Re: [FFmpeg-devel] [PATCH] libavfilter: zscale performance optimization >4x

From: Victoria Zhislina <niva213@gmail.com>
To: FFmpeg development discussions and patches
	<ffmpeg-devel@ffmpeg.org>,
	 Victoria Zhislina <Victoria.Zhislina-at-intel.com@ffmpeg.org>
Subject: Re: [FFmpeg-devel] [PATCH] libavfilter: zscale performance optimization >4x
Date: Mon, 21 Feb 2022 18:22:02 +0300
Message-ID: <CAAPJamDWdc839c+HkvDfYRoF95ESj96-x2NzDbWBq=XW=Qx-3Q@mail.gmail.com> (raw)
In-Reply-To: <164544253625.19727.14104805163737761889@lain.red.khirnov.net>

Hi, Anton. Thanks for your input. But the patch does the single thing
described in the commit message - improves performance >4 :)  Sorry.
This patch creation is based on real experience and on real measurement.
 Please notice  I don't write ">40x" so it is not an advertisement :).
 Right you are, under the hood it does 2 main things and one small
additional one combined to achieve the performance gain mentioned in the
commit message :)
However the changes are extremely local - they cover just a couple of
functions in a single file and  It doesn't make sense to split them. It
seems to me split will make ffmpeg-devel mail list and ffmpeg  git log
dirtier, not cleaner.
So let's wait for the Paul B Mahol opinion - it is his code that I've
modified.

On Mon, Feb 21, 2022 at 2:22 PM Anton Khirnov <anton@khirnov.net> wrote:

> > libavfilter: zscale performance optimization >4x
>
> This reads like an advertisement rather than a useful description. It
> should say what the patch does, performance improvement numbers should
> be mentioned in the commit message body.
>
> Quoting Victoria Zhislina (2022-02-21 09:20:55)
> > By ffmpeg threading support implementation via frame slicing and doing
> > zimg_filter_graph_build that used to take 30-60% of each frame processig
> > only if necessary (some parameters changed)
> > the performance increase vs original version
> > in video downscale and color conversion  >4x is seen
> > on 64 cores Intel Xeon, 3x on i7-6700K (4 cores with HT)
>
> This implies the patch does multiple unrelated things. Then it should be
> split in multiple patches, unless some important factor prevents that
> (then that factor should be described in the commit message).
>
> --
> Anton Khirnov
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel@ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
>
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".