On Thu, 2 Apr 2026, Tamar Christina wrote:
> param_vect_scalar_cost_multiplier is support to make all vector loops
> profitable
> and does so by scaling up the cost of a scalar iteration. However the new
> check
> for uncounted loop profitability doesn't trigger it when we couldn't estimate
> the scalar trip count.
Why's that? min_profitable_estimate should be adjusted by the param, no?
At least that was the whole idea about the check you patch.
Richard.
> This adjust the condition to check if the parameter is set to <= 100% scaling,
> which means we want to make vector code more profitable and so bypass the
> check.
>
> I've also considered that maybe the check should be something like <= 500% to
> allow someone to use param_vect_scalar_cost_multiplier without immediately
> enabling uncounted loops forcefully, but with a high multiplier enable it.
>
> Most backends use a 100x multiplier so using <= 500% would work too if we
> think
> it's useful.
>
> Bootstrapped Regtested on aarch64-none-linux-gnu,
> arm-none-linux-gnueabihf, x86_64-pc-linux-gnu
> -m32, -m64 and no issues.
>
> Any objections?
>
> Thanks,
> Tamar
>
> gcc/ChangeLog:
>
> * tree-vect-loop.cc (vect_analyze_loop_costing): Check scalar
> cost multiplier.
>
> gcc/testsuite/ChangeLog:
>
> * gcc.target/aarch64/sve/cost_model_19.c: New test.
>
> ---
> diff --git a/gcc/testsuite/gcc.target/aarch64/sve/cost_model_19.c
> b/gcc/testsuite/gcc.target/aarch64/sve/cost_model_19.c
> new file mode 100644
> index
> 0000000000000000000000000000000000000000..9e4421d630f2b04833100f83dec378428b3d170f
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/sve/cost_model_19.c
> @@ -0,0 +1,12 @@
> +/* { dg-do compile } */
> +/* { dg-options "-Ofast -march=armv8-a+sve -mmax-vectorization
> -ffreestanding -fdump-tree-vect-details" } */
> +
> +unsigned long f(const char *s)
> +{
> + unsigned long i = 0;
> + while (*s++)
> + i++;
> + return i;
> +}
> +
> +/* { dg-final { scan-tree-dump "vectorized 1 loops in function" "vect" } } */
> diff --git a/gcc/tree-vect-loop.cc b/gcc/tree-vect-loop.cc
> index
> 7121edb8d81904d227ba510f20a776fa7c9c8224..37fbbd1dc94e61d52b951eb4dd246776beeaad86
> 100644
> --- a/gcc/tree-vect-loop.cc
> +++ b/gcc/tree-vect-loop.cc
> @@ -1964,7 +1964,8 @@ vect_analyze_loop_costing (loop_vec_info loop_vinfo,
> likely alignment prologue). */
> if (LOOP_VINFO_NITERS_UNCOUNTED_P (loop_vinfo)
> && estimated_niter == -1
> - && min_profitable_estimate > (int) vect_vf_for_cost (loop_vinfo))
> + && min_profitable_estimate > (int) vect_vf_for_cost (loop_vinfo)
> + && param_vect_scalar_cost_multiplier <= 100)
> {
> if (dump_enabled_p ())
> dump_printf_loc (MSG_NOTE, vect_location,
>
>
>
--
Richard Biener <[email protected]>
SUSE Software Solutions Germany GmbH, Frankenstrasse 146, 90461 Nuernberg,
Germany; GF: Ivo Totev, Andrew Myers, Andrew McDonald, Boudien Moerman;
HRB 36809 (AG Nuernberg)
