GCC has no idea about optimization even if the optimization is turned
to the maximum:
unsigned long long x(unsigned lo, unsigned hi)
{
return ((unsigned long long)hi << 32) | lo;
}
gcc -O3 -c -o a.o a.c; objdump -d a.o:
0: 55 push %ebp
1: 89 e5 mov %esp,%ebp
3: 8b 4d 0c mov 0xc(%ebp),%ecx
6: 53 push %ebx
7: 89 ca mov %ecx,%edx
9: 31 db xor %ebx,%ebx
b: 8b 4d 08 mov 0x8(%ebp),%ecx
e: 31 c0 xor %eax,%eax
10: 09 da or %ebx,%edx
^^^ %ebx is zero here, can be thrown out
12: 09 c8 or %ecx,%eax
^^^ %ax is zero here, can be replaced with mov
%ecx, %eax
14: 8b 1c 24 mov (%esp),%ebx
17: c9 leave
18: c3 ret
After seeing this I am not sure if GCC has a peephole optimizer but if they
have they have to add following rules:
or reg1, reg2 where reg1 is containing 0 can be thrown out
or reg1, reg2 where reg2 is containing 0 can be replaced with mov reg1, reg2
and possibly further peephole optimized
After some manual rewrite the function shrinks significantly to:
55 push %ebp
89 e5 mov %esp,%ebp
8b 4d 0c mov 0xc(%ebp),%edx
53 push %ebx
8b 4d 08 mov 0x8(%ebp),%eax
8b 1c 24 mov (%esp),%ebx
c9 leave
c3 ret
CL<
