Merge branch 'ebourne_static_analysis_improvements' into 'main'

Static analysis improvements

See merge request gysela-developpers/gyselalibxx!674

ci_tools/gyselalib_static_analysis.py

@@ -41,6 +41,8 @@
 
 mirror_functions = {'create_mirror', 'create_mirror_and_copy', 'create_mirror_view', 'create_mirror_view_and_copy'}
 
+parallel_functions = ['parallel_for', 'parallel_for_each', 'parallel_transform_reduce']
+
 HOME_DIR = Path(__file__).parent.parent.absolute()
 global_folders = [HOME_DIR / f for f in ('src', 'simulations', 'tests')]
 
@@ -224,8 +226,8 @@ def search_for_unnecessary_auto(file):
                       for v,t in zip(variables, variable_types) if t is not None and t.attrib['str'] == 'auto']
 
     # Find the name and location of the first use of the auto variables
-    var_data_idx = [file.data.index(v.attrib) for v in auto_variables if v]
-    var_names = [v.attrib['str'] for v in auto_variables if v]
+    var_data_idx = [file.data.index(v.attrib) for v in auto_variables if v is not None]
+    var_names = [v.attrib['str'] for v in auto_variables if v is not None]
     for idx, var_name in enumerate(var_names):
         start = var_data_idx[idx]
         end = next(i for i,v in enumerate(file.data[start:], start) if v['str'] == ';')
@@ -240,6 +242,27 @@ def search_for_unnecessary_auto(file):
         if not any(v['str'] in chain(mirror_functions, auto_functions) for v in file.data[start:end]):
             report_error(ERROR, file, file.data[start]['linenr'], f"Please use explicit types instead of auto ({var_name})")
 
+    # Find auto in arguments of lambda functions
+    for elem in file.data_xml.findall(".token[@str='[']"):
+        start_idx = file.data.index(elem.attrib)+1
+        end_idx = next(j for j,a in enumerate(file.data[start_idx:], start_idx) if a['str'] == ']')
+        keys = ''.join(a['str'] for a in file.data[start_idx:end_idx]).split(',')
+        if not all(k[0] in ('&', '=') for k in keys):
+            continue
+        end_args_idx = next(j for j,a in enumerate(file.data[end_idx:], end_idx) if a['str'] == ')')
+        lambda_args = ' '.join(a['str'] for a in file.data[end_idx+2:end_args_idx]).split(',')
+        for a in lambda_args:
+            if 'auto' in a:
+                report_error(ERROR, file, file.data[start]['linenr'], f"Please use explicit types instead of auto ({a})")
+
+    for elem in chain(file.data_xml.findall(".token[@str='KOKKOS_CLASS_LAMBDA']"), file.data_xml.findall(".token[@str='KOKKOS_LAMBDA']")):
+        start_idx = file.data.index(elem.attrib)+3
+        end_args_idx = next(j for j,a in enumerate(file.data[start_idx:], start_idx) if a['str'] == ')')
+        lambda_args = ' '.join(a['str'] for a in file.data[start_idx:end_args_idx]).split(',')
+        for a in lambda_args:
+            if 'auto' in a.split():
+                report_error(ERROR, file, file.data[start]['linenr'], f"Please use explicit types instead of auto ({a})")
+
 def search_for_bad_create_mirror(file):
     """
     Test for instances where create_mirror functions are called incorrectly.
@@ -406,6 +429,52 @@ def check_directives(file):
         elif not possible_matches and include_str == '"':
             report_error(STYLE, file, linenr, f'Angle brackets should be used to include files from external libraries ({include_str}-><{include_file}>)')
 
+def check_kokkos_lambda_use(file):
+    """
+    Check that KOKKOS_LAMBDA expressions are present and that they are correctly used.
+
+    The checks are:
+    - Check that lambda functions passed to ddc::parallel_X functions use KOKKOS_LAMBDA or KOKKOS_CLASS_LAMBDA
+    - Check that class variables are not used in lambda functions passed to ddc::parallel_X functions.
+    """
+    for p in parallel_functions:
+        for elem in file.data_xml.findall(f".token[@str='{p}']"):
+            idx = file.data.index(elem.attrib)
+            open_brackets = file.data[idx+1]
+            close_brackets_id = open_brackets['link']
+            scope = open_brackets['scope']
+            end_idx = file.data.index(file.data_xml.find(f".token[@id='{close_brackets_id}'][@scope='{scope}']").attrib)
+            arg_splitters = [idx+1]
+            i = idx+3
+            while i < end_idx:
+                if file.data[i]['str'] == ',':
+                    arg_splitters.append(i)
+                    i+=1
+                elif file.data[i]['str'] in ('(','{', '['):
+                    close_brackets_id = file.data[i]['link']
+                    scope = file.data[i]['scope']
+                    i = file.data.index(file.data_xml.find(f".token[@id='{close_brackets_id}'][@scope='{scope}']").attrib)+1
+                else:
+                    i+=1
+
+            if file.data[arg_splitters[-1]+1]['str'] == '[':
+                last_arg = ' '.join(a['str'] for a in file.data[arg_splitters[-1]+1:end_idx])
+                correct_last_arg = 'KOKKOS_LAMBDA '+last_arg[last_arg.find(']')+1:]
+                report_error(ERROR, file, elem.attrib['linenr'], f'The lambda function passed to the function ddc::{p} must be a KOKKOS_LAMBDA\n{last_arg}\nShould be:\n{correct_last_arg}')
+
+            lambda_body_start = next(i for i,a in enumerate(file.data[arg_splitters[-1]+1:], arg_splitters[-1]+1) if a['str'] == '{')
+            lambda_body_end_id = file.data[lambda_body_start]['link']
+            lambda_body_scope = file.data[lambda_body_start]['scope']
+            end_body_idx = file.data.index(file.data_xml.find(f".token[@id='{lambda_body_end_id}'][@scope='{lambda_body_scope}']").attrib)
+            for a in file.data[lambda_body_start+1:end_body_idx]:
+                var_id = a.get('variable', None)
+                if var_id:
+                    var = next(v for v in file.variables if v['id'] == var_id)
+                    var_scope = var['scope']
+                    scope_info = file.root.find(f"./dump/scopes/scope[@id='{var_scope}']").attrib
+                    if scope_info['type'] == 'Class':
+                        report_error(ERROR, file, a['linenr'], f'Please create a local variable with the suffix "_proxy" to store the class variable that is used in a parallel loop ({a["str"]})')
+
 def check_licence_presence(file):
     """
     Test to ensure that all .hpp files have the license information in the first line.
@@ -454,5 +523,6 @@ def check_licence_presence(file):
         search_for_bad_memory(myfile)
         check_directives(myfile)
         check_licence_presence(myfile)
+        check_kokkos_lambda_use(myfile)
 
     sys.exit(error_level)