gcc堆栈优化？

Question

I just wrote a code sample in C and tried disassembling it. Following is the code sample.

void start() {
    char phone[100];
    strcmp(phone, "12312312313");

    char name[100];
    strcmp(name, "eQuiNoX");

    char contact[100];
    strcmp(contact, "PM twitter.com/eQuiNoX__");
}

When I disassemble the start function I get the following:-

08048414 <start>:
 8048414: 55                    push   ebp
 8048415: 89 e5                 mov    ebp,esp
 8048417: 81 ec 58 01 00 00     sub    esp,0x158
 804841d: c9                    leave  
 804841e: c3                    ret   

1. I have not enabled any kind of optimization. Could someone explain why I get 158 subtracted from esp rather than assembly code which pushes values onto the stack and calls the strcmp method? Is it because it does not depend on any user input?
2. Also, is there any way I could generate the extended assembly (im not sure if thats the right term, i just wish to see the assembly code for pushing values onto the stack and the calling of the strcmp function). Is there any way I could do that?
3. Is this kind of behavior specific to processor architectures or gcc versions or both?

Answer 1

First, strcmp is a standard library function, so gcc is free to have special knowledge about how it works. In fact, it does; it'll seldom generate a library call. You can try -fno-builtin to disable.

Second, you're comparing to unitialized values. This is, I believe undefined behavior. So the compiler may do anything it pleases, including producing random code.

You can try the -S option to gcc get more detailed disassembly (or, rather, lack of assembly); alternatively, if you compile with -g (debugging), objdump -S will display the source along with the assembled code.

Here is an example I compiled with gcc -fno-builtin -g -O0 test.c -c and then dumped with objdump -S test.o :

test.o:     file format elf64-x86-64


Disassembly of section .text:

0000000000000000 <main>:
#include <string.h>

int main() {
   0:   55                      push   %rbp
   1:   48 89 e5                mov    %rsp,%rbp
   4:   48 83 ec 10             sub    $0x10,%rsp
    const char foo[] = "foo";
   8:   8b 05 00 00 00 00       mov    0x0(%rip),%eax        # e <main+0xe>
   e:   89 45 f0                mov    %eax,-0x10(%rbp)
    return strcmp(foo, "bar");
  11:   48 8d 45 f0             lea    -0x10(%rbp),%rax
  15:   be 00 00 00 00          mov    $0x0,%esi
  1a:   48 89 c7                mov    %rax,%rdi
  1d:   e8 00 00 00 00          callq  22 <main+0x22>
}
  22:   c9                      leaveq 
  23:   c3                      retq   

Answer 2

Regarding the sub esp,0x158 instruction, rather than generate a boatload of push operations (which involve copying the operand to the stack too, not just reserving space), typically the compiler will just reserve enough space for all local variables by moving the stack pointer just once. That's what this instruction is doing. 0x158 is 344 in decimal, so it's reserving 300 bytes for the arrays and probably some extra space for compiler generated structures (or maybe to put the strcmp operands on the stack too).

Answer 3

1. Because your code has no effect in the program execution. All variables are used in function calls whose return values are discarded, thus the compiler tagged it as unused code, and felt it should remove. If you want to keep the unused code, be sure you're not using any optimisations - compile with -O0 .

2. See my point above.

3. I suspect most compilers would perform this optimisation, independent of architecture.

Answer 4

It seems that it really was optimized. You can try compiling with -O0 flag to ensure that no optimization is made (however I'm not sure it will work).

Or you can just return cmp results from the start function to show the compiler that you really use them:

int start() {
    char phone[] = "43423432";
    return strcmp(phone, "12312312313");
}