numpy build fails on powerpc ydl
My build of numpy fails under Yellow Dog Linux 2.1, running on a powerpc multiprocessor board from Curtiss-Wright. Its kernel is 2.4.19-Asmp tailored by the vendor. The gcc compiler is configured as ppc-yellowdog-linux with version number 2.95.3 20010111. The python I'm using is Python 2.5.1 (r251:54863) installed as python2. Plain /usr/bin/python is 1.5.x . The numpy version I'm trying to build is r4003 for v1.0.4 . The setup fails compiling build/src.linux-ppc-2.5/numpy/core/src/umathmodule.c with a long list of error messages of the following two kinds. warning: conflicting types for built-in function `sinl' repeated for `cosl', `fabsl', and `sqrtl', triggered by line 442. inconsistent operand constraints in an `asm', triggered by lines 1100, 1124, 1150, 1755, 1785, and 1834. I cannot see on those source lines what causes such a message; I suspect there is some long complicated cpp macro or asm statement in some include file which I don't find. Has anyone tried building numpy on Yellow Dog Linux or on a PowerPC with gcc? Vincent Broman broman@spawar.navy.mil
Vincent Broman wrote:
My build of numpy fails under Yellow Dog Linux 2.1, running on a powerpc multiprocessor board from Curtiss-Wright.
Its kernel is 2.4.19-Asmp tailored by the vendor. The gcc compiler is configured as ppc-yellowdog-linux with version number 2.95.3 20010111. The python I'm using is Python 2.5.1 (r251:54863) installed as python2. Plain /usr/bin/python is 1.5.x . The numpy version I'm trying to build is r4003 for v1.0.4 .
The setup fails compiling build/src.linux-ppc-2.5/numpy/core/src/umathmodule.c with a long list of error messages of the following two kinds.
warning: conflicting types for built-in function `sinl' repeated for `cosl', `fabsl', and `sqrtl', triggered by line 442.
You may be the first one to build on this platform. What needs to happen is that the correct config.h file needs to be set up for that platform. The long-float versions of certain functions are being incorrectly identified. Would you be willing to help get the config.h file set up correctly? -Travis
On 8/30/07, Vincent Broman <broman@spawar.navy.mil> wrote:
My build of numpy fails under Yellow Dog Linux 2.1, running on a powerpc multiprocessor board from Curtiss-Wright.
Its kernel is 2.4.19-Asmp tailored by the vendor.
Which vendor? The gcc compiler is configured as ppc-yellowdog-linux with
version number 2.95.3 20010111.
That compiler is really, I mean really, ancient. And the API changed in newer gcc (> 3.x.x), so code compiled with later versions isn't binary compatible. Hmmm. Curtiss-Wright now supports Linux and kernel 2.6.16 on some of their newer hardware, you might want to check with them or install a more current distro from Fedora or someone else who supports the PPC. The python I'm using is Python 2.5.1 (r251:54863) installed as python2.
Plain /usr/bin/python is 1.5.x . The numpy version I'm trying to build is r4003 for v1.0.4 .
The setup fails compiling build/src.linux-ppc-2.5 /numpy/core/src/umathmodule.c with a long list of error messages of the following two kinds.
warning: conflicting types for built-in function `sinl' repeated for `cosl', `fabsl', and `sqrtl', triggered by line 442.
Any more detail on these? What causes the conflict. I've got to wonder about the the libc/libm versions also. Does the include file math.h say anything about the prototypes for these functions? I expect cosl et.al. to be potential problems on the PPC anyway due to the way long doubles were implemented. inconsistent operand constraints in an `asm',
triggered by lines 1100, 1124, 1150, 1755, 1785, and 1834.
I cannot see on those source lines what causes such a message; I suspect there is some long complicated cpp macro or asm statement in some include file which I don't find.
What is the PPC model number? Chuck
participants (3)
-
Charles R Harris -
Travis E. Oliphant -
Vincent Broman