Undocumented Amiga hardware stuff - Page 4

yaqube · 22 December 2009, 19:47

Quote:

Originally Posted by Toni Wilen

I thought Denise hpos range (standard PAL) is 0 to 227.5 * 2 = 455.

For the short line (227 CCKs) it should be from 0 to 226*2+1 (= 453) but as it's advanced by 2 it is in range from 2 to 455.

Quote:

Unfortunately it is practically impossible to confirm without knowing internals of the chip

I tested this indirectly by changing sprite trigger position (the sprite trigerred on any hpos value between 2 and 455). The same test may be conducted on ECS Denise horizontal display window: if diwstop <= 455 (and >= 2) and diwstrt < 2 the display is blanked. Alternatively if diwstart is in 2 to 455 range and diwstop out of this range the display is never blanked horizontally.

Quote:

Confirmed, A1200 returns 0x00F8, ECS Denise 0xFFFC (was 0x00FC in UAE)

A4000 returns 0xFCF8. Actually A1200 misses the second shift register and the high byte is always 0.

Photon · 22 December 2009, 20:30

Yeah, I noticed this when I applied my maximum overscan mentioned in the overscan thread on my A600

According to HRM, 00 and ff should be valid DIWSTRT and DIWSTOP - it would be interesting to know if there's a HRM updated for ECS that mentions why ECS has this "limitation". 00 and ff works fine on OCS, which is what the HRM is written for, I guess. Until I see / hear otherwise

Maybe this nice thread will end up as a HRM addendum

Toni Wilen · 22 December 2009, 20:36

Quote:

Originally Posted by Photon

Yeah, I noticed this when I applied my maximum overscan mentioned in the overscan thread on my A600

According to HRM, 00 and ff should be valid DIWSTRT and DIWSTOP - it would be interesting to know if there's a HRM updated for ECS that mentions why ECS has this "limitation". 00 and ff works fine on OCS, which is what the HRM is written for, I guess. Until I see / hear otherwise

Maybe this nice thread will end up as a HRM addendum

Third edition includes "appendix C Enhanced Chip Set". Nothing about changed DIW limits. (except related DIWHIGH change)

I guess I need to put OCS Denise back and do some tests

Photon · 23 December 2009, 00:45

IIRC I had to go up pretty far ($28 or so??) to make DIWSTRT work? Could be mistaken.

Toni Wilen · 24 December 2009, 14:01

I finally did some tests with OCS Denise and as usual I was wrong..

horizontal diwstart 0-1 does not normally work (just like ECS Denise) but it seems horizontal diw does open if vpos is between 0 and 8 (inclusive)

Unfortunately logic analyzer is not connected to OCS Denise Amiga so I can't really be sure which lines have special behavior. (and why?)

More tests to do..

[few more tests done]

It looks like at least one line (8? 9?) counts to 445 (not 455) because changing diwstop hpos from 0xBE to 0xBD during this line (and then in next line to 0xC8 or larger) causes display to blank.

455 - 445 = 10 / 2 = 5. Hmm..

[and some more test]

I created copper list that moves hdiwstart so that it never matches (wait for 1/3 of scanline, hstart=2, wait for 2/3 of scanline, hstart=middle of scanline and so on..)

Test 1: line 1: make sure hdiw is open by setting hstrt to some normal value, hstop=0xff, wait 1-7 lines, run above copperlist: no display

Test 2: same but wait to line 9 (or more): hidw open, visible display

There is something weird happenin on line 8 (or all lines 0 to 8)

[4th]

Same tests with ECS Denise and no more weirdness, hdiw won't mysteriously close during early raster lines.

yaqube · 26 December 2009, 20:56

I have made some tests with an OCS Denise and a logic analyzer. I was wrong thinking that the OCS display window generation is considerably different than the ECS.

I can confirm all Toni's observations. During normal display hdiwstrt < 2 doesn't enable the display window. Also hdiwstop > 455 ($1C7) doesn't disable it. Exactly the same way acts an ECS Denise.

As Tony has noticed something interesting happens during display lines 0-8. The display window is enabled there. And I think I have found an explanation.

During first 9 display lines (0 - 8) register $038 (STREQU) is written (the write takes place during first refresh slot). That's a time when serration and equalization pulses are generated for CSYNC. Then during lines 9 - 25 (PAL) register $03A (STRVBL) is written.

