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 17-include config.mk 18 19ifeq (,$(filter-out undefined default,$(origin CC))) 20 CC = gcc 21endif 22ifeq (,$(filter-out undefined default,$(origin CXX))) 23 CXX = g++ 24endif 25ifeq (,$(filter-out undefined default,$(origin FC))) 26 FC = gfortran 27endif 28ifeq (,$(filter-out undefined default,$(origin LINK))) 29 LINK = $(CC) 30endif 31NVCC ?= $(CUDA_DIR)/bin/nvcc 32NVCC_CXX ?= $(CXX) 33 34# ASAN must be left empty if you don't want to use it 35ASAN ?= 36 37LDFLAGS ?= 38UNDERSCORE ?= 1 39 40# MFEM_DIR env variable should point to sibling directory 41ifneq ($(wildcard ../mfem/libmfem.*),) 42 MFEM_DIR ?= ../mfem 43endif 44 45# NEK5K_DIR env variable should point to sibling directory 46ifneq ($(wildcard ../Nek5000/*),) 47 NEK5K_DIR ?= $(abspath ../Nek5000) 48endif 49export NEK5K_DIR 50MPI ?= 1 51 52# CEED_DIR env for NEK5K testing 53export CEED_DIR = $(abspath .) 54 55# XSMM_DIR env variable should point to XSMM master (github.com/hfp/libxsmm) 56XSMM_DIR ?= ../libxsmm 57 58# OCCA_DIR env variable should point to OCCA master (github.com/libocca/occa) 59OCCA_DIR ?= ../occa 60 61# env variable MAGMA_DIR can be used too 62MAGMA_DIR ?= ../magma 63# If CUDA_DIR is not set, check for nvcc, or resort to /usr/local/cuda 64CUDA_DIR ?= $(or $(patsubst %/,%,$(dir $(patsubst %/,%,$(dir \ 65 $(shell which nvcc 2> /dev/null))))),/usr/local/cuda) 66 67# Check for PETSc in ../petsc 68ifneq ($(wildcard ../petsc/lib/libpetsc.*),) 69 PETSC_DIR ?= ../petsc 70endif 71 72# Warning: SANTIZ options still don't run with /gpu/occa 73# export LSAN_OPTIONS=suppressions=.asanignore 74AFLAGS = -fsanitize=address #-fsanitize=undefined -fno-omit-frame-pointer 75 76MARCHFLAG := $(if $(shell $(CC) -E -march=native -x c /dev/null > /dev/null 2>&1 && echo 1),-march=native,-mtune=native) 77 78OPT = -O -g $(MARCHFLAG) -ffp-contract=fast -fopenmp-simd 79CFLAGS = -std=c99 $(OPT) -Wall -Wextra -Wno-unused-parameter -fPIC -MMD -MP 80CXXFLAGS = -std=c++11 $(OPT) -Wall -Wextra -Wno-unused-parameter -fPIC -MMD -MP 81NVCCFLAGS = -ccbin $(CXX) -Xcompiler "$(OPT)" -Xcompiler -fPIC 82# If using the IBM XL Fortran (xlf) replace FFLAGS appropriately: 83ifneq ($(filter %xlf %xlf_r,$(FC)),) 84 FFLAGS = $(OPT) -ffree-form -qpreprocess -qextname -qpic -MMD -DSOURCE_DIR='"$(abspath $(<D))/"' 85else # gfortran/Intel-style options 86 FFLAGS = -cpp $(OPT) -Wall -Wextra -Wno-unused-parameter -Wno-unused-dummy-argument -fPIC -MMD -MP -DSOURCE_DIR='"$(abspath $(<D))/"' 87endif 88 89ifeq ($(UNDERSCORE), 1) 90 CPPFLAGS += -DUNDERSCORE 91endif 92 93ifeq ($(COVERAGE), 1) 94 CFLAGS += --coverage 95 LDFLAGS += --coverage 96endif 97 98CFLAGS += $(if $(ASAN),$(AFLAGS)) 99FFLAGS += $(if $(ASAN),$(AFLAGS)) 100LDFLAGS += $(if $(ASAN),$(AFLAGS)) 101CPPFLAGS += -I./include 102LDLIBS = -lm 103OBJDIR := build 104LIBDIR := lib 105 106# Installation variables 107prefix ?