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 $(CONFIG) 19 20ifeq (,$(filter-out undefined default,$(origin CC))) 21 CC = gcc 22endif 23ifeq (,$(filter-out undefined default,$(origin CXX))) 24 CXX = g++ 25endif 26ifeq (,$(filter-out undefined default,$(origin FC))) 27 FC = gfortran 28endif 29ifeq (,$(filter-out undefined default,$(origin LINK))) 30 LINK = $(CC) 31endif 32NVCC ?= $(CUDA_DIR)/bin/nvcc 33NVCC_CXX ?= $(CXX) 34 35# ASAN must be left empty if you don't want to use it 36ASAN ?= 37 38LDFLAGS ?= 39UNDERSCORE ?= 1 40 41# MFEM_DIR env variable should point to sibling directory 42ifneq ($(wildcard ../mfem/libmfem.*),) 43 MFEM_DIR ?= ../mfem 44endif 45 46# NEK5K_DIR env variable should point to sibling directory 47ifneq ($(wildcard ../Nek5000/*),) 48 NEK5K_DIR ?= $(abspath ../Nek5000) 49endif 50export NEK5K_DIR 51MPI ?= 1 52 53# CEED_DIR env for NEK5K testing 54export CEED_DIR = $(abspath .) 55 56# XSMM_DIR env variable should point to XSMM master (github.com/hfp/libxsmm) 57XSMM_DIR ?= ../libxsmm 58 59# OCCA_DIR env variable should point to OCCA master (github.com/libocca/occa) 60OCCA_DIR ?= ../occa 61 62# env variable MAGMA_DIR can be used too 63MAGMA_DIR ?= ../magma 64# If CUDA_DIR is not set, check for nvcc, or resort to /usr/local/cuda 65CUDA_DIR ?= $(or $(patsubst %/,%,$(dir $(patsubst %/,%,$(dir \ 66 $(shell which nvcc 2> /dev/null))))),/usr/local/cuda) 67 68# Check for PETSc in ../petsc 69ifneq ($(wildcard ../petsc/lib/libpetsc.*),) 70 PETSC_DIR ?= ../petsc 71endif 72 73# Warning: SANTIZ options still don't run with /gpu/occa 74# export LSAN_OPTIONS=suppressions=.asanignore 75AFLAGS = -fsanitize=address #-fsanitize=undefined -fno-omit-frame-pointer 76 77CC_VENDOR := $(patsubst gcc%,gcc,$(firstword $(filter gcc% clang icc XL,$(shell $(CC) --version)))) 78FC_VENDOR := $(firstword $(filter GNU ifort XL,$(shell $(FC) --version 2>&1 || $(FC) -qversion))) 79 80# Default extra flags by vendor 81MARCHFLAG.gcc := -march=native 82MARCHFLAG.clang := $(MARCHFLAG.gcc) 83MARCHFLAG.icc := 84OMP_SIMD_FLAG.gcc := -fopenmp-simd 85OMP_SIMD_FLAG.clang := $(OMP_SIMD_FLAG.gcc) 86OMP_SIMD_FLAG.icc := -qopenmp-simd 87OPT.gcc := -ffp-contract=fast 88OPT.clang := $(OPT.gcc) 89CFLAGS.gcc := -fPIC -std=c99 -Wall -Wextra -Wno-unused-parameter -MMD -MP 90CFLAGS.clang := $(CFLAGS.gcc) 91CFLAGS.icc := $(CFLAGS.gcc) 92CFLAGS.XL := -qpic -MMD 93CXXFLAGS.gcc := -fPIC -std=c++11 -Wall -Wextra -Wno-unused-parameter -MMD -MP 94CXXFLAGS.clang := $(CXXFLAGS.gcc) 95CXXFLAGS.icc := $(CXXFLAGS.gcc) 96CXXFLAGS.XL := -qpic -MMD 97FFLAGS.GNU := -fPIC -cpp -Wall -Wextra -Wno-unused-parameter -Wno-unused-dummy-argument -MMD -MP 98FFLAGS.ifort := -fPIC -cpp 99FFLAGS.XL := -qpic -ffree-form -qpreprocess -qextname -MMD 100 101# This check works with compilers that use gcc and clang. It fails with some 102# compilers; e.g., xlc apparently ignores all options when -E is passed, thus 103# succeeds with any flags. Users can pass MARCHFLAG=... if desired. 