2001-01-31-UniversalIRIdea.txt

Date: Wed, 31 Jan 2001 12:04:33 -0600
From: Vikram S. Adve <vadve@cs.uiuc.edu>
To: Chris Lattner <lattner@cs.uiuc.edu>
Subject: another thought

I have a budding idea about making LLVM a little more ambitious: a
customizable runtime system that can be used to implement language-specific
virtual machines for many different languages.  E.g., a C vm, a C++ vm, a
Java vm, a Lisp vm, ..

The idea would be that LLVM would provide a standard set of runtime features
(some low-level like standard assembly instructions with code generation and
static and runtime optimization; some higher-level like type-safety and
perhaps a garbage collection library).  Each language vm would select the
runtime features needed for that language, extending or customizing them as
needed.  Most of the machine-dependent code-generation and optimization
features as well as low-level machine-independent optimizations (like PRE)
could be provided by LLVM and should be sufficient for any language,
simplifying the language compiler.  (This would also help interoperability
between languages.)  Also, some or most of the higher-level
machine-independent features like type-safety and access safety should be
reusable by different languages, with minor extensions.  The language
compiler could then focus on language-specific analyses and optimizations.

The risk is that this sounds like a universal IR -- something that the
compiler community has tried and failed to develop for decades, and is
universally skeptical about.  No matter what we say, we won't be able to
convince anyone that we have a universal IR that will work.  We need to
think about whether LLVM is different or if has something novel that might
convince people.  E.g., the idea of providing a package of separable
features that different languages select from.  Also, using SSA with or
without type-safety as the intermediate representation.

One interesting starting point would be to discuss how a JVM would be
implemented on top of LLVM a bit more.  That might give us clues on how to
structure LLVM to support one or more language VMs.

--Vikram