2001-07-30 09:04:34 +00:00
|
|
|
/*
|
|
|
|
|
* The simplest mpeg encoder (well, it was the simplest!)
|
2009-01-19 15:46:40 +00:00
|
|
|
* Copyright (c) 2000,2001 Fabrice Bellard
|
2001-07-30 09:04:34 +00:00
|
|
|
*
|
2007-07-05 10:37:29 +00:00
|
|
|
* Optimized for ia32 CPUs by Nick Kurshev <nickols_k@mail.ru>
|
|
|
|
|
* h263, mpeg1, mpeg2 dequantizer & draw_edges by Michael Niedermayer <michaelni@gmx.at>
|
|
|
|
|
*
|
2006-10-07 15:30:46 +00:00
|
|
|
* This file is part of FFmpeg.
|
|
|
|
|
*
|
|
|
|
|
* FFmpeg is free software; you can redistribute it and/or
|
2002-05-25 22:45:33 +00:00
|
|
|
* modify it under the terms of the GNU Lesser General Public
|
|
|
|
|
* License as published by the Free Software Foundation; either
|
2006-10-07 15:30:46 +00:00
|
|
|
* version 2.1 of the License, or (at your option) any later version.
|
2001-07-30 09:04:34 +00:00
|
|
|
*
|
2006-10-07 15:30:46 +00:00
|
|
|
* FFmpeg is distributed in the hope that it will be useful,
|
2001-07-30 09:04:34 +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.
|
2001-07-30 09:04:34 +00:00
|
|
|
*
|
2002-05-25 22:45:33 +00:00
|
|
|
* You should have received a copy of the GNU Lesser General Public
|
2006-10-07 15:30:46 +00:00
|
|
|
* License along with FFmpeg; 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
|
2001-07-30 09:04:34 +00:00
|
|
|
*/
|
|
|
|
|
|
2010-09-08 15:07:14 +00:00
|
|
|
#include "libavutil/cpu.h"
|
2008-05-09 11:56:36 +00:00
|
|
|
#include "libavutil/x86_cpu.h"
|
|
|
|
|
#include "libavcodec/avcodec.h"
|
|
|
|
|
#include "libavcodec/dsputil.h"
|
|
|
|
|
#include "libavcodec/mpegvideo.h"
|
2007-11-27 22:23:34 +00:00
|
|
|
#include "dsputil_mmx.h"
|
2002-09-29 22:44:22 +00:00
|
|
|
|
2012-02-15 12:26:15 +02:00
|
|
|
extern uint16_t ff_inv_zigzag_direct16[64];
|
2002-01-13 04:59:37 +00:00
|
|
|
|
2001-11-03 02:22:48 +00:00
|
|
|
|
2003-12-14 02:57:00 +00:00
|
|
|
static void dct_unquantize_h263_intra_mmx(MpegEncContext *s,
|
2001-11-03 02:22:48 +00:00
|
|
|
DCTELEM *block, int n, int qscale)
|
|
|
|
|
{
|
2008-05-08 21:11:24 +00:00
|
|
|
x86_reg level, qmul, qadd, nCoeffs;
|
2002-09-29 22:44:22 +00:00
|
|
|
|
|
|
|
|
qmul = qscale << 1;
|
2001-11-03 02:22:48 +00:00
|
|
|
|
2003-12-07 01:33:45 +00:00
|
|
|
assert(s->block_last_index[n]>=0 || s->h263_aic);
|
2005-12-17 18:14:38 +00:00
|
|
|
|
2003-12-14 02:57:00 +00:00
|
|
|
if (!s->h263_aic) {
|
|
|
|
|
if (n < 4)
|
|
|
|
|
level = block[0] * s->y_dc_scale;
|
|
|
|
|
else
|
|
|
|
|
level = block[0] * s->c_dc_scale;
|
|
|
|
|
qadd = (qscale - 1) | 1;
|
|
|
|
|
}else{
|
|
|
|
|
qadd = 0;
|
|
|
|
|
level= block[0];
|
2001-11-03 02:22:48 +00:00
|
|
|
}
|
2003-12-14 02:57:00 +00:00
|
|
|
if(s->ac_pred)
|
|
|
|
|
nCoeffs=63;
|
|
|
|
|
else
|
|
|
|
|
nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
|
|
|
|
|
//printf("%d %d ", qmul, qadd);
|
2008-10-16 13:34:09 +00:00
|
|
|
__asm__ volatile(
|
2005-12-22 01:10:11 +00:00
|
|
|
"movd %1, %%mm6 \n\t" //qmul
|
|
|
|
|
"packssdw %%mm6, %%mm6 \n\t"
|
|
|
|
|
"packssdw %%mm6, %%mm6 \n\t"
|
|
|
|
|
"movd %2, %%mm5 \n\t" //qadd
|
|
|
|
|
"pxor %%mm7, %%mm7 \n\t"
|
|
|
|
|
"packssdw %%mm5, %%mm5 \n\t"
|
|
|
|
|
"packssdw %%mm5, %%mm5 \n\t"
|
|
|
|
|
"psubw %%mm5, %%mm7 \n\t"
|
|
|
|
|
"pxor %%mm4, %%mm4 \n\t"
|
2011-01-15 22:09:35 +00:00
|
|
|
".p2align 4 \n\t"
|
2005-12-22 01:10:11 +00:00
|
|
|
"1: \n\t"
|
|
|
|
|
"movq (%0, %3), %%mm0 \n\t"
|
|
|
|
|
"movq 8(%0, %3), %%mm1 \n\t"
|
|
|
|
|
|
|
|
|
|
"pmullw %%mm6, %%mm0 \n\t"
|
|
|
|
|
"pmullw %%mm6, %%mm1 \n\t"
|
|
|
|
|
|
|
|
|
|
"movq (%0, %3), %%mm2 \n\t"
|
|
|
|
|
"movq 8(%0, %3), %%mm3 \n\t"
|
|
|
|
|
|
|
|
|
|
"pcmpgtw %%mm4, %%mm2 \n\t" // block[i] < 0 ? -1 : 0
|
|
|
|
|
"pcmpgtw %%mm4, %%mm3 \n\t" // block[i] < 0 ? -1 : 0
|
