This could be related to the previous bug in the Fortran-C interop features I reported, but since the conditions around the bug and symptoms are different I believe this is two different problems. Therefore I make a new thread about this one.
I have a main Fortran program program.F90:
PROGRAM main
USE, INTRINSIC :: ISO_C_BINDING, ONLY: c_float
IMPLICIT NONE
INTERFACE
SUBROUTINE check_array(res) BIND(C)
IMPORT :: c_float
REAL(c_float), INTENT(inout), ALLOCATABLE :: res(:, :, :)
END SUBROUTINE check_array
END INTERFACE
REAL, ALLOCATABLE :: my_array(:, :, :)
CALL check_array(my_array)
WRITE(*, *) "First element of my_array: ", my_array(1, 1, 1)
WRITE(*, *) "Lower bound of my_array: ", LBOUND(my_array)
WRITE(*, *) "Size of my_array: ", SHAPE(my_array)
WRITE(*, *) "Last element of my_array: ", my_array(SIZE(my_array, 1), &
SIZE(my_array, 2), SIZE(my_array, 3))
END PROGRAM main
and an accompanying C++ routine in check_array.cpp:
#include <ISO_Fortran_binding.h>
#include <iostream>
#include <cassert>
extern "C" {
void check_array(CFI_cdesc_t* data) {
assert (data->rank == 3);
assert (data->type == CFI_type_float);
// If already allocated - free
if (data->base_addr) CFI_deallocate(data);
// Allocate the Fortran array
const CFI_index_t lower_bounds[3] = {1, 1, 1};
const CFI_index_t upper_bounds[3] = {3, 3, 1996};
int ierr = CFI_allocate(data, lower_bounds, upper_bounds, 0);
assert (ierr == CFI_SUCCESS);
const size_t nelems =
(data->dim[0].extent - data->dim[0].lower_bound + 1) *
(data->dim[1].extent - data->dim[1].lower_bound + 1) *
(data->dim[2].extent - data->dim[2].lower_bound + 1);
std::cout << "nelems: " << nelems << std::endl;
// Fill array with some data
float* ptr = reinterpret_cast<float*>(data->base_addr);
for (size_t i = 0; i < nelems; ++i) {
ptr[i] = static_cast<float>(i);
}
}
}
Again, this works with Intel, GNU and NAG compilers. The output should be something like:
$ ./test_cfi
nelems: 17964
First element of my_array: 0.00000000
Lower bound of my_array: 1 1 1
Size of my_array: 3 3 1996
Last element of my_array: 17963.0000
When compiling with nvfortran and nvc++ I get a segfault:
$ ./test_cfi
nelems: 17964
Segmentation fault (core dumped)
The segfault occur when writing into the allocated array (ptr[i] = static_cast<float>(i);). I strongly suspect that CFI_allocate is not allocating the correct memory…
I am using the following nvfortran version:
$ nvfortran --version
nvfortran 24.3-0 64-bit target on x86-64 Linux -tp alderlake
NVIDIA Compilers and Tools
Copyright (c) 2024, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
Again, thanks for your effort.