+/*
+ Copyright (C) 2001-2002 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.
-// Software scaling and colorspace conversion routines for MPlayer
+ 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.
-// Orginal C implementation by A'rpi/ESP-team <arpi@thot.banki.hu>
-// current version mostly by Michael Niedermayer (michaelni@gmx.at)
-// the parts written by michael are under GNU GPL
+ 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
+*/
+
+/*
+ supported Input formats: YV12 (grayscale soon too)
+ supported output formats: YV12, BGR15, BGR16, BGR24, BGR32 (grayscale soon too)
+*/
#include <inttypes.h>
#include <string.h>
+#include <math.h>
+#include <stdio.h>
#include "../config.h"
+#include "../mangle.h"
+#ifdef HAVE_MALLOC_H
+#include <malloc.h>
+#endif
#include "swscale.h"
-#include "../mmx_defs.h"
+#include "../cpudetect.h"
+#include "../libvo/img_format.h"
+#undef MOVNTQ
+#undef PAVGB
//#undef HAVE_MMX2
+//#define HAVE_3DNOW
//#undef HAVE_MMX
//#undef ARCH_X86
#define DITHER1XBPP
-int fullUVIpol=0;
-//disables the unscaled height version
-int allwaysIpol=0;
#define RET 0xC3 //near return opcode
+
+#ifdef MP_DEBUG
+#define ASSERT(x) if(!(x)) { printf("ASSERT " #x " failed\n"); *((int*)0)=0; }
+#else
+#define ASSERT(x) ;
+#endif
+
+#ifdef M_PI
+#define PI M_PI
+#else
+#define PI 3.14159265358979323846
+#endif
+
+extern int verbose; // defined in mplayer.c
/*
NOTES
known BUGS with known cause (no bugreports please!, but patches are welcome :) )
-horizontal MMX2 scaler reads 1-7 samples too much (might cause a sig11)
+horizontal fast_bilinear MMX2 scaler reads 1-7 samples too much (might cause a sig11)
-Supported output formats BGR15 BGR16 BGR24 BGR32
+Supported output formats BGR15 BGR16 BGR24 BGR32 YV12
BGR15 & BGR16 MMX verions support dithering
Special versions: fast Y 1:1 scaling (no interpolation in y direction)
TODO
more intelligent missalignment avoidance for the horizontal scaler
+change the distance of the u & v buffer
+write special vertical cubic upscale version
+Optimize C code (yv12 / minmax)
+add support for packed pixel yuv input & output
+add support for Y8 input & output
+add BGR4 output support
+add BGR32 / BGR24 input support
*/
#define ABS(a) ((a) > 0 ? (a) : (-(a)))
#define MIN(a,b) ((a) > (b) ? (b) : (a))
#define MAX(a,b) ((a) < (b) ? (b) : (a))
-#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
-
-#ifdef HAVE_MMX2
-#define MOVNTQ(a,b) "movntq " #a ", " #b " \n\t"
-#else
-#define MOVNTQ(a,b) "movq " #a ", " #b " \n\t"
+#ifdef ARCH_X86
+#define CAN_COMPILE_X86_ASM
#endif
-
-#ifdef HAVE_MMX
+#ifdef CAN_COMPILE_X86_ASM
static uint64_t __attribute__((aligned(8))) yCoeff= 0x2568256825682568LL;
static uint64_t __attribute__((aligned(8))) vrCoeff= 0x3343334333433343LL;
static uint64_t __attribute__((aligned(8))) ubCoeff= 0x40cf40cf40cf40cfLL;
static uint64_t __attribute__((aligned(8))) vgCoeff= 0xE5E2E5E2E5E2E5E2LL;
static uint64_t __attribute__((aligned(8))) ugCoeff= 0xF36EF36EF36EF36ELL;
+static uint64_t __attribute__((aligned(8))) bF8= 0xF8F8F8F8F8F8F8F8LL;
+static uint64_t __attribute__((aligned(8))) bFC= 0xFCFCFCFCFCFCFCFCLL;
static uint64_t __attribute__((aligned(8))) w400= 0x0400040004000400LL;
static uint64_t __attribute__((aligned(8))) w80= 0x0080008000800080LL;
static uint64_t __attribute__((aligned(8))) w10= 0x0010001000100010LL;
+static uint64_t __attribute__((aligned(8))) w02= 0x0002000200020002LL;
static uint64_t __attribute__((aligned(8))) bm00001111=0x00000000FFFFFFFFLL;
static uint64_t __attribute__((aligned(8))) bm00000111=0x0000000000FFFFFFLL;
static uint64_t __attribute__((aligned(8))) bm11111000=0xFFFFFFFFFF000000LL;
-static uint64_t __attribute__((aligned(8))) b16Dither= 0x0004000400040004LL;
-static uint64_t __attribute__((aligned(8))) b16Dither1=0x0004000400040004LL;
-static uint64_t __attribute__((aligned(8))) b16Dither2=0x0602060206020602LL;
-static uint64_t __attribute__((aligned(8))) g16Dither= 0x0002000200020002LL;
-static uint64_t __attribute__((aligned(8))) g16Dither1=0x0002000200020002LL;
-static uint64_t __attribute__((aligned(8))) g16Dither2=0x0301030103010301LL;
+static volatile uint64_t __attribute__((aligned(8))) b5Dither;
+static volatile uint64_t __attribute__((aligned(8))) g5Dither;
+static volatile uint64_t __attribute__((aligned(8))) g6Dither;
+static volatile uint64_t __attribute__((aligned(8))) r5Dither;
+
+static uint64_t __attribute__((aligned(8))) dither4[2]={
+ 0x0103010301030103LL,
+ 0x0200020002000200LL,};
+
+static uint64_t __attribute__((aligned(8))) dither8[2]={
+ 0x0602060206020602LL,
+ 0x0004000400040004LL,};
static uint64_t __attribute__((aligned(8))) b16Mask= 0x001F001F001F001FLL;
static uint64_t __attribute__((aligned(8))) g16Mask= 0x07E007E007E007E0LL;
static uint64_t __attribute__((aligned(8))) g15Mask= 0x03E003E003E003E0LL;
static uint64_t __attribute__((aligned(8))) r15Mask= 0x7C007C007C007C00LL;
-static uint64_t __attribute__((aligned(8))) temp0;
+static uint64_t __attribute__((aligned(8))) M24A= 0x00FF0000FF0000FFLL;
+static uint64_t __attribute__((aligned(8))) M24B= 0xFF0000FF0000FF00LL;
+static uint64_t __attribute__((aligned(8))) M24C= 0x0000FF0000FF0000LL;
+
+// FIXME remove
static uint64_t __attribute__((aligned(8))) asm_yalpha1;
static uint64_t __attribute__((aligned(8))) asm_uvalpha1;
#endif
-// temporary storage for 4 yuv lines:
-// 16bit for now (mmx likes it more compact)
-#ifdef HAVE_MMX
-static uint16_t __attribute__((aligned(8))) pix_buf_y[4][2048];
-static uint16_t __attribute__((aligned(8))) pix_buf_uv[2][2048*2];
-#else
-static uint16_t pix_buf_y[4][2048];
-static uint16_t pix_buf_uv[2][2048*2];
-#endif
-
// clipping helper table for C implementations:
static unsigned char clip_table[768];
+static unsigned short clip_table16b[768];
+static unsigned short clip_table16g[768];
+static unsigned short clip_table16r[768];
+static unsigned short clip_table15b[768];
+static unsigned short clip_table15g[768];
+static unsigned short clip_table15r[768];
+
// yuv->rgb conversion tables:
static int yuvtab_2568[256];
static int yuvtab_3343[256];
static int yuvtab_0c92[256];
static int yuvtab_1a1e[256];
static int yuvtab_40cf[256];
+// Needed for cubic scaler to catch overflows
+static int clip_yuvtab_2568[768];
+static int clip_yuvtab_3343[768];
+static int clip_yuvtab_0c92[768];
+static int clip_yuvtab_1a1e[768];
+static int clip_yuvtab_40cf[768];
+//global sws_flags from the command line
+int sws_flags=0;
-static uint8_t funnyYCode[10000];
-static uint8_t funnyUVCode[10000];
-
-static int canMMX2BeUsed=0;
-
-#define FULL_YSCALEYUV2RGB \
- "pxor %%mm7, %%mm7 \n\t"\
- "movd %6, %%mm6 \n\t" /*yalpha1*/\
- "punpcklwd %%mm6, %%mm6 \n\t"\
- "punpcklwd %%mm6, %%mm6 \n\t"\
- "movd %7, %%mm5 \n\t" /*uvalpha1*/\
- "punpcklwd %%mm5, %%mm5 \n\t"\
- "punpcklwd %%mm5, %%mm5 \n\t"\
- "xorl %%eax, %%eax \n\t"\
- "1: \n\t"\
- "movq (%0, %%eax, 2), %%mm0 \n\t" /*buf0[eax]*/\
- "movq (%1, %%eax, 2), %%mm1 \n\t" /*buf1[eax]*/\
- "movq (%2, %%eax,2), %%mm2 \n\t" /* uvbuf0[eax]*/\
- "movq (%3, %%eax,2), %%mm3 \n\t" /* uvbuf1[eax]*/\
- "psubw %%mm1, %%mm0 \n\t" /* buf0[eax] - buf1[eax]*/\
- "psubw %%mm3, %%mm2 \n\t" /* uvbuf0[eax] - uvbuf1[eax]*/\
- "pmulhw %%mm6, %%mm0 \n\t" /* (buf0[eax] - buf1[eax])yalpha1>>16*/\
- "pmulhw %%mm5, %%mm2 \n\t" /* (uvbuf0[eax] - uvbuf1[eax])uvalpha1>>16*/\
- "psraw $4, %%mm1 \n\t" /* buf0[eax] - buf1[eax] >>4*/\
- "movq 4096(%2, %%eax,2), %%mm4 \n\t" /* uvbuf0[eax+2048]*/\
- "psraw $4, %%mm3 \n\t" /* uvbuf0[eax] - uvbuf1[eax] >>4*/\
- "paddw %%mm0, %%mm1 \n\t" /* buf0[eax]yalpha1 + buf1[eax](1-yalpha1) >>16*/\
- "movq 4096(%3, %%eax,2), %%mm0 \n\t" /* uvbuf1[eax+2048]*/\
- "paddw %%mm2, %%mm3 \n\t" /* uvbuf0[eax]uvalpha1 - uvbuf1[eax](1-uvalpha1)*/\
- "psubw %%mm0, %%mm4 \n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048]*/\
- "psubw w80, %%mm1 \n\t" /* 8(Y-16)*/\
- "psubw w400, %%mm3 \n\t" /* 8(U-128)*/\
- "pmulhw yCoeff, %%mm1 \n\t"\
-\
-\
- "pmulhw %%mm5, %%mm4 \n\t" /* (uvbuf0[eax+2048] - uvbuf1[eax+2048])uvalpha1>>16*/\
- "movq %%mm3, %%mm2 \n\t" /* (U-128)8*/\
- "pmulhw ubCoeff, %%mm3 \n\t"\
- "psraw $4, %%mm0 \n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048] >>4*/\
- "pmulhw ugCoeff, %%mm2 \n\t"\
- "paddw %%mm4, %%mm0 \n\t" /* uvbuf0[eax+2048]uvalpha1 - uvbuf1[eax+2048](1-uvalpha1)*/\
- "psubw w400, %%mm0 \n\t" /* (V-128)8*/\
-\
-\
- "movq %%mm0, %%mm4 \n\t" /* (V-128)8*/\
- "pmulhw vrCoeff, %%mm0 \n\t"\
- "pmulhw vgCoeff, %%mm4 \n\t"\
- "paddw %%mm1, %%mm3 \n\t" /* B*/\
- "paddw %%mm1, %%mm0 \n\t" /* R*/\
- "packuswb %%mm3, %%mm3 \n\t"\
-\
- "packuswb %%mm0, %%mm0 \n\t"\
- "paddw %%mm4, %%mm2 \n\t"\
- "paddw %%mm2, %%mm1 \n\t" /* G*/\
-\
- "packuswb %%mm1, %%mm1 \n\t"
-
-#define YSCALEYUV2RGB \
- "movd %6, %%mm6 \n\t" /*yalpha1*/\
- "punpcklwd %%mm6, %%mm6 \n\t"\
- "punpcklwd %%mm6, %%mm6 \n\t"\
- "movq %%mm6, asm_yalpha1 \n\t"\
- "movd %7, %%mm5 \n\t" /*uvalpha1*/\
- "punpcklwd %%mm5, %%mm5 \n\t"\
- "punpcklwd %%mm5, %%mm5 \n\t"\
- "movq %%mm5, asm_uvalpha1 \n\t"\
- "xorl %%eax, %%eax \n\t"\
- "1: \n\t"\
- "movq (%2, %%eax), %%mm2 \n\t" /* uvbuf0[eax]*/\
- "movq (%3, %%eax), %%mm3 \n\t" /* uvbuf1[eax]*/\
- "movq 4096(%2, %%eax), %%mm5 \n\t" /* uvbuf0[eax+2048]*/\
- "movq 4096(%3, %%eax), %%mm4 \n\t" /* uvbuf1[eax+2048]*/\
- "psubw %%mm3, %%mm2 \n\t" /* uvbuf0[eax] - uvbuf1[eax]*/\
- "psubw %%mm4, %%mm5 \n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048]*/\
- "movq asm_uvalpha1, %%mm0 \n\t"\
- "pmulhw %%mm0, %%mm2 \n\t" /* (uvbuf0[eax] - uvbuf1[eax])uvalpha1>>16*/\
- "pmulhw %%mm0, %%mm5 \n\t" /* (uvbuf0[eax+2048] - uvbuf1[eax+2048])uvalpha1>>16*/\
- "psraw $4, %%mm3 \n\t" /* uvbuf0[eax] - uvbuf1[eax] >>4*/\
- "psraw $4, %%mm4 \n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048] >>4*/\
- "paddw %%mm2, %%mm3 \n\t" /* uvbuf0[eax]uvalpha1 - uvbuf1[eax](1-uvalpha1)*/\
- "paddw %%mm5, %%mm4 \n\t" /* uvbuf0[eax+2048]uvalpha1 - uvbuf1[eax+2048](1-uvalpha1)*/\
- "psubw w400, %%mm3 \n\t" /* (U-128)8*/\
- "psubw w400, %%mm4 \n\t" /* (V-128)8*/\
- "movq %%mm3, %%mm2 \n\t" /* (U-128)8*/\
- "movq %%mm4, %%mm5 \n\t" /* (V-128)8*/\
- "pmulhw ugCoeff, %%mm3 \n\t"\
