Makefile

At first, we need to comment out the variable STATIC_LIB and add a new one

DYNAMIC_LIB_BASE_NAME := cutil

That’s all for this file.

common.mk

Add this part before the #compiler section

# Build the final library name
DYNAMIC_LIB := lib$(DYNAMIC_LIB_BASE_NAME)
ifeq ($(emu), 1)
 DYNAMIC_LIB := $(DYNAMIC_LIB)Emu
endif

ifneq ($(DARWIN),)
 DYNAMIC_LIB := $(DYNAMIC_LIB)-apple
endif

DYNAMIC_LIB := $(DYNAMIC_LIB)_x86

ifneq "$(strip $(HP_64))" ""
 DYNAMIC_LIB := $(DYNAMIC_LIB)_64
endif

LIB_EXTENSION :=
ifneq ($(DARWIN),)
 LIB_EXTENSION := .dylib
else
 LIB_EXTENSION := .so
endif
DYNAMIC_LIB := $(DYNAMIC_LIB)$(LIB_EXTENSION)

Next, we modify the #Lib standard section so that it can work in both x86 and x86-64 environments.

# Libs
# standard
ifneq ($(DARWIN),)
 LIB      += -L$(CUDA_INSTALL_PATH)/lib -L$(LIBDIR) -L$(COMMONDIR)/lib -L$(COMMONDIR)/lib/darwin -lcuda -lcudart
else
 ifneq "$(strip $(HP_64))" ""
   LIB      += -L$(CUDA_INSTALL_PATH)/lib64 -L$(LIBDIR) -L$(COMMONDIR)/lib -L$(COMMONDIR)/lib/linux  -lcuda -lcudart
 else
   LIB      += -L$(CUDA_INSTALL_PATH)/lib -L$(LIBDIR) -L$(COMMONDIR)/lib -L$(COMMONDIR)/lib/linux  -lcuda -lcudart
 endif
endif

Now, we come to the section for NVCCFLAGS.

# append optional arch/SM version flags (such as -arch sm_11)
NVCCFLAGS += $(SMVERSIONFLAGS)
NVCCFLAGS += -gencode=arch=compute_10,code=sm_10 -gencode=arch=compute_10,code=compute_10
NVCCFLAGS += -gencode=arch=compute_13,code=sm_13 -gencode=arch=compute_13,code=sm_20

The fourth part is the section to set the flag for 32-bit or 64-bit machines

# architecture flag for cubin build
ifneq "$(strip $(HP_64))" ""
 CUBIN_ARCH_FLAG := -m64
else
 CUBIN_ARCH_FLAG := -m32
endif

The fifth part is the #Lib section

#Lib
LIB    +=  $(PARAMGLLIB) $(CUDPPLIB) ${LIB}

Finally, the most important part is the section we tell how to make a static library or a shared library

# Lib/exe configuration
ifneq ($(STATIC_LIB),)   # make static library
 ifeq ($(emu), 1)
   NVCCFLAGS   += -deviceemu
   CUDACCFLAGS +=
   BINSUBDIR   := emu$(BINSUBDIR)
   LIBSUFFIX   := _$(LIB_ARCH)$(LIBSUFFIX)_emu
 # consistency, makes developing easier
   CXXFLAGS    += -D__DEVICE_EMULATION__
   CFLAGS        += -D__DEVICE_EMULATION__
 endif
 TARGETDIR := $(LIBDIR)
 TARGET   := $(subst .a,$(LIBSUFFIX).a,$(LIBDIR)/$(STATIC_LIB))
 LINKLINE  = ar qv $(TARGET) $(OBJS); ranlib $(TARGET)
else ifneq ($(DYNAMIC_LIB_BASE_NAME),)       # make shared library
 TARGETDIR := $(LIBDIR)
 TARGET    := $(LIBDIR)/$(DYNAMIC_LIB)
# Compilers
 NVCC       := nvcc
 CXX        := g++ -fPIC
 CC         := gcc
 LINK       := g++ -fPIC

