43s status

[Claudio L.]

(06-26-2014 08:47 AM)brouhaha Wrote:  It's an Energy Micro EFM32GG380F1024, with 1MiB of flash and 128KiB of RAM. The EFM32GG can support additional external memory, but we don't have enough PCB space for that, and it would significantly increase the power consumption. To the best of my knowledge, there is no SoC with 2MiB of flash and 512KiB of RAM that doesn't draw at least an order of magnitude more power than the EFM32GG, so there is basically no chance that we will switch to something with more memory.

Fair enough. That was an ideal case, to run without changes. So far newRPL works very well with about 8 kbytes of ram for TempOb (this is what I use for testing under frequent GC). However, it starts failing with you set the precision high (at 2007 digits, each real number takes about 900 bytes).
I'll make sure to keep things flexible enough so we can run with 128 kbytes. The amount of flash would require the most changes: we'd have to reduce the maximum system precision by a large amount. Right now the tables for transcendental functions alone are about 1 MB (for 2000 digits precision), but using 500 digits would cut that to 250 kbytes, very manageable with 1 MB of flash.

(06-26-2014 08:47 AM)brouhaha Wrote:  Also we're not about to throw away our hard-won design expertise with the EFM32 and start from scratch (again!) with some other SoC. We've already restarted the project several times, and if we keep doing that we'll never ship anything.

I wouldn't ask you for that, I simply didn't know that you had a SoC selected. This is the first time you mention the chipset, so we are all learning for the first time the specs of your hardware. It's not so far from the Cortex-M4 that me and other people were thinking of as the "ideal" candidate, so in the end we were all thinking alike.

(06-26-2014 08:47 AM)brouhaha Wrote:  The HP-28C clearly had nowhere near enough RAM, but the HP-28S was a pretty usable RPL machine with 128KiB of ROM and 32KiB of RAM. Our hardware has eight times as much flash (vs 28S ROM), and four times as much RAM. Even with the increased memory consumption of 32-bit pointers on native ARM vs. 20-bit on Saturn (or Saturn emulated on ARM), its seems that it should be possible to run RPL on it. However, as I said, RPL isn't our target.

It's definitely possible, we'll just have to keep it more humble. Perhaps the only drawback is the lack of an MMU won't let us optimize the use of memory, we'll have to use fixed partitions or work around it moving big blocks of memory.

(06-26-2014 08:47 AM)brouhaha Wrote:  We do expect customers to be interested in at least two different firmware choices for our hardware, Free42 and WP-43s. We also have the HP-41CX microcode running on it, but that doesn't map well to the extended Pioneer keyboard layout.

We'll keep our minds open (and our code portable), who knows, maybe when you release you might have a third firmware choice, but we are not there yet.

Claudio

[Paul Dale]

(06-26-2014 01:09 PM)Claudio L. Wrote:  However, it starts failing with you set the precision high (at 2007 digits, each real number takes about 900 bytes).

It should be possible to run at a lot less digits than this. Especially if you are willing to accept 1 ULP accuracy.


- Pauli

[Paul Dale]

For an unlimited duration on device with limited memory, you must implement a compacting garbage collector.


- Pauli

[Claudio L.]

(06-26-2014 02:06 PM)Paul Dale Wrote:  For an unlimited duration on device with limited memory, you must implement a compacting garbage collector.


- Pauli

Yes, it's been finished for a couple of months now and running solid. I keep the RAM limited artificially to force the GC to work often, so if we introduce any bugs related to GC as we develop new commands, they will become apparent quickly.

Claudio

[pito]

(06-26-2014 01:09 PM)Claudio L. Wrote:  I wouldn't ask you for that, I simply didn't know that you had a SoC selected. This is the first time you mention the chipset, so we are all learning for the first time the specs of your hardware. It's not so far from the Cortex-M4 that me and other people were thinking of as the "ideal" candidate, so in the end we were all thinking alike.
..
Claudio

EFM32GG380F1024 is a Cortex M3 (basically "the same" as the CM4). CM4 usually includes an FPU, which is of no use for your calculator (single precision only) and some CM4 (ie STM32F4xx) include a flash accelerator to eliminate the flash/rom waitstates.

[Claudio L.]

(06-26-2014 02:04 PM)Paul Dale Wrote:  

(06-26-2014 01:09 PM)Claudio L. Wrote:  However, it starts failing with you set the precision high (at 2007 digits, each real number takes about 900 bytes).

It should be possible to run at a lot less digits than this. Especially if you are willing to accept 1 ULP accuracy.


- Pauli

The default is set to 36 digits right now, but the user can select up to 2007 ("2007 SETPREC" on the demo). The main reason to have high number of digits is because large numbers will become a replacement for the infinite precision integers used by the CAS.
Numbers in newRPL are just that: numbers. The user need not distinguish integers from reals or infinite precision integers like it did on userRPL, and having lots of digits is key to achieve this.
Still, there will be limitations. Computing 1000! exactly requires 2568 digits. We won't be able to provide an exact answer while the 50g does.

Claudio

[Massimo Gnerucci]

(06-26-2014 01:09 PM)Claudio L. Wrote:  I wouldn't ask you for that, I simply didn't know that you had a SoC selected. This is the first time you mention the chipset, so we are all learning for the first time the specs of your hardware.

Well, not really true: sa-penguin already unveiled it.

[walter b]

You can't beat a good memory.

d:-)

[pguertin]

(06-19-2014 03:02 PM)Sylvain Cote Wrote:  Another alternative is to use the "U.S. International" keyboard layout.
I have been using this layout for years on Windows, OSX & Linux.

