dsputil_mmx.c

Trolltech公司发布的图形界面操作系统。可在qt-embedded-2.3.10平台上编译为嵌入式图形界面操作系统。

            "paddw %2, %%mm1		\n\t"\
            "psraw $5, %%mm0		\n\t"\
            "psraw $5, %%mm1		\n\t"\
            "packuswb %%mm1, %%mm0	\n\t"\
            OP_3DNOW(%%mm0, 8(%1), %%mm1, q)\
            :: "r"(temp), "r"(dst), "m"(ROUNDER)\
            : "memory"\
        );\
        dst+=dstStride;\
        src+=srcStride;\
    }\
}\
\
static void OPNAME ## mpeg4_qpel8_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
    uint64_t temp;\
\
    asm volatile(\
        "pxor %%mm7, %%mm7		\n\t"\
        "1:				\n\t"\
        "movq  (%0), %%mm0		\n\t" /* ABCDEFGH */\
        "movq %%mm0, %%mm1		\n\t" /* ABCDEFGH */\
        "movq %%mm0, %%mm2		\n\t" /* ABCDEFGH */\
        "punpcklbw %%mm7, %%mm0		\n\t" /* 0A0B0C0D */\
        "punpckhbw %%mm7, %%mm1		\n\t" /* 0E0F0G0H */\
        "pshufw $0x90, %%mm0, %%mm5	\n\t" /* 0A0A0B0C */\
        "pshufw $0x41, %%mm0, %%mm6	\n\t" /* 0B0A0A0B */\
        "movq %%mm2, %%mm3		\n\t" /* ABCDEFGH */\
        "movq %%mm2, %%mm4		\n\t" /* ABCDEFGH */\
        "psllq $8, %%mm2		\n\t" /* 0ABCDEFG */\
        "psllq $16, %%mm3		\n\t" /* 00ABCDEF */\
        "psllq $24, %%mm4		\n\t" /* 000ABCDE */\
        "punpckhbw %%mm7, %%mm2		\n\t" /* 0D0E0F0G */\
        "punpckhbw %%mm7, %%mm3		\n\t" /* 0C0D0E0F */\
        "punpckhbw %%mm7, %%mm4		\n\t" /* 0B0C0D0E */\
        "paddw %%mm3, %%mm5		\n\t" /* b */\
        "paddw %%mm2, %%mm6		\n\t" /* c */\
        "paddw %%mm5, %%mm5		\n\t" /* 2b */\
        "psubw %%mm5, %%mm6		\n\t" /* c - 2b */\
        "pshufw $0x06, %%mm0, %%mm5	\n\t" /* 0C0B0A0A */\
        "pmullw "MANGLE(ff_pw_3)", %%mm6		\n\t" /* 3c - 6b */\
        "paddw %%mm4, %%mm0		\n\t" /* a */\
        "paddw %%mm1, %%mm5		\n\t" /* d */\
        "pmullw "MANGLE(ff_pw_20)", %%mm0		\n\t" /* 20a */\
        "psubw %%mm5, %%mm0		\n\t" /* 20a - d */\
        "paddw %6, %%mm6		\n\t"\
        "paddw %%mm6, %%mm0		\n\t" /* 20a - 6b + 3c - d */\
        "psraw $5, %%mm0		\n\t"\
        /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\
        \
        "movd 5(%0), %%mm5		\n\t" /* FGHI */\
        "punpcklbw %%mm7, %%mm5		\n\t" /* 0F0G0H0I */\
        "pshufw $0xF9, %%mm5, %%mm6	\n\t" /* 0G0H0I0I */\
        "paddw %%mm5, %%mm1		\n\t" /* a */\
        "paddw %%mm6, %%mm2		\n\t" /* b */\
        "pshufw $0xBE, %%mm5, %%mm6	\n\t" /* 0H0I0I0H */\
        "pshufw $0x6F, %%mm5, %%mm5	\n\t" /* 0I0I0H0G */\
        "paddw %%mm6, %%mm3		\n\t" /* c */\
        "paddw %%mm5, %%mm4		\n\t" /* d */\
        "paddw %%mm2, %%mm2		\n\t" /* 2b */\
        "psubw %%mm2, %%mm3		\n\t" /* c - 2b */\
        "pmullw "MANGLE(ff_pw_20)", %%mm1		\n\t" /* 20a */\
        "pmullw "MANGLE(ff_pw_3)", %%mm3		\n\t" /* 3c - 6b */\
        "psubw %%mm4, %%mm3		\n\t" /* -6b + 3c - d */\
        "paddw %6, %%mm1		\n\t"\
        "paddw %%mm1, %%mm3		\n\t" /* 20a - 6b + 3c - d */\
        "psraw $5, %%mm3		\n\t"\
        "packuswb %%mm3, %%mm0		\n\t"\
        OP_MMX2(%%mm0, (%1), %%mm4, q)\
        \
        "addl %3, %0			\n\t"\
        "addl %4, %1			\n\t"\
        "decl %2			\n\t"\
        " jnz 1b			\n\t"\
        : "+a"(src), "+c"(dst), "+m"(h)\
        : "S"(srcStride), "D"(dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(temp), "m"(ROUNDER)\
        : "memory"\
    );\
}\
\
static void OPNAME ## mpeg4_qpel8_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
    int i;\
    int16_t temp[8];\
    /* quick HACK, XXX FIXME MUST be optimized */\
    for(i=0; i<h; i++)\
    {\
        temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\
        temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\
        temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\
        temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\
        temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\
        temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 8]);\
        temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 8])*3 - (src[ 3]+src[ 7]);\
        temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 8])*6 + (src[ 5]+src[ 7])*3 - (src[ 4]+src[ 6]);\
        asm volatile(\
            "movq (%0), %%mm0		\n\t"\
            "movq 8(%0), %%mm1		\n\t"\
            "paddw %2, %%mm0		\n\t"\
            "paddw %2, %%mm1		\n\t"\
            "psraw $5, %%mm0		\n\t"\
            "psraw $5, %%mm1		\n\t"\
            "packuswb %%mm1, %%mm0	\n\t"\
            OP_3DNOW(%%mm0, (%1), %%mm1, q)\
            :: "r"(temp), "r"(dst), "m"(ROUNDER)\
            :"memory"\
        );\
        dst+=dstStride;\
        src+=srcStride;\
    }\
}

#define QPEL_OP(OPNAME, ROUNDER, RND, OP, MMX)\
\
static void OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
    uint64_t temp[17*4];\
    uint64_t *temp_ptr= temp;\
    int count= 17;\
\
    /*FIXME unroll */\
    asm volatile(\
        "pxor %%mm7, %%mm7		\n\t"\
        "1:				\n\t"\
        "movq (%0), %%mm0		\n\t"\
        "movq (%0), %%mm1		\n\t"\
        "movq 8(%0), %%mm2		\n\t"\
        "movq 8(%0), %%mm3		\n\t"\
        "punpcklbw %%mm7, %%mm0		\n\t"\
        "punpckhbw %%mm7, %%mm1		\n\t"\
        "punpcklbw %%mm7, %%mm2		\n\t"\
        "punpckhbw %%mm7, %%mm3		\n\t"\
        "movq %%mm0, (%1)		\n\t"\
        "movq %%mm1, 17*8(%1)		\n\t"\
        "movq %%mm2, 2*17*8(%1)		\n\t"\
        "movq %%mm3, 3*17*8(%1)		\n\t"\
        "addl $8, %1			\n\t"\
        "addl %3, %0			\n\t"\
        "decl %2			\n\t"\
        " jnz 1b			\n\t"\
        : "+r" (src), "+r" (temp_ptr), "+r"(count)\
        : "r" (srcStride)\
        : "memory"\
    );\
    \
    temp_ptr= temp;\
    count=4;\
    \
/*FIXME reorder for speed */\
    asm volatile(\
        /*"pxor %%mm7, %%mm7		\n\t"*/\
        "1:				\n\t"\
        "movq (%0), %%mm0		\n\t"\
        "movq 8(%0), %%mm1		\n\t"\
        "movq 16(%0), %%mm2		\n\t"\
        "movq 24(%0), %%mm3		\n\t"\
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0),  8(%0),   (%0), 32(%0), (%1), OP)\
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5,  8(%0),   (%0),   (%0), 40(%0), (%1, %3), OP)\
        "addl %4, %1			\n\t"\
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5,   (%0),   (%0),  8(%0), 48(%0), (%1), OP)\
        \
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5,   (%0),  8(%0), 16(%0), 56(%0), (%1, %3), OP)\
        "addl %4, %1			\n\t"\
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5,  8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 72(%0), (%1, %3), OP)\
        "addl %4, %1			\n\t"\
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 80(%0), (%1), OP)\
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 88(%0), (%1, %3), OP)\
        "addl %4, %1			\n\t"\
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 40(%0), 48(%0), 56(%0), 96(%0), (%1), OP)\
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 48(%0), 56(%0), 64(%0),104(%0), (%1, %3), OP)\
        "addl %4, %1			\n\t"\
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 56(%0), 64(%0), 72(%0),112(%0), (%1), OP)\
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 64(%0), 72(%0), 80(%0),120(%0), (%1, %3), OP)\
        "addl %4, %1			\n\t"\
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 72(%0), 80(%0), 88(%0),128(%0), (%1), OP)\
