BGI/BASE/src/2bwt/Makefile

   1 #-------------------------------------------------------------------------------
   2 #
   3 # Copyright 2010-2012 Duane Merrill
   4 #
   5 # Licensed under the Apache License, Version 2.0 (the "License");
   6 # you may not use this file except in compliance with the License.
   7 # You may obtain a copy of the License at
   8 #
   9 #     http://www.apache.org/licenses/LICENSE-2.0
  10 #
  11 # Unless required by applicable law or agreed to in writing, software
  12 # distributed under the License is distributed on an "AS IS" BASIS,
  13 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  14 # See the License for the specific language governing permissions and
  15 # limitations under the License.
  16 #
  17 #  For more information, see our Google Code project site:
  18 #  http://code.google.com/p/back40computing/
  19 #
  20 #-------------------------------------------------------------------------------
  21
  22 #-------------------------------------------------------------------------------
  23 # Build script for project
  24 #-------------------------------------------------------------------------------
  25
  26 CC = g++
  27 DEFINE =
  28 CFLAGS = -O3 -funroll-loops -march=nocona -maccumulate-outgoing-args -Wno-write-strings -Wno-unused-result -Wall
  29 LIBS = -lm -lz
  30
  31 ifeq ($(use_gpu), 1)
  32         NVCC = "$(shell which nvcc)"
  33         NVCC_VERSION = $(strip $(shell nvcc --version | grep release | sed 's/.*release //' |  sed 's/,.*//'))
  34 endif
  35
  36 KERNELS =
  37
  38 # detect OS
  39 OSUPPER = $(shell uname -s 2>/dev/null | tr [:lower:] [:upper:])
  40
  41 #-------------------------------------------------------------------------------
  42 # Gen targets
  43 #-------------------------------------------------------------------------------
  44
  45 GEN_SM30 = -gencode=arch=compute_30,code=\"sm_30,compute_30\"
  46 GEN_SM20 = -gencode=arch=compute_20,code=\"sm_20,compute_20\"
  47 GEN_SM13 = -gencode=arch=compute_13,code=\"sm_13,compute_13\"
  48 GEN_SM11 = -gencode=arch=compute_11,code=\"sm_11,compute_11\"
  49 GEN_SM10 = -gencode=arch=compute_10,code=\"sm_10,compute_10\"
  50 #SM_TARGETS = $(GEN_SM20) $(GEN_SM13) $(GEN_SM10)
  51 SM_TARGETS = $(GEN_SM30)
  52
  53 #-------------------------------------------------------------------------------
  54 # Tune arch
  55 #-------------------------------------------------------------------------------
  56
  57 ifdef tunearch
  58     TUNE_ARCH = $(tunearch)
  59 else
  60         TUNE_ARCH = 300
  61 endif
  62
  63 ifeq ($(TUNE_ARCH), 300)
  64         TUNE_SM_TARGETS = $(GEN_SM30)
  65 endif
  66 ifeq ($(TUNE_ARCH), 200)
  67         TUNE_SM_TARGETS = $(GEN_SM20)
  68 endif
  69 ifeq ($(TUNE_ARCH), 130)
  70         TUNE_SM_TARGETS = $(GEN_SM13)
  71 endif
  72 ifeq ($(TUNE_ARCH), 100)
  73         TUNE_SM_TARGETS = $(GEN_SM10)
  74 endif
  75
  76 #-------------------------------------------------------------------------------
  77 # Tune size
  78 #-------------------------------------------------------------------------------
  79
  80 ifdef tunesize
  81     TUNE_SIZE = $(tunesize)
  82 else
  83         TUNE_SIZE = 4
  84 endif
  85
  86
  87 #-------------------------------------------------------------------------------
  88 # Libs
  89 #-------------------------------------------------------------------------------
  90
  91
  92 #-------------------------------------------------------------------------------
  93 # Includes
  94 #-------------------------------------------------------------------------------
