1 /*****************************************************************************
2 * memcpyaltivec.c : AltiVec memcpy module
3 *****************************************************************************
4 * Copyright (C) 2001 VideoLAN
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., 59 Temple Place - Suite 330, Boston, MA 02111, USA.
22 *****************************************************************************/
24 #ifndef __BUILD_ALTIVEC_ASM__
26 /*****************************************************************************
28 *****************************************************************************/
38 /*****************************************************************************
40 *****************************************************************************/
41 static void * fast_memcpy ( void * to, const void * from, size_t len );
43 /*****************************************************************************
45 *****************************************************************************/
46 static int Activate ( vlc_object_t *p_this )
48 p_this->p_vlc->pf_memcpy = fast_memcpy;
52 /*****************************************************************************
54 *****************************************************************************/
56 set_description( _("AltiVec memcpy") );
57 set_category( CAT_ADVANCED );
58 set_subcategory( SUBCAT_ADVANCED_MISC );
59 add_requirement( ALTIVEC );
60 set_capability( "memcpy", 100 );
61 set_callbacks( Activate, NULL );
62 add_shortcut( "altivec" );
66 typedef unsigned long size_t;
67 #endif /* __BUILD_ALTIVEC_ASM__ */
69 #if defined(CAN_COMPILE_C_ALTIVEC) || defined( __BUILD_ALTIVEC_ASM__ )
71 #define vector_s16_t vector signed short
72 #define vector_u16_t vector unsigned short
73 #define vector_s8_t vector signed char
74 #define vector_u8_t vector unsigned char
75 #define vector_s32_t vector signed int
76 #define vector_u32_t vector unsigned int
79 #define SMALL_MEMCPY(to, from, len) \
81 unsigned char * end = to + len; \
88 static void * fast_memcpy( void * _to, const void * _from, size_t len )
91 unsigned char * to = (unsigned char *)_to;
92 unsigned char * from = (unsigned char *)_from;
96 /* Align destination to MMREG_SIZE -boundary */
97 register unsigned long int delta;
99 delta = ((unsigned long)to)&(MMREG_SIZE-1);
102 delta = MMREG_SIZE - delta;
104 SMALL_MEMCPY(to, from, delta);
107 if( len & ~(MMREG_SIZE-1) )
109 vector_u8_t perm, ref0, ref1, tmp;
111 perm = vec_lvsl( 0, from );
112 ref0 = vec_ld( 0, from );
113 ref1 = vec_ld( 15, from );
116 tmp = vec_perm( ref0, ref1, perm );
117 while( len & ~(MMREG_SIZE-1) )
119 ref0 = vec_ld( 0, from );
120 ref1 = vec_ld( 15, from );
123 vec_st( tmp, 0, to );
124 tmp = vec_perm( ref0, ref1, perm );
127 vec_st( tmp, 0, to );
134 SMALL_MEMCPY( to, from, len );
142 #if !defined(CAN_COMPILE_C_ALTIVEC) && !defined(__BUILD_ALTIVEC_ASM__)
145 * The asm code is generated with:
147 * gcc-2.95 -fvec -D__BUILD_ALTIVEC_ASM__ -O9 -fomit-frame-pointer -mregnames -S * memcpyaltivec.c
149 * sed 's/.L/._L/g' memcpyaltivec.s |
150 * awk '{args=""; len=split ($2, arg, ",");
151 * for (i=1; i<=len; i++) { a=arg[i]; if (i<len) a=a",";
152 * args = args sprintf ("%-6s", a) }
153 * printf ("\t\"\t%-16s%-24s\\n\"\n", $1, args) }' |
157 static void * fast_memcpy( void * _to, const void * _from, size_t len )
160 " cmplwi %cr0, %r5, 16 \n"
163 " andi. %r0, %r3, 15 \n"
165 " subfic %r0, %r0, 16 \n"
166 " add %r11, %r3, %r0 \n"
167 " cmplw %cr0, %r3, %r11 \n"
168 " subf %r5, %r0, %r5 \n"
171 " lbz %r0, 0(%r4) \n"
172 " stb %r0, 0(%r9) \n"
173 " addi %r9, %r9, 1 \n"
174 " cmplw %cr0, %r9, %r11 \n"
175 " addi %r4, %r4, 1 \n"
178 " rlwinm. %r0, %r5, 0, 0, 27 \n"
180 " addi %r5, %r5, -16 \n"
182 " lvsl %v12, 0, %r4 \n"
183 " lvx %v1, 0, %r4 \n"
184 " lvx %v0, %r11, %r4 \n"
185 " rlwinm. %r0, %r5, 0, 0, 27 \n"
186 " vperm %v13, %v1, %v0, %v12 \n"
187 " addi %r4, %r4, 16 \n"
188 " bc 12, 2, ._L11 \n"
190 " addi %r5, %r5, -16 \n"
192 " lvx %v1, 0, %r4 \n"
193 " lvx %v0, %r11, %r4 \n"
194 " rlwinm. %r0, %r5, 0, 0, 27 \n"
195 " stvx %v13, 0, %r9 \n"
196 " vperm %v13, %v1, %v0, %v12 \n"
197 " addi %r4, %r4, 16 \n"
198 " addi %r9, %r9, 16 \n"
201 " stvx %v13, 0, %r9 \n"
202 " addi %r9, %r9, 16 \n"
204 " cmpwi %cr0, %r5, 0 \n"
206 " add %r5, %r9, %r5 \n"
207 " cmplw %cr0, %r9, %r5 \n"
210 " lbz %r0, 0(%r4) \n"
211 " stb %r0, 0(%r9) \n"
212 " addi %r9, %r9, 1 \n"
213 " cmplw %cr0, %r9, %r5 \n"
214 " addi %r4, %r4, 1 \n"
215 " bc 12, 0, ._L17 \n"