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 17CONFIG ?= config.mk 18-include $(CONFIG) 19 20ifeq (,$(filter-out undefined default,$(origin CC))) 21 CC = gcc 22endif 23ifeq (,$(filter-out undefined default,$(origin CXX))) 24 CXX = g++ 25endif 26ifeq (,$(filter-out undefined default,$(origin FC))) 27 FC = gfortran 28endif 29ifeq (,$(filter-out undefined default,$(origin LINK))) 30 LINK = $(CC) 31endif 32NVCC ?= $(CUDA_DIR)/bin/nvcc 33NVCC_CXX ?= $(CXX) 34HIPCC ?= $(HIP_DIR)/bin/hipcc 35 36# ASAN must be left empty if you don't want to use it 37ASAN ?= 38 39LDFLAGS ?= 40UNDERSCORE ?= 1 41 42# MFEM_DIR env variable should point to sibling directory 43ifneq ($(wildcard ../mfem/libmfem.*),) 44 MFEM_DIR ?= ../mfem 45endif 46 47# NEK5K_DIR env variable should point to sibling directory 48ifneq ($(wildcard ../Nek5000/*),) 49 NEK5K_DIR ?= $(abspath ../Nek5000) 50endif 51export NEK5K_DIR 52MPI ?= 1 53 54# CEED_DIR env for NEK5K testing 55export CEED_DIR = $(abspath .) 56 57# XSMM_DIR env variable should point to XSMM master (github.com/hfp/libxsmm) 58XSMM_DIR ?= ../libxsmm 59 60# OCCA_DIR env variable should point to OCCA master (github.com/libocca/occa) 61OCCA_DIR ?= ../occa 62 63# env variable MAGMA_DIR can be used too 64MAGMA_DIR ?= ../magma 65# If CUDA_DIR is not set, check for nvcc, or resort to /usr/local/cuda 66CUDA_DIR ?= $(or $(patsubst %/,%,$(dir $(patsubst %/,%,$(dir \ 67 $(shell which nvcc 2> /dev/null))))),/usr/local/cuda) 68HIP_DIR ?= /opt/rocm/hip 69 70# Check for PETSc in ../petsc 71ifneq ($(wildcard ../petsc/lib/libpetsc.*),) 72 PETSC_DIR ?= ../petsc 73endif 74 75# Warning: SANTIZ options still don't run with /gpu/occa 76# export LSAN_OPTIONS=suppressions=.asanignore 77AFLAGS = -fsanitize=address #-fsanitize=undefined -fno-omit-frame-pointer 78 79CC_VENDOR := $(patsubst gcc%,gcc,$(firstword $(filter gcc% clang icc XL,$(shell $(CC) --version)))) 80FC_VENDOR := $(firstword $(filter GNU ifort XL,$(shell $(FC) --version 2>&1 || $(FC) -qversion))) 81 82# Default extra flags by vendor 83MARCHFLAG.gcc := -march=native 84MARCHFLAG.clang := $(MARCHFLAG.gcc) 85MARCHFLAG.icc := 86OMP_SIMD_FLAG.gcc := -fopenmp-simd 87OMP_SIMD_FLAG.clang := $(OMP_SIMD_FLAG.gcc) 88OMP_SIMD_FLAG.icc := -qopenmp-simd 89OPT.gcc := -ffp-contract=fast 90OPT.clang := $(OPT.gcc) 91CFLAGS.gcc := -fPIC -std=c99 -Wall -Wextra -Wno-unused-parameter -MMD -MP 92CFLAGS.clang := $(CFLAGS.gcc) 93CFLAGS.icc := $(CFLAGS.gcc) 94CFLAGS.XL := -qpic -MMD 95CXXFLAGS.gcc := -fPIC -std=c++11 -Wall -Wextra -Wno-unused-parameter -MMD -MP 96CXXFLAGS.clang := $(CXXFLAGS.gcc) 97CXXFLAGS.icc := $(CXXFLAGS.gcc) 98CXXFLAGS.XL := -qpic -MMD 99FFLAGS.GNU := -fPIC -cpp -Wall -Wextra -Wno-unused-parameter -Wno-unused-dummy-argument -MMD -MP 100FFLAGS.ifort := -fPIC -cpp 101FFLAGS.XL := -qpic -ffree-form -qpreprocess -qextname -MMD 102 103# This check works with compilers that use gcc and clang. It fails with some 104# compilers; e.g., xlc apparently ignores all options when -E is passed, thus 105# succeeds with any flags. Users can pass MARCHFLAG=... if desired. 106cc_check_flag = $(shell $(CC) -E -Werror $(1) -x c /dev/null > /dev/null 2>&1 && echo 1) 107MARCHFLAG := $(MARCHFLAG.$(CC_VENDOR)) 108MARCHFLAG := $(if $(call cc_check_flag,$(MARCHFLAG)),$(MARCHFLAG),-mcpu=native) 109MARCHFLAG := $(if $(call cc_check_flag,$(MARCHFLAG)),$(MARCHFLAG)) 110 111OMP_SIMD_FLAG := $(OMP_SIMD_FLAG.$(CC_VENDOR)) 112OMP_SIMD_FLAG := $(if $(call cc_check_flag,$(OMP_SIMD_FLAG)),$(OMP_SIMD_FLAG)) 113 114OPT ?= -O -g $(MARCHFLAG) $(OPT.$(CC_VENDOR)) $(OMP_SIMD_FLAG) 115CFLAGS ?= $(OPT) $(CFLAGS.$(CC_VENDOR)) 116CXXFLAGS ?= $(OPT) $(CXXFLAGS.$(CC_VENDOR)) 117NVCCFLAGS ?= -ccbin $(CXX) -Xcompiler "$(OPT)" -Xcompiler -fPIC 118HIPCCFLAGS ?= $(OPT) -fPIC 119FFLAGS ?= $(OPT) $(FFLAGS.$(FC_VENDOR)) 120 121ifeq ($(COVERAGE), 1) 122 CFLAGS += --coverage 123 LDFLAGS += --coverage 124endif 125 126CFLAGS += $(if $(ASAN),$(AFLAGS)) 127FFLAGS += $(if $(ASAN),$(AFLAGS)) 128LDFLAGS += $(if $(ASAN),$(AFLAGS)) 129CPPFLAGS += -I./include 130LDLIBS = -lm 131OBJDIR := build 132LIBDIR := lib 133 134# Installation variables 135prefix ?= /usr/local 136bindir = $(prefix)/bin 137libdir = $(prefix)/lib 138okldir = $(libdir)/okl 139includedir = $(prefix)/include 140pkgconfigdir = $(libdir)/pkgconfig 141INSTALL = install 142INSTALL_PROGRAM = $(INSTALL) 143INSTALL_DATA = $(INSTALL) -m644 144 145# Get number of processors of the machine 146NPROCS := $(shell getconf _NPROCESSORS_ONLN) 147# prepare make options to run in parallel 148MFLAGS := -j $(NPROCS) --warn-undefined-variables \ 149 --no-print-directory --no-keep-going 150 151PYTHON ?= python3 152PROVE ?= prove 153PROVE_OPTS ?= -j $(NPROCS) 154DARWIN := $(filter Darwin,$(shell uname -s)) 155SO_EXT := $(if $(DARWIN),dylib,so) 156 157ceed.pc := $(LIBDIR)/pkgconfig/ceed.pc 158libceed := $(LIBDIR)/libceed.$(SO_EXT) 159CEED_LIBS = -lceed 160libceed.c := $(wildcard interface/ceed*.c) 161gallery.c := $(wildcard gallery/*/ceed*.c) 162libceed.c += $(gallery.c) 163libceed_test := $(LIBDIR)/libceed_test.$(SO_EXT) 164libceeds = $(libceed) $(libceed_test) 165BACKENDS_BUILTIN := /cpu/self/ref/serial /cpu/self/ref/blocked /cpu/self/opt/serial /cpu/self/opt/blocked 166BACKENDS := $(BACKENDS_BUILTIN) 167 168# Tests 169tests.c := $(sort $(wildcard tests/t[0-9][0-9][0-9]-*.c)) 170tests.f := $(sort $(wildcard tests/t[0-9][0-9][0-9]-*.f90)) 171tests := $(tests.c:tests/%.c=$(OBJDIR)/%) 172ctests := $(tests) 173tests += $(tests.f:tests/%.f90=$(OBJDIR)/%) 174# Examples 175examples.c := $(sort $(wildcard examples/ceed/*.c)) 176examples.f := $(sort $(wildcard examples/ceed/*.f)) 177examples := $(examples.c:examples/ceed/%.c=$(OBJDIR)/%) 178examples += $(examples.f:examples/ceed/%.f=$(OBJDIR)/%) 179# MFEM Examples 180mfemexamples.cpp := $(sort $(wildcard examples/mfem/*.cpp)) 181mfemexamples := $(mfemexamples.cpp:examples/mfem/%.cpp=$(OBJDIR)/mfem-%) 182# Nek5K Examples 183nekexamples := $(OBJDIR)/nek-bps 184# PETSc Examples 185petscexamples.c := $(wildcard examples/petsc/*.c) 186petscexamples := $(petscexamples.c:examples/petsc/%.c=$(OBJDIR)/petsc-%) 187# Fluid Dynamics Examples 188fluidsexamples.c := $(sort $(wildcard examples/fluids/*.c)) 189fluidsexamples := $(fluidsexamples.c:examples/fluids/%.c=$(OBJDIR)/fluids-%) 190# Solid Mechanics Examples 191solidsexamples.c := $(sort $(wildcard examples/solids/*.c)) 192solidsexamples := $(solidsexamples.c:examples/solids/%.c=$(OBJDIR)/solids-%) 