https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96700
Jim Wilson <wilson at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Resolution|--- |INVALID
Status|UNCONFIRMED |RESOLVED
--- Comment #3 from Jim Wilson <wilson at gcc dot gnu.org> ---
I took another look at this.
I can get this to fail with an x86_64 gcc, but that is the system compiler
which is gcc-7. A little testing shows that the testcase works for x86_64
gcc-8 and later and fails for gcc-7. However, since this is a testcase for
gcc-11, there is no expectation that it should work with older gcc versions.
We sometimes need to modify testcases to work when gcc changes.
I also tried this with rv32-elf compilers, gcc-11, gcc-10, gcc-9, and I
couldn't get the testcase to fail for any of them, with or without -flto.
So I can't reproduce the original poster's problem, which gets me back to my
original claim that I don't think that this is a gcc bug.
Actually, on second thought, I realized that I'm using the original testcase
from the gcc source tree gcc.dg/tree-ssa/buitlin-sprintf.c not the attachment.
Looking at the attachment, I see that some numbers have been modified. So that
is why it fails. It is a bogus testcase. Looking further, I see that you are
using an old version of the testcase before it was changed in 2018. See
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86274
which both changed an optimization and changed the testcase to match the new
optimization. The fix was put on mainlnie (gcc-9) and backported to gcc-8, so
the testcase should fail for any gcc-8 or later.
This is invalid. You can't take a testcase from gcc-X and expect it to work
with gcc-Y.
