MPLAPACK is a multi-precision linear algebra package based on BLAS and LAPACK, implemented in C++ (C++17 required). It supports a range of high-precision arithmetic libraries including GMP, MPFR, and QD, allowing users to select the backend that best suits their requirements. MPLAPACK is free software distributed under a 2-clause BSD-style license, supplemental to the original LAPACK license.

• 2026-04-09 MPLAPACK 2.1.1 released. Patch release: GCC 15 support (external/gmp C23 fix, {C,R}drgvx IPA-modref workaround), arm64 promoted to Tier 1 (Ubuntu arm64 and macOS Apple Silicon), musl/Alpine ARM64 build fix, MinGW laed3 build fix, binary128 std::abs(__float128) ambiguity fix on aarch64-apple-darwin, and a latent DD miscompilation fix (propagating -ffp-contract=off to all DD build targets). OpenBLAS bumped to 0.3.32. ABI/source compatible with 2.1.0.
• 2026-03-05 MPLAPACK 2.1.0 released. C++17 now required. binary128/binary80 naming unified, MPFR emin/emax auto-adjustment, extended build matrix. LAPACK 3.9.1 Fortran sources are now mechanically converted to idiomatic C++ via Fable and bundled in the release tarball. See CHANGES.md and MIGRATION.md.
• 2022-09-12 MPLAPACK 2.0.1 released, featuring CUDA versions of Rgemm (dd) and Rsyrk (dd) for Volta and Ampere architectures (~450 GFlops on V100), and Windows DLLs for MinGW-w64.
• 2022-07-26 MPLAPACK 2.0.0 released. All routines (except mixed-precision) functional and tested across all supported precisions.

• MPBLAS: All BLAS routines in multiple-precision arithmetic.
• MPLAPACK: All LAPACK routines in multiple-precision arithmetic (except mixed-precision routines):
  • Linear Equations
  • Linear Least Squares (LLS) Problems
  • Generalized Linear Least Squares (LSE and GLM) Problems
  • Standard Eigenvalue and Singular Value Problems
  • Symmetric Eigenproblems (SEP)
  • Nonsymmetric Eigenproblems (NEP)
  • Singular Value Decomposition (SVD)
  • Generalized Eigenvalue and Singular Value Problems
  • Generalized Symmetric Definite Eigenproblems (GSEP)
  • Generalized Nonsymmetric Eigenproblems (GNEP)
  • Generalized Singular Value Decomposition (GSVD)

• MPFR + MPC https://www.mpfr.org/ / http://www.multiprecision.org/mpc/ (arbitrary precision with IEEE-like rounding; primary backend)
• GMP https://gmplib.org/ (arbitrary precision)
• double (binary64)
• DD, QD https://www.davidhbailey.com/dhbsoftware/ (DD ≈ binary128, QD ≈ binary256)
• binary128 (IEEE 754-2018; compiler and platform support is complex — see binary128_binary80_type_support.md)
• binary80 (80-bit extended precision; Intel/AMD x86 only)

Clang users: binary128 is not supported. Use GCC for binary128. GCC 15: supported as of 2.1.1. On musl-based distros (e.g. Alpine), the binary128 backend stays on __float128 + libquadmath because musl does not ship strfromf128/strfromf64x.

Tier 1/2 build scripts are in misc/:

misc/buildtest_tier1_amd64_macOS.sh
misc/buildtest_tier1_arm64_macOS.sh
misc/buildtest_tier1_amd64_mingw64.sh
misc/buildtest_tier1_amd64_ubuntu.sh
misc/buildtest_tier1_arm64_ubuntu.sh
misc/buildtest_tier2_amd64_ubuntu_intel.sh
misc/buildtest_tier2_i386_debian.sh

• GCC / G++ / GFortran (C++17 is required; GCC 15 supported as of 2.1.1)
• Standard autotools: autoconf, automake, libtool
• wget or curl (to fetch the tarball)

