<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 --- - inline asm incorrectly handles output operands sometimes"
href="http://llvm.org/bugs/show_bug.cgi?id=21273">21273</a>
</td>
</tr>
<tr>
<th>Summary</th>
<td>inline asm incorrectly handles output operands sometimes
</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>All
</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>dimitry@andric.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>Depending on the optimization level, clang trunk r219624 seems to sometimes
handle output operands incorrectly. This was reported to me by FreeBSD kernel
developers, who attempted to compile the following:
int ivy_rng_store(long *buf)
{
long tmp;
int retry;
retry = 10;
__asm __volatile(
"1:\n\t"
"rdrand %2\n\t" /* read randomness into tmp */
"jb 2f\n\t" /* CF is set on success, exit retry loop */
"dec %0\n\t" /* otherwise, retry-- */
"jne 1b\n\t" /* and loop if retries are not exhausted */
"jmp 3f\n" /* failure, retry is 0, used as return value */
"2:\n\t"
"mov %2,%1\n\t" /* *buf = tmp */
"3:"
: "+q" (retry), "=m" (*buf), "=q" (tmp) : : "cc");
return (retry);
}
E.g., the intent is that 'tmp' is just used for output, but the actual value is
not used outside the inline asm. It is stored to *buf instead.
However, clang -O0 seems to have trouble keeping the two apart, as the
resulting assembly is:
ivy_rng_store: # @ivy_rng_store
.cfi_startproc
# BB#0: # %entry
pushq %rbp
.Ltmp0:
.cfi_def_cfa_offset 16
.Ltmp1:
.cfi_offset %rbp, -16
movq %rsp, %rbp
.Ltmp2:
.cfi_def_cfa_register %rbp
movq %rdi, -8(%rbp)
movl $10, -20(%rbp)
movl -20(%rbp), %eax
movq -8(%rbp), %rdi
#APP
.Ltmp3:
rdrandq %rdi
jb .Ltmp4
decl %eax
jne .Ltmp3
jmp .Ltmp5
.Ltmp4:
movq %rdi, (%rdi)
.Ltmp5:
#NO_APP
movl %eax, -20(%rbp)
movq %rdi, -16(%rbp)
movl -20(%rbp), %eax
popq %rbp
retq
Clearly, the movq %rdi, (%rdi) is incorrect. This seems to be magically solved
by enabling optimization, e.g. at -O1 or higher:
ivy_rng_store: # @ivy_rng_store
.cfi_startproc
# BB#0: # %entry
pushq %rbp
.Ltmp0:
.cfi_def_cfa_offset 16
.Ltmp1:
.cfi_offset %rbp, -16
movq %rsp, %rbp
.Ltmp2:
.cfi_def_cfa_register %rbp
movl $10, %eax
#APP
.Ltmp3:
rdrandq %rcx
jb .Ltmp4
decl %eax
jne .Ltmp3
jmp .Ltmp5
.Ltmp4:
movq %rcx, (%rdi)
.Ltmp5:
#NO_APP
popq %rbp
retq
Something similar happens when targeting i386 at -O0:
ivy_rng_store: # @ivy_rng_store
# BB#0: # %entry
pushl %ebp
movl %esp, %ebp
subl $12, %esp
movl 8(%ebp), %eax
movl %eax, -4(%ebp)
movl $10, -12(%ebp)
movl -12(%ebp), %eax
movl -4(%ebp), %ecx
#APP
.Ltmp0:
rdrandl %ecx
jb .Ltmp1
decl %eax
jne .Ltmp0
jmp .Ltmp2
.Ltmp1:
movl %ecx, (%ecx)
.Ltmp2:
#NO_APP
movl %eax, -12(%ebp)
movl %ecx, -8(%ebp)
movl -12(%ebp), %eax
addl $12, %esp
popl %ebp
retl
However, on i386 optimization does not fix it, e.g. at -O1 or higher:
ivy_rng_store: # @ivy_rng_store
# BB#0: # %entry
pushl %ebp
movl %esp, %ebp
movl 8(%ebp), %ecx
movl $10, %eax
#APP
.Ltmp0:
rdrandl %ecx
jb .Ltmp1
decl %eax
jne .Ltmp0
jmp .Ltmp2
.Ltmp1:
movl %ecx, (%ecx)
.Ltmp2:
#NO_APP
popl %ebp
retl
Changing the output constraint on 'tmp' to "+q" seems to help on amd64, but on
i386 it still produces incorrect output at -O1 or higher optimization.
I tested the above code with different versions of gcc (4.7 through 5.0), but
the resulting assembly was always as expected, at any optimization level.</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>