Makefile.rule

###################################################################################################
#                                                                                                 #
# This file is part of BLASFEO.                                                                   #
#                                                                                                 #
# BLASFEO -- BLAS for embedded optimization.                                                      #
# Copyright (C) 2019 by Gianluca Frison.                                                          #
# Developed at IMTEK (University of Freiburg) under the supervision of Moritz Diehl.              #
# All rights reserved.                                                                            #
#                                                                                                 #
# The 2-Clause BSD License                                                                        #
#                                                                                                 #
# Redistribution and use in source and binary forms, with or without                              #
# modification, are permitted provided that the following conditions are met:                     #
#                                                                                                 #
# 1. Redistributions of source code must retain the above copyright notice, this                  #
#    list of conditions and the following disclaimer.                                             #
# 2. Redistributions in binary form must reproduce the above copyright notice,                    #
#    this list of conditions and the following disclaimer in the documentation                    #
#    and/or other materials provided with the distribution.                                       #
#                                                                                                 #
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND                 #
# ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED                   #
# WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE                          #
# DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR                 #
# ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES                  #
# (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;                    #
# LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND                     #
# ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT                      #
# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS                   #
# SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.                                    #
#                                                                                                 #
# Author: Gianluca Frison, gianluca.frison (at) imtek.uni-freiburg.de                             #
#                                                                                                 #
###################################################################################################

# Do something in this makefile
$(info Parsing Makefile.rule)

# Get path of Makefile.rule as main project directory 
#CURRENT_DIR  := $(dir $(lastword $(MAKEFILE_LIST)))
MAKEFILE_PATH := $(abspath $(lastword $(MAKEFILE_LIST)))
CURRENT_DIR := $(patsubst %/,%,$(dir $(MAKEFILE_PATH)))



#################################################
### main makefile options
#################################################

# Select target architecture (TARGET)
#
# X64_INTEL_SKYLAKE_X: x86_64 architecture with AVX512 (F+VL) ISA (64 bit OS)
# Code optimized for Intel Skylake-X and following, AMD Zen 5 architectures with 2 512-bit FMA pipes.
#
# X64_INTEL_HASWELL: x86_64 architecture with AVX2 and FMA ISAs (64 bit OS)
# Code optimized for Intel Haswell, Intel Skylake and following, AMD Zen, AMD Zen2, AMD Zen3 and AMD Zen4 architectures.
#
# X64_INTEL_SANDY_BRIDGE : x86_64 architecture with AVX ISA (64 bit OS)
# Code optimized for Intel Sandy-Bridge architecture.
#
# X64_INTEL_CORE : x86_64 architecture with SSE3 ISA (64 bit OS)
# Code optimized for Intel Core archiecture.
#
# X64_AMD_BULLDOZER : x86_64 architecture with AVX and FMA ISAs (64 bit OS)
# Code optimized for AMD Bulldozer.
#
# X86_AMD_JAGUAR : x86 architecture with AVX ISA (32 bit OS)
# Code optimized for AMD Jaguar.
#
# X86_AMD_BARCELONA : x86 architecture with SSE3 ISA (32 bit OS)
# Code optimized for AMD Barcelona.
#
# ARMV8A_APPLE_M1 : ARMv8A architecture with NEON (64 bit OS)
# Code optimized for Apple M1, M2.
#
# ARMV8A_ARM_CORTEX_A76 : ARMv8A architecture with NEON (64 bit OS)
# Code optimized for ARM Cortex A76.
#
# ARMV8A_ARM_CORTEX_A73 : ARMv8A architecture with NEON (64 bit OS)
# Code optimized for ARM Cortex A73.
#
# ARMV8A_ARM_CORTEX_A57 : ARMv8A architecture with NEON (64 bit OS)
# Code optimized for ARM Cortex A57, A72.
#
# ARMV8A_ARM_CORTEX_A55 : ARMv8A architecture with NEON (64 bit OS)
# Code optimized for ARM Cortex A55.
#
# ARMV8A_ARM_CORTEX_A53 : ARMv8A architecture with NEON (64 bit OS)
# Code optimized for ARM Cortex A53.
#
# ARMV7A_ARM_CORTEX_A15 : ARMv7A architecture with NEON-VFPv4 ISA (32 bit OS)
# Code optimized for ARM Cortex A15.
#
# ARMV7A_ARM_CORTEX_A9 : ARMv7A architecture with NEON-VFPv3 ISA (32 bit OS)
# Code optimized for ARM Cortex A9.
#
# ARMV7A_ARM_CORTEX_A7 : ARMv7A architecture with NEON-VFPv4 ISA (32 bit OS)
# Code optimized for ARM Cortex A7.
#
# GENERIC : generic architecture, plain C code running on virtually every target but not specialized to any.
# Choose one of the above optimized targets for better performance on specific architectures.
#
#
# TARGET = X64_INTEL_SKYLAKE_X
TARGET = X64_INTEL_HASWELL
# TARGET = X64_INTEL_SANDY_BRIDGE
# TARGET = X64_INTEL_CORE
#
# TARGET = X64_AMD_BULLDOZER
# TARGET = X86_AMD_JAGUAR
# TARGET = X86_AMD_BARCELONA
#
# TARGET = ARMV8A_APPLE_M1
#
# TARGET = ARMV8A_ARM_CORTEX_A76
# TARGET = ARMV8A_ARM_CORTEX_A73
# TARGET = ARMV8A_ARM_CORTEX_A57
# TARGET = ARMV8A_ARM_CORTEX_A55
# TARGET = ARMV8A_ARM_CORTEX_A53
# TARGET = ARMV7A_ARM_CORTEX_A15
# TARGET = ARMV7A_ARM_CORTEX_A9
# TARGET = ARMV7A_ARM_CORTEX_A7
#
# TARGET = GENERIC

