for the CPU version correct the index access, GPU version still needs to

be done

Author: roiser

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

@@ -26,7 +26,7 @@
       fptype_sv& MEs_sv = E_ACCESS::kernelAccess( MEs );
       MEs_sv = fptype_sv{ 0 };
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-      fptype* numerators = NUM_ACCESS::ieventAccessRecord( allNumerators, ievt0 );
+      fptype* numerators = NUM_ACCESS::ieventAccessRecord( allNumerators, ievt0 * %(ndiagrams)d );
       fptype* denominators = DEN_ACCESS::ieventAccessRecord( allDenominators, ievt0 );
       fptype_sv* numerators_sv = NUM_ACCESS::kernelAccessP( numerators );
       fptype_sv& denominators_sv = DEN_ACCESS::kernelAccess( denominators );
@@ -333,7 +333,7 @@
       if( allChannelIds != nullptr ) // fix segfault #892 (not 'channelIds[0] != 0')
       {
         const unsigned int channelId = getChannelId( allChannelIds, ievt0, false );
-        fptype* numerators = NUM_ACCESS::ieventAccessRecord( allNumerators, ievt0 );
+        fptype* numerators = NUM_ACCESS::ieventAccessRecord( allNumerators, ievt0 * %(ndiagrams)d );
         fptype* denominators = DEN_ACCESS::ieventAccessRecord( allDenominators, ievt0 );
         fptype_sv* numerators_sv = NUM_ACCESS::kernelAccessP( numerators );
         fptype_sv& denominators_sv = DEN_ACCESS::kernelAccess( denominators );

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

@@ -1882,6 +1882,7 @@ def super_get_matrix_element_calls(self, matrix_element, color_amplitudes, multi
         ###misc.sprint(multi_channel_map)
         res = []
         ###res.append('for(int i=0;i<%s;i++){jamp[i] = cxtype(0.,0.);}' % len(color_amplitudes))
+        ndiagrams = len(me)
         res.append("""//constexpr size_t nxcoup = ndcoup + nicoup; // both dependent and independent couplings (BUG #823)
       constexpr size_t nxcoup = ndcoup + nIPC; // both dependent and independent couplings (FIX #823)
       const fptype* allCOUPs[nxcoup];
@@ -1918,7 +1919,7 @@ def super_get_matrix_element_calls(self, matrix_element, color_amplitudes, multi
         COUPs[ndcoup + iicoup] = allCOUPs[ndcoup + iicoup]; // independent couplings, fixed for all events
       fptype* MEs = E_ACCESS::ieventAccessRecord( allMEs, ievt0 );
 #ifdef MGONGPU_SUPPORTS_MULTICHANNEL
-      fptype* numerators = NUM_ACCESS::ieventAccessRecord( allNumerators, ievt0 );
+      fptype* numerators = NUM_ACCESS::ieventAccessRecord( allNumerators, ievt0 * %d );
       fptype* denominators = DEN_ACCESS::ieventAccessRecord( allDenominators, ievt0 );
 #endif
 #endif
@@ -1930,7 +1931,7 @@ def super_get_matrix_element_calls(self, matrix_element, color_amplitudes, multi
       // Numerators and denominators for the current event (CUDA) or SIMD event page (C++)
       fptype_sv* numerators_sv = NUM_ACCESS::kernelAccessP( numerators );
       fptype_sv& denominators_sv = DEN_ACCESS::kernelAccess( denominators );
-#endif""")
+#endif""" % ndiagrams)
         diagrams = matrix_element.get('diagrams')
         diag_to_config = {}
         if multi_channel_map: