Fedora 8 packages to exclude

GNU Radio
1

Just need a clearification: the gnuradio FC8 installation procedure

http://gnuradio.utah.edu/trac/wiki/PS3FC8Install

says to put

exclude=blas blas-devel oprofile-debuginfo oprofile numactl nuactl-devel

in /etc/yum.conf, but the SDK3.0 installation guide, on p.35 at the
bottom states to remove any such exclude? I’m guessing it all means to
block and remove those guys before installing the sdk, disable the
fedora repos, then un-exclude and install the sdk? On a previous attempt
part of the sdk complained about missing one.

TIA

–Chuck

2

On Tue, Jul 01, 2008 at 06:05:28PM -0400, Charles S. wrote:

block and remove those guys before installing the sdk, disable the
fedora repos, then un-exclude and install the sdk? On a previous attempt
part of the sdk complained about missing one.

Yes. That’s the theory. In theory the SDK has better code than the
Fedora (and that may have been true in FC7). I’m not sure that’s
still the case. The numactl* stuff doesn’t work on the PS3, so it
doesn’t matter. I haven’t looked to see if they’re really installing
a cell optimized blas or not. Last I checked, oprofile on F8 was
better than the SDK at least when running on the QS21 blades (no
hypervisor). You’ll have to see how oprofile works on the PS3.
There’s also another ps3 specific performance tool in Geoff’s stuff,
though I don’t remember what it’s called.

Eric

3

On Tue, 2008-07-01 at 20:00 -0400, Charles S. wrote:

Ok - I’ve redone this several times with variations; this last time was
very carefuly ‘by the book’ and I run into this after a fresh fc8,
updates, sdk3.0, Devel Tools, kernel 2.6.25.6-27.fc8, not cross
compiling:

Appearently it is picking up the wrong g++ (4.1.2) which does not match
the sdk headers in /usr/lib/gcc/ppu/4.1.1/include/

… but I’m at a loss how to fix that. export CPP=/usr/bin/ppu-g++
makes configure fail sanity check.

Opps, make it export CXX=/usr/bin/ppu-g++ and it gets past that spot.

–Chuck

4

On Tue, 2008-07-01 at 15:22 -0700, Eric B. wrote:

On Tue, Jul 01, 2008 at 06:05:28PM -0400, Charles S. wrote:

Just need a clearification: the gnuradio FC8 installation procedure

http://gnuradio.utah.edu/trac/wiki/PS3FC8Install

Yes. That’s the theory. In theory the SDK has better code than the
Fedora (and that may have been true in FC7). I’m not sure that’s
still the case. The numactl* stuff doesn’t work on the PS3, so it

Ok - I’ve redone this several times with variations; this last time was
very carefuly ‘by the book’ and I run into this after a fresh fc8,
updates, sdk3.0, Devel Tools, kernel 2.6.25.6-27.fc8, not cross
compiling:

make[5]: Entering directory `/usr/src/trunk/gcell/src/lib/runtime’
/bin/sh …/…/…/…/libtool --tag=CXX --mode=compile g++
-DHAVE_CONFIG_H -I. -I…/…/…/… -DOMNITHREAD_POSIX=1
-I/usr/src/trunk/omnithread -I/usr/src/trunk/mblock/src/lib
-I/usr/src/trunk/gcell/src/include
-I/usr/src/trunk/gcell/src/lib/runtime
-I/usr/src/trunk/gcell/src/lib/general
-I/usr/src/trunk/gcell/src/lib/wrapper
-I…/…/…/…/gcell/src/ibm/sync/ppu_source -g -O2 -Wall
-Woverloaded-virtual -pthread -MT gc_job_manager_impl.lo -MD -MP
-MF .deps/gc_job_manager_impl.Tpo -c -o gc_job_manager_impl.lo
gc_job_manager_impl.cc
g++ -DHAVE_CONFIG_H -I. -I…/…/…/… -DOMNITHREAD_POSIX=1
-I/usr/src/trunk/omnithread -I/usr/src/trunk/mblock/src/lib
-I/usr/src/trunk/gcell/src/include
-I/usr/src/trunk/gcell/src/lib/runtime
-I/usr/src/trunk/gcell/src/lib/general
-I/usr/src/trunk/gcell/src/lib/wrapper
-I…/…/…/…/gcell/src/ibm/sync/ppu_source -g -O2 -Wall
-Woverloaded-virtual -pthread -MT gc_job_manager_impl.lo -MD -MP
-MF .deps/gc_job_manager_impl.Tpo -c gc_job_manager_impl.cc -fPIC -DPIC
-o .libs/gc_job_manager_impl.o
In file included from gc_job_manager_impl.cc:32:
…/…/…/…/gcell/src/ibm/sync/ppu_source/atomic_dec_if_positive.h:44:28:
error: ppu_intrinsics.h: No such file or directory

Appearently it is picking up the wrong g++ (4.1.2) which does not match
the sdk headers in /usr/lib/gcc/ppu/4.1.1/include/

… but I’m at a loss how to fix that. export CPP=/usr/bin/ppu-g++
makes configure fail sanity check.

