C++ Template Linear Algebra PACKage
First things to know about <T>LAPACK:
- We write <T>LAPACK whenever it is possible. This includes all software documentation, discussions, and presentation.
- We say it T-L-A-PACK .
- We use
tlapack
for files, folders, and links, to make it more portable and easier to use.
<T>LAPACK provides:
- Precision-neutral function template implementation
Supported in part by NSF ACI 2004850.
<T>LAPACK is a work in progress (WIP) project. This is a list of the current functionality:
- BLAS
- Level 1, except SDSDOT.
- Level 2, except the routines for banded and packed formats (xGBMV, xHBMV, xHPMV, xSBMV, xSPMV, xTBMV, xTPMV, xTBSV, xTPSV, xHPR, xHPR2, xSPR, xSPR2).
- Level 3.
- Linear solvers
- Safe scaling triangular solve
- Safe scaling solver for Sylvester equations
- Matrix factorizations
- Cholesky
- Recursive
- Blocked
- Fully pivoted QR
- Fully pivoted RQ
- Fully pivoted LQ
- Fully pivoted QL
- Hessenberg reduction
- Level-2
- Blocked
- Householder QR
- Level-2
- Blocked
- Householder RQ
- Level-2
- Blocked
- Householder LQ
- Level-2
- Blocked
- Householder QL
- Level-2
- Blocked
- Householder bidiagonalization
- Level-2
- Blocked
- LU with partial pivoting
- Level-0
- Recursive
- Cholesky
- Matrix inversion
- Norms of general, hermitian, symmetric and triangular matrices
- 1-norm
- Infinity-norm
- Frobenius-norm
- Max-norm
- Nonsymmetric Eigenvalue Problem
- Schur decomposition
- Double-shift implicit QR
- Multishift implicit QR with Aggressive Early Deflation (AED)
- Eigenvector computation
- Swap eigenvalues in Schur form
- Generalized Schur problem
- Generalized eigenvalue problem
- Schur decomposition
- Symmetric Eigenvalue Problem
- Tridiagonal reduction
- Eigenvector computation
- Singular Value Decomposition (SVD)
- Standard singular value problem
- Generalized singular value problem
- Linear least squares
- Using QR factorization
- Using SVD
- Additional kernels
- Order 1 and 2 Sylvester equation solver
- In-place upper times lower triangular matrix multiplication for general and hermitian matrices
- In-place lower times upper triangular matrix multiplication
- In-place transpose of a matrix
The complete documentation and implementation for the routines that are available are listed in the API documentation.
<T>LAPACK is built and installed with CMake.
You can either download binaries for the latest stable or previous release of CMake, or build the current development distribution from source. CMake is also available in the APT repository on Ubuntu 16.04 or higher.
<T>LAPACK can be build following the standard CMake recipe
cmake -B build . # configuration step
cmake --build build # build step
To install <T>LAPACK in your system, use
cmake -B build -D CMAKE_INSTALL_PREFIX=/path/to/install . # configuration step
cmake --build build --target install # install step
Standard environment variables affect CMake. Some examples are
CXX C++ compiler
CXXFLAGS C++ compiler flags
LDFLAGS linker flags
The Fortran and C wrappers to <T>LAPACK also use, among others,
CC C compiler
CFLAGS C compiler flags
FC Fortran compiler
FFLAGS Fortran compiler flags
- This page lists the environment variables that have special meaning to CMake.
It is also possible to pass variables to CMake during the configuration step using the -D
flag.
The following example builds <T>LAPACK in debug mode inside the directory build
mkdir build
cmake -B build -DCMAKE_BUILD_TYPE=Debug .
cmake --build build
- This page documents variables that are provided by CMake or have meaning to CMake when set by project code.
Here are the <T>LAPACK specific options and their default values
# Option # Default
BUILD_BLASPP_TESTS OFF
Use BLAS++ tests to test <T>LAPACK templates.
