cavsdsp_mmx.c

mediastreamer2是开源的网络传输媒体流的库

        "pmullw %6, %%mm6           \n\t"\
        "movq "#D", %%mm7           \n\t"\
        "pmullw %5, %%mm7           \n\t"\
        "psllw $3, "#B"             \n\t"\
        "psubw "#B", %%mm6          \n\t"\
        "psraw $3, "#B"             \n\t"\
        "paddw %%mm7, %%mm6         \n\t"\
        "paddw "#B", %%mm6          \n\t"\
        "paddw "#E", "#E"           \n\t"\
        "pxor %%mm7, %%mm7          \n\t"\
        "add %2, %0                 \n\t"\
        "punpcklbw %%mm7, "#F"      \n\t"\
        "psubw "#E", %%mm6          \n\t"\
        "psraw $1, "#E"             \n\t"\
        "psubw "#F", %%mm6          \n\t"\
        "paddw %4, %%mm6            \n\t"\
        "psraw $7, %%mm6            \n\t"\
        "packuswb %%mm6, %%mm6      \n\t"\
        OP(%%mm6, (%1), A, d)            \
        "add %3, %1                 \n\t"


#define QPEL_CAVSVNUM(VOP,OP,ADD,MUL1,MUL2)\
    int w= 2;\
    src -= 2*srcStride;\
    \
    while(w--){\
      asm volatile(\
        "pxor %%mm7, %%mm7          \n\t"\
        "movd (%0), %%mm0           \n\t"\
        "add %2, %0                 \n\t"\
        "movd (%0), %%mm1           \n\t"\
        "add %2, %0                 \n\t"\
        "movd (%0), %%mm2           \n\t"\
        "add %2, %0                 \n\t"\
        "movd (%0), %%mm3           \n\t"\
        "add %2, %0                 \n\t"\
        "movd (%0), %%mm4           \n\t"\
        "add %2, %0                 \n\t"\
        "punpcklbw %%mm7, %%mm0     \n\t"\
        "punpcklbw %%mm7, %%mm1     \n\t"\
        "punpcklbw %%mm7, %%mm2     \n\t"\
        "punpcklbw %%mm7, %%mm3     \n\t"\
        "punpcklbw %%mm7, %%mm4     \n\t"\
        VOP(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
        VOP(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
        VOP(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
        VOP(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
        VOP(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\
        VOP(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\
        VOP(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
        VOP(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
        \
        : "+a"(src), "+c"(dst)\
        : "S"((long)srcStride), "D"((long)dstStride), "m"(ADD), "m"(MUL1), "m"(MUL2)\
        : "memory"\
     );\
     if(h==16){\
        asm volatile(\
            VOP(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
            VOP(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
            VOP(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\
            VOP(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\
            VOP(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\
            VOP(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\
            VOP(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
            VOP(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
            \
           : "+a"(src), "+c"(dst)\
           : "S"((long)srcStride), "D"((long)dstStride), "m"(ADD),  "m"(MUL1), "m"(MUL2)\
           : "memory"\
        );\
     }\
     src += 4-(h+5)*srcStride;\
     dst += 4-h*dstStride;\
   }

