Merge remote-tracking branch 'upstream/master' (including PR madgraph…

…5#368 adding _cu.o and removing gXXX.cu links) into rocrand

Fix conflicts:
	epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/cudacpp.mk
	epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/model_handling.py

The conflicts were in the interplay of adding Hiprand and adding _cu.o (so for instance gHipRandRandomNumberKernel.o becomes HipRandRandomNumberKernel_cu.o)

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

@@ -602,7 +602,7 @@ $(BUILDDIR)/.build.$(TAG):
 	@if [ "$(oldtagsb)" != "" ]; then echo "Cannot build for tag=$(TAG) as old builds exist for other tags:"; echo "  $(oldtagsb)"; echo "Please run 'make clean' first\nIf 'make clean' is not enough: run 'make clean USEBUILDDIR=1 AVX=$(AVX) FPTYPE=$(FPTYPE)' or 'make cleanall'"; exit 1; fi
 	@touch $(BUILDDIR)/.build.$(TAG)
 
-# Generic target and build rules: objects from CUDA compilation
+# Generic target and build rules: objects from CUDA or HIP compilation
 # NB: CCBUILDRULEFLAGS includes "-x cu" for nvcc and "-x hip" for hipcc (#810)
 ifneq ($(GPUCC),)
 $(BUILDDIR)/%%.o : %%.cu *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
@@ -620,7 +620,7 @@ $(BUILDDIR)/%%.o : %%.cc *.h ../../src/*.h $(BUILDDIR)/.build.$(TAG)
 	@if [ ! -d $(BUILDDIR) ]; then echo "mkdir -p $(BUILDDIR)"; mkdir -p $(BUILDDIR); fi
 	$(CXX) $(CPPFLAGS) $(CXXFLAGS) -fPIC -c $< -o $@
 
-# Apply special build flags only to CrossSectionKernel.cc and gCrossSectionKernel.cu (no fast math, see #117 and #516)
+# Apply special build flags only to CrossSectionKernel[_cu].o (no fast math, see #117 and #516)
 # Added edgecase for HIP compilation
 ifeq ($(shell $(CXX) --version | grep ^nvc++),)
 $(BUILDDIR)/CrossSectionKernels.o: CXXFLAGS := $(filter-out -ffast-math,$(CXXFLAGS))
@@ -632,17 +632,17 @@ else
 endif
 endif
 
-# Apply special build flags only to check_sa.o and gcheck_sa.o (NVTX in timermap.h, #679)
+# Apply special build flags only to check_sa[_cu].o (NVTX in timermap.h, #679)
 $(BUILDDIR)/check_sa.o: CXXFLAGS += $(USE_NVTX) $(CUINC)
-$(BUILDDIR)/gcheck_sa.o: CXXFLAGS += $(USE_NVTX) $(CUINC)
+$(BUILDDIR)/check_sa_cu.o: CXXFLAGS += $(USE_NVTX) $(CUINC)
 
-# Apply special build flags only to check_sa and (Cu|Roc)randRandomNumberKernel
+# Apply special build flags only to check_sa[_cu].o and (Cu|Hip)randRandomNumberKernel[_cu].o
 $(BUILDDIR)/check_sa.o: CXXFLAGS += $(RNDCXXFLAGS)
-$(BUILDDIR)/gcheck_sa.o: CUFLAGS += $(RNDCXXFLAGS)
+$(BUILDDIR)/check_sa_cu.o: CUFLAGS += $(RNDCXXFLAGS)
 $(BUILDDIR)/CurandRandomNumberKernel.o: CXXFLAGS += $(RNDCXXFLAGS)
-$(BUILDDIR)/gCurandRandomNumberKernel.o: CUFLAGS += $(RNDCXXFLAGS)
+$(BUILDDIR)/CurandRandomNumberKernel_cu.o: CUFLAGS += $(RNDCXXFLAGS)
 $(BUILDDIR)/HiprandRandomNumberKernel.o: CXXFLAGS += $(RNDCXXFLAGS)
-$(BUILDDIR)/gHiprandRandomNumberKernel.o: CUFLAGS += $(RNDCXXFLAGS)
+$(BUILDDIR)/HiprandRandomNumberKernel_cu.o: CUFLAGS += $(RNDCXXFLAGS)
 ifeq ($(HASCURAND),hasCurand) # curand headers, #679
 $(BUILDDIR)/CurandRandomNumberKernel.o: CXXFLAGS += $(CUINC)
 endif
@@ -666,10 +666,10 @@ endif
 ###endif
 ###endif
 
