(3 in pipeline) - runcard fixes, multi-backend gridpacks, python/bash profiling in tlau

epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/CHANGELOG.md

@@ -6,12 +6,23 @@ The format is loosely based on [Keep a Changelog](https://keepachangelog.com).
 
 --------------------------------------------------------------------------------
 
-## [Unreleased] - 2024-10-06
+## [Unreleased] - 2024-10-08
+
+### Added
+
+- Infrastructure issues
+  - AV ([#945]) Add cudacpp_bldall runcard to produce multi-backend gridpacks.
+  - AV ([#700]) Add cudacpp_helinl and cudacpp_hrdcod runcards to support HELINL=1 and HRDCOD=1 builds.
+  - AV ([#957]) In internal "tlau" tests, instrument python code in gridpacks to provide timing profiles for event generation.
+  - AV Enhance internal "tlau" tests and add a few test logs and various scripts to analyse them.
 
 ### Changed
 
 - Updated cudacpp version to 1.00.01.
 
+- Infrastructure issues
+  - AV ([#700]) Renamed the floating_type runcard as cudacpp_fptype for consistency with other cudacpp runcards.
+
 ### Fixed
 
 - Platform-specific issues
@@ -56,9 +67,12 @@ The format is loosely based on [Keep a Changelog](https://keepachangelog.com).
 [Unreleased]: https://github.com/madgraph5/madgraph4gpu/releases/compare/cudacpp_for3.6.0_v1.00.00...HEAD
 
 [#601]: https://github.com/madgraph5/madgraph4gpu/issues/601
+[#700]: https://github.com/madgraph5/madgraph4gpu/issues/700
 [#846]: https://github.com/madgraph5/madgraph4gpu/issues/846
 [#942]: https://github.com/madgraph5/madgraph4gpu/issues/942
 [#944]: https://github.com/madgraph5/madgraph4gpu/issues/944
+[#945]: https://github.com/madgraph5/madgraph4gpu/issues/945
+[#957]: https://github.com/madgraph5/madgraph4gpu/issues/957
 [#959]: https://github.com/madgraph5/madgraph4gpu/issues/959
 [#993]: https://github.com/madgraph5/madgraph4gpu/issues/993
 [#995]: https://github.com/madgraph5/madgraph4gpu/issues/995

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

@@ -1,12 +1,17 @@
 diff --git b/epochX/cudacpp/gg_tt.mad/SubProcesses/makefile a/epochX/cudacpp/gg_tt.mad/SubProcesses/makefile
-index 348c283be..49e6800ff 100644
+index 348c283be..2e5bff046 100644
 --- b/epochX/cudacpp/gg_tt.mad/SubProcesses/makefile
 +++ a/epochX/cudacpp/gg_tt.mad/SubProcesses/makefile
-@@ -1,6 +1,37 @@
+@@ -1,6 +1,42 @@
 +SHELL := /bin/bash
 +
  include ../../Source/make_opts
 +
++# Export the relevant settings from make_opts to builds using the cudacpp makefile (#700)
++export FPTYPE
++export HELINL
++export HRDCOD
++
 +# Determine CUDACPP_BUILDDIR based on the user-defined choices of BACKEND, FPTYPE, HELINL, HRDCOD and USEBUILDDIR (#829)
 +# Stop with an error if BACKEND=cuda and nvcc is missing or if BACKEND=hip and hipcc is missing
 +include ../../src/cudacpp_config.mk
@@ -40,7 +45,7 @@ index 348c283be..49e6800ff 100644
  # Load additional dependencies of the bias module, if present
  ifeq (,$(wildcard ../bias_dependencies))
  BIASDEPENDENCIES =
-@@ -24,7 +55,20 @@ else
+@@ -24,7 +60,20 @@ else
      MADLOOP_LIB =
  endif
 
@@ -62,7 +67,7 @@ index 348c283be..49e6800ff 100644
 
  LIBS = $(LIBDIR)libbias.$(libext) $(LIBDIR)libdhelas.$(libext) $(LIBDIR)libdsample.$(libext) $(LIBDIR)libgeneric.$(libext) $(LIBDIR)libpdf.$(libext) $(LIBDIR)libgammaUPC.$(libext) $(LIBDIR)libmodel.$(libext) $(LIBDIR)libcernlib.$(libext) $(MADLOOP_LIB) $(LOOP_LIBS)
 
