Overview Features Coding ApolloOS Performance Forum Downloads Products Order Contact

Welcome to the Apollo Forum

This forum is for people interested in the APOLLO CPU.
Please read the forum usage manual.
Please visit our Apollo-Discord Server for support.



All TopicsNewsPerformanceGamesDemosApolloVampireAROSWorkbenchATARIReleases
Information about the Apollo CPU and FPU.

Status of the FPUpage  1 2 3 4 5 6 7 8 9 10 11 12 13 

Gunnar von Boehn
(Apollo Team Member)
Posts 6207
19 May 2017 10:14


There was the question regarding the status of the FPU.
 
Unfortunately the thread became unreadable with partly untechnical comments. Therefore we start here fresh with proper information.

Please only comment on technical level.
 
 
The status is:
 
APOLLO has a fully working FPU, supporting SINGLE, DOUBLE and EXTENDED precision calculations.
 
While the APOLLO FPU was successfully used in some customer project, for usage in 68K AMIGA OS some more testing was found required.
The Apollo-team did ask for AMIGA coders community to help testing 12month ago - in this time no coders applied for this.
So it became clear that the FPU priority in AMIGA world is not very high.
This makes sense as the majority of AMIGA software does not use FPU at all.
 
While we have a FPU now, we also know that the FPGA technology will allow to reach floating point performance 1000 times faster than existing Motorola 68060 CPUs we need to look for a solution which we can use long time and which allow benefit from these FPGA technologies. Therefore its advisable to verify and review the design to make sure that its future-proof.
 
If 68k coders are interesting in supporting this then please contact us. Ideally via IRC or via email EXTERNAL LINK   
Please note that coders willing to support need to be willing to put in significant time, and that all development will be done with 68k ASM.
 
 
Many thanks


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
19 Jun 2017 06:16


We deleted some post based on misunderstanding our original post.

We absolutely do NOT need help to test the FPU.
We are perfectly able to do this on our own.

 
We look for someone interested in helping to write a new high end FPU application - which shows that FPU performance can be pushed to new limits.
 
The core calculation routines in this application need to be written in ASM.

Please reply only if you are willing to write such app.


Matthew Langtry

Posts 199
19 Jun 2017 19:42


would a ray tracing program be able show off fpu performance?


Olivier Landemarre

Posts 147
19 Jun 2017 22:47


Gunnar von Boehn wrote:

We deleted some post based on misunderstanding our original post.
 
  We absolutely do NOT need help to test the FPU.
  We are perfectly able to do this on our own.
 
 
  We look for someone interested in helping to write a new high end FPU application - which shows that FPU performance can be pushed to new limits.
 
  The core calculation routines in this application need to be written in ASM.
 
  Please reply only if you are willing to write such app.

I have probably the application to do this, this is a small raytracer, I not write it myself but do several port of it. It is not very fast raytracer but it is quite easy to speed up I think and this software can run independently of any system (run already on 68K, x86, PPC). Most mathematical subroutine are in separate file as simple function such like this:
VOID v3s_trnsf( V3S *v, V3S *result, MAT4 m )
{
  V3S  hv;

  hv.x = v->x * m[0][0] + v->y * m[1][0] + v->z * m[2][0] + m[3][0];
  hv.y = v->x * m[0][1] + v->y * m[1][1] + v->z * m[2][1] + m[3][1];
  hv.z = v->x * m[0][2] + v->y * m[1][2] + v->z * m[2][2] + m[3][2];

  *result = hv;
}

include ASM inside can probably highly speed up the software but there is 2 issue:
- I'm not able to code in ASM
- I have no appolo card, I hope buy a standalone when available.

I can propose to prepare the test case, optimize a bit the actual C code and if someone is interested send him the source code for adjustment.

EXTERNAL LINK 
Olivier



Cyber Gorf
(Needs Verification)
Posts 39/ 1
26 Jun 2017 13:48


Dear Apollo Team!
 
I often hear from your side, one should buy the Vampire for the features it offers now and not for things, that may come (or not come). That is a very good advice.
 
But I do not understand, why you still advertise features, that are not yet built in.
You are deliberately misleading new customers by doing so:
 
On this very page you make following statements:
 
  *Overview*
 
"Apollo Core 68080 is the natural and modern evolution of latest 68000 processors. It's 100% code compatible, ...."
 
  *Features*
 
Matrix: "Integrated FPU" and "Pipelined FPU" are both green (should be yellow at best)
Apollo Core 68080 advantages:
"Fully pipelined, double/extended FPU"
 
  *Products*
 
Vampire 600 V2 and Vampire 500 V2+ contain the "Apollo 68080 CPU"
 
 
Every visitor of this page and every potential new customer has therefore to assume, that a 100% compatible, fully pipelined and integrated  FPU is part of every Vampire Card.
There is no hint, that makes you think otherwise.
 
It is not the first time, someone mentions this here. It would take unter 5 min. to change this.
You can not expect a new customer to search though thousands of entries in a forum, to make sure that a clearly advertised feature is really there.
 
I just do not understand, why you act this way?
The Vampire is a great product as it is - you do not have to make false statements about it.