It seems that writing $038 register doesn't reset (exactly doesn't set to 2) the hpos counter inside Denise. This counter is probably 9-bit wide and rolls over a few times during this period. That's why start condition of hdiwstart < 2 is satisfied. Also any hdiwstop is satisfied but at the end of the 9th line the hpos counter counts only to (455 + 9*454) % 512 = 445 so if hdiwstop > 445 the display is enabled.

If you wonder what the above numbers mean here is a brief explanation:
455 - last hpos value during the last display line
9 - number of lines when $038 strobe register is written
454 - line length (227 CCKs)

In ECS Denise this design omission has been fixed and that's why hdiwstrt < 2 doesn't enable the display window.

Toni Wilen · 26 December 2009, 21:36

Quote:

Originally Posted by yaqube

(455 + 9*454) % 512 = 445

Thanks. This explains it perfectly and is simple and logical

(I should have noticed this but I never checked what was the difference in strobes between lines 0 to 8 and 9+)

Toni Wilen · 27 December 2009, 14:45

Quick NTSC mode tests. Last "non-working" line is 9. Line 9 max hdiw is 397. 398 = flickers (long/short line), 400 is first non-flickering frame.

I think I am missing something because I can't find how it is supposed to be 400.. perhaps in NTSC first line(s) have different strobes than in PAL mode? (2 lines would look more or less perfect)

yaqube · 27 December 2009, 15:37

Indeed in the NTSC mode the first 10 lines are strobed by $038 register.
So if the last line of the previous frame was a short line (227 CCKs) then the hpos counter at the end of the 10th line is equal to:

(455 + 5*454 + 5*456) % 512 = 397

455 - last hpos value at the end of the previous frame (assuming a short line)
5 - number of short and long lines during the first 10 display lines
454 - length of a short line (227 CCKs)
456 - length of a long line (228 CCKs)

If the previous frame ends with long line (228 CCKs) then this value is 399.

Toni Wilen · 09 February 2010, 10:51

OCS sprites outside of bitplane area is possible (officially supported on/off bit only in AGA): write to BPL1DAT once somewhere in the beginning of scanline and sprites appear normally until end of scanline. (I don't remember if BPLCON0 needs at least one enabled bitplane but at least this works outside of DDFSTRT-DDFSTOP range)

Can be combined with normal bitplanes (and can cause some display "alignment" issues) Useful if you want to save DMA cycles but for example still want to have something visible in left or right border.

Toni Wilen · 09 February 2010, 10:57

A1000 DIP Agnus (also in first A2000 revision) has undocumented "bug", display field's first or last (depending on your viewpoint) strobe isn't vblank probe which causes vblank interrupt to trigger one line later than in any other Amiga model. (yaqube found this one)

Everything else (including DFF004/DFF006) works normally, only vblank interrupt gets delayed.

Not common technical info: Paula sets the interrupt when it sees first vblank strobe in RGA bus, not Agnus. Only "real" Agnus interrupt is blitter finished and that uses separate pin that connects to Paula.

Photon · 09 February 2010, 11:03

Last paragraph: still only regarding A1000?

BPL1DAT trick: fails on ECS/AGA?

Toni Wilen · 09 February 2010, 15:52

Quote:

Originally Posted by Photon

Last paragraph: still only regarding A1000?

No, all Amigas, even AGA use same inter-chip communication. (in theory you could replace Lisa with Denise and display would still work as long all display modes and color depths are ECS compatible)

Quote:

BPL1DAT trick: fails on ECS/AGA?

It should work (at least it works on ECS), see also above reply

andreas · 14 May 2010, 01:15

Quote:

Originally Posted by Toni Wilen

OCS sprites outside of bitplane area is possible (officially supported on/off bit only in AGA)

Wow. That's really interesting.
Have you ever stumbled on any pre-1990 game that actually does make use of this?

Toni Wilen · 14 May 2010, 09:29

Quote:

Originally Posted by andreas

Wow. That's really interesting.
Have you ever stumbled on any pre-1990 game that actually does make use of this?

No. Only known program so far is demo Sequential by Andromeda. ("magnet" and "18 frames!" parts) But there has to be others..

andreas · 14 May 2010, 18:08

There ought to be some. To say the least, I'm almost a bit shocked!!
Well, this could've opened lots of new ways in coding for game programmers even in A500 times at the end of the 1980s, and there was nobody of them who would've bothered using it! From a "quick 'n dirty" point of view: quite a loss!

Toni Wilen · 15 May 2010, 11:55

BPLCON4 (AGA-only) sprite and bitplane color changes are delayed by 1 lores pixel compared to direct color register color changes.

