2 # Beginning of user configuration
5 # This library's version
8 # If you set the suffix, the library name will be libopenblas_$(LIBNAMESUFFIX).a
9 # and libopenblas_$(LIBNAMESUFFIX).so. Meanwhile, the soname in shared library
10 # is libopenblas_$(LIBNAMESUFFIX).so.0.
13 # You can specify the target architecture, otherwise it's
14 # automatically detected.
17 # If you want to support multiple architecture in one binary
20 # If you want the full list of x86_64 architectures supported in DYNAMIC_ARCH
21 # mode (including individual optimizied codes for PENRYN, DUNNINGTON, OPTERON,
22 # OPTERON_SSE3, ATOM and NANO rather than fallbacks to older architectures)
25 # C compiler including binary type(32bit / 64bit). Default is gcc.
26 # Don't use Intel Compiler or PGI, it won't generate right codes as I expect.
29 # Fortran compiler. Default is g77.
32 # Even you can specify cross compiler. Meanwhile, please set HOSTCC.
34 # cross compiler for Windows
35 # CC = x86_64-w64-mingw32-gcc
36 # FC = x86_64-w64-mingw32-gfortran
38 # cross compiler for 32bit ARM
39 # CC = arm-linux-gnueabihf-gcc
40 # FC = arm-linux-gnueabihf-gfortran
42 # cross compiler for 64bit ARM
43 # CC = aarch64-linux-gnu-gcc
44 # FC = aarch64-linux-gnu-gfortran
47 # If you use the cross compiler, please set this host compiler.
50 # If you need 32bit binary, define BINARY=32, otherwise define BINARY=64
53 # About threaded BLAS. It will be automatically detected if you don't
55 # For force setting for single threaded, specify USE_THREAD = 0
56 # For force setting for multi threaded, specify USE_THREAD = 1
59 # If you're going to use this library with OpenMP, please comment it in.
60 # This flag is always set for POWER8. Don't modify the flag
63 # The OpenMP scheduler to use - by default this is "static" and you
64 # will normally not want to change this unless you know that your main
65 # workload will involve tasks that have highly unbalanced running times
66 # for individual threads. Changing away from "static" may also adversely
67 # affect memory access locality in NUMA systems. Setting to "runtime" will
68 # allow you to select the scheduler from the environment variable OMP_SCHEDULE
69 # CCOMMON_OPT += -DOMP_SCHED=dynamic
71 # You can define maximum number of threads. Basically it should be
72 # less than actual number of cores. If you don't specify one, it's
73 # automatically detected by the the script.
76 # If you have enabled USE_OPENMP and your application would call
77 # OpenBLAS's calculation API from multi threads, please comment it in.
78 # This flag defines how many instances of OpenBLAS's calculation API can
79 # actually run in parallel. If more threads call OpenBLAS's calculation API,
80 # they need to wait for the preceding API calls to finish or risk data corruption.
83 # if you don't need to install the static library, please comment it in.
86 # if you don't need generate the shared library, please comment it in.
89 # If you don't need CBLAS interface, please comment it in.
92 # If you only want CBLAS interface without installing Fortran compiler,
93 # please comment it in.
96 # If you don't need LAPACK, please comment it in.
97 # If you set NO_LAPACK=1, the library automatically sets NO_LAPACKE=1.
100 # If you don't need LAPACKE (C Interface to LAPACK), please comment it in.
103 # Build LAPACK Deprecated functions since LAPACK 3.6.0
104 BUILD_LAPACK_DEPRECATED = 1
106 # Build RecursiveLAPACK on top of LAPACK
109 # If you want to use legacy threaded Level 3 implementation.
110 # USE_SIMPLE_THREADED_LEVEL3 = 1
112 # If you want to use the new, still somewhat experimental code that uses
113 # thread-local storage instead of a central memory buffer in memory.c
114 # Note that if your system uses GLIBC, it needs to have at least glibc 2.21
118 # If you want to drive whole 64bit region by BLAS. Not all Fortran
119 # compiler supports this. It's safe to keep comment it out if you
120 # are not sure(equivalent to "-i8" option).