        \
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 80(%0), 88(%0), 96(%0),128(%0), (%1, %3), OP)\
        "addl %4, %1			\n\t"  \
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 88(%0), 96(%0),104(%0),120(%0), (%1), OP)\
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 96(%0),104(%0),112(%0),112(%0), (%1, %3), OP)\
        \
        "addl $136, %0			\n\t"\
        "addl %6, %1			\n\t"\
        "decl %2			\n\t"\
        " jnz 1b			\n\t"\
        \
        : "+r"(temp_ptr), "+r"(dst), "+g"(count)\
        : "r"(dstStride), "r"(2*dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(ROUNDER), "g"(4-14*dstStride)\
        :"memory"\
    );\
}\
\
static void OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
    uint64_t temp[9*4];\
    uint64_t *temp_ptr= temp;\
    int count= 9;\
\
    /*FIXME unroll */\
    asm volatile(\
        "pxor %%mm7, %%mm7		\n\t"\
        "1:				\n\t"\
        "movq (%0), %%mm0		\n\t"\
        "movq (%0), %%mm1		\n\t"\
        "punpcklbw %%mm7, %%mm0		\n\t"\
        "punpckhbw %%mm7, %%mm1		\n\t"\
        "movq %%mm0, (%1)		\n\t"\
        "movq %%mm1, 9*8(%1)		\n\t"\
        "addl $8, %1			\n\t"\
        "addl %3, %0			\n\t"\
        "decl %2			\n\t"\
        " jnz 1b			\n\t"\
        : "+r" (src), "+r" (temp_ptr), "+r"(count)\
        : "r" (srcStride)\
        : "memory"\
    );\
    \
    temp_ptr= temp;\
    count=2;\
    \
/*FIXME reorder for speed */\
    asm volatile(\
        /*"pxor %%mm7, %%mm7		\n\t"*/\
        "1:				\n\t"\
        "movq (%0), %%mm0		\n\t"\
        "movq 8(%0), %%mm1		\n\t"\
        "movq 16(%0), %%mm2		\n\t"\
        "movq 24(%0), %%mm3		\n\t"\
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0),  8(%0),   (%0), 32(%0), (%1), OP)\
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5,  8(%0),   (%0),   (%0), 40(%0), (%1, %3), OP)\
        "addl %4, %1			\n\t"\
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5,   (%0),   (%0),  8(%0), 48(%0), (%1), OP)\
        \
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5,   (%0),  8(%0), 16(%0), 56(%0), (%1, %3), OP)\
        "addl %4, %1			\n\t"\
        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5,  8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\
        \
        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 64(%0), (%1, %3), OP)\
        "addl %4, %1			\n\t"\
        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 56(%0), (%1), OP)\
        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 48(%0), (%1, %3), OP)\
                \
        "addl $72, %0			\n\t"\
        "addl %6, %1			\n\t"\
        "decl %2			\n\t"\
        " jnz 1b			\n\t"\
         \
        : "+r"(temp_ptr), "+r"(dst), "+g"(count)\
        : "r"(dstStride), "r"(2*dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(ROUNDER), "g"(4-6*dstStride)\
        : "memory"\
   );\
}\
\
static void OPNAME ## qpel8_mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
    OPNAME ## pixels8_mmx(dst, src, stride, 8);\
}\
\
static void OPNAME ## qpel8_mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    uint64_t temp[8];\
    uint8_t * const half= (uint8_t*)temp;\
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\
    OPNAME ## pixels8_l2_mmx(dst, src, half, stride, stride, 8);\
}\
\
static void OPNAME ## qpel8_mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    OPNAME ## mpeg4_qpel8_h_lowpass_ ## MMX(dst, src, stride, stride, 8);\
}\
\
static void OPNAME ## qpel8_mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    uint64_t temp[8];\
    uint8_t * const half= (uint8_t*)temp;\
