On Friday, 20 May 2016 at 23:16:01 UTC, Richard Delorme wrote:
The source can be compiled with dmd, ldc or gdc, but the best
performance are obtained with the latter (almost twice faster).
Allowing LDC to do cross-module optimisations (by adding the
-singleobj flag) and make use of popcnt increased the performance
by 47% for one specific benchmark, which should bring it into the
same ballpark as GDC. See https://github.com/abulmo/amoeba/pull/2
for more details.
PGO might still give the latter a bit of an edge, though, as
LDC's implementation is still experimental and there is a lot of
unused potential for making use of the profile data in LLVM's