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 $(shell which nvcc 2> /dev/null))))),/usr/local/cuda) 47 48# Check for PETSc in ../petsc 49ifneq ($(wildcard ../petsc/lib/libpetsc.*),) 50 PETSC_DIR ?= ../petsc 51endif 52 53# Warning: SANTIZ options still don't run with /gpu/occa 54# export LSAN_OPTIONS=suppressions=.asanignore 55AFLAGS = -fsanitize=address #-fsanitize=undefined -fno-omit-frame-pointer 56 57OPT = -O -g -march=native -ffp-contract=fast 58CFLAGS = -std=c99 $(OPT) -Wall -Wextra -Wno-unused-parameter -fPIC -MMD -MP 59NVCCFLAGS = -Xcompiler "$(OPT)" -Xcompiler -fPIC 60# If using the IBM XL Fortran (xlf) replace FFLAGS appropriately: 61ifneq ($(filter %xlf %xlf_r,$(FC)),) 62 FFLAGS = $(OPT) -ffree-form -qpreprocess -qextname -qpic -MMD 63else # gfortran/Intel-style options 64 FFLAGS = -cpp $(OPT) -ffree-form -Wall -Wextra -Wno-unused-parameter -Wno-unused-dummy-argument -fPIC -MMD -MP 65endif 66 67ifeq ($(UNDERSCORE), 1) 68 CFLAGS += -DUNDERSCORE 69endif 70 71ifeq ($(COVERAGE), 1) 72 CFLAGS += --coverage 73 LDFLAGS += --coverage 74endif 75 76CFLAGS += $(if $(ASAN),$(AFLAGS)) 77FFLAGS += $(if $(ASAN),$(AFLAGS)) 78LDFLAGS += $(if $(ASAN),$(AFLAGS)) 79CPPFLAGS = -I./include 80LDLIBS = -lm 81OBJDIR := build 82LIBDIR := lib 83 84# Installation variables 85prefix ?= /usr/local 86bindir = $(prefix)/bin 87libdir = $(prefix)/lib 88okldir = $(libdir)/okl 89includedir = $(prefix)/include 90pkgconfigdir = $(libdir)/pkgconfig 91INSTALL = install 92INSTALL_PROGRAM = $(INSTALL) 93INSTALL_DATA = $(INSTALL) -m644 94 95# Get number of processors of the machine 96NPROCS := $(shell getconf _NPROCESSORS_ONLN) 97# prepare make options to run in parallel 98MFLAGS := -j $(NPROCS) --warn-undefined-variables \ 99 --no-print-directory --no-keep-going 100 101PROVE ?= prove 102PROVE_OPTS ?= -j $(NPROCS) 103DARWIN := $(filter Darwin,$(shell uname -s)) 104SO_EXT := $(if $(DARWIN),dylib,so) 105 106ceed.pc := $(LIBDIR)/pkgconfig/ceed.pc 107libceed := $(LIBDIR)/libceed.$(SO_EXT) 108libceed.c := $(wildcard interface/ceed*.c) 109BACKENDS_BUILTIN := /cpu/self/ref /cpu/self/tmpl /cpu/self/blocked 110BACKENDS := $(BACKENDS_BUILTIN) 111 112# Tests 113tests.c := $(sort $(wildcard tests/t[0-9][0-9][0-9]-*.c)) 114tests.f := $(sort $(wildcard tests/t[0-9][0-9][0-9]-*.f)) 115tests := $(tests.c:tests/%.c=$(OBJDIR)/%) 116ctests := $(tests) 117tests += $(tests.f:tests/%.f=$(OBJDIR)/%) 118#examples 119examples.c := $(sort $(wildcard examples/ceed/*.c)) 120examples.f := $(sort $(wildcard examples/ceed/*.f)) 121examples := $(examples.c:examples/ceed/%.c=$(OBJDIR)/%) 122examples += $(examples.f:examples/ceed/%.f=$(OBJDIR)/%) 123#mfemexamples 124mfemexamples.cpp := $(sort $(wildcard examples/mfem/*.cpp)) 125mfemexamples := $(mfemexamples.cpp:examples/mfem/%.cpp=$(OBJDIR)/mfem-%) 