Re: Any reason to keep libsdsl in experimental (Was: Please help creating shared *and* static library with cmake)

[Adam Borowski <kilobyte@angband.pl>]

On Sun, Apr 10, 2016 at 07:20:50PM +0200, Tomasz Buchert wrote:
> > ] Architecture: amd64 arm64 ppc64el s390x
> > 
> > But why would you artificially restrict this to just these?
> > 
> > 2.0.3-1 successfully built also on:
> > alpha kfreebsd-amd64 mips64el ppc64 sparc64 x32
> > 
> > Of these, only alpha can be considered a doorstop architecture.
> 
> Damn, I only looked at the officially released archs, sorry.
> I'll upload the less restricted version soon.

Thanks!

I don't know the real use of other 2nd class architectures, but at least for
x32 the benefits over amd64 are:

* drastically lower memory use: relevant for server consolidation and
  vhosting, not really for scientific software
* modest speed gains.  Sometimes 40%, sometimes 0%, a realistic figure on
  some piece of diverse code I've tested is only 7%.

A speed gain of only 7% is typically something only a "gentoo ricer" would
look at, with one exception: when you build a $1M cluster you get to save
$70k, all for the cost of debootstrap+recompile (and go back to amd64 in case
of any problems).

Which seems to be just the purpose of current users of libsdsl.


-- 
A tit a day keeps the vet away.