PPU frame timing

The following behavior is tested by the ppu_vbl_nmi_timing test ROMs. Only the NTSC PPU is covered, though most probably applies to the PAL PPU.

Even/Odd Frames

• The PPU has an even/odd flag that is toggled every frame, regardless of whether the BG is enabled or disabled.
• With BG disabled, each PPU frame is 341*262=89342 PPU clocks long. There is no skipped clock every other frame.
• With BG enabled, each odd PPU frame is one PPU clock shorter than normal. I've timed this to occur around PPU clock 328 on the pre-render scanline, but haven't written a test ROM for it yet.
• By keeping BG disabled until after the time when the clock is skipped on odd frames, you can get a different color dot crawl pattern than normal (it looks more like that of interlace, where colors flicker between two states rather than the normal three). Presumably Battletoads (and others) encounter this, since it keeps the BG disabled until well after this time each frame.

CPU-PPU Clock Alignment

• The NTSC PPU runs at 3 times the CPU clock rate, so for a given power-up PPU events can occur on one of three relative alignments with the CPU clock they fall within. The PPU uses a different master clock divider, so there are more than just three alignments possible before powering up. The results below only cover one particular set of alignments (the one with the fewest special cases).

• Synchronizing with the PPU clock: If BG rendering is off, each frame will be 29830.6667 CPU clocks long. If the CPU checks the VBL flag in a loop every 29831 clocks until it is set, at some point it will end when the VBL flag is set just before the CPU reads it:

	   PPU      29830.7   29830.7  29830.7  29830.7
	-C--P-------C--P-------C-P-------CP-------*
	CPU       29831      29831     29831    29831
	
* The above synchronization will result in an exact PPU/CPU alignment.
	
	 -|--.--V--|-     -|--V--.--|-     -V--.--.--|- 
	Read             Read             Read
	                                 Loop will stop here

VBL Flag Timing

See also: NMI

• Reading $2002 within a few PPU clocks of when VBL is set results in special-case behavior. Reading one PPU clock before reads it as clear and never sets the flag or generates NMI for that frame. Reading on the same PPU clock or one later reads it as set, clears it, and suppresses the NMI for that frame. Reading two or more PPU clocks before/after it's set behaves normally (reads flag's value, clears it, and doesn't affect NMI operation).
• On an NTSC machine, the VBL flag is cleared 6820 PPU clocks, or exactly 20 scanlines, after it is set. In other words, it's cleared at the start of the pre-render scanline. (TO DO: confirmation on PAL NES and common PAL famiclone)