126petscexamples.c := $(sort $(wildcard examples/petsc/*.c)) 127petscexamples := $(petscexamples.c:examples/petsc/%.c=$(OBJDIR)/petsc-%) 128 129# backends/[ref, template, blocked, avx, occa, magma] 130ref.c := $(sort $(wildcard backends/ref/*.c)) 131template.c := $(sort $(wildcard backends/template/*.c)) 132cuda.c := $(sort $(wildcard backends/cuda/*.c)) 133cuda.cu := $(sort $(wildcard backends/cuda/*.cu)) 134blocked.c := $(sort $(wildcard backends/blocked/*.c)) 135avx.c := $(sort $(wildcard backends/avx/*.c)) 136xsmm.c := $(sort $(wildcard backends/xsmm/*.c)) 137occa.c := $(sort $(wildcard backends/occa/*.c)) 138magma_preprocessor := python backends/magma/gccm.py 139magma_pre_src := $(filter-out %_tmp.c, $(wildcard backends/magma/ceed-*.c)) 140magma_dsrc := $(wildcard backends/magma/magma_d*.c) 141magma_tmp.c := $(magma_pre_src:%.c=%_tmp.c) 142magma_tmp.cu := $(magma_pre_src:%.c=%_cuda.cu) 143magma_allsrc.c := $(magma_dsrc) $(magma_tmp.c) 144magma_allsrc.cu:= $(magma_tmp.cu) 145 146# Output using the 216-color rules mode 147rule_file = $(notdir $(1)) 148rule_path = $(patsubst %/,%,$(dir $(1))) 149last_path = $(notdir $(patsubst %/,%,$(dir $(1)))) 150ansicolor = $(shell echo $(call last_path,$(1)) | cksum | cut -b1-2 | xargs -IS expr 2 \* S + 17) 151emacs_out = @printf " %10s %s/%s\n" $(1) $(call rule_path,$(2)) $(call rule_file,$(2)) 152color_out = @if [ -t 1 ]; then \ 153 printf " %10s \033[38;5;%d;1m%s\033[m/%s\n" \ 154 $(1) $(call ansicolor,$(2)) \ 155 $(call rule_path,$(2)) $(call rule_file,$(2)); else \ 156 printf " %10s %s\n" $(1) $(2); fi 157# if TERM=dumb, use it, otherwise switch to the term one 158output = $(if $(TERM:dumb=),$(call color_out,$1,$2),$(call emacs_out,$1,$2)) 159 160# if V is set to non-nil, turn the verbose mode 161quiet = $(if $(V),$($(1)),$(call output,$1,$@);$($(1))) 162 163# Cancel built-in and old-fashioned implicit rules which we don't use 164.SUFFIXES: 165 166.SECONDEXPANSION: # to expand $$(@D)/.DIR 167 168.SECONDARY: $(magma_tmp.c) $(magma_tmp.cu) 169 170%/.DIR : 171 @mkdir -p $(@D) 172 @touch $@ 173 174.PRECIOUS: %/.DIR 175 176lib: $(libceed) $(ceed.pc) 177# run 'lib' target in parallel 178par:;@$(MAKE) $(MFLAGS) V=$(V) lib 179backend_status = $(if $(filter $1,$(BACKENDS)), [backends: $1], [not found]) 180info: 181 $(info ------------------------------------) 182 $(info CC = $(CC)) 183 $(info FC = $(FC)) 184 $(info CPPFLAGS = $(CPPFLAGS)) 185 $(info CFLAGS = $(value CFLAGS)) 186 $(info FFLAGS = $(value FFLAGS)) 187 $(info NVCCFLAGS = $(value NVCCFLAGS)) 188 $(info LDFLAGS = $(value LDFLAGS)) 189 $(info LDLIBS = $(LDLIBS)) 190 $(info OPT = $(OPT)) 191 $(info AFLAGS = $(AFLAGS)) 192 $(info ASAN = $(or $(ASAN),(empty))) 193 $(info V = $(or $(V),(empty)) [verbose=$(if $(V),on,off)]) 194 $(info ------------------------------------) 195 $(info AVX_STATUS = $(AVX_STATUS)$(call backend_status,/cpu/self/avx)) 196 $(info XSMM_DIR = $(XSMM_DIR)$(call backend_status,/cpu/xsmm/serial /cpu/xsmm/blocked)) 197 $(info OCCA_DIR = $(OCCA_DIR)$(call backend_status,/cpu/occa /gpu/occa /omp/occa)) 198 $(info MAGMA_DIR = $(MAGMA_DIR)$(call backend_status,/gpu/magma)) 199 $(info CUDA_DIR = $(CUDA_DIR)$(call backend_status,/gpu/magma)) 200 $(info ------------------------------------) 201 $(info MFEM_DIR = $(MFEM_DIR)) 202 $(info PETSC_DIR = $(PETSC_DIR)) 203 $(info ------------------------------------) 204 $(info prefix = $(prefix)) 205 $(info includedir = $(value includedir)) 206 $(info libdir = $(value libdir)) 207 $(info okldir = $(value okldir)) 208 $(info pkgconfigdir = $(value pkgconfigdir)) 209 $(info ------------------------------------) 210 @true 211info-backends: 212 $(info make: 'lib' with optional backends: $(filter-out $(BACKENDS_BUILTIN),$(BACKENDS))) 213.PHONY: lib all par info info-backends 214 215$(libceed) : LDFLAGS += $(if $(DARWIN), -install_name @rpath/$(notdir $(libceed))) 216 217# Standard Backends 218libceed.c += $(ref.c) 219libceed.c += $(template.c) 220libceed.c += $(blocked.c) 221 222# AVX Backed 223AVX_STATUS = Disabled 224AVX := $(shell $(CC) $(CFLAGS) -v -E - < /dev/null 2>&1 | grep -c avx) 225ifeq ($(AVX),1) 226 AVX_STATUS = Enabled 227 libceed.c += $(avx.c) 228 BACKENDS += /cpu/self/avx 229endif 230 231# libXSMM Backends 232ifneq ($(wildcard $(XSMM_DIR)/lib/libxsmm.*),) 233 $(libceed) : LDFLAGS += -L$(XSMM_DIR)/lib -Wl,-rpath,$(abspath $(XSMM_DIR)/lib) 234 $(libceed) : LDLIBS += -lxsmm -ldl -lblas 235 libceed.c += $(xsmm.c) 236 $(xsmm.c:%.c=$(OBJDIR)/%.o) : CFLAGS += -I$(XSMM_DIR)/include 237 BACKENDS += /cpu/self/xsmm/serial /cpu/self/xsmm/blocked 238endif 239 240# OCCA Backends 241ifneq ($(wildcard $(OCCA_DIR)/lib/libocca.*),) 242 $(libceed) : LDFLAGS += -L$(OCCA_DIR)/lib -Wl,-rpath,$(abspath $(OCCA_DIR)/lib) 243 $(libceed) : LDLIBS += -locca 244 libceed.c += $(occa.c) 245 $(occa.c:%.c=$(OBJDIR)/%.o) : CFLAGS += -I$(OCCA_DIR)/include 246 BACKENDS += /cpu/occa /gpu/occa /omp/occa 247endif 248 249# Cuda Backend 250CUDA_LIB_DIR := $(wildcard $(foreach d,lib lib64,$(CUDA_DIR)/$d/libcudart.${SO_EXT})) 251CUDA_LIB_DIR := $(patsubst %/,%,$(dir $(firstword $(CUDA_LIB_DIR)))) 252ifneq ($(CUDA_LIB_DIR),) 253 $(libceed) : CFLAGS += -I$(CUDA_DIR)/include 254 $(libceed) : LDFLAGS += -L$(CUDA_LIB_DIR) -Wl,-rpath,$(abspath $(CUDA_LIB_DIR)) 255 $(libceed) : LDLIBS += -lcudart -lnvrtc -lcuda 256 libceed.c += $(cuda.c) 257 libceed.cu += $(cuda.cu) 258 BACKENDS += /gpu/cuda 259endif 260 261# MAGMA Backend 262ifneq ($(wildcard $(MAGMA_DIR)/lib/libmagma.