GCC 7.0 vs. LLVM Clang 4.0 Performance (January 2017)

[AsuMagic]

AFAIK, today many projects use llvm during development but ship their production builds build with gcc. Would it be possible to compare the performance of real life applications such as firefox, dorktable or some open source games?

dorktable... that's rude
I don't think you will get lots of interest on firefox with browser benchmarks, since I assume the work is mostly done by the JIT'd code.
But for open source games, that would be pretty cool indeed.

[wizard69]

Looks like a guy can hardly go wrong here choosing a compiler,

[Michael]

The problem is that this benchmark is now measuring any bugs in the governor, the CPU frequency potentially jumping more than 50%

Given the same governor was used throughout testing and that each benchmark is run a minimum of three times and increases if std dev is too great, and the deviations are shown on the graphs and nothing appeared awkward with these results, I don't think it's too much concern.

[rx80]

Ubuntu default...

The problem is that this benchmark is now measuring any bugs in the governor, the CPU frequency potentially jumping more than 50%

[Michael]

Why are the tests done with powersave mode? "Scaling Governor: intel_pstate powersave"

Ubuntu default...

[rx80]

Why are the tests done with powersave mode? "Scaling Governor: intel_pstate powersave"

[dispat0r]

AFAIK, today many projects use llvm during development but ship their production builds build with gcc. Would it be possible to compare the performance of real life applications such as firefox, dorktable or some open source games?

This, and maybe with the use of LTO.

[howarth]

AFAIK, today many projects use llvm during development but ship their production builds build with gcc. Would it be possible to compare the performance of real life applications such as firefox, dorktable or some open source games?

FYI, the OpenMandriva linux distribution, https://www.openmandriva.org, is built almost entirely with clang 3.9.1. https://wiki.openmandriva.org/en/3.01/Release_Notes.

[treba]

AFAIK, today many projects use llvm during development but ship their production builds build with gcc. Would it be possible to compare the performance of real life applications such as firefox, dorktable or some open source games?

[Michael]

I like the win/loss report; it gives a nice quick overview and basically helps summarize results to add value above and beyond the "here's all the raw data"

With PTS7 git: phoronix-test-suite winners-and-losers <result file> if any other metrics/stats you would find interesting, can easily add. If it ends up adding more stuff to it, will probably rename from winners-and-losers.