Alpha s part2

.gitignore

@@ -26,8 +26,10 @@ test/googletest
 *.local
 *~
 local/
+**/build/
 *.log
 *.output
 *.error
 *.cobaltlog
 *.BKP
+.DS_Store

epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/CMake/Compilers.txt

@@ -0,0 +1,2 @@
+set(CMAKE_CXX_STANDARD 17)
+set(CMAKE_CXX_STANDARD_REQUIRED True)

epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/CMake/Macros.txt

@@ -0,0 +1,10 @@
+MACRO(SUBDIRLIST result)
+  FILE(GLOB children RELATIVE ${CMAKE_CURRENT_SOURCE_DIR} ${CMAKE_CURRENT_SOURCE_DIR}/*)
+  SET(dirlist "")
+  FOREACH(child ${children})
+    IF(IS_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}/${child})
+      LIST(APPEND dirlist ${child})
+    ENDIF()
+  ENDFOREACH()
+  SET(${result} ${dirlist})
+ENDMACRO()

epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/CMake/Platforms.txt

@@ -0,0 +1,3 @@
+if (CMAKE_HOST_APPLE)
+  add_definitions(-DMGONGPU_HAS_NO_CURAND)
+endif(CMAKE_HOST_APPLE)

epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/CMakeLists.txt

@@ -0,0 +1,14 @@
+# Minimal CMake configuration to build a functional CPU version
+
+cmake_minimum_required(VERSION 3.22)
+
+project(Madgraph4GPU)
+
+include(${PROJECT_SOURCE_DIR}/CMake/Platforms.txt)
+include(${PROJECT_SOURCE_DIR}/CMake/Compilers.txt)
+include(${PROJECT_SOURCE_DIR}/CMake/Macros.txt)
+
+set(PROJECT_GITROOT_DIR ${PROJECT_SOURCE_DIR}/../../..)
+
+add_subdirectory(src)
+add_subdirectory(SubProcesses)

epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/SubProcesses/CMakeLists.txt

@@ -0,0 +1,4 @@
+SUBDIRLIST(SUBDIRS)
+FOREACH(subdir ${SUBDIRS})
+  ADD_SUBDIRECTORY(${subdir})
+ENDFOREACH()

epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/CMakeLists_P.txt

@@ -0,0 +1,15 @@
+get_filename_component(basename ${CMAKE_CURRENT_SOURCE_DIR} NAME)
+string(TOLOWER ${basename} targadd)
+
+file(GLOB_RECURSE HEADERS "*.h")
+
+set(libname mg5amc_cxx_${targadd})
+add_library(${libname} BridgeKernels.cc CPPProcess.cc CrossSectionKernels.cc MatrixElementKernels.cc
+                       RamboSamplingKernels.cc RandomNumberKernels.cc ${HEADERS})
+target_include_directories(${libname} PUBLIC "${PROJECT_SOURCE_DIR}/src"
+                                                        "${PROJECT_GITROOT_DIR}/tools")
+
+set(execname check_${targadd}.exe)
+add_executable(${execname} check_sa.cc)
+target_link_libraries(${execname} PUBLIC mg5amc_common ${libname})
+target_include_directories(${execname} PRIVATE "${PROJECT_SOURCE_DIR}/src")

epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/src/CMakeLists.txt

@@ -0,0 +1,2 @@
+file(GLOB_RECURSE HEADERS "*.h")
+add_library(mg5amc_common Parameters_sm.cc read_slha.cc ${HEADERS})

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

@@ -937,7 +937,7 @@ def get_process_function_definitions(self, write=True):
         if len(self.params2order) == 0: # remove all IPD occurrences (issue #349)
             file_lines = file.split('\n')
             file_lines = [l.replace('cIPC,cIPD','cIPC') for l in file_lines] # remove cIPD from OpenMP pragma
-            file_lines = [l for l in file_lines if 'IPD' not in l] # remove all other lines matchin IPD           
+            file_lines = [l for l in file_lines if 'IPD' not in l] # remove all other lines matchin IPD
             file = '\n'.join( file_lines )
         return file
 
@@ -1054,6 +1054,7 @@ def generate_process_files(self):
             self.matrix_elements[0].set('has_mirror_process', False)
             self.nprocesses/=2
         super(export_cpp.OneProcessExporterGPU, self).generate_process_files()
+        self.edit_CMakeLists()
         self.edit_check_sa()
         self.edit_mgonGPU()
         self.edit_processidfile() # AV new file (NB this is Sigma-specific, should not be a symlink to Subprocesses)
@@ -1067,6 +1068,16 @@ def generate_process_files(self):
         files.ln(pjoin(self.path, 'RandomNumberKernels.cc'), self.path, 'gRandomNumberKernels.cu')
         files.ln(pjoin(self.path, 'BridgeKernels.cc'), self.path, 'gBridgeKernels.cu')
 
