comparison ppc/dsputil_altivec.c @ 1352:e8ff4783f188 libavcodec

1) remove TBL support in PPC performance. It's much more useful to use the PMCs, and with Apple's CHUD it's fairly easy too. No reason to keep useless code around 2) make the PPC perf stuff a configure option 3) make put_pixels16_altivec a bit faster by unrolling the loop by 4 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)

1351:0fc1a6f8ca94

}

/* next one assumes that ((line_size % 16) == 0) */
void put_pixels16_altivec(uint8_t *block, const uint8_t *pixels, int line_size, int h)
{
POWERPC_TBL_DECLARE(altivec_put_pixels16_num, 1);
#ifdef ALTIVEC_USE_REFERENCE_C_CODE
    int i;

POWERPC_TBL_START_COUNT(altivec_put_pixels16_num, 1);

    for(i=0; i<h; i++) {
      *((uint32_t*)(block )) = (((const struct unaligned_32 *) (pixels))->l);
      *((uint32_t*)(block+4)) = (((const struct unaligned_32 *) (pixels+4))->l);
      *((uint32_t*)(block+8)) = (((const struct unaligned_32 *) (pixels+8))->l);
      *((uint32_t*)(block+12)) = (((const struct unaligned_32 *) (pixels+12))->l);
      pixels+=line_size;
      block +=line_size;
    }

POWERPC_TBL_STOP_COUNT(altivec_put_pixels16_num, 1);

#else /* ALTIVEC_USE_REFERENCE_C_CODE */
    register vector unsigned char pixelsv1, pixelsv2;
    register vector unsigned char perm = vec_lvsl(0, pixels);
    int i;

POWERPC_TBL_START_COUNT(altivec_put_pixels16_num, 1);

    for(i=0; i<h; i++) {
      pixelsv1 = vec_ld(0, (unsigned char*)pixels);
      pixelsv2 = vec_ld(16, (unsigned char*)pixels);
      vec_st(vec_perm(pixelsv1, pixelsv2, perm),
             0, (unsigned char*)block);
      pixels+=line_size;
      block +=line_size;
    }

POWERPC_TBL_STOP_COUNT(altivec_put_pixels16_num, 1);

#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
}

/* next one assumes that ((line_size % 16) == 0) */
#define op_avg(a,b)  a = ( ((a)|(b)) - ((((a)^(b))&0xFEFEFEFEUL)>>1) )
void avg_pixels16_altivec(uint8_t *block, const uint8_t *pixels, int line_size, int h)
{
POWERPC_TBL_DECLARE(altivec_avg_pixels16_num, 1);
#ifdef ALTIVEC_USE_REFERENCE_C_CODE
    int i;

POWERPC_TBL_START_COUNT(altivec_avg_pixels16_num, 1);

    for(i=0; i<h; i++) {
      op_avg(*((uint32_t*)(block)),(((const struct unaligned_32 *)(pixels))->l));
      op_avg(*((uint32_t*)(block+4)),(((const struct unaligned_32 *)(pixels+4))->l));
      op_avg(*((uint32_t*)(block+8)),(((const struct unaligned_32 *)(pixels+8))->l));
      op_avg(*((uint32_t*)(block+12)),(((const struct unaligned_32 *)(pixels+12))->l));
      pixels+=line_size;
      block +=line_size;
    }

POWERPC_TBL_STOP_COUNT(altivec_avg_pixels16_num, 1);

#else /* ALTIVEC_USE_REFERENCE_C_CODE */
    register vector unsigned char pixelsv1, pixelsv2, pixelsv, blockv;
    register vector unsigned char perm = vec_lvsl(0, pixels);
    int i;

POWERPC_TBL_START_COUNT(altivec_avg_pixels16_num, 1);

    for(i=0; i<h; i++) {
      pixelsv1 = vec_ld(0, (unsigned char*)pixels);
      pixelsv2 = vec_ld(16, (unsigned char*)pixels);
      blockv = vec_ld(0, block);

      vec_st(blockv, 0, (unsigned char*)block);
      pixels+=line_size;
      block +=line_size;
    }

POWERPC_TBL_STOP_COUNT(altivec_avg_pixels16_num, 1);

