问题：
为什么CUDA的示例makefile找不到CUDA库？
王渊
我运行的是Arch Linux，并从存储库中安装了cuda sdk和cuda工具包。我已经编译了/opt/cudasdk/CUDALibraries中的库。
不，我运行make in /opt/cuda-sdk/C编译sdk示例，得到以下错误：
# make
make[1]: Entering directory `/opt/cuda-sdk/C/common'
make[1]: Leaving directory `/opt/cuda-sdk/C/common'
make[1]: Entering directory `/opt/cuda-sdk/C/common'
make[1]: Leaving directory `/opt/cuda-sdk/C/common'
make[1]: Entering directory `/opt/cuda-sdk/C/common'
make[1]: Leaving directory `/opt/cuda-sdk/C/common'
make[1]: Entering directory `/opt/cuda-sdk/shared'
make[1]: Leaving directory `/opt/cuda-sdk/shared'
make[1]: Entering directory `/opt/cuda-sdk/C/src/newdelete'
make[1]: Leaving directory `/opt/cuda-sdk/C/src/newdelete'
make[1]: Entering directory `/opt/cuda-sdk/C/src/simpleTextureDrv'
/usr/bin/ld: cannot find -lcuda
collect2: error: ld returned 1 exit status
make[1]: *** [../../bin/linux/release/simpleTextureDrv] Error 1
make[1]: Leaving directory `/opt/cuda-sdk/C/src/simpleTextureDrv'
make: *** [src/simpleTextureDrv/Makefile.ph_build] Error 2
Makefile本身似乎只包含文件/opt/cudasdk/C/common/common。mk，即：
################################################################################
#
# Copyright 1993-2011 NVIDIA Corporation. All rights reserved.
#
# NVIDIA Corporation and its licensors retain all intellectual property and
# proprietary rights in and to this software and related documentation.
# Any use, reproduction, disclosure, or distribution of this software
# and related documentation without an express license agreement from
# NVIDIA Corporation is strictly prohibited.
#
# Please refer to the applicable NVIDIA end user license agreement (EULA)
# associated with this source code for terms and conditions that govern
# your use of this NVIDIA software.
#
################################################################################
#
# Common build script for CUDA source projects for Linux and Mac platforms
#
################################################################################
.SUFFIXES : .cu .cu_dbg.o .c_dbg.o .cpp_dbg.o .cu_rel.o .c_rel.o .cpp_rel.o .cubin .ptx
# Add new SM Versions here as devices with new Compute Capability are released
SM_VERSIONS := 10 11 12 13 20 21 30
CUDA_INSTALL_PATH ?= /opt/cuda-toolkit
ifdef cuda-install
CUDA_INSTALL_PATH := $(cuda-install)
endif
# detect OS
OSUPPER = $(shell uname -s 2>/dev/null | tr [:lower:] [:upper:])
OSLOWER = $(shell uname -s 2>/dev/null | tr [:upper:] [:lower:])
# 'linux' is output for Linux system, 'darwin' for OS X
DARWIN = $(strip $(findstring DARWIN, $(OSUPPER)))
ifneq ($(DARWIN),)
SNOWLEOPARD = $(strip $(findstring 10.6, $(shell egrep "<string>10\.6" /System/Library/CoreServices/SystemVersion.plist)))
LION = $(strip $(findstring 10.7, $(shell egrep "<string>10\.7" /System/Library/CoreServices/SystemVersion.plist)))
endif
# detect 32-bit or 64-bit platform
HP_64 = $(shell uname -m | grep 64)
OSARCH= $(shell uname -m)
# Basic directory setup for SDK
# (override directories only if they are not already defined)
SRCDIR ?=
ROOTDIR ?= ..
