unsigned char *src, unsigned char *dest,
ogg_uint32_t stride);
void (*recon_intra8x8) (unsigned char *ReconPtr, ogg_int16_t *ChangePtr,
ogg_uint32_t LineStep);
void (*recon_inter8x8) (unsigned char *ReconPtr, unsigned char *RefPtr,
ogg_int16_t *ChangePtr, ogg_uint32_t LineStep);
void (*recon_inter8x8_half) (unsigned char *ReconPtr, unsigned char *RefPtr1,
unsigned char *RefPtr2, ogg_int16_t *ChangePtr,
ogg_uint32_t LineStep);
void (*fdct_short) (ogg_int16_t *InputData, ogg_int16_t *OutputData);
ogg_uint32_t (*row_sad8) (unsigned char *Src1, unsigned char *Src2);
ogg_uint32_t (*col_sad8x8) (unsigned char *Src1, unsigned char *Src2,
ogg_uint32_t stride);
ogg_uint32_t (*sad8x8) (unsigned char *ptr1, ogg_uint32_t stride1,
unsigned char *ptr2, ogg_uint32_t stride2);
ogg_uint32_t (*sad8x8_thres) (unsigned char *ptr1, ogg_uint32_t stride1,
unsigned char *ptr2, ogg_uint32_t stride2,
ogg_uint32_t thres);
ogg_uint32_t (*sad8x8_xy2_thres)(unsigned char *SrcData, ogg_uint32_t SrcStride,
unsigned char *RefDataPtr1,
unsigned char *RefDataPtr2, ogg_uint32_t RefStride,
ogg_uint32_t thres);
ogg_uint32_t (*intra8x8_err) (unsigned char *DataPtr, ogg_uint32_t Stride);
ogg_uint32_t (*inter8x8_err) (unsigned char *SrcData, ogg_uint32_t SrcStride,
unsigned char *RefDataPtr, ogg_uint32_t RefStride);
ogg_uint32_t (*inter8x8_err_xy2)(unsigned char *SrcData, ogg_uint32_t SrcStride,
unsigned char *RefDataPtr1,
unsigned char *RefDataPtr2, ogg_uint32_t RefStride);
void (*LoopFilter) (PB_INSTANCE *pbi, int FLimit);
void (*FilterVert) (unsigned char * PixelPtr,
ogg_int32_t LineLength, ogg_int16_t *BoundingValuePtr);
void (*IDctSlow) (ogg_int16_t *InputData,
ogg_int16_t *QuantMatrix, ogg_int16_t *OutputData);
void (*IDct3) (ogg_int16_t *InputData,
ogg_int16_t *QuantMatrix, ogg_int16_t *OutputData);
void (*IDct10) (ogg_int16_t *InputData,
ogg_int16_t *QuantMatrix, ogg_int16_t *OutputData);
} DspFunctions;
extern void dsp_dct_init(DspFunctions *funcs, ogg_uint32_t cpu_flags);
extern void dsp_recon_init (DspFunctions *funcs, ogg_uint32_t cpu_flags);
extern void dsp_dct_decode_init(DspFunctions *funcs, ogg_uint32_t cpu_flags);
extern void dsp_idct_init(DspFunctions *funcs, ogg_uint32_t cpu_flags);
void dsp_init(DspFunctions *funcs);
void dsp_static_init(DspFunctions *funcs);
#if defined(USE_ASM) && (defined(__i386__) || defined(__x86_64__) || defined(WIN32))
extern void dsp_mmx_init(DspFunctions *funcs);
extern void dsp_mmxext_init(DspFunctions *funcs);
extern void dsp_mmx_fdct_init(DspFunctions *funcs);
extern void dsp_mmx_recon_init(DspFunctions *funcs);
extern void dsp_mmx_dct_decode_init(DspFunctions *funcs);
extern void dsp_mmx_idct_init(DspFunctions *funcs);
#endif
#define dsp_save_fpu(funcs) (funcs.save_fpu ())
#define dsp_restore_fpu(funcs) (funcs.restore_fpu ())
#define dsp_sub8x8(funcs,a1,a2,a3,a4,a5) (funcs.sub8x8 (a1,a2,a3,a4,a5))
#define dsp_sub8x8_128(funcs,a1,a2,a3) (funcs.sub8x8_128 (a1,a2,a3))
#define dsp_sub8x8avg2(funcs,a1,a2,a3,a4,a5,a6) (funcs.sub8x8avg2 (a1,a2,a3,a4,a5,a6))
#define dsp_copy8x8(funcs,ptr1,ptr2,str1) (funcs.copy8x8 (ptr1,ptr2,str1))
#define dsp_recon_intra8x8(funcs,ptr1,ptr2,str1) (funcs.recon_intra8x8 (ptr1,ptr2,str1))
#define dsp_recon_inter8x8(funcs,ptr1,ptr2,ptr3,str1) \
(funcs.recon_inter8x8 (ptr1,ptr2,ptr3,str1))
#define dsp_recon_inter8x8_half(funcs,ptr1,ptr2,ptr3,ptr4,str1) \
(funcs.recon_inter8x8_half (ptr1,ptr2,ptr3,ptr4,str1))
#define dsp_fdct_short(funcs,in,out) (funcs.fdct_short (in,out))
#define dsp_row_sad8(funcs,ptr1,ptr2) (funcs.row_sad8 (ptr1,ptr2))
#define dsp_col_sad8x8(funcs,ptr1,ptr2,str1) (funcs.col_sad8x8 (ptr1,ptr2,str1))
#define dsp_sad8x8(funcs,ptr1,str1,ptr2,str2) (funcs.sad8x8 (ptr1,str1,ptr2,str2))
#define dsp_sad8x8_thres(funcs,ptr1,str1,ptr2,str2,t) (funcs.sad8x8_thres (ptr1,str1,ptr2,str2,t))
#define dsp_sad8x8_xy2_thres(funcs,ptr1,str1,ptr2,ptr3,str2,t) \
(funcs.sad8x8_xy2_thres (ptr1,str1,ptr2,ptr3,str2,t))
#define dsp_intra8x8_err(funcs,ptr1,str1) (funcs.intra8x8_err (ptr1,str1))
#define dsp_inter8x8_err(funcs,ptr1,str1,ptr2,str2) \
(funcs.inter8x8_err (ptr1,str1,ptr2,str2))
#define dsp_inter8x8_err_xy2(funcs,ptr1,str1,ptr2,ptr3,str2) \
(funcs.inter8x8_err_xy2 (ptr1,str1,ptr2,ptr3,str2))
#define dsp_LoopFilter(funcs, ptr1, i) \
(funcs.LoopFilter(ptr1, i))
#define dsp_IDctSlow(funcs, ptr1, ptr2, ptr3) \
(funcs.IDctSlow(ptr1, ptr2, ptr3))
#define dsp_IDct3(funcs, ptr1, ptr2, ptr3) \
(funcs.IDctSlow(ptr1, ptr2, ptr3))
#define dsp_IDct10(funcs, ptr1, ptr2, ptr3) \
(funcs.IDctSlow(ptr1, ptr2, ptr3))
#endif /* DSP_H */

1141
Engine/lib/libtheora/lib/enc/encapiwrapper.c Normal file
View file

File diff suppressed because it is too large Load diff

1479
Engine/lib/libtheora/lib/enc/encode.c Normal file
View file

File diff suppressed because it is too large Load diff

310
Engine/lib/libtheora/lib/enc/encoder_huffman.c Normal file
View file

@ -0,0 +1,310 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: encoder_huffman.c 13884 2007-09-22 08:38:10Z giles $
********************************************************************/
#include <stdlib.h>
#include <stdio.h>
#include "codec_internal.h"
#include "hufftables.h"
static void CreateHuffmanList(HUFF_ENTRY ** HuffRoot,
ogg_uint32_t HIndex,
const ogg_uint32_t *FreqList ) {
int i;
HUFF_ENTRY *entry_ptr;
HUFF_ENTRY *search_ptr;
/* Create a HUFF entry for token zero. */
HuffRoot[HIndex] = (HUFF_ENTRY *)_ogg_calloc(1,sizeof(*HuffRoot[HIndex]));
HuffRoot[HIndex]->Previous = NULL;
HuffRoot[HIndex]->Next = NULL;
HuffRoot[HIndex]->ZeroChild = NULL;
HuffRoot[HIndex]->OneChild = NULL;
HuffRoot[HIndex]->Value = 0;
HuffRoot[HIndex]->Frequency = FreqList[0];
if ( HuffRoot[HIndex]->Frequency == 0 )
HuffRoot[HIndex]->Frequency = 1;
/* Now add entries for all the other possible tokens. */
for ( i = 1; i < MAX_ENTROPY_TOKENS; i++ ) {
entry_ptr = (HUFF_ENTRY *)_ogg_calloc(1,sizeof(*entry_ptr));
entry_ptr->Value = i;
entry_ptr->Frequency = FreqList[i];
entry_ptr->ZeroChild = NULL;
entry_ptr->OneChild = NULL;
/* Force min value of 1. This prevents the tree getting too deep. */
if ( entry_ptr->Frequency == 0 )
entry_ptr->Frequency = 1;
if ( entry_ptr->Frequency <= HuffRoot[HIndex]->Frequency ){
entry_ptr->Next = HuffRoot[HIndex];
HuffRoot[HIndex]->Previous = entry_ptr;
entry_ptr->Previous = NULL;
HuffRoot[HIndex] = entry_ptr;
}else{
search_ptr = HuffRoot[HIndex];
while ( (search_ptr->Next != NULL) &&
(search_ptr->Frequency < entry_ptr->Frequency) ){
search_ptr = (HUFF_ENTRY *)search_ptr->Next;
}
if ( search_ptr->Frequency < entry_ptr->Frequency ){
entry_ptr->Next = NULL;
entry_ptr->Previous = search_ptr;
search_ptr->Next = entry_ptr;
}else{
entry_ptr->Next = search_ptr;
entry_ptr->Previous = search_ptr->Previous;
search_ptr->Previous->Next = entry_ptr;
search_ptr->Previous = entry_ptr;
}
}
}
}
static void CreateCodeArray( HUFF_ENTRY * HuffRoot,
ogg_uint32_t *HuffCodeArray,
unsigned char *HuffCodeLengthArray,
ogg_uint32_t CodeValue,
unsigned char CodeLength ) {
/* If we are at a leaf then fill in a code array entry. */
if ( ( HuffRoot->ZeroChild == NULL ) && ( HuffRoot->OneChild == NULL ) ){
HuffCodeArray[HuffRoot->Value] = CodeValue;
HuffCodeLengthArray[HuffRoot->Value] = CodeLength;
}else{
/* Recursive calls to scan down the tree. */
CodeLength++;
CreateCodeArray(HuffRoot->ZeroChild, HuffCodeArray, HuffCodeLengthArray,
((CodeValue << 1) + 0), CodeLength);
CreateCodeArray(HuffRoot->OneChild, HuffCodeArray, HuffCodeLengthArray,
((CodeValue << 1) + 1), CodeLength);
}
}
static void BuildHuffmanTree( HUFF_ENTRY **HuffRoot,
ogg_uint32_t *HuffCodeArray,
unsigned char *HuffCodeLengthArray,
ogg_uint32_t HIndex,
const ogg_uint32_t *FreqList ){
HUFF_ENTRY *entry_ptr;
HUFF_ENTRY *search_ptr;
/* First create a sorted linked list representing the frequencies of
each token. */
CreateHuffmanList( HuffRoot, HIndex, FreqList );
/* Now build the tree from the list. */
/* While there are at least two items left in the list. */
while ( HuffRoot[HIndex]->Next != NULL ){
/* Create the new node as the parent of the first two in the list. */
entry_ptr = (HUFF_ENTRY *)_ogg_calloc(1,sizeof(*entry_ptr));
entry_ptr->Value = -1;
entry_ptr->Frequency = HuffRoot[HIndex]->Frequency +
HuffRoot[HIndex]->Next->Frequency ;
entry_ptr->ZeroChild = HuffRoot[HIndex];
entry_ptr->OneChild = HuffRoot[HIndex]->Next;
/* If there are still more items in the list then insert the new
node into the list. */
if (entry_ptr->OneChild->Next != NULL ){
/* Set up the provisional 'new root' */
HuffRoot[HIndex] = entry_ptr->OneChild->Next;
HuffRoot[HIndex]->Previous = NULL;
/* Now scan through the remaining list to insert the new entry
at the appropriate point. */
if ( entry_ptr->Frequency <= HuffRoot[HIndex]->Frequency ){
entry_ptr->Next = HuffRoot[HIndex];
HuffRoot[HIndex]->Previous = entry_ptr;
entry_ptr->Previous = NULL;
HuffRoot[HIndex] = entry_ptr;
}else{
search_ptr = HuffRoot[HIndex];
while ( (search_ptr->Next != NULL) &&
(search_ptr->Frequency < entry_ptr->Frequency) ){
search_ptr = search_ptr->Next;
}
if ( search_ptr->Frequency < entry_ptr->Frequency ){
entry_ptr->Next = NULL;
entry_ptr->Previous = search_ptr;
search_ptr->Next = entry_ptr;
}else{
entry_ptr->Next = search_ptr;
entry_ptr->Previous = search_ptr->Previous;
search_ptr->Previous->Next = entry_ptr;
search_ptr->Previous = entry_ptr;
}
}
}else{
/* Build has finished. */
entry_ptr->Next = NULL;
entry_ptr->Previous = NULL;
HuffRoot[HIndex] = entry_ptr;
}
/* Delete the Next/Previous properties of the children (PROB NOT NEC). */
entry_ptr->ZeroChild->Next = NULL;
entry_ptr->ZeroChild->Previous = NULL;
entry_ptr->OneChild->Next = NULL;
entry_ptr->OneChild->Previous = NULL;
}
/* Now build a code array from the tree. */
CreateCodeArray( HuffRoot[HIndex], HuffCodeArray,
HuffCodeLengthArray, 0, 0);
}
static void DestroyHuffTree(HUFF_ENTRY *root_ptr){
if (root_ptr){
if ( root_ptr->ZeroChild )
DestroyHuffTree(root_ptr->ZeroChild);
if ( root_ptr->OneChild )
DestroyHuffTree(root_ptr->OneChild);
_ogg_free(root_ptr);
}
}
void ClearHuffmanSet( PB_INSTANCE *pbi ){
int i;
ClearHuffmanTrees(pbi->HuffRoot_VP3x);
for ( i = 0; i < NUM_HUFF_TABLES; i++ )
if (pbi->HuffCodeArray_VP3x[i])
_ogg_free (pbi->HuffCodeArray_VP3x[i]);
for ( i = 0; i < NUM_HUFF_TABLES; i++ )
if (pbi->HuffCodeLengthArray_VP3x[i])
_ogg_free (pbi->HuffCodeLengthArray_VP3x[i]);
}
void InitHuffmanSet( PB_INSTANCE *pbi ){
int i;
ClearHuffmanSet(pbi);
pbi->ExtraBitLengths_VP3x = ExtraBitLengths_VP31;
for ( i = 0; i < NUM_HUFF_TABLES; i++ ){
pbi->HuffCodeArray_VP3x[i] =
_ogg_calloc(MAX_ENTROPY_TOKENS,
sizeof(*pbi->HuffCodeArray_VP3x[i]));
pbi->HuffCodeLengthArray_VP3x[i] =
_ogg_calloc(MAX_ENTROPY_TOKENS,
sizeof(*pbi->HuffCodeLengthArray_VP3x[i]));
BuildHuffmanTree( pbi->HuffRoot_VP3x,
pbi->HuffCodeArray_VP3x[i],
pbi->HuffCodeLengthArray_VP3x[i],
i, FrequencyCounts_VP3[i]);
}
}
static int ReadHuffTree(HUFF_ENTRY * HuffRoot, int depth,
oggpack_buffer *opb) {
long bit;
long ret;
theora_read(opb,1,&bit);
if(bit < 0) return OC_BADHEADER;
else if(!bit) {
int ret;
if (++depth > 32) return OC_BADHEADER;
HuffRoot->ZeroChild = (HUFF_ENTRY *)_ogg_calloc(1, sizeof(HUFF_ENTRY));
ret = ReadHuffTree(HuffRoot->ZeroChild, depth, opb);
if (ret < 0) return ret;
HuffRoot->OneChild = (HUFF_ENTRY *)_ogg_calloc(1, sizeof(HUFF_ENTRY));
ret = ReadHuffTree(HuffRoot->OneChild, depth, opb);
if (ret < 0) return ret;
HuffRoot->Value = -1;
} else {
HuffRoot->ZeroChild = NULL;
HuffRoot->OneChild = NULL;
theora_read(opb,5,&ret);
HuffRoot->Value=ret;;
if (HuffRoot->Value < 0) return OC_BADHEADER;
}
return 0;
}
int ReadHuffmanTrees(codec_setup_info *ci, oggpack_buffer *opb) {
int i;
for (i=0; i<NUM_HUFF_TABLES; i++) {
int ret;
ci->HuffRoot[i] = (HUFF_ENTRY *)_ogg_calloc(1, sizeof(HUFF_ENTRY));
ret = ReadHuffTree(ci->HuffRoot[i], 0, opb);
if (ret) return ret;
}
return 0;
}
static void WriteHuffTree(HUFF_ENTRY *HuffRoot, oggpack_buffer *opb) {
if (HuffRoot->Value >= 0) {
oggpackB_write(opb, 1, 1);
oggpackB_write(opb, HuffRoot->Value, 5);
} else {
oggpackB_write(opb, 0, 1);
WriteHuffTree(HuffRoot->ZeroChild, opb);
WriteHuffTree(HuffRoot->OneChild, opb);
}
}
void WriteHuffmanTrees(HUFF_ENTRY *HuffRoot[NUM_HUFF_TABLES],
oggpack_buffer *opb) {
int i;
for(i=0; i<NUM_HUFF_TABLES; i++) {
WriteHuffTree(HuffRoot[i], opb);
}
}
static HUFF_ENTRY *CopyHuffTree(const HUFF_ENTRY *HuffSrc) {
if(HuffSrc){
HUFF_ENTRY *HuffDst;
HuffDst = (HUFF_ENTRY *)_ogg_calloc(1, sizeof(HUFF_ENTRY));
HuffDst->Value = HuffSrc->Value;
if (HuffSrc->Value < 0) {
HuffDst->ZeroChild = CopyHuffTree(HuffSrc->ZeroChild);
HuffDst->OneChild = CopyHuffTree(HuffSrc->OneChild);
}
return HuffDst;
}
return NULL;
}
void InitHuffmanTrees(PB_INSTANCE *pbi, const codec_setup_info *ci) {
int i;
pbi->ExtraBitLengths_VP3x = ExtraBitLengths_VP31;
for(i=0; i<NUM_HUFF_TABLES; i++){
pbi->HuffRoot_VP3x[i] = CopyHuffTree(ci->HuffRoot[i]);
}
}
void ClearHuffmanTrees(HUFF_ENTRY *HuffRoot[NUM_HUFF_TABLES]){
int i;
for(i=0; i<NUM_HUFF_TABLES; i++) {
DestroyHuffTree(HuffRoot[i]);
HuffRoot[i] = NULL;
}
}

74
Engine/lib/libtheora/lib/enc/encoder_huffman.h Normal file
View file

@ -0,0 +1,74 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: encoder_huffman.h 13884 2007-09-22 08:38:10Z giles $
********************************************************************/
/********************************************************************
* Constants
********************************************************************/
#define NUM_HUFF_TABLES 80
#define DC_HUFF_OFFSET 0
#define AC_HUFF_OFFSET 16
#define AC_TABLE_2_THRESH 5
#define AC_TABLE_3_THRESH 14
#define AC_TABLE_4_THRESH 27
#define DC_HUFF_CHOICES 16
#define DC_HUFF_CHOICE_BITS 4
#define AC_HUFF_CHOICES 16
#define AC_HUFF_CHOICE_BITS 4
/* Constants assosciated with entropy tokenisation. */
#define MAX_SINGLE_TOKEN_VALUE 6
#define DCT_VAL_CAT2_MIN 3
#define DCT_VAL_CAT3_MIN 7
#define DCT_VAL_CAT4_MIN 9
#define DCT_VAL_CAT5_MIN 13
#define DCT_VAL_CAT6_MIN 21
#define DCT_VAL_CAT7_MIN 37
#define DCT_VAL_CAT8_MIN 69
#define DCT_EOB_TOKEN 0
#define DCT_EOB_PAIR_TOKEN 1
#define DCT_EOB_TRIPLE_TOKEN 2
#define DCT_REPEAT_RUN_TOKEN 3
#define DCT_REPEAT_RUN2_TOKEN 4
#define DCT_REPEAT_RUN3_TOKEN 5
#define DCT_REPEAT_RUN4_TOKEN 6
#define DCT_SHORT_ZRL_TOKEN 7
#define DCT_ZRL_TOKEN 8
#define ONE_TOKEN 9 /* Special tokens for -1,1,-2,2 */
#define MINUS_ONE_TOKEN 10
#define TWO_TOKEN 11
#define MINUS_TWO_TOKEN 12
#define LOW_VAL_TOKENS (MINUS_TWO_TOKEN + 1)
#define DCT_VAL_CATEGORY3 (LOW_VAL_TOKENS + 4)
#define DCT_VAL_CATEGORY4 (DCT_VAL_CATEGORY3 + 1)
#define DCT_VAL_CATEGORY5 (DCT_VAL_CATEGORY4 + 1)
#define DCT_VAL_CATEGORY6 (DCT_VAL_CATEGORY5 + 1)
#define DCT_VAL_CATEGORY7 (DCT_VAL_CATEGORY6 + 1)
#define DCT_VAL_CATEGORY8 (DCT_VAL_CATEGORY7 + 1)
#define DCT_RUN_CATEGORY1 (DCT_VAL_CATEGORY8 + 1)
#define DCT_RUN_CATEGORY1B (DCT_RUN_CATEGORY1 + 5)
#define DCT_RUN_CATEGORY1C (DCT_RUN_CATEGORY1B + 1)
#define DCT_RUN_CATEGORY2 (DCT_RUN_CATEGORY1C + 1)
/* 32 */
#define MAX_ENTROPY_TOKENS (DCT_RUN_CATEGORY2 + 2)

572
Engine/lib/libtheora/lib/enc/encoder_idct.c Normal file
View file

@ -0,0 +1,572 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation http://www.xiph.org/ *
* *
********************************************************************
function: C implementation of the Theora iDCT
last mod: $Id: encoder_idct.c 14714 2008-04-12 01:04:43Z giles $
********************************************************************/
#include <string.h>
#include "codec_internal.h"
#include "quant_lookup.h"
#define IdctAdjustBeforeShift 8
/* cos(n*pi/16) or sin(8-n)*pi/16) */
#define xC1S7 64277
#define xC2S6 60547
#define xC3S5 54491
#define xC4S4 46341
#define xC5S3 36410
#define xC6S2 25080
#define xC7S1 12785
/* compute the 16 bit signed 1D inverse DCT - spec version */
/*
static void idct_short__c ( ogg_int16_t * InputData, ogg_int16_t * OutputData ) {
ogg_int32_t t[8], r;
ogg_int16_t *y = InputData;
ogg_int16_t *x = OutputData;
t[0] = y[0] + y[4];
t[0] &= 0xffff;
t[0] = (xC4S4 * t[0]) >> 16;
t[1] = y[0] - y[4];
t[1] &= 0xffff;
t[1] = (xC4S4 * t[1]) >> 16;
t[2] = ((xC6S2 * y[2]) >> 16) - ((xC2S6 * y[6]) >> 16);
t[3] = ((xC2S6 * y[2]) >> 16) + ((xC6S2 * y[6]) >> 16);
t[4] = ((xC7S1 * y[1]) >> 16) - ((xC1S7 * y[7]) >> 16);
t[5] = ((xC3S5 * y[5]) >> 16) - ((xC5S3 * y[3]) >> 16);
t[6] = ((xC5S3 * y[5]) >> 16) + ((xC3S5 * y[3]) >> 16);
t[7] = ((xC1S7 * y[1]) >> 16) + ((xC7S1 * y[7]) >> 16);
r = t[4] + t[5];
t[5] = t[4] - t[5];
t[5] &= 0xffff;
t[5] = (xC4S4 * (-t[5])) >> 16;
t[4] = r;
r = t[7] + t[6];
t[6] = t[7] - t[6];
t[6] &= 0xffff;
t[6] = (xC4S4 * t[6]) >> 16;
t[7] = r;
r = t[0] + t[3];
t[3] = t[0] - t[3];
t[0] = r;
r = t[1] + t[2];
t[2] = t[1] - t[2];
t[1] = r;
r = t[6] + t[5];
t[5] = t[6] - t[5];
t[6] = r;
r = t[0] + t[7];
r &= 0xffff;
x[0] = r;
r = t[1] + t[6];
r &= 0xffff;
x[1] = r;
r = t[2] + t[5];
r &= 0xffff;
x[2] = r;
r = t[3] + t[4];
r &= 0xffff;
x[3] = r;
r = t[3] - t[4];
r &= 0xffff;
x[4] = r;
r = t[2] - t[5];
r &= 0xffff;
x[5] = r;
r = t[1] - t[6];
r &= 0xffff;
x[6] = r;
r = t[0] - t[7];
r &= 0xffff;
x[7] = r;
}
*/
static void dequant_slow( ogg_int16_t * dequant_coeffs,
ogg_int16_t * quantized_list,
ogg_int32_t * DCT_block) {
int i;
for(i=0;i<64;i++)
DCT_block[dezigzag_index[i]] = quantized_list[i] * dequant_coeffs[i];
}
void IDctSlow__c( Q_LIST_ENTRY * InputData,
ogg_int16_t *QuantMatrix,
ogg_int16_t * OutputData ) {
ogg_int32_t IntermediateData[64];
ogg_int32_t * ip = IntermediateData;
ogg_int16_t * op = OutputData;
ogg_int32_t _A, _B, _C, _D, _Ad, _Bd, _Cd, _Dd, _E, _F, _G, _H;
ogg_int32_t _Ed, _Gd, _Add, _Bdd, _Fd, _Hd;
ogg_int32_t t1, t2;
int loop;
dequant_slow( QuantMatrix, InputData, IntermediateData);
/* Inverse DCT on the rows now */
for ( loop = 0; loop < 8; loop++){
/* Check for non-zero values */
if ( ip[0] | ip[1] | ip[2] | ip[3] | ip[4] | ip[5] | ip[6] | ip[7] ) {
t1 = (xC1S7 * ip[1]);
t2 = (xC7S1 * ip[7]);
t1 >>= 16;
t2 >>= 16;
_A = t1 + t2;
t1 = (xC7S1 * ip[1]);
t2 = (xC1S7 * ip[7]);
t1 >>= 16;
t2 >>= 16;
_B = t1 - t2;
t1 = (xC3S5 * ip[3]);
t2 = (xC5S3 * ip[5]);
t1 >>= 16;
t2 >>= 16;
_C = t1 + t2;
t1 = (xC3S5 * ip[5]);
t2 = (xC5S3 * ip[3]);
t1 >>= 16;
t2 >>= 16;
_D = t1 - t2;
t1 = (xC4S4 * (ogg_int16_t)(_A - _C));
t1 >>= 16;
_Ad = t1;
t1 = (xC4S4 * (ogg_int16_t)(_B - _D));
t1 >>= 16;
_Bd = t1;
_Cd = _A + _C;
_Dd = _B + _D;
t1 = (xC4S4 * (ogg_int16_t)(ip[0] + ip[4]));
t1 >>= 16;
_E = t1;
t1 = (xC4S4 * (ogg_int16_t)(ip[0] - ip[4]));
t1 >>= 16;
_F = t1;
t1 = (xC2S6 * ip[2]);
t2 = (xC6S2 * ip[6]);
t1 >>= 16;
t2 >>= 16;
_G = t1 + t2;
t1 = (xC6S2 * ip[2]);
t2 = (xC2S6 * ip[6]);
t1 >>= 16;
t2 >>= 16;
_H = t1 - t2;
_Ed = _E - _G;
_Gd = _E + _G;
_Add = _F + _Ad;
_Bdd = _Bd - _H;
_Fd = _F - _Ad;
_Hd = _Bd + _H;
/* Final sequence of operations over-write original inputs. */
ip[0] = (ogg_int16_t)((_Gd + _Cd ) >> 0);
ip[7] = (ogg_int16_t)((_Gd - _Cd ) >> 0);
ip[1] = (ogg_int16_t)((_Add + _Hd ) >> 0);
ip[2] = (ogg_int16_t)((_Add - _Hd ) >> 0);
ip[3] = (ogg_int16_t)((_Ed + _Dd ) >> 0);
ip[4] = (ogg_int16_t)((_Ed - _Dd ) >> 0);
ip[5] = (ogg_int16_t)((_Fd + _Bdd ) >> 0);
ip[6] = (ogg_int16_t)((_Fd - _Bdd ) >> 0);
}
ip += 8; /* next row */
}
ip = IntermediateData;
for ( loop = 0; loop < 8; loop++){
/* Check for non-zero values (bitwise or faster than ||) */
if ( ip[0 * 8] | ip[1 * 8] | ip[2 * 8] | ip[3 * 8] |
ip[4 * 8] | ip[5 * 8] | ip[6 * 8] | ip[7 * 8] ) {
t1 = (xC1S7 * ip[1*8]);
t2 = (xC7S1 * ip[7*8]);
t1 >>= 16;
t2 >>= 16;
_A = t1 + t2;
t1 = (xC7S1 * ip[1*8]);
t2 = (xC1S7 * ip[7*8]);
t1 >>= 16;
t2 >>= 16;
_B = t1 - t2;
t1 = (xC3S5 * ip[3*8]);
t2 = (xC5S3 * ip[5*8]);
t1 >>= 16;
t2 >>= 16;
_C = t1 + t2;
t1 = (xC3S5 * ip[5*8]);
t2 = (xC5S3 * ip[3*8]);
t1 >>= 16;
t2 >>= 16;
_D = t1 - t2;
t1 = (xC4S4 * (ogg_int16_t)(_A - _C));
t1 >>= 16;
_Ad = t1;
t1 = (xC4S4 * (ogg_int16_t)(_B - _D));
t1 >>= 16;
_Bd = t1;
_Cd = _A + _C;
_Dd = _B + _D;
t1 = (xC4S4 * (ogg_int16_t)(ip[0*8] + ip[4*8]));
t1 >>= 16;
_E = t1;
t1 = (xC4S4 * (ogg_int16_t)(ip[0*8] - ip[4*8]));
t1 >>= 16;
_F = t1;
t1 = (xC2S6 * ip[2*8]);
t2 = (xC6S2 * ip[6*8]);
t1 >>= 16;
t2 >>= 16;
_G = t1 + t2;
t1 = (xC6S2 * ip[2*8]);
t2 = (xC2S6 * ip[6*8]);
t1 >>= 16;
t2 >>= 16;
_H = t1 - t2;
_Ed = _E - _G;
_Gd = _E + _G;
_Add = _F + _Ad;
_Bdd = _Bd - _H;
_Fd = _F - _Ad;
_Hd = _Bd + _H;
_Gd += IdctAdjustBeforeShift;
_Add += IdctAdjustBeforeShift;
_Ed += IdctAdjustBeforeShift;
_Fd += IdctAdjustBeforeShift;
/* Final sequence of operations over-write original inputs. */
op[0*8] = (ogg_int16_t)((_Gd + _Cd ) >> 4);
op[7*8] = (ogg_int16_t)((_Gd - _Cd ) >> 4);
op[1*8] = (ogg_int16_t)((_Add + _Hd ) >> 4);
op[2*8] = (ogg_int16_t)((_Add - _Hd ) >> 4);
op[3*8] = (ogg_int16_t)((_Ed + _Dd ) >> 4);
op[4*8] = (ogg_int16_t)((_Ed - _Dd ) >> 4);
op[5*8] = (ogg_int16_t)((_Fd + _Bdd ) >> 4);
op[6*8] = (ogg_int16_t)((_Fd - _Bdd ) >> 4);
}else{
op[0*8] = 0;
op[7*8] = 0;
op[1*8] = 0;
op[2*8] = 0;
op[3*8] = 0;
op[4*8] = 0;
op[5*8] = 0;
op[6*8] = 0;
}
ip++; /* next column */
op++;
}
}
/************************
x x x x 0 0 0 0
x x x 0 0 0 0 0
x x 0 0 0 0 0 0
x 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
*************************/
static void dequant_slow10( ogg_int16_t * dequant_coeffs,
ogg_int16_t * quantized_list,
ogg_int32_t * DCT_block){
int i;
memset(DCT_block,0, 128);
for(i=0;i<10;i++)
DCT_block[dezigzag_index[i]] = quantized_list[i] * dequant_coeffs[i];
}
void IDct10__c( Q_LIST_ENTRY * InputData,
ogg_int16_t *QuantMatrix,
ogg_int16_t * OutputData ){
ogg_int32_t IntermediateData[64];
ogg_int32_t * ip = IntermediateData;
ogg_int16_t * op = OutputData;
ogg_int32_t _A, _B, _C, _D, _Ad, _Bd, _Cd, _Dd, _E, _F, _G, _H;
ogg_int32_t _Ed, _Gd, _Add, _Bdd, _Fd, _Hd;
ogg_int32_t t1, t2;
int loop;
dequant_slow10( QuantMatrix, InputData, IntermediateData);
/* Inverse DCT on the rows now */
for ( loop = 0; loop < 4; loop++){
/* Check for non-zero values */
if ( ip[0] | ip[1] | ip[2] | ip[3] ){
t1 = (xC1S7 * ip[1]);
t1 >>= 16;
_A = t1;
t1 = (xC7S1 * ip[1]);
t1 >>= 16;
_B = t1 ;
t1 = (xC3S5 * ip[3]);
t1 >>= 16;
_C = t1;
t2 = (xC5S3 * ip[3]);
t2 >>= 16;
_D = -t2;
t1 = (xC4S4 * (ogg_int16_t)(_A - _C));
t1 >>= 16;
_Ad = t1;
t1 = (xC4S4 * (ogg_int16_t)(_B - _D));
t1 >>= 16;
_Bd = t1;
_Cd = _A + _C;
_Dd = _B + _D;
t1 = (xC4S4 * ip[0] );
t1 >>= 16;
_E = t1;
_F = t1;
t1 = (xC2S6 * ip[2]);
t1 >>= 16;
_G = t1;
t1 = (xC6S2 * ip[2]);
t1 >>= 16;
_H = t1 ;
_Ed = _E - _G;
_Gd = _E + _G;
_Add = _F + _Ad;
_Bdd = _Bd - _H;
_Fd = _F - _Ad;
_Hd = _Bd + _H;
/* Final sequence of operations over-write original inputs. */
ip[0] = (ogg_int16_t)((_Gd + _Cd ) >> 0);
ip[7] = (ogg_int16_t)((_Gd - _Cd ) >> 0);
ip[1] = (ogg_int16_t)((_Add + _Hd ) >> 0);
ip[2] = (ogg_int16_t)((_Add - _Hd ) >> 0);
ip[3] = (ogg_int16_t)((_Ed + _Dd ) >> 0);
ip[4] = (ogg_int16_t)((_Ed - _Dd ) >> 0);
ip[5] = (ogg_int16_t)((_Fd + _Bdd ) >> 0);
ip[6] = (ogg_int16_t)((_Fd - _Bdd ) >> 0);
}
ip += 8; /* next row */
}
ip = IntermediateData;
for ( loop = 0; loop < 8; loop++) {
/* Check for non-zero values (bitwise or faster than ||) */
if ( ip[0 * 8] | ip[1 * 8] | ip[2 * 8] | ip[3 * 8] ) {
t1 = (xC1S7 * ip[1*8]);
t1 >>= 16;
_A = t1 ;
t1 = (xC7S1 * ip[1*8]);
t1 >>= 16;
_B = t1 ;
t1 = (xC3S5 * ip[3*8]);
t1 >>= 16;
_C = t1 ;
t2 = (xC5S3 * ip[3*8]);
t2 >>= 16;
_D = - t2;
t1 = (xC4S4 * (ogg_int16_t)(_A - _C));
t1 >>= 16;
_Ad = t1;
t1 = (xC4S4 * (ogg_int16_t)(_B - _D));
t1 >>= 16;
_Bd = t1;
_Cd = _A + _C;
_Dd = _B + _D;
t1 = (xC4S4 * ip[0*8]);
t1 >>= 16;
_E = t1;
_F = t1;
t1 = (xC2S6 * ip[2*8]);
t1 >>= 16;
_G = t1;
t1 = (xC6S2 * ip[2*8]);
t1 >>= 16;
_H = t1;
_Ed = _E - _G;
_Gd = _E + _G;
_Add = _F + _Ad;
_Bdd = _Bd - _H;
_Fd = _F - _Ad;
_Hd = _Bd + _H;
_Gd += IdctAdjustBeforeShift;
_Add += IdctAdjustBeforeShift;
_Ed += IdctAdjustBeforeShift;
_Fd += IdctAdjustBeforeShift;
/* Final sequence of operations over-write original inputs. */
op[0*8] = (ogg_int16_t)((_Gd + _Cd ) >> 4);
op[7*8] = (ogg_int16_t)((_Gd - _Cd ) >> 4);
op[1*8] = (ogg_int16_t)((_Add + _Hd ) >> 4);
op[2*8] = (ogg_int16_t)((_Add - _Hd ) >> 4);
op[3*8] = (ogg_int16_t)((_Ed + _Dd ) >> 4);
op[4*8] = (ogg_int16_t)((_Ed - _Dd ) >> 4);
op[5*8] = (ogg_int16_t)((_Fd + _Bdd ) >> 4);
op[6*8] = (ogg_int16_t)((_Fd - _Bdd ) >> 4);
}else{
op[0*8] = 0;
op[7*8] = 0;
op[1*8] = 0;
op[2*8] = 0;
op[3*8] = 0;
op[4*8] = 0;
op[5*8] = 0;
op[6*8] = 0;
}
ip++; /* next column */
op++;
}
}
/***************************
x 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
**************************/
void IDct1( Q_LIST_ENTRY * InputData,
ogg_int16_t *QuantMatrix,
ogg_int16_t * OutputData ){
int loop;
ogg_int16_t OutD;
OutD=(ogg_int16_t) ((ogg_int32_t)(InputData[0]*QuantMatrix[0]+15)>>5);
for(loop=0;loop<64;loop++)
OutputData[loop]=OutD;
}
void dsp_idct_init (DspFunctions *funcs, ogg_uint32_t cpu_flags)
{
funcs->IDctSlow = IDctSlow__c;
funcs->IDct10 = IDct10__c;
funcs->IDct3 = IDct10__c;
#if defined(USE_ASM)
// todo: make mmx encoder idct for MSC one day...
#if !defined (_MSC_VER)
if (cpu_flags & OC_CPU_X86_MMX) {
dsp_mmx_idct_init(funcs);
}
#endif
#endif
}

120
Engine/lib/libtheora/lib/enc/encoder_lookup.h Normal file
View file

@ -0,0 +1,120 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation http://www.xiph.org/ *
* *
********************************************************************
function: simple static lookups for VP3 frame encoder
last mod: $Id: encoder_lookup.h 15323 2008-09-19 19:43:59Z giles $
********************************************************************/
#include "codec_internal.h"
static const ogg_uint32_t MvPattern[(MAX_MV_EXTENT * 2) + 1] = {
0x000000ff, 0x000000fd, 0x000000fb, 0x000000f9,
0x000000f7, 0x000000f5, 0x000000f3, 0x000000f1,
0x000000ef, 0x000000ed, 0x000000eb, 0x000000e9,
0x000000e7, 0x000000e5, 0x000000e3, 0x000000e1,
0x0000006f, 0x0000006d, 0x0000006b, 0x00000069,
0x00000067, 0x00000065, 0x00000063, 0x00000061,
0x0000002f, 0x0000002d, 0x0000002b, 0x00000029,
0x00000009, 0x00000007, 0x00000002, 0x00000000,
0x00000001, 0x00000006, 0x00000008, 0x00000028,
0x0000002a, 0x0000002c, 0x0000002e, 0x00000060,
0x00000062, 0x00000064, 0x00000066, 0x00000068,
0x0000006a, 0x0000006c, 0x0000006e, 0x000000e0,
0x000000e2, 0x000000e4, 0x000000e6, 0x000000e8,
0x000000ea, 0x000000ec, 0x000000ee, 0x000000f0,
0x000000f2, 0x000000f4, 0x000000f6, 0x000000f8,
0x000000fa, 0x000000fc, 0x000000fe,
};
static const ogg_uint32_t MvBits[(MAX_MV_EXTENT * 2) + 1] = {
8, 8, 8, 8, 8, 8, 8, 8,
8, 8, 8, 8, 8, 8, 8, 8,
7, 7, 7, 7, 7, 7, 7, 7,
6, 6, 6, 6, 4, 4, 3, 3,
3, 4, 4, 6, 6, 6, 6, 7,
7, 7, 7, 7, 7, 7, 7, 8,
8, 8, 8, 8, 8, 8, 8, 8,
8, 8, 8, 8, 8, 8, 8,
};
static const ogg_uint32_t MvPattern2[(MAX_MV_EXTENT * 2) + 1] = {
0x0000003f, 0x0000003d, 0x0000003b, 0x00000039,
0x00000037, 0x00000035, 0x00000033, 0x00000031,
0x0000002f, 0x0000002d, 0x0000002b, 0x00000029,
0x00000027, 0x00000025, 0x00000023, 0x00000021,
0x0000001f, 0x0000001d, 0x0000001b, 0x00000019,
0x00000017, 0x00000015, 0x00000013, 0x00000011,
0x0000000f, 0x0000000d, 0x0000000b, 0x00000009,
0x00000007, 0x00000005, 0x00000003, 0x00000000,
0x00000002, 0x00000004, 0x00000006, 0x00000008,
0x0000000a, 0x0000000c, 0x0000000e, 0x00000010,
0x00000012, 0x00000014, 0x00000016, 0x00000018,
0x0000001a, 0x0000001c, 0x0000001e, 0x00000020,
0x00000022, 0x00000024, 0x00000026, 0x00000028,
0x0000002a, 0x0000002c, 0x0000002e, 0x00000030,
0x00000032, 0x00000034, 0x00000036, 0x00000038,
0x0000003a, 0x0000003c, 0x0000003e,
};
static const ogg_uint32_t MvBits2[(MAX_MV_EXTENT * 2) + 1] = {
6, 6, 6, 6, 6, 6, 6, 6,
6, 6, 6, 6, 6, 6, 6, 6,
6, 6, 6, 6, 6, 6, 6, 6,
6, 6, 6, 6, 6, 6, 6, 6,
6, 6, 6, 6, 6, 6, 6, 6,
6, 6, 6, 6, 6, 6, 6, 6,
6, 6, 6, 6, 6, 6, 6, 6,
6, 6, 6, 6, 6, 6, 6,
};
static const ogg_uint32_t ModeBitPatterns[MAX_MODES] = {
0x00, 0x02, 0x06, 0x0E, 0x1E, 0x3E, 0x7E, 0x7F };
static const ogg_int32_t ModeBitLengths[MAX_MODES] = {
1, 2, 3, 4, 5, 6, 7, 7 };
static const unsigned char ModeSchemes[MODE_METHODS-2][MAX_MODES] = {
/* Last Mv dominates */
{ 3, 4, 2, 0, 1, 5, 6, 7 }, /* L P M N I G GM 4 */
{ 2, 4, 3, 0, 1, 5, 6, 7 }, /* L P N M I G GM 4 */
{ 3, 4, 1, 0, 2, 5, 6, 7 }, /* L M P N I G GM 4 */
{ 2, 4, 1, 0, 3, 5, 6, 7 }, /* L M N P I G GM 4 */
/* No MV dominates */
{ 0, 4, 3, 1, 2, 5, 6, 7 }, /* N L P M I G GM 4 */
{ 0, 5, 4, 2, 3, 1, 6, 7 }, /* N G L P M I GM 4 */
};
static const ogg_uint32_t MvThreshTable[Q_TABLE_SIZE] = {
65, 65, 65, 65, 50, 50, 50, 50,
40, 40, 40, 40, 40, 40, 40, 40,
30, 30, 30, 30, 30, 30, 30, 30,
20, 20, 20, 20, 20, 20, 20, 20,
15, 15, 15, 15, 15, 15, 15, 15,
10, 10, 10, 10, 10, 10, 10, 10,
5, 5, 5, 5, 5, 5, 5, 5,
0, 0, 0, 0, 0, 0, 0, 0
};
static const ogg_uint32_t MVChangeFactorTable[Q_TABLE_SIZE] = {
11, 11, 11, 11, 12, 12, 12, 12,
13, 13, 13, 13, 13, 13, 13, 13,
14, 14, 14, 14, 14, 14, 14, 14,
14, 14, 14, 14, 14, 14, 14, 14,
14, 14, 14, 14, 14, 14, 14, 14,
14, 14, 14, 14, 14, 14, 14, 14,
15, 15, 15, 15, 15, 15, 15, 15,
15, 15, 15, 15, 15, 15, 15, 15
};