-@@ -43,41 +87,148 @@ ifeq ($(strip $(MATRIX_HEL)),)
+@@ -43,41 +92,150 @@ ifeq ($(strip $(MATRIX_HEL)),)
  endif
 
 
@@ -153,32 +158,34 @@ index 348c283be..49e6800ff 100644
 +	rm -f $(PROG)
 +	ln -s $(PROG)_fortran $(PROG)
 +
-+madevent_cuda_link:
-+	$(MAKE) USEGTEST=0 BACKEND=cuda $(CUDACPP_BUILDDIR)/$(PROG)_cuda
++madevent_cudacpp_link: all
 +	rm -f $(PROG)
++ifeq ($(BACKEND),cuda)
 +	ln -s $(CUDACPP_BUILDDIR)/$(PROG)_cuda $(PROG)
-+
-+madevent_hip_link:
-+	$(MAKE) USEGTEST=0 BACKEND=hip $(CUDACPP_BUILDDIR)/$(PROG)_hip
-+	rm -f $(PROG)
++else ifeq ($(BACKEND),hip)
 +	ln -s $(CUDACPP_BUILDDIR)/$(PROG)_hip $(PROG)
-+
-+madevent_cpp_link:
-+	$(MAKE) USEGTEST=0 BACKEND=cppauto $(CUDACPP_BUILDDIR)/$(PROG)_cpp
-+	rm -f $(PROG)
++else
 +	ln -s $(CUDACPP_BUILDDIR)/$(PROG)_cpp $(PROG)
++endif
++
++madevent_cuda_link:
++	$(MAKE) USEGTEST=0 USEBUILDDIR=1 BACKEND=cuda madevent_cudacpp_link
 
 -$(LIBDIR)libpdf.$(libext): 
 -	cd ../../Source/PDF; make
-+override SUPPORTED_AVXS = cppnone cppsse4 cppavx2 cpp512y cpp512z cppauto
-+madevent_%_link:
-+	@if [ '$(words $(filter $*, $(SUPPORTED_AVXS)))' != '1' ]; then echo "ERROR! Invalid target '$@' (supported madevent_cpp*_link targets are: $(foreach avx,$(SUPPORTED_AVXS),'madevent_cpp$(avx)_link'))"; exit 1; fi
-+	$(MAKE) USEGTEST=0 BACKEND=$* $(CUDACPP_BUILDDIR)/$(PROG)_cpp
-+	rm -f $(PROG)
-+	ln -s $(CUDACPP_BUILDDIR)/$(PROG)_cpp $(PROG)
++madevent_hip_link:
++	$(MAKE) USEGTEST=0 USEBUILDDIR=1 BACKEND=hip madevent_cudacpp_link
 
 -$(LIBDIR)libgammaUPC.$(libext):
 -	cd ../../Source/PDF/gammaUPC; make
++madevent_cpp_link:
++	$(MAKE) USEGTEST=0 USEBUILDDIR=1 BACKEND=cppauto madevent_cudacpp_link
++
++override SUPPORTED_AVXS = cppnone cppsse4 cppavx2 cpp512y cpp512z cppauto
++madevent_%_link:
++	@if [ '$(words $(filter $*, $(SUPPORTED_AVXS)))' != '1' ]; then echo "ERROR! Invalid target '$@' (supported madevent_cpp*_link targets are: $(foreach avx,$(SUPPORTED_AVXS),'madevent_$(avx)_link'))"; exit 1; fi
++	$(MAKE) USEGTEST=0 USEBUILDDIR=1 BACKEND=$* madevent_cudacpp_link
++
 +# Building $(PROG)_cpp no longer builds $(PROG)_cuda if CUDACPP_BACKENDLIB for cuda exists (this was the case in the past to allow cpp-only builds #503)
 +$(CUDACPP_BUILDDIR)/$(PROG)_cpp: $(PROCESS) $(DSIG_cudacpp) auto_dsig.o $(LIBS) $(MATRIX) counters.o ompnumthreads.o $(CUDACPP_BUILDDIR)/.cudacpplibs
 +	$(FC) -o $@ $(PROCESS) $(DSIG_cudacpp) auto_dsig.o $(MATRIX) $(LINKLIBS) $(BIASDEPENDENCIES) $(OMPFLAGS) counters.o ompnumthreads.o -L$(LIBDIR)/$(CUDACPP_BUILDDIR) -l$(CUDACPP_COMMONLIB) -l$(CUDACPP_BACKENDLIB) $(LIBFLAGSRPATH) $(LDFLAGS)
@@ -227,7 +234,7 @@ index 348c283be..49e6800ff 100644
 
