<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 - widening of atomic loads"
href="https://bugs.llvm.org/show_bug.cgi?id=33426">33426</a>
</td>
</tr>
<tr>
<th>Summary</th>
<td>widening of atomic loads
</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>richard-llvm@metafoo.co.uk
</td>
</tr>
<tr>
<th>CC</th>
<td>llvm-bugs@lists.llvm.org
</td>
</tr></table>
<p>
<div>
<pre>Testcase:
#include <atomic>
#include <cstring>
#include <x86intrin.h>
alignas(16) std::atomic<char> data[16];
void store(int n, char v) { data[n].store(std::memory_order_relaxed); }
__m128i load() {
char copy[16];
for (int n = 0; n != 16; ++n)
copy[n] = data[n].load(std::memory_order_relaxed);
__m128i result;
std::memcpy(&result, copy, 16);
return result;
}
When targeting x86_64, LLVM optimizes the 'load' function to a sequence of 16
load/store pairs:
movb data(%rip), %al
movb %al, -24(%rsp)
movb data+1(%rip), %al
movb %al, -23(%rsp)
movb data+2(%rip), %al
movb %al, -22(%rsp)
movb data+3(%rip), %al
movb %al, -21(%rsp)
movb data+4(%rip), %al
movb %al, -20(%rsp)
movb data+5(%rip), %al
movb %al, -19(%rsp)
movb data+6(%rip), %al
movb %al, -18(%rsp)
movb data+7(%rip), %al
movb %al, -17(%rsp)
movb data+8(%rip), %al
movb %al, -16(%rsp)
movb data+9(%rip), %al
movb %al, -15(%rsp)
movb data+10(%rip), %al
movb %al, -14(%rsp)
movb data+11(%rip), %al
movb %al, -13(%rsp)
movb data+12(%rip), %al
movb %al, -12(%rsp)
movb data+13(%rip), %al
movb %al, -11(%rsp)
movb data+14(%rip), %al
movb %al, -10(%rsp)
movb data+15(%rip), %al
movb %al, -9(%rsp)
movaps -24(%rsp), %xmm0
retq
The function should be optimized to a single 16-byte load.</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>