Re: A fast scheme implementation?
- From: Andrew Reilly <andrew-newspost@xxxxxxxxxxxxxxxxxxxxx>
- Date: 14 Dec 2007 00:57:56 GMT
On Thu, 13 Dec 2007 14:47:26 -0800, William D Clinger wrote:
Not entirely true. If the compiler squirrels away a serialized
representation of its IR along with the object code, then when invoked
to do linking it can perform whole-program optimizations.
Such things have been known in commercial C compilers for some time;
more recently the LLVM project has been working in this area, and a
separate effort is known to exist for GCC.
I stand corrected. Should I count the time spent waiting for this to be
implemented in gcc as part of the compile time?
Only if the competition was specific to GCC, rather than just "C". I
remember using a MIPS toolchain back in the late 80s/early 90s that had a
mode where it would emit some sort of half-compiled tree representation,
rather than object code, and the "linker" would then do whole-program
optimization. I never used it, because it was (a) slow, and (b) I tended
to compile all of my programs in a "whole program" way anyway, they
didn't depend much on separately compiled libraries.
It would be interesting to see how well a modern Java implementation
managed it, too. One of the things that current versions of HotSpot are
supposed to do is inlining and specialization.
Cheers,
--
Andrew
.
- Follow-Ups:
- Re: A fast scheme implementation?
- From: Jeffrey Mark Siskind
- Re: A fast scheme implementation?
- References:
- A fast scheme implementation?
- From: duaneb
- Re: A fast scheme implementation?
- From: William D Clinger
- Re: A fast scheme implementation?
- From: Barak A. Pearlmutter
- Re: A fast scheme implementation?
- From: Joel J. Adamson
- Re: A fast scheme implementation?
- From: Barak A. Pearlmutter
- Re: A fast scheme implementation?
- From: Adrian
- Re: A fast scheme implementation?
- From: William D Clinger
- Re: A fast scheme implementation?
- From: Jed Davis
- Re: A fast scheme implementation?
- From: William D Clinger
- A fast scheme implementation?
- Prev by Date: Re: A fast scheme implementation?
- Next by Date: Re: A fast scheme implementation?
- Previous by thread: Re: A fast scheme implementation?
- Next by thread: Re: A fast scheme implementation?
- Index(es):
Relevant Pages
|