[FFmpeg-devel] [PATCH 1/2] libavutil/cpu: Adds fast gather detection.

[James Almer]

On 7/16/2021 10:44 AM, Alan Kelly wrote:
> Broadwell and later and Zen3 and later have fast gather instructions.
> ---
>   Haswell is now excluded from EXTERNAL_AVX2_FAST as discussed in the
>   email thread.

I was very explicit about this not being ok. We're not disabling all ymm 
usage for Haswell just for one or two swscale functions using gathers.

Lets go with Lynne's latest suggestion and not change the flags at all 
and use gathers on Haswell, same as other arches, by looking at the 
AVX2_FAST flag.

>   libavutil/cpu.h     |  1 +
>   libavutil/x86/cpu.c | 11 ++++++++++-
>   2 files changed, 11 insertions(+), 1 deletion(-)
> 
> diff --git a/libavutil/cpu.h b/libavutil/cpu.h
> index c069076439..ec3073d021 100644
> --- a/libavutil/cpu.h
> +++ b/libavutil/cpu.h
> @@ -113,6 +113,7 @@ void av_force_cpu_count(int count);
>    *  av_set_cpu_flags_mask(), then this function will behave as if AVX is not
>    *  present.
>    */
> +
>   size_t av_cpu_max_align(void);
>   
>   #endif /* AVUTIL_CPU_H */
> diff --git a/libavutil/x86/cpu.c b/libavutil/x86/cpu.c
> index bcd41a50a2..158e2170c4 100644
> --- a/libavutil/x86/cpu.c
> +++ b/libavutil/x86/cpu.c
> @@ -146,8 +146,17 @@ int ff_get_cpu_flags_x86(void)
>       if (max_std_level >= 7) {
>           cpuid(7, eax, ebx, ecx, edx);
>   #if HAVE_AVX2
> -        if ((rval & AV_CPU_FLAG_AVX) && (ebx & 0x00000020))
> +        if ((rval & AV_CPU_FLAG_AVX) && (ebx & 0x00000020)){
>               rval |= AV_CPU_FLAG_AVX2;
> +
> +            cpuid(1, eax, ebx, ecx, std_caps);
> +            family = ((eax >> 8) & 0xf) + ((eax >> 20) & 0xff);
> +            model  = ((eax >> 4) & 0xf) + ((eax >> 12) & 0xf0);
> +            // Haswell and earlier has slow gather
> +            if(family == 6 && model < 70)
> +                rval |= AV_CPU_FLAG_AVXSLOW;
> +        }
> +
>   #if HAVE_AVX512 /* F, CD, BW, DQ, VL */
>           if ((xcr0_lo & 0xe0) == 0xe0) { /* OPMASK/ZMM state */
>               if ((rval & AV_CPU_FLAG_AVX2) && (ebx & 0xd0030000) == 0xd0030000)
>