softusb: minimize time between SETUP/OUT and DATAx

By not waiting until the transmitter is idle, we can reduce the
time between the end of the SE0 indicating EOP of SETUP or OUT
and the first transition of the following DATAx to a mere two
bit times (as oppose to ~6 bit times before).

While I haven't been able to find anything in the USB standard
that would require such tight timing, AVR USB chips (i.e., the
ATmega32U2 of atusb and the AT90USB162 of the Faderfox LV3)
flat out ignore any transfers with longer delays.

softusb-input/main.c

@@ -119,7 +119,7 @@ static void make_usb_token(unsigned char pid, unsigned int elevenbits, unsigned
 	out[2] |= usb_crc5(out[1], out[2]) << 3;
 }
 
-static void usb_tx(const unsigned char *buf, unsigned char len)
+static void usb_tx_nowait(const unsigned char *buf, unsigned char len)
 {
 	unsigned char i;
 
@@ -130,9 +130,15 @@ static void usb_tx(const unsigned char *buf, unsigned char len)
 	}
 	while(rio8(SIE_TX_PENDING));
 	wio8(SIE_TX_VALID, 0);
+}
+
+static void usb_tx(const unsigned char *buf, unsigned char len)
+{
+	usb_tx_nowait(buf, len);
 	while(rio8(SIE_TX_BUSY));
 }
 
+
 static inline void usb_ack(void)
 {
 	wio8(SIE_TX_DATA, 0x80); /* send SYNC */
@@ -302,7 +308,7 @@ static char usb_out(unsigned addr, const unsigned char *buf, unsigned char len)
 
 	/* send OUT */
 	make_usb_token(USB_PID_OUT, addr, out);
-	usb_tx(out, 3);
+	usb_tx_nowait(out, 3);
 
 	/* send DATAx */
 	usb_tx(buf, len);
@@ -346,7 +352,7 @@ static int control_transfer(unsigned char addr, struct setup_packet *p,
 	TRIGGER_ON();
 
 	/* send them back-to-back */
-	usb_tx(setup, 3);
+	usb_tx_nowait(setup, 3);
 	usb_tx(usb_buffer, 11);
 
 	TRIGGER_OFF();