void structAtomicLoad() {
struct foo f = __c11_atomic_load(&bigAtomic, 5); // expected-error {{atomic load requires runtime support that is not available for this target}}
struct bar b;
__atomic_load(&smallThing, &b, 5);
__atomic_load(&bigThing, &f, 5);
}
IODataQueueEntry * IOSharedDataQueue::peek()
{
IODataQueueEntry *entry = 0;
UInt32 headOffset;
UInt32 tailOffset;
if (!dataQueue) {
return NULL;
}
// Read head and tail with acquire barrier
// See rdar://problem/40780584 for an explanation of relaxed/acquire barriers
headOffset = __c11_atomic_load((_Atomic UInt32 *)&dataQueue->head, __ATOMIC_RELAXED);
tailOffset = __c11_atomic_load((_Atomic UInt32 *)&dataQueue->tail, __ATOMIC_ACQUIRE);
if (headOffset != tailOffset) {
IODataQueueEntry * head = 0;
UInt32 headSize = 0;
UInt32 headOffset = dataQueue->head;
UInt32 queueSize = getQueueSize();
if (headOffset >= queueSize) {
return NULL;
}
head = (IODataQueueEntry *)((char *)dataQueue->queue + headOffset);
headSize = head->size;
// Check if there's enough room before the end of the queue for a header.
// If there is room, check if there's enough room to hold the header and
// the data.
if ((headOffset > UINT32_MAX - DATA_QUEUE_ENTRY_HEADER_SIZE) ||
(headOffset + DATA_QUEUE_ENTRY_HEADER_SIZE > queueSize) ||
(headOffset + DATA_QUEUE_ENTRY_HEADER_SIZE > UINT32_MAX - headSize) ||
(headOffset + headSize + DATA_QUEUE_ENTRY_HEADER_SIZE > queueSize)) {
// No room for the header or the data, wrap to the beginning of the queue.
// Note: wrapping even with the UINT32_MAX checks, as we have to support
// queueSize of UINT32_MAX
entry = dataQueue->queue;
} else {
entry = head;
}
}
return entry;
}
__host__ __device__
typename enable_if<
sizeof(Integer64) == 8,
Integer64
>::type
atomic_load(const Integer64 *x)
{
#if defined(__CUDA_ARCH__)
return atomicAdd(const_cast<Integer64*>(x), Integer64(0));
#elif defined(__GNUC__)
return atomic_load_n(x, __ATOMIC_SEQ_CST);
#elif defined(_MSC_VER)
return InterlockedExchangeAdd(x, Integer64(0));
#elif defined(__clang__)
return __c11_atomic_load(x);
#else
#error "No atomic_load_n implementation."
#endif
}
long use_count() const // nothrow
{
return __c11_atomic_load( const_cast< atomic_int_least32_t* >( &use_count_ ), __ATOMIC_ACQUIRE );
}
Boolean IOSharedDataQueue::dequeue(void *data, UInt32 *dataSize)
{
Boolean retVal = TRUE;
IODataQueueEntry * entry = 0;
UInt32 entrySize = 0;
UInt32 headOffset = 0;
UInt32 tailOffset = 0;
UInt32 newHeadOffset = 0;
if (!dataQueue || (data && !dataSize)) {
return false;
}
// Read head and tail with acquire barrier
// See rdar://problem/40780584 for an explanation of relaxed/acquire barriers
headOffset = __c11_atomic_load((_Atomic UInt32 *)&dataQueue->head, __ATOMIC_RELAXED);
tailOffset = __c11_atomic_load((_Atomic UInt32 *)&dataQueue->tail, __ATOMIC_ACQUIRE);
if (headOffset != tailOffset) {
IODataQueueEntry * head = 0;
UInt32 headSize = 0;
UInt32 queueSize = getQueueSize();
if (headOffset > queueSize) {
return false;
}
head = (IODataQueueEntry *)((char *)dataQueue->queue + headOffset);
headSize = head->size;
// we wrapped around to beginning, so read from there
// either there was not even room for the header
if ((headOffset > UINT32_MAX - DATA_QUEUE_ENTRY_HEADER_SIZE) ||
(headOffset + DATA_QUEUE_ENTRY_HEADER_SIZE > queueSize) ||
// or there was room for the header, but not for the data
(headOffset + DATA_QUEUE_ENTRY_HEADER_SIZE > UINT32_MAX - headSize) ||
(headOffset + headSize + DATA_QUEUE_ENTRY_HEADER_SIZE > queueSize)) {
// Note: we have to wrap to the beginning even with the UINT32_MAX checks
// because we have to support a queueSize of UINT32_MAX.
