dsputil_mmx.c

ffmpeg的完整源代码和作者自己写的文档。不但有在Linux的工程哦

      "punpcklbw %%mm7,%%mm1\n"
      "punpckhbw %%mm7,%%mm5\n"
      "punpckhbw %%mm7,%%mm3\n"
      "psubw %%mm1, %%mm4\n"
      "psubw %%mm3, %%mm5\n"
      "psubw %%mm4, %%mm0\n"
      "psubw %%mm5, %%mm2\n"
      "pxor %%mm3, %%mm3\n"
      "pxor %%mm1, %%mm1\n"
      "pcmpgtw %%mm0, %%mm3\n\t"
      "pcmpgtw %%mm2, %%mm1\n\t"
      "pxor %%mm3, %%mm0\n"
      "pxor %%mm1, %%mm2\n"
      "psubw %%mm3, %%mm0\n"
      "psubw %%mm1, %%mm2\n"
      "paddw %%mm0, %%mm2\n"
      "paddw %%mm2, %%mm6\n"

      "add %2,%0\n"
      "subl $2, %%ecx\n"
      " jnz 1b\n"

      "movq %%mm6, %%mm0\n"
      "punpcklwd %%mm7,%%mm0\n"
      "punpckhwd %%mm7,%%mm6\n"
      "paddd %%mm0, %%mm6\n"

      "movq %%mm6,%%mm0\n"
      "psrlq $32, %%mm6\n"
      "paddd %%mm6,%%mm0\n"
      "movd %%mm0,%1\n"
      : "+r" (pix1), "=r"(tmp)
      : "r" ((long)line_size) , "g" (h-2)
      : "%ecx");
      return tmp + hf_noise8_mmx(pix+8, line_size, h);
}

static int nsse16_mmx(void *p, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
    MpegEncContext *c = p;
    int score1, score2;

    if(c) score1 = c->dsp.sse[0](c, pix1, pix2, line_size, h);
    else  score1 = sse16_mmx(c, pix1, pix2, line_size, h);
    score2= hf_noise16_mmx(pix1, line_size, h) - hf_noise16_mmx(pix2, line_size, h);

    if(c) return score1 + FFABS(score2)*c->avctx->nsse_weight;
    else  return score1 + FFABS(score2)*8;
}

static int nsse8_mmx(void *p, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
    MpegEncContext *c = p;
    int score1= sse8_mmx(c, pix1, pix2, line_size, h);
    int score2= hf_noise8_mmx(pix1, line_size, h) - hf_noise8_mmx(pix2, line_size, h);

    if(c) return score1 + FFABS(score2)*c->avctx->nsse_weight;
    else  return score1 + FFABS(score2)*8;
}

static int vsad_intra16_mmx(void *v, uint8_t * pix, uint8_t * dummy, int line_size, int h) {
    int tmp;

    assert( (((int)pix) & 7) == 0);
    assert((line_size &7) ==0);

#define SUM(in0, in1, out0, out1) \
      "movq (%0), %%mm2\n"\
      "movq 8(%0), %%mm3\n"\
      "add %2,%0\n"\
      "movq %%mm2, " #out0 "\n"\
      "movq %%mm3, " #out1 "\n"\
      "psubusb " #in0 ", %%mm2\n"\
      "psubusb " #in1 ", %%mm3\n"\
      "psubusb " #out0 ", " #in0 "\n"\
      "psubusb " #out1 ", " #in1 "\n"\
      "por %%mm2, " #in0 "\n"\
      "por %%mm3, " #in1 "\n"\
      "movq " #in0 ", %%mm2\n"\
      "movq " #in1 ", %%mm3\n"\
      "punpcklbw %%mm7, " #in0 "\n"\
      "punpcklbw %%mm7, " #in1 "\n"\
      "punpckhbw %%mm7, %%mm2\n"\
      "punpckhbw %%mm7, %%mm3\n"\
      "paddw " #in1 ", " #in0 "\n"\
      "paddw %%mm3, %%mm2\n"\
      "paddw %%mm2, " #in0 "\n"\
      "paddw " #in0 ", %%mm6\n"


  asm volatile (
      "movl %3,%%ecx\n"
      "pxor %%mm6,%%mm6\n"
      "pxor %%mm7,%%mm7\n"
      "movq (%0),%%mm0\n"
      "movq 8(%0),%%mm1\n"
      "add %2,%0\n"
      "subl $2, %%ecx\n"
      SUM(%%mm0, %%mm1, %%mm4, %%mm5)
      "1:\n"

