<html>
<head>
<base href="https://bugs.llvm.org/">
</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 - miscompilation of small integer bit math"
href="https://bugs.llvm.org/show_bug.cgi?id=34349">34349</a>
</td>
</tr>
<tr>
<th>Summary</th>
<td>miscompilation of small integer bit math
</td>
</tr>
<tr>
<th>Product</th>
<td>libraries
</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>enhancement
</td>
</tr>
<tr>
<th>Priority</th>
<td>P
</td>
</tr>
<tr>
<th>Component</th>
<td>Scalar Optimizations
</td>
</tr>
<tr>
<th>Assignee</th>
<td>unassignedbugs@nondot.org
</td>
</tr>
<tr>
<th>Reporter</th>
<td>andrew.b.adams@gmail.com
</td>
</tr>
<tr>
<th>CC</th>
<td>llvm-bugs@lists.llvm.org
</td>
</tr></table>
<p>
<div>
<pre>Sometime in between 311774 and 311796 the following ll started producing
different results with and without running opt. Perhaps r311789 uncovered
something? It's a routine to do unsigned 8-bit int division by 201 using bit
tricks. I do not believe it exploits any UB.
Before opt:
define i8 @fast_div_201(i8 %p) {
entry:
%v3 = zext i8 %p to i16
%v4 = mul i16 %v3, 71
%v5 = lshr i16 %v4, 8
%v6 = trunc i16 %v5 to i8
%v7 = sub i8 %p, %v6
%v8 = lshr i8 %v7, 1
%v9 = zext i8 %p to i16
%v10 = mul i16 %v9, 71
%v11 = lshr i16 %v10, 8
%v12 = trunc i16 %v11 to i8
%v13 = add i8 %v12, %v8
%v14 = lshr i8 %v13, 7
ret i8 %v14
}
opt -O3 with trunk unconditionally returns zero:
define i8 @fast_div_201(i8 %p) local_unnamed_addr #0 {
entry:
ret i8 0
}
opt -O3 with llvm 3.8 just cleans up the common subexpressions:
define i8 @fast_div_201(i8 %p) #0 {
entry:
%v3 = zext i8 %p to i16
%v4 = mul nuw nsw i16 %v3, 71
%v5 = lshr i16 %v4, 8
%v6 = trunc i16 %v5 to i8
%v7 = sub i8 %p, %v6
%v8 = lshr i8 %v7, 1
%v13 = add nuw i8 %v8, %v6
%v14 = lshr i8 %v13, 7
ret i8 %v14
}
As an example for why zero is the wrong output, I'll plug in the largest
possible value of the input (255) and see what each value is:
define i8 @fast_div_201(i8 %p) {
entry:
%v3 = zext i8 %p to i16 // u16(255)
%v4 = mul i16 %v3, 71 // u16(18105)
%v5 = lshr i16 %v4, 8 // u16(70)
%v6 = trunc i16 %v5 to i8 // u8(70)
%v7 = sub i8 %p, %v6 // u8(185)
%v8 = lshr i8 %v7, 1 // u8(92)
%v9 = zext i8 %p to i16 // u16(255)
%v10 = mul i16 %v9, 71 // u16(18105)
%v11 = lshr i16 %v10, 8 // u16(70)
%v12 = trunc i16 %v11 to i8 // u8(70)
%v13 = add i8 %v12, %v8 // u8(162)
%v14 = lshr i8 %v13, 7 // u8(1)
ret i8 %v14 // 255/201 == 1
}</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>