https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86416
Tobias Burnus <burnus at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |burnus at gcc dot gnu.org,
| |rguenth at gcc dot gnu.org
--- Comment #4 from Tobias Burnus <burnus at gcc dot gnu.org> ---
Cross ref: See also https://github.com/SOLLVE/sollve_vv/issues/41
(In reply to Jose Manuel Monsalve Diaz from comment #0)
> The problems:
> 1. DOUBLE COMPLEX values produce a compiler LTO error:
The problem is that COMPLEX(kind=16) alias COMPLEX*32 is a REAL128 complex
floating-point number, which is supported on the host – but not on the offload
target (here: nvptx-none).
I think the output by LTO is correct but not very helpful:
lto1: fatal error: unsupported mode 'XF' (for long double)
lto1: fatal error: unsupported mode 'TF' (for __float128)
I think when generating code for offloading, this must be more explicit like:
nvptx-none does not support 80-bit floating-point numbers
nvptx-none does not support 128-bit floating-point numbers
Or something along those lines which actually helps the user to understand what
goes wrong.
Richard, Jakub: Thoughts?
C example – untested – which illustrates the issue and can be turned into a
test case, once LTO has a nicer error.
-----------------------------------
/* { dg-do link } */
/* { dg-require-effective-target large_long_double } */
/* { dg-require-effective-target offload_device } */
/* { dg-error "fatal error: unsupported mode 'XF'" */
#include <stdlib.h> /* For abort. */
long double foo (long double x)
{
#pragma omp target map(tofrom:x)
x *= 2.0;
return x;
}
int main
{
long double v = foo (10.0q) - 20.0q;
if (v > 1.0e-5 || v < -1.0e-5) abort();
return 0;
}
-----------------------------------
/* { dg-do link { target __float128 } } */
/* { dg-add-options __float128 } */
/* { dg-require-effective-target offload_device } */
/* { dg-error "fatal error: unsupported mode 'TF'" */
__float128 foo(__float128 y)
{
#pragma omp target map(tofrom: y)
y *= 4.0;
return y;
}
int main
{
__float128 v = foo (5.0L) - 20.0L;
if (v > 1.0e-5 || v < -1.0e-5) abort();
return 0;
}
-----------------------------------
(Assumes that for offload_device, neither data type is available; if it is, one
needs to exclude those offload-targets.)
> 2. CHARACTER default mapping not correct according to the specs.
See PR 92920.
