Re: [Mingw-users] Really slow executables generated by Msys/Mingw gcc

[Douglas V. <dou...@dy...>]

> 
> > Malloc is not the problem.
> 
> Right -- I added another time() and it shows that the malloc calls
> do usually add 0 sec and occasionally 1 sec.
> 
> > I wouldn't know enough about the assembled code to know what was going=20
> > on and decipher it.  I would have thought though that gcc would create=20
> > similar code.
> 
> I have tried your code and the gcc invocations on Linux with both
> 3.3.5 as well as 4.0.2.
> 
> 3.3.5 mostly reflects your results (about 23 sec on my machine).
> However the -ffast-math version with 4.0.2 results in about 9 sec
> on my machine which is pretty similar to your 8.9 sec using .NET


> 
> I have not bothered to check on the generated assembly however
> all but the -ffast-math version requires -lm on Linux which probably
> implies there is a subroutine call involved.
> 

Can a subroutine call add that much overhead.  Yes I understand that
there is some overhead, but certainly it seems to me that the
calculation of a trig function is alot more time consuming than such
overhead.

This example with the trig functions was just a small one I put together
so other people could test.  If I were to use a Gaussian elimination
code to solve a large matrix  i would get comparable results that Cases
C and D are much slower than any of the others.  This would not involve
any complex trig function just straight forward adding and
multiplaction.  I can do this and make all these runs again if I need to
to show that its not just associated with the trig function.

> Therefor I guess Greg Chicares assumption that the difference in
> speed is due to FSIN inline (or something similar) as opposed to
> actually calling a subroutine has some probability.
> 
> However according to the gcc docs -ffast-math option impacts
> reproduceability of the results on different platforms etc.

As I said, I don't know the implications of using -ffast-math and don't
necessarly want to use it as "the solution".  But still, if you compare
Case A1 with Case C1 or D1 there is a large difference in run time.  And
as just stated above, this slowness is not just associated with trig
functions.   If i use one that just uses basic it math there is still a
large runtime difference.

> 
> Best,
> Michael