Re: [PATCH 13/19] <sys/platform/x86.h>: Add AMX-FP16 support

[Noah Goldstein <goldstein.w.n@gmail.com>]

On Wed, Apr 5, 2023 at 11:28 AM H.J. Lu via Libc-alpha
<libc-alpha@sourceware.org> wrote:
>
> Add AMX-FP16 support to <sys/platform/x86.h>.
> ---
>  manual/platform.texi               | 3 +++
>  sysdeps/x86/bits/platform/x86.h    | 1 +
>  sysdeps/x86/cpu-features.c         | 2 ++
>  sysdeps/x86/include/cpu-features.h | 3 +++
>  sysdeps/x86/tst-get-cpu-features.c | 2 ++
>  5 files changed, 11 insertions(+)
>
> diff --git a/manual/platform.texi b/manual/platform.texi
> index af79f5eb4d..7d4aa3d339 100644
> --- a/manual/platform.texi
> +++ b/manual/platform.texi
> @@ -200,6 +200,9 @@ The supported processor features are:
>  @item
>  @code{AMX_INT8} -- Tile computational operations on 8-bit numbers.
>
> +@item
> +@code{AMX_FP16} -- Tile computational operations on FP16 numbers.
> +
>  @item
>  @code{AMX_TILE} -- Tile architecture.
>
> diff --git a/sysdeps/x86/bits/platform/x86.h b/sysdeps/x86/bits/platform/x86.h
> index 2a15ad937a..2776c69b16 100644
> --- a/sysdeps/x86/bits/platform/x86.h
> +++ b/sysdeps/x86/bits/platform/x86.h
> @@ -298,6 +298,7 @@ enum
>    x86_cpu_FSRS                 = x86_cpu_index_7_ecx_1_eax + 11,
>    x86_cpu_FSRCS                        = x86_cpu_index_7_ecx_1_eax + 12,
>    x86_cpu_WRMSRNS              = x86_cpu_index_7_ecx_1_eax + 19,
> +  x86_cpu_AMX_FP16             = x86_cpu_index_7_ecx_1_eax + 21,
>    x86_cpu_HRESET               = x86_cpu_index_7_ecx_1_eax + 22,
>    x86_cpu_LAM                  = x86_cpu_index_7_ecx_1_eax + 26,
>
> diff --git a/sysdeps/x86/cpu-features.c b/sysdeps/x86/cpu-features.c
> index da04ad0b00..6c1b5efc5f 100644
> --- a/sysdeps/x86/cpu-features.c
> +++ b/sysdeps/x86/cpu-features.c
> @@ -213,6 +213,8 @@ update_active (struct cpu_features *cpu_features)
>           CPU_FEATURE_SET_ACTIVE (cpu_features, AMX_TILE);
>           /* Determine if AMX_INT8 is usable.  */
>           CPU_FEATURE_SET_ACTIVE (cpu_features, AMX_INT8);
> +         /* Determine if AMX_FP16 is usable.  */
> +         CPU_FEATURE_SET_ACTIVE (cpu_features, AMX_FP16);
>         }
>
>        /* These features are usable only when OSXSAVE is enabled.  */
> diff --git a/sysdeps/x86/include/cpu-features.h b/sysdeps/x86/include/cpu-features.h
> index 4e40fe0482..07c841c1d4 100644
> --- a/sysdeps/x86/include/cpu-features.h
> +++ b/sysdeps/x86/include/cpu-features.h
> @@ -309,6 +309,7 @@ enum
>  #define bit_cpu_FZLRM          (1u << 10)
>  #define bit_cpu_FSRS           (1u << 11)
>  #define bit_cpu_FSRCS          (1u << 12)
> +#define bit_cpu_AMX_FP16       (1u << 21)
>  #define bit_cpu_HRESET         (1u << 22)
>  #define bit_cpu_LAM            (1u << 26)
>
> @@ -546,6 +547,7 @@ enum
>  #define index_cpu_FZLRM                CPUID_INDEX_7_ECX_1
>  #define index_cpu_FSRS         CPUID_INDEX_7_ECX_1
>  #define index_cpu_FSRCS                CPUID_INDEX_7_ECX_1
> +#define index_cpu_AMX_FP16     CPUID_INDEX_7_ECX_1
>  #define index_cpu_HRESET       CPUID_INDEX_7_ECX_1
>  #define index_cpu_LAM          CPUID_INDEX_7_ECX_1
>
> @@ -783,6 +785,7 @@ enum
>  #define reg_FZLRM              eax
>  #define reg_FSRS               eax
>  #define reg_FSRCS              eax
> +#define reg_AMX_FP16           eax
>  #define reg_HRESET             eax
>  #define reg_LAM                        eax
>
> diff --git a/sysdeps/x86/tst-get-cpu-features.c b/sysdeps/x86/tst-get-cpu-features.c
> index 9c436eaa64..c0f222cb77 100644
> --- a/sysdeps/x86/tst-get-cpu-features.c
> +++ b/sysdeps/x86/tst-get-cpu-features.c
> @@ -210,6 +210,7 @@ do_test (void)
>    CHECK_CPU_FEATURE_PRESENT (FSRS);
>    CHECK_CPU_FEATURE_PRESENT (FSRCS);
>    CHECK_CPU_FEATURE_PRESENT (WRMSRNS);
> +  CHECK_CPU_FEATURE_PRESENT (AMX_FP16);
>    CHECK_CPU_FEATURE_PRESENT (HRESET);
>    CHECK_CPU_FEATURE_PRESENT (LAM);
>    CHECK_CPU_FEATURE_PRESENT (AESKLE);
> @@ -377,6 +378,7 @@ do_test (void)
>    CHECK_CPU_FEATURE_ACTIVE (FZLRM);
>    CHECK_CPU_FEATURE_ACTIVE (FSRS);
>    CHECK_CPU_FEATURE_ACTIVE (FSRCS);
> +  CHECK_CPU_FEATURE_ACTIVE (AMX_FP16);
>    CHECK_CPU_FEATURE_ACTIVE (AESKLE);
>    CHECK_CPU_FEATURE_ACTIVE (WIDE_KL);
>    CHECK_CPU_FEATURE_ACTIVE (PTWRITE);
> --
> 2.39.2
>

LGTM
Reviewed-by: Noah Goldstein <goldstein.w.n@gmail.com>