KOKKOS_INLINE_FUNCTION
void operator()( size_type i ) const
{
const size_type invalid_index = map_type::invalid_index;
uint32_t length = 0;
size_type min_index = ~0u, max_index = 0;
for (size_type curr = m_map.m_hash_lists(i); curr != invalid_index; curr = m_map.m_next_index[curr]) {
++length;
min_index = (curr < min_index) ? curr : min_index;
max_index = (max_index < curr) ? curr : max_index;
}
size_type distance = (0u < length) ? max_index - min_index : 0u;
size_type blocks = (0u < length) ? max_index/32u - min_index/32u : 0u;
// normalize data
length = length < 100u ? length : 99u;
distance = distance < 100u ? distance : 99u;
blocks = blocks < 100u ? blocks : 99u;
if (0u < length)
{
atomic_fetch_add( &m_length(length), 1);
atomic_fetch_add( &m_distance(distance), 1);
atomic_fetch_add( &m_block_distance(blocks), 1);
}
}
void
test_add ()
{
v = 0;
count = 1;
atomic_fetch_add (&v, count);
if (v != 1)
abort ();
atomic_fetch_add_explicit (&v, count, memory_order_consume);
if (v != 2)
abort ();
atomic_fetch_add (&v, 1);
if (v != 3)
abort ();
atomic_fetch_add_explicit (&v, 1, memory_order_release);
if (v != 4)
abort ();
atomic_fetch_add (&v, 1);
if (v != 5)
abort ();
atomic_fetch_add_explicit (&v, count, memory_order_seq_cst);
if (v != 6)
abort ();
}
/*******************************************************************************
* Run tests
******************************************************************************/
int main(int argc, char **argv)
{
char *test_name = "---gnu_gcc_atomics---";
char *csv_header = ":functionName: runtime(ns)";
char *csv_path = "./builtin_atomic_tests.csv"; // Touch to create
ct_info_t *info;
uint64_t iters, start_time, stop_time; // ~4*10^6 iterations
info = ct_init_info(test_name, csv_path);
if (argc != 2) {
iters = 1 << 22;
} else {
iters = atoi(argv[1]);
}
ct_add_line(info, "IterationsOfEachFunction:%lu\n", iters);
ct_add_line(info, "%-s\n", csv_header, FIELD_WIDTH);
RUN_TIMED_TEST("C_postfix_increment", POST_FIX_INCREMENT, iters);
RUN_TIMED_TEST("atomic_init", INIT, iters);
RUN_TIMED_TEST("atomic_thread_fence", THREAD_FENCE, iters);
RUN_TIMED_TEST("atomic_signal_fence", SIGNAL_FENCE, iters);
RUN_TIMED_TEST("atomic_is_lock_free", IS_LOCK_FREE, iters);
RUN_TIMED_TEST("atomic_store_explicit", STORE_EXPLICIT, iters);
RUN_TIMED_TEST("atomic_store", STORE, iters);
RUN_TIMED_TEST("atomic_load_explicit", LOAD_EXPLICIT, iters);
RUN_TIMED_TEST("atomic_load", LOAD, iters);
RUN_TIMED_TEST("atomic_exchange_explicit", EXCHANGE_EXPLICIT, iters);
RUN_TIMED_TEST("atomic_exchange", EXCHANGE, iters);
/* RUN_TIMED_TEST("atomic_compare_exchange_strong_explicit", */
/* COMPARE_EXCHANGE_STRONG_EXPLICIT, iters); */
/* RUN_TIMED_TEST("atomic_compare_exchange_strong", */
/* COMPARE_EXCHANGE_STRONG, iters); */
/* RUN_TIMED_TEST("atomic_compare_exchange_weak_explicit", */
/* COMPARE_EXCHANGE_WEAK_EXPLICIT, iters); */
/* RUN_TIMED_TEST("atomic_compare_exchange_weak", COMPARE_EXCHANGE_WEAK, */
/* iters); */
RUN_TIMED_TEST("atomic_fetch_add", FETCH_ADD, iters);
RUN_TIMED_TEST("atomic_fetch_add_explicit", FETCH_ADD_EXPLICIT, iters);
RUN_TIMED_TEST("atomic_fetch_sub", FETCH_SUB, iters);
RUN_TIMED_TEST("atomic_fetch_sub_explicit", FETCH_SUB_EXPLICIT, iters);
RUN_TIMED_TEST("atomic_fetch_or", FETCH_OR, iters);
