Reputation: 6355
int a = (int)5;
Does VS optimize that (remove the cast)? The situation above is trivial but I'm writing some template classes which take arbitrary type arguments in constructor:
template <typename U>
MyClass(U argument)
{
T a = (T)argument;
}
In most cases the cast is needed to avoid compiler warnings but when T = U then the cast is redundant. Or maybe is there a better way to implement that?
Upvotes: 2
Views: 272
Reputation: 56479
Following Oded's hint from comments, I did a test in gcc-4.7.2 and MVSC-2012:
template <typename U, typename T>
void assign1(const T& t, U &u)
{
u = (U) t; // CAST
}
template <typename U, typename T>
void assign2(const T& t, U &u)
{
u = t; // WITHOUT CAST
}
int main()
{
{
int t = 12;
int u = 1;
assign1(t, u);
}
{
int t = 12;
int u = 1;
assign2(t, u);
}
}
assign1
assembly code (gcc):
!{
! u = (U) t;
assign1<int, int>(int const&, int&)+3: mov 0x8(%ebp),%eax
assign1<int, int>(int const&, int&)+6: mov (%eax),%edx
assign1<int, int>(int const&, int&)+8: mov 0xc(%ebp),%eax
assign1<int, int>(int const&, int&)+11: mov %edx,(%eax)
!}
assign2
assembly code(gcc):
!{
! u = t;
assign2<int, int>(int const&, int&)+3: mov 0x8(%ebp),%eax
assign2<int, int>(int const&, int&)+6: mov (%eax),%edx
assign2<int, int>(int const&, int&)+8: mov 0xc(%ebp),%eax
assign2<int, int>(int const&, int&)+11: mov %edx,(%eax)
!}
They are same in gcc.
assign1
assembly code(MSVC):
001413EE mov eax,dword ptr [u]
001413F1 mov ecx,dword ptr [t]
001413F4 mov edx,dword ptr [ecx]
001413F6 mov dword ptr [eax],edx
assign2
assembly code(MSVC):
0014142E mov eax,dword ptr [u]
00141431 mov ecx,dword ptr [t]
00141434 mov edx,dword ptr [ecx]
00141436 mov dword ptr [eax],edx
They are same in MSVC, too.
So, both compilers omit the cast.
Upvotes: 2