SetTile
==================================
Any Link Below is truncated to the root of SDK Manual,
NINTENDO%2064%20SDK%205_CDROM/man/allman50
Can Also be downloaded at romhacking
Compiled By Trevor
Credits to Zoinkity for Original Find
|
typedef struct {
int cmd:8;
unsigned int fmt:3;
unsigned int siz:2;
unsigned int pad0:1;
unsigned int line:9;
unsigned int tmem:9;
unsigned int pad1:5;
unsigned int tile:3;
unsigned int palette:4;
unsigned int ct:1;
unsigned int mt:1;
unsigned int maskt:4;
unsigned int shiftt:4;
unsigned int cs:1;
unsigned int ms:1;
unsigned int masks:4;
unsigned int shifts:4;
} Gsettile;
For refrence I have included Zoinkity's Original Findings.
upper word
format 0x00000000 000 rgba
0x00200000 001 yuv
0x00400000 010 ci
0x00600000 011 ia
0x00800000 100 i
size 0x00000000 000 4bit
0x00080000 01 8bit
0x00100000 10 16bit
0x00180000 11 32bit
0x0003FE00 line
0x000001FF tmem TMEM offset words (64bit)
lower word
07000000 tile
00F00000 palette
00080000 clamp t
00040000 mirror t
0003C000 mask t
00003C00 shift t
00000200 clamp s
00000100 mirror s
000000F0 mask s
0000000F shift s
|
BitMap: C=Clamp, M=Mirror, PAL=Pallet
| 1111 0101 |
100 |
1 0 |
00 |
0 0000 000 |
0 0010 0000 |
0000 0 |
111 |
0000 |
0 |
0 |
00 00 |
00 00 |
0 |
0 |
0000 |
0000 |
| CMD |
FMT |
SZE |
|
LINE |
TMEM |
PAD |
TILE |
PAL |
C |
M |
MSK S |
SHFTS |
C |
M |
MSK T |
SHFTT |
|
000 rgba
001 yuv
010 ci
011 ia
100 i |
00 4bit
01 8bit
10 16bit
11 32bit |
|
Words
per
Line |
Address
offset
Words |
Null |
0-7 |
0-F |
|
|
0-F |
0-F |
|
|
0-F |
0-F |
Applies image assuming given attibutes.
ie. F5600600 00FD8360 would load a 48x48 intensity+alpha image with 4bit pixel data and clamping the corners to the s/t given in the point data.
If palettes are loaded, you can set the palette bits to swap palettes at runtime.
How does this work... Line = 000000011 =3... how does 3 = 48?
OK, oops, worked it out, 3 * 16 = 48. 16 because 64/4bit = 16. Each line is 16 texels
Mip-Maps must set shift values(hex) to their level (may vary for non-square textures).
EG, Tile 0 shift 0 (1)
Tile 1 Shift 1 (0.5)
Tile 2 Shift 2 (0.25)
etc.
This allows 1 texel of Tile1 to cover the same area as 2x2 texels of Tile0 and 1 texel of Tile2 to cover 4x4 texels of Tile0.
To allow an effect where distance warrants more varieation but less sharp detail like a field you can set 2 32x32 textures to the same shift values above.
Alternativly you can set a 64x32 texture to mip-map down and double the shift value. This will get blurry very quickly.
Mapping this texture on Tile 0 and 1 with a shift of 4 makes
Result is the same as using Detail mode, except that in Detail Mode you UV the "bigger" image, in this mode you UV the "Smaller" image.
Also different is that Detail uses another texture while LOD can re-use a tile Given this example
Tile 0 = Shift 0, TMEM ADD = 0
Tile 1 = Shift 4, TMEM ADD = 0
Tile 2 = Shift 5, TMEM ADD = +1 (assuming Tile0 was 64x32x4 this add would be 1024)
Tile 3 = Shift 6, TMEM ADD = +2 (1,536)
Tile 4 = Shift 7, TMEM ADD = +3 (1,792)
Tile 5 = Shift 8, TMEM ADD = +4 (1,920)
Tile 6 = Shift 9, TMEM ADD = +5 (1,984)
Visualisation of 4bit CI Mode Tile Storage in TMEM for 2 textures or 1 texture + detail.
Textures are 64x32 and 32x32 mip-mapped (32x32 not mip-mapped as detail)
Each texture has 3 levels instead of 5 but is sufficiant to minimise moiring.
8Bit Ci can only have 1 pallet at a time
4bit IA can have double 1legnth but more moiring will occur since still only 3 tiles allowed each.
eg, 64x64 and 64x32