2001-10-11 00:13:27 +02:00
|
|
|
|
/*
|
|
|
|
|
Copyright (C) 2001 Michael Niedermayer (michaelni@gmx.at)
|
|
|
|
|
|
|
|
|
|
This program is free software; you can redistribute it and/or modify
|
|
|
|
|
it under the terms of the GNU General Public License as published by
|
|
|
|
|
the Free Software Foundation; either version 2 of the License, or
|
|
|
|
|
(at your option) any later version.
|
|
|
|
|
|
|
|
|
|
This program is distributed in the hope that it will be useful,
|
|
|
|
|
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
|
|
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
|
|
|
GNU General Public License for more details.
|
|
|
|
|
|
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
|
|
|
along with this program; if not, write to the Free Software
|
|
|
|
|
Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
|
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
/*
|
2001-10-15 05:01:08 +02:00
|
|
|
|
C MMX MMX2 3DNow
|
2001-10-11 00:13:27 +02:00
|
|
|
|
isVertDC Ec Ec
|
|
|
|
|
isVertMinMaxOk Ec Ec
|
2001-10-15 05:01:08 +02:00
|
|
|
|
doVertLowPass E e e
|
2001-10-11 00:13:27 +02:00
|
|
|
|
doVertDefFilter Ec Ec Ec
|
|
|
|
|
isHorizDC Ec Ec
|
|
|
|
|
isHorizMinMaxOk a
|
2001-10-15 05:01:08 +02:00
|
|
|
|
doHorizLowPass E a a
|
2001-10-11 00:21:19 +02:00
|
|
|
|
doHorizDefFilter E ac ac
|
2001-10-11 00:13:27 +02:00
|
|
|
|
deRing
|
2001-10-15 05:01:08 +02:00
|
|
|
|
Vertical RKAlgo1 E a a
|
|
|
|
|
Vertical X1 a E E
|
|
|
|
|
Horizontal X1 a E E
|
2001-10-17 22:42:07 +02:00
|
|
|
|
LinIpolDeinterlace e E E*
|
|
|
|
|
CubicIpolDeinterlace a e e*
|
|
|
|
|
LinBlendDeinterlace e E E*
|
2001-10-16 04:31:14 +02:00
|
|
|
|
MedianDeinterlace Ec Ec
|
2001-10-12 00:35:45 +02:00
|
|
|
|
|
2001-10-11 00:13:27 +02:00
|
|
|
|
|
2001-10-11 00:21:19 +02:00
|
|
|
|
* i dont have a 3dnow CPU -> its untested
|
2001-10-11 00:13:27 +02:00
|
|
|
|
E = Exact implementation
|
2001-10-17 22:42:07 +02:00
|
|
|
|
e = allmost exact implementation (slightly different rounding,...)
|
2001-10-11 00:13:27 +02:00
|
|
|
|
a = alternative / approximate impl
|
|
|
|
|
c = checked against the other implementations (-vo md5)
|
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
TODO:
|
2001-10-13 04:31:15 +02:00
|
|
|
|
verify that everything workes as it should (how?)
|
2001-10-11 00:13:27 +02:00
|
|
|
|
reduce the time wasted on the mem transfer
|
|
|
|
|
implement dering
|
2001-10-11 00:21:19 +02:00
|
|
|
|
implement everything in C at least (done at the moment but ...)
|
2001-10-11 00:13:27 +02:00
|
|
|
|
unroll stuff if instructions depend too much on the prior one
|
|
|
|
|
we use 8x8 blocks for the horizontal filters, opendivx seems to use 8x4?
|
|
|
|
|
move YScale thing to the end instead of fixing QP
|
2001-10-11 00:21:19 +02:00
|
|
|
|
write a faster and higher quality deblocking filter :)
|
2001-10-12 00:35:45 +02:00
|
|
|
|
do something about the speed of the horizontal filters
|
|
|
|
|
make the mainloop more flexible (variable number of blocks at once
|
|
|
|
|
(the if/else stuff per block is slowing things down)
|
2001-10-13 04:31:15 +02:00
|
|
|
|
compare the quality & speed of all filters
|
|
|
|
|
split this huge file
|
2001-10-15 05:01:08 +02:00
|
|
|
|
fix warnings (unused vars, ...)
|
2001-10-16 04:31:14 +02:00
|
|
|
|
noise reduction filters
|
2001-10-11 00:13:27 +02:00
|
|
|
|
...
|
|
|
|
|
|
|
|
|
|
Notes:
|
|
|
|
|
|
2001-10-11 00:21:19 +02:00
|
|
|
|
*/
|
|
|
|
|
|
2001-10-16 04:31:14 +02:00
|
|
|
|
//Changelog: use the CVS log
|
2001-10-11 00:13:27 +02:00
|
|
|
|
|
|
|
|
|
#include <inttypes.h>
|
|
|
|
|
#include <stdio.h>
|
2001-10-12 00:35:45 +02:00
|
|
|
|
#include <stdlib.h>
|
2001-10-23 01:36:35 +02:00
|
|
|
|
#include <string.h>
|
2001-10-11 00:13:27 +02:00
|
|
|
|
#include "../config.h"
|
|
|
|
|
//#undef HAVE_MMX2
|
2001-10-11 00:21:19 +02:00
|
|
|
|
//#define HAVE_3DNOW
|
2001-10-11 00:13:27 +02:00
|
|
|
|
//#undef HAVE_MMX
|
2001-10-11 00:21:19 +02:00
|
|
|
|
#include "postprocess.h"
|
2001-10-11 00:13:27 +02:00
|
|
|
|
|
2001-10-13 17:53:24 +02:00
|
|
|
|
#define MIN(a,b) ((a) > (b) ? (b) : (a))
|
|
|
|
|
#define MAX(a,b) ((a) < (b) ? (b) : (a))
|
|
|
|
|
#define ABS(a) ((a) > 0 ? (a) : (-(a)))
|
|
|
|
|
#define SIGN(a) ((a) > 0 ? 1 : -1)
|
|
|
|
|
|
|
|
|
|
#ifdef HAVE_MMX2
|
|
|
|
|
#define PAVGB(a,b) "pavgb " #a ", " #b " \n\t"
|
|
|
|
|
#elif defined (HAVE_3DNOW)
|
|
|
|
|
#define PAVGB(a,b) "pavgusb " #a ", " #b " \n\t"
|
|
|
|
|
#endif
|
2001-10-11 00:13:27 +02:00
|
|
|
|
|
2001-10-23 01:36:35 +02:00
|
|
|
|
#define GET_MODE_BUFFER_SIZE 500
|
|
|
|
|
#define OPTIONS_ARRAY_SIZE 10
|
|
|
|
|
|
|
|
|
|
|
2001-10-11 00:13:27 +02:00
|
|
|
|
static uint64_t packedYOffset= 0x0000000000000000LL;
|
|
|
|
|
static uint64_t packedYScale= 0x0100010001000100LL;
|
|
|
|
|
static uint64_t w05= 0x0005000500050005LL;
|
|
|
|
|
static uint64_t w20= 0x0020002000200020LL;
|
|
|
|
|
static uint64_t w1400= 0x1400140014001400LL;
|
|
|
|
|
static uint64_t bm00000001= 0x00000000000000FFLL;
|
|
|
|
|
static uint64_t bm00010000= 0x000000FF00000000LL;
|
|
|
|
|
static uint64_t bm00001000= 0x00000000FF000000LL;
|
|
|
|
|
static uint64_t bm10000000= 0xFF00000000000000LL;
|
|
|
|
|
static uint64_t bm10000001= 0xFF000000000000FFLL;
|
|
|
|
|
static uint64_t bm11000011= 0xFFFF00000000FFFFLL;
|
2001-10-11 00:21:19 +02:00
|
|
|
|
static uint64_t bm00000011= 0x000000000000FFFFLL;
|
2001-10-13 04:31:15 +02:00
|
|
|
|
static uint64_t bm11111110= 0xFFFFFFFFFFFFFF00LL;
|
2001-10-11 00:21:19 +02:00
|
|
|
|
static uint64_t bm11000000= 0xFFFF000000000000LL;
|
2001-10-11 00:13:27 +02:00
|
|
|
|
static uint64_t bm00011000= 0x000000FFFF000000LL;
|
|
|
|
|
static uint64_t bm00110011= 0x0000FFFF0000FFFFLL;
|
|
|
|
|
static uint64_t bm11001100= 0xFFFF0000FFFF0000LL;
|
|
|
|
|
static uint64_t b00= 0x0000000000000000LL;
|
2001-10-13 04:31:15 +02:00
|
|
|
|
static uint64_t b01= 0x0101010101010101LL;
|
2001-10-11 00:13:27 +02:00
|
|
|
|
static uint64_t b02= 0x0202020202020202LL;
|
|
|
|
|
static uint64_t b0F= 0x0F0F0F0F0F0F0F0FLL;
|
|
|
|
|
static uint64_t bFF= 0xFFFFFFFFFFFFFFFFLL;
|
2001-10-11 00:21:19 +02:00
|
|
|
|
static uint64_t b20= 0x2020202020202020LL;
|
|
|
|
|
static uint64_t b80= 0x8080808080808080LL;
|
2001-10-11 00:13:27 +02:00
|
|
|
|
static uint64_t b7E= 0x7E7E7E7E7E7E7E7ELL;
|
|
|
|
|
static uint64_t b7C= 0x7C7C7C7C7C7C7C7CLL;
|
|
|
|
|
static uint64_t b3F= 0x3F3F3F3F3F3F3F3FLL;
|
|
|
|
|
static uint64_t temp0=0;
|
|
|
|
|
static uint64_t temp1=0;
|
|
|
|
|
static uint64_t temp2=0;
|
|
|
|
|
static uint64_t temp3=0;
|
|
|
|
|
static uint64_t temp4=0;
|
|
|
|
|
static uint64_t temp5=0;
|
|
|
|
|
static uint64_t pQPb=0;
|
2001-10-19 15:41:38 +02:00
|
|
|
|
static uint8_t tempBlock[16*16]; //used so the horizontal code gets aligned data
|
2001-10-11 00:13:27 +02:00
|
|
|
|
|
|
|
|
|
int hFlatnessThreshold= 56 - 16;
|
|
|
|
|
int vFlatnessThreshold= 56 - 16;
|
|
|
|
|
|
|
|
|
|
//amount of "black" u r willing to loose to get a brightness corrected picture
|
|
|
|
|
double maxClippedThreshold= 0.01;
|
|
|
|
|
|
2001-10-23 01:36:35 +02:00
|
|
|
|
int maxAllowedY=234;
|
2001-10-19 15:41:38 +02:00
|
|
|
|
int minAllowedY=16;
|
2001-10-11 00:13:27 +02:00
|
|
|
|
|
2001-10-23 01:36:35 +02:00
|
|
|
|
static struct PPFilter filters[]=
|
|
|
|
|
{
|
|
|
|
|
{"hb", "hdeblock", 1, 1, 3, H_DEBLOCK},
|
|
|
|
|
{"vb", "vdeblock", 1, 2, 4, V_DEBLOCK},
|
|
|
|
|
{"vr", "rkvdeblock", 1, 2, 4, H_RK1_FILTER},
|
|
|
|
|
{"h1", "x1hdeblock", 1, 1, 3, H_X1_FILTER},
|
|
|
|
|
{"v1", "x1vdeblock", 1, 2, 4, V_X1_FILTER},
|
|
|
|
|
{"dr", "dering", 1, 5, 6, DERING},
|
|
|
|
|
{"al", "autolevels", 0, 1, 2, LEVEL_FIX},
|
|
|
|
|
{"lb", "linblenddeint", 0, 1, 6, LINEAR_BLEND_DEINT_FILTER},
|
|
|
|
|
{"li", "linipoldeint", 0, 1, 6, LINEAR_IPOL_DEINT_FILTER},
|
|
|
|
|
{"ci", "cubicipoldeint", 0, 1, 6, CUBIC_IPOL_DEINT_FILTER},
|
|
|
|
|
{"md", "mediandeint", 0, 1, 6, MEDIAN_DEINT_FILTER},
|
|
|
|
|
{NULL, NULL,0,0,0,0} //End Marker
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
static char *replaceTable[]=
|
|
|
|
|
{
|
|
|
|
|
"default", "hdeblock:a,vdeblock:a,dering:a,autolevels",
|
|
|
|
|
"de", "hdeblock:a,vdeblock:a,dering:a,autolevels",
|
|
|
|
|
"fast", "x1hdeblock:a,x1vdeblock:a,dering:a,autolevels",
|
|
|
|
|
"fa", "x1hdeblock:a,x1vdeblock:a,dering:a,autolevels",
|
|
|
|
|
NULL //End Marker
|
|
|
|
|
};
|
|
|
|
|
|
2001-10-16 04:31:14 +02:00
|
|
|
|
#ifdef TIMING
|
2001-10-11 00:13:27 +02:00
|
|
|
|
static inline long long rdtsc()
|
|
|
|
|
{
|
|
|
|
|
long long l;
|
|
|
|
|
asm volatile( "rdtsc\n\t"
|
|
|
|
|
: "=A" (l)
|
|
|
|
|
);
|
|
|
|
|
// printf("%d\n", int(l/1000));
|
|
|
|
|
return l;
|
|
|
|
|
}
|
2001-10-13 17:36:04 +02:00
|
|
|
|
#endif
|
2001-10-11 00:13:27 +02:00
|
|
|
|
|
2001-10-13 17:36:04 +02:00
|
|
|
|
#ifdef HAVE_MMX2
|
2001-10-11 00:13:27 +02:00
|
|
|
|
static inline void prefetchnta(void *p)
|
|
|
|
|
{
|
|
|
|
|
asm volatile( "prefetchnta (%0)\n\t"
|
|
|
|
|
: : "r" (p)
|
|
|
|
|
);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static inline void prefetcht0(void *p)
|
|
|
|
|
{
|
|
|
|
|
asm volatile( "prefetcht0 (%0)\n\t"
|
|
|
|
|
: : "r" (p)
|
|
|
|
|
);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static inline void prefetcht1(void *p)
|
|
|
|
|
{
|
|
|
|
|
asm volatile( "prefetcht1 (%0)\n\t"
|
|
|
|
|
: : "r" (p)
|
|
|
|
|
);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static inline void prefetcht2(void *p)
|
|
|
|
|
{
|
|
|
|
|
asm volatile( "prefetcht2 (%0)\n\t"
|
|
|
|
|
: : "r" (p)
|
|
|
|
|
);
|
|
|
|
|
}
|
2001-10-13 17:36:04 +02:00
|
|
|
|
#endif
|
2001-10-11 00:13:27 +02:00
|
|
|
|
|
|
|
|
|
//FIXME? |255-0| = 1 (shouldnt be a problem ...)
|
|
|
|
|
/**
|
2001-10-17 22:42:07 +02:00
|
|
|
|
* Check if the middle 8x8 Block in the given 8x16 block is flat
|
2001-10-11 00:13:27 +02:00
|
|
|
|
*/
|
2001-10-12 00:35:45 +02:00
|
|
|
|
static inline int isVertDC(uint8_t src[], int stride){
|
2001-10-11 00:13:27 +02:00
|
|
|
|
int numEq= 0;
|
2001-10-12 00:35:45 +02:00
|
|
|
|
int y;
|
2001-10-17 22:42:07 +02:00
|
|
|
|
src+= stride*4; // src points to begin of the 8x8 Block
|
2001-10-11 00:13:27 +02:00
|
|
|
|
#ifdef HAVE_MMX
|
|
|
|
|
asm volatile(
|
|
|
|
|
"pushl %1\n\t"
|
|
|
|
|
"movq b7E, %%mm7 \n\t" // mm7 = 0x7F
|
|
|
|
|
"movq b7C, %%mm6 \n\t" // mm6 = 0x7D
|
|
|
|
|
"movq (%1), %%mm0 \n\t"
|
|
|
|
|
"addl %2, %1 \n\t"
|
|
|
|
|
"movq (%1), %%mm1 \n\t"
|
|
|
|
|
"psubb %%mm1, %%mm0 \n\t" // mm0 = differnece
|
|
|
|
|
"paddb %%mm7, %%mm0 \n\t"
|
|
|
|
|
"pcmpgtb %%mm6, %%mm0 \n\t"
|
|
|
|
|
|
|
|
|
|
"addl %2, %1 \n\t"
|
|
|
|
|
"movq (%1), %%mm2 \n\t"
|
|
|
|
|
"psubb %%mm2, %%mm1 \n\t"
|
|
|
|
|
"paddb %%mm7, %%mm1 \n\t"
|
|
|
|
|
"pcmpgtb %%mm6, %%mm1 \n\t"
|
|
|
|
|
"paddb %%mm1, %%mm0 \n\t"
|
|
|
|
|
|
|
|
|
|
"addl %2, %1 \n\t"
|
|
|
|
|
"movq (%1), %%mm1 \n\t"
|
|
|
|
|
"psubb %%mm1, %%mm2 \n\t"
|
|
|
|
|
"paddb %%mm7, %%mm2 \n\t"
|
|
|
|
|
"pcmpgtb %%mm6, %%mm2 \n\t"
|
|
|
|
|
"paddb %%mm2, %%mm0 \n\t"
|
|
|
|
|
|
|
|
|
|
"addl %2, %1 \n\t"
|
|
|
|
|
"movq (%1), %%mm2 \n\t"
|
|
|
|
|
"psubb %%mm2, %%mm1 \n\t"
|
|
|
|
|
"paddb %%mm7, %%mm1 \n\t"
|
|
|
|
|
"pcmpgtb %%mm6, %%mm1 \n\t"
|
|
|
|
|
"paddb %%mm1, %%mm0 \n\t"
|
|
|
|
|
|
|
|
|
|
"addl %2, %1 \n\t"
|
|
|
|
|
"movq (%1), %%mm1 \n\t"
|
|
|
|
|
"psubb %%mm1, %%mm2 \n\t"
|
|
|
|
|
"paddb %%mm7, %%mm2 \n\t"
|
|
|
|
|
"pcmpgtb %%mm6, %%mm2 \n\t"
|
|
|
|
|
"paddb %%mm2, %%mm0 \n\t"
|
|
|
|
|
|
|
|
|
|
"addl %2, %1 \n\t"
|
|
|
|
|
"movq (%1), %%mm2 \n\t"
|
|
|
|
|
"psubb %%mm2, %%mm1 \n\t"
|
|
|
|
|
"paddb %%mm7, %%mm1 \n\t"
|
|
|
|
|
"pcmpgtb %%mm6, %%mm1 \n\t"
|
|
|
|
|
"paddb %%mm1, %%mm0 \n\t"
|
|
|
|
|
|
|
|
|
|
"addl %2, %1 \n\t"
|
|
|
|
|
"movq (%1), %%mm1 \n\t"
|
|
|
|
|
"psubb %%mm1, %%mm2 \n\t"
|
|
|
|
|
"paddb %%mm7, %%mm2 \n\t"
|
|
|
|
|
"pcmpgtb %%mm6, %%mm2 \n\t"
|
|
|
|
|
"paddb %%mm2, %%mm0 \n\t"
|
|
|
|
|
|
|
|
|
|
" \n\t"
|
|
|
|
|
"movq %%mm0, %%mm1 \n\t"
|
|
|
|
|
"psrlw $8, %%mm0 \n\t"
|
|
|
|
|
"paddb %%mm1, %%mm0 \n\t"
|
|
|
|
|
"movq %%mm0, %%mm1 \n\t"
|
|
|
|
|
"psrlq $16, %%mm0 \n\t"
|
|
|
|
|
"paddb %%mm1, %%mm0 \n\t"
|
|
|
|
|
"movq %%mm0, %%mm1 \n\t"
|
|
|
|
|
"psrlq $32, %%mm0 \n\t"
|
|
|
|
|
"paddb %%mm1, %%mm0 \n\t"
|
|
|
|
|
"popl %1\n\t"
|
|
|
|
|
"movd %%mm0, %0 \n\t"
|
|
|
|
|
: "=r" (numEq)
|
|
|
|
|
: "r" (src), "r" (stride)
|
|
|
|
|
);
|
|
|
|
|
// printf("%d\n", numEq);
|
|
|
|
|
numEq= (256 - (numEq & 0xFF)) &0xFF;
|
|
|
|
|
|
|
|
|
|
// int asmEq= numEq;
|
|
|
|
|
// numEq=0;
|
|
|
|
|
// uint8_t *temp= src;
|
|
|
|
|
|
|
|
|
|
#else
|
2001-10-12 00:35:45 +02:00
|
|
|
|
for(y=0; y<BLOCK_SIZE-1; y++)
|
2001-10-11 00:13:27 +02:00
|
|
|
|
{
|
|
|
|
|
if(((src[0] - src[0+stride] + 1)&0xFFFF) < 3) numEq++;
|
|
|
|
|
if(((src[1] - src[1+stride] + 1)&0xFFFF) < 3) numEq++;
|
|
|
|
|
if(((src[2] - src[2+stride] + 1)&0xFFFF) < 3) numEq++;
|
|
|
|
|
if(((src[3] - src[3+stride] + 1)&0xFFFF) < 3) numEq++;
|
|
|
|
|
if(((src[4] - src[4+stride] + 1)&0xFFFF) < 3) numEq++;
|
|
|
|
|
if(((src[5] - src[5+stride] + 1)&0xFFFF) < 3) numEq++;
|
|
|
|
|
if(((src[6] - src[6+stride] + 1)&0xFFFF) < 3) numEq++;
|
|
|
|
|
if(((src[7] - src[7+stride] + 1)&0xFFFF) < 3) numEq++;
|
|
|
|
|
src+= stride;
|
|
|
|
|
}
|
|
|
|
|
#endif
|
|
|
|
|
/* if(abs(numEq - asmEq) > 0)
|
|
|
|
|
{
|
|
|
|
|
printf("\nasm:%d c:%d\n", asmEq, numEq);
|
|
|
|
|
for(int y=0; y<8; y++)
|
|
|
|
|
{
|
|
|
|
|
for(int x=0; x<8; x++)
|
|
|
|
|
{
|
|
|
|
|
printf("%d ", temp[x + y*stride]);
|
|
|
|
|
}
|
|
|
|
|
printf("\n");
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
*/
|
2001-10-12 00:35:45 +02:00
|
|
|
|
// for(int i=0; i<numEq/8; i++) src[i]=255;
|
|
|
|
|
return (numEq > vFlatnessThreshold) ? 1 : 0;
|
2001-10-11 00:13:27 +02:00
|
|
|
|
}
|
|
|
|
|
|
2001-10-12 00:35:45 +02:00
|
|
|
|
static inline int isVertMinMaxOk(uint8_t src[], int stride, int QP)
|
2001-10-11 00:13:27 +02:00
|
|
|
|
{
|
|
|
|
|
#ifdef HAVE_MMX
|
|
|
|
|
int isOk;
|
2001-10-17 22:42:07 +02:00
|
|
|
|
src+= stride*3;
|
2001-10-11 00:13:27 +02:00
|
|
|
|
asm volatile(
|
|
|
|
|
// "int $3 \n\t"
|
|
|
|
|
"movq (%1, %2), %%mm0 \n\t"
|
|
|
|
|
"movq (%1, %2, 8), %%mm1 \n\t"
|
|
|
|
|
"movq %%mm0, %%mm2 \n\t"
|
|
|
|
|
"psubusb %%mm1, %%mm0 \n\t"
|
|
|
|
|
"psubusb %%mm2, %%mm1 \n\t"
|
|
|
|
|
"por %%mm1, %%mm0 \n\t" // ABS Diff
|
|
|
|
|
|
|
|
|
|
"movq pQPb, %%mm7 \n\t" // QP,..., QP
|
|
|
|
|
"paddusb %%mm7, %%mm7 \n\t" // 2QP ... 2QP
|
|
|
|
|
"psubusb %%mm7, %%mm0 \n\t" // Diff <= 2QP -> 0
|
|
|
|
|
"pcmpeqd b00, %%mm0 \n\t"
|
|
|
|
|
"psrlq $16, %%mm0 \n\t"
|
|
|
|
|
"pcmpeqd bFF, %%mm0 \n\t"
|
|
|
|
|
// "movd %%mm0, (%1, %2, 4)\n\t"
|
|
|
|
|
"movd %%mm0, %0 \n\t"
|
|
|
|
|
: "=r" (isOk)
|
|
|
|
|
: "r" (src), "r" (stride)
