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# NEK5K_DIR env variable should point to sibling directory 37ifneq ($(wildcard ../Nek5000/*),) 38 NEK5K_DIR ?= ../Nek5000 39endif 40 41# XSMM_DIR env variable should point to XSMM master (github.com/hfp/libxsmm) 42XSMM_DIR ?= ../libxsmm 43 44# OCCA_DIR env variable should point to OCCA master (github.com/libocca/occa) 45OCCA_DIR ?= ../occa 46 47# env variable MAGMA_DIR can be used too 48MAGMA_DIR ?= ../magma 49# If CUDA_DIR is not set, check for nvcc, or resort to /usr/local/cuda 50CUDA_DIR ?= $(or $(patsubst %/,%,$(dir $(patsubst %/,%,$(dir \ 51 $(shell which nvcc 2> /dev/null))))),/usr/local/cuda) 52 53# Check for PETSc in ../petsc 54ifneq ($(wildcard ../petsc/lib/libpetsc.*),) 55 PETSC_DIR ?= ../petsc 56endif 57 58# Warning: SANTIZ options still don't run with /gpu/occa 59# export LSAN_OPTIONS=suppressions=.asanignore 60AFLAGS = -fsanitize=address #-fsanitize=undefined -fno-omit-frame-pointer 61 62OPT = -O -g -march=native -ffp-contract=fast 63CFLAGS = -std=c99 $(OPT) -Wall -Wextra -Wno-unused-parameter -fPIC -MMD -MP 64NVCCFLAGS = -Xcompiler "$(OPT)" -Xcompiler -fPIC 65# If using the IBM XL Fortran (xlf) replace FFLAGS appropriately: 66ifneq ($(filter %xlf %xlf_r,$(FC)),) 67 FFLAGS = $(OPT) -ffree-form -qpreprocess -qextname -qpic -MMD 68else # gfortran/Intel-style options 69 FFLAGS = -cpp $(OPT) -Wall -Wextra -Wno-unused-parameter -Wno-unused-dummy-argument -fPIC -MMD -MP 70endif 71 72ifeq ($(UNDERSCORE), 1) 73 CFLAGS += -DUNDERSCORE 74endif 75 76ifeq ($(COVERAGE), 1) 77 CFLAGS += --coverage 78 LDFLAGS += --coverage 79endif 80 81CFLAGS += $(if $(ASAN),$(AFLAGS)) 82FFLAGS += $(if $(ASAN),$(AFLAGS)) 83LDFLAGS += $(if $(ASAN),$(AFLAGS)) 84CPPFLAGS = -I./include 85LDLIBS = -lm 86OBJDIR := build 87LIBDIR := lib 88 89# Installation variables 90prefix ?= /usr/local 91bindir = $(prefix)/bin 92libdir = $(prefix)/lib 93okldir = $(libdir)/okl 94includedir = $(prefix)/include 95pkgconfigdir = $(libdir)/pkgconfig 96INSTALL = install 97INSTALL_PROGRAM = $(INSTALL) 98INSTALL_DATA = $(INSTALL) -m644 99 100# Get number of processors of the machine 101NPROCS := $(shell getconf _NPROCESSORS_ONLN) 102# prepare make options to run in parallel 103MFLAGS := -j $(NPROCS) --warn-undefined-variables \ 104 --no-print-directory --no-keep-going 105 106PROVE ?= prove 107PROVE_OPTS ?= -j $(NPROCS) 108DARWIN := $(filter Darwin,$(shell uname -s)) 109SO_EXT := $(if $(DARWIN),dylib,so) 110 111ceed.pc := $(LIBDIR)/pkgconfig/ceed.pc 112libceed := $(LIBDIR)/libceed.$(SO_EXT) 113libceed.c := $(wildcard interface/ceed*.c) 114BACKENDS_BUILTIN := /cpu/self/ref /cpu/self/tmpl /cpu/self/blocked 115BACKENDS := $(BACKENDS_BUILTIN) 116 117# Tests 118tests.c := $(sort $(wildcard tests/t[0-9][0-9][0-9]-*.c)) 119tests.f := $(sort $(wildcard tests/t[0-9][0-9][0-9]-*.f90)) 120tests := $(tests.c:tests/%.c=$(OBJDIR)/%) 121ctests := $(tests) 122tests += $(tests.f:tests/%.f90=$(OBJDIR)/%) 123#examples 124examples.c := $(sort $(wildcard examples/ceed/*.c)) 125examples.f := $(sort $(wildcard examples/ceed/*.f)) 126examples := $(examples.c:examples/ceed/%.c=$(OBJDIR)/%) 127examples += $(examples.f:examples/ceed/%.f=$(OBJDIR)/%) 128#mfemexamples 129mfemexamples.cpp := $(sort $(wildcard examples/mfem/*.cpp)) 130mfemexamples := $(mfemexamples.cpp:examples/mfem/%.cpp=$(OBJDIR)/mfem-%) 131#nekexamples 132nekexamples.usr := $(sort $(wildcard examples/nek5000/*.usr)) 133nekexamples := $(nekexamples.usr:examples/nek5000/%.usr=nek-%) 134#petscexamples 135petscexamples.c := $(sort $(wildcard examples/petsc/*.c)) 136petscexamples := $(petscexamples.c:examples/petsc/%.c=$(OBJDIR)/petsc-%) 137 138# backends/[ref, template, blocked, avx, occa, magma] 139ref.c := $(sort $(wildcard backends/ref/*.c)) 140template.c := $(sort $(wildcard backends/template/*.c)) 141cuda.c := $(sort $(wildcard backends/cuda/*.c)) 142cuda.cu := $(sort $(wildcard backends/cuda/*.cu)) 143blocked.c := $(sort $(wildcard backends/blocked/*.c)) 144avx.c := $(sort $(wildcard backends/avx/*.c)) 145xsmm.c := $(sort $(wildcard backends/xsmm/*.c)) 146occa.c := $(sort $(wildcard backends/occa/*.c)) 147magma_preprocessor := python backends/magma/gccm.py 148magma_pre_src := $(filter-out %_tmp.c, $(wildcard backends/magma/ceed-*.c)) 149magma_dsrc := $(wildcard backends/magma/magma_d*.c) 150magma_tmp.c := $(magma_pre_src:%.c=%_tmp.c) 151magma_tmp.cu := $(magma_pre_src:%.c=%_cuda.cu) 152magma_allsrc.c := $(magma_dsrc) $(magma_tmp.c) 153magma_allsrc.cu:= $(magma_tmp.cu) 154 155# Output using the 216-color rules mode 156rule_file = $(notdir $(1)) 157rule_path = $(patsubst %/,%,$(dir $(1))) 158last_path = $(notdir $(patsubst %/,%,$(dir $(1)))) 159ansicolor = $(shell echo $(call last_path,$(1)) | cksum | cut -b1-2 | xargs -IS expr 2 \* S + 17) 160emacs_out = @printf " %10s %s/%s\n" $(1) $(call rule_path,$(2)) $(call rule_file,$(2)) 161color_out = @if [ -t 1 ]; then \ 162 printf " %10s \033[38;5;%d;1m%s\033[m/%s\n" \ 163 $(1) $(call ansicolor,$(2)) \ 164 $(call rule_path,$(2)) $(call rule_file,$(2)); else \ 165 printf " %10s %s\n" $(1) $(2); fi 166# if TERM=dumb, use it, otherwise switch to the term one 167output = $(if $(TERM:dumb=),$(call color_out,$1,$2),$(call emacs_out,$1,$2)) 168 169# if V is set to non-nil, turn the verbose mode 170quiet = $(if $(V),$($(1)),$(call output,$1,$@);$($(1))) 171 172# Cancel built-in and old-fashioned implicit rules which we don't use 173.SUFFIXES: 174 175.SECONDEXPANSION: # to expand $$(@D)/.DIR 176 177.SECONDARY: $(magma_tmp.c) $(magma_tmp.cu) 178 179%/.DIR : 180 @mkdir -p $(@D) 181 @touch $@ 182 183.PRECIOUS: %/.DIR 184 185lib: $(libceed) $(ceed.pc) 186# run 'lib' target in parallel 187par:;@$(MAKE) $(MFLAGS) V=$(V) lib 188backend_status = $(if $(filter $1,$(BACKENDS)), [backends: $1], [not found]) 189info: 190 $(info ------------------------------------) 191 $(info CC = $(CC)) 192 $(info FC = $(FC)) 193 $(info CPPFLAGS = $(CPPFLAGS)) 194 $(info CFLAGS = $(value CFLAGS)) 195 $(info