RUN_TIMED_TEST("atomic_fetch_or_explicit", FETCH_OR_EXPLICIT, iters);
RUN_TIMED_TEST("atomic_fetch_xor", FETCH_XOR, iters);
RUN_TIMED_TEST("atomic_fetch_xor_explicit", FETCH_XOR_EXPLICIT, iters);
RUN_TIMED_TEST("atomic_fetch_and", FETCH_AND, iters);
RUN_TIMED_TEST("atomic_fetch_and_explicit", FETCH_AND_EXPLICIT, iters);
RUN_TIMED_TEST("atomic_flag_test_and_set", FLAG_TEST_AND_SET, iters);
RUN_TIMED_TEST("atomic_flag_test_and_set_explicit",
FLAG_TEST_AND_SET_EXPLICIT, iters);
RUN_TIMED_TEST("atomic_flag_clear", FLAG_CLEAR, iters);
RUN_TIMED_TEST("atomic_flag_clear_explicit", FLAG_CLEAR_EXPLICIT, iters);
atomic_init(&dummy_atomic_var, atomic_fetch_add(&atomic_var, 0));
printf("dummy_atomic_var = %d\n", atomic_fetch_add(&dummy_atomic_var, 0));
ct_fini_info(info);
}
/*****************************************************************************
* aout_DecPlay : filter & mix the decoded buffer
*****************************************************************************/
int aout_DecPlay (audio_output_t *aout, block_t *block, int input_rate)
{
aout_owner_t *owner = aout_owner (aout);
assert (input_rate >= INPUT_RATE_DEFAULT / AOUT_MAX_INPUT_RATE);
assert (input_rate <= INPUT_RATE_DEFAULT * AOUT_MAX_INPUT_RATE);
assert (block->i_pts >= VLC_TS_0);
block->i_length = CLOCK_FREQ * block->i_nb_samples
/ owner->input_format.i_rate;
aout_OutputLock (aout);
int ret = aout_CheckReady (aout);
if (unlikely(ret == AOUT_DEC_FAILED))
goto drop; /* Pipeline is unrecoverably broken :-( */
const mtime_t now = mdate (), advance = block->i_pts - now;
if (advance < -AOUT_MAX_PTS_DELAY)
{ /* Late buffer can be caused by bugs in the decoder, by scheduling
* latency spikes (excessive load, SIGSTOP, etc.) or if buffering is
* insufficient. We assume the PTS is wrong and play the buffer anyway:
* Hopefully video has encountered a similar PTS problem as audio. */
msg_Warn (aout, "buffer too late (%"PRId64" us): dropped", advance);
goto drop;
}
if (advance > AOUT_MAX_ADVANCE_TIME)
{ /* Early buffers can only be caused by bugs in the decoder. */
msg_Err (aout, "buffer too early (%"PRId64" us): dropped", advance);
goto drop;
}
if (block->i_flags & BLOCK_FLAG_DISCONTINUITY)
owner->sync.discontinuity = true;
block = aout_FiltersPlay (owner->filters, block, input_rate);
if (block == NULL)
goto lost;
/* Software volume */
aout_volume_Amplify (owner->volume, block);
/* Drift correction */
aout_DecSynchronize (aout, block->i_pts, input_rate);
/* Output */
owner->sync.end = block->i_pts + block->i_length + 1;
owner->sync.discontinuity = false;
aout_OutputPlay (aout, block);
atomic_fetch_add(&owner->buffers_played, 1);
out:
aout_OutputUnlock (aout);
return ret;
drop:
owner->sync.discontinuity = true;
block_Release (block);
lost:
atomic_fetch_add(&owner->buffers_lost, 1);
goto out;
}
// Reserves a free slot in ReserveBMap
qitem* queue_get_item(queue *q)
{
int i;
qitem *buf = (qitem*)(q->buf + q->mapbytes*2);
atomic_llong *map = (atomic_llong*)q->buf;
int zbit;
qitem *r;
while(1)
{
// Reserve space
for (i = 0; i < q->map_elements; i++)
{
long long int mval = atomic_load(&map[i]);
long long int nval;
// if no zero bit go to next element
if (!(zbit = ffz(mval)))
{
continue;
}
nval = mval | (((long long int)1) << (--zbit));
// update map val with bit set.