REQUIRES: BUILD_TESTING=ON
BUILD_EXAMPLES ON
Build examples
BUILD_LAPACKPP_TESTS OFF
Use LAPACK++ tests to test <T>LAPACK templates.
REQUIRES: BUILD_TESTING=ON
BUILD_STANDALONE_TESTS OFF
Build tests as standalone executables in addition to tests for CTest.
The executables are installed in the test directory.
The executables do not use Catch2.
REQUIRES: BUILD_TESTING=ON
BUILD_testBLAS_TESTS ON
Build testBLAS tests.
REQUIRES: BUILD_TESTING=ON
BUILD_TESTING ON
Build the testing tree
BUILD_C_WRAPPERS OFF
Build and install C wrappers (Work In Progress)
BUILD_CBLAS_WRAPPERS OFF
Build and install CBLAS wrappers (Work In Progress)
BUILD_Fortran_WRAPPERS OFF
Build and install Fortran wrappers (Work In Progress)
TLAPACK_CHECK_INPUT ON
Enable checks on input arguments.
REQUIRES: TLAPACK_NDEBUG=OFF
TLAPACK_DEFAULT_INFCHECK ON
Default behavior of checks for Infs. Checks can be activated/deactivated at runtime.
REQUIRES: TLAPACK_NDEBUG=OFF
TLAPACK_ENABLE_INFCHECK=ON
TLAPACK_DEFAULT_NANCHECK ON
Default behavior of checks for NaNs. Checks can be activated/deactivated at runtime.
REQUIRES: TLAPACK_NDEBUG=OFF
TLAPACK_ENABLE_NANCHECK=ON
TLAPACK_ENABLE_INFCHECK OFF
Enable check for Infs as specified in the documentation of each routine.
REQUIRES: TLAPACK_NDEBUG=OFF
TLAPACK_ENABLE_NANCHECK OFF
Enable check for NaNs as specified in the documentation of each routine.
REQUIRES: TLAPACK_NDEBUG=OFF
TLAPACK_INT_T int64_t
Type of all non size-related integers in libtlapack_c, libtlapack_cblas, libtlapack_fortran, and in the routines of the legacy API. It is the type
used for the array increments, e.g., incx and incy.
Supported types:
int, short, long, long long, int8_t, int16_t,
int32_t, int64_t, int_least8_t, int_least16_t,
int_least32_t, int_least64_t, int_fast8_t,
int_fast16_t, int_fast32_t, int_fast64_t,
intmax_t, intptr_t, ptrdiff_t
NOTE: TLAPACK_INT_T=int64_t if TLAPACK_USE_LAPACKPP=ON
TLAPACK_NDEBUG OFF
Disable all error checks.
TLAPACK_SIZE_T size_t
Type of all size-related integers in libtlapack_c, libtlapack_cblas, libtlapack_fortran, and in the routines of the legacy API.
Supported types:
int, short, long, long long, int8_t, int16_t,
int32_t, int64_t, int_least8_t, int_least16_t,
int_least32_t, int_least64_t, int_fast8_t,
int_fast16_t, int_fast32_t, int_fast64_t,
intmax_t, intptr_t, ptrdiff_t,
size_t, uint8_t, uint16_t, uint32_t, uint64_t
NOTE: TLAPACK_SIZE_T=int64_t if TLAPACK_USE_LAPACKPP=ON
TLAPACK_USE_LAPACKPP OFF
Use LAPACK++ wrappers to link with optimized BLAS and LAPACK libraries.
Mind that LAPACK++ needs BLAS++.
