Performance note:  I realized later that changing the "counts" map to a "vector<pair<uint64_t, uint64_t> >" results in a big speed boost (30%) for the gem5.fast build.  I had originally used the map just as an easy way to sort and store.  Pushing to the back of a vector and sorting once is a much better method.  It's a 3 line change.

src/cpu/BaseCPU.py (Diff revision 1)

I'm new to gem5 and curious about this patch. My question is what the function of this instruction is? Is it related to simulator events?  

Show all issues

Ship It!

src/cpu/simple/atomic.cc (Diff revision 1)

Should here be:
"info.count++;" ?
Or info.count will become the BBV number timed by instruction number of this basic block.

Show all issues

1. Mitch Hayenga April 2, 2013, 6:24 a.m. (April 2, 2013, 6:24 a.m.)

   This was intentional.  I can't find where in the Simpoint papers/tech reports where it says this right now, but it's also the way Valgrind does simpoint analysis.
   
   http://valgrind.org/docs/manual/bbv-manual.html#bbv-manual.fileformat
   "The frequency count is multiplied by the number of instructions that are in the basic block, in order to weigh the count so that instructions in small basic blocks aren't counted as more important than instructions in large basic blocks."

2. Meng Wang April 2, 2013, 10:32 a.m. (April 2, 2013, 10:32 a.m.)

   Thanks. This explain makes sense. I understand now.