      SUM(%%mm4, %%mm5, %%mm0, %%mm1)

      SUM(%%mm0, %%mm1, %%mm4, %%mm5)

      "subl $2, %%ecx\n"
      "jnz 1b\n"

      "movq %%mm6,%%mm0\n"
      "psrlq $32, %%mm6\n"
      "paddw %%mm6,%%mm0\n"
      "movq %%mm0,%%mm6\n"
      "psrlq $16, %%mm0\n"
      "paddw %%mm6,%%mm0\n"
      "movd %%mm0,%1\n"
      : "+r" (pix), "=r"(tmp)
      : "r" ((long)line_size) , "m" (h)
      : "%ecx");
    return tmp & 0xFFFF;
}
#undef SUM

static int vsad_intra16_mmx2(void *v, uint8_t * pix, uint8_t * dummy, int line_size, int h) {
    int tmp;

    assert( (((int)pix) & 7) == 0);
    assert((line_size &7) ==0);

#define SUM(in0, in1, out0, out1) \
      "movq (%0), " #out0 "\n"\
      "movq 8(%0), " #out1 "\n"\
      "add %2,%0\n"\
      "psadbw " #out0 ", " #in0 "\n"\
      "psadbw " #out1 ", " #in1 "\n"\
      "paddw " #in1 ", " #in0 "\n"\
      "paddw " #in0 ", %%mm6\n"

  asm volatile (
      "movl %3,%%ecx\n"
      "pxor %%mm6,%%mm6\n"
      "pxor %%mm7,%%mm7\n"
      "movq (%0),%%mm0\n"
      "movq 8(%0),%%mm1\n"
      "add %2,%0\n"
      "subl $2, %%ecx\n"
      SUM(%%mm0, %%mm1, %%mm4, %%mm5)
      "1:\n"

      SUM(%%mm4, %%mm5, %%mm0, %%mm1)

      SUM(%%mm0, %%mm1, %%mm4, %%mm5)

      "subl $2, %%ecx\n"
      "jnz 1b\n"

      "movd %%mm6,%1\n"
      : "+r" (pix), "=r"(tmp)
      : "r" ((long)line_size) , "m" (h)
      : "%ecx");
    return tmp;
}
#undef SUM

static int vsad16_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
    int tmp;

    assert( (((int)pix1) & 7) == 0);
    assert( (((int)pix2) & 7) == 0);
    assert((line_size &7) ==0);

#define SUM(in0, in1, out0, out1) \
      "movq (%0),%%mm2\n"\
      "movq (%1)," #out0 "\n"\
      "movq 8(%0),%%mm3\n"\
      "movq 8(%1)," #out1 "\n"\
      "add %3,%0\n"\
      "add %3,%1\n"\
      "psubb " #out0 ", %%mm2\n"\
      "psubb " #out1 ", %%mm3\n"\
      "pxor %%mm7, %%mm2\n"\
      "pxor %%mm7, %%mm3\n"\
      "movq %%mm2, " #out0 "\n"\
      "movq %%mm3, " #out1 "\n"\
      "psubusb " #in0 ", %%mm2\n"\
      "psubusb " #in1 ", %%mm3\n"\
      "psubusb " #out0 ", " #in0 "\n"\
      "psubusb " #out1 ", " #in1 "\n"\
      "por %%mm2, " #in0 "\n"\
      "por %%mm3, " #in1 "\n"\
      "movq " #in0 ", %%mm2\n"\
      "movq " #in1 ", %%mm3\n"\
      "punpcklbw %%mm7, " #in0 "\n"\
      "punpcklbw %%mm7, " #in1 "\n"\
      "punpckhbw %%mm7, %%mm2\n"\
      "punpckhbw %%mm7, %%mm3\n"\
      "paddw " #in1 ", " #in0 "\n"\
      "paddw %%mm3, %%mm2\n"\
      "paddw %%mm2, " #in0 "\n"\
      "paddw " #in0 ", %%mm6\n"


