On Fri, Jul 11, 2025 at 11:03 AM Robin Dapp <rdapp....@gmail.com> wrote:
>
> This encapsulates the IFN and the builtin-function way of handling
> gather/scatter via three defines:
>
> GATHER_SCATTER_IFN_P
> GATHER_SCATTER_LEGACY_P
> GATHER_SCATTER_EMULATED_P
>
> and introduces a helper define for SLP operand handling as well.
OK.
> gcc/ChangeLog:
>
> * tree-vect-slp.cc (GATHER_SCATTER_OFFSET): New define.
> (vect_get_and_check_slp_defs): Use.
> * tree-vectorizer.h (GATHER_SCATTER_LEGACY_P): New define.
> (GATHER_SCATTER_IFN_P): Ditto.
> (GATHER_SCATTER_EMULATED_P): Ditto.
> * tree-vect-stmts.cc (vectorizable_store): Use.
> (vectorizable_load): Use.
> ---
> gcc/tree-vect-slp.cc | 12 +++++++-----
> gcc/tree-vect-stmts.cc | 19 +++++++++----------
> gcc/tree-vectorizer.h | 8 ++++++++
> 3 files changed, 24 insertions(+), 15 deletions(-)
>
> diff --git a/gcc/tree-vect-slp.cc b/gcc/tree-vect-slp.cc
> index ad75386926a..0c95ed946bb 100644
> --- a/gcc/tree-vect-slp.cc
> +++ b/gcc/tree-vect-slp.cc
> @@ -507,6 +507,8 @@ vect_def_types_match (enum vect_def_type dta, enum
> vect_def_type dtb)
> && (dtb == vect_external_def || dtb == vect_constant_def)));
> }
>
> +#define GATHER_SCATTER_OFFSET (-3)
> +
> static const int no_arg_map[] = { 0 };
> static const int arg0_map[] = { 1, 0 };
> static const int arg1_map[] = { 1, 1 };
> @@ -516,10 +518,10 @@ static const int arg1_arg4_arg5_map[] = { 3, 1, 4, 5 };
> static const int arg1_arg3_arg4_map[] = { 3, 1, 3, 4 };
> static const int arg3_arg2_map[] = { 2, 3, 2 };
> static const int op1_op0_map[] = { 2, 1, 0 };
> -static const int off_map[] = { 1, -3 };
> -static const int off_op0_map[] = { 2, -3, 0 };
> -static const int off_arg2_arg3_map[] = { 3, -3, 2, 3 };
> -static const int off_arg3_arg2_map[] = { 3, -3, 3, 2 };
> +static const int off_map[] = { 1, GATHER_SCATTER_OFFSET };
> +static const int off_op0_map[] = { 2, GATHER_SCATTER_OFFSET, 0 };
> +static const int off_arg2_arg3_map[] = { 3, GATHER_SCATTER_OFFSET, 2, 3 };
> +static const int off_arg3_arg2_map[] = { 3, GATHER_SCATTER_OFFSET, 3, 2 };
> static const int mask_call_maps[6][7] = {
> { 1, 1, },
> { 2, 1, 2, },
> @@ -691,7 +693,7 @@ vect_get_and_check_slp_defs (vec_info *vinfo, unsigned
> char swap,
> {
> oprnd_info = (*oprnds_info)[i];
> int opno = map ? map[i] : int (i);
> - if (opno == -3)
> + if (opno == GATHER_SCATTER_OFFSET)
> {
> gcc_assert (STMT_VINFO_GATHER_SCATTER_P (stmt_info));
> if (!is_a <loop_vec_info> (vinfo)
> diff --git a/gcc/tree-vect-stmts.cc b/gcc/tree-vect-stmts.cc
> index 4aa69da2218..57942f43c3b 100644
> --- a/gcc/tree-vect-stmts.cc
> +++ b/gcc/tree-vect-stmts.cc
> @@ -2455,7 +2455,7 @@ get_load_store_type (vec_info *vinfo, stmt_vec_info
> stmt_info,
> If that failed for some reason (e.g. because another pattern
> took priority), just handle cases in which the offset already
> has the right type. */
> - else if (gs_info->ifn != IFN_LAST
> + else if (GATHER_SCATTER_IFN_P (*gs_info)
