1 # This Makefile compiles a library containing the C++ runtime for the TensorFlow 2 # library. It's designed for use on platforms with limited resources where 3 # running a full Bazel build would be prohibitive, or for cross-compilation onto 4 # embedded systems. It includes only a bare-bones set of functionality. 5 # 6 # The default setup below is aimed at Unix-like devices, and should work on 7 # modern Linux and OS X distributions without changes. 8 # 9 # If you have another platform, you'll need to take a careful look at the 10 # compiler flags and folders defined below. They're separated into two sections, 11 # the first for the host (the machine you're compiling on) and the second for 12 # the target (the machine you want the program to run on). 13 14 SHELL := /bin/bash 15 16 # Host compilation settings 17 18 # Find where we're running from, so we can store generated files here. 19 ifeq ($(origin MAKEFILE_DIR), undefined) 20 MAKEFILE_DIR := $(shell dirname $(realpath $(lastword $(MAKEFILE_LIST)))) 21 endif 22 23 HAS_GEN_HOST_PROTOC := \ 24 $(shell test -f $(MAKEFILE_DIR)/gen/protobuf-host/bin/protoc && echo "true" ||\ 25 echo "false") 26 27 # Hexagon integration 28 ifdef HEXAGON_LIBS 29 LIBGEMM_WRAPPER := $(HEXAGON_LIBS)/libhexagon_controller.so 30 ifeq ($(shell test -f $(LIBGEMM_WRAPPER) 2> /dev/null; echo $$?), 0) 31 $(info "Use hexagon libs at " $(LIBGEMM_WRAPPER)) 32 else 33 $(error "hexagon libs not found at " $(LIBGEMM_WRAPPER)) 34 endif 35 ifdef HEXAGON_INCLUDE 36 ifeq ($(shell test -d $(HEXAGON_INCLUDE) 2> /dev/null; echo $$?), 0) 37 $(info "Use hexagon libs at " $(HEXAGON_INCLUDE)) 38 else 39 $(error "hexagon libs not found at " $(HEXAGON_INCLUDE)) 40 endif 41 else 42 $(error "HEXAGON_INCLUDE is not set.") 43 endif 44 ifneq ($(TARGET),ANDROID) 45 $(error "hexagon is only supported on Android") 46 endif 47 endif # HEXAGON_LIBS 48 49 # If ANDROID_TYPES is not set assume __ANDROID_TYPES_SLIM__ 50 ifeq ($(ANDROID_TYPES),) 51 ANDROID_TYPES := -D__ANDROID_TYPES_SLIM__ 52 endif 53 54 # Try to figure out the host system 55 HOST_OS := 56 ifeq ($(OS),Windows_NT) 57 HOST_OS = WINDOWS 58 else 59 UNAME_S := $(shell uname -s) 60 ifeq ($(UNAME_S),Linux) 61 HOST_OS := LINUX 62 endif 63 ifeq ($(UNAME_S),Darwin) 64 HOST_OS := OSX 65 endif 66 endif 67 68 HOST_ARCH := $(shell if [[ $(shell uname -m) =~ i[345678]86 ]]; then echo x86_32; else echo $(shell uname -m); fi) 69 70 # Where compiled objects are stored. 71 HOST_OBJDIR := $(MAKEFILE_DIR)/gen/host_obj/ 72 HOST_BINDIR := $(MAKEFILE_DIR)/gen/host_bin/ 73 HOST_GENDIR := $(MAKEFILE_DIR)/gen/host_obj/ 74 75 # Settings for the host compiler. 76 HOST_CXX := $(CC_PREFIX) gcc 77 HOST_CXXFLAGS := --std=c++11 78 HOST_LDOPTS := 79 ifeq ($(HAS_GEN_HOST_PROTOC),true) 80 HOST_LDOPTS += -L$(MAKEFILE_DIR)/gen/protobuf-host/lib 81 endif 82 HOST_LDOPTS += -L/usr/local/lib 83 84 HOST_INCLUDES := \ 85 -I. \ 86 -I$(MAKEFILE_DIR)/../../../ \ 87 -I$(MAKEFILE_DIR)/downloads/ \ 88 -I$(MAKEFILE_DIR)/downloads/eigen \ 89 -I$(MAKEFILE_DIR)/downloads/gemmlowp \ 90 -I$(MAKEFILE_DIR)/downloads/nsync/public \ 91 -I$(MAKEFILE_DIR)/downloads/fft2d \ 92 -I$(HOST_GENDIR) 93 ifeq ($(HAS_GEN_HOST_PROTOC),true) 94 HOST_INCLUDES += -I$(MAKEFILE_DIR)/gen/protobuf-host/include 95 endif 96 # This is at the end so any globally-installed frameworks like protobuf don't 97 # override local versions in the source tree. 98 HOST_INCLUDES += -I/usr/local/include 99 100 HOST_LIBS := \ 101 $(HOST_NSYNC_LIB) \ 102 -lstdc++ \ 103 -lprotobuf \ 104 -lpthread \ 105 -lm \ 106 -lz 107 108 # If we're on Linux, also link in the dl library. 