  # Dependencies
 
-@@ -97,5 +248,80 @@ unwgt.o: genps.inc nexternal.inc symswap.inc cluster.inc run.inc message.inc \
+@@ -97,5 +255,80 @@ unwgt.o: genps.inc nexternal.inc symswap.inc cluster.inc run.inc message.inc \
  	 run_config.inc
  initcluster.o: message.inc
 

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

@@ -33,10 +33,16 @@ def compile(self, *args, **opts):
         if 'cwd' in opts and os.path.basename(opts['cwd']) == 'Source':
             path = pjoin(opts['cwd'], 'make_opts')
             common_run_interface.CommonRunCmd.update_make_opts_full(path,
-                {'FPTYPE': self.run_card['floating_type'] })
-            misc.sprint('FPTYPE checked')
+                {'FPTYPE': self.run_card['cudacpp_fptype'],
+                 'HELINL': self.run_card['cudacpp_helinl'],
+                 'HRDCOD': self.run_card['cudacpp_hrdcod'] })
+            misc.sprint('FPTYPE, HELINL, HRDCOD checked')
         cudacpp_supported_backends = [ 'fortran', 'cuda', 'hip', 'cpp', 'cppnone', 'cppsse4', 'cppavx2', 'cpp512y', 'cpp512z', 'cppauto' ]
         if args and args[0][0] == 'madevent' and hasattr(self, 'run_card'):            
+            if self.run_card['cudacpp_bldall'] == True: # pre-build all backends #945
+                logger.info("Pre-building madevent in madevent_interface.py with ALL matrix elements")
+                args[0][0] = 'bldall'
+                misc.compile(nb_core=self.options['nb_core'], *args, **opts)
             cudacpp_backend = self.run_card['cudacpp_backend'].lower() # the default value is defined in launch_plugin.py
             logger.info("Building madevent in madevent_interface.py with '%s' matrix elements"%cudacpp_backend)
             if cudacpp_backend in cudacpp_supported_backends :
@@ -47,14 +53,18 @@ def compile(self, *args, **opts):
         else:
             return misc.compile(nb_core=self.options['nb_core'], *args, **opts)
 
-# Phase-Space Optimization ------------------------------------------------------------------------------------
+# CUDACPP runcard block ------------------------------------------------------------------------------------
 template_on = \
 """#***********************************************************************
 # SIMD/GPU configuration for the CUDACPP plugin
 #************************************************************************
- %(cudacpp_backend)s = cudacpp_backend ! CUDACPP backend: fortran, cuda, hip, cpp, cppnone, cppsse4, cppavx2, cpp512y, cpp512z, cppauto
+ %(cudacpp_backend)s = cudacpp_backend ! CUDACPP backend: fortran, cuda, hip, cpp (DEFAULT), cppnone, cppsse4, cppavx2, cpp512y, cpp512z, cppauto
+#*** WARNING! The following cudacpp runcards are experimental! Users should normally change only the cudacpp_backend card ***
+ %(cudacpp_fptype)s = cudacpp_fptype ! CUDACPP floating point precision: f (single), d (double), m (mixed, DEFAULT: double for amplitudes, single for colors)
+ %(cudacpp_hrdcod)s = cudacpp_hrdcod ! CUDACPP parameter hardcoding: 0 (DEFAULT, parameters not hardcoded: read param_card.dat at runtime), 1 (hardcoded parameters)
+ %(cudacpp_helinl)s = cudacpp_helinl ! CUDACPP helicity amplitude inlining: 0 (DEFAULT, ordinary inlining of templates), 1 (aggressive inlining with 'always inline')
+ %(cudacpp_bldall)s = cudacpp_bldall ! CUDACPP build all available backends in separate build directories: False, True
 """
-
 template_off = ''
 plugin_block = banner_mod.RunBlock('simd', template_on=template_on, template_off=template_off)
 
