re PR tree-optimization/81410 (-O3 breaks code)

2017-06-18  Richard Biener  <rguenther@suse.de>

	PR tree-optimization/81410
	* tree-vect-stmts.c (vectorizable_load): Properly adjust for
	the gap in the ! slp_perm SLP case after each group.

	* gcc.dg/vect/pr81410.c: New testcase.

From-SVN: r250312
This commit is contained in:
Richard Biener 2017-07-18 13:55:47 +00:00 committed by Richard Biener
parent 35896b18ec
commit b267968e67
4 changed files with 77 additions and 4 deletions

View File

@ -1,3 +1,9 @@
2017-06-18 Richard Biener <rguenther@suse.de>
PR tree-optimization/81410
* tree-vect-stmts.c (vectorizable_load): Properly adjust for
the gap in the ! slp_perm SLP case after each group.
2017-07-18 Jan Hubicka <hubicka@ucw.cz>
PR middle-end/81463

View File

@ -1,3 +1,8 @@
2017-06-18 Richard Biener <rguenther@suse.de>
PR tree-optimization/81410
* gcc.dg/vect/pr81410.c: New testcase.
2017-07-18 Jan Hubicka <hubicka@ucw.cz>
PR middle-end/81462

View File

@ -0,0 +1,38 @@
/* { dg-do run } */
/* { dg-require-effective-target vect_long_long } */
#include "tree-vect.h"
typedef long long uint64_t;
uint64_t x[24];
uint64_t y[16];
uint64_t z[8];
void __attribute__((noinline)) foo()
{
for (int i = 0; i < 8; ++i)
{
y[2*i] = x[3*i];
y[2*i + 1] = x[3*i + 1];
z[i] = 1;
}
}
int main()
{
check_vect ();
for (int i = 0; i < 24; ++i)
{
x[i] = i;
__asm__ volatile ("" : : : "memory");
}
foo ();
for (int i = 0; i < 8; ++i)
if (y[2*i] != 3*i || y[2*i+1] != 3*i + 1)
__builtin_abort ();
return 0;
}
/* { dg-final { scan-tree-dump "vectorized 1 loops" "vect" } } */

View File

@ -7118,6 +7118,7 @@ vectorizable_load (gimple *stmt, gimple_stmt_iterator *gsi, gimple **vec_stmt,
{
first_stmt = GROUP_FIRST_ELEMENT (stmt_info);
group_size = GROUP_SIZE (vinfo_for_stmt (first_stmt));
int group_gap = GROUP_GAP (vinfo_for_stmt (first_stmt));
/* For SLP vectorization we directly vectorize a subchain
without permutation. */
if (slp && ! SLP_TREE_LOAD_PERMUTATION (slp_node).exists ())
@ -7153,10 +7154,15 @@ vectorizable_load (gimple *stmt, gimple_stmt_iterator *gsi, gimple **vec_stmt,
not only the number of vector stmts the permutation result
fits in. */
if (slp_perm)
vec_num = (group_size * vf + nunits - 1) / nunits;
{
vec_num = (group_size * vf + nunits - 1) / nunits;
group_gap_adj = vf * group_size - nunits * vec_num;
}
else
vec_num = SLP_TREE_NUMBER_OF_VEC_STMTS (slp_node);
group_gap_adj = vf * group_size - nunits * vec_num;
{
vec_num = SLP_TREE_NUMBER_OF_VEC_STMTS (slp_node);
group_gap_adj = group_gap;
}
}
else
vec_num = group_size;
@ -7316,6 +7322,7 @@ vectorizable_load (gimple *stmt, gimple_stmt_iterator *gsi, gimple **vec_stmt,
aggr_type = vectype;
prev_stmt_info = NULL;
int group_elt = 0;
for (j = 0; j < ncopies; j++)
{
/* 1. Create the vector or array pointer update chain. */
@ -7603,10 +7610,27 @@ vectorizable_load (gimple *stmt, gimple_stmt_iterator *gsi, gimple **vec_stmt,
/* Store vector loads in the corresponding SLP_NODE. */
if (slp && !slp_perm)
SLP_TREE_VEC_STMTS (slp_node).quick_push (new_stmt);
/* With SLP permutation we load the gaps as well, without
we need to skip the gaps after we manage to fully load
all elements. group_gap_adj is GROUP_SIZE here. */
group_elt += nunits;
if (group_gap_adj != 0 && ! slp_perm
&& group_elt == group_size - group_gap_adj)
{
bool ovf;
tree bump
= wide_int_to_tree (sizetype,
wi::smul (TYPE_SIZE_UNIT (elem_type),
group_gap_adj, &ovf));
dataref_ptr = bump_vector_ptr (dataref_ptr, ptr_incr, gsi,
stmt, bump);
group_elt = 0;
}
}
/* Bump the vector pointer to account for a gap or for excess
elements loaded for a permuted SLP load. */
if (group_gap_adj != 0)
if (group_gap_adj != 0 && slp_perm)
{
bool ovf;
tree bump