FUNDAL

Fortran UNified Device Acceleration Library

a pure Fortran library providing a unified API for GPU/device memory management over OpenACC and OpenMP backends.

🔀 Unified API OpenACC or OpenMP backend selected at compile time — no conditional code in user programs	🧱 Structured memory Device-only pointer arrays via dev_alloc, dev_free, dev_memcpy_*	📦 Unstructured memory Host allocatable arrays mapped to the device via dev_alloc_unstr	🔄 Assignment-style copy dev_assign_to/from_device with automatic reallocation on size change
🖥️ Device handling Query device type, ID, count, memory, and properties	🌐 MPI multi-device mpih_object class for multi-GPU parallel scenarios	📐 Full rank & kind coverage Array ranks 1–7; numeric kinds R8P, R4P, I8P, I4P, I2P, I1P	🏗️ Multiple build systems FoBiS, make

Documentation

For full documentation (guide, API reference, examples, etc...) see the FUNDAL website.

---

Authors

• Stefano Zaghi — stefano.zaghi@cnr.it
• Giacomo Rossi — giacomo.rossi@amd.com
• Andrea di Mascio — andrea.dimascio@univaq.it
• Francesco Salvadore — f.salvadore@cineca.it

Contributions are welcome — see the Contributing page.

Copyrights

This project is distributed under a multi-licensing system:

• FOSS projects: GPL v3
• Closed source / commercial: BSD 2-Clause, BSD 3-Clause, or MIT

Anyone interested in using, developing, or contributing to FUNDAL is welcome — pick the license that best fits your needs.

---

Quick start

program fundal_taste
use, intrinsic :: iso_fortran_env, only : I4P=>int32, R8P=>real64
use            :: fundal

implicit none
real(R8P), pointer     :: a_dev(:,:,:)=>null()  ! device memory
real(R8P), allocatable :: b_hos(:,:,:)           ! host   memory
integer(I4P)           :: ierr, i, j, k

call dev_init  ! initialise device environment

call dev_alloc(fptr_dev=a_dev, lbounds=[-1,-2,-3], ubounds=[1,2,3], ierr=ierr)
allocate(b_hos(-1:1,-2:2,-3:3))
b_hos = -3._R8P

call dev_memcpy_to_device(dst=a_dev, src=b_hos)

!$acc parallel loop independent deviceptr(a_dev) collapse(3)
!$omp target teams distribute parallel do collapse(3) has_device_addr(a_dev)
do k=-3,3; do j=-2,2; do i=-1,1
  a_dev(i,j,k) = a_dev(i,j,k) / 2._R8P
enddo; enddo; enddo

call dev_memcpy_from_device(dst=b_hos, src=a_dev)
print*, b_hos

call dev_free(a_dev)
endprogram fundal_taste

See src/tests/ for more examples including MPI multi-device and unstructured memory patterns.

---

Install

FoBiS

git clone https://github.com/szaghi/FUNDAL && cd FUNDAL
FoBiS.py build -mode fundal-test-oac-nvf   # NVIDIA nvfortran + OpenACC
FoBiS.py build -mode fundal-test-omp-ifx   # Intel IFX + OpenMP
FoBiS.py build -mode fundal-test-oac-gnu   # GNU gfortran + OpenACC (partial)
FoBiS.py build -mode fundal-test-omp-amd   # AMD Flang + OpenMP

Run all tests:

utils/run_tests.sh

Each test prints test passed on success.

install.sh

Download and build in one step using the bundled install script from the latest release:

wget $(curl -s https://api.github.com/repos/szaghi/FUNDAL/releases/latest \
  | jq -r '.assets[] | select(.name | test("install.sh";"i")) | .browser_download_url')
chmod +x install.sh
./install.sh --download wget --build fobis