<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 - Make -thinlto-index-only work well with thin static archives"
href="https://bugs.llvm.org/show_bug.cgi?id=42717">42717</a>
</td>
</tr>
<tr>
<th>Summary</th>
<td>Make -thinlto-index-only work well with thin static archives
</td>
</tr>
<tr>
<th>Product</th>
<td>lld
</td>
</tr>
<tr>
<th>Version</th>
<td>unspecified
</td>
</tr>
<tr>
<th>Hardware</th>
<td>PC
</td>
</tr>
<tr>
<th>OS</th>
<td>Windows NT
</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>All Bugs
</td>
</tr>
<tr>
<th>Assignee</th>
<td>akhuang@google.com
</td>
</tr>
<tr>
<th>Reporter</th>
<td>rnk@google.com
</td>
</tr>
<tr>
<th>CC</th>
<td>llvm-bugs@lists.llvm.org, llvm@inglorion.net, peter.smith@linaro.org
</td>
</tr></table>
<p>
<div>
<pre>Many build systems (such as LLVM's) are organized as collections of static
archives. Currently, the recommended way to use -thinlto-index-only to
distribute LTO backend actions is to use --start-lib / --end-lib on the command
line:
<a href="http://lists.llvm.org/pipermail/llvm-dev/2019-June/133145.html">http://lists.llvm.org/pipermail/llvm-dev/2019-June/133145.html</a>
To ease adoption of thinlto in legacy build systems, LLD should at least handle
thin archives, if not true static archives. There is a discussion of some of
these issues and how to solve them here:
<a href="https://reviews.llvm.org/D64461#1586526">https://reviews.llvm.org/D64461#1586526</a>
It boils down to coming up with a good path for the .thinlto.bc file calculated
here:
<a href="http://llvm-cs.pcc.me.uk/lib/LTO/LTO.cpp#1206">http://llvm-cs.pcc.me.uk/lib/LTO/LTO.cpp#1206</a>
With a very Chromium / LLVM perspective on things, I think the ideal index file
name would be something like:
${pathtoobj}/${object_noext}.o # stage1 output
${pathtoobj}/${object_noext}.${binary_noext}.thinlto.bc # index file
${pathtoobj}/${object_noext}.${binary_noext}.imports # import list
${pathtoobj}/${object_noext}.${binary_noext}.o # native output
So, for chromium on Windows, we'd have:
obj/v8/v8_libbase/bits.obj
obj/v8/v8_libbase/bits.d8.thinlto.bc
obj/v8/v8_libbase/bits.d8.o
obj/v8/v8_libbase/bits.chrome_child.thinlto.bc
obj/v8/v8_libbase/bits.chrome_child.o
... one per thinlto binary
I (think?) today the current behavior is that every link will attempt to write
obj/v8/v8_libbase/bits.thinlto.bc, which won't work if the same object is
compiled into two binaries.
Amy expressed interest in working on this.</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>