- "pmulhw vgCoeff, %%mm4 \n\t"\
- /* mm2=(U-128)8, mm3=ug, mm4=vg mm5=(V-128)8 */\
- "movq (%0, %%eax, 2), %%mm0 \n\t" /*buf0[eax]*/\
- "movq (%1, %%eax, 2), %%mm1 \n\t" /*buf1[eax]*/\
- "movq 8(%0, %%eax, 2), %%mm6 \n\t" /*buf0[eax]*/\
- "movq 8(%1, %%eax, 2), %%mm7 \n\t" /*buf1[eax]*/\
- "psubw %%mm1, %%mm0 \n\t" /* buf0[eax] - buf1[eax]*/\
- "psubw %%mm7, %%mm6 \n\t" /* buf0[eax] - buf1[eax]*/\
- "pmulhw asm_yalpha1, %%mm0 \n\t" /* (buf0[eax] - buf1[eax])yalpha1>>16*/\
- "pmulhw asm_yalpha1, %%mm6 \n\t" /* (buf0[eax] - buf1[eax])yalpha1>>16*/\
- "psraw $4, %%mm1 \n\t" /* buf0[eax] - buf1[eax] >>4*/\
- "psraw $4, %%mm7 \n\t" /* buf0[eax] - buf1[eax] >>4*/\
- "paddw %%mm0, %%mm1 \n\t" /* buf0[eax]yalpha1 + buf1[eax](1-yalpha1) >>16*/\
- "paddw %%mm6, %%mm7 \n\t" /* buf0[eax]yalpha1 + buf1[eax](1-yalpha1) >>16*/\
- "pmulhw ubCoeff, %%mm2 \n\t"\
- "pmulhw vrCoeff, %%mm5 \n\t"\
- "psubw w80, %%mm1 \n\t" /* 8(Y-16)*/\
- "psubw w80, %%mm7 \n\t" /* 8(Y-16)*/\
- "pmulhw yCoeff, %%mm1 \n\t"\
- "pmulhw yCoeff, %%mm7 \n\t"\
- /* mm1= Y1, mm2=ub, mm3=ug, mm4=vg mm5=vr, mm7=Y2 */\
- "paddw %%mm3, %%mm4 \n\t"\
- "movq %%mm2, %%mm0 \n\t"\
- "movq %%mm5, %%mm6 \n\t"\
- "movq %%mm4, %%mm3 \n\t"\
- "punpcklwd %%mm2, %%mm2 \n\t"\
- "punpcklwd %%mm5, %%mm5 \n\t"\
- "punpcklwd %%mm4, %%mm4 \n\t"\
- "paddw %%mm1, %%mm2 \n\t"\
- "paddw %%mm1, %%mm5 \n\t"\
- "paddw %%mm1, %%mm4 \n\t"\
- "punpckhwd %%mm0, %%mm0 \n\t"\
- "punpckhwd %%mm6, %%mm6 \n\t"\
- "punpckhwd %%mm3, %%mm3 \n\t"\
- "paddw %%mm7, %%mm0 \n\t"\
- "paddw %%mm7, %%mm6 \n\t"\
- "paddw %%mm7, %%mm3 \n\t"\
- /* mm0=B1, mm2=B2, mm3=G2, mm4=G1, mm5=R1, mm6=R2 */\
- "packuswb %%mm0, %%mm2 \n\t"\
- "packuswb %%mm6, %%mm5 \n\t"\
- "packuswb %%mm3, %%mm4 \n\t"\
- "pxor %%mm7, %%mm7 \n\t"
-
-#define YSCALEYUV2RGB1 \
- "xorl %%eax, %%eax \n\t"\
- "1: \n\t"\
- "movq (%2, %%eax), %%mm3 \n\t" /* uvbuf0[eax]*/\
- "movq 4096(%2, %%eax), %%mm4 \n\t" /* uvbuf0[eax+2048]*/\
- "psraw $4, %%mm3 \n\t" /* uvbuf0[eax] - uvbuf1[eax] >>4*/\
- "psraw $4, %%mm4 \n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048] >>4*/\
- "psubw w400, %%mm3 \n\t" /* (U-128)8*/\
- "psubw w400, %%mm4 \n\t" /* (V-128)8*/\
- "movq %%mm3, %%mm2 \n\t" /* (U-128)8*/\
- "movq %%mm4, %%mm5 \n\t" /* (V-128)8*/\
- "pmulhw ugCoeff, %%mm3 \n\t"\
- "pmulhw vgCoeff, %%mm4 \n\t"\
- /* mm2=(U-128)8, mm3=ug, mm4=vg mm5=(V-128)8 */\
- "movq (%1, %%eax, 2), %%mm1 \n\t" /*buf0[eax]*/\
- "movq 8(%1, %%eax, 2), %%mm7 \n\t" /*buf0[eax]*/\
- "psraw $4, %%mm1 \n\t" /* buf0[eax] - buf1[eax] >>4*/\
- "psraw $4, %%mm7 \n\t" /* buf0[eax] - buf1[eax] >>4*/\
- "pmulhw ubCoeff, %%mm2 \n\t"\
- "pmulhw vrCoeff, %%mm5 \n\t"\
- "psubw w80, %%mm1 \n\t" /* 8(Y-16)*/\
- "psubw w80, %%mm7 \n\t" /* 8(Y-16)*/\
- "pmulhw yCoeff, %%mm1 \n\t"\
- "pmulhw yCoeff, %%mm7 \n\t"\
- /* mm1= Y1, mm2=ub, mm3=ug, mm4=vg mm5=vr, mm7=Y2 */\
- "paddw %%mm3, %%mm4 \n\t"\
- "movq %%mm2, %%mm0 \n\t"\
- "movq %%mm5, %%mm6 \n\t"\
- "movq %%mm4, %%mm3 \n\t"\
- "punpcklwd %%mm2, %%mm2 \n\t"\
- "punpcklwd %%mm5, %%mm5 \n\t"\
- "punpcklwd %%mm4, %%mm4 \n\t"\
- "paddw %%mm1, %%mm2 \n\t"\
- "paddw %%mm1, %%mm5 \n\t"\
- "paddw %%mm1, %%mm4 \n\t"\
- "punpckhwd %%mm0, %%mm0 \n\t"\
- "punpckhwd %%mm6, %%mm6 \n\t"\
- "punpckhwd %%mm3, %%mm3 \n\t"\
- "paddw %%mm7, %%mm0 \n\t"\
- "paddw %%mm7, %%mm6 \n\t"\
- "paddw %%mm7, %%mm3 \n\t"\
- /* mm0=B1, mm2=B2, mm3=G2, mm4=G1, mm5=R1, mm6=R2 */\
- "packuswb %%mm0, %%mm2 \n\t"\
- "packuswb %%mm6, %%mm5 \n\t"\
- "packuswb %%mm3, %%mm4 \n\t"\
- "pxor %%mm7, %%mm7 \n\t"
-
-#define WRITEBGR32 \
- /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */\
- "movq %%mm2, %%mm1 \n\t" /* B */\
- "movq %%mm5, %%mm6 \n\t" /* R */\
- "punpcklbw %%mm4, %%mm2 \n\t" /* GBGBGBGB 0 */\
- "punpcklbw %%mm7, %%mm5 \n\t" /* 0R0R0R0R 0 */\
- "punpckhbw %%mm4, %%mm1 \n\t" /* GBGBGBGB 2 */\
- "punpckhbw %%mm7, %%mm6 \n\t" /* 0R0R0R0R 2 */\
- "movq %%mm2, %%mm0 \n\t" /* GBGBGBGB 0 */\
- "movq %%mm1, %%mm3 \n\t" /* GBGBGBGB 2 */\
- "punpcklwd %%mm5, %%mm0 \n\t" /* 0RGB0RGB 0 */\
- "punpckhwd %%mm5, %%mm2 \n\t" /* 0RGB0RGB 1 */\
- "punpcklwd %%mm6, %%mm1 \n\t" /* 0RGB0RGB 2 */\
- "punpckhwd %%mm6, %%mm3 \n\t" /* 0RGB0RGB 3 */\
-\
- MOVNTQ(%%mm0, (%4, %%eax, 4))\
- MOVNTQ(%%mm2, 8(%4, %%eax, 4))\
- MOVNTQ(%%mm1, 16(%4, %%eax, 4))\
- MOVNTQ(%%mm3, 24(%4, %%eax, 4))\
-\
- "addl $8, %%eax \n\t"\
- "cmpl %5, %%eax \n\t"\
- " jb 1b \n\t"
-
-#define WRITEBGR16 \
- "movq %%mm2, %%mm1 \n\t" /* B */\
- "movq %%mm4, %%mm3 \n\t" /* G */\
- "movq %%mm5, %%mm6 \n\t" /* R */\
-\
- "punpcklbw %%mm7, %%mm3 \n\t" /* 0G0G0G0G */\
- "punpcklbw %%mm7, %%mm2 \n\t" /* 0B0B0B0B */\
- "punpcklbw %%mm7, %%mm5 \n\t" /* 0R0R0R0R */\
-\
- "psrlw $3, %%mm2 \n\t"\
- "psllw $3, %%mm3 \n\t"\
- "psllw $8, %%mm5 \n\t"\
-\
- "pand g16Mask, %%mm3 \n\t"\
- "pand r16Mask, %%mm5 \n\t"\
-\
- "por %%mm3, %%mm2 \n\t"\
- "por %%mm5, %%mm2 \n\t"\
-\
- "punpckhbw %%mm7, %%mm4 \n\t" /* 0G0G0G0G */\
- "punpckhbw %%mm7, %%mm1 \n\t" /* 0B0B0B0B */\
- "punpckhbw %%mm7, %%mm6 \n\t" /* 0R0R0R0R */\
-\
- "psrlw $3, %%mm1 \n\t"\
- "psllw $3, %%mm4 \n\t"\
- "psllw $8, %%mm6 \n\t"\
-\
- "pand g16Mask, %%mm4 \n\t"\
- "pand r16Mask, %%mm6 \n\t"\
-\
- "por %%mm4, %%mm1 \n\t"\
- "por %%mm6, %%mm1 \n\t"\
-\
- MOVNTQ(%%mm2, (%4, %%eax, 2))\
- MOVNTQ(%%mm1, 8(%4, %%eax, 2))\
-\
- "addl $8, %%eax \n\t"\
- "cmpl %5, %%eax \n\t"\
- " jb 1b \n\t"
-
-#define WRITEBGR15 \
- "movq %%mm2, %%mm1 \n\t" /* B */\
- "movq %%mm4, %%mm3 \n\t" /* G */\
- "movq %%mm5, %%mm6 \n\t" /* R */\
-\
- "punpcklbw %%mm7, %%mm3 \n\t" /* 0G0G0G0G */\
- "punpcklbw %%mm7, %%mm2 \n\t" /* 0B0B0B0B */\
- "punpcklbw %%mm7, %%mm5 \n\t" /* 0R0R0R0R */\
-\
- "psrlw $3, %%mm2 \n\t"\
- "psllw $2, %%mm3 \n\t"\
- "psllw $7, %%mm5 \n\t"\
-\
- "pand g15Mask, %%mm3 \n\t"\
- "pand r15Mask, %%mm5 \n\t"\
-\
- "por %%mm3, %%mm2 \n\t"\
- "por %%mm5, %%mm2 \n\t"\
-\
- "punpckhbw %%mm7, %%mm4 \n\t" /* 0G0G0G0G */\
- "punpckhbw %%mm7, %%mm1 \n\t" /* 0B0B0B0B */\
- "punpckhbw %%mm7, %%mm6 \n\t" /* 0R0R0R0R */\
-\
- "psrlw $3, %%mm1 \n\t"\
- "psllw $2, %%mm4 \n\t"\
- "psllw $7, %%mm6 \n\t"\
-\
- "pand g15Mask, %%mm4 \n\t"\
- "pand r15Mask, %%mm6 \n\t"\
-\
- "por %%mm4, %%mm1 \n\t"\
- "por %%mm6, %%mm1 \n\t"\
-\
- MOVNTQ(%%mm2, (%4, %%eax, 2))\
- MOVNTQ(%%mm1, 8(%4, %%eax, 2))\
-\
- "addl $8, %%eax \n\t"\
- "cmpl %5, %%eax \n\t"\
- " jb 1b \n\t"
-// FIXME find a faster way to shuffle it to BGR24
-#define WRITEBGR24 \
- /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */\
- "movq %%mm2, %%mm1 \n\t" /* B */\
- "movq %%mm5, %%mm6 \n\t" /* R */\
- "punpcklbw %%mm4, %%mm2 \n\t" /* GBGBGBGB 0 */\
- "punpcklbw %%mm7, %%mm5 \n\t" /* 0R0R0R0R 0 */\
- "punpckhbw %%mm4, %%mm1 \n\t" /* GBGBGBGB 2 */\
- "punpckhbw %%mm7, %%mm6 \n\t" /* 0R0R0R0R 2 */\
- "movq %%mm2, %%mm0 \n\t" /* GBGBGBGB 0 */\
- "movq %%mm1, %%mm3 \n\t" /* GBGBGBGB 2 */\
- "punpcklwd %%mm5, %%mm0 \n\t" /* 0RGB0RGB 0 */\
- "punpckhwd %%mm5, %%mm2 \n\t" /* 0RGB0RGB 1 */\
- "punpcklwd %%mm6, %%mm1 \n\t" /* 0RGB0RGB 2 */\
- "punpckhwd %%mm6, %%mm3 \n\t" /* 0RGB0RGB 3 */\
-\
- "movq %%mm0, %%mm4 \n\t" /* 0RGB0RGB 0 */\
- "psrlq $8, %%mm0 \n\t" /* 00RGB0RG 0 */\
- "pand bm00000111, %%mm4 \n\t" /* 00000RGB 0 */\
- "pand bm11111000, %%mm0 \n\t" /* 00RGB000 0.5 */\
- "por %%mm4, %%mm0 \n\t" /* 00RGBRGB 0 */\
- "movq %%mm2, %%mm4 \n\t" /* 0RGB0RGB 1 */\
- "psllq $48, %%mm2 \n\t" /* GB000000 1 */\
- "por %%mm2, %%mm0 \n\t" /* GBRGBRGB 0 */\
-\
- "movq %%mm4, %%mm2 \n\t" /* 0RGB0RGB 1 */\
- "psrld $16, %%mm4 \n\t" /* 000R000R 1 */\
- "psrlq $24, %%mm2 \n\t" /* 0000RGB0 1.5 */\
- "por %%mm4, %%mm2 \n\t" /* 000RRGBR 1 */\
- "pand bm00001111, %%mm2 \n\t" /* 0000RGBR 1 */\
- "movq %%mm1, %%mm4 \n\t" /* 0RGB0RGB 2 */\
- "psrlq $8, %%mm1 \n\t" /* 00RGB0RG 2 */\
- "pand bm00000111, %%mm4 \n\t" /* 00000RGB 2 */\
- "pand bm11111000, %%mm1 \n\t" /* 00RGB000 2.5 */\
- "por %%mm4, %%mm1 \n\t" /* 00RGBRGB 2 */\
- "movq %%mm1, %%mm4 \n\t" /* 00RGBRGB 2 */\
- "psllq $32, %%mm1 \n\t" /* BRGB0000 2 */\
- "por %%mm1, %%mm2 \n\t" /* BRGBRGBR 1 */\
-\
- "psrlq $32, %%mm4 \n\t" /* 000000RG 2.5 */\
- "movq %%mm3, %%mm5 \n\t" /* 0RGB0RGB 3 */\
- "psrlq $8, %%mm3 \n\t" /* 00RGB0RG 3 */\
- "pand bm00000111, %%mm5 \n\t" /* 00000RGB 3 */\
- "pand bm11111000, %%mm3 \n\t" /* 00RGB000 3.5 */\
- "por %%mm5, %%mm3 \n\t" /* 00RGBRGB 3 */\
- "psllq $16, %%mm3 \n\t" /* RGBRGB00 3 */\
- "por %%mm4, %%mm3 \n\t" /* RGBRGBRG 2.5 */\
-\
- "leal (%%eax, %%eax, 2), %%ebx \n\t"\
- MOVNTQ(%%mm0, (%4, %%ebx))\
- MOVNTQ(%%mm2, 8(%4, %%ebx))\
- MOVNTQ(%%mm3, 16(%4, %%ebx))\
-\
- "addl $8, %%eax \n\t"\
- "cmpl %5, %%eax \n\t"\
- " jb 1b \n\t"
-
-
-static inline void yuv2yuv(uint16_t *buf0, uint16_t *buf1, uint16_t *uvbuf0, uint16_t *uvbuf1,
- uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstw, int yalpha, int uvalpha)
-{
- int yalpha1=yalpha^4095;
- int uvalpha1=uvalpha^4095;
- int i;
+//global srcFilter
+SwsFilter src_filter= {NULL, NULL, NULL, NULL};
+
+float sws_lum_gblur= 0.0;
+float sws_chr_gblur= 0.0;
+int sws_chr_vshift= 0;
+int sws_chr_hshift= 0;
+float sws_chr_sharpen= 0.0;
+float sws_lum_sharpen= 0.0;
+
+/* cpuCaps combined from cpudetect and whats actually compiled in
+ (if there is no support for something compiled in it wont appear here) */
+static CpuCaps cpuCaps;
- asm volatile ("\n\t"::: "memory");
+void (*swScale)(SwsContext *context, uint8_t* src[], int srcStride[], int srcSliceY,
+ int srcSliceH, uint8_t* dst[], int dstStride[])=NULL;
- for(i=0;i<dstw;i++)
+static SwsVector *getConvVec(SwsVector *a, SwsVector *b);
+
+#ifdef CAN_COMPILE_X86_ASM
+void in_asm_used_var_warning_killer()
+{
+ volatile int i= yCoeff+vrCoeff+ubCoeff+vgCoeff+ugCoeff+bF8+bFC+w400+w80+w10+
+ bm00001111+bm00000111+bm11111000+b16Mask+g16Mask+r16Mask+b15Mask+g15Mask+r15Mask+asm_yalpha1+ asm_uvalpha1+
+ M24A+M24B+M24C+w02 + b5Dither+g5Dither+r5Dither+g6Dither+dither4[0]+dither8[0];
+ if(i) i=0;
+}
+#endif
+
+static inline void yuv2yuvXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
+ int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
+ uint8_t *dest, uint8_t *uDest, uint8_t *vDest, int dstW)
+{
+ //FIXME Optimize (just quickly writen not opti..)
