SWDEV-447405 - Reset the last SDMA engine after every few copies

The copies can get blocked if the last SDMA engine is used by another
copy and this can lead to perf drop in some of the tests like Gromacs.
Resetting the last engine by checking the engine status and fetching the
new mask after few copies can avoid this.

Change-Id: I8fe8ea678db508d291c6242f3741fa9215e99921


[ROCm/clr commit: 1b25484f0f]

projects/clr/rocclr/device/rocm/rocblit.cpp

@@ -32,7 +32,8 @@ DmaBlitManager::DmaBlitManager(VirtualGPU& gpu, Setup setup)
     : HostBlitManager(gpu, setup),
       MinSizeForPinnedTransfer(dev().settings().pinnedMinXferSize_),
       completeOperation_(false),
-      context_(nullptr) {}
+      context_(nullptr),
+      sdmaEngineRetainCount_(0) {}
 
 inline void DmaBlitManager::synchronize() const {
   if (syncOperation_) {
@@ -684,6 +685,7 @@ bool DmaBlitManager::hsaCopy(const Memory& srcMemory, const Memory& dstMemory,
   uint32_t copyMask = 0;
   uint32_t freeEngineMask = 0;
   bool kUseRegularCopyApi = 0;
+  constexpr size_t kRetainCountThreshold = 8;
   bool forceSDMA = (copyMetadata.copyEnginePreference_ ==
                       amd::CopyMetadata::CopyEnginePreference::SDMA);
   HwQueueEngine engine = HwQueueEngine::Unknown;
@@ -694,10 +696,21 @@ bool DmaBlitManager::hsaCopy(const Memory& srcMemory, const Memory& dstMemory,
       (dstAgent.handle != dev().getCpuAgent().handle)) {
     engine = HwQueueEngine::SdmaWrite;
     copyMask = kUseRegularCopyApi ? 0 : dev().fetchSDMAMask(this, false);
+    if (copyMask == 0) {
+      // Track the HtoD copies and increment the count. The last used SDMA engine might be busy
+      // and using it everytime can cause contention. When the count exceeds the threshold,
+      // reset it so as to check the engine status and fetch the new mask.
+      sdmaEngineRetainCount_ = (sdmaEngineRetainCount_ > kRetainCountThreshold)
+                               ? 0 : sdmaEngineRetainCount_++;
+    }
   } else if ((srcAgent.handle != dev().getCpuAgent().handle) &&
              (dstAgent.handle == dev().getCpuAgent().handle)) {
     engine = HwQueueEngine::SdmaRead;
     copyMask = kUseRegularCopyApi ? 0 : dev().fetchSDMAMask(this, true);
+    if (copyMask == 0 && sdmaEngineRetainCount_ > 0) {
+      // Track the DtoH copies and decrement the count.
+      sdmaEngineRetainCount_--;
+    }
   }
 
   if (engine == HwQueueEngine::Unknown && forceSDMA) {
@@ -714,9 +727,11 @@ bool DmaBlitManager::hsaCopy(const Memory& srcMemory, const Memory& dstMemory,
 
   if (!kUseRegularCopyApi && engine != HwQueueEngine::Unknown) {
     if (copyMask == 0) {
-      // Check if there a recently used SDMA engine for the stream
-      copyMask = gpu().getLastUsedSdmaEngine();
-      ClPrint(amd::LOG_DEBUG, amd::LOG_COPY, "Last copy mask 0x%x", copyMask);
+      if (sdmaEngineRetainCount_) {
+        // Check if there a recently used SDMA engine for the stream
+        copyMask = gpu().getLastUsedSdmaEngine();
+        ClPrint(amd::LOG_DEBUG, amd::LOG_COPY, "Last copy mask 0x%x", copyMask);
+      }
       if (copyMask == 0) {
         // Check SDMA engine status
         status = hsa_amd_memory_copy_engine_status(dstAgent, srcAgent, &freeEngineMask);

projects/clr/rocclr/device/rocm/rocblit.hpp

@@ -236,6 +236,8 @@ class DmaBlitManager : public device::HostBlitManager {
   const size_t MinSizeForPinnedTransfer;
   bool completeOperation_;                    //!< DMA blit manager must complete operation
   amd::Context* context_;                     //!< A dummy context
+  mutable size_t sdmaEngineRetainCount_;      //!< Keeps track of memcopies to either get the last
+                                              //!< used SDMA engine or fetch the new mask
 
  private:
   //! Disable copy constructor