On 28.04.2011 18:19, Daniel Gibson wrote: > 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.
But applicability of language features is a bit limited, I would say. I think, this is not exactly fair - to take any domain-specific task (like "matrix multiplication") and use it as a "language benchmark". In this particular case (matrix multiplication) - yes, we may gain from specific features. But another example - I need to write MIME headers parses in D - this involves (mostly) scanning of strings, comparisons etc - it is hardly possible to use some feature which could really help and will be fast (std.string is mostly slow as it is a bit too generic, regexps are even more slow). So, I've to write my own version - using only core features of the language (as there is no highly-optimized nor any version in standard library anyway). Now, when this is done - what it will be - "compiler benchmark" or "language benchmark"? To me, of course it does matter - how good is generated code, as it directly affect performance of my application. /Alexander