Branches compatible with <T>LAPACK:
https://bitbucket.org/weslleyspereira/blaspp/branch/tlapack
https://bitbucket.org/weslleyspereira/lapackpp/branch/tlapack
<T>LAPACK currently depends on the following projects:
Project | Version | When |
---|---|---|
CMake | >= v3.14 | Always |
ClangFormat | 10 | Always |
Catch2 | >= 3.0.1 | BUILD_TESTING=ON |
Git | - | BUILD_BLASPP_TESTS=ON OR BUILD_LAPACKPP_TESTS=ON |
github.com/icl-utk-edu/testsweeper | >= 2021.04.00 | BUILD_BLASPP_TESTS=ON OR BUILD_LAPACKPP_TESTS=ON |
LAPACK | >= 3.9.0 | BUILD_BLASPP_TESTS=ON OR BUILD_LAPACKPP_TESTS=ON |
ICL BLAS++ test suite | 2023.06.00 | BUILD_BLASPP_TESTS=ON |
ICL LAPACK++ test suite | 2022.07.00 | BUILD_LAPACKPP_TESTS=ON |
LAPACKE | >= 3.9.0 | BUILD_LAPACKPP_TESTS=ON |
github.com/tlapack/testBLAS | Latest | BUILD_testBLAS_TESTS=ON |
Note that:
- The CMake build system is prepared to download Catch2 and testsweeper whenever they are not available in the system.
- ICL BLAS++ and LAPACK++ test suites and testBLAS are Git submodules of <T>LAPACK. The are downloaded and built when the corresponding options are set to
ON
.
<T>LAPACK may also use the following projects when available:
Project | Version tested | When |
---|---|---|
ICL BLAS++ | 2023.06.00 | TLAPACK_USE_LAPACKPP=ON |
ICL LAPACK++ | 2023.06.00 | TLAPACK_USE_LAPACKPP=ON |
kokkos mdspan | 0.4.0 | TLAPACK_TEST_MDSPAN=ON or in some examples |
Eigen | commit: 2873916f | TLAPACK_TEST_EIGEN=ON or in some examples |
GNU MPFR C++ | Latest in APT | TLAPACK_TEST_MPFR=ON or in some examples |
GNU libquad | Latest in APT | TLAPACK_TEST_QUAD=ON |
StarPU | 1.4.1 | Running StarPU examples |
We also continuously test <T>LAPACK with optimized BLAS and LAPACK implementations: OpenBLAS, Intel MKL, Flame BLIS, LAPACK, Netlib BLAS.
Note
<T>LAPACK supports Eigen v3.4.0 after applying the patch 9210e71f. This commit fixes an ADL conflict between Eigen::internal::size()
and std::size()
.
The documentation of <T>LAPACK is generated using Doxygen. The documentation is available online at https://tlapack.github.io/tlapack. Alternatively, you can generate the documentation in your local machine. To do so, follow the steps below:
-
Install Doxygen in your local machine. See the Doxygen installation page for more details.
-
In the top directory of <T>LAPACK, run
doxygen docs/Doxyfile
to generate the <T>LAPACK documentation via Doxygen in your local machine.
Additional information about the software can be found in the Wiki pages of the project.
Please read CONTRIBUTING.md for details on how to contribute to <T>LAPACK.
<T>LAPACK is continuously tested on Ubuntu, MacOS and Windows using GNU compilers. See the latest test results in the Github Actions webpage for <T>LAPACK. The tests split into three categories:
-
Test routines in test/src using
-
various precision types:
float
,double
,std::complex<float>
,std::complex<double>
andEigen::half
. -
various matrix and vector data structures:
tlapack::LegacyMatrix
,Eigen::Matrix
andstd::experimental::mdspan
(the latter fromhttps://github.com/kokkos/mdspan
).
-
-
Tests from testBLAS for good conformance with BLAS standards.
-
BLAS++ testers and LAPACK++ testers for measuring performance and accuracy compared to LAPACKE.
To test <T>LAPACK, build with BUILD_TESTING=ON
. Then, run ctest
inside the build directory.
BSD 3-Clause License
Copyright (c) 2021-2023, University of Colorado Denver. All rights reserved.
Copyright (c) 2017-2021, University of Tennessee. All rights reserved.
Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:
-
Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
-
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.
-
Neither the name of the copyright holder nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission.
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 HOLDER 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.
[1] Higham, N. J. (2002). Accuracy and stability of numerical algorithms. Society for industrial and applied mathematics.