KBaba7/llama.cpp
0
1ifndef LLAMA_MAKEFILE2$(error The Makefile build is deprecated. Use the CMake build instead. For more details, see https://github.com/ggerganov/llama.cpp/blob/master/docs/build.md)3endif4 5# Define the default target now so that it is always the first target6BUILD_TARGETS = \7 libllava.a \8 llama-batched \9 llama-batched-bench \10 llama-bench \11 llama-cli \12 llama-convert-llama2c-to-ggml \13 llama-embedding \14 llama-eval-callback \15 llama-export-lora \16 llama-gbnf-validator \17 llama-gguf \18 llama-gguf-hash \19 llama-gguf-split \20 llama-gritlm \21 llama-imatrix \22 llama-infill \23 llama-llava-cli \24 llama-minicpmv-cli\25 llama-qwen2vl-cli\26 llama-lookahead \27 llama-lookup \28 llama-lookup-create \29 llama-lookup-merge \30 llama-lookup-stats \31 llama-parallel \32 llama-passkey \33 llama-perplexity \34 llama-q8dot \35 llama-quantize \36 llama-quantize-stats \37 llama-retrieval \38 llama-save-load-state \39 llama-server \40 llama-simple \41 llama-simple-chat \42 llama-run \43 llama-speculative \44 llama-tokenize \45 llama-vdot \46 llama-cvector-generator \47 llama-gen-docs \48 tests/test-c.o49 50# Binaries only useful for tests51TEST_TARGETS = \52 tests/test-arg-parser \53 tests/test-autorelease \54 tests/test-backend-ops \55 tests/test-chat \56 tests/test-chat-template \57 tests/test-double-float \58 tests/test-grammar-integration \59 tests/test-grammar-parser \60 tests/test-json-schema-to-grammar \61 tests/test-llama-grammar \62 tests/test-log \63 tests/test-model-load-cancel \64 tests/test-quantize-fns \65 tests/test-quantize-perf \66 tests/test-rope \67 tests/test-sampling \68 tests/test-tokenizer-0 \69 tests/test-tokenizer-1-bpe \70 tests/test-tokenizer-1-spm71# tests/test-opt \72 73# Legacy build targets that were renamed in #7809, but should still be removed when the project is cleaned74LEGACY_TARGETS_CLEAN = main quantize quantize-stats perplexity imatrix embedding vdot q8dot convert-llama2c-to-ggml \75 simple batched batched-bench save-load-state server gguf gguf-split eval-callback llama-bench libllava.a llava-cli baby-llama \76 retrieval speculative infill tokenize parallel export-lora lookahead lookup passkey gritlm77 78# Legacy build targets that were renamed in #7809, but we want to build binaries that for them that output a deprecation warning if people try to use them.79# We don't want to clutter things too much, so we only build replacements for the most commonly used binaries.80LEGACY_TARGETS_BUILD = main quantize perplexity embedding server81 82# Deprecation aliases83ifdef LLAMA_CUBLAS84$(error LLAMA_CUBLAS is removed. Use GGML_CUDA instead.)85endif86 87ifdef LLAMA_CUDA88GGML_CUDA := 189DEPRECATE_WARNING := 190endif91 92ifdef LLAMA_KOMPUTE93GGML_KOMPUTE := 194DEPRECATE_WARNING := 195endif96 97ifdef LLAMA_METAL98GGML_METAL := 199DEPRECATE_WARNING := 1100endif101 102ifdef LLAMA_RPC103GGML_RPC := 1104DEPRECATE_WARNING := 1105endif106 107ifdef LLAMA_SYCL108GGML_SYCL := 1109DEPRECATE_WARNING := 1110endif111 112ifdef LLAMA_SYCL_F16113GGML_SYCL_F16 := 1114DEPRECATE_WARNING := 1115endif116 117ifdef LLAMA_OPENBLAS118GGML_OPENBLAS := 1119DEPRECATE_WARNING := 1120endif121 122ifdef LLAMA_OPENBLAS64123GGML_OPENBLAS64 := 1124DEPRECATE_WARNING := 1125endif126 127ifdef LLAMA_BLIS128GGML_BLIS := 1129DEPRECATE_WARNING := 1130endif131 132ifdef LLAMA_NO_LLAMAFILE133GGML_NO_LLAMAFILE := 1134DEPRECATE_WARNING := 1135endif136 137ifdef LLAMA_NO_ACCELERATE138GGML_NO_ACCELERATE := 1139DEPRECATE_WARNING := 1140endif141 142ifdef LLAMA_NO_OPENMP143GGML_NO_OPENMP := 1144DEPRECATE_WARNING := 1145endif146 147ifdef LLAMA_NO_METAL148GGML_NO_METAL := 1149DEPRECATE_WARNING := 1150endif151 152ifdef LLAMA_DISABLE_LOGS153REMOVE_WARNING := 1154endif155 156ifdef LLAMA_SERVER_VERBOSE157REMOVE_WARNING := 1158endif159 160ifndef UNAME_S161UNAME_S := $(shell uname -s)162endif163 164ifndef UNAME_P165UNAME_P := $(shell uname -p)166endif167 168ifndef UNAME_M169UNAME_M := $(shell uname -m)170endif171 172# In GNU make default CXX is g++ instead of c++. Let's fix that so that users173# of non-gcc compilers don't have to provide g++ alias or wrapper.174DEFCC := cc175DEFCXX := c++176ifeq ($(origin CC),default)177CC := $(DEFCC)178endif179ifeq ($(origin CXX),default)180CXX := $(DEFCXX)181endif182 183# Mac OS + Arm can report x86_64184# ref: https://github.com/ggerganov/whisper.cpp/issues/66#issuecomment-1282546789185ifeq ($(UNAME_S),Darwin)186 ifndef GGML_NO_METAL187 GGML_METAL := 1188 endif189 190 GGML_NO_OPENMP := 1191 192 ifneq ($(UNAME_P),arm)193 SYSCTL_M := $(shell sysctl -n hw.optional.arm64 2>/dev/null)194 ifeq ($(SYSCTL_M),1)195 # UNAME_P := arm196 # UNAME_M := arm64197 warn := $(warning Your arch is announced as x86_64, but it seems to actually be ARM64. Not fixing that can lead to bad performance. For more info see: https://github.com/ggerganov/whisper.cpp/issues/66\#issuecomment-1282546789)198 endif199 endif200endif201 202ifdef GGML_METAL203 GGML_METAL_EMBED_LIBRARY := 1204endif205 206ifdef GGML_RPC207 BUILD_TARGETS += rpc-server208endif209 210ifdef GGML_VULKAN211 BUILD_TARGETS += vulkan-shaders-gen212endif213 214default: $(BUILD_TARGETS) $(LEGACY_TARGETS_BUILD)215 216test: $(TEST_TARGETS)217 @failures=0; \218 for test_target in $(TEST_TARGETS); do \219 if [ "$$test_target" = "tests/test-tokenizer-0" ]; then \220 ./$$test_target $(CURDIR)/models/ggml-vocab-llama-spm.gguf; \221 ./$$test_target $(CURDIR)/models/ggml-vocab-llama-bpe.gguf; \222 ./$$test_target $(CURDIR)/models/ggml-vocab-phi-3.gguf; \223 ./$$test_target $(CURDIR)/models/ggml-vocab-falcon.gguf; \224 ./$$test_target $(CURDIR)/models/ggml-vocab-bert-bge.gguf; \225 ./$$test_target $(CURDIR)/models/ggml-vocab-starcoder.gguf; \226 ./$$test_target $(CURDIR)/models/ggml-vocab-gpt-2.gguf; \227 ./$$test_target $(CURDIR)/models/ggml-vocab-refact.gguf; \228 elif [ "$$test_target" = "tests/test-tokenizer-1-spm" ]; then \229 continue; \230 elif [ "$$test_target" = "tests/test-tokenizer-1-bpe" ]; then \231 continue; \232 else \233 echo "Running test $$test_target..."; \234 ./$$test_target; \235 fi; \236 if [ $$? -ne 0 ]; then \237 printf 'Test %s FAILED!\n\n' $$test_target; \238 failures=$$(( failures + 1 )); \239 else \240 printf 'Test %s passed.\n\n' $$test_target; \241 fi; \242 done; \243 if [ $$failures -gt 0 ]; then \244 printf '\n%s tests failed.\n' $$failures; \245 exit 1; \246 fi247 @echo 'All tests passed.'248 249all: $(BUILD_TARGETS) $(TEST_TARGETS) $(LEGACY_TARGETS_BUILD)250 251ifdef RISCV_CROSS_COMPILE252CC := riscv64-unknown-linux-gnu-gcc253CXX := riscv64-unknown-linux-gnu-g++254endif255 256#257# Compile flags258#259 260# keep standard at C11 and C++17261MK_CPPFLAGS = -Iggml/include -Iggml/src -Iinclude -Isrc -Icommon -DGGML_USE_CPU262MK_CFLAGS = -std=c11 -fPIC263MK_CXXFLAGS = -std=c++17 -fPIC264MK_NVCCFLAGS = -std=c++17265 266ifdef LLAMA_NO_CCACHE267GGML_NO_CCACHE := 1268DEPRECATE_WARNING := 1269endif270 271ifndef GGML_NO_CCACHE272CCACHE := $(shell which ccache)273ifdef CCACHE274export CCACHE_SLOPPINESS = time_macros275$(info I ccache found, compilation results will be cached. Disable with GGML_NO_CCACHE.)276CC := $(CCACHE) $(CC)277CXX := $(CCACHE) $(CXX)278else279$(info I ccache not found. Consider installing it for faster compilation.)280endif # CCACHE281endif # GGML_NO_CCACHE282 283# clock_gettime came in POSIX.1b (1993)284# CLOCK_MONOTONIC came in POSIX.1-2001 / SUSv3 as optional285# posix_memalign came in POSIX.1-2001 / SUSv3286# M_PI is an XSI extension since POSIX.1-2001 / SUSv3, came in XPG1 (1985)287MK_CPPFLAGS += -D_XOPEN_SOURCE=600288 289# Somehow in OpenBSD whenever POSIX conformance is specified290# some string functions rely on locale_t availability,291# which was introduced in POSIX.1-2008, forcing us to go higher292ifeq ($(UNAME_S),OpenBSD)293 MK_CPPFLAGS += -U_XOPEN_SOURCE -D_XOPEN_SOURCE=700294endif295 296# Data types, macros and functions related to controlling CPU affinity and297# some memory allocation are available on Linux through GNU extensions in libc298ifeq ($(UNAME_S),Linux)299 MK_CPPFLAGS += -D_GNU_SOURCE300 MK_LDFLAGS += -ldl301endif302 303# RLIMIT_MEMLOCK came in BSD, is not specified in POSIX.1,304# and on macOS its availability depends on enabling Darwin extensions305# similarly on DragonFly, enabling BSD extensions is necessary306ifeq ($(UNAME_S),Darwin)307 MK_CPPFLAGS += -D_DARWIN_C_SOURCE308endif309ifeq ($(UNAME_S),DragonFly)310 MK_CPPFLAGS += -D__BSD_VISIBLE311endif312 313# alloca is a non-standard interface that is not visible on BSDs when314# POSIX conformance is specified, but not all of them provide a clean way315# to enable it in such cases316ifeq ($(UNAME_S),FreeBSD)317 MK_CPPFLAGS += -D__BSD_VISIBLE318endif319ifeq ($(UNAME_S),NetBSD)320 MK_CPPFLAGS += -D_NETBSD_SOURCE321endif322ifeq ($(UNAME_S),OpenBSD)323 MK_CPPFLAGS += -D_BSD_SOURCE324endif325 326ifdef GGML_SCHED_MAX_COPIES327 MK_CPPFLAGS += -DGGML_SCHED_MAX_COPIES=$(GGML_SCHED_MAX_COPIES)328endif329 330ifdef LLAMA_DEBUG331 MK_CFLAGS += -O0 -g332 MK_CXXFLAGS += -O0 -g333 MK_LDFLAGS += -g334 MK_NVCCFLAGS += -O0 -g335 336 ifeq ($(UNAME_S),Linux)337 MK_CPPFLAGS += -D_GLIBCXX_ASSERTIONS338 endif339else340 MK_CPPFLAGS += -DNDEBUG341 MK_CFLAGS += -O3 -g342 MK_CXXFLAGS += -O3 -g343 MK_NVCCFLAGS += -O3 -g344endif345 346ifdef