|
|
|
|
|
);
|
2001-10-12 00:35:45 +02:00
|
|
|
|
return isOk ? 1 : 0;
|
2001-10-11 00:13:27 +02:00
|
|
|
|
#else
|
|
|
|
|
|
2001-10-12 00:35:45 +02:00
|
|
|
|
int isOk2= 1;
|
|
|
|
|
int x;
|
2001-10-17 22:42:07 +02:00
|
|
|
|
src+= stride*3;
|
2001-10-12 00:35:45 +02:00
|
|
|
|
for(x=0; x<BLOCK_SIZE; x++)
|
2001-10-11 00:13:27 +02:00
|
|
|
|
{
|
2001-10-12 00:35:45 +02:00
|
|
|
|
if(abs((int)src[x + stride] - (int)src[x + (stride<<3)]) > 2*QP) isOk2=0;
|
2001-10-11 00:13:27 +02:00
|
|
|
|
}
|
|
|
|
|
/* if(isOk && !isOk2 || !isOk && isOk2)
|
|
|
|
|
{
|
|
|
|
|
printf("\nasm:%d c:%d QP:%d\n", isOk, isOk2, QP);
|
|
|
|
|
for(int y=0; y<9; y++)
|
|
|
|
|
{
|
|
|
|
|
for(int x=0; x<8; x++)
|
|
|
|
|
{
|
|
|
|
|
printf("%d ", src[x + y*stride]);
|
|
|
|
|
}
|
|
|
|
|
printf("\n");
|
|
|
|
|
}
|
|
|
|
|
} */
|
|
|
|
|
|
|
|
|
|
return isOk2;
|
|
|
|
|
#endif
|
|
|
|
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/**
|
2001-10-17 22:42:07 +02:00
|
|
|
|
* Do a vertical low pass filter on the 8x16 block (only write to the 8x8 block in the middle)
|
2001-10-16 04:31:14 +02:00
|
|
|
|
* using the 9-Tap Filter (1,1,2,2,4,2,2,1,1)/16
|
2001-10-11 00:13:27 +02:00
|
|
|
|
*/
|
|
|
|
|
static inline void doVertLowPass(uint8_t *src, int stride, int QP)
|
|
|
|
|
{
|
2001-10-11 00:21:19 +02:00
|
|
|
|
#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
|
2001-10-17 22:42:07 +02:00
|
|
|
|
src+= stride*3;
|
2001-10-11 00:13:27 +02:00
|
|
|
|
asm volatile( //"movv %0 %1 %2\n\t"
|
|
|
|
|
"pushl %0 \n\t"
|
|
|
|
|
"movq pQPb, %%mm0 \n\t" // QP,..., QP
|
|
|
|
|
|
|
|
|
|
"movq (%0), %%mm6 \n\t"
|
|
|
|
|
"movq (%0, %1), %%mm5 \n\t"
|
|
|
|
|
"movq %%mm5, %%mm1 \n\t"
|
|
|
|
|
"movq %%mm6, %%mm2 \n\t"
|
|
|
|
|
"psubusb %%mm6, %%mm5 \n\t"
|
|
|
|
|
"psubusb %%mm1, %%mm2 \n\t"
|
|
|
|
|
"por %%mm5, %%mm2 \n\t" // ABS Diff of lines
|
|
|
|
|
"psubusb %%mm0, %%mm2 \n\t" // diff <= QP -> 0
|
|
|
|
|
"pcmpeqb b00, %%mm2 \n\t" // diff <= QP -> FF
|
|
|
|
|
|
|
|
|
|
"pand %%mm2, %%mm6 \n\t"
|
|
|
|
|
"pandn %%mm1, %%mm2 \n\t"
|
|
|
|
|
"por %%mm2, %%mm6 \n\t"// First Line to Filter
|
|
|
|
|
|
|
|
|
|
"movq (%0, %1, 8), %%mm5 \n\t"
|
|
|
|
|
"leal (%0, %1, 4), %%eax \n\t"
|
|
|
|
|
"leal (%0, %1, 8), %%ebx \n\t"
|
|
|
|
|
"subl %1, %%ebx \n\t"
|
|
|
|
|
"addl %1, %0 \n\t" // %0 points to line 1 not 0
|
|
|
|
|
"movq (%0, %1, 8), %%mm7 \n\t"
|
|
|
|
|
"movq %%mm5, %%mm1 \n\t"
|
|
|
|
|
"movq %%mm7, %%mm2 \n\t"
|
|
|
|
|
"psubusb %%mm7, %%mm5 \n\t"
|
|
|
|
|
"psubusb %%mm1, %%mm2 \n\t"
|
|
|
|
|
"por %%mm5, %%mm2 \n\t" // ABS Diff of lines
|
|
|
|
|
"psubusb %%mm0, %%mm2 \n\t" // diff <= QP -> 0
|
|
|
|
|
"pcmpeqb b00, %%mm2 \n\t" // diff <= QP -> FF
|
|
|
|
|
|
|
|
|
|
"pand %%mm2, %%mm7 \n\t"
|
|
|
|
|
"pandn %%mm1, %%mm2 \n\t"
|
|
|
|
|
"por %%mm2, %%mm7 \n\t" // First Line to Filter
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
// 1 2 3 4 5 6 7 8
|
|
|
|
|
// %0 %0+%1 %0+2%1 eax %0+4%1 eax+2%1 ebx eax+4%1
|
|
|
|
|
// 6 4 2 2 1 1
|
|
|
|
|
// 6 4 4 2
|
|
|
|
|
// 6 8 2
|
2001-10-17 22:42:07 +02:00
|
|
|
|
|
2001-10-11 00:13:27 +02:00
|
|
|
|
"movq (%0, %1), %%mm0 \n\t" // 1
|
|
|
|
|
"movq %%mm0, %%mm1 \n\t" // 1
|
2001-10-11 00:21:19 +02:00
|
|
|
|
PAVGB(%%mm6, %%mm0) //1 1 /2
|
|
|
|
|
PAVGB(%%mm6, %%mm0) //3 1 /4
|
2001-10-11 00:13:27 +02:00
|
|
|
|
|
|
|
|
|
"movq (%0, %1, 4), %%mm2 \n\t" // 1
|
|
|
|
|
"movq %%mm2, %%mm5 \n\t" // 1
|
2001-10-11 00:21:19 +02:00
|
|
|
|
PAVGB((%%eax), %%mm2) // 11 /2
|
|
|
|
|
PAVGB((%0, %1, 2), %%mm2) // 211 /4
|
2001-10-11 00:13:27 +02:00
|
|
|
|
"movq %%mm2, %%mm3 \n\t" // 211 /4
|
|
|
|
|
"movq (%0), %%mm4 \n\t" // 1
|
2001-10-11 00:21:19 +02:00
|
|
|
|
PAVGB(%%mm4, %%mm3) // 4 211 /8
|
|
|
|
|
PAVGB(%%mm0, %%mm3) //642211 /16
|
2001-10-11 00:13:27 +02:00
|
|
|
|
"movq %%mm3, (%0) \n\t" // X
|
|
|
|
|
// mm1=2 mm2=3(211) mm4=1 mm5=5 mm6=0 mm7=9
|
|
|
|
|
"movq %%mm1, %%mm0 \n\t" // 1
|
2001-10-11 00:21:19 +02:00
|
|
|
|
PAVGB(%%mm6, %%mm0) //1 1 /2
|
2001-10-11 00:13:27 +02:00
|
|
|
|
"movq %%mm4, %%mm3 \n\t" // 1
|
2001-10-11 00:21:19 +02:00
|
|
|
|
PAVGB((%0,%1,2), %%mm3) // 1 1 /2
|
|
|
|
|
PAVGB((%%eax,%1,2), %%mm5) // 11 /2
|
|
|
|
|
PAVGB((%%eax), %%mm5) // 211 /4
|
|
|
|
|
PAVGB(%%mm5, %%mm3) // 2 2211 /8
|
|
|
|
|
PAVGB(%%mm0, %%mm3) //4242211 /16
|
2001-10-11 00:13:27 +02:00
|
|
|
|
"movq %%mm3, (%0,%1) \n\t" // X
|
|
|
|
|
// mm1=2 mm2=3(211) mm4=1 mm5=4(211) mm6=0 mm7=9
|
2001-10-11 00:21:19 +02:00
|
|
|
|
PAVGB(%%mm4, %%mm6) //11 /2
|
2001-10-11 00:13:27 +02:00
|
|
|
|
"movq (%%ebx), %%mm0 \n\t" // 1
|
2001-10-11 00:21:19 +02:00
|
|
|
|
PAVGB((%%eax, %1, 2), %%mm0) // 11/2
|
2001-10-11 00:13:27 +02:00
|
|
|
|
"movq %%mm0, %%mm3 \n\t" // 11/2
|
2001-10-11 00:21:19 +02:00
|
|
|
|
PAVGB(%%mm1, %%mm0) // 2 11/4
|
|
|
|
|
PAVGB(%%mm6, %%mm0) //222 11/8
|
|
|
|
|
PAVGB(%%mm2, %%mm0) //22242211/16
|
2001-10-11 00:13:27 +02:00
|
|
|
|
"movq (%0, %1, 2), %%mm2 \n\t" // 1
|
|
|
|
|
"movq %%mm0, (%0, %1, 2) \n\t" // X
|
|
|
|
|
// mm1=2 mm2=3 mm3=6(11) mm4=1 mm5=4(211) mm6=0(11) mm7=9
|
|
|
|
|
"movq (%%eax, %1, 4), %%mm0 \n\t" // 1
|
2001-10-11 00:21:19 +02:00
|
|
|
|
PAVGB((%%ebx), %%mm0) // 11 /2
|
|
|
|
|
PAVGB(%%mm0, %%mm6) //11 11 /4
|
|
|
|
|
PAVGB(%%mm1, %%mm4) // 11 /2
|
|
|
|
|
PAVGB(%%mm2, %%mm1) // 11 /2
|
|
|
|
|
PAVGB(%%mm1, %%mm6) //1122 11 /8
|
|
|
|
|
PAVGB(%%mm5, %%mm6) //112242211 /16
|
2001-10-11 00:13:27 +02:00
|
|
|
|
"movq (%%eax), %%mm5 \n\t" // 1
|
|
|
|
|
"movq %%mm6, (%%eax) \n\t" // X
|
|
|
|
|
// mm0=7(11) mm1=2(11) mm2=3 mm3=6(11) mm4=1(11) mm5=4 mm7=9
|
|
|
|
|
"movq (%%eax, %1, 4), %%mm6 \n\t" // 1
|
2001-10-11 00:21:19 +02:00
|
|
|
|
PAVGB(%%mm7, %%mm6) // 11 /2
|
|
|
|
|
PAVGB(%%mm4, %%mm6) // 11 11 /4
|
|
|
|
|
PAVGB(%%mm3, %%mm6) // 11 2211 /8
|
|
|
|
|
PAVGB(%%mm5, %%mm2) // 11 /2
|
2001-10-11 00:13:27 +02:00
|
|
|
|
"movq (%0, %1, 4), %%mm4 \n\t" // 1
|
2001-10-11 00:21:19 +02:00
|
|
|
|
PAVGB(%%mm4, %%mm2) // 112 /4
|
|
|
|
|
PAVGB(%%mm2, %%mm6) // 112242211 /16
|
2001-10-11 00:13:27 +02:00
|
|
|
|
"movq %%mm6, (%0, %1, 4) \n\t" // X
|
|
|
|
|
// mm0=7(11) mm1=2(11) mm2=3(112) mm3=6(11) mm4=5 mm5=4 mm7=9
|
2001-10-11 00:21:19 +02:00
|
|
|
|
PAVGB(%%mm7, %%mm1) // 11 2 /4
|
|
|
|
|
PAVGB(%%mm4, %%mm5) // 11 /2
|
|
|
|
|
PAVGB(%%mm5, %%mm0) // 11 11 /4
|
2001-10-11 00:13:27 +02:00
|
|
|
|
"movq (%%eax, %1, 2), %%mm6 \n\t" // 1
|
2001-10-11 00:21:19 +02:00
|
|
|
|
PAVGB(%%mm6, %%mm1) // 11 4 2 /8
|
|
|
|
|
PAVGB(%%mm0, %%mm1) // 11224222 /16
|
2001-10-11 00:13:27 +02:00
|
|
|
|
"movq %%mm1, (%%eax, %1, 2) \n\t" // X
|
|
|
|
|
// mm2=3(112) mm3=6(11) mm4=5 mm5=4(11) mm6=6 mm7=9
|
2001-10-11 00:21:19 +02:00
|
|
|
|
PAVGB((%%ebx), %%mm2) // 112 4 /8
|
2001-10-11 00:13:27 +02:00
|
|
|
|
"movq (%%eax, %1, 4), %%mm0 \n\t" // 1
|
2001-10-11 00:21:19 +02:00
|
|
|
|
PAVGB(%%mm0, %%mm6) // 1 1 /2
|
|
|
|
|
PAVGB(%%mm7, %%mm6) // 1 12 /4
|
|
|
|
|
PAVGB(%%mm2, %%mm6) // 1122424 /4
|
2001-10-11 00:13:27 +02:00
|
|
|
|
"movq %%mm6, (%%ebx) \n\t" // X
|
|
|
|
|
// mm0=8 mm3=6(11) mm4=5 mm5=4(11) mm7=9
|
2001-10-11 00:21:19 +02:00
|
|
|
|
PAVGB(%%mm7, %%mm5) // 11 2 /4
|
|
|
|
|
PAVGB(%%mm7, %%mm5) // 11 6 /8
|
2001-10-11 00:13:27 +02:00
|
|
|
|
|
2001-10-11 00:21:19 +02:00
|
|
|
|
PAVGB(%%mm3, %%mm0) // 112 /4
|
|
|
|
|
PAVGB(%%mm0, %%mm5) // 112246 /16
|
2001-10-11 00:13:27 +02:00
|
|
|
|
"movq %%mm5, (%%eax, %1, 4) \n\t" // X
|
|
|
|
|
"popl %0\n\t"
|
|
|
|
|
|
|
|
|
|
:
|
|
|
|
|
: "r" (src), "r" (stride)
|
|
|
|
|
: "%eax", "%ebx"
|
|
|
|
|
);
|
|
|
|
|
#else
|
|
|
|
|
const int l1= stride;
|
|
|
|
|
const int l2= stride + l1;
|
|
|
|
|
const int l3= stride + l2;
|
|
|
|
|
const int l4= stride + l3;
|
|
|
|
|
const int l5= stride + l4;
|
|
|
|
|
const int l6= stride + l5;
|
|
|
|
|
const int l7= stride + l6;
|
|
|
|
|
const int l8= stride + l7;
|
|
|
|
|
const int l9= stride + l8;
|
2001-10-12 00:35:45 +02:00
|
|
|
|
int x;
|
2001-10-17 22:42:07 +02:00
|
|
|
|
src+= stride*3;
|
2001-10-12 00:35:45 +02:00
|
|
|
|
for(x=0; x<BLOCK_SIZE; x++)
|
2001-10-11 00:13:27 +02:00
|
|
|
|
{
|
|
|
|
|
const int first= ABS(src[0] - src[l1]) < QP ? src[0] : src[l1];
|
|
|
|
|
const int last= ABS(src[l8] - src[l9]) < QP ? src[l9] : src[l8];
|
|
|
|
|
|
|
|
|
|
int sums[9];
|
|
|
|
|
sums[0] = first + src[l1];
|
|
|
|
|
sums[1] = src[l1] + src[l2];
|
|
|
|
|
sums[2] = src[l2] + src[l3];
|
|
|
|
|
sums[3] = src[l3] + src[l4];
|
|
|
|
|
sums[4] = src[l4] + src[l5];
|
|
|
|
|
sums[5] = src[l5] + src[l6];
|
|
|
|
|
sums[6] = src[l6] + src[l7];
|
|
|
|
|
sums[7] = src[l7] + src[l8];
|
|
|
|
|
sums[8] = src[l8] + last;
|
|
|
|
|
|
|
|
|
|
src[l1]= ((sums[0]<<2) + ((first + sums[2])<<1) + sums[4] + 8)>>4;
|
|
|
|
|
src[l2]= ((src[l2]<<2) + (first + sums[0] + sums[3]<<1) + sums[5] + 8)>>4;
|
|
|
|
|
src[l3]= ((src[l3]<<2) + (first + sums[1] + sums[4]<<1) + sums[6] + 8)>>4;
|
|
|
|
|
src[l4]= ((src[l4]<<2) + (sums[2] + sums[5]<<1) + sums[0] + sums[7] + 8)>>4;
|
|
|
|
|
src[l5]= ((src[l5]<<2) + (sums[3] + sums[6]<<1) + sums[1] + sums[8] + 8)>>4;
|
|
|
|
|
src[l6]= ((src[l6]<<2) + (last + sums[7] + sums[4]<<1) + sums[2] + 8)>>4;
|
|
|
|
|
src[l7]= ((last + src[l7]<<2) + (src[l8] + sums[5]<<1) + sums[3] + 8)>>4;
|
|
|
|
|
src[l8]= ((sums[8]<<2) + (last + sums[6]<<1) + sums[4] + 8)>>4;
|
|
|
|
|
|
|
|
|
|
src++;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
#endif
|
|
|
|
|
}
|
|
|
|
|
|
2001-10-11 00:21:19 +02:00
|
|
|
|
/**
|
|
|
|
|
* Experimental implementation of the filter (Algorithm 1) described in a paper from Ramkishor & Karandikar
|
|
|
|
|
* values are correctly clipped (MMX2)
|
|
|
|
|
* values are wraparound (C)
|
|
|
|
|
* conclusion: its fast, but introduces ugly horizontal patterns if there is a continious gradient
|
|
|
|
|
0 8 16 24
|
|
|
|
|
x = 8
|
|
|
|
|
x/2 = 4
|
|
|
|
|
x/8 = 1
|
|
|
|
|
1 12 12 23
|
|
|
|
|
*/
|
2001-10-13 04:31:15 +02:00
|
|
|
|
static inline void vertRK1Filter(uint8_t *src, int stride, int QP)
|
2001-10-11 00:21:19 +02:00
|
|
|
|
{
|
2001-10-12 00:35:45 +02:00
|
|
|
|
#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
|
2001-10-17 22:42:07 +02:00
|
|
|
|
src+= stride*3;
|
2001-10-11 00:21:19 +02:00
|
|
|
|
// FIXME rounding
|
|
|
|
|
asm volatile(
|
|
|
|
|
"pxor %%mm7, %%mm7 \n\t" // 0
|
|
|
|
|
"movq b80, %%mm6 \n\t" // MIN_SIGNED_BYTE
|
|
|
|
|
"leal (%0, %1), %%eax \n\t"
|
|
|
|
|
"leal (%%eax, %1, 4), %%ebx \n\t"
|
|
|
|
|
// 0 1 2 3 4 5 6 7 8 9
|
|
|
|
|
// %0 eax eax+%1 eax+2%1 %0+4%1 ebx ebx+%1 ebx+2%1 %0+8%1 ebx+4%1
|
|
|
|
|
"movq pQPb, %%mm0 \n\t" // QP,..., QP
|
|
|
|
|
"movq %%mm0, %%mm1 \n\t" // QP,..., QP
|
|
|
|
|
"paddusb b02, %%mm0 \n\t"
|
|
|
|
|
"psrlw $2, %%mm0 \n\t"
|
|
|
|
|
"pand b3F, %%mm0 \n\t" // QP/4,..., QP/4
|
|
|
|
|
"paddusb %%mm1, %%mm0 \n\t" // QP*1.25 ...
|
|
|
|
|
"movq (%0, %1, 4), %%mm2 \n\t" // line 4
|
|
|
|
|
"movq (%%ebx), %%mm3 \n\t" // line 5
|
|
|
|
|
"movq %%mm2, %%mm4 \n\t" // line 4
|
|
|
|
|
"pcmpeqb %%mm5, %%mm5 \n\t" // -1
|
|
|
|
|
"pxor %%mm2, %%mm5 \n\t" // -line 4 - 1
|
2001-10-12 00:35:45 +02:00
|
|
|
|
PAVGB(%%mm3, %%mm5)
|
2001-10-11 00:21:19 +02:00
|
|
|
|
"paddb %%mm6, %%mm5 \n\t" // (l5-l4)/2
|
|
|
|
|
"psubusb %%mm3, %%mm4 \n\t"
|
|
|
|
|
"psubusb %%mm2, %%mm3 \n\t"
|
|
|
|
|
"por %%mm3, %%mm4 \n\t" // |l4 - l5|
|
|
|
|
|
"psubusb %%mm0, %%mm4 \n\t"
|
|
|
|
|
"pcmpeqb %%mm7, %%mm4 \n\t"
|
|
|
|
|
"pand %%mm4, %%mm5 \n\t" // d/2
|
|
|
|
|
|
|
|
|
|
// "paddb %%mm6, %%mm2 \n\t" // line 4 + 0x80
|
|
|
|
|
"paddb %%mm5, %%mm2 \n\t"
|
|
|
|
|
// "psubb %%mm6, %%mm2 \n\t"
|
|
|
|
|
"movq %%mm2, (%0,%1, 4) \n\t"
|
|
|
|
|
|
|
|
|
|
"movq (%%ebx), %%mm2 \n\t"
|
|
|
|
|
// "paddb %%mm6, %%mm2 \n\t" // line 5 + 0x80
|
|
|
|
|
"psubb %%mm5, %%mm2 \n\t"
|
|
|
|
|
// "psubb %%mm6, %%mm2 \n\t"
|
|
|
|
|
"movq %%mm2, (%%ebx) \n\t"
|
|
|
|
|
|
|
|
|
|
"paddb %%mm6, %%mm5 \n\t"
|
|
|
|
|
"psrlw $2, %%mm5 \n\t"
|
|
|
|
|
"pand b3F, %%mm5 \n\t"
|
|
|
|
|
"psubb b20, %%mm5 \n\t" // (l5-l4)/8
|
|
|
|
|
|
|
|
|
|
"movq (%%eax, %1, 2), %%mm2 \n\t"
|
|
|
|
|
"paddb %%mm6, %%mm2 \n\t" // line 3 + 0x80
|
|
|
|
|
"paddsb %%mm5, %%mm2 \n\t"
|
|
|
|
|
"psubb %%mm6, %%mm2 \n\t"
|
|
|
|
|
"movq %%mm2, (%%eax, %1, 2) \n\t"
|
|
|
|
|
|
|
|
|
|
"movq (%%ebx, %1), %%mm2 \n\t"
|
|
|
|
|
"paddb %%mm6, %%mm2 \n\t" // line 6 + 0x80
|
|
|
|
|
"psubsb %%mm5, %%mm2 \n\t"
|
|
|
|
|
"psubb %%mm6, %%mm2 \n\t"
|
|
|
|
|
"movq %%mm2, (%%ebx, %1) \n\t"
|
|
|
|
|
|
|
|
|
|
:
|
|
|
|
|
: "r" (src), "r" (stride)
|
|
|
|
|
: "%eax", "%ebx"
|
|
|
|
|
);
|
|
|
|
|
#else
|
|
|
|
|
const int l1= stride;
|
|
|
|
|
const int l2= stride + l1;
|
|
|
|
|
const int l3= stride + l2;
|
|
|
|
|
const int l4= stride + l3;
|
|
|
|
|
const int l5= stride + l4;
|
|
|
|
|
const int l6= stride + l5;
|
|
|
|
|
const int l7= stride + l6;
|
|
|
|
|
const int l8= stride + l7;
|
|
|
|
|
const int l9= stride + l8;
|
2001-10-12 00:35:45 +02:00
|
|
|
|
int x;
|
2001-10-17 22:42:07 +02:00
|
|
|
|
src+= stride*3;
|
2001-10-12 00:35:45 +02:00
|
|
|
|
for(x=0; x<BLOCK_SIZE; x++)
|
2001-10-11 00:21:19 +02:00
|
|
|
|
{
|
|
|
|
|
if(ABS(src[l4]-src[l5]) < QP + QP/4)
|
|
|
|
|
{
|
2001-10-12 00:35:45 +02:00
|
|
|
|
int v = (src[l5] - src[l4]);
|
|
|
|
|
|
|
|
|
|
src[l3] +=v/8;
|
|
|
|
|
src[l4] +=v/2;
|
|
|
|
|
src[l5] -=v/2;
|
|
|
|
|
src[l6] -=v/8;
|
2001-10-11 00:21:19 +02:00
|
|
|
|
|
|
|
|
|
}
|
|
|
|
|
src++;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
#endif
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Experimental Filter 1
|
2001-10-13 04:31:15 +02:00
|
|
|
|
* will not damage linear gradients
|
|
|
|
|
* Flat blocks should look like they where passed through the (1,1,2,2,4,2,2,1,1) 9-Tap filter
|
2001-10-12 00:35:45 +02:00
|
|
|
|
* can only smooth blocks at the expected locations (it cant smooth them if they did move)
|
|
|
|
|
* MMX2 version does correct clipping C version doesnt
|
2001-10-11 00:21:19 +02:00
|
|
|
|
*/
|
|
|
|
|
static inline void vertX1Filter(uint8_t *src, int stride, int QP)
|
|
|
|
|
{
|
2001-10-12 00:35:45 +02:00
|
|
|
|
#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
|
2001-10-17 22:42:07 +02:00
|
|
|
|
src+= stride*3;
|
|
|
|
|
|
2001-10-11 00:21:19 +02:00
|
|
|
|
asm volatile(
|
2001-10-12 00:35:45 +02:00
|
|
|
|
"pxor %%mm7, %%mm7 \n\t" // 0
|
|
|
|
|
// "movq b80, %%mm6 \n\t" // MIN_SIGNED_BYTE
|
|
|
|
|
"leal (%0, %1), %%eax \n\t"
|
|
|
|
|
"leal (%%eax, %1, 4), %%ebx \n\t"
|
|
|
|
|
// 0 1 2 3 4 5 6 7 8 9
|
|
|
|
|
// %0 eax eax+%1 eax+2%1 %0+4%1 ebx ebx+%1 ebx+2%1 %0+8%1 ebx+4%1
|
|
|
|
|
"movq (%%eax, %1, 2), %%mm0 \n\t" // line 3
|
|
|
|
|
"movq (%0, %1, 4), %%mm1 \n\t" // line 4
|
|
|
|
|
"movq %%mm1, %%mm2 \n\t" // line 4
|
|
|
|
|
"psubusb %%mm0, %%mm1 \n\t"
|
|
|
|
|
"psubusb %%mm2, %%mm0 \n\t"
|
|
|
|
|
"por %%mm1, %%mm0 \n\t" // |l2 - l3|
