* Support for 68020/68030
- It is broken again :-(

* Support for other platforms: SPARC
- cleanups in compemu_support.cpp to get rid of little-endian assumptions
- set up another Virtual Register for the m68k regstruct base address

* Implement Dynamic Code Patches (see MAE from Apple)
- should be associated with a cpu core that first does pretranslation before
  executing the instruction handlers. e.g. a dynamic recompiler to synthetic
  instructions, like people from ARDI have been doing for Syn68k

* Reversed address space
- reimplement that scheme in the standard Basilisk II distribution but in
  a cleaner way that time. Then, there should be just little work to add
  support for it to the JIT compiler

* Lazy Flusher 2: a lazier implementation
- goal; postpone as much as possible adjustments to jump/branch addresses
  in the source blocks
- planned process:

# a soft cache flush occurs
for any basic block compiled; do
  if (first instruction is not ChecksumOp) then
    save that first instruction;
    patch the first instruction with ChecksumOp;
  fi
end

# ChecksumOp is an illegal (host) instruction that will be caught by an
appropriate signal handler

# Next time we call a compiled block, we fall into the SIGILL handler
Since we know the address of the illegal instruction, we can retrieve
the corresponding blockinfo structure. Then, inside the SIGILL handler,
we compute the block checksum:

- If it matches the one computed at translation time, we remove the
ChecksumOp and restore the old but valid instruction. More precisely,
the N bytes that a ChecksumOp instruction takes.

- If checksums mismatch, we have to invalidate the block. We generate a
jump to recompile_block() in place of the ChecksumOp instruction. We
return from the SIGILL handler and fall into recompile_block().

recompile_block(): do recompile the block but now patches the other
blocks that will have to call the new one. This is achieved thanks to
the dependency list.

Note: for any given block, we need to now the list of all its possible
predecessors.

-- 
Gwenole Beauchesne
gb@dial.oleane.com