[PATCH v2 2.32] x86: Optimizing memcpy for AMD Zen architecture.
H.J. Lu
hjl.tools@gmail.com
Fri Oct 30 11:28:21 GMT 2020
On Fri, Oct 30, 2020 at 1:45 AM Florian Weimer <fweimer@redhat.com> wrote:
>
> From: Sajan Karumanchi <sajan.karumanchi@amd.com>
>
> Modifying the shareable cache '__x86_shared_cache_size', which is a
> factor in computing the non-temporal threshold parameter
> '__x86_shared_non_temporal_threshold' to optimize memcpy for AMD Zen
> architectures.
> In the existing implementation, the shareable cache is computed as 'L3
> per thread, L2 per core'. Recomputing this shareable cache as 'L3 per
> CCX(Core-Complex)' has brought in performance gains.
> As per the large bench variant results, this patch also addresses the
> regression problem on AMD Zen architectures.
>
> Backport of commit 59803e81f96b479c17f583b31eac44b57591a1bf upstream,
> with the fix from cb3a749a22a55645dc6a52659eea765300623f98 ("x86:
> Restore processing of cache size tunables in init_cacheinfo") applied.
>
> Reviewed-by: Premachandra Mallappa <premachandra.mallappa@amd.com>
> Co-Authored-by: Florian Weimer <fweimer@redhat.com>
>
> ---
> sysdeps/x86/cacheinfo.c | 20 +++++++++++++++++---
> 1 file changed, 17 insertions(+), 3 deletions(-)
>
> diff --git a/sysdeps/x86/cacheinfo.c b/sysdeps/x86/cacheinfo.c
> index dadec5d58f..3fb4a028d8 100644
> --- a/sysdeps/x86/cacheinfo.c
> +++ b/sysdeps/x86/cacheinfo.c
> @@ -808,7 +808,7 @@ init_cacheinfo (void)
> threads = 1 << ((ecx >> 12) & 0x0f);
> }
>
> - if (threads == 0)
> + if (threads == 0 || cpu_features->basic.family >= 0x17)
> {
> /* If APIC ID width is not available, use logical
> processor count. */
> @@ -823,8 +823,22 @@ init_cacheinfo (void)
> if (threads > 0)
> shared /= threads;
>
> - /* Account for exclusive L2 and L3 caches. */
> - shared += core;
> + /* Get shared cache per ccx for Zen architectures. */
> + if (cpu_features->basic.family >= 0x17)
> + {
> + unsigned int eax;
> +
> + /* Get number of threads share the L3 cache in CCX. */
> + __cpuid_count (0x8000001D, 0x3, eax, ebx, ecx, edx);
> +
> + unsigned int threads_per_ccx = ((eax >> 14) & 0xfff) + 1;
> + shared *= threads_per_ccx;
> + }
> + else
> + {
> + /* Account for exclusive L2 and L3 caches. */
> + shared += core;
> + }
> }
> }
>
>
LGTM.
Thanks.
--
H.J.
More information about the Libc-alpha
mailing list