Re: [dpdk-dev] [PATCH v2] lib/librte_eal: support SVE flag on ARM64

[Jerin Jacob <jerinjacobk@gmail.com>]

On Mon, Aug 17, 2020 at 6:17 PM Wei Hu (Xavier)
<huwei013@chinasoftinc.com> wrote:
>
> From: "Wei Hu (Xavier)" <xavier.huwei@huawei.com>
>
> SVE is the next-generation SIMD extension of the ARMv8-A AArch64
> instruction set.
> The related marco definition can be found in linux kernel:
>   arch/arm64/include/uapi/asm/hwcap.h
>
> This patch supports getting cpu SVE feature on ARM64 platform.
>
> Signed-off-by: Chengwen Feng <fengchengwen@huawei.com>
> Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>

Change the git commit like

eal/arm64: update CPU flags




> ---
> v1 -> v2:
>         Adds more sve-related definition to rte_cpu_feature_table,
>         sunch as SVE2, etc.
> ---
>  lib/librte_eal/arm/include/rte_cpuflags_64.h |  1 +
>  lib/librte_eal/arm/rte_cpuflags.c            | 11 +++++++++++
>  2 files changed, 12 insertions(+)
>
> diff --git a/lib/librte_eal/arm/include/rte_cpuflags_64.h b/lib/librte_eal/arm/include/rte_cpuflags_64.h
> index 95cc01474..069844ddb 100644
> --- a/lib/librte_eal/arm/include/rte_cpuflags_64.h
> +++ b/lib/librte_eal/arm/include/rte_cpuflags_64.h
> @@ -22,6 +22,7 @@ enum rte_cpu_flag_t {
>         RTE_CPUFLAG_SHA2,
>         RTE_CPUFLAG_CRC32,
>         RTE_CPUFLAG_ATOMICS,
> +       RTE_CPUFLAG_SVE,

Please intrdouce the flag for all newly added items as well.

>         RTE_CPUFLAG_AARCH64,
>         /* The last item */
>         RTE_CPUFLAG_NUMFLAGS,/**< This should always be the last! */
> diff --git a/lib/librte_eal/arm/rte_cpuflags.c b/lib/librte_eal/arm/rte_cpuflags.c
> index caf3dc83a..97a9fcfd4 100644
> --- a/lib/librte_eal/arm/rte_cpuflags.c
> +++ b/lib/librte_eal/arm/rte_cpuflags.c
> @@ -95,6 +95,17 @@ const struct feature_entry rte_cpu_feature_table[] = {
>         FEAT_DEF(SHA2,          REG_HWCAP,    6)
>         FEAT_DEF(CRC32,         REG_HWCAP,    7)
>         FEAT_DEF(ATOMICS,       REG_HWCAP,    8)
> +       FEAT_DEF(SVE,           REG_HWCAP,    22)
> +       FEAT_DEF(SVE2,          REG_HWCAP2,   1)
> +       FEAT_DEF(SVEAES,        REG_HWCAP2,   2)
> +       FEAT_DEF(SVEPMULL,      REG_HWCAP2,   3)
> +       FEAT_DEF(SVEBITPERM,    REG_HWCAP2,   4)
> +       FEAT_DEF(SVESHA3,       REG_HWCAP2,   5)
> +       FEAT_DEF(SVESM4,        REG_HWCAP2,   6)

Following stuff is missing
HWCAP2_FLAGM2           (1 << 7)
HWCAP2_FRINT            (1 << 8)

> +       FEAT_DEF(SVEI8MM,       REG_HWCAP2,   9)
> +       FEAT_DEF(SVEF32MM,      REG_HWCAP2,   10)
> +       FEAT_DEF(SVEF64MM,      REG_HWCAP2,   11)
> +       FEAT_DEF(SVEBF16,       REG_HWCAP2,   12)
>         FEAT_DEF(AARCH64,       REG_PLATFORM, 1)
>  };
>  #endif /* RTE_ARCH */
> --
> 2.27.0
>