#include <string.h>
#include <math.h>
#include <stdio.h>
+#include <unistd.h>
#include "../config.h"
#include "../mangle.h"
#include <assert.h>
#else
#include <stdlib.h>
#endif
+#ifdef HAVE_SYS_MMAN_H
+#include <sys/mman.h>
+#if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)
+#define MAP_ANONYMOUS MAP_ANON
+#endif
+#endif
#include "swscale.h"
#include "swscale_internal.h"
#include "../cpudetect.h"
//FIXME replace this with something faster
#define isPlanarYUV(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_YVU9 \
+ || (x)==IMGFMT_NV12 || (x)==IMGFMT_NV21 \
|| (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P)
#define isYUV(x) ((x)==IMGFMT_UYVY || (x)==IMGFMT_YUY2 || isPlanarYUV(x))
#define isGray(x) ((x)==IMGFMT_Y800)
#define isSupportedOut(x) ((x)==IMGFMT_YV12 || (x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY\
|| (x)==IMGFMT_444P || (x)==IMGFMT_422P || (x)==IMGFMT_411P\
|| isRGB(x) || isBGR(x)\
+ || (x)==IMGFMT_NV12 || (x)==IMGFMT_NV21\
|| (x)==IMGFMT_Y800 || (x)==IMGFMT_YVU9)
#define isPacked(x) ((x)==IMGFMT_YUY2 || (x)==IMGFMT_UYVY ||isRGB(x) || isBGR(x))
#define MIN(a,b) ((a) > (b) ? (b) : (a))
#define MAX(a,b) ((a) < (b) ? (b) : (a))
-#ifdef ARCH_X86
-static uint64_t __attribute__((aligned(8))) bF8= 0xF8F8F8F8F8F8F8F8LL;
-static uint64_t __attribute__((aligned(8))) bFC= 0xFCFCFCFCFCFCFCFCLL;
+#if defined(ARCH_X86) || defined(ARCH_X86_64)
+static uint64_t attribute_used __attribute__((aligned(8))) bF8= 0xF8F8F8F8F8F8F8F8LL;
+static uint64_t attribute_used __attribute__((aligned(8))) bFC= 0xFCFCFCFCFCFCFCFCLL;
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))) bm01010101=0x00FF00FF00FF00FFLL;
+static uint64_t attribute_used __attribute__((aligned(8))) w02= 0x0002000200020002LL;
+static uint64_t attribute_used __attribute__((aligned(8))) bm00001111=0x00000000FFFFFFFFLL;
+static uint64_t attribute_used __attribute__((aligned(8))) bm00000111=0x0000000000FFFFFFLL;
+static uint64_t attribute_used __attribute__((aligned(8))) bm11111000=0xFFFFFFFFFF000000LL;
+static uint64_t attribute_used __attribute__((aligned(8))) bm01010101=0x00FF00FF00FF00FFLL;
-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 volatile uint64_t attribute_used __attribute__((aligned(8))) b5Dither;
+static volatile uint64_t attribute_used __attribute__((aligned(8))) g5Dither;
+static volatile uint64_t attribute_used __attribute__((aligned(8))) g6Dither;
+static volatile uint64_t attribute_used __attribute__((aligned(8))) r5Dither;
static uint64_t __attribute__((aligned(8))) dither4[2]={
0x0103010301030103LL,
0x0004000400040004LL,};
static uint64_t __attribute__((aligned(8))) b16Mask= 0x001F001F001F001FLL;
-static uint64_t __attribute__((aligned(8))) g16Mask= 0x07E007E007E007E0LL;
-static uint64_t __attribute__((aligned(8))) r16Mask= 0xF800F800F800F800LL;
+static uint64_t attribute_used __attribute__((aligned(8))) g16Mask= 0x07E007E007E007E0LL;
+static uint64_t attribute_used __attribute__((aligned(8))) r16Mask= 0xF800F800F800F800LL;
static uint64_t __attribute__((aligned(8))) b15Mask= 0x001F001F001F001FLL;
