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 17ifeq (,$(filter-out undefined default,$(origin CC))) 18 CC = gcc 19endif 20ifeq (,$(filter-out undefined default,$(origin FC))) 21 FC = gfortran 22endif 23NVCC = $(CUDA_DIR)/bin/nvcc 24 25# ASAN must be left empty if you don't want to use it 26ASAN ?= 27 28LDFLAGS ?= 29UNDERSCORE ?= 1 30 31# MFEM_DIR env variable should point to sibling directory 32ifneq ($(wildcard ../mfem/libmfem.*),) 33 MFEM_DIR?=../mfem 34endif 35 36# OCCA_DIR env variable should point to OCCA master (github.com/libocca/occa) 37OCCA_DIR ?= ../occa 38 39# env variable MAGMA_DIR can be used too 40MAGMA_DIR ?= ../magma 41# If CUDA_DIR is not set, check for nvcc, or resort to /usr/local/cuda 42CUDA_DIR ?= $(or $(patsubst %/,%,$(dir $(patsubst %/,%,$(dir \ 43 $(shell which nvcc 2> /dev/null))))),/usr/local/cuda) 44 45# Warning: SANTIZ options still don't run with /gpu/occa 46# export LSAN_OPTIONS=suppressions=.asanignore 47AFLAGS = -fsanitize=address #-fsanitize=undefined -fno-omit-frame-pointer 48 49OPT = -O -g -march=native -ffp-contract=fast 50CFLAGS = -std=c99 $(OPT) -Wall -Wextra -Wno-unused-parameter -fPIC -MMD -MP 51NVCCFLAGS = -Xcompiler "$(OPT)" -Xcompiler -fPIC 52# If using the IBM XL Fortran (xlf) replace FFLAGS appropriately: 53ifneq ($(filter %xlf %xlf_r,$(FC)),) 54 FFLAGS = $(OPT) -ffree-form -qpreprocess -qextname -qpic -MMD 55else # gfortran/Intel-style options 56 FFLAGS = -cpp $(OPT) -ffree-form -Wall -Wextra -Wno-unused-parameter -Wno-unused-dummy-argument -fPIC -MMD -MP 57endif 58 59ifeq ($(UNDERSCORE), 1) 60 CFLAGS += -DUNDERSCORE 61endif 62 63ifeq ($(COVERAGE), 1) 64 CFLAGS += --coverage 65 LDFLAGS += --coverage 66endif 67 68CFLAGS += $(if $(ASAN),$(AFLAGS)) 69FFLAGS += $(if $(ASAN),$(AFLAGS)) 70LDFLAGS += $(if $(ASAN),$(AFLAGS)) 71CPPFLAGS = -I./include 72LDLIBS = -lm 73OBJDIR := build 74LIBDIR := lib 75 76# Installation variables 77prefix ?= /usr/local 78bindir = $(prefix)/bin 79libdir = $(prefix)/lib 80okldir = $(libdir)/okl 81includedir = $(prefix)/include 82pkgconfigdir = $(libdir)/pkgconfig 83INSTALL = install 84INSTALL_PROGRAM = $(INSTALL) 85INSTALL_DATA = $(INSTALL) -m644 86 87# Get number of processors of the machine 88NPROCS := $(shell getconf _NPROCESSORS_ONLN) 89# prepare make options to run in parallel 90MFLAGS := -j $(NPROCS) --warn-undefined-variables \ 91 --no-print-directory --no-keep-going 92 93PROVE ?= prove 94PROVE_OPTS ?= -j $(NPROCS) 95DARWIN := $(filter Darwin,$(shell uname -s)) 96SO_EXT := $(if $(DARWIN),dylib,so) 97 98ceed.pc := $(LIBDIR)/pkgconfig/ceed.pc 99libceed := $(LIBDIR)/libceed.