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
26 Jun 2017 14:21


Cyber Gorf wrote:

But I do not understand, why you still advertise features, that are not yet built in.
 
You are deliberately misleading new customers by doing so:

 
You bark to the wrong tree.
You mis-understand the difference between a CPU manufacturer and a System manufacturer.
 

Please understand that we here:

We produce the APOLLO CORE and the SAGA chipset.
We here do _NOT_ produce the Vampire2.

 

Its a fact that: 
APOLLO CORE has a FPU. 
APOLLO CORE supports also 64bit and can address GIGABYTES of MEMORY.
 

Cyber Gorf wrote:

The Vampire is a great product as it is - you do not have to make false statements about it.

 
Actually we make no statements about the Vampire at all.
Frankly, you are confused on which website you are. 
 

Cyber Gorf wrote:

You are deliberately misleading new customer

 
Actually you are not our customer.
 
If you want to become our customer than please build an FPGA system and please license our core. If you FPGA is big enough you can also license the FPU.

Looking forward to your product.


Vojin Vidanovic

Posts 770
26 Jun 2017 17:31


Mr Thelier wishes to code an softFPU

Can he be aided?
EXTERNAL LINK


Dusko Kovacic

Posts 21
26 Jun 2017 21:00


Cyber Gorf wrote:

  But I do not understand, why you still advertise features, that are not yet built in.
  You are deliberately misleading new customers by doing so:
   

To understand what apollo core really is, it takes more than just a few clicks.




Jari Eskelinen

Posts 23
27 Jun 2017 06:12


Vojin Vidanovic wrote:

Mr Thelier wishes to code an softFPU
 
  Can he be aided?
  EXTERNAL LINK 

As a personal asm learning project I have started working with soft FPU. Suprisingly it is already able to run Mandelbrot generator (though under UAE only, some problems with Vampire, need to discuss with the team).

Perhaps Mr Thelier would like to collaborate... I do not have account on amigans.net, could you ask him if he would care to come over here?


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
27 Jun 2017 06:21


Jari Eskelinen wrote:

As a personal asm learning project I have started working with soft FPU. Suprisingly it is already able to run Mandelbrot generator (though under UAE only, some problems with Vampire, need to discuss with the team).
 
Perhaps Mr Thelier would like to collaborate... I do not have account on amigans.net, could you ask him if he would care to come over here?

Excellent initiative Jari!
We congratulate you.

We will support your fully.
Please ping us in IRC.



Wawa T

Posts 695
27 Jun 2017 08:50


could you ask him if he would care to come over here?

alain is posting here.


Thellier Alain

Posts 141
28 Jun 2017 09:52


Hello

To be clear: I am not currently coding a Soft Fpu
I was only investigating about "How coding a Soft FPU for Vampire could be done ?"
As it is not a domain (traps & fpu) that I know

Also I dont have time for coding much nor specially the envy to code anymore especially such a boring code...

Once said I am thinking that for someone that knows the fpu & traps domain recompiling shoebill/SoftFloat for Vampire is perhaps not so complicated...
Of course the resulting SoftFpu will be slow but will allow to run NOW legacy productivity software made for Amiga4040 & co ...

Alain Thellier


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
28 Jun 2017 22:40


Here is Jari's SoftFPU running
There is of course a lot more to do, but his SoftFPU really works and some programs mandelbrot, fractal etc run 100% correctly already.
 



David Wright

Posts 373
28 Jun 2017 22:46


Nice to see progres in all the areas. Makes this project more interesting as I explore the improved Amiga.
Ist es alles gut.


Vojin Vidanovic

Posts 770
28 Jun 2017 23:11


2.08 MFLOPS is better then zero.

Could it be a public download until 8881 compatibility is introduced?


Gregthe Canuck

Posts 274
29 Jun 2017 00:37


Very cool update from Jari.

What MFLOPS values would a stock A500 and A1200 present?


Steve Ferrell

Posts 424
29 Jun 2017 02:41


gregthe canuck wrote:

Very cool update from Jari.
 
  What MFLOPS values would a stock A500 and A1200 present?

I have an M1230XA accelerator running a 68030/68882 at 50Mhz and I get 1.34 MFlops.



Gregthe Canuck

Posts 274
29 Jun 2017 04:29


Hi Steve - thanks for the info.
 
Not too bad for software emulation then - faster than a 68882.

I am curious if anyone knows... Where would all the overhead be in this type of emulation? The trap and call to the emulation code, or the emulation routine itself?

 
 
 
 


Roman S.

Posts 149
29 Jun 2017 07:39


I bet the routine :)

I hope it will use the HyperThreading - AFAIK the software optimized by hand often utilizes the fact that real FPU can execute instructions in parallel to the CPU.

Anyway - good news. In reality we probably still won't be able to run the 68060 demos on Vampire, but at least all productivity software will work.




Simo Koivukoski
(Apollo Team Member)
Posts 601
29 Jun 2017 08:06


A very early Software Floating Point Unit bring-up test:

EXTERNAL LINK

posts 254page  1 2 3 4 5 6 7 8 9 10 11 12 13