THE ENGINE IS NOT PROGRAMMED WITH THE STAGING ADDRESS.

Logged every write to the accelerator's control registers for a white and a black
export, with the writing PC:

  fill=ff
    0x5600020: 0248@3F3B6090  0248@3F3B6A00  0258@3FB16EF0
    0x5600030: 0032@3F3B60E0  0033@3F3B6420  0032@3F3B6C00
               0432@3FB29FD0  0633@3FB2A130  0233@3FB2A1E0
               0032@3FB2A790  0032@3FB16F40  0233@3FB17260
  fill=00
    identical sequence (0x248 / 0x258 / 0x32 / 0x33 / 0x432 / 0x633 / 0x233 ...)

Neither register, in either fill, carries anything derived from the staging
address 0x004FE790 (which would appear as the halves 0x4FE7 / 0x9070).  The
sequences are fill-independent.

So the "engine fetches the staged buffer" hypothesis is dead: there is nothing to
add to accel_w.  The engine is programmed with small scalars only, and the staged
pixels are discarded.

STATE
  * NO client-chosen XIE image has ever been displayed;
  * fill-dependent proof: client -> photomap -> FIFO -> engine output -> CPU store
    into staging (283 x 0xFFFF for white vs mostly 0x0000 for black);
  * staging is overwritten; VRAM never differs (0 bytes / 128 KiB);
  * no register carries a source/destination address for the engine;
  * model opt-in (VT1200_ACCEL=1); default configuration unchanged;
  * xshow (core X) is the only viewer that has shown an image.
