[Bug c/39620] New: wrong assumption of clobbered registers of inline assembly

[Bug c/39620] New: wrong assumption of clobbered registers of inline assembly

[codemasterhs at yahoo dot de]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain, Size: 1041 bytes --]

I have the following code:

static inline void zeromem4b(uint32t *dst, uint32t count) {
        asm volatile("xor %%eax,%%eax\n\trep stosl"
                :
                :"c"(count),"D"(dst)
                :"%eax", "cc", "memory");
}

If I call it and after it want to, e.g. print out the value of *dst it gives me
the value which it has after running the function. That means that gcc thinks
that "%edi" hasn´t changed, but it did and I can not put it into the clobbered
list, because this gives me an error.

I compile with "-O2 -Wall -fno-builtin -march=i586".


-- 
           Summary: wrong assumption of clobbered registers of inline
                    assembly
           Product: gcc
           Version: 4.3.3
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: codemasterhs at yahoo dot de
  GCC host triplet: cygwin
GCC target triplet: i586 elf


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39620

[Bug c/39620] wrong assumption of clobbered registers of inline assembly

[jakub at gcc dot gnu dot org]

------- Comment #1 from jakub at gcc dot gnu dot org  2009-04-03 08:11 -------
User error, if you don't tell gcc that %edi is clobbered, obviously it can
assume it hasn't.
As rep stosl modifies both %ecx and %edi, you can write e.g.:
int tmp1, tmp2;
asm volatile("xor %%eax,%%eax\n\trep stosl" : "=c" (tmp1), "=D" (tmp2) : "0"
(count), "1" (dst) : "%eax", "cc", "memory");
or:
asm volatile("xor %%eax,%%eax\n\trep stosl" : "+c" (count), "+D" (dst) : :
"%eax", "cc", "memory");
if you don't mind that the count and dst variables will change.
Also note that __builtin_memset (dst, '\0', count); will in most cases
result in more optimal code for your CPU, hardcoding these in assembly is
usually a bad idea.


-- 

jakub at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|                            |INVALID


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39620

[Bug c/39620] wrong assumption of clobbered registers of inline assembly

[codemasterhs at yahoo dot de]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain, Size: 1472 bytes --]



------- Comment #2 from codemasterhs at yahoo dot de  2009-04-03 08:30 -------
Subject: Re:  wrong assumption of clobbered registers of inline
 assembly

The code is for my loader/os and so I can´t and do not want to use the 
builtin functions.

Your 1st example works, but I find it a bit strange that I have to write 
such code only to tell gcc, that 2 registers are clobbered! Your 2nd 
example doesn´t work, because of the plus, this gives me an error.

I think the better way would be either gcc assumes that edi and ecx are 
clobbered (which isn´t so) or I can specify that in the clobbered list 
(which is not possible)!

jakub at gcc dot gnu dot org wrote:
> ------- Comment #1 from jakub at gcc dot gnu dot org  2009-04-03 08:11 -------
> User error, if you don't tell gcc that %edi is clobbered, obviously it can
> assume it hasn't.
> As rep stosl modifies both %ecx and %edi, you can write e.g.:
> int tmp1, tmp2;
> asm volatile("xor %%eax,%%eax\n\trep stosl" : "=c" (tmp1), "=D" (tmp2) : "0"
> (count), "1" (dst) : "%eax", "cc", "memory");
> or:
> asm volatile("xor %%eax,%%eax\n\trep stosl" : "+c" (count), "+D" (dst) : :
> "%eax", "cc", "memory");
> if you don't mind that the count and dst variables will change.
> Also note that __builtin_memset (dst, '\0', count); will in most cases
> result in more optimal code for your CPU, hardcoding these in assembly is
> usually a bad idea.
>
>
>   


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39620