193 194# Backends/[ref, blocked, template, memcheck, opt, avx, occa, magma] 195ref.c := $(sort $(wildcard backends/ref/*.c)) 196blocked.c := $(sort $(wildcard backends/blocked/*.c)) 197template.c := $(sort $(wildcard backends/template/*.c)) 198ceedmemcheck.c := $(sort $(wildcard backends/memcheck/*.c)) 199opt.c := $(sort $(wildcard backends/opt/*.c)) 200avx.c := $(sort $(wildcard backends/avx/*.c)) 201xsmm.c := $(sort $(wildcard backends/xsmm/*.c)) 202cuda.c := $(sort $(wildcard backends/cuda/*.c)) 203cuda.cpp := $(sort $(wildcard backends/cuda/*.cpp)) 204cuda.cu := $(sort $(wildcard backends/cuda/*.cu)) 205cuda-reg.c := $(sort $(wildcard backends/cuda-reg/*.c)) 206cuda-reg.cu := $(sort $(wildcard backends/cuda-reg/*.cu)) 207cuda-shared.c := $(sort $(wildcard backends/cuda-shared/*.c)) 208cuda-shared.cu := $(sort $(wildcard backends/cuda-shared/*.cu)) 209cuda-gen.c := $(sort $(wildcard backends/cuda-gen/*.c)) 210cuda-gen.cpp := $(sort $(wildcard backends/cuda-gen/*.cpp)) 211cuda-gen.cu := $(sort $(wildcard backends/cuda-gen/*.cu)) 212occa.c := $(sort $(wildcard backends/occa/*.c)) 213magma.c := $(sort $(wildcard backends/magma/*.c)) 214magma.cu := $(sort $(wildcard backends/magma/*.cu)) 215hip.hip := $(sort $(wildcard backends/hip/*.hip.cpp)) 216hip.cpp := $(sort $(filter-out $(hip.hip),$(wildcard backends/hip/*.cpp))) 217hip.c := $(sort $(wildcard backends/hip/*.c)) 218 219# Output using the 216-color rules mode 220rule_file = $(notdir $(1)) 221rule_path = $(patsubst %/,%,$(dir $(1))) 222last_path = $(notdir $(patsubst %/,%,$(dir $(1)))) 223ansicolor = $(shell echo $(call last_path,$(1)) | cksum | cut -b1-2 | xargs -IS expr 2 \* S + 17) 224emacs_out = @printf " %10s %s/%s\n" $(1) $(call rule_path,$(2)) $(call rule_file,$(2)) 225color_out = @if [ -t 1 ]; then \ 226 printf " %10s \033[38;5;%d;1m%s\033[m/%s\n" \ 227 $(1) $(call ansicolor,$(2)) \ 228 $(call rule_path,$(2)) $(call rule_file,$(2)); else \ 229 printf " %10s %s\n" $(1) $(2); fi 230# if TERM=dumb, use it, otherwise switch to the term one 231output = $(if $(TERM:dumb=),$(call color_out,$1,$2),$(call emacs_out,$1,$2)) 232 233# if V is set to non-nil, turn the verbose mode 234quiet = $(if $(V),$($(1)),$(call output,$1,$@);$($(1))) 235 236# Cancel built-in and old-fashioned implicit rules which we don't use 237.SUFFIXES: 238 239.SECONDEXPANSION: # to expand $$(@D)/.DIR 240 241%/.DIR : 242 @mkdir -p $(@D) 243 @touch $@ 244 245.PRECIOUS: %/.DIR 246 247lib: $(libceed) $(ceed.pc) 248# run 'lib' target in parallel 249par:;@$(MAKE) $(MFLAGS) V=$(V) lib 250backend_status = $(if $(filter $1,$(BACKENDS)), [backends: $1], [not found]) 251info: 252 $(info ------------------------------------) 253 $(info CC = $(CC)) 254 $(info CXX = $(CXX)) 255 $(info FC = $(FC)) 256 $(info CPPFLAGS = $(CPPFLAGS)) 257 $(info CFLAGS = $(value CFLAGS)) 258 $(info CXXFLAGS = $(value CXXFLAGS)) 259 $(info FFLAGS = $(value FFLAGS)) 260 $(info NVCCFLAGS = $(value NVCCFLAGS)) 261 $(info LDFLAGS = $(value LDFLAGS)) 262 $(info LDLIBS = $(LDLIBS)) 263 $(info OPT = $(OPT)) 264 $(info AFLAGS = $(AFLAGS)) 265 $(info ASAN = $(or $(ASAN),(empty))) 266 $(info V = $(or $(V),(empty)) [verbose=$(if $(V),on,off)]) 