// If successful we are done.
if (atomic_compare_exchange_strong(&map[i], &mval, nval))
{
// printf("ZBIT %d %lld %lld\n",zbit,mval, sizeof(mval));
atomic_fetch_add(&q->size, 1);
break;
}
// Unable to exchange, go again for same index.
else
{
atomic_fetch_add(&q->ct, 1);
i--;
}
}
if (i < q->map_elements)
{
r = &buf[i*64+zbit];
break;
}
else
{
usleep(100);
}
}
return r;
}
static void *Worker( void *arg ) {
TYPE id = (size_t)arg;
uint64_t entry;
#ifdef FAST
unsigned int cnt = 0, oid = id;
#endif // FAST
for ( int r = 0; r < RUNS; r += 1 ) {
entry = 0;
while ( atomic_load(&stop) == 0 ) {
atomic_store(&states[id*PADRATIO], LOCKED);
while (1) {
int lturn = atomic_load(&turn);
if (!validate_left(id, lturn)) {
atomic_store(&states[id*PADRATIO], WAITING);
while (1) {
if (validate_left(id, lturn) && lturn == atomic_load_explicit(&turn, memory_order_acquire)) break;
Pause();
lturn = atomic_load_explicit(&turn, memory_order_acquire);
}
atomic_store(&states[id*PADRATIO], LOCKED);
continue;
}
while (lturn == atomic_load_explicit(&turn, memory_order_acquire)) {
if (validate_right(id, lturn)) break;
Pause();
}
if (lturn == atomic_load_explicit(&turn, memory_order_acquire)) break;
}
CriticalSection( id ); // critical section
int lturn = (atomic_load_explicit(&turn, memory_order_relaxed)+1) % N;
atomic_store_explicit(&turn, lturn, memory_order_relaxed);
atomic_store_explicit(&states[id*PADRATIO], UNLOCKED, memory_order_release); // exit protocol
#ifdef FAST
id = startpoint( cnt ); // different starting point each experiment
cnt = cycleUp( cnt, NoStartPoints );
#endif // FAST
entry += 1;
} // while
#ifdef FAST
id = oid;
#endif // FAST
entries[r][id] = entry;
atomic_fetch_add( &Arrived, 1 );
while ( atomic_load(&stop) != 0 ) Pause();
atomic_fetch_add( &Arrived, -1 );
} // for
return NULL;
} // Worker
static void ffmmal_stop_decoder(AVCodecContext *avctx)
{
MMALDecodeContext *ctx = avctx->priv_data;
MMAL_COMPONENT_T *decoder = ctx->decoder;
MMAL_BUFFER_HEADER_T *buffer;
mmal_port_disable(decoder->input[0]);
mmal_port_disable(decoder->output[0]);
mmal_port_disable(decoder->control);
mmal_port_flush(decoder->input[0]);
mmal_port_flush(decoder->output[0]);
mmal_port_flush(decoder->control);
while ((buffer = mmal_queue_get(ctx->queue_decoded_frames)))
mmal_buffer_header_release(buffer);
while (ctx->waiting_buffers) {
FFBufferEntry *buffer = ctx->waiting_buffers;
ctx->waiting_buffers = buffer->next;
if (buffer->flags & MMAL_BUFFER_HEADER_FLAG_FRAME_END)
atomic_fetch_add(&ctx->packets_buffered, -1);
av_buffer_unref(&buffer->ref);
av_free(buffer);
}
ctx->waiting_buffers_tail = NULL;
av_assert0(atomic_load(&ctx->packets_buffered) == 0);
ctx->frames_output = ctx->eos_received = ctx->eos_sent = ctx->packets_sent = ctx->extradata_sent = 0;
}
void mm_physical_increment_count(addr_t page)
{
if (!frames)
return;
if ((page / PAGE_SIZE) < maximum_page_number)
return atomic_fetch_add(&frames[page / PAGE_SIZE].count, 1);
}
ring_buffer_size_t PaUtil_AdvanceRingBufferReadIndex( PaUtilRingBuffer *rbuf, ring_buffer_size_t elementCount )
{
/* ensure that previous reads (copies out of the ring buffer) are always completed before updating (writing) the read index.
