M
Michaelo
Have a good health.
I think that today small companies as Transmeta(tm) have no chances
on the market of general purposes microprocessors.
I suspect they have no chances in obvious future too.
Because the market actually divided between strong players.
I think that Transmeta(tm) was doomed for fault from the beginning.
My treatment of Technical reasons of Transmeta(tm)'s fate I'll give
later.
Now let me tell you the story.
At the beginning of 90xx I though about problems of compatibility of
soviet's amateur's personal / home computers most of which was built on
soviet's copy of i8080 CPU.
As a result I invented a principle, which however I forgot soon.
Later I thought how to run applications for i286 and i386 CPUs on i8088
CPU.
i reminded that principle.
Finally in 1994 or 1995 i read some articles about RISC and artificial
intelligence.
And I invent a conception of technology which I named "codemorfing"
(I read it as "coddemmo'rfin'g")
However I did not implement it and I was not able to do it.
I tried of course. But that year of great depression in Ukraine
I wasn't able even to find documentation and fix my computer.
Political situation in Ukraine was also hard.
I understood that something's going wrong,
and my expressions in address of government was danger for me.
I suspected that somebody is watching me.
I decided to do a horrible trick.
I planed to compose a good looking project of my ideas
and tell about it in several places.
I though such.
1. In best case I'll find a people to whom this project will be
interesting,
and who:
1) Invest in it.
2) or employ me
3) or joint me
4) or at least give me moral support
2. In worse case somebody may implement it without me (as it was
popular)
but I must compose the project such that it must have hidden troubles
which have to be solved with my assistance.
And them the person who implement my project will
1) employ me
2) or pay me for consulting
3) or should explain how did he /she /they got it since all
people who
heard it from me showed no interest for it.
So they collected information about me, why ?
It proves that somebody was watching me. Why ?
And persons who grab my ideas have no law to complain
me in giving improper information.
3. In worst an more probable case:
Project will get fault,
and I'll distribute information about born troubles of the
project.
I thought - How can I do this ?
But finally I decided to survive (remain alive physically) I must
follow the habits of this horrible land.
I.e. if I won't fool people I'll DIE.
That is an abstract of information which I shared:
1. Personal computer have different conceptions of requirements to the
performance than
other kind of computers. Personal computer about 80-98 % of time execute
very small
pieces of code. Maximal performance of machine needed in small pieces of
time.
So performance when it needed may be paid by performance when it do not
needed.
I named it "new conception of performance calculation for personal
computers",
I also compared it with accelerator level of the car.
2. RISC processors are more effective than CISC processors.
Instruction set of RISC processor are optimized for machine generated
code.
So effective conversion code of CISC processor
into the code of RISC processor gives more performance
in some cases even with weaker processors.
3. Codemorfing is a technology of converting machine code
of one microprocessor into code of other microprocessor
directly during execution of the program made by code of first
microprocessor.
Codemorfing allow to effectively convert small pieces of code,
which frequently repeats (in command flow).
(How does it do it read later.
By the way this principle is not my invention indeed,
it was implemented in 80xx, as people explained me)
So 1st + 3rd says that codemorfing is good for personal computer
but not good for system of automatic control.
4. Since performance of processor is required for small pieces of time
(or in general requirements to performance is different in different
moments)
both voltage and frequency variation of processor for optimal
temperature regime
and performance needs
gives preferences in chip lifetime and respect to requests of
performance.
Codemorfing may give an input data for calculation of needed
performance.
For bit of second during the searching entry in data massive
voltage may be rose from 5 V to 12 V and frequency twice.
(Before it I heard about systems with dynamic frequency variation,
and about systems with very slow voltage and frequency variation,
and also about system with frequent (dynamic) changing sleep and and
work modes.
But best results gives exactly dynamic calculation and variation of both
parameters.)
5. Codemorfing works such: special algorithm (morfer)
one by one converts small sequences of instructions
of microprocessor for which program of those instructions was created
into code of microprocessor on which this all executes.