FFLAGS = $(value FFLAGS)) 196 $(info NVCCFLAGS = $(value NVCCFLAGS)) 197 $(info LDFLAGS = $(value LDFLAGS)) 198 $(info LDLIBS = $(LDLIBS)) 199 $(info OPT = $(OPT)) 200 $(info AFLAGS = $(AFLAGS)) 201 $(info ASAN = $(or $(ASAN),(empty))) 202 $(info V = $(or $(V),(empty)) [verbose=$(if $(V),on,off)]) 203 $(info ------------------------------------) 204 $(info AVX_STATUS = $(AVX_STATUS)$(call backend_status,/cpu/self/avx)) 205 $(info XSMM_DIR = $(XSMM_DIR)$(call backend_status,/cpu/xsmm/serial /cpu/xsmm/blocked)) 206 $(info OCCA_DIR = $(OCCA_DIR)$(call backend_status,/cpu/occa /gpu/occa /omp/occa)) 207 $(info MAGMA_DIR = $(MAGMA_DIR)$(call backend_status,/gpu/magma)) 208 $(info CUDA_DIR = $(CUDA_DIR)$(call backend_status,/gpu/magma)) 209 $(info ------------------------------------) 210 $(info MFEM_DIR = $(MFEM_DIR)) 211 $(info NEK5K_DIR = $(NEK5K_DIR)) 212 $(info PETSC_DIR = $(PETSC_DIR)) 213 $(info ------------------------------------) 214 $(info prefix = $(prefix)) 215 $(info includedir = $(value includedir)) 216 $(info libdir = $(value libdir)) 217 $(info okldir = $(value okldir)) 218 $(info pkgconfigdir = $(value pkgconfigdir)) 219 $(info ------------------------------------) 220 @true 221info-backends: 222 $(info make: 'lib' with optional backends: $(filter-out $(BACKENDS_BUILTIN),$(BACKENDS))) 223.PHONY: lib all par info info-backends 224 225$(libceed) : LDFLAGS += $(if $(DARWIN), -install_name @rpath/$(notdir $(libceed))) 226 227# Standard Backends 228libceed.c += $(ref.c) 229libceed.c += $(template.c) 230libceed.c += $(blocked.c) 231 232# AVX Backed 233AVX_STATUS = Disabled 234AVX := $(shell $(CC) $(CFLAGS) -v -E - < /dev/null 2>&1 | grep -c avx) 235ifeq ($(AVX),1) 236 AVX_STATUS = Enabled 237 libceed.c += $(avx.c) 238 BACKENDS += /cpu/self/avx 239endif 240 241# libXSMM Backends 242ifneq ($(wildcard $(XSMM_DIR)/lib/libxsmm.*),) 243 $(libceed) : LDFLAGS += -L$(XSMM_DIR)/lib -Wl,-rpath,$(abspath $(XSMM_DIR)/lib) 244 $(libceed) : LDLIBS += -lxsmm -ldl -lblas 245 libceed.c += $(xsmm.c) 246 $(xsmm.c:%.c=$(OBJDIR)/%.o) : CFLAGS += -I$(XSMM_DIR)/include 247 BACKENDS += /cpu/self/xsmm/serial /cpu/self/xsmm/blocked 248endif 249 250# OCCA Backends 251ifneq ($(wildcard $(OCCA_DIR)/lib/libocca.*),) 252 $(libceed) : LDFLAGS += -L$(OCCA_DIR)/lib -Wl,-rpath,$(abspath $(OCCA_DIR)/lib) 253 $(libceed) : LDLIBS += -locca 254 libceed.c += $(occa.c) 255 $(occa.c:%.c=$(OBJDIR)/%.o) : CFLAGS += -I$(OCCA_DIR)/include 256 BACKENDS += /cpu/occa /gpu/occa /omp/occa 257endif 258 259# Cuda Backend 260CUDA_LIB_DIR := $(wildcard $(foreach d,lib lib64,$(CUDA_DIR)/$d/libcudart.${SO_EXT})) 261CUDA_LIB_DIR := $(patsubst %/,%,$(dir $(firstword $(CUDA_LIB_DIR)))) 262ifneq ($(CUDA_LIB_DIR),) 263 $(libceed) : CFLAGS += -I$(CUDA_DIR)/include 264 $(libceed) : LDFLAGS += -L$(CUDA_LIB_DIR) -Wl,-rpath,$(abspath $(CUDA_LIB_DIR)) 265 $(libceed) : LDLIBS += -lcudart -lnvrtc -lcuda 266 libceed.c += $(cuda.c) 267 libceed.cu += $(cuda.cu) 268 BACKENDS += /gpu/cuda 269endif 270 271# MAGMA Backend 272ifneq ($(wildcard $(MAGMA_DIR)/lib/libmagma.