void libbenchmark_benchmark_queue_umm_liblfds700_lockfree_enqueue1_dequeue1_cleanup( struct lfds710_list_aso_state *logical_processor_set,
enum libbenchmark_topology_numa_mode numa_mode,
struct libbenchmark_results_state *rs,
struct libbenchmark_threadset_state *tsets )
{
struct lfds710_list_asu_element
*lasue = NULL;
struct libbenchmark_benchmark_queue_umm_liblfds700_lockfree_enqueue1_dequeue1_overall_benchmark_state
*obs;
struct libbenchmark_benchmark_queue_umm_liblfds700_lockfree_enqueue1_dequeue1_per_thread_benchmark_state
*ptbs;
struct libbenchmark_threadset_per_thread_state
*pts;
LFDS710_PAL_ASSERT( logical_processor_set != NULL );
// TRD : numa_mode can be any value in its range
LFDS710_PAL_ASSERT( rs != NULL );
LFDS710_PAL_ASSERT( tsets != NULL );
while( LFDS710_LIST_ASU_GET_START_AND_THEN_NEXT(tsets->list_of_per_thread_states,lasue) )
{
pts = LFDS710_LIST_ASU_GET_VALUE_FROM_ELEMENT( *lasue );
ptbs = LIBBENCHMARK_THREADSET_PER_THREAD_STATE_GET_USERS_PER_THREAD_STATE( *pts );
libbenchmark_results_put_result( rs,
LIBBENCHMARK_DATASTRUCTURE_ID_QUEUE_UMM,
LIBBENCHMARK_BENCHMARK_ID_ENQUEUE_UMM1_THEN_DEQUEUE_UMM1,
LIBBENCHMARK_LOCK_ID_LIBLFDS700_LOCKFREE,
numa_mode,
logical_processor_set,
LIBBENCHMARK_TOPOLOGY_NODE_GET_LOGICAL_PROCESSOR_NUMBER( *pts->tns_lp ),
LIBBENCHMARK_TOPOLOGY_NODE_GET_WINDOWS_GROUP_NUMBER( *pts->tns_lp ),
ptbs->operation_count );
}
obs = tsets->users_threadset_state;
lfds700_queue_cleanup( obs->qs, NULL );
return;
}
void test_lfds700_queue_enqueuing( struct lfds700_list_asu_state *list_of_logical_processors, lfds700_pal_uint_t memory_in_megabytes )
{
enum lfds700_misc_validity
dvs = LFDS700_MISC_VALIDITY_VALID;
lfds700_pal_uint_t
*per_thread_counters,
loop,
number_elements,
number_logical_processors;
struct lfds700_list_asu_element
*lasue;
struct lfds700_misc_prng_state
ps;
struct lfds700_queue_element
dummy_qe,
*qe;
struct lfds700_queue_state
qs;
struct lfds700_misc_validation_info
vi;
struct test_pal_logical_processor
*lp;
struct util_thread_starter_state
*tts;
struct test_element
*te;
struct test_state
*ts;
test_pal_thread_state_t
*thread_handles;
assert( list_of_logical_processors != NULL );
// TRD : memory_in_megabytes can be any value in its range
/* TRD : create an empty queue
then run one thread per CPU
where each thread busy-works, enqueuing elements from a freelist (one local freelist per thread)
until 100000 elements are enqueued, per thread
each element's void pointer of user data is a struct containing thread number and element number
where element_number is a thread-local counter starting at 0
when we're done, we check that all the elements are present
and increment on a per-thread basis
*/
internal_display_test_name( "Enqueuing" );
lfds700_list_asu_query( list_of_logical_processors, LFDS700_LIST_ASU_QUERY_GET_POTENTIALLY_INACCURATE_COUNT, NULL, (void **) &number_logical_processors );
lfds700_misc_prng_init( &ps );
number_elements = ( memory_in_megabytes * ONE_MEGABYTE_IN_BYTES ) / ( sizeof(struct test_element) * number_logical_processors );
lfds700_queue_init_valid_on_current_logical_core( &qs, &dummy_qe, &ps, NULL );
ts = util_malloc_wrapper( sizeof(struct test_state) * number_logical_processors );