+
+    # sr - generate CMakeLists.txt file inside the P* directory
+    def edit_CMakeLists(self):
+        """Generate CMakeLists.txt"""
+        misc.sprint('Entering PLUGIN_OneProcessExporter.edit_CMakeLists')
+        template = open(pjoin(self.template_path,'gpu','CMakeLists_P.txt'),'r').read()
+        ff = open(pjoin(self.path, 'CMakeLists.txt'),'w')
+        ff.write(template)
+        ff.close()
+
     # 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"""

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

@@ -25,12 +25,12 @@ def PLUGIN_make_unique(input, keepordering=None):
             printordering = False
             misc.sprint('keepordering (default): %s'%keepordering) # AV - add a printout only in the first call
     else:
-        misc.sprint('keepordering (argument): %s'%keepordering) # AV - add a printout at every call only if it is an argument	
+        misc.sprint('keepordering (argument): %s'%keepordering) # AV - add a printout at every call only if it is an argument
     ###sprint(keepordering) # AV - remove the printout at every call
     if not keepordering:
         return list(set(input))
     else:
-        return list(dict.fromkeys(input)) 
+        return list(dict.fromkeys(input))
 
 DEFAULT_make_unique = misc.make_unique
 misc.make_unique = PLUGIN_make_unique
@@ -65,12 +65,12 @@ class PLUGIN_ProcessExporter(export_cpp.ProcessExporterGPU):
     #      Note: only change class attribute
     #  - PLUGIN_ProcessExporter(ProcessExporterGPU)
     #      This class
-    
+
     # Below are the class variable that are defined in export_v4.VirtualExporter
     # AV - keep defaults from export_v4.VirtualExporter
     # Check status of the directory. Remove it if already exists
-    ###check = True 
-    # Output type: [Template/dir/None] copy the Template (via copy_template), just create dir or do nothing 
+    ###check = True
+    # Output type: [Template/dir/None] copy the Template (via copy_template), just create dir or do nothing
     ###output = 'Template'
 
