New interface

epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/MG5aMC_patches/PROD/patch.common

@@ -78,12 +78,12 @@ index 348c283be..74db44d84 100644
 +CUDACPP_MAKEENV:=$(shell echo '$(.VARIABLES)' | tr " " "\n" | egrep "(USEBUILDDIR|AVX|FPTYPE|HELINL|HRDCOD)")
 +###$(info CUDACPP_MAKEENV=$(CUDACPP_MAKEENV))
 +###$(info $(foreach v,$(CUDACPP_MAKEENV),$(v)="$($(v))"))
-+CUDACPP_BUILDDIR:=$(shell $(MAKE) $(foreach v,$(CUDACPP_MAKEENV),$(v)="$($(v))") -f $(CUDACPP_MAKEFILE) -pn |& awk '/Building/{print $$3}' | sed s/BUILDDIR=//)
-+ifeq ($(CUDACPP_BUILDDIR),)
-+$(error CUDACPP_BUILDDIR='$(CUDACPP_BUILDDIR)' should not be empty!)
-+else
++CUDACPP_BUILDDIR:=$(shell $(MAKE) $(foreach v,$(CUDACPP_MAKEENV),$(v)="$($(v))") -f $(CUDACPP_MAKEFILE) -pn 2>/dev/null | awk '/Building/{print $$3}' | sed s/BUILDDIR=//)
++#ifeq ($(CUDACPP_BUILDDIR),)
++#$(error CUDACPP_BUILDDIR='$(CUDACPP_BUILDDIR)' should not be empty!)
++#else
 +$(info CUDACPP_BUILDDIR='$(CUDACPP_BUILDDIR)')
-+endif
++#endif
 +CUDACPP_COMMONLIB=mg5amc_common
 +CUDACPP_CXXLIB=mg5amc_$(processid_short)_cpp
 +CUDACPP_CULIB=mg5amc_$(processid_short)_cuda
@@ -110,7 +110,7 @@ index 348c283be..74db44d84 100644
 
 -$(PROG): $(PROCESS) auto_dsig.o $(LIBS) $(MATRIX)
 -	$(FC) -o $(PROG) $(PROCESS) $(MATRIX) $(LINKLIBS) $(LDFLAGS) $(BIASDEPENDENCIES) -fopenmp
-+LDFLAGS+=-Wl,--no-relax # avoid 'failed to convert GOTPCREL relocation' error #458
++#LDFLAGS+=-Wl,--no-relax # avoid 'failed to convert GOTPCREL relocation' error #458 (flag not universal -> skip?)
 
 -$(PROG)_forhel: $(PROCESS) auto_dsig.o $(LIBS) $(MATRIX_HEL)
 -	$(FC) -o $(PROG)_forhel $(PROCESS) $(MATRIX_HEL) $(LINKLIBS) $(LDFLAGS) $(BIASDEPENDENCIES) -fopenmp

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/MatrixElementKernels.cc

@@ -112,10 +112,17 @@ namespace mg5amcCpu
     // See https://community.arm.com/arm-community-blogs/b/operating-systems-blog/posts/runtime-detection-of-cpu-features-on-an-armv8-a-cpu
     bool ok = true; // this is just an assumption!
     const std::string tag = "arm neon (128bit as in SSE4.2)";
-#else
+#elif  defined(__x86_64__) || defined(__i386__)
     bool known = true;
     bool ok = __builtin_cpu_supports( "sse4.2" );
     const std::string tag = "nehalem (SSE4.2)";
+#else
+     bool known = false; // __builtin_cpu_supports is not supported
+    // See https://gcc.gnu.org/onlinedocs/gcc/Basic-PowerPC-Built-in-Functions-Available-on-all-Configurations.html
+    // See https://stackoverflow.com/q/62783908
+    // See https://community.arm.com/arm-community-blogs/b/operating-systems-blog/posts/runtime-detection-of-cpu-features-on-an-armv8-a-cpu
+    bool ok = true; // this is just an assumption!
+    const std::string tag = "arm neon (128bit as in SSE4.2)";
 #endif
 #else
     bool known = true;

epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/cudacpp.mk

@@ -27,6 +27,7 @@ UNAME_S := $(shell uname -s)
 UNAME_P := $(shell uname -p)
 ###$(info UNAME_P='$(UNAME_P)')
 
