PPU registers

Register types:
 * R - Readable
 * W - Writeable
 * 8 - 8-bit access only
 * 16 - 8-bit access to either address, or 16-bit access to the lower address.
 * 24 - 8-bit or 16-bit access to 3 registers.
 * 8x2 - An internal 2-byte state accessed by two 8-bit read or writes (LSB first).

INIDISP - Screen display ($2100 write)
7 bit  0 F... BBBB |   |||| |    ++++- Screen brightness (linear steps from 0 = none to $F = full) +- Force blanking

BGMODE - BG mode and Character size ($2105 write)
7 bit  0 4321 PMMM |||| |||| |||| |+++- BG mode (see below) |||| + Mode 1 BG3 priority (0 = normal, 1 = high) |||+-- BG1 character size (0 = 8x8, 1 = 16x16) ||+--- BG2 character size (0 = 8x8, 1 = 16x16) |+ BG3 character size (0 = 8x8, 1 = 16x16) +- BG4 character size (0 = 8x8, 1 = 16x16)

BG Modes Mode| BG bit depth |Offsets |     Priorities (front -> back)       |                     Notes |BG1 BG2 BG3 BG4|per tile|                                     | 0 | 2   2   2   2 |   No   |   S3 1H 2H S2 1L 2L S1 3H 4H S0 3L 4L| 1 | 4   4   2     |   No   |   S3 1H 2H S2 1L 2L S1 3H    S0 3L   |BG3 priority = 0 |              |        |3H S3 1H 2H S2 1L 2L S1       S0 3L   |BG3 priority = 1 2 | 4   4         |  Yes   |   S3 1H    S2 2H    S1 1L    S0 2L   | 3 | 8   4         |   No   |   S3 1H    S2 2H    S1 1L    S0 2L   | 4 | 8   2         |  Yes   |   S3 1H    S2 2H    S1 1L    S0 2L   | 5 | 4   2         |   No   |   S3 1H    S2 2H    S1 1L    S0 2L   |Fixed 16 pixel char width. Forced high-res mode. 6 | 4             |  Yes   |   S3 1H    S2       S1 1L    S0      |Fixed 16 pixel char width. Forced high-res mode. 7 | 8             |   No   |   S3       S2       S1 1L    S0      |Fixed 8x8 char size. 7EXT| 8  7         |   No   |   S3       S2 2H    S1 1L    S0 2L   |Fixed 8x8 char size. BG2 bit 7 acts as priority.

MOSAIC - Screen pixelation ($2106 write)
7 bit  0 SSSS 4321 |||| |||| |||| |||+- Enable BG1 mosaic |||| ||+-- Enable BG2 mosaic |||| |+--- Enable BG3 mosaic |||| + Enable BG4 mosaic ++++-- Mosaic size in pixels (0 = 1x1, ..., 15 = 16x16)

BGnSC - BG1-4 tilemap address and size ($2107-$210A write)
7 bit  0 AAAA AAYX |||| |||| |||| |||+- Horizontal tilemap count (0 = 1 tilemap, 1 = 2 tilemaps) |||| ||+-- Vertical tilemap count (0 = 1 tilemap, 1 = 2 tilemaps) ++++-++--- Tilemap VRAM address (address = AAAAAA << 10)

Tilemaps may be placed at any 2 KiB page.

CHR word base address
The tile base address for background CHR can start at any 4 KiB page.

Tilemap offsets that go past the end of VRAM are allowed to wrap around to the beginning.

BG12NBA - BG1 and BG2 CHR word base address ($210B write)
7 bit  0 BBBB AAAA |||| |||| |||| ++++- BG1 CHR word base address (address = AAAA << 12) ++++-- BG2 CHR word base address (address = BBBB << 12)

BG34NBA - BG3 and BG4 CHR word base address ($210C write)
7 bit  0 DDDD CCCC |||| |||| |||| ++++- BG3 CHR word base address (address = CCCC << 12) ++++-- BG4 CHR word base address (address = DDDD << 12)

Scroll
Each of these scroll registers is normally updated by two single-byte writes to the same address. After two consecutive writes the scroll value is fully updated.

