When we are unlucky operand canonicalization can end up presenting
us with different order, making a possible SLP reduction group
not match up. The following allows swapping operands in this case.
Bootstrapped and tested on x86_64-unknown-linux-gnu, pushed.
* tree-vect-slp.cc (vect_get_operand_map): Handle commutative
operands when swapping is requested.
(vect_build_slp_tree_1): Allow STMT_VINFO_REDUC_IDX differences
when operand swapping makes them match and request swapping.
(vect_build_slp_instance): Indicate we have successfully
discovered a SLP reduction group.
* gcc.dg/vect/slp-reduc-13.c: New testcase.
Co-authored-by: Eric Botcazou <[email protected]>
---
gcc/testsuite/gcc.dg/vect/slp-reduc-13.c | 66 ++++++++++++++++++++++++
gcc/tree-vect-slp.cc | 23 ++++++++-
2 files changed, 87 insertions(+), 2 deletions(-)
create mode 100644 gcc/testsuite/gcc.dg/vect/slp-reduc-13.c
diff --git a/gcc/testsuite/gcc.dg/vect/slp-reduc-13.c
b/gcc/testsuite/gcc.dg/vect/slp-reduc-13.c
new file mode 100644
index 00000000000..00e91fc6251
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/vect/slp-reduc-13.c
@@ -0,0 +1,66 @@
+/* { dg-do compile } */
+/* { dg-require-effective-target vect_int } */
+/* { dg-additional-options "-fgimple" } */
+
+int q[2];
+
+void __GIMPLE (ssa,guessed_local(16535624),startwith("loop"))
+foo (int * r)
+{
+ int i;
+ int sum2;
+ int sum1;
+ int _1;
+ long unsigned int _2;
+ long unsigned int _3;
+ int * _4;
+ int _24;
+ __SIZETYPE__ _6;
+ __SIZETYPE__ _7;
+ int * _8;
+ int _9;
+ int _13;
+ unsigned int _30;
+ unsigned int _31;
+
+ __BB(2,guessed_local(16535624)):
+ goto __BB3(precise(134217728));
+
+ __BB(3,loop_header(1),guessed_local(1057206200)):
+ sum1_5 = __PHI (__BB5: sum1_18, __BB2: 0);
+ sum2_26 = __PHI (__BB5: sum2_19, __BB2: 0);
+ i_28 = __PHI (__BB5: i_20, __BB2: 0);
+ _31 = __PHI (__BB5: _30, __BB2: 64u);
+ _1 = i_28 * 2;
+ _2 = (long unsigned int) _1;
+ _3 = _2 * 4ul;
+ _4 = r_17(D) + _3;
+ _24 = __MEM <int> (_4);
+ /* Deliberately have swapped operands here */
+ sum1_18 = sum1_5 + _24;
+ _13 = _1 + 1;
+ _6 = (__SIZETYPE__) _13;
+ _7 = _6 * 4ul;
+ _8 = r_17(D) + _7;
+ _9 = __MEM <int> (_8);
+ /* versus here. */
+ sum2_19 = _9 + sum2_26;
+ i_20 = i_28 + 1;
+ _30 = _31 - 1u;
+ if (_30 != 0u)
+ goto __BB5(guessed(132118446));
+ else
+ goto __BB4(guessed(2099282));
+
+ __BB(5,guessed_local(1040670576)):
+ goto __BB3(precise(134217728));
+
+ __BB(4,guessed_local(16535624)):
+ sum1_33 = __PHI (__BB3: sum1_18);
+ sum2_32 = __PHI (__BB3: sum2_19);
+ q[0] = sum1_33;
+ q[1] = sum2_32;
+ return;
+}
+
+/* { dg-final { scan-tree-dump "SLP discovery of size 2 reduction group
succeeded" "vect" } } */
diff --git a/gcc/tree-vect-slp.cc b/gcc/tree-vect-slp.cc
index f64f874abff..aa6c3e2e041 100644
--- a/gcc/tree-vect-slp.cc
+++ b/gcc/tree-vect-slp.cc
@@ -558,7 +558,8 @@ vect_get_operand_map (const gimple *stmt, bool
gather_scatter_p = false,
if (gimple_assign_rhs_code (assign) == COND_EXPR
&& COMPARISON_CLASS_P (gimple_assign_rhs1 (assign)))
gcc_unreachable ();
- if (TREE_CODE_CLASS (gimple_assign_rhs_code (assign)) == tcc_comparison
+ if ((TREE_CODE_CLASS (gimple_assign_rhs_code (assign)) == tcc_comparison
+ || commutative_tree_code (gimple_assign_rhs_code (assign)))
&& swap)
return op1_op0_map;
if (gather_scatter_p)
@@ -1352,7 +1353,12 @@ vect_build_slp_tree_1 (vec_info *vinfo, unsigned char
*swap,
uniform but only that of the first stmt matters. */
&& !(first_reduc_idx != -1
&& STMT_VINFO_REDUC_IDX (stmt_info) != -1
- && REDUC_GROUP_FIRST_ELEMENT (stmt_info)))
+ && REDUC_GROUP_FIRST_ELEMENT (stmt_info))
+ && !(first_reduc_idx != -1
+ && STMT_VINFO_REDUC_IDX (stmt_info) != -1
+ && rhs_code.is_tree_code ()
+ && commutative_tree_code (tree_code (rhs_code))
+ && first_reduc_idx == 1 - STMT_VINFO_REDUC_IDX (stmt_info)))
{
if (dump_enabled_p ())
{
@@ -1617,6 +1623,15 @@ vect_build_slp_tree_1 (vec_info *vinfo, unsigned char
*swap,
&& (swap_tree_comparison ((tree_code)first_stmt_code)
== (tree_code)rhs_code))
swap[i] = 1;
+
+ if (i != 0
+ && first_reduc_idx != STMT_VINFO_REDUC_IDX (stmt_info)
+ && first_reduc_idx != -1
+ && STMT_VINFO_REDUC_IDX (stmt_info) != -1
+ && rhs_code.is_tree_code ()
+ && commutative_tree_code (tree_code (rhs_code))
+ && first_reduc_idx == 1 - STMT_VINFO_REDUC_IDX (stmt_info))
+ swap[i] = 1;
}
matches[i] = true;
@@ -4164,6 +4179,10 @@ vect_build_slp_instance (vec_info *vinfo,
if (dump_enabled_p ())
{
+ if (kind == slp_inst_kind_reduc_group)
+ dump_printf_loc (MSG_NOTE, vect_location,
+ "SLP discovery of size %d reduction group "
+ "succeeded\n", group_size);
dump_printf_loc (MSG_NOTE, vect_location,
"Final SLP tree for instance %p:\n",
(void *) new_instance);
--
2.51.0
