rewrite of f32 fma, and other optmization of simd kernel ops

StgSharp.Common/HighPerformance/M512.cs

@@ -31,6 +31,7 @@ using System.Linq;
 using System.Numerics;
 using System.Runtime.CompilerServices;
 using System.Runtime.InteropServices;
+using System.Runtime.Intrinsics;
 using System.Text;
 using System.Threading.Tasks;
 
@@ -41,12 +42,18 @@ namespace StgSharp.HighPerformance
     {
 
         [FieldOffset(0)]private fixed byte _buffer[64];
+        [FieldOffset(0)]private Vector512<float> Vec;
 
         public ref T AsRef<T>() where T: unmanaged, INumber<T>
         {
             return ref Unsafe.As<byte, T>(ref _buffer[0]);
         }
 
+        public void BroadCastFrom<T>(T value) where T: unmanaged, INumber<T>
+        {
+            Vec = Vector512.CreateScalarUnsafe(value).As<T, float>();
+        }
+
         public ref T Member<T>(int index) where T: unmanaged, INumber<T>
         {
             return ref Unsafe.As<byte, T>(ref _buffer[0]);