profile

perf stat -B -e cache-references,cache-misses,cycles,instructions,branches,faults,migrations,branch-misses -g --delay 250 ./bench/run_func data/twitter.json.xz dumps

before:
        96,346,033      cache-references
         2,396,956      cache-misses              #    2.488 % of all cache refs
     2,819,207,654      cycles
     6,903,811,887      instructions              #    2.45  insn per cycle
     1,527,040,323      branches
           175,425      faults
                 0      migrations
         9,638,830      branch-misses             #    0.63% of all branches

       0.995690355 seconds time elapsed

       0.949150000 seconds user
       0.978124000 seconds sys

after:
        58,807,500      cache-references
         2,276,541      cache-misses              #    3.871 % of all cache refs
     1,535,556,207      cycles
     5,080,111,489      instructions              #    3.31  insn per cycle
     1,088,522,837      branches
               220      faults
                 0      migrations
         6,015,137      branch-misses             #    0.55% of all branches

       0.656489187 seconds time elapsed

       0.655558000 seconds user
       0.133799000 seconds sys