Boost logo

Boost :

From: Kevin Sopp (baraclese_at_[hidden])
Date: 2008-08-21 13:29:35

> Ah, I see. 32 bit is certainly good enough. Speaking of assembler and
> platform specific stuff, do you know how to trap integer overflow events
> in C++ to handle them as exceptions? It would be a big performance
> advantage to try a calculation that might overflow, catch the overflow
> exception thrown by the event handler and then execute high
> precision/complex calculations instead to get the correct answer. This
> would need to coexist with code that doesn't catch overflow, so the
> event handler would need to inspect a dynamic policy that informs it
> whether to ignore or throw in it's currently context.
> overflow_policy.push(THROW);
> try {
> a = b * c * d;
> } catch overthrow_exception {
> a = foo(b, c, d);
> };
> overflow_policy.pop();

It's an interesting idea. I'm not an ASM expert, I just learned some
for mp_int to play around a little. It could be done if the CPU was
able to inspect the carry flag automatically after each arithmetic
instruction and then issue an interrupt on overflow.

> After looking at your mp_int data structure this sounds like something
> you could implement quite easily by reusing a large portion of your
> existing code.

Yes, that can be done. I don't know when it will happen. Maybe someone
else will do it ;-)


Boost list run by bdawes at, gregod at, cpdaniel at, john at