+ int i;
+ for(i=0; i<dstW; i++)
{
- ((uint8_t*)dest)[0] = (buf0[i]*yalpha1+buf1[i]*yalpha)>>19;
- dest++;
+ int val=0;
+ int j;
+ for(j=0; j<lumFilterSize; j++)
+ val += lumSrc[j][i] * lumFilter[j];
+
+ dest[i]= MIN(MAX(val>>19, 0), 255);
}
- if(uvalpha != -1)
- {
- for(i=0; i<dstw/2; i++)
+ if(uDest != NULL)
+ for(i=0; i<(dstW>>1); i++)
{
- ((uint8_t*)uDest)[0] = (uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>19;
- ((uint8_t*)vDest)[0] = (uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19;
- uDest++;
- vDest++;
+ int u=0;
+ int v=0;
+ int j;
+ for(j=0; j<chrFilterSize; j++)
+ {
+ u += chrSrc[j][i] * chrFilter[j];
+ v += chrSrc[j][i + 2048] * chrFilter[j];
+ }
+
+ uDest[i]= MIN(MAX(u>>19, 0), 255);
+ vDest[i]= MIN(MAX(v>>19, 0), 255);
}
- }
}
-/**
- * vertical scale YV12 to RGB
- */
-static inline void yuv2rgbX(uint16_t *buf0, uint16_t *buf1, uint16_t *uvbuf0, uint16_t *uvbuf1,
- uint8_t *dest, int dstw, int yalpha, int uvalpha, int dstbpp)
+static inline void yuv2rgbXinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
+ int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
+ uint8_t *dest, int dstW, int dstFormat)
{
- int yalpha1=yalpha^4095;
- int uvalpha1=uvalpha^4095;
- int i;
-
- if(fullUVIpol)
+ if(dstFormat==IMGFMT_BGR32)
+ {
+ int i;
+ for(i=0; i<(dstW>>1); i++){
+ int j;
+ int Y1=0;
+ int Y2=0;
+ int U=0;
+ int V=0;
+ int Cb, Cr, Cg;
+ for(j=0; j<lumFilterSize; j++)
+ {
+ Y1 += lumSrc[j][2*i] * lumFilter[j];
+ Y2 += lumSrc[j][2*i+1] * lumFilter[j];
+ }
+ for(j=0; j<chrFilterSize; j++)
+ {
+ U += chrSrc[j][i] * chrFilter[j];
+ V += chrSrc[j][i+2048] * chrFilter[j];
+ }
+ Y1= clip_yuvtab_2568[ (Y1>>19) + 256 ];
+ Y2= clip_yuvtab_2568[ (Y2>>19) + 256 ];
+ U >>= 19;
+ V >>= 19;
+
+ Cb= clip_yuvtab_40cf[U+ 256];
+ Cg= clip_yuvtab_1a1e[V+ 256] + yuvtab_0c92[U+ 256];
+ Cr= clip_yuvtab_3343[V+ 256];
+
+ dest[8*i+0]=clip_table[((Y1 + Cb) >>13)];
+ dest[8*i+1]=clip_table[((Y1 + Cg) >>13)];
+ dest[8*i+2]=clip_table[((Y1 + Cr) >>13)];
+
+ dest[8*i+4]=clip_table[((Y2 + Cb) >>13)];
+ dest[8*i+5]=clip_table[((Y2 + Cg) >>13)];
+ dest[8*i+6]=clip_table[((Y2 + Cr) >>13)];
+ }
+ }
+ else if(dstFormat==IMGFMT_BGR24)
+ {
+ int i;
+ for(i=0; i<(dstW>>1); i++){
+ int j;
+ int Y1=0;
+ int Y2=0;
+ int U=0;
+ int V=0;
+ int Cb, Cr, Cg;
+ for(j=0; j<lumFilterSize; j++)
+ {
+ Y1 += lumSrc[j][2*i] * lumFilter[j];
+ Y2 += lumSrc[j][2*i+1] * lumFilter[j];
+ }
+ for(j=0; j<chrFilterSize; j++)
+ {
+ U += chrSrc[j][i] * chrFilter[j];
+ V += chrSrc[j][i+2048] * chrFilter[j];
+ }
+ Y1= clip_yuvtab_2568[ (Y1>>19) + 256 ];
+ Y2= clip_yuvtab_2568[ (Y2>>19) + 256 ];
+ U >>= 19;
+ V >>= 19;
+
+ Cb= clip_yuvtab_40cf[U+ 256];
+ Cg= clip_yuvtab_1a1e[V+ 256] + yuvtab_0c92[U+ 256];
+ Cr= clip_yuvtab_3343[V+ 256];
+
+ dest[0]=clip_table[((Y1 + Cb) >>13)];
+ dest[1]=clip_table[((Y1 + Cg) >>13)];
+ dest[2]=clip_table[((Y1 + Cr) >>13)];
+
+ dest[3]=clip_table[((Y2 + Cb) >>13)];
+ dest[4]=clip_table[((Y2 + Cg) >>13)];
+ dest[5]=clip_table[((Y2 + Cr) >>13)];
+ dest+=6;
+ }
+ }
+ else if(dstFormat==IMGFMT_BGR16)
{
+ int i;
+#ifdef DITHER1XBPP
+ static int ditherb1=1<<14;
+ static int ditherg1=1<<13;
+ static int ditherr1=2<<14;
+ static int ditherb2=3<<14;
+ static int ditherg2=3<<13;
+ static int ditherr2=0<<14;
+
+ ditherb1 ^= (1^2)<<14;
+ ditherg1 ^= (1^2)<<13;
+ ditherr1 ^= (1^2)<<14;
+ ditherb2 ^= (3^0)<<14;
+ ditherg2 ^= (3^0)<<13;
+ ditherr2 ^= (3^0)<<14;
+#else
+ const int ditherb1=0;
+ const int ditherg1=0;
+ const int ditherr1=0;
+ const int ditherb2=0;
+ const int ditherg2=0;
+ const int ditherr2=0;
+#endif
+ for(i=0; i<(dstW>>1); i++){
+ int j;
+ int Y1=0;
+ int Y2=0;
+ int U=0;
+ int V=0;
+ int Cb, Cr, Cg;
+ for(j=0; j<lumFilterSize; j++)
+ {
+ Y1 += lumSrc[j][2*i] * lumFilter[j];
+ Y2 += lumSrc[j][2*i+1] * lumFilter[j];
+ }
+ for(j=0; j<chrFilterSize; j++)
+ {
+ U += chrSrc[j][i] * chrFilter[j];
+ V += chrSrc[j][i+2048] * chrFilter[j];
+ }
+ Y1= clip_yuvtab_2568[ (Y1>>19) + 256 ];
+ Y2= clip_yuvtab_2568[ (Y2>>19) + 256 ];
+ U >>= 19;
+ V >>= 19;
+
+ Cb= clip_yuvtab_40cf[U+ 256];
+ Cg= clip_yuvtab_1a1e[V+ 256] + yuvtab_0c92[U+ 256];
+ Cr= clip_yuvtab_3343[V+ 256];
+
+ ((uint16_t*)dest)[2*i] =
+ clip_table16b[(Y1 + Cb + ditherb1) >>13] |
+ clip_table16g[(Y1 + Cg + ditherg1) >>13] |
+ clip_table16r[(Y1 + Cr + ditherr1) >>13];
+
+ ((uint16_t*)dest)[2*i+1] =
+ clip_table16b[(Y2 + Cb + ditherb2) >>13] |
+ clip_table16g[(Y2 + Cg + ditherg2) >>13] |
+ clip_table16r[(Y2 + Cr + ditherr2) >>13];
+ }
+ }
+ else if(dstFormat==IMGFMT_BGR15)
+ {
+ int i;
+#ifdef DITHER1XBPP
+ static int ditherb1=1<<14;
+ static int ditherg1=1<<14;
+ static int ditherr1=2<<14;
+ static int ditherb2=3<<14;
+ static int ditherg2=3<<14;
+ static int ditherr2=0<<14;
+
+ ditherb1 ^= (1^2)<<14;
+ ditherg1 ^= (1^2)<<14;
+ ditherr1 ^= (1^2)<<14;
+ ditherb2 ^= (3^0)<<14;
+ ditherg2 ^= (3^0)<<14;
+ ditherr2 ^= (3^0)<<14;
+#else
+ const int ditherb1=0;
+ const int ditherg1=0;
+ const int ditherr1=0;
+ const int ditherb2=0;
+ const int ditherg2=0;
+ const int ditherr2=0;
+#endif
+ for(i=0; i<(dstW>>1); i++){
+ int j;
+ int Y1=0;
+ int Y2=0;
+ int U=0;
+ int V=0;
+ int Cb, Cr, Cg;
+ for(j=0; j<lumFilterSize; j++)
+ {
+ Y1 += lumSrc[j][2*i] * lumFilter[j];
+ Y2 += lumSrc[j][2*i+1] * lumFilter[j];
+ }
+ for(j=0; j<chrFilterSize; j++)
+ {
+ U += chrSrc[j][i] * chrFilter[j];
+ V += chrSrc[j][i+2048] * chrFilter[j];
+ }
+ Y1= clip_yuvtab_2568[ (Y1>>19) + 256 ];
+ Y2= clip_yuvtab_2568[ (Y2>>19) + 256 ];
+ U >>= 19;
+ V >>= 19;
+
+ Cb= clip_yuvtab_40cf[U+ 256];
+ Cg= clip_yuvtab_1a1e[V+ 256] + yuvtab_0c92[U+ 256];
+ Cr= clip_yuvtab_3343[V+ 256];
+
+ ((uint16_t*)dest)[2*i] =
+ clip_table15b[(Y1 + Cb + ditherb1) >>13] |
+ clip_table15g[(Y1 + Cg + ditherg1) >>13] |
+ clip_table15r[(Y1 + Cr + ditherr1) >>13];
+
+ ((uint16_t*)dest)[2*i+1] =
+ clip_table15b[(Y2 + Cb + ditherb2) >>13] |
+ clip_table15g[(Y2 + Cg + ditherg2) >>13] |
+ clip_table15r[(Y2 + Cr + ditherr2) >>13];
+ }
+ }
+}
-#ifdef HAVE_MMX
- if(dstbpp == 32)
- {
- asm volatile(
+//Note: we have C, X86, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one
+//Plain C versions
+#if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT)
+#define COMPILE_C
+#endif
-FULL_YSCALEYUV2RGB
- "punpcklbw %%mm1, %%mm3 \n\t" // BGBGBGBG
- "punpcklbw %%mm7, %%mm0 \n\t" // R0R0R0R0
+#ifdef CAN_COMPILE_X86_ASM
- "movq %%mm3, %%mm1 \n\t"
- "punpcklwd %%mm0, %%mm3 \n\t" // BGR0BGR0
- "punpckhwd %%mm0, %%mm1 \n\t" // BGR0BGR0
+#if (defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
+#define COMPILE_MMX
+#endif
- MOVNTQ(%%mm3, (%4, %%eax, 4))
- MOVNTQ(%%mm1, 8(%4, %%eax, 4))
+#if defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT)
+#define COMPILE_MMX2
+#endif
- "addl $4, %%eax \n\t"
- "cmpl %5, %%eax \n\t"
- " jb 1b \n\t"
+#if (defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
+#define COMPILE_3DNOW
+#endif
+#endif //CAN_COMPILE_X86_ASM
+
+#undef HAVE_MMX
+#undef HAVE_MMX2
+#undef HAVE_3DNOW
+
+#ifdef COMPILE_C
+#undef HAVE_MMX
+#undef HAVE_MMX2
+#undef HAVE_3DNOW
+#define RENAME(a) a ## _C
+#include "swscale_template.c"
+#endif
+#ifdef CAN_COMPILE_X86_ASM
- :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
- "m" (yalpha1), "m" (uvalpha1)
- : "%eax"
- );
- }
- else if(dstbpp==24)
- {
- asm volatile(
+//X86 versions
+/*
+#undef RENAME
+#undef HAVE_MMX
+#undef HAVE_MMX2
+#undef HAVE_3DNOW
+#define ARCH_X86
+#define RENAME(a) a ## _X86
+#include "swscale_template.c"
+*/
+//MMX versions
+#ifdef COMPILE_MMX
+#undef RENAME
+#define HAVE_MMX
+#undef HAVE_MMX2
+#undef HAVE_3DNOW
+#define RENAME(a) a ## _MMX
+#include "swscale_template.c"
+#endif
-FULL_YSCALEYUV2RGB
+//MMX2 versions
+#ifdef COMPILE_MMX2
+#undef RENAME
+#define HAVE_MMX
+#define HAVE_MMX2
+#undef HAVE_3DNOW
+#define RENAME(a) a ## _MMX2
+#include "swscale_template.c"
+#endif
- // lsb ... msb
- "punpcklbw %%mm1, %%mm3 \n\t" // BGBGBGBG
- "punpcklbw %%mm7, %%mm0 \n\t" // R0R0R0R0
+//3DNOW versions
+#ifdef COMPILE_3DNOW
+#undef RENAME
+#define HAVE_MMX
+#undef HAVE_MMX2
+#define HAVE_3DNOW
+#define RENAME(a) a ## _3DNow
+#include "swscale_template.c"
+#endif
- "movq %%mm3, %%mm1 \n\t"
- "punpcklwd %%mm0, %%mm3 \n\t" // BGR0BGR0
- "punpckhwd %%mm0, %%mm1 \n\t" // BGR0BGR0
+#endif //CAN_COMPILE_X86_ASM
- "movq %%mm3, %%mm2 \n\t" // BGR0BGR0
- "psrlq $8, %%mm3 \n\t" // GR0BGR00
- "pand bm00000111, %%mm2 \n\t" // BGR00000
- "pand bm11111000, %%mm3 \n\t" // 000BGR00
- "por %%mm2, %%mm3 \n\t" // BGRBGR00
- "movq %%mm1, %%mm2 \n\t"
- "psllq $48, %%mm1 \n\t" // 000000BG
- "por %%mm1, %%mm3 \n\t" // BGRBGRBG
+// minor note: the HAVE_xyz is messed up after that line so dont use it
- "movq %%mm2, %%mm1 \n\t" // BGR0BGR0
- "psrld $16, %%mm2 \n\t" // R000R000
- "psrlq $24, %%mm1 \n\t" // 0BGR0000
- "por %%mm2, %%mm1 \n\t" // RBGRR000
- "movl %4, %%ebx \n\t"
- "addl %%eax, %%ebx \n\t"
+// old global scaler, dont use for new code, unless it uses only the stuff from the command line
+// will use sws_flags from the command line
+void SwScale_YV12slice(unsigned char* src[], int srcStride[], int srcSliceY ,
+ int srcSliceH, uint8_t* dst[], int dstStride, int dstbpp,
+ int srcW, int srcH, int dstW, int dstH){
-#ifdef HAVE_MMX2
- //FIXME Alignment
- "movntq %%mm3, (%%ebx, %%eax, 2)\n\t"
- "movntq %%mm1, 8(%%ebx, %%eax, 2)\n\t"
-#else
- "movd %%mm3, (%%ebx, %%eax, 2) \n\t"
- "psrlq $32, %%mm3 \n\t"
- "movd %%mm3, 4(%%ebx, %%eax, 2) \n\t"
- "movd %%mm1, 8(%%ebx, %%eax, 2) \n\t"
-#endif
- "addl $4, %%eax \n\t"
- "cmpl %5, %%eax \n\t"
- " jb 1b \n\t"
-
- :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "m" (dest), "m" (dstw),
- "m" (yalpha1), "m" (uvalpha1)
- : "%eax", "%ebx"
- );
- }
- else if(dstbpp==15)
- {
- asm volatile(
+ static SwsContext *context=NULL;
+ int dstFormat;
+ int flags=0;
+ static int firstTime=1;
+ int dstStride3[3]= {dstStride, dstStride>>1, dstStride>>1};
-FULL_YSCALEYUV2RGB
-#ifdef DITHER1XBPP
- "paddusb b16Dither, %%mm1 \n\t"
- "paddusb b16Dither, %%mm0 \n\t"
