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 66MFEM_EX_DIR := examples/mfem 67LIBDIR := lib 68 69# Installation variables 70prefix ?= /usr/local 71bindir = $(prefix)/bin 72libdir = $(prefix)/lib 73okldir = $(prefix)/lib/okl 74includedir = $(prefix)/include 75pkgconfigdir = $(libdir)/pkgconfig 76INSTALL = install 77INSTALL_PROGRAM = $(INSTALL) 78INSTALL_DATA = $(INSTALL) -m644 79 80# Get number of processors of the machine 81NPROCS := $(shell getconf _NPROCESSORS_ONLN) 82# prepare make options to run in parallel 83MFLAGS := -j $(NPROCS) --warn-undefined-variables \ 84 --no-print-directory --no-keep-going 85 86PROVE ?= prove 87PROVE_OPTS ?= -j $(NPROCS) 88DARWIN := $(filter Darwin,$(shell uname -s)) 89SO_EXT := $(if $(DARWIN),dylib,so) 90 91ceed.pc := $(LIBDIR)/pkgconfig/ceed.pc 92libceed := $(LIBDIR)/libceed.$(SO_EXT) 93libceed.c := $(wildcard ceed*.c) 94BACKENDS := /cpu/self/ref /cpu/self/tmpl /cpu/self/opt 95 96# Tests 97tests.c := $(sort $(wildcard tests/t[0-9][0-9]-*.c)) 98tests.f := $(sort $(wildcard tests/t[0-9][0-9]-*.f)) 99tests := $(tests.c:tests/%.c=$(OBJDIR)/%) 100ctests := $(tests) 101tests += $(tests.f:tests/%.f=$(OBJDIR)/%) 102#examples 103examples.c := $(sort $(wildcard examples/ceed/*.c)) 104examples.f := $(sort $(wildcard examples/ceed/*.f)) 105examples := $(examples.c:examples/ceed/%.c=$(OBJDIR)/%) 106examples += $(examples.f:examples/ceed/%.f=$(OBJDIR)/%) 107#mfemexamples 108mfemexamples.cpp := $(sort $(wildcard examples/mfem/*.cpp)) 109mfemexamples := $(mfemexamples.cpp:examples/mfem/%.cpp=$(MFEM_EX_DIR)/%) 110# backends/[ref & occa & magma] 111ref.c := $(sort $(wildcard backends/ref/*.c)) 112template.c := $(sort $(wildcard backends/template/*.c)) 113optimized.c:= $(sort $(wildcard backends/optimized/*.c)) 114occa.c := $(sort $(wildcard backends/occa/*.c)) 115magma_preprocessor := python backends/magma/gccm.py 116magma_pre_src := $(filter-out %_tmp.c, $(wildcard backends/magma/ceed-*.c)) 117magma_dsrc := $(wildcard backends/magma/magma_d*.c) 118magma_tmp.c := $(magma_pre_src:%.c=%_tmp.c) 119magma_tmp.cu := $(magma_pre_src:%.c=%_cuda.cu) 120magma_allsrc.c := $(magma_dsrc) $(magma_tmp.c) 121magma_allsrc.cu:= $(magma_tmp.cu) 122 123# Output using the 216-color rules mode 124rule_file = $(notdir $(1)) 125rule_path = $(patsubst %/,%,$(dir $(1))) 126last_path = $(notdir $(patsubst %/,%,$(dir $(1)))) 127ansicolor = $(shell echo $(call last_path,$(1)) | cksum | cut -b1-2 | xargs -IS expr 2 \* S + 17) 128emacs_out = @printf " %10s %s/%s\n" $(1) $(call rule_path,$(2)) $(call rule_file,$(2)) 129color_out = @if [ -t 1 ]; then \ 130 printf " %10s \033[38;5;%d;1m%s\033[m/%s\n" \ 131 $(1) $(call ansicolor,$(2)) \ 132 $(call rule_path,$(2)) $(call rule_file,$(2)); else \ 133 printf " %10s %s\n" $(1) $(2); fi 134# if TERM=dumb, use it, otherwise switch to the term one 135output = $(if $(TERM:dumb=),$(call color_out,$1,$2),$(call emacs_out,$1,$2)) 136 137# if V is set to non-nil, turn the verbose mode 138quiet = $(if $(V),$($(1)),$(call output,$1,$@);$($(1))) 139 140.SUFFIXES: 141.SUFFIXES: .c .o .d 142.SECONDEXPANSION: # to expand $$(@D)/.DIR 143 144.SECONDARY: $(magma_tmp.c) $(magma_tmp.cu) 145 146%/.DIR : 147 @mkdir -p $(@D) 148 @touch $@ 149 150.PRECIOUS: %/.DIR 151 152this: $(libceed) $(ceed.pc) 153# run 'this' target in parallel 154all:;@$(MAKE) $(MFLAGS) V=$(V) this 155 156$(libceed) : LDFLAGS += $(if $(DARWIN), -install_name @rpath/$(notdir $(libceed))) 157 158libceed.c += $(ref.c) 159libceed.c += $(template.c) 160libceed.c += $(optimized.c) 161ifneq ($(wildcard $(OCCA_DIR)/lib/libocca.