2002-01-20 14:48:02 +00:00
|
|
|
/*
|
|
|
|
* Alpha optimized DSP utils
|
|
|
|
* Copyright (c) 2002 Falk Hueffner <falk@debian.org>
|
|
|
|
*
|
2002-05-25 22:45:33 +00:00
|
|
|
* This library is free software; you can redistribute it and/or
|
|
|
|
* modify it under the terms of the GNU Lesser General Public
|
|
|
|
* License as published by the Free Software Foundation; either
|
|
|
|
* version 2 of the License, or (at your option) any later version.
|
2002-01-20 14:48:02 +00:00
|
|
|
*
|
2002-05-25 22:45:33 +00:00
|
|
|
* This library is distributed in the hope that it will be useful,
|
2002-01-20 14:48:02 +00:00
|
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
2002-05-25 22:45:33 +00:00
|
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
|
* Lesser General Public License for more details.
|
2002-01-20 14:48:02 +00:00
|
|
|
*
|
2002-05-25 22:45:33 +00:00
|
|
|
* You should have received a copy of the GNU Lesser General Public
|
|
|
|
* License along with this library; if not, write to the Free Software
|
2006-01-12 22:43:26 +00:00
|
|
|
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
2002-01-20 14:48:02 +00:00
|
|
|
*/
|
|
|
|
|
|
|
|
#include "asm.h"
|
|
|
|
#include "../dsputil.h"
|
|
|
|
#include "../mpegvideo.h"
|
|
|
|
|
2003-12-15 09:21:28 +00:00
|
|
|
static void dct_unquantize_h263_intra_axp(MpegEncContext *s, DCTELEM *block,
|
2002-07-03 03:01:06 +00:00
|
|
|
int n, int qscale)
|
2002-01-20 14:48:02 +00:00
|
|
|
{
|
2002-07-03 03:01:06 +00:00
|
|
|
int i, n_coeffs;
|
|
|
|
uint64_t qmul, qadd;
|
|
|
|
uint64_t correction;
|
|
|
|
DCTELEM *orig_block = block;
|
2005-12-12 01:42:06 +00:00
|
|
|
DCTELEM block0; /* might not be used uninitialized */
|
2002-06-24 22:22:39 +00:00
|
|
|
|
2002-10-07 02:03:21 +00:00
|
|
|
qadd = WORD_VEC((qscale - 1) | 1);
|
|
|
|
qmul = qscale << 1;
|
2005-12-17 18:14:38 +00:00
|
|
|
/* This mask kills spill from negative subwords to the next subword. */
|
2002-10-07 02:03:21 +00:00
|
|
|
correction = WORD_VEC((qmul - 1) + 1); /* multiplication / addition */
|
|
|
|
|
2003-12-15 09:21:28 +00:00
|
|
|
if (!s->h263_aic) {
|
2005-12-17 18:14:38 +00:00
|
|
|
if (n < 4)
|
2003-12-15 09:21:28 +00:00
|
|
|
block0 = block[0] * s->y_dc_scale;
|
|
|
|
else
|
|
|
|
block0 = block[0] * s->c_dc_scale;
|
2002-01-20 14:48:02 +00:00
|
|
|
} else {
|
2003-12-15 09:21:28 +00:00
|
|
|
qadd = 0;
|
2002-01-20 14:48:02 +00:00
|
|
|
}
|
2005-12-17 18:14:38 +00:00
|
|
|
n_coeffs = 63; // does not always use zigzag table
|
2002-07-03 03:01:06 +00:00
|
|
|
|
2002-10-07 02:03:21 +00:00
|
|
|
for(i = 0; i <= n_coeffs; block += 4, i += 4) {
|
2002-07-03 03:01:06 +00:00
|
|
|
uint64_t levels, negmask, zeros, add;
|
|
|
|
|
|
|
|
levels = ldq(block);
|
|
|
|
if (levels == 0)
|
|
|
|
continue;
|
|
|
|
|
2002-07-03 23:30:32 +00:00
|
|
|
#ifdef __alpha_max__
|
|
|
|
/* I don't think the speed difference justifies runtime
|
|
|
|
detection. */
|
2002-07-03 03:01:06 +00:00
|
|
|
negmask = maxsw4(levels, -1); /* negative -> ffff (-1) */
|
|
|
|
negmask = minsw4(negmask, 0); /* positive -> 0000 (0) */
