dsputil_mmx.c

linux下的MPEG1

    "psubw " #a1 ", " #b1 "           \n\t"\
    "psubw " #a2 ", " #b2 "           \n\t"

#define HADAMARD48\
        LBUTTERFLY2(%%mm0, %%mm1, %%mm2, %%mm3)\
        LBUTTERFLY2(%%mm4, %%mm5, %%mm6, %%mm7)\
        LBUTTERFLY2(%%mm0, %%mm2, %%mm1, %%mm3)\
        LBUTTERFLY2(%%mm4, %%mm6, %%mm5, %%mm7)\
        LBUTTERFLY2(%%mm0, %%mm4, %%mm1, %%mm5)\
        LBUTTERFLY2(%%mm2, %%mm6, %%mm3, %%mm7)\

#define MMABS(a,z)\
    "pxor " #z ", " #z "              \n\t"\
    "pcmpgtw " #a ", " #z "           \n\t"\
    "pxor " #z ", " #a "              \n\t"\
    "psubw " #z ", " #a "             \n\t"

#define MMABS_SUM(a,z, sum)\
    "pxor " #z ", " #z "              \n\t"\
    "pcmpgtw " #a ", " #z "           \n\t"\
    "pxor " #z ", " #a "              \n\t"\
    "psubw " #z ", " #a "             \n\t"\
    "paddusw " #a ", " #sum "         \n\t"

#define MMABS_MMX2(a,z)\
    "pxor " #z ", " #z "              \n\t"\
    "psubw " #a ", " #z "             \n\t"\
    "pmaxsw " #z ", " #a "            \n\t"

#define MMABS_SUM_MMX2(a,z, sum)\
    "pxor " #z ", " #z "              \n\t"\
    "psubw " #a ", " #z "             \n\t"\
    "pmaxsw " #z ", " #a "            \n\t"\
    "paddusw " #a ", " #sum "         \n\t"

#define TRANSPOSE4(a,b,c,d,t)\
    SBUTTERFLY(a,b,t,wd) /* a=aebf t=cgdh */\
    SBUTTERFLY(c,d,b,wd) /* c=imjn b=kolp */\
    SBUTTERFLY(a,c,d,dq) /* a=aeim d=bfjn */\
    SBUTTERFLY(t,b,c,dq) /* t=cgko c=dhlp */

#define LOAD4(o, a, b, c, d)\
        "movq "#o"(%1), " #a "        \n\t"\
        "movq "#o"+16(%1), " #b "     \n\t"\
        "movq "#o"+32(%1), " #c "     \n\t"\
        "movq "#o"+48(%1), " #d "     \n\t"

#define STORE4(o, a, b, c, d)\
        "movq "#a", "#o"(%1)          \n\t"\
        "movq "#b", "#o"+16(%1)       \n\t"\
        "movq "#c", "#o"+32(%1)       \n\t"\
        "movq "#d", "#o"+48(%1)       \n\t"\

static int hadamard8_diff_mmx(void *s, uint8_t *src1, uint8_t *src2, int stride, int h){
    DECLARE_ALIGNED_8(uint64_t, temp[16]);
    int sum=0;

    assert(h==8);

    diff_pixels_mmx((DCTELEM*)temp, src1, src2, stride);

    asm volatile(
        LOAD4(0 , %%mm0, %%mm1, %%mm2, %%mm3)
        LOAD4(64, %%mm4, %%mm5, %%mm6, %%mm7)

        HADAMARD48

        "movq %%mm7, 112(%1)            \n\t"

        TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm7)
        STORE4(0 , %%mm0, %%mm3, %%mm7, %%mm2)

        "movq 112(%1), %%mm7            \n\t"
        TRANSPOSE4(%%mm4, %%mm5, %%mm6, %%mm7, %%mm0)
        STORE4(64, %%mm4, %%mm7, %%mm0, %%mm6)

        LOAD4(8 , %%mm0, %%mm1, %%mm2, %%mm3)
        LOAD4(72, %%mm4, %%mm5, %%mm6, %%mm7)

