https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109128
--- Comment #2 from Tobias Burnus <burnus at gcc dot gnu.org> ---
(In reply to Thomas Schwinge from comment #1)
> See also "Allow the accelerator to have more offloaded functions than the
> host".
Which was:
- if (num_target_entries != num_funcs + num_vars)
+ if (num_target_entries < num_funcs + num_vars)
I think this approach only works if functions are missing "at the end".
The code does:
gomp_load_image_to_device (..., const void *host_table,
const void *target_data, ...)
...
void **host_func_table = ((void ***) host_table)[0];
void **host_funcs_end = ((void ***) host_table)[1];
int num_funcs = host_funcs_end - host_func_table;
...
num_target_entries
= devicep->load_image_func (devicep->target_id, version,
target_data, &target_table,
...
k->host_start = (uintptr_t) host_func_table[i];
...
k->tgt_offset = target_table[i].start;
Thus, it strictly depends on the order. That is: If some function "in the
middle" is not linked in / part of 'host_table' alias '__OFFLOAD_TABLE__', all
remaining host<->device function and/or variable assignments are messed up.
