libavcodec/alpha/mpegvideo_alpha.c

   1 /*
   2  * Alpha optimized DSP utils
   3  * Copyright (c) 2002 Falk Hueffner <falk@debian.org>
   4  *
   5  * This library is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU Lesser General Public
   7  * License as published by the Free Software Foundation; either
   8  * version 2 of the License, or (at your option) any later version.
   9  *
  10  * This library is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * Lesser General Public License for more details.
  14  *
  15  * You should have received a copy of the GNU Lesser General Public
  16  * License along with this library; if not, write to the Free Software
  17  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  18  */
  19
  20 #include "asm.h"
  21 #include "../dsputil.h"
  22 #include "../mpegvideo.h"
  23
  24 static void dct_unquantize_h263_intra_axp(MpegEncContext *s, DCTELEM *block,
  25                                     int n, int qscale)
  26 {
  27     int i, n_coeffs;
  28     uint64_t qmul, qadd;
  29     uint64_t correction;
  30     DCTELEM *orig_block = block;
  31     DCTELEM block0;
  32
  33     qadd = WORD_VEC((qscale - 1) | 1);
  34     qmul = qscale << 1;
  35     /* This mask kills spill from negative subwords to the next subword.  */
  36     correction = WORD_VEC((qmul - 1) + 1); /* multiplication / addition */
  37
  38     if (!s->h263_aic) {
  39         if (n < 4)
  40             block0 = block[0] * s->y_dc_scale;
  41         else
  42             block0 = block[0] * s->c_dc_scale;
  43     } else {
  44         qadd = 0;
  45     }
  46     n_coeffs = 63; // does not always use zigzag table
  47
  48     for(i = 0; i <= n_coeffs; block += 4, i += 4) {
  49         uint64_t levels, negmask, zeros, add;
  50
  51         levels = ldq(block);
  52         if (levels == 0)
  53             continue;
  54
  55 #ifdef __alpha_max__
  56         /* I don't think the speed difference justifies runtime
  57            detection.  */
  58         negmask = maxsw4(levels, -1); /* negative -> ffff (-1) */
  59         negmask = minsw4(negmask, 0); /* positive -> 0000 (0) */
  60 #else
  61         negmask = cmpbge(WORD_VEC(0x7fff), levels);
  62         negmask &= (negmask >> 1) | (1 << 7);
  63         negmask = zap(-1, negmask);
  64 #endif
  65
  66         zeros = cmpbge(0, levels);
  67         zeros &= zeros >> 1;
  68         /* zeros |= zeros << 1 is not needed since qadd <= 255, so
  69            zapping the lower byte suffices.  */
  70
  71         levels *= qmul;
  72         levels -= correction & (negmask << 16);
  73
  74         /* Negate qadd for negative levels.  */
  75         add = qadd ^ negmask;
  76         add += WORD_VEC(0x0001) & negmask;
  77         /* Set qadd to 0 for levels == 0.  */
  78         add = zap(add, zeros);
  79
  80         levels += add;
  81
  82         stq(levels, block);
  83     }
  84
  85     if (s->mb_intra && !s->h263_aic)
  86         orig_block[0] = block0;
  87 }
  88
  89 static void dct_unquantize_h263_inter_axp(MpegEncContext *s, DCTELEM *block,
  90                                     int n, int qscale)
  91 {
  92     int i, n_coeffs;
  93     uint64_t qmul, qadd;
  94     uint64_t correction;
  95     DCTELEM *orig_block = block;
  96     DCTELEM block0;
  97
  98     qadd = WORD_VEC((qscale - 1) | 1);
  99     qmul = qscale << 1;
 100     /* This mask kills spill from negative subwords to the next subword.  */
 101     correction = WORD_VEC((qmul - 1) + 1); /* multiplication / addition */
 102
 103     n_coeffs = s->intra_scantable.raster_end[s->block_last_index[n]];
 104
 105     for(i = 0; i <= n_coeffs; block += 4, i += 4) {
 106         uint64_t levels, negmask, zeros, add;
 107
 108         levels = ldq(block);
 109         if (levels == 0)
 110             continue;
 111
 112 #ifdef __alpha_max__
 113         /* I don't think the speed difference justifies runtime
 114            detection.  */
 115         negmask = maxsw4(levels, -1); /* negative -> ffff (-1) */
 116         negmask = minsw4(negmask, 0); /* positive -> 0000 (0) */
 117 #else
 118         negmask = cmpbge(WORD_VEC(0x7fff), levels);
 119         negmask &= (negmask >> 1) | (1 << 7);
 120         negmask = zap(-1, negmask);
 121 #endif
 122
 123         zeros = cmpbge(0, levels);
 124         zeros &= zeros >> 1;
 125         /* zeros |= zeros << 1 is not needed since qadd <= 255, so
 126            zapping the lower byte suffices.  */
 127
 128         levels *= qmul;
 129         levels -= correction & (negmask << 16);
 130
 131         /* Negate qadd for negative levels.  */
 132         add = qadd ^ negmask;
 133         add += WORD_VEC(0x0001) & negmask;
 134         /* Set qadd to 0 for levels == 0.  */
 135         add = zap(add, zeros);
 136
 137         levels += add;
 138
 139         stq(levels, block);
 140     }
 141 }
 142
 143 void MPV_common_init_axp(MpegEncContext *s)
 144 {
 145     s->dct_unquantize_h263_intra = dct_unquantize_h263_intra_axp;
 146     s->dct_unquantize_h263_inter = dct_unquantize_h263_inter_axp;
 147 }