RadeonHD V.5 driver

@K-L

Quote:
You're welcome.

These days, AMD's own developers work on the Linux drivers, so there's a fully paid team involved. A lot of money gets spent on Linux development...

Quote:
True. I don't have the luxury of starting side projects of this scale. So, for me to embark on such a project I'd have to be confident that it'll do more than just pay the bills (and I'd need to convince an investor or two that it's worth spending money on).

Hans

@Hans

Can you set up Go Fund me, Patreon etc? or something.

@PixelHi

Quote:
I'd better finish some of my existing projects first...

A few other things to bear in mind in terms of developing a GPU specifically for AmigaOS 4.x:
- An FPGA based GPU is unlikely to reach anything near commercial GPU performance (e.g., have a look at the Apollo Maggie 3D demo). We're not talking about slightly slower, but a lot slower than what we're using now
- Performance could be greatly improved my making an ASIC, which requires lots of $$$
- The GPU alone wouldn't provide hardware accelerated video encoding/decoding. Video codecs are complicated, and I have no idea if a video codec core could be obtained from somewhere

Such a GPU would be more suited to the electronics hobbyists crowd, and modern 8/16-bit retro computers where the needs and expectations are much lower. At least, unless some major funding could be put behind it (which you'd need a compelling case for given the GPUs already available).

Hans

@Hans

Another idea, get a contract from the AMD crowd, sign an NDA, work with and for the Linux driver source and port that (you'd still be a one man show, but you wouldn't need to do guesswork as much *and* you could pester the Linux devs if something doesn't work)

Probably completely bull as i don't know how driver development works

@Hans

I'm little surprised that various software/driver developers don't have a donate "button" next on thier websites.

I was wondering if Os4Depot could add "donate" button next to download. I'm using e.g. driver for Sound Blaster or it become to complicated due to various issue due to donors wanting e.g. refunds), sure it free but even if I wanted I have no way support it. Or take e.g. ignition development.

Not my area of expertise.

Your work (I hope!) is supported by sale of Enhancer /v5 driver, but what about supporting RadeonRx development? Is it going to be sold separately as RadeonHD driver?

@Raziel
Quote:
No thanks. I have no desire to work for a large corporation like AMD, nor do I want to make graphics drivers my life's work. Besides, we can already use the Linux AMDGPU source code, and AMD have a GPU driver dev. email address where I can ask for help.

What I'd love to see, is a linux driver wrapper so that we can use their code almost unchanged. This would reduce the workload to get drivers working considerably.

One snag: we're almost the only ones still using big-endian, so both the GPU hardware and drivers no longer support it. This adds to the workload of porting driver code. I've been suggesting we switch to little-endian for this reason.


@PixelHi
Quote:
That's a question for A-EON to answer. My impression is that it'll continue to be bundled with the Enhancer Pack.

Hans

So, I got MSI Radeon RX 550 and it will not work with my Sam460. Black screen. System boots fine if I swap to my 7750.

Is there anything I need to change in a Uboot/AmigaOS before swapping GPUs?

@PixelHi

Check the following re the Radeon RX driver installation :


In the file System:Kickstart/kicklayout make sure you that have the following line :

MODULE Kickstart/RadeonRX.chip

in the same directory System:Kickstart/

Make sure you have the following file :

RadeonRX.chip

@Spectre660

Thank you! It works now. Looks like installer from Enchancer did not add this line to kicklayout file, driver was copied there already.

So, RX 550 works great, all my Radeon HD works fine with exeption of three R7 240 and 250 will not.

@Hans
Quote:Using an endian-swapping version of GCC might help, like the one which was used for building native x86/x64 Amithlon software.
That way you don't need to adopt the sources, such special GCC versions will add the required code to swap any 16, 32 and 64 bit reads and writes, and since most (or even all?) PowerPC and POWER CPUs have special endian swapping instructions there wouldn't even be a speed penalty unlike with the Amithlon x86/x64 GCC which had to add additional instructions for it.

@joerg

I tried use the reverse load and store instructions I was disappointed with results. If remember correct, I got as poor results as bit mask bits, then shift & Or result code. Yeh it might be I don’t have the instruction, or maybe some trap instruction interrupt bug, some forgot check what CPU has or has not the instructions, not sure how that works.

Wont you just get lot of conversions all over place, where did not need them, if use LE GCC compiler, most cases you load 32bit and store 32bit, the order of what is stored and loaded is irrelevant, because you’re not doing anything with content.

Also if part of the code is LE format (the driver), and other part is BE format (the OS), you need to do some conversion between the two. You have only moved problem from between GFX card and the Driver to between the driver and OS.

Also pretty sure the GCC linker will refuse to link LE code with BE code, as detect its not correct ELF format.

Big brother “Power” I believe can map some memory ranges to LE or BE format, where if you put a struct in one memory range it can be LE format, and in other memory range BE format, this approach be simplest, can it be heck’ed in with a fast small FPGA? Or approach like give too much latency. Just telling CPU that the PCIe bus is LE format, that be sweet.

@PixelHi

If the R7 240 and 250 show U-Boot then there is a chance
that they may work with RadeonHD drivers after V5.14 .

@LiveForIt

You can't use standard LE GCC versions but would have to build a special one like the Amithlon GCC. May be a lot of work to port more than 20 years old GCC code to current GGC versions, but would be "just" a one-time-job, unlike the changes Hans has to to for each update of the LE Linux AMD sources.

Mixed endian isn't not only supported by POWER CPUs (like the one used in the X1000), at least the 7xy and 74xy PowerPC CPUs (as used in the EyeTech A1 and Pegasos 1+2) had segment (IIRC 256 MB on AmigaOS 4.x) based endian flags as well, i.e. parts of the memory (or PCI I/O space) can use big endian while others use little endian accesses instead.
Not sure about the 6xy (classic Amiga CyberStormPPC and BlizzardPPC) and 4x0 CPUs (SAM), there might be a problem with them, and I don't know if QorIQ CPUs (X50[24]0) support it.

@Spectre660

Is there a way to get V5 for Sam460? AmiStore only have versions for X1000 and X5000.

@Hans

Is lack GART support for Sam460 (whatever that is) limitation of Sam460 or its just not easy to implement for it comparing x1000/5000?

In other words will it ever work for Sam460 for RadeonHD/RX?

@PixelHi

Can't answer that.
But Looks like the Sam460ex version may not be available via Amistore until some bug fixes are released.
I would say that is a good think though.

Quote:

I think it a glitch in AMIStore. When I run Updater it shows there is a 5.14 driver for Sam460 available as an update.

@PixelHi

GART is a Memory Management Unit which allows Direct Memory Access access from graphics card to the main memory.

But it will require cache coherency to work properly. And this is not implemented/supported on the SAM4x0.

The X1000 and X5000 offer cache coherency, so no need for manual cache handling.

@geennaam

Sounds like a dead end for Sam460 performance.

Also, I installed v5.14 driver and now my 7750 and r7 240 will not work. I'm getting and error saying: object not found. System is frozen.

I used installer and driver is installed.

Good that my Rx is working, so I can still fix it. Can someone please help me here 😔.

EDIT: After cold boot, Rx will not show screen at all. Time to do fresh install...

Usually you don’t have to reinstall an Amiga. Have you checked your Kicklayout?