手动编译Parboil

写在最前

这里是使用Parboil自带的脚本编译和使用的教程：https://blog.csdn.net/FishSeeker/article/details/79479714

最近因为跑实验的原因需要修改以及合并部分代码然后再编译，所以无法使用Parboil自带的编译功能，只能自己编译。

编译脚本

下面是cuda通用的编译脚本，大部分东西都是固定的不需要改。详情见注释

# Location of the CUDA Toolkit，默认的路径即可
CUDA_PATH ?= "/usr/local/cuda"


# architecture
HOST_ARCH   := $(shell uname -m)
TARGET_ARCH ?= $(HOST_ARCH)
ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 ppc64le armv7l))
    ifneq ($(TARGET_ARCH),$(HOST_ARCH))
        ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 ppc64le))
            TARGET_SIZE := 64
        else ifneq (,$(filter $(TARGET_ARCH),armv7l))
            TARGET_SIZE := 32
        endif
    else
        TARGET_SIZE := $(shell getconf LONG_BIT)
    endif
else
    $(error ERROR - unsupported value $(TARGET_ARCH) for TARGET_ARCH!)
endif
ifneq ($(TARGET_ARCH),$(HOST_ARCH))
    ifeq (,$(filter $(HOST_ARCH)-$(TARGET_ARCH),aarch64-armv7l x86_64-armv7l x86_64-aarch64 x86_64-ppc64le))
        $(error ERROR - cross compiling from $(HOST_ARCH) to $(TARGET_ARCH) is not supported!)
    endif
endif

# When on native aarch64 system with userspace of 32-bit, change TARGET_ARCH to armv7l
ifeq ($(HOST_ARCH)-$(TARGET_ARCH)-$(TARGET_SIZE),aarch64-aarch64-32)
    TARGET_ARCH = armv7l
endif

# operating system
HOST_OS   := $(shell uname -s 2>/dev/null | tr "[:upper:]" "[:lower:]")
TARGET_OS ?= $(HOST_OS)
ifeq (,$(filter $(TARGET_OS),linux darwin qnx android))
    $(error ERROR - unsupported value $(TARGET_OS) for TARGET_OS!)
endif

HOST_COMPILER ?= g++
NVCC          := $(CUDA_PATH)/bin/nvcc -ccbin $(HOST_COMPILER)

# internal flags
NVCCFLAGS   := -m${TARGET_SIZE}
CCFLAGS     :=
LDFLAGS     :=


ifneq ($(TARGET_ARCH),$(HOST_ARCH))
    ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-linux)
        ifneq ($(TARGET_FS),)
            GCCVERSIONLTEQ46 := $(shell expr `$(HOST_COMPILER) -dumpversion` \<= 4.6)
            ifeq ($(GCCVERSIONLTEQ46),1)
                CCFLAGS += --sysroot=$(TARGET_FS)
            endif
            LDFLAGS += --sysroot=$(TARGET_FS)
            LDFLAGS += -rpath-link=$(TARGET_FS)/lib
            LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib
            LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib/arm-linux-gnueabihf
        endif
    endif
endif

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

#　这里添加编译参数，比如-keep， -Xptxas 等等
ALL_CCFLAGS := 
ALL_CCFLAGS += $(NVCCFLAGS)
ALL_CCFLAGS += $(EXTRA_NVCCFLAGS)
ALL_CCFLAGS += $(addprefix -Xcompiler ,$(CCFLAGS))
ALL_CCFLAGS += $(addprefix -Xcompiler ,$(EXTRA_CCFLAGS))

SAMPLE_ENABLED := 1

ALL_LDFLAGS :=
ALL_LDFLAGS += $(ALL_CCFLAGS)
ALL_LDFLAGS += $(addprefix -Xlinker ,$(LDFLAGS))
ALL_LDFLAGS += $(addprefix -Xlinker ,$(EXTRA_LDFLAGS))

# 这里添加include库
INCLUDES  := -I../../common/inc
INCLUDES  += -I/home/fish/GPU/benchmark/Parboil/parboil/common/include
LIBRARIES :=

################################################################################

# 这里写你的GPU计算能力
SMS ?= 61

ifeq ($(SMS),)
$(info >>> WARNING - no SM architectures have been specified - waiving sample <<<)
SAMPLE_ENABLED := 0
endif

ifeq ($(GENCODE_FLAGS),)
# Generate SASS code for each SM architecture listed in $(SMS)
$(foreach sm,$(SMS),$(eval GENCODE_FLAGS += -gencode arch=compute_$(sm),code=sm_$(sm)))

# Generate PTX code from the highest SM architecture in $(SMS) to guarantee forward-compatibility
HIGHEST_SM := $(lastword $(sort $(SMS)))
ifneq ($(HIGHEST_SM),)
GENCODE_FLAGS += -gencode arch=compute_$(HIGHEST_SM),code=compute_$(HIGHEST_SM)
endif
endif

ifeq ($(SAMPLE_ENABLED),0)
EXEC ?= @echo "[@]"
endif

################################################################################

# Target rules
all: build

# 这里写最后生成的二进制名字
build: mm

check.deps:
ifeq ($(SAMPLE_ENABLED),0)
	@echo "Sample will be waived due to the above missing dependencies"
else
	@echo "Sample is ready - all dependencies have been met"
endif

# 这里写二进制的依赖
mm: mm.o io.o parboil_cuda.o
	$(EXEC) $(NVCC) $(ALL_LDFLAGS) $(GENCODE_FLAGS) -o $@ $+ $(LIBRARIES)


# 这里写obj文件的依赖
mm.o:main.cu
	$(EXEC) $(NVCC) $(INCLUDES) $(ALL_CCFLAGS) $(GENCODE_FLAGS) -o $@ -c $<

io.o:io.cc
	$(EXEC) $(NVCC) $(INCLUDES) $(ALL_CCFLAGS) $(GENCODE_FLAGS) -o $@ -c $<

# 这个对于Parboil每个app都是必须的，路径要写对
parboil_cuda.o:/home/fish/GPU/benchmark/Parboil/parboil/common/src/parboil_cuda.c
	$(EXEC) $(NVCC) $(INCLUDES) $(ALL_CCFLAGS) $(GENCODE_FLAGS) -o $@ -c $<


run: build
	$(EXEC) ./mm
# 这里make clean时删除什么
clean:
	rm -f mm mm.o io.o parboil_cuda.o

clobber: clean

编译方法

修改上述编译脚本，然后在Parboil文件夹的src文件夹中建立一个Makefile文件，把这个脚本复制进去执行make即可。如果不知道需要那些.o的文件，可以看看parboil各个app内的build文件夹内的文件，大概就能知道需要什么.o文件了