Benchmarking Mercury As The "Fastest Firefox Fork" With AVX, AES, LTO + PGO

[phoronix]

Phoronix: Benchmarking Mercury As The "Fastest Firefox Fork" With AVX, AES, LTO + PGO

Following the news last week of Firefox outperforming Chrome in SunSpider, a Phoronix reader pointed out Mercury that is an open-source web browser claiming to be the "fastest Firefox fork" and making use of Advanced Vector Extensions (AVX) and AES instructions along with compiler features like Link-Time Optimizations (LTO) and Profile-Guided Optimizations (PGO). The project advertises as being 8-20% faster than upstream Firefox. Curious I ran a couple benchmarks on my end of this Firefox fork.

https://www.phoronix.com/review/mercury-firefox-perf

[geearf]

I'd like to see Thorium vs Chromium vs some other fork as well.
Seeing how Mercury compares make me wonder how that would be.

[edxposed]

Chromium:

[bachchain]

Oh, I get it. The mercury people were looking at their graphs backwards.

[andyprough]

I don't have a horse in the race as I don't use Firefox or Mercury, but continuing to use a bloated beast that labors under its own weight like Ubuntu for benchmarking browsers seems like a strange choice. I don't know if the results are applicable to anything I would use.

[chromer]

I think benchmarking browsers on Linux VS mac VS Windows , is a better approach, to see on which OS firefox and chrome have best performance.

[oleid]

I don't have a horse in the race as I don't use Firefox or Mercury, but continuing to use a bloated beast that labors under its own weight like Ubuntu for benchmarking browsers seems like a strange choice. I don't know if the results are applicable to anything I would use.

As bloated as Ubuntu is, I highly doubt it would make any difference since all browsers run on the same system.

[oleid]

Doesn't Firefox use LTO+PGO as well?

[marios]

I think benchmarking browsers on Linux VS mac VS Windows , is a better approach, to see on which OS firefox and chrome have best performance.

Things are strange with macOS, especially for M* macs. Since there are only 2 microarchitectures (that are not too different) and their variants, binaries built for M* macs are compiled with mcpu=apple-m1 or something similar. It can only be compared to march=native on x86 CPUs, which can only happen if you built Firefox and the entire system yourself.

PS. For M2 the flags could be suboptimal, but compared to the shitshow of x86 march=generic it is not nearly as bad.

[Danny3]

Doesn't Firefox use LTO+PGO as well?

Not when it's built by Debian's maintainers as that build definitely sucks as it much slower than the build built by Mozilla.