On Tue, 29 Jul 2025, Richard Biener wrote:
> This should be present only on SLP nodes now. The RISC-V changes
> are mechanical along the line of the SLP_TREE_TYPE changes. The
> aarch64 changes are for removing vect_mem_access_type and gating
> one problematical test with !m_costing_for_scalar (rather than
> on a NULL node).
>
> Bootstrapped and tested on x86_64-unknown-linux-gnu.
I have now pushed this variant.
Richard.
> * tree-vectorizer.h (_stmt_vec_info::memory_access_type): Remove.
> (STMT_VINFO_MEMORY_ACCESS_TYPE): Likewise.
> (vect_mem_access_type): Likewise.
> * tree-vect-stmts.cc (vectorizable_store): Do not set
> STMT_VINFO_MEMORY_ACCESS_TYPE. Fix SLP_TREE_MEMORY_ACCESS_TYPE
> usage.
> * tree-vect-loop.cc (update_epilogue_loop_vinfo): Remove
> checking of memory access type.
> * config/riscv/riscv-vector-costs.cc (costs::compute_local_live_ranges):
> Use SLP_TREE_MEMORY_ACCESS_TYPE.
> (costs::need_additional_vector_vars_p): Likewise.
> (segment_loadstore_group_size): Get SLP node as argument,
> use SLP_TREE_MEMORY_ACCESS_TYPE.
> (costs::adjust_stmt_cost): Pass down SLP node.
> * config/aarch64/aarch64.cc (aarch64_ld234_st234_vectors): Use
> SLP_TREE_MEMORY_ACCESS_TYPE instead of vect_mem_access_type.
> (aarch64_detect_vector_stmt_subtype): Likewise.
> (aarch64_vector_costs::count_ops): Likewise.
> (aarch64_vector_costs::add_stmt_cost): Likewise.
> ---
> gcc/config/aarch64/aarch64.cc | 15 ++++++++-------
> gcc/config/riscv/riscv-vector-costs.cc | 16 ++++++++--------
> gcc/tree-vect-loop.cc | 6 ++----
> gcc/tree-vect-stmts.cc | 3 +--
> gcc/tree-vectorizer.h | 17 -----------------
> 5 files changed, 19 insertions(+), 38 deletions(-)
>
> diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc
> index cb1699ab7c5..8acf24947ea 100644
> --- a/gcc/config/aarch64/aarch64.cc
> +++ b/gcc/config/aarch64/aarch64.cc
> @@ -17166,8 +17166,8 @@ aarch64_ld234_st234_vectors (vect_cost_for_stmt kind,
> stmt_vec_info stmt_info,
> && STMT_VINFO_DATA_REF (stmt_info))
> {
> stmt_info = DR_GROUP_FIRST_ELEMENT (stmt_info);
> - if (stmt_info
> - && vect_mem_access_type (stmt_info, node) == VMAT_LOAD_STORE_LANES)
> + if (node
> + && SLP_TREE_MEMORY_ACCESS_TYPE (node) == VMAT_LOAD_STORE_LANES)
> return DR_GROUP_SIZE (stmt_info);
> }
> return 0;
> @@ -17439,7 +17439,7 @@ aarch64_detect_vector_stmt_subtype (vec_info *vinfo,
> vect_cost_for_stmt kind,
> cost by the number of elements in the vector. */
> if (kind == scalar_load
> && sve_costs
> - && vect_mem_access_type (stmt_info, node) == VMAT_GATHER_SCATTER)
> + && SLP_TREE_MEMORY_ACCESS_TYPE (node) == VMAT_GATHER_SCATTER)
> {
> unsigned int nunits = vect_nunits_for_cost (vectype);
> /* Test for VNx2 modes, which have 64-bit containers. */
> @@ -17452,7 +17452,7 @@ aarch64_detect_vector_stmt_subtype (vec_info *vinfo,
> vect_cost_for_stmt kind,
> in a scatter operation. */
> if (kind == scalar_store
> && sve_costs
> - && vect_mem_access_type (stmt_info, node) == VMAT_GATHER_SCATTER)
> + && SLP_TREE_MEMORY_ACCESS_TYPE (node) == VMAT_GATHER_SCATTER)
> return sve_costs->scatter_store_elt_cost;
>
> /* Detect cases in which vec_to_scalar represents an in-loop reduction. */
> @@ -17708,7 +17708,7 @@ aarch64_vector_costs::count_ops (unsigned int count,
> vect_cost_for_stmt kind,
> if (stmt_info
> && kind == vec_to_scalar
> && (m_vec_flags & VEC_ADVSIMD)
> - && vect_mem_access_type (stmt_info, node) == VMAT_GATHER_SCATTER)
> + && SLP_TREE_MEMORY_ACCESS_TYPE (node) == VMAT_GATHER_SCATTER)
> {
> auto dr = STMT_VINFO_DATA_REF (stmt_info);
> tree dr_ref = DR_REF (dr);
> @@ -17823,7 +17823,7 @@ aarch64_vector_costs::count_ops (unsigned int count,
> vect_cost_for_stmt kind,
> if (stmt_info
> && sve_issue
> && (kind == scalar_load || kind == scalar_store)
> - && vect_mem_access_type (stmt_info, node) == VMAT_GATHER_SCATTER)
> + && SLP_TREE_MEMORY_ACCESS_TYPE (node) == VMAT_GATHER_SCATTER)
> {
> unsigned int pairs = CEIL (count, 2);
> ops->pred_ops += sve_issue->gather_scatter_pair_pred_ops * pairs;
> @@ -17978,9 +17978,10 @@ aarch64_vector_costs::add_stmt_cost (int count,
> vect_cost_for_stmt kind,
>
> /* Check if we've seen an SVE gather/scatter operation and which size.
