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 ?= 27NDEBUG ?= 1 28 29LDFLAGS ?= 30UNDERSCORE ?= 1 31 32# OCCA_DIR env variable should point to OCCA master (github.com/libocca/occa) 33OCCA_DIR ?= ../occa 34 35# env variable MAGMA_DIR can be used too 36MAGMA_DIR ?= ../magma 37# If CUDA_DIR is not set, check for nvcc, or resort to /usr/local/cuda 38CUDA_DIR ?= $(or $(patsubst %/,%,$(dir $(patsubst %/,%,$(dir \ 39 $(shell which nvcc 2> /dev/null))))),/usr/local/cuda) 40 41# Warning: SANTIZ options still don't run with /gpu/occa 42# export LSAN_OPTIONS=suppressions=.asanignore 43AFLAGS = -fsanitize=address #-fsanitize=undefined -fno-omit-frame-pointer 44 45CFLAGS = -std=c99 -Wall -Wextra -Wno-unused-parameter -fPIC -MMD -MP 46FFLAGS = -cpp -Wall -Wextra -Wno-unused-parameter -Wno-unused-dummy-argument -fPIC -MMD -MP 47# If using the IBM XL Fortran (xlf) replace FFLAGS appropriately: 48ifneq ($(filter %xlf %xlf_r,$(FC)),) 49 FFLAGS = -qpreprocess -qextname -qpic -MMD 50endif 51 52CFLAGS += $(if $(NDEBUG),-O2 -DNDEBUG=1,-g) 53 54ifeq ($(UNDERSCORE), 1) 55 CFLAGS += -DUNDERSCORE 56endif 57 58FFLAGS += $(if $(NDEBUG),-O2 -DNDEBUG=1,-g) 59 60CFLAGS += $(if $(ASAN),$(AFLAGS)) 61FFLAGS += $(if $(ASAN),$(AFLAGS)) 62LDFLAGS += $(if $(ASAN),$(AFLAGS)) 63CPPFLAGS = -I./include 64LDLIBS = -lm 65OBJDIR := build 66LIBDIR := lib 67 68# Installation variables 69prefix ?= /usr/local 70bindir = $(prefix)/bin 71libdir = $(prefix)/lib 72okldir = $(prefix)/lib/okl 73includedir = $(prefix)/include 74pkgconfigdir = $(libdir)/pkgconfig 75INSTALL = install 76INSTALL_PROGRAM = $(INSTALL) 77INSTALL_DATA = $(INSTALL) -m644 78 79# Get number of processors of the machine 80NPROCS := $(shell getconf _NPROCESSORS_ONLN) 81# prepare make options to run in parallel 82MFLAGS := -j $(NPROCS) --warn-undefined-variables \ 83 --no-print-directory --no-keep-going 84 85PROVE ?= prove 86PROVE_OPTS ?= -j $(NPROCS) 87DARWIN := $(filter Darwin,$(shell uname -s)) 88SO_EXT := $(if $(DARWIN),dylib,so) 89 90ceed.pc := $(LIBDIR)/pkgconfig/ceed.pc 91libceed := $(LIBDIR)/libceed.$(SO_EXT) 92libceed.c := $(wildcard ceed*.c) 93BACKENDS := /cpu/self/ref /cpu/self/tmpl /cpu/self/opt 94 95# Tests 96tests.c := $(sort $(wildcard tests/t[0-9][0-9]-*.c)) 97tests.f := $(sort $(wildcard tests/t[0-9][0-9]-*.f)) 98tests := $(tests.c:tests/%.c=$(OBJDIR)/%) 99ctests := $(tests) 100tests += $(tests.f:tests/%.f=$(OBJDIR)/%) 101#examples 102examples.c := $(sort $(wildcard examples/ceed/*.c)) 103examples.f := $(sort $(wildcard examples/ceed/*.f)) 104examples := $(examples.c:examples/ceed/%.c=$(OBJDIR)/%) 105examples += $(examples.f:examples/ceed/%.f=$(OBJDIR)/%) 106#mfemexamples 