[Bug lto/86344] GCC 8.1 ICEd at LTO stage
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86344 Richard Biener changed: What|Removed |Added Status|WAITING |RESOLVED Resolution|--- |WORKSFORME --- Comment #5 from Richard Biener --- Reporter says he figured a way that works for him. Since GCC 8 is no longer maintained, lets close this.
[Bug lto/86344] GCC 8.1 ICEd at LTO stage
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86344 --- Comment #4 from Oliver Ddin --- So, I grabbed GCC8.1.1 from trunk and built it. Crash reoccured. However, I noticed that GCC wasn't doing LTO through the linker plugin, so after rebuilding with the correct configuration to enable -fuse-linker-plugin to work, I then rebuilt the project and this time, LTO succeeded without error. So, I suspect this is specifically a bug that manifests when doing LTO without using the linker plugin. I unfortunately don't have anything better for you at this point and since I now have a working build, I'll probably not get the time to find a minimal case that reproduces this in the non-linker-plugin scenario.
[Bug lto/86344] GCC 8.1 ICEd at LTO stage
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86344 --- Comment #3 from Andrew Pinski --- One thing which you can do (note on the trunk, you need to add an extra option; I don't remember what it is currently), is link with -r and reduce the number of object files to the ones which crash. And then add -save-temps to the orginal gcc and reduce the preprocessed source using multi-delta (or other tools).
[Bug lto/86344] GCC 8.1 ICEd at LTO stage
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86344 --- Comment #2 from Richard Biener --- You should probably also try the HEAD from the gcc 8 branch, a bug with similar symptoms has been fixed recently.
[Bug lto/86344] GCC 8.1 ICEd at LTO stage
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86344 Martin Liška changed: What|Removed |Added Status|UNCONFIRMED |WAITING Last reconfirmed||2018-06-28 Ever confirmed|0 |1 --- Comment #1 from Martin Liška --- Ok, if you add --save-temps --verbose. You should see command line for lto1 invocation. Then it would be good to put a breakpoint and dump the symbol which triggers the assert: node1->debug() Anyway, a reduced test-case would be best to handle it.