/// <summary>
/// Apply floating point IDCT transformation into dest, using a temporary block 'temp' provided by the caller (optimization).
/// Ported from https://github.com/norishigefukushima/dct_simd/blob/master/dct/dct8x8_simd.cpp#L239
/// </summary>
/// <param name="src">Source</param>
/// <param name="dest">Destination</param>
/// <param name="temp">Temporary block provided by the caller</param>
public static void TransformIDCT(ref Block8x8F src, ref Block8x8F dest, ref Block8x8F temp)
{
// TODO: Transpose is a bottleneck now. We need full AVX support to optimize it:
// https://github.com/dotnet/corefx/issues/22940
src.TransposeInto(ref temp);
IDCT8x4_LeftPart(ref temp, ref dest);
IDCT8x4_RightPart(ref temp, ref dest);
dest.TransposeInto(ref temp);
IDCT8x4_LeftPart(ref temp, ref dest);
IDCT8x4_RightPart(ref temp, ref dest);
// TODO: What if we leave the blocks in a scaled-by-x8 state until final color packing?
dest.MultiplyInplace(C_0_125);
}
/// <summary>
/// Apply floating point IDCT transformation into dest, using a temporary block 'temp' provided by the caller (optimization)
/// </summary>
/// <param name="src">Source</param>
/// <param name="dest">Destination</param>
/// <param name="temp">Temporary block provided by the caller</param>
/// <param name="offsetSourceByNeg128">If true, a constant -128.0 offset is applied for all values before FDCT </param>
public static void TransformFDCT(
ref Block8x8F src,
ref Block8x8F dest,
ref Block8x8F temp,
bool offsetSourceByNeg128 = true)
{
src.TransposeInto(ref temp);
if (offsetSourceByNeg128)
{
temp.AddToAllInplace(new Vector4(-128));
}
FDCT8x4_LeftPart(ref temp, ref dest);
FDCT8x4_RightPart(ref temp, ref dest);
dest.TransposeInto(ref temp);
FDCT8x4_LeftPart(ref temp, ref dest);
FDCT8x4_RightPart(ref temp, ref dest);
dest.MultiplyInplace(C_0_125);
}