Crafty22.10 numa and Lenny

[frankp]

Bob

Just upgraded to Debian Lenny, tried to compile crafty-22.10 and got the numa related error below.

(Crafty works fine compiled with numa disabled.).

Frank


main.c: In function ‘main’:
main.c:3763: warning: passing argument 2 of ‘numa_node_to_cpus’ from incompatible pointer type
main.c:3763: error: too many arguments to function ‘numa_node_to_cpus’
make[2]: *** [crafty.o] Error 1
make[2]: Leaving directory `/home/fp/crafty/crafty-22.10'
make[1]: *** [crafty-make] Error 2
make[1]: Leaving directory `/home/fp/crafty/crafty-22.10'
make: *** [linux-amd64] Error 2
fp@debian:~/crafty/crafty-22.10$

[okoli]

I don't have any atm time so I make it short:

Put the sources in your apt-sources.list to (root):
deb-src http://ftp.de.debian.org/debian/ sid main contrib non-free

Then do a (user):

apt-get source crafty

cd into the crafty dir and do a:

dpkg-buildpackage -rfakeroot

You will receive a neatly done crfty package.

Install it as root.

Oliver

[frankp]

Thanks.

I was wondering if the numa library had been updated from that assumed in crafty.

[bob]

Bob

Just upgraded to Debian Lenny, tried to compile crafty-22.10 and got the numa related error below.

(Crafty works fine compiled with numa disabled.).

Frank


main.c: In function ‘main’:
main.c warning: passing argument 2 of ‘numa_node_to_cpus’ from incompatible pointer type
main.c error: too many arguments to function ‘numa_node_to_cpus’
make[2]: *** [crafty.o] Error 1
make[2]: Leaving directory `/home/fp/crafty/crafty-22.10'
make[1]: *** [crafty-make] Error 2
make[1]: Leaving directory `/home/fp/crafty/crafty-22.10'
make: *** [linux-amd64] Error 2
fp@debian:~/crafty/crafty-22.10$

The libnuma stuff might have changed. First and most obvious question is are you using an AMD box with multiple cpus? That is the only place NUMA is needed... Intel boxen are not NUMA at present. At least not the 2-4 cpu boxes most can actually afford.

[bob]

Thanks.

I was wondering if the numa library had been updated from that assumed in crafty.

I am checking on this now. Last time I used this code was on the 8-way opteron boxes I used in some of the past CCT events...

[bob]

Bob

Just upgraded to Debian Lenny, tried to compile crafty-22.10 and got the numa related error below.

(Crafty works fine compiled with numa disabled.).

Frank


main.c: In function ‘main’:
main.c warning: passing argument 2 of ‘numa_node_to_cpus’ from incompatible pointer type
main.c error: too many arguments to function ‘numa_node_to_cpus’
make[2]: *** [crafty.o] Error 1
make[2]: Leaving directory `/home/fp/crafty/crafty-22.10'
make[1]: *** [crafty-make] Error 2
make[1]: Leaving directory `/home/fp/crafty/crafty-22.10'
make: *** [linux-amd64] Error 2
fp@debian:~/crafty/crafty-22.10$

OK, I looked around and the definition was changed around November 2008, but it is not clear how it is going to be modified as this change has broken a lot of code and the author is apparently working on a backward-compatibility fix to solve this. At the moment, a fix is unclear...

[frankp]

Bob

Just upgraded to Debian Lenny, tried to compile crafty-22.10 and got the numa related error below.

(Crafty works fine compiled with numa disabled.).

Frank


main.c: In function ‘main’:
main.c:3763: warning: passing argument 2 of ‘numa_node_to_cpus’ from incompatible pointer type
main.c:3763: error: too many arguments to function ‘numa_node_to_cpus’
make[2]: *** [crafty.o] Error 1
make[2]: Leaving directory `/home/fp/crafty/crafty-22.10'
make[1]: *** [crafty-make] Error 2
make[1]: Leaving directory `/home/fp/crafty/crafty-22.10'
make: *** [linux-amd64] Error 2
fp@debian:~/crafty/crafty-22.10$

The libnuma stuff might have changed. First and most obvious question is are you using an AMD box with multiple cpus? That is the only place NUMA is needed... Intel boxen are not NUMA at present. At least not the 2-4 cpu boxes most can actually afford.

No: just quad core. As noted, compiles and works fine with NUMA disabled. Just seems that some definitions in (the later) libnuma had changed from that assumed in the code.