The two-write mechanism internally keeps shared latch values, so these registers should not normally be written in mixed order. Complete both writes to one register before moving on to the next.

The scroll offset is always relative to the top-left of the screen, even when updating mid-frame with HDMA.

BGnHOFS - BG1-4 horizontal scroll offset ($210D, $210F, $2111, $2113 write twice)
15 bit  8   7  bit  0 .... ..XX  XXXX XXXX ||  |||| ||||         ++---++++-++++- BGn horizontal scroll On write: BGnHOFS = (value << 8) | (bgofs_latch & ~7) | (bghofs_latch & 7) bgofs_latch = value bghofs_latch = value Note: BG1HOFS uses the same address as M7HOFS

BGnVOFS - BG1-4 vertical scroll offset ($210E, $2110, $2112, $2114 write twice)
15 bit  8   7  bit  0 .... ..YY  YYYY YYYY ||  |||| ||||         ++---++++-++++- BGn vertical scroll On write: BGnVOFS = (value << 8) | bgofs_latch bgofs_latch = value Note: BG1VOFS uses the same address as M7VOFS

TM - Main screen layer enable ($212C write)
7 bit  0 ...O 4321 | ||||   | |||+- Enable BG1 on main screen | ||+-- Enable BG2 on main screen | |+--- Enable BG3 on main screen | + Enable BG4 on main screen +-- Enable OBJ on main screen

TS - Subscreen layer enable ($212D write)
7 bit  0 ...O 4321 | ||||   | |||+- Enable BG1 on subscreen | ||+-- Enable BG2 on subscreen | |+--- Enable BG3 on subscreen | + Enable BG4 on subscreen +-- Enable OBJ on subscreen

SETINI - Screen Mode/Video Select ($2133 write)
7 bit  0 EX.. HOiI ||  |||| ||   |||+- Screen interlacing ||  ||+-- OBJ interlacing ||  |+--- Overscan mode ||  + High-res mode |+ EXTBG mode +- External sync


 * Screen interlacing causes every odd frame to lower its picture scanlines half a line between the even frames. When enabled, this produces a 480i picture composed of 2 frames (fields), instead of the default 240p progressive picture where each frame appears at the same vertical level.
 * STAT78 ($213F) can be used to check whether the current frame is an even or odd field.
 * When interlacing is enabled for BG mode 5 or 6, the BG layers are automatically interlaced to give a view of the background that has double the vertical resolution in 480i, effectively making every BG pixel half as tall.
 * OBJ interlacing interlaces the sprites to double their vertical resolution in 480i. Sprite pixels will appear half as tall.
 * Overscan mode enables the full 240 line picture when set, instead of only 224. On NTSC televisions this extra area is not normally visible, but on PAL it is very visible. Setting this causes NMI/vblank to begin 8 lines later, and end 8 lines earlier, dramatically reducing the vblank length in NTSC. Sprite and scroll positions are relative to the end of the blanking period, so enabling this automatically shifts everything up 8 lines. Using this feature makes the SNES drawing positions similar to the NES.
 * High-res mode doubles the horizontal output resolution from 256 to 512 pixels.
 * In most BG modes this causes the main screen to render pixels on even columns, and the sub screen to render on odd columns. This is sometimes called "pseudo-hires". Some games use this for a transparency effect (Kirby's Dreamland 3, Jurassic Park), relying on blurring from the composite video signal to blend the columns.
 * In BG modes 5 and 6, this high-res is forced, but the BG layers are automatically interleaved to double their horizontal resolution, making every BG pixel half as wide.
 * EXTBG controls a second-layer effect in BG mode 7 only.
 * External sync is used for super-imposing images from an external device. Normally 0.

