[CTS] Fix urEnqueueKernelLaunchIncrementMultiDeviceMultiThreadTest test

Do not run multiple kernels in parallel that modify the same
memory. To make this work, atomics would have to be used.

Instead, use separate memory for each kernel.

Author: igchor

test/conformance/enqueue/urEnqueueKernelLaunchAndMemcpyInOrder.cpp

@@ -343,8 +343,8 @@ TEST_P(urEnqueueKernelLaunchIncrementMultiDeviceMultiThreadTest, Success) {
             constexpr size_t n_dimensions = 1;
 
             auto queue = queuePerThread ? queues[i] : queues.back();
-            auto kernel = queuePerThread ? kernels[i] : kernels.back();
-            auto sharedPtr = queuePerThread ? SharedMem[i] : SharedMem.back();
+            auto kernel = kernels[i];
+            auto sharedPtr = SharedMem[i];
 
             std::vector<uur::raii::Event> Events(numOpsPerThread + 1);
             for (size_t j = 0; j < numOpsPerThread; j++) {
@@ -381,29 +381,12 @@ TEST_P(urEnqueueKernelLaunchIncrementMultiDeviceMultiThreadTest, Success) {
             size_t ExpectedValue = InitialValue;
             ExpectedValue += numOpsPerThread;
             for (uint32_t j = 0; j < ArraySize; ++j) {
-                if (queuePerThread) {
-                    ASSERT_EQ(data[j], ExpectedValue);
-                } else {
-                    // All threads write to the same memory, so the value might be greater
-                    ASSERT_GE(data[j], ExpectedValue);
-                }
+                ASSERT_EQ(data[j], ExpectedValue);
             }
         });
     }
 
     for (auto &thread : threads) {
         thread.join();
     }
-
-    if (!queuePerThread) {
-        size_t ExpectedValue = InitialValue;
-        ExpectedValue += numOpsPerThread * numThreads;
-
-        for (size_t i = 0; i < devices.size(); i++) {
-            for (uint32_t j = 0; j < ArraySize; ++j) {
-                ASSERT_EQ(reinterpret_cast<uint32_t *>(SharedMem.back())[j],
-                          ExpectedValue);
-            }
-        }
-    }
 }