#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
}

/* next one assumes that ((line_size % 8) == 0) */
void avg_pixels8_altivec(uint8_t * block, const uint8_t * pixels, int line_size, int h)
{
POWERPC_TBL_DECLARE(altivec_avg_pixels8_num, 1);
#ifdef ALTIVEC_USE_REFERENCE_C_CODE
    int i;
POWERPC_TBL_START_COUNT(altivec_avg_pixels8_num, 1);
    for (i = 0; i < h; i++) {
        *((uint32_t *) (block)) =
            (((*((uint32_t *) (block))) |
              ((((const struct unaligned_32 *) (pixels))->l))) -
             ((((*((uint32_t *) (block))) ^

                                                  4))->
                  l))) & 0xFEFEFEFEUL) >> 1));
        pixels += line_size;
        block += line_size;
    }
POWERPC_TBL_STOP_COUNT(altivec_avg_pixels8_num, 1);

#else /* ALTIVEC_USE_REFERENCE_C_CODE */
    register vector unsigned char pixelsv1, pixelsv2, pixelsv, blockv;
    int i;

POWERPC_TBL_START_COUNT(altivec_avg_pixels8_num, 1);
 
   for (i = 0; i < h; i++) {
     /*
       block is 8 bytes-aligned, so we're either in the
       left block (16 bytes-aligned) or in the right block (not)

     
     pixels += line_size;
     block += line_size;
   }
   
POWERPC_TBL_STOP_COUNT(altivec_avg_pixels8_num, 1);
 
#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
}

/* next one assumes that ((line_size % 8) == 0) */
void put_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, int line_size, int h)
{
POWERPC_TBL_DECLARE(altivec_put_pixels8_xy2_num, 1);
#ifdef ALTIVEC_USE_REFERENCE_C_CODE
    int j;
POWERPC_TBL_START_COUNT(altivec_put_pixels8_xy2_num, 1);
    for (j = 0; j < 2; j++) {
      int i;
      const uint32_t a = (((const struct unaligned_32 *) (pixels))->l);
      const uint32_t b =
        (((const struct unaligned_32 *) (pixels + 1))->l);

        block += line_size;
      } pixels += 4 - line_size * (h + 1);
      block += 4 - line_size * h;
    }

POWERPC_TBL_STOP_COUNT(altivec_put_pixels8_xy2_num, 1);

#else /* ALTIVEC_USE_REFERENCE_C_CODE */
   register int i;
   register vector unsigned char
     pixelsv1, pixelsv2,

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

     
     block += line_size;
     pixels += line_size;
   }
   
POWERPC_TBL_STOP_COUNT(altivec_put_pixels8_xy2_num, 1);
#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
}

/* 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_TBL_DECLARE(altivec_put_no_rnd_pixels8_xy2_num, 1);
#ifdef ALTIVEC_USE_REFERENCE_C_CODE
    int j;
POWERPC_TBL_START_COUNT(altivec_put_no_rnd_pixels8_xy2_num, 1);
    for (j = 0; j < 2; j++) {
      int i;
      const uint32_t a = (((const struct unaligned_32 *) (pixels))->l);
      const uint32_t b =
        (((const struct unaligned_32 *) (pixels + 1))->l);

        block += line_size;
      } pixels += 4 - line_size * (h + 1);
      block += 4 - line_size * h;
    }
    
POWERPC_TBL_STOP_COUNT(altivec_put_no_rnd_pixels8_xy2_num, 1);

#else /* ALTIVEC_USE_REFERENCE_C_CODE */
   register int i;
   register vector unsigned char
     pixelsv1, pixelsv2,

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

     
     block += line_size;
     pixels += line_size;
   }
   
POWERPC_TBL_STOP_COUNT(altivec_put_no_rnd_pixels8_xy2_num, 1);
#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
}

/* 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_TBL_DECLARE(altivec_put_pixels16_xy2_num, 1);
#ifdef ALTIVEC_USE_REFERENCE_C_CODE
    int j;
POWERPC_TBL_START_COUNT(altivec_put_pixels16_xy2_num, 1);
      for (j = 0; j < 4; j++) {
      int i;
      const uint32_t a = (((const struct unaligned_32 *) (pixels))->l);
      const uint32_t b =
        (((const struct unaligned_32 *) (pixels + 1))->l);

        block += line_size;
      } pixels += 4 - line_size * (h + 1);
      block += 4 - line_size * h;
    }

POWERPC_TBL_STOP_COUNT(altivec_put_pixels16_xy2_num, 1);