     # If sa_symmetry is true, generate fewer matrix elements
@@ -80,10 +80,10 @@ class PLUGIN_ProcessExporter(export_cpp.ProcessExporterGPU):
     # Below are the class variable that are defined in export_cpp.ProcessExporterGPU
     # AV - keep defaults from export_cpp.ProcessExporterGPU
     # Decide which type of merging is used [madevent/madweight]
-    ###grouped_mode = False 
+    ###grouped_mode = False
     # Other options
     ###default_opt = {'clean': False, 'complex_mass':False, 'export_format':'madevent', 'mp': False, 'v5_model': True }
-    
+
     # AV - keep defaults from export_cpp.ProcessExporterGPU
     # AV - used in MadGraphCmd.do_output to assign export_cpp.ExportCPPFactory to MadGraphCmd._curr_exporter (if cpp or gpu)
     # AV - used in MadGraphCmd.export to assign helas_call_writers.(CPPUFO|GPUFO)HelasCallWriter to MadGraphCmd._curr_helas_model (if cpp or gpu)
@@ -93,16 +93,18 @@ class PLUGIN_ProcessExporter(export_cpp.ProcessExporterGPU):
     # AV - use a custom OneProcessExporter
     ###oneprocessclass = export_cpp.OneProcessExporterGPU # responsible for P directory
     oneprocessclass = model_handling.PLUGIN_OneProcessExporter
-    
+
     # Information to find the template file that we want to include from madgraph
     # you can include additional file from the plugin directory as well
     # AV - use template files from PLUGINDIR instead of MG5DIR and add gpu/mgOnGpuVectors.h
     # [NB: mgOnGpuConfig.h, check_sa.cc and fcheck_sa.f are handled through dedicated methods]
     ###s = MG5DIR + '/madgraph/iolibs/template_files/'
     s = PLUGINDIR + '/madgraph/iolibs/template_files/'
-    from_template = {'.': [s+'.clang-format'],
+    from_template = {'.': [s+'.clang-format', s+'CMakeLists.txt'],
+                     'CMake': [s+'CMake/Compilers.txt', s+'CMake/Platforms.txt', s+'CMake/Macros.txt'],
                      'src': [s+'gpu/rambo.h', s+'read_slha.h', s+'read_slha.cc',
-                             s+'gpu/mgOnGpuFptypes.h', s+'gpu/mgOnGpuCxtypes.h', s+'gpu/mgOnGpuVectors.h'],
+                             s+'gpu/mgOnGpuFptypes.h', s+'gpu/mgOnGpuCxtypes.h', s+'gpu/mgOnGpuVectors.h',
+                             s+'src/CMakeLists.txt'],
                      'SubProcesses': [s+'gpu/nvtx.h', s+'gpu/timer.h', s+'gpu/timermap.h', s+'gpu/CudaRuntime.h',
                                       s+'gpu/MemoryBuffers.h', s+'gpu/MemoryAccessHelpers.h', s+'gpu/MemoryAccessVectors.h',
                                       s+'gpu/MemoryAccessMatrixElements.h', s+'gpu/MemoryAccessMomenta.h',
@@ -120,11 +122,13 @@ class PLUGIN_ProcessExporter(export_cpp.ProcessExporterGPU):
                                       s+'gpu/MadgraphTest.h', s+'gpu/runTest.cc',
                                       s+'gpu/testmisc.cc', s+'gpu/testxxx_cc_ref.txt',
                                       s+'gpu/perf.py', s+'gpu/profile.sh']}
+                                      s+'gpu/testmisc.cc', s+'gpu/testxxx_cc_ref.txt',
+                                      s+'gpu/perf.py', s+'gpu/profile.sh', s+'SubProcesses/CMakeLists.txt']}
     to_link_in_P = ['nvtx.h', 'timer.h', 'timermap.h', 'CudaRuntime.h',
                     'MemoryBuffers.h', 'MemoryAccessHelpers.h', 'MemoryAccessVectors.h',
                     'MemoryAccessMatrixElements.h', 'MemoryAccessMomenta.h',
-                    'MemoryAccessRandomNumbers.h', 'MemoryAccessWeights.h', 
-		    'MemoryAccessAmplitudes.h', 'MemoryAccessWavefunctions.h',
+                    'MemoryAccessRandomNumbers.h', 'MemoryAccessWeights.h',
+	                'MemoryAccessAmplitudes.h', 'MemoryAccessWavefunctions.h',
                     'EventStatistics.h',
                     'CrossSectionKernels.cc', 'CrossSectionKernels.h',
                     'MatrixElementKernels.cc', 'MatrixElementKernels.h',
@@ -146,9 +150,9 @@ class PLUGIN_ProcessExporter(export_cpp.ProcessExporterGPU):
 
     # AV - use a custom UFOModelConverter (model/aloha exporter)
     ###create_model_class =  export_cpp.UFOModelConverterGPU
-    import PLUGIN.CUDACPP_SA_OUTPUT.model_handling as model_handling 
+    import PLUGIN.CUDACPP_SA_OUTPUT.model_handling as model_handling
     create_model_class = model_handling.PLUGIN_UFOModelConverter
-    
+
     # AV - "aloha_exporter" is not used anywhere!
     # (OM: "typically not defined but useful for this tutorial - the class for writing helas routine")
     ###aloha_exporter = None

epochX/cudacpp/gg_tt/CMake/Compilers.txt

@@ -0,0 +1,2 @@
+set(CMAKE_CXX_STANDARD 17)
+set(CMAKE_CXX_STANDARD_REQUIRED True)

epochX/cudacpp/gg_tt/CMake/Macros.txt

@@ -0,0 +1,10 @@
+MACRO(SUBDIRLIST result)
+  FILE(GLOB children RELATIVE ${CMAKE_CURRENT_SOURCE_DIR} ${CMAKE_CURRENT_SOURCE_DIR}/*)
+  SET(dirlist "")
+  FOREACH(child ${children})
+    IF(IS_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}/${child})
+      LIST(APPEND dirlist ${child})
+    ENDIF()
+  ENDFOREACH()
+  SET(${result} ${dirlist})
+ENDMACRO()

epochX/cudacpp/gg_tt/CMake/Platforms.txt

@@ -0,0 +1,3 @@
+if (CMAKE_HOST_APPLE)
+  add_definitions(-DMGONGPU_HAS_NO_CURAND)
+endif(CMAKE_HOST_APPLE)

epochX/cudacpp/gg_tt/CMakeLists.txt

@@ -0,0 +1,14 @@
+# Minimal CMake configuration to build a functional CPU version
+
+cmake_minimum_required(VERSION 3.22)
+
+project(Madgraph4GPU)
+
+include(${PROJECT_SOURCE_DIR}/CMake/Platforms.txt)
+include(${PROJECT_SOURCE_DIR}/CMake/Compilers.txt)
+include(${PROJECT_SOURCE_DIR}/CMake/Macros.txt)
+
+set(PROJECT_GITROOT_DIR ${PROJECT_SOURCE_DIR}/../../..)
+
+add_subdirectory(src)
+add_subdirectory(SubProcesses)

epochX/cudacpp/gg_tt/SubProcesses/Bridge.h

@@ -98,14 +98,16 @@ namespace mg5amcCpu
      */
     void gpu_sequence( const FORTRANFPTYPE* momenta, FORTRANFPTYPE* mes, const bool goodHelOnly = false );
 
