|
|
|
|
|
|
by zxcdw
4636 days ago
|
|
|
The article doesn't mention reasoning for using -O2 instead of -O3 or -Ofast. It seems that they give better performance in this case, as does adding -funroll-all-loops(when combined with -Ofast gives almost 10 % speedup). I am compiling with GCC 4.8.1, so perhaps authors 4.6.3 (what?) behaves differently. |
|
|