ZORR ARM

_________________
I HAVE ABS OF STEEL
--
CAN YOU SEE ME? CAN YOU HEAR ME? OK FOR WORK

@MEGA_RJ_MICAL

ZZ9000.

@Hans

Quote:

While ARM is officially bi-endian, I think we'd be better off making the switch to little-endian. Little-endian won, and it's time to accept it.

This is the smartest thing anyone has ever said about this subject.

I've always preferred big endian. It's intuitive and human understandable for values laid out it memory and maybe it's also potentially better for load store machines that don't do operations on memory directly. However the disconnect comes when you realise your registers behave like a little endian union and on little endian machines, unions in memory behave exactly the same way.

_________________
Doing stupid things for fun...

@Hammer

I only "think" of it as a co-processor - I did mention that for many it is considered impure or none-native as it essentially hijacks and emulates but for me that is just a cleverly designed parallel system with deep integration.

Quote:

PiStorm+Raspberry Pi+EMU68 is not a co-processor for the Amiga since this product transparently takes over from the main 68K CPU.

I would say that it is more than an "accelerator" as it also handles net, ide and hdmi.

My way of thinking is that the PI-Storm is a set of extra processors and chips that are now available to the AMIGA. ROM replacement was a genius move and I am looking to fit one in my A1200/030 but genuinely don't know what to do with my Blizzard MK III if I do!

I dreamed of a full colour AMIGA A1200 in the 90s and now I can afford one (definitely cannot afford a vampire or AOS4 device!)

Back on topic - the PI-Storm approach is certainly a valid way of extending the possibilities of the older platforms and almost certainly soon we will start seeing RISC-V based approaches. I now have a stack of RISC-V boards that I have been programming and think that they are the perfect platform for these types of projects, the PICO for example could handle most of what the PI-Storm does (probably not the HDMI out?) for < $10 including hardware costs.

@Hans

That was @Hammer 's point from his post, I was agreeing with him

@matthey

Very thoughtful and informative post. Thanks!

Without giving too much away, the long term goal is to create entirely open source systems that can be fabricated in any factory without having to pay any licensing and that although cannot compete in terms of IPS (or sadly also wifi performance :( ) with other competitors, is more than adequately fast for most modern computing.

In terms of GPUs, we are still a long way off from fast custom silicon but there are plenty of projects working on it and we already have performance that is useful enough - not ready for the latest video games - but more than fast enough for your STB, watch or retro computer.

You are absolutely right in mentioning that the documentation is letting us down but if you happen to speak Chinese you may not be so disheartened ;)

Everyday more documentation gets written and more frameworks created.

It is already a very stable and enjoyable environment to develop for.

This is complete speculation - but I suspect that the instruction set will at some point produce a massive performance increase as the chips get smaller and mass manufacturing optimises the designs.

@xe54

1. A RISC-V CPU has the same 'wrong' byte ordering like a x86
2. Existing silicone is rare, under powered or overpriced.
3. Soldering a rare and exotic CPU on to a custom made motherboard will result in a very costly system

> Could this be the ultimate AMIGA hardware?

So the 'Amiga' is a computer that was build with an 68k CPU and after the bankruptcy of Amiga Technologies there where no new systems. It would not made sense building an accelerator for an Amiga with such an alien CPU.

If you refer to AmigaOS4 as Amiga, there is only one company that has currently the right of porting it, they have no money and never supported other systems than a few PowerPC boards.

If you are just hyping the RISC V for being an open source CPU design, there are tons of them that where implemented in FPGA. Even the MIPS CPU, that could be called a RISC-V predecessor and was used on the PS1, is OpenSource today.

xe54 Quote:

Very thoughtful and informative post. Thanks! Without giving too much away, the long term goal is to create entirely open source systems that can be fabricated in any factory without having to pay any licensing and that although cannot compete in terms of IPS (or sadly also wifi performance :( ) with other competitors, is more than adequately fast for most modern computing.

There are open hardware supporting groups and businesses, including RISC-V developers, that want to eliminate all licensing and royalty requirements for chip fabrication. Some of the RISC-V hardware comes close if this has not been achieved already. Early fully open hardware is at a disadvantage as there are highly optimized IP blocks for caches, SerDes, etc. that do not currently have modern open hardware options available but eventually this should change. ARM and chip foundries have high quality and fully debugged reasonably priced a la carte IP blocks which will take time to replace. Eventually, open hardware could do for hardware what Linux did for software. The competition should lower the cost to make custom ASICs although specialized licensed IP will still be available to customers that want to buy an advantage over open hardware offerings.

xe54 Quote:

In terms of GPUs, we are still a long way off from fast custom silicon but there are plenty of projects working on it and we already have performance that is useful enough - not ready for the latest video games - but more than fast enough for your STB, watch or retro computer.