|
|
|
|
|
|
|
|
|
"pxor %%mm2, %%mm0 \n\t"
|
|
|
|
|
"pxor %%mm3, %%mm1 \n\t"
|
|
|
|
|
|
|
|
|
|
"paddw %%mm7, %%mm0 \n\t"
|
|
|
|
|
"paddw %%mm7, %%mm1 \n\t"
|
|
|
|
|
|
|
|
|
|
"pxor %%mm0, %%mm2 \n\t"
|
|
|
|
|
"pxor %%mm1, %%mm3 \n\t"
|
|
|
|
|
|
|
|
|
|
"pcmpeqw %%mm7, %%mm0 \n\t" // block[i] == 0 ? -1 : 0
|
|
|
|
|
"pcmpeqw %%mm7, %%mm1 \n\t" // block[i] == 0 ? -1 : 0
|
|
|
|
|
|
|
|
|
|
"pandn %%mm2, %%mm0 \n\t"
|
|
|
|
|
"pandn %%mm3, %%mm1 \n\t"
|
|
|
|
|
|
|
|
|
|
"movq %%mm0, (%0, %3) \n\t"
|
|
|
|
|
"movq %%mm1, 8(%0, %3) \n\t"
|
|
|
|
|
|
|
|
|
|
"add $16, %3 \n\t"
|
|
|
|
|
"jng 1b \n\t"
|
2008-08-24 08:41:20 +00:00
|
|
|
::"r" (block+nCoeffs), "rm"(qmul), "rm" (qadd), "r" (2*(-nCoeffs))
|
2005-12-22 01:10:11 +00:00
|
|
|
: "memory"
|
|
|
|
|
);
|
2003-12-14 02:57:00 +00:00
|
|
|
block[0]= level;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
static void dct_unquantize_h263_inter_mmx(MpegEncContext *s,
|
|
|
|
|
DCTELEM *block, int n, int qscale)
|
|
|
|
|
{
|
2008-05-08 21:11:24 +00:00
|
|
|
x86_reg qmul, qadd, nCoeffs;
|
2003-12-14 02:57:00 +00:00
|
|
|
|
|
|
|
|
qmul = qscale << 1;
|
|
|
|
|
qadd = (qscale - 1) | 1;
|
|
|
|
|
|
|
|
|
|
assert(s->block_last_index[n]>=0 || s->h263_aic);
|
2005-12-17 18:14:38 +00:00
|
|
|
|
2003-12-14 02:57:00 +00:00
|
|
|
nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
|
2002-01-13 04:59:37 +00:00
|
|
|
//printf("%d %d ", qmul, qadd);
|
2008-10-16 13:34:09 +00:00
|
|
|
__asm__ volatile(
|
2005-12-22 01:10:11 +00:00
|
|
|
"movd %1, %%mm6 \n\t" //qmul
|
|
|
|
|
"packssdw %%mm6, %%mm6 \n\t"
|
|
|
|
|
"packssdw %%mm6, %%mm6 \n\t"
|
|
|
|
|
"movd %2, %%mm5 \n\t" //qadd
|
|
|
|
|
"pxor %%mm7, %%mm7 \n\t"
|
|
|
|
|
"packssdw %%mm5, %%mm5 \n\t"
|
|
|
|
|
"packssdw %%mm5, %%mm5 \n\t"
|
|
|
|
|
"psubw %%mm5, %%mm7 \n\t"
|
|
|
|
|
"pxor %%mm4, %%mm4 \n\t"
|
2011-01-15 22:09:35 +00:00
|
|
|
".p2align 4 \n\t"
|
2005-12-22 01:10:11 +00:00
|
|
|
"1: \n\t"
|
|
|
|
|
"movq (%0, %3), %%mm0 \n\t"
|
|
|
|
|
"movq 8(%0, %3), %%mm1 \n\t"
|
|
|
|
|
|
|
|
|
|
"pmullw %%mm6, %%mm0 \n\t"
|
|
|
|
|
"pmullw %%mm6, %%mm1 \n\t"
|
|
|
|
|
|
|
|
|
|
"movq (%0, %3), %%mm2 \n\t"
|
|
|
|
|
"movq 8(%0, %3), %%mm3 \n\t"
|
|
|
|
|
|
|
|
|
|
"pcmpgtw %%mm4, %%mm2 \n\t" // block[i] < 0 ? -1 : 0
|
|
|
|
|
"pcmpgtw %%mm4, %%mm3 \n\t" // block[i] < 0 ? -1 : 0
|
|
|
|
|
|
|
|
|
|
"pxor %%mm2, %%mm0 \n\t"
|
|
|
|
|
"pxor %%mm3, %%mm1 \n\t"
|
|
|
|
|
|
|
|
|
|
"paddw %%mm7, %%mm0 \n\t"
|
|
|
|
|
"paddw %%mm7, %%mm1 \n\t"
|
|
|
|
|
|
|
|
|
|
"pxor %%mm0, %%mm2 \n\t"
|
|
|
|
|
"pxor %%mm1, %%mm3 \n\t"
|
|
|
|
|
|
|
|
|
|
"pcmpeqw %%mm7, %%mm0 \n\t" // block[i] == 0 ? -1 : 0
|
|
|
|
|
"pcmpeqw %%mm7, %%mm1 \n\t" // block[i] == 0 ? -1 : 0
|
|
|
|
|
|
|
|
|
|
"pandn %%mm2, %%mm0 \n\t"
|
|
|
|
|
"pandn %%mm3, %%mm1 \n\t"
|
|
|
|
|
|
|
|
|
|
"movq %%mm0, (%0, %3) \n\t"
|
|
|
|
|
"movq %%mm1, 8(%0, %3) \n\t"
|
|
|
|
|
|
|
|
|
|
"add $16, %3 \n\t"
|
|
|
|
|
"jng 1b \n\t"
|
2008-08-24 08:41:20 +00:00
|
|
|
::"r" (block+nCoeffs), "rm"(qmul), "rm" (qadd), "r" (2*(-nCoeffs))
|
2005-12-22 01:10:11 +00:00
|
|
|
: "memory"
|
|
|
|
|
);
|
2001-11-03 02:22:48 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
2001-07-30 09:04:34 +00:00
|
|
|
/*
|
|
|
|
|
NK:
|
|
|
|
|
Note: looking at PARANOID:
|
|
|
|
|
"enable all paranoid tests for rounding, overflows, etc..."
|
|
|
|
|
|
|
|
|
|
#ifdef PARANOID
|
|
|
|
|
if (level < -2048 || level > 2047)
|
|
|
|
|
fprintf(stderr, "unquant error %d %d\n", i, level);
|
|
|
|
|
#endif
|
2007-12-01 22:21:04 +00:00
|
|
|
We can suppose that result of two multiplications can't be greater than 0xFFFF
|
2001-07-30 09:04:34 +00:00
|
|
|
i.e. is 16-bit, so we use here only PMULLW instruction and can avoid
|
|
|
|
|
a complex multiplication.
