The Fast3DEX display list commands are microcode that are passed to the RSP for rendering. They are all 8 bytes long.

Stalls the signal processor (the RSP), and as a consequence, the RDP too. This should only be used for debugging purposes.

00 00 00 00 00 00 00 00

Apply transformation matrix. Used in HUD text printing code to translate textures, as well as copious other transformations including animations.

01 [PP] 00 00 [AA AA AA AA]

Parameters (can be OR'd together):

Example: Push matrix and and multiply by matrix at 0x00213DF8; Multiply by matrix at 0x00213DB8

01 04 00 40 00 21 3D F8
01 00 00 40 00 21 3D B8

Takes a block of memory from an address and puts it in the location pointed to by an index and an offset.

03 [nn] [oo] [ii] [aa aa aa aa]

i is an index into a table of addresses of DMEM. Given enumerations for i are:

• G_MV_MMTX = 2
• G_MV_PMTX = 6
• G_MV_VIEWPORT = 8
• G_MV_LIGHT = 10
• G_MV_POINT = 12
• G_MV_MATRIX = 14

Note however that only VIEWPORT, LIGHT, and MATRIX are used by any of the macros given to programmers.

Example: Loads diffuse RGBA from 0x0 in Segment 0x0E; loads ambient RGBA from 0x08 in Segment 0x0E

03 86 00 10 0E 00 00 00
03 88 00 10 0E 00 00 08

Fills the vertex buffer with vertex information (ex. coordinates, color values). Max amount of bytes to load in F3DEX is 0x200 (32 vertices).

04 [II] [xx xx] [SS SS SS SS]

xx xx = NNNN NNLL LLLL LLLL

Example: Load 30 (0x1E0 / 0x10) vertices from 0x0E000780 and put them into the vertex buffer, starting at position 0.

04 00 79 DF 0E 00 07 80

Signifies the start of a Display List. May be used to link data and branch the current DL.

06 [AA] 00 00 [BB BB BB BB]

Example: Loads a display list from 0xA50 in segment 0x07.

06 00 00 00 07 00 0A 50

Loads a new microcode executable into the RSP.

B4 00 00 00 [dd dd dd dd]
AF 00 [ss ss] [tt tt tt tt]

d points to the start of the data section and t to the start of the text section, with s specifying the size of the data section. After loading the new microcode, the RCP is reset as appropriate.

Note that d is stored in the high half of the “RDP word”, the opcode for which is shown in italics before the actual AF opcode.

Tests the Z value of the vertex at a buffer index against a given z-value.

B4 00 00 00 [dd dd dd dd]

B0 [aa a][b bb] [zz zz zz zz]

If the given vertex's Z value ≤ z, then the processor switches over to the display list at address d (equivalent to 06 jump with no return stored). Otherwise continues through the display list. Useful for LOD-related model processing, where several occurrences of this can be stacked to branch to progressively closer versions of the model.

Note that the B0 opcode actually pulls the address of the branching display list from the high half of the “RDP word” (16 bytes used for generic value storage, as far as is known). The italicized opcode given above is set by the “basic function” listed for this opcode, immediately before the actual B0 opcode.

Although not stated in the documentation, the vertex buffer index is presumably limited to the range 0 ≤ index ≤ 31 just like other buffer indices. It's unknown why the opcode needs the index twice, and multiplied by different amounts.

Also, z is usually calculated for the N64 programmer (only the *Zraw form specifies a raw z value). The formula used in these cases is as shown, written as C code:

  return (int32_t)( x * 0x00010000);

Where zval is the Z value the programmer is thinking of comparing against (whereas the “real” Z value handed to the opcode is calculated by this function), near and far denote the distances of the near and far clipping planes, respectively, from the viewpoint. zmin and zmax are the minimum and maximum possible Z values, respectively. flag decides on the projection, taking either of these enumerations:

• G_BZ_PERSP = 0
• G_BZ_ORTHO = 1

As their names would suggest, PERSP handles perspective projection, ORTHO handles orthographic.
Note that in a default setup, zmin is 0, and zmax is 0x03FF.

This formula cannot be reversed for the value provided in the microcode, since the necessary information is absent. (Since the values are provided in the macro call generating this opcode, there's no guarantee any of the values match what they're “supposed” to be, e.g. near could be set to something far different from the location of the actual near clipping plane.)