#else /* ALTIVEC_USE_REFERENCE_C_CODE */
   register int i;
   register vector unsigned char
     pixelsv1, pixelsv2, pixelsv3, pixelsv4;

     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_TBL_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)

     
     block += line_size;
     pixels += line_size;
   }
   
POWERPC_TBL_STOP_COUNT(altivec_put_pixels16_xy2_num, 1);
#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
}

/* 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_TBL_DECLARE(altivec_put_no_rnd_pixels16_xy2_num, 1);
#ifdef ALTIVEC_USE_REFERENCE_C_CODE
    int j;
POWERPC_TBL_START_COUNT(altivec_put_no_rnd_pixels16_xy2_num, 1);
      for (j = 0; j < 4; j++) {
      int i;
      const uint32_t a = (((const struct unaligned_32 *) (pixels))->l);
      const uint32_t b =
        (((const struct unaligned_32 *) (pixels + 1))->l);

        block += line_size;
      } pixels += 4 - line_size * (h + 1);
      block += 4 - line_size * h;
    }

POWERPC_TBL_STOP_COUNT(altivec_put_no_rnd_pixels16_xy2_num, 1);

#else /* ALTIVEC_USE_REFERENCE_C_CODE */
   register int i;
   register vector unsigned char
     pixelsv1, pixelsv2, pixelsv3, pixelsv4;

     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_TBL_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)

     
     block += line_size;
     pixels += line_size;
   }
   
POWERPC_TBL_STOP_COUNT(altivec_put_no_rnd_pixels16_xy2_num, 1);
#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
}

int has_altivec(void)
{

1352:e8ff4783f188

}

/* next one assumes that ((line_size % 16) == 0) */
void put_pixels16_altivec(uint8_t *block, const uint8_t *pixels, int line_size, int h)
{
POWERPC_PERF_DECLARE(altivec_put_pixels16_num, 1);
#ifdef ALTIVEC_USE_REFERENCE_C_CODE
    int i;

POWERPC_PERF_START_COUNT(altivec_put_pixels16_num, 1);

    for(i=0; i<h; i++) {
      *((uint32_t*)(block )) = (((const struct unaligned_32 *) (pixels))->l);
      *((uint32_t*)(block+4)) = (((const struct unaligned_32 *) (pixels+4))->l);
      *((uint32_t*)(block+8)) = (((const struct unaligned_32 *) (pixels+8))->l);
      *((uint32_t*)(block+12)) = (((const struct unaligned_32 *) (pixels+12))->l);
      pixels+=line_size;
      block +=line_size;
    }

POWERPC_PERF_STOP_COUNT(altivec_put_pixels16_num, 1);

#else /* ALTIVEC_USE_REFERENCE_C_CODE */
    register vector unsigned char pixelsv1, pixelsv2;
    register vector unsigned char pixelsv1B, pixelsv2B;
    register vector unsigned char pixelsv1C, pixelsv2C;
    register vector unsigned char pixelsv1D, pixelsv2D;

    register vector unsigned char perm = vec_lvsl(0, pixels);
    int i;
    register int line_size_2 = line_size << 1;
    register int line_size_3 = line_size + line_size_2;
    register int line_size_4 = line_size << 2;

POWERPC_PERF_START_COUNT(altivec_put_pixels16_num, 1);
// hand-unrolling the loop by 4 gains about 15%
// mininum execution time goes from 74 to 60 cycles
// it's faster than -funroll-loops, but using
// -funroll-loops w/ this is bad - 74 cycles again.
// all this is on a 7450, tuning for the 7450
#if 0
    for(i=0; i<h; i++) {
      pixelsv1 = vec_ld(0, (unsigned char*)pixels);
      pixelsv2 = vec_ld(16, (unsigned char*)pixels);
      vec_st(vec_perm(pixelsv1, pixelsv2, perm),
             0, (unsigned char*)block);
      pixels+=line_size;
      block +=line_size;
    }
#else
    for(i=0; i<h; i+=4) {
      pixelsv1 = vec_ld(0, (unsigned char*)pixels);
      pixelsv2 = vec_ld(16, (unsigned char*)pixels);
      pixelsv1B = vec_ld(line_size, (unsigned char*)pixels);
      pixelsv2B = vec_ld(16 + line_size, (unsigned char*)pixels);
      pixelsv1C = vec_ld(line_size_2, (unsigned char*)pixels);
      pixelsv2C = vec_ld(16 + line_size_2, (unsigned char*)pixels);
      pixelsv1D = vec_ld(line_size_3, (unsigned char*)pixels);
      pixelsv2D = vec_ld(16 + line_size_3, (unsigned char*)pixels);
      vec_st(vec_perm(pixelsv1, pixelsv2, perm),
             0, (unsigned char*)block);
      vec_st(vec_perm(pixelsv1B, pixelsv2B, perm),
             line_size, (unsigned char*)block);
      vec_st(vec_perm(pixelsv1C, pixelsv2C, perm),
             line_size_2, (unsigned char*)block);
      vec_st(vec_perm(pixelsv1D, pixelsv2D, perm),
             line_size_3, (unsigned char*)block);
      pixels+=line_size_4;
      block +=line_size_4;
    }
#endif
POWERPC_PERF_STOP_COUNT(altivec_put_pixels16_num, 1);

