[LLVMdev] [LNT] Question about results reliability in LNT infrustructure
renato.golin at linaro.org
Mon Jul 1 12:26:09 PDT 2013
On 1 July 2013 19:13, Jakob Stoklund Olesen <stoklund at 2pi.dk> wrote:
> "We find that the performance impact of -O3 over -O2 optimizations is
> indistinguishable from random noise.”
Yes, we circulated that paper a few weeks ago on the list.
This reminds me of an interesting project I saw that runs a genetic
algorithm on the LLVM optimization passes using opt and you can get some
interesting results from them, that would help you distinguish O2 from O3,
but on a test-by-test basis. There's no way yet to know beforehand what is
the best combination of passes for each specific input.
That is, if I'm not mistaken, one of the reasons to enable the loop
vectorizer on -O2 as -Os by default, as well as -O3, on LLVM.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the llvm-dev