= /usr/local 108bindir = $(prefix)/bin 109libdir = $(prefix)/lib 110okldir = $(libdir)/okl 111includedir = $(prefix)/include 112pkgconfigdir = $(libdir)/pkgconfig 113INSTALL = install 114INSTALL_PROGRAM = $(INSTALL) 115INSTALL_DATA = $(INSTALL) -m644 116 117# Get number of processors of the machine 118NPROCS := $(shell getconf _NPROCESSORS_ONLN) 119# prepare make options to run in parallel 120MFLAGS := -j $(NPROCS) --warn-undefined-variables \ 121 --no-print-directory --no-keep-going 122 123PYTHON ?= python3 124PROVE ?= prove 125PROVE_OPTS ?= -j $(NPROCS) 126DARWIN := $(filter Darwin,$(shell uname -s)) 127SO_EXT := $(if $(DARWIN),dylib,so) 128 129ceed.pc := $(LIBDIR)/pkgconfig/ceed.pc 130libceed := $(LIBDIR)/libceed.$(SO_EXT) 131CEED_LIBS = -lceed 132libceed.c := $(wildcard interface/ceed*.c) 133gallery.c := $(wildcard gallery/*/ceed*.c) 134libceed.c += $(gallery.c) 135libceed_test := $(LIBDIR)/libceed_test.$(SO_EXT) 136libceeds = $(libceed) $(libceed_test) 137BACKENDS_BUILTIN := /cpu/self/ref/serial /cpu/self/ref/blocked /cpu/self/opt/serial /cpu/self/opt/blocked 138BACKENDS := $(BACKENDS_BUILTIN) 139 140# Tests 141tests.c := $(sort $(wildcard tests/t[0-9][0-9][0-9]-*.c)) 142tests.f := $(sort $(wildcard tests/t[0-9][0-9][0-9]-*.f90)) 143tests := $(tests.c:tests/%.c=$(OBJDIR)/%) 144ctests := $(tests) 145tests += $(tests.f:tests/%.f90=$(OBJDIR)/%) 146# Examples 147examples.c := $(sort $(wildcard examples/ceed/*.c)) 148examples.f := $(sort $(wildcard examples/ceed/*.f)) 149examples := $(examples.c:examples/ceed/%.c=$(OBJDIR)/%) 150examples += $(examples.f:examples/ceed/%.f=$(OBJDIR)/%) 151# MFEM Examples 152mfemexamples.cpp := $(sort $(wildcard examples/mfem/*.cpp)) 153mfemexamples := $(mfemexamples.cpp:examples/mfem/%.cpp=$(OBJDIR)/mfem-%) 154# Nek5K Examples 155nekexamples := $(OBJDIR)/nek-bps 156# PETSc Examples 157petscexamples.c := $(wildcard examples/petsc/*.c) 158petscexamples := $(petscexamples.c:examples/petsc/%.c=$(OBJDIR)/petsc-%) 159# Navier-Stokes Examples 160nsexamples.c := $(sort $(wildcard examples/navier-stokes/*.c)) 161nsexamples := $(nsexamples.c:examples/navier-stokes/%.c=$(OBJDIR)/ns-%) 162 163# Backends/[ref, blocked, template, memcheck, opt, avx, occa, magma] 164ref.c := $(sort $(wildcard backends/ref/*.c)) 165blocked.c := $(sort $(wildcard backends/blocked/*.c)) 166template.c := $(sort $(wildcard backends/template/*.c)) 167ceedmemcheck.c := $(sort $(wildcard backends/memcheck/*.c)) 168opt.c := $(sort $(wildcard backends/opt/*.c)) 169avx.c := $(sort $(wildcard backends/avx/*.c)) 170xsmm.c := $(sort $(wildcard backends/xsmm/*.c)) 171cuda.c := $(sort $(wildcard backends/cuda/*.c)) 172cuda.cpp := $(sort $(wildcard backends/cuda/*.cpp)) 173cuda.cu := $(sort $(wildcard backends/cuda/*.cu)) 174cuda-reg.c := $(sort $(wildcard backends/cuda-reg/*.c)) 175cuda-reg.cu := $(sort $(wildcard backends/cuda-reg/*.cu)) 176cuda-shared.c := $(sort $(wildcard backends/cuda-shared/*.c)) 177cuda-shared.cu := $(sort $(wildcard backends/cuda-shared/*.cu)) 178cuda-gen.c := $(sort $(wildcard backends/cuda-gen/*.c)) 179cuda-gen.cpp := $(sort $(wildcard backends/cuda-gen/*.cpp)) 180cuda-gen.cu := $(sort $(wildcard backends/cuda-gen/*.cu)) 181occa.c := $(sort $(wildcard backends/occa/*.c)) 182magma_preprocessor := python backends/magma/gccm.py 183magma_pre_src := $(filter-out %ceed-magma.c %_tmp.c, $(wildcard backends/magma/ceed-*.c)) 184magma_dsrc := $(wildcard backends/magma/magma_d*.c) 185magma_dsrc += backends/magma/ceed-magma.c 186magma_tmp.c := $(magma_pre_src:%.c=%_tmp.c) 187magma_tmp.cu := $(magma_pre_src:%.c=%_cuda.cu) 188magma_allsrc.c := $(magma_dsrc) $(magma_tmp.c) 189magma_allsrc.cu:= $(magma_tmp.cu) backends/magma/magma_devptr.cu 190magma_allsrc.cu+= backends/magma/magma_dbasisApply_grad.cu backends/magma/magma_dbasisApply_interp.cu backends/magma/magma_dbasisApply_weight.cu 191magma_allsrc.cu+= backends/magma/magma_drestrictApply.cu 192 193# Output using the 216-color rules mode 194rule_file = $(notdir $(1)) 195rule_path = $(patsubst %/,%,$(dir $(1))) 196last_path = $(notdir $(patsubst %/,%,$(dir $(1)))) 197ansicolor = $(shell echo $(call last_path,$(1)) | cksum | cut -b1-2 | xargs -IS expr 2 \* S + 17) 198emacs_out = @printf " %10s %s/%s\n" $(1) $(call rule_path,$(2)) $(call rule_file,$(2)) 199color_out = @if [ -t 1 ]; then \ 200 printf " %10s \033[38;5;%d;1m%s\033[m/%s\n" \ 201 $(1) $(call ansicolor,$(2)) \ 202 $(call rule_path,$(2)) $(call rule_file,$(2)); else \ 203 printf " %10s %s\n" $(1) $(2); fi 204# if TERM=dumb, use it, otherwise switch to the term one 205output = $(if $(TERM:dumb=),$(call color_out,$1,$2),$(call emacs_out,$1,$2)) 206 207# if V is set to non-nil, turn the verbose mode 208quiet = $(if $(V),$($(1)),$(call output,$1,$@);$($(1))) 209 210# Cancel built-in and old-fashioned implicit rules which we don't use 211.SUFFIXES: 212 213.SECONDEXPANSION: # to expand $$(@D)/.DIR 214 215.SECONDARY: $(magma_tmp.c) $(magma_tmp.cu) 216 217%/.DIR : 218 @mkdir -p $(@D) 219 @touch $@ 220 221.PRECIOUS: %/.DIR 222 223lib: $(libceed) $(ceed.pc) 224# run 'lib' target in parallel 225par:;@$(MAKE) $(MFLAGS) V=$(V) lib 226backend_status = $(if $(filter $1,$(BACKENDS)), [backends: $1], [not found]) 227info: 228 $(info ------------------------------------) 229 $(info CC = $(CC)) 230 $(info CXX = $(CXX)) 231 $(info FC = $(FC)) 232 $(info CPPFLAGS = $(CPPFLAGS)) 233 $(info CFLAGS = $(value CFLAGS)) 234 $(info CXXFLAGS = $(value CXXFLAGS)) 235 $(info FFLAGS = $(value FFLAGS)) 236 $(info NVCCFLAGS = $(value NVCCFLAGS)) 237 $(info LDFLAGS = $(value LDFLAGS)) 238 $(info LDLIBS = $(LDLIBS)) 239 $(info OPT = $(OPT)) 240 $(info AFLAGS = $(AFLAGS)) 241 $(info ASAN = $(or $(ASAN),(empty))) 242 $(info V = $(or $(V),(empty)) [verbose=$(if $(V),on,off)]) 243 $(info ------------------------------------) 244 $(info MEMCHK_STATUS = $(MEMCHK_STATUS)$(call backend_status,/cpu/self/memcheck/serial /cpu/sef/memcheck/blocked)) 245 $(info AVX_STATUS = $(AVX_STATUS)$(call backend_status,/cpu/self/avx/serial /cpu/self/avx/blocked)) 246 $(info XSMM_DIR = $(XSMM_DIR)$(call backend_status,/cpu/self/xsmm/serial /cpu/self/xsmm/blocked)) 247 $(info OCCA_DIR = $(OCCA_DIR)$(call backend_status,/cpu/occa /gpu/occa /omp/occa)) 248 $(info MAGMA_DIR = $(MAGMA_DIR)$(call backend_status,/gpu/magma)) 249 $(info CUDA_DIR = $(CUDA_DIR)$(call backend_status,$(CUDA_BACKENDS))) 250 $(info ------------------------------------) 251 $(info MFEM_DIR = $(MFEM_DIR)) 252 $(info NEK5K_DIR = $(NEK5K_DIR)) 253 $(info PETSC_DIR = $(PETSC_DIR)) 254 $(info ------------------------------------) 255 $(info prefix = $(prefix)) 256 $(info includedir = $(value includedir)) 257 $(info libdir = $(value libdir)) 258 $(info okldir = $(value okldir)) 259 $(info pkgconfigdir = $(value pkgconfigdir)) 260 $(info ------------------------------------) 261 @true 262info-backends: 263 $(info make: 'lib' with optional backends: $(filter-out $(BACKENDS_BUILTIN),$(BACKENDS))) 264.PHONY: lib all par info info-backends 265 266$(libceed) : LDFLAGS += $(if $(DARWIN), -install_name @rpath/$(notdir $(libceed))) 267$(libceed_test) : LDFLAGS += $(if $(DARWIN), -install_name @rpath/$(notdir $(libceed_test))) 268 269# Standard Backends 270libceed.c += $(ref.c) 271libceed.c += $(blocked.c) 272libceed.c += $(opt.c) 273 274# Testing Backends 275test_backends.c := $(template.c) 276TEST_BACKENDS := /cpu/self/tmpl /cpu/self/tmpl/sub 277 278# Memcheck Backend 279MEMCHK_STATUS = Disabled 280MEMCHK := $(shell echo "\#include <valgrind/memcheck.h>" | $(CC) $(CPPFLAGS) -E - >/dev/null 2>&1 && echo 1) 281ifeq ($(MEMCHK),1) 282 MEMCHK_STATUS = Enabled 283 libceed.c += $(ceedmemcheck.c) 284 BACKENDS += /cpu/self/memcheck/serial /cpu/self/memcheck/blocked 285endif 286 287# AVX Backed 288AVX_STATUS = Disabled 289AVX := $(shell $(CC) $(OPT) -v -E - < /dev/null 2>&1 | grep -c ' -mavx') 290ifeq ($(AVX),1) 291 AVX_STATUS = Enabled 292 libceed.c += $(avx.c) 293 BACKENDS += /cpu/self/avx/serial /cpu/self/avx/blocked 294endif 295 296# libXSMM Backends 297ifneq ($(wildcard $(XSMM_DIR)/lib/libxsmm.*),) 298 $(libceeds) : LDFLAGS += -L$(XSMM_DIR)/lib -Wl,-rpath,$(abspath $(XSMM_DIR)/lib) 299 $(libceeds) : LDLIBS += -lxsmm -ldl 300 MKL ?= 301 ifeq (,$(MKL)$(MKLROOT)) 302 BLAS_LIB = -lblas 303 else 304 ifneq ($(MKLROOT),) 305 # Some installs put everything inside an intel64 subdirectory, others not 306 MKL_LIBDIR = $(dir $(firstword $(wildcard $(MKLROOT)/lib/intel64/libmkl_sequential.