Teensy 4.0 First Beta Test

Status
Not open for further replies.
defragster said:
Kurt - that example shows answers to some questions I had - passing the &Serial# nicely. Will give it a look.

Question - what is the Rx and Tx connect scheme for the ports?
Click to expand...
Currently T1->r2, t2->R3, t3->r5, t5->r6, t6->r7, t7->R8, T8->R1
Note I skipped Serial4
Serial4 I have hooked up to PC with USB to Serial adapter... Currently Adafruit CP2104 version...
 
I tried 12.5mhz but that didn't work. Will try 4 and 8Mhz as a test then play with delay. May have to dig deeper into what they are doing with SPI in their library. Now to rehook up the BME280 chip.
Click to expand...
Another possible approach would be to compare T3.2 scope of SPI CLK talking to the BME820 to the scope of T4 SPI CLK and see if you can notice gaps or delays that are controlled by CCR.

Yes, Paul added code for setting SPI CLK, though I didn't know he updated the deprecated setFrequency. The current T4 SPI lib does not have code to set polarity and bit order, so i'm not surprised the deprecated functions haven't been updated. Registers default to Mode0 and MSB which is what most devices use (including BME20). With T4 so fast, it still might be a problem with setting CS, so a delayMicros(10) after CS HIGH and before CS LOW might work, and maybe delays between spixfer() ....
 
Kurt -- Tim

Since SD.h is working decided to hook up my EMIC2 text-2-speech board to the T4 and put a file on the sd card to be played by the EMIC2, figure good verification for both. Anyway, it worked, it found the card, the correct file and played it. I am attaching a recording from my phone (frankly its not a very good recording), need something to do better recordings, just for fun.View attachment Voice 001.zip

I have it hooked up to an amplified speaker so getting some static. I modified the library awhile ago to work with Teensies, don't even know if it would work on a avr board anymore :)
 
KurtE said:
Serial port area: You have 6 pins per Serial port, With RX/TX/GND - Wondering what the other three pin are...
Click to expand...

On the breakout board, the other 3 pins have no connection.

I supposed I could have used only 3 pin headers and wires, but I used 6 pins with 3 unused so this FTDI cable with a 6 pin socket can plug in.

TTL-232R1.jpg
 
Another possible approach would be to compare T3.2 scope of SPI CLK talking to the BME820 to the scope of T4 SPI CLK and see if you can notice gaps or delays that are controlled by CCR.

Yes, Paul added code for setting SPI CLK, though I didn't know he updated the deprecated setFrequency. The current T4 SPI lib does not have code to set polarity and bit order, so i'm not surprised the deprecated functions haven't been updated. Registers default to Mode0 and MSB which is what most devices use (including BME20). With T4 so fast, it still might be a problem with setting CS, so a delayMicros(10) after CS HIGH and before CS LOW might work, and maybe delays between spixfer() ....
Click to expand...
Will play with the BME280 more tomorrow - getting tired and getting late.
 
KurtE said:
Currently T1->r2, t2->R3, t3->r5, t5->r6, t6->r7, t7->R8, T8->R1
Note I skipped Serial4
Serial4 I have hooked up to PC with USB to Serial adapter... Currently Adafruit CP2104 version...
Click to expand...

Thanks Kurt, Okay that is how I read the open comment yesterday - then I wired 8 serials as loopback with each Tx><Rx and - prepared to edit the code - but that worked with FrankB's code - so I went with it as it was simple to not wire but jumper :) I even started to expand the code to have loopBack() and loopCross() in the same code to do either selectively with the same sketch and had to remove that ...

You version does not work as wired so will get wires out. I've hacked out debug as it wasn't good for much except 'status spew' as USB was working. I put it through a Teensy as that seemed way more cool.

FrankB - I'd vote for the EvResponder.disable()- taking out overhead and just doing stuff that could be done as well or better under user control. With a T4 leaving loop so many more times/sec seems overkill - instead of 1-2 million on T_3.x it can be 10-20 million loop() exits/sec. Maybe it should have an .enable() in case a user wants to selectively pause it during some portion of the code.
 
defragster said:
You version does not work as wired so will get wires out.
Click to expand...

What I did, and figured out that my T4 was lifting off, was start off the program and have T1 connected to R1, Should do the send/receive no errors... Assuming that works hook up
T1->R2, T2->R1 see if it still works, move from R1, to R3 and then try T3 to R1... and if works move from R1 to R5...
 