$(SO_EXT) 100libceed.c := $(wildcard interface/ceed*.c) 101BACKENDS_BUILTIN := /cpu/self/ref /cpu/self/tmpl /cpu/self/blocked 102BACKENDS := $(BACKENDS_BUILTIN) 103 104# Tests 105tests.c := $(sort $(wildcard tests/t[0-9][0-9][0-9]-*.c)) 106tests.f := $(sort $(wildcard tests/t[0-9][0-9][0-9]-*.f)) 107tests := $(tests.c:tests/%.c=$(OBJDIR)/%) 108ctests := $(tests) 109tests += $(tests.f:tests/%.f=$(OBJDIR)/%) 110#examples 111examples.c := $(sort $(wildcard examples/ceed/*.c)) 112examples.f := $(sort $(wildcard examples/ceed/*.f)) 113examples := $(examples.c:examples/ceed/%.c=$(OBJDIR)/%) 114examples += $(examples.f:examples/ceed/%.f=$(OBJDIR)/%) 115#mfemexamples 116mfemexamples.cpp := $(sort $(wildcard examples/mfem/*.cpp)) 117mfemexamples := $(mfemexamples.cpp:examples/mfem/%.cpp=$(OBJDIR)/mfem-%) 118petscexamples.c := $(sort $(wildcard examples/petsc/*.c)) 119petscexamples := $(petscexamples.c:examples/petsc/%.c=$(OBJDIR)/petsc-%) 120 121# backends/[ref, template, blocked, avx, occa, magma] 122ref.c := $(sort $(wildcard backends/ref/*.c)) 123template.c := $(sort $(wildcard backends/template/*.c)) 124cuda.c := $(sort $(wildcard backends/cuda/*.c)) 125cuda.cu := $(sort $(wildcard backends/cuda/*.cu)) 126blocked.c := $(sort $(wildcard backends/blocked/*.c)) 127avx.c := $(sort $(wildcard backends/avx/*.c)) 128occa.c := $(sort $(wildcard backends/occa/*.c)) 129magma_preprocessor := python backends/magma/gccm.py 130magma_pre_src := $(filter-out %_tmp.c, $(wildcard backends/magma/ceed-*.c)) 131magma_dsrc := $(wildcard backends/magma/magma_d*.c) 132magma_tmp.c := $(magma_pre_src:%.c=%_tmp.c) 133magma_tmp.cu := $(magma_pre_src:%.c=%_cuda.cu) 134magma_allsrc.c := $(magma_dsrc) $(magma_tmp.c) 135magma_allsrc.cu:= $(magma_tmp.cu) 136 137# Output using the 216-color rules mode 138rule_file = $(notdir $(1)) 139rule_path = $(patsubst %/,%,$(dir $(1))) 140last_path = $(notdir $(patsubst %/,%,$(dir $(1)))) 141ansicolor = $(shell echo $(call last_path,$(1)) | cksum | cut -b1-2 | xargs -IS expr 2 \* S + 17) 142emacs_out = @printf " %10s %s/%s\n" $(1) $(call rule_path,$(2)) $(call rule_file,$(2)) 143color_out = @if [ -t 1 ]; then \ 144 printf " %10s \033[38;5;%d;1m%s\033[m/%s\n" \ 145 $(1) $(call ansicolor,$(2)) \ 146 $(call rule_path,$(2)) $(call rule_file,$(2)); else \ 147 printf " %10s %s\n" $(1) $(2); fi 148# if TERM=dumb, use it, otherwise switch to the term one 149output = $(if $(TERM:dumb=),$(call color_out,$1,$2),$(call emacs_out,$1,$2)) 150 151# if V is set to non-nil, turn the verbose mode 152quiet = $(if $(V),$($(1)),$(call output,$1,$@);$($(1))) 153 154# Cancel built-in and old-fashioned implicit rules which we don't use 155.SUFFIXES: 156 157.SECONDEXPANSION: # to expand $$(@D)/.DIR 158 159.SECONDARY: $(magma_tmp.c) $(magma_tmp.cu) 160 161%/.DIR : 162 @mkdir -p $(@D) 163 @touch $@ 164 165.PRECIOUS: %/.DIR 166 167lib: $(libceed) $(ceed.pc) 168# run 'lib' target in parallel 169all:;@$(MAKE) $(MFLAGS) V=$(V) lib 170backend_status = $(if $(filter $1,$(BACKENDS)), [backends: $1], [not found]) 171info: 172 $(info ------------------------------------) 173 $(info CC = $(CC)) 174 $(info FC = $(FC)) 175 $(info CPPFLAGS = $(CPPFLAGS)) 176 $(info CFLAGS = $(value CFLAGS)) 177 $(info FFLAGS = $(value FFLAGS)) 178 $(info NVCCFLAGS = $(value NVCCFLAGS)) 179 $(info LDFLAGS = $(value LDFLAGS)) 180 $(info LDLIBS = $(LDLIBS)) 181 $(info OPT = $(OPT)) 182 $(info AFLAGS = $(AFLAGS)) 183 $(info ASAN = $(or $(ASAN),(empty))) 184 $(info V = $(or $(V),(empty)) [verbose=$(if $(V),on,off)]) 185 $(info ------------------------------------) 186 $(info AVX_STATUS = $(AVX_STATUS)$(call backend_status,/cpu/self/avx)) 187 $(info OCCA_DIR = $(OCCA_DIR)$(call backend_status,/cpu/occa /gpu/occa /omp/occa)) 188 $(info MAGMA_DIR = $(MAGMA_DIR)$(call backend_status,/gpu/magma)) 189 $(info CUDA_DIR = $(CUDA_DIR)$(call backend_status,/gpu/magma)) 190 $(info ------------------------------------) 191 $(info MFEM_DIR = $(MFEM_DIR)) 192 $(info PETSC_DIR = $(PETSC_DIR)) 193 $(info ------------------------------------) 194 $(info prefix = $(prefix)) 195 $(info includedir = $(value includedir)) 196 $(info libdir = $(value libdir)) 197 $(info okldir = $(value okldir)) 198 $(info pkgconfigdir = $(value pkgconfigdir)) 199 $(info ------------------------------------) 200 @true 201info-backends: 202 $(info make: 'lib' with optional backends: $(filter-out $(BACKENDS_BUILTIN),$(BACKENDS))) 203.PHONY: lib all info info-backends 204 205$(libceed) : LDFLAGS += $(if $(DARWIN), -install_name @rpath/$(notdir $(libceed))) 206 207# Standard Backends 208libceed.c += $(ref.c) 209libceed.c += $(template.c) 210libceed.c += $(blocked.c) 211 212# AVX Backed 213AVX_STATUS = Disabled 214AVX := $(shell $(CC) $(CFLAGS) -v -E - < /dev/null 2>&1 | grep -c avx) 215ifeq ($(AVX),1) 216 AVX_STATUS = Enabled 217 libceed.c += $(avx.c) 218 BACKENDS += /cpu/self/avx 219endif 220 221# OCCA Backends 222ifneq ($(wildcard $(OCCA_DIR)/lib/libocca.*),) 223 $(libceed) : LDFLAGS += -L$(OCCA_DIR)/lib -Wl,-rpath,$(abspath $(OCCA_DIR)/lib) 224 $(libceed) : LDLIBS += -locca 225 libceed.c += $(occa.c) 226 $(occa.c:%.c=$(OBJDIR)/%.o) : CFLAGS += -I$(OCCA_DIR)/include 227 BACKENDS += /cpu/occa /gpu/occa /omp/occa 228endif 229 230# Cuda Backend 231CUDA_LIB_DIR := $(wildcard $(foreach d,lib lib64,$(CUDA_DIR)/$d/libcudart.${SO_EXT})) 232CUDA_LIB_DIR := $(patsubst %/,%,$(dir $(firstword $(CUDA_LIB_DIR)))) 233ifneq ($(CUDA_LIB_DIR),) 234 $(libceed) : CFLAGS += -I$(CUDA_DIR)/include 235 $(libceed) : LDFLAGS += -L$(CUDA_LIB_DIR) -Wl,-rpath,$(abspath $(CUDA_LIB_DIR)) 236 $(libceed) : LDLIBS += -lcudart -lnvrtc -lcuda 237 libceed.c += $(cuda.c) 238 libceed.cu += $(cuda.cu) 239 BACKENDS += /gpu/cuda 240endif 241 242# MAGMA Backend 243ifneq ($(wildcard $(MAGMA_DIR)/lib/libmagma.