<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][AVX] v8i32 SETCC promotion causes unnecessary truncation/extensions"
href="https://bugs.llvm.org/show_bug.cgi?id=42025">42025</a>
</td>
</tr>
<tr>
<th>Summary</th>
<td>[X86][AVX] v8i32 SETCC promotion causes unnecessary truncation/extensions
</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: X86
</td>
</tr>
<tr>
<th>Assignee</th>
<td>unassignedbugs@nondot.org
</td>
</tr>
<tr>
<th>Reporter</th>
<td>llvm-dev@redking.me.uk
</td>
</tr>
<tr>
<th>CC</th>
<td>craig.topper@gmail.com, lebedev.ri@gmail.com, llvm-bugs@lists.llvm.org, llvm-dev@redking.me.uk, nikita.ppv@gmail.com, spatel+llvm@rotateright.com
</td>
</tr></table>
<p>
<div>
<pre>define i8 @v8i32_or_and(<8 x i32> %a, <8 x i32> %b, <8 x i32> %c, <8 x i32> %d,
<8 x i32> %e, <8 x i32> %f) {
%x0 = icmp sgt <8 x i32> %a, %b
%x1 = icmp slt <8 x i32> %c, %d
%x2 = icmp eq <8 x i32> %e, %f
%y = or <8 x i1> %x0, %x1
%z = and <8 x i1> %y, %x2
%res = bitcast <8 x i1> %z to i8
ret i8 %res
}
<a href="https://godbolt.org/z/Pn8KgQ">https://godbolt.org/z/Pn8KgQ</a>
AVX2:
v8i32_or_and: # @v8i32_or_and
vpcmpgtd %ymm1, %ymm0, %ymm0
vpcmpgtd %ymm2, %ymm3, %ymm1
vpor %ymm1, %ymm0, %ymm0
vpcmpeqd %ymm5, %ymm4, %ymm1
vextracti128 $1, %ymm1, %xmm2
vpackssdw %xmm2, %xmm1, %xmm1
vextracti128 $1, %ymm0, %xmm2
vpackssdw %xmm2, %xmm0, %xmm0
vpand %xmm1, %xmm0, %xmm0
vpacksswb %xmm0, %xmm0, %xmm0
vpmovmskb %xmm0, %eax
vzeroupper
retq
But should simplify to:
v8i32_or_and: # @v8i32_or_and
vpcmpgtd %ymm1, %ymm0, %ymm0
vpcmpgtd %ymm2, %ymm3, %ymm1
vpor %ymm1, %ymm0, %ymm0
vpcmpeqd %ymm5, %ymm4, %ymm1
vpand %ymm0, %ymm0, %ymm0
vmovmskps %ymm0, %eax
vzeroupper
retq
AVX1 is trickier but should be able to simplify in a similar manner (but via
xmm PCMPGT/PCMPEQ ops):
v8i32_or_and: # @v8i32_or_and
vpcmpgtd %xmm1, %xmm0, %xmm6
vextractf128 $1, %ymm1, %xmm1
vextractf128 $1, %ymm0, %xmm0
vpcmpgtd %xmm1, %xmm0, %xmm0
vpcmpgtd %xmm2, %xmm3, %xmm1
vextractf128 $1, %ymm2, %xmm2
vextractf128 $1, %ymm3, %xmm3
vpcmpgtd %xmm2, %xmm3, %xmm2
vpor %xmm1, %xmm6, %xmm1
vpcmpeqd %xmm5, %xmm4, %xmm3
vpor %xmm2, %xmm0, %xmm0
vextractf128 $1, %ymm4, %xmm2
vpackssdw %xmm0, %xmm1, %xmm0
vextractf128 $1, %ymm5, %xmm1
vpcmpeqd %xmm1, %xmm2, %xmm1
vpackssdw %xmm1, %xmm3, %xmm1
vpand %xmm1, %xmm0, %xmm0
vpacksswb %xmm0, %xmm0, %xmm0
vpmovmskb %xmm0, %eax
retq</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>