1 |
Volker Armin Hemmann wrote: |
2 |
> |
3 |
> install the gcc-porting and toolchain overlays. Have fun. Or feel the pain. |
4 |
> look at this first: |
5 |
> http://bugs.gentoo.org/show_bug.cgi?id=249226 |
6 |
> |
7 |
> |
8 |
|
9 |
Thanks for the answer. I emerged it and it seems to be working, even |
10 |
though I am using pretty aggressive flags: |
11 |
|
12 |
CFLAGS="-march=native -O3 -pipe -fgcse-sm -fgcse-las -fivopts |
13 |
-ftree-loop-linear -floop-interchange -floop-strip-mine -floop-block |
14 |
-ftree-loop-distribution -ftracer -funroll-loops -fprefetch-loop-arrays |
15 |
-freorder-blocks-and-partition -funswitch-loops -fsee -ftree-loop-im |
16 |
-ftree-loop-ivcanon -fvariable-expansion-in-unroller" |
17 |
|
18 |
|
19 |
I tried also using -ftree-parallelize-loops=4, but with it compiler |
20 |
fails when I try to compile practically anything. |
21 |
|
22 |
|
23 |
BTW: How much of a performance gain would be if one could compile |
24 |
everything with -profile-generate, run resulting program on some |
25 |
representative test data and then use profile files to generate final |
26 |
binary with -fprofile-use ? |