swscale_template.c

mediastreamer2是开源的网络传输媒体流的库

    ASMALIGN(4)\
    "2:                                             \n\t"\
    "movq  (%%"REG_S", %%"REG_a", 2), %%mm0         \n\t" /* Y1srcData */\
    "movq 8(%%"REG_S", %%"REG_a", 2), %%mm2         \n\t" /* Y2srcData */\
    "mov                4(%%"REG_d"), %%"REG_S"     \n\t"\
    "movq  (%%"REG_S", %%"REG_a", 2), %%mm4         \n\t" /* Y1srcData */\
    "movq                      %%mm0, %%mm3         \n\t"\
    "punpcklwd                 %%mm4, %%mm0         \n\t"\
    "punpckhwd                 %%mm4, %%mm3         \n\t"\
    "movq               8(%%"REG_d"), %%mm4         \n\t" /* filterCoeff */\
    "pmaddwd                   %%mm4, %%mm0         \n\t"\
    "pmaddwd                   %%mm4, %%mm3         \n\t"\
    "paddd                     %%mm0, %%mm1         \n\t"\
    "paddd                     %%mm3, %%mm5         \n\t"\
    "movq 8(%%"REG_S", %%"REG_a", 2), %%mm3         \n\t" /* Y2srcData */\
    "mov               16(%%"REG_d"), %%"REG_S"     \n\t"\
    "add                         $16, %%"REG_d"     \n\t"\
    "test                  %%"REG_S", %%"REG_S"     \n\t"\
    "movq                      %%mm2, %%mm0         \n\t"\
    "punpcklwd                 %%mm3, %%mm2         \n\t"\
    "punpckhwd                 %%mm3, %%mm0         \n\t"\
    "pmaddwd                   %%mm4, %%mm2         \n\t"\
    "pmaddwd                   %%mm4, %%mm0         \n\t"\
    "paddd                     %%mm2, %%mm7         \n\t"\
    "paddd                     %%mm0, %%mm6         \n\t"\
    " jnz                         2b                \n\t"\
    "psrad                       $16, %%mm1         \n\t"\
    "psrad                       $16, %%mm5         \n\t"\
    "psrad                       $16, %%mm7         \n\t"\
    "psrad                       $16, %%mm6         \n\t"\
    "movq      "VROUNDER_OFFSET"(%0), %%mm0         \n\t"\
    "packssdw                  %%mm5, %%mm1         \n\t"\
    "packssdw                  %%mm6, %%mm7         \n\t"\
    "paddw                     %%mm0, %%mm1         \n\t"\
    "paddw                     %%mm0, %%mm7         \n\t"\
    "movq               "U_TEMP"(%0), %%mm3         \n\t"\
    "movq               "V_TEMP"(%0), %%mm4         \n\t"\

#define YSCALEYUV2RGBX \
    "psubw  "U_OFFSET"(%0), %%mm3       \n\t" /* (U-128)8*/\
    "psubw  "V_OFFSET"(%0), %%mm4       \n\t" /* (V-128)8*/\
    "movq            %%mm3, %%mm2       \n\t" /* (U-128)8*/\
    "movq            %%mm4, %%mm5       \n\t" /* (V-128)8*/\
    "pmulhw "UG_COEFF"(%0), %%mm3       \n\t"\
    "pmulhw "VG_COEFF"(%0), %%mm4       \n\t"\
/* mm2=(U-128)8, mm3=ug, mm4=vg mm5=(V-128)8 */\
    "pmulhw "UB_COEFF"(%0), %%mm2       \n\t"\
    "pmulhw "VR_COEFF"(%0), %%mm5       \n\t"\
    "psubw  "Y_OFFSET"(%0), %%mm1       \n\t" /* 8(Y-16)*/\
    "psubw  "Y_OFFSET"(%0), %%mm7       \n\t" /* 8(Y-16)*/\
    "pmulhw  "Y_COEFF"(%0), %%mm1       \n\t"\
    "pmulhw  "Y_COEFF"(%0), %%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"
#if 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"\
    "xor              %%"REG_a", %%"REG_a"  \n\t"\
    ASMALIGN(4)\
    "1:                                 \n\t"\
    "movq     (%0, %%"REG_a",2), %%mm0  \n\t" /*buf0[eax]*/\
    "movq     (%1, %%"REG_a",2), %%mm1  \n\t" /*buf1[eax]*/\
    "movq     (%2, %%"REG_a",2), %%mm2  \n\t" /* uvbuf0[eax]*/\
    "movq     (%3, %%"REG_a",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, %%"REG_a",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, %%"REG_a",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        "MANGLE(w80)", %%mm1  \n\t" /* 8(Y-16)*/\
    "psubw       "MANGLE(w400)", %%mm3  \n\t" /* 8(U-128)*/\
    "pmulhw    "MANGLE(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   "MANGLE(ubCoeff)", %%mm3  \n\t"\
    "psraw                   $4, %%mm0  \n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048] >>4*/\
    "pmulhw   "MANGLE(ugCoeff)", %%mm2  \n\t"\
    "paddw                %%mm4, %%mm0  \n\t" /* uvbuf0[eax+2048]uvalpha1 - uvbuf1[eax+2048](1-uvalpha1)*/\
    "psubw       "MANGLE(w400)", %%mm0  \n\t" /* (V-128)8*/\
\
\
    "movq                 %%mm0, %%mm4  \n\t" /* (V-128)8*/\
    "pmulhw   "MANGLE(vrCoeff)", %%mm0  \n\t"\
    "pmulhw   "MANGLE(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"
#endif

