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) 19COMMON ?= common.mk 20-include $(COMMON) 21 22ifeq (,$(filter-out undefined default,$(origin CC))) 23 CC = gcc 24endif 25ifeq (,$(filter-out undefined default,$(origin CXX))) 26 CXX = g++ 27endif 28ifeq (,$(filter-out undefined default,$(origin FC))) 29 FC = gfortran 30endif 31ifeq (,$(filter-out undefined default,$(origin LINK))) 32 LINK = $(CC) 33endif 34NVCC ?= $(CUDA_DIR)/bin/nvcc 35NVCC_CXX ?= $(CXX) 36HIPCC ?= $(HIP_DIR)/bin/hipcc 37 38# ASAN must be left empty if you don't want to use it 39ASAN ?= 40 41LDFLAGS ?= 42UNDERSCORE ?= 1 43 44# MFEM_DIR env variable should point to sibling directory 45ifneq ($(wildcard ../mfem/libmfem.*),) 46 MFEM_DIR ?= ../mfem 47endif 48 49# NEK5K_DIR env variable should point to sibling directory 50ifneq ($(wildcard ../Nek5000/*),) 51 NEK5K_DIR ?= $(abspath ../Nek5000) 52endif 53export NEK5K_DIR 54MPI ?= 1 55 56# CEED_DIR env for NEK5K testing 57export CEED_DIR = $(abspath .) 58 59# XSMM_DIR env variable should point to XSMM master (github.com/hfp/libxsmm) 60XSMM_DIR ?= ../libxsmm 61 62# OCCA_DIR env variable should point to OCCA master (github.com/libocca/occa) 63OCCA_DIR ?= ../occa 64 65# env variable MAGMA_DIR can be used too 66MAGMA_DIR ?= ../magma 67# If CUDA_DIR is not set, check for nvcc, or resort to /usr/local/cuda 68CUDA_DIR ?= $(or $(patsubst %/,%,$(dir $(patsubst %/,%,$(dir \ 69 $(shell which nvcc 2> /dev/null))))),/usr/local/cuda) 70HIP_DIR ?= /opt/rocm 71 72# Check for PETSc in ../petsc 73ifneq ($(wildcard ../petsc/lib/libpetsc.*),) 74 PETSC_DIR ?= ../petsc 75endif 76 77# Warning: SANTIZ options still don't run with /gpu/occa 78# export LSAN_OPTIONS=suppressions=.asanignore 79AFLAGS = -fsanitize=address #-fsanitize=undefined -fno-omit-frame-pointer 80 81CC_VENDOR := $(patsubst gcc%,gcc,$(firstword $(filter gcc% clang icc XL,$(shell $(CC) --version)))) 82FC_VENDOR := $(firstword $(filter GNU ifort XL,$(shell $(FC) --version 2>&1 || $(FC) -qversion))) 83 84# Default extra flags by vendor 85MARCHFLAG.gcc := -march=native 86MARCHFLAG.clang := $(MARCHFLAG.gcc) 87MARCHFLAG.icc := 88OMP_SIMD_FLAG.gcc := -fopenmp-simd 89OMP_SIMD_FLAG.clang := $(OMP_SIMD_FLAG.gcc) 90OMP_SIMD_FLAG.icc := -qopenmp-simd 91OPT.gcc := -ffp-contract=fast 92OPT.clang := $(OPT.gcc) 93CFLAGS.gcc := -fPIC -std=c99 -Wall -Wextra -Wno-unused-parameter -MMD -MP 94CFLAGS.clang := $(CFLAGS.gcc) 95CFLAGS.icc := $(CFLAGS.gcc) 96CFLAGS.XL := -qpic -MMD 97CXXFLAGS.gcc := -fPIC -std=c++11 -Wall -Wextra -Wno-unused-parameter -MMD -MP 98CXXFLAGS.clang := $(CXXFLAGS.gcc) 99CXXFLAGS.icc := $(CXXFLAGS.gcc) 100CXXFLAGS.XL := -qpic -MMD 101FFLAGS.GNU := -fPIC -cpp -Wall -Wextra -Wno-unused-parameter -Wno-unused-dummy-argument -MMD -MP 102FFLAGS.ifort := -fPIC -cpp 103FFLAGS.XL := -qpic -ffree-form -qpreprocess -qextname -MMD 104 105# This check works with compilers that use gcc and clang. It fails with some 106# compilers; e.g., xlc apparently ignores all options when -E is passed, thus 107# succeeds with any flags. Users can pass MARCHFLAG=... if desired. 108cc_check_flag = $(shell $(CC) -E -Werror $(1) -x c /dev/null > /dev/null 2>&1 && echo 1) 109MARCHFLAG := $(MARCHFLAG.