https://github.com/dlang/dmd/pull/6176
I'm happy to report that DMD has (finally!) gotten some significant new
optimizations! Specifically, 'slicing' a two register wide aggregate into two
register-sized variables, enabling much better enregistering.
Given the code:
void foo(int[] a, int[] b, int[] c) {
foreach (i; 0 .. a.length)
a[i] = b[i] + c[i];
}
the inner loop formerly compiled to:
LA: mov EAX,018h[ESP]
mov EDX,010h[ESP]
mov ECX,[EBX*4][EAX]
add ECX,[EBX*4][EDX]
mov ESI,020h[ESP]
mov [EBX*4][ESI],ECX
inc EBX
cmp EBX,01Ch[ESP]
jb LA
and now:
L1A: mov ECX,[EBX*4][EDI]
add ECX,[EBX*4][ESI]
mov 0[EBX*4][EBP],ECX
inc EBX
cmp EBX,EDX
jb L1A
I've been wanting to do this for years, and finally got around to it. (I also
thought of a simpler way to implement it, which helped a lot.)
Further work will be in widening what this applies to.