All third-party libraries (GMP, MPFR, MPC, QD, OpenBLAS, dlfcn-win32) are bundled and built automatically. No separate installation of these libraries is required.

mkdir -p $HOME/tmp && cd $HOME/tmp
wget https://github.com/nakatamaho/mplapack/releases/download/v2.1.1/mplapack-2.1.1.tar.xz
tar xvf mplapack-2.1.1.tar.xz
cd mplapack-2.1.1
export CXX=g++ CC=gcc FC=gfortran
./configure \
    --prefix=$HOME/MPLAPACK \
    --enable-gmp=yes \
    --enable-mpfr=yes \
    --enable-binary128=yes \
    --enable-qd=yes \
    --enable-dd=yes \
    --enable-double=yes \
    --enable-test=yes \
    --enable-benchmark=yes
make -j$(nproc)
make install

To also enable binary80 (Intel/AMD x86 only), add --enable-binary80=yes:

./configure \
    --prefix=$HOME/MPLAPACK \
    --enable-binary80=yes \
    --enable-binary128=yes \
    --enable-gmp=yes \
    --enable-mpfr=yes \
    --enable-qd=yes \
    --enable-dd=yes \
    --enable-double=yes \
    --enable-test=yes \
    --enable-benchmark=yes

FSF GCC is required. The default Apple Clang does not support binary128. On Apple Silicon (arm64), omit --enable-binary80=yes (binary80 is x86-only).

sudo port install gcc15 coreutils git gsed
mkdir -p $HOME/tmp && cd $HOME/tmp
wget https://github.com/nakatamaho/mplapack/releases/download/v2.1.1/mplapack-2.1.1.tar.xz
tar xvf mplapack-2.1.1.tar.xz
cd mplapack-2.1.1
export CXX=g++-mp-15 CC=gcc-mp-15 FC=gfortran-mp-15
./configure \
    --prefix=$HOME/MPLAPACK \
    --enable-gmp=yes \
    --enable-mpfr=yes \
    --enable-binary128=yes \
    --enable-qd=yes \
    --enable-dd=yes \
    --enable-double=yes \
    --enable-test=yes \
    --enable-benchmark=yes
# On Intel macs, additionally pass --enable-binary80=yes
make -j$(sysctl -n hw.logicalcpu)
make install

sudo apt-get install gcc-mingw-w64-x86-64 g++-mingw-w64-x86-64 gfortran-mingw-w64-x86-64
mkdir -p $HOME/tmp && cd $HOME/tmp
wget https://github.com/nakatamaho/mplapack/releases/download/v2.1.1/mplapack-2.1.1.tar.xz
tar xvf mplapack-2.1.1.tar.xz
cd mplapack-2.1.1
export CXX=x86_64-w64-mingw32-g++
export CC=x86_64-w64-mingw32-gcc
export FC=x86_64-w64-mingw32-gfortran
./configure \
    --host=x86_64-w64-mingw32 \
    --prefix=$HOME/MPLAPACK \
    --enable-gmp=yes \
    --enable-mpfr=yes \
    --enable-binary128=yes \
    --enable-binary80=yes \
    --enable-qd=yes \
    --enable-dd=yes \
    --enable-double=yes \
    --enable-test=yes
make -j$(nproc)
make install

make check

Test results are summarized automatically by misc/summarize_mplapack_tests.py.

Note: The Fable conversion pipeline is not included in the release tarball. It is available only via Git clone. Regenerating sources also requires expanding the bundled LAPACK 3.9.1 source under external/lapack/ before running the scripts.

fable/ is a top-level standalone component providing automated Fortran-to-C++ conversion of LAPACK 3.9.1 sources via Fable and FEM (Fortran Emulator).