558
Engine/lib/libtheora/lib/enc/encoder_quant.c Normal file
View file

@ -0,0 +1,558 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2005 *
* by the Xiph.Org Foundation http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: encoder_quant.c 15153 2008-08-04 18:37:55Z tterribe $
********************************************************************/
#include <stdlib.h>
#include <string.h>
#include "codec_internal.h"
#include "quant_lookup.h"
#define OC_QUANT_MAX (1024<<2)
static const unsigned DC_QUANT_MIN[2]={4<<2,8<<2};
static const unsigned AC_QUANT_MIN[2]={2<<2,4<<2};
#define OC_MAXI(_a,_b) ((_a)<(_b)?(_b):(_a))
#define OC_MINI(_a,_b) ((_a)>(_b)?(_b):(_a))
#define OC_CLAMPI(_a,_b,_c) (OC_MAXI(_a,OC_MINI(_b,_c)))
static int ilog(unsigned _v){
int ret;
for(ret=0;_v;ret++)_v>>=1;
return ret;
}
void WriteQTables(PB_INSTANCE *pbi,oggpack_buffer* _opb) {
th_quant_info *_qinfo = &pbi->quant_info;
const th_quant_ranges *qranges;
const th_quant_base *base_mats[2*3*64];
int indices[2][3][64];
int nbase_mats;
int nbits;
int ci;
int qi;
int qri;
int qti;
int pli;
int qtj;
int plj;
int bmi;
int i;
/*Unlike the scale tables, we can't assume the maximum value will be in
index 0, so search for it here.*/
i=_qinfo->loop_filter_limits[0];
for(qi=1;qi<64;qi++)i=OC_MAXI(i,_qinfo->loop_filter_limits[qi]);
nbits=ilog(i);
oggpackB_write(_opb,nbits,3);
for(qi=0;qi<64;qi++){
oggpackB_write(_opb,_qinfo->loop_filter_limits[qi],nbits);
}
/* 580 bits for VP3.*/
nbits=OC_MAXI(ilog(_qinfo->ac_scale[0]),1);
oggpackB_write(_opb,nbits-1,4);
for(qi=0;qi<64;qi++)oggpackB_write(_opb,_qinfo->ac_scale[qi],nbits);
/* 516 bits for VP3.*/
nbits=OC_MAXI(ilog(_qinfo->dc_scale[0]),1);
oggpackB_write(_opb,nbits-1,4);
for(qi=0;qi<64;qi++)oggpackB_write(_opb,_qinfo->dc_scale[qi],nbits);
/*Consolidate any duplicate base matrices.*/
nbase_mats=0;
for(qti=0;qti<2;qti++)for(pli=0;pli<3;pli++){
qranges=_qinfo->qi_ranges[qti]+pli;
for(qri=0;qri<=qranges->nranges;qri++){
for(bmi=0;;bmi++){
if(bmi>=nbase_mats){
base_mats[bmi]=qranges->base_matrices+qri;
indices[qti][pli][qri]=nbase_mats++;
break;
}
else if(memcmp(base_mats[bmi][0],qranges->base_matrices[qri],
sizeof(base_mats[bmi][0]))==0){
indices[qti][pli][qri]=bmi;
break;
}
}
}
}
/*Write out the list of unique base matrices.
1545 bits for VP3 matrices.*/
oggpackB_write(_opb,nbase_mats-1,9);
for(bmi=0;bmi<nbase_mats;bmi++){
for(ci=0;ci<64;ci++)oggpackB_write(_opb,base_mats[bmi][0][ci],8);
}
/*Now store quant ranges and their associated indices into the base matrix
list.
46 bits for VP3 matrices.*/
nbits=ilog(nbase_mats-1);
for(i=0;i<6;i++){
qti=i/3;
pli=i%3;
qranges=_qinfo->qi_ranges[qti]+pli;
if(i>0){
if(qti>0){
if(qranges->nranges==_qinfo->qi_ranges[qti-1][pli].nranges&&
memcmp(qranges->sizes,_qinfo->qi_ranges[qti-1][pli].sizes,
qranges->nranges*sizeof(qranges->sizes[0]))==0&&
memcmp(indices[qti][pli],indices[qti-1][pli],
(qranges->nranges+1)*sizeof(indices[qti][pli][0]))==0){
oggpackB_write(_opb,1,2);
continue;
}
}
qtj=(i-1)/3;
plj=(i-1)%3;
if(qranges->nranges==_qinfo->qi_ranges[qtj][plj].nranges&&
memcmp(qranges->sizes,_qinfo->qi_ranges[qtj][plj].sizes,
qranges->nranges*sizeof(qranges->sizes[0]))==0&&
memcmp(indices[qti][pli],indices[qtj][plj],
(qranges->nranges+1)*sizeof(indices[qti][pli][0]))==0){
oggpackB_write(_opb,0,1+(qti>0));
continue;
}
oggpackB_write(_opb,1,1);
}
oggpackB_write(_opb,indices[qti][pli][0],nbits);
for(qi=qri=0;qi<63;qri++){
oggpackB_write(_opb,qranges->sizes[qri]-1,ilog(62-qi));
qi+=qranges->sizes[qri];
oggpackB_write(_opb,indices[qti][pli][qri+1],nbits);
}
}
}
/* a copied/reconciled version of derf's theora-exp code; redundancy
should be eliminated at some point */
void InitQTables( PB_INSTANCE *pbi ){
int qti; /* coding mode: intra or inter */
int pli; /* Y U V */
th_quant_info *qinfo = &pbi->quant_info;
pbi->QThreshTable = pbi->quant_info.ac_scale;
for(qti=0;qti<2;qti++){
for(pli=0;pli<3;pli++){
int qi; /* quality index */
int qri; /* range iterator */
for(qi=0,qri=0; qri<=qinfo->qi_ranges[qti][pli].nranges; qri++){
th_quant_base base;
ogg_uint32_t q;
int qi_start;
int qi_end;
int ci;
memcpy(base,qinfo->qi_ranges[qti][pli].base_matrices[qri],
sizeof(base));
qi_start=qi;
if(qri==qinfo->qi_ranges[qti][pli].nranges)
qi_end=qi+1;
else
qi_end=qi+qinfo->qi_ranges[qti][pli].sizes[qri];
/* Iterate over quality indicies in this range */
for(;;){
/*Scale DC the coefficient from the proper table.*/
q=((ogg_uint32_t)qinfo->dc_scale[qi]*base[0]/100)<<2;
q=OC_CLAMPI(DC_QUANT_MIN[qti],q,OC_QUANT_MAX);
pbi->quant_tables[qti][pli][qi][0]=(ogg_uint16_t)q;
/*Now scale AC coefficients from the proper table.*/
for(ci=1;ci<64;ci++){
q=((ogg_uint32_t)qinfo->ac_scale[qi]*base[ci]/100)<<2;
q=OC_CLAMPI(AC_QUANT_MIN[qti],q,OC_QUANT_MAX);
pbi->quant_tables[qti][pli][qi][ci]=(ogg_uint16_t)q;
}
if(++qi>=qi_end)break;
/*Interpolate the next base matrix.*/
for(ci=0;ci<64;ci++){
base[ci]=(unsigned char)
((2*((qi_end-qi)*qinfo->qi_ranges[qti][pli].base_matrices[qri][ci]+
(qi-qi_start)*qinfo->qi_ranges[qti][pli].base_matrices[qri+1][ci])
+qinfo->qi_ranges[qti][pli].sizes[qri])/
(2*qinfo->qi_ranges[qti][pli].sizes[qri]));
}
}
}
}
}
}
static void BuildZigZagIndex(PB_INSTANCE *pbi){
ogg_int32_t i,j;
/* invert the row to zigzag coeffient order lookup table */
for ( i = 0; i < BLOCK_SIZE; i++ ){
j = dezigzag_index[i];
pbi->zigzag_index[j] = i;
}
}
static void init_quantizer ( CP_INSTANCE *cpi,
unsigned char QIndex ){
int i;
double ZBinFactor;
double RoundingFactor;
double temp_fp_quant_coeffs;
double temp_fp_quant_round;
double temp_fp_ZeroBinSize;
PB_INSTANCE *pbi = &cpi->pb;
const ogg_uint16_t * temp_Y_coeffs;
const ogg_uint16_t * temp_U_coeffs;
const ogg_uint16_t * temp_V_coeffs;
const ogg_uint16_t * temp_Inter_Y_coeffs;
const ogg_uint16_t * temp_Inter_U_coeffs;
const ogg_uint16_t * temp_Inter_V_coeffs;
ogg_uint16_t scale_factor = cpi->pb.quant_info.ac_scale[QIndex];
/* Notes on setup of quantisers. The initial multiplication by
the scale factor is done in the ogg_int32_t domain to insure that the
precision in the quantiser is the same as in the inverse
quantiser where all calculations are integer. The "<< 2" is a
normalisation factor for the forward DCT transform. */
temp_Y_coeffs = pbi->quant_tables[0][0][QIndex];
temp_U_coeffs = pbi->quant_tables[0][1][QIndex];
temp_V_coeffs = pbi->quant_tables[0][2][QIndex];
temp_Inter_Y_coeffs = pbi->quant_tables[1][0][QIndex];
temp_Inter_U_coeffs = pbi->quant_tables[1][1][QIndex];
temp_Inter_V_coeffs = pbi->quant_tables[1][2][QIndex];
ZBinFactor = 0.9;
switch(cpi->pb.info.sharpness){
case 0:
ZBinFactor = 0.65;
if ( scale_factor <= 50 )
RoundingFactor = 0.499;
else
RoundingFactor = 0.46;
break;
case 1:
ZBinFactor = 0.75;
if ( scale_factor <= 50 )
RoundingFactor = 0.476;
else
RoundingFactor = 0.400;
break;
default:
ZBinFactor = 0.9;
if ( scale_factor <= 50 )
RoundingFactor = 0.476;
else
RoundingFactor = 0.333;
break;
}
/* Use fixed multiplier for intra Y DC */
temp_fp_quant_coeffs = temp_Y_coeffs[0];
temp_fp_quant_round = temp_fp_quant_coeffs * RoundingFactor;
pbi->fp_quant_Y_round[0] = (ogg_int32_t) (0.5 + temp_fp_quant_round);
temp_fp_ZeroBinSize = temp_fp_quant_coeffs * ZBinFactor;
pbi->fp_ZeroBinSize_Y[0] = (ogg_int32_t) (0.5 + temp_fp_ZeroBinSize);
temp_fp_quant_coeffs = 1.0 / temp_fp_quant_coeffs;
pbi->fp_quant_Y_coeffs[0] = (0.5 + SHIFT16 * temp_fp_quant_coeffs);
/* Intra U */
temp_fp_quant_coeffs = temp_U_coeffs[0];
temp_fp_quant_round = temp_fp_quant_coeffs * RoundingFactor;
pbi->fp_quant_U_round[0] = (0.5 + temp_fp_quant_round);
temp_fp_ZeroBinSize = temp_fp_quant_coeffs * ZBinFactor;
pbi->fp_ZeroBinSize_U[0] = (0.5 + temp_fp_ZeroBinSize);
temp_fp_quant_coeffs = 1.0 / temp_fp_quant_coeffs;
pbi->fp_quant_U_coeffs[0]= (0.5 + SHIFT16 * temp_fp_quant_coeffs);
/* Intra V */
temp_fp_quant_coeffs = temp_V_coeffs[0];
temp_fp_quant_round = temp_fp_quant_coeffs * RoundingFactor;
pbi->fp_quant_V_round[0] = (0.5 + temp_fp_quant_round);
temp_fp_ZeroBinSize = temp_fp_quant_coeffs * ZBinFactor;
pbi->fp_ZeroBinSize_V[0] = (0.5 + temp_fp_ZeroBinSize);
temp_fp_quant_coeffs = 1.0 / temp_fp_quant_coeffs;
pbi->fp_quant_V_coeffs[0]= (0.5 + SHIFT16 * temp_fp_quant_coeffs);
/* Inter Y */
temp_fp_quant_coeffs = temp_Inter_Y_coeffs[0];
temp_fp_quant_round = temp_fp_quant_coeffs * RoundingFactor;
pbi->fp_quant_Inter_Y_round[0]= (0.5 + temp_fp_quant_round);
temp_fp_ZeroBinSize = temp_fp_quant_coeffs * ZBinFactor;
pbi->fp_ZeroBinSize_Inter_Y[0]= (0.5 + temp_fp_ZeroBinSize);
temp_fp_quant_coeffs= 1.0 / temp_fp_quant_coeffs;
pbi->fp_quant_Inter_Y_coeffs[0]= (0.5 + SHIFT16 * temp_fp_quant_coeffs);
/* Inter U */
temp_fp_quant_coeffs = temp_Inter_U_coeffs[0];
temp_fp_quant_round = temp_fp_quant_coeffs * RoundingFactor;
pbi->fp_quant_Inter_U_round[0]= (0.5 + temp_fp_quant_round);
temp_fp_ZeroBinSize = temp_fp_quant_coeffs * ZBinFactor;
pbi->fp_ZeroBinSize_Inter_U[0]= (0.5 + temp_fp_ZeroBinSize);
temp_fp_quant_coeffs= 1.0 / temp_fp_quant_coeffs;
pbi->fp_quant_Inter_U_coeffs[0]= (0.5 + SHIFT16 * temp_fp_quant_coeffs);
/* Inter V */
temp_fp_quant_coeffs = temp_Inter_V_coeffs[0];
temp_fp_quant_round = temp_fp_quant_coeffs * RoundingFactor;
pbi->fp_quant_Inter_V_round[0]= (0.5 + temp_fp_quant_round);
temp_fp_ZeroBinSize = temp_fp_quant_coeffs * ZBinFactor;
pbi->fp_ZeroBinSize_Inter_V[0]= (0.5 + temp_fp_ZeroBinSize);
temp_fp_quant_coeffs= 1.0 / temp_fp_quant_coeffs;
pbi->fp_quant_Inter_V_coeffs[0]= (0.5 + SHIFT16 * temp_fp_quant_coeffs);
for ( i = 1; i < 64; i++ ){
/* Intra Y */
temp_fp_quant_coeffs = temp_Y_coeffs[i];
temp_fp_quant_round = temp_fp_quant_coeffs * RoundingFactor;
pbi->fp_quant_Y_round[i] = (0.5 + temp_fp_quant_round);
temp_fp_ZeroBinSize = temp_fp_quant_coeffs * ZBinFactor;
pbi->fp_ZeroBinSize_Y[i] = (0.5 + temp_fp_ZeroBinSize);
temp_fp_quant_coeffs = 1.0 / temp_fp_quant_coeffs;
pbi->fp_quant_Y_coeffs[i] = (0.5 + SHIFT16 * temp_fp_quant_coeffs);
/* Intra U */
temp_fp_quant_coeffs = temp_U_coeffs[i];
temp_fp_quant_round = temp_fp_quant_coeffs * RoundingFactor;
pbi->fp_quant_U_round[i] = (0.5 + temp_fp_quant_round);
temp_fp_ZeroBinSize = temp_fp_quant_coeffs * ZBinFactor;
pbi->fp_ZeroBinSize_U[i] = (0.5 + temp_fp_ZeroBinSize);
temp_fp_quant_coeffs = 1.0 / temp_fp_quant_coeffs;
pbi->fp_quant_U_coeffs[i]= (0.5 + SHIFT16 * temp_fp_quant_coeffs);
/* Intra V */
temp_fp_quant_coeffs = temp_V_coeffs[i];
temp_fp_quant_round = temp_fp_quant_coeffs * RoundingFactor;
pbi->fp_quant_V_round[i] = (0.5 + temp_fp_quant_round);
temp_fp_ZeroBinSize = temp_fp_quant_coeffs * ZBinFactor;
pbi->fp_ZeroBinSize_V[i] = (0.5 + temp_fp_ZeroBinSize);
temp_fp_quant_coeffs = 1.0 / temp_fp_quant_coeffs;
pbi->fp_quant_V_coeffs[i]= (0.5 + SHIFT16 * temp_fp_quant_coeffs);
/* Inter Y */
temp_fp_quant_coeffs = temp_Inter_Y_coeffs[i];
temp_fp_quant_round = temp_fp_quant_coeffs * RoundingFactor;
pbi->fp_quant_Inter_Y_round[i]= (0.5 + temp_fp_quant_round);
temp_fp_ZeroBinSize = temp_fp_quant_coeffs * ZBinFactor;
pbi->fp_ZeroBinSize_Inter_Y[i]= (0.5 + temp_fp_ZeroBinSize);
temp_fp_quant_coeffs = 1.0 / temp_fp_quant_coeffs;
pbi->fp_quant_Inter_Y_coeffs[i]= (0.5 + SHIFT16 * temp_fp_quant_coeffs);
/* Inter U */
temp_fp_quant_coeffs = temp_Inter_U_coeffs[i];
temp_fp_quant_round = temp_fp_quant_coeffs * RoundingFactor;
pbi->fp_quant_Inter_U_round[i]= (0.5 + temp_fp_quant_round);
temp_fp_ZeroBinSize = temp_fp_quant_coeffs * ZBinFactor;
pbi->fp_ZeroBinSize_Inter_U[i]= (0.5 + temp_fp_ZeroBinSize);
temp_fp_quant_coeffs = 1.0 / temp_fp_quant_coeffs;
pbi->fp_quant_Inter_U_coeffs[i]= (0.5 + SHIFT16 * temp_fp_quant_coeffs);
/* Inter V */
temp_fp_quant_coeffs = temp_Inter_V_coeffs[i];
temp_fp_quant_round = temp_fp_quant_coeffs * RoundingFactor;
pbi->fp_quant_Inter_V_round[i]= (0.5 + temp_fp_quant_round);
temp_fp_ZeroBinSize = temp_fp_quant_coeffs * ZBinFactor;
pbi->fp_ZeroBinSize_Inter_V[i]= (0.5 + temp_fp_ZeroBinSize);
temp_fp_quant_coeffs = 1.0 / temp_fp_quant_coeffs;
pbi->fp_quant_Inter_V_coeffs[i]= (0.5 + SHIFT16 * temp_fp_quant_coeffs);
}
pbi->fquant_coeffs = pbi->fp_quant_Y_coeffs;
}
void select_quantiser(PB_INSTANCE *pbi, int type) {
/* select a quantiser according to what plane has to be coded in what
* mode. Could be extended to a more sophisticated scheme. */
switch(type) {
case BLOCK_Y:
pbi->fquant_coeffs = pbi->fp_quant_Y_coeffs;
pbi->fquant_round = pbi->fp_quant_Y_round;
pbi->fquant_ZbSize = pbi->fp_ZeroBinSize_Y;
break;
case BLOCK_U:
pbi->fquant_coeffs = pbi->fp_quant_U_coeffs;
pbi->fquant_round = pbi->fp_quant_U_round;
pbi->fquant_ZbSize = pbi->fp_ZeroBinSize_U;
break;
case BLOCK_V:
pbi->fquant_coeffs = pbi->fp_quant_V_coeffs;
pbi->fquant_round = pbi->fp_quant_V_round;
pbi->fquant_ZbSize = pbi->fp_ZeroBinSize_V;
break;
case BLOCK_INTER_Y:
pbi->fquant_coeffs = pbi->fp_quant_Inter_Y_coeffs;
pbi->fquant_round = pbi->fp_quant_Inter_Y_round;
pbi->fquant_ZbSize = pbi->fp_ZeroBinSize_Inter_Y;
break;
case BLOCK_INTER_U:
pbi->fquant_coeffs = pbi->fp_quant_Inter_U_coeffs;
pbi->fquant_round = pbi->fp_quant_Inter_U_round;
pbi->fquant_ZbSize = pbi->fp_ZeroBinSize_Inter_U;
break;
case BLOCK_INTER_V:
pbi->fquant_coeffs = pbi->fp_quant_Inter_V_coeffs;
pbi->fquant_round = pbi->fp_quant_Inter_V_round;
pbi->fquant_ZbSize = pbi->fp_ZeroBinSize_Inter_V;
break;
}
}
void quantize( PB_INSTANCE *pbi,
ogg_int16_t * DCT_block,
Q_LIST_ENTRY * quantized_list){
ogg_uint32_t i; /* Row index */
Q_LIST_ENTRY val; /* Quantised value. */
ogg_int32_t * FquantRoundPtr = pbi->fquant_round;
ogg_int32_t * FquantCoeffsPtr = pbi->fquant_coeffs;
ogg_int32_t * FquantZBinSizePtr = pbi->fquant_ZbSize;
ogg_int16_t * DCT_blockPtr = DCT_block;
ogg_uint32_t * ZigZagPtr = (ogg_uint32_t *)pbi->zigzag_index;
ogg_int32_t temp;
/* Set the quantized_list to default to 0 */
memset( quantized_list, 0, 64 * sizeof(Q_LIST_ENTRY) );
/* Note that we add half divisor to effect rounding on positive number */
for( i = 0; i < VFRAGPIXELS; i++) {
int col;
/* Iterate through columns */
for( col = 0; col < 8; col++) {
if ( DCT_blockPtr[col] >= FquantZBinSizePtr[col] ) {
temp = FquantCoeffsPtr[col] * ( DCT_blockPtr[col] + FquantRoundPtr[col] ) ;
val = (Q_LIST_ENTRY) (temp>>16);
quantized_list[ZigZagPtr[col]] = ( val > 511 ) ? 511 : val;
} else if ( DCT_blockPtr[col] <= -FquantZBinSizePtr[col] ) {
temp = FquantCoeffsPtr[col] *
( DCT_blockPtr[col] - FquantRoundPtr[col] ) + MIN16;
val = (Q_LIST_ENTRY) (temp>>16);
quantized_list[ZigZagPtr[col]] = ( val < -511 ) ? -511 : val;
}
}
FquantRoundPtr += 8;
FquantCoeffsPtr += 8;
FquantZBinSizePtr += 8;
DCT_blockPtr += 8;
ZigZagPtr += 8;
}
}
static void init_dequantizer ( PB_INSTANCE *pbi,
unsigned char QIndex ){
int i, j;
ogg_uint16_t * InterY_coeffs;
ogg_uint16_t * InterU_coeffs;
ogg_uint16_t * InterV_coeffs;
ogg_uint16_t * Y_coeffs;
ogg_uint16_t * U_coeffs;
ogg_uint16_t * V_coeffs;
Y_coeffs = pbi->quant_tables[0][0][QIndex];
U_coeffs = pbi->quant_tables[0][1][QIndex];
V_coeffs = pbi->quant_tables[0][2][QIndex];
InterY_coeffs = pbi->quant_tables[1][0][QIndex];
InterU_coeffs = pbi->quant_tables[1][1][QIndex];
InterV_coeffs = pbi->quant_tables[1][2][QIndex];
/* invert the dequant index into the quant index
the dxer has a different order than the cxer. */
BuildZigZagIndex(pbi);
/* Reorder dequantisation coefficients into dct zigzag order. */
for ( i = 0; i < BLOCK_SIZE; i++ ) {
j = pbi->zigzag_index[i];
pbi->dequant_Y_coeffs[j] = Y_coeffs[i];
}
for ( i = 0; i < BLOCK_SIZE; i++ ) {
j = pbi->zigzag_index[i];
pbi->dequant_U_coeffs[j] = U_coeffs[i];
}
for ( i = 0; i < BLOCK_SIZE; i++ ) {
j = pbi->zigzag_index[i];
pbi->dequant_V_coeffs[j] = V_coeffs[i];
}
for ( i = 0; i < BLOCK_SIZE; i++ ){
j = pbi->zigzag_index[i];
pbi->dequant_InterY_coeffs[j] = InterY_coeffs[i];
}
for ( i = 0; i < BLOCK_SIZE; i++ ){
j = pbi->zigzag_index[i];
pbi->dequant_InterU_coeffs[j] = InterU_coeffs[i];
}
for ( i = 0; i < BLOCK_SIZE; i++ ){
j = pbi->zigzag_index[i];
pbi->dequant_InterV_coeffs[j] = InterV_coeffs[i];
}
pbi->dequant_coeffs = pbi->dequant_Y_coeffs;
}
void UpdateQ( PB_INSTANCE *pbi, int NewQIndex ){
ogg_uint32_t qscale;
/* clamp to legal bounds */
if (NewQIndex >= Q_TABLE_SIZE) NewQIndex = Q_TABLE_SIZE - 1;
else if (NewQIndex < 0) NewQIndex = 0;
pbi->FrameQIndex = NewQIndex;
qscale = pbi->quant_info.ac_scale[NewQIndex];
pbi->ThisFrameQualityValue = qscale;
/* Re-initialise the Q tables for forward and reverse transforms. */
init_dequantizer ( pbi, (unsigned char) pbi->FrameQIndex );
}
void UpdateQC( CP_INSTANCE *cpi, ogg_uint32_t NewQ ){
ogg_uint32_t qscale;
PB_INSTANCE *pbi = &cpi->pb;
/* Do bounds checking and convert to a float. */
qscale = NewQ;
if ( qscale < pbi->quant_info.ac_scale[Q_TABLE_SIZE-1] )
qscale = pbi->quant_info.ac_scale[Q_TABLE_SIZE-1];
else if ( qscale > pbi->quant_info.ac_scale[0] )
qscale = pbi->quant_info.ac_scale[0];
/* Set the inter/intra descision control variables. */
pbi->FrameQIndex = Q_TABLE_SIZE - 1;
while ((ogg_int32_t) pbi->FrameQIndex >= 0 ) {
if ( (pbi->FrameQIndex == 0) ||
( pbi->quant_info.ac_scale[pbi->FrameQIndex] >= NewQ) )
break;
pbi->FrameQIndex --;
}
/* Re-initialise the Q tables for forward and reverse transforms. */
init_quantizer ( cpi, pbi->FrameQIndex );
init_dequantizer ( pbi, pbi->FrameQIndex );
}

1447
Engine/lib/libtheora/lib/enc/encoder_toplevel.c Normal file
View file

File diff suppressed because it is too large Load diff

243
Engine/lib/libtheora/lib/enc/frarray.c Normal file
View file

@ -0,0 +1,243 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: frarray.c 15153 2008-08-04 18:37:55Z tterribe $
********************************************************************/
#include <string.h>
#include "codec_internal.h"
#include "block_inline.h"
/* Long run bit string coding */
static ogg_uint32_t FrArrayCodeSBRun( CP_INSTANCE *cpi, ogg_uint32_t value){
ogg_uint32_t CodedVal = 0;
ogg_uint32_t CodedBits = 0;
/* Coding scheme:
Codeword RunLength
0 1
10x 2-3
110x 4-5
1110xx 6-9
11110xxx 10-17
111110xxxx 18-33
111111xxxxxxxxxxxx 34-4129 */
if ( value == 1 ){
CodedVal = 0;
CodedBits = 1;
} else if ( value <= 3 ) {
CodedVal = 0x0004 + (value - 2);
CodedBits = 3;
} else if ( value <= 5 ) {
CodedVal = 0x000C + (value - 4);
CodedBits = 4;
} else if ( value <= 9 ) {
CodedVal = 0x0038 + (value - 6);
CodedBits = 6;
} else if ( value <= 17 ) {
CodedVal = 0x00F0 + (value - 10);
CodedBits = 8;
} else if ( value <= 33 ) {
CodedVal = 0x03E0 + (value - 18);
CodedBits = 10;
} else {
CodedVal = 0x3F000 + (value - 34);
CodedBits = 18;
}
/* Add the bits to the encode holding buffer. */
oggpackB_write( cpi->oggbuffer, CodedVal, CodedBits );
return CodedBits;
}
/* Short run bit string coding */
static ogg_uint32_t FrArrayCodeBlockRun( CP_INSTANCE *cpi,
ogg_uint32_t value ) {
ogg_uint32_t CodedVal = 0;
ogg_uint32_t CodedBits = 0;
/* Coding scheme:
Codeword RunLength
0x 1-2
10x 3-4
110x 5-6
1110xx 7-10
11110xx 11-14
11111xxxx 15-30 */
if ( value <= 2 ) {
CodedVal = value - 1;
CodedBits = 2;
} else if ( value <= 4 ) {
CodedVal = 0x0004 + (value - 3);
CodedBits = 3;
} else if ( value <= 6 ) {
CodedVal = 0x000C + (value - 5);
CodedBits = 4;
} else if ( value <= 10 ) {
CodedVal = 0x0038 + (value - 7);
CodedBits = 6;
} else if ( value <= 14 ) {
CodedVal = 0x0078 + (value - 11);
CodedBits = 7;
} else {
CodedVal = 0x01F0 + (value - 15);
CodedBits = 9;
}
/* Add the bits to the encode holding buffer. */
oggpackB_write( cpi->oggbuffer, CodedVal, CodedBits );
return CodedBits;
}
void PackAndWriteDFArray( CP_INSTANCE *cpi ){
ogg_uint32_t i;
unsigned char val;
ogg_uint32_t run_count;
ogg_uint32_t SB, MB, B; /* Block, MB and SB loop variables */
ogg_uint32_t BListIndex = 0;
ogg_uint32_t LastSbBIndex = 0;
ogg_int32_t DfBlockIndex; /* Block index in display_fragments */
/* Initialise workspaces */
memset( cpi->pb.SBFullyFlags, 1, cpi->pb.SuperBlocks);
memset( cpi->pb.SBCodedFlags, 0, cpi->pb.SuperBlocks );
memset( cpi->PartiallyCodedFlags, 0, cpi->pb.SuperBlocks );
memset( cpi->BlockCodedFlags, 0, cpi->pb.UnitFragments);
for( SB = 0; SB < cpi->pb.SuperBlocks; SB++ ) {
/* Check for coded blocks and macro-blocks */
for ( MB=0; MB<4; MB++ ) {
/* If MB in frame */
if ( QuadMapToMBTopLeft(cpi->pb.BlockMap,SB,MB) >= 0 ) {
for ( B=0; B<4; B++ ) {
DfBlockIndex = QuadMapToIndex1( cpi->pb.BlockMap,SB, MB, B );
/* Does Block lie in frame: */
if ( DfBlockIndex >= 0 ) {
/* In Frame: If it is not coded then this SB is only
partly coded.: */
if ( cpi->pb.display_fragments[DfBlockIndex] ) {
cpi->pb.SBCodedFlags[SB] = 1; /* SB at least partly coded */
cpi->BlockCodedFlags[BListIndex] = 1; /* Block is coded */
}else{
cpi->pb.SBFullyFlags[SB] = 0; /* SB not fully coded */
cpi->BlockCodedFlags[BListIndex] = 0; /* Block is not coded */
}
BListIndex++;
}
}
}
}
/* Is the SB fully coded or uncoded.
If so then backup BListIndex and MBListIndex */
if ( cpi->pb.SBFullyFlags[SB] || !cpi->pb.SBCodedFlags[SB] ) {
BListIndex = LastSbBIndex; /* Reset to values from previous SB */
}else{
cpi->PartiallyCodedFlags[SB] = 1; /* Set up list of partially
coded SBs */
LastSbBIndex = BListIndex;
}
}
/* Code list of partially coded Super-Block. */
val = cpi->PartiallyCodedFlags[0];
oggpackB_write( cpi->oggbuffer, (ogg_uint32_t)val, 1);
i = 0;
while ( i < cpi->pb.SuperBlocks ) {
run_count = 0;
while ( (i<cpi->pb.SuperBlocks) &&
(cpi->PartiallyCodedFlags[i]==val) &&
run_count<4129 ) {
i++;
run_count++;
}
/* Code the run */
FrArrayCodeSBRun( cpi, run_count);
if(run_count >= 4129 && i < cpi->pb.SuperBlocks ){
val = cpi->PartiallyCodedFlags[i];
oggpackB_write( cpi->oggbuffer, (ogg_uint32_t)val, 1);
}else
val = ( val == 0 ) ? 1 : 0;
}
/* RLC Super-Block fully/not coded. */
i = 0;
/* Skip partially coded blocks */
while( (i < cpi->pb.SuperBlocks) && cpi->PartiallyCodedFlags[i] )
i++;
if ( i < cpi->pb.SuperBlocks ) {
val = cpi->pb.SBFullyFlags[i];
oggpackB_write( cpi->oggbuffer, (ogg_uint32_t)val, 1);
while ( i < cpi->pb.SuperBlocks ) {
run_count = 0;
while ( (i < cpi->pb.SuperBlocks) &&
(cpi->pb.SBFullyFlags[i] == val) &&
run_count < 4129) {
i++;
/* Skip partially coded blocks */
while( (i < cpi->pb.SuperBlocks) && cpi->PartiallyCodedFlags[i] )
i++;
run_count++;
}
/* Code the run */
FrArrayCodeSBRun( cpi, run_count );
if(run_count >= 4129 && i < cpi->pb.SuperBlocks ){
val = cpi->PartiallyCodedFlags[i];
oggpackB_write( cpi->oggbuffer, (ogg_uint32_t)val, 1);
}else
val = ( val == 0 ) ? 1 : 0;
}
}
/* Now code the block flags */
if ( BListIndex > 0 ) {
/* Code the block flags start value */
val = cpi->BlockCodedFlags[0];
oggpackB_write( cpi->oggbuffer, (ogg_uint32_t)val, 1);
/* Now code the block flags. */
for ( i = 0; i < BListIndex; ) {
run_count = 0;
while ( (i < BListIndex) && (cpi->BlockCodedFlags[i] == val) ) {
i++;
run_count++;
}
FrArrayCodeBlockRun( cpi, run_count );
val = ( val == 0 ) ? 1 : 0;
}
}
}

392
Engine/lib/libtheora/lib/enc/frinit.c Normal file
View file

@ -0,0 +1,392 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: frinit.c 15153 2008-08-04 18:37:55Z tterribe $
********************************************************************/
#include <stdlib.h>
#include "codec_internal.h"
void InitializeFragCoordinates(PB_INSTANCE *pbi){
ogg_uint32_t i, j;
ogg_uint32_t HorizFrags = pbi->HFragments;
ogg_uint32_t VertFrags = pbi->VFragments;
ogg_uint32_t StartFrag = 0;
/* Y */
for(i = 0; i< VertFrags; i++){
for(j = 0; j< HorizFrags; j++){
ogg_uint32_t ThisFrag = i * HorizFrags + j;
pbi->FragCoordinates[ ThisFrag ].x=j * BLOCK_HEIGHT_WIDTH;
pbi->FragCoordinates[ ThisFrag ].y=i * BLOCK_HEIGHT_WIDTH;
}
}
/* U */
HorizFrags >>= 1;
VertFrags >>= 1;
StartFrag = pbi->YPlaneFragments;
for(i = 0; i< VertFrags; i++) {
for(j = 0; j< HorizFrags; j++) {
ogg_uint32_t ThisFrag = StartFrag + i * HorizFrags + j;
pbi->FragCoordinates[ ThisFrag ].x=j * BLOCK_HEIGHT_WIDTH;
pbi->FragCoordinates[ ThisFrag ].y=i * BLOCK_HEIGHT_WIDTH;
}
}
/* V */
StartFrag = pbi->YPlaneFragments + pbi->UVPlaneFragments;
for(i = 0; i< VertFrags; i++) {
for(j = 0; j< HorizFrags; j++) {
ogg_uint32_t ThisFrag = StartFrag + i * HorizFrags + j;
pbi->FragCoordinates[ ThisFrag ].x=j * BLOCK_HEIGHT_WIDTH;
pbi->FragCoordinates[ ThisFrag ].y=i * BLOCK_HEIGHT_WIDTH;
}
}
}
static void CalcPixelIndexTable( PB_INSTANCE *pbi){
ogg_uint32_t i;
ogg_uint32_t * PixelIndexTablePtr;
/* Calculate the pixel index table for normal image buffers */
PixelIndexTablePtr = pbi->pixel_index_table;
for ( i = 0; i < pbi->YPlaneFragments; i++ ) {
PixelIndexTablePtr[ i ] =
((i / pbi->HFragments) * VFRAGPIXELS *
pbi->info.width);
PixelIndexTablePtr[ i ] +=
((i % pbi->HFragments) * HFRAGPIXELS);
}
PixelIndexTablePtr = &pbi->pixel_index_table[pbi->YPlaneFragments];
for ( i = 0; i < ((pbi->HFragments >> 1) * pbi->VFragments); i++ ) {
PixelIndexTablePtr[ i ] =
((i / (pbi->HFragments / 2) ) *
(VFRAGPIXELS *
(pbi->info.width / 2)) );
PixelIndexTablePtr[ i ] +=
((i % (pbi->HFragments / 2) ) *
HFRAGPIXELS) + pbi->YPlaneSize;
}
/************************************************************************/
/* Now calculate the pixel index table for image reconstruction buffers */
PixelIndexTablePtr = pbi->recon_pixel_index_table;
for ( i = 0; i < pbi->YPlaneFragments; i++ ){
PixelIndexTablePtr[ i ] =
((i / pbi->HFragments) * VFRAGPIXELS *
pbi->YStride);
PixelIndexTablePtr[ i ] +=
((i % pbi->HFragments) * HFRAGPIXELS) +
pbi->ReconYDataOffset;
}
/* U blocks */
PixelIndexTablePtr = &pbi->recon_pixel_index_table[pbi->YPlaneFragments];
for ( i = 0; i < pbi->UVPlaneFragments; i++ ) {
PixelIndexTablePtr[ i ] =
((i / (pbi->HFragments / 2) ) *
(VFRAGPIXELS * (pbi->UVStride)) );
PixelIndexTablePtr[ i ] +=
((i % (pbi->HFragments / 2) ) *
HFRAGPIXELS) + pbi->ReconUDataOffset;
}
/* V blocks */
PixelIndexTablePtr =
&pbi->recon_pixel_index_table[pbi->YPlaneFragments +
pbi->UVPlaneFragments];
for ( i = 0; i < pbi->UVPlaneFragments; i++ ) {
PixelIndexTablePtr[ i ] =
((i / (pbi->HFragments / 2) ) *
(VFRAGPIXELS * (pbi->UVStride)) );
PixelIndexTablePtr[ i ] +=
((i % (pbi->HFragments / 2) ) * HFRAGPIXELS) +
pbi->ReconVDataOffset;
}
}
void ClearFragmentInfo(PB_INSTANCE * pbi){
/* free prior allocs if present */
if(pbi->display_fragments) _ogg_free(pbi->display_fragments);
if(pbi->pixel_index_table) _ogg_free(pbi->pixel_index_table);
if(pbi->recon_pixel_index_table) _ogg_free(pbi->recon_pixel_index_table);
if(pbi->FragTokenCounts) _ogg_free(pbi->FragTokenCounts);
if(pbi->CodedBlockList) _ogg_free(pbi->CodedBlockList);
if(pbi->FragMVect) _ogg_free(pbi->FragMVect);
if(pbi->FragCoeffs) _ogg_free(pbi->FragCoeffs);
if(pbi->FragCoefEOB) _ogg_free(pbi->FragCoefEOB);
if(pbi->skipped_display_fragments) _ogg_free(pbi->skipped_display_fragments);
if(pbi->QFragData) _ogg_free(pbi->QFragData);
if(pbi->TokenList) _ogg_free(pbi->TokenList);
if(pbi->FragCodingMethod) _ogg_free(pbi->FragCodingMethod);
if(pbi->FragCoordinates) _ogg_free(pbi->FragCoordinates);
if(pbi->FragQIndex) _ogg_free(pbi->FragQIndex);
if(pbi->PPCoefBuffer) _ogg_free(pbi->PPCoefBuffer);
if(pbi->FragmentVariances) _ogg_free(pbi->FragmentVariances);
if(pbi->BlockMap) _ogg_free(pbi->BlockMap);
if(pbi->SBCodedFlags) _ogg_free(pbi->SBCodedFlags);
if(pbi->SBFullyFlags) _ogg_free(pbi->SBFullyFlags);
if(pbi->MBFullyFlags) _ogg_free(pbi->MBFullyFlags);
if(pbi->MBCodedFlags) _ogg_free(pbi->MBCodedFlags);
if(pbi->_Nodes) _ogg_free(pbi->_Nodes);
pbi->_Nodes = 0;
pbi->QFragData = 0;
pbi->TokenList = 0;
pbi->skipped_display_fragments = 0;
pbi->FragCoeffs = 0;
pbi->FragCoefEOB = 0;
pbi->display_fragments = 0;
pbi->pixel_index_table = 0;
pbi->recon_pixel_index_table = 0;
pbi->FragTokenCounts = 0;
pbi->CodedBlockList = 0;
pbi->FragCodingMethod = 0;
pbi->FragMVect = 0;
pbi->MBCodedFlags = 0;
pbi->MBFullyFlags = 0;
pbi->BlockMap = 0;
pbi->SBCodedFlags = 0;
pbi->SBFullyFlags = 0;
pbi->QFragData = 0;
pbi->TokenList = 0;
pbi->skipped_display_fragments = 0;
pbi->FragCoeffs = 0;
pbi->FragCoefEOB = 0;
pbi->display_fragments = 0;
pbi->pixel_index_table = 0;
pbi->recon_pixel_index_table = 0;
pbi->FragTokenCounts = 0;
pbi->CodedBlockList = 0;
pbi->FragCodingMethod = 0;
pbi->FragCoordinates = 0;
pbi->FragMVect = 0;
pbi->PPCoefBuffer=0;
pbi->PPCoefBuffer=0;
pbi->FragQIndex = 0;
pbi->FragQIndex = 0;
pbi->FragmentVariances= 0;
pbi->FragmentVariances = 0 ;
}
void InitFragmentInfo(PB_INSTANCE * pbi){
/* clear any existing info */
ClearFragmentInfo(pbi);
/* Perform Fragment Allocations */
pbi->display_fragments =
_ogg_malloc(pbi->UnitFragments * sizeof(*pbi->display_fragments));
pbi->pixel_index_table =
_ogg_malloc(pbi->UnitFragments * sizeof(*pbi->pixel_index_table));
pbi->recon_pixel_index_table =
_ogg_malloc(pbi->UnitFragments * sizeof(*pbi->recon_pixel_index_table));
pbi->FragTokenCounts =
_ogg_malloc(pbi->UnitFragments * sizeof(*pbi->FragTokenCounts));
pbi->CodedBlockList =
_ogg_malloc(pbi->UnitFragments * sizeof(*pbi->CodedBlockList));
pbi->FragMVect =
_ogg_malloc(pbi->UnitFragments * sizeof(*pbi->FragMVect));
pbi->FragCoeffs =
_ogg_malloc(pbi->UnitFragments * sizeof(*pbi->FragCoeffs));
pbi->FragCoefEOB =
_ogg_malloc(pbi->UnitFragments * sizeof(*pbi->FragCoefEOB));
pbi->skipped_display_fragments =
_ogg_malloc(pbi->UnitFragments * sizeof(*pbi->skipped_display_fragments));
pbi->QFragData =
_ogg_malloc(pbi->UnitFragments * sizeof(*pbi->QFragData));
pbi->TokenList =
_ogg_malloc(pbi->UnitFragments * sizeof(*pbi->TokenList));
pbi->FragCodingMethod =
_ogg_malloc(pbi->UnitFragments * sizeof(*pbi->FragCodingMethod));
pbi->FragCoordinates =
_ogg_malloc(pbi->UnitFragments * sizeof(*pbi->FragCoordinates));
pbi->FragQIndex =
_ogg_malloc(pbi->UnitFragments * sizeof(*pbi->FragQIndex));
pbi->PPCoefBuffer =
_ogg_malloc(pbi->UnitFragments * sizeof(*pbi->PPCoefBuffer));
pbi->FragmentVariances =
_ogg_malloc(pbi->UnitFragments * sizeof(*pbi->FragmentVariances));
pbi->_Nodes =
_ogg_malloc(pbi->UnitFragments * sizeof(*pbi->_Nodes));
/* Super Block Initialization */
pbi->SBCodedFlags =
_ogg_malloc(pbi->SuperBlocks * sizeof(*pbi->SBCodedFlags));
pbi->SBFullyFlags =
_ogg_malloc(pbi->SuperBlocks * sizeof(*pbi->SBFullyFlags));
/* Macro Block Initialization */
pbi->MBCodedFlags =
_ogg_malloc(pbi->MacroBlocks * sizeof(*pbi->MBCodedFlags));
pbi->MBFullyFlags =
_ogg_malloc(pbi->MacroBlocks * sizeof(*pbi->MBFullyFlags));
pbi->BlockMap =
_ogg_malloc(pbi->SuperBlocks * sizeof(*pbi->BlockMap));
}
void ClearFrameInfo(PB_INSTANCE * pbi){
if(pbi->ThisFrameRecon )
_ogg_free(pbi->ThisFrameRecon );
if(pbi->GoldenFrame)
_ogg_free(pbi->GoldenFrame);
if(pbi->LastFrameRecon)
_ogg_free(pbi->LastFrameRecon);
if(pbi->PostProcessBuffer)
_ogg_free(pbi->PostProcessBuffer);
pbi->ThisFrameRecon = 0;
pbi->GoldenFrame = 0;
pbi->LastFrameRecon = 0;
pbi->PostProcessBuffer = 0;
pbi->ThisFrameRecon = 0;
pbi->GoldenFrame = 0;
pbi->LastFrameRecon = 0;
pbi->PostProcessBuffer = 0;
}
void InitFrameInfo(PB_INSTANCE * pbi, unsigned int FrameSize){
/* clear any existing info */
ClearFrameInfo(pbi);
/* allocate frames */
pbi->ThisFrameRecon =
_ogg_malloc(FrameSize*sizeof(*pbi->ThisFrameRecon));
pbi->GoldenFrame =
_ogg_malloc(FrameSize*sizeof(*pbi->GoldenFrame));
pbi->LastFrameRecon =
_ogg_malloc(FrameSize*sizeof(*pbi->LastFrameRecon));
pbi->PostProcessBuffer =
_ogg_malloc(FrameSize*sizeof(*pbi->PostProcessBuffer));
}
void InitFrameDetails(PB_INSTANCE *pbi){
int FrameSize;
/*pbi->PostProcessingLevel = 0;
pbi->PostProcessingLevel = 4;
pbi->PostProcessingLevel = 5;
pbi->PostProcessingLevel = 6;*/
pbi->PostProcessingLevel = 0;
/* Set the frame size etc. */
pbi->YPlaneSize = pbi->info.width *
pbi->info.height;
pbi->UVPlaneSize = pbi->YPlaneSize / 4;
pbi->HFragments = pbi->info.width / HFRAGPIXELS;
pbi->VFragments = pbi->info.height / VFRAGPIXELS;
pbi->UnitFragments = ((pbi->VFragments * pbi->HFragments)*3)/2;
pbi->YPlaneFragments = pbi->HFragments * pbi->VFragments;
pbi->UVPlaneFragments = pbi->YPlaneFragments / 4;
pbi->YStride = (pbi->info.width + STRIDE_EXTRA);
pbi->UVStride = pbi->YStride / 2;
pbi->ReconYPlaneSize = pbi->YStride *
(pbi->info.height + STRIDE_EXTRA);
pbi->ReconUVPlaneSize = pbi->ReconYPlaneSize / 4;
FrameSize = pbi->ReconYPlaneSize + 2 * pbi->ReconUVPlaneSize;
pbi->YDataOffset = 0;
pbi->UDataOffset = pbi->YPlaneSize;
pbi->VDataOffset = pbi->YPlaneSize + pbi->UVPlaneSize;
pbi->ReconYDataOffset =
(pbi->YStride * UMV_BORDER) + UMV_BORDER;
pbi->ReconUDataOffset = pbi->ReconYPlaneSize +
(pbi->UVStride * (UMV_BORDER/2)) + (UMV_BORDER/2);
pbi->ReconVDataOffset = pbi->ReconYPlaneSize + pbi->ReconUVPlaneSize +
(pbi->UVStride * (UMV_BORDER/2)) + (UMV_BORDER/2);
/* Image dimensions in Super-Blocks */
pbi->YSBRows = (pbi->info.height/32) +
( pbi->info.height%32 ? 1 : 0 );
pbi->YSBCols = (pbi->info.width/32) +
( pbi->info.width%32 ? 1 : 0 );
pbi->UVSBRows = ((pbi->info.height/2)/32) +
( (pbi->info.height/2)%32 ? 1 : 0 );
pbi->UVSBCols = ((pbi->info.width/2)/32) +
( (pbi->info.width/2)%32 ? 1 : 0 );
/* Super-Blocks per component */
pbi->YSuperBlocks = pbi->YSBRows * pbi->YSBCols;
pbi->UVSuperBlocks = pbi->UVSBRows * pbi->UVSBCols;
pbi->SuperBlocks = pbi->YSuperBlocks+2*pbi->UVSuperBlocks;
/* Useful externals */
pbi->MacroBlocks = ((pbi->VFragments+1)/2)*((pbi->HFragments+1)/2);
InitFragmentInfo(pbi);
InitFrameInfo(pbi, FrameSize);
InitializeFragCoordinates(pbi);
/* Configure mapping between quad-tree and fragments */
CreateBlockMapping ( pbi->BlockMap, pbi->YSuperBlocks,
pbi->UVSuperBlocks, pbi->HFragments, pbi->VFragments);
/* Re-initialise the pixel index table. */
CalcPixelIndexTable( pbi );
}