At the most basic GPU level, 3D isn't really needed and can be faked with software for limited use. Even the minimalist original Amiga 68kCPU+chipset could be shrunk to a modern chip process and would be low power and have a small footprint. A 3D GPU is more complex and challenging for low power devices though. Part of the problem is that more modern 3D software and APIs practically require a flexible rendering pipeline. Early 3D hardware often used fixed function rendering pipelines and VLIW cores which were low power but not flexible. Think Silicon used VLIW NEMA cores before they switched to the RV64X NEOX cores which gained them flexibility and may partially be the reason they upgraded. The RV64X vector GPU is relatively simple, easy to program for a GPU, scalable and flexible which is appealing for low power devices. It will likely be adequate for simple low power 3D GPU needs but has more potential if the specialized GPU vector instructions provide enough benefit. It's probably not going to compete with highly specialized and tuned commercial 3D GPUs when scaled up but it will be interesting to see how close it can come. Some technology from mobile GPUs have already influenced most GPUs like tile based rendering (see the Samsung link Hans posted which is amazing documentation). Tile based rendering is less flexible and has more limitations than immediate mode rendering but is worth it because of the memory bandwidth reduction which directly translates to power savings. Mobile GPUs often use a fraction of the memory bandwidth of desktop GPUs. APIs and programming styles have been changed to better take advantage of tile rendering. Not only have the number of mobile computer devices overtaken desktops but low power devices have cost advantages. This is why much of the disruptive GPU technology is coming from mobile GPUs. Would we even recognize it though?

https://blog.imaginationtech.com/imagination-technologies-the-ray-tracing-pioneers/ Quote:

Efficiency vs brute force A key difference, of course, is that while the power consumption of a desktop NVIDIA RTX card has been measured drawing 225W while gaming, our solution was designed specifically with mobile power envelopes in mind. Our chip operates at just two watts and the demo board at around 10 watts. What’s more, this was this was built using older 28nm process technology, demonstrating how our solution operated at an order of magnitude of lower power. The peak rate was 300MRay/sec, which considering the power envelope compares favourably with the new NVIDIA cards. We believe that heat and power consumption will be critical in the long-term for ray traced devices. Ray tracing powered by dedicated fixed-function hardware will be far more efficient than doing so using traditional rasterization or general compute hardware. With both bandwidth efficiency and superior quality, our unique high-efficiency focussed approach is a win-win. And while AR and VR are yet to break into the mainstream there is still a lot of belief out there that they will eventually do so. When it comes to VR, to keep everything smooth, techniques such as variable sample rates and foveated rendering need to be employed and these are easier too with our hybrid ray tracing.

If these claims by Imagination Technologies for their ray tracing GPUs are true, then this may be disruptive technology enabling not only mobile devices with 3D ray tracing but also low cost devices like a Raspberry Pi with this technology standard. The Amiga would miss this technology because they don't create custom SoCs. How much could standard 3D ray tracing benefit Amiga marketing anyway? Licensing the Imagination Technologies ray tracing GPU would likely put a "little endian" RISC-V GPU "management" core in an Amiga anyway.

xe54 Quote:

You are absolutely right in mentioning that the documentation is letting us down but if you happen to speak Chinese you may not be so disheartened ;) Everyday more documentation gets written and more frameworks created. It is already a very stable and enjoyable environment to develop for.

The Chinese translation of the documentation may be highly accurate but this shows how technical knowledge is needed too. Contrast this with Samsung that we know hires employees from the U.S. and produces the kind of documentation that attracts customers.

xe54 Quote:

This is complete speculation - but I suspect that the instruction set will at some point produce a massive performance increase as the chips get smaller and mass manufacturing optimises the designs.

Improved chip processes should improve performance but I'm not so sure that improved RISC-V core designs can fully overcome the inherently weak ISA. Breaking code down into more smaller instructions allows smaller cores using less power but is bad for a high performance core. Instructions have to be executed at an increased pace to keep up and more stalls are created between instructions. OoO execution and code fusion/folding can only partially fix this and OoO execution greatly increases the size of the core and power requirements. For something like the RV64X GPU vector extensions to work, small cores are needed (this is why the Larrabee architecture used in-order x86-64 cores based on the P54C Pentium). It's easy to add many weak RISC cores but most code is not parallel enough to use them. New high performance designs for MIPS, SPARC, PA-RISC, etc. were not attempted anymore after it became obvious that they would not be competitive. Even the more powerful PPC ISA stopped seeing high performance new designs after the disappointing G5 CPU and there were several mediocre performance but power efficient core designs. ARM created AArch64 which is closer to CISC instruction complexity and power to better compete in performance with x86-64.