# Step 1: clone the repository
git clone https://github.com/nakatamaho/mplapack
cd mplapack

# Step 2: expand bundled LAPACK sources
cd external/lapack
tar xvf lapack-3.9.1.tar.gz
cd ../..

# Step 3: run the conversion pipeline
bash fable/go.sh          # library routines (BLAS/LAPACK C++ sources + headers + patches)
bash fable/go_testing.sh  # test programs (EIG/LIN/MATGEN)

• https://github.com/nakatamaho/mplapack/tree/master/mplapack/test/lin/results
• https://github.com/nakatamaho/mplapack/tree/master/mplapack/test/eig/results

• https://github.com/nakatamaho/mplapack/tree/master/benchmark/results/2022

• https://arxiv.org/abs/2109.13406v2
• https://raw.githubusercontent.com/nakatamaho/mplapack/master/doc/manual/manual.pdf

@misc{2109.13406v2,
  Author = {Maho Nakata},
  Title  = {MPLAPACK version 2.0.1 user manual},
  Year   = {2022},
  Eprint = {arXiv:2109.13406v2},
}

• https://www.youtube.com/watch?v=M76wHwckNPU (created by Ge Baolai)

• https://github.com/nakatamaho/mplapack/blob/master/doc/presentation/2023-06-01%20CMSI%E6%95%99%E8%82%B2%E8%A8%88%E7%AE%97%E7%A7%91%E5%AD%A6%E6%8A%80%E8%A1%93%E7%89%B9%E8%AB%96A%20%E7%AC%AC7%E5%9B%9E%20%E4%B8%AD%E7%94%B0%E7%9C%9F%E7%A7%80.pdf (in Japanese)
• https://github.com/nakatamaho/mplapack/blob/v2.0/doc/presentation/20211128_%E7%B2%BE%E5%BA%A6%E4%BF%9D%E8%A8%98meeting.pdf (in Japanese)

2.1.1 is a patch release. No API/ABI changes vs. 2.1.0. Highlights:

• GCC 15 support. external/gmp patched for the C23 default (void g(){} → properly typed prototype); patch is gated on GCC >= 15 at configure time. {C,R}drgvx test drivers carry an __attribute__((optimize("O1"))) workaround for a g++-15 IPA-modref miscompilation.
• arm64 promoted to Tier 1: Ubuntu arm64 and macOS Apple Silicon.
• musl/Alpine ARM64 build fix (M_PIl fallback).
• MinGW lapack/laed3 build fix (LAMC3 prototype, max macro guard).
• binary128 sign() ambiguity on aarch64-apple-darwin / MacPorts GCC 15 fixed (unqualified abs()).
• DD miscompilation hardening: -ffp-contract=off (and -fp-model strict for icpx) is now propagated to all DD build targets — reference and optimized BLAS/LAPACK, tests, and benchmarks. CUDA DD targets are intentionally excluded (nvcc rejects the flag).
• configure fix: a stale CXXFLAGS="$SAVE_CXXFLAGS" line was silently clearing user-supplied CXXFLAGS (e.g. -fsanitize=address).
• std::abs(__float128) probe tightened to require an exact-overload match instead of trusting __SIZEOF_FLOAT128__.
• dd/qd test comparisons switched from diff to misc/num_diff.py (relative tolerance 1e-30 for dd, 1e-60 for qd) to absorb 1-bit FMA-induced rounding noise.
• make distcheck fixes: external/openblas/Makefile moved into the benchmark-conditional AC_CONFIG_FILES block; dd/qd test scripts use ${srcdir} for VPATH builds; missing misc/ scripts added to EXTRA_DIST.
• OpenBLAS updated to 0.3.32 (also fixes the macOS arm64 build).
• configure build summary: now reports compiler commands and versions, target CPU and integer model, OpenMP runtime, enabled backends, binary128/binary80 type / I/O / math / literal-suffix / interop, and std::abs(__float128) availability.
• Release build matrix broadened across 77 (OS, arch, toolchain) configurations: Alpine 3.19–3.23, Debian 11–13, Ubuntu 18.04–24.04, Rocky 8–10, Fedora 42/43, openSUSE Leap 15/16, openSUSE Tumbleweed, on x86_64, i386/i686, aarch64, ppc64le, s390x, riscv64, mips64le, plus Intel oneAPI (icpx 2025.3.2) and MinGW-w64 cross targets.
• Enterprise lifecycle note: Rocky 8/9, openSUSE Leap 15, and Ubuntu 18/20/22 stay on the libquadmath path for their full support window — libquadmath support cannot be dropped yet.