VMAIN - Video Port Control ($2115 write)
7 bit  0 M... RRII |   |||| |    ||++- Address increment amount: |   ||     0: Increment by 1 word |   ||     1: Increment by 32 words |   ||     2: Increment by 128 words |   ||     3: Increment by 128 words |   ++--- Address remapping: |          0: None |          1: Remap rrrrrrrr YYYccccc -> rrrrrrrr cccccYYY (2bpp) |          2: Remap rrrrrrrY YYcccccP -> rrrrrrrc ccccPYYY (4bpp) |          3: Remap rrrrrrYY YcccccPP -> rrrrrrcc cccPPYYY (8bpp) +- Address increment mode: 0: Increment after writing $2118 or reading $2139 1: Increment after writing $2119 or reading $213A


 * Address remapping allows redirection of the write address to update 32-tile rows horizontally when using II = 0. Within a 32-tile group, sequential access iterates through the same 8-pixel row of each tile horizontally. After 32 spans, it returns to the second row of the first tile. Finally after a group of 32 tiles has been updated, it advances to the next group of 32 tiles..
 * This is suitable for a 32x32 tilemap in 8x8 tile mode. By filling each row of the tilemap with sequential values, each group of 32 tiles now corresponds to a contiguous horizontal span of pixels.
 * P = tile bitplane-word, c = group column, Y = tile pixel row, r = group row.
 * When setting the starting address, the starting tile of a 32-tile group will always be the at the same position as its remapped address.
 * With 4bpp or 8bpp modes, each increment advances through the 2 or 4 plane-words of a single tile before advancing to the next tile.
 * Simplified explanation:
 * 1. Write all planes for an 8 pixel span before proceeding horizontally to the next.
 * 2. After completing a row of 256 pixels (32 spans), proceed vertically to the next.

VMADDL, VMADDH - VRAM word address ($2116, $2117 write)
VMADDH     VMADDL $2117      $2116 7  bit  0   7  bit  0 hHHH HHHH  LLLL LLLL |||| ||||  |||| |||| ++++-++++---++++-++++- VRAM word address On write: Update VMADD vram_latch = [VMADD]

Because the SNES only has 64 KiB of VRAM, VRAM address bit 15 has no effect.

VMDATAL, VMDATAH - VRAM data write ($2118, $2119 write)
VMDATAH    VMDATAL $2119      $2118 7  bit  0   7  bit  0 HHHH HHHH  LLLL LLLL |||| ||||  |||| |||| ++++-++++---++++-++++- VRAM data word On $2118 write: If address increment mode == 0, increment VMADD On $2119 write: If address increment mode == 1, increment VMADD

VMDATALREAD, VMDATAHREAD - VRAM data read ($2139, $213A read)
VMDATAHREAD VMDATALREAD $213A      $2139 7 bit  0   7  bit  0 HHHH HHHH  LLLL LLLL |||| ||||  |||| ||||  ++++-++++---++++-++++- VRAM data word from vram_latch On $2139 read: value = vram_latch.low If address increment mode == 0: vram_latch = [VMADD] Increment VMADD On $213A read: value = vram_latch.high If address increment mode == 1, vram_latch = [VMADD] Increment VMADD

CGADD - CGRAM word address ($2121 write)
7 bit  0 AAAA AAAA |||| |||| ++++-++++- CGRAM word address On write: cgram_byte = 0

CGDATA - CGRAM data write ($2122 write twice)
15 bit  8   7  bit  0 .BBB BBGG  GGGR RRRR ||| ||||  |||| ||||   ||| ||||   |||+-++++- Red component ||| ||++---+++--- Green component +++-++-- Blue component On write: If cgram_byte == 0, cgram_latch = value If cgram_byte == 1, CGDATA = (value << 8) | cgram_latch cgram_byte = ~cgram_byte

Two single-byte writes to this register will update a single CGRAM word. The effect is applied only once the second byte is written.

Each write will increment the internal byte address. After two writes it will automatically have incremented to the next word.

