Cameron Kaiser
Cameron Kaiser
> Checked for prior ppcle64 tickets, like #2789: looks like ppcle64 was healthy as of Sept/Oct 2023 timeframe. > > Maybe a regression you (or @classilla) can bisect or isolate...
@runlevel5 , can you try compiling it with dynrec off and PIE on so I can narrow down whether the problem is the JIT or C code? I'm off site...
What @weirddan455 said. My guess is that we have hardcoded addresses I'll need to deal with.
Thank you, @runlevel5 . I will do some further analysis this weekend, but I think I have an idea what's wrong.
Just an update: this is stalled because tip FTBFS due to a vendored libpng problem (not properly including VSX files leading to link failure). System libs work. I am working...
Upstreamed `libpng` fix to https://github.com/mesonbuild/wrapdb/pull/1538
Okay, it *wasn't* what I thought. It turns out this is a very old bug that PIE just happens to unmask. It does not affect the 32-bit PPC dynrec. @runlevel5...
@oturpe , this fix can be cherry-picked and applied to prior versions; it does not have any dependencies.
Probably should always be an option, not the default, since it wouldn't be backwards compatible (the documentation even mentions that the pseudo-ops don't issue these instructions, so having them do...
No, I understand that. I'm concerned that's not backwards compatible either. I'm trying to think of a situation where you change the processor bit with (RS)EP but succeeding code assumes...