| 1 |
Hi James, |
| 2 |
|
| 3 |
It is as I suspected according to your log file - |
| 4 |
|
| 5 |
> --disable-libgcj |
| 6 |
> --enable-languages=c,c++,fortran |
| 7 |
|
| 8 |
Fortran has been a bit problematic with crossdev - I usually have to |
| 9 |
build my toolchains without fortran support, and that requires: |
| 10 |
|
| 11 |
USE="-fortran" |
| 12 |
ECONF_EXTRA="--with-languages=c,c++" |
| 13 |
|
| 14 |
When compiling GCC, the fortran component of gcc looks for libgmp and |
| 15 |
also libmfpr (or one of them at least), both in the "rpath". |
| 16 |
|
| 17 |
The "rpath" is usually /lib or /usr/lib (not ${SYSROOT}/lib or any other |
| 18 |
default compiler search path). |
| 19 |
|
| 20 |
Obviously, when cross-compiling, the libgmp / libmfpr that it finds is a |
| 21 |
.so file for i686 or whatever your build architecture is, and they are |
| 22 |
definitely not compatible with arm object code. |
| 23 |
|
| 24 |
I would only suggest trying to add a fortran compiler when you're |
| 25 |
compiling gcc _natively_ on your board, although that can be quite a |
| 26 |
tedious process and you'll likely need some swap in the form of an SD |
| 27 |
card or USB memory stick. |
| 28 |
|
| 29 |
By the way, I still haven't done any hardware floating point |
| 30 |
acceleration testing because I haven't had all that much spare time |
| 31 |
lately. So if you are up for it, try it out. |
| 32 |
|
| 33 |
Martin Guy has a few benchmarks and the average speedups with LAME are |
| 34 |
|
| 35 |
OABI -> EABI: ~11 |
| 36 |
EABI/softfp -> EABI/hardfp: 2 |
| 37 |
|
| 38 |
That means OABI -> EABI/hardfp: ~22. |
| 39 |
|
| 40 |
|
| 41 |
~/Chris |
| 42 |
-- |
| 43 |
gentoo-embedded@l.g.o mailing list |
Archives