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

   1 /*****************************************************************************
   2  * memcpyaltivec.c : AltiVec memcpy module
   3  *****************************************************************************
   4  * Copyright (C) 2001 VideoLAN
   5  * $Id: memcpyaltivec.c,v 1.2 2002/08/11 12:27:11 sam Exp $
   6  *
   7  * Authors: 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., 59 Temple Place - Suite 330, Boston, MA  02111, USA.
  22  *****************************************************************************/
  23
  24 #ifndef __BUILD_ALTIVEC_ASM__
  25
  26 /*****************************************************************************
  27  * Preamble
  28  *****************************************************************************/
  29 #include <stdlib.h>
  30 #include <string.h>
  31
  32 #include <vlc/vlc.h>
  33
  34 /*****************************************************************************
  35  * Local prototypes.
  36  *****************************************************************************/
  37 static void * fast_memcpy ( void * to, const void * from, size_t len );
  38
  39 /*****************************************************************************
  40  * Module initializer.
  41  *****************************************************************************/
  42 static int Activate ( vlc_object_t *p_this )
  43 {
  44     p_this->p_vlc->pf_memcpy = fast_memcpy;
  45     return VLC_SUCCESS;
  46 }
  47
  48 /*****************************************************************************
  49  * Module descriptor.
  50  *****************************************************************************/
  51 vlc_module_begin();
  52     set_description( _("AltiVec memcpy module") );
  53     add_requirement( ALTIVEC );
  54     set_capability( "memcpy", 100 );
  55     set_callbacks( Activate, NULL );
  56     add_shortcut( "altivec" );
  57 vlc_module_end();
  58
  59 #else
  60 typedef unsigned long size_t;
  61 #endif /* __BUILD_ALTIVEC_ASM__ */
  62
  63 #if defined(CAN_COMPILE_C_ALTIVEC) || defined( __BUILD_ALTIVEC_ASM__ )
  64
  65 #define vector_s16_t vector signed short
  66 #define vector_u16_t vector unsigned short
  67 #define vector_s8_t vector signed char
  68 #define vector_u8_t vector unsigned char
  69 #define vector_s32_t vector signed int
  70 #define vector_u32_t vector unsigned int
  71 #define MMREG_SIZE 16
  72
  73 #define SMALL_MEMCPY(to, from, len)                                         \
  74 {                                                                           \
  75     unsigned char * end = to + len;                                         \
  76     while( to < end )                                                       \
  77     {                                                                       \
  78         *to++ = *from++;                                                    \
  79     }                                                                       \
  80 }
  81
  82 static void * fast_memcpy( void * _to, const void * _from, size_t len )
  83 {
  84     void * retval = _to;
  85     unsigned char * to = (unsigned char *)_to;
  86     unsigned char * from = (unsigned char *)_from;
  87
  88     if( len > 16 )
  89     {
  90         /* Align destination to MMREG_SIZE -boundary */
  91         register unsigned long int delta;
  92
  93         delta = ((unsigned long)to)&(MMREG_SIZE-1);
  94         if( delta )
  95         {
  96             delta = MMREG_SIZE - delta;
  97             len -= delta;
  98             SMALL_MEMCPY(to, from, delta);
  99         }
 100
 101         if( len & ~(MMREG_SIZE-1) )
 102         {
 103             vector_u8_t perm, ref0, ref1, tmp;
 104
 105             perm = vec_lvsl( 0, from );
 106             ref0 = vec_ld( 0, from );
 107             ref1 = vec_ld( 15, from );
 108             from += 16;
 109             len -= 16;
 110             tmp = vec_perm( ref0, ref1, perm );
 111             while( len & ~(MMREG_SIZE-1) )
 112             {
 113                 ref0 = vec_ld( 0, from );
 114                 ref1 = vec_ld( 15, from );
 115                 from += 16;
 116                 len -= 16;
 117                 vec_st( tmp, 0, to );
 118                 tmp = vec_perm( ref0, ref1, perm );
 119                 to += 16;
 120             }
 121             vec_st( tmp, 0, to );
 122             to += 16;
 123         }
 124     }
 125
 126     if( len )
 127     {