Renders two triangles according to the vertices inside the vertex buffer

B1 [AA] [BB] [CC] 00 [DD] [EE] [FF]

Example: Render a triangle using the vertex data specified at the vertex buffer positions 0, 1 (2/2) and 2 (4/2), along with a second triangle at vertex buffer positions 3 (6/2), 4 (8/2), and 5 (0x0A/2).

B1 00 02 04 00 06 08 0A

Modifies a four-byte portion of the vertex specified.

B2 ww nn nn vv vv vv vv

The portion modified is specified by w, and the new value is given in v. Lighting calculations (if enabled) and position transformations are not calculated by the RSP after use of this command, so modifications modify final color and vertices.

The valid values for w have names as follows:

• G_MWO_POINT_RGBA = 0x10 Modifies the color of the vertex
• G_MWO_POINT_ST = 0x14 Modifies the texture coordinates
• G_MWO_POINT_XYSCREEN = 0x18 Modifies the X and Y position
• G_MWO_POINT_ZSCREEN = 0x1C Modifies the Z position (lower four nybbles of v should always be zero for this modification)

The exact nature of these values is unclear. The SDK documentation describes them as “byte offsets”, however they don't match offsets in the vertex structure.

This opcode sets the low half of the generic RDP word.

B3 00 00 00 [LL LL LL LL]

Note: the referenced function also generates opcode B4. See the section on B4 for info on that.

This opcode sets the higher half of the generic RDP word.

B4 00 00 00 [hh hh hh hh]

Note: the referenced function also generates opcode B3. See the section on B3 for info on that.

Renders one quad according to the vertices inside the vertex buffer

B5 [AA] [BB] [CC] 00 [DD] [EE] [FF]

Example: Render a quad using the vertex data specified at the vertex buffer positions 0, 1 (2/2), 2 (4/2), 0, 2 (4/2) and 3 (6/2).

B5 00 02 04 00 00 04 06

Note: It's possible this command is actually G_Line3D. Needs to be tested on hardware or LLE.

Disables certain geometry parameters (ex. lighting, front-/backface culling, Z-buffer).

B6 00 00 00 [AA AA AA AA]

Parameters:
See RSP Geometry Mode under RCP Structs.

Examples:

B6 00 00 00 00 02 10 00 : Disable G_LIGHTING and G_CULL_FRONT
B6 00 00 00 00 02 00 00 : Disable G_LIGHTING
B6 00 00 00 00 00 00 00 : Disable nothing
B6 00 00 00 00 00 22 00 : Disable G_CULL_BACK and G_SHADING_SMOOTH

Enables certain geometry parameters (ex. lighting, front-/backface culling, Z-buffer).

B7 00 00 00 [AA AA AA AA]

Parameters:
See RSP Geometry Mode under RCP structs.

Examples:

B7 00 00 00 00 02 10 00 : Enable G_LIGHTING and G_CULL_FRONT
B7 00 00 00 00 02 00 00 : Enable G_LIGHTING
B7 00 00 00 00 00 00 00 : Enable nothing
B7 00 00 00 00 00 22 00 : Enable G_CULL_BACK and G_SHADING_SMOOTH

Terminates the current Display List

B8 00 00 00 00 00 00 00

Modifies various bits of the lower half of the RDP Other Modes word.

B9 00 [ss] [nn] [dd dd dd dd]

The operation performed by the opcode can be expressed as such, assuming LO to stand for the lower word of the RDP other modes:

Or, in English, it clears all the bits designated to be modified (determined by s and n), and then sets those bits according to data. Important to note that mode-bits are preshifted. That is, it is up to the caller to shift data appropriately. The enumerations provided for data are pre-shifted to match this requirement.
The enumerations provided for s are:

• G_MDSFT_ALPHACOMPARE = 0
• G_MDSFT_ZSRCSEL = 2
• G_MDSFT_RENDERMODE = 3

The structure of the lower modes is provided in the RCP structs page.

Modifies various bits of the lower half of the RDP Other Modes word.

