rutgers-apl / taskprof Goto Github PK

We tried to reproduce the results using the benchmark provided, and got similar results as described in the paper except for Delaunay Triangulation. The annotated regions show almost no parallelism. The output is as follows:

region_info.csv:

4228347,IO.h,184
4228158,IO.h,176
4227901,IO.h,159
4227424,IO.h,184
4227218,IO.h,176
4226941,IO.h,159
4254744,delaunay.C,275
4254603,delaunay.C,263
4206013,IO.h,89 

region_all.csv:

Regions Optimization factor, Whole program parallelism
50,1.75187
100,1.75639
200,1.75866 
400,1.7598 

build.sh detects the "cpu-cycles" event exists or not and decides whether to compile or not. Yet the name of PMU event seems to be changed to "cpu-cycles OR cpu/cpu-cycles" on a Xeon E5 machine which runs Ubuntu 16.04.