Convert 16x2 chars display to big TFT display

On 9/21/2021 11:49 PM, pozz wrote:
> Il 22/09/2021 04:34, Don Y ha scritto:
>> On 9/21/2021 3:31 AM, pozz wrote:
>> You will likely find there are issues that you've not even considered
>> that the original software relies on (directly or indirectly).
> 
> Yes, it could be possible.
> 
>> Can your emulation handle a *steady* stream of reads/writes?  Or,
>> are you expecting the host to be gentle with the display interface?
> 
> The host is a 16-bits MCU from Fujitsu. I will measure timings by an 
> oscilloscope, but I don't think it writes so fast.

But you will only be able to observe the activities that *happen*
while you are watching.  Without access to the sources, you won't know
if this is representative of every possible set of interactions with
the display (and, thus, YOUR display).

> And I can implement a busy flag trick: after any instruction, I can simulate a 
> busy state... if the host really use busy flag check (I suspect yes, because I 
> saw R/W signal goes high and low often).

Again, you are assuming that the host software always checks for this
and honors it.  Are you sure the developer didn't "forget" in some
cases?  (in which case, you have to emulate the "forgotten" case)?
Or, didn't implicitly rely on knowledge of how the (old) display
device would react in a given situation?  (e.g., "I know that if I
issue an XYZ command, the display controller will finish processing
it in 1.3 microseconds... and, the time it takes for my code to
advance to the next display access happens to be 1.8 microseconds
so no need to check the busy flag")

I can't know that the code does this.  Nor can you -- without
a look at the source code.

>> Keep in mind, that you have to run a fair bit of code to paint
>> pels onto the display from "character codes".
> 
> I will send sniffed character codes by a serial line to a "smart TFT".
> 
>> Does the host rely on the timing of the display's "busy" indication
>> in any way? 
> 
> Most probably the host polls for busy flag and I think I can replicate this.

See above.

>> Are there any bug in the display controller that
>> the host code works around -- that you'd have to emulate (but
>> aren't disclosed in the datasheet because they *are* bugs)?
> 
> I wrote some code for HD44780 in the past and I don't remember of any 
> workaround. Of course it depends on how many features of controller the host uses.

Which you won't know without a look through the code...

>> Are the visual characteristics of your display similar to those
>> of the original display?  Cursor/character blink and other
>> attributes?
> 
> If I can sniff these things, I think I can replicate on the TFT.
> 
>> Put a pair of back-to-back latches on the data bus and some
>> steering logic and don't sweat the fine-grained timing issues.
> 
> Yes, it could be done in this way.
> 
> And I was thinking to leave the original display connected and put the 
> interface as a passive sniffer. In this way I don't need to replicate the exact 
> behaviour of the controller, but only detect the stream of data and commands. 
> Of course, in this case I have to be fast enough as the controller (I can't 
> stop communication by busy flag).

I'm not sure I see what that buys you.  It certainly doesn't help
the cost.  And, the sniffed signals *should* be identical to your
emulation of them.  What will you do if there is a discrepancy,
at run time, 3 months from now?  Light a big "error" indicator?

If I was tackling something like this, I'd try to build a model of
the controller's internal state in the "emulator".  Then, design
a hardware interface that handled the fine-grained timing issues
(to decouple the emulator as much as possible).  Then, see how
fast I could get the interface between that hardware and the
*model* to run.