for( loop = 0 ; loop < number_logical_processors ; loop++ )
{
(ts+loop)->qs = &qs;
(ts+loop)->thread_number = loop;
(ts+loop)->number_elements = number_elements;
}
thread_handles = util_malloc_wrapper( sizeof(test_pal_thread_state_t) * number_logical_processors );
util_thread_starter_new( &tts, number_logical_processors );
LFDS700_MISC_BARRIER_STORE;
lfds700_misc_force_store();
loop = 0;
lasue = NULL;
while( LFDS700_LIST_ASU_GET_START_AND_THEN_NEXT(*list_of_logical_processors, lasue) )
{
lp = LFDS700_LIST_ASU_GET_VALUE_FROM_ELEMENT( *lasue );
util_thread_starter_start( tts, &thread_handles[loop], loop, lp, thread_simple_enqueuer, ts+loop );
loop++;
}
util_thread_starter_run( tts );
for( loop = 0 ; loop < number_logical_processors ; loop++ )
test_pal_thread_wait( thread_handles[loop] );
util_thread_starter_delete( tts );
free( thread_handles );
LFDS700_MISC_BARRIER_LOAD;
/* TRD : first, validate the queue
then dequeue
we expect to find element numbers increment on a per thread basis
*/
vi.min_elements = vi.max_elements = number_elements * number_logical_processors;
lfds700_queue_query( &qs, LFDS700_QUEUE_QUERY_SINGLETHREADED_VALIDATE, &vi, &dvs );
per_thread_counters = util_malloc_wrapper( sizeof(lfds700_pal_uint_t) * number_logical_processors );
for( loop = 0 ; loop < number_logical_processors ; loop++ )
*(per_thread_counters+loop) = 0;
while( dvs == LFDS700_MISC_VALIDITY_VALID and lfds700_queue_dequeue(&qs, &qe, &ps) )
{
te = LFDS700_QUEUE_GET_VALUE_FROM_ELEMENT( *qe );
if( te->thread_number >= number_logical_processors )
{
dvs = LFDS700_MISC_VALIDITY_INVALID_TEST_DATA;
break;
}
if( te->counter > per_thread_counters[te->thread_number] )
dvs = LFDS700_MISC_VALIDITY_INVALID_MISSING_ELEMENTS;
if( te->counter < per_thread_counters[te->thread_number] )
dvs = LFDS700_MISC_VALIDITY_INVALID_ADDITIONAL_ELEMENTS;
if( te->counter == per_thread_counters[te->thread_number] )
per_thread_counters[te->thread_number]++;
}
free( per_thread_counters );
for( loop = 0 ; loop < number_logical_processors ; loop++ )
util_aligned_free( (ts+loop)->te_array );
free( ts );
lfds700_queue_cleanup( &qs, NULL );
internal_display_test_result( 1, "queue", dvs );
return;
}
void test_lfds700_queue_enqueuing_with_malloc_and_dequeuing_with_free( struct lfds700_list_asu_state *list_of_logical_processors )
{
enum lfds700_misc_validity
dvs = LFDS700_MISC_VALIDITY_VALID;
lfds700_pal_uint_t
loop,
number_logical_processors;
struct lfds700_list_asu_element
*lasue;
struct lfds700_misc_prng_state
ps;
struct lfds700_queue_element
*qe;
struct lfds700_queue_state
qs;
struct lfds700_misc_validation_info
vi;
struct test_pal_logical_processor
*lp;
struct util_thread_starter_state
*tts;
struct test_state
*ts;
test_pal_thread_state_t
*thread_handles;
assert( list_of_logical_processors != NULL );
// TRD : qt can be any value in its range
/* TRD : one thread per logical core
each thread loops for ten seconds
mallocs and enqueues 1k elements, then dequeues and frees 1k elements
*/
internal_display_test_name( "Enqueuing with malloc dequeuing with free (%d seconds)", TEST_DURATION_IN_SECONDS );
lfds700_list_asu_query( list_of_logical_processors, LFDS700_LIST_ASU_QUERY_GET_POTENTIALLY_INACCURATE_COUNT, NULL, (void **) &number_logical_processors );