+include ../../Source/make_opts
 #-------------------------------------------------------------------------------
 
 #=== Configure common compiler flags for C++ and CUDA
@@ -220,7 +221,7 @@ else ifneq ($(shell $(CXX) --version | egrep '^(clang)'),)
 override OMPFLAGS = -fopenmp
 ###override OMPFLAGS = # disable OpenMP MT on clang (was not ok without or with nvcc before #578)
 else ifneq ($(shell $(CXX) --version | egrep '^(Apple clang)'),)
-override OMPFLAGS = # disable OpenMP MT on Apple clang (builds fail in the CI #578)
+override OMPFLAGS = -fopenmp # disable OpenMP MT on Apple clang (builds fail in the CI #578)
 else
 override OMPFLAGS = -fopenmp
 ###override OMPFLAGS = # disable OpenMP MT (default before #575)
@@ -554,7 +555,7 @@ endif
 $(LIBDIR)/lib$(MG5AMC_CXXLIB).so: $(BUILDDIR)/fbridge.o
 $(LIBDIR)/lib$(MG5AMC_CXXLIB).so: cxx_objects_lib += $(BUILDDIR)/fbridge.o
 $(LIBDIR)/lib$(MG5AMC_CXXLIB).so: $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_objects_lib)
-	$(CXX) -shared -o $@ $(cxx_objects_lib) $(CXXLIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB)
+	$(CXX) -shared -o $@ $(cxx_objects_lib) $(CXXLIBFLAGSRPATH2) -L$(LIBDIR) -l$(MG5AMC_COMMONLIB) $(LIBFLAGS) -fopenmp
 
 ifneq ($(NVCC),)
 $(LIBDIR)/lib$(MG5AMC_CULIB).so: $(BUILDDIR)/fbridge_cu.o

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

@@ -726,7 +726,6 @@ def write_hardcoded_parameters(self, params):
             for par in pars.split(','):
                 ###print(len(pardef_lines), par) # for debugging
                 pardef_lines[par] = ( 'constexpr ' + type + ' ' + par )
-        misc.sprint( 'pardef_lines size =', len(pardef_lines), ', keys size =', len(pardef_lines.keys()) )
         ###print( pardef_lines ) # for debugging
         ###for line in pardef_lines: misc.sprint(line) # for debugging
         parset_pars = []
@@ -1036,14 +1035,14 @@ class PLUGIN_OneProcessExporter(PLUGIN_export_cpp.OneProcessExporterGPU):
 
     # AV - overload export_cpp.OneProcessExporterGPU constructor (rename gCPPProcess to CPPProcess, set include_multi_channel)
     def __init__(self, *args, **kwargs):
-        misc.sprint('Entering PLUGIN_OneProcessExporter.__init__')
+#        misc.sprint('Entering PLUGIN_OneProcessExporter.__init__')
         for kwarg in kwargs: misc.sprint( 'kwargs[%s] = %s' %( kwarg, kwargs[kwarg] ) )
         super().__init__(*args, **kwargs)
         self.process_class = 'CPPProcess'
         ###if self.in_madevent_mode: proc_id = kwargs['prefix']+1 # madevent+cudacpp (NB: HERE SELF.IN_MADEVENT_MODE DOES NOT WORK!)
         if 'prefix' in kwargs: proc_id = kwargs['prefix']+1 # madevent+cudacpp (ime+1 from ProcessExporterFortranMEGroup.generate_subprocess_directory)
         else: proc_id = 0 # standalone_cudacpp
-        misc.sprint(proc_id)
+#        misc.sprint(proc_id)
         self.proc_id = proc_id
 
     # AV - overload export_cpp.OneProcessExporterGPU method (indent comments in process_lines)
@@ -1141,12 +1140,19 @@ def get_process_function_definitions(self, write=True):
 
