[init] Adding CUDA language/compiler and CodePrinter (#32)

This PR aims to make the C code compilable using nvcc. The cuda language was added as well as a CudaCodePrinter.

Changes to stdlib:

Wrapped expressions using complex types in an `ifndef __NVCC__` to avoid processing them with the nvcc compiler

---------

Co-authored-by: Mouad Elalj, EmilyBourne

.dict_custom.txt

@@ -106,5 +106,6 @@ Valgrind
 variadic
 subclasses
 oneAPI
+Cuda
 getter
 setter

.github/actions/pytest_parallel/action.yml

@@ -10,8 +10,8 @@ runs:
   steps:
     - name: Test with pytest
       run: |
-        mpiexec -n 4 ${MPI_OPTS} python -m pytest epyccel/test_parallel_epyccel.py -v -m parallel -rXx
-        #mpiexec -n 4 ${MPI_OPTS} python -m pytest epyccel -v -m parallel -rXx
+        mpiexec -n 4 ${MPI_OPTS} python -m pytest epyccel/test_parallel_epyccel.py -v -m "parallel and not cuda" -rXx
+        #mpiexec -n 4 ${MPI_OPTS} python -m pytest epyccel -v -m "parallel and not cuda" -rXx
       shell: ${{ inputs.shell_cmd }}
       working-directory: ./tests
 

.github/actions/pytest_run/action.yml

@@ -51,13 +51,13 @@ runs:
       working-directory: ./tests
       id: pytest_3
     - name: Test Fortran translations
-      run: python -m pytest -n auto -rX ${FLAGS} -m "not (parallel or xdist_incompatible) and not (c or python or ccuda) ${{ inputs.pytest_mark }}" --ignore=symbolic --ignore=ndarrays 2>&1 | tee s4_outfile.out
+      run: python -m pytest -n auto -rX ${FLAGS} -m "not (parallel or xdist_incompatible) and not (c or python or cuda) ${{ inputs.pytest_mark }}" --ignore=symbolic --ignore=ndarrays 2>&1 | tee s4_outfile.out
       shell: ${{ inputs.shell_cmd }}
       working-directory: ./tests
       id: pytest_4
     - name: Test multi-file Fortran translations
       run: |
-        python -m pytest -rX ${FLAGS} -m "xdist_incompatible and not parallel and not (c or python or ccuda) ${{ inputs.pytest_mark }}" --ignore=symbolic --ignore=ndarrays | tee s5_outfile.out
+        python -m pytest -rX ${FLAGS} -m "xdist_incompatible and not parallel and not (c or python or cuda) ${{ inputs.pytest_mark }}" --ignore=symbolic --ignore=ndarrays | tee s5_outfile.out
         pyccel-clean
       shell: ${{ inputs.shell_cmd }}
       working-directory: ./tests

.github/actions/pytest_run_cuda/action.yml

@@ -1,4 +1,4 @@
-name: 'Pyccel pytest commands generating Ccuda'
+name: 'Pyccel pytest commands generating Cuda'
 inputs:
   shell_cmd:
     description: 'Specifies the shell command (different for anaconda)'
@@ -11,7 +11,14 @@ runs:
     - name: Ccuda tests with pytest
       run: |
         # Catch exit 5 (no tests found)
-        sh -c 'python -m pytest -n auto -rx -m "not (parallel or xdist_incompatible) and ccuda" --ignore=symbolic --ignore=ndarrays; ret=$?; [ $ret = 5 ] && exit 0 || exit $ret'
+        python -m pytest -rX ${FLAGS} -m "not (xdist_incompatible or parallel) and cuda ${{ inputs.pytest_mark }}" --ignore=symbolic --ignore=ndarrays 2>&1 | tee s1_outfile.out
         pyccel-clean
       shell: ${{ inputs.shell_cmd }}
       working-directory: ./tests
+    - name: Final step
+      if: always()
+      id: status
+      run:
+        python ci_tools/json_pytest_output.py -t "Cuda Test Summary" --tests "Cuda tests:${{ steps.pytest_1.outcome }}:tests/s1_outfile.out"
+
+      shell: ${{ inputs.shell_cmd }}

CHANGELOG.md

@@ -1,6 +1,12 @@
 # Change Log
 All notable changes to this project will be documented in this file.
 
