Commit 7921a90e by Richard Biener Committed by Richard Biener

re PR tree-optimization/91178 (Infinite recursion in split_constant_offset in slp after r260289)

2019-07-17  Richard Biener  <rguenther@suse.de>

	PR tree-optimization/91178
	* tree-vect-stmts.c (get_group_load_store_type): For SLP
	loads with a gap larger than the vector size always use
	VMAT_STRIDED_SLP.
	(vectorizable_load): For VMAT_STRIDED_SLP with a permutation
	avoid loading vectors that are only contained in the gap
	and thus are not needed.

	* gcc.dg/torture/pr91178.c: New testcase.

From-SVN: r273549
parent 9505acd8
2019-07-17 Richard Biener <rguenther@suse.de> 2019-07-17 Richard Biener <rguenther@suse.de>
PR tree-optimization/91178
* tree-vect-stmts.c (get_group_load_store_type): For SLP
loads with a gap larger than the vector size always use
VMAT_STRIDED_SLP.
(vectorizable_load): For VMAT_STRIDED_SLP with a permutation
avoid loading vectors that are only contained in the gap
and thus are not needed.
2019-07-17 Richard Biener <rguenther@suse.de>
PR tree-optimization/91180 PR tree-optimization/91180
* tree-ssa-sccvn.c (vn_reference_lookup_3): Fix offset * tree-ssa-sccvn.c (vn_reference_lookup_3): Fix offset
computation for memset partial defs. computation for memset partial defs.
......
2019-07-17 Richard Biener <rguenther@suse.de> 2019-07-17 Richard Biener <rguenther@suse.de>
PR tree-optimization/91178
* gcc.dg/torture/pr91178.c: New testcase.
2019-07-17 Richard Biener <rguenther@suse.de>
PR tree-optimization/91180 PR tree-optimization/91180
* gcc.dg/torture/pr91180.c: New testcase. * gcc.dg/torture/pr91180.c: New testcase.
......
/* { dg-do compile } */
int a;
extern int f[10][91125];
int b[50];
void c()
{
for (int d = 6; d <= a; d++)
for (int e = 16; e <= 24; e++)
b[e] -= f[d][d];
}
...@@ -2267,6 +2267,14 @@ get_group_load_store_type (stmt_vec_info stmt_info, tree vectype, bool slp, ...@@ -2267,6 +2267,14 @@ get_group_load_store_type (stmt_vec_info stmt_info, tree vectype, bool slp,
/ vect_get_scalar_dr_size (first_dr_info))) / vect_get_scalar_dr_size (first_dr_info)))
overrun_p = false; overrun_p = false;
/* If the gap at the end of the group exceeds a whole vector
in size use the strided SLP code which can skip code-generation
for the gap. */
if (vls_type == VLS_LOAD && known_gt (gap, nunits))
*memory_access_type = VMAT_STRIDED_SLP;
else
*memory_access_type = VMAT_CONTIGUOUS;
/* If the gap splits the vector in half and the target /* If the gap splits the vector in half and the target
can do half-vector operations avoid the epilogue peeling can do half-vector operations avoid the epilogue peeling
by simply loading half of the vector only. Usually by simply loading half of the vector only. Usually
...@@ -2274,7 +2282,8 @@ get_group_load_store_type (stmt_vec_info stmt_info, tree vectype, bool slp, ...@@ -2274,7 +2282,8 @@ get_group_load_store_type (stmt_vec_info stmt_info, tree vectype, bool slp,
dr_alignment_support alignment_support_scheme; dr_alignment_support alignment_support_scheme;
scalar_mode elmode = SCALAR_TYPE_MODE (TREE_TYPE (vectype)); scalar_mode elmode = SCALAR_TYPE_MODE (TREE_TYPE (vectype));
machine_mode vmode; machine_mode vmode;
if (overrun_p if (*memory_access_type == VMAT_CONTIGUOUS
&& overrun_p
&& !masked_p && !masked_p
&& (((alignment_support_scheme && (((alignment_support_scheme
= vect_supportable_dr_alignment (first_dr_info, false))) = vect_supportable_dr_alignment (first_dr_info, false)))
...@@ -2297,7 +2306,6 @@ get_group_load_store_type (stmt_vec_info stmt_info, tree vectype, bool slp, ...@@ -2297,7 +2306,6 @@ get_group_load_store_type (stmt_vec_info stmt_info, tree vectype, bool slp,
"Peeling for outer loop is not supported\n"); "Peeling for outer loop is not supported\n");
return false; return false;
} }
*memory_access_type = VMAT_CONTIGUOUS;
} }
} }
else else
...@@ -8732,6 +8740,7 @@ vectorizable_load (stmt_vec_info stmt_info, gimple_stmt_iterator *gsi, ...@@ -8732,6 +8740,7 @@ vectorizable_load (stmt_vec_info stmt_info, gimple_stmt_iterator *gsi,
/* Checked by get_load_store_type. */ /* Checked by get_load_store_type. */
unsigned int const_nunits = nunits.to_constant (); unsigned int const_nunits = nunits.to_constant ();
unsigned HOST_WIDE_INT cst_offset = 0; unsigned HOST_WIDE_INT cst_offset = 0;
unsigned int group_gap = 0;
gcc_assert (!LOOP_VINFO_FULLY_MASKED_P (loop_vinfo)); gcc_assert (!LOOP_VINFO_FULLY_MASKED_P (loop_vinfo));
gcc_assert (!nested_in_vect_loop); gcc_assert (!nested_in_vect_loop);
...@@ -8749,6 +8758,7 @@ vectorizable_load (stmt_vec_info stmt_info, gimple_stmt_iterator *gsi, ...@@ -8749,6 +8758,7 @@ vectorizable_load (stmt_vec_info stmt_info, gimple_stmt_iterator *gsi,
if (slp && grouped_load) if (slp && grouped_load)
{ {
group_size = DR_GROUP_SIZE (first_stmt_info); group_size = DR_GROUP_SIZE (first_stmt_info);
group_gap = DR_GROUP_GAP (first_stmt_info);
ref_type = get_group_alias_ptr_type (first_stmt_info); ref_type = get_group_alias_ptr_type (first_stmt_info);
} }
else else
...@@ -8892,6 +8902,14 @@ vectorizable_load (stmt_vec_info stmt_info, gimple_stmt_iterator *gsi, ...@@ -8892,6 +8902,14 @@ vectorizable_load (stmt_vec_info stmt_info, gimple_stmt_iterator *gsi,
if (nloads > 1) if (nloads > 1)
vec_alloc (v, nloads); vec_alloc (v, nloads);
stmt_vec_info new_stmt_info = NULL; stmt_vec_info new_stmt_info = NULL;
if (slp && slp_perm
&& (group_el % group_size) > group_size - group_gap
&& (group_el % group_size) + nloads * lnel < group_size)
{
dr_chain.quick_push (NULL_TREE);
group_el += nloads * lnel;
continue;
}
for (i = 0; i < nloads; i++) for (i = 0; i < nloads; i++)
{ {
tree this_off = build_int_cst (TREE_TYPE (alias_off), tree this_off = build_int_cst (TREE_TYPE (alias_off),
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment