This removes the non-SLP paths from vectorizable_call, propagates
out ncopies == 1 and removes empty loops resulting from that.
Bootstrapped and tested on x86_64-unknown-linux-gnu, will squash
and push.
* tree-vect-stmts.cc (vectorizable_call): Remove non-SLP path.
---
gcc/tree-vect-stmts.cc | 124 +++++++++++++++++++----------------------
1 file changed, 56 insertions(+), 68 deletions(-)
diff --git a/gcc/tree-vect-stmts.cc b/gcc/tree-vect-stmts.cc
index 7a89bc0e5d4..0799e4fd664 100644
--- a/gcc/tree-vect-stmts.cc
+++ b/gcc/tree-vect-stmts.cc
@@ -3381,7 +3381,6 @@ vectorizable_call (vec_info *vinfo,
tree vectypes[ARRAY_SIZE (dt)] = {};
slp_tree slp_op[ARRAY_SIZE (dt)] = {};
int ndts = ARRAY_SIZE (dt);
- int j;
auto_vec<tree, 8> vargs;
enum { NARROW, NONE, WIDEN } modifier;
size_t i, nargs;
@@ -3510,13 +3509,14 @@ vectorizable_call (vec_info *vinfo,
return false;
}
- if (vect_emulated_vector_p (vectype_in) || vect_emulated_vector_p
(vectype_out))
- {
+ if (vect_emulated_vector_p (vectype_in)
+ || vect_emulated_vector_p (vectype_out))
+ {
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"use emulated vector type for call\n");
return false;
- }
+ }
/* FORNOW */
nunits_in = TYPE_VECTOR_SUBPARTS (vectype_in);
@@ -3693,32 +3693,28 @@ vectorizable_call (vec_info *vinfo,
tree prev_res = NULL_TREE;
vargs.safe_grow (vect_nargs, true);
auto_vec<vec<tree> > vec_defs (nargs);
- for (j = 0; j < 1; ++j)
- {
- /* Build argument list for the vectorized call. */
- if (cfn == CFN_GOMP_SIMD_LANE)
- {
- for (i = 0; i < SLP_TREE_NUMBER_OF_VEC_STMTS (slp_node); ++i)
- {
- /* ??? For multi-lane SLP we'd need to build
- { 0, 0, .., 1, 1, ... }. */
- tree cst = build_index_vector (vectype_out,
- i * nunits_out, 1);
- tree new_var
- = vect_get_new_ssa_name (vectype_out, vect_simple_var,
- "cst_");
- gimple *init_stmt = gimple_build_assign (new_var, cst);
- vect_init_vector_1 (vinfo, stmt_info, init_stmt, NULL);
- new_temp = make_ssa_name (vec_dest);
- gimple *new_stmt
- = gimple_build_assign (new_temp, new_var);
- vect_finish_stmt_generation (vinfo, stmt_info, new_stmt,
- gsi);
- slp_node->push_vec_def (new_stmt);
- }
- continue;
- }
+ /* Build argument list for the vectorized call. */
+ if (cfn == CFN_GOMP_SIMD_LANE)
+ {
+ for (i = 0; i < SLP_TREE_NUMBER_OF_VEC_STMTS (slp_node); ++i)
+ {
+ /* ??? For multi-lane SLP we'd need to build
+ { 0, 0, .., 1, 1, ... }. */
+ tree cst = build_index_vector (vectype_out,
+ i * nunits_out, 1);
+ tree new_var
+ = vect_get_new_ssa_name (vectype_out, vect_simple_var, "cst_");
+ gimple *init_stmt = gimple_build_assign (new_var, cst);
+ vect_init_vector_1 (vinfo, stmt_info, init_stmt, NULL);
+ new_temp = make_ssa_name (vec_dest);
+ gimple *new_stmt = gimple_build_assign (new_temp, new_var);
+ vect_finish_stmt_generation (vinfo, stmt_info, new_stmt, gsi);
+ slp_node->push_vec_def (new_stmt);
+ }
+ }
+ else
+ {
vec<tree> vec_oprnds0;
vect_get_slp_defs (vinfo, slp_node, &vec_defs);
vec_oprnds0 = vec_defs[0];
@@ -3730,9 +3726,8 @@ vectorizable_call (vec_info *vinfo,