Morfer replace all jumps (GOTOs) on itself and at the end of each
of those pieces of code also put jump on itself. Before jump
morfer set the special identifier (register) which determines
the point from which was done jump.
Morfer converts only that areas of code which are needed for execution.
Morfer conduct statistics of usage of pieces of code,
and after certain number of runs of certain pieces of code
optimize em or convert em again with better quality.
Morfer may replays jumps on itself by jumps on respective
points when further converting or optimization are not needed or
impossible.
Morfer may move pieces of code in memory, with respective swapping
the jumps, restore jumps on itself (but I did not tell why).
During conversion and optimization of pieces of code
morfer may install and remove "agents" (this word I used now) in code
(for making statistics, for instance).
6. Design and manufacturing microprocessors optimized for codemorfing
and supported dynamic voltage and frequency variation together with
respective soft ware may gift big money and glory.
)
Much than glory it may allow you to feel yourself a Robin Hood of
industry.
You break the chains processor dependency of software,
in best case actually any program may be ran on your system.
I told it about 10 years ago not such properly, sometimes it sounded
as speech of madman, and I told different parts of this
information to different people.
(I had no Internet, no even modem, and finally no working computer for
the moment.)
But what I told may be turn into sweet illusion without understanding
some hidden traps.
But about this later.
Finally I forgot about those project, I decided that my plan did not
work.
And people really had me as madman.
Then in my life was depressions, thinks about suicide, inferiority
complex.
But in about 1999 I heard about russian project E2K(tm).
Which reminded me my.
Finally I heard that teams which worked at E2K(tm)
crossed with team which worked in mysterious company named
Transmeta(tm).
(By the way, "codemorfing" I also called "codemetamorfing".)
Finally I read about the product of Transmeta in magazine.
I read MY IDEAS with MY thinking way with MY naming.
Yes I found a lot of changing with my original conceptions.
For instance treatment of voltage regulation of Transmeta(tm)'s
Crusoe(tm)
was different. Probably chip does not allow to change voltage such fast.
And conception of performance calculation was also adapted to
multimedia requirements. But ground ideas was same.
Well since I did not patent all those stuff,
for which I had no money, no experience,
but what is most important that was impossible to patent that all
together.
I decided to make a deal with Transmeta(tm):
my assistance for their aid .
Finally I decided to assist them when, as for my mind,
my assistance will be only anger laugh at them.
Which however do not take off the question - How that conception came to
them ?
And now let me bait at them.
I think.
The algorithm of morfer which I described is not effective.
It of course allows to replace say 386 by PowerPC style CPU.
1. But it may not give sensitive gain.
After certain point it stops and do not accelerate any more.
It because what do you mean under "optimization".
You can not optimize whole program but just only piece each time.
To do such optimization effective you need to implement a special
way of optimization.
Of course, you can have few megabytes of pre optimized algorithms.
But in this case such computer may be turn in perfect player with one
key play,
which can not run formats which it does not know.
Same about engines in games, math. modeling, ect..
However video and sound run from the beginning, without stops and
breaks
in first moments (when morfer does it job).
You can add it to new conception ...
But I meant not this.
First of all it must remove unused CISC functionality.
(Some CISC instructions make operations result of which are not used)
And second level unused functionality actually replace original
fragments
of algorithm by branches of algorithm where each new branch will be
just a case of fragment algorithm.
And vice versa, original fragment will be general case of each branch.
(So if for calculation of original fragment needed 100 clicks, why not
replace
it with two fragments where most frequent will be executed for 80
clicks.)
Order of branch checks may be changed, order of some instructions
or jointed chains of instructions (linked by input - output).
Oh yes about branching, how to create it ?
Where it is obviously it's not a trouble,
but where it is no general solution you may experiment
I mean make an experimental branches in random order
(but functionally correct) and check em during the execution.
But RISC is good, VLIW is better, but best of all multy core
systems.