- "paddusb b16Dither, %%mm3 \n\t"
+ if(firstTime)
+ {
+#ifdef ARCH_X86
+ if(gCpuCaps.hasMMX)
+ asm volatile("emms\n\t"::: "memory"); //FIXME this shouldnt be required but it IS (even for non mmx versions)
#endif
- "punpcklbw %%mm7, %%mm1 \n\t" // 0G0G0G0G
- "punpcklbw %%mm7, %%mm3 \n\t" // 0B0B0B0B
- "punpcklbw %%mm7, %%mm0 \n\t" // 0R0R0R0R
+ flags= SWS_PRINT_INFO;
+ firstTime=0;
- "psrlw $3, %%mm3 \n\t"
- "psllw $2, %%mm1 \n\t"
- "psllw $7, %%mm0 \n\t"
- "pand g15Mask, %%mm1 \n\t"
- "pand r15Mask, %%mm0 \n\t"
+ if(src_filter.lumH) freeVec(src_filter.lumH);
+ if(src_filter.lumV) freeVec(src_filter.lumV);
+ if(src_filter.chrH) freeVec(src_filter.chrH);
+ if(src_filter.chrV) freeVec(src_filter.chrV);
- "por %%mm3, %%mm1 \n\t"
- "por %%mm1, %%mm0 \n\t"
+ if(sws_lum_gblur!=0.0){
+ src_filter.lumH= getGaussianVec(sws_lum_gblur, 3.0);
+ src_filter.lumV= getGaussianVec(sws_lum_gblur, 3.0);
+ }else{
+ src_filter.lumH= getIdentityVec();
+ src_filter.lumV= getIdentityVec();
+ }
- MOVNTQ(%%mm0, (%4, %%eax, 2))
+ if(sws_chr_gblur!=0.0){
+ src_filter.chrH= getGaussianVec(sws_chr_gblur, 3.0);
+ src_filter.chrV= getGaussianVec(sws_chr_gblur, 3.0);
+ }else{
+ src_filter.chrH= getIdentityVec();
+ src_filter.chrV= getIdentityVec();
+ }
- "addl $4, %%eax \n\t"
- "cmpl %5, %%eax \n\t"
- " jb 1b \n\t"
+ if(sws_chr_sharpen!=0.0){
+ SwsVector *g= getConstVec(-1.0, 3);
+ SwsVector *id= getConstVec(10.0/sws_chr_sharpen, 1);
+ g->coeff[1]=2.0;
+ addVec(id, g);
+ convVec(src_filter.chrH, id);
+ convVec(src_filter.chrV, id);
+ freeVec(g);
+ freeVec(id);
+ }
- :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
- "m" (yalpha1), "m" (uvalpha1)
- : "%eax"
- );
+ if(sws_lum_sharpen!=0.0){
+ SwsVector *g= getConstVec(-1.0, 3);
+ SwsVector *id= getConstVec(10.0/sws_lum_sharpen, 1);
+ g->coeff[1]=2.0;
+ addVec(id, g);
+ convVec(src_filter.lumH, id);
+ convVec(src_filter.lumV, id);
+ freeVec(g);
+ freeVec(id);
}
- else if(dstbpp==16)
- {
- asm volatile(
-FULL_YSCALEYUV2RGB
-#ifdef DITHER1XBPP
- "paddusb g16Dither, %%mm1 \n\t"
- "paddusb b16Dither, %%mm0 \n\t"
- "paddusb b16Dither, %%mm3 \n\t"
-#endif
- "punpcklbw %%mm7, %%mm1 \n\t" // 0G0G0G0G
- "punpcklbw %%mm7, %%mm3 \n\t" // 0B0B0B0B
- "punpcklbw %%mm7, %%mm0 \n\t" // 0R0R0R0R
+ if(sws_chr_hshift)
+ shiftVec(src_filter.chrH, sws_chr_hshift);
+
+ if(sws_chr_vshift)
+ shiftVec(src_filter.chrV, sws_chr_vshift);
+
+ normalizeVec(src_filter.chrH, 1.0);
+ normalizeVec(src_filter.chrV, 1.0);
+ normalizeVec(src_filter.lumH, 1.0);
+ normalizeVec(src_filter.lumV, 1.0);
- "psrlw $3, %%mm3 \n\t"
- "psllw $3, %%mm1 \n\t"
- "psllw $8, %%mm0 \n\t"
- "pand g16Mask, %%mm1 \n\t"
- "pand r16Mask, %%mm0 \n\t"
+ if(verbose > 1) printVec(src_filter.chrH);
+ if(verbose > 1) printVec(src_filter.lumH);
+ }
- "por %%mm3, %%mm1 \n\t"
- "por %%mm1, %%mm0 \n\t"
+ switch(dstbpp)
+ {
+ case 8 : dstFormat= IMGFMT_Y8; break;
+ case 12: dstFormat= IMGFMT_YV12; break;
+ case 15: dstFormat= IMGFMT_BGR15; break;
+ case 16: dstFormat= IMGFMT_BGR16; break;
+ case 24: dstFormat= IMGFMT_BGR24; break;
+ case 32: dstFormat= IMGFMT_BGR32; break;
+ default: return;
+ }
- MOVNTQ(%%mm0, (%4, %%eax, 2))
+ switch(sws_flags)
+ {
+ case 0: flags|= SWS_FAST_BILINEAR; break;
+ case 1: flags|= SWS_BILINEAR; break;
+ case 2: flags|= SWS_BICUBIC; break;
+ case 3: flags|= SWS_X; break;
+ case 4: flags|= SWS_POINT; break;
+ default:flags|= SWS_BILINEAR; break;
+ }
- "addl $4, %%eax \n\t"
- "cmpl %5, %%eax \n\t"
- " jb 1b \n\t"
+ if(!context) context=getSwsContext(srcW, srcH, IMGFMT_YV12, dstW, dstH, dstFormat, flags, &src_filter, NULL);
- :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
- "m" (yalpha1), "m" (uvalpha1)
- : "%eax"
- );
- }
-#else
- asm volatile ("\n\t"::: "memory");
- if(dstbpp==32 || dstbpp==24)
+ swScale(context, src, srcStride, srcSliceY, srcSliceH, dst, dstStride3);
+}
+
+static inline void initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
+ int srcW, int dstW, int filterAlign, int one, int flags,
+ SwsVector *srcFilter, SwsVector *dstFilter)
+{
+ int i;
+ int filterSize;
+ int filter2Size;
+ int minFilterSize;
+ double *filter=NULL;
+ double *filter2=NULL;
+#ifdef ARCH_X86
+ if(gCpuCaps.hasMMX)
+ asm volatile("emms\n\t"::: "memory"); //FIXME this shouldnt be required but it IS (even for non mmx versions)
+#endif
+
+ *filterPos = (int16_t*)memalign(8, dstW*sizeof(int16_t));
+ if(ABS(xInc - 0x10000) <10) // unscaled
+ {
+ int i;
+ filterSize= 1;
+ filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
+ for(i=0; i<dstW*filterSize; i++) filter[i]=0;
+
+ for(i=0; i<dstW; i++)
{
- for(i=0;i<dstw;i++){
- // vertical linear interpolation && yuv2rgb in a single step:
- int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
- int U=((uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>19);
- int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19);
- dest[0]=clip_table[((Y + yuvtab_40cf[U]) >>13)];
- dest[1]=clip_table[((Y + yuvtab_1a1e[V] + yuvtab_0c92[U]) >>13)];
- dest[2]=clip_table[((Y + yuvtab_3343[V]) >>13)];
- dest+=dstbpp>>3;
- }
+ filter[i*filterSize]=1;
+ (*filterPos)[i]=i;
}
- else if(dstbpp==16)
+
+ }
+ else if(flags&SWS_POINT) // lame looking point sampling mode
+ {
+ int i;
+ int xDstInSrc;
+ filterSize= 1;
+ filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
+
+ xDstInSrc= xInc/2 - 0x8000;
+ for(i=0; i<dstW; i++)
{
- for(i=0;i<dstw;i++){
- // vertical linear interpolation && yuv2rgb in a single step:
- int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
- int U=((uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>19);
- int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19);
-
- ((uint16_t*)dest)[0] =
- (clip_table[(Y + yuvtab_40cf[U]) >>13]>>3) |
- ((clip_table[(Y + yuvtab_1a1e[V] + yuvtab_0c92[U]) >>13]<<3)&0x07E0) |
- ((clip_table[(Y + yuvtab_3343[V]) >>13]<<8)&0xF800);
- dest+=2;
- }
+ int xx= (xDstInSrc>>16) - (filterSize>>1) + 1;
+
+ (*filterPos)[i]= xx;
+ filter[i]= 1.0;
+ xDstInSrc+= xInc;
}
- else if(dstbpp==15)
+ }
+ else if(xInc <= (1<<16) || (flags&SWS_FAST_BILINEAR)) // upscale
+ {
+ int i;
+ int xDstInSrc;
+ if (flags&SWS_BICUBIC) filterSize= 4;
+ else if(flags&SWS_X ) filterSize= 4;
+ else filterSize= 2;
+// printf("%d %d %d\n", filterSize, srcW, dstW);
+ filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
+
+ xDstInSrc= xInc/2 - 0x8000;
+ for(i=0; i<dstW; i++)
{
- for(i=0;i<dstw;i++){
- // vertical linear interpolation && yuv2rgb in a single step:
- int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
- int U=((uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>19);
- int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>19);
-
- ((uint16_t*)dest)[0] =
- (clip_table[(Y + yuvtab_40cf[U]) >>13]>>3) |
- ((clip_table[(Y + yuvtab_1a1e[V] + yuvtab_0c92[U]) >>13]<<2)&0x03E0) |
- ((clip_table[(Y + yuvtab_3343[V]) >>13]<<7)&0x7C00);
- dest+=2;
+ int xx= (xDstInSrc>>16) - (filterSize>>1) + 1;
+ int j;
+
+ (*filterPos)[i]= xx;
+ if((flags & SWS_BICUBIC) || (flags & SWS_X))
+ {
+ double d= ABS(((xx+1)<<16) - xDstInSrc)/(double)(1<<16);
+ double y1,y2,y3,y4;
+ double A= -0.6;
+ if(flags & SWS_BICUBIC){
+ // Equation is from VirtualDub
+ y1 = ( + A*d - 2.0*A*d*d + A*d*d*d);
+ y2 = (+ 1.0 - (A+3.0)*d*d + (A+2.0)*d*d*d);
+ y3 = ( - A*d + (2.0*A+3.0)*d*d - (A+2.0)*d*d*d);
+ y4 = ( + A*d*d - A*d*d*d);
+ }else{
+ // cubic interpolation (derived it myself)
+ y1 = ( -2.0*d + 3.0*d*d - 1.0*d*d*d)/6.0;
+ y2 = (6.0 -3.0*d - 6.0*d*d + 3.0*d*d*d)/6.0;
+ y3 = ( +6.0*d + 3.0*d*d - 3.0*d*d*d)/6.0;
+ y4 = ( -1.0*d + 1.0*d*d*d)/6.0;
+ }
+
+// printf("%d %d %d \n", coeff, (int)d, xDstInSrc);
+ filter[i*filterSize + 0]= y1;
+ filter[i*filterSize + 1]= y2;
+ filter[i*filterSize + 2]= y3;
+ filter[i*filterSize + 3]= y4;
+// printf("%1.3f %1.3f %1.3f %1.3f %1.3f\n",d , y1, y2, y3, y4);
}
+ else
+ {
+ for(j=0; j<filterSize; j++)
+ {
+ double d= ABS((xx<<16) - xDstInSrc)/(double)(1<<16);
+ double coeff= 1.0 - d;
+ if(coeff<0) coeff=0;
+ // printf("%d %d %d \n", coeff, (int)d, xDstInSrc);
+ filter[i*filterSize + j]= coeff;
+ xx++;
+ }
+ }
+ xDstInSrc+= xInc;
}
-#endif
- }//FULL_UV_IPOL
- else
+ }
+ else // downscale
{
-#ifdef HAVE_MMX
- if(dstbpp == 32)
+ int xDstInSrc;
+ if(flags&SWS_BICUBIC) filterSize= (int)ceil(1 + 4.0*srcW / (double)dstW);
+ else if(flags&SWS_X) filterSize= (int)ceil(1 + 4.0*srcW / (double)dstW);
+ else filterSize= (int)ceil(1 + 2.0*srcW / (double)dstW);
+// printf("%d %d %d\n", *filterSize, srcW, dstW);
+ filter= (double*)memalign(8, dstW*sizeof(double)*filterSize);
+
+ xDstInSrc= xInc/2 - 0x8000;
+ for(i=0; i<dstW; i++)
{
- asm volatile(
- YSCALEYUV2RGB
- WRITEBGR32
-
- :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
- "m" (yalpha1), "m" (uvalpha1)
- : "%eax"
- );
+ int xx= (int)((double)xDstInSrc/(double)(1<<16) - (filterSize-1)*0.5 + 0.5);
+ int j;
+ (*filterPos)[i]= xx;
+ for(j=0; j<filterSize; j++)
+ {
+ double d= ABS((xx<<16) - xDstInSrc)/(double)xInc;
+ double coeff;
+ if((flags & SWS_BICUBIC) || (flags & SWS_X))
+ {
+ double A= -0.75;
+// d*=2;
+ // Equation is from VirtualDub
+ if(d<1.0)
+ coeff = (1.0 - (A+3.0)*d*d + (A+2.0)*d*d*d);
+ else if(d<2.0)
+ coeff = (-4.0*A + 8.0*A*d - 5.0*A*d*d + A*d*d*d);
+ else
+ coeff=0.0;
+ }
+/* else if(flags & SWS_X)
+ {
+ }*/
+ else
+ {
+ coeff= 1.0 - d;
+ if(coeff<0) coeff=0;
+ }
+// printf("%1.3f %d %d \n", coeff, (int)d, xDstInSrc);
+ filter[i*filterSize + j]= coeff;
+ xx++;
+ }
+ xDstInSrc+= xInc;
}
- else if(dstbpp==24)
+ }
+
+ /* apply src & dst Filter to filter -> filter2
+ free(filter);
+ */
+ filter2Size= filterSize;
+ if(srcFilter) filter2Size+= srcFilter->length - 1;
+ if(dstFilter) filter2Size+= dstFilter->length - 1;
+ filter2= (double*)memalign(8, filter2Size*dstW*sizeof(double));
+
+ for(i=0; i<dstW; i++)
+ {
+ int j;
+ SwsVector scaleFilter;
+ SwsVector *outVec;
+
+ scaleFilter.coeff= filter + i*filterSize;
+ scaleFilter.length= filterSize;
+
+ if(srcFilter) outVec= getConvVec(srcFilter, &scaleFilter);
