imdct.c

从FFMPEG转换而来的H264解码程序,VC下编译..

/* ***** BEGIN LICENSE BLOCK *****
 *
 * Portions Copyright (c) 1995-2005 RealNetworks, Inc. All Rights Reserved.
 *
 * The contents of this file, and the files included with this file,
 * are subject to the current version of the RealNetworks Public
 * Source License (the "RPSL") available at
 * http://www.helixcommunity.org/content/rpsl unless you have licensed
 * the file under the current version of the RealNetworks Community
 * Source License (the "RCSL") available at
 * http://www.helixcommunity.org/content/rcsl, in which case the RCSL
 * will apply. You may also obtain the license terms directly from
 * RealNetworks.  You may not use this file except in compliance with
 * the RPSL or, if you have a valid RCSL with RealNetworks applicable
 * to this file, the RCSL.  Please see the applicable RPSL or RCSL for
 * the rights, obligations and limitations governing use of the
 * contents of the file.
 *
 * This file is part of the Helix DNA Technology. RealNetworks is the
 * developer of the Original Code and owns the copyrights in the
 * portions it created.
 *
 * This file, and the files included with this file, is distributed
 * and made available on an 'AS IS' basis, WITHOUT WARRANTY OF ANY
 * KIND, EITHER EXPRESS OR IMPLIED, AND REALNETWORKS HEREBY DISCLAIMS
 * ALL SUCH WARRANTIES, INCLUDING WITHOUT LIMITATION, ANY WARRANTIES
 * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE, QUIET
 * ENJOYMENT OR NON-INFRINGEMENT.
 *
 * Technology Compatibility Kit Test Suite(s) Location:
 *    http://www.helixcommunity.org/content/tck
 *
 * Contributor(s):
 *
 * ***** END LICENSE BLOCK ***** */

/**************************************************************************************
 * Fixed-point HE-AAC decoder
 * Jon Recker ([e-mail])
 * February 2005
 *
 * imdct.c - inverse MDCT
 **************************************************************************************/

#include "coder.h"
#include "assembly.h"

#define RND_VAL		(1 << (FBITS_OUT_IMDCT-1))

#ifndef AAC_ENABLE_SBR

/**************************************************************************************
 * Function:    DecWindowOverlap
 *
 * Description: apply synthesis window, do overlap-add, clip to 16-bit PCM,
 *                for winSequence LONG-LONG
 *
 * Inputs:      input buffer (output of type-IV DCT)
 *              overlap buffer (saved from last time)
 *              number of channels
 *              window type (sin or KBD) for input buffer
 *              window type (sin or KBD) for overlap buffer
 *
 * Outputs:     one channel, one frame of 16-bit PCM, interleaved by nChans
 *
 * Return:      none
 *
 * Notes:       this processes one channel at a time, but skips every other sample in
 *                the output buffer (pcm) for stereo interleaving
 *              this should fit in registers on ARM
 *
 * TODO:        ARM5E version with saturating overlap/add (QADD)
 *              asm code with free pointer updates, better load scheduling
 **************************************************************************************/
static void DecWindowOverlap(int *buf0, int *over0, short *pcm0, int nChans, int winTypeCurr, int winTypePrev)
{
	int in, w0, w1, f0, f1;
	int *buf1, *over1;
	short *pcm1;
	const int *wndPrev, *wndCurr;

	buf0 += (1024 >> 1);
	buf1  = buf0  - 1;
	pcm1  = pcm0 + (1024 - 1) * nChans;
	over1 = over0 + 1024 - 1;

	wndPrev = (winTypePrev == 1 ? kbdWindow + kbdWindowOffset[1] : sinWindow + sinWindowOffset[1]);
	if (winTypeCurr == winTypePrev) {
		/* cut window loads in half since current and overlap sections use same symmetric window */
		do {
			w0 = *wndPrev++;
			w1 = *wndPrev++;
			in = *buf0++;

			f0 = MULSHIFT32(w0, in);
			f1 = MULSHIFT32(w1, in);

