<html>
<head>
<base href="http://llvm.org/bugs/" />
</head>
<body><table border="1" cellspacing="0" cellpadding="8">
<tr>
<th>Bug ID</th>
<td><a class="bz_bug_link
bz_status_NEW "
title="NEW --- - [x86_64] CodeGen generates an instruction that is impossible to encode in x86_64."
href="http://llvm.org/bugs/show_bug.cgi?id=17902">17902</a>
</td>
</tr>
<tr>
<th>Summary</th>
<td>[x86_64] CodeGen generates an instruction that is impossible to encode in x86_64.
</td>
</tr>
<tr>
<th>Product</th>
<td>new-bugs
</td>
</tr>
<tr>
<th>Version</th>
<td>trunk
</td>
</tr>
<tr>
<th>Hardware</th>
<td>PC
</td>
</tr>
<tr>
<th>OS</th>
<td>Linux
</td>
</tr>
<tr>
<th>Status</th>
<td>NEW
</td>
</tr>
<tr>
<th>Severity</th>
<td>normal
</td>
</tr>
<tr>
<th>Priority</th>
<td>P
</td>
</tr>
<tr>
<th>Component</th>
<td>new bugs
</td>
</tr>
<tr>
<th>Assignee</th>
<td>unassignedbugs@nondot.org
</td>
</tr>
<tr>
<th>Reporter</th>
<td>madhusudancs@gmail.com
</td>
</tr>
<tr>
<th>CC</th>
<td>llvmbugs@cs.uiuc.edu
</td>
</tr>
<tr>
<th>Classification</th>
<td>Unclassified
</td>
</tr></table>
<p>
<div>
<pre>Created <span class=""><a href="attachment.cgi?id=11527" name="attach_11527" title="C snippet to reproduce the bug">attachment 11527</a> <a href="attachment.cgi?id=11527&action=edit" title="C snippet to reproduce the bug">[details]</a></span>
C snippet to reproduce the bug
I ran into a subtle bug related to x86_64 instruction encoding when using LLVM
JIT. (I am using LLVM 3.3 release version. There is no 3.3 in the Version field
above, so I have chosen trunk as the version).
I am attaching the super-reduced test case (the C file) that reproduces the
bug. Just to be on the same page, I ran the following commands to obtain the
.ll, .o and .s files:
$ clang -cc1 -O1 -emit-llvm x86encodingbug.c
$ llc -O0 --regalloc=fast -filetype=obj x86encodingbug.ll
$ llc -O0 --regalloc=fast -filetype=asm x86encodingbug.ll
If we compare line x86encodingbug.s:62 with the equivalent instruction in the
x86encodingbug.o file (through objdump -S) the bug is kind of evident.
Intended instruction: movb %ah, (%rdi,%r8)
Actual instruction: mov %spl,(%rdi,%r8,1)
Note the %ah vs %spl register.
Running this code through lli using the following command line arguments:
$ lli -O0 --regalloc=fast x86encodingbug.ll 0 a
String: �
where the expected result was "String: 7" (ASCII code of 'a' is 97).
Notes in page 977 (or Vol. 2A 3-503) of Intel's manual:
<a href="http://www.intel.com/content/dam/www/public/us/en/documents/manuals/64-ia-32-architectures-software-developer-manual-325462.pdf">http://www.intel.com/content/dam/www/public/us/en/documents/manuals/64-ia-32-architectures-software-developer-manual-325462.pdf</a>
states that this instruction cannot be encoded:
"""
***In 64-bit mode, r/m8 can not be encoded to access the following byte
registers if a REX prefix is used: AH, BH, CH, DH.
"""
x86's RegisterInfo Target Descriptor
(llvm/lib/Target/X86/X86RegisterInfo.td:60) also mentions this and yet an
instruction which cannot be encoded is generated.
This was probably a regression introduced by the commits that fixed <a class="bz_bug_link
bz_status_RESOLVED bz_closed"
title="RESOLVED FIXED - llc -O3: invalid assembler: Error: can't encode register '%dh' in an instruction requiring REX prefix."
href="show_bug.cgi?id=6151">bug #6151</a>.</pre>
</div>
</p>
<hr>
<span>You are receiving this mail because:</span>
<ul>
<li>You are on the CC list for the bug.</li>
</ul>
</body>
</html>