PPU nametables

From NESdev Wiki
Revision as of 00:39, 3 April 2011 by Drag (talk | contribs) (Individual bytes of CHR tiles are 8x1 pixels)
Jump to navigationJump to search
     (0,0)     (256,0)     (511,0)
       +-----------+-----------+
       |           |           |
       |           |           |
       |   $2000   |   $2400   |
       |           |           |
       |           |           |
(0,240)+-----------+-----------+(511,240)
       |           |           |
       |           |           |
       |   $2800   |   $2C00   |
       |           |           |
       |           |           |
       +-----------+-----------+
     (0,479)   (256,479)   (511,479)

Mirroring

Main article: Mirroring

The NES has four nametables, arranged in a 2x2 pattern. Each occupies a 1 KiB chunk of PPU address space, starting at $2000 at the top left, $2400 at the top right, $2800 at the bottom left, and $2C00 at the bottom right. Each byte in the nametable controls one 8x8 pixel character cell, and each nametable has 30 rows of 32 tiles each, for 960 ($3C0) bytes; the rest is used by each nametable's attribute table. With each tile being 8x8 pixels, this makes a total of 256x240 pixels in one map, the same size as one full screen.

But the NES system board itself has only 2 KiB of VRAM (called CIRAM), enough for two nametables; hardware on the cartridge controls address bit 10 of CIRAM to map one nametable on top of another.

  • Vertical mirroring: $2000 equals $2800 and $2400 equals $2C00 (e.g. Super Mario Bros.)
  • Horizontal mirroring: $2000 equals $2400 and $2800 equals $2C00 (e.g. Kid Icarus)
  • One-screen mirroring: All nametables refer to the same memory at any given time, and the mapper directly manipulates CIRAM address bit 10 (e.g. many Rare games using AxROM)
  • Four-screen mirroring: CIRAM is disabled, and the cartridge contains additional VRAM used for all nametables (e.g. Gauntlet, Rad Racer 2)

Background evaluation

Conceptually, the PPU does this 33 times for each scanline:

  1. Fetch a nametable entry from $2000-$2FBF.
  2. Fetch the corresponding attribute table entry from $23C0-$2FFF and increment the current VRAM address within the same row.
  3. Fetch the low-order byte of a 8x1 pixel sliver of pattern table from $0000-$0FF7 or $1000-$1FF7.
  4. Fetch the high-order byte of this sliver from an address 8 bytes higher.
  5. Turn the attribute data and the pattern table data into palette indices, and combine them with data from sprite data using priority.

It also does a fetch of a 34th (nametable, attribute, pattern) tuple that is never used, but some mappers rely on this fetch for timing purposes.