Micro/minor g adjust optimisation
David Dickinson
Branch: micro/minor_g_adjust_optimisation
Branch: next
Merged
Merged pull request
Merged in micro/minor_g_adjust_optimisation (pull request #417)
Merged in micro/minor_g_adjust_optimisation (pull request #417)
Avoid making a variable saved in g_adjust
Slightly optimise g_adjust by pulling repeat calculations out of inner loop
Local testing shows this to be 10-20% faster than next when compiled with gnu-5 using
-O3
. It seems next can approach this performance if compiled with-march=native -Ofast
. This suggests that these changes are perhaps doing what the compiler would like to do but can’t at O3 as it would reorder the floating point calculations.