In my emulator I have this function:
fn vram_read(self: *Self, addr: u16) u8 {
return self.vram[self.mirror_vram_addr(addr)];
}
vram is defined as vram: [2048]u8. In the profiler this simple function is taking 18% of the time because of a memcpy.
I asked codex where this memcpy was coming from and according to it, Zig generated code that was copying the 2KB array every time I read it from the array.
Codex suggested slicing the array (self.vram[0..][self.mirror_vram_addr(addr)] ) before reading from it to fix the issue. And, indeed that fixed the issue, the memcpy is entirely gone.
So why does Zig need a memcpy to read from the array?
The project was compiled with ReleaseFast.