     # AV - modify export_cpp.OneProcessExporterGPU method (add debug printouts for multichannel #342)
     def get_sigmaKin_lines(self, color_amplitudes, write=True):
-        misc.sprint('Entering PLUGIN_OneProcessExporter.get_sigmaKin_lines')
-        misc.sprint(self.include_multi_channel)
-        misc.sprint(self.support_multichannel)
+#        misc.sprint('Entering PLUGIN_OneProcessExporter.get_sigmaKin_lines')
+#        misc.sprint(self.include_multi_channel)
+#        misc.sprint(self.support_multichannel)
         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)
@@ -1160,14 +1166,14 @@ def get_all_sigmaKin_lines(self, color_amplitudes, class_name):
         ret_lines = []
         if self.single_helicities:
             ###assert self.include_multi_channel # remove this assert: must handle both cases and produce two different code bases (#473)
-            misc.sprint(type(self.helas_call_writer))
-            misc.sprint(self.support_multichannel, self.include_multi_channel)
+#            misc.sprint(type(self.helas_call_writer))
+#            misc.sprint(self.support_multichannel, self.include_multi_channel)
             multi_channel = None
             if self.include_multi_channel:
                 if not self.support_multichannel:
                     raise Exception("link with madevent not supported")
                 multi_channel = self.get_multi_channel_dictionary(self.matrix_elements[0].get('diagrams'), self.include_multi_channel)
-                misc.sprint(multi_channel)
+#                misc.sprint(multi_channel)
             ###misc.sprint( 'before get_matrix_element_calls', self.matrix_elements[0].get_number_of_wavefunctions() ) # WRONG value of nwf, eg 7 for gg_tt
             helas_calls = self.helas_call_writer.get_matrix_element_calls(\
                                                     self.matrix_elements[0],
@@ -1295,14 +1301,12 @@ def get_process_info_lines(self, matrix_element):
     # AV - replace the export_cpp.OneProcessExporterGPU method (invert .cc/.cu, add debug printouts)
     def generate_process_files(self):
         """Generate mgOnGpuConfig.h, CPPProcess.cc, CPPProcess.h, check_sa.cc, gXXX.cu links"""
-        misc.sprint('Entering PLUGIN_OneProcessExporter.generate_process_files')
-        if self.include_multi_channel:
-            misc.sprint('self.include_multi_channel is already defined: this is madevent+second_exporter mode') # FIXME? use self.in_madevent_mode instead?
-        else:
-            misc.sprint('self.include_multi_channel is not yet defined: this is standalone_cudacpp mode') # see issue #473
-        if self.matrix_elements[0].get('has_mirror_process'):
-            self.matrix_elements[0].set('has_mirror_process', False)
-            self.nprocesses/=2
+#        misc.sprint('Entering PLUGIN_OneProcessExporter.generate_process_files')
+        if not self.include_multi_channel:
+            #this condition is likely wrong and need to be removed
+            if self.matrix_elements[0].get('has_mirror_process'):
+                self.matrix_elements[0].set('has_mirror_process', False)
+                self.nprocesses/=2
         super(PLUGIN_export_cpp.OneProcessExporterGPU, self).generate_process_files()
         self.edit_CMakeLists()
         self.edit_check_sa()
@@ -1333,15 +1337,15 @@ def generate_process_files(self):
         ###template_ref = 'dump_CPUTest.'+self.process_name+'.txt'
         template_ref = self.template_path + '/../../../test/ref/' + 'dump_CPUTest.' + self.process_name + '.txt'
         if os.path.exists( template_ref ):
-            misc.sprint( 'Copying test reference file: ', template_ref )
+#            misc.sprint( 'Copying test reference file: ', template_ref )
             PLUGIN_export_cpp.cp( template_ref, self.path + '/../../test/ref' )
-        else:
-            misc.sprint( 'Test reference file does not exist and will not be copied: ', template_ref )
+#        else:
+#            misc.sprint( 'Test reference file does not exist and will not be copied: ', template_ref )
 
