From: Robin Dapp <[email protected]>
With the vectorizer now passing vect_gather_load and vect_scatter_store,
the aarch64 code needs a bit of updating, which this patch does.
gcc/ChangeLog:
* config/aarch64/aarch64.cc (aarch64_detect_vector_stmt_subtype):
Simplify gather/scatter detection.
(aarch64_vector_costs::count_ops): Get units from vectype
instead of count.
(aarch64_vector_costs::add_stmt_cost): Simplify detection.
---
gcc/config/aarch64/aarch64.cc | 30 ++++++++++++------------------
1 file changed, 12 insertions(+), 18 deletions(-)
diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc
index 6c1229d0a95..4dfb205ee77 100644
--- a/gcc/config/aarch64/aarch64.cc
+++ b/gcc/config/aarch64/aarch64.cc
@@ -18398,27 +18398,22 @@ aarch64_detect_vector_stmt_subtype (vec_info *vinfo,
vect_cost_for_stmt kind,
if (vect_is_store_elt_extraction (kind, stmt_info))
return simd_costs->store_elt_extra_cost;
- /* Detect SVE gather loads, which are costed as a single scalar_load
- for each element. We therefore need to divide the full-instruction
- cost by the number of elements in the vector. */
- if (kind == scalar_load
+ /* Detect SVE gather loads. */
+ if (kind == vector_gather_load
&& node
- && sve_costs
- && mat_gather_scatter_p (SLP_TREE_MEMORY_ACCESS_TYPE (node)))
+ && sve_costs)
{
- unsigned int nunits = vect_nunits_for_cost (vectype);
/* Test for VNx2 modes, which have 64-bit containers. */
if (known_eq (GET_MODE_NUNITS (TYPE_MODE (vectype)), aarch64_sve_vg))
- return { sve_costs->gather_load_x64_cost, nunits };
- return { sve_costs->gather_load_x32_cost, nunits };
+ return sve_costs->gather_load_x64_cost;
+ return sve_costs->gather_load_x32_cost;
}
/* Detect cases in which a scalar_store is really storing one element
in a scatter operation. */
- if (kind == scalar_store
+ if (kind == vector_scatter_store
&& node
- && sve_costs
- && mat_gather_scatter_p (SLP_TREE_MEMORY_ACCESS_TYPE (node)))
+ && sve_costs)
return sve_costs->scatter_store_elt_cost;
/* Detect cases in which vec_to_scalar represents an in-loop reduction. */
@@ -18872,10 +18867,10 @@ aarch64_vector_costs::count_ops (unsigned int count,
vect_cost_for_stmt kind,
/* Add any overhead associated with gather loads and scatter stores. */
if (stmt_info
&& sve_issue
- && (kind == scalar_load || kind == scalar_store)
- && mat_gather_scatter_p (SLP_TREE_MEMORY_ACCESS_TYPE (node)))
+ && (kind == vector_gather_load || kind == vector_scatter_store))
{
- unsigned int pairs = CEIL (count, 2);
+ unsigned int pairs = CEIL (vect_nunits_for_cost
+ (STMT_VINFO_VECTYPE (stmt_info)), 2);
ops->pred_ops += sve_issue->gather_scatter_pair_pred_ops * pairs;
ops->general_ops += sve_issue->gather_scatter_pair_general_ops * pairs;
}
@@ -19270,11 +19265,10 @@ aarch64_vector_costs::add_stmt_cost (int count,
vect_cost_for_stmt kind,
stmt_cost);
/* Check if we've seen an SVE gather/scatter operation and which size.
*/
- if (kind == scalar_load
+ if (kind == vector_gather_load
&& node
&& vectype
- && aarch64_sve_mode_p (TYPE_MODE (vectype))
- && mat_gather_scatter_p (SLP_TREE_MEMORY_ACCESS_TYPE (node)))
+ && aarch64_sve_mode_p (TYPE_MODE (vectype)))
{
const sve_vec_cost *sve_costs = aarch64_tune_params.vec_costs->sve;
if (sve_costs)
--
2.56.0
