uncommented debug statements but set default debug flags to false and…

… added some additional compile time debug flags

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

@@ -1,7 +1,7 @@
 # Copyright (C) 2020-2024 CERN and UCLouvain.
 # Licensed under the GNU Lesser General Public License (version 3 or later).
 # Created by: O. Mattelaer (Sep 2021) for the MG5aMC CUDACPP plugin.
-# Further modified by: O. Mattelaer, A. Valassi (2021-2024) for the MG5aMC CUDACPP plugin.
+# Further modified by: O. Mattelaer, A. Valassi, Z. Wettersten (2021-2024) for the MG5aMC CUDACPP plugin.
 
 # AV - Rename the plugin as CUDACPP_OUTPUT (even if the madgraph4gpu directory is still called CUDACPP_SA_OUTPUT)
 # This can be used in mg5amcnlo in one of two ways:

epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/Bridge.h

@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2024 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Roiser (Nov 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2024) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2021-2024) for the MG5aMC CUDACPP plugin.
 
 #ifndef BRIDGE_H
 #define BRIDGE_H 1
@@ -255,19 +255,22 @@ namespace mg5amcCpu
         throw std::logic_error( "Bridge constructor: FIXME! cannot choose gputhreads" ); // this should never happen!
       m_gpublocks = m_nevt / m_gputhreads;
     }
-    //std::cout << "WARNING! Instantiate device Bridge (nevt=" << m_nevt << ", gpublocks=" << m_gpublocks << ", gputhreads=" << m_gputhreads
-    //          << ", gpublocks*gputhreads=" << m_gpublocks * m_gputhreads << ")" << std::endl;
+#ifdef MGONGPU_VERBOSE_BRIDGE
+    std::cout << "WARNING! Instantiate device Bridge (nevt=" << m_nevt << ", gpublocks=" << m_gpublocks << ", gputhreads=" << m_gputhreads
+              << ", gpublocks*gputhreads=" << m_gpublocks * m_gputhreads << ")" << std::endl;
+#endif
     m_pmek.reset( new MatrixElementKernelDevice( m_devMomentaC, m_devGs, m_devRndHel, m_devRndCol, m_devChannelIds, m_devMEs, m_devSelHel, m_devSelCol, m_gpublocks, m_gputhreads ) );
 #else
-    //std::cout << "WARNING! Instantiate host Bridge (nevt=" << m_nevt << ")" << std::endl;
+#ifdef MGONGPU_VERBOSE_BRIDGE
+    std::cout << "WARNING! Instantiate host Bridge (nevt=" << m_nevt << ")" << std::endl;
+#endif
     m_pmek.reset( new MatrixElementKernelHost( m_hstMomentaC, m_hstGs, m_hstRndHel, m_hstRndCol, m_hstChannelIds, m_hstMEs, m_hstSelHel, m_hstSelCol, m_nevt ) );
 #endif // MGONGPUCPP_GPUIMPL
     // Create a process object, read param card and set parameters
     // FIXME: the process instance can happily go out of scope because it is only needed to read parameters?
     // FIXME: the CPPProcess should really be a singleton? what if fbridgecreate is called from several Fortran threads?
     CPPProcess process( /*verbose=*/false );
-    std::string paramCard = "../../Cards/param_card.dat";
-    std::string paramCardTrex = "../Cards/param_card.dat";
+    std::string paramCard = "../Cards/param_card.dat"; // ZW: change default param_card.dat location to one dir down
     /*
 #ifdef __HIPCC__
     if( !std::experimental::filesystem::exists( paramCard ) ) paramCard = "../" + paramCard;
@@ -279,8 +282,12 @@ namespace mg5amcCpu
     //if( !( stat( paramCard.c_str(), &dummyBuffer ) == 0 ) ) paramCard = "../" + paramCard; //
     auto fileExists = []( std::string& fileName )
     { struct stat buffer; return stat( fileName.c_str(), &buffer ) == 0; };
-    if( fileExists( paramCardTrex ) ) paramCard = paramCardTrex; // ZW: override param_card.dat to be one dir down since trex runs from the SubProcesses dir directory
-    if( !fileExists( paramCard ) ) paramCard = "../" + paramCard; // bypass std::filesystem #803
+    size_t paramCardCheck = 2; // ZW: check for paramCard up to 2 directories up
+    for( size_t k = 0 ; k < paramCardCheck ; ++k )
+    {
+      if( fileExists( paramCard ) ) break;  // bypass std::filesystem #803
+      paramCard = "../" + paramCard;
+    }
     process.initProc( paramCard );
   }
 