|
2002-07-03 23:30:32 +00:00
|
|
|
#else
|
|
|
|
negmask = cmpbge(WORD_VEC(0x7fff), levels);
|
|
|
|
negmask &= (negmask >> 1) | (1 << 7);
|
|
|
|
negmask = zap(-1, negmask);
|
|
|
|
#endif
|
2002-07-03 03:01:06 +00:00
|
|
|
|
|
|
|
zeros = cmpbge(0, levels);
|
|
|
|
zeros &= zeros >> 1;
|
|
|
|
/* zeros |= zeros << 1 is not needed since qadd <= 255, so
|
|
|
|
zapping the lower byte suffices. */
|
|
|
|
|
|
|
|
levels *= qmul;
|
|
|
|
levels -= correction & (negmask << 16);
|
|
|
|
|
|
|
|
/* Negate qadd for negative levels. */
|
|
|
|
add = qadd ^ negmask;
|
|
|
|
add += WORD_VEC(0x0001) & negmask;
|
|
|
|
/* Set qadd to 0 for levels == 0. */
|
|
|
|
add = zap(add, zeros);
|
|
|
|
|
|
|
|
levels += add;
|
|
|
|
|
|
|
|
stq(levels, block);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (s->mb_intra && !s->h263_aic)
|
|
|
|
orig_block[0] = block0;
|
2002-01-20 14:48:02 +00:00
|
|
|
}
|
|
|
|
|
2003-12-15 09:21:28 +00:00
|
|
|
static void dct_unquantize_h263_inter_axp(MpegEncContext *s, DCTELEM *block,
|
|
|
|
int n, int qscale)
|
|
|
|
{
|
|
|
|
int i, n_coeffs;
|
|
|
|
uint64_t qmul, qadd;
|
|
|
|
uint64_t correction;
|
|
|
|
|
|
|
|
qadd = WORD_VEC((qscale - 1) | 1);
|
|
|
|
qmul = qscale << 1;
|
2005-12-17 18:14:38 +00:00
|
|
|
/* This mask kills spill from negative subwords to the next subword. */
|
2003-12-15 09:21:28 +00:00
|
|
|
correction = WORD_VEC((qmul - 1) + 1); /* multiplication / addition */
|
|
|
|
|
|
|
|
n_coeffs = s->intra_scantable.raster_end[s->block_last_index[n]];
|
|
|
|
|
|
|
|
for(i = 0; i <= n_coeffs; block += 4, i += 4) {
|
|
|
|
uint64_t levels, negmask, zeros, add;
|
|
|
|
|
|
|
|
levels = ldq(block);
|
|
|
|
if (levels == 0)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
#ifdef __alpha_max__
|
|
|
|
/* I don't think the speed difference justifies runtime
|
|
|
|
detection. */
|
|
|
|
negmask = maxsw4(levels, -1); /* negative -> ffff (-1) */
|
|
|
|
negmask = minsw4(negmask, 0); /* positive -> 0000 (0) */
|
|
|
|
#else
|
|
|
|
negmask = cmpbge(WORD_VEC(0x7fff), levels);
|
|
|
|
negmask &= (negmask >> 1) | (1 << 7);
|
|
|
|
negmask = zap(-1, negmask);
|
|
|
|
#endif
|
|
|
|
|
|
|
|
zeros = cmpbge(0, levels);
|
|
|
|
zeros &= zeros >> 1;
|
|
|
|
/* zeros |= zeros << 1 is not needed since qadd <= 255, so
|
|
|
|
zapping the lower byte suffices. */
|
|
|
|
|
|
|
|
levels *= qmul;
|
|
|
|
levels -= correction & (negmask << 16);
|
|
|
|
|
|
|
|
/* Negate qadd for negative levels. */
|
|
|
|
add = qadd ^ negmask;
|
|
|
|
add += WORD_VEC(0x0001) & negmask;
|
|
|
|
/* Set qadd to 0 for levels == 0. */
|
|
|
|
add = zap(add, zeros);
|
|
|
|
|
|
|
|
levels += add;
|
|
|
|
|
|
|
|
stq(levels, block);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2002-01-20 14:48:02 +00:00
|
|
|
void MPV_common_init_axp(MpegEncContext *s)
|
|
|
|
{
|
2003-12-15 09:21:28 +00:00
|
|
|
s->dct_unquantize_h263_intra = dct_unquantize_h263_intra_axp;
|
|
|
|
s->dct_unquantize_h263_inter = dct_unquantize_h263_inter_axp;
|
2002-01-20 14:48:02 +00:00
|
|
|
}
|