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# XSMM_DIR env variable should point to XSMM master (github.com/hfp/libxsmm) 37XSMM_DIR ?= ../libxsmm 38 39# OCCA_DIR env variable should point to OCCA master (github.com/libocca/occa) 40OCCA_DIR ?= ../occa 41 42# env variable MAGMA_DIR can be used too 43MAGMA_DIR ?= ../magma 44# If CUDA_DIR is not set, check for nvcc, or resort to /usr/local/cuda 45CUDA_DIR ?= $(or $(patsubst %/,%,$(dir $(patsubst %/,%,$(dir \ 46 $(shell which nvcc 2> /dev/null))))),/usr/local/cuda) 47 48# Warning: SANTIZ options still don't run with /gpu/occa 49# export LSAN_OPTIONS=suppressions=.asanignore 50AFLAGS = -fsanitize=address #-fsanitize=undefined -fno-omit-frame-pointer 51 52OPT = -O -g -march=native -ffp-contract=fast 53CFLAGS = -std=c99 $(OPT) -Wall -Wextra -Wno-unused-parameter -fPIC -MMD -MP 54NVCCFLAGS = -Xcompiler "$(OPT)" -Xcompiler -fPIC 55# If using the IBM XL Fortran (xlf) replace FFLAGS appropriately: 56ifneq ($(filter %xlf %xlf_r,$(FC)),) 57 FFLAGS = $(OPT) -ffree-form -qpreprocess -qextname -qpic -MMD 58else # gfortran/Intel-style options 59 FFLAGS = -cpp $(OPT) -ffree-form -Wall -Wextra -Wno-unused-parameter -Wno-unused-dummy-argument -fPIC -MMD -MP 60endif 61 62ifeq ($(UNDERSCORE), 1) 63 CFLAGS += -DUNDERSCORE 64endif 65 66ifeq ($(COVERAGE), 1) 67 CFLAGS += --coverage 68 LDFLAGS += --coverage 69endif 70 71CFLAGS += $(if $(ASAN),$(AFLAGS)) 72FFLAGS += $(if $(ASAN),$(AFLAGS)) 73LDFLAGS += $(if $(ASAN),$(AFLAGS)) 74CPPFLAGS = -I./include 75LDLIBS = -lm 76OBJDIR := build 77LIBDIR := lib 78 79# Installation variables 80prefix ?= /usr/local 81bindir = $(prefix)/bin 82libdir = $(prefix)/lib 83okldir = $(libdir)/okl 84includedir = $(prefix)/include 85pkgconfigdir = $(libdir)/pkgconfig 86INSTALL = install 87INSTALL_PROGRAM = $(INSTALL) 88INSTALL_DATA = $(INSTALL) -m644 89 90# Get number of processors of the machine 91NPROCS := $(shell getconf _NPROCESSORS_ONLN) 92# prepare make options to run in parallel 93MFLAGS := -j $(NPROCS) --warn-undefined-variables \ 94 --no-print-directory --no-keep-going 95 96PROVE ?= prove 97PROVE_OPTS ?= -j $(NPROCS) 98DARWIN := $(filter Darwin,$(shell uname -s)) 99SO_EXT := $(if $(DARWIN),dylib,so) 100 101ceed.pc := $(LIBDIR)/pkgconfig/ceed.pc 102libceed := $(LIBDIR)/libceed.$(SO_EXT) 103libceed.c := $(wildcard interface/ceed*.c) 104BACKENDS_BUILTIN := /cpu/self/ref /cpu/self/tmpl /cpu/self/blocked 105BACKENDS := $(BACKENDS_BUILTIN) 106 107# Tests 108tests.c := $(sort $(wildcard tests/t[0-9][0-9][0-9]-*.c)) 109tests.f := $(sort $(wildcard tests/t[0-9][0-9][0-9]-*.f)) 110tests := $(tests.c:tests/%.c=$(OBJDIR)/%) 111ctests := $(tests) 112tests += $(tests.f:tests/%.f=$(OBJDIR)/%) 113#examples 114examples.c := $(sort $(wildcard examples/ceed/*.c)) 