% configure -GNU % make ; make install
% configure -pdt=/usr/packages/pdt-xxx -mpiinc=/usr/packages/mpi-xxx/include -mpilib=/usr/packages/mpi-xxx/lib % make install
% configure.pl +mpich +no_fortran +tau goThis step creates make.inc
% setenv TAU_MAKEFILE /usr/packages/tau-xxx/i386_linux/lib/Makefile.tau % setenv MPI_ROOT /usr/packages/mpi-xxx/ % make
The above image shows the profile of the acceptance testing benchmark when it executes on four processors (top left window). Racy, TAU's profile browser, shows that a significant portion of time is spent in the MPI_Sendrecv() routine (bottom left). On node 0, context 0, thread 0, the histogram of the exclusive time spent in all routines is shown (right window).
By clicking the middle mouse button over the "mean" tag in the main racy window, we can see the text window that shows detailed information for the entire run averaged over the four processors. This information can be sorted in several ways.
The function legend shows all routines for a given execution.
The above image shows the breakdown of key routines across all nodes of execution. We see the relative contribution of MPI_Recv(), MPI_Waitany(), and upsp_gs_garther_scatter() routines.
For more information on TAU, please visit the TAU webpage.