Then, in the background, run a task that examines the state of
the model and "paints" the new display, accordingly.  Hopefully,
the slippage won't cost you any features or result in display
artifacts that aren't already part of the original design
(e.g., updating the contents of the display asynchronously
wrt it's being painted into the hardware)

On Tuesday, September 21, 2021 at 10:34:52 PM UTC-4, Don Y wrote:
> On 9/21/2021 3:31 AM, pozz wrote: 
> > I have a CPU board that interfaces to a standard 16x2 chars display, based on 
> > HD44780 controller. 
> > 
> > I'd like to replace this display with a full-color big TFT display, maybe 10", 
> > but I can't change the firmware of the CPU (because I don't have the source 
> > code and I can't rewrite it, because of cost and time). 
> > 
> > The simple idea is to create an interface between the current CPU board and the 
> > TFT display. This interface should be able to detect the signals from the CPU 
> > and that was directed to HD44780, interpret the instructions (write CGRAM/DDRAM 
> > and so on) and write the same things to the big TFT display. 
> > 
> > Of course, exactly the same texts will be written on TFT display (of course 
> > with a bigger and visible font) and this is acceptable. 
> > 
> > So the MCU on the interface should be able to read 8 data bits, RS and R/W 
> > signals syncronous to signal E. 
> > 
> > I'm thinking to configure E signal as an external interrupt on the falling 
> > edge. In the interrupt, I read data bits, R/W and RS and decode the instruction 
> > or data. 
> > 
> > Any suggestions? Maybe a similar code is already online.
> You will likely find there are issues that you've not even considered 
> that the original software relies on (directly or indirectly). 
> 
> Can your emulation handle a *steady* stream of reads/writes? Or, 
> are you expecting the host to be gentle with the display interface? 
> Keep in mind, that you have to run a fair bit of code to paint 
> pels onto the display from "character codes". 
> 
> Does the host rely on the timing of the display's "busy" indication 
> in any way? Are there any bug in the display controller that 
> the host code works around -- that you'd have to emulate (but 
> aren't disclosed in the datasheet because they *are* bugs)? 
> Are the visual characteristics of your display similar to those 
> of the original display? Cursor/character blink and other 
> attributes? 
> 
> Put a pair of back-to-back latches on the data bus and some 
> steering logic and don't sweat the fine-grained timing issues.

No need for adding hardware.  Read the data lines and the enable signal (and R/W, RS if you want) in a single read operation.  Save the previous value in a memory location and watch the new value for E to drop from 1 to 0.  Then read the previous sample where E was high.  If your reads are fast enough it will contain valid write data.  The tDSW is only 80 ns with 5V power, so the interface reads will need to be faster than 12.5 MHz to assure valid data.  With a 3.3V unit the timing is 195 ns, so a bit faster than 5 MHz will do the job.  If you are emulating a 4 bit data bus the entire set of signals fits in a single byte port. 

Otherwise a rather simple FPGA can do the job.  If you want to emulate the full set of features of the HD44780 it might take a bit of work since it does some fancy cursor operations and has an internal buffer for user defined characters.  

The HD44780 has a few bugs that require software work arounds in the drivers.  The big one is the reset issue that requires an arcane sequence to initialize the device.  This does not need to be emulated since the work around sequence will work even if the chip doesn't have the bug.  The operation of the cursor and character writing is a bit clumsy, but should not be hard to emulate.  I don't recall any timing you will be obligated to emulate other than meeting the bus speeds which is obvious.  The various delays in the chip are handled in the interface software either by reading a flag or simply by waiting some minimum time.  If your emulation runs faster it won't matter.  I'm pretty sure there is no "too fast", but it's been a year or so since I worked with it and don't recall every detail. 

-- 

Rick C.

- Get 1,000 miles of free Supercharging
- Tesla referral code - https://ts.la/richard11209

