🚀 go-pugleaf

Thomas Sippel - Dau (vulture@imperial.ac.uk) wrote:

> Essentially that is because the gcc implementation of templates is along
> the lines of "proovide the minimum necessary to get it going", while
> the ATT code (on which the SGI compiler is based) used a more general
> approach.

Umm, the SGI compiler is NOT ATT-based (at least, not since
Irix 5.3/Irix 6.0.1).

I assume the original poster meant that the generated code ran slowly, not
that the compilation itself was slower..

There *is* a known bug that I found recently (for which there won't be a fix
in Irix 6.2, since we found the problem too late to get into that release - it
will be fixed in a patch), where if you declare a template member function
in the following way, it's NOT inlined:

  template <class T> class Foo {
      ...
      void memfunc();	/* no inline specification here */
      ...
  };

  template <class T> inline Foo<T>::memfunc()
  {
    /* defined as inline, but compiler will fail to inline it */
  }


The workaround for this problem is to stick the inline keyword before the
declaration of the member function inside the template class.

This *may* be the reason for the slowness.  In all other respects, there
should be no difference between template and non-template code.
--
Shankar Unni				E-Mail:	shankar@sgi.com
Silicon Graphics Inc.			Phone:	+1-415-933-2072
URL:	http://reality.sgi.com/employees/shankar