-static uint64_t __attribute__((aligned(8))) g15Mask= 0x03E003E003E003E0LL;
-static uint64_t __attribute__((aligned(8))) r15Mask= 0x7C007C007C007C00LL;
+static uint64_t attribute_used __attribute__((aligned(8))) g15Mask= 0x03E003E003E003E0LL;
+static uint64_t attribute_used __attribute__((aligned(8))) r15Mask= 0x7C007C007C007C00LL;
-static uint64_t __attribute__((aligned(8))) M24A= 0x00FF0000FF0000FFLL;
-static uint64_t __attribute__((aligned(8))) M24B= 0xFF0000FF0000FF00LL;
-static uint64_t __attribute__((aligned(8))) M24C= 0x0000FF0000FF0000LL;
+static uint64_t attribute_used __attribute__((aligned(8))) M24A= 0x00FF0000FF0000FFLL;
+static uint64_t attribute_used __attribute__((aligned(8))) M24B= 0xFF0000FF0000FF00LL;
+static uint64_t attribute_used __attribute__((aligned(8))) M24C= 0x0000FF0000FF0000LL;
#ifdef FAST_BGR2YV12
-static const uint64_t bgr2YCoeff __attribute__((aligned(8))) = 0x000000210041000DULL;
-static const uint64_t bgr2UCoeff __attribute__((aligned(8))) = 0x0000FFEEFFDC0038ULL;
-static const uint64_t bgr2VCoeff __attribute__((aligned(8))) = 0x00000038FFD2FFF8ULL;
+static const uint64_t bgr2YCoeff attribute_used __attribute__((aligned(8))) = 0x000000210041000DULL;
+static const uint64_t bgr2UCoeff attribute_used __attribute__((aligned(8))) = 0x0000FFEEFFDC0038ULL;
+static const uint64_t bgr2VCoeff attribute_used __attribute__((aligned(8))) = 0x00000038FFD2FFF8ULL;
#else
-static const uint64_t bgr2YCoeff __attribute__((aligned(8))) = 0x000020E540830C8BULL;
-static const uint64_t bgr2UCoeff __attribute__((aligned(8))) = 0x0000ED0FDAC23831ULL;
-static const uint64_t bgr2VCoeff __attribute__((aligned(8))) = 0x00003831D0E6F6EAULL;
+static const uint64_t bgr2YCoeff attribute_used __attribute__((aligned(8))) = 0x000020E540830C8BULL;
+static const uint64_t bgr2UCoeff attribute_used __attribute__((aligned(8))) = 0x0000ED0FDAC23831ULL;
+static const uint64_t bgr2VCoeff attribute_used __attribute__((aligned(8))) = 0x00003831D0E6F6EAULL;
#endif
-static const uint64_t bgr2YOffset __attribute__((aligned(8))) = 0x1010101010101010ULL;
-static const uint64_t bgr2UVOffset __attribute__((aligned(8)))= 0x8080808080808080ULL;
-static const uint64_t w1111 __attribute__((aligned(8))) = 0x0001000100010001ULL;
+static const uint64_t bgr2YOffset attribute_used __attribute__((aligned(8))) = 0x1010101010101010ULL;
+static const uint64_t bgr2UVOffset attribute_used __attribute__((aligned(8)))= 0x8080808080808080ULL;
+static const uint64_t w1111 attribute_used __attribute__((aligned(8))) = 0x0001000100010001ULL;
#endif
// clipping helper table for C implementations:
extern const uint8_t dither_8x8_73[8][8];
extern const uint8_t dither_8x8_220[8][8];
-#ifdef ARCH_X86
+#if defined(ARCH_X86) || defined(ARCH_X86_64)
void in_asm_used_var_warning_killer()
{
volatile int i= bF8+bFC+w10+
}
}
+static inline void yuv2nv12XinC(int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize,
+ int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize,
+ uint8_t *dest, uint8_t *uDest, int dstW, int chrDstW, int dstFormat)
+{
+ //FIXME Optimize (just quickly writen not opti..)