CGDATA - CGRAM data read ($213B read twice)
15 bit  8   7  bit  0 xBBB BBGG  GGGR RRRR |||| ||||  |||| ||||  |||| ||||   |||+-++++- Red component |||| ||++---+++--- Green component |+++-++-- Blue component + PPU2 open bus On read: If cgram_byte == 0, value = CGDATA.low If cgram_byte == 1, value = CGDATA.high cgram_byte = ~cgram_byte

OBSEL - Object size and Character address ($2101 write)
7 bit  0 SSSN NbBB |||| |||| |||| |+++- Name base address (word address = bBB << 13) |||+-+ Name select (word offset = (NN+1) << 12) +++--- Object size: 0: 8x8  and 16x16 1: 8x8  and 32x32 2: 8x8  and 64x64 3: 16x16 and 32x32 4: 16x16 and 64x64 5: 32x32 and 64x64 6: 16x32 and 32x64 7: 16x32 and 32x32


 * Name base address selects a 16 KiB-aligned quarter of VRAM for the first 8 KiB of available sprite tiles. Bit 2 was reserved for a planned but never implemented expansion to 128 KiB VRAM, so is normally 0.
 * Name select controls a relative offset from the name base address in NN+1 8 KiB increments, selecting a second 8 KiB of available sprite tiles. With name select of 0, the second half follows the base 8 KiB contiguously.
 * Object size controls the sizes available for sprites. The two modes featuring rectangular sizes (6, 7) were not documented by the SNES development manual.

OAMADDL, OAMADDH - OAM word address ($2102, $2103 write)
OAMADDH    OAMADDL $2102      $2103 7  bit  0   7  bit  0 P... ...B  AAAA AAAA |      |   |||| |||| |       |   ++++-++++- OAM word address |      +- OAM table select (0 = 256 word table, 1 = 16 word table) +- OAM priority rotation (1 = enable) On write: Update OAMADD internal_oamadd = (OAMADD & $1FF) << 1

OAMDATA - OAM data write ($2104 write)
7 bit  0 DDDD DDDD |||| |||| ++++-++++- OAM data On write: If (internal_oamadd & 1) == 0, oam_latch = value If internal_oamadd < $200 and (internal_oamadd & 1) == 1: [internal_oamadd-1] = oam_latch [internal_oamadd] = value If internal_oamadd >= $200, [internal_oamadd] = value internal_oamadd = internal_oamadd + 1

When the OAM byte address is less than 512:
 * Two single-byte writes to this register will update a single OAM word. The effect is applied only once the second byte is written.

When the OAM byte address is 512 or above:
 * Each write immediately applies to the current byte.

Each write will increment the internal byte address.

OAMDATAREAD - OAM data read ($2138 read)
7 bit  0 DDDD DDDD |||| |||| ++++-++++- OAM data On read: value = [internal_oamadd] internal_oamadd = internal_oamadd + 1

M7SEL - Mode 7 settings ($211A write)
7 bit  0 RF.. ..YX ||    || ||    |+- Flip screen horizontally (backgrounds only) ||    +-- Flip screen vertically (backgrounds only) |+ Non-tilemap fill (0 = transparent, 1 = character 0) +- Tilemap repeat (0 = tilemap repeats, 1 = Non-tilemap fill beyond tilemap boundaries)

M7HOFS - Mode 7 horizontal scroll offset ($210D write twice)
15 bit  8   7  bit  0 ...X XXXX  XXXX XXXX | ||||  |||| ||||     +-++++---++++-++++- Mode 7 horizontal scroll (signed) On write: M7HOFS = (value << 8) | mode7_latch mode7_latch = value Note: This register uses the same address as BG1HOFS

M7VOFS - Mode 7 vertical scroll offset ($210E write twice)
15 bit  8   7  bit  0 ...Y YYYY  YYYY YYYY | ||||  |||| ||||     +-++++---++++-++++- Mode 7 vertical scroll (signed) On write: M7VOFS = (value << 8) | mode7_latch mode7_latch = value Note: This register uses the same address as BG1VOFS

M7A - Mode 7 matrix A and Multiplication factor 1 ($211B write twice)
15 bit  8   7  bit  0 DDDD DDDD  dddd dddd |||| ||||  |||| ||||  ++++-++++---++++-++++- Mode 7 matrix A (8.8 fixed point) ++++-++++---++++-++++- 16-bit multiplication factor On write: M7A = (value << 8) | mode7_latch mode7_latch = value

