Fast VLOAD

[ZeroByteOrg]

This speeds up VLOAD by enabling block reads directly into VRAM in the LOAD routine using MACPTR. This is over ten times faster than the previous byte-by-byte loading that was done.

This requires a potentially-breaking change for software that uses MACPTR directly. MACPTR now uses the C flag to select between normal loading behavior, and a new copy loop which does not increment the memory destination pointer as data is read from the buffer.

I have implemented the new behavior by creating a new copy loop in fat32_read. The new loop is required, since the previous copy loop uses count-down-to-zero behavior on the loop index, which would reverse the order of bytes being sent into VRAM. Modifying the existing loop would have also required a check on each iteration whether to increment the dst counter, which would be slower, so I opted for a purpose-built copy loop instead.

Kernal's own calls to MACPTR are patched to work with the new API functionality on MACPTR, but end-user code that calls MACPTR will need to ensure the carry flag is clear prior to calling MACPTR or else it will cause data loading to behave incorrectly.

[ZeroByteOrg]

Accompanying emulator patch is PR#432

[mist64]

Once this is merged:

• https://github.com/commanderx16/x16-emulator/pull/432 will mirror this for the Host FS in the emulator
• we need to document it

[mist64]

I cleaned up the label magic in fce973e79f0930db770ef96cf74ff14061737f1e.