Roots 2.0 / Sanity "flower zoomer" part has gaps if this is not correctly handled.

meynaf · 20 May 2010, 10:31

If you like strange things i suggest you get 020 or 030 (real hardware) and try to predict what's in the CCR's N bit after a CMP2 instruction. Good luck :-)
Btw. Mot' manuals say "undefined", of course.

Toni Wilen · 20 May 2010, 10:33

Quote:

Originally Posted by meynaf

If you like strange things i suggest you get 020 or 030 (real hardware) and try to predict what's in the CCR's N bit after a CMP2 instruction. Good luck :-)
Btw. Mot' manuals say "undefined", of course.

CPU (non-68000) is not as interesting as chipset

meynaf · 20 May 2010, 14:52

Quote:

Originally Posted by Toni Wilen

CPU (non-68000) is not as interesting as chipset

Why ?

24 December 2009, 14:01	#65
Toni Wilen WinUAE developer Join Date: Aug 2001 Location: Hämeenlinna/Finland Age: 49 Posts: 26,502	I finally did some tests with OCS Denise and as usual I was wrong.. horizontal diwstart 0-1 does not normally work (just like ECS Denise) but it seems horizontal diw does open if vpos is between 0 and 8 (inclusive) Unfortunately logic analyzer is not connected to OCS Denise Amiga so I can't really be sure which lines have special behavior. (and why?) More tests to do.. [few more tests done] It looks like at least one line (8? 9?) counts to 445 (not 455) because changing diwstop hpos from 0xBE to 0xBD during this line (and then in next line to 0xC8 or larger) causes display to blank. 455 - 445 = 10 / 2 = 5. Hmm.. [and some more test] I created copper list that moves hdiwstart so that it never matches (wait for 1/3 of scanline, hstart=2, wait for 2/3 of scanline, hstart=middle of scanline and so on..) Test 1: line 1: make sure hdiw is open by setting hstrt to some normal value, hstop=0xff, wait 1-7 lines, run above copperlist: no display Test 2: same but wait to line 9 (or more): hidw open, visible display There is something weird happenin on line 8 (or all lines 0 to 8) [4th] Same tests with ECS Denise and no more weirdness, hdiw won't mysteriously close during early raster lines. Last edited by Toni Wilen; 25 December 2009 at 13:30.

26 December 2009, 20:56	#66
yaqube Registered User Join Date: Mar 2008 Location: Poland Posts: 159	I have made some tests with an OCS Denise and a logic analyzer. I was wrong thinking that the OCS display window generation is considerably different than the ECS. I can confirm all Toni's observations. During normal display hdiwstrt < 2 doesn't enable the display window. Also hdiwstop > 455 ($1C7) doesn't disable it. Exactly the same way acts an ECS Denise. As Tony has noticed something interesting happens during display lines 0-8. The display window is enabled there. And I think I have found an explanation. During first 9 display lines (0 - 8) register $038 (STREQU) is written (the write takes place during first refresh slot). That's a time when serration and equalization pulses are generated for CSYNC. Then during lines 9 - 25 (PAL) register $03A (STRVBL) is written. It seems that writing $038 register doesn't reset (exactly doesn't set to 2) the hpos counter inside Denise. This counter is probably 9-bit wide and rolls over a few times during this period. That's why start condition of hdiwstart < 2 is satisfied. Also any hdiwstop is satisfied but at the end of the 9th line the hpos counter counts only to (455 + 9454) % 512 = 445 so if hdiwstop > 445 the display is enabled. If you wonder what the above numbers mean here is a brief explanation: 455 - last hpos value during the last display line 9 - number of lines when $038 strobe register is written 454 - line length (227 CCKs) In ECS Denise this design omission has been fixed and that's why hdiwstrt < 2 doesn't enable the display window. Last edited by yaqube; 27 December 2009 at 09:31.*

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
who can provide hardware to create ADFs of some old rare stuff like Nautilus...?	Bernd	support.Other	3	19 August 2011 23:41
Stuff for sale amiga a1200 plus more retro stuff	blast	MarketPlace	23	22 June 2010 19:05
Action Replay Undocumented Features	deicidal	support.Hardware	0	01 March 2010 17:15
I've got some Amiga stuff...I want your SNES stuff!	Fingerlickin_B	MarketPlace	14	20 February 2009 01:33
Amiga stuff for trade for Atari Stuff	8bitguy1	MarketPlace	0	12 February 2009 05:36

22 December 2009, 20:30	#62
Photon Moderator Join Date: Nov 2004 Location: Eksjö / Sweden Posts: 5,602	Yeah, I noticed this when I applied my maximum overscan mentioned in the overscan thread on my A600 According to HRM, 00 and ff should be valid DIWSTRT and DIWSTOP - it would be interesting to know if there's a HRM updated for ECS that mentions why ECS has this "limitation". 00 and ff works fine on OCS, which is what the HRM is written for, I guess. Until I see / hear otherwise Maybe this nice thread will end up as a HRM addendum

23 December 2009, 00:45	#64
Photon Moderator Join Date: Nov 2004 Location: Eksjö / Sweden Posts: 5,602	IIRC I had to go up pretty far ($28 or so??) to make DIWSTRT work? Could be mistaken.