LLAMA_SANITIZE_THREAD347 MK_CFLAGS += -fsanitize=thread -g348 MK_CXXFLAGS += -fsanitize=thread -g349 MK_LDFLAGS += -fsanitize=thread -g350endif351 352ifdef LLAMA_SANITIZE_ADDRESS353 MK_CFLAGS += -fsanitize=address -fno-omit-frame-pointer -g354 MK_CXXFLAGS += -fsanitize=address -fno-omit-frame-pointer -g355 MK_LDFLAGS += -fsanitize=address -fno-omit-frame-pointer -g356endif357 358ifdef LLAMA_SANITIZE_UNDEFINED359 MK_CFLAGS += -fsanitize=undefined -g360 MK_CXXFLAGS += -fsanitize=undefined -g361 MK_LDFLAGS += -fsanitize=undefined -g362endif363 364ifdef LLAMA_SERVER_SSL365 MK_CPPFLAGS += -DCPPHTTPLIB_OPENSSL_SUPPORT366 MK_LDFLAGS += -lssl -lcrypto367endif368 369ifndef GGML_NO_CPU_AARCH64370 MK_CPPFLAGS += -DGGML_USE_CPU_AARCH64371endif372 373# warnings374WARN_FLAGS = \375 -Wall \376 -Wextra \377 -Wpedantic \378 -Wcast-qual \379 -Wno-unused-function380 381MK_CFLAGS += \382 $(WARN_FLAGS) \383 -Wshadow \384 -Wstrict-prototypes \385 -Wpointer-arith \386 -Wmissing-prototypes \387 -Werror=implicit-int \388 -Werror=implicit-function-declaration389 390MK_CXXFLAGS += \391 $(WARN_FLAGS) \392 -Wmissing-declarations \393 -Wmissing-noreturn394 395ifeq ($(LLAMA_FATAL_WARNINGS),1)396 MK_CFLAGS += -Werror397 MK_CXXFLAGS += -Werror398endif399 400# this version of Apple ld64 is buggy401ifneq '' '$(findstring dyld-1015.7,$(shell $(CC) $(LDFLAGS) -Wl,-v 2>&1))'402 MK_CPPFLAGS += -DHAVE_BUGGY_APPLE_LINKER403endif404 405# OS specific406# TODO: support Windows407ifneq '' '$(filter $(UNAME_S),Linux Darwin FreeBSD NetBSD OpenBSD Haiku)'408 MK_CFLAGS += -pthread409 MK_CXXFLAGS += -pthread410endif411 412# detect Windows413ifneq ($(findstring _NT,$(UNAME_S)),)414 _WIN32 := 1415endif416 417# library name prefix418ifneq ($(_WIN32),1)419 LIB_PRE := lib420endif421 422# Dynamic Shared Object extension423ifneq ($(_WIN32),1)424 DSO_EXT := .so425else426 DSO_EXT := .dll427endif428 429# Windows Sockets 2 (Winsock) for network-capable apps430ifeq ($(_WIN32),1)431 LWINSOCK2 := -lws2_32432endif433 434ifdef LLAMA_GPROF435 MK_CFLAGS += -pg436 MK_CXXFLAGS += -pg437endif438 439# Architecture specific440# TODO: probably these flags need to be tweaked on some architectures441# feel free to update the Makefile for your architecture and send a pull request or issue442 443ifndef RISCV_CROSS_COMPILE444 445ifeq ($(UNAME_M),$(filter $(UNAME_M),x86_64 i686 amd64))446 # Use all CPU extensions that are available:447 MK_CFLAGS += -march=native -mtune=native448 HOST_CXXFLAGS += -march=native -mtune=native449 450 # Usage AMX build test451 #MK_CFLAGS += -march=graniterapids -mtune=graniterapids452 #HOST_CXXFLAGS += -march=graniterapids -mtune=graniterapids453 454 # Usage AVX-only455 #MK_CFLAGS += -mfma -mf16c -mavx456 #MK_CXXFLAGS += -mfma -mf16c -mavx457 458 # Usage SSSE3-only (Not is SSE3!)459 #MK_CFLAGS += -mssse3460 #MK_CXXFLAGS += -mssse3461endif462 463ifneq '' '$(findstring mingw,$(shell $(CC) -dumpmachine))'464 # The stack is only 16-byte aligned on Windows, so don't let gcc emit aligned moves.465 # https://gcc.gnu.org/bugzilla/show_bug.cgi?id=54412466 # https://github.com/ggerganov/llama.cpp/issues/2922467 MK_CFLAGS += -Xassembler -muse-unaligned-vector-move468 MK_CXXFLAGS += -Xassembler -muse-unaligned-vector-move469 470 # Target Windows 8 for PrefetchVirtualMemory471 MK_CPPFLAGS += -D_WIN32_WINNT=0x602472endif473 474ifneq ($(filter aarch64%,$(UNAME_M)),)475 # Apple M1, M2, etc.476 # Raspberry Pi 3, 4, Zero 2 (64-bit)477 # Nvidia Jetson478 MK_CFLAGS += -mcpu=native479 MK_CXXFLAGS += -mcpu=native480 JETSON_RELEASE_INFO = $(shell jetson_release)481 ifdef JETSON_RELEASE_INFO482 ifneq ($(filter TX2%,$(JETSON_RELEASE_INFO)),)483 JETSON_EOL_MODULE_DETECT = 1484 CC = aarch64-unknown-linux-gnu-gcc485 cxx = aarch64-unknown-linux-gnu-g++486 endif487 endif488endif489 490ifneq ($(filter armv6%,$(UNAME_M)),)491 # Raspberry Pi 1, Zero492 MK_CFLAGS += -mfpu=neon-fp-armv8 -mfp16-format=ieee -mno-unaligned-access493 MK_CXXFLAGS += -mfpu=neon-fp-armv8 -mfp16-format=ieee -mno-unaligned-access494endif495 496ifneq ($(filter armv7%,$(UNAME_M)),)497 # Raspberry Pi 2498 MK_CFLAGS += -mfpu=neon-fp-armv8 -mfp16-format=ieee -mno-unaligned-access -funsafe-math-optimizations499 MK_CXXFLAGS += -mfpu=neon-fp-armv8 -mfp16-format=ieee -mno-unaligned-access -funsafe-math-optimizations500endif501 502ifneq ($(filter armv8%,$(UNAME_M)),)503 # Raspberry Pi 3, 4, Zero 2 (32-bit)504 MK_CFLAGS += -mfp16-format=ieee -mno-unaligned-access505 MK_CXXFLAGS += -mfp16-format=ieee -mno-unaligned-access506endif507 508ifneq ($(filter ppc64%,$(UNAME_M)),)509 POWER9_M := $(shell grep "POWER9" /proc/cpuinfo)510 ifneq (,$(findstring POWER9,$(POWER9_M)))511 