if (masked_loop_p && reduc_idx >= 0)
{
unsigned int vec_num = vec_oprnds0.length ();
- vargs[varg++] = vect_get_loop_mask (loop_vinfo,
- gsi, masks, vec_num,
- vectype_out, i);
+ vargs[varg++] = vect_get_loop_mask (loop_vinfo, gsi, masks,
+ vec_num, vectype_out, i);
}
size_t k;
for (k = 0; k < nargs; k++)
@@ -3782,12 +3777,11 @@ vectorizable_call (vec_info *vinfo,
else if (mask_opno >= 0 && masked_loop_p)
{
unsigned int vec_num = vec_oprnds0.length ();
- tree mask = vect_get_loop_mask (loop_vinfo,
- gsi, masks, vec_num,
- vectype_out, i);
- vargs[mask_opno] = prepare_vec_mask
- (loop_vinfo, TREE_TYPE (mask), mask,
- vargs[mask_opno], gsi);
+ tree mask = vect_get_loop_mask (loop_vinfo, gsi, masks,
+ vec_num, vectype_out, i);
+ vargs[mask_opno]
+ = prepare_vec_mask (loop_vinfo, TREE_TYPE (mask), mask,
+ vargs[mask_opno], gsi);
}
gcall *call;
@@ -3804,6 +3798,7 @@ vectorizable_call (vec_info *vinfo,
slp_node->push_vec_def (new_stmt);
}
}
+
for (i = 0; i < nargs; i++)
{
vec<tree> vec_oprndsi = vec_defs[i];
@@ -3815,41 +3810,34 @@ vectorizable_call (vec_info *vinfo,
auto_vec<vec<tree> > vec_defs (nargs);
/* We don't define any narrowing conditional functions at present. */
gcc_assert (mask_opno < 0);
- for (j = 0; j < 1; ++j)
- {
- /* Build argument list for the vectorized call. */
- if (j == 0)
- vargs.create (nargs * 2);
- else
- vargs.truncate (0);
- vec<tree> vec_oprnds0;
+ /* Build argument list for the vectorized call. */
+ vargs.create (nargs * 2);
- vect_get_slp_defs (vinfo, slp_node, &vec_defs);
- vec_oprnds0 = vec_defs[0];
+ vect_get_slp_defs (vinfo, slp_node, &vec_defs);
+ vec<tree> vec_oprnds0 = vec_defs[0];
- /* Arguments are ready. Create the new vector stmt. */
- for (i = 0; vec_oprnds0.iterate (i, &vec_oprnd0); i += 2)
+ /* Arguments are ready. Create the new vector stmt. */
+ for (i = 0; vec_oprnds0.iterate (i, &vec_oprnd0); i += 2)
+ {
+ size_t k;
+ vargs.truncate (0);
+ for (k = 0; k < nargs; k++)
{
- size_t k;
- vargs.truncate (0);
- for (k = 0; k < nargs; k++)
- {
- vec<tree> vec_oprndsk = vec_defs[k];
- vargs.quick_push (vec_oprndsk[i]);
- vargs.quick_push (vec_oprndsk[i + 1]);
- }
- gcall *call;
- if (ifn != IFN_LAST)
- call = gimple_build_call_internal_vec (ifn, vargs);
- else
- call = gimple_build_call_vec (fndecl, vargs);
- new_temp = make_ssa_name (vec_dest, call);
- gimple_call_set_lhs (call, new_temp);
- gimple_call_set_nothrow (call, true);
- vect_finish_stmt_generation (vinfo, stmt_info, call, gsi);
- slp_node->push_vec_def (call);
+ vec<tree> vec_oprndsk = vec_defs[k];
+ vargs.quick_push (vec_oprndsk[i]);
+ vargs.quick_push (vec_oprndsk[i + 1]);
}
+ gcall *call;
+ if (ifn != IFN_LAST)
+ call = gimple_build_call_internal_vec (ifn, vargs);
+ else
+ call = gimple_build_call_vec (fndecl, vargs);
+ new_temp = make_ssa_name (vec_dest, call);
+ gimple_call_set_lhs (call, new_temp);
+ gimple_call_set_nothrow (call, true);
+ vect_finish_stmt_generation (vinfo, stmt_info, call, gsi);
+ slp_node->push_vec_def (call);
}
for (i = 0; i < nargs; i++)
--
2.43.0