109 ifeq ($(HOST_OS),LINUX) 110 HOST_LIBS += -ldl -lpthread 111 endif 112 113 # If we're on a Pi, link in pthreads and dl 114 ifeq ($(HOST_OS),PI) 115 HOST_LIBS += -ldl -lpthread 116 endif 117 118 119 # proto_text is a tool that converts protobufs into a form we can use more 120 # compactly within TensorFlow. It's a bit like protoc, but is designed to 121 # produce a much more minimal result so we can save binary space. 122 # We have to build it on the host system first so that we can create files 123 # that are needed for the runtime building. 124 PROTO_TEXT := $(HOST_BINDIR)proto_text 125 # The list of dependencies is derived from the Bazel build file by running 126 # the gen_file_lists.sh script on a system with a working Bazel setup. 127 PROTO_TEXT_CC_FILES := $(shell cat $(MAKEFILE_DIR)/proto_text_cc_files.txt) 128 PROTO_TEXT_PB_CC_LIST := $(shell cat $(MAKEFILE_DIR)/proto_text_pb_cc_files.txt) 129 PROTO_TEXT_PB_H_LIST := $(shell cat $(MAKEFILE_DIR)/proto_text_pb_h_files.txt) 130 131 # Locations of the intermediate files proto_text generates. 132 PROTO_TEXT_PB_H_FILES := $(addprefix $(HOST_GENDIR), $(PROTO_TEXT_PB_H_LIST)) 133 PROTO_TEXT_CC_OBJS := $(addprefix $(HOST_OBJDIR), $(PROTO_TEXT_CC_FILES:.cc=.o)) 134 PROTO_TEXT_PB_OBJS := $(addprefix $(HOST_OBJDIR), $(PROTO_TEXT_PB_CC_LIST:.cc=.o)) 135 PROTO_TEXT_OBJS := $(PROTO_TEXT_CC_OBJS) $(PROTO_TEXT_PB_OBJS) 136 137 # Target device settings. 138 139 # Default to running on the same system we're compiling on. 140 # You should override TARGET on the command line if you're cross-compiling, e.g. 141 # make -f tensorflow/contrib/makefile/Makefile TARGET=ANDROID 142 TARGET := $(HOST_OS) 143 144 # Where compiled objects are stored. 145 GENDIR := $(MAKEFILE_DIR)/gen/ 146 OBJDIR := $(GENDIR)obj/ 147 LIBDIR := $(GENDIR)lib/ 148 BINDIR := $(GENDIR)bin/ 149 PBTGENDIR := $(GENDIR)proto_text/ 150 PROTOGENDIR := $(GENDIR)proto/ 151 DEPDIR := $(GENDIR)dep/ 152 $(shell mkdir -p $(DEPDIR) >/dev/null) 153 154 # Settings for the target compiler. 155 CXX := $(CC_PREFIX) gcc 156 OPTFLAGS := -O2 157 158 ifneq ($(TARGET),ANDROID) 159 OPTFLAGS += -march=native 160 endif 161 162 CXXFLAGS := --std=c++11 -DIS_SLIM_BUILD -fno-exceptions -DNDEBUG $(OPTFLAGS) 163 LDFLAGS := \ 164 -L/usr/local/lib 165 DEPFLAGS = -MT $@ -MMD -MP -MF $(DEPDIR)/$*.Td 166 167 INCLUDES := \ 168 -I. \ 169 -I$(MAKEFILE_DIR)/downloads/ \ 170 -I$(MAKEFILE_DIR)/downloads/eigen \ 171 -I$(MAKEFILE_DIR)/downloads/gemmlowp \ 172 -I$(MAKEFILE_DIR)/downloads/nsync/public \ 173 -I$(MAKEFILE_DIR)/downloads/fft2d \ 174 -I$(PROTOGENDIR) \ 175 -I$(PBTGENDIR) 176 ifeq ($(HAS_GEN_HOST_PROTOC),true) 177 INCLUDES += -I$(MAKEFILE_DIR)/gen/protobuf-host/include 178 endif 179 # This is at the end so any globally-installed frameworks like protobuf don't 180 # override local versions in the source tree. 181 INCLUDES += -I/usr/local/include 182 183 LIBS := \ 184 $(TARGET_NSYNC_LIB) \ 185 -lstdc++ \ 186 -lprotobuf \ 187 -lz \ 188 -lm 189 190 ifeq ($(HAS_GEN_HOST_PROTOC),true) 191 PROTOC := $(MAKEFILE_DIR)/gen/protobuf-host/bin/protoc 192 else 193 PROTOC := protoc 194 endif 195 196 $(info PROTOC = "$(PROTOC)") 197 $(info CC_PREFIX = "$(CC_PREFIX)") 198 199 PROTOCFLAGS := 200 AR := ar 201 ARFLAGS := -r 202 LIBFLAGS := 203 204 # If we're on OS X, make sure that globals aren't stripped out. 205 ifeq ($(TARGET),OSX) 206 ifeq ($(HAS_GEN_HOST_PROTOC),true) 207 LIBFLAGS += -L$(MAKEFILE_DIR)/gen/protobuf-host/lib 208 export LD_LIBRARY_PATH=$(MAKEFILE_DIR)/gen/protobuf-host/lib 209 endif 210 LDFLAGS += -all_load 211 endif 212 # Make sure that we don't strip global constructors on Linux. 