This is an awesome case history of exactly what we are trying to achieve with D.
Any chance you could make an article out of this? Brief is ok. Essentially
showing the different language versions, then what you did to the D one that
kicked ass.
On 5/3/2011 8:42 PM, Adam D. Ruppe wrote:
Walter Bright wrote:
My experience in tuning D code for speed is that if you write "C++"
code in D, you will get the same performance as C++. However, D
makes it a *lot* easier to reorganize/refactor code to try and make
it run faster, and this makes for better algorithms and hence
faster code.
Anecdote: Friday, I was asked to justify my decision to use D to a
company exec.
He pointed me to some benchmarks that had ruby vs python vs php vs C++ and asked
how D would stack up. (The benchmark was sorting
an array of integers, nothing really fancy.)
Of them, Python had the shortest code, but C++ was the fastest, by
far. Not really surprising.
I first ported C++ to D. Equal lines of code, equal performance.
The optimized builds did show D lagging ~20% behind g++ though.
But, 20% behind g++ is still about 8x faster than the interpreted
languages, so not bad.
But, then I made a few minor tweaks. Slicing instead of copying.
Array operations and foreach instead of iterator loops, stuff like
that.
The D code was now the shortest of all - even beating Python by a
couple lines - while being just as readable and editable as the
"productivity languages".
And, to my surprise, the standard compile actually *beat* the
optimized C++ build, running twice as fast! I was expecting
it to be about equal, but instead, it just spanked it.
Shorter code, easier to read code *AND* faster code, it beat all
the competition in all the categories. Very pleasing result for
everyone involved.
Now, I dislike benchmarks as a general rule. Real world code is
a lot more complex than these trivial things realize. But hell,
when they work /for/ me, I won't complain.
(Also, while a weaker comparison, I find this same trend continues
to apps as a whole too. My D web app is about 1/8 the size in code
of the smallest open source PHP program I found with similar
features. 40% faster too. But Perpetually Hideous Programs are,
well, perpetually hideous. Beating PHP is no great feat.)