|
|
|
|
|
"movq (%%ebx), %%mm3 \n\t" // line 5
|
|
|
|
|
"movq (%%ebx, %1), %%mm4 \n\t" // line 6
|
|
|
|
|
"movq %%mm3, %%mm5 \n\t" // line 5
|
|
|
|
|
"psubusb %%mm4, %%mm3 \n\t"
|
|
|
|
|
"psubusb %%mm5, %%mm4 \n\t"
|
|
|
|
|
"por %%mm4, %%mm3 \n\t" // |l5 - l6|
|
|
|
|
|
PAVGB(%%mm3, %%mm0) // (|l2 - l3| + |l5 - l6|)/2
|
|
|
|
|
"movq %%mm2, %%mm1 \n\t" // line 4
|
|
|
|
|
"psubusb %%mm5, %%mm2 \n\t"
|
|
|
|
|
"movq %%mm2, %%mm4 \n\t"
|
|
|
|
|
"pcmpeqb %%mm7, %%mm2 \n\t" // (l4 - l5) <= 0 ? -1 : 0
|
|
|
|
|
"psubusb %%mm1, %%mm5 \n\t"
|
|
|
|
|
"por %%mm5, %%mm4 \n\t" // |l4 - l5|
|
|
|
|
|
"psubusb %%mm0, %%mm4 \n\t" //d = MAX(0, |l4-l5| - (|l2-l3| + |l5-l6|)/2)
|
|
|
|
|
"movq %%mm4, %%mm3 \n\t" // d
|
|
|
|
|
"psubusb pQPb, %%mm4 \n\t"
|
|
|
|
|
"pcmpeqb %%mm7, %%mm4 \n\t" // d <= QP ? -1 : 0
|
2001-10-13 04:31:15 +02:00
|
|
|
|
"psubusb b01, %%mm3 \n\t"
|
2001-10-12 00:35:45 +02:00
|
|
|
|
"pand %%mm4, %%mm3 \n\t" // d <= QP ? d : 0
|
|
|
|
|
|
|
|
|
|
PAVGB(%%mm7, %%mm3) // d/2
|
2001-10-13 04:31:15 +02:00
|
|
|
|
"movq %%mm3, %%mm1 \n\t" // d/2
|
|
|
|
|
PAVGB(%%mm7, %%mm3) // d/4
|
|
|
|
|
PAVGB(%%mm1, %%mm3) // 3*d/8
|
2001-10-12 00:35:45 +02:00
|
|
|
|
|
|
|
|
|
"movq (%0, %1, 4), %%mm0 \n\t" // line 4
|
|
|
|
|
"pxor %%mm2, %%mm0 \n\t" //(l4 - l5) <= 0 ? -l4-1 : l4
|
|
|
|
|
"psubusb %%mm3, %%mm0 \n\t"
|
|
|
|
|
"pxor %%mm2, %%mm0 \n\t"
|
|
|
|
|
"movq %%mm0, (%0, %1, 4) \n\t" // line 4
|
|
|
|
|
|
|
|
|
|
"movq (%%ebx), %%mm0 \n\t" // line 5
|
|
|
|
|
"pxor %%mm2, %%mm0 \n\t" //(l4 - l5) <= 0 ? -l5-1 : l5
|
|
|
|
|
"paddusb %%mm3, %%mm0 \n\t"
|
|
|
|
|
"pxor %%mm2, %%mm0 \n\t"
|
|
|
|
|
"movq %%mm0, (%%ebx) \n\t" // line 5
|
|
|
|
|
|
2001-10-13 04:31:15 +02:00
|
|
|
|
PAVGB(%%mm7, %%mm1) // d/4
|
2001-10-12 00:35:45 +02:00
|
|
|
|
|
|
|
|
|
"movq (%%eax, %1, 2), %%mm0 \n\t" // line 3
|
|
|
|
|
"pxor %%mm2, %%mm0 \n\t" //(l4 - l5) <= 0 ? -l4-1 : l4
|
2001-10-13 04:31:15 +02:00
|
|
|
|
"psubusb %%mm1, %%mm0 \n\t"
|
2001-10-12 00:35:45 +02:00
|
|
|
|
"pxor %%mm2, %%mm0 \n\t"
|
|
|
|
|
"movq %%mm0, (%%eax, %1, 2) \n\t" // line 3
|
|
|
|
|
|
|
|
|
|
"movq (%%ebx, %1), %%mm0 \n\t" // line 6
|
|
|
|
|
"pxor %%mm2, %%mm0 \n\t" //(l4 - l5) <= 0 ? -l5-1 : l5
|
2001-10-13 04:31:15 +02:00
|
|
|
|
"paddusb %%mm1, %%mm0 \n\t"
|
2001-10-12 00:35:45 +02:00
|
|
|
|
"pxor %%mm2, %%mm0 \n\t"
|
|
|
|
|
"movq %%mm0, (%%ebx, %1) \n\t" // line 6
|
|
|
|
|
|
2001-10-13 04:31:15 +02:00
|
|
|
|
PAVGB(%%mm7, %%mm1) // d/8
|
2001-10-12 00:35:45 +02:00
|
|
|
|
|
|
|
|
|
"movq (%%eax, %1), %%mm0 \n\t" // line 2
|
|
|
|
|
"pxor %%mm2, %%mm0 \n\t" //(l4 - l5) <= 0 ? -l2-1 : l2
|
2001-10-13 04:31:15 +02:00
|
|
|
|
"psubusb %%mm1, %%mm0 \n\t"
|
2001-10-12 00:35:45 +02:00
|
|
|
|
"pxor %%mm2, %%mm0 \n\t"
|
|
|
|
|
"movq %%mm0, (%%eax, %1) \n\t" // line 2
|
|
|
|
|
|
|
|
|
|
"movq (%%ebx, %1, 2), %%mm0 \n\t" // line 7
|
|
|
|
|
"pxor %%mm2, %%mm0 \n\t" //(l4 - l5) <= 0 ? -l7-1 : l7
|
2001-10-13 04:31:15 +02:00
|
|
|
|
"paddusb %%mm1, %%mm0 \n\t"
|
2001-10-12 00:35:45 +02:00
|
|
|
|
"pxor %%mm2, %%mm0 \n\t"
|
|
|
|
|
"movq %%mm0, (%%ebx, %1, 2) \n\t" // line 7
|
2001-10-11 00:21:19 +02:00
|
|
|
|
|
|
|
|
|
:
|
|
|
|
|
: "r" (src), "r" (stride)
|
|
|
|
|
: "%eax", "%ebx"
|
|
|
|
|
);
|
|
|
|
|
#else
|
2001-10-12 00:35:45 +02:00
|
|
|
|
|
|
|
|
|
const int l1= stride;
|
|
|
|
|
const int l2= stride + l1;
|
|
|
|
|
const int l3= stride + l2;
|
|
|
|
|
const int l4= stride + l3;
|
|
|
|
|
const int l5= stride + l4;
|
|
|
|
|
const int l6= stride + l5;
|
|
|
|
|
const int l7= stride + l6;
|
|
|
|
|
const int l8= stride + l7;
|
|
|
|
|
const int l9= stride + l8;
|
|
|
|
|
int x;
|
2001-10-17 22:42:07 +02:00
|
|
|
|
|
|
|
|
|
src+= stride*3;
|
2001-10-12 00:35:45 +02:00
|
|
|
|
for(x=0; x<BLOCK_SIZE; x++)
|
|
|
|
|
{
|
|
|
|
|
int a= src[l3] - src[l4];
|
|
|
|
|
int b= src[l4] - src[l5];
|
2001-10-13 04:31:15 +02:00
|
|
|
|
int c= src[l5] - src[l6];
|
2001-10-12 00:35:45 +02:00
|
|
|
|
|
|
|
|
|
int d= MAX(ABS(b) - (ABS(a) + ABS(c))/2, 0);
|
|
|
|
|
|
|
|
|
|
if(d < QP)
|
|
|
|
|
{
|
|
|
|
|
int v = d * SIGN(-b);
|
|
|
|
|
|
|
|
|
|
src[l2] +=v/8;
|
|
|
|
|
src[l3] +=v/4;
|
2001-10-13 04:31:15 +02:00
|
|
|
|
src[l4] +=3*v/8;
|
|
|
|
|
src[l5] -=3*v/8;
|
2001-10-12 00:35:45 +02:00
|
|
|
|
src[l6] -=v/4;
|
|
|
|
|
src[l7] -=v/8;
|
|
|
|
|
|
|
|
|
|
}
|
|
|
|
|
src++;
|
|
|
|
|
}
|
|
|
|
|
/*
|
2001-10-11 00:21:19 +02:00
|
|
|
|
const int l1= stride;
|
|
|
|
|
const int l2= stride + l1;
|
|
|
|
|
const int l3= stride + l2;
|
|
|
|
|
const int l4= stride + l3;
|
|
|
|
|
const int l5= stride + l4;
|
|
|
|
|
const int l6= stride + l5;
|
|
|
|
|
const int l7= stride + l6;
|
|
|
|
|
const int l8= stride + l7;
|
|
|
|
|
const int l9= stride + l8;
|
|
|
|
|
for(int x=0; x<BLOCK_SIZE; x++)
|
|
|
|
|
{
|
|
|
|
|
int v2= src[l2];
|
|
|
|
|
int v3= src[l3];
|
|
|
|
|
int v4= src[l4];
|
|
|
|
|
int v5= src[l5];
|
|
|
|
|
int v6= src[l6];
|
|
|
|
|
int v7= src[l7];
|
|
|
|
|
|
|
|
|
|
if(ABS(v4-v5)<QP && ABS(v4-v5) - (ABS(v3-v4) + ABS(v5-v6))>0 )
|
|
|
|
|
{
|
|
|
|
|
src[l3] = (6*v2 + 4*v3 + 3*v4 + 2*v5 + v6 )/16;
|
|
|
|
|
src[l4] = (3*v2 + 3*v3 + 4*v4 + 3*v5 + 2*v6 + v7 )/16;
|
|
|
|
|
src[l5] = (1*v2 + 2*v3 + 3*v4 + 4*v5 + 3*v6 + 3*v7)/16;
|
|
|
|
|
src[l6] = ( 1*v3 + 2*v4 + 3*v5 + 4*v6 + 6*v7)/16;
|
|
|
|
|
}
|
|
|
|
|
src++;
|
|
|
|
|
}
|
2001-10-12 00:35:45 +02:00
|
|
|
|
*/
|
2001-10-11 00:21:19 +02:00
|
|
|
|
#endif
|
|
|
|
|
}
|
|
|
|
|
|
2001-10-13 04:31:15 +02:00
|
|
|
|
/**
|
|
|
|
|
* Experimental Filter 1 (Horizontal)
|
|
|
|
|
* will not damage linear gradients
|
|
|
|
|
* Flat blocks should look like they where passed through the (1,1,2,2,4,2,2,1,1) 9-Tap filter
|
|
|
|
|
* can only smooth blocks at the expected locations (it cant smooth them if they did move)
|
|
|
|
|
* MMX2 version does correct clipping C version doesnt
|
|
|
|
|
* not identical with the vertical one
|
|
|
|
|
*/
|
|
|
|
|
static inline void horizX1Filter(uint8_t *src, int stride, int QP)
|
|
|
|
|
{
|
|
|
|
|
int y;
|
|
|
|
|
static uint64_t *lut= NULL;
|
|
|
|
|
if(lut==NULL)
|
|
|
|
|
{
|
|
|
|
|
int i;
|
|
|
|
|
lut= (uint64_t*)memalign(8, 256*8);
|
|
|
|
|
for(i=0; i<256; i++)
|
|
|
|
|
{
|
|
|
|
|
int v= i < 128 ? 2*i : 2*(i-256);
|
|
|
|
|
/*
|
|
|
|
|
//Simulate 112242211 9-Tap filter
|
|
|
|
|
uint64_t a= (v/16) & 0xFF;
|
|
|
|
|
uint64_t b= (v/8) & 0xFF;
|
|
|
|
|
uint64_t c= (v/4) & 0xFF;
|
|
|
|
|
uint64_t d= (3*v/8) & 0xFF;
|
|
|
|
|
*/
|
|
|
|
|
//Simulate piecewise linear interpolation
|
|
|
|
|
uint64_t a= (v/16) & 0xFF;
|
|
|
|
|
uint64_t b= (v*3/16) & 0xFF;
|
|
|
|
|
uint64_t c= (v*5/16) & 0xFF;
|
|
|
|
|
uint64_t d= (7*v/16) & 0xFF;
|
|
|
|
|
uint64_t A= (0x100 - a)&0xFF;
|
|
|
|
|
uint64_t B= (0x100 - b)&0xFF;
|
|
|
|
|
uint64_t C= (0x100 - c)&0xFF;
|
|
|
|
|
uint64_t D= (0x100 - c)&0xFF;
|
|
|
|
|
|
|
|
|
|
lut[i] = (a<<56) | (b<<48) | (c<<40) | (d<<32) |
|
|
|
|
|
(D<<24) | (C<<16) | (B<<8) | (A);
|
|
|
|
|
//lut[i] = (v<<32) | (v<<24);
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
|
|
|
|
|
asm volatile(
|
|
|
|
|
"pxor %%mm7, %%mm7 \n\t" // 0
|
|
|
|
|
// "movq b80, %%mm6 \n\t" // MIN_SIGNED_BYTE
|
|
|
|
|
"leal (%0, %1), %%eax \n\t"
|
|
|
|
|
"leal (%%eax, %1, 4), %%ebx \n\t"
|
|
|
|
|
|
|
|
|
|
"movq b80, %%mm6 \n\t"
|
2001-10-13 11:30:07 +02:00
|
|
|
|
"movd pQPb, %%mm5 \n\t" // QP
|
2001-10-13 04:31:15 +02:00
|
|
|
|
"movq %%mm5, %%mm4 \n\t"
|
|
|
|
|
"paddusb %%mm5, %%mm5 \n\t" // 2QP
|
|
|
|
|
"paddusb %%mm5, %%mm4 \n\t" // 3QP
|
|
|
|
|
"pxor %%mm5, %%mm5 \n\t" // 0
|
|
|
|
|
"psubb %%mm4, %%mm5 \n\t" // -3QP
|
|
|
|
|
"por bm11111110, %%mm5 \n\t" // ...,FF,FF,-3QP
|
|
|
|
|
"psllq $24, %%mm5 \n\t"
|
|
|
|
|
|
|
|
|
|
// 0 1 2 3 4 5 6 7 8 9
|
|
|
|
|
// %0 eax eax+%1 eax+2%1 %0+4%1 ebx ebx+%1 ebx+2%1 %0+8%1 ebx+4%1
|
|
|
|
|
|
|
|
|
|
#define HX1old(a) \
|
|
|
|
|
"movd " #a ", %%mm0 \n\t"\
|
|
|
|
|
"movd 4" #a ", %%mm1 \n\t"\
|
|
|
|
|
"punpckldq %%mm1, %%mm0 \n\t"\
|
|
|
|
|
"movq %%mm0, %%mm1 \n\t"\
|
|
|
|
|
"movq %%mm0, %%mm2 \n\t"\
|
|
|
|
|
"psrlq $8, %%mm1 \n\t"\
|
|
|
|
|
"psubusb %%mm1, %%mm2 \n\t"\
|
|
|
|
|
"psubusb %%mm0, %%mm1 \n\t"\
|
|
|
|
|
"por %%mm2, %%mm1 \n\t" /* p<>x = |px - p(x+1)| */\
|
|
|
|
|
"pcmpeqb %%mm7, %%mm2 \n\t" /* p<>x = sgn[px - p(x+1)] */\
|
|
|
|
|
"pshufw $0x00, %%mm1, %%mm3 \n\t" /* p<>5 = |p1 - p2| */\
|
|
|
|
|
PAVGB(%%mm1, %%mm3) /* p<>5 = (|p2-p1| + |p6-p5|)/2 */\
|
|
|
|
|
"psrlq $16, %%mm3 \n\t" /* p<>3 = (|p2-p1| + |p6-p5|)/2 */\
|
|
|
|
|
"psubusb %%mm3, %%mm1 \n\t" /* |p3-p4|-(|p2-p1| + |p6-p5|)/2 */\
|
|
|
|
|
"paddb %%mm5, %%mm1 \n\t"\
|
|
|
|
|
"psubusb %%mm5, %%mm1 \n\t"\
|
|
|
|
|
PAVGB(%%mm7, %%mm1)\
|
|
|
|
|
"pxor %%mm2, %%mm1 \n\t"\
|
|
|
|
|
"psubb %%mm2, %%mm1 \n\t"\
|
|
|
|
|
"psrlq $24, %%mm1 \n\t"\
|
|
|
|
|
"movd %%mm1, %%ecx \n\t"\
|
|
|
|
|
"paddb %%mm6, %%mm0 \n\t"\
|
|
|
|
|
"paddsb (%3, %%ecx, 8), %%mm0 \n\t"\
|
|
|
|
|
"paddb %%mm6, %%mm0 \n\t"\
|
|
|
|
|
"movq %%mm0, " #a " \n\t"\
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
HX1old((%0))
|
|
|
|
|
HX1old((%%eax))
|
|
|
|
|
HX1old((%%eax, %1))
|
|
|
|
|
HX1old((%%eax, %1, 2))
|
|
|
|
|
HX1old((%0, %1, 4))
|
|
|
|
|
HX1old((%%ebx))
|
|
|
|
|
HX1old((%%ebx, %1))
|
|
|
|
|
HX1old((%%ebx, %1, 2))
|
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
//FIXME add some comments, its unreadable ...
|
|
|
|
|
#define HX1b(a, c, b, d) \
|
|
|
|
|
"movd " #a ", %%mm0 \n\t"\
|
|
|
|
|
"movd 4" #a ", %%mm1 \n\t"\
|
|
|
|
|
"punpckldq %%mm1, %%mm0 \n\t"\
|
|
|
|
|
"movd " #b ", %%mm4 \n\t"\
|
|
|
|
|
"movq %%mm0, %%mm1 \n\t"\
|
|
|
|
|
"movq %%mm0, %%mm2 \n\t"\
|
|
|
|
|
"psrlq $8, %%mm1 \n\t"\
|
|
|
|
|
"movd 4" #b ", %%mm3 \n\t"\
|
|
|
|
|
"psubusb %%mm1, %%mm2 \n\t"\
|
|
|
|
|
"psubusb %%mm0, %%mm1 \n\t"\
|
|
|
|
|
"por %%mm2, %%mm1 \n\t" /* p<>x = |px - p(x+1)| */\
|
|
|
|
|
"pcmpeqb %%mm7, %%mm2 \n\t" /* p<>x = sgn[px - p(x+1)] */\
|
|
|
|
|
"punpckldq %%mm3, %%mm4 \n\t"\
|
|
|
|
|
"movq %%mm1, %%mm3 \n\t"\
|
|
|
|
|
"psllq $32, %%mm3 \n\t" /* p<>5 = |p1 - p2| */\
|
|
|
|
|
PAVGB(%%mm1, %%mm3) /* p<>5 = (|p2-p1| + |p6-p5|)/2 */\
|
|
|
|
|
"paddb %%mm6, %%mm0 \n\t"\
|
|
|
|
|
"psrlq $16, %%mm3 \n\t" /* p<>3 = (|p2-p1| + |p6-p5|)/2 */\
|
|
|
|
|
"psubusb %%mm3, %%mm1 \n\t" /* |p3-p4|-(|p2-p1| + |p6-p5|)/2 */\
|
|
|
|
|
"movq %%mm4, %%mm3 \n\t"\
|
|
|
|
|
"paddb %%mm5, %%mm1 \n\t"\
|
|
|
|
|
"psubusb %%mm5, %%mm1 \n\t"\
|
|
|
|
|
"psrlq $8, %%mm3 \n\t"\
|
|
|
|
|
PAVGB(%%mm7, %%mm1)\
|
|
|
|
|
"pxor %%mm2, %%mm1 \n\t"\
|
|
|
|
|
"psubb %%mm2, %%mm1 \n\t"\
|
|
|
|
|
"movq %%mm4, %%mm2 \n\t"\
|
|
|
|
|
"psrlq $24, %%mm1 \n\t"\
|
|
|
|
|
"psubusb %%mm3, %%mm2 \n\t"\
|
|
|
|
|
"movd %%mm1, %%ecx \n\t"\
|
|
|
|
|
"psubusb %%mm4, %%mm3 \n\t"\
|
2001-10-13 11:30:07 +02:00
|
|
|
|
"paddsb (%2, %%ecx, 8), %%mm0 \n\t"\
|
2001-10-13 04:31:15 +02:00
|
|
|
|
"por %%mm2, %%mm3 \n\t" /* p<>x = |px - p(x+1)| */\
|
|
|
|
|
"paddb %%mm6, %%mm0 \n\t"\
|
|
|
|
|
"pcmpeqb %%mm7, %%mm2 \n\t" /* p<>x = sgn[px - p(x+1)] */\
|
|
|
|
|
"movq %%mm3, %%mm1 \n\t"\
|
|
|
|
|
"psllq $32, %%mm1 \n\t" /* p<>5 = |p1 - p2| */\
|
|
|
|
|
"movq %%mm0, " #a " \n\t"\
|
|
|
|
|
PAVGB(%%mm3, %%mm1) /* p<>5 = (|p2-p1| + |p6-p5|)/2 */\
|
|
|
|
|
"paddb %%mm6, %%mm4 \n\t"\
|
|
|
|
|
"psrlq $16, %%mm1 \n\t" /* p<>3 = (|p2-p1| + |p6-p5|)/2 */\
|
|
|
|
|
"psubusb %%mm1, %%mm3 \n\t" /* |p3-p4|-(|p2-p1| + |p6-p5|)/2 */\
|
|
|
|
|
"paddb %%mm5, %%mm3 \n\t"\
|
|
|
|
|
"psubusb %%mm5, %%mm3 \n\t"\
|
|
|
|
|
PAVGB(%%mm7, %%mm3)\
|
|
|
|
|
"pxor %%mm2, %%mm3 \n\t"\
|
|
|
|
|
"psubb %%mm2, %%mm3 \n\t"\
|
|
|
|
|
"psrlq $24, %%mm3 \n\t"\
|
|
|
|
|
"movd " #c ", %%mm0 \n\t"\
|
|
|
|
|
"movd 4" #c ", %%mm1 \n\t"\
|
|
|
|
|
"punpckldq %%mm1, %%mm0 \n\t"\
|
|
|
|
|
"paddb %%mm6, %%mm0 \n\t"\
|
2001-10-13 11:30:07 +02:00
|
|
|
|
"paddsb (%2, %%ecx, 8), %%mm0 \n\t"\
|
2001-10-13 04:31:15 +02:00
|
|
|
|
"paddb %%mm6, %%mm0 \n\t"\
|
|
|
|
|
"movq %%mm0, " #c " \n\t"\
|
|
|
|
|
"movd %%mm3, %%ecx \n\t"\
|
|
|
|
|
"movd " #d ", %%mm0 \n\t"\
|
2001-10-13 11:30:07 +02:00
|
|
|
|
"paddsb (%2, %%ecx, 8), %%mm4 \n\t"\
|
2001-10-13 04:31:15 +02:00
|
|
|
|
"movd 4" #d ", %%mm1 \n\t"\
|
|
|
|
|
"paddb %%mm6, %%mm4 \n\t"\
|
|
|
|
|
"punpckldq %%mm1, %%mm0 \n\t"\
|
|
|
|
|
"movq %%mm4, " #b " \n\t"\
|
|
|
|
|
"paddb %%mm6, %%mm0 \n\t"\
|
2001-10-13 11:30:07 +02:00
|
|
|
|
"paddsb (%2, %%ecx, 8), %%mm0 \n\t"\
|
2001-10-13 04:31:15 +02:00
|
|
|
|
"paddb %%mm6, %%mm0 \n\t"\
|
|
|
|
|
"movq %%mm0, " #d " \n\t"\
|
|
|
|
|
|
|
|
|
|
HX1b((%0),(%%eax),(%%eax, %1),(%%eax, %1, 2))
|
|
|
|
|
HX1b((%0, %1, 4),(%%ebx),(%%ebx, %1),(%%ebx, %1, 2))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
:
|
2001-10-13 11:30:07 +02:00
|
|
|
|
: "r" (src), "r" (stride), "r" (lut)
|
2001-10-13 04:31:15 +02:00
|
|
|
|
: "%eax", "%ebx", "%ecx"
|
|
|
|
|
);
|
|
|
|
|
#else
|
|
|
|
|
|
|
|
|
|
//FIXME (has little in common with the mmx2 version)
|
|
|
|
|
for(y=0; y<BLOCK_SIZE; y++)
|
|
|
|
|
{
|
|
|
|
|
int a= src[1] - src[2];
|
|
|
|
|
int b= src[3] - src[4];
|
|
|
|
|
int c= src[5] - src[6];
|
|
|
|
|
|
|
|
|
|
int d= MAX(ABS(b) - (ABS(a) + ABS(c))/2, 0);
|
|
|
|
|
|
|
|
|
|
if(d < QP)
|
|
|
|
|
{
|
|
|
|
|
int v = d * SIGN(-b);
|
|
|
|
|
|
|
|
|
|
src[1] +=v/8;
|
|
|
|
|
src[2] +=v/4;
|
|
|
|
|
src[3] +=3*v/8;
|
|
|
|
|
src[4] -=3*v/8;
|
|
|
|
|
src[5] -=v/4;
|
|
|
|
|
src[6] -=v/8;
|
|
|
|
|
|
|
|
|
|
}
|
|
|
|
|
src+=stride;
|
|
|
|
|
}
|
|
|
|
|
#endif
|
|
|
|
|
}
|
|
|
|
|
|
2001-10-11 00:21:19 +02:00
|
|
|
|
|
2001-10-11 00:13:27 +02:00
|
|
|
|
static inline void doVertDefFilter(uint8_t src[], int stride, int QP)
|
|
|
|
|
{
|
|
|
|
|
#ifdef HAVE_MMX
|
2001-10-17 22:42:07 +02:00
|
|
|
|
src+= stride*4;
|
2001-10-11 00:13:27 +02:00
|
|
|
|
//FIXME try pmul for *5 stuff
|
|
|
|
|
// src[0]=0;
|
|
|
|
|
asm volatile(
|
|
|
|
|
"pxor %%mm7, %%mm7 \n\t"
|
|
|
|
|
"leal (%0, %1), %%eax \n\t"
|
|
|
|
|
"leal (%%eax, %1, 4), %%ebx \n\t"
|
|
|
|
|
// 0 1 2 3 4 5 6 7
|
|
|
|
|
// %0 %0+%1 %0+2%1 eax+2%1 %0+4%1 eax+4%1 ebx+%1 ebx+2%1
|
|
|
|
|
// %0 eax eax+%1 eax+2%1 %0+4%1 ebx ebx+%1 ebx+2%1
|
|
|
|
|
|
|
|
|
|
"movq (%0), %%mm0 \n\t"
|
|
|
|
|
"movq %%mm0, %%mm1 \n\t"
|
|
|
|
|
"punpcklbw %%mm7, %%mm0 \n\t" // low part of line 0
|
|
|
|
|
"punpckhbw %%mm7, %%mm1 \n\t" // high part of line 0
|
|
|
|
|
|
|
|
|
|
"movq (%%eax), %%mm2 \n\t"
|
|
|
|
|
"movq %%mm2, %%mm3 \n\t"
|
|
|
|
|
"punpcklbw %%mm7, %%mm2 \n\t" // low part of line 1
|
|
|
|
|
"punpckhbw %%mm7, %%mm3 \n\t" // high part of line 1
|
|
|
|
|
|
|
|
|
|
"movq (%%eax, %1), %%mm4 \n\t"
|
|
|
|
|
"movq %%mm4, %%mm5 \n\t"
|
|
|
|
|
"punpcklbw %%mm7, %%mm4 \n\t" // low part of line 2
|
|
|
|
|
"punpckhbw %%mm7, %%mm5 \n\t" // high part of line 2
|
|
|
|
|
|
|
|
|
|
"paddw %%mm0, %%mm0 \n\t" // 2L0
|
|
|
|
|
"paddw %%mm1, %%mm1 \n\t" // 2H0