213 ifeq ($(TARGET),LINUX) 214 ifeq ($(HAS_GEN_HOST_PROTOC),true) 215 LIBFLAGS += -L$(MAKEFILE_DIR)/gen/protobuf-host/lib 216 export LD_LIBRARY_PATH=$(MAKEFILE_DIR)/gen/protobuf-host/lib 217 endif 218 CXXFLAGS += -fPIC 219 LIBFLAGS += -Wl,--allow-multiple-definition -Wl,--whole-archive 220 LDFLAGS := -Wl,--no-whole-archive 221 endif 222 # If we're on Linux, also link in the dl library. 223 ifeq ($(TARGET),LINUX) 224 LIBS += -ldl -lpthread 225 endif 226 # If we're cross-compiling for the Raspberry Pi, use the right gcc. 227 ifeq ($(TARGET),PI) 228 CXXFLAGS += $(ANDROID_TYPES) -DRASPBERRY_PI 229 LDFLAGS := -Wl,--no-whole-archive 230 LIBS += -ldl -lpthread 231 LIBFLAGS += -Wl,--allow-multiple-definition -Wl,--whole-archive 232 endif 233 234 # Set up Android building 235 # LINT.IfChange 236 ifeq ($(TARGET),ANDROID) 237 # Override NDK_ROOT on the command line with your own NDK location, e.g. 238 # make -f tensorflow/contrib/makefile/Makefile TARGET=ANDROID \ 239 # NDK_ROOT=/path/to/your/ndk 240 # You need to have an Android version of the protobuf libraries compiled to link 241 # in. The compile_android_protobuf.sh script may help. 242 243 ANDROID_HOST_OS_ARCH := 244 ifeq ($(HOST_OS),LINUX) 245 ANDROID_HOST_OS_ARCH=linux 246 endif 247 ifeq ($(HOST_OS),OSX) 248 ANDROID_HOST_OS_ARCH=darwin 249 endif 250 ifeq ($(HOST_OS),WINDOWS) 251 $(error "windows is not supported.") 252 endif 253 254 ifeq ($(HOST_ARCH),x86_32) 255 ANDROID_HOST_OS_ARCH := $(ANDROID_HOST_OS_ARCH)-x86 256 else 257 ANDROID_HOST_OS_ARCH := $(ANDROID_HOST_OS_ARCH)-$(HOST_ARCH) 258 endif 259 260 ifndef ANDROID_ARCH 261 ANDROID_ARCH := armeabi-v7a 262 endif 263 264 ifeq ($(ANDROID_ARCH),arm64-v8a) 265 TOOLCHAIN := aarch64-linux-android-4.9 266 SYSROOT_ARCH := arm64 267 BIN_PREFIX := aarch64-linux-android 268 MARCH_OPTION := 269 endif 270 ifeq ($(ANDROID_ARCH),armeabi) 271 TOOLCHAIN := arm-linux-androideabi-4.9 272 SYSROOT_ARCH := arm 273 BIN_PREFIX := arm-linux-androideabi 274 MARCH_OPTION := 275 endif 276 ifeq ($(ANDROID_ARCH),armeabi-v7a) 277 TOOLCHAIN := arm-linux-androideabi-4.9 278 SYSROOT_ARCH := arm 279 BIN_PREFIX := arm-linux-androideabi 280 MARCH_OPTION := -march=armv7-a -mfloat-abi=softfp -mfpu=neon 281 endif 282 ifeq ($(ANDROID_ARCH),mips) 283 TOOLCHAIN := mipsel-linux-android-4.9 284 SYSROOT_ARCH := mips 285 BIN_PREFIX := mipsel-linux-android 286 MARCH_OPTION := 287 endif 288 ifeq ($(ANDROID_ARCH),mips64) 289 TOOLCHAIN := mips64el-linux-android-4.9 290 SYSROOT_ARCH := mips64 291 BIN_PREFIX := mips64el-linux-android 292 MARCH_OPTION := 293 endif 294 ifeq ($(ANDROID_ARCH),x86) 295 TOOLCHAIN := x86-4.9 296 SYSROOT_ARCH := x86 297 BIN_PREFIX := i686-linux-android 298 MARCH_OPTION := 299 endif 300 ifeq ($(ANDROID_ARCH),x86_64) 301 TOOLCHAIN := x86_64-4.9 302 SYSROOT_ARCH := x86_64 303 BIN_PREFIX := x86_64-linux-android 304 MARCH_OPTION := 305 endif 306 307 ifndef NDK_ROOT 308 $(error "NDK_ROOT is not defined.") 309 endif 310 CXX := $(CC_PREFIX) $(NDK_ROOT)/toolchains/$(TOOLCHAIN)/prebuilt/$(ANDROID_HOST_OS_ARCH)/bin/$(BIN_PREFIX)-g++ 311 CC := $(CC_PREFIX) $(NDK_ROOT)/toolchains/$(TOOLCHAIN)/prebuilt/$(ANDROID_HOST_OS_ARCH)/bin/$(BIN_PREFIX)-gcc 312 CXXFLAGS +=\ 313 --sysroot $(NDK_ROOT)/platforms/android-21/arch-$(SYSROOT_ARCH) \ 314 -Wno-narrowing \ 315 -fomit-frame-pointer \ 316 $(MARCH_OPTION) \ 317 -fPIE \ 318 -fPIC 319 INCLUDES = \ 320 -I$(NDK_ROOT)/sources/android/support/include \ 321 -I$(NDK_ROOT)/sources/cxx-stl/gnu-libstdc++/4.9/include \ 322 -I$(NDK_ROOT)/sources/cxx-stl/gnu-libstdc++/4.9/libs/$(ANDROID_ARCH)/include \ 323 -I. \ 324 -I$(MAKEFILE_DIR)/downloads/ \ 325 -I$(MAKEFILE_DIR)/downloads/eigen \ 326 -I$(MAKEFILE_DIR)/downloads/gemmlowp \ 327 -I$(MAKEFILE_DIR)/downloads/nsync/public \ 328 -I$(MAKEFILE_DIR)/downloads/fft2d \ 329 -I$(MAKEFILE_DIR)/gen/protobuf_android/$(ANDROID_ARCH)/include \ 330 -I$(PROTOGENDIR) \ 