Description of problem:

gcc on Fedora 16 fails to generate "memcpy" code when compiling a function calling a static inline function containing __builtin_memcpy with an immediate value as the destination argument.

Version-Release number of selected component (if applicable):

gcc-4.6.2-1.fc16.x86_64

How reproducible:

always

Steps to Reproduce:
1. compile the following function with "gcc -O -S"

----
static inline void *cp(void* d,void* s,int l)
{
    return __builtin_memcpy(d,s,l);
}
void main(void *p)
{
    cp(0,p,256);
}
----

When we change to "cp(p,0,256)", it generates expected code.
 
Actual results:

...
main:
.LFB1:
        .cfi_startproc
        rep
        ret
        .cfi_endproc
.LFE1:
...

Expected results:

...
main:
.LFB1:
        .cfi_startproc
        movq    %rdi, %rdx
        movl    $0, %esi
        movl    $256, %eax
(code for memcpy)
...

Additional info: