Skip to content

[SYCL][L0] Honor property::queue::enable_profiling #5543

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 4 commits into from
Feb 16, 2022
Merged

[SYCL][L0] Honor property::queue::enable_profiling #5543

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
416c471
[SYCL][L0] Honor property::queue::enable_profiling
smaslov-intel Feb 10, 2022
d537cc5
Add zePrint to be able to check in E2E test
smaslov-intel Feb 10, 2022
43f0c7f
restart testing
smaslov-intel Feb 15, 2022
459356b
fix reported regressions
smaslov-intel Feb 16, 2022
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
125 changes: 71 additions & 54 deletions sycl/plugins/level_zero/pi_level_zero.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -32,8 +32,8 @@ extern "C" {
// Forward declarartions.
static pi_result EventRelease(pi_event Event, pi_queue LockedQueue);
static pi_result QueueRelease(pi_queue Queue, pi_queue LockedQueue);
static pi_result EventCreate(pi_context Context, bool HostVisible,
pi_event *RetEvent);
static pi_result EventCreate(pi_context Context, pi_queue Queue,
bool HostVisible, pi_event *RetEvent);
}

namespace {
Expand Down Expand Up @@ -428,20 +428,13 @@ pi_result _pi_mem::removeMapping(void *MappedTo, Mapping &MapInfo) {

pi_result
_pi_context::getFreeSlotInExistingOrNewPool(ze_event_pool_handle_t &Pool,
size_t &Index, bool HostVisible) {
size_t &Index, bool HostVisible,
bool ProfilingEnabled) {
// Lock while updating event pool machinery.
std::lock_guard<std::mutex> Lock(ZeEventPoolCacheMutex);

// Setup for host-visible pool as needed.
ze_event_pool_flag_t ZePoolFlag = {};
std::list<ze_event_pool_handle_t> *ZePoolCache;

if (HostVisible) {
ZePoolFlag = ZE_EVENT_POOL_FLAG_HOST_VISIBLE;
ZePoolCache = &ZeHostVisibleEventPoolCache;
} else {
ZePoolCache = &ZeDeviceScopeEventPoolCache;
}
std::list<ze_event_pool_handle_t> *ZePoolCache =
getZeEventPoolCache(HostVisible, ProfilingEnabled);

// Remove full pool from the cache.
if (!ZePoolCache->empty()) {
Expand All @@ -460,7 +453,12 @@ _pi_context::getFreeSlotInExistingOrNewPool(ze_event_pool_handle_t &Pool,
if (*ZePool == nullptr) {
ZeStruct<ze_event_pool_desc_t> ZeEventPoolDesc;
ZeEventPoolDesc.count = MaxNumEventsPerPool;
ZeEventPoolDesc.flags = ZePoolFlag | ZE_EVENT_POOL_FLAG_KERNEL_TIMESTAMP;
ZeEventPoolDesc.flags = 0;
if (HostVisible)
ZeEventPoolDesc.flags |= ZE_EVENT_POOL_FLAG_HOST_VISIBLE;
if (ProfilingEnabled)
ZeEventPoolDesc.flags |= ZE_EVENT_POOL_FLAG_KERNEL_TIMESTAMP;
zePrint("ze_event_pool_desc_t flags set to: %d\n", ZeEventPoolDesc.flags);

std::vector<ze_device_handle_t> ZeDevices;
std::for_each(Devices.begin(), Devices.end(),
Expand All @@ -486,12 +484,8 @@ pi_result _pi_context::decrementUnreleasedEventsInPool(pi_event Event) {
return PI_SUCCESS;
}

std::list<ze_event_pool_handle_t> *ZePoolCache;
if (Event->IsHostVisible()) {
ZePoolCache = &ZeHostVisibleEventPoolCache;
} else {
ZePoolCache = &ZeDeviceScopeEventPoolCache;
}
std::list<ze_event_pool_handle_t> *ZePoolCache =
getZeEventPoolCache(Event->isHostVisible(), Event->isProfilingEnabled());