107mfemexamples.cpp := $(sort $(wildcard examples/mfem/*.cpp)) 108mfemexamples := $(mfemexamples.cpp:examples/mfem/%.cpp=$(OBJDIR)/mfem-%) 109# backends/[ref & occa & magma] 110ref.c := $(sort $(wildcard backends/ref/*.c)) 111template.c := $(sort $(wildcard backends/template/*.c)) 112optimized.c:= $(sort $(wildcard backends/optimized/*.c)) 113occa.c := $(sort $(wildcard backends/occa/*.c)) 114magma_preprocessor := python backends/magma/gccm.py 115magma_pre_src := $(filter-out %_tmp.c, $(wildcard backends/magma/ceed-*.c)) 116magma_dsrc := $(wildcard backends/magma/magma_d*.c) 117magma_tmp.c := $(magma_pre_src:%.c=%_tmp.c) 118magma_tmp.cu := $(magma_pre_src:%.c=%_cuda.cu) 119magma_allsrc.c := $(magma_dsrc) $(magma_tmp.c) 120magma_allsrc.cu:= $(magma_tmp.cu) 121 122# Output using the 216-color rules mode 123rule_file = $(notdir $(1)) 124rule_path = $(patsubst %/,%,$(dir $(1))) 125last_path = $(notdir $(patsubst %/,%,$(dir $(1)))) 126ansicolor = $(shell echo $(call last_path,$(1)) | cksum | cut -b1-2 | xargs -IS expr 2 \* S + 17) 127emacs_out = @printf " %10s %s/%s\n" $(1) $(call rule_path,$(2)) $(call rule_file,$(2)) 128color_out = @if [ -t 1 ]; then \ 129 printf " %10s \033[38;5;%d;1m%s\033[m/%s\n" \ 130 $(1) $(call ansicolor,$(2)) \ 131 $(call rule_path,$(2)) $(call rule_file,$(2)); else \ 132 printf " %10s %s\n" $(1) $(2); fi 133# if TERM=dumb, use it, otherwise switch to the term one 134output = $(if $(TERM:dumb=),$(call color_out,$1,$2),$(call emacs_out,$1,$2)) 135 136# if V is set to non-nil, turn the verbose mode 137quiet = $(if $(V),$($(1)),$(call output,$1,$@);$($(1))) 138 139.SUFFIXES: 140.SUFFIXES: .c .o .d 141.SECONDEXPANSION: # to expand $$(@D)/.DIR 142 143.SECONDARY: $(magma_tmp.c) $(magma_tmp.cu) 144 145%/.DIR : 146 @mkdir -p $(@D) 147 @touch $@ 148 149.PRECIOUS: %/.DIR 150 151this: $(libceed) $(ceed.pc) 152# run 'this' target in parallel 153all:;@$(MAKE) $(MFLAGS) V=$(V) this 154 155$(libceed) : LDFLAGS += $(if $(DARWIN), -install_name @rpath/$(notdir $(libceed))) 156 157libceed.c += $(ref.c) 158libceed.c += $(template.c) 159libceed.c += $(optimized.c) 160ifneq ($(wildcard $(OCCA_DIR)/lib/libocca.*),) 161 $(libceed) : LDFLAGS += -L$(OCCA_DIR)/lib -Wl,-rpath,$(abspath $(OCCA_DIR)/lib) 162 $(libceed) : LDLIBS += -locca 163 libceed.c += $(occa.c) 164 $(occa.c:%.c=$(OBJDIR)/%.o) : CFLAGS += -I$(OCCA_DIR)/include 165 BACKENDS += /cpu/occa /gpu/occa /omp/occa /ocl/occa 166endif 167ifneq ($(wildcard $(MAGMA_DIR)/lib/libmagma.*),) 168 CUDA_LIB_DIR := $(wildcard $(foreach d,lib lib64,$(CUDA_DIR)/$d/libcudart.