Re: FW: [avrdude-dev] Windows port (was: Butterfly proposal: Exit upgradefunction)

[Jan-Hinnerk Reichert]

Hi again,

I have done some tests and reading on the changes regarding 
serial-ports. I even read stuff in the microsoft database ;-(

The serial open should be safe the way it is ;-)

Now for the speed testing. I have just done read tests on an AVR910. 
Benefits on STK500 should be unnoticeable, because of the better 
protocol.

Reading flash of ATmega161:
avrdude 4.3.0, linux "-low_latency": 81.93s
avrdude 4.3.0, linux "low_latency":  31.60s
avrdude 4.3.0, windows:              82.32s
avrdude-Martin:                      31.59s

So the use of Windows-API is equivalent to switching serial ports to 
low latency under Linux. This is a great benefit.

In addition this makes it possible to gain even more speed by reducing 
the FIFO-threshold.

avrdude-Martin, receive_fifo_threshold=1: 14.53s

Although it looks like the udelay()-stuff needs some further 
investigation, I think we should merge the changes regarding serial 
interface as soon as possible.

BTW: The FIFO-thresholds can be adjusted in Device-Manager (hope this 
is the right term). Although it doesn't say so, WinXP needs a restart 
before changes take effect ;-(

Cheers
  Jan-Hinnerk