Thanks for clarifying. I looked at the description, and it seems that they
have a fairly 1:1 encoding from IR instructions to bitcode instructions,
such that the majority of bitcode instructions translate 1:1 into target
instructions on RISC machines. That would make it a DBT scheme in my mind.

For x86 instructions they have to do more work, or they could choose to eat
emitting multiple target instructions for each bitcode instruction.

Working at the binary instruction level like this doesn't preclude doing
sophisticated dynamic translation. What it does is enable a fast-and-light
translation mode. When the instruction set is at this level, you can
translate a bitcode instruction to a target instruction in 12-20 cycles per
instruction (which is *very* low), and sometimes faster. That's low enough
that trace linearization erases the costs of translation.

Come to that, this format is easier to decode than x86, so it's probably
easier to translate than what HDTrans needed for pentium->pentium
translation.



On Fri, Oct 25, 2013 at 8:14 AM, Ben Kloosterman <[email protected]> wrote:

>
> Bit code . IR files become .bc files which are processed by the execution
> engine..
> http://llvm.org/docs/BitCodeFormat.html
>
> It can also  run almost fully compiled assemblies  which is probably a DBT
> engine but i havent followed that through .
>
> Ben
>
> _______________________________________________
> bitc-dev mailing list
> [email protected]
> http://www.coyotos.org/mailman/listinfo/bitc-dev
>
>
_______________________________________________
bitc-dev mailing list
[email protected]
http://www.coyotos.org/mailman/listinfo/bitc-dev

Reply via email to