Bugzilla – Bug 313
[X86] JIT miscompiles unsigned short to floating point cast
Last modified: 2004-04-06 14:30:54
You need to log in before you can comment on or make changes to this bug.
Wow, this bug has been here for a LONG time. This is causing cfrac to fail with the JIT, and maybe 'ed' as well. The problem is that the X86 backend is emitting a movzx instruction from 8 to 16 bits instead of 16 to 32 bits. Because it puts the right registers into the instruction, LLC works fine (one movzx is the same as another movzx), but the encoding of the instructions are different, so it fails in the JIT. -Chris
Fix here: http://mail.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20040405/013549.html MallocBench/cfrac now works with the X86 JIT. Thanks to bugpoint and the loop extractor for reducing this down to something that I could handle. :) -Chris