*),) 244 ifneq ($(CUDA_LIB_DIR),) 245 cuda_link = -Wl,-rpath,$(CUDA_LIB_DIR) -L$(CUDA_LIB_DIR) -lcublas -lcusparse -lcudart 246 omp_link = -fopenmp 247 magma_link_static = -L$(MAGMA_DIR)/lib -lmagma $(cuda_link) $(omp_link) 248 magma_link_shared = -L$(MAGMA_DIR)/lib -Wl,-rpath,$(abspath $(MAGMA_DIR)/lib) -lmagma 249 magma_link := $(if $(wildcard $(MAGMA_DIR)/lib/libmagma.${SO_EXT}),$(magma_link_shared),$(magma_link_static)) 250 $(libceed) : LDLIBS += $(magma_link) 251 $(tests) $(examples) : LDLIBS += $(magma_link) 252 libceed.c += $(magma_allsrc.c) 253 libceed.cu += $(magma_allsrc.cu) 254 $(magma_allsrc.c:%.c=$(OBJDIR)/%.o) : CFLAGS += -DADD_ -I$(MAGMA_DIR)/include -I$(CUDA_DIR)/include 255 $(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 256 BACKENDS += /gpu/magma 257 endif 258endif 259 260export BACKENDS 261 262# generate magma_tmp.c and magma_cuda.cu from magma.c 263%_tmp.c %_cuda.cu : %.c 264 $(magma_preprocessor) $< 265 266libceed.o = $(libceed.c:%.c=$(OBJDIR)/%.o) $(libceed.cu:%.cu=$(OBJDIR)/%.o) 267$(libceed.o): | info-backends 268$(libceed) : $(libceed.o) | $$(@D)/.DIR 269 $(call quiet,CC) $(LDFLAGS) -shared -o $@ $^ $(LDLIBS) 270 271$(OBJDIR)/%.o : $(CURDIR)/%.c | $$(@D)/.DIR 272 $(call quiet,CC) $(CPPFLAGS) $(CFLAGS) -c -o $@ $(abspath $<) 273 274$(OBJDIR)/%.o : $(CURDIR)/%.cu | $$(@D)/.DIR 275 $(call quiet,NVCC) $(CPPFLAGS) $(NVCCFLAGS) -c -o $@ $(abspath $<) 276 277$(OBJDIR)/% : tests/%.c | $$(@D)/.DIR 278 $(call quiet,LINK.c) -o $@ $(abspath $<) -lceed $(LDLIBS) 279 280$(OBJDIR)/% : tests/%.f | $$(@D)/.DIR 281 $(call quiet,LINK.F) -o $@ $(abspath $<) -lceed $(LDLIBS) 282 283$(OBJDIR)/% : examples/ceed/%.c | $$(@D)/.DIR 284 $(call quiet,LINK.c) -o $@ $(abspath $<) -lceed $(LDLIBS) 285 286$(OBJDIR)/% : examples/ceed/%.f | $$(@D)/.DIR 287 $(call quiet,LINK.F) -o $@ $(abspath $<) -lceed $(LDLIBS) 288 289$(OBJDIR)/mfem-% : examples/mfem/%.cpp $(libceed) | $$(@D)/.DIR 290 $(MAKE) -C examples/mfem CEED_DIR=`pwd` $* 291 mv examples/mfem/$* $@ 292 293$(OBJDIR)/petsc-% : examples/petsc/%.c $(libceed) $(ceed.pc) | $$(@D)/.DIR 294 $(MAKE) -C examples/petsc CEED_DIR=`pwd` $* 295 mv examples/petsc/$* $@ 296 297$(tests) $(examples) : $(libceed) 298$(tests) $(examples) : LDFLAGS += -Wl,-rpath,$(abspath $(LIBDIR)) -L$(LIBDIR) 299 300run-% : $(OBJDIR)/% 301 @tests/tap.sh $(<:build/%=%) 302# Test core libCEED 303test : $(tests:$(OBJDIR)/%=run-%) $(examples:$(OBJDIR)/%=run-%) 304 305# run test target in parallel 306tst : ;@$(MAKE) $(MFLAGS) V=$(V) test 