267 $(info ------------------------------------) 268 $(info MEMCHK_STATUS = $(MEMCHK_STATUS)$(call backend_status,/cpu/self/memcheck/serial /cpu/sef/memcheck/blocked)) 269 $(info AVX_STATUS = $(AVX_STATUS)$(call backend_status,/cpu/self/avx/serial /cpu/self/avx/blocked)) 270 $(info XSMM_DIR = $(XSMM_DIR)$(call backend_status,/cpu/self/xsmm/serial /cpu/self/xsmm/blocked)) 271 $(info OCCA_DIR = $(OCCA_DIR)$(call backend_status,/cpu/occa /gpu/occa /omp/occa)) 272 $(info MAGMA_DIR = $(MAGMA_DIR)$(call backend_status,/gpu/magma /gpu/magma/det)) 273 $(info CUDA_DIR = $(CUDA_DIR)$(call backend_status,$(CUDA_BACKENDS))) 274 $(info HIP_DIR = $(HIP_DIR)$(call backend_status,$(HIP_BACKENDS))) 275 $(info ------------------------------------) 276 $(info MFEM_DIR = $(MFEM_DIR)) 277 $(info NEK5K_DIR = $(NEK5K_DIR)) 278 $(info PETSC_DIR = $(PETSC_DIR)) 279 $(info ------------------------------------) 280 $(info prefix = $(prefix)) 281 $(info includedir = $(value includedir)) 282 $(info libdir = $(value libdir)) 283 $(info okldir = $(value okldir)) 284 $(info pkgconfigdir = $(value pkgconfigdir)) 285 $(info ------------------------------------) 286 @true 287info-backends: 288 $(info make: 'lib' with optional backends: $(filter-out $(BACKENDS_BUILTIN),$(BACKENDS))) 289.PHONY: lib all par info info-backends 290 291$(libceed) : LDFLAGS += $(if $(DARWIN), -install_name @rpath/$(notdir $(libceed))) 292$(libceed_test) : LDFLAGS += $(if $(DARWIN), -install_name @rpath/$(notdir $(libceed_test))) 293 294# Standard Backends 295libceed.c += $(ref.c) 296libceed.c += $(blocked.c) 297libceed.c += $(opt.c) 298 299# Testing Backends 300test_backends.c := $(template.c) 301TEST_BACKENDS := /cpu/self/tmpl /cpu/self/tmpl/sub 302 303# Memcheck Backend 304MEMCHK_STATUS = Disabled 305MEMCHK := $(shell echo "\#include <valgrind/memcheck.h>" | $(CC) $(CPPFLAGS) -E - >/dev/null 2>&1 && echo 1) 306ifeq ($(MEMCHK),1) 307 MEMCHK_STATUS = Enabled 308 libceed.c += $(ceedmemcheck.c) 309 BACKENDS += /cpu/self/memcheck/serial /cpu/self/memcheck/blocked 310endif 311 312# AVX Backed 313AVX_STATUS = Disabled 314AVX_FLAG := $(if $(filter clang,$(CC_VENDOR)),+avx,-mavx) 315AVX := $(filter $(AVX_FLAG),$(shell $(CC) $(OPT) -v -E -x c /dev/null 2>&1)) 316ifneq ($(AVX),) 317 AVX_STATUS = Enabled 318 libceed.c += $(avx.c) 319 BACKENDS += /cpu/self/avx/serial /cpu/self/avx/blocked 320endif 321 322# libXSMM Backends 323ifneq ($(wildcard $(XSMM_DIR)/lib/libxsmm.*),) 324 $(libceeds) : LDFLAGS += -L$(XSMM_DIR)/lib -Wl,-rpath,$(abspath $(XSMM_DIR)/lib) 325 $(libceeds) : LDLIBS += -lxsmm -ldl 326 MKL ?= 327 ifeq (,$(MKL)$(MKLROOT)) 328 BLAS_LIB = -lblas 329 else 330 ifneq ($(MKLROOT),) 331 # Some installs put everything inside an intel64 subdirectory, others not 332 MKL_LIBDIR = $(dir $(firstword $(wildcard $(MKLROOT)/lib/intel64/libmkl_sequential.* $(MKLROOT)/lib/libmkl_sequential.*))) 333 MKL_LINK = -L$(MKL_LIBDIR) -Wl,-rpath,$(MKL_LIBDIR) 334 endif 335 BLAS_LIB = $(MKL_LINK) -Wl,--no-as-needed -lmkl_intel_lp64 -lmkl_sequential -lmkl_core -lpthread -lm -ldl 336 endif 337 $(libceeds) : LDLIBS += $(BLAS_LIB) 338 libceed.c += $(xsmm.c) 339 $(xsmm.c:%.c=$(OBJDIR)/%.o) $(xsmm.c:%=%.tidy) : CPPFLAGS += -I$(XSMM_DIR)/include 340 BACKENDS += /cpu/self/xsmm/serial /cpu/self/xsmm/blocked 341endif 342 343# OCCA Backends 344ifneq ($(wildcard $(OCCA_DIR)/lib/libocca.