Tier 1 platforms run the full pipeline including make distcheck. Tier 2 platforms run build only.

Tier 1 (release blockers): make distcheck must pass on all Tier 1 platforms. Tier 2 (build guarantee): build only; not release-blocking. Tier 3 (patches accepted): no CI coverage.

• version 2.0.1: https://github.com/nakatamaho/mplapack/blob/master/doc/Release2.0.1.md
• version 2.0.0: https://github.com/nakatamaho/mplapack/blob/master/doc/Release2.0.0.md
• version 1.0.0: https://github.com/nakatamaho/mplapack/blob/master/doc/Release1.0.0.md

• 2026-04-09 MPLAPACK 2.1.1 released. GCC 15 support, arm64 promoted to Tier 1 (Ubuntu arm64, macOS Apple Silicon), DD -ffp-contract=off propagation fix, binary128 / MinGW / musl build fixes, OpenBLAS 0.3.32. ABI compatible with 2.1.0.
• 2026-03-05 MPLAPACK 2.1.0 released. binary128/binary80 naming unified, MPFR emin/emax auto-adjustment, extended build matrix (Alpine, Rocky, Debian i386, CUDA 13.1.1). LAPACK 3.9.1 Fortran sources mechanically converted to idiomatic C++ via Fable and bundled in the release tarball.
• 2022-09-12 MPLAPACK 2.0.1 released.
• 2022-07-26 MPLAPACK 2.0.0 released.
• 2022-06-14 MPLAPACK 2.0.0 alpha released.
• 2021-11-01 1.0.1 release. Fixed DD and QD arithmetic with Intel oneAPI.
• 2021-10-01 1.0.0 release. All real LAPACK routines available; SVD and non-symmetric eigenproblem solvers added.
• 2021-04-11 0.9.3 release. CentOS 7 AArch64 support.
• 2021-04-01 0.9.0 release. Renamed to mplapack.
• 2012-12-25 MPACK 0.8.0. NVIDIA C2050 support for Rgemm (double-double).
• 2010-08-20 MPACK 0.6.7. Condition number estimators added (Rgecon, Rpocon). License changed to 2-clause BSD.
• 2009-11-24 MPACK 0.6.0.
• 2008-07-15 mpack-0.0.1. Now configurable and installable.
• 2008-06-24 Project page created.

http://mplapack.sourceforge.net/

This work has been supported by: The Special Postdoctoral Researchers' Program of RIKEN (2008, 2009), Grant-in-Aid for Scientific Research (B) 21300017 from the Japan Society for the Promotion of Science (2009–2011), Microsoft Research CORE6 (2010), JSPS KAKENHI Grant no. 18H03206, and TIS inc.

M.N. would like to thank Dr. Imamura Toshiyuki, Dr. Nakasato Naohito, Dr. Fujisawa Katsuki, Dr. Kouya Tomonori, Dr. Takahashi Daisuke, Dr. Goto Kazushige, Dr. Himeno Ryutaro, Dr. Hishimuna Toshiaki, Dr. Katagiri Takahiro, Dr. Ogita Takeshi, Dr. Kashiwagi Masaaki, Dr. Yuasa Fukuko, Dr. Ishikawa Tadashi, Dr. Geshi Masaaki, and Mr. Minato Yuichiro for warm encouragement.

@misc{2109.13406v2,
  Author = {Maho Nakata},
  Title  = {MPLAPACK version 2.0.1 user manual},
  Year   = {2022},
  Eprint = {arXiv:2109.13406v2},
}

NAKATA Maho [email protected] [email protected]