!LANG= cpupower frequency-info

analyzing CPU 7:
  driver: intel_pstate
  CPUs which run at the same hardware frequency: 7
  CPUs which need to have their frequency coordinated by software: 7
  maximum transition latency:  Cannot determine or is not supported.
  hardware limits: 400 MHz - 3.90 GHz
  available cpufreq governors: performance powersave
  current policy: frequency should be within 400 MHz and 3.90 GHz.
                  The governor "performance" may decide which speed to use
                  within this range.
  current CPU frequency: Unable to call hardware
  current CPU frequency: 3.64 GHz (asserted by call to kernel)
  boost state support:
    Supported: yes
    Active: yes

!rm -Rf tmp
!mkdir -p tmp

%%writefile tmp/pipeline-perf.c

#include <stdio.h>
#include "timing.h"


int main()
{
  int result = 0;

  {
    int a = 0, b = 0;

    timestamp_type t1;
    get_timestamp(&t1);

    for (int ntrips = 0; ntrips < 1000; ++ntrips)
      for (int i = 0; i< 400*1000; ++i)
      {
        a += i;
        a += i;
      }

    timestamp_type t2;
    get_timestamp(&t2);

    printf("a, a: elapsed time %g s\n",
        timestamp_diff_in_seconds(t1, t2));
    result += a+b;
  }

  return result;
}

Writing tmp/pipeline-perf.c

!cd tmp; gcc -std=gnu99 -lrt -I.. -opipeline-perf pipeline-perf.c
!tmp/pipeline-perf

a, a: elapsed time 1.14691 s

# !objdump --disassemble tmp/pipeline-perf

!cd tmp; gcc -std=gnu99 -lrt -I.. -opipeline-perf pipeline-perf.c
!tmp/pipeline-perf

a, a: elapsed time 1.13467 s

%%writefile tmp/pipeline-perf.c

#include <stdio.h>
#include "timing.h"


int main()
{
  int result = 0;

  {
    int a = 0, b = 0;

    timestamp_type t1;
    get_timestamp(&t1);

    for (int ntrips = 0; ntrips < 1000; ++ntrips)
      for (int i = 0; i< 1000*1000; ++i)
      {
        a += i;
        a += i;
      }

    timestamp_type t2;
    get_timestamp(&t2);

    printf("a, a: elapsed time %g s\n",
        timestamp_diff_in_seconds(t1, t2));
    result += a+b;
  }

  {
    int a = 0, b = 0;

    timestamp_type t1;
    get_timestamp(&t1);

    for (int ntrips = 0; ntrips < 1000; ++ntrips)
      for (int i = 0; i< 1000*1000; ++i)
      {
        a += i;
        b += i;
      }

    timestamp_type t2;
    get_timestamp(&t2);

    printf("a, b: elapsed time %g s\n",
        timestamp_diff_in_seconds(t1, t2));
    result += a+b;
  }

  {
    int a = 0, b = 0;

    timestamp_type t1;
    get_timestamp(&t1);

    for (int ntrips = 0; ntrips < 1000; ++ntrips)
      for (int i = 0; i< 250*1000; ++i)
      {
        a += i;
        a += i;
 
        a += i;
        a += i;
 
        a += i;
        a += i;
 
        a += i;
        a += i;
 
      }

    timestamp_type t2;
    get_timestamp(&t2);

    printf("a, a unrolled: elapsed time %g s\n",
        timestamp_diff_in_seconds(t1, t2));
    result += a+b;
  }
  {
    int a = 0, b = 0;

    timestamp_type t1;
    get_timestamp(&t1);

    for (int ntrips = 0; ntrips < 1000; ++ntrips)
      for (int i = 0; i< 250*1000; ++i)
      {
        a += i;
        a += i;
        a += i;
        a += i;
          
        b += i;
        b += i;
        b += i;
        b += i;
      }

    timestamp_type t2;
    get_timestamp(&t2);

    printf("aa, bb unrolled: elapsed time %g s\n",
        timestamp_diff_in_seconds(t1, t2));
    result += a+b;
  }

  {
    int a = 0, b = 0;

    timestamp_type t1;
    get_timestamp(&t1);

    for (int ntrips = 0; ntrips < 1000; ++ntrips)
      for (int i = 0; i< 250*1000; ++i)
      {
        a += i;
        b += i;

        a += i;
        b += i;

        a += i;
        b += i;

        a += i;
        b += i;
      }

    timestamp_type t2;
    get_timestamp(&t2);

    printf("a, b unrolled: elapsed time %g s\n",
        timestamp_diff_in_seconds(t1, t2));
    result += a+b;
  }

  return result;
}

Overwriting tmp/pipeline-perf.c

!cd tmp; gcc -std=gnu99 -lrt -I.. -opipeline-perf pipeline-perf.c 
!tmp/pipeline-perf

a, a: elapsed time 2.85316 s
a, b: elapsed time 1.53108 s
a, a unrolled: elapsed time 2.862 s
aa, bb unrolled: elapsed time 1.46449 s
a, b unrolled: elapsed time 1.49071 s

Pipeline Performance Mysteries¶