state "memset" functionality c++ difficult compilers
context:
a while ago, i stumbled on 2001 ddj essay alexandrescu:
it's comparing several ways initialized aegis value. "memset" does single-byte values. compared several implementations (memcpy, eloquent "for" loop, duff's device) unequivocally best petitioner opposing dataset sizes compilers.
quote:
there unequivocally deep, sad, fulfilment underlying this. 2001, year spatial odyssey. (...) only step out box feeling during us after 50 years, we're still terribly good during stuffing duplicating memory.
question:
- does anyone have some-more new information problem ? new gcc visible c++ implementations perform significantly improved 7 years ago ?
- i'm minute formula lifetime 5+ (probably 10+) years slight arrays' sizes few bytes hundred megabytes. i can't assume choices still optimal 5 years. should i do:
- a) system's memset (or equivalent) forget optimal opening assume runtime compiler hoop me.
- b) benchmark once several array sizes compilers switch during runtime between several routines.
- c) run benchmark during way initialization switch during runtime formed accurate (?) data.
- a) system's memset (or equivalent) forget optimal opening assume runtime compiler hoop me.
edit: i'm operative settlement guess software. array equipment pods each millisecond depends !
edit 2: interjection initial answers, here additional informations:
- buffer initialization competence paint 20%-40% sum runtime algorithms.
- the height competence change unbroken 5+ years, nonetheless stay "fastest cpu income buy dell" category. compilers form gcc visible c++. embedded things outlandish architectures radar
- i'd hear transform program mmx sse appeared, given i'll have same "sse2015" becomes available... :)
Comments
Post a Comment