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\
    OPNAME ## pixels8_l2_mmx(dst, src+1, half, stride, stride, 8);\
}\
\
static void OPNAME ## qpel8_mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    uint64_t temp[8];\
    uint8_t * const half= (uint8_t*)temp;\
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride);\
    OPNAME ## pixels8_l2_mmx(dst, src, half, stride, stride, 8);\
}\
\
static void OPNAME ## qpel8_mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, src, stride, stride);\
}\
\
static void OPNAME ## qpel8_mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    uint64_t temp[8];\
    uint8_t * const half= (uint8_t*)temp;\
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride);\
    OPNAME ## pixels8_l2_mmx(dst, src+stride, half, stride, stride, 8);\
}\
static void OPNAME ## qpel8_mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    uint64_t half[8 + 9];\
    uint8_t * const halfH= ((uint8_t*)half) + 64;\
    uint8_t * const halfHV= ((uint8_t*)half);\
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
    put ## RND ## pixels8_l2_mmx(halfH, src, halfH, 8, stride, 9);\
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
    OPNAME ## pixels8_l2_mmx(dst, halfH, halfHV, stride, 8, 8);\
}\
static void OPNAME ## qpel8_mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    uint64_t half[8 + 9];\
    uint8_t * const halfH= ((uint8_t*)half) + 64;\
    uint8_t * const halfHV= ((uint8_t*)half);\
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
    put ## RND ## pixels8_l2_mmx(halfH, src+1, halfH, 8, stride, 9);\
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
    OPNAME ## pixels8_l2_mmx(dst, halfH, halfHV, stride, 8, 8);\
}\
static void OPNAME ## qpel8_mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    uint64_t half[8 + 9];\
    uint8_t * const halfH= ((uint8_t*)half) + 64;\
    uint8_t * const halfHV= ((uint8_t*)half);\
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
    put ## RND ## pixels8_l2_mmx(halfH, src, halfH, 8, stride, 9);\
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
    OPNAME ## pixels8_l2_mmx(dst, halfH+8, halfHV, stride, 8, 8);\
}\
static void OPNAME ## qpel8_mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    uint64_t half[8 + 9];\
    uint8_t * const halfH= ((uint8_t*)half) + 64;\
    uint8_t * const halfHV= ((uint8_t*)half);\
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
    put ## RND ## pixels8_l2_mmx(halfH, src+1, halfH, 8, stride, 9);\
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
    OPNAME ## pixels8_l2_mmx(dst, halfH+8, halfHV, stride, 8, 8);\
}\
static void OPNAME ## qpel8_mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    uint64_t half[8 + 9];\
    uint8_t * const halfH= ((uint8_t*)half) + 64;\
    uint8_t * const halfHV= ((uint8_t*)half);\
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
    OPNAME ## pixels8_l2_mmx(dst, halfH, halfHV, stride, 8, 8);\
}\
static void OPNAME ## qpel8_mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    uint64_t half[8 + 9];\
    uint8_t * const halfH= ((uint8_t*)half) + 64;\
    uint8_t * const halfHV= ((uint8_t*)half);\
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
    OPNAME ## pixels8_l2_mmx(dst, halfH+8, halfHV, stride, 8, 8);\
}\
static void OPNAME ## qpel8_mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    uint64_t half[8 + 9];\
    uint8_t * const halfH= ((uint8_t*)half);\
    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
    put ## RND ## pixels8_l2_mmx(halfH, src, halfH, 8, stride, 9);\
    OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\
}\