An optimum RISC-V core is like a piston internal combustion engine (ICE). It's simple, reliable and energy efficient but it is not what you want for high performance. More valves can be added, timing changed, combustion chamber shape changed, etc. but it still has a bottleneck when pumping air past valves and an ICE is a big air pump. The big bottleneck for RISC is memory access. CISC cores don't have the bottleneck to memory so are inherently more powerful like a gas turbine or Wankel engine. They can't be as energy efficient for a simple fuel efficient design but they become more competitive in energy efficiency for a high performance design. The same is true for CISC cores where x86-64 cores can do more work for the same power, get the job done faster and return to a low power sleep mode which is why they are competitive for servers.

@MEGA_RJ_MICAL

Fair point

@OneTimer1

Quote:

If you are just hyping the RISC V for being an open source CPU design, there are tons of them that where implemented in FPGA. Even the MIPS CPU, that could be called a RISC-V predecessor and was used on the PS1, is OpenSource today.

Yes I think this was my main point, to introduce RISC-V real hardware as a possible path. The fact is RISC-V is being actively developed now, can be fabricated anywhere and you don't have to pay royalties and licenses - it is far more capable and future proof than a discontinued chip from the playstation.

Soon the chips will be extremely cheap and more than fast enough.

Quote:

If you refer to AmigaOS4 as Amiga, there is only one company that has currently the right of porting it, they have no money and never supported other systems than a few PowerPC boards.

True. That would have to change to move forwards... or we use the alternates that already exist... AROS, MORPH, SCALOS, COFFIN. I trust Trevor more than anybody that came before him but also having been in this limbo state since the 90s am under no illusion about the future of AMIGA or lack thereof.

@xe54

This really is pointless when it'll be hard enough to get Hyperion to port AmigaOS to the OpenPPC laptop project and that will probably take 5 years even if they want to!

WE HAVE VERY FEW RESOURCES AS A PLATFORM AND THE ONLY OPEN SOURCE OPTION CURRENTLY IS AROS ON X86-64! WHY MUDDY THE WATERS FURTHER? SERIOUSLY?

_________________
"Art challenges technology. Technology inspires the art."
John Lasseter, Co-Founder of Pixar Animation Studios

@BigD

Quote:

This really is pointless when it'll be hard enough to get Hyperion to port AmigaOS to the OpenPPC laptop project and that will probably take 5 years even if they want to!

5 years is probably optimistic. A laptop has a lot of custom parts where you do not have any existing drivers yet. It will require wifi support as well as power management. Nothing that exists today in os4.

Hyperion can not even port os4 to Tabor in 5 years. An it is just a standard ppc motherboard with a slightly different cpu, that mostly just affects a few compiler settings and not much actual code.

Hyperion will never port AOS4 to a different platform. PPC or not. They do not have anyone with the skills to do so.

For those struggling with cognitive dissonance when it comes to this subject matter:

Anything 'Amiga' in the future that would be leveraging a yet to be designed and produced RISC-V CPU, would have absolutely N O T H I N G to do with Hyperion, AmigaOS 4.x, A-Eon, or anything that people consider to fall under the catch-all term 'Amiga' today.

A clean slate, or green fields future 'Amiga' would sever all ties with legacy architecture. Yes, it could then just as easily be based on ARM or x64, but since we're spitballing, there is some merit for it to be based on RISC-V.

_________________
All the way, with 68k

@xe54
Quote:

Yes I think this was my main point, to introduce RISC-V real hardware as a possible path. The fact is RISC-V is being actively developed now, can be fabricated anywhere and you don't have to pay royalties and licenses - it is far more capable and future proof than a discontinued chip from the playstation.

My NVIDIA GTX 1650 Super (TU116), RTX 3070 Ti (GA104), and two 3080 Ti (GA102) GPUs have customized RISC V and it does nothing for accelerating Amiga 68K legacy.

GSP (NV-RISCV) has been used as GPU's controller core.

NVIDIA's Tesla T10 is based on the year 2009 GT200B ASIC has NV-RISCV. GT200B is used SKUs such as GeForce GTX 260, https://www.techpowerup.com/gpu-specs/nvidia-gt200b.g142

NVIDIA has enabled GSP (NV-RISC) for Tesla T10.


PlayStation's MIPS CPU still continues into MIPS64 Release 6.06. MIPS CPU failed to create a unified desktop computer platform.