331 -I$(PBTGENDIR) 332 333 LIBS := \ 334 $(TARGET_NSYNC_LIB) \ 335 -lgnustl_static \ 336 -lprotobuf \ 337 -llog \ 338 -lz \ 339 -lm \ 340 -ldl \ 341 -latomic 342 343 LD := $(NDK_ROOT)/toolchains/$(TOOLCHAIN)/prebuilt/$(ANDROID_HOST_OS_ARCH)/$(BIN_PREFIX)/bin/ld 344 345 LDFLAGS := \ 346 $(MARCH_OPTION) \ 347 -L$(MAKEFILE_DIR)/gen/protobuf_android/$(ANDROID_ARCH)/lib \ 348 -L$(NDK_ROOT)/sources/cxx-stl/gnu-libstdc++/4.9/libs/$(ANDROID_ARCH) \ 349 -fPIE \ 350 -pie \ 351 -v 352 353 AR := $(NDK_ROOT)/toolchains/$(TOOLCHAIN)/prebuilt/$(ANDROID_HOST_OS_ARCH)/bin/$(BIN_PREFIX)-ar 354 ARFLAGS := r 355 LIBFLAGS += -Wl,--allow-multiple-definition -Wl,--whole-archive 356 357 ifdef HEXAGON_LIBS 358 INCLUDES += -I$(HEXAGON_INCLUDE) 359 LIBS += -lhexagon_controller 360 LDFLAGS += -L$(HEXAGON_LIBS) 361 CXXFLAGS += -DUSE_HEXAGON_LIBS 362 363 # CAVEAT: We should disable TENSORFLOW_DISABLE_META while running 364 # quantized_matmul on Android because it crashes in 365 # MultiThreadGemm in tensorflow/core/kernels/meta_support.cc 366 # See http://b/33270149 367 # TODO(satok): Remove once it's fixed 368 CXXFLAGS += -DTENSORFLOW_DISABLE_META 369 370 # Declare __ANDROID_TYPES_FULL__ to enable required types for hvx 371 CXXFLAGS += -D__ANDROID_TYPES_FULL__ 372 endif 373 374 ifdef ENABLE_EXPERIMENTAL_HEXNN_OPS 375 CXXFLAGS += -DENABLE_EXPERIMENTAL_HEXNN_OPS 376 endif 377 378 ifeq ($(BUILD_FOR_TEGRA),1) 379 NVCC := $(JETPACK)/cuda/bin/nvcc 380 NVCCFLAGS := -x=cu -D__CUDACC__ -DNVCC -DANDROID_TEGRA -ccbin $(NDK_ROOT)/toolchains/$(TOOLCHAIN)/prebuilt/$(ANDROID_HOST_OS_ARCH)/bin/$(BIN_PREFIX)-g++ --std c++11 --expt-relaxed-constexpr -m64 -gencode arch=compute_53,\"code=sm_53\" -gencode arch=compute_62,\"code=sm_62\" -DEIGEN_AVOID_STL_ARRAY -DTENSORFLOW_USE_EIGEN_THREADPOOL -DLANG_CXX11 -DEIGEN_HAS_C99_MATH -DGOOGLE_CUDA=1 -DTF_EXTRA_CUDA_CAPABILITIES=5.3 381 CXXFLAGS4NVCC =\ 382 -DIS_SLIM_BUILD \ 383 -DANDROID_TEGRA \ 384 -fno-exceptions \ 385 -DNDEBUG $(OPTFLAGS) \ 386 -march=armv8-a \ 387 -fPIE \ 388 -D__ANDROID_TYPES_FULL__ \ 389 --sysroot $(NDK_ROOT)/platforms/android-21/arch-arm64 390 391 CXXFLAGS +=\ 392 -DGOOGLE_CUDA=1 \ 393 -D__ANDROID_TYPES_FULL__ \ 394 -DANDROID_TEGRA \ 395 -DEIGEN_AVOID_STL_ARRAY \ 396 -DEIGEN_HAS_C99_MATH \ 397 -DLANG_CXX11 -DTENSORFLOW_USE_EIGEN_THREADPOOL -DTF_EXTRA_CUDA_CAPABILITIES=5.3 398 399 INCLUDES += \ 400 -Itensorflow/core/kernels \ 401 -I$(MAKEFILE_DIR)/downloads/cub \ 402 -I$(MAKEFILE_DIR)/downloads/cub/cub_archive/cub/device \ 403 -Ithird_party/toolchains/gpus/cuda \ 404 -I$(JETPACK)/cuda/include \ 405 -I$(JETPACK) \ 406 -I$(JETPACK)/cuDNN/aarch64 \ 407 -I$(JETPACK)/cuda/extras/CUPTI/include 408 409 410 CUDA_LIBS := \ 411 -ltfcuda \ 412 -lcudart_static \ 413 -lcudnn \ 414 -lcublas_static \ 415 -lcufftw_static \ 416 -lcusolver_static \ 417 -lcusparse_static \ 418 -lcufft \ 419 -lcuda \ 420 -lculibos \ 421 -lcurand_static 422 423 OBJDIR := $(OBJDIR)android_arm64-v8a/ 424 LIBDIR := $(LIBDIR)android_arm64-v8a/ 425 BINDIR := $(BINDIR)android_arm64-v8a/ 426 DEPDIR := $(DEPDIR)android_arm64-v8a/ 427 428 TEGRA_LIBS := \ 429 -L$(JETPACK)/cuda/targets/aarch64-linux-androideabi/lib \ 430 -L$(JETPACK)/cuda/targets/aarch64-linux-androideabi/lib/stubs \ 431 -L$(JETPACK)/cuda/targets/aarch64-linux-androideabi/lib64 \ 432 -L$(JETPACK)/cuda/targets/aarch64-linux-androideabi/lib64/stubs \ 433 -L$(JETPACK)/cuDNN/aarch64/cuda/lib64 \ 434 -L$(LIBDIR) 435 436 CUDA_LIB_DEPS := $(LIBDIR)libtfcuda.a 437 else 438 OBJDIR := $(OBJDIR)android_$(ANDROID_ARCH)/ 439 LIBDIR := $(LIBDIR)android_$(ANDROID_ARCH)/ 440 BINDIR := $(BINDIR)android_$(ANDROID_ARCH)/ 441 DEPDIR := $(DEPDIR)android_$(ANDROID_ARCH)/ 442 endif # ifeq ($(BUILD_FOR_TEGRA),1) 