#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
}

/* next one assumes that ((line_size % 16) == 0) */
#define op_avg(a,b)  a = ( ((a)|(b)) - ((((a)^(b))&0xFEFEFEFEUL)>>1) )
void avg_pixels16_altivec(uint8_t *block, const uint8_t *pixels, int line_size, int h)
{
POWERPC_PERF_DECLARE(altivec_avg_pixels16_num, 1);
#ifdef ALTIVEC_USE_REFERENCE_C_CODE
    int i;

POWERPC_PERF_START_COUNT(altivec_avg_pixels16_num, 1);

    for(i=0; i<h; i++) {
      op_avg(*((uint32_t*)(block)),(((const struct unaligned_32 *)(pixels))->l));
      op_avg(*((uint32_t*)(block+4)),(((const struct unaligned_32 *)(pixels+4))->l));
      op_avg(*((uint32_t*)(block+8)),(((const struct unaligned_32 *)(pixels+8))->l));
      op_avg(*((uint32_t*)(block+12)),(((const struct unaligned_32 *)(pixels+12))->l));
      pixels+=line_size;
      block +=line_size;
    }

POWERPC_PERF_STOP_COUNT(altivec_avg_pixels16_num, 1);

#else /* ALTIVEC_USE_REFERENCE_C_CODE */
    register vector unsigned char pixelsv1, pixelsv2, pixelsv, blockv;
    register vector unsigned char perm = vec_lvsl(0, pixels);
    int i;

POWERPC_PERF_START_COUNT(altivec_avg_pixels16_num, 1);

    for(i=0; i<h; i++) {
      pixelsv1 = vec_ld(0, (unsigned char*)pixels);
      pixelsv2 = vec_ld(16, (unsigned char*)pixels);
      blockv = vec_ld(0, block);

      vec_st(blockv, 0, (unsigned char*)block);
      pixels+=line_size;
      block +=line_size;
    }

POWERPC_PERF_STOP_COUNT(altivec_avg_pixels16_num, 1);

#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
}

/* next one assumes that ((line_size % 8) == 0) */
void avg_pixels8_altivec(uint8_t * block, const uint8_t * pixels, int line_size, int h)
{
POWERPC_PERF_DECLARE(altivec_avg_pixels8_num, 1);
#ifdef ALTIVEC_USE_REFERENCE_C_CODE
    int i;
POWERPC_PERF_START_COUNT(altivec_avg_pixels8_num, 1);
    for (i = 0; i < h; i++) {
        *((uint32_t *) (block)) =
            (((*((uint32_t *) (block))) |
              ((((const struct unaligned_32 *) (pixels))->l))) -
             ((((*((uint32_t *) (block))) ^

                                                  4))->
                  l))) & 0xFEFEFEFEUL) >> 1));
        pixels += line_size;
        block += line_size;
    }
POWERPC_PERF_STOP_COUNT(altivec_avg_pixels8_num, 1);

#else /* ALTIVEC_USE_REFERENCE_C_CODE */
    register vector unsigned char pixelsv1, pixelsv2, pixelsv, blockv;
    int i;

POWERPC_PERF_START_COUNT(altivec_avg_pixels8_num, 1);
 
   for (i = 0; i < h; i++) {
     /*
       block is 8 bytes-aligned, so we're either in the
       left block (16 bytes-aligned) or in the right block (not)

     
     pixels += line_size;
     block += line_size;
   }
   
POWERPC_PERF_STOP_COUNT(altivec_avg_pixels8_num, 1);
 
