On Fri, 15 Mar 2002, Charles Shannon Hendrix wrote:

: What I'm talking about is calls to things like this *outside* of a
: library.  How about static code?

Static code is not the norm these days, but you can feel free to compile
things any way you want.  Since the main scope of the libc.so change in this
thread is to help out the base NetBSD distribution, and there are few
CPU-intensive static binaries in the base NetBSD distro, this part isn't
really relevant.

: How about functions you write yourself that are not in a library anywhere?

Uh, in order to be able to *run* this code, you have to link it against
*something*.  If it links against libc, as all C, C++, and F77 code
currently does, you get the .[u]{mul,div,mod} overlay, so your code gets the
speedup too.  No need for -mv8 to speed up these ops.

(Any C, C++, or F77 function you compile with the default options that has a
mul, div, or mod will create calls to these functions, and you have to link
against libc to get them.)

: Also, -mv8 is supposed to also do instruction ordering to speed things up.
: You can see this pretty simply by looking at the assembly output.

Yes, I know this.  However, my suggestion is in the scope of providing a
speedup that is generic to all NetBSD/sparc users, without having to
splinter off a whole separate binary distro of "NetBSD/sparcv8".

Sure, you can compile the userland all yourself.  I'm trying to deal with
the more general case....  8-)

-- 
-- Todd Vierling <tv@wasabisystems.com>  *  Wasabi & NetBSD:  Run with it.
-- CDs, Integration, Embedding, Support -- http://www.wasabisystems.com/