|
|
|
|
|
=====================================================
|
|
|
|
|
Full formula for multiplication of 2 integer numbers
|
|
|
|
|
which are represent as high:low words:
|
|
|
|
|
input: value1 = high1:low1
|
|
|
|
|
value2 = high2:low2
|
|
|
|
|
output: value3 = value1*value2
|
|
|
|
|
value3=high3:low3 (on overflow: modulus 2^32 wrap-around)
|
|
|
|
|
this mean that for 0x123456 * 0x123456 correct result is 0x766cb0ce4
|
|
|
|
|
but this algorithm will compute only 0x66cb0ce4
|
|
|
|
|
this limited by 16-bit size of operands
|
|
|
|
|
---------------------------------
|
|
|
|
|
tlow1 = high1*low2
|
|
|
|
|
tlow2 = high2*low1
|
|
|
|
|
tlow1 = tlow1 + tlow2
|
|
|
|
|
high3:low3 = low1*low2
|
|
|
|
|
high3 += tlow1
|
|
|
|
|
*/
|
2003-12-14 02:57:00 +00:00
|
|
|
static void dct_unquantize_mpeg1_intra_mmx(MpegEncContext *s,
|
2001-07-30 23:28:35 +00:00
|
|
|
DCTELEM *block, int n, int qscale)
|
2001-07-30 09:04:34 +00:00
|
|
|
{
|
2008-05-08 21:11:24 +00:00
|
|
|
x86_reg nCoeffs;
|
2003-02-11 16:35:48 +00:00
|
|
|
const uint16_t *quant_matrix;
|
2003-12-14 02:57:00 +00:00
|
|
|
int block0;
|
2002-09-29 22:44:22 +00:00
|
|
|
|
|
|
|
|
assert(s->block_last_index[n]>=0);
|
|
|
|
|
|
|
|
|
|
nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ]+1;
|
2002-01-13 04:59:37 +00:00
|
|
|
|
2005-12-17 18:14:38 +00:00
|
|
|
if (n < 4)
|
2003-12-14 02:57:00 +00:00
|
|
|
block0 = block[0] * s->y_dc_scale;
|
|
|
|
|
else
|
|
|
|
|
block0 = block[0] * s->c_dc_scale;
|
|
|
|
|
/* XXX: only mpeg1 */
|
|
|
|
|
quant_matrix = s->intra_matrix;
|
2008-10-16 13:34:09 +00:00
|
|
|
__asm__ volatile(
|
2005-12-22 01:10:11 +00:00
|
|
|
"pcmpeqw %%mm7, %%mm7 \n\t"
|
|
|
|
|
"psrlw $15, %%mm7 \n\t"
|
|
|
|
|
"movd %2, %%mm6 \n\t"
|
|
|
|
|
"packssdw %%mm6, %%mm6 \n\t"
|
|
|
|
|
"packssdw %%mm6, %%mm6 \n\t"
|
|
|
|
|
"mov %3, %%"REG_a" \n\t"
|
2011-01-15 22:09:35 +00:00
|
|
|
".p2align 4 \n\t"
|
2005-12-22 01:10:11 +00:00
|
|
|
"1: \n\t"
|
|
|
|
|
"movq (%0, %%"REG_a"), %%mm0 \n\t"
|
|
|
|
|
"movq 8(%0, %%"REG_a"), %%mm1 \n\t"
|
|
|
|
|
"movq (%1, %%"REG_a"), %%mm4 \n\t"
|
|
|
|
|
"movq 8(%1, %%"REG_a"), %%mm5 \n\t"
|
|
|
|
|
"pmullw %%mm6, %%mm4 \n\t" // q=qscale*quant_matrix[i]
|
|
|
|
|
"pmullw %%mm6, %%mm5 \n\t" // q=qscale*quant_matrix[i]
|
|
|
|
|
"pxor %%mm2, %%mm2 \n\t"
|
|
|
|
|
"pxor %%mm3, %%mm3 \n\t"
|
|
|
|
|
"pcmpgtw %%mm0, %%mm2 \n\t" // block[i] < 0 ? -1 : 0
|
|
|
|
|
"pcmpgtw %%mm1, %%mm3 \n\t" // block[i] < 0 ? -1 : 0
|
|
|
|
|
"pxor %%mm2, %%mm0 \n\t"
|
|
|
|
|
"pxor %%mm3, %%mm1 \n\t"
|
|
|
|
|
"psubw %%mm2, %%mm0 \n\t" // abs(block[i])
|
|
|
|
|
"psubw %%mm3, %%mm1 \n\t" // abs(block[i])
|
|
|
|
|
"pmullw %%mm4, %%mm0 \n\t" // abs(block[i])*q
|
|
|
|
|
"pmullw %%mm5, %%mm1 \n\t" // abs(block[i])*q
|
|
|
|
|
"pxor %%mm4, %%mm4 \n\t"
|
|
|
|
|
"pxor %%mm5, %%mm5 \n\t" // FIXME slow
|
|
|
|
|
"pcmpeqw (%0, %%"REG_a"), %%mm4 \n\t" // block[i] == 0 ? -1 : 0
|
|
|
|
|
"pcmpeqw 8(%0, %%"REG_a"), %%mm5\n\t" // block[i] == 0 ? -1 : 0
|
|
|
|
|
"psraw $3, %%mm0 \n\t"
|
|
|
|
|
"psraw $3, %%mm1 \n\t"
|
|
|
|
|
"psubw %%mm7, %%mm0 \n\t"
|
|
|
|
|
"psubw %%mm7, %%mm1 \n\t"
|
|
|
|
|
"por %%mm7, %%mm0 \n\t"
|
|
|
|
|
"por %%mm7, %%mm1 \n\t"
|
|
|
|
|
"pxor %%mm2, %%mm0 \n\t"
|
|
|
|
|
"pxor %%mm3, %%mm1 \n\t"
|
|
|
|
|
"psubw %%mm2, %%mm0 \n\t"
|
|
|
|
|
"psubw %%mm3, %%mm1 \n\t"
|
|
|
|
|
"pandn %%mm0, %%mm4 \n\t"
|
|
|
|
|
"pandn %%mm1, %%mm5 \n\t"
|
|
|
|
|
"movq %%mm4, (%0, %%"REG_a") \n\t"
|
|
|
|
|
"movq %%mm5, 8(%0, %%"REG_a") \n\t"
|
|
|
|
|
|
|
|
|
|