> && !is_gimple_call (stmt_info->stmt)
> && !tree_nop_conversion_p (TREE_TYPE (gs_info->offset),
> TREE_TYPE
> (gs_info->offset_vectype)))
> @@ -8368,7 +8368,8 @@ vectorizable_store (vec_info *vinfo,
> }
> else if (memory_access_type != VMAT_LOAD_STORE_LANES
> && (memory_access_type != VMAT_GATHER_SCATTER
> - || (gs_info.decl && !VECTOR_BOOLEAN_TYPE_P
> (mask_vectype))))
> + || (GATHER_SCATTER_LEGACY_P (gs_info)
> + && !VECTOR_BOOLEAN_TYPE_P (mask_vectype))))
> {
> if (dump_enabled_p ())
> dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
> @@ -8376,8 +8377,7 @@ vectorizable_store (vec_info *vinfo,
> return false;
> }
> else if (memory_access_type == VMAT_GATHER_SCATTER
> - && gs_info.ifn == IFN_LAST
> - && !gs_info.decl)
> + && GATHER_SCATTER_EMULATED_P (gs_info))
> {
> if (dump_enabled_p ())
> dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
> @@ -9103,7 +9103,7 @@ vectorizable_store (vec_info *vinfo,
> final_mask, vec_mask, gsi);
> }
>
> - if (gs_info.ifn != IFN_LAST)
> + if (GATHER_SCATTER_IFN_P (gs_info))
> {
> if (costing_p)
> {
> @@ -9166,7 +9166,7 @@ vectorizable_store (vec_info *vinfo,
> vect_finish_stmt_generation (vinfo, stmt_info, call, gsi);
> new_stmt = call;
> }
> - else if (gs_info.decl)
> + else if (GATHER_SCATTER_LEGACY_P (gs_info))
> {
> /* The builtin decls path for scatter is legacy, x86 only. */
> gcc_assert (nunits.is_constant ()
> @@ -10077,8 +10077,7 @@ vectorizable_load (vec_info *vinfo,
> return false;
> }
> else if (memory_access_type == VMAT_GATHER_SCATTER
> - && gs_info.ifn == IFN_LAST
> - && !gs_info.decl)
> + && GATHER_SCATTER_EMULATED_P (gs_info))
> {
> if (dump_enabled_p ())
> dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
> @@ -11040,7 +11039,7 @@ vectorizable_load (vec_info *vinfo,
>
> /* 2. Create the vector-load in the loop. */
> unsigned HOST_WIDE_INT align;
> - if (gs_info.ifn != IFN_LAST)
> + if (GATHER_SCATTER_IFN_P (gs_info))
> {
> if (costing_p)
> {
> @@ -11112,7 +11111,7 @@ vectorizable_load (vec_info *vinfo,
> new_stmt = call;
> data_ref = NULL_TREE;
> }
> - else if (gs_info.decl)
> + else if (GATHER_SCATTER_LEGACY_P (gs_info))
> {
> /* The builtin decls path for gather is legacy, x86 only. */
> gcc_assert (!final_len && nunits.is_constant ());
> diff --git a/gcc/tree-vectorizer.h b/gcc/tree-vectorizer.h
> index 799d5fed7a9..7b927491b1c 100644
> --- a/gcc/tree-vectorizer.h
> +++ b/gcc/tree-vectorizer.h
> @@ -1655,6 +1655,14 @@ struct gather_scatter_info {
> #define PURE_SLP_STMT(S) ((S)->slp_type == pure_slp)
> #define STMT_SLP_TYPE(S) (S)->slp_type
>
> +#define GATHER_SCATTER_LEGACY_P(info) ((info).decl != NULL_TREE \
> + && (info).ifn == IFN_LAST)
> +#define GATHER_SCATTER_IFN_P(info) ((info).decl == NULL_TREE \
> + && (info).ifn != IFN_LAST)
> +#define GATHER_SCATTER_EMULATED_P(info) ((info).decl == NULL_TREE \
> + && (info).ifn == IFN_LAST)
> +
> +
> /* Contains the scalar or vector costs for a vec_info. */
> class vector_costs
> {
> --
> 2.50.0
>