BA 00 [ss] [nn] [dd dd dd dd]

Like with B9, data is preshifted, and the opcode's procedure can be formulised in the same way (where OH stands for the high word of the other modes):

The enumerations provided for s are:

• G_MDSFT_BLENDMASK = 0
• G_MDSFT_ALPHADITHER = 4
• G_MDSFT_RGBDITHER = 6
• G_MDSFT_COMBKEY = 8
• G_MDSFT_TEXTCONV = 9
• G_MDSFT_TEXTFILT = 12
• G_MDSFT_TEXTLUT = 14
• G_MDSFT_TEXTLOD = 16
• G_MDSFT_TEXTDETAIL = 17
• G_MDSFT_TEXTPERSP = 19
• G_MDSFT_CYCLETYPE = 20
• G_MDSFT_COLORDITHER = 22
• G_MDSFT_PIPELINE = 23

The structure of the higher modes is provided in the RCP structs page.

Sets the texture scaling factor.

BB 00 [xx] [nn] [ss ss] [tt tt]

xx = 00LL Lddd

Note that scaling factors are binary fractional values with an implied 0.; for example, a scaleS of 0x8000 means 0b0.1000_0000_0000_0000, or 0d0.5.

Examples:

BB 00 00 01 FF FF FF FF : Standard - at start for extremely close to 1 scaling enabled on tile desc 0.
BB 00 00 01 80 00 80 00 : Standard - at start for 0.5 scaling enabled on tile desc 0.
BB 00 00 01 0F 80 07 C0 : Scale for environment mapping 64x32 and enable tile desc 0.
BB 00 00 00 FF FF FF FF : Disable tile descriptor 0 (end of DL).

Loads a new 32-bit value data to the location specified by index and offset.

BC oo oo ii dd dd dd dd

i is an index into a table of DMEM addresses. The enumerations given for this are:

The offset is, as the name suggests, an offset from the address that index resolves to. Enumerations are also provided for this value, for each of the enumerated index values. (XXX where to put them? It's a bit of a list.)

Pops num matrices from the stack specified by modelview matrix stack.

BD380002 aaaaaaaa

BE 00 vv vv 00 00 ww ww

This command takes the vertices in the vertex buffer from vfirst through vlast as describing the volume of the object being rendered (called the “bounding value”). If the bounding volume does not intersect with the current viewing volume (aka if the bounding volume is entirely offscreen), then the display list ends, equivalent to a B8 opcode. Otherwise, the display list continues as though nothing happened (equivalent to 00, then).

For F3DEX, Each of vfirst and vlast must be in the range 0 ≤ x ≤ 31. Additionally, vfirst < vlast must be true (thus a minimum of two vertices must be specified, and range of vertices cannot be reversed). It's not specified what the behavior is when all the vertices are coplanar.

Renders one triangle according to the vertices inside the vertex buffer

BF 00 00 00 00 [AA] [BB] [CC]

Example: Render a triangle using the vertex data specified at the vertex buffer positions 0, 1 (0x02 / 0x02) and 2 (0x04 / 0x02).

BF 00 00 00 00 00 02 04

Does nothing. Seemingly different from 00; judging by the names and position of this opcode and 00, this opcode stalls the RDP, whereas the other stalls the RSP.

C0 00 00 00 00 00 00 00

Draws a textured 2D rectangle on the screen

E4 [xx x][y yy] 0[I] [XX X][Y YY]
B4 00 00 00 [SS SS] [TT TT]
B3 00 00 00 [DD DD] [EE EE]

Draws a textured 2D rectangle on the screen

E5 [xx x][y yy] 0[I] [XX X][Y YY]
B4 00 00 00 [SS SS] [TT TT]
B3 00 00 00 [DD DD] [EE EE]

This is similar to opcode E4, with the only difference being that the S and T coordinates of a texture are flipped in rendering, so that S coordinates are along the Y axis and T along the X axis. This essentially flips the texture about the diagonal line (XXX,YYY),(xxx,yyy).

This forces a wait for a texture to load, in order to synchronize with pixel rendering. This ensures that loading a new texture won't disrupt the rendering of primitives mid-render.

E6 00 00 00 00 00 00 00

This is not the “start of display list” command. There is no such thing.

Waits for the RDP to finish rendering its currently-rendering primitive, before updating RDP attributes. This avoids altering the rendering of a primitive in the middle of its render.

E7 00 00 00 00 00 00 00

Historically, several model viewers have interpreted this command as somehow starting a display list. This is not true. Display lists have no command that starts a display list. To find display lists, the most accurate way would be to find the end display list command (B8) and work backwards, until an invalid display list command is encountered, or to parse byte-code that points to display lists.