#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
}

/* next one assumes that ((line_size % 8) == 0) */
void put_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, int line_size, int h)
{
POWERPC_PERF_DECLARE(altivec_put_pixels8_xy2_num, 1);
#ifdef ALTIVEC_USE_REFERENCE_C_CODE
    int j;
POWERPC_PERF_START_COUNT(altivec_put_pixels8_xy2_num, 1);
    for (j = 0; j < 2; j++) {
      int i;
      const uint32_t a = (((const struct unaligned_32 *) (pixels))->l);
      const uint32_t b =
        (((const struct unaligned_32 *) (pixels + 1))->l);

        block += line_size;
      } pixels += 4 - line_size * (h + 1);
      block += 4 - line_size * h;
    }

POWERPC_PERF_STOP_COUNT(altivec_put_pixels8_xy2_num, 1);

#else /* ALTIVEC_USE_REFERENCE_C_CODE */
   register int i;
   register vector unsigned char
     pixelsv1, pixelsv2,

   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);

     
     block += line_size;
     pixels += line_size;
   }
   
POWERPC_PERF_STOP_COUNT(altivec_put_pixels8_xy2_num, 1);
#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
}

/* 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);
#ifdef ALTIVEC_USE_REFERENCE_C_CODE
    int j;
POWERPC_PERF_START_COUNT(altivec_put_no_rnd_pixels8_xy2_num, 1);
    for (j = 0; j < 2; j++) {
      int i;
      const uint32_t a = (((const struct unaligned_32 *) (pixels))->l);
      const uint32_t b =
        (((const struct unaligned_32 *) (pixels + 1))->l);

        block += line_size;
      } pixels += 4 - line_size * (h + 1);
      block += 4 - line_size * h;
    }
    
POWERPC_PERF_STOP_COUNT(altivec_put_no_rnd_pixels8_xy2_num, 1);

#else /* ALTIVEC_USE_REFERENCE_C_CODE */
   register int i;
   register vector unsigned char
     pixelsv1, pixelsv2,

   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);

     
     block += line_size;
     pixels += line_size;
   }
   
POWERPC_PERF_STOP_COUNT(altivec_put_no_rnd_pixels8_xy2_num, 1);
#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
}

/* 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);
#ifdef ALTIVEC_USE_REFERENCE_C_CODE
    int j;
POWERPC_PERF_START_COUNT(altivec_put_pixels16_xy2_num, 1);
      for (j = 0; j < 4; j++) {
      int i;
      const uint32_t a = (((const struct unaligned_32 *) (pixels))->l);
      const uint32_t b =
        (((const struct unaligned_32 *) (pixels + 1))->l);

        block += line_size;
      } pixels += 4 - line_size * (h + 1);
      block += 4 - line_size * h;
    }

POWERPC_PERF_STOP_COUNT(altivec_put_pixels16_xy2_num, 1);

#else /* ALTIVEC_USE_REFERENCE_C_CODE */
   register int i;
   register vector unsigned char
     pixelsv1, pixelsv2, pixelsv3, pixelsv4;

     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)

     
     block += line_size;
     pixels += line_size;
   }
   
POWERPC_PERF_STOP_COUNT(altivec_put_pixels16_xy2_num, 1);
#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
}

/* 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);
#ifdef ALTIVEC_USE_REFERENCE_C_CODE
    int j;
POWERPC_PERF_START_COUNT(altivec_put_no_rnd_pixels16_xy2_num, 1);
      for (j = 0; j < 4; j++) {
      int i;
      const uint32_t a = (((const struct unaligned_32 *) (pixels))->l);
      const uint32_t b =
        (((const struct unaligned_32 *) (pixels + 1))->l);

        block += line_size;
      } pixels += 4 - line_size * (h + 1);
      block += 4 - line_size * h;
    }

POWERPC_PERF_STOP_COUNT(altivec_put_no_rnd_pixels16_xy2_num, 1);

#else /* ALTIVEC_USE_REFERENCE_C_CODE */
   register int i;
   register vector unsigned char
     pixelsv1, pixelsv2, pixelsv3, pixelsv4;

     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)

     
     block += line_size;
     pixels += line_size;
   }
   
POWERPC_PERF_STOP_COUNT(altivec_put_no_rnd_pixels16_xy2_num, 1);
#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
}

int has_altivec(void)
{