[bob]

Bob

Just upgraded to Debian Lenny, tried to compile crafty-22.10 and got the numa related error below.

(Crafty works fine compiled with numa disabled.).

Frank


main.c: In function ‘main’:
main.c warning: passing argument 2 of ‘numa_node_to_cpus’ from incompatible pointer type
main.c error: too many arguments to function ‘numa_node_to_cpus’
make[2]: *** [crafty.o] Error 1
make[2]: Leaving directory `/home/fp/crafty/crafty-22.10'
make[1]: *** [crafty-make] Error 2
make[1]: Leaving directory `/home/fp/crafty/crafty-22.10'
make: *** [linux-amd64] Error 2
fp@debian:~/crafty/crafty-22.10$

The libnuma stuff might have changed. First and most obvious question is are you using an AMD box with multiple cpus? That is the only place NUMA is needed... Intel boxen are not NUMA at present. At least not the 2-4 cpu boxes most can actually afford.

No: just quad core. As noted, compiles and works fine with NUMA disabled. Just seems that some definitions in (the later) libnuma had changed from that assumed in the code.

They did. and it has caused a lot of complaints. In any case, you don't want it defined anyway since your box is not NUMA and enabling the NUMA code won't add a thing for you...

[mhull]

Bob

Just upgraded to Debian Lenny, tried to compile crafty-22.10 and got the numa related error below.

(Crafty works fine compiled with numa disabled.).

Frank


main.c: In function ‘main’:
main.c warning: passing argument 2 of ‘numa_node_to_cpus’ from incompatible pointer type
main.c error: too many arguments to function ‘numa_node_to_cpus’
make[2]: *** [crafty.o] Error 1
make[2]: Leaving directory `/home/fp/crafty/crafty-22.10'
make[1]: *** [crafty-make] Error 2
make[1]: Leaving directory `/home/fp/crafty/crafty-22.10'
make: *** [linux-amd64] Error 2
fp@debian:~/crafty/crafty-22.10$

The libnuma stuff might have changed. First and most obvious question is are you using an AMD box with multiple cpus? That is the only place NUMA is needed... Intel boxen are not NUMA at present. At least not the 2-4 cpu boxes most can actually afford.

No: just quad core. As noted, compiles and works fine with NUMA disabled. Just seems that some definitions in (the later) libnuma had changed from that assumed in the code.

They did. and it has caused a lot of complaints. In any case, you don't want it defined anyway since your box is not NUMA and enabling the NUMA code won't add a thing for you...

Does your dual-quad CCT-11 hardware need NUMA?

[bob]

Bob

Just upgraded to Debian Lenny, tried to compile crafty-22.10 and got the numa related error below.

(Crafty works fine compiled with numa disabled.).

Frank


main.c: In function ‘main’:
main.c warning: passing argument 2 of ‘numa_node_to_cpus’ from incompatible pointer type
main.c error: too many arguments to function ‘numa_node_to_cpus’
make[2]: *** [crafty.o] Error 1
make[2]: Leaving directory `/home/fp/crafty/crafty-22.10'
make[1]: *** [crafty-make] Error 2
make[1]: Leaving directory `/home/fp/crafty/crafty-22.10'
make: *** [linux-amd64] Error 2
fp@debian:~/crafty/crafty-22.10$

The libnuma stuff might have changed. First and most obvious question is are you using an AMD box with multiple cpus? That is the only place NUMA is needed... Intel boxen are not NUMA at present. At least not the 2-4 cpu boxes most can actually afford.

No: just quad core. As noted, compiles and works fine with NUMA disabled. Just seems that some definitions in (the later) libnuma had changed from that assumed in the code.

They did. and it has caused a lot of complaints. In any case, you don't want it defined anyway since your box is not NUMA and enabling the NUMA code won't add a thing for you...

Does your dual-quad CCT-11 hardware need NUMA?

No. At present, if you stick with the X86-64 world and 2-way or 4-way boxes, AMD machines are the only ones that use NUMA. And, in fact, without giving up too much, you can ignore the NUMA issues and even force the hardware to interleave the memory address space so that consecutive memory pages are placed on consecutive nodes rather than the usual "first 1/4 of memory goes on node 0, ..." approach for a NUMA box.

So far, the usual intel boxes, such as the 70 node cluster where I used one of the nodes for the last couple of CCT events, are all pure SMP with a single shared memory rather than having the memory distributed across the individual processors.