<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 - Failure to hoist repeated vextractf128 from switch cases"
href="https://bugs.llvm.org/show_bug.cgi?id=38917">38917</a>
</td>
</tr>
<tr>
<th>Summary</th>
<td>Failure to hoist repeated vextractf128 from switch cases
</td>
</tr>
<tr>
<th>Product</th>
<td>new-bugs
</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>new bugs
</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>llvm-bugs@lists.llvm.org
</td>
</tr></table>
<p>
<div>
<pre>Simplified (masked store style) example:
<a href="https://gcc.godbolt.org/z/g7URz3">https://gcc.godbolt.org/z/g7URz3</a>
void store07(__m256i a, __m256i v, int *f) {
switch(_mm256_movemask_ps(a)) {
case 0xff:
f[7] = _mm256_extract_epi32(v, 7);
case 0x7f:
f[6] = _mm256_extract_epi32(v, 6);
case 0x3f:
f[5] = _mm256_extract_epi32(v, 5);
case 0x1f:
f[4] = _mm256_extract_epi32(v, 4);
case 0x0f:
f[3] = _mm256_extract_epi32(v, 3);
case 0x07:
f[2] = _mm256_extract_epi32(v, 2);
case 0x03:
f[1] = _mm256_extract_epi32(v, 1);
case 0x01:
f[0] = _mm256_extract_epi32(v, 0);
break;
}
}
_Z7store07Dv4_xS_Pi: # @_Z7store07Dv4_xS_Pi
vmovmskps %ymm0, %eax
cmpl $62, %eax
jg .LBB0_3
decl %eax
cmpl $30, %eax
ja .LBB0_14
jmpq *.LJTI0_0(,%rax,8)
.LBB0_3:
cmpl $63, %eax
je .LBB0_8
cmpl $255, %eax
je .LBB0_6
cmpl $127, %eax
je .LBB0_7
.LBB0_14:
retq
.LBB0_6:
vextractf128 $1, %ymm1, %xmm0
vextractps $3, %xmm0, 28(%rdi)
.LBB0_7:
vextractf128 $1, %ymm1, %xmm0
vextractps $2, %xmm0, 24(%rdi)
.LBB0_8:
vextractf128 $1, %ymm1, %xmm0
vextractps $1, %xmm0, 20(%rdi)
.LBB0_9:
vextractf128 $1, %ymm1, %xmm0
vmovss %xmm0, 16(%rdi)
.LBB0_10:
vextractps $3, %xmm1, 12(%rdi)
.LBB0_11:
vextractps $2, %xmm1, 8(%rdi)
.LBB0_12:
vextractps $1, %xmm1, 4(%rdi)
.LBB0_13:
vmovss %xmm1, (%rdi)
retq
Shouldn't CSE be able to hoist the repeated vextractf128 instructions out from
the switch cases?</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>