Fix paintScreen() for IDE 1.6.10

paintScreen() used instruction padding for a delay to allow a byte to be sent
to the display using SPI. This delay became to short with the new gcc version
in Arduino IDE 1.6.10

The function has been changed to test for the serial transfer to complete
before sending the next byte.

src/ArduboyCore.cpp

@@ -244,20 +244,26 @@ void ArduboyCore::paintScreen(const uint8_t *image)
 // will be used by any buffer based subclass
 void ArduboyCore::paintScreen(uint8_t image[])
 {
-  for (int i = 0; i < (HEIGHT*WIDTH)/8; i++)
+  uint8_t c;
-  {
+  int i = 0;
-    // SPI.transfer(image[i]);
 
-    // we need to burn 18 cycles between sets of SPDR
+  SPDR = image[i++]; // set the first SPI data byte to get things started
-    // 4 clock cycles
+
-    SPDR = image[i];
+  // the code to iterate the loop and get the next byte from the buffer is
-    // 7 clock cycles
+  // executed while the previous byte is being sent out by the SPI controller
-    asm volatile(
+  while (i < (HEIGHT * WIDTH) / 8)
-      "mul __zero_reg__, __zero_reg__ \n" // 2 cycles
+  {
-      "mul __zero_reg__, __zero_reg__ \n" // 2 cycles
+    // get the next byte. It's put in a local variable so it can be sent as
-      "mul __zero_reg__, __zero_reg__ \n" // 2 cycles
+    // as soon as possible after the sending of the previous byte has completed
-      );
+    c = image[i++];
+
+    while (!(SPSR & _BV(SPIF))) { } // wait for the previous byte to be sent
+
+    // put the next byte in the SPI data register. The SPI controller will
+    // clock it out while the loop continues and gets the next byte ready
+    SPDR = c;
   }
+  while (!(SPSR & _BV(SPIF))) { } // wait for the last byte to be sent
 }
 
 void ArduboyCore::blank()