Last edited by Hammer on 02-Aug-2022 at 04:35 AM.
Last edited by Hammer on 02-Aug-2022 at 04:25 AM.

_________________
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB
Amiga 1200 (Rev 1D1, KS 3.2, PiStorm32lite/RPi 4B 4GB/Emu68)
Amiga 500 (Rev 6A, KS 3.2, PiStorm/RPi 3a/Emu68)

@Hans

Quote:

Hans wrote: What HAL are you talking about? ExecSG has a Hardware Abstraction Layer (HAL), needed in part because of low-level variations between PowerPC chips (e.g., certain config registers). So, that's not a barrier to getting the kernel working on ARM. Getting existing software to work is a different matter, because it's all PowerPC and 68K. While ARM is officially bi-endian, I think we'd be better off making the switch to little-endian. Little-endian won, and it's time to accept it. Hans

Static Windows ACPI HAL targets "designed for Windows" firmware HAL ACPI.

To boot/run ACPI-enabled Windows NT, AMD supplied "designed for Windows" firmware HAL ACPI for PS5-based AMD 4700S solution.

Windows 7/8/10 are not specifically aware of incoming AM5 B650/X670 motherboards and Windows 7/8/10 will boot and run on these incoming motherboards.

ExecSG having Hardware Abstraction Layer (HAL) is meaningless when my existing AmigaOS 4.1 FE can't boot and run on Power9 Raptor!

I have AmigaOS 4.1 for SAM440 and classic Amiga PPC hardware. How many times I would pay for AmigaOS 4.1 licenses for each hardware variation? PowerPC's unified ecosystem promise is BS and I'm tired of unified promises from the non-X86 camp.


AmigaOS 3.1/3.1.4/3.2 68K for A1200.. it can support CPUs from 68020 to AC68080 to PiStorm32-Pi 4 CM. SAM440/SAM460 has a dead-end road map worst than classic Amiga hardware.


AmigaOS 4.0 ExecSG HAL argument wouldn't match Windows 2000 running on the X570 motherboard since AmigaOS 4.0 wouldn't run on the Power9 Raptor motherboard.

Power9 Raptor PCIe 4.0 motherboard is equivalent to AMD X570 PCIe 4.0 motherboard. The PowerPC world has a current-gen PCIe 4.0 motherboard and AmigaOS 4.1 doesn't run on it! LOL

Last edited by Hammer on 02-Aug-2022 at 07:10 AM.
Last edited by Hammer on 02-Aug-2022 at 07:07 AM.
Last edited by Hammer on 02-Aug-2022 at 07:01 AM.
Last edited by Hammer on 02-Aug-2022 at 05:18 AM.
Last edited by Hammer on 02-Aug-2022 at 05:12 AM.
Last edited by Hammer on 02-Aug-2022 at 05:11 AM.
Last edited by Hammer on 02-Aug-2022 at 05:06 AM.
Last edited by Hammer on 02-Aug-2022 at 05:05 AM.
Last edited by Hammer on 02-Aug-2022 at 05:02 AM.

_________________
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB
Amiga 1200 (Rev 1D1, KS 3.2, PiStorm32lite/RPi 4B 4GB/Emu68)
Amiga 500 (Rev 6A, KS 3.2, PiStorm/RPi 3a/Emu68)

@agami

Quote:

For those struggling with cognitive dissonance when it comes to this subject matter: Anything 'Amiga' in the future that would be leveraging a yet to be designed and produced RISC-V CPU, would have absolutely N O T H I N G to do with Hyperion, AmigaOS 4.x, A-Eon, or anything that people consider to fall under the catch-all term 'Amiga' today. A clean slate, or green fields future 'Amiga' would sever all ties with legacy architecture. Yes, it could then just as easily be based on ARM or x64, but since we're spitballing, there is some merit for it to be based on RISC-V.

68K drop-in replacement solutions such as PiStorm+Pi-3a (or Pi Zero 2W) or AC68080 will run with C= Commodore-Amiga Inc's classic Amiga 500 hardware and Hyperion's AmigaOS 3.14/3.2.x or Amiga Inc/Cloanto's AmigaOS 3.X.

For Hyperion, AmigaOS 3.1.4/3.2.x 68K sales are beating AmigaOS 4.x PowerPC camp.

A-EON Technology Ltd's AmigaOne A1220 (with Freescale QorIQ P1022 e500V2, dual-core, 1.2 GHz, 32-bit, non-standard 64-bit SIMD).

vs

C= Commodore-Amiga Inc's Amiga 1200 with PiStorm32-Pi 4CM (with quad-core ARM Cortex-A72 processors @ 1.5 GHz, 64-bit ISA, 128-bit NEON SIMD). Still runs with Commodore-Amiga designed AGA chipset. 68K CPU component is not from Commodore-Amiga since it's from Motorola.

