<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][SSE] Failure to SimplifyDemandedElts through ADDSUB intrinsics"
href="https://bugs.llvm.org/show_bug.cgi?id=46277">46277</a>
</td>
</tr>
<tr>
<th>Summary</th>
<td>[X86][SSE] Failure to SimplifyDemandedElts through ADDSUB intrinsics
</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, llvm-bugs@lists.llvm.org, llvm-dev@redking.me.uk, spatel+llvm@rotateright.com
</td>
</tr></table>
<p>
<div>
<pre><a href="https://godbolt.org/z/KPaVBx">https://godbolt.org/z/KPaVBx</a>
We are failing to remove undemanded elements through addsub intrinsics. This
most likely needs to be handled in InstCombiner::SimplifyDemandedVectorElts,
but we might still need it in
X86TargetLowering::SimplifyDemandedVectorEltsForTargetNode as well, as we do
generate X86ISD::ADDSUB on the fly, and the SHUFFLE(FADD,FSUB) pattern might
mean that hasOneUse tests will have failed before the combine.
#include <x86intrin.h>
void test_fadd(float x, float y, float z, float w, __m128 right, float *out) {
__m128 left = _mm_setr_ps(x, y, z, w);
__m128 result = _mm_add_ps(left, right);
*out++ = _mm_cvtss_f32(result);
*out++ = _mm_cvtss_f32(_mm_shuffle_ps(result, result, _MM_SHUFFLE(1,1,1,1)));
}
void test_faddsub(float x, float y, float z, float w, __m128 right, float *out)
{
__m128 left = _mm_setr_ps(x, y, z, w);
__m128 result = _mm_addsub_ps(left, right);
*out++ = _mm_cvtss_f32(result);
*out++ = _mm_cvtss_f32(_mm_shuffle_ps(result, result, _MM_SHUFFLE(1,1,1,1)));
}
test_fadd:
vinsertps $16, %xmm1, %xmm0, %xmm0 # xmm0 = xmm0[0],xmm1[0],xmm0[2,3]
vaddps %xmm4, %xmm0, %xmm0
vmovss %xmm0, (%rdi)
vextractps $1, %xmm0, 4(%rdi)
retq
test_faddsub:
vinsertps $16, %xmm1, %xmm0, %xmm0 # xmm0 = xmm0[0],xmm1[0],xmm0[2,3]
vinsertps $32, %xmm2, %xmm0, %xmm0 # xmm0 = xmm0[0,1],xmm2[0],xmm0[3]
vinsertps $48, %xmm3, %xmm0, %xmm0 # xmm0 = xmm0[0,1,2],xmm3[0]
vaddsubps %xmm4, %xmm0, %xmm0
vmovss %xmm0, (%rdi)
vextractps $1, %xmm0, 4(%rdi)
retq
we also fail to fold the stores into movsd/movlps....</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>