On Mon, Jan 4, 2021 at 12:50 PM Richard Sandiford via Gcc-patches
<gcc-patches@gcc.gnu.org> wrote:
>
> The IFN_MASK* functions take two leading arguments: a load or
> store pointer and a “cookie”. The type of the cookie is the
> type of the access for TBAA purposes (like for MEM_REFs)
> while the value of the cookie is the alignment of the access.
> This PR was caused by a disagreement about whether the alignment
> is measured in bits or bytes.
>
> It looks like this goes back to PR68786, which made the
> vectoriser create its own cookie argument rather than reusing
> the one created by ifcvt. The alignment value of the new cookie
> was measured in bytes (as needed by set_ptr_info_alignment)
> while the existing code expected it to be measured in bits.
> The folds I added for IFN_MASK_LOAD and STORE then made
> things worse.
>
> Tested on aarch64-linux-gnu, aarch64_be-elf and x86_64-linux-gnu.
> OK for the explow bits?
OK.
Richard.
> Richard
>
>
> gcc/
> PR tree-vectorization/95401
> * config/aarch64/aarch64-sve-builtins.cc
> (gimple_folder::load_store_cookie): Use bits rather than bytes
> for the alignment argument to IFN_MASK_LOAD and IFN_MASK_STORE.
> * gimple-fold.c (gimple_fold_mask_load_store_mem_ref): Likewise.
> * tree-vect-stmts.c (vectorizable_store): Likewise.
> (vectorizable_load): Likewise.
>
> gcc/testsuite/
> PR tree-vectorization/95401
> * g++.dg/vect/pr95401.cc: New test.
> * g++.dg/vect/pr95401a.cc: Likewise.
> ---
> gcc/config/aarch64/aarch64-sve-builtins.cc | 2 +-
> gcc/gimple-fold.c | 2 +-
> gcc/testsuite/g++.dg/vect/pr95401.cc | 13 +++++++++++++
> gcc/testsuite/g++.dg/vect/pr95401a.cc | 13 +++++++++++++
> gcc/tree-vect-stmts.c | 14 ++++++++------
> 5 files changed, 36 insertions(+), 8 deletions(-)
> create mode 100644 gcc/testsuite/g++.dg/vect/pr95401.cc
> create mode 100644 gcc/testsuite/g++.dg/vect/pr95401a.cc
>
> diff --git a/gcc/config/aarch64/aarch64-sve-builtins.cc
> b/gcc/config/aarch64/aarch64-sve-builtins.cc
> index e73aa9ad8a9..6589438855a 100644
> --- a/gcc/config/aarch64/aarch64-sve-builtins.cc
> +++ b/gcc/config/aarch64/aarch64-sve-builtins.cc
> @@ -2580,7 +2580,7 @@ gimple_folder::fold_contiguous_base (gimple_seq &stmts,
> tree vectype)
> tree
> gimple_folder::load_store_cookie (tree type)
> {
> - return build_int_cst (build_pointer_type (type), TYPE_ALIGN_UNIT (type));
> + return build_int_cst (build_pointer_type (type), TYPE_ALIGN (type));
> }
>
> /* Fold the call to a call to INSTANCE, with the same arguments. */
> diff --git a/gcc/gimple-fold.c b/gcc/gimple-fold.c
> index 3148c6b84d9..1f3b7d0818d 100644
> --- a/gcc/gimple-fold.c
> +++ b/gcc/gimple-fold.c
> @@ -5201,7 +5201,7 @@ gimple_fold_mask_load_store_mem_ref (gcall *call, tree
> vectype)
> if (!tree_fits_uhwi_p (alias_align) || !integer_all_onesp (mask))
> return NULL_TREE;
>
> - unsigned HOST_WIDE_INT align = tree_to_uhwi (alias_align) * BITS_PER_UNIT;
> + unsigned HOST_WIDE_INT align = tree_to_uhwi (alias_align);
> if (TYPE_ALIGN (vectype) != align)
> vectype = build_aligned_type (vectype, align);
> tree offset = build_zero_cst (TREE_TYPE (alias_align));
> diff --git a/gcc/testsuite/g++.dg/vect/pr95401.cc
> b/gcc/testsuite/g++.dg/vect/pr95401.cc
> new file mode 100644
> index 00000000000..6a56dab0957
> --- /dev/null