104cc_check_flag = $(shell $(CC) -E -Werror $(1) -x c /dev/null > /dev/null 2>&1 && echo 1) 105MARCHFLAG := $(MARCHFLAG.$(CC_VENDOR)) 106MARCHFLAG := $(if $(call cc_check_flag,$(MARCHFLAG)),$(MARCHFLAG),-mcpu=native) 107MARCHFLAG := $(if $(call cc_check_flag,$(MARCHFLAG)),$(MARCHFLAG)) 108 109OMP_SIMD_FLAG := $(OMP_SIMD_FLAG.$(CC_VENDOR)) 110OMP_SIMD_FLAG := $(if $(call cc_check_flag,$(OMP_SIMD_FLAG)),$(OMP_SIMD_FLAG)) 111 112OPT ?= -O -g $(MARCHFLAG) $(OPT.$(CC_VENDOR)) $(OMP_SIMD_FLAG) 113CFLAGS ?= $(OPT) $(CFLAGS.$(CC_VENDOR)) 114CXXFLAGS ?= $(OPT) $(CXXFLAGS.$(CC_VENDOR)) 115NVCCFLAGS ?= -ccbin $(CXX) -Xcompiler "$(OPT)" -Xcompiler -fPIC 116FFLAGS ?= $(OPT) $(FFLAGS.$(FC_VENDOR)) 117 118ifeq ($(COVERAGE), 1) 119 CFLAGS += --coverage 120 LDFLAGS += --coverage 121endif 122 123CFLAGS += $(if $(ASAN),$(AFLAGS)) 124FFLAGS += $(if $(ASAN),$(AFLAGS)) 125LDFLAGS += $(if $(ASAN),$(AFLAGS)) 126CPPFLAGS += -I./include 127LDLIBS = -lm 128OBJDIR := build 129LIBDIR := lib 130 131# Installation variables 132prefix ?= /usr/local 133bindir = $(prefix)/bin 134libdir = $(prefix)/lib 135okldir = $(libdir)/okl 136includedir = $(prefix)/include 137pkgconfigdir = $(libdir)/pkgconfig 138INSTALL = install 139INSTALL_PROGRAM = $(INSTALL) 140INSTALL_DATA = $(INSTALL) -m644 141 142# Get number of processors of the machine 143NPROCS := $(shell getconf _NPROCESSORS_ONLN) 144# prepare make options to run in parallel 145MFLAGS := -j $(NPROCS) --warn-undefined-variables \ 146 --no-print-directory --no-keep-going 147 148PYTHON ?= python3 149PROVE ?= prove 150PROVE_OPTS ?= -j $(NPROCS) 151DARWIN := $(filter Darwin,$(shell uname -s)) 152SO_EXT := $(if $(DARWIN),dylib,so) 153 154ceed.pc := $(LIBDIR)/pkgconfig/ceed.pc 155libceed := $(LIBDIR)/libceed.$(SO_EXT) 156CEED_LIBS = -lceed 157libceed.c := $(wildcard interface/ceed*.c) 158gallery.c := $(wildcard gallery/*/ceed*.c) 159libceed.c += $(gallery.c) 160libceed_test := $(LIBDIR)/libceed_test.$(SO_EXT) 161libceeds = $(libceed) $(libceed_test) 162BACKENDS_BUILTIN := /cpu/self/ref/serial /cpu/self/ref/blocked /cpu/self/opt/serial /cpu/self/opt/blocked 163BACKENDS := $(BACKENDS_BUILTIN) 164 165# Tests 166tests.c := $(sort $(wildcard tests/t[0-9][0-9][0-9]-*.c)) 167tests.f := $(sort $(wildcard tests/t[0-9][0-9][0-9]-*.f90)) 168tests := $(tests.c:tests/%.c=$(OBJDIR)/%) 169ctests := $(tests) 170tests += $(tests.f:tests/%.f90=$(OBJDIR)/%) 171# Examples 172examples.c := $(sort $(wildcard examples/ceed/*.c)) 173examples.f := $(sort $(wildcard examples/ceed/*.f)) 174examples := $(examples.c:examples/ceed/%.c=$(OBJDIR)/%) 175examples += $(examples.f:examples/ceed/%.f=$(OBJDIR)/%) 176# MFEM Examples 177mfemexamples.cpp := $(sort $(wildcard examples/mfem/*.cpp)) 178mfemexamples := $(mfemexamples.cpp:examples/mfem/%.cpp=$(OBJDIR)/mfem-%) 179# Nek5K Examples 180nekexamples := $(OBJDIR)/nek-bps 181# PETSc Examples 182petscexamples.c := $(wildcard examples/petsc/*.c) 183petscexamples := $(petscexamples.c:examples/petsc/%.c=$(OBJDIR)/petsc-%) 184# Fluid Dynamics Examples 185fluidsexamples.c := $(sort $(wildcard examples/fluids/*.c)) 186fluidsexamples := $(fluidsexamples.c:examples/fluids/%.c=$(OBJDIR)/fluids-%) 187# Solid Mechanics Examples 188solidsexamples.c := $(sort $(wildcard examples/solids/*.c)) 189solidsexamples := $(solidsexamples.c:examples/solids/%.c=$(OBJDIR)/solids-%) 190 191# Backends/[ref, blocked, template, memcheck, opt, avx, occa, magma] 192ref.c := $(sort $(wildcard backends/ref/*.c)) 193blocked.c := $(sort $(wildcard backends/blocked/*.c)) 194template.c := $(sort $(wildcard backends/template/*.c)) 195ceedmemcheck.c := $(sort $(wildcard backends/memcheck/*.c)) 196opt.c := $(sort $(wildcard backends/opt/*.c)) 197avx.c := $(sort $(wildcard backends/avx/*.c)) 198xsmm.c := $(sort $(wildcard backends/xsmm/*.c)) 199cuda.c := $(sort $(wildcard backends/cuda/*.c)) 200cuda.cpp := $(sort $(wildcard backends/cuda/*.cpp)) 201cuda.cu := $(sort $(wildcard backends/cuda/*.cu)) 202cuda-reg.c := $(sort $(wildcard backends/cuda-reg/*.c)) 203cuda-reg.cu := $(sort $(wildcard backends/cuda-reg/*.cu)) 204cuda-shared.c := $(sort $(wildcard backends/cuda-shared/*.c)) 205cuda-shared.cu := $(sort $(wildcard backends/cuda-shared/*.cu)) 206cuda-gen.c := $(sort $(wildcard backends/cuda-gen/*.c)) 207cuda-gen.cpp := $(sort $(wildcard backends/cuda-gen/*.cpp)) 208cuda-gen.cu := $(sort $(wildcard backends/cuda-gen/*.cu)) 209occa.c := $(sort $(wildcard backends/occa/*.c)) 210magma.c := $(sort $(wildcard backends/magma/*.c)) 211magma.cu := $(sort $(wildcard backends/magma/*.cu)) 212 213# Output using the 216-color rules mode 214rule_file = $(notdir $(1)) 215rule_path = $(patsubst %/,%,$(dir $(1))) 216last_path = $(notdir $(patsubst %/,%,$(dir $(1)))) 217ansicolor = $(shell echo $(call last_path,$(1)) | cksum | cut -b1-2 | xargs -IS expr 2 \* S + 17) 218emacs_out = @printf " %10s %s/%s\n" $(1) $(call rule_path,$(2)) $(call rule_file,$(2)) 219color_out = @if [ -t 1 ]; then \ 220 printf " %10s \033[38;5;%d;1m%s\033[m/%s\n" \ 221 $(1) $(call ansicolor,$(2)) \ 222 $(call rule_path,$(2)) $(call rule_file,$(2)); else \ 223 printf " %10s %s\n" $(1) $(2); fi 224# if TERM=dumb, use it, otherwise switch to the term one 225output = $(if $(TERM:dumb=),$(call color_out,$1,$2),$(call emacs_out,$1,$2)) 226 227# if V is set to non-nil, turn the verbose mode 228quiet = $(if $(V),$($(1)),$(call output,$1,$@);$($(1))) 229 230# Cancel built-in and old-fashioned implicit rules which we don't use 231.SUFFIXES: 232 233.SECONDEXPANSION: # to expand $$(@D)/.DIR 234 235%/.DIR : 236 @mkdir -p $(@D) 237 @touch $@ 238 239.PRECIOUS: %/.DIR 240 241lib: $(libceed) $(ceed.pc) 242# run 'lib' target in parallel 243par:;@$(MAKE) $(MFLAGS) V=$(V) lib 244backend_status = $(if $(filter $1,$(BACKENDS)), [backends: $1], [not found]) 245info: 246 $(info ------------------------------------) 247 $(info CC = $(CC)) 248 $(info CXX = $(CXX)) 249 $(info FC = $(FC)) 250 $(info CPPFLAGS = $(CPPFLAGS)) 251 $(info CFLAGS = $(value CFLAGS)) 252 $(info CXXFLAGS = $(value CXXFLAGS)) 253 $(info FFLAGS = $(value FFLAGS)) 254 $(info NVCCFLAGS = $(value NVCCFLAGS)) 255 $(info LDFLAGS = $(value LDFLAGS)) 256 $(info LDLIBS = $(LDLIBS)) 257 $(info OPT = $(OPT)) 258 $(info AFLAGS = $(AFLAGS)) 259 $(info ASAN = $(or $(ASAN),(empty))) 260 $(info V = $(or $(V),(empty)) [verbose=$(if $(V),on,off)]) 261 $(info ------------------------------------) 262 $(info MEMCHK_STATUS = $(MEMCHK_STATUS)$(call backend_status,/cpu/self/memcheck/serial /cpu/sef/memcheck/blocked)) 263 $(info AVX_STATUS = $(AVX_STATUS)$(call backend_status,/cpu/self/avx/serial /cpu/self/avx/blocked)) 264 $(info XSMM_DIR = $(XSMM_DIR)$(call backend_status,/cpu/self/xsmm/serial /cpu/self/xsmm/blocked)) 265 $(info OCCA_DIR = $(OCCA_DIR)$(call backend_status,/cpu/occa /gpu/occa /omp/occa)) 266 $(info MAGMA_DIR = $(MAGMA_DIR)$(call backend_status,/gpu/magma)) 267 $(info CUDA_DIR = $(CUDA_DIR)$(call backend_status,$(CUDA_BACKENDS))) 268 $(info ------------------------------------) 269 $(info MFEM_DIR = $(MFEM_DIR)) 270 $(info NEK5K_DIR = $(NEK5K_DIR)) 271 $(info PETSC_DIR = $(PETSC_DIR)) 272 $(info ------------------------------------) 273 $(info prefix = $(prefix)) 274 $(info includedir = $(value includedir)) 275 $(info libdir = $(value libdir)) 276 $(info okldir = $(value okldir)) 277 $(info pkgconfigdir = $(value pkgconfigdir)) 278 $(info ------------------------------------) 279 @true 280info-backends: 281 $(info make: 'lib' with optional backends: $(filter-out $(BACKENDS_BUILTIN),$(BACKENDS))) 282.PHONY: lib all par info info-backends 283 284$(libceed) : LDFLAGS += $(if $(DARWIN), -install_name @rpath/$(notdir $(libceed))) 285$(libceed_test) : LDFLAGS += $(if $(DARWIN), -install_name @rpath/$(notdir $(libceed_test))) 286 287# Standard Backends 288libceed.c += $(ref.c) 289libceed.c += $(blocked.c) 290libceed.c += $(opt.c) 291 292# Testing Backends 293test_backends.c := $(template.c) 294TEST_BACKENDS := /cpu/self/tmpl /cpu/self/tmpl/sub 295 296# Memcheck Backend 297MEMCHK_STATUS = Disabled 298MEMCHK := $(shell echo "\#include <valgrind/memcheck.h>" | $(CC) $(CPPFLAGS) -E - >/dev/null 2>&1 && echo 1) 299ifeq ($(MEMCHK),1) 300 MEMCHK_STATUS = Enabled 301 libceed.c += $(ceedmemcheck.c) 302 BACKENDS += /cpu/self/memcheck/serial /cpu/self/memcheck/blocked 303endif 304 305# AVX Backed 306AVX_STATUS = Disabled 307AVX_FLAG := $(if $(filter clang,$(CC_VENDOR)),+avx,-mavx) 308AVX := $(filter $(AVX_FLAG),$(shell $(CC) $(OPT) -v -E -x c /dev/null 2>&1)) 309ifneq ($(AVX),) 310 AVX_STATUS = Enabled 311 libceed.c += $(avx.c) 312 BACKENDS += /cpu/self/avx/serial /cpu/self/avx/blocked 313endif 314 315# libXSMM Backends 316ifneq ($(wildcard $(XSMM_DIR)/lib/libxsmm.*),) 317 $(libceeds) : LDFLAGS += -L$(XSMM_DIR)/lib -Wl,-rpath,$(abspath $(XSMM_DIR)/lib) 318 $(libceeds) : LDLIBS += -lxsmm -ldl 319 MKL ?= 320 ifeq (,$(MKL)$(MKLROOT)) 321 BLAS_LIB = -lblas 322 else 323 ifneq ($(MKLROOT),) 324 # Some installs put everything inside an intel64 subdirectory, others not 325 MKL_LIBDIR = $(dir $(firstword $(wildcard $(MKLROOT)/lib/intel64/libmkl_sequential.