<html>
<head>
<base href="https://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 --- - [mc] image_atomic -- issues with unorm (and other modifiers) and dmask."
href="https://llvm.org/bugs/show_bug.cgi?id=28239">28239</a>
</td>
</tr>
<tr>
<th>Summary</th>
<td>[mc] image_atomic -- issues with unorm (and other modifiers) and dmask.
</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>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>Backend: AMDGPU
</td>
</tr>
<tr>
<th>Assignee</th>
<td>artem.tamazov@amd.com
</td>
</tr>
<tr>
<th>Reporter</th>
<td>artem.tamazov@amd.com
</td>
</tr>
<tr>
<th>CC</th>
<td>llvm-bugs@lists.llvm.org, nikolay.haustov@amd.com
</td>
</tr>
<tr>
<th>Classification</th>
<td>Unclassified
</td>
</tr></table>
<p>
<div>
<pre>Created <span class=""><a href="attachment.cgi?id=16616" name="attach_16616" title="06156.tests_tg_gfx7_asm_lit1_bug.06.08.min.image_atomic_add_etc-modifiers-dmask.zip">attachment 16616</a> <a href="attachment.cgi?id=16616&action=edit" title="06156.tests_tg_gfx7_asm_lit1_bug.06.08.min.image_atomic_add_etc-modifiers-dmask.zip">[details]</a></span>
06156.tests_tg_gfx7_asm_lit1_bug.06.08.min.image_atomic_add_etc-modifiers-dmask.zip
It seems that unorm (and other modifiers) cannot be used without dmask:
<span class="quote">> image_atomic_add v0, v0, s[0:7] dmask:0xF unorm // ok
> image_atomic_add v0, v0, s[0:7] unorm // failed</span >
it is even more strange taking into account the fact that dmask may be omitted:
<span class="quote">> image_atomic_add v0, v0, s[0:7] // ok</span >
Another catch is that assembler produces the *same* encoding for V1 and V4
varieties of instructions, which is definitely wrong. Example:
<span class="quote">> image_atomic_umin v0, v0 , s[0:7] dmask:0x1 unorm glc ; encoding: [0x00,0x31,0x54,0xf0,0x00,0x00,0x00,0x00]
>image_atomic_umin v0, v[0:3], s[0:7] dmask:0x1 unorm glc ; encoding: [0x00,0x31,0x54,0xf0,0x00,0x00,0x00,0x00]</span >
The issue needs to be carefully verified/investigated.
Some GFX7 lit tests attached.
These tests are not guaranteed to be valid, though.
Tested with r273139 (git # 4a94acfea8bfd6eb6b5809dd40f03e1fa42e673a)</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>