entry = dataQueue->queue;
entrySize = entry->size;
if ((entrySize > UINT32_MAX - DATA_QUEUE_ENTRY_HEADER_SIZE) ||
(entrySize + DATA_QUEUE_ENTRY_HEADER_SIZE > queueSize)) {
return false;
}
newHeadOffset = entrySize + DATA_QUEUE_ENTRY_HEADER_SIZE;
// else it is at the end
} else {
entry = head;
entrySize = entry->size;
if ((entrySize > UINT32_MAX - DATA_QUEUE_ENTRY_HEADER_SIZE) ||
(entrySize + DATA_QUEUE_ENTRY_HEADER_SIZE > UINT32_MAX - headOffset) ||
(entrySize + DATA_QUEUE_ENTRY_HEADER_SIZE + headOffset > queueSize)) {
return false;
}
newHeadOffset = headOffset + entrySize + DATA_QUEUE_ENTRY_HEADER_SIZE;
}
} else {
// empty queue
return false;
}
if (data) {
if (entrySize > *dataSize) {
// not enough space
return false;
}
memcpy(data, &(entry->data), entrySize);
*dataSize = entrySize;
}
__c11_atomic_store((_Atomic UInt32 *)&dataQueue->head, newHeadOffset, __ATOMIC_RELEASE);
if (newHeadOffset == tailOffset) {
//
// If we are making the queue empty, then we need to make sure
// that either the enqueuer notices, or we notice the enqueue
// that raced with our making of the queue empty.
//
__c11_atomic_thread_fence(__ATOMIC_SEQ_CST);
}
return retVal;
}
Boolean IOSharedDataQueue::enqueue(void * data, UInt32 dataSize)
{
UInt32 head;
UInt32 tail;
UInt32 newTail;
const UInt32 entrySize = dataSize + DATA_QUEUE_ENTRY_HEADER_SIZE;
IODataQueueEntry * entry;
// Force a single read of head and tail
// See rdar://problem/40780584 for an explanation of relaxed/acquire barriers
tail = __c11_atomic_load((_Atomic UInt32 *)&dataQueue->tail, __ATOMIC_RELAXED);
head = __c11_atomic_load((_Atomic UInt32 *)&dataQueue->head, __ATOMIC_ACQUIRE);
// Check for overflow of entrySize
if (dataSize > UINT32_MAX - DATA_QUEUE_ENTRY_HEADER_SIZE) {
return false;
}
// Check for underflow of (getQueueSize() - tail)
if (getQueueSize() < tail || getQueueSize() < head) {
return false;
}
if ( tail >= head )
{
// Is there enough room at the end for the entry?
if ((entrySize <= UINT32_MAX - tail) &&
((tail + entrySize) <= getQueueSize()) )
{
entry = (IODataQueueEntry *)((UInt8 *)dataQueue->queue + tail);
entry->size = dataSize;
memcpy(&entry->data, data, dataSize);
// The tail can be out of bound when the size of the new entry
// exactly matches the available space at the end of the queue.
// The tail can range from 0 to dataQueue->queueSize inclusive.
newTail = tail + entrySize;
}
else if ( head > entrySize ) // Is there enough room at the beginning?
{
// Wrap around to the beginning, but do not allow the tail to catch
// up to the head.
dataQueue->queue->size = dataSize;
// We need to make sure that there is enough room to set the size before
// doing this. The user client checks for this and will look for the size
// at the beginning if there isn't room for it at the end.
if ( ( getQueueSize() - tail ) >= DATA_QUEUE_ENTRY_HEADER_SIZE )
{
((IODataQueueEntry *)((UInt8 *)dataQueue->queue + tail))->size = dataSize;
}
memcpy(&dataQueue->queue->data, data, dataSize);
newTail = entrySize;
}
else
{
return false; // queue is full
}
}
else
{
// Do not allow the tail to catch up to the head when the queue is full.
// That's why the comparison uses a '>' rather than '>='.
if ( (head - tail) > entrySize )
{
entry = (IODataQueueEntry *)((UInt8 *)dataQueue->queue + tail);
entry->size = dataSize;
memcpy(&entry->data, data, dataSize);
newTail = tail + entrySize;
}
else
{
return false; // queue is full
}
}
// Publish the data we just enqueued
__c11_atomic_store((_Atomic UInt32 *)&dataQueue->tail, newTail, __ATOMIC_RELEASE);
if (tail != head) {
//
// The memory barrier below paris with the one in ::dequeue
// so that either our store to the tail cannot be missed by
// the next dequeue attempt, or we will observe the dequeuer
// making the queue empty.
//
// Of course, if we already think the queue is empty,
// there's no point paying this extra cost.
//
__c11_atomic_thread_fence(__ATOMIC_SEQ_CST);
head = __c11_atomic_load((_Atomic UInt32 *)&dataQueue->head, __ATOMIC_RELAXED);
}
if (tail == head) {
// Send notification (via mach message) that data is now available.
sendDataAvailableNotification();
}
return true;
}
// CHECK: void test13() {
// CHECK: _Atomic(int) i;
// CHECK: __c11_atomic_init(&i, 0);
// CHECK: __c11_atomic_load(&i, 0);
// CHECK: }
void test13() {
_Atomic(int) i;
__c11_atomic_init(&i, 0);
__c11_atomic_load(&i, 0);
}