> */
> if (kind == scalar_load
> + && !m_costing_for_scalar
> && vectype
> && aarch64_sve_mode_p (TYPE_MODE (vectype))
> - && vect_mem_access_type (stmt_info, node) == VMAT_GATHER_SCATTER)
> + && SLP_TREE_MEMORY_ACCESS_TYPE (node) == VMAT_GATHER_SCATTER)
> {
> const sve_vec_cost *sve_costs = aarch64_tune_params.vec_costs->sve;
> if (sve_costs)
> diff --git a/gcc/config/riscv/riscv-vector-costs.cc
> b/gcc/config/riscv/riscv-vector-costs.cc
> index 1c6bc25c2ad..44ef44a1435 100644
> --- a/gcc/config/riscv/riscv-vector-costs.cc
> +++ b/gcc/config/riscv/riscv-vector-costs.cc
> @@ -400,7 +400,7 @@ costs::compute_local_live_ranges (
> pair &live_range
> = live_ranges->get_or_insert (lhs, &existed_p);
> gcc_assert (!existed_p);
> - if (STMT_VINFO_MEMORY_ACCESS_TYPE (program_point.stmt_info)
> + if (SLP_TREE_MEMORY_ACCESS_TYPE (*node)
> == VMAT_LOAD_STORE_LANES)
> point = get_first_lane_point (program_points,
> program_point.stmt_info);
> @@ -418,8 +418,7 @@ costs::compute_local_live_ranges (
> bool existed_p = false;
> pair &live_range
> = live_ranges->get_or_insert (var, &existed_p);
> - if (STMT_VINFO_MEMORY_ACCESS_TYPE (
> - program_point.stmt_info)
> + if (SLP_TREE_MEMORY_ACCESS_TYPE (*node)
> == VMAT_LOAD_STORE_LANES)
> point = get_last_lane_point (program_points,
> program_point.stmt_info);
> @@ -608,7 +607,7 @@ costs::need_additional_vector_vars_p (stmt_vec_info
> stmt_info,
> if (type == load_vec_info_type || type == store_vec_info_type)
> {
> if (STMT_VINFO_GATHER_SCATTER_P (stmt_info)
> - && STMT_VINFO_MEMORY_ACCESS_TYPE (stmt_info) == VMAT_GATHER_SCATTER)
> + && SLP_TREE_MEMORY_ACCESS_TYPE (node) == VMAT_GATHER_SCATTER)
> return true;
>
> machine_mode mode = TYPE_MODE (STMT_VINFO_VECTYPE (stmt_info));
> @@ -1086,7 +1085,7 @@ costs::better_main_loop_than_p (const vector_costs
> *uncast_other) const
> load/store. */
> static int
> segment_loadstore_group_size (enum vect_cost_for_stmt kind,
> - stmt_vec_info stmt_info)
> + stmt_vec_info stmt_info, slp_tree node)
> {
> if (stmt_info
> && (kind == vector_load || kind == vector_store)
> @@ -1094,7 +1093,7 @@ segment_loadstore_group_size (enum vect_cost_for_stmt
> kind,
> {
> stmt_info = DR_GROUP_FIRST_ELEMENT (stmt_info);
> if (stmt_info
> - && STMT_VINFO_MEMORY_ACCESS_TYPE (stmt_info) == VMAT_LOAD_STORE_LANES)
> + && SLP_TREE_MEMORY_ACCESS_TYPE (node) == VMAT_LOAD_STORE_LANES)
> return DR_GROUP_SIZE (stmt_info);
> }
> return 0;
> @@ -1108,7 +1107,7 @@ segment_loadstore_group_size (enum vect_cost_for_stmt
> kind,
> unsigned
> costs::adjust_stmt_cost (enum vect_cost_for_stmt kind, loop_vec_info loop,
> stmt_vec_info stmt_info,
> - slp_tree, tree vectype, int stmt_cost)
> + slp_tree node, tree vectype, int stmt_cost)
> {
> const cpu_vector_cost *costs = get_vector_costs ();
> switch (kind)
> @@ -1131,7 +1130,8 @@ costs::adjust_stmt_cost (enum vect_cost_for_stmt kind,
> loop_vec_info loop,
> each vector in the group. Here we additionally add permute
> costs for each. */
> /* TODO: Indexed and ordered/unordered cost. */
> - int group_size = segment_loadstore_group_size (kind, stmt_info);
> + int group_size = segment_loadstore_group_size (kind, stmt_info,
> + node);
> if (group_size > 1)
> {
> switch (group_size)
> diff --git a/gcc/tree-vect-loop.cc b/gcc/tree-vect-loop.cc
> index 830fece33c6..ce08909ab83 100644
> --- a/gcc/tree-vect-loop.cc
> +++ b/gcc/tree-vect-loop.cc
> @@ -11294,11 +11294,9 @@ update_epilogue_loop_vinfo (class loop *epilogue,