  asm volatile (
      "movl %4,%%ecx\n"
      "pxor %%mm6,%%mm6\n"
      "pcmpeqw %%mm7,%%mm7\n"
      "psllw $15, %%mm7\n"
      "packsswb %%mm7, %%mm7\n"
      "movq (%0),%%mm0\n"
      "movq (%1),%%mm2\n"
      "movq 8(%0),%%mm1\n"
      "movq 8(%1),%%mm3\n"
      "add %3,%0\n"
      "add %3,%1\n"
      "subl $2, %%ecx\n"
      "psubb %%mm2, %%mm0\n"
      "psubb %%mm3, %%mm1\n"
      "pxor %%mm7, %%mm0\n"
      "pxor %%mm7, %%mm1\n"
      SUM(%%mm0, %%mm1, %%mm4, %%mm5)
      "1:\n"

      SUM(%%mm4, %%mm5, %%mm0, %%mm1)

      SUM(%%mm0, %%mm1, %%mm4, %%mm5)

      "subl $2, %%ecx\n"
      "jnz 1b\n"

      "movq %%mm6,%%mm0\n"
      "psrlq $32, %%mm6\n"
      "paddw %%mm6,%%mm0\n"
      "movq %%mm0,%%mm6\n"
      "psrlq $16, %%mm0\n"
      "paddw %%mm6,%%mm0\n"
      "movd %%mm0,%2\n"
      : "+r" (pix1), "+r" (pix2), "=r"(tmp)
      : "r" ((long)line_size) , "m" (h)
      : "%ecx");
    return tmp & 0x7FFF;
}
#undef SUM

static int vsad16_mmx2(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
    int tmp;

    assert( (((int)pix1) & 7) == 0);
    assert( (((int)pix2) & 7) == 0);
    assert((line_size &7) ==0);

#define SUM(in0, in1, out0, out1) \
      "movq (%0)," #out0 "\n"\
      "movq (%1),%%mm2\n"\
      "movq 8(%0)," #out1 "\n"\
      "movq 8(%1),%%mm3\n"\
      "add %3,%0\n"\
      "add %3,%1\n"\
      "psubb %%mm2, " #out0 "\n"\
      "psubb %%mm3, " #out1 "\n"\
      "pxor %%mm7, " #out0 "\n"\
      "pxor %%mm7, " #out1 "\n"\
      "psadbw " #out0 ", " #in0 "\n"\
      "psadbw " #out1 ", " #in1 "\n"\
      "paddw " #in1 ", " #in0 "\n"\
      "paddw " #in0 ", %%mm6\n"

  asm volatile (
      "movl %4,%%ecx\n"
      "pxor %%mm6,%%mm6\n"
      "pcmpeqw %%mm7,%%mm7\n"
      "psllw $15, %%mm7\n"
      "packsswb %%mm7, %%mm7\n"
      "movq (%0),%%mm0\n"
      "movq (%1),%%mm2\n"
      "movq 8(%0),%%mm1\n"
      "movq 8(%1),%%mm3\n"
      "add %3,%0\n"
      "add %3,%1\n"
      "subl $2, %%ecx\n"
      "psubb %%mm2, %%mm0\n"
      "psubb %%mm3, %%mm1\n"
      "pxor %%mm7, %%mm0\n"
      "pxor %%mm7, %%mm1\n"
      SUM(%%mm0, %%mm1, %%mm4, %%mm5)
      "1:\n"

      SUM(%%mm4, %%mm5, %%mm0, %%mm1)

      SUM(%%mm0, %%mm1, %%mm4, %%mm5)

      "subl $2, %%ecx\n"
      "jnz 1b\n"

