SWDEV-508004 - Improve hipEventRecord

- Resolve stream once for event record. We should avoid calling
  getStream again in addMarker

Change-Id: I78448c4f151ae10a5c8e8c248b2f4078b84191cb


[ROCm/clr commit: a22c45d635]

projects/clr/hipamd/src/hip_event.cpp

@@ -237,7 +237,9 @@ hipError_t Event::recordCommand(amd::Command*& command, amd::HostQueue* stream,
 // ================================================================================================
 hipError_t Event::enqueueRecordCommand(hipStream_t stream, amd::Command* command, bool record) {
   command->enqueue();
-  if (event_ == &command->event()) return hipSuccess;
+  if (event_ == &command->event()) {
+    return hipSuccess;
+  }
   if (event_ != nullptr) {
     event_->release();
   }
@@ -250,7 +252,9 @@ hipError_t Event::enqueueRecordCommand(hipStream_t stream, amd::Command* command
 // ================================================================================================
 hipError_t Event::addMarker(hipStream_t stream, amd::Command* command,
                             bool record, bool batch_flush) {
-  hip::Stream* hip_stream = hip::getStream(stream);
+  // Skip wait as we should not be resolving stream in this sub
+  constexpr bool kSkipWait = true;
+  hip::Stream* hip_stream = hip::getStream(stream, kSkipWait);
   // Keep the lock always at the beginning of this to avoid a race. SWDEV-277847
   amd::ScopedLock lock(lock_);
   hipError_t status = recordCommand(command, hip_stream, 0, batch_flush);

projects/clr/hipamd/src/hip_event.hpp

@@ -106,7 +106,6 @@ class Event {
 
   Event(uint32_t flags) : flags_(flags), lock_(true) /* hipEvent_t lock*/,
                               event_(nullptr), unrecorded_(false), stream_(nullptr) {
-    // No need to init event_ here as addMarker does that
     device_id_ = hip::getCurrentDevice()->deviceId();  // Created in current device ctx
   }
 

projects/clr/hipamd/src/hip_mempool_impl.cpp

@@ -97,7 +97,7 @@ Heap::SortedMap::iterator Heap::EraseAllocaton(Heap::SortedMap::iterator& it) {
   const device::Memory* dev_mem = memory->getDeviceMemory(*device_->devices()[0]);
   void* dev_mem_vaddr = reinterpret_cast<void*>(dev_mem->virtualAddress());
   total_size_ -= it->first.first;
-  
+
   if (dev_mem_vaddr != nullptr) {
     amd::SvmBuffer::free(memory->getContext(), dev_mem_vaddr);
   } else {