git.sesse.net Git - vlc/blob - modules/misc/memcpy/memcpyaltivec.c

   1 /*****************************************************************************
   2  * memcpyaltivec.c : AltiVec memcpy module
   3  *****************************************************************************
   4  * Copyright (C) 2001 the VideoLAN team
   5  * $Id$
   6  *
   7  * Author: Christophe Massiot <massiot@via.ecp.fr>
   8  *
   9  * This program is free software; you can redistribute it and/or modify
  10  * it under the terms of the GNU General Public License as published by
  11  * the Free Software Foundation; either version 2 of the License, or
  12  * (at your option) any later version.
  13  *
  14  * This program 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
  17  * GNU General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU General Public License
  20  * along with this program; if not, write to the Free Software
  21  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston MA 02110-1301, USA.
  22  *****************************************************************************/
  23
  24 #ifndef __BUILD_ALTIVEC_ASM__
  25
  26 /*****************************************************************************
  27  * Preamble
  28  *****************************************************************************/
  29
  30 #include <vlc/vlc.h>
  31
  32 #ifdef HAVE_ALTIVEC_H
  33 #   include <altivec.h>
  34 #endif
  35
  36 /*****************************************************************************
  37  * Local prototypes.
  38  *****************************************************************************/
  39 static void * fast_memcpy ( void * to, const void * from, size_t len );
  40
  41 /*****************************************************************************
  42  * Module initializer.
  43  *****************************************************************************/
  44 static int Activate ( vlc_object_t *p_this )
  45 {
  46     p_this->p_libvlc->pf_memcpy = fast_memcpy;
  47     return VLC_SUCCESS;
  48 }
  49
  50 /*****************************************************************************
  51  * Module descriptor.
  52  *****************************************************************************/
  53 vlc_module_begin();
  54     set_description( _("AltiVec memcpy") );
  55     set_category( CAT_ADVANCED );
  56     set_subcategory( SUBCAT_ADVANCED_MISC );
  57     add_requirement( ALTIVEC );
  58     set_capability( "memcpy", 100 );
  59     set_callbacks( Activate, NULL );
  60     add_shortcut( "altivec" );
  61 vlc_module_end();
  62
  63 #else
  64 typedef unsigned long size_t;
  65 #endif /* __BUILD_ALTIVEC_ASM__ */
  66
  67 #if defined(CAN_COMPILE_C_ALTIVEC) || defined( __BUILD_ALTIVEC_ASM__ )
  68
  69 #define vector_s16_t vector signed short
  70 #define vector_u16_t vector unsigned short
  71 #define vector_s8_t vector signed char
  72 #define vector_u8_t vector unsigned char
  73 #define vector_s32_t vector signed int
  74 #define vector_u32_t vector unsigned int
  75 #define MMREG_SIZE 16
  76
  77 #define SMALL_MEMCPY(to, from, len)                                         \
  78 {                                                                           \
  79     unsigned char * end = to + len;                                         \
  80     while( to < end )                                                       \
  81     {                                                                       \
  82         *to++ = *from++;                                                    \
  83     }                                                                       \
  84 }
  85
  86 static void * fast_memcpy( void * _to, const void * _from, size_t len )
  87 {
  88     void * retval = _to;
  89     unsigned char * to = (unsigned char *)_to;
  90     unsigned char * from = (unsigned char *)_from;
  91
  92     if( len > 16 )
  93     {
  94         /* Align destination to MMREG_SIZE -boundary */
  95         register unsigned long int delta;
  96
  97         delta = ((unsigned long)to)&(MMREG_SIZE-1);
  98         if( delta )
  99         {
 100             delta = MMREG_SIZE - delta;
 101             len -= delta;
 102             SMALL_MEMCPY(to, from, delta);
 103         }
 104
 105         if( len & ~(MMREG_SIZE-1) )
 106         {
 107             vector_u8_t perm, ref0, ref1, tmp;
 108
 109             perm = vec_lvsl( 0, from );
 110             ref0 = vec_ld( 0, from );
 111             ref1 = vec_ld( 15, from );
 112             from += 16;
 113             len -= 16;
 114             tmp = vec_perm( ref0, ref1, perm );
 115             while( len & ~(MMREG_SIZE-1) )
 116             {
 117                 ref0 = vec_ld( 0, from );
 118                 ref1 = vec_ld( 15, from );
 119                 from += 16;