 128         SMALL_MEMCPY( to, from, len );
 129     }
 130
 131     return retval;
 132 }
 133
 134 #endif
 135
 136 #if !defined(CAN_COMPILE_C_ALTIVEC) && !defined(__BUILD_ALTIVEC_ASM__)
 137
 138 /*
 139  * The asm code is generated with:
 140  *
 141  * gcc-2.95 -fvec -D__BUILD_ALTIVEC_ASM__ -O9 -fomit-frame-pointer -mregnames -S *      memcpyaltivec.c
 142  *
 143  * sed 's/.L/._L/g' memcpyaltivec.s |
 144  * awk '{args=""; len=split ($2, arg, ",");
 145  *      for (i=1; i<=len; i++) { a=arg[i]; if (i<len) a=a",";
 146  *                               args = args sprintf ("%-6s", a) }
 147  *      printf ("\t\"\t%-16s%-24s\\n\"\n", $1, args) }' |
 148  * unexpand -a
 149  */
 150
 151 static void * fast_memcpy( void * _to, const void * _from, size_t len )
 152 {
 153     asm ("                                              \n"
 154         "       cmplwi          %cr0, %r5,  16          \n"
 155         "       mr              %r9,  %r3               \n"
 156         "       bc              4,    1,    ._L3        \n"
 157         "       andi.           %r0,  %r3,  15          \n"
 158         "       bc              12,   2,    ._L4        \n"
 159         "       subfic          %r0,  %r0,  16          \n"
 160         "       add             %r11, %r3,  %r0         \n"
 161         "       cmplw           %cr0, %r3,  %r11        \n"
 162         "       subf            %r5,  %r0,  %r5         \n"
 163         "       bc              4,    0,    ._L4        \n"
 164         "       ._L7:                                   \n"
 165         "       lbz             %r0,  0(%r4)            \n"
 166         "       stb             %r0,  0(%r9)            \n"
 167         "       addi            %r9,  %r9,  1           \n"
 168         "       cmplw           %cr0, %r9,  %r11        \n"
 169         "       addi            %r4,  %r4,  1           \n"
 170         "       bc              12,   0,    ._L7        \n"
 171         "       ._L4:                                   \n"
 172         "       rlwinm.         %r0,  %r5,  0,    0,    27    \n"
 173         "       bc              12,   2,    ._L3        \n"
 174         "       addi            %r5,  %r5,  -16         \n"
 175         "       li              %r11, 15                \n"
 176         "       lvsl            %v12, 0,    %r4         \n"
 177         "       lvx             %v1,  0,    %r4         \n"
 178         "       lvx             %v0,  %r11, %r4         \n"
 179         "       rlwinm.         %r0,  %r5,  0,    0,    27    \n"
 180         "       vperm           %v13, %v1,  %v0,  %v12  \n"
 181         "       addi            %r4,  %r4,  16          \n"
 182         "       bc              12,   2,    ._L11       \n"
 183         "       ._L12:                                  \n"
 184         "       addi            %r5,  %r5,  -16         \n"
 185         "       li              %r11, 15                \n"
 186         "       lvx             %v1,  0,    %r4         \n"
 187         "       lvx             %v0,  %r11, %r4         \n"
 188         "       rlwinm.         %r0,  %r5,  0,    0,    27    \n"
 189         "       stvx            %v13, 0,    %r9         \n"
 190         "       vperm           %v13, %v1,  %v0,  %v12  \n"
 191         "       addi            %r4,  %r4,  16          \n"
 192         "       addi            %r9,  %r9,  16          \n"
 193         "       bc              4,    2,    ._L12       \n"
 194         "       ._L11:                                  \n"
 195         "       stvx            %v13, 0,    %r9         \n"
 196         "       addi            %r9,  %r9,  16          \n"
 197         "       ._L3:                                   \n"
 198         "       cmpwi           %cr0, %r5,  0           \n"
 199         "       bclr            12,   2                 \n"
 200         "       add             %r5,  %r9,  %r5         \n"
 201         "       cmplw           %cr0, %r9,  %r5         \n"
 202         "       bclr            4,    0                 \n"
 203         "       ._L17:                                  \n"
 204         "       lbz             %r0,  0(%r4)            \n"
 205         "       stb             %r0,  0(%r9)            \n"
 206         "       addi            %r9,  %r9,  1           \n"
 207         "       cmplw           %cr0, %r9,  %r5         \n"
 208         "       addi            %r4,  %r4,  1           \n"
 209         "       bc              12,   0,    ._L17       \n"
 210         );
 211 }
 212
 213 #endif