$(CC_VENDOR)) 110MARCHFLAG := $(if $(call cc_check_flag,$(MARCHFLAG)),$(MARCHFLAG),-mcpu=native) 111MARCHFLAG := $(if $(call cc_check_flag,$(MARCHFLAG)),$(MARCHFLAG)) 112 113OMP_SIMD_FLAG := $(OMP_SIMD_FLAG.$(CC_VENDOR)) 114OMP_SIMD_FLAG := $(if $(call cc_check_flag,$(OMP_SIMD_FLAG)),$(OMP_SIMD_FLAG)) 115 116OPT ?= -O -g $(MARCHFLAG) $(OPT.$(CC_VENDOR)) $(OMP_SIMD_FLAG) 117CFLAGS ?= $(OPT) $(CFLAGS.$(CC_VENDOR)) 118CXXFLAGS ?= $(OPT) $(CXXFLAGS.$(CC_VENDOR)) 119NVCCFLAGS ?= -ccbin $(CXX) -Xcompiler "$(OPT)" -Xcompiler -fPIC 120HIPCCFLAGS ?= $(OPT) -fPIC 121FFLAGS ?= $(OPT) $(FFLAGS.$(FC_VENDOR)) 122 123ifeq ($(COVERAGE), 1) 124 CFLAGS += --coverage 125 LDFLAGS += --coverage 126endif 127 128CFLAGS += $(if $(ASAN),$(AFLAGS)) 129FFLAGS += $(if $(ASAN),$(AFLAGS)) 130LDFLAGS += $(if $(ASAN),$(AFLAGS)) 131CPPFLAGS += -I./include 132LDLIBS = -lm 133OBJDIR := build 134LIBDIR := lib 135 136# Installation variables 137prefix ?= /usr/local 138bindir = $(prefix)/bin 139libdir = $(prefix)/lib 140includedir = $(prefix)/include 141pkgconfigdir = $(libdir)/pkgconfig 142INSTALL = install 143INSTALL_PROGRAM = $(INSTALL) 144INSTALL_DATA = $(INSTALL) -m644 145 146# Get number of processors of the machine 147NPROCS := $(shell getconf _NPROCESSORS_ONLN) 148# prepare make options to run in parallel 149MFLAGS := -j $(NPROCS) --warn-undefined-variables \ 150 --no-print-directory --no-keep-going 151 152PYTHON ?= python3 153PROVE ?= prove 154PROVE_OPTS ?= -j $(NPROCS) 155DARWIN := $(filter Darwin,$(shell uname -s)) 156SO_EXT := $(if $(DARWIN),dylib,so) 157 158ceed.pc := $(LIBDIR)/pkgconfig/ceed.pc 159libceed := $(LIBDIR)/libceed.$(SO_EXT) 160CEED_LIBS = -lceed 161libceed.c := $(wildcard interface/ceed*.c) 162gallery.c := $(wildcard gallery/*/ceed*.c) 163libceed.c += $(gallery.c) 164libceed_test := $(LIBDIR)/libceed_test.$(SO_EXT) 165libceeds = $(libceed) $(libceed_test) 166BACKENDS_BUILTIN := /cpu/self/ref/serial /cpu/self/ref/blocked /cpu/self/opt/serial /cpu/self/opt/blocked 167BACKENDS := $(BACKENDS_BUILTIN) 168 169# Tests 170tests.c := $(sort $(wildcard tests/t[0-9][0-9][0-9]-*.c)) 171tests.f := $(sort $(wildcard tests/t[0-9][0-9][0-9]-*.f90)) 172tests := $(tests.c:tests/%.c=$(OBJDIR)/%) 173ctests := $(tests) 174tests += $(tests.f:tests/%.f90=$(OBJDIR)/%) 175# Examples 176examples.c := $(sort $(wildcard examples/ceed/*.c)) 177examples.f := $(sort $(wildcard examples/ceed/*.f)) 178examples := $(examples.c:examples/ceed/%.c=$(OBJDIR)/%) 179examples += $(examples.f:examples/ceed/%.f=$(OBJDIR)/%) 180# MFEM Examples 181mfemexamples.cpp := $(sort $(wildcard examples/mfem/*.cpp)) 182mfemexamples := $(mfemexamples.cpp:examples/mfem/%.cpp=$(OBJDIR)/mfem-%) 183# Nek5K Examples 184nekexamples := $(OBJDIR)/nek-bps 185# PETSc Examples 186petscexamples.c := $(wildcard examples/petsc/*.c) 187petscexamples := $(petscexamples.c:examples/petsc/%.c=$(OBJDIR)/petsc-%) 188# Fluid Dynamics Examples 189fluidsexamples.c := $(sort $(wildcard examples/fluids/*.c)) 190fluidsexamples := $(fluidsexamples.c:examples/fluids/%.c=$(OBJDIR)/fluids-%) 191# Solid Mechanics Examples 192solidsexamples.c := $(sort $(wildcard examples/solids/*.c)) 193solidsexamples := $(solidsexamples.c:examples/solids/%.c=$(OBJDIR)/solids-%) 194 195# Backends/[ref, blocked, template, memcheck, opt, avx, occa, magma] 196ref.c := $(sort $(wildcard