How to reference variables inside an "asm"?

I've got -masm=intel on so I can use good ol' Intel syntax. I need to pop a value off the stack into my unsigned short "cur_val", how do I reference "cur_val"?

Code:

struct bigint_t init_bigint_t(unsigned long count, ...)
{
    struct bigint_t new_bigint;
    for(;count > 0;count--)
    {
        block cur_val;
        asm("pop word cur_val\n\t"); /* Error: junk `cur_val' after expression */

    }
}

Although it's Intel syntax, it's not Fasm dialect, I'm a little confused.

Well that depends entirely on your compiler.
There is NO standard for what asm does, so you should at least start with the manual.

I'm using GCC, so I guess that would mean I'm really using GAS in Intel mode().

GCC-Inline-Assembly-HOWTO

Yeah, I can use Google to. The hard part is finding documentation with Intel syntax turned on. It really does matther.
Intel(Fasm):

Code:

lea eax, dword ptr [esi + ebx * 4 - 0x10]

See how nice and neat that is, with it's continuity and no stupid "%"s.

AT&T:

Code:

leal -0x10(%esi, %ebx, 4), %eax

Just seeing it... Ewww... I just threw up a little... AT&T syntax is sickening.

Code:

struct bigint_t init_bigint_t(unsigned long count, ...)
{
    struct bigint_t new_bigint;
    for(;count > 0;count--)
    {
        block cur_val asm("_cur_val");
        asm volatile
        (
        "pop word ptr [cur_val]#"
        );
    }
}

I'm getting cur_val undefined, why? Am I supposed to use some mangled form of "cur_val"?

Code:

struct bigint_t init_bigint_t(unsigned long count, ...)
{
    va_list blocks;
    struct bigint_t new_bigint;
    new_bigint.count = count;
    va_start(blocks, count);
    new_bigint.root = (struct bigint_node*)malloc(sizeof(struct bigint_node)); /* Alloc space for root */
    struct bigint_node *cur_node = new_bigint.root; /* Set current node to root */
    for(;count > 0;count--)
    {
        cur_node->val = va_arg(blocks, block); /* Get block */
        cur_node->next = (struct bigint_node*)malloc(0); /* Allocate next node */
        cur_node = cur_node->next; /* Set current node to next node(the node just allocated) */
    }
    va_end(blocks);
    return new_bigint;
}

I went ahead and just did it with stdarg instead. Wasn't as hard as I was expecting either.

malloc(0) is asking for trouble...

Why does the syntax matter?

What matters is getting the constraint strings right. The actual instruction between the first pair of quotes doesn't really matter.

This "works" (compiles without complaint)

Code:

$ cat foo.c
#include <stdio.h>


void foo ( unsigned long count, ...)
{
    for(;count > 0;count--)
    {
        int cur_val;
        asm("pop %0\n" : :"m"(cur_val) );
        printf("Value=%x\n", cur_val);
    }
}

int main()
{
    foo(3,22,44,66);
    return 0;
}
$ gcc -masm=intel foo.c
$ ./a.out 
Value=0
Value=8048500
Value=8048500

It doesn't work because the top of the stack are not parameters, but the return address (and other things).
The variadic parameters would be accessible though ebp, not esp.

The generated code around the asm looks like this

Code:

.globl foo
	.type	foo, @function
foo:
	push	ebp
	mov	ebp, esp
	sub	esp, 24
	jmp	.L2
.L3:
#APP
# 9 "foo.c" 1
	pop DWORD PTR [ebp-4]

# 0 "" 2
#NO_APP
	mov	eax, DWORD PTR [ebp-4]

The malloc(0) thing was just an accident, it was supposed to be "malloc(sizeof(struct bigint_node))".

I'm not sure what I was thinking wanting to pop a param off the stack on my 64 bit box(Not that it would work in cdecl, as you've pointed out.). I guess I forgot that they use a weird register based calling convention.