High-performance and low-latency implementation of a wavelet-based image compression scheme

An apparatus for graphics processing at a second device, comprising:

• a memory;
• and
• a processor coupled to the memory and, based on information stored in the memory, the processor is configured to:
• obtain a bitstream associated with an image frame associated with a first device, wherein the bitstream includes a set of serialized wavelet coefficients associated with the image frame;
• deserialize the set of serialized wavelet coefficients, wherein the set of deserialized wavelet coefficients includes a set of additional bits corresponding to a set of wavelet coefficients associated with the set of serialized wavelet coefficients;
• reconstruct, in a single pass for each of a plurality of horizontal and vertical divisions of the image frame, the image frame based on the set of deserialized wavelet coefficients based on an inverse discrete wavelet transform (iDWT), wherein to reconstruct the image frame, the processor is configured to:
• perform the iDWT on the image frame using one or more 16-bit precision computations;
• and
• allocate the set of deserialized wavelet coefficients to a 12-10-10-bit storage format based on the one or more 16-bit precision computations;
• and
• output an indication of the reconstructed image frame.

A method of graphics processing at a first device, comprising:

• generating, in a single pass for each of a plurality of horizontal and vertical divisions of an image frame, a set of wavelet coefficients for the image frame based on a discrete wavelet transform (DWT), wherein generating the set of wavelet coefficients comprises:
• performing the DWT on the image frame based on one or more 16-bit precision computations in order to generate the set of wavelet coefficients;
• and
• allocating the set of wavelet coefficients to a 12-10-10-bit storage format based on the one or more 16-bit precision computations;
• selecting a set of bits associated with the set of wavelet coefficients to remove based on an entropy coding process;
• serializing the set of wavelet coefficients, wherein the set of serialized wavelet coefficients do not include the set of bits;
• and
• outputting, for a second device, a bitstream including the set of serialized wavelet coefficients.

A method of graphics processing at a second device, comprising:

• obtaining a bitstream associated with an image frame associated with a first device, wherein the bitstream includes a set of serialized wavelet coefficients associated with the image frame;
• deserializing the set of serialized wavelet coefficients, wherein the set of deserialized wavelet coefficients includes a set of additional bits corresponding to a set of wavelet coefficients associated with the set of serialized wavelet coefficients;
• reconstructing, in a single pass for each of a plurality of horizontal and vertical divisions of the image frame, the image frame based on the set of deserialized wavelet coefficients based on an inverse discrete wavelet transform (iDWT), wherein reconstructing the image frame comprises:
• performing the iDWT on the image frame using one or more 16-bit precision computations;
• and
• allocating the set of deserialized wavelet coefficients to a 12-10-10-bit storage format based on the one or more 16-bit precision computations;
• and
• outputting an indication of the reconstructed image frame.