Re: [boost] Suggest: detail functions for full built-in arithmetic

From: Phil Endecott (spam_from_boost_dev_at_[hidden])
Date: 2007-09-09 13:35:39

Guillaume Melquiond wrote:
> Le samedi 08 septembre 2007 ? 23:10 -0400, Daryle Walker a ?crit :
>> Maybe we should add a header to our detail (and/or pending) folder 
>> encapsulating the full versions of built-in integer arithmetic 
>> operations, that most(?) processors have, but neither C nor C++ let us 
>> access.  I'm talking about functions like:
>> 
>> std::pair<unsigned, bool>  full_add( unsigned a1, unsigned a2 );
>> 
>> The processor has a "carry" flag whenever an integer addition overflows, 
>> but we can't access it (without assembly-language hacks).
>
> Is it really worth it? Nowadays, compilers are smart enough. They
> generate better code when details are not obfuscated by hand-crafted
> assembly code. I tested the following code with GCC:
>
>         #include <algorithm>
>         
>         std::pair<unsigned, bool> full_add(unsigned a, unsigned b)
>         { return std::make_pair(a + b, a + b < a); }
>         
>         bool no_overflow(unsigned a, unsigned b)
>         { return !full_add(a, b).second; }

Here's what I get on ARM:

_Z8full_addjj:
        adds	r2, r1, r2
        movcc	r1, #0
        movcs	r1, #1
        str	r2, [r0, #0]
        strb	r1, [r0, #4]
        mov	pc, lr

_Z11no_overflowjj:
        cmn	r0, r1
        movcs	r0, #0
        movcc	r0, #1
        mov	pc, lr

which looks pretty good, and probably optimal.

But I have encountered cases where compilers don't (can't?) do a very 
good job; for example, I don't think I've seen a way to get optimal 
code for 128-bit addition.  Even 64-bit arithmetic is not so great on 
32-bit processors, with library functions used rather than inline 
adds/adc sequences.  But maybe this is better fixed by improving the 
compiler.  Daryle, what application did you have in mind?

Phil.