1034
Engine/lib/libtheora/lib/enc/hufftables.h Normal file
View file

File diff suppressed because it is too large Load diff

767
Engine/lib/libtheora/lib/enc/mcomp.c Normal file
View file

@ -0,0 +1,767 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: mcomp.c 15153 2008-08-04 18:37:55Z tterribe $
********************************************************************/
#include <stdlib.h>
#include <stdio.h>
#include "codec_internal.h"
/* Initialises motion compentsation. */
void InitMotionCompensation ( CP_INSTANCE *cpi ){
int i;
int SearchSite=0;
int Len;
int LineStepY = (ogg_int32_t)cpi->pb.YStride;
Len=((MAX_MV_EXTENT/2)+1)/2;
/* How many search stages are there. */
cpi->MVSearchSteps = 0;
/* Set up offsets arrays used in half pixel correction. */
cpi->HalfPixelRef2Offset[0] = -LineStepY - 1;
cpi->HalfPixelRef2Offset[1] = -LineStepY;
cpi->HalfPixelRef2Offset[2] = -LineStepY + 1;
cpi->HalfPixelRef2Offset[3] = - 1;
cpi->HalfPixelRef2Offset[4] = 0;
cpi->HalfPixelRef2Offset[5] = 1;
cpi->HalfPixelRef2Offset[6] = LineStepY - 1;
cpi->HalfPixelRef2Offset[7] = LineStepY;
cpi->HalfPixelRef2Offset[8] = LineStepY + 1;
cpi->HalfPixelXOffset[0] = -1;
cpi->HalfPixelXOffset[1] = 0;
cpi->HalfPixelXOffset[2] = 1;
cpi->HalfPixelXOffset[3] = -1;
cpi->HalfPixelXOffset[4] = 0;
cpi->HalfPixelXOffset[5] = 1;
cpi->HalfPixelXOffset[6] = -1;
cpi->HalfPixelXOffset[7] = 0;
cpi->HalfPixelXOffset[8] = 1;
cpi->HalfPixelYOffset[0] = -1;
cpi->HalfPixelYOffset[1] = -1;
cpi->HalfPixelYOffset[2] = -1;
cpi->HalfPixelYOffset[3] = 0;
cpi->HalfPixelYOffset[4] = 0;
cpi->HalfPixelYOffset[5] = 0;
cpi->HalfPixelYOffset[6] = 1;
cpi->HalfPixelYOffset[7] = 1;
cpi->HalfPixelYOffset[8] = 1;
/* Generate offsets for 8 search sites per step. */
while ( Len>0 ) {
/* Another step. */
cpi->MVSearchSteps += 1;
/* Compute offsets for search sites. */
cpi->MVOffsetX[SearchSite] = -Len;
cpi->MVOffsetY[SearchSite++] = -Len;
cpi->MVOffsetX[SearchSite] = 0;
cpi->MVOffsetY[SearchSite++] = -Len;
cpi->MVOffsetX[SearchSite] = Len;
cpi->MVOffsetY[SearchSite++] = -Len;
cpi->MVOffsetX[SearchSite] = -Len;
cpi->MVOffsetY[SearchSite++] = 0;
cpi->MVOffsetX[SearchSite] = Len;
cpi->MVOffsetY[SearchSite++] = 0;
cpi->MVOffsetX[SearchSite] = -Len;
cpi->MVOffsetY[SearchSite++] = Len;
cpi->MVOffsetX[SearchSite] = 0;
cpi->MVOffsetY[SearchSite++] = Len;
cpi->MVOffsetX[SearchSite] = Len;
cpi->MVOffsetY[SearchSite++] = Len;
/* Contract. */
Len /= 2;
}
/* Compute pixel index offsets. */
for ( i=SearchSite-1; i>=0; i-- )
cpi->MVPixelOffsetY[i] = (cpi->MVOffsetY[i]*LineStepY) + cpi->MVOffsetX[i];
}
static ogg_uint32_t GetInterErr (CP_INSTANCE *cpi, unsigned char * NewDataPtr,
unsigned char * RefDataPtr1,
unsigned char * RefDataPtr2,
ogg_uint32_t PixelsPerLine ) {
ogg_int32_t DiffVal;
ogg_int32_t RefOffset = (int)(RefDataPtr1 - RefDataPtr2);
ogg_uint32_t RefPixelsPerLine = PixelsPerLine + STRIDE_EXTRA;
/* Mode of interpolation chosen based upon on the offset of the
second reference pointer */
if ( RefOffset == 0 ) {
DiffVal = dsp_inter8x8_err (cpi->dsp, NewDataPtr, PixelsPerLine,
RefDataPtr1, RefPixelsPerLine);
}else{
DiffVal = dsp_inter8x8_err_xy2 (cpi->dsp, NewDataPtr, PixelsPerLine,
RefDataPtr1,
RefDataPtr2, RefPixelsPerLine);
}
/* Compute and return population variance as mis-match metric. */
return DiffVal;
}
static ogg_uint32_t GetHalfPixelSumAbsDiffs (CP_INSTANCE *cpi,
unsigned char * SrcData,
unsigned char * RefDataPtr1,
unsigned char * RefDataPtr2,
ogg_uint32_t PixelsPerLine,
ogg_uint32_t ErrorSoFar,
ogg_uint32_t BestSoFar ) {
ogg_uint32_t DiffVal = ErrorSoFar;
ogg_int32_t RefOffset = (int)(RefDataPtr1 - RefDataPtr2);
ogg_uint32_t RefPixelsPerLine = PixelsPerLine + STRIDE_EXTRA;
if ( RefOffset == 0 ) {
/* Simple case as for non 0.5 pixel */
DiffVal += dsp_sad8x8 (cpi->dsp, SrcData, PixelsPerLine,
RefDataPtr1, RefPixelsPerLine);
} else {
DiffVal += dsp_sad8x8_xy2_thres (cpi->dsp, SrcData, PixelsPerLine,
RefDataPtr1,
RefDataPtr2, RefPixelsPerLine, BestSoFar);
}
return DiffVal;
}
ogg_uint32_t GetMBIntraError (CP_INSTANCE *cpi, ogg_uint32_t FragIndex,
ogg_uint32_t PixelsPerLine ) {
ogg_uint32_t LocalFragIndex = FragIndex;
ogg_uint32_t IntraError = 0;
dsp_save_fpu (cpi->dsp);
/* Add together the intra errors for those blocks in the macro block
that are coded (Y only) */
if ( cpi->pb.display_fragments[LocalFragIndex] )
IntraError +=
dsp_intra8x8_err (cpi->dsp, &cpi->
ConvDestBuffer[cpi->pb.pixel_index_table[LocalFragIndex]],
PixelsPerLine);
LocalFragIndex++;
if ( cpi->pb.display_fragments[LocalFragIndex] )
IntraError +=
dsp_intra8x8_err (cpi->dsp, &cpi->
ConvDestBuffer[cpi->pb.pixel_index_table[LocalFragIndex]],
PixelsPerLine);
LocalFragIndex = FragIndex + cpi->pb.HFragments;
if ( cpi->pb.display_fragments[LocalFragIndex] )
IntraError +=
dsp_intra8x8_err (cpi->dsp, &cpi->
ConvDestBuffer[cpi->pb.pixel_index_table[LocalFragIndex]],
PixelsPerLine);
LocalFragIndex++;
if ( cpi->pb.display_fragments[LocalFragIndex] )
IntraError +=
dsp_intra8x8_err (cpi->dsp, &cpi->
ConvDestBuffer[cpi->pb.pixel_index_table[LocalFragIndex]],
PixelsPerLine);
dsp_restore_fpu (cpi->dsp);
return IntraError;
}
ogg_uint32_t GetMBInterError (CP_INSTANCE *cpi,
unsigned char * SrcPtr,
unsigned char * RefPtr,
ogg_uint32_t FragIndex,
ogg_int32_t LastXMV,
ogg_int32_t LastYMV,
ogg_uint32_t PixelsPerLine ) {
ogg_uint32_t RefPixelsPerLine = cpi->pb.YStride;
ogg_uint32_t LocalFragIndex = FragIndex;
ogg_int32_t PixelIndex;
ogg_int32_t RefPixelIndex;
ogg_int32_t RefPixelOffset;
ogg_int32_t RefPtr2Offset;
ogg_uint32_t InterError = 0;
unsigned char * SrcPtr1;
unsigned char * RefPtr1;
dsp_save_fpu (cpi->dsp);
/* Work out pixel offset into source buffer. */
PixelIndex = cpi->pb.pixel_index_table[LocalFragIndex];
/* Work out the pixel offset in reference buffer for the default
motion vector */
RefPixelIndex = cpi->pb.recon_pixel_index_table[LocalFragIndex];
RefPixelOffset = ((LastYMV/2) * RefPixelsPerLine) + (LastXMV/2);
/* Work out the second reference pointer offset. */
RefPtr2Offset = 0;
if ( LastXMV % 2 ) {
if ( LastXMV > 0 )
RefPtr2Offset += 1;
else
RefPtr2Offset -= 1;
}
if ( LastYMV % 2 ) {
if ( LastYMV > 0 )
RefPtr2Offset += RefPixelsPerLine;
else
RefPtr2Offset -= RefPixelsPerLine;
}
/* Add together the errors for those blocks in the macro block that
are coded (Y only) */
if ( cpi->pb.display_fragments[LocalFragIndex] ) {
SrcPtr1 = &SrcPtr[PixelIndex];
RefPtr1 = &RefPtr[RefPixelIndex + RefPixelOffset];
InterError += GetInterErr(cpi, SrcPtr1, RefPtr1,
&RefPtr1[RefPtr2Offset], PixelsPerLine );
}
LocalFragIndex++;
if ( cpi->pb.display_fragments[LocalFragIndex] ) {
PixelIndex = cpi->pb.pixel_index_table[LocalFragIndex];
RefPixelIndex = cpi->pb.recon_pixel_index_table[LocalFragIndex];
SrcPtr1 = &SrcPtr[PixelIndex];
RefPtr1 = &RefPtr[RefPixelIndex + RefPixelOffset];
InterError += GetInterErr(cpi, SrcPtr1, RefPtr1,
&RefPtr1[RefPtr2Offset], PixelsPerLine );
}
LocalFragIndex = FragIndex + cpi->pb.HFragments;
if ( cpi->pb.display_fragments[LocalFragIndex] ) {
PixelIndex = cpi->pb.pixel_index_table[LocalFragIndex];
RefPixelIndex = cpi->pb.recon_pixel_index_table[LocalFragIndex];
SrcPtr1 = &SrcPtr[PixelIndex];
RefPtr1 = &RefPtr[RefPixelIndex + RefPixelOffset];
InterError += GetInterErr(cpi, SrcPtr1, RefPtr1,
&RefPtr1[RefPtr2Offset], PixelsPerLine );
}
LocalFragIndex++;
if ( cpi->pb.display_fragments[LocalFragIndex] ) {
PixelIndex = cpi->pb.pixel_index_table[LocalFragIndex];
RefPixelIndex = cpi->pb.recon_pixel_index_table[LocalFragIndex];
SrcPtr1 = &SrcPtr[PixelIndex];
RefPtr1 = &RefPtr[RefPixelIndex + RefPixelOffset];
InterError += GetInterErr(cpi, SrcPtr1, RefPtr1,
&RefPtr1[RefPtr2Offset], PixelsPerLine );
}
dsp_restore_fpu (cpi->dsp);
return InterError;
}
ogg_uint32_t GetMBMVInterError (CP_INSTANCE *cpi,
unsigned char * RefFramePtr,
ogg_uint32_t FragIndex,
ogg_uint32_t PixelsPerLine,
ogg_int32_t *MVPixelOffset,
MOTION_VECTOR *MV ) {
ogg_uint32_t Error = 0;
ogg_uint32_t MinError;
ogg_uint32_t InterMVError = 0;
ogg_int32_t i;
ogg_int32_t x=0, y=0;
ogg_int32_t step;
ogg_int32_t SearchSite=0;
unsigned char *SrcPtr[4] = {NULL,NULL,NULL,NULL};
unsigned char *RefPtr=NULL;
unsigned char *CandidateBlockPtr=NULL;
unsigned char *BestBlockPtr=NULL;
ogg_uint32_t RefRow2Offset = cpi->pb.YStride * 8;
int MBlockDispFrags[4];
/* Half pixel variables */
ogg_int32_t HalfPixelError;
ogg_int32_t BestHalfPixelError;
unsigned char BestHalfOffset;
unsigned char * RefDataPtr1;
unsigned char * RefDataPtr2;
dsp_save_fpu (cpi->dsp);
/* Note which of the four blocks in the macro block are to be
included in the search. */
MBlockDispFrags[0] =
cpi->pb.display_fragments[FragIndex];
MBlockDispFrags[1] =
cpi->pb.display_fragments[FragIndex + 1];
MBlockDispFrags[2] =
cpi->pb.display_fragments[FragIndex + cpi->pb.HFragments];
MBlockDispFrags[3] =
cpi->pb.display_fragments[FragIndex + cpi->pb.HFragments + 1];
/* Set up the source pointers for the four source blocks. */
SrcPtr[0] = &cpi->ConvDestBuffer[cpi->pb.pixel_index_table[FragIndex]];
SrcPtr[1] = SrcPtr[0] + 8;
SrcPtr[2] = SrcPtr[0] + (PixelsPerLine * 8);
SrcPtr[3] = SrcPtr[2] + 8;
/* Set starting reference point for search. */
RefPtr = &RefFramePtr[cpi->pb.recon_pixel_index_table[FragIndex]];
/* Check the 0,0 candidate. */
if ( MBlockDispFrags[0] ) {
Error += dsp_sad8x8 (cpi->dsp, SrcPtr[0], PixelsPerLine, RefPtr,
PixelsPerLine + STRIDE_EXTRA);
}
if ( MBlockDispFrags[1] ) {
Error += dsp_sad8x8 (cpi->dsp, SrcPtr[1], PixelsPerLine, RefPtr + 8,
PixelsPerLine + STRIDE_EXTRA);
}
if ( MBlockDispFrags[2] ) {
Error += dsp_sad8x8 (cpi->dsp, SrcPtr[2], PixelsPerLine, RefPtr + RefRow2Offset,
PixelsPerLine + STRIDE_EXTRA);
}
if ( MBlockDispFrags[3] ) {
Error += dsp_sad8x8 (cpi->dsp, SrcPtr[3], PixelsPerLine, RefPtr + RefRow2Offset + 8,
PixelsPerLine + STRIDE_EXTRA);
}
/* Set starting values to results of 0, 0 vector. */
MinError = Error;
BestBlockPtr = RefPtr;
x = 0;
y = 0;
MV->x = 0;
MV->y = 0;
/* Proceed through N-steps. */
for ( step=0; step<cpi->MVSearchSteps; step++ ) {
/* Search the 8-neighbours at distance pertinent to current step.*/
for ( i=0; i<8; i++ ) {
/* Set pointer to next candidate matching block. */
CandidateBlockPtr = RefPtr + MVPixelOffset[SearchSite];
/* Reset error */
Error = 0;
/* Get the score for the current offset */
if ( MBlockDispFrags[0] ) {
Error += dsp_sad8x8 (cpi->dsp, SrcPtr[0], PixelsPerLine, CandidateBlockPtr,
PixelsPerLine + STRIDE_EXTRA);
}
if ( MBlockDispFrags[1] && (Error < MinError) ) {
Error += dsp_sad8x8_thres (cpi->dsp, SrcPtr[1], PixelsPerLine, CandidateBlockPtr + 8,
PixelsPerLine + STRIDE_EXTRA, MinError);
}
if ( MBlockDispFrags[2] && (Error < MinError) ) {
Error += dsp_sad8x8_thres (cpi->dsp, SrcPtr[2], PixelsPerLine, CandidateBlockPtr + RefRow2Offset,
PixelsPerLine + STRIDE_EXTRA, MinError);
}
if ( MBlockDispFrags[3] && (Error < MinError) ) {
Error += dsp_sad8x8_thres (cpi->dsp, SrcPtr[3], PixelsPerLine, CandidateBlockPtr + RefRow2Offset + 8,
PixelsPerLine + STRIDE_EXTRA, MinError);
}
if ( Error < MinError ) {
/* Remember best match. */
MinError = Error;
BestBlockPtr = CandidateBlockPtr;
/* Where is it. */
x = MV->x + cpi->MVOffsetX[SearchSite];
y = MV->y + cpi->MVOffsetY[SearchSite];
}
/* Move to next search location. */
SearchSite += 1;
}
/* Move to best location this step. */
RefPtr = BestBlockPtr;
MV->x = x;
MV->y = y;
}
/* Factor vectors to 1/2 pixel resoultion. */
MV->x = (MV->x * 2);
MV->y = (MV->y * 2);
/* Now do the half pixel pass */
BestHalfOffset = 4; /* Default to the no offset case. */
BestHalfPixelError = MinError;
/* Get the half pixel error for each half pixel offset */
for ( i=0; i < 9; i++ ) {
HalfPixelError = 0;
if ( MBlockDispFrags[0] ) {
RefDataPtr1 = BestBlockPtr;
RefDataPtr2 = RefDataPtr1 + cpi->HalfPixelRef2Offset[i];
HalfPixelError =
GetHalfPixelSumAbsDiffs(cpi, SrcPtr[0], RefDataPtr1, RefDataPtr2,
PixelsPerLine, HalfPixelError, BestHalfPixelError );
}
if ( MBlockDispFrags[1] && (HalfPixelError < BestHalfPixelError) ) {
RefDataPtr1 = BestBlockPtr + 8;
RefDataPtr2 = RefDataPtr1 + cpi->HalfPixelRef2Offset[i];
HalfPixelError =
GetHalfPixelSumAbsDiffs(cpi, SrcPtr[1], RefDataPtr1, RefDataPtr2,
PixelsPerLine, HalfPixelError, BestHalfPixelError );
}
if ( MBlockDispFrags[2] && (HalfPixelError < BestHalfPixelError) ) {
RefDataPtr1 = BestBlockPtr + RefRow2Offset;
RefDataPtr2 = RefDataPtr1 + cpi->HalfPixelRef2Offset[i];
HalfPixelError =
GetHalfPixelSumAbsDiffs(cpi, SrcPtr[2], RefDataPtr1, RefDataPtr2,
PixelsPerLine, HalfPixelError, BestHalfPixelError );
}
if ( MBlockDispFrags[3] && (HalfPixelError < BestHalfPixelError) ) {
RefDataPtr1 = BestBlockPtr + RefRow2Offset + 8;
RefDataPtr2 = RefDataPtr1 + cpi->HalfPixelRef2Offset[i];
HalfPixelError =
GetHalfPixelSumAbsDiffs(cpi, SrcPtr[3], RefDataPtr1, RefDataPtr2,
PixelsPerLine, HalfPixelError, BestHalfPixelError );
}
if ( HalfPixelError < BestHalfPixelError ) {
BestHalfOffset = (unsigned char)i;
BestHalfPixelError = HalfPixelError;
}
}
/* Half pixel adjust the MV */
MV->x += cpi->HalfPixelXOffset[BestHalfOffset];
MV->y += cpi->HalfPixelYOffset[BestHalfOffset];
/* Get the error score for the chosen 1/2 pixel offset as a variance. */
InterMVError = GetMBInterError( cpi, cpi->ConvDestBuffer, RefFramePtr,
FragIndex, MV->x, MV->y, PixelsPerLine );
dsp_restore_fpu (cpi->dsp);
/* Return score of best matching block. */
return InterMVError;
}
ogg_uint32_t GetMBMVExhaustiveSearch (CP_INSTANCE *cpi,
unsigned char * RefFramePtr,
ogg_uint32_t FragIndex,
ogg_uint32_t PixelsPerLine,
MOTION_VECTOR *MV ) {
ogg_uint32_t Error = 0;
ogg_uint32_t MinError = HUGE_ERROR;
ogg_uint32_t InterMVError = 0;
ogg_int32_t i, j;
ogg_int32_t x=0, y=0;
unsigned char *SrcPtr[4] = {NULL,NULL,NULL,NULL};
unsigned char *RefPtr;
unsigned char *CandidateBlockPtr=NULL;
unsigned char *BestBlockPtr=NULL;
ogg_uint32_t RefRow2Offset = cpi->pb.YStride * 8;
int MBlockDispFrags[4];
/* Half pixel variables */
ogg_int32_t HalfPixelError;
ogg_int32_t BestHalfPixelError;
unsigned char BestHalfOffset;
unsigned char * RefDataPtr1;
unsigned char * RefDataPtr2;
dsp_save_fpu (cpi->dsp);
/* Note which of the four blocks in the macro block are to be
included in the search. */
MBlockDispFrags[0] = cpi->
pb.display_fragments[FragIndex];
MBlockDispFrags[1] = cpi->
pb.display_fragments[FragIndex + 1];
MBlockDispFrags[2] = cpi->
pb.display_fragments[FragIndex + cpi->pb.HFragments];
MBlockDispFrags[3] = cpi->
pb.display_fragments[FragIndex + cpi->pb.HFragments + 1];
/* Set up the source pointers for the four source blocks. */
SrcPtr[0] = &cpi->
ConvDestBuffer[cpi->pb.pixel_index_table[FragIndex]];
SrcPtr[1] = SrcPtr[0] + 8;
SrcPtr[2] = SrcPtr[0] + (PixelsPerLine * 8);
SrcPtr[3] = SrcPtr[2] + 8;
RefPtr = &RefFramePtr[cpi->pb.recon_pixel_index_table[FragIndex]];
RefPtr = RefPtr - ((MAX_MV_EXTENT/2) * cpi->
pb.YStride) - (MAX_MV_EXTENT/2);
/* Search each pixel alligned site */
for ( i = 0; i < (ogg_int32_t)MAX_MV_EXTENT; i ++ ) {
/* Starting position in row */
CandidateBlockPtr = RefPtr;
for ( j = 0; j < (ogg_int32_t)MAX_MV_EXTENT; j++ ) {
/* Reset error */
Error = 0;
/* Summ errors for each block. */
if ( MBlockDispFrags[0] ) {
Error += dsp_sad8x8 (cpi->dsp, SrcPtr[0], PixelsPerLine, CandidateBlockPtr,
PixelsPerLine + STRIDE_EXTRA);
}
if ( MBlockDispFrags[1] ){
Error += dsp_sad8x8 (cpi->dsp, SrcPtr[1], PixelsPerLine, CandidateBlockPtr + 8,
PixelsPerLine + STRIDE_EXTRA);
}
if ( MBlockDispFrags[2] ){
Error += dsp_sad8x8 (cpi->dsp, SrcPtr[2], PixelsPerLine, CandidateBlockPtr + RefRow2Offset,
PixelsPerLine + STRIDE_EXTRA);
}
if ( MBlockDispFrags[3] ){
Error += dsp_sad8x8 (cpi->dsp, SrcPtr[3], PixelsPerLine, CandidateBlockPtr + RefRow2Offset + 8,
PixelsPerLine + STRIDE_EXTRA);
}
/* Was this the best so far */
if ( Error < MinError ) {
MinError = Error;
BestBlockPtr = CandidateBlockPtr;
x = 16 + j - MAX_MV_EXTENT;
y = 16 + i - MAX_MV_EXTENT;
}
/* Move the the next site */
CandidateBlockPtr ++;
}
/* Move on to the next row. */
RefPtr += cpi->pb.YStride;
}
/* Factor vectors to 1/2 pixel resoultion. */
MV->x = (x * 2);
MV->y = (y * 2);
/* Now do the half pixel pass */
BestHalfOffset = 4; /* Default to the no offset case. */
BestHalfPixelError = MinError;
/* Get the half pixel error for each half pixel offset */
for ( i=0; i < 9; i++ ) {
HalfPixelError = 0;
if ( MBlockDispFrags[0] ) {
RefDataPtr1 = BestBlockPtr;
RefDataPtr2 = RefDataPtr1 + cpi->HalfPixelRef2Offset[i];
HalfPixelError =
GetHalfPixelSumAbsDiffs(cpi, SrcPtr[0], RefDataPtr1, RefDataPtr2,
PixelsPerLine, HalfPixelError, BestHalfPixelError );
}
if ( MBlockDispFrags[1] && (HalfPixelError < BestHalfPixelError) ) {
RefDataPtr1 = BestBlockPtr + 8;
RefDataPtr2 = RefDataPtr1 + cpi->HalfPixelRef2Offset[i];
HalfPixelError =
GetHalfPixelSumAbsDiffs(cpi, SrcPtr[1], RefDataPtr1, RefDataPtr2,
PixelsPerLine, HalfPixelError, BestHalfPixelError );
}
if ( MBlockDispFrags[2] && (HalfPixelError < BestHalfPixelError) ) {
RefDataPtr1 = BestBlockPtr + RefRow2Offset;
RefDataPtr2 = RefDataPtr1 + cpi->HalfPixelRef2Offset[i];
HalfPixelError =
GetHalfPixelSumAbsDiffs(cpi, SrcPtr[2], RefDataPtr1, RefDataPtr2,
PixelsPerLine, HalfPixelError, BestHalfPixelError );
}
if ( MBlockDispFrags[3] && (HalfPixelError < BestHalfPixelError) ) {
RefDataPtr1 = BestBlockPtr + RefRow2Offset + 8;
RefDataPtr2 = RefDataPtr1 + cpi->HalfPixelRef2Offset[i];
HalfPixelError =
GetHalfPixelSumAbsDiffs(cpi, SrcPtr[3], RefDataPtr1, RefDataPtr2,
PixelsPerLine, HalfPixelError, BestHalfPixelError );
}
if ( HalfPixelError < BestHalfPixelError ){
BestHalfOffset = (unsigned char)i;
BestHalfPixelError = HalfPixelError;
}
}
/* Half pixel adjust the MV */
MV->x += cpi->HalfPixelXOffset[BestHalfOffset];
MV->y += cpi->HalfPixelYOffset[BestHalfOffset];
/* Get the error score for the chosen 1/2 pixel offset as a variance. */
InterMVError = GetMBInterError( cpi, cpi->ConvDestBuffer, RefFramePtr,
FragIndex, MV->x, MV->y, PixelsPerLine );
dsp_restore_fpu (cpi->dsp);
/* Return score of best matching block. */
return InterMVError;
}
static ogg_uint32_t GetBMVExhaustiveSearch (CP_INSTANCE *cpi,
unsigned char * RefFramePtr,
ogg_uint32_t FragIndex,
ogg_uint32_t PixelsPerLine,
MOTION_VECTOR *MV ) {
ogg_uint32_t Error = 0;
ogg_uint32_t MinError = HUGE_ERROR;
ogg_uint32_t InterMVError = 0;
ogg_int32_t i, j;
ogg_int32_t x=0, y=0;
unsigned char *SrcPtr = NULL;
unsigned char *RefPtr;
unsigned char *CandidateBlockPtr=NULL;
unsigned char *BestBlockPtr=NULL;
/* Half pixel variables */
ogg_int32_t HalfPixelError;
ogg_int32_t BestHalfPixelError;
unsigned char BestHalfOffset;
unsigned char * RefDataPtr2;
/* Set up the source pointer for the block. */
SrcPtr = &cpi->
ConvDestBuffer[cpi->pb.pixel_index_table[FragIndex]];
RefPtr = &RefFramePtr[cpi->pb.recon_pixel_index_table[FragIndex]];
RefPtr = RefPtr - ((MAX_MV_EXTENT/2) *
cpi->pb.YStride) - (MAX_MV_EXTENT/2);
/* Search each pixel alligned site */
for ( i = 0; i < (ogg_int32_t)MAX_MV_EXTENT; i ++ ) {
/* Starting position in row */
CandidateBlockPtr = RefPtr;
for ( j = 0; j < (ogg_int32_t)MAX_MV_EXTENT; j++ ){
/* Get the block error score. */
Error = dsp_sad8x8 (cpi->dsp, SrcPtr, PixelsPerLine, CandidateBlockPtr,
PixelsPerLine + STRIDE_EXTRA);
/* Was this the best so far */
if ( Error < MinError ) {
MinError = Error;
BestBlockPtr = CandidateBlockPtr;
x = 16 + j - MAX_MV_EXTENT;
y = 16 + i - MAX_MV_EXTENT;
}
/* Move the the next site */
CandidateBlockPtr ++;
}
/* Move on to the next row. */
RefPtr += cpi->pb.YStride;
}
/* Factor vectors to 1/2 pixel resoultion. */
MV->x = (x * 2);
MV->y = (y * 2);
/* Now do the half pixel pass */
BestHalfOffset = 4; /* Default to the no offset case. */
BestHalfPixelError = MinError;
/* Get the half pixel error for each half pixel offset */
for ( i=0; i < 9; i++ ) {
RefDataPtr2 = BestBlockPtr + cpi->HalfPixelRef2Offset[i];
HalfPixelError =
GetHalfPixelSumAbsDiffs(cpi, SrcPtr, BestBlockPtr, RefDataPtr2,
PixelsPerLine, 0, BestHalfPixelError );
if ( HalfPixelError < BestHalfPixelError ){
BestHalfOffset = (unsigned char)i;
BestHalfPixelError = HalfPixelError;
}
}
/* Half pixel adjust the MV */
MV->x += cpi->HalfPixelXOffset[BestHalfOffset];
MV->y += cpi->HalfPixelYOffset[BestHalfOffset];
/* Get the variance score at the chosen offset */
RefDataPtr2 = BestBlockPtr + cpi->HalfPixelRef2Offset[BestHalfOffset];
InterMVError =
GetInterErr(cpi, SrcPtr, BestBlockPtr, RefDataPtr2, PixelsPerLine );
/* Return score of best matching block. */
return InterMVError;
}
ogg_uint32_t GetFOURMVExhaustiveSearch (CP_INSTANCE *cpi,
unsigned char * RefFramePtr,
ogg_uint32_t FragIndex,
ogg_uint32_t PixelsPerLine,
MOTION_VECTOR *MV ) {
ogg_uint32_t InterMVError;
dsp_save_fpu (cpi->dsp);
/* For the moment the 4MV mode is only deemed to be valid
if all four Y blocks are to be updated */
/* This may be adapted later. */
if ( cpi->pb.display_fragments[FragIndex] &&
cpi->pb.display_fragments[FragIndex + 1] &&
cpi->pb.display_fragments[FragIndex + cpi->pb.HFragments] &&
cpi->pb.display_fragments[FragIndex + cpi->pb.HFragments + 1] ) {
/* Reset the error score. */
InterMVError = 0;
/* Get the error component from each coded block */
InterMVError +=
GetBMVExhaustiveSearch(cpi, RefFramePtr, FragIndex,
PixelsPerLine, &(MV[0]) );
InterMVError +=
GetBMVExhaustiveSearch(cpi, RefFramePtr, (FragIndex + 1),
PixelsPerLine, &(MV[1]) );
InterMVError +=
GetBMVExhaustiveSearch(cpi, RefFramePtr,
(FragIndex + cpi->pb.HFragments),
PixelsPerLine, &(MV[2]) );
InterMVError +=
GetBMVExhaustiveSearch(cpi, RefFramePtr,
(FragIndex + cpi->pb.HFragments + 1),
PixelsPerLine, &(MV[3]) );
}else{
InterMVError = HUGE_ERROR;
}
dsp_restore_fpu (cpi->dsp);
/* Return score of best matching block. */
return InterMVError;
}

339
Engine/lib/libtheora/lib/enc/misc_common.c Normal file
View file

@ -0,0 +1,339 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: misc_common.c 15323 2008-09-19 19:43:59Z giles $
********************************************************************/
#include <string.h>
#include "codec_internal.h"
#include "block_inline.h"
#define FIXED_Q 150
#define MAX_UP_REG_LOOPS 2
/* Gives the initial bytes per block estimate for each Q value */
static const double BpbTable[Q_TABLE_SIZE] = {
0.42, 0.45, 0.46, 0.49, 0.51, 0.53, 0.56, 0.58,
0.61, 0.64, 0.68, 0.71, 0.74, 0.77, 0.80, 0.84,
0.89, 0.92, 0.98, 1.01, 1.04, 1.13, 1.17, 1.23,
1.28, 1.34, 1.41, 1.45, 1.51, 1.59, 1.69, 1.80,
1.84, 1.94, 2.02, 2.15, 2.23, 2.34, 2.44, 2.50,
2.69, 2.80, 2.87, 3.04, 3.16, 3.29, 3.59, 3.66,
3.86, 3.94, 4.22, 4.50, 4.64, 4.70, 5.24, 5.34,
5.61, 5.87, 6.11, 6.41, 6.71, 6.99, 7.36, 7.69
};
static const double KfBpbTable[Q_TABLE_SIZE] = {
0.74, 0.81, 0.88, 0.94, 1.00, 1.06, 1.14, 1.19,
1.27, 1.34, 1.42, 1.49, 1.54, 1.59, 1.66, 1.73,
1.80, 1.87, 1.97, 2.01, 2.08, 2.21, 2.25, 2.36,
2.39, 2.50, 2.55, 2.65, 2.71, 2.82, 2.95, 3.01,
3.11, 3.19, 3.31, 3.42, 3.58, 3.66, 3.78, 3.89,
4.11, 4.26, 4.36, 4.39, 4.63, 4.76, 4.85, 5.04,
5.26, 5.29, 5.47, 5.64, 5.76, 6.05, 6.35, 6.67,
6.91, 7.17, 7.40, 7.56, 8.02, 8.45, 8.86, 9.38
};
double GetEstimatedBpb( CP_INSTANCE *cpi, ogg_uint32_t TargetQ ){
ogg_uint32_t i;
ogg_int32_t ThreshTableIndex = Q_TABLE_SIZE - 1;
double BytesPerBlock;
/* Search for the Q table index that matches the given Q. */
for ( i = 0; i < Q_TABLE_SIZE; i++ ) {
if ( TargetQ >= cpi->pb.QThreshTable[i] ) {
ThreshTableIndex = i;
break;
}
}
/* Adjust according to Q shift and type of frame */
if ( cpi->pb.FrameType == KEY_FRAME ) {
/* Get primary prediction */
BytesPerBlock = KfBpbTable[ThreshTableIndex];
} else {
/* Get primary prediction */
BytesPerBlock = BpbTable[ThreshTableIndex];
BytesPerBlock = BytesPerBlock * cpi->BpbCorrectionFactor;
}
return BytesPerBlock;
}
static void UpRegulateMB( CP_INSTANCE *cpi, ogg_uint32_t RegulationQ,
ogg_uint32_t SB, ogg_uint32_t MB, int NoCheck ) {
ogg_int32_t FragIndex;
ogg_uint32_t B;
/* Variables used in calculating corresponding row,col and index in
UV planes */
ogg_uint32_t UVRow;
ogg_uint32_t UVColumn;
ogg_uint32_t UVFragOffset;
/* There may be MB's lying out of frame which must be ignored. For
these MB's Top left block will have a negative Fragment Index. */
if ( QuadMapToMBTopLeft(cpi->pb.BlockMap, SB, MB ) >= 0 ) {
/* Up regulate the component blocks Y then UV. */
for ( B=0; B<4; B++ ){
FragIndex = QuadMapToIndex1( cpi->pb.BlockMap, SB, MB, B );
if ( ( !cpi->pb.display_fragments[FragIndex] ) &&
( (NoCheck) || (cpi->FragmentLastQ[FragIndex] > RegulationQ) ) ){
cpi->pb.display_fragments[FragIndex] = 1;
cpi->extra_fragments[FragIndex] = 1;
cpi->FragmentLastQ[FragIndex] = RegulationQ;
cpi->MotionScore++;
}
}
/* Check the two UV blocks */
FragIndex = QuadMapToMBTopLeft(cpi->pb.BlockMap, SB, MB );
UVRow = (FragIndex / (cpi->pb.HFragments * 2));
UVColumn = (FragIndex % cpi->pb.HFragments) / 2;
UVFragOffset = (UVRow * (cpi->pb.HFragments / 2)) + UVColumn;
FragIndex = cpi->pb.YPlaneFragments + UVFragOffset;
if ( ( !cpi->pb.display_fragments[FragIndex] ) &&
( (NoCheck) || (cpi->FragmentLastQ[FragIndex] > RegulationQ) ) ) {
cpi->pb.display_fragments[FragIndex] = 1;
cpi->extra_fragments[FragIndex] = 1;
cpi->FragmentLastQ[FragIndex] = RegulationQ;
cpi->MotionScore++;
}
FragIndex += cpi->pb.UVPlaneFragments;
if ( ( !cpi->pb.display_fragments[FragIndex] ) &&
( (NoCheck) || (cpi->FragmentLastQ[FragIndex] > RegulationQ) ) ) {
cpi->pb.display_fragments[FragIndex] = 1;
cpi->extra_fragments[FragIndex] = 1;
cpi->FragmentLastQ[FragIndex] = RegulationQ;
cpi->MotionScore++;
}
}
}
static void UpRegulateBlocks (CP_INSTANCE *cpi, ogg_uint32_t RegulationQ,
ogg_int32_t RecoveryBlocks,
ogg_uint32_t * LastSB, ogg_uint32_t * LastMB ) {
ogg_uint32_t LoopTimesRound = 0;
ogg_uint32_t MaxSB = cpi->pb.YSBRows *
cpi->pb.YSBCols; /* Tot super blocks in image */
ogg_uint32_t SB, MB; /* Super-Block and macro block indices. */
/* First scan for blocks for which a residue update is outstanding. */
while ( (cpi->MotionScore < RecoveryBlocks) &&
(LoopTimesRound < MAX_UP_REG_LOOPS) ) {
LoopTimesRound++;
for ( SB = (*LastSB); SB < MaxSB; SB++ ) {
/* Check its four Macro-Blocks */
for ( MB=(*LastMB); MB<4; MB++ ) {
/* Mark relevant blocks for update */
UpRegulateMB( cpi, RegulationQ, SB, MB, 0 );
/* Keep track of the last refresh MB. */
(*LastMB) += 1;
if ( (*LastMB) == 4 )
(*LastMB) = 0;
/* Termination clause */
if (cpi->MotionScore >= RecoveryBlocks) {
/* Make sure we don't stall at SB level */
if ( *LastMB == 0 )
SB++;
break;
}
}
/* Termination clause */
if (cpi->MotionScore >= RecoveryBlocks)
break;
}
/* Update super block start index */
if ( SB >= MaxSB){
(*LastSB) = 0;
}else{
(*LastSB) = SB;
}
}
}
void UpRegulateDataStream (CP_INSTANCE *cpi, ogg_uint32_t RegulationQ,
ogg_int32_t RecoveryBlocks ) {
ogg_uint32_t LastPassMBPos = 0;
ogg_uint32_t StdLastMBPos = 0;
ogg_uint32_t MaxSB = cpi->pb.YSBRows *
cpi->pb.YSBCols; /* Tot super blocks in image */
ogg_uint32_t SB=0; /* Super-Block index */
ogg_uint32_t MB; /* Macro-Block index */
/* Decduct the number of blocks in an MB / 2 from the recover block count.
This will compensate for the fact that once we start checking an MB
we test every block in that macro block */
if ( RecoveryBlocks > 3 )
RecoveryBlocks -= 3;
/* Up regulate blocks last coded at higher Q */
UpRegulateBlocks( cpi, RegulationQ, RecoveryBlocks,
&cpi->LastEndSB, &StdLastMBPos );
/* If we have still not used up the minimum number of blocks and are
at the minimum Q then run through a final pass of the data to
insure that each block gets a final refresh. */
if ( (RegulationQ == VERY_BEST_Q) &&
(cpi->MotionScore < RecoveryBlocks) ) {
if ( cpi->FinalPassLastPos < MaxSB ) {
for ( SB = cpi->FinalPassLastPos; SB < MaxSB; SB++ ) {
/* Check its four Macro-Blocks */
for ( MB=LastPassMBPos; MB<4; MB++ ) {
/* Mark relevant blocks for update */
UpRegulateMB( cpi, RegulationQ, SB, MB, 1 );
/* Keep track of the last refresh MB. */
LastPassMBPos += 1;
if ( LastPassMBPos == 4 ) {
LastPassMBPos = 0;
/* Increment SB index */
cpi->FinalPassLastPos += 1;
}
/* Termination clause */
if (cpi->MotionScore >= RecoveryBlocks)
break;
}
/* Termination clause */
if (cpi->MotionScore >= RecoveryBlocks)
break;
}
}
}
}
void RegulateQ( CP_INSTANCE *cpi, ogg_int32_t UpdateScore ) {
double PredUnitScoreBytes;
ogg_uint32_t QIndex = Q_TABLE_SIZE - 1;
ogg_uint32_t i;
if ( UpdateScore > 0 ) {
double TargetUnitScoreBytes = (double)cpi->ThisFrameTargetBytes /
(double)UpdateScore;
double LastBitError = 10000.0; /* Silly high number */
/* Search for the best Q for the target bitrate. */
for ( i = 0; i < Q_TABLE_SIZE; i++ ) {
PredUnitScoreBytes = GetEstimatedBpb( cpi, cpi->pb.QThreshTable[i] );
if ( PredUnitScoreBytes > TargetUnitScoreBytes ) {
if ( (PredUnitScoreBytes - TargetUnitScoreBytes) <= LastBitError ) {
QIndex = i;
} else {
QIndex = i - 1;
}
break;
} else {
LastBitError = TargetUnitScoreBytes - PredUnitScoreBytes;
}
}
}
/* QIndex should now indicate the optimal Q. */
cpi->pb.ThisFrameQualityValue = cpi->pb.QThreshTable[QIndex];
/* Apply range restrictions for key frames. */
if ( cpi->pb.FrameType == KEY_FRAME ) {
if ( cpi->pb.ThisFrameQualityValue > cpi->pb.QThreshTable[20] )
cpi->pb.ThisFrameQualityValue = cpi->pb.QThreshTable[20];
else if ( cpi->pb.ThisFrameQualityValue < cpi->pb.QThreshTable[50] )
cpi->pb.ThisFrameQualityValue = cpi->pb.QThreshTable[50];
}
/* Limit the Q value to the maximum available value */
if (cpi->pb.ThisFrameQualityValue >
cpi->pb.QThreshTable[cpi->Configuration.ActiveMaxQ]) {
cpi->pb.ThisFrameQualityValue =
(ogg_uint32_t)cpi->pb.QThreshTable[cpi->Configuration.ActiveMaxQ];
}
if(cpi->FixedQ) {
if ( cpi->pb.FrameType == KEY_FRAME ) {
cpi->pb.ThisFrameQualityValue = cpi->pb.QThreshTable[43];
cpi->pb.ThisFrameQualityValue = cpi->FixedQ;
} else {
cpi->pb.ThisFrameQualityValue = cpi->FixedQ;
}
}
/* If the quantizer value has changed then re-initialise it */
if ( cpi->pb.ThisFrameQualityValue != cpi->pb.LastFrameQualityValue ) {
/* Initialise quality tables. */
UpdateQC( cpi, cpi->pb.ThisFrameQualityValue );
cpi->pb.LastFrameQualityValue = cpi->pb.ThisFrameQualityValue;
}
}
void CopyBackExtraFrags(CP_INSTANCE *cpi){
ogg_uint32_t i,j;
unsigned char * SrcPtr;
unsigned char * DestPtr;
ogg_uint32_t PlaneLineStep;
ogg_uint32_t PixelIndex;
/* Copy back for Y plane. */
PlaneLineStep = cpi->pb.info.width;
for ( i = 0; i < cpi->pb.YPlaneFragments; i++ ) {
/* We are only interested in updated fragments. */
if ( cpi->extra_fragments[i] ) {
/* Get the start index for the fragment. */
PixelIndex = cpi->pb.pixel_index_table[i];
SrcPtr = &cpi->yuv1ptr[PixelIndex];
DestPtr = &cpi->ConvDestBuffer[PixelIndex];
for ( j = 0; j < VFRAGPIXELS; j++ ) {
memcpy( DestPtr, SrcPtr, HFRAGPIXELS);
SrcPtr += PlaneLineStep;
DestPtr += PlaneLineStep;
}
}
}
/* Now the U and V planes */
PlaneLineStep = cpi->pb.info.width / 2;
for ( i = cpi->pb.YPlaneFragments;
i < (cpi->pb.YPlaneFragments + (2 * cpi->pb.UVPlaneFragments)) ;
i++ ) {
/* We are only interested in updated fragments. */
if ( cpi->extra_fragments[i] ) {
/* Get the start index for the fragment. */
PixelIndex = cpi->pb.pixel_index_table[i];
SrcPtr = &cpi->yuv1ptr[PixelIndex];
DestPtr = &cpi->ConvDestBuffer[PixelIndex];
for ( j = 0; j < VFRAGPIXELS; j++ ) {
memcpy( DestPtr, SrcPtr, HFRAGPIXELS);
SrcPtr += PlaneLineStep;
DestPtr += PlaneLineStep;
}
}
}
}

89
Engine/lib/libtheora/lib/enc/pb.c Normal file
View file

@ -0,0 +1,89 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: pb.c 14372 2008-01-05 23:52:28Z giles $
********************************************************************/
#include <stdlib.h>
#include <string.h>
#include "codec_internal.h"
void ClearTmpBuffers(PB_INSTANCE * pbi){
if(pbi->ReconDataBuffer)
_ogg_free(pbi->ReconDataBuffer);
if(pbi->DequantBuffer)
_ogg_free(pbi->DequantBuffer);
if(pbi->TmpDataBuffer)
_ogg_free(pbi->TmpDataBuffer);
if(pbi->TmpReconBuffer)
_ogg_free(pbi->TmpReconBuffer);
pbi->ReconDataBuffer=0;
pbi->DequantBuffer = 0;
pbi->TmpDataBuffer = 0;
pbi->TmpReconBuffer = 0;
}
void InitTmpBuffers(PB_INSTANCE * pbi){
/* clear any existing info */
ClearTmpBuffers(pbi);
/* Adjust the position of all of our temporary */
pbi->ReconDataBuffer =
_ogg_malloc(64*sizeof(*pbi->ReconDataBuffer));
pbi->DequantBuffer =
_ogg_malloc(64 * sizeof(*pbi->DequantBuffer));
pbi->TmpDataBuffer =
_ogg_malloc(64 * sizeof(*pbi->TmpDataBuffer));
pbi->TmpReconBuffer =
_ogg_malloc(64 * sizeof(*pbi->TmpReconBuffer));
}
void ClearPBInstance(PB_INSTANCE *pbi){
if(pbi){
ClearTmpBuffers(pbi);
if (pbi->opb) {
_ogg_free(pbi->opb);
}
}
}
void InitPBInstance(PB_INSTANCE *pbi){
/* initialize whole structure to 0 */
memset(pbi, 0, sizeof(*pbi));
InitTmpBuffers(pbi);
/* allocate memory for the oggpack_buffer */
pbi->opb = _ogg_malloc(sizeof(oggpack_buffer));
/* variables needing initialization (not being set to 0) */
pbi->ModifierPointer[0] = &pbi->Modifier[0][255];
pbi->ModifierPointer[1] = &pbi->Modifier[1][255];
pbi->ModifierPointer[2] = &pbi->Modifier[2][255];
pbi->ModifierPointer[3] = &pbi->Modifier[3][255];
pbi->DecoderErrorCode = 0;
pbi->KeyFrameType = DCT_KEY_FRAME;
pbi->FramesHaveBeenSkipped = 0;
}

