I think that having no "o" constraint is asm for i386 target is unfair, because in effect all the "m" values are actually offsetable. For example, when I write: movl 4+%5, %2 - I always get correct code. The only problem here is that I get gas warning each time %5 happens to be a base register without an offset, i.e. something like (%ebx) turns the above line into movl 4+(%ebx), %eax If there was "o" constraing that behaves like "m" in most cases but prints zero in case of zero offset, some people like me would be happy. What I want, once again: %5 == (%ebx) "m" prints (%ebx) "o" prints 0(%ebx) -> quite correct movl 4+0(%ebx), %eax in "o" case. %5 == 12(%ebx) both "m" and "o" print (%ebx) -> still correct movl 4+12(%ebx), %eax code. I hope that's easy to do. Thanks in advance :)
So this is what you want: int f(int *i){int t; asm("a %1":"=r"(t):"o"(*i)); return t;} 0(%eax) and not (%eax) which it currently is now, right?
(In reply to comment #1) > So this is what you want: > int f(int *i){int t; asm("a %1":"=r"(t):"o"(*i)); return t;} > 0(%eax) and not (%eax) which it currently is now, right? Yes, that's right. But that's not enough. Nowadays gcc doesn't like "o" at all, it says: "error: impossible constraint in `asm'". /usr/local/egcs-3/bin/c++ -O -Wall mo.cpp mo.cpp: In function `int main(int, char**)': mo.cpp:3: error: impossible constraint in `asm' ==> mo.cpp <== int main(int argc, char *[]) { long long A = argc; asm volatile( "" : "=o" (A) : : "cc" ); return A; } With "m", it compiles fine. (note -O switch)