*),) 345 $(libceeds) : LDFLAGS += -L$(OCCA_DIR)/lib -Wl,-rpath,$(abspath $(OCCA_DIR)/lib) 346 $(libceeds) : LDLIBS += -locca 347 libceed.c += $(occa.c) 348 $(occa.c:%.c=$(OBJDIR)/%.o) $(occa.c:%=%.tidy) : CPPFLAGS += -I$(OCCA_DIR)/include 349 BACKENDS += /cpu/occa /gpu/occa /omp/occa 350endif 351 352# CUDA Backends 353CUDA_LIB_DIR := $(wildcard $(foreach d,lib lib64,$(CUDA_DIR)/$d/libcudart.${SO_EXT})) 354CUDA_LIB_DIR := $(patsubst %/,%,$(dir $(firstword $(CUDA_LIB_DIR)))) 355CUDA_LIB_DIR_STUBS := $(CUDA_LIB_DIR)/stubs 356CUDA_BACKENDS = /gpu/cuda/ref /gpu/cuda/reg /gpu/cuda/shared /gpu/cuda/gen 357ifneq ($(CUDA_LIB_DIR),) 358 $(libceeds) : CPPFLAGS += -I$(CUDA_DIR)/include 359 $(libceeds) : LDFLAGS += -L$(CUDA_LIB_DIR) -Wl,-rpath,$(abspath $(CUDA_LIB_DIR)) 360 $(libceeds) : LDLIBS += -lcudart -lnvrtc -lcuda -lcublas 361 $(libceeds) : LINK = $(CXX) 362 libceed.c += $(cuda.c) $(cuda-reg.c) $(cuda-shared.c) $(cuda-gen.c) 363 libceed.cpp += $(cuda.cpp) $(cuda-gen.cpp) 364 libceed.cu += $(cuda.cu) $(cuda-reg.cu) $(cuda-shared.cu) $(cuda-gen.cu) 365 BACKENDS += $(CUDA_BACKENDS) 366endif 367 368# HIP Backends 369ROCM_DIR = $(HIP_DIR)/.. 370HIP_LIB_DIR := $(wildcard $(foreach d,lib lib64,$(HIP_DIR)/$d/libhiprtc.${SO_EXT})) 371HIP_LIB_DIR := $(patsubst %/,%,$(dir $(firstword $(HIP_LIB_DIR)))) 372HIP_BACKENDS = /gpu/hip/ref 373ifneq ($(HIP_LIB_DIR),) 374 $(libceeds) : HIPCCFLAGS += -I./include 375 ifneq ($(CXX), $(HIPCC)) 376 CPPFLAGS += $(subst =,,$(shell $(HIP_DIR)/bin/hipconfig -C)) 377 endif 378 CPPFLAGS += -I$(ROCM_DIR)/hipblas/include -Wno-unused-function 379 $(libceeds) : LDFLAGS += -L$(HIP_LIB_DIR) -Wl,-rpath,$(abspath $(HIP_LIB_DIR)) 380 $(libceeds) : LDFLAGS += -L$(ROCM_DIR)/hipblas/lib -Wl,-rpath,$(abspath $(ROCM_DIR)/hipblas/lib) 381 $(libceeds) : LDLIBS += -lhip_hcc -lhiprtc -lhipblas 382 $(libceeds) : LINK = $(CXX) 383 libceed.hip += $(hip.hip) 384 libceed.cpp += $(hip.cpp) 385 libceed.c += $(hip.c) 386 BACKENDS += $(HIP_BACKENDS) 387endif 388 389# MAGMA Backend 390ifneq ($(wildcard $(MAGMA_DIR)/lib/libmagma.*),) 391 ifneq ($(CUDA_LIB_DIR),) 392 cuda_link = -Wl,-rpath,$(CUDA_LIB_DIR) -L$(CUDA_LIB_DIR) -lcublas -lcusparse -lcudart 393 omp_link = -fopenmp 394 magma_link_static = -L$(MAGMA_DIR)/lib -lmagma $(cuda_link) $(omp_link) 395 magma_link_shared = -L$(MAGMA_DIR)/lib -Wl,-rpath,$(abspath $(MAGMA_DIR)/lib) -lmagma 396 magma_link := $(if $(wildcard $(MAGMA_DIR)/lib/libmagma.${SO_EXT}),$(magma_link_shared),$(magma_link_static)) 397 $(libceeds) : LDLIBS += $(magma_link) 398 $(tests) $(examples) : LDLIBS += $(magma_link) 399 libceed.c += $(magma.c) 400 libceed.cu += $(magma.cu) 401 $(magma.c:%.c=$(OBJDIR)/%.o) $(magma.c:%=%.tidy) : CPPFLAGS += -DADD_ -I$(MAGMA_DIR)/include -I$(CUDA_DIR)/include 402 $(magma.cu:%.cu=$(OBJDIR)/%.o) : CPPFLAGS += --compiler-options=-fPIC -DADD_ -I$(MAGMA_DIR)/include -I$(MAGMA_DIR)/magmablas -I$(MAGMA_DIR)/control -I$(CUDA_DIR)/include 403 BACKENDS += /gpu/magma /gpu/magma/det 404 endif 405endif 406 407export BACKENDS 408 409libceed.o = $(libceed.c:%.c=$(OBJDIR)/%.o) $(libceed.cpp:%.cpp=$(OBJDIR)/%.o) $(libceed.cu:%.cu=$(OBJDIR)/%.o) $(libceed.hip:%.hip.cpp=$(OBJDIR)/%.o) 410$(filter %fortran.o,$(libceed.o)) : CPPFLAGS += $(if $(filter 1,$(UNDERSCORE)),-DUNDERSCORE) 411$(libceed.o): | info-backends 412$(libceed) : $(libceed.o) | $$(@D)/.DIR 413 $(call quiet,LINK) $(LDFLAGS) -shared -o $@ $^ $(LDLIBS) 414 415$(OBJDIR)/%.o : $(CURDIR)/%.c | $$(@D)/.DIR 416 $(call quiet,CC) $(CPPFLAGS) $(CFLAGS) -c -o $@ $(abspath $<) 417 418$(OBJDIR)/%.o : $(CURDIR)/%.cpp | $$(@D)/.DIR 419 $(call quiet,CXX) $(CPPFLAGS) $(CXXFLAGS) -c -o $@ $(abspath $<) 420 421$(OBJDIR)/%.o : $(CURDIR)/%.cu | $$(@D)/.DIR 422 $(call quiet,NVCC) $(CPPFLAGS) $(NVCCFLAGS) -c -o $@ $(abspath $<) 423 424$(OBJDIR)/%.o : $(CURDIR)/%.hip.cpp | $$(@D)/.DIR 425 $(call quiet,HIPCC) $(HIPCCFLAGS) -c -o $@ $(abspath $<) 426 427$(OBJDIR)/% : tests/%.c | $$(@D)/.DIR 428 $(call quiet,LINK.c) $(CEED_LDFLAGS) -o $@ $(abspath $<) $(CEED_LIBS) $(LDLIBS) 429 430$(OBJDIR)/% : tests/%.f90 | $$(@D)/.DIR 431 $(call quiet,LINK.F) -DSOURCE_DIR='"$(abspath $(<D))/"' $(CEED_LDFLAGS) -o $@ $(abspath $<) $(CEED_LIBS) $(LDLIBS) 432 433$(OBJDIR)/% : examples/ceed/%.c | $$(@D)/.DIR 434 $(call quiet,LINK.c) $(CEED_LDFLAGS) -o $@ $(abspath $<) $(CEED_LIBS) $(LDLIBS) 435 436$(OBJDIR)/% : examples/ceed/%.f | $$(@D)/.DIR 437 $(call quiet,LINK.F) -DSOURCE_DIR='"$(abspath $(<D))/"' $(CEED_LDFLAGS) -o $@ $(abspath $<) $(CEED_LIBS) $(LDLIBS) 438 439$(OBJDIR)/mfem-% : examples/mfem/%.cpp $(libceed) | $$(@D)/.DIR 440 +$(MAKE) -C examples/mfem CEED_DIR=`pwd` \ 441 MFEM_DIR="$(abspath $(MFEM_DIR))" CXX=$(CXX) $* 442 mv examples/mfem/$* $@ 443 444# Note: Multiple Nek files cannot be built in parallel. The '+' here enables 445# this single Nek bps file to be built in parallel with other examples, 446# such as when calling `make prove-all -j2`. 447$(OBJDIR)/nek-bps : examples/nek/bps/bps.usr examples/nek/nek-examples.sh $(libceed) | $$(@D)/.DIR 448 +$(MAKE) -C examples MPI=$(MPI) CEED_DIR=`pwd` NEK5K_DIR="$(abspath $(NEK5K_DIR))" nek 449 mv examples/nek/build/bps $(OBJDIR)/bps 450 cp examples/nek/nek-examples.sh $(OBJDIR)/nek-bps 451 452$(OBJDIR)/petsc-% : examples/petsc/%.c $(libceed) $(ceed.pc) | $$(@D)/.DIR 453 +$(MAKE) -C examples/petsc CEED_DIR=`pwd` \ 454 PETSC_DIR="$(abspath $(PETSC_DIR))" OPT="$(OPT)" $* 455 mv examples/petsc/$* $@ 456 457$(OBJDIR)/fluids-% : examples/fluids/%.c $(libceed) $(ceed.pc) | $$(@D)/.DIR 458 +$(MAKE) -C examples/fluids CEED_DIR=`pwd` \ 459 PETSC_DIR="$(abspath $(PETSC_DIR))" OPT="$(OPT)" $* 460 mv examples/fluids/$* $@ 461 462$(OBJDIR)/solids-% : examples/solids/%.c $(libceed) $(ceed.pc) | $$(@D)/.DIR 463 +$(MAKE) -C examples/solids CEED_DIR=`pwd` \ 464 PETSC_DIR="$(abspath $(PETSC_DIR))" OPT="$(OPT)" $* 465 mv examples/solids/$* $@ 466 467libceed_test.o = $(test_backends.c:%.c=$(OBJDIR)/%.o) 468$(libceed_test) : $(libceed.o) $(libceed_test.o) | $$(@D)/.DIR 469 $(call quiet,LINK) $(LDFLAGS) -shared -o $@ $^ $(LDLIBS) 470 471$(examples) : $(libceed) 472$(tests) : $(libceed_test) 473$(tests) : CEED_LIBS = -lceed_test 474$(tests) $(examples) : LDFLAGS += -Wl,-rpath,$(abspath $(LIBDIR)) -L$(LIBDIR) 475 476run-t% : BACKENDS += $(TEST_BACKENDS) 477run-% : $(OBJDIR)/% 