@@ -349,7 +356,9 @@ namespace mg5amcCpu
     if( goodHelOnly ) return;
     m_pmek->computeMatrixElements( useChannelIds );
     copyHostFromDevice( m_hstMEs, m_devMEs );
-    //flagAbnormalMEs( m_hstMEs.data(), m_nevt );
+#ifdef MGONGPU_VERBOSE_BRIDGE
+    flagAbnormalMEs( m_hstMEs.data(), m_nevt );
+#endif
     copyHostFromDevice( m_hstSelHel, m_devSelHel );
     copyHostFromDevice( m_hstSelCol, m_devSelCol );
     if constexpr( std::is_same_v<FORTRANFPTYPE, fptype> )
@@ -402,7 +411,9 @@ namespace mg5amcCpu
     }
     if( goodHelOnly ) return;
     m_pmek->computeMatrixElements( useChannelIds );
-    //flagAbnormalMEs( m_hstMEs.data(), m_nevt );
+#ifdef MGONGPU_VERBOSE_BRIDGE
+    flagAbnormalMEs( m_hstMEs.data(), m_nevt );
+#endif
     if constexpr( std::is_same_v<FORTRANFPTYPE, fptype> )
     {
       memcpy( mes, m_hstMEs.data(), m_hstMEs.bytes() );

epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/GpuRuntime.h

@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2024 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: J. Teig (Jun 2023, based on earlier work by S. Roiser) for the MG5aMC CUDACPP plugin.
-// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2024) for the MG5aMC CUDACPP plugin.
+// Further modified by: O. Mattelaer, S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2020-2024) for the MG5aMC CUDACPP plugin.
 
 #ifndef MG5AMC_GPURUNTIME_H
 #define MG5AMC_GPURUNTIME_H 1
@@ -38,7 +38,7 @@ namespace mg5amcGpu
   // *** FIXME! This will all need to be designed differently when going to multi-GPU nodes! ***
   struct GpuRuntime final
   {
-    GpuRuntime( const bool debug = false )
+    GpuRuntime( const bool debug = false ) // ZW: default debug to false
       : m_debug( debug ) { setUp( m_debug ); }
     ~GpuRuntime() { tearDown( m_debug ); }
     GpuRuntime( const GpuRuntime& ) = delete;
@@ -50,7 +50,7 @@ namespace mg5amcGpu
     // Set up CUDA application
     // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
     // Calling cudaSetDevice on startup is useful to properly book-keep the time spent in CUDA initialization
-    static void setUp( const bool debug = false )
+    static void setUp( const bool debug = false ) // ZW: default debug to false
     {
       // ** NB: it is useful to call cudaSetDevice, or cudaFree, to properly book-keep the time spent in CUDA initialization
       // ** NB: otherwise, the first CUDA operation (eg a cudaMemcpyToSymbol in CPPProcess ctor) appears to take much longer!
@@ -71,7 +71,7 @@ namespace mg5amcGpu
     // ** NB: strictly speaking this is not needed when using the CUDA runtime API **
     // Calling cudaDeviceReset on shutdown is only needed for checking memory leaks in cuda-memcheck
     // See https://docs.nvidia.com/cuda/cuda-memcheck/index.html#leak-checking
-    static void tearDown( const bool debug = false )
+    static void tearDown( const bool debug = false ) // ZW: default debug to false
     {
       if( debug ) std::cout << "__GpuRuntime: calling GpuDeviceReset()" << std::endl;
       checkGpu( gpuDeviceReset() );

epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MatrixElementKernels.cc

@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2024 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: A. Valassi (Jan 2022) for the MG5aMC CUDACPP plugin.
-// Further modified by: J. Teig, A. Valassi (2022-2024) for the MG5aMC CUDACPP plugin.
+// Further modified by: J. Teig, A. Valassi, Z. Wettersten (2022-2024) for the MG5aMC CUDACPP plugin.
 
