1# Copyright (c) 2017-2018, Lawrence Livermore National Security, LLC. 2# Produced at the Lawrence Livermore National Laboratory. LLNL-CODE-734707. 3# All Rights reserved. See files LICENSE and NOTICE for details. 4# 5# This file is part of CEED, a collection of benchmarks, miniapps, software 6# libraries and APIs for efficient high-order finite element and spectral 7# element discretizations for exascale applications. For more information and 8# source code availability see http://github.com/ceed. 9# 10# The CEED research is supported by the Exascale Computing Project 17-SC-20-SC, 11# a collaborative effort of two U.S. Department of Energy organizations (Office 12# of Science and the National Nuclear Security Administration) responsible for 13# the planning and preparation of a capable exascale ecosystem, including 14# software, applications, hardware, advanced system engineering and early 15# testbed platforms, in support of the nation's exascale computing imperative. 16 17CONFIG ?= config.mk 18-include $(CONFIG) 19COMMON ?= common.mk 20-include $(COMMON) 21 22ifeq (,$(filter-out undefined default,$(origin CC))) 23 CC = gcc 24endif 25ifeq (,$(filter-out undefined default,$(origin CXX))) 26 CXX = g++ 27endif 28ifeq (,$(filter-out undefined default,$(origin FC))) 29 FC = gfortran 30endif 31ifeq (,$(filter-out undefined default,$(origin LINK))) 32 LINK = $(CC) 33endif 34NVCC ?= $(CUDA_DIR)/bin/nvcc 35NVCC_CXX ?= $(CXX) 36HIPCC ?= $(HIP_DIR)/bin/hipcc 37 38# ASAN must be left empty if you don't want to use it 39ASAN ?= 40 41LDFLAGS ?= 42UNDERSCORE ?= 1 43 44# MFEM_DIR env variable should point to sibling directory 45ifneq ($(wildcard ../mfem/libmfem.*),) 46 MFEM_DIR ?= ../mfem 47endif 48 49# NEK5K_DIR env variable should point to sibling directory 50ifneq ($(wildcard ../Nek5000/*),) 51 NEK5K_DIR ?= $(abspath ../Nek5000) 52endif 53export NEK5K_DIR 54MPI ?= 1 55 56# CEED_DIR env for NEK5K testing 57export CEED_DIR = $(abspath .) 58 59# XSMM_DIR env variable should point to XSMM master (github.com/hfp/libxsmm) 60XSMM_DIR ?= ../libxsmm 61 62# OCCA_DIR env variable should point to OCCA master (github.com/libocca/occa) 63OCCA_DIR ?= ../occa 64 65# env variable MAGMA_DIR can be used too 66MAGMA_DIR ?= ../magma 67# If CUDA_DIR is not set, check for nvcc, or resort to /usr/local/cuda 68CUDA_DIR ?= $(or $(patsubst %/,%,$(dir $(patsubst %/,%,$(dir \ 69 $(shell which nvcc 2> /dev/null))))),/usr/local/cuda) 70HIP_DIR ?= /opt/rocm 71 72# Check for PETSc in ../petsc 73ifneq ($(wildcard ../petsc/lib/libpetsc.*),) 74 PETSC_DIR ?= ../petsc 75endif 76 77# Warning: SANTIZ options still don't run with /gpu/occa 78# export LSAN_OPTIONS=suppressions=.asanignore 79AFLAGS = -fsanitize=address #-fsanitize=undefined -fno-omit-frame-pointer 80 81# Note: Intel oneAPI C/C++ compiler is now icx/icpx 82CC_VENDOR := $(subst oneAPI,icc,$(patsubst gcc%,gcc,$(firstword $(filter gcc% clang icc oneAPI XL,$(shell $(CC) --version))))) 83FC_VENDOR := $(firstword $(filter GNU ifort XL,$(shell $(FC) --version 2>&1 || $(FC) -qversion))) 84 85# Default extra flags by vendor 86MARCHFLAG.gcc := -march=native 87MARCHFLAG.clang := $(MARCHFLAG.gcc) 88MARCHFLAG.icc := 89OMP_SIMD_FLAG.gcc := -fopenmp-simd 90OMP_SIMD_FLAG.clang := $(OMP_SIMD_FLAG.gcc) 91OMP_SIMD_FLAG.icc := -qopenmp-simd 92OPT.gcc := -ffp-contract=fast 93OPT.clang := $(OPT.gcc) 94CFLAGS.gcc := -fPIC -std=c99 -Wall -Wextra -Wno-unused-parameter -MMD -MP 95CFLAGS.clang := $(CFLAGS.gcc) 96CFLAGS.icc := $(CFLAGS.gcc) 97CFLAGS.XL := -qpic -MMD 98CXXFLAGS.gcc := -fPIC -std=c++11 -Wall -Wextra -Wno-unused-parameter -MMD -MP 99CXXFLAGS.clang := $(CXXFLAGS.gcc) 100CXXFLAGS.icc := $(CXXFLAGS.gcc) 101CXXFLAGS.XL := -qpic -MMD 102FFLAGS.GNU := -fPIC -cpp -Wall -Wextra -Wno-unused-parameter -Wno-unused-dummy-argument -MMD -MP 103FFLAGS.ifort := -fPIC -cpp 104FFLAGS.XL := -qpic -ffree-form -qpreprocess -qextname -MMD 105 106# This check works with compilers that use gcc and clang. It fails with some 107# compilers; e.g., xlc apparently ignores all options when -E is passed, thus 108# succeeds with any flags. Users can pass MARCHFLAG=... if desired. 109cc_check_flag = $(shell $(CC) -E -Werror $(1) -x c /dev/null > /dev/null 2>&1 && echo 1) 110MARCHFLAG := $(MARCHFLAG.$(CC_VENDOR)) 111MARCHFLAG := $(if $(call cc_check_flag,$(MARCHFLAG)),$(MARCHFLAG),-mcpu=native) 112MARCHFLAG := $(if $(call cc_check_flag,$(MARCHFLAG)),$(MARCHFLAG)) 113 114OMP_SIMD_FLAG := $(OMP_SIMD_FLAG.$(CC_VENDOR)) 115OMP_SIMD_FLAG := $(if $(call cc_check_flag,$(OMP_SIMD_FLAG)),$(OMP_SIMD_FLAG)) 116 117OPT ?= -O -g $(MARCHFLAG) $(OPT.$(CC_VENDOR)) $(OMP_SIMD_FLAG) 118CFLAGS ?= $(OPT) $(CFLAGS.$(CC_VENDOR)) 119CXXFLAGS ?= $(OPT) $(CXXFLAGS.$(CC_VENDOR)) 120NVCCFLAGS ?= -ccbin $(CXX) -Xcompiler "$(OPT)" -Xcompiler -fPIC 121HIPCCFLAGS ?= $(OPT) -fPIC 122FFLAGS ?= $(OPT) $(FFLAGS.$(FC_VENDOR)) 123 124ifeq ($(COVERAGE), 1) 125 CFLAGS += --coverage 126 CXXFLAGS += --coverage 127 LDFLAGS += --coverage 128endif 129 130CFLAGS += $(if $(ASAN),$(AFLAGS)) 131FFLAGS += $(if $(ASAN),$(AFLAGS)) 132LDFLAGS += $(if $(ASAN),$(AFLAGS)) 133CPPFLAGS += -I./include 134LDLIBS = -lm 135OBJDIR := build 136LIBDIR := lib 137 138# Installation variables 139prefix ?= /usr/local 140bindir = $(prefix)/bin 141libdir = $(prefix)/lib 142includedir = $(prefix)/include 143pkgconfigdir = $(libdir)/pkgconfig 144INSTALL = install 145INSTALL_PROGRAM = $(INSTALL) 146INSTALL_DATA = $(INSTALL) -m644 147 148# Get number of processors of the machine 149NPROCS := $(shell getconf _NPROCESSORS_ONLN) 150# prepare make options to run in parallel 151MFLAGS := -j $(NPROCS) --warn-undefined-variables \ 152 --no-print-directory --no-keep-going 153 154PYTHON ?= python3 155PROVE ?= prove 156PROVE_OPTS ?= -j $(NPROCS) 157DARWIN := $(filter Darwin,$(shell uname -s)) 158SO_EXT := $(if $(DARWIN),dylib,so) 159 160ceed.pc := $(LIBDIR)/pkgconfig/ceed.pc 161libceed := $(LIBDIR)/libceed.$(SO_EXT) 162CEED_LIBS = -lceed 163libceed.c := $(filter-out interface/ceed-cuda.c interface/ceed-hip.c, $(wildcard interface/ceed*.c)) 164gallery.c := $(wildcard gallery/*/ceed*.c) 165libceed.c += $(gallery.c) 166libceed_test := $(LIBDIR)/libceed_test.