+ int i;
+ for(i=0; i<dstW; i++)
+ {
+ int val=1<<18;
+ int j;
+ for(j=0; j<lumFilterSize; j++)
+ val += lumSrc[j][i] * lumFilter[j];
+
+ dest[i]= MIN(MAX(val>>19, 0), 255);
+ }
+
+ if(uDest == NULL)
+ return;
+
+ if(dstFormat == IMGFMT_NV12)
+ for(i=0; i<chrDstW; i++)
+ {
+ int u=1<<18;
+ int v=1<<18;
+ int j;
+ for(j=0; j<chrFilterSize; j++)
+ {
+ u += chrSrc[j][i] * chrFilter[j];
+ v += chrSrc[j][i + 2048] * chrFilter[j];
+ }
+
+ uDest[2*i]= MIN(MAX(u>>19, 0), 255);
+ uDest[2*i+1]= MIN(MAX(v>>19, 0), 255);
+ }
+ else
+ for(i=0; i<chrDstW; i++)
+ {
+ int u=1<<18;
+ int v=1<<18;
+ int j;
+ for(j=0; j<chrFilterSize; j++)
+ {
+ u += chrSrc[j][i] * chrFilter[j];
+ v += chrSrc[j][i + 2048] * chrFilter[j];
+ }
+
+ uDest[2*i]= MIN(MAX(v>>19, 0), 255);
+ uDest[2*i+1]= MIN(MAX(u>>19, 0), 255);
+ }
+}
#define YSCALE_YUV_2_PACKEDX_C(type) \
for(i=0; i<(dstW>>1); i++){\
((uint8_t*)dest)[3]= r[Y2];\
((uint8_t*)dest)[4]= g[Y2];\
((uint8_t*)dest)[5]= b[Y2];\
- ((uint8_t*)dest)+=6;\
+ dest+=6;\
}\
break;\
case IMGFMT_BGR24:\
((uint8_t*)dest)[3]= b[Y2];\
((uint8_t*)dest)[4]= g[Y2];\
((uint8_t*)dest)[5]= r[Y2];\
- ((uint8_t*)dest)+=6;\
+ dest+=6;\
}\
break;\
case IMGFMT_RGB16:\
acc+= acc + g[((buf0[i+6]*yalpha1+buf1[i+6]*yalpha)>>19) + d128[6]];\
acc+= acc + g[((buf0[i+7]*yalpha1+buf1[i+7]*yalpha)>>19) + d128[7]];\
((uint8_t*)dest)[0]= acc;\
- ((uint8_t*)dest)++;\
+ dest++;\
}\
\
/*\
((uint8_t*)dest)[3]= r[Y2];
((uint8_t*)dest)[4]= g[Y2];
((uint8_t*)dest)[5]= b[Y2];
- ((uint8_t*)dest)+=6;
+ dest+=6;
}
break;
case IMGFMT_BGR24:
((uint8_t*)dest)[3]= b[Y2];
((uint8_t*)dest)[4]= g[Y2];
((uint8_t*)dest)[5]= r[Y2];
- ((uint8_t*)dest)+=6;
+ dest+=6;
}
break;
case IMGFMT_RGB16:
acc+= acc + g[Y2+d128[(i+1)&7]];
if((i&7)==6){
((uint8_t*)dest)[0]= acc;
- ((uint8_t*)dest)++;
+ dest++;
}
}
}
#endif //HAVE_ALTIVEC
#endif //ARCH_POWERPC
-#ifdef ARCH_X86
+#if defined(ARCH_X86) || defined(ARCH_X86_64)
#if (defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
#define COMPILE_MMX
#if (defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
#define COMPILE_3DNOW
#endif
-#endif //ARCH_X86
+#endif //ARCH_X86 || ARCH_X86_64
#undef HAVE_MMX
#undef HAVE_MMX2
#endif
#endif //ARCH_POWERPC
-#ifdef ARCH_X86
+#if defined(ARCH_X86) || defined(ARCH_X86_64)
//X86 versions
/*
#include "swscale_template.c"
#endif
-#endif //ARCH_X86
+#endif //ARCH_X86 || ARCH_X86_64
// minor note: the HAVE_xyz is messed up after that line so don't use it
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)
+ SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
