#cuda compile

CUDA_PATH ?= /usr/local/cuda
TARGET_SIZE := 64 # 32
HOST_COMPILER ?= g++
NVCC := $(CUDA_PATH)/bin/nvcc -ccbin $(HOST_COMPILER)
NVCCFLAGS := -m${TARGET_SIZE} -Xcompiler -fPIC -Xcompiler -fno-omit-frame-pointer -g
LDFLAGS = -lcuda -L $(CUDA_PATH)/targets/x86_64-linux/lib -lcudart

SMS ?= 30 35 37 50 52 60 61 70 75
$(foreach sm,$(SMS),$(eval GENCODE_FLAGS += -gencode arch=compute_$(sm),code=sm_$(sm)))
HIGHEST_SM := $(lastword $(sort $(SMS)))
GENCODE_FLAGS += -gencode arch=compute_$(HIGHEST_SM),code=compute_$(HIGHEST_SM)

# samples source access BEWARE!
# contains includes which conflict: search /usr/include first
CFLAGS += -I/usr/include -I$(CUDA_PATH)/samples
CFLAGS += -I$(CUDA_PATH)/samples/common/inc
CFLAGS += -I$(CUDA_PATH)/targets/x86_64-linux/include
CFLAGS += -Wno-unused-function

INCLUDES += -I$(TOPDIR)/cinelerra -I$(TOPDIR)/guicast
INCLUDES += -I$(CUDA_PATH)/samples
INCLUDES += -I$(CUDA_PATH)/samples/common/inc
INCLUDES += -I$(CUDA_PATH)/targets/x86_64-linux/include

CUFLAGS := $(INCLUDES) $(NVCCFLAGS) $(GENCODE_FLAGS)
$(shell mkdir -p $(OBJDIR))
$(shell echo $(CUFLAGS) > $(OBJDIR)/cu_flags)

$(OBJDIR)/%.o: %.cu
	$(NVCC) `cat $(OBJDIR)/cu_flags` $(BFLAGS) -DMSGQUAL=$* -c $< -o $@