Ivan Godard <ivan@millcomputing.com> writes:

> On 4/29/2023 9:55 AM, Anton Ertl wrote:
>> EricP <ThatWouldBeTelling@thevillage.com> writes:
>>> I do wonder just what is being licensed with an "architectural license"
>>> as any new design (microarchitecture) would be the designer's own I.P.
>>> and I don't think an ISA can be protected by copyright or patent.
>>
>> It seems that the industry agrees that an ISA can be monopolized
>> ("protected") in some way. What I imagine is that individual
>> instructions are patented; ARM has some innovative ones in the A64
>> instruction set, and patent law as practiced seems to set a low bar
>> for making things patentable.
>>
>> Maybe the encoding is monopolized in some way. Could it be copyright?
>> Maybe the argument is that the decoder in the CPU core is a
>> derived work of the copyrighted encoding.
>
> A suitably novel encoding method is patentable. Mill has a patent on
> its bi-directional encoding.

You undersell the value of the split instruction stream. IIRC in your
presentation you said the extra stream doubles the amount of
instructions you can read at once, which to me seemed like a dubious
benefit for the complexity it brings. But the real value is you get to
use fixed length on one side and variable length on the other. So you
get to be many times faster on the fixed side while still getting the
advantages of variable length instructions.

I didn't realize that until I independently invented something similar.
I was thinking about the possibility of storing all immediate values
backwards from the last jump target and having an addressing mode that
references last jump target - x. The idea being to get all the variable
length stuff out of the main instruction stream. After a while I
realized this was basically a split instruction stream where the
backwards stream was all immediate load instructions. When I realized
that I immediately looked up Mill's split stream instructions and sure
enough, it was fixed length on one side. I had missed the entire point
of split stream until that point!

Ivan Godard <ivan@millcomputing.com> schrieb:
> On 4/29/2023 9:55 AM, Anton Ertl wrote:
>> EricP <ThatWouldBeTelling@thevillage.com> writes:
>>> I do wonder just what is being licensed with an "architectural license"
>>> as any new design (microarchitecture) would be the designer's own I.P.
>>> and I don't think an ISA can be protected by copyright or patent.
>>
>> It seems that the industry agrees that an ISA can be monopolized
>> ("protected") in some way. What I imagine is that individual
>> instructions are patented; ARM has some innovative ones in the A64
>> instruction set, and patent law as practiced seems to set a low bar
>> for making things patentable.
>>
>> Maybe the encoding is monopolized in some way. Could it be copyright?
>> Maybe the argument is that the decoder in the CPU core is a
>> derived work of the copyrighted encoding.
>
> A suitably novel encoding method is patentable. Mill has a patent on
> its bi-directional encoding.
>

Do you have a patent number for that handy? I'd like to take a look
(and see if I can even read it - patentese is sometimes opaque).

John Levine <johnl@taugh.com> schrieb:
> According to Thomas Koenig <tkoenig@netcologne.de>:
>>Terje Mathisen <terje.mathisen@tmsw.no> schrieb:
>>
>>> Afaik, there _are_ no similar bodies anywhere in the
>>> (western/civilized?) world. The USPTO seems to use the classic GitHub PR
>>> approach: LGTM! I.e. just grant the patent, leaving it up to the lawyers
>>> and the court system to figure out which of them should not be valid.
>>
>>I think the US relented in 2012 and allowed opposition procedures,
>>same as everybody else (only they call it revision, but nevermind).
>
> US patents have always been examined to ensure that they cover
> patentable subject matter, are novel, and not obvious. The first
> patent examiner was Thomas Jefferson, as part of his duties as
> Secretary of State.

Sure, but the lack of opposition led to some strange situations
when the examiner missed something, for example prior art.

One of the strategies for the US was then to prepare a legal
opinion that the patent is invalid, and wait to be sued.

And prior art can be tricky. I once took part as an expert witness
in an opposition procedure before the EPA where somebody had claimed
a certain relationship between process and a certain property
of the prodct. The relationship was entirely trivial (to me),
but the patent office had decided that this did not amount to a
lack of inventive step.

So, we took an old patent, where both property and process
were described. Problem was that the method for measuring the
property that the patent we opposed had used didn't exist at the
time that the original one was written, so we had to establish
that relationship.

In the oral proceedings, our patent attorney had to do a lot to
keep me quiet. She even passed me a note at one time that what
the other party's patent attorney was saying was actually beneficial
to us.

> For software patents there has been a long standing problem that the
> quality of the examination was poor. A large part of that is that the
> novelty check was mostly done by looking at previous patents, but the
> vast majority of software techniques are not patented. The 2012 change
> allowed third-party challenges in addition to examination.
>
>>What I found really funny were "dormant" patents, which were granted,
>>but not published (which makes nonsense of the idea of a patent).
>
> Other than a handful of patents that are classified, there's no such
> thing.

That one was indeed classified.

> What does exist is so-called submarine patents where the
> applicant keeps amending the application to stretch out the time
> until it's granted. That's been somewhat fixed by publishing most
> applications after 18 months, whether or not they're granted, and
> changing the term to 20 years from filing rather than 17 years from
> grant.

Again, the US following the rest of the world after a suitably large
number of decades :-)

On 2023-05-03 8:30, Thomas Koenig wrote:
> Ivan Godard <ivan@millcomputing.com> schrieb:
>> On 4/29/2023 9:55 AM, Anton Ertl wrote:
>>> EricP <ThatWouldBeTelling@thevillage.com> writes:
>>>> I do wonder just what is being licensed with an "architectural license"
>>>> as any new design (microarchitecture) would be the designer's own I.P.
>>>> and I don't think an ISA can be protected by copyright or patent.
>>>
>>> It seems that the industry agrees that an ISA can be monopolized
>>> ("protected") in some way. What I imagine is that individual
>>> instructions are patented; ARM has some innovative ones in the A64
>>> instruction set, and patent law as practiced seems to set a low bar
>>> for making things patentable.
>>>
>>> Maybe the encoding is monopolized in some way. Could it be copyright?
>>> Maybe the argument is that the decoder in the CPU core is a
>>> derived work of the copyrighted encoding.
>>
>> A suitably novel encoding method is patentable. Mill has a patent on
>> its bi-directional encoding.
>>
>
> Do you have a patent number for that handy? I'd like to take a look
> (and see if I can even read it - patentese is sometimes opaque).

U.S. Patent 9,513,920 - Computer Processor Employing Split-Stream Encoding

The text seems to be on the Mill web:

https://millcomputing.com/blog/wp-content/uploads/2016/10/uspto9513920.pdf