Avx512 convert opt

[anthonycanino]

Overview

This PR demonstrates a few things which I think are useful for learnings (and optimize several cases below that ARM typically had optimized and beat us in micros):

1. The creation of a table driven intrinsic, e.g., NI_AVX512DQ_VL_ConvertToDouble128, which allows to map a lowering between a method (though in this case, there are no frontend APIs added) and an argument type (for example, Vector128<long>) to a specific hardware instruction vcvtps2udq.

2. The mapping of a vector method to a named intrinsic, for example, Vector128.ConvertToDouble(Vector128<ulong> v) to NI_AVX512DQ_VL_ConvertToDouble128.

3. The mapping of a variable length vector method to a named intrinsic, for example Vector.ConvertToDouble(Vector<ulong> v) to NI_AVX512DQ_VL_ConvertToDouble128.

4. The addition of several new AVX512 instructions to the instruction tables.

Not all intrinsics must map to a single instruction --- several will actually lower to a series of instructions, which we'd probably call it "codegen" versus a hardware intrinsic, though those terms are loosely defined.

Optimized Cases

Cases that have been optimized to use one of the single AVX512VL instructions...

        // -----------------------------------------
	// Cases are Vector long => double, ulong => double
	// -----------------------------------------

	// VCVTUQQ2PD
	[MethodImpl(MethodImplOptions.NoInlining)]
	public static Vector128<double> VLongToDouble(long i)
	{
		Vector128<long> v1 = Vector128.Create(i);
		return Vector128.ConvertToDouble(v1);
	}

	// VCVTQQ2PD
	[MethodImpl(MethodImplOptions.NoInlining)]
	public static Vector128<double> VUlongToDouble(ulong i)
	{
		Vector128<ulong> v1 = Vector128.Create(i);
		return Vector128.ConvertToDouble(v1);
	}

	// VCVTUQQ2PD
	[MethodImpl(MethodImplOptions.NoInlining)]
	public static Vector<double> VLongToDoubleVT(long i)
	{
		Vector<long> v1 = new Vector<long>(i);
		return Vector.ConvertToDouble(v1);
	}

	// VCVTQQ2PD
	[MethodImpl(MethodImplOptions.NoInlining)]
	public static Vector<double> VUlongToDoubleVT(ulong i)
	{
		Vector<ulong> v1 = new Vector<ulong>(i);
		return Vector.ConvertToDouble(v1);
	}

	// -----------------------------------------
	// Cases are Vector double => long
	// -----------------------------------------

	[MethodImpl(MethodImplOptions.NoInlining)]
	public static Vector128<long> VDoubleToLong(double i)
	{
		Vector128<double> v1 = Vector128.Create(i);
		return Vector128.ConvertToInt64(v1);
	}

	// VCVTUQQ2PD
	[MethodImpl(MethodImplOptions.NoInlining)]
	public static Vector<long> VDoubleToLongVT(double i)
	{
		Vector<double> v1 = new Vector<double>(i);
		return Vector.ConvertToInt64(v1);
	}

	// -----------------------------------------
	// Cases are Vector double => ulong
	// -----------------------------------------

	[MethodImpl(MethodImplOptions.NoInlining)]
	public static Vector128<ulong> VDoubleToUlong(double i)
	{
		Vector128<double> v1 = Vector128.Create(i);
		return Vector128.ConvertToUInt64(v1);
	}

	// VCVTUQQ2PD
	[MethodImpl(MethodImplOptions.NoInlining)]
	public static Vector<ulong> VDoubleToUlongVT(double i)
	{
		Vector<double> v1 = new Vector<double>(i);
		return Vector.ConvertToUInt64(v1);
	}

	// -----------------------------------------
	// Cases are Vector float => uint
	// -----------------------------------------

	[MethodImpl(MethodImplOptions.NoInlining)]
	public static Vector128<uint> VFloatToUint(float i)
	{
		Vector128<float> v1 = Vector128.Create(i);
		return Vector128.ConvertToUInt32(v1);
	}

	[MethodImpl(MethodImplOptions.NoInlining)]
	public static Vector<uint> VFloatToUintVT(float i)
	{
		Vector<float> v1 = new Vector<float>(i);
		return Vector.ConvertToUInt32(v1);
	}

	[MethodImpl(MethodImplOptions.NoInlining)]
	public static Vector128<float> VUintToFloat(uint i)
	{
		Vector128<uint> v1 = Vector128.Create(i);
		return Vector128.ConvertToSingle(v1);
	}

	[MethodImpl(MethodImplOptions.NoInlining)]
	public static Vector<float> VUintToFloatVT(uint i)
	{
		Vector<uint> v1 = new Vector<uint>(i);
		return Vector.ConvertToSingle(v1);
	}