Hi

The Apollo team is happy to announce the immediate disponibility of the Gold 2.7 core for all Vampires 500, 500+ and 600. You can right now dowload the appropriated core on this link :

http://wiki.apollo-accelerators.com/doku.php/start#latest_core

The Gold 2.7 core has many inprovements, the most notable one being of course the hardware FPU. But this is not the only one new thing ! Here is the changelog :

[GOLD2.7] (02.03.2018)
* Added fully pipelined hard FPU with peak 78 MFlops
* Added native support for all 68882 FPU operations without need of extra libraries
* Added new high performance memory controller, reaching 600 MB/sec memory speed
* Added MapRom support
* Added Workbench hardware sprite support
* Added Scanlines optional mode for retro look of games on HDMI output
* Added RangerMem (A500 slowmem) for better compatibility with some old games/demos
* Added new boot pic
* Improved MPU (memory protection) with catching memory access error Enabled
* Improved AMMX 2 instruction set with acceleration features for video decoding, image decoding, 2D-games, 3D-games
* Improved WHDLOAD compatibility
* Improved Faster Boot time

Like always, we are waiting your feedback on IRC :
chat.freenode.net 6697 (SSL)/6667 (non SSL) #Apollo-Team

We wish you a good installation !

Last edited by guibrush on 02-Mar-2018 at 12:23 PM.

@guibrush

Quote:

Workbench hardware sprite support

This interests me. What's special about this? It almost sounds like a pointer. But I wonder what's so special about the Workbench here? To me it shouldn't matter what software is using sprites, as long as there are hard sprites.

@guibrush

Nice update. The Scanlines addition is a nice touch. The Apollo Team has done a great Job. Thanks!

Drewlio77

@Hypex

I am just speculating here, but... hardware sprite on RTG perhaps, assumption being that all users use RTG for Workbench?

Last edited by kolla on 02-Mar-2018 at 02:23 PM.

_________________
B5D6A1D019D5D45BCC56F4782AC220D8B3E2A6CC

@guibrush

Thanks for the hard work you do for the Vampire accelerators. I'm mainly curious about the Vampire 1200 and if you guys are getting closer to finishing it? Thanks.

Ed

_________________
Amiga 1200 - ACA 1233 68030 128MB Ram
8GB CF With tons of Classics

AmigaOne X5000

Raspberry PI 400 - PiMiga 1.5
"That which doesn't kill you, only makes you stronger" - Someone important, but I forgot who

@guibrush

The FPU seems to have some issues, as some simple test cases reveal:

https://blog.alb42.de/2018/03/02/vampire-v2-7-with-fpu/

Rant: Is it really that hard to run a semi-serious compiler testsuite to validate your instruction set implementation? Did they make too many corner cuts to squeeze the FPU in the FPGA? To be seen...

_________________
MorphOS, classic Amiga, demoscene, and stuff
"When a bridge is not enough, build a Viaduct!"
"Strip the Amiga community of speculation and we can fit every forum on a 720k floppy" (by resle)

@Chain-Q
Actually, after a bit deeper investigation with Alb42, it looks like the Vampire FPU strongly prefers 32bit floats over higher precisions. I cannot yet tell if it simply rounds everything to 32bit and does all operations in 32bit precision, or it's only their default load-store precision is different and internally keeps the higher precision, but either way, its current behavior is totally incompatible to a real 68k FPU.

But at least the benchmark results are amazing, aren't they?

_________________
MorphOS, classic Amiga, demoscene, and stuff
"When a bridge is not enough, build a Viaduct!"
"Strip the Amiga community of speculation and we can fit every forum on a 720k floppy" (by resle)

@Chain-Q

I guess they're well aware of it, in fact it was their choice to use less precision in order to make the FPU fit into the currently used FPGA.
It is documented here where it is also specified that the current Apollo FPU is intended to be used to run mostly games and demos.

Of course a full precision FPU would be more than welcome, but I guess we'll have to wait for a bigger FPGA implementation.

_________________
God created Paula so that Allister Brimble and Dave Whittaker could do music

Check my Amiga gameplays (ITA)!

@Chain-Q

Quote:

Chain-Q wrote: @Chain-Q either way, its current behavior is totally incompatible to a real 68k FPU. But at least the benchmark results are amazing, aren't they?

Yeah totally incompatible. Running Quake, Demos and other stuff with no problems sure makes it unusable.

@WOSPUPOS4

Quote:

WOSPUPOS4 wrote: @Chain-Q Quote: Chain-Q wrote: @Chain-Q either way, its current behavior is totally incompatible to a real 68k FPU. But at least the benchmark results are amazing, aren't they? Yeah totally incompatible. Running Quake, Demos and other stuff with no problems sure makes it unusable.

Let's not forget Imagine and Lightwave

Last edited by ErikBauer on 03-Mar-2018 at 11:31 AM.

_________________
God created Paula so that Allister Brimble and Dave Whittaker could do music

Check my Amiga gameplays (ITA)!

Mjnurney recording of his expriences with the X12 core 2.7.

https://www.youtube.com/watch?v=Vb5fL1ivNMo

@ErikBauer
The float precision is not something you should debate or negotiate, it's not like "yeah, well, it's a bit less precise, who cares".

Same FPU instruction with the same operands with the same hardware settings gives different output than on any other hardware 68k FPU implementation. In my book that's called incompatible.

Remember the Intel FDIV bug in the Pentium? There one in 9 billion division gave different results than the expected one, if you used FDIV with random parameters. Intel was still forced to call back the chips, and was ridiculed by competing platforms (including Amiga users) for decades.