        HADAMARD48

        "movq %%mm7, 120(%1)            \n\t"

        TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm7)
        STORE4(8 , %%mm0, %%mm3, %%mm7, %%mm2)

        "movq 120(%1), %%mm7            \n\t"
        TRANSPOSE4(%%mm4, %%mm5, %%mm6, %%mm7, %%mm0)
        "movq %%mm7, %%mm5              \n\t"//FIXME remove
        "movq %%mm6, %%mm7              \n\t"
        "movq %%mm0, %%mm6              \n\t"
//        STORE4(72, %%mm4, %%mm7, %%mm0, %%mm6) //FIXME remove

        LOAD4(64, %%mm0, %%mm1, %%mm2, %%mm3)
//        LOAD4(72, %%mm4, %%mm5, %%mm6, %%mm7)

        HADAMARD48
        "movq %%mm7, 64(%1)             \n\t"
        MMABS(%%mm0, %%mm7)
        MMABS_SUM(%%mm1, %%mm7, %%mm0)
        MMABS_SUM(%%mm2, %%mm7, %%mm0)
        MMABS_SUM(%%mm3, %%mm7, %%mm0)
        MMABS_SUM(%%mm4, %%mm7, %%mm0)
        MMABS_SUM(%%mm5, %%mm7, %%mm0)
        MMABS_SUM(%%mm6, %%mm7, %%mm0)
        "movq 64(%1), %%mm1             \n\t"
        MMABS_SUM(%%mm1, %%mm7, %%mm0)
        "movq %%mm0, 64(%1)             \n\t"

        LOAD4(0 , %%mm0, %%mm1, %%mm2, %%mm3)
        LOAD4(8 , %%mm4, %%mm5, %%mm6, %%mm7)

        HADAMARD48
        "movq %%mm7, (%1)               \n\t"
        MMABS(%%mm0, %%mm7)
        MMABS_SUM(%%mm1, %%mm7, %%mm0)
        MMABS_SUM(%%mm2, %%mm7, %%mm0)
        MMABS_SUM(%%mm3, %%mm7, %%mm0)
        MMABS_SUM(%%mm4, %%mm7, %%mm0)
        MMABS_SUM(%%mm5, %%mm7, %%mm0)
        MMABS_SUM(%%mm6, %%mm7, %%mm0)
        "movq (%1), %%mm1               \n\t"
        MMABS_SUM(%%mm1, %%mm7, %%mm0)
        "movq 64(%1), %%mm1             \n\t"
        MMABS_SUM(%%mm1, %%mm7, %%mm0)

        "movq %%mm0, %%mm1              \n\t"
        "psrlq $32, %%mm0               \n\t"
        "paddusw %%mm1, %%mm0           \n\t"
        "movq %%mm0, %%mm1              \n\t"
        "psrlq $16, %%mm0               \n\t"
        "paddusw %%mm1, %%mm0           \n\t"
        "movd %%mm0, %0                 \n\t"

        : "=r" (sum)
        : "r"(temp)
    );
    return sum&0xFFFF;
}

static int hadamard8_diff_mmx2(void *s, uint8_t *src1, uint8_t *src2, int stride, int h){
    DECLARE_ALIGNED_8(uint64_t, temp[16]);
    int sum=0;

    assert(h==8);

    diff_pixels_mmx((DCTELEM*)temp, src1, src2, stride);

    asm volatile(
        LOAD4(0 , %%mm0, %%mm1, %%mm2, %%mm3)
        LOAD4(64, %%mm4, %%mm5, %%mm6, %%mm7)

        HADAMARD48

        "movq %%mm7, 112(%1)            \n\t"

        TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm7)
        STORE4(0 , %%mm0, %%mm3, %%mm7, %%mm2)

        "movq 112(%1), %%mm7            \n\t"
        TRANSPOSE4(%%mm4, %%mm5, %%mm6, %%mm7, %%mm0)
        STORE4(64, %%mm4, %%mm7, %%mm0, %%mm6)

