https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92149
Bug ID: 92149
Summary: Enefficient x86_64 code
Product: gcc
Version: 9.2.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: maxim.yegorushkin at gmail dot com
Target Milestone: ---
The following code:
#include <array>
#include <cstring>
#include <cstdint>
using std::uint64_t;
using A = std::array<unsigned char, 6>;
template<class T>
constexpr A pack(T v) {
using C = unsigned char;
return {C(v), C(v >> 8), C(v >> 16), C(v >> 24), C(v >> 32), C(v >>
40)};
}
template<class T>
constexpr A pack2(T v) {
A r{};
std::memcpy(r.data(), &v, 6);
return r;
}
A f(uint64_t a) { return pack(a); }
A f2(uint64_t a) { return pack2(a); }
When compiled with `clang++-9.0 -std=gnu++17 -O3 -march=skylake` produces the
most efficient assembly:
f(unsigned long):
mov rax, rdi
ret
f2(unsigned long):
mov rax, rdi
ret
Whereas when compiled with `g++-9.2 -std=gnu++17 -O3 -march=skylake` the
assembly is inefficient:
f(unsigned long):
mov rax, rdi
shr rax, 32
mov BYTE PTR [rsp-2], al
mov rax, rdi
shr rax, 40
mov BYTE PTR [rsp-1], al
movzx eax, WORD PTR [rsp-2]
sal rax, 32
mov rdx, rax
mov eax, edi
or rax, rdx
ret
f2(unsigned long):
movabs rax, 281474976710655
and rax, rdi
ret
Why does g++ emit such verbose code please?
