Changes for GPU

m4/ax_cuda.m4

@@ -41,31 +41,43 @@ AC_DEFUN([AX_CUDA],
 AC_CHECK_PROG([NVCC_EXIST],[nvcc],["yes"],["no"])
 AS_IF([test "x$NVCC_EXIST" = "xno"],[],[
           NVCC=`which nvcc`
-          
+
           # Set CUDA_CFLAGS to $NVCC, where substring "bin/nvcc"
           # is substituted by "include".
-          CUDA_CFLAGS=" -I${NVCC%bin//nvcc}"
-          CUDA_CFLAGS=" -I${CUDA_CFLAGS%bin/nvcc}"
+          CUDA_CFLAGS=" ${NVCC%bin//nvcc}"
+          CUDA_CFLAGS=" ${CUDA_CFLAGS%bin/nvcc}"
           CUDA_CFLAGS=" -I${CUDA_CFLAGS}include"
 
           #Set CUDA_CFLAGS to $NVCC, where substring "bin/nvcc"
           #is substituted by "lib".
-          CUDA_LIBS=" -L${NVCC%bin//nvcc}"
-          CUDA_LIBS=" -L${CUDA_LIBS%bin/nvcc}"
+          CUDA_LIBS="${NVCC%bin//nvcc}"
+          CUDA_LIBS="${CUDA_LIBS%bin/nvcc}"
+          CUDA_PATH=$CUDA_LIBS
           CUDA_LIBS=" -L${CUDA_LIBS}lib"
 
           # If $build_cpu contains "_64", append "64" to CUDA_LIBS
           AS_IF([echo $build_cpu | grep -q "_64"],
-                [CUDA_LIBS+="64"])
-                
+                [
+                 AS_IF([ test -d $CUDA_PATH/lib64 ], [ CUDA_LIBS+="64" ], [])
+                 # Be carefull the return code 0 mean true return code 1 mean false
+                 AS_IF([ command -v bumblebeed >/dev/null ], [ CUDA_LIBS+=" -L/usr/lib64/nvidia-bumblebee/ "  ],
+                                                             [
+                                                               echo "bumblebee, NVIDIA optimus,  not found"
+                                                             ])
+                 AS_IF([ test -d /usr/local/cuda/lib64  ], [ CUDA_LIBS+=" -L/usr/local/cuda/lib64 "  ],
+                       [
+                        AS_IF([ test -d /usr/local/cuda/lib ],[ CUDA_LIBS+=" -L/usr/local/cuda/lib  "  ])
+                       ])
+                ])
+
           # Append " -lcuda -lcudart" to CUDA_LIBS
           CUDA_LIBS+=" -lcuda -lcudart"
-
+          
           # Make variables available in Makefile.am
-          AC_SUBST(CUDA_CFLAGS)
-          AC_SUBST(CUDA_LIBS)
+          AC_SUBST([CUDA_CFLAGS])
+          AC_SUBST([CUDA_LIBS])
           echo $NVCC
-          AC_SUBST(NVCC)
+          AC_SUBST([NVCC])
           AC_DEFINE([NVCC],[],[NVCC compiling])
 ])dnl
 

src/Makefile.am

 
-LINKLIBS = $(DEFAULT_LIB) $(PTHREAD_LIBS) $(OPT_LIBS) $(HDF5_LDFLAGS) $(HDF5_LIBS) $(BOOST_LDFLAGS)
+LINKLIBS = $(DEFAULT_LIB) $(PTHREAD_LIBS) $(OPT_LIBS) $(HDF5_LDFLAGS) $(HDF5_LIBS) $(BOOST_LDFLAGS) $(CUDA_LIBS)
+
+if BUILDCUDA
+ CUDA_SOURCES=../../openfpm_devices/src/memory/CudaMemory.cu
+else
+ CUDA_SOURCES=
+endif
 