        LOAD4(8 , %%mm0, %%mm1, %%mm2, %%mm3)
        LOAD4(72, %%mm4, %%mm5, %%mm6, %%mm7)

        HADAMARD48

        "movq %%mm7, 120(%1)            \n\t"

        TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm7)
        STORE4(8 , %%mm0, %%mm3, %%mm7, %%mm2)

        "movq 120(%1), %%mm7            \n\t"
        TRANSPOSE4(%%mm4, %%mm5, %%mm6, %%mm7, %%mm0)
        "movq %%mm7, %%mm5              \n\t"//FIXME remove
        "movq %%mm6, %%mm7              \n\t"
        "movq %%mm0, %%mm6              \n\t"
//        STORE4(72, %%mm4, %%mm7, %%mm0, %%mm6) //FIXME remove

        LOAD4(64, %%mm0, %%mm1, %%mm2, %%mm3)
//        LOAD4(72, %%mm4, %%mm5, %%mm6, %%mm7)

        HADAMARD48
        "movq %%mm7, 64(%1)             \n\t"
        MMABS_MMX2(%%mm0, %%mm7)
        MMABS_SUM_MMX2(%%mm1, %%mm7, %%mm0)
        MMABS_SUM_MMX2(%%mm2, %%mm7, %%mm0)
        MMABS_SUM_MMX2(%%mm3, %%mm7, %%mm0)
        MMABS_SUM_MMX2(%%mm4, %%mm7, %%mm0)
        MMABS_SUM_MMX2(%%mm5, %%mm7, %%mm0)
        MMABS_SUM_MMX2(%%mm6, %%mm7, %%mm0)
        "movq 64(%1), %%mm1             \n\t"
        MMABS_SUM_MMX2(%%mm1, %%mm7, %%mm0)
        "movq %%mm0, 64(%1)             \n\t"

        LOAD4(0 , %%mm0, %%mm1, %%mm2, %%mm3)
        LOAD4(8 , %%mm4, %%mm5, %%mm6, %%mm7)

        HADAMARD48
        "movq %%mm7, (%1)               \n\t"
        MMABS_MMX2(%%mm0, %%mm7)
        MMABS_SUM_MMX2(%%mm1, %%mm7, %%mm0)
        MMABS_SUM_MMX2(%%mm2, %%mm7, %%mm0)
        MMABS_SUM_MMX2(%%mm3, %%mm7, %%mm0)
        MMABS_SUM_MMX2(%%mm4, %%mm7, %%mm0)
        MMABS_SUM_MMX2(%%mm5, %%mm7, %%mm0)
        MMABS_SUM_MMX2(%%mm6, %%mm7, %%mm0)
        "movq (%1), %%mm1               \n\t"
        MMABS_SUM_MMX2(%%mm1, %%mm7, %%mm0)
        "movq 64(%1), %%mm1             \n\t"
        MMABS_SUM_MMX2(%%mm1, %%mm7, %%mm0)

        "pshufw $0x0E, %%mm0, %%mm1     \n\t"
        "paddusw %%mm1, %%mm0           \n\t"
        "pshufw $0x01, %%mm0, %%mm1     \n\t"
        "paddusw %%mm1, %%mm0           \n\t"
        "movd %%mm0, %0                 \n\t"

        : "=r" (sum)
        : "r"(temp)
    );
    return sum&0xFFFF;
}


WARPER8_16_SQ(hadamard8_diff_mmx, hadamard8_diff16_mmx)
WARPER8_16_SQ(hadamard8_diff_mmx2, hadamard8_diff16_mmx2)
#endif //CONFIG_ENCODERS

#define put_no_rnd_pixels8_mmx(a,b,c,d) put_pixels8_mmx(a,b,c,d)
#define put_no_rnd_pixels16_mmx(a,b,c,d) put_pixels16_mmx(a,b,c,d)

