This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
[PATCH] Avoid creating dead code in the vectorizer
- From: Richard Biener <rguenther at suse dot de>
- To: gcc-patches at gcc dot gnu dot org
- Date: Wed, 2 Jan 2013 15:05:52 +0100 (CET)
- Subject: [PATCH] Avoid creating dead code in the vectorizer
When analyzing PR55334 further I noticed that when vectorizing
an invariant load we still create a regular (but dead) vector
load and an associated pointer induction variable. That's of
course pointless.
Fix is simple.
Bootstrapped (with -O3) and tested on x86_64-unknown-linux-gnu, applied to
trunk.
Richard.
2013-01-02 Richard Biener <rguenther@suse.de>
* tree-vect-stmts.c (vectorizable_load): When vectorizing an
invariant load do not generate a vector load from the scalar
location.
Index: gcc/tree-vect-stmts.c
===================================================================
*** gcc/tree-vect-stmts.c (revision 194787)
--- gcc/tree-vect-stmts.c (working copy)
*************** vectorizable_load (gimple stmt, gimple_s
*** 4988,4993 ****
--- 4988,5006 ----
/* Record the mapping between SSA_NAMEs and statements. */
vect_record_grouped_load_vectors (stmt, dr_chain);
}
+ /* Handle invariant-load. */
+ else if (inv_p && !bb_vinfo)
+ {
+ gimple_stmt_iterator gsi2 = *gsi;
+ gcc_assert (!grouped_load && !slp_perm);
+ gsi_next (&gsi2);
+ new_temp = vect_init_vector (stmt, scalar_dest,
+ vectype, &gsi2);
+ new_stmt = SSA_NAME_DEF_STMT (new_temp);
+ /* Store vector loads in the corresponding SLP_NODE. */
+ if (slp)
+ SLP_TREE_VEC_STMTS (slp_node).quick_push (new_stmt);
+ }
else
{
for (i = 0; i < vec_num; i++)
*************** vectorizable_load (gimple stmt, gimple_s
*** 5135,5151 ****
}
}
- /* 4. Handle invariant-load. */
- if (inv_p && !bb_vinfo)
- {
- gimple_stmt_iterator gsi2 = *gsi;
- gcc_assert (!grouped_load);
- gsi_next (&gsi2);
- new_temp = vect_init_vector (stmt, scalar_dest,
- vectype, &gsi2);
- new_stmt = SSA_NAME_DEF_STMT (new_temp);
- }
-
if (negative)
{
tree perm_mask = perm_mask_for_reverse (vectype);
--- 5148,5153 ----