Creating a C++ pre-increment operation with CLANG inline assembly

Question

I am trying to do the equivalent of b = ++a; with inline assembly, but I'm getting strange values in my variables after executing the code. I am using clang++ (g++ compatible) to compile inline assembly. Here is what I get so far:

#include <iostream>

using std::endl;
using std::cout;

int main()
{
    uint64_t a = 0;
    uint64_t b = 0;

    asm volatile(
    "pushq %%rbp;"
    "movq %%rsp, %%rbp;"
    "movl $0, -4(%%rbp);"
    "movl $0, -8(%%rbp);"
    "addq $1, -4(%%rbp);"
    "mov -4(%%rbp), %%rax;"
    "mov %%rax, -8(%%rbp);"
    "mov -4(%%rbp), %0;"
    "mov -8(%%rbp), %1;"
    "movq %%rbp, %%rsp;"
    "popq %%rbp"
    :"=r" (a), "=r" (b)
    :
    :"%rax", "%rbp", "%rsp"
    );
    cout << "a = " << a << ", b = " << b << endl;
    return 0;
}

Apple clang version 11.0.0 (clang-1100.0.33.12) Target: x86_64-apple-darwin18.7.0 — Oleg, Nov 20 '19 at 16:17
`movl $0, -4(%%rbp)` only writes the low 32 bits hence the top 32 are random garbage. Also, a 64 bit integer is 8 bytes. So you want `-8(%rbp)` and `-16(%rbp)`. Your code is of course broken in other ways, such as overwriting the red zone of the containing function and doesn't make much sense anyway. — Jester, Nov 20 '19 at 16:31
Thanks and yes, it is working with -8 and -16, but what should I change to be ok? — Oleg, Nov 20 '19 at 16:58

Michael Petch · Accepted Answer · 2019-11-20T19:12:56.467

Your code is overly complex and it appears you might have looked at compiler generated code from elsewhere to create an answer.

What does b=++a; do? It first increments a and then assigns that value to b. In such an expression a is both used as input and output. The value of a is read, incremented by 1, saved and the result copied to b. With GCC's extended inline assembly you can treat a as an input and output operand using the + modifier on the constraint. b can be used with an output only operand using the = modifier on the constraint. The INC instruction can be used to increment a and the MOV instruction to copy that value b.

The inline assembly could simply look like:

#include <iostream>

using std::endl;
using std::cout;

int main()
{
    uint64_t a = 0;
    uint64_t b = 0;

    asm ("inc %0\n\t"
         "mov %0, %1"
         : "+r" (a), "=r" (b)
    );
    cout << "a = " << a << ", b = " << b << endl;
    return 0;
}

The output should be:

a = 1, b = 1

Note: Since there are no side effects other than the registers we told the compiler we'd modify it isn't necessary to use volatile on the asm statement.

Thanks, yes this is correct but what is the red zone that I am modifying in my example? How to fix that? — Oleg, Nov 21 '19 at 08:10
@oleg The [red zone](https://en.wikipedia.org/wiki/Red_zone_(computing)) is the 128 bytes immediately below the current RSP. You have to subtract 128 from RSP before using the stack and add 128 back when finished. An example is here: https://stackoverflow.com/a/37503773/3857942 — Michael Petch, Nov 21 '19 at 08:23
You only have to worry about the red zone if you are putting data on the stack in 64-bit code in your inline assembly. The red zone doesn't exist in 32-bit Linux. In my example I don't use the stack so I don't need to concern myself with the red zone. — Michael Petch, Nov 21 '19 at 08:24

Creating a C++ pre-increment operation with CLANG inline assembly

1 Answers1