Talk:The skinny on NES scrolling
I made the following changes:
- I removed the first example, since it appeared to be broken w.r.t. fine Y scroll.
- I added a note about $2007 having side effects on v, please elaborate or correct it if anything is wrong. I note that nintendulator does a weird Y-1 thing if rendering is enabled when it is used? I'm not sure what is going on there.
- Bit 14 of t was inconsistently sometimes referred to as bit 15.
- Added a note about Y increment of v, which appears to occur on pixel 250 in nintendulator? I find this confusing, so I would appreciate if someone with authoritative knowledge could correctly explain the timing for the Y increment of v. Complete wrapping logic would also be helpful.
- The imaginary bit 15 was removed from the examples at the top of the page, for clarity and consistency.
- Added examples for simple scrolling, where only X, or no split is needed.
- Revised the $2006, $2005, $2005, $2006 example to be a more practical example, explaining how to set the scroll for a specific X/Y/nametable.
- Explanation of why the registers work this way, trying to make it easier to remember.
Please check for errors, and make corrections as necessary. We could also use some information on tile/attribute fetching, timings and which/whether can be changed mid-scanline.
- Rainwarrior 03:31, 8 January 2013 (MST)
Does x also have a corresponding temporary/reload register that gets copied into x at the beginning of each line?
Correct me if I'm wrong, but can't I write $2005 anywhere in the scanline, and x should be reloaded to the expected value at the start of the next line. If there was no reload value for it, and x is immediately set by $2005, it would be absolutely critical where in the scanline I happen to set it, and I don't remember this ever being the case when trying it on an NES.
- Rainwarrior 05:01, 8 January 2013 (MST)
- I guess this one doesn't matter. If it is reloaded we can just refer to the reload value as x. - Rainwarrior 06:42, 8 January 2013 (MST)
Making scroll writes during hblank... Some threads have suggested this, but is it possible, and how? When I've tried to do scrolling in the past, I could never seem to get timings that accurate, so it was always necessary to pull it back from the right edge and accept some amount of glitching on the end of the line. Glitches like this seem to be very normal for games with split scrolling... can it actually be avoided? Is it bad advice to suggest timing it during hblank?
- Rainwarrior 05:33, 8 January 2013 (MST)
- Okay, enough information has been gathered on the forum to answer this. The article is updated. - Rainwarrior 18:03, 8 January 2013 (MST)
Is it worth describing short sequences for quicker (and sometimes easier to calculate) but incomplete scrolling (i.e. 2005/2006, 2006/2006)? —Lidnariq 19:57, 8 January 2013 (MST)
- I don't see why not. Maybe put them in an "advanced examples" section? - Rainwarrior 23:22, 8 January 2013 (MST)
Actual timing for V/T updates during rendering
Some of this was mentioned in the Visual 2C02 forum thread, but I've just double-checked it in the simulator:
- Pre-render scanline, all dots 280 thru 304 - copy non-horizontal bits (yyyN.YYYYY.....) from T to V
- Every scanline, dots 328/336/8/16/24/32/.../240/248/256 - increment horizontal bits in V
- Every scanline, dot 256 - increment vertical bits in V
- Every scanline, dot 257 - copy horizontal bits from T to V
- Wow that's some great information, thanks! I will integrate it into the article when I have some time, if somebody else doesn't get to it first. - Rainwarrior 02:49, 11 January 2013 (MST)
Split X/Y scroll simplification
The example given in the article for mid-frame X/Y scroll updates uses the write sequence $2006 + $2005 + $2005 + $2006, where both of the $2006 write values need to be calculated ahead of time. Since only the last write is necessary to make the Y-scroll update take place, you can replace the first $2006 write with a $2005 write (if the base nametable isn't going to change, otherwise you'd need an additional write to $2000) which can be a bit simpler to actually code since you'd only need 3 distinct values, allowing you to safely fit the entire sequence within HBLANK (STX $2005; STY $2005; STX $2005; STA $2006). --Quietust (talk) 14:18, 8 August 2013 (MDT)
- On the one hand, now all four writes can fit in hblank. On the other hand, now all four writes have to be in hblank to avoid visible glitches. —Lidnariq (talk) 16:05, 8 August 2013 (MDT)
- Correct. There's a window of about 20 CPU cycles when v can be modified, namely during sprite fetch. A sequence of n consecutive writes takes 4n - 3 cycles from the start of the first write to the end of the last: 4n because STA $200x takes four cycles, and - 3 because the first opcode fetch doesn't count toward the timing. The $2006-$2005-$2005-$2006 sequence requires only the last two writes cycle of to be in this window, resulting in a write sequence that occupies 5 of 20 cycles. The simplification requires the whole thing because of the immediate effect of fine X changes, resulting in 13/20, and 7-cycle jitter is at the very limit of what can be achieved with sprite 0 or NMI alone. --Tepples (talk) 12:28, 9 August 2013 (MDT)