Re: bzero optimization rarely does

[Paul Koning <pkoning at equallogic dot com>]

>>>>> "Richard" == Richard Henderson <rth@redhat.com> writes:

 Richard> On Wed, Jul 10, 2002 at 05:51:20PM -0400, Paul Koning wrote:
 >> Would people agree that bzero should become straight code for
 >> cases that are more than just a single store instruction?  If yes,
 >> how does one go about changing that?  Who might do it?

 Richard> This should be controled by the MOVE_MAX target macro.

I don't think the problem is with MOVE_MAX as such; the problem is
that MOVE_MAX specifies what a single instruction can do, and bzero
appears to be limiting its work to just that small amount.  It should
limit it to something quite a lot larger.

The curious thing is that memset (for non-constant source) does
generate sequences of several instructions, which makes sense.

	 paul