MK_CFLAGS += -mcpu=power9512 MK_CXXFLAGS += -mcpu=power9513 endif514endif515 516ifneq ($(filter ppc64le%,$(UNAME_M)),)517 MK_CFLAGS += -mcpu=powerpc64le518 MK_CXXFLAGS += -mcpu=powerpc64le519 CUDA_POWER_ARCH = 1520endif521 522ifneq ($(filter loongarch64%,$(UNAME_M)),)523 MK_CFLAGS += -mlasx524 MK_CXXFLAGS += -mlasx525endif526 527ifneq ($(filter riscv64%,$(UNAME_M)),)528 MK_CFLAGS += -march=rv64gcv -mabi=lp64d529 MK_CXXFLAGS += -march=rv64gcv -mabi=lp64d530endif531 532else # RISC-V CROSS COMPILATION533 MK_CFLAGS += -march=rv64gcv -mabi=lp64d534 MK_CXXFLAGS += -march=rv64gcv -mabi=lp64d535endif536 537ifndef GGML_NO_ACCELERATE538 # Mac OS - include Accelerate framework.539 # `-framework Accelerate` works both with Apple Silicon and Mac Intel540 ifeq ($(UNAME_S),Darwin)541 MK_CPPFLAGS += -DGGML_USE_ACCELERATE -DGGML_USE_BLAS -DGGML_BLAS_USE_ACCELERATE542 MK_CPPFLAGS += -DACCELERATE_NEW_LAPACK543 MK_CPPFLAGS += -DACCELERATE_LAPACK_ILP64544 MK_LDFLAGS += -framework Accelerate545 OBJ_GGML_EXT += ggml/src/ggml-blas/ggml-blas.o546 endif547endif # GGML_NO_ACCELERATE548 549ifndef GGML_NO_OPENMP550 MK_CPPFLAGS += -DGGML_USE_OPENMP551 MK_CFLAGS += -fopenmp552 MK_CXXFLAGS += -fopenmp553endif # GGML_NO_OPENMP554 555ifdef GGML_OPENBLAS556 MK_CPPFLAGS += -DGGML_USE_BLAS $(shell pkg-config --cflags-only-I openblas)557 MK_CFLAGS += $(shell pkg-config --cflags-only-other openblas)558 MK_LDFLAGS += $(shell pkg-config --libs openblas)559 OBJ_GGML_EXT += ggml/src/ggml-blas/ggml-blas.o560endif # GGML_OPENBLAS561 562ifdef GGML_OPENBLAS64563 MK_CPPFLAGS += -DGGML_USE_BLAS $(shell pkg-config --cflags-only-I openblas64)564 MK_CFLAGS += $(shell pkg-config --cflags-only-other openblas64)565 MK_LDFLAGS += $(shell pkg-config --libs openblas64)566 OBJ_GGML_EXT += ggml/src/ggml-blas/ggml-blas.o567endif # GGML_OPENBLAS64568 569ifdef GGML_BLIS570 MK_CPPFLAGS += -DGGML_USE_BLAS -DGGML_BLAS_USE_BLIS -I/usr/local/include/blis -I/usr/include/blis571 MK_LDFLAGS += -lblis -L/usr/local/lib572 OBJ_GGML_EXT += ggml/src/ggml-blas/ggml-blas.o573endif # GGML_BLIS574 575ifdef GGML_NVPL576 MK_CPPFLAGS += -DGGML_USE_BLAS -DGGML_BLAS_USE_NVPL -DNVPL_ILP64 -I/usr/local/include/nvpl_blas -I/usr/include/nvpl_blas577 MK_LDFLAGS += -L/usr/local/lib -lnvpl_blas_core -lnvpl_blas_ilp64_gomp578 OBJ_GGML_EXT += ggml/src/ggml-blas/ggml-blas.o579endif # GGML_NVPL580 581ifndef GGML_NO_LLAMAFILE582 MK_CPPFLAGS += -DGGML_USE_LLAMAFILE583 OBJ_GGML_EXT += ggml/src/ggml-cpu/llamafile/sgemm.o584endif585 586ifndef GGML_NO_AMX587 MK_CPPFLAGS += -DGGML_USE_AMX588 OBJ_GGML_EXT += ggml/src/ggml-cpu/amx/amx.o ggml/src/ggml-cpu/amx/mmq.o589endif590 591# only necessary for the CPU backend files592MK_CPPFLAGS += -Iggml/src/ggml-cpu593 594ifdef GGML_RPC595 MK_CPPFLAGS += -DGGML_USE_RPC596 OBJ_GGML_EXT += ggml/src/ggml-rpc.o597endif # GGML_RPC598 599OBJ_CUDA_TMPL = $(patsubst %.cu,%.o,$(wildcard ggml/src/ggml-cuda/template-instances/fattn-mma*.cu))600OBJ_CUDA_TMPL += $(patsubst %.cu,%.o,$(wildcard ggml/src/ggml-cuda/template-instances/mmq*.cu))601 602ifdef GGML_CUDA_FA_ALL_QUANTS603 OBJ_CUDA_TMPL += $(patsubst %.cu,%.o,$(wildcard ggml/src/ggml-cuda/template-instances/fattn-vec*.cu))604else605 OBJ_CUDA_TMPL += $(patsubst %.cu,%.o,$(wildcard ggml/src/ggml-cuda/template-instances/fattn-vec*q4_0-q4_0.cu))606 OBJ_CUDA_TMPL += $(patsubst %.cu,%.o,$(wildcard ggml/src/ggml-cuda/template-instances/fattn-vec*q8_0-q8_0.cu))607 OBJ_CUDA_TMPL += $(patsubst %.cu,%.o,$(wildcard ggml/src/ggml-cuda/template-instances/fattn-vec*f16-f16.cu))608endif # GGML_CUDA_FA_ALL_QUANTS609 610ifdef GGML_CUDA611 ifneq ('', '$(wildcard /opt/cuda)')612 CUDA_PATH ?= /opt/cuda613 else614 CUDA_PATH ?= /usr/local/cuda615 endif616 617 MK_CPPFLAGS += -DGGML_USE_CUDA -DGGML_CUDA_USE_GRAPHS -I$(CUDA_PATH)/include -I$(CUDA_PATH)/targets/$(UNAME_M)-linux/include618 MK_LDFLAGS += -lcuda -lcublas -lculibos -lcudart -lcublasLt -lpthread -ldl -lrt -L$(CUDA_PATH)/lib64 -L/usr/lib64 -L$(CUDA_PATH)/targets/$(UNAME_M)-linux/lib -L$(CUDA_PATH)/lib64/stubs -L/usr/lib/wsl/lib619 MK_NVCCFLAGS += -use_fast_math620 621 OBJ_GGML_EXT += ggml/src/ggml-cuda/ggml-cuda.o622 OBJ_GGML_EXT += $(patsubst %.cu,%.o,$(wildcard ggml/src/ggml-cuda/*.cu))623 OBJ_GGML_EXT += $(OBJ_CUDA_TMPL)624 625ifdef LLAMA_FATAL_WARNINGS626 MK_NVCCFLAGS += -Werror all-warnings627endif # LLAMA_FATAL_WARNINGS628 629ifndef JETSON_EOL_MODULE_DETECT630 MK_NVCCFLAGS += --forward-unknown-to-host-compiler631endif # JETSON_EOL_MODULE_DETECT632 633ifdef LLAMA_DEBUG634 MK_NVCCFLAGS += -lineinfo635endif # LLAMA_DEBUG636 637ifdef GGML_CUDA_DEBUG638 MK_NVCCFLAGS += --device-debug639endif # GGML_CUDA_DEBUG640 641ifdef GGML_CUDA_NVCC642 