http://software.intel.com/en-us/articles/visualizing-parallel-speedup-with-cilkview en-us ). And ultimately what you need to get out of the application is the DAG plus reading the cycle counter at the start end of serial regions between cilk operations (spawn sync), right? There would certainly be overhead to logging all that, but it seems doable just by changing mode within libcilkrts. So what am I missing? Is the problem just the extra complexity -- is there anything here that would require modifying the C compiler (to change the generated code for spawns)? Thanks, -Ryan ]]> http://software.intel.com/en-us/articles/visualizing-parallel-speedup-with-cilkview/#comment-72107 Tue, 27 Mar 2012 06:50:24 -0700 http://software.intel.com/en-us/articles/visualizing-parallel-speedup-with-cilkview/#comment-72107