TheBackShed.com

Forum Index : Microcontroller and PC projects : Y'all ready for this?

Author

Message

PhenixRising
Guru

Joined: 07/11/2023
Location: United Kingdom
Posts: 1937

Posted: 10:06am 22 May 2026

Copy link to clipboard

Print this post

Betas for Pico, Pico 2 and Micro:bit

I have the RP2350 uf2 running on Pete's DIL module.

NOTE: This is the very first release for Pico so don't expect accurate documentation. One minute you think you're reading about the Pico but you ain't

mozzie
Guru

Joined: 15/06/2020
Location: Australia
Posts: 370

Posted: 08:04am 23 May 2026

Copy link to clipboard

Print this post

G'day Phenix,
Thanks for the link, only had a quick look but for simple stuff and speed this looks very good.

Also has the one thing the PicoMite lacks, a proper low power sleep mode. Assuming it works of course.

More research required.

Regards, Lyle.

Mixtel90

Guru

Joined: 05/10/2019
Location: United Kingdom
Posts: 8877

Posted: 09:49am 23 May 2026

Copy link to clipboard

Print this post

TBH I'm not sure what I'm looking at... :)

The various low power modes on the RPnnnn drive me bog-eyed when I look at the data sheet!

I have a MicroBit somewhere. It's programmed as a thermometer, but I'm not sure where it is.

Mick

Zilog Inside! nascom.info for Nascom & Gemini
Preliminary MMBasic docs & my PCB designs

PhenixRising
Guru

Joined: 07/11/2023
Location: United Kingdom
Posts: 1937

Posted: 10:04am 23 May 2026

Copy link to clipboard

Print this post

Hi Lyle,

Yes the SLEEP is what caught my eye because others have mentioned it.

It's so nice to settle on the RP2350 DIL module, running MMBasic and having another, running ARMbasic as a super high-speed peripheral.

I'm currently testing and not surprisingly, it's FAST

PhenixRising
Guru

Joined: 07/11/2023
Location: United Kingdom
Posts: 1937

Posted: 11:55am 23 May 2026

Copy link to clipboard

Print this post

Quote
Note:
1) Not overclocked
2) Times are for all 1000 iterations

BM1:
2040: 0.383 msec
2350: 0.277 msec

BM2:
2040: 0.332 msec
2350: 0.230 msec

BM3:
2040: 0.859 msec
2350: 0.523 msec

BM4:
2040: 1.283 msec
2350: 0.402 msec

BM5:
2040: 1.362 msec
2350: 0.486 msec

BM6:
2040: 3.111 msec
2350: 1.584 msec

BM7:
2040: 3.721 msec
2350: 1.957 msec

BM8:
2040: 564.388 msec
2350: 95.297 msec

#define include_trig
#include <AB_Math.bas>

sub printTime (usec)
dim flt_usec as single

if usec <1000 then
Print "0.";(usec);" msec"
Else
flt_usec = usec / 1000
print flt_usec;" msec"
endif
print
endsub

Main:
Print "ARMbasic Benchmark tests"
Print " "

Print "Benchmark 1"
A=Timer
For j = 1 To 1000
Next j
printTime(Timer-A)

Print "Benchmark 2"
A=Timer
j =j0
BM2:
j = j+1
If j < 1000 then GoTo BM2
printTime(Timer-A)

Print "Benchmark 3"
A=Timer
j = 0
BM3:
j = j+1
aa = j/j*j+j-j
If j < 1000 then GoTo BM3
printTime(Timer-A)

Print "Benchmark 4"
A=Timer
j = 0
BM4:
j = j+1
aa = j/2*3+4-5
If j < 1000 then GoTo BM4
printTime(Timer-A)

Print "Benchmark 5"
A=Timer
j = 0
BM5:
j = j+1
m = j/2*3+4-5
Gosub fourthou
If j < 1000 then GoTo BM5
printTime(Timer-A)

Print "Benchmark 6"
A=Timer
j = 0
dim ray(5)
BM6:
j = j+1
m = j/2*3+4-5
Gosub fourthou
For q = 1 TO 5
Next q
If j < 1000 then GoTo BM6
printTime(Timer-A)

Print "Benchmark 7"
A=Timer
j = 0
dim ray2(5)
BM7:
j = j+1
m = j/2*3+4-5
Gosub fourthou
For q = 1 TO 5
ray2(q) = m
Next q
If j < 1000 then GoTo BM7
printTime(Timer-A)

Print "Benchmark 8"
A=Timer
j = 0
BM8:
j = j+1
m = pow(j,2)
blog = LOG(j)
csin = SIN(j)
If j < 1000 then GoTo BM8
printTime(Timer-A)

End

fourthou:
RETURN

The label "main:" wasn't mentioned in the documentation and if you don't have it, the compiler gets all upset.

zeitfest
Guru

Joined: 31/07/2019
Location: Australia
Posts: 680

Posted: 02:54am 24 May 2026

Copy link to clipboard

Print this post

AFAIK the pico does not use a dedicated arithmetic unit like other processors,
but instead uses optimised software routines. So the interpreter and compiled version
probably execute the same underlying stuff, as opposed to a compiler's code accessing a arithmetic/logic unit. The times are slow compared to a more usual processor.
Even crappy laptops are pushing teraflops now.

