Add multiEngine field to command queue with debug variable to override it.

Change-Id: I3c1e424a7ad545e166e178d1726595e6d9502ca7

runtime/command_queue/command_queue.cpp

@@ -71,6 +71,10 @@ CommandQueue::CommandQueue(Context *context, Device *deviceId, const cl_queue_pr
     }
 
     processProperties(properties);
+
+    if (DebugManager.flags.ForceMultiEngineQueue.get() > -1) {
+        this->multiEngineQueue = DebugManager.flags.ForceMultiEngineQueue.get();
+    }
 }
 
 CommandQueue::~CommandQueue() {

runtime/command_queue/command_queue.h

@@ -406,6 +406,8 @@ class CommandQueue : public BaseObject<_cl_command_queue> {
         return requiresCacheFlushAfterWalker;
     }
 
+    bool isMultiEngineQueue() { return this->multiEngineQueue; }
+
     // taskCount of last task
     uint32_t taskCount = 0;
 
@@ -456,6 +458,7 @@ class CommandQueue : public BaseObject<_cl_command_queue> {
     bool mapDcFlushRequired = false;
     bool isSpecialCommandQueue = false;
     bool requiresCacheFlushAfterWalker = false;
+    bool multiEngineQueue = false;
 
     std::unique_ptr<TimestampPacketContainer> timestampPacketContainer;
 

runtime/os_interface/debug_variables_base.inl

@@ -122,5 +122,6 @@ DECLARE_DEBUG_VARIABLE(int32_t, NodeOrdinal, -1, "-1: default do not override, 0
 DECLARE_DEBUG_VARIABLE(int32_t, OverrideThreadArbitrationPolicy, -1, "-1 (dont override) or any valid config (0: Age Based, 1: Round Robin)")
 DECLARE_DEBUG_VARIABLE(int32_t, OverrideAubDeviceId, -1, "-1 dont override, any other: use this value for AUB generation device id")
 DECLARE_DEBUG_VARIABLE(int32_t, EnableTimestampPacket, -1, "-1: default, 0: disable, 1:enable. Write Timestamp Packet for each set of gpu walkers")
+DECLARE_DEBUG_VARIABLE(int32_t, ForceMultiEngineQueue, -1, "-1 no change, 0: force false 1:force true. Forces multi engine queue flag in command queue")
 DECLARE_DEBUG_VARIABLE(bool, UseMaxSimdSizeToDeduceMaxWorkgroupSize, false, "With this flag on, max workgroup size is deduced using SIMD32 instead of SIMD8, this causes the max wkg size to be 4 times bigger")
 DECLARE_DEBUG_VARIABLE(bool, ReturnRawGpuTimestamps, false, "Driver returns raw GPU tiemstamps instead of calculated ones.")

unit_tests/command_queue/command_queue_tests.cpp

@@ -23,6 +23,7 @@
 #include "unit_tests/fixtures/memory_management_fixture.h"
 #include "unit_tests/fixtures/buffer_fixture.h"
 #include "unit_tests/helpers/unit_test_helper.h"
+#include "unit_tests/helpers/debug_manager_state_restore.h"
 #include "unit_tests/libult/ult_command_stream_receiver.h"
 #include "unit_tests/mocks/mock_memory_manager.h"
 #include "unit_tests/mocks/mock_command_queue.h"
@@ -976,3 +977,20 @@ TEST(CommandQueueDestructorTest, whenCommandQueueIsDestroyedThenDestroysTimestam
     context->release();
     EXPECT_EQ(1, context->getRefInternalCount());
 }
+
+TEST(CommandQueuePropertiesTests, whenDefaultCommandQueueIsCreatedThenItIsNotMultiEngineQueue) {
+    MockCommandQueue queue;
+    EXPECT_FALSE(queue.multiEngineQueue);
+    EXPECT_FALSE(queue.isMultiEngineQueue());
+    queue.multiEngineQueue = true;
+    EXPECT_TRUE(queue.isMultiEngineQueue());
+}
+TEST(CommandQueuePropertiesTests, whenDebugVariableOverridesMultiEngineVariableThenItIsSetToTrue) {
+    DebugManagerStateRestore restore;
+    DebugManager.flags.ForceMultiEngineQueue.set(1u);
+    MockCommandQueue queue;
+    EXPECT_TRUE(queue.isMultiEngineQueue());
+    DebugManager.flags.ForceMultiEngineQueue.set(0u);
+    MockCommandQueue queue2;
+    EXPECT_FALSE(queue2.isMultiEngineQueue());
+}

unit_tests/mocks/mock_command_queue.h

@@ -17,6 +17,7 @@ namespace OCLRT {
 class MockCommandQueue : public CommandQueue {
   public:
     using CommandQueue::device;
+    using CommandQueue::multiEngineQueue;
     using CommandQueue::obtainNewTimestampPacketNodes;
     using CommandQueue::requiresCacheFlushAfterWalker;
     using CommandQueue::throttle;

unit_tests/test_files/igdrcl.config

@@ -108,3 +108,4 @@ EnableCacheFlushAfterWalker = 0
 EnableHostPtrTracking = 1
 DisableDcFlushInEpilogue = 0
 EnableCacheFlushAfterWalkerForAllQueues = 0
+ForceMultiEngineQueue = -1