@@ -75,23 +85,49 @@ def reset_simd(self, old_value, new_value, name):
     def reset_makeopts(self, old_value, new_value, name):
         if not hasattr(self, 'path'):
             raise Exception
-        if name == 'floating_type':
+        if name == 'cudacpp_fptype':
             common_run_interface.CommonRunCmd.update_make_opts_full({'FPTYPE': new_value})
+        elif name == 'cudacpp_hrdcod':
+            raise Exception('Cannot change cudacpp_hrdcod')
+        elif name == 'cudacpp_helinl':
+            raise Exception('Cannot change cudacpp_helinl')
+        elif name == 'cudacpp_bldall':
+            raise Exception('Cannot change cudacpp_bldall')
         else:
             raise Exception
         Sourcedir = pjoin(os.path.dirname(os.path.dirname(self.path)), 'Source')
         subprocess.call(['make', 'cleanall'], cwd=Sourcedir, stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL)
 
     def default_setup(self):
         super().default_setup()
-        self.add_param('floating_type', 'm', include=False, hidden=True,
-                       fct_mod=(self.reset_makeopts,(),{}),
-                       allowed=['m','d','f'],
-                       comment='floating point precision: f (single), d (double), m (mixed: double for amplitudes, single for colors)'
-                       )
         cudacpp_supported_backends = [ 'fortran', 'cuda', 'hip', 'cpp', 'cppnone', 'cppsse4', 'cppavx2', 'cpp512y', 'cpp512z', 'cppauto' ]
-        self.add_param('cudacpp_backend', 'cpp', include=False, hidden=False,
+        self.add_param('cudacpp_backend', 'cpp',
+                       include=False, # AV: 'include=True' would add "CUDACPP_BACKEND = 'cpp'" to run_card.inc
+                       hidden=False, # AV: keep cudacpp_backend in runcard template and keep 'hidden='False'
                        allowed=cudacpp_supported_backends)
+        self.add_param('cudacpp_fptype', 'm',
+                       include=False, # AV: 'include=True' would add "CUDACPP_FPTYPE = 'm'" to run_card.inc (if fct_mod is removed, else codegen fails)
+                       hidden=False, # AV: add cudacpp_backend to runcard template and keep 'hidden='False'
+                       fct_mod=(self.reset_makeopts,(),{}), # AV: I assume this forces a 'make cleanavx' if FPTYPE changes?
+                       allowed=['m','d','f']
+                       )
+        self.add_param('cudacpp_helinl', '0',
+                       include=False,  # AV: no need to add this parameter to run_card.inc
+                       hidden=False, # AV: add cudacpp_helinl to runcard template and keep 'hidden='False'
+                       fct_mod=(self.reset_makeopts,(),{}), # AV: I assume this raises an exception if cudacpp_helinl changes?
+                       allowed=['0','1']
+                       )
+        self.add_param('cudacpp_hrdcod', '0',
+                       include=False,  # AV: no need to add this parameter to run_card.inc
+                       hidden=False, # AV: add cudacpp_hrdcod to runcard template and keep 'hidden='False'
+                       fct_mod=(self.reset_makeopts,(),{}), # AV: I assume this raises an exception if cudacpp_hrdcod changes?
+                       allowed=['0','1']
+                       )
+        self.add_param('cudacpp_bldall', False,
+                       include=False, # AV: no need to add this parameter to run_card.inc
+                       hidden=False, # AV: add cudacpp_bldall to runcard template and keep 'hidden='False'
+                       fct_mod=(self.reset_makeopts,(),{}), # AV: I assume this raises an exception if cudacpp_bldall changes?
+                       )
         self['vector_size'] = 16 # already setup in default class (just change value)
         self['aloha_flag'] = '--fast-math'
         self['matrix_flag'] = '-O3'

epochX/cudacpp/CODEGEN/generateAndCompare.sh

@@ -199,6 +199,38 @@ function codeGenAndDiff()
     susy_gg_ulul)
       cmd="import model MSSM_SLHA2; generate g g > ul ul~"
       ;;
+    pp_dy012j) # from Jin #942
+      cmd="import model sm-no_b_mass
+      define p = u d c s b u~ d~ c~ s~ b~ g
+      define j = p
+      define ell+ = e+ mu+ ta+
+      define ell- = e- mu- ta-
+      define nu = ve vm vt
+      define nubar = ve~ vm~ vt~
+      generate p p > ell+ ell- @0
+      add process p p > ell+ ell- j @1
+      add process p p > ell+ ell- j j @2"
+      ;;
+    pp_dy3j) # from Jin #942
+      cmd="import model sm-no_b_mass
+      define p = u d c s b u~ d~ c~ s~ b~ g
+      define j = p
+      define ell+ = e+ mu+ ta+
+      define ell- = e- mu- ta-
+      define nu = ve vm vt
+      define nubar = ve~ vm~ vt~
+      generate p p > ell+ ell- j j j @0"
+      ;;
+    pp_dy4j) # extend dy3j from Jin #942
+      cmd="import model sm-no_b_mass
+      define p = u d c s b u~ d~ c~ s~ b~ g
+      define j = p
+      define ell+ = e+ mu+ ta+
+      define ell- = e- mu- ta-
+      define nu = ve vm vt
+      define nubar = ve~ vm~ vt~
+      generate p p > ell+ ell- j j j j @0"
+      ;;
     atlas)
       cmd="import model sm-no_b_mass
       define p = g u c d s b u~ c~ d~ s~ b~