* $(MKLROOT)/lib/libmkl_sequential.*))) 326 MKL_LINK = -L$(MKL_LIBDIR) -Wl,-rpath,$(MKL_LIBDIR) 327 endif 328 BLAS_LIB = $(MKL_LINK) -Wl,--no-as-needed -lmkl_intel_lp64 -lmkl_sequential -lmkl_core -lpthread -lm -ldl 329 endif 330 $(libceeds) : LDLIBS += $(BLAS_LIB) 331 libceed.c += $(xsmm.c) 332 $(xsmm.c:%.c=$(OBJDIR)/%.o) $(xsmm.c:%=%.tidy) : CPPFLAGS += -I$(XSMM_DIR)/include 333 BACKENDS += /cpu/self/xsmm/serial /cpu/self/xsmm/blocked 334endif 335 336# OCCA Backends 337ifneq ($(wildcard $(OCCA_DIR)/lib/libocca.*),) 338 $(libceeds) : LDFLAGS += -L$(OCCA_DIR)/lib -Wl,-rpath,$(abspath $(OCCA_DIR)/lib) 339 $(libceeds) : LDLIBS += -locca 340 libceed.c += $(occa.c) 341 $(occa.c:%.c=$(OBJDIR)/%.o) $(occa.c:%=%.tidy) : CPPFLAGS += -I$(OCCA_DIR)/include 342 BACKENDS += /cpu/occa /gpu/occa /omp/occa 343endif 344 345# CUDA Backends 346CUDA_LIB_DIR := $(wildcard $(foreach d,lib lib64,$(CUDA_DIR)/$d/libcudart.${SO_EXT})) 347CUDA_LIB_DIR := $(patsubst %/,%,$(dir $(firstword $(CUDA_LIB_DIR)))) 348CUDA_LIB_DIR_STUBS := $(CUDA_LIB_DIR)/stubs 349CUDA_BACKENDS = /gpu/cuda/ref /gpu/cuda/reg /gpu/cuda/shared /gpu/cuda/gen 350ifneq ($(CUDA_LIB_DIR),) 351 $(libceeds) : CPPFLAGS += -I$(CUDA_DIR)/include 352 $(libceeds) : LDFLAGS += -L$(CUDA_LIB_DIR) -Wl,-rpath,$(abspath $(CUDA_LIB_DIR)) 353 $(libceeds) : LDLIBS += -lcudart -lnvrtc -lcuda 354 $(libceeds) : LINK = $(CXX) 355 libceed.c += $(cuda.c) $(cuda-reg.c) $(cuda-shared.c) $(cuda-gen.c) 356 libceed.cpp += $(cuda.cpp) $(cuda-gen.cpp) 357 libceed.cu += $(cuda.cu) $(cuda-reg.cu) $(cuda-shared.cu) $(cuda-gen.cu) 358 BACKENDS += $(CUDA_BACKENDS) 359endif 360 361# MAGMA Backend 362ifneq ($(wildcard $(MAGMA_DIR)/lib/libmagma.*),) 363 ifneq ($(CUDA_LIB_DIR),) 364 cuda_link = -Wl,-rpath,$(CUDA_LIB_DIR) -L$(CUDA_LIB_DIR) -lcublas -lcusparse -lcudart 365 omp_link = -fopenmp 366 magma_link_static = -L$(MAGMA_DIR)/lib -lmagma $(cuda_link) $(omp_link) 367 magma_link_shared = -L$(MAGMA_DIR)/lib -Wl,-rpath,$(abspath $(MAGMA_DIR)/lib) -lmagma 368 magma_link := $(if $(wildcard $(MAGMA_DIR)/lib/libmagma.${SO_EXT}),$(magma_link_shared),$(magma_link_static)) 369 $(libceeds) : LDLIBS += $(magma_link) 370 $(tests) $(examples) : LDLIBS += $(magma_link) 371 libceed.c += $(magma.c) 372 libceed.cu += $(magma.cu) 373 $(magma.c:%.c=$(OBJDIR)/%.o) $(magma.c:%=%.tidy) : CPPFLAGS += -DADD_ -I$(MAGMA_DIR)/include -I$(CUDA_DIR)/include 374 $(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 375 BACKENDS += /gpu/magma 376 endif 377endif 378 379export BACKENDS 380 381libceed.o = $(libceed.c:%.c=$(OBJDIR)/%.o) $(libceed.cpp:%.cpp=$(OBJDIR)/%.o) $(libceed.cu:%.cu=$(OBJDIR)/%.o) 382$(filter %fortran.o,$(libceed.o)) : CPPFLAGS += $(if $(filter 1,$(UNDERSCORE)),-DUNDERSCORE) 383$(libceed.o): | info-backends 384$(libceed) : $(libceed.o) | $$(@D)/.DIR 385 $(call quiet,LINK) $(LDFLAGS) -shared -o $@ $^ $(LDLIBS) 386 387$(OBJDIR)/%.o : $(CURDIR)/%.c | $$(@D)/.DIR 388 $(call quiet,CC) $(CPPFLAGS) $(CFLAGS) -c -o $@ $(abspath $<) 389 390$(OBJDIR)/%.o : $(CURDIR)/%.cpp | $$(@D)/.DIR 391 $(call quiet,CXX) $(CPPFLAGS) $(CXXFLAGS) -c -o $@ $(abspath $<) 392 393$(OBJDIR)/%.o : $(CURDIR)/%.cu | $$(@D)/.DIR 394 $(call quiet,NVCC) $(CPPFLAGS) $(NVCCFLAGS) -c -o $@ $(abspath $<) 395 396$(OBJDIR)/% : tests/%.c | $$(@D)/.DIR 397 $(call quiet,LINK.c) $(CEED_LDFLAGS) -o $@ $(abspath $<) $(CEED_LIBS) $(LDLIBS) 398 399$(OBJDIR)/% : tests/%.f90 | $$(@D)/.DIR 400 $(call quiet,LINK.F) -DSOURCE_DIR='"$(abspath $(<D))/"' $(CEED_LDFLAGS) -o $@ $(abspath $<) $(CEED_LIBS) $(LDLIBS) 401 402$(OBJDIR)/% : examples/ceed/%.c | $$(@D)/.DIR 403 $(call quiet,LINK.c) $(CEED_LDFLAGS) -o $@ $(abspath $<) $(CEED_LIBS) $(LDLIBS) 404 405$(OBJDIR)/% : examples/ceed/%.f | $$(@D)/.DIR 406 $(call quiet,LINK.F) -DSOURCE_DIR='"$(abspath $(<D))/"' $(CEED_LDFLAGS) -o $@ $(abspath $<) $(CEED_LIBS) $(LDLIBS) 407 408$(OBJDIR)/mfem-% : examples/mfem/%.cpp $(libceed) | $$(@D)/.DIR 409 +$(MAKE) -C examples/mfem CEED_DIR=`pwd` \ 410 MFEM_DIR="$(abspath $(MFEM_DIR))" CXX=$(CXX) $* 411 mv examples/mfem/$* $@ 412 413# Note: Multiple Nek files cannot be built in parallel. The '+' here enables 414# this single Nek bps file to be built in parallel with other examples, 415# such as when calling `make prove-all -j2`. 416$(OBJDIR)/nek-bps : examples/nek/bps/bps.usr examples/nek/nek-examples.sh $(libceed) | $$(@D)/.DIR 417 +$(MAKE) -C examples MPI=$(MPI) CEED_DIR=`pwd` NEK5K_DIR="$(abspath $(NEK5K_DIR))" nek 418 mv examples/nek/build/bps $(OBJDIR)/bps 419 cp examples/nek/nek-examples.sh $(OBJDIR)/nek-bps 420 421$(OBJDIR)/petsc-% : examples/petsc/%.c $(libceed) $(ceed.pc) | $$(@D)/.DIR 422 +$(MAKE) -C examples/petsc CEED_DIR=`pwd` \ 423 PETSC_DIR="$(abspath $(PETSC_DIR))" OPT="$(OPT)" $* 424 mv examples/petsc/$* $@ 425 426$(OBJDIR)/fluids-% : examples/fluids/%.c $(libceed) $(ceed.pc) | $$(@D)/.DIR 427 +$(MAKE) -C examples/fluids CEED_DIR=`pwd` \ 428 PETSC_DIR="$(abspath $(PETSC_DIR))" OPT="$(OPT)" $* 429 mv examples/fluids/$* $@ 430 431$(OBJDIR)/solids-% : examples/solids/%.c $(libceed) $(ceed.pc) | $$(@D)/.DIR 432 +$(MAKE) -C examples/solids CEED_DIR=`pwd` \ 433 PETSC_DIR="$(abspath $(PETSC_DIR))" OPT="$(OPT)" $* 434 mv examples/solids/$* $@ 435 436libceed_test.o = $(test_backends.c:%.c=$(OBJDIR)/%.o) 437$(libceed_test) : $(libceed.o) $(libceed_test.o) | $$(@D)/.DIR 438 $(call quiet,LINK) $(LDFLAGS) -shared -o $@ $^ $(LDLIBS) 439 440$(examples) : $(libceed) 441$(tests) : $(libceed_test) 442$(tests) : CEED_LIBS = -lceed_test 443$(tests) $(examples) : LDFLAGS += -Wl,-rpath,$(abspath $(LIBDIR)) -L$(LIBDIR) 444 445run-t% : BACKENDS += $(TEST_BACKENDS) 446run-% : $(OBJDIR)/% 447 @tests/tap.sh $(<:$(OBJDIR)/%=%) 448 449external_examples := \ 450 $(if $(MFEM_DIR),$(mfemexamples)) \ 451 $(if $(PETSC_DIR),$(petscexamples)) \ 452 $(if $(NEK5K_DIR),$(nekexamples)) \ 453 $(if $(PETSC_DIR),$(fluidsexamples)) \ 454 $(if $(PETSC_DIR),$(solidsexamples)) 455 456allexamples = $(examples) $(external_examples) 457 458# The test and prove targets can be controlled via pattern searches. The 459# default is to run tests and those examples that have no external dependencies. 