|
|
|
|
|
"psubw %%mm4, %%mm2 \n\t" // L1 - L2
|
|
|
|
|
"psubw %%mm5, %%mm3 \n\t" // H1 - H2
|
|
|
|
|
"psubw %%mm2, %%mm0 \n\t" // 2L0 - L1 + L2
|
|
|
|
|
"psubw %%mm3, %%mm1 \n\t" // 2H0 - H1 + H2
|
|
|
|
|
|
|
|
|
|
"psllw $2, %%mm2 \n\t" // 4L1 - 4L2
|
|
|
|
|
"psllw $2, %%mm3 \n\t" // 4H1 - 4H2
|
|
|
|
|
"psubw %%mm2, %%mm0 \n\t" // 2L0 - 5L1 + 5L2
|
|
|
|
|
"psubw %%mm3, %%mm1 \n\t" // 2H0 - 5H1 + 5H2
|
|
|
|
|
|
|
|
|
|
"movq (%%eax, %1, 2), %%mm2 \n\t"
|
|
|
|
|
"movq %%mm2, %%mm3 \n\t"
|
|
|
|
|
"punpcklbw %%mm7, %%mm2 \n\t" // L3
|
|
|
|
|
"punpckhbw %%mm7, %%mm3 \n\t" // H3
|
|
|
|
|
|
|
|
|
|
"psubw %%mm2, %%mm0 \n\t" // 2L0 - 5L1 + 5L2 - L3
|
|
|
|
|
"psubw %%mm3, %%mm1 \n\t" // 2H0 - 5H1 + 5H2 - H3
|
|
|
|
|
"psubw %%mm2, %%mm0 \n\t" // 2L0 - 5L1 + 5L2 - 2L3
|
|
|
|
|
"psubw %%mm3, %%mm1 \n\t" // 2H0 - 5H1 + 5H2 - 2H3
|
|
|
|
|
"movq %%mm0, temp0 \n\t" // 2L0 - 5L1 + 5L2 - 2L3
|
|
|
|
|
"movq %%mm1, temp1 \n\t" // 2H0 - 5H1 + 5H2 - 2H3
|
|
|
|
|
|
|
|
|
|
"movq (%0, %1, 4), %%mm0 \n\t"
|
|
|
|
|
"movq %%mm0, %%mm1 \n\t"
|
|
|
|
|
"punpcklbw %%mm7, %%mm0 \n\t" // L4
|
|
|
|
|
"punpckhbw %%mm7, %%mm1 \n\t" // H4
|
|
|
|
|
|
|
|
|
|
"psubw %%mm0, %%mm2 \n\t" // L3 - L4
|
|
|
|
|
"psubw %%mm1, %%mm3 \n\t" // H3 - H4
|
|
|
|
|
"movq %%mm2, temp2 \n\t" // L3 - L4
|
|
|
|
|
"movq %%mm3, temp3 \n\t" // H3 - H4
|
|
|
|
|
"paddw %%mm4, %%mm4 \n\t" // 2L2
|
|
|
|
|
"paddw %%mm5, %%mm5 \n\t" // 2H2
|
|
|
|
|
"psubw %%mm2, %%mm4 \n\t" // 2L2 - L3 + L4
|
|
|
|
|
"psubw %%mm3, %%mm5 \n\t" // 2H2 - H3 + H4
|
|
|
|
|
|
|
|
|
|
"psllw $2, %%mm2 \n\t" // 4L3 - 4L4
|
|
|
|
|
"psllw $2, %%mm3 \n\t" // 4H3 - 4H4
|
|
|
|
|
"psubw %%mm2, %%mm4 \n\t" // 2L2 - 5L3 + 5L4
|
|
|
|
|
"psubw %%mm3, %%mm5 \n\t" // 2H2 - 5H3 + 5H4
|
|
|
|
|
//50 opcodes so far
|
|
|
|
|
"movq (%%ebx), %%mm2 \n\t"
|
|
|
|
|
"movq %%mm2, %%mm3 \n\t"
|
|
|
|
|
"punpcklbw %%mm7, %%mm2 \n\t" // L5
|
|
|
|
|
"punpckhbw %%mm7, %%mm3 \n\t" // H5
|
|
|
|
|
"psubw %%mm2, %%mm4 \n\t" // 2L2 - 5L3 + 5L4 - L5
|
|
|
|
|
"psubw %%mm3, %%mm5 \n\t" // 2H2 - 5H3 + 5H4 - H5
|
|
|
|
|
"psubw %%mm2, %%mm4 \n\t" // 2L2 - 5L3 + 5L4 - 2L5
|
|
|
|
|
"psubw %%mm3, %%mm5 \n\t" // 2H2 - 5H3 + 5H4 - 2H5
|
|
|
|
|
|
|
|
|
|
"movq (%%ebx, %1), %%mm6 \n\t"
|
|
|
|
|
"punpcklbw %%mm7, %%mm6 \n\t" // L6
|
|
|
|
|
"psubw %%mm6, %%mm2 \n\t" // L5 - L6
|
|
|
|
|
"movq (%%ebx, %1), %%mm6 \n\t"
|
|
|
|
|
"punpckhbw %%mm7, %%mm6 \n\t" // H6
|
|
|
|
|
"psubw %%mm6, %%mm3 \n\t" // H5 - H6
|
|
|
|
|
|
|
|
|
|
"paddw %%mm0, %%mm0 \n\t" // 2L4
|
|
|
|
|
"paddw %%mm1, %%mm1 \n\t" // 2H4
|
|
|
|
|
"psubw %%mm2, %%mm0 \n\t" // 2L4 - L5 + L6
|
|
|
|
|
"psubw %%mm3, %%mm1 \n\t" // 2H4 - H5 + H6
|
|
|
|
|
|
|
|
|
|
"psllw $2, %%mm2 \n\t" // 4L5 - 4L6
|
|
|
|
|
"psllw $2, %%mm3 \n\t" // 4H5 - 4H6
|
|
|
|
|
"psubw %%mm2, %%mm0 \n\t" // 2L4 - 5L5 + 5L6
|
|
|
|
|
"psubw %%mm3, %%mm1 \n\t" // 2H4 - 5H5 + 5H6
|
|
|
|
|
|
|
|
|
|
"movq (%%ebx, %1, 2), %%mm2 \n\t"
|
|
|
|
|
"movq %%mm2, %%mm3 \n\t"
|
|
|
|
|
"punpcklbw %%mm7, %%mm2 \n\t" // L7
|
|
|
|
|
"punpckhbw %%mm7, %%mm3 \n\t" // H7
|
|
|
|
|
|
|
|
|
|
"paddw %%mm2, %%mm2 \n\t" // 2L7
|
|
|
|
|
"paddw %%mm3, %%mm3 \n\t" // 2H7
|
|
|
|
|
"psubw %%mm2, %%mm0 \n\t" // 2L4 - 5L5 + 5L6 - 2L7
|
|
|
|
|
"psubw %%mm3, %%mm1 \n\t" // 2H4 - 5H5 + 5H6 - 2H7
|
|
|
|
|
|
|
|
|
|
"movq temp0, %%mm2 \n\t" // 2L0 - 5L1 + 5L2 - 2L3
|
|
|
|
|
"movq temp1, %%mm3 \n\t" // 2H0 - 5H1 + 5H2 - 2H3
|
|
|
|
|
//FIXME pxor, psubw, pmax for abs
|
|
|
|
|
"movq %%mm7, %%mm6 \n\t" // 0
|
|
|
|
|
"pcmpgtw %%mm0, %%mm6 \n\t"
|
|
|
|
|
"pxor %%mm6, %%mm0 \n\t"
|
|
|
|
|
"psubw %%mm6, %%mm0 \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
|
|
|
|
|
"movq %%mm7, %%mm6 \n\t" // 0
|
|
|
|
|
"pcmpgtw %%mm1, %%mm6 \n\t"
|
|
|
|
|
"pxor %%mm6, %%mm1 \n\t"
|
|
|
|
|
"psubw %%mm6, %%mm1 \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
|
|
|
|
|
|
|
|
|
|
"movq %%mm7, %%mm6 \n\t" // 0
|
|
|
|
|
"pcmpgtw %%mm2, %%mm6 \n\t"
|
|
|
|
|
"pxor %%mm6, %%mm2 \n\t"
|
|
|
|
|
"psubw %%mm6, %%mm2 \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
|
|
|
|
|
"movq %%mm7, %%mm6 \n\t" // 0
|
|
|
|
|
"pcmpgtw %%mm3, %%mm6 \n\t"
|
|
|
|
|
"pxor %%mm6, %%mm3 \n\t"
|
|
|
|
|
"psubw %%mm6, %%mm3 \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
|
|
|
|
|
|
|
|
|
|
#ifdef HAVE_MMX2
|
|
|
|
|
"pminsw %%mm2, %%mm0 \n\t"
|
|
|
|
|
"pminsw %%mm3, %%mm1 \n\t"
|
|
|
|
|
#else
|
|
|
|
|
"movq %%mm0, %%mm6 \n\t"
|
|
|
|
|
"psubusw %%mm2, %%mm6 \n\t"
|
|
|
|
|
"psubw %%mm6, %%mm0 \n\t"
|
|
|
|
|
"movq %%mm1, %%mm6 \n\t"
|
|
|
|
|
"psubusw %%mm3, %%mm6 \n\t"
|
|
|
|
|
"psubw %%mm6, %%mm1 \n\t"
|
|
|
|
|
#endif
|
|
|
|
|
|
|
|
|
|
"movq %%mm7, %%mm6 \n\t" // 0
|
|
|
|
|
"pcmpgtw %%mm4, %%mm6 \n\t" // sign(2L2 - 5L3 + 5L4 - 2L5)
|
|
|
|
|
"pxor %%mm6, %%mm4 \n\t"
|
|
|
|
|
"psubw %%mm6, %%mm4 \n\t" // |2L2 - 5L3 + 5L4 - 2L5|
|
|
|
|
|
"pcmpgtw %%mm5, %%mm7 \n\t" // sign(2H2 - 5H3 + 5H4 - 2H5)
|
|
|
|
|
"pxor %%mm7, %%mm5 \n\t"
|
|
|
|
|
"psubw %%mm7, %%mm5 \n\t" // |2H2 - 5H3 + 5H4 - 2H5|
|
|
|
|
|
// 100 opcodes
|
|
|
|
|
"movd %2, %%mm2 \n\t" // QP
|
|
|
|
|
"punpcklwd %%mm2, %%mm2 \n\t"
|
|
|
|
|
"punpcklwd %%mm2, %%mm2 \n\t"
|
|
|
|
|
"psllw $3, %%mm2 \n\t" // 8QP
|
|
|
|
|
"movq %%mm2, %%mm3 \n\t" // 8QP
|
|
|
|
|
"pcmpgtw %%mm4, %%mm2 \n\t"
|
|
|
|
|
"pcmpgtw %%mm5, %%mm3 \n\t"
|
|
|
|
|
"pand %%mm2, %%mm4 \n\t"
|
|
|
|
|
"pand %%mm3, %%mm5 \n\t"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
"psubusw %%mm0, %%mm4 \n\t" // hd
|
|
|
|
|
"psubusw %%mm1, %%mm5 \n\t" // ld
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
"movq w05, %%mm2 \n\t" // 5
|
|
|
|
|
"pmullw %%mm2, %%mm4 \n\t"
|
|
|
|
|
"pmullw %%mm2, %%mm5 \n\t"
|
|
|
|
|
"movq w20, %%mm2 \n\t" // 32
|
|
|
|
|
"paddw %%mm2, %%mm4 \n\t"
|
|
|
|
|
"paddw %%mm2, %%mm5 \n\t"
|
|
|
|
|
"psrlw $6, %%mm4 \n\t"
|
|
|
|
|
"psrlw $6, %%mm5 \n\t"
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
"movq w06, %%mm2 \n\t" // 6
|
|
|
|
|
"paddw %%mm2, %%mm4 \n\t"
|
|
|
|
|
"paddw %%mm2, %%mm5 \n\t"
|
|
|
|
|
"movq w1400, %%mm2 \n\t" // 1400h = 5120 = 5/64*2^16
|
|
|
|
|
//FIXME if *5/64 is supposed to be /13 then we should use 5041 instead of 5120
|
|
|
|
|
"pmulhw %%mm2, %%mm4 \n\t" // hd/13
|
|
|
|
|
"pmulhw %%mm2, %%mm5 \n\t" // ld/13
|
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
"movq temp2, %%mm0 \n\t" // L3 - L4
|
|
|
|
|
"movq temp3, %%mm1 \n\t" // H3 - H4
|
|
|
|
|
|
|
|
|
|
"pxor %%mm2, %%mm2 \n\t"
|
|
|
|
|
"pxor %%mm3, %%mm3 \n\t"
|
|
|
|
|
|
|
|
|
|
// FIXME rounding error
|
|
|
|
|
"psraw $1, %%mm0 \n\t" // (L3 - L4)/2
|
|
|
|
|
"psraw $1, %%mm1 \n\t" // (H3 - H4)/2
|
|
|
|
|
"pcmpgtw %%mm0, %%mm2 \n\t" // sign (L3-L4)
|
|
|
|
|
"pcmpgtw %%mm1, %%mm3 \n\t" // sign (H3-H4)
|
|
|
|
|
"pxor %%mm2, %%mm0 \n\t"
|
|
|
|
|
"pxor %%mm3, %%mm1 \n\t"
|
|
|
|
|
"psubw %%mm2, %%mm0 \n\t" // |L3-L4|
|
|
|
|
|
"psubw %%mm3, %%mm1 \n\t" // |H3-H4|
|
|
|
|
|
// "psrlw $1, %%mm0 \n\t" // |L3 - L4|/2
|
|
|
|
|
// "psrlw $1, %%mm1 \n\t" // |H3 - H4|/2
|
|
|
|
|
|
|
|
|
|
"pxor %%mm6, %%mm2 \n\t"
|
|
|
|
|
"pxor %%mm7, %%mm3 \n\t"
|
|
|
|
|
"pand %%mm2, %%mm4 \n\t"
|
|
|
|
|
"pand %%mm3, %%mm5 \n\t"
|
|
|
|
|
|
|
|
|
|
#ifdef HAVE_MMX2
|
|
|
|
|
"pminsw %%mm0, %%mm4 \n\t"
|
|
|
|
|
"pminsw %%mm1, %%mm5 \n\t"
|
|
|
|
|
#else
|
|
|
|
|
"movq %%mm4, %%mm2 \n\t"
|
|
|
|
|
"psubusw %%mm0, %%mm2 \n\t"
|
|
|
|
|
"psubw %%mm2, %%mm4 \n\t"
|
|
|
|
|
"movq %%mm5, %%mm2 \n\t"
|
|
|
|
|
"psubusw %%mm1, %%mm2 \n\t"
|
|
|
|
|
"psubw %%mm2, %%mm5 \n\t"
|
|
|
|
|
#endif
|
|
|
|
|
"pxor %%mm6, %%mm4 \n\t"
|
|
|
|
|
"pxor %%mm7, %%mm5 \n\t"
|
|
|
|
|
"psubw %%mm6, %%mm4 \n\t"
|
|
|
|
|
"psubw %%mm7, %%mm5 \n\t"
|
|
|
|
|
"packsswb %%mm5, %%mm4 \n\t"
|
|
|
|
|
"movq (%%eax, %1, 2), %%mm0 \n\t"
|
|
|
|
|
"paddb %%mm4, %%mm0 \n\t"
|
|
|
|
|
"movq %%mm0, (%%eax, %1, 2) \n\t"
|
|
|
|
|
"movq (%0, %1, 4), %%mm0 \n\t"
|
|
|
|
|
"psubb %%mm4, %%mm0 \n\t"
|
|
|
|
|
"movq %%mm0, (%0, %1, 4) \n\t"
|
|
|
|
|
|
|
|
|
|
:
|
|
|
|
|
: "r" (src), "r" (stride), "r" (QP)
|
|
|
|
|
: "%eax", "%ebx"
|
|
|
|
|
);
|
|
|
|
|
#else
|
|
|
|
|
const int l1= stride;
|
|
|
|
|
const int l2= stride + l1;
|
|
|
|
|
const int l3= stride + l2;
|
|
|
|
|
const int l4= stride + l3;
|
|
|
|
|
const int l5= stride + l4;
|
|
|
|
|
const int l6= stride + l5;
|
|
|
|
|
const int l7= stride + l6;
|
|
|
|
|
const int l8= stride + l7;
|
|
|
|
|
// const int l9= stride + l8;
|
2001-10-12 00:35:45 +02:00
|
|
|
|
int x;
|
2001-10-17 22:42:07 +02:00
|
|
|
|
src+= stride*3;
|
2001-10-12 00:35:45 +02:00
|
|
|
|
for(x=0; x<BLOCK_SIZE; x++)
|
2001-10-11 00:13:27 +02:00
|
|
|
|
{
|
|
|
|
|
const int middleEnergy= 5*(src[l5] - src[l4]) + 2*(src[l3] - src[l6]);
|
|
|
|
|
if(ABS(middleEnergy) < 8*QP)
|
|
|
|
|
{
|
|
|
|
|
const int q=(src[l4] - src[l5])/2;
|
|
|
|
|
const int leftEnergy= 5*(src[l3] - src[l2]) + 2*(src[l1] - src[l4]);
|
|
|
|
|
const int rightEnergy= 5*(src[l7] - src[l6]) + 2*(src[l5] - src[l8]);
|
|
|
|
|
|
|
|
|
|
int d= ABS(middleEnergy) - MIN( ABS(leftEnergy), ABS(rightEnergy) );
|
|
|
|
|
d= MAX(d, 0);
|
|
|
|
|
|
|
|
|
|
d= (5*d + 32) >> 6;
|
|
|
|
|
d*= SIGN(-middleEnergy);
|
|
|
|
|
|
|
|
|
|
if(q>0)
|
|
|
|
|
{
|
|
|
|
|
d= d<0 ? 0 : d;
|
|
|
|
|
d= d>q ? q : d;
|
|
|
|
|
}
|
|
|
|
|
else
|
|
|
|
|
{
|
|
|
|
|
d= d>0 ? 0 : d;
|
|
|
|
|
d= d<q ? q : d;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
src[l4]-= d;
|
|
|
|
|
src[l5]+= d;
|
|
|
|
|
}
|
|
|
|
|
src++;
|
|
|
|
|
}
|
|
|
|
|
#endif
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
//FIXME? |255-0| = 1
|
|
|
|
|
/**
|
|
|
|
|
* Check if the given 8x8 Block is mostly "flat" and copy the unaliged data into tempBlock.
|
|
|
|
|
*/
|
2001-10-12 00:35:45 +02:00
|
|
|
|
static inline int isHorizDCAndCopy2Temp(uint8_t src[], int stride)
|
2001-10-11 00:13:27 +02:00
|
|
|
|
{
|
|
|
|
|
// src++;
|
|
|
|
|
int numEq= 0;
|
|
|
|
|
#ifdef HAVE_MMX
|
|
|
|
|
asm volatile (
|
|
|
|
|
// "int $3 \n\t"
|
|
|
|
|
"pushl %1\n\t"
|
|
|
|
|
"movq b7E, %%mm7 \n\t" // mm7 = 0x7F
|
|
|
|
|
"movq b7C, %%mm6 \n\t" // mm6 = 0x7D
|
|
|
|
|
"leal tempBlock, %%eax \n\t"
|
|
|
|
|
"pxor %%mm0, %%mm0 \n\t"
|
|
|
|
|
|
|
|
|
|
#define HDC_CHECK_AND_CPY(i) \
|
|
|
|
|
"movq -4(%1), %%mm2 \n\t"\
|
|
|
|
|
"psrlq $32, %%mm2 \n\t"\
|
|
|
|
|
"punpckldq 4(%1), %%mm2 \n\t" /* (%1) */\
|
|
|
|
|
"movq %%mm2, %%mm1 \n\t"\
|
|
|
|
|
"psrlq $8, %%mm2 \n\t"\
|
|
|
|
|
"psubb %%mm1, %%mm2 \n\t"\
|
|
|
|
|
"paddb %%mm7, %%mm2 \n\t"\
|
|
|
|
|
"pcmpgtb %%mm6, %%mm2 \n\t"\
|
|
|
|
|
"paddb %%mm2, %%mm0 \n\t"\
|
|
|
|
|
"movq %%mm1," #i "(%%eax) \n\t"
|
|
|
|
|
|
|
|
|
|
HDC_CHECK_AND_CPY(0)
|
|
|
|
|
"addl %2, %1 \n\t"
|
|
|
|
|
HDC_CHECK_AND_CPY(8)
|
|
|
|
|
"addl %2, %1 \n\t"
|
|
|
|
|
HDC_CHECK_AND_CPY(16)
|
|
|
|
|
"addl %2, %1 \n\t"
|
|
|
|
|
HDC_CHECK_AND_CPY(24)
|
|
|
|
|
"addl %2, %1 \n\t"
|
|
|
|
|
HDC_CHECK_AND_CPY(32)
|
|
|
|
|
"addl %2, %1 \n\t"
|
|
|
|
|
HDC_CHECK_AND_CPY(40)
|
|
|
|
|
"addl %2, %1 \n\t"
|
|
|
|
|
HDC_CHECK_AND_CPY(48)
|
|
|
|
|
"addl %2, %1 \n\t"
|
|
|
|
|
HDC_CHECK_AND_CPY(56)
|
|
|
|
|
|
|
|
|
|
"psllq $8, %%mm0 \n\t" // remove dummy value
|
|
|
|
|
"movq %%mm0, %%mm1 \n\t"
|
|
|
|
|
"psrlw $8, %%mm0 \n\t"
|
|
|
|
|
"paddb %%mm1, %%mm0 \n\t"
|
|
|
|
|
"movq %%mm0, %%mm1 \n\t"
|
|
|
|
|
"psrlq $16, %%mm0 \n\t"
|
|
|
|
|
"paddb %%mm1, %%mm0 \n\t"
|
|
|
|
|
"movq %%mm0, %%mm1 \n\t"
|
|
|
|
|
"psrlq $32, %%mm0 \n\t"
|
|
|
|
|
"paddb %%mm1, %%mm0 \n\t"
|
|
|
|
|
"popl %1\n\t"
|
|
|
|
|
"movd %%mm0, %0 \n\t"
|
|
|
|
|
: "=r" (numEq)
|
|
|
|
|
: "r" (src), "r" (stride)
|
|
|
|
|
: "%eax"
|
|
|
|
|
);
|
|
|
|
|
// printf("%d\n", numEq);
|
|
|
|
|
numEq= (256 - (numEq & 0xFF)) &0xFF;
|
|
|
|
|
#else
|
2001-10-12 00:35:45 +02:00
|
|
|
|
int y;
|
|
|
|
|
for(y=0; y<BLOCK_SIZE; y++)
|
2001-10-11 00:13:27 +02:00
|
|
|
|
{
|
|
|
|
|
if(((src[0] - src[1] + 1) & 0xFFFF) < 3) numEq++;
|
|
|
|
|
if(((src[1] - src[2] + 1) & 0xFFFF) < 3) numEq++;
|
|
|
|
|
if(((src[2] - src[3] + 1) & 0xFFFF) < 3) numEq++;
|
|
|
|
|
if(((src[3] - src[4] + 1) & 0xFFFF) < 3) numEq++;
|
|
|
|
|
if(((src[4] - src[5] + 1) & 0xFFFF) < 3) numEq++;
|
|
|
|
|
if(((src[5] - src[6] + 1) & 0xFFFF) < 3) numEq++;
|
|
|
|
|
if(((src[6] - src[7] + 1) & 0xFFFF) < 3) numEq++;
|
|
|
|
|
tempBlock[0 + y*TEMP_STRIDE] = src[0];
|
|
|
|
|
tempBlock[1 + y*TEMP_STRIDE] = src[1];
|
|
|
|
|
tempBlock[2 + y*TEMP_STRIDE] = src[2];
|
|
|
|
|
tempBlock[3 + y*TEMP_STRIDE] = src[3];
|
|
|
|
|
tempBlock[4 + y*TEMP_STRIDE] = src[4];
|
|
|
|
|
tempBlock[5 + y*TEMP_STRIDE] = src[5];
|
|
|
|
|
tempBlock[6 + y*TEMP_STRIDE] = src[6];
|
|
|
|
|
tempBlock[7 + y*TEMP_STRIDE] = src[7];
|
|
|
|
|
src+= stride;
|
|
|
|
|
}
|
|
|
|
|
#endif
|
|
|
|
|
/* if(abs(numEq - asmEq) > 0)
|
|
|
|
|
{
|
|
|
|
|
// printf("\nasm:%d c:%d\n", asmEq, numEq);
|
|
|
|
|
for(int y=0; y<8; y++)
|
|
|
|
|
{
|
|
|
|
|
for(int x=0; x<8; x++)
|
|
|
|
|
{
|
|
|
|
|
printf("%d ", src[x + y*stride]);
|
|
|
|
|
}
|
|
|
|
|
printf("\n");
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
*/
|
|
|
|
|
// printf("%d\n", numEq);
|
|
|
|
|
return numEq > hFlatnessThreshold;
|
|
|
|
|
}
|
|
|
|
|
|
2001-10-12 00:35:45 +02:00
|
|
|
|
static inline int isHorizMinMaxOk(uint8_t src[], int stride, int QP)
|
2001-10-11 00:13:27 +02:00
|
|
|
|
{
|
|
|
|
|
#ifdef MMX_FIXME
|
|
|
|
|
FIXME
|
|
|
|
|
int isOk;
|
|
|
|
|
asm volatile(
|
|
|
|
|
// "int $3 \n\t"
|
|
|
|
|
"movq (%1, %2), %%mm0 \n\t"
|
|
|
|
|
"movq (%1, %2, 8), %%mm1 \n\t"
|
|
|
|
|
"movq %%mm0, %%mm2 \n\t"
|
|
|
|
|
"psubusb %%mm1, %%mm0 \n\t"
|
|
|
|
|
"psubusb %%mm2, %%mm1 \n\t"
|
|
|
|
|
"por %%mm1, %%mm0 \n\t" // ABS Diff
|
|
|
|
|
|
|
|
|
|
"movq pQPb, %%mm7 \n\t" // QP,..., QP
|
|
|
|
|
"paddusb %%mm7, %%mm7 \n\t" // 2QP ... 2QP
|
|
|
|
|
"psubusb %%mm7, %%mm0 \n\t" // Diff <= 2QP -> 0
|
|
|
|
|
"pcmpeqd b00, %%mm0 \n\t"
|
|
|
|
|
"psrlq $16, %%mm0 \n\t"
|
|
|
|
|
"pcmpeqd bFF, %%mm0 \n\t"
|
|
|
|
|
// "movd %%mm0, (%1, %2, 4)\n\t"
|
|
|
|
|
"movd %%mm0, %0 \n\t"
|
|
|
|
|
: "=r" (isOk)
|
|
|
|
|
: "r" (src), "r" (stride)
|
|
|
|
|
);
|
|
|
|
|
return isOk;
|
|
|
|
|
#else
|
2001-10-12 00:35:45 +02:00
|
|
|
|
if(abs(src[0] - src[7]) > 2*QP) return 0;
|
2001-10-11 00:13:27 +02:00
|
|
|
|
|
2001-10-12 00:35:45 +02:00
|
|
|
|
return 1;
|
2001-10-11 00:13:27 +02:00
|
|
|
|
#endif
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static inline void doHorizDefFilterAndCopyBack(uint8_t dst[], int stride, int QP)
|
|
|
|
|
{
|
2001-10-11 00:21:19 +02:00
|
|
|
|
#ifdef HAVE_MMX
|
2001-10-11 00:13:27 +02:00
|
|
|
|
asm volatile(
|
|
|
|
|
"pushl %0 \n\t"
|
|
|
|
|
"pxor %%mm7, %%mm7 \n\t"
|
|
|
|
|
"movq bm00001000, %%mm6 \n\t"
|
|
|
|
|
"movd %2, %%mm5 \n\t" // QP
|
|
|
|
|
"movq %%mm5, %%mm4 \n\t"
|
|
|
|
|
"paddusb %%mm5, %%mm5 \n\t" // 2QP
|