# Select back-end linear lagebra version (LA) to implement BLASFEO API:
# HIGH_PERFORMANCE : target-tailored; performance-optimized for cache resident matrices; panel-major matrix format
# REFERENCE : target-unspecific lightly-optimized; small code footprint; {panel,column}-major matrix format(s)
# EXTERNAL_BLAS_WRAPPER : call to external BLAS and LAPACK libraries; column-major matrix format
#
LA = HIGH_PERFORMANCE
# LA = REFERENCE
# LA = EXTERNAL_BLAS_WRAPPER

# Export additional back-ends with different naming
# BLASFEO reference with blasfeo_ref_xxx naming
# BLASFEO_REF_API = 0
BLASFEO_REF_API = 1
# BLASFEO high-performance with blasfeo_hp_xxx naming
BLASFEO_HP_API = 0
# BLASFEO_HP_API = 1

# Select the Matrix Format internally used in the blasfeo_{d,s}mat structure
# COLMAJ : column-major matrix format
# PANELMAJ : panel-major matrix format
# MF = COLMAJ
MF = PANELMAJ

# Compile the BLAS API routines provided by BLASFEO
#
# BLAS_API = 0
BLAS_API = 1

# Export standard FORTRAN namings for BLAS API routines
# 0 : routines namings are in the form blasfeo_blas_dgemm and blasfeo_lapack_dpotrf
# 1 : routines namings are in the form dgemm_ and dpotrf_
# and export standard C namings for CBLAS API routines
# 0 : routines namings are in the form blasfeo_cblas_dgemm
# 1 : routines namings are in the form cblas_dgemm
#
FORTRAN_BLAS_API = 0
# FORTRAN_BLAS_API = 1

# Complement the BLAS_API with the Netlib BLAS (only for FORTAN_BLAS_API=1)
COMPLEMENT_WITH_NETLIB_BLAS = 0
# COMPLEMENT_WITH_NETLIB_BLAS = 1

# Complement the BLAS_API with the Netlib LAPACK (only for FORTAN_BLAS_API=1)
COMPLEMENT_WITH_NETLIB_LAPACK = 0
# COMPLEMENT_WITH_NETLIB_LAPACK = 1

