xmlvm-users

[xmlvm-users] JVM to LLVM or XMLVM to LLVM

[Steve H. <st...@we...>]

Hi Arno et al.,

I'm curious if anyone has considered writing an XMLVM converter to go from
JVM byte code to LLVM?  I have been analyzing the java to c path, then
compiled the resulting C code to LLVM IR code and compared that IR code to
the original JVM bytecode instructions.

On a cursory review of some methods I have found that the IR code generally
contains about 10 times the number of instructions as the corresponding JVM
byte code.  This factor also corresponds closely with the benchmark results
comparing Oracle's JVM to XMLVM (using some JSON parsing tests I have set
up).

I'm thinking that optimal results can probably be attained with there is
roughly a 1 to 1 correspondence between JVM instructions and LLVM
instructions.

It seems to me that XMLVM is well suited for being able to output LLVM, but
I'm sort of just thinking out loud at this point.

Has this been thought of or tried by anyone yet?  Is my goal of seeking a 1
to 1 mapping of JVM instructions to LLVM instructions (or as close as
possible) as a means of performance maximization off base?  Are there
technical reasons why such a conversion either can't be done or would not
be useful?

Thanks for any insight you can provide.

Best regards

Steve

-- 
Steve Hannah
Web Lite Solutions Corp.

Re: [xmlvm-users] JVM to LLVM or XMLVM to LLVM

[Arno P. <ar...@pu...>]

I think this is definitely a good idea. What you are proposing is a Java 
frontend to LLVM. There seems to be at least one such project:
https://llvm.org/svn/llvm-project/java/trunk/docs/java-frontend.txt

Reading through that page, they discuss some difficulties in modeling 
the operand stack of the JVM. It might be interesting to investigate a 
DEX frontend to LLVM since it uses a register-based instruction set 
(XMLVM also uses DEX for the C backend). A cursory search didn't show 
any hits.

Arno


On 9/10/13 4:45 AM, Steve Hannah wrote:
> Hi Arno et al.,
>
> I'm curious if anyone has considered writing an XMLVM converter to go
> from JVM byte code to LLVM?  I have been analyzing the java to c path,
> then compiled the resulting C code to LLVM IR code and compared that IR
> code to the original JVM bytecode instructions.
>
> On a cursory review of some methods I have found that the IR code
> generally contains about 10 times the number of instructions as the
> corresponding JVM byte code.  This factor also corresponds closely with
> the benchmark results comparing Oracle's JVM to XMLVM (using some JSON
> parsing tests I have set up).
>
> I'm thinking that optimal results can probably be attained with there is
> roughly a 1 to 1 correspondence between JVM instructions and LLVM
> instructions.
>
> It seems to me that XMLVM is well suited for being able to output LLVM,
> but I'm sort of just thinking out loud at this point.
>
> Has this been thought of or tried by anyone yet?  Is my goal of seeking
> a 1 to 1 mapping of JVM instructions to LLVM instructions (or as close
> as possible) as a means of performance maximization off base?  Are there
> technical reasons why such a conversion either can't be done or would
> not be useful?
>
> Thanks for any insight you can provide.
>
> Best regards
>
> Steve
>
> --
> Steve Hannah
> Web Lite Solutions Corp.
>
>
> ------------------------------------------------------------------------------
> How ServiceNow helps IT people transform IT departments:
> 1. Consolidate legacy IT systems to a single system of record for IT
> 2. Standardize and globalize service processes across IT
> 3. Implement zero-touch automation to replace manual, redundant tasks
> http://pubads.g.doubleclick.net/gampad/clk?id=51271111&iu=/4140/ostg.clktrk
>
>
>
> _______________________________________________
> xmlvm-users mailing list
> xml...@li...
> https://lists.sourceforge.net/lists/listinfo/xmlvm-users
>