|
|
|
|
"paddusb %%mm5, %%mm4 \n\t" // 3QP
|
|
|
|
|
"psllq $24, %%mm4 \n\t"
|
|
|
|
|
"pxor %%mm5, %%mm5 \n\t" // 0
|
|
|
|
|
"psubb %%mm4, %%mm5 \n\t" // -QP
|
|
|
|
|
"leal tempBlock, %%eax \n\t"
|
|
|
|
|
|
|
|
|
|
//FIXME? "unroll by 2" and mix
|
2001-10-11 00:21:19 +02:00
|
|
|
|
#ifdef HAVE_MMX2
|
|
|
|
|
#define HDF(i) \
|
|
|
|
|
"movq " #i "(%%eax), %%mm0 \n\t"\
|
|
|
|
|
"movq %%mm0, %%mm1 \n\t"\
|
|
|
|
|
"movq %%mm0, %%mm2 \n\t"\
|
|
|
|
|
"psrlq $8, %%mm1 \n\t"\
|
|
|
|
|
"psubusb %%mm1, %%mm2 \n\t"\
|
|
|
|
|
"psubusb %%mm0, %%mm1 \n\t"\
|
|
|
|
|
"por %%mm2, %%mm1 \n\t" /* p<>x = |px - p(x+1)| */\
|
|
|
|
|
"pcmpeqb %%mm7, %%mm2 \n\t" /* p<>x = sgn[px - p(x+1)] */\
|
|
|
|
|
"pshufw $0x00, %%mm1, %%mm3 \n\t" /* p<>5 = |p1 - p2| */\
|
|
|
|
|
"pminub %%mm1, %%mm3 \n\t" /* p<>5 = min(|p2-p1|, |p6-p5|)*/\
|
|
|
|
|
"psrlq $16, %%mm3 \n\t" /* p<>3 = min(|p2-p1|, |p6-p5|)*/\
|
|
|
|
|
"psubusb %%mm3, %%mm1 \n\t" /* |p3-p4|-min(|p1-p2|,|p5-p6|) */\
|
|
|
|
|
"paddb %%mm5, %%mm1 \n\t"\
|
|
|
|
|
"psubusb %%mm5, %%mm1 \n\t"\
|
|
|
|
|
"psrlw $2, %%mm1 \n\t"\
|
|
|
|
|
"pxor %%mm2, %%mm1 \n\t"\
|
|
|
|
|
"psubb %%mm2, %%mm1 \n\t"\
|
|
|
|
|
"pand %%mm6, %%mm1 \n\t"\
|
|
|
|
|
"psubb %%mm1, %%mm0 \n\t"\
|
|
|
|
|
"psllq $8, %%mm1 \n\t"\
|
|
|
|
|
"paddb %%mm1, %%mm0 \n\t"\
|
|
|
|
|
"movd %%mm0, (%0) \n\t"\
|
|
|
|
|
"psrlq $32, %%mm0 \n\t"\
|
|
|
|
|
"movd %%mm0, 4(%0) \n\t"
|
|
|
|
|
#else
|
|
|
|
|
#define HDF(i)\
|
|
|
|
|
"movq " #i "(%%eax), %%mm0 \n\t"\
|
2001-10-11 00:13:27 +02:00
|
|
|
|
"movq %%mm0, %%mm1 \n\t"\
|
|
|
|
|
"movq %%mm0, %%mm2 \n\t"\
|
|
|
|
|
"psrlq $8, %%mm1 \n\t"\
|
|
|
|
|
"psubusb %%mm1, %%mm2 \n\t"\
|
|
|
|
|
"psubusb %%mm0, %%mm1 \n\t"\
|
2001-10-11 00:21:19 +02:00
|
|
|
|
"por %%mm2, %%mm1 \n\t" /* p<>x = |px - p(x+1)| */\
|
|
|
|
|
"pcmpeqb %%mm7, %%mm2 \n\t" /* p<>x = sgn[px - p(x+1)] */\
|
|
|
|
|
"movq %%mm1, %%mm3 \n\t"\
|
|
|
|
|
"psllq $32, %%mm3 \n\t"\
|
|
|
|
|
"movq %%mm3, %%mm4 \n\t"\
|
|
|
|
|
"psubusb %%mm1, %%mm4 \n\t"\
|
|
|
|
|
"psubb %%mm4, %%mm3 \n\t"\
|
|
|
|
|
"psrlq $16, %%mm3 \n\t" /* p<>3 = min(|p2-p1|, |p6-p5|)*/\
|
2001-10-11 00:13:27 +02:00
|
|
|
|
"psubusb %%mm3, %%mm1 \n\t" /* |p3-p4|-min(|p1-p2|,|p5,<2C>6|) */\
|
|
|
|
|
"paddb %%mm5, %%mm1 \n\t"\
|
|
|
|
|
"psubusb %%mm5, %%mm1 \n\t"\
|
|
|
|
|
"psrlw $2, %%mm1 \n\t"\
|
|
|
|
|
"pxor %%mm2, %%mm1 \n\t"\
|
|
|
|
|
"psubb %%mm2, %%mm1 \n\t"\
|
|
|
|
|
"pand %%mm6, %%mm1 \n\t"\
|
|
|
|
|
"psubb %%mm1, %%mm0 \n\t"\
|
|
|
|
|
"psllq $8, %%mm1 \n\t"\
|
|
|
|
|
"paddb %%mm1, %%mm0 \n\t"\
|
|
|
|
|
"movd %%mm0, (%0) \n\t"\
|
|
|
|
|
"psrlq $32, %%mm0 \n\t"\
|
|
|
|
|
"movd %%mm0, 4(%0) \n\t"
|
2001-10-11 00:21:19 +02:00
|
|
|
|
#endif
|
2001-10-11 00:13:27 +02:00
|
|
|
|
HDF(0)
|
|
|
|
|
"addl %1, %0 \n\t"
|
|
|
|
|
HDF(8)
|
|
|
|
|
"addl %1, %0 \n\t"
|
|
|
|
|
HDF(16)
|
|
|
|
|
"addl %1, %0 \n\t"
|
|
|
|
|
HDF(24)
|
|
|
|
|
"addl %1, %0 \n\t"
|
|
|
|
|
HDF(32)
|
|
|
|
|
"addl %1, %0 \n\t"
|
|
|
|
|
HDF(40)
|
|
|
|
|
"addl %1, %0 \n\t"
|
|
|
|
|
HDF(48)
|
|
|
|
|
"addl %1, %0 \n\t"
|
|
|
|
|
HDF(56)
|
|
|
|
|
"popl %0 \n\t"
|
|
|
|
|
:
|
|
|
|
|
: "r" (dst), "r" (stride), "r" (QP)
|
|
|
|
|
: "%eax"
|
|
|
|
|
);
|
|
|
|
|
#else
|
|
|
|
|
uint8_t *src= tempBlock;
|
|
|
|
|
|
2001-10-12 00:35:45 +02:00
|
|
|
|
int y;
|
|
|
|
|
for(y=0; y<BLOCK_SIZE; y++)
|
2001-10-11 00:13:27 +02:00
|
|
|
|
{
|
2001-10-14 19:11:17 +02:00
|
|
|
|
const int middleEnergy= 5*(src[4] - src[5]) + 2*(src[2] - src[5]);
|
|
|
|
|
|
2001-10-11 00:13:27 +02:00
|
|
|
|
dst[0] = src[0];
|
|
|
|
|
dst[1] = src[1];
|
|
|
|
|
dst[2] = src[2];
|
|
|
|
|
dst[3] = src[3];
|
|
|
|
|
dst[4] = src[4];
|
|
|
|
|
dst[5] = src[5];
|
|
|
|
|
dst[6] = src[6];
|
|
|
|
|
dst[7] = src[7];
|
|
|
|
|
|
|
|
|
|
if(ABS(middleEnergy) < 8*QP)
|
|
|
|
|
{
|
|
|
|
|
const int q=(src[3] - src[4])/2;
|
|
|
|
|
const int leftEnergy= 5*(src[2] - src[1]) + 2*(src[0] - src[3]);
|
|
|
|
|
const int rightEnergy= 5*(src[6] - src[5]) + 2*(src[4] - src[7]);
|
|
|
|
|
|
|
|
|
|
int d= ABS(middleEnergy) - MIN( ABS(leftEnergy), ABS(rightEnergy) );
|
|
|
|
|
d= MAX(d, 0);
|
|
|
|
|
|
|
|
|
|
d= (5*d + 32) >> 6;
|
|
|
|
|
d*= SIGN(-middleEnergy);
|
|
|
|
|
|
|
|
|
|
if(q>0)
|
|
|
|
|
{
|
|
|
|
|
d= d<0 ? 0 : d;
|
|
|
|
|
d= d>q ? q : d;
|
|
|
|
|
}
|
|
|
|
|
else
|
|
|
|
|
{
|
|
|
|
|
d= d>0 ? 0 : d;
|
|
|
|
|
d= d<q ? q : d;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
dst[3]-= d;
|
|
|
|
|
dst[4]+= d;
|
|
|
|
|
}
|
|
|
|
|
dst+= stride;
|
|
|
|
|
src+= TEMP_STRIDE;
|
|
|
|
|
}
|
|
|
|
|
#endif
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/**
|
2001-10-15 05:01:08 +02:00
|
|
|
|
* Do a horizontal low pass filter on the 10x8 block (dst points to middle 8x8 Block)
|
2001-10-16 04:31:14 +02:00
|
|
|
|
* using the 9-Tap Filter (1,1,2,2,4,2,2,1,1)/16 (C version)
|
|
|
|
|
* using the 7-Tap Filter (2,2,2,4,2,2,2)/16 (MMX2/3DNOW version)
|
2001-10-11 00:13:27 +02:00
|
|
|
|
*/
|
|
|
|
|
static inline void doHorizLowPassAndCopyBack(uint8_t dst[], int stride, int QP)
|
|
|
|
|
{
|
|
|
|
|
//return;
|
2001-10-11 00:21:19 +02:00
|
|
|
|
#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
|
2001-10-11 00:13:27 +02:00
|
|
|
|
asm volatile( //"movv %0 %1 %2\n\t"
|
|
|
|
|
"pushl %0\n\t"
|
|
|
|
|
"pxor %%mm7, %%mm7 \n\t"
|
|
|
|
|
"leal tempBlock, %%eax \n\t"
|
2001-10-11 00:21:19 +02:00
|
|
|
|
/*
|
2001-10-11 00:13:27 +02:00
|
|
|
|
#define HLP1 "movq (%0), %%mm0 \n\t"\
|
|
|
|
|
"movq %%mm0, %%mm1 \n\t"\
|
|
|
|
|
"psllq $8, %%mm0 \n\t"\
|
2001-10-11 00:21:19 +02:00
|
|
|
|
PAVGB(%%mm1, %%mm0)\
|
2001-10-11 00:13:27 +02:00
|
|
|
|
"psrlw $8, %%mm0 \n\t"\
|
|
|
|
|
"pxor %%mm1, %%mm1 \n\t"\
|
|
|
|
|
"packuswb %%mm1, %%mm0 \n\t"\
|
|
|
|
|
"movq %%mm0, %%mm1 \n\t"\
|
|
|
|
|
"movq %%mm0, %%mm2 \n\t"\
|
|
|
|
|
"psllq $32, %%mm0 \n\t"\
|
|
|
|
|
"paddb %%mm0, %%mm1 \n\t"\
|
|
|
|
|
"psllq $16, %%mm2 \n\t"\
|
2001-10-11 00:21:19 +02:00
|
|
|
|
PAVGB(%%mm2, %%mm0)\
|
2001-10-11 00:13:27 +02:00
|
|
|
|
"movq %%mm0, %%mm3 \n\t"\
|
|
|
|
|
"pand bm11001100, %%mm0 \n\t"\
|
|
|
|
|
"paddusb %%mm0, %%mm3 \n\t"\
|
|
|
|
|
"psrlq $8, %%mm3 \n\t"\
|
2001-10-11 00:21:19 +02:00
|
|
|
|
PAVGB(%%mm1, %%mm4)\
|
|
|
|
|
PAVGB(%%mm3, %%mm2)\
|
2001-10-11 00:13:27 +02:00
|
|
|
|
"psrlq $16, %%mm2 \n\t"\
|
|
|
|
|
"punpcklbw %%mm2, %%mm2 \n\t"\
|
|
|
|
|
"movq %%mm2, (%0) \n\t"\
|
|
|
|
|
|
|
|
|
|
#define HLP2 "movq (%0), %%mm0 \n\t"\
|
|
|
|
|
"movq %%mm0, %%mm1 \n\t"\
|
|
|
|
|
"psllq $8, %%mm0 \n\t"\
|
2001-10-11 00:21:19 +02:00
|
|
|
|
PAVGB(%%mm1, %%mm0)\
|
2001-10-11 00:13:27 +02:00
|
|
|
|
"psrlw $8, %%mm0 \n\t"\
|
|
|
|
|
"pxor %%mm1, %%mm1 \n\t"\
|
|
|
|
|
"packuswb %%mm1, %%mm0 \n\t"\
|
|
|
|
|
"movq %%mm0, %%mm2 \n\t"\
|
|
|
|
|
"psllq $32, %%mm0 \n\t"\
|
|
|
|
|
"psllq $16, %%mm2 \n\t"\
|
2001-10-11 00:21:19 +02:00
|
|
|
|
PAVGB(%%mm2, %%mm0)\
|
2001-10-11 00:13:27 +02:00
|
|
|
|
"movq %%mm0, %%mm3 \n\t"\
|
|
|
|
|
"pand bm11001100, %%mm0 \n\t"\
|
|
|
|
|
"paddusb %%mm0, %%mm3 \n\t"\
|
|
|
|
|
"psrlq $8, %%mm3 \n\t"\
|
2001-10-11 00:21:19 +02:00
|
|
|
|
PAVGB(%%mm3, %%mm2)\
|
2001-10-11 00:13:27 +02:00
|
|
|
|
"psrlq $16, %%mm2 \n\t"\
|
|
|
|
|
"punpcklbw %%mm2, %%mm2 \n\t"\
|
|
|
|
|
"movq %%mm2, (%0) \n\t"\
|
2001-10-11 00:21:19 +02:00
|
|
|
|
*/
|
2001-10-11 00:13:27 +02:00
|
|
|
|
// approximately a 7-Tap Filter with Vector (1,2,3,4,3,2,1)/16
|
|
|
|
|
/*
|
|
|
|
|
Implemented Exact 7-Tap
|
|
|
|
|
9421 A321
|
|
|
|
|
36421 64321
|
|
|
|
|
334321 =
|
|
|
|
|
1234321 =
|
|
|
|
|
1234321 =
|
|
|
|
|
123433 =
|
|
|
|
|
12463 12346
|
|
|
|
|
1249 123A
|
|
|
|
|
|
|
|
|
|
*/
|
2001-10-15 05:01:08 +02:00
|
|
|
|
|
2001-10-11 00:21:19 +02:00
|
|
|
|
#ifdef HAVE_MMX2
|
2001-10-11 00:13:27 +02:00
|
|
|
|
#define HLP3(i) "movq " #i "(%%eax), %%mm0 \n\t"\
|
|
|
|
|
"movq %%mm0, %%mm1 \n\t"\
|
|
|
|
|
"movq %%mm0, %%mm2 \n\t"\
|
|
|
|
|
"movq %%mm0, %%mm3 \n\t"\
|
|
|
|
|
"movq %%mm0, %%mm4 \n\t"\
|
|
|
|
|
"psllq $8, %%mm1 \n\t"\
|
|
|
|
|
"psrlq $8, %%mm2 \n\t"\
|
|
|
|
|
"pand bm00000001, %%mm3 \n\t"\
|
|
|
|
|
"pand bm10000000, %%mm4 \n\t"\
|
|
|
|
|
"por %%mm3, %%mm1 \n\t"\
|
|
|
|
|
"por %%mm4, %%mm2 \n\t"\
|
2001-10-11 00:21:19 +02:00
|
|
|
|
PAVGB(%%mm2, %%mm1)\
|
|
|
|
|
PAVGB(%%mm1, %%mm0)\
|
2001-10-11 00:13:27 +02:00
|
|
|
|
\
|
|
|
|
|
"pshufw $0xF9, %%mm0, %%mm3 \n\t"\
|
|
|
|
|
"pshufw $0x90, %%mm0, %%mm4 \n\t"\
|
2001-10-11 00:21:19 +02:00
|
|
|
|
PAVGB(%%mm3, %%mm4)\
|
|
|
|
|
PAVGB(%%mm4, %%mm0)\
|
2001-10-11 00:13:27 +02:00
|
|
|
|
"movd %%mm0, (%0) \n\t"\
|
|
|
|
|
"psrlq $32, %%mm0 \n\t"\
|
2001-10-11 00:21:19 +02:00
|
|
|
|
"movd %%mm0, 4(%0) \n\t"
|
|
|
|
|
#else
|
|
|
|
|
#define HLP3(i) "movq " #i "(%%eax), %%mm0 \n\t"\
|
|
|
|
|
"movq %%mm0, %%mm1 \n\t"\
|
|
|
|
|
"movq %%mm0, %%mm2 \n\t"\
|
2001-10-15 05:01:08 +02:00
|
|
|
|
"movd -4(%0), %%mm3 \n\t" /*0001000*/\
|
|
|
|
|
"movd 8(%0), %%mm4 \n\t" /*0001000*/\
|
2001-10-11 00:21:19 +02:00
|
|
|
|
"psllq $8, %%mm1 \n\t"\
|
|
|
|
|
"psrlq $8, %%mm2 \n\t"\
|
2001-10-15 05:01:08 +02:00
|
|
|
|
"psrlq $24, %%mm3 \n\t"\
|
|
|
|
|
"psllq $56, %%mm4 \n\t"\
|
2001-10-11 00:21:19 +02:00
|
|
|
|
"por %%mm3, %%mm1 \n\t"\
|
|
|
|
|
"por %%mm4, %%mm2 \n\t"\
|
|
|
|
|
PAVGB(%%mm2, %%mm1)\
|
|
|
|
|
PAVGB(%%mm1, %%mm0)\
|
|
|
|
|
\
|
|
|
|
|
"movq %%mm0, %%mm3 \n\t"\
|
|
|
|
|
"movq %%mm0, %%mm4 \n\t"\
|
|
|
|
|
"movq %%mm0, %%mm5 \n\t"\
|
|
|
|
|
"psrlq $16, %%mm3 \n\t"\
|
|
|
|
|
"psllq $16, %%mm4 \n\t"\
|
|
|
|
|
"pand bm11000000, %%mm5 \n\t"\
|
|
|
|
|
"por %%mm5, %%mm3 \n\t"\
|
|
|
|
|
"movq %%mm0, %%mm5 \n\t"\
|
|
|
|
|
"pand bm00000011, %%mm5 \n\t"\
|
|
|
|
|
"por %%mm5, %%mm4 \n\t"\
|
|
|
|
|
PAVGB(%%mm3, %%mm4)\
|
|
|
|
|
PAVGB(%%mm4, %%mm0)\
|
|
|
|
|
"movd %%mm0, (%0) \n\t"\
|
|
|
|
|
"psrlq $32, %%mm0 \n\t"\
|
|
|
|
|
"movd %%mm0, 4(%0) \n\t"
|
|
|
|
|
#endif
|
2001-10-11 00:13:27 +02:00
|
|
|
|
|
2001-10-15 05:01:08 +02:00
|
|
|
|
/* uses the 7-Tap Filter: 1112111 */
|
|
|
|
|
#define NEW_HLP(i)\
|
|
|
|
|
"movq " #i "(%%eax), %%mm0 \n\t"\
|
|
|
|
|
"movq %%mm0, %%mm1 \n\t"\
|
|
|
|
|
"movq %%mm0, %%mm2 \n\t"\
|
|
|
|
|
"movd -4(%0), %%mm3 \n\t" /*0001000*/\
|
|
|
|
|
"movd 8(%0), %%mm4 \n\t" /*0001000*/\
|
|
|
|
|
"psllq $8, %%mm1 \n\t"\
|
|
|
|
|
"psrlq $8, %%mm2 \n\t"\
|
|
|
|
|
"psrlq $24, %%mm3 \n\t"\
|
|
|
|
|
"psllq $56, %%mm4 \n\t"\
|
|
|
|
|
"por %%mm3, %%mm1 \n\t"\
|
|
|
|
|
"por %%mm4, %%mm2 \n\t"\
|
|
|
|
|
"movq %%mm1, %%mm5 \n\t"\
|
|
|
|
|
PAVGB(%%mm2, %%mm1)\
|
|
|
|
|
PAVGB(%%mm1, %%mm0)\
|
|
|
|
|
"psllq $8, %%mm5 \n\t"\
|
|
|
|
|
"psrlq $8, %%mm2 \n\t"\
|
|
|
|
|
"por %%mm3, %%mm5 \n\t"\
|
|
|
|
|
"por %%mm4, %%mm2 \n\t"\
|
|
|
|
|
"movq %%mm5, %%mm1 \n\t"\
|
|
|
|
|
PAVGB(%%mm2, %%mm5)\
|
|
|
|
|
"psllq $8, %%mm1 \n\t"\
|
|
|
|
|
"psrlq $8, %%mm2 \n\t"\
|
|
|
|
|
"por %%mm3, %%mm1 \n\t"\
|
|
|
|
|
"por %%mm4, %%mm2 \n\t"\
|
|
|
|
|
PAVGB(%%mm2, %%mm1)\
|
|
|
|
|
PAVGB(%%mm1, %%mm5)\
|
|
|
|
|
PAVGB(%%mm5, %%mm0)\
|
|
|
|
|
"movd %%mm0, (%0) \n\t"\
|
|
|
|
|
"psrlq $32, %%mm0 \n\t"\
|
|
|
|
|
"movd %%mm0, 4(%0) \n\t"
|
|
|
|
|
|
|
|
|
|
/* uses the 9-Tap Filter: 112242211 */
|
|
|
|
|
#define NEW_HLP2(i)\
|
|
|
|
|
"movq " #i "(%%eax), %%mm0 \n\t" /*0001000*/\
|
|
|
|
|
"movq %%mm0, %%mm1 \n\t" /*0001000*/\
|
|
|
|
|
"movq %%mm0, %%mm2 \n\t" /*0001000*/\
|
|
|
|
|
"movd -4(%0), %%mm3 \n\t" /*0001000*/\
|
|
|
|
|
"movd 8(%0), %%mm4 \n\t" /*0001000*/\
|
|
|
|
|
"psllq $8, %%mm1 \n\t"\
|
|
|
|
|
"psrlq $8, %%mm2 \n\t"\
|
|
|
|
|
"psrlq $24, %%mm3 \n\t"\
|
|
|
|
|
"psllq $56, %%mm4 \n\t"\
|
|
|
|
|
"por %%mm3, %%mm1 \n\t" /*0010000*/\
|
|
|
|
|
"por %%mm4, %%mm2 \n\t" /*0000100*/\
|
|
|
|
|
"movq %%mm1, %%mm5 \n\t" /*0010000*/\
|
|
|
|
|
PAVGB(%%mm2, %%mm1) /*0010100*/\
|
|
|
|
|
PAVGB(%%mm1, %%mm0) /*0012100*/\
|
|
|
|
|
"psllq $8, %%mm5 \n\t"\
|
|
|
|
|
"psrlq $8, %%mm2 \n\t"\
|
|
|
|
|
"por %%mm3, %%mm5 \n\t" /*0100000*/\
|
|
|
|
|
"por %%mm4, %%mm2 \n\t" /*0000010*/\
|
|
|
|
|
"movq %%mm5, %%mm1 \n\t" /*0100000*/\
|
|
|
|
|
PAVGB(%%mm2, %%mm5) /*0100010*/\
|
|
|
|
|
"psllq $8, %%mm1 \n\t"\
|
|
|
|
|
"psrlq $8, %%mm2 \n\t"\
|
|
|
|
|
"por %%mm3, %%mm1 \n\t" /*1000000*/\
|
|
|
|
|
"por %%mm4, %%mm2 \n\t" /*0000001*/\
|
|
|
|
|
"movq %%mm1, %%mm6 \n\t" /*1000000*/\
|
|
|
|
|
PAVGB(%%mm2, %%mm1) /*1000001*/\
|
|
|
|
|
"psllq $8, %%mm6 \n\t"\
|
|
|
|
|
"psrlq $8, %%mm2 \n\t"\
|
|
|
|
|
"por %%mm3, %%mm6 \n\t"/*100000000*/\
|
|
|
|
|
"por %%mm4, %%mm2 \n\t"/*000000001*/\
|
|
|
|
|
PAVGB(%%mm2, %%mm6) /*100000001*/\
|
|
|
|
|
PAVGB(%%mm6, %%mm1) /*110000011*/\
|
|
|
|
|
PAVGB(%%mm1, %%mm5) /*112000211*/\
|
|
|
|
|
PAVGB(%%mm5, %%mm0) /*112242211*/\
|
|
|
|
|
"movd %%mm0, (%0) \n\t"\
|
|
|
|
|
"psrlq $32, %%mm0 \n\t"\
|
|
|
|
|
"movd %%mm0, 4(%0) \n\t"
|
|
|
|
|
|
|
|
|
|
#define HLP(i) NEW_HLP(i)
|
2001-10-11 00:13:27 +02:00
|
|
|
|
|
|
|
|
|
HLP(0)
|
|
|
|
|
"addl %1, %0 \n\t"
|
|
|
|
|
HLP(8)
|
|
|
|
|
"addl %1, %0 \n\t"
|
|
|
|
|
HLP(16)
|
|
|
|
|
"addl %1, %0 \n\t"
|
|
|
|
|
HLP(24)
|
|
|
|
|
"addl %1, %0 \n\t"
|
|
|
|
|
HLP(32)
|
|
|
|
|
"addl %1, %0 \n\t"
|
|
|
|
|
HLP(40)
|
|
|
|
|
"addl %1, %0 \n\t"
|
|
|
|
|
HLP(48)
|
|
|
|
|
"addl %1, %0 \n\t"
|
|
|
|
|
HLP(56)
|
|
|
|
|
|
|
|
|
|
"popl %0\n\t"
|
|
|
|
|
:
|
|
|
|
|
: "r" (dst), "r" (stride)
|
|
|
|
|
: "%eax", "%ebx"
|
|
|
|
|
);
|
|
|
|
|
|
|
|
|
|
#else
|
|
|
|
|
uint8_t *temp= tempBlock;
|
2001-10-12 00:35:45 +02:00
|
|
|
|
int y;
|
|
|
|
|
for(y=0; y<BLOCK_SIZE; y++)
|
2001-10-11 00:13:27 +02:00
|
|
|
|
{
|
|
|
|
|
const int first= ABS(dst[-1] - dst[0]) < QP ? dst[-1] : dst[0];
|
|
|
|
|
const int last= ABS(dst[8] - dst[7]) < QP ? dst[8] : dst[7];
|
|
|
|
|
|
|
|
|
|
int sums[9];
|
|
|
|
|
sums[0] = first + temp[0];
|
|
|
|
|
sums[1] = temp[0] + temp[1];
|
|
|
|
|
sums[2] = temp[1] + temp[2];
|
|
|
|
|
sums[3] = temp[2] + temp[3];
|
|
|
|
|
sums[4] = temp[3] + temp[4];
|
|
|
|
|
sums[5] = temp[4] + temp[5];
|
|
|
|
|
sums[6] = temp[5] + temp[6];
|
|
|
|
|
sums[7] = temp[6] + temp[7];
|
|
|
|
|
sums[8] = temp[7] + last;
|
|
|
|
|
|
|
|
|
|
dst[0]= ((sums[0]<<2) + ((first + sums[2])<<1) + sums[4] + 8)>>4;
|
|
|
|
|
dst[1]= ((dst[1]<<2) + (first + sums[0] + sums[3]<<1) + sums[5] + 8)>>4;
|
|
|
|
|
dst[2]= ((dst[2]<<2) + (first + sums[1] + sums[4]<<1) + sums[6] + 8)>>4;
|
|
|
|
|
dst[3]= ((dst[3]<<2) + (sums[2] + sums[5]<<1) + sums[0] + sums[7] + 8)>>4;
|
|
|
|
|
dst[4]= ((dst[4]<<2) + (sums[3] + sums[6]<<1) + sums[1] + sums[8] + 8)>>4;
|
|
|
|
|
dst[5]= ((dst[5]<<2) + (last + sums[7] + sums[4]<<1) + sums[2] + 8)>>4;
|
|
|
|
|
dst[6]= ((last + dst[6]<<2) + (dst[7] + sums[5]<<1) + sums[3] + 8)>>4;
|
|
|
|
|
dst[7]= ((sums[8]<<2) + (last + sums[6]<<1) + sums[4] + 8)>>4;
|
|
|
|
|
|
|
|
|
|
dst+= stride;
|
|
|
|
|
temp+= TEMP_STRIDE;
|
|
|
|
|
}
|
|
|
|
|
#endif
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