478 @tests/tap.sh $(<:$(OBJDIR)/%=%) 479 480external_examples := \ 481 $(if $(MFEM_DIR),$(mfemexamples)) \ 482 $(if $(PETSC_DIR),$(petscexamples)) \ 483 $(if $(NEK5K_DIR),$(nekexamples)) \ 484 $(if $(PETSC_DIR),$(fluidsexamples)) \ 485 $(if $(PETSC_DIR),$(solidsexamples)) 486 487allexamples = $(examples) $(external_examples) 488 489# The test and prove targets can be controlled via pattern searches. The 490# default is to run tests and those examples that have no external dependencies. 491# Examples of finer grained control: 492# 493# make test search='petsc mfem' # PETSc and MFEM examples 494# make prove search='t3' # t3xx series tests 495# make junit search='ex petsc' # core ex* and PETSc tests 496search ?= t ex 497realsearch = $(search:%=%%) 498matched = $(foreach pattern,$(realsearch),$(filter $(OBJDIR)/$(pattern),$(tests) $(allexamples))) 499 500# Test core libCEED 501test : $(matched:$(OBJDIR)/%=run-%) 502 503# Run test target in parallel 504tst : ;@$(MAKE) $(MFLAGS) V=$(V) test 505# CPU C tests only for backend % 506ctc-% : $(ctests);@$(foreach tst,$(ctests),$(tst) /cpu/$*;) 507 508prove : BACKENDS += $(TEST_BACKENDS) 509prove : $(matched) 510 $(info Testing backends: $(BACKENDS)) 511 $(PROVE) $(PROVE_OPTS) --exec 'tests/tap.sh' $(matched:$(OBJDIR)/%=%) 512# Run prove target in parallel 513prv : ;@$(MAKE) $(MFLAGS) V=$(V) prove 514 515prove-all : 516 +$(MAKE) prove realsearch=% 517 518junit-t% : BACKENDS += $(TEST_BACKENDS) 519junit-% : $(OBJDIR)/% 520 @printf " %10s %s\n" TEST $(<:$(OBJDIR)/%=%); $(PYTHON) tests/junit.py $(<:$(OBJDIR)/%=%) 521 522junit : $(matched:$(OBJDIR)/%=junit-%) 523 524all: $(alltests) 525 526examples : $(allexamples) 527ceedexamples : $(examples) 528nekexamples : $(nekexamples) 529mfemexamples : $(mfemexamples) 530petscexamples : $(petscexamples) 531 532# Benchmarks 533allbenchmarks = petsc-bps 534bench_targets = $(addprefix bench-,$(allbenchmarks)) 535.PHONY: $(bench_targets) benchmarks 536$(bench_targets): bench-%: $(OBJDIR)/% 537 cd benchmarks && ./benchmark.sh --ceed "$(BACKENDS)" -r $(*).sh 538benchmarks: $(bench_targets) 539 540$(ceed.pc) : pkgconfig-prefix = $(abspath .) 541$(OBJDIR)/ceed.pc : pkgconfig-prefix = $(prefix) 542.INTERMEDIATE : $(OBJDIR)/ceed.pc 543%/ceed.pc : ceed.pc.template | $$(@D)/.DIR 544 @sed "s:%prefix%:$(pkgconfig-prefix):" $< > $@ 545 546OCCA := $(OCCA_DIR)/bin/occa 547OKL_KERNELS := $(wildcard backends/occa/*.okl) 548 549okl-cache : 550 $(OCCA) cache ceed $(OKL_KERNELS) 551 552okl-clear: 553 $(OCCA) clear -y -l ceed 554 555install : $(libceed) $(OBJDIR)/ceed.pc 556 $(INSTALL) -d $(addprefix $(if $(DESTDIR),"$(DESTDIR)"),"$(includedir)"\ 557 "$(libdir)" "$(pkgconfigdir)" $(if $(OCCA_ON),"$(okldir)")) 558 $(INSTALL_DATA) include/ceed.h "$(DESTDIR)$(includedir)/" 559 $(INSTALL_DATA) include/ceedf.h "$(DESTDIR)$(includedir)/" 560 $(INSTALL_DATA) $(libceed) "$(DESTDIR)$(libdir)/" 561 $(INSTALL_DATA) $(OBJDIR)/ceed.pc "$(DESTDIR)$(pkgconfigdir)/" 562 $(if $(OCCA_ON),$(INSTALL_DATA) $(OKL_KERNELS) "$(DESTDIR)$(okldir)/") 563 564.PHONY : cln clean doxygen doc lib install all print test tst prove prv prove-all junit examples style style-c style-py tidy okl-cache okl-clear info info-backends 565 566cln clean : 567 $(RM) -r $(OBJDIR) $(LIBDIR) dist *egg* .pytest_cache *cffi* 568 $(MAKE) -C examples clean NEK5K_DIR="$(abspath $(NEK5K_DIR))" 569 $(MAKE) -C tests/python clean 570 $(RM) benchmarks/*output.txt 571 572distclean : clean 573 $(RM) -r doc/html doc/sphinx/build $(CONFIG) 574 575DOXYGEN ?