951
Engine/lib/libtheora/lib/enc/pp.c Normal file
View file

@ -0,0 +1,951 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: pp.c 15057 2008-06-22 21:07:32Z xiphmont $
********************************************************************/
#include <stdlib.h>
#include <string.h>
#include "codec_internal.h"
#include "pp.h"
#include "dsp.h"
#define MAX(a, b) ((a>b)?a:b)
#define MIN(a, b) ((a<b)?a:b)
#define PP_QUALITY_THRESH 49
static const ogg_int32_t SharpenModifier[ Q_TABLE_SIZE ] =
{ -12, -11, -10, -10, -9, -9, -9, -9,
-6, -6, -6, -6, -6, -6, -6, -6,
-4, -4, -4, -4, -4, -4, -4, -4,
-2, -2, -2, -2, -2, -2, -2, -2,
-2, -2, -2, -2, -2, -2, -2, -2,
0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0
};
static const ogg_uint32_t DcQuantScaleV1[ Q_TABLE_SIZE ] = {
22, 20, 19, 18, 17, 17, 16, 16,
15, 15, 14, 14, 13, 13, 12, 12,
11, 11, 10, 10, 9, 9, 9, 8,
8, 8, 7, 7, 7, 6, 6, 6,
6, 5, 5, 5, 5, 4, 4, 4,
4, 4, 3, 3, 3, 3, 3, 3,
3, 2, 2, 2, 2, 2, 2, 2,
2, 1, 1, 1, 1, 1, 1, 1
};
static const ogg_uint32_t * const DeringModifierV1=DcQuantScaleV1;
static void PClearFrameInfo(PP_INSTANCE * ppi){
int i;
if(ppi->ScanPixelIndexTable) _ogg_free(ppi->ScanPixelIndexTable);
ppi->ScanPixelIndexTable=0;
if(ppi->ScanDisplayFragments) _ogg_free(ppi->ScanDisplayFragments);
ppi->ScanDisplayFragments=0;
for(i = 0 ; i < MAX_PREV_FRAMES ; i ++)
if(ppi->PrevFragments[i]){
_ogg_free(ppi->PrevFragments[i]);
ppi->PrevFragments[i]=0;
}
if(ppi->FragScores) _ogg_free(ppi->FragScores);
ppi->FragScores=0;
if(ppi->SameGreyDirPixels) _ogg_free(ppi->SameGreyDirPixels);
ppi->SameGreyDirPixels=0;
if(ppi->FragDiffPixels) _ogg_free(ppi->FragDiffPixels);
ppi->FragDiffPixels=0;
if(ppi->BarBlockMap) _ogg_free(ppi->BarBlockMap);
ppi->BarBlockMap=0;
if(ppi->TmpCodedMap) _ogg_free(ppi->TmpCodedMap);
ppi->TmpCodedMap=0;
if(ppi->RowChangedPixels) _ogg_free(ppi->RowChangedPixels);
ppi->RowChangedPixels=0;
if(ppi->PixelScores) _ogg_free(ppi->PixelScores);
ppi->PixelScores=0;
if(ppi->PixelChangedMap) _ogg_free(ppi->PixelChangedMap);
ppi->PixelChangedMap=0;
if(ppi->ChLocals) _ogg_free(ppi->ChLocals);
ppi->ChLocals=0;
if(ppi->yuv_differences) _ogg_free(ppi->yuv_differences);
ppi->yuv_differences=0;
}
void PInitFrameInfo(PP_INSTANCE * ppi){
int i;
PClearFrameInfo(ppi);
ppi->ScanPixelIndexTable =
_ogg_malloc(ppi->ScanFrameFragments*sizeof(*ppi->ScanPixelIndexTable));
ppi->ScanDisplayFragments =
_ogg_malloc(ppi->ScanFrameFragments*sizeof(*ppi->ScanDisplayFragments));
for(i = 0 ; i < MAX_PREV_FRAMES ; i ++)
ppi->PrevFragments[i] =
_ogg_malloc(ppi->ScanFrameFragments*sizeof(*ppi->PrevFragments));
ppi->FragScores =
_ogg_malloc(ppi->ScanFrameFragments*sizeof(*ppi->FragScores));
ppi->SameGreyDirPixels =
_ogg_malloc(ppi->ScanFrameFragments*sizeof(*ppi->SameGreyDirPixels));
ppi->FragDiffPixels =
_ogg_malloc(ppi->ScanFrameFragments*sizeof(*ppi->FragScores));
ppi->BarBlockMap=
_ogg_malloc(3 * ppi->ScanHFragments*sizeof(*ppi->BarBlockMap));
ppi->TmpCodedMap =
_ogg_malloc(ppi->ScanHFragments*sizeof(*ppi->TmpCodedMap));
ppi->RowChangedPixels =
_ogg_malloc(3 * ppi->ScanConfig.VideoFrameHeight*
sizeof(*ppi->RowChangedPixels));
ppi->PixelScores =
_ogg_malloc(ppi->ScanConfig.VideoFrameWidth*
sizeof(*ppi->PixelScores) * PSCORE_CB_ROWS);
ppi->PixelChangedMap =
_ogg_malloc(ppi->ScanConfig.VideoFrameWidth*
sizeof(*ppi->PixelChangedMap) * PMAP_CB_ROWS);
ppi->ChLocals =
_ogg_malloc(ppi->ScanConfig.VideoFrameWidth*
sizeof(*ppi->ChLocals) * CHLOCALS_CB_ROWS);
ppi->yuv_differences =
_ogg_malloc(ppi->ScanConfig.VideoFrameWidth*
sizeof(*ppi->yuv_differences) * YDIFF_CB_ROWS);
}
void ClearPPInstance(PP_INSTANCE *ppi){
PClearFrameInfo(ppi);
}
void InitPPInstance(PP_INSTANCE *ppi, DspFunctions *funcs){
memset(ppi,0,sizeof(*ppi));
memcpy(&ppi->dsp, funcs, sizeof(DspFunctions));
/* Initializations */
ppi->PrevFrameLimit = 3; /* Must not exceed MAX_PREV_FRAMES (Note
that this number includes the current
frame so "1 = no effect") */
/* Scan control variables. */
ppi->HFragPixels = 8;
ppi->VFragPixels = 8;
ppi->SRFGreyThresh = 4;
ppi->SRFColThresh = 5;
ppi->NoiseSupLevel = 3;
ppi->SgcLevelThresh = 3;
ppi->SuvcLevelThresh = 4;
/* Variables controlling S.A.D. breakouts. */
ppi->GrpLowSadThresh = 10;
ppi->GrpHighSadThresh = 64;
ppi->PrimaryBlockThreshold = 5;
ppi->SgcThresh = 16; /* (Default values for 8x8 blocks). */
ppi->UVBlockThreshCorrection = 1.25;
ppi->UVSgcCorrection = 1.5;
ppi->MaxLineSearchLen = MAX_SEARCH_LINE_LEN;
}
static void DeringBlockStrong(unsigned char *SrcPtr,
unsigned char *DstPtr,
ogg_int32_t Pitch,
ogg_uint32_t FragQIndex,
const ogg_uint32_t *QuantScale){
ogg_int16_t UDMod[72];
ogg_int16_t LRMod[72];
unsigned int j,k,l;
const unsigned char * Src;
unsigned int QValue = QuantScale[FragQIndex];
unsigned char p;
unsigned char pl;
unsigned char pr;
unsigned char pu;
unsigned char pd;
int al;
int ar;
int au;
int ad;
int atot;
int B;
int newVal;
const unsigned char *curRow = SrcPtr - 1; /* avoid negative array indexes */
unsigned char *dstRow = DstPtr;
const unsigned char *lastRow = SrcPtr-Pitch;
const unsigned char *nextRow = SrcPtr+Pitch;
unsigned int rowOffset = 0;
unsigned int round = (1<<6);
int High;
int Low;
int TmpMod;
int Sharpen = SharpenModifier[FragQIndex];
High = 3 * QValue;
if(High>32)High=32;
Low = 0;
/* Initialize the Mod Data */
Src = SrcPtr-Pitch;
for(k=0;k<9;k++){
for(j=0;j<8;j++){
TmpMod = 32 + QValue - (abs(Src[j+Pitch]-Src[j]));
if(TmpMod< -64)
TmpMod = Sharpen;
else if(TmpMod<Low)
TmpMod = Low;
else if(TmpMod>High)
TmpMod = High;
UDMod[k*8+j] = (ogg_int16_t)TmpMod;
}
Src +=Pitch;
}
Src = SrcPtr-1;
for(k=0;k<8;k++){
for(j=0;j<9;j++){
TmpMod = 32 + QValue - (abs(Src[j+1]-Src[j]));
if(TmpMod< -64 )
TmpMod = Sharpen;
else if(TmpMod<0)
TmpMod = Low;
else if(TmpMod>High)
TmpMod = High;
LRMod[k*9+j] = (ogg_int16_t)TmpMod;
}
Src+=Pitch;
}
for(k=0;k<8;k++){
/* In the case that this function called with same buffer for
source and destination, To keep the c and the mmx version to have
consistant results, intermediate buffer is used to store the
eight pixel value before writing them to destination
(i.e. Overwriting souce for the speical case) */
for(l=0;l<8;l++){
atot = 128;
B = round;
p = curRow[ rowOffset +l +1];
pl = curRow[ rowOffset +l];
al = LRMod[k*9+l];
atot -= al;
B += al * pl;
pu = lastRow[ rowOffset +l];
au = UDMod[k*8+l];
atot -= au;
B += au * pu;
pd = nextRow[ rowOffset +l];
ad = UDMod[(k+1)*8+l];
atot -= ad;
B += ad * pd;
pr = curRow[ rowOffset +l+2];
ar = LRMod[k*9+l+1];
atot -= ar;
B += ar * pr;
newVal = ( atot * p + B) >> 7;
dstRow[ rowOffset +l]= clamp255( newVal );
}
rowOffset += Pitch;
}
}
static void DeringBlockWeak(unsigned char *SrcPtr,
unsigned char *DstPtr,
ogg_int32_t Pitch,
ogg_uint32_t FragQIndex,
const ogg_uint32_t *QuantScale){
ogg_int16_t UDMod[72];
ogg_int16_t LRMod[72];
unsigned int j,k;
const unsigned char * Src;
unsigned int QValue = QuantScale[FragQIndex];
unsigned char p;
unsigned char pl;
unsigned char pr;
unsigned char pu;
unsigned char pd;
int al;
int ar;
int au;
int ad;
int atot;
int B;
int newVal;
const unsigned char *curRow = SrcPtr-1;
unsigned char *dstRow = DstPtr;
const unsigned char *lastRow = SrcPtr-Pitch;
const unsigned char *nextRow = SrcPtr+Pitch;
unsigned int rowOffset = 0;
unsigned int round = (1<<6);
int High;
int Low;
int TmpMod;
int Sharpen = SharpenModifier[FragQIndex];
High = 3 * QValue;
if(High>24)
High=24;
Low = 0 ;
/* Initialize the Mod Data */
Src=SrcPtr-Pitch;
for(k=0;k<9;k++) {
for(j=0;j<8;j++) {
TmpMod = 32 + QValue - 2*(abs(Src[j+Pitch]-Src[j]));
if(TmpMod< -64)
TmpMod = Sharpen;
else if(TmpMod<Low)
TmpMod = Low;
else if(TmpMod>High)
TmpMod = High;
UDMod[k*8+j] = (ogg_int16_t)TmpMod;
}
Src +=Pitch;
}
Src = SrcPtr-1;
for(k=0;k<8;k++){
for(j=0;j<9;j++){
TmpMod = 32 + QValue - 2*(abs(Src[j+1]-Src[j]));
if(TmpMod< -64 )
TmpMod = Sharpen;
else if(TmpMod<Low)
TmpMod = Low;
else if(TmpMod>High)
TmpMod = High;
LRMod[k*9+j] = (ogg_int16_t)TmpMod;
}
Src+=Pitch;
}
for(k=0;k<8;k++) {
for(j=0;j<8;j++){
atot = 128;
B = round;
p = curRow[ rowOffset +j+1];
pl = curRow[ rowOffset +j];
al = LRMod[k*9+j];
atot -= al;
B += al * pl;
pu = lastRow[ rowOffset +j];
au = UDMod[k*8+j];
atot -= au;
B += au * pu;
pd = nextRow[ rowOffset +j];
ad = UDMod[(k+1)*8+j];
atot -= ad;
B += ad * pd;
pr = curRow[ rowOffset +j+2];
ar = LRMod[k*9+j+1];
atot -= ar;
B += ar * pr;
newVal = ( atot * p + B) >> 7;
dstRow[ rowOffset +j] = clamp255( newVal );
}
rowOffset += Pitch;
}
}
static void DeringFrame(PB_INSTANCE *pbi,
unsigned char *Src, unsigned char *Dst){
ogg_uint32_t col,row;
unsigned char *SrcPtr;
unsigned char *DestPtr;
ogg_uint32_t BlocksAcross,BlocksDown;
const ogg_uint32_t *QuantScale;
ogg_uint32_t Block;
ogg_uint32_t LineLength;
ogg_int32_t Thresh1,Thresh2,Thresh3,Thresh4;
Thresh1 = 384;
Thresh2 = 4 * Thresh1;
Thresh3 = 5 * Thresh2/4;
Thresh4 = 5 * Thresh2/2;
QuantScale = DeringModifierV1;
BlocksAcross = pbi->HFragments;
BlocksDown = pbi->VFragments;
SrcPtr = Src + pbi->ReconYDataOffset;
DestPtr = Dst + pbi->ReconYDataOffset;
LineLength = pbi->YStride;
Block = 0;
for ( row = 0 ; row < BlocksDown; row ++){
for (col = 0; col < BlocksAcross; col ++){
ogg_uint32_t Quality = pbi->FragQIndex[Block];
ogg_int32_t Variance = pbi->FragmentVariances[Block];
if( pbi->PostProcessingLevel >5 && Variance > Thresh3 ){
DeringBlockStrong(SrcPtr + 8 * col, DestPtr + 8 * col,
LineLength,Quality,QuantScale);
if( (col > 0 &&
pbi->FragmentVariances[Block-1] > Thresh4 ) ||
(col + 1 < BlocksAcross &&
pbi->FragmentVariances[Block+1] > Thresh4 ) ||
(row + 1 < BlocksDown &&
pbi->FragmentVariances[Block+BlocksAcross] > Thresh4) ||
(row > 0 &&
pbi->FragmentVariances[Block-BlocksAcross] > Thresh4) ){
DeringBlockStrong(SrcPtr + 8 * col, DestPtr + 8 * col,
LineLength,Quality,QuantScale);
DeringBlockStrong(SrcPtr + 8 * col, DestPtr + 8 * col,
LineLength,Quality,QuantScale);
}
} else if(Variance > Thresh2 ) {
DeringBlockStrong(SrcPtr + 8 * col, DestPtr + 8 * col,
LineLength,Quality,QuantScale);
} else if(Variance > Thresh1 ) {
DeringBlockWeak(SrcPtr + 8 * col, DestPtr + 8 * col,
LineLength,Quality,QuantScale);
} else {
dsp_copy8x8(pbi->dsp, SrcPtr + 8 * col, DestPtr + 8 * col, LineLength);
}
++Block;
}
SrcPtr += 8 * LineLength;
DestPtr += 8 * LineLength;
}
/* Then U */
BlocksAcross /= 2;
BlocksDown /= 2;
LineLength /= 2;
SrcPtr = Src + pbi->ReconUDataOffset;
DestPtr = Dst + pbi->ReconUDataOffset;
for ( row = 0 ; row < BlocksDown; row ++) {
for (col = 0; col < BlocksAcross; col ++) {
ogg_uint32_t Quality = pbi->FragQIndex[Block];
ogg_int32_t Variance = pbi->FragmentVariances[Block];
if( pbi->PostProcessingLevel >5 && Variance > Thresh4 ) {
DeringBlockStrong(SrcPtr + 8 * col, DestPtr + 8 * col,
LineLength,Quality,QuantScale);
DeringBlockStrong(SrcPtr + 8 * col, DestPtr + 8 * col,
LineLength,Quality,QuantScale);
DeringBlockStrong(SrcPtr + 8 * col, DestPtr + 8 * col,
LineLength,Quality,QuantScale);
}else if(Variance > Thresh2 ){
DeringBlockStrong(SrcPtr + 8 * col, DestPtr + 8 * col,
LineLength,Quality,QuantScale);
}else if(Variance > Thresh1 ){
DeringBlockWeak(SrcPtr + 8 * col, DestPtr + 8 * col,
LineLength,Quality,QuantScale);
}else{
dsp_copy8x8(pbi->dsp, SrcPtr + 8 * col, DestPtr + 8 * col, LineLength);
}
++Block;
}
SrcPtr += 8 * LineLength;
DestPtr += 8 * LineLength;
}
/* Then V */
SrcPtr = Src + pbi->ReconVDataOffset;
DestPtr = Dst + pbi->ReconVDataOffset;
for ( row = 0 ; row < BlocksDown; row ++){
for (col = 0; col < BlocksAcross; col ++){
ogg_uint32_t Quality = pbi->FragQIndex[Block];
ogg_int32_t Variance = pbi->FragmentVariances[Block];
if( pbi->PostProcessingLevel >5 && Variance > Thresh4 ) {
DeringBlockStrong(SrcPtr + 8 * col, DestPtr + 8 * col,
LineLength,Quality,QuantScale);
DeringBlockStrong(SrcPtr + 8 * col, DestPtr + 8 * col,
LineLength,Quality,QuantScale);
DeringBlockStrong(SrcPtr + 8 * col, DestPtr + 8 * col,
LineLength,Quality,QuantScale);
}else if(Variance > Thresh2 ){
DeringBlockStrong(SrcPtr + 8 * col, DestPtr + 8 * col,
LineLength,Quality,QuantScale);
}else if(Variance > Thresh1 ){
DeringBlockWeak(SrcPtr + 8 * col, DestPtr + 8 * col,
LineLength,Quality,QuantScale);
}else{
dsp_copy8x8(pbi->dsp, SrcPtr + 8 * col, DestPtr + 8 * col, LineLength);
}
++Block;
}
SrcPtr += 8 * LineLength;
DestPtr += 8 * LineLength;
}
}
void UpdateFragQIndex(PB_INSTANCE *pbi){
ogg_uint32_t ThisFrameQIndex;
ogg_uint32_t i;
/* Check this frame quality index */
ThisFrameQIndex = pbi->FrameQIndex;
/* It is not a key frame, so only reset those are coded */
for( i = 0; i < pbi->UnitFragments; i++ )
if( pbi->display_fragments[i])
pbi->FragQIndex[i] = ThisFrameQIndex;
}
static void DeblockLoopFilteredBand(PB_INSTANCE *pbi,
unsigned char *SrcPtr,
unsigned char *DesPtr,
ogg_uint32_t PlaneLineStep,
ogg_uint32_t FragsAcross,
ogg_uint32_t StartFrag,
const ogg_uint32_t *QuantScale){
ogg_uint32_t j,k;
ogg_uint32_t CurrentFrag=StartFrag;
ogg_int32_t QStep;
ogg_int32_t FLimit;
unsigned char *Src, *Des;
ogg_int32_t x[10];
ogg_int32_t Sum1, Sum2;
while(CurrentFrag < StartFrag + FragsAcross){
Src=SrcPtr+8*(CurrentFrag-StartFrag)-PlaneLineStep*5;
Des=DesPtr+8*(CurrentFrag-StartFrag)-PlaneLineStep*4;
QStep = QuantScale[pbi->FragQIndex[CurrentFrag+FragsAcross]];
FLimit = ( QStep * 3 ) >> 2;
for( j=0; j<8 ; j++){
x[0] = Src[0];
x[1] = Src[PlaneLineStep];
x[2] = Src[PlaneLineStep*2];
x[3] = Src[PlaneLineStep*3];
x[4] = Src[PlaneLineStep*4];
x[5] = Src[PlaneLineStep*5];
x[6] = Src[PlaneLineStep*6];
x[7] = Src[PlaneLineStep*7];
x[8] = Src[PlaneLineStep*8];
x[9] = Src[PlaneLineStep*9];
Sum1=Sum2=0;
for(k=1;k<=4;k++){
Sum1 += abs(x[k]-x[k-1]);
Sum2 += abs(x[k+4]-x[k+5]);
}
pbi->FragmentVariances[CurrentFrag] +=((Sum1>255)?255:Sum1);
pbi->FragmentVariances[CurrentFrag + FragsAcross] += ((Sum2>255)?255:Sum2);
if( Sum1 < FLimit &&
Sum2 < FLimit &&
(x[5] - x[4]) < QStep &&
(x[4] - x[5]) < QStep ){
/* low pass filtering (LPF7: 1 1 1 2 1 1 1) */
Des[0 ] = (x[0] + x[0] +x[0] + x[1] * 2 +
x[2] + x[3] +x[4] + 4) >> 3;
Des[PlaneLineStep ] = (x[0] + x[0] +x[1] + x[2] * 2 +
x[3] + x[4] +x[5] + 4) >> 3;
Des[PlaneLineStep*2] = (x[0] + x[1] +x[2] + x[3] * 2 +
x[4] + x[5] +x[6] + 4) >> 3;
Des[PlaneLineStep*3] = (x[1] + x[2] +x[3] + x[4] * 2 +
x[5] + x[6] +x[7] + 4) >> 3;
Des[PlaneLineStep*4] = (x[2] + x[3] +x[4] + x[5] * 2 +
x[6] + x[7] +x[8] + 4) >> 3;
Des[PlaneLineStep*5] = (x[3] + x[4] +x[5] + x[6] * 2 +
x[7] + x[8] +x[9] + 4) >> 3;
Des[PlaneLineStep*6] = (x[4] + x[5] +x[6] + x[7] * 2 +
x[8] + x[9] +x[9] + 4) >> 3;
Des[PlaneLineStep*7] = (x[5] + x[6] +x[7] + x[8] * 2 +
x[9] + x[9] +x[9] + 4) >> 3;
}else {
/* copy the pixels to destination */
Des[0 ]= (unsigned char)x[1];
Des[PlaneLineStep ]= (unsigned char)x[2];
Des[PlaneLineStep*2]= (unsigned char)x[3];
Des[PlaneLineStep*3]= (unsigned char)x[4];
Des[PlaneLineStep*4]= (unsigned char)x[5];
Des[PlaneLineStep*5]= (unsigned char)x[6];
Des[PlaneLineStep*6]= (unsigned char)x[7];
Des[PlaneLineStep*7]= (unsigned char)x[8];
}
Src ++;
Des ++;
}
/* done with filtering the horizontal edge, now let's do the
vertical one */
/* skip the first one */
if(CurrentFrag==StartFrag)
CurrentFrag++;
else{
Des=DesPtr-8*PlaneLineStep+8*(CurrentFrag-StartFrag);
Src=Des-5;
Des-=4;
QStep = QuantScale[pbi->FragQIndex[CurrentFrag]];
FLimit = ( QStep * 3 ) >> 2;
for( j=0; j<8 ; j++){
x[0] = Src[0];
x[1] = Src[1];
x[2] = Src[2];
x[3] = Src[3];
x[4] = Src[4];
x[5] = Src[5];
x[6] = Src[6];
x[7] = Src[7];
x[8] = Src[8];
x[9] = Src[9];
Sum1=Sum2=0;
for(k=1;k<=4;k++){
Sum1 += abs(x[k]-x[k-1]);
Sum2 += abs(x[k+4]-x[k+5]);
}
pbi->FragmentVariances[CurrentFrag-1] += ((Sum1>255)?255:Sum1);
pbi->FragmentVariances[CurrentFrag] += ((Sum2>255)?255:Sum2);
if( Sum1 < FLimit &&
Sum2 < FLimit &&
(x[5] - x[4]) < QStep &&
(x[4] - x[5]) < QStep ){
/* low pass filtering (LPF7: 1 1 1 2 1 1 1) */
Des[0] = (x[0] + x[0] +x[0] + x[1] * 2 + x[2] + x[3] +x[4] + 4) >> 3;
Des[1] = (x[0] + x[0] +x[1] + x[2] * 2 + x[3] + x[4] +x[5] + 4) >> 3;
Des[2] = (x[0] + x[1] +x[2] + x[3] * 2 + x[4] + x[5] +x[6] + 4) >> 3;
Des[3] = (x[1] + x[2] +x[3] + x[4] * 2 + x[5] + x[6] +x[7] + 4) >> 3;
Des[4] = (x[2] + x[3] +x[4] + x[5] * 2 + x[6] + x[7] +x[8] + 4) >> 3;
Des[5] = (x[3] + x[4] +x[5] + x[6] * 2 + x[7] + x[8] +x[9] + 4) >> 3;
Des[6] = (x[4] + x[5] +x[6] + x[7] * 2 + x[8] + x[9] +x[9] + 4) >> 3;
Des[7] = (x[5] + x[6] +x[7] + x[8] * 2 + x[9] + x[9] +x[9] + 4) >> 3;
}
Src += PlaneLineStep;
Des += PlaneLineStep;
}
CurrentFrag ++;
}
}
}
static void DeblockVerticalEdgesInLoopFilteredBand(PB_INSTANCE *pbi,
unsigned char *SrcPtr,
unsigned char *DesPtr,
ogg_uint32_t PlaneLineStep,
ogg_uint32_t FragsAcross,
ogg_uint32_t StartFrag,
const ogg_uint32_t *QuantScale){
ogg_uint32_t j,k;
ogg_uint32_t CurrentFrag=StartFrag;
ogg_int32_t QStep;
ogg_int32_t FLimit;
unsigned char *Src, *Des;
ogg_int32_t x[10];
ogg_int32_t Sum1, Sum2;
while(CurrentFrag < StartFrag + FragsAcross-1) {
Src=SrcPtr+8*(CurrentFrag-StartFrag+1)-5;
Des=DesPtr+8*(CurrentFrag-StartFrag+1)-4;
QStep = QuantScale[pbi->FragQIndex[CurrentFrag+1]];
FLimit = ( QStep * 3)>>2 ;
for( j=0; j<8 ; j++){
x[0] = Src[0];
x[1] = Src[1];
x[2] = Src[2];
x[3] = Src[3];
x[4] = Src[4];
x[5] = Src[5];
x[6] = Src[6];
x[7] = Src[7];
x[8] = Src[8];
x[9] = Src[9];
Sum1=Sum2=0;
for(k=1;k<=4;k++){
Sum1 += abs(x[k]-x[k-1]);
Sum2 += abs(x[k+4]-x[k+5]);
}
pbi->FragmentVariances[CurrentFrag] += ((Sum1>255)?255:Sum1);
pbi->FragmentVariances[CurrentFrag+1] += ((Sum2>255)?255:Sum2);
if( Sum1 < FLimit &&
Sum2 < FLimit &&
(x[5] - x[4]) < QStep &&
(x[4] - x[5]) < QStep ){
/* low pass filtering (LPF7: 1 1 1 2 1 1 1) */
Des[0] = (x[0] + x[0] +x[0] + x[1] * 2 + x[2] + x[3] +x[4] + 4) >> 3;
Des[1] = (x[0] + x[0] +x[1] + x[2] * 2 + x[3] + x[4] +x[5] + 4) >> 3;
Des[2] = (x[0] + x[1] +x[2] + x[3] * 2 + x[4] + x[5] +x[6] + 4) >> 3;
Des[3] = (x[1] + x[2] +x[3] + x[4] * 2 + x[5] + x[6] +x[7] + 4) >> 3;
Des[4] = (x[2] + x[3] +x[4] + x[5] * 2 + x[6] + x[7] +x[8] + 4) >> 3;
Des[5] = (x[3] + x[4] +x[5] + x[6] * 2 + x[7] + x[8] +x[9] + 4) >> 3;
Des[6] = (x[4] + x[5] +x[6] + x[7] * 2 + x[8] + x[9] +x[9] + 4) >> 3;
Des[7] = (x[5] + x[6] +x[7] + x[8] * 2 + x[9] + x[9] +x[9] + 4) >> 3;
}
Src +=PlaneLineStep;
Des +=PlaneLineStep;
}
CurrentFrag ++;
}
}
static void DeblockPlane(PB_INSTANCE *pbi,
unsigned char *SourceBuffer,
unsigned char *DestinationBuffer,
ogg_uint32_t Channel ){
ogg_uint32_t i,k;
ogg_uint32_t PlaneLineStep=0;
ogg_uint32_t StartFrag =0;
ogg_uint32_t PixelIndex=0;
unsigned char * SrcPtr=0, * DesPtr=0;
ogg_uint32_t FragsAcross=0;
ogg_uint32_t FragsDown=0;
const ogg_uint32_t *QuantScale=0;
switch( Channel ){
case 0:
/* Get the parameters */
PlaneLineStep = pbi->YStride;
FragsAcross = pbi->HFragments;
FragsDown = pbi->VFragments;
StartFrag = 0;
PixelIndex = pbi->ReconYDataOffset;
SrcPtr = & SourceBuffer[PixelIndex];
DesPtr = & DestinationBuffer[PixelIndex];
break;
case 1:
/* Get the parameters */
PlaneLineStep = pbi->UVStride;
FragsAcross = pbi->HFragments / 2;
FragsDown = pbi->VFragments / 2;
StartFrag = pbi->YPlaneFragments;
PixelIndex = pbi->ReconUDataOffset;
SrcPtr = & SourceBuffer[PixelIndex];
DesPtr = & DestinationBuffer[PixelIndex];
break;
default:
/* Get the parameters */
PlaneLineStep = pbi->UVStride;
FragsAcross = pbi->HFragments / 2;
FragsDown = pbi->VFragments / 2;
StartFrag = pbi->YPlaneFragments + pbi->UVPlaneFragments;
PixelIndex = pbi->ReconVDataOffset;
SrcPtr = & SourceBuffer[PixelIndex];
DesPtr = & DestinationBuffer[PixelIndex];
break;
}
QuantScale = DcQuantScaleV1;
for(i=0;i<4;i++)
memcpy(DesPtr+i*PlaneLineStep, SrcPtr+i*PlaneLineStep, PlaneLineStep);
k = 1;
while( k < FragsDown ){
SrcPtr += 8*PlaneLineStep;
DesPtr += 8*PlaneLineStep;
/* Filter both the horizontal and vertical block edges inside the band */
DeblockLoopFilteredBand(pbi, SrcPtr, DesPtr, PlaneLineStep,
FragsAcross, StartFrag, QuantScale);
/* Move Pointers */
StartFrag += FragsAcross;
k ++;
}
/* The Last band */
for(i=0;i<4;i++)
memcpy(DesPtr+(i+4)*PlaneLineStep,
SrcPtr+(i+4)*PlaneLineStep,
PlaneLineStep);
DeblockVerticalEdgesInLoopFilteredBand(pbi,SrcPtr,DesPtr,PlaneLineStep,
FragsAcross,StartFrag,QuantScale);
}
static void DeblockFrame(PB_INSTANCE *pbi, unsigned char *SourceBuffer,
unsigned char *DestinationBuffer){
memset(pbi->FragmentVariances, 0 , sizeof(ogg_int32_t) * pbi->UnitFragments);
UpdateFragQIndex(pbi);
/* Y */
DeblockPlane( pbi, SourceBuffer, DestinationBuffer, 0);
/* U */
DeblockPlane( pbi, SourceBuffer, DestinationBuffer, 1);
/* V */
DeblockPlane( pbi, SourceBuffer, DestinationBuffer, 2);
}
void PostProcess(PB_INSTANCE *pbi){
switch (pbi->PostProcessingLevel){
case 8:
/* on a slow machine, use a simpler and faster deblocking filter */
DeblockFrame(pbi, pbi->LastFrameRecon,pbi->PostProcessBuffer);
break;
case 6:
DeblockFrame(pbi, pbi->LastFrameRecon,pbi->PostProcessBuffer);
UpdateUMVBorder(pbi, pbi->PostProcessBuffer );
DeringFrame(pbi, pbi->PostProcessBuffer, pbi->PostProcessBuffer);
break;
case 5:
DeblockFrame(pbi, pbi->LastFrameRecon,pbi->PostProcessBuffer);
UpdateUMVBorder(pbi, pbi->PostProcessBuffer );
DeringFrame(pbi, pbi->PostProcessBuffer, pbi->PostProcessBuffer);
break;
case 4:
DeblockFrame(pbi, pbi->LastFrameRecon, pbi->PostProcessBuffer);
break;
case 1:
UpdateFragQIndex(pbi);
break;
case 0:
break;
default:
DeblockFrame(pbi, pbi->LastFrameRecon, pbi->PostProcessBuffer);
UpdateUMVBorder(pbi, pbi->PostProcessBuffer );
DeringFrame(pbi, pbi->PostProcessBuffer, pbi->PostProcessBuffer);
break;
}
}

48
Engine/lib/libtheora/lib/enc/pp.h Normal file
View file

@ -0,0 +1,48 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: pp.h 13884 2007-09-22 08:38:10Z giles $
********************************************************************/
/* Constants. */
#define INTERNAL_BLOCK_HEIGHT 8
#define INTERNAL_BLOCK_WIDTH 8
/* NEW Line search values. */
#define UP 0
#define DOWN 1
#define LEFT 2
#define RIGHT 3
#define FIRST_ROW 0
#define NOT_EDGE_ROW 1
#define LAST_ROW 2
#define YDIFF_CB_ROWS (INTERNAL_BLOCK_HEIGHT * 3)
#define CHLOCALS_CB_ROWS (INTERNAL_BLOCK_HEIGHT * 3)
#define PMAP_CB_ROWS (INTERNAL_BLOCK_HEIGHT * 3)
#define PSCORE_CB_ROWS (INTERNAL_BLOCK_HEIGHT * 4)
/* Status values in block coding map */
#define CANDIDATE_BLOCK_LOW -2
#define CANDIDATE_BLOCK -1
#define BLOCK_NOT_CODED 0
#define BLOCK_CODED_BAR 3
#define BLOCK_CODED_SGC 4
#define BLOCK_CODED_LOW 4
#define BLOCK_CODED 5
#define MAX_PREV_FRAMES 16
#define MAX_SEARCH_LINE_LEN 7

43
Engine/lib/libtheora/lib/enc/quant_lookup.h Normal file
View file

@ -0,0 +1,43 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: quant_lookup.h 13884 2007-09-22 08:38:10Z giles $
********************************************************************/
#include "codec_internal.h"
#define MIN16 ((1<<16)-1)
#define SHIFT16 (1<<16)
#define MIN_LEGAL_QUANT_ENTRY 8
#define MIN_DEQUANT_VAL 2
#define IDCT_SCALE_FACTOR 2 /* Shift left bits to improve IDCT precision */
#define OLD_SCHEME 1
/******************************
* lookup table for DCT coefficient zig-zag ordering
* ****************************/
static const ogg_uint32_t dezigzag_index[64] = {
0, 1, 8, 16, 9, 2, 3, 10,
17, 24, 32, 25, 18, 11, 4, 5,
12, 19, 26, 33, 40, 48, 41, 34,
27, 20, 13, 6, 7, 14, 21, 28,
35, 42, 49, 56, 57, 50, 43, 36,
29, 22, 15, 23, 30, 37, 44, 51,
58, 59, 52, 45, 38, 31, 39, 46,
53, 60, 61, 54, 47, 55, 62, 63
};

110
Engine/lib/libtheora/lib/enc/reconstruct.c Normal file
View file

@ -0,0 +1,110 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: reconstruct.c 13884 2007-09-22 08:38:10Z giles $
********************************************************************/
#include "codec_internal.h"
static void copy8x8__c (unsigned char *src,
unsigned char *dest,
unsigned int stride)
{
int j;
for ( j = 0; j < 8; j++ ){
((ogg_uint32_t*)dest)[0] = ((ogg_uint32_t*)src)[0];
((ogg_uint32_t*)dest)[1] = ((ogg_uint32_t*)src)[1];
src+=stride;
dest+=stride;
}
}
static void recon_intra8x8__c (unsigned char *ReconPtr, ogg_int16_t *ChangePtr,
ogg_uint32_t LineStep)
{
ogg_uint32_t i;
for (i = 8; i; i--){
/* Convert the data back to 8 bit unsigned */
/* Saturate the output to unsigend 8 bit values */
ReconPtr[0] = clamp255( ChangePtr[0] + 128 );
ReconPtr[1] = clamp255( ChangePtr[1] + 128 );
ReconPtr[2] = clamp255( ChangePtr[2] + 128 );
ReconPtr[3] = clamp255( ChangePtr[3] + 128 );
ReconPtr[4] = clamp255( ChangePtr[4] + 128 );
ReconPtr[5] = clamp255( ChangePtr[5] + 128 );
ReconPtr[6] = clamp255( ChangePtr[6] + 128 );
ReconPtr[7] = clamp255( ChangePtr[7] + 128 );
ReconPtr += LineStep;
ChangePtr += 8;
}
}
static void recon_inter8x8__c (unsigned char *ReconPtr, unsigned char *RefPtr,
ogg_int16_t *ChangePtr, ogg_uint32_t LineStep)
{
ogg_uint32_t i;
for (i = 8; i; i--){
ReconPtr[0] = clamp255(RefPtr[0] + ChangePtr[0]);
ReconPtr[1] = clamp255(RefPtr[1] + ChangePtr[1]);
ReconPtr[2] = clamp255(RefPtr[2] + ChangePtr[2]);
ReconPtr[3] = clamp255(RefPtr[3] + ChangePtr[3]);
ReconPtr[4] = clamp255(RefPtr[4] + ChangePtr[4]);
ReconPtr[5] = clamp255(RefPtr[5] + ChangePtr[5]);
ReconPtr[6] = clamp255(RefPtr[6] + ChangePtr[6]);
ReconPtr[7] = clamp255(RefPtr[7] + ChangePtr[7]);
ChangePtr += 8;
ReconPtr += LineStep;
RefPtr += LineStep;
}
}
static void recon_inter8x8_half__c (unsigned char *ReconPtr, unsigned char *RefPtr1,
unsigned char *RefPtr2, ogg_int16_t *ChangePtr,
ogg_uint32_t LineStep)
{
ogg_uint32_t i;
for (i = 8; i; i--){
ReconPtr[0] = clamp255((((int)RefPtr1[0] + (int)RefPtr2[0]) >> 1) + ChangePtr[0] );
ReconPtr[1] = clamp255((((int)RefPtr1[1] + (int)RefPtr2[1]) >> 1) + ChangePtr[1] );
ReconPtr[2] = clamp255((((int)RefPtr1[2] + (int)RefPtr2[2]) >> 1) + ChangePtr[2] );
ReconPtr[3] = clamp255((((int)RefPtr1[3] + (int)RefPtr2[3]) >> 1) + ChangePtr[3] );
ReconPtr[4] = clamp255((((int)RefPtr1[4] + (int)RefPtr2[4]) >> 1) + ChangePtr[4] );
ReconPtr[5] = clamp255((((int)RefPtr1[5] + (int)RefPtr2[5]) >> 1) + ChangePtr[5] );
ReconPtr[6] = clamp255((((int)RefPtr1[6] + (int)RefPtr2[6]) >> 1) + ChangePtr[6] );
ReconPtr[7] = clamp255((((int)RefPtr1[7] + (int)RefPtr2[7]) >> 1) + ChangePtr[7] );
ChangePtr += 8;
ReconPtr += LineStep;
RefPtr1 += LineStep;
RefPtr2 += LineStep;
}
}
void dsp_recon_init (DspFunctions *funcs, ogg_uint32_t cpu_flags)
{
funcs->copy8x8 = copy8x8__c;
funcs->recon_intra8x8 = recon_intra8x8__c;
funcs->recon_inter8x8 = recon_inter8x8__c;
funcs->recon_inter8x8_half = recon_inter8x8_half__c;
#if defined(USE_ASM)
if (cpu_flags & OC_CPU_X86_MMX) {
dsp_mmx_recon_init(funcs);
}
#endif
}

2301
Engine/lib/libtheora/lib/enc/scan.c Normal file
View file

File diff suppressed because it is too large Load diff

40
Engine/lib/libtheora/lib/enc/toplevel_lookup.h Normal file
View file

@ -0,0 +1,40 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: toplevel_lookup.h 13884 2007-09-22 08:38:10Z giles $
********************************************************************/
#include "codec_internal.h"
const ogg_uint32_t PriorKeyFrameWeight[KEY_FRAME_CONTEXT] = { 1,2,3,4,5 };
/* Data structures controlling addition of residue blocks */
const ogg_uint32_t ResidueErrorThresh[Q_TABLE_SIZE] = {
750, 700, 650, 600, 590, 580, 570, 560,
550, 540, 530, 520, 510, 500, 490, 480,
470, 460, 450, 440, 430, 420, 410, 400,
390, 380, 370, 360, 350, 340, 330, 320,
310, 300, 290, 280, 270, 260, 250, 245,
240, 235, 230, 225, 220, 215, 210, 205,
200, 195, 190, 185, 180, 175, 170, 165,
160, 155, 150, 145, 140, 135, 130, 130 };
const ogg_uint32_t ResidueBlockFactor[Q_TABLE_SIZE] = {
3, 3, 3, 3, 3, 3, 3, 3,
3, 3, 3, 3, 3, 3, 3, 3,
3, 3, 3, 3, 3, 3, 3, 3,
3, 3, 3, 3, 3, 3, 3, 3,
2, 2, 2, 2, 2, 2, 2, 2,
2, 2, 2, 2, 2, 2, 2, 2,
2, 2, 2, 2, 2, 2, 2, 2,
2, 2, 2, 2, 2, 2, 2, 2 };