lfds700_misc_prng_init( &ps );
qe = util_aligned_malloc( sizeof(struct lfds700_queue_element), LFDS700_PAL_ATOMIC_ISOLATION_IN_BYTES );
lfds700_queue_init_valid_on_current_logical_core( &qs, qe, &ps, NULL );
ts = util_malloc_wrapper( sizeof(struct test_state) * number_logical_processors );
for( loop = 0 ; loop < number_logical_processors ; loop++ )
(ts+loop)->qs = &qs;
thread_handles = util_malloc_wrapper( sizeof(test_pal_thread_state_t) * number_logical_processors );
util_thread_starter_new( &tts, number_logical_processors );
LFDS700_MISC_BARRIER_STORE;
lfds700_misc_force_store();
loop = 0;
lasue = NULL;
while( LFDS700_LIST_ASU_GET_START_AND_THEN_NEXT(*list_of_logical_processors, lasue) )
{
lp = LFDS700_LIST_ASU_GET_VALUE_FROM_ELEMENT( *lasue );
util_thread_starter_start( tts, &thread_handles[loop], loop, lp, thread_enqueuer_with_malloc_and_dequeuer_with_free, ts+loop );
loop++;
}
util_thread_starter_run( tts );
for( loop = 0 ; loop < number_logical_processors ; loop++ )
test_pal_thread_wait( thread_handles[loop] );
util_thread_starter_delete( tts );
free( thread_handles );
LFDS700_MISC_BARRIER_LOAD;
vi.min_elements = vi.max_elements = 0;
lfds700_queue_query( &qs, LFDS700_QUEUE_QUERY_SINGLETHREADED_VALIDATE, &vi, &dvs );
free( ts );
lfds700_queue_cleanup( &qs, queue_element_cleanup_callback );
internal_display_test_result( 1, "queue", dvs );
return;
}
void test_lfds700_queue_rapid_enqueuing_and_dequeuing( struct lfds700_list_asu_state *list_of_logical_processors, lfds700_pal_uint_t memory_in_megabytes )
{
enum lfds700_misc_validity
dvs = LFDS700_MISC_VALIDITY_VALID;
lfds700_pal_uint_t
loop,
number_elements_with_dummy_element,
number_elements_without_dummy_element,
number_logical_processors,
*per_thread_counters;
struct lfds700_list_asu_element
*lasue;
struct lfds700_misc_prng_state
ps;
struct lfds700_queue_element
*qe;
struct lfds700_misc_validation_info
vi;
struct lfds700_queue_state
qs;
struct test_pal_logical_processor
*lp;
struct util_thread_starter_state
*tts;
struct test_element
*te_array,
*te;
struct test_state
*ts;
test_pal_thread_state_t
*thread_handles;
assert( list_of_logical_processors != NULL );
// TRD : memory_in_megabytes can be any value in its range
/* TRD : we create a single queue with 50,000 elements
we don't want too many elements, so we ensure plenty of element re-use
each thread simply loops dequeuing and enqueuing
where the user data indicates thread number and an increment counter
vertification is that the counter increments on a per-thread basis
*/
internal_display_test_name( "Rapid enqueuing and dequeuing (%d seconds)", TEST_DURATION_IN_SECONDS );
lfds700_list_asu_query( list_of_logical_processors, LFDS700_LIST_ASU_QUERY_GET_POTENTIALLY_INACCURATE_COUNT, NULL, (void **) &number_logical_processors );
lfds700_misc_prng_init( &ps );
number_elements_with_dummy_element = ( memory_in_megabytes * ONE_MEGABYTE_IN_BYTES ) / sizeof(struct test_element);
if( number_elements_with_dummy_element > (10000 * number_logical_processors) + 1 )
number_elements_with_dummy_element = (10000 * number_logical_processors) + 1;
number_elements_without_dummy_element = number_elements_with_dummy_element - 1;