*),) 273 ifneq ($(CUDA_LIB_DIR),) 274 cuda_link = -Wl,-rpath,$(CUDA_LIB_DIR) -L$(CUDA_LIB_DIR) -lcublas -lcusparse -lcudart 275 omp_link = -fopenmp 276 magma_link_static = -L$(MAGMA_DIR)/lib -lmagma $(cuda_link) $(omp_link) 277 magma_link_shared = -L$(MAGMA_DIR)/lib -Wl,-rpath,$(abspath $(MAGMA_DIR)/lib) -lmagma 278 magma_link := $(if $(wildcard $(MAGMA_DIR)/lib/libmagma.${SO_EXT}),$(magma_link_shared),$(magma_link_static)) 279 $(libceed) : LDLIBS += $(magma_link) 280 $(tests) $(examples) : LDLIBS += $(magma_link) 281 libceed.c += $(magma_allsrc.c) 282 libceed.cu += $(magma_allsrc.cu) 283 $(magma_allsrc.c:%.c=$(OBJDIR)/%.o) : CFLAGS += -DADD_ -I$(MAGMA_DIR)/include -I$(CUDA_DIR)/include 284 $(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 285 BACKENDS += /gpu/magma 286 endif 287endif 288 289export BACKENDS 290 291# generate magma_tmp.c and magma_cuda.cu from magma.c 292%_tmp.c %_cuda.cu : %.c 293 $(magma_preprocessor) $< 294 295libceed.o = $(libceed.c:%.c=$(OBJDIR)/%.o) $(libceed.cu:%.cu=$(OBJDIR)/%.o) 296$(libceed.o): | info-backends 297$(libceed) : $(libceed.o) | $$(@D)/.DIR 298 $(call quiet,CC) $(LDFLAGS) -shared -o $@ $^ $(LDLIBS) 299 300$(OBJDIR)/%.o : $(CURDIR)/%.c | $$(@D)/.DIR 301 $(call quiet,CC) $(CPPFLAGS) $(CFLAGS) -c -o $@ $(abspath $<) 302 303$(OBJDIR)/%.o : $(CURDIR)/%.cu | $$(@D)/.DIR 304 $(call quiet,NVCC) $(CPPFLAGS) $(NVCCFLAGS) -c -o $@ $(abspath $<) 305 306$(OBJDIR)/% : tests/%.c | $$(@D)/.DIR 307 $(call quiet,LINK.c) -o $@ $(abspath $<) -lceed $(LDLIBS) 308 309$(OBJDIR)/% : tests/%.f90 | $$(@D)/.DIR 310 $(call quiet,LINK.F) -o $@ $(abspath $<) -lceed $(LDLIBS) 311 312$(OBJDIR)/% : examples/ceed/%.c | $$(@D)/.DIR 313 $(call quiet,LINK.c) -o $@ $(abspath $<) -lceed $(LDLIBS) 314 315$(OBJDIR)/% : examples/ceed/%.f | $$(@D)/.DIR 316 $(call quiet,LINK.F) -o $@ $(abspath $<) -lceed $(LDLIBS) 317 318$(OBJDIR)/mfem-% : examples/mfem/%.cpp $(libceed) | $$(@D)/.DIR 319 $(MAKE) -C examples/mfem CEED_DIR=`pwd` \ 320 MFEM_DIR="$(abspath $(MFEM_DIR))" $* 321 mv examples/mfem/$* $@ 322 323$(OBJDIR)/petsc-% : examples/petsc/%.c $(libceed) $(ceed.pc) | $$(@D)/.DIR 324 $(MAKE) -C examples/petsc CEED_DIR=`pwd` \ 325 PETSC_DIR="$(abspath $(PETSC_DIR))" $* 326 mv examples/petsc/$* $@ 327 328$(tests) $(examples) : $(libceed) 329$(tests) $(examples) : LDFLAGS += -Wl,-rpath,$(abspath $(LIBDIR)) -L$(LIBDIR) 330 331run-% : $(OBJDIR)/% 332 @tests/tap.sh $(<:build/%=%) 333# Test core libCEED 334test : $(tests:$(OBJDIR)/%=run-%) $(examples:$(OBJDIR)/%=run-%) 335 336# run test target in parallel 337tst : ;@$(MAKE) $(MFLAGS) V=$(V) test 338# CPU C tests only for backend % 339ctc-% : $(ctests);@$(foreach tst,$(ctests),$(tst) /cpu/$*;) 340 341prove : $(tests) $(examples) 342 $(info Testing backends: $(BACKENDS)) 343 $(PROVE) $(PROVE_OPTS) --exec 'tests/tap.sh' $(tests:$(OBJDIR)/%=%) $(examples:$(OBJDIR)/%=%) 344# run prove target in parallel 345prv : ;@$(MAKE) $(MFLAGS) V=$(V) prove 