			in = *over0;
			*pcm0 = CLIPTOSHORT( (in - f0 + RND_VAL) >> FBITS_OUT_IMDCT );
			pcm0 += nChans;

			in = *over1;
			*pcm1 = CLIPTOSHORT( (in + f1 + RND_VAL) >> FBITS_OUT_IMDCT );
			pcm1 -= nChans;

			in = *buf1--;
			*over1-- = MULSHIFT32(w0, in);
			*over0++ = MULSHIFT32(w1, in);
		} while (over0 < over1);
	} else {
		/* different windows for current and overlap parts - should still fit in registers on ARM w/o stack spill */
		wndCurr = (winTypeCurr == 1 ? kbdWindow + kbdWindowOffset[1] : sinWindow + sinWindowOffset[1]);
		do {
			w0 = *wndPrev++;
			w1 = *wndPrev++;
			in = *buf0++;

			f0 = MULSHIFT32(w0, in);
			f1 = MULSHIFT32(w1, in);

			in = *over0;
			*pcm0 = CLIPTOSHORT( (in - f0 + RND_VAL) >> FBITS_OUT_IMDCT );
			pcm0 += nChans;

			in = *over1;
			*pcm1 = CLIPTOSHORT( (in + f1 + RND_VAL) >> FBITS_OUT_IMDCT );
			pcm1 -= nChans;

			w0 = *wndCurr++;
			w1 = *wndCurr++;
			in = *buf1--;

			*over1-- = MULSHIFT32(w0, in);
			*over0++ = MULSHIFT32(w1, in);
		} while (over0 < over1);
	}
}

/**************************************************************************************
 * Function:    DecWindowOverlapLongStart
 *
 * Description: apply synthesis window, do overlap-add, clip to 16-bit PCM,
 *                for winSequence LONG-START
 *
 * Inputs:      input buffer (output of type-IV DCT)
 *              overlap buffer (saved from last time)
 *              number of channels
 *              window type (sin or KBD) for input buffer
 *              window type (sin or KBD) for overlap buffer
 *
 * Outputs:     one channel, one frame of 16-bit PCM, interleaved by nChans
 *
 * Return:      none
 *
 * Notes:       this processes one channel at a time, but skips every other sample in
 *                the output buffer (pcm) for stereo interleaving
 *              this should fit in registers on ARM
 *
 * TODO:        ARM5E version with saturating overlap/add (QADD)
 *              asm code with free pointer updates, better load scheduling
 **************************************************************************************/
static void DecWindowOverlapLongStart(int *buf0, int *over0, short *pcm0, int nChans, int winTypeCurr, int winTypePrev)
{
	int i,  in, w0, w1, f0, f1;
	int *buf1, *over1;
	short *pcm1;
	const int *wndPrev, *wndCurr;

	buf0 += (1024 >> 1);
	buf1  = buf0  - 1;
	pcm1  = pcm0 + (1024 - 1) * nChans;
	over1 = over0 + 1024 - 1;

	wndPrev = (winTypePrev == 1 ? kbdWindow + kbdWindowOffset[1] : sinWindow + sinWindowOffset[1]);
	i = 448;	/* 2 outputs, 2 overlaps per loop */
	do {
		w0 = *wndPrev++;
		w1 = *wndPrev++;
		in = *buf0++;

		f0 = MULSHIFT32(w0, in);
		f1 = MULSHIFT32(w1, in);

		in = *over0;
		*pcm0 = CLIPTOSHORT( (in - f0 + RND_VAL) >> FBITS_OUT_IMDCT );
		pcm0 += nChans;

		in = *over1;
		*pcm1 = CLIPTOSHORT( (in + f1 + RND_VAL) >> FBITS_OUT_IMDCT );
		pcm1 -= nChans;

		in = *buf1--;

		*over1-- = 0;		/* Wn = 0 for n = (2047, 2046, ... 1600) */
		*over0++ = in >> 1;	/* Wn = 1 for n = (1024, 1025, ... 1471) */
	} while (--i);

	wndCurr = (winTypeCurr == 1 ? kbdWindow + kbdWindowOffset[0] : sinWindow + sinWindowOffset[0]);

