| 1 |
Date: Tue, 13 Feb 2001 18:25:42 -0600 |
| 2 |
From: Vikram S. Adve <[email protected]> |
| 3 |
To: Chris Lattner <[email protected]> |
| 4 |
Subject: RE: LLVM Concerns... |
| 5 |
|
| 6 |
> 1. Reference types |
| 7 |
> Right now, I've spec'd out the language to have a pointer type, which |
| 8 |
> works fine for lots of stuff... except that Java really has |
| 9 |
> references: constrained pointers that cannot be manipulated: added and |
| 10 |
> subtracted, moved, etc... Do we want to have a type like this? It |
| 11 |
> could be very nice for analysis (pointer always points to the start of |
| 12 |
> an object, etc...) and more closely matches Java semantics. The |
| 13 |
> pointer type would be kept for C++ like semantics. Through analysis, |
| 14 |
> C++ pointers could be promoted to references in the LLVM |
| 15 |
> representation. |
| 16 |
|
| 17 |
|
| 18 |
You're right, having references would be useful. Even for C++ the *static* |
| 19 |
compiler could generate references instead of pointers with fairly |
| 20 |
straightforward analysis. Let's include a reference type for now. But I'm |
| 21 |
also really concerned that LLVM is becoming big and complex and (perhaps) |
| 22 |
too high-level. After we get some initial performance results, we may have |
| 23 |
a clearer idea of what our goals should be and we should revisit this |
| 24 |
question then. |
| 25 |
|
| 26 |
> 2. Our "implicit" memory references in assembly language: |
| 27 |
> After thinking about it, this model has two problems: |
| 28 |
> A. If you do pointer analysis and realize that two stores are |
| 29 |
> independent and can share the same memory source object, |
| 30 |
|
| 31 |
not sure what you meant by "share the same memory source object" |
| 32 |
|
| 33 |
> there is |
| 34 |
> no way to represent this in either the bytecode or assembly. |
| 35 |
> B. When parsing assembly/bytecode, we effectively have to do a full |
| 36 |
> SSA generation/PHI node insertion pass to build the dependencies |
| 37 |
> when we don't want the "pinned" representation. This is not |
| 38 |
> cool. |
| 39 |
|
| 40 |
I understand the concern. But again, let's focus on the performance first |
| 41 |
and then look at the language design issues. E.g., it would be good to know |
| 42 |
how big the bytecode files are before expanding them further. I am pretty |
| 43 |
keen to explore the implications of LLVM for mobile devices. Both bytecode |
| 44 |
size and power consumption are important to consider there. |
| 45 |
|
| 46 |
--Vikram |
| 47 |
|