Future Work
Short Term
- Detailed user manual:
-
A detailed user manual describing the interfaces, algorithms,
and examples on how to put together code generators.
- x86 code generator:
- The x86 poses several
challenges which can be met quite effectively, I believe, by the
register allocation technology used in MLRISC.
- Support for GC:
- There is a strong interaction
with support for GC and global code motion. MLRISC aims at
providing a generic framework for code generators, and finding
the right level of information to support GC and global code
motion is an issue. I think we have several solutions to address
this that need more evaluation.
- Other architectures:
- There is the need to port
to other architectures like the MIPS and SPARC.
Long Term
- Predicated VLIW:
- Currently, there are no optimizations
that will support predication. In principle there is no reason why this
should be the case.
- Other compilers:
- I would really like to see some
major compiler effort bootstrapped with an MLRISC backend.
- Robustness of scheduling
- It is extremely difficult to
debug errors in modules that perform aggressive code
reorganizations. Ideas from formal methods such as typed assembly
language (TAL) or Proof Carrying Code (PCC) are worth investigating.
Lal George
Last modified: Thu May 14 13:41:05 EDT 1998