Instrument multiple sections of code

You can also instrument multiple sections of code.

Instrumenting multiple code blocks in C

The next scenario is to instrument a single section of code in C.

The API is slightly different, but the concept is the same.

For multiple code segments the first two steps and cleanup are the same but the start and stop functions are slightly different because they include markers to indicate which segment you are profiling.

Here are the steps for multiple segments:

• Include 2 header files (same)
• Initialize the plugin by calling pmuv3_bundle_init() with the bundle number as an argument (same)
• Use get_next_index() to obtain the next segment number, the numbers count up from 0
• Start counting by calling the function get_start_count() and pass the segment number
• Call the function get_end_count() and pass the segment number to stop counting for that segment
• Write the collected data to a CSV file by calling process_data() with the same bundle number
• Clean up with shutdown_resources()

You can repeat for additional segments by getting the next segment number and using the start and stop functions again.

The example below collects the initialize_vectors() function and the calculate_result() functions separately instead of collecting the data for both of them as in the previous example.

Use a text editor to create a file test2.c in the test directory with the contents below:

    

        
        
#include <stdio.h>
#include <stdint.h>
#include <math.h>

#include "pmuv3_plugin_bundle.h"
#include "processing.h"

#define VECTOR_SIZE 10000

void initialize_vectors(double vector_a[], double vector_b[], int size) {
    for (int i = 0; i < size; i++) {
        vector_a[i] = sin(i);
        vector_b[i] = cos(i);
    }
}

void calculate_result(double result[], double vector_a[], double vector_b[], int size) {
    for (int i = 0; i < size; i++) {
        result[i] = vector_a[i] + vector_b[i];
    }
}

int main(int argc, char **argv) {
    uint64_t section1, section2;
    int bundle_number;
    struct CountData counter_data;

    /* Bundle number is passed as an argument */
    if (argc != 2) {
        fprintf(stderr, "Usage: %s <bundle number>\n", argv[0]);
        return EXIT_FAILURE;
    }
    bundle_number = atoi(argv[1]);

    /* Initialize the bundle of events */
    if (pmuv3_bundle_init(bundle_number) != 0) {
        fprintf(stderr, "Failed to initialize PMU bundle\n");
        return EXIT_FAILURE;
    }


    /* Declare vectors */
    double vector_a[VECTOR_SIZE], vector_b[VECTOR_SIZE], result[VECTOR_SIZE];

    /* Initialize vectors */
    section1 = get_next_index();
    printf("section 1: %ld\n", section1);
    get_start_count(&counter_data, "SECTION_1" , section1);
    initialize_vectors(vector_a, vector_b, VECTOR_SIZE);
    get_end_count(&counter_data, "SECTION_1", section1);

    /* Calculate vectors */
    section2 = get_next_index();
    printf("section 2: %ld\n", section2);
    get_start_count(&counter_data, "SECTION_2" , section2);
    calculate_result(result, vector_a, vector_b, VECTOR_SIZE);
    get_end_count(&counter_data, "SECTION_2", section2);


    /* Write CSV file and clean up */
    process_data(bundle_number);
    shutdown_resources();

    return EXIT_SUCCESS;
}

    

Build the application:

    

        
        
gcc -I ../linux/tools/lib/perf/include  -I ../PMUv3_plugin/ test2.c -o test2 -L ../PMUv3_plugin/  -lpmuv3_plugin_bundle -lperf -lapi -lm

    

Run the application and pass the bundle number of 3 (to capture the stall information):

    

        
        
sudo ./test2 3

    

The output now contains data for 2 sections and prints:

    

        
        - running pmuv3_plugin_bundle.c...OK
section 1: 0
section 2: 1
End is 4295746437, Start is 4295069884, diff is 676553
End is 4294968118, Start is 4294967724, diff is 394
End is 4295307089, Start is 4294993064, diff is 314025
End is 4295295692, Start is 4294986720, diff is 308972
End is 4294967867, Start is 4294967499, diff is 368
End is 4295245791, Start is 4294976253, diff is 269538
End is 4295012643, Start is 4294968164, diff is 44479
End is 4295863211, Start is 4295753132, diff is 110079
End is 4294968215, Start is 4294968196, diff is 19
End is 4295331822, Start is 4295308901, diff is 22921
End is 4295319761, Start is 4295297198, diff is 22563
End is 4294967953, Start is 4294967918, diff is 35
End is 4295268632, Start is 4295247245, diff is 21387
End is 4295014550, Start is 4295012949, diff is 1601

        
    

The results are captured in the file bundle3.csv.

Display the text file to see the contents:

    

        
        
cat bundle3.csv

    

The data shows the metrics on the first line and the values for section 1 on the second line and the values for section 2 on the third line.

    

        
        CONTEXT,CPU_CYCLES,BR_MIS_PRED,BR_PRED,BR_RETIRED,BR_MIS_PRED_RETIRED,BR_IMMED_SPEC,BR_INDIRECT_SPEC
SECTION_1,679996,517,313451,307426,485,268808,44711
SECTION_2,110975,33,24423,24000,41,22507,1925

        
    

You can use this methodology to instrument multiple sections of code and generate the data for all bundles by modifying the run.sh file from the single section instrumentation. All you need to do is change the command from test1 to test2 and invoke the run.sh script again.