+ else outVec= &scaleFilter;
+
+ ASSERT(outVec->length == filter2Size)
+ //FIXME dstFilter
+
+ for(j=0; j<outVec->length; j++)
{
- asm volatile(
- YSCALEYUV2RGB
- WRITEBGR24
-
- :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
- "m" (yalpha1), "m" (uvalpha1)
- : "%eax", "%ebx"
- );
+ filter2[i*filter2Size + j]= outVec->coeff[j];
}
- else if(dstbpp==15)
- {
- asm volatile(
- YSCALEYUV2RGB
- /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
-#ifdef DITHER1XBPP
- "paddusb b16Dither, %%mm2 \n\t"
- "paddusb b16Dither, %%mm4 \n\t"
- "paddusb b16Dither, %%mm5 \n\t"
-#endif
- WRITEBGR15
+ (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
- :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
- "m" (yalpha1), "m" (uvalpha1)
- : "%eax"
- );
- }
- else if(dstbpp==16)
+ if(outVec != &scaleFilter) freeVec(outVec);
+ }
+ free(filter); filter=NULL;
+
+ /* try to reduce the filter-size (step1 find size and shift left) */
+ // Assume its near normalized (*0.5 or *2.0 is ok but * 0.001 is not)
+ minFilterSize= 0;
+ for(i=dstW-1; i>=0; i--)
+ {
+ int min= filter2Size;
+ int j;
+ double cutOff=0.0;
+
+ /* get rid off near zero elements on the left by shifting left */
+ for(j=0; j<filter2Size; j++)
{
- asm volatile(
- YSCALEYUV2RGB
- /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
-#ifdef DITHER1XBPP
- "paddusb g16Dither, %%mm2 \n\t"
- "paddusb b16Dither, %%mm4 \n\t"
- "paddusb b16Dither, %%mm5 \n\t"
-#endif
+ int k;
+ cutOff += ABS(filter2[i*filter2Size]);
- WRITEBGR16
+ if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
- :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
- "m" (yalpha1), "m" (uvalpha1)
- : "%eax"
- );
+ /* preserve Monotonicity because the core cant handle the filter otherwise */
+ if(i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
+
+ // Move filter coeffs left
+ for(k=1; k<filter2Size; k++)
+ filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
+ filter2[i*filter2Size + k - 1]= 0.0;
+ (*filterPos)[i]++;
}
-#else
-//FIXME unroll C loop and dont recalculate UV
- asm volatile ("\n\t"::: "memory");
- if(dstbpp==32 || dstbpp==24)
+ cutOff=0.0;
+ /* count near zeros on the right */
+ for(j=filter2Size-1; j>0; j--)
{
- for(i=0;i<dstw;i++){
- // vertical linear interpolation && yuv2rgb in a single step:
- int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
- int U=((uvbuf0[i/2]*uvalpha1+uvbuf1[i/2]*uvalpha)>>19);
- int V=((uvbuf0[i/2+2048]*uvalpha1+uvbuf1[i/2+2048]*uvalpha)>>19);
- dest[0]=clip_table[((Y + yuvtab_40cf[U]) >>13)];
- dest[1]=clip_table[((Y + yuvtab_1a1e[V] + yuvtab_0c92[U]) >>13)];
- dest[2]=clip_table[((Y + yuvtab_3343[V]) >>13)];
- dest+=dstbpp>>3;
- }
+ cutOff += ABS(filter2[i*filter2Size + j]);
+
+ if(cutOff > SWS_MAX_REDUCE_CUTOFF) break;
+ min--;
}
- else if(dstbpp==16)
+
+ if(min>minFilterSize) minFilterSize= min;
+ }
+
+ /* try to reduce the filter-size (step2 reduce it) */
+ for(i=0; i<dstW; i++)
+ {
+ int j;
+
+ for(j=0; j<minFilterSize; j++)
+ filter2[i*minFilterSize + j]= filter2[i*filter2Size + j];
+ }
+ if((flags&SWS_PRINT_INFO) && verbose)
+ printf("SwScaler: reducing filtersize %d -> %d\n", filter2Size, minFilterSize);
+ filter2Size= minFilterSize;
+ ASSERT(filter2Size > 0)
+
+ //FIXME try to align filterpos if possible
+
+ //fix borders
+ for(i=0; i<dstW; i++)
+ {
+ int j;
+ if((*filterPos)[i] < 0)
{
- for(i=0;i<dstw;i++){
- // vertical linear interpolation && yuv2rgb in a single step:
- int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
- int U=((uvbuf0[i/2]*uvalpha1+uvbuf1[i/2]*uvalpha)>>19);
- int V=((uvbuf0[i/2+2048]*uvalpha1+uvbuf1[i/2+2048]*uvalpha)>>19);
-
- ((uint16_t*)dest)[0] =
- (clip_table[(Y + yuvtab_40cf[U]) >>13]>>3) |
- ((clip_table[(Y + yuvtab_1a1e[V] + yuvtab_0c92[U]) >>13]<<3)&0x07E0) |
- ((clip_table[(Y + yuvtab_3343[V]) >>13]<<8)&0xF800);
- dest+=2;
+ // Move filter coeffs left to compensate for filterPos
+ for(j=1; j<filter2Size; j++)
+ {
+ int left= MAX(j + (*filterPos)[i], 0);
+ filter2[i*filter2Size + left] += filter2[i*filter2Size + j];
+ filter2[i*filter2Size + j]=0;
}
+ (*filterPos)[i]= 0;
}
- else if(dstbpp==15)
+
+ if((*filterPos)[i] + filter2Size > srcW)
{
- for(i=0;i<dstw;i++){
- // vertical linear interpolation && yuv2rgb in a single step:
- int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>19)];
- int U=((uvbuf0[i/2]*uvalpha1+uvbuf1[i/2]*uvalpha)>>19);
- int V=((uvbuf0[i/2+2048]*uvalpha1+uvbuf1[i/2+2048]*uvalpha)>>19);
-
- ((uint16_t*)dest)[0] =
- (clip_table[(Y + yuvtab_40cf[U]) >>13]>>3) |
- ((clip_table[(Y + yuvtab_1a1e[V] + yuvtab_0c92[U]) >>13]<<2)&0x03E0) |
- ((clip_table[(Y + yuvtab_3343[V]) >>13]<<7)&0x7C00);
- dest+=2;
+ int shift= (*filterPos)[i] + filter2Size - srcW;
+ // Move filter coeffs right to compensate for filterPos
+ for(j=filter2Size-2; j>=0; j--)
+ {
+ int right= MIN(j + shift, filter2Size-1);
+ filter2[i*filter2Size +right] += filter2[i*filter2Size +j];
+ filter2[i*filter2Size +j]=0;
}
+ (*filterPos)[i]= srcW - filter2Size;
}
-#endif
- } //!FULL_UV_IPOL
-}
+ }
-/**
- * YV12 to RGB without scaling or interpolating
- */
-static inline void yuv2rgb1(uint16_t *buf0, uint16_t *buf1, uint16_t *uvbuf0, uint16_t *uvbuf1,
- uint8_t *dest, int dstw, int yalpha, int uvalpha, int dstbpp)
-{
- int yalpha1=yalpha^4095;
- int uvalpha1=uvalpha^4095;
- int i;
- if(fullUVIpol || allwaysIpol)
+
+ *outFilterSize= (filter2Size +(filterAlign-1)) & (~(filterAlign-1));
+ *outFilter= (int16_t*)memalign(8, *outFilterSize*dstW*sizeof(int16_t));
+ memset(*outFilter, 0, *outFilterSize*dstW*sizeof(int16_t));
+
+ /* Normalize & Store in outFilter */
+ for(i=0; i<dstW; i++)
{
- yuv2rgbX(buf0, buf1, uvbuf0, uvbuf1, dest, dstw, yalpha, uvalpha, dstbpp);
- return;
- }
-#ifdef HAVE_MMX
- if(dstbpp == 32)
- {
- asm volatile(
- YSCALEYUV2RGB1
- WRITEBGR32
- :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
- "m" (yalpha1), "m" (uvalpha1)
- : "%eax"
- );
- }
- else if(dstbpp==24)
+ int j;
+ double sum=0;
+ double scale= one;
+ for(j=0; j<filter2Size; j++)
{
- asm volatile(
- YSCALEYUV2RGB1
- WRITEBGR24
- :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
- "m" (yalpha1), "m" (uvalpha1)
- : "%eax", "%ebx"
- );
+ sum+= filter2[i*filter2Size + j];
}
- else if(dstbpp==15)
+ scale/= sum;
+ for(j=0; j<filter2Size; j++)
{
- asm volatile(
- YSCALEYUV2RGB1
- /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
-#ifdef DITHER1XBPP
- "paddusb b16Dither, %%mm2 \n\t"
- "paddusb b16Dither, %%mm4 \n\t"
- "paddusb b16Dither, %%mm5 \n\t"
-#endif
- WRITEBGR15
- :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
- "m" (yalpha1), "m" (uvalpha1)
- : "%eax"
- );
+ (*outFilter)[i*(*outFilterSize) + j]= (int)(filter2[i*filter2Size + j]*scale);
}
- else if(dstbpp==16)
- {
- asm volatile(
- YSCALEYUV2RGB1
- /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
-#ifdef DITHER1XBPP
- "paddusb g16Dither, %%mm2 \n\t"
- "paddusb b16Dither, %%mm4 \n\t"
- "paddusb b16Dither, %%mm5 \n\t"
-#endif
+ }
- WRITEBGR16
- :: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
- "m" (yalpha1), "m" (uvalpha1)
- : "%eax"
- );
- }
-#else
-//FIXME unroll C loop and dont recalculate UV
- asm volatile ("\n\t"::: "memory");
+ free(filter2);
+}
- if(dstbpp==32 || dstbpp==24)
- {
- for(i=0;i<dstw;i++){
- // vertical linear interpolation && yuv2rgb in a single step:
- int Y=yuvtab_2568[buf0[i]>>7];
- int U=((uvbuf0[i/2]*uvalpha1+uvbuf1[i/2]*uvalpha)>>19);
- int V=((uvbuf0[i/2+2048]*uvalpha1+uvbuf1[i/2+2048]*uvalpha)>>19);
- dest[0]=clip_table[((Y + yuvtab_40cf[U]) >>13)];
- dest[1]=clip_table[((Y + yuvtab_1a1e[V] + yuvtab_0c92[U]) >>13)];
- dest[2]=clip_table[((Y + yuvtab_3343[V]) >>13)];
- dest+=dstbpp>>3;
- }
- }
- else if(dstbpp==16)
- {
- for(i=0;i<dstw;i++){
- // vertical linear interpolation && yuv2rgb in a single step:
- int Y=yuvtab_2568[buf0[i]>>7];
- int U=((uvbuf0[i/2]*uvalpha1+uvbuf1[i/2]*uvalpha)>>19);
- int V=((uvbuf0[i/2+2048]*uvalpha1+uvbuf1[i/2+2048]*uvalpha)>>19);
-
- ((uint16_t*)dest)[0] =
- (clip_table[(Y + yuvtab_40cf[U]) >>13]>>3) |
- ((clip_table[(Y + yuvtab_1a1e[V] + yuvtab_0c92[U]) >>13]<<3)&0x07E0) |
- ((clip_table[(Y + yuvtab_3343[V]) >>13]<<8)&0xF800);
- dest+=2;
- }
- }
- else if(dstbpp==15)
+#ifdef ARCH_X86
+static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode)
+{
+ uint8_t *fragment;
+ int imm8OfPShufW1;
+ int imm8OfPShufW2;
+ int fragmentLength;
+
+ int xpos, i;
+
+ // create an optimized horizontal scaling routine
+
+ //code fragment
+
+ asm volatile(
+ "jmp 9f \n\t"
+ // Begin
+ "0: \n\t"
+ "movq (%%esi), %%mm0 \n\t" //FIXME Alignment
+ "movq %%mm0, %%mm1 \n\t"
+ "psrlq $8, %%mm0 \n\t"
+ "punpcklbw %%mm7, %%mm1 \n\t"
+ "movq %%mm2, %%mm3 \n\t"
+ "punpcklbw %%mm7, %%mm0 \n\t"
+ "addw %%bx, %%cx \n\t" //2*xalpha += (4*lumXInc)&0xFFFF
+ "pshufw $0xFF, %%mm1, %%mm1 \n\t"
+ "1: \n\t"
+ "adcl %%edx, %%esi \n\t" //xx+= (4*lumXInc)>>16 + carry
+ "pshufw $0xFF, %%mm0, %%mm0 \n\t"
+ "2: \n\t"
+ "psrlw $9, %%mm3 \n\t"
+ "psubw %%mm1, %%mm0 \n\t"
+ "pmullw %%mm3, %%mm0 \n\t"
+ "paddw %%mm6, %%mm2 \n\t" // 2*alpha += xpos&0xFFFF
+ "psllw $7, %%mm1 \n\t"
+ "paddw %%mm1, %%mm0 \n\t"
+
+ "movq %%mm0, (%%edi, %%eax) \n\t"
+
+ "addl $8, %%eax \n\t"
+ // End
+ "9: \n\t"
+// "int $3\n\t"
+ "leal 0b, %0 \n\t"
+ "leal 1b, %1 \n\t"
+ "leal 2b, %2 \n\t"
+ "decl %1 \n\t"
+ "decl %2 \n\t"
+ "subl %0, %1 \n\t"
+ "subl %0, %2 \n\t"
+ "leal 9b, %3 \n\t"
+ "subl %0, %3 \n\t"
+ :"=r" (fragment), "=r" (imm8OfPShufW1), "=r" (imm8OfPShufW2),
+ "=r" (fragmentLength)
+ );
+
+ xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
+
+ for(i=0; i<dstW/8; i++)
+ {
+ int xx=xpos>>16;
+
+ if((i&3) == 0)
{
- for(i=0;i<dstw;i++){
- // vertical linear interpolation && yuv2rgb in a single step:
- int Y=yuvtab_2568[buf0[i]>>7];
- int U=((uvbuf0[i/2]*uvalpha1+uvbuf1[i/2]*uvalpha)>>19);
- int V=((uvbuf0[i/2+2048]*uvalpha1+uvbuf1[i/2+2048]*uvalpha)>>19);
-
- ((uint16_t*)dest)[0] =
- (clip_table[(Y + yuvtab_40cf[U]) >>13]>>3) |
