[PATCH v1 1/1] x86: Lower non-temporal copy threshold for Hygon

[H.J. Lu]

On Sat, May 9, 2026 at 2:26 PM Jiamei Xie <xiejiamei@hygon.cn> wrote:
>
> From: xiejiamei <xiejiamei@hygon.cn>
>
> Benchmarks on Hygon processors show that the default non-temporal
> threshold is higher than ideal for large copy workloads. As a result,
> memcpy and memmove may continue to use the temporal copy path for
> longer than is beneficial, increasing cache pollution and reducing
> throughput for large copies.
>
> Lower the copy non-temporal threshold to 3/8 of the shared cache size
> per thread on Hygon. This allows the non-temporal copy path to be
> selected earlier while leaving the memset non-temporal threshold
> unchanged.
>
> Signed-off-by: xiejiamei <xiejiamei@hygon.cn>
> ---
>  sysdeps/x86/dl-cacheinfo.h | 7 +++++++
>  1 file changed, 7 insertions(+)
>
> diff --git a/sysdeps/x86/dl-cacheinfo.h b/sysdeps/x86/dl-cacheinfo.h
> index b6e17b0e32..84817a84fd 100644
> --- a/sysdeps/x86/dl-cacheinfo.h
> +++ b/sysdeps/x86/dl-cacheinfo.h
> @@ -1244,6 +1244,13 @@ dl_init_cacheinfo (struct cpu_features *cpu_features)
>    if (!CPU_FEATURES_ARCH_P (cpu_features, Avoid_Non_Temporal_Memset))
>      memset_non_temporal_threshold = non_temporal_threshold;
>
> +  /* Hygon benefits from entering the non-temporal copy path earlier.
> +     Use 3/8 of the shared cache size per thread to reduce cache
> +     pollution and improve throughput for large copies. Keep the memset
> +     non-temporal threshold unchanged.  */
> +  if (cpu_features->basic.kind == arch_kind_hygon)
> +    non_temporal_threshold = shared_per_thread * 3 / 8;

Please put this on the else branch when setting non_temporal_threshold tunable.

> +
>    tunable_size = TUNABLE_GET (x86_non_temporal_threshold, long int, NULL);
>    if (tunable_size > minimum_non_temporal_threshold
>        && tunable_size <= maximum_non_temporal_threshold)
> --
> 2.43.0
>
>


-- 
H.J.