https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114523
--- Comment #13 from Jose E. Marchesi <jemarch at gcc dot gnu.org> ---
Thanks. The new title is way better. And thank you for the further analysis
and the reproducer that also makes clang to generate the no-verifiable code!
I wonder, is the issue also there when -mno-alu32 is used? In that case the
generated code doesn't involve "subregs" (or 32-bit operations in assembly-like
syntax):
.file "foo.c"
.text
.align 3
.global foo
.type foo, @function
foo:
call bar
lddw %r1,baz
mov %r0,%r0
and %r0,0xffffffff
ldxw %r2,[%r1+0]
add %r0,-1
neg %r2
xor %r0,%r2
rsh %r0,63
exit
.size foo, .-foo
.global baz
.type baz, @object
.lcomm baz,4,4
.ident "GCC: (GNU) 14.0.1 20240226 (experimental)"
Cuper, is the verifier able to track proper values through the xor in this
case?
