Skip to content

[libc][CPP] correct cmpxchg failure ordering inference #133127

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 5 commits into from
Mar 26, 2025
Merged

[libc][CPP] correct cmpxchg failure ordering inference #133127

Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
46f0710
[libc][CPP] correct cmpxchg failure ordering inference
SchrodingerZhu Mar 26, 2025
6c9f699
address comments
SchrodingerZhu Mar 26, 2025
4d23450
address comments
SchrodingerZhu Mar 26, 2025
c3811dd
unify spacing
SchrodingerZhu Mar 26, 2025
d50e1f6
update
SchrodingerZhu Mar 26, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
164 changes: 92 additions & 72 deletions libc/src/__support/CPP/atomic.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -40,6 +40,28 @@ enum class MemoryScope : int {
#endif
};

namespace impl {
LIBC_INLINE constexpr int order(MemoryOrder mem_ord) {
return static_cast<int>(mem_ord);
}

LIBC_INLINE constexpr int scope(MemoryScope mem_scope) {
return static_cast<int>(mem_scope);
}

template <class T> LIBC_INLINE T *addressof(T &ref) {
return __builtin_addressof(ref);
}

LIBC_INLINE constexpr int infer_failure_order(MemoryOrder mem_ord) {
if (mem_ord == MemoryOrder::RELEASE)
return order(MemoryOrder::RELAXED);
if (mem_ord == MemoryOrder::ACQ_REL)
return order(MemoryOrder::ACQUIRE);
return order(mem_ord);
}
} // namespace impl

template <typename T> struct Atomic {
static_assert(is_trivially_copyable_v<T> && is_copy_constructible_v<T> &&
is_move_constructible_v<T> && is_copy_assignable_v<T> &&
Expand All @@ -54,15 +76,6 @@ template <typename T> struct Atomic {

private:
// type conversion helper to avoid long c++ style casts
LIBC_INLINE static int order(MemoryOrder mem_ord) {
return static_cast<int>(mem_ord);
}

LIBC_INLINE static int scope(MemoryScope mem_scope) {
return static_cast<int>(mem_scope);
}

LIBC_INLINE static T *addressof(T &ref) { return __builtin_addressof(ref); }

// Require types that are 1, 2, 4, 8, or 16 bytes in length to be aligned to
// at least their size to be potentially used lock-free.
Expand Down Expand Up @@ -98,10 +111,11 @@ template <typename T> struct Atomic {
[[maybe_unused]] MemoryScope mem_scope = MemoryScope::DEVICE) {
T res;
#if __has_builtin(__scoped_atomic_load)
__scoped_atomic_load(addressof(val), addressof(res), order(mem_ord),
scope(mem_scope));
__scoped_atomic_load(impl::addressof(val), impl::addressof(res),
impl::order(mem_ord), impl::scope(mem_scope));
#else
__atomic_load(addressof(val), addressof(res), order(mem_ord));
__atomic_load(impl::addressof(val), impl::addressof(res),
impl::order(mem_ord));
#endif
return res;
}
Expand All @@ -116,20 +130,22 @@ template <typename T> struct Atomic {
store(T rhs, MemoryOrder mem_ord = MemoryOrder::SEQ_CST,
[[maybe_unused]] MemoryScope mem_scope = MemoryScope::DEVICE) {
#if __has_builtin(__scoped_atomic_store)
__scoped_atomic_store(addressof(val), addressof(rhs), order(mem_ord),
scope(mem_scope));
__scoped_atomic_store(impl::addressof(val), impl::addressof(rhs),
impl::order(mem_ord), impl::scope(mem_scope));
#else
__atomic_store(addressof(val), addressof(rhs), order(mem_ord));
__atomic_store(impl::addressof(val), impl::addressof(rhs),
impl::order(mem_ord));
#endif
}

// Atomic compare exchange
LIBC_INLINE bool compare_exchange_strong(
T &expected, T desired, MemoryOrder mem_ord = MemoryOrder::SEQ_CST,
[[maybe_unused]] MemoryScope mem_scope = MemoryScope::DEVICE) {
return __atomic_compare_exchange(addressof(val), addressof(expected),
addressof(desired), false, order(mem_ord),
order(mem_ord));
return __atomic_compare_exchange(
impl::addressof(val), impl::addressof(expected),
impl::addressof(desired), false, impl::order(mem_ord),
impl::infer_failure_order(mem_ord));
}