- ((clip_table[(Y + yuvtab_1a1e[V] + yuvtab_0c92[U]) >>13]<<2)&0x03E0) |
- ((clip_table[(Y + yuvtab_3343[V]) >>13]<<7)&0x7C00);
- dest+=2;
- }
+ int a=0;
+ int b=((xpos+xInc)>>16) - xx;
+ int c=((xpos+xInc*2)>>16) - xx;
+ int d=((xpos+xInc*3)>>16) - xx;
+
+ memcpy(funnyCode + fragmentLength*i/4, fragment, fragmentLength);
+
+ funnyCode[fragmentLength*i/4 + imm8OfPShufW1]=
+ funnyCode[fragmentLength*i/4 + imm8OfPShufW2]=
+ a | (b<<2) | (c<<4) | (d<<6);
+
+ // if we dont need to read 8 bytes than dont :), reduces the chance of
+ // crossing a cache line
+ if(d<3) funnyCode[fragmentLength*i/4 + 1]= 0x6E;
+
+ funnyCode[fragmentLength*(i+4)/4]= RET;
}
-#endif
+ xpos+=xInc;
+ }
}
+#endif // ARCH_X86
+//FIXME remove
+void SwScale_Init(){
+}
-static inline void hyscale(uint16_t *dst, int dstWidth, uint8_t *src, int srcWidth, int xInc)
-{
- int i;
- unsigned int xpos=0;
- // *** horizontal scale Y line to temp buffer
-#ifdef ARCH_X86
+static void globalInit(){
+ // generating tables:
+ int i;
+ for(i=0; i<768; i++){
+ int c= MIN(MAX(i-256, 0), 255);
+ clip_table[i]=c;
+ yuvtab_2568[c]= clip_yuvtab_2568[i]=(0x2568*(c-16))+(256<<13);
+ yuvtab_3343[c]= clip_yuvtab_3343[i]=0x3343*(c-128);
+ yuvtab_0c92[c]= clip_yuvtab_0c92[i]=-0x0c92*(c-128);
+ yuvtab_1a1e[c]= clip_yuvtab_1a1e[i]=-0x1a1e*(c-128);
+ yuvtab_40cf[c]= clip_yuvtab_40cf[i]=0x40cf*(c-128);
+ }
-#ifdef HAVE_MMX2
- if(canMMX2BeUsed)
- {
- asm volatile(
- "pxor %%mm7, %%mm7 \n\t"
- "pxor %%mm2, %%mm2 \n\t" // 2*xalpha
- "movd %5, %%mm6 \n\t" // xInc&0xFFFF
- "punpcklwd %%mm6, %%mm6 \n\t"
- "punpcklwd %%mm6, %%mm6 \n\t"
- "movq %%mm6, %%mm2 \n\t"
- "psllq $16, %%mm2 \n\t"
- "paddw %%mm6, %%mm2 \n\t"
- "psllq $16, %%mm2 \n\t"
- "paddw %%mm6, %%mm2 \n\t"
- "psllq $16, %%mm2 \n\t" //0,t,2t,3t t=xInc&0xFF
- "movq %%mm2, temp0 \n\t"
- "movd %4, %%mm6 \n\t" //(xInc*4)&0xFFFF
- "punpcklwd %%mm6, %%mm6 \n\t"
- "punpcklwd %%mm6, %%mm6 \n\t"
- "xorl %%eax, %%eax \n\t" // i
- "movl %0, %%esi \n\t" // src
- "movl %1, %%edi \n\t" // buf1
- "movl %3, %%edx \n\t" // (xInc*4)>>16
- "xorl %%ecx, %%ecx \n\t"
- "xorl %%ebx, %%ebx \n\t"
- "movw %4, %%bx \n\t" // (xInc*4)&0xFFFF
-
-#define FUNNY_Y_CODE \
- PREFETCH" 1024(%%esi) \n\t"\
- PREFETCH" 1056(%%esi) \n\t"\
- PREFETCH" 1088(%%esi) \n\t"\
- "call funnyYCode \n\t"\
- "movq temp0, %%mm2 \n\t"\
- "xorl %%ecx, %%ecx \n\t"
-
-FUNNY_Y_CODE
-FUNNY_Y_CODE
-FUNNY_Y_CODE
-FUNNY_Y_CODE
-FUNNY_Y_CODE
-FUNNY_Y_CODE
-FUNNY_Y_CODE
-FUNNY_Y_CODE
-
- :: "m" (src), "m" (dst), "m" (dstWidth), "m" ((xInc*4)>>16),
- "m" ((xInc*4)&0xFFFF), "m" (xInc&0xFFFF)
- : "%eax", "%ebx", "%ecx", "%edx", "%esi", "%edi"
- );
- for(i=dstWidth-1; (i*xInc)>>16 >=srcWidth-1; i--) dst[i] = src[srcWidth-1]*128;
- }
+ for(i=0; i<768; i++)
+ {
+ int v= clip_table[i];
+ clip_table16b[i]= v>>3;
+ clip_table16g[i]= (v<<3)&0x07E0;
+ clip_table16r[i]= (v<<8)&0xF800;
+ clip_table15b[i]= v>>3;
+ clip_table15g[i]= (v<<2)&0x03E0;
+ clip_table15r[i]= (v<<7)&0x7C00;
+ }
+
+cpuCaps= gCpuCaps;
+
+#ifdef RUNTIME_CPUDETECT
+#ifdef CAN_COMPILE_X86_ASM
+ // ordered per speed fasterst first
+ if(gCpuCaps.hasMMX2)
+ swScale= swScale_MMX2;
+ else if(gCpuCaps.has3DNow)
+ swScale= swScale_3DNow;
+ else if(gCpuCaps.hasMMX)
+ swScale= swScale_MMX;
else
- {
+ swScale= swScale_C;
+
+#else
+ swScale= swScale_C;
+ cpuCaps.hasMMX2 = cpuCaps.hasMMX = cpuCaps.has3DNow = 0;
#endif
- //NO MMX just normal asm ...
- asm volatile(
- "xorl %%eax, %%eax \n\t" // i
- "xorl %%ebx, %%ebx \n\t" // xx
- "xorl %%ecx, %%ecx \n\t" // 2*xalpha
- "1: \n\t"
- "movzbl (%0, %%ebx), %%edi \n\t" //src[xx]
- "movzbl 1(%0, %%ebx), %%esi \n\t" //src[xx+1]
- "subl %%edi, %%esi \n\t" //src[xx+1] - src[xx]
- "imull %%ecx, %%esi \n\t" //(src[xx+1] - src[xx])*2*xalpha
- "shll $16, %%edi \n\t"
- "addl %%edi, %%esi \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
- "movl %1, %%edi \n\t"
- "shrl $9, %%esi \n\t"
- "movw %%si, (%%edi, %%eax, 2) \n\t"
- "addw %4, %%cx \n\t" //2*xalpha += xInc&0xFF
- "adcl %3, %%ebx \n\t" //xx+= xInc>>8 + carry
-
- "movzbl (%0, %%ebx), %%edi \n\t" //src[xx]
- "movzbl 1(%0, %%ebx), %%esi \n\t" //src[xx+1]
- "subl %%edi, %%esi \n\t" //src[xx+1] - src[xx]
- "imull %%ecx, %%esi \n\t" //(src[xx+1] - src[xx])*2*xalpha
- "shll $16, %%edi \n\t"
- "addl %%edi, %%esi \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
- "movl %1, %%edi \n\t"
- "shrl $9, %%esi \n\t"
- "movw %%si, 2(%%edi, %%eax, 2) \n\t"
- "addw %4, %%cx \n\t" //2*xalpha += xInc&0xFF
- "adcl %3, %%ebx \n\t" //xx+= xInc>>8 + carry
-
-
- "addl $2, %%eax \n\t"
- "cmpl %2, %%eax \n\t"
- " jb 1b \n\t"
-
-
- :: "r" (src), "m" (dst), "m" (dstWidth), "m" (xInc>>16), "m" (xInc&0xFFFF)
- : "%eax", "%ebx", "%ecx", "%edi", "%esi"
- );
+#else //RUNTIME_CPUDETECT
#ifdef HAVE_MMX2
- } //if MMX2 cant be used
-#endif
+ swScale= swScale_MMX2;
+ cpuCaps.has3DNow = 0;
+#elif defined (HAVE_3DNOW)
+ swScale= swScale_3DNow;
+ cpuCaps.hasMMX2 = 0;
+#elif defined (HAVE_MMX)
+ swScale= swScale_MMX;
+ cpuCaps.hasMMX2 = cpuCaps.has3DNow = 0;
#else
- for(i=0;i<dstWidth;i++){
- register unsigned int xx=xpos>>16;
- register unsigned int xalpha=(xpos&0xFFFF)>>9;
- dst[i]=(src[xx]*(xalpha^127)+src[xx+1]*xalpha);
- xpos+=xInc;
- }
+ swScale= swScale_C;
+ cpuCaps.hasMMX2 = cpuCaps.hasMMX = cpuCaps.has3DNow = 0;
#endif
+#endif //!RUNTIME_CPUDETECT
}
-inline static void hcscale(uint16_t *dst, int dstWidth,
- uint8_t *src1, uint8_t *src2, int srcWidth, int xInc)
-{
- int xpos=0;
+
+SwsContext *getSwsContext(int srcW, int srcH, int srcFormat, int dstW, int dstH, int dstFormat, int flags,
+ SwsFilter *srcFilter, SwsFilter *dstFilter){
+
+ const int widthAlign= dstFormat==IMGFMT_YV12 ? 16 : 8;
+ SwsContext *c;
int i;
+ SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
+
#ifdef ARCH_X86
-#ifdef HAVE_MMX2
- if(canMMX2BeUsed)
+ if(gCpuCaps.hasMMX)
+ asm volatile("emms\n\t"::: "memory");
+#endif
+
+ if(swScale==NULL) globalInit();
+
+ /* sanity check */
+ if(srcW<1 || srcH<1 || dstW<1 || dstH<1) return NULL;
+
+/* FIXME
+ if(dstStride[0]%widthAlign !=0 )
{
- asm volatile(
- "pxor %%mm7, %%mm7 \n\t"
- "pxor %%mm2, %%mm2 \n\t" // 2*xalpha
- "movd %5, %%mm6 \n\t" // xInc&0xFFFF
- "punpcklwd %%mm6, %%mm6 \n\t"
- "punpcklwd %%mm6, %%mm6 \n\t"
- "movq %%mm6, %%mm2 \n\t"
- "psllq $16, %%mm2 \n\t"
- "paddw %%mm6, %%mm2 \n\t"
- "psllq $16, %%mm2 \n\t"
- "paddw %%mm6, %%mm2 \n\t"
- "psllq $16, %%mm2 \n\t" //0,t,2t,3t t=xInc&0xFFFF
- "movq %%mm2, temp0 \n\t"
- "movd %4, %%mm6 \n\t" //(xInc*4)&0xFFFF
- "punpcklwd %%mm6, %%mm6 \n\t"
- "punpcklwd %%mm6, %%mm6 \n\t"
- "xorl %%eax, %%eax \n\t" // i
- "movl %0, %%esi \n\t" // src
- "movl %1, %%edi \n\t" // buf1
- "movl %3, %%edx \n\t" // (xInc*4)>>16
- "xorl %%ecx, %%ecx \n\t"
- "xorl %%ebx, %%ebx \n\t"
- "movw %4, %%bx \n\t" // (xInc*4)&0xFFFF
-
-#define FUNNYUVCODE \
- PREFETCH" 1024(%%esi) \n\t"\
- PREFETCH" 1056(%%esi) \n\t"\
- PREFETCH" 1088(%%esi) \n\t"\
- "call funnyUVCode \n\t"\
- "movq temp0, %%mm2 \n\t"\
- "xorl %%ecx, %%ecx \n\t"
-
-FUNNYUVCODE
-FUNNYUVCODE
-FUNNYUVCODE
-FUNNYUVCODE
-
-FUNNYUVCODE
-FUNNYUVCODE
-FUNNYUVCODE
-FUNNYUVCODE
-
-
- "xorl %%eax, %%eax \n\t" // i
- "movl %6, %%esi \n\t" // src
- "movl %1, %%edi \n\t" // buf1
- "addl $4096, %%edi \n\t"
-
-FUNNYUVCODE
-FUNNYUVCODE
-FUNNYUVCODE
-FUNNYUVCODE
-
-FUNNYUVCODE
-FUNNYUVCODE
-FUNNYUVCODE
-FUNNYUVCODE
-
- :: "m" (src1), "m" (dst), "m" (dstWidth), "m" ((xInc*4)>>16),
- "m" ((xInc*4)&0xFFFF), "m" (xInc&0xFFFF), "m" (src2)
- : "%eax", "%ebx", "%ecx", "%edx", "%esi", "%edi"
- );
- for(i=dstWidth-1; (i*xInc)>>16 >=srcWidth/2-1; i--)
+ if(flags & SWS_PRINT_INFO)
+ fprintf(stderr, "SwScaler: Warning: dstStride is not a multiple of %d!\n"
+ "SwScaler: ->cannot do aligned memory acesses anymore\n",
+ widthAlign);
+ }
+*/
+ if(!dstFilter) dstFilter= &dummyFilter;
+ if(!srcFilter) srcFilter= &dummyFilter;
+
+ c= memalign(64, sizeof(SwsContext));
+ memset(c, 0, sizeof(SwsContext));
+
+ c->srcW= srcW;
+ c->srcH= srcH;
+ c->dstW= dstW;
+ c->dstH= dstH;
+ c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
+ c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
+ c->flags= flags;
+ c->dstFormat= dstFormat;
+ c->srcFormat= srcFormat;
+
+ if(cpuCaps.hasMMX2)
+ {
+ c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
+ if(!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR))
{
- dst[i] = src1[srcWidth/2-1]*128;
- dst[i+2048] = src2[srcWidth/2-1]*128;
+ if(flags&SWS_PRINT_INFO)
+ fprintf(stderr, "SwScaler: output Width is not a multiple of 32 -> no MMX2 scaler\n");
}
}
else
+ c->canMMX2BeUsed=0;
+
+ // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
+ // but only for the FAST_BILINEAR mode otherwise do correct scaling
+ // n-2 is the last chrominance sample available
+ // this is not perfect, but noone shuld notice the difference, the more correct variant
+ // would be like the vertical one, but that would require some special code for the
+ // first and last pixel
+ if(flags&SWS_FAST_BILINEAR)
+ {
+ if(c->canMMX2BeUsed) c->lumXInc+= 20;
+ //we dont use the x86asm scaler if mmx is available
+ else if(cpuCaps.hasMMX) c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
+ }
+
+ /* set chrXInc & chrDstW */
+ if((flags&SWS_FULL_UV_IPOL) && dstFormat!=IMGFMT_YV12)
+ c->chrXInc= c->lumXInc>>1, c->chrDstW= dstW;
+ else
+ c->chrXInc= c->lumXInc, c->chrDstW= (dstW+1)>>1;
+
+ /* set chrYInc & chrDstH */
+ if(dstFormat==IMGFMT_YV12) c->chrYInc= c->lumYInc, c->chrDstH= (dstH+1)>>1;
+ else c->chrYInc= c->lumYInc>>1, c->chrDstH= dstH;
+
+ /* precalculate horizontal scaler filter coefficients */
{
+ const int filterAlign= cpuCaps.hasMMX ? 4 : 1;
+
+ initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
+ srcW , dstW, filterAlign, 1<<14, flags,
+ srcFilter->lumH, dstFilter->lumH);
+ initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
+ (srcW+1)>>1, c->chrDstW, filterAlign, 1<<14, flags,
+ srcFilter->chrH, dstFilter->chrH);
+
+#ifdef ARCH_X86
+// cant downscale !!!
