public static void TransformFDCT(
ref JpegBlock8x8F src,
ref JpegBlock8x8F dest,
ref JpegBlock8x8F temp)
{
src.TransposeInto(ref temp);
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);
}
public static void TransformIDCT(ref JpegBlock8x8F src, ref JpegBlock8x8F dest, ref JpegBlock8x8F 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);
}
public static void TransformFDCT(
ref JpegBlock8x8F src,
ref JpegBlock8x8F dest,
ref JpegBlock8x8F temp,
bool offsetSourceByNeg128)
{
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);
}