3

已解决：请参阅下面的评论

我一直在研究 GPU 分形生成器，并且一直在使用 CUDA 来访问 GPU，并使用 libpng 来生成输出图像。我正在运行 OSX 10.7.4、Cuda 版本 4.2、V0.2.1221、libpng15（我认为是 1.5.11）

代码的第一个版本（非常混乱和hacky）只有一个源文件GPUkernel.cu，它具有使用CUDA库计算分形的函数，并且还使用libpng生成png。这用命令编译得很好

all:    src/GPUkernel.cu
        nvcc -o base src/GPUkernel.cu -lpng15 -lz

它产生了可执行的基础，一切运行良好。

新版本：

现在，随着这个项目变得越来越大，为了尝试使代码更易于管理，我将源代码拆分为两个文件，GPUkernel.cu 和 GPUfractal.cpp。GPUkernel.cu 包含所有 CUDA GPU 函数，GPUfractal.cpp 包含 main() 和 libpng 函数。

当我尝试编译此代码时遇到一些问题，与 libpng 和 CUDA 使用的不同架构有关（我认为）。我的makefile是：

CUDA_INSTALL_PATH ?= /Developer/GPU\ Computing/CUDALibraries/common

ARCH=-arch i386 -arch x86_64

CXX := g++
CC := gcc
LINK := g++ -fPIC $(ARCH)
NVCC := nvcc -ccbin /usr/bin

# Includes
INCLUDES = -I. -I$(CUDA_INSTALL_PATH)/inc -I/usr/local/cuda/include 

# Common flags
COMMONFLAGS += $(INCLUDES)
NVCCFLAGS += $(COMMONFLAGS)
CXXFLAGS += $(COMMONFLAGS)
CFLAGS += $(COMMONFLAGS)

LIB_CUDA := -L$(CUDA_INSTALL_PATH)/lib `libpng-config --cflags --ldflags`

OBJS = GPUkernel.cu.o GPUfractal.cpp.o
TARGET = base
LINKLINE = $(LINK) -o $(TARGET) $(OBJS) $(LIB_CUDA)

.SUFFIXES: .c .cpp .cu .o

%.c.o: %.c
    $(CC) $(CFLAGS) -c $< -o $@

%.cu.o: %.cu
    $(NVCC) $(NVCCFLAGS) -c $< -o $@

%.cpp.o: %.cpp
    $(CXX) $(CXXFLAGS) -c $< -o $@

$(TARGET): $(OBJS) Makefile
    $(LINKLINE)

