Repository navigation
JIT: (bug) Vector128.Dot with a scalar-widened operand hits a JIT assert #133551
Description
Repro
using System;
using System.Runtime.CompilerServices;
using System.Runtime.Intrinsics;
using System.Runtime.Intrinsics.X86;
class Program
{
[MethodImpl(MethodImplOptions.NoInlining | MethodImplOptions.AggressiveOptimization)]
static float Test(float x, Vector128<float> v) =>
Vector128.Dot(Vector128.CreateScalarUnsafe(x), v);
static void Main()
{
if (!Avx.IsSupported)
throw new PlatformNotSupportedException("AVX is required.");
Console.WriteLine(Test(3.0f, Vector128.Create(2.0f)));
}
}Expected
6
Actual
Assert failure(PID 21972 [0x000055d4], Thread: 123408 [0x1e210]): Assertion failed 'false && "found use of a node that is not in the LIR sequence"' in 'Program:Test(float,System.Runtime.Intrinsics.Vector128`1[float]):float' during 'Lowering nodeinfo' (IL size 13; hash 0x1ad19734; FullOpts)
File: C:\prj\runtime-main3\src\coreclr\jit\lir.cpp:1692
The same assert fires for Vector128.Dot(v, Vector128.CreateScalarUnsafe(x)), for the double
forms, and for the Vector256 forms. Running with DOTNET_EnableAVX=0 prints the expected 6.
Platform
Windows x64 with AVX, local Checked build of dotnet/runtime main (ef97eb5).
Activity
dotnet-policy-service commentedon Sep 10, 2026
Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch
See info in area-owners.md if you want to be subscribed.
Fix scalar-typed SIMD operands in Dot and mask inversion lowering (#1…
EgorBo commentedon Sep 18, 2026
Affected SDK versions:
Fix scalar-typed SIMD operands in Dot and mask inversion lowering (do…
EgorBo commentedon Sep 18, 2026
@tannergooding would you consider backporting the fix from #133709 to
release/11.0?It currently only exists in
main(12.0). The repro still fails on the shipped 11.0 RC2 SDK (11.0.100-rc.2.26467.112), while 8.0.425, 9.0.318 and 10.0.401 all behave correctly, so this is a regression introduced during the 11.0 cycle rather than a long-standing bug — 11.0 would ship with it.Note: this comment was produced by an AI agent; the repro was re-run against all four released SDKs to confirm the results above.
[release/11.0] Fix scalar-typed SIMD operands in Dot and mask inversi…