// Put the empty pool to the cache of the pools.
std::lock_guard<std::mutex> Lock(ZeEventPoolCacheMutex);
Expand Down Expand Up @@ -611,13 +605,15 @@ inline static void piQueueRetainNoLock(pi_queue Queue) { Queue->RefCount++; }
// \param Event a pointer to hold the newly created pi_event
// \param CommandType various command type determined by the caller
// \param CommandList is the command list where the event is added
inline static pi_result
createEventAndAssociateQueue(pi_queue Queue, pi_event *Event,
pi_command_type CommandType,
pi_command_list_ptr_t CommandList) {
pi_result Res = piEventCreate(Queue->Context, Event);
if (Res != PI_SUCCESS)
return Res;
// \param ForceHostVisible tells if the event must be created in
// the host-visible pool
inline static pi_result createEventAndAssociateQueue(
pi_queue Queue, pi_event *Event, pi_command_type CommandType,
pi_command_list_ptr_t CommandList, bool ForceHostVisible = false) {

PI_CALL(EventCreate(Queue->Context, Queue,
ForceHostVisible ? true : EventsScope == AllHostVisible,
Event));

(*Event)->Queue = Queue;
(*Event)->CommandType = CommandType;
Expand Down Expand Up @@ -806,13 +802,11 @@ pi_result _pi_context::finalize() {
// For example, event pool caches would be still alive.
{
std::lock_guard<std::mutex> Lock(ZeEventPoolCacheMutex);
for (auto &ZePool : ZeDeviceScopeEventPoolCache)
ZE_CALL(zeEventPoolDestroy, (ZePool));
for (auto &ZePool : ZeHostVisibleEventPoolCache)
ZE_CALL(zeEventPoolDestroy, (ZePool));

ZeDeviceScopeEventPoolCache.clear();
ZeHostVisibleEventPoolCache.clear();
for (auto &ZePoolCache : ZeEventPoolCache) {
for (auto &ZePool : ZePoolCache)
ZE_CALL(zeEventPoolDestroy, (ZePool));
ZePoolCache.clear();
}
}

// Destroy the command list used for initializations
Expand Down Expand Up @@ -841,8 +835,7 @@ pi_result _pi_context::finalize() {

bool _pi_queue::isInOrderQueue() const {
// If out-of-order queue property is not set, then this is a in-order queue.
return ((this->PiQueueProperties & PI_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE) ==
0);
return ((this->Properties & PI_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE) == 0);
}

pi_result _pi_queue::resetCommandList(pi_command_list_ptr_t CommandList,
Expand Down Expand Up @@ -1032,11 +1025,10 @@ static const zeCommandListBatchConfig ZeCommandListBatchCopyConfig = [] {
_pi_queue::_pi_queue(ze_command_queue_handle_t Queue,
std::vector<ze_command_queue_handle_t> &CopyQueues,
pi_context Context, pi_device Device,
bool OwnZeCommandQueue,
pi_queue_properties PiQueueProperties)
bool OwnZeCommandQueue, pi_queue_properties Properties)
: ZeComputeCommandQueue{Queue}, ZeCopyCommandQueues{CopyQueues},
Context{Context}, Device{Device}, OwnZeCommandQueue{OwnZeCommandQueue},
PiQueueProperties(PiQueueProperties) {
Properties(Properties) {
ComputeCommandBatch.OpenCommandList = CommandListMap.end();
CopyCommandBatch.OpenCommandList = CommandListMap.end();
ComputeCommandBatch.QueueBatchSize =
Expand Down Expand Up @@ -1350,7 +1342,10 @@ pi_result _pi_queue::executeCommandList(pi_command_list_ptr_t CommandList,
// Create a "proxy" host-visible event.
//
pi_event HostVisibleEvent;
PI_CALL(EventCreate(Context, true, &HostVisibleEvent));
auto Res = createEventAndAssociateQueue(
this, &HostVisibleEvent, PI_COMMAND_TYPE_USER, CommandList, true);
if (Res)
return Res;

// Update each command's event in the command-list to "see" this
// proxy event as a host-visible counterpart.
Expand All @@ -1359,10 +1354,14 @@ pi_result _pi_queue::executeCommandList(pi_command_list_ptr_t CommandList,
PI_CALL(piEventRetain(HostVisibleEvent));
}

