On 4/28/11 12:57 PM, Jens Mueller wrote:
Walter Bright wrote:
On 4/28/2011 9:19 AM, Daniel Gibson wrote:
But there's no need for a D compiler to optimize loops that just copy
parts of an array into another array (and similar stuff), because in D
you use slices for that - they're (probably) faster and easier to use.
So IMHO it's fair to use slices where possible.
(And they're a language feature and not just part of the library)
Furthermore this particular benchmark is a "programming language
benchmark" and not a compiler benchmark, so it's fair to use every
feature of the language.
For example, a lot of effort is expended in C and Fortran compilers
"reverse engineering" loops so they can be recompiled and optimized
as vector operations. I don't see ever bothering with this in D
compilers, as D offers a vector notation.
If somebody wants to read the elaborate version of the above:
http://drdobbs.com/blogs/229300270
Walter's Dr. Dobb's articles are very good. I only read the new ones
posted. But I have to read the older ones if I find some time.
BTW
In the book "Coder at Work" Fran Allen takes a quite strong position
against C. She was deeply into Fortran compilers. She says: "We have
seriously regressed, since C developed. C has destroyed our ability to
advance the state of the art in automatic optimization, automatic
parallelization, automatic mapping of a high-level language to the
machine."
When I read this some time ago I found that position quite interesting.
But unfortunately I cannot tell how valid it is. Walter with all of his
experience in developing compilers may give a well informed evaluation
on Allen's statement. I'm just curious.
Jens
It's simple - Fortran can assume that all data is unaliased (although
aliasing is still technically possible). C, with its intensive
pointer-based ethos that makes many aliasing-related optimizations
difficult, has annoyed many a Fortran aficionados.
Andrei