115examples.f := $(sort $(wildcard examples/ceed/*.f)) 116examples := $(examples.c:examples/ceed/%.c=$(OBJDIR)/%) 117examples += $(examples.f:examples/ceed/%.f=$(OBJDIR)/%) 118#mfemexamples 119mfemexamples.cpp := $(sort $(wildcard examples/mfem/*.cpp)) 120mfemexamples := $(mfemexamples.cpp:examples/mfem/%.cpp=$(OBJDIR)/mfem-%) 121petscexamples.c := $(sort $(wildcard examples/petsc/*.c)) 122petscexamples := $(petscexamples.c:examples/petsc/%.c=$(OBJDIR)/petsc-%) 123 124# backends/[ref, template, blocked, avx, occa, magma] 125ref.c := $(sort $(wildcard backends/ref/*.c)) 126template.c := $(sort $(wildcard backends/template/*.c)) 127cuda.c := $(sort $(wildcard backends/cuda/*.c)) 128cuda.cu := $(sort $(wildcard backends/cuda/*.cu)) 129blocked.c := $(sort $(wildcard backends/blocked/*.c)) 130avx.c := $(sort $(wildcard backends/avx/*.c)) 131xsmm.c := $(sort $(wildcard backends/xsmm/*.c)) 132occa.c := $(sort $(wildcard backends/occa/*.c)) 133magma_preprocessor := python backends/magma/gccm.py 134magma_pre_src := $(filter-out %_tmp.c, $(wildcard backends/magma/ceed-*.c)) 135magma_dsrc := $(wildcard backends/magma/magma_d*.c) 136magma_tmp.c := $(magma_pre_src:%.c=%_tmp.c) 137magma_tmp.cu := $(magma_pre_src:%.c=%_cuda.cu) 138magma_allsrc.c := $(magma_dsrc) $(magma_tmp.c) 139magma_allsrc.cu:= $(magma_tmp.cu) 140 141# Output using the 216-color rules mode 142rule_file = $(notdir $(1)) 143rule_path = $(patsubst %/,%,$(dir $(1))) 144last_path = $(notdir $(patsubst %/,%,$(dir $(1)))) 145ansicolor = $(shell echo $(call last_path,$(1)) | cksum | cut -b1-2 | xargs -IS expr 2 \* S + 17) 146emacs_out = @printf " %10s %s/%s\n" $(1) $(call rule_path,$(2)) $(call rule_file,$(2)) 147color_out = @if [ -t 1 ]; then \ 148 printf " %10s \033[38;5;%d;1m%s\033[m/%s\n" \ 149 $(1) $(call ansicolor,$(2)) \ 150 $(call rule_path,$(2)) $(call rule_file,$(2)); else \ 151 printf " %10s %s\n" $(1) $(2); fi 152# if TERM=dumb, use it, otherwise switch to the term one 153output = $(if $(TERM:dumb=),$(call color_out,$1,$2),$(call emacs_out,$1,$2)) 154 155# if V is set to non-nil, turn the verbose mode 156quiet = $(if $(V),$($(1)),$(call output,$1,$@);$($(1))) 157 158# Cancel built-in and old-fashioned implicit rules which we don't use 159.SUFFIXES: 160 161.SECONDEXPANSION: # to expand $$(@D)/.DIR 162 163.SECONDARY: $(magma_tmp.c) $(magma_tmp.cu) 164 165%/.DIR : 166 @mkdir -p $(@D) 167 @touch $@ 168 169.PRECIOUS: %/.DIR 170 171lib: $(libceed) $(ceed.pc) 172# run 'lib' target in parallel 173all:;@$(MAKE) $(MFLAGS) V=$(V) lib 174backend_status = $(if $(filter $1,$(BACKENDS)), [backends: $1], [not found]) 175info: 176 $(info ------------------------------------) 177 $(info CC = $(CC)) 178 $(info FC = $(FC)) 179 $(info CPPFLAGS = $(CPPFLAGS)) 180 $(info CFLAGS = $(value CFLAGS)) 181 $(info FFLAGS = $(value FFLAGS)) 182 $(info NVCCFLAGS = $(value NVCCFLAGS)) 183 $(info LDFLAGS = $(value LDFLAGS)) 184 $(info LDLIBS = $(LDLIBS)) 185 $(info OPT = $(OPT)) 186 $(info AFLAGS = $(AFLAGS)) 187 $(info ASAN = $(or $(ASAN),(empty))) 188 $(info V = $(or $(V),(empty)) [verbose=$(if $(V),on,off)]) 189 $(info ------------------------------------) 190 $(info AVX_STATUS = $(AVX_STATUS)$(call backend_status,/cpu/self/avx)) 191 $(info XSMM_DIR = $(XSMM_DIR)$(call backend_status,/cpu/xsmm/serial /cpu/xsmm/blocked)) 192 $(info OCCA_DIR = $(OCCA_DIR)$(call backend_status,/cpu/occa /gpu/occa /omp/occa)) 193 $(info MAGMA_DIR = $(MAGMA_DIR)$(call backend_status,/gpu/magma)) 194 $(info CUDA_DIR = $(CUDA_DIR)$(call backend_status,/gpu/magma)) 195 $(info ------------------------------------) 196 $(info MFEM_DIR = $(MFEM_DIR)) 197 $(info PETSC_DIR = $(PETSC_DIR)) 198 $(info ------------------------------------) 199 $(info prefix = $(prefix)) 200 $(info includedir = $(value includedir)) 201 $(info libdir = $(value libdir)) 202 $(info okldir = $(value okldir)) 203 $(info pkgconfigdir = $(value pkgconfigdir)) 204 $(info ------------------------------------) 205 @true 206info-backends: 207 $(info make: 'lib' with optional backends: $(filter-out $(BACKENDS_BUILTIN),$(BACKENDS))) 208.PHONY: lib all info info-backends 209 210$(libceed) : LDFLAGS += $(if $(DARWIN), -install_name @rpath/$(notdir $(libceed))) 211 212# Standard Backends 213libceed.c += $(ref.c) 214libceed.c += $(template.c) 215libceed.c += $(blocked.c) 216 217# AVX Backed 218AVX_STATUS = Disabled 219AVX := $(shell $(CC) $(CFLAGS) -v -E - < /dev/null 2>&1 | grep -c avx) 220ifeq ($(AVX),1) 221 AVX_STATUS = Enabled 222 libceed.c += $(avx.c) 223 BACKENDS += /cpu/self/avx 224endif 225 226# libXSMM Backends 227ifneq ($(wildcard $(XSMM_DIR)/lib/libxsmm.*),) 228 $(libceed) : LDFLAGS += -L$(XSMM_DIR)/lib -Wl,-rpath,$(abspath $(XSMM_DIR)/lib) 229 $(libceed) : LDLIBS += -lxsmm -ldl -lblas 230 libceed.c += $(xsmm.c) 231 $(xsmm.c:%.c=$(OBJDIR)/%.o) : CFLAGS += -I$(XSMM_DIR)/include 232 BACKENDS += /cpu/self/xsmm/serial /cpu/self/xsmm/blocked 233endif 234 235# OCCA Backends 236ifneq ($(wildcard $(OCCA_DIR)/lib/libocca.*),) 237 $(libceed) : LDFLAGS += -L$(OCCA_DIR)/lib -Wl,-rpath,$(abspath $(OCCA_DIR)/lib) 238 $(libceed) : LDLIBS += -locca 239 libceed.c += $(occa.c) 240 $(occa.c:%.c=$(OBJDIR)/%.o) : CFLAGS += -I$(OCCA_DIR)/include 241 BACKENDS += /cpu/occa /gpu/occa /omp/occa 242endif 243 244# Cuda Backend 245CUDA_LIB_DIR := $(wildcard $(foreach d,lib lib64,$(CUDA_DIR)/$d/libcudart.