Why is LDC so much slower than dmd...

[MartinNowak]

...and how can we make it faster?

The difference in compilation time is almost 2x, see dub Travis-CI.
I'm a little fed up by dmd's bad optimizer (and codegen bugs), but switching the compiler would be a much simpler choice if LDC compiled debug builds almost as fast as dmd.
I guess you guys know what the problems are. Have you already tried to optimize compile times?

[Trass3r]

#539
Almost all of the compile time is spent in llvm.

[HK47196]

FWIW, using a parallel makefile I find ldc to be -much- faster than DMD - LDC's optimized compile times can often beat DMD without optimization when using make. DMD seems to choke on input that isn't grouped together in one pass so it doesn't see much of a speedup at all from parallel make.

[MartinNowak]

Really?
That sounds weird because LDC uses the same frontend and also suffers from the overhead of separate compilation. With D's module system, it's inherently more efficient to compile multiple modules at once, because you common imports only get parsed once.

[Temtaime]

Maybe there's some unoptimized IR is passed to llvm ?
There should be a way to reduce compilation time.
I'm new with llvm but'll try to look.