X86 JIT misencodes cbw and cwd

[llvmbot]

Bugzilla Link	1130
Resolution	FIXED
Resolved on	Feb 22, 2010 12:54
Version	trunk
OS	All
Attachments	Byte-code demonstrating the problem
Reporter	LLVM Bugzilla Contributor

Extended Description

$ lli f.bc
lli((anonymous namespace)::PrintStackTrace()+0x1a)[0x878977a]
lli((anonymous namespace)::SignalHandler(int)+0x112)[0x8789a40]
[0xb7f04420]
lli(llvm::ExecutionEngine::runFunctionAsMain(llvm::Function*,
std::vector<std::basic_string<char, std::char_traits,
std::allocator >, std::allocator<std::basic_string<char,
std::char_traits, std::allocator > > > const&, char const*
const*)+0x23a)[0x84cc066]
lli(main+0x2ba)[0x83b6056]
/lib/tls/i686/cmov/libc.so.6(__libc_start_main+0xdc)[0xb7c8cebc]
lli[0x83b58e1]
Floating point exception

vs

$ lli -force-interpreter f.bc
-128 rd 1 = 0

The C back-end agrees with lli -force-interpreter (and so do I!).
Running "bugpoint --run-jit f.bc" reduces it a bit.

I send the byte-code in an attachment in a moment.

PS: Probably I should report this for one of the libraries, but
it is not clear where x86 JIT bugs should go.

[llvmbot]

assigned to @lattner

[lattner]

Fixed, patch here:
http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20070122/043184.html

Very very nice catch, thanks!

-Chris

[lattner]

mentioned in issue #1335