Switching (back) to MSVC (Re: [Vm-dev] What generates disabledPlugins.c?0

[Andreas Raab]

On 7/22/2010 9:00 AM, Juan Vuletich wrote:
> I heard that the MS tools can't reasonably be used with less than 2Gb of
> ram and a fast processor. I've been regularly using the Win VM building
> setup suggested by you for years on a P3 with 1Gb of memory. And my
> current Win environment is a 1Gb VirtualBox inside a 2.5Gb Mac. I know,
> I could dual boot to Win, to have the full 2.5Gb for it...

That's the tools. You don't need to use much of the tools if you only 
want to recompile (in fact, hopefully none). In other words, even if 
this is true you should be no worse off than today.

> Besides, Cog does not generate native code for every CompiledMethod in
> the system, so I wonder if there would be a visible performance loss
> leaving GCC.

That remains to be seen, but I don't think this is going to be an issue. 
The performance differences have been historically high in micro 
benchmarks (2x) but much less pointed in macro benchmarks (10%). In 
fact, when measuring performance without the core interpreter loop 
(i.e., just the primitives) the MS compiler historically fared better 
than GCC (noticable for example for complex BitBlt rules). The big gain 
of GCC was static register assignments for interpret() which has been 
severely broken in GCC 3 and 4 (which is the reason I'm sticking with 
2.95.2). And since the core interpreter loop is precisely what Cog 
optimizes I would actually expect a net benefit but we'll have to run 
the benchmarks to be sure.

Cheers,
   - Andreas