NVCC = $(CCACHE) $(GGML_CUDA_NVCC)643else644 NVCC = $(CCACHE) nvcc645endif # GGML_CUDA_NVCC646 647ifdef CUDA_DOCKER_ARCH648 MK_NVCCFLAGS += -Wno-deprecated-gpu-targets -arch=$(CUDA_DOCKER_ARCH)649else ifndef CUDA_POWER_ARCH650 MK_NVCCFLAGS += -arch=native651endif # CUDA_DOCKER_ARCH652 653ifdef GGML_CUDA_FORCE_MMQ654 MK_NVCCFLAGS += -DGGML_CUDA_FORCE_MMQ655endif # GGML_CUDA_FORCE_MMQ656 657ifdef GGML_CUDA_FORCE_CUBLAS658 MK_NVCCFLAGS += -DGGML_CUDA_FORCE_CUBLAS659endif # GGML_CUDA_FORCE_CUBLAS660 661ifdef GGML_CUDA_F16662 MK_NVCCFLAGS += -DGGML_CUDA_F16663endif # GGML_CUDA_F16664 665ifdef GGML_CUDA_DMMV_F16666 MK_NVCCFLAGS += -DGGML_CUDA_F16667endif # GGML_CUDA_DMMV_F16668 669ifdef GGML_CUDA_PEER_MAX_BATCH_SIZE670 MK_NVCCFLAGS += -DGGML_CUDA_PEER_MAX_BATCH_SIZE=$(GGML_CUDA_PEER_MAX_BATCH_SIZE)671else672 MK_NVCCFLAGS += -DGGML_CUDA_PEER_MAX_BATCH_SIZE=128673endif # GGML_CUDA_PEER_MAX_BATCH_SIZE674 675ifdef GGML_CUDA_NO_PEER_COPY676 MK_NVCCFLAGS += -DGGML_CUDA_NO_PEER_COPY677endif # GGML_CUDA_NO_PEER_COPY678 679ifdef GGML_CUDA_CCBIN680 MK_NVCCFLAGS += -ccbin $(GGML_CUDA_CCBIN)681endif # GGML_CUDA_CCBIN682 683ifdef GGML_CUDA_FA_ALL_QUANTS684 MK_NVCCFLAGS += -DGGML_CUDA_FA_ALL_QUANTS685endif # GGML_CUDA_FA_ALL_QUANTS686 687ifdef JETSON_EOL_MODULE_DETECT688define NVCC_COMPILE689 $(NVCC) -I. -Icommon -D_XOPEN_SOURCE=600 -D_GNU_SOURCE -DNDEBUG -DGGML_USE_CUDA -I/usr/local/cuda/include -I/opt/cuda/include -I/usr/local/cuda/targets/aarch64-linux/include -std=c++11 -O3 $(NVCCFLAGS) $(CPPFLAGS) -Xcompiler "$(CUDA_CXXFLAGS)" -c $< -o $@690endef # NVCC_COMPILE691else692define NVCC_COMPILE693 $(NVCC) $(NVCCFLAGS) $(CPPFLAGS) -Xcompiler "$(CUDA_CXXFLAGS)" -c $< -o $@694endef # NVCC_COMPILE695endif # JETSON_EOL_MODULE_DETECT696 697ggml/src/ggml-cuda/%.o: \698 ggml/src/ggml-cuda/%.cu \699 ggml/include/ggml.h \700 ggml/src/ggml-common.h \701 ggml/src/ggml-cuda/common.cuh702 $(NVCC_COMPILE)703 704ggml/src/ggml-cuda/ggml-cuda.o: \705 ggml/src/ggml-cuda/ggml-cuda.cu \706 ggml/include/ggml-cuda.h \707 ggml/include/ggml.h \708 ggml/include/ggml-backend.h \709 ggml/src/ggml-backend-impl.h \710 ggml/src/ggml-common.h \711 $(wildcard ggml/src/ggml-cuda/*.cuh)712 $(NVCC_COMPILE)713endif # GGML_CUDA714 715ifdef GGML_VULKAN716 MK_CPPFLAGS += -DGGML_USE_VULKAN717 MK_LDFLAGS += $(shell pkg-config --libs vulkan)718 OBJ_GGML_EXT += ggml/src/ggml-vulkan.o ggml/src/ggml-vulkan-shaders.o719 720ifdef GGML_VULKAN_CHECK_RESULTS721 MK_CPPFLAGS += -DGGML_VULKAN_CHECK_RESULTS722endif723 724ifdef GGML_VULKAN_DEBUG725 MK_CPPFLAGS += -DGGML_VULKAN_DEBUG726endif727 728ifdef GGML_VULKAN_MEMORY_DEBUG729 MK_CPPFLAGS += -DGGML_VULKAN_MEMORY_DEBUG730endif731 732ifdef GGML_VULKAN_PERF733 MK_CPPFLAGS += -DGGML_VULKAN_PERF734endif735 736ifdef GGML_VULKAN_VALIDATE737 MK_CPPFLAGS += -DGGML_VULKAN_VALIDATE738endif739 740ifdef GGML_VULKAN_RUN_TESTS741 MK_CPPFLAGS += -DGGML_VULKAN_RUN_TESTS742endif743 744GLSLC_CMD = glslc745_ggml_vk_genshaders_cmd = $(shell pwd)/vulkan-shaders-gen746_ggml_vk_header = ggml/src/ggml-vulkan-shaders.hpp747_ggml_vk_source = ggml/src/ggml-vulkan-shaders.cpp748_ggml_vk_input_dir = ggml/src/ggml-vulkan/vulkan-shaders749_ggml_vk_shader_deps = $(echo $(_ggml_vk_input_dir)/*.comp)750 751ggml/src/ggml-vulkan.o: ggml/src/ggml-vulkan/ggml-vulkan.cpp ggml/include/ggml-vulkan.h $(_ggml_vk_header) $(_ggml_vk_source)752 $(CXX) $(CXXFLAGS) $(shell pkg-config --cflags vulkan) -c $< -o $@753 754$(_ggml_vk_header): $(_ggml_vk_source)755 756$(_ggml_vk_source): $(_ggml_vk_shader_deps) vulkan-shaders-gen757 $(_ggml_vk_genshaders_cmd) \758 --glslc $(GLSLC_CMD) \759 --input-dir $(_ggml_vk_input_dir) \760 --target-hpp $(_ggml_vk_header) \761 --target-cpp $(_ggml_vk_source)762 763vulkan-shaders-gen: ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp764 $(CXX) $(CXXFLAGS) -o $@ $(LDFLAGS) ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp765 766endif # GGML_VULKAN767 768ifdef GGML_HIP769 ifeq ($(wildcard /opt/rocm),)770 ROCM_PATH ?= /usr771 AMDGPU_TARGETS ?= $(shell $(shell which amdgpu-arch))772 else773 ROCM_PATH ?= /opt/rocm774 AMDGPU_TARGETS ?= $(shell $(ROCM_PATH)/llvm/bin/amdgpu-arch)775 endif776 777 MK_CPPFLAGS += -DGGML_USE_HIP -DGGML_USE_CUDA778 779ifdef GGML_HIP_UMA780 MK_CPPFLAGS += -DGGML_HIP_UMA781endif # GGML_HIP_UMA782 783 MK_LDFLAGS += -L$(ROCM_PATH)/lib -Wl,-rpath=$(ROCM_PATH)/lib784 MK_LDFLAGS += -L$(ROCM_PATH)/lib64 -Wl,-rpath=$(ROCM_PATH)/lib64785 MK_LDFLAGS += -lhipblas -lamdhip64 -lrocblas786 787 HIPCC ?