> tree advance)
> updated offset we set using ADVANCE. Instead we have to make sure the
> reference in the data references point to the corresponding copy of
> the original in the epilogue. Make sure to update both
> - gather/scatters recognized by dataref analysis and also other
> - refs that get_load_store_type classified as VMAT_GATHER_SCATTER. */
> + gather/scatters recognized by dataref analysis. */
> auto vstmt_vinfo = vect_stmt_to_vectorize (stmt_vinfo);
> - if (STMT_VINFO_MEMORY_ACCESS_TYPE (vstmt_vinfo) == VMAT_GATHER_SCATTER
> - || STMT_VINFO_STRIDED_P (vstmt_vinfo)
> + if (STMT_VINFO_STRIDED_P (vstmt_vinfo)
> || STMT_VINFO_GATHER_SCATTER_P (vstmt_vinfo))
> {
> /* ??? As we copy epilogues from the main loop incremental
> diff --git a/gcc/tree-vect-stmts.cc b/gcc/tree-vect-stmts.cc
> index 4cf6c3b1a6f..1dfb148e803 100644
> --- a/gcc/tree-vect-stmts.cc
> +++ b/gcc/tree-vect-stmts.cc
> @@ -7934,7 +7934,6 @@ vectorizable_store (vec_info *vinfo,
> bool costing_p = cost_vec;
> if (costing_p) /* transformation not required. */
> {
> - STMT_VINFO_MEMORY_ACCESS_TYPE (stmt_info) = memory_access_type;
> SLP_TREE_MEMORY_ACCESS_TYPE (slp_node) = memory_access_type;
>
> if (loop_vinfo
> @@ -7965,7 +7964,7 @@ vectorizable_store (vec_info *vinfo,
>
> SLP_TREE_TYPE (slp_node) = store_vec_info_type;
> }
> - gcc_assert (memory_access_type == SLP_TREE_MEMORY_ACCESS_TYPE (stmt_info));
> + gcc_assert (memory_access_type == SLP_TREE_MEMORY_ACCESS_TYPE (slp_node));
>
> /* Transform. */
>
> diff --git a/gcc/tree-vectorizer.h b/gcc/tree-vectorizer.h
> index 76ff2cfea2f..7e7b2b4b836 100644
> --- a/gcc/tree-vectorizer.h
> +++ b/gcc/tree-vectorizer.h
> @@ -1437,10 +1437,6 @@ public:
> /* For both loads and stores. */
> unsigned simd_lane_access_p : 3;
>
> - /* Classifies how the load or store is going to be implemented
> - for loop vectorization. */
> - vect_memory_access_type memory_access_type;
> -
> /* For INTEGER_INDUC_COND_REDUCTION, the initial value to be used. */
> tree induc_cond_initial_val;
>
> @@ -1583,7 +1579,6 @@ struct gather_scatter_info {
> #define STMT_VINFO_DATA_REF(S) ((S)->dr_aux.dr + 0)
> #define STMT_VINFO_GATHER_SCATTER_P(S) (S)->gather_scatter_p
> #define STMT_VINFO_STRIDED_P(S) (S)->strided_p
> -#define STMT_VINFO_MEMORY_ACCESS_TYPE(S) (S)->memory_access_type
> #define STMT_VINFO_SIMD_LANE_ACCESS_P(S) (S)->simd_lane_access_p
> #define STMT_VINFO_VEC_INDUC_COND_INITIAL_VAL(S) (S)->induc_cond_initial_val
> #define STMT_VINFO_REDUC_EPILOGUE_ADJUSTMENT(S)
> (S)->reduc_epilogue_adjustment
> @@ -2832,18 +2827,6 @@ vect_is_reduction (stmt_vec_info stmt_info)
> return STMT_VINFO_REDUC_IDX (stmt_info) >= 0;
> }
>
> -/* Returns the memory acccess type being used to vectorize the statement. If
> - SLP this is read from NODE, otherwise it's read from the STMT_VINFO. */
> -
> -inline vect_memory_access_type
> -vect_mem_access_type (stmt_vec_info stmt_info, slp_tree node)
> -{
> - if (node)
> - return SLP_TREE_MEMORY_ACCESS_TYPE (node);
> - else
> - return STMT_VINFO_MEMORY_ACCESS_TYPE (stmt_info);
> -}
> -
> /* If STMT_INFO describes a reduction, return the vect_reduction_type
> of the reduction it describes, otherwise return -1. */
> inline int
>
--
Richard Biener <rguent...@suse.de>
SUSE Software Solutions Germany GmbH,
Frankenstrasse 146, 90461 Nuernberg, Germany;
GF: Ivo Totev, Andrew McDonald, Werner Knoblich; (HRB 36809, AG Nuernberg)