{
int i;
int filterSize;
int minFilterSize;
double *filter=NULL;
double *filter2=NULL;
-#ifdef ARCH_X86
+#if defined(ARCH_X86) || defined(ARCH_X86_64)
if(flags & SWS_CPU_CAPS_MMX)
asm volatile("emms\n\t"::: "memory"); //FIXME this shouldnt be required but it IS (even for non mmx versions)
#endif
double xDstInSrc;
double sizeFactor, filterSizeInSrc;
const double xInc1= (double)xInc / (double)(1<<16);
- int param= (flags&SWS_PARAM_MASK)>>SWS_PARAM_SHIFT;
if (flags&SWS_BICUBIC) sizeFactor= 4.0;
else if(flags&SWS_X) sizeFactor= 8.0;
else if(flags&SWS_AREA) sizeFactor= 1.0; //downscale only, for upscale it is bilinear
else if(flags&SWS_GAUSS) sizeFactor= 8.0; // infinite ;)
- else if(flags&SWS_LANCZOS) sizeFactor= param ? 2.0*param : 6.0;
+ else if(flags&SWS_LANCZOS) sizeFactor= param[0] != SWS_PARAM_DEFAULT ? 2.0*param[0] : 6.0;
else if(flags&SWS_SINC) sizeFactor= 20.0; // infinite ;)
else if(flags&SWS_SPLINE) sizeFactor= 20.0; // infinite ;)
else if(flags&SWS_BILINEAR) sizeFactor= 2.0;
double coeff;
if(flags & SWS_BICUBIC)
{
- double A= param ? -param*0.01 : -0.60;
-
- // Equation is from VirtualDub
- if(d<1.0)
- coeff = (1.0 - (A+3.0)*d*d + (A+2.0)*d*d*d);
+ double B= param[0] != SWS_PARAM_DEFAULT ? param[0] : 0.0;
+ double C= param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6;
+
+ if(d<1.0)
+ coeff = (12-9*B-6*C)*d*d*d + (-18+12*B+6*C)*d*d + 6-2*B;
else if(d<2.0)
- coeff = (-4.0*A + 8.0*A*d - 5.0*A*d*d + A*d*d*d);
+ coeff = (-B-6*C)*d*d*d + (6*B+30*C)*d*d + (-12*B-48*C)*d +8*B+24*C;
else
coeff=0.0;
}
}*/
else if(flags & SWS_X)
{
- double A= param ? param*0.1 : 1.0;
+ double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
if(d<1.0)
coeff = cos(d*PI);
}
else if(flags & SWS_GAUSS)
{
- double p= param ? param*0.1 : 3.0;
+ double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
coeff = pow(2.0, - p*d*d);
}
else if(flags & SWS_SINC)
}
else if(flags & SWS_LANCZOS)
{
- double p= param ? param : 3.0;
+ double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
coeff = d ? sin(d*PI)*sin(d*PI/p)/(d*d*PI*PI/p) : 1.0;
if(d>p) coeff=0;
}
if(min>minFilterSize) minFilterSize= min;
}
+ if (flags & SWS_CPU_CAPS_ALTIVEC) {
+ // we can handle the special case 4,
+ // so we don't want to go to the full 8
+ if (minFilterSize < 5)
+ filterAlign = 4;
+
+ // we really don't want to waste our time
+ // doing useless computation, so fall-back on
+ // the scalar C code for very small filter.
+ // vectorizing is worth it only if you have
+ // decent-sized vector.