*),) 162 $(libceed) : LDFLAGS += -L$(OCCA_DIR)/lib -Wl,-rpath,$(abspath $(OCCA_DIR)/lib) 163 $(libceed) : LDLIBS += -locca 164 libceed.c += $(occa.c) 165 $(occa.c:%.c=$(OBJDIR)/%.o) : CFLAGS += -I$(OCCA_DIR)/include 166 BACKENDS += /cpu/occa /gpu/occa /omp/occa /ocl/occa 167endif 168ifneq ($(wildcard $(MAGMA_DIR)/lib/libmagma.*),) 169 CUDA_LIB_DIR := $(wildcard $(foreach d,lib lib64,$(CUDA_DIR)/$d/libcudart.${SO_EXT})) 170 CUDA_LIB_DIR := $(patsubst %/,%,$(dir $(firstword $(CUDA_LIB_DIR)))) 171 ifneq ($(CUDA_LIB_DIR),) 172 cuda_link = -Wl,-rpath,$(CUDA_LIB_DIR) -L$(CUDA_LIB_DIR) -lcublas -lcusparse -lcudart 173 omp_link = -fopenmp 174 magma_link_static = -L$(MAGMA_DIR)/lib -lmagma $(cuda_link) $(omp_link) 175 magma_link_shared = -L$(MAGMA_DIR)/lib -Wl,-rpath,$(abspath $(MAGMA_DIR)/lib) -lmagma 176 magma_link := $(if $(wildcard $(MAGMA_DIR)/lib/libmagma.${SO_EXT}),$(magma_link_shared),$(magma_link_static)) 177 magma_allsrc.o = $(magma_allsrc.c:%.c=$(OBJDIR)/%.o) $(magma_allsrc.cu:%.cu=$(OBJDIR)/%.o) 178 $(libceed) : LDLIBS += $(magma_link) 179 $(tests) $(examples) : LDLIBS += $(magma_link) 180 $(libceed) : $(magma_allsrc.o) 181 libceed.c += $(magma_allsrc.c) 182 libceed.cu += $(magma_allsrc.cu) 183 $(magma_allsrc.c:%.c=$(OBJDIR)/%.o) : CFLAGS += -DADD_ -I$(MAGMA_DIR)/include -I$(CUDA_DIR)/include 184 $(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 185 BACKENDS += /gpu/magma 186 endif 187endif 188 189export BACKENDS 190 191# generate magma_tmp.c and magma_cuda.cu from magma.c 192$(magma_tmp.c) $(magma_tmp.cu): $(magma_pre_src) | $$(@D)/.DIR 193 $(magma_preprocessor) $< 194 195$(libceed) : $(libceed.c:%.c=$(OBJDIR)/%.o) $(libceed.cu:%.cu=$(OBJDIR)/%.o) | $$(@D)/.DIR 196 $(call quiet,CC) $(LDFLAGS) -shared -o $@ $^ $(LDLIBS) 197 198$(OBJDIR)/%.o : %.c | $$(@D)/.DIR 199 $(call quiet,CC) $(CPPFLAGS) $(CFLAGS) -c -o $@ $(abspath $<) 200 201$(OBJDIR)/%.o : %.cu | $$(@D)/.DIR 202 $(call quiet,NVCC) $(CPPFLAGS) $(NVCCFLAGS) -c -o $@ $(abspath $<) 203 204$(OBJDIR)/% : tests/%.c | $$(@D)/.DIR 205 $(call quiet,CC) $(CPPFLAGS) $(CFLAGS) $(LDFLAGS) -o $@ $(abspath $<) -lceed $(LDLIBS) 206 207$(OBJDIR)/% : tests/%.f | $$(@D)/.DIR 208 $(call quiet,FC) $(CPPFLAGS) $(FFLAGS) $(LDFLAGS) -o $@ $(abspath $<) -lceed $(LDLIBS) 209 210$(OBJDIR)/% : examples/ceed/%.c | $$(@D)/.DIR 211 $(call quiet,CC) $(CPPFLAGS) $(CFLAGS) $(LDFLAGS) -o $@ $(abspath $<) -lceed $(LDLIBS) 212 213$(OBJDIR)/% : examples/ceed/%.f | $$(@D)/.DIR 214 $(call quiet,FC) $(CPPFLAGS) $(FFLAGS) $(LDFLAGS) -o $@ $(abspath $<) -lceed $(LDLIBS) 215 216$(tests) $(examples) : $(libceed) 217$(tests) $(examples) : LDFLAGS += -Wl,-rpath,$(abspath $(LIBDIR)) -L$(LIBDIR) 218 219# Build MFEM