443 endif # ANDROID 444 # LINT.ThenChange(//tensorflow/contrib/android/cmake/CMakeLists.txt) 445 446 # Settings for iOS. 447 ifeq ($(TARGET),IOS) 448 IPHONEOS_PLATFORM := $(shell xcrun --sdk iphoneos --show-sdk-platform-path) 449 IPHONEOS_SYSROOT := $(shell xcrun --sdk iphoneos --show-sdk-path) 450 IPHONESIMULATOR_PLATFORM := $(shell xcrun --sdk iphonesimulator \ 451 --show-sdk-platform-path) 452 IPHONESIMULATOR_SYSROOT := $(shell xcrun --sdk iphonesimulator \ 453 --show-sdk-path) 454 IOS_SDK_VERSION := $(shell xcrun --sdk iphoneos --show-sdk-version) 455 MIN_SDK_VERSION := 9.0 456 # Override IOS_ARCH with ARMV7, ARMV7S, ARM64, or I386. 457 IOS_ARCH := X86_64 458 ifeq ($(IOS_ARCH),ARMV7) 459 CXXFLAGS += -miphoneos-version-min=$(MIN_SDK_VERSION) \ 460 -arch armv7 \ 461 -fembed-bitcode \ 462 -D__thread=thread_local \ 463 -DUSE_GEMM_FOR_CONV \ 464 -Wno-c++11-narrowing \ 465 -mno-thumb \ 466 -DTF_LEAN_BINARY \ 467 $(ANDROID_TYPES) \ 468 -fno-exceptions \ 469 -isysroot \ 470 ${IPHONEOS_SYSROOT} 471 LDFLAGS := -arch armv7 \ 472 -fembed-bitcode \ 473 -miphoneos-version-min=${MIN_SDK_VERSION} \ 474 -framework Accelerate \ 475 -Xlinker -S \ 476 -Xlinker -x \ 477 -Xlinker -dead_strip \ 478 -all_load \ 479 -L$(GENDIR)protobuf_ios/lib \ 480 -lz 481 endif 482 ifeq ($(IOS_ARCH),ARMV7S) 483 CXXFLAGS += -miphoneos-version-min=$(MIN_SDK_VERSION) \ 484 -arch armv7s \ 485 -fembed-bitcode \ 486 -D__thread=thread_local \ 487 -DUSE_GEMM_FOR_CONV \ 488 -Wno-c++11-narrowing \ 489 -mno-thumb \ 490 -DTF_LEAN_BINARY \ 491 $(ANDROID_TYPES) \ 492 -fno-exceptions \ 493 -isysroot \ 494 ${IPHONEOS_SYSROOT} 495 LDFLAGS := -arch armv7s \ 496 -fembed-bitcode \ 497 -miphoneos-version-min=${MIN_SDK_VERSION} \ 498 -framework Accelerate \ 499 -Xlinker -S \ 500 -Xlinker -x \ 501 -Xlinker -dead_strip \ 502 -all_load \ 503 -L$(GENDIR)protobuf_ios/lib \ 504 -lz 505 endif 506 ifeq ($(IOS_ARCH),ARM64) 507 CXXFLAGS += -miphoneos-version-min=$(MIN_SDK_VERSION) \ 508 -arch arm64 \ 509 -fembed-bitcode \ 510 -D__thread=thread_local \ 511 -DUSE_GEMM_FOR_CONV \ 512 -Wno-c++11-narrowing \ 513 -DTF_LEAN_BINARY \ 514 $(ANDROID_TYPES) \ 515 -fno-exceptions \ 516 -isysroot \ 517 ${IPHONEOS_SYSROOT} 518 LDFLAGS := -arch arm64 \ 519 -fembed-bitcode \ 520 -miphoneos-version-min=${MIN_SDK_VERSION} \ 521 -framework Accelerate \ 522 -Xlinker -S \ 523 -Xlinker -x \ 524 -Xlinker -dead_strip \ 525 -all_load \ 526 -L$(GENDIR)protobuf_ios/lib \ 527 -lz 528 endif 529 ifeq ($(IOS_ARCH),I386) 530 CXXFLAGS += -mios-simulator-version-min=$(MIN_SDK_VERSION) \ 531 -arch i386 \ 532 -mno-sse \ 533 -fembed-bitcode \ 534 -D__thread=thread_local \ 535 -DUSE_GEMM_FOR_CONV \ 536 -Wno-c++11-narrowing \ 537 -DTF_LEAN_BINARY \ 538 $(ANDROID_TYPES) \ 539 -fno-exceptions \ 540 -isysroot \ 541 ${IPHONESIMULATOR_SYSROOT} 542 LDFLAGS := -arch i386 \ 543 -fembed-bitcode \ 544 -mios-simulator-version-min=${MIN_SDK_VERSION} \ 545 -framework Accelerate \ 546 -Xlinker -S \ 547 -Xlinker -x \ 548 -Xlinker -dead_strip \ 549 -all_load \ 550 -L$(GENDIR)protobuf_ios/lib \ 551 -lz 552 endif 553 ifeq ($(IOS_ARCH),X86_64) 554 CXXFLAGS += -mios-simulator-version-min=$(MIN_SDK_VERSION) \ 555 -arch x86_64 \ 556 -fembed-bitcode \ 557 -D__thread=thread_local \ 558 -DUSE_GEMM_FOR_CONV \ 559 -Wno-c++11-narrowing \ 560 -DTF_LEAN_BINARY \ 561 $(ANDROID_TYPES) \ 562 -fno-exceptions \ 563 -isysroot \ 564 ${IPHONESIMULATOR_SYSROOT} 565 LDFLAGS := -arch x86_64 \ 566 -fembed-bitcode \ 567 -mios-simulator-version-min=${MIN_SDK_VERSION} \ 568 -framework Accelerate \ 569 -Xlinker -S \ 570 -Xlinker -x \ 571 -Xlinker -dead_strip \ 572 -all_load \ 573 -L$(GENDIR)protobuf_ios/lib \ 574 -lz 575 endif 576 OBJDIR := $(OBJDIR)ios_$(IOS_ARCH)/ 577 LIBDIR := $(LIBDIR)ios_$(IOS_ARCH)/ 578 BINDIR := $(BINDIR)ios_$(IOS_ARCH)/ 579 DEPDIR := $(DEPDIR)ios_$(IOS_ARCH)/ 580 endif 581 582 # This library is the main target for this makefile. It will contain a minimal 583 # runtime that can be linked in to other programs. 