[Beowulf] newbie

Nifty Tom Mitchell niftyompi at niftyegg.com
Tue May 12 11:37:14 PDT 2009

On Thu, May 07, 2009 at 02:49:18PM -0700, Greg Lindahl wrote:
> On Thu, May 07, 2009 at 05:43:02PM -0400, Mark Hahn wrote:
> >> Probably AMD had been thinking hard on this and decided to make compilers at
> >> last. http://developer.amd.com/cpu/open64/pages/default.aspx
> >
> > interesting.  it wasn't obvious at a glance how this actually differed 
> > from gcc.  besides a comparison on real code, it would
> > be interesting to know the reason (political, practical?)
> > what keeps open64 from simply contributing to mainstream gcc.
> > is there some conflicting infrastructure?

I went and looked at the source package. 
Folks this is for all practical purposes the Pathscale compiler.

Many Pathscale compiler users know that the Pathscale compiler is built
upon the SGI open sourced compiler base. The .iso package from Pathscale
does contain source as per the GPL.   I should note that the Pathscale
compiler does have some non GPL library support that the package at AMD
may have alternate GPL versions for ....

A simple minded scan of the source tree found:
	* SGI 11427 lines
	* Pathscale  6284 lines
	* QLogic 440 lines
	* Intel 167 lines
	* AMD  151 lines.

It is unclear how well integrated changes are over time and
what the deltas are.   I do see a 2008 Pathscale copyright
so some recent changes are in.

Bugfixes might be expected to trail the Pathscale/SiCortex updates
by months....

I guess if you are a fan of the technical superiority and usability
improvements of CentOS in contrast to RHEL then this game is for you.

  FWIW the mainstream gcc and open64 compilers have very different internal
structures and designs.   It is very hard to pull from one to the other.
Some work was done prior to gcc4.x to sell the open64 philosophy to the
gcc team but it did not take.
  Perhaps the most important omission in the open64 family of bits is the inability
to switch to the Linux kernel ABI... i.e. it will not build the Linux kernel.  User
space is in good shape except for the odd places where 'spiffy new gcc tricks' and
inventions are required.

  Fortran shines, very much so in 64bit mode.
  C is darn good,
  C++ is solid (open64x is written mostly in C++) some new C++ tricks are up for grabs.

	T o m  M i t c h e l l 
	Found me a new hat, now what?

More information about the Beowulf mailing list