Forces a wait for rendering to finish before updating tile descriptor attributes, so as to not disrupt rendering of primitives mid-render.

E8 00 00 00 00 00 00 00

Generates an interrupt for the main CPU when the RDP has finished doing everything. This is typically the last opcode before the “end display list” opcode (B8).

E9 00 00 00 00 00 00 00

Sets the center, scale, and width parameters for the green and blue components for chroma key (see EB for red).

EA [ww w][x xx] [cc] [ss] [dd] [tt]

w and x are in an unsigned fixed-point 4.8 format. The other parameters are normal 8-bit unsigned integers.

Sets the scale, center, and width parameters for the red component for chroma key (see EA for blue and green).

EB 00 00 00 0[w ww] [cc] [ss]

Sets the scale, center, and width parameters for the red component for chroma key (see EA for blue and green).

w is an unsigned fixed-point 4.8 number. c and s are 8-bit unsigned integers.

Sets the terms need to perform conversion from YUV to RGB.

   E    C  00aa aaaa  aaab bbbb  bbbb cccc
cccc cddd  dddd ddee  eeee eeef  ffff ffff

All of the parameters are signed 9-bit numbers, with range -256 ≤ k ≤ 255.

Sets the scissoring rectangle.

ED [xx x][y yy] 0[m] [vv v][w ww]

Valid values for m are:

• G_SC_NON_INTERLACE = 0
• G_SC_EVEN_INTERLACE = 2
• G_SC_ODD_INTERLACE = 3

NON_INTERLACE draws all scanlines, while EVEN and ODD draw only even or odd scanlines, respectively.

This sets the Z value for the entire primitive to be rendered, when the source for its depth component is set to be taken from this value. (This could be thought of as the depth version of “flat shading”.)

EE 00 00 00 zz zz dd dd

z and d are signed 16-bit values. z sets the actual depth for the primitive, while d helps rendering of anti-aliased and decal objects (this is usually 0).

Sets the other mode bits of the RDP.

EF [HH HH HH] [LL LL LL LL]

H is similar to using BA, and L is similar to using B9.

The difference from either BA or B9 is that this command sets all the bits in both halves of the RDP word (that is, this opcode has no way of limiting what bits are actually affected).

The RSP Geometry Mode word bits' parameters can be found under the RCP structs page.

Tells the RDP to load a number of colors from the texture loaded to a given tile descriptor. The tile descriptor's attributes and other state related to textures inform the RDP how to load the palette, and where to load it to.

F0 00 00 00 0[t] [cc c]0 00

(Note: the count is quadrupled likely due to how palettes are stored in TMEM. Not sure where details on that would be appropriate, though.)

Sets the texture coordinates and size

F2 [SS S][T TT] 0[I] [WW W][H HH]

Examples:

F2 00 00 00 00 07 C0 7C : 32x32 textures
F2 00 00 00 00 0F C0 7C : 64x32
F2 00 00 00 00 07 C0 FC : 32x64

Determines how much data to load after SETTIMG

F3 [SS S][T TT] 0[I] [XX X][D DD]

Some notes on dxt:

Examples:

F3 00 00 00 07 7F F1 00 : 32x64 16-bit Textures
F3 00 00 00 07 7F F0 80 : 64x32 16-bit Textures
F3 00 00 00 07 3F F1 00 : 32x32 16-bit Textures 
F3 00 00 00 07 FF F1 00 : 64x64 8-bit Textures 

Loads a rectangular portion of the texture being pointed to in DRAM for given tile descriptor into TMEM.

F4 [ss s][t tt] 0[i] [uu u][v vv]

(s, t) specifies the upper-left corner of the texture in RAM, offset from its earlier-specified origin, and (u, v) specifies the lower-right corner of the texture to load.

All coordinate values are in unsigned fixed-point 10.2 format (range 0 ≤ n ≤ 1023.75).