$(SO_EXT) 167libceeds = $(libceed) $(libceed_test) 168BACKENDS_BUILTIN := /cpu/self/ref/serial /cpu/self/ref/blocked /cpu/self/opt/serial /cpu/self/opt/blocked 169BACKENDS := $(BACKENDS_BUILTIN) 170 171# Tests 172tests.c := $(sort $(wildcard tests/t[0-9][0-9][0-9]-*.c)) 173tests.f := $(sort $(wildcard tests/t[0-9][0-9][0-9]-*.f90)) 174tests := $(tests.c:tests/%.c=$(OBJDIR)/%) 175ctests := $(tests) 176tests += $(tests.f:tests/%.f90=$(OBJDIR)/%) 177# Examples 178examples.c := $(sort $(wildcard examples/ceed/*.c)) 179examples.f := $(sort $(wildcard examples/ceed/*.f)) 180examples := $(examples.c:examples/ceed/%.c=$(OBJDIR)/%) 181examples += $(examples.f:examples/ceed/%.f=$(OBJDIR)/%) 182# MFEM Examples 183mfemexamples.cpp := $(sort $(wildcard examples/mfem/*.cpp)) 184mfemexamples := $(mfemexamples.cpp:examples/mfem/%.cpp=$(OBJDIR)/mfem-%) 185# Nek5K Examples 186nekexamples := $(OBJDIR)/nek-bps 187# PETSc Examples 188petscexamples.c := $(wildcard examples/petsc/*.c) 189petscexamples := $(petscexamples.c:examples/petsc/%.c=$(OBJDIR)/petsc-%) 190# Fluid Dynamics Examples 191fluidsexamples.c := $(sort $(wildcard examples/fluids/*.c)) 192fluidsexamples := $(fluidsexamples.c:examples/fluids/%.c=$(OBJDIR)/fluids-%) 193# Solid Mechanics Examples 194solidsexamples.c := $(sort $(wildcard examples/solids/*.c)) 195solidsexamples := $(solidsexamples.c:examples/solids/%.c=$(OBJDIR)/solids-%) 196 197# Backends/[ref, blocked, template, memcheck, opt, avx, occa, magma] 198ref.c := $(sort $(wildcard backends/ref/*.c)) 199blocked.c := $(sort $(wildcard backends/blocked/*.c)) 200template.c := $(sort $(wildcard backends/template/*.c)) 201ceedmemcheck.c := $(sort $(wildcard backends/memcheck/*.c)) 202opt.c := $(sort $(wildcard backends/opt/*.c)) 203avx.c := $(sort $(wildcard backends/avx/*.c)) 204xsmm.c := $(sort $(wildcard backends/xsmm/*.c)) 205cuda.c := $(sort $(wildcard backends/cuda/*.c)) 206cuda.cpp := $(sort $(wildcard backends/cuda/*.cpp)) 207cuda.cu := $(sort $(wildcard backends/cuda/kernels/*.cu)) 208cuda-shared.c := $(sort $(wildcard backends/cuda-shared/*.c)) 209cuda-shared.cu := $(sort $(wildcard backends/cuda-shared/kernels/*.cu)) 210cuda-gen.c := $(sort $(wildcard backends/cuda-gen/*.c)) 211cuda-gen.cpp := $(sort $(wildcard backends/cuda-gen/*.cpp)) 212cuda-gen.cu := $(sort $(wildcard backends/cuda-gen/kernels/*.cu)) 213occa.cpp := $(sort $(shell find backends/occa -type f -name *.cpp)) 214magma.c := $(sort $(wildcard backends/magma/*.c)) 215magma.cu := $(sort $(wildcard backends/magma/kernels/*.cu)) 216hip.c := $(sort $(wildcard backends/hip/*.c)) 217hip.cpp := $(sort $(wildcard backends/hip/*.cpp)) 218hip.hip := $(sort $(wildcard backends/hip/kernels/*.hip.cpp)) 219 220# Quiet, color output 221quiet ?= $($(1)) 222 223# Cancel built-in and old-fashioned implicit rules which we don't use 224.SUFFIXES: 225 226.SECONDEXPANSION: # to expand $$(@D)/.DIR 227 228%/.DIR : 229 @mkdir -p $(@D) 230 @touch $@ 231 232.PRECIOUS: %/.DIR 233 234lib: $(libceed) $(ceed.pc) 235# run 'lib' target in parallel 236par:;@$(MAKE) $(MFLAGS) V=$(V) lib 237backend_status = $(if $(filter $1,$(BACKENDS)), [backends: $1], [not found]) 238info: 239 $(info ------------------------------------) 240 $(info CC = $(CC)) 241 $(info CXX = $(CXX)) 242 $(info FC = $(FC)) 243 $(info CPPFLAGS = $(CPPFLAGS)) 244 $(info CFLAGS = $(value CFLAGS)) 245 $(info CXXFLAGS = $(value CXXFLAGS)) 246 $(info FFLAGS = $(value FFLAGS)) 247 $(info NVCCFLAGS = $(value NVCCFLAGS)) 248 $(info HIPCCFLAGS = $(value HIPCCFLAGS)) 249 $(info LDFLAGS = $(value LDFLAGS)) 250 $(info LDLIBS = $(LDLIBS)) 251 $(info OPT = $(OPT)) 252 $(info AFLAGS = $(AFLAGS)) 253 $(info ASAN = $(or $(ASAN),(empty))) 