Refactor performance counters interface

Signed-off-by: Zbigniew Zdanowicz <[email protected]>

Author: zzdanowicz

opencl/source/command_queue/gpgpu_walker_bdw_and_later.inl

@@ -192,7 +192,7 @@ size_t EnqueueOperation<GfxFamily>::getSizeRequiredCSKernel(bool reserveProfilin
     if (reserveProfilingCmdsSpace) {
         size += 2 * sizeof(PIPE_CONTROL) + 2 * sizeof(typename GfxFamily::MI_STORE_REGISTER_MEM);
     }
-    size += PerformanceCounters::getGpuCommandsSize(commandQueue, reservePerfCounters);
+    size += PerformanceCounters::getGpuCommandsSize(commandQueue.getPerfCounters(), commandQueue.getGpgpuEngine().osContext->getEngineType(), reservePerfCounters);
     size += GpgpuWalkerHelper<GfxFamily>::getSizeForWADisableLSQCROPERFforOCL(pKernel);
     size += GpgpuWalkerHelper<GfxFamily>::getSizeForWaDisableRccRhwoOptimization(pKernel);
 

opencl/source/command_queue/gpgpu_walker_xehp_and_later.inl

@@ -159,7 +159,7 @@ size_t EnqueueOperation<GfxFamily>::getSizeRequiredCSKernel(bool reserveProfilin
         size += static_cast<size_t>(ImplicitScalingDispatch<GfxFamily>::getSize(false, staticPartitioning, devices, groupStart, groupCount));
     }
 
-    size += PerformanceCounters::getGpuCommandsSize(commandQueue, reservePerfCounters);
+    size += PerformanceCounters::getGpuCommandsSize(commandQueue.getPerfCounters(), commandQueue.getGpgpuEngine().osContext->getEngineType(), reservePerfCounters);
 
     return size;
 }

shared/source/os_interface/performance_counters.cpp

@@ -162,14 +162,11 @@ void PerformanceCounters::deleteQuery(QueryHandle_1_0 &handle) {
 //////////////////////////////////////////////////////
 // PerformanceCounters::getGpuCommandsSize
 //////////////////////////////////////////////////////
-uint32_t PerformanceCounters::getGpuCommandsSize(CommandQueue &commandQueue, const bool reservePerfCounters) {
-
+uint32_t PerformanceCounters::getGpuCommandsSize(PerformanceCounters *performanceCounters, aub_stream::EngineType engineType, const bool reservePerfCounters) {
     uint32_t size = 0;
 
     if (reservePerfCounters) {
-
-        const auto performanceCounters = commandQueue.getPerfCounters();
-        const auto commandBufferType = EngineHelpers::isCcs(commandQueue.getGpgpuEngine().osContext->getEngineType())
+        const auto commandBufferType = EngineHelpers::isCcs(engineType)
                                            ? MetricsLibraryApi::GpuCommandBufferType::Compute
                                            : MetricsLibraryApi::GpuCommandBufferType::Render;
 

shared/source/os_interface/performance_counters.h

@@ -9,6 +9,8 @@
 #include "shared/source/utilities/metrics_library.h"
 #include "shared/source/utilities/perf_counter.h"
 
+#include "engine_node.h"
+
 #include <mutex>
 
 namespace NEO {
@@ -17,7 +19,6 @@ namespace NEO {
 // Forward declaration.
 //////////////////////////////////////////////////////
 class TagNodeBase;
-class CommandQueue;
 
 //////////////////////////////////////////////////////
 // Performance counters implementation.
@@ -47,7 +48,7 @@ class PerformanceCounters {
     //////////////////////////////////////////////////////
     // Gpu commands.
     //////////////////////////////////////////////////////
-    static uint32_t getGpuCommandsSize(CommandQueue &commandQueue, const bool reservePerfCounters);
+    static uint32_t getGpuCommandsSize(PerformanceCounters *performanceCounters, aub_stream::EngineType engineType, const bool reservePerfCounters);
     uint32_t getGpuCommandsSize(const MetricsLibraryApi::GpuCommandBufferType commandBufferType, const bool begin);
     bool getGpuCommands(const MetricsLibraryApi::GpuCommandBufferType commandBufferType, TagNodeBase &performanceCounters, const bool begin, const uint32_t bufferSize, void *pBuffer);