Perf 로 프로파일링 후 코드 최적화하기

설치

case 2 (for system 2)
Wait, that is not the end. If returning result of $uname -r is like this 4.4.77, the previous way may not work. If you do $sudo apt-get install linux-tools-4.4.77 but cannot get the tool installed. In this case, download the required linux kernel version (our case is v4.4.77) and go to ‘linux->tools->perf’ to install it.

$git clone git://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git

After download finished,

$cd linux-stable/

$git checkout v4.4.77 (you can also check your git status git status)

$cd tools/perf/

$make (If error came and asked for installation of ‘flex’ and ‘bison’, $sudo apt-get install flex and $sudo apt-get install bison)

$sudo cp perf /usr/bin

After doing this, you can try $perf --version and hope it works for you now.

After make, you can also try make install and then check the perf --version.

레퍼런스 : https://xiaoyanzhuo.github.io/2019/01/18/Perf-Tool.html

Usage

perf stat -p 49298
perf record -e cpu-clock -p 49298

 Performance counter stats for process id '49298':

         26,955.53 msec task-clock:u              #    1.037 CPUs utilized
                 0      context-switches:u        #    0.000 K/sec
                 0      cpu-migrations:u          #    0.000 K/sec
           179,775      page-faults:u             # 6669.449 M/sec
    46,899,747,932      cycles:u                  # 1739927.580 GHz
    59,990,994,484      instructions:u            #    1.28  insn per cycle
    13,567,779,293      branches:u                # 503349259.618 M/sec
        72,828,426      branch-misses:u           #    0.54% of all branches

      26.003937135 seconds time elapsed

레퍼런스 : https://varblog.org/blog/2018/05/24/profiling-and-optimizing-machine-learning-model-training-with-pytorch/

작성 중