https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109362
Bug ID: 109362
Summary: codegen adds unnecessary extra add when reading atomic
member
Product: gcc
Version: 12.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c++
Assignee: unassigned at gcc dot gnu.org
Reporter: barry.revzin at gmail dot com
Target Milestone: ---
This program:
#include <atomic>
struct S {
long size;
std::atomic<char*> read_ptr;
auto peek() const -> const char* {
return read_ptr.load(std::memory_order_acquire);
}
};
auto with_atomic(S const& v) {
while (true) {
if (v.peek()) {
return true;
}
}
}
emits (on gcc 12.2 -O3):
with_atomic(S const&):
add rdi, 8
.L2:
mov rax, QWORD PTR [rdi]
test rax, rax
je .L2
mov eax, 1
ret
But that add is completely necessary, the mov could just be:
mov rax, QWORD PTR [rdi + 8]
which is what clang (16.0 -O3) generates:
with_atomic(S const&): # @with_atomic(S const&)
.LBB0_1: # =>This Inner Loop Header: Depth=1
mov rax, qword ptr [rdi + 8]
test rax, rax
je .LBB0_1
mov al, 1
ret
It's not just an extra add, it's consuming an extra register - which has more
downstream optimization effects.