static inline void dering(uint8_t src[], int stride, int QP)
|
|
|
|
|
{
|
|
|
|
|
//FIXME
|
|
|
|
|
|
|
|
|
|
#ifdef HAVE_MMX2X
|
|
|
|
|
asm volatile(
|
|
|
|
|
"leal (%0, %1), %%eax \n\t"
|
|
|
|
|
"leal (%%eax, %1, 4), %%ebx \n\t"
|
|
|
|
|
// 0 1 2 3 4 5 6 7 8 9
|
|
|
|
|
// %0 eax eax+%1 eax+2%1 %0+4%1 ebx ebx+%1 ebx+2%1 %0+8%1 ebx+4%1
|
|
|
|
|
|
|
|
|
|
"pcmpeq %%mm6, %%mm6 \n\t"
|
|
|
|
|
"pxor %%mm7, %%mm7 \n\t"
|
|
|
|
|
|
|
|
|
|
#define FIND_MIN_MAX(addr)\
|
|
|
|
|
"movq (" #addr "), %%mm0, \n\t"\
|
|
|
|
|
"pminub %%mm0, %%mm6 \n\t"\
|
|
|
|
|
"pmaxub %%mm0, %%mm7 \n\t"
|
|
|
|
|
|
|
|
|
|
FIND_MIN_MAX(%0)
|
|
|
|
|
FIND_MIN_MAX(%%eax)
|
|
|
|
|
FIND_MIN_MAX(%%eax, %1)
|
|
|
|
|
FIND_MIN_MAX(%%eax, %1, 2)
|
|
|
|
|
FIND_MIN_MAX(%0, %1, 4)
|
|
|
|
|
FIND_MIN_MAX(%%ebx)
|
|
|
|
|
FIND_MIN_MAX(%%ebx, %1)
|
|
|
|
|
FIND_MIN_MAX(%%ebx, %1, 2)
|
|
|
|
|
FIND_MIN_MAX(%0, %1, 8)
|
|
|
|
|
FIND_MIN_MAX(%%ebx, %1, 2)
|
|
|
|
|
|
|
|
|
|
"movq %%mm6, %%mm4 \n\t"
|
|
|
|
|
"psrlq $32, %%mm6 \n\t"
|
|
|
|
|
"pminub %%mm4, %%mm6 \n\t"
|
|
|
|
|
"movq %%mm6, %%mm4 \n\t"
|
|
|
|
|
"psrlq $16, %%mm6 \n\t"
|
|
|
|
|
"pminub %%mm4, %%mm6 \n\t"
|
|
|
|
|
"movq %%mm6, %%mm4 \n\t"
|
|
|
|
|
"psrlq $8, %%mm6 \n\t"
|
|
|
|
|
"pminub %%mm4, %%mm6 \n\t" // min of pixels
|
|
|
|
|
|
|
|
|
|
"movq %%mm7, %%mm4 \n\t"
|
|
|
|
|
"psrlq $32, %%mm7 \n\t"
|
|
|
|
|
"pmaxub %%mm4, %%mm7 \n\t"
|
|
|
|
|
"movq %%mm7, %%mm4 \n\t"
|
|
|
|
|
"psrlq $16, %%mm7 \n\t"
|
|
|
|
|
"pmaxub %%mm4, %%mm7 \n\t"
|
|
|
|
|
"movq %%mm7, %%mm4 \n\t"
|
|
|
|
|
"psrlq $8, %%mm7 \n\t"
|
|
|
|
|
"pmaxub %%mm4, %%mm7 \n\t" // max of pixels
|
2001-10-11 00:21:19 +02:00
|
|
|
|
PAVGB(%%mm6, %%mm7) // (max + min)/2
|
2001-10-11 00:13:27 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
: : "r" (src), "r" (stride), "r" (QP)
|
|
|
|
|
: "%eax", "%ebx"
|
|
|
|
|
);
|
|
|
|
|
#else
|
|
|
|
|
|
|
|
|
|
//FIXME
|
|
|
|
|
#endif
|
|
|
|
|
}
|
|
|
|
|
|
2001-10-15 05:01:08 +02:00
|
|
|
|
/**
|
|
|
|
|
* Deinterlaces the given block
|
2001-10-17 22:42:07 +02:00
|
|
|
|
* will be called for every 8x8 block, and can read & write into an 8x16 block
|
2001-10-15 05:01:08 +02:00
|
|
|
|
*/
|
|
|
|
|
static inline void deInterlaceInterpolateLinear(uint8_t src[], int stride)
|
|
|
|
|
{
|
|
|
|
|
#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
|
|
|
|
|
asm volatile(
|
|
|
|
|
"leal (%0, %1), %%eax \n\t"
|
|
|
|
|
"leal (%%eax, %1, 4), %%ebx \n\t"
|
|
|
|
|
// 0 1 2 3 4 5 6 7 8 9
|
|
|
|
|
// %0 eax eax+%1 eax+2%1 %0+4%1 ebx ebx+%1 ebx+2%1 %0+8%1 ebx+4%1
|
|
|
|
|
|
|
|
|
|
"movq (%0), %%mm0 \n\t"
|
|
|
|
|
"movq (%%eax, %1), %%mm1 \n\t"
|
2001-10-17 22:42:07 +02:00
|
|
|
|
PAVGB(%%mm1, %%mm0)
|
2001-10-15 05:01:08 +02:00
|
|
|
|
"movq %%mm0, (%%eax) \n\t"
|
|
|
|
|
"movq (%0, %1, 4), %%mm0 \n\t"
|
2001-10-17 22:42:07 +02:00
|
|
|
|
PAVGB(%%mm0, %%mm1)
|
2001-10-15 05:01:08 +02:00
|
|
|
|
"movq %%mm1, (%%eax, %1, 2) \n\t"
|
|
|
|
|
"movq (%%ebx, %1), %%mm1 \n\t"
|
2001-10-17 22:42:07 +02:00
|
|
|
|
PAVGB(%%mm1, %%mm0)
|
2001-10-15 05:01:08 +02:00
|
|
|
|
"movq %%mm0, (%%ebx) \n\t"
|
|
|
|
|
"movq (%0, %1, 8), %%mm0 \n\t"
|
2001-10-17 22:42:07 +02:00
|
|
|
|
PAVGB(%%mm0, %%mm1)
|
2001-10-15 05:01:08 +02:00
|
|
|
|
"movq %%mm1, (%%ebx, %1, 2) \n\t"
|
|
|
|
|
|
|
|
|
|
: : "r" (src), "r" (stride)
|
|
|
|
|
: "%eax", "%ebx"
|
|
|
|
|
);
|
|
|
|
|
#else
|
|
|
|
|
int x;
|
|
|
|
|
for(x=0; x<8; x++)
|
|
|
|
|
{
|
|
|
|
|
src[stride] = (src[0] + src[stride*2])>>1;
|
|
|
|
|
src[stride*3] = (src[stride*2] + src[stride*4])>>1;
|
|
|
|
|
src[stride*5] = (src[stride*4] + src[stride*6])>>1;
|
|
|
|
|
src[stride*7] = (src[stride*6] + src[stride*8])>>1;
|
|
|
|
|
src++;
|
|
|
|
|
}
|
|
|
|
|
#endif
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Deinterlaces the given block
|
2001-10-17 22:42:07 +02:00
|
|
|
|
* will be called for every 8x8 block, and can read & write into an 8x16 block
|
|
|
|
|
* no cliping in C version
|
2001-10-15 05:01:08 +02:00
|
|
|
|
*/
|
2001-10-17 22:42:07 +02:00
|
|
|
|
static inline void deInterlaceInterpolateCubic(uint8_t src[], int stride)
|
2001-10-15 05:01:08 +02:00
|
|
|
|
{
|
|
|
|
|
#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
|
|
|
|
|
asm volatile(
|
|
|
|
|
"leal (%0, %1), %%eax \n\t"
|
|
|
|
|
"leal (%%eax, %1, 4), %%ebx \n\t"
|
2001-10-17 22:42:07 +02:00
|
|
|
|
"leal (%%ebx, %1, 4), %%ecx \n\t"
|
|
|
|
|
"addl %1, %%ecx \n\t"
|
|
|
|
|
"pxor %%mm7, %%mm7 \n\t"
|
|
|
|
|
// 0 1 2 3 4 5 6 7 8 9 10
|
|
|
|
|
// %0 eax eax+%1 eax+2%1 %0+4%1 ebx ebx+%1 ebx+2%1 %0+8%1 ebx+4%1 ecx
|
2001-10-15 05:01:08 +02:00
|
|
|
|
|
2001-10-17 22:42:07 +02:00
|
|
|
|
#define DEINT_CUBIC(a,b,c,d,e)\
|
|
|
|
|
"movq " #a ", %%mm0 \n\t"\
|
|
|
|
|
"movq " #b ", %%mm1 \n\t"\
|
|
|
|
|
"movq " #d ", %%mm2 \n\t"\
|
|
|
|
|
"movq " #e ", %%mm3 \n\t"\
|
|
|
|
|
PAVGB(%%mm2, %%mm1) /* (b+d) /2 */\
|
|
|
|
|
PAVGB(%%mm3, %%mm0) /* a(a+e) /2 */\
|
|
|
|
|
"movq %%mm0, %%mm2 \n\t"\
|
|
|
|
|
"punpcklbw %%mm7, %%mm0 \n\t"\
|
|
|
|
|
"punpckhbw %%mm7, %%mm2 \n\t"\
|
|
|
|
|
"movq %%mm1, %%mm3 \n\t"\
|
|
|
|
|
"punpcklbw %%mm7, %%mm1 \n\t"\
|
|
|
|
|
"punpckhbw %%mm7, %%mm3 \n\t"\
|
|
|
|
|
"psubw %%mm1, %%mm0 \n\t" /* L(a+e - (b+d))/2 */\
|
|
|
|
|
"psubw %%mm3, %%mm2 \n\t" /* H(a+e - (b+d))/2 */\
|
|
|
|
|
"psraw $3, %%mm0 \n\t" /* L(a+e - (b+d))/16 */\
|
|
|
|
|
"psraw $3, %%mm2 \n\t" /* H(a+e - (b+d))/16 */\
|
|
|
|
|
"psubw %%mm0, %%mm1 \n\t" /* L(9b + 9d - a - e)/16 */\
|
|
|
|
|
"psubw %%mm2, %%mm3 \n\t" /* H(9b + 9d - a - e)/16 */\
|
|
|
|
|
"packuswb %%mm3, %%mm1 \n\t"\
|
|
|
|
|
"movq %%mm1, " #c " \n\t"
|
|
|
|
|
|
|
|
|
|
DEINT_CUBIC((%0), (%%eax, %1), (%%eax, %1, 2), (%0, %1, 4), (%%ebx, %1))
|
|
|
|
|
DEINT_CUBIC((%%eax, %1), (%0, %1, 4), (%%ebx), (%%ebx, %1), (%0, %1, 8))
|
|
|
|
|
DEINT_CUBIC((%0, %1, 4), (%%ebx, %1), (%%ebx, %1, 2), (%0, %1, 8), (%%ecx))
|
|
|
|
|
DEINT_CUBIC((%%ebx, %1), (%0, %1, 8), (%%ebx, %1, 4), (%%ecx), (%%ecx, %1, 2))
|
2001-10-15 05:01:08 +02:00
|
|
|
|
|
|
|
|
|
: : "r" (src), "r" (stride)
|
2001-10-17 22:42:07 +02:00
|
|
|
|
: "%eax", "%ebx", "ecx"
|
2001-10-15 05:01:08 +02:00
|
|
|
|
);
|
|
|
|
|
#else
|
|
|
|
|
int x;
|
|
|
|
|
for(x=0; x<8; x++)
|
|
|
|
|
{
|
2001-10-17 22:42:07 +02:00
|
|
|
|
src[stride*3] = (-src[0] + 9*src[stride*2] + 9*src[stride*4] - src[stride*6])>>4;
|
|
|
|
|
src[stride*5] = (-src[stride*2] + 9*src[stride*4] + 9*src[stride*6] - src[stride*8])>>4;
|
|
|
|
|
src[stride*7] = (-src[stride*4] + 9*src[stride*6] + 9*src[stride*8] - src[stride*10])>>4;
|
|
|
|
|
src[stride*9] = (-src[stride*6] + 9*src[stride*8] + 9*src[stride*10] - src[stride*12])>>4;
|
2001-10-15 05:01:08 +02:00
|
|
|
|
src++;
|
|
|
|
|
}
|
|
|
|
|
#endif
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Deinterlaces the given block
|
2001-10-17 22:42:07 +02:00
|
|
|
|
* will be called for every 8x8 block, and can read & write into an 8x16 block
|
2001-10-15 05:01:08 +02:00
|
|
|
|
* will shift the image up by 1 line (FIXME if this is a problem)
|
|
|
|
|
*/
|
|
|
|
|
static inline void deInterlaceBlendLinear(uint8_t src[], int stride)
|
|
|
|
|
{
|
|
|
|
|
#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
|
|
|
|
|
asm volatile(
|
|
|
|
|
"leal (%0, %1), %%eax \n\t"
|
|
|
|
|
"leal (%%eax, %1, 4), %%ebx \n\t"
|
|
|
|
|
// 0 1 2 3 4 5 6 7 8 9
|
|
|
|
|
// %0 eax eax+%1 eax+2%1 %0+4%1 ebx ebx+%1 ebx+2%1 %0+8%1 ebx+4%1
|
|
|
|
|
|
|
|
|
|
"movq (%0), %%mm0 \n\t" // L0
|
|
|
|
|
"movq (%%eax, %1), %%mm1 \n\t" // L2
|
|
|
|
|
PAVGB(%%mm1, %%mm0) // L0+L2
|
|
|
|
|
"movq (%%eax), %%mm2 \n\t" // L1
|
|
|
|
|
PAVGB(%%mm2, %%mm0)
|
|
|
|
|
"movq %%mm0, (%0) \n\t"
|
|
|
|
|
"movq (%%eax, %1, 2), %%mm0 \n\t" // L3
|
|
|
|
|
PAVGB(%%mm0, %%mm2) // L1+L3
|
|
|
|
|
PAVGB(%%mm1, %%mm2) // 2L2 + L1 + L3
|
|
|
|
|
"movq %%mm2, (%%eax) \n\t"
|
|
|
|
|
"movq (%0, %1, 4), %%mm2 \n\t" // L4
|
|
|
|
|
PAVGB(%%mm2, %%mm1) // L2+L4
|
|
|
|
|
PAVGB(%%mm0, %%mm1) // 2L3 + L2 + L4
|
|
|
|
|
"movq %%mm1, (%%eax, %1) \n\t"
|
|
|
|
|
"movq (%%ebx), %%mm1 \n\t" // L5
|
|
|
|
|
PAVGB(%%mm1, %%mm0) // L3+L5
|
|
|
|
|
PAVGB(%%mm2, %%mm0) // 2L4 + L3 + L5
|
|
|
|
|
"movq %%mm0, (%%eax, %1, 2) \n\t"
|
|
|
|
|
"movq (%%ebx, %1), %%mm0 \n\t" // L6
|
|
|
|
|
PAVGB(%%mm0, %%mm2) // L4+L6
|
|
|
|
|
PAVGB(%%mm1, %%mm2) // 2L5 + L4 + L6
|
|
|
|
|
"movq %%mm2, (%0, %1, 4) \n\t"
|
|
|
|
|
"movq (%%ebx, %1, 2), %%mm2 \n\t" // L7
|
|
|
|
|
PAVGB(%%mm2, %%mm1) // L5+L7
|
|
|
|
|
PAVGB(%%mm0, %%mm1) // 2L6 + L5 + L7
|
|
|
|
|
"movq %%mm1, (%%ebx) \n\t"
|
|
|
|
|
"movq (%0, %1, 8), %%mm1 \n\t" // L8
|
|
|
|
|
PAVGB(%%mm1, %%mm0) // L6+L8
|
|
|
|
|
PAVGB(%%mm2, %%mm0) // 2L7 + L6 + L8
|
|
|
|
|
"movq %%mm0, (%%ebx, %1) \n\t"
|
|
|
|
|
"movq (%%ebx, %1, 4), %%mm0 \n\t" // L9
|
|
|
|
|
PAVGB(%%mm0, %%mm2) // L7+L9
|
|
|
|
|
PAVGB(%%mm1, %%mm2) // 2L8 + L7 + L9
|
|
|
|
|
"movq %%mm2, (%%ebx, %1, 2) \n\t"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
: : "r" (src), "r" (stride)
|
|
|
|
|
: "%eax", "%ebx"
|
|
|
|
|
);
|
|
|
|
|
#else
|
|
|
|
|
int x;
|
|
|
|
|
for(x=0; x<8; x++)
|
|
|
|
|
{
|
|
|
|
|
src[0 ] = (src[0 ] + 2*src[stride ] + src[stride*2])>>2;
|
|
|
|
|
src[stride ] = (src[stride ] + 2*src[stride*2] + src[stride*3])>>2;
|
|
|
|
|
src[stride*2] = (src[stride*2] + 2*src[stride*3] + src[stride*4])>>2;
|
|
|
|
|
src[stride*3] = (src[stride*3] + 2*src[stride*4] + src[stride*5])>>2;
|
|
|
|
|
src[stride*4] = (src[stride*4] + 2*src[stride*5] + src[stride*6])>>2;
|
|
|
|
|
src[stride*5] = (src[stride*5] + 2*src[stride*6] + src[stride*7])>>2;
|
|
|
|
|
src[stride*6] = (src[stride*6] + 2*src[stride*7] + src[stride*8])>>2;
|
|
|
|
|
src[stride*7] = (src[stride*7] + 2*src[stride*8] + src[stride*9])>>2;
|
|
|
|
|
src++;
|
|
|
|
|
}
|
|
|
|
|
#endif
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Deinterlaces the given block
|
|
|
|
|
* will be called for every 8x8 block, except the last row, and can read & write into an 8x16 block
|
|
|
|
|
*/
|
|
|
|
|
static inline void deInterlaceMedian(uint8_t src[], int stride)
|
|
|
|
|
{
|
2001-10-16 04:31:14 +02:00
|
|
|
|
#ifdef HAVE_MMX
|
|
|
|
|
#ifdef HAVE_MMX2
|
2001-10-15 05:01:08 +02:00
|
|
|
|
asm volatile(
|
|
|
|
|
"leal (%0, %1), %%eax \n\t"
|
|
|
|
|
"leal (%%eax, %1, 4), %%ebx \n\t"
|
|
|
|
|
// 0 1 2 3 4 5 6 7 8 9
|
|
|
|
|
// %0 eax eax+%1 eax+2%1 %0+4%1 ebx ebx+%1 ebx+2%1 %0+8%1 ebx+4%1
|
|
|
|
|
|
|
|
|
|
"movq (%0), %%mm0 \n\t" //
|
|
|
|
|
"movq (%%eax, %1), %%mm2 \n\t" //
|
|
|
|
|
"movq (%%eax), %%mm1 \n\t" //
|
|
|
|
|
"movq %%mm0, %%mm3 \n\t"
|
|
|
|
|
"pmaxub %%mm1, %%mm0 \n\t" //
|
|
|
|
|
"pminub %%mm3, %%mm1 \n\t" //
|
|
|
|
|
"pmaxub %%mm2, %%mm1 \n\t" //
|
|
|
|
|
"pminub %%mm1, %%mm0 \n\t"
|
|
|
|
|
"movq %%mm0, (%%eax) \n\t"
|
|
|
|
|
|
|
|
|
|
"movq (%0, %1, 4), %%mm0 \n\t" //
|
|
|
|
|
"movq (%%eax, %1, 2), %%mm1 \n\t" //
|
|
|
|
|
"movq %%mm2, %%mm3 \n\t"
|
|
|
|
|
"pmaxub %%mm1, %%mm2 \n\t" //
|
|
|
|
|
"pminub %%mm3, %%mm1 \n\t" //
|
|
|
|
|
"pmaxub %%mm0, %%mm1 \n\t" //
|
|
|
|
|
"pminub %%mm1, %%mm2 \n\t"
|
|
|
|
|
"movq %%mm2, (%%eax, %1, 2) \n\t"
|
|
|
|
|
|
|
|
|
|
"movq (%%ebx), %%mm2 \n\t" //
|
|
|
|
|
"movq (%%ebx, %1), %%mm1 \n\t" //
|
|
|
|
|
"movq %%mm2, %%mm3 \n\t"
|
|
|
|
|
"pmaxub %%mm0, %%mm2 \n\t" //
|
|
|
|
|
"pminub %%mm3, %%mm0 \n\t" //
|
|
|
|
|
"pmaxub %%mm1, %%mm0 \n\t" //
|
|
|
|
|
"pminub %%mm0, %%mm2 \n\t"
|
|
|
|
|
"movq %%mm2, (%%ebx) \n\t"
|
|
|
|
|
|
|
|
|
|
"movq (%%ebx, %1, 2), %%mm2 \n\t" //
|
|
|
|
|
"movq (%0, %1, 8), %%mm0 \n\t" //
|
|
|
|
|
"movq %%mm2, %%mm3 \n\t"
|
|
|
|
|
"pmaxub %%mm0, %%mm2 \n\t" //
|
|
|
|
|
"pminub %%mm3, %%mm0 \n\t" //
|
|
|
|
|
"pmaxub %%mm1, %%mm0 \n\t" //
|
|
|
|
|
"pminub %%mm0, %%mm2 \n\t"
|
|
|
|
|
"movq %%mm2, (%%ebx, %1, 2) \n\t"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
: : "r" (src), "r" (stride)
|
|
|
|
|
: "%eax", "%ebx"
|
|
|
|
|
);
|
2001-10-16 04:31:14 +02:00
|
|
|
|
|
|
|
|
|
#else // MMX without MMX2
|
|
|
|
|
asm volatile(
|
|
|
|
|
"leal (%0, %1), %%eax \n\t"
|
|
|
|
|
"leal (%%eax, %1, 4), %%ebx \n\t"
|
|
|
|
|
// 0 1 2 3 4 5 6 7 8 9
|
|
|
|
|
// %0 eax eax+%1 eax+2%1 %0+4%1 ebx ebx+%1 ebx+2%1 %0+8%1 ebx+4%1
|
|
|
|
|
"pxor %%mm7, %%mm7 \n\t"
|
|
|
|
|
|
|
|
|
|
#define MEDIAN(a,b,c)\
|
|
|
|
|
"movq " #a ", %%mm0 \n\t"\
|
|
|
|
|
"movq " #b ", %%mm2 \n\t"\
|
|
|
|
|
"movq " #c ", %%mm1 \n\t"\
|
|
|
|
|
"movq %%mm0, %%mm3 \n\t"\
|
|
|
|
|
"movq %%mm1, %%mm4 \n\t"\
|
|
|
|
|
"movq %%mm2, %%mm5 \n\t"\
|
|
|
|
|
"psubusb %%mm1, %%mm3 \n\t"\
|
|
|
|
|
"psubusb %%mm2, %%mm4 \n\t"\
|
|
|
|
|
"psubusb %%mm0, %%mm5 \n\t"\
|
|
|
|
|
"pcmpeqb %%mm7, %%mm3 \n\t"\
|
|
|
|
|
"pcmpeqb %%mm7, %%mm4 \n\t"\
|
|
|
|
|
"pcmpeqb %%mm7, %%mm5 \n\t"\
|
|
|
|
|
"movq %%mm3, %%mm6 \n\t"\
|
|
|
|
|
"pxor %%mm4, %%mm3 \n\t"\
|
|
|
|
|
"pxor %%mm5, %%mm4 \n\t"\
|
|
|
|
|
"pxor %%mm6, %%mm5 \n\t"\
|
|
|
|
|
"por %%mm3, %%mm1 \n\t"\
|
|
|
|
|
"por %%mm4, %%mm2 \n\t"\
|
|
|
|
|
"por %%mm5, %%mm0 \n\t"\
|
|
|
|
|
"pand %%mm2, %%mm0 \n\t"\
|
|
|
|
|
"pand %%mm1, %%mm0 \n\t"\
|
|
|
|
|
"movq %%mm0, " #b " \n\t"
|
|
|
|
|
|
|
|
|
|
MEDIAN((%0), (%%eax), (%%eax, %1))
|
|
|
|
|
MEDIAN((%%eax, %1), (%%eax, %1, 2), (%0, %1, 4))
|
|
|
|
|
MEDIAN((%0, %1, 4), (%%ebx), (%%ebx, %1))
|
|
|
|
|
MEDIAN((%%ebx, %1), (%%ebx, %1, 2), (%0, %1, 8))
|
|
|
|
|
|
|
|
|
|
: : "r" (src), "r" (stride)
|
|
|
|
|
: "%eax", "%ebx"
|
|
|
|
|
);
|
|
|
|
|
#endif // MMX
|
2001-10-15 05:01:08 +02:00
|
|
|
|
#else
|
|
|
|
|
//FIXME
|
|
|
|
|
int x;
|
|
|
|
|
for(x=0; x<8; x++)
|
|
|
|
|
{
|
|
|
|
|
src[0 ] = (src[0 ] + 2*src[stride ] + src[stride*2])>>2;
|
|
|
|
|
src[stride ] = (src[stride ] + 2*src[stride*2] + src[stride*3])>>2;
|
|
|
|
|
src[stride*2] = (src[stride*2] + 2*src[stride*3] + src[stride*4])>>2;
|
|
|
|
|
src[stride*3] = (src[stride*3] + 2*src[stride*4] + src[stride*5])>>2;
|
|
|
|
|
src[stride*4] = (src[stride*4] + 2*src[stride*5] + src[stride*6])>>2;
|
|
|
|
|
src[stride*5] = (src[stride*5] + 2*src[stride*6] + src[stride*7])>>2;
|
|
|
|
|
src[stride*6] = (src[stride*6] + 2*src[stride*7] + src[stride*8])>>2;
|
|
|
|
|
src[stride*7] = (src[stride*7] + 2*src[stride*8] + src[stride*9])>>2;
|
|
|
|
|
src++;
|
|
|
|
|
}
|
|
|
|
|
#endif
|
|
|
|
|
}
|
|
|
|
|
|
2001-10-13 17:36:04 +02:00
|
|
|
|
#ifdef HAVE_ODIVX_POSTPROCESS
|
|
|
|
|
#include "../opendivx/postprocess.h"
|
|
|
|
|
int use_old_pp=0;
|
|
|
|
|
#endif
|
2001-10-11 00:21:19 +02:00
|
|
|
|
|
2001-10-13 17:36:04 +02:00
|
|
|
|
static void postProcess(uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
|
|
|
|
|
QP_STORE_T QPs[], int QPStride, int isColor, int mode);
|
2001-10-11 00:21:19 +02:00
|
|
|
|
|
2001-10-23 01:36:35 +02:00
|
|
|
|
/* -pp Command line Help
|
|
|
|
|
NOTE/FIXME: put this at an appropriate place (--help, html docs, man mplayer)?