For watchdog, should we make a library or make it part of the core? I updated the definitions in imxrt.h and submitted a pull request, and have working code, and enabled the vector table for the callback

Note:
Watchdog Interrupt Count Time-out (WICT) field determines, how long before the counter time-out must
the interrupt occur. The reset value is 0x04 implies interrupt will occur 2 seconds before time-out. The
maximum value that can be programmed to WICT field is 127.5 seconds with a resolution of 0.5 seconds.
Click to expand...

I keep it set at 2 seconds, and able to print to Serial4 output before the reset occurs. This very good to safely handle tasks before reset. :)

Basically with a watchdog of 10 seconds and interrupt of 2 seconds before, the interrupt fires at 8 seconds and final reset happens at 10 seconds. Theoretically your timeout is more like 8 seconds max to feed the dog. :)
 
Last edited:
KurtE said:
What I did, and figured out that my T4 was lifting off, was start off the program and have T1 connected to R1, Should do the send/receive no errors... Assuming that works hook up
T1->R2, T2->R1 see if it still works, move from R1, to R3 and then try T3 to R1... and if works move from R1 to R5...
Click to expand...

The problem is my Jumper of each port looping back to itself - I haven't grabbed the needed wires to chain them along yet - just had an idea for an easy to remove scheme ... but the row to row spacing it not on 0.1" spacing :(

Found set of short F to F jumper wires ... now I need to copy this machine to my laptop ...

Update: Jumper wires installed and working perfectly - #define SPD 2400000 Delta is 2 or 3 ms:
Loop: 62011 Delta time: 2, CB Out: 320, CB In: 320
Click to expand...
Output to Serial and Serial4 included in line ports 1 to 8 with debug off.
 
Last edited:
mjs513 said:
Will play with the BME280 more tomorrow - getting tired and getting late.
Click to expand...

data sheet says max SPI is 10mhz

EVK CCR settings for 8mhz 0x696968xx
(xx EVK SPI peripheral clock frequency differs from T4) 08 for T4
 
Last edited:
tonton81 said:
For watchdog, should we make a library or make it part of the core? I updated the definitions in imxrt.h and submitted a pull request, and have working code, and enabled the vector table for the callback

Note:

I keep it set at 2 seconds, and able to print to Serial4 output before the reset occurs. This very good to safely handle tasks before reset. :)

Basically with a watchdog of 10 seconds and interrupt of 2 seconds before, the interrupt fires at 8 seconds and final reset happens at 10 seconds. Theoretically your timeout is more like 8 seconds max to feed the dog. :)
Click to expand...

Just tested another thing, you can actually feed the watchdog from within the ISR to prevent the maximum threshold from being reset as well
 
tonton81 said:
For watchdog, should we make a library or make it part of the core? I updated the definitions in imxrt.h and submitted a pull request, and have working code, and enabled the vector table for the callback
...
Click to expand...

That sounds like it could be useful for some set of users - it was a visited topic of interest on T_3.x at least once. If the core is going to include a uniform way to handle high and over temp then having uniform a way to setup and feed a watchdog seems like it could be similarly useful.

Anther note I wanted to make was to consider dropping the CPU speed down on an unhandled fault - when it goes into a while(1) it may as well drop the speed and voltage. If/when/as I revisit my debug_t4 I might see how that plays out.
 
Attention, Windows only:

In case someone wants to try GNU Arm Embedded Toolchain 8-2018-q4-major, I uploaded a working version here: https://drive.google.com/file/d/1ORKyt_FSqn2kgP0ngY3UguLcdzgfQJ6u/view
The problem with the original version is, that objcopy throws an error "64-bit address xyz out of range for Intel Hex file", due to a bug. I took the objcopy from gcc-7 major and now it works.
The download includes the additional teensy libraries, too.

1) Extract the zip to Arduino\hardware\tools
2) To use it, edit the board.txt file, and set teensy4b.build.toolchain=arm8/bin/

I tried a few sketches, it works suprisingly good - some throw additional warnings (which is good).

@mjs513, thank you!
Code: 
File size 5MB
Buffer size 100 bytes
Starting write test.  Please wait up to a minute
Write 257.67 KB/sec
Maximum latency: 149300 usec, Minimum Latency: 0 usec, Avg Latency: 387 usec

Starting read test.  Please wait up to a minute
Read 786.29 KB/sec
Maximum latency: 4090 usec, Minimum Latency: 0 usec, Avg Latency: 127 usec
- half speed of T3.6 ... oops.. ok, there is some potential to optimize it. it just uses the SPI-library functions.
 