584 LIB_NAME := libtensorflow-core.a 585 LIB_PATH := $(LIBDIR)$(LIB_NAME) 586 587 # A small example program that shows how to link against the library. 588 BENCHMARK_NAME := $(BINDIR)benchmark 589 590 # What sources we want to compile, derived from the main Bazel build using the 591 # gen_file_lists.sh script. 592 593 CORE_CC_ALL_SRCS := \ 594 $(wildcard tensorflow/core/*.cc) \ 595 $(wildcard tensorflow/core/common_runtime/*.cc) \ 596 $(wildcard tensorflow/core/framework/*.cc) \ 597 $(wildcard tensorflow/core/graph/*.cc) \ 598 $(wildcard tensorflow/core/grappler/*.cc) \ 599 $(wildcard tensorflow/core/grappler/*/*.cc) \ 600 $(wildcard tensorflow/core/lib/*/*.cc) \ 601 $(wildcard tensorflow/core/platform/*.cc) \ 602 $(wildcard tensorflow/core/platform/*/*.cc) \ 603 $(wildcard tensorflow/core/platform/*/*/*.cc) \ 604 $(wildcard tensorflow/core/util/*.cc) \ 605 $(wildcard tensorflow/core/util/*/*.cc) \ 606 tensorflow/core/util/version_info.cc 607 # Remove duplicates (for version_info.cc) 608 CORE_CC_ALL_SRCS := $(sort $(CORE_CC_ALL_SRCS)) 609 610 CORE_CC_EXCLUDE_SRCS_NON_GPU := \ 611 $(wildcard tensorflow/core/*/*test.cc) \ 612 $(wildcard tensorflow/core/*/*testutil*) \ 613 $(wildcard tensorflow/core/*/*testlib*) \ 614 $(wildcard tensorflow/core/*/*main.cc) \ 615 $(wildcard tensorflow/core/*/*/*test.cc) \ 616 $(wildcard tensorflow/core/*/*/*testutil*) \ 617 $(wildcard tensorflow/core/*/*/*testlib*) \ 618 $(wildcard tensorflow/core/*/*/*main.cc) \ 619 $(wildcard tensorflow/core/debug/*.cc) \ 620 $(wildcard tensorflow/core/framework/op_gen_lib.cc) \ 621 $(wildcard tensorflow/core/graph/dot.*) \ 622 $(wildcard tensorflow/core/lib/db/*) \ 623 $(wildcard tensorflow/core/lib/gif/*) \ 624 $(wildcard tensorflow/core/lib/io/zlib*) \ 625 $(wildcard tensorflow/core/lib/io/record*) \ 626 $(wildcard tensorflow/core/lib/jpeg/*) \ 627 $(wildcard tensorflow/core/lib/png/*) \ 628 $(wildcard tensorflow/core/util/events_writer.*) \ 629 $(wildcard tensorflow/core/util/reporter.*) \ 630 $(wildcard tensorflow/core/platform/default/test_benchmark.*) \ 631 $(wildcard tensorflow/core/platform/cloud/*) \ 632 $(wildcard tensorflow/core/platform/google/*) \ 633 $(wildcard tensorflow/core/platform/google/*/*) \ 634 $(wildcard tensorflow/core/platform/jpeg.*) \ 635 $(wildcard tensorflow/core/platform/png.*) \ 636 $(wildcard tensorflow/core/platform/s3/*) \ 637 $(wildcard tensorflow/core/platform/windows/*) \ 638 $(wildcard tensorflow/core/grappler/inputs/trivial_test_graph_input_yielder.*) \ 639 $(wildcard tensorflow/core/grappler/inputs/file_input_yielder.*) \ 640 $(wildcard tensorflow/core/grappler/clusters/single_machine.*) \ 641 tensorflow/core/util/cuda_kernel_helper_test.cu.cc 642 643 CORE_CC_EXCLUDE_SRCS := \ 644 $(CORE_CC_EXCLUDE_SRCS_NON_GPU) \ 645 $(wildcard tensorflow/core/platform/stream_executor.*) \ 646 $(wildcard tensorflow/core/platform/default/cuda_libdevice_path.*) \ 647 $(wildcard tensorflow/core/platform/cuda.h) \ 648 $(wildcard tensorflow/core/platform/cuda_libdevice_path.*) \ 649 $(wildcard tensorflow/core/user_ops/*.cu.cc) \ 650 $(wildcard tensorflow/core/common_runtime/gpu/*) \ 651 $(wildcard tensorflow/core/common_runtime/gpu_device_factory.