1 /*****************************************************************************
2 * memcpyaltivec.c : AltiVec memcpy module
3 *****************************************************************************
4 * Copyright (C) 2001 the VideoLAN team
7 * Author: Christophe Massiot <massiot@via.ecp.fr>
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.
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.
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 *****************************************************************************/
24 #ifndef __BUILD_ALTIVEC_ASM__
26 /*****************************************************************************
28 *****************************************************************************/
35 #include <vlc_plugin.h>
41 /*****************************************************************************
43 *****************************************************************************/
44 static void * fast_memcpy ( void * to, const void * from, size_t len );
46 /*****************************************************************************
48 *****************************************************************************/
49 static int Activate ( vlc_object_t *p_this )
51 vlc_fastmem_register( fast_memcpy, NULL );
55 /*****************************************************************************
57 *****************************************************************************/
59 set_description( N_("AltiVec memcpy") );
60 set_category( CAT_ADVANCED );
61 set_subcategory( SUBCAT_ADVANCED_MISC );
62 add_requirement( ALTIVEC );
63 set_capability( "memcpy", 100 );
64 set_callbacks( Activate, NULL );
65 add_shortcut( "altivec" );
69 typedef unsigned long size_t;
70 #endif /* __BUILD_ALTIVEC_ASM__ */
72 #if defined(CAN_COMPILE_C_ALTIVEC) || defined( __BUILD_ALTIVEC_ASM__ )
74 #define vector_s16_t vector signed short
75 #define vector_u16_t vector unsigned short
76 #define vector_s8_t vector signed char
77 #define vector_u8_t vector unsigned char
78 #define vector_s32_t vector signed int
79 #define vector_u32_t vector unsigned int
82 #define SMALL_MEMCPY(to, from, len) \
84 unsigned char * end = to + len; \
91 static void * fast_memcpy( void * _to, const void * _from, size_t len )
94 unsigned char * to = (unsigned char *)_to;
95 unsigned char * from = (unsigned char *)_from;
99 /* Align destination to MMREG_SIZE -boundary */
100 register unsigned long int delta;
102 delta = ((unsigned long)to)&(MMREG_SIZE-1);
105 delta = MMREG_SIZE - delta;
107 SMALL_MEMCPY(to, from, delta);
110 if( len & ~(MMREG_SIZE-1) )
112 vector_u8_t perm, ref0, ref1, tmp;
114 perm = vec_lvsl( 0, from );
115 ref0 = vec_ld( 0, from );
116 ref1 = vec_ld( 15, from );
119 tmp = vec_perm( ref0, ref1, perm );
120 while( len & ~(MMREG_SIZE-1) )
122 ref0 = vec_ld( 0, from );
123 ref1 = vec_ld( 15, from );
126 vec_st( tmp, 0, to );
127 tmp = vec_perm( ref0, ref1, perm );
130 vec_st( tmp, 0, to );
137 SMALL_MEMCPY( to, from, len );
145 #if !defined(CAN_COMPILE_C_ALTIVEC) && !defined(__BUILD_ALTIVEC_ASM__)
148 * The asm code is generated with:
150 * gcc-2.95 -fvec -D__BUILD_ALTIVEC_ASM__ -O9 -fomit-frame-pointer -mregnames -S * memcpyaltivec.c
152 * sed 's/.L/._L/g' memcpyaltivec.s |
153 * awk '{args=""; len=split ($2, arg, ",");
154 * for (i=1; i<=len; i++) { a=arg[i]; if (i<len) a=a",";
155 * args = args sprintf ("%-6s", a) }
156 * printf ("\t\"\t%-16s%-24s\\n\"\n", $1, args) }' |
160 static void * fast_memcpy( void * _to, const void * _from, size_t len )
163 " cmplwi %cr0, %r5, 16 \n"
166 " andi. %r0, %r3, 15 \n"
168 " subfic %r0, %r0, 16 \n"
169 " add %r11, %r3, %r0 \n"
170 " cmplw %cr0, %r3, %r11 \n"
171 " subf %r5, %r0, %r5 \n"
174 " lbz %r0, 0(%r4) \n"
175 " stb %r0, 0(%r9) \n"
176 " addi %r9, %r9, 1 \n"
177 " cmplw %cr0, %r9, %r11 \n"
178 " addi %r4, %r4, 1 \n"
181 " rlwinm. %r0, %r5, 0, 0, 27 \n"
183 " addi %r5, %r5, -16 \n"
185 " lvsl %v12, 0, %r4 \n"
186 " lvx %v1, 0, %r4 \n"
187 " lvx %v0, %r11, %r4 \n"
188 " rlwinm. %r0, %r5, 0, 0, 27 \n"
189 " vperm %v13, %v1, %v0, %v12 \n"
190 " addi %r4, %r4, 16 \n"
191 " bc 12, 2, ._L11 \n"
193 " addi %r5, %r5, -16 \n"
195 " lvx %v1, 0, %r4 \n"
196 " lvx %v0, %r11, %r4 \n"
197 " rlwinm. %r0, %r5, 0, 0, 27 \n"
198 " stvx %v13, 0, %r9 \n"
199 " vperm %v13, %v1, %v0, %v12 \n"
200 " addi %r4, %r4, 16 \n"
201 " addi %r9, %r9, 16 \n"
204 " stvx %v13, 0, %r9 \n"
205 " addi %r9, %r9, 16 \n"
207 " cmpwi %cr0, %r5, 0 \n"
209 " add %r5, %r9, %r5 \n"
210 " cmplw %cr0, %r9, %r5 \n"
213 " lbz %r0, 0(%r4) \n"
214 " stb %r0, 0(%r9) \n"
215 " addi %r9, %r9, 1 \n"
216 " cmplw %cr0, %r9, %r5 \n"
217 " addi %r4, %r4, 1 \n"
218 " bc 12, 0, ._L17 \n"