"add $16, %%"REG_a" \n\t"
|
|
|
|
|
"js 1b \n\t"
|
2008-08-24 08:41:20 +00:00
|
|
|
::"r" (block+nCoeffs), "r"(quant_matrix+nCoeffs), "rm" (qscale), "g" (-2*nCoeffs)
|
2005-12-22 01:10:11 +00:00
|
|
|
: "%"REG_a, "memory"
|
|
|
|
|
);
|
2003-12-14 02:57:00 +00:00
|
|
|
block[0]= block0;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void dct_unquantize_mpeg1_inter_mmx(MpegEncContext *s,
|
|
|
|
|
DCTELEM *block, int n, int qscale)
|
|
|
|
|
{
|
2008-05-08 21:11:24 +00:00
|
|
|
x86_reg nCoeffs;
|
2003-12-14 02:57:00 +00:00
|
|
|
const uint16_t *quant_matrix;
|
|
|
|
|
|
|
|
|
|
assert(s->block_last_index[n]>=0);
|
|
|
|
|
|
|
|
|
|
nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ]+1;
|
2002-04-17 16:30:31 +00:00
|
|
|
|
2002-04-27 12:30:26 +00:00
|
|
|
quant_matrix = s->inter_matrix;
|
2008-10-16 13:34:09 +00:00
|
|
|
__asm__ volatile(
|
2005-12-22 01:10:11 +00:00
|
|
|
"pcmpeqw %%mm7, %%mm7 \n\t"
|
|
|
|
|
"psrlw $15, %%mm7 \n\t"
|
|
|
|
|
"movd %2, %%mm6 \n\t"
|
|
|
|
|
"packssdw %%mm6, %%mm6 \n\t"
|
|
|
|
|
"packssdw %%mm6, %%mm6 \n\t"
|
|
|
|
|
"mov %3, %%"REG_a" \n\t"
|
2011-01-15 22:09:35 +00:00
|
|
|
".p2align 4 \n\t"
|
2005-12-22 01:10:11 +00:00
|
|
|
"1: \n\t"
|
|
|
|
|
"movq (%0, %%"REG_a"), %%mm0 \n\t"
|
|
|
|
|
"movq 8(%0, %%"REG_a"), %%mm1 \n\t"
|
|
|
|
|
"movq (%1, %%"REG_a"), %%mm4 \n\t"
|
|
|
|
|
"movq 8(%1, %%"REG_a"), %%mm5 \n\t"
|
|
|
|
|
"pmullw %%mm6, %%mm4 \n\t" // q=qscale*quant_matrix[i]
|
|
|
|
|
"pmullw %%mm6, %%mm5 \n\t" // q=qscale*quant_matrix[i]
|
|
|
|
|
"pxor %%mm2, %%mm2 \n\t"
|
|
|
|
|
"pxor %%mm3, %%mm3 \n\t"
|
|
|
|
|
"pcmpgtw %%mm0, %%mm2 \n\t" // block[i] < 0 ? -1 : 0
|
|
|
|
|
"pcmpgtw %%mm1, %%mm3 \n\t" // block[i] < 0 ? -1 : 0
|
|
|
|
|
"pxor %%mm2, %%mm0 \n\t"
|
|
|
|
|
"pxor %%mm3, %%mm1 \n\t"
|
|
|
|
|
"psubw %%mm2, %%mm0 \n\t" // abs(block[i])
|
|
|
|
|
"psubw %%mm3, %%mm1 \n\t" // abs(block[i])
|
|
|
|
|
"paddw %%mm0, %%mm0 \n\t" // abs(block[i])*2
|
|
|
|
|
"paddw %%mm1, %%mm1 \n\t" // abs(block[i])*2
|
|
|
|
|
"paddw %%mm7, %%mm0 \n\t" // abs(block[i])*2 + 1
|
|
|
|
|
"paddw %%mm7, %%mm1 \n\t" // abs(block[i])*2 + 1
|
|
|
|
|
"pmullw %%mm4, %%mm0 \n\t" // (abs(block[i])*2 + 1)*q
|
|
|
|
|
"pmullw %%mm5, %%mm1 \n\t" // (abs(block[i])*2 + 1)*q
|
|
|
|
|
"pxor %%mm4, %%mm4 \n\t"
|
|
|
|
|
"pxor %%mm5, %%mm5 \n\t" // FIXME slow
|
|
|
|
|
"pcmpeqw (%0, %%"REG_a"), %%mm4 \n\t" // block[i] == 0 ? -1 : 0
|
|
|
|
|
"pcmpeqw 8(%0, %%"REG_a"), %%mm5\n\t" // block[i] == 0 ? -1 : 0
|
|
|
|
|
"psraw $4, %%mm0 \n\t"
|
|
|
|
|
"psraw $4, %%mm1 \n\t"
|
|
|
|
|
"psubw %%mm7, %%mm0 \n\t"
|
|
|
|
|
"psubw %%mm7, %%mm1 \n\t"
|
|
|
|
|
"por %%mm7, %%mm0 \n\t"
|
|
|
|
|
"por %%mm7, %%mm1 \n\t"
|
|
|
|
|
"pxor %%mm2, %%mm0 \n\t"
|
|
|
|
|
"pxor %%mm3, %%mm1 \n\t"
|
|
|
|
|
"psubw %%mm2, %%mm0 \n\t"
|
|
|
|
|
"psubw %%mm3, %%mm1 \n\t"
|
|
|
|
|
"pandn %%mm0, %%mm4 \n\t"
|
|
|
|
|
"pandn %%mm1, %%mm5 \n\t"
|
|
|
|
|
"movq %%mm4, (%0, %%"REG_a") \n\t"
|
|
|
|
|
"movq %%mm5, 8(%0, %%"REG_a") \n\t"
|
|
|
|
|
|
|
|
|
|
"add $16, %%"REG_a" \n\t"
|
|
|
|
|
"js 1b \n\t"
|
2008-08-24 08:41:20 +00:00
|
|
|
::"r" (block+nCoeffs), "r"(quant_matrix+nCoeffs), "rm" (qscale), "g" (-2*nCoeffs)
|
2005-12-22 01:10:11 +00:00
|
|
|
: "%"REG_a, "memory"
|
|
|
|
|
);
|
2002-04-17 16:30:31 +00:00
|
|
|
}
|
|
|
|
|
|
2003-12-14 02:57:00 +00:00
|
|
|
static void dct_unquantize_mpeg2_intra_mmx(MpegEncContext *s,
|
2002-04-17 16:30:31 +00:00
|
|
|
DCTELEM *block, int n, int qscale)
|
|
|
|
|
{
|
2008-05-08 21:11:24 +00:00
|
|
|
x86_reg nCoeffs;
|
2003-02-11 16:35:48 +00:00