254 $(info V = $(or $(V),(empty)) [verbose=$(if $(V),on,off)]) 255 $(info ------------------------------------) 256 $(info MEMCHK_STATUS = $(MEMCHK_STATUS)$(call backend_status,/cpu/self/memcheck/serial /cpu/self/memcheck/blocked)) 257 $(info AVX_STATUS = $(AVX_STATUS)$(call backend_status,/cpu/self/avx/serial /cpu/self/avx/blocked)) 258 $(info XSMM_DIR = $(XSMM_DIR)$(call backend_status,/cpu/self/xsmm/serial /cpu/self/xsmm/blocked)) 259 $(info OCCA_DIR = $(OCCA_DIR)$(call backend_status,$(OCCA_BACKENDS))) 260 $(info MAGMA_DIR = $(MAGMA_DIR)$(call backend_status,/gpu/magma /gpu/magma/det)) 261 $(info CUDA_DIR = $(CUDA_DIR)$(call backend_status,$(CUDA_BACKENDS))) 262 $(info HIP_DIR = $(HIP_DIR)$(call backend_status,$(HIP_BACKENDS))) 263 $(info ------------------------------------) 264 $(info MFEM_DIR = $(MFEM_DIR)) 265 $(info NEK5K_DIR = $(NEK5K_DIR)) 266 $(info PETSC_DIR = $(PETSC_DIR)) 267 $(info ------------------------------------) 268 $(info prefix = $(prefix)) 269 $(info includedir = $(value includedir)) 270 $(info libdir = $(value libdir)) 271 $(info pkgconfigdir = $(value pkgconfigdir)) 272 $(info ------------------------------------) 273 @true 274info-backends: 275 $(info make: 'lib' with optional backends: $(filter-out $(BACKENDS_BUILTIN),$(BACKENDS))) 276.PHONY: lib all par info info-backends 277 278$(libceed) : LDFLAGS += $(if $(DARWIN), -install_name @rpath/$(notdir $(libceed))) 279$(libceed_test) : LDFLAGS += $(if $(DARWIN), -install_name @rpath/$(notdir $(libceed_test))) 280 281# Standard Backends 282libceed.c += $(ref.c) 283libceed.c += $(blocked.c) 284libceed.c += $(opt.c) 285 286# Testing Backends 287test_backends.c := $(template.c) 288TEST_BACKENDS := /cpu/self/tmpl /cpu/self/tmpl/sub 289 290# Memcheck Backend 291MEMCHK_STATUS = Disabled 292MEMCHK := $(shell echo "\#include <valgrind/memcheck.h>" | $(CC) $(CPPFLAGS) -E - >/dev/null 2>&1 && echo 1) 293ifeq ($(MEMCHK),1) 294 MEMCHK_STATUS = Enabled 295 libceed.c += $(ceedmemcheck.c) 296 BACKENDS += /cpu/self/memcheck/serial /cpu/self/memcheck/blocked 297endif 298 299# AVX Backed 300AVX_STATUS = Disabled 301AVX_FLAG := $(if $(filter clang,$(CC_VENDOR)),+avx,-mavx) 302AVX := $(filter $(AVX_FLAG),$(shell $(CC) $(OPT) -v -E -x c /dev/null 2>&1)) 303ifneq ($(AVX),) 304 AVX_STATUS = Enabled 305 libceed.c += $(avx.c) 306 BACKENDS += /cpu/self/avx/serial /cpu/self/avx/blocked 307endif 308 309# libXSMM Backends 310ifneq ($(wildcard $(XSMM_DIR)/lib/libxsmm.*),) 311 $(libceeds) : LDFLAGS += -L$(XSMM_DIR)/lib -Wl,-rpath,$(abspath $(XSMM_DIR)/lib) 312 $(libceeds) : LDLIBS += -lxsmm -ldl 313 MKL ?= 314 ifeq (,$(MKL)$(MKLROOT)) 315 BLAS_LIB = -lblas 316 else 317 ifneq ($(MKLROOT),) 318 # Some installs put everything inside an intel64 subdirectory, others not 319 MKL_LIBDIR = $(dir $(firstword $(wildcard $(MKLROOT)/lib/intel64/libmkl_sequential.* $(MKLROOT)/lib/libmkl_sequential.*))) 320 MKL_LINK = -L$(MKL_LIBDIR) -Wl,-rpath,$(MKL_LIBDIR) 321 endif 322 BLAS_LIB = $(MKL_LINK) -Wl,--no-as-needed -lmkl_intel_lp64 -lmkl_sequential -lmkl_core -lpthread -lm -ldl 323 endif 324 $(libceeds) : LDLIBS += $(BLAS_LIB) 325 libceed.c += $(xsmm.c) 326 $(xsmm.c:%.c=$(OBJDIR)/%.o) $(xsmm.c:%=%.tidy) : CPPFLAGS += -I$(XSMM_DIR)/include 327 BACKENDS += /cpu/self/xsmm/serial /cpu/self/xsmm/blocked 328endif 329 330# OCCA Backends 331OCCA_BACKENDS = /cpu/self/occa 332ifneq ($(wildcard $(OCCA_DIR)/lib/libocca.