FindBLAS

Finds the installed Basic Linear Algebra Subprograms (BLAS) Fortran library, which implements the BLAS linear-algebra interface:

find_package(BLAS [...])

At least one of the C, CXX, or Fortran languages must be enabled.

Imported Targets

This module provides the following Imported Targets:

BLAS::BLAS

Added in version 3.18.

Target encapsulating the libraries and usage requirements to use BLAS, available only if BLAS is found.

Result Variables

This module defines the following variables:

BLAS_FOUND

Boolean indicating whether the library implementing the BLAS interface is found.

BLAS_LINKER_FLAGS

Uncached list of required linker flags (excluding -l and -L).

BLAS_LIBRARIES

Uncached list of libraries (using full path name) to link against to use BLAS (may be empty if compiler implicitly links BLAS).

BLAS95_LIBRARIES

Uncached list of libraries (using full path name) to link against to use BLAS95 interface.

BLAS95_FOUND

Boolean indicating whether the library implementing the BLAS95 interface is found.

Input Variables

The following variables may be set to influence this module's behavior:

BLA_STATIC

If ON, the static linkage will be used.

BLA_VENDOR

Set to one of the BLAS/LAPACK Vendors to search for BLAS only from the specified vendor. If not set, all vendors are considered.

BLA_F95

If ON, the module tries to find the BLAS95 interfaces.

BLA_PREFER_PKGCONFIG

Added in version 3.11.

If set, pkg-config will be used to search for a BLAS library first and if one is found that is preferred

BLA_PKGCONFIG_BLAS

Added in version 3.25.

If set, the pkg-config method will look for this module name instead of just blas.

BLA_SIZEOF_INTEGER

Added in version 3.22.

Specify the BLAS/LAPACK library integer size:

4

Search for a BLAS/LAPACK with 32-bit integer interfaces.

8

Search for a BLAS/LAPACK with 64-bit integer interfaces.

ANY

Search for any BLAS/LAPACK. Most likely, a BLAS/LAPACK with 32-bit integer interfaces will be found.

BLA_THREAD

Added in version 4.1.

Specify the BLAS/LAPACK threading model:

SEQ

Sequential model

OMP

OpenMP model

ANY

Search for any BLAS/LAPACK, if both are available most likely OMP will be found.

This is currently only supported by NVIDIA NVPL.

BLAS/LAPACK Vendors

Generic

Generic reference implementation.

ACML, ACML_MP, ACML_GPU

AMD Core Math Library

AOCL, AOCL_mt

Added in version 3.27.

AMD Optimizing CPU Libraries.

Apple, NAS

Apple BLAS (Accelerate), and Apple NAS (vecLib).

Arm, Arm_mp, Arm_ilp64, Arm_ilp64_mp

Added in version 3.18.

Arm Performance Libraries.

ATLAS

Automatically Tuned Linear Algebra Software.

CXML, DXML

Compaq/Digital Extended Math Library.

EML, EML_mt

Added in version 3.20.

Elbrus Math Library.

FLAME

Added in version 3.11.

BLIS Framework.

FlexiBLAS

Added in version 3.19.

Fujitsu_SSL2, Fujitsu_SSL2BLAMP, Fujitsu_SSL2SVE, Fujitsu_SSL2BLAMPSVE

Added in version 3.20.

Fujitsu SSL2 serial and parallel blas/lapack with SVE instructions.

Goto

GotoBLAS.

IBMESSL, IBMESSL_SMP

IBM Engineering and Scientific Subroutine Library.

Intel

Intel MKL 32 bit and 64 bit obsolete versions.

Intel10_32

Intel MKL v10 32 bit, threaded code.

Intel10_64lp

Intel MKL v10+ 64 bit, threaded code, lp64 model.

Intel10_64lp_seq

Intel MKL v10+ 64 bit, sequential code, lp64 model.

Intel10_64ilp

Added in version 3.13.

Intel MKL v10+ 64 bit, threaded code, ilp64 model.

Intel10_64ilp_seq

Added in version 3.13.

Intel MKL v10+ 64 bit, sequential code, ilp64 model.

Intel10_64_dyn

Added in version 3.17.

Intel MKL v10+ 64 bit, single dynamic library.

libblastrampoline

Added in version 3.30.

A BLAS/LAPACK demuxing library using PLT trampolines.

NVPL

Added in version 4.1.

NVIDIA Performance Libraries.

NVHPC

Added in version 3.21.

NVIDIA HPC SDK.

OpenBLAS

Added in version 3.6.

PhiPACK

Portable High Performance ANSI C (PHiPAC).

SCSL, SCSL_mp

Scientific Computing Software Library.

SGIMATH

SGI Scientific Mathematical Library.

SunPerf

Sun Performance Library.

Intel MKL

To use the Intel MKL implementation of BLAS, a project must enable at least one of the C or CXX languages. Set BLA_VENDOR to an Intel MKL variant either on the command-line as -DBLA_VENDOR=Intel10_64lp or in project code:

set(BLA_VENDOR Intel10_64lp)
find_package(BLAS)

In order to build a project using Intel MKL, and end user must first establish an Intel MKL environment:

Intel oneAPI

Source the full Intel environment script:

. /opt/intel/oneapi/setvars.sh

Or, source the MKL component environment script:

. /opt/intel/oneapi/mkl/latest/env/vars.sh

Intel Classic

Source the full Intel environment script:

. /opt/intel/bin/compilervars.sh intel64

Or, source the MKL component environment script:

. /opt/intel/mkl/bin/mklvars.sh intel64

The above environment scripts set the MKLROOT environment variable to the top of the MKL installation. They also add the location of the runtime libraries to the dynamic library loader environment variable for your platform (e.g. LD_LIBRARY_PATH). This is necessary for programs linked against MKL to run.

Note

As of Intel oneAPI 2021.2, loading only the MKL component does not make all of its dependencies available. In particular, the iomp5 library must be available separately, or provided by also loading the compiler component environment:

. /opt/intel/oneapi/compiler/latest/env/vars.sh

Examples

Finding BLAS and linking it to a project target:

find_package(BLAS)
target_link_libraries(example PRIVATE BLAS::BLAS)