On Tue, Apr 21, 2020 at 2:13 PM Richard Sandiford
<richard.sandif...@arm.com> wrote:
>
> The type compatibility handling in simplify_vector_constructor is
> based on the number of elements and on element type compatibility,
> but that's no longer enough to ensure that two vector types are
> compatible. This patch uses a VIEW_CONVERT_EXPR if the permutation
> type and result type are distinct.
>
> Tested on aarch64-linux-gnu and x86_64-linux-gnu. OK to install?
OK.
> Richard
>
>
> 2020-04-21 Richard Sandiford <richard.sandif...@arm.com>
>
> gcc/
> PR tree-optimization/94683
> * tree-ssa-forwprop.c (simplify_vector_constructor): Use a
> VIEW_CONVERT_EXPR to handle mixtures of similarly-structured
> but distinct vector types.
>
> gcc/testsuite/
> PR tree-optimization/94683
> * gcc.target/aarch64/sve/acle/general/pr94683.c: New test.
> ---
> .../aarch64/sve/acle/general/pr94683.c | 29 +++++++++++++++++++
> gcc/tree-ssa-forwprop.c | 5 ++++
> 2 files changed, 34 insertions(+)
> create mode 100644
> gcc/testsuite/gcc.target/aarch64/sve/acle/general/pr94683.c
>
> diff --git a/gcc/tree-ssa-forwprop.c b/gcc/tree-ssa-forwprop.c
> index 3d8acf7eb03..1a50045b367 100644
> --- a/gcc/tree-ssa-forwprop.c
> +++ b/gcc/tree-ssa-forwprop.c
> @@ -2598,6 +2598,11 @@ simplify_vector_constructor (gimple_stmt_iterator *gsi)
> res, TYPE_SIZE (type), bitsize_zero_node);
> if (conv_code != ERROR_MARK)
> res = gimple_build (&stmts, conv_code, type, res);
> + else if (!useless_type_conversion_p (type, TREE_TYPE (res)))
> + {
> + gcc_assert (!targetm.compatible_vector_types_p (type, perm_type));
> + res = gimple_build (&stmts, VIEW_CONVERT_EXPR, type, res);
> + }
> /* Blend in the actual constant. */
> if (converted_orig1)
> res = gimple_build (&stmts, VEC_PERM_EXPR, type,
> diff --git a/gcc/testsuite/gcc.target/aarch64/sve/acle/general/pr94683.c
> b/gcc/testsuite/gcc.target/aarch64/sve/acle/general/pr94683.c
> new file mode 100644
> index 00000000000..fb7c0e479cf
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/sve/acle/general/pr94683.c
> @@ -0,0 +1,29 @@
> +/* { dg-options "-O2 -msve-vector-bits=256" } */
> +/* { dg-final { check-function-bodies "**" "" } } */
> +
> +#include <arm_sve.h>
> +
> +typedef float v8sf __attribute__((vector_size(32)));
> +
> +#ifdef __cplusplus
> +extern "C" {
> +#endif
> +
> +/*
> +** test:
> +** fadd z0\.s, p0/m, z0\.s, #1.0
> +** trn1 z0\.s, z0\.s, z0\.s
> +** fdiv z0\.s, p0/m, z0\.s, z1\.s
> +** ret
> +*/
> +svfloat32_t
> +test (svbool_t pg, svfloat32_t x, svfloat32_t y)
> +{
> + v8sf a = svadd_x (pg, x, 1);
> + v8sf b = { a[0], a[0], a[2], a[2], a[4], a[4], a[6], a[6] };
> + return svdiv_x (pg, b, y);
> +}
> +
> +#ifdef __cplusplus
> +}
> +#endif
