gistfile1.txt

$ perf stat ./python.default -m performance run -b mako -r

 Performance counter stats for './python.default -m performance run -b mako -r':

      19307.778115      task-clock (msec)         #    0.585 CPUs utilized
               632      context-switches          #    0.033 K/sec
               191      cpu-migrations            #    0.010 K/sec
            945163      page-faults               #    0.049 M/sec
      102219040792      cycles                    #    5.294 GHz
       27777883730      stalled-cycles-frontend   #   27.17% frontend cycles idle
       15855030490      stalled-cycles-backend    #   15.51% backend cycles idle
      193523548833      instructions              #    1.89  insn per cycle
                                                  #    0.14  stalled cycles per insn
       40264135620      branches                  # 2085.384 M/sec
         347597640      branch-misses             #    0.86% of all branches

      33.026918905 seconds time elapsed

$ perf stat ./python.patched -m performance run -b mako -r

 Performance counter stats for './python.patched -m performance run -b mako -r':

      15079.263993      task-clock (msec)         #    0.423 CPUs utilized
               639      context-switches          #    0.042 K/sec
               194      cpu-migrations            #    0.013 K/sec
            944512      page-faults               #    0.063 M/sec
      110264367110      cycles                    #    7.312 GHz
       36176583497      stalled-cycles-frontend   #   32.81% frontend cycles idle
       23302398028      stalled-cycles-backend    #   21.13% backend cycles idle
      189321878449      instructions              #    1.72  insn per cycle
                                                  #    0.19  stalled cycles per insn
       39432308748      branches                  # 2615.002 M/sec
         352216634      branch-misses             #    0.89% of all branches

      35.629047349 seconds time elapsed

NOTE: Number of instructions are decreased slightly.  The problem is insn / cycle.


 Performance counter stats for './python.default -m performance run -b mako -r':

         666761928      L1-dcache-load-misses
         613456531      L1-icache-load-misses
          49339774      cache-misses
         309471477      L1-dcache-store-misses

      32.978880977 seconds time elapsed

 Performance counter stats for './python.patched -m performance run -b mako -r':

         670959723      L1-dcache-load-misses
         697039948      L1-icache-load-misses
          49139047      cache-misses
         308375381      L1-dcache-store-misses

      35.501321551 seconds time elapsed

OK, L1-icache-load-misses is problem.