+## \[Cuda - UNRELEASED\]
+
+### Added
+
+-   #32 : add support for `nvcc` Compiler and `cuda` language as a possible option.
+
 ## \[UNRELEASED\]
 
 ### Added

ci_tools/json_pytest_output.py

@@ -61,7 +61,7 @@ def     mini_md_summary(title, outcome, failed_tests):
     summary = ""
 
     failed_pattern = re.compile(r".*FAILED.*")
-    languages = ('c', 'fortran', 'python')
+    languages = ('c', 'fortran', 'python', 'cuda')
     pattern = {lang: re.compile(r".*\["+lang+r"\]\ \_.*") for lang in languages}
 
     for i in p_args.tests:

pyccel/codegen/codegen.py

@@ -9,34 +9,41 @@
 from pyccel.codegen.printing.fcode  import FCodePrinter
 from pyccel.codegen.printing.ccode  import CCodePrinter
 from pyccel.codegen.printing.pycode import PythonCodePrinter
+from pyccel.codegen.printing.cucode import CudaCodePrinter
 
 from pyccel.ast.core      import FunctionDef, Interface, ModuleHeader
 from pyccel.errors.errors import Errors
 from pyccel.utilities.stage import PyccelStage
 
-_extension_registry = {'fortran': 'f90', 'c':'c',  'python':'py'}
-_header_extension_registry = {'fortran': None, 'c':'h',  'python':None}
+_extension_registry = {'fortran': 'f90', 'c':'c',  'python':'py', 'cuda':'cu'}
+_header_extension_registry = {'fortran': None, 'c':'h',  'python':None, 'cuda':'h'}
 printer_registry    = {
                         'fortran':FCodePrinter,
                         'c':CCodePrinter,
-                        'python':PythonCodePrinter
+                        'python':PythonCodePrinter,
+                        'cuda':CudaCodePrinter
                       }
 
 pyccel_stage = PyccelStage()
 
 class Codegen(object):
 
-    """Abstract class for code generator."""
+    """
+    Class which handles the generation of code.
 
-    def __init__(self, parser, name):
-        """Constructor for Codegen.
-
-        parser: pyccel parser
+    The class which handles the generation of code. This is done by creating an appropriate class
+    inheriting from `CodePrinter` and using it to create strings describing the code that should
+    be printed. This class then takes care of creating the necessary files.
 
+    Parameters
+    ----------
+    parser : SemanticParser
+        The Pyccel Semantic parser node.
+    name : str
+        Name of the generated module or program.
+    """
 
-        name: str
-            name of the generated module or program.
-        """
+    def __init__(self, parser, name):
         pyccel_stage.set_stage('codegen')
         self._parser   = parser
         self._ast      = parser.ast
@@ -135,12 +142,22 @@ def language(self):
         return self._language
 
     def set_printer(self, **settings):
-        """ Set the current codeprinter instance"""
+        """
+        Set the current codeprinter instance.
+        
+        Getting the language that will be used (default language used is fortran),
+        Then instantiating the codePrinter with the corresponding language.
+
+        Parameters
+        ----------
+        **settings : dict
+            Any additional arguments which are necessary for CCodePrinter.
+        """
         # Get language used (default language used is fortran)
         language = settings.pop('language', 'fortran')
 
         # Set language
-        if not language in ['fortran', 'c', 'python']:
+        if not language in ['fortran', 'c', 'python', 'cuda']:
             raise ValueError('{} language is not available'.format(language))
         self._language = language
 

pyccel/codegen/compiling/compilers.py

@@ -441,7 +441,10 @@ def compile_shared_library(self, compile_obj, output_folder, verbose = False, sh
         # Collect compile information
         exec_cmd, includes, libs_flags, libdirs_flags, m_code = \
                 self._get_compile_components(compile_obj, accelerators)
-        linker_libdirs_flags = ['-Wl,-rpath' if l == '-L' else l for l in libdirs_flags]
+        if self._info['exec'] == 'nvcc':
+            linker_libdirs_flags = ['-Xcompiler' if l == '-L' else f'"-Wl,-rpath,{l}"' for l in libdirs_flags]
+        else:
+            linker_libdirs_flags = ['-Wl,-rpath' if l == '-L' else l for l in libdirs_flags]
 
         flags.insert(0,"-shared")
 

pyccel/codegen/pipeline.py

@@ -179,9 +179,10 @@ def handle_error(stage):
     if language is None:
         language = 'fortran'
 
-    # Choose Fortran compiler
+    # Choose Default compiler
     if compiler is None:
-        compiler = os.environ.get('PYCCEL_DEFAULT_COMPILER', 'GNU')
+        default_compiler_family = 'nvidia' if language == 'cuda' else 'GNU'
+        compiler = os.environ.get('PYCCEL_DEFAULT_COMPILER', default_compiler_family)
 
     fflags = [] if fflags is None else fflags.split()
     wrapper_flags = [] if wrapper_flags is None else wrapper_flags.split()

