Linux下 Float 与 Int 类型计算性能对比

代码

#include <stdio.h>
#include <stdlib.h>
#include <time.h>

#include <iostream>

static constexpr int64_t kCalculateNum = 100 * 1000 * 1000L;

static int64_t testFloat()  // return void* in POSIX
{
  struct timespec start_ts, end_ts;
  clock_gettime(CLOCK_MONOTONIC, &start_ts);
  volatile float d = 999999.f;
  for (int n = 0; n < kCalculateNum; ++n) {
    d = d * 0.7f;
  }
  clock_gettime(CLOCK_MONOTONIC, &end_ts);
  auto diff_ns = (end_ts.tv_sec - start_ts.tv_sec) * 1e9 +
                 (end_ts.tv_nsec - start_ts.tv_nsec);
  return diff_ns / 1000;
}

static int64_t testInt()  // return void* in POSIX
{
  struct timespec start_ts, end_ts;
  clock_gettime(CLOCK_MONOTONIC, &start_ts);
  volatile int32_t d = 999999;
  for (int n = 0; n < kCalculateNum; ++n) {
    d = d * 7;
  }
  clock_gettime(CLOCK_MONOTONIC, &end_ts);
  auto diff_ns = (end_ts.tv_sec - start_ts.tv_sec) * 1e9 +
                 (end_ts.tv_nsec - start_ts.tv_nsec);
  return diff_ns / 1000;
}

int main(void) {
  auto float_duration_us = testFloat();
  auto int_duration_us = testInt();
  std::cout << "Float calculate times: " << kCalculateNum
            << " took time: " << float_duration_us << "us.\n";
  std::cout << "Int calculate times: " << kCalculateNum
            << " took time: " << int_duration_us << "us.\n";
  std::cout << "Float / Int radio: "
            << (float_duration_us / int_duration_us * 1.0f) << std::endl;
  return 0;
}

执行结果

编译

g++ test_float.cpp -g3 -o test_float.exe

结果

多次执行，结果相当，计算 float 类型耗时大约是 int 整形的 62倍，即性能是 int 的 1/62 左右。

gobert@DESKTOP-EJV14VH:~/develop/debug$ g++ test_float.cpp -g3 -o test_float.exe
gobert@DESKTOP-EJV14VH:~/develop/debug$ ./test_float.exe
Float calculate times: 100000000 took time: 2537060us.
Int calculate times: 100000000 took time: 40442us.
Float / Int radio: 62

以上仅供参考！！

（采用 CC BY-NC-SA 4.0 许可协议进行授权）

本文标题：《 float 和 int 在 Linux 下的计算性能对比》

本文链接：https://gbcpp.github.io/float-performance.html

本文最后一次更新为天前，文章中的某些内容可能已过时！

Linux下 Float 与 Int 类型计算性能对比

代码

执行结果

目录