 noinst_PROGRAMS = vcluster_test
-vcluster_test_SOURCES = main.cpp VCluster/VCluster.cpp ../../openfpm_devices/src/memory/HeapMemory.cpp ../../openfpm_devices/src/memory/PtrMemory.cpp ../../openfpm_devices/src/Memleak_check.cpp
-vcluster_test_CXXFLAGS = $(AM_CXXFLAGS) $(INCLUDES_PATH)  $(BOOST_CPPFLAGS)
+vcluster_test_SOURCES = main.cpp VCluster/VCluster.cpp  ../../openfpm_devices/src/memory/HeapMemory.cpp ../../openfpm_devices/src/memory/PtrMemory.cpp ../../openfpm_devices/src/Memleak_check.cpp $(CUDA_SOURCES)
+vcluster_test_CXXFLAGS = $(AM_CXXFLAGS) $(INCLUDES_PATH)  $(BOOST_CPPFLAGS) $(CUDA_CFLAGS)
 vcluster_test_CFLAGS = $(CUDA_CFLAGS)
 vcluster_test_LDADD = $(LINKLIBS)
 
 lib_LIBRARIES = libvcluster.a
 libvcluster_a_SOURCES = VCluster/VCluster.cpp
-libvcluster_a_CXXFLAGS = $(AM_CXXFLAGS) $(INCLUDES_PATH) $(BOOST_CPPFLAGS)
+libvcluster_a_CXXFLAGS = $(AM_CXXFLAGS) $(INCLUDES_PATH) $(BOOST_CPPFLAGS) $(CUDA_CFLAGS)
 libvcluster_a_CFLAGS =
 
 nobase_include_HEADERS = MPI_wrapper/MPI_IallreduceW.hpp MPI_wrapper/MPI_IrecvW.hpp MPI_wrapper/MPI_IBcastW.hpp  MPI_wrapper/MPI_IsendW.hpp MPI_wrapper/MPI_util.hpp MPI_wrapper/MPI_IAllGather.hpp  \
@@ -17,7 +23,7 @@ VCluster/VCluster_base.hpp VCluster/VCluster.hpp VCluster/VCluster_meta_function
 util/Vcluster_log.hpp
 
 .cu.o :
-	$(NVCC) $(NVCCFLAGS) -o $@ -c $<
+	$(NVCC) $(NVCCFLAGS) $(INCLUDES_PATH) -o $@ -c $<
 
 test: vcluster_test
 	source $(HOME)/openfpm_vars &&  cd .. && mpirun -np 3 ./src/vcluster_test && mpirun -np 4 ./src/vcluster_test

src/VCluster/VCluster_semantic_unit_tests.hpp

@@ -1480,6 +1480,37 @@ BOOST_AUTO_TEST_CASE (Vcluster_semantic_sendrecv_6)
 	}
 }
 
+BOOST_AUTO_TEST_CASE( Vcluster_semantic_ssend_recv_layout_switch )
+{
+	auto & v_cl = create_vcluster();
+
+	if (v_cl.size() > 10)	{return;}
+
+	openfpm::vector<openfpm::vector_gpu<aggregate<float,float[3]>>> vd;
+	openfpm::vector_gpu<aggregate<float,float[3]>> collect;
+	openfpm::vector<size_t> prc_send;
+    openfpm::vector<size_t> prc_recv;
+    openfpm::vector<size_t> sz_recv;
+
+	vd.resize(v_cl.size());
+
+	for (size_t i = 0 ; i < vd.size() ; i++)
+	{
+		vd.get(i).resize(100);
+
+		for (size_t j = 0 ; j < vd.get(i).size() ; j++)
+		{
+			vd.get(i).template get<0>(j) = 10000*i + v_cl.rank()*100 + j;
+
+			vd.get(i).template get<1>(j)[0] = 400000 + 10000*i + v_cl.rank()*100 + j;
+			vd.get(i).template get<1>(j)[1] = 400000 + 10000*i + v_cl.rank()*100 + j;
+			vd.get(i).template get<1>(j)[2] = 400000 + 10000*i + v_cl.rank()*100 + j;
+		}
+	}
+
+	v_cl.SSendRecv<openfpm::vector_gpu<aggregate<float,float[3]>>,decltype(collect),memory_traits_inte>(vd,collect,prc_send, prc_recv,sz_recv);
+}
+
 /*BOOST_AUTO_TEST_CASE (Vcluster_semantic_bench_all_all)
 {
                 Vcluster & vcl = create_vcluster();