If openmp is used wrap the likwid markers calls in omp blocks

pystencils_benchmark/benchmark.py

@@ -6,7 +6,7 @@ from jinja2 import Environment, PackageLoader, StrictUndefined
 import numpy as np
 
 from pystencils.backends.cbackend import generate_c, get_headers
-from pystencils.astnodes import KernelFunction
+from pystencils.astnodes import KernelFunction, PragmaBlock
 from pystencils.enums import Backend
 from pystencils.typing import get_base_type
 from pystencils.sympyextensions import prod
@@ -98,7 +98,7 @@ def kernel_main(kernels_ast: List[KernelFunction], *,
     Returns:
         C code as string
     """
-    Kernel = namedtuple('Kernel', ['name', 'constants', 'fields', 'call_parameters', 'call_argument_list'])
+    Kernel = namedtuple('Kernel', ['name', 'constants', 'fields', 'call_parameters', 'call_argument_list', 'openmp'])
     kernels = []
     includes = set()
     for kernel in kernels_ast:
@@ -107,6 +107,8 @@ def kernel_main(kernels_ast: List[KernelFunction], *,
         constants = []
         fields = []
         call_parameters = []
+        # TODO: Think about it maybe there is a better way to detect openmp
+        openmp = isinstance(kernel.body.args[0], PragmaBlock)
         for p in kernel.get_parameters():
             if not p.is_field_parameter:
                 constants.append((p.symbol.name, str(p.symbol.dtype)))
@@ -139,8 +141,10 @@ def kernel_main(kernels_ast: List[KernelFunction], *,
                     align = 0
                 fields.append((p.field_name, dtype, elements, size, offset, align))
                 call_parameters.append(p.field_name)
+
+        # TODO: Think about openmp detection again
         kernels.append(Kernel(name=name, fields=fields, constants=constants, call_parameters=call_parameters,
-                              call_argument_list=",".join(call_parameters)))
+                              call_argument_list=",".join(call_parameters), openmp=openmp))
 
         includes.add(name)
 

pystencils_benchmark/templates/main.c

@@ -39,6 +39,9 @@ int main(int argc, char **argv)
         {% else %}
         {{dataType}}* {{field_name}} = ({{dataType}} *) malloc({{size}});
         {% endif %}
+        {% if kernel.openmp %}
+        #pragma omp parallel for schedule(static)
+        {% endif %}
         for (unsigned long long i = 0; i < {{elements}}; ++i)
             {{field_name}}[i] = 0.23;
         {% endfor %}
@@ -50,7 +53,14 @@ int main(int argc, char **argv)
         {% endfor %}
 
         {% if likwid %}
+        {% if kernel.openmp %}
+        #pragma omp parallel
+        {
+        {% endif %}
         LIKWID_MARKER_REGISTER("{{kernel.name}}");
+        {% if kernel.openmp %}
+        }
+        {% endif %}
         {% endif %}
 
         for(int warmup = 1; warmup >= 0; --warmup) {
@@ -58,7 +68,14 @@ int main(int argc, char **argv)
             if(warmup == 0) {
                 repeat = n_repeat;
                 {% if likwid %}
+                {% if kernel.openmp %}
+                #pragma omp parallel
+                {
+                {% endif %}
                 LIKWID_MARKER_START("{{kernel.name}}");
+                {% if kernel.openmp %}
+                }
+                {% endif %}
                 {% endif %}
             }
 
@@ -80,13 +97,20 @@ int main(int argc, char **argv)
             {% endif %}
         }
 
-        {% for field_name, dataType, elements, size, offset, alignment in kernel.fields %}
-        free({{field_name}});
-        {% endfor %}
-
         {% if likwid %}
+        {% if kernel.openmp %}
+        #pragma omp parallel
+        {
+        {% endif %}
         LIKWID_MARKER_STOP("{{kernel.name}}");
+        {% if kernel.openmp %}
+        }
+        {% endif %}
         {% endif %}
+
+        {% for field_name, dataType, elements, size, offset, alignment in kernel.fields %}
+        free({{field_name}});
+        {% endfor %}
     }
     {% endfor %}