Sets the texture properties (ex. mirroring)

F5 [XX XX XX XX XX XX XX]

Bit-wise parameters:

  [F  5]  fffi i0nn nnnn nnnm mmmm mmmm

0000 0ttt pppp ccaa aass ssdd bbbb uuuu

Or simply Bitsize = 4*2^(i)

Examples:

F5 10 00 00 07 00 00 00 : Settile for RGBA16, followed by another F5 command (rendertile)
F5 70 00 00 07 00 00 00 : Settile for IA16, followed by another F5 command (rendertile)
F5 10 10 00 07 01 40 50 : Rendertile for RGBA16 32x32 textures
F5 10 20 00 07 01 40 60 : Rendertile for RGBA16 64x32 textures
F5 70 10 00 00 01 40 50 : Rendertile for IA16 32x32 textures
F5 00 01 00 01 00 00 00 : Settile for palettes loading into TLUT0 in tile descriptor 1
F5 40 08 00 00 09 80 60 : Rendertile for CI4 64x64 textures with clamping on the T (vertical) axis

Draws a colored rectangle on the screen. Use command 0xF7 to set the color of the rectangle.

F6 [VV V][W WW] 00 [XX X][Y YY]

Sets the fill color for use in fill mode.

F7 00 00 00 [CC CC CC CC]

Sets the fog color

F8 00 00 00 [RR] [GG] [BB] [AA]

Example: set the fog for the following polygons to pure green

F8 00 00 00 00 FF 00 FF

Sets the RDP blender's blend color, a general purpose color register available for various purposes in the blender.

F9 00 00 00 [RR] [GG] [BB] [AA]

Example: make blend colour green and opaque

FB 00 00 00 00 FF 00 FF

Sets the primitive color register of the RDP's color combiner. Also sets two additional values available to the color combiner, the minimum LOD level and an LOD fraction.

FA 00 [mm] [ff] [RR] [GG] [BB] [AA]

m and f are both fixed-point unsigned 0.8 numbers, meaning a range of 0 ≤ n ≤ 0.99609375. m defines the minimum possible value for LOD to have, when the LOD calculated for a particular part of the primitive is less than 1.0. In other words, the LOD of any part of the primitive is clamped at the lower end to max(m, LOD).

f specifies a fraction that the programmer can specify for use in the color combiner of the RDP. It's meant to offer a further refinement (XXX verify that, not 100% sure on it) on the linear filtering of two mipmaps.

Sets the environment color for combiner

FB 00 00 00 [RR] [GG] [BB] [AA]

Example: make everything loaded underneath the command red, and semi-transparent

FB 00 00 00 FF 00 00 8C

Performs combining operations (ex. multi-texturing). The N64's Color Combiner calculates the output color by combining the various source colors and source alphas using the following equation: outputcolor = ( A - B ) * C + D.

FC [XX XX XX XX XX XX XX]

Bitwise mask for X:

[aaaa] [ccccc] [eee] [ggg] [iiii] [kkkkk] [bbbb] [jjjj] [mmm] [ooo] [ddd] [fff] [hhh] [lll] [nnn] [ppp]

More information here:

N64's Color Combiner

Examples:

FC 12 7F FF FF FF F8 38 : Standard usage for solid RGBA textures
FC 12 18 24 FF 33 FF FF : Standard usage for alpha RGBA textures

Sets the texture image offset

FD [xx] 00 00 [bb bb bb bb]

[xx] = fffi i000

Or simply Bitsize = 4*2^(i)

Example: Load texture position from 0x90 in RAM bank 0x04 for RGBA16

FD 10 00 00 04 00 00 90

Sets the location of the depth buffer

FE 00 00 00 [ii ii ii ii]

Since the format of the depth buffer is fixed (14-bit Z value and 2-bit dZ value), those parameters are not here, unlike for FD and FF. The width of the depth buffer is shared with the color buffer (see FF), and thus isn't set here.

Sets the location of the color framebuffer, usually a framebuffer that's displayed to the screen.

FF [xx] 0[w ww] [ii ii ii ii]

xx = fffs s000 (binary)

The color buffer is set to RAM location i, with a width of w. f and s are the typical image format enumerations, described in G_Settile.

The location is usually some place set as the framebuffer to show (which is not handled by display lists), but oftentimes will be set to a depth buffer. This is usually done to fill (aka clear) the depth buffer in fill mode.

• http://www.smwcentral.net/?p=viewthread&t=74389
• http://origami64.net/showthread.php?tid=114
• http://n64.icequake.net/doc/n64intro/kantan/step2/4-3.html
• http://jul.rustedlogic.net/thread.php?id=1234
• http://level42.ca/projects/ultra64/Documentation/man/pro-man/pro25/25-01.html
• http://wiki.spinout182.com/w/F3DEX2
• http://wiki.cloudmodding.com/oot/F3DZEX/Opcode_Details