<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 - Poor handling of float3 within a float4 type"
href="https://bugs.llvm.org/show_bug.cgi?id=50920">50920</a>
</td>
</tr>
<tr>
<th>Summary</th>
<td>Poor handling of float3 within a float4 type
</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>Common Code Generator Code
</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>a.bataev@hotmail.com, llvm-bugs@lists.llvm.org, spatel+llvm@rotateright.com
</td>
</tr></table>
<p>
<div>
<pre><a href="https://godbolt.org/z/7jxnGdd7d">https://godbolt.org/z/7jxnGdd7d</a>
#include <cstdint>
#include <cfloat>
#include <x86intrin.h>
__m128 invert_float3(__m128 x) {
for (int i = 0; i != 3; ++i) {
x[i] = x[i] == 0 ? FLT_MAX : 1.0 / x[i];
}
return x;
}
define <4 x float> @invert_float3(<4 x float> %0) {
%2 = extractelement <4 x float> %0, i32 0
%3 = fcmp oeq float %2, 0.000000e+00
%4 = fdiv float 1.000000e+00, %2
%5 = select i1 %3, float 0x47EFFFFFE0000000, float %4
%6 = shufflevector <4 x float> %0, <4 x float> undef, <2 x i32> <i32 1, i32
2>
%7 = fcmp oeq <2 x float> %6, zeroinitializer
%8 = fdiv <2 x float> <float 1.000000e+00, float 1.000000e+00>, %6
%9 = select <2 x i1> %7, <2 x float> <float 0x47EFFFFFE0000000, float
0x47EFFFFFE0000000>, <2 x float> %8
%10 = insertelement <4 x float> %0, float %5, i32 0
%11 = shufflevector <2 x float> %9, <2 x float> poison, <4 x i32> <i32 0, i32
1, i32 undef, i32 undef>
%12 = shufflevector <4 x float> %10, <4 x float> %11, <4 x i32> <i32 0, i32
4, i32 5, i32 3>
ret <4 x float> %12
}
invert_float3(float __vector(4)):
vxorps %xmm1, %xmm1, %xmm1
vucomiss %xmm1, %xmm0
jne .LBB0_2
jp .LBB0_2
vmovss .LCPI0_0(%rip), %xmm1 # xmm1 = mem[0],zero,zero,zero
jmp .LBB0_3
.LBB0_2:
vmovss .LCPI0_1(%rip), %xmm1 # xmm1 = mem[0],zero,zero,zero
vdivss %xmm0, %xmm1, %xmm1
.LBB0_3:
vmovaps .LCPI0_2(%rip), %xmm4 # xmm4 = <1.0E+0,1.0E+0,u,u>
vpermilps $233, %xmm0, %xmm2 # xmm2 = xmm0[1,2,2,3]
vxorps %xmm3, %xmm3, %xmm3
vblendps $3, %xmm1, %xmm0, %xmm0 # xmm0 =
xmm1[0,1],xmm0[2,3]
vcmpeqps %xmm3, %xmm2, %xmm3
vdivps %xmm2, %xmm4, %xmm2
vblendvps %xmm3, .LCPI0_3(%rip), %xmm2, %xmm2
vshufps $76, %xmm2, %xmm0, %xmm0 # xmm0 = xmm0[0,3],xmm2[0,1]
vpermilps $120, %xmm0, %xmm0 # xmm0 = xmm0[0,2,3,1]
retq
Many vector math libraries embed float3 vectors inside a float4 register
vector, and just ignore the [3] element. For instance, in this safe fdiv code,
only the first 3 elements are inverted.
I'm hoping that D57059 will help with this, and ideally will generate a <float
x 3> pattern that blends with the [3] pass-through value.</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>