Merge commit 'fbdacbd54' into oct23av

This adds the following commits by Olivier (final part of PR madgraph5#764, I had already merged individually previous commits)
fbdacbd       Stefan Roiser   Fri Sep 15 09:36:36 2023 +0200  Merge pull request madgraph5#764 from madgraph5/new_interface
87afdc2       Olivier Mattelaer       Thu Sep 14 09:13:25 2023 +0200  change default gpu size to 16k
f27f44d       Olivier Mattelaer       Tue Sep 12 23:33:24 2023 +0200  adding shortcut for the SIMD/GPU mode
22fd349       Olivier Mattelaer       Tue Sep 12 16:53:39 2023 +0200  allowing run_card to not include the vector_size (then use the value defined at output time)
5b594c7       Olivier Mattelaer       Tue Sep 12 16:08:42 2023 +0200  fix compilation issue
0dd8542       oliviermattelaer        Tue Sep 12 15:29:00 2023 +0200  Merge branch 'master' into new_interface
c839c47       Olivier Mattelaer       Tue Sep 12 15:07:03 2023 +0200  update to version supporting vector_size to be setup via the run_card

Fix conflicts:
epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/output.py
epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/patchMad.sh

Author: valassi

MG5aMC/mg5amcnlo

@@ -25,16 +25,19 @@ index 617f10b93..dbe08b846 100644
  	$(RM) *.o $(LIBRARIES) $(BINARIES)
  	cd PDF; make clean; cd ..
  	cd PDF/gammaUPC; make clean; cd ../../
-@@ -132,4 +132,9 @@ clean:
+@@ -132,4 +132,11 @@ clean:
  	cd BIAS/ptj_bias; make clean; cd ../..
  	if [ -d $(CUTTOOLSDIR) ]; then cd $(CUTTOOLSDIR); make clean; cd ..; fi
  	if [ -d $(IREGIDIR) ]; then cd $(IREGIDIR); make clean; cd ..; fi
 +
 +clean: cleanSource
  	for i in `ls -d ../SubProcesses/P*`; do cd $$i; make clean; cd -; done;
 +
-+cleanall: cleanSource
++cleanavx:
 +	for i in `ls -d ../SubProcesses/P*`; do cd $$i; make cleanavxs; cd -; done;
++cleanall: cleanSource # THIS IS THE ONE
++	for i in `ls -d ../SubProcesses/P*`; do cd $$i; make cleanavxs; cd -; done;
++
 diff --git b/epochX/cudacpp/gg_tt.mad/SubProcesses/makefile a/epochX/cudacpp/gg_tt.mad/SubProcesses/makefile
 index 348c283be..74db44d84 100644
 --- b/epochX/cudacpp/gg_tt.mad/SubProcesses/makefile
@@ -283,44 +286,6 @@ index 348c283be..74db44d84 100644
 +
 +distclean: cleanall # Clean all fortran and cudacpp builds as well as the googletest installation
 +	$(MAKE) -f $(CUDACPP_MAKEFILE) distclean
-diff --git b/epochX/cudacpp/gg_tt.mad/Source/make_opts a/epochX/cudacpp/gg_tt.mad/Source/make_opts
-index 57f5f7bb9..bd3c24228 100644
---- b/epochX/cudacpp/gg_tt.mad/Source/make_opts
-+++ a/epochX/cudacpp/gg_tt.mad/Source/make_opts
-@@ -1,12 +1,17 @@
-+pdlabel1=
-+pdlabel2=
-+lhapdf=
-+PYTHIA8_PATH=NotInstalled
-+MG5AMC_VERSION=3.5.0_lo_vect
- GLOBAL_FLAG=-O3 -ffast-math -fbounds-check
-+ALOHA_FLAG=
-+MATRIX_FLAG=
- DEFAULT_CPP_COMPILER=g++
- MACFLAG=
- STDLIB=-lstdc++
- STDLIB_FLAG=
- DEFAULT_F_COMPILER=gfortran
- DEFAULT_F2PY_COMPILER=f2py3
--MG5AMC_VERSION=SpecifiedByMG5aMCAtRunTime
--PYTHIA8_PATH=NotInstalled
- #end_of_make_opts_variables
- 
- BIASLIBDIR=../../../lib/
-diff --git b/epochX/cudacpp/gg_tt.mad/bin/internal/banner.py a/epochX/cudacpp/gg_tt.mad/bin/internal/banner.py
-index 8f8df219d..7624b9f55 100755
---- b/epochX/cudacpp/gg_tt.mad/bin/internal/banner.py
-+++ a/epochX/cudacpp/gg_tt.mad/bin/internal/banner.py
-@@ -4187,7 +4187,8 @@ class RunCardLO(RunCard):
-         self.add_param('mxxmin4pdg',[-1.], system=True)
-         self.add_param('mxxpart_antipart', [False], system=True)
-                      
--        
-+        # CUDACPP parameters
-+        self.add_param('cudacpp_backend', 'CPP', include=False, hidden=False)
-              
-     def check_validity(self):
-         """ """
 diff --git b/epochX/cudacpp/gg_tt.mad/bin/internal/gen_ximprove.py a/epochX/cudacpp/gg_tt.mad/bin/internal/gen_ximprove.py
 index 4dd71db86..3b8ec3121 100755
 --- b/epochX/cudacpp/gg_tt.mad/bin/internal/gen_ximprove.py

