<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 - Better codegen for strided load"
href="https://bugs.llvm.org/show_bug.cgi?id=48046">48046</a>
</td>
</tr>
<tr>
<th>Summary</th>
<td>Better codegen for strided load
</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>david.bolvansky@gmail.com
</td>
</tr>
<tr>
<th>CC</th>
<td>craig.topper@gmail.com, llvm-bugs@lists.llvm.org, llvm-dev@redking.me.uk, pengfei.wang@intel.com, spatel+llvm@rotateright.com
</td>
</tr></table>
<p>
<div>
<pre>#define N 4
float x[4*N], y[N];
void foo (int p)
{
int i;
for (i = 0; i < N; i++)
y[i] = x[p + 3*i];
}
Clang -O3 -mavx2:
foo(int): # @foo(int)
movsxd rax, edi
vmovss xmm0, dword ptr [4*rax + x] # xmm0 = mem[0],zero,zero,zero
vmovss dword ptr [rip + y], xmm0
vmovss xmm0, dword ptr [4*rax + x+12] # xmm0 = mem[0],zero,zero,zero
vmovss dword ptr [rip + y+4], xmm0
vmovss xmm0, dword ptr [4*rax + x+24] # xmm0 = mem[0],zero,zero,zero
vmovss dword ptr [rip + y+8], xmm0
vmovss xmm0, dword ptr [4*rax + x+36] # xmm0 = mem[0],zero,zero,zero
vmovss dword ptr [rip + y+12], xmm0
ret
ICC -O3 -mavx2:
foo(int):
movsxd rdi, edi #6.1
vmovss xmm16, DWORD PTR [12+x+rdi*4] #10.12
vmovss xmm17, DWORD PTR [x+rdi*4] #10.12
vinsertps xmm1, xmm16, DWORD PTR [36+x+rdi*4], 16 #10.12
vinsertps xmm0, xmm17, DWORD PTR [24+x+rdi*4], 16 #10.12
vunpcklps xmm2, xmm0, xmm1 #10.12
vmovups XMMWORD PTR y[rip], xmm2 #10.5
ret #11.1
x:
y:
GCC -O3 -mavx2:
foo(int):
movsx rdi, edi
vmovss xmm0, DWORD PTR x[0+rdi*4]
vinsertps xmm0, xmm0, DWORD PTR x[12+rdi*4], 0x10
vmovlps QWORD PTR y[rip], xmm0
vmovss xmm0, DWORD PTR x[24+rdi*4]
vinsertps xmm0, xmm0, DWORD PTR x[36+rdi*4], 0x10
vmovlps QWORD PTR y[rip+8], xmm0
ret
y:
GCC has the best Block RThroughput value - 2.5.
<a href="https://godbolt.org/z/Pc1TWz">https://godbolt.org/z/Pc1TWz</a></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>