(write-after-read) => full barrier
*/
return (atomic_fetch_add(&rbuf->readIndex,elementCount) +elementCount) & rbuf->bigMask;
}
RatResult
rat_pool_put(RatPool* pool, RatFunction function, void* arg)
{
RAT_ASSERT(pool);
RAT_ASSERT(function);
RatResult result;
// acquire the pool lock
result = rat_lock_acquire(pool->lock);
RAT_ASSERT(result == RAT_RESULT_SUCCESS);
// put the work in the pool list
atomic_fetch_add(&pool->total_work, 1);
RatResult put_result = rat_work_put(
&pool->work,
&pool->free_work,
function,
arg
);
// wake the pool thread
if (put_result == RAT_RESULT_SUCCESS)
{
result = rat_condition_signal(pool->wake_condition);
RAT_ASSERT(result == RAT_RESULT_SUCCESS);
}
// release the pool lock
result = rat_lock_release(pool->lock);
RAT_ASSERT(result == RAT_RESULT_SUCCESS);
return put_result;
}
input_item_t *input_item_Hold( input_item_t *p_item )
{
input_item_owner_t *owner = item_owner(p_item);
atomic_fetch_add( &owner->refs, (atomic_uint)1 ); // sunqueen modify
return p_item;
}
ring_buffer_size_t PaUtil_AdvanceRingBufferWriteIndex( PaUtilRingBuffer *rbuf, ring_buffer_size_t elementCount )
{
/* ensure that previous writes are seen before we update the write index
(write after write)
*/
return (atomic_fetch_add ( &rbuf->writeIndex,elementCount) + elementCount) & rbuf->bigMask;;
}
struct udev *
_udev_ref(struct udev *udev)
{
atomic_fetch_add(&udev->refcount, 1);
return udev;
}
// Sets UsedBMap to signal consumer it can be used.
