GCC Status Report (2004-03-09)

[Paul Koning]

>>>>> "Zack" == Zack Weinberg <zack@codesourcery.com> writes:

 Zack> Ian Lance Taylor <ian@wasabisystems.com> writes:
 >> Paul Koning <pkoning@equallogic.com> writes:
 >>> Perhaps we need asm("":::"barrier")?
 >> Seconded.  For high performance device manipulation, it would be
 >> good to have a way to say "do not schedule instructions across
 >> this point" without also saying "clobber everything".

 Zack> __builtin_fence() (possibly with an argument to request various
 Zack> sorts of memory barriers that various architectures have) seems
 Zack> preferable to me over asm()...

Good point.  You'd want to be able to ask for memory barrier
instructions, or not -- in some cases all you need is to block
scheduler actions, and the ordering rules of the target machine are
otherwise sufficient for the purpose.

	  paul