460# Examples of finer grained control: 461# 462# make test search='petsc mfem' # PETSc and MFEM examples 463# make prove search='t3' # t3xx series tests 464# make junit search='ex petsc' # core ex* and PETSc tests 465search ?= t ex 466realsearch = $(search:%=%%) 467matched = $(foreach pattern,$(realsearch),$(filter $(OBJDIR)/$(pattern),$(tests) $(allexamples))) 468 469# Test core libCEED 470test : $(matched:$(OBJDIR)/%=run-%) 471 472# Run test target in parallel 473tst : ;@$(MAKE) $(MFLAGS) V=$(V) test 474# CPU C tests only for backend % 475ctc-% : $(ctests);@$(foreach tst,$(ctests),$(tst) /cpu/$*;) 476 477prove : BACKENDS += $(TEST_BACKENDS) 478prove : $(matched) 479 $(info Testing backends: $(BACKENDS)) 480 $(PROVE) $(PROVE_OPTS) --exec 'tests/tap.sh' $(matched:$(OBJDIR)/%=%) 481# Run prove target in parallel 482prv : ;@$(MAKE) $(MFLAGS) V=$(V) prove 483 484prove-all : 485 +$(MAKE) prove realsearch=% 486 487junit-t% : BACKENDS += $(TEST_BACKENDS) 488junit-% : $(OBJDIR)/% 489 @printf " %10s %s\n" TEST $(<:$(OBJDIR)/%=%); $(PYTHON) tests/junit.py $(<:$(OBJDIR)/%=%) 490 491junit : $(matched:$(OBJDIR)/%=junit-%) 492 493all: $(alltests) 494 495examples : $(allexamples) 496ceedexamples : $(examples) 497nekexamples : $(nekexamples) 498mfemexamples : $(mfemexamples) 499petscexamples : $(petscexamples) 500 501# Benchmarks 502allbenchmarks = petsc-bps 503bench_targets = $(addprefix bench-,$(allbenchmarks)) 504.PHONY: $(bench_targets) benchmarks 505$(bench_targets): bench-%: $(OBJDIR)/% 506 cd benchmarks && ./benchmark.sh --ceed "$(BACKENDS)" -r $(*).sh 507benchmarks: $(bench_targets) 508 509$(ceed.pc) : pkgconfig-prefix = $(abspath .) 510$(OBJDIR)/ceed.pc : pkgconfig-prefix = $(prefix) 511.INTERMEDIATE : $(OBJDIR)/ceed.pc 512%/ceed.pc : ceed.pc.template | $$(@D)/.DIR 513 @sed "s:%prefix%:$(pkgconfig-prefix):" $< > $@ 514 515OCCA := $(OCCA_DIR)/bin/occa 516OKL_KERNELS := $(wildcard backends/occa/*.okl) 517 518okl-cache : 519 $(OCCA) cache ceed $(OKL_KERNELS) 520 521okl-clear: 522 $(OCCA) clear -y -l ceed 523 524install : $(libceed) $(OBJDIR)/ceed.pc 525 $(INSTALL) -d $(addprefix $(if $(DESTDIR),"$(DESTDIR)"),"$(includedir)"\ 526 "$(libdir)" "$(pkgconfigdir)" $(if $(OCCA_ON),"$(okldir)")) 527 $(INSTALL_DATA) include/ceed.h "$(DESTDIR)$(includedir)/" 528 $(INSTALL_DATA) include/ceedf.h "$(DESTDIR)$(includedir)/" 529 $(INSTALL_DATA) $(libceed) "$(DESTDIR)$(libdir)/" 530 $(INSTALL_DATA) $(OBJDIR)/ceed.pc "$(DESTDIR)$(pkgconfigdir)/" 531 $(if $(OCCA_ON),$(INSTALL_DATA) $(OKL_KERNELS) "$(DESTDIR)$(okldir)/") 532 533.PHONY : cln clean doxygen doc lib install all print test tst prove prv prove-all junit examples style style-c style-py tidy okl-cache okl-clear info info-backends 534 535cln clean : 536 $(RM) -r $(OBJDIR) $(LIBDIR) dist *egg* .pytest_cache *cffi* 537 $(MAKE) -C examples clean NEK5K_DIR="$(abspath $(NEK5K_DIR))" 538 $(MAKE) -C tests/python clean 539 $(RM) benchmarks/*output.txt 540 541distclean : clean 542 $(RM) -r doc/html doc/sphinx/build $(CONFIG) 543 544DOXYGEN ?