|
|---|
| | 
A1200 Coder
Posts 74
08 May 2017 10:11
| Regarding the demoscene, and those using 68060+FPU and AGA for demos, I dont think they are doing much realtime stuff, but prerender scenes that last 10-20 secs. That's why those demos require 64 MB or more memory. This is not impressive, you need to calculate stuff in realtime to be useful, for e.g. games. A 68060 is too slow for anything else than simple affine texturemapping in low-res and realtime, and that you can do with fixed point arithmetic. When one needs perspective correct texturemapping an FPU starts to be handy, or at least thats the way it went on the PC side.
| |
| | 
Mr Niding
Posts 459
08 May 2017 10:27
| Im just going to copy paste; Making 'high-end' Amiga demos - Michal 'Kiero' Wozniak - Demobit 2017 EXTERNAL LINK
| |
| | 
Gunnar von Boehn
(Apollo Team Member)
Posts 6222
08 May 2017 12:11
| A1200 coder wrote:
|
Regarding the demoscene, and those using 68060+FPU and AGA for demos, I dont think they are doing much realtime stuff, but prerender scenes that last 10-20 secs. That's why those demos require 64 MB or more memory. This is not impressive, you need to calculate stuff in realtime to be useful, for e.g. games.
A 68060 is too slow for anything else than simple affine texturemapping in low-res and realtime, and that you can do with fixed point arithmetic. When one needs perspective correct texturemapping an FPU starts to be handy, or at least thats the way it went on the PC side.
|
Thanks for the excellent post!
| |
| | 
Przemyslaw Tkaczyk
Posts 155
08 May 2017 12:31
| A1200 coder wrote:
|
Regarding the demoscene, and those using 68060+FPU and AGA for demos, I dont think they are doing much realtime stuff, but prerender scenes that last 10-20 secs. That's why those demos require 64 MB or more memory. This is not impressive, you need to calculate stuff in realtime to be useful, for e.g. games.
|
CITATION NEEDED. Give some actual examples, and don't play a guessing game (because you are wrong). What you think doesn't make it true.
A1200 coder wrote:
|
A 68060 is too slow for anything else than simple affine texturemapping in low-res and realtime, and that you can do with fixed point arithmetic. When one needs perspective correct texturemapping an FPU starts to be handy, or at least thats the way it went on the PC side.
|
Writing advanced, elegant math code depends on the FPU heavily. Just because it can be done in integer doesn't make it optimal. And don't assume that all we want to do is "simple texturemapping" in low-res. Again, talk with Kiero/Elude and ask him about his physics in later demos.
| |
| | 
Chain Q
Posts 19
08 May 2017 12:42
| | Regarding the demoscene, and those using 68060+FPU and AGA for demos, I dont think they are doing much realtime stuff, but prerender scenes that last 10-20 secs. |
Except, most of them don't do that.| That's why those demos require 64 MB or more memory. |
I can count on a single hand how many demos need 64 MB. Most of the 060 demos run with 16MB, or 32MB just fine. (And quite a chunk of that is used for the streaming music.)| This is not impressive, you need to calculate stuff in realtime to be useful, for e.g. games. |
Quake runs pretty well on a 060. It's realtime. And it needs FPU. Do the math. (Pun intended.)| A 68060 is too slow for anything else than simple affine texturemapping in low-res and realtime, and that you can do with fixed point arithmetic. |
Except, it isn't. You can do perspective correction in realtime on a 060 just fine. (Again: Quake...) Also, the texture mapper is not the part which needs the FPU (that much) in a well written engine.| When one needs perspective correct texturemapping an FPU starts to be handy, or at least thats the way it went on the PC side. |
I was involved in making softrender PC demoengine back around 1998-2000. It used MMX in the texture mapper, but not an FPU. Not in the inner loops of the mapper at least. (The FPU was too slow for that, even on a ~300-400 Mhz x86.)Bottom line: what you wrote is an insult to most 060 demomakers. But fortunately it only shows that you've got little clue how those demos work internally.
| |
| |
Chain Q
Posts 19
08 May 2017 13:14
| Gunnar von Boehn wrote:
| | Thanks for the excellent post! |
Excellent, huh? And also, completely wrong.The IEEE 754 floating point is a standard. The 68k FPU is a standard. Support it or STFU. It's that simple. If you can't support it, because a full-fledged 68k FPU don't fit into the Vampire's FPGA, or you simply can't implement it for $whatever reason, just admit it, and move on, but these fairy tales with the conclusion of "the FPU is bad, m'kay?" are so damn unnecessary it hurts. Especially when most people are just daydreaming up random reasons, because they clearly don't even have the slightest clue how it works, and what it is used for. *sigh* (Note: the poster wrote a 68k/ColdFire codegenerator including '881/2 and CFv4 FPU support to an open source compiler, running on Amiga and Atari. The poster also wrote various 68k demoscene demos using the FPU.)
| |
| | 
Daniel Sevo
Posts 299
08 May 2017 16:28
| @Gunnar
From a user perspective, a lot has already been said about why a compatible FPU would make sense. If there wont be one then, well..Some will miss it some wont...