epochX/cudacpp/bldall.sh

@@ -0,0 +1,32 @@
+#!/bin/bash
+
+if [ "$1" == "" ] || [ "$2" != "" ]; then
+  echo "Usage: $0 <process_directory>"
+  exit 1
+fi
+spdir=${1}/SubProcesses
+
+if [ ! -d ${spdir} ]; then
+  echo "ERROR! Directory not found: ${spdir}"
+  exit 1
+fi
+cd ${spdir}
+
+START=$(date +%s)
+
+TMP=${START}
+for pdir in P*; do
+  echo "--------------------------------------------------------------------------------"
+  cd ${pdir}
+  pwd
+  sleep 1
+  make -j bldall
+  cd - > /dev/null
+  END=$(date +%s)
+  echo "ELAPSED: $((END-TMP)) seconds"
+  TMP=${END}
+done
+echo "================================================================================"
+
+END=$(date +%s)
+echo "ELAPSED: $((END-START)) seconds"

epochX/cudacpp/ee_mumu.mad/Cards/run_card.dat

@@ -203,5 +203,10 @@ systematics = systematics_program ! none, systematics [python], SysCalc [deprece
 #***********************************************************************
 # SIMD/GPU configuration for the CUDACPP plugin
 #************************************************************************
- cpp = cudacpp_backend ! CUDACPP backend: fortran, cuda, hip, cpp, cppnone, cppsse4, cppavx2, cpp512y, cpp512z, cppauto
+ cpp = cudacpp_backend ! CUDACPP backend: fortran, cuda, hip, cpp (DEFAULT), cppnone, cppsse4, cppavx2, cpp512y, cpp512z, cppauto
+#*** WARNING! The following cudacpp runcards are experimental! Users should normally change only the cudacpp_backend card ***
+ m = cudacpp_fptype ! CUDACPP floating point precision: f (single), d (double), m (mixed, DEFAULT: double for amplitudes, single for colors)
+ 0 = cudacpp_hrdcod ! CUDACPP parameter hardcoding: 0 (DEFAULT, parameters not hardcoded: read param_card.dat at runtime), 1 (hardcoded parameters)
+ 0 = cudacpp_helinl ! CUDACPP helicity amplitude inlining: 0 (DEFAULT, ordinary inlining of templates), 1 (aggressive inlining with 'always inline')
+ False = cudacpp_bldall ! CUDACPP build all available backends in separate build directories: False, True
 

epochX/cudacpp/ee_mumu.mad/Cards/run_card_default.dat

@@ -203,5 +203,10 @@ systematics = systematics_program ! none, systematics [python], SysCalc [deprece
 #***********************************************************************
 # SIMD/GPU configuration for the CUDACPP plugin
 #************************************************************************
- cpp = cudacpp_backend ! CUDACPP backend: fortran, cuda, hip, cpp, cppnone, cppsse4, cppavx2, cpp512y, cpp512z, cppauto
+ cpp = cudacpp_backend ! CUDACPP backend: fortran, cuda, hip, cpp (DEFAULT), cppnone, cppsse4, cppavx2, cpp512y, cpp512z, cppauto
+#*** WARNING! The following cudacpp runcards are experimental! Users should normally change only the cudacpp_backend card ***
+ m = cudacpp_fptype ! CUDACPP floating point precision: f (single), d (double), m (mixed, DEFAULT: double for amplitudes, single for colors)
+ 0 = cudacpp_hrdcod ! CUDACPP parameter hardcoding: 0 (DEFAULT, parameters not hardcoded: read param_card.dat at runtime), 1 (hardcoded parameters)
+ 0 = cudacpp_helinl ! CUDACPP helicity amplitude inlining: 0 (DEFAULT, ordinary inlining of templates), 1 (aggressive inlining with 'always inline')
+ False = cudacpp_bldall ! CUDACPP build all available backends in separate build directories: False, True
 

