|
358
|
1 /*
|
|
|
2 This part of code was taken by from Linux-2.4.3 and slightly modified
|
|
|
3 for MMX2 instruction set. I have done it since linux uses page aligned
|
|
|
4 blocks but mplayer uses weakly ordered data and original sources can not
|
|
376
|
5 speedup their. Only using prefetchnta and movntq together have effect!
|
|
358
|
6 If you have questions please contact with me: Nick Kurshev: nickols_k@mail.ru.
|
|
|
7 */
|
|
376
|
8 #ifdef HAVE_MMX2
|
|
|
9 /* for small memory blocks (<256 bytes) this version is faster */
|
|
|
10 #define small_memcpy(to,from,n)\
|
|
|
11 {\
|
|
|
12 __asm__ __volatile__(\
|
|
|
13 "rep ; movsb\n"\
|
|
|
14 ::"D" (to), "S" (from),"c" (n)\
|
|
|
15 : "memory");\
|
|
|
16 }
|
|
358
|
17
|
|
370
|
18 inline static void * fast_memcpy(void * to, const void * from, unsigned len)
|
|
358
|
19 {
|
|
|
20 void *p;
|
|
|
21 int i;
|
|
|
22
|
|
|
23 if(len >= 0x200) /* 512-byte blocks */
|
|
|
24 {
|
|
|
25 p = to;
|
|
|
26 i = len >> 6; /* len/64 */
|
|
376
|
27 len&=63;
|
|
|
28
|
|
358
|
29 __asm__ __volatile__ (
|
|
376
|
30 "1: prefetchnta (%0)\n" /* This set is 28 bytes */
|
|
|
31 " prefetchnta 64(%0)\n"
|
|
|
32 " prefetchnta 128(%0)\n"
|
|
|
33 " prefetchnta 192(%0)\n"
|
|
|
34 " prefetchnta 256(%0)\n"
|
|
|
35 #if 0
|
|
358
|
36 "2: \n"
|
|
|
37 ".section .fixup, \"ax\"\n"
|
|
|
38 "3: movw $0x1AEB, 1b\n" /* jmp on 26 bytes */
|
|
|
39 " jmp 2b\n"
|
|
|
40 ".previous\n"
|
|
|
41 ".section __ex_table,\"a\"\n"
|
|
|
42 " .align 4\n"
|
|
|
43 " .long 1b, 3b\n"
|
|
|
44 ".previous"
|
|
376
|
45 #endif
|
|
358
|
46 : : "r" (from) );
|
|
|
47
|
|
|
48
|
|
|
49 for(; i>0; i--)
|
|
|
50 {
|
|
|
51 __asm__ __volatile__ (
|
|
376
|
52 "1: prefetchnta 320(%0)\n"
|
|
358
|
53 "2: movq (%0), %%mm0\n"
|
|
|
54 " movq 8(%0), %%mm1\n"
|
|
|
55 " movq 16(%0), %%mm2\n"
|
|
|
56 " movq 24(%0), %%mm3\n"
|
|
|
57 " movntq %%mm0, (%1)\n"
|
|
|
58 " movntq %%mm1, 8(%1)\n"
|
|
|
59 " movntq %%mm2, 16(%1)\n"
|
|
|
60 " movntq %%mm3, 24(%1)\n"
|
|
|
61 " movq 32(%0), %%mm0\n"
|
|
|
62 " movq 40(%0), %%mm1\n"
|
|
|
63 " movq 48(%0), %%mm2\n"
|
|
|
64 " movq 56(%0), %%mm3\n"
|
|
|
65 " movntq %%mm0, 32(%1)\n"
|
|
|
66 " movntq %%mm1, 40(%1)\n"
|
|
|
67 " movntq %%mm2, 48(%1)\n"
|
|
|
68 " movntq %%mm3, 56(%1)\n"
|
|
376
|
69 #if 0
|
|
358
|
70 ".section .fixup, \"ax\"\n"
|
|
|
71 "3: movw $0x05EB, 1b\n" /* jmp on 5 bytes */
|
|
|
72 " jmp 2b\n"
|
|
|
73 ".previous\n"
|
|
|
74 ".section __ex_table,\"a\"\n"
|
|
|
75 " .align 4\n"
|
|
|
76 " .long 1b, 3b\n"
|
|
|
77 ".previous"
|
|
376
|
78 #endif
|
|
358
|
79 : : "r" (from), "r" (to) : "memory");
|
|
|
80 from+=64;
|
|
|
81 to+=64;
|
|
|
82 }
|
|
|
83 __asm__ __volatile__ ("emms":::"memory");
|
|
|
84 }
|
|
|
85 /*
|
|
|
86 * Now do the tail of the block
|
|
|
87 */
|
|
376
|
88 small_memcpy(to, from, len);
|
|
358
|
89 return p;
|
|
|
90 }
|
|
376
|
91 #define memcpy(a,b,c) fast_memcpy(a,b,c)
|
|
358
|
92 #endif
|
|
|
93
|
|
370
|
94
|
