work in progress

Author: smazouz42

pyccel/codegen/printing/ccode.py

@@ -46,6 +46,7 @@
 from pyccel.ast.numpytypes import NumpyFloat32Type, NumpyFloat64Type, NumpyComplex64Type, NumpyComplex128Type
 from pyccel.ast.numpytypes import NumpyNDArrayType, numpy_precision_map
 from pyccel.ast.cudatypes  import CudaArrayType
+from pyccel.ast.cudaext    import CudaFull
 
 from pyccel.ast.utilities import expand_to_loops
 
@@ -59,6 +60,7 @@
 
 from pyccel.codegen.printing.codeprinter import CodePrinter
 
+
 from pyccel.errors.errors   import Errors
 from pyccel.errors.messages import (PYCCEL_RESTRICTION_TODO, INCOMPATIBLE_TYPEVAR_TO_FUNC,
                                     PYCCEL_RESTRICTION_IS_ISNOT, UNSUPPORTED_ARRAY_RANK)
@@ -2181,6 +2183,9 @@ def _print_Assign(self, expr):
         # Inhomogenous tuples are unravelled and therefore do not exist in the c printer
         if isinstance(rhs, (NumpyArray, PythonTuple)):
             return prefix_code+self.copy_NumpyArray_Data(expr)
+        if(isinstance(rhs, (CudaFull))):
+            # TODO add support for CudaFull
+            return " \n"
         if isinstance(rhs, (NumpyFull)):
             return prefix_code+self.arrayFill(expr)
         lhs = self._print(expr.lhs)

pyccel/codegen/printing/cucode.py

@@ -116,8 +116,7 @@ def _print_CudaSynchronize(self, expr):
         return 'cudaDeviceSynchronize();\n'
 
     def _print_CudaEmpty(self, expr):
-        print(expr)
-        return 'cudaDeviceSynchronize();\n'
+        return 'cuda_array_create(1,  (int64_t[]){INT64_C(10)}, nd_double, false,allocateMemoryOnHost);\n'
     def _print_ModuleHeader(self, expr):
         self.set_scope(expr.module.scope)
         self._in_header = True
@@ -158,7 +157,7 @@ def _print_Allocate(self, expr):
         else:
             raise NotImplementedError(f"Don't know how to index {variable.class_type} type")
         shape_dtype = self.get_c_type(NumpyInt64Type())
-        shape_Assign = "("+ shape_dtype +"[]){" + shape + "}"
+        shape_Assign = "int64_t shape_Assign [] = {" + shape + "};\n"
         is_view = 'false' if variable.on_heap else 'true'
         memory_location = expr.variable.memory_location
         if memory_location in ('device', 'host'):
@@ -167,8 +166,8 @@ def _print_Allocate(self, expr):
             memory_location = 'managedMemory'
         self.add_import(c_imports['cuda_ndarrays'])
         self.add_import(c_imports['ndarrays'])
-        alloc_code = f"{self._print(expr.variable)} = cuda_array_create({variable.rank},  {shape_Assign}, {dtype}, {is_view},{memory_location});\n"
-        return f'{alloc_code}'
+        alloc_code = f"{self._print(expr.variable)} = cuda_array_create({variable.rank},  shape_Assign, {dtype}, {is_view},{memory_location});\n"
+        return f'{shape_Assign} {alloc_code}'
 
     def _print_Deallocate(self, expr):
         var_code = self._print(expr.variable)

pyccel/stdlib/cuda_ndarrays/cuda_ndarrays.h

@@ -5,6 +5,11 @@
 # include <iostream>
 #include "../ndarrays/ndarrays.h"
 
+t_ndarray   cuda_array_create(int32_t nd, int64_t *shape, enum e_types type, bool is_view ,
+enum e_memory_locations location);
+int32_t cuda_free_host(t_ndarray arr);
+
+
 using namespace std;