 NVCCFLAGS   += -Xcompiler '-fPIC'
 LINKLINE  = $(CC) $(CUBIN_ARCH_FLAG) -shared -fPIC -o $(TARGET) $(OBJS) $(LIB)
else    # make executable file
 LIB += -lcutil_$(LIB_ARCH)$(LIBSUFFIX)
 # Device emulation configuration
 ifeq ($(emu), 1)
 NVCCFLAGS   += -deviceemu
 CUDACCFLAGS +=
 BINSUBDIR   := emu$(BINSUBDIR)
 # consistency, makes developing easier
 CXXFLAGS    += -D__DEVICE_EMULATION__
 CFLAGS        += -D__DEVICE_EMULATION__
 endif
 TARGETDIR := $(BINDIR)/$(BINSUBDIR)
 TARGET    := $(TARGETDIR)/$(EXECUTABLE)
 LINKLINE  = $(LINK) -o $(TARGET) $(OBJS) $(LIB)
endif

That’s all, we will have libcutil_x86_64.so in the ../lib folder.

References:

1. http://forums.nvidia.com/lofiversion/index.php?t156326.html
2. http://www.evl.uic.edu/sjames/cs525/apps/common.mk

• CUDA 2.3
• gcc 4.3.3

Installation

First, do the followings

1. Unzip to /usr/local/cudpp
2. If you’re using a 64-bit system, modify all common.mk files to point to $CUDA_INSTALL_PATH/lib64, instead of $CUDA_INSTALL_PATH/lib
3. Jump to cudpp/common, type : make

If you download the zip file, you may get errors, so update the following files by #include the appropriate header files

• in common/src/cmd_arg_reader.cpp: <typeinfo>
• in common/inc/cmd_arg_reader.h: <typeinfo>
• in common/inc/exception.h: <cstdlib>
• in common/src/cutil.cpp: <cstring>

It’s better to check out the code from SVN using

svn checkout http://cudpp.googlecode.com/svn/trunk/ /usr/local/cudpp

That’s all, type

make clean

make

to build CUDA Utility Library (libcutil). The result looks like

ar: creating ./../lib/libcutil.a
a - obj/release/bank_checker.cpp_o
a - obj/release/cmd_arg_reader.cpp_o
a - obj/release/cutil.cpp_o
a - obj/release/stopwatch.cpp_o
a - obj/release/stopwatch_linux.cpp_o
a - obj/release/findFile.cpp_o

Then, jump to ../cudpp and type make. The result looks like

ar: creating ./../lib/libcudpp64.a
a - obj/release/cudpp.cpp_o
a - obj/release/cudpp_plan.cpp_o
a - obj/release/cudpp_plan_manager.cpp_o
a - obj/release/cudpp_maximal_launch.cpp_o
a - obj/release/compact_app.cu_o
a - obj/release/scan_app.cu_o
a - obj/release/segmented_scan_app.cu_o
a - obj/release/spmvmult_app.cu_o
a - obj/release/radixsort_app.cu_o
a - obj/release/rand_app.cu_o

You may need to change the common.mk file if you get linked error

CXX        := g++ -fPIC
CC         := gcc -fPIC
LINK       := g++ -fPIC

NOTE: There are two common.mk files, they are in linux_build/ folder and common/ folder.

Sample codes

Compile cudpp_testrig: you may get errors

sparse.h:46: error: ‘malloc’ was not declared in this scope
spmvmult_gold.cpp: In function ‘void readMatrixMarket(MMMatrix*, const char*)’:
spmvmult_gold.cpp:94: error: ‘exit’ was not declared in this scope
spmvmult_gold.cpp:122: error: ‘qsort’ was not declared in this scope

Then, you need to #include <cstdlib> to the file spmvmult_gold.cpp

1. CUDA (NVIDIA): original C, now extend to many languages
2. Stream framework (AMD ATI): now deprecated by OpenCL

C++

1. CUDA
2. Sieve C++ programming model
3. Intel Threading Building Block (TBB)
4. Intel Concurrent Collection (CnC)
5. RapidMind multi-core development platform (RapidMind Inc –> Intel)

Fortran

1. CUDA
2. PGI Accelerate

MultiLange:

1. OpenMP
2. MPI
3. OpenCL
4. CUDA: C, C++, Python, Fortran, Matlab