https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94611
Bug ID: 94611
Summary: gccgo hangs (infinite loop) on complex projects,
seemingly in simplify-rtx.c/simplify_plus_minus
Product: gcc
Version: 9.3.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: go
Assignee: ian at airs dot com
Reporter: gcc at octaforge dot org
CC: cmang at google dot com
Target Milestone: ---
Created attachment 48283
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=48283&action=edit
debugger steps
Attempting to compile complicated projects, such as bootstrapping the official
Go compiler (regardless of version), using gccgo will result in the gc1 process
entering infinite loop.
Configuration:
- GCC 9.3.0 (as far as I know, all 9.x series are known to do this, but I had
it work on some version of 8.x some time ago)
- ppc64le architecture (seems to affect at very least all other PowerPC)
- Linux
The gc1 process will keep running forever and keep growing in RSS.
By attaching into the process, I was able to narrow it down to
'simplify_plus_minus' from simplify-rtx.c.
Unfortunately, I wasn't able to isolate a trivial test case. It seems to happen
when building any sufficiently complex project. Trivial things build alright,
and gccgo can also bootstrap itself (when building a cross-toolchain).
I've never seen this happen with C/C++, so I suspect the gccgo frontend is
generating bogus data that the backend can't process.
Attached is the debugger output stepping through the call, one whole iteration,
on GCC 9.3.0.