${SO_EXT})) 246CUDA_LIB_DIR := $(patsubst %/,%,$(dir $(firstword $(CUDA_LIB_DIR)))) 247ifneq ($(CUDA_LIB_DIR),) 248 $(libceed) : CFLAGS += -I$(CUDA_DIR)/include 249 $(libceed) : LDFLAGS += -L$(CUDA_LIB_DIR) -Wl,-rpath,$(abspath $(CUDA_LIB_DIR)) 250 $(libceed) : LDLIBS += -lcudart -lnvrtc -lcuda 251 libceed.c += $(cuda.c) 252 libceed.cu += $(cuda.cu) 253 BACKENDS += /gpu/cuda 254endif 255 256# MAGMA Backend 257ifneq ($(wildcard $(MAGMA_DIR)/lib/libmagma.*),) 258 ifneq ($(CUDA_LIB_DIR),) 259 cuda_link = -Wl,-rpath,$(CUDA_LIB_DIR) -L$(CUDA_LIB_DIR) -lcublas -lcusparse -lcudart 260 omp_link = -fopenmp 261 magma_link_static = -L$(MAGMA_DIR)/lib -lmagma $(cuda_link) $(omp_link) 262 magma_link_shared = -L$(MAGMA_DIR)/lib -Wl,-rpath,$(abspath $(MAGMA_DIR)/lib) -lmagma 263 magma_link := $(if $(wildcard $(MAGMA_DIR)/lib/libmagma.${SO_EXT}),$(magma_link_shared),$(magma_link_static)) 264 $(libceed) : LDLIBS += $(magma_link) 265 $(tests) $(examples) : LDLIBS += $(magma_link) 266 libceed.c += $(magma_allsrc.c) 267 libceed.cu += $(magma_allsrc.cu) 268 $(magma_allsrc.c:%.c=$(OBJDIR)/%.o) : CFLAGS += -DADD_ -I$(MAGMA_DIR)/include -I$(CUDA_DIR)/include 269 $(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 270 BACKENDS += /gpu/magma 271 endif 272endif 273 274export BACKENDS 275 276# generate magma_tmp.c and magma_cuda.cu from magma.c 277%_tmp.c %_cuda.cu : %.c 278 $(magma_preprocessor) $< 279 280libceed.o = $(libceed.c:%.c=$(OBJDIR)/%.o) $(libceed.cu:%.cu=$(OBJDIR)/%.o) 281$(libceed.o): | info-backends 282$(libceed) : $(libceed.o) | $$(@D)/.DIR 283 $(call quiet,CC) $(LDFLAGS) -shared -o $@ $^ $(LDLIBS) 284 285$(OBJDIR)/%.o : $(CURDIR)/%.c | $$(@D)/.DIR 286 $(call quiet,CC) $(CPPFLAGS) $(CFLAGS) -c -o $@ $(abspath $<) 287 288$(OBJDIR)/%.o : $(CURDIR)/%.cu | $$(@D)/.DIR 289 $(call quiet,NVCC) $(CPPFLAGS) $(NVCCFLAGS) -c -o $@ $(abspath $<) 290 291$(OBJDIR)/% : tests/%.c | $$(@D)/.DIR 292 $(call quiet,LINK.c) -o $@ $(abspath $<) -lceed $(LDLIBS) 293 294$(OBJDIR)/% : tests/%.f | $$(@D)/.DIR 295 $(call quiet,LINK.F) -o $@ $(abspath $<) -lceed $(LDLIBS) 296 297$(OBJDIR)/% : examples/ceed/%.c | $$(@D)/.DIR 298 $(call quiet,LINK.c) -o $@ $(abspath $<) -lceed $(LDLIBS) 299 300$(OBJDIR)/% : examples/ceed/%.f | $$(@D)/.DIR 301 $(call quiet,LINK.F) -o $@ $(abspath $<) -lceed $(LDLIBS) 302 303$(OBJDIR)/mfem-% : examples/mfem/%.cpp $(libceed) | $$(@D)/.DIR 304 $(MAKE) -C examples/mfem CEED_DIR=`pwd` $* 305 mv examples/mfem/$* $@ 306 307$(OBJDIR)/petsc-% : examples/petsc/%.c $(libceed) $(ceed.pc) | $$(@D)/.DIR 308 $(MAKE) -C examples/petsc CEED_DIR=`pwd` $* 309 mv examples/petsc/$* $@ 310 311$(tests) $(examples) : $(libceed) 312$(tests) $(examples) : LDFLAGS += -Wl,-rpath,$(abspath $(LIBDIR)) -L$(LIBDIR) 313 314run-% : $(OBJDIR)/% 315 @tests/tap.sh $(<:build/%=%) 316# Test core libCEED 317test : $(tests:$(OBJDIR)/%=run-%) $(examples:$(OBJDIR)/%=run-%) 318 319# run test target in parallel 320tst : ;@$(MAKE) $(MFLAGS) V=$(V) test 321# CPU C tests only for backend % 322ctc-% : $(ctests);@$(foreach tst,$(ctests),$(tst) /cpu/$*;) 323 324prove : $(tests) $(examples) 325 $(info Testing backends: $(BACKENDS)) 326 $(PROVE) $(PROVE_OPTS) --exec 'tests/tap.sh' $(tests:$(OBJDIR)/%=%) $(examples:$(OBJDIR)/%=%) 327# run prove target in parallel 328prv : ;@$(MAKE) $(MFLAGS) V=$(V) prove 329 330alltests := $(tests) $(examples) $(if $(MFEM_DIR),$(mfemexamples)) $(if $(PETSC_DIR),$(petscexamples)) 331prove-all : $(alltests) 332 $(info Testing backends: $(BACKENDS)) 333 $(PROVE) $(PROVE_OPTS) --exec 'tests/tap.sh' $(alltests:$(OBJDIR)/%=%) 334 335examples : $(examples) 336 337$(ceed.pc) : pkgconfig-prefix = $(abspath .) 338$(OBJDIR)/ceed.pc : pkgconfig-prefix = $(prefix) 339.INTERMEDIATE : $(OBJDIR)/ceed.pc 340%/ceed.pc : ceed.pc.template | $$(@D)/.DIR 341 @sed "s:%prefix%:$(pkgconfig-prefix):" $< > $@ 342 343OCCA := $(OCCA_DIR)/bin/occa 344OKL_KERNELS := $(wildcard backends/occa/*.okl) 345 346okl-cache : 347 $(OCCA) cache ceed $(OKL_KERNELS) 348 349okl-clear: 350 $(OCCA) clear -y -l ceed 351 352install : $(libceed) $(OBJDIR)/ceed.pc 353 $(INSTALL) -d $(addprefix $(if $(DESTDIR),"$(DESTDIR)"),"$(includedir)"\ 354 "$(libdir)" "$(pkgconfigdir)" $(if $(OCCA_ON),"$(okldir)")) 355 $(INSTALL_DATA) include/ceed.h "$(DESTDIR)$(includedir)/" 356 $(INSTALL_DATA) include/ceedf.h "$(DESTDIR)$(includedir)/" 357 $(INSTALL_DATA) $(libceed) "$(DESTDIR)$(libdir)/" 358 $(INSTALL_DATA) $(OBJDIR)/ceed.pc "$(DESTDIR)$(pkgconfigdir)/" 359 $(if $(OCCA_ON),$(INSTALL_DATA) $(OKL_KERNELS) "$(DESTDIR)$(okldir)/") 360 361.PHONY : cln clean print test tst prove prv examples style install doc okl-cache okl-clear 362 363cln clean : 364 $(RM) -r $(OBJDIR) $(LIBDIR) 365 $(MAKE) -C examples/ceed clean 366 $(MAKE) -C examples/mfem clean 367 $(MAKE) -C examples/petsc clean 368 (cd examples/nek5000 && bash make-nek-examples.sh clean) 369 $(RM) $(magma_tmp.c) $(magma_tmp.cu) backends/magma/*~ backends/magma/*.o 370 371distclean : clean 372 $(RM) -r doc/html 373 374doc : 375 doxygen Doxyfile 376 377style : 378 astyle --style=google --indent=spaces=2 --max-code-length=80 \ 379 --keep-one-line-statements --keep-one-line-blocks --lineend=linux \ 380 --suffix=none --preserve-date --formatted \ 381 --exclude=include/ceedf.h --exclude=tests/t310-basis-f.h \ 382 include/*.h interface/*.[ch] tests/*.[ch] backends/*/*.[ch] \ 383 examples/*/*.[ch] examples/*/*.[ch]pp -i 384 385print : 386 @echo $(VAR)=$($(VAR)) 387 388print-% : 389 $(info [ variable name]: $*) 390 $(info [ origin]: $(origin $*)) 391 $(info [ value]: $(value $*)) 392 $(info [expanded value]: $($*)) 393 $(info ) 394 @true 395 396-include $(libceed.c:%.c=build/%.d) $(tests.c:tests/%.c=build/%.d) 397