// Decrement the reference count by 1 so all the remaining references
// are from the other commands in this batch. This host-visible event
// will be destroyed after all events in the batch are gone.
// Decrement the reference count of the event such that all the remaining
// references are from the other commands in this batch. This host-visible
// event will not be waited/release by SYCL RT, so it must be destroyed
// after all events in the batch are gone.
PI_CALL(piEventRelease(HostVisibleEvent));
PI_CALL(piEventRelease(HostVisibleEvent));
PI_CALL(piEventRelease(HostVisibleEvent));

// Indicate no cleanup is needed for this PI event as it is special.
HostVisibleEvent->CleanedUp = true;

Expand Down Expand Up @@ -2105,7 +2104,7 @@ pi_result piDevicesGet(pi_platform Platform, pi_device_type DeviceType,
*NumDevices = ZeDeviceCount;

if (NumEntries == 0) {
// Devices should be nullptr when querying the number of devices
// Devices should be nullptr when querying the number of devices.
PI_ASSERT(Devices == nullptr, PI_INVALID_VALUE);
return PI_SUCCESS;
}
Expand Down Expand Up @@ -4954,10 +4953,6 @@ _pi_event::getOrCreateHostVisibleEvent(ze_event_handle_t &ZeHostVisibleEvent) {
if (EventsScope != OnDemandHostVisibleProxy)
die("getOrCreateHostVisibleEvent: missing host-visible event");

// Create a "proxy" host-visible event on demand.
PI_CALL(EventCreate(Context, true, &HostVisibleEvent));
HostVisibleEvent->CleanedUp = true;

// Submit the command(s) signalling the proxy event to the queue.
// We have to first submit a wait for the device-only event for which this
// proxy is created.
Expand All @@ -4974,6 +4969,13 @@ _pi_event::getOrCreateHostVisibleEvent(ze_event_handle_t &ZeHostVisibleEvent) {
Queue, CommandList, false /* UseCopyEngine */, OkToBatch))
return Res;

// Create a "proxy" host-visible event.
auto Res = createEventAndAssociateQueue(
Queue, &HostVisibleEvent, PI_COMMAND_TYPE_USER, CommandList, true);
// HostVisibleEvent->CleanedUp = true;
if (Res != PI_SUCCESS)
return Res;

ZE_CALL(zeCommandListAppendWaitOnEvents,
(CommandList->first, 1, &ZeEvent));
ZE_CALL(zeCommandListAppendSignalEvent,
Expand All @@ -4988,12 +4990,21 @@ _pi_event::getOrCreateHostVisibleEvent(ze_event_handle_t &ZeHostVisibleEvent) {
return PI_SUCCESS;
}

static pi_result EventCreate(pi_context Context, bool HostVisible,
pi_event *RetEvent) {
// Helper function for creating a PI event.
// The "Queue" argument specifies the PI queue where a command is submitted.
// The "HostVisible" argument specifies if event needs to be allocated from
// a host-visible pool.
//
static pi_result EventCreate(pi_context Context, pi_queue Queue,
bool HostVisible, pi_event *RetEvent) {

bool ProfilingEnabled =
!Queue || (Queue->Properties & PI_QUEUE_PROFILING_ENABLE) != 0;

size_t Index = 0;
ze_event_pool_handle_t ZeEventPool = {};
if (auto Res = Context->getFreeSlotInExistingOrNewPool(ZeEventPool, Index,
HostVisible))
if (auto Res = Context->getFreeSlotInExistingOrNewPool(
ZeEventPool, Index, HostVisible, ProfilingEnabled))
return Res;

ze_event_handle_t ZeEvent;
Expand Down Expand Up @@ -5034,8 +5045,9 @@ static pi_result EventCreate(pi_context Context, bool HostVisible,
return PI_SUCCESS;
}

// Exteral PI API entry
pi_result piEventCreate(pi_context Context, pi_event *RetEvent) {
return EventCreate(Context, EventsScope == AllHostVisible, RetEvent);
return EventCreate(Context, nullptr, EventsScope == AllHostVisible, RetEvent);
}

pi_result piEventGetInfo(pi_event Event, pi_event_info ParamName,
Expand Down Expand Up @@ -5101,6 +5113,11 @@ pi_result piEventGetProfilingInfo(pi_event Event, pi_profiling_info ParamName,

