Le 24/11/2010 07:21, Christopher Samuel a écrit : > This is the error on x86-64: > > ==13648== 6 errors in context 1 of 1: > ==13648== Conditional jump or move depends on uninitialised value(s) > ==13648== at 0x40B678: hwloc_alloc (bind.c:451)
Just fixed it. > ==6259== 1 errors in context 1 of 18: > ==6259== Syscall param get_mempolicy(nodemask) points to unaddressable > byte(s) > ==6259== at 0xFBE6FE0: syscall (in /lib/power6x/libc-2.4.so) > ==6259== by 0xFC9BDCB: get_mempolicy (in /usr/lib/libnuma.so.1) > ==6259== by 0x10016A7B: hwloc_linux_get_thisthread_membind > (topology-linux.c:1064) This one looks strange to me, this nodemask was allocated and checked just above get_mempolicy. Could be a valgrind problem. > ==6259== 1 errors in context 2 of 18: > ==6259== Invalid write of size 1 > ==6259== at 0xFC005F4: __vsnprintf_chk@@GLIBC_2.4 (in > /lib/power6x/libc-2.4.so) > ==6259== by 0xFC004F3: __snprintf_chk@@GLIBC_2.4 (in > /lib/power6x/libc-2.4.so) > ==6259== by 0xFC9B4D3: numa_init (in /usr/lib/libnuma.so.1) Not from us. > ==6259== 12 errors in context 7 of 18: > ==6259== Invalid write of size 1 > ==6259== at 0xFFBA678: memset (mc_replace_strmem.c:626) > ==6259== by 0x10016867: hwloc_linux_set_thisthread_membind > (topology-linux.c:1001) Memset just got fixed. The others are duplicates of these. Brice
