[Bug target/57896] [4.7/4.8/4.9 Regression] ICE in expand_expr_real_2

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=57896

--- Comment #29 from Uroš Bizjak <ubizjak at gmail dot com> ---
(In reply to Jakub Jelinek from comment #28)
> Perhaps just make it gcc_checking_assert?

On release branches perhaps, but the consequences of
crtl->emit.regno_pointer_align_length == 0 warrant ICE even in the released
compiler. We are talking about "ICE in gen_reg_rtx ..." vs. "some random ICE
due to corrupted internal data" (as is this PR) reports. From maintainer's POV,
I'd like to have gcc_assert in development and release branches.