PI_ASSERT(Event, PI_INVALID_EVENT);

if (Event->Queue &&
(Event->Queue->Properties & PI_QUEUE_PROFILING_ENABLE) == 0) {
return PI_PROFILING_INFO_NOT_AVAILABLE;
}

uint64_t ZeTimerResolution =
Event->Queue
? Event->Queue->Device->ZeDeviceProperties->timerResolution
Expand Down Expand Up @@ -5375,7 +5392,7 @@ static pi_result EventRelease(pi_event Event, pi_queue LockedQueue) {
// and release a reference to it.
if (Event->HostVisibleEvent && Event->HostVisibleEvent != Event) {
// Decrement ref-count of the host-visible proxy event.
PI_CALL(piEventRelease(Event->HostVisibleEvent));
PI_CALL(EventRelease(Event->HostVisibleEvent, LockedQueue));
}

auto Context = Event->Context;
Expand Down
27 changes: 20 additions & 7 deletions sycl/plugins/level_zero/pi_level_zero.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -562,9 +562,11 @@ struct _pi_context : _pi_object {

// Get index of the free slot in the available pool. If there is no available
// pool then create new one. The HostVisible parameter tells if we need a
// slot for a host-visible event.
// slot for a host-visible event. The ProfilingEnabled tells is we need a
// slot for an event with profiling capabilities.
pi_result getFreeSlotInExistingOrNewPool(ze_event_pool_handle_t &, size_t &,
bool HostVisible);
bool HostVisible,
bool ProfilingEnabled);

// Decrement number of events living in the pool upon event destroy
// and return the pool to the cache if there are no unreleased events.
Expand Down Expand Up @@ -601,9 +603,14 @@ struct _pi_context : _pi_object {
// head. In case there is no next pool, a new pool is created and made the
// head.
//
std::list<ze_event_pool_handle_t> ZeDeviceScopeEventPoolCache;
// Cache of event pools to which host-visible events are added to.
std::list<ze_event_pool_handle_t> ZeHostVisibleEventPoolCache;
std::vector<std::list<ze_event_pool_handle_t>> ZeEventPoolCache{4};
auto getZeEventPoolCache(bool HostVisible, bool WithProfiling) {
if (HostVisible)
return WithProfiling ? &ZeEventPoolCache[0] : &ZeEventPoolCache[1];
else
return WithProfiling ? &ZeEventPoolCache[2] : &ZeEventPoolCache[3];
}

// This map will be used to determine if a pool is full or not
// by storing number of empty slots available in the pool.
Expand All @@ -625,7 +632,7 @@ struct _pi_queue : _pi_object {
_pi_queue(ze_command_queue_handle_t Queue,
std::vector<ze_command_queue_handle_t> &CopyQueues,
pi_context Context, pi_device Device, bool OwnZeCommandQueue,
pi_queue_properties PiQueueProperties = 0);
pi_queue_properties Properties = 0);

// Level Zero compute command queue handle.
ze_command_queue_handle_t ZeComputeCommandQueue;
Expand Down Expand Up @@ -731,7 +738,7 @@ struct _pi_queue : _pi_object {
bool isBatchingAllowed(bool IsCopy) const;

// Keeps the properties of this queue.
pi_queue_properties PiQueueProperties;
pi_queue_properties Properties;

// Returns true if the queue is a in-order queue.
bool isInOrderQueue() const;
Expand Down Expand Up @@ -986,11 +993,17 @@ struct _pi_event : _pi_object {
// than by just this one event, depending on the mode (see EventsScope).
//
pi_event HostVisibleEvent = {nullptr};
bool IsHostVisible() const { return this == HostVisibleEvent; }
bool isHostVisible() const { return this == HostVisibleEvent; }

// Get the host-visible event or create one and enqueue its signal.
pi_result getOrCreateHostVisibleEvent(ze_event_handle_t &HostVisibleEvent);

// Tells if this event is with profiling capabilities.
bool isProfilingEnabled() const {
return !Queue || // tentatively assume user events are profiling enabled
(Queue->Properties & PI_QUEUE_PROFILING_ENABLE) != 0;
}

// Level Zero command list where the command signaling this event was appended
// to. This is currently used to remember/destroy the command list after all
// commands in it are completed, i.e. this event signaled.
Expand Down