BLITTER RE-CREATION THOUGHTS

[Cyprian]

Someone from Poland that it is also working on some of those ATX ST motherboards. I understand he tested it both on original hardware and an ATX motherboard.

do you remember the nickname?
I asked x_angel (ATX ST author) but he said it wasn't him

[ijor]

do you remember the nickname? I asked x_angel (ATX ST author) but he said it wasn't him

It is not x_angel. Well, at least, I think it is not him. I'm not sure about the nickname, but even if I knew it, I don't think it would be correct to post here personal details about a private email exchange. Honestly, I don't even know if his work is public or not.

[exxos]

@Cyprian Going back to these tests viewtopic.php?p=51449#p51449

These 2 images were original and our blitter. What exactly is "out" on our core ? Looks like some timing is slightly to fast to slightly to slow causing some odd screw in the test ?

Original in STFM - program 1.jpeg (91.91 KiB) Viewed 1895 times

Sparkalaphobia in STFM - program 1.jpeg (83.54 KiB) Viewed 1895 times

[Cyprian]

@exxos I see that the second the BLiTTER pass starts 4 cycles earlier, but it could be a result of miscounting the bus cycles in the first pass.
Sparkalaphobia on the right side, upper arrow - the first the BLiTTER pass , the bottom arrow - the second pass:

20221029_sparklaphobia.png (130.9 KiB) Viewed 1861 times

May I ask you to do one more test? It provides some additional details like, when the BLiTTER touches the bus ( black/blue/green strips) and how it interacts with a long the CPU instructions like "DIWS.W" (white spaces between each the BLiTTER pass)

BLiT_H6a4h.png (15.64 KiB) Viewed 1861 times

[exxos]

May I ask you to do one more test? It provides some additional details like, when the BLiTTER touches the bus ( black/blue/green strips) and how it interacts with a long the CPU instructions like "DIWS.W" (white spaces between each the BLiTTER pass)

They were done in this thread viewtopic.php?p=51449#p51449

So if our glitter is taking the bus too fast, it doesn't make sense why our glitter is slower than the original.

[Icky]

We, @ijor and I, have a bit of an update on the Phoenix Sparkalaphobia aka the Hardware FPGA Blitter with Ijor's Blitter core. We have been working in the background on the core and testing. Ijor has been instrumental in helping get the timings and plumbing in place. I have learnt a great deal more about FPGAs through this experience.

We have performed extensive testing and are now confident to announce we have a working Blitter core on real hardware that matches the original Blitter.



Now for the details

Our tests have been on the following hardware: an original STFM and Phoenix H5 motherboard and we have tested on TOS102 and TOS104.

The software we used:

• The Blitter pattern programs from @Cyprian
  - viewtopic.php?p=24716#p24716
  - viewtopic.php?f=80&t=79&start=260#p28439
  - viewtopic.php?f=80&t=79&start=300#p51391
• Blitter BGACK delay test program
• GEMBench 6.31

Using the above software we got matches of all the patterns from Cyprian's programs both original and FPGA version matched exactly. Ijor's delay test program gave the same results as an original Blitter on an STFM and H5.

However the GEMBench tests spawned tangential discussions with @exxos and Ijor on GEMBench timing (rounding errors). We were seeing a 0.005 difference in running on TOS102 which is negligible but GEMBench rounding was causing 99% score. That said GEMBench scores on TOS104 are 100% between original and FPGA.

Below are some example results. You may also notice that there are timing differences between TOS102 and TOS104 - Ijor mentioned in the thread Blitter on H5 Test the Blitter code section is similar in TOS104 - TOS206 but different in TOS102. There is also more discussion in that thread about the GEMBench timings.

• Original STFM with original Blitter on TOS102 the Blitter timing is 3.570
• Original STFM with Sparkalaphobia Blitter on TOS102 the Blitter timing is 3.575
• Original STFM with original Blitter on TOS104 the Blitter timing is 3.590
• Original STFM with Sparkalaphobia Blitter on TOS104 the Blitter timing is 3.590

What next?

Well with the shortage of FPGAs currently there won't be many Sparkalaphobia boards being made other than a few DEV ones for continued testing etc. However we now have a solution for a future Blitter chip replacement for original and Phoenix motherboards.

[exxos]

Congratulations @Icky and @ijor

Give yourselves a pat on the back. We don't have one of those icons, so this will have to do..

[EvilFranky]

Great news!

What's planned for added features?

[exxos]

What's planned for added features?

32MHz

[EvilFranky]

What's planned for added features?

32MHz