IVGCVSW-2059: Remove CL-specific code from Profiler

Change-Id: I279986b83c7a38b6d395f8a496382ffaebb9ed08
diff --git a/src/armnn/Profiling.cpp b/src/armnn/Profiling.cpp
index b749c80..4caa319 100644
--- a/src/armnn/Profiling.cpp
+++ b/src/armnn/Profiling.cpp
@@ -9,10 +9,6 @@
 #include <streamline_annotate.h>
 #endif
 
-#if ARMCOMPUTECL_ENABLED
-#include <arm_compute/runtime/CL/CLFunctions.h>
-#endif
-
 #include <algorithm>
 #include <iomanip>
 #include <iostream>
@@ -42,12 +38,6 @@
 // It can be convenient for local tests.
 constexpr bool g_WriteReportToStdOutOnProfilerDestruction = false;
 
-// Whether events denoting operations running on the GPU should force a sync before/after the event.
-// This is hardcoded to true for now as the profiling timings are not very useful without it.
-#if ARMCOMPUTECL_ENABLED
-constexpr bool g_ProfilingForceGpuSync = true;
-#endif
-
 Measurement FindMeasurement(const std::string& name, const Event* event)
 {
 
@@ -206,9 +196,6 @@
 
 Event* Profiler::BeginEvent(Compute compute, const std::string& label, std::vector<InstrumentPtr>&& instruments)
 {
-    // We need to sync just before the begin event to not include time before the period we want to time.
-    WaitForDevice(compute);
-
     Event* parent = m_Parents.empty() ? nullptr : m_Parents.top();
     m_EventSequence.push_back(std::make_unique<Event>(label, this, parent, compute, std::move(instruments)));
     Event* event = m_EventSequence.back().get();
@@ -478,16 +465,6 @@
     }
 }
 
-void Profiler::WaitForDevice(Compute compute) const
-{
-#if ARMCOMPUTECL_ENABLED
-    if(compute == Compute::GpuAcc && g_ProfilingForceGpuSync)
-    {
-        arm_compute::CLScheduler::get().sync();
-    }
-#endif
-}
-
 std::uint32_t Profiler::GetEventColor(Compute compute) const
 {
     switch(compute)
diff --git a/src/armnn/Profiling.hpp b/src/armnn/Profiling.hpp
index 758b753..ef6bfd5 100644
--- a/src/armnn/Profiling.hpp
+++ b/src/armnn/Profiling.hpp
@@ -74,10 +74,6 @@
         uint32_t m_Count;
     };
 
-    // Waits for a compute device to finish working to guarantee correct timings.
-    // Currently used exclusively when emitting profiling events denoting GPU work.
-    void WaitForDevice(Compute compute) const;
-
     template<typename EventIterType>
     void AnalyzeEventSequenceAndWriteResults(EventIterType first, EventIterType last, std::ostream& outStream) const;