dsputil_altivec.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

   pixelsv2 = vec_mergeh(vczero, pixelsv2);
   pixelssum1 = vec_add((vector unsigned short)pixelsv1,
                        (vector unsigned short)pixelsv2);
   pixelssum1 = vec_add(pixelssum1, vctwo);

POWERPC_PERF_START_COUNT(altivec_put_pixels8_xy2_num, 1);
   for (i = 0; i < h ; i++) {
     int rightside = ((unsigned long)block & 0x0000000F);
     blockv = vec_ld(0, block);

     temp1 = vec_ld(line_size, pixels);
     temp2 = vec_ld(line_size + 16, pixels);
     pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(line_size, pixels));
     if (((((unsigned long)pixels) + line_size) & 0x0000000F) ==  0x0000000F)
     {
       pixelsv2 = temp2;
     }
     else
     {
       pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(line_size + 1, pixels));
     }

     pixelsv1 = vec_mergeh(vczero, pixelsv1);
     pixelsv2 = vec_mergeh(vczero, pixelsv2);
     pixelssum2 = vec_add((vector unsigned short)pixelsv1,
                          (vector unsigned short)pixelsv2);
     temp3 = vec_add(pixelssum1, pixelssum2);
     temp3 = vec_sra(temp3, vctwo);
     pixelssum1 = vec_add(pixelssum2, vctwo);
     pixelsavg = vec_packsu(temp3, (vector unsigned short) vczero);

     if (rightside)
     {
       blockv = vec_perm(blockv, pixelsavg, vcprm(0, 1, s0, s1));
     }
     else
     {
       blockv = vec_perm(blockv, pixelsavg, vcprm(s0, s1, 2, 3));
     }

     vec_st(blockv, 0, block);

     block += line_size;
     pixels += line_size;
   }

POWERPC_PERF_STOP_COUNT(altivec_put_pixels8_xy2_num, 1);
}

/* next one assumes that ((line_size % 8) == 0) */
void put_no_rnd_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, int line_size, int h)
{
POWERPC_PERF_DECLARE(altivec_put_no_rnd_pixels8_xy2_num, 1);
   register int i;
   register vector unsigned char
     pixelsv1, pixelsv2,
     pixelsavg;
   register vector unsigned char
     blockv, temp1, temp2;
   register vector unsigned short
     pixelssum1, pixelssum2, temp3;
   register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0);
   register const vector unsigned short vcone = (const vector unsigned short)vec_splat_u16(1);
   register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2);

   temp1 = vec_ld(0, pixels);
   temp2 = vec_ld(16, pixels);
   pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(0, pixels));
   if ((((unsigned long)pixels) & 0x0000000F) ==  0x0000000F)
   {
     pixelsv2 = temp2;
   }
   else
   {
     pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(1, pixels));
   }
   pixelsv1 = vec_mergeh(vczero, pixelsv1);
   pixelsv2 = vec_mergeh(vczero, pixelsv2);
   pixelssum1 = vec_add((vector unsigned short)pixelsv1,
                        (vector unsigned short)pixelsv2);
   pixelssum1 = vec_add(pixelssum1, vcone);

POWERPC_PERF_START_COUNT(altivec_put_no_rnd_pixels8_xy2_num, 1);
   for (i = 0; i < h ; i++) {
     int rightside = ((unsigned long)block & 0x0000000F);
     blockv = vec_ld(0, block);

     temp1 = vec_ld(line_size, pixels);
     temp2 = vec_ld(line_size + 16, pixels);
     pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(line_size, pixels));
     if (((((unsigned long)pixels) + line_size) & 0x0000000F) ==  0x0000000F)
     {
       pixelsv2 = temp2;
     }
     else
     {
       pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(line_size + 1, pixels));
     }

     pixelsv1 = vec_mergeh(vczero, pixelsv1);
     pixelsv2 = vec_mergeh(vczero, pixelsv2);
     pixelssum2 = vec_add((vector unsigned short)pixelsv1,
                          (vector unsigned short)pixelsv2);
     temp3 = vec_add(pixelssum1, pixelssum2);
     temp3 = vec_sra(temp3, vctwo);
     pixelssum1 = vec_add(pixelssum2, vcone);
     pixelsavg = vec_packsu(temp3, (vector unsigned short) vczero);