*),) 333 OCCA_MODES := $(shell $(OCCA_DIR)/bin/occa modes) 334 OCCA_BACKENDS += $(if $(filter OpenMP,$(OCCA_MODES)),/cpu/openmp/occa) 335# OCCA_BACKENDS += $(if $(filter OpenCL,$(OCCA_MODES)),/gpu/opencl/occa) 336 OCCA_BACKENDS += $(if $(filter HIP,$(OCCA_MODES)),/gpu/hip/occa) 337 OCCA_BACKENDS += $(if $(filter CUDA,$(OCCA_MODES)),/gpu/cuda/occa) 338 339 $(libceeds) : CPPFLAGS += -I$(OCCA_DIR)/include 340 $(libceeds) : LDFLAGS += -L$(OCCA_DIR)/lib -Wl,-rpath,$(abspath $(OCCA_DIR)/lib) 341 $(libceeds) : LDLIBS += -locca 342 $(libceeds) : LINK = $(CXX) 343 libceed.cpp += $(occa.cpp) 344 BACKENDS += $(OCCA_BACKENDS) 345endif 346 347# CUDA Backends 348CUDA_LIB_DIR := $(wildcard $(foreach d,lib lib64,$(CUDA_DIR)/$d/libcudart.${SO_EXT})) 349CUDA_LIB_DIR := $(patsubst %/,%,$(dir $(firstword $(CUDA_LIB_DIR)))) 350CUDA_LIB_DIR_STUBS := $(CUDA_LIB_DIR)/stubs 351CUDA_BACKENDS = /gpu/cuda/ref /gpu/cuda/shared /gpu/cuda/gen 352ifneq ($(CUDA_LIB_DIR),) 353 $(libceeds) : CPPFLAGS += -I$(CUDA_DIR)/include 354 $(libceeds) : LDFLAGS += -L$(CUDA_LIB_DIR) -Wl,-rpath,$(abspath $(CUDA_LIB_DIR)) 355 $(libceeds) : LDLIBS += -lcudart -lnvrtc -lcuda -lcublas 356 $(libceeds) : LINK = $(CXX) 357 libceed.c += interface/ceed-cuda.c 358 libceed.c += $(cuda.c) $(cuda-shared.c) $(cuda-gen.c) 359 libceed.cpp += $(cuda.cpp) $(cuda-gen.cpp) 360 libceed.cu += $(cuda.cu) $(cuda-shared.cu) $(cuda-gen.cu) 361 BACKENDS += $(CUDA_BACKENDS) 362endif 363 364# HIP Backends 365HIP_LIB_DIR := $(wildcard $(foreach d,lib lib64,$(HIP_DIR)/$d/libamdhip64.${SO_EXT})) 366HIP_LIB_DIR := $(patsubst %/,%,$(dir $(firstword $(HIP_LIB_DIR)))) 367HIP_BACKENDS = /gpu/hip/ref 368ifneq ($(HIP_LIB_DIR),) 369 $(libceeds) : HIPCCFLAGS += -I./include 370 ifneq ($(CXX), $(HIPCC)) 371 CPPFLAGS += $(subst =,,$(shell $(HIP_DIR)/bin/hipconfig -C)) 372 endif 373 $(libceeds) : CPPFLAGS += -I$(HIP_DIR)/include -Wno-unused-function 374 $(libceeds) : LDFLAGS += -L$(HIP_LIB_DIR) -Wl,-rpath,$(abspath $(HIP_LIB_DIR)) 375 $(libceeds) : LDLIBS += -lamdhip64 -lhipblas 376 $(libceeds) : LINK = $(CXX) 377 libceed.c += interface/ceed-hip.c 378 libceed.c += $(hip.c) 379 libceed.cpp += $(hip.cpp) 380 libceed.hip += $(hip.hip) 381 BACKENDS += $(HIP_BACKENDS) 382endif 383 384# MAGMA Backend 385ifneq ($(wildcard $(MAGMA_DIR)/lib/libmagma.*),) 386 ifneq ($(CUDA_LIB_DIR),) 387 cuda_link = -Wl,-rpath,$(CUDA_LIB_DIR) -L$(CUDA_LIB_DIR) -lcublas -lcusparse -lcudart 388 omp_link = -fopenmp 389 magma_link_static = -L$(MAGMA_DIR)/lib -lmagma $(cuda_link) $(omp_link) 390 magma_link_shared = -L$(MAGMA_DIR)/lib -Wl,-rpath,$(abspath $(MAGMA_DIR)/lib) -lmagma 391 magma_link := $(if $(wildcard $(MAGMA_DIR)/lib/libmagma.