comparison libswscale/internal_bfin.S @ 27156:23f1738030fc

whitespace cosmetics

27155:d5d7ecbbf0d8

 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */


/*
    YUV420 to RGB565 conversion.  This routine takes a YUV 420 planar macroblock
    and converts it to RGB565.  R:5 bits, G:6 bits, B:5 bits.. packed into shorts


    The following calculation is used for the conversion:

      r = clipz((y-oy)*cy  + crv*(v-128))
      g = clipz((y-oy)*cy  + cgv*(v-128) + cgu*(u-128))
      b = clipz((y-oy)*cy  + cbu*(u-128))

    y,u,v are pre scaled by a factor of 4 i.e. left shifted to gain precision.


    New factorization to eliminate the truncation error which was
    occuring due to the byteop3p.


  1) use the bytop16m to subtract quad bytes we use this in U8 this
   then so the offsets need to be renormalized to 8bits.

  2) scale operands up by a factor of 4 not 8 because Blackfin
     multiplies include a shift.

  3) compute into the accumulators cy*yx0, cy*yx1

  4) compute each of the linear equations
      r = clipz((y-oy)*cy  + crv*(v-128))

      g = clipz((y-oy)*cy  + cgv*(v-128) + cgu*(u-128))

      b = clipz((y-oy)*cy  + cbu*(u-128))

     reuse of the accumulators requires that we actually multiply
     twice once with addition and the second time with a subtaction.

     because of this we need to compute the equations in the order R B
     then G saving the writes for B in the case of 24/32 bit color
     formats.

    api: yuv2rgb_kind (uint8_t *Y, uint8_t *U, uint8_t *V, int *out,
                       int dW, uint32_t *coeffs);

        A          B
        ---        ---
        i2 = cb    i3 = cr
        i1 = coeff i0 = y

  Where coeffs have the following layout in memory.

  uint32_t oy,oc,zero,cy,crv,rmask,cbu,bmask,cgu,cgv;

  coeffs is a pointer to oy.

  the {rgb} masks are only utilized by the 565 packing algorithm. Note the data
  replication is used to simplify the internal algorithms for the dual mac architecture
  of BlackFin.

  All routines are exported with _ff_bfin_ as a symbol prefix

  rough performance gain compared against -O3:

  2779809/1484290 187.28%

  which translates to ~33c/pel to ~57c/pel for the reference vs 17.5
  c/pel for the optimized implementations. Not sure why there is such a
  huge variation on the reference codes on Blackfin I guess it must have
  to do with the memory system.

*/

#define mL3 .text
#ifdef __FDPIC__
#define mL1 .l1.text

27156:23f1738030fc

 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */


/*
YUV420 to RGB565 conversion.  This routine takes a YUV 420 planar macroblock
and converts it to RGB565.  R:5 bits, G:6 bits, B:5 bits.. packed into shorts


The following calculation is used for the conversion:

  r = clipz((y-oy)*cy  + crv*(v-128))
  g = clipz((y-oy)*cy  + cgv*(v-128) + cgu*(u-128))
  b = clipz((y-oy)*cy  + cbu*(u-128))

y,u,v are pre scaled by a factor of 4 i.e. left shifted to gain precision.


New factorization to eliminate the truncation error which was
occuring due to the byteop3p.


1) use the bytop16m to subtract quad bytes we use this in U8 this
 then so the offsets need to be renormalized to 8bits.

2) scale operands up by a factor of 4 not 8 because Blackfin
   multiplies include a shift.

3) compute into the accumulators cy*yx0, cy*yx1

4) compute each of the linear equations
     r = clipz((y - oy) * cy  + crv * (v - 128))

     g = clipz((y - oy) * cy  + cgv * (v - 128) + cgu * (u - 128))

     b = clipz((y - oy) * cy  + cbu * (u - 128))

   reuse of the accumulators requires that we actually multiply
   twice once with addition and the second time with a subtaction.

   because of this we need to compute the equations in the order R B
   then G saving the writes for B in the case of 24/32 bit color
   formats.

   api: yuv2rgb_kind (uint8_t *Y, uint8_t *U, uint8_t *V, int *out,
                      int dW, uint32_t *coeffs);

       A          B
       ---        ---
       i2 = cb    i3 = cr
       i1 = coeff i0 = y

Where coeffs have the following layout in memory.

uint32_t oy,oc,zero,cy,crv,rmask,cbu,bmask,cgu,cgv;

coeffs is a pointer to oy.

the {rgb} masks are only utilized by the 565 packing algorithm. Note the data
replication is used to simplify the internal algorithms for the dual mac architecture
of BlackFin.

All routines are exported with _ff_bfin_ as a symbol prefix

rough performance gain compared against -O3:

2779809/1484290 187.28%

which translates to ~33c/pel to ~57c/pel for the reference vs 17.5
c/pel for the optimized implementations. Not sure why there is such a
huge variation on the reference codes on Blackfin I guess it must have
to do with the memory system.
*/

#define mL3 .text
#ifdef __FDPIC__
#define mL1 .l1.text