Ben Hutchings wrote:
David M Garza wrote:
gdb's stack trace says:
#0 0x400f8e1 in boost::detail::atomic_increment (pw=0x40006cc4) at sp_counted_base_gcc_ia64.hpp:38
Since gdb is printing the pointers as 32-bit values, and pw is not 64-bit-aligned, it appears that your compiler is using the ILP32 convention (32-bit int, long and pointer types). I assumed the LP64 convention (64-bit long and pointer types, 32-bit int) when writing the assembly-language code in sp_counted_base_gcc_ia64.hpp because I wasn't aware that other conventions were in use on IA64. If the C++ code is compiled with an ILP32 convention then the assembly-language code will be completely broken. I may be able to adjust the code to work to detect the model at compile time and use 32-bit memory accesses if appropriate.
Another option is to just switch to 'int' and 32-bit accesses (as we did for PPC64 IIRC.)