+ if (minFilterSize < 3)
+ filterAlign = 1;
+ }
+
ASSERT(minFilterSize > 0)
filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
ASSERT(filterSize > 0)
free(filter);
}
-#ifdef ARCH_X86
+#if defined(ARCH_X86) || defined(ARCH_X86_64)
static void initMMX2HScaler(int dstW, int xInc, uint8_t *funnyCode, int16_t *filter, int32_t *filterPos, int numSplits)
{
uint8_t *fragmentA;
- int imm8OfPShufW1A;
- int imm8OfPShufW2A;
- int fragmentLengthA;
+ long imm8OfPShufW1A;
+ long imm8OfPShufW2A;
+ long fragmentLengthA;
uint8_t *fragmentB;
- int imm8OfPShufW1B;
- int imm8OfPShufW2B;
- int fragmentLengthB;
+ long imm8OfPShufW1B;
+ long imm8OfPShufW2B;
+ long fragmentLengthB;
int fragmentPos;
int xpos, i;
"jmp 9f \n\t"
// Begin
"0: \n\t"
- "movq (%%edx, %%eax), %%mm3 \n\t"
- "movd (%%ecx, %%esi), %%mm0 \n\t"
- "movd 1(%%ecx, %%esi), %%mm1 \n\t"
+ "movq (%%"REG_d", %%"REG_a"), %%mm3\n\t"
+ "movd (%%"REG_c", %%"REG_S"), %%mm0\n\t"
+ "movd 1(%%"REG_c", %%"REG_S"), %%mm1\n\t"
"punpcklbw %%mm7, %%mm1 \n\t"
"punpcklbw %%mm7, %%mm0 \n\t"
"pshufw $0xFF, %%mm1, %%mm1 \n\t"
"pshufw $0xFF, %%mm0, %%mm0 \n\t"
"2: \n\t"
"psubw %%mm1, %%mm0 \n\t"
- "movl 8(%%ebx, %%eax), %%esi \n\t"
+ "movl 8(%%"REG_b", %%"REG_a"), %%esi\n\t"
"pmullw %%mm3, %%mm0 \n\t"
"psllw $7, %%mm1 \n\t"
"paddw %%mm1, %%mm0 \n\t"
- "movq %%mm0, (%%edi, %%eax) \n\t"
+ "movq %%mm0, (%%"REG_D", %%"REG_a")\n\t"
- "addl $8, %%eax \n\t"
+ "add $8, %%"REG_a" \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"
+ "lea 0b, %0 \n\t"
+ "lea 1b, %1 \n\t"
+ "lea 2b, %2 \n\t"
+ "dec %1 \n\t"
+ "dec %2 \n\t"
+ "sub %0, %1 \n\t"
+ "sub %0, %2 \n\t"
+ "lea 9b, %3 \n\t"
+ "sub %0, %3 \n\t"
:"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
"jmp 9f \n\t"
// Begin
"0: \n\t"
- "movq (%%edx, %%eax), %%mm3 \n\t"
- "movd (%%ecx, %%esi), %%mm0 \n\t"
+ "movq (%%"REG_d", %%"REG_a"), %%mm3\n\t"
+ "movd (%%"REG_c", %%"REG_S"), %%mm0\n\t"
"punpcklbw %%mm7, %%mm0 \n\t"
"pshufw $0xFF, %%mm0, %%mm1 \n\t"
"1: \n\t"
"pshufw $0xFF, %%mm0, %%mm0 \n\t"
"2: \n\t"
"psubw %%mm1, %%mm0 \n\t"
- "movl 8(%%ebx, %%eax), %%esi \n\t"
+ "movl 8(%%"REG_b", %%"REG_a"), %%esi\n\t"
"pmullw %%mm3, %%mm0 \n\t"
"psllw $7, %%mm1 \n\t"
"paddw %%mm1, %%mm0 \n\t"
- "movq %%mm0, (%%edi, %%eax) \n\t"
+ "movq %%mm0, (%%"REG_D", %%"REG_a")\n\t"
- "addl $8, %%eax \n\t"
+ "add $8, %%"REG_a" \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"
+ "lea 0b, %0 \n\t"
+ "lea 1b, %1 \n\t"
+ "lea 2b, %2 \n\t"
+ "dec %1 \n\t"
+ "dec %2 \n\t"
+ "sub %0, %1 \n\t"
+ "sub %0, %2 \n\t"
+ "lea 9b, %3 \n\t"
+ "sub %0, %3 \n\t"
:"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
}
filterPos[i/2]= xpos>>16; // needed to jump to the next part
}
-#endif // ARCH_X86
+#endif // ARCH_X86 || ARCH_X86_64
static void globalInit(){
// generating tables:
static SwsFunc getSwsFunc(int flags){
#ifdef RUNTIME_CPUDETECT
-#ifdef ARCH_X86
+#if defined(ARCH_X86) || defined(ARCH_X86_64)
// ordered per speed fasterst first
if(flags & SWS_CPU_CAPS_MMX2)
return swScale_MMX2;
uint8_t *dstPtr= dst;
for(i=0; i<srcSliceH; i++)
{
- memcpy(dstPtr, srcPtr, srcStride[0]);
+ memcpy(dstPtr, srcPtr, c->srcW);
srcPtr+= srcStride[0];
dstPtr+= dstStride[0];
}
}
- dst = dstParam[1] + dstStride[1]*srcSliceY;
- interleaveBytes( src[1],src[2],dst,c->srcW,srcSliceH,srcStride[1],srcStride[2],dstStride[0] );