ROOTBINDIR ?= $(ROOTDIR)/../bin
BINDIR ?= $(ROOTBINDIR)/$(OSLOWER)
ROOTOBJDIR ?= obj
LIBDIR := $(ROOTDIR)/../lib
COMMONDIR := $(ROOTDIR)/../common
SHAREDDIR := $(ROOTDIR)/../../shared/
# Compilers
NVCC := $(CUDA_INSTALL_PATH)/bin/nvcc
CXX := g++ -fPIC
CC := gcc -fPIC
LINK := g++ -fPIC
# Includes
INCLUDES += -I. -I$(CUDA_INSTALL_PATH)/include -I$(COMMONDIR)/inc -I$(SHAREDDIR)/inc
# Warning flags
CXXWARN_FLAGS := \
-W -Wall \
-Wimplicit \
-Wswitch \
-Wformat \
-Wchar-subscripts \
-Wparentheses \
-Wmultichar \
-Wtrigraphs \
-Wpointer-arith \
-Wcast-align \
-Wreturn-type \
-Wno-unused-function \
$(SPACE)
CWARN_FLAGS := $(CXXWARN_FLAGS) \
-Wstrict-prototypes \
-Wmissing-prototypes \
-Wmissing-declarations \
-Wnested-externs \
-Wmain \
# architecture flag for nvcc and gcc compilers build
CUBIN_ARCH_FLAG :=
CXX_ARCH_FLAGS :=
NVCCFLAGS :=
LIB_ARCH := $(OSARCH)
# Determining the necessary Cross-Compilation Flags
# 32-bit OS, but we target 64-bit cross compilation
ifeq ($(x86_64),1)
NVCCFLAGS += -m64
LIB_ARCH = x86_64
ifneq ($(DARWIN),)
CXX_ARCH_FLAGS += -arch x86_64
else
CXX_ARCH_FLAGS += -m64
endif
else
# 64-bit OS, and we target 32-bit cross compilation
ifeq ($(i386),1)
NVCCFLAGS += -m32
LIB_ARCH = i386
ifneq ($(DARWIN),)
CXX_ARCH_FLAGS += -arch i386
else
CXX_ARCH_FLAGS += -m32
endif
else
ifeq "$(strip $(HP_64))" ""
LIB_ARCH = i386
NVCCFLAGS += -m32
ifneq ($(DARWIN),)
CXX_ARCH_FLAGS += -arch i386
else
CXX_ARCH_FLAGS += -m32
endif
else
LIB_ARCH = x86_64
NVCCFLAGS += -m64
ifneq ($(DARWIN),)
CXX_ARCH_FLAGS += -arch x86_64
else
CXX_ARCH_FLAGS += -m64
endif
endif
endif
endif
# Compiler-specific flags (by default, we always use sm_10, sm_20, and sm_30), unless we use the SMVERSION template
GENCODE_SM10 := -gencode=arch=compute_10,code=\"sm_10,compute_10\"
GENCODE_SM20 := -gencode=arch=compute_20,code=\"sm_20,compute_20\"
GENCODE_SM30 := -gencode=arch=compute_30,code=\"sm_30,compute_30\"
CXXFLAGS += $(CXXWARN_FLAGS) $(CXX_ARCH_FLAGS)
CFLAGS += $(CWARN_FLAGS) $(CXX_ARCH_FLAGS)
LINKFLAGS +=
LINK += $(LINKFLAGS) $(CXX_ARCH_FLAGS)
# This option for Mac allows CUDA applications to work without requiring to set DYLD_LIBRARY_PATH
ifneq ($(DARWIN),)
LINK += -Xlinker -rpath $(CUDA_INSTALL_PATH)/lib
endif
# Common flags
COMMONFLAGS += $(INCLUDES) -DUNIX
# If we are enabling GPU based debugging, then we want to use -G, warning that this
# May have a significant impact on GPU device code, since optimizations are turned off
ifeq ($(gpudbg),1)
NVCCFLAGS += -G
dbg = $(gpudbg)
endif
# Debug/release configuration
ifeq ($(dbg),1)
COMMONFLAGS += -g
NVCCFLAGS += -D_DEBUG
CXXFLAGS += -D_DEBUG
CFLAGS += -D_DEBUG
BINSUBDIR := debug
LIBSUFFIX := D
else
COMMONFLAGS += -O2
BINSUBDIR := release
LIBSUFFIX :=
NVCCFLAGS += --compiler-options -fno-strict-aliasing
CXXFLAGS += -fno-strict-aliasing
CFLAGS += -fno-strict-aliasing
endif
# architecture flag for cubin build
CUBIN_ARCH_FLAG :=
# OpenGL is used or not (if it is used, then it is necessary to include GLEW)
ifeq ($(USEGLLIB),1)
ifneq ($(DARWIN),)
