[Issue 16200] Faster pow implementation for integral exponents

d-bugmail at puremagic.com d-bugmail at puremagic.com
Tue Feb 2 11:10:25 UTC 2021


https://issues.dlang.org/show_bug.cgi?id=16200

Berni44 <bugzilla at bernis-buecher.de> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |bugzilla at bernis-buecher.de

--- Comment #2 from Berni44 <bugzilla at bernis-buecher.de> ---
Results on my computer:

dmd:
0 1 sec, 331 ms, 327 μs, and 6 hnsecs
1 2 secs, 940 ms, 293 μs, and 9 hnsecs
2 7 secs, 428 ms, 953 μs, and 5 hnsecs

ldc:
0 330 ms, 3 μs, and 2 hnsecs
1 117 ms, 824 μs, and 1 hnsec
2 1 sec, 324 ms, 530 μs, and 6 hnsecs

IMHO the program compares apples with oranges, that is doubles with reals.
Replacing double with real I get:

dmd:
0 2 secs, 335 ms, 325 μs, and 8 hnsecs
1 2 secs, 942 ms, 572 μs, and 9 hnsecs
2 7 secs, 421 ms, 349 μs, and 3 hnsecs

ldc:
0 530 ms, 947 μs, and 1 hnsec
1 117 ms, 768 μs, and 1 hnsec
2 1 sec, 324 ms, 124 μs, and 9 hnsecs

Not so clear, that the new algorithm is superior...

(I did a lot of more speed checking with these algorithms - I will write the
results down and post a link here, once I'm done.)

See also https://github.com/dlang/phobos/pull/7783 for results on checking
correctness.

--


More information about the Digitalmars-d-bugs mailing list