#define QPEL_CAVS(OPNAME, OP, MMX)\
static void OPNAME ## cavs_qpel8_h_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
    int h=8;\
    asm volatile(\
        "pxor %%mm7, %%mm7          \n\t"\
        "movq %5, %%mm6             \n\t"\
        "1:                         \n\t"\
        "movq    (%0), %%mm0        \n\t"\
        "movq   1(%0), %%mm2        \n\t"\
        "movq %%mm0, %%mm1          \n\t"\
        "movq %%mm2, %%mm3          \n\t"\
        "punpcklbw %%mm7, %%mm0     \n\t"\
        "punpckhbw %%mm7, %%mm1     \n\t"\
        "punpcklbw %%mm7, %%mm2     \n\t"\
        "punpckhbw %%mm7, %%mm3     \n\t"\
        "paddw %%mm2, %%mm0         \n\t"\
        "paddw %%mm3, %%mm1         \n\t"\
        "pmullw %%mm6, %%mm0        \n\t"\
        "pmullw %%mm6, %%mm1        \n\t"\
        "movq   -1(%0), %%mm2       \n\t"\
        "movq    2(%0), %%mm4       \n\t"\
        "movq %%mm2, %%mm3          \n\t"\
        "movq %%mm4, %%mm5          \n\t"\
        "punpcklbw %%mm7, %%mm2     \n\t"\
        "punpckhbw %%mm7, %%mm3     \n\t"\
        "punpcklbw %%mm7, %%mm4     \n\t"\
        "punpckhbw %%mm7, %%mm5     \n\t"\
        "paddw %%mm4, %%mm2         \n\t"\
        "paddw %%mm3, %%mm5         \n\t"\
        "psubw %%mm2, %%mm0         \n\t"\
        "psubw %%mm5, %%mm1         \n\t"\
        "movq %6, %%mm5             \n\t"\
        "paddw %%mm5, %%mm0         \n\t"\
        "paddw %%mm5, %%mm1         \n\t"\
        "psraw $3, %%mm0            \n\t"\
        "psraw $3, %%mm1            \n\t"\
        "packuswb %%mm1, %%mm0      \n\t"\
        OP(%%mm0, (%1),%%mm5, 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_5), "m"(ff_pw_4)\
        : "memory"\
    );\
}\
\
static inline void OPNAME ## cavs_qpel8or16_v1_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
  QPEL_CAVSVNUM(QPEL_CAVSV1,OP,ff_pw_64,ff_pw_96,ff_pw_42)      \
}\
\
static inline void OPNAME ## cavs_qpel8or16_v2_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
  QPEL_CAVSVNUM(QPEL_CAVSV2,OP,ff_pw_4,ff_pw_5,ff_pw_5)         \
}\
\
static inline void OPNAME ## cavs_qpel8or16_v3_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
  QPEL_CAVSVNUM(QPEL_CAVSV3,OP,ff_pw_64,ff_pw_96,ff_pw_42)      \
}\
\
static void OPNAME ## cavs_qpel8_v1_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
    OPNAME ## cavs_qpel8or16_v1_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
}\
static void OPNAME ## cavs_qpel16_v1_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
    OPNAME ## cavs_qpel8or16_v1_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
    OPNAME ## cavs_qpel8or16_v1_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
}\
\
static void OPNAME ## cavs_qpel8_v2_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
    OPNAME ## cavs_qpel8or16_v2_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
}\
static void OPNAME ## cavs_qpel16_v2_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
    OPNAME ## cavs_qpel8or16_v2_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
    OPNAME ## cavs_qpel8or16_v2_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
}\
\
static void OPNAME ## cavs_qpel8_v3_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
    OPNAME ## cavs_qpel8or16_v3_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
}\
static void OPNAME ## cavs_qpel16_v3_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
    OPNAME ## cavs_qpel8or16_v3_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
    OPNAME ## cavs_qpel8or16_v3_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
}\
\
static void OPNAME ## cavs_qpel16_h_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
    OPNAME ## cavs_qpel8_h_ ## MMX(dst  , src  , dstStride, srcStride);\
    OPNAME ## cavs_qpel8_h_ ## MMX(dst+8, src+8, dstStride, srcStride);\
    src += 8*srcStride;\
    dst += 8*dstStride;\
    OPNAME ## cavs_qpel8_h_ ## MMX(dst  , src  , dstStride, srcStride);\
    OPNAME ## cavs_qpel8_h_ ## MMX(dst+8, src+8, dstStride, srcStride);\
}\

#define CAVS_MC(OPNAME, SIZE, MMX) \
static void ff_ ## OPNAME ## cavs_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    OPNAME ## cavs_qpel ## SIZE ## _h_ ## MMX(dst, src, stride, stride);\
}\
\
static void ff_ ## OPNAME ## cavs_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    OPNAME ## cavs_qpel ## SIZE ## _v1_ ## MMX(dst, src, stride, stride);\
}\
\
static void ff_ ## OPNAME ## cavs_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    OPNAME ## cavs_qpel ## SIZE ## _v2_ ## MMX(dst, src, stride, stride);\
}\
\
static void ff_ ## OPNAME ## cavs_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    OPNAME ## cavs_qpel ## SIZE ## _v3_ ## MMX(dst, src, stride, stride);\
}\