vi.min_elements = number_elements_without_dummy_element;
vi.max_elements = number_elements_without_dummy_element;
te_array = util_aligned_malloc( sizeof(struct test_element) * number_elements_with_dummy_element, LFDS700_PAL_ATOMIC_ISOLATION_IN_BYTES );
lfds700_queue_init_valid_on_current_logical_core( &qs, &(te_array+number_elements_without_dummy_element)->qe, &ps, NULL );
// TRD : we assume the test will iterate at least once (or we'll have a false negative)
for( loop = 0 ; loop < number_elements_without_dummy_element ; loop++ )
{
(te_array+loop)->thread_number = loop;
(te_array+loop)->counter = 0;
LFDS700_QUEUE_SET_VALUE_IN_ELEMENT( (te_array+loop)->qe, te_array+loop );
lfds700_queue_enqueue( &qs, &(te_array+loop)->qe, &ps );
}
ts = util_malloc_wrapper( sizeof(struct test_state) * number_logical_processors );
for( loop = 0 ; loop < number_logical_processors ; loop++ )
{
(ts+loop)->qs = &qs;
(ts+loop)->thread_number = loop;
(ts+loop)->counter = 0;
}
thread_handles = util_malloc_wrapper( sizeof(test_pal_thread_state_t) * number_logical_processors );
util_thread_starter_new( &tts, number_logical_processors );
LFDS700_MISC_BARRIER_STORE;
lfds700_misc_force_store();
loop = 0;
lasue = NULL;
while( LFDS700_LIST_ASU_GET_START_AND_THEN_NEXT(*list_of_logical_processors, lasue) )
{
lp = LFDS700_LIST_ASU_GET_VALUE_FROM_ELEMENT( *lasue );
util_thread_starter_start( tts, &thread_handles[loop], loop, lp, thread_rapid_enqueuer_and_dequeuer, ts+loop );
loop++;
}
util_thread_starter_run( tts );
for( loop = 0 ; loop < number_logical_processors ; loop++ )
test_pal_thread_wait( thread_handles[loop] );
util_thread_starter_delete( tts );
free( thread_handles );
LFDS700_MISC_BARRIER_LOAD;
lfds700_queue_query( &qs, LFDS700_QUEUE_QUERY_SINGLETHREADED_VALIDATE, &vi, &dvs );
// TRD : now check results
per_thread_counters = util_malloc_wrapper( sizeof(lfds700_pal_uint_t) * number_logical_processors );
for( loop = 0 ; loop < number_logical_processors ; loop++ )
*(per_thread_counters+loop) = 0;
while( dvs == LFDS700_MISC_VALIDITY_VALID and lfds700_queue_dequeue(&qs, &qe, &ps) )
{
te = LFDS700_QUEUE_GET_VALUE_FROM_ELEMENT( *qe );
if( te->thread_number >= number_logical_processors )
{
dvs = LFDS700_MISC_VALIDITY_INVALID_TEST_DATA;
break;
}
if( per_thread_counters[te->thread_number] == 0 )
per_thread_counters[te->thread_number] = te->counter;
if( te->counter > per_thread_counters[te->thread_number] )
dvs = LFDS700_MISC_VALIDITY_INVALID_MISSING_ELEMENTS;
if( te->counter < per_thread_counters[te->thread_number] )
dvs = LFDS700_MISC_VALIDITY_INVALID_ADDITIONAL_ELEMENTS;
if( te->counter == per_thread_counters[te->thread_number] )
per_thread_counters[te->thread_number]++;
}
free( per_thread_counters );
lfds700_queue_cleanup( &qs, NULL );
util_aligned_free( te_array );
free( ts );
internal_display_test_result( 1, "queue", dvs );
return;
}
void test_lfds700_queue_dequeuing( struct lfds700_list_asu_state *list_of_logical_processors, lfds700_pal_uint_t memory_in_megabytes )
{
enum lfds700_misc_validity
dvs = LFDS700_MISC_VALIDITY_VALID;
lfds700_pal_uint_t
loop,
number_elements_with_dummy_element,
number_elements_without_dummy_element,
number_logical_processors;
struct lfds700_list_asu_element