backends/ref/*.c)) 197blocked.c := $(sort $(wildcard backends/blocked/*.c)) 198template.c := $(sort $(wildcard backends/template/*.c)) 199ceedmemcheck.c := $(sort $(wildcard backends/memcheck/*.c)) 200opt.c := $(sort $(wildcard backends/opt/*.c)) 201avx.c := $(sort $(wildcard backends/avx/*.c)) 202xsmm.c := $(sort $(wildcard backends/xsmm/*.c)) 203cuda.c := $(sort $(wildcard backends/cuda/*.c)) 204cuda.cpp := $(sort $(wildcard backends/cuda/*.cpp)) 205cuda.cu := $(sort $(wildcard backends/cuda/kernels/*.cu)) 206cuda-shared.c := $(sort $(wildcard backends/cuda-shared/*.c)) 207cuda-shared.cu := $(sort $(wildcard backends/cuda-shared/kernels/*.cu)) 208cuda-gen.c := $(sort $(wildcard backends/cuda-gen/*.c)) 209cuda-gen.cpp := $(sort $(wildcard backends/cuda-gen/*.cpp)) 210cuda-gen.cu := $(sort $(wildcard backends/cuda-gen/kernels/*.cu)) 211occa.cpp := $(sort $(shell find backends/occa -type f -name *.cpp)) 212magma.c := $(sort $(wildcard backends/magma/*.c)) 213magma.cu := $(sort $(wildcard backends/magma/kernels/cuda/*.cu)) 214magma.hip := $(sort $(wildcard backends/magma/kernels/hip/*.hip.cpp)) 215hip.c := $(sort $(wildcard backends/hip/*.c)) 216hip.cpp := $(sort $(wildcard backends/hip/*.cpp)) 217hip.hip := $(sort $(wildcard backends/hip/kernels/*.hip.cpp)) 218 219# Quiet, color output 220quiet ?= $($(1)) 221 222# Cancel built-in and old-fashioned implicit rules which we don't use 223.SUFFIXES: 224 225.SECONDEXPANSION: # to expand $$(@D)/.DIR 226 227%/.DIR : 228 @mkdir -p $(@D) 229 @touch $@ 230 231.PRECIOUS: %/.DIR 232 233lib: $(libceed) $(ceed.pc) 234# run 'lib' target in parallel 235par:;@$(MAKE) $(MFLAGS) V=$(V) lib 236backend_status = $(if $(filter $1,$(BACKENDS)), [backends: $1], [not found]) 237info: 238 $(info ------------------------------------) 239 $(info CC = $(CC)) 240 $(info CXX = $(CXX)) 241 $(info FC = $(FC)) 242 $(info CPPFLAGS = $(CPPFLAGS)) 243 $(info CFLAGS = $(value CFLAGS)) 244 $(info CXXFLAGS = $(value CXXFLAGS)) 245 $(info FFLAGS = $(value FFLAGS)) 246 $(info NVCCFLAGS = $(value NVCCFLAGS)) 247 $(info HIPCCFLAGS = $(value HIPCCFLAGS)) 248 $(info LDFLAGS = $(value LDFLAGS)) 249 $(info LDLIBS = $(LDLIBS)) 250 $(info OPT = $(OPT)) 251 $(info AFLAGS = $(AFLAGS)) 252 $(info ASAN = $(or $(ASAN),(empty))) 253 $(info V = $(or $(V),(empty)) [verbose=$(if $(V),on,off)]) 254 $(info ------------------------------------) 255 $(info MEMCHK_STATUS = $(MEMCHK_STATUS)$(call backend_status,/cpu/self/memcheck/serial /cpu/sef/memcheck/blocked)) 256 $(info AVX_STATUS = $(AVX_STATUS)$(call backend_status,/cpu/self/avx/serial /cpu/self/avx/blocked)) 257 $(info XSMM_DIR = $(XSMM_DIR)$(call backend_status,/cpu/self/xsmm/serial /cpu/self/xsmm/blocked)) 258 $(info OCCA_DIR = $(OCCA_DIR)$(call backend_status,$(OCCA_BACKENDS))) 259 $(info MAGMA_DIR = $(MAGMA_DIR)$(call backend_status,/gpu/magma /gpu/magma/det)) 260 $(info CUDA_DIR = $(CUDA_DIR)$(call backend_status,$(CUDA_BACKENDS))) 261 $(info HIP_DIR = $(HIP_DIR)$(call backend_status,$(HIP_BACKENDS))) 262 $(info ------------------------------------) 263 $(info MFEM_DIR = $(MFEM_DIR)) 264 $(info NEK5K_DIR = $(NEK5K_DIR)) 265 $(info PETSC_DIR = $(PETSC_DIR)) 266 $(info ------------------------------------) 267 $(info prefix = $(prefix)) 268 $(info