Which is the fastest software C2P 1x1 routine

[matthey]

@JimDrew
Back in the day, did you ever ask C= for chunky modes or why they didn't add them? Or were they oblivious to the success and potential of your Mac emulation?

Of course we can do it the right way today with chunky and an enhanced Amiga chipset in the fpga. Wasn't MikeJ going to upgrade the fpga for the fpgaArcade plus anyway? If he would put a descent sized Altera fpga on there, we would be set .

[JimDrew]

Just replaced the original upload I did with a newer version with a slightly faster PCx C2P routine (.5ms or so faster). It was the last version I had in all of my code, and it matches the code in the video drivers for the Mac and PC emulations, so that is what we released officially for EMPLANT, FUSION, and PCx.

Originally Posted by matthey

@JimDrew
Back in the day, did you ever ask C= for chunky modes or why they didn't add them? Or were they oblivious to the success and potential of your Mac emulation?

Of course we can do it the right way today with chunky and an enhanced Amiga chipset in the fpga. Wasn't MikeJ going to upgrade the fpga for the fpgaArcade plus anyway? If he would put a descent sized Altera fpga on there, we would be set .

I think the bitplanes is a hold over from the original Amiga design, based around the old school way of doing things. I think video chips are easier to design in single planes because of the memory usage?? I don't know for sure, but there was no desire on CBM's part to add modes other than proprietary things like HAM and HAM8. I guess that would be a question to ask Dave Haynie or Bill Heard.

I gave Commodore my C2P routines to be used in some project that they were working on. That was kind of an exchange for them giving me permission to use the foundation of the kernal's multitasking code that had to be modified to allow the Mac to run peacefully with the Amiga. I had a meeting with John Dilulo (then CEO I believe) about the Mac emulation and of course they were thrilled that video production houses were buying Amigas to run the Mac's Avid video suite via my emulation, but CBM was (rightfully) afraid of Apple's reaction if Commodore were to try to pursue this themselves as an official CBM product. I ran into the same issue when I met with the CEO of Sony while in Japan - Sony wanted a tiny PC capable of running my Mac emulation, but Apple was making noises so a bundle never occurred. When I actually met with John Sculley and the Apple legal team in Cupertino to discuss a bundle deal, they really didn't care what I was doing as long as it didn't some how make the Mac look slow or fail to run apps. It was actually the opposite, where we could run more apps (and faster) than any single Mac could. That too posed a problem, and Apple decided it was in their best interest to not bring attention to my product and they surprisingly admitted that they had no legal means to stop it because the emulation was clearly not using any Apple code or violating any copyrights. It was a friendly departure at the end of the day... which was a blessing. I was fearing the worst!

I have not looked at the synth lately but I think there is still plenty of space in Mike's FPGA Arcade Replay to add things like C2P conversion.

[matthey]

Originally Posted by JimDrew

I think the bitplanes is a hold over from the original Amiga design, based around the old school way of doing things. I think video chips are easier to design in single planes because of the memory usage?? I don't know for sure, but there was no desire on CBM's part to add modes other than proprietary things like HAM and HAM8. I guess that would be a question to ask Dave Haynie or Bill Heard.

Yes, bit planes are efficient for 16 colors and less. It made sense to start with them in the dinosaur days of computing.

Originally Posted by JimDrew

I gave Commodore my C2P routines to be used in some project that they were working on. That was kind of an exchange for them giving me permission to use the foundation of the kernal's multitasking code that had to be modified to allow the Mac to run peacefully with the Amiga. I had a meeting with John Dilulo (then CEO I believe) about the Mac emulation and of course they were thrilled that video production houses were buying Amigas to run the Mac's Avid video suite via my emulation, but CBM was (rightfully) afraid of Apple's reaction if Commodore were to try to pursue this themselves as an official CBM product. I ran into the same issue when I met with the CEO of Sony while in Japan - Sony wanted a tiny PC capable of running my Mac emulation, but Apple was making noises so a bundle never occurred. When I actually met with John Sculley and the Apple legal team in Cupertino to discuss a bundle deal, they really didn't care what I was doing as long as it didn't some how make the Mac look slow or fail to run apps. It was actually the opposite, where we could run more apps (and faster) than any single Mac could. That too posed a problem, and Apple decided it was in their best interest to not bring attention to my product and they surprisingly admitted that they had no legal means to stop it because the emulation was clearly not using any Apple code or violating any copyrights. It was a friendly departure at the end of the day... which was a blessing. I was fearing the worst!

