mpegvideo_altivec.c

   1 /*
   2  * Copyright (c) 2002 Dieter Shirley
   3  *
   4  * dct_unquantize_h263_altivec:
   5  * Copyright (c) 2003 Romain Dolbeau <romain@dolbeau.org>
   6  *
   7  * This file is part of FFmpeg.
   8  *
   9  * FFmpeg is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public
  11  * License as published by the Free Software Foundation; either
  12  * version 2.1 of the License, or (at your option) any later version.
  13  *
  14  * FFmpeg is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with FFmpeg; if not, write to the Free Software
  21  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  22  */
  23
  24 #include <stdlib.h>
  25 #include <stdio.h>
  26
  27 #include "config.h"
  28 #include "libavutil/attributes.h"
  29 #include "libavutil/cpu.h"
  30 #include "libavutil/ppc/cpu.h"
  31 #include "libavutil/ppc/types_altivec.h"
  32 #include "libavutil/ppc/util_altivec.h"
  33 #include "libavcodec/mpegvideo.h"
  34
  35 #if HAVE_ALTIVEC
  36
  37 /* AltiVec version of dct_unquantize_h263
  38    this code assumes `block' is 16 bytes-aligned */
  39 static void dct_unquantize_h263_altivec(MpegEncContext *s,
  40                                  int16_t *block, int n, int qscale)
  41 {
  42     int i, level, qmul, qadd;
  43     int nCoeffs;
  44
  45     assert(s->block_last_index[n]>=0);
  46
  47     qadd = (qscale - 1) | 1;
  48     qmul = qscale << 1;
  49
  50     if (s->mb_intra) {
  51         if (!s->h263_aic) {
  52             if (n < 4)
  53                 block[0] = block[0] * s->y_dc_scale;
  54             else
  55                 block[0] = block[0] * s->c_dc_scale;
  56         }else
  57             qadd = 0;
  58         i = 1;
  59         nCoeffs= 63; //does not always use zigzag table
  60     } else {
  61         i = 0;
  62         nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ];
  63     }
  64
  65     {
  66         register const vector signed short vczero = (const vector signed short)vec_splat_s16(0);
  67         DECLARE_ALIGNED(16, short, qmul8) = qmul;
  68         DECLARE_ALIGNED(16, short, qadd8) = qadd;
  69         register vector signed short blockv, qmulv, qaddv, nqaddv, temp1;
  70         register vector bool short blockv_null, blockv_neg;
  71         register short backup_0 = block[0];
  72         register int j = 0;
  73
  74         qmulv = vec_splat((vec_s16)vec_lde(0, &qmul8), 0);
  75         qaddv = vec_splat((vec_s16)vec_lde(0, &qadd8), 0);
  76         nqaddv = vec_sub(vczero, qaddv);
  77
  78         // vectorize all the 16 bytes-aligned blocks
  79         // of 8 elements
  80         for(; (j + 7) <= nCoeffs ; j+=8) {
  81             blockv = vec_ld(j << 1, block);
  82             blockv_neg = vec_cmplt(blockv, vczero);
  83             blockv_null = vec_cmpeq(blockv, vczero);
  84             // choose between +qadd or -qadd as the third operand
  85             temp1 = vec_sel(qaddv, nqaddv, blockv_neg);
  86             // multiply & add (block{i,i+7} * qmul [+-] qadd)
  87             temp1 = vec_mladd(blockv, qmulv, temp1);
  88             // put 0 where block[{i,i+7} used to have 0
  89             blockv = vec_sel(temp1, blockv, blockv_null);
  90             vec_st(blockv, j << 1, block);
  91         }
  92
  93         // if nCoeffs isn't a multiple of 8, finish the job
  94         // using good old scalar units.
  95         // (we could do it using a truncated vector,
  96         // but I'm not sure it's worth the hassle)
  97         for(; j <= nCoeffs ; j++) {
  98             level = block[j];
  99             if (level) {
 100                 if (level < 0) {
 101                     level = level * qmul - qadd;
 102                 } else {
 103                     level = level * qmul + qadd;
 104                 }
 105                 block[j] = level;
 106             }
 107         }
 108
 109         if (i == 1) {
 110             // cheat. this avoid special-casing the first iteration
 111             block[0] = backup_0;
 112         }
 113     }
 114 }
 115
 116 #endif /* HAVE_ALTIVEC */
 117
 118 av_cold void ff_mpv_common_init_ppc(MpegEncContext *s)
 119 {
 120 #if HAVE_ALTIVEC
 121     if (!PPC_ALTIVEC(av_get_cpu_flags()))
 122         return;
 123
 124     if ((s->avctx->dct_algo == FF_DCT_AUTO) ||
 125         (s->avctx->dct_algo == FF_DCT_ALTIVEC)) {
 126         s->dct_unquantize_h263_intra = dct_unquantize_h263_altivec;
 127         s->dct_unquantize_h263_inter = dct_unquantize_h263_altivec;
 128     }
 129 #endif /* HAVE_ALTIVEC */
 130 }