display in Cuda

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi everybody,

Sorry for my English, it’s not very good.

So, I use Cuda associated with C++ to develop a program on GPU and CPU.

I would like display an answers in the function device host but I don’t know how I do.

And I don’t understand how work cuPrint? Maybe, there are others possibility?

I hope someone can help me.

Alaninho

2

For the later cards and cudatoolkit printf command works from device, global and host functions.

3

If your card is of computability more than 2.0 you can use printf.
You have to define sm >=20 when you compile your code to be able to use printf()
After what you can use printf like in a cpu program.

4

just put printf in the code where you need to check and then when you compile add the flag -arch=sm_20 . If you put printf in a device or global function each thread will execute it, so you will get lots of prints.

5

How I know that my card is a computable with more 2.0?

And I have to write sm>=20 like option for the nvcc?

Sorry but I begin with cuda.

Alaninho

6

you can check the the nvidia webpage (CUDA GPUs - Compute Capability | NVIDIA Developer) or you can run the devicequery example from the SDK examples and check the compute capability of your device.

7

pasoleatis explained better than me, use a flag for compilation.

I was not really clear it is true, I will try to explain better next time.

If you know the name of your graphic card, have a look in google to know the compute capability. I don’t know a direct link with all the compute capability for all graphics cards.

Edit: pasoleatis definitively gave you the right answer.

8

Thank you, the compute capability is 2.1, thus, I can use printf?

I don’t understand the flag history? I must to write -arch=sm_20 in the line with nvcc?

Can you give me a example please?

Thank you pasoleatis and Dext for you help

9

If you have linux you compile somethong like nvcc -O2 -arch=sm_20 your_code.cu On windows you have projects and I have no idea.

10

I have a Makefile but I don’t where I can write -02 -arch=sm_20 my_code.cu

This my Makefile :

NVCC= nvcc

CPPC= g++

LIBEASEA=$(EZ_PATH)libeasea/

CXXFLAGS+=-g -Wall -O2 -I$(LIBEASEA)include -I$(EZ_PATH)boost

LDFLAGS=$(EZ_PATH)boost/program_options.a $(LIBEASEA)libeasea.a -lpthread

pemsa_arbre_SRC= pemsa_arbreIndividual.cpp

pemsa_arbre_MAIN_HDR= pemsa_arbre.cpp

pemsa_arbre_UC_HDR= pemsa_arbreIndividual.hpp

pemsa_arbre_HDR= $(pemsa_arbre_SRC:.cpp=.hpp) 

SRC= $(pemsa_arbre_SRC) $(pemsa_arbre_MAIN_HDR)

CUDA_SRC = pemsa_arbreIndividual.cu

HDR= $(pemsa_arbre_HDR) $(pemsa_arbre_UC_HDR)

OBJ= $(pemsa_arbre_SRC:.cpp=.o) $(pemsa_arbre_MAIN_HDR:.cpp=.o)

#USER MAKEFILE OPTIONS :

CPPFLAGS+=

#END OF USER MAKEFILE OPTIONS

CPPFLAGS+= -I$(LIBEASEA)include -I$(EZ_PATH)boost

NVCCFLAGS+= --compiler-options -fpermissive

BIN= pemsa_arbre

all:$(BIN)

$(BIN):$(OBJ)

	$(NVCC)  $^ -o $@ $(LDFLAGS) 

%.o:%.cu

	$(NVCC) $(NVCCFLAGS) -o $@ $< -c -DTIMING $(CPPFLAGS) -g -Xcompiler -Wall

easeaclean: clean

	rm -f Makefile pemsa_arbre.prm $(SRC) $(HDR) pemsa_arbre.mak $(CUDA_SRC) *.linkinfo pemsa_arbre.png pemsa_arbre.dat pemsa_arbre.vcproj pemsa_arbre.plot pemsa_arbre.r pemsa_arbre.csv pemsa_arbre.pop

clean:

	rm -f $(OBJ) $(BIN)
11

Try by adding this line, it should works. NVCCFLAGS means you add the flag you want to use for the nvcc compiler

#USER MAKEFILE OPTIONS :

NVCCFLAGS+= -arch=sm_20

#END OF USER MAKEFILE OPTIONS

If it works also try with -arch=sm_21 (I don’t know if that one works, so say me if it does work. I am curious)