https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123870

--- Comment #37 from Jeffrey A. Law <law at gcc dot gnu.org> ---
Given that I fixed at least one bug that looks like this:

2. GCC 15.2.0 -O2, crashes with ICE "validate_change_or_fail"

I'd like to go ahead and get a testcase for that carved out as a distinct bug. 
After some wandering there's a reasonable chance this is a duplicate of
pr120642.  Once I've got a testcase it should be fairly easy to determine if
it's a duplicate.

What is generally best for a bug of that nature is to extract the cpp output
rather than giving us a package to build.  Building packages relies on the
precise header files and such on any given system which is prone to differences
that can make a bug go latent.  It also means we've got to do at least some of
the reproduction steps on a native system (which are dramatically slower than
crosses using modern x86 hardware).

Instead I would suggest you do the build Ilya.  At some point you'll trip the
error on one or more input files.  Add "-save-temps" to whatever command line
is used to build the problematic input file.  That will generate a .i or .ii
file (for C and C++ respectively).  Include that generated .i/.ii file in the
new bug report along with the command line you used.

In that form the test is self-contained and we can debug with cross compilers
which is much faster than current RISC-V native debugging.

I would use the same proccess for:

3. GCC 14.3.0 -O2, uses invalid th.vmv2r instruction for xtheadvector

I know I fixed a bug of that nature as well.  IIRC it happened with
multiply-add/multiply-accumulate style instrutions.  The procedure above will
provide me with enough info to quickly determine if it's the same bug I already
fixed.

Reply via email to