For the record, there was a time when we tried to intercept the ioctl() that checked whether you were connected to a tty or not and always indicate a tty without asking the host. This gave you repeatable results. I'm not sure what happened to that though... could be it's still there in the Tru64 code but never got ported to Linux. It's also possible (though less likely) that it used to work for Linux but a newer version of libc uses a different syscall to check.
Steve On Thu, Apr 17, 2008 at 6:56 PM, Steve Reinhardt <[EMAIL PROTECTED]> wrote: > Yea, this is a long-standing issue for any system that does syscall > emulation. At runtime, the I/O library checks the output device to see if > it's line-buffered or block buffered (i.e., a tty or not), and that affects > whether or not it flushes the output buffer on a newline. Since SE mode > just passes the syscalls through to the underlying host, the target's I/O > library can tell whether the output is being sent to the tty or to a pipe, > and the code path will change slightly depending on that. > > Ideally for real simulations you'd always redirect the target's I/O to a > file and then you'll get consistent results. > > Steve > > > On Thu, Apr 17, 2008 at 6:41 PM, Nicolas Zea <[EMAIL PROTECTED]> > wrote: > > > I've noticed that the final trace time in a normal run: > > > > build/ALPHA_SE/m5.opt configs/example/se.py -c > > tests/test-progs/hello/bin/alpha/linux/hello > > > > and a run that pipes the results to a file > > > > build/ALPHA_SE/m5.opt configs/example/se.py -c > > tests/test-progs/hello/bin/alpha/linux/hello | tee hello_output > > > > have different cycle times. In the case of an unmodified m5_b5, they are > > 3208500 and 2871500 respectively. Any idea what may cause this difference? > > (or does this not happen for you, in which case something very odd is going > > on on my end) > > > > -Nick_______________________________________________ > > m5-users mailing list > > [email protected] > > http://m5sim.org/cgi-bin/mailman/listinfo/m5-users > > > >
_______________________________________________ m5-users mailing list [email protected] http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
