GCC 10.1 Compiler Optimization Benchmarks

[AHSauge]

Ok, so I am not allowed to be interested in that. Got it.

I'm asking you why this is useful, and your response is "so I am not allowed to be interested in that"? That makes a lot of sense to me!

Follow up question: Do you even do programming?

[CochainComplex]

pgo will be interessting especially the new -fprofile-partial-training might change the Performance with cases where no profile for everything exists. whithout this flag code code not civered by a profile will be optimized for size. this flag just compiles that code generic.

[rene]

Now a comparison with the previous gcc and llvm/clang version would have been interesting. Freestanding this is a bit plumbed.

[Pentarctagon]

[FONT=Courier New]It's been irrelevant for ages now. Hardly faster than -O1.

Smaller executable size for Os than O1 though.

[FireBurn]

Please could we see the difference with -march=native at -O1 -O2 -O3 with and without lto

I'm running Gentoo with -O3 -march=native -flto and I'd be curious to know if -O2 would be better and if -flto is worth the added compiling / linking time

[oleid]

Please could we see the difference with -march=native at -O1 -O2 -O3 with and without lto

I'm running Gentoo with -O3 -march=native -flto and I'd be curious to know if -O2 would be better and if -flto is worth the added compiling / linking time

I didn't use gentoo for a long time. But using O3 system wide wasn't a good idea back then. Sometimes binaries were even slower due to cache misses and the startup time was worse.

Is it nowadays more pleasant to use lto system wide?

[FireBurn]

It works fine here, but it's not benchmarked

[greyseek3r]

Given that -O2 is at times faster than -O3 on generic x86_64, it would be interesting to see a comparison between -O2 -march=native and -O3 -march=native, and -O3 -march=native with PGO and LTO. AFAIK LTO has limited benefits with O2, right?

[Azpegath]

Please could we see the difference with -march=native at -O1 -O2 -O3 with and without lto

I'm running Gentoo with -O3 -march=native -flto and I'd be curious to know if -O2 would be better and if -flto is worth the added compiling / linking time

I agree with this, Michael any chance we can get to see that?

[JustinTurdeau]

It's been irrelevant for ages now. Hardly faster than -O1.

Obvious troll...