346 347allexamples := $(examples) $(if $(MFEM_DIR),$(mfemexamples)) $(if $(PETSC_DIR),$(petscexamples)) 348alltests := $(tests) $(allexamples) 349fulltestlist = $(alltests) $(if $(NEK5K_DIR), $(nekexamples)) 350prepnektests: 351 (export CC FC && cd examples && make prepnektests) 352prove-all : $(alltests) $(if $(NEK5K_DIR), prepnektests) 353 $(info Testing backends: $(BACKENDS)) 354 $(PROVE) $(PROVE_OPTS) --exec 'tests/tap.sh' $(fulltestlist:$(OBJDIR)/%=%) 355 356all: $(alltests) 357 358examples : $(allexamples) 359 360# Benchmarks 361allbenchmarks = petsc-bp1 petsc-bp3 362bench_targets = $(addprefix bench-,$(allbenchmarks)) 363.PHONY: $(bench_targets) benchmarks 364$(bench_targets): bench-%: $(OBJDIR)/% 365 cd benchmarks && ./benchmark.sh --ceed "$(BACKENDS)" -r $(*).sh 366benchmarks: $(bench_targets) 367 368$(ceed.pc) : pkgconfig-prefix = $(abspath .) 369$(OBJDIR)/ceed.pc : pkgconfig-prefix = $(prefix) 370.INTERMEDIATE : $(OBJDIR)/ceed.pc 371%/ceed.pc : ceed.pc.template | $$(@D)/.DIR 372 @sed "s:%prefix%:$(pkgconfig-prefix):" $< > $@ 373 374OCCA := $(OCCA_DIR)/bin/occa 375OKL_KERNELS := $(wildcard backends/occa/*.okl) 376 377okl-cache : 378 $(OCCA) cache ceed $(OKL_KERNELS) 379 380okl-clear: 381 $(OCCA) clear -y -l ceed 382 383install : $(libceed) $(OBJDIR)/ceed.pc 384 $(INSTALL) -d $(addprefix $(if $(DESTDIR),"$(DESTDIR)"),"$(includedir)"\ 385 "$(libdir)" "$(pkgconfigdir)" $(if $(OCCA_ON),"$(okldir)")) 386 $(INSTALL_DATA) include/ceed.h "$(DESTDIR)$(includedir)/" 387 $(INSTALL_DATA) include/ceedf.h "$(DESTDIR)$(includedir)/" 388 $(INSTALL_DATA) $(libceed) "$(DESTDIR)$(libdir)/" 389 $(INSTALL_DATA) $(OBJDIR)/ceed.pc "$(DESTDIR)$(pkgconfigdir)/" 390 $(if $(OCCA_ON),$(INSTALL_DATA) $(OKL_KERNELS) "$(DESTDIR)$(okldir)/") 391 392.PHONY : cln clean doc lib install all print test tst prove prv prove-all examples style okl-cache okl-clear info info-backends 393 394cln clean : 395 $(RM) -r $(OBJDIR) $(LIBDIR) 396 $(MAKE) -C examples/ceed clean 397 $(MAKE) -C examples/mfem clean 398 $(MAKE) -C examples/petsc clean 399 (cd examples/nek5000 && bash make-nek-examples.sh clean) 400 $(RM) $(magma_tmp.c) $(magma_tmp.cu) backends/magma/*~ backends/magma/*.o 401 $(RM) -rf benchmarks/*output.txt 402 403distclean : clean 404 $(RM) -r doc/html 405 406doc : 407 doxygen Doxyfile 408 409style : 410 astyle --style=google --indent=spaces=2 --max-code-length=80 \ 411 --keep-one-line-statements --keep-one-line-blocks --lineend=linux \ 412 --suffix=none --preserve-date --formatted \ 413 --exclude=include/ceedf.h --exclude=tests/t310-basis-f.h \ 414 include/*.h interface/*.[ch] tests/*.[ch] backends/*/*.[ch] \ 415 examples/*/*.[ch] examples/*/*.[ch]pp -i 416 417print : 418 @echo $(VAR)=$($(VAR)) 419 420print-% : 421 $(info [ variable name]: $*) 422 $(info [ origin]: $(origin $*)) 423 $(info [ value]: $(value $*)) 424 $(info [expanded value]: $($*)) 425 $(info ) 426 @true 427 428-include $(libceed.c:%.c=build/%.d) $(tests.c:tests/%.c=build/%.d) 429