includedir = $(value includedir)) 269 $(info libdir = $(value libdir)) 270 $(info pkgconfigdir = $(value pkgconfigdir)) 271 $(info ------------------------------------) 272 @true 273info-backends: 274 $(info make: 'lib' with optional backends: $(filter-out $(BACKENDS_BUILTIN),$(BACKENDS))) 275.PHONY: lib all par info info-backends 276 277$(libceed) : LDFLAGS += $(if $(DARWIN), -install_name @rpath/$(notdir $(libceed))) 278$(libceed_test) : LDFLAGS += $(if $(DARWIN), -install_name @rpath/$(notdir $(libceed_test))) 279 280# Standard Backends 281libceed.c += $(ref.c) 282libceed.c += $(blocked.c) 283libceed.c += $(opt.c) 284 285# Testing Backends 286test_backends.c := $(template.c) 287TEST_BACKENDS := /cpu/self/tmpl /cpu/self/tmpl/sub 288 289# Memcheck Backend 290MEMCHK_STATUS = Disabled 291MEMCHK := $(shell echo "\#include <valgrind/memcheck.h>" | $(CC) $(CPPFLAGS) -E - >/dev/null 2>&1 && echo 1) 292ifeq ($(MEMCHK),1) 293 MEMCHK_STATUS = Enabled 294 libceed.c += $(ceedmemcheck.c) 295 BACKENDS += /cpu/self/memcheck/serial /cpu/self/memcheck/blocked 296endif 297 298# AVX Backed 299AVX_STATUS = Disabled 300AVX_FLAG := $(if $(filter clang,$(CC_VENDOR)),+avx,-mavx) 301AVX := $(filter $(AVX_FLAG),$(shell $(CC) $(OPT) -v -E -x c /dev/null 2>&1)) 302ifneq ($(AVX),) 303 AVX_STATUS = Enabled 304 libceed.c += $(avx.c) 305 BACKENDS += /cpu/self/avx/serial /cpu/self/avx/blocked 306endif 307 308# libXSMM Backends 309ifneq ($(wildcard $(XSMM_DIR)/lib/libxsmm.*),) 310 $(libceeds) : LDFLAGS += -L$(XSMM_DIR)/lib -Wl,-rpath,$(abspath $(XSMM_DIR)/lib) 311 $(libceeds) : LDLIBS += -lxsmm -ldl 312 MKL ?= 313 ifeq (,$(MKL)$(MKLROOT)) 314 BLAS_LIB = -lblas 315 else 316 ifneq ($(MKLROOT),) 317 # Some installs put everything inside an intel64 subdirectory, others not 318 MKL_LIBDIR = $(dir $(firstword $(wildcard $(MKLROOT)/lib/intel64/libmkl_sequential.* $(MKLROOT)/lib/libmkl_sequential.*))) 319 MKL_LINK = -L$(MKL_LIBDIR) -Wl,-rpath,$(MKL_LIBDIR) 320 endif 321 BLAS_LIB = $(MKL_LINK) -Wl,--no-as-needed -lmkl_intel_lp64 -lmkl_sequential -lmkl_core -lpthread -lm -ldl 322 endif 323 $(libceeds) : LDLIBS += $(BLAS_LIB) 324 libceed.c += $(xsmm.c) 325 $(xsmm.c:%.c=$(OBJDIR)/%.o) $(xsmm.c:%=%.tidy) : CPPFLAGS += -I$(XSMM_DIR)/include 326 BACKENDS += /cpu/self/xsmm/serial /cpu/self/xsmm/blocked 327endif 328 329# OCCA Backends 330OCCA_BACKENDS = /cpu/occa/serial 331ifneq ($(wildcard $(OCCA_DIR)/lib/libocca.*),) 332 OCCA_MODES := $(shell $(OCCA_DIR)/bin/occa modes) 333 OCCA_BACKENDS += $(if $(filter OpenMP,$(OCCA_MODES)),/cpu/occa/openmp) 334# OCCA_BACKENDS += $(if $(filter OpenCL,$(OCCA_MODES)),/opencl/occa) 335 OCCA_BACKENDS += $(if $(filter HIP,$(OCCA_MODES)),/gpu/occa/hip) 336 OCCA_BACKENDS += $(if $(filter CUDA,$(OCCA_MODES)),/gpu/occa/cuda) 337 338 $(libceeds) : CPPFLAGS += -I$(OCCA_DIR)/include 339 $(libceeds) : LDFLAGS += -L$(OCCA_DIR)/lib -Wl,-rpath,$(abspath $(OCCA_DIR)/lib) 340 $(libceeds) : LDLIBS += -locca 341 $(libceeds) : LINK = $(CXX) 342 libceed.cpp += $(occa.cpp) 343 BACKENDS += $(OCCA_BACKENDS) 344endif 345 346# CUDA Backends 347CUDA_LIB_DIR := $(wildcard $(foreach d,lib lib64,$(CUDA_DIR)/$d/libcudart.