|
|
|
|
|
|
|
|
|
|
-pp <filterName>[:<option>[:<option>...]][,[-]<filterName>[:<option>...]]...
|
|
|
|
|
|
|
|
|
|
long form example:
|
|
|
|
|
-pp vdeblock:autoq,hdeblock:autoq,linblenddeint -pp default,-vdeblock
|
|
|
|
|
short form example:
|
|
|
|
|
-pp vb:a,hb:a,lb -pp de,-vb
|
|
|
|
|
|
|
|
|
|
Filters Options
|
|
|
|
|
short long name short long option Description
|
|
|
|
|
* * a autoq cpu power dependant enabler
|
|
|
|
|
c chrom chrominance filtring enabled
|
|
|
|
|
y nochrom chrominance filtring disabled
|
|
|
|
|
hb hdeblock horizontal deblocking filter
|
|
|
|
|
vb vdeblock vertical deblocking filter
|
|
|
|
|
vr rkvdeblock
|
|
|
|
|
h1 x1hdeblock Experimental horizontal deblock filter 1
|
|
|
|
|
v1 x1vdeblock Experimental vertical deblock filter 1
|
|
|
|
|
dr dering not implemented yet
|
|
|
|
|
al autolevels automatic brightness / contrast fixer
|
|
|
|
|
f fullyrange stretch luminance range to (0..255)
|
|
|
|
|
lb linblenddeint linear blend deinterlacer
|
|
|
|
|
li linipoldeint linear interpolating deinterlacer
|
|
|
|
|
ci cubicipoldeint cubic interpolating deinterlacer
|
|
|
|
|
md mediandeint median deinterlacer
|
|
|
|
|
de default hdeblock:a,vdeblock:a,dering:a,autolevels
|
|
|
|
|
fa fast x1hdeblock:a,x1vdeblock:a,dering:a,autolevels
|
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* returns a PPMode struct which will have a non 0 error variable if an error occured
|
|
|
|
|
* name is the string after "-pp" on the command line
|
|
|
|
|
* quality is a number from 0 to GET_PP_QUALITY_MAX
|
|
|
|
|
*/
|
|
|
|
|
struct PPMode getPPModeByNameAndQuality(char *name, int quality)
|
|
|
|
|
{
|
|
|
|
|
char temp[GET_MODE_BUFFER_SIZE];
|
|
|
|
|
char *p= temp;
|
|
|
|
|
char *filterDelimiters= ",";
|
|
|
|
|
char *optionDelimiters= ":";
|
|
|
|
|
struct PPMode ppMode= {0,0,0,0,0,0};
|
|
|
|
|
char *filterToken;
|
|
|
|
|
|
|
|
|
|
strncpy(temp, name, GET_MODE_BUFFER_SIZE);
|
|
|
|
|
|
|
|
|
|
for(;;){
|
|
|
|
|
char *p2;
|
|
|
|
|
char *filterName;
|
|
|
|
|
int q= GET_PP_QUALITY_MAX;
|
|
|
|
|
int chrom=-1;
|
|
|
|
|
char *option;
|
|
|
|
|
char *options[OPTIONS_ARRAY_SIZE];
|
|
|
|
|
int i;
|
|
|
|
|
int filterNameOk=0;
|
|
|
|
|
int numOfUnknownOptions=0;
|
|
|
|
|
int enable=1; //does the user want us to enabled or disabled the filter
|
|
|
|
|
|
|
|
|
|
filterToken= strtok(p, filterDelimiters);
|
|
|
|
|
if(filterToken == NULL) break;
|
|
|
|
|
p+= strlen(filterToken) + 1;
|
|
|
|
|
filterName= strtok(filterToken, optionDelimiters);
|
|
|
|
|
printf("%s::%s\n", filterToken, filterName);
|
|
|
|
|
|
|
|
|
|
if(*filterName == '-')
|
|
|
|
|
{
|
|
|
|
|
enable=0;
|
|
|
|
|
filterName++;
|
|
|
|
|
}
|
|
|
|
|
for(;;){ //for all options
|
|
|
|
|
option= strtok(NULL, optionDelimiters);
|
|
|
|
|
if(option == NULL) break;
|
|
|
|
|
|
|
|
|
|
printf("%s\n", option);
|
|
|
|
|
if(!strcmp("autoq", option) || !strcmp("a", option)) q= quality;
|
|
|
|
|
else if(!strcmp("nochrom", option) || !strcmp("y", option)) chrom=0;
|
|
|
|
|
else if(!strcmp("chrom", option) || !strcmp("c", option)) chrom=1;
|
|
|
|
|
else
|
|
|
|
|
{
|
|
|
|
|
options[numOfUnknownOptions] = option;
|
|
|
|
|
numOfUnknownOptions++;
|
|
|
|
|
options[numOfUnknownOptions] = NULL;
|
|
|
|
|
}
|
|
|
|
|
if(numOfUnknownOptions >= OPTIONS_ARRAY_SIZE-1) break;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* replace stuff from the replace Table */
|
|
|
|
|
for(i=0; replaceTable[2*i]!=NULL; i++)
|
|
|
|
|
{
|
|
|
|
|
if(!strcmp(replaceTable[2*i], filterName))
|
|
|
|
|
{
|
|
|
|
|
int newlen= strlen(replaceTable[2*i + 1]);
|
|
|
|
|
int plen;
|
|
|
|
|
int spaceLeft;
|
|
|
|
|
|
|
|
|
|
if(p==NULL) p= temp, *p=0; //last filter
|
|
|
|
|
else p--, *p=','; //not last filter
|
|
|
|
|
|
|
|
|
|
plen= strlen(p);
|
|
|
|
|
spaceLeft= (int)p - (int)temp + plen;
|
|
|
|
|
if(spaceLeft + newlen >= GET_MODE_BUFFER_SIZE)
|
|
|
|
|
{
|
|
|
|
|
ppMode.error++;
|
|
|
|
|
break;
|
|
|
|
|
}
|
|
|
|
|
memmove(p + newlen, p, plen+1);
|
|
|
|
|
memcpy(p, replaceTable[2*i + 1], newlen);
|
|
|
|
|
filterNameOk=1;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
for(i=0; filters[i].shortName!=NULL; i++)
|
|
|
|
|
{
|
|
|
|
|
if( !strcmp(filters[i].longName, filterName)
|
|
|
|
|
|| !strcmp(filters[i].shortName, filterName))
|
|
|
|
|
{
|
|
|
|
|
ppMode.lumMode &= ~filters[i].mask;
|
|
|
|
|
ppMode.chromMode &= ~filters[i].mask;
|
|
|
|
|
|
|
|
|
|
filterNameOk=1;
|
|
|
|
|
if(!enable) break; // user wants to disable it
|
|
|
|
|
|
|
|
|
|
if(q >= filters[i].minLumQuality)
|
|
|
|
|
ppMode.lumMode|= filters[i].mask;
|
|
|
|
|
if(chrom==1 || (chrom==-1 && filters[i].chromDefault))
|
|
|
|
|
if(q >= filters[i].minChromQuality)
|
|
|
|
|
ppMode.chromMode|= filters[i].mask;
|
|
|
|
|
|
|
|
|
|
if(filters[i].mask == LEVEL_FIX)
|
|
|
|
|
{
|
|
|
|
|
int o;
|
|
|
|
|
ppMode.minAllowedY= 16;
|
|
|
|
|
ppMode.maxAllowedY= 234;
|
|
|
|
|
for(o=0; options[o]!=NULL; o++)
|
|
|
|
|
if( !strcmp(options[o],"fullyrange")
|
|
|
|
|
||!strcmp(options[o],"f"))
|
|
|
|
|
{
|
|
|
|
|
ppMode.minAllowedY= 0;
|
|
|
|
|
ppMode.maxAllowedY= 255;
|
|
|
|
|
numOfUnknownOptions--;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
if(!filterNameOk) ppMode.error++;
|
|
|
|
|
ppMode.error += numOfUnknownOptions;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if(ppMode.lumMode & H_DEBLOCK) ppMode.oldMode |= PP_DEBLOCK_Y_H;
|
|
|
|
|
if(ppMode.lumMode & V_DEBLOCK) ppMode.oldMode |= PP_DEBLOCK_Y_V;
|
|
|
|
|
if(ppMode.chromMode & H_DEBLOCK) ppMode.oldMode |= PP_DEBLOCK_C_H;
|
|
|
|
|
if(ppMode.chromMode & V_DEBLOCK) ppMode.oldMode |= PP_DEBLOCK_C_V;
|
|
|
|
|
if(ppMode.lumMode & DERING) ppMode.oldMode |= PP_DERING_Y;
|
|
|
|
|
if(ppMode.chromMode & DERING) ppMode.oldMode |= PP_DERING_C;
|
|
|
|
|
|
|
|
|
|
return ppMode;
|
|
|
|
|
}
|
|
|
|
|
|
2001-10-11 00:13:27 +02:00
|
|
|
|
/**
|
|
|
|
|
* ...
|
|
|
|
|
*/
|
|
|
|
|
void postprocess(unsigned char * src[], int src_stride,
|
|
|
|
|
unsigned char * dst[], int dst_stride,
|
|
|
|
|
int horizontal_size, int vertical_size,
|
|
|
|
|
QP_STORE_T *QP_store, int QP_stride,
|
|
|
|
|
int mode)
|
|
|
|
|
{
|
2001-10-23 01:36:35 +02:00
|
|
|
|
/*
|
|
|
|
|
static int qual=0;
|
|
|
|
|
|
|
|
|
|
struct PPMode ppMode= getPPModeByNameAndQuality("fast,default,-hdeblock,-vdeblock", qual);
|
|
|
|
|
qual++;
|
|
|
|
|
qual%=7;
|
|
|
|
|
printf("\n%d %d %d %d\n", ppMode.lumMode, ppMode.chromMode, ppMode.oldMode, ppMode.error);
|
|
|
|
|
postprocess2(src, src_stride, dst, dst_stride,
|
|
|
|
|
horizontal_size, vertical_size, QP_store, QP_stride, &ppMode);
|
|
|
|
|
|
|
|
|
|
return;
|
|
|
|
|
*/
|
2001-10-11 00:21:19 +02:00
|
|
|
|
|
2001-10-13 17:36:04 +02:00
|
|
|
|
#ifdef HAVE_ODIVX_POSTPROCESS
|
|
|
|
|
// Note: I could make this shit outside of this file, but it would mean one
|
|
|
|
|
// more function call...
|
|
|
|
|
if(use_old_pp){
|
|
|
|
|
odivx_postprocess(src,src_stride,dst,dst_stride,horizontal_size,vertical_size,QP_store,QP_stride,mode);
|
|
|
|
|
return;
|
|
|
|
|
}
|
|
|
|
|
#endif
|
|
|
|
|
|
2001-10-11 00:21:19 +02:00
|
|
|
|
postProcess(src[0], src_stride, dst[0], dst_stride,
|
2001-10-12 00:35:45 +02:00
|
|
|
|
horizontal_size, vertical_size, QP_store, QP_stride, 0, mode);
|
2001-10-11 00:13:27 +02:00
|
|
|
|
|
|
|
|
|
horizontal_size >>= 1;
|
|
|
|
|
vertical_size >>= 1;
|
|
|
|
|
src_stride >>= 1;
|
|
|
|
|
dst_stride >>= 1;
|
2001-10-13 04:31:15 +02:00
|
|
|
|
mode= ((mode&0xFF)>>4) | (mode&0xFFFFFF00);
|
2001-10-11 00:13:27 +02:00
|
|
|
|
|
|
|
|
|
if(1)
|
|
|
|
|
{
|
2001-10-11 00:21:19 +02:00
|
|
|
|
postProcess(src[1], src_stride, dst[1], dst_stride,
|
2001-10-13 04:31:15 +02:00
|
|
|
|
horizontal_size, vertical_size, QP_store, QP_stride, 1, mode);
|
2001-10-11 00:21:19 +02:00
|
|
|
|
postProcess(src[2], src_stride, dst[2], dst_stride,
|
2001-10-23 01:36:35 +02:00
|
|
|
|
horizontal_size, vertical_size, QP_store, QP_stride, 2, mode);
|
2001-10-11 00:13:27 +02:00
|
|
|
|
}
|
|
|
|
|
else
|
|
|
|
|
{
|
|
|
|
|
memcpy(dst[1], src[1], src_stride*horizontal_size);
|
|
|
|
|
memcpy(dst[2], src[2], src_stride*horizontal_size);
|
|
|
|
|
}
|
|
|
|
|
}
|
2001-10-13 17:36:04 +02:00
|
|
|
|
|
2001-10-23 01:36:35 +02:00
|
|
|
|
void postprocess2(unsigned char * src[], int src_stride,
|
|
|
|
|
unsigned char * dst[], int dst_stride,
|
|
|
|
|
int horizontal_size, int vertical_size,
|
|
|
|
|
QP_STORE_T *QP_store, int QP_stride,
|
|
|
|
|
struct PPMode *mode)
|
|
|
|
|
{
|
|
|
|
|
|
|
|
|
|
#ifdef HAVE_ODIVX_POSTPROCESS
|
|
|
|
|
// Note: I could make this shit outside of this file, but it would mean one
|
|
|
|
|
// more function call...
|
|
|
|
|
if(use_old_pp){
|
|
|
|
|
odivx_postprocess(src,src_stride,dst,dst_stride,horizontal_size,vertical_size,QP_store,QP_stride,
|
|
|
|
|
mode->oldMode);
|
|
|
|
|
return;
|
|
|
|
|
}
|
|
|
|
|
#endif
|
|
|
|
|
|
|
|
|
|
postProcess(src[0], src_stride, dst[0], dst_stride,
|
|
|
|
|
horizontal_size, vertical_size, QP_store, QP_stride, 0, mode->lumMode);
|
|
|
|
|
|
|
|
|
|
horizontal_size >>= 1;
|
|
|
|
|
vertical_size >>= 1;
|
|
|
|
|
src_stride >>= 1;
|
|
|
|
|
dst_stride >>= 1;
|
|
|
|
|
|
|
|
|
|
postProcess(src[1], src_stride, dst[1], dst_stride,
|
|
|
|
|
horizontal_size, vertical_size, QP_store, QP_stride, 1, mode->chromMode);
|
|
|
|
|
postProcess(src[2], src_stride, dst[2], dst_stride,
|
|
|
|
|
horizontal_size, vertical_size, QP_store, QP_stride, 2, mode->chromMode);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
2001-10-11 00:21:19 +02:00
|
|
|
|
/**
|
|
|
|
|
* gets the mode flags for a given quality (larger values mean slower but better postprocessing)
|
2001-10-13 17:36:04 +02:00
|
|
|
|
* 0 <= quality <= 6
|
2001-10-11 00:21:19 +02:00
|
|
|
|
*/
|
2001-10-13 17:36:04 +02:00
|
|
|
|
int getPpModeForQuality(int quality){
|
|
|
|
|
int modes[1+GET_PP_QUALITY_MAX]= {
|
|
|
|
|
0,
|
|
|
|
|
#if 1
|
|
|
|
|
// horizontal filters first
|
|
|
|
|
LUM_H_DEBLOCK,
|
|
|
|
|
LUM_H_DEBLOCK | LUM_V_DEBLOCK,
|
|
|
|
|
LUM_H_DEBLOCK | LUM_V_DEBLOCK | CHROM_H_DEBLOCK,
|
|
|
|
|
LUM_H_DEBLOCK | LUM_V_DEBLOCK | CHROM_H_DEBLOCK | CHROM_V_DEBLOCK,
|
|
|
|
|
LUM_H_DEBLOCK | LUM_V_DEBLOCK | CHROM_H_DEBLOCK | CHROM_V_DEBLOCK | LUM_DERING,
|
|
|
|
|
LUM_H_DEBLOCK | LUM_V_DEBLOCK | CHROM_H_DEBLOCK | CHROM_V_DEBLOCK | LUM_DERING | CHROM_DERING
|
|
|
|
|
#else
|
|
|
|
|
// vertical filters first
|
2001-10-11 00:21:19 +02:00
|
|
|
|
LUM_V_DEBLOCK,
|
|
|
|
|
LUM_V_DEBLOCK | LUM_H_DEBLOCK,
|
|
|
|
|
LUM_V_DEBLOCK | LUM_H_DEBLOCK | CHROM_V_DEBLOCK,
|
|
|
|
|
LUM_V_DEBLOCK | LUM_H_DEBLOCK | CHROM_V_DEBLOCK | CHROM_H_DEBLOCK,
|
|
|
|
|
LUM_V_DEBLOCK | LUM_H_DEBLOCK | CHROM_V_DEBLOCK | CHROM_H_DEBLOCK | LUM_DERING,
|
|
|
|
|
LUM_V_DEBLOCK | LUM_H_DEBLOCK | CHROM_V_DEBLOCK | CHROM_H_DEBLOCK | LUM_DERING | CHROM_DERING
|
2001-10-13 17:36:04 +02:00
|
|
|
|
#endif
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
#ifdef HAVE_ODIVX_POSTPROCESS
|
|
|
|
|
int odivx_modes[1+GET_PP_QUALITY_MAX]= {
|
|
|
|
|
0,
|
|
|
|
|
PP_DEBLOCK_Y_H,
|
|
|
|
|
PP_DEBLOCK_Y_H|PP_DEBLOCK_Y_V,
|
|
|
|
|
PP_DEBLOCK_Y_H|PP_DEBLOCK_Y_V|PP_DEBLOCK_C_H,
|
|
|
|
|
PP_DEBLOCK_Y_H|PP_DEBLOCK_Y_V|PP_DEBLOCK_C_H|PP_DEBLOCK_C_V,
|
|
|
|
|
PP_DEBLOCK_Y_H|PP_DEBLOCK_Y_V|PP_DEBLOCK_C_H|PP_DEBLOCK_C_V|PP_DERING_Y,
|
|
|
|
|
PP_DEBLOCK_Y_H|PP_DEBLOCK_Y_V|PP_DEBLOCK_C_H|PP_DEBLOCK_C_V|PP_DERING_Y|PP_DERING_C
|
|
|
|
|
};
|
|
|
|
|
if(use_old_pp) return odivx_modes[quality];
|
|
|
|
|
#endif
|
|
|
|
|
return modes[quality];
|
2001-10-11 00:13:27 +02:00
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Copies a block from src to dst and fixes the blacklevel
|
2001-10-12 00:35:45 +02:00
|
|
|
|
* numLines must be a multiple of 4
|
|
|
|
|
* levelFix == 0 -> dont touch the brighness & contrast
|
2001-10-11 00:13:27 +02:00
|
|
|
|
*/
|
2001-10-12 00:35:45 +02:00
|
|
|
|
static inline void blockCopy(uint8_t dst[], int dstStride, uint8_t src[], int srcStride,
|
|
|
|
|
int numLines, int levelFix)
|
2001-10-11 00:13:27 +02:00
|
|
|
|
{
|
2001-10-12 00:35:45 +02:00
|
|
|
|
int i;
|
|
|
|
|
if(levelFix)
|
|
|
|
|
{
|
2001-10-11 00:13:27 +02:00
|
|
|
|
#ifdef HAVE_MMX
|
|
|
|
|
asm volatile(
|
|
|
|
|
"pushl %0 \n\t"
|
|
|
|
|
"pushl %1 \n\t"
|
|
|
|
|
"leal (%2,%2), %%eax \n\t"
|
|
|
|
|
"leal (%3,%3), %%ebx \n\t"
|
|
|
|
|
"movq packedYOffset, %%mm2 \n\t"
|
|
|
|
|
"movq packedYScale, %%mm3 \n\t"
|
2001-10-13 13:58:41 +02:00
|
|
|
|
"pxor %%mm4, %%mm4 \n\t"
|
2001-10-11 00:13:27 +02:00
|
|
|
|
|
|
|
|
|
#define SCALED_CPY \
|
|
|
|
|
"movq (%0), %%mm0 \n\t"\
|
2001-10-23 03:46:50 +02:00
|
|
|
|
"movq (%0), %%mm5 \n\t"\
|
2001-10-13 13:58:41 +02:00
|
|
|
|
"punpcklbw %%mm4, %%mm0 \n\t"\
|
|
|
|
|
"punpckhbw %%mm4, %%mm5 \n\t"\
|
2001-10-23 02:03:19 +02:00
|
|
|
|
"psubw %%mm2, %%mm0 \n\t"\
|
|
|
|
|
"psubw %%mm2, %%mm5 \n\t"\
|
2001-10-23 03:46:50 +02:00
|
|
|
|
"movq (%0,%2), %%mm1 \n\t"\
|
2001-10-23 02:03:19 +02:00
|
|
|
|
"psllw $6, %%mm0 \n\t"\
|
|
|
|
|
"psllw $6, %%mm5 \n\t"\
|
2001-10-13 13:58:41 +02:00
|
|
|
|
"pmulhw %%mm3, %%mm0 \n\t"\
|
2001-10-23 03:46:50 +02:00
|
|
|
|
"movq (%0,%2), %%mm6 \n\t"\
|
2001-10-13 13:58:41 +02:00
|
|
|
|
"pmulhw %%mm3, %%mm5 \n\t"\
|
|
|
|
|
"punpcklbw %%mm4, %%mm1 \n\t"\
|
2001-10-23 03:46:50 +02:00
|
|
|
|
"punpckhbw %%mm4, %%mm6 \n\t"\
|
2001-10-23 02:03:19 +02:00
|
|
|
|
"psubw %%mm2, %%mm1 \n\t"\
|
2001-10-23 03:46:50 +02:00
|
|
|
|
"psubw %%mm2, %%mm6 \n\t"\
|
2001-10-23 02:03:19 +02:00
|
|
|
|
"psllw $6, %%mm1 \n\t"\
|
2001-10-23 03:46:50 +02:00
|
|
|
|
"psllw $6, %%mm6 \n\t"\
|
2001-10-13 13:58:41 +02:00
|
|
|
|
"pmulhw %%mm3, %%mm1 \n\t"\
|
2001-10-23 03:46:50 +02:00
|
|
|
|
"pmulhw %%mm3, %%mm6 \n\t"\
|
|
|
|
|
"addl %%eax, %0 \n\t"\
|
|
|
|
|
"packuswb %%mm5, %%mm0 \n\t"\
|
|
|
|
|
"packuswb %%mm6, %%mm1 \n\t"\
|
|
|
|
|
"movq %%mm0, (%1) \n\t"\
|
2001-10-13 13:58:41 +02:00
|
|
|
|
"movq %%mm1, (%1, %3) \n\t"\
|
2001-10-11 00:13:27 +02:00
|
|
|
|
|
2001-10-12 00:35:45 +02:00
|
|
|
|
SCALED_CPY
|
2001-10-11 00:13:27 +02:00
|
|
|
|
"addl %%ebx, %1 \n\t"
|
2001-10-12 00:35:45 +02:00
|
|
|
|
SCALED_CPY
|
2001-10-11 00:13:27 +02:00
|
|
|
|
"addl %%ebx, %1 \n\t"
|
2001-10-23 03:46:50 +02:00
|
|
|
|
SCALED_CPY
|
|
|
|
|
"addl %%ebx, %1 \n\t"
|
|
|
|
|
SCALED_CPY
|
2001-10-12 00:35:45 +02:00
|
|
|
|
|
|
|
|
|
"popl %1 \n\t"
|
|
|
|
|
"popl %0 \n\t"
|
|
|
|
|
: : "r" (src),
|
|
|
|
|
"r" (dst),
|
|
|
|
|
"r" (srcStride),
|
2001-10-23 03:46:50 +02:00
|
|
|
|
"r" (dstStride)
|
2001-10-12 00:35:45 +02:00
|
|
|
|
: "%eax", "%ebx"
|
|
|
|
|
);
|
|
|
|
|
#else
|
|
|
|
|
for(i=0; i<numLines; i++)
|
|
|
|
|
memcpy( &(dst[dstStride*i]),
|
|
|
|
|
&(src[srcStride*i]), BLOCK_SIZE);
|
|
|
|
|
#endif
|
|
|
|
|
}
|
|
|
|
|
else
|
|
|
|
|
{
|
|
|
|
|
#ifdef HAVE_MMX
|
|
|
|
|
asm volatile(
|
|
|
|
|
"movl %4, %%eax \n\t"
|
|
|
|
|
"movl %%eax, temp0\n\t"
|
|
|
|
|
"pushl %0 \n\t"
|
|
|
|
|
"pushl %1 \n\t"
|
|
|
|
|
