llvm dev - May 2015 - [LLVMdev] convert GetElemtPtr result to pointer on element?

Hi all,
I'm still struggling with getElementPtr.

I have a global variable (array of doubles).
I can use ConstantExpr::getGetElementPtr(myGlobalVariable, indices) and use 
the result as argument of e.g. a function. I read the GEP docu and know that 
I need anextra leading '0' index to dereference the global's
pointer.
So far so good.
But I can not succeed in using the getGetElementPtr result in constructing 
an initalizer for another global value (which expects a ConstantFP* and not 
a ConstantExpr*).

I guess I have a missconception, but I can't figure it out...

Alex

> But I can not succeed in using the getGetElementPtr result in constructing
> an initalizer for another global value (which expects a ConstantFP* and not
> a ConstantExpr*).
The result of a Constant GEP is going to be the address of a
floating-point value rather than the value itself, so you can't use it
to initialise something that expects a float. For example:

    @arr = constant [2 x float] [float 0.0, float 1.0]
    @bad = global float getelementptr([2 x float], [2 x float]* @arr,
i32 0, i32 1)
    @good = global float* getelementptr([2 x float], [2 x float]*
@arr, i32 0, i32 1)

LLVM doesn't really do value aliases, so you'd usually just write
"@bad = global float 1.0" for the second line. You might extract that
1.0 value by looking into the @arr definition directly using the LLVM
API -- if you can't do that (say because it's not a compile-time
constant) then the chances are what you're trying to write wouldn't be
valid anyway.

Cheers.

Tim.

Hi Tim,

I forgot to say that I try to do all this in the LLVM API, not IR.
I read in a bc'ed program and try to edit it using the API.

Alex


Tim Northover wrote:
>> But I can not succeed in using the getGetElementPtr result in
>> constructing an initalizer for another global value (which expects a
>> ConstantFP* and not a ConstantExpr*).
> 
> The result of a Constant GEP is going to be the address of a
> floating-point value rather than the value itself, so you can't use it
> to initialise something that expects a float. For example:
> 
>     @arr = constant [2 x float] [float 0.0, float 1.0]
>     @bad = global float getelementptr([2 x float], [2 x float]* @arr,
> i32 0, i32 1)
>     @good = global float* getelementptr([2 x float], [2 x float]*
> @arr, i32 0, i32 1)
> 
> LLVM doesn't really do value aliases, so you'd usually just write
> "@bad = global float 1.0" for the second line. You might extract
that
> 1.0 value by looking into the @arr definition directly using the LLVM
> API -- if you can't do that (say because it's not a compile-time
> constant) then the chances are what you're trying to write wouldn't
be
> valid anyway.
> 
> Cheers.
> 
> Tim.