${SO_EXT})) 348CUDA_LIB_DIR := $(patsubst %/,%,$(dir $(firstword $(CUDA_LIB_DIR)))) 349CUDA_LIB_DIR_STUBS := $(CUDA_LIB_DIR)/stubs 350CUDA_BACKENDS = /gpu/cuda/ref /gpu/cuda/shared /gpu/cuda/gen 351ifneq ($(CUDA_LIB_DIR),) 352 $(libceeds) : CPPFLAGS += -I$(CUDA_DIR)/include 353 $(libceeds) : LDFLAGS += -L$(CUDA_LIB_DIR) -Wl,-rpath,$(abspath $(CUDA_LIB_DIR)) 354 $(libceeds) : LDLIBS += -lcudart -lnvrtc -lcuda -lcublas 355 $(libceeds) : LINK = $(CXX) 356 libceed.c += $(cuda.c) $(cuda-shared.c) $(cuda-gen.c) 357 libceed.cpp += $(cuda.cpp) $(cuda-gen.cpp) 358 libceed.cu += $(cuda.cu) $(cuda-shared.cu) $(cuda-gen.cu) 359 BACKENDS += $(CUDA_BACKENDS) 360endif 361 362# HIP Backends 363HIP_LIB_DIR := $(wildcard $(foreach d,lib lib64,$(HIP_DIR)/$d/libamdhip64.${SO_EXT})) 364HIP_LIB_DIR := $(patsubst %/,%,$(dir $(firstword $(HIP_LIB_DIR)))) 365HIP_BACKENDS = /gpu/hip/ref 366ifneq ($(HIP_LIB_DIR),) 367 $(libceeds) : HIPCCFLAGS += -I./include 368 ifneq ($(CXX), $(HIPCC)) 369 CPPFLAGS += $(subst =,,$(shell $(HIP_DIR)/bin/hipconfig -C)) 370 endif 371 CPPFLAGS += -I$(HIP_DIR)/include -Wno-unused-function 372 $(libceeds) : LDFLAGS += -L$(HIP_LIB_DIR) -Wl,-rpath,$(abspath $(HIP_LIB_DIR)) 373 $(libceeds) : LDLIBS += -lamdhip64 -lhipblas 374 $(libceeds) : LINK = $(CXX) 375 libceed.hip += $(hip.hip) 376 libceed.cpp += $(hip.cpp) 377 libceed.c += $(hip.c) 378 BACKENDS += $(HIP_BACKENDS) 379endif 380 381# MAGMA Backend 382ifneq ($(wildcard $(MAGMA_DIR)/lib/libmagma.*),) 383 MAGMA_ARCH=$(shell nm -g $(MAGMA_DIR)/lib/libmagma.* | grep -c "hipblas") 384 ifeq ($(MAGMA_ARCH), 0) #CUDA MAGMA 385 ifneq ($(CUDA_LIB_DIR),) 386 cuda_link = -Wl,-rpath,$(CUDA_LIB_DIR) -L$(CUDA_LIB_DIR) -lcublas -lcusparse -lcudart 387 omp_link = -fopenmp 388 magma_link_static = -L$(MAGMA_DIR)/lib -lmagma $(cuda_link) $(omp_link) 389 magma_link_shared = -L$(MAGMA_DIR)/lib -Wl,-rpath,$(abspath $(MAGMA_DIR)/lib) -lmagma 390 magma_link := $(if $(wildcard $(MAGMA_DIR)/lib/libmagma.${SO_EXT}),$(magma_link_shared),$(magma_link_static)) 391 $(libceeds) : LDLIBS += $(magma_link) 392 $(tests) $(examples) : LDLIBS += $(magma_link) 393 libceed.c += $(magma.c) 394 libceed.cu += $(magma.cu) 395 $(magma.c:%.c=$(OBJDIR)/%.o) $(magma.c:%=%.tidy) : CPPFLAGS += -DADD_ -I$(MAGMA_DIR)/include -I$(CUDA_DIR)/include 396 $(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 397 BACKENDS += /gpu/magma /gpu/magma/det 398 endif 399 else # HIP MAGMA 400 ifneq ($(HIP_LIB_DIR),) 401 hip_link = -Wl,-rpath,$(HIP_LIB_DIR) -L$(HIP_LIB_DIR) -lhipblas -lhipsparse -lamdhip64 402 omp_link = -fopenmp 403 magma_link_static = -L$(MAGMA_DIR)/lib -lmagma $(hip_link) $(omp_link) 404 magma_link_shared = -L$(MAGMA_DIR)/lib -Wl,-rpath,$(abspath $(MAGMA_DIR)/lib) -lmagma 405 magma_link := $(if $(wildcard $(MAGMA_DIR)/lib/libmagma.${SO_EXT}),$(magma_link_shared),$(magma_link_static)) 406 $(libceeds) : LDLIBS += $(magma_link) 407 $(tests) $(examples) : LDLIBS += $(magma_link) 408 libceed.c += $(magma.c) 409 libceed.hip += $(magma.hip) 410 ifneq ($(CXX), $(HIPCC)) 411 $(magma.c:%.c=$(OBJDIR)/%.o) $(magma.c:%=%.tidy) : CPPFLAGS += -I$(MAGMA_DIR)/include -I$(HIP_DIR)/include -DHAVE_HIP -DADD_ 412 else 413 $(magma.c:%.c=$(OBJDIR)/%.o) $(magma.c:%=%.tidy) : HIPCCFLAGS += -I$(MAGMA_DIR)/include -I$(HIP_DIR)/include -DHAVE_HIP -DADD_ 414 endif 415 $(magma.hip:%.hip.cpp=$(OBJDIR)/%.o) : HIPCCFLAGS += -I$(MAGMA_DIR)/include -I$(MAGMA_DIR)/magmablas -I$(MAGMA_DIR)/control -I$(HIP_DIR)/include -DHAVE_HIP -DADD_ 416 BACKENDS += /gpu/magma 417 endif 418 endif 419endif 420 421export BACKENDS 422 423libceed.o = $(libceed.c:%.c=$(OBJDIR)/%.o) $(libceed.cpp:%.cpp=$(OBJDIR)/%.o) $(libceed.cu:%.cu=$(OBJDIR)/%.o) $(libceed.hip:%.hip.cpp=$(OBJDIR)/%.o) 424$(filter %fortran.o,$(libceed.o)) : CPPFLAGS += $(if $(filter 1,$(UNDERSCORE)),-DUNDERSCORE) 425$(libceed.o): | info-backends 426$(libceed) : $(libceed.o) | $$(@D)/.DIR 427 $(call quiet,LINK) $(LDFLAGS) -shared -o $@ $^ $(LDLIBS) 428 429$(OBJDIR)/%.o : $(CURDIR)/%.c | $$(@D)/.DIR 430 $(call quiet,CC) $(CPPFLAGS) $(CFLAGS) -c -o $@ $(abspath $<) 431 432$(OBJDIR)/%.o : $(CURDIR)/%.cpp | $$(@D)/.DIR 433 $(call quiet,CXX) $(CPPFLAGS) $(CXXFLAGS) -c -o $@ $(abspath $<) 434 435$(OBJDIR)/%.o : $(CURDIR)/%.cu | $$(@D)/.DIR 436 $(call quiet,NVCC) $(filter-out -Wno-unused-function, $(CPPFLAGS)) $(NVCCFLAGS) -c -o $@ $(abspath $<) 437 438$(OBJDIR)/%.o : $(CURDIR)/%.hip.cpp | $$(@D)/.DIR 439 $(call quiet,HIPCC) $(HIPCCFLAGS) -c -o $@ $(abspath $<) 440 441$(OBJDIR)/% : tests/%.c | $$(@D)/.DIR 442 $(call quiet,LINK.c) $(CEED_LDFLAGS) -o $@ $(abspath $<) $(CEED_LIBS) $(LDLIBS) 443 444$(OBJDIR)/% : tests/%.f90 | $$(@D)/.DIR 445 $(call quiet,LINK.F) -DSOURCE_DIR='"$(abspath $(<D))/"' $(CEED_LDFLAGS) -o $@ $(abspath $<) $(CEED_LIBS) $(LDLIBS) 446 447$(OBJDIR)/% : examples/ceed/%.c | $$(@D)/.DIR 448 $(call quiet,LINK.c) $(CEED_LDFLAGS) -o $@ $(abspath $<) $(CEED_LIBS) $(LDLIBS) 449 450$(OBJDIR)/% : examples/ceed/%.f | $$(@D)/.DIR 451 $(call quiet,LINK.F) -DSOURCE_DIR='"$(abspath $(<D))/"' $(CEED_LDFLAGS) -o $@ $(abspath $<) $(CEED_LIBS) $(LDLIBS) 452 453$(OBJDIR)/mfem-% : examples/mfem/%.cpp $(libceed) | $$(@D)/.DIR 454 +$(MAKE) -C examples/mfem CEED_DIR=`pwd` \ 455 MFEM_DIR="$(abspath $(MFEM_DIR))" CXX=$(CXX) $* 456 mv examples/mfem/$* $@ 457 458# Note: Multiple Nek files cannot be built in parallel. The '+' here enables 459# this single Nek bps file to be built in parallel with other examples, 460# such as when calling `make prove-all -j2`. 461$(OBJDIR)/nek-bps : examples/nek/bps/bps.usr examples/nek/nek-examples.sh $(libceed) | $$(@D)/.DIR 462 +$(MAKE) -C examples MPI=$(MPI) CEED_DIR=`pwd` NEK5K_DIR="$(abspath $(NEK5K_DIR))" nek 463 mv examples/nek/build/bps $(OBJDIR)/bps 464 cp examples/nek/nek-examples.sh $(OBJDIR)/nek-bps 465 466$(OBJDIR)/petsc-% : examples/petsc/%.c $(libceed) $(ceed.pc) | $$(@D)/.DIR 467 +$(MAKE) -C examples/petsc CEED_DIR=`pwd` \ 468 PETSC_DIR="$(abspath $(PETSC_DIR))" OPT="$(OPT)" $* 469 mv examples/petsc/$* $@ 470 471$(OBJDIR)/fluids-% : examples/fluids/%.c $(libceed) $(ceed.pc) | $$(@D)/.DIR 472 +$(MAKE) -C examples/fluids CEED_DIR=`pwd` \ 473 PETSC_DIR="$(abspath $(PETSC_DIR))" OPT="$(OPT)" $* 474 mv examples/fluids/$* $@ 475 476$(OBJDIR)/solids-% : examples/solids/%.c $(libceed) $(ceed.pc) | $$(@D)/.DIR 477 +$(MAKE) -C examples/solids CEED_DIR=`pwd` \ 478 PETSC_DIR="$(abspath $(PETSC_DIR))" OPT="$(OPT)" $* 479 mv examples/solids/$* $@ 480 481libceed_test.o = $(test_backends.c:%.c=$(OBJDIR)/%.o) 482$(libceed_test) : $(libceed.o) $(libceed_test.o) | $$(@D)/.DIR 483 $(call quiet,LINK) $(LDFLAGS) -shared -o $@ $^ $(LDLIBS) 484 485$(examples) : $(libceed) 