Makefile.Cuda

# AutoDock-GPU Cuda Makefile

# ------------------------------------------------------
# Note that environment variables must be defined
# before compiling
# DEVICE?
# if DEVICE=CPU: CPU_INCLUDE_PATH?, CPU_LIBRARY_PATH?
# if DEVICE=GPU: GPU_INCLUDE_PATH?, GPU_LIBRARY_PATH?

# ------------------------------------------------------
# Choose OpenCL device
# Valid values: CPU, GPU

NVCC = nvcc
CPP = g++
UNAME := $(shell uname)

TARGETS = 52 60 61 70 80

ifeq ($(TENSOR), ON)
	NVTENSOR=-DUSE_NVTENSOR
	NVTENSOR+=-I./wmma_extension/include
	TARGETS = 80
endif

CUDA_TARGETS=$(foreach target,$(TARGETS),-gencode arch=compute_$(target),code=sm_$(target))

ifeq ($(DEVICE), CPU)
	DEV =-DCPU_DEVICE
else ifeq ($(DEVICE), GPU)
	DEV =-DGPU_DEVICE
endif

# ------------------------------------------------------
# Project directories
# opencl_lvs: wrapper for OpenCL APIs
COMMON_DIR=./common
HOST_INC_DIR=./host/inc
HOST_SRC_DIR=./host/src
KRNL_DIR=./cuda
KCMN_DIR=$(COMMON_DIR)
BIN_DIR=./bin
LIB_CUDA = kernels.o -lcurand -lcudart -DUSE_CUDA

TARGET := autodock
TOOL_TARGET := adgpu_analysis

IFLAGS=-I$(COMMON_DIR) -I$(HOST_INC_DIR) -I$(GPU_INCLUDE_PATH) -I$(KRNL_DIR)
LFLAGS=-L$(GPU_LIBRARY_PATH) -Wl,-rpath=$(GPU_LIBRARY_PATH):$(CPU_LIBRARY_PATH)
CFLAGS=-std=c++17 $(IFLAGS) $(LFLAGS)
TOOL_CFLAGS=-std=c++17 -I$(COMMON_DIR) -I$(HOST_INC_DIR)

ifeq ($(DEVICE), CPU)
	TARGET:=$(TARGET)_cpu
else ifeq ($(DEVICE), GPU)
	NWI=-DN64WI
	TARGET:=$(TARGET)_gpu
endif

ifeq ($(OVERLAP), ON)
	PIPELINE=-DUSE_PIPELINE -fopenmp
endif

BIN := $(wildcard $(TARGET)*)

# ------------------------------------------------------
# Number of work-items (wi)
# Valid values: 32, 64, 128, 256
NUMWI=

ifeq ($(NUMWI), 32)
ifeq ($(TENSOR), ON)
$(error NUMWI needs to be at least 64 with TENSOR=ON)
endif
	NWI=-DN32WI
	TARGET:=$(TARGET)_32wi
else ifeq ($(NUMWI), 64)
	NWI=-DN64WI
	TARGET:=$(TARGET)_64wi
else ifeq ($(NUMWI), 128)
	NWI=-DN128WI
	TARGET:=$(TARGET)_128wi
else ifeq ($(NUMWI), 256)
		NWI=-DN256WI
		TARGET:=$(TARGET)_256wi
else
	ifeq ($(DEVICE), CPU)
		NWI=-DN16WI
		TARGET:=$(TARGET)_16wi
	else ifeq ($(DEVICE), GPU)
		NWI=-DN128WI
		TARGET:=$(TARGET)_128wi
	endif
endif

# ------------------------------------------------------
# Configuration
# FDEBUG (full) : enables debugging on both host + device
# LDEBUG (light): enables debugging on host
# RELEASE
CONFIG=RELEASE
#CONFIG=FDEBUG

ifeq ($(CONFIG),FDEBUG)
	OPT =-O0 -g3 -Wall -DDOCK_DEBUG
	CUDA_FLAGS = -G -use_fast_math --ptxas-options="-v" $(CUDA_TARGETS) -std=c++11
else ifeq ($(CONFIG),LDEBUG)
	OPT =-O0 -g3 -Wall
	CUDA_FLAGS = -use_fast_math --ptxas-options="-v" $(CUDA_TARGETS) -std=c++11
else ifeq ($(CONFIG),RELEASE)
	OPT =-O3
	CUDA_FLAGS = -use_fast_math --ptxas-options="-v" $(CUDA_TARGETS) -std=c++11
else
	OPT =
	CUDA_FLAGS = -use_fast_math --ptxas-options="-v" $(CUDA_TARGETS) -std=c++11
endif