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

@@ -32,7 +32,12 @@
     #    allows the command "output myformat PATH" in madgraph.
     #    MYCLASS should inherit from class madgraph.iolibs.export_v4.VirtualExporter
     import PLUGIN.CUDACPP_OUTPUT.output as output
-    new_output = { 'standalone_cudacpp' : output.PLUGIN_ProcessExporter }
+    new_output = { 'madevent_simd' : output.SIMD_ProcessExporter,
+                   'madevent_gpu' : output.GPU_ProcessExporter,
+                   'standalone_cudacpp' : output.PLUGIN_ProcessExporter,
+                   'standalone_simd' :  output.SIMD_ProcessExporter,
+                   'standalone_cuda' :  output.GPU_ProcessExporter,
+                  }
 
     # 2. Define new way to handle the cluster.
     #    Example: new_cluster = {'mycluster': MYCLUSTERCLASS}

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

@@ -1,6 +1,8 @@
 
 import logging
-
+import os
+import subprocess
+pjoin = os.path.join
 logger = logging.getLogger('cmdprint') # for stdout
 
 try:
@@ -9,20 +11,30 @@
     import internal.madevent_interface as madevent_interface
     import internal.misc as misc
     import internal.extended_cmd as extended_cmd
+    import internal.banner as banner_mod
 else:
     import madgraph.interface.madevent_interface as madevent_interface
     import madgraph.various.misc as misc
     import madgraph.interface.extended_cmd as extended_cmd
+    import madgraph.various.banner as banner_mod
 
 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'):
+            import pathlib
+            import os
+            pjoin = os.path.join
+            
+
+
+            
             cudacpp_backend = self.run_card['cudacpp_backend'].upper() # 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':
@@ -36,5 +48,53 @@ def compile(self, *args, **opts):
             return misc.compile(nb_core=self.options['nb_core'], *args, **opts)
         else:
             return misc.compile(nb_core=self.options['nb_core'], *args, **opts)
-    
+
+class CPPRunCard(banner_mod.RunCardLO):
+
+    def reset_simd(self, old_value, new_value, name):
+        if not hasattr(self, 'path'):
+            raise Exception
+
+        if name == "vector_size" and new_value <= int(old_value):
+            # code can handle the new size -> do not recompile
+            return
+        
+        Sourcedir = pjoin(os.path.dirname(os.path.dirname(self.path)), 'Source')
+        subprocess.call(['make', 'cleanavx'], cwd=Sourcedir, stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL)
+        
+    def plugin_input(self, finput):
+        return
+
+    def default_setup(self):
+        super().default_setup()
+        self.add_param('cudacpp_backend', 'CPP', include=False, hidden=False)
+        
+
+    def write_one_include_file(self, output_dir, incname, output_file=None):
+        """write one include file at the time"""
+
+        if incname == "vector.inc" and 'vector_size' not in self.user_set:
+            return
+        super().write_one_include_file(output_dir, incname, output_file)
+
+
+    def check_validity(self):
+        """ensure that PLUGIN information are consistent"""
+        
+        super().check_validity()
+
+        if self['SDE_strategy'] != 1:
+            logger.warning('SDE_strategy different of 1 is not supported with SMD/GPU mode')
+            self['sde_strategy'] = 1
+
+        if self['hel_recycling']:
+            self['hel_recycling'] = False
+            
+class GPURunCard(CPPRunCard):
+
+    def default_setup(self):
+        super(CPPRunCard, self).default_setup()
+        self.add_param('cudacpp_backend', 'CUDA', include=False, hidden=False)
+        
 MEINTERFACE = CPPMEInterface
+RunCard = CPPRunCard

