vect: Verify that GET_MODE_NUNITS is power-of-2
Commit Message
While working on autovectorizing for the RISCV port I encountered an issue
where can_duplicate_and_interleave_p assumes that GET_MODE_NUNITS is a
power of two. The RISC-V target has vector modes (e.g. VNx1DImode) that
are not a power of two.
Tested on RISCV and x86_64-linux-gnu. Okay?
2023-03-09 Michael Collison <collison@rivosinc.com>
* poly-int.h (exact_div_p): New function to
verify that argument is a power of 2 poly_int.
* tree-vect-slp.cc (can_duplicate_and_interleave_p):
Check that GET_MODE_NUNITS is a power of 2.
---
gcc/poly-int.h | 17 +++++++++++++++++
gcc/tree-vect-slp.cc | 3 ++-
2 files changed, 19 insertions(+), 1 deletion(-)
Comments
On Fri, Mar 10, 2023 at 9:16 PM Michael Collison <collison@rivosinc.com> wrote:
>
> While working on autovectorizing for the RISCV port I encountered an issue
> where can_duplicate_and_interleave_p assumes that GET_MODE_NUNITS is a
> power of two. The RISC-V target has vector modes (e.g. VNx1DImode) that
> are not a power of two.
We do not support vector types that do not have a power-of-two
element count, see TYPE_VECTOR_SUBPARTS.
Also your test below verifies that nunits is divisible by two, not that it
is power-of-two? So maybe what you want to know is whether
known_gt (nunits, 1)?
> Tested on RISCV and x86_64-linux-gnu. Okay?
>
> 2023-03-09 Michael Collison <collison@rivosinc.com>
>
> * poly-int.h (exact_div_p): New function to
> verify that argument is a power of 2 poly_int.
> * tree-vect-slp.cc (can_duplicate_and_interleave_p):
> Check that GET_MODE_NUNITS is a power of 2.
> ---
> gcc/poly-int.h | 17 +++++++++++++++++
> gcc/tree-vect-slp.cc | 3 ++-
> 2 files changed, 19 insertions(+), 1 deletion(-)
>
> diff --git a/gcc/poly-int.h b/gcc/poly-int.h
> index 12571455081..d09632f341f 100644
> --- a/gcc/poly-int.h
> +++ b/gcc/poly-int.h
> @@ -2219,6 +2219,23 @@ multiple_p (const poly_int_pod<N, Ca> &a, const poly_int_pod<N, Cb> &b,
> return constant_multiple_p (a, b, multiple);
> }
>
> +/* Return true, if A is known to be a multiple of B. */
> +
> +template<unsigned int N, typename Ca, typename Cb>
> +inline bool
> +exact_div_p (const poly_int_pod<N, Ca> &a, Cb b)
> +{
> + typedef POLY_CONST_COEFF (Ca, Cb) C;
> + poly_int<N, C> r;
> + for (unsigned int i = 0; i < N; i++)
> + {
> + if ((a.coeffs[i] % b) != 0)
> + return false;
> +
> + }
> + return true;
> +}
> +
> /* Return A / B, given that A is known to be a multiple of B. */
>
> template<unsigned int N, typename Ca, typename Cb>
> diff --git a/gcc/tree-vect-slp.cc b/gcc/tree-vect-slp.cc
> index 9a4e000925e..6be2036a13a 100644
> --- a/gcc/tree-vect-slp.cc
> +++ b/gcc/tree-vect-slp.cc
> @@ -426,7 +426,8 @@ can_duplicate_and_interleave_p (vec_info *vinfo, unsigned int count,
> if (vector_type
> && VECTOR_MODE_P (TYPE_MODE (vector_type))
> && known_eq (GET_MODE_SIZE (TYPE_MODE (vector_type)),
> - GET_MODE_SIZE (base_vector_mode)))
> + GET_MODE_SIZE (base_vector_mode))
> + && exact_div_p (GET_MODE_NUNITS (TYPE_MODE (vector_type)), 2))
> {
> /* Try fusing consecutive sequences of COUNT / NVECTORS elements
> together into elements of type INT_TYPE and using the result
> --
> 2.34.1
>
@@ -2219,6 +2219,23 @@ multiple_p (const poly_int_pod<N, Ca> &a, const poly_int_pod<N, Cb> &b,
return constant_multiple_p (a, b, multiple);
}
+/* Return true, if A is known to be a multiple of B. */
+
+template<unsigned int N, typename Ca, typename Cb>
+inline bool
+exact_div_p (const poly_int_pod<N, Ca> &a, Cb b)
+{
+ typedef POLY_CONST_COEFF (Ca, Cb) C;
+ poly_int<N, C> r;
+ for (unsigned int i = 0; i < N; i++)
+ {
+ if ((a.coeffs[i] % b) != 0)
+ return false;
+
+ }
+ return true;
+}
+
/* Return A / B, given that A is known to be a multiple of B. */
template<unsigned int N, typename Ca, typename Cb>
@@ -426,7 +426,8 @@ can_duplicate_and_interleave_p (vec_info *vinfo, unsigned int count,
if (vector_type
&& VECTOR_MODE_P (TYPE_MODE (vector_type))
&& known_eq (GET_MODE_SIZE (TYPE_MODE (vector_type)),
- GET_MODE_SIZE (base_vector_mode)))
+ GET_MODE_SIZE (base_vector_mode))
+ && exact_div_p (GET_MODE_NUNITS (TYPE_MODE (vector_type)), 2))
{
/* Try fusing consecutive sequences of COUNT / NVECTORS elements
together into elements of type INT_TYPE and using the result