How many CPU ticks a malloc() costs?

cost-of-malloc.md

Cost of malloc

How many CPU ticks it costs to allocate a piece of memory in heap via malloc()?

TL;DR Around 200 ticks.

Here is the code:

#include <iostream>
#include <chrono>
using namespace std::chrono;

class Foo {
  int x;
};

int sum(int total) {
  auto start = high_resolution_clock::now();
  int sum = 0;
  for (int i = 0; i < total; i++) {
    sum += 1;
    sum += 1;
    sum += 1;
    sum += 1;
    sum += 1;
    sum += 1;
    sum += 1;
    sum += 1;
    sum += 1;
    sum += 1;
  }
  auto stop = high_resolution_clock::now();
  auto duration = duration_cast<microseconds>(stop - start);
  return duration.count();
}

int heap(int total) {
  auto start = high_resolution_clock::now();
  for (int i = 0; i < total; i++) {
    new Foo();
    new Foo();
    new Foo();
    new Foo();
    new Foo();
    new Foo();
    new Foo();
    new Foo();
    new Foo();
    new Foo();
  }
  auto stop = high_resolution_clock::now();
  auto duration = duration_cast<microseconds>(stop - start);
  return duration.count();
}

int main() {
  int total = 100000000;
  int m1 = sum(total);
  int m2 = heap(total);
  std::cout << (m2 / m1) << std::endl;
  return 0;
}

Compile it like this and run:

$ clang++ -O0 a.cpp && ./a.out

I'm getting numbers between 60 and 80.

Thus, a single malloc costs about 180-240 CPU ticks, because sum += 1 is exactly three Assembly instructions. You can check it by compiling the code above into Assembly:

$ clang++ -S -mllvm --x86-asm-syntax=intel -O0 a.cpp

The file a.s will contain Assembly, where sum += 1 is represented with the following three:

ldur	w8, [x29, #-20]
add	w8, w8, #1
stur	w8, [x29, #-20]

By the way, it is MacBook Pro M2 with macOS Ventura 13.3.1:

$ clang -v
Homebrew clang version 15.0.7
Target: arm64-apple-darwin22.4.0
Thread model: posix
InstalledDir: /opt/homebrew/opt/llvm/bin