This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC 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: gcc compile-time performance

> 
> Do you have a feeling for the amount of precision needed?  Eg, if we did a 
> simple floating-type code that used 16-bit fraction and 16-bit exponent, 
> then operations like multiply and divide would become very cheap (we could 
> simply cast to long for the tricky bits): the statistics show that 
> multiplication and division out-number addition and subtraction, but there 
> is one place where we have
> 
>       REAL_ARITHMETIC (real_almost_one, MINUS_EXPR, real_one, 
> real_almost_one);
> 
> Which might be too small.
It is not that small one.  The "almost values" are 1/10000 off.

Honza
> 
> R.
Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]