#define QPEL_V_LOW(m3,m4,m5,m6, pw_20, pw_3, rnd, in0, in1, in2, in7, out, OP)\
        "paddw " #m4 ", " #m3 "           \n\t" /* x1 */\
        "movq "MANGLE(ff_pw_20)", %%mm4   \n\t" /* 20 */\
        "pmullw " #m3 ", %%mm4            \n\t" /* 20x1 */\
        "movq "#in7", " #m3 "             \n\t" /* d */\
        "movq "#in0", %%mm5               \n\t" /* D */\
        "paddw " #m3 ", %%mm5             \n\t" /* x4 */\
        "psubw %%mm5, %%mm4               \n\t" /* 20x1 - x4 */\
        "movq "#in1", %%mm5               \n\t" /* C */\
        "movq "#in2", %%mm6               \n\t" /* B */\
        "paddw " #m6 ", %%mm5             \n\t" /* x3 */\
        "paddw " #m5 ", %%mm6             \n\t" /* x2 */\
        "paddw %%mm6, %%mm6               \n\t" /* 2x2 */\
        "psubw %%mm6, %%mm5               \n\t" /* -2x2 + x3 */\
        "pmullw "MANGLE(ff_pw_3)", %%mm5  \n\t" /* -6x2 + 3x3 */\
        "paddw " #rnd ", %%mm4            \n\t" /* x2 */\
        "paddw %%mm4, %%mm5               \n\t" /* 20x1 - 6x2 + 3x3 - x4 */\
        "psraw $5, %%mm5                  \n\t"\
        "packuswb %%mm5, %%mm5            \n\t"\
        OP(%%mm5, out, %%mm7, d)