*) 652 653 ifeq ($(BUILD_FOR_TEGRA),1) 654 CORE_CC_ALL_SRCS := $(CORE_CC_ALL_SRCS) \ 655 tensorflow/core/kernels/concat_lib_gpu.cc \ 656 tensorflow/core/kernels/cuda_solvers.cc \ 657 tensorflow/core/kernels/cudnn_pooling_gpu.cc \ 658 tensorflow/core/kernels/dense_update_functor.cc \ 659 tensorflow/core/kernels/fractional_avg_pool_op.cc \ 660 tensorflow/core/kernels/fractional_max_pool_op.cc \ 661 tensorflow/core/kernels/fractional_pool_common.cc \ 662 tensorflow/core/kernels/pooling_ops_3d.cc \ 663 tensorflow/core/kernels/sparse_fill_empty_rows_op.cc \ 664 tensorflow/core/kernels/list_kernels.cc \ 665 $(wildcard tensorflow/core/common_runtime/gpu/*.cc) \ 666 $(wildcard tensorflow/stream_executor/*.cc) \ 667 $(wildcard tensorflow/stream_executor/*/*.cc) 668 669 CORE_CC_EXCLUDE_SRCS := \ 670 $(CORE_CC_EXCLUDE_SRCS_NON_GPU) 671 672 CUDA_CC_SRCS := $(wildcard tensorflow/core/kernels/*.cu.cc) 673 CUDA_CC_OBJS := $(addprefix $(OBJDIR), $(CUDA_CC_SRCS:.cc=.o)) 674 endif # TEGRA 675 676 # Filter out all the excluded files. 677 TF_CC_SRCS := $(filter-out $(CORE_CC_EXCLUDE_SRCS), $(CORE_CC_ALL_SRCS)) 678 # Add in any extra files that don't fit the patterns easily 679 TF_CC_SRCS += tensorflow/contrib/makefile/downloads/fft2d/fftsg.c 680 # Also include the op and kernel definitions. 681 TF_CC_SRCS += $(shell cat $(MAKEFILE_DIR)/tf_op_files.txt) 682 PBT_CC_SRCS := $(shell cat $(MAKEFILE_DIR)/tf_pb_text_files.txt) 683 PROTO_SRCS := $(shell cat $(MAKEFILE_DIR)/tf_proto_files.txt) 684 BENCHMARK_SRCS := \ 685 tensorflow/core/util/reporter.cc \ 686 tensorflow/tools/benchmark/benchmark_model.cc \ 687 tensorflow/tools/benchmark/benchmark_model_main.cc 688 689 ifdef HEXAGON_LIBS 690 TF_CC_SRCS += \ 691 tensorflow/cc/framework/scope.cc \ 692 tensorflow/cc/framework/ops.cc \ 693 tensorflow/cc/ops/const_op.cc \ 694 tensorflow/core/kernels/hexagon/graph_transfer_utils.cc \ 695 tensorflow/core/kernels/hexagon/graph_transferer.cc \ 696 tensorflow/core/kernels/hexagon/hexagon_control_wrapper.cc \ 697 tensorflow/core/kernels/hexagon/hexagon_ops_definitions.cc \ 698 tensorflow/core/kernels/hexagon/hexagon_remote_fused_graph_executor_build.cc 699 endif 700 701 # File names of the intermediate files target compilation generates. 702 TF_CC_OBJS := $(addprefix $(OBJDIR), \ 703 $(patsubst %.cc,%.o,$(patsubst %.c,%.o,$(TF_CC_SRCS)))) 704 PBT_GEN_FILES := $(addprefix $(PBTGENDIR), $(PBT_CC_SRCS)) 705 PBT_OBJS := $(addprefix $(OBJDIR), $(PBT_CC_SRCS:.cc=.o)) 706 PROTO_CC_SRCS := $(addprefix $(PROTOGENDIR), $(PROTO_SRCS:.proto=.pb.cc)) 707 PROTO_OBJS := $(addprefix $(OBJDIR), $(PROTO_SRCS:.proto=.pb.o)) 708 LIB_OBJS := $(PROTO_OBJS) $(TF_CC_OBJS) $(PBT_OBJS) 709 BENCHMARK_OBJS := $(addprefix $(OBJDIR), $(BENCHMARK_SRCS:.cc=.o)) 710 711 .PHONY: clean cleantarget 712 713 # The target that's compiled if there's no command-line arguments. 714 all: $(LIB_PATH) $(BENCHMARK_NAME) 715 716 # Rules for target compilation. 717 718 719 .phony_version_info: 720 tensorflow/core/util/version_info.cc: .phony_version_info 721 tensorflow/tools/git/gen_git_source.sh $@ 722 723 # Gathers together all the objects we've compiled into a single '.a' archive. 724 $(LIB_PATH): $(LIB_OBJS) 725 @mkdir -p $(dir $@) 726 $(AR) $(ARFLAGS) $(LIB_PATH) $(LIB_OBJS) 727 728 $(BENCHMARK_NAME): $(BENCHMARK_OBJS) $(LIB_PATH) $(CUDA_LIB_DEPS) 729 @mkdir -p $(dir $@) 730 $(CXX) $(CXXFLAGS) $(INCLUDES) \ 731 -o $(BENCHMARK_NAME) $(BENCHMARK_OBJS) \ 732 $(LIBFLAGS) $(TEGRA_LIBS) $(LIB_PATH) $(LDFLAGS) $(LIBS) $(CUDA_LIBS) 733 734 # NVCC compilation rules for Tegra 735 ifeq ($(BUILD_FOR_TEGRA),1) 736 $(OBJDIR)%.cu.o: %.cu.cc 737 @mkdir -p $(dir $@) 738 @mkdir -p $(dir $(DEPDIR)$*) 739 $(NVCC) $(NVCCFLAGS) -Xcompiler "$(CXXFLAGS4NVCC) $(DEPFLAGS)" $(INCLUDES) -c $< -o $@ 740 741 $(LIBDIR)libtfcuda.a: $(CUDA_CC_OBJS) 742 @mkdir -p $(dir $@) 743 $(AR) $(ARFLAGS) $@ $(CUDA_CC_OBJS) 744 endif 745 746 # Matches on the normal hand-written TensorFlow C++ source files. 