How translate sequential program on two or more processors ?
Again same conceptions of new calculation.
All processors are needed not always, and
for algorithms which in most cases consists of repeats of smaller
algorithms.
Abstract the loop, and push what remained on processors,
each with respective parameters, and with own loops and with threads.
If there are a lot of loops leave only most useable of them.
In worst case you can divide a chain of instructions
on two or more OCCAM - like chunks.
But there's very interesting moment.
Morfer must fulfill very special function even with paralleled code,
it must restore functionality of program when
it became not able to be executed in parallel mode.
(Because of any reason, but mostly because of input data.)
After it change something in paralleling way.
Again you may use here experiments.
So proper work of paralleling here must not to be guaranteed from the
beginning,
it's dynamic.
And in all of that you may use statistics, and even simple neural
networks
for optimizations.
I wished to write something important but forgot what....
Oh yes.
Algorithm which seams goodly optimized and is already linked
(when it used) may be cut and optimized again, with experiments for
instance.
And finally.
Algorithm (or certain or random its parts)
may be trashed on small logical units and optimized
with general methods of optimization (but very slow).
There is no need to stop execution for all operations (translation and
optimization)
morfer may execute small part of each of them during each call of
code under which it makes.
Results of optimization and transformation of algorithm may be saved in
file,
(if morfer embedded in OS). Gains of morfer in OS is another story.
I just can say that it may help not only with processor dependencies
but any hardware and in some cases software.
How that conception has came to them ?
I admit a telepathy and / or random play,
I even admit that this world is just big Matrix(tm)
and some or all people are agents.
But I tend to think that special services had been watching me,
for having at least some political position, in period of total
depression,
and faster of all I remained alive because I was playing a moron sins
childhood.
So what did I get ?
--Michaelo Mitrofanov (e-mail address removed)
All trademarks given in this text are property of their respective
owners.
"codemorfing" is not the same word with "codemorphing".
I wrote "f" because I always had problems with languages and spelling.
I think that today small companies as Transmeta(tm) have no chances
on the market of general purposes microprocessors.
I suspect they have no chances in obvious future too.
Because the market actually divided between strong players.
I think that Transmeta(tm) was doomed for fault from the beginning.
My treatment of Technical reasons of Transmeta(tm)'s fate I'll give
later.
Now let me tell you the story.
At the beginning of 90xx I though about problems of compatibility of
soviet's amateur's personal / home computers most of which was built on
soviet's copy of i8080 CPU.
As a result I invented a principle, which however I forgot soon.
Later I thought how to run applications for i286 and i386 CPUs on i8088
CPU.
i reminded that principle.
Finally in 1994 or 1995 i read some articles about RISC and artificial
intelligence.
And I invent a conception of technology which I named "codemorfing"
(I read it as "coddemmo'rfin'g")
However I did not implement it and I was not able to do it.
I tried of course. But that year of great depression in Ukraine
I wasn't able even to find documentation and fix my computer.
Political situation in Ukraine was also hard.
I understood that something's going wrong,
and my expressions in address of government was danger for me.
I suspected that somebody is watching me.
I decided to do a horrible trick.
I planed to compose a good looking project of my ideas
and tell about it in several places.
I though such.
1. In best case I'll find a people to whom this project will be
interesting,
and who:
1) Invest in it.
2) or employ me
3) or joint me
4) or at least give me moral support
2. In worse case somebody may implement it without me (as it was
popular)
but I must compose the project such that it must have hidden troubles
which have to be solved with my assistance.
And them the person who implement my project will
1) employ me
2) or pay me for consulting
3) or should explain how did he /she /they got it since all
people who
heard it from me showed no interest for it.
So they collected information about me, why ?
It proves that somebody was watching me. Why ?
And persons who grab my ideas have no law to complain
me in giving improper information.
3. In worst an more probable case:
Project will get fault,
and I'll distribute information about born troubles of the
project.