–Chuck

5

On Tue, Jul 01, 2008 at 08:00:31PM -0400, Charles S. wrote:

-I/usr/src/trunk/omnithread -I/usr/src/trunk/mblock/src/lib
-I/usr/src/trunk/gcell/src/include

Appearently it is picking up the wrong g++ (4.1.2) which does not match
the sdk headers in /usr/lib/gcc/ppu/4.1.1/include/

… but I’m at a loss how to fix that. export CPP=/usr/bin/ppu-g++
makes configure fail sanity check.

CPP’s the C preprocessor :wink:

You’ll want CXX=…

However, I suggest that you just use

./configure-cell-cross --prefix=FOO

when building for the cell, even if native.

Eric

6

On Tue, Jul 01, 2008 at 08:08:06PM -0400, Charles S. wrote:

the sdk headers in /usr/lib/gcc/ppu/4.1.1/include/

… but I’m at a loss how to fix that. export CPP=/usr/bin/ppu-g++
makes configure fail sanity check.

Opps, make it export CXX=/usr/bin/ppu-g++ and it gets past that spot.

ppu32-g++ is even better. That builds 32-bit executables, instead of
64-bit. On the PS3, with it’s limited memory, there’s no reason to
use 64-bit.

Eric

7

On Tue, Jul 01, 2008 at 11:17:41PM -0400, Charles S. wrote:

PROBLEM IS HERE:
the ps3, but of course not for the cell). Example
in /opt/cell/sdk/src/tutorial/simple compiles just fine using ppu32-gcc.

What could be the major malfunction or missing peice?

–Chuck

On the PS3, did you run:

ps3$ sudo
/mnt/share/gr/trunk/dtools/bin/tweak-cell-for-cross-compiling

(From http://gnuradio.org/trac/wiki/CrossCompilingForCell)

Eric

8

On Tue, 2008-07-01 at 20:08 -0400, Charles S. wrote:

On Tue, 2008-07-01 at 20:00 -0400, Charles S. wrote:

Ok - I’ve redone this several times with variations; this last time was
very carefuly ‘by the book’ and I run into this after a fresh fc8,

Ok, the ps3 runs fine but of course takes forever on the big
gnuradio_swig_py_general.cc file so, setting up the cross-compile
machine with F8, updates, even uninstall/install sdk3.0 but keep running
into issue with configure. From config.log:

Target: ppu
Configured with: …/toolchain/gcc/configure --prefix=/opt/cell/toolchain
–mandir=/opt/cell/toolchain/man --infodir=/opt/cell/toolchain/info
–with-sysroot=/opt/cell/sysroot
–with-as=/opt/cell/toolchain/bin/ppu-as
–with-ld=/opt/cell/toolchain/bin/ppu-ld --enable-threads
–with-system-zlib --disable-checking --enable-__cxa_atexit
–disable-libunwind-exceptions --enable-languages=c,c++,fortran,ada
–disable-nls --enable-clocale=gnu
–enable-version-specific-runtime-libs --with-long-double-128
–program-prefix=ppu- --disable-bootstrap --build=i686-pc-linux-gnu
–target=ppu
Thread model: posix
gcc version 4.1.1
configure:3367: $? = 0
configure:3374: ppu32-gcc -V >&5
ppu-gcc: ‘-V’ must come at the start of the command line
configure:3377: $? = 1
configure:3400: checking for C compiler default output file name

PROBLEM IS HERE:
configure:3427: ppu32-gcc -I/mnt/cell-root/usr/include
-L/mnt/cell-root/lib -L/mnt/cell-root/usr/lib conftest.c >&5
/opt/cell/toolchain/bin/ppu-ld: skipping incompatible /lib/libc.so.6
when searching for /lib/libc.so.6

/opt/cell/toolchain/bin/ppu-ld: cannot find /lib/libc.so.6
collect2: ld returned 1 exit status

There IS no other libc.so.6 (it is a link to libc-2.7.so just like on
the ps3, but of course not for the cell). Example
in /opt/cell/sdk/src/tutorial/simple compiles just fine using ppu32-gcc.

What could be the major malfunction or missing peice?

–Chuck