Re: 64 bit codes

[Mok-Kong Shen <mok-kong.shen@t-online.de>]

Am 27.09.2011 15:09, schrieb Ludovic Brenta:
> Mok-Kong Shen wrote on comp.lang.ada:
>> I have some computations that would be much more efficient, if done
>> with 64-bit arithmetics ("natively") than with 32-bit arithmetics
>> and I don't like to take the trouble/"risk" of having a 2nd OS on my
>> PC (I don't think currently to possess even the proper knowledge to
>> do that at all).
>
> I'm not sure what you mean exactly by "64-bit arithmetic".  If you
> require 64-bit addresses, then yes you need a 64-bit compiler.  But if
> you only need 64-bit floating point numbers, the 32-bit compiler can
> provide them just fine.  In fact, on Intel hardware, Long_Floats are
> normally 80 bits wide IIRC.  Similarly, I think Long_Long_Integer is
> 64 bits wide even with a 32-bit compiler.

On 32-bit gcc, arithmetic operations (+, *, etc.) on "unsigned long
long int" certainly all work, but it's inefficient. Anyway the runtime,
as I have observed in the past, is much higher than I would expect.
For lack of knowledge I have not attempted to see the actual code
generated, but I suppose that's done with the help of some low-level
macros. (By definition, there aren't 64-bit arithmetic instructions on
32-bit hardware.) For on 64-bit hardware a native 64-bit arithmetic
operation shouldn't last very much longer than a corresponding 32-bit
opeartion, I would think.

M. K. Shen