build for specific gpu architectures

make.device

@@ -0,0 +1,42 @@
+# make.device
+# add gpu specific compilation flags
+
+
+ifneq ($(findstring Kepler, $(GPU_TARGET)),)
+MIN_ARCH ?= 300
+NV_SM += -gencode arch=compute_30,code=sm_30
+NV_SM += -gencode arch=compute_35,code=sm_35
+NV_COMP := -gencode arch=compute_35,code=compute_35
+endif
+
+ifneq ($(findstring Maxwell, $(GPU_TARGET)),)
+MIN_ARCH ?= 500
+NV_SM += -gencode arch=compute_50,code=sm_50
+NV_SM += -gencode arch=compute_52,code=sm_52
+NV_COMP := -gencode arch=compute_52,code=compute_52
+endif
+
+ifneq ($(findstring Pascal, $(GPU_TARGET)),)
+MIN_ARCH ?= 600
+NV_SM += -gencode arch=compute_60,code=sm_60
+NV_SM += -gencode arch=compute_61,code=sm_61
+NV_COMP := -gencode arch=compute_61,code=compute_61
+endif
+
+ifneq ($(findstring Volta, $(GPU_TARGET)),)
+MIN_ARCH ?= 700
+NV_SM += -gencode arch=compute_70,code=sm_70
+NV_COMP := -gencode arch=compute_70,code=compute_70
+endif
+
+ifneq ($(findstring Tesla, $(GPU_TARGET)),)
+MIN_ARCH ?= 750
+NV_SM += -gencode arch=compute_75,code=sm_75
+NV_COMP := -gencode arch=compute_75,code=compute_75
+endif
+
+
+ifeq ($(NV_COMP),)
+$(error GPU_TARGET, is $(GPU_TARGET), must contain one or more of Kepler, Maxwell, Pascal, Volta, and Tesla)
+endif
+

makefile

@@ -24,6 +24,7 @@ LIBS =
 # use nvcc to determine if we should compile for gpu or not
 USE_CUDA = 0
 ifneq ($(shell which nvcc),)
+include make.device
 CUDA_MACRO = -D_HAS_CUDA_
 INC += -I$(CUDADIR)/include
 LIBDIRS += -L$(CUDADIR)/lib64
@@ -40,7 +41,7 @@ CXX_VERSION ?= -std=c++11
 
 # the entire flags for compilation
 CXXFLAGS := $(OPTIMIZATION_LEVEL) $(WARNINGS) $(CXX_VERSION) $(CUDA_MACRO) $(FPIC) -MMD
-NVCCFLAGS := $(CXX_VERSION) $(OPTIMIZATION_LEVEL) -Xcompiler "$(CXXFLAGS)"
+NVCCFLAGS := $(CXX_VERSION) $(OPTIMIZATION_LEVEL) -Xcompiler "$(CXXFLAGS)" $(NV_SM) $(NV_COMP)
 
 
 # make these available to child makefiles