FlexBASIC feature request

Mickster

I do this a lot:

'set timeout to 5ms
timeout = timer + 5

do
    if timer > timeout then exit do
loop

Or is this the way to do it?:

timeout = getcnt() + (clkfreq/200)

do
  if getcnt() > timeout then exit do
loop

Would there be problems with getcnt() rollover?

localroger

Yes, cnt rollover will getcha in the second example. Not sure this is correct FB format but here's the logic you need to use, doing the subtraction each loop:

timeout = clkfreq / 200
timestart = getcnt()
do
  if getcnt() - timestart > timeout then exit do
loop

This way even if the subtraction spans a rollover, the result will still be positive and compare properly against the timeout value.

Mickster

Took a while for me to get my head around this....many thanks.

I still need a general purpose ms timer as I like to monitor the response times of many actuators for diagnostic purposes.

One of my processes only needs a 1ms scan so I think I will increment a variable in that cog, after a waitcnt and use that as a reference

localroger

Even if you use a millisecond timer you have the rollover problem every 50 days or so.

Mickster

Not a problem. A typical production cycle is less than one minute so I simply reset the counter to zero.

If I name it Timer:

Timer=0

Execute some process code 

Timetaken=Timer

Micromite BASIC has a 64bit ms counter which I would never have to worry about but I still zero it because I can

JonnyMac

The Arduino has a convenient function called millis() that returns the milliseconds since reset. We don't yet have that in Spin, so I coded it like this:

pub millis() : result | now0, now1, cf

  org
                getct     now1                          wc      ' snapshot 64-bit cnt register
                getct     now0

                stalli                                          ' hold interrupts during division
                rdlong    cf, #$44                              ' read ticks per second
                qdiv      cf, ##1000                            ' calc ticks per ms
                getqx     cf
                allowi                                          ' enable interrupts
                
                nop                                             ' let interrupts be processed

                stalli
                setq      now1                                  ' divide system counter by ticks/ms
                qdiv      now0, cf    
                getqx     result
                allowi
  end

Since FlexBASIC allows inline assembly, you should be able to adjust this -- I don't know FlexBASIC well enough to do that. I have tested this method in Spin and it works (it's based on Chip's code getsec() code in the interpreter).

JonnyMac

As a mental exercise, I gave it a whack. After looking at the compiler listing to see which scratch variables I could use, I came up with this.

function millis() as uinteger

  asm
                getct     arg02                         wc      ' snapshot 64-bit cnt register
                getct     arg01

                stalli                                          ' hold interrupts during division
                rdlong    arg03, #$44                           ' read ticks per second
                qdiv      arg03, ##1000                         ' calc ticks per ms
                getqx     arg03
                allowi                                          ' enable interrupts
                
                nop                                             ' let interrupts be processed

                stalli
                setq      arg02                                 ' divide system counter by ticks/ms
                qdiv      arg01, arg03    
                getqx     result1
                allowi  
  end asm

end function

It compiles, but always returns 0. Looking at the new listing I see why: the compiler replaces the getqx result1 line with mov result1, #0.

005dc                 | ' function millis() as uinteger
005dc                 | _millis
005dc                 | ' 
005dc                 | '   asm
005dc     1A FA 70 FD | 	getct	arg02 wc
005e0     1A F8 60 FD | 	getct	arg01
005e4     24 42 60 FD | 	stalli
005e8     44 FC 04 FB | 	rdlong	arg03, #68
005ec     01 00 00 FF 
005f0     E8 FD 14 FD | 	qdiv	arg03, ##1000
005f4     18 FC 60 FD | 	getqx	arg03
005f8     24 40 60 FD | 	allowi
005fc     00 00 00 00 | 	nop
00600     24 42 60 FD | 	stalli
00604     28 FA 60 FD | 	setq	arg02
00608     7E F8 10 FD | 	qdiv	arg01, arg03
0060c     24 40 60 FD | 	allowi
00610     00 EA 04 F6 | 	mov	result1, #0
00614                 | _millis_ret
00614     2D 00 64 FD | 	ret

I don't know why. It seems odd that the compiler would arbitrarily change inline assembly code. Maybe Eric can assist -- or even add the millis() function to FlexBASIC.

Mickster

@JonnyMac

Hey Jon.

Isn't that solution still prone to falling into the rollover trap?

BTW, congrats on the excellent N&V article. If it was mentioned here then I missed it. I found the link on another forum.
The bonus was the comments section 😂😂😂. Some guy made a snide remark and I'm pretty sure the reply came from @ManAtWork, right? It was brilliant

As they say; the article needs to go viral

JonnyMac

Until there is a language with 64-bit variables, yes. Treating the millis() value as an unsigned long, the roll-over is 49.7 days. I have a variation that let's me reset a sync point if I want, but that same capability is easily done in high-level code. As Roger pointed out above, if you do the subtraction in the loop the roll-over is not a problem.

Using the 64-bit version of cnt gives you more range -- but you have to do this from assembly. You can use the lower half of cnt (getct() in the P2) if your events are very short. I run at 200MHz, so using the lower half of cnt means I have to access that register faster than every 21.4s, otherwise I will get a roll-over error.

I'm glad you enjoyed the article.

JonnyMac

Got it! I think... we'll have to check with @ersmith on the rules of ASM..END ASM. That said, this works. Demo program is attached.

function millis() as uinteger

  dim as uinteger now1, now0
  dim as uinteger cf
  dim as uinteger ms

  asm
                getct     now1                          wc      ' snapshot 64-bit cnt register
                getct     now0

                rdlong    cf, #20                               ' read ticks per second
                qdiv      cf, ##1000                            ' calc ticks per ms
                getqx     cf
                
                setq      now1                                  ' divide system counter by ticks/ms
                qdiv      now0, cf    
                getqx     ms
  end asm
  
  return ms

end function

Once I defined local variables and looked in the right place for the system frequency (I checked the listing file for getsec()), it came together.

ersmith

@JonnyMac : thanks for posting that code. I think you've groked asm/end asm just fine .

Ironically, there is a builtin _getms() function, but I forgot to document it and/or provide an alias for BASIC. @Mickster : Jon's function above should work, or you could call "_getms()" (underscore at the beginning) which does the same thing: provides a 32 bit millisecond timer.

The next version of FlexBASIC will have a "getms()" function (no underscore). The underscore version will continue to be supported.

JonnyMac

@Mickster If you throw Spin2 into your mix of programming languages, Chip just showed a way to add custom interrupt handlers. I took his blinking LED example and converted it to a free running timer that is installed in the Spin interpreter cog. I have two variations: one that does milliseconds, the other that is a 24-hour RTC with hours, minutes, seconds, and 1/100ths seconds. Both allow the timer to be put on hold and updated.

See this thread: http://forums.parallax.com/discussion/172453/free-running-timer-in-the-spin-interpreter-cog#latest

With FlexBASIC, you could probably do something similar by creating a timer cog with access methods that suited your needs. Given that the Flex system compiles to native PASM, you don't have the Spin cog like I do to install an ISR within.

Mickster

@ersmith
@JonnyMac

Thanks guys, it's all looking great.

Away working right now. Didn't bring my Prop stuff (easily sidetracked)