* $(MKLROOT)/lib/libmkl_sequential.*))) 307 MKL_LINK = -L$(MKL_LIBDIR) -Wl,-rpath,$(MKL_LIBDIR) 308 endif 309 BLAS_LIB = $(MKL_LINK) -Wl,--no-as-needed -lmkl_intel_lp64 -lmkl_sequential -lmkl_core -lpthread -lm -ldl 310 endif 311 $(libceeds) : LDLIBS += $(BLAS_LIB) 312 libceed.c += $(xsmm.c) 313 $(xsmm.c:%.c=$(OBJDIR)/%.o) $(xsmm.c:%=%.tidy) : CPPFLAGS += -I$(XSMM_DIR)/include 314 BACKENDS += /cpu/self/xsmm/serial /cpu/self/xsmm/blocked 315endif 316 317# OCCA Backends 318ifneq ($(wildcard $(OCCA_DIR)/lib/libocca.*),) 319 $(libceeds) : LDFLAGS += -L$(OCCA_DIR)/lib -Wl,-rpath,$(abspath $(OCCA_DIR)/lib) 320 $(libceeds) : LDLIBS += -locca 321 libceed.c += $(occa.c) 322 $(occa.c:%.c=$(OBJDIR)/%.o) $(occa.c:%=%.tidy) : CPPFLAGS += -I$(OCCA_DIR)/include 323 BACKENDS += /cpu/occa /gpu/occa /omp/occa 324endif 325 326# CUDA Backends 327CUDA_LIB_DIR := $(wildcard $(foreach d,lib lib64,$(CUDA_DIR)/$d/libcudart.${SO_EXT})) 328CUDA_LIB_DIR := $(patsubst %/,%,$(dir $(firstword $(CUDA_LIB_DIR)))) 329CUDA_LIB_DIR_STUBS := $(CUDA_LIB_DIR)/stubs 330CUDA_BACKENDS = /gpu/cuda/ref /gpu/cuda/reg /gpu/cuda/shared /gpu/cuda/gen 331ifneq ($(CUDA_LIB_DIR),) 332 $(libceeds) : CFLAGS += -I$(CUDA_DIR)/include 333 $(libceeds) : CPPFLAGS += -I$(CUDA_DIR)/include 334 $(libceeds) : LDFLAGS += -L$(CUDA_LIB_DIR) -Wl,-rpath,$(abspath $(CUDA_LIB_DIR)) 335 $(libceeds) : LDLIBS += -lcudart -lnvrtc -lcuda 336 $(libceeds) : LINK = $(CXX) 337 libceed.c += $(cuda.c) $(cuda-reg.c) $(cuda-shared.c) $(cuda-gen.c) 338 libceed.cpp += $(cuda.cpp) $(cuda-gen.cpp) 339 libceed.cu += $(cuda.cu) $(cuda-reg.cu) $(cuda-shared.cu) $(cuda-gen.cu) 340 BACKENDS += $(CUDA_BACKENDS) 341endif 342 343# MAGMA Backend 344ifneq ($(wildcard $(MAGMA_DIR)/lib/libmagma.*),) 345 ifneq ($(CUDA_LIB_DIR),) 346 cuda_link = -Wl,-rpath,$(CUDA_LIB_DIR) -L$(CUDA_LIB_DIR) -lcublas -lcusparse -lcudart 347 omp_link = -fopenmp 348 magma_link_static = -L$(MAGMA_DIR)/lib -lmagma $(cuda_link) $(omp_link) 349 magma_link_shared = -L$(MAGMA_DIR)/lib -Wl,-rpath,$(abspath $(MAGMA_DIR)/lib) -lmagma 350 magma_link := $(if $(wildcard $(MAGMA_DIR)/lib/libmagma.${SO_EXT}),$(magma_link_shared),$(magma_link_static)) 351 $(libceeds) : LDLIBS += $(magma_link) 352 $(tests) $(examples) : LDLIBS += $(magma_link) 353 libceed.c += $(magma_allsrc.c) 354 libceed.cu += $(magma_allsrc.cu) 355 $(magma_allsrc.c:%.c=$(OBJDIR)/%.o) $(magma_allsrc.c:%=%.tidy) : CPPFLAGS += -DADD_ -I$(MAGMA_DIR)/include -I$(CUDA_DIR)/include 356 $(magma_allsrc.cu:%.cu=$(OBJDIR)/%.o) : NVCCFLAGS += --compiler-options=-fPIC -DADD_ -I$(MAGMA_DIR)/include -I$(MAGMA_DIR)/magmablas -I$(MAGMA_DIR)/control -I$(CUDA_DIR)/include 357 BACKENDS += /gpu/magma 358 endif 359endif 360 361export BACKENDS 362 363# Generate magma_tmp.c and magma_cuda.cu from magma.c 364%_tmp.c %_cuda.cu : %.c 365 $(magma_preprocessor) $< 366 367libceed.o = $(libceed.c:%.c=$(OBJDIR)/%.o) $(libceed.cpp:%.cpp=$(OBJDIR)/%.o) $(libceed.cu:%.cu=$(OBJDIR)/%.o) 368$(libceed.o): | info-backends 369$(libceed) : $(libceed.o) | $$(@D)/.DIR 370 $(call quiet,LINK) $(LDFLAGS) -shared -o $@ $^ $(LDLIBS) 371 372$(OBJDIR)/%.o : $(CURDIR)/%.c | $$(@D)/.DIR 373 $(call quiet,CC) $(CPPFLAGS) $(CFLAGS) -c -o $@ $(abspath $<) 374 375$(OBJDIR)/%.o : $(CURDIR)/%.cpp | $$(@D)/.DIR 376 $(call quiet,CXX) $(CPPFLAGS) $(CXXFLAGS) -c -o $@ $(abspath $<) 377 378$(OBJDIR)/%.o : $(CURDIR)/%.cu | $$(@D)/.DIR 379 $(call quiet,NVCC) $(CPPFLAGS) $(NVCCFLAGS) -c -o $@ $(abspath $<) 380 381$(OBJDIR)/% : tests/%.c | $$(@D)/.DIR 382 $(call quiet,LINK.c) -o $@ $(abspath $<) $(CEED_LIBS) $(LDLIBS) 383 384$(OBJDIR)/% : tests/%.f90 | $$(@D)/.DIR 385 $(call quiet,LINK.F) -o $@ $(abspath $<) $(CEED_LIBS) $(LDLIBS) 386 387$(OBJDIR)/% : examples/ceed/%.c | $$(@D)/.DIR 388 $(call quiet,LINK.c) -o $@ $(abspath $<) $(CEED_LIBS) $(LDLIBS) 389 390$(OBJDIR)/% : examples/ceed/%.f | $$(@D)/.DIR 391 $(call quiet,LINK.F) -o $@ $(abspath $<) $(CEED_LIBS) $(LDLIBS) 392 393$(OBJDIR)/mfem-% : examples/mfem/%.cpp $(libceed) | $$(@D)/.DIR 394 +$(MAKE) -C examples/mfem CEED_DIR=`pwd` \ 395 MFEM_DIR="$(abspath $(MFEM_DIR))" $* 396 mv examples/mfem/$* $@ 397 398# Note: Multiple Nek files cannot be built in parallel. The '+' here enables 399# this single Nek bps file to be built in parallel with other examples, 400# such as when calling `make prove-all -j2`. 401$(OBJDIR)/nek-bps : examples/nek/bps/bps.usr examples/nek/nek-examples.sh $(libceed) | $$(@D)/.DIR 402 +$(MAKE) -C examples MPI=$(MPI) CEED_DIR=`pwd` NEK5K_DIR="$(abspath $(NEK5K_DIR))" nek 403 mv examples/nek/build/bps $(OBJDIR)/bps 404 cp examples/nek/nek-examples.sh $(OBJDIR)/nek-bps 405 406$(OBJDIR)/petsc-% : examples/petsc/%.c $(libceed) $(ceed.pc) | $$(@D)/.DIR 407 +$(MAKE) -C examples/petsc CEED_DIR=`pwd` \ 408 PETSC_DIR="$(abspath $(PETSC_DIR))" $* 409 mv examples/petsc/$* $@ 410 411$(OBJDIR)/ns-% : examples/navier-stokes/%.c $(libceed) $(ceed.pc) | $$(@D)/.DIR 412 +$(MAKE) -C examples/navier-stokes CEED_DIR=`pwd` \ 413 PETSC_DIR="$(abspath $(PETSC_DIR))" $* 414 mv examples/navier-stokes/$* $@ 415 416libceed_test.o = $(test_backends.c:%.c=$(OBJDIR)/%.o) 417$(libceed_test) : $(libceed.o) $(libceed_test.o) | $$(@D)/.DIR 418 $(call quiet,LINK) $(LDFLAGS) -shared -o $@ $^ $(LDLIBS) 419 420$(examples) : $(libceed) 421$(tests) : $(libceed_test) 422$(tests) : CEED_LIBS = -lceed_test 423$(tests) $(examples) : LDFLAGS += -Wl,-rpath,$(abspath $(LIBDIR)) -L$(LIBDIR) 424 425run-t% : BACKENDS += $(TEST_BACKENDS) 426run-% : $(OBJDIR)/% 427 @tests/tap.sh $(<:$(OBJDIR)/%=%) 428 429external_examples := \ 430 $(if $(MFEM_DIR),$(mfemexamples)) \ 431 $(if $(PETSC_DIR),$(petscexamples)) \ 432 $(if $(NEK5K_DIR),$(nekexamples)) \ 433 $(if $(PETSC_DIR),$(nsexamples)) 434 435allexamples = $(examples) $(external_examples) 436 437# The test and prove targets can be controlled via pattern searches. The 438# default is to run tests and those examples that have no external dependencies. 