On Mar 15, 2011, at 1:50 PM, Andreas Färber wrote:> Hello, > > Running the Debug+Asserts version instead of the Release should give you the file and line of the "unreachable" code. > > AndreasThanks for the suggestion. I recompiled with Debug+Asserts and it shows that the unreachable is in CallingConvLower.cpp:162. Here is the error message: Call result #2 has unhandled type f32UNREACHABLE executed at CallingConvLower.cpp:162! I don't think my backend is modified enough from the MBlaze backend that is in the release to be causing this error. I am however looking through the various files of the backend to try to find where the calling convention might be causing problems with f32 data types. Josef -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20110315/eb02e676/attachment.html>
> I don't think my backend is modified enough from the MBlaze backend that is > in the release to be causing this error. I am however looking through the > various files of the backend to try to find where the calling convention > might be causing problems with f32 data types.Form the backtrace it seems like you haven't defined how to return f32 stuff out of the function. -- With best regards, Anton Korobeynikov Faculty of Mathematics and Mechanics, Saint Petersburg State University
>> I don't think my backend is modified enough from the MBlaze backend that is >> in the release to be causing this error. I am however looking through the >> various files of the backend to try to find where the calling convention >> might be causing problems with f32 data types. > Form the backtrace it seems like you haven't defined how to return f32 > stuff out of the function.Could it be that the microblaze backend only has 2 return registers and we've written a function call that wants 3? If that could be the problem I'd guess it is highly likely because a lot of our code if 3 element vectors being passed around. Josef