OPENGLLIB := -L/System/Library/Frameworks/OpenGL.framework/Libraries
OPENGLLIB += -lGL -lGLU $(COMMONDIR)/lib/$(OSLOWER)/libGLEW.a
else
# this case for linux platforms
OPENGLLIB := -lGL -lGLU -lX11 -lXi -lXmu
# check if x86_64 flag has been set, otherwise, check HP_64 is i386/x86_64
ifeq ($(x86_64),1)
OPENGLLIB += -lGLEW_x86_64 -L/usr/X11R6/lib64
else
ifeq ($(i386),)
ifeq "$(strip $(HP_64))" ""
OPENGLLIB += -lGLEW -L/usr/X11R6/lib
else
OPENGLLIB += -lGLEW_x86_64 -L/usr/X11R6/lib64
endif
endif
endif
# check if i386 flag has been set, otehrwise check HP_64 is i386/x86_64
ifeq ($(i386),1)
OPENGLLIB += -lGLEW -L/usr/X11R6/lib
else
ifeq ($(x86_64),)
ifeq "$(strip $(HP_64))" ""
OPENGLLIB += -lGLEW -L/usr/X11R6/lib
else
OPENGLLIB += -lGLEW_x86_64 -L/usr/X11R6/lib64
endif
endif
endif
endif
endif
ifeq ($(USEGLUT),1)
ifneq ($(DARWIN),)
OPENGLLIB += -framework GLUT
else
ifeq ($(x86_64),1)
OPENGLLIB += -lglut -L/usr/lib64
endif
ifeq ($(i386),1)
OPENGLLIB += -lglut -L/usr/lib
endif
ifeq ($(x86_64),)
ifeq ($(i386),)
OPENGLLIB += -lglut
endif
endif
endif
endif
ifeq ($(USEPARAMGL),1)
PARAMGLLIB := -lparamgl_$(LIB_ARCH)$(LIBSUFFIX)
endif
ifeq ($(USERENDERCHECKGL),1)
RENDERCHECKGLLIB := -lrendercheckgl_$(LIB_ARCH)$(LIBSUFFIX)
endif
ifeq ($(USENVCUVID), 1)
ifneq ($(DARWIN),)
NVCUVIDLIB := -L../../common/lib/darwin -lnvcuvid
endif
endif
# Libs
ifneq ($(DARWIN),)
LIB := -L$(CUDA_INSTALL_PATH)/lib -L$(LIBDIR) -L$(COMMONDIR)/lib/$(OSLOWER) -L$(SHAREDDIR)/lib $(NVCUVIDLIB)
else
ifeq "$(strip $(HP_64))" ""
ifeq ($(x86_64),1)
LIB := -L$(CUDA_INSTALL_PATH)/lib64 -L$(LIBDIR) -L$(COMMONDIR)/lib/$(OSLOWER) -L$(SHAREDDIR)/lib
else
LIB := -L$(CUDA_INSTALL_PATH)/lib -L$(LIBDIR) -L$(COMMONDIR)/lib/$(OSLOWER) -L$(SHAREDDIR)/lib
endif
else
ifeq ($(i386),1)
LIB := -L$(CUDA_INSTALL_PATH)/lib -L$(LIBDIR) -L$(COMMONDIR)/lib/$(OSLOWER) -L$(SHAREDDIR)/lib
else
LIB := -L$(CUDA_INSTALL_PATH)/lib64 -L$(LIBDIR) -L$(COMMONDIR)/lib/$(OSLOWER) -L$(SHAREDDIR)/lib
endif
endif
endif
# If dynamically linking to CUDA and CUDART, we exclude the libraries from the LIB
ifeq ($(USECUDADYNLIB),1)
LIB += ${OPENGLLIB} $(PARAMGLLIB) $(RENDERCHECKGLLIB) ${LIB} -ldl -rdynamic
else
# static linking, we will statically link against CUDA and CUDART
ifeq ($(USEDRVAPI),1)
LIB += -lcuda ${OPENGLLIB} $(PARAMGLLIB) $(RENDERCHECKGLLIB) ${LIB}
else
ifeq ($(emu),1)
LIB += -lcudartemu
else
LIB += -lcudart
endif
LIB += ${OPENGLLIB} $(PARAMGLLIB) $(RENDERCHECKGLLIB) ${LIB}
endif
endif
ifeq ($(USECUFFT),1)
ifeq ($(emu),1)
LIB += -lcufftemu
else
LIB += -lcufft
endif
endif
ifeq ($(USECUBLAS),1)
ifeq ($(emu),1)
LIB += -lcublasemu
else
LIB += -lcublas
endif
endif
ifeq ($(USECURAND),1)
LIB += -lcurand
endif
ifeq ($(USECUSPARSE),1)
LIB += -lcusparse
endif
# Lib/exe configuration
# Lib/exe configuration
# Lib/exe configuration
ifneq ($(STATIC_LIB),)
TARGETDIR := $(LIBDIR)
TARGET := $(subst .a,_$(LIB_ARCH)$(LIBSUFFIX).a,$(LIBDIR)/$(STATIC_LIB))
LINKLINE = ar rucv $(TARGET) $(OBJS)
else
ifneq ($(OMIT_CUTIL_LIB),1)
LIB += -lcutil_$(LIB_ARCH)$(LIBSUFFIX)
