Editing Pixar Image Computer (section)

== PXR File Format ==
Pixar Image Computer use 'Pixar storage standard' PXR File format. PXR file have structure:<ref>Pixar Image Computer Users Manual v1.2 Dec 1986</ref>

1. Header (512 bytes)

2. Tile pointer table (8 × numberTiles bytes)

3. Picture data

Origin of image is left top corner. All data in file use small endian byte order.

=== Header (512 bytes) ===
{| class="wikitable"
|-
! Offset (hex)
! Length
! Field name
! Description
|-
| 0 (0x0000)
| 4 bytes
| file signature/magic number
| {{tt|0x80 0xe8 0x00 0x00}}
|-
| 4 (0x0004)
| 2 bytes
| Version
| File format version (only have version 0)
|-
| 6 (0x0006)
| 246 bytes
| Ascii description
| String for file description
|-
| 252 (0x00fc)
| 4
| Label Pointer
| For continue label if need label longer than 246 bytes. Pointer address from start of file.
|-
| 256 (0x0100)
| 160 bytes
| colspan=2 | Unused
|-
| 416 (0x01a0)
| 2 bytes
| Picture height
| Pixel height of whole picture
|-
| 418 (0x01a2)
| 2 bytes
| Picture width
| Pixel width of full picture
|-
| 420 (0x01a4)
| 2 bytes
| Tile height
| Pixel height of each tile (≤ full picture height)
|-
| 422 (0x01a6)
| 2 bytes
| Tile width
| Pixel width of each tile (≤ full picture width)
|-
| 424 (0x01a8)
| 2 bytes
| Picture format
| 4 bits for tell which channel is in file: <code>R</code> = bit 0, <code>G</code> = bit 1, <code>B</code> = bit 2, <code>A</code> = bit 3; 4 low bits in low byte. 
|-
| 426 (0x01aa)
| 2 bytes
| Picture storage
| Picture encoding and number bits. Have 4 values: <code>0</code> = 8 bit encoded, <code>1</code> = 12 bit encoded, <code>2</code> = 8 bit dumped, <code>3</code> = 12 bit dumped
|-
| 428 (0x01ac)
| 2 bytes
| Blocking factor
| Optimum disk transfer chunk
|-
| 430 (0x01ae)
| 2 bytes
| Alpha mode
| Have 2 value: <code>0</code> = matted-to-black (pre multiply); <code>1</code> = unassociated
|-
| 432 (0x01b0)
| 2 bytes
| x offset
| x offset for picture
|-
| 434 (0x01b2)
| 2 bytes
| y offset
| y offset for picture
|-
| 436 (0x01b4)
| 76 bytes
| colspan=2 |Unused
|}

=== Tile pointer table (8 × numberTiles bytes) ===
The tile pointer table starts at byte 512 from the start of the file and has 1+ tile pointers.

{| class="wikitable"
|-
! 512 (0x200)
| Tile pointer 1
! Address of tile 0 (4 bytes)
! Length of tile 0 data (4 bytes)
|-
! 520 (0x208)
| Tile pointer 2
! Address of tile 1 (4 bytes)
! Length of tile 1 data (4 bytes)
|-
! ...
! ...
! ...
! ...
|}

To calculate the number of tiles for x direction:
 1 + (pictureWidth − 1)/tileWidth

and y direction:
 1 + (pictureHeight − 1)/tileHeight
 
Tile order is sweep from left to right, from top to bottom; first tile 0 is at left top of picture and last tile (totalNumberTiles − 1) is at right bottom of image.

=== Dumped & Encoded Pixel Data ===

Picture storage (at 2 bytes at byte 426 from start of file) can have values:

1. 0 - 8 bit channels encoded
2. 1 - 12 bit channels encoded
3. 2 - 8 bit channels dumped
4. 3 - 12 bit channels dumped

Dumped tiles store pixel data direct with order RGBA (4 channels), RGB (3 channels), or R (1 channel, grey image) for whole tile and no have extra byte or mark show end of scan line. Encoded tiles use packets for store image data. Each packet have 2 byte header and encoded pixel data after. Two byte header have flag and count for packet, flag is 4 low bits of byte 2, count is top 4 bits of byte 2 (top bits) combine with 8 bits of byte 1 (low bits) create 12 bit count. Flag have values:

{| class="wikitable"
|-
! Flag
! Count
! RGBA
! RGB
! R
! comment
|-
! 0
! no use
! 
!
! 
! end of disk block
|-
! 1
! use
! RGBA RGBA ...
! RGB RGB ...
! R ...
! full channel dump
|-
! 2
! use
! λRGBA λRGBA ...
! λRGB λRGB ...
! λR λR ...
! full channel run
|-
! 3
! use
! A RGB RGB ...
! no use
! no use
! partial channel dump
|-
! 4
! use
! A λRGB λRGB ...
! no use
! no use
! partial channel run
|}

For flag value 1 and 3, count is ''p'' - 1 number of pixels dumped in packet. For flag value 2 and 4, count is ''n'' - 1 number of run lengths in packet, λ is 1 byte (8 bits) store ''p'' - 1 number times repeat same pixel. Flag value 3 and 4 only use one A value for that packet and store only RGBA (4 channels) data. One packet never store pixel data from different scan lines but each scan line can use different/any flag.

=== Pixar 12 bit pixel format ===
Pixar Image Computer and PXR use special 12 bit fix point format have range [-1.5; 2.5), with 10 bits of precision (the documentation describes the range as [-0.5; 1.5) plus an "overflow bit").<ref>[http://www.bitsavers.org/pdf/pixar/Pixar_Image_Computer_Users_Manual_v1.2_Dec1986.pdf]</ref>

{| class="wikitable"
! Bits !! Value !! Overflow
|-
|1010 0000 0000||&minus;1.5||Y
|-
|1100 0000 0000||&minus;1.0||Y
|-
|1110 0000 0000||&minus;0.5||
|-
|0000 0000 0000||0||
|-
|0010 0000 0000||0.5||
|-
|0100 0000 0000||1.0||
|-
|0110 0000 0000||1.5||Y
|-
|1000 0000 0000||2.0||Y
|-
|1001 1111 1111||2.4995...||Y
|}

=== 16 bit coefficient values ===
Pixar Image Computer also have special 16 bit fixed point data type (for mảtix value, etc.) have name 'coefficient'. It have range [-2.0; 2.0) and use first 14 bits for fraction. For value ≥ 0, last 2 bit for whole number. For value < 0, last 2 bit for sign and whole number (use 2 complement).

{| class="wikitable"
|-
! decimal
| hex
! binary
! value
|-
! 32 768
| 0x8000
! 1000 0000 0000 0000
! -2.0
|-
! 40 960
| 0xa000
! 1010 0000 0000 0000
! -1.5
|-
! 49 152
| 0xc000
! 1100 0000 0000 0000
! -1.0
|-
! 57 344
| 0xe000
! 1110 0000 0000 0000
! -0.5
|-
! 65 536
| 0xfffff
! 1111 1111 1111 1111
! -0.00006103515625
|-
! 0
| 0x0000
! 0000 0000 0000 0000
! 0.0
|-
! 8192
| 0x2000
! 0010 0000 0000 0000
! 0.5
|-
! 16384
| 0x4000
! 0100 0000 0000 0000
! 1.0
|-
! 24 576
| 0x6000
! 0110 0000 0000 0000
! 1.5
|-
! 32767
| 0x7fff
! 0111 1111 1111 1111
! 1.99993896484375
|}