-#### Apply special build flags only to CPPProcess.cc (-flto)
+#### Apply special build flags only to CPPProcess.o (-flto)
 ###$(BUILDDIR)/CPPProcess.o: CXXFLAGS += -flto
 
-#### Apply special build flags only to CPPProcess.cc (AVXFLAGS)
+#### Apply special build flags only to CPPProcess.o (AVXFLAGS)
 ###$(BUILDDIR)/CPPProcess.o: CXXFLAGS += $(AVXFLAGS)
 
 #-------------------------------------------------------------------------------
@@ -691,8 +691,8 @@ cxx_objects_exe=$(BUILDDIR)/CommonRandomNumberKernel.o $(BUILDDIR)/RamboSampling
 
 ifneq ($(GPUCC),)
 MG5AMC_CULIB = mg5amc_$(processid_short)_cuda
-cu_objects_lib=$(BUILDDIR)/gCPPProcess.o $(BUILDDIR)/gMatrixElementKernels.o $(BUILDDIR)/gBridgeKernels.o $(BUILDDIR)/gCrossSectionKernels.o
-cu_objects_exe=$(BUILDDIR)/gCommonRandomNumberKernel.o $(BUILDDIR)/gRamboSamplingKernels.o
+cu_objects_lib=$(BUILDDIR)/CPPProcess_cu.o $(BUILDDIR)/MatrixElementKernels_cu.o $(BUILDDIR)/BridgeKernels_cu.o $(BUILDDIR)/CrossSectionKernels_cu.o
+cu_objects_exe=$(BUILDDIR)/CommonRandomNumberKernel_cu.o $(BUILDDIR)/RamboSamplingKernels_cu.o
 endif
 
 # Target (and build rules): C++ and CUDA shared libraries

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

@@ -1030,8 +1030,7 @@ class PLUGIN_OneProcessExporter(PLUGIN_export_cpp.OneProcessExporterGPU):
     # AV - change defaults from export_cpp.OneProcessExporterGPU
     # [NB process_class = "CPPProcess" is set in OneProcessExporterCPP.__init__]
     # [NB process_class = "gCPPProcess" is set in OneProcessExporterGPU.__init__]
-    ###cc_ext = 'cu' # create gCPPProcess.cu (and symlink it as CPPProcess.cc)
-    cc_ext = 'cc' # create CPPProcess.cc (and symlink it as gCPPProcess.cu)
+    cc_ext = 'cc' # create CPPProcess.cc (build it also as CPPProcess_cu.so, no longer symlink it as gCPPProcess.cu)
 
     # AV - keep defaults from export_cpp.OneProcessExporterGPU
     ###process_dir = '.'
@@ -1079,7 +1078,7 @@ def get_process_class_definitions(self, write=True):
         file = '\n'.join( file.split('\n')[8:] ) # skip first 8 lines in process_class.inc (copyright)
         return file
 
-    # AV - replace export_cpp.OneProcessExporterGPU method (fix gCPPProcess.cu)
+    # AV - replace export_cpp.OneProcessExporterGPU method (fix CPPProcess.cc)
     def get_process_function_definitions(self, write=True):
         """The complete class definition for the process"""
         replace_dict = super(PLUGIN_export_cpp.OneProcessExporterGPU,self).get_process_function_definitions(write=False) # defines replace_dict['initProc_lines']
@@ -1178,9 +1177,9 @@ def get_sigmaKin_lines(self, color_amplitudes, write=True):
         else:
             return replace_dict
 
