TRANSFER ROUTINE ARGUMENTS -- SOURCE FOUND, DESTINATION ABSENT.

Breakpoint at file 0x1253FA (the MOVE A7,*A6 inside the transfer routine):

  XFER a0(src)=00919850  a7=00000000  a5=05600080  a6=05600050
       b3=05600100  b4=05600030  b5=0560010   b6=05600200

  and the copy loop confirms the source walks forward one word at a time:
  COPY a0=00919850 a5=05600080
  COPY a0=00919860 a5=05600080
  COPY a0=00919870 a5=05600080   (steps of 0x10 bits = 1 word)

So:
  A0 = 0x00919850  -- a MAIN RAM bit address (byte 0x0012330A): the source pixels,
                      i.e. the photomap data (the same buffer verified to hold
                      0xFF white / 0x00 black).
  A7 = 0           -- the word handed to the data register.
  A5/A6/B3..B6     -- exactly the register pointers (FIFO, data, command,
                      control, status, engine output).

NO DESTINATION IS PASSED TO THIS ROUTINE.  There is no pointer to VRAM
(>= 0x04000000) anywhere in the arguments, the register writes or the table.
The engine is expected to know where to put the pixels -- the placement must be
programmed before this call, through the parameter/command ports
(0x5600080 table, 0x5600020 = 0x248/0x258, 0x5600100 = 0008/0200/0201/0202), and
that encoding is not yet decoded.

WHAT IS PROVEN NOW
  * the export completes without error;
  * the source pixels are streamed into the accelerator FIFO (2200 writes for a
    white tile, containing 0xFFFF; 4400 across two exports; 0x0000 dominates for
    a black tile);
  * the transfer routine, its arguments and the full register interface are
    identified.

WHAT IS NOT
  * where the engine is told to put them -- hence nothing reaches VRAM and no
    client-chosen XIE image has been displayed;
  * the model is opt-in (VT1200_ACCEL=1); the default configuration is unchanged;
  * xshow (core X) remains the only viewer that has shown an image.
