This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [tuples][patch] Dead Store Elimination
- From: Diego Novillo <dnovillo at google dot com>
- To: Jakub Staszak <kuba at et dot pl>
- Cc: GCC <gcc-patches at gcc dot gnu dot org>, Oleg Ryjkov <olegr at google dot com>
- Date: Tue, 25 Mar 2008 09:15:53 -0400
- Subject: Re: [tuples][patch] Dead Store Elimination
- References: <B04D4810-1B9C-494C-A58B-EE2361F808E7@et.pl>
On 3/24/08 2:58 PM, Jakub Staszak wrote:
* tree-ssa-dse.c (execute_simple_dse): #if 0 removed, bitmap_empty_p
condition
added. (get_stmt_uid): Removed. (record_voperand_set): Arguments changed.
(dse_initialize_block_local_data, memory_ssa_name_same,
memory_address_same,
get_kill_of_stmt_lhs, dse_possible_dead_store_p, dse_optimize_stmt,
dse_record_phis,
dse_finalize_block, tree_ssa_dse): Tuplified.
* passes.c (init_optimization_passes): pass_simple_dse and pass_dse
enabled.
* testsuite/gcc.dg/tree-ssa/ssa-dse-10.c: {dg-final} changed.
* testsuite/gcc.dg/tree-ssa/pr30375.c: {dg-final} changed.
I would rather use the original bitmaps for keeping track of store
statements. Setting a flag on statements also works, but we are now
going through every statement and PHI in the block instead of just the
marked ones. That may slow things down.
You can add a UID field to gimple_statement_base. We will need it
anyway, for LTO.
Thanks. Diego.