drivers: st7789: Optimize RLE decoding loop

Migrate the filling of the line buffer into a seperate function.
This does naturally reduce the cost of the loop management but
much more importantly allows us to use viper native code
generator.

wasp/boards/simulator/micropython.py

@@ -1,5 +1,16 @@
-def const(x):
+def const(fn):
-    return x
+    return fn
+
+def native(fn):
+    return fn
+
+def viper(fn):
+    def ptr8(buf):
+        return buf
+
+    # This is a bit of a hack since the scope for ptr8 won't be right
+    # but it does mean no changes to the client
+    fn.__globals__['ptr8'] = ptr8
+
+    return fn
 
-def native(x):
-    return x

wasp/drivers/st7789.py

@@ -19,6 +19,16 @@ _RAMWR              = const(0x2c)
 _COLMOD             = const(0x3a)
 _MADCTL             = const(0x36)
 
+@micropython.viper
+def fastfill(mv, color: int, count: int, offset: int):
+    p = ptr8(mv)
+    colorhi = color >> 8
+    colorlo = color & 0xff
+
+    for x in range(count):
+        p[2*(x+offset)    ] = colorhi
+        p[2*(x+offset) + 1] = colorlo
+
 class ST7789(object):
     def __init__(self, width, height):
         self.width = width
@@ -111,12 +121,12 @@ class ST7789(object):
 
         for rl in rle:
             while rl:
-                buf[bp] = color >> 8
+                count = min(sx - bp, rl)
-                buf[bp+1] = color & 0xff
+                fastfill(buf, color, count, bp)
-                bp += 2
+                bp += count
-                rl -= 1
+                rl -= count
 
-                if bp >= (2*sx):
+                if bp >= sx:
                     self.write_data(buf)
                     bp = 0