Return error when Kernel SIMD size is not in expected range

Change-Id: Ic4411535cd97f6e4e2c0252c43ab90c78713c5c5
Signed-off-by: Dunajski, Bartosz <bartosz.dunajski@intel.com>

Author: BartoszDunajski

Jenkinsfile

@@ -1,5 +1,5 @@
 #!groovy
 dependenciesRevision='b67325e72bcd5baa03886d87cac9afe6aea8cebd-1371'
 strategy='EQUAL'
-allowedCD=219
+allowedCD=221
 allowedF=11

runtime/kernel/kernel.cpp

@@ -195,6 +195,13 @@ cl_int Kernel::initialize() {
         const auto &patchInfo = kernelInfo.patchInfo;
 
         reconfigureKernel();
+        auto &hwInfo = device.getHardwareInfo();
+        auto &hwHelper = HwHelper::get(hwInfo.platform.eRenderCoreFamily);
+        auto maxSimdSize = getKernelInfo().getMaxSimdSize();
+
+        if (maxSimdSize != 1 && maxSimdSize < hwHelper.getMinimalSIMDSize()) {
+            return CL_INVALID_KERNEL;
+        }
 
         crossThreadDataSize = patchInfo.dataParameterStream
                                   ? patchInfo.dataParameterStream->DataParameterStreamSize
@@ -242,8 +249,8 @@ cl_int Kernel::initialize() {
             preferredWkgMultipleOffset = workloadInfo.preferredWkgMultipleOffset != WorkloadInfo::undefinedOffset ? ptrOffset(crossThread, workloadInfo.preferredWkgMultipleOffset) : preferredWkgMultipleOffset;
 
             *maxWorkGroupSizeForCrossThreadData = maxKernelWorkGroupSize;
-            *dataParameterSimdSize = getKernelInfo().getMaxSimdSize();
-            *preferredWkgMultipleOffset = getKernelInfo().getMaxSimdSize();
+            *dataParameterSimdSize = maxSimdSize;
+            *preferredWkgMultipleOffset = maxSimdSize;
             *parentEventOffset = WorkloadInfo::invalidParentEvent;
         }
 
@@ -359,7 +366,7 @@ cl_int Kernel::initialize() {
                 usingBuffers = true;
                 allBufferArgsStateful &= static_cast<uint32_t>(argInfo.pureStatefulBufferAccess);
                 this->auxTranslationRequired |= !kernelInfo.kernelArgInfo[i].pureStatefulBufferAccess &&
-                                                HwHelper::renderCompressedBuffersSupported(getDevice().getHardwareInfo());
+                                                HwHelper::renderCompressedBuffersSupported(hwInfo);
             } else if (argInfo.isDeviceQueue) {
                 kernelArgHandlers[i] = &Kernel::setArgDevQueue;
                 kernelArguments[i].type = DEVICE_QUEUE_OBJ;
@@ -368,7 +375,7 @@ cl_int Kernel::initialize() {
             }
         }
 
-        auxTranslationRequired &= HwHelper::get(device.getHardwareInfo().platform.eRenderCoreFamily).requiresAuxResolves();
+        auxTranslationRequired &= hwHelper.requiresAuxResolves();
 
         if (DebugManager.flags.DisableAuxTranslation.get()) {
             auxTranslationRequired = false;

unit_tests/kernel/kernel_tests.cpp

@@ -2732,6 +2732,36 @@ TEST(KernelTest, whenNullAllocationThenAssignNullPointerToCacheFlushVector) {
     EXPECT_EQ(nullptr, kernel.mockKernel->kernelArgRequiresCacheFlush[0]);
 }
 
+TEST(KernelTest, givenKernelCompiledWithSimdSizeLowerThanExpectedWhenInitializingThenReturnError) {
+    auto device = std::make_unique<MockClDevice>(MockDevice::createWithNewExecutionEnvironment<MockDevice>(platformDevices[0]));
+    auto minSimd = HwHelper::get(device->getHardwareInfo().platform.eRenderCoreFamily).getMinimalSIMDSize();
+    MockKernelWithInternals kernel(*device);
+    kernel.executionEnvironment.CompiledSIMD32 = 0;
+    kernel.executionEnvironment.CompiledSIMD16 = 0;
+    kernel.executionEnvironment.CompiledSIMD8 = 1;
+
+    cl_int retVal = kernel.mockKernel->initialize();
+
+    if (minSimd > 8) {
+        EXPECT_EQ(CL_INVALID_KERNEL, retVal);
+    } else {
+        EXPECT_EQ(CL_SUCCESS, retVal);
+    }
+}
+
+TEST(KernelTest, givenKernelCompiledWithSimdOneWhenInitializingThenReturnError) {
+    auto device = std::make_unique<MockClDevice>(MockDevice::createWithNewExecutionEnvironment<MockDevice>(platformDevices[0]));
+    MockKernelWithInternals kernel(*device);
+    kernel.executionEnvironment.CompiledSIMD32 = 0;
+    kernel.executionEnvironment.CompiledSIMD16 = 0;
+    kernel.executionEnvironment.CompiledSIMD8 = 0;
+    kernel.executionEnvironment.LargestCompiledSIMDSize = 1;
+
+    cl_int retVal = kernel.mockKernel->initialize();
+
+    EXPECT_EQ(CL_SUCCESS, retVal);
+}
+
 TEST(KernelTest, whenAllocationRequiringCacheFlushThenAssignAllocationPointerToCacheFlushVector) {
     MockGraphicsAllocation mockAllocation;
     auto device = std::make_unique<MockClDevice>(MockDevice::createWithNewExecutionEnvironment<MockDevice>(platformDevices[0]));