     # SR - generate CMakeLists.txt file inside the P* directory
     def edit_CMakeLists(self):
         """Generate CMakeLists.txt"""
-        misc.sprint('Entering PLUGIN_OneProcessExporter.edit_CMakeLists')
+#        misc.sprint('Entering PLUGIN_OneProcessExporter.edit_CMakeLists')
         template = open(pjoin(self.template_path,'CMake/SubProcesses/CMakeLists_P.txt'),'r').read()
         ff = open(pjoin(self.path, 'CMakeLists.txt'),'w')
         ff.write(template)
@@ -1350,7 +1354,7 @@ def edit_CMakeLists(self):
     # AV - replace the export_cpp.OneProcessExporterGPU method (invert .cc/.cu, add debug printouts)
     def edit_check_sa(self):
         """Generate check_sa.cc and fcheck_sa.f"""
-        misc.sprint('Entering PLUGIN_OneProcessExporter.edit_check_sa')
+#        misc.sprint('Entering PLUGIN_OneProcessExporter.edit_check_sa')
         ff = open(pjoin(self.path, 'check_sa.cc'),'w')
         template = open(pjoin(self.template_path,'gpu','check_sa.cc'),'r').read()
         ff.write(template) # nothing to replace in check_sa.cc
@@ -1367,7 +1371,7 @@ def edit_check_sa(self):
     # AV - replace the export_cpp.OneProcessExporterGPU method (add debug printouts and multichannel handling #473) 
     def edit_mgonGPU(self):
         """Generate mgOnGpuConfig.h"""
-        misc.sprint('Entering PLUGIN_OneProcessExporter.edit_mgonGPU')
+#        misc.sprint('Entering PLUGIN_OneProcessExporter.edit_mgonGPU')
         template = open(pjoin(self.template_path,'gpu','mgOnGpuConfig.h'),'r').read()
         replace_dict = {}
         nexternal, nincoming = self.matrix_elements[0].get_nexternal_ninitial()
@@ -1387,7 +1391,7 @@ def edit_mgonGPU(self):
     # AV - new method
     def edit_processidfile(self):
         """Generate epoch_process_id.h"""
-        misc.sprint('Entering PLUGIN_OneProcessExporter.edit_processidfile')
+#        misc.sprint('Entering PLUGIN_OneProcessExporter.edit_processidfile')
         template = open(pjoin(self.template_path,'gpu','epoch_process_id.h'),'r').read()
         replace_dict = {}
         replace_dict['processid'] = self.get_process_name()
@@ -1399,7 +1403,7 @@ def edit_processidfile(self):
     # AV - new method
     def edit_coloramps(self):
         """Generate coloramps.h"""
-        misc.sprint('Entering PLUGIN_OneProcessExporter.edit_coloramps')
+#        misc.sprint('Entering PLUGIN_OneProcessExporter.edit_coloramps')
         template = open(pjoin(self.template_path,'gpu','coloramps.h'),'r').read()
         ff = open(pjoin(self.path, 'coloramps.h'),'w')
         # The following five lines from OneProcessExporterCPP.get_sigmaKin_lines (using OneProcessExporterCPP.get_icolamp_lines)
@@ -1417,7 +1421,7 @@ def edit_coloramps(self):
     # AV - new method
     def edit_testxxx(self):
         """Generate testxxx.cc"""
-        misc.sprint('Entering PLUGIN_OneProcessExporter.edit_testxxx')
+#        misc.sprint('Entering PLUGIN_OneProcessExporter.edit_testxxx')
         template = open(pjoin(self.template_path,'gpu','testxxx.cc'),'r').read()
         replace_dict = {}
         replace_dict['model_name'] = self.model_name
@@ -1428,7 +1432,7 @@ def edit_testxxx(self):
     # AV - new method
     def edit_memorybuffers(self):
         """Generate MemoryBuffers.h"""
-        misc.sprint('Entering PLUGIN_OneProcessExporter.edit_memorybuffers')
+#        misc.sprint('Entering PLUGIN_OneProcessExporter.edit_memorybuffers')
         template = open(pjoin(self.template_path,'gpu','MemoryBuffers.h'),'r').read()
         replace_dict = {}
         replace_dict['model_name'] = self.model_name
@@ -1439,7 +1443,7 @@ def edit_memorybuffers(self):
     # AV - new method
     def edit_memoryaccesscouplings(self):
         """Generate MemoryAccessCouplings.h"""
