SWDEV-301667 - Add cache state for a device

- Add a global cache state for a device to indicate scopes of submitted
AQL packets
- Remove scopes for TS marker if hipEventReleaseToDevice is passed. Set
env ROC_EVENT_NO_FLUSH=1 to use NOP AQL for event records.
It would flush caches by default with system scope release.
- Calling finish() should ensure if caches are flushed, if not queue a
marker

Change-Id: Ibbbdbb1cd7ac61cb35649169212142545be159e0

rocclr/platform/command.cpp

@@ -55,8 +55,14 @@ Event::Event(HostQueue& queue)
 }
 
 // ================================================================================================
-Event::Event() : callbacks_(NULL), status_(CL_SUBMITTED),
-    hw_event_(nullptr), notify_event_(nullptr), device_(nullptr) { notified_.clear(); }
+Event::Event()
+    : callbacks_(NULL),
+      status_(CL_SUBMITTED),
+      hw_event_(nullptr),
+      notify_event_(nullptr),
+      device_(nullptr) {
+  notified_.clear();
+}
 
 // ================================================================================================
 Event::~Event() {
@@ -241,7 +247,7 @@ bool Event::awaitCompletion() {
       return false;
     }
 
-    ClPrint(LOG_DEBUG, LOG_WAIT, "waiting for event %p to complete, current status %d",
+    ClPrint(LOG_DEBUG, LOG_WAIT, "Waiting for event %p to complete, current status %d",
       this, status());
     auto* queue = command().queue();
     if ((queue != nullptr) && queue->vdev()->ActiveWait()) {
@@ -256,7 +262,7 @@ bool Event::awaitCompletion() {
         lock_.wait();
       }
     }
-    ClPrint(LOG_DEBUG, LOG_WAIT, "event %p wait completed", this);
+    ClPrint(LOG_DEBUG, LOG_WAIT, "Event %p wait completed", this);
   }
 
   return status() == CL_COMPLETE;
@@ -277,7 +283,7 @@ bool Event::notifyCmdQueue(bool cpu_wait) {
         notified_.clear();
         return false;
       }
-      ClPrint(LOG_DEBUG, LOG_CMD, "queue marker to command queue: %p", queue);
+      ClPrint(LOG_DEBUG, LOG_CMD, "Queue marker to command queue: %p", queue);
       command->enqueue();
       // Save notification, associated with the current event
       notify_event_ = command;
@@ -290,7 +296,7 @@ bool Event::notifyCmdQueue(bool cpu_wait) {
         notified_.clear();
         return false;
       }
-      ClPrint(LOG_DEBUG, LOG_CMD, "queue marker to command queue: %p", queue);
+      ClPrint(LOG_DEBUG, LOG_CMD, "Queue marker to command queue: %p", queue);
       command->enqueue();
       command->release();
     }
@@ -336,7 +342,7 @@ void Command::enqueue() {
     Agent::postEventCreate(as_cl(static_cast<Event*>(this)), type_);
   }
 
-  ClPrint(LOG_DEBUG, LOG_CMD, "command is enqueued: %p", this);
+  ClPrint(LOG_DEBUG, LOG_CMD, "Command enqueued: %p", this);
 
   // Direct dispatch logic below will submit the command immediately, but the command status
   // update will occur later after flush() with a wait
@@ -360,7 +366,7 @@ void Command::enqueue() {
       EnableProfiling();
     }
 
-    if (isMarker && !profilingInfo().marker_ts_) {
+    if (isMarker && (profilingInfo().marker_ts_ == 0)) {
       // Update batch head for the current marker. Hence the status of all commands can be
       // updated upon the marker completion
       SetBatchHead(queue_->GetSubmittionBatch());
@@ -414,6 +420,7 @@ NDRangeKernelCommand::NDRangeKernelCommand(HostQueue& queue, const EventWaitList
     profilingInfo_.enabled_ = true;
     profilingInfo_.clear();
     profilingInfo_.callback_ = nullptr;
+    profilingInfo_.marker_ts_ = 1;
   }
   kernel_.retain();
 }

rocclr/platform/command.hpp

@@ -102,7 +102,7 @@ class Event : public RuntimeObject {
   static const EventWaitList nullWaitList;
 
   struct ProfilingInfo {
-    ProfilingInfo(bool enabled = false) : enabled_(enabled), waves_(0), marker_ts_(false) {
+    ProfilingInfo(bool enabled = false) : enabled_(enabled), waves_(0), marker_ts_(0) {
       if (enabled) {
         clear();
         callback_ = nullptr;
@@ -113,10 +113,11 @@ class Event : public RuntimeObject {
     uint64_t submitted_;
     uint64_t start_;
     uint64_t end_;
-    bool enabled_;    //!< Profiling enabled for the wave limiter
-    uint32_t waves_;  //!< The number of waves used in a dispatch
+    bool enabled_;        //!< Profiling enabled for the wave limiter
+    uint32_t waves_;      //!< The number of waves used in a dispatch
     ProfilingCallback* callback_;
-    bool marker_ts_;
+    uint32_t marker_ts_;  //!< Marker with release scope
+                          //!< 5 - system scope, 3 - device scope, 1 - no scopes
     void clear() {
       queued_ = 0ULL;
       submitted_ = 0ULL;

rocclr/platform/commandqueue.cpp

@@ -106,20 +106,20 @@ bool HostQueue::terminate() {
 
 void HostQueue::finish() {
   Command* command = nullptr;
+  bool isCacheFlushed = device().IsCacheFlushed(Device::CacheState::kCacheStateSystem);
   if (IS_HIP) {
     command = getLastQueuedCommand(true);
-    // Check if the queue has nothing to process and return
-    if (AMD_DIRECT_DISPATCH &&  command == nullptr) {
+    if (AMD_DIRECT_DISPATCH && isCacheFlushed && command == nullptr) {
       return;
     }
   }
-  if (nullptr == command) {
+  if (nullptr == command || !isCacheFlushed) {
     // Send a finish to make sure we finished all commands
     command = new Marker(*this, false);
     if (command == NULL) {
       return;
     }
-    ClPrint(LOG_DEBUG, LOG_CMD, "marker is queued");
+    ClPrint(LOG_DEBUG, LOG_CMD, "Marker queued, Cache Flushed = %d", isCacheFlushed);
     command->enqueue();
   }
   // Check HW status of the ROCcrl event. Note: not all ROCclr modes support HW status
@@ -194,7 +194,7 @@ void HostQueue::loop(device::VirtualDevice* virtualDevice) {
       continue;
     }
 
-    ClPrint(LOG_DEBUG, LOG_CMD, "command (%s) is submitted: %p", getOclCommandKindString(command->type()), command);
+    ClPrint(LOG_DEBUG, LOG_CMD, "Command (%s) submitted: %p", getOclCommandKindString(command->type()), command);
 
     command->setStatus(CL_SUBMITTED);