Re: [omr-dev] Is this correct IL codegen?

["Mark Stoodley" <mstoodle@xxxxxxxxxx>]

Hi Dibyendu,

> I guess
part of the question is whether indirect load/store IL
supports array offsets of the form base[index] or do I need to compute
the address?

The first operand of the indirect load/store
for an array access is the address of the
element that should be accessed. As
such, it's up to the IL to do the multiplication
of the index by the size of each element.
We do that to give the compiler visibility
of the calculations so it can optimize
them where possible.

So, if your "indexNode" parameters
come in pre-multiplied, then it looks roughly
right, though I confess I didn't check
your code in great detail. But I suspect you
aren't expecting pre-multiplied indexNode
parameters, so in that case you'll want
to generate something like the following
for the first child:
        aiadd
         
      <your base address>
         
      imul
         
              <your
index>
         
              iconst
<TR::DataType;:getSize(dt)>

That's for 32-bit indices, of course.
In a 64-bit system, you'd normally use aladd,
lmul, and lconst . Depending on how
flexible you want to be, you may need to
do some integer conversion from the
data type of your index node to the expected
integer type of the index calculation
(i or l). Make sense?

You're probably working with C arrays,
so no header to add in, but if there were one
you would also have to add that into
the first operand of the xaload|xastore .

Hope that helps!

Mark Stoodley		8200 Warden Avenue
Senior Software Developer		Markham, L6G 1C7
IBM Runtime Technologies		Canada
Phone:	+1-905-413-5831
e-mail:	mstoodle@xxxxxxxxxx
We cannot solve our problems with the same thinking we used when we created them - Albert Einstein