#define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b "    \n\t"
#define AVG_3DNOW_OP(a,b,temp, size) \
"mov" #size " " #b ", " #temp "   \n\t"\
"pavgusb " #temp ", " #a "        \n\t"\
"mov" #size " " #a ", " #b "      \n\t"
#define AVG_MMX2_OP(a,b,temp, size) \
"mov" #size " " #b ", " #temp "   \n\t"\
"pavgb " #temp ", " #a "          \n\t"\
"mov" #size " " #a ", " #b "      \n\t"

QPEL_CAVS(put_,       PUT_OP, 3dnow)
QPEL_CAVS(avg_, AVG_3DNOW_OP, 3dnow)
QPEL_CAVS(put_,       PUT_OP, mmx2)
QPEL_CAVS(avg_,  AVG_MMX2_OP, mmx2)

CAVS_MC(put_, 8, 3dnow)
CAVS_MC(put_, 16,3dnow)
CAVS_MC(avg_, 8, 3dnow)
CAVS_MC(avg_, 16,3dnow)
CAVS_MC(put_, 8, mmx2)
CAVS_MC(put_, 16,mmx2)
CAVS_MC(avg_, 8, mmx2)
CAVS_MC(avg_, 16,mmx2)

void ff_put_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride);
void ff_avg_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride);
void ff_put_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride);
void ff_avg_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride);

void ff_cavsdsp_init_mmx2(DSPContext* c, AVCodecContext *avctx) {
#define dspfunc(PFX, IDX, NUM) \
    c->PFX ## _pixels_tab[IDX][ 0] = ff_ ## PFX ## NUM ## _mc00_mmx2; \
    c->PFX ## _pixels_tab[IDX][ 2] = ff_ ## PFX ## NUM ## _mc20_mmx2; \
    c->PFX ## _pixels_tab[IDX][ 4] = ff_ ## PFX ## NUM ## _mc01_mmx2; \
    c->PFX ## _pixels_tab[IDX][ 8] = ff_ ## PFX ## NUM ## _mc02_mmx2; \
    c->PFX ## _pixels_tab[IDX][12] = ff_ ## PFX ## NUM ## _mc03_mmx2; \

    dspfunc(put_cavs_qpel, 0, 16);
    dspfunc(put_cavs_qpel, 1, 8);
    dspfunc(avg_cavs_qpel, 0, 16);
    dspfunc(avg_cavs_qpel, 1, 8);
#undef dspfunc
    c->cavs_idct8_add = cavs_idct8_add_mmx;
}

void ff_cavsdsp_init_3dnow(DSPContext* c, AVCodecContext *avctx) {
#define dspfunc(PFX, IDX, NUM) \
    c->PFX ## _pixels_tab[IDX][ 0] = ff_ ## PFX ## NUM ## _mc00_mmx2; \
    c->PFX ## _pixels_tab[IDX][ 2] = ff_ ## PFX ## NUM ## _mc20_3dnow; \
    c->PFX ## _pixels_tab[IDX][ 4] = ff_ ## PFX ## NUM ## _mc01_3dnow; \
    c->PFX ## _pixels_tab[IDX][ 8] = ff_ ## PFX ## NUM ## _mc02_3dnow; \
    c->PFX ## _pixels_tab[IDX][12] = ff_ ## PFX ## NUM ## _mc03_3dnow; \

    dspfunc(put_cavs_qpel, 0, 16);
    dspfunc(put_cavs_qpel, 1, 8);
    dspfunc(avg_cavs_qpel, 0, 16);
    dspfunc(avg_cavs_qpel, 1, 8);
#undef dspfunc
    c->cavs_idct8_add = cavs_idct8_add_mmx;
}