> +++ b/gcc/testsuite/g++.dg/vect/pr95401.cc
> @@ -0,0 +1,13 @@
> +// { dg-additional-options "-mavx2 -O3" { target avx2_runtime } }
> +// { dg-additional-sources pr95401a.cc }
> +
> +extern int var_9;
> +extern unsigned var_14;
> +extern int arr_16[];
> +#include <algorithm>
> +void test() {
> + for (short a = 0; a < (short)var_9; a += 12140)
> + for (short b = 0; b < 8; b++)
> + if (std::max(var_14, 1U))
> + arr_16[a + b] = 0;
> +}
> diff --git a/gcc/testsuite/g++.dg/vect/pr95401a.cc
> b/gcc/testsuite/g++.dg/vect/pr95401a.cc
> new file mode 100644
> index 00000000000..71b054c7621
> --- /dev/null
> +++ b/gcc/testsuite/g++.dg/vect/pr95401a.cc
> @@ -0,0 +1,13 @@
> +// { dg-do compile }
> +
> +#include "../../gcc.dg/vect/tree-vect.h"
> +
> +int var_9 = 1693986256, var_14;
> +int arr_16[11];
> +void test();
> +int main()
> +{
> + check_vect();
> + test();
> + return 0;
> +}
> diff --git a/gcc/tree-vect-stmts.c b/gcc/tree-vect-stmts.c
> index 11737a38a56..90822aba0ee 100644
> --- a/gcc/tree-vect-stmts.c
> +++ b/gcc/tree-vect-stmts.c
> @@ -7964,7 +7964,7 @@ vectorizable_store (vec_info *vinfo,
> /* Emit:
> MASK_STORE_LANES (DATAREF_PTR, ALIAS_PTR, VEC_MASK,
> VEC_ARRAY). */
> - unsigned int align = TYPE_ALIGN_UNIT (TREE_TYPE (vectype));
> + unsigned int align = TYPE_ALIGN (TREE_TYPE (vectype));
> tree alias_ptr = build_int_cst (ref_type, align);
> call = gimple_build_call_internal (IFN_MASK_STORE_LANES, 4,
> dataref_ptr, alias_ptr,
> @@ -8079,7 +8079,7 @@ vectorizable_store (vec_info *vinfo,
> if (final_mask)
> {
> align = least_bit_hwi (misalign | align);
> - tree ptr = build_int_cst (ref_type, align);
> + tree ptr = build_int_cst (ref_type, align * BITS_PER_UNIT);
> gcall *call
> = gimple_build_call_internal (IFN_MASK_STORE, 4,
> dataref_ptr, ptr,
> @@ -8094,7 +8094,7 @@ vectorizable_store (vec_info *vinfo,
> = vect_get_loop_len (loop_vinfo, loop_lens,
> vec_num * ncopies, vec_num * j + i);
> align = least_bit_hwi (misalign | align);
> - tree ptr = build_int_cst (ref_type, align);
> + tree ptr = build_int_cst (ref_type, align * BITS_PER_UNIT);
> machine_mode vmode = TYPE_MODE (vectype);
> opt_machine_mode new_ovmode
> = get_len_load_store_mode (vmode, false);
> @@ -9235,7 +9235,7 @@ vectorizable_load (vec_info *vinfo,
> /* Emit:
> VEC_ARRAY = MASK_LOAD_LANES (DATAREF_PTR, ALIAS_PTR,
> VEC_MASK). */
> - unsigned int align = TYPE_ALIGN_UNIT (TREE_TYPE (vectype));
> + unsigned int align = TYPE_ALIGN (TREE_TYPE (vectype));
> tree alias_ptr = build_int_cst (ref_type, align);
> call = gimple_build_call_internal (IFN_MASK_LOAD_LANES, 3,
> dataref_ptr, alias_ptr,
> @@ -9336,7 +9336,8 @@ vectorizable_load (vec_info *vinfo,
> if (final_mask)
> {
> align = least_bit_hwi (misalign | align);
> - tree ptr = build_int_cst (ref_type, align);
> + tree ptr = build_int_cst (ref_type,
> + align * BITS_PER_UNIT);
> gcall *call
> = gimple_build_call_internal (IFN_MASK_LOAD, 3,
> dataref_ptr, ptr,
> @@ -9352,7 +9353,8 @@ vectorizable_load (vec_info *vinfo,
> vec_num * ncopies,
> vec_num * j + i);
> align = least_bit_hwi (misalign | align);
> - tree ptr = build_int_cst (ref_type, align);
> + tree ptr = build_int_cst (ref_type,
> + align * BITS_PER_UNIT);
> gcall *call
> = gimple_build_call_internal (IFN_LEN_LOAD, 3,
> dataref_ptr, ptr,
