public inbox for gcc-cvs@sourceware.org
help / color / mirror / Atom feed
* [gcc r12-2273] vect: Create array_slice of live-out stmts
@ 2021-07-13 9:18 Richard Sandiford
0 siblings, 0 replies; only message in thread
From: Richard Sandiford @ 2021-07-13 9:18 UTC (permalink / raw)
To: gcc-cvs
https://gcc.gnu.org/g:b68eb70bd6df8c4b846bddb4f0aeae9054b932bc
commit r12-2273-gb68eb70bd6df8c4b846bddb4f0aeae9054b932bc
Author: Richard Sandiford <richard.sandiford@arm.com>
Date: Tue Jul 13 10:17:36 2021 +0100
vect: Create array_slice of live-out stmts
This patch constructs an array_slice of the scalar statements that
produce live-out reduction results in the original unvectorised loop.
There are three cases:
- SLP reduction chains: the final SLP stmt is live-out
- full SLP reductions: all SLP stmts are live-out
- non-SLP reductions: the single scalar stmt is live-out
This is a slight simplification on its own, mostly because it maans
“group_size” has a consistent meaning throughout the function.
The main justification though is that it helps with later patches.
gcc/
* tree-vect-loop.c (vect_create_epilog_for_reduction): Truncate
scalar_results to group_size elements after reducing down from
N*group_size elements. Construct an array_slice of the live-out
stmts and assert that there is one stmt per scalar result.
Diff:
---
gcc/tree-vect-loop.c | 61 ++++++++++++++++++----------------------------------
1 file changed, 21 insertions(+), 40 deletions(-)
diff --git a/gcc/tree-vect-loop.c b/gcc/tree-vect-loop.c
index 7c3e3352b43..8390ac80ca0 100644
--- a/gcc/tree-vect-loop.c
+++ b/gcc/tree-vect-loop.c
@@ -5010,7 +5010,12 @@ vect_create_epilog_for_reduction (loop_vec_info loop_vinfo,
auto_vec<tree> scalar_results;
unsigned int group_size = 1, k;
auto_vec<gimple *> phis;
- bool slp_reduc = false;
+ /* SLP reduction without reduction chain, e.g.,
+ # a1 = phi <a2, a0>
+ # b1 = phi <b2, b0>
+ a2 = operation (a1)
+ b2 = operation (b1) */
+ bool slp_reduc = (slp_node && !REDUC_GROUP_FIRST_ELEMENT (stmt_info));
bool direct_slp_reduc;
tree new_phi_result;
tree induction_index = NULL_TREE;
@@ -5050,6 +5055,16 @@ vect_create_epilog_for_reduction (loop_vec_info loop_vinfo,
adjustment_def = STMT_VINFO_REDUC_EPILOGUE_ADJUSTMENT (reduc_info);
}
+ stmt_vec_info single_live_out_stmt[] = { stmt_info };
+ array_slice<const stmt_vec_info> live_out_stmts = single_live_out_stmt;
+ if (slp_reduc)
+ /* All statements produce live-out values. */
+ live_out_stmts = SLP_TREE_SCALAR_STMTS (slp_node);
+ else if (slp_node)
+ /* The last statement in the reduction chain produces the live-out
+ value. */
+ single_live_out_stmt[0] = SLP_TREE_SCALAR_STMTS (slp_node)[group_size - 1];
+
unsigned vec_num;
int ncopies;
if (slp_node)
@@ -5248,13 +5263,6 @@ vect_create_epilog_for_reduction (loop_vec_info loop_vinfo,
new_scalar_dest = vect_create_destination_var (scalar_dest, NULL);
bitsize = TYPE_SIZE (scalar_type);
- /* SLP reduction without reduction chain, e.g.,
- # a1 = phi <a2, a0>
- # b1 = phi <b2, b0>
- a2 = operation (a1)
- b2 = operation (b1) */
- slp_reduc = (slp_node && !REDUC_GROUP_FIRST_ELEMENT (stmt_info));
-
/* True if we should implement SLP_REDUC using native reduction operations
instead of scalar operations. */
direct_slp_reduc = (reduc_fn != IFN_LAST
@@ -5877,6 +5885,7 @@ vect_create_epilog_for_reduction (loop_vec_info loop_vinfo,
first_res, res);
scalar_results[j % group_size] = new_res;
}
+ scalar_results.truncate (group_size);
for (k = 0; k < group_size; k++)
scalar_results[k] = gimple_convert (&stmts, scalar_type,
scalar_results[k]);
@@ -5969,39 +5978,11 @@ vect_create_epilog_for_reduction (loop_vec_info loop_vinfo,
use <s_out4>
use <s_out4> */
-
- /* In SLP reduction chain we reduce vector results into one vector if
- necessary, hence we set here REDUC_GROUP_SIZE to 1. SCALAR_DEST is the
- LHS of the last stmt in the reduction chain, since we are looking for
- the loop exit phi node. */
- if (REDUC_GROUP_FIRST_ELEMENT (stmt_info))
- {
- stmt_vec_info dest_stmt_info
- = vect_orig_stmt (SLP_TREE_SCALAR_STMTS (slp_node)[group_size - 1]);
- scalar_dest = gimple_assign_lhs (dest_stmt_info->stmt);
- group_size = 1;
- }
-
- /* In SLP we may have several statements in NEW_PHIS and REDUCTION_PHIS (in
- case that REDUC_GROUP_SIZE is greater than vectorization factor).
- Therefore, we need to match SCALAR_RESULTS with corresponding statements.
- The first (REDUC_GROUP_SIZE / number of new vector stmts) scalar results
- correspond to the first vector stmt, etc.
- (RATIO is equal to (REDUC_GROUP_SIZE / number of new vector stmts)). */
- if (group_size > new_phis.length ())
- gcc_assert (!(group_size % new_phis.length ()));
-
- for (k = 0; k < group_size; k++)
+ gcc_assert (live_out_stmts.size () == scalar_results.length ());
+ for (k = 0; k < live_out_stmts.size (); k++)
{
- if (slp_reduc)
- {
- stmt_vec_info scalar_stmt_info = SLP_TREE_SCALAR_STMTS (slp_node)[k];
-
- orig_stmt_info = STMT_VINFO_RELATED_STMT (scalar_stmt_info);
- /* SLP statements can't participate in patterns. */
- gcc_assert (!orig_stmt_info);
- scalar_dest = gimple_assign_lhs (scalar_stmt_info->stmt);
- }
+ stmt_vec_info scalar_stmt_info = vect_orig_stmt (live_out_stmts[k]);
+ scalar_dest = gimple_assign_lhs (scalar_stmt_info->stmt);
phis.create (3);
/* Find the loop-closed-use at the loop exit of the original scalar
^ permalink raw reply [flat|nested] only message in thread
only message in thread, other threads:[~2021-07-13 9:18 UTC | newest]
Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-07-13 9:18 [gcc r12-2273] vect: Create array_slice of live-out stmts Richard Sandiford
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).