[Runtime] Support arbitrary bitwidth integer in LLVM execution engine (…

…#37)

* Add T test

* Add make_any_bitwidth_np_array

* Reorg new_arg

* Fix reverting struct array to int array

* Fix output type

* Add i8 i16 return

* Add support for i<16

* Add typedef to global_var

* Fix return_dtype

* Add np_supported_types

* Fix np_type

* Cast np_type

* Fix pylint

* Add UInt in builder

* Update tests

* Fix return arg

* Fix return same

allo/customize.py

@@ -34,6 +34,7 @@
 from .ir.builder import ASTTransformer
 from .ir.infer import TypeInferer
 from .ir.transform import get_affine_loop_nests, find_loop_in_bands
+from .ir.types import AlloType
 from .build_module import _mlir_lower_pipeline, lower_linalg_and_attach_names
 from .module import LLVMModule, HLSModule
 
@@ -54,9 +55,10 @@ def _get_global_vars(_func):
     global_vars = _func.__globals__.copy()
 
     # Get back to the outer-most scope (user-defined function)
-    # Mainly used to get the annotation definitions, which are probably not defined in __globals__
+    # Mainly used to get the annotation definitions (shape and type),
+    # which are probably not defined in __globals__
     for name, var in inspect.stack()[2][0].f_locals.items():
-        if isinstance(var, (int, float)) or inspect.isfunction(var):
+        if isinstance(var, (int, float, AlloType)) or inspect.isfunction(var):
             global_vars[name] = var
 
     freevar_names = _func.__code__.co_freevars

allo/ir/builder.py

@@ -303,16 +303,19 @@ def build_general_binop(ctx, node, lhs, rhs):
             ast.Add: {
                 Float: arith_d.AddFOp,
                 Int: arith_d.AddIOp,
+                UInt: arith_d.AddIOp,
                 Fixed: hcl_d.AddFixedOp,
             },
             ast.Sub: {
                 Float: arith_d.SubFOp,
                 Int: arith_d.SubIOp,
+                UInt: arith_d.SubIOp,
                 Fixed: hcl_d.SubFixedOp,
             },
             ast.Mult: {
                 Float: arith_d.MulFOp,
                 Int: arith_d.MulIOp,
+                UInt: arith_d.MulIOp,
                 Fixed: hcl_d.MulFixedOp,
             },
             ast.Div: {

allo/ir/infer.py

@@ -225,9 +225,6 @@ def visit_AnnAssign(ctx, node):
                 rhs = TypeInferer.visit_constant_tensor(ctx, node)
             else:
                 raise RuntimeError("Unsupported data type")
-            # assert (
-            #     rhs.dtype == target_dtype
-            # ), f"Type mismatch, got {rhs.dtype} and {target_dtype} for {node.__class__.__name__} `{node.target.id}`"
             if not isinstance(node.value, ast.Constant):
                 assert (
                     rhs.shape == target_shape