add handle for custom device (#61)

This pull request addresses issue
#41 by implementing a new
feature in Pyccel that allows users to define a custom device

**Commit Summary**

- Adding handler for custom device and its code generation.
- Adding test

---------

Co-authored-by: EmilyBourne <[email protected]>

Author: smazouz42

CHANGELOG.md

@@ -10,6 +10,7 @@ All notable changes to this project will be documented in this file.
 -   #59 : Updated `cuda` clash checker.
 -   #42 : Add support for custom kernel in`cuda`.
 -   #42 : Add Cuda module to Pyccel. Add support for `cuda.synchronize` function.
+-   #41 : Add support for custom device in`cuda`.
 
 ## \[UNRELEASED\]
 

docs/cuda.md

@@ -20,4 +20,27 @@ threadsperblock = 1
 # Call your kernel function
 my_kernel[blockspergrid, threadsperblock]()
 
-```
+```
+
+### device
+
+Device functions are similar to kernels, but are executed within the context of a kernel. They can be called only from kernels or device functions, and are typically used for operations that are too small to justify launching a separate kernel, or for operations that need to be performed repeatedly within the context of a kernel.
+
+```python
+from pyccel.decorators import device, kernel
+
+@device
+def add(x, y):
+    return x + y
+
+@kernel
+def my_kernel():
+    x = 1
+    y = 2
+    z = add(x, y)
+    print(z)
+
+my_kernel[1, 1]()
+
+```
+

pyccel/codegen/printing/cucode.py

@@ -86,9 +86,10 @@ def function_signature(self, expr, print_arg_names = True):
         str
             Signature of the function.
         """
-        cuda_decorater = '__global__' if 'kernel' in expr.decorators else ''
+        cuda_decorator = '__global__' if 'kernel' in expr.decorators else \
+        '__device__' if 'device' in expr.decorators else ''
         c_function_signature = super().function_signature(expr, print_arg_names)
-        return f'{cuda_decorater} {c_function_signature}'
+        return f'{cuda_decorator} {c_function_signature}'
 
     def _print_KernelCall(self, expr):
         func = expr.funcdef
@@ -109,7 +110,7 @@ def _print_ModuleHeader(self, expr):
         cuda_headers = ""
         for f in expr.module.funcs:
             if not f.is_inline:
-                if 'kernel' in f.decorators:  # Checking for 'kernel' decorator
+                if 'kernel' in f.decorators or 'device' in f.decorators:
                     cuda_headers += self.function_signature(f) + ';\n'
                 else:
                     funcs += self.function_signature(f) + ';\n'

pyccel/decorators.py

@@ -11,6 +11,7 @@
 __all__ = (
     'allow_negative_index',
     'bypass',
+    'device',
     'elemental',
     'inline',
     'private',
@@ -141,3 +142,21 @@ def __getitem__(self, args):
             return self._f
 
     return KernelAccessor(f)
+
+def device(f):
+    """
+    Decorator for marking a function as a GPU device function.
+
+    This decorator is used to mark a Python function as a GPU device function.
+
+    Parameters
+    ----------
+    f : Function
+        The function to be marked as a device.
+
+    Returns
+    -------
+    f
+        The function marked as a device.
+    """
+    return f

pyccel/errors/messages.py

@@ -165,7 +165,7 @@
 INVALID_KERNEL_LAUNCH_CONFIG = 'Expected exactly 2 parameters for kernel launch'
 INVALID_KERNEL_CALL_BP_GRID = 'Invalid Block per grid parameter for Kernel call'
 INVALID_KERNEL_CALL_TP_BLOCK = 'Invalid Thread per Block parameter for Kernel call'
-
+INVAlID_DEVICE_CALL = 'A function decorated with "device" should be called only from a "kernel" or another "device" function.'
 
 
 

pyccel/parser/semantic.py

@@ -140,9 +140,10 @@
         UNUSED_DECORATORS, UNSUPPORTED_POINTER_RETURN_VALUE, PYCCEL_RESTRICTION_OPTIONAL_NONE,
         PYCCEL_RESTRICTION_PRIMITIVE_IMMUTABLE, PYCCEL_RESTRICTION_IS_ISNOT,
         FOUND_DUPLICATED_IMPORT, UNDEFINED_WITH_ACCESS, MACRO_MISSING_HEADER_OR_FUNC, PYCCEL_RESTRICTION_INHOMOG_SET,
-        MISSING_KERNEL_CONFIGURATION,
+        MISSING_KERNEL_CONFIGURATION, INVAlID_DEVICE_CALL,
         INVALID_KERNEL_LAUNCH_CONFIG, INVALID_KERNEL_CALL_BP_GRID, INVALID_KERNEL_CALL_TP_BLOCK)
 
+
 from pyccel.parser.base      import BasicParser
 from pyccel.parser.syntactic import SyntaxParser
 from pyccel.parser.syntax.headers import types_meta
@@ -1134,6 +1135,10 @@ def _handle_function(self, expr, func, args, *, is_method = False, use_build_fun
         FunctionCall/PyccelFunction
             The semantic representation of the call.
         """
+
+        if isinstance(func, FunctionDef) and 'device' in func.decorators:
+            if 'kernel' not in self.scope.decorators and 'device' not in self.scope.decorators:
+                errors.report(INVAlID_DEVICE_CALL,symbol=expr, severity='fatal')
         if isinstance(func, PyccelFunctionDef):
             if use_build_functions:
                 annotation_method = '_build_' + func.cls_name.__name__

tests/cuda/test_device_semantic.py

@@ -0,0 +1,31 @@
+# pylint: disable=missing-function-docstring, missing-module-docstring
+import pytest
+
+from pyccel import epyccel
+from pyccel.decorators import device
+from pyccel.errors.errors import Errors, PyccelSemanticError
+from pyccel.errors.messages import (INVAlID_DEVICE_CALL,)
+
+
+@pytest.mark.cuda
+def test_invalid_device_call():
+    def invalid_device_call():
+        @device
+        def device_call():
+            pass
+        def fake_kernel_call():
+            device_call()
+
+        fake_kernel_call()
+
+    errors = Errors()
+
+    with pytest.raises(PyccelSemanticError):
+        epyccel(invalid_device_call, language="cuda")
+
+    assert errors.has_errors()
+
+    assert errors.num_messages() == 1
+
+    error_info = [*errors.error_info_map.values()][0][0]
+    assert INVAlID_DEVICE_CALL == error_info.message

tests/pyccel/scripts/kernel/device_test.py

@@ -0,0 +1,18 @@
+# pylint: disable=missing-function-docstring, missing-module-docstring
+from pyccel.decorators import device, kernel
+from pyccel import cuda
+
+@device
+def device_call():
+    print("Hello from device")
+
+@kernel
+def kernel_call():
+    device_call()
+
+def f():
+    kernel_call[1,1]()
+    cuda.synchronize()
+
+if __name__ == '__main__':
+    f()

tests/pyccel/test_pyccel.py

@@ -709,6 +709,14 @@ def test_kernel_collision(gpu_available):
     pyccel_test("scripts/kernel/kernel_name_collision.py",
             language="cuda", execute_code=gpu_available)
 
+#------------------------------------------------------------------------------
+
+@pytest.mark.cuda
+def test_device_call(gpu_available):
+    types = str
+    pyccel_test("scripts/kernel/device_test.py",
+            language="cuda", output_dtype=types, execute_code=gpu_available)
+
 #------------------------------------------------------------------------------
 def test_print_strings(language):
     types = str