439# Examples of finer grained control: 440# 441# make test search='petsc mfem' # PETSc and MFEM examples 442# make prove search='t3' # t3xx series tests 443# make junit search='ex petsc' # core ex* and PETSc tests 444search ?= t ex 445realsearch = $(search:%=%%) 446matched = $(foreach pattern,$(realsearch),$(filter $(OBJDIR)/$(pattern),$(tests) $(allexamples))) 447 448# Test core libCEED 449test : $(matched:$(OBJDIR)/%=run-%) 450 451# Run test target in parallel 452tst : ;@$(MAKE) $(MFLAGS) V=$(V) test 453# CPU C tests only for backend % 454ctc-% : $(ctests);@$(foreach tst,$(ctests),$(tst) /cpu/$*;) 455 456prove : BACKENDS += $(TEST_BACKENDS) 457prove : $(matched) 458 $(info Testing backends: $(BACKENDS)) 459 $(PROVE) $(PROVE_OPTS) --exec 'tests/tap.sh' $(matched:$(OBJDIR)/%=%) 460# Run prove target in parallel 461prv : ;@$(MAKE) $(MFLAGS) V=$(V) prove 462 463prove-all : 464 +$(MAKE) prove realsearch=% 465 466junit-t% : BACKENDS += $(TEST_BACKENDS) 467junit-% : $(OBJDIR)/% 468 @printf " %10s %s\n" TEST $(<:$(OBJDIR)/%=%); $(PYTHON) tests/junit.py $(<:$(OBJDIR)/%=%) 469 470junit : $(matched:$(OBJDIR)/%=junit-%) 471 472all: $(alltests) 473 474examples : $(allexamples) 475ceedexamples : $(examples) 476nekexamples : $(nekexamples) 477mfemexamples : $(mfemexamples) 478petscexamples : $(petscexamples) 479 480# Benchmarks 481allbenchmarks = petsc-bps 482bench_targets = $(addprefix bench-,$(allbenchmarks)) 483.PHONY: $(bench_targets) benchmarks 484$(bench_targets): bench-%: $(OBJDIR)/% 485 cd benchmarks && ./benchmark.sh --ceed "$(BACKENDS)" -r $(*).sh 486benchmarks: $(bench_targets) 487 488$(ceed.pc) : pkgconfig-prefix = $(abspath .) 489$(OBJDIR)/ceed.pc : pkgconfig-prefix = $(prefix) 490.INTERMEDIATE : $(OBJDIR)/ceed.pc 491%/ceed.pc : ceed.pc.template | $$(@D)/.DIR 492 @sed "s:%prefix%:$(pkgconfig-prefix):" $< > $@ 493 494OCCA := $(OCCA_DIR)/bin/occa 495OKL_KERNELS := $(wildcard backends/occa/*.okl) 496 497okl-cache : 498 $(OCCA) cache ceed $(OKL_KERNELS) 499 500okl-clear: 501 $(OCCA) clear -y -l ceed 502 503install : $(libceed) $(OBJDIR)/ceed.pc 504 $(INSTALL) -d $(addprefix $(if $(DESTDIR),"$(DESTDIR)"),"$(includedir)"\ 505 "$(libdir)" "$(pkgconfigdir)" $(if $(OCCA_ON),"$(okldir)")) 506 $(INSTALL_DATA) include/ceed.h "$(DESTDIR)$(includedir)/" 507 $(INSTALL_DATA) include/ceedf.h "$(DESTDIR)$(includedir)/" 508 $(INSTALL_DATA) $(libceed) "$(DESTDIR)$(libdir)/" 509 $(INSTALL_DATA) $(OBJDIR)/ceed.pc "$(DESTDIR)$(pkgconfigdir)/" 510 $(if $(OCCA_ON),$(INSTALL_DATA) $(OKL_KERNELS) "$(DESTDIR)$(okldir)/") 511 512.PHONY : cln clean doxygen doc lib install all print test tst