	/* do 64 more loops - 2 outputs, 2 overlaps per loop */
	do {
		w0 = *wndPrev++;
		w1 = *wndPrev++;
		in = *buf0++;

		f0 = MULSHIFT32(w0, in);
		f1 = MULSHIFT32(w1, in);

		in = *over0;
		*pcm0 = CLIPTOSHORT( (in - f0 + RND_VAL) >> FBITS_OUT_IMDCT );
		pcm0 += nChans;

		in = *over1;
		*pcm1 = CLIPTOSHORT( (in + f1 + RND_VAL) >> FBITS_OUT_IMDCT );
		pcm1 -= nChans;

		w0 = *wndCurr++;	/* W[0], W[1], ... --> W[255], W[254], ... */
		w1 = *wndCurr++;	/* W[127], W[126], ... --> W[128], W[129], ... */
		in = *buf1--;

		*over1-- = MULSHIFT32(w0, in);	/* Wn = short window for n = (1599, 1598, ... , 1536) */
		*over0++ = MULSHIFT32(w1, in);	/* Wn = short window for n = (1472, 1473, ... , 1535) */
	} while (over0 < over1);
}

/**************************************************************************************
 * Function:    DecWindowOverlapLongStop
 *
 * Description: apply synthesis window, do overlap-add, clip to 16-bit PCM,
 *                for winSequence LONG-STOP
 *
 * Inputs:      input buffer (output of type-IV DCT)
 *              overlap buffer (saved from last time)
 *              number of channels
 *              window type (sin or KBD) for input buffer
 *              window type (sin or KBD) for overlap buffer
 *
 * Outputs:     one channel, one frame of 16-bit PCM, interleaved by nChans
 *
 * Return:      none
 *
 * Notes:       this processes one channel at a time, but skips every other sample in
 *                the output buffer (pcm) for stereo interleaving
 *              this should fit in registers on ARM
 *
 * TODO:        ARM5E version with saturating overlap/add (QADD)
 *              asm code with free pointer updates, better load scheduling
 **************************************************************************************/
static void DecWindowOverlapLongStop(int *buf0, int *over0, short *pcm0, int nChans, int winTypeCurr, int winTypePrev)
{
	int i, in, w0, w1, f0, f1;
	int *buf1, *over1;
	short *pcm1;
	const int *wndPrev, *wndCurr;

	buf0 += (1024 >> 1);
	buf1  = buf0  - 1;
	pcm1  = pcm0 + (1024 - 1) * nChans;
	over1 = over0 + 1024 - 1;

	wndPrev = (winTypePrev == 1 ? kbdWindow + kbdWindowOffset[0] : sinWindow + sinWindowOffset[0]);
	wndCurr = (winTypeCurr == 1 ? kbdWindow + kbdWindowOffset[1] : sinWindow + sinWindowOffset[1]);

	i = 448;	/* 2 outputs, 2 overlaps per loop */
	do {
		/* Wn = 0 for n = (0, 1, ... 447) */
		/* Wn = 1 for n = (576, 577, ... 1023) */
		in = *buf0++;
		f1 = in >> 1;	/* scale since skipping multiply by Q31 */

		in = *over0;
		*pcm0 = CLIPTOSHORT( (in + RND_VAL) >> FBITS_OUT_IMDCT );
		pcm0 += nChans;

		in = *over1;
		*pcm1 = CLIPTOSHORT( (in + f1 + RND_VAL) >> FBITS_OUT_IMDCT );
		pcm1 -= nChans;

		w0 = *wndCurr++;
		w1 = *wndCurr++;
		in = *buf1--;

		*over1-- = MULSHIFT32(w0, in);
		*over0++ = MULSHIFT32(w1, in);
	} while (--i);

	/* do 64 more loops - 2 outputs, 2 overlaps per loop */
	do {
		w0 = *wndPrev++;	/* W[0], W[1], ...W[63] */
		w1 = *wndPrev++;	/* W[127], W[126], ... W[64] */
		in = *buf0++;

		f0 = MULSHIFT32(w0, in);
		f1 = MULSHIFT32(w1, in);