Alejandro Martinez Vicente <alejandro.martinezvice...@arm.com> writes:
> Hi,
>
> This patch fixes bug 90681. It was caused by trying to SLP vectorize a non
> grouped load. We've fixed it by tweaking a bit the implementation: mark
> masked loads as not vectorizable, but support them as an special case. Then
> the detect them in the test for normal non-grouped loads that was already
> there.
>
> The bug reproducer now works and the performance test we added is still happy.
>
> Alejandro
>
> gcc/ChangeLog:
>
> 2019-05-31 Alejandro Martinez <alejandro.martinezvice...@arm.com>
>
> PR tree-optimization/90681
> * internal-fn.c (mask_load_direct): Mark as non-vectorizable again.
> * tree-vect-slp.c (vect_build_slp_tree_1): Add masked loads as a
> special case for SLP, but fail on non-grouped loads.
>
>
> 2019-05-31 Alejandro Martinez <alejandro.martinezvice...@arm.com>
>
> gcc/testsuite/
>
> PR tree-optimization/90681
> * gfortran.dg/vect/pr90681.f: Bug reproducer.
OK, thanks.
Richard
> diff --git a/gcc/internal-fn.c b/gcc/internal-fn.c
> index 3051a7a..04081f3 100644
> --- a/gcc/internal-fn.c
> +++ b/gcc/internal-fn.c
> @@ -100,7 +100,7 @@ init_internal_fns ()
> /* Create static initializers for the information returned by
> direct_internal_fn. */
> #define not_direct { -2, -2, false }
> -#define mask_load_direct { -1, 2, true }
> +#define mask_load_direct { -1, 2, false }
> #define load_lanes_direct { -1, -1, false }
> #define mask_load_lanes_direct { -1, -1, false }
> #define gather_load_direct { -1, -1, false }
> diff --git a/gcc/testsuite/gfortran.dg/vect/pr90681.f
> b/gcc/testsuite/gfortran.dg/vect/pr90681.f
> new file mode 100644
> index 0000000..03d3987
> --- /dev/null
> +++ b/gcc/testsuite/gfortran.dg/vect/pr90681.f
> @@ -0,0 +1,13 @@
> +C { dg-do compile }
> +C { dg-additional-options "-march=armv8.2-a+sve" { target { aarch64*-*-* } }
> }
> + SUBROUTINE HMU (H1)
> + COMMON DD(107)
> + DIMENSION H1(NORBS,*)
> + DO 70 J1 = IA,I1
> + H1(I1,J1) = 0
> + JO1 = J1
> + IF (JO1.EQ.1) THEN
> + H1(I1,J1) = DD(NI)
> + END IF
> + 70 CONTINUE
> + END
> diff --git a/gcc/tree-vect-slp.c b/gcc/tree-vect-slp.c
> index 884db33..23a8a20 100644
> --- a/gcc/tree-vect-slp.c
> +++ b/gcc/tree-vect-slp.c
> @@ -661,6 +661,7 @@ vect_build_slp_tree_1 (unsigned char *swap,
> machine_mode optab_op2_mode;
> machine_mode vec_mode;
> stmt_vec_info first_load = NULL, prev_first_load = NULL;
> + bool load_p = false;
>
> /* For every stmt in NODE find its def stmt/s. */
> stmt_vec_info stmt_info;
> @@ -714,7 +715,10 @@ vect_build_slp_tree_1 (unsigned char *swap,
> if (gcall *call_stmt = dyn_cast <gcall *> (stmt))
> {
> rhs_code = CALL_EXPR;
> - if ((gimple_call_internal_p (call_stmt)
> +
> + if (gimple_call_internal_p (stmt, IFN_MASK_LOAD))
> + load_p = true;
> + else if ((gimple_call_internal_p (call_stmt)
> && (!vectorizable_internal_fn_p
> (gimple_call_internal_fn (call_stmt))))
> || gimple_call_tail_p (call_stmt)
> @@ -732,7 +736,10 @@ vect_build_slp_tree_1 (unsigned char *swap,
> }
> }
> else
> - rhs_code = gimple_assign_rhs_code (stmt);
> + {
> + rhs_code = gimple_assign_rhs_code (stmt);
> + load_p = TREE_CODE_CLASS (rhs_code) == tcc_reference;
> + }
>
> /* Check the operation. */
> if (i == 0)
> @@ -899,7 +906,7 @@ vect_build_slp_tree_1 (unsigned char *swap,
> } /* Grouped access. */
> else
> {
> - if (TREE_CODE_CLASS (rhs_code) == tcc_reference)
> + if (load_p)
> {
> /* Not grouped load. */
> if (dump_enabled_p ())