486$(tests) : $(libceed_test) 487$(tests) : CEED_LIBS = -lceed_test 488$(tests) $(examples) : LDFLAGS += -Wl,-rpath,$(abspath $(LIBDIR)) -L$(LIBDIR) 489 490run-t% : BACKENDS += $(TEST_BACKENDS) 491run-% : $(OBJDIR)/% 492 @tests/tap.sh $(<:$(OBJDIR)/%=%) 493 494external_examples := \ 495 $(if $(MFEM_DIR),$(mfemexamples)) \ 496 $(if $(PETSC_DIR),$(petscexamples)) \ 497 $(if $(NEK5K_DIR),$(nekexamples)) \ 498 $(if $(PETSC_DIR),$(fluidsexamples)) \ 499 $(if $(PETSC_DIR),$(solidsexamples)) 500 501allexamples = $(examples) $(external_examples) 502 503# The test and prove targets can be controlled via pattern searches. The 504# default is to run tests and those examples that have no external dependencies. 505# Examples of finer grained control: 506# 507# make test search='petsc mfem' # PETSc and MFEM examples 508# make prove search='t3' # t3xx series tests 509# make junit search='ex petsc' # core ex* and PETSc tests 510search ?= t ex 511realsearch = $(search:%=%%) 512matched = $(foreach pattern,$(realsearch),$(filter $(OBJDIR)/$(pattern),$(tests) $(allexamples))) 513 514# Test core libCEED 515test : $(matched:$(OBJDIR)/%=run-%) 516 517# Run test target in parallel 518tst : ;@$(MAKE) $(MFLAGS) V=$(V) test 519# CPU C tests only for backend % 520ctc-% : $(ctests);@$(foreach tst,$(ctests),$(tst) /cpu/$*;) 521 522prove : BACKENDS += $(TEST_BACKENDS) 523prove : $(matched) 524 $(info Testing backends: $(BACKENDS)) 525 $(PROVE) $(PROVE_OPTS) --exec 'tests/tap.sh' $(matched:$(OBJDIR)/%=%) 526# Run prove target in parallel 527prv : ;@$(MAKE) $(MFLAGS) V=$(V) prove 528 529prove-all : 530 +$(MAKE) prove realsearch=% 531 532junit-t% : BACKENDS += $(TEST_BACKENDS) 533junit-% : $(OBJDIR)/% 534 @printf " %10s %s\n" TEST $(<:$(OBJDIR)/%=%); $(PYTHON) tests/junit.py $(<:$(OBJDIR)/%=%) 535 536junit : $(matched:$(OBJDIR)/%=junit-%) 537 538all: $(alltests) 539 540examples : $(allexamples) 541ceedexamples : $(examples) 542nekexamples : $(nekexamples) 543mfemexamples : $(mfemexamples) 544petscexamples : $(petscexamples) 545 546# Benchmarks 547allbenchmarks = petsc-bps 548bench_targets = $(addprefix bench-,$(allbenchmarks)) 549.PHONY: $(bench_targets) benchmarks 550$(bench_targets): bench-%: $(OBJDIR)/% 551 cd benchmarks && ./benchmark.sh --ceed "$(BACKENDS)" -r $(*).sh 552benchmarks: $(bench_targets) 553 554$(ceed.pc) : pkgconfig-prefix = $(abspath .) 555$(OBJDIR)/ceed.pc : pkgconfig-prefix = $(prefix) 556.INTERMEDIATE : $(OBJDIR)/ceed.pc 557%/ceed.pc : ceed.pc.template | $$(@D)/.DIR 558 @sed "s:%prefix%:$(pkgconfig-prefix):" $< > $@ 559 560install : $(libceed) $(OBJDIR)/ceed.pc 561 $(INSTALL) -d $(addprefix $(if $(DESTDIR),"$(DESTDIR)"),"$(includedir)"\ 562 "$(libdir)" "$(pkgconfigdir)") 563 $(INSTALL_DATA) include/ceed.h "$(DESTDIR)$(includedir)/" 564 $(INSTALL_DATA) include/ceedf.h "$(DESTDIR)$(includedir)/" 565 $(INSTALL_DATA) $(libceed) "$(DESTDIR)$(libdir)/" 566 $(INSTALL_DATA) $(OBJDIR)/ceed.pc "$(DESTDIR)$(pkgconfigdir)/" 567 568.PHONY : cln clean doxygen doc lib install all print test tst prove prv prove-all junit examples style style-c style-py tidy info info-backends 569 570cln clean : 571 $(RM) -r $(OBJDIR) $(LIBDIR) dist *egg* .pytest_cache *cffi* 572 $(MAKE) -C examples clean NEK5K_DIR="$(abspath $(NEK5K_DIR))" 573 $(MAKE) -C tests/python clean 574 $(RM) benchmarks/*output.txt 575 576distclean : clean 577 $(RM) -r doc/html doc/sphinx/build $(CONFIG) 578 579DOXYGEN ?