This fixes double-scaling of the inner loop scalar cost caused
by routing the scalar costs through the add_stmt_cost hook and
using vect_body as the location. The issue makes almost every
outer loop vectorization profitable.
Bootstrapped and tested on x86_64-unknown-linux-gnu, pushed.
2021-08-23 Richard Biener <rguent...@suse.de>
* tree-vect-loop.c (vect_compute_single_scalar_iteration_cost):
Properly scale the inner loop cost only once.
---
gcc/tree-vect-loop.c | 12 ++++++++----
1 file changed, 8 insertions(+), 4 deletions(-)
diff --git a/gcc/tree-vect-loop.c b/gcc/tree-vect-loop.c
index 995d143dbbd..c521b43a47c 100644
--- a/gcc/tree-vect-loop.c
+++ b/gcc/tree-vect-loop.c
@@ -1284,6 +1284,8 @@ vect_compute_single_scalar_iteration_cost (loop_vec_info
loop_vinfo)
else
kind = scalar_stmt;
+ /* We are using vect_prologue here to avoid scaling twice
+ by the inner loop factor. */
record_stmt_cost (&LOOP_VINFO_SCALAR_ITERATION_COST (loop_vinfo),
factor, kind, stmt_info, 0, vect_prologue);
}
@@ -1297,11 +1299,13 @@ vect_compute_single_scalar_iteration_cost
(loop_vec_info loop_vinfo)
j, si)
(void) add_stmt_cost (loop_vinfo, target_cost_data, si->count,
si->kind, si->stmt_info, si->vectype,
- si->misalign, vect_body);
- unsigned dummy, body_cost = 0;
- finish_cost (target_cost_data, &dummy, &body_cost, &dummy);
+ si->misalign, si->where);
+ unsigned prologue_cost = 0, body_cost = 0, epilogue_cost = 0;
+ finish_cost (target_cost_data, &prologue_cost, &body_cost,
+ &epilogue_cost);
destroy_cost_data (target_cost_data);
- LOOP_VINFO_SINGLE_SCALAR_ITERATION_COST (loop_vinfo) = body_cost;
+ LOOP_VINFO_SINGLE_SCALAR_ITERATION_COST (loop_vinfo)
+ = prologue_cost + body_cost + epilogue_cost;
}
--
2.31.1