Heck, I reported much smaller issues to QEMU-m68k for example, they were immediately fixed. But if the Vampire does it, somehow it's just fine, because Imagine now works, and we're building The Better Future, and the laymen don't care anyway, so we can get away with breaking the entire instruction set...

_________________
MorphOS, classic Amiga, demoscene, and stuff
"When a bridge is not enough, build a Viaduct!"
"Strip the Amiga community of speculation and we can fit every forum on a 720k floppy" (by resle)

@Chain-Q

You are 100% right, not telling you're not.
Perhaps we see Apollo with two different perspectives, and I'll explain you mine:
Today Apollo core implementation sold with Vampire V2 is far to be complete and it clearly needs more powerful FPGA HW to run perfectly. Today Apollo Core is a live Alpha of what I believe the final result will be. Reading Gunnar's and the rest of the team's posts it is clear they are well aware of this and they continue to repeat it to everyone. It is clearly stated in the FPU documentation they published.

This, I can understand and accept. V2 is a step towards something more complete, and is a step I like and appreciate. Moreover it is a step that lets you play a bit with it and clearly tells you what you shouldn't expect from it.

Last edited by ErikBauer on 03-Mar-2018 at 12:10 PM.

_________________
God created Paula so that Allister Brimble and Dave Whittaker could do music

Check my Amiga gameplays (ITA)!

@Chain-Q
Actually, it turns out, the Apollo Core team even documented this in their FPU specs, by calling it "reducing the width of the FPU", to "either 64bit or a little less". Which is nowhere explicit to be taken seriously, and a serious downplay on such a fundamental issue.

_________________
MorphOS, classic Amiga, demoscene, and stuff
"When a bridge is not enough, build a Viaduct!"
"Strip the Amiga community of speculation and we can fit every forum on a 720k floppy" (by resle)

@Chain-Q

Again, you're right. That single sentence does not tell anything specific.
And again, it is a live Alpha, and I'm confident that in future implementations (If not V4, maybe V5) there will be enough room for full bit precision.

_________________
God created Paula so that Allister Brimble and Dave Whittaker could do music

Check my Amiga gameplays (ITA)!

@ErikBauer
My problem is, I'm developing a developer tool. From now on, there's one more issue to care about. People will run their machines with this FPU enabled, and all the bugreports in perfectly well written software caused by a misbehaving/broken/incompatible/reduced-width/name-it-what-you-want FPU will pour down on me, and people who use the developer tools I provided. And I have zero interest or intent in supporting this abomination.

People just have so little concept of how a CPU and an architecture works. The OS4 world ran into their incompatible FPU mess with the Tabor SPE, now we have to deal with similar sh*t on classic, thanks to the Vampire team. Let me be not cheered up by this.

But I stop commenting on the issue, because no one cares anyway...

_________________
MorphOS, classic Amiga, demoscene, and stuff
"When a bridge is not enough, build a Viaduct!"
"Strip the Amiga community of speculation and we can fit every forum on a 720k floppy" (by resle)

@Chain-Q

Problem is, most people will not understand Vampire FPU is far from being a complete and fully compatible FPU. Yes I understand your concerning.

Have you tried to address it somewhat with the team?

Even if, to me, the sentence Quote:

People shall use the FPU for what it is intended to on the Vampire, that is to run most of the Amiga Demos and Games requiring an FPU.

alone should clearly be intended as "this FPU is not fully compatible and you should not hassle developing tool developers for software not running on it".

_________________
God created Paula so that Allister Brimble and Dave Whittaker could do music

Check my Amiga gameplays (ITA)!

im not sure where precission has significant impact on compatibility, other than on corner cases such as rounding, where an incorrect result can prevent the application from performing correctly. of course as soon as operations stack upon results of each other and precission decays in the process, the final results may differ significantly. though since floating point is always only approximation of the actual number an application should imho be prepared to deal with this.

@Chain-Q

People care, but as they say; size matters, especially with regards to FPGA. Then it becomes a tradeoff between what features can be included to get the best expirience. Obviously from your perspective as an active developer, it presents a problem/pain/frustration.

Or something to adapt to (easy for me as a non developer to say, I agree). Question to you; How active are the legacy FPU being used by actual users of said tool(s)? Democoders use FPU, and for the most part old and current demos works just fine.

Or something to be ignored, which seems to be your decision.

@Overflow
How actively a certain instruction set or a feature being used - you cannot tell with a developer tool. That's the whole point of it. People will just use for whatever the heck they want. But actually, at least one application developed by Alb42 heavily uses it, and it depends on the features of it. Fortunately, we do have or software floating point library, so we can continue to use that (although on the expense of being much slower), just as people can still run FEmu....

@Thread:
BTW, Alb42 summed up some of his findings here:
https://blog.alb42.de/2018/03/03/vampire-2-7-fpu

Basically, it seems to use some sort of a single-double hybrid, with 23 bits mantissa and 11 bits exponent (or something near those values), so it can go up to very high numbers, but loses terribly on the precision front, of course. And it's not IEEE 754 compliant. And you don't need to do extreme numbers or wait until the error "accumulates" to end up with precision loss, due to the nature of floats. As I wrote on Alb42's blog in a comment:

moveq.l #5,d2
fmove.w d2,fp2
fmul.d #1.3,fp2

This instruction stream (copied from an actual test app) essentially multiplies the integer value of 5 by 1.3. This should result in 6.5. On the Vampire FPU, this results in 6.499999523, according to our tests. It's one single instruction, and you already see the problem.

Last edited by Chain-Q on 03-Mar-2018 at 04:14 PM.

_________________
MorphOS, classic Amiga, demoscene, and stuff
"When a bridge is not enough, build a Viaduct!"
"Strip the Amiga community of speculation and we can fit every forum on a 720k floppy" (by resle)