On 3/31/20 5:30 PM, data pulverizer wrote:
I've observed large differences in timing performance while filling arrays using different methods (for vs foreach vs arr[] = x) and don't know why. I've looked at array.d (https://github.com/dlang/dmd/blob/9792735c82ac997d11d7fe6c3d6c604389b3f5bd/src/dmd/root/array.d) but I'm still none the wiser.
for: you are indexing an array on each step, incurring a bounds check, and also doing a pointer calculation on each access.
foreach: Better, iterates a pointer over the whole array, so it's going to be faster.
slice assign: Best, this uses whatever tricks are possible in the given architecture. Most likely memset, or memcpy. These are insanely optimized functions, which is why they perform so well.
-Steve