int queue_push(queue *q, qitem* item)
{
qitem *buf = (qitem*)(q->buf + q->mapbytes*2);
atomic_llong *map = (atomic_llong*)(q->buf + q->mapbytes);
const int diff = (item - buf);
const int i = diff / 64;
const int zbit = diff % 64;
while (1)
{
long long int mval = atomic_load(&map[i]);
long long int nval;
nval = mval | (((long long int)1) << zbit);
if (atomic_compare_exchange_strong(&map[i], &mval, nval))
{
// printf("PUSHING ON POS i=%d zbit=%d diff=%d rdiff=%lld\n",i, zbit, diff, item-buf);
break;
}
else
atomic_fetch_add(&q->ct, 1);
usleep(DELAY_BY);
}
return 1;
}
void
test_fetch_add ()
{
v = 0;
count = 1;
if (atomic_fetch_add_explicit (&v, count, memory_order_relaxed) != 0)
abort ();
if (atomic_fetch_add_explicit (&v, 1, memory_order_consume) != 1)
abort ();
if (atomic_fetch_add_explicit (&v, count, memory_order_acquire) != 2)
abort ();
if (atomic_fetch_add_explicit (&v, 1, memory_order_release) != 3)
abort ();
if (atomic_fetch_add_explicit (&v, count, memory_order_acq_rel) != 4)
abort ();
if (atomic_fetch_add_explicit (&v, 1, memory_order_seq_cst) != 5)
abort ();
if (atomic_fetch_add (&v, 1) != 6)
abort ();
}
void cpu_interrupt_irq_entry(struct registers *regs, int int_no)
{
cpu_interrupt_set(0);
atomic_fetch_add_explicit(&interrupt_counts[int_no], 1, memory_order_relaxed);
int already_in_kernel = 0;
if(!current_thread->regs)
current_thread->regs = regs;
else
already_in_kernel = 1;
atomic_fetch_add(¤t_thread->interrupt_level, 1);
/* Now, run through the stage1 handlers, and see if we need any
* stage2 handlers to run later. */
int s1started = timer_start(&interrupt_timers[int_no]);
for(int i = 0; i < MAX_HANDLERS; i++)
{
if(interrupt_handlers[int_no][i].fn)
(interrupt_handlers[int_no][i].fn)(regs, int_no, 0);
}
if(s1started)
timer_stop(&interrupt_timers[int_no]);
cpu_interrupt_set(0);
atomic_fetch_sub(¤t_thread->interrupt_level, 1);
if(!already_in_kernel) {
__setup_signal_handler(regs);
current_thread->regs = 0;
ASSERT(!current_thread || !current_process || (current_process == kernel_process) || current_thread->held_locks == 0 || (current_thread->flags & THREAD_KERNEL));
}
}
/* Called from render callbacks. No lock, wait, and IO here */
void
ca_Render(audio_output_t *p_aout, uint8_t *p_output, size_t i_requested)
{
struct aout_sys_common *p_sys = (struct aout_sys_common *) p_aout->sys;
if (atomic_load_explicit(&p_sys->b_paused, memory_order_relaxed))
{
memset(p_output, 0, i_requested);
return;
}
/* Pull audio from buffer */
int32_t i_available;
void *p_data = TPCircularBufferTail(&p_sys->circular_buffer,
&i_available);
if (i_available < 0)
i_available = 0;
size_t i_tocopy = __MIN(i_requested, (size_t) i_available);
if (i_tocopy > 0)
{
memcpy(p_output, p_data, i_tocopy);
TPCircularBufferConsume(&p_sys->circular_buffer, i_tocopy);
}
/* Pad with 0 */
if (i_requested > i_tocopy)
{
atomic_fetch_add(&p_sys->i_underrun_size, i_requested - i_tocopy);
memset(&p_output[i_tocopy], 0, i_requested - i_tocopy);
}
}
picture_t *picture_pool_Wait(picture_pool_t *pool)
{
unsigned i;
vlc_mutex_lock(&pool->lock);
assert(pool->refs > 0);
while (pool->available == 0)
vlc_cond_wait(&pool->wait, &pool->lock);
i = ffsll(pool->available);