|
|
|
const uint16_t *quant_matrix;
|
2003-12-14 02:57:00 +00:00
|
|
|
int block0;
|
2005-12-17 18:14:38 +00:00
|
|
|
|
2002-09-29 22:44:22 +00:00
|
|
|
assert(s->block_last_index[n]>=0);
|
|
|
|
|
|
|
|
|
|
if(s->alternate_scan) nCoeffs= 63; //FIXME
|
|
|
|
|
else nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ];
|
2002-04-17 16:30:31 +00:00
|
|
|
|
2005-12-17 18:14:38 +00:00
|
|
|
if (n < 4)
|
2003-12-14 02:57:00 +00:00
|
|
|
block0 = block[0] * s->y_dc_scale;
|
|
|
|
|
else
|
|
|
|
|
block0 = block[0] * s->c_dc_scale;
|
|
|
|
|
quant_matrix = s->intra_matrix;
|
2008-10-16 13:34:09 +00:00
|
|
|
__asm__ volatile(
|
2005-12-22 01:10:11 +00:00
|
|
|
"pcmpeqw %%mm7, %%mm7 \n\t"
|
|
|
|
|
"psrlw $15, %%mm7 \n\t"
|
|
|
|
|
"movd %2, %%mm6 \n\t"
|
|
|
|
|
"packssdw %%mm6, %%mm6 \n\t"
|
|
|
|
|
"packssdw %%mm6, %%mm6 \n\t"
|
|
|
|
|
"mov %3, %%"REG_a" \n\t"
|
2011-01-15 22:09:35 +00:00
|
|
|
".p2align 4 \n\t"
|
2005-12-22 01:10:11 +00:00
|
|
|
"1: \n\t"
|
|
|
|
|
"movq (%0, %%"REG_a"), %%mm0 \n\t"
|
|
|
|
|
"movq 8(%0, %%"REG_a"), %%mm1 \n\t"
|
|
|
|
|
"movq (%1, %%"REG_a"), %%mm4 \n\t"
|
|
|
|
|
"movq 8(%1, %%"REG_a"), %%mm5 \n\t"
|
|
|
|
|
"pmullw %%mm6, %%mm4 \n\t" // q=qscale*quant_matrix[i]
|
|
|
|
|
"pmullw %%mm6, %%mm5 \n\t" // q=qscale*quant_matrix[i]
|
|
|
|
|
"pxor %%mm2, %%mm2 \n\t"
|
|
|
|
|
"pxor %%mm3, %%mm3 \n\t"
|
|
|
|
|
"pcmpgtw %%mm0, %%mm2 \n\t" // block[i] < 0 ? -1 : 0
|
|
|
|
|
"pcmpgtw %%mm1, %%mm3 \n\t" // block[i] < 0 ? -1 : 0
|
|
|
|
|
"pxor %%mm2, %%mm0 \n\t"
|
|
|
|
|
"pxor %%mm3, %%mm1 \n\t"
|
|
|
|
|
"psubw %%mm2, %%mm0 \n\t" // abs(block[i])
|
|
|
|
|
"psubw %%mm3, %%mm1 \n\t" // abs(block[i])
|
|
|
|
|
"pmullw %%mm4, %%mm0 \n\t" // abs(block[i])*q
|
|
|
|
|
"pmullw %%mm5, %%mm1 \n\t" // abs(block[i])*q
|
|
|
|
|
"pxor %%mm4, %%mm4 \n\t"
|
|
|
|
|
"pxor %%mm5, %%mm5 \n\t" // FIXME slow
|
|
|
|
|
"pcmpeqw (%0, %%"REG_a"), %%mm4 \n\t" // block[i] == 0 ? -1 : 0
|
|
|
|
|
"pcmpeqw 8(%0, %%"REG_a"), %%mm5\n\t" // block[i] == 0 ? -1 : 0
|
|
|
|
|
"psraw $3, %%mm0 \n\t"
|
|
|
|
|
"psraw $3, %%mm1 \n\t"
|
|
|
|
|
"pxor %%mm2, %%mm0 \n\t"
|
|
|
|
|
"pxor %%mm3, %%mm1 \n\t"
|
|
|
|
|
"psubw %%mm2, %%mm0 \n\t"
|
|
|
|
|
"psubw %%mm3, %%mm1 \n\t"
|
|
|
|
|
"pandn %%mm0, %%mm4 \n\t"
|
|
|
|
|
"pandn %%mm1, %%mm5 \n\t"
|
|
|
|
|
"movq %%mm4, (%0, %%"REG_a") \n\t"
|
|
|
|
|
"movq %%mm5, 8(%0, %%"REG_a") \n\t"
|
|
|
|
|
|
|
|
|
|
"add $16, %%"REG_a" \n\t"
|
|
|
|
|
"jng 1b \n\t"
|
2008-08-24 08:41:20 +00:00
|
|
|
::"r" (block+nCoeffs), "r"(quant_matrix+nCoeffs), "rm" (qscale), "g" (-2*nCoeffs)
|
2005-12-22 01:10:11 +00:00
|
|
|
: "%"REG_a, "memory"
|
|
|
|
|
);
|
2003-12-14 02:57:00 +00:00
|
|
|
block[0]= block0;
|
2007-06-12 09:29:25 +00:00
|
|
|
//Note, we do not do mismatch control for intra as errors cannot accumulate
|
2003-12-14 02:57:00 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void dct_unquantize_mpeg2_inter_mmx(MpegEncContext *s,
|
|
|
|
|
DCTELEM *block, int n, int qscale)
|
|
|
|
|
{
|
2008-05-08 21:11:24 +00:00
|
|
|
x86_reg nCoeffs;
|
2003-12-14 02:57:00 +00:00
|
|
|
const uint16_t *quant_matrix;
|
2005-12-17 18:14:38 +00:00
|
|
|
|
2003-12-14 02:57:00 +00:00
|
|
|
assert(s->block_last_index[n]>=0);
|
|
|
|
|
|
|
|
|
|
if(s->alternate_scan) nCoeffs= 63; //FIXME
|
|
|
|
|
else nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ];
|
2002-04-17 16:30:31 +00:00
|
|
|
|
2002-04-27 12:30:26 +00:00
|
|
|
quant_matrix = s->inter_matrix;
|
2008-10-16 13:34:09 +00:00
|
|
|
__asm__ volatile(
|
2005-12-22 01:10:11 +00:00