// Atomic compare exchange (separate success and failure memory orders)
Expand All @@ -138,17 +154,19 @@ template <typename T> struct Atomic {
MemoryOrder failure_order,
[[maybe_unused]] MemoryScope mem_scope = MemoryScope::DEVICE) {
return __atomic_compare_exchange(
addressof(val), addressof(expected), addressof(desired), false,
order(success_order), order(failure_order));
impl::addressof(val), impl::addressof(expected),
impl::addressof(desired), false, impl::order(success_order),
impl::order(failure_order));
}

// Atomic compare exchange (weak version)
LIBC_INLINE bool compare_exchange_weak(
T &expected, T desired, MemoryOrder mem_ord = MemoryOrder::SEQ_CST,
[[maybe_unused]] MemoryScope mem_scope = MemoryScope::DEVICE) {
return __atomic_compare_exchange(addressof(val), addressof(expected),
addressof(desired), true, order(mem_ord),
order(mem_ord));
return __atomic_compare_exchange(
impl::addressof(val), impl::addressof(expected),
impl::addressof(desired), true, impl::order(mem_ord),
impl::infer_failure_order(mem_ord));
}

// Atomic compare exchange (weak version with separate success and failure
Expand All @@ -158,20 +176,22 @@ template <typename T> struct Atomic {
MemoryOrder failure_order,
[[maybe_unused]] MemoryScope mem_scope = MemoryScope::DEVICE) {
return __atomic_compare_exchange(
addressof(val), addressof(expected), addressof(desired), true,
order(success_order), order(failure_order));
impl::addressof(val), impl::addressof(expected),
impl::addressof(desired), true, impl::order(success_order),
impl::order(failure_order));
}

LIBC_INLINE T
exchange(T desired, MemoryOrder mem_ord = MemoryOrder::SEQ_CST,
[[maybe_unused]] MemoryScope mem_scope = MemoryScope::DEVICE) {
T ret;
#if __has_builtin(__scoped_atomic_exchange)
__scoped_atomic_exchange(addressof(val), addressof(desired), addressof(ret),
order(mem_ord), scope(mem_scope));
__scoped_atomic_exchange(impl::addressof(val), impl::addressof(desired),
impl::addressof(ret), impl::order(mem_ord),
impl::scope(mem_scope));
#else
__atomic_exchange(addressof(val), addressof(desired), addressof(ret),
order(mem_ord));
__atomic_exchange(impl::addressof(val), impl::addressof(desired),
impl::addressof(ret), impl::order(mem_ord));
#endif
return ret;
}
Expand All @@ -181,10 +201,12 @@ template <typename T> struct Atomic {
[[maybe_unused]] MemoryScope mem_scope = MemoryScope::DEVICE) {
static_assert(cpp::is_integral_v<T>, "T must be an integral type.");
#if __has_builtin(__scoped_atomic_fetch_add)
return __scoped_atomic_fetch_add(addressof(val), increment, order(mem_ord),
scope(mem_scope));
return __scoped_atomic_fetch_add(impl::addressof(val), increment,
impl::order(mem_ord),
impl::scope(mem_scope));
#else
return __atomic_fetch_add(addressof(val), increment, order(mem_ord));
return __atomic_fetch_add(impl::addressof(val), increment,
impl::order(mem_ord));
#endif
}

Expand All @@ -193,10 +215,11 @@ template <typename T> struct Atomic {
[[maybe_unused]] MemoryScope mem_scope = MemoryScope::DEVICE) {
static_assert(cpp::is_integral_v<T>, "T must be an integral type.");
#if __has_builtin(__scoped_atomic_fetch_or)
return __scoped_atomic_fetch_or(addressof(val), mask, order(mem_ord),
scope(mem_scope));
return __scoped_atomic_fetch_or(impl::addressof(val), mask,
impl::order(mem_ord),
impl::scope(mem_scope));
#else
return __atomic_fetch_or(addressof(val), mask, order(mem_ord));
return __atomic_fetch_or(impl::addressof(val), mask, impl::order(mem_ord));
#endif
}