409
Engine/lib/libtheora/lib/enc/x86_32/dct_decode_mmx.c Normal file
View file

@ -0,0 +1,409 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2008 *
* by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: dct_decode_mmx.c 15400 2008-10-15 12:10:58Z tterribe $
********************************************************************/
#include <stdlib.h>
#include "../codec_internal.h"
#if defined(USE_ASM)
static const __attribute__((aligned(8),used)) ogg_int64_t OC_V3=
0x0003000300030003LL;
static const __attribute__((aligned(8),used)) ogg_int64_t OC_V4=
0x0004000400040004LL;
static void loop_filter_v(unsigned char *_pix,int _ystride,
const ogg_int16_t *_ll){
long esi;
_pix-=_ystride*2;
__asm__ __volatile__(
/*mm0=0*/
"pxor %%mm0,%%mm0\n\t"
/*esi=_ystride*3*/
"lea (%[ystride],%[ystride],2),%[s]\n\t"
/*mm7=_pix[0...8]*/
"movq (%[pix]),%%mm7\n\t"
/*mm4=_pix[0...8+_ystride*3]*/
"movq (%[pix],%[s]),%%mm4\n\t"
/*mm6=_pix[0...8]*/
"movq %%mm7,%%mm6\n\t"
/*Expand unsigned _pix[0...3] to 16 bits.*/
"punpcklbw %%mm0,%%mm6\n\t"
"movq %%mm4,%%mm5\n\t"
/*Expand unsigned _pix[4...8] to 16 bits.*/
"punpckhbw %%mm0,%%mm7\n\t"
/*Expand other arrays too.*/
"punpcklbw %%mm0,%%mm4\n\t"
"punpckhbw %%mm0,%%mm5\n\t"
/*mm7:mm6=_p[0...8]-_p[0...8+_ystride*3]:*/
"psubw %%mm4,%%mm6\n\t"
"psubw %%mm5,%%mm7\n\t"
/*mm5=mm4=_pix[0...8+_ystride]*/
"movq (%[pix],%[ystride]),%%mm4\n\t"
/*mm1=mm3=mm2=_pix[0..8]+_ystride*2]*/
"movq (%[pix],%[ystride],2),%%mm2\n\t"
"movq %%mm4,%%mm5\n\t"
"movq %%mm2,%%mm3\n\t"
"movq %%mm2,%%mm1\n\t"
/*Expand these arrays.*/
"punpckhbw %%mm0,%%mm5\n\t"
"punpcklbw %%mm0,%%mm4\n\t"
"punpckhbw %%mm0,%%mm3\n\t"
"punpcklbw %%mm0,%%mm2\n\t"
/*Preload...*/
"movq %[OC_V3],%%mm0\n\t"
/*mm3:mm2=_pix[0...8+_ystride*2]-_pix[0...8+_ystride]*/
"psubw %%mm5,%%mm3\n\t"
"psubw %%mm4,%%mm2\n\t"
/*Scale by 3.*/
"pmullw %%mm0,%%mm3\n\t"
"pmullw %%mm0,%%mm2\n\t"
/*Preload...*/
"movq %[OC_V4],%%mm0\n\t"
/*f=mm3:mm2==_pix[0...8]-_pix[0...8+_ystride*3]+
3*(_pix[0...8+_ystride*2]-_pix[0...8+_ystride])*/
"paddw %%mm7,%%mm3\n\t"
"paddw %%mm6,%%mm2\n\t"
/*Add 4.*/
"paddw %%mm0,%%mm3\n\t"
"paddw %%mm0,%%mm2\n\t"
/*"Divide" by 8.*/
"psraw $3,%%mm3\n\t"
"psraw $3,%%mm2\n\t"
/*Now compute lflim of mm3:mm2 cf. Section 7.10 of the sepc.*/
/*Free up mm5.*/
"packuswb %%mm5,%%mm4\n\t"
/*mm0=L L L L*/
"movq (%[ll]),%%mm0\n\t"
/*if(R_i<-2L||R_i>2L)R_i=0:*/
"movq %%mm2,%%mm5\n\t"
"pxor %%mm6,%%mm6\n\t"
"movq %%mm0,%%mm7\n\t"
"psubw %%mm0,%%mm6\n\t"
"psllw $1,%%mm7\n\t"
"psllw $1,%%mm6\n\t"
/*mm2==R_3 R_2 R_1 R_0*/
/*mm5==R_3 R_2 R_1 R_0*/
/*mm6==-2L -2L -2L -2L*/
/*mm7==2L 2L 2L 2L*/
"pcmpgtw %%mm2,%%mm7\n\t"
"pcmpgtw %%mm6,%%mm5\n\t"
"pand %%mm7,%%mm2\n\t"
"movq %%mm0,%%mm7\n\t"
"pand %%mm5,%%mm2\n\t"
"psllw $1,%%mm7\n\t"
"movq %%mm3,%%mm5\n\t"
/*mm3==R_7 R_6 R_5 R_4*/
/*mm5==R_7 R_6 R_5 R_4*/
/*mm6==-2L -2L -2L -2L*/
/*mm7==2L 2L 2L 2L*/
"pcmpgtw %%mm3,%%mm7\n\t"
"pcmpgtw %%mm6,%%mm5\n\t"
"pand %%mm7,%%mm3\n\t"
"movq %%mm0,%%mm7\n\t"
"pand %%mm5,%%mm3\n\t"
/*if(R_i<-L)R_i'=R_i+2L;
if(R_i>L)R_i'=R_i-2L;
if(R_i<-L||R_i>L)R_i=-R_i':*/
"psraw $1,%%mm6\n\t"
"movq %%mm2,%%mm5\n\t"
"psllw $1,%%mm7\n\t"
/*mm2==R_3 R_2 R_1 R_0*/
/*mm5==R_3 R_2 R_1 R_0*/
/*mm6==-L -L -L -L*/
/*mm0==L L L L*/
/*mm5=R_i>L?FF:00*/
"pcmpgtw %%mm0,%%mm5\n\t"
/*mm6=-L>R_i?FF:00*/
"pcmpgtw %%mm2,%%mm6\n\t"
/*mm7=R_i>L?2L:0*/
"pand %%mm5,%%mm7\n\t"
/*mm2=R_i>L?R_i-2L:R_i*/
"psubw %%mm7,%%mm2\n\t"
"movq %%mm0,%%mm7\n\t"
/*mm5=-L>R_i||R_i>L*/
"por %%mm6,%%mm5\n\t"
"psllw $1,%%mm7\n\t"
/*mm7=-L>R_i?2L:0*/
"pand %%mm6,%%mm7\n\t"
"pxor %%mm6,%%mm6\n\t"
/*mm2=-L>R_i?R_i+2L:R_i*/
"paddw %%mm7,%%mm2\n\t"
"psubw %%mm0,%%mm6\n\t"
/*mm5=-L>R_i||R_i>L?-R_i':0*/
"pand %%mm2,%%mm5\n\t"
"movq %%mm0,%%mm7\n\t"
/*mm2=-L>R_i||R_i>L?0:R_i*/
"psubw %%mm5,%%mm2\n\t"
"psllw $1,%%mm7\n\t"
/*mm2=-L>R_i||R_i>L?-R_i':R_i*/
"psubw %%mm5,%%mm2\n\t"
"movq %%mm3,%%mm5\n\t"
/*mm3==R_7 R_6 R_5 R_4*/
/*mm5==R_7 R_6 R_5 R_4*/
/*mm6==-L -L -L -L*/
/*mm0==L L L L*/
/*mm6=-L>R_i?FF:00*/
"pcmpgtw %%mm3,%%mm6\n\t"
/*mm5=R_i>L?FF:00*/
"pcmpgtw %%mm0,%%mm5\n\t"
/*mm7=R_i>L?2L:0*/
"pand %%mm5,%%mm7\n\t"
/*mm2=R_i>L?R_i-2L:R_i*/
"psubw %%mm7,%%mm3\n\t"
"psllw $1,%%mm0\n\t"
/*mm5=-L>R_i||R_i>L*/
"por %%mm6,%%mm5\n\t"
/*mm0=-L>R_i?2L:0*/
"pand %%mm6,%%mm0\n\t"
/*mm3=-L>R_i?R_i+2L:R_i*/
"paddw %%mm0,%%mm3\n\t"
/*mm5=-L>R_i||R_i>L?-R_i':0*/
"pand %%mm3,%%mm5\n\t"
/*mm2=-L>R_i||R_i>L?0:R_i*/
"psubw %%mm5,%%mm3\n\t"
/*mm2=-L>R_i||R_i>L?-R_i':R_i*/
"psubw %%mm5,%%mm3\n\t"
/*Unfortunately, there's no unsigned byte+signed byte with unsigned
saturation op code, so we have to promote things back 16 bits.*/
"pxor %%mm0,%%mm0\n\t"
"movq %%mm4,%%mm5\n\t"
"punpcklbw %%mm0,%%mm4\n\t"
"punpckhbw %%mm0,%%mm5\n\t"
"movq %%mm1,%%mm6\n\t"
"punpcklbw %%mm0,%%mm1\n\t"
"punpckhbw %%mm0,%%mm6\n\t"
/*_pix[0...8+_ystride]+=R_i*/
"paddw %%mm2,%%mm4\n\t"
"paddw %%mm3,%%mm5\n\t"
/*_pix[0...8+_ystride*2]-=R_i*/
"psubw %%mm2,%%mm1\n\t"
"psubw %%mm3,%%mm6\n\t"
"packuswb %%mm5,%%mm4\n\t"
"packuswb %%mm6,%%mm1\n\t"
/*Write it back out.*/
"movq %%mm4,(%[pix],%[ystride])\n\t"
"movq %%mm1,(%[pix],%[ystride],2)\n\t"
:[s]"=&S"(esi)
:[pix]"r"(_pix),[ystride]"r"((long)_ystride),[ll]"r"(_ll),
[OC_V3]"m"(OC_V3),[OC_V4]"m"(OC_V4)
:"memory"
);
}
/*This code implements the bulk of loop_filter_h().
Data are striped p0 p1 p2 p3 ... p0 p1 p2 p3 ..., so in order to load all
four p0's to one register we must transpose the values in four mmx regs.
When half is done we repeat this for the rest.*/
static void loop_filter_h4(unsigned char *_pix,long _ystride,
const ogg_int16_t *_ll){
long esi;
long edi;
__asm__ __volatile__(
/*x x x x 3 2 1 0*/
"movd (%[pix]),%%mm0\n\t"
/*esi=_ystride*3*/
"lea (%[ystride],%[ystride],2),%[s]\n\t"
/*x x x x 7 6 5 4*/
"movd (%[pix],%[ystride]),%%mm1\n\t"
/*x x x x B A 9 8*/
"movd (%[pix],%[ystride],2),%%mm2\n\t"
/*x x x x F E D C*/
"movd (%[pix],%[s]),%%mm3\n\t"
/*mm0=7 3 6 2 5 1 4 0*/
"punpcklbw %%mm1,%%mm0\n\t"
/*mm2=F B E A D 9 C 8*/
"punpcklbw %%mm3,%%mm2\n\t"
/*mm1=7 3 6 2 5 1 4 0*/
"movq %%mm0,%%mm1\n\t"
/*mm0=F B 7 3 E A 6 2*/
"punpckhwd %%mm2,%%mm0\n\t"
/*mm1=D 9 5 1 C 8 4 0*/
"punpcklwd %%mm2,%%mm1\n\t"
"pxor %%mm7,%%mm7\n\t"
/*mm5=D 9 5 1 C 8 4 0*/
"movq %%mm1,%%mm5\n\t"
/*mm1=x C x 8 x 4 x 0==pix[0]*/
"punpcklbw %%mm7,%%mm1\n\t"
/*mm5=x D x 9 x 5 x 1==pix[1]*/
"punpckhbw %%mm7,%%mm5\n\t"
/*mm3=F B 7 3 E A 6 2*/
"movq %%mm0,%%mm3\n\t"
/*mm0=x E x A x 6 x 2==pix[2]*/
"punpcklbw %%mm7,%%mm0\n\t"
/*mm3=x F x B x 7 x 3==pix[3]*/
"punpckhbw %%mm7,%%mm3\n\t"
/*mm1=mm1-mm3==pix[0]-pix[3]*/
"psubw %%mm3,%%mm1\n\t"
/*Save a copy of pix[2] for later.*/
"movq %%mm0,%%mm4\n\t"
/*mm0=mm0-mm5==pix[2]-pix[1]*/
"psubw %%mm5,%%mm0\n\t"
/*Scale by 3.*/
"pmullw %[OC_V3],%%mm0\n\t"
/*f=mm1==_pix[0]-_pix[3]+ 3*(_pix[2]-_pix[1])*/
"paddw %%mm1,%%mm0\n\t"
/*Add 4.*/
"paddw %[OC_V4],%%mm0\n\t"
/*"Divide" by 8, producing the residuals R_i.*/
"psraw $3,%%mm0\n\t"
/*Now compute lflim of mm0 cf. Section 7.10 of the sepc.*/
/*mm6=L L L L*/
"movq (%[ll]),%%mm6\n\t"
/*if(R_i<-2L||R_i>2L)R_i=0:*/
"movq %%mm0,%%mm1\n\t"
"pxor %%mm2,%%mm2\n\t"
"movq %%mm6,%%mm3\n\t"
"psubw %%mm6,%%mm2\n\t"
"psllw $1,%%mm3\n\t"
"psllw $1,%%mm2\n\t"
/*mm0==R_3 R_2 R_1 R_0*/
/*mm1==R_3 R_2 R_1 R_0*/
/*mm2==-2L -2L -2L -2L*/
/*mm3==2L 2L 2L 2L*/
"pcmpgtw %%mm0,%%mm3\n\t"
"pcmpgtw %%mm2,%%mm1\n\t"
"pand %%mm3,%%mm0\n\t"
"pand %%mm1,%%mm0\n\t"
/*if(R_i<-L)R_i'=R_i+2L;
if(R_i>L)R_i'=R_i-2L;
if(R_i<-L||R_i>L)R_i=-R_i':*/
"psraw $1,%%mm2\n\t"
"movq %%mm0,%%mm1\n\t"
"movq %%mm6,%%mm3\n\t"
/*mm0==R_3 R_2 R_1 R_0*/
/*mm1==R_3 R_2 R_1 R_0*/
/*mm2==-L -L -L -L*/
/*mm6==L L L L*/
/*mm2=-L>R_i?FF:00*/
"pcmpgtw %%mm0,%%mm2\n\t"
/*mm1=R_i>L?FF:00*/
"pcmpgtw %%mm6,%%mm1\n\t"
/*mm3=2L 2L 2L 2L*/
"psllw $1,%%mm3\n\t"
/*mm6=2L 2L 2L 2L*/
"psllw $1,%%mm6\n\t"
/*mm3=R_i>L?2L:0*/
"pand %%mm1,%%mm3\n\t"
/*mm6=-L>R_i?2L:0*/
"pand %%mm2,%%mm6\n\t"
/*mm0=R_i>L?R_i-2L:R_i*/
"psubw %%mm3,%%mm0\n\t"
/*mm1=-L>R_i||R_i>L*/
"por %%mm2,%%mm1\n\t"
/*mm0=-L>R_i?R_i+2L:R_i*/
"paddw %%mm6,%%mm0\n\t"
/*mm1=-L>R_i||R_i>L?R_i':0*/
"pand %%mm0,%%mm1\n\t"
/*mm0=-L>R_i||R_i>L?0:R_i*/
"psubw %%mm1,%%mm0\n\t"
/*mm0=-L>R_i||R_i>L?-R_i':R_i*/
"psubw %%mm1,%%mm0\n\t"
/*_pix[1]+=R_i;*/
"paddw %%mm0,%%mm5\n\t"
/*_pix[2]-=R_i;*/
"psubw %%mm0,%%mm4\n\t"
/*mm5=x x x x D 9 5 1*/
"packuswb %%mm7,%%mm5\n\t"
/*mm4=x x x x E A 6 2*/
"packuswb %%mm7,%%mm4\n\t"
/*mm5=E D A 9 6 5 2 1*/
"punpcklbw %%mm4,%%mm5\n\t"
/*edi=6 5 2 1*/
"movd %%mm5,%%edi\n\t"
"movw %%di,1(%[pix])\n\t"
/*Why is there such a big stall here?*/
"psrlq $32,%%mm5\n\t"
"shrl $16,%%edi\n\t"
"movw %%di,1(%[pix],%[ystride])\n\t"
/*edi=E D A 9*/
"movd %%mm5,%%edi\n\t"
"movw %%di,1(%[pix],%[ystride],2)\n\t"
"shrl $16,%%edi\n\t"
"movw %%di,1(%[pix],%[s])\n\t"
:[s]"=&S"(esi),[d]"=&D"(edi),
[pix]"+r"(_pix),[ystride]"+r"(_ystride),[ll]"+r"(_ll)
:[OC_V3]"m"(OC_V3),[OC_V4]"m"(OC_V4)
:"memory"
);
}
static void loop_filter_h(unsigned char *_pix,int _ystride,
const ogg_int16_t *_ll){
_pix-=2;
loop_filter_h4(_pix,_ystride,_ll);
loop_filter_h4(_pix+(_ystride<<2),_ystride,_ll);
}
static void loop_filter_mmx(PB_INSTANCE *pbi, int FLimit){
int j;
ogg_int16_t __attribute__((aligned(8))) ll[4];
unsigned char *cp = pbi->display_fragments;
ogg_uint32_t *bp = pbi->recon_pixel_index_table;
if ( FLimit == 0 ) return;
ll[0]=ll[1]=ll[2]=ll[3]=FLimit;
for ( j = 0; j < 3 ; j++){
ogg_uint32_t *bp_begin = bp;
ogg_uint32_t *bp_end;
int stride;
int h;
switch(j) {
case 0: /* y */
bp_end = bp + pbi->YPlaneFragments;
h = pbi->HFragments;
stride = pbi->YStride;
break;
default: /* u,v, 4:20 specific */
bp_end = bp + pbi->UVPlaneFragments;
h = pbi->HFragments >> 1;
stride = pbi->UVStride;
break;
}
while(bp<bp_end){
ogg_uint32_t *bp_left = bp;
ogg_uint32_t *bp_right = bp + h;
while(bp<bp_right){
if(cp[0]){
if(bp>bp_left)
loop_filter_h(&pbi->LastFrameRecon[bp[0]],stride,ll);
if(bp_left>bp_begin)
loop_filter_v(&pbi->LastFrameRecon[bp[0]],stride,ll);
if(bp+1<bp_right && !cp[1])
loop_filter_h(&pbi->LastFrameRecon[bp[0]]+8,stride,ll);
if(bp+h<bp_end && !cp[h])
loop_filter_v(&pbi->LastFrameRecon[bp[h]],stride,ll);
}
bp++;
cp++;
}
}
}
__asm__ __volatile__("emms\n\t");
}
/* install our implementation in the function table */
void dsp_mmx_dct_decode_init(DspFunctions *funcs)
{
funcs->LoopFilter = loop_filter_mmx;
}
#endif /* USE_ASM */

666
Engine/lib/libtheora/lib/enc/x86_32/dsp_mmx.c Normal file
View file

@ -0,0 +1,666 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: dsp_mmx.c 15153 2008-08-04 18:37:55Z tterribe $
********************************************************************/
#include <stdlib.h>
#include "../codec_internal.h"
#include "../dsp.h"
#if defined(USE_ASM)
static const __attribute__ ((aligned(8),used)) ogg_int64_t V128 = 0x0080008000800080LL;
#define DSP_OP_AVG(a,b) ((((int)(a)) + ((int)(b)))/2)
#define DSP_OP_DIFF(a,b) (((int)(a)) - ((int)(b)))
#define DSP_OP_ABS_DIFF(a,b) abs((((int)(a)) - ((int)(b))))
#define SUB_LOOP \
" movq (%0), %%mm0 \n\t" /* mm0 = FiltPtr */ \
" movq (%1), %%mm1 \n\t" /* mm1 = ReconPtr */ \
" movq %%mm0, %%mm2 \n\t" /* dup to prepare for up conversion */\
" movq %%mm1, %%mm3 \n\t" /* dup to prepare for up conversion */\
/* convert from UINT8 to INT16 */ \
" punpcklbw %%mm7, %%mm0 \n\t" /* mm0 = INT16(FiltPtr) */ \
" punpcklbw %%mm7, %%mm1 \n\t" /* mm1 = INT16(ReconPtr) */ \
" punpckhbw %%mm7, %%mm2 \n\t" /* mm2 = INT16(FiltPtr) */ \
" punpckhbw %%mm7, %%mm3 \n\t" /* mm3 = INT16(ReconPtr) */ \
/* start calculation */ \
" psubw %%mm1, %%mm0 \n\t" /* mm0 = FiltPtr - ReconPtr */ \
" psubw %%mm3, %%mm2 \n\t" /* mm2 = FiltPtr - ReconPtr */ \
" movq %%mm0, (%2) \n\t" /* write answer out */ \
" movq %%mm2, 8(%2) \n\t" /* write answer out */ \
/* Increment pointers */ \
" add $16, %2 \n\t" \
" add %3, %0 \n\t" \
" add %4, %1 \n\t"
static void sub8x8__mmx (unsigned char *FiltPtr, unsigned char *ReconPtr,
ogg_int16_t *DctInputPtr, ogg_uint32_t PixelsPerLine,
ogg_uint32_t ReconPixelsPerLine)
{
__asm__ __volatile__ (
" .p2align 4 \n\t"
" pxor %%mm7, %%mm7 \n\t"
SUB_LOOP
SUB_LOOP
SUB_LOOP
SUB_LOOP
SUB_LOOP
SUB_LOOP
SUB_LOOP
SUB_LOOP
: "+r" (FiltPtr),
"+r" (ReconPtr),
"+r" (DctInputPtr)
: "m" (PixelsPerLine),
"m" (ReconPixelsPerLine)
: "memory"
);
}
#define SUB_128_LOOP \
" movq (%0), %%mm0 \n\t" /* mm0 = FiltPtr */ \
" movq %%mm0, %%mm2 \n\t" /* dup to prepare for up conversion */\
/* convert from UINT8 to INT16 */ \
" punpcklbw %%mm7, %%mm0 \n\t" /* mm0 = INT16(FiltPtr) */ \
" punpckhbw %%mm7, %%mm2 \n\t" /* mm2 = INT16(FiltPtr) */ \
/* start calculation */ \
" psubw %%mm1, %%mm0 \n\t" /* mm0 = FiltPtr - 128 */ \
" psubw %%mm1, %%mm2 \n\t" /* mm2 = FiltPtr - 128 */ \
" movq %%mm0, (%1) \n\t" /* write answer out */ \
" movq %%mm2, 8(%1) \n\t" /* write answer out */ \
/* Increment pointers */ \
" add $16, %1 \n\t" \
" add %2, %0 \n\t"
static void sub8x8_128__mmx (unsigned char *FiltPtr, ogg_int16_t *DctInputPtr,
ogg_uint32_t PixelsPerLine)
{
__asm__ __volatile__ (
" .p2align 4 \n\t"
" pxor %%mm7, %%mm7 \n\t"
" movq %[V128], %%mm1 \n\t"
SUB_128_LOOP
SUB_128_LOOP
SUB_128_LOOP
SUB_128_LOOP
SUB_128_LOOP
SUB_128_LOOP
SUB_128_LOOP
SUB_128_LOOP
: "+r" (FiltPtr),
"+r" (DctInputPtr)
: "m" (PixelsPerLine),
[V128] "m" (V128)
: "memory"
);
}
#define SUB_AVG2_LOOP \
" movq (%0), %%mm0 \n\t" /* mm0 = FiltPtr */ \
" movq (%1), %%mm1 \n\t" /* mm1 = ReconPtr1 */ \
" movq (%2), %%mm4 \n\t" /* mm1 = ReconPtr2 */ \
" movq %%mm0, %%mm2 \n\t" /* dup to prepare for up conversion */\
" movq %%mm1, %%mm3 \n\t" /* dup to prepare for up conversion */\
" movq %%mm4, %%mm5 \n\t" /* dup to prepare for up conversion */\
/* convert from UINT8 to INT16 */ \
" punpcklbw %%mm7, %%mm0 \n\t" /* mm0 = INT16(FiltPtr) */ \
" punpcklbw %%mm7, %%mm1 \n\t" /* mm1 = INT16(ReconPtr1) */ \
" punpcklbw %%mm7, %%mm4 \n\t" /* mm1 = INT16(ReconPtr2) */ \
" punpckhbw %%mm7, %%mm2 \n\t" /* mm2 = INT16(FiltPtr) */ \
" punpckhbw %%mm7, %%mm3 \n\t" /* mm3 = INT16(ReconPtr1) */ \
" punpckhbw %%mm7, %%mm5 \n\t" /* mm3 = INT16(ReconPtr2) */ \
/* average ReconPtr1 and ReconPtr2 */ \
" paddw %%mm4, %%mm1 \n\t" /* mm1 = ReconPtr1 + ReconPtr2 */ \
" paddw %%mm5, %%mm3 \n\t" /* mm3 = ReconPtr1 + ReconPtr2 */ \
" psrlw $1, %%mm1 \n\t" /* mm1 = (ReconPtr1 + ReconPtr2) / 2 */ \
" psrlw $1, %%mm3 \n\t" /* mm3 = (ReconPtr1 + ReconPtr2) / 2 */ \
" psubw %%mm1, %%mm0 \n\t" /* mm0 = FiltPtr - ((ReconPtr1 + ReconPtr2) / 2) */ \
" psubw %%mm3, %%mm2 \n\t" /* mm2 = FiltPtr - ((ReconPtr1 + ReconPtr2) / 2) */ \
" movq %%mm0, (%3) \n\t" /* write answer out */ \
" movq %%mm2, 8(%3) \n\t" /* write answer out */ \
/* Increment pointers */ \
" add $16, %3 \n\t" \
" add %4, %0 \n\t" \
" add %5, %1 \n\t" \
" add %5, %2 \n\t"
static void sub8x8avg2__mmx (unsigned char *FiltPtr, unsigned char *ReconPtr1,
unsigned char *ReconPtr2, ogg_int16_t *DctInputPtr,
ogg_uint32_t PixelsPerLine,
ogg_uint32_t ReconPixelsPerLine)
{
__asm__ __volatile__ (
" .p2align 4 \n\t"
" pxor %%mm7, %%mm7 \n\t"
SUB_AVG2_LOOP
SUB_AVG2_LOOP
SUB_AVG2_LOOP
SUB_AVG2_LOOP
SUB_AVG2_LOOP
SUB_AVG2_LOOP
SUB_AVG2_LOOP
SUB_AVG2_LOOP
: "+r" (FiltPtr),
"+r" (ReconPtr1),
"+r" (ReconPtr2),
"+r" (DctInputPtr)
: "m" (PixelsPerLine),
"m" (ReconPixelsPerLine)
: "memory"
);
}
static ogg_uint32_t row_sad8__mmx (unsigned char *Src1, unsigned char *Src2)
{
ogg_uint32_t MaxSad;
__asm__ __volatile__ (
" .p2align 4 \n\t"
" pxor %%mm6, %%mm6 \n\t" /* zero out mm6 for unpack */
" pxor %%mm7, %%mm7 \n\t" /* zero out mm7 for unpack */
" movq (%1), %%mm0 \n\t" /* take 8 bytes */
" movq (%2), %%mm1 \n\t"
" movq %%mm0, %%mm2 \n\t"
" psubusb %%mm1, %%mm0 \n\t" /* A - B */
" psubusb %%mm2, %%mm1 \n\t" /* B - A */
" por %%mm1, %%mm0 \n\t" /* and or gives abs difference */
" movq %%mm0, %%mm1 \n\t"
" punpcklbw %%mm6, %%mm0 \n\t" /* ; unpack low four bytes to higher precision */
" punpckhbw %%mm7, %%mm1 \n\t" /* ; unpack high four bytes to higher precision */
" movq %%mm0, %%mm2 \n\t"
" movq %%mm1, %%mm3 \n\t"
" psrlq $32, %%mm2 \n\t" /* fold and add */
" psrlq $32, %%mm3 \n\t"
" paddw %%mm2, %%mm0 \n\t"
" paddw %%mm3, %%mm1 \n\t"
" movq %%mm0, %%mm2 \n\t"
" movq %%mm1, %%mm3 \n\t"
" psrlq $16, %%mm2 \n\t"
" psrlq $16, %%mm3 \n\t"
" paddw %%mm2, %%mm0 \n\t"
" paddw %%mm3, %%mm1 \n\t"
" psubusw %%mm0, %%mm1 \n\t"
" paddw %%mm0, %%mm1 \n\t" /* mm1 = max(mm1, mm0) */
" movd %%mm1, %0 \n\t"
" andl $0xffff, %0 \n\t"
: "=m" (MaxSad),
"+r" (Src1),
"+r" (Src2)
:
: "memory"
);
return MaxSad;
}
static ogg_uint32_t col_sad8x8__mmx (unsigned char *Src1, unsigned char *Src2,
ogg_uint32_t stride)
{
ogg_uint32_t MaxSad;
__asm__ __volatile__ (
" .p2align 4 \n\t"
" pxor %%mm3, %%mm3 \n\t" /* zero out mm3 for unpack */
" pxor %%mm4, %%mm4 \n\t" /* mm4 low sum */
" pxor %%mm5, %%mm5 \n\t" /* mm5 high sum */
" pxor %%mm6, %%mm6 \n\t" /* mm6 low sum */
" pxor %%mm7, %%mm7 \n\t" /* mm7 high sum */
" mov $4, %%edi \n\t" /* 4 rows */
"1: \n\t"
" movq (%1), %%mm0 \n\t" /* take 8 bytes */
" movq (%2), %%mm1 \n\t" /* take 8 bytes */
" movq %%mm0, %%mm2 \n\t"
" psubusb %%mm1, %%mm0 \n\t" /* A - B */
" psubusb %%mm2, %%mm1 \n\t" /* B - A */
" por %%mm1, %%mm0 \n\t" /* and or gives abs difference */
" movq %%mm0, %%mm1 \n\t"
" punpcklbw %%mm3, %%mm0 \n\t" /* unpack to higher precision for accumulation */
" paddw %%mm0, %%mm4 \n\t" /* accumulate difference... */
" punpckhbw %%mm3, %%mm1 \n\t" /* unpack high four bytes to higher precision */
" paddw %%mm1, %%mm5 \n\t" /* accumulate difference... */
" add %3, %1 \n\t" /* Inc pointer into the new data */
" add %3, %2 \n\t" /* Inc pointer into the new data */
" dec %%edi \n\t"
" jnz 1b \n\t"
" mov $4, %%edi \n\t" /* 4 rows */
"2: \n\t"
" movq (%1), %%mm0 \n\t" /* take 8 bytes */
" movq (%2), %%mm1 \n\t" /* take 8 bytes */
" movq %%mm0, %%mm2 \n\t"
" psubusb %%mm1, %%mm0 \n\t" /* A - B */
" psubusb %%mm2, %%mm1 \n\t" /* B - A */
" por %%mm1, %%mm0 \n\t" /* and or gives abs difference */
" movq %%mm0, %%mm1 \n\t"
" punpcklbw %%mm3, %%mm0 \n\t" /* unpack to higher precision for accumulation */
" paddw %%mm0, %%mm6 \n\t" /* accumulate difference... */
" punpckhbw %%mm3, %%mm1 \n\t" /* unpack high four bytes to higher precision */
" paddw %%mm1, %%mm7 \n\t" /* accumulate difference... */
" add %3, %1 \n\t" /* Inc pointer into the new data */
" add %3, %2 \n\t" /* Inc pointer into the new data */
" dec %%edi \n\t"
" jnz 2b \n\t"
" psubusw %%mm6, %%mm7 \n\t"
" paddw %%mm6, %%mm7 \n\t" /* mm7 = max(mm7, mm6) */
" psubusw %%mm4, %%mm5 \n\t"
" paddw %%mm4, %%mm5 \n\t" /* mm5 = max(mm5, mm4) */
" psubusw %%mm5, %%mm7 \n\t"
" paddw %%mm5, %%mm7 \n\t" /* mm7 = max(mm5, mm7) */
" movq %%mm7, %%mm6 \n\t"
" psrlq $32, %%mm6 \n\t"
" psubusw %%mm6, %%mm7 \n\t"
" paddw %%mm6, %%mm7 \n\t" /* mm7 = max(mm5, mm7) */
" movq %%mm7, %%mm6 \n\t"
" psrlq $16, %%mm6 \n\t"
" psubusw %%mm6, %%mm7 \n\t"
" paddw %%mm6, %%mm7 \n\t" /* mm7 = max(mm5, mm7) */
" movd %%mm7, %0 \n\t"
" andl $0xffff, %0 \n\t"
: "=r" (MaxSad),
"+r" (Src1),
"+r" (Src2)
: "r" (stride)
: "memory", "edi"
);
return MaxSad;
}
#define SAD_LOOP \
" movq (%1), %%mm0 \n\t" /* take 8 bytes */ \
" movq (%2), %%mm1 \n\t" \
" movq %%mm0, %%mm2 \n\t" \
" psubusb %%mm1, %%mm0 \n\t" /* A - B */ \
" psubusb %%mm2, %%mm1 \n\t" /* B - A */ \
" por %%mm1, %%mm0 \n\t" /* and or gives abs difference */ \
" movq %%mm0, %%mm1 \n\t" \
" punpcklbw %%mm6, %%mm0 \n\t" /* unpack to higher precision for accumulation */ \
" paddw %%mm0, %%mm7 \n\t" /* accumulate difference... */ \
" punpckhbw %%mm6, %%mm1 \n\t" /* unpack high four bytes to higher precision */ \
" add %3, %1 \n\t" /* Inc pointer into the new data */ \
" paddw %%mm1, %%mm7 \n\t" /* accumulate difference... */ \
" add %4, %2 \n\t" /* Inc pointer into ref data */
static ogg_uint32_t sad8x8__mmx (unsigned char *ptr1, ogg_uint32_t stride1,
unsigned char *ptr2, ogg_uint32_t stride2)
{
ogg_uint32_t DiffVal;
__asm__ __volatile__ (
" .p2align 4 \n\t"
" pxor %%mm6, %%mm6 \n\t" /* zero out mm6 for unpack */
" pxor %%mm7, %%mm7 \n\t" /* mm7 contains the result */
SAD_LOOP
SAD_LOOP
SAD_LOOP
SAD_LOOP
SAD_LOOP
SAD_LOOP
SAD_LOOP
SAD_LOOP
" movq %%mm7, %%mm0 \n\t"
" psrlq $32, %%mm7 \n\t"
" paddw %%mm0, %%mm7 \n\t"
" movq %%mm7, %%mm0 \n\t"
" psrlq $16, %%mm7 \n\t"
" paddw %%mm0, %%mm7 \n\t"
" movd %%mm7, %0 \n\t"
" andl $0xffff, %0 \n\t"
: "=m" (DiffVal),
"+r" (ptr1),
"+r" (ptr2)
: "r" (stride1),
"r" (stride2)
: "memory"
);
return DiffVal;
}
static ogg_uint32_t sad8x8_thres__mmx (unsigned char *ptr1, ogg_uint32_t stride1,
unsigned char *ptr2, ogg_uint32_t stride2,
ogg_uint32_t thres)
{
return sad8x8__mmx (ptr1, stride1, ptr2, stride2);
}
static ogg_uint32_t sad8x8_xy2_thres__mmx (unsigned char *SrcData, ogg_uint32_t SrcStride,
unsigned char *RefDataPtr1,
unsigned char *RefDataPtr2, ogg_uint32_t RefStride,
ogg_uint32_t thres)
{
ogg_uint32_t DiffVal;
__asm__ __volatile__ (
" .p2align 4 \n\t"
" pcmpeqd %%mm5, %%mm5 \n\t" /* fefefefefefefefe in mm5 */
" paddb %%mm5, %%mm5 \n\t"
" pxor %%mm6, %%mm6 \n\t" /* zero out mm6 for unpack */
" pxor %%mm7, %%mm7 \n\t" /* mm7 contains the result */
" mov $8, %%edi \n\t" /* 8 rows */
"1: \n\t"
" movq (%1), %%mm0 \n\t" /* take 8 bytes */
" movq (%2), %%mm2 \n\t"
" movq (%3), %%mm3 \n\t" /* take average of mm2 and mm3 */
" movq %%mm2, %%mm1 \n\t"
" pand %%mm3, %%mm1 \n\t"
" pxor %%mm2, %%mm3 \n\t"
" pand %%mm5, %%mm3 \n\t"
" psrlq $1, %%mm3 \n\t"
" paddb %%mm3, %%mm1 \n\t"
" movq %%mm0, %%mm2 \n\t"
" psubusb %%mm1, %%mm0 \n\t" /* A - B */
" psubusb %%mm2, %%mm1 \n\t" /* B - A */
" por %%mm1, %%mm0 \n\t" /* and or gives abs difference */
" movq %%mm0, %%mm1 \n\t"
" punpcklbw %%mm6, %%mm0 \n\t" /* unpack to higher precision for accumulation */
" paddw %%mm0, %%mm7 \n\t" /* accumulate difference... */
" punpckhbw %%mm6, %%mm1 \n\t" /* unpack high four bytes to higher precision */
" add %4, %1 \n\t" /* Inc pointer into the new data */
" paddw %%mm1, %%mm7 \n\t" /* accumulate difference... */
" add %5, %2 \n\t" /* Inc pointer into ref data */
" add %5, %3 \n\t" /* Inc pointer into ref data */
" dec %%edi \n\t"
" jnz 1b \n\t"
" movq %%mm7, %%mm0 \n\t"
" psrlq $32, %%mm7 \n\t"
" paddw %%mm0, %%mm7 \n\t"
" movq %%mm7, %%mm0 \n\t"
" psrlq $16, %%mm7 \n\t"
" paddw %%mm0, %%mm7 \n\t"
" movd %%mm7, %0 \n\t"
" andl $0xffff, %0 \n\t"
: "=m" (DiffVal),
"+r" (SrcData),
"+r" (RefDataPtr1),
"+r" (RefDataPtr2)
: "m" (SrcStride),
"m" (RefStride)
: "edi", "memory"
);
return DiffVal;
}
static ogg_uint32_t intra8x8_err__mmx (unsigned char *DataPtr, ogg_uint32_t Stride)
{
ogg_uint32_t XSum;
ogg_uint32_t XXSum;
__asm__ __volatile__ (
" .p2align 4 \n\t"
" pxor %%mm5, %%mm5 \n\t"
" pxor %%mm6, %%mm6 \n\t"
" pxor %%mm7, %%mm7 \n\t"
" mov $8, %%edi \n\t"
"1: \n\t"
" movq (%2), %%mm0 \n\t" /* take 8 bytes */
" movq %%mm0, %%mm2 \n\t"
" punpcklbw %%mm6, %%mm0 \n\t"
" punpckhbw %%mm6, %%mm2 \n\t"
" paddw %%mm0, %%mm5 \n\t"
" paddw %%mm2, %%mm5 \n\t"
" pmaddwd %%mm0, %%mm0 \n\t"
" pmaddwd %%mm2, %%mm2 \n\t"
" paddd %%mm0, %%mm7 \n\t"
" paddd %%mm2, %%mm7 \n\t"
" add %3, %2 \n\t" /* Inc pointer into src data */
" dec %%edi \n\t"
" jnz 1b \n\t"
" movq %%mm5, %%mm0 \n\t"
" psrlq $32, %%mm5 \n\t"
" paddw %%mm0, %%mm5 \n\t"
" movq %%mm5, %%mm0 \n\t"
" psrlq $16, %%mm5 \n\t"
" paddw %%mm0, %%mm5 \n\t"
" movd %%mm5, %%edi \n\t"
" movsx %%di, %%edi \n\t"
" movl %%edi, %0 \n\t"
" movq %%mm7, %%mm0 \n\t"
" psrlq $32, %%mm7 \n\t"
" paddd %%mm0, %%mm7 \n\t"
" movd %%mm7, %1 \n\t"
: "=r" (XSum),
"=r" (XXSum),
"+r" (DataPtr)
: "r" (Stride)
: "edi", "memory"
);
/* Compute population variance as mis-match metric. */
return (( (XXSum<<6) - XSum*XSum ) );
}
static ogg_uint32_t inter8x8_err__mmx (unsigned char *SrcData, ogg_uint32_t SrcStride,
unsigned char *RefDataPtr, ogg_uint32_t RefStride)
{
ogg_uint32_t XSum;
ogg_uint32_t XXSum;
__asm__ __volatile__ (
" .p2align 4 \n\t"
" pxor %%mm5, %%mm5 \n\t"
" pxor %%mm6, %%mm6 \n\t"
" pxor %%mm7, %%mm7 \n\t"
" mov $8, %%edi \n\t"
"1: \n\t"
" movq (%2), %%mm0 \n\t" /* take 8 bytes */
" movq (%3), %%mm1 \n\t"
" movq %%mm0, %%mm2 \n\t"
" movq %%mm1, %%mm3 \n\t"
" punpcklbw %%mm6, %%mm0 \n\t"
" punpcklbw %%mm6, %%mm1 \n\t"
" punpckhbw %%mm6, %%mm2 \n\t"
" punpckhbw %%mm6, %%mm3 \n\t"
" psubsw %%mm1, %%mm0 \n\t"
" psubsw %%mm3, %%mm2 \n\t"
" paddw %%mm0, %%mm5 \n\t"
" paddw %%mm2, %%mm5 \n\t"
" pmaddwd %%mm0, %%mm0 \n\t"
" pmaddwd %%mm2, %%mm2 \n\t"
" paddd %%mm0, %%mm7 \n\t"
" paddd %%mm2, %%mm7 \n\t"
" add %4, %2 \n\t" /* Inc pointer into src data */
" add %5, %3 \n\t" /* Inc pointer into ref data */
" dec %%edi \n\t"
" jnz 1b \n\t"
" movq %%mm5, %%mm0 \n\t"
" psrlq $32, %%mm5 \n\t"
" paddw %%mm0, %%mm5 \n\t"
" movq %%mm5, %%mm0 \n\t"
" psrlq $16, %%mm5 \n\t"
" paddw %%mm0, %%mm5 \n\t"
" movd %%mm5, %%edi \n\t"
" movsx %%di, %%edi \n\t"
" movl %%edi, %0 \n\t"
" movq %%mm7, %%mm0 \n\t"
" psrlq $32, %%mm7 \n\t"
" paddd %%mm0, %%mm7 \n\t"
" movd %%mm7, %1 \n\t"
: "=m" (XSum),
"=m" (XXSum),
"+r" (SrcData),
"+r" (RefDataPtr)
: "m" (SrcStride),
"m" (RefStride)
: "edi", "memory"
);
/* Compute and return population variance as mis-match metric. */
return (( (XXSum<<6) - XSum*XSum ));
}
static ogg_uint32_t inter8x8_err_xy2__mmx (unsigned char *SrcData, ogg_uint32_t SrcStride,
unsigned char *RefDataPtr1,
unsigned char *RefDataPtr2, ogg_uint32_t RefStride)
{
ogg_uint32_t XSum;
ogg_uint32_t XXSum;
__asm__ __volatile__ (
" .p2align 4 \n\t"
" pcmpeqd %%mm4, %%mm4 \n\t" /* fefefefefefefefe in mm4 */
" paddb %%mm4, %%mm4 \n\t"
" pxor %%mm5, %%mm5 \n\t"
" pxor %%mm6, %%mm6 \n\t"
" pxor %%mm7, %%mm7 \n\t"
" mov $8, %%edi \n\t"
"1: \n\t"
" movq (%2), %%mm0 \n\t" /* take 8 bytes */
" movq (%3), %%mm2 \n\t"
" movq (%4), %%mm3 \n\t" /* take average of mm2 and mm3 */
" movq %%mm2, %%mm1 \n\t"
" pand %%mm3, %%mm1 \n\t"
" pxor %%mm2, %%mm3 \n\t"
" pand %%mm4, %%mm3 \n\t"
" psrlq $1, %%mm3 \n\t"
" paddb %%mm3, %%mm1 \n\t"
" movq %%mm0, %%mm2 \n\t"
" movq %%mm1, %%mm3 \n\t"
" punpcklbw %%mm6, %%mm0 \n\t"
" punpcklbw %%mm6, %%mm1 \n\t"
" punpckhbw %%mm6, %%mm2 \n\t"
" punpckhbw %%mm6, %%mm3 \n\t"
" psubsw %%mm1, %%mm0 \n\t"
" psubsw %%mm3, %%mm2 \n\t"
" paddw %%mm0, %%mm5 \n\t"
" paddw %%mm2, %%mm5 \n\t"
" pmaddwd %%mm0, %%mm0 \n\t"
" pmaddwd %%mm2, %%mm2 \n\t"
" paddd %%mm0, %%mm7 \n\t"
" paddd %%mm2, %%mm7 \n\t"
" add %5, %2 \n\t" /* Inc pointer into src data */
" add %6, %3 \n\t" /* Inc pointer into ref data */
" add %6, %4 \n\t" /* Inc pointer into ref data */
" dec %%edi \n\t"
" jnz 1b \n\t"
" movq %%mm5, %%mm0 \n\t"
" psrlq $32, %%mm5 \n\t"
" paddw %%mm0, %%mm5 \n\t"
" movq %%mm5, %%mm0 \n\t"
" psrlq $16, %%mm5 \n\t"
" paddw %%mm0, %%mm5 \n\t"
" movd %%mm5, %%edi \n\t"
" movsx %%di, %%edi \n\t"
" movl %%edi, %0 \n\t"
" movq %%mm7, %%mm0 \n\t"
" psrlq $32, %%mm7 \n\t"
" paddd %%mm0, %%mm7 \n\t"
" movd %%mm7, %1 \n\t"
: "=m" (XSum),
"=m" (XXSum),
"+r" (SrcData),
"+r" (RefDataPtr1),
"+r" (RefDataPtr2)
: "m" (SrcStride),
"m" (RefStride)
: "edi", "memory"
);
/* Compute and return population variance as mis-match metric. */
return (( (XXSum<<6) - XSum*XSum ));
}
static void restore_fpu (void)
{
__asm__ __volatile__ (
" emms \n\t"
);
}
void dsp_mmx_init(DspFunctions *funcs)
{
funcs->restore_fpu = restore_fpu;
funcs->sub8x8 = sub8x8__mmx;
funcs->sub8x8_128 = sub8x8_128__mmx;
funcs->sub8x8avg2 = sub8x8avg2__mmx;
funcs->row_sad8 = row_sad8__mmx;
funcs->col_sad8x8 = col_sad8x8__mmx;
funcs->sad8x8 = sad8x8__mmx;
funcs->sad8x8_thres = sad8x8_thres__mmx;
funcs->sad8x8_xy2_thres = sad8x8_xy2_thres__mmx;
funcs->intra8x8_err = intra8x8_err__mmx;
funcs->inter8x8_err = inter8x8_err__mmx;
funcs->inter8x8_err_xy2 = inter8x8_err_xy2__mmx;
}
#endif /* USE_ASM */

347
Engine/lib/libtheora/lib/enc/x86_32/dsp_mmxext.c Normal file
View file

@ -0,0 +1,347 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: dsp_mmxext.c 15153 2008-08-04 18:37:55Z tterribe $
********************************************************************/
#include <stdlib.h>
#include "../codec_internal.h"
#include "../dsp.h"
#if defined(USE_ASM)
#define SAD_MMXEXT_LOOP \
" movq (%1), %%mm0 \n\t" /* take 8 bytes */ \
" movq (%2), %%mm1 \n\t" \
" psadbw %%mm1, %%mm0 \n\t" \
" add %3, %1 \n\t" /* Inc pointer into the new data */ \
" paddw %%mm0, %%mm7 \n\t" /* accumulate difference... */ \
" add %4, %2 \n\t" /* Inc pointer into ref data */
static ogg_uint32_t sad8x8__mmxext (unsigned char *ptr1, ogg_uint32_t stride1,
unsigned char *ptr2, ogg_uint32_t stride2)
{
ogg_uint32_t DiffVal;
__asm__ __volatile__ (
" .p2align 4 \n\t"
" pxor %%mm7, %%mm7 \n\t" /* mm7 contains the result */
SAD_MMXEXT_LOOP
SAD_MMXEXT_LOOP
SAD_MMXEXT_LOOP
SAD_MMXEXT_LOOP
SAD_MMXEXT_LOOP
SAD_MMXEXT_LOOP
SAD_MMXEXT_LOOP
" movq (%1), %%mm0 \n\t" /* take 8 bytes */
" movq (%2), %%mm1 \n\t"
" psadbw %%mm1, %%mm0 \n\t"
" paddw %%mm0, %%mm7 \n\t" /* accumulate difference... */
" movd %%mm7, %0 \n\t"
: "=r" (DiffVal),
"+r" (ptr1),
"+r" (ptr2)
: "r" (stride1),
"r" (stride2)
: "memory"
);
return DiffVal;
}
#define SAD_TRES_LOOP \
" movq (%1), %%mm0 \n\t" /* take 8 bytes */ \
" movq (%2), %%mm1 \n\t" \
" psadbw %%mm1, %%mm0 \n\t" \
" add %3, %1 \n\t" /* Inc pointer into the new data */ \
" paddw %%mm0, %%mm7 \n\t" /* accumulate difference... */ \
" add %4, %2 \n\t" /* Inc pointer into ref data */
static ogg_uint32_t sad8x8_thres__mmxext (unsigned char *ptr1, ogg_uint32_t stride1,
unsigned char *ptr2, ogg_uint32_t stride2,
ogg_uint32_t thres)
{
ogg_uint32_t DiffVal;
__asm__ __volatile__ (
" .p2align 4 \n\t"
" pxor %%mm7, %%mm7 \n\t" /* mm7 contains the result */
SAD_TRES_LOOP
SAD_TRES_LOOP
SAD_TRES_LOOP
SAD_TRES_LOOP
SAD_TRES_LOOP
SAD_TRES_LOOP
SAD_TRES_LOOP
SAD_TRES_LOOP
" movd %%mm7, %0 \n\t"
: "=r" (DiffVal),
"+r" (ptr1),
"+r" (ptr2)
: "r" (stride1),
"r" (stride2)
: "memory"
);
return DiffVal;
}
#define SAD_XY2_TRES \
" movq (%1), %%mm0 \n\t" /* take 8 bytes */ \
" movq (%2), %%mm1 \n\t" \
" movq (%3), %%mm2 \n\t" \
" pavgb %%mm2, %%mm1 \n\t" \
" psadbw %%mm1, %%mm0 \n\t" \
\
" add %4, %1 \n\t" /* Inc pointer into the new data */ \
" paddw %%mm0, %%mm7 \n\t" /* accumulate difference... */ \
" add %5, %2 \n\t" /* Inc pointer into ref data */ \
" add %5, %3 \n\t" /* Inc pointer into ref data */
static ogg_uint32_t sad8x8_xy2_thres__mmxext (unsigned char *SrcData, ogg_uint32_t SrcStride,
unsigned char *RefDataPtr1,
unsigned char *RefDataPtr2, ogg_uint32_t RefStride,
ogg_uint32_t thres)
{
ogg_uint32_t DiffVal;
__asm__ __volatile__ (
" .p2align 4 \n\t"
" pxor %%mm7, %%mm7 \n\t" /* mm7 contains the result */
SAD_XY2_TRES
SAD_XY2_TRES
SAD_XY2_TRES
SAD_XY2_TRES
SAD_XY2_TRES
SAD_XY2_TRES
SAD_XY2_TRES
SAD_XY2_TRES
" movd %%mm7, %0 \n\t"
: "=m" (DiffVal),
"+r" (SrcData),
"+r" (RefDataPtr1),
"+r" (RefDataPtr2)
: "m" (SrcStride),
"m" (RefStride)
: "memory"
);
return DiffVal;
}
static ogg_uint32_t row_sad8__mmxext (unsigned char *Src1, unsigned char *Src2)
{
ogg_uint32_t MaxSad;
__asm__ __volatile__ (
" .p2align 4 \n\t"
" movd (%1), %%mm0 \n\t"
" movd (%2), %%mm1 \n\t"
" psadbw %%mm0, %%mm1 \n\t"
" movd 4(%1), %%mm2 \n\t"
" movd 4(%2), %%mm3 \n\t"
" psadbw %%mm2, %%mm3 \n\t"
" pmaxsw %%mm1, %%mm3 \n\t"
" movd %%mm3, %0 \n\t"
" andl $0xffff, %0 \n\t"
: "=m" (MaxSad),
"+r" (Src1),
"+r" (Src2)
:
: "memory"
);
return MaxSad;
}
static ogg_uint32_t col_sad8x8__mmxext (unsigned char *Src1, unsigned char *Src2,
ogg_uint32_t stride)
{
ogg_uint32_t MaxSad;
__asm__ __volatile__ (
" .p2align 4 \n\t"
" pxor %%mm3, %%mm3 \n\t" /* zero out mm3 for unpack */
" pxor %%mm4, %%mm4 \n\t" /* mm4 low sum */
" pxor %%mm5, %%mm5 \n\t" /* mm5 high sum */
" pxor %%mm6, %%mm6 \n\t" /* mm6 low sum */
" pxor %%mm7, %%mm7 \n\t" /* mm7 high sum */
" mov $4, %%edi \n\t" /* 4 rows */
"1: \n\t"
" movq (%1), %%mm0 \n\t" /* take 8 bytes */
" movq (%2), %%mm1 \n\t" /* take 8 bytes */
" movq %%mm0, %%mm2 \n\t"
" psubusb %%mm1, %%mm0 \n\t" /* A - B */
" psubusb %%mm2, %%mm1 \n\t" /* B - A */
" por %%mm1, %%mm0 \n\t" /* and or gives abs difference */
" movq %%mm0, %%mm1 \n\t"
" punpcklbw %%mm3, %%mm0 \n\t" /* unpack to higher precision for accumulation */
" paddw %%mm0, %%mm4 \n\t" /* accumulate difference... */
" punpckhbw %%mm3, %%mm1 \n\t" /* unpack high four bytes to higher precision */
" paddw %%mm1, %%mm5 \n\t" /* accumulate difference... */
" add %3, %1 \n\t" /* Inc pointer into the new data */
" add %3, %2 \n\t" /* Inc pointer into the new data */
" dec %%edi \n\t"
" jnz 1b \n\t"
" mov $4, %%edi \n\t" /* 4 rows */
"2: \n\t"
" movq (%1), %%mm0 \n\t" /* take 8 bytes */
" movq (%2), %%mm1 \n\t" /* take 8 bytes */
" movq %%mm0, %%mm2 \n\t"
" psubusb %%mm1, %%mm0 \n\t" /* A - B */
" psubusb %%mm2, %%mm1 \n\t" /* B - A */
" por %%mm1, %%mm0 \n\t" /* and or gives abs difference */
" movq %%mm0, %%mm1 \n\t"
" punpcklbw %%mm3, %%mm0 \n\t" /* unpack to higher precision for accumulation */
" paddw %%mm0, %%mm6 \n\t" /* accumulate difference... */
" punpckhbw %%mm3, %%mm1 \n\t" /* unpack high four bytes to higher precision */
" paddw %%mm1, %%mm7 \n\t" /* accumulate difference... */
" add %3, %1 \n\t" /* Inc pointer into the new data */
" add %3, %2 \n\t" /* Inc pointer into the new data */
" dec %%edi \n\t"
" jnz 2b \n\t"
" pmaxsw %%mm6, %%mm7 \n\t"
" pmaxsw %%mm4, %%mm5 \n\t"
" pmaxsw %%mm5, %%mm7 \n\t"
" movq %%mm7, %%mm6 \n\t"
" psrlq $32, %%mm6 \n\t"
" pmaxsw %%mm6, %%mm7 \n\t"
" movq %%mm7, %%mm6 \n\t"
" psrlq $16, %%mm6 \n\t"
" pmaxsw %%mm6, %%mm7 \n\t"
" movd %%mm7, %0 \n\t"
" andl $0xffff, %0 \n\t"
: "=r" (MaxSad),
"+r" (Src1),
"+r" (Src2)
: "r" (stride)
: "memory", "edi"
);
return MaxSad;
}
static ogg_uint32_t inter8x8_err_xy2__mmxext (unsigned char *SrcData, ogg_uint32_t SrcStride,
unsigned char *RefDataPtr1,
unsigned char *RefDataPtr2, ogg_uint32_t RefStride)
{
ogg_uint32_t XSum;
ogg_uint32_t XXSum;
__asm__ __volatile__ (
" .p2align 4 \n\t"
" pxor %%mm4, %%mm4 \n\t"
" pxor %%mm5, %%mm5 \n\t"
" pxor %%mm6, %%mm6 \n\t"
" pxor %%mm7, %%mm7 \n\t"
" mov $8, %%edi \n\t"
"1: \n\t"
" movq (%2), %%mm0 \n\t" /* take 8 bytes */
" movq (%3), %%mm2 \n\t"
" movq (%4), %%mm1 \n\t" /* take average of mm2 and mm1 */
" pavgb %%mm2, %%mm1 \n\t"
" movq %%mm0, %%mm2 \n\t"
" movq %%mm1, %%mm3 \n\t"
" punpcklbw %%mm6, %%mm0 \n\t"
" punpcklbw %%mm4, %%mm1 \n\t"
" punpckhbw %%mm6, %%mm2 \n\t"
" punpckhbw %%mm4, %%mm3 \n\t"
" psubsw %%mm1, %%mm0 \n\t"
" psubsw %%mm3, %%mm2 \n\t"
" paddw %%mm0, %%mm5 \n\t"
" paddw %%mm2, %%mm5 \n\t"
" pmaddwd %%mm0, %%mm0 \n\t"
" pmaddwd %%mm2, %%mm2 \n\t"
" paddd %%mm0, %%mm7 \n\t"
" paddd %%mm2, %%mm7 \n\t"
" add %5, %2 \n\t" /* Inc pointer into src data */
" add %6, %3 \n\t" /* Inc pointer into ref data */
" add %6, %4 \n\t" /* Inc pointer into ref data */
" dec %%edi \n\t"
" jnz 1b \n\t"
" movq %%mm5, %%mm0 \n\t"
" psrlq $32, %%mm5 \n\t"
" paddw %%mm0, %%mm5 \n\t"
" movq %%mm5, %%mm0 \n\t"
" psrlq $16, %%mm5 \n\t"
" paddw %%mm0, %%mm5 \n\t"
" movd %%mm5, %%edi \n\t"
" movsx %%di, %%edi \n\t"
" movl %%edi, %0 \n\t"
" movq %%mm7, %%mm0 \n\t"
" psrlq $32, %%mm7 \n\t"
" paddd %%mm0, %%mm7 \n\t"
" movd %%mm7, %1 \n\t"
: "=m" (XSum),
"=m" (XXSum),
"+r" (SrcData),
"+r" (RefDataPtr1),
"+r" (RefDataPtr2)
: "m" (SrcStride),
"m" (RefStride)
: "edi", "memory"
);
/* Compute and return population variance as mis-match metric. */
return (( (XXSum<<6) - XSum*XSum ));
}
void dsp_mmxext_init(DspFunctions *funcs)
{
funcs->row_sad8 = row_sad8__mmxext;
funcs->col_sad8x8 = col_sad8x8__mmxext;
funcs->sad8x8 = sad8x8__mmxext;
funcs->sad8x8_thres = sad8x8_thres__mmxext;
funcs->sad8x8_xy2_thres = sad8x8_xy2_thres__mmxext;
funcs->inter8x8_err_xy2 = inter8x8_err_xy2__mmxext;
}
#endif /* USE_ASM */

