On 10/21/21 6:10 PM, Tom Kacvinsky via Gcc wrote:
On Thu, Oct 21, 2021 at 8:06 PM Martin Sebor via Gcc <gcc@gcc.gnu.org>
wrote:
I put #pragma GCC optimize "0" at the top of gimplify.c to help
me debug something in a bootstrapped compiler. The file failed
to compile with many assembler errors like this:
/tmp/ccL9zcXD.s: Assembler messages:
/tmp/ccL9zcXD.s:9: Error: CFI instruction used without previous
.cfi_startproc
I've done this before and had no problems. Is this supposed to
work or was I just lucky when it did before?
I know the binutils people have been doing work with CFI stuff, so perhaps
this is a binutils
issue? Which version of binutils are you using? A newer version
of binutils (or perhaps older)
might make a difference.
I don't think the problem is Binutils. The assembly emitted
by GCC changes with the pragma: the functions are missing
.cfi_startproc and .cfi_endproc directives, but have other
.cfi directives. This happens even when I remove the #pragma
from the file, and the only difference is that I'm compiling
the file with a compiler that includes the same file compiled
with #pragma optimize "0". So GCC emits different assembly
depending on whether or not its gimplify.c was compiled
optimized. That seems like a symptom of miscompilation
to me. The GCC binary, by the way, is stage 2.
Martin
Regards,
Tom
