https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67278
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Keywords| |ice-checking
Target| |x86_64-*-*
Status|NEW |ASSIGNED
Component|middle-end |target
Assignee|unassigned at gcc dot gnu.org |rguenth at gcc dot
gnu.org
--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
Ok, looks like the ABI passes V1XF in 16bytes on the stack which means this is
a verifier issue (works "correctly" w/o checking).
On i?86 you'll get
t.c:1:1: error: vector size not an integral multiple of component size
typedef long double a __attribute__((vector_size (16)));
^~~~~~~
Changing the testcase to
typedef long double a __attribute__((vector_size (32)));
a
sum (a first, a second)
{
return first + second;
}
and thus trying to create V2XF yields
t.c: In function ?sum?:
t.c:5:1: internal compiler error: in type_natural_mode, at
config/i386/i386.c:7890
{
^
0x1211845 type_natural_mode
/space/rguenther/src/svn/trunk3/gcc/config/i386/i386.c:7890
0x1215c87 ix86_return_in_memory
/space/rguenther/src/svn/trunk3/gcc/config/i386/i386.c:9765
0xaadcc7 aggregate_value_p(tree_node const*, tree_node const*)
/space/rguenther/src/svn/trunk3/gcc/function.c:2081
0xab677c allocate_struct_function(tree_node*, bool)
/space/rguenther/src/svn/trunk3/gcc/function.c:4914
0x76cc78 store_parm_decls()
I'm going to fix the verifier and leave the above to target maintainers.
Looks like we have no way of rejecting component types by the backend here btw.
