Re: [Amforth] AmForth & AI (Expert Systems)

[John S. <jsa...@gm...>]

Hi Everyone,
Forgot to mention that an assembler abstraction layer is still being used
to make easier porting to
other chips (this part still works reliably). The difference being the use
of assembler macros instead of C macros for this layer.
>From what I am seeing is the riscv toolchain isn't reliable enough through
the chain. The riscv assembler on its own has better
reliability (at least for now).



On Mon, Mar 30, 2026 at 11:06 AM John Sarabacha <jsa...@gm...>
wrote:

> Hi Everyone,
> For those still following this thread, to improve the reliability of this
> work some changes were made to glanForth(derived from AmForth).
> A hybrid of C generated and  assembler code using asm(...) and C data
> structures is no longer being used, pure assembler generated code is being
> used instead, due to problems relating to the riscv toolchain used by
> MounRiver Studio (even the most up to date toolchain had these issues -
> still experimental???). The best results were obtained by using the pure
> riscv assembly code for amForth. Also to get consistent amforth dictionary
> alignment on 32 bit boundaries the NEXT macro wasn't being used with the
> primary words, instead a macro was used for each primary word along with a
> branch instruction to the next code in the ITC interpreter.
> These results were verified not by simulation but by actual hardware (chip
> level) observations.
> The latest build for a functioning riscv amforth (almost the complete code
> base) was 10,043 bytes for the CH32X033 and can function interactively or
> can co-exist with other C or assembler routines.
> An experimental version of the Expert System CLIPS V7 will be used to
> generate and verify forth code sequences with this build.
> Very little of the available SRAM has been used in this build, which can
> be used to store significant amounts of user forth programs.
>
> Best Regards,
> John S
>
> On Wed, Mar 4, 2026 at 9:56 AM John Sarabacha <jsa...@gm...>
> wrote:
>
>> Hi Everyone,
>> For those interested in this topic and are using AmForth, an experimental
>> version of the Expert System CLIPS V7 has been successfully built and will
>> be used with glanForth(derived from AmForth) to generate forth code
>> sequences (based on defined knowledge rules and 2012 standard forth syntax)
>> across a large network of glannForth (non-derived runtime embedded forth)
>> nodes.
>> Currently glanForth(derived from AmForth) is a hybrid of assembler code
>> and C data structures.
>>
>> Regards to all,
>> John S
>>
>>