Re: dynamic bintrans

heya guys,


> I'm interested in dynamic binary translation. I've done some small
> tests of my own, and they work pretty good. Now I want to make the
> compiler retargetable. For this, I have to choose an IR that can


You Don't HAVE to. If You Translate Using Matching Tables you can do
without an IR and map emulated instructions directly to (a sequence of)
host instructions.


> describe targets which has specific characteristics (such as delay
> slots, condition codes (x86), condition insns (arm, ia-64), ..).
> Standard RTL, such as used by GCC, seems to be too generic to
> describe for example the condition codes on x86. It also has
> troubles describing conditional insns.


An IR That Compiles Is easy to achieve. but to emit good code and still
be "retargetable" (define that...) is hard. but since you won't have
much time for sophisticated optimizations anyway and speed really
matters, i'd suggest to build an easy compiler backend and focus on fast
instruction selection etc. i got heaps of info about the whole topic
sitting on my desk, so feel free to contact me, and in addition have a
look at


      http://savage.light-speed.de/


yirr-ma, qemu and lightning might be of interest for you :)


cheers,
jens

Return to the comp.compilers page.
Search the comp.compilers archives again.