-    # AV - modify export_cpp.OneProcessExporterGPU method (fix gCPPProcess.cu)
+    # AV - modify export_cpp.OneProcessExporterGPU method (fix CPPProcess.cc)
     def get_all_sigmaKin_lines(self, color_amplitudes, class_name):
-        """Get sigmaKin_process for all subprocesses for gCPPProcess.cu"""
+        """Get sigmaKin_process for all subprocesses for CPPProcess.cc"""
         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)
@@ -1339,16 +1338,6 @@ def generate_process_files(self):
         self.edit_testxxx() # AV new file (NB this is generic in Subprocesses and then linked in Sigma-specific)
         self.edit_memorybuffers() # AV new file (NB this is generic in Subprocesses and then linked in Sigma-specific)
         self.edit_memoryaccesscouplings() # AV new file (NB this is generic in Subprocesses and then linked in Sigma-specific)
-        # Add symbolic links in the P1 directory
-        files.ln(pjoin(self.path, 'check_sa.cc'), self.path, 'gcheck_sa.cu')
-        files.ln(pjoin(self.path, 'CPPProcess.cc'), self.path, 'gCPPProcess.cu')
-        files.ln(pjoin(self.path, 'CrossSectionKernels.cc'), self.path, 'gCrossSectionKernels.cu')
-        files.ln(pjoin(self.path, 'MatrixElementKernels.cc'), self.path, 'gMatrixElementKernels.cu')
-        files.ln(pjoin(self.path, 'RamboSamplingKernels.cc'), self.path, 'gRamboSamplingKernels.cu')
-        files.ln(pjoin(self.path, 'CommonRandomNumberKernel.cc'), self.path, 'gCommonRandomNumberKernel.cu')
-        files.ln(pjoin(self.path, 'CurandRandomNumberKernel.cc'), self.path, 'gCurandRandomNumberKernel.cu')
-        files.ln(pjoin(self.path, 'HiprandRandomNumberKernel.cc'), self.path, 'gHiprandRandomNumberKernel.cu')
-        files.ln(pjoin(self.path, 'BridgeKernels.cc'), self.path, 'gBridgeKernels.cu')
         # NB: symlink of cudacpp.mk to makefile is overwritten by madevent makefile if this exists (#480)
         # NB: this relies on the assumption that cudacpp code is generated before madevent code
         files.ln(pjoin(self.path, 'cudacpp.mk'), self.path, 'makefile')
@@ -1477,7 +1466,7 @@ def edit_memoryaccesscouplings(self):
     # AV - overload the export_cpp.OneProcessExporterGPU method (add debug printout and truncate last \n)
     # [*NB export_cpp.UFOModelConverterGPU.write_process_h_file is not called!*]
     def write_process_h_file(self, writer):
-        """Generate final gCPPProcess.h"""
+        """Generate final CPPProcess.h"""
         ###misc.sprint('Entering PLUGIN_OneProcessExporter.write_process_h_file')
         out = super().write_process_h_file(writer)
         writer.seek(-1, os.SEEK_CUR)
@@ -1561,7 +1550,7 @@ def get_color_matrix_lines(self, matrix_element):
 
     # AV - replace the export_cpp.OneProcessExporterGPU method (improve formatting)
     def get_initProc_lines(self, matrix_element, color_amplitudes):
-        """Get initProc_lines for function definition for gCPPProcess::initProc"""
+        """Get initProc_lines for function definition for CPPProcess::initProc"""
         initProc_lines = []
         initProc_lines.append('// Set external particle masses for this matrix element')
         for part in matrix_element.get_external_wavefunctions():
@@ -1607,7 +1596,7 @@ class PLUGIN_GPUFOHelasCallWriter(helas_call_writers.GPUFOHelasCallWriter):
     #  - PLUGIN_GPUFOHelasCallWriter(GPUFOHelasCallWriter)
     #      This class
 
-    # AV - replace helas_call_writers.GPUFOHelasCallWriter method (improve formatting of gCPPProcess.cu)
+    # AV - replace helas_call_writers.GPUFOHelasCallWriter method (improve formatting of CPPProcess.cc)
     # [GPUFOHelasCallWriter.format_coupling is called by GPUFOHelasCallWriter.get_external_line/generate_helas_call]
     # [GPUFOHelasCallWriter.get_external_line is called by GPUFOHelasCallWriter.get_external]
     # [GPUFOHelasCallWriter.get_external (adding #ifdef CUDA) is called by GPUFOHelasCallWriter.generate_helas_call]