 120                 len -= 16;
 121                 vec_st( tmp, 0, to );
 122                 tmp = vec_perm( ref0, ref1, perm );
 123                 to += 16;
 124             }
 125             vec_st( tmp, 0, to );
 126             to += 16;
 127         }
 128     }
 129
 130     if( len )
 131     {
 132         SMALL_MEMCPY( to, from, len );
 133     }
 134
 135     return retval;
 136 }
 137
 138 #endif
 139
 140 #if !defined(CAN_COMPILE_C_ALTIVEC) && !defined(__BUILD_ALTIVEC_ASM__)
 141
 142 /*
 143  * The asm code is generated with:
 144  *
 145  * gcc-2.95 -fvec -D__BUILD_ALTIVEC_ASM__ -O9 -fomit-frame-pointer -mregnames -S *      memcpyaltivec.c
 146  *
 147  * sed 's/.L/._L/g' memcpyaltivec.s |
 148  * awk '{args=""; len=split ($2, arg, ",");
 149  *      for (i=1; i<=len; i++) { a=arg[i]; if (i<len) a=a",";
 150  *                               args = args sprintf ("%-6s", a) }
 151  *      printf ("\t\"\t%-16s%-24s\\n\"\n", $1, args) }' |
 152  * unexpand -a
 153  */
 154
 155 static void * fast_memcpy( void * _to, const void * _from, size_t len )
 156 {
 157     asm ("                                              \n"
 158         "       cmplwi          %cr0, %r5,  16          \n"
 159         "       mr              %r9,  %r3               \n"
 160         "       bc              4,    1,    ._L3        \n"
 161         "       andi.           %r0,  %r3,  15          \n"
 162         "       bc              12,   2,    ._L4        \n"
 163         "       subfic          %r0,  %r0,  16          \n"
 164         "       add             %r11, %r3,  %r0         \n"
 165         "       cmplw           %cr0, %r3,  %r11        \n"
 166         "       subf            %r5,  %r0,  %r5         \n"
 167         "       bc              4,    0,    ._L4        \n"
 168         "       ._L7:                                   \n"
 169         "       lbz             %r0,  0(%r4)            \n"
 170         "       stb             %r0,  0(%r9)            \n"
 171         "       addi            %r9,  %r9,  1           \n"
 172         "       cmplw           %cr0, %r9,  %r11        \n"
 173         "       addi            %r4,  %r4,  1           \n"
 174         "       bc              12,   0,    ._L7        \n"
 175         "       ._L4:                                   \n"
 176         "       rlwinm.         %r0,  %r5,  0,    0,    27    \n"
 177         "       bc              12,   2,    ._L3        \n"
 178         "       addi            %r5,  %r5,  -16         \n"
 179         "       li              %r11, 15                \n"
 180         "       lvsl            %v12, 0,    %r4         \n"
 181         "       lvx             %v1,  0,    %r4         \n"
 182         "       lvx             %v0,  %r11, %r4         \n"
 183         "       rlwinm.         %r0,  %r5,  0,    0,    27    \n"
 184         "       vperm           %v13, %v1,  %v0,  %v12  \n"
 185         "       addi            %r4,  %r4,  16          \n"
 186         "       bc              12,   2,    ._L11       \n"
 187         "       ._L12:                                  \n"
 188         "       addi            %r5,  %r5,  -16         \n"
 189         "       li              %r11, 15                \n"
 190         "       lvx             %v1,  0,    %r4         \n"
 191         "       lvx             %v0,  %r11, %r4         \n"
 192         "       rlwinm.         %r0,  %r5,  0,    0,    27    \n"
 193         "       stvx            %v13, 0,    %r9         \n"
 194         "       vperm           %v13, %v1,  %v0,  %v12  \n"
 195         "       addi            %r4,  %r4,  16          \n"
 196         "       addi            %r9,  %r9,  16          \n"
 197         "       bc              4,    2,    ._L12       \n"
 198         "       ._L11:                                  \n"
 199         "       stvx            %v13, 0,    %r9         \n"
 200         "       addi            %r9,  %r9,  16          \n"
 201         "       ._L3:                                   \n"
 202         "       cmpwi           %cr0, %r5,  0           \n"
 203         "       bclr            12,   2                 \n"
 204         "       add             %r5,  %r9,  %r5         \n"
 205         "       cmplw           %cr0, %r9,  %r5         \n"
 206         "       bclr            4,    0                 \n"
 207         "       ._L17:                                  \n"
 208         "       lbz             %r0,  0(%r4)            \n"
 209         "       stb             %r0,  0(%r9)            \n"
 210         "       addi            %r9,  %r9,  1           \n"
 211         "       cmplw           %cr0, %r9,  %r5         \n"
 212         "       addi            %r4,  %r4,  1           \n"
 213         "       bc              12,   0,    ._L17       \n"
 214         );
 215 }
 216
 217 #endif