<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 --- - SLP Vectorizer too aggressive with simple stores"
href="https://llvm.org/bugs/show_bug.cgi?id=31116">31116</a>
</td>
</tr>
<tr>
<th>Summary</th>
<td>SLP Vectorizer too aggressive with simple stores
</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>normal
</td>
</tr>
<tr>
<th>Priority</th>
<td>P
</td>
</tr>
<tr>
<th>Component</th>
<td>Scalar Optimizations
</td>
</tr>
<tr>
<th>Assignee</th>
<td>unassignedbugs@nondot.org
</td>
</tr>
<tr>
<th>Reporter</th>
<td>matze@braunis.de
</td>
</tr>
<tr>
<th>CC</th>
<td>llvm-bugs@lists.llvm.org
</td>
</tr>
<tr>
<th>Classification</th>
<td>Unclassified
</td>
</tr></table>
<p>
<div>
<pre>In a situation like the following (modeled after what happens in a C++
constructor with two base classes), the SLP vectorizer seems too aggressive:
struct C {
char *BaseA;
char *BaseB;
};
extern char VTab[];
void initC(C *c) {
c->BaseA = VTab + 48;
c->BaseB = VTab + 16;
}
The SLP Vectorizer transforms:
%BaseA = getelementptr inbounds %struct.C, %struct.C* %c, i64 0, i32 0
store i8* getelementptr inbounds ([0 x i8], [0 x i8]* @VTab, i64 0, i64 48),
i8** %BaseA, align 8, !tbaa !2
%BaseB = getelementptr inbounds %struct.C, %struct.C* %c, i64 0, i32 1
store i8* getelementptr inbounds ([0 x i8], [0 x i8]* @VTab, i64 0, i64 16),
i8** %BaseB, align 8, !tbaa !7
into:
%BaseA = getelementptr inbounds %struct.C, %struct.C* %c, i64 0, i32 0
%BaseB = getelementptr inbounds %struct.C, %struct.C* %c, i64 0, i32 1
%0 = bitcast i8** %BaseA to <2 x i8*>*
store <2 x i8*> <i8* getelementptr inbounds ([0 x i8], [0 x i8]* @VTab, i64
0, i64 48), i8* getelementptr inbounds ([0 x i8], [0 x i8]* @VTab, i64 0, i64
16)>, <2 x i8*>* %0, align 8, !tbaa !2
ret void
which results in this code on X86:
movq _VTab@GOTPCREL(%rip), %rax
leaq 16(%rax), %rcx
movd %rcx, %xmm0
addq $48, %rax
movd %rax, %xmm1
punpcklqdq %xmm0, %xmm1 ## xmm1 = xmm1[0],xmm0[0]
movdqu %xmm1, (%rdi)
while it can be this if the SLP Vectorizer doesn't do anything:
movq _VTab@GOTPCREL(%rip), %rax
leaq 48(%rax), %rcx
movq %rcx, (%rdi)
addq $16, %rax
movq %rax, 8(%rdi)</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>