Create work partition allocation

Related-To: NEO-5546
Resolves: NEO-5561
Signed-off-by: Maciej Dziuban <maciej.dziuban@intel.com>

shared/source/command_stream/command_stream_receiver.cpp

@@ -48,6 +48,10 @@ CommandStreamReceiver::CommandStreamReceiver(ExecutionEnvironment &executionEnvi
         indirectHeap[i] = nullptr;
     }
     internalAllocationStorage = std::make_unique<InternalAllocationStorage>(*this);
+
+    if (DebugManager.flags.EnableStaticPartitioning.get() == 1) {
+        this->staticWorkPartitioningEnabled = true;
+    }
 }
 
 CommandStreamReceiver::~CommandStreamReceiver() {
@@ -226,6 +230,11 @@ void CommandStreamReceiver::cleanupResources() {
         getMemoryManager()->freeGraphicsMemory(clearColorAllocation);
         clearColorAllocation = nullptr;
     }
+
+    if (workPartitionAllocation) {
+        getMemoryManager()->freeGraphicsMemory(workPartitionAllocation);
+        workPartitionAllocation = nullptr;
+    }
 }
 
 bool CommandStreamReceiver::waitForCompletionWithTimeout(bool enableTimeout, int64_t timeoutMicroseconds, uint32_t taskCountToWait) {
@@ -474,6 +483,37 @@ bool CommandStreamReceiver::initializeTagAllocation() {
     return true;
 }
 
+bool CommandStreamReceiver::createWorkPartitionAllocation(const Device &device) {
+    if (!staticWorkPartitioningEnabled) {
+        return false;
+    }
+    UNRECOVERABLE_IF(device.getNumAvailableDevices() < 2);
+
+    AllocationProperties properties{this->rootDeviceIndex, true, 4096u, GraphicsAllocation::AllocationType::WORK_PARTITION_SURFACE, true, false, deviceBitfield};
+    this->workPartitionAllocation = getMemoryManager()->allocateGraphicsMemoryWithProperties(properties);
+    if (this->workPartitionAllocation == nullptr) {
+        return false;
+    }
+
+    for (uint32_t deviceIndex = 0; deviceIndex < deviceBitfield.size(); deviceIndex++) {
+        if (!deviceBitfield.test(deviceIndex)) {
+            continue;
+        }
+
+        const uint32_t copySrc = deviceIndex;
+        const Vec3<size_t> copySrcSize = {sizeof(copySrc), 1, 1};
+        DeviceBitfield copyBitfield{};
+        copyBitfield.set(deviceIndex);
+        BlitOperationResult blitResult = BlitHelper::blitMemoryToAllocationBanks(device, workPartitionAllocation, 0, &copySrc, copySrcSize, copyBitfield);
+
+        if (blitResult != BlitOperationResult::Success) {
+            return false;
+        }
+    }
+
+    return true;
+}
+
 bool CommandStreamReceiver::createGlobalFenceAllocation() {
     auto hwInfo = executionEnvironment.rootDeviceEnvironments[rootDeviceIndex]->getHardwareInfo();
     if (!HwHelper::get(hwInfo->platform.eRenderCoreFamily).isFenceAllocationRequired(*hwInfo)) {

shared/source/command_stream/command_stream_receiver.h

@@ -142,6 +142,7 @@ class CommandStreamReceiver {
     GraphicsAllocation *allocateDebugSurface(size_t size);
     GraphicsAllocation *getPreemptionAllocation() const { return preemptionAllocation; }
     GraphicsAllocation *getGlobalFenceAllocation() const { return globalFenceAllocation; }
+    GraphicsAllocation *getWorkPartitionAllocation() const { return workPartitionAllocation; }
 
     void requestStallingPipeControlOnNextFlush() { stallingPipeControlOnNextFlushRequired = true; }
     bool isStallingPipeControlOnNextFlushRequired() const { return stallingPipeControlOnNextFlushRequired; }
@@ -168,6 +169,7 @@ class CommandStreamReceiver {
     void setExperimentalCmdBuffer(std::unique_ptr<ExperimentalCommandBuffer> &&cmdBuffer);
 
     bool initializeTagAllocation();
+    MOCKABLE_VIRTUAL bool createWorkPartitionAllocation(const Device &device);
     MOCKABLE_VIRTUAL bool createGlobalFenceAllocation();
     MOCKABLE_VIRTUAL bool createPreemptionAllocation();
     MOCKABLE_VIRTUAL bool createPerDssBackedBuffer(Device &device);
@@ -273,6 +275,7 @@ class CommandStreamReceiver {
     GraphicsAllocation *debugSurface = nullptr;
     GraphicsAllocation *perDssBackedBuffer = nullptr;
     GraphicsAllocation *clearColorAllocation = nullptr;
+    GraphicsAllocation *workPartitionAllocation = nullptr;
 
     IndirectHeap *indirectHeap[IndirectHeap::NUM_TYPES];
     OsContext *osContext = nullptr;
@@ -318,6 +321,7 @@ class CommandStreamReceiver {
     bool lastVmeSubslicesConfig = false;
     bool stallingPipeControlOnNextFlushRequired = false;
     bool timestampPacketWriteEnabled = false;
+    bool staticWorkPartitioningEnabled = false;
     bool nTo1SubmissionModelEnabled = false;
     bool lastSpecialPipelineSelectMode = false;
     bool requiresInstructionCacheFlush = false;