This is the mail archive of the libstdc++@gcc.gnu.org mailing list for the libstdc++ project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: RFC: fp printing speedup patch


Paolo Carlini wrote:

... hum, now I see that this can be easily explained considering that your performance
numbers are only for the formatting part not for the complete output (right?). If,
for instance, we suppose that in current mainline roughly half the total time is spent
for writing the buffer and half for formatting, everything would be consistent...

... if this is correct (I think so), it also helps putting the whole issue in better
perspective, since implies that overall we go from ~6 secs to ~4 (if we are "lucky",
that is small numbers) or from ~6 to ~9 (big numbers).


Therefore, the user, which can only see the complete result, may experience a 33% speed
improvement or a 50% speed decrease, respectively.


The differences are of course bigger both directions when the low level write component
weigh less, that is for faster devices than mine (a vanilla EIDE 80G disk) or for
stringbuf.


Paolo.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]