cuda dynamic link makefile ,compiles but not excuted

Accelerated Computing CUDA CUDA Programming and Performance
#1

Hello, I have a project that runs just fine from nsight eclipse.
I created a makefile and it compiles fine but when I try to run it ,it gives segmentaion fault.
I noticed this:

In my code , I print some values :

NbOfCDV : 286752
NbOfSTD : 286720
NbOfNOI : 32

TotalSize: 298713088 bytes

When the code is being executed from makefile ,it gives :

NbOfCDV : 286752
NbOfSTD : 0
NbOfNOI : 0

TotalSize: 8386658489117799521 bytes

!!!

What am I doing wrong??

Makefile:

PROJECT_NAME = Calc

# Location of the CUDA Toolkit
CUDA_PATH     = /usr/local/cuda-6.5
CUDA_BIN_PATH ?= $(CUDA_PATH)/bin

# compiler
GCC = gcc
NVCC = $(CUDA_BIN_PATH)/nvcc

# Compiler flags
CCFLAGS 		:= -m64 -std=c99 
NVCCFLAGS   	:= --target-cpu-architecture x86 -m64 -rdc=true 
LDFLAGS     	:= -lcudart -lcublas_device -lcudadevrt -lcublas -lcuda -lm 
DLINKFLAGS 		:= -lcublas_device -lcudadevrt -lcublas -lcudart

#debug
DEBUG ?= 0
ifeq (DEBUG, 1)
    CCFLAGS 	+= -DDEBUG -g 
    NVCCFLAGS 	+= -DDEBUG -g -G
else
    CCFLAGS 	+= -DNDEBUG
    NVCCFLAGS 	+= -DNDEBUG
endif

# Common includes and library paths 
INCLUDES  := -I/usr/local/cuda-6.5/include ...

LIBRARIES := -L/usr/local/cuda-6.5/lib64 ...

# CUDA code generation flags
GENCODE_SM35    := -gencode arch=compute_35,code=sm_35
GENCODE_FLAGS   := $(GENCODE_SM35)

# Target rules
.PHONY: all
all: build

build: $(PROJECT_NAME)

Calc.o:Calc.cu	
	$(NVCC)	$(NVCCFLAGS) $(MKLFLAGS) $(GENCODE_FLAGS) $(INCLUDES)  -c $<
	$(NVCC) -dlink $(GENCODE_FLAGS)  $(DLINKFLAGS) -o Calc_link.o $@ 
			
PhListFile.o:PhListFile.c		
	$(GCC) $(CCFLAGS) $(INCLUDES) -c $<
			
PhFilesUtil.o:PhFilesUtil.c			
	$(GCC) $(CCFLAGS) $(INCLUDES) -c $<
	
	
$(PROJECT_NAME):Calc.o	Calc_link.o	 PhListFile.o	PhFilesUtil.o		    	 
	$(GCC) $(CCFLAGS) -o  $@ $+   $(LIBRARIES) $(LDFLAGS)
		

run: build
	./$(PROJECT_NAME)

.PHONY: clean
clean:
	rm -f *.o $(PROJECT_NAME)
#2

Any ideas for this?

Thanks!

#3

A seg fault occurs in host code, not device code. Have you tracked down the line in your host code that is causing the seg fault? If not, do that first.

#4

But , when I debug through nsight everything is ok…
Not even a warning.

If ,after “make” ,I use from command line either cuda-gdb or gdb ,it just shows :

Program received signal SIGSEGV, Segmentation fault.
0x00000034d64897cb in memcpy () from /lib64/libc.so.6

backtrace

#0  0x00000034d64897cb in memcpy () from /lib64/libc.so.6
#1  0x000000000045d237 in LoadSample ()
#2  0x0000000000450429 in TestList ()
#3  0x000000000044f5e0 in main ()

I don’t know if/how to extract more information from this.

The LoadSample is in TestList and main calls TestList.

But I repeat , the program runs fine through nsight!

#5

Use printf to subdivide the program that seg faults into pieces. Eventually you will get down to the line that is causing the seg fault.