123 # Unfortunately most of kernel won't give us high quality buffer.
124 # BLAS tries to find the best region before entering main function,
125 # but it will consume time. If you don't like it, you can disable one.
128 # If you want to disable CPU/Memory affinity on Linux.
131 # if you are compiling for Linux and you have more than 16 numa nodes or more than 256 cpus
134 # Don't use AVX kernel on Sandy Bridge. It is compatible with old compilers
135 # and OS. However, the performance is low.
138 # Don't use Haswell optimizations if binutils is too old (e.g. RHEL6)
141 # Don't use parallel make.
142 # NO_PARALLEL_MAKE = 1
144 # Force number of make jobs. The default is the number of logical CPU of the host.
145 # This is particularly useful when using distcc.
146 # A negative value will disable adding a -j flag to make, allowing to use a parent
147 # make -j value. This is useful to call OpenBLAS make from an other project
151 # If you would like to know minute performance report of GotoBLAS.
152 # FUNCTION_PROFILE = 1
154 # Support for IEEE quad precision(it's *real* REAL*16)( under testing)
155 # This option should not be used - it is a holdover from unfinished code present
156 # in the original GotoBLAS2 library that may be usable as a starting point but
157 # is not even expected to compile in its present form.
160 # Theads are still working for a while after finishing BLAS operation
161 # to reduce thread activate/deactivate overhead. You can determine
162 # time out to improve performance. This number should be from 4 to 30
163 # which corresponds to (1 << n) cycles. For example, if you set to 26,
164 # thread will be running for (1 << 26) cycles(about 25ms on 3.0GHz
165 # system). Also you can control this mumber by THREAD_TIMEOUT
166 # CCOMMON_OPT += -DTHREAD_TIMEOUT=26
168 # Using special device driver for mapping physically contigous memory
169 # to the user space. If bigphysarea is enabled, it will use it.
170 # DEVICEDRIVER_ALLOCATION = 1
172 # If you need to synchronize FP CSR between threads (for x86/x86_64 only).
173 # CONSISTENT_FPCSR = 1
175 # If any gemm arguement m, n or k is less or equal this threshold, gemm will be execute
176 # with single thread. (Actually in recent versions this is a factor proportional to the
177 # number of floating point operations necessary for the given problem size, no longer
178 # an individual dimension). You can use this setting to avoid the overhead of multi-
179 # threading in small matrix sizes. The default value is 4, but values as high as 50 have
180 # been reported to be optimal for certain workloads (50 is the recommended value for Julia).
181 # GEMM_MULTITHREAD_THRESHOLD = 4
183 # If you need santy check by comparing reference BLAS. It'll be very
184 # slow (Not implemented yet).
187 # The installation directory.
188 # PREFIX = /opt/OpenBLAS
190 # Common Optimization Flag;
191 # The default -O2 is enough.
192 # Flags for POWER8 are defined in Makefile.power. Don't modify COMMON_OPT
195 # gfortran option for LAPACK to improve thread-safety
196 # It is enabled by default in Makefile.system for gfortran
197 # Flags for POWER8 are defined in Makefile.power. Don't modify FCOMMON_OPT
198 # FCOMMON_OPT = -frecursive
203 # Build Debug version
206 # Set maximum stack allocation.
207 # The default value is 2048. 0 disable stack allocation a may reduce GER and GEMV
208 # performance. For details, https://github.com/xianyi/OpenBLAS/pull/482
210 # MAX_STACK_ALLOC = 0
212 # Add a prefix or suffix to all exported symbol names in the shared library.
213 # Avoid conflicts with other BLAS libraries, especially when using
214 # 64 bit integer interfaces in OpenBLAS.
215 # For details, https://github.com/xianyi/OpenBLAS/pull/459
217 # The same prefix and suffix are also added to the library name,
218 # i.e. you get lib$(SYMBOLPREFIX)openblas$(SYMBOLSUFFIX) rather than libopenblas
224 # End of user configuration