 #include "MatrixElementKernels.h"
 
@@ -60,7 +60,9 @@ namespace mg5amcCpu
 #ifdef MGONGPU_CHANNELID_DEBUG
     MatrixElementKernelBase::dumpNevtProcessedByChannel();
 #endif
-   // MatrixElementKernelBase::dumpSignallingFPEs();
+#ifdef MGONGPU_VERBOSE_FPES
+    MatrixElementKernelBase::dumpSignallingFPEs();
+#endif
   }
 
   //--------------------------------------------------------------------------
@@ -130,7 +132,9 @@ namespace mg5amcCpu
     if( std::fetestexcept( FE_OVERFLOW ) ) fpes += " FE_OVERFLOW";
     if( std::fetestexcept( FE_UNDERFLOW ) ) fpes += " FE_UNDERFLOW";
     //if( std::fetestexcept( FE_INEXACT ) ) fpes += " FE_INEXACT"; // do not print this out: this would almost always signal!
-    if( fpes != "" )
+    if( fpes == "" )
+      std::cout << "INFO: No Floating Point Exceptions have been reported" << std::endl;
+    else
       std::cerr << "INFO: The following Floating Point Exceptions have been reported:" << fpes << std::endl;
   }
 
@@ -272,14 +276,15 @@ namespace mg5amcCpu
 #endif
     if( verbose )
     {
-      if( tag != "none" ){
-        //std::cout << "INFO: The application does not require the host to support any AVX feature" << std::endl;
-      if( ok && !known )
+      if( tag == "none" )
+        std::cout << "INFO: The application does not require the host to support any AVX feature" << std::endl;
+      else if( ok && known )
+        std::cout << "INFO: The application is built for " << tag << " and the host supports it" << std::endl;
+      else if( ok )
         std::cout << "WARNING: The application is built for " << tag << " but it is unknown if the host supports it" << std::endl;
-      else if ( !ok && known )
+      else
         std::cout << "ERROR! The application is built for " << tag << " but the host does not support it" << std::endl;
     }
-    }
     return ok;
   }
 

epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/MatrixElementKernels.h

@@ -134,7 +134,7 @@ namespace mg5amcCpu
 
     // Does this host system support the SIMD used in the matrix element calculation?
     // [NB: this is private, SIMD vectorization in mg5amc C++ code is currently only used in the ME calculations below MatrixElementKernelHost!]
-    static bool hostSupportsSIMD( const bool verbose = true );
+    static bool hostSupportsSIMD( const bool verbose = false ); // ZW: set verbose to false by default
 
   private:
 

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

@@ -1,7 +1,7 @@
 # Copyright (C) 2020-2024 CERN and UCLouvain.
 # Licensed under the GNU Lesser General Public License (version 3 or later).
 # Created by: A. Valassi (Mar 2024) for the MG5aMC CUDACPP plugin.
-# Further modified by: A. Valassi (2024) for the MG5aMC CUDACPP plugin.
+# Further modified by: A. Valassi, Z. Wettersten (2024) for the MG5aMC CUDACPP plugin.
 
 #-------------------------------------------------------------------------------
 
@@ -13,7 +13,8 @@ ifeq ($(BACKEND),)
   override BACKEND = gpucpp
 endif
 
-# Stop immediately if BACKEND=cuda but nvcc is missing
+# ZW: gpucpp backend checks if there is a GPU backend available before going to SIMD
+# prioritises CUDA over HIP
 ifeq ($(BACKEND),gpucpp)
   ifeq ($(shell which nvcc 2>/dev/null),)
     ifeq ($(shell which hipcc 2>/dev/null),)

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

@@ -1,7 +1,7 @@
 # Copyright (C) 2020-2024 CERN and UCLouvain.
 # Licensed under the GNU Lesser General Public License (version 3 or later).
 # Created by: S. Roiser (Feb 2020) for the MG5aMC CUDACPP plugin.
-# Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi (2020-2024) for the MG5aMC CUDACPP plugin.
+# Further modified by: S. Hageboeck, O. Mattelaer, S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2020-2024) for the MG5aMC CUDACPP plugin.
 