On Wednesday, September 22, 2021 at 2:49:47 AM UTC-4, pozz wrote:
> Il 22/09/2021 04:34, Don Y ha scritto: 
> > On 9/21/2021 3:31 AM, pozz wrote: 
> >> I have a CPU board that interfaces to a standard 16x2 chars display, 
> >> based on HD44780 controller. 
> >> 
> >> I'd like to replace this display with a full-color big TFT display, 
> >> maybe 10", but I can't change the firmware of the CPU (because I don't 
> >> have the source code and I can't rewrite it, because of cost and time). 
> >> 
> >> The simple idea is to create an interface between the current CPU 
> >> board and the TFT display. This interface should be able to detect the 
> >> signals from the CPU and that was directed to HD44780, interpret the 
> >> instructions (write CGRAM/DDRAM and so on) and write the same things 
> >> to the big TFT display. 
> >> 
> >> Of course, exactly the same texts will be written on TFT display (of 
> >> course with a bigger and visible font) and this is acceptable. 
> >> 
> >> So the MCU on the interface should be able to read 8 data bits, RS and 
> >> R/W signals syncronous to signal E. 
> >> 
> >> I'm thinking to configure E signal as an external interrupt on the 
> >> falling edge. In the interrupt, I read data bits, R/W and RS and 
> >> decode the instruction or data. 
> >> 
> >> Any suggestions? Maybe a similar code is already online. 
> > 
> > You will likely find there are issues that you've not even considered 
> > that the original software relies on (directly or indirectly).
> Yes, it could be possible.
> > Can your emulation handle a *steady* stream of reads/writes? Or, 
> > are you expecting the host to be gentle with the display interface?
> The host is a 16-bits MCU from Fujitsu. I will measure timings by an 
> oscilloscope, but I don't think it writes so fast. 
> And I can implement a busy flag trick: after any instruction, I can 
> simulate a busy state... if the host really use busy flag check (I 
> suspect yes, because I saw R/W signal goes high and low often).
> > Keep in mind, that you have to run a fair bit of code to paint 
> > pels onto the display from "character codes".
> I will send sniffed character codes by a serial line to a "smart TFT".
> > Does the host rely on the timing of the display's "busy" indication 
> > in any way?
> Most probably the host polls for busy flag and I think I can replicate this.
> > Are there any bug in the display controller that 
> > the host code works around -- that you'd have to emulate (but 
> > aren't disclosed in the datasheet because they *are* bugs)?
> I wrote some code for HD44780 in the past and I don't remember of any 
> workaround. Of course it depends on how many features of controller the 
> host uses.
> > Are the visual characteristics of your display similar to those 
> > of the original display? Cursor/character blink and other 
> > attributes?
> If I can sniff these things, I think I can replicate on the TFT.
> > Put a pair of back-to-back latches on the data bus and some 
> > steering logic and don't sweat the fine-grained timing issues.
> Yes, it could be done in this way. 
> 
> And I was thinking to leave the original display connected and put the 
> interface as a passive sniffer. In this way I don't need to replicate 
> the exact behaviour of the controller, but only detect the stream of 
> data and commands. Of course, in this case I have to be fast enough as 
> the controller (I can't stop communication by busy flag).

The timing of the HD44780U and the various emulations is specified assuming a given internal clock speed set by an RC.  If you look at the timing details of the instructions as the clock speed varies you can get timings that can let software work without reading the READY flag.  The data sheet timings are based on the ideal timing of the internal clock and are not accurate for production variations as well as temperature and voltage.  But since there is never any harm in the controller working faster than the given timings, you don't really need to worry with this as long as your MCU runs faster than the HD44780U with a 270 kHz clock.  I can't think of a reason why your emulator device would not work much faster than the HD44780U.  

-- 

Rick C.

+ Get 1,000 miles of free Supercharging
+ Tesla referral code - https://ts.la/richard11209

On 2021-09-21, pozz <pozzugno@gmail.com> wrote:
> The simple idea is to create an interface between the current CPU board 
> and the TFT display. This interface should be able to detect the signals 
> from the CPU and that was directed to HD44780, interpret the 
> instructions (write CGRAM/DDRAM and so on) and write the same things to 
> the big TFT display.
[...]
> Any suggestions? Maybe a similar code is already online.

It's not HD44780, but

https://github.com/bigtreetech/BIGTREETECH-TFT35-V3.0

does exactly that for a 12864 LCD (which has a parallel interface with
similar timing requirements). Look for "Marlin Mode" in the source code.

cu
Michael