      "movd %%mm6,%2\n"
      : "+r" (pix1), "+r" (pix2), "=r"(tmp)
      : "r" ((long)line_size) , "m" (h)
      : "%ecx");
    return tmp;
}
#undef SUM

static void diff_bytes_mmx(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){
    long i=0;
    asm volatile(
        "1:                             \n\t"
        "movq  (%2, %0), %%mm0          \n\t"
        "movq  (%1, %0), %%mm1          \n\t"
        "psubb %%mm0, %%mm1             \n\t"
        "movq %%mm1, (%3, %0)           \n\t"
        "movq 8(%2, %0), %%mm0          \n\t"
        "movq 8(%1, %0), %%mm1          \n\t"
        "psubb %%mm0, %%mm1             \n\t"
        "movq %%mm1, 8(%3, %0)          \n\t"
        "add $16, %0                    \n\t"
        "cmp %4, %0                     \n\t"
        " jb 1b                         \n\t"
        : "+r" (i)
        : "r"(src1), "r"(src2), "r"(dst), "r"((long)w-15)
    );
    for(; i<w; i++)
        dst[i+0] = src1[i+0]-src2[i+0];
}

static void sub_hfyu_median_prediction_mmx2(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top){
    long i=0;
    uint8_t l, lt;

    asm volatile(
        "1:                             \n\t"
        "movq  -1(%1, %0), %%mm0        \n\t" // LT
        "movq  (%1, %0), %%mm1          \n\t" // T
        "movq  -1(%2, %0), %%mm2        \n\t" // L
        "movq  (%2, %0), %%mm3          \n\t" // X
        "movq %%mm2, %%mm4              \n\t" // L
        "psubb %%mm0, %%mm2             \n\t"
        "paddb %%mm1, %%mm2             \n\t" // L + T - LT
        "movq %%mm4, %%mm5              \n\t" // L
        "pmaxub %%mm1, %%mm4            \n\t" // max(T, L)
        "pminub %%mm5, %%mm1            \n\t" // min(T, L)
        "pminub %%mm2, %%mm4            \n\t"
        "pmaxub %%mm1, %%mm4            \n\t"
        "psubb %%mm4, %%mm3             \n\t" // dst - pred
        "movq %%mm3, (%3, %0)           \n\t"
        "add $8, %0                     \n\t"
        "cmp %4, %0                     \n\t"
        " jb 1b                         \n\t"
        : "+r" (i)
        : "r"(src1), "r"(src2), "r"(dst), "r"((long)w)
    );

    l= *left;
    lt= *left_top;

    dst[0]= src2[0] - mid_pred(l, src1[0], (l + src1[0] - lt)&0xFF);

    *left_top= src1[w-1];
    *left    = src2[w-1];
}

#define DIFF_PIXELS_1(m,a,t,p1,p2)\
    "mov"#m" "#p1", "#a"              \n\t"\
    "mov"#m" "#p2", "#t"              \n\t"\
    "punpcklbw "#a", "#t"             \n\t"\
    "punpcklbw "#a", "#a"             \n\t"\
    "psubw     "#t", "#a"             \n\t"\

#define DIFF_PIXELS_8(m0,m1,mm,p1,p2,stride,temp) {\
    uint8_t *p1b=p1, *p2b=p2;\
    asm volatile(\
        DIFF_PIXELS_1(m0, mm##0, mm##7, (%1), (%2))\
        DIFF_PIXELS_1(m0, mm##1, mm##7, (%1,%3), (%2,%3))\
        DIFF_PIXELS_1(m0, mm##2, mm##7, (%1,%3,2), (%2,%3,2))\
        "add %4, %1                   \n\t"\
        "add %4, %2                   \n\t"\
        DIFF_PIXELS_1(m0, mm##3, mm##7, (%1), (%2))\
        DIFF_PIXELS_1(m0, mm##4, mm##7, (%1,%3), (%2,%3))\
        DIFF_PIXELS_1(m0, mm##5, mm##7, (%1,%3,2), (%2,%3,2))\
        DIFF_PIXELS_1(m0, mm##6, mm##7, (%1,%4), (%2,%4))\
        "mov"#m1" "#mm"0, %0          \n\t"\
        DIFF_PIXELS_1(m0, mm##7, mm##0, (%1,%3,4), (%2,%3,4))\
        "mov"#m1" %0, "#mm"0          \n\t"\
        : "=m"(temp), "+r"(p1b), "+r"(p2b)\
        : "r"((long)stride), "r"((long)stride*3)\
    );\
}

#define DIFF_PIXELS_4x8(p1,p2,stride,temp) DIFF_PIXELS_8(d, q,   %%mm,  p1, p2, stride, temp)
#define DIFF_PIXELS_8x8(p1,p2,stride,temp) DIFF_PIXELS_8(q, dqa, %%xmm, p1, p2, stride, temp)