endif
ifneq ($(OMIT_SHRUTIL_LIB),1)
LIB += -lshrutil_$(LIB_ARCH)$(LIBSUFFIX)
endif
# Device emulation configuration
ifeq ($(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) $(LIB)
endif
# check if verbose
ifeq ($(verbose), 1)
VERBOSE :=
else
VERBOSE := @
endif
################################################################################
# Check for input flags and set compiler flags appropriately
################################################################################
ifeq ($(fastmath), 1)
NVCCFLAGS += -use_fast_math
endif
ifeq ($(keep), 1)
NVCCFLAGS += -keep
NVCC_KEEP_CLEAN := *.i* *.cubin *.cu.c *.cudafe* *.fatbin.c *.ptx
endif
ifdef maxregisters
NVCCFLAGS += -maxrregcount $(maxregisters)
endif
ifeq ($(ptxas), 1)
NVCCFLAGS += --ptxas-options=-v
endif
# Add cudacc flags
NVCCFLAGS += $(CUDACCFLAGS)
# Add common flags
NVCCFLAGS += $(COMMONFLAGS)
CXXFLAGS += $(COMMONFLAGS)
CFLAGS += $(COMMONFLAGS)
ifeq ($(nvcc_warn_verbose),1)
NVCCFLAGS += $(addprefix --compiler-options ,$(CXXWARN_FLAGS))
NVCCFLAGS += --compiler-options -fno-strict-aliasing
endif
################################################################################
# Set up object files
################################################################################
OBJDIR := $(ROOTOBJDIR)/$(LIB_ARCH)/$(BINSUBDIR)
OBJS += $(patsubst %.cpp,$(OBJDIR)/%.cpp.o,$(notdir $(CCFILES)))
OBJS += $(patsubst %.c,$(OBJDIR)/%.c.o,$(notdir $(CFILES)))
OBJS += $(patsubst %.cu,$(OBJDIR)/%.cu.o,$(notdir $(CUFILES)))
################################################################################
# Set up cubin output files
################################################################################
CUBINDIR := $(SRCDIR)data
CUBINS += $(patsubst %.cu,$(CUBINDIR)/%.cubin,$(notdir $(CUBINFILES)))
################################################################################
# Set up PTX output files
################################################################################
PTXDIR := $(SRCDIR)data
PTXBINS += $(patsubst %.cu,$(PTXDIR)/%.ptx,$(notdir $(PTXFILES)))
################################################################################
# Rules
################################################################################
$(OBJDIR)/%.c.o : $(SRCDIR)%.c $(C_DEPS)
$(VERBOSE)$(CC) $(CFLAGS) -o $@ -c $<
$(OBJDIR)/%.cpp.o : $(SRCDIR)%.cpp $(C_DEPS)
$(VERBOSE)$(CXX) $(CXXFLAGS) -o $@ -c $<
# Default arch includes gencode for sm_10, sm_20, sm_30, and other archs from GENCODE_ARCH declared in the makefile
$(OBJDIR)/%.cu.o : $(SRCDIR)%.cu $(CU_DEPS)
$(VERBOSE)$(NVCC) $(GENCODE_SM10) $(GENCODE_ARCH) $(GENCODE_SM20) $(GENCODE_SM30) $(NVCCFLAGS) $(SMVERSIONFLAGS) -o $@ -c $<
# Default arch includes gencode for sm_10, sm_20, sm_30, and other archs from GENCODE_ARCH declared in the makefile
$(CUBINDIR)/%.cubin : $(SRCDIR)%.cu cubindirectory
$(VERBOSE)$(NVCC) $(GENCODE_SM10) $(GENCODE_ARCH) $(GENCODE_SM20) $(GENCODE_SM30) $(CUBIN_ARCH_FLAG) $(NVCCFLAGS) $(SMVERSIONFLAGS) -o $@ -cubin $<
