Weeknight 'Virtual' Dev session #9, Thursday 14th January 2021, 7pm - 10pm (GMT)

[Arcadian]

As the new games and demo coding sections of the forum are currently very active, we are now hosting weeknight "Dev sessions" on Zoom, that are primarily focused on the development of new games and graphics/music demos for Acorn computers as well as development tools, emulators and debuggers.

They take place the week in-between the 2 x Tuesday night 'general' sessions on the Thursday evening, between 7pm and 10pm.

The sessions are intended for developers to share progress updates and request/offer help and feedback. We hope that these sessions will help motivate and encourage developers to see projects through to completion.

Just to be clear - these sessions are open to everyone: programmers, graphics artists and musicians, even if you've just got an interest/appreciation of - or ideas for - new games and demos, please come along for moral support - there may be opportunities to beta-test new versions of upcoming projects . There really isn't a limit to the number of places available - well, the maximum we can host in a Zoom meeting is 100 but it's unlikely we'll ever get anywhere near that figure!

If there are any issues you'd like help with - or would like to demonstrate something - please leave details in the thread and we'll add them to the lists of items to be covered.

If you are new to Zoom, forum member JudgeBeeb has written a handy guide to getting set up and logged in.

I've already sent invites out to most of the Dev Night regulars, but should anybody else want to attend - or if you haven't received the Zoom logins - please post in this thread and I will send them to you.

Attended:

1. Arcadian (Dave)
2. kieranhj (Kieran)
3. RobC (Rob)
4. sPhilMainwaring (Phil)
5. MarkDRyan (Mark)
6. tricky (Richard)
7. dudleysoft71 (James)
8. BigEd (Ed)
9. oss003 (Kees)
10. DaveFootitt (Dave)
11. chrisb (Chris)
12. IanS (Ian)
13. billcarr2005
14. dominic (Dominic B)
15. Pernod (Nigel)
16. chrisn (Chris)
17. dp11 (Dominic)
18. JIM (James) with son (Lewis)
19. ChrisSawer (Chris)
20. EdwardianDuck (Jeremy)
21. SarahWalker (Sarah)
22. flibble (Peter)
23. Tom Williamson
24. damiensturdy (Damien)
25. Leonardo Vettori
26. Francesco Fiorentini - http://retromagazine.net

[kieranhj]

After a few months away from the (retro) keyboard, my new year's resolution is to start on some new projects and do some more Masterclasses.

To that end, I've prepared An Intro to CRTC Registers (part 1) for Thursday evening. In this session we can cover:

• A high-level overview of what the CRTC actually does
• CRTC registers and how to program them
• Custom modes / changing screen dimensions to save RAM
• Hardware scrolling and CRTC address wrap around

If all goes well I'm planning to do a follow up session to cover some more advanced techniques, namely Vertical Rupture and Smooth Vertical Scrolling, but one step at a time!

Look forward to seeing you all there.

[ChrisB]

As requested by Arcadian I can come along (but only after 9pm) to give an update/brief demo of my Untitled Dungeon Game.

[Arcadian]

Ah that's fantastic thanks Chris.

So as for the schedule it's still quite open, but let's go with the following for now:

7:00pm - 7.30pm: General Chat

7.30pm - 8.00pm: BBC Micro co-processor graphics compression (Dudleysoft71 - James Watson)

8.00pm - 9.00pm: An Intro to CRTC Registers - Part 1 (kieranhj)

9.00pm - 9.15pm: New 3D Bomb Alley-inspired game (JIM - James Watson)

9.15pm - 9.45pm: Untitled Dungeon Game update (Chris B)

9.45pm - 10.00pm: New Rally X update (Tricky)

Would anybody else like to take a slot, either to demo their project(s) or to ask for help/advice on what you're currently working on?

[tricky]

I'll give a brief update on my new rally-x game.

[Arcadian]

Ok super I've put you in for 9.45 - 10.00pm - is that ok?

If anybody else would like a slot, we can make one available 7.45pm - 8.00 ...

[dudleysoft71]

I don't really need a slot, but I could give a quick demo of my ray tracing demo i did.

[BigEd]

I don't really need a slot, but I could give a quick demo of my ray tracing demo i did.

It might be interesting to talk about your VDU handler running on the host: that's 6502 code, it's small (presumably) and any saved cycles will help performance.

[dudleysoft71]

I can definitely do that, it's a fairly simple system, but it's probably possible to speed it up a bit more, it's been a while since I wrote the 6502 code so it might definitely benefit from a couple more eyes on it.

[markdryan]

It might be interesting to talk about your VDU handler running on the host: that's 6502 code, it's small (presumably) and any saved cycles will help performance.

It might also be interesting to show the code that installs the 6502 handler on the copro side. For some reason I'd assumed that this would be really complicated to do but it turns out to be pretty straight forward. It's just a matter of issuing a few sys calls from the copro side.

[hoglet]

I can definitely do that, it's a fairly simple system, but it's probably possible to speed it up a bit more, it's been a while since I wrote the 6502 code so it might definitely benefit from a couple more eyes on it.

Is your library published anywhere?

Ed and I took a quick look at it a few days ago (by running the host side in the ICE and single stepping throught it).

From memory, the inner loop was:

Code: Select all

.loop
LDA &FEE1    / 4
STA &XXXX,X  / 6
INX          / 2
DEY          / 2
BNE loop     / 3

There is one small optimization you could do. If you sent each block of data backwards, you could do away with one of the loop counters, which reduces the loop from 17 cycles to 15 cycles.

Dave

[dudleysoft71]

Yeah, I have thought about that, I've just updated my code to remove the INX, and to send the block backwards, my new code is actually 1 byte shorter than the previous code.