*lasue;
struct lfds700_misc_prng_state
ps;
struct lfds700_queue_state
qs;
struct lfds700_misc_validation_info
vi = { 0, 0 };
struct test_pal_logical_processor
*lp;
struct util_thread_starter_state
*tts;
struct test_element
*te_array;
struct test_state
*ts;
test_pal_thread_state_t
*thread_handles;
assert( list_of_logical_processors != NULL );
// TRD : memory_in_megabytes can be any value in its range
/* TRD : create a queue, add 1,000,000 elements
use a single thread to enqueue every element
each elements user data is an incrementing counter
then run one thread per CPU
where each busy-works dequeuing
when an element is dequeued, we check (on a per-thread basis) the
value dequeued is greater than the element previously dequeued
note we have no variation in the test for CAS+GC vs DWCAS
this is because all we do is dequeue
what we actually want to stress test is the queue
not CAS
so it's better to let the dequeue run as fast as possible
*/
internal_display_test_name( "Dequeuing" );
lfds700_list_asu_query( list_of_logical_processors, LFDS700_LIST_ASU_QUERY_GET_POTENTIALLY_INACCURATE_COUNT, NULL, (void **) &number_logical_processors );
lfds700_misc_prng_init( &ps );
number_elements_with_dummy_element = ( memory_in_megabytes * ONE_MEGABYTE_IN_BYTES ) / sizeof(struct test_element);
number_elements_without_dummy_element = number_elements_with_dummy_element - 1;
te_array = util_aligned_malloc( sizeof(struct test_element) * number_elements_with_dummy_element, LFDS700_PAL_ATOMIC_ISOLATION_IN_BYTES );
lfds700_queue_init_valid_on_current_logical_core( &qs, &(te_array + number_elements_without_dummy_element)->qe, &ps, NULL );
for( loop = 0 ; loop < number_elements_without_dummy_element ; loop++ )
{
LFDS700_QUEUE_SET_VALUE_IN_ELEMENT( (te_array+loop)->qe, loop );
lfds700_queue_enqueue( &qs, &(te_array+loop)->qe, &ps );
}
ts = util_malloc_wrapper( sizeof(struct test_state) * number_logical_processors );
for( loop = 0 ; loop < number_logical_processors ; loop++ )
{
(ts+loop)->qs = &qs;
(ts+loop)->error_flag = LOWERED;
}
thread_handles = util_malloc_wrapper( sizeof(test_pal_thread_state_t) * number_logical_processors );
util_thread_starter_new( &tts, number_logical_processors );
LFDS700_MISC_BARRIER_STORE;
lfds700_misc_force_store();
loop = 0;
lasue = NULL;
while( LFDS700_LIST_ASU_GET_START_AND_THEN_NEXT(*list_of_logical_processors, lasue) )
{
lp = LFDS700_LIST_ASU_GET_VALUE_FROM_ELEMENT( *lasue );
util_thread_starter_start( tts, &thread_handles[loop], loop, lp, thread_simple_dequeuer, ts+loop );
loop++;
}
util_thread_starter_run( tts );
for( loop = 0 ; loop < number_logical_processors ; loop++ )
test_pal_thread_wait( thread_handles[loop] );
util_thread_starter_delete( tts );
free( thread_handles );
LFDS700_MISC_BARRIER_LOAD;
// TRD : check queue is empty
lfds700_queue_query( &qs, LFDS700_QUEUE_QUERY_SINGLETHREADED_VALIDATE, &vi, &dvs );
// TRD : check for raised error flags
for( loop = 0 ; loop < number_logical_processors ; loop++ )
if( (ts+loop)->error_flag == RAISED )
dvs = LFDS700_MISC_VALIDITY_INVALID_TEST_DATA;
free( ts );
util_aligned_free( te_array );
lfds700_queue_cleanup( &qs, NULL );
internal_display_test_result( 1, "queue", dvs );
return;
}