${SO_EXT}),$(magma_link_shared),$(magma_link_static)) 392 $(libceeds) : LDLIBS += $(magma_link) 393 $(tests) $(examples) : LDLIBS += $(magma_link) 394 libceed.c += $(magma.c) 395 libceed.cu += $(magma.cu) 396 $(magma.c:%.c=$(OBJDIR)/%.o) $(magma.c:%=%.tidy) : CPPFLAGS += -DADD_ -I$(MAGMA_DIR)/include -I$(CUDA_DIR)/include 397 $(magma.cu:%.cu=$(OBJDIR)/%.o) : CPPFLAGS += --compiler-options=-fPIC -DADD_ -I$(MAGMA_DIR)/include -I$(MAGMA_DIR)/magmablas -I$(MAGMA_DIR)/control -I$(CUDA_DIR)/include 398 BACKENDS += /gpu/cuda/magma /gpu/cuda/magma/det 399 endif 400endif 401 402export BACKENDS 403 404libceed.o = $(libceed.c:%.c=$(OBJDIR)/%.o) $(libceed.cpp:%.cpp=$(OBJDIR)/%.o) $(libceed.cu:%.cu=$(OBJDIR)/%.o) $(libceed.hip:%.hip.cpp=$(OBJDIR)/%.o) 405$(filter %fortran.o,$(libceed.o)) : CPPFLAGS += $(if $(filter 1,$(UNDERSCORE)),-DUNDERSCORE) 406$(libceed.o): | info-backends 407$(libceed) : $(libceed.o) | $$(@D)/.DIR 408 $(call quiet,LINK) $(LDFLAGS) -shared -o $@ $^ $(LDLIBS) 409 410$(OBJDIR)/%.o : $(CURDIR)/%.c | $$(@D)/.DIR 411 $(call quiet,CC) $(CPPFLAGS) $(CFLAGS) -c -o $@ $(abspath $<) 412 413$(OBJDIR)/%.o : $(CURDIR)/%.cpp | $$(@D)/.DIR 414 $(call quiet,CXX) $(CPPFLAGS) $(CXXFLAGS) -c -o $@ $(abspath $<) 415 416$(OBJDIR)/%.o : $(CURDIR)/%.cu | $$(@D)/.DIR 417 $(call quiet,NVCC) $(filter-out -Wno-unused-function, $(CPPFLAGS)) $(NVCCFLAGS) -c -o $@ $(abspath $<) 418 419$(OBJDIR)/%.o : $(CURDIR)/%.hip.cpp | $$(@D)/.DIR 420 $(call quiet,HIPCC) $(HIPCCFLAGS) -c -o $@ $(abspath $<) 421 422$(OBJDIR)/% : tests/%.c | $$(@D)/.DIR 423 $(call quiet,LINK.c) $(CEED_LDFLAGS) -o $@ $(abspath $<) $(CEED_LIBS) $(LDLIBS) 424 425$(OBJDIR)/% : tests/%.f90 | $$(@D)/.DIR 426 $(call quiet,LINK.F) -DSOURCE_DIR='"$(abspath $(<D))/"' $(CEED_LDFLAGS) -o $@ $(abspath $<) $(CEED_LIBS) $(LDLIBS) 427 428$(OBJDIR)/% : examples/ceed/%.c | $$(@D)/.DIR 429 $(call quiet,LINK.c) $(CEED_LDFLAGS) -o $@ $(abspath $<) $(CEED_LIBS) $(LDLIBS) 430 431$(OBJDIR)/% : examples/ceed/%.f | $$(@D)/.DIR 432 $(call quiet,LINK.F) -DSOURCE_DIR='"$(abspath $(<D))/"' $(CEED_LDFLAGS) -o $@ $(abspath $<) $(CEED_LIBS) $(LDLIBS) 433 434$(OBJDIR)/mfem-% : examples/mfem/%.cpp $(libceed) | $$(@D)/.DIR 435 +$(MAKE) -C examples/mfem CEED_DIR=`pwd` \ 436 MFEM_DIR="$(abspath $(MFEM_DIR))" CXX=$(CXX) $* 437 mv examples/mfem/$* $@ 438 439# Note: Multiple Nek files cannot be built in parallel. The '+' here enables 440# this single Nek bps file to be built in parallel with other examples, 441# such as when calling `make prove-all -j2`. 442$(OBJDIR)/nek-bps : examples/nek/bps/bps.usr examples/nek/nek-examples.sh $(libceed) | $$(@D)/.DIR 443 +$(MAKE) -C examples MPI=$(MPI) CEED_DIR=`pwd` NEK5K_DIR="$(abspath $(NEK5K_DIR))" nek 444 mv examples/nek/build/bps $(OBJDIR)/bps 445 cp examples/nek/nek-examples.sh $(OBJDIR)/nek-bps 446 447$(OBJDIR)/petsc-% : examples/petsc/%.c $(libceed) $(ceed.pc) | $$(@D)/.DIR 448 +$(MAKE) -C examples/petsc CEED_DIR=`pwd` \ 449 PETSC_DIR="$(abspath $(PETSC_DIR))" OPT="$(OPT)" $* 450 mv examples/petsc/$* $@ 451 452$(OBJDIR)/fluids-% : examples/fluids/%.c $(libceed) $(ceed.pc) | $$(@D)/.DIR 453 +$(MAKE) -C examples/fluids CEED_DIR=`pwd` \ 454 PETSC_DIR="$(abspath $(PETSC_DIR))" OPT="$(OPT)" $* 455 mv examples/fluids/$* $@ 456 457$(OBJDIR)/solids-% : examples/solids/%.c $(libceed) $(ceed.pc) | $$(@D)/.DIR 458 +$(MAKE) -C examples/solids CEED_DIR=`pwd` \ 459 PETSC_DIR="$(abspath $(PETSC_DIR))" OPT="$(OPT)" $* 460 mv examples/solids/$* $@ 461 462libceed_test.o = $(test_backends.c:%.c=$(OBJDIR)/%.o) 463$(libceed_test) : $(libceed.o) $(libceed_test.o) | $$(@D)/.DIR 464 $(call quiet,LINK) $(LDFLAGS) -shared -o $@ $^ $(LDLIBS) 465 466$(examples) : $(libceed) 467$(tests) : $(libceed_test) 468$(tests) : CEED_LIBS = -lceed_test 469$(tests) $(examples) : LDFLAGS += -Wl,-rpath,$(abspath $(LIBDIR)) -L$(LIBDIR) 470 471run-t% : BACKENDS += $(TEST_BACKENDS) 472run-% : $(OBJDIR)/% 473 @tests/tap.sh $(<:$(OBJDIR)/%=%) 474 475external_examples := \ 476 $(if $(MFEM_DIR),$(mfemexamples)) \ 477 $(if $(PETSC_DIR),$(petscexamples)) \ 478 $(if $(NEK5K_DIR),$(nekexamples)) \ 479 $(if $(PETSC_DIR),$(fluidsexamples)) \ 480 $(if $(PETSC_DIR),$(solidsexamples)) 481 482allexamples = $(examples) $(external_examples) 483 484# The test and prove targets can be controlled via pattern searches. The 485# default is to run tests and those examples that have no external dependencies. 486# Examples of finer grained control: 487# 488# make test search='petsc mfem' # PETSc and MFEM examples 489# make prove search='t3' # t3xx series tests 490# make junit search='ex petsc' # core ex* and PETSc tests 491search ?= t ex 492realsearch = $(search:%=%%) 493matched = $(foreach pattern,$(realsearch),$(filter $(OBJDIR)/$(pattern),$(tests) $(allexamples))) 494 495# Test core libCEED 496test : $(matched:$(OBJDIR)/%=run-%) 497 498# Run test target in parallel 499tst : ;@$(MAKE) $(MFLAGS) V=$(V) test 500# CPU C tests only for backend % 501ctc-% : $(ctests);@$(foreach tst,$(ctests),$(tst) /cpu/$*;) 502 503prove : BACKENDS += $(TEST_BACKENDS) 504prove : $(matched) 505 $(info Testing backends: $(BACKENDS)) 506 $(PROVE) $(PROVE_OPTS) --exec 'tests/tap.sh' $(matched:$(OBJDIR)/%=%) 507# Run prove target in parallel 508prv : ;@$(MAKE) $(MFLAGS) V=$(V) prove 509 510prove-all : 511 +$(MAKE) prove realsearch=% 512 513junit-t% : BACKENDS += $(TEST_BACKENDS) 514junit-% : $(OBJDIR)/% 515 @printf " %10s %s\n" TEST $(<:$(OBJDIR)/%=%); $(PYTHON) tests/junit.py $(<:$(OBJDIR)/%=%) 516 517junit : $(matched:$(OBJDIR)/%=junit-%) 518 519all: $(alltests) 520 521examples : $(allexamples) 522ceedexamples : $(examples) 523nekexamples : $(nekexamples) 524mfemexamples : $(mfemexamples) 525petscexamples : $(petscexamples) 526 527# Benchmarks 528allbenchmarks = petsc-bps 529bench_targets = $(addprefix bench-,$(allbenchmarks)) 530.PHONY: $(bench_targets) benchmarks 531$(bench_targets): bench-%: $(OBJDIR)/% 532 cd benchmarks && ./benchmark.sh --ceed "$(BACKENDS)" -r $(*).sh 533benchmarks: $(bench_targets) 534 535$(ceed.pc) : pkgconfig-prefix = $(abspath .) 536$(OBJDIR)/ceed.pc : pkgconfig-prefix = $(prefix) 537.INTERMEDIATE : $(OBJDIR)/ceed.pc 538%/ceed.pc : ceed.pc.template | $$(@D)/.DIR 539 @sed "s:%prefix%:$(pkgconfig-prefix):" $< > $@ 540 541install : $(libceed) $(OBJDIR)/ceed.pc 542 $(INSTALL) -d $(addprefix $(if $(DESTDIR),"$(DESTDIR)"),"$(includedir)"\ 543 "$(libdir)" "$(pkgconfigdir)") 544 $(INSTALL_DATA) include/ceed.h "$(DESTDIR)$(includedir)/" 545 $(INSTALL_DATA) include/ceedf.h "$(DESTDIR)$(includedir)/" 546 $(INSTALL_DATA) include/ceed-hash.h "$(DESTDIR)$(includedir)/" 547 $(INSTALL_DATA) include/ceed-khash.h "$(DESTDIR)$(includedir)/" 548 $(INSTALL_DATA) $(libceed) "$(DESTDIR)$(libdir)/" 549 $(INSTALL_DATA) $(OBJDIR)/ceed.pc "$(DESTDIR)$(pkgconfigdir)/" 550 551.PHONY : cln clean doxygen doc lib install all print test tst prove prv prove-all junit examples style style-c style-py tidy info info-backends 552 553cln clean : 554 $(RM) -r $(OBJDIR) $(LIBDIR) dist *egg* .pytest_cache *cffi* 555 $(MAKE) -C examples clean NEK5K_DIR="$(abspath $(NEK5K_DIR))" 556 $(MAKE) -C tests/python clean 557 $(RM) benchmarks/*output.txt 558 559distclean : clean 560 $(RM) -r doc/html doc/sphinx/build $(CONFIG) 561 562DOXYGEN ?