The following adds support for single-lane SLP .GOMP_SIMD_LANE
vectorization.
This doesn't handle much, esp. g++.dg/vect/simd-*.cc with their
'inscan' uses are unhandled.
* tree-vect-slp.cc (no_arg_map): New.
(vect_get_operand_map): Handle IFN_GOMP_SIMD_LANE.
(vect_build_slp_tree_1): Likewise.
* tree-vect-stmts.cc (vectorizable_call): Handle single-lane SLP
for .GOMP_SIMD_LANE calls.
---
gcc/tree-vect-slp.cc | 11 +++++++++++
gcc/tree-vect-stmts.cc | 27 +++++++++++++++++++--------
2 files changed, 30 insertions(+), 8 deletions(-)
diff --git a/gcc/tree-vect-slp.cc b/gcc/tree-vect-slp.cc
index 3138a815da7..f3743997e9c 100644
--- a/gcc/tree-vect-slp.cc
+++ b/gcc/tree-vect-slp.cc
@@ -505,6 +505,7 @@ static const int cond_expr_maps[3][5] = {
{ 4, -2, -1, 1, 2 },
{ 4, -1, -2, 2, 1 }
};
+static const int no_arg_map[] = { 0 };
static const int arg0_map[] = { 1, 0 };
static const int arg1_map[] = { 1, 1 };
static const int arg2_map[] = { 1, 2 };
@@ -585,6 +586,9 @@ vect_get_operand_map (const gimple *stmt, bool
gather_scatter_p = false,
case IFN_CTZ:
return arg0_map;
+ case IFN_GOMP_SIMD_LANE:
+ return no_arg_map;
+
default:
break;
}
@@ -1168,6 +1172,8 @@ vect_build_slp_tree_1 (vec_info *vinfo, unsigned char
*swap,
ldst_p = true;
rhs_code = CFN_MASK_STORE;
}
+ else if (cfn == CFN_GOMP_SIMD_LANE)
+ ;
else if ((cfn != CFN_LAST
&& cfn != CFN_MASK_CALL
&& internal_fn_p (cfn)
@@ -1271,6 +1277,11 @@ vect_build_slp_tree_1 (vec_info *vinfo, unsigned char
*swap,
need_same_oprnds = true;
first_op1 = gimple_call_arg (call_stmt, 1);
}
+ else if (rhs_code == CFN_GOMP_SIMD_LANE)
+ {
+ need_same_oprnds = true;
+ first_op1 = gimple_call_arg (call_stmt, 1);
+ }
}
else
{
diff --git a/gcc/tree-vect-stmts.cc b/gcc/tree-vect-stmts.cc
index 840ff8a3406..270c5a5dd34 100644
--- a/gcc/tree-vect-stmts.cc
+++ b/gcc/tree-vect-stmts.cc
@@ -3341,7 +3341,7 @@ vectorizable_call (vec_info *vinfo,
if (ifn == IFN_LAST && !fndecl)
{
if (cfn == CFN_GOMP_SIMD_LANE
- && !slp_node
+ && (!slp_node || SLP_TREE_LANES (slp_node) == 1)
&& loop_vinfo
&& LOOP_VINFO_LOOP (loop_vinfo)->simduid
&& TREE_CODE (gimple_call_arg (stmt, 0)) == SSA_NAME
@@ -3487,18 +3487,15 @@ vectorizable_call (vec_info *vinfo,
/* Build argument list for the vectorized call. */
if (slp_node)
{
- vec<tree> vec_oprnds0;
-
+ unsigned int vec_num = SLP_TREE_NUMBER_OF_VEC_STMTS (slp_node);
vect_get_slp_defs (vinfo, slp_node, &vec_defs);
- vec_oprnds0 = vec_defs[0];
/* Arguments are ready. Create the new vector stmt. */
- FOR_EACH_VEC_ELT (vec_oprnds0, i, vec_oprnd0)
+ for (i = 0; i < vec_num; ++i)
{
int varg = 0;
if (masked_loop_p && reduc_idx >= 0)
{
- unsigned int vec_num = vec_oprnds0.length ();
/* Always true for SLP. */
gcc_assert (ncopies == 1);
vargs[varg++] = vect_get_loop_mask (loop_vinfo,
@@ -3539,11 +3536,26 @@ vectorizable_call (vec_info *vinfo,
vect_finish_stmt_generation (vinfo, stmt_info,
new_stmt, gsi);
}
+ else if (cfn == CFN_GOMP_SIMD_LANE)
+ {
+ /* ??? 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);
+ new_stmt = gimple_build_assign (new_temp, new_var);
+ vect_finish_stmt_generation (vinfo, stmt_info, new_stmt,
+ gsi);
+ }
else
{
if (len_opno >= 0 && len_loop_p)
{
- unsigned int vec_num = vec_oprnds0.length ();
/* Always true for SLP. */
gcc_assert (ncopies == 1);
tree len
@@ -3557,7 +3569,6 @@ vectorizable_call (vec_info *vinfo,
}
else if (mask_opno >= 0 && masked_loop_p)
{
- unsigned int vec_num = vec_oprnds0.length ();
/* Always true for SLP. */
gcc_assert (ncopies == 1);
tree mask = vect_get_loop_mask (loop_vinfo,
--
2.43.0