prove prv prove-all junit examples style tidy okl-cache okl-clear info info-backends 513 514cln clean : 515 $(RM) -r $(OBJDIR) $(LIBDIR) dist *egg* .pytest_cache *cffi* 516 $(MAKE) -C examples clean NEK5K_DIR="$(abspath $(NEK5K_DIR))" 517 $(MAKE) -C tests/python clean 518 $(RM) $(magma_tmp.c) $(magma_tmp.cu) backends/magma/*~ backends/magma/*.o 519 $(RM) benchmarks/*output.txt 520 521distclean : clean 522 $(RM) -r doc/html config.mk 523 524DOXYGEN ?= doxygen 525doxygen : 526 $(DOXYGEN) Doxyfile 527 528# This is a real file, but it doesn't depend on the state of a file, 529# so we make it phony to force it. We sort the shortlog by author 530# last name. 531.PHONY: AUTHORS 532AUTHORS: 533 git shortlog -s | awk '{$$1 = "placeholder"; print $$NF,$$0}' | sort | cut -d\ -f3- > $@ 534 535doc-html doc-latexpdf doc-epub : doc-% : doxygen AUTHORS 536 make -C doc/sphinx $* 537 538doc : doc-html 539 540style : 541 @astyle --options=.astylerc \ 542 $(filter-out include/ceedf.h tests/t320-basis-f.h, \ 543 $(wildcard include/*.h interface/*.[ch] tests/*.[ch] backends/*/*.[ch] \ 544 examples/*/*.[ch] examples/*/*.[ch]pp gallery/*/*.[ch])) 545 546CLANG_TIDY ?= clang-tidy 547%.c.tidy : %.c 548 $(CLANG_TIDY) $^ -- $(CPPFLAGS) --std=c99 549 550tidy : $(libceed.c:%=%.tidy) 551 552print : 553 @echo $(VAR)=$($(VAR)) 554 555print-% : 556 $(info [ variable name]: $*) 557 $(info [ origin]: $(origin $*)) 558 $(info [ value]: $(value $*)) 559 $(info [expanded value]: $($*)) 560 $(info ) 561 @true 562 563# "make configure" will autodetect any variables not passed on the 564# command line, caching the result in config.mk to be used on any 565# subsequent invocations of make. For example, 566# 567# make configure CC=/path/to/my/cc CUDA_DIR=/opt/cuda 568# make 569# make prove 570configure : 571 @: > config.mk 572 @echo "CC = $(CC)" | tee -a config.mk 573 @echo "CXX = $(CXX)" | tee -a config.mk 574 @echo "FC = $(FC)" | tee -a config.mk 575 @echo "NVCC = $(NVCC)" | tee -a config.mk 576 @echo "NVCC_CXX = $(NVCC_CXX)" | tee -a config.mk 577 @echo "CFLAGS = $(CFLAGS)" | tee -a config.mk 578 @echo "CPPFLAGS = $(CPPFLAGS)" | tee -a config.mk 579 @echo "FFLAGS = $(FFLAGS)" | tee -a config.mk 580 @echo "NVCCFLAGS = $(NVCCFLAGS)" | tee -a config.mk 581 @echo "LDFLAGS = $(LDFLAGS)" | tee -a config.mk 582 @echo "LDLIBS = $(LDLIBS)" | tee -a config.mk 583 @echo "MAGMA_DIR = $(MAGMA_DIR)" | tee -a config.mk 584 @echo "XSMM_DIR = $(XSMM_DIR)" | tee -a config.mk 585 @echo "CUDA_DIR = $(CUDA_DIR)" | tee -a config.mk 586 @echo "MFEM_DIR = $(MFEM_DIR)" | tee -a config.mk 587 @echo "PETSC_DIR = $(PETSC_DIR)" | tee -a config.mk 588 @echo "NEK5K_DIR = $(NEK5K_DIR)" | tee -a config.mk 589 @echo "Configuration cached in config.mk" 590 591.PHONY : configure 592 593-include $(libceed.c:%.c=$(OBJDIR)/%.d) $(tests.c:tests/%.c=$(OBJDIR)/%.d) 594