1 |
On 9/5/11 11:43 PM, Stefan G. Weichinger wrote: |
2 |
> |
3 |
> Anyone else using Intel's compiler, icc? |
4 |
|
5 |
Hi Stefan, |
6 |
|
7 |
try to stick to gcc as most pacakges will compile with it. |
8 |
|
9 |
I personally use icc/ifort for some sience packages and see speedups of |
10 |
calculation between 2-25x depending on the *FLAGS. But this needs much |
11 |
optimization of the flags. Nevertheless interesting for performance |
12 |
critical apps. |
13 |
|
14 |
Another compiler which was recently released after a long time as closed |
15 |
source app into the open source world is the ekopath compiler suite |
16 |
(ekopath(-bin) and path64) which proofed to have the best optimization |
17 |
of all compilers in benchmarks. But same as icc, it might not work with |
18 |
some packages. |
19 |
|
20 |
And never use it with the kernel. |
21 |
|
22 |
justin |
23 |
|
24 |
|
25 |
> |
26 |
> I do for quite a while now. |
27 |
> Followed http://www.gentoo-wiki.info/HOWTO_ICC_and_Portage |
28 |
> |
29 |
> I still prefer gcc over icc so I use |
30 |
> |
31 |
> http://www.gentoo-wiki.info/HOWTO_ICC_and_Portage#.2Fetc.2Fportage.2Fbashrc |
32 |
> |
33 |
> to only use icc for stuff I list in /etc/portage/package.icc |
34 |
> |
35 |
> - |
36 |
> |
37 |
> Lately I get compilation-errors for packages that aren't listed in that |
38 |
> file, for example dev-python/numpy-1.6.1 |
39 |
> |
40 |
> It fails with log-lines telling me that icc was used to (try to) compile it: |
41 |
> |
42 |
> [..] |
43 |
> icc: command line warning #10156: ignoring option '-fp'; no argument |
44 |
> required |
45 |
> icc: error #10236: File not found: 'l1-cache-line-size=64' |
46 |
> icc: command line warning #10156: ignoring option '-fp'; no argument |
47 |
> required |
48 |
> icc: error #10236: File not found: 'l2-cache-size=4096' |
49 |
> error: Command "icc -pthread -shared -Wl,-O1 -Wl,--as-needed -shared -O2 |
50 |
> -D_FORTIFY_SOURCE=2 -march=core2 -mcx16 -msahf --param l1-cache-size=32 |
51 |
> --param l1-cache-line-size=64 --param l2-cache-size=4096 -mtune=core2 |
52 |
> -fno-strict-aliasing |
53 |
> build-2.7/temp.linux-x86_64-2.7/build-2.7/src.linux-x86_64-2.7/numpy/core/src/_sortmodule.o |
54 |
> -L/usr/lib64 -Lbuild-2.7/temp.linux-x86_64-2.7 -lnpymath -lm -lpython2.7 |
55 |
> -o build-2.7/lib.linux-x86_64-2.7/numpy/core/_sort.so" failed with exit |
56 |
> status 1 |
57 |
> |
58 |
> |
59 |
> # grep numpy /etc/portage/* |
60 |
> # |
61 |
> |
62 |
> The shell is bash: |
63 |
> |
64 |
> # echo $SHELL |
65 |
> /bin/bash |
66 |
> |
67 |
> hmm |
68 |
> |
69 |
> Yeah, I know, I could simply get rid of icc again. |
70 |
> |
71 |
> But maybe someone in here has an idea why this fails. |
72 |
> Maybe it somehow needs some small fix somewhere. |
73 |
> |
74 |
> Stefan |
75 |
> |