  95
  96 CUDA_INC = "$(shell dirname $(NVCC))/../include"
  97 INC = -I$(CUDA_INC) -I.
  98
  99 #-------------------------------------------------------------------------------
 100 # Defines
 101 #-------------------------------------------------------------------------------
 102
 103 DEFINES =
 104
 105 #-------------------------------------------------------------------------------
 106 # Compiler Flags
 107 #-------------------------------------------------------------------------------
 108
 109 force64 = 1
 110 ifneq ($(force64), 1)
 111 # Compile with 32-bit device pointers by default
 112         ARCH_SUFFIX = i386
 113         ARCH = -m32
 114 else
 115         ARCH_SUFFIX = x86_64
 116         ARCH = -m64
 117 endif
 118
 119 NVCCFLAGS = -Xptxas -v -Xcudafe -\# -cuda --ptxas-options=-v
 120
 121 ifeq (WIN_NT, $(findstring WIN_NT, $(OSUPPER)))
 122         NVCCFLAGS += -Xcompiler /bigobj -Xcompiler /Zm500
 123 endif
 124
 125 ifeq (,$(findstring 3.0, $(NVCC_VERSION)))
 126 ifneq ($(abi), 1)
 127 # Disable the ABI by default for 3.1+
 128         NVCCFLAGS += -Xptxas -abi=no
 129 endif
 130 endif
 131
 132 ifeq ($(verbose), 1)
 133     NVCCFLAGS += -v
 134 endif
 135
 136 ifeq ($(keep), 1)
 137     NVCCFLAGS += -keep
 138 endif
 139
 140 ifdef maxregisters
 141     NVCCFLAGS += -maxrregcount $(maxregisters)
 142 endif
 143
 144 ifeq (4.1, $(findstring 4.1, $(NVCC_VERSION)))
 145 # Disable nvvm for 4.1
 146         NVCCFLAGS += -open64
 147 endif
 148
 149 #-------------------------------------------------------------------------------
 150 # Dependency Lists
 151 #-------------------------------------------------------------------------------
 152
 153 DEPS =                  ./Makefile \
 154                                 $(wildcard b40c/util/*.cuh) \
 155                                 $(wildcard b40c/util/**/*.cuh) \
 156                                 $(wildcard b40c/scan/*.cuh) \
 157                                 $(wildcard b40c/scan/**/*.cuh) \
 158                                 $(wildcard b40c/partition/*.cuh) \
 159                                 $(wildcard b40c/partition/**/*.cuh) \
 160                                 $(wildcard b40c/radix_sort/*.cuh) \
 161                                 $(wildcard b40c/radix_sort/**/*.cuh)
 162
 163 #-------------------------------------------------------------------------------
 164 # g++ and its options
 165 #-------------------------------------------------------------------------------
 166
 167 CUDALIBFLAG = -L/usr/local/cuda/lib64/ -lcuda -lcudart
 168 CFLAGS = -O3 -Wall -funroll-loops -fprefetch-loop-arrays -fopenmp -std=c++0x -lm
 169 ZLIB = -lz
 170 ifneq ($(disablempopcnt), 1)
 171         CFLAGS += -mpopcnt
 172 endif
 173 DEPS = Makefile
 174 BIN_DIR = ./bin/
 175
 176
 177 #-------------------------------------------------------------------------------
 178 # CPU & GPU version
 179 #-------------------------------------------------------------------------------
 180
 181 ifeq ($(use_gpu), 1)
 182 all:  2bwt_builder_gpu 2bwt_builder_cpu AsiicBWT2BWT
 183 #2BWT-Builder 2BWT-Interface.o Sample overlap-sample
 184 else
 185 all:  2bwt_builder_cpu AsiicBWT2BWT
 186 # 2BWT-Builder 2BWT-Interface.o Sample overlap-sample
 187 endif
 188
 189
 190 #-------------------------------------------------------------------------------
 191 # CPU objectives
 192 #-------------------------------------------------------------------------------
 193
 194 %.o: %.cpp %.h $(DEPS)
 195         $(CC) $(CFLAGS) -c $< -o $@
 196
 197 2bwt_builder_cpu: readbwt.cpp lv2_cpu_sort.h $(DEPS)
 198         $(CC) $(CFLAGS) -D DISABLE_GPU readbwt.cpp $(ZLIB) -o 2bwt_builder_cpu
 199
 200 ifeq ($(use_gpu), 1)
 201 .lv2_gpu_sort_$(NVCC_VERSION)_$(ARCH_SUFFIX).cpp: lv2_gpu_sort.cu lv2_gpu_sort.h $(DEPS)
 202         $(NVCC) $(SM_TARGETS) lv2_gpu_sort.cu $(NVCCFLAGS) $(ARCH) $(INC) $(LIBS) -O3 -c -o .lv2_gpu_sort_$(NVCC_VERSION)_$(ARCH_SUFFIX).cpp
 203 # cpp -> o
 204 .lv2_gpu_sort_$(NVCC_VERSION)_$(ARCH_SUFFIX).o: .lv2_gpu_sort_$(NVCC_VERSION)_$(ARCH_SUFFIX).cpp $(DEPS)
 205         $(CC) $(CFLAGS) -c .lv2_gpu_sort_$(NVCC_VERSION)_$(ARCH_SUFFIX).cpp -o .lv2_gpu_sort_$(NVCC_VERSION)_$(ARCH_SUFFIX).o
 206
 207 2bwt_builder_gpu: readbwt.cpp lv2_cpu_sort.h .lv2_gpu_sort_$(NVCC_VERSION)_$(ARCH_SUFFIX).o $(DEPS)
 208         $(CC) $(CFLAGS) $(CUDALIBFLAG) readbwt.cpp .lv2_gpu_sort_$(NVCC_VERSION)_$(ARCH_SUFFIX).o $(ZLIB) -o 2bwt_builder_gpu
 209 endif
 210
 211
 212 AsiicBWT2BWT: AsiicBWT2BWT.c
 213         $(CC) $(CFLAGS) AsiicBWT2BWT.c -o AsiicBWT2BWT $(LIBS)
 214
 215 clean :
 216         rm -f *.i* *.cubin *.cu.c *.cudafe* *.fatbin.c *.ptx *.hash *.cu.cpp *.o 2bwt_builder_*pu AsiicBWT2BWT