-        misc.sprint('Entering PLUGIN_OneProcessExporter.edit_memoryaccesscouplings')
+#        misc.sprint('Entering PLUGIN_OneProcessExporter.edit_memoryaccesscouplings')
         template = open(pjoin(self.template_path,'gpu','MemoryAccessCouplings.h'),'r').read()
         replace_dict = {}
         replace_dict['model_name'] = self.model_name
@@ -1451,7 +1455,7 @@ def edit_memoryaccesscouplings(self):
     # [*NB export_cpp.UFOModelConverterGPU.write_process_h_file is not called!*]
     def write_process_h_file(self, writer):
         """Generate final gCPPProcess.h"""
-        misc.sprint('Entering PLUGIN_OneProcessExporter.write_process_h_file')
+#        misc.sprint('Entering PLUGIN_OneProcessExporter.write_process_h_file')
         out = super().write_process_h_file(writer)
         writer.seek(-1, os.SEEK_CUR)
         writer.truncate()
@@ -1473,7 +1477,7 @@ def super_write_process_cc_file(self, writer):
     # AV - overload the export_cpp.OneProcessExporterGPU method (add debug printout and truncate last \n)
     def write_process_cc_file(self, writer):
         """Generate CPPProcess.cc"""
-        misc.sprint('Entering PLUGIN_OneProcessExporter.write_process_cc_file')
+#        misc.sprint('Entering PLUGIN_OneProcessExporter.write_process_cc_file')
         ###out = super().write_process_cc_file(writer)
         out = self.super_write_process_cc_file(writer)
         writer.seek(-1, os.SEEK_CUR)
@@ -1652,7 +1656,7 @@ def super_get_matrix_element_calls(self, matrix_element, color_amplitudes, multi
                 color[namp][njamp] = coeff
         me = matrix_element.get('diagrams')
         matrix_element.reuse_outdated_wavefunctions(me)
-        misc.sprint(multi_channel_map)
+#        misc.sprint(multi_channel_map)
         res = []
         ###res.append('for(int i=0;i<%s;i++){jamp[i] = cxtype(0.,0.);}' % len(color_amplitudes))
         res.append("""constexpr size_t nxcoup = ndcoup + nicoup; // both dependent and independent couplings
@@ -1707,7 +1711,7 @@ def super_get_matrix_element_calls(self, matrix_element, color_amplitudes, multi
                                   sum([diagrams[idiag].get('amplitudes') for \
                                        idiag in multi_channel_map[config]], [])]
                 diag_to_config[amp[0]] = config
-        misc.sprint(diag_to_config)
+#        misc.sprint(diag_to_config)
         id_amp = 0
         for diagram in matrix_element.get('diagrams'):
             ###print('DIAGRAM %3d: #wavefunctions=%3d, #diagrams=%3d' %
@@ -1819,13 +1823,13 @@ def get_external_line(self, wf, argument):
                                  wf.get('me_id')-1,
                                  wf.get('number_external')-1)
             elif argument.is_boson():
-                misc.sprint(call)
-                misc.sprint( (wf.get('mass'),
-                                 wf.get('number_external')-1,
-                                 # For boson, need initial/final here
-                                 (-1) ** (wf.get('state') == 'initial'),
-                                 wf.get('me_id')-1,
-                                 wf.get('number_external')-1))
+#                misc.sprint(call)
+#                misc.sprint( (wf.get('mass'),
+#                                 wf.get('number_external')-1,
+#                                 # For boson, need initial/final here
+#                                 (-1) ** (wf.get('state') == 'initial'),
+#                                 wf.get('me_id')-1,
+#                                 wf.get('number_external')-1))
                 return  self.format_coupling(call % \
                                 (wf.get('mass'),
                                  wf.get('number_external')-1,