$(PTXDIR)/%.ptx : $(SRCDIR)%.cu ptxdirectory
$(VERBOSE)$(NVCC) $(CUBIN_ARCH_FLAG) $(NVCCFLAGS) $(SMVERSIONFLAGS) -o $@ -ptx $<
#
# The following definition is a template that gets instantiated for each SM
# version (sm_10, sm_13, etc.) stored in SMVERSIONS. It does 2 things:
# 1. It adds to OBJS a .cu_sm_XX.o for each .cu file it finds in CUFILES_sm_XX.
# 2. It generates a rule for building .cu_sm_XX.o files from the corresponding
# .cu file.
#
# The intended use for this is to allow Makefiles that use common.mk to compile
# files to different Compute Capability targets (aka SM arch version). To do
# so, in the Makefile, list files for each SM arch separately, like so:
# This will be used over the default rule abov
#
# CUFILES_sm_10 := mycudakernel_sm10.cu app.cu
# CUFILES_sm_12 := anothercudakernel_sm12.cu
#
define SMVERSION_template
#OBJS += $(patsubst %.cu,$(OBJDIR)/%.cu_$(1).o,$(notdir $(CUFILES_$(1))))
OBJS += $(patsubst %.cu,$(OBJDIR)/%.cu_$(1).o,$(notdir $(CUFILES_sm_$(1))))
$(OBJDIR)/%.cu_$(1).o : $(SRCDIR)%.cu $(CU_DEPS)
# $(VERBOSE)$(NVCC) -o $$@ -c $$< $(NVCCFLAGS) $(1)
$(VERBOSE)$(NVCC) -gencode=arch=compute_$(1),code=\"sm_$(1),compute_$(1)\" $(GENCODE_SM20) $(GENCODE_SM30) -o $$@ -c $$< $(NVCCFLAGS)
endef
# This line invokes the above template for each arch version stored in
# SM_VERSIONS. The call function invokes the template, and the eval
# function interprets it as make commands.
$(foreach smver,$(SM_VERSIONS),$(eval $(call SMVERSION_template,$(smver))))
$(TARGET): makedirectories $(OBJS) $(CUBINS) $(PTXBINS) Makefile
$(VERBOSE)$(LINKLINE)
cubindirectory:
$(VERBOSE)mkdir -p $(CUBINDIR)
ptxdirectory:
$(VERBOSE)mkdir -p $(PTXDIR)
makedirectories:
$(VERBOSE)mkdir -p $(LIBDIR)
$(VERBOSE)mkdir -p $(OBJDIR)
$(VERBOSE)mkdir -p $(TARGETDIR)
tidy :
$(VERBOSE)find . | egrep "#" | xargs rm -f
$(VERBOSE)find . | egrep "\~" | xargs rm -f
clean : tidy
$(VERBOSE)rm -f *.stub.c *.gpu *.cu.cpp *.i *.ii
$(VERBOSE)rm -f *.cubin *.ptx *.fatbin.c *.hash
$(VERBOSE)rm -f *.cudafe1.c *.cudafe2.c *.cudafe1.cpp *.cudafe2.cpp
$(VERBOSE)rm -f $(OBJS)
$(VERBOSE)rm -f $(CUBINS)
