[Libre-soc-dev] Inverse DCT

[Luke Kenneth Casson Leighton]

On 7/30/21, Luke Kenneth Casson Leighton <lkcl at lkcl.net> wrote:
> next step, putting in a yield-based inverse DCT, was successful.
>
> next step is to link it into instructions and write a simulator unit test

done, successfully.

the problem comes with the LD instruction.

FFT: bitreverse with shift
DCT: recursive halfswap then bitreverse
iDCT: bitreverse then inverse recursive halfswap

this is just too much to fit into SVP64 24 bit prefix, and the
LD-byterev is actually interfering with applying REMAP.

what i am thinking of doing is removing bytereversing ftom the
augmented LD and just having LD-with-shift, to which 3  REMAP modes
above can be applied.

this makes FFT about 13 instructions rather than 11 but pffh.

l.