Yet another alternative if you don't like dead keys and want to access
seldom-used special characters (this wouldn't be fun for typing whole
paragraphs of French) is to assign one key to be a Compose key
(a.k.a Multi-key). Just type Compose, then a few characters (usually
two) that make a good mnemonic. So for example:

• Compose / o gives ø
• Compose - D gives Đ
• Compose t m gives ™
• Compose _ a gives ā
• Compose ( 4 2 ) gives ㊷
  

Choosing a Compose key is easy to do in several distributions of Linux
(Preferences/Keyboard/Options) and a bit harder in others (but you enjoy
tweaking xkb configuration files by hand, don't you?). Don't know about
MacOS and Windows, but I'm pretty sure there are ways to do this as
well. Linux comes with hundreds of ready-made Compose key combos
but you can also define your own.

[jracca]

(06-16-2014 07:22 PM)walter b Wrote:  

(06-16-2014 06:42 PM)jracca Wrote:  The 42s was fantastic for what it was... The biggest downside is no IO capability.

The only other thing that would have been nice and easy considering the technology of the time is fractions like the 32Sii

The HP-42S is still HP's most powerful RPN calculator you can get (and presumably HP's calculator with the most power per cm^3). There are several threads in the forum antiquum speculating about the reasons why it wasn't allowed to feature more I/O capabilities - please look there if you are interested. And the fraction business began years later, maybe when turning from students to students

How can wikipedia be wrong? The display makes it so much easier to use than the 15C, I never was efficient without the manual on the 15 to do matrix math, but the 42s is great unless you need symbolics in your matrix... then the 50g works

[ElectroDuende]

(06-15-2014 08:14 AM)walter b Wrote:  Thus, calculators either work with one printer (protocol) or none. Thus, we need to point out THE printer - it may well become the same as for the HP-42S, -48x, etc. again.

d:-)

RS-232 9600 bps ticket printer?

You have tons of compatible models...
as an example: http://www.custom.biz/mobile-printers/my...er-bt-19-5

[jebem]

Is this the year of the large scale reptiles coming out to the sunshine?

[x34]

"Walter B words: This calculator will be the first non-HP RPN calculator designed from scratch after quite some decades."

Hmm. What about DM-series by SwissMicros and Elektronika-MK152/161 by Semico?

x34

[Thomas Radtke]

(09-11-2014 05:44 AM)x34 Wrote:  Hmm. What about DM-series by SwissMicros and Elektronika-MK152/161 by Semico?

The DMs are based on HP-firmware, and the MKs evolved from the MK-52 design. They certainly weren't designed 'from scratch'.

[x34]

(09-11-2014 06:15 AM)Thomas Radtke Wrote:  

(07-17-2014 06:40 PM)jracca Wrote:  Hmm. What about DM-series by SwissMicros and Elektronika-MK152/161 by Semico?

The DMs are based on HP-firmware, and the MKs evolved from the MK-52 design. They certainly weren't designed 'from scratch'.

DM-15C uses sufficiently reworked firmware, and internal structure (CPU, display, etc) is far from HP-15C.

Sure, MK's evolved from MK-52, but use completely different and wastly extended command set. They were certainly designed from scratch.

Besides, I wonder if WP-43 is not evolved from HP-42?

[Thomas Radtke]

Judging from recent keyboard polls and questions about e.g. equation editors, the 43s is or was completely open.

From my own hard- and software projects, I know designing is more than just programming and soldering, so yes, the 43s started from scratch while the projects you mentioned were not.

[Paul Dale]

(09-11-2014 09:09 AM)Thomas Radtke Wrote:  From my own hard- and software projects, I know designing is more than just programming and soldering, so yes, the 43s started from scratch while the projects you mentioned were not.

The software is the (relatively) easy bit and it still takes years to get right and more years to debug properly (I figure we're about halfway through this process for the 34S). The UI definition is another massive task -- thanks Walter, even though we had lots of discussions (aka "arguments") along the way. The hardware design is also problematic and the physical design far more so -- thanks HP for making such an open platform as the 30b despite its well reported shortcomings.

I'm not denying that the software makes the product but there are plenty of other factors involved.


- Pauli

[walter b]

Well said.

(09-11-2014 11:25 AM)Paul Dale Wrote:  ... even though we had lots of discussions (aka "arguments") along the way.

The nice side of most technical men is you can discuss with them fiercely and eventually reach a better solution for all of us (sorry, ladies, no flame intended - just my experience). That will continue with the 43S for sure. If we'd all have the same opinion, only a tiny fraction of mankind would be required.

d:-)

[jracca]

(09-11-2014 06:15 AM)Thomas Radtke Wrote:  

(07-17-2014 06:40 PM)jracca Wrote:  Hmm. What about DM-series by SwissMicros and Elektronika-MK152/161 by Semico?

The DMs are based on HP-firmware, and the MKs evolved from the MK-52 design. They certainly weren't designed 'from scratch'.

That is not my quote, Never head of an Elektronika MK-153

[Thomas Radtke]

(09-17-2014 06:56 PM)jracca Wrote:  

(09-11-2014 06:15 AM)Thomas Radtke Wrote:  The DMs are based on HP-firmware, and the MKs evolved from the MK-52 design. They certainly weren't designed 'from scratch'.

That is not my quote, Never head of an Elektronika MK-153

Yes, sorry. I've corrected the quote.