https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120447
--- Comment #9 from Richard Sandiford <rsandifo at gcc dot gnu.org> ---
I think the ICE is caused by a bad interaction between the AArch64 optimisation
and r16-718, which added more checks for paradoxical subregs. The testcase
works if r16-718 is reverted.
I think the should go with something like this:
diff --git a/gcc/emit-rtl.cc b/gcc/emit-rtl.cc
index 3f453cda67e..df7fa912ebb 100644
--- a/gcc/emit-rtl.cc
+++ b/gcc/emit-rtl.cc
@@ -998,10 +998,11 @@ validate_subreg (machine_mode omode, machine_mode imode,
&& known_le (osize, isize))
return false;
- /* The outer size must be ordered wrt the register size, otherwise
- we wouldn't know at compile time how many registers the outer
- mode occupies. */
- if (!ordered_p (osize, regsize))
+ /* If the inner size is greater than the register size, the inner value
+ is split into blocks of size REGSIZE. The outer size must then be
+ ordered wrt the register size, otherwise we wouldn't know at compile
+ time how many blocks the outer mode occupies. */
+ if (maybe_ge (isize, regsize) && !ordered_p (osize, regsize))
return false;
/* For normal pseudo registers, we want most of the same checks. Namely:
Mind if I take the PR to test that?
