<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 - llvm.experimental.vector.reduce.{fadd,fmul} select fails when passed a non-undef accumulator"
href="https://bugs.llvm.org/show_bug.cgi?id=36732">36732</a>
</td>
</tr>
<tr>
<th>Summary</th>
<td>llvm.experimental.vector.reduce.{fadd,fmul} select fails when passed a non-undef accumulator
</td>
</tr>
<tr>
<th>Product</th>
<td>new-bugs
</td>
</tr>
<tr>
<th>Version</th>
<td>unspecified
</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>enhancement
</td>
</tr>
<tr>
<th>Priority</th>
<td>P
</td>
</tr>
<tr>
<th>Component</th>
<td>new bugs
</td>
</tr>
<tr>
<th>Assignee</th>
<td>unassignedbugs@nondot.org
</td>
</tr>
<tr>
<th>Reporter</th>
<td>gonzalobg88@gmail.com
</td>
</tr>
<tr>
<th>CC</th>
<td>chandlerc@gmail.com, hfinkel@anl.gov, llvm-bugs@lists.llvm.org, llvm-dev@redking.me.uk
</td>
</tr></table>
<p>
<div>
<pre>The docs of the llvm.experimental.vector.reduce.{fadd,fmul} intrinsics state:
<span class="quote">> If the intrinsic call has fast-math flags, then the reduction will not preserve the associativity of an equivalent scalarized counterpart. If it does not have fast-math flags, then the reduction will be ordered, implying that the operation respects the associativity of a scalarized reduction.
>
> Arguments:
> The first argument to this intrinsic is a scalar accumulator value, which is only used when there are no fast-math flags attached. This argument may be undef when fast-math flags are used.</span >
While undef + fast-math flags works fine, I haven't been able to get the
non-fast-math + accumulator version to work. They fail to select. I'm using
LLVM6 with assertions enabled.
For example:
declare float @llvm.experimental.vector.reduce.fadd.f32.f32.v4f32(float, <4 x
float>)
define internal float
@_ZN32simd_intrinsic_generic_reduction3foo17ha7e2b586cf5567bdE(<4 x float>*
noalias nocapture dereferenceable(16)) unnamed_addr #0 {
%2 = alloca float, align 4
%3 = load <4 x float>, <4 x float>* %0, align 16
%4 = call float @llvm.experimental.vector.reduce.fadd.f32.f32.v4f32(float
1.000000e+00, <4 x float> %3)
store float %4, float* %2, align 4
%5 = load float, float* %2, align 4
br label %6
; <label>:6: ; preds = %1
ret float %5
}
produces
LLVM ERROR: Cannot select: 0x3f478b0: f32 = vecreduce_strict_fadd 0x3f479e8,
0x3f477e0
0x3f479e8: f32,ch = load<LD4[ConstantPool]> 0x3ea4c28, 0x3f47b88, undef:i64
0x3f47b88: i64 = X86ISD::Wrapper TargetConstantPool:i64<float 1.000000e+00>
0
0x3f47848: i64 = TargetConstantPool<float 1.000000e+00> 0
0x3f47778: i64 = undef
0x3f477e0: v4f32,ch = load<LD16[%0](dereferenceable)> 0x3ea4c28, 0x3f476a8,
undef:i64
0x3f476a8: i64,ch = CopyFromReg 0x3ea4c28, Register:i64 %1
0x3f47640: i64 = Register %1
0x3f47778: i64 = undef
In function: _ZN32simd_intrinsic_generic_reduction3foo17ha7e2b586cf5567bdE
Compiler returned: 1
and
declare float @llvm.experimental.vector.reduce.fmul.f32.f32.v4f32(float, <4 x
float>)
define internal float
@_ZN32simd_intrinsic_generic_reduction3foo17ha7e2b586cf5567bdE(<4 x float>*
noalias nocapture dereferenceable(16)) unnamed_addr #0 {
%2 = alloca float, align 4
%3 = load <4 x float>, <4 x float>* %0, align 16
%4 = call float @llvm.experimental.vector.reduce.fmul.f32.f32.v4f32(float
1.000000e+00, <4 x float> %3)
store float %4, float* %2, align 4
%5 = load float, float* %2, align 4
br label %6
; <label>:6: ; preds = %1
ret float %5
}
produces this
LLVM ERROR: Cannot select: 0x4da88a0: f32 = vecreduce_strict_fmul 0x4da89d8,
0x4da87d0
0x4da89d8: f32,ch = load<LD4[ConstantPool]> 0x4d05c28, 0x4da8b78, undef:i64
0x4da8b78: i64 = X86ISD::Wrapper TargetConstantPool:i64<float 1.000000e+00>
0
0x4da8838: i64 = TargetConstantPool<float 1.000000e+00> 0
0x4da8768: i64 = undef
0x4da87d0: v4f32,ch = load<LD16[%0](dereferenceable)> 0x4d05c28, 0x4da8698,
undef:i64
0x4da8698: i64,ch = CopyFromReg 0x4d05c28, Register:i64 %1
0x4da8630: i64 = Register %1
0x4da8768: i64 = undef
In function: _ZN32simd_intrinsic_generic_reduction3foo17ha7e2b586cf5567bdE
Compiler returned: 1</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>