+    void cpu_depCouplings( const FORTRANFPTYPE* gs );
+
     /**
      * Sequence to be executed for the vectorized CPU matrix element calculation
      *
      * @param momenta the pointer to the input 4-momenta
      * @param mes the pointer to the output matrix elements
      * @param goodHelOnly quit after computing good helicities?
      */
-    void cpu_sequence( const FORTRANFPTYPE* momenta, FORTRANFPTYPE* mes, const bool goodHelOnly = false );
+    void cpu_sequence( const FORTRANFPTYPE* momenta, const FORTRANFPTYPE* gs, FORTRANFPTYPE* mes, const bool goodHelOnly = false );
 
   private:
     int m_nevt;                // number of events
@@ -122,6 +124,7 @@ namespace mg5amcCpu
     static constexpr int s_gputhreadsmin = 32; // minimum number of gpu threads
 #else
     mg5amcCpu::HostBufferMomenta m_hstMomentaC;
+    mg5amcCpu::HostBufferGs m_hstGsC;
     mg5amcCpu::HostBufferMatrixElements m_hstMEsC;
     std::unique_ptr<mg5amcCpu::MatrixElementKernelHost> m_pmek;
 #endif
@@ -163,6 +166,7 @@ namespace mg5amcCpu
     , m_hstMEsC( m_nevt )
 #else
     , m_hstMomentaC( m_nevt )
+    , m_hstGsC( m_nevt )
     , m_hstMEsC( m_nevt )
 #endif
     , m_pmek( nullptr )
@@ -186,7 +190,7 @@ namespace mg5amcCpu
 #else
     std::cout << "WARNING! Instantiate host Bridge (nevt=" << m_nevt << ")" << std::endl;
     mg5amcCpu::CPPProcess process( /*verbose=*/false );
-    m_pmek.reset( new mg5amcCpu::MatrixElementKernelHost( m_hstMomentaC, m_hstMEsC, m_nevt ) );
+    m_pmek.reset( new mg5amcCpu::MatrixElementKernelHost( m_hstMomentaC, m_hstGsC, m_hstMEsC, m_nevt ) );
 #endif // __CUDACC__
     process.initProc( "../../Cards/param_card.dat" );
   }
@@ -235,8 +239,16 @@ namespace mg5amcCpu
 #endif
 
 #ifndef __CUDACC__
+
+  template<typename FORTRANFPTYPE>
+  void Bridge<FORTRANFPTYPE>::cpu_depCouplings( const FORTRANFPTYPE* gs )
+  {
+    memcpy( m_hstGsC.data(), gs, m_nevt * sizeof( FORTRANFPTYPE ) );
+    m_pmek->computeDependentCouplings();
+  }
+
   template<typename FORTRANFPTYPE>
-  void Bridge<FORTRANFPTYPE>::cpu_sequence( const FORTRANFPTYPE* momenta, FORTRANFPTYPE* mes, const bool goodHelOnly )
+  void Bridge<FORTRANFPTYPE>::cpu_sequence( const FORTRANFPTYPE* momenta, const FORTRANFPTYPE* gs, FORTRANFPTYPE* mes, const bool goodHelOnly )
   {
     hst_transposeMomentaF2C( momenta, m_hstMomentaC.data(), m_nevt );
     if( !m_goodHelsCalculated )

epochX/cudacpp/gg_tt/SubProcesses/BridgeKernels.cc

@@ -18,9 +18,10 @@ namespace mg5amcCpu
   //--------------------------------------------------------------------------
 
   BridgeKernelBase::BridgeKernelBase( const BufferMomenta& momenta,         // input: momenta
+                                      const BufferGs& gs,                   // input: gs for alphaS
                                       BufferMatrixElements& matrixElements, // output: matrix elements
                                       const size_t nevt )
-    : MatrixElementKernelBase( momenta, matrixElements )
+    : MatrixElementKernelBase( momenta, gs, matrixElements )
     , NumberOfEvents( nevt )
     , m_bridge( nevt, npar, np4 )
   {
@@ -42,9 +43,10 @@ namespace mg5amcCpu
   //--------------------------------------------------------------------------
 
