<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 - [x86] try harder to reduce shuffle with zero operand"
href="https://bugs.llvm.org/show_bug.cgi?id=40318">40318</a>
</td>
</tr>
<tr>
<th>Summary</th>
<td>[x86] try harder to reduce shuffle with zero operand
</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>All
</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: X86
</td>
</tr>
<tr>
<th>Assignee</th>
<td>unassignedbugs@nondot.org
</td>
</tr>
<tr>
<th>Reporter</th>
<td>spatel+llvm@rotateright.com
</td>
</tr>
<tr>
<th>CC</th>
<td>craig.topper@gmail.com, llvm-bugs@lists.llvm.org, llvm-dev@redking.me.uk, spatel+llvm@rotateright.com
</td>
</tr></table>
<p>
<div>
<pre>As noted here:
<a href="https://reviews.llvm.org/D56281">https://reviews.llvm.org/D56281</a>
...we should be able to do better than this:
define <8 x i16> @shift_zext(<8 x i16> %x) {
%r = shufflevector <8 x i16> %x, <8 x i16> zeroinitializer, <8 x i32> <i32 1,
i32 9, i32 10, i32 11, i32 undef, i32 undef, i32 undef, i32 undef>
ret <8 x i16> %r
}
$ llc -o - shufsse.ll
pxor %xmm1, %xmm1
punpcklwd %xmm0, %xmm1 ## xmm1 =
xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
pshufhw $232, %xmm1, %xmm0 ## xmm0 = xmm1[0,1,2,3,4,6,6,7]
pshufd $230, %xmm0, %xmm0 ## xmm0 = xmm0[2,1,2,3]
pshuflw $75, %xmm0, %xmm0 ## xmm0 = xmm0[3,2,0,1,4,5,6,7]
retq
Or an even worse example:
define <8 x i16> @shift_zext_worse
(<8 x i16> %x) {
%r = shufflevector <8 x i16> %x, <8 x i16> zeroinitializer, <8 x i32> <i32 2,
i32 9, i32 10, i32 11, i32 undef, i32 undef, i32 undef, i32 undef>
ret <8 x i16> %r
}
$ llc -o - shufsse.ll
pxor %xmm1, %xmm1
punpcklwd %xmm0, %xmm1 ## xmm1 =
xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3]
pshufd $39, %xmm1, %xmm0 ## xmm0 = xmm1[3,1,2,0]
pshuflw $232, %xmm0, %xmm0 ## xmm0 = xmm0[0,2,2,3,4,5,6,7]
pshufd $232, %xmm0, %xmm0 ## xmm0 = xmm0[0,2,2,3]
pshuflw $39, %xmm0, %xmm0 ## xmm0 = xmm0[3,1,2,0,4,5,6,7]
retq
----------------------------------------------------------------------------
When shuffling with a zero operand, we should try harder to use logical shifts?
In all cases, 2 opposing shifts can put zero elements into all lanes.</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>