Expand All @@ -205,10 +228,11 @@ template <typename T> struct Atomic {
[[maybe_unused]] MemoryScope mem_scope = MemoryScope::DEVICE) {
static_assert(cpp::is_integral_v<T>, "T must be an integral type.");
#if __has_builtin(__scoped_atomic_fetch_and)
return __scoped_atomic_fetch_and(addressof(val), mask, order(mem_ord),
scope(mem_scope));
return __scoped_atomic_fetch_and(impl::addressof(val), mask,
impl::order(mem_ord),
impl::scope(mem_scope));
#else
return __atomic_fetch_and(addressof(val), mask, order(mem_ord));
return __atomic_fetch_and(impl::addressof(val), mask, impl::order(mem_ord));
#endif
}

Expand All @@ -217,10 +241,12 @@ template <typename T> struct Atomic {
[[maybe_unused]] MemoryScope mem_scope = MemoryScope::DEVICE) {
static_assert(cpp::is_integral_v<T>, "T must be an integral type.");
#if __has_builtin(__scoped_atomic_fetch_sub)
return __scoped_atomic_fetch_sub(addressof(val), decrement, order(mem_ord),
scope(mem_scope));
return __scoped_atomic_fetch_sub(impl::addressof(val), decrement,
impl::order(mem_ord),
impl::scope(mem_scope));
#else
return __atomic_fetch_sub(addressof(val), decrement, order(mem_ord));
return __atomic_fetch_sub(impl::addressof(val), decrement,
impl::order(mem_ord));
#endif
}

Expand All @@ -244,14 +270,6 @@ template <typename T> struct AtomicRef {
private:
T *ptr;

LIBC_INLINE static int order(MemoryOrder mem_ord) {
return static_cast<int>(mem_ord);
}

LIBC_INLINE static int scope(MemoryScope mem_scope) {
return static_cast<int>(mem_scope);
}

public:
// Constructor from T reference
LIBC_INLINE explicit constexpr AtomicRef(T &obj) : ptr(&obj) {}
Expand All @@ -270,9 +288,10 @@ template <typename T> struct AtomicRef {
[[maybe_unused]] MemoryScope mem_scope = MemoryScope::DEVICE) const {
T res;
#if __has_builtin(__scoped_atomic_load)
__scoped_atomic_load(ptr, &res, order(mem_ord), scope(mem_scope));
__scoped_atomic_load(ptr, &res, impl::order(mem_ord),
impl::scope(mem_scope));
#else
__atomic_load(ptr, &res, order(mem_ord));
__atomic_load(ptr, &res, impl::order(mem_ord));
#endif
return res;
}
Expand All @@ -287,9 +306,10 @@ template <typename T> struct AtomicRef {
store(T rhs, MemoryOrder mem_ord = MemoryOrder::SEQ_CST,
[[maybe_unused]] MemoryScope mem_scope = MemoryScope::DEVICE) const {
#if __has_builtin(__scoped_atomic_store)
__scoped_atomic_store(ptr, &rhs, order(mem_ord), scope(mem_scope));
__scoped_atomic_store(ptr, &rhs, impl::order(mem_ord),
impl::scope(mem_scope));
#else
__atomic_store(ptr, &rhs, order(mem_ord));
__atomic_store(ptr, &rhs, impl::order(mem_ord));
#endif
}

Expand All @@ -298,7 +318,8 @@ template <typename T> struct AtomicRef {
T &expected, T desired, MemoryOrder mem_ord = MemoryOrder::SEQ_CST,
[[maybe_unused]] MemoryScope mem_scope = MemoryScope::DEVICE) const {
return __atomic_compare_exchange(ptr, &expected, &desired, false,
order(mem_ord), order(mem_ord));
impl::order(mem_ord),
impl::infer_failure_order(mem_ord));
}

// Atomic compare exchange (strong, separate success/failure memory orders)
Expand All @@ -307,8 +328,8 @@ template <typename T> struct AtomicRef {
MemoryOrder failure_order,
[[maybe_unused]] MemoryScope mem_scope = MemoryScope::DEVICE) const {
return __atomic_compare_exchange(ptr, &expected, &desired, false,
order(success_order),
order(failure_order));
impl::order(success_order),
impl::order(failure_order));
}

