Mercurial > libavcodec.hg

comparison mem.c @ 2979:bfabfdf9ce55 libavcodec

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
COSMETICS: tabs --> spaces, some prettyprinting
author diego
date Thu, 22 Dec 2005 01:10:11 +0000
parents ef2149182f1c
children e23d2bddf122
comparison
equal deleted inserted replaced
2978:403183bbb505 2979:bfabfdf9ce55
62 ptr = memalign(16,size); 62 ptr = memalign(16,size);
63 /* Why 64? 63 /* Why 64?
64 Indeed, we should align it: 64 Indeed, we should align it:
65 on 4 for 386 65 on 4 for 386
66 on 16 for 486 66 on 16 for 486
67 on 32 for 586, PPro - k6-III 67 on 32 for 586, PPro - k6-III
68 on 64 for K7 (maybe for P3 too). 68 on 64 for K7 (maybe for P3 too).
69 Because L1 and L2 caches are aligned on those values. 69 Because L1 and L2 caches are aligned on those values.
70 But I don't want to code such logic here! 70 But I don't want to code such logic here!
71 */ 71 */
72 /* Why 16? 72 /* Why 16?
73 because some cpus need alignment, for example SSE2 on P4, & most RISC cpus 73 because some cpus need alignment, for example SSE2 on P4, & most RISC cpus
74 it will just trigger an exception and the unaligned load will be done in the 74 it will just trigger an exception and the unaligned load will be done in the
75 exception handler or it will just segfault (SSE2 on P4) 75 exception handler or it will just segfault (SSE2 on P4)
76 Why not larger? because i didnt see a difference in benchmarks ... 76 Why not larger? because i didnt see a difference in benchmarks ...
77 */ 77 */
78 /* benchmarks with p3 78 /* benchmarks with p3
79 memalign(64)+1 3071,3051,3032 79 memalign(64)+1 3071,3051,3032
80 memalign(64)+2 3051,3032,3041 80 memalign(64)+2 3051,3032,3041
81 memalign(64)+4 2911,2896,2915 81 memalign(64)+4 2911,2896,2915
82 memalign(64)+8 2545,2554,2550 82 memalign(64)+8 2545,2554,2550
83 memalign(64)+16 2543,2572,2563 83 memalign(64)+16 2543,2572,2563
84 memalign(64)+32 2546,2545,2571 84 memalign(64)+32 2546,2545,2571
85 memalign(64)+64 2570,2533,2558 85 memalign(64)+64 2570,2533,2558
86 86
87 btw, malloc seems to do 8 byte alignment by default here 87 btw, malloc seems to do 8 byte alignment by default here
88 */ 88 */
89 #else 89 #else
90 ptr = malloc(size); 90 ptr = malloc(size);