*),) 263 ifneq ($(CUDA_LIB_DIR),) 264 cuda_link = -Wl,-rpath,$(CUDA_LIB_DIR) -L$(CUDA_LIB_DIR) -lcublas -lcusparse -lcudart 265 omp_link = -fopenmp 266 magma_link_static = -L$(MAGMA_DIR)/lib -lmagma $(cuda_link) $(omp_link) 267 magma_link_shared = -L$(MAGMA_DIR)/lib -Wl,-rpath,$(abspath $(MAGMA_DIR)/lib) -lmagma 268 magma_link := $(if $(wildcard $(MAGMA_DIR)/lib/libmagma.${SO_EXT}),$(magma_link_shared),$(magma_link_static)) 269 $(libceed) : LDLIBS += $(magma_link) 270 $(tests) $(examples) : LDLIBS += $(magma_link) 271 libceed.c += $(magma_allsrc.c) 272 libceed.cu += $(magma_allsrc.cu) 273 $(magma_allsrc.c:%.c=$(OBJDIR)/%.o) : CFLAGS += -DADD_ -I$(MAGMA_DIR)/include -I$(CUDA_DIR)/include 274 $(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 275 BACKENDS += /gpu/magma 276 endif 277endif 278 279export BACKENDS 280 281# generate magma_tmp.c and magma_cuda.cu from magma.c 282%_tmp.c %_cuda.cu : %.c 283 $(magma_preprocessor) $< 284 285libceed.o = $(libceed.c:%.c=$(OBJDIR)/%.o) $(libceed.cu:%.cu=$(OBJDIR)/%.o) 286$(libceed.o): | info-backends 287$(libceed) : $(libceed.o) | $$(@D)/.DIR 288 $(call quiet,CC) $(LDFLAGS) -shared -o $@ $^ $(LDLIBS) 289 290$(OBJDIR)/%.o : $(CURDIR)/%.c | $$(@D)/.DIR 291 $(call quiet,CC) $(CPPFLAGS) $(CFLAGS) -c -o $@ $(abspath $<) 292 293$(OBJDIR)/%.o : $(CURDIR)/%.cu | $$(@D)/.DIR 294 $(call quiet,NVCC) $(CPPFLAGS) $(NVCCFLAGS) -c -o $@ $(abspath $<) 295 296$(OBJDIR)/% : tests/%.c | $$(@D)/.DIR 297 $(call quiet,LINK.c) -o $@ $(abspath $<) -lceed $(LDLIBS) 298 299$(OBJDIR)/% : tests/%.f | $$(@D)/.DIR 300 $(call quiet,LINK.F) -o $@ $(abspath $<) -lceed $(LDLIBS) 301 302$(OBJDIR)/% : examples/ceed/%.c | $$(@D)/.DIR 303 $(call quiet,LINK.c) -o $@ $(abspath $<) -lceed $(LDLIBS) 304 305$(OBJDIR)/% : examples/ceed/%.f | $$(@D)/.DIR 306 $(call quiet,LINK.F) -o $@ $(abspath $<) -lceed $(LDLIBS) 307 308$(OBJDIR)/mfem-% : examples/mfem/%.cpp $(libceed) | $$(@D)/.DIR 309 $(MAKE) -C examples/mfem CEED_DIR=`pwd` \ 310 MFEM_DIR="$(abspath $(MFEM_DIR))" $* 311 mv examples/mfem/$* $@ 312 313$(OBJDIR)/petsc-% : examples/petsc/%.c $(libceed) $(ceed.pc) | $$(@D)/.DIR 314 $(MAKE) -C examples/petsc CEED_DIR=`pwd` \ 315 PETSC_DIR="$(abspath $(PETSC_DIR))" $* 316 mv examples/petsc/$* $@ 317 318$(tests) $(examples) : $(libceed) 319$(tests) $(examples) : LDFLAGS += -Wl,-rpath,$(abspath $(LIBDIR)) -L$(LIBDIR) 320 321run-% : $(OBJDIR)/% 322 @tests/tap.sh $(<:build/%=%) 323# Test core libCEED 324test : $(tests:$(OBJDIR)/%=run-%) $(examples:$(OBJDIR)/%=run-%) 325 326# run test target in parallel 327tst : ;@$(MAKE) $(MFLAGS) V=$(V) test 328# CPU C tests only for backend % 329ctc-% : $(ctests);@$(foreach tst,$(ctests),$(tst) /cpu/$*;) 330 331prove : $(tests) $(examples) 332 $(info Testing backends: $(BACKENDS)) 333 $(PROVE) $(PROVE_OPTS) --exec 'tests/tap.sh' $(tests:$(OBJDIR)/%=%) $(examples:$(OBJDIR)/%=%) 334# run prove target in parallel 335prv : ;@$(MAKE) $(MFLAGS) V=$(V) prove 336 337allexamples := $(examples) $(if $(MFEM_DIR),$(mfemexamples)) $(if $(PETSC_DIR),$(petscexamples)) 338alltests := $(tests) $(allexamples) 339prove-all : $(alltests) 340 $(info Testing backends: $(BACKENDS)) 341 $(PROVE) $(PROVE_OPTS) --exec 'tests/tap.sh' $(alltests:$(OBJDIR)/%=%) 342 343all: $(alltests) 344 345examples : $(allexamples) 346 347# Benchmarks 348allbenchmarks = petsc-bp1 petsc-bp3 349bench-targets = $(addprefix bench-,$(allbenchmarks)) 350.PHONY: $(bench-targets) benchmarks 351$(bench-targets): bench-%: $(OBJDIR)/% 352 cd benchmarks && ./benchmark.sh --ceed "$(BACKENDS)" -r $(*).sh 353benchmarks: $(bench-targets) 354 355$(ceed.pc) : pkgconfig-prefix = $(abspath .) 356$(OBJDIR)/ceed.pc : pkgconfig-prefix = $(prefix) 357.INTERMEDIATE : $(OBJDIR)/ceed.pc 358%/ceed.pc : ceed.pc.template | $$(@D)/.DIR 359 @sed "s:%prefix%:$(pkgconfig-prefix):" $< > $@ 360 361OCCA := $(OCCA_DIR)/bin/occa 362OKL_KERNELS := $(wildcard backends/occa/*.okl) 363 364okl-cache : 365 $(OCCA) cache ceed $(OKL_KERNELS) 366 367okl-clear: 368 $(OCCA) clear -y -l ceed 369 370install : $(libceed) $(OBJDIR)/ceed.pc 371 $(INSTALL) -d $(addprefix $(if $(DESTDIR),"$(DESTDIR)"),"$(includedir)"\ 372 "$(libdir)" "$(pkgconfigdir)" $(if $(OCCA_ON),"$(okldir)")) 373 $(INSTALL_DATA) include/ceed.h "$(DESTDIR)$(includedir)/" 374 $(INSTALL_DATA) include/ceedf.h "$(DESTDIR)$(includedir)/" 375 $(INSTALL_DATA) $(libceed) "$(DESTDIR)$(libdir)/" 376 $(INSTALL_DATA) $(OBJDIR)/ceed.pc "$(DESTDIR)$(pkgconfigdir)/" 377 $(if $(OCCA_ON),$(INSTALL_DATA) $(OKL_KERNELS) "$(DESTDIR)$(okldir)/") 378 379.PHONY : cln clean print test tst prove prv examples style install doc okl-cache okl-clear 380 381cln clean : 382 $(RM) -r $(OBJDIR) $(LIBDIR) 383 $(MAKE) -C examples/ceed clean 384 $(MAKE) -C examples/mfem clean 385 $(MAKE) -C examples/petsc clean 386 (cd examples/nek5000 && bash make-nek-examples.sh clean) 387 $(RM) $(magma_tmp.c) $(magma_tmp.cu) backends/magma/*~ backends/magma/*.o 388 $(RM) -rf benchmarks/*output.txt 389 390distclean : clean 391 $(RM) -r doc/html 392 393doc : 394 doxygen Doxyfile 395 396style : 397 astyle --style=google --indent=spaces=2 --max-code-length=80 \ 398 --keep-one-line-statements --keep-one-line-blocks --lineend=linux \ 399 --suffix=none --preserve-date --formatted \ 400 --exclude=include/ceedf.h --exclude=tests/t310-basis-f.h \ 401 include/*.h interface/*.[ch] tests/*.[ch] backends/*/*.[ch] \ 402 examples/*/*.[ch] examples/*/*.[ch]pp -i 403 404print : 405 @echo $(VAR)=$($(VAR)) 406 407print-% : 408 $(info [ variable name]: $*) 409 $(info [ origin]: $(origin $*)) 410 $(info [ value]: $(value $*)) 411 $(info [expanded value]: $($*)) 412 $(info ) 413 @true 414 415-include $(libceed.c:%.c=build/%.d) $(tests.c:tests/%.c=build/%.d) 416