I haven't published it anywhere yet, I want to tidy it up a bit, it's still referencing specific stuff from BAGI, including some host library code which needs to be separated out into it's own ASM source file, and needs turning into a proper library with it's own makefile. Those are on my TODO list, but life keeps getting in the way of doing it.

[tricky]

I'm not sure what size these transfers are, but would it be worth unrolling the loop 2X and jumping into the middle if it is an odd number?
Obviously, it can be unrolled as many times as you like, but the jumping in gets a little more expensive.

[dudleysoft71]

I've just had a bit of a sort out of the code, I've made a few changes both host and parasite side, the practial upshot of this is that the screen transfer routine should be a lot faster, I've changed the following things:

• Changed blocks are sent in reverse order, this removes the INX call, reducing the inner loop by 2 cycles.
• Changed code 128 to send a new start address, this allows skips larger than 127 bytes to happen quicker than sending 2 skip instructions.
• Parasite code has been changed to ignore any changes between the last change and the end of screen memory, meaning it wont send lots of skips down to the end of screen memory.

[dudleysoft71]

I'm not sure what size these transfers are, but would it be worth unrolling the loop 2X and jumping into the middle if it is an odd number?
Obviously, it can be unrolled as many times as you like, but the jumping in gets a little more expensive.

The address is self modifying so it would take more overhead to sotre and increment both addresses, since the largest block is 127 bytes it seems like a lot of extra code for not a lot of saving, I'm not sure of where the breakeven point would come with this change.

[BigEd]

From a learning perspective, it would be really good to see both the before and the after code, dudleysoft. If you could just put this code in a gist or a git repo, that would be a great record of what can be done and how to do it.

[dp11]

unrolling a few times will be a certain win even with self modifying code. Another option to consider is to interleave the data for example to receive 1K you end up with :

Code: Select all

.loop
LDA &FEE1    / 4
STA &3000,X  / 6

LDA &FEE1    / 4
STA &3100,X  / 6

LDA &FEE1    / 4
STA &3200,X  / 6

LDA &FEE1    / 4
STA &3300,X  / 6

INX          / 2
BNE loop     / 3

[tricky]

I'm not sure what size these transfers are, but would it be worth unrolling the loop 2X and jumping into the middle if it is an odd number?
Obviously, it can be unrolled as many times as you like, but the jumping in gets a little more expensive.

The address is self modifying so it would take more overhead to sotre and increment both addresses, since the largest block is 127 bytes it seems like a lot of extra code for not a lot of saving, I'm not sure of where the breakeven point would come with this change.

For 2X
Calc addr as usual, but when storing first addr, store second as well.

Code: Select all

sta odd+3+1 ; 4
sta odd+3+2 ; 4
tya ; 2
lsr A ; 2
bcs odd ; 2.5

.loop
LDA &FEE1    / 4
STA &XXXX,X  / 6
DEY          / 2
.odd
LDA &FEE1    / 4
STA &XXXX,X  / 6
DEY          / 2
BNE loop     / 3

So, one off overhead for loop is 14.5 (14 or 15) and saving is ~1.5Y (3*(Y-1)/2) so it is cheaper for > 10 bytes - unless I have miss-calculated.

[dudleysoft71]

I'd have to check how much space I've got left for the code, I'm trying to keep it in one page at &C00, I'm trying to leave as much of the normal low memory as possible, to allow the parasite to use whatever features of the host, and since its doing it's own rendering i felt that the UDCs were the best place for the host code.

[dp11]

show us your code compressing the other 200 odd bytes to get space to speed up the transfer sounds fun. You can also put the transfer loop in zero page and the self modifying code becomes faster as the address being modified is in zero page

[Arcadian]

Sounds like we may need to dedicate a chunk of time to James (dudleysoft71) as there is clearly a lot of interest and potentially questions!

Shall we put down 7.30pm - 8.00pm? (And if there are any coding help requests - which there haven't been any of so far - could be moved to 7.00 - 7.30pm?)

[dudleysoft71]

show us your code compressing the other 200 odd bytes to get space to speed up the transfer sounds fun. You can also put the transfer loop in zero page and the self modifying code becomes faster as the address being modified is in zero page

I've just created a github repository for the project here which has the latest build.

I'm not sure how much zero page is available while using the tube host, I notice that the tube itself is using some of ZP for it's own code.

The code for the host is here.

[BigEd]

Many thanks for sharing!

[dp11]

If you want to spin of a new thread that is fine. Can you send the pallet data backwards too? and the length *2

[dudleysoft71]

If you want to spin of a new thread that is fine. Can you send the pallet data backwards too? and the length *2

I have done, you can find it here.

[kieranhj]

I’m hoping my session will be closer to 30 minutes than an hour but I’ve yet to run through it all. There are some slides but want to make sure there’s more time spent on practical examples. This may or may not generate questions depending on whether I’m lecturing to the already knowledgeable!

EDIT: Hopelessly optimistic, it's probably an hour.

[IanS]

I've already sent invites out to most of the Dev Night regulars, but should anybody else want to attend - or if you haven't received the Zoom logins - please post in this thread and I will send them to you.

Can you send me an invite, I will try and join a bit later on.

[oss003]

@Dave, are the invitations already sent for tonights meeting?

Greetings
Kees

[billcarr2005]

I've already sent invites out to most of the Dev Night regulars, but should anybody else want to attend - or if you haven't received the Zoom logins - please post in this thread and I will send them to you.

I might get chance to drop in sometime during proceedings, if i could get myself a login

[Arcadian]

Hi

Kees yes the invites to Dev Night regulars were issued 10th December, I've checked and you should have received one. There have been plenty of acceptances so far.

IanS and BillCarr I will PM you the meeting info now. If you'd like to regularly receive the Dev Night invites let me know.