LZ4 Compiler Optimization Flags

[Shawn]

I have been testing using the fullbench executable in the tests directory, and changed around the -O3 flag in the Makefile to -O2. I noticed that after testing both flags a few times, the -O2 flag was consistently faster then the -O3 flag by around 5%. Is there any specific reason for the -O3 flag being called instead of the -O2 flags?

[Cyan]

It used to be that -O3 produced significantly faster binaries than -O2.

It changed a few versions ago,

both as a consequence of source code modifications, which are friendlier to compiler's optimizations,

and as a consequence of compiler's improvements.

Note though that this result (-O2 faster than -O3) is not a sure outcome.

You will find that such difference vary, depending on compiler type and version.

If I do remember correctly, I considered switching to -O2 a while ago,

but shelved the idea after finding a few configurations in which -O2 was actually slower.

I'm likely to reconsider it in the future.