comparison ppc/gmc_altivec.c @ 2967:ef2149182f1c libavcodec

COSMETICS: Remove all trailing whitespace.

2966:564788471dd4

    const int C=(16-x16)*(   y16);
    const int D=(   x16)*(   y16);
    int i;

POWERPC_PERF_START_COUNT(altivec_gmc1_num, GMC1_PERF_COND);
    
    for(i=0; i<h; i++)
    {
        dst[0]= (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1] + rounder)>>8;
        dst[1]= (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2] + rounder)>>8;
        dst[2]= (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3] + rounder)>>8;

    Bv = vec_splat(tempA, 1);
    Cv = vec_splat(tempA, 2);
    Dv = vec_splat(tempA, 3);

    rounderV = vec_ld(0, (unsigned short*)rounder_a);
    
    // we'll be able to pick-up our 9 char elements
    // at src from those 32 bytes
    // we load the first batch here, as inside the loop
    // we can re-use 'src+stride' from one iteration
    // as the 'src' of the next.
    src_0 = vec_ld(0, src);
    src_1 = vec_ld(16, src);
    srcvA = vec_perm(src_0, src_1, vec_lvsl(0, src));
    
    if (src_really_odd != 0x0000000F)
    { // if src & 0xF == 0xF, then (src+1) is properly aligned on the second vector.
      srcvB = vec_perm(src_0, src_1, vec_lvsl(1, src));
    }
    else
    {
      srcvB = src_1;
    }
    srcvA = vec_mergeh(vczero, srcvA);
    srcvB = vec_mergeh(vczero, srcvB);
    
    for(i=0; i<h; i++)
    {
      dst_odd = (unsigned long)dst & 0x0000000F;
      src_really_odd = (((unsigned long)src) + stride) & 0x0000000F;
      
      dstv = vec_ld(0, dst);
      
      // we we'll be able to pick-up our 9 char elements
      // at src + stride from those 32 bytes
      // then reuse the resulting 2 vectors srvcC and srcvD
      // as the next srcvA and srcvB
      src_0 = vec_ld(stride + 0, src);
      src_1 = vec_ld(stride + 16, src);
      srcvC = vec_perm(src_0, src_1, vec_lvsl(stride + 0, src));
      
      if (src_really_odd != 0x0000000F)
      { // if src & 0xF == 0xF, then (src+1) is properly aligned on the second vector.
        srcvD = vec_perm(src_0, src_1, vec_lvsl(stride + 1, src));
      }
      else
      {
        srcvD = src_1;
      }
      
      srcvC = vec_mergeh(vczero, srcvC);
      srcvD = vec_mergeh(vczero, srcvD);
      

      // OK, now we (finally) do the math :-)
      // those four instructions replaces 32 int muls & 32 int adds.
      // isn't AltiVec nice ?
      tempA = vec_mladd((vector unsigned short)srcvA, Av, rounderV);
      tempB = vec_mladd((vector unsigned short)srcvB, Bv, tempA);
      tempC = vec_mladd((vector unsigned short)srcvC, Cv, tempB);
      tempD = vec_mladd((vector unsigned short)srcvD, Dv, tempC);
      
      srcvA = srcvC;
      srcvB = srcvD;
      
      tempD = vec_sr(tempD, vcsr8);
      
      dstv2 = vec_pack(tempD, (vector unsigned short)vczero);
      
      if (dst_odd)
      {
        dstv2 = vec_perm(dstv, dstv2, vcprm(0,1,s0,s1));
      }
      else
      {
        dstv2 = vec_perm(dstv, dstv2, vcprm(s0,s1,2,3));
      }
      
      vec_st(dstv2, 0, dst);
      
      dst += stride;
      src += stride;
    }

POWERPC_PERF_STOP_COUNT(altivec_gmc1_num, GMC1_PERF_COND);

2967:ef2149182f1c

    const int C=(16-x16)*(   y16);
    const int D=(   x16)*(   y16);
    int i;

POWERPC_PERF_START_COUNT(altivec_gmc1_num, GMC1_PERF_COND);

    for(i=0; i<h; i++)
    {
        dst[0]= (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1] + rounder)>>8;
        dst[1]= (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2] + rounder)>>8;
        dst[2]= (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3] + rounder)>>8;

    Bv = vec_splat(tempA, 1);
    Cv = vec_splat(tempA, 2);
    Dv = vec_splat(tempA, 3);

    rounderV = vec_ld(0, (unsigned short*)rounder_a);

    // we'll be able to pick-up our 9 char elements
    // at src from those 32 bytes
    // we load the first batch here, as inside the loop
    // we can re-use 'src+stride' from one iteration
    // as the 'src' of the next.
    src_0 = vec_ld(0, src);
    src_1 = vec_ld(16, src);
    srcvA = vec_perm(src_0, src_1, vec_lvsl(0, src));

    if (src_really_odd != 0x0000000F)
    { // if src & 0xF == 0xF, then (src+1) is properly aligned on the second vector.
      srcvB = vec_perm(src_0, src_1, vec_lvsl(1, src));
    }
    else
    {
      srcvB = src_1;
    }
    srcvA = vec_mergeh(vczero, srcvA);
    srcvB = vec_mergeh(vczero, srcvB);

    for(i=0; i<h; i++)
    {
      dst_odd = (unsigned long)dst & 0x0000000F;
      src_really_odd = (((unsigned long)src) + stride) & 0x0000000F;

      dstv = vec_ld(0, dst);

      // we we'll be able to pick-up our 9 char elements
      // at src + stride from those 32 bytes
      // then reuse the resulting 2 vectors srvcC and srcvD
      // as the next srcvA and srcvB
      src_0 = vec_ld(stride + 0, src);
      src_1 = vec_ld(stride + 16, src);
      srcvC = vec_perm(src_0, src_1, vec_lvsl(stride + 0, src));

      if (src_really_odd != 0x0000000F)
      { // if src & 0xF == 0xF, then (src+1) is properly aligned on the second vector.
        srcvD = vec_perm(src_0, src_1, vec_lvsl(stride + 1, src));
      }
      else
      {
        srcvD = src_1;
      }

      srcvC = vec_mergeh(vczero, srcvC);
      srcvD = vec_mergeh(vczero, srcvD);


      // OK, now we (finally) do the math :-)
      // those four instructions replaces 32 int muls & 32 int adds.
      // isn't AltiVec nice ?
      tempA = vec_mladd((vector unsigned short)srcvA, Av, rounderV);
      tempB = vec_mladd((vector unsigned short)srcvB, Bv, tempA);
      tempC = vec_mladd((vector unsigned short)srcvC, Cv, tempB);
      tempD = vec_mladd((vector unsigned short)srcvD, Dv, tempC);

      srcvA = srcvC;
      srcvB = srcvD;

      tempD = vec_sr(tempD, vcsr8);

      dstv2 = vec_pack(tempD, (vector unsigned short)vczero);

      if (dst_odd)
      {
        dstv2 = vec_perm(dstv, dstv2, vcprm(0,1,s0,s1));
      }
      else
      {
        dstv2 = vec_perm(dstv, dstv2, vcprm(s0,s1,2,3));
      }

      vec_st(dstv2, 0, dst);

      dst += stride;
      src += stride;
    }

POWERPC_PERF_STOP_COUNT(altivec_gmc1_num, GMC1_PERF_COND);