Add performance hints when kernel's argument requires aux translation

Resolves: NEO-2931

Change-Id: I3756265d0d8a774805b0b35088b7477b09b5a7bf
Signed-off-by: Jobczyk, Lukasz <lukasz.jobczyk@intel.com>

Author: LukaszJobczyk

runtime/context/driver_diagnostics.cpp

@@ -53,6 +53,7 @@ const char *DriverDiagnostics::hintFormat[] = {
     "Performance hint: Local workgroup sizes { %u, %u, %u } selected for this workload ( kernel name: %s ) may not be optimal, consider using following local workgroup size: { %u, %u, %u }.",                                           //BAD_LOCAL_WORKGROUP_SIZE
     "Performance hint: Kernel %s register pressure is too high, spill fills will be generated, additional surface needs to be allocated of size %u, consider simplifying your kernel.",                                                   //REGISTER_PRESSURE_TOO_HIGH
     "Performance hint: Kernel %s private memory usage is too high and exhausts register space, additional surface needs to be allocated of size %u, consider reducing amount of private memory used, avoid using private memory arrays.", //PRIVATE_MEMORY_USAGE_TOO_HIGH
-    "Performance hint: Kernel %s submission requires coherency with CPU; this will impact performance."                                                                                                                                   //KERNEL_REQUIRES_COHERENCY
+    "Performance hint: Kernel %s submission requires coherency with CPU; this will impact performance.",                                                                                                                                  //KERNEL_REQUIRES_COHERENCY
+    "Performance hint: Kernel %s requires aux translation on argument [%u] = \"%s\""                                                                                                                                                      //KERNEL_ARGUMENT_AUX_TRANSLATION
 };
 } // namespace NEO

runtime/context/driver_diagnostics.h

@@ -46,7 +46,8 @@ enum PerformanceHints {
     BAD_LOCAL_WORKGROUP_SIZE,
     REGISTER_PRESSURE_TOO_HIGH,
     PRIVATE_MEMORY_USAGE_TOO_HIGH,
-    KERNEL_REQUIRES_COHERENCY
+    KERNEL_REQUIRES_COHERENCY,
+    KERNEL_ARGUMENT_AUX_TRANSLATION
 };
 
 class DriverDiagnostics {

runtime/kernel/kernel.cpp

@@ -2149,6 +2149,12 @@ void Kernel::fillWithBuffersForAuxTranslation(MemObjsForAuxTranslation &memObjsF
             auto buffer = castToObject<Buffer>(getKernelArg(i));
             if (buffer && buffer->getGraphicsAllocation()->getAllocationType() == GraphicsAllocation::AllocationType::BUFFER_COMPRESSED) {
                 memObjsForAuxTranslation.insert(buffer);
+
+                auto &context = this->program->getContext();
+                if (context.isProvidingPerformanceHints()) {
+                    context.providePerformanceHint(CL_CONTEXT_DIAGNOSTICS_LEVEL_BAD_INTEL, KERNEL_ARGUMENT_AUX_TRANSLATION,
+                                                   kernelInfo.name.c_str(), i, kernelInfo.kernelArgInfo.at(i).name.c_str());
+                }
             }
         }
     }

unit_tests/context/driver_diagnostics_tests.cpp

@@ -424,6 +424,35 @@ TEST_F(PerformanceHintTest, givenPrintDriverDiagnosticsDebugModeEnabledWhenConte
     context->release();
 }
 
+TEST_F(PerformanceHintTest, givenPrintDriverDiagnosticsDebugModeEnabledWhenCallFillWithBuffersForAuxTranslationThenContextProvidesProperHint) {
+    DebugManagerStateRestore dbgRestore;
+    DebugManager.flags.PrintDriverDiagnostics.set(1);
+
+    auto pDevice = castToObject<Device>(devices[0]);
+    MockKernelWithInternals mockKernel(*pDevice, context);
+    MockBuffer buffer;
+    cl_mem clMem = &buffer;
+
+    buffer.getGraphicsAllocation()->setAllocationType(GraphicsAllocation::AllocationType::BUFFER_COMPRESSED);
+    mockKernel.kernelInfo.kernelArgInfo.resize(1);
+    mockKernel.kernelInfo.kernelArgInfo.at(0).kernelArgPatchInfoVector.resize(1);
+    mockKernel.kernelInfo.kernelArgInfo.at(0).pureStatefulBufferAccess = false;
+    mockKernel.mockKernel->initialize();
+    mockKernel.mockKernel->auxTranslationRequired = true;
+    mockKernel.mockKernel->setArgBuffer(0, sizeof(cl_mem *), &clMem);
+
+    testing::internal::CaptureStdout();
+    MemObjsForAuxTranslation memObjects;
+    mockKernel.mockKernel->fillWithBuffersForAuxTranslation(memObjects);
+
+    snprintf(expectedHint, DriverDiagnostics::maxHintStringSize, DriverDiagnostics::hintFormat[KERNEL_ARGUMENT_AUX_TRANSLATION],
+             mockKernel.mockKernel->getKernelInfo().name.c_str(), 0, mockKernel.mockKernel->getKernelInfo().kernelArgInfo.at(0).name.c_str());
+
+    std::string output = testing::internal::GetCapturedStdout();
+    EXPECT_NE(0u, output.size());
+    EXPECT_TRUE(containsHint(expectedHint, userData));
+}
+
 TEST_P(PerformanceHintKernelTest, GivenSpillFillWhenKernelIsInitializedThenContextProvidesProperHint) {
 
     auto pDevice = castToObject<Device>(devices[0]);