# Compile the CBLAS API routines provided by BLASFEO
#
CBLAS_API = 0
# CBLAS_API = 1

#Compile the LAPACKE API from Netlib
LAPACKE_API = 0
# LAPACKE_API = 1

# Select external BLAS and LAPACK implementation (to be provided by the user).
# Edit Makefile.external_blas to specify installation location (default /opt).
# It is used by the BLASFEO library if LA=EXTERNAL_BLAS_WRAPPER.
# It may also be used as a comparison in some benchmarks and tests.
#
EXTERNAL_BLAS = 0
# EXTERNAL_BLAS = SYSTEM
# EXTERNAL_BLAS = OPENBLAS
# EXTERNAL_BLAS = NETLIB
# EXTERNAL_BLAS = MKL
# EXTERNAL_BLAS = BLIS
# EXTERNAL_BLAS = ATLAS
# EXTERNAL_BLAS = ARMPL
# EXTERNAL_BLAS = ACCELERATE
include $(CURRENT_DIR)/Makefile.external_blas

# Select operating system (automatic selection for LINUX and MAC)
#
UNAME_S := $(shell uname -s)
ifeq ($(UNAME_S), Linux)
    OS = LINUX
endif
ifeq ($(UNAME_S), Darwin)
    OS = MAC
endif
#
# Select operating system (manual selection)
#
# OS = LINUX
# OS = MAC
# OS = WINDOWS



#################################################
### other makefile options and settings
#################################################

# In BLAS API, fallback to external BLAS library for some not-yet-implemented routines
#
FALLBACK_TO_EXTERNAL_BLAS = 0
# FALLBACK_TO_EXTERNAL_BLAS = 1

# Maximum inner product length K for buffer allocation on stack (decrease this value if stack size is exceeded)
#
K_MAX_STACK = 300

# Macro level (code size vs performance in assembly kernels):
# 0 : no macro (min code size)
# 1 : all macro but gemm kernel (sweet spot)
# 2 : all macro (max performance)
#
# MACRO_LEVEL = 2
MACRO_LEVEL = 1
# MACRO_LEVEL = 0

# Use C99 extension to math library
#
# USE_C99_MATH = 0
USE_C99_MATH = 1

# Compile auxiliary functions with external dependencies (for memory allocation, printing and timing)
#
# EXT_DEP = 0
EXT_DEP = 1

# Enables the compilation of sandbox (experimental)
#
SANDBOX_MODE = 0
# SANDBOX_MODE = 1

# Enables the compilation of experimental routines
#
EXPERIMENTAL = 0
# EXPERIMENTAL = 1

# Enable on-line checks for matrix and vector dimensions (experimental)
#
RUNTIME_CHECKS = 0
# RUNTIME_CHECKS = 1

# Print name of BLAS API routines when called (for debugging purposes)
#
PRINT_NAME = 0
# PRINT_NAME = 1

# Select the packing algorithm (limited to colmaj dgemm and sgemm ATM)
# AUTO   : automatic switching between packing algorithms
# ALG_0  : no packing
# ALG_M1 : pack A
# ALG_N1 : pack B
# ALG_2  : pack A and B
#
PACKING_ALG = AUTO
# PACKING_ALG = ALG_0
# PACKING_ALG = ALG_M1
# PACKING_ALG = ALG_N1
# PACKING_ALG = ALG_2

# C Compiler
#
CC ?= gcc
# CC = clang
# CC = x86_64-w64-mingw32-gcc
# CC = arm-linux-gnueabihf-gcc
# CC = /opt/ndk/bin/aarch64-linux-android-gcc
# CC = /home/gianluca/android-ndk-r23b/toolchains/llvm/prebuilt/linux-x86_64/bin/aarch64-linux-android31-clang

