ceph_perf_local

在ceph 源码ceph-master\ceph-master\src\test\CMakeLists.txt 中定义了一个ceph_perf_local的可执行文件，用于测试
不同平台对ceph性能的影响
#ceph_perf_local
add_executable(ceph_perf_local 
  perf_local.cc
  perf_helper.cc)
if(HAVE_SSE)
  set(PERF_LOCAL_FLAGS ${SSE3_FLAGS})
endif(HAVE_SSE)
if(HAVE_NEON)
  set(PERF_LOCAL_FLAGS ${ARM_NEON_FLAGS})
endif(HAVE_NEON)
if(PERF_LOCAL_FLAGS)
  set_target_properties(ceph_perf_local PROPERTIES COMPILE_FLAGS
    ${PERF_LOCAL_FLAGS})
endif()
target_link_libraries(ceph_perf_local os global ${UNITTEST_LIBS})
从makefile中可以看出ceph_perf_local的源文件有两个，分别是perf_local.cc/perf_helper.cc.
其编译好的执行bin的name为ceph_perf_local，从makefile中同样可以看出可以使用x86的sse或者arm的neon 
指令来加速
这个tool的入口函数如下：
int main(int argc, char *argv[])
{
  vector<const char*> args;
  argv_to_vec(argc, (const char **)argv, args);

  auto cct = global_init(NULL, args, CEPH_ENTITY_TYPE_CLIENT,
			 CODE_ENVIRONMENT_UTILITY, 0);
  common_init_finish(g_ceph_context);
  Cycles::init();

  bind_thread_to_cpu(3);
  #测试命令可以带一个参数，如果不带参数的话，则测试所有的item项
  if (argc == 1) {
    // No test names specified; run all tests.
    for (size_t i = 0; i < sizeof(tests)/sizeof(TestInfo); ++i) {
      run_test(tests[i]);
    }
  } else {
    // Run only the tests that were specified on the command line.
    for (int i = 1; i < argc; i++) {
      bool found_test = false;
      for (size_t j = 0; j < sizeof(tests)/sizeof(TestInfo); ++j) {
        if (strcmp(argv[i], tests[j].name) == 0) {
          found_test = true;
          run_test(tests[j]);
          break;
        }
      }
      if (!found_test) {
        int width = printf("%-24s ??", argv[i]);
        printf("%*s No such test\n", 32-width, "");
      }
    }
  }
}
从main函数中可以看出测试的item项在tests 这个数组中
TestInfo tests[] = {
  {"atomic_int_cmp", atomic_int_cmp,
    "atomic_t::compare_and_swap"},
  {"atomic_int_inc", atomic_int_inc,
    "atomic_t::inc"},}

}
这里以atomic_int_inc的测试为例
double atomic_int_inc()
{
  int count = 1000000;
  #定义一个原子变量
  std::atomic<int64_t> value = { 11 };
  #对原子变量操作前获取时间
  uint64_t start = Cycles::rdtsc();
  #对这个原子变量自加1000000
  for (int i = 0; i < count; i++) {
    value++;
  }
  #原子变量自加完成后，读取时间
  uint64_t stop = Cycles::rdtsc();
  // printf("Final value: %d\n", value.load());
  #计算本次原子变量自加占用的时间
  return Cycles::to_seconds(stop - start)/count;
}