I thought - How can I do this ?
But finally I decided to survive (remain alive physically) I must
follow the habits of this horrible land.
I.e. if I won't fool people I'll DIE.
That is an abstract of information which I shared:
1. Personal computer have different conceptions of requirements to the
performance than
other kind of computers. Personal computer about 80-98 % of time execute
very small
pieces of code. Maximal performance of machine needed in small pieces of
time.
So performance when it needed may be paid by performance when it do not
needed.
I named it "new conception of performance calculation for personal
computers",
I also compared it with accelerator level of the car.
2. RISC processors are more effective than CISC processors.
Instruction set of RISC processor are optimized for machine generated
code.
So effective conversion code of CISC processor
into the code of RISC processor gives more performance
in some cases even with weaker processors.
3. Codemorfing is a technology of converting machine code
of one microprocessor into code of other microprocessor
directly during execution of the program made by code of first
microprocessor.
Codemorfing allow to effectively convert small pieces of code,
which frequently repeats (in command flow).
(How does it do it read later.
By the way this principle is not my invention indeed,
it was implemented in 80xx, as people explained me)
So 1st + 3rd says that codemorfing is good for personal computer
but not good for system of automatic control.
4. Since performance of processor is required for small pieces of time
(or in general requirements to performance is different in different
moments)
both voltage and frequency variation of processor for optimal
temperature regime
and performance needs
gives preferences in chip lifetime and respect to requests of
performance.
Codemorfing may give an input data for calculation of needed
performance.
For bit of second during the searching entry in data massive
voltage may be rose from 5 V to 12 V and frequency twice.
(Before it I heard about systems with dynamic frequency variation,
and about systems with very slow voltage and frequency variation,
and also about system with frequent (dynamic) changing sleep and and
work modes.
But best results gives exactly dynamic calculation and variation of both
parameters.)
5. Codemorfing works such: special algorithm (morfer)
one by one converts small sequences of instructions
of microprocessor for which program of those instructions was created
into code of microprocessor on which this all executes.
Morfer replace all jumps (GOTOs) on itself and at the end of each
of those pieces of code also put jump on itself. Before jump
morfer set the special identifier (register) which determines
the point from which was done jump.
Morfer converts only that areas of code which are needed for execution.
Morfer conduct statistics of usage of pieces of code,
and after certain number of runs of certain pieces of code
optimize em or convert em again with better quality.
Morfer may replays jumps on itself by jumps on respective
points when further converting or optimization are not needed or
impossible.
Morfer may move pieces of code in memory, with respective swapping
the jumps, restore jumps on itself (but I did not tell why).
During conversion and optimization of pieces of code
morfer may install and remove "agents" (this word I used now) in code
(for making statistics, for instance).
6. Design and manufacturing microprocessors optimized for codemorfing
and supported dynamic voltage and frequency variation together with
respective soft ware may gift big money and glory.
)Much than glory it may allow you to feel yourself a Robin Hood of
industry.
You break the chains processor dependency of software,
in best case actually any program may be ran on your system.
I told it about 10 years ago not such properly, sometimes it sounded
as speech of madman, and I told different parts of this
information to different people.
(I had no Internet, no even modem, and finally no working computer for
the moment.)
But what I told may be turn into sweet illusion without understanding
some hidden traps.
But about this later.
Finally I forgot about those project, I decided that my plan did not
work.
And people really had me as madman.
Then in my life was depressions, thinks about suicide, inferiority
complex.
But in about 1999 I heard about russian project E2K(tm).
Which reminded me my.
Finally I heard that teams which worked at E2K(tm)
crossed with team which worked in mysterious company named
Transmeta(tm).
(By the way, "codemorfing" I also called "codemetamorfing".)
Finally I read about the product of Transmeta in magazine.
I read MY IDEAS with MY thinking way with MY naming.
Yes I found a lot of changing with my original conceptions.