BME280 and SPI

@manitou

Yes, Paul added code for setting SPI CLK, though I didn't know he updated the deprecated setFrequency. The current T4 SPI lib does not have code to set polarity and bit order, so i'm not surprised the deprecated functions haven't been updated. Registers default to Mode0 and MSB which is what most devices use (including BME20). With T4 so fast, it still might be a problem with setting CS, so a delayMicros(10) after CS HIGH and before CS LOW might work, and maybe delays between spixfer() ....
Click to expand...
Looking at SPI.cpp looks like it is on the todo list, setClock,
// TODO: make these implement settings - for now, just fixed config
Click to expand...
How for the good news, I added the delayMicros to the library and now its working. So that seemed to do the trick :) Thanks for the suggestion. Maybe that needs to get into the core to fix the issues with timings.

EDIT: Haven't tested this yet but going through the SPI.h file I extracted Paul's method for setting the divider and CCR and put it into standalone function as well. Just wanted to get a sanity check before I screwed it up:
Code: 
void setClockFreq(uint32_t clock)
{
	uint32_t ccr; // clock config, pg 2660 (RT1050 ref, rev 2)
	uint32_t tcr; // transmit command, pg 2664 (RT1050 ref, rev 2)

	uint32_t d, div, clkhz = 528000000/7;  // LPSPI peripheral clock
	if (clock == 0) clock =1;
		d= clkhz/clock;
	if (d && clkhz/d > clock) d++;
	if (d > 257) d= 257;  // max div
	if (d > 2) {
		div = d-2;
	} else {
		div =0;
	}
	LPSPI4_CCR = LPSPI_CCR_SCKDIV(div);
	LPSPI4_TCR = LPSPI_TCR_FRAMESZ(7);    // TCR has polarity and bit order too
}

EDIT2:
Code: 
data sheet says max SPI is 10mhz

EVK CCR settings for 8mhz 0x696968xx
(xx EVK SPI peripheral clock frequency differs from T4) 08 for T4
If I put 10Mhz into the above function it returns 06 vs the 08? d returns 8. Not sure why the d-2?
 
Last edited:
And for what it is worth... I have tried SPI at 30mhz... It appears like it is running at about 25...

Here is test run with 30mhz and LSBFIRST...

screenshot.jpg
 
mjs513 said:
If I put 10Mhz into the above function it returns 06 vs the 08? d returns 8. Not sure why the d-2?
Click to expand...


Yes, I provided setClockFreq snippet (post #454). If you look at the ref manual, the CCR DIV is treated as +2, so if DIV is 0, peripheral clock is divided by 2. If DIV is 1, clock is divided by 3 ... So if you want the clock divided by 8, you put 6 in the CCR DIV field.
 
KurtE said:
And for what it is worth... I have tried SPI at 30mhz... It appears like it is running at about 25...
Click to expand...
The discrete SPI dividers for 75 mhz are 2,3,4,5 .... which gives SPI CLK of 37.5, 25, 18.75 ... the setClockFreq algorithm picks the divider less than or equal to user-requested frequency
 
@KurtE
Finally learning details about SPI that I never got into before. I had downloaded your changes in the Pull request awhile ago and changing the clock in SPIsettings is working perfectly. Put together a test sketch on using transactions (mirrors the one on teensy SPI page) and tested changing the clock against the various clock speeds (60mhz down to 500000hz).

at 40Mhz the clk pulses look more like sine waves with freq of about 38Mhz, MOSI looks pretty good, just getting spikes on the leading and trailing edges
Code: 
at 30Mhz reading 25.5Mhz, clk beginning to look like nice squares waves, MOSI still good (yep same as you got)
at 10 MHz reading 9.43Mhz, guess would call it ringing in the square wave pulse for the clock, MOSI still good but still with the spikes
at 20 Mhz reading 18.86Mhz, clk getting better, looking like a nice square wave, MOSI still good with the spikes
at 4 Mhz reading 3.97 Mhz, clock nice square wave,  MOSI still good but still with the spikes

I tried using a modified version of the DigitalPotControl example but that didn't seem to work, didn't see anything on the scope. Still trying to figure that out. By the time T5 comes out maybe I know what I am doing :) Thanks for the patience.

Mike

EDIT: was wrong. Still doesn't work

EDIT2: Just a quick comparison of waveforms between the T3.2 and T4. On the T3.2 at 30Mhz looks more like the waveform on the T4 at 20Mhz
 
Last edited:
Status
Not open for further replies.
Back
Top