------- Comment #7 from pluto at agmk dot net 2010-04-18 19:04 ------- (In reply to comment #5) > (In reply to comment #4) > > during analysis the boost::thread::start_thread() function which causes > > invalid writes detected by valgrind i've noticed that gcc-4.5 generates > > bigger and more complex code of this function with few more lock'ed opcodes. > > afaics gcc-4.5 produces some mess for boost::shared_ptr. > > could please someone look at this? it may be a missed optimization or other > > bug. > > It seems to be a completely different implementation. > > This bug lacks a testcase. >
i've atatched the first testcase for this bug. gcc43 and gcc44/45 generate completely different code for the boost::thread::start_thread function. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39979