= doxygen 545doxygen : 546 $(DOXYGEN) Doxyfile 547 548doc-html doc-latexpdf doc-epub : doc-% : doxygen 549 make -C doc/sphinx $* 550 551doc : doc-html 552 553style-c : 554 @astyle --options=.astylerc \ 555 $(filter-out include/ceedf.h tests/t320-basis-f.h, \ 556 $(wildcard include/*.h interface/*.[ch] tests/*.[ch] backends/*/*.[ch] \ 557 examples/*/*/*.[ch] examples/*/*.[ch] examples/*/*.[ch]pp gallery/*/*.[ch])) 558 559AUTOPEP8 = autopep8 560style-py : AUTOPEP8_ARGS = --in-place --aggressive 561style-py : 562 @$(AUTOPEP8) $(AUTOPEP8_ARGS) $(wildcard *.py python**/*.py tests/python**/*.py examples**/*.py doc/sphinx/source**/*.py benchmarks/*.py) 563 564style : style-c style-py 565 566CLANG_TIDY ?= clang-tidy 567%.c.tidy : %.c 568 $(CLANG_TIDY) $(TIDY_OPTS) $^ -- $(CPPFLAGS) --std=c99 -I$(CUDA_DIR)/include 569 570tidy : $(libceed.c:%=%.tidy) 571 572print : 573 @echo $(VAR)=$($(VAR)) 574 575print-% : 576 $(info [ variable name]: $*) 577 $(info [ origin]: $(origin $*)) 578 $(info [ flavor]: $(flavor $*)) 579 $(info [ value]: $(value $*)) 580 $(info [expanded value]: $($*)) 581 $(info ) 582 @true 583 584# "make configure" detects any variables passed on the command line or 585# previously set in config.mk, caching them in config.mk as simple 586# (:=) variables. Variables set in config.mk or on the command line 587# take precedence over the defaults provided in the file. Typical 588# usage: 589# 590# make configure CC=/path/to/my/cc CUDA_DIR=/opt/cuda 591# make 592# make prove 593# 594# The values in the file can be updated by passing them on the command 595# line, e.g., 596# 597# make configure CC=/path/to/other/clang 598 599# All variables to consider for caching 600CONFIG_VARS = CC CXX FC NVCC NVCC_CXX \ 601 OPT CFLAGS CPPFLAGS CXXFLAGS FFLAGS NVCCFLAGS \ 602 LDFLAGS LDLIBS \ 603 MAGMA_DIR XSMM_DIR CUDA_DIR MFEM_DIR PETSC_DIR NEK5K_DIR 604 605# $(call needs_save,CFLAGS) returns true (a nonempty string) if CFLAGS 606# was set on the command line or in config.mk (where it will appear as 607# a simple variable). 608needs_save = $(or $(filter command line,$(origin $(1))),$(filter simple,$(flavor $(1)))) 609 610configure : 611 $(file > $(CONFIG)) 612 $(foreach v,$(CONFIG_VARS),$(if $(call needs_save,$(v)),$(file >> $(CONFIG),$(v) := $($(v))))) 613 @echo "Configuration cached in $(CONFIG):" 614 @cat $(CONFIG) 615 616wheel : export MARCHFLAG = -march=generic 617wheel : export WHEEL_PLAT = manylinux2010_x86_64 618wheel : 619 docker run -it --user $(shell id -u):$(shell id -g) --rm -v $(PWD):/io -w /io \ 620 -e MARCHFLAG -e WHEEL_PLAT \ 621 quay.io/pypa/$(WHEEL_PLAT) python/make-wheels.sh 622 623.PHONY : configure wheel 624 625# Include *.d deps when not -B = --always-make: useful if the paths are wonky in a container 626-include $(if $(filter B,$(MAKEFLAGS)),,$(libceed.c:%.c=$(OBJDIR)/%.d) $(tests.c:tests/%.c=$(OBJDIR)/%.d)) 627