Ping. Ok for mainline?
On Thu, Apr 25, 2024 at 09:26:45AM +0200, Stefan Schulze Frielinghaus wrote:
> Bitcount operations popcount, clz, and ctz are emulated for narrow modes
> in case an operation is only supported for wider modes. Beside that ctz
> may be emulated via clz in expand_ctz. Reflect this in
> expression_expensive_p.
>
> I considered the emulation of ctz via clz as not expensive since this
> basically reduces to ctz (x) = c - (clz (x & ~x)) where c is the mode
> precision minus 1 which should be faster than a loop.
>
> Bootstrapped and regtested on x86_64 and s390. Though, this is probably
> stage1 material?
>
> gcc/ChangeLog:
>
> PR tree-optimization/110490
> * tree-scalar-evolution.cc (expression_expensive_p): Also
> consider mode widening for popcount, clz, and ctz.
> ---
> gcc/tree-scalar-evolution.cc | 23 +++++++++++++++++++++++
> 1 file changed, 23 insertions(+)
>
> diff --git a/gcc/tree-scalar-evolution.cc b/gcc/tree-scalar-evolution.cc
> index b0a5e09a77c..622c7246c1b 100644
> --- a/gcc/tree-scalar-evolution.cc
> +++ b/gcc/tree-scalar-evolution.cc
> @@ -3458,6 +3458,28 @@ bitcount_call:
> && (optab_handler (optab, word_mode)
> != CODE_FOR_nothing))
> break;
> + /* If popcount is available for a wider mode, we emulate the
> + operation for a narrow mode by first zero-extending the value
> + and then computing popcount in the wider mode. Analogue for
> + ctz. For clz we do the same except that we additionally have
> + to subtract the difference of the mode precisions from the
> + result. */
> + if (is_a <scalar_int_mode> (mode, &int_mode))
> + {
> + machine_mode wider_mode_iter;
> + FOR_EACH_WIDER_MODE (wider_mode_iter, mode)
> + if (optab_handler (optab, wider_mode_iter)
> + != CODE_FOR_nothing)
> + goto check_call_args;
> + /* Operation ctz may be emulated via clz in expand_ctz. */
> + if (optab == ctz_optab)
> + {
> + FOR_EACH_WIDER_MODE_FROM (wider_mode_iter, mode)
> + if (optab_handler (clz_optab, wider_mode_iter)
> + != CODE_FOR_nothing)
> + goto check_call_args;
> + }
> + }
> return true;
> }
> break;
> @@ -3469,6 +3491,7 @@ bitcount_call:
> break;
> }
>
> +check_call_args:
> FOR_EACH_CALL_EXPR_ARG (arg, iter, expr)
> if (expression_expensive_p (arg, cond_overflow_p, cache, op_cost))
> return true;
> --
> 2.44.0
>