# archive routine
#
AR = ar

# Installation directory
#
PREFIX = /opt

# compiler / assembler / linker flags
#
# CFLAGS  =
ASFLAGS =
LDFLAGS =

# Common optimization flags
#
CFLAGS ?= -O2
CFLAGS += -fPIC
#CFLAGS += -Wuninitialized
#CFLAGS += -Wjump-misses-init

# Debugging flags
#
CFLAGS  += #-g #-Wall -pedantic -Wfloat-equal #-pg
ASFLAGS += #-g

# Profiling flags
#
#CFLAGS += --coverage
#CFLAGS += -fopenmp



# Installation directory
TOP = $(CURRENT_DIR)

# Support local options
# TODO move somewhere else ???
-include $(CURRENT_DIR)/Makefile.local

# search directories
CFLAGS += -I$(TOP)/include



# Conditional definitions and checks

# LA high performance
ifeq ($(LA), HIGH_PERFORMANCE)
CFLAGS  += -DLA_HIGH_PERFORMANCE
ASFLAGS  += -DLA_HIGH_PERFORMANCE
BINARY_DIR = build/$(LA)/$(TARGET)
endif
# LA reference
ifeq ($(LA), REFERENCE)
CFLAGS  += -DLA_REFERENCE
BINARY_DIR = build/$(LA)/$(TARGET)
endif
# LA blas wrapper
ifeq ($(LA), EXTERNAL_BLAS_WRAPPER)
ifeq ($(EXTERNAL_BLAS), 0)
$(error No EXTERNAL_BLAS selected for LA=EXTERNAL_BLAS_WRAPPER)
endif
ifeq ($(MF), PANELMAJ)
$(error As of now, with LA=EXTERNAL_BLAS_WRAPPER only MF=COLMAJ is supported)
endif
CFLAGS  += -DLA_EXTERNAL_BLAS_WRAPPER
BINARY_DIR = build/$(LA)/$(EXTERNAL_BLAS)
endif
# TODO remove and fix tests
# CFLAGS  += -DBLASFEO_LA=$(LA)

ifeq ($(BLASFEO_REF_API), 1)
CFLAGS += -DBLASFEO_REF_API
endif

ifeq ($(MF), COLMAJ)
CFLAGS  += -DMF_COLMAJ
ASFLAGS  += -DMF_COLMAJ
endif
ifeq ($(MF), PANELMAJ)
CFLAGS  += -DMF_PANELMAJ
ASFLAGS  += -DMF_PANELMAJ
endif

ifeq ($(BLAS_API), 1)
CFLAGS += -DBLAS_API
ASFLAGS += -DBLAS_API
ifeq ($(FORTRAN_BLAS_API), 1)
CFLAGS += -DFORTRAN_BLAS_API
ASFLAGS +=
endif # FORTRAN_BLAS_API
ifeq ($(FALLBACK_TO_EXTERNAL_BLAS), 1)
CFLAGS += -DFALLBACK_TO_EXTERNAL_BLAS
ASFLAGS += -DFALLBACK_TO_EXTERNAL_BLAS
endif # FALLBACK_TO_EXTERNAL_BLAS
endif # BLAS_API

#ifeq ($(CBLAS_API), 1)
#ifeq ($(FORTRAN_BLAS_API), 0)
#$(error Cannot expose non-FORTRAN style BLAS_API when building CBLAS_API)
#endif
#endif

ifeq ($(LAPACKE_API), 1) 
ifeq ($(FORTRAN_BLAS_API), 0)
$(error Cannot expose non-FORTRAN style BLAS_API when building LAPACKE_API)
endif
endif

ifeq ($(COMPLEMENT_WITH_NETLIB_BLAS), 1)
ifeq ($(FORTRAN_BLAS_API), 0)
$(error Cannot expose non-FORTRAN style BLAS_API when complementing with Netlib BLAS)
endif
endif

ifeq ($(COMPLEMENT_WITH_NETLIB_LAPACK), 1)
ifeq ($(FORTRAN_BLAS_API), 0)
$(error Cannot expose non-FORTRAN style BLAS_API when complementing with Netlib LAPACK)
endif
endif

STACK_SIZE := $(shell ulimit -s)
ifneq ($(STACK_SIZE), unlimited)
STACK_SIZE_EXCEEDED := $(shell echo $(K_MAX_STACK)*12*8*2 \> $(STACK_SIZE)*1024 | bc )
ifeq ($(STACK_SIZE_EXCEEDED), 1)
$(error stack size likely to be exceeded, please decrease the value of K_MAX_STACK )
endif
endif
CFLAGS  += -DK_MAX_STACK=$(K_MAX_STACK)