Thanks for the great history. It really makes it sound like you were dealing with the big wigs after what Apple has become today. It's definitely scarey when a whole legal department can sue your small company just to annoy and scare you into doing what they want. I'm surprised they would admit they had no legal grounds to stop you instead of bluffing. That must have been before Apple's change of heart put an end to the Mac clones.

Originally Posted by JimDrew

I have not looked at the synth lately but I think there is still plenty of space in Mike's FPGA Arcade Replay to add things like C2P conversion.

I'm sure there is room for C2P although there is RTG chunky already. There is enough fpga space for the retro 8 bit and some simpler 16 bit computer emulation but space is tight for more complex systems and processors that need more performance. Emulation of Genesis/MegaDrive, NeoGeo, X68000, Nintendo, SNES, etc is what would really put the fpga Arcade on the map. Then there are Amiga and AtariST enthusiasts that want new faster and more advanced computers. I think Mike's concept is good and I like the standalone fpga board but I don't think the current fpga size has hit the sweet spot where the possibilities really feel unlimited.

[JimDrew]

Interesting. There is quite a bit of space left with the Amiga core. There are many different emulations planned or already being ported to the replay board. Mike has an Atari ST emulator, and I have already seen SNES, Nintendo, and other cores in the works. It's always good to have a larger capacity device but Mike designed this thing years ago when the latest and greatest (affordable) FPGA was being used. It's too bad that they don't make FPGAs like they do micros - 100% pin compatible with more code/core space!

Yeah, Apple had a change of heart between when I wrote the Mac emulation for the Amiga and the PC version. I am confident that they ripped my code apart looking for anything they could possibly claim. I am sure they learned a few things about the Amiga too.

[Gunnar]

Originally Posted by matthey

Yes, bit planes are efficient for 16 colors and less.

Also with more bits planar can be nice.
Imagine AGA would have had a 9 plane EHB mode.
It would not have costed any more color registers.
And would have costed only 12% more DMA bandwidth to double the colors on the screen.

Originally Posted by matthey

I'm sure there is room for C2P although there is RTG chunky already.

The C2P instruction which we added yesterday.
Did only cost about 60 LE.
This is nearly free.

[matthey]

Originally Posted by JimDrew

Interesting. There is quite a bit of space left with the Amiga core. There are many different emulations planned or already being ported to the replay board. Mike has an Atari ST emulator, and I have already seen SNES, Nintendo, and other cores in the works. It's always good to have a larger capacity device but Mike designed this thing years ago when the latest and greatest (affordable) FPGA was being used. It's too bad that they don't make FPGAs like they do micros - 100% pin compatible with more code/core space!

The Vampire 600 accelerator only has a Cyclone II which is smaller than the Xilinx fpga in the fpga Arcade. The Phoenix 68k core is really crammed in and bottled up in there. Just 2x more fpga space would allow superscalar execution, more optimal pipelining, reasonable sized caches, etc. without breaking the bank. An fpga Arcade sized fgpa may allow 68030 performance, a Vampire 600 sized fpga may allow 68040 performance and an fpga that is 50% larger yet should easily give 68060+ performance. I think it's worth another $20 in fpga cost to have 68060 performance.

I thought it would be nice if fpgas could be bought on a DIMM like module (consistent for fpgas with the same number of pins). When prices drop, you just pop out the old and pop in the new. It's the same logic as a CPU socket but it makes more sense with fpga technology that is changing faster than CPU technology today. Do you think we could make money producing and marketing them?

