Add delayMicroseconds

This isn't a good version of it, but it has the capability of delaying a
lot shorter times than `delay` can. We could use the register too, but
right now this is fine. Really just need to find a better way to get the
necessary assembly and a nice way to measure 16MHz pulses.

src/LiquidCrystal.zig

@@ -143,7 +143,7 @@ pub fn LiquidCrystal(comptime rs: u8, comptime enable: u8, comptime d0: u8, comp
             write4bits(0x03);
             time.delay(5);
             write4bits(0x03);
-            time.delay(1);
+            time.delayMicroseconds(150);
             write4bits(0x02);
 
             command(function_set | lcd_config.functionBits());
@@ -292,10 +292,10 @@ pub fn LiquidCrystal(comptime rs: u8, comptime enable: u8, comptime d0: u8, comp
         fn pulseEnable() void {
             gpio.digitalWrite(enable_pin, .low);
             // TODO: This should be 1 us.. but we're slow here.
-            time.delay(1);
+            time.delayMicroseconds(1);
             gpio.digitalWrite(enable_pin, .high);
             // Enable just needs to be high for >450 ns so we good.
-            time.delay(1);
+            time.delayMicroseconds(1);
             gpio.digitalWrite(enable_pin, .low);
             // This'll be 100us not 1us like others to settle
             time.delay(1);

src/time.zig

@@ -72,3 +72,33 @@ pub fn delay(ms: u64) void {
         }
     }
 }
+
+// u32 not u64 to easily get signed version so we can have more efficient
+// loop later. This is really trial and error and should be done better when
+// I am able.
+pub fn delayMicroseconds(us: u32) void {
+    // TODO: Care about clock frequency
+    // We'll assume <= 1us overhead of function call is enough
+    if (us <= 1) return;
+
+    // This is signed so later we can subtract an amount larger than 1
+    // and check if it's greater than 0 without underflow.
+    var us_var: i64 = us;
+
+    // Trying to get the loop later exactly right... and this x4 to get
+    // us to actually be us and not ms, but this was trial and error anyway
+    us_var <<= 2;
+    // Subtract an amount of time. Arduino libraries have 5 so I'll do that,
+    // but I think copying the variable will add a load in there. But then
+    // we aren't doing the shift first at the moment so it may even out.
+    us_var -= 5;
+
+    // After messing around with inline ASM for forever, I think Zig/LLVM
+    // don't fully work with AVR inline assembly constraints like =w - or
+    // probably any 32 bit or 64 bit types. Instead, just try out a busy
+    // wait with a loop.
+    while (us_var > 0) {
+        us_var -= 9;
+        asm volatile ("nop");
+    }
+}