747 $(OBJDIR)%.o: %.cc | $(PBT_GEN_FILES) 748 @mkdir -p $(dir $@) 749 @mkdir -p $(dir $(DEPDIR)$*) 750 $(CXX) $(CXXFLAGS) $(DEPFLAGS) $(INCLUDES) -c $< -o $@ 751 @mv -f $(DEPDIR)/$*.Td $(DEPDIR)/$*.d 752 753 # Matches on plain C files. 754 $(OBJDIR)%.o: %.c 755 @mkdir -p $(dir $@) 756 @mkdir -p $(dir $(DEPDIR)$*) 757 $(CXX) $(patsubst --std=c++11,--std=c99, $(CXXFLAGS)) -x c $(DEPFLAGS) \ 758 $(INCLUDES) -c $< -o $@ 759 @mv -f $(DEPDIR)/$*.Td $(DEPDIR)/$*.d 760 761 # Compiles C++ source files that have been generated by protoc. 762 $(OBJDIR)%.pb.o: $(PROTOGENDIR)%.pb.cc 763 @mkdir -p $(dir $@) 764 $(CXX) $(CXXFLAGS) $(INCLUDES) -c $< -o $@ 765 766 # Builds C++ code from proto files using protoc. 767 $(PROTOGENDIR)%.pb.cc $(PROTOGENDIR)%.pb.h: %.proto 768 @mkdir -p $(dir $@) 769 $(PROTOC) $(PROTOCFLAGS) $< --cpp_out $(PROTOGENDIR) 770 771 # Uses proto_text to generate minimal pb_text C++ files from protos. 772 $(PBTGENDIR)%.pb_text.cc $(PBTGENDIR)%.pb_text.h $(PBTGENDIR)%.pb_text-impl.h: %.proto | $(PROTO_TEXT) 773 @mkdir -p $(dir $@) 774 $(PROTO_TEXT) \ 775 $(PBTGENDIR)tensorflow/core \ 776 tensorflow/core/ \ 777 tensorflow/tools/proto_text/placeholder.txt \ 778 $< 779 780 # Compiles the C++ source files created by proto_text. 781 $(OBJDIR)%.pb_text.o: $(PBTGENDIR)%.pb_text.cc 782 @mkdir -p $(dir $@) 783 $(CXX) $(CXXFLAGS) $(INCLUDES) -c $< -o $@ 784 785 # Makes sure that we don't compile the protoc-generated C++ sources before they 786 # and the proto_text files have been created. 787 $(PROTO_OBJS): $(PROTO_CC_SRCS) $(PBT_GEN_FILES) 788 789 # Host compilation rules. 790 791 # For normal manually-created TensorFlow C++ source files. 792 $(HOST_OBJDIR)%.o: %.cc 793 @mkdir -p $(dir $@) 794 $(HOST_CXX) $(HOST_CXXFLAGS) $(HOST_INCLUDES) -c $< -o $@ 795 796 # Compiles object code from protoc-built C++ source files. 797 $(HOST_OBJDIR)%.pb.o: $(HOST_GENDIR)%.pb.cc 798 @mkdir -p $(dir $@) 799 $(HOST_CXX) $(HOST_CXXFLAGS) $(HOST_INCLUDES) -c $< -o $@ 800 801 # Ensures we wait until proto_text has generated the .h files from protos before 802 # we compile the C++. 803 $(PROTO_TEXT_OBJS) : $(PROTO_TEXT_PB_H_FILES) 804 805 # Runs proto_text to generate C++ source files from protos. 806 $(PROTO_TEXT): $(PROTO_TEXT_OBJS) $(PROTO_TEXT_PB_H_FILES) 807 @mkdir -p $(dir $@) 808 $(HOST_CXX) $(HOST_CXXFLAGS) $(HOST_INCLUDES) \ 809 -o $(PROTO_TEXT) $(PROTO_TEXT_OBJS) $(HOST_LDOPTS) $(HOST_LIBS) 810 811 # Compiles the C++ source files from protos using protoc. 812 $(HOST_GENDIR)%.pb.cc $(HOST_GENDIR)%.pb.h: %.proto 813 @mkdir -p $(dir $@) 814 $(PROTOC) $(PROTOCFLAGS) $< --cpp_out $(HOST_GENDIR) 815 816 # Gets rid of all generated files. 817 clean: 818 rm -rf $(MAKEFILE_DIR)/gen 819 rm -rf tensorflow/core/util/version_info.cc 820 821 # Gets rid of all generated files except protobuf libs generated 822 # before calling make. This allows users not to recompile proto libs everytime. 823 clean_except_protobuf_libs: 824 find $(MAKEFILE_DIR)/gen -mindepth 1 -maxdepth 1 ! -name "protobuf*" -exec rm -r "{}" \; 825 rm -rf tensorflow/core/util/version_info.cc 826 827 # Gets rid of target files only, leaving the host alone. Also leaves the lib 828 # directory untouched deliberately, so we can persist multiple architectures 829 # across builds for iOS and Android. 830 cleantarget: 831 rm -rf $(OBJDIR) 832 rm -rf $(BINDIR) 833 rm -rf $(LIBDIR) 834 835 $(DEPDIR)/%.d: ; 836 .PRECIOUS: $(DEPDIR)/%.d 837 838 -include $(patsubst %,$(DEPDIR)/%.d,$(basename $(TF_CC_SRCS))) 839 840 ifdef SUB_MAKEFILES 841 $(warning "include sub makefiles, must not contain white spaces in the path:" $(SUB_MAKEFILES)) 842 include $(SUB_MAKEFILES) 843 endif 844