${SO_EXT})) 169 CUDA_LIB_DIR := $(patsubst %/,%,$(dir $(firstword $(CUDA_LIB_DIR)))) 170 ifneq ($(CUDA_LIB_DIR),) 171 cuda_link = -Wl,-rpath,$(CUDA_LIB_DIR) -L$(CUDA_LIB_DIR) -lcublas -lcusparse -lcudart 172 omp_link = -fopenmp 173 magma_link_static = -L$(MAGMA_DIR)/lib -lmagma $(cuda_link) $(omp_link) 174 magma_link_shared = -L$(MAGMA_DIR)/lib -Wl,-rpath,$(abspath $(MAGMA_DIR)/lib) -lmagma 175 magma_link := $(if $(wildcard $(MAGMA_DIR)/lib/libmagma.${SO_EXT}),$(magma_link_shared),$(magma_link_static)) 176 magma_allsrc.o = $(magma_allsrc.c:%.c=$(OBJDIR)/%.o) $(magma_allsrc.cu:%.cu=$(OBJDIR)/%.o) 177 $(libceed) : LDLIBS += $(magma_link) 178 $(tests) $(examples) : LDLIBS += $(magma_link) 179 $(libceed) : $(magma_allsrc.o) 180 libceed.c += $(magma_allsrc.c) 181 libceed.cu += $(magma_allsrc.cu) 182 $(magma_allsrc.c:%.c=$(OBJDIR)/%.o) : CFLAGS += -DADD_ -I$(MAGMA_DIR)/include -I$(CUDA_DIR)/include 183 $(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 184 BACKENDS += /gpu/magma 185 endif 186endif 187 188export BACKENDS 189 190# generate magma_tmp.c and magma_cuda.cu from magma.c 191$(magma_tmp.c) $(magma_tmp.cu): $(magma_pre_src) | $$(@D)/.DIR 192 $(magma_preprocessor) $< 193 194$(libceed) : $(libceed.c:%.c=$(OBJDIR)/%.o) $(libceed.cu:%.cu=$(OBJDIR)/%.o) | $$(@D)/.DIR 195 $(call quiet,CC) $(LDFLAGS) -shared -o $@ $^ $(LDLIBS) 196 197$(OBJDIR)/%.o : %.c | $$(@D)/.DIR 198 $(call quiet,CC) $(CPPFLAGS) $(CFLAGS) -c -o $@ $(abspath $<) 199 200$(OBJDIR)/%.o : %.cu | $$(@D)/.DIR 201 $(call quiet,NVCC) $(CPPFLAGS) $(NVCCFLAGS) -c -o $@ $(abspath $<) 202 203$(OBJDIR)/% : tests/%.c | $$(@D)/.DIR 204 $(call quiet,CC) $(CPPFLAGS) $(CFLAGS) $(LDFLAGS) -o $@ $(abspath $<) -lceed $(LDLIBS) 205 206$(OBJDIR)/% : tests/%.f | $$(@D)/.DIR 207 $(call quiet,FC) $(CPPFLAGS) $(FFLAGS) $(LDFLAGS) -o $@ $(abspath $<) -lceed $(LDLIBS) 208 209$(OBJDIR)/% : examples/ceed/%.c | $$(@D)/.DIR 210 $(call quiet,CC) $(CPPFLAGS) $(CFLAGS) $(LDFLAGS) -o $@ $(abspath $<) -lceed $(LDLIBS) 211 212$(OBJDIR)/% : examples/ceed/%.f | $$(@D)/.DIR 213 $(call quiet,FC) $(CPPFLAGS) $(FFLAGS) $(LDFLAGS) -o $@ $(abspath $<) -lceed $(LDLIBS) 214 215$(OBJDIR)/mfem-% : examples/mfem/%.cpp $(libceed) | $$(@D)/.DIR 216 $(MAKE) -C examples/mfem CEED_DIR=`pwd` $* 217 mv examples/mfem/$* $@ 218 219$(tests) $(examples) : $(libceed) 220$(tests) $(examples) : LDFLAGS += -Wl,-rpath,$(abspath $(LIBDIR)) -L$(LIBDIR) 221 222run-% : $(OBJDIR)/% 223 @tests/tap.sh $(<:build/%=%) 224# Test core libCEED 225test : $(tests:$(OBJDIR)/%=run-%) $(examples:$(OBJDIR)/%=run-%) 226 227# run test target in parallel 228tst : ;@$(MAKE) $(MFLAGS) V=$(V) test 229# CPU C tests only for backend % 230ctc-% : $(ctests);@$(foreach tst,$(ctests),$(tst) /cpu/$*;) 231 232prove : $(tests) $(examples) 