339
Engine/lib/libtheora/lib/enc/x86_32/fdct_mmx.c Normal file
View file

@ -0,0 +1,339 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: fdct_mmx.c 15153 2008-08-04 18:37:55Z tterribe $
********************************************************************/
/* mmx fdct implementation */
#include "theora/theora.h"
#include "../codec_internal.h"
#include "../dsp.h"
#if defined(USE_ASM)
static const __attribute__ ((aligned(8),used)) ogg_int64_t xC1S7 = 0x0fb15fb15fb15fb15LL;
static const __attribute__ ((aligned(8),used)) ogg_int64_t xC2S6 = 0x0ec83ec83ec83ec83LL;
static const __attribute__ ((aligned(8),used)) ogg_int64_t xC3S5 = 0x0d4dbd4dbd4dbd4dbLL;
static const __attribute__ ((aligned(8),used)) ogg_int64_t xC4S4 = 0x0b505b505b505b505LL;
static const __attribute__ ((aligned(8),used)) ogg_int64_t xC5S3 = 0x08e3a8e3a8e3a8e3aLL;
static const __attribute__ ((aligned(8),used)) ogg_int64_t xC6S2 = 0x061f861f861f861f8LL;
static const __attribute__ ((aligned(8),used)) ogg_int64_t xC7S1 = 0x031f131f131f131f1LL;
/* execute stage 1 of forward DCT */
#define Fdct_mmx(ip0,ip1,ip2,ip3,ip4,ip5,ip6,ip7,temp) \
" movq " #ip0 ", %%mm0 \n\t" \
" movq " #ip1 ", %%mm1 \n\t" \
" movq " #ip3 ", %%mm2 \n\t" \
" movq " #ip5 ", %%mm3 \n\t" \
" movq %%mm0, %%mm4 \n\t" \
" movq %%mm1, %%mm5 \n\t" \
" movq %%mm2, %%mm6 \n\t" \
" movq %%mm3, %%mm7 \n\t" \
\
" paddsw " #ip7 ", %%mm0 \n\t" /* mm0 = ip0 + ip7 = is07 */ \
" paddsw " #ip2 ", %%mm1 \n\t" /* mm1 = ip1 + ip2 = is12 */ \
" paddsw " #ip4 ", %%mm2 \n\t" /* mm2 = ip3 + ip4 = is34 */ \
" paddsw " #ip6 ", %%mm3 \n\t" /* mm3 = ip5 + ip6 = is56 */ \
" psubsw " #ip7 ", %%mm4 \n\t" /* mm4 = ip0 - ip7 = id07 */ \
" psubsw " #ip2 ", %%mm5 \n\t" /* mm5 = ip1 - ip2 = id12 */ \
\
" psubsw %%mm2, %%mm0 \n\t" /* mm0 = is07 - is34 */ \
\
" paddsw %%mm2, %%mm2 \n\t" \
\
" psubsw " #ip4 ", %%mm6 \n\t" /* mm6 = ip3 - ip4 = id34 */ \
\
" paddsw %%mm0, %%mm2 \n\t" /* mm2 = is07 + is34 = is0734 */ \
" psubsw %%mm3, %%mm1 \n\t" /* mm1 = is12 - is56 */ \
" movq %%mm0," #temp " \n\t" /* Save is07 - is34 to free mm0; */ \
" paddsw %%mm3, %%mm3 \n\t" \
" paddsw %%mm1, %%mm3 \n\t" /* mm3 = is12 + 1s56 = is1256 */ \
\
" psubsw " #ip6 ", %%mm7 \n\t" /* mm7 = ip5 - ip6 = id56 */ \
/* ------------------------------------------------------------------- */ \
" psubsw %%mm7, %%mm5 \n\t" /* mm5 = id12 - id56 */ \
" paddsw %%mm7, %%mm7 \n\t" \
" paddsw %%mm5, %%mm7 \n\t" /* mm7 = id12 + id56 */ \
/* ------------------------------------------------------------------- */ \
" psubsw %%mm3, %%mm2 \n\t" /* mm2 = is0734 - is1256 */ \
" paddsw %%mm3, %%mm3 \n\t" \
\
" movq %%mm2, %%mm0 \n\t" /* make a copy */ \
" paddsw %%mm2, %%mm3 \n\t" /* mm3 = is0734 + is1256 */ \
\
" pmulhw %[xC4S4], %%mm0 \n\t" /* mm0 = xC4S4 * ( is0734 - is1256 ) - ( is0734 - is1256 ) */ \
" paddw %%mm2, %%mm0 \n\t" /* mm0 = xC4S4 * ( is0734 - is1256 ) */ \
" psrlw $15, %%mm2 \n\t" \
" paddw %%mm2, %%mm0 \n\t" /* Truncate mm0, now it is op[4] */ \
\
" movq %%mm3, %%mm2 \n\t" \
" movq %%mm0," #ip4 " \n\t" /* save ip4, now mm0,mm2 are free */ \
\
" movq %%mm3, %%mm0 \n\t" \
" pmulhw %[xC4S4], %%mm3 \n\t" /* mm3 = xC4S4 * ( is0734 +is1256 ) - ( is0734 +is1256 ) */ \
\
" psrlw $15, %%mm2 \n\t" \
" paddw %%mm0, %%mm3 \n\t" /* mm3 = xC4S4 * ( is0734 +is1256 ) */ \
" paddw %%mm2, %%mm3 \n\t" /* Truncate mm3, now it is op[0] */ \
\
" movq %%mm3," #ip0 " \n\t" \
/* ------------------------------------------------------------------- */ \
" movq " #temp ", %%mm3 \n\t" /* mm3 = irot_input_y */ \
" pmulhw %[xC2S6], %%mm3 \n\t" /* mm3 = xC2S6 * irot_input_y - irot_input_y */ \
\
" movq " #temp ", %%mm2 \n\t" \
" movq %%mm2, %%mm0 \n\t" \
\
" psrlw $15, %%mm2 \n\t" /* mm3 = xC2S6 * irot_input_y */ \
" paddw %%mm0, %%mm3 \n\t" \
\
" paddw %%mm2, %%mm3 \n\t" /* Truncated */ \
" movq %%mm5, %%mm0 \n\t" \
\
" movq %%mm5, %%mm2 \n\t" \
" pmulhw %[xC6S2], %%mm0 \n\t" /* mm0 = xC6S2 * irot_input_x */ \
\
" psrlw $15, %%mm2 \n\t" \
" paddw %%mm2, %%mm0 \n\t" /* Truncated */ \
\
" paddsw %%mm0, %%mm3 \n\t" /* ip[2] */ \
" movq %%mm3," #ip2 " \n\t" /* Save ip2 */ \
\
" movq %%mm5, %%mm0 \n\t" \
" movq %%mm5, %%mm2 \n\t" \
\
" pmulhw %[xC2S6], %%mm5 \n\t" /* mm5 = xC2S6 * irot_input_x - irot_input_x */ \
" psrlw $15, %%mm2 \n\t" \
\
" movq " #temp ", %%mm3 \n\t" \
" paddw %%mm0, %%mm5 \n\t" /* mm5 = xC2S6 * irot_input_x */ \
\
" paddw %%mm2, %%mm5 \n\t" /* Truncated */ \
" movq %%mm3, %%mm2 \n\t" \
\
" pmulhw %[xC6S2], %%mm3 \n\t" /* mm3 = xC6S2 * irot_input_y */ \
" psrlw $15, %%mm2 \n\t" \
\
" paddw %%mm2, %%mm3 \n\t" /* Truncated */ \
" psubsw %%mm5, %%mm3 \n\t" \
\
" movq %%mm3," #ip6 " \n\t" \
/* ------------------------------------------------------------------- */ \
" movq %[xC4S4], %%mm0 \n\t" \
" movq %%mm1, %%mm2 \n\t" \
" movq %%mm1, %%mm3 \n\t" \
\
" pmulhw %%mm0, %%mm1 \n\t" /* mm0 = xC4S4 * ( is12 - is56 ) - ( is12 - is56 ) */ \
" psrlw $15, %%mm2 \n\t" \
\
" paddw %%mm3, %%mm1 \n\t" /* mm0 = xC4S4 * ( is12 - is56 ) */ \
" paddw %%mm2, %%mm1 \n\t" /* Truncate mm1, now it is icommon_product1 */ \
\
" movq %%mm7, %%mm2 \n\t" \
" movq %%mm7, %%mm3 \n\t" \
\
" pmulhw %%mm0, %%mm7 \n\t" /* mm7 = xC4S4 * ( id12 + id56 ) - ( id12 + id56 ) */ \
" psrlw $15, %%mm2 \n\t" \
\
" paddw %%mm3, %%mm7 \n\t" /* mm7 = xC4S4 * ( id12 + id56 ) */ \
" paddw %%mm2, %%mm7 \n\t" /* Truncate mm7, now it is icommon_product2 */ \
/* ------------------------------------------------------------------- */ \
" pxor %%mm0, %%mm0 \n\t" /* Clear mm0 */ \
" psubsw %%mm6, %%mm0 \n\t" /* mm0 = - id34 */ \
\
" psubsw %%mm7, %%mm0 \n\t" /* mm0 = - ( id34 + idcommon_product2 ) */ \
" paddsw %%mm6, %%mm6 \n\t" \
" paddsw %%mm0, %%mm6 \n\t" /* mm6 = id34 - icommon_product2 */ \
\
" psubsw %%mm1, %%mm4 \n\t" /* mm4 = id07 - icommon_product1 */ \
" paddsw %%mm1, %%mm1 \n\t" \
" paddsw %%mm4, %%mm1 \n\t" /* mm1 = id07 + icommon_product1 */ \
/* ------------------------------------------------------------------- */ \
" movq %[xC1S7], %%mm7 \n\t" \
" movq %%mm1, %%mm2 \n\t" \
\
" movq %%mm1, %%mm3 \n\t" \
" pmulhw %%mm7, %%mm1 \n\t" /* mm1 = xC1S7 * irot_input_x - irot_input_x */ \
\
" movq %[xC7S1], %%mm7 \n\t" \
" psrlw $15, %%mm2 \n\t" \
\
" paddw %%mm3, %%mm1 \n\t" /* mm1 = xC1S7 * irot_input_x */ \
" paddw %%mm2, %%mm1 \n\t" /* Trucated */ \
\
" pmulhw %%mm7, %%mm3 \n\t" /* mm3 = xC7S1 * irot_input_x */ \
" paddw %%mm2, %%mm3 \n\t" /* Truncated */ \
\
" movq %%mm0, %%mm5 \n\t" \
" movq %%mm0, %%mm2 \n\t" \
\
" movq %[xC1S7], %%mm7 \n\t" \
" pmulhw %%mm7, %%mm0 \n\t" /* mm0 = xC1S7 * irot_input_y - irot_input_y */ \
\
" movq %[xC7S1], %%mm7 \n\t" \
" psrlw $15, %%mm2 \n\t" \
\
" paddw %%mm5, %%mm0 \n\t" /* mm0 = xC1S7 * irot_input_y */ \
" paddw %%mm2, %%mm0 \n\t" /* Truncated */ \
\
" pmulhw %%mm7, %%mm5 \n\t" /* mm5 = xC7S1 * irot_input_y */ \
" paddw %%mm2, %%mm5 \n\t" /* Truncated */ \
\
" psubsw %%mm5, %%mm1 \n\t" /* mm1 = xC1S7 * irot_input_x - xC7S1 * irot_input_y = ip1 */ \
" paddsw %%mm0, %%mm3 \n\t" /* mm3 = xC7S1 * irot_input_x - xC1S7 * irot_input_y = ip7 */ \
\
" movq %%mm1," #ip1 " \n\t" \
" movq %%mm3," #ip7 " \n\t" \
/* ------------------------------------------------------------------- */ \
" movq %[xC3S5], %%mm0 \n\t" \
" movq %[xC5S3], %%mm1 \n\t" \
\
" movq %%mm6, %%mm5 \n\t" \
" movq %%mm6, %%mm7 \n\t" \
\
" movq %%mm4, %%mm2 \n\t" \
" movq %%mm4, %%mm3 \n\t" \
\
" pmulhw %%mm0, %%mm4 \n\t" /* mm4 = xC3S5 * irot_input_x - irot_input_x */ \
" pmulhw %%mm1, %%mm6 \n\t" /* mm6 = xC5S3 * irot_input_y - irot_input_y */ \
\
" psrlw $15, %%mm2 \n\t" \
" psrlw $15, %%mm5 \n\t" \
\
" paddw %%mm3, %%mm4 \n\t" /* mm4 = xC3S5 * irot_input_x */ \
" paddw %%mm7, %%mm6 \n\t" /* mm6 = xC5S3 * irot_input_y */ \
\
" paddw %%mm2, %%mm4 \n\t" /* Truncated */ \
" paddw %%mm5, %%mm6 \n\t" /* Truncated */ \
\
" psubsw %%mm6, %%mm4 \n\t" /* ip3 */ \
" movq %%mm4," #ip3 " \n\t" \
\
" movq %%mm3, %%mm4 \n\t" \
" movq %%mm7, %%mm6 \n\t" \
\
" pmulhw %%mm1, %%mm3 \n\t" /* mm3 = xC5S3 * irot_input_x - irot_input_x */ \
" pmulhw %%mm0, %%mm7 \n\t" /* mm7 = xC3S5 * irot_input_y - irot_input_y */ \
\
" paddw %%mm2, %%mm4 \n\t" \
" paddw %%mm5, %%mm6 \n\t" \
\
" paddw %%mm4, %%mm3 \n\t" /* mm3 = xC5S3 * irot_input_x */ \
" paddw %%mm6, %%mm7 \n\t" /* mm7 = xC3S5 * irot_input_y */ \
\
" paddw %%mm7, %%mm3 \n\t" /* ip5 */ \
" movq %%mm3," #ip5 " \n\t"
#define Transpose_mmx(ip0,ip1,ip2,ip3,ip4,ip5,ip6,ip7, \
op0,op1,op2,op3,op4,op5,op6,op7) \
" movq " #ip0 ", %%mm0 \n\t" /* mm0 = a0 a1 a2 a3 */ \
" movq " #ip4 ", %%mm4 \n\t" /* mm4 = e4 e5 e6 e7 */ \
" movq " #ip1 ", %%mm1 \n\t" /* mm1 = b0 b1 b2 b3 */ \
" movq " #ip5 ", %%mm5 \n\t" /* mm5 = f4 f5 f6 f7 */ \
" movq " #ip2 ", %%mm2 \n\t" /* mm2 = c0 c1 c2 c3 */ \
" movq " #ip6 ", %%mm6 \n\t" /* mm6 = g4 g5 g6 g7 */ \
" movq " #ip3 ", %%mm3 \n\t" /* mm3 = d0 d1 d2 d3 */ \
" movq %%mm1," #op1 " \n\t" /* save b0 b1 b2 b3 */ \
" movq " #ip7 ", %%mm7 \n\t" /* mm7 = h0 h1 h2 h3 */ \
/* Transpose 2x8 block */ \
" movq %%mm4, %%mm1 \n\t" /* mm1 = e3 e2 e1 e0 */ \
" punpcklwd %%mm5, %%mm4 \n\t" /* mm4 = f1 e1 f0 e0 */ \
" movq %%mm0," #op0 " \n\t" /* save a3 a2 a1 a0 */ \
" punpckhwd %%mm5, %%mm1 \n\t" /* mm1 = f3 e3 f2 e2 */ \
" movq %%mm6, %%mm0 \n\t" /* mm0 = g3 g2 g1 g0 */ \
" punpcklwd %%mm7, %%mm6 \n\t" /* mm6 = h1 g1 h0 g0 */ \
" movq %%mm4, %%mm5 \n\t" /* mm5 = f1 e1 f0 e0 */ \
" punpckldq %%mm6, %%mm4 \n\t" /* mm4 = h0 g0 f0 e0 = MM4 */ \
" punpckhdq %%mm6, %%mm5 \n\t" /* mm5 = h1 g1 f1 e1 = MM5 */ \
" movq %%mm1, %%mm6 \n\t" /* mm6 = f3 e3 f2 e2 */ \
" movq %%mm4," #op4 " \n\t" \
" punpckhwd %%mm7, %%mm0 \n\t" /* mm0 = h3 g3 h2 g2 */ \
" movq %%mm5," #op5 " \n\t" \
" punpckhdq %%mm0, %%mm6 \n\t" /* mm6 = h3 g3 f3 e3 = MM7 */ \
" movq " #op0 ", %%mm4 \n\t" /* mm4 = a3 a2 a1 a0 */ \
" punpckldq %%mm0, %%mm1 \n\t" /* mm1 = h2 g2 f2 e2 = MM6 */ \
" movq " #op1 ", %%mm5 \n\t" /* mm5 = b3 b2 b1 b0 */ \
" movq %%mm4, %%mm0 \n\t" /* mm0 = a3 a2 a1 a0 */ \
" movq %%mm6," #op7 " \n\t" \
" punpcklwd %%mm5, %%mm0 \n\t" /* mm0 = b1 a1 b0 a0 */ \
" movq %%mm1," #op6 " \n\t" \
" punpckhwd %%mm5, %%mm4 \n\t" /* mm4 = b3 a3 b2 a2 */ \
" movq %%mm2, %%mm5 \n\t" /* mm5 = c3 c2 c1 c0 */ \
" punpcklwd %%mm3, %%mm2 \n\t" /* mm2 = d1 c1 d0 c0 */ \
" movq %%mm0, %%mm1 \n\t" /* mm1 = b1 a1 b0 a0 */ \
" punpckldq %%mm2, %%mm0 \n\t" /* mm0 = d0 c0 b0 a0 = MM0 */ \
" punpckhdq %%mm2, %%mm1 \n\t" /* mm1 = d1 c1 b1 a1 = MM1 */ \
" movq %%mm4, %%mm2 \n\t" /* mm2 = b3 a3 b2 a2 */ \
" movq %%mm0," #op0 " \n\t" \
" punpckhwd %%mm3, %%mm5 \n\t" /* mm5 = d3 c3 d2 c2 */ \
" movq %%mm1," #op1 " \n\t" \
" punpckhdq %%mm5, %%mm4 \n\t" /* mm4 = d3 c3 b3 a3 = MM3 */ \
" punpckldq %%mm5, %%mm2 \n\t" /* mm2 = d2 c2 b2 a2 = MM2 */ \
" movq %%mm4," #op3 " \n\t" \
" movq %%mm2," #op2 " \n\t"
/* This performs a 2D Forward DCT on an 8x8 block with short
coefficients. We try to do the truncation to match the C
version. */
static void fdct_short__mmx ( ogg_int16_t *InputData, ogg_int16_t *OutputData)
{
ogg_int16_t __attribute__((aligned(8))) temp[8*8];
__asm__ __volatile__ (
" .p2align 4 \n\t"
/*
* Input data is an 8x8 block. To make processing of the data more efficent
* we will transpose the block of data to two 4x8 blocks???
*/
Transpose_mmx ( (%0), 16(%0), 32(%0), 48(%0), 8(%0), 24(%0), 40(%0), 56(%0),
(%1), 16(%1), 32(%1), 48(%1), 8(%1), 24(%1), 40(%1), 56(%1))
Fdct_mmx ( (%1), 16(%1), 32(%1), 48(%1), 8(%1), 24(%1), 40(%1), 56(%1), (%2))
Transpose_mmx (64(%0), 80(%0), 96(%0),112(%0), 72(%0), 88(%0),104(%0),120(%0),
64(%1), 80(%1), 96(%1),112(%1), 72(%1), 88(%1),104(%1),120(%1))
Fdct_mmx (64(%1), 80(%1), 96(%1),112(%1), 72(%1), 88(%1),104(%1),120(%1), (%2))
Transpose_mmx ( 0(%1), 16(%1), 32(%1), 48(%1), 64(%1), 80(%1), 96(%1),112(%1),
0(%1), 16(%1), 32(%1), 48(%1), 64(%1), 80(%1), 96(%1),112(%1))
Fdct_mmx ( 0(%1), 16(%1), 32(%1), 48(%1), 64(%1), 80(%1), 96(%1),112(%1), (%2))
Transpose_mmx ( 8(%1), 24(%1), 40(%1), 56(%1), 72(%1), 88(%1),104(%1),120(%1),
8(%1), 24(%1), 40(%1), 56(%1), 72(%1), 88(%1),104(%1),120(%1))
Fdct_mmx ( 8(%1), 24(%1), 40(%1), 56(%1), 72(%1), 88(%1),104(%1),120(%1), (%2))
" emms \n\t"
: "+r" (InputData),
"+r" (OutputData)
: "r" (temp),
[xC1S7] "m" (xC1S7), /* gcc 3.1+ allows named asm parameters */
[xC2S6] "m" (xC2S6),
[xC3S5] "m" (xC3S5),
[xC4S4] "m" (xC4S4),
[xC5S3] "m" (xC5S3),
[xC6S2] "m" (xC6S2),
[xC7S1] "m" (xC7S1)
: "memory"
);
}
/* install our implementation in the function table */
void dsp_mmx_fdct_init(DspFunctions *funcs)
{
funcs->fdct_short = fdct_short__mmx;
}
#endif /* USE_ASM */

1452
Engine/lib/libtheora/lib/enc/x86_32/idct_mmx.c Normal file
View file

File diff suppressed because it is too large Load diff

182
Engine/lib/libtheora/lib/enc/x86_32/recon_mmx.c Normal file
View file

@ -0,0 +1,182 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: recon_mmx.c 15153 2008-08-04 18:37:55Z tterribe $
********************************************************************/
#include "../codec_internal.h"
#if defined(USE_ASM)
static const __attribute__ ((aligned(8),used)) ogg_int64_t V128 = 0x8080808080808080LL;
static void copy8x8__mmx (unsigned char *src,
unsigned char *dest,
unsigned int stride)
{
__asm__ __volatile__ (
" .p2align 4 \n\t"
" lea (%2, %2, 2), %%edi \n\t"
" movq (%1), %%mm0 \n\t"
" movq (%1, %2), %%mm1 \n\t"
" movq (%1, %2, 2), %%mm2 \n\t"
" movq (%1, %%edi), %%mm3 \n\t"
" lea (%1, %2, 4), %1 \n\t"
" movq %%mm0, (%0) \n\t"
" movq %%mm1, (%0, %2) \n\t"
" movq %%mm2, (%0, %2, 2) \n\t"
" movq %%mm3, (%0, %%edi) \n\t"
" lea (%0, %2, 4), %0 \n\t"
" movq (%1), %%mm0 \n\t"
" movq (%1, %2), %%mm1 \n\t"
" movq (%1, %2, 2), %%mm2 \n\t"
" movq (%1, %%edi), %%mm3 \n\t"
" movq %%mm0, (%0) \n\t"
" movq %%mm1, (%0, %2) \n\t"
" movq %%mm2, (%0, %2, 2) \n\t"
" movq %%mm3, (%0, %%edi) \n\t"
: "+a" (dest)
: "c" (src),
"d" (stride)
: "memory", "edi"
);
}
static void recon_intra8x8__mmx (unsigned char *ReconPtr, ogg_int16_t *ChangePtr,
ogg_uint32_t LineStep)
{
__asm__ __volatile__ (
" .p2align 4 \n\t"
" movq %[V128], %%mm0 \n\t" /* Set mm0 to 0x8080808080808080 */
" lea 128(%1), %%edi \n\t" /* Endpoint in input buffer */
"1: \n\t"
" movq (%1), %%mm2 \n\t" /* First four input values */
" packsswb 8(%1), %%mm2 \n\t" /* pack with next(high) four values */
" por %%mm0, %%mm0 \n\t"
" pxor %%mm0, %%mm2 \n\t" /* Convert result to unsigned (same as add 128) */
" lea 16(%1), %1 \n\t" /* Step source buffer */
" cmp %%edi, %1 \n\t" /* are we done */
" movq %%mm2, (%0) \n\t" /* store results */
" lea (%0, %2), %0 \n\t" /* Step output buffer */
" jc 1b \n\t" /* Loop back if we are not done */
: "+r" (ReconPtr)
: "r" (ChangePtr),
"r" (LineStep),
[V128] "m" (V128)
: "memory", "edi"
);
}
static void recon_inter8x8__mmx (unsigned char *ReconPtr, unsigned char *RefPtr,
ogg_int16_t *ChangePtr, ogg_uint32_t LineStep)
{
__asm__ __volatile__ (
" .p2align 4 \n\t"
" pxor %%mm0, %%mm0 \n\t"
" lea 128(%1), %%edi \n\t"
"1: \n\t"
" movq (%2), %%mm2 \n\t" /* (+3 misaligned) 8 reference pixels */
" movq (%1), %%mm4 \n\t" /* first 4 changes */
" movq %%mm2, %%mm3 \n\t"
" movq 8(%1), %%mm5 \n\t" /* last 4 changes */
" punpcklbw %%mm0, %%mm2 \n\t" /* turn first 4 refs into positive 16-bit #s */
" paddsw %%mm4, %%mm2 \n\t" /* add in first 4 changes */
" punpckhbw %%mm0, %%mm3 \n\t" /* turn last 4 refs into positive 16-bit #s */
" paddsw %%mm5, %%mm3 \n\t" /* add in last 4 changes */
" add %3, %2 \n\t" /* next row of reference pixels */
" packuswb %%mm3, %%mm2 \n\t" /* pack result to unsigned 8-bit values */
" lea 16(%1), %1 \n\t" /* next row of changes */
" cmp %%edi, %1 \n\t" /* are we done? */
" movq %%mm2, (%0) \n\t" /* store result */
" lea (%0, %3), %0 \n\t" /* next row of output */
" jc 1b \n\t"
: "+r" (ReconPtr)
: "r" (ChangePtr),
"r" (RefPtr),
"r" (LineStep)
: "memory", "edi"
);
}
static void recon_inter8x8_half__mmx (unsigned char *ReconPtr, unsigned char *RefPtr1,
unsigned char *RefPtr2, ogg_int16_t *ChangePtr,
ogg_uint32_t LineStep)
{
__asm__ __volatile__ (
" .p2align 4 \n\t"
" pxor %%mm0, %%mm0 \n\t"
" lea 128(%1), %%edi \n\t"
"1: \n\t"
" movq (%2), %%mm2 \n\t" /* (+3 misaligned) 8 reference pixels */
" movq (%3), %%mm4 \n\t" /* (+3 misaligned) 8 reference pixels */
" movq %%mm2, %%mm3 \n\t"
" punpcklbw %%mm0, %%mm2 \n\t" /* mm2 = start ref1 as positive 16-bit #s */
" movq %%mm4, %%mm5 \n\t"
" movq (%1), %%mm6 \n\t" /* first 4 changes */
" punpckhbw %%mm0, %%mm3 \n\t" /* mm3 = end ref1 as positive 16-bit #s */
" movq 8(%1), %%mm7 \n\t" /* last 4 changes */
" punpcklbw %%mm0, %%mm4 \n\t" /* mm4 = start ref2 as positive 16-bit #s */
" punpckhbw %%mm0, %%mm5 \n\t" /* mm5 = end ref2 as positive 16-bit #s */
" paddw %%mm4, %%mm2 \n\t" /* mm2 = start (ref1 + ref2) */
" paddw %%mm5, %%mm3 \n\t" /* mm3 = end (ref1 + ref2) */
" psrlw $1, %%mm2 \n\t" /* mm2 = start (ref1 + ref2)/2 */
" psrlw $1, %%mm3 \n\t" /* mm3 = end (ref1 + ref2)/2 */
" paddw %%mm6, %%mm2 \n\t" /* add changes to start */
" paddw %%mm7, %%mm3 \n\t" /* add changes to end */
" lea 16(%1), %1 \n\t" /* next row of changes */
" packuswb %%mm3, %%mm2 \n\t" /* pack start|end to unsigned 8-bit */
" add %4, %2 \n\t" /* next row of reference pixels */
" add %4, %3 \n\t" /* next row of reference pixels */
" movq %%mm2, (%0) \n\t" /* store result */
" add %4, %0 \n\t" /* next row of output */
" cmp %%edi, %1 \n\t" /* are we done? */
" jc 1b \n\t"
: "+r" (ReconPtr)
: "r" (ChangePtr),
"r" (RefPtr1),
"r" (RefPtr2),
"m" (LineStep)
: "memory", "edi"
);
}
void dsp_mmx_recon_init(DspFunctions *funcs)
{
funcs->copy8x8 = copy8x8__mmx;
funcs->recon_intra8x8 = recon_intra8x8__mmx;
funcs->recon_inter8x8 = recon_inter8x8__mmx;
funcs->recon_inter8x8_half = recon_inter8x8_half__mmx;
}
#endif /* USE_ASM */

1605
Engine/lib/libtheora/lib/enc/x86_32_vs/dsp_mmx.c Normal file
View file

File diff suppressed because it is too large Load diff

333
Engine/lib/libtheora/lib/enc/x86_32_vs/fdct_mmx.c Normal file
View file

@ -0,0 +1,333 @@
;//==========================================================================
;//
;// THIS CODE AND INFORMATION IS PROVIDED "AS IS" WITHOUT WARRANTY OF ANY
;// KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE
;// IMPLIED WARRANTIES OF MERCHANTABILITY AND/OR FITNESS FOR A PARTICULAR
;// PURPOSE.
;//
;// Copyright (c) 1999 - 2001 On2 Technologies Inc. All Rights Reserved.
;//
;//--------------------------------------------------------------------------
#include "theora/theora.h"
#include "../codec_internal.h"
#include "../dsp.h"
static const ogg_int64_t xC1S7 = 0x0fb15fb15fb15fb15;
static const ogg_int64_t xC2S6 = 0x0ec83ec83ec83ec83;
static const ogg_int64_t xC3S5 = 0x0d4dbd4dbd4dbd4db;
static const ogg_int64_t xC4S4 = 0x0b505b505b505b505;
static const ogg_int64_t xC5S3 = 0x08e3a8e3a8e3a8e3a;
static const ogg_int64_t xC6S2 = 0x061f861f861f861f8;
static const ogg_int64_t xC7S1 = 0x031f131f131f131f1;
static __inline void Transpose_mmx( ogg_int16_t *InputData1, ogg_int16_t *OutputData1,
ogg_int16_t *InputData2, ogg_int16_t *OutputData2)
{
__asm {
align 16
mov eax, InputData1
mov ebx, InputData2
mov ecx, OutputData1
mov edx, OutputData2
movq mm0, [eax] ; /* mm0 = a0 a1 a2 a3 */
movq mm4, [ebx] ; /* mm4 = e4 e5 e6 e7 */
movq mm1, [16 + eax] ; /* mm1 = b0 b1 b2 b3 */
movq mm5, [16 + ebx] ; /* mm5 = f4 f5 f6 f7 */
movq mm2, [32 + eax] ; /* mm2 = c0 c1 c2 c3 */
movq mm6, [32 + ebx] ; /* mm6 = g4 g5 g6 g7 */
movq mm3, [48 + eax] ; /* mm3 = d0 d1 d2 d3 */
movq [16 + ecx], mm1 ; /* save b0 b1 b2 b3 */
movq mm7, [48 + ebx] ; /* mm7 = h0 h1 h2 h3 */
; /* Transpose 2x8 block */
movq mm1, mm4 ; /* mm1 = e3 e2 e1 e0 */
punpcklwd mm4, mm5 ; /* mm4 = f1 e1 f0 e0 */
movq [ecx], mm0 ; /* save a3 a2 a1 a0 */
punpckhwd mm1, mm5 ; /* mm1 = f3 e3 f2 e2 */
movq mm0, mm6 ; /* mm0 = g3 g2 g1 g0 */
punpcklwd mm6, mm7 ; /* mm6 = h1 g1 h0 g0 */
movq mm5, mm4 ; /* mm5 = f1 e1 f0 e0 */
punpckldq mm4, mm6 ; /* mm4 = h0 g0 f0 e0 = MM4 */
punpckhdq mm5, mm6 ; /* mm5 = h1 g1 f1 e1 = MM5 */
movq mm6, mm1 ; /* mm6 = f3 e3 f2 e2 */
movq [edx], mm4 ;
punpckhwd mm0, mm7 ; /* mm0 = h3 g3 h2 g2 */
movq [16 + edx], mm5 ;
punpckhdq mm6, mm0 ; /* mm6 = h3 g3 f3 e3 = MM7 */
movq mm4, [ecx] ; /* mm4 = a3 a2 a1 a0 */
punpckldq mm1, mm0 ; /* mm1 = h2 g2 f2 e2 = MM6 */
movq mm5, [16 + ecx] ; /* mm5 = b3 b2 b1 b0 */
movq mm0, mm4 ; /* mm0 = a3 a2 a1 a0 */
movq [48 + edx], mm6 ;
punpcklwd mm0, mm5 ; /* mm0 = b1 a1 b0 a0 */
movq [32 + edx], mm1 ;
punpckhwd mm4, mm5 ; /* mm4 = b3 a3 b2 a2 */
movq mm5, mm2 ; /* mm5 = c3 c2 c1 c0 */
punpcklwd mm2, mm3 ; /* mm2 = d1 c1 d0 c0 */
movq mm1, mm0 ; /* mm1 = b1 a1 b0 a0 */
punpckldq mm0, mm2 ; /* mm0 = d0 c0 b0 a0 = MM0 */
punpckhdq mm1, mm2 ; /* mm1 = d1 c1 b1 a1 = MM1 */
movq mm2, mm4 ; /* mm2 = b3 a3 b2 a2 */
movq [ecx], mm0 ;
punpckhwd mm5, mm3 ; /* mm5 = d3 c3 d2 c2 */
movq [16 + ecx], mm1 ;
punpckhdq mm4, mm5 ; /* mm4 = d3 c3 b3 a3 = MM3 */
punpckldq mm2, mm5 ; /* mm2 = d2 c2 b2 a2 = MM2 */
movq [48 + ecx], mm4 ;
movq [32 + ecx], mm2 ;
};
}
static __inline void Fdct_mmx( ogg_int16_t *InputData1, ogg_int16_t *InputData2, ogg_int16_t *temp)
{
__asm {
align 16
mov eax, InputData1
mov ebx, InputData2
mov ecx, temp
movq mm0, [eax] ;
movq mm1, [16 + eax] ;
movq mm2, [48 + eax] ;
movq mm3, [16 + ebx] ;
movq mm4, mm0 ;
movq mm5, mm1 ;
movq mm6, mm2 ;
movq mm7, mm3 ;
;
paddsw mm0, [48 + ebx] ; /* mm0 = ip0 + ip7 = is07 */
paddsw mm1, [32 + eax] ; /* mm1 = ip1 + ip2 = is12 */
paddsw mm2, [ebx] ; /* mm2 = ip3 + ip4 = is34 */
paddsw mm3, [32 + ebx] ; /* mm3 = ip5 + ip6 = is56 */
psubsw mm4, [48 + ebx] ; /* mm4 = ip0 - ip7 = id07 */
psubsw mm5, [32 + eax] ; /* mm5 = ip1 - ip2 = id12 */
;
psubsw mm0, mm2 ; /* mm0 = is07 - is34 */
;
paddsw mm2, mm2 ;
;
psubsw mm6, [ebx] ; /* mm6 = ip3 - ip4 = id34 */
;
paddsw mm2, mm0 ; /* mm2 = is07 + is34 = is0734 */
psubsw mm1, mm3 ; /* mm1 = is12 - is56 */
movq [ecx], mm0 ; /* Save is07 - is34 to free mm0; */
paddsw mm3, mm3 ;
paddsw mm3, mm1 ; /* mm3 = is12 + 1s56 = is1256 */
;
psubsw mm7, [32 + ebx] ; /* mm7 = ip5 - ip6 = id56 */
; /* ------------------------------------------------------------------- */
psubsw mm5, mm7 ; /* mm5 = id12 - id56 */
paddsw mm7, mm7 ;
paddsw mm7, mm5 ; /* mm7 = id12 + id56 */
; /* ------------------------------------------------------------------- */
psubsw mm2, mm3 ; /* mm2 = is0734 - is1256 */
paddsw mm3, mm3 ;
;
movq mm0, mm2 ; /* make a copy */
paddsw mm3, mm2 ; /* mm3 = is0734 + is1256 */
;
pmulhw mm0, xC4S4 ; /* mm0 = xC4S4 * ( is0734 - is1256 ) - ( is0734 - is1256 ) */
paddw mm0, mm2 ; /* mm0 = xC4S4 * ( is0734 - is1256 ) */
psrlw mm2, 15 ;
paddw mm0, mm2 ; /* Truncate mm0, now it is op[4] */
;
movq mm2, mm3 ;
movq [ebx], mm0 ; /* save ip4, now mm0,mm2 are free */
;
movq mm0, mm3 ;
pmulhw mm3, xC4S4 ; /* mm3 = xC4S4 * ( is0734 +is1256 ) - ( is0734 +is1256 ) */
;
psrlw mm2, 15 ;
paddw mm3, mm0 ; /* mm3 = xC4S4 * ( is0734 +is1256 ) */
paddw mm3, mm2 ; /* Truncate mm3, now it is op[0] */
;
movq [eax], mm3 ;
; /* ------------------------------------------------------------------- */
movq mm3, [ecx] ; /* mm3 = irot_input_y */
pmulhw mm3, xC2S6 ; /* mm3 = xC2S6 * irot_input_y - irot_input_y */
;
movq mm2, [ecx] ;
movq mm0, mm2 ;
;
psrlw mm2, 15 ; /* mm3 = xC2S6 * irot_input_y */
paddw mm3, mm0 ;
;
paddw mm3, mm2 ; /* Truncated */
movq mm0, mm5 ;
;
movq mm2, mm5 ;
pmulhw mm0, xC6S2 ; /* mm0 = xC6S2 * irot_input_x */
;
psrlw mm2, 15 ;
paddw mm0, mm2 ; /* Truncated */
;
paddsw mm3, mm0 ; /* ip[2] */
movq [32 + eax], mm3 ; /* Save ip2 */
;
movq mm0, mm5 ;
movq mm2, mm5 ;
;
pmulhw mm5, xC2S6 ; /* mm5 = xC2S6 * irot_input_x - irot_input_x */
psrlw mm2, 15 ;
;
movq mm3, [ecx] ;
paddw mm5, mm0 ; /* mm5 = xC2S6 * irot_input_x */
;
paddw mm5, mm2 ; /* Truncated */
movq mm2, mm3 ;
;
pmulhw mm3, xC6S2 ; /* mm3 = xC6S2 * irot_input_y */
psrlw mm2, 15 ;
;
paddw mm3, mm2 ; /* Truncated */
psubsw mm3, mm5 ;
;
movq [32 + ebx], mm3 ;
; /* ------------------------------------------------------------------- */
movq mm0, xC4S4 ;
movq mm2, mm1 ;
movq mm3, mm1 ;
;
pmulhw mm1, mm0 ; /* mm0 = xC4S4 * ( is12 - is56 ) - ( is12 - is56 ) */
psrlw mm2, 15 ;
;
paddw mm1, mm3 ; /* mm0 = xC4S4 * ( is12 - is56 ) */
paddw mm1, mm2 ; /* Truncate mm1, now it is icommon_product1 */
;
movq mm2, mm7 ;
movq mm3, mm7 ;
;
pmulhw mm7, mm0 ; /* mm7 = xC4S4 * ( id12 + id56 ) - ( id12 + id56 ) */
psrlw mm2, 15 ;
;
paddw mm7, mm3 ; /* mm7 = xC4S4 * ( id12 + id56 ) */
paddw mm7, mm2 ; /* Truncate mm7, now it is icommon_product2 */
; /* ------------------------------------------------------------------- */
pxor mm0, mm0 ; /* Clear mm0 */
psubsw mm0, mm6 ; /* mm0 = - id34 */
;
psubsw mm0, mm7 ; /* mm0 = - ( id34 + idcommon_product2 ) */
paddsw mm6, mm6 ;
paddsw mm6, mm0 ; /* mm6 = id34 - icommon_product2 */
;
psubsw mm4, mm1 ; /* mm4 = id07 - icommon_product1 */
paddsw mm1, mm1 ;
paddsw mm1, mm4 ; /* mm1 = id07 + icommon_product1 */
; /* ------------------------------------------------------------------- */
movq mm7, xC1S7 ;
movq mm2, mm1 ;
;
movq mm3, mm1 ;
pmulhw mm1, mm7 ; /* mm1 = xC1S7 * irot_input_x - irot_input_x */
;
movq mm7, xC7S1 ;
psrlw mm2, 15 ;
;
paddw mm1, mm3 ; /* mm1 = xC1S7 * irot_input_x */
paddw mm1, mm2 ; /* Trucated */
;
pmulhw mm3, mm7 ; /* mm3 = xC7S1 * irot_input_x */
paddw mm3, mm2 ; /* Truncated */
;
movq mm5, mm0 ;
movq mm2, mm0 ;
;
movq mm7, xC1S7 ;
pmulhw mm0, mm7 ; /* mm0 = xC1S7 * irot_input_y - irot_input_y */
;
movq mm7, xC7S1 ;
psrlw mm2, 15 ;
;
paddw mm0, mm5 ; /* mm0 = xC1S7 * irot_input_y */
paddw mm0, mm2 ; /* Truncated */
;
pmulhw mm5, mm7 ; /* mm5 = xC7S1 * irot_input_y */
paddw mm5, mm2 ; /* Truncated */
;
psubsw mm1, mm5 ; /* mm1 = xC1S7 * irot_input_x - xC7S1 * irot_input_y = ip1 */
paddsw mm3, mm0 ; /* mm3 = xC7S1 * irot_input_x - xC1S7 * irot_input_y = ip7 */
;
movq [16 + eax], mm1 ;
movq [48 + ebx], mm3 ;
; /* ------------------------------------------------------------------- */
movq mm0, xC3S5 ;
movq mm1, xC5S3 ;
;
movq mm5, mm6 ;
movq mm7, mm6 ;
;
movq mm2, mm4 ;
movq mm3, mm4 ;
;
pmulhw mm4, mm0 ; /* mm4 = xC3S5 * irot_input_x - irot_input_x */
pmulhw mm6, mm1 ; /* mm6 = xC5S3 * irot_input_y - irot_input_y */
;
psrlw mm2, 15 ;
psrlw mm5, 15 ;
;
paddw mm4, mm3 ; /* mm4 = xC3S5 * irot_input_x */
paddw mm6, mm7 ; /* mm6 = xC5S3 * irot_input_y */
;
paddw mm4, mm2 ; /* Truncated */
paddw mm6, mm5 ; /* Truncated */
;
psubsw mm4, mm6 ; /* ip3 */
movq [48 + eax], mm4 ;
;
movq mm4, mm3 ;
movq mm6, mm7 ;
;
pmulhw mm3, mm1 ; /* mm3 = xC5S3 * irot_input_x - irot_input_x */
pmulhw mm7, mm0 ; /* mm7 = xC3S5 * irot_input_y - irot_input_y */
;
paddw mm4, mm2 ;
paddw mm6, mm5 ;
;
paddw mm3, mm4 ; /* mm3 = xC5S3 * irot_input_x */
paddw mm7, mm6 ; /* mm7 = xC3S5 * irot_input_y */
;
paddw mm3, mm7 ; /* ip5 */
movq [16 + ebx], mm3 ;
};
}
static void fdct_short__mmx ( ogg_int16_t *InputData, ogg_int16_t *OutputData)
{
static ogg_int16_t tmp[32];
ogg_int16_t* align_tmp = (ogg_int16_t*)((unsigned char*)tmp + (16 - ((int)tmp)&15));
Transpose_mmx(InputData, OutputData, InputData + 4, OutputData + 4);
Fdct_mmx(OutputData, OutputData + 4, align_tmp);
Transpose_mmx(InputData + 32, OutputData + 32, InputData + 36, OutputData + 36);
Fdct_mmx(OutputData+32, OutputData + 36, align_tmp);
Transpose_mmx(OutputData, OutputData, OutputData + 32, OutputData + 32);
Fdct_mmx(OutputData, OutputData + 32, align_tmp);
Transpose_mmx(OutputData + 4, OutputData + 4, OutputData + 36, OutputData + 36);
Fdct_mmx(OutputData + 4, OutputData + 36, align_tmp);
__asm emms
}
void dsp_mmx_fdct_init(DspFunctions *funcs)
{
funcs->fdct_short = fdct_short__mmx;
}