#define QPEL_BASE(OPNAME, ROUNDER, RND, OP_MMX2, OP_3DNOW)\
static void OPNAME ## mpeg4_qpel16_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"\
        "movq %%mm0, %5                   \n\t"\
        /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\
        \
        "movq 5(%0), %%mm0                \n\t" /* FGHIJKLM */\
        "movq %%mm0, %%mm5                \n\t" /* FGHIJKLM */\
        "movq %%mm0, %%mm6                \n\t" /* FGHIJKLM */\
        "psrlq $8, %%mm0                  \n\t" /* GHIJKLM0 */\
        "psrlq $16, %%mm5                 \n\t" /* HIJKLM00 */\
        "punpcklbw %%mm7, %%mm0           \n\t" /* 0G0H0I0J */\
        "punpcklbw %%mm7, %%mm5           \n\t" /* 0H0I0J0K */\
        "paddw %%mm0, %%mm2               \n\t" /* b */\
        "paddw %%mm5, %%mm3               \n\t" /* c */\
        "paddw %%mm2, %%mm2               \n\t" /* 2b */\
        "psubw %%mm2, %%mm3               \n\t" /* c - 2b */\
        "movq %%mm6, %%mm2                \n\t" /* FGHIJKLM */\
        "psrlq $24, %%mm6                 \n\t" /* IJKLM000 */\
        "punpcklbw %%mm7, %%mm2           \n\t" /* 0F0G0H0I */\
        "punpcklbw %%mm7, %%mm6           \n\t" /* 0I0J0K0L */\
        "pmullw "MANGLE(ff_pw_3)", %%mm3  \n\t" /* 3c - 6b */\
        "paddw %%mm2, %%mm1               \n\t" /* a */\
        "paddw %%mm6, %%mm4               \n\t" /* d */\
        "pmullw "MANGLE(ff_pw_20)", %%mm1 \n\t" /* 20a */\
        "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"\
        "movq %5, %%mm1                   \n\t"\
        "packuswb %%mm3, %%mm1            \n\t"\
        OP_MMX2(%%mm1, (%1),%%mm4, q)\
        /* mm0= GHIJ, mm2=FGHI, mm5=HIJK, mm6=IJKL, mm7=0 */\
        \
        "movq 9(%0), %%mm1                \n\t" /* JKLMNOPQ */\
        "movq %%mm1, %%mm4                \n\t" /* JKLMNOPQ */\
        "movq %%mm1, %%mm3                \n\t" /* JKLMNOPQ */\
        "psrlq $8, %%mm1                  \n\t" /* KLMNOPQ0 */\
        "psrlq $16, %%mm4                 \n\t" /* LMNOPQ00 */\
        "punpcklbw %%mm7, %%mm1           \n\t" /* 0K0L0M0N */\
        "punpcklbw %%mm7, %%mm4           \n\t" /* 0L0M0N0O */\
        "paddw %%mm1, %%mm5               \n\t" /* b */\
        "paddw %%mm4, %%mm0               \n\t" /* c */\
        "paddw %%mm5, %%mm5               \n\t" /* 2b */\
        "psubw %%mm5, %%mm0               \n\t" /* c - 2b */\
        "movq %%mm3, %%mm5                \n\t" /* JKLMNOPQ */\
        "psrlq $24, %%mm3                 \n\t" /* MNOPQ000 */\
        "pmullw "MANGLE(ff_pw_3)", %%mm0  \n\t" /* 3c - 6b */\
        "punpcklbw %%mm7, %%mm3           \n\t" /* 0M0N0O0P */\
        "paddw %%mm3, %%mm2               \n\t" /* d */\
        "psubw %%mm2, %%mm0               \n\t" /* -6b + 3c - d */\
        "movq %%mm5, %%mm2                \n\t" /* JKLMNOPQ */\
        "punpcklbw %%mm7, %%mm2           \n\t" /* 0J0K0L0M */\
        "punpckhbw %%mm7, %%mm5           \n\t" /* 0N0O0P0Q */\
        "paddw %%mm2, %%mm6               \n\t" /* a */\
        "pmullw "MANGLE(ff_pw_20)", %%mm6 \n\t" /* 20a */\
        "paddw %6, %%mm0                  \n\t"\
        "paddw %%mm6, %%mm0               \n\t" /* 20a - 6b + 3c - d */\
        "psraw $5, %%mm0                  \n\t"\
        /* mm1=KLMN, mm2=JKLM, mm3=MNOP, mm4=LMNO, mm5=NOPQ mm7=0 */\
        \
        "paddw %%mm5, %%mm3               \n\t" /* a */\
        "pshufw $0xF9, %%mm5, %%mm6       \n\t" /* 0O0P0Q0Q */\
        "paddw %%mm4, %%mm6               \n\t" /* b */\
        "pshufw $0xBE, %%mm5, %%mm4       \n\t" /* 0P0Q0Q0P */\
        "pshufw $0x6F, %%mm5, %%mm5       \n\t" /* 0Q0Q0P0O */\
        "paddw %%mm1, %%mm4               \n\t" /* c */\
        "paddw %%mm2, %%mm5               \n\t" /* d */\
        "paddw %%mm6, %%mm6               \n\t" /* 2b */\
        "psubw %%mm6, %%mm4               \n\t" /* c - 2b */\
        "pmullw "MANGLE(ff_pw_20)", %%mm3 \n\t" /* 20a */\
        "pmullw "MANGLE(ff_pw_3)", %%mm4  \n\t" /* 3c - 6b */\
        "psubw %%mm5, %%mm3               \n\t" /* -6b + 3c - d */\
        "paddw %6, %%mm4                  \n\t"\
        "paddw %%mm3, %%mm4               \n\t" /* 20a - 6b + 3c - d */\
        "psraw $5, %%mm4                  \n\t"\
        "packuswb %%mm4, %%mm0            \n\t"\
        OP_MMX2(%%mm0, 8(%1), %%mm4, q)\
        \
        "add %3, %0                       \n\t"\
        "add %4, %1                       \n\t"\
        "decl %2                          \n\t"\
        " jnz 1b                          \n\t"\
        : "+a"(src), "+c"(dst), "+m"(h)\
        : "d"((long)srcStride), "S"((long)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(temp), "m"(ROUNDER)\
        : "memory"\
    );\
}\
\
static void OPNAME ## mpeg4_qpel16_h_