233 $(info Testing backends: $(BACKENDS)) 234 $(PROVE) $(PROVE_OPTS) --exec 'tests/tap.sh' $(tests:$(OBJDIR)/%=%) $(examples:$(OBJDIR)/%=%) 235# run prove target in parallel 236prv : ;@$(MAKE) $(MFLAGS) V=$(V) prove 237 238prove-allexamples : $(tests) $(examples) $(mfemexamples) 239 $(info Testing backends: $(BACKENDS)) 240 $(PROVE) $(PROVE_OPTS) --exec 'tests/tap.sh' $(tests:$(OBJDIR)/%=%) $(examples:$(OBJDIR)/%=%) $(mfemexamples:$(OBJDIR)/%=%) 241 242examples : $(examples) 243 244$(ceed.pc) : pkgconfig-prefix = $(abspath .) 245$(OBJDIR)/ceed.pc : pkgconfig-prefix = $(prefix) 246.INTERMEDIATE : $(OBJDIR)/ceed.pc 247%/ceed.pc : ceed.pc.template | $$(@D)/.DIR 248 @sed "s:%prefix%:$(pkgconfig-prefix):" $< > $@ 249 250# The occa executable is not linked with RPATH by default, so we need it to find its libocca.so 251OCCA := $(if $(DARWIN),DYLD_LIBRARY_PATH,LD_LIBRARY_PATH)=$(OCCA_DIR)/lib $(OCCA_DIR)/bin/occa 252OKL_KERNELS := $(wildcard backends/occa/*.okl) 253 254okl-cache : 255 $(OCCA) cache ceed $(OKL_KERNELS) 256 257okl-clear: 258 $(OCCA) clear -y -l ceed 259 260install : $(libceed) $(OBJDIR)/ceed.pc 261 $(INSTALL) -d "$(DESTDIR)$(includedir)" "$(DESTDIR)$(libdir)" "$(DESTDIR)$(okldir)" "$(DESTDIR)$(pkgconfigdir)" 262 $(INSTALL_DATA) include/ceed.h "$(DESTDIR)$(includedir)/" 263 $(INSTALL_DATA) include/ceedf.h "$(DESTDIR)$(includedir)/" 264 $(INSTALL_DATA) $(libceed) "$(DESTDIR)$(libdir)/" 265 $(INSTALL_DATA) $(OBJDIR)/ceed.pc "$(DESTDIR)$(pkgconfigdir)/" 266 $(INSTALL_DATA) $(OKL_KERNELS) "$(DESTDIR)$(okldir)/" 267 268.PHONY : all cln clean print test tst prove prv examples style install doc okl-cache okl-clear 269 270cln clean : 271 $(RM) *.o *.d $(libceed) 272 $(RM) -r *.dSYM $(OBJDIR) $(LIBDIR)/pkgconfig 273 $(MAKE) -C examples/ceed clean 274 $(MAKE) -C examples/mfem clean 275 $(MAKE) -C examples/petsc clean 276 (cd examples/nek5000 && bash make-nek-examples.sh clean) 277 $(RM) $(magma_tmp.c) $(magma_tmp.cu) backends/magma/*~ backends/magma/*.o 278 279distclean : clean 280 rm -rf doc/html 281 282doc : 283 doxygen Doxyfile 284 285style : 286 astyle --style=google --indent=spaces=2 --max-code-length=80 \ 287 --keep-one-line-statements --keep-one-line-blocks --lineend=linux \ 288 --suffix=none --preserve-date --formatted \ 289 *.[ch] tests/*.[ch] backends/*/*.[ch] examples/*/*.[ch] examples/*/*.[ch]pp 290 291print : 292 @echo $(VAR)=$($(VAR)) 293 294print-% : 295 $(info [ variable name]: $*) 296 $(info [ origin]: $(origin $*)) 297 $(info [ value]: $(value $*)) 298 $(info [expanded value]: $($*)) 299 $(info ) 300 @true 301 302-include $(libceed.c:%.c=build/%.d) $(tests.c:tests/%.c=build/%.d) 303