这是我的终端输出：

pikachu:GPUfractal tom$ make
nvcc -ccbin /usr/bin -I. -I/Developer/GPU\ Computing/CUDALibraries/common/inc -I/usr/local/cuda/include  -c GPUkernel.cu -o GPUkernel.cu.o
ptxas /tmp/tmpxft_00000847_00000000-2_GPUkernel.ptx, line 105; warning : Double is not supported. Demoting to float
g++ -I. -I/Developer/GPU\ Computing/CUDALibraries/common/inc -I/usr/local/cuda/include  -c GPUfractal.cpp -o GPUfractal.cpp.o
GPUfractal.cpp: In function ‘int renderImage(int*, float*, int, int, std::string, bool, int)’:
GPUfractal.cpp:180: warning: deprecated conversion from string constant to ‘char*’
GPUfractal.cpp:232: warning: deprecated conversion from string constant to ‘char*’
g++ -fPIC -arch i386 -arch x86_64 -o base GPUkernel.cu.o GPUfractal.cpp.o -L/Developer/GPU\ Computing/CUDALibraries/common/lib `libpng-config --cflags --ldflags`
ld: warning: ignoring file GPUfractal.cpp.o, file was built for unsupported file format which is not the architecture being linked (i386)
Undefined symbols for architecture i386:
  "_main", referenced from:
      start in crt1.10.6.o
  "_cudaGetLastError", referenced from:
      calculateFractal(int*, float*, int, int, float, float, float, float, int, bool, float, float)in GPUkernel.cu.o
  "_cudaGetErrorString", referenced from:
      calculateFractal(int*, float*, int, int, float, float, float, float, int, bool, float, float)in GPUkernel.cu.o
  "_cudaMemset", referenced from:
      calculateFractal(int*, float*, int, int, float, float, float, float, int, bool, float, float)in GPUkernel.cu.o
  "_cudaConfigureCall", referenced from:
      calculateFractal(int*, float*, int, int, float, float, float, float, int, bool, float, float)in GPUkernel.cu.o
  "_cudaMemcpy", referenced from:
      calculateFractal(int*, float*, int, int, float, float, float, float, int, bool, float, float)in GPUkernel.cu.o
  "_cudaFree", referenced from:
      calculateFractal(int*, float*, int, int, float, float, float, float, int, bool, float, float)in GPUkernel.cu.o
  "___cudaRegisterFatBinary", referenced from:
      __sti____cudaRegisterAll_44_tmpxft_00000847_00000000_4_GPUkernel_cpp1_ii_ad246568()      in GPUkernel.cu.o
  "___cudaRegisterFunction", referenced from:
      __sti____cudaRegisterAll_44_tmpxft_00000847_00000000_4_GPUkernel_cpp1_ii_ad246568()      in GPUkernel.cu.o
  "___cudaUnregisterFatBinary", referenced from:
      __cudaUnregisterBinaryUtil()      in GPUkernel.cu.o
  "_cudaLaunch", referenced from:
      cudaError cudaLaunch<char>(char*)in GPUkernel.cu.o
  "_cudaSetupArgument", referenced from:
      __device_stub__Z19calculateMandlebrotPiPfffffibff(int*, float*, float, float, float, float, int, bool, float, float)in GPUkernel.cu.o
  "_cudaMalloc", referenced from:
      cudaError cudaMalloc<int>(int**, unsigned long)in GPUkernel.cu.o
      cudaError cudaMalloc<float>(float**, unsigned long)in GPUkernel.cu.o
ld: symbol(s) not found for architecture i386
collect2: ld returned 1 exit status
ld: warning: ignoring file GPUkernel.cu.o, file was built for i386 which is not the architecture being linked (x86_64)
ld: warning: ignoring file /usr/local/lib/libpng15.a, file was built for archive which is not the architecture being linked (x86_64)
Undefined symbols for architecture x86_64:
  "_png_create_write_struct", referenced from:
      renderImage(int*, float*, int, int, std::basic_string<char, std::char_traits<char>, std::allocator<char> >, bool, int)in GPUfractal.cpp.o
  "_png_create_info_struct", referenced from:
      renderImage(int*, float*, int, int, std::basic_string<char, std::char_traits<char>, std::allocator<char> >, bool, int)in GPUfractal.cpp.o
  "_png_set_longjmp_fn", referenced from:
      renderImage(int*, float*, int, int, std::basic_string<char, std::char_traits<char>, std::allocator<char> >, bool, int)in GPUfractal.cpp.o
  "_png_init_io", referenced from:
      renderImage(int*, float*, int, int, std::basic_string<char, std::char_traits<char>, std::allocator<char> >, bool, int)in GPUfractal.cpp.o
  "_png_set_IHDR", referenced from:
      renderImage(int*, float*, int, int, std::basic_string<char, std::char_traits<char>, std::allocator<char> >, bool, int)in GPUfractal.cpp.o
  "_png_set_text", referenced from:
      renderImage(int*, float*, int, int, std::basic_string<char, std::char_traits<char>, std::allocator<char> >, bool, int)in GPUfractal.cpp.o
  "_png_write_info", referenced from:
      renderImage(int*, float*, int, int, std::basic_string<char, std::char_traits<char>, std::allocator<char> >, bool, int)in GPUfractal.cpp.o
  "_png_write_row", referenced from:
      renderImage(int*, float*, int, int, std::basic_string<char, std::char_traits<char>, std::allocator<char> >, bool, int)in GPUfractal.cpp.o
  "_png_write_end", referenced from:
      renderImage(int*, float*, int, int, std::basic_string<char, std::char_traits<char>, std::allocator<char> >, bool, int)in GPUfractal.cpp.o
  "_png_free_data", referenced from:
      renderImage(int*, float*, int, int, std::basic_string<char, std::char_traits<char>, std::allocator<char> >, bool, int)in GPUfractal.cpp.o
  "_png_destroy_write_struct", referenced from:
      renderImage(int*, float*, int, int, std::basic_string<char, std::char_traits<char>, std::allocator<char> >, bool, int)in GPUfractal.cpp.o
ld: symbol(s) not found for architecture x86_64
collect2: ld returned 1 exit status
lipo: can't open input file: /var/folders/rw/344w_6js7d7dhg8pch7zws180000gn/T//ccZYSoRL.out (No such file or directory)
make: *** [base] Error 1

我已将最新版本的代码上传到github，您还可以在此处的存档文件夹中查看旧（工作）版本的代码（由于 noob 用户限制，链接已删除，但您可以在 github 上的“存档”中找到它/0.1'）。

更多信息：我在代码的第一个版本中遇到了与未定义符号类似的问题，但是通过大量谷歌搜索，发现这篇文章通过添加 libpng-config --cflags --ldflags 到 makefile解决了我的问题

感谢您阅读本文，希望第一篇文章不会太糟糕:)

感谢您的帮助！

4

2 回答 2

1

您似乎正在为 32 位和 64 位构建主机代码。在 OS X 上，如果您想将主机代码 (.cpp) 编译为 64 位，则需要-arch x86_64，或者如果您想要 32 位，则需要-arch i386，但我认为您不应同时指定两者。

对于您的设备代码，如果您想要 64 位，那么您需要添加-m64到 nvcc 命令行。对于 32 位，使用-m32.

您还需要链接到适当的 CUDA 库（32 位或 64 位）。如果您查看 CUDA 安装路径，您会看到有 32 位和 64 位目录以及相应的库。仅包含这些链接路径之一（使用-L），具体取决于您构建的是 32 位还是 64 位。如果您正在使用 CUDA 运行时 API（可能），那么您还必须链接 libcuda ( -lcuda) 和 libcudart ( -lcudart)。

您还需要对您链接的任何其他库（libpng 或其他）执行相同的操作。

于 2012-09-06T03:13:57.777 回答
0

我相信您需要链接 cuda 库和 cuda 运行时库。链接器 (g++) 不了解 CUDA，因此您需要告诉它与 cuda 库链接。

我有一个混合的 C++ 和 CUDA 项目，我使用 g++ 编译 C++ 文件和 nvcc 编译我的 CUDA 文件。然后我调用

g++ -o exec_name *.o -lcuda -lcudart

于 2012-09-06T03:14:43.137 回答