Minor follow-up to this topic:
I submitted test results to CDash with gcc6.3 on Ubuntu and extra release
flags https://cdash.kyomu.43-1.org/viewTest.php?onlyfailed&buildid=4886
-march=native -O3 -funsafe-loop-optimizations -flto
A few tests fail (some with binary output, some with roundoff
A useful flag that I use in release mode is -funsafe-loop-optimizations,
which assumes that loop counters cannot overflow and other such things. I
don't remember the exact numbers but I do remember that using this flag
made a small, but measurable, performance improvement.
On Thursday, Septembe
Thanks Bruno and Martin for the swift reply. For starters i will try with
-march=native .
Cheers,
Denis.
On Thursday, September 15, 2016 at 2:04:58 PM UTC+2, Denis Davydov wrote:
>
> Dear all,
>
> From the matrix-free tutorials 37 and 48, I see that the recommended flags
> for VectorizedArrays
Denis,
I usually use -march=native, -O3, and -flto I never dare to use -ffast-math
because my understanding is that your code can give wrong results
https://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html
Best,
Bruno
On Thursday, September 15, 2016 at 8:04:58 AM UTC-4, Denis Davydov wrote:
>
