https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116145
Richard Sandiford <rsandifo at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |rsandifo at gcc dot gnu.org
--- Comment #6 from Richard Sandiford <rsandifo at gcc dot gnu.org> ---
The problem seems to be that may_trap_or_fault_p returns true for the memory,
which stops loop2_invariant from hoisting the load. The code there is:
if (/* MEM_NOTRAP_P only relates to the actual position of the memory
reference; moving it out of context such as when moving code
when optimizing, might cause its address to become invalid. */
code_changed
|| !MEM_NOTRAP_P (x))
{
poly_int64 size = MEM_SIZE_KNOWN_P (x) ? MEM_SIZE (x) : -1;
return rtx_addr_can_trap_p_1 (XEXP (x, 0), 0, size,
GET_MODE (x), code_changed);
}
(where code_changed is true). Since the SVE load doesn't support the full
LO_SUM range, there is no hint that the address refers to invariant memory.
I'm not sure the comment in the code is correct for MEM_READONLY_P, which only
refers to _statically allocated_ readonly memory. (At least, the comment
doesn't apply to constant memory created by force_const_mem, which creates
things that cannot trap regardless of context.) Adding:
if (MEM_READONLY_P (x) && MEM_NOTRAP_P (x))
return false;
makes the test work for me.