#define REAL_YSCALEYUV2PACKED(index, c) \
    "movq "CHR_MMX_FILTER_OFFSET"+8("#c"), %%mm0              \n\t"\
    "movq "LUM_MMX_FILTER_OFFSET"+8("#c"), %%mm1              \n\t"\
    "psraw                $3, %%mm0                           \n\t"\
    "psraw                $3, %%mm1                           \n\t"\
    "movq              %%mm0, "CHR_MMX_FILTER_OFFSET"+8("#c") \n\t"\
    "movq              %%mm1, "LUM_MMX_FILTER_OFFSET"+8("#c") \n\t"\
    "xor            "#index", "#index"                        \n\t"\
    ASMALIGN(4)\
    "1:                                 \n\t"\
    "movq     (%2, "#index"), %%mm2     \n\t" /* uvbuf0[eax]*/\
    "movq     (%3, "#index"), %%mm3     \n\t" /* uvbuf1[eax]*/\
    "movq 4096(%2, "#index"), %%mm5     \n\t" /* uvbuf0[eax+2048]*/\
    "movq 4096(%3, "#index"), %%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 "CHR_MMX_FILTER_OFFSET"+8("#c"), %%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                $7, %%mm3     \n\t" /* uvbuf0[eax] - uvbuf1[eax] >>4*/\
    "psraw                $7, %%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)*/\
    "movq  (%0, "#index", 2), %%mm0     \n\t" /*buf0[eax]*/\
    "movq  (%1, "#index", 2), %%mm1     \n\t" /*buf1[eax]*/\
    "movq 8(%0, "#index", 2), %%mm6     \n\t" /*buf0[eax]*/\
    "movq 8(%1, "#index", 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 "LUM_MMX_FILTER_OFFSET"+8("#c"), %%mm0  \n\t" /* (buf0[eax] - buf1[eax])yalpha1>>16*/\
    "pmulhw "LUM_MMX_FILTER_OFFSET"+8("#c"), %%mm6  \n\t" /* (buf0[eax] - buf1[eax])yalpha1>>16*/\
    "psraw                $7, %%mm1     \n\t" /* buf0[eax] - buf1[eax] >>4*/\
    "psraw                $7, %%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*/\

#define YSCALEYUV2PACKED(index, c)  REAL_YSCALEYUV2PACKED(index, c)