 #=== Determine the name of this makefile (https://ftp.gnu.org/old-gnu/Manuals/make-3.80/html_node/make_17.html)
 #=== NB: assume that the same name (e.g. cudacpp.mk, Makefile...) is used in the Subprocess and src directories

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

@@ -552,6 +552,7 @@ endif
 #-------------------------------------------------------------------------------
 
 # Target (and build rules): C++ rwgt libraries
+# ZW: the -Bsymbolic flag ensures that function calls will be handled internally by the library, rather than going to global context
 cxx_rwgtfiles := $(BUILDDIR)/rwgt_runner_cpp.o $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(BUILDDIR)/fbridge_cpp.o $(cxx_objects_lib) $(cxx_objects_exe) $(BUILDDIR)/CurandRandomNumberKernel_cpp.o $(BUILDDIR)/HiprandRandomNumberKernel_cpp.o
 $(cxx_rwgtlib): LIBFLAGS += $(CXXLIBFLAGSRPATH)
 $(cxx_rwgtlib): $(LIBDIR)/lib$(MG5AMC_COMMONLIB).so $(cxx_rwgtfiles) $(cxx_objects_lib)

epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/fbridge.cc

@@ -1,7 +1,7 @@
 // Copyright (C) 2020-2024 CERN and UCLouvain.
 // Licensed under the GNU Lesser General Public License (version 3 or later).
 // Created by: S. Roiser (Oct 2021) for the MG5aMC CUDACPP plugin.
-// Further modified by: S. Roiser, J. Teig, A. Valassi, Z. Wettersten (2021-2024) for the MG5aMC CUDACPP plugin.
+// Further modified by: S. Roiser, J. Teig, A. Valassi (2021-2024) for the MG5aMC CUDACPP plugin.
 
 #include "Bridge.h"
 #include "CPPProcess.h"

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

@@ -62,10 +62,11 @@ fpeEnable()
   //std::cout << "fpeEnable:     FE_INVALID is" << ( ( fpes & FE_INVALID ) ? " " : " NOT " ) << "enabled" << std::endl;
   //std::cout << "fpeEnable:     FE_OVERFLOW is" << ( ( fpes & FE_OVERFLOW ) ? " " : " NOT " ) << "enabled" << std::endl;
   //std::cout << "fpeEnable:     FE_UNDERFLOW is" << ( ( fpes & FE_UNDERFLOW ) ? " " : " NOT " ) << "enabled" << std::endl;
-  constexpr bool enableFPE = true; // this is hardcoded and no longer controlled by getenv( "CUDACPP_RUNTIME_ENABLEFPE" )
+  constexpr bool enableFPE = false; // this is hardcoded and no longer controlled by getenv( "CUDACPP_RUNTIME_ENABLEFPE" )
+  // ZW: hardcode enableFPE to false
   if( enableFPE )
   {
-    //std::cout << "INFO: The following Floating Point Exceptions will cause SIGFPE program aborts: FE_DIVBYZERO, FE_INVALID, FE_OVERFLOW" << std::endl;
+    std::cout << "INFO: The following Floating Point Exceptions will cause SIGFPE program aborts: FE_DIVBYZERO, FE_INVALID, FE_OVERFLOW" << std::endl;
     feenableexcept( FE_INVALID | FE_DIVBYZERO | FE_OVERFLOW ); // new strategy #831 (do not enable FE_UNDERFLOW)
     //fpes = fegetexcept();
     //std::cout << "fpeEnable: analyse fegetexcept()=" << fpes << std::endl;

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

@@ -117,7 +117,7 @@ namespace mg5amcCpu
 #else
     memcpy( cHel, tHel, ncomb * npar * sizeof( short ) );
 #endif
-    //fpeEnable(); // enable SIGFPE traps for Floating Point Exceptions
+    fpeEnable(); // enable SIGFPE traps for Floating Point Exceptions
   }
 
   //--------------------------------------------------------------------------

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

@@ -1,7 +1,7 @@
 # Copyright (C) 2020-2024 CERN and UCLouvain.
 # Licensed under the GNU Lesser General Public License (version 3 or later).
 # Created by: O. Mattelaer (Sep 2021) for the MG5aMC CUDACPP plugin.
-# Further modified by: O. Mattelaer, J. Teig, A. Valassi (2021-2024) for the MG5aMC CUDACPP plugin.
+# Further modified by: O. Mattelaer, J. Teig, A. Valassi, Z. Wettersten (2021-2024) for the MG5aMC CUDACPP plugin.
 
 import os
 import sys

tools/REX/rwgt_runner.cc

@@ -11,8 +11,8 @@
 // A class for reweighting matrix elements for
 %(process_lines)s
 //--------------------------------------------------------------------------
-#ifndef _LIBCOMP_
-#define _LIBCOMP_
+#ifndef _TREX_
+#define _TREX_
 #endif
 #include "rwgt_instance.h"
 #include "fbridge.h"