197
Engine/lib/libtheora/lib/enc/x86_32_vs/recon_mmx.c Normal file
View file

@ -0,0 +1,197 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: reconstruct.c,v 1.6 2003/12/03 08:59:41 arc Exp $
********************************************************************/
#include "../codec_internal.h"
static const unsigned __int64 V128 = 0x8080808080808080;
static void copy8x8__mmx (unsigned char *src,
unsigned char *dest,
unsigned int stride)
{
//Is this even the fastest way to do this?
__asm {
align 16
mov eax, src
mov ebx, dest
mov ecx, stride
lea edi, [ecx + ecx * 2]
movq mm0, [eax]
movq mm1, [eax + ecx]
movq mm2, [eax + ecx * 2]
movq mm3, [eax + edi]
lea eax, [eax + ecx * 4]
movq [ebx], mm0
movq [ebx + ecx], mm1
movq [ebx + ecx * 2], mm2
movq [ebx + edi], mm3
lea ebx, [ebx + ecx * 4]
movq mm0, [eax]
movq mm1, [eax + ecx]
movq mm2, [eax + ecx * 2]
movq mm3, [eax + edi]
movq [ebx], mm0
movq [ebx + ecx], mm1
movq [ebx + ecx * 2], mm2
movq [ebx + edi], mm3
};
}
static void recon_intra8x8__mmx (unsigned char *ReconPtr, ogg_int16_t *ChangePtr,
ogg_uint32_t LineStep)
{
__asm {
align 16
mov eax, ReconPtr
mov ebx, ChangePtr
mov ecx, LineStep
movq mm0, V128
lea edi, [128 + ebx]
loop_start:
movq mm2, [ebx]
packsswb mm2, [8 + ebx]
por mm0, mm0
pxor mm2, mm0
lea ebx, [16 + ebx]
cmp ebx, edi
movq [eax], mm2
lea eax, [eax + ecx]
jc loop_start
};
}
static void recon_inter8x8__mmx (unsigned char *ReconPtr, unsigned char *RefPtr,
ogg_int16_t *ChangePtr, ogg_uint32_t LineStep)
{
__asm {
align 16
mov eax, ReconPtr
mov ebx, ChangePtr
mov ecx, LineStep
mov edx, RefPtr
pxor mm0, mm0
lea edi, [128 + ebx]
loop_start:
movq mm2, [edx]
movq mm4, [ebx]
movq mm3, mm2
movq mm5, [8 + ebx]
punpcklbw mm2, mm0
paddsw mm2, mm4
punpckhbw mm3, mm0
paddsw mm3, mm5
add edx, ecx
packuswb mm2, mm3
lea ebx, [16 + ebx]
cmp ebx, edi
movq [eax], mm2
lea eax, [eax + ecx]
jc loop_start
};
}
static void recon_inter8x8_half__mmx (unsigned char *ReconPtr, unsigned char *RefPtr1,
unsigned char *RefPtr2, ogg_int16_t *ChangePtr,
ogg_uint32_t LineStep)
{
__asm {
align 16
mov eax, ReconPtr
mov ebx, ChangePtr
mov ecx, RefPtr1
mov edx, RefPtr2
pxor mm0, mm0
lea edi, [128 + ebx]
loop_start:
movq mm2, [ecx]
movq mm4, [edx]
movq mm3, mm2
punpcklbw mm2, mm0
movq mm5, mm4
movq mm6, [ebx]
punpckhbw mm3, mm0
movq mm7, [8 + ebx]
punpcklbw mm4, mm0
punpckhbw mm5, mm0
paddw mm2, mm4
paddw mm3, mm5
psrlw mm2, 1
psrlw mm3, 1
paddw mm2, mm6
paddw mm3, mm7
lea ebx, [16 + ebx]
packuswb mm2, mm3
add ecx, LineStep
add edx, LineStep
movq [eax], mm2
add eax, LineStep
cmp ebx, edi
jc loop_start
};
}
void dsp_mmx_recon_init(DspFunctions *funcs)
{
funcs->copy8x8 = copy8x8__mmx;
funcs->recon_intra8x8 = recon_intra8x8__mmx;
funcs->recon_inter8x8 = recon_inter8x8__mmx;
funcs->recon_inter8x8_half = recon_inter8x8_half__mmx;
}

409
Engine/lib/libtheora/lib/enc/x86_64/dct_decode_mmx.c Normal file
View file

@ -0,0 +1,409 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2008 *
* by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: dct_decode_mmx.c 15400 2008-10-15 12:10:58Z tterribe $
********************************************************************/
#include <stdlib.h>
#include "../codec_internal.h"
#if defined(USE_ASM)
static const __attribute__((aligned(8),used)) ogg_int64_t OC_V3=
0x0003000300030003LL;
static const __attribute__((aligned(8),used)) ogg_int64_t OC_V4=
0x0004000400040004LL;
static void loop_filter_v(unsigned char *_pix,int _ystride,
const ogg_int16_t *_ll){
long esi;
_pix-=_ystride*2;
__asm__ __volatile__(
/*mm0=0*/
"pxor %%mm0,%%mm0\n\t"
/*esi=_ystride*3*/
"lea (%[ystride],%[ystride],2),%[s]\n\t"
/*mm7=_pix[0...8]*/
"movq (%[pix]),%%mm7\n\t"
/*mm4=_pix[0...8+_ystride*3]*/
"movq (%[pix],%[s]),%%mm4\n\t"
/*mm6=_pix[0...8]*/
"movq %%mm7,%%mm6\n\t"
/*Expand unsigned _pix[0...3] to 16 bits.*/
"punpcklbw %%mm0,%%mm6\n\t"
"movq %%mm4,%%mm5\n\t"
/*Expand unsigned _pix[4...8] to 16 bits.*/
"punpckhbw %%mm0,%%mm7\n\t"
/*Expand other arrays too.*/
"punpcklbw %%mm0,%%mm4\n\t"
"punpckhbw %%mm0,%%mm5\n\t"
/*mm7:mm6=_p[0...8]-_p[0...8+_ystride*3]:*/
"psubw %%mm4,%%mm6\n\t"
"psubw %%mm5,%%mm7\n\t"
/*mm5=mm4=_pix[0...8+_ystride]*/
"movq (%[pix],%[ystride]),%%mm4\n\t"
/*mm1=mm3=mm2=_pix[0..8]+_ystride*2]*/
"movq (%[pix],%[ystride],2),%%mm2\n\t"
"movq %%mm4,%%mm5\n\t"
"movq %%mm2,%%mm3\n\t"
"movq %%mm2,%%mm1\n\t"
/*Expand these arrays.*/
"punpckhbw %%mm0,%%mm5\n\t"
"punpcklbw %%mm0,%%mm4\n\t"
"punpckhbw %%mm0,%%mm3\n\t"
"punpcklbw %%mm0,%%mm2\n\t"
/*Preload...*/
"movq %[OC_V3],%%mm0\n\t"
/*mm3:mm2=_pix[0...8+_ystride*2]-_pix[0...8+_ystride]*/
"psubw %%mm5,%%mm3\n\t"
"psubw %%mm4,%%mm2\n\t"
/*Scale by 3.*/
"pmullw %%mm0,%%mm3\n\t"
"pmullw %%mm0,%%mm2\n\t"
/*Preload...*/
"movq %[OC_V4],%%mm0\n\t"
/*f=mm3:mm2==_pix[0...8]-_pix[0...8+_ystride*3]+
3*(_pix[0...8+_ystride*2]-_pix[0...8+_ystride])*/
"paddw %%mm7,%%mm3\n\t"
"paddw %%mm6,%%mm2\n\t"
/*Add 4.*/
"paddw %%mm0,%%mm3\n\t"
"paddw %%mm0,%%mm2\n\t"
/*"Divide" by 8.*/
"psraw $3,%%mm3\n\t"
"psraw $3,%%mm2\n\t"
/*Now compute lflim of mm3:mm2 cf. Section 7.10 of the sepc.*/
/*Free up mm5.*/
"packuswb %%mm5,%%mm4\n\t"
/*mm0=L L L L*/
"movq (%[ll]),%%mm0\n\t"
/*if(R_i<-2L||R_i>2L)R_i=0:*/
"movq %%mm2,%%mm5\n\t"
"pxor %%mm6,%%mm6\n\t"
"movq %%mm0,%%mm7\n\t"
"psubw %%mm0,%%mm6\n\t"
"psllw $1,%%mm7\n\t"
"psllw $1,%%mm6\n\t"
/*mm2==R_3 R_2 R_1 R_0*/
/*mm5==R_3 R_2 R_1 R_0*/
/*mm6==-2L -2L -2L -2L*/
/*mm7==2L 2L 2L 2L*/
"pcmpgtw %%mm2,%%mm7\n\t"
"pcmpgtw %%mm6,%%mm5\n\t"
"pand %%mm7,%%mm2\n\t"
"movq %%mm0,%%mm7\n\t"
"pand %%mm5,%%mm2\n\t"
"psllw $1,%%mm7\n\t"
"movq %%mm3,%%mm5\n\t"
/*mm3==R_7 R_6 R_5 R_4*/
/*mm5==R_7 R_6 R_5 R_4*/
/*mm6==-2L -2L -2L -2L*/
/*mm7==2L 2L 2L 2L*/
"pcmpgtw %%mm3,%%mm7\n\t"
"pcmpgtw %%mm6,%%mm5\n\t"
"pand %%mm7,%%mm3\n\t"
"movq %%mm0,%%mm7\n\t"
"pand %%mm5,%%mm3\n\t"
/*if(R_i<-L)R_i'=R_i+2L;
if(R_i>L)R_i'=R_i-2L;
if(R_i<-L||R_i>L)R_i=-R_i':*/
"psraw $1,%%mm6\n\t"
"movq %%mm2,%%mm5\n\t"
"psllw $1,%%mm7\n\t"
/*mm2==R_3 R_2 R_1 R_0*/
/*mm5==R_3 R_2 R_1 R_0*/
/*mm6==-L -L -L -L*/
/*mm0==L L L L*/
/*mm5=R_i>L?FF:00*/
"pcmpgtw %%mm0,%%mm5\n\t"
/*mm6=-L>R_i?FF:00*/
"pcmpgtw %%mm2,%%mm6\n\t"
/*mm7=R_i>L?2L:0*/
"pand %%mm5,%%mm7\n\t"
/*mm2=R_i>L?R_i-2L:R_i*/
"psubw %%mm7,%%mm2\n\t"
"movq %%mm0,%%mm7\n\t"
/*mm5=-L>R_i||R_i>L*/
"por %%mm6,%%mm5\n\t"
"psllw $1,%%mm7\n\t"
/*mm7=-L>R_i?2L:0*/
"pand %%mm6,%%mm7\n\t"
"pxor %%mm6,%%mm6\n\t"
/*mm2=-L>R_i?R_i+2L:R_i*/
"paddw %%mm7,%%mm2\n\t"
"psubw %%mm0,%%mm6\n\t"
/*mm5=-L>R_i||R_i>L?-R_i':0*/
"pand %%mm2,%%mm5\n\t"
"movq %%mm0,%%mm7\n\t"
/*mm2=-L>R_i||R_i>L?0:R_i*/
"psubw %%mm5,%%mm2\n\t"
"psllw $1,%%mm7\n\t"
/*mm2=-L>R_i||R_i>L?-R_i':R_i*/
"psubw %%mm5,%%mm2\n\t"
"movq %%mm3,%%mm5\n\t"
/*mm3==R_7 R_6 R_5 R_4*/
/*mm5==R_7 R_6 R_5 R_4*/
/*mm6==-L -L -L -L*/
/*mm0==L L L L*/
/*mm6=-L>R_i?FF:00*/
"pcmpgtw %%mm3,%%mm6\n\t"
/*mm5=R_i>L?FF:00*/
"pcmpgtw %%mm0,%%mm5\n\t"
/*mm7=R_i>L?2L:0*/
"pand %%mm5,%%mm7\n\t"
/*mm2=R_i>L?R_i-2L:R_i*/
"psubw %%mm7,%%mm3\n\t"
"psllw $1,%%mm0\n\t"
/*mm5=-L>R_i||R_i>L*/
"por %%mm6,%%mm5\n\t"
/*mm0=-L>R_i?2L:0*/
"pand %%mm6,%%mm0\n\t"
/*mm3=-L>R_i?R_i+2L:R_i*/
"paddw %%mm0,%%mm3\n\t"
/*mm5=-L>R_i||R_i>L?-R_i':0*/
"pand %%mm3,%%mm5\n\t"
/*mm2=-L>R_i||R_i>L?0:R_i*/
"psubw %%mm5,%%mm3\n\t"
/*mm2=-L>R_i||R_i>L?-R_i':R_i*/
"psubw %%mm5,%%mm3\n\t"
/*Unfortunately, there's no unsigned byte+signed byte with unsigned
saturation op code, so we have to promote things back 16 bits.*/
"pxor %%mm0,%%mm0\n\t"
"movq %%mm4,%%mm5\n\t"
"punpcklbw %%mm0,%%mm4\n\t"
"punpckhbw %%mm0,%%mm5\n\t"
"movq %%mm1,%%mm6\n\t"
"punpcklbw %%mm0,%%mm1\n\t"
"punpckhbw %%mm0,%%mm6\n\t"
/*_pix[0...8+_ystride]+=R_i*/
"paddw %%mm2,%%mm4\n\t"
"paddw %%mm3,%%mm5\n\t"
/*_pix[0...8+_ystride*2]-=R_i*/
"psubw %%mm2,%%mm1\n\t"
"psubw %%mm3,%%mm6\n\t"
"packuswb %%mm5,%%mm4\n\t"
"packuswb %%mm6,%%mm1\n\t"
/*Write it back out.*/
"movq %%mm4,(%[pix],%[ystride])\n\t"
"movq %%mm1,(%[pix],%[ystride],2)\n\t"
:[s]"=&S"(esi)
:[pix]"r"(_pix),[ystride]"r"((long)_ystride),[ll]"r"(_ll),
[OC_V3]"m"(OC_V3),[OC_V4]"m"(OC_V4)
:"memory"
);
}
/*This code implements the bulk of loop_filter_h().
Data are striped p0 p1 p2 p3 ... p0 p1 p2 p3 ..., so in order to load all
four p0's to one register we must transpose the values in four mmx regs.
When half is done we repeat this for the rest.*/
static void loop_filter_h4(unsigned char *_pix,long _ystride,
const ogg_int16_t *_ll){
long esi;
long edi;
__asm__ __volatile__(
/*x x x x 3 2 1 0*/
"movd (%[pix]),%%mm0\n\t"
/*esi=_ystride*3*/
"lea (%[ystride],%[ystride],2),%[s]\n\t"
/*x x x x 7 6 5 4*/
"movd (%[pix],%[ystride]),%%mm1\n\t"
/*x x x x B A 9 8*/
"movd (%[pix],%[ystride],2),%%mm2\n\t"
/*x x x x F E D C*/
"movd (%[pix],%[s]),%%mm3\n\t"
/*mm0=7 3 6 2 5 1 4 0*/
"punpcklbw %%mm1,%%mm0\n\t"
/*mm2=F B E A D 9 C 8*/
"punpcklbw %%mm3,%%mm2\n\t"
/*mm1=7 3 6 2 5 1 4 0*/
"movq %%mm0,%%mm1\n\t"
/*mm0=F B 7 3 E A 6 2*/
"punpckhwd %%mm2,%%mm0\n\t"
/*mm1=D 9 5 1 C 8 4 0*/
"punpcklwd %%mm2,%%mm1\n\t"
"pxor %%mm7,%%mm7\n\t"
/*mm5=D 9 5 1 C 8 4 0*/
"movq %%mm1,%%mm5\n\t"
/*mm1=x C x 8 x 4 x 0==pix[0]*/
"punpcklbw %%mm7,%%mm1\n\t"
/*mm5=x D x 9 x 5 x 1==pix[1]*/
"punpckhbw %%mm7,%%mm5\n\t"
/*mm3=F B 7 3 E A 6 2*/
"movq %%mm0,%%mm3\n\t"
/*mm0=x E x A x 6 x 2==pix[2]*/
"punpcklbw %%mm7,%%mm0\n\t"
/*mm3=x F x B x 7 x 3==pix[3]*/
"punpckhbw %%mm7,%%mm3\n\t"
/*mm1=mm1-mm3==pix[0]-pix[3]*/
"psubw %%mm3,%%mm1\n\t"
/*Save a copy of pix[2] for later.*/
"movq %%mm0,%%mm4\n\t"
/*mm0=mm0-mm5==pix[2]-pix[1]*/
"psubw %%mm5,%%mm0\n\t"
/*Scale by 3.*/
"pmullw %[OC_V3],%%mm0\n\t"
/*f=mm1==_pix[0]-_pix[3]+ 3*(_pix[2]-_pix[1])*/
"paddw %%mm1,%%mm0\n\t"
/*Add 4.*/
"paddw %[OC_V4],%%mm0\n\t"
/*"Divide" by 8, producing the residuals R_i.*/
"psraw $3,%%mm0\n\t"
/*Now compute lflim of mm0 cf. Section 7.10 of the sepc.*/
/*mm6=L L L L*/
"movq (%[ll]),%%mm6\n\t"
/*if(R_i<-2L||R_i>2L)R_i=0:*/
"movq %%mm0,%%mm1\n\t"
"pxor %%mm2,%%mm2\n\t"
"movq %%mm6,%%mm3\n\t"
"psubw %%mm6,%%mm2\n\t"
"psllw $1,%%mm3\n\t"
"psllw $1,%%mm2\n\t"
/*mm0==R_3 R_2 R_1 R_0*/
/*mm1==R_3 R_2 R_1 R_0*/
/*mm2==-2L -2L -2L -2L*/
/*mm3==2L 2L 2L 2L*/
"pcmpgtw %%mm0,%%mm3\n\t"
"pcmpgtw %%mm2,%%mm1\n\t"
"pand %%mm3,%%mm0\n\t"
"pand %%mm1,%%mm0\n\t"
/*if(R_i<-L)R_i'=R_i+2L;
if(R_i>L)R_i'=R_i-2L;
if(R_i<-L||R_i>L)R_i=-R_i':*/
"psraw $1,%%mm2\n\t"
"movq %%mm0,%%mm1\n\t"
"movq %%mm6,%%mm3\n\t"
/*mm0==R_3 R_2 R_1 R_0*/
/*mm1==R_3 R_2 R_1 R_0*/
/*mm2==-L -L -L -L*/
/*mm6==L L L L*/
/*mm2=-L>R_i?FF:00*/
"pcmpgtw %%mm0,%%mm2\n\t"
/*mm1=R_i>L?FF:00*/
"pcmpgtw %%mm6,%%mm1\n\t"
/*mm3=2L 2L 2L 2L*/
"psllw $1,%%mm3\n\t"
/*mm6=2L 2L 2L 2L*/
"psllw $1,%%mm6\n\t"
/*mm3=R_i>L?2L:0*/
"pand %%mm1,%%mm3\n\t"
/*mm6=-L>R_i?2L:0*/
"pand %%mm2,%%mm6\n\t"
/*mm0=R_i>L?R_i-2L:R_i*/
"psubw %%mm3,%%mm0\n\t"
/*mm1=-L>R_i||R_i>L*/
"por %%mm2,%%mm1\n\t"
/*mm0=-L>R_i?R_i+2L:R_i*/
"paddw %%mm6,%%mm0\n\t"
/*mm1=-L>R_i||R_i>L?R_i':0*/
"pand %%mm0,%%mm1\n\t"
/*mm0=-L>R_i||R_i>L?0:R_i*/
"psubw %%mm1,%%mm0\n\t"
/*mm0=-L>R_i||R_i>L?-R_i':R_i*/
"psubw %%mm1,%%mm0\n\t"
/*_pix[1]+=R_i;*/
"paddw %%mm0,%%mm5\n\t"
/*_pix[2]-=R_i;*/
"psubw %%mm0,%%mm4\n\t"
/*mm5=x x x x D 9 5 1*/
"packuswb %%mm7,%%mm5\n\t"
/*mm4=x x x x E A 6 2*/
"packuswb %%mm7,%%mm4\n\t"
/*mm5=E D A 9 6 5 2 1*/
"punpcklbw %%mm4,%%mm5\n\t"
/*edi=6 5 2 1*/
"movd %%mm5,%%edi\n\t"
"movw %%di,1(%[pix])\n\t"
/*Why is there such a big stall here?*/
"psrlq $32,%%mm5\n\t"
"shrl $16,%%edi\n\t"
"movw %%di,1(%[pix],%[ystride])\n\t"
/*edi=E D A 9*/
"movd %%mm5,%%edi\n\t"
"movw %%di,1(%[pix],%[ystride],2)\n\t"
"shrl $16,%%edi\n\t"
"movw %%di,1(%[pix],%[s])\n\t"
:[s]"=&S"(esi),[d]"=&D"(edi),
[pix]"+r"(_pix),[ystride]"+r"(_ystride),[ll]"+r"(_ll)
:[OC_V3]"m"(OC_V3),[OC_V4]"m"(OC_V4)
:"memory"
);
}
static void loop_filter_h(unsigned char *_pix,int _ystride,
const ogg_int16_t *_ll){
_pix-=2;
loop_filter_h4(_pix,_ystride,_ll);
loop_filter_h4(_pix+(_ystride<<2),_ystride,_ll);
}
static void loop_filter_mmx(PB_INSTANCE *pbi, int FLimit){
int j;
ogg_int16_t __attribute__((aligned(8))) ll[4];
unsigned char *cp = pbi->display_fragments;
ogg_uint32_t *bp = pbi->recon_pixel_index_table;
if ( FLimit == 0 ) return;
ll[0]=ll[1]=ll[2]=ll[3]=FLimit;
for ( j = 0; j < 3 ; j++){
ogg_uint32_t *bp_begin = bp;
ogg_uint32_t *bp_end;
int stride;
int h;
switch(j) {
case 0: /* y */
bp_end = bp + pbi->YPlaneFragments;
h = pbi->HFragments;
stride = pbi->YStride;
break;
default: /* u,v, 4:20 specific */
bp_end = bp + pbi->UVPlaneFragments;
h = pbi->HFragments >> 1;
stride = pbi->UVStride;
break;
}
while(bp<bp_end){
ogg_uint32_t *bp_left = bp;
ogg_uint32_t *bp_right = bp + h;
while(bp<bp_right){
if(cp[0]){
if(bp>bp_left)
loop_filter_h(&pbi->LastFrameRecon[bp[0]],stride,ll);
if(bp_left>bp_begin)
loop_filter_v(&pbi->LastFrameRecon[bp[0]],stride,ll);
if(bp+1<bp_right && !cp[1])
loop_filter_h(&pbi->LastFrameRecon[bp[0]]+8,stride,ll);
if(bp+h<bp_end && !cp[h])
loop_filter_v(&pbi->LastFrameRecon[bp[h]],stride,ll);
}
bp++;
cp++;
}
}
}
__asm__ __volatile__("emms\n\t");
}
/* install our implementation in the function table */
void dsp_mmx_dct_decode_init(DspFunctions *funcs)
{
funcs->LoopFilter = loop_filter_mmx;
}
#endif /* USE_ASM */

303
Engine/lib/libtheora/lib/enc/x86_64/dsp_mmx.c Normal file
View file

@ -0,0 +1,303 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: dsp_mmx.c 15397 2008-10-14 02:06:24Z tterribe $
********************************************************************/
#include <stdlib.h>
#include "../codec_internal.h"
#include "../dsp.h"
#if defined(USE_ASM)
typedef unsigned long long ogg_uint64_t;
static const __attribute__ ((aligned(8),used)) ogg_int64_t V128 = 0x0080008000800080LL;
#define DSP_OP_AVG(a,b) ((((int)(a)) + ((int)(b)))/2)
#define DSP_OP_DIFF(a,b) (((int)(a)) - ((int)(b)))
#define DSP_OP_ABS_DIFF(a,b) abs((((int)(a)) - ((int)(b))))
static void sub8x8__mmx (unsigned char *FiltPtr, unsigned char *ReconPtr,
ogg_int16_t *DctInputPtr, ogg_uint32_t PixelsPerLine,
ogg_uint32_t ReconPixelsPerLine)
{
__asm__ __volatile__ (
" .balign 16 \n\t"
" pxor %%mm7, %%mm7 \n\t"
".rept 8 \n\t"
" movq (%0), %%mm0 \n\t" /* mm0 = FiltPtr */
" movq (%1), %%mm1 \n\t" /* mm1 = ReconPtr */
" movq %%mm0, %%mm2 \n\t" /* dup to prepare for up conversion */
" movq %%mm1, %%mm3 \n\t" /* dup to prepare for up conversion */
/* convert from UINT8 to INT16 */
" punpcklbw %%mm7, %%mm0 \n\t" /* mm0 = INT16(FiltPtr) */
" punpcklbw %%mm7, %%mm1 \n\t" /* mm1 = INT16(ReconPtr) */
" punpckhbw %%mm7, %%mm2 \n\t" /* mm2 = INT16(FiltPtr) */
" punpckhbw %%mm7, %%mm3 \n\t" /* mm3 = INT16(ReconPtr) */
/* start calculation */
" psubw %%mm1, %%mm0 \n\t" /* mm0 = FiltPtr - ReconPtr */
" psubw %%mm3, %%mm2 \n\t" /* mm2 = FiltPtr - ReconPtr */
" movq %%mm0, (%2) \n\t" /* write answer out */
" movq %%mm2, 8(%2) \n\t" /* write answer out */
/* Increment pointers */
" add $16, %2 \n\t"
" add %3, %0 \n\t"
" add %4, %1 \n\t"
".endr \n\t"
: "+r" (FiltPtr),
"+r" (ReconPtr),
"+r" (DctInputPtr)
: "r" ((ogg_uint64_t)PixelsPerLine),
"r" ((ogg_uint64_t)ReconPixelsPerLine)
: "memory"
);
}
static void sub8x8_128__mmx (unsigned char *FiltPtr, ogg_int16_t *DctInputPtr,
ogg_uint32_t PixelsPerLine)
{
ogg_uint64_t ppl = PixelsPerLine;
__asm__ __volatile__ (
" .balign 16 \n\t"
" pxor %%mm7, %%mm7 \n\t"
" movq %[V128], %%mm1 \n\t"
".rept 8 \n\t"
" movq (%0), %%mm0 \n\t" /* mm0 = FiltPtr */
" movq %%mm0, %%mm2 \n\t" /* dup to prepare for up conversion */
/* convert from UINT8 to INT16 */
" punpcklbw %%mm7, %%mm0 \n\t" /* mm0 = INT16(FiltPtr) */
" punpckhbw %%mm7, %%mm2 \n\t" /* mm2 = INT16(FiltPtr) */
/* start calculation */
" psubw %%mm1, %%mm0 \n\t" /* mm0 = FiltPtr - 128 */
" psubw %%mm1, %%mm2 \n\t" /* mm2 = FiltPtr - 128 */
" movq %%mm0, (%1) \n\t" /* write answer out */
" movq %%mm2, 8(%1) \n\t" /* write answer out */
/* Increment pointers */
" add $16, %1 \n\t"
" add %2, %0 \n\t"
".endr \n\t"
: "+r" (FiltPtr),
"+r" (DctInputPtr)
: "r" (ppl), /* gcc bug? a cast won't work here, e.g. (ogg_uint64_t)PixelsPerLine */
[V128] "m" (V128)
: "memory"
);
}
static void sub8x8avg2__mmx (unsigned char *FiltPtr, unsigned char *ReconPtr1,
unsigned char *ReconPtr2, ogg_int16_t *DctInputPtr,
ogg_uint32_t PixelsPerLine,
ogg_uint32_t ReconPixelsPerLine)
{
__asm__ __volatile__ (
" .balign 16 \n\t"
" pxor %%mm7, %%mm7 \n\t"
".rept 8 \n\t"
" movq (%0), %%mm0 \n\t" /* mm0 = FiltPtr */
" movq (%1), %%mm1 \n\t" /* mm1 = ReconPtr1 */
" movq (%2), %%mm4 \n\t" /* mm1 = ReconPtr2 */
" movq %%mm0, %%mm2 \n\t" /* dup to prepare for up conversion */
" movq %%mm1, %%mm3 \n\t" /* dup to prepare for up conversion */
" movq %%mm4, %%mm5 \n\t" /* dup to prepare for up conversion */
/* convert from UINT8 to INT16 */
" punpcklbw %%mm7, %%mm0 \n\t" /* mm0 = INT16(FiltPtr) */
" punpcklbw %%mm7, %%mm1 \n\t" /* mm1 = INT16(ReconPtr1) */
" punpcklbw %%mm7, %%mm4 \n\t" /* mm1 = INT16(ReconPtr2) */
" punpckhbw %%mm7, %%mm2 \n\t" /* mm2 = INT16(FiltPtr) */
" punpckhbw %%mm7, %%mm3 \n\t" /* mm3 = INT16(ReconPtr1) */
" punpckhbw %%mm7, %%mm5 \n\t" /* mm3 = INT16(ReconPtr2) */
/* average ReconPtr1 and ReconPtr2 */
" paddw %%mm4, %%mm1 \n\t" /* mm1 = ReconPtr1 + ReconPtr2 */
" paddw %%mm5, %%mm3 \n\t" /* mm3 = ReconPtr1 + ReconPtr2 */
" psrlw $1, %%mm1 \n\t" /* mm1 = (ReconPtr1 + ReconPtr2) / 2 */
" psrlw $1, %%mm3 \n\t" /* mm3 = (ReconPtr1 + ReconPtr2) / 2 */
" psubw %%mm1, %%mm0 \n\t" /* mm0 = FiltPtr - ((ReconPtr1 + ReconPtr2) / 2) */
" psubw %%mm3, %%mm2 \n\t" /* mm2 = FiltPtr - ((ReconPtr1 + ReconPtr2) / 2) */
" movq %%mm0, (%3) \n\t" /* write answer out */
" movq %%mm2, 8(%3) \n\t" /* write answer out */
/* Increment pointers */
" add $16, %3 \n\t"
" add %4, %0 \n\t"
" add %5, %1 \n\t"
" add %5, %2 \n\t"
".endr \n\t"
: "+r" (FiltPtr),
"+r" (ReconPtr1),
"+r" (ReconPtr2),
"+r" (DctInputPtr)
: "r" ((ogg_uint64_t)PixelsPerLine),
"r" ((ogg_uint64_t)ReconPixelsPerLine)
: "memory"
);
}
static ogg_uint32_t intra8x8_err__mmx (unsigned char *DataPtr, ogg_uint32_t Stride)
{
ogg_uint64_t XSum;
ogg_uint64_t XXSum;
__asm__ __volatile__ (
" .balign 16 \n\t"
" pxor %%mm5, %%mm5 \n\t"
" pxor %%mm6, %%mm6 \n\t"
" pxor %%mm7, %%mm7 \n\t"
" mov $8, %%rdi \n\t"
"1: \n\t"
" movq (%2), %%mm0 \n\t" /* take 8 bytes */
" movq %%mm0, %%mm2 \n\t"
" punpcklbw %%mm6, %%mm0 \n\t"
" punpckhbw %%mm6, %%mm2 \n\t"
" paddw %%mm0, %%mm5 \n\t"
" paddw %%mm2, %%mm5 \n\t"
" pmaddwd %%mm0, %%mm0 \n\t"
" pmaddwd %%mm2, %%mm2 \n\t"
" paddd %%mm0, %%mm7 \n\t"
" paddd %%mm2, %%mm7 \n\t"
" add %3, %2 \n\t" /* Inc pointer into src data */
" dec %%rdi \n\t"
" jnz 1b \n\t"
" movq %%mm5, %%mm0 \n\t"
" psrlq $32, %%mm5 \n\t"
" paddw %%mm0, %%mm5 \n\t"
" movq %%mm5, %%mm0 \n\t"
" psrlq $16, %%mm5 \n\t"
" paddw %%mm0, %%mm5 \n\t"
" movd %%mm5, %%rdi \n\t"
" movsx %%di, %%rdi \n\t"
" mov %%rdi, %0 \n\t"
" movq %%mm7, %%mm0 \n\t"
" psrlq $32, %%mm7 \n\t"
" paddd %%mm0, %%mm7 \n\t"
" movd %%mm7, %1 \n\t"
: "=r" (XSum),
"=r" (XXSum),
"+r" (DataPtr)
: "r" ((ogg_uint64_t)Stride)
: "rdi", "memory"
);
/* Compute population variance as mis-match metric. */
return (( (XXSum<<6) - XSum*XSum ) );
}
static ogg_uint32_t inter8x8_err__mmx (unsigned char *SrcData, ogg_uint32_t SrcStride,
unsigned char *RefDataPtr, ogg_uint32_t RefStride)
{
ogg_uint64_t XSum;
ogg_uint64_t XXSum;
__asm__ __volatile__ (
" .balign 16 \n\t"
" pxor %%mm5, %%mm5 \n\t"
" pxor %%mm6, %%mm6 \n\t"
" pxor %%mm7, %%mm7 \n\t"
" mov $8, %%rdi \n\t"
"1: \n\t"
" movq (%2), %%mm0 \n\t" /* take 8 bytes */
" movq (%3), %%mm1 \n\t"
" movq %%mm0, %%mm2 \n\t"
" movq %%mm1, %%mm3 \n\t"
" punpcklbw %%mm6, %%mm0 \n\t"
" punpcklbw %%mm6, %%mm1 \n\t"
" punpckhbw %%mm6, %%mm2 \n\t"
" punpckhbw %%mm6, %%mm3 \n\t"
" psubsw %%mm1, %%mm0 \n\t"
" psubsw %%mm3, %%mm2 \n\t"
" paddw %%mm0, %%mm5 \n\t"
" paddw %%mm2, %%mm5 \n\t"
" pmaddwd %%mm0, %%mm0 \n\t"
" pmaddwd %%mm2, %%mm2 \n\t"
" paddd %%mm0, %%mm7 \n\t"
" paddd %%mm2, %%mm7 \n\t"
" add %4, %2 \n\t" /* Inc pointer into src data */
" add %5, %3 \n\t" /* Inc pointer into ref data */
" dec %%rdi \n\t"
" jnz 1b \n\t"
" movq %%mm5, %%mm0 \n\t"
" psrlq $32, %%mm5 \n\t"
" paddw %%mm0, %%mm5 \n\t"
" movq %%mm5, %%mm0 \n\t"
" psrlq $16, %%mm5 \n\t"
" paddw %%mm0, %%mm5 \n\t"
" movd %%mm5, %%rdi \n\t"
" movsx %%di, %%rdi \n\t"
" mov %%rdi, %0 \n\t"
" movq %%mm7, %%mm0 \n\t"
" psrlq $32, %%mm7 \n\t"
" paddd %%mm0, %%mm7 \n\t"
" movd %%mm7, %1 \n\t"
: "=m" (XSum),
"=m" (XXSum),
"+r" (SrcData),
"+r" (RefDataPtr)
: "r" ((ogg_uint64_t)SrcStride),
"r" ((ogg_uint64_t)RefStride)
: "rdi", "memory"
);
/* Compute and return population variance as mis-match metric. */
return (( (XXSum<<6) - XSum*XSum ));
}
static void restore_fpu (void)
{
__asm__ __volatile__ (
" emms \n\t"
);
}
void dsp_mmx_init(DspFunctions *funcs)
{
funcs->restore_fpu = restore_fpu;
funcs->sub8x8 = sub8x8__mmx;
funcs->sub8x8_128 = sub8x8_128__mmx;
funcs->sub8x8avg2 = sub8x8avg2__mmx;
funcs->intra8x8_err = intra8x8_err__mmx;
funcs->inter8x8_err = inter8x8_err__mmx;
}
#endif /* USE_ASM */

323
Engine/lib/libtheora/lib/enc/x86_64/dsp_mmxext.c Normal file
View file

@ -0,0 +1,323 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: dsp_mmxext.c 15397 2008-10-14 02:06:24Z tterribe $
********************************************************************/
#include <stdlib.h>
#include "../codec_internal.h"
#include "../dsp.h"
#if defined(USE_ASM)
typedef unsigned long long ogg_uint64_t;
static ogg_uint32_t sad8x8__mmxext (unsigned char *ptr1, ogg_uint32_t stride1,
unsigned char *ptr2, ogg_uint32_t stride2)
{
ogg_uint32_t DiffVal;
__asm__ __volatile__ (
" .balign 16 \n\t"
" pxor %%mm7, %%mm7 \n\t" /* mm7 contains the result */
".rept 7 \n\t"
" movq (%1), %%mm0 \n\t" /* take 8 bytes */
" movq (%2), %%mm1 \n\t"
" psadbw %%mm1, %%mm0 \n\t"
" add %3, %1 \n\t" /* Inc pointer into the new data */
" paddw %%mm0, %%mm7 \n\t" /* accumulate difference... */
" add %4, %2 \n\t" /* Inc pointer into ref data */
".endr \n\t"
" movq (%1), %%mm0 \n\t" /* take 8 bytes */
" movq (%2), %%mm1 \n\t"
" psadbw %%mm1, %%mm0 \n\t"
" paddw %%mm0, %%mm7 \n\t" /* accumulate difference... */
" movd %%mm7, %0 \n\t"
: "=r" (DiffVal),
"+r" (ptr1),
"+r" (ptr2)
: "r" ((ogg_uint64_t)stride1),
"r" ((ogg_uint64_t)stride2)
: "memory"
);
return DiffVal;
}
static ogg_uint32_t sad8x8_thres__mmxext (unsigned char *ptr1, ogg_uint32_t stride1,
unsigned char *ptr2, ogg_uint32_t stride2,
ogg_uint32_t thres)
{
ogg_uint32_t DiffVal;
__asm__ __volatile__ (
" .balign 16 \n\t"
" pxor %%mm7, %%mm7 \n\t" /* mm7 contains the result */
".rept 8 \n\t"
" movq (%1), %%mm0 \n\t" /* take 8 bytes */
" movq (%2), %%mm1 \n\t"
" psadbw %%mm1, %%mm0 \n\t"
" add %3, %1 \n\t" /* Inc pointer into the new data */
" paddw %%mm0, %%mm7 \n\t" /* accumulate difference... */
" add %4, %2 \n\t" /* Inc pointer into ref data */
".endr \n\t"
" movd %%mm7, %0 \n\t"
: "=r" (DiffVal),
"+r" (ptr1),
"+r" (ptr2)
: "r" ((ogg_uint64_t)stride1),
"r" ((ogg_uint64_t)stride2)
: "memory"
);
return DiffVal;
}
static ogg_uint32_t sad8x8_xy2_thres__mmxext (unsigned char *SrcData, ogg_uint32_t SrcStride,
unsigned char *RefDataPtr1,
unsigned char *RefDataPtr2, ogg_uint32_t RefStride,
ogg_uint32_t thres)
{
ogg_uint32_t DiffVal;
__asm__ __volatile__ (
" .balign 16 \n\t"
" pxor %%mm7, %%mm7 \n\t" /* mm7 contains the result */
".rept 8 \n\t"
" movq (%1), %%mm0 \n\t" /* take 8 bytes */
" movq (%2), %%mm1 \n\t"
" movq (%3), %%mm2 \n\t"
" pavgb %%mm2, %%mm1 \n\t"
" psadbw %%mm1, %%mm0 \n\t"
" add %4, %1 \n\t" /* Inc pointer into the new data */
" paddw %%mm0, %%mm7 \n\t" /* accumulate difference... */
" add %5, %2 \n\t" /* Inc pointer into ref data */
" add %5, %3 \n\t" /* Inc pointer into ref data */
".endr \n\t"
" movd %%mm7, %0 \n\t"
: "=m" (DiffVal),
"+r" (SrcData),
"+r" (RefDataPtr1),
"+r" (RefDataPtr2)
: "r" ((ogg_uint64_t)SrcStride),
"r" ((ogg_uint64_t)RefStride)
: "memory"
);
return DiffVal;
}
static ogg_uint32_t row_sad8__mmxext (unsigned char *Src1, unsigned char *Src2)
{
ogg_uint32_t MaxSad;
__asm__ __volatile__ (
" .balign 16 \n\t"
" movd (%1), %%mm0 \n\t"
" movd (%2), %%mm1 \n\t"
" psadbw %%mm0, %%mm1 \n\t"
" movd 4(%1), %%mm2 \n\t"
" movd 4(%2), %%mm3 \n\t"
" psadbw %%mm2, %%mm3 \n\t"
" pmaxsw %%mm1, %%mm3 \n\t"
" movd %%mm3, %0 \n\t"
" andl $0xffff, %0 \n\t"
: "=m" (MaxSad),
"+r" (Src1),
"+r" (Src2)
:
: "memory"
);
return MaxSad;
}
static ogg_uint32_t col_sad8x8__mmxext (unsigned char *Src1, unsigned char *Src2,
ogg_uint32_t stride)
{
ogg_uint32_t MaxSad;
__asm__ __volatile__ (
" .balign 16 \n\t"
" pxor %%mm3, %%mm3 \n\t" /* zero out mm3 for unpack */
" pxor %%mm4, %%mm4 \n\t" /* mm4 low sum */
" pxor %%mm5, %%mm5 \n\t" /* mm5 high sum */
" pxor %%mm6, %%mm6 \n\t" /* mm6 low sum */
" pxor %%mm7, %%mm7 \n\t" /* mm7 high sum */
" mov $4, %%rdi \n\t" /* 4 rows */
"1: \n\t"
" movq (%1), %%mm0 \n\t" /* take 8 bytes */
" movq (%2), %%mm1 \n\t" /* take 8 bytes */
" movq %%mm0, %%mm2 \n\t"
" psubusb %%mm1, %%mm0 \n\t" /* A - B */
" psubusb %%mm2, %%mm1 \n\t" /* B - A */
" por %%mm1, %%mm0 \n\t" /* and or gives abs difference */
" movq %%mm0, %%mm1 \n\t"
" punpcklbw %%mm3, %%mm0 \n\t" /* unpack to higher precision for accumulation */
" paddw %%mm0, %%mm4 \n\t" /* accumulate difference... */
" punpckhbw %%mm3, %%mm1 \n\t" /* unpack high four bytes to higher precision */
" paddw %%mm1, %%mm5 \n\t" /* accumulate difference... */
" add %3, %1 \n\t" /* Inc pointer into the new data */
" add %3, %2 \n\t" /* Inc pointer into the new data */
" dec %%rdi \n\t"
" jnz 1b \n\t"
" mov $4, %%rdi \n\t" /* 4 rows */
"2: \n\t"
" movq (%1), %%mm0 \n\t" /* take 8 bytes */
" movq (%2), %%mm1 \n\t" /* take 8 bytes */
" movq %%mm0, %%mm2 \n\t"
" psubusb %%mm1, %%mm0 \n\t" /* A - B */
" psubusb %%mm2, %%mm1 \n\t" /* B - A */
" por %%mm1, %%mm0 \n\t" /* and or gives abs difference */
" movq %%mm0, %%mm1 \n\t"
" punpcklbw %%mm3, %%mm0 \n\t" /* unpack to higher precision for accumulation */
" paddw %%mm0, %%mm6 \n\t" /* accumulate difference... */
" punpckhbw %%mm3, %%mm1 \n\t" /* unpack high four bytes to higher precision */
" paddw %%mm1, %%mm7 \n\t" /* accumulate difference... */
" add %3, %1 \n\t" /* Inc pointer into the new data */
" add %3, %2 \n\t" /* Inc pointer into the new data */
" dec %%rdi \n\t"
" jnz 2b \n\t"
" pmaxsw %%mm6, %%mm7 \n\t"
" pmaxsw %%mm4, %%mm5 \n\t"
" pmaxsw %%mm5, %%mm7 \n\t"
" movq %%mm7, %%mm6 \n\t"
" psrlq $32, %%mm6 \n\t"
" pmaxsw %%mm6, %%mm7 \n\t"
" movq %%mm7, %%mm6 \n\t"
" psrlq $16, %%mm6 \n\t"
" pmaxsw %%mm6, %%mm7 \n\t"
" movd %%mm7, %0 \n\t"
" andl $0xffff, %0 \n\t"
: "=r" (MaxSad),
"+r" (Src1),
"+r" (Src2)
: "r" ((ogg_uint64_t)stride)
: "memory", "rdi"
);
return MaxSad;
}
static ogg_uint32_t inter8x8_err_xy2__mmxext (unsigned char *SrcData, ogg_uint32_t SrcStride,
unsigned char *RefDataPtr1,
unsigned char *RefDataPtr2, ogg_uint32_t RefStride)
{
ogg_uint64_t XSum;
ogg_uint64_t XXSum;
__asm__ __volatile__ (
" .balign 16 \n\t"
" pxor %%mm4, %%mm4 \n\t"
" pxor %%mm5, %%mm5 \n\t"
" pxor %%mm6, %%mm6 \n\t"
" pxor %%mm7, %%mm7 \n\t"
" mov $8, %%rdi \n\t"
"1: \n\t"
" movq (%2), %%mm0 \n\t" /* take 8 bytes */
" movq (%3), %%mm2 \n\t"
" movq (%4), %%mm1 \n\t" /* take average of mm2 and mm1 */
" pavgb %%mm2, %%mm1 \n\t"
" movq %%mm0, %%mm2 \n\t"
" movq %%mm1, %%mm3 \n\t"
" punpcklbw %%mm6, %%mm0 \n\t"
" punpcklbw %%mm4, %%mm1 \n\t"
" punpckhbw %%mm6, %%mm2 \n\t"
" punpckhbw %%mm4, %%mm3 \n\t"
" psubsw %%mm1, %%mm0 \n\t"
" psubsw %%mm3, %%mm2 \n\t"
" paddw %%mm0, %%mm5 \n\t"
" paddw %%mm2, %%mm5 \n\t"
" pmaddwd %%mm0, %%mm0 \n\t"
" pmaddwd %%mm2, %%mm2 \n\t"
" paddd %%mm0, %%mm7 \n\t"
" paddd %%mm2, %%mm7 \n\t"
" add %5, %2 \n\t" /* Inc pointer into src data */
" add %6, %3 \n\t" /* Inc pointer into ref data */
" add %6, %4 \n\t" /* Inc pointer into ref data */
" dec %%rdi \n\t"
" jnz 1b \n\t"
" movq %%mm5, %%mm0 \n\t"
" psrlq $32, %%mm5 \n\t"
" paddw %%mm0, %%mm5 \n\t"
" movq %%mm5, %%mm0 \n\t"
" psrlq $16, %%mm5 \n\t"
" paddw %%mm0, %%mm5 \n\t"
" movd %%mm5, %%edi \n\t"
" movsx %%di, %%edi \n\t"
" movl %%edi, %0 \n\t"
" movq %%mm7, %%mm0 \n\t"
" psrlq $32, %%mm7 \n\t"
" paddd %%mm0, %%mm7 \n\t"
" movd %%mm7, %1 \n\t"
: "=m" (XSum),
"=m" (XXSum),
"+r" (SrcData),
"+r" (RefDataPtr1),
"+r" (RefDataPtr2)
: "r" ((ogg_uint64_t)SrcStride),
"r" ((ogg_uint64_t)RefStride)
: "rdi", "memory"
);
/* Compute and return population variance as mis-match metric. */
return (( (XXSum<<6) - XSum*XSum ));
}
void dsp_mmxext_init(DspFunctions *funcs)
{
funcs->row_sad8 = row_sad8__mmxext;
funcs->col_sad8x8 = col_sad8x8__mmxext;
funcs->sad8x8 = sad8x8__mmxext;
funcs->sad8x8_thres = sad8x8_thres__mmxext;
funcs->sad8x8_xy2_thres = sad8x8_xy2_thres__mmxext;
funcs->inter8x8_err_xy2 = inter8x8_err_xy2__mmxext;
}
#endif /* USE_ASM */

