kicad-developers team mailing list archive
Mailing list archive
Re: SVN revs greater than 1753 require bootst 1.36 or higher
al davis <ad10@...>
Sun, 10 May 2009 09:13:59 -0400
KMail/1.11.2 (Linux/2.6.26-1-amd64; KDE/4.2.2; x86_64; ; )
On Sunday 10 May 2009, Dick Hollenbeck wrote:
> Has anyone verified that the C++ compiler is able to optimize
> this macro down to something concise?
I have not looked at *this* macro, but a few years ago I made
some tests on various uses of STL (the C++ standard template
I found that in all cases, gcc did a good job of optimizing for
speed, and a not-so-good job at optimizing for size of the
executable. I also found that the compiler often took a long
time to do it.
I compared the vector class against C-style arrays and found
that there was no speed penalty for using the vector class, even
in speed-critical loops.
I found that the string class seems to always perform at least
as good as C-style strings, in both speed and space.
The only downsides of using STL that I found are compile time
and code size. Code size is bigger because the templates are
expanded in line, not as called functions. This means there is
a lot of duplication when a template is used often.
I also looked at the run time cost of exceptions, and found that
when exceptions are not called, there is no run time cost, as
compared to other methods of accomplishing the task. The
compiler does generate extra code to handle the exceptions,
including calling destructors and unrolling the stack, but it is
only used when an exception is actually thrown.
I also looked at the run-time cost of virtual functions, and
found that it always compares favorably to other ways of
accomplishing the task.
I think my application (circuit simulation, Gnucap) is more
critical on speed issues than anything in Kicad
My conclusion of all of this is that these features work well
and should be used when they do what is needed.