assert(i > 0);
pool->available &= ~(1ULL << (i - 1));
vlc_mutex_unlock(&pool->lock);
picture_t *picture = pool->picture[i - 1];
if (pool->pic_lock != NULL && pool->pic_lock(picture) != 0) {
vlc_mutex_lock(&pool->lock);
pool->available |= 1ULL << (i - 1);
vlc_cond_signal(&pool->wait);
vlc_mutex_unlock(&pool->lock);
return NULL;
}
picture_t *clone = picture_pool_ClonePicture(pool, i - 1);
if (clone != NULL) {
assert(clone->p_next == NULL);
atomic_fetch_add(&pool->refs, 1);
}
return clone;
}
void net_notify_packet_ready(struct net_dev *nd)
{
atomic_fetch_add(&nd->rx_pending, 1);
if(nd->callbacks->poll)
tm_thread_resume(nd->rec_thread.thread);
/* TODO: notify CPU to schedule this process NOW */
}
void picture_pool_Delete(picture_pool_t *pool)
{
for (int i = 0; i < pool->picture_count; i++) {
picture_t *picture = pool->picture[i];
if (pool->master) {
for (int j = 0; j < pool->master->picture_count; j++) {
if (pool->master->picture[j] == picture)
pool->master->picture_reserved[j] = false;
}
} else {
picture_gc_sys_t *gc_sys = picture->gc.p_sys;
assert(!pool->picture_reserved[i]);
/* Restore the original garbage collector */
if (atomic_fetch_add(&picture->gc.refcount, 1) == 0)
{ /* Simple case: the picture is not locked, destroy it now. */
picture->gc.pf_destroy = gc_sys->destroy;
picture->gc.p_sys = gc_sys->destroy_sys;
free(gc_sys);
}
else /* Intricate case: the picture is still locked and the gc
cannot be modified (w/o memory synchronization). */
atomic_store(&gc_sys->zombie, true);
picture_Release(picture);
}
}
free(pool->picture_reserved);
free(pool->picture);
free(pool);
}
picture_t *picture_pool_Get(picture_pool_t *pool)
{
vlc_mutex_lock(&pool->lock);
assert(pool->refs > 0);
for (unsigned i = ffsll(pool->available); i; i = fnsll(pool->available, i))
{
pool->available &= ~(1ULL << (i - 1));
vlc_mutex_unlock(&pool->lock);
picture_t *picture = pool->picture[i - 1];
if (pool->pic_lock != NULL && pool->pic_lock(picture) != 0) {
vlc_mutex_lock(&pool->lock);
pool->available |= 1ULL << (i - 1);
continue;
}
picture_t *clone = picture_pool_ClonePicture(pool, i - 1);
if (clone != NULL) {
assert(clone->p_next == NULL);
atomic_fetch_add(&pool->refs, 1);
}
return clone;
}
vlc_mutex_unlock(&pool->lock);
return NULL;
}
/**
* Create a stream
*
* Allocate and initialize memory for a stream.
*
* @return pointer to the created stream
*/
lpel_stream_t *LpelStreamCreate(int size)
{
assert( size >= 0);
if (0==size) size = STREAM_BUFFER_SIZE;
lpel_stream_t *s;
s = LpelWorkerGetStream(); // try to get from the free list first
if (s == NULL) {
s = (lpel_stream_t *) malloc( sizeof(lpel_stream_t) ); // allocate if fail
LpelBufferInit( &s->buffer, size);
}
assert(LpelBufferIsEmpty(&s->buffer));
s->uid = atomic_fetch_add( &stream_seq, 1);
PRODLOCK_INIT( &s->prod_lock );
atomic_init( &s->n_sem, 0);
atomic_init( &s->e_sem, size);
s->is_poll = 0;
s->prod_sd = NULL;
s->cons_sd = NULL;
s->usr_data = NULL;
s->type = LPEL_STREAM_MIDDLE;
s->next = NULL;
s->read_cnt = 0;
s->write_cnt = 0;
return s;
}
/**
* Blocking, consuming read from a stream
*
* If the stream is empty, the task is suspended until
* a producer writes an item to the stream.