pyccel/codegen/printing/cucode.py

@@ -0,0 +1,74 @@
+# coding: utf-8
+#------------------------------------------------------------------------------------------#
+# This file is part of Pyccel which is released under MIT License. See the LICENSE file or #
+# go to https://github.com/pyccel/pyccel/blob/master/LICENSE for full license details.     #
+#------------------------------------------------------------------------------------------#
+"""
+Provide tools for generating and handling CUDA code.
+This module is designed to interface Pyccel's Abstract Syntax Tree (AST) with CUDA,
+enabling the direct translation of high-level Pyccel expressions into CUDA code.
+"""
+
+from pyccel.codegen.printing.ccode import CCodePrinter, c_library_headers
+
+from pyccel.ast.core        import Import, Module
+
+from pyccel.errors.errors   import Errors
+
+
+errors = Errors()
+
+__all__ = ["CudaCodePrinter"]
+
+class CudaCodePrinter(CCodePrinter):
+    """
+    Print code in CUDA format.
+
+    This printer converts Pyccel's Abstract Syntax Tree (AST) into strings of CUDA code.
+    Navigation through this file utilizes _print_X functions,
+    as is common with all printers.
+
+    Parameters
+    ----------
+    filename : str
+            The name of the file being pyccelised.
+    prefix_module : str
+            A prefix to be added to the name of the module.
+    """
+    language = "cuda"
+
+    def __init__(self, filename, prefix_module = None):
+
+        errors.set_target(filename, 'file')
+
+        super().__init__(filename)
+
+    def _print_Module(self, expr):
+        self.set_scope(expr.scope)
+        self._current_module = expr.name
+        body = ''.join(self._print(i) for i in expr.body)
+
+        global_variables = ''.join(self._print(d) for d in expr.declarations)
+
+        # Print imports last to be sure that all additional_imports have been collected
+        imports = [Import(expr.name, Module(expr.name,(),())), *self._additional_imports.values()]
+        c_headers_imports = ''
+        local_imports = ''
+
+        for imp in imports:
+            if imp.source in c_library_headers:
+                c_headers_imports += self._print(imp)
+            else:
+                local_imports += self._print(imp)
+
+        imports = f'{c_headers_imports}\
+                    extern "C"{{\n\
+                    {local_imports}\
+                    }}'
+
+        code = f'{imports}\n\
+                 {global_variables}\n\
+                 {body}\n'
+
+        self.exit_scope()
+        return code

pyccel/commands/console.py

@@ -80,7 +80,7 @@ def pyccel(files=None, mpi=None, openmp=None, openacc=None, output_dir=None, com
     # ... backend compiler options
     group = parser.add_argument_group('Backend compiler options')
 
-    group.add_argument('--language', choices=('fortran', 'c', 'python'), help='Generated language')
+    group.add_argument('--language', choices=('fortran', 'c', 'python', 'cuda'), help='Generated language')
 
     group.add_argument('--compiler', help='Compiler family or json file containing a compiler description {GNU,intel,PGI}')
 

pyccel/compilers/default_compilers.py

@@ -185,6 +185,15 @@
                 },
             'family': 'nvidia',
             }
+#------------------------------------------------------------
+nvcc_info = {'exec'         : 'nvcc',
+             'language'     : 'cuda',
+             'debug_flags'  : ("-g",),
+             'release_flags': ("-O3",),
+             'general_flags': ('--compiler-options', '-fPIC',),
+             'family'       : 'nvidia'
+            }
+
 
 #------------------------------------------------------------
 def change_to_lib_flag(lib):
@@ -288,6 +297,7 @@ def change_to_lib_flag(lib):
 pgfortran_info.update(python_info)
 nvc_info.update(python_info)
 nvfort_info.update(python_info)
+nvcc_info.update(python_info)
 
 available_compilers = {('GNU', 'c') : gcc_info,
                        ('GNU', 'fortran') : gfort_info,
@@ -296,6 +306,7 @@ def change_to_lib_flag(lib):
                        ('PGI', 'c') : pgcc_info,
                        ('PGI', 'fortran') : pgfortran_info,
                        ('nvidia', 'c') : nvc_info,
-                       ('nvidia', 'fortran') : nvfort_info}
+                       ('nvidia', 'fortran') : nvfort_info,
+                       ('nvidia', 'cuda'): nvcc_info}
 
 vendors = ('GNU','intel','PGI','nvidia')

pyccel/naming/__init__.py

@@ -10,7 +10,9 @@
 from .fortrannameclashchecker import FortranNameClashChecker
 from .cnameclashchecker import CNameClashChecker
 from .pythonnameclashchecker import PythonNameClashChecker
+from .cudanameclashchecker import CudaNameClashChecker
 
 name_clash_checkers = {'fortran':FortranNameClashChecker(),
         'c':CNameClashChecker(),
-        'python':PythonNameClashChecker()}
+        'python':PythonNameClashChecker(),
+        'cuda':CudaNameClashChecker()}