[gclist] Prior work?

Charles Fiterman cef@geodesic.com
Wed, 09 Feb 2000 13:11:41 -0600


At 06:55 PM 2/8/00 -0500, you wrote:
>
>I'm interested in any prior work to tailoring a GC for a particular
>program. I'm only aware of work by Stephen Thomas and Richard Jones for Lazy
>ML. I'd like to know if anyone has tried this on a larger scale, and what
>the performance tradeoffs are.

The real work is tailoring the compiler to the GC not the other way around.

1. Zeroing unused pointers.

2. Custom mark routines for things like stacks.

3. Inline object creation including partial evaluating the creation of
execution frames at compile time.

4. Adjusting allocation for cache efficiency. e.g. the object is 56 bytes
long but if you give 64 bytes it will end up cache aligned. (Assume 30
cycles for divide and 256 for a cache miss)

Life is a lot simpler in languages with single inheritance.