I just discovered this Chinese Univeristy of Hong Kong's researcher project:
"DWT shader that runs on existing GPU hardwares. We implement the convolutionary based DWT and propose an indirect addressing technique that supports various boundary extension schemes. The proposed technique also unifies the computation of both forward and inverse DWT, hence making the shaders for FDWT and IDWT are basically the same. This may simplify the future hardware implementation. The software has already been successfully integrated into the well-known JPEG2000 codec, JasPer."
They also claims:
To popularize the usage of DWT on GPU, we release both the full source code and binary. The software is released in two forms:
1) JasPer unofficial GPU extension (both source code and compiled binary),
2) Standalone DWT-GPU C++ Class (source code with example program).
Hope that helps !