[llvm-dev] JIT: Mapping global variable in JIT'ted code to variable in running program

[Lang Hames via llvm-dev]

Hi Tarun,

My first thought is that you might be adding a definition (e.g. @x = global
i32 7) to your module, rather than an external declaration (e.g. @x =
external global i32). Make sure you have the latter: If you provide a
definition in your module it will shadow any definitions in your mapping,
which would explain the behavior you're seeing.

Cheers,
Lang.


On Mon, Sep 14, 2015 at 2:55 PM, Tarun Prabhu via llvm-dev <
llvm-dev at lists.llvm.org> wrote:

> Hi,
>
> I think this is probably easiest to explain with code (I only provided the
> essentials for clarity):
>
> // begin file jit.cpp
>
> int myglobal;
>
> void printMyGlobal() {
>   printf("myglobal: %d\n", myglobal);
> }
>
> int main(int argc, char *argv[]) {
>
>   // This file, jit.cpp has been compiled to bitcode (clang -S -emit-llvm
> jit.cpp)
>   // and is read into Module M here
>   Module *M = ...
>   ExecutionEngine *ee = ...
>
>   myglobal = 42;
>
>   ee->addGlobalMapping(M->getGlobalVariable("myglobal"), (void*)&myglobal);
>   ee->finalizeObject();
>
>   void(*fptr)() = (void(*)())ee->getFunctionAddress("printMyGlobal");
>   fptr();
>   printMyGlobal();
> }
>
> // end file jit.cpp
>
> I compile this file (jit.cpp) into bitcode which I then load int M. M
>  also contains a global variable called myglobal. I want to "map" the
> myglobal variable in M to the global variable in the running program. The
> idea is that the value of myglobal inside the JIT'ted module will also be
> 42. However, as it is right now, when I run the code, I get:
>
> myglobal: 0
> myglobal: 42
>
> The first is the JIT'ted printMyGlobal() and the second is the AOT'ed
> printMyGlobal(). But when I lookup the address of the global variable
> myglobal (ee->getGlobalValueAddress()), it is mapped to the correct
> address. Similarly, if I lookup ee->getGlobalValueAtAddress(), I get the
> expected result.
>
> Am I missing an important step? Is what I am trying to do even possible?
>
> Thanks for the help.
>
> Tarun
>
>
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20150914/da5aab1f/attachment.html>