+ dst = dstParam[1] + dstStride[1]*srcSliceY/2;
+ if (c->dstFormat == IMGFMT_NV12)
+ interleaveBytes( src[1],src[2],dst,c->srcW/2,srcSliceH/2,srcStride[1],srcStride[2],dstStride[0] );
+ else
+ interleaveBytes( src[2],src[1],dst,c->srcW/2,srcSliceH/2,srcStride[2],srcStride[1],dstStride[0] );
return srcSliceH;
}
sortedStride[0]= stride[0];
sortedStride[1]= stride[1];
sortedStride[2]= stride[2];
+ }
+ else if(format == IMGFMT_NV12 || format == IMGFMT_NV21)
+ {
+ sortedP[0]= p[0];
+ sortedP[1]= p[1];
+ sortedP[2]= NULL;
+ sortedStride[0]= stride[0];
+ sortedStride[1]= stride[1];
+ sortedStride[2]= 0;
}else{
MSG_ERR("internal error in orderYUV\n");
}
break;
case IMGFMT_YV12:
case IMGFMT_Y800: //FIXME remove after different subsamplings are fully implemented
+ case IMGFMT_NV12:
+ case IMGFMT_NV21:
*h=1;
*v=1;
break;
yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
//FIXME factorize
-
+
+#ifdef HAVE_ALTIVEC
+ yuv2rgb_altivec_init_tables (c, inv_table, brightness, contrast, saturation);
+#endif
return 0;
}
}
SwsContext *sws_getContext(int srcW, int srcH, int origSrcFormat, int dstW, int dstH, int origDstFormat, int flags,
- SwsFilter *srcFilter, SwsFilter *dstFilter){
+ SwsFilter *srcFilter, SwsFilter *dstFilter, double *param){
SwsContext *c;
int i;
int unscaled, needsDither;
int srcFormat, dstFormat;
SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
-#ifdef ARCH_X86
+#if defined(ARCH_X86) || defined(ARCH_X86_64)
if(flags & SWS_CPU_CAPS_MMX)
asm volatile("emms\n\t"::: "memory");
#endif
if((isBGR(srcFormat) || isRGB(srcFormat)) && !(flags&SWS_FULL_CHR_H_INP))
c->chrSrcHSubSample=1;
+ if(param){
+ c->param[0] = param[0];
+ c->param[1] = param[1];
+ }else{
+ c->param[0] =
+ c->param[1] = SWS_PARAM_DEFAULT;
+ }
+
c->chrIntHSubSample= c->chrDstHSubSample;
c->chrIntVSubSample= c->chrSrcVSubSample;
if(unscaled && !usesHFilter && !usesVFilter)
{
/* yv12_to_nv12 */
- if(srcFormat == IMGFMT_YV12 && dstFormat == IMGFMT_NV12)
+ if(srcFormat == IMGFMT_YV12 && (dstFormat == IMGFMT_NV12 || dstFormat == IMGFMT_NV21))
{
c->swScale= PlanarToNV12Wrapper;
}
}
}
+#ifdef HAVE_ALTIVEC
+ if ((c->flags & SWS_CPU_CAPS_ALTIVEC) &&
+ ((srcFormat == IMGFMT_YV12 &&
+ (dstFormat == IMGFMT_YUY2 || dstFormat == IMGFMT_UYVY)))) {
+ // unscaled YV12 -> packed YUV, we want speed
+ if (dstFormat == IMGFMT_YUY2)
+ c->swScale= yv12toyuy2_unscaled_altivec;
+ else
+ c->swScale= yv12touyvy_unscaled_altivec;
+ }
+#endif
+
/* simple copy */
if( srcFormat == dstFormat
|| (isPlanarYUV(srcFormat) && isGray(dstFormat))
/* precalculate horizontal scaler filter coefficients */
{
- const int filterAlign= (flags & SWS_CPU_CAPS_MMX) ? 4 : 1;
+ const int filterAlign=
+ (flags & SWS_CPU_CAPS_MMX) ? 4 :
+ (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
+ 1;
initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
srcW , dstW, filterAlign, 1<<14,
(flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
- srcFilter->lumH, dstFilter->lumH);
+ srcFilter->lumH, dstFilter->lumH, c->param);
initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
(flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
- srcFilter->chrH, dstFilter->chrH);
+ srcFilter->chrH, dstFilter->chrH, c->param);
-#ifdef ARCH_X86
+#if defined(ARCH_X86) || defined(ARCH_X86_64)
// can't downscale !!!