Mixtel90

Guru

Joined: 05/10/2019
Location: United Kingdom
Posts: 8877

Posted: 07:02am 24 May 2026

Copy link to clipboard

Print this post

It sort of has...
From the Datasheet:

Quote Each Cortex-M33 CPU core is equipped with two instances of a double-precision coprocessor that provides acceleration
of double-precision floating point operations including add, subtract, multiply, divide and square root. The design is
implemented in just a few thousand gates and so occupies much less silicon die area than a full double-precision
floating-point unit. Nevertheless, these coprocessors considerably speed up basic double-precision operations
compared to pure software implementations. The coprocessors also offer support for some single-precision operations
and conversions.

The two coprocessor instances are assigned to the Secure and Non-secure domains. Coprocessor number 4 always
maps to the coprocessor used for the current processor security state. Coprocessor number 5 always maps to the Non-
secure coprocessor instance, but is accessible only from Secure code. This duplication avoids saving and restoring the
coprocessor context during Secure/Non-secure state transitions.

Note that there is no speed gain from using integers in BASIC or from a compiler optimising code as integers or single precision. However, where some hardware coprocessors fall back into software mode for double precision the M3 doesn't. These coprocessors are *very* fast as they are not complex structures.

Quote Unlike a conventional FPU, the accelerator does not contain a full register bank. Not only does this save die area, it also
means that saving and restoring the coprocessor’s state is very fast: in fact, the entire state fits within six 32-bit words
and hence can be saved to, or restored from, the CPU in three instructions.
The accelerator contains a wide adder, capable of adding two mantissa values and three exponent values
simultaneously. There is also a shifter that can either perform a logical right shift by a given amount, or normalise a
denormalised mantissa and report the amount of shift required to do so. A considerable amount of hardware in the
shifter is shared between these two operating modes.

It's pretty impressive really. Section 3.6.2 of the RP2350 Datasheet.

Mick

Zilog Inside! nascom.info for Nascom & Gemini
Preliminary MMBasic docs & my PCB designs

PhenixRising
Guru

Joined: 07/11/2023
Location: United Kingdom
Posts: 1937

Posted: 10:16am 24 May 2026

Copy link to clipboard

Print this post

zeitfest said
Even crappy laptops are pushing teraflops now.

But they don't run @150Mhz

PhenixRising
Guru

Joined: 07/11/2023
Location: United Kingdom
Posts: 1937

Posted: 10:18am 24 May 2026

Copy link to clipboard

Print this post

Mixtel90 said
It's pretty impressive really. Section 3.6.2 of the RP2350 Datasheet.

No kidding. I remember thinking that I could speed something up by going all integer but the difference was negligible.

zeitfest
Guru

Joined: 31/07/2019
Location: Australia
Posts: 680

Posted: 10:29am 24 May 2026

Copy link to clipboard

Print this post

They seem to be describing something that is a few thousand gates used as an accelerator for software doing math, as a math coprocessor.
Usually a math coprocessor is something that does math ops in a few instruction cycles and is much faster.
(ed) I gather the pico 2 does do some math ops in a few cycles but the speeds above are still slow.
Edited 2026-05-24 21:00 by zeitfest

PhenixRising
Guru

Joined: 07/11/2023
Location: United Kingdom
Posts: 1937

Posted: 11:07am 24 May 2026

Copy link to clipboard

Print this post

zeitfest said but the speeds above are still slow.

The benchmarks on our table are per iteration (time taken/1000)

The benchmarks above are for the FULL 1000 iterations (stated at the top).

Mixtel90

Guru

Joined: 05/10/2019
Location: United Kingdom
Posts: 8877

Posted: 03:13pm 24 May 2026

Copy link to clipboard

Print this post

I think they referred to it as a "coprocessor" as that's what it sort of does. They also refer to it as an "accelerator", which is probably a more accurate description.

Oh, the evils that are benchmarks!
a) make sure that the compiler for your chosen hardware is optimised to the hilt for it.
b) use a general purpose compiler for your competition's hardware, preferably a non-optimising one that generates lots of labels.
c) bask in the glory that comes from one-upmanship!

If you hand the accelerator a single command it will probably be faster than any true "math coprocessor" at the same CPU clock speed. Anything more complex and it will get increasingly slower, but by how much may not always be relevant.

Mick

Zilog Inside! nascom.info for Nascom & Gemini
Preliminary MMBasic docs & my PCB designs

zeitfest
Guru

Joined: 31/07/2019
Location: Australia
Posts: 680

Posted: 07:47am 26 May 2026

Copy link to clipboard

Print this post

I had a squiz at the datasheet

Apparently the ARM processor uses a FPU for single-precision. Double precision math uses the accelerator which splits it into chunks and feeds the chunks back to the main processor to be digested using the single precision FPU.
Clever enough I guess but kind of mediocre IMHO.

Mixtel90

Guru

Joined: 05/10/2019
Location: United Kingdom
Posts: 8877

Posted: 09:48am 26 May 2026

Copy link to clipboard

Print this post

A fraction of the complexity, size and cost of a a full double precision FPU though. The ARM Cortex series cores are optimised for low cost and high efficiency, not highest performance. There have to be trade-offs. A big FPU sits there doing nothing for most of the time in many cases.

Mick

Zilog Inside! nascom.info for Nascom & Gemini
Preliminary MMBasic docs & my PCB designs

PhenixRising
Guru

Joined: 07/11/2023
Location: United Kingdom
Posts: 1937

Posted: 09:50am 26 May 2026

Copy link to clipboard

Print this post

ARMbasic:
Had a bit of a play with it but there is no support for the UARTs which means it's a non-starter for me

Print this page

To reply to this topic, you need to log in.