This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
More rewrite_into_ssa/rewrite_ssa_into_ssa compile-time imrovements
- From: Jeffrey A Law <law at redhat dot com>
- To: gcc-patches at gcc dot gnu dot org
- Date: Mon, 31 Jan 2005 22:26:35 -0700
- Subject: More rewrite_into_ssa/rewrite_ssa_into_ssa compile-time imrovements
- Organization: Red Hat, Inc
- Reply-to: law at redhat dot com
Yet another compile-time performance issue I noticed while working on
an unrelated change.
The SSA rewriting code currently implements KILLS with a simple bitmap
which is shared across the calls to mark_def_sites and friends. That
shared sbitmap is (of course) cleared as we start processing a new
block.
While working on some unrelated changes, I noticed that we're spending
quite a bit of time in sbitmap_zero.
My experiments show that the clearing of KILLS is _far_ more costly than
the searches through blocks in sparse bitmaps for the typical access
patterns we see in KILLS -- changing the KILLS into sparse bitmaps
improves the compile time of insn-attrtab by 1.1% and the components of
cc1 by 0.7%.
Bootstrapped and regression tested on i686-pc-linux-gnu.