--- Comment #6 from 2012-10-04 15:50:29 PDT ---
(In reply to comment #5)

> could you please compare again against everything?

I have re-compiled druntime and Phobos (but not dmd, because in the meantime it
has not changed), and my run-time for the write1 with N=2_000_000 is about the
same. The run-time with N=10_000_000 seems a little lower, 5.55 seconds (but
changes with similar magnitude are sometimes just noise, so it's not

Those timings are just one data point (32 bit Windows, Core 2). If we you want
to optimize textual output, then probably more data points are needed, with a
Core i7, on 64 bits, and on a Linux. And probably a profiling is useful.


The optimization stages of GCC and Clang contain an algorithm that turn small
dividend (division and modulus) operations in a faster combination of simpler
operations. This is handy in all kind of code, and it's also quite useful in
code like:

auto c = cast(uint) (v % 10);
v /= 10;

Some info:

Configure issuemail:
------- You are receiving this mail because: -------

Reply via email to