     if (rightside)
     {
       blockv = vec_perm(blockv, pixelsavg, vcprm(0, 1, s0, s1));
     }
     else
     {
       blockv = vec_perm(blockv, pixelsavg, vcprm(s0, s1, 2, 3));
     }

     vec_st(blockv, 0, block);

     block += line_size;
     pixels += line_size;
   }

POWERPC_PERF_STOP_COUNT(altivec_put_no_rnd_pixels8_xy2_num, 1);
}

/* next one assumes that ((line_size % 16) == 0) */
void put_pixels16_xy2_altivec(uint8_t * block, const uint8_t * pixels, int line_size, int h)
{
POWERPC_PERF_DECLARE(altivec_put_pixels16_xy2_num, 1);
   register int i;
   register vector unsigned char
     pixelsv1, pixelsv2, pixelsv3, pixelsv4;
   register vector unsigned char
     blockv, temp1, temp2;
   register vector unsigned short
     pixelssum1, pixelssum2, temp3,
     pixelssum3, pixelssum4, temp4;
   register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0);
   register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2);

POWERPC_PERF_START_COUNT(altivec_put_pixels16_xy2_num, 1);

   temp1 = vec_ld(0, pixels);
   temp2 = vec_ld(16, pixels);
   pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(0, pixels));
   if ((((unsigned long)pixels) & 0x0000000F) ==  0x0000000F)
   {
     pixelsv2 = temp2;
   }
   else
   {
     pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(1, pixels));
   }
   pixelsv3 = vec_mergel(vczero, pixelsv1);
   pixelsv4 = vec_mergel(vczero, pixelsv2);
   pixelsv1 = vec_mergeh(vczero, pixelsv1);
   pixelsv2 = vec_mergeh(vczero, pixelsv2);
   pixelssum3 = vec_add((vector unsigned short)pixelsv3,
                        (vector unsigned short)pixelsv4);
   pixelssum3 = vec_add(pixelssum3, vctwo);
   pixelssum1 = vec_add((vector unsigned short)pixelsv1,
                        (vector unsigned short)pixelsv2);
   pixelssum1 = vec_add(pixelssum1, vctwo);

   for (i = 0; i < h ; i++) {
     blockv = vec_ld(0, block);

     temp1 = vec_ld(line_size, pixels);
     temp2 = vec_ld(line_size + 16, pixels);
     pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(line_size, pixels));
     if (((((unsigned long)pixels) + line_size) & 0x0000000F) ==  0x0000000F)
     {
       pixelsv2 = temp2;
     }
     else
     {
       pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(line_size + 1, pixels));
     }

     pixelsv3 = vec_mergel(vczero, pixelsv1);
     pixelsv4 = vec_mergel(vczero, pixelsv2);
     pixelsv1 = vec_mergeh(vczero, pixelsv1);
     pixelsv2 = vec_mergeh(vczero, pixelsv2);

     pixelssum4 = vec_add((vector unsigned short)pixelsv3,
                          (vector unsigned short)pixelsv4);
     pixelssum2 = vec_add((vector unsigned short)pixelsv1,
                          (vector unsigned short)pixelsv2);
     temp4 = vec_add(pixelssum3, pixelssum4);
     temp4 = vec_sra(temp4, vctwo);
     temp3 = vec_add(pixelssum1, pixelssum2);
     temp3 = vec_sra(temp3, vctwo);

     pixelssum3 = vec_add(pixelssum4, vctwo);
     pixelssum1 = vec_add(pixelssum2, vctwo);

     blockv = vec_packsu(temp3, temp4);

     vec_st(blockv, 0, block);

     block += line_size;
     pixels += line_size;
   }

POWERPC_PERF_STOP_COUNT(altivec_put_pixels16_xy2_num, 1);
}

/* next one assumes that ((line_size % 16) == 0) */
void put_no_rnd_pixels16_xy2_altivec(uint8_t * block, const uint8_t * pixels, int line_size, int h)
{
POWERPC_PERF_DECLARE(altivec_put_no_rnd_pixels16_xy2_num, 1);
   register int i;
   register vector unsigned char
     pixelsv1, pixelsv2, pixelsv3, pixelsv4;
   register vector unsigned char
     blockv, temp1, temp2;
   register vector unsigned short
     pixelssum1, pixelssum2, temp3,
     pixelssum3, pixelssum4, temp4;
   register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0);
   register const vector unsigned short vcone = (const vector unsigned short)vec_splat_u16(1);
   register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2);

