https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95384
Bug ID: 95384
Summary: Poor codegen cause by using base class instead of
member for Optional construction
Product: gcc
Version: 10.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c++
Assignee: unassigned at gcc dot gnu.org
Reporter: barry.revzin at gmail dot com
Target Milestone: ---
Following up on #95383:
struct nullopt_t {} inline constexpr nullopt{};
template <typename T>
struct OptionalStorage {
struct Empty { };
union {
Empty _;
T value;
};
bool engaged;
OptionalStorage(nullopt_t) : _(), engaged(false) { }
OptionalStorage(T v) : value(v), engaged(true) { }
};
template <typename T>
struct OptionalWithBase : OptionalStorage<T> {
using OptionalStorage<T>::OptionalStorage;
};
template <typename T>
struct OptionalWithMember {
OptionalStorage<T> o;
OptionalWithMember(nullopt_t) : o(nullopt) { }
OptionalWithMember(T v) : o(v) { }
};
OptionalWithBase<int> foo_with_base(bool b) {
if (b) {
return 42;
}
return nullopt;
}
OptionalWithMember<int> foo_with_member(bool b) {
if (b) {
return 42;
}
return nullopt;
}
OptionalWithBase<T> and OptionalWithMember<T> should be the same thing. It's
just that one inherits from its storage and the other has it as a member. But
the codegen is very different (https://godbolt.org/z/j8m68Y), probably
something to do with tail padding?
gcc 10.1 -O2:
foo_with_base(bool):
test dil, dil
je .L2
mov DWORD PTR [rsp-8], 42
mov BYTE PTR [rsp-4], 1
mov rax, QWORD PTR [rsp-8]
ret
.L2:
mov BYTE PTR [rsp-4], 0
mov rax, QWORD PTR [rsp-8]
ret
foo_with_member(bool):
test dil, dil
mov eax, 0
movabs rdx, 4294967338
cmovne rax, rdx
ret
gcc 10.2 -O3 or gcc trunk -O2:
foo_with_base(bool):
xor eax, eax
test dil, dil
je .L2
mov DWORD PTR [rsp-8], 42
mov eax, 1
.L2:
mov BYTE PTR [rsp-4], al
mov rax, QWORD PTR [rsp-8]
ret
foo_with_member(bool):
xor edx, edx
mov ecx, 42
test dil, dil
cmovne rax, rcx
mov ecx, 1
cmovne rdx, rcx
movabs rcx, -1095216660481
and rax, rcx
sal rdx, 32
or rax, rdx
ret
clang 10, -O2 or -O3:
foo_with_base(bool): # @foo_with_base(bool)
shl rdi, 32
lea rax, [rdi + 42]
ret
foo_with_member(bool): # @foo_with_member(bool)
shl rdi, 32
lea rax, [rdi + 42]
ret
