Re: [PATCH v3 3/9] Introduce can_vector_compare_p function

[Richard Sandiford <richard.sandiford@arm.com>]

Ilya Leoshkevich <iii@linux.ibm.com> writes:
> @@ -329,6 +332,34 @@ expand_vec_cmp_expr_p (tree value_type, tree mask_type, enum tree_code code)
>    return false;
>  }
>  
> +/* Return TRUE iff vcond_optab/vcondu_optab support the given tree
> +   comparison.  */

Nit: better to use "true" rather than "TRUE" in modern comments;
TRUE is a hold-over from C90 days.

> +
> +static bool
> +vcond_icode_p (tree value_type, tree cmp_op_type, enum tree_code code)
> +{
> +  enum rtx_code rcode = get_rtx_code_safe (code, TYPE_UNSIGNED (cmp_op_type));
> +  if (rcode == LAST_AND_UNUSED_RTX_CODE)
> +    /* This may happen, for example, if code == SSA_NAME, in which case we
> +       cannot be certain whether a vector insn is available.  */
> +    return false;

IMO expand_vec_cond_expr_p should exit early for SSA names before
calling this function (but after checking get_vcond_mask_icode).
Asserting for a comparison seems reasonable here.

> diff --git a/gcc/optabs.c b/gcc/optabs.c
> index cdd07f3e53f..479789570a9 100644
> --- a/gcc/optabs.c
> +++ b/gcc/optabs.c
> @@ -3819,6 +3819,30 @@ can_compare_p (enum rtx_code code, machine_mode mode,
>    return 0;
>  }
>  
> +/* Return whether back-end can emit a vector comparison insn(s) using a give> +   CODE, with operands with CMP_OP_MODE, producing a result with VALUE_MODE,
> +   in order to achieve a PURPOSE.  */

Maybe:

/* Return whether the backend can emit a vector comparison for code CODE,
   comparing operands of mode CMP_OP_MODE and producing a result
   with VALUE_MODE.  */

> +
> +bool
> +can_vector_compare_p (enum rtx_code code, machine_mode value_mode,
> +		      machine_mode cmp_op_mode,
> +		      enum can_vector_compare_purpose purpose)

I think we should leave out the purpose argument until it's needed.
(Sorry if someone else said the opposite, can't remember now.)

Maybe can_vcond_compare_p would be more consistent with the other
function names in the patch.

Thanks,
Richard