   BridgeKernelHost::BridgeKernelHost( const BufferMomenta& momenta,         // input: momenta
+                                      const BufferGs& gs,                   // input: Gs for alphaS
                                       BufferMatrixElements& matrixElements, // output: matrix elements
                                       const size_t nevt )
-    : BridgeKernelBase( momenta, matrixElements, nevt )
+    : BridgeKernelBase( momenta, gs, matrixElements, nevt )
     , m_fortranMomenta( nevt )
   {
   }
@@ -58,18 +60,25 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
+  void BridgeKernelHost::computeDependentCouplings()
+  {
+    m_bridge.cpu_depCouplings( m_gs.data() );
+  }
+
+  //--------------------------------------------------------------------------
+
   void BridgeKernelHost::computeGoodHelicities()
   {
     constexpr bool goodHelOnly = true;
-    m_bridge.cpu_sequence( m_fortranMomenta.data(), m_matrixElements.data(), goodHelOnly );
+    m_bridge.cpu_sequence( m_fortranMomenta.data(), m_gs.data(), m_matrixElements.data(), goodHelOnly );
   }
 
   //--------------------------------------------------------------------------
 
   void BridgeKernelHost::computeMatrixElements()
   {
     constexpr bool goodHelOnly = false;
-    m_bridge.cpu_sequence( m_fortranMomenta.data(), m_matrixElements.data(), goodHelOnly );
+    m_bridge.cpu_sequence( m_fortranMomenta.data(), m_gs.data(), m_matrixElements.data(), goodHelOnly );
   }
 
   //--------------------------------------------------------------------------

epochX/cudacpp/gg_tt/SubProcesses/BridgeKernels.h

@@ -22,6 +22,7 @@ namespace mg5amcCpu
 
     // Constructor from existing input and output buffers
     BridgeKernelBase( const BufferMomenta& momenta,         // input: momenta
+                      const BufferGs& gs,                   // input: gs for alphaS
                       BufferMatrixElements& matrixElements, // output: matrix elements
                       const size_t nevt );
 
@@ -47,12 +48,16 @@ namespace mg5amcCpu
 
     // Constructor from existing input and output buffers
     BridgeKernelHost( const BufferMomenta& momenta,         // input: momenta
+                      const BufferGs& gs,                   // input: gs for alphaS
                       BufferMatrixElements& matrixElements, // output: matrix elements
                       const size_t nevt );
 
     // Destructor
     virtual ~BridgeKernelHost() {}
 
+    // Calculate dependent couplings from Gs transferred via the Bridge
+    void computeDependentCouplings() override final;
+
     // Transpose input momenta from C to Fortran before the matrix element calculation in the Bridge
     void transposeInputMomentaC2F() override final;
 

epochX/cudacpp/gg_tt/SubProcesses/CMakeLists.txt

@@ -0,0 +1,4 @@
+SUBDIRLIST(SUBDIRS)
+FOREACH(subdir ${SUBDIRS})
+  ADD_SUBDIRECTORY(${subdir})
+ENDFOREACH()

epochX/cudacpp/gg_tt/SubProcesses/MatrixElementKernels.cc

@@ -16,9 +16,10 @@ namespace mg5amcCpu
   //--------------------------------------------------------------------------
 
   MatrixElementKernelHost::MatrixElementKernelHost( const BufferMomenta& momenta,         // input: momenta
+                                                    const BufferGs& gs,                   // input: gs for alphaS
                                                     BufferMatrixElements& matrixElements, // output: matrix elements
                                                     const size_t nevt )
-    : MatrixElementKernelBase( momenta, matrixElements )
+    : MatrixElementKernelBase( momenta, gs, matrixElements )
     , NumberOfEvents( nevt )
   {
     if( m_momenta.isOnDevice() ) throw std::runtime_error( "MatrixElementKernelHost: momenta must be a host array" );
@@ -43,6 +44,13 @@ namespace mg5amcCpu
 
   //--------------------------------------------------------------------------
 
+  void MatrixElementKernelHost::computeDependentCouplings()
+  {
+    dependentCouplings( m_gs.data(), m_gs.size() );
+  }
+
+  //--------------------------------------------------------------------------
+
   void MatrixElementKernelHost::computeGoodHelicities()
   {
     using mgOnGpu::ncomb; // the number of helicity combinations