# Location of the CUDA Toolkit
CUDA_PATH ?= /usr/local/cuda-10.1

# Location of BOINC libraries
BOINC_DIR = /home/eric/BOINC_Project/boinc
BOINC_API_DIR = $(BOINC_DIR)/api
BOINC_LIB_DIR = $(BOINC_DIR)/lib
BOINC_SCHED_DIR = $(BOINC_DIR)/sched
BOINC_DB_DIR = $(BOINC_DIR)/db

# Location of PARI library
PARI = /home/eric/BOINC_Project/Pari/pari-2.8-1711-ge5c317c


# 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
ifeq ($(TARGET_OS),darwin)
    ifeq ($(shell expr `xcodebuild -version | grep -i xcode | awk '{print $$2}' | cut -d'.' -f1` \>= 5),1)
        HOST_COMPILER ?= clang++
    endif
else ifneq ($(TARGET_ARCH),$(HOST_ARCH))
    ifeq ($(HOST_ARCH)-$(TARGET_ARCH),x86_64-armv7l)
        ifeq ($(TARGET_OS),linux)
            HOST_COMPILER ?= arm-linux-gnueabihf-g++
        else ifeq ($(TARGET_OS),qnx)
            ifeq ($(QNX_HOST),)
                $(error ERROR - QNX_HOST must be passed to the QNX host toolchain)
            endif
            ifeq ($(QNX_TARGET),)
                $(error ERROR - QNX_TARGET must be passed to the QNX target toolchain)
            endif
            export QNX_HOST
            export QNX_TARGET
            HOST_COMPILER ?= $(QNX_HOST)/usr/bin/arm-unknown-nto-qnx6.6.0eabi-g++
        else ifeq ($(TARGET_OS),android)
            HOST_COMPILER ?= arm-linux-androideabi-g++
        endif
    else ifeq ($(TARGET_ARCH),aarch64)
        ifeq ($(TARGET_OS), linux)
            HOST_COMPILER ?= aarch64-linux-gnu-g++
        else ifeq ($(TARGET_OS),qnx)
            ifeq ($(QNX_HOST),)
                $(error ERROR - QNX_HOST must be passed to the QNX host toolchain)
            endif
            ifeq ($(QNX_TARGET),)
                $(error ERROR - QNX_TARGET must be passed to the QNX target toolchain)
            endif
            export QNX_HOST
            export QNX_TARGET
            HOST_COMPILER ?= $(QNX_HOST)/usr/bin/aarch64-unknown-nto-qnx7.0.0-g++
        else ifeq ($(TARGET_OS), android)
            HOST_COMPILER ?= aarch64-linux-android-clang++
        endif
    else ifeq ($(TARGET_ARCH),ppc64le)
        HOST_COMPILER ?= powerpc64le-linux-gnu-g++
    endif
endif
HOST_COMPILER ?= g++
NVCC          := $(CUDA_PATH)/bin/nvcc -ccbin $(HOST_COMPILER)


# internal flags
NVCCFLAGS   := -m${TARGET_SIZE}
CCFLAGS     := -O3
LDFLAGS     :=


# build flags
ifeq ($(TARGET_OS),darwin)
    LDFLAGS += -rpath $(CUDA_PATH)/lib
    CCFLAGS += -arch $(HOST_ARCH)
else ifeq ($(HOST_ARCH)-$(TARGET_ARCH)-$(TARGET_OS),x86_64-armv7l-linux)
    LDFLAGS += --dynamic-linker=/lib/ld-linux-armhf.so.3
    CCFLAGS += -mfloat-abi=hard
else ifeq ($(TARGET_OS),android)
    LDFLAGS += -pie
    CCFLAGS += -fpie -fpic -fexceptions
endif

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
    ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-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 -L $(TARGET_FS)/lib
            LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib -L $(TARGET_FS)/usr/lib
            LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib/aarch64-linux-gnu -L $(TARGET_FS)/usr/lib/aarch64-linux-gnu
            LDFLAGS += --unresolved-symbols=ignore-in-shared-libs
            CCFLAGS += -isystem=$(TARGET_FS)/usr/include
            CCFLAGS += -isystem=$(TARGET_FS)/usr/include/aarch64-linux-gnu
        endif
    endif
endif

ifeq ($(TARGET_OS),qnx)
    CCFLAGS += -DWIN_INTERFACE_CUSTOM
    LDFLAGS += -lsocket
endif


