fixing issue of stefan with nprocesses>1

epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/launch_plugin.py

@@ -0,0 +1,38 @@
+
+import madgraph.various.misc as misc
+import madgraph.interface.extended_cmd as extended_cmd
+import logging
+
+logger = logging.getLogger('cmdprint') # for stdout
+
+try:
+    import madgraph
+except:
+    import internal.madevent_interface as madevent_interface
+else:
+    import madgraph.interface.madevent_interface as madevent_interface
+
+class CPPMEInterface(madevent_interface.MadEventCmdShell):
+
+    def compile(self, *args, **opts):
+        """ """
+        import multiprocessing
+        if not self.options['nb_core'] or self.options['nb_core'] == 'None':
+            self.options['nb_core'] = multiprocessing.cpu_count()
+
+        if args and args[0][0] == 'madevent' and hasattr(self, 'run_card'):
+            cudacpp_backend = self.run_card['cudacpp_backend'] # the default value is defined in banner.py
+            logger.info("Building madevent in madevent_interface.py with '%s' matrix elements"%cudacpp_backend)
+            if cudacpp_backend == 'FORTRAN':
+                args[0][0] = 'madevent_fortran_link'
+            elif cudacpp_backend == 'CPP':
+                args[0][0] = 'madevent_cpp_link'
+            elif cudacpp_backend == 'CUDA':
+                args[0][0] = 'madevent_cuda_link'
+            else:
+                raise Exception("Invalid cudacpp_backend='%s': only 'FORTRAN', 'CPP', 'CUDA' are supported")
+            return misc.compile(nb_core=self.options['nb_core'], *args, **opts)
+        else:
+            return misc.compile(nb_core=self.options['nb_core'], *args, **opts)
+
+MEINTERFACE = CPPMEInterface

epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/process_function_definitions.inc

@@ -405,8 +405,7 @@ namespace mg5amcCpu
     {
       // nprocesses>1 was last observed for "mirror processes" in uux_ttx in the 270 branch (see issue #343 and PRs #360 and #396)
       constexpr int nprocesses = %(nproc)i;
-      static_assert( nprocesses == 1, "Assume nprocesses == 1" );
-      // process_id corresponds to the index of DSIG1 Fortran functions (must be 1 because cudacpp is unable to handle DSIG2)
+      static_assert( nprocesses == 1 || nprocesses == 2, "Assume nprocesses == 1 or 2" );
       constexpr int process_id = %(proc_id)i; // code generation source: %(proc_id_source)s
       static_assert( process_id == 1, "Assume process_id == 1" );
     }

epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/model_handling.py

@@ -1146,6 +1146,13 @@ def get_sigmaKin_lines(self, color_amplitudes, write=True):
         replace_dict = super().get_sigmaKin_lines(color_amplitudes, write=False)
         replace_dict['proc_id'] = self.proc_id if self.proc_id>0 else 1
         replace_dict['proc_id_source'] = 'madevent + cudacpp exporter' if self.proc_id>0 else 'standalone_cudacpp' # FIXME? use self.in_madevent_mode instead?
+
+        # Extract denominator (avoid to extend size for mirroring)
+        den_factors = [str(me.get_denominator_factor()) for me in \
+                            self.matrix_elements]
+        replace_dict['den_factors'] = ",".join(den_factors)
+        misc.sprint(replace_dict['den_factors'])
+
         if write:
             file = self.read_template_file(self.process_sigmaKin_function_template) % replace_dict
             file = '\n'.join( file.split('\n')[8:] ) # skip first 8 lines in process_sigmaKin_function.inc (copyright)