// Atomic exchange
Expand All @@ -317,10 +338,10 @@ template <typename T> struct AtomicRef {
[[maybe_unused]] MemoryScope mem_scope = MemoryScope::DEVICE) const {
T ret;
#if __has_builtin(__scoped_atomic_exchange)
__scoped_atomic_exchange(ptr, &desired, &ret, order(mem_ord),
scope(mem_scope));
__scoped_atomic_exchange(ptr, &desired, &ret, impl::order(mem_ord),
impl::scope(mem_scope));
#else
__atomic_exchange(ptr, &desired, &ret, order(mem_ord));
__atomic_exchange(ptr, &desired, &ret, impl::order(mem_ord));
#endif
return ret;
}
Expand All @@ -330,10 +351,10 @@ template <typename T> struct AtomicRef {
[[maybe_unused]] MemoryScope mem_scope = MemoryScope::DEVICE) const {
static_assert(cpp::is_integral_v<T>, "T must be an integral type.");
#if __has_builtin(__scoped_atomic_fetch_add)
return __scoped_atomic_fetch_add(ptr, increment, order(mem_ord),
scope(mem_scope));
return __scoped_atomic_fetch_add(ptr, increment, impl::order(mem_ord),
impl::scope(mem_scope));
#else
return __atomic_fetch_add(ptr, increment, order(mem_ord));
return __atomic_fetch_add(ptr, increment, impl::order(mem_ord));
#endif
}

Expand All @@ -342,10 +363,10 @@ template <typename T> struct AtomicRef {
[[maybe_unused]] MemoryScope mem_scope = MemoryScope::DEVICE) const {
static_assert(cpp::is_integral_v<T>, "T must be an integral type.");
#if __has_builtin(__scoped_atomic_fetch_or)
return __scoped_atomic_fetch_or(ptr, mask, order(mem_ord),
scope(mem_scope));
return __scoped_atomic_fetch_or(ptr, mask, impl::order(mem_ord),
impl::scope(mem_scope));
#else
return __atomic_fetch_or(ptr, mask, order(mem_ord));
return __atomic_fetch_or(ptr, mask, impl::order(mem_ord));
#endif
}

Expand All @@ -354,10 +375,10 @@ template <typename T> struct AtomicRef {
[[maybe_unused]] MemoryScope mem_scope = MemoryScope::DEVICE) const {
static_assert(cpp::is_integral_v<T>, "T must be an integral type.");
#if __has_builtin(__scoped_atomic_fetch_and)
return __scoped_atomic_fetch_and(ptr, mask, order(mem_ord),
scope(mem_scope));
return __scoped_atomic_fetch_and(ptr, mask, impl::order(mem_ord),
impl::scope(mem_scope));
#else
return __atomic_fetch_and(ptr, mask, order(mem_ord));
return __atomic_fetch_and(ptr, mask, impl::order(mem_ord));
#endif
}

Expand All @@ -366,10 +387,10 @@ template <typename T> struct AtomicRef {
[[maybe_unused]] MemoryScope mem_scope = MemoryScope::DEVICE) const {
static_assert(cpp::is_integral_v<T>, "T must be an integral type.");
#if __has_builtin(__scoped_atomic_fetch_sub)
return __scoped_atomic_fetch_sub(ptr, decrement, order(mem_ord),
scope(mem_scope));
return __scoped_atomic_fetch_sub(ptr, decrement, impl::order(mem_ord),
impl::scope(mem_scope));
#else
return __atomic_fetch_sub(ptr, decrement, order(mem_ord));
return __atomic_fetch_sub(ptr, decrement, impl::order(mem_ord));
#endif
}
};
Expand Down Expand Up @@ -402,7 +423,6 @@ LIBC_INLINE void atomic_signal_fence([[maybe_unused]] MemoryOrder mem_ord) {
asm volatile("" ::: "memory");
#endif
}

} // namespace cpp
} // namespace LIBC_NAMESPACE_DECL

Expand Down