"leal (%2,%2), %%eax \n\t"
|
|
|
|
|
"leal (%3,%3), %%ebx \n\t"
|
|
|
|
|
"movq packedYOffset, %%mm2 \n\t"
|
|
|
|
|
"movq packedYScale, %%mm3 \n\t"
|
|
|
|
|
|
|
|
|
|
#define SIMPLE_CPY \
|
|
|
|
|
"movq (%0), %%mm0 \n\t"\
|
|
|
|
|
"movq (%0,%2), %%mm1 \n\t"\
|
|
|
|
|
"movq %%mm0, (%1) \n\t"\
|
|
|
|
|
"movq %%mm1, (%1, %3) \n\t"\
|
|
|
|
|
|
|
|
|
|
"1: \n\t"
|
|
|
|
|
SIMPLE_CPY
|
|
|
|
|
"addl %%eax, %0 \n\t"
|
|
|
|
|
"addl %%ebx, %1 \n\t"
|
|
|
|
|
SIMPLE_CPY
|
2001-10-11 00:13:27 +02:00
|
|
|
|
"addl %%eax, %0 \n\t"
|
|
|
|
|
"addl %%ebx, %1 \n\t"
|
2001-10-12 00:35:45 +02:00
|
|
|
|
"decl temp0 \n\t"
|
|
|
|
|
"jnz 1b \n\t"
|
|
|
|
|
|
2001-10-11 00:13:27 +02:00
|
|
|
|
"popl %1 \n\t"
|
|
|
|
|
"popl %0 \n\t"
|
|
|
|
|
: : "r" (src),
|
|
|
|
|
"r" (dst),
|
|
|
|
|
"r" (srcStride),
|
2001-10-12 00:35:45 +02:00
|
|
|
|
"r" (dstStride),
|
|
|
|
|
"m" (numLines>>2)
|
2001-10-11 00:13:27 +02:00
|
|
|
|
: "%eax", "%ebx"
|
|
|
|
|
);
|
|
|
|
|
#else
|
2001-10-12 00:35:45 +02:00
|
|
|
|
for(i=0; i<numLines; i++)
|
2001-10-11 00:13:27 +02:00
|
|
|
|
memcpy( &(dst[dstStride*i]),
|
|
|
|
|
&(src[srcStride*i]), BLOCK_SIZE);
|
|
|
|
|
#endif
|
2001-10-12 00:35:45 +02:00
|
|
|
|
}
|
2001-10-11 00:13:27 +02:00
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* Filters array of bytes (Y or U or V values)
|
|
|
|
|
*/
|
2001-10-13 17:36:04 +02:00
|
|
|
|
static void postProcess(uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
|
2001-10-12 00:35:45 +02:00
|
|
|
|
QP_STORE_T QPs[], int QPStride, int isColor, int mode)
|
2001-10-11 00:13:27 +02:00
|
|
|
|
{
|
2001-10-12 00:35:45 +02:00
|
|
|
|
int x,y;
|
|
|
|
|
/* we need 64bit here otherwise we<77>ll going to have a problem
|
|
|
|
|
after watching a black picture for 5 hours*/
|
|
|
|
|
static uint64_t *yHistogram= NULL;
|
|
|
|
|
int black=0, white=255; // blackest black and whitest white in the picture
|
2001-10-11 00:13:27 +02:00
|
|
|
|
|
2001-10-17 22:42:07 +02:00
|
|
|
|
/* Temporary buffers for handling the last row(s) */
|
|
|
|
|
static uint8_t *tempDst= NULL;
|
|
|
|
|
static uint8_t *tempSrc= NULL;
|
|
|
|
|
|
2001-10-19 15:41:38 +02:00
|
|
|
|
/* Temporary buffers for handling the last block */
|
|
|
|
|
static uint8_t *tempDstBlock= NULL;
|
|
|
|
|
static uint8_t *tempSrcBlock= NULL;
|
|
|
|
|
|
|
|
|
|
uint8_t *dstBlockPtrBackup;
|
|
|
|
|
uint8_t *srcBlockPtrBackup;
|
|
|
|
|
|
2001-10-16 04:31:14 +02:00
|
|
|
|
#ifdef TIMING
|
2001-10-11 00:13:27 +02:00
|
|
|
|
long long T0, T1, memcpyTime=0, vertTime=0, horizTime=0, sumTime, diffTime=0;
|
|
|
|
|
sumTime= rdtsc();
|
|
|
|
|
#endif
|
|
|
|
|
|
2001-10-17 22:42:07 +02:00
|
|
|
|
if(tempDst==NULL)
|
|
|
|
|
{
|
|
|
|
|
tempDst= (uint8_t*)memalign(8, 1024*24);
|
|
|
|
|
tempSrc= (uint8_t*)memalign(8, 1024*24);
|
2001-10-19 15:41:38 +02:00
|
|
|
|
tempDstBlock= (uint8_t*)memalign(8, 1024*24);
|
|
|
|
|
tempSrcBlock= (uint8_t*)memalign(8, 1024*24);
|
2001-10-17 22:42:07 +02:00
|
|
|
|
}
|
|
|
|
|
|
2001-10-11 00:13:27 +02:00
|
|
|
|
if(!yHistogram)
|
|
|
|
|
{
|
2001-10-12 00:35:45 +02:00
|
|
|
|
int i;
|
|
|
|
|
yHistogram= (uint64_t*)malloc(8*256);
|
|
|
|
|
for(i=0; i<256; i++) yHistogram[i]= width*height/64*15/256;
|
2001-10-19 15:41:38 +02:00
|
|
|
|
|
|
|
|
|
if(mode & FULL_Y_RANGE)
|
|
|
|
|
{
|
|
|
|
|
maxAllowedY=255;
|
|
|
|
|
minAllowedY=0;
|
|
|
|
|
}
|
2001-10-11 00:13:27 +02:00
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if(!isColor)
|
|
|
|
|
{
|
|
|
|
|
uint64_t sum= 0;
|
2001-10-12 00:35:45 +02:00
|
|
|
|
int i;
|
|
|
|
|
static int framenum= -1;
|
|
|
|
|
uint64_t maxClipped;
|
|
|
|
|
uint64_t clipped;
|
|
|
|
|
double scale;
|
|
|
|
|
|
|
|
|
|
framenum++;
|
|
|
|
|
if(framenum == 1) yHistogram[0]= width*height/64*15/256;
|
|
|
|
|
|
|
|
|
|
for(i=0; i<256; i++)
|
|
|
|
|
{
|
2001-10-11 00:13:27 +02:00
|
|
|
|
sum+= yHistogram[i];
|
2001-10-12 00:35:45 +02:00
|
|
|
|
// printf("%d ", yHistogram[i]);
|
|
|
|
|
}
|
|
|
|
|
// printf("\n\n");
|
|
|
|
|
|
|
|
|
|
/* we allways get a completly black picture first */
|
|
|
|
|
maxClipped= (uint64_t)(sum * maxClippedThreshold);
|
2001-10-11 00:13:27 +02:00
|
|
|
|
|
2001-10-12 00:35:45 +02:00
|
|
|
|
clipped= sum;
|
2001-10-11 00:13:27 +02:00
|
|
|
|
for(black=255; black>0; black--)
|
|
|
|
|
{
|
|
|
|
|
if(clipped < maxClipped) break;
|
|
|
|
|
clipped-= yHistogram[black];
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
clipped= sum;
|
|
|
|
|
for(white=0; white<256; white++)
|
|
|
|
|
{
|
|
|
|
|
if(clipped < maxClipped) break;
|
|
|
|
|
clipped-= yHistogram[white];
|
|
|
|
|
}
|
|
|
|
|
|
2001-10-23 02:03:19 +02:00
|
|
|
|
packedYOffset= (black - minAllowedY) & 0xFFFF;
|
2001-10-11 00:13:27 +02:00
|
|
|
|
packedYOffset|= packedYOffset<<32;
|
|
|
|
|
packedYOffset|= packedYOffset<<16;
|
|
|
|
|
|
2001-10-12 00:35:45 +02:00
|
|
|
|
scale= (double)(maxAllowedY - minAllowedY) / (double)(white-black);
|
2001-10-11 00:13:27 +02:00
|
|
|
|
|
2001-10-23 02:03:19 +02:00
|
|
|
|
packedYScale= (uint16_t)(scale*1024.0 + 0.5);
|
2001-10-11 00:13:27 +02:00
|
|
|
|
packedYScale|= packedYScale<<32;
|
|
|
|
|
packedYScale|= packedYScale<<16;
|
|
|
|
|
}
|
|
|
|
|
else
|
|
|
|
|
{
|
|
|
|
|
packedYScale= 0x0100010001000100LL;
|
|
|
|
|
packedYOffset= 0;
|
|
|
|
|
}
|
|
|
|
|
|
2001-10-17 22:42:07 +02:00
|
|
|
|
/* copy first row of 8x8 blocks */
|
2001-10-12 00:35:45 +02:00
|
|
|
|
for(x=0; x<width; x+=BLOCK_SIZE)
|
|
|
|
|
blockCopy(dst + x, dstStride, src + x, srcStride, 8, mode & LEVEL_FIX);
|
2001-10-11 00:13:27 +02:00
|
|
|
|
|
2001-10-17 22:42:07 +02:00
|
|
|
|
for(y=0; y<height; y+=BLOCK_SIZE)
|
2001-10-11 00:13:27 +02:00
|
|
|
|
{
|
|
|
|
|
//1% speedup if these are here instead of the inner loop
|
|
|
|
|
uint8_t *srcBlock= &(src[y*srcStride]);
|
|
|
|
|
uint8_t *dstBlock= &(dst[y*dstStride]);
|
2001-10-17 22:42:07 +02:00
|
|
|
|
|
|
|
|
|
/* can we mess with a 8x16 block from srcBlock/dstBlock downwards, if not
|
|
|
|
|
than use a temporary buffer */
|
|
|
|
|
if(y+15 >= height)
|
|
|
|
|
{
|
|
|
|
|
/* copy from line 5 to 12 of src, these will e copied with
|
|
|
|
|
blockcopy to dst later */
|
|
|
|
|
memcpy(tempSrc + srcStride*5, srcBlock + srcStride*5,
|
|
|
|
|
srcStride*MAX(height-y-5, 0) );
|
|
|
|
|
|
|
|
|
|
/* duplicate last line to fill the void upto line 12 */
|
|
|
|
|
if(y+12 >= height)
|
|
|
|
|
{
|
|
|
|
|
int i;
|
|
|
|
|
for(i=height-y; i<=12; i++)
|
|
|
|
|
memcpy(tempSrc + srcStride*i,
|
|
|
|
|
src + srcStride*(height-1), srcStride);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
/* copy up to 5 lines of dst */
|
|
|
|
|
memcpy(tempDst, dstBlock, dstStride*MIN(height-y, 5) );
|
|
|
|
|
dstBlock= tempDst;
|
|
|
|
|
srcBlock= tempSrc;
|
|
|
|
|
}
|
2001-10-11 00:13:27 +02:00
|
|
|
|
|
2001-10-19 15:41:38 +02:00
|
|
|
|
// From this point on it is guranteed that we can read and write 16 lines downward
|
2001-10-11 00:13:27 +02:00
|
|
|
|
// finish 1 block before the next otherwise we<77>ll might have a problem
|
|
|
|
|
// with the L1 Cache of the P4 ... or only a few blocks at a time or soemthing
|
2001-10-12 00:35:45 +02:00
|
|
|
|
for(x=0; x<width; x+=BLOCK_SIZE)
|
2001-10-11 00:13:27 +02:00
|
|
|
|
{
|
2001-10-12 00:35:45 +02:00
|
|
|
|
const int stride= dstStride;
|
2001-10-11 00:13:27 +02:00
|
|
|
|
int QP= isColor ?
|
|
|
|
|
QPs[(y>>3)*QPStride + (x>>3)]:
|
2001-10-13 11:30:07 +02:00
|
|
|
|
QPs[(y>>4)*QPStride + (x>>4)];
|
|
|
|
|
if(!isColor && (mode & LEVEL_FIX)) QP= (QP* (packedYScale &0xFFFF))>>8;
|
2001-10-11 00:13:27 +02:00
|
|
|
|
#ifdef HAVE_MMX
|
2001-10-17 22:42:07 +02:00
|
|
|
|
asm volatile(
|
|
|
|
|
"movd %0, %%mm7 \n\t"
|
|
|
|
|
"packuswb %%mm7, %%mm7 \n\t" // 0, 0, 0, QP, 0, 0, 0, QP
|
|
|
|
|
"packuswb %%mm7, %%mm7 \n\t" // 0,QP, 0, QP, 0,QP, 0, QP
|
|
|
|
|
"packuswb %%mm7, %%mm7 \n\t" // QP,..., QP
|
|
|
|
|
"movq %%mm7, pQPb \n\t"
|
|
|
|
|
: : "r" (QP)
|
|
|
|
|
);
|
2001-10-11 00:13:27 +02:00
|
|
|
|
#endif
|
|
|
|
|
|
2001-10-16 04:31:14 +02:00
|
|
|
|
#ifdef MORE_TIMING
|
2001-10-17 22:42:07 +02:00
|
|
|
|
T0= rdtsc();
|
2001-10-11 00:13:27 +02:00
|
|
|
|
#endif
|
|
|
|
|
|
2001-10-11 00:21:19 +02:00
|
|
|
|
#ifdef HAVE_MMX2
|
2001-10-17 22:42:07 +02:00
|
|
|
|
prefetchnta(srcBlock + (((x>>3)&3) + 5)*srcStride + 32);
|
|
|
|
|
prefetchnta(srcBlock + (((x>>3)&3) + 9)*srcStride + 32);
|
|
|
|
|
prefetcht0(dstBlock + (((x>>3)&3) + 5)*dstStride + 32);
|
|
|
|
|
prefetcht0(dstBlock + (((x>>3)&3) + 9)*dstStride + 32);
|
2001-10-11 00:21:19 +02:00
|
|
|
|
#elif defined(HAVE_3DNOW)
|
|
|
|
|
//FIXME check if this is faster on an 3dnow chip or if its faster without the prefetch or ...
|
2001-10-17 22:42:07 +02:00
|
|
|
|
/* prefetch(srcBlock + (((x>>3)&3) + 5)*srcStride + 32);
|
|
|
|
|
prefetch(srcBlock + (((x>>3)&3) + 9)*srcStride + 32);
|
|
|
|
|
prefetchw(dstBlock + (((x>>3)&3) + 5)*dstStride + 32);
|
|
|
|
|
prefetchw(dstBlock + (((x>>3)&3) + 9)*dstStride + 32);
|
2001-10-11 00:21:19 +02:00
|
|
|
|
*/
|
2001-10-11 00:13:27 +02:00
|
|
|
|
#endif
|
2001-10-17 22:42:07 +02:00
|
|
|
|
|
|
|
|
|
if(!isColor) yHistogram[ srcBlock[srcStride*5] ]++;
|
|
|
|
|
|
2001-10-19 15:56:12 +02:00
|
|
|
|
#ifdef PP_FUNNY_STRIDE
|
2001-10-19 15:41:38 +02:00
|
|
|
|
//can we mess with a 8x16 block, if not use a temp buffer, yes again
|
|
|
|
|
if(x+7 >= width)
|
|
|
|
|
{
|
|
|
|
|
int i;
|
|
|
|
|
dstBlockPtrBackup= dstBlock;
|
|
|
|
|
srcBlockPtrBackup= srcBlock;
|
|
|
|
|
|
|
|
|
|
for(i=0;i<BLOCK_SIZE*2; i++)
|
|
|
|
|
{
|
|
|
|
|
memcpy(tempSrcBlock+i*srcStride, srcBlock+i*srcStride, width-x);
|
|
|
|
|
memcpy(tempDstBlock+i*dstStride, dstBlock+i*dstStride, width-x);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
dstBlock= tempDstBlock;
|
|
|
|
|
srcBlock= tempSrcBlock;
|
|
|
|
|
}
|
2001-10-19 15:56:12 +02:00
|
|
|
|
#endif
|
2001-10-19 15:41:38 +02:00
|
|
|
|
|
2001-10-17 22:42:07 +02:00
|
|
|
|
blockCopy(dstBlock + dstStride*5, dstStride,
|
|
|
|
|
srcBlock + srcStride*5, srcStride, 8, mode & LEVEL_FIX);
|
|
|
|
|
|
|
|
|
|
if(mode & LINEAR_IPOL_DEINT_FILTER)
|
|
|
|
|
deInterlaceInterpolateLinear(dstBlock, dstStride);
|
|
|
|
|
else if(mode & LINEAR_BLEND_DEINT_FILTER)
|
|
|
|
|
deInterlaceBlendLinear(dstBlock, dstStride);
|
|
|
|
|
else if(mode & MEDIAN_DEINT_FILTER)
|
|
|
|
|
deInterlaceMedian(dstBlock, dstStride);
|
|
|
|
|
else if(mode & CUBIC_IPOL_DEINT_FILTER)
|
|
|
|
|
deInterlaceInterpolateCubic(dstBlock, dstStride);
|
|
|
|
|
/* else if(mode & CUBIC_BLEND_DEINT_FILTER)
|
|
|
|
|
deInterlaceBlendCubic(dstBlock, dstStride);
|
2001-10-15 05:01:08 +02:00
|
|
|
|
*/
|
2001-10-11 00:13:27 +02:00
|
|
|
|
|
2001-10-17 22:42:07 +02:00
|
|
|
|
/* only deblock if we have 2 blocks */
|
|
|
|
|
if(y + 8 < height)
|
|
|
|
|
{
|
2001-10-16 04:31:14 +02:00
|
|
|
|
#ifdef MORE_TIMING
|
2001-10-11 00:13:27 +02:00
|
|
|
|
T1= rdtsc();
|
|
|
|
|
memcpyTime+= T1-T0;
|
|
|
|
|
T0=T1;
|
|
|
|
|
#endif
|
2001-10-19 19:03:51 +02:00
|
|
|
|
if(mode & V_RK1_FILTER)
|
|
|
|
|
vertRK1Filter(dstBlock, stride, QP);
|
|
|
|
|
else if(mode & V_X1_FILTER)
|
|
|
|
|
vertX1Filter(dstBlock, stride, QP);
|
|
|
|
|
else if(mode & V_DEBLOCK)
|
2001-10-11 00:13:27 +02:00
|
|
|
|
{
|
2001-10-19 19:03:51 +02:00
|
|
|
|
if( isVertDC(dstBlock, stride))
|
2001-10-11 00:21:19 +02:00
|
|
|
|
{
|
2001-10-19 19:03:51 +02:00
|
|
|
|
if(isVertMinMaxOk(dstBlock, stride, QP))
|
|
|
|
|
doVertLowPass(dstBlock, stride, QP);
|
2001-10-11 00:21:19 +02:00
|
|
|
|
}
|
2001-10-19 19:03:51 +02:00
|
|
|
|
else
|
|
|
|
|
doVertDefFilter(dstBlock, stride, QP);
|
2001-10-11 00:13:27 +02:00
|
|
|
|
}
|
2001-10-16 04:31:14 +02:00
|
|
|
|
#ifdef MORE_TIMING
|
2001-10-11 00:13:27 +02:00
|
|
|
|
T1= rdtsc();
|
|
|
|
|
vertTime+= T1-T0;
|
|
|
|
|
T0=T1;
|
|
|
|
|
#endif
|
|
|
|
|
}
|
|
|
|
|
|
2001-10-17 22:42:07 +02:00
|
|
|
|
/* check if we have a previous block to deblock it with dstBlock */
|
2001-10-19 15:41:38 +02:00
|
|
|
|
if(x - 8 >= 0)
|
2001-10-11 00:13:27 +02:00
|
|
|
|
{
|
2001-10-16 04:31:14 +02:00
|
|
|
|
#ifdef MORE_TIMING
|
2001-10-11 00:13:27 +02:00
|
|
|
|
T0= rdtsc();
|
|
|
|
|
#endif
|
2001-10-19 19:03:51 +02:00
|
|
|
|
if(mode & H_X1_FILTER)
|
|
|
|
|
horizX1Filter(dstBlock-4, stride, QP);
|
|
|
|
|
else if(mode & H_DEBLOCK)
|
2001-10-11 00:13:27 +02:00
|
|
|
|
{
|
2001-10-19 19:03:51 +02:00
|
|
|
|
if( isHorizDCAndCopy2Temp(dstBlock-4, stride))
|
2001-10-11 00:21:19 +02:00
|
|
|
|
{
|
2001-10-19 19:03:51 +02:00
|
|
|
|
if(isHorizMinMaxOk(tempBlock, TEMP_STRIDE, QP))
|
|
|
|
|
doHorizLowPassAndCopyBack(dstBlock-4, stride, QP);
|
2001-10-11 00:21:19 +02:00
|
|
|
|
}
|
2001-10-19 19:03:51 +02:00
|
|
|
|
else
|
|
|
|
|
doHorizDefFilterAndCopyBack(dstBlock-4, stride, QP);
|
2001-10-11 00:13:27 +02:00
|
|
|
|
}
|
2001-10-16 04:31:14 +02:00
|
|
|
|
#ifdef MORE_TIMING
|
2001-10-11 00:13:27 +02:00
|
|
|
|
T1= rdtsc();
|
|
|
|
|
horizTime+= T1-T0;
|
|
|
|
|
T0=T1;
|
|
|
|
|
#endif
|
|
|
|
|
dering(dstBlock - 9 - stride, stride, QP);
|
|
|
|
|
}
|
|
|
|
|
else if(y!=0)
|
|
|
|
|
dering(dstBlock - stride*9 + width-9, stride, QP);
|
|
|
|
|
//FIXME dering filter will not be applied to last block (bottom right)
|
|
|
|
|
|
2001-10-19 15:56:12 +02:00
|
|
|
|
#ifdef PP_FUNNY_STRIDE
|
2001-10-19 15:41:38 +02:00
|
|
|
|
/* did we use a tmp-block buffer */
|
|
|
|
|
if(x+7 >= width)
|
|
|
|
|
{
|
|
|
|
|
int i;
|
|
|
|
|
dstBlock= dstBlockPtrBackup;
|
|
|
|
|
srcBlock= srcBlockPtrBackup;
|
|
|
|
|
|
|
|
|
|
for(i=0;i<BLOCK_SIZE*2; i++)
|
|
|
|
|
{
|
|
|
|
|
memcpy(dstBlock+i*dstStride, tempDstBlock+i*dstStride, width-x);
|
|
|
|
|
}
|
|
|
|
|
}
|
2001-10-19 15:56:12 +02:00
|
|
|
|
#endif
|
2001-10-19 15:41:38 +02:00
|
|
|
|
|
2001-10-11 00:13:27 +02:00
|
|
|
|
dstBlock+=8;
|
|
|
|
|
srcBlock+=8;
|
2001-10-17 22:42:07 +02:00
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* did we use a tmp buffer */
|
2001-10-19 15:41:38 +02:00
|
|
|
|
if(y+15 >= height)
|
2001-10-17 22:42:07 +02:00
|
|
|
|
{
|
|
|
|
|
uint8_t *dstBlock= &(dst[y*dstStride]);
|
|
|
|
|
memcpy(dstBlock, tempDst, dstStride*(height-y) );
|
2001-10-11 00:13:27 +02:00
|
|
|
|
}
|
|
|
|
|
}
|
2001-10-11 00:21:19 +02:00
|
|
|
|
#ifdef HAVE_3DNOW
|
|
|
|
|
asm volatile("femms");
|
|
|
|
|
#elif defined (HAVE_MMX)
|
2001-10-11 00:13:27 +02:00
|
|
|
|
asm volatile("emms");
|
|
|
|
|
#endif
|
|
|
|
|
|
2001-10-16 04:31:14 +02:00
|
|
|
|
#ifdef TIMING
|
2001-10-11 00:13:27 +02:00
|
|
|
|
// FIXME diff is mostly the time spent for rdtsc (should subtract that but ...)
|
|
|
|
|
sumTime= rdtsc() - sumTime;
|
|
|
|
|
if(!isColor)
|
|
|
|
|
printf("cpy:%4dk, vert:%4dk, horiz:%4dk, sum:%4dk, diff:%4dk, color: %d/%d \r",
|
2001-10-12 00:35:45 +02:00
|
|
|
|
(int)(memcpyTime/1000), (int)(vertTime/1000), (int)(horizTime/1000),
|
|
|
|
|
(int)(sumTime/1000), (int)((sumTime-memcpyTime-vertTime-horizTime)/1000)
|
2001-10-11 00:13:27 +02:00
|
|
|
|
, black, white);
|
|
|
|
|
#endif
|
|
|
|
|
}
|