#define REAL_YSCALEYUV2RGB(index, c) \
    "xor            "#index", "#index"  \n\t"\
    ASMALIGN(4)\
    "1:                                 \n\t"\
    "movq     (%2, "#index"), %%mm2     \n\t" /* uvbuf0[eax]*/\
    "movq     (%3, "#index"), %%mm3     \n\t" /* uvbuf1[eax]*/\
    "movq 4096(%2, "#index"), %%mm5     \n\t" /* uvbuf0[eax+2048]*/\
    "movq 4096(%3, "#index"), %%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 "CHR_MMX_FILTER_OFFSET"+8("#c"), %%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  "U_OFFSET"("#c"), %%mm3     \n\t" /* (U-128)8*/\
    "psubw  "V_OFFSET"("#c"), %%mm4     \n\t" /* (V-128)8*/\
    "movq              %%mm3, %%mm2     \n\t" /* (U-128)8*/\
    "movq              %%mm4, %%mm5     \n\t" /* (V-128)8*/\
    "pmulhw "UG_COEFF"("#c"), %%mm3     \n\t"\
    "pmulhw "VG_COEFF"("#c"), %%mm4     \n\t"\
    /* mm2=(U-128)8, mm3=ug, mm4=vg mm5=(V-128)8 */\
    "movq  (%0, "#index", 2), %%mm0     \n\t" /*buf0[eax]*/\
    "movq  (%1, "#index", 2), %%mm1     \n\t" /*buf1[eax]*/\
    "movq 8(%0, "#index", 2), %%mm6     \n\t" /*buf0[eax]*/\
    "movq 8(%1, "#index", 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 "LUM_MMX_FILTER_OFFSET"+8("#c"), %%mm0  \n\t" /* (buf0[eax] - buf1[eax])yalpha1>>16*/\
    "pmulhw "LUM_MMX_FILTER_OFFSET"+8("#c"), %%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 "UB_COEFF"("#c"), %%mm2     \n\t"\
    "pmulhw "VR_COEFF"("#c"), %%mm5     \n\t"\
    "psubw  "Y_OFFSET"("#c"), %%mm1     \n\t" /* 8(Y-16)*/\
    "psubw  "Y_OFFSET"("#c"), %%mm7     \n\t" /* 8(Y-16)*/\
    "pmulhw  "Y_COEFF"("#c"), %%mm1     \n\t"\
    "pmulhw  "Y_COEFF"("#c"), %%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 YSCALEYUV2RGB(index, c)  REAL_YSCALEYUV2RGB(index, c)

#define REAL_YSCALEYUV2PACKED1(index, c) \
    "xor            "#index", "#index"  \n\t"\
    ASMALIGN(4)\
    "1:                                 \n\t"\
    "movq     (%2, "#index"), %%mm3     \n\t" /* uvbuf0[eax]*/\
    "movq 4096(%2, "#index"), %%mm4     \n\t" /* uvbuf0[eax+2048]*/\
    "psraw                $7, %%mm3     \n\t" \
    "psraw                $7, %%mm4     \n\t" \
    "movq  (%0, "#index", 2), %%mm1     \n\t" /*buf0[eax]*/\
    "movq 8(%0, "#index", 2), %%mm7     \n\t" /*buf0[eax]*/\
    "psraw                $7, %%mm1     \n\t" \
    "psraw                $7, %%mm7     \n\t" \

#define YSCALEYUV2PACKED1(index, c)  REAL_YSCALEYUV2PACKED1(index, c)

#define REAL_YSCALEYUV2RGB1(index, c) \
    "xor            "#index", "#index"  \n\t"\
    ASMALIGN(4)\
    "1:                                 \n\t"\
    "movq     (%2, "#index"), %%mm3     \n\t" /* uvbuf0[eax]*/\
    "movq 4096(%2, "#index"), %%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  "U_OFFSET"("#c"), %%mm3     \n\t" /* (U-128)8*/\
    "psubw  "V_OFFSET"("#c"), %%mm4     \n\t" /* (V-128)8*/\
    "movq              %%mm3, %%mm2     \n\t" /* (U-128)8*/\
    "movq              %%mm4, %%mm5     \n\t" /* (V-128)8*/\
    "pmulhw "UG_COEFF"("#c"), %%mm3     \n\t"\
    "pmulhw "VG_COEFF"("#c"), %%mm4     \n\t"\
    /* mm2=(U-128)8, mm3=ug, mm4=vg mm5=(V-128)8 */\
    "movq  (%0, "#index", 2), %%mm1     \n\t" /*buf0[eax]*/\
    "movq 8(%0, "#index", 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 "UB_COEFF"("#c"), %%mm2     \n\t"\
    "pmulhw "VR_COEFF"("#c"), %%mm5     \n\t"\
    "psubw  "Y_OFFSET"("#c"), %%mm1     \n\t" /* 8(Y-16)*/\
    "psubw  "Y_OFFSET"("#c"), %%mm7     \n\t" /* 8(Y-16)*/\
    "pmulhw  "Y_COEFF"("#c"), %%mm1     \n\t"\
    "pmulhw  "Y_COEFF"("#c"), %%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"