+ if(c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR))
+ {
+ initMMX2HScaler( dstW, c->lumXInc, c->funnyYCode);
+ initMMX2HScaler(c->chrDstW, c->chrXInc, c->funnyUVCode);
+ }
#endif
- asm volatile(
- "xorl %%eax, %%eax \n\t" // i
- "xorl %%ebx, %%ebx \n\t" // xx
- "xorl %%ecx, %%ecx \n\t" // 2*xalpha
- "1: \n\t"
- "movl %0, %%esi \n\t"
- "movzbl (%%esi, %%ebx), %%edi \n\t" //src[xx]
- "movzbl 1(%%esi, %%ebx), %%esi \n\t" //src[xx+1]
- "subl %%edi, %%esi \n\t" //src[xx+1] - src[xx]
- "imull %%ecx, %%esi \n\t" //(src[xx+1] - src[xx])*2*xalpha
- "shll $16, %%edi \n\t"
- "addl %%edi, %%esi \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
- "movl %1, %%edi \n\t"
- "shrl $9, %%esi \n\t"
- "movw %%si, (%%edi, %%eax, 2) \n\t"
-
- "movzbl (%5, %%ebx), %%edi \n\t" //src[xx]
- "movzbl 1(%5, %%ebx), %%esi \n\t" //src[xx+1]
- "subl %%edi, %%esi \n\t" //src[xx+1] - src[xx]
- "imull %%ecx, %%esi \n\t" //(src[xx+1] - src[xx])*2*xalpha
- "shll $16, %%edi \n\t"
- "addl %%edi, %%esi \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
- "movl %1, %%edi \n\t"
- "shrl $9, %%esi \n\t"
- "movw %%si, 4096(%%edi, %%eax, 2)\n\t"
-
- "addw %4, %%cx \n\t" //2*xalpha += xInc&0xFF
- "adcl %3, %%ebx \n\t" //xx+= xInc>>8 + carry
- "addl $1, %%eax \n\t"
- "cmpl %2, %%eax \n\t"
- " jb 1b \n\t"
-
- :: "m" (src1), "m" (dst), "m" (dstWidth), "m" (xInc>>16), "m" (xInc&0xFFFF),
- "r" (src2)
- : "%eax", "%ebx", "%ecx", "%edi", "%esi"
- );
-#ifdef HAVE_MMX2
- } //if MMX2 cant be used
+ } // Init Horizontal stuff
+
+
+
+ /* precalculate vertical scaler filter coefficients */
+ initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
+ srcH , dstH, 1, (1<<12)-4, flags,
+ srcFilter->lumV, dstFilter->lumV);
+ initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
+ (srcH+1)>>1, c->chrDstH, 1, (1<<12)-4, flags,
+ srcFilter->chrV, dstFilter->chrV);
+
+ // Calculate Buffer Sizes so that they wont run out while handling these damn slices
+ c->vLumBufSize= c->vLumFilterSize;
+ c->vChrBufSize= c->vChrFilterSize;
+ for(i=0; i<dstH; i++)
+ {
+ int chrI= i*c->chrDstH / dstH;
+ int nextSlice= MAX(c->vLumFilterPos[i ] + c->vLumFilterSize - 1,
+ ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<1));
+ nextSlice&= ~1; // Slices start at even boundaries
+ if(c->vLumFilterPos[i ] + c->vLumBufSize < nextSlice)
+ c->vLumBufSize= nextSlice - c->vLumFilterPos[i ];
+ if(c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>1))
+ c->vChrBufSize= (nextSlice>>1) - c->vChrFilterPos[chrI];
+ }
+
+ // allocate pixbufs (we use dynamic allocation because otherwise we would need to
+ // allocate several megabytes to handle all possible cases)
+ c->lumPixBuf= (int16_t**)memalign(4, c->vLumBufSize*2*sizeof(int16_t*));
+ c->chrPixBuf= (int16_t**)memalign(4, c->vChrBufSize*2*sizeof(int16_t*));
+ for(i=0; i<c->vLumBufSize; i++)
+ c->lumPixBuf[i]= c->lumPixBuf[i+c->vLumBufSize]= (uint16_t*)memalign(8, 4000);
+ for(i=0; i<c->vChrBufSize; i++)
+ c->chrPixBuf[i]= c->chrPixBuf[i+c->vChrBufSize]= (uint16_t*)memalign(8, 8000);
+
+ //try to avoid drawing green stuff between the right end and the stride end
+ for(i=0; i<c->vLumBufSize; i++) memset(c->lumPixBuf[i], 0, 4000);
+ for(i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, 8000);
+
+ ASSERT(c->chrDstH <= dstH)
+
+ // pack filter data for mmx code
+ if(cpuCaps.hasMMX)
+ {
+ c->lumMmxFilter= (int16_t*)memalign(8, c->vLumFilterSize* dstH*4*sizeof(int16_t));
+ c->chrMmxFilter= (int16_t*)memalign(8, c->vChrFilterSize*c->chrDstH*4*sizeof(int16_t));
+ for(i=0; i<c->vLumFilterSize*dstH; i++)
+ c->lumMmxFilter[4*i]=c->lumMmxFilter[4*i+1]=c->lumMmxFilter[4*i+2]=c->lumMmxFilter[4*i+3]=
+ c->vLumFilter[i];
+ for(i=0; i<c->vChrFilterSize*c->chrDstH; i++)
+ c->chrMmxFilter[4*i]=c->chrMmxFilter[4*i+1]=c->chrMmxFilter[4*i+2]=c->chrMmxFilter[4*i+3]=
+ c->vChrFilter[i];
+ }
+
+ if(flags&SWS_PRINT_INFO)
+ {
+#ifdef DITHER1XBPP
+ char *dither= " dithered";
+#else
+ char *dither= "";
#endif
+ if(flags&SWS_FAST_BILINEAR)
+ fprintf(stderr, "\nSwScaler: FAST_BILINEAR scaler ");
+ else if(flags&SWS_BILINEAR)
+ fprintf(stderr, "\nSwScaler: BILINEAR scaler ");
+ else if(flags&SWS_BICUBIC)
+ fprintf(stderr, "\nSwScaler: BICUBIC scaler ");
+ else if(flags&SWS_POINT)
+ fprintf(stderr, "\nSwScaler: POINT scaler ");
+ else
+ fprintf(stderr, "\nSwScaler: ehh flags invalid?! ");
+
+ if(dstFormat==IMGFMT_BGR15)
+ fprintf(stderr, "with%s BGR15 output ", dither);
+ else if(dstFormat==IMGFMT_BGR16)
+ fprintf(stderr, "with%s BGR16 output ", dither);
+ else if(dstFormat==IMGFMT_BGR24)
+ fprintf(stderr, "with BGR24 output ");
+ else if(dstFormat==IMGFMT_BGR32)
+ fprintf(stderr, "with BGR32 output ");
+ else if(dstFormat==IMGFMT_YV12)
+ fprintf(stderr, "with YV12 output ");
+ else
+ fprintf(stderr, "without output ");
+
+ if(cpuCaps.hasMMX2)
+ fprintf(stderr, "using MMX2\n");
+ else if(cpuCaps.has3DNow)
+ fprintf(stderr, "using 3DNOW\n");
+ else if(cpuCaps.hasMMX)
+ fprintf(stderr, "using MMX\n");
+ else
+ fprintf(stderr, "using C\n");
+ }
+
+ if((flags & SWS_PRINT_INFO) && verbose)
+ {
+ if(cpuCaps.hasMMX)
+ {
+ if(c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
+ printf("SwScaler: using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
+ else
+ {
+ if(c->hLumFilterSize==4)
+ printf("SwScaler: using 4-tap MMX scaler for horizontal luminance scaling\n");
+ else if(c->hLumFilterSize==8)
+ printf("SwScaler: using 8-tap MMX scaler for horizontal luminance scaling\n");
+ else
+ printf("SwScaler: using n-tap MMX scaler for horizontal luminance scaling\n");
+
+ if(c->hChrFilterSize==4)
+ printf("SwScaler: using 4-tap MMX scaler for horizontal chrominance scaling\n");
+ else if(c->hChrFilterSize==8)
+ printf("SwScaler: using 8-tap MMX scaler for horizontal chrominance scaling\n");
+ else
+ printf("SwScaler: using n-tap MMX scaler for horizontal chrominance scaling\n");
+ }
+ }
+ else
+ {
+#ifdef ARCH_X86
+ printf("SwScaler: using X86-Asm scaler for horizontal scaling\n");
#else
- for(i=0;i<dstWidth;i++){
- register unsigned int xx=xpos>>16;
- register unsigned int xalpha=(xpos&0xFFFF)>>9;
- dst[i]=(src1[xx]*(xalpha^127)+src1[xx+1]*xalpha);
- dst[i+2048]=(src2[xx]*(xalpha^127)+src2[xx+1]*xalpha);
- xpos+=xInc;
- }
+ if(flags & SWS_FAST_BILINEAR)
+ printf("SwScaler: using FAST_BILINEAR C scaler for horizontal scaling\n");
+ else
+ printf("SwScaler: using C scaler for horizontal scaling\n");
#endif
-}
+ }
+ if(dstFormat==IMGFMT_YV12)
+ {
+ if(c->vLumFilterSize==1)
+ printf("SwScaler: using 1-tap %s \"scaler\" for vertical scaling (YV12)\n", cpuCaps.hasMMX ? "MMX" : "C");
+ else
+ printf("SwScaler: using n-tap %s scaler for vertical scaling (YV12)\n", cpuCaps.hasMMX ? "MMX" : "C");
+ }
+ else
+ {
+ if(c->vLumFilterSize==1 && c->vChrFilterSize==2)
+ printf("SwScaler: using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
+ "SwScaler: 2-tap scaler for vertical chrominance scaling (BGR)\n",cpuCaps.hasMMX ? "MMX" : "C");
+ else if(c->vLumFilterSize==2 && c->vChrFilterSize==2)
+ printf("SwScaler: using 2-tap linear %s scaler for vertical scaling (BGR)\n", cpuCaps.hasMMX ? "MMX" : "C");
+ else
+ printf("SwScaler: using n-tap %s scaler for vertical scaling (BGR)\n", cpuCaps.hasMMX ? "MMX" : "C");
+ }
+ if(dstFormat==IMGFMT_BGR24)
+ printf("SwScaler: using %s YV12->BGR24 Converter\n",
+ cpuCaps.hasMMX2 ? "MMX2" : (cpuCaps.hasMMX ? "MMX" : "C"));
+ else if(dstFormat==IMGFMT_BGR32)
+ printf("SwScaler: using %s YV12->BGR32 Converter\n", cpuCaps.hasMMX ? "MMX" : "C");
+ else if(dstFormat==IMGFMT_BGR16)
+ printf("SwScaler: using %s YV12->BGR16 Converter\n", cpuCaps.hasMMX ? "MMX" : "C");
+ else if(dstFormat==IMGFMT_BGR15)
+ printf("SwScaler: using %s YV12->BGR15 Converter\n", cpuCaps.hasMMX ? "MMX" : "C");
+
+ printf("SwScaler: %dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
+ }
-// *** bilinear scaling and yuv->rgb or yuv->yuv conversion of yv12 slices:
-// *** Note: it's called multiple times while decoding a frame, first time y==0
-// *** Designed to upscale, but may work for downscale too.
-// s_xinc = (src_width << 16) / dst_width
-// s_yinc = (src_height << 16) / dst_height
-void SwScale_YV12slice(unsigned char* srcptr[],int stride[], int y, int h,
- uint8_t* dstptr[], int dststride, int dstw, int dstbpp,
- unsigned int s_xinc,unsigned int s_yinc){
+ return c;
+}
-// scaling factors:
-//static int s_yinc=(vo_dga_src_height<<16)/vo_dga_vp_height;
-//static int s_xinc=(vo_dga_src_width<<8)/vo_dga_vp_width;
+/**
+ * returns a normalized gaussian curve used to filter stuff
+ * quality=3 is high quality, lowwer is lowwer quality
+ */
-unsigned int s_xinc2;
+SwsVector *getGaussianVec(double variance, double quality){
+ const int length= (int)(variance*quality + 0.5) | 1;
+ int i;
+ double *coeff= memalign(sizeof(double), length*sizeof(double));
+ double middle= (length-1)*0.5;
+ SwsVector *vec= malloc(sizeof(SwsVector));
-static int s_srcypos; // points to the dst Pixels center in the source (0 is the center of pixel 0,0 in src)
-static int s_ypos;
+ vec->coeff= coeff;
+ vec->length= length;
-// last horzontally interpolated lines, used to avoid unnecessary calculations
-static int s_last_ypos;
-static int s_last_y1pos;
+ for(i=0; i<length; i++)
+ {
+ double dist= i-middle;
+ coeff[i]= exp( -dist*dist/(2*variance*variance) ) / sqrt(2*variance*PI);
+ }
-static int static_dstw;
+ normalizeVec(vec, 1.0);
-#ifdef HAVE_MMX2
-// used to detect a horizontal size change
-static int old_dstw= -1;
-static int old_s_xinc= -1;
-#endif
+ return vec;
+}
-int srcWidth= (dstw*s_xinc + 0x8000)>>16;
-int dstUVw= fullUVIpol ? dstw : dstw/2;
+SwsVector *getConstVec(double c, int length){
+ int i;
+ double *coeff= memalign(sizeof(double), length*sizeof(double));
+ SwsVector *vec= malloc(sizeof(SwsVector));
+ vec->coeff= coeff;
+ vec->length= length;
-#ifdef HAVE_MMX2
-canMMX2BeUsed= (s_xinc <= 0x10000 && (dstw&31)==0 && (srcWidth&15)==0) ? 1 : 0;
-#endif
+ for(i=0; i<length; i++)
+ coeff[i]= c;
-// match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
-// n-2 is the last chrominance sample available
-// FIXME this is not perfect, but noone shuld notice the difference, the more correct variant
-// would be like the vertical one, but that would require some special code for the
-// first and last pixel
-if(canMMX2BeUsed) s_xinc+= 20;
-else s_xinc = ((srcWidth-2)<<16)/(dstw-2) - 20;
-
-if(fullUVIpol && !dstbpp==12) s_xinc2= s_xinc>>1;
-else s_xinc2= s_xinc;
- // force calculation of the horizontal interpolation of the first line
-
- if(y==0){
- s_last_ypos=-99;
- s_last_y1pos=-99;
- s_srcypos= s_yinc/2 - 0x8000;
- s_ypos=0;
-#ifdef HAVE_MMX2
-// cant downscale !!!
