lua-users home
lua-l archive

[Date Prev][Date Next][Thread Prev][Thread Next] [Date Index] [Thread Index]


Hi,

Rici Lake wrote:
> >  [In case anyone wonders: isnan() is only defined in C99.]
> 
> I'm not convinced by this. (I agree with making the test explicit, just 
> not with the avoidance of isnan()). isnan() is in POSIX (since Issue 3, 
> it says here); furthermore, it has quite a long history of being 
> implemented, including in all the IEEE-754 emulation libraries I've 
> used. (It may require #include <ieee.h>, or some other non-standard 
> header.)

There is no benefit from using isnan(). It's not inlined
(probably because nobody ever bothered) and it's slower.
Especially for the common case with a non-NaN operand.

> On the other hand, I've certainly seen C compilers (though I 
> admit not for a long time) which would cheerfully optimize away the 
> (a)!=(a) check (which certainly should be optimized away if a is an 
> integer type.)

It's clearly a violation of the standard to optimize this
comparison away for floating point numbers. Abandon all
hope that such a compiler gets the other subtle issues
of FP arithmetic right.

BTW: At least gcc 3.x has no problems optimizing the integer
variant. Dito for the check for integerness (convert to int,
compare to original number -- see arrayindex, luaH_get). Maybe
this should be converted to an explicit macro, too. The integer
variants could then be defined to '0' or '1' which would help
less gifted compilers.

[LuaJIT has a highly optimized variant of the integerness test
in x86 assembler, but it's not easily reusable.]

Bye,
     Mike