= doxygen 580doxygen : 581 $(DOXYGEN) Doxyfile 582 583doc-html doc-latexpdf doc-epub : doc-% : doxygen 584 make -C doc/sphinx $* 585 586doc : doc-html 587 588style-c : 589 @astyle --options=.astylerc \ 590 $(filter-out include/ceedf.h $(wildcard tests/t*-f.h), \ 591 $(wildcard include/*.h interface/*.[ch] tests/*.[ch] backends/*/*.[ch] \ 592 examples/*/*/*.[ch] examples/*/*.[ch] examples/*/*.[ch]pp gallery/*/*.[ch])) 593 594AUTOPEP8 = autopep8 595style-py : AUTOPEP8_ARGS = --in-place --aggressive 596style-py : 597 @$(AUTOPEP8) $(AUTOPEP8_ARGS) $(wildcard *.py python**/*.py tests/python**/*.py examples**/*.py doc/sphinx/source**/*.py benchmarks/*.py) 598 599style : style-c style-py 600 601CLANG_TIDY ?= clang-tidy 602 603%.c.tidy : %.c 604 $(CLANG_TIDY) $(TIDY_OPTS) $^ -- $(CPPFLAGS) --std=c99 -I$(CUDA_DIR)/include 605 606%.cpp.tidy : %.cpp 607 $(CLANG_TIDY) $(TIDY_OPTS) $^ -- $(CPPFLAGS) --std=c++11 -I$(CUDA_DIR)/include -I$(OCCA_DIR)/include 608 609tidy_c : $(libceed.c:%=%.tidy) 610tidy_cpp : $(libceed.cpp:%=%.tidy) 611 612tidy : tidy_c tidy_cpp 613 614print : 615 @echo $(VAR)=$($(VAR)) 616 617print-% : 618 $(info [ variable name]: $*) 619 $(info [ origin]: $(origin $*)) 620 $(info [ flavor]: $(flavor $*)) 621 $(info [ value]: $(value $*)) 622 $(info [expanded value]: $($*)) 623 $(info ) 624 @true 625 626# "make configure" detects any variables passed on the command line or 627# previously set in config.mk, caching them in config.mk as simple 628# (:=) variables. Variables set in config.mk or on the command line 629# take precedence over the defaults provided in the file. Typical 630# usage: 631# 632# make configure CC=/path/to/my/cc CUDA_DIR=/opt/cuda 633# make 634# make prove 635# 636# The values in the file can be updated by passing them on the command 637# line, e.g., 638# 639# make configure CC=/path/to/other/clang 640 641# All variables to consider for caching 642CONFIG_VARS = CC CXX FC NVCC NVCC_CXX HIPCC \ 643 OPT CFLAGS CPPFLAGS CXXFLAGS FFLAGS NVCCFLAGS HIPCCFLAGS \ 644 LDFLAGS LDLIBS \ 645 MAGMA_DIR XSMM_DIR CUDA_DIR MFEM_DIR PETSC_DIR NEK5K_DIR HIP_DIR 646 647# $(call needs_save,CFLAGS) returns true (a nonempty string) if CFLAGS 648# was set on the command line or in config.mk (where it will appear as 649# a simple variable). 650needs_save = $(or $(filter command line,$(origin $(1))),$(filter simple,$(flavor $(1)))) 651 652configure : 653 $(file > $(CONFIG)) 654 $(foreach v,$(CONFIG_VARS),$(if $(call needs_save,$(v)),$(file >> $(CONFIG),$(v) := $($(v))))) 655 @echo "Configuration cached in $(CONFIG):" 656 @cat $(CONFIG) 657 658wheel : export MARCHFLAG = -march=generic 659wheel : export WHEEL_PLAT = manylinux2010_x86_64 660wheel : 661 docker run -it --user $(shell id -u):$(shell id -g) --rm -v $(PWD):/io -w /io \ 662 -e MARCHFLAG -e WHEEL_PLAT \ 663 quay.io/pypa/$(WHEEL_PLAT) python/make-wheels.sh 664 665.PHONY : configure wheel 666 667# Include *.d deps when not -B = --always-make: useful if the paths are wonky in a container 668-include $(if $(filter B,$(MAKEFLAGS)),,$(libceed.c:%.c=$(OBJDIR)/%.d) $(tests.c:tests/%.c=$(OBJDIR)/%.d)) 669