Enable link time optimization for ffmpeg

[Bond-009]

No description provided.

[joshuaboniface]

What is the benefit/drawback/impact of this?

[nyanmisaka]

What is the benefit/drawback/impact of this?

https://johnysswlab.com/link-time-optimizations-new-way-to-do-compiler-optimizations/

Compilation takes double the time with LTO enabled for both compilers. Linking time is huge with LTO for both compilers, on GCC it’s 86 times slower, on CLANG 48 times slower.

This experiment has shown that, for ffmpeg at least, enabling LTO doesn’t bring expected speed ups.

[Bond-009]

What is the benefit/drawback/impact of this?

All my test are done on ffmpeg master commit 33db0cb
with gcc (GCC) 10.2.0 and clang version 11.1.0
on a Ryzen 7 2700 with 32GB of ram

	GCC	GCC LTO	Clang
build time (s)	105	4,813	95
ffmpeg size	279,584	279,576	275,480
ffprobe size	168,040	155,712	180,344

Benchmark #1: ./ffmpeg_gcc -threads 1 -i BigBuckBunny-AV1.webm -benchmark -f null -
  Time (mean ± σ):      1.188 s ±  0.015 s    [User: 1.175 s, System: 0.013 s]
  Range (min … max):    1.163 s …  1.212 s    10 runs

Benchmark #1: ./ffmpeg_gcc_lto -threads 1 -i BigBuckBunny-AV1.webm -benchmark -f null -
  Time (mean ± σ):      1.180 s ±  0.014 s    [User: 1.167 s, System: 0.012 s]
  Range (min … max):    1.156 s …  1.201 s    10 runs

Benchmark #1: ./ffmpeg_clang -threads 1 -i BigBuckBunny-AV1.webm -benchmark -f null -
  Time (mean ± σ):      1.165 s ±  0.013 s    [User: 1.156 s, System: 0.009 s]
  Range (min … max):    1.150 s …  1.188 s    10 runs

The build time difference on GitHub actions was an increase of 3-5min (on a 20min build)

I'd say it's worth it, even if the advantages are (currently) limited, they will only improve with newer compilers.

[nyanmisaka]

Don't forget to add it for windows build.

jellyfin-ffmpeg/docker-build-win64.sh

Line 514 in 99b6c47

--enable-gpl \

[nyanmisaka]

LTGM, even though it will slow down the build speed