- if((old_s_xinc != s_xinc || old_dstw!=dstw) && canMMX2BeUsed)
- {
- uint8_t *fragment;
- int imm8OfPShufW1;
- int imm8OfPShufW2;
- int fragmentLength;
-
- int xpos, xx, xalpha, i;
-
- old_s_xinc= s_xinc;
- old_dstw= dstw;
-
- static_dstw= dstw;
-
- // create an optimized horizontal scaling routine
-
- //code fragment
-
- asm volatile(
- "jmp 9f \n\t"
- // Begin
- "0: \n\t"
- "movq (%%esi), %%mm0 \n\t" //FIXME Alignment
- "movq %%mm0, %%mm1 \n\t"
- "psrlq $8, %%mm0 \n\t"
- "punpcklbw %%mm7, %%mm1 \n\t"
- "movq %%mm2, %%mm3 \n\t"
- "punpcklbw %%mm7, %%mm0 \n\t"
- "addw %%bx, %%cx \n\t" //2*xalpha += (4*s_xinc)&0xFFFF
- "pshufw $0xFF, %%mm1, %%mm1 \n\t"
- "1: \n\t"
- "adcl %%edx, %%esi \n\t" //xx+= (4*s_xinc)>>16 + carry
- "pshufw $0xFF, %%mm0, %%mm0 \n\t"
- "2: \n\t"
- "psrlw $9, %%mm3 \n\t"
- "psubw %%mm1, %%mm0 \n\t"
- "pmullw %%mm3, %%mm0 \n\t"
- "paddw %%mm6, %%mm2 \n\t" // 2*alpha += xpos&0xFFFF
- "psllw $7, %%mm1 \n\t"
- "paddw %%mm1, %%mm0 \n\t"
-
- "movq %%mm0, (%%edi, %%eax) \n\t"
-
- "addl $8, %%eax \n\t"
- // End
- "9: \n\t"
-// "int $3\n\t"
- "leal 0b, %0 \n\t"
- "leal 1b, %1 \n\t"
- "leal 2b, %2 \n\t"
- "decl %1 \n\t"
- "decl %2 \n\t"
- "subl %0, %1 \n\t"
- "subl %0, %2 \n\t"
- "leal 9b, %3 \n\t"
- "subl %0, %3 \n\t"
- :"=r" (fragment), "=r" (imm8OfPShufW1), "=r" (imm8OfPShufW2),
- "=r" (fragmentLength)
- );
-
- xpos= 0; //s_xinc/2 - 0x8000; // difference between pixel centers
-
- /* choose xinc so that all 8 parts fit exactly
- Note: we cannot use just 1 part because it would not fit in the code cache */
-// s_xinc2_diff= -((((s_xinc2*(dstw/8))&0xFFFF))/(dstw/8))-10;
-// s_xinc_diff= -((((s_xinc*(dstw/8))&0xFFFF))/(dstw/8));
-#ifdef ALT_ERROR
-// s_xinc2_diff+= ((0x10000/(dstw/8)));
-#endif
-// s_xinc_diff= s_xinc2_diff*2;
+ return vec;
+}
-// s_xinc2+= s_xinc2_diff;
-// s_xinc+= s_xinc_diff;
-// old_s_xinc= s_xinc;
+SwsVector *getIdentityVec(void){
+ double *coeff= memalign(sizeof(double), sizeof(double));
+ SwsVector *vec= malloc(sizeof(SwsVector));
+ coeff[0]= 1.0;
- for(i=0; i<dstw/8; i++)
- {
- int xx=xpos>>16;
+ vec->coeff= coeff;
+ vec->length= 1;
- if((i&3) == 0)
- {
- int a=0;
- int b=((xpos+s_xinc)>>16) - xx;
- int c=((xpos+s_xinc*2)>>16) - xx;
- int d=((xpos+s_xinc*3)>>16) - xx;
+ return vec;
+}
- memcpy(funnyYCode + fragmentLength*i/4, fragment, fragmentLength);
+void normalizeVec(SwsVector *a, double height){
+ int i;
+ double sum=0;
+ double inv;
- funnyYCode[fragmentLength*i/4 + imm8OfPShufW1]=
- funnyYCode[fragmentLength*i/4 + imm8OfPShufW2]=
- a | (b<<2) | (c<<4) | (d<<6);
+ for(i=0; i<a->length; i++)
+ sum+= a->coeff[i];
- // if we dont need to read 8 bytes than dont :), reduces the chance of
- // crossing a cache line
- if(d<3) funnyYCode[fragmentLength*i/4 + 1]= 0x6E;
+ inv= height/sum;
- funnyYCode[fragmentLength*(i+4)/4]= RET;
- }
- xpos+=s_xinc;
- }
+ for(i=0; i<a->length; i++)
+ a->coeff[i]*= height;
+}
- xpos= 0; //s_xinc2/2 - 0x10000; // difference between centers of chrom samples
- for(i=0; i<dstUVw/8; i++)
- {
- int xx=xpos>>16;
+void scaleVec(SwsVector *a, double scalar){
+ int i;
- if((i&3) == 0)
- {
- int a=0;
- int b=((xpos+s_xinc2)>>16) - xx;
- int c=((xpos+s_xinc2*2)>>16) - xx;
- int d=((xpos+s_xinc2*3)>>16) - xx;
+ for(i=0; i<a->length; i++)
+ a->coeff[i]*= scalar;
+}
- memcpy(funnyUVCode + fragmentLength*i/4, fragment, fragmentLength);
+static SwsVector *getConvVec(SwsVector *a, SwsVector *b){
+ int length= a->length + b->length - 1;
+ double *coeff= memalign(sizeof(double), length*sizeof(double));
+ int i, j;
+ SwsVector *vec= malloc(sizeof(SwsVector));
- funnyUVCode[fragmentLength*i/4 + imm8OfPShufW1]=
- funnyUVCode[fragmentLength*i/4 + imm8OfPShufW2]=
- a | (b<<2) | (c<<4) | (d<<6);
+ vec->coeff= coeff;
+ vec->length= length;
- // if we dont need to read 8 bytes than dont :), reduces the chance of
- // crossing a cache line
- if(d<3) funnyUVCode[fragmentLength*i/4 + 1]= 0x6E;
+ for(i=0; i<length; i++) coeff[i]= 0.0;
- funnyUVCode[fragmentLength*(i+4)/4]= RET;
- }
- xpos+=s_xinc2;
+ for(i=0; i<a->length; i++)
+ {
+ for(j=0; j<b->length; j++)
+ {
+ coeff[i+j]+= a->coeff[i]*b->coeff[j];
}
-// funnyCode[0]= RET;
}
-#endif // HAVE_MMX2
- } // reset counters
-
- while(1){
- unsigned char *dest =dstptr[0]+dststride*s_ypos;
- unsigned char *uDest=dstptr[1]+(dststride>>1)*(s_ypos>>1);
- unsigned char *vDest=dstptr[2]+(dststride>>1)*(s_ypos>>1);
-
- int y0=(s_srcypos + 0xFFFF)>>16; // first luminance source line number below the dst line
- // points to the dst Pixels center in the source (0 is the center of pixel 0,0 in src)
- int srcuvpos= dstbpp==12 ? s_srcypos + s_yinc/2 - 0x8000 :
- s_srcypos - 0x8000;
- int y1=(srcuvpos + 0x1FFFF)>>17; // first chrominance source line number below the dst line
- int yalpha=((s_srcypos-1)&0xFFFF)>>4;
- int uvalpha=((srcuvpos-1)&0x1FFFF)>>5;
- uint16_t *buf0=pix_buf_y[y0&1]; // top line of the interpolated slice
- uint16_t *buf1=pix_buf_y[((y0+1)&1)]; // bottom line of the interpolated slice
- uint16_t *uvbuf0=pix_buf_uv[y1&1]; // top line of the interpolated slice
- uint16_t *uvbuf1=pix_buf_uv[(y1+1)&1]; // bottom line of the interpolated slice
- int i;
+ return vec;
+}
- if(y0>=y+h) break; // FIXME wrong, skips last lines, but they are dupliactes anyway
+static SwsVector *sumVec(SwsVector *a, SwsVector *b){
+ int length= MAX(a->length, b->length);
+ double *coeff= memalign(sizeof(double), length*sizeof(double));
+ int i;
+ SwsVector *vec= malloc(sizeof(SwsVector));
+
+ vec->coeff= coeff;
+ vec->length= length;
+
+ for(i=0; i<length; i++) coeff[i]= 0.0;
+
+ for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
+ for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
+
+ return vec;
+}
- if((y0&1) && dstbpp==12) uvalpha=-1; // there is no alpha if there is no line
+static SwsVector *diffVec(SwsVector *a, SwsVector *b){
+ int length= MAX(a->length, b->length);
+ double *coeff= memalign(sizeof(double), length*sizeof(double));
+ int i;
+ SwsVector *vec= malloc(sizeof(SwsVector));
+
+ vec->coeff= coeff;
+ vec->length= length;
+
+ for(i=0; i<length; i++) coeff[i]= 0.0;
- s_ypos++; s_srcypos+=s_yinc;
+ for(i=0; i<a->length; i++) coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
+ for(i=0; i<b->length; i++) coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
- //only interpolate the src line horizontally if we didnt do it allready
- if(s_last_ypos!=y0)
+ return vec;
+}
+
+/* shift left / or right if "shift" is negative */
+static SwsVector *getShiftedVec(SwsVector *a, int shift){
+ int length= a->length + ABS(shift)*2;
+ double *coeff= memalign(sizeof(double), length*sizeof(double));
+ int i;
+ SwsVector *vec= malloc(sizeof(SwsVector));
+
+ vec->coeff= coeff;
+ vec->length= length;
+
+ for(i=0; i<length; i++) coeff[i]= 0.0;
+
+ for(i=0; i<a->length; i++)
{
- unsigned char *src;
- // skip if first line has been horiz scaled alleady
- if(s_last_ypos != y0-1)
- {
- // check if first line is before any available src lines
- if(y0-1 < y) src=srcptr[0]+(0 )*stride[0];
- else src=srcptr[0]+(y0-y-1)*stride[0];
+ coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
+ }
- hyscale(buf0, dstw, src, srcWidth, s_xinc);
- }
- // check if second line is after any available src lines
- if(y0-y >= h) src=srcptr[0]+(h-1)*stride[0];
- else src=srcptr[0]+(y0-y)*stride[0];
+ return vec;
+}
+
+void shiftVec(SwsVector *a, int shift){
+ SwsVector *shifted= getShiftedVec(a, shift);
+ free(a->coeff);
+ a->coeff= shifted->coeff;
+ a->length= shifted->length;
+ free(shifted);
+}
+
+void addVec(SwsVector *a, SwsVector *b){
+ SwsVector *sum= sumVec(a, b);
+ free(a->coeff);
+ a->coeff= sum->coeff;
+ a->length= sum->length;
+ free(sum);
+}
+
+void subVec(SwsVector *a, SwsVector *b){
+ SwsVector *diff= diffVec(a, b);
+ free(a->coeff);
+ a->coeff= diff->coeff;
+ a->length= diff->length;
+ free(diff);
+}
+
+void convVec(SwsVector *a, SwsVector *b){
+ SwsVector *conv= getConvVec(a, b);
+ free(a->coeff);
+ a->coeff= conv->coeff;
+ a->length= conv->length;
+ free(conv);
+}
+
+SwsVector *cloneVec(SwsVector *a){
+ double *coeff= memalign(sizeof(double), a->length*sizeof(double));
+ int i;
+ SwsVector *vec= malloc(sizeof(SwsVector));
+
+ vec->coeff= coeff;
+ vec->length= a->length;
+
+ for(i=0; i<a->length; i++) coeff[i]= a->coeff[i];
+
+ return vec;
+}
+
+void printVec(SwsVector *a){
+ int i;
+ double max=0;
+ double min=0;
+ double range;
+
+ for(i=0; i<a->length; i++)
+ if(a->coeff[i]>max) max= a->coeff[i];
+
+ for(i=0; i<a->length; i++)
+ if(a->coeff[i]<min) min= a->coeff[i];
- // the min() is required to avoid reuseing lines which where not available
- s_last_ypos= MIN(y0, y+h-1);
- hyscale(buf1, dstw, src, srcWidth, s_xinc);
+ range= max - min;
+
+ for(i=0; i<a->length; i++)
+ {
+ int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
+ printf("%1.3f ", a->coeff[i]);
+ for(;x>0; x--) printf(" ");
+ printf("|\n");
}
-// printf("%d %d %d %d\n", y, y1, s_last_y1pos, h);
- // *** horizontal scale U and V lines to temp buffer
- if(s_last_y1pos!=y1)
+}
+
+void freeVec(SwsVector *a){
+ if(!a) return;
+ if(a->coeff) free(a->coeff);
+ a->coeff=NULL;
+ a->length=0;
+ free(a);
+}
+
+void freeSwsContext(SwsContext *c){
+ int i;
+
+ if(!c) return;
+
+ if(c->lumPixBuf)
{
- uint8_t *src1, *src2;
- // skip if first line has been horiz scaled alleady
- if(s_last_y1pos != y1-1)
+ for(i=0; i<c->vLumBufSize*2; i++)
{
- // check if first line is before any available src lines
- if(y1-y/2-1 < 0)
- {
- src1= srcptr[1]+(0)*stride[1];
- src2= srcptr[2]+(0)*stride[2];
- }else{
- src1= srcptr[1]+(y1-y/2-1)*stride[1];
- src2= srcptr[2]+(y1-y/2-1)*stride[2];
- }
- hcscale(uvbuf0, dstUVw, src1, src2, srcWidth, s_xinc2);
+ if(c->lumPixBuf[i]) free(c->lumPixBuf[i]);
+ c->lumPixBuf[i]=NULL;
}
+ free(c->lumPixBuf);
+ c->lumPixBuf=NULL;
+ }
- // check if second line is after any available src lines
- if(y1 - y/2 >= h/2)
+ if(c->chrPixBuf)
+ {
+ for(i=0; i<c->vChrBufSize*2; i++)
{
- src1= srcptr[1]+(h/2-1)*stride[1];
- src2= srcptr[2]+(h/2-1)*stride[2];
- }else{
- src1= srcptr[1]+(y1-y/2)*stride[1];
- src2= srcptr[2]+(y1-y/2)*stride[2];
+ if(c->chrPixBuf[i]) free(c->chrPixBuf[i]);
+ c->chrPixBuf[i]=NULL;
}
- hcscale(uvbuf1, dstUVw, src1, src2, srcWidth, s_xinc2);
-
- // the min() is required to avoid reuseing lines which where not available
- s_last_y1pos= MIN(y1, y/2+h/2-1);
+ free(c->chrPixBuf);
+ c->chrPixBuf=NULL;
}
- if(dstbpp==12) //YV12
- yuv2yuv(buf0, buf1, uvbuf0, uvbuf1, dest, uDest, vDest, dstw, yalpha, uvalpha);
- else if(ABS(s_yinc - 0x10000) < 10)
- yuv2rgb1(buf0, buf1, uvbuf0, uvbuf1, dest, dstw, yalpha, uvalpha, dstbpp);
- else
- yuv2rgbX(buf0, buf1, uvbuf0, uvbuf1, dest, dstw, yalpha, uvalpha, dstbpp);
-
-#ifdef HAVE_MMX
- b16Dither= b16Dither1;
- b16Dither1= b16Dither2;
- b16Dither2= b16Dither;
-
- g16Dither= g16Dither1;
- g16Dither1= g16Dither2;
- g16Dither2= g16Dither;
-#endif
- }
-__asm __volatile(SFENCE:::"memory");
-__asm __volatile(EMMS:::"memory");
+ if(c->vLumFilter) free(c->vLumFilter);
+ c->vLumFilter = NULL;
+ if(c->vChrFilter) free(c->vChrFilter);
+ c->vChrFilter = NULL;
+ if(c->hLumFilter) free(c->hLumFilter);
+ c->hLumFilter = NULL;
+ if(c->hChrFilter) free(c->hChrFilter);
+ c->hChrFilter = NULL;
+
+ if(c->vLumFilterPos) free(c->vLumFilterPos);
+ c->vLumFilterPos = NULL;
+ if(c->vChrFilterPos) free(c->vChrFilterPos);
+ c->vChrFilterPos = NULL;
+ if(c->hLumFilterPos) free(c->hLumFilterPos);
+ c->hLumFilterPos = NULL;
+ if(c->hChrFilterPos) free(c->hChrFilterPos);
+ c->hChrFilterPos = NULL;
+
+ if(c->lumMmxFilter) free(c->lumMmxFilter);
+ c->lumMmxFilter = NULL;
+ if(c->chrMmxFilter) free(c->chrMmxFilter);
+ c->chrMmxFilter = NULL;
+
+ free(c);
}
-void SwScale_Init(){
- // generating tables:
- int i;
- for(i=0;i<256;i++){
- clip_table[i]=0;
- clip_table[i+256]=i;
- clip_table[i+512]=255;
- yuvtab_2568[i]=(0x2568*(i-16))+(256<<13);
- yuvtab_3343[i]=0x3343*(i-128);
- yuvtab_0c92[i]=-0x0c92*(i-128);
- yuvtab_1a1e[i]=-0x1a1e*(i-128);
- yuvtab_40cf[i]=0x40cf*(i-128);
- }
-
-}
projects / ffmpeg / blobdiff