2001-02-13-Reference-MemoryResponse.txt

Date: Tue, 13 Feb 2001 18:25:42 -0600
From: Vikram S. Adve <vadve@cs.uiuc.edu>
To: Chris Lattner <sabre@nondot.org>
Subject: RE: LLVM Concerns...

> 1. Reference types
>    Right now, I've spec'd out the language to have a pointer type, which
>    works fine for lots of stuff... except that Java really has
>    references: constrained pointers that cannot be manipulated: added and
>    subtracted, moved, etc... Do we want to have a type like this?  It
>    could be very nice for analysis (pointer always points to the start of
>    an object, etc...) and more closely matches Java semantics.  The
>    pointer type would be kept for C++ like semantics.  Through analysis,
>    C++ pointers could be promoted to references in the LLVM
>    representation.


You're right, having references would be useful.  Even for C++ the *static*
compiler could generate references instead of pointers with fairly
straightforward analysis.  Let's include a reference type for now.  But I'm
also really concerned that LLVM is becoming big and complex and (perhaps)
too high-level.  After we get some initial performance results, we may have
a clearer idea of what our goals should be and we should revisit this
question then.

> 2. Our "implicit" memory references in assembly language:
>    After thinking about it, this model has two problems:
>       A. If you do pointer analysis and realize that two stores are
>          independent and can share the same memory source object,

not sure what you meant by "share the same memory source object"

> there is
>          no way to represent this in either the bytecode or assembly.
>       B. When parsing assembly/bytecode, we effectively have to do a full
>          SSA generation/PHI node insertion pass to build the dependencies
>          when we don't want the "pinned" representation.  This is not
>          cool.

I understand the concern.  But again, let's focus on the performance first
and then look at the language design issues.  E.g., it would be good to know
how big the bytecode files are before expanding them further.  I am pretty
keen to explore the implications of LLVM for mobile devices.  Both bytecode
size and power consumption are important to consider there.

--Vikram