For instance treatment of voltage regulation of Transmeta(tm)'s
Crusoe(tm)
was different. Probably chip does not allow to change voltage such fast.
And conception of performance calculation was also adapted to
multimedia requirements. But ground ideas was same.
Well since I did not patent all those stuff,
for which I had no money, no experience,
but what is most important that was impossible to patent that all
together.
I decided to make a deal with Transmeta(tm):
my assistance for their aid .
Finally I decided to assist them when, as for my mind,
my assistance will be only anger laugh at them.
Which however do not take off the question - How that conception came to
them ?
And now let me bait at them.
I think.
The algorithm of morfer which I described is not effective.
It of course allows to replace say 386 by PowerPC style CPU.
1. But it may not give sensitive gain.
After certain point it stops and do not accelerate any more.
It because what do you mean under "optimization".
You can not optimize whole program but just only piece each time.
To do such optimization effective you need to implement a special
way of optimization.
Of course, you can have few megabytes of pre optimized algorithms.
But in this case such computer may be turn in perfect player with one
key play,
which can not run formats which it does not know.
Same about engines in games, math. modeling, ect..
However video and sound run from the beginning, without stops and
breaks
in first moments (when morfer does it job).
You can add it to new conception ...
But I meant not this.
First of all it must remove unused CISC functionality.
(Some CISC instructions make operations result of which are not used)
And second level unused functionality actually replace original
fragments
of algorithm by branches of algorithm where each new branch will be
just a case of fragment algorithm.
And vice versa, original fragment will be general case of each branch.
(So if for calculation of original fragment needed 100 clicks, why not
replace
it with two fragments where most frequent will be executed for 80
clicks.)
Order of branch checks may be changed, order of some instructions
or jointed chains of instructions (linked by input - output).
Oh yes about branching, how to create it ?
Where it is obviously it's not a trouble,
but where it is no general solution you may experiment
I mean make an experimental branches in random order
(but functionally correct) and check em during the execution.
But RISC is good, VLIW is better, but best of all multy core
systems.
How translate sequential program on two or more processors ?
Again same conceptions of new calculation.
All processors are needed not always, and
for algorithms which in most cases consists of repeats of smaller
algorithms.
Abstract the loop, and push what remained on processors,
each with respective parameters, and with own loops and with threads.
If there are a lot of loops leave only most useable of them.
In worst case you can divide a chain of instructions
on two or more OCCAM - like chunks.
But there's very interesting moment.
Morfer must fulfill very special function even with paralleled code,
it must restore functionality of program when
it became not able to be executed in parallel mode.
(Because of any reason, but mostly because of input data.)
After it change something in paralleling way.
Again you may use here experiments.
So proper work of paralleling here must not to be guaranteed from the
beginning,
it's dynamic.
And in all of that you may use statistics, and even simple neural
networks
for optimizations.
I wished to write something important but forgot what....
Oh yes.
Algorithm which seams goodly optimized and is already linked
(when it used) may be cut and optimized again, with experiments for
instance.
And finally.
Algorithm (or certain or random its parts)
may be trashed on small logical units and optimized
with general methods of optimization (but very slow).
There is no need to stop execution for all operations (translation and
optimization)
morfer may execute small part of each of them during each call of
code under which it makes.
Results of optimization and transformation of algorithm may be saved in
file,
(if morfer embedded in OS). Gains of morfer in OS is another story.
I just can say that it may help not only with processor dependencies
but any hardware and in some cases software.
How that conception has came to them ?
I admit a telepathy and / or random play,
I even admit that this world is just big Matrix(tm)
and some or all people are agents.
But I tend to think that special services had been watching me,
for having at least some political position, in period of total
depression,
and faster of all I remained alive because I was playing a moron sins
childhood.
So what did I get ?
--Michaelo Mitrofanov (e-mail address removed)
All trademarks given in this text are property of their respective
owners.
"codemorfing" is not the same word with "codemorphing".
I wrote "f" because I always had problems with languages and spelling.