[SYCL][NATIVECPU] remove Comdat from kernel whose name was taken/renamed (#18726)

Taking the kernel name and renaming the original function afterwards
caused recently a number of check-sycl tests to fail on Windows:
```
SYCL :: check_device_code/native_cpu/native_cpu_builtins.cpp
SYCL :: check_device_code/native_cpu/vectorization.cpp
SYCL :: native_cpu/link-noinline.cpp
SYCL :: native_cpu/no-opt.cpp
SYCL :: native_cpu/vector-add.cpp
```
This is a quick fix for these fails by removing the Comdat from such
kernels. Maintaining the Comdat info properly for such kernels will be
addressed in a subsequent PR.
A new test was added to ensure launches from functions with multiple
definitions work.

Author: uwedolinsky

llvm/lib/SYCLNativeCPUUtils/PrepareSYCLNativeCPU.cpp

@@ -357,6 +357,7 @@ PreservedAnalyses PrepareSYCLNativeCPUPass::run(Module &M,
           if (OrigF->use_empty()) {
             RemovableFuncs.insert(OrigF);
           } else {
+            OrigF->setComdat(nullptr);
             OrigF->setName(Name + ".orig");
           }
         } else {

sycl/test/native_cpu/multiple_definitions.cpp

@@ -0,0 +1,49 @@
+// check kernel launches from function with multiple definitions work/link
+// REQUIRES: native_cpu
+// RUN: %clangxx -fsycl -fsycl-targets=native_cpu -DSOURCE1 %s -fno-inline -c -o %t1.o
+// RUN: %clangxx -fsycl -fsycl-targets=native_cpu -DSOURCE2 %s -fno-inline -c -o %t2.o
+// RUN: %clangxx -fsycl -fsycl-targets=native_cpu %t1.o %t2.o -fno-inline -mllvm -sycl-native-cpu-vecz-width=4 -o %t
+// RUN: env ONEAPI_DEVICE_SELECTOR="native_cpu:cpu" %t
+// RUN: %clangxx -fsycl -fsycl-targets=native_cpu %t1.o %t2.o -mllvm -sycl-native-cpu-vecz-width=4 -o %t2
+// RUN: env ONEAPI_DEVICE_SELECTOR="native_cpu:cpu" %t2
+
+#include <sycl/sycl.hpp>
+
+constexpr unsigned N = 5;
+using _Array = int[N];
+
+using namespace ::sycl;
+
+inline int func() {
+  queue deviceQueue;
+  sycl::range<1> range{N};
+  const device dev = deviceQueue.get_device();
+  const context ctx = deviceQueue.get_context();
+  auto OUT_acc = (_Array *)malloc_shared(sizeof(_Array), dev, ctx);
+  for (int i = 0; i < N; i++)
+    (*OUT_acc)[i] = 2;
+
+  deviceQueue.submit([&](handler &cgh) {
+    cgh.parallel_for<class Test>(
+        range, [=](sycl::id<1> ID) { (*OUT_acc)[ID.get(0)] = 1; });
+  });
+
+  deviceQueue.wait();
+
+  for (int i = 0; i < N; i++) {
+    if ((*OUT_acc)[i] != 1)
+      return 1;
+  }
+
+  sycl::free(OUT_acc, deviceQueue);
+  return 0;
+}
+
+#ifdef SOURCE1
+int (*fref)() = &func;
+#endif
+#ifdef SOURCE2
+int (*fref2)() = &func;
+extern int (*fref)();
+int main(void) { return fref() + fref2(); }
+#endif // SOURCE2