ifneq ($(PACKING_ALG), AUTO)
CFLAGS  += -DPACKING_$(PACKING_ALG)
endif

ifeq ($(USE_C99_MATH), 1)
CFLAGS += -DUSE_C99_MATH
endif

ifeq ($(RUNTIME_CHECKS), 1)
CFLAGS += -DDIM_CHECK
endif

ifeq ($(EXT_DEP), 1)
CFLAGS += -DEXT_DEP
endif

ifeq ($(SANDBOX_MODE), 1)
CFLAGS += -DSANDBOX_MODE
endif

ifeq ($(MACRO_LEVEL), 1)
ASFLAGS += -DMACRO_LEVEL=1
endif
ifeq ($(MACRO_LEVEL), 2)
ASFLAGS += -DMACRO_LEVEL=2
endif

ifeq ($(PRINT_NAME), 1)
CFLAGS += -DPRINT_NAME
endif

ifeq ($(OS), LINUX)
CFLAGS  += -DOS_LINUX
ASFLAGS += -DOS_LINUX
endif
ifeq ($(OS), MAC)
CFLAGS  += -DOS_MAC
ASFLAGS += -DOS_MAC
endif
ifeq ($(OS), WINDOWS)
CFLAGS  += -DOS_WINDOWS
ASFLAGS += -DOS_WINDOWS
endif
ifeq ($(SOC), DSPACE)
CFLAGS  += -D__DSPACE__
ASFLAGS += -D__DSPACE__
endif
ifeq ($(SOC), BACHMANN)
CFLAGS  += -D__BACHMANN__
ASFLAGS += -D__BACHMANN__
endif

# EXTERNAL_BLAS

ifndef EXTERNAL_BLAS
    EXTERNAL_BLAS = 0
endif

CFLAGS += $(INCLUDE_EXTERNAL_BLAS)
ifeq ($(EXTERNAL_BLAS), 0)
CFLAGS  += -DEXTERNAL_BLAS_NONE
endif
ifeq ($(EXTERNAL_BLAS), SYSTEM)
CFLAGS  += -DEXTERNAL_BLAS_SYSTEM
endif
ifeq ($(EXTERNAL_BLAS), OPENBLAS)
CFLAGS  += -DEXTERNAL_BLAS_OPENBLAS
endif
ifeq ($(EXTERNAL_BLAS), BLIS)
CFLAGS  += -DEXTERNAL_BLAS_BLIS -std=gnu99
endif
ifeq ($(EXTERNAL_BLAS), NETLIB)
CFLAGS  += -DEXTERNAL_BLAS_NETLIB
endif
ifeq ($(EXTERNAL_BLAS), MKL)
CFLAGS  += -DEXTERNAL_BLAS_MKL -std=c99 -m64 -DMKL_DIRECT_CALL_SEQ
endif
ifeq ($(EXTERNAL_BLAS), ATLAS)
CFLAGS  += -DEXTERNAL_BLAS_ATLAS
endif
ifeq ($(EXTERNAL_BLAS), ARMPL)
CFLAGS  += -DEXTERNAL_BLAS_ARMPL
endif
ifeq ($(EXTERNAL_BLAS), ACCELERATE)
CFLAGS  += -DEXTERNAL_BLAS_ACCELERATE
endif
# TODO remove and fix tests
# CFLAGS  += -DEXTERNAL_BLAS=$(EXTERNAL_BLAS)