307# CPU C tests only for backend % 308ctc-% : $(ctests);@$(foreach tst,$(ctests),$(tst) /cpu/$*;) 309 310prove : $(tests) $(examples) 311 $(info Testing backends: $(BACKENDS)) 312 $(PROVE) $(PROVE_OPTS) --exec 'tests/tap.sh' $(tests:$(OBJDIR)/%=%) $(examples:$(OBJDIR)/%=%) 313# run prove target in parallel 314prv : ;@$(MAKE) $(MFLAGS) V=$(V) prove 315 316alltests := $(tests) $(examples) $(if $(MFEM_DIR),$(mfemexamples)) $(if $(PETSC_DIR),$(petscexamples)) 317prove-all : $(alltests) 318 $(info Testing backends: $(BACKENDS)) 319 $(PROVE) $(PROVE_OPTS) --exec 'tests/tap.sh' $(alltests:$(OBJDIR)/%=%) 320 321examples : $(examples) 322 323$(ceed.pc) : pkgconfig-prefix = $(abspath .) 324$(OBJDIR)/ceed.pc : pkgconfig-prefix = $(prefix) 325.INTERMEDIATE : $(OBJDIR)/ceed.pc 326%/ceed.pc : ceed.pc.template | $$(@D)/.DIR 327 @sed "s:%prefix%:$(pkgconfig-prefix):" $< > $@ 328 329OCCA := $(OCCA_DIR)/bin/occa 330OKL_KERNELS := $(wildcard backends/occa/*.okl) 331 332okl-cache : 333 $(OCCA) cache ceed $(OKL_KERNELS) 334 335okl-clear: 336 $(OCCA) clear -y -l ceed 337 338install : $(libceed) $(OBJDIR)/ceed.pc 339 $(INSTALL) -d $(addprefix $(if $(DESTDIR),"$(DESTDIR)"),"$(includedir)"\ 340 "$(libdir)" "$(pkgconfigdir)" $(if $(OCCA_ON),"$(okldir)")) 341 $(INSTALL_DATA) include/ceed.h "$(DESTDIR)$(includedir)/" 342 $(INSTALL_DATA) include/ceedf.h "$(DESTDIR)$(includedir)/" 343 $(INSTALL_DATA) $(libceed) "$(DESTDIR)$(libdir)/" 344 $(INSTALL_DATA) $(OBJDIR)/ceed.pc "$(DESTDIR)$(pkgconfigdir)/" 345 $(if $(OCCA_ON),$(INSTALL_DATA) $(OKL_KERNELS) "$(DESTDIR)$(okldir)/") 346 347.PHONY : cln clean print test tst prove prv examples style install doc okl-cache okl-clear 348 349cln clean : 350 $(RM) -r $(OBJDIR) $(LIBDIR) 351 $(MAKE) -C examples/ceed clean 352 $(MAKE) -C examples/mfem clean 353 $(MAKE) -C examples/petsc clean 354 (cd examples/nek5000 && bash make-nek-examples.sh clean) 355 $(RM) $(magma_tmp.c) $(magma_tmp.cu) backends/magma/*~ backends/magma/*.o 356 357distclean : clean 358 $(RM) -r doc/html 359 360doc : 361 doxygen Doxyfile 362 363style : 364 astyle --style=google --indent=spaces=2 --max-code-length=80 \ 365 --keep-one-line-statements --keep-one-line-blocks --lineend=linux \ 366 --suffix=none --preserve-date --formatted \ 367 --exclude=include/ceedf.h --exclude=tests/t310-basis-f.h \ 368 include/*.h interface/*.[ch] tests/*.[ch] backends/*/*.[ch] \ 369 examples/*/*.[ch] examples/*/*.[ch]pp -i 370 371print : 372 @echo $(VAR)=$($(VAR)) 373 374print-% : 375 $(info [ variable name]: $*) 376 $(info [ origin]: $(origin $*)) 377 $(info [ value]: $(value $*)) 378 $(info [expanded value]: $($*)) 379 $(info ) 380 @true 381 382-include $(libceed.c:%.c=build/%.d) $(tests.c:tests/%.c=build/%.d) 383