comparison ppc/mpegvideo_altivec.c @ 2967:ef2149182f1c libavcodec

COSMETICS: Remove all trailing whitespace.

2966:564788471dd4

#include <stdio.h>
#include "../dsputil.h"
#include "../mpegvideo.h"

#include "gcc_fixes.h"
 
#include "dsputil_altivec.h"

// Swaps two variables (used for altivec registers)
#define SWAP(a,b) \
do { \

#define FOUROF(a) (a)
#else
// slower, for dumb non-apple GCC
#define FOUROF(a) {a,a,a,a}
#endif
int dct_quantize_altivec(MpegEncContext* s, 
                        DCTELEM* data, int n,
                        int qscale, int* overflow)
{
    int lastNonZero;
    vector float row0, row1, row2, row3, row4, row5, row6, row7;

            }

            if (whichPass == 1)
            {
                // transpose the data for the second pass
                 
                // First, block transpose the upper right with lower left.
                SWAP(row4, alt0);
                SWAP(row5, alt1);
                SWAP(row6, alt2);
                SWAP(row7, alt3);

                    vec_cmpgt(alt6, zero));
            alt7 = vec_sel(vec_madd(alt7, q7, negBias), vec_madd(alt7, q7, bias),
                    vec_cmpgt(alt7, zero));
        }

 
    }

    // Store the data back into the original block
    {
        vector signed short data0, data1, data2, data3, data4, data5, data6, data7;

        vec_ste(scanIndices_01, 0, &lastNonZeroChar);

        lastNonZero = lastNonZeroChar;
        
        // While the data is still in vectors we check for the transpose IDCT permute
        // and handle it using the vector unit if we can.  This is the permute used
        // by the altivec idct, so it is common when using the altivec dct.

        if ((lastNonZero > 0) && (s->dsp.idct_permutation_type == FF_TRANSPOSE_IDCT_PERM))

/*
  AltiVec version of dct_unquantize_h263
  this code assumes `block' is 16 bytes-aligned
*/
void dct_unquantize_h263_altivec(MpegEncContext *s, 
                                 DCTELEM *block, int n, int qscale)
{
POWERPC_PERF_DECLARE(altivec_dct_unquantize_h263_num, 1);
    int i, level, qmul, qadd;
    int nCoeffs;
    
    assert(s->block_last_index[n]>=0);

POWERPC_PERF_START_COUNT(altivec_dct_unquantize_h263_num, 1);
    
    qadd = (qscale - 1) | 1;
    qmul = qscale << 1;
    
    if (s->mb_intra) {
        if (!s->h263_aic) {
            if (n < 4) 
                block[0] = block[0] * s->y_dc_scale;
            else
                block[0] = block[0] * s->c_dc_scale;
        }else
            qadd = 0;
        i = 1;
        nCoeffs= 63; //does not allways use zigzag table 
    } else {
        i = 0;
        nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ];
    }

          };
      register vector signed short blockv, qmulv, qaddv, nqaddv, temp1;
      register vector bool short blockv_null, blockv_neg;
      register short backup_0 = block[0];
      register int j = 0;
      
      qmulv = vec_ld(0, qmul8);
      qaddv = vec_ld(0, qadd8);
      nqaddv = vec_ld(0, nqadd8);

#if 0 // block *is* 16 bytes-aligned, it seems.

            }
            block[j] = level;
        }
      }
#endif
      
      // vectorize all the 16 bytes-aligned blocks
      // of 8 elements
      for(; (j + 7) <= nCoeffs ; j+=8)
      {
        blockv = vec_ld(j << 1, block);

                level = level * qmul + qadd;
            }
            block[j] = level;
        }
      }
      
      if (i == 1)
      { // cheat. this avoid special-casing the first iteration
        block[0] = backup_0;
      }
    }

2967:ef2149182f1c

#include <stdio.h>
#include "../dsputil.h"
#include "../mpegvideo.h"

#include "gcc_fixes.h"

#include "dsputil_altivec.h"

// Swaps two variables (used for altivec registers)
#define SWAP(a,b) \
do { \

#define FOUROF(a) (a)
#else
// slower, for dumb non-apple GCC
#define FOUROF(a) {a,a,a,a}
#endif
int dct_quantize_altivec(MpegEncContext* s,
                        DCTELEM* data, int n,
                        int qscale, int* overflow)
{
    int lastNonZero;
    vector float row0, row1, row2, row3, row4, row5, row6, row7;

            }

            if (whichPass == 1)
            {
                // transpose the data for the second pass

                // First, block transpose the upper right with lower left.
                SWAP(row4, alt0);
                SWAP(row5, alt1);
                SWAP(row6, alt2);
                SWAP(row7, alt3);

                    vec_cmpgt(alt6, zero));
            alt7 = vec_sel(vec_madd(alt7, q7, negBias), vec_madd(alt7, q7, bias),
                    vec_cmpgt(alt7, zero));
        }


    }

    // Store the data back into the original block
    {
        vector signed short data0, data1, data2, data3, data4, data5, data6, data7;

        vec_ste(scanIndices_01, 0, &lastNonZeroChar);

        lastNonZero = lastNonZeroChar;

        // While the data is still in vectors we check for the transpose IDCT permute
        // and handle it using the vector unit if we can.  This is the permute used
        // by the altivec idct, so it is common when using the altivec dct.

        if ((lastNonZero > 0) && (s->dsp.idct_permutation_type == FF_TRANSPOSE_IDCT_PERM))

/*
  AltiVec version of dct_unquantize_h263
  this code assumes `block' is 16 bytes-aligned
*/
void dct_unquantize_h263_altivec(MpegEncContext *s,
                                 DCTELEM *block, int n, int qscale)
{
POWERPC_PERF_DECLARE(altivec_dct_unquantize_h263_num, 1);
    int i, level, qmul, qadd;
    int nCoeffs;

    assert(s->block_last_index[n]>=0);

POWERPC_PERF_START_COUNT(altivec_dct_unquantize_h263_num, 1);

    qadd = (qscale - 1) | 1;
    qmul = qscale << 1;

    if (s->mb_intra) {
        if (!s->h263_aic) {
            if (n < 4)
                block[0] = block[0] * s->y_dc_scale;
            else
                block[0] = block[0] * s->c_dc_scale;
        }else
            qadd = 0;
        i = 1;
        nCoeffs= 63; //does not allways use zigzag table
    } else {
        i = 0;
        nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ];
    }

          };
      register vector signed short blockv, qmulv, qaddv, nqaddv, temp1;
      register vector bool short blockv_null, blockv_neg;
      register short backup_0 = block[0];
      register int j = 0;

      qmulv = vec_ld(0, qmul8);
      qaddv = vec_ld(0, qadd8);
      nqaddv = vec_ld(0, nqadd8);

#if 0 // block *is* 16 bytes-aligned, it seems.

            }
            block[j] = level;
        }
      }
#endif

      // vectorize all the 16 bytes-aligned blocks
      // of 8 elements
      for(; (j + 7) <= nCoeffs ; j+=8)
      {
        blockv = vec_ld(j << 1, block);

                level = level * qmul + qadd;
            }
            block[j] = level;
        }
      }

      if (i == 1)
      { // cheat. this avoid special-casing the first iteration
        block[0] = backup_0;
      }
    }