[FFmpeg-devel] [PATCH] x86/cpu: set avxslow cpuflag on bdver4 and btver2 CPUs
Hendrik Leppkes
h.leppkes at gmail.com
Sun Feb 7 17:38:02 CET 2016
On Sun, Feb 7, 2016 at 4:05 AM, James Almer <jamrial at gmail.com> wrote:
> They are also slow when using 256 bit wide registers
>
> Signed-off-by: James Almer <jamrial at gmail.com>
> ---
> See https://gcc.gnu.org/git/?p=gcc.git;a=commitdiff;h=3a33f4ce8ea1efdebec7f7138d00e9be9a12d630
> libavutil/x86/cpu.c | 10 ++++------
> 1 file changed, 4 insertions(+), 6 deletions(-)
>
> diff --git a/libavutil/x86/cpu.c b/libavutil/x86/cpu.c
> index f57d72d..bb63daa 100644
> --- a/libavutil/x86/cpu.c
> +++ b/libavutil/x86/cpu.c
> @@ -182,13 +182,11 @@ int ff_get_cpu_flags_x86(void)
>
> /* Similar to the above but for AVX functions on AMD processors.
> This is necessary only for functions using YMM registers on Bulldozer
> - based CPUs as they lack 256-bits execution units. SSE/AVX functions
> - using XMM registers are always faster on them.
> + and Jaguar based CPUs as they lack 256-bits execution units. SSE/AVX
> + functions using XMM registers are always faster on them.
> AV_CPU_FLAG_AVX and AV_CPU_FLAG_AVXSLOW are both set so that AVX is
> - used unless explicitly disabled by checking AV_CPU_FLAG_AVXSLOW.
> - TODO: Confirm if Excavator is affected or not by this once it's
> - released, and update the check if necessary. Same for btver2. */
> - if (family == 0x15 && (rval & AV_CPU_FLAG_AVX))
> + used unless explicitly disabled by checking AV_CPU_FLAG_AVXSLOW. */
> + if ((family == 0x15 || family == 0x16) && (rval & AV_CPU_FLAG_AVX))
> rval |= AV_CPU_FLAG_AVXSLOW;
> }
>
LGTM.
More information about the ffmpeg-devel
mailing list