= doxygen 576doxygen : 577 $(DOXYGEN) Doxyfile 578 579doc-html doc-latexpdf doc-epub : doc-% : doxygen 580 make -C doc/sphinx $* 581 582doc : doc-html 583 584style-c : 585 @astyle --options=.astylerc \ 586 $(filter-out include/ceedf.h tests/t320-basis-f.h, \ 587 $(wildcard include/*.h interface/*.[ch] tests/*.[ch] backends/*/*.[ch] \ 588 examples/*/*/*.[ch] examples/*/*.[ch] examples/*/*.[ch]pp gallery/*/*.[ch])) 589 590AUTOPEP8 = autopep8 591style-py : AUTOPEP8_ARGS = --in-place --aggressive 592style-py : 593 @$(AUTOPEP8) $(AUTOPEP8_ARGS) $(wildcard *.py python**/*.py tests/python**/*.py examples**/*.py doc/sphinx/source**/*.py benchmarks/*.py) 594 595style : style-c style-py 596 597CLANG_TIDY ?= clang-tidy 598%.c.tidy : %.c 599 $(CLANG_TIDY) $(TIDY_OPTS) $^ -- $(CPPFLAGS) --std=c99 -I$(CUDA_DIR)/include 600 601tidy : $(libceed.c:%=%.tidy) 602 603print : 604 @echo $(VAR)=$($(VAR)) 605 606print-% : 607 $(info [ variable name]: $*) 608 $(info [ origin]: $(origin $*)) 609 $(info [ flavor]: $(flavor $*)) 610 $(info [ value]: $(value $*)) 611 $(info [expanded value]: $($*)) 612 $(info ) 613 @true 614 615# "make configure" detects any variables passed on the command line or 616# previously set in config.mk, caching them in config.mk as simple 617# (:=) variables. Variables set in config.mk or on the command line 618# take precedence over the defaults provided in the file. Typical 619# usage: 620# 621# make configure CC=/path/to/my/cc CUDA_DIR=/opt/cuda 622# make 623# make prove 624# 625# The values in the file can be updated by passing them on the command 626# line, e.g., 627# 628# make configure CC=/path/to/other/clang 629 630# All variables to consider for caching 631CONFIG_VARS = CC CXX FC NVCC NVCC_CXX HIPCC \ 632 OPT CFLAGS CPPFLAGS CXXFLAGS FFLAGS NVCCFLAGS HIPCCFLAGS \ 633 LDFLAGS LDLIBS \ 634 MAGMA_DIR XSMM_DIR CUDA_DIR MFEM_DIR PETSC_DIR NEK5K_DIR HIP_DIR 635 636# $(call needs_save,CFLAGS) returns true (a nonempty string) if CFLAGS 637# was set on the command line or in config.mk (where it will appear as 638# a simple variable). 639needs_save = $(or $(filter command line,$(origin $(1))),$(filter simple,$(flavor $(1)))) 640 641configure : 642 $(file > $(CONFIG)) 643 $(foreach v,$(CONFIG_VARS),$(if $(call needs_save,$(v)),$(file >> $(CONFIG),$(v) := $($(v))))) 644 @echo "Configuration cached in $(CONFIG):" 645 @cat $(CONFIG) 646 647wheel : export MARCHFLAG = -march=generic 648wheel : export WHEEL_PLAT = manylinux2010_x86_64 649wheel : 650 docker run -it --user $(shell id -u):$(shell id -g) --rm -v $(PWD):/io -w /io \ 651 -e MARCHFLAG -e WHEEL_PLAT \ 652 quay.io/pypa/$(WHEEL_PLAT) python/make-wheels.sh 653 654.PHONY : configure wheel 655 656# Include *.d deps when not -B = --always-make: useful if the paths are wonky in a container 657-include $(if $(filter B,$(MAKEFLAGS)),,$(libceed.c:%.c=$(OBJDIR)/%.d) $(tests.c:tests/%.c=$(OBJDIR)/%.d)) 658