= doxygen 563doxygen : 564 $(DOXYGEN) Doxyfile 565 566doc-html doc-latexpdf doc-epub : doc-% : doxygen 567 make -C doc/sphinx $* 568 569doc : doc-html 570 571style-c : 572 @astyle --options=.astylerc \ 573 $(filter-out include/ceedf.h $(wildcard tests/t*-f.h), \ 574 $(wildcard include/*.h interface/*.[ch] tests/*.[ch] backends/*/*.[ch] \ 575 examples/*/*/*.[ch] examples/*/*.[ch] examples/*/*.[ch]pp gallery/*/*.[ch])) 576 577AUTOPEP8 = autopep8 578style-py : AUTOPEP8_ARGS = --in-place --aggressive 579style-py : 580 @$(AUTOPEP8) $(AUTOPEP8_ARGS) $(wildcard *.py python**/*.py tests/python**/*.py examples**/*.py doc/sphinx/source**/*.py benchmarks/*.py) 581 582style : style-c style-py 583 584CLANG_TIDY ?= clang-tidy 585 586%.c.tidy : %.c 587 $(CLANG_TIDY) $(TIDY_OPTS) $^ -- $(CPPFLAGS) --std=c99 -I$(CUDA_DIR)/include 588 589%.cpp.tidy : %.cpp 590 $(CLANG_TIDY) $(TIDY_OPTS) $^ -- $(CPPFLAGS) --std=c++11 -I$(CUDA_DIR)/include -I$(OCCA_DIR)/include 591 592tidy_c : $(libceed.c:%=%.tidy) 593tidy_cpp : $(libceed.cpp:%=%.tidy) 594 595tidy : tidy_c tidy_cpp 596 597print : 598 @echo $(VAR)=$($(VAR)) 599 600print-% : 601 $(info [ variable name]: $*) 602 $(info [ origin]: $(origin $*)) 603 $(info [ flavor]: $(flavor $*)) 604 $(info [ value]: $(value $*)) 605 $(info [expanded value]: $($*)) 606 $(info ) 607 @true 608 609# "make configure" detects any variables passed on the command line or 610# previously set in config.mk, caching them in config.mk as simple 611# (:=) variables. Variables set in config.mk or on the command line 612# take precedence over the defaults provided in the file. Typical 613# usage: 614# 615# make configure CC=/path/to/my/cc CUDA_DIR=/opt/cuda 616# make 617# make prove 618# 619# The values in the file can be updated by passing them on the command 620# line, e.g., 621# 622# make configure CC=/path/to/other/clang 623 624# All variables to consider for caching 625CONFIG_VARS = CC CXX FC NVCC NVCC_CXX HIPCC \ 626 OPT CFLAGS CPPFLAGS CXXFLAGS FFLAGS NVCCFLAGS HIPCCFLAGS \ 627 LDFLAGS LDLIBS \ 628 MAGMA_DIR XSMM_DIR CUDA_DIR MFEM_DIR PETSC_DIR NEK5K_DIR HIP_DIR 629 630# $(call needs_save,CFLAGS) returns true (a nonempty string) if CFLAGS 631# was set on the command line or in config.mk (where it will appear as 632# a simple variable). 633needs_save = $(or $(filter command line,$(origin $(1))),$(filter simple,$(flavor $(1)))) 634 635configure : 636 $(file > $(CONFIG)) 637 $(foreach v,$(CONFIG_VARS),$(if $(call needs_save,$(v)),$(file >> $(CONFIG),$(v) := $($(v))))) 638 @echo "Configuration cached in $(CONFIG):" 639 @cat $(CONFIG) 640 641wheel : export MARCHFLAG = -march=generic 642wheel : export WHEEL_PLAT = manylinux2010_x86_64 643wheel : 644 docker run -it --user $(shell id -u):$(shell id -g) --rm -v $(PWD):/io -w /io \ 645 -e MARCHFLAG -e WHEEL_PLAT \ 646 quay.io/pypa/$(WHEEL_PLAT) python/make-wheels.sh 647 648.PHONY : configure wheel 649 650# Include *.d deps when not -B = --always-make: useful if the paths are wonky in a container 651-include $(if $(filter B,$(MAKEFLAGS)),,$(libceed.c:%.c=$(OBJDIR)/%.d) $(tests.c:tests/%.c=$(OBJDIR)/%.d)) 652