BPs 220libceed-mfem : 221 make && cd examples && make mfem && cd ../ 222 223run-% : $(OBJDIR)/% 224 @tests/tap.sh $(<:build/%=%) 225# Test core libCEED 226test : $(tests:$(OBJDIR)/%=run-%) $(examples:$(OBJDIR)/%=run-%) 227 228# run test target in parallel 229tst : ;@$(MAKE) $(MFLAGS) V=$(V) test 230# CPU C tests only for backend % 231ctc-% : $(ctests);@$(foreach tst,$(ctests),$(tst) /cpu/$*;) 232 233prove : $(tests) $(examples) 234 $(info Testing backends: $(BACKENDS)) 235 $(PROVE) $(PROVE_OPTS) --exec 'tests/tap.sh' $(tests:$(OBJDIR)/%=%) $(examples:$(OBJDIR)/%=%) 236# run prove target in parallel 237prv : ;@$(MAKE) $(MFLAGS) V=$(V) prove 238 239prove-mfem : $(tests) $(examples) 240 make libceed-mfem 241 $(info Testing backends: $(BACKENDS)) 242 $(PROVE) $(PROVE_OPTS) --exec 'tests/tap.sh' $(tests:$(OBJDIR)/%=%) $(examples:$(OBJDIR)/%=%) 243 $(PROVE) $(PROVE_OPTS) --exec 'tests/tap.sh' $(mfemexamples:$(MFEM_EX_DIR)/%=%) 244 245examples : $(examples) 246 247$(ceed.pc) : pkgconfig-prefix = $(abspath .) 248$(OBJDIR)/ceed.pc : pkgconfig-prefix = $(prefix) 249.INTERMEDIATE : $(OBJDIR)/ceed.pc 250%/ceed.pc : ceed.pc.template | $$(@D)/.DIR 251 @sed "s:%prefix%:$(pkgconfig-prefix):" $< > $@ 252 253# The occa executable is not linked with RPATH by default, so we need it to find its libocca.so 254OCCA := $(if $(DARWIN),DYLD_LIBRARY_PATH,LD_LIBRARY_PATH)=$(OCCA_DIR)/lib $(OCCA_DIR)/bin/occa 255OKL_KERNELS := $(wildcard backends/occa/*.okl) 256 257okl-cache : 258 $(OCCA) cache ceed $(OKL_KERNELS) 259 260okl-clear: 261 $(OCCA) clear -y -l ceed 262 263install : $(libceed) $(OBJDIR)/ceed.pc 264 $(INSTALL) -d "$(DESTDIR)$(includedir)" "$(DESTDIR)$(libdir)" "$(DESTDIR)$(okldir)" "$(DESTDIR)$(pkgconfigdir)" 265 $(INSTALL_DATA) include/ceed.h "$(DESTDIR)$(includedir)/" 266 $(INSTALL_DATA) include/ceedf.h "$(DESTDIR)$(includedir)/" 267 $(INSTALL_DATA) $(libceed) "$(DESTDIR)$(libdir)/" 268 $(INSTALL_DATA) $(OBJDIR)/ceed.pc "$(DESTDIR)$(pkgconfigdir)/" 269 $(INSTALL_DATA) $(OKL_KERNELS) "$(DESTDIR)$(okldir)/" 270 271.PHONY : all cln clean print test tst prove prv examples style install doc okl-cache okl-clear 272 273cln clean : 274 $(RM) *.o *.d $(libceed) 275 $(RM) -r *.dSYM $(OBJDIR) $(LIBDIR)/pkgconfig 276 $(MAKE) -C examples/ceed clean 277 $(MAKE) -C examples/mfem clean 278 $(MAKE) -C examples/petsc clean 279 (cd examples/nek5000 && bash make-nek-examples.sh clean) 280 $(RM) $(magma_tmp.c) $(magma_tmp.cu) backends/magma/*~ backends/magma/*.o 281 282distclean : clean 283 rm -rf doc/html 284 285doc : 286 doxygen Doxyfile 287 288style : 289 astyle --style=google --indent=spaces=2 --max-code-length=80 \ 290 --keep-one-line-statements --keep-one-line-blocks --lineend=linux \ 291 --suffix=none --preserve-date --formatted \ 292 *.[ch] tests/*.[ch] backends/*/*.[ch] examples/*/*.[ch] examples/*/*.[ch]pp 293 294print : 295 @echo $(VAR)=$($(VAR)) 296 297print-% : 298 $(info [ variable name]: $*) 299 $(info [ origin]: $(origin $*)) 300 $(info [ value]: $(value $*)) 301 $(info [expanded value]: $($*)) 302 $(info ) 303 @true 304 305-include $(libceed.c:%.c=build/%.d) $(tests.c:tests/%.c=build/%.d) 306