POWERPC_PERF_START_COUNT(altivec_put_no_rnd_pixels16_xy2_num, 1);

   temp1 = vec_ld(0, pixels);
   temp2 = vec_ld(16, pixels);
   pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(0, pixels));
   if ((((unsigned long)pixels) & 0x0000000F) ==  0x0000000F)
   {
     pixelsv2 = temp2;
   }
   else
   {
     pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(1, pixels));
   }
   pixelsv3 = vec_mergel(vczero, pixelsv1);
   pixelsv4 = vec_mergel(vczero, pixelsv2);
   pixelsv1 = vec_mergeh(vczero, pixelsv1);
   pixelsv2 = vec_mergeh(vczero, pixelsv2);
   pixelssum3 = vec_add((vector unsigned short)pixelsv3,
                        (vector unsigned short)pixelsv4);
   pixelssum3 = vec_add(pixelssum3, vcone);
   pixelssum1 = vec_add((vector unsigned short)pixelsv1,
                        (vector unsigned short)pixelsv2);
   pixelssum1 = vec_add(pixelssum1, vcone);

   for (i = 0; i < h ; i++) {
     blockv = vec_ld(0, block);

     temp1 = vec_ld(line_size, pixels);
     temp2 = vec_ld(line_size + 16, pixels);
     pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(line_size, pixels));
     if (((((unsigned long)pixels) + line_size) & 0x0000000F) ==  0x0000000F)
     {
       pixelsv2 = temp2;
     }
     else
     {
       pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(line_size + 1, pixels));
     }

     pixelsv3 = vec_mergel(vczero, pixelsv1);
     pixelsv4 = vec_mergel(vczero, pixelsv2);
     pixelsv1 = vec_mergeh(vczero, pixelsv1);
     pixelsv2 = vec_mergeh(vczero, pixelsv2);

     pixelssum4 = vec_add((vector unsigned short)pixelsv3,
                          (vector unsigned short)pixelsv4);
     pixelssum2 = vec_add((vector unsigned short)pixelsv1,
                          (vector unsigned short)pixelsv2);
     temp4 = vec_add(pixelssum3, pixelssum4);
     temp4 = vec_sra(temp4, vctwo);
     temp3 = vec_add(pixelssum1, pixelssum2);
     temp3 = vec_sra(temp3, vctwo);

     pixelssum3 = vec_add(pixelssum4, vcone);
     pixelssum1 = vec_add(pixelssum2, vcone);

     blockv = vec_packsu(temp3, temp4);

     vec_st(blockv, 0, block);

     block += line_size;
     pixels += line_size;
   }

POWERPC_PERF_STOP_COUNT(altivec_put_no_rnd_pixels16_xy2_num, 1);
}

int hadamard8_diff8x8_altivec(/*MpegEncContext*/ void *s, uint8_t *dst, uint8_t *src, int stride, int h){
POWERPC_PERF_DECLARE(altivec_hadamard8_diff8x8_num, 1);
    int sum;
    register const vector unsigned char vzero =
                            (const vector unsigned char)vec_splat_u8(0);
    register vector signed short temp0, temp1, temp2, temp3, temp4,
                                 temp5, temp6, temp7;
POWERPC_PERF_START_COUNT(altivec_hadamard8_diff8x8_num, 1);
  {
    register const vector signed short vprod1 =(const vector signed short)
                                        AVV( 1,-1, 1,-1, 1,-1, 1,-1);
    register const vector signed short vprod2 =(const vector signed short)
                                        AVV( 1, 1,-1,-1, 1, 1,-1,-1);
    register const vector signed short vprod3 =(const vector signed short)
                                        AVV( 1, 1, 1, 1,-1,-1,-1,-1);
    register const vector unsigned char perm1 = (const vector unsigned char)
      AVV(0x02, 0x03, 0x00, 0x01, 0x06, 0x07, 0x04, 0x05,
          0x0A, 0x0B, 0x08, 0x09, 0x0E, 0x0F, 0x0C, 0x0D);
    register const vector unsigned char perm2 = (const vector unsigned char)
      AVV(0x04, 0x05, 0x06, 0x07, 0x00, 0x01, 0x02, 0x03,
          0x0C, 0x0D, 0x0E, 0x0F, 0x08, 0x09, 0x0A, 0x0B);
    register const vector unsigned char perm3 = (const vector unsigned char)
      AVV(0x08, 0x09, 0x0A, 0x0B, 0x0C, 0x0D, 0x0E, 0x0F,
          0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07);