27 December 2009, 14:45	#68
Toni Wilen WinUAE developer Join Date: Aug 2001 Location: Hämeenlinna/Finland Age: 49 Posts: 26,502	Quick NTSC mode tests. Last "non-working" line is 9. Line 9 max hdiw is 397. 398 = flickers (long/short line), 400 is first non-flickering frame. I think I am missing something because I can't find how it is supposed to be 400.. perhaps in NTSC first line(s) have different strobes than in PAL mode? (2 lines would look more or less perfect)

27 December 2009, 15:37	#69
yaqube Registered User Join Date: Mar 2008 Location: Poland Posts: 159	Indeed in the NTSC mode the first 10 lines are strobed by $038 register. So if the last line of the previous frame was a short line (227 CCKs) then the hpos counter at the end of the 10th line is equal to: (455 + 5454 + 5456) % 512 = 397 455 - last hpos value at the end of the previous frame (assuming a short line) 5 - number of short and long lines during the first 10 display lines 454 - length of a short line (227 CCKs) 456 - length of a long line (228 CCKs) If the previous frame ends with long line (228 CCKs) then this value is 399.

09 February 2010, 10:51	#70
Toni Wilen WinUAE developer Join Date: Aug 2001 Location: Hämeenlinna/Finland Age: 49 Posts: 26,502	OCS sprites outside of bitplane area is possible (officially supported on/off bit only in AGA): write to BPL1DAT once somewhere in the beginning of scanline and sprites appear normally until end of scanline. (I don't remember if BPLCON0 needs at least one enabled bitplane but at least this works outside of DDFSTRT-DDFSTOP range) Can be combined with normal bitplanes (and can cause some display "alignment" issues) Useful if you want to save DMA cycles but for example still want to have something visible in left or right border.

09 February 2010, 10:57	#71
Toni Wilen WinUAE developer Join Date: Aug 2001 Location: Hämeenlinna/Finland Age: 49 Posts: 26,502	A1000 DIP Agnus (also in first A2000 revision) has undocumented "bug", display field's first or last (depending on your viewpoint) strobe isn't vblank probe which causes vblank interrupt to trigger one line later than in any other Amiga model. (yaqube found this one) Everything else (including DFF004/DFF006) works normally, only vblank interrupt gets delayed. Not common technical info: Paula sets the interrupt when it sees first vblank strobe in RGA bus, not Agnus. Only "real" Agnus interrupt is blitter finished and that uses separate pin that connects to Paula.

09 February 2010, 11:03	#72
Photon Moderator Join Date: Nov 2004 Location: Eksjö / Sweden Posts: 5,602	Last paragraph: still only regarding A1000? BPL1DAT trick: fails on ECS/AGA?

14 May 2010, 18:08	#76
andreas Zone Friend Join Date: Jun 2001 Location: Germany Age: 50 Posts: 5,857	There ought to be some. To say the least, I'm almost a bit shocked!! Well, this could've opened lots of new ways in coding for game programmers even in A500 times at the end of the 1980s, and there was nobody of them who would've bothered using it! From a "quick 'n dirty" point of view: quite a loss!

15 May 2010, 11:55	#77
Toni Wilen WinUAE developer Join Date: Aug 2001 Location: Hämeenlinna/Finland Age: 49 Posts: 26,502	BPLCON4 (AGA-only) sprite and bitplane color changes are delayed by 1 lores pixel compared to direct color register color changes. Roots 2.0 / Sanity "flower zoomer" part has gaps if this is not correctly handled.

20 May 2010, 10:31	#78
meynaf son of 68k Join Date: Nov 2007 Location: Lyon / France Age: 51 Posts: 5,322	If you like strange things i suggest you get 020 or 030 (real hardware) and try to predict what's in the CCR's N bit after a CMP2 instruction. Good luck :-) Btw. Mot' manuals say "undefined", of course.

Currently Active Users Viewing This Thread: 1 (0 members and 1 guests)