# Architecture-specific flags
ifeq ($(TARGET), X64_INTEL_SKYLAKE_X)
CFLAGS  += -m64 -mavx512f -mavx512vl -mfma -DTARGET_X64_INTEL_SKYLAKE_X
endif
ifeq ($(TARGET), X64_INTEL_HASWELL)
CFLAGS  += -m64 -mavx2 -mfma -DTARGET_X64_INTEL_HASWELL
endif
ifeq ($(TARGET), X64_INTEL_SANDY_BRIDGE)
CFLAGS  += -m64 -mavx -DTARGET_X64_INTEL_SANDY_BRIDGE
endif
ifeq ($(TARGET), X64_INTEL_CORE)
CFLAGS  += -m64 -msse3 -DTARGET_X64_INTEL_CORE
endif
ifeq ($(TARGET), X64_AMD_BULLDOZER)
CFLAGS  += -m64 -mavx -mfma -DTARGET_X64_AMD_BULLDOZER
endif
ifeq ($(TARGET), X86_AMD_JAGUAR)
CFLAGS  += -m32 -mavx -DTARGET_X86_AMD_JAGUAR
ASFLAGS += -m32 -mavx -DTARGET_X86_AMD_JAGUAR
endif
ifeq ($(TARGET), X86_AMD_BARCELONA)
CFLAGS  += -m32 -msse3 -DTARGET_X86_AMD_BARCELONA
ASFLAGS += -m32 -msse3 -DTARGET_X86_AMD_BARCELONA
endif
ifeq ($(TARGET), ARMV8A_APPLE_M1)
CFLAGS  += -march=armv8-a+crc+crypto+simd -DTARGET_ARMV8A_APPLE_M1 -DTARGET_ARMV8A_ARM_CORTEX_A57
ASFLAGS += -DTARGET_ARMV8A_APPLE_M1 -DTARGET_ARMV8A_ARM_CORTEX_A57
endif
ifeq ($(TARGET), ARMV8A_ARM_CORTEX_A76)
CFLAGS  += -march=armv8-a+crc+crypto+simd -DTARGET_ARMV8A_ARM_CORTEX_A76 -DTARGET_ARMV8A_ARM_CORTEX_A57
ASFLAGS += -DTARGET_ARMV8A_ARM_CORTEX_A76 -DTARGET_ARMV8A_ARM_CORTEX_A57
endif
ifeq ($(TARGET), ARMV8A_ARM_CORTEX_A73)
CFLAGS  += -march=armv8-a+crc+crypto+simd -DTARGET_ARMV8A_ARM_CORTEX_A73 -DTARGET_ARMV8A_ARM_CORTEX_A57
ASFLAGS += -DTARGET_ARMV8A_ARM_CORTEX_A73 -DTARGET_ARMV8A_ARM_CORTEX_A57
endif
ifeq ($(TARGET), ARMV8A_ARM_CORTEX_A57)
CFLAGS  += -march=armv8-a+crc+crypto+simd -DTARGET_ARMV8A_ARM_CORTEX_A57
ASFLAGS += -DTARGET_ARMV8A_ARM_CORTEX_A57
endif
ifeq ($(TARGET), ARMV8A_ARM_CORTEX_A55)
CFLAGS  += -march=armv8-a+crc+crypto+simd -DTARGET_ARMV8A_ARM_CORTEX_A55 -DTARGET_ARMV8A_ARM_CORTEX_A53
ASFLAGS += -DTARGET_ARMV8A_ARM_CORTEX_A55 -DTARGET_ARMV8A_ARM_CORTEX_A53
endif
ifeq ($(TARGET), ARMV8A_ARM_CORTEX_A53)
CFLAGS  += -march=armv8-a+crc+crypto+simd -DTARGET_ARMV8A_ARM_CORTEX_A53
ASFLAGS += -DTARGET_ARMV8A_ARM_CORTEX_A53
endif
ifeq ($(TARGET), ARMV7A_ARM_CORTEX_A15)
CFLAGS  += -marm -mfloat-abi=hard -mfpu=neon-vfpv4 -DTARGET_ARMV7A_ARM_CORTEX_A15 # -mcpu=cortex-a15
ASFLAGS += -mfpu=neon-vfpv4 -DTARGET_ARMV7A_ARM_CORTEX_A15
endif
ifeq ($(TARGET), ARMV7A_ARM_CORTEX_A9)
CFLAGS  += -marm -mfloat-abi=hard -mfpu=neon -DTARGET_ARMV7A_ARM_CORTEX_A9 # -mcpu=cortex-a9
ASFLAGS += -mfpu=neon -DTARGET_ARMV7A_ARM_CORTEX_A9
endif
ifeq ($(TARGET), ARMV7A_ARM_CORTEX_A7)
CFLAGS  += -marm -mfloat-abi=hard -mfpu=neon-vfpv4 -DTARGET_ARMV7A_ARM_CORTEX_A7 # -mcpu=cortex-a7
ASFLAGS += -mfpu=neon-vfpv4 -DTARGET_ARMV7A_ARM_CORTEX_A7
endif
ifeq ($(TARGET), GENERIC)
CFLAGS  += -DTARGET_GENERIC
endif
# TODO remove and fix tests
# CFLAGS  += -DBLASFEO_TARGET=$(TARGET)