#define ONEITERBUTTERFLY(i, res)                                        \
    {                                                                   \
      register vector unsigned char src1, src2, srcO;                   \
      register vector unsigned char dst1, dst2, dstO;                   \
      register vector signed short srcV, dstV;                          \
      register vector signed short but0, but1, but2, op1, op2, op3;     \
      src1 = vec_ld(stride * i, src);                                   \
      src2 = vec_ld((stride * i) + 15, src);                            \
      srcO = vec_perm(src1, src2, vec_lvsl(stride * i, src));           \
      dst1 = vec_ld(stride * i, dst);                                   \
      dst2 = vec_ld((stride * i) + 15, dst);                            \
      dstO = vec_perm(dst1, dst2, vec_lvsl(stride * i, dst));           \
      /* promote the unsigned chars to signed shorts */                 \
      /* we're in the 8x8 function, we only care for the first 8 */     \
      srcV =                                                            \
        (vector signed short)vec_mergeh((vector signed char)vzero,      \
        (vector signed char)srcO);                                      \
      dstV =                                                            \
        (vector signed short)vec_mergeh((vector signed char)vzero,      \
        (vector signed char)dstO);                                      \
      /* substractions inside the first butterfly */                    \
      but0 = vec_sub(srcV, dstV);                                       \
      op1 = vec_perm(but0, but0, perm1);                                \
      but1 = vec_mladd(but0, vprod1, op1);                              \
      op2 = vec_perm(but1, but1, perm2);                                \
      but2 = vec_mladd(but1, vprod2, op2);                              \
      op3 = vec_perm(but2, but2, perm3);                                \
      res = vec_mladd(but2, vprod3, op3);                               \
    }
    ONEITERBUTTERFLY(0, temp0);
    ONEITERBUTTERFLY(1, temp1);
    ONEITERBUTTERFLY(2, temp2);
    ONEITERBUTTERFLY(3, temp3);
    ONEITERBUTTERFLY(4, temp4);
    ONEITERBUTTERFLY(5, temp5);
    ONEITERBUTTERFLY(6, temp6);
    ONEITERBUTTERFLY(7, temp7);
  }
#undef ONEITERBUTTERFLY
  {
    register vector signed int vsum;
    register vector signed short line0 = vec_add(temp0, temp1);
    register vector signed short line1 = vec_sub(temp0, temp1);
    register vector signed short line2 = vec_add(temp2, temp3);
    register vector signed short line3 = vec_sub(temp2, temp3);
    register vector signed short line4 = vec_add(temp4, temp5);
    register vector signed short line5 = vec_sub(temp4, temp5);
    register vector signed short line6 = vec_add(temp6, temp7);
    register vector signed short line7 = vec_sub(temp6, temp7);

    register vector signed short line0B = vec_add(line0, line2);
    register vector signed short line2B = vec_sub(line0, line2);
    register vector signed short line1B = vec_add(line1, line3);
    register vector signed short line3B = vec_sub(line1, line3);
    register vector signed short line4B = vec_add(line4, line6);
    register vector signed short line6B = vec_sub(line4, line6);
    register vector signed short line5B = vec_add(line5, line7);
    register vector signed short line7B = vec_sub(line5, line7);

    register vector signed short line0C = vec_add(line0B, line4B);
    register vector signed short line4C = vec_sub(line0B, line4B);
    register vector signed short line1C = vec_add(line1B, line5B);
    register vector signed short line5C = vec_sub(line1B, line5B);
    register vector signed short line2C = vec_add(line2B, line6B);