Makefile leads to compilation error

Question

I'm trying to compile a program (which isn't mine):

make -f makefile

... using the following makefile:

# Compiler for .cpp files
CPP  = g++

# Use nvcc to compile .cu files
NVCC = nvcc
NVCCFLAGS = -arch sm_20 # For fermi's in keeneland

# Add CUDA Paths
ICUDA = /usr/lib/nvidia-cuda-toolkit/include
LCUDA = /usr/lib/nvidia-cuda-toolkit/lib64

# Add CUDA libraries to the link line
LFLAGS += -lcuda -lcudart -L$(LCUDA) -lgomp

# Include standard optimization flags
CPPFLAGS = -O3 -c -I $(ICUDA) -Xcompiler -fopenmp -DTHRUST_DEVICE_BACKEND=THRUST_DEVICE_BACKEND_OMP

# List of all the objects you need
OBJECTS  = timer.o ar1.o kGrid.o vfInit.o parameters.o

# Rule that tells make how to make the program from the objects
main :  main.o $(OBJECTS)
        $(CPP) -o main main.o $(OBJECTS) $(LFLAGS)

# Rule that tells make how to turn a .cu file into a .o
%.o: %.cu
                $(NVCC) ${NVCCFLAGS} $(CPPFLAGS) -c $<

# How does make know how to turn a .cpp into a .o?  It's built-in!
# but if you wanted to type it out it would look like:
# %.o: %.cpp
#       $(CPP) $(CPPFLAGS) -c $<

clean :
        rm -f *.o
        rm -f core core.*

veryclean :
        rm -f *.o
        rm -f core core.*
        rm -f main

Which results in the following commands:

nvcc -arch sm_20  -O3 -c -I /usr/lib/nvidia-cuda-toolkit/include -Xcompiler -fopenmp -DTHRUST_DEVICE_BACKEND=THRUST_DEVICE_BACKEND_OMP -c main.cu
g++  -O3 -c -I /usr/lib/nvidia-cuda-toolkit/include -Xcompiler -fopenmp -DTHRUST_DEVICE_BACKEND=THRUST_DEVICE_BACKEND_OMP  -c -o timer.o timer.cpp
g++: error: unrecognized command line option â-Xcompilerâ
make: *** [timer.o] Error 1

I don't understand the makefile: the -xCompiler flag (in the variable CPPFLAGS) should be used only by the nvcc compiler, not g++. Therefore, I understand why I am getting an error. However, I don't understand, from my basic understanding of the makefile above, why at some point the variable CPPFLAGS follows g++ (variable CPP). I don't see any such sequence in the makefile.

Answer 1

Your main rule requires timer.o. There is no explicit rule for timer.o so make uses a built in implicit rule (as mentioned in the comment at the end of your makefile). The implicit rule for converting .cpp files into .o files has the form

$(CPP) $(CPPFLAGS) -c $<

So it's compiling using the options in CPPFLAGS which contains -Xcompiler. You probably want the -Xcompiler flag to be in NVCCFLAGS and not CPPFLAGS.