[Gunnar]

Originally Posted by matthey

The Vampire 600 accelerator only has a Cyclone II and it is bigger than the Xilinx fpga in the fpga Arcade.

I always forget what size the FPGA ARCADE FPGA has.
How many LE does it have again?

All I remember is that the MIST FPGA has a very good size.
It should be big enough for the chipset and the CPU.

[matthey]

Originally Posted by Gunnar

I always forget what size the FPGA ARCADE FPGA has.
How many LE does it have again?

The fpga Arcade uses a Xilinx XC3S1600E in BGA320 pin package with 1.6M gates (33192 Equivalent
Logic Cells, 14,752 slices). Maybe it is big enough.

Edit: Added manufacturer claims for ELC and slices although I've heard that they are over stated compared to Altera fpgas. In any case, this is not a cheap fgpa.

[Gunnar]

Originally Posted by matthey

The fpga Arcade uses a Xilinx XC3S1600E in BGA320 pin package with 1.6M gates. It's almost big enough.

The 1.6 million gates is "bla-bla" of the marketing department.
The really important info is the number of slices or LE equivalents.

Can someone with a replay say how many slices / LE are free?
I mean how many are available after loading the chipset.

[Thorham]

Originally Posted by matthey

33192 Equivalent Logic Cells, 14,752 slices

What an odd numbers. How many usable transistors does that amount to? I say strange, because I have no idea how this works, and 68000s are made of 68000 transistors, so those numbers seem on the low side

[matthey]

Originally Posted by Thorham

What an odd numbers. How many usable transistors does that amount to? I say strange, because I have no idea how this works, and 68000s are made of 68000 transistors, so those numbers seem on the low side

I believe each logic element would take several transistors. There are also other fpga resources using significant transistors, especially memory. I would guess that the fpga gate count would be the closest equivalent to chip transistors or maybe it's just "bla-bla" marketing crap .

[JimDrew]

Originally Posted by matthey

The fpga Arcade uses a Xilinx XC3S1600E in BGA320 pin package with 1.6M gates (33192 Equivalent
Logic Cells, 14,752 slices). It's almost big enough.

Edit: Added manufacturer claims for ELC and slices although I've heard that they are over stated compared to Altera fpgas. In any case, this is not a cheap fgpa.

Funny, the Xilinx people say the same about the Altera parts.

Originally Posted by matthey

The Vampire 600 accelerator only has a Cyclone II and it is bigger than the Xilinx fpga in the fpga Arcade.

It is? The BOM for the Vampire 600 shows it using a 138 pin Altera part EP2C8Q208C8, which has 8,256 LE's. That's a far cry from the FPGA Arcade's 320 pin Spartan with 33,192 LE's.

I just saw some sysinfo screens and watched a video of the Vampire 600. It's way slower than FPGA Arcade's full AGA core. Did the FPGA change from the BOM? From what Mike and Wolfgang have stated, there is enough space in the Spartan for several Amiga cores.

[Gunnar]

Originally Posted by JimDrew

The BOM for the Vampire 600 shows it using a 138 pin Altera part EP2C8Q208C8, which has 8,256 LE's. That's a far cry from the FPGA Arcade's 320 pin Spartan with 33,192 LE's.

Yes the Cyclone 2 on the Vampire is smaller.
If I see the info correct the Vmapire has 8K flipflops and the Replay has 28K flipflops.

This is good news.
As this means the Replay could upgrade the CPU core.
It seems we remembered some older/wrong info about the replay.

Originally Posted by JimDrew

I just saw some sysinfo screens and watched a video of the Vampire 600. It's way slower than FPGA Arcade's full AGA core.

The video shows the old TG68 Softcore.
The new Phoenix CPU core gives a good speed boost.