M7B - Mode 7 matrix B and Multiplication factor 2 ($211C write twice)
15 bit  8   7  bit  0 DDDD DDDD  dddd dddd |||| ||||  |||| ||||  ++++-++++---++++-++++- Mode 7 matrix B (8.8 fixed point) ++++-++++- 8-bit multiplication factor On write: M7B = (value << 8) | mode7_latch mode7_latch = value

M7n - Mode 7 matrix C-D ($211D-211E write twice)
15 bit  8   7  bit  0 DDDD DDDD  dddd dddd |||| ||||  |||| ||||  ++++-++++---++++-++++- Mode 7 matrix n (8.8 fixed point) On write: M7n = (value << 8) | mode7_latch mode7_latch = value

M7X - Mode 7 center X ($211F write twice)
15 bit  8   7  bit  0 ...X XXXX  XXXX XXXX | ||||  |||| ||||     +-++++---++++-++++- Mode 7 center X (signed) On write: M7X = (value << 8) | mode7_latch mode7_latch = value

M7Y - Mode 7 center Y ($2120 write twice)
15 bit  8   7  bit  0 ...Y YYYY  YYYY YYYY | ||||  |||| ||||     +-++++---++++-++++- Mode 7 center Y (signed) On write: M7Y = (value << 8) | mode7_latch mode7_latch = value

W12SEL - Window Mask Settings for BG1 and BG2 ($2123 write)
7 bit  0 DdCc BbAa |||| |||| |||| |||+- Invert window 1 for BG1 |||| ||+-- Enable window 1 for BG1 |||| |+--- Invert window 2 for BG1 |||| + Enable window 2 for BG1 |||+-- Invert window 1 for BG2 ||+--- Enable window 1 for BG2 |+ Invert window 2 for BG2 +- Enable window 2 for BG2

W34SEL - Window Mask Settings for BG3 and BG4 ($2124 write)
7 bit  0 HhGg FfEe |||| |||| |||| |||+- Invert window 1 for BG3 |||| ||+-- Enable window 1 for BG3 |||| |+--- Invert window 2 for BG3 |||| + Enable window 2 for BG3 |||+-- Invert window 1 for BG4 ||+--- Enable window 1 for BG4 |+ Invert window 2 for BG4 +- Enable window 2 for BG4

WOBJSEL - Window Mask Settings for OBJ and Color Window ($2125 write)
7 bit  0 LlKk JjIi |||| |||| |||| |||+- Invert window 1 for OBJ |||| ||+-- Enable window 1 for OBJ |||| |+--- Invert window 2 for OBJ |||| + Enable window 2 for OBJ |||+-- Invert window 1 for color math ||+--- Enable window 1 for color math |+ Invert window 2 for color math +- Enable window 2 for color math

WH0 - Window 1 left position ($2126 write)
7 bit  0 LLLL LLLL |||| |||| ++++-++++- Window 1 left edge position

WH1 - Window 1 right position ($2127 write)
7 bit  0 RRRR RRRR |||| |||| ++++-++++- Window 1 right edge position

WH2 - Window 2 left position ($2128 write)
7 bit  0 LLLL LLLL |||| |||| ++++-++++- Window 2 left edge position

WH3 - Window 2 right position ($2129 write)
7 bit  0 RRRR RRRR |||| |||| ++++-++++- Window 2 left edge position

WBGLOG - Window BG mask logic ($212A write)
7 bit  0 4433 2211 |||| |||| |||| ||++- BG1 window mask logic |||| ++--- BG2 window mask logic ||++-- BG3 window mask logic ++ BG4 window mask logic

WOBJLOG - Window OBJ and color math mask logic ($212B write)
7 bit  0 .... CCOO ||||     ||++- OBJ window mask logic ++--- Color window mask logic

Mask logic types Value|Logic 0 | OR   1 | AND 2 | XOR 3 | XNOR

TMW - Main screen layer window enable ($212E write)
7 bit  0 ...O 4321 | ||||   | |||+- Apply enabled windows to main screen BG1 | ||+-- Apply enabled windows to main screen BG2 | |+--- Apply enabled windows to main screen BG3 | + Apply enabled windows to main screen BG4 +-- Apply enabled windows to main screen OBJ