|
|
|
"pcmpeqw %%mm7, %%mm7 \n\t"
|
|
|
|
|
"psrlq $48, %%mm7 \n\t"
|
|
|
|
|
"movd %2, %%mm6 \n\t"
|
|
|
|
|
"packssdw %%mm6, %%mm6 \n\t"
|
|
|
|
|
"packssdw %%mm6, %%mm6 \n\t"
|
|
|
|
|
"mov %3, %%"REG_a" \n\t"
|
2011-01-15 22:09:35 +00:00
|
|
|
".p2align 4 \n\t"
|
2005-12-22 01:10:11 +00:00
|
|
|
"1: \n\t"
|
|
|
|
|
"movq (%0, %%"REG_a"), %%mm0 \n\t"
|
|
|
|
|
"movq 8(%0, %%"REG_a"), %%mm1 \n\t"
|
|
|
|
|
"movq (%1, %%"REG_a"), %%mm4 \n\t"
|
|
|
|
|
"movq 8(%1, %%"REG_a"), %%mm5 \n\t"
|
|
|
|
|
"pmullw %%mm6, %%mm4 \n\t" // q=qscale*quant_matrix[i]
|
|
|
|
|
"pmullw %%mm6, %%mm5 \n\t" // q=qscale*quant_matrix[i]
|
|
|
|
|
"pxor %%mm2, %%mm2 \n\t"
|
|
|
|
|
"pxor %%mm3, %%mm3 \n\t"
|
|
|
|
|
"pcmpgtw %%mm0, %%mm2 \n\t" // block[i] < 0 ? -1 : 0
|
|
|
|
|
"pcmpgtw %%mm1, %%mm3 \n\t" // block[i] < 0 ? -1 : 0
|
|
|
|
|
"pxor %%mm2, %%mm0 \n\t"
|
|
|
|
|
"pxor %%mm3, %%mm1 \n\t"
|
|
|
|
|
"psubw %%mm2, %%mm0 \n\t" // abs(block[i])
|
|
|
|
|
"psubw %%mm3, %%mm1 \n\t" // abs(block[i])
|
|
|
|
|
"paddw %%mm0, %%mm0 \n\t" // abs(block[i])*2
|
|
|
|
|
"paddw %%mm1, %%mm1 \n\t" // abs(block[i])*2
|
|
|
|
|
"pmullw %%mm4, %%mm0 \n\t" // abs(block[i])*2*q
|
|
|
|
|
"pmullw %%mm5, %%mm1 \n\t" // abs(block[i])*2*q
|
|
|
|
|
"paddw %%mm4, %%mm0 \n\t" // (abs(block[i])*2 + 1)*q
|
|
|
|
|
"paddw %%mm5, %%mm1 \n\t" // (abs(block[i])*2 + 1)*q
|
|
|
|
|
"pxor %%mm4, %%mm4 \n\t"
|
|
|
|
|
"pxor %%mm5, %%mm5 \n\t" // FIXME slow
|
|
|
|
|
"pcmpeqw (%0, %%"REG_a"), %%mm4 \n\t" // block[i] == 0 ? -1 : 0
|
|
|
|
|
"pcmpeqw 8(%0, %%"REG_a"), %%mm5\n\t" // block[i] == 0 ? -1 : 0
|
|
|
|
|
"psrlw $4, %%mm0 \n\t"
|
|
|
|
|
"psrlw $4, %%mm1 \n\t"
|
|
|
|
|
"pxor %%mm2, %%mm0 \n\t"
|
|
|
|
|
"pxor %%mm3, %%mm1 \n\t"
|
|
|
|
|
"psubw %%mm2, %%mm0 \n\t"
|
|
|
|
|
"psubw %%mm3, %%mm1 \n\t"
|
|
|
|
|
"pandn %%mm0, %%mm4 \n\t"
|
|
|
|
|
"pandn %%mm1, %%mm5 \n\t"
|
|
|
|
|
"pxor %%mm4, %%mm7 \n\t"
|
|
|
|
|
"pxor %%mm5, %%mm7 \n\t"
|
|
|
|
|
"movq %%mm4, (%0, %%"REG_a") \n\t"
|
|
|
|
|
"movq %%mm5, 8(%0, %%"REG_a") \n\t"
|
|
|
|
|
|
|
|
|
|
"add $16, %%"REG_a" \n\t"
|
|
|
|
|
"jng 1b \n\t"
|
|
|
|
|
"movd 124(%0, %3), %%mm0 \n\t"
|
|
|
|
|
"movq %%mm7, %%mm6 \n\t"
|
|
|
|
|
"psrlq $32, %%mm7 \n\t"
|
|
|
|
|
"pxor %%mm6, %%mm7 \n\t"
|
|
|
|
|
"movq %%mm7, %%mm6 \n\t"
|
|
|
|
|
"psrlq $16, %%mm7 \n\t"
|
|
|
|
|
"pxor %%mm6, %%mm7 \n\t"
|
|
|
|
|
"pslld $31, %%mm7 \n\t"
|
|
|
|
|
"psrlq $15, %%mm7 \n\t"
|
|
|
|
|
"pxor %%mm7, %%mm0 \n\t"
|
|
|
|
|
"movd %%mm0, 124(%0, %3) \n\t"
|
|
|
|
|
|
2008-08-24 08:41:20 +00:00
|
|
|
::"r" (block+nCoeffs), "r"(quant_matrix+nCoeffs), "rm" (qscale), "r" (-2*nCoeffs)
|
2005-12-22 01:10:11 +00:00
|
|
|
: "%"REG_a, "memory"
|
|
|
|
|
);
|
2001-07-30 09:04:34 +00:00
|
|
|
}
|
|
|
|
|
|
2004-01-02 22:29:08 +00:00
|
|
|
static void denoise_dct_mmx(MpegEncContext *s, DCTELEM *block){
|
|
|
|
|
const int intra= s->mb_intra;
|
|
|
|
|
int *sum= s->dct_error_sum[intra];
|
|
|
|
|
uint16_t *offset= s->dct_offset[intra];
|
|
|
|
|
|
|
|
|
|
s->dct_count[intra]++;
|
|
|
|
|
|
2008-10-16 13:34:09 +00:00
|
|
|
__asm__ volatile(
|
2005-12-22 01:10:11 +00:00
|
|
|
"pxor %%mm7, %%mm7 \n\t"
|
|
|
|
|
"1: \n\t"
|
|
|
|
|
"pxor %%mm0, %%mm0 \n\t"
|
|
|
|
|
"pxor %%mm1, %%mm1 \n\t"
|
|
|
|
|
"movq (%0), %%mm2 \n\t"
|
|
|
|
|
"movq 8(%0), %%mm3 \n\t"
|
|
|
|
|
"pcmpgtw %%mm2, %%mm0 \n\t"
|
|
|
|
|
"pcmpgtw %%mm3, %%mm1 \n\t"
|
|
|
|
|
"pxor %%mm0, %%mm2 \n\t"
|
|
|
|
|
"pxor %%mm1, %%mm3 \n\t"
|
|
|
|
|
"psubw %%mm0, %%mm2 \n\t"
|
|
|
|
|
"psubw %%mm1, %%mm3 \n\t"
|
|
|
|