*
* @param sd stream descriptor
* @return the next item of the stream
* @pre current task is single reader
*/
void *LpelStreamRead( lpel_stream_desc_t *sd)
{
void *item;
lpel_task_t *self = sd->task;
assert( sd->mode == 'r');
/* MONITORING CALLBACK */
#ifdef USE_TASK_EVENT_LOGGING
if (sd->mon && MON_CB(stream_readprepare)) {
MON_CB(stream_readprepare)(sd->mon);
}
#endif
/* quasi P(n_sem) */
if ( atomic_fetch_sub( &sd->stream->n_sem, 1) == 0) {
#ifdef USE_TASK_EVENT_LOGGING
/* MONITORING CALLBACK */
if (sd->mon && MON_CB(stream_blockon)) {
MON_CB(stream_blockon)(sd->mon);
}
#endif
/* wait on stream: */
LpelTaskBlockStream( self);
}
/* read the top element */
item = LpelBufferTop( &sd->stream->buffer);
assert( item != NULL);
/* pop off the top element */
LpelBufferPop( &sd->stream->buffer);
/* quasi V(e_sem) */
if ( atomic_fetch_add( &sd->stream->e_sem, 1) < 0) {
/* e_sem was -1 */
lpel_task_t *prod = sd->stream->prod_sd->task;
/* wakeup producer: make ready */
LpelTaskUnblock( self, prod);
/* MONITORING CALLBACK */
#ifdef USE_TASK_EVENT_LOGGING
if (sd->mon && MON_CB(stream_wakeup)) {
MON_CB(stream_wakeup)(sd->mon);
}
#endif
}
/* MONITORING CALLBACK */
#ifdef USE_TASK_EVENT_LOGGING
if (sd->mon && MON_CB(stream_readfinish)) {
MON_CB(stream_readfinish)(sd->mon, item);
}
#endif
return item;
}
void report(std::thread::id tid, const char* format, ...) {
Report report;
report.time = chrono::duration_cast<chrono::duration<double>>(
chrono::high_resolution_clock::now() - logStartTime).count();
if (queueThreadMap.find(tid) == queueThreadMap.end()) {
// this makes n and queueThreadMap thread-safe
//
queueThreadMapMutex.lock();
queueThreadMap[tid] = atomic_fetch_add(&n, 1);
queueThreadMapMutex.unlock();
assert(n < NUMBER_OF_QUEUES);
}
int q = queueThreadMap[tid];
// XXX bug here! user can give long strings that are not handled!
char buffer[256];
va_list args;
va_start(args, format);
vsprintf(buffer, format, args);
// perror(buffer);
va_end(args);
sprintf(report.text, "%.8lf|%d|%s", report.time, q, buffer);
report.text[REPORT_LENGTH - 1] = '\0';
queue[q].push(report);
}
input_item_t *input_item_Hold( input_item_t *p_item )
{
input_item_owner_t *owner = item_owner(p_item);
atomic_fetch_add( &owner->refs, 1 );
return p_item;
}
/*
* This function just provide calculated blocktime per cpu and trace it.
* Total blocktime is calculated in mark_postcopy_blocktime_end.
*
*
* Assume we have 3 CPU
*
* S1 E1 S1 E1
* -----***********------------xxx***************------------------------> CPU1
*
* S2 E2
* ------------****************xxx---------------------------------------> CPU2
*
* S3 E3
* ------------------------****xxx********-------------------------------> CPU3
*
* We have sequence S1,S2,E1,S3,S1,E2,E3,E1
* S2,E1 - doesn't match condition due to sequence S1,S2,E1 doesn't include CPU3
* S3,S1,E2 - sequence includes all CPUs, in this case overlap will be S1,E2 -
* it's a part of total blocktime.