But Gunnar, if you only like to read about post that say we don't need FPU , I'm wondering if you haven't made up your mind already? ;-)Don't get me wrong, we are all grateful for the *amazing* work of your team to have gotten us this far with hardware we never thought would see the light of day. And it will surely enable a lot of nice new software coming to Amiga in the future even without FPU in current gen Vampire - and a lot of ppl will be perfectly happy.. From that perspective, your priorities might be just right. (Was it Star Trek? "The needs of the many outweigh the needs of the few") That being said, I believe it would be fair to all those who pre- ordered to give a conclusive answer whether the current gen Vamp will be a 68EC080 or a full 68080, especially considering that it was implied that FPU was "coming soon" at previous occasions.
For some users, an EC *simply will not* replace their 060 no matter what opinion ppl might have about the usefulness or quantity of Amiga FPU software. ;-)
| |
| |
Gunnar von Boehn
(Apollo Team Member)
Posts 6222
08 May 2017 16:57
| Chain Q wrote:
|
The poster also wrote various 68k demoscene demos using the FPU.)
|
And I design CPU and SIMD units for a living.
I wrote Vector/FPU code for SuperComputers in the TOP10 list.
I wrote Vectorlibs which are used by IBM Power Customers worldwide.
* Trust me the 68881/2 is not a good FPU.
* Trust me the 68060 FPU has a lot room for improving.
* Trust we can be do this much better and 68K will really rock again.
What I speak about has nothing to do with being incompatible!
You misread me when you think I said this.
What I said is that it would help us if people are interesting in doing something "great" again.
So if you like to help do something really awesome on AMIGA please step forward!
| |
| | 
Olaf Schoenweiss
Posts 690
08 May 2017 16:59
| I do not understand what discussion is about gunnar already wrote that FPU is on to-do list but not top priority except someone offers a use case that persuades them to change priority. At the moment SAGA and 16bit sound are more important and I guess most users will agree
| |
| | 
Thierry Atheist
Posts 644
08 May 2017 17:14
| Gunnar von Boehn wrote:
| | What I said is that it would help us if people are interesting in doing something "great" again. |
THIS, THIS.... it has to be THIS!!!!!
| |
| | 
Aksel Andersen
Posts 120
08 May 2017 17:20
| What would the requirements be for such a test suite?
Surely there must be some existing suites for this made in the past?
| |
| |
Chain Q
Posts 19
08 May 2017 17:22
| Gunnar von Boehn wrote:
|
Trust me the 68881/2 is not a good FPU.
Trust me the 68060 FPU has a lot room for improving.
Trust me this can be done much better and 68K can really rock.
|
The 6888x was a decent design for its time (if you ever looked at the direct competitor x87 line... well, thanks but no thanks), although in retrospect it's quite overengineered. Also, they clearly shoot for features instead of speed. The '060 was clearly let down by its FPU, compared to the competition and the rest of its design, but it's still a feature which was invaluable along the years and it still runs pretty okay. Also it's 20+ years old at this point. So indeed, it's about time that someone does it BETTER. What I speak about has nothing to do with being incompatible. You misread me when you think I said this.
|
Then just do it, if you know what and how to do. And don't say "excellent" to posts, which are clearly wrong. :)
| |
| |
Gunnar von Boehn
(Apollo Team Member)
Posts 6222
08 May 2017 17:23
| Aksel Andersen wrote:
|
What would the requirements be for such a test suite?
Surely there must be some existing suites for this made in the past?
|
No no we don't need a testsuite we have done millions of FPU test already.We talk here about a usecase..
Something to show the superiority...
Like riva did for AMMX
| |
| |
Aksel Andersen
Posts 120
08 May 2017 17:26
| Gunnar von Boehn wrote:
|
Aksel Andersen wrote:
|
What would the requirements be for such a test suite?
Surely there must be some existing suites for this made in the past?
|
No no we don't need a testsuite we have done millions of FPU test already.
We talk here about a usecase..
Something to show the superiority...
Like riva did for AMMX
|
Ah... Ok. What about Demos, demos and demos? ;) A perfect showcase for the vampire imo..
| |
| |
Gunnar von Boehn
(Apollo Team Member)
Posts 6222
08 May 2017 17:30
| Aksel Andersen wrote:
|
What about Demos, demos and demos? ;)
A perfect showcase for the vampire imo..
|
Yes if its a demo which was never there before.
| |
| |
Aksel Andersen
Posts 120
08 May 2017 17:32
| Gunnar von Boehn wrote:
|
Aksel Andersen wrote:
|
What about Demos, demos and demos? ;)
A perfect showcase for the vampire imo..
|
Yes if its a demo which was never there before.
|
Now you are just being difficult.
| |
| |
Gunnar von Boehn
(Apollo Team Member)
Posts 6222
08 May 2017 17:34
| Aksel Andersen wrote:
|
Now you are just being difficult.
|
Do you really not understand what I mean?
| |
| | 
Lord Aga
(Apollo Team Member)
Posts 119
08 May 2017 17:35
| Chain Q wrote:
|
Support it or STFU.
|
Some basic decency wouldn't hurt in this discussion.
| |
| |
Olaf Schoenweiss
Posts 690
08 May 2017 17:36
| how about raytracer?
dbwrender is part of aros contributions and cli basedyou could compare speed of normal version and vampire version
| |
| |
Aksel Andersen
Posts 120
08 May 2017 17:38
| Gunnar von Boehn wrote:
|
Aksel Andersen wrote:
|
Now you are just being difficult.
|
Do you really not understand what I mean?
|
Well I am confused. Many of us are about this. Don't this thread show this? But let's not go into a flame war here guys. :)
| |
|
Information about the Apollo CPU and FPU.