[matthey]

Originally Posted by Gunnar

Yes the Cyclone 2 on the Vampire is smaller.
If I see the info correct the Vmapire has 8K flipflops and the Replay has 28K flipflops.

This is good news.
As this means the Replay could upgrade the CPU core.
It seems we remembered some older/wrong info about the replay.

My mistake. I stand corrected. The fpgaArcade fpga is bigger and faster than the Vampire's Cyclone II and probably a little bigger than the Mist's Cyclone III. I knew something was up when I saw the price. My understanding is still that the Xilinx LE claim is high compared to Altera so the Xilinx XC3S1600E may be more like 28k to 30k LE in Altera fpga LEs? Xilinx fpgas also generally have less internal clock timing/PLLs although the fpga Arcade board may have what is needed externally? In any case, there should be adequate room for the Apollo core and caches with the Amiga chipset. I know it would be a pain to switch "tooling" for Xilinx fpgas but it's probably worth evaluating. The fpgaArcade is a high quality board with good Amiga support and a good marketing plan.

[JimDrew]

Where can I get info on the Apollo core? Is it open source? There would be no need to switch tooling as the Xilinx ISE can use either Verilog or VHDL, and its not that difficult to convert one to the other if you absolutely needed to.

I am the North American distributor for the FPGA Arcade Replay.

[matthey]

Originally Posted by JimDrew

Where can I get info on the Apollo core? Is it open source? There would be no need to switch tooling as the Xilinx ISE can use either Verilog or VHDL, and its not that difficult to convert one to the other if you absolutely needed to.

The Apollo web site is here although there may be some dust:

http://www.apollo-core.com/

Apollo is a superscalar, highly pipelined mostly 68020 compatible CPU. Phoenix is scalar with a shortened pipeline and currently lacks 68020 compatibility because of size limitations in the Cyclone II. Phoenix is complete and being debugged in Majsta's Vampire 600. Apollo has only been tested in simulation but bug fixes of Phoenix should improve Apollo as many of the units are the same.

The Apollo and Phoenix cores are written in VHDL. It wouldn't be rocket science to convert but would probably require substantial tweaking and optimizing. The less familiar Xilinx software may take some time to get used to also. Gunnar could explain better I'm sure.

There has been discussion of some kind of fairly open license but the cores are not currently open source. They aren't quite ready for distribution yet.

[JimDrew]

Had a nice hour long chat with Gunnar about this core today. Thanks for the link!

[tolkien]

Wooow...Gunnar and Jim Drew speaking together. Only good things can happends from here.

[mc6809e]

Originally Posted by matthey

Yes, bit planes are efficient for 16 colors and less. It made sense to start with them in the dinosaur days of computing.

Actually bitplanes are more efficient for every non power of 2 colors display.

A linear increase in memory usage allows for an exponential increase in color. For example, going from 16 to 32 colors takes 25% more memory but doubles the number of colors. And an 8 color image wastes memory access slots if the display hardware requires 4 bits per pixel. The bitplanes scheme can also help area fill operations since a single plane can be area filled and this single plane can then quickly be written (sometimes inverted) into display planes to produce a poly of a particular color.

At the time the Amiga was designed, most displays were actually chunky in the sense that they might use 2 adjacent bits or 4 adjacent bits per pixel for color selection and memory was still very expensive per bit. The initial design of the Amiga actually called for something like 128k of chipram.

Had the Amiga not been delayed by corporate politics, the machine would have been released earlier and the bitplane idea would have been seen as a great advantage.

But as we all know memory densities continued to increase rapidly.

One of the worst thing about bitplanes is the difficulty of taking advantage of page mode in DRAM. While it takes 280ns for a chip ram memory access on a standard Amiga, a page mode access on the same ram might take 150ns or even 120ns, but bitplanes in separate areas of memory generally mean the page changes on every access so page mode can't be used. The AGA chipset does take some advantage of this but at the expense of silicon used for larger buffers.

[JimDrew]

Creative use of bitplanes allows you efficient updates of background data quickly too, like scrolling.