* S1 - here is last_begin
* Legend of the picture is following:
* * - means blocktime per vCPU
* x - means overlapped blocktime (total blocktime)
*
* @addr: host virtual address
*/
static void mark_postcopy_blocktime_end(uintptr_t addr)
{
MigrationIncomingState *mis = migration_incoming_get_current();
PostcopyBlocktimeContext *dc = mis->blocktime_ctx;
int i, affected_cpu = 0;
bool vcpu_total_blocktime = false;
uint32_t read_vcpu_time, low_time_offset;
if (!dc) {
return;
}
low_time_offset = get_low_time_offset(dc);
/* lookup cpu, to clear it,
* that algorithm looks straighforward, but it's not
* optimal, more optimal algorithm is keeping tree or hash
* where key is address value is a list of */
for (i = 0; i < smp_cpus; i++) {
uint32_t vcpu_blocktime = 0;
read_vcpu_time = atomic_fetch_add(&dc->page_fault_vcpu_time[i], 0);
if (atomic_fetch_add(&dc->vcpu_addr[i], 0) != addr ||
read_vcpu_time == 0) {
continue;
}
atomic_xchg(&dc->vcpu_addr[i], 0);
vcpu_blocktime = low_time_offset - read_vcpu_time;
affected_cpu += 1;
/* we need to know is that mark_postcopy_end was due to
* faulted page, another possible case it's prefetched
* page and in that case we shouldn't be here */
if (!vcpu_total_blocktime &&
atomic_fetch_add(&dc->smp_cpus_down, 0) == smp_cpus) {
vcpu_total_blocktime = true;
}
/* continue cycle, due to one page could affect several vCPUs */
dc->vcpu_blocktime[i] += vcpu_blocktime;
}
atomic_sub(&dc->smp_cpus_down, affected_cpu);
if (vcpu_total_blocktime) {
dc->total_blocktime += low_time_offset - atomic_fetch_add(
&dc->last_begin, 0);
}
trace_mark_postcopy_blocktime_end(addr, dc, dc->total_blocktime,
affected_cpu);
}
static int atomic_hash_table_update_one_file(english_word *word){
uint64_t hashv=fnv_hash(word->str,word->len);
uint64_t index=hashv%global_hash_table_size;
//this next line results in a lot of cache misses
//for obvious reasons
if(!global_hash_table[index]){//word isn't in the hash table, add it
uint8_t *mem=xmalloc(word->len);
word->str=(char*)my_strcpy(mem,(uint8_t*)word->str,word->len);
void *prev=global_hash_table[index];
int test=atomic_compare_exchange_n(global_hash_table+index,&prev,word);
if(test){
//we added the word
//this needs to be atomic to prevent two threads writing different
//values to the same index of indices
uint64_t old_indices_index=atomic_fetch_add(&indices_index,1);
//this doesn't need to be atomic, since indices_index will never be
//decremented, so no one else will change this
hash_table_indices[old_indices_index]=index;
goto end1;
}
//else, someone else changed the value of global_hash_table[index] before us
}
while(1){
do {
//see if the value in the table is the same as our value
//if so update the value already in the table
if(string_compare(global_hash_table[index],word)){
//atomically increment word count
atomic_add(&global_hash_table[index]->count,1);
goto end0;
}
} while(global_hash_table[++index]);
//not in the table use next free index (if we can)
void *prev=global_hash_table[index];
int test=atomic_compare_exchange_n(global_hash_table+index,&prev,word);
if(test){
uint64_t old_indices_index=atomic_fetch_add(&indices_index,1);
hash_table_indices[old_indices_index]=index;
goto end1;
}
//if !test the compare exchange failed and we need to keep looping
}
end0:
return 0;
end1:
return 1;
}
/*
* Locks the mutex
* Progress Condition: Blocking
*/
void ticket_mutex_lock(ticket_mutex_t * self)
{
long lingress = atomic_fetch_add(&self->ingress, 1);
while (lingress != atomic_load(&self->egress)) {
sched_yield(); // Replace this with thrd_yield() if you use <threads.h>
}
// This thread has acquired the lock on the mutex
}
/* increase a refcount on an inode which is already owned (has a non-zero refcount). */
void vfs_inode_get(struct inode *node)
{
mutex_acquire(ic_lock);
atomic_fetch_add(&node->count, 1);
assert(node->count > 1);
assert((node->flags & INODE_INUSE));
mutex_release(ic_lock);
}
void writer(void) // Writes positive values; ends with a negative value.
{
const int N = 20; // Number of data values to write.
for( int n = 0; n <= N; ++n)
{
int d = n < N ? 10+n : -1; // Prepare data or end marker.
// When no readers are busy, lock the semaphore (count = -1):
while( atomic_fetch_sub(&count,MAX_READERS+1) != MAX_READERS)
atomic_fetch_add(&count, MAX_READERS+1);
printf("Writer is writing %d\n", d), // Critical section.
data = d;
atomic_fetch_add(&count, MAX_READERS+1); // Release the
// semaphores.
thrd_sleep(&ms,NULL); // Simulate data production.
}
}