epochX/cudacpp/ee_mumu.mad/SubProcesses/makefile

@@ -2,6 +2,11 @@ SHELL := /bin/bash
 
 include ../../Source/make_opts
 
+# Export the relevant settings from make_opts to builds using the cudacpp makefile (#700)
+export FPTYPE
+export HELINL
+export HRDCOD
+
 # Determine CUDACPP_BUILDDIR based on the user-defined choices of BACKEND, FPTYPE, HELINL, HRDCOD and USEBUILDDIR (#829)
 # Stop with an error if BACKEND=cuda and nvcc is missing or if BACKEND=hip and hipcc is missing
 include ../../src/cudacpp_config.mk
@@ -162,27 +167,29 @@ madevent_fortran_link: $(PROG)_fortran
 	rm -f $(PROG)
 	ln -s $(PROG)_fortran $(PROG)
 
-madevent_cuda_link:
-	$(MAKE) USEGTEST=0 BACKEND=cuda $(CUDACPP_BUILDDIR)/$(PROG)_cuda
+madevent_cudacpp_link: all
 	rm -f $(PROG)
+ifeq ($(BACKEND),cuda)
 	ln -s $(CUDACPP_BUILDDIR)/$(PROG)_cuda $(PROG)
+else ifeq ($(BACKEND),hip)
+	ln -s $(CUDACPP_BUILDDIR)/$(PROG)_hip $(PROG)
+else
+	ln -s $(CUDACPP_BUILDDIR)/$(PROG)_cpp $(PROG)
+endif
+
+madevent_cuda_link:
+	$(MAKE) USEGTEST=0 USEBUILDDIR=1 BACKEND=cuda madevent_cudacpp_link
 
 madevent_hip_link:
-	$(MAKE) USEGTEST=0 BACKEND=hip $(CUDACPP_BUILDDIR)/$(PROG)_hip
-	rm -f $(PROG)
-	ln -s $(CUDACPP_BUILDDIR)/$(PROG)_hip $(PROG)
+	$(MAKE) USEGTEST=0 USEBUILDDIR=1 BACKEND=hip madevent_cudacpp_link
 
 madevent_cpp_link:
-	$(MAKE) USEGTEST=0 BACKEND=cppauto $(CUDACPP_BUILDDIR)/$(PROG)_cpp
-	rm -f $(PROG)
-	ln -s $(CUDACPP_BUILDDIR)/$(PROG)_cpp $(PROG)
+	$(MAKE) USEGTEST=0 USEBUILDDIR=1 BACKEND=cppauto madevent_cudacpp_link
 
 override SUPPORTED_AVXS = cppnone cppsse4 cppavx2 cpp512y cpp512z cppauto
 madevent_%_link:
-	@if [ '$(words $(filter $*, $(SUPPORTED_AVXS)))' != '1' ]; then echo "ERROR! Invalid target '$@' (supported madevent_cpp*_link targets are: $(foreach avx,$(SUPPORTED_AVXS),'madevent_cpp$(avx)_link'))"; exit 1; fi
-	$(MAKE) USEGTEST=0 BACKEND=$* $(CUDACPP_BUILDDIR)/$(PROG)_cpp
-	rm -f $(PROG)
-	ln -s $(CUDACPP_BUILDDIR)/$(PROG)_cpp $(PROG)
+	@if [ '$(words $(filter $*, $(SUPPORTED_AVXS)))' != '1' ]; then echo "ERROR! Invalid target '$@' (supported madevent_cpp*_link targets are: $(foreach avx,$(SUPPORTED_AVXS),'madevent_$(avx)_link'))"; exit 1; fi
+	$(MAKE) USEGTEST=0 USEBUILDDIR=1 BACKEND=$* madevent_cudacpp_link
 
 # Building $(PROG)_cpp no longer builds $(PROG)_cuda if CUDACPP_BACKENDLIB for cuda exists (this was the case in the past to allow cpp-only builds #503)
 $(CUDACPP_BUILDDIR)/$(PROG)_cpp: $(PROCESS) $(DSIG_cudacpp) auto_dsig.o $(LIBS) $(MATRIX) counters.o ompnumthreads.o $(CUDACPP_BUILDDIR)/.cudacpplibs