342
Engine/lib/libtheora/lib/enc/x86_64/fdct_mmx.c Normal file
View file

@ -0,0 +1,342 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 1999-2006 *
* by the Xiph.Org Foundation http://www.xiph.org/ *
* *
********************************************************************/
/* mmx fdct implementation for x86_64 */
/* $Id: fdct_mmx.c 15397 2008-10-14 02:06:24Z tterribe $ */
#include "theora/theora.h"
#include "../codec_internal.h"
#include "../dsp.h"
#if defined(USE_ASM)
static const __attribute__ ((aligned(8),used)) ogg_int64_t xC1S7 = 0x0fb15fb15fb15fb15LL;
static const __attribute__ ((aligned(8),used)) ogg_int64_t xC2S6 = 0x0ec83ec83ec83ec83LL;
static const __attribute__ ((aligned(8),used)) ogg_int64_t xC3S5 = 0x0d4dbd4dbd4dbd4dbLL;
static const __attribute__ ((aligned(8),used)) ogg_int64_t xC4S4 = 0x0b505b505b505b505LL;
static const __attribute__ ((aligned(8),used)) ogg_int64_t xC5S3 = 0x08e3a8e3a8e3a8e3aLL;
static const __attribute__ ((aligned(8),used)) ogg_int64_t xC6S2 = 0x061f861f861f861f8LL;
static const __attribute__ ((aligned(8),used)) ogg_int64_t xC7S1 = 0x031f131f131f131f1LL;
#if defined(__MINGW32__) || defined(__CYGWIN__) || \
defined(__OS2__) || (defined (__OpenBSD__) && !defined(__ELF__))
# define M(a) "_" #a
#else
# define M(a) #a
#endif
/* execute stage 1 of forward DCT */
#define Fdct_mmx(ip0,ip1,ip2,ip3,ip4,ip5,ip6,ip7,temp) \
" movq " #ip0 ", %%mm0 \n\t" \
" movq " #ip1 ", %%mm1 \n\t" \
" movq " #ip3 ", %%mm2 \n\t" \
" movq " #ip5 ", %%mm3 \n\t" \
" movq %%mm0, %%mm4 \n\t" \
" movq %%mm1, %%mm5 \n\t" \
" movq %%mm2, %%mm6 \n\t" \
" movq %%mm3, %%mm7 \n\t" \
\
" paddsw " #ip7 ", %%mm0 \n\t" /* mm0 = ip0 + ip7 = is07 */ \
" paddsw " #ip2 ", %%mm1 \n\t" /* mm1 = ip1 + ip2 = is12 */ \
" paddsw " #ip4 ", %%mm2 \n\t" /* mm2 = ip3 + ip4 = is34 */ \
" paddsw " #ip6 ", %%mm3 \n\t" /* mm3 = ip5 + ip6 = is56 */ \
" psubsw " #ip7 ", %%mm4 \n\t" /* mm4 = ip0 - ip7 = id07 */ \
" psubsw " #ip2 ", %%mm5 \n\t" /* mm5 = ip1 - ip2 = id12 */ \
\
" psubsw %%mm2, %%mm0 \n\t" /* mm0 = is07 - is34 */ \
\
" paddsw %%mm2, %%mm2 \n\t" \
\
" psubsw " #ip4 ", %%mm6 \n\t" /* mm6 = ip3 - ip4 = id34 */ \
\
" paddsw %%mm0, %%mm2 \n\t" /* mm2 = is07 + is34 = is0734 */ \
" psubsw %%mm3, %%mm1 \n\t" /* mm1 = is12 - is56 */ \
" movq %%mm0," #temp " \n\t" /* Save is07 - is34 to free mm0; */ \
" paddsw %%mm3, %%mm3 \n\t" \
" paddsw %%mm1, %%mm3 \n\t" /* mm3 = is12 + 1s56 = is1256 */ \
\
" psubsw " #ip6 ", %%mm7 \n\t" /* mm7 = ip5 - ip6 = id56 */ \
/* ------------------------------------------------------------------- */ \
" psubsw %%mm7, %%mm5 \n\t" /* mm5 = id12 - id56 */ \
" paddsw %%mm7, %%mm7 \n\t" \
" paddsw %%mm5, %%mm7 \n\t" /* mm7 = id12 + id56 */ \
/* ------------------------------------------------------------------- */ \
" psubsw %%mm3, %%mm2 \n\t" /* mm2 = is0734 - is1256 */ \
" paddsw %%mm3, %%mm3 \n\t" \
\
" movq %%mm2, %%mm0 \n\t" /* make a copy */ \
" paddsw %%mm2, %%mm3 \n\t" /* mm3 = is0734 + is1256 */ \
\
" pmulhw %[xC4S4], %%mm0 \n\t" /* mm0 = xC4S4 * ( is0734 - is1256 ) - ( is0734 - is1256 ) */ \
" paddw %%mm2, %%mm0 \n\t" /* mm0 = xC4S4 * ( is0734 - is1256 ) */ \
" psrlw $15, %%mm2 \n\t" \
" paddw %%mm2, %%mm0 \n\t" /* Truncate mm0, now it is op[4] */ \
\
" movq %%mm3, %%mm2 \n\t" \
" movq %%mm0," #ip4 " \n\t" /* save ip4, now mm0,mm2 are free */ \
\
" movq %%mm3, %%mm0 \n\t" \
" pmulhw %[xC4S4], %%mm3 \n\t" /* mm3 = xC4S4 * ( is0734 +is1256 ) - ( is0734 +is1256 ) */ \
\
" psrlw $15, %%mm2 \n\t" \
" paddw %%mm0, %%mm3 \n\t" /* mm3 = xC4S4 * ( is0734 +is1256 ) */ \
" paddw %%mm2, %%mm3 \n\t" /* Truncate mm3, now it is op[0] */ \
\
" movq %%mm3," #ip0 " \n\t" \
/* ------------------------------------------------------------------- */ \
" movq " #temp ", %%mm3 \n\t" /* mm3 = irot_input_y */ \
" pmulhw %[xC2S6], %%mm3 \n\t" /* mm3 = xC2S6 * irot_input_y - irot_input_y */ \
\
" movq " #temp ", %%mm2 \n\t" \
" movq %%mm2, %%mm0 \n\t" \
\
" psrlw $15, %%mm2 \n\t" /* mm3 = xC2S6 * irot_input_y */ \
" paddw %%mm0, %%mm3 \n\t" \
\
" paddw %%mm2, %%mm3 \n\t" /* Truncated */ \
" movq %%mm5, %%mm0 \n\t" \
\
" movq %%mm5, %%mm2 \n\t" \
" pmulhw %[xC6S2], %%mm0 \n\t" /* mm0 = xC6S2 * irot_input_x */ \
\
" psrlw $15, %%mm2 \n\t" \
" paddw %%mm2, %%mm0 \n\t" /* Truncated */ \
\
" paddsw %%mm0, %%mm3 \n\t" /* ip[2] */ \
" movq %%mm3," #ip2 " \n\t" /* Save ip2 */ \
\
" movq %%mm5, %%mm0 \n\t" \
" movq %%mm5, %%mm2 \n\t" \
\
" pmulhw %[xC2S6], %%mm5 \n\t" /* mm5 = xC2S6 * irot_input_x - irot_input_x */ \
" psrlw $15, %%mm2 \n\t" \
\
" movq " #temp ", %%mm3 \n\t" \
" paddw %%mm0, %%mm5 \n\t" /* mm5 = xC2S6 * irot_input_x */ \
\
" paddw %%mm2, %%mm5 \n\t" /* Truncated */ \
" movq %%mm3, %%mm2 \n\t" \
\
" pmulhw %[xC6S2], %%mm3 \n\t" /* mm3 = xC6S2 * irot_input_y */ \
" psrlw $15, %%mm2 \n\t" \
\
" paddw %%mm2, %%mm3 \n\t" /* Truncated */ \
" psubsw %%mm5, %%mm3 \n\t" \
\
" movq %%mm3," #ip6 " \n\t" \
/* ------------------------------------------------------------------- */ \
" movq %[xC4S4], %%mm0 \n\t" \
" movq %%mm1, %%mm2 \n\t" \
" movq %%mm1, %%mm3 \n\t" \
\
" pmulhw %%mm0, %%mm1 \n\t" /* mm0 = xC4S4 * ( is12 - is56 ) - ( is12 - is56 ) */ \
" psrlw $15, %%mm2 \n\t" \
\
" paddw %%mm3, %%mm1 \n\t" /* mm0 = xC4S4 * ( is12 - is56 ) */ \
" paddw %%mm2, %%mm1 \n\t" /* Truncate mm1, now it is icommon_product1 */ \
\
" movq %%mm7, %%mm2 \n\t" \
" movq %%mm7, %%mm3 \n\t" \
\
" pmulhw %%mm0, %%mm7 \n\t" /* mm7 = xC4S4 * ( id12 + id56 ) - ( id12 + id56 ) */ \
" psrlw $15, %%mm2 \n\t" \
\
" paddw %%mm3, %%mm7 \n\t" /* mm7 = xC4S4 * ( id12 + id56 ) */ \
" paddw %%mm2, %%mm7 \n\t" /* Truncate mm7, now it is icommon_product2 */ \
/* ------------------------------------------------------------------- */ \
" pxor %%mm0, %%mm0 \n\t" /* Clear mm0 */ \
" psubsw %%mm6, %%mm0 \n\t" /* mm0 = - id34 */ \
\
" psubsw %%mm7, %%mm0 \n\t" /* mm0 = - ( id34 + idcommon_product2 ) */ \
" paddsw %%mm6, %%mm6 \n\t" \
" paddsw %%mm0, %%mm6 \n\t" /* mm6 = id34 - icommon_product2 */ \
\
" psubsw %%mm1, %%mm4 \n\t" /* mm4 = id07 - icommon_product1 */ \
" paddsw %%mm1, %%mm1 \n\t" \
" paddsw %%mm4, %%mm1 \n\t" /* mm1 = id07 + icommon_product1 */ \
/* ------------------------------------------------------------------- */ \
" movq %[xC1S7], %%mm7 \n\t" \
" movq %%mm1, %%mm2 \n\t" \
\
" movq %%mm1, %%mm3 \n\t" \
" pmulhw %%mm7, %%mm1 \n\t" /* mm1 = xC1S7 * irot_input_x - irot_input_x */ \
\
" movq %[xC7S1], %%mm7 \n\t" \
" psrlw $15, %%mm2 \n\t" \
\
" paddw %%mm3, %%mm1 \n\t" /* mm1 = xC1S7 * irot_input_x */ \
" paddw %%mm2, %%mm1 \n\t" /* Trucated */ \
\
" pmulhw %%mm7, %%mm3 \n\t" /* mm3 = xC7S1 * irot_input_x */ \
" paddw %%mm2, %%mm3 \n\t" /* Truncated */ \
\
" movq %%mm0, %%mm5 \n\t" \
" movq %%mm0, %%mm2 \n\t" \
\
" movq %[xC1S7], %%mm7 \n\t" \
" pmulhw %%mm7, %%mm0 \n\t" /* mm0 = xC1S7 * irot_input_y - irot_input_y */ \
\
" movq %[xC7S1], %%mm7 \n\t" \
" psrlw $15, %%mm2 \n\t" \
\
" paddw %%mm5, %%mm0 \n\t" /* mm0 = xC1S7 * irot_input_y */ \
" paddw %%mm2, %%mm0 \n\t" /* Truncated */ \
\
" pmulhw %%mm7, %%mm5 \n\t" /* mm5 = xC7S1 * irot_input_y */ \
" paddw %%mm2, %%mm5 \n\t" /* Truncated */ \
\
" psubsw %%mm5, %%mm1 \n\t" /* mm1 = xC1S7 * irot_input_x - xC7S1 * irot_input_y = ip1 */ \
" paddsw %%mm0, %%mm3 \n\t" /* mm3 = xC7S1 * irot_input_x - xC1S7 * irot_input_y = ip7 */ \
\
" movq %%mm1," #ip1 " \n\t" \
" movq %%mm3," #ip7 " \n\t" \
/* ------------------------------------------------------------------- */ \
" movq %[xC3S5], %%mm0 \n\t" \
" movq %[xC5S3], %%mm1 \n\t" \
\
" movq %%mm6, %%mm5 \n\t" \
" movq %%mm6, %%mm7 \n\t" \
\
" movq %%mm4, %%mm2 \n\t" \
" movq %%mm4, %%mm3 \n\t" \
\
" pmulhw %%mm0, %%mm4 \n\t" /* mm4 = xC3S5 * irot_input_x - irot_input_x */ \
" pmulhw %%mm1, %%mm6 \n\t" /* mm6 = xC5S3 * irot_input_y - irot_input_y */ \
\
" psrlw $15, %%mm2 \n\t" \
" psrlw $15, %%mm5 \n\t" \
\
" paddw %%mm3, %%mm4 \n\t" /* mm4 = xC3S5 * irot_input_x */ \
" paddw %%mm7, %%mm6 \n\t" /* mm6 = xC5S3 * irot_input_y */ \
\
" paddw %%mm2, %%mm4 \n\t" /* Truncated */ \
" paddw %%mm5, %%mm6 \n\t" /* Truncated */ \
\
" psubsw %%mm6, %%mm4 \n\t" /* ip3 */ \
" movq %%mm4," #ip3 " \n\t" \
\
" movq %%mm3, %%mm4 \n\t" \
" movq %%mm7, %%mm6 \n\t" \
\
" pmulhw %%mm1, %%mm3 \n\t" /* mm3 = xC5S3 * irot_input_x - irot_input_x */ \
" pmulhw %%mm0, %%mm7 \n\t" /* mm7 = xC3S5 * irot_input_y - irot_input_y */ \
\
" paddw %%mm2, %%mm4 \n\t" \
" paddw %%mm5, %%mm6 \n\t" \
\
" paddw %%mm4, %%mm3 \n\t" /* mm3 = xC5S3 * irot_input_x */ \
" paddw %%mm6, %%mm7 \n\t" /* mm7 = xC3S5 * irot_input_y */ \
\
" paddw %%mm7, %%mm3 \n\t" /* ip5 */ \
" movq %%mm3," #ip5 " \n\t"
#define Transpose_mmx(ip0,ip1,ip2,ip3,ip4,ip5,ip6,ip7, \
op0,op1,op2,op3,op4,op5,op6,op7) \
" movq " #ip0 ", %%mm0 \n\t" /* mm0 = a0 a1 a2 a3 */ \
" movq " #ip4 ", %%mm4 \n\t" /* mm4 = e4 e5 e6 e7 */ \
" movq " #ip1 ", %%mm1 \n\t" /* mm1 = b0 b1 b2 b3 */ \
" movq " #ip5 ", %%mm5 \n\t" /* mm5 = f4 f5 f6 f7 */ \
" movq " #ip2 ", %%mm2 \n\t" /* mm2 = c0 c1 c2 c3 */ \
" movq " #ip6 ", %%mm6 \n\t" /* mm6 = g4 g5 g6 g7 */ \
" movq " #ip3 ", %%mm3 \n\t" /* mm3 = d0 d1 d2 d3 */ \
" movq %%mm1," #op1 " \n\t" /* save b0 b1 b2 b3 */ \
" movq " #ip7 ", %%mm7 \n\t" /* mm7 = h0 h1 h2 h3 */ \
/* Transpose 2x8 block */ \
" movq %%mm4, %%mm1 \n\t" /* mm1 = e3 e2 e1 e0 */ \
" punpcklwd %%mm5, %%mm4 \n\t" /* mm4 = f1 e1 f0 e0 */ \
" movq %%mm0," #op0 " \n\t" /* save a3 a2 a1 a0 */ \
" punpckhwd %%mm5, %%mm1 \n\t" /* mm1 = f3 e3 f2 e2 */ \
" movq %%mm6, %%mm0 \n\t" /* mm0 = g3 g2 g1 g0 */ \
" punpcklwd %%mm7, %%mm6 \n\t" /* mm6 = h1 g1 h0 g0 */ \
" movq %%mm4, %%mm5 \n\t" /* mm5 = f1 e1 f0 e0 */ \
" punpckldq %%mm6, %%mm4 \n\t" /* mm4 = h0 g0 f0 e0 = MM4 */ \
" punpckhdq %%mm6, %%mm5 \n\t" /* mm5 = h1 g1 f1 e1 = MM5 */ \
" movq %%mm1, %%mm6 \n\t" /* mm6 = f3 e3 f2 e2 */ \
" movq %%mm4," #op4 " \n\t" \
" punpckhwd %%mm7, %%mm0 \n\t" /* mm0 = h3 g3 h2 g2 */ \
" movq %%mm5," #op5 " \n\t" \
" punpckhdq %%mm0, %%mm6 \n\t" /* mm6 = h3 g3 f3 e3 = MM7 */ \
" movq " #op0 ", %%mm4 \n\t" /* mm4 = a3 a2 a1 a0 */ \
" punpckldq %%mm0, %%mm1 \n\t" /* mm1 = h2 g2 f2 e2 = MM6 */ \
" movq " #op1 ", %%mm5 \n\t" /* mm5 = b3 b2 b1 b0 */ \
" movq %%mm4, %%mm0 \n\t" /* mm0 = a3 a2 a1 a0 */ \
" movq %%mm6," #op7 " \n\t" \
" punpcklwd %%mm5, %%mm0 \n\t" /* mm0 = b1 a1 b0 a0 */ \
" movq %%mm1," #op6 " \n\t" \
" punpckhwd %%mm5, %%mm4 \n\t" /* mm4 = b3 a3 b2 a2 */ \
" movq %%mm2, %%mm5 \n\t" /* mm5 = c3 c2 c1 c0 */ \
" punpcklwd %%mm3, %%mm2 \n\t" /* mm2 = d1 c1 d0 c0 */ \
" movq %%mm0, %%mm1 \n\t" /* mm1 = b1 a1 b0 a0 */ \
" punpckldq %%mm2, %%mm0 \n\t" /* mm0 = d0 c0 b0 a0 = MM0 */ \
" punpckhdq %%mm2, %%mm1 \n\t" /* mm1 = d1 c1 b1 a1 = MM1 */ \
" movq %%mm4, %%mm2 \n\t" /* mm2 = b3 a3 b2 a2 */ \
" movq %%mm0," #op0 " \n\t" \
" punpckhwd %%mm3, %%mm5 \n\t" /* mm5 = d3 c3 d2 c2 */ \
" movq %%mm1," #op1 " \n\t" \
" punpckhdq %%mm5, %%mm4 \n\t" /* mm4 = d3 c3 b3 a3 = MM3 */ \
" punpckldq %%mm5, %%mm2 \n\t" /* mm2 = d2 c2 b2 a2 = MM2 */ \
" movq %%mm4," #op3 " \n\t" \
" movq %%mm2," #op2 " \n\t"
/* This performs a 2D Forward DCT on an 8x8 block with short
coefficients. We try to do the truncation to match the C
version. */
static void fdct_short__mmx ( ogg_int16_t *InputData, ogg_int16_t *OutputData)
{
ogg_int16_t __attribute__((aligned(8))) temp[8*8];
__asm__ __volatile__ (
" .balign 16 \n\t"
/*
* Input data is an 8x8 block. To make processing of the data more efficent
* we will transpose the block of data to two 4x8 blocks???
*/
Transpose_mmx ( (%0), 16(%0), 32(%0), 48(%0), 8(%0), 24(%0), 40(%0), 56(%0),
(%1), 16(%1), 32(%1), 48(%1), 8(%1), 24(%1), 40(%1), 56(%1))
Fdct_mmx ( (%1), 16(%1), 32(%1), 48(%1), 8(%1), 24(%1), 40(%1), 56(%1), (%2))
Transpose_mmx (64(%0), 80(%0), 96(%0),112(%0), 72(%0), 88(%0),104(%0),120(%0),
64(%1), 80(%1), 96(%1),112(%1), 72(%1), 88(%1),104(%1),120(%1))
Fdct_mmx (64(%1), 80(%1), 96(%1),112(%1), 72(%1), 88(%1),104(%1),120(%1), (%2))
Transpose_mmx ( 0(%1), 16(%1), 32(%1), 48(%1), 64(%1), 80(%1), 96(%1),112(%1),
0(%1), 16(%1), 32(%1), 48(%1), 64(%1), 80(%1), 96(%1),112(%1))
Fdct_mmx ( 0(%1), 16(%1), 32(%1), 48(%1), 64(%1), 80(%1), 96(%1),112(%1), (%2))
Transpose_mmx ( 8(%1), 24(%1), 40(%1), 56(%1), 72(%1), 88(%1),104(%1),120(%1),
8(%1), 24(%1), 40(%1), 56(%1), 72(%1), 88(%1),104(%1),120(%1))
Fdct_mmx ( 8(%1), 24(%1), 40(%1), 56(%1), 72(%1), 88(%1),104(%1),120(%1), (%2))
" emms \n\t"
: "+r" (InputData),
"+r" (OutputData)
: "r" (temp),
[xC1S7] "m" (xC1S7), /* gcc 3.1+ allows named asm parameters */
[xC2S6] "m" (xC2S6),
[xC3S5] "m" (xC3S5),
[xC4S4] "m" (xC4S4),
[xC5S3] "m" (xC5S3),
[xC6S2] "m" (xC6S2),
[xC7S1] "m" (xC7S1)
: "memory"
);
}
/* install our implementation in the function table */
void dsp_mmx_fdct_init(DspFunctions *funcs)
{
funcs->fdct_short = fdct_short__mmx;
}
#endif /* USE_ASM */

27
Engine/lib/libtheora/lib/enc/x86_64/idct_mmx.c Normal file
View file

@ -0,0 +1,27 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: idct_mmx.c 15397 2008-10-14 02:06:24Z tterribe $
********************************************************************/
#include "../codec_internal.h"
#if defined(USE_ASM)
/* nothing implemented right now */
void dsp_mmx_idct_init(DspFunctions *funcs)
{
}
#endif /* USE_ASM */

184
Engine/lib/libtheora/lib/enc/x86_64/recon_mmx.c Normal file
View file

@ -0,0 +1,184 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: recon_mmx.c 15397 2008-10-14 02:06:24Z tterribe $
********************************************************************/
#include "../codec_internal.h"
#if defined(USE_ASM)
typedef unsigned long long ogg_uint64_t;
static const __attribute__ ((aligned(8),used)) ogg_int64_t V128 = 0x8080808080808080LL;
static void copy8x8__mmx (unsigned char *src,
unsigned char *dest,
ogg_uint32_t stride)
{
__asm__ __volatile__ (
" .balign 16 \n\t"
" lea (%2, %2, 2), %%rdi \n\t"
" movq (%1), %%mm0 \n\t"
" movq (%1, %2), %%mm1 \n\t"
" movq (%1, %2, 2), %%mm2 \n\t"
" movq (%1, %%rdi), %%mm3 \n\t"
" lea (%1, %2, 4), %1 \n\t"
" movq %%mm0, (%0) \n\t"
" movq %%mm1, (%0, %2) \n\t"
" movq %%mm2, (%0, %2, 2) \n\t"
" movq %%mm3, (%0, %%rdi) \n\t"
" lea (%0, %2, 4), %0 \n\t"
" movq (%1), %%mm0 \n\t"
" movq (%1, %2), %%mm1 \n\t"
" movq (%1, %2, 2), %%mm2 \n\t"
" movq (%1, %%rdi), %%mm3 \n\t"
" movq %%mm0, (%0) \n\t"
" movq %%mm1, (%0, %2) \n\t"
" movq %%mm2, (%0, %2, 2) \n\t"
" movq %%mm3, (%0, %%rdi) \n\t"
: "+a" (dest)
: "c" (src),
"d" ((ogg_uint64_t)stride)
: "memory", "rdi"
);
}
static void recon_intra8x8__mmx (unsigned char *ReconPtr, ogg_int16_t *ChangePtr,
ogg_uint32_t LineStep)
{
__asm__ __volatile__ (
" .balign 16 \n\t"
" movq %[V128], %%mm0 \n\t" /* Set mm0 to 0x8080808080808080 */
" lea 128(%1), %%rdi \n\t" /* Endpoint in input buffer */
"1: \n\t"
" movq (%1), %%mm2 \n\t" /* First four input values */
" packsswb 8(%1), %%mm2 \n\t" /* pack with next(high) four values */
" por %%mm0, %%mm0 \n\t"
" pxor %%mm0, %%mm2 \n\t" /* Convert result to unsigned (same as add 128) */
" lea 16(%1), %1 \n\t" /* Step source buffer */
" cmp %%rdi, %1 \n\t" /* are we done */
" movq %%mm2, (%0) \n\t" /* store results */
" lea (%0, %2), %0 \n\t" /* Step output buffer */
" jc 1b \n\t" /* Loop back if we are not done */
: "+r" (ReconPtr)
: "r" (ChangePtr),
"r" ((ogg_uint64_t)LineStep),
[V128] "m" (V128)
: "memory", "rdi"
);
}
static void recon_inter8x8__mmx (unsigned char *ReconPtr, unsigned char *RefPtr,
ogg_int16_t *ChangePtr, ogg_uint32_t LineStep)
{
__asm__ __volatile__ (
" .balign 16 \n\t"
" pxor %%mm0, %%mm0 \n\t"
" lea 128(%1), %%rdi \n\t"
"1: \n\t"
" movq (%2), %%mm2 \n\t" /* (+3 misaligned) 8 reference pixels */
" movq (%1), %%mm4 \n\t" /* first 4 changes */
" movq %%mm2, %%mm3 \n\t"
" movq 8(%1), %%mm5 \n\t" /* last 4 changes */
" punpcklbw %%mm0, %%mm2 \n\t" /* turn first 4 refs into positive 16-bit #s */
" paddsw %%mm4, %%mm2 \n\t" /* add in first 4 changes */
" punpckhbw %%mm0, %%mm3 \n\t" /* turn last 4 refs into positive 16-bit #s */
" paddsw %%mm5, %%mm3 \n\t" /* add in last 4 changes */
" add %3, %2 \n\t" /* next row of reference pixels */
" packuswb %%mm3, %%mm2 \n\t" /* pack result to unsigned 8-bit values */
" lea 16(%1), %1 \n\t" /* next row of changes */
" cmp %%rdi, %1 \n\t" /* are we done? */
" movq %%mm2, (%0) \n\t" /* store result */
" lea (%0, %3), %0 \n\t" /* next row of output */
" jc 1b \n\t"
: "+r" (ReconPtr)
: "r" (ChangePtr),
"r" (RefPtr),
"r" ((ogg_uint64_t)LineStep)
: "memory", "rdi"
);
}
static void recon_inter8x8_half__mmx (unsigned char *ReconPtr, unsigned char *RefPtr1,
unsigned char *RefPtr2, ogg_int16_t *ChangePtr,
ogg_uint32_t LineStep)
{
__asm__ __volatile__ (
" .balign 16 \n\t"
" pxor %%mm0, %%mm0 \n\t"
" lea 128(%1), %%rdi \n\t"
"1: \n\t"
" movq (%2), %%mm2 \n\t" /* (+3 misaligned) 8 reference pixels */
" movq (%3), %%mm4 \n\t" /* (+3 misaligned) 8 reference pixels */
" movq %%mm2, %%mm3 \n\t"
" punpcklbw %%mm0, %%mm2 \n\t" /* mm2 = start ref1 as positive 16-bit #s */
" movq %%mm4, %%mm5 \n\t"
" movq (%1), %%mm6 \n\t" /* first 4 changes */
" punpckhbw %%mm0, %%mm3 \n\t" /* mm3 = end ref1 as positive 16-bit #s */
" movq 8(%1), %%mm7 \n\t" /* last 4 changes */
" punpcklbw %%mm0, %%mm4 \n\t" /* mm4 = start ref2 as positive 16-bit #s */
" punpckhbw %%mm0, %%mm5 \n\t" /* mm5 = end ref2 as positive 16-bit #s */
" paddw %%mm4, %%mm2 \n\t" /* mm2 = start (ref1 + ref2) */
" paddw %%mm5, %%mm3 \n\t" /* mm3 = end (ref1 + ref2) */
" psrlw $1, %%mm2 \n\t" /* mm2 = start (ref1 + ref2)/2 */
" psrlw $1, %%mm3 \n\t" /* mm3 = end (ref1 + ref2)/2 */
" paddw %%mm6, %%mm2 \n\t" /* add changes to start */
" paddw %%mm7, %%mm3 \n\t" /* add changes to end */
" lea 16(%1), %1 \n\t" /* next row of changes */
" packuswb %%mm3, %%mm2 \n\t" /* pack start|end to unsigned 8-bit */
" add %4, %2 \n\t" /* next row of reference pixels */
" add %4, %3 \n\t" /* next row of reference pixels */
" movq %%mm2, (%0) \n\t" /* store result */
" add %4, %0 \n\t" /* next row of output */
" cmp %%rdi, %1 \n\t" /* are we done? */
" jc 1b \n\t"
: "+r" (ReconPtr)
: "r" (ChangePtr),
"r" (RefPtr1),
"r" (RefPtr2),
"r" ((ogg_uint64_t)LineStep)
: "memory", "rdi"
);
}
void dsp_mmx_recon_init(DspFunctions *funcs)
{
funcs->copy8x8 = copy8x8__mmx;
funcs->recon_intra8x8 = recon_intra8x8__mmx;
funcs->recon_inter8x8 = recon_inter8x8__mmx;
funcs->recon_inter8x8_half = recon_inter8x8_half__mmx;
}
#endif /* USE_ASM */

491
Engine/lib/libtheora/lib/internal.h Normal file
View file

@ -0,0 +1,491 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2007 *
* by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
* *
********************************************************************
function:
last mod: $Id: internal.h 15469 2008-10-30 12:49:42Z tterribe $
********************************************************************/
#if !defined(_internal_H)
# define _internal_H (1)
# include <stdlib.h>
# if defined(HAVE_CONFIG_H)
# include <config.h>
# endif
# include "theora/codec.h"
# include "theora/theora.h"
# include "dec/ocintrin.h"
# include "dec/huffman.h"
# include "dec/quant.h"
/*Thank you Microsoft, I know the order of operations.*/
# if defined(_MSC_VER)
# pragma warning(disable:4554) /* order of operations */
# pragma warning(disable:4799) /* disable missing EMMS warnings */
# endif
/*This library's version.*/
# define OC_VENDOR_STRING "Xiph.Org libTheora I 20081020 3 2 1"
/*Theora bitstream version.*/
# define TH_VERSION_MAJOR (3)
# define TH_VERSION_MINOR (2)
# define TH_VERSION_SUB (1)
# define TH_VERSION_CHECK(_info,_maj,_min,_sub) \
((_info)->version_major>(_maj)||(_info)->version_major==(_maj)&& \
((_info)->version_minor>(_min)||(_info)->version_minor==(_min)&& \
(_info)->version_subminor>=(_sub)))
/*A keyframe.*/
#define OC_INTRA_FRAME (0)
/*A predicted frame.*/
#define OC_INTER_FRAME (1)
/*A frame of unknown type (frame type decision has not yet been made).*/
#define OC_UNKWN_FRAME (-1)
/*The amount of padding to add to the reconstructed frame buffers on all
sides.
This is used to allow unrestricted motion vectors without special casing.
This must be a multiple of 2.*/
#define OC_UMV_PADDING (16)
/*Frame classification indices.*/
/*The previous golden frame.*/
#define OC_FRAME_GOLD (0)
/*The previous frame.*/
#define OC_FRAME_PREV (1)
/*The current frame.*/
#define OC_FRAME_SELF (2)
/*The input or output buffer.*/
#define OC_FRAME_IO (3)
/*Macroblock modes.*/
/*Macro block is invalid: It is never coded.*/
#define OC_MODE_INVALID (-1)
/*Encoded difference from the same macro block in the previous frame.*/
#define OC_MODE_INTER_NOMV (0)
/*Encoded with no motion compensated prediction.*/
#define OC_MODE_INTRA (1)
/*Encoded difference from the previous frame offset by the given motion
vector.*/
#define OC_MODE_INTER_MV (2)
/*Encoded difference from the previous frame offset by the last coded motion
vector.*/
#define OC_MODE_INTER_MV_LAST (3)
/*Encoded difference from the previous frame offset by the second to last
coded motion vector.*/
#define OC_MODE_INTER_MV_LAST2 (4)
/*Encoded difference from the same macro block in the previous golden
frame.*/
#define OC_MODE_GOLDEN_NOMV (5)
/*Encoded difference from the previous golden frame offset by the given motion
vector.*/
#define OC_MODE_GOLDEN_MV (6)
/*Encoded difference from the previous frame offset by the individual motion
vectors given for each block.*/
#define OC_MODE_INTER_MV_FOUR (7)
/*The number of (coded) modes.*/
#define OC_NMODES (8)
/*Macro block is not coded.*/
#define OC_MODE_NOT_CODED (8)
/*Predictor bit flags.*/
/*Left.*/
#define OC_PL (1)
/*Upper-left.*/
#define OC_PUL (2)
/*Up.*/
#define OC_PU (4)
/*Upper-right.*/
#define OC_PUR (8)
/*Constants for the packet state machine common between encoder and decoder.*/
/*Next packet to emit/read: Codec info header.*/
#define OC_PACKET_INFO_HDR (-3)
/*Next packet to emit/read: Comment header.*/
#define OC_PACKET_COMMENT_HDR (-2)
/*Next packet to emit/read: Codec setup header.*/
#define OC_PACKET_SETUP_HDR (-1)
/*No more packets to emit/read.*/
#define OC_PACKET_DONE (INT_MAX)
typedef struct oc_theora_state oc_theora_state;
/*A map from a super block to fragment numbers.*/
typedef int oc_sb_map[4][4];
/*A map from a macro block to fragment numbers.*/
typedef int oc_mb_map[3][4];
/*A motion vector.*/
typedef signed char oc_mv[2];
/*Super block information.
Super blocks are 32x32 segments of pixels in a single color plane indexed
in image order.
Internally, super blocks are broken up into four quadrants, each of which
contains a 2x2 pattern of blocks, each of which is an 8x8 block of pixels.
Quadrants, and the blocks within them, are indexed in a special order called
a "Hilbert curve" within the super block.
In order to differentiate between the Hilbert-curve indexing strategy and
the regular image order indexing strategy, blocks indexed in image order
are called "fragments".
Fragments are indexed in image order, left to right, then bottom to top,
from Y plane to Cb plane to Cr plane.*/
typedef struct{
unsigned coded_fully:1;
unsigned coded_partially:1;
unsigned quad_valid:4;
oc_sb_map map;
}oc_sb;
/*Macro block information.
The co-located fragments in all image planes corresponding to the location of
a single luma plane super block quadrant forms a macro block.
Thus there is only a single set of macro blocks for all planes, which
contains between 6 and 12 fragments, depending on the pixel format.
Therefore macro block information is kept in a separate array from super
blocks, to avoid unused space in the other planes.*/
typedef struct{
/*The current macro block mode.
A negative number indicates the macro block lies entirely outside the
coded frame.*/
int mode;
/*The X location of the macro block's upper-left hand pixel.*/
int x;
/*The Y location of the macro block's upper-right hand pixel.*/
int y;
/*The fragments that belong to this macro block in each color plane.
Fragments are stored in image order (left to right then top to bottom).
When chroma components are decimated, the extra fragments have an index of
-1.*/
oc_mb_map map;
}oc_mb;
/*Information about a fragment which intersects the border of the displayable
region.
This marks which pixels belong to the displayable region, and is used to
ensure that pixels outside of this region are never referenced.
This allows applications to pass in buffers that are really the size of the
displayable region without causing a seg fault.*/
typedef struct{
/*A bit mask marking which pixels are in the displayable region.
Pixel (x,y) corresponds to bit (y<<3|x).*/
ogg_int64_t mask;
/*The number of pixels in the displayable region.
This is always positive, and always less than 64.*/
int npixels;
}oc_border_info;
/*Fragment information.*/
typedef struct{
/*A flag indicating whether or not this fragment is coded.*/
unsigned coded:1;
/*A flag indicating that all of this fragment lies outside the displayable
region of the frame.
Note the contrast with an invalid macro block, which is outside the coded
frame, not just the displayable one.*/
unsigned invalid:1;
/*The quality index used for this fragment's AC coefficients.*/
unsigned qi:6;
/*The mode of the macroblock this fragment belongs to.
Note that the C standard requires an explicit signed keyword for bitfield
types, since some compilers may treat them as unsigned without it.*/
signed int mbmode:8;
/*The prediction-corrected DC component.
Note that the C standard requires an explicit signed keyword for bitfield
types, since some compilers may treat them as unsigned without it.*/
signed int dc:16;
/*A pointer to the portion of an image covered by this fragment in several
images.
The first three are reconstructed frame buffers, while the last is the
input image buffer.
The appropriate stride value is determined by the color plane the fragment
belongs in.*/
unsigned char *buffer[4];
/*Information for fragments which lie partially outside the displayable
region.
For fragments completely inside or outside this region, this is NULL.*/
oc_border_info *border;
/*The motion vector used for this fragment.*/
oc_mv mv;
}oc_fragment;
/*A description of each fragment plane.*/
typedef struct{
/*The number of fragments in the horizontal direction.*/
int nhfrags;
/*The number of fragments in the vertical direction.*/
int nvfrags;
/*The offset of the first fragment in the plane.*/
int froffset;
/*The total number of fragments in the plane.*/
int nfrags;
/*The number of super blocks in the horizontal direction.*/
int nhsbs;
/*The number of super blocks in the vertical direction.*/
int nvsbs;
/*The offset of the first super block in the plane.*/
int sboffset;
/*The total number of super blocks in the plane.*/
int nsbs;
}oc_fragment_plane;
/*The shared (encoder and decoder) functions that have accelerated variants.*/
typedef struct{
void (*frag_recon_intra)(unsigned char *_dst,int _dst_ystride,
const ogg_int16_t *_residue);
void (*frag_recon_inter)(unsigned char *_dst,int _dst_ystride,
const unsigned char *_src,int _src_ystride,const ogg_int16_t *_residue);
void (*frag_recon_inter2)(unsigned char *_dst,int _dst_ystride,
const unsigned char *_src1,int _src1_ystride,const unsigned char *_src2,
int _src2_ystride,const ogg_int16_t *_residue);
void (*state_frag_copy)(const oc_theora_state *_state,
const int *_fragis,int _nfragis,int _dst_frame,int _src_frame,int _pli);
void (*state_frag_recon)(oc_theora_state *_state,oc_fragment *_frag,
int _pli,ogg_int16_t _dct_coeffs[128],int _last_zzi,int _ncoefs,
ogg_uint16_t _dc_iquant,const ogg_uint16_t _ac_iquant[64]);
void (*restore_fpu)(void);
void (*state_loop_filter_frag_rows)(oc_theora_state *_state,int *_bv,
int _refi,int _pli,int _fragy0,int _fragy_end);
}oc_base_opt_vtable;
/*Common state information between the encoder and decoder.*/
struct oc_theora_state{
/*The stream information.*/
th_info info;
/*Table for shared accelerated functions.*/
oc_base_opt_vtable opt_vtable;
/*CPU flags to detect the presence of extended instruction sets.*/
ogg_uint32_t cpu_flags;
/*The fragment plane descriptions.*/
oc_fragment_plane fplanes[3];
/*The total number of fragments in a single frame.*/
int nfrags;
/*The list of fragments, indexed in image order.*/
oc_fragment *frags;
/*The total number of super blocks in a single frame.*/
int nsbs;
/*The list of super blocks, indexed in image order.*/
oc_sb *sbs;
/*The number of macro blocks in the X direction.*/
int nhmbs;
/*The number of macro blocks in the Y direction.*/
int nvmbs;
/*The total number of macro blocks.*/
int nmbs;
/*The list of macro blocks, indexed in super block order.
That is, the macro block corresponding to the macro block mbi in (luma
plane) super block sbi is (sbi<<2|mbi).*/
oc_mb *mbs;
/*The list of coded fragments, in coded order.*/
int *coded_fragis;
/*The number of coded fragments in each plane.*/
int ncoded_fragis[3];
/*The list of uncoded fragments.
This just past the end of the list, which is in reverse order, and
uses the same block of allocated storage as the coded_fragis list.*/
int *uncoded_fragis;
/*The number of uncoded fragments in each plane.*/
int nuncoded_fragis[3];
/*The list of coded macro blocks in the Y plane, in coded order.*/
int *coded_mbis;
/*The number of coded macro blocks in the Y plane.*/
int ncoded_mbis;
/*A copy of the image data used to fill the input pointers in each fragment.
If the data pointers or strides change, these input pointers must be
re-populated.*/
th_ycbcr_buffer input;
/*The number of unique border patterns.*/
int nborders;
/*The storage for the border info for all border fragments.
This data is pointed to from the appropriate fragments.*/
oc_border_info borders[16];
/*The index of the buffers being used for each OC_FRAME_* reference frame.*/
int ref_frame_idx[3];
/*The actual buffers used for the previously decoded frames.*/
th_ycbcr_buffer ref_frame_bufs[3];
/*The storage for the reference frame buffers.*/
unsigned char *ref_frame_data;
/*The frame number of the last keyframe.*/
ogg_int64_t keyframe_num;
/*The frame number of the current frame.*/
ogg_int64_t curframe_num;
/*The granpos of the current frame.*/
ogg_int64_t granpos;
/*The type of the current frame.*/
int frame_type;
/*The quality indices of the current frame.*/
int qis[3];
/*The number of quality indices used in the current frame.*/
int nqis;
/*The dequantization tables.*/
oc_quant_table *dequant_tables[2][3];
oc_quant_tables dequant_table_data[2][3];
/*Loop filter strength parameters.*/
unsigned char loop_filter_limits[64];
};
/*The function type used to fill in the chroma plane motion vectors for a
macro block when 4 different motion vectors are specified in the luma
plane.
_cbmvs: The chroma block-level motion vectors to fill in.
_lmbmv: The luma macro-block level motion vector to fill in for use in
prediction.
_lbmvs: The luma block-level motion vectors.*/
typedef void (*oc_set_chroma_mvs_func)(oc_mv _cbmvs[4],const oc_mv _lbmvs[4]);
/*A map from the index in the zig zag scan to the coefficient number in a
block.
The extra 64 entries send out of bounds indexes to index 64.
This is used to safely ignore invalid zero runs when decoding
coefficients.*/
extern const int OC_FZIG_ZAG[128];
/*A map from the coefficient number in a block to its index in the zig zag
scan.*/
extern const int OC_IZIG_ZAG[64];
/*The predictor frame to use for each macro block mode.*/
extern const int OC_FRAME_FOR_MODE[OC_NMODES];
/*A map from physical macro block ordering to bitstream macro block
ordering within a super block.*/
extern const int OC_MB_MAP[2][2];
/*A list of the indices in the oc_mb.map array that can be valid for each of
the various chroma decimation types.*/
extern const int OC_MB_MAP_IDXS[TH_PF_NFORMATS][12];
/*The number of indices in the oc_mb.map array that can be valid for each of
the various chroma decimation types.*/
extern const int OC_MB_MAP_NIDXS[TH_PF_NFORMATS];
/*A table of functions used to fill in the Cb,Cr plane motion vectors for a
macro block when 4 different motion vectors are specified in the luma
plane.*/
extern const oc_set_chroma_mvs_func OC_SET_CHROMA_MVS_TABLE[TH_PF_NFORMATS];
int oc_ilog(unsigned _v);
void **oc_malloc_2d(size_t _height,size_t _width,size_t _sz);
void **oc_calloc_2d(size_t _height,size_t _width,size_t _sz);
void oc_free_2d(void *_ptr);
void oc_ycbcr_buffer_flip(th_ycbcr_buffer _dst,
const th_ycbcr_buffer _src);
int oc_dct_token_skip(int _token,int _extra_bits);
int oc_frag_pred_dc(const oc_fragment *_frag,
const oc_fragment_plane *_fplane,int _x,int _y,int _pred_last[3]);
int oc_state_init(oc_theora_state *_state,const th_info *_info);
void oc_state_clear(oc_theora_state *_state);
void oc_state_vtable_init_c(oc_theora_state *_state);
void oc_state_borders_fill_rows(oc_theora_state *_state,int _refi,int _pli,
int _y0,int _yend);
void oc_state_borders_fill_caps(oc_theora_state *_state,int _refi,int _pli);
void oc_state_borders_fill(oc_theora_state *_state,int _refi);
void oc_state_fill_buffer_ptrs(oc_theora_state *_state,int _buf_idx,
th_ycbcr_buffer _img);
int oc_state_mbi_for_pos(oc_theora_state *_state,int _mbx,int _mby);
int oc_state_get_mv_offsets(oc_theora_state *_state,int *_offsets,
int _dx,int _dy,int _ystride,int _pli);
int oc_state_loop_filter_init(oc_theora_state *_state,int *_bv);
void oc_state_loop_filter(oc_theora_state *_state,int _frame);
#if defined(OC_DUMP_IMAGES)
int oc_state_dump_frame(const oc_theora_state *_state,int _frame,
const char *_suf);
#endif
/*Shared accelerated functions.*/
void oc_frag_recon_intra(const oc_theora_state *_state,
unsigned char *_dst,int _dst_ystride,const ogg_int16_t *_residue);
void oc_frag_recon_inter(const oc_theora_state *_state,
unsigned char *_dst,int _dst_ystride,
const unsigned char *_src,int _src_ystride,const ogg_int16_t *_residue);
void oc_frag_recon_inter2(const oc_theora_state *_state,
unsigned char *_dst,int _dst_ystride,
const unsigned char *_src1,int _src1_ystride,const unsigned char *_src2,
int _src2_ystride,const ogg_int16_t *_residue);
void oc_state_frag_copy(const oc_theora_state *_state,const int *_fragis,
int _nfragis,int _dst_frame,int _src_frame,int _pli);
void oc_state_frag_recon(oc_theora_state *_state,oc_fragment *_frag,
int _pli,ogg_int16_t _dct_coeffs[128],int _last_zzi,int _ncoefs,
ogg_uint16_t _dc_iquant,const ogg_uint16_t _ac_iquant[64]);
void oc_state_loop_filter_frag_rows(oc_theora_state *_state,int *_bv,
int _refi,int _pli,int _fragy0,int _fragy_end);
void oc_restore_fpu(const oc_theora_state *_state);
/*Default pure-C implementations.*/
void oc_frag_recon_intra_c(unsigned char *_dst,int _dst_ystride,
const ogg_int16_t *_residue);
void oc_frag_recon_inter_c(unsigned char *_dst,int _dst_ystride,
const unsigned char *_src,int _src_ystride,const ogg_int16_t *_residue);
void oc_frag_recon_inter2_c(unsigned char *_dst,int _dst_ystride,
const unsigned char *_src1,int _src1_ystride,const unsigned char *_src2,
int _src2_ystride,const ogg_int16_t *_residue);
void oc_state_frag_copy_c(const oc_theora_state *_state,const int *_fragis,
int _nfragis,int _dst_frame,int _src_frame,int _pli);
void oc_state_frag_recon_c(oc_theora_state *_state,oc_fragment *_frag,
int _pli,ogg_int16_t _dct_coeffs[128],int _last_zzi,int _ncoefs,
ogg_uint16_t _dc_iquant,const ogg_uint16_t _ac_iquant[64]);
void oc_state_loop_filter_frag_rows_c(oc_theora_state *_state,int *_bv,
int _refi,int _pli,int _fragy0,int _fragy_end);
void oc_restore_fpu_c(void);
/*We need a way to call a few encoder functions without introducing a link-time
dependency into the decoder, while still allowing the old alpha API which
does not distinguish between encoder and decoder objects to be used.
We do this by placing a function table at the start of the encoder object
which can dispatch into the encoder library.
We do a similar thing for the decoder in case we ever decide to split off a
common base library.*/
typedef void (*oc_state_clear_func)(theora_state *_th);
typedef int (*oc_state_control_func)(theora_state *th,int req,
void *buf,size_t buf_sz);
typedef ogg_int64_t (*oc_state_granule_frame_func)(theora_state *_th,
ogg_int64_t _granulepos);
typedef double (*oc_state_granule_time_func)(theora_state *_th,
ogg_int64_t _granulepos);
typedef struct oc_state_dispatch_vtbl oc_state_dispatch_vtbl;
struct oc_state_dispatch_vtbl{
oc_state_clear_func clear;
oc_state_control_func control;
oc_state_granule_frame_func granule_frame;
oc_state_granule_time_func granule_time;
};
#endif