fix: add missing checks to return proper OpenCL error from API

changes affect clEnqueueNDCountKernelINTEL and
clGetKernelMaxConcurrentWorkGroupCountINTEL

Related-To: NEO-9688

Signed-off-by: Aleksander Czerwionka <aleksander.czerwionka@intel.com>

opencl/source/api/api.cpp

@@ -6066,6 +6066,12 @@ cl_int CL_API_CALL clGetKernelMaxConcurrentWorkGroupCountINTEL(cl_command_queue
         return retVal;
     }
 
+    for (size_t i = 0; i < workDim; i++) {
+        if (localWorkSize[i] == 0) {
+            return CL_INVALID_WORK_GROUP_SIZE;
+        }
+    }
+
     withCastToInternal(commandQueue, &pCommandQueue);
     *suggestedWorkGroupCount = pKernel->getMaxWorkGroupCount(workDim, localWorkSize, pCommandQueue);
 
@@ -6112,6 +6118,9 @@ cl_int CL_API_CALL clEnqueueNDCountKernelINTEL(cl_command_queue commandQueue,
     pKernel = pMultiDeviceKernel->getKernel(rootDeviceIndex);
     size_t globalWorkSize[3];
     for (size_t i = 0; i < workDim; i++) {
+        if (localWorkSize[i] == 0) {
+            return CL_INVALID_WORK_GROUP_SIZE;
+        }
         globalWorkSize[i] = workgroupCount[i] * localWorkSize[i];
     }
 

opencl/test/unit_test/api/cl_get_kernel_max_concurrent_work_group_count_intel_tests.inl

@@ -52,6 +52,10 @@ TEST_F(clGetKernelMaxConcurrentWorkGroupCountTests, GivenInvalidInputWhenCalling
     retVal = clGetKernelMaxConcurrentWorkGroupCountINTEL(pCommandQueue, pMultiDeviceKernel, workDim,
                                                          globalWorkOffset, nullptr, &suggestedWorkGroupCount);
     EXPECT_EQ(CL_INVALID_WORK_GROUP_SIZE, retVal);
+
+    retVal = clGetKernelMaxConcurrentWorkGroupCountINTEL(pCommandQueue, pMultiDeviceKernel, workDim,
+                                                         globalWorkOffset, localWorkSize, &suggestedWorkGroupCount);
+    EXPECT_EQ(CL_INVALID_WORK_GROUP_SIZE, retVal);
 }
 
 TEST_F(clGetKernelMaxConcurrentWorkGroupCountTests, GivenVariousInputWhenGettingMaxConcurrentWorkGroupCountThenCorrectValuesAreReturned) {

opencl/test/unit_test/command_queue/enqueue_kernel_1_tests.cpp

@@ -391,6 +391,20 @@ TEST_F(EnqueueKernelTest, givenKernelWhenAllArgsAreSetThenClEnqueueNDCountKernel
     EXPECT_EQ(CL_SUCCESS, retVal);
 }
 
+TEST_F(EnqueueKernelTest, givenLocalWorkSizeEqualZeroThenClEnqueueNDCountKernelINTELReturnsError) {
+    size_t workgroupCount[3] = {1, 1, 1};
+    size_t localWorkSize[3] = {0, 1, 1};
+    cl_int retVal = CL_SUCCESS;
+    std::unique_ptr<MultiDeviceKernel> pMultiDeviceKernel(MultiDeviceKernel::create(pProgram, pProgram->getKernelInfosForKernel("CopyBuffer"), retVal));
+
+    retVal = clEnqueueNDCountKernelINTEL(pCmdQ, pMultiDeviceKernel.get(), 1, nullptr, workgroupCount, localWorkSize, 0, nullptr, nullptr);
+    EXPECT_EQ(CL_INVALID_WORK_GROUP_SIZE, retVal);
+
+    pMultiDeviceKernel.get()->setKernelExecutionType(CL_KERNEL_EXEC_INFO_CONCURRENT_TYPE_INTEL);
+    retVal = clEnqueueNDCountKernelINTEL(pCmdQ, pMultiDeviceKernel.get(), 1, nullptr, workgroupCount, localWorkSize, 0, nullptr, nullptr);
+    EXPECT_EQ(CL_INVALID_WORK_GROUP_SIZE, retVal);
+}
+
 TEST_F(EnqueueKernelTest, givenKernelWhenNotAllArgsAreSetButSetKernelArgIsCalledTwiceThenClEnqueueNDCountKernelINTELReturnsError) {
     const size_t n = 512;
     size_t workgroupCount[3] = {2, 1, 1};

shared/source/helpers/kernel_helpers.cpp

@@ -45,7 +45,7 @@ uint32_t KernelHelper::getMaxWorkGroupCount(const RootDeviceEnvironment &rootDev
     for (uint32_t i = 1; i < workDim; i++) {
         workGroupSize *= localWorkSize[i];
     }
-
+    UNRECOVERABLE_IF(workGroupSize == 0);
     auto numThreadsPerThreadGroup = static_cast<uint32_t>(Math::divideAndRoundUp(workGroupSize, kernelDescriptor.kernelAttributes.simdSize));
     auto maxWorkGroupsCount = availableThreadCount / numThreadsPerThreadGroup;