math - Optimize mathematical library (libm) -
have tried compile glibc
-march=corei7
see if there's performance improvement on version comes default linux x68_64 distribution? gcc compiled -march=i686
. think (not sure) mathematical library compiled same way. can confirm this?
most linux distributions x86 compile using i686 instructions, asking scheduling them later processors. haven't followed later developments.
a long while different versions of system libraries according processor lines common, performance differences deemed small cost. , machines got more uniform in performance meanwhile.
one thing has remembered today's machines memory bound. i.e., today memory access takes few hundred times longer instruction, , gap growing. not mention machine (an oldish laptop, top-of-the-line 2 years back) has 4 cores (8 threads), battling data/instructions memory. making code run tiny bit faster, cpu can wait longer ram, isn't productive.
Comments
Post a Comment