|
"movq %%mm2, %%mm4 \n\t"
|
|
|
|
|
"movq %%mm3, %%mm5 \n\t"
|
|
|
|
|
"psubusw (%2), %%mm2 \n\t"
|
|
|
|
|
"psubusw 8(%2), %%mm3 \n\t"
|
|
|
|
|
"pxor %%mm0, %%mm2 \n\t"
|
|
|
|
|
"pxor %%mm1, %%mm3 \n\t"
|
|
|
|
|
"psubw %%mm0, %%mm2 \n\t"
|
|
|
|
|
"psubw %%mm1, %%mm3 \n\t"
|
|
|
|
|
"movq %%mm2, (%0) \n\t"
|
|
|
|
|
"movq %%mm3, 8(%0) \n\t"
|
|
|
|
|
"movq %%mm4, %%mm2 \n\t"
|
|
|
|
|
"movq %%mm5, %%mm3 \n\t"
|
|
|
|
|
"punpcklwd %%mm7, %%mm4 \n\t"
|
|
|
|
|
"punpckhwd %%mm7, %%mm2 \n\t"
|
|
|
|
|
"punpcklwd %%mm7, %%mm5 \n\t"
|
|
|
|
|
"punpckhwd %%mm7, %%mm3 \n\t"
|
|
|
|
|
"paddd (%1), %%mm4 \n\t"
|
|
|
|
|
"paddd 8(%1), %%mm2 \n\t"
|
|
|
|
|
"paddd 16(%1), %%mm5 \n\t"
|
|
|
|
|
"paddd 24(%1), %%mm3 \n\t"
|
|
|
|
|
"movq %%mm4, (%1) \n\t"
|
|
|
|
|
"movq %%mm2, 8(%1) \n\t"
|
|
|
|
|
"movq %%mm5, 16(%1) \n\t"
|
|
|
|
|
"movq %%mm3, 24(%1) \n\t"
|
|
|
|
|
"add $16, %0 \n\t"
|
|
|
|
|
"add $32, %1 \n\t"
|
|
|
|
|
"add $16, %2 \n\t"
|
|
|
|
|
"cmp %3, %0 \n\t"
|
|
|
|
|
" jb 1b \n\t"
|
2004-01-02 22:29:08 +00:00
|
|
|
: "+r" (block), "+r" (sum), "+r" (offset)
|
|
|
|
|
: "r"(block+64)
|
|
|
|
|
);
|
|
|
|
|
}
|
|
|
|
|
|
2004-01-03 20:14:45 +00:00
|
|
|
static void denoise_dct_sse2(MpegEncContext *s, DCTELEM *block){
|
|
|
|
|
const int intra= s->mb_intra;
|
|
|
|
|
int *sum= s->dct_error_sum[intra];
|
|
|
|
|
uint16_t *offset= s->dct_offset[intra];
|
|
|
|
|
|
|
|
|
|
s->dct_count[intra]++;
|
|
|
|
|
|
2008-10-16 13:34:09 +00:00
|
|
|
__asm__ volatile(
|
2005-12-22 01:10:11 +00:00
|
|
|
"pxor %%xmm7, %%xmm7 \n\t"
|
|
|
|
|
"1: \n\t"
|
|
|
|
|
"pxor %%xmm0, %%xmm0 \n\t"
|
|
|
|
|
"pxor %%xmm1, %%xmm1 \n\t"
|
|
|
|
|
"movdqa (%0), %%xmm2 \n\t"
|
|
|
|
|
"movdqa 16(%0), %%xmm3 \n\t"
|
|
|
|
|
"pcmpgtw %%xmm2, %%xmm0 \n\t"
|
|
|
|
|
"pcmpgtw %%xmm3, %%xmm1 \n\t"
|
|
|
|
|
"pxor %%xmm0, %%xmm2 \n\t"
|
|
|
|
|
"pxor %%xmm1, %%xmm3 \n\t"
|
|
|
|
|
"psubw %%xmm0, %%xmm2 \n\t"
|
|
|
|
|
"psubw %%xmm1, %%xmm3 \n\t"
|
|
|
|
|
"movdqa %%xmm2, %%xmm4 \n\t"
|
|
|
|
|
"movdqa %%xmm3, %%xmm5 \n\t"
|
|
|
|
|
"psubusw (%2), %%xmm2 \n\t"
|
|
|
|
|
"psubusw 16(%2), %%xmm3 \n\t"
|
|
|
|
|
"pxor %%xmm0, %%xmm2 \n\t"
|
|
|
|
|
"pxor %%xmm1, %%xmm3 \n\t"
|
|
|
|
|
"psubw %%xmm0, %%xmm2 \n\t"
|
|
|
|
|
"psubw %%xmm1, %%xmm3 \n\t"
|
|
|
|
|
"movdqa %%xmm2, (%0) \n\t"
|
|
|
|
|
"movdqa %%xmm3, 16(%0) \n\t"
|
|
|
|
|
"movdqa %%xmm4, %%xmm6 \n\t"
|
|
|
|
|
"movdqa %%xmm5, %%xmm0 \n\t"
|
|
|
|
|
"punpcklwd %%xmm7, %%xmm4 \n\t"
|
|
|
|
|
"punpckhwd %%xmm7, %%xmm6 \n\t"
|
|
|
|
|
"punpcklwd %%xmm7, %%xmm5 \n\t"
|
|
|
|
|
"punpckhwd %%xmm7, %%xmm0 \n\t"
|
|
|
|
|
"paddd (%1), %%xmm4 \n\t"
|
|
|
|
|
"paddd 16(%1), %%xmm6 \n\t"
|
|
|
|
|
"paddd 32(%1), %%xmm5 \n\t"
|
|
|
|
|
"paddd 48(%1), %%xmm0 \n\t"
|
|
|
|
|
"movdqa %%xmm4, (%1) \n\t"
|
|
|
|
|
"movdqa %%xmm6, 16(%1) \n\t"
|
|
|
|
|
"movdqa %%xmm5, 32(%1) \n\t"
|
|
|
|
|
"movdqa %%xmm0, 48(%1) \n\t"
|
|
|
|
|
"add $32, %0 \n\t"
|
|
|
|
|
"add $64, %1 \n\t"
|
|
|
|
|
"add $32, %2 \n\t"
|
|
|
|
|
"cmp %3, %0 \n\t"
|
|
|
|
|
" jb 1b \n\t"
|
2004-01-03 20:14:45 +00:00
|
|
|
: "+r" (block), "+r" (sum), "+r" (offset)
|
|
|
|
|
: "r"(block+64)
|
2010-10-31 13:15:16 +00:00
|
|
|
XMM_CLOBBERS_ONLY("%xmm0", "%xmm1", "%xmm2", "%xmm3",
|
|
|
|
|
"%xmm4", "%xmm5", "%xmm6", "%xmm7")
|
2004-01-03 20:14:45 +00:00
|
|
|
);
|
|
|
|
|
}
|
|
|
|
|
|
2009-01-13 23:44:16 +00:00
|
|
|
#if HAVE_SSSE3
|
2007-05-12 05:55:09 +00:00
|
|
|
#define HAVE_SSSE3_BAK
|
|
|
|
|
#endif
|
|
|
|
|
#undef HAVE_SSSE3
|
2009-01-13 23:44:16 +00:00
|
|
|
#define HAVE_SSSE3 0
