CUDA动态并行性MakeFile

5
这是我第一次使用动态并行性编写程序,但我无法编译代码。我需要在大学的研究项目中运行它,非常感谢任何帮助:
我遇到了以下错误:
/cm/shared/apps/cuda50/toolkit/5.0.35/bin/nvcc -m64 -dc  -gencode arch=compute_35,code=sm_35 -rdc=true -dlink -po maxrregcount=16 -I/cm/shared/apps/cuda50/toolkit/5.0.35 -I. -I.. -I../../common/inc -o BlackScholes.o -c BlackScholes.cu
g++ -m64 -I/cm/shared/apps/cuda50/toolkit/5.0.35 -I. -I.. -I../../common/inc -o BlackScholes_gold.o -c BlackScholes_gold.cpp
g++ -m64 -o BlackScholes BlackScholes.o BlackScholes_gold.o -L/cm/shared/apps/cuda50/toolkit/5.0.35/lib64 -lcudart -lcudadevrt
BlackScholes.o: In function `__sti____cudaRegisterAll_47_tmpxft_000059cb_00000000_6_BlackScholes_cpp1_ii_c58990ec()':
tmpxft_000059cb_00000000-3_BlackScholes.cudafe1.cpp:(.text+0x1354): undefined reference to `__cudaRegisterLinkedBinary_47_tmpxft_000059cb_00000000_6_BlackScholes_cpp1_ii_c58990ec'
collect2: ld returned 1 exit status
make: *** [BlackScholes] Error 1

我有一个cpp文件、一个cu文件和一个cuh文件。下面是我的makefile中重要的部分:

# CUDA code generation flags
#GENCODE_SM10    := -gencode arch=compute_10,code=sm_10
GENCODE_SM20     := -gencode arch=compute_20,code=sm_20
GENCODE_SM30     := -gencode arch=compute_30,code=sm_30 -gencode arch=compute_35,code=sm_35
GENCODE_SM35     := -gencode arch=compute_35,code=sm_35
#GENCODE_FLAGS    := $(GENCODE_SM10) $(GENCODE_SM20) $(GENCODE_SM30)
GENCODE_FLAGS    := $(GENCODE_SM35)

# OS-specific build flags
ifneq ($(DARWIN),)
      LDFLAGS   := -Xlinker -rpath $(CUDA_LIB_PATH) -L$(CUDA_LIB_PATH) -lcudart -lcudadevrt
      CCFLAGS   := -arch $(OS_ARCH)
else
  ifeq ($(OS_SIZE),32)
      LDFLAGS   := -L$(CUDA_LIB_PATH) -lcudart -lcudadevrt
      CCFLAGS   := -m32
  else
      LDFLAGS   := -L$(CUDA_LIB_PATH) -lcudart -lcudadevrt
      CCFLAGS   := -m64
  endif
endif

# OS-architecture specific flags
ifeq ($(OS_SIZE),32)
      NVCCFLAGS := -m32 -dc
else
      NVCCFLAGS := -m64 -dc
endif

# Debug build flags
ifeq ($(dbg),1)
      CCFLAGS   += -g
      NVCCFLAGS += -g -G
      TARGET := debug
else
      TARGET := release
endif


# Common includes and paths for CUDA
INCLUDES      := -I$(CUDA_INC_PATH) -I. -I.. -I../../common/inc

# Additional parameters
MAXRREGCOUNT  :=  -po maxrregcount=16

# Target rules
all: build

build: BlackScholes

BlackScholes.o: BlackScholes.cu
        $(NVCC) $(NVCCFLAGS) $(EXTRA_NVCCFLAGS) $(GENCODE_FLAGS) -rdc=true -dlink $(MAXRREGCOUNT) $(INCLUDES) -o $@ -c $<

BlackScholes_gold.o: BlackScholes_gold.cpp
        $(GCC) $(CCFLAGS) $(INCLUDES) -o $@ -c $<

BlackScholes: BlackScholes.o BlackScholes_gold.o
        $(GCC) $(CCFLAGS) -o $@ $+ $(LDFLAGS) $(EXTRA_LDFLAGS)
        mkdir -p ../../bin/$(OSLOWER)/$(TARGET)
        cp $@ ../../bin/$(OSLOWER)/$(TARGET)
    enter code here

run: build
        ./BlackScholes
1个回答

6
使用主机链接器(g++)进行可执行文件的最终链接时,如果使用可重定位设备代码(nvcc -dc),则需要进行中间设备代码链接步骤。
根据文档
If you want to invoke the device and host linker separately, you can do:

nvcc –arch=sm_20 –dc a.cu b.cu
nvcc –arch=sm_20 –dlink a.o b.o –o link.o
g++ a.o b.o link.o –L<path> -lcudart

由于您在编译行中指定了-dc,因此您正在进行仅编译操作(就像如果您向 g++ 指定了 -c 一样)。

以下是经过修改/压缩的 Makefile,可以帮助您了解所涉及的内容:

GENCODE_SM35     := -gencode arch=compute_35,code=sm_35
GENCODE_FLAGS    := $(GENCODE_SM35)

LDFLAGS   := -L/usr/local/cuda/lib64 -lcudart -lcudadevrt
CCFLAGS   := -m64

NVCCFLAGS := -m64 -dc

NVCC := nvcc
GCC := g++

# Debug build flags
ifeq ($(dbg),1)
      CCFLAGS   += -g
      NVCCFLAGS += -g -G
      TARGET := debug
else
      TARGET := release
endif


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

# Additional parameters
MAXRREGCOUNT  :=  -po maxrregcount=16

# Target rules
all: build

build: BlackScholes

BlackScholes.o: BlackScholes.cu
        $(NVCC) $(NVCCFLAGS) $(EXTRA_NVCCFLAGS) $(GENCODE_FLAGS) $(MAXRREGCOUNT) $(INCLUDES) -o $@ $<
        $(NVCC) -dlink  $(GENCODE_FLAGS) $(MAXRREGCOUNT)  -o bs_link.o $@

BlackScholes_gold.o: BlackScholes_gold.cpp
        $(GCC) $(CCFLAGS) $(INCLUDES) -o $@ -c $<

BlackScholes: BlackScholes.o BlackScholes_gold.o bs_link.o
        $(GCC) $(CCFLAGS) -o $@ $+ $(LDFLAGS) $(EXTRA_LDFLAGS)

run: build
        ./BlackScholes

我需要将“-lcudadevrt”附加到“Command”的字符串中(即“nvcc -lcudadevrt”),之后它就可以工作了。非常感谢您的帮助Robert。 :-) - Kanishk Kanoria
是的,我认为中间步骤也需要添加,这意味着我的 Makefile 可以通过在中间(-dlink)链接步骤中指定 $(LDFLAGS) 进行修改。 - Robert Crovella

网页内容由stack overflow 提供, 点击上面的
可以查看英文原文,
原文链接