dead writes elimination patch

Jeffrey A Law law@cygnus.com
Mon Jan 18 03:38:00 GMT 1999


  In message < 199901141408.PAA09438@eux100.sgp.st.com >you write:
  > Jeff, thanks for your suggestions, here is the new version of 
  > the patch.
Thanks.  In hind sight, I should have suggested using an EXPR_LIST to track
interesting stores instead of an array.  It's a little easier to manage and
a little bit more efficient from a runtime standpoint (insertion of a store
onto the list is cheap as is flushing the list).

It also occurred to me that we should be using "anti_dependence", not
"true_dependence" in mark_used_regs -- X in that case is a read, which
must have occurred *before* the writes on the mem_set_list.  This is a minor
optimization.

I reimplemented your optimization using an EXPR_LIST.  This also avoids the
need to deal with the usual copyright assignment issues (though if you plan
on contributing regularly, we should go ahead and get an assignment for you
on file).

I've checked in the updated code.

Thanks!

jeff




More information about the Gcc-patches mailing list