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

@@ -203,30 +203,33 @@ def finalize(self, matrix_element, cmdhistory, MG5options, outputflag):
 	    cmdhistory is the list of command used so far.
 	    MG5options are all the options of the main interface
 	    outputflags is a list of options provided when doing the output command"""
-        misc.sprint('Entering PLUGIN_ProcessExporter.finalize', self.in_madevent_mode)
+        misc.sprint('Entering PLUGIN_ProcessExporter.finalize', self.in_madevent_mode, type(self))
         if self.in_madevent_mode:
             self.add_input_for_banner()
             if 'CUDACPP_CODEGEN_PATCHLEVEL' in os.environ: patchlevel = os.environ['CUDACPP_CODEGEN_PATCHLEVEL']
             else: patchlevel = ''
+<<<<<<< HEAD
             # OLDEST implementation (AV)
             #path = os.path.realpath(os.curdir + os.sep + 'PLUGIN' + os.sep + 'CUDACPP_OUTPUT')
             #misc.sprint(path)
             #if os.system(path + os.sep + 'patchMad.sh ' + self.dir_path + ' PROD ' + patchlevel) != 0:
             #    logger.debug("####### \n stdout is \n %s", stdout)
             #    logger.info("####### \n stderr is \n %s", stderr)
             #    raise Exception('ERROR! the O/S call to patchMad.sh failed')            
-            # OLD implementation (OM)
-            #plugin_path = os.path.dirname(os.path.realpath( __file__ ))
-            #p = subprocess.Popen([pjoin(plugin_path, 'patchMad.sh'), self.dir_path , 'PROD', str(patchlevel)])
-            #stdout, stderr = p.communicate()
-            #if not p.returncode:
+            # OLD implementation (SH PR #762)
+            #if os.system(PLUGINDIR + os.sep + 'patchMad.sh ' + self.dir_path + ' PROD ' + patchlevel) != 0:
             #    logger.debug("####### \n stdout is \n %s", stdout)
             #    logger.info("####### \n stderr is \n %s", stderr)
-            #    raise Exception('ERROR! the O/S call to patchMad.sh failed')            
-            # NEW implementation (SH)
-            if os.system(PLUGINDIR + os.sep + 'patchMad.sh ' + self.dir_path + ' PROD ' + patchlevel) != 0:
+            #    raise Exception('ERROR! the O/S call to patchMad.sh failed')
+            # NEW implementation (OM PR #764)
+            plugin_path = os.path.dirname(os.path.realpath( __file__ ))
+            p = subprocess.Popen([pjoin(plugin_path, 'patchMad.sh'), self.dir_path , 'PROD', str(patchlevel)],
+                                 stdout=subprocess.PIPE, stderr=subprocess.PIPE)
+            stdout, stderr = p.communicate()
+            if p.returncode != 0:
                 logger.debug("####### \n stdout is \n %s", stdout)
                 logger.info("####### \n stderr is \n %s", stderr)
+                logger.info("return code is %s\n", p.returncode)
                 raise Exception('ERROR! the O/S call to patchMad.sh failed')
             # Additional patching (OM)
             self.add_madevent_plugin_fct() # Added by OM
@@ -267,3 +270,36 @@ def add_madevent_plugin_fct(self):
         files.ln( pjoin(self.dir_path, 'lib'),  pjoin(self.dir_path, 'SubProcesses'))
 
 #------------------------------------------------------------------------------------
+class SIMD_ProcessExporter(PLUGIN_ProcessExporter):
+
+    def change_output_args(args, cmd):
+        """ """
+        cmd._export_format = "madevent"
+        args.append('--hel_recycling=False')
+        args.append('--me_exporter=standalone_simd')
+        if 'vector_size' not in ''.join(args):
+            args.append('--vector_size=16')
+        return args
+
+        
+    
+class GPU_ProcessExporter(PLUGIN_ProcessExporter):
+    
+    def change_output_args(args, cmd):
+        """ """
+        cmd._export_format = "madevent"
+        args.append('--hel_recycling=False')
+        args.append('--me_exporter=standalone_cuda')
+        if 'vector_size' not in ''.join(args):
+            args.append('--vector_size=16384')
+        return args
+        
+    def finalize(self, matrix_element, cmdhistory, MG5options, outputflag):
+
+        misc.sprint("enter dedicated function")
+        out = super().finalize(matrix_element, cmdhistory, MG5options, outputflag)
+        #change RunCard class to have default for GPU
+        text = open(pjoin(self.dir_path, 'bin', 'internal', 'launch_plugin.py'), 'r').read()
+        text = text.replace('RunCard = CPPRunCard', 'RunCard = GPURunCard')
+        open(pjoin(self.dir_path, 'bin', 'internal', 'launch_plugin.py'), 'w').write(text)
+        return out

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

@@ -81,7 +81,7 @@ fi
 # Patch the default Fortran code to provide the integration with the cudacpp plugin
 # (1) Process-independent patches
 touch ${dir}/Events/.keep # this file should already be present (mg5amcnlo copies it from Template/LO/Events/.keep) 
-cp -pr ${scrdir}/MG5aMC_patches/${dir_patches}/fbridge_common.inc ${dir}/SubProcesses # new file
+\cp -pr ${scrdir}/MG5aMC_patches/${dir_patches}/fbridge_common.inc ${dir}/SubProcesses # new file
 if [ "${tmadmode}" != "0" ]; then
   sed -i 's/2  = sde_strategy/1  = sde_strategy/' ${dir}/Cards/run_card.dat # use strategy SDE=1 in multichannel mode (see #419)
   sed -i 's/SDE_STRAT = 2/SDE_STRAT = 1/' ${dir}/Source/run_card.inc # use strategy SDE=1 in multichannel mode (see #419)