TSW - Subscreen layer window enable ($212F write)
7 bit  0 ...O 4321 | ||||   | |||+- Apply enabled windows to subscreen BG1 | ||+-- Apply enabled windows to subscreen BG2 | |+--- Apply enabled windows to subscreen BG3 | + Apply enabled windows to subscreen BG4 +-- Apply enabled windows to subscreen OBJ

CGWSEL - Color addition select ($2130 write)
7 bit  0 BBMM ..AD ||||  || ||||  |+- Direct color mode ||||  +-- Addend (0 = fixed color, 1 = subscreen) ||++-- Color math disable region ++ Clip colors to black before math region

Region types Value|Region 0 |Nowhere 1 |Outside color window 2 |Inside color window 3 |Everywhere

CGADSUB - Color math designation ($2131 write)
7 bit  0 MHBO 4321 |||| |||| |||| |||+- BG1 color math enable |||| ||+-- BG2 color math enable |||| |+--- BG3 color math enable |||| + BG4 color math enable |||+-- OBJ color math enable ||+--- Backdrop color math enable |+ Half color math +- Operator type (0 = add, 1 = subtract)

COLDATA - Fixed color data ($2132 write)
7 bit  0 BGRC CCCC |||| |||| |||+-++++- Color value ||+--- Write color value to blue channel |+ Write color value to green channel +- Write color value to red channel

MPYL, MPYM, MPYH - Multiplication result ($2134, $2135, $2136 read)
MPYH       MPYM        MPYL $2136      $2135       $2134 7  bit  0   7  bit  0   7  bit  0 HHHH HHHH  MMMM MMMM   LLLL LLLL |||| ||||  |||| ||||   |||| |||| ++++-++++---++++-++++---++++-++++- Multiplication result

This result may be read back after writing M7A with a 16-bit value (write twice), and M7B with an 8-bit value (write once). These two values will be multiplied to produce the value read here.

SLHV - Software latch for H/V counters ($2137 read)
7 bit  0 xxxx xxxx |||| |||| ++++-++++- Open bus On read: counter_latch = 1

OPHCT - Output horizontal counter ($213C read twice)
15 bit  8   7  bit  0 xxxx xxxX  XXXX XXXX |||| ||||  |||| ||||  |||| |||+---++++-++++- Horizontal counter value ++++-+++-- PPU2 open bus On read: If ophct_byte == 0, value = OPHCT.low If ophct_byte == 1, value = OPHCT.high ophct_byte = ~ophct_byte

OPVCT - Output vertical counter ($213D read twice)
15 bit  8   7  bit  0 xxxx xxxY  YYYY YYYY |||| ||||  |||| ||||  |||| |||+---++++-++++- Vertical counter value ++++-+++-- PPU2 open bus On read: If opvct_byte == 0, value = OPVCT.low If opvct_byte == 1, value = OPVCT.high opvct_byte = ~opvct_byte

When counter_latch transitions from 0 to 1, these registers are latched with the current counter values. counter_latch is set when SLHV is read or /EXTLATCH (PPU2 pin 29) is asserted, and is cleared when STAT78 is read. /EXTLATCH is connected to joypad IO D7 and can be controlled by the CPU via WRIO or by a joypad.

counter_latch behavior has not been fully confirmed.

STAT77 - PPU1 status flags and version ($213E read)
7 bit  0 TRMx VVVV |||| |||| |||| ++++- PPU1 version |||+-- PPU1 open bus ||+--- Master/slave mode (PPU1 pin 25) |+ Range over flag (sprite tile overflow) +- Time over flag (sprite overflow)

STAT78 - PPU2 status flags and version ($213F read)
7 bit  0 FLxM VVVV |||| |||| |||| ++++- PPU2 version |||+-- NTSC/PAL mode (0 = NTSC, 1 = PAL) (PPU2 pin 30) ||+--- PPU2 open bus |+ Counter latch value +- Interlace field On read: counter_latch = 0 ophct_byte = 0 opvct_byte = 0

If a condition that sets counter_latch is active when STAT78 is read, it is not known if counter_latch is cleared. Existing documentation suggests it is not cleared and the counters are not relatched.