_________________
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB
Amiga 1200 (Rev 1D1, KS 3.2, PiStorm32lite/RPi 4B 4GB/Emu68)
Amiga 500 (Rev 6A, KS 3.2, PiStorm/RPi 3a/Emu68)

@Hammer

Quote:

ExecSG having Hardware Abstraction Layer (HAL) is meaningless when my existing AmigaOS 4.1 FE can't boot and run on Power9 Raptor!

A HAL doesn't magically make the OS work on any CPU. All it does is keep all CPU/motherboard specific code in one place. Work still needs to be done for the various hardware, whether that work is on OS code, or firmware code (like the "designed for Windows' firmware code you mentioned).

I'm sorry that AmigaOS 4.x doesn't work on the Power9 Raptor motherboard, but someone needs to write the HAL code (and drivers) to make it possible. In the PC world, hardware manufacturers will do the work themselves for Windows compatibility. AmigaOS' market is so tiny that we don't have that luxury.**

I see no technical barriers to providing the HAL code for all supported motherboards on a single OS installation CD (other OSes do it). Making per-motherboard-type distributions was Hyperion decision. Changing their mind on this is likely to be difficult...

Hans


** Hence, why you have one guy writing the graphics drivers instead of AMD's development team providing the drivers "out of the box."

_________________
http://hdrlab.org.nz/ - Amiga OS 4 projects, programming articles and more. Home of the RadeonHD driver for Amiga OS 4.x project.
https://keasigmadelta.com/ - More of my work.

@Hammer

Quote:

For Hyperion, AmigaOS 3.1.4/3.2.x 68K sales are beating AmigaOS 4.x PowerPC camp.

That is probably not surprising. Do you remember AEOv1 and how it got dissolved? Because allegedly someone took money that Trevor put in there, earmarked for Amiga stuff, and allegedly used it to fund an unrelated lawsuit?
And in the aftermath after AEONv2 was started, allegedly offer was made to someone that either we take this to court and someone goes to prison for theft or we get FREE AOS4 licenses for every AmigaOne motherboard we produce, forever.
And allegedly, someone picked the "not prison" option.

So, exactly how many OS4 licenses do you think they have been selling for actual revenue lately in the last 5 years?
10?

Chances are high that I might have loose change in my "travel jar" that exceeds Hyperions sales of OS4 for the last several years.

Last edited by V8 on 02-Aug-2022 at 09:02 AM.

@Hans

Quote:

A HAL doesn't magically make the OS work on any CPU. All it does is keep all CPU/motherboard specific code in one place. Work still needs to be done for the various hardware, whether that work is on OS code, or firmware code (like the "designed for Windows' firmware code you mentioned). I'm sorry that AmigaOS 4.x doesn't work on the Power9 Raptor motherboard, but someone needs to write the HAL code (and drivers) to make it possible. In the PC world, hardware manufacturers will do the work themselves for Windows compatibility. AmigaOS' market is so tiny that we don't have that luxury.** I see no technical barriers to providing the HAL code for all supported motherboards on a single OS installation CD (other OSes do it). Making per-motherboard-type distributions was Hyperion decision. Changing their mind on this is likely to be difficult... Hans

FYI, Windows 11 X64's APCI X64 HAL is supplied by Microsoft.

The original year 2009 Windows 7 build still boots, installs, and works on July 2019 era AM4 X570 motherboard. AM4 socket and X370 chipset-based motherboards were released in September 2016. ACPI firmware assures compability with Windows NT ACPI HAL target.


https://www.youtube.com/watch?v=iRMSry8UlqE
Windows 3.1/MS-DOS/basic VGA still boots and runs on a modern UEFI/ACPI-based X86-64 PC with Intel Core i5-7400 Skylake Refresh.

Windows 3.1/MS-DOS is not aware of the new features.


https://www.youtube.com/watch?v=QVdW_SQG5ZY
Windows NT 3.51 still boots and runs on a modern UEFI/ACPI-based X86-64 PC with Intel Core i5-6500 Skylake.

Last edited by Hammer on 03-Aug-2022 at 07:28 AM.

_________________
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB
Amiga 1200 (Rev 1D1, KS 3.2, PiStorm32lite/RPi 4B 4GB/Emu68)
Amiga 500 (Rev 6A, KS 3.2, PiStorm/RPi 3a/Emu68)

PADDING

_________________
I HAVE ABS OF STEEL
--
CAN YOU SEE ME? CAN YOU HEAR ME? OK FOR WORK