quicktime/ffmpeg/libavcodec/ps2/dsputil_mmi.c

   1 /*
   2  * MMI optimized DSP utils
   3  * Copyright (c) 2000, 2001 Fabrice Bellard.
   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  * MMI optimization by Leon van Stuivenberg
  20  * clear_blocks_mmi() by BroadQ
  21  */
  22
  23 #include "../dsputil.h"
  24 #include "mmi.h"
  25
  26 void ff_mmi_idct_put(uint8_t *dest, int line_size, DCTELEM *block);
  27 void ff_mmi_idct_add(uint8_t *dest, int line_size, DCTELEM *block);
  28 void ff_mmi_idct(DCTELEM *block);
  29
  30 static void clear_blocks_mmi(DCTELEM * blocks)
  31 {
  32         asm volatile(
  33         ".set noreorder    \n"
  34         "addiu $9, %0, 768 \n"
  35         "nop               \n"
  36         "1:                \n"
  37         "sq $0, 0(%0)      \n"
  38         "move $8, %0       \n"
  39         "addi %0, %0, 64   \n"
  40         "sq $0, 16($8)     \n"
  41         "slt $10, %0, $9   \n"
  42         "sq $0, 32($8)     \n"
  43         "bnez $10, 1b      \n"
  44         "sq $0, 48($8)     \n"
  45         ".set reorder      \n"
  46         : "+r" (blocks) ::  "$8", "$9", "memory" );
  47 }
  48
  49
  50 static void get_pixels_mmi(DCTELEM *block, const uint8_t *pixels, int line_size)
  51 {
  52         asm volatile(
  53         ".set   push            \n\t"
  54         ".set   mips3           \n\t"
  55         "ld     $8, 0(%0)       \n\t"
  56         "add    %0, %0, %2      \n\t"
  57         "ld     $9, 0(%0)       \n\t"
  58         "add    %0, %0, %2      \n\t"
  59         "ld     $10, 0(%0)      \n\t"
  60         "pextlb $8, $0, $8      \n\t"
  61         "sq     $8, 0(%1)       \n\t"
  62         "add    %0, %0, %2      \n\t"
  63         "ld     $8, 0(%0)       \n\t"
  64         "pextlb $9, $0, $9      \n\t"
  65         "sq     $9, 16(%1)      \n\t"
  66         "add    %0, %0, %2      \n\t"
  67         "ld     $9, 0(%0)       \n\t"
  68         "pextlb $10, $0, $10    \n\t"
  69         "sq     $10, 32(%1)     \n\t"
  70         "add    %0, %0, %2      \n\t"
  71         "ld     $10, 0(%0)      \n\t"
  72         "pextlb $8, $0, $8      \n\t"
  73         "sq     $8, 48(%1)      \n\t"
  74         "add    %0, %0, %2      \n\t"
  75         "ld     $8, 0(%0)       \n\t"
  76         "pextlb $9, $0, $9      \n\t"
  77         "sq     $9, 64(%1)      \n\t"
  78         "add    %0, %0, %2      \n\t"
  79         "ld     $9, 0(%0)       \n\t"
  80         "pextlb $10, $0, $10    \n\t"
  81         "sq     $10, 80(%1)     \n\t"
  82         "pextlb $8, $0, $8      \n\t"
  83         "sq     $8, 96(%1)      \n\t"
  84         "pextlb $9, $0, $9      \n\t"
  85         "sq     $9, 112(%1)     \n\t"
  86         ".set   pop             \n\t"
  87         : "+r" (pixels) : "r" (block), "r" (line_size) : "$8", "$9", "$10", "memory" );
  88 }
  89
  90
  91 static void put_pixels8_mmi(uint8_t *block, const uint8_t *pixels, int line_size, int h)
  92 {
  93         asm volatile(
  94         ".set   push            \n\t"
  95         ".set   mips3           \n\t"
  96         "1:                     \n\t"
  97         "ldr    $8, 0(%1)       \n\t"
  98         "addiu  %2, %2, -1      \n\t"
  99         "ldl    $8, 7(%1)       \n\t"
 100         "add    %1, %1, %3      \n\t"
 101         "sd     $8, 0(%0)       \n\t"
 102         "add    %0, %0, %3      \n\t"
 103         "bgtz   %2, 1b          \n\t"
 104         ".set   pop             \n\t"
 105         : "+r" (block), "+r" (pixels), "+r" (h) : "r" (line_size)
 106         : "$8", "memory" );
 107 }
 108
 109
 110 static void put_pixels16_mmi(uint8_t *block, const uint8_t *pixels, int line_size, int h)
 111 {
 112         asm volatile (
 113         ".set   push            \n\t"
 114         ".set   mips3           \n\t"
 115         "1:                     \n\t"
 116         "ldr    $8, 0(%1)       \n\t"
 117         "add    $11, %1, %3     \n\t"
 118         "ldl    $8, 7(%1)       \n\t"
 119         "add    $10, %0, %3     \n\t"
 120         "ldr    $9, 8(%1)       \n\t"
 121         "ldl    $9, 15(%1)      \n\t"
 122         "ldr    $12, 0($11)     \n\t"
 123         "add    %1, $11, %3     \n\t"
 124         "ldl    $12, 7($11)     \n\t"
 125         "pcpyld $8, $9, $8      \n\t"
 126         "sq     $8, 0(%0)       \n\t"
 127         "ldr    $13, 8($11)     \n\t"
 128         "addiu  %2, %2, -2      \n\t"
 129         "ldl    $13, 15($11)    \n\t"
 130         "add    %0, $10, %3     \n\t"
 131         "pcpyld $12, $13, $12   \n\t"
 132         "sq     $12, 0($10)     \n\t"
 133         "bgtz   %2, 1b          \n\t"
 134         ".set   pop             \n\t"
 135         : "+r" (block), "+r" (pixels), "+r" (h) : "r" (line_size)
 136         : "$8", "$9", "$10", "$11", "$12", "$13", "memory" );
 137 }
 138
 139
 140 void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx)
 141 {
 142     const int idct_algo= avctx->idct_algo;
 143
 144     c->clear_blocks = clear_blocks_mmi;
 145
 146     c->put_pixels_tab[1][0] = put_pixels8_mmi;
 147     c->put_no_rnd_pixels_tab[1][0] = put_pixels8_mmi;
 148
 149     c->put_pixels_tab[0][0] = put_pixels16_mmi;
 150     c->put_no_rnd_pixels_tab[0][0] = put_pixels16_mmi;
 151
 152     c->get_pixels = get_pixels_mmi;
 153
 154     if(idct_algo==FF_IDCT_AUTO || idct_algo==FF_IDCT_PS2){
 155         c->idct_put= ff_mmi_idct_put;
 156         c->idct_add= ff_mmi_idct_add;
 157         c->idct    = ff_mmi_idct;
 158         c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM;
 159     }
 160 }
 161