Thread: [Inline Assembly] What am I doing wrong in this function?

I'm trying to learn assembly and I'm trying to covert a function using inline assembly. The function I want to convert is the following:

Code:

void no_asm_ver() {
  int first_number = 0, second_number = 1, third_number, i;

  for(i = 2; i < 20; ++i) {
    third_number = first_number + second_number;
    printf("%d ", third_number);
    first_number = second_number;
    second_number = third_number;
  } fflush(stdout); write(1, "\n", 1);
}

So I want to convert this to assembly (except for the loop and the last line where it flushes "stdout" and prints a newline). I tried to do the following:

Code:

void asm_ver() {
  const void* prompt = "%d ";
  asm (
    "mov $0, %r15;" // frist_number
    "mov $1, %r14;" // second_number
    // "mov $0, %r13;" // third_number
  );

  int i;

  for(i = 2; i < 20; ++i) {
    asm volatile (
      // third_number = first_number + second_number
      "mov %r15, %r13;"
      "add %r14, %r13;"

      // printf("%d", third_number);
      "mov %r13, %rsi;"
      "call printf;"

      // first_number = second_number
      "mov %r14, %r15;"

      // second_number = third_number
      "mov %r13, %r14;"

      :: "D" (prompt)
    );
  } fflush(stdout); write(1, "\n", 1);
}

I've used "r13-r15" for the variables. For "printf", I've seen a post (link in the bottom of this post) about the calling convention for functions so I'm trying to pass the first parameter (prompt) in the "rdi" register and the second parameter (third_num) in the "rsi" register and then call "printf". If I remove the last line from the "asm" statement (:: "D" (prompt)) then it will work. But now, I'm getting the following error message in GCC:

Code:

test.c: In function ‘asm_ver’:
test.c:67:5: error: invalid 'asm': operand number out of range
   67 |     asm volatile (
      |     ^~~
test.c:67:5: error: invalid 'asm': operand number out of range
test.c:67:5: error: invalid 'asm': operand number out of range
test.c:67:5: error: invalid 'asm': operand number out of range
test.c:67:5: error: invalid 'asm': operand number out of range
test.c:67:5: error: invalid 'asm': operand number missing after %-letter
test.c:67:5: error: invalid 'asm': operand number out of range
test.c:67:5: error: invalid 'asm': operand number out of range
test.c:67:5: error: invalid 'asm': operand number out of range
test.c:67:5: error: invalid 'asm': operand number out of range

In TCC, the message is the following:

Code:

test.c:83: error: invalid operand reference after %

So none of them are very informative about what I'm doing wrong. Any ideas? Is there a chane that it doesn't like that I'm using the "r" registers...

Link for the post talking about the calling conversion: X86-64 assembly from scratch - Conrad Kleinespel

Originally Posted by Salem

Extended Asm (Using the GNU Compiler Collection (GCC))
This suggests you need to add some \n\t to get proper lines and indentation.

Otherwise the assembler just sees one long line and barfs on it.

I added them but this doesn't seem to fix anything. I also don't understand why seeing it as a long line will not work as there are semicolons to indicate the end of an instruction.

Originally Posted by Salem

Maybe your x86 doesn't have registers called r13, r14, r15 etc.

My whaaaaaat????? My CPU is Ryzen 5 2400G

Also, I know that I was the one that originally thought that but tbh, it doesn't make sense as it works without the last line of the asm statement so I don't think that this is the problem...

Originally Posted by john.c

Does this work:

Code:

#include <stdio.h>
 
int main()
{
    int src = 1, dst;
 
    __asm__(
        "mov %1, %0\n\t"  // dst = src
        "add $1, %0"      // ++dst
        : "=r" (dst)
        : "r" (src));
 
    printf("%d\n", dst);
    return 0;
}

Thanks you, this works but not for my example. "r" let's the compiler choose a register and I want to use specif ones in this case. I mean, this would probably be OK to make my code but It will be a good practice to find and fix my code so I can properly learn. In any case, thank you!

Originally Posted by Salem

This seemed to work, doubling up the %
But only in the 2nd example.

Code:

void asm_ver() {
  const void* prompt = "%d ";
  asm (
    "mov $0, %r15\n\t" 
    "mov $1, %r14\n\t"
  );
 
  int i;
 
  for(i = 2; i < 20; ++i) {
    asm volatile (
      "mov %%r15, %%r13\n\t"
      "add %%r14, %%r13\n\t"
 
      "mov %%r13, %%rsi\n\t"
      "call printf\n\t"
 
      "mov %%r14, %%r15\n\t"
 
      "mov %%r13, %%r14\n\t"
 
      :: "D" (prompt)
    );
  }
}

$ gcc -c -S foo.c
$ cat foo.s
<< snipped >>
#APP
# 3 "foo.c" 1
    mov $0, %r15
    mov $1, %r14
    
# 0 "" 2
#NO_APP

#APP
# 11 "foo.c" 1
    mov %r15, %r13
    add %r14, %r13
    mov %r13, %rsi
    call printf
    mov %r14, %r15
    mov %r13, %r14
    
# 0 "" 2
#NO_APP

This.... WORKS! What? Why it works now? The documentation needs to be bigger and include way more examples... Anyway, I plan on making my own compiler and I'm learning how to create ELF files so I should probably not bother with GCC inline assembly at all.

Once again, thank you for your help!