为什么 C++ 什么都不做时比 C 慢

Question

C++ is almost 4 times slower than C at doing nothing (at least on my machine). When the following file is compiled by g++, it is 4 times slower than with gcc:

int main() {}

With time sh -c 'for i in $(seq 0 1000); do./a.out; done' time sh -c 'for i in $(seq 0 1000); do./a.out; done' time sh -c 'for i in $(seq 0 1000); do./a.out; done' , I get 0.515s for the C version, and 2s for the C++ one. Why is that?

There isn't a single thing done in this program, and the disassembled program is basically the same in both versions, so why is the C++ version roughly 4 times slower?

My guess is that the C++ library takes a longer time to load, but the assembly is so similar that I can't think of a reason for C++ to be so much slower.

Edit: Well, it appears that I only partially disassembled the program ( objdump -d instead of objdump -D ), resulting in me not seeing the library being loaded. It is pretty obvious now seeing the disassembled output that the C++ version just takes longer to load its libraries. What I first meant by "slightly differing" was that the addresses used in mov or call diverged without calling other functions (if I understand correctly the meaning of objdump 's output).

So the main question is solved, but 2 seconds is huge for a program that does nothing. I'm on a x86_64 processor if that helps, and I built the executable gcc empty.c and g++ empty.cpp . I think the issue really is me using a HDD, thus taking more time to load libstd++ . I also built with the -static flag, and am now getting very similar results (0.246s and 0.241s). The C binary is also exactly the same as the C++ one (if that's not a diff bug).

Answer 1

C++ standard library initial startup is heavier than the C runtime.

C++ ABI needs to initialize some structures for basic language features. Mutex locks for thread-safe initialization of data , exeptions , thread local storage , RTTI , etc.

This is why an empty executable created with a C++ compiler will start slower than an empty C program. Likewise, an empty program created with Assembly language will start much faster then a similar program created with C with a full-featured C runtime (CRT, GNU LIBC, etc).

Answer 2

In fact, you compare nothing at all, because your program does nothing at all. What you got is not the program execution time , which in your case is null, but the process execution time , ie load+run+terminate, which is longer if your program needs more resources.

Some numbers could lead some light on this:

say we have emptyc and emptycpp , respectively compiled with gcc-10.3 and g++-10.3

On my system:

• emptyc :

  • file size: 13232
  • elf dynamic section: 17 entries (1 shared lib)
  • truss syscall report: 48 syscalls including 6 open and 11 mmap

• emptycpp

  • file size: 13312
  • elf dynamic section: 20 entries (4 shared libs)
  • truss syscall report: 88 syscalls including 15 open and 29 mmap

Conclusion :

A program compiled with g++ needs more resources than the same program compiled with gcc. A process generating 88 syscalls is probably slower than a process generating 48.