function_call_overhead.c

// gcc -O2 function_call_overhead.c

#include <stdio.h>
#include <inttypes.h>

#define concat_lower_uint32(high, low) (high = (high << 32) | low);

int __attribute__ ((noinline)) foo() {
	return 42;
}

int main() {
	uint64_t start0 = 0, start1 = 0, end0 = 0, end1 = 0;

	// rdtsc - read time-stamp counter, increases by one every clock cycle
	__asm__ __volatile__("rdtsc" : "=a"(start0), "=d"(start1));

	foo();

	__asm__ __volatile__("rdtsc" : "=a"(end0), "=d"(end1));

	concat_lower_uint32(start1, start0);
	concat_lower_uint32(end1, end0);

	printf("%" PRIu64 " cycles\n", end1 - start1);
	return 0;
}

method_call_overhead.rb

# gem install RubyInline
require "inline"

class Test
  def foo
    42
  end

  inline do |builder|
    builder.include "<stdint.h>"

    builder.prefix <<-C
      #define concat_lower_uint32(high, low) (high = (high << 32) | low);
      const int iterations = 100;
    C

    builder.c <<-C
      VALUE measure_cycles() {
        uint64_t start0 = 0, start1 = 0, end0 = 0, end1 = 0;
        ID foo = rb_intern("foo");
        VALUE results = rb_ary_new_capa(iterations);

        for (int i = 0; i < iterations; i++) {
          // rdtsc - read time-stamp counter, increases by one every clock cycle
          __asm__ __volatile__("rdtsc" : "=a"(start0), "=d"(start1));

          rb_funcall(self, foo, 0);

          __asm__ __volatile__("rdtsc" : "=a"(end0), "=d"(end1));

          concat_lower_uint32(start1, start0);
          concat_lower_uint32(end1, end0);

          rb_ary_push(results, LONG2NUM((long)(end1 - start1)));
        }

        return results;
      }
    C
  end
end

p Test.new.measure_cycles