|
2007-05-12 05:55:09 +00:00
|
|
|
|
|
|
|
|
#undef HAVE_SSE2
|
2002-01-27 13:30:18 +00:00
|
|
|
#undef HAVE_MMX2
|
2009-01-13 23:44:16 +00:00
|
|
|
#define HAVE_SSE2 0
|
|
|
|
|
#define HAVE_MMX2 0
|
2002-01-27 13:30:18 +00:00
|
|
|
#define RENAME(a) a ## _MMX
|
2003-10-23 13:22:20 +00:00
|
|
|
#define RENAMEl(a) a ## _mmx
|
2002-01-27 13:30:18 +00:00
|
|
|
#include "mpegvideo_mmx_template.c"
|
|
|
|
|
|
2009-01-13 23:44:16 +00:00
|
|
|
#undef HAVE_MMX2
|
|
|
|
|
#define HAVE_MMX2 1
|
2002-01-27 13:30:18 +00:00
|
|
|
#undef RENAME
|
2003-11-02 23:19:47 +00:00
|
|
|
#undef RENAMEl
|
2002-01-27 13:30:18 +00:00
|
|
|
#define RENAME(a) a ## _MMX2
|
2003-10-23 13:22:20 +00:00
|
|
|
#define RENAMEl(a) a ## _mmx2
|
2002-01-27 13:30:18 +00:00
|
|
|
#include "mpegvideo_mmx_template.c"
|
2002-01-15 04:07:29 +00:00
|
|
|
|
2009-01-13 23:44:16 +00:00
|
|
|
#undef HAVE_SSE2
|
|
|
|
|
#define HAVE_SSE2 1
|
2004-01-29 01:15:25 +00:00
|
|
|
#undef RENAME
|
|
|
|
|
#undef RENAMEl
|
|
|
|
|
#define RENAME(a) a ## _SSE2
|
|
|
|
|
#define RENAMEl(a) a ## _sse2
|
|
|
|
|
#include "mpegvideo_mmx_template.c"
|
|
|
|
|
|
2007-05-12 05:55:09 +00:00
|
|
|
#ifdef HAVE_SSSE3_BAK
|
2009-01-13 23:44:16 +00:00
|
|
|
#undef HAVE_SSSE3
|
|
|
|
|
#define HAVE_SSSE3 1
|
2007-05-12 05:55:09 +00:00
|
|
|
#undef RENAME
|
|
|
|
|
#undef RENAMEl
|
|
|
|
|
#define RENAME(a) a ## _SSSE3
|
|
|
|
|
#define RENAMEl(a) a ## _sse2
|
|
|
|
|
#include "mpegvideo_mmx_template.c"
|
|
|
|
|
#endif
|
|
|
|
|
|
2012-02-15 14:10:33 +02:00
|
|
|
void ff_MPV_common_init_mmx(MpegEncContext *s)
|
2001-07-30 09:04:34 +00:00
|
|
|
{
|
2010-09-08 15:07:14 +00:00
|
|
|
int mm_flags = av_get_cpu_flags();
|
2010-08-24 17:47:05 +00:00
|
|
|
|
2010-09-04 09:59:08 +00:00
|
|
|
if (mm_flags & AV_CPU_FLAG_MMX) {
|
2002-09-29 22:44:22 +00:00
|
|
|
const int dct_algo = s->avctx->dct_algo;
|
2005-12-17 18:14:38 +00:00
|
|
|
|
2003-12-14 02:57:00 +00:00
|
|
|
s->dct_unquantize_h263_intra = dct_unquantize_h263_intra_mmx;
|
|
|
|
|
s->dct_unquantize_h263_inter = dct_unquantize_h263_inter_mmx;
|
|
|
|
|
s->dct_unquantize_mpeg1_intra = dct_unquantize_mpeg1_intra_mmx;
|
|
|
|
|
s->dct_unquantize_mpeg1_inter = dct_unquantize_mpeg1_inter_mmx;
|
2006-04-28 17:03:52 +00:00
|
|
|
if(!(s->flags & CODEC_FLAG_BITEXACT))
|
|
|
|
|
s->dct_unquantize_mpeg2_intra = dct_unquantize_mpeg2_intra_mmx;
|
2003-12-14 02:57:00 +00:00
|
|
|
s->dct_unquantize_mpeg2_inter = dct_unquantize_mpeg2_inter_mmx;
|
2002-04-07 02:03:32 +00:00
|
|
|
|
2010-09-04 09:59:08 +00:00
|
|
|
if (mm_flags & AV_CPU_FLAG_SSE2) {
|
2005-12-22 01:10:11 +00:00
|
|
|
s->denoise_dct= denoise_dct_sse2;
|
|
|
|
|
} else {
|
|
|
|
|
s->denoise_dct= denoise_dct_mmx;
|
|
|
|
|
}
|
2002-01-27 13:30:18 +00:00
|
|
|
|
2002-08-29 23:55:32 +00:00
|
|
|
if(dct_algo==FF_DCT_AUTO || dct_algo==FF_DCT_MMX){
|
2009-01-13 23:44:16 +00:00
|
|
|
#if HAVE_SSSE3
|
2010-09-04 09:59:08 +00:00
|
|
|
if(mm_flags & AV_CPU_FLAG_SSSE3){
|
2007-05-12 05:55:09 +00:00
|
|
|
s->dct_quantize= dct_quantize_SSSE3;
|
|
|
|
|
} else
|
|
|
|
|
#endif
|
2010-09-04 09:59:08 +00:00
|
|
|
if(mm_flags & AV_CPU_FLAG_SSE2){
|
2004-01-29 01:15:25 +00:00
|
|
|
s->dct_quantize= dct_quantize_SSE2;
|
2010-09-04 09:59:08 +00:00
|
|
|
} else if(mm_flags & AV_CPU_FLAG_MMX2){
|
2002-08-29 23:55:32 +00:00
|
|
|
s->dct_quantize= dct_quantize_MMX2;
|
|
|
|
|
} else {
|
|
|
|
|
s->dct_quantize= dct_quantize_MMX;
|
|
|
|
|
}
|
2002-05-02 04:39:45 +00:00
|
|
|
}
|
2001-07-30 23:28:35 +00:00
|
|
|
}
|
2001-07-30 09:04:34 +00:00
|
|
|
}
|