= $(CCACHE) $(ROCM_PATH)/bin/hipcc788 789 HIPFLAGS += $(addprefix --offload-arch=,$(AMDGPU_TARGETS))790 791ifdef GGML_CUDA_FORCE_MMQ792 HIPFLAGS += -DGGML_CUDA_FORCE_MMQ793endif # GGML_CUDA_FORCE_MMQ794 795ifdef GGML_CUDA_FORCE_CUBLAS796 HIPFLAGS += -DGGML_CUDA_FORCE_CUBLAS797endif # GGML_CUDA_FORCE_CUBLAS798 799ifdef GGML_CUDA_NO_PEER_COPY800 HIPFLAGS += -DGGML_CUDA_NO_PEER_COPY801endif # GGML_CUDA_NO_PEER_COPY802 803 OBJ_GGML_EXT += ggml/src/ggml-cuda/ggml-cuda.o804 OBJ_GGML_EXT += $(patsubst %.cu,%.o,$(wildcard ggml/src/ggml-cuda/*.cu))805 OBJ_GGML_EXT += $(OBJ_CUDA_TMPL)806 807ggml/src/ggml-cuda/ggml-cuda.o: \808 ggml/src/ggml-cuda/ggml-cuda.cu \809 ggml/include/ggml-cuda.h \810 ggml/include/ggml.h \811 ggml/include/ggml-backend.h \812 ggml/src/ggml-backend-impl.h \813 ggml/src/ggml-common.h \814 $(wildcard ggml/src/ggml-cuda/*.cuh)815 $(HIPCC) $(CXXFLAGS) $(HIPFLAGS) -x hip -c -o $@ $<816 817ggml/src/ggml-cuda/%.o: \818 ggml/src/ggml-cuda/%.cu \819 ggml/include/ggml.h \820 ggml/src/ggml-common.h \821 ggml/src/ggml-cuda/common.cuh822 $(HIPCC) $(CXXFLAGS) $(HIPFLAGS) -x hip -c -o $@ $<823endif # GGML_HIP824 825ifdef GGML_MUSA826 ifeq ($(wildcard /opt/musa),)827 MUSA_PATH ?= /usr/local/musa828 else829 MUSA_PATH ?= /opt/musa830 endif831 MUSA_ARCHITECTURES ?= 21;22832 833 MK_CPPFLAGS += -DGGML_USE_MUSA -DGGML_USE_CUDA834 MK_LDFLAGS += -L$(MUSA_PATH)/lib -Wl,-rpath=$(MUSA_PATH)/lib835 MK_LDFLAGS += -lmusa -lmusart -lmublas836 837 ifndef GGML_NO_OPENMP838 # For Ubuntu Focal839 MK_CPPFLAGS += -I/usr/lib/llvm-10/include/openmp840 MK_LDFLAGS += -L/usr/lib/llvm-10/lib841 # For Ubuntu Jammy842 MK_CPPFLAGS += -I/usr/lib/llvm-14/lib/clang/14.0.0/include843 MK_LDFLAGS += -L/usr/lib/llvm-14/lib844 endif # GGML_NO_OPENMP845 846 CC := $(MUSA_PATH)/bin/clang847 CXX := $(MUSA_PATH)/bin/clang++848 MCC := $(CCACHE) $(MUSA_PATH)/bin/mcc849 850 MUSAFLAGS = -x musa -mtgpu851 MUSAFLAGS += $(foreach arch,$(subst ;, ,$(MUSA_ARCHITECTURES)),--cuda-gpu-arch=mp_$(arch))852 853ifdef GGML_CUDA_FORCE_MMQ854 MUSAFLAGS += -DGGML_CUDA_FORCE_MMQ855endif # GGML_CUDA_FORCE_MMQ856 857ifdef GGML_CUDA_FORCE_CUBLAS858 MUSAFLAGS += -DGGML_CUDA_FORCE_CUBLAS859endif # GGML_CUDA_FORCE_CUBLAS860 861ifdef GGML_CUDA_F16862 MUSAFLAGS += -DGGML_CUDA_F16863endif # GGML_CUDA_F16864 865ifdef GGML_CUDA_DMMV_F16866 MUSAFLAGS += -DGGML_CUDA_F16867endif # GGML_CUDA_DMMV_F16868 869ifdef GGML_CUDA_PEER_MAX_BATCH_SIZE870 MUSAFLAGS += -DGGML_CUDA_PEER_MAX_BATCH_SIZE=$(GGML_CUDA_PEER_MAX_BATCH_SIZE)871else872 MUSAFLAGS += -DGGML_CUDA_PEER_MAX_BATCH_SIZE=128873endif # GGML_CUDA_PEER_MAX_BATCH_SIZE874 875ifdef GGML_CUDA_NO_PEER_COPY876 MUSAFLAGS += -DGGML_CUDA_NO_PEER_COPY877endif # GGML_CUDA_NO_PEER_COPY878 879ifdef GGML_CUDA_FA_ALL_QUANTS880 MUSAFLAGS += -DGGML_CUDA_FA_ALL_QUANTS881endif # GGML_CUDA_FA_ALL_QUANTS882 883 OBJ_GGML_EXT += ggml/src/ggml-cuda/ggml-cuda.o884 OBJ_GGML_EXT += $(patsubst %.cu,%.o,$(wildcard ggml/src/ggml-cuda/*.cu))885 OBJ_GGML_EXT += $(OBJ_CUDA_TMPL)886 887ggml/src/ggml-cuda/ggml-cuda.o: \888 ggml/src/ggml-cuda/ggml-cuda.cu \889 ggml/include/ggml-cuda.h \890 ggml/include/ggml.h \891 ggml/include/ggml-backend.h \892 ggml/src/ggml-backend-impl.h \893 ggml/src/ggml-common.h \894 $(wildcard ggml/src/ggml-cuda/*.cuh)895 $(MCC) $(CXXFLAGS) $(MUSAFLAGS) -c -o $@ $<896 897ggml/src/ggml-cuda/%.o: \898 ggml/src/ggml-cuda/%.cu \899 ggml/include/ggml.h \900 ggml/src/ggml-common.h \901 ggml/src/ggml-cuda/common.cuh902 $(MCC) $(CXXFLAGS) $(MUSAFLAGS) -c -o $@ $<903endif # GGML_MUSA904 905ifdef GGML_METAL906 MK_CPPFLAGS += -DGGML_USE_METAL907 MK_LDFLAGS += -framework Foundation -framework Metal -framework MetalKit908 OBJ_GGML_EXT += ggml/src/ggml-metal/ggml-metal.o909 910ifdef GGML_METAL_USE_BF16911 MK_CPPFLAGS += -DGGML_METAL_USE_BF16912endif # GGML_METAL_USE_BF16913ifdef GGML_METAL_NDEBUG914 MK_CPPFLAGS += -DGGML_METAL_NDEBUG915endif916ifdef GGML_METAL_EMBED_LIBRARY917 MK_CPPFLAGS += -DGGML_METAL_EMBED_LIBRARY918 OBJ_GGML_EXT += ggml/src/ggml-metal-embed.o919endif920endif # GGML_METAL921 922ifdef GGML_METAL923ggml/src/ggml-metal/ggml-metal.o: \924 ggml/src/ggml-metal/ggml-metal.m \925 ggml/src/ggml-metal/ggml-metal-impl.h \926 ggml/include/ggml-metal.h \927 ggml/include/ggml.h928 $(CC) $(CFLAGS) -c $< -o $@929 930ifdef GGML_METAL_EMBED_LIBRARY931ggml/src/ggml-metal-embed.o: \932 ggml/src/ggml-metal/ggml-metal.metal \933 ggml/src/ggml-metal/ggml-metal-impl.h \934 ggml/src/ggml-common.h935 @echo "Embedding Metal library"936 @sed -e '/__embed_ggml-common.h__/r ggml/src/ggml-common.h' -e '/__embed_ggml-common.h__/d' < ggml/src/ggml-metal/ggml-metal.metal > ggml/src/ggml-metal/ggml-metal-embed.metal.tmp937 @sed -e '/#include "ggml-metal-impl.h"/r ggml/src/ggml-metal/ggml-metal-impl.h' -e '/#include "ggml-metal-impl.h"/d' < ggml/src/ggml-metal/ggml-metal-embed.metal.tmp > ggml/src/ggml-metal/ggml-metal-embed.metal938 $(eval TEMP_ASSEMBLY=$(shell mktemp -d))939 @echo ".section __DATA, __ggml_metallib" > $(TEMP_ASSEMBLY)/ggml-metal-embed.s940 @echo ".globl _ggml_metallib_start" >> $(TEMP_ASSEMBLY)/ggml-metal-embed.s941 @echo "_ggml_metallib_start:" >> $(TEMP_ASSEMBLY)/ggml-metal-embed.s942 @echo ".incbin \"ggml/src/ggml-metal/ggml-metal-embed.metal\"" >> $(TEMP_ASSEMBLY)/ggml-metal-embed.s943 @echo ".globl _ggml_metallib_end" >> $(TEMP_ASSEMBLY)/ggml-metal-embed.s944 @echo "_ggml_metallib_end:" >> $(TEMP_ASSEMBLY)/ggml-metal-embed.s945 $(CC) $(CFLAGS) -c $(TEMP_ASSEMBLY)/ggml-metal-embed.s -o $@946 @rm -f ${TEMP_ASSEMBLY}/ggml-metal-embed.s947 @rmdir ${TEMP_ASSEMBLY}948endif949endif # GGML_METAL950 951DIR_GGML = ggml952DIR_LLAMA = src953DIR_COMMON = common954 955OBJ_GGML = \956 $(DIR_GGML)/src/ggml.o \957 $(DIR_GGML)/src/ggml-alloc.o \958 $(DIR_GGML)/src/ggml-backend.o \959 $(DIR_GGML)/src/ggml-backend-reg.o \960 $(DIR_GGML)/src/ggml-opt.o \961 $(DIR_GGML)/src/ggml-quants.o \962 $(DIR_GGML)/src/ggml-threading.o \963 $(DIR_GGML)/src/ggml-cpu/ggml-cpu.o \964 $(DIR_GGML)/src/ggml-cpu/ggml-cpu_cpp.o \965 $(DIR_GGML)/src/ggml-cpu/ggml-cpu-aarch64.o \966 $(DIR_GGML)/src/ggml-cpu/ggml-cpu-hbm.o \967 $(DIR_GGML)/src/ggml-cpu/ggml-cpu-quants.o \968 $(DIR_GGML)/src/ggml-cpu/ggml-cpu-traits.o \969 $(OBJ_GGML_EXT)970 971OBJ_LLAMA = \972 $(DIR_LLAMA)/llama.o \973 $(DIR_LLAMA)/llama-vocab.o \974 $(DIR_LLAMA)/llama-grammar.o \975 $(DIR_LLAMA)/llama-sampling.o \976 $(DIR_LLAMA)/unicode.o \977 $(DIR_LLAMA)/unicode-data.o978 979OBJ_COMMON = \980 $(DIR_COMMON)/common.o \981 $(DIR_COMMON)/arg.o \982 $(DIR_COMMON)/log.o \983 $(DIR_COMMON)/console.o \984 $(DIR_COMMON)/ngram-cache.o \985 $(DIR_COMMON)/sampling.o \986 $(DIR_COMMON)/speculative.o \987 $(DIR_COMMON)/chat.o \988 $(DIR_COMMON)/build-info.o \989 $(DIR_COMMON)/json-schema-to-grammar.o990 991OBJ_ALL = $(OBJ_GGML) $(OBJ_LLAMA) $(OBJ_COMMON)992 993LIB_GGML = $(LIB_PRE)ggml$(DSO_EXT)994LIB_GGML_S = $(LIB_PRE)ggml.a995 996LIB_LLAMA = $(LIB_PRE)llama$(DSO_EXT)997LIB_LLAMA_S = $(LIB_PRE)llama.a998 999LIB_COMMON = $(LIB_PRE)common$(DSO_EXT)1000LIB_COMMON_S = $(LIB_PRE)common.a1001 1002LIB_ALL = $(LIB_GGML) $(LIB_LLAMA) $(LIB_COMMON)1003LIB_ALL_S = $(LIB_GGML_S) $(LIB_LLAMA_S) $(LIB_COMMON_S)1004 1005GF_CC := $(CC)1006include scripts/get-flags.mk1007 1008# combine build flags with cmdline overrides1009override CPPFLAGS := $(MK_CPPFLAGS) $(CPPFLAGS)1010override CFLAGS := $(CPPFLAGS) $(MK_CFLAGS) $(GF_CFLAGS) $(CFLAGS)1011BASE_CXXFLAGS := $(MK_CXXFLAGS) $(CXXFLAGS)1012override CXXFLAGS := $(BASE_CXXFLAGS) $(HOST_CXXFLAGS) $(GF_CXXFLAGS) $(CPPFLAGS)1013override NVCCFLAGS := $(MK_NVCCFLAGS) $(NVCCFLAGS)1014override LDFLAGS := $(MK_LDFLAGS) $(LDFLAGS)1015 1016# identify CUDA host compiler1017ifdef GGML_CUDA1018GF_CC := $(NVCC) $(NVCCFLAGS) 2>/dev/null .c -Xcompiler1019include scripts/get-flags.mk1020CUDA_CXXFLAGS := $(BASE_CXXFLAGS) $(GF_CXXFLAGS) -Wno-pedantic1021endif1022 1023ifdef LLAMA_CURL1024override CXXFLAGS := $(CXXFLAGS) -DLLAMA_USE_CURL1025override LDFLAGS := $(LDFLAGS) -lcurl1026endif1027 1028#1029# Print build information1030#1031 1032$(info I llama.cpp build info: )1033$(info I UNAME_S: $(UNAME_S))1034$(info I UNAME_P: $(UNAME_P))1035$(info I UNAME_M: $(UNAME_M))1036$(info I CFLAGS: $(CFLAGS))1037$(info I CXXFLAGS: $(CXXFLAGS))1038$(info I NVCCFLAGS: $(NVCCFLAGS))1039$(info I LDFLAGS: $(LDFLAGS))1040$(info I CC: $(shell $(CC) --version | head -n 1))1041$(info I CXX: $(shell $(CXX) --version | head -n 1))1042ifdef GGML_CUDA1043$(info I NVCC: $(shell $(NVCC) --version | tail -n 1))1044CUDA_VERSION := $(shell $(NVCC) --version | grep -oP 'release (\K[0-9]+\.[0-9])')1045ifeq ($(shell awk -v "v=$(CUDA_VERSION)" 'BEGIN { print (v < 11.7) }'),1)1046 1047ifndef CUDA_DOCKER_ARCH1048ifndef CUDA_POWER_ARCH1049$(error I ERROR: For CUDA versions < 11.7 a target CUDA architecture must be explicitly provided via environment variable CUDA_DOCKER_ARCH, e.g. by running "export CUDA_DOCKER_ARCH=compute_XX" on Unix-like systems, where XX is the minimum compute capability that the code needs to run on. A list with compute capabilities can be found here: https://developer.nvidia.com/cuda-gpus )1050endif # CUDA_POWER_ARCH1051endif # CUDA_DOCKER_ARCH1052 1053endif # eq ($(shell echo "$(CUDA_VERSION) < 11.7" | bc),1)1054endif # GGML_CUDA1055$(info )1056 1057ifdef DEPRECATE_WARNING1058$(info !!! DEPRECATION WARNING !!!)1059$(info The following LLAMA_ options are deprecated and will be removed in the future. Use the GGML_ prefix instead)1060$(info - LLAMA_CUDA)1061$(info - LLAMA_METAL)1062$(info - LLAMA_METAL_EMBED_LIBRARY)1063$(info - LLAMA_OPENMP)1064$(info - LLAMA_RPC)1065$(info - LLAMA_SYCL)1066$(info - LLAMA_SYCL_F16)1067$(info - LLAMA_OPENBLAS)1068$(info - LLAMA_OPENBLAS64)1069$(info - LLAMA_BLIS)1070$(info - LLAMA_NO_LLAMAFILE)1071$(info - LLAMA_NO_ACCELERATE)1072$(info - LLAMA_NO_OPENMP)1073$(info - LLAMA_NO_METAL)1074$(info - LLAMA_NO_CCACHE)1075$(info )1076endif1077 1078ifdef REMOVE_WARNING1079$(info !!! REMOVAL WARNING !!!)1080$(info The following LLAMA_ options have been removed and are no longer supported)1081$(info - LLAMA_DISABLE_LOGS (https://github.com/ggerganov/llama.cpp/pull/9418))1082$(info - LLAMA_SERVER_VERBOSE (https://github.com/ggerganov/llama.cpp/pull/9418))1083$(info )1084endif1085 1086#1087# Build libraries1088#1089 1090# Libraries1091LIB_GGML = libggml.so1092LIB_GGML_S = libggml.a1093 1094LIB_LLAMA = libllama.so1095LIB_LLAMA_S = libllama.a1096 1097LIB_COMMON = libcommon.so1098LIB_COMMON_S = libcommon.a1099 1100# Targets1101BUILD_TARGETS += $(LIB_GGML) $(LIB_GGML_S) $(LIB_LLAMA) $(LIB_LLAMA_S) $(LIB_COMMON) $(LIB_COMMON_S)1102 1103# Dependency files1104DEP_FILES = $(OBJ_GGML:.o=.d) $(OBJ_LLAMA:.o=.d) $(OBJ_COMMON:.o=.d)1105 1106# Default target1107all: $(BUILD_TARGETS)1108 1109# force c++ build for source file that have same name as c file1110# Note: need this exception because `ggml-cpu.c` and `ggml-cpu.cpp` both produce the same obj/dep files1111$(DIR_GGML)/%_cpp.o: $(DIR_GGML)/%.cpp1112 $(CXX) $(CXXFLAGS) -MMD -c $< -o $@1113 1114# Rules for building object files1115$(DIR_GGML)/%.o: $(DIR_GGML)/%.c1116 $(CC) $(CFLAGS) -MMD -c $< -o $@1117 1118$(DIR_GGML)/%.o: $(DIR_GGML)/%.cpp1119 $(CXX) $(CXXFLAGS) -MMD -c $< -o $@1120 1121$(DIR_LLAMA)/%.o: $(DIR_LLAMA)/%.cpp1122 $(CXX) $(CXXFLAGS) -MMD -c $< -o $@1123 1124$(DIR_COMMON)/%.o: $(DIR_COMMON)/%.cpp1125 $(CXX) $(CXXFLAGS) -MMD -c $< -o $@1126 1127# Rules for building libraries1128$(LIB_GGML): $(OBJ_GGML)1129 $(CXX) $(CXXFLAGS) -shared -fPIC -o $@ $^ $(LDFLAGS)1130 1131$(LIB_GGML_S): $(OBJ_GGML)1132 ar rcs $(LIB_GGML_S) $^1133 1134$(LIB_LLAMA): $(OBJ_LLAMA) $(LIB_GGML)1135 $(CXX) $(CXXFLAGS) -shared -fPIC -o $@ $^ $(LDFLAGS)1136 1137$(LIB_LLAMA_S): $(OBJ_LLAMA)1138 ar rcs $(LIB_LLAMA_S) $^1139 1140$(LIB_COMMON): $(OBJ_COMMON) $(LIB_LLAMA) $(LIB_GGML)1141 $(CXX) $(CXXFLAGS) -shared -fPIC -o $@ $^ $(LDFLAGS)1142 1143$(LIB_COMMON_S): $(OBJ_COMMON)1144 ar rcs $(LIB_COMMON_S) $^1145 1146# Include dependency files1147-include $(DEP_FILES)1148 1149# Clean generated server assets1150clean-server-assets:1151 find examples/server -type f -name "*.js.hpp" -delete1152 find examples/server -type f -name "*.mjs.hpp" -delete1153 find examples/server -type f -name "*.css.hpp" -delete1154 find examples/server -type f -name "*.html.hpp" -delete1155 1156# Clean rule1157clean: clean-server-assets1158 rm -vrf $(BUILD_TARGETS) $(TEST_TARGETS)1159 rm -rvf *.a *.dll *.so *.dot1160 find ggml src common tests examples pocs -type f -name "*.o" -delete1161 find ggml src common tests examples pocs -type f -name "*.d" -delete1162 1163#1164# Examples1165#1166 1167# $< is the first prerequisite, i.e. the source file.1168# Explicitly compile this to an object file so that it can be cached with ccache.1169# The source file is then filtered out from $^ (the list of all prerequisites) and the object file is added instead.1170 1171# Helper function that replaces .c, .cpp, and .cu file endings with .o:1172GET_OBJ_FILE = $(patsubst %.c,%.o,$(patsubst %.cpp,%.o,$(patsubst %.cu,%.o,$(1))))1173 1174llama-cli: examples/main/main.cpp \1175 $(OBJ_ALL)1176 $(CXX) $(CXXFLAGS) -c $< -o $(call GET_OBJ_FILE, $<)1177 $(CXX) $(CXXFLAGS) $(filter-out %.h $<,$^) $(call GET_OBJ_FILE, $<) -o $@ $(LDFLAGS)1178 @echo1179 @echo '==== Run ./llama-cli -h for help. ===='1180 @echo1181 1182llama-infill: examples/infill/infill.cpp \1183 $(OBJ_ALL)1184 $(CXX) $(CXXFLAGS) -c $< -o $(call GET_OBJ_FILE, $<)1185 $(CXX) $(CXXFLAGS) $(filter-out %.h $<,$^) $(call GET_OBJ_FILE, $<) -o $@ $(LDFLAGS)1186 1187llama-run: examples/run/run.cpp \1188 $(OBJ_ALL)1189 $(CXX) $(CXXFLAGS) -c $< -o $(call GET_OBJ_FILE, $<)1190 $(CXX) $(CXXFLAGS) $(filter-out %.h $<,$^) $(call GET_OBJ_FILE, $<) -o $@ $(LDFLAGS)1191 1192llama-simple: examples/simple/simple.cpp \1193 $(OBJ_ALL)1194 $(CXX) $(CXXFLAGS) -c $< -o $(call GET_OBJ_FILE, $<)1195 $(CXX) $(CXXFLAGS) $(filter-out %.h $<,$^) $(call GET_OBJ_FILE, $<) -o $@ $(LDFLAGS)1196 1197llama-simple-chat: examples/simple-chat/simple-chat.cpp \1198 $(OBJ_ALL)1199 $(CXX) $(CXXFLAGS) -c $< -o $(call GET_OBJ_FILE, $<)1200 $(CXX) $(CXXFLAGS) $(filter-out %.h $<,$^) $(call GET_OBJ_FILE, $<) -o $@ $(LDFLAGS)