Get rid of gpu kernel api stubs (#203)

Author: Hardcode84

dpcomp_gpu_runtime/CMakeLists.txt

@@ -22,7 +22,6 @@ endif()
 
 set(SOURCES_LIST
     lib/gpu_runtime.cpp
-    lib/kernel_api_stubs.cpp
     )
 set(HEADERS_LIST
     lib/level_zero_printing.hpp

dpcomp_gpu_runtime/lib/kernel_api_stubs.cpp

@@ -45,18 +45,25 @@ def _register_funcs():
             mlir_func_name("get_local_id"),
             mlir_func_name("get_global_size"),
             mlir_func_name("get_local_size"),
+            mlir_func_name("kernel_barrier"),
+            mlir_func_name("kernel_mem_fence"),
         ]
 
-        _atomic_ops = ["add", "sub"]
-        _atomic_ops_types = ["int32", "int64", "float32", "float64"]
-
         from itertools import product
 
-        for o, t in product(_atomic_ops, _atomic_ops_types):
-            _funcs.append(mlir_func_name("atomic_" + o + "_" + t))
+        _types = ["int32", "int64", "float32", "float64"]
+        _atomic_ops = ["add", "sub"]
+        for o, t in product(_atomic_ops, _types):
+            _funcs.append(mlir_func_name(f"atomic_{o}_{t}"))
+
+        for n, t in product(range(8), _types):
+            _funcs.append(mlir_func_name(f"local_array_{t}_{n}"))
 
         for name in _funcs:
-            func = getattr(runtime_lib, name)
+            if hasattr(runtime_lib, name):
+                func = getattr(runtime_lib, name)
+            else:
+                func = 1
             register_cfunc(ll, name, func)
 
         _alloc_func = runtime_lib.dpcompGpuSetMemInfoAllocFunc

numba_dpcomp/numba_dpcomp/mlir/gpu_runtime.py

@@ -135,7 +135,6 @@ def _gen_tests():
         "test_namedtuple3",  # namedtuple
         "test_issue6102",  # list support
         "test_oversized_tuple_as_arg_to_kernel",  # UnsupportedParforsError not raised
-        "test_issue5942_2",  # invalid result
         "test_parfor_ufunc_typing",  # np.isinf
         "test_issue_5098",  # list support and more
         "test_ufunc_expr",  # np.bitwise_and(