Use kwset in pickaxe
authorFredrik Kuivinen <frekui@gmail.com>
Sat, 20 Aug 2011 22:41:57 +0000 (00:41 +0200)
committerJunio C Hamano <gitster@pobox.com>
Sun, 21 Aug 2011 05:33:57 +0000 (22:33 -0700)
commitb95c5ada990ee2afb83af9be5b82d2804618a59d
treedaff546c4f0615a48fe16ab7afecda626d17152c
parentfca65d4b260912448c4f103ab3321fa2b82bf907
Use kwset in pickaxe

Benchmarks in the hot cache case:

before:
$ perf stat --repeat=5 git log -Sqwerty

Performance counter stats for 'git log -Sqwerty' (5 runs):

       47,092,744 cache-misses             #      2.825 M/sec   ( +-   1.607% )
      123,368,389 cache-references         #      7.400 M/sec   ( +-   0.812% )
      330,040,998 branch-misses            #      3.134 %       ( +-   0.257% )
   10,530,896,750 branches                 #    631.663 M/sec   ( +-   0.121% )
   62,037,201,030 instructions             #      1.399 IPC     ( +-   0.142% )
   44,331,294,321 cycles                   #   2659.073 M/sec   ( +-   0.326% )
           96,794 page-faults              #      0.006 M/sec   ( +-  11.952% )
               25 CPU-migrations           #      0.000 M/sec   ( +-  25.266% )
            1,424 context-switches         #      0.000 M/sec   ( +-   0.540% )
     16671.708650 task-clock-msecs         #      0.997 CPUs    ( +-   0.343% )

      16.728692052  seconds time elapsed   ( +-   0.344% )

after:
$ perf stat --repeat=5 git log -Sqwerty

Performance counter stats for 'git log -Sqwerty' (5 runs):

       51,385,522 cache-misses             #      4.619 M/sec   ( +-   0.565% )
      129,177,880 cache-references         #     11.611 M/sec   ( +-   0.219% )
      319,222,775 branch-misses            #      6.946 %       ( +-   0.134% )
    4,595,913,233 branches                 #    413.086 M/sec   ( +-   0.112% )
   31,395,042,533 instructions             #      1.062 IPC     ( +-   0.129% )
   29,558,348,598 cycles                   #   2656.740 M/sec   ( +-   0.204% )
           93,224 page-faults              #      0.008 M/sec   ( +-   4.487% )
               19 CPU-migrations           #      0.000 M/sec   ( +-  10.425% )
              950 context-switches         #      0.000 M/sec   ( +-   0.360% )
     11125.796039 task-clock-msecs         #      0.997 CPUs    ( +-   0.239% )

      11.164216599  seconds time elapsed   ( +-   0.240% )

So the kwset code is about 33% faster.

Signed-off-by: Fredrik Kuivinen <frekui@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Makefile
diffcore-pickaxe.c