SWDEV-301667 - Fix SDMA mask reuse

If we are using the mask returned by getLastUsedSdmaEngine() then we
need to apply the SDMA Read/Write mask to it before using with HSA
copy_on_engine API.

Change-Id: I6e5dc6c187eeb3c61ee159e9d2a0fa7b4737c06e

rocclr/device/rocm/rocblit.cpp

@@ -33,7 +33,9 @@ DmaBlitManager::DmaBlitManager(VirtualGPU& gpu, Setup setup)
       MinSizeForPinnedTransfer(dev().settings().pinnedMinXferSize_),
       completeOperation_(false),
       context_(nullptr),
-      sdmaEngineRetainCount_(0) {}
+      sdmaEngineRetainCount_(0) {
+        dev().getSdmaRWMasks(&sdmaEngineReadMask_, &sdmaEngineWriteMask_);
+      }
 
 inline void DmaBlitManager::synchronize() const {
   if (syncOperation_) {
@@ -731,6 +733,8 @@ bool DmaBlitManager::hsaCopy(const Memory& srcMemory, const Memory& dstMemory,
         // 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);
+        copyMask &= (engine == HwQueueEngine::SdmaRead ?
+                    sdmaEngineReadMask_ : sdmaEngineWriteMask_);
       }
       if (copyMask == 0) {
         // Check SDMA engine status

rocclr/device/rocm/rocblit.hpp

@@ -238,6 +238,8 @@ class DmaBlitManager : public device::HostBlitManager {
   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
+  uint32_t sdmaEngineReadMask_;               //!< SDMA Engine Read Mask
+  uint32_t sdmaEngineWriteMask_;              //!< SDMA Engine Write Mask
 
  private:
   //! Disable copy constructor

rocclr/device/rocm/rocdevice.cpp

@@ -3473,6 +3473,12 @@ uint32_t Device::fetchSDMAMask(const device::BlitManager* handle, bool readEngin
   return (readEngine ? maxSdmaReadMask_ : maxSdmaWriteMask_) & engine;
 }
 
+// ================================================================================================
+void Device::getSdmaRWMasks(uint32_t* readMask, uint32_t* writeMask) const {
+  *readMask = maxSdmaReadMask_;
+  *writeMask = maxSdmaWriteMask_;
+}
+
 // ================================================================================================
 void Device::resetSDMAMask(const device::BlitManager* handle) const {
   amd::ScopedLock lock(vgpusAccess());

rocclr/device/rocm/rocdevice.hpp

@@ -583,7 +583,8 @@ class Device : public NullDevice {
   void HiddenHeapAlloc(const VirtualGPU& gpu);
 
   uint32_t fetchSDMAMask(const device::BlitManager* handle, bool readEngine = true) const;
-  void resetSDMAMask(const device::BlitManager* handle) const ;
+  void resetSDMAMask(const device::BlitManager* handle) const;
+  void getSdmaRWMasks(uint32_t* readMask, uint32_t* writeMask) const;
   bool isXgmi() const { return isXgmi_; }
 
  private: