<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 - [SLP] Allow vectorization for vec4 ops for avx2+"
href="https://bugs.llvm.org/show_bug.cgi?id=47576">47576</a>
</td>
</tr>
<tr>
<th>Summary</th>
<td>[SLP] Allow vectorization for vec4 ops for avx2+
</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>Scalar Optimizations
</td>
</tr>
<tr>
<th>Assignee</th>
<td>unassignedbugs@nondot.org
</td>
</tr>
<tr>
<th>Reporter</th>
<td>david.bolvansky@gmail.com
</td>
</tr>
<tr>
<th>CC</th>
<td>llvm-bugs@lists.llvm.org
</td>
</tr></table>
<p>
<div>
<pre>typedef unsigned char pixel;
typedef unsigned short dctcoef;
void pixel_sub_wxh( dctcoef *__restrict diff, pixel *__restrict pix1, int
i_pix1, pixel *__restrict pix2, int i_pix2 )
{
diff[0] = pix1[0] + pix2[0];
diff[1] = pix1[1] + pix2[1];
diff[2] = pix1[2] + pix2[2];
diff[3] = pix1[3] + pix2[3];
}
Clang -mavx2 -O3 :
pixel_sub_wxh(unsigned short*, unsigned char*, int, unsigned char*, int):
# @pixel_sub_wxh(unsigned short*, unsigned char*, int, unsigned char*,
int)
movzx eax, byte ptr [rsi]
movzx edx, byte ptr [rcx]
add edx, eax
mov word ptr [rdi], dx
movzx eax, byte ptr [rsi + 1]
movzx edx, byte ptr [rcx + 1]
add edx, eax
mov word ptr [rdi + 2], dx
movzx eax, byte ptr [rsi + 2]
movzx edx, byte ptr [rcx + 2]
add edx, eax
mov word ptr [rdi + 4], dx
movzx eax, byte ptr [rsi + 3]
movzx ecx, byte ptr [rcx + 3]
add ecx, eax
mov word ptr [rdi + 6], cx
ret
ICC -mavx2 -O3 (since 19.00) :
pixel_sub_wxh(unsigned short*, unsigned char*, int, unsigned char*, int):
vmovd xmm0, DWORD PTR [rsi] #6.15
vmovd xmm1, DWORD PTR [rcx] #6.25
vpmovzxbw xmm0, xmm0 #6.15
vpmovzxbw xmm1, xmm1 #6.25
vpaddw xmm2, xmm0, xmm1 #6.25
vmovq QWORD PTR [rdi], xmm2 #6.5
ret
<a href="https://gcc.godbolt.org/z/a9rxaM">https://gcc.godbolt.org/z/a9rxaM</a>
Cost model issue?</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>