# ------------------------------------------------------
# Reproduce results (remove randomness)
REPRO=NO

ifeq ($(REPRO),YES)
	REP =-DREPRO
else
	REP =
endif
# ------------------------------------------------------

all: otool odock

check-env-dev:
	@if test -z "$$DEVICE"; then \
		echo "Please set DEVICE to either CPU, GPU, CUDA, or OCLGPU to build docking software."; \
		exit 1; \
	else \
		if [ "$$DEVICE" = "CPU" ]; then \
			echo "DEVICE is set to $$DEVICE which is not a valid Cuda device."; \
			exit 1; \
		else \
			if [ "$$DEVICE" = "GPU" ]; then \
				echo "DEVICE is set to $$DEVICE"; \
			else \
				echo "DEVICE value is invalid. Please set DEVICE to either CPU, GPU, CUDA, or OCLGPU"; \
				exit 1; \
			fi; \
		fi; \
	fi; \
	echo " "

check-env-cpu:
	@if test -z "$$CPU_INCLUDE_PATH"; then \
		echo "CPU_INCLUDE_PATH is undefined"; \
	else \
		echo "CPU_INCLUDE_PATH is set to $$CPU_INCLUDE_PATH"; \
	fi; \
	if test -z "$$CPU_LIBRARY_PATH"; then \
		echo "CPU_LIBRARY_PATH is undefined"; \
	else \
		echo "CPU_LIBRARY_PATH is set to $$CPU_LIBRARY_PATH"; \
	fi; \
	echo " "

check-env-gpu:
	@if test -z "$$GPU_INCLUDE_PATH"; then \
		echo "GPU_INCLUDE_PATH is undefined"; \
	else \
		echo "GPU_INCLUDE_PATH is set to $$GPU_INCLUDE_PATH"; \
	fi; \
	if test -z "$$GPU_LIBRARY_PATH"; then \
		echo "GPU_LIBRARY_PATH is undefined"; \
	else \
		echo "GPU_LIBRARY_PATH is set to $$GPU_LIBRARY_PATH"; \
	fi; \
	echo " "

check-env-all: check-env-dev check-env-cpu check-env-gpu

# ------------------------------------------------------
# Priting out its git version hash

GIT_VERSION := $(shell ./version_string.sh)

CFLAGS+=-DVERSION=\"$(GIT_VERSION)\"
TOOL_CFLAGS+=-DVERSION=\"$(GIT_VERSION)\"

# ------------------------------------------------------

kernels: $(KERNEL_SRC)
	$(NVCC) $(NWI) $(REP) $(CUDA_FLAGS) $(IFLAGS) $(CUDA_INCLUDES) $(NVTENSOR) -c $(KRNL_DIR)/kernels.cu

otool:
	@echo "Building" $(TOOL_TARGET) "..."
	$(CPP) \
	$(shell ls $(HOST_SRC_DIR)/*.cpp) \
	$(TOOL_CFLAGS) \
	-o$(BIN_DIR)/$(TOOL_TARGET) \
	$(PIPELINE) $(NVTENSOR) $(OPT) -DTOOLMODE $(REP)

odock: check-env-all kernels
	@echo "Building" $(TARGET) "..."
	$(CPP) \
	$(shell ls $(HOST_SRC_DIR)/*.cpp) \
	$(CFLAGS) \
	$(LIB_CUDA) \
	-o$(BIN_DIR)/$(TARGET) \
	$(DEV) $(NWI) $(PIPELINE) $(NVTENSOR) $(OPT) $(DD) $(REP) $(KFLAGS)

# Example
# 1ac8: for testing gradients of translation and rotation genes
# 7cpa: for testing gradients of torsion genes (15 torsions) 
# 3tmn: for testing gradients of torsion genes (1 torsion)

PDB      := 3ce3
NRUN     := 100
NGEN     := 27000
POPSIZE  := 150
TESTNAME := test
TESTLS   := ad

test: odock
	$(BIN_DIR)/$(TARGET) \
	-ffile ./input/$(PDB)/derived/$(PDB)_protein.maps.fld \
	-lfile ./input/$(PDB)/derived/$(PDB)_ligand.pdbqt \
	-nrun $(NRUN) \
	-ngen $(NGEN) \
	-psize $(POPSIZE) \
	-resnam $(TESTNAME) \
	-gfpop 0 \
	-lsmet $(TESTLS)

.PHONY: clean