# Install directory of different arch
CUDA_INSTALL_TARGET_DIR :=
ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-linux)
    CUDA_INSTALL_TARGET_DIR = targets/armv7-linux-gnueabihf/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-linux)
    CUDA_INSTALL_TARGET_DIR = targets/aarch64-linux/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-android)
    CUDA_INSTALL_TARGET_DIR = targets/armv7-linux-androideabi/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-android)
    CUDA_INSTALL_TARGET_DIR = targets/aarch64-linux-androideabi/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-qnx)
    CUDA_INSTALL_TARGET_DIR = targets/ARMv7-linux-QNX/
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-qnx)
    CUDA_INSTALL_TARGET_DIR = targets/aarch64-qnx/
else ifeq ($(TARGET_ARCH),ppc64le)
    CUDA_INSTALL_TARGET_DIR = targets/ppc64le-linux/
endif


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


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))

# Common includes and paths for CUDA
NVCC_INCLUDES  := -I$(CUDA_PATH)/samples/common/inc/
LIBRARIES :=


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

# Gencode arguments
SMS ?= 30 35 37 50 52 60 61 70 75
#SMS ?= 50 52 60 61 70 75
#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

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






system  := $(shell uname -s | tr A-Z a-z )
hw      := $(shell uname -m)
OBase   := O$(system)-$(hw)
O       := $(OBase)-Cuda

CC  = g++

CUDA_INC = /usr/local/cuda/targets/x86_64-linux/include
INCLUDES = -I$(BOINC_DIR) -I$(BOINC_LIB_DIR) -I$(BOINC_API_DIR) -I$(BOINC_SCHED_DIR) -I$(BOINC_DB_DIR) -I$(PARI)/src/headers -I$(PARI)/$(OBase) -I$(CUDA_INC)
#INCLUDES = -I$(BOINC_DIR) -I$(BOINC_LIB_DIR) -I$(BOINC_API_DIR) -I$(BOINC_SCHED_DIR) -I$(BOINC_DB_DIR) -I$(PARI)/src/headers -I$(PARI)/$(OBase)

#LIBS =  -L$(BOINC_LIB_DIR) -L$(BOINC_API_DIR) -L$(PARI)/$(OBase) -L/usr/local/lib -lboinc_api -lboinc -lpari -lm -lgmp
LIBS =  -L$(BOINC_LIB_DIR) -L$(BOINC_API_DIR) -L$(PARI)/$(OBase) -L/usr/local/lib -lboinc_api -lboinc -l:libpari.a -lm -l:libgmp.a

#CXXFLAGS = -O3 -Wall -Werror -fmax-errors=4 $(INCLUDES)
CXXFLAGS = -O3 -Wall -Wno-unknown-pragmas -Wno-class-memaccess -fmax-errors=4 $(INCLUDES)

# Using the -static flag causes the cuda routines to crash.
#STATIC_FLAGS = -Xcompiler -static-libgcc -Xcompiler -static-libstdc++ -Xcompiler -static
STATIC_FLAGS = -Xcompiler -static-libgcc -Xcompiler -static-libstdc++

# Use this to get line numbers with cuda-memcheck
#ALL_CCFLAGS += -lineinfo -Xptxas --warn-on-spills -Xptxas --verbose -Xcompiler -fmax-errors=4

PREPROC_DEFS = -DAPP_VERSION_GPU_CUDA -DCUDA

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

# Target rules

all: build

build: $(O) $(O)/GetDecics

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


$(O)/polDiscTest_gpuCuda.o: polDiscTest_gpuCuda.cu gpuMultiPrec128.h
	$(EXEC) $(NVCC) $(NVCC_INCLUDES) $(ALL_CCFLAGS) $(GENCODE_FLAGS) -o $@ -c $<

$(O)/polDiscTest_cpu.o: polDiscTest_cpu.cpp
	$(CC) $(CXXFLAGS) -c -o $@ $< -pthread

$(O)/TgtMartinet.o: TgtMartinet.cpp TgtMartinet.h $(O)/polDiscTest_cpu.o
	$(CC) $(CXXFLAGS) $(PREPROC_DEFS) -c -o $@ $< -pthread

$(O)/cuda_GetDecics.o: cuda_GetDecics.cpp cuda_GetDecics.h
	$(CC) $(CXXFLAGS) $(PREPROC_DEFS) -c -o $@ $< -pthread

$(O)/GetDecics.o: GetDecics.cpp GetDecics.h
	$(CC) $(CXXFLAGS) $(PREPROC_DEFS) -c -o $@ $< -pthread

$(O)/GetDecics: $(O)/GetDecics.o $(O)/cuda_GetDecics.o $(O)/TgtMartinet.o $(O)/polDiscTest_gpuCuda.o $(O)/polDiscTest_cpu.o
	$(EXEC) $(NVCC) $(STATIC_FLAGS) $(ALL_LDFLAGS) $(GENCODE_FLAGS) -o $@ $+ $(LIBS) $(LIBRARIES) -Xcompiler -pthread

$(O): ;	mkdir -p $(O)

clean:
	rm -r -f $(O)