[Issue 5650] A RedBlackTree performance problem

[d-bugmail at puremagic.com]

http://d.puremagic.com/issues/show_bug.cgi?id=5650



--- Comment #27 from bearophile_hugs at eml.cc 2013-10-08 12:11:30 PDT ---
(In reply to comment #26)

> The current situation on my PC is the following:
> 
> [1]   Java version: 0.26
> [2]   D version: 0.28
> [2]   D version + GC.disable() at the top: 0.28
> [3]   D version: 0.45
> [3]   D version + GC.disable() at the top: 0.21
> [4]   C++ version: 0.11
> [5]   C++ version: 0.12


My current timings are improved over the #Comment16 (32 bit OS), this is good:


D:    1.13 (dmd, -O -release -inline -noboundscheck)
D:    1.11 (ldmd2, same switches)
D:    0.56 (No GC)
D:    0.51 (No GC, ldc2)
Java: 0.46
C++:  0.29 (GCC 4.9, g++ -Ofast -s -flto)


> There are other reports for more specific RedBlack tree performance issues.

What ones?


> The performance seems to be in the ballpark you were looking for (on my PC.)
> Can this bug be closed?

Here Java is about as fast as D without GC, and about twice faster than D with
GC. C++ is about four times faster than D with GC. (a missing timing is the D
code compiled with ldc2 that allocates from the C heap (using or not using an
arena allocator)).

Is four times slower than C++ fast enough to close this bug report? Is such
lower performance going to be gained back using an arena allocator from the
Andrei allocators?

The situation is better than before, but perhaps it's not yet the time to close
the bug report.

-- 
Configure issuemail: http://d.puremagic.com/issues/userprefs.cgi?tab=email
------- You are receiving this mail because: -------