$(VERBOSE)rm -f $(PTXBINS)
$(VERBOSE)rm -f $(TARGET)
$(VERBOSE)rm -f $(NVCC_KEEP_CLEAN)
$(VERBOSE)rm -f $(ROOTBINDIR)/$(OSLOWER)/$(BINSUBDIR)/*.ppm
$(VERBOSE)rm -f $(ROOTBINDIR)/$(OSLOWER)/$(BINSUBDIR)/*.pgm
$(VERBOSE)rm -f $(ROOTBINDIR)/$(OSLOWER)/$(BINSUBDIR)/*.bin
$(VERBOSE)rm -f $(ROOTBINDIR)/$(OSLOWER)/$(BINSUBDIR)/*.bmp
$(VERBOSE)rm -f $(ROOTBINDIR)/$(OSLOWER)/$(BINSUBDIR)/*.txt
$(VERBOSE)rm -f $(CUBINDIR)/*.cubin $(PTXDIR)/*.ptx
$(VERBOSE)rm -rf $(ROOTOBJDIR)
$(VERBOSE)rm -rf $(LIBDIR)
$(VERBOSE)rm -rf $(OBJDIR)
$(VERBOSE)rm -rf $(TARGETDIR)
clobber : clean
$(VERBOSE)rm -rf $(COMMONDIR)/lib/*.a
$(VERBOSE)rm -rf $(SHAREDDIR)/lib/*.a
$(VERBOSE)rm -rf $(COMMONDIR)/obj
$(VERBOSE)rm -rf $(SHAREDDIR)/obj
Echo$LD_LIBRARY_PATH的输出为空。
我已经检查了libcuda，可以在我认为应该的地方找到它：
# ls /usr/lib | grep libcuda
libcuda.so.1
libcuda.so.304.32
我是否需要做一些明显的事情来编译这些文件？这是CUDA makefile的问题还是我的系统的问题？
共有1个答案
令狐宜民
类似资料：
-
CUDA 工具包是一种针对支持CUDA功能的GPU（图形处理器）的C语言开发环境。CUDA 开发环境包括: nvcc C语言编译器 适用于GPU（图形处理器）的CUDA FFT和BLAS库 分析器 适用于GPU（图形处理器）的gdb调试器（在2008年3月推出alpha版） CUDA运行时（CUDA runtime）驱动程序（目前在标准的NVIDIA GPU驱动中也提供） CUDA编程手册 CUD
-
问题内容： 当我使用固定内存进行CUDA数据传输时，我发现数据传输速度大大提高。在linux上，实现此目标的底层系统调用是mlock。从mlock的手册页中可以看出，锁定该页可防止将其换出： mlock（）将页面锁定在地址范围内，该地址范围从addr开始并持续len个字节。当调用成功返回时，保证所有包含指定地址范围一部分的页面都驻留在RAM中； 在测试中，我的系统上有几千个可用内存，因此从不存在内
-
我有一台配置了NVIDIA GeForce1080 GTX和CentOS 7作为操作系统的Intel Xeon机器。我已经安装了NVIDIA驱动程序410.93和cuda toolkit 10.0。编译完cuda示例后，我尝试运行/设备查询。但它是这样扔的 一些命令输出 lspci | grep VGA nvidia-smi nvcc版本 路径 LSAMD|grep nvidia lsmod|gr
-
我正在使用Visual Studio2013进行CUDA(7.0)项目。该项目为64bit。我正在使用驱动程序API，需要从ptx或cubin文件加载模块。但我找不到文件。在VS I go properties->cuda C/C++->common->nvcc编译类型中，将其更改为-cubin或ptx。编译器完成ok但我找不到那个文件。我只能在output debug目录中看到kernel.cu
-
译者：片刻 torch.cuda 用于设置和运行 CUDA 操作。它会跟踪当前选定的GPU，并且默认情况下会在该设备上创建您分配的所有 CUDA tensors。可以使用 torch.cuda.device 上下文管理器更改所选设备。 但是，一旦分配了 tensor，就可以对其进行操作而不管所选择的设备如何，结果将始终与 tensor 放在同一设备上。 默认情况下不允许跨 GPU 操作，除了 co
-
torch.cuda会记录当前选择的GPU，并且分配的所有CUDA张量将在上面创建。可以使用torch.cuda.device上下文管理器更改所选设备。 但是，一旦张量被分配，您可以直接对其进行操作，而不考虑所选择的设备，结果将始终放在与张量相同的设备上。 默认情况下，不支持跨GPU操作，唯一的例外是copy_()。 除非启用对等存储器访问，否则对分布不同设备上的张量任何启动操作的尝试都将会引发错