if(c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR))
{
+#define MAX_FUNNY_CODE_SIZE 10000
+#ifdef MAP_ANONYMOUS
+ c->funnyYCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
+ c->funnyUVCode = (uint8_t*)mmap(NULL, MAX_FUNNY_CODE_SIZE, PROT_EXEC | PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
+#else
+ c->funnyYCode = (uint8_t*)memalign(32, MAX_FUNNY_CODE_SIZE);
+ c->funnyUVCode = (uint8_t*)memalign(32, MAX_FUNNY_CODE_SIZE);
+#endif
+
c->lumMmx2Filter = (int16_t*)memalign(8, (dstW /8+8)*sizeof(int16_t));
c->chrMmx2Filter = (int16_t*)memalign(8, (c->chrDstW /4+8)*sizeof(int16_t));
c->lumMmx2FilterPos= (int32_t*)memalign(8, (dstW /2/8+8)*sizeof(int32_t));
/* precalculate vertical scaler filter coefficients */
- initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
- srcH , dstH, 1, (1<<12)-4,
- (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
- srcFilter->lumV, dstFilter->lumV);
- initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
- c->chrSrcH, c->chrDstH, 1, (1<<12)-4,
- (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
- srcFilter->chrV, dstFilter->chrV);
+ {
+ const int filterAlign=
+ (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
+ 1;
+
+ initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
+ srcH , dstH, filterAlign, (1<<12)-4,
+ (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags,
+ srcFilter->lumV, dstFilter->lumV, c->param);
+ initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
+ c->chrSrcH, c->chrDstH, filterAlign, (1<<12)-4,
+ (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
+ srcFilter->chrV, dstFilter->chrV, c->param);
+ }
// Calculate Buffer Sizes so that they won't run out while handling these damn slices
c->vLumBufSize= c->vLumFilterSize;
}
else
{
-#ifdef ARCH_X86
+#if defined(ARCH_X86) || defined(ARCH_X86_64)
MSG_V("SwScaler: using X86-Asm scaler for horizontal scaling\n");
#else
if(flags & SWS_FAST_BILINEAR)
if(c->hChrFilterPos) free(c->hChrFilterPos);
c->hChrFilterPos = NULL;
+#if defined(ARCH_X86) || defined(ARCH_X86_64)
+#ifdef MAP_ANONYMOUS
+ if(c->funnyYCode) munmap(c->funnyYCode, MAX_FUNNY_CODE_SIZE);
+ if(c->funnyUVCode) munmap(c->funnyUVCode, MAX_FUNNY_CODE_SIZE);
+#else
+ if(c->funnyYCode) free(c->funnyYCode);
+ if(c->funnyUVCode) free(c->funnyUVCode);
+#endif
+ c->funnyYCode=NULL;
+ c->funnyUVCode=NULL;
+#endif
+
if(c->lumMmx2Filter) free(c->lumMmx2Filter);
c->lumMmx2Filter=NULL;
if(c->chrMmx2Filter) free(c->chrMmx2Filter);
projects / ffmpeg / blobdiff