# Characteristics of Nintendo 64 compiler output Or: heuristics for distinguishing them. ## General compiler stuff ### IDO * Unconditional branches use `b`. * Float literals can be loaded with `lui` followed by `mtc1` or from rodata with `lui` and `lwc1`. These can be reordered among other instructions in the function. (It seems to never use `lui`, `ori`, `mtc1` but this needs to be verified) * This is the only compiler where an instruction that isn't `mfhi` or `mflo` can follow a `break 6`. * The expansion of division and modulus for ido in general allows for reorderings that the other compilers don't have, but this is the easiest to check. * (-g only) The compiler will emit `b after; nop; after:` frequently. These are the words 0x10000001 0x00000000. ### KMC GCC (2.7.2) * Unconditional branches use `j`. * Float literals will always be loaded with a sequence of `lui`, an optional `ori`, and `mtc1` with the GPR used for all three being $at. These are always directly sequential. * `break 6` will always be followed by `mfhi` or `mflo`. `break 7` will always be followed by `mfhi`, `mflo`, or `addiu $at, $zero, -0x1` (exactly). ### GCC 2.8(?) * Function epilogue (`addiu $sp, $sp, X`, `jr $ra`) can be omitted after an infinite loop. ### SN64 * Unconditional branches use `j`. * Float literals will always be loaded with a sequence of `lui` and `lwc1` with the GPR being used for both being $at. These are always directly sequential. * `mtc1 $at, ...` is *never* emitted. * `break 6` will always be followed by `mfhi` or `mflo`. `break 7` will always be followed by `mfhi`, `mflo`, or `addiu $at, $zero, -0x1` (exactly). ## MIPS levels ### MIPS I * requires a `nop` after certain loads (TODO: which?) ### MIPS III * An unencumbered `-mips3` (not patched as in e.g. KMC) will use `daddiu` etc. for loads (see e.g. iQue) # Entrypoints ## makerom Original assembly looks like ```mips la $8 <BssStart> li $9 <bssSize> 1: sw $0, 0($8) sw $0, 4($8) addi $8, 8 addi $9, 0xfff8 bne $9, $0, 1b la $29 <bootStack> la $10 <bootEntry> j $10 ``` The symbols in `<>` are all written as numeric constants. IDO will expand `la` to `(lui)/addiu` and `li` to `(lui)/ori` in such cases; `j` expands to `jr`. ## mild ```mips la $8,_%sSegmentBssStart la $9,_%sSegmentBssSize 1: sw $0, 0($8) sw $0, 4($8) addi $8, 8 addi $9, 0xfff8 bne $9, $0, 1b la $10,%s # bootEntry la $29,%s # bootStack jr $10 ``` (not 100% confirmed but very likely from reading it from the binary). The instructions all use *symbols*, which GCC expands as `lui/addiu`. ## SN64 There are several versions of the entrypoint for SN64. * The main distinguishing common factors between them is that they all have `li $sp, 0x803FFFF0` (aka `lui $sp, 0x803F` followed by `ori $sp, $sp, 0xFFF0`) and that they use a `jal` for calling the bootproc instead of `jr`. * `break` also appears frequently, although some custom versions do not use it.
Last changed by  
 
EllipticEllipsis
114

Read more

IDO compilation flowchart

digraph{ c [label="C code .c", shape=parallelogram] p [label="Pascal code .p", shape=parallelogram] s [label="Assembly .s", shape=parallelogram] ucode [label="UCode", shape=parallelogram] cc[shape=square] c -> cc p -> cc s -> cc

5/4/2023
Ocarina of Time Gateway changes

Introduction The build date of the newly-discovered Gateway rom is 99-01-19 11:37:17. This is later than every released Nintendo 64 version (the last of these being in November 1998): the last NTSC version (commonly referred to as 1.2 or NN2) was built on 98-11-12, and the last PAL version (the second PAL version, referred to as NP1) on 98-11-18. Summary The vast majority of the rom is the same as NN2. A small number of the changes made between NN2 and NP1 are present, but plenty of others are not (such as the Gerudo who gives you Gerudo card's clothes, which are patched to be stay a single colour in NP1 and later). These are noted below. The most visible change is that the new Gerudo symbol is used instead of the original crescent and star, in both scene model textures and actor model textures. This change was originally thought to come from Majora's Mask first; this rom is now the earliest known build of a Zelda game to use it.[^gerudo] Most interestingly, this also includes textures in Dampe's race and Iron Knuckles, which are not patched in any other known rom version. The file choice screen is changed to only display a single file, and no copy/erase functionality. A new file (which we may guess is called gateway_title_static) contains 3 new textures for this, a Start button, and textures for Please Select., Press A button to decide., Press B button to exit.. After entering a name, the game starts immediately END is pressed. The Save option is disabled in the pause menu, and the Game Over screen goes straight to the "Continue playing?" choice, skipping the save screen. Since these are the only way to save, saving is impossible. No text is changed, so the tutorial signs still explain how to save the game. Rumble is completely disabled (so just as in the Virtual Console release, the Stone of Agony is completely useless).

4/26/2023
Majora's Mask Gateway changes

Introduction The build date of the newly-discovered Gateway rom is 00-08-08 09:25:41. This is between the US retail version (00-07-31, referred to as NE0) and the first PAL version (00-09-25, referred to as NP0) Summary Most of the rom is the same as NE0. A number of patches/changes previously only known in NP0 and after are also present, surprising considering that this rom was built only 10 days after NE0. Several unused functions are removed from various files, as they are in NP0. Several patches from NP0 (in z_eventmgr, z_kaleido_setup), but not all. The file choice screen is changed to only display a single file, and no copy/erase functionality. A new file (which we may guess is called gateway_title_static) contains 3 new textures for this, a Start button, and textures for Please Select., Press A button to decide., Press B button to exit.. The Options screen is changed to remove the Dolby logo. After entering a name, the game starts immediately END is pressed. Other standard gateway-compliant changes (removing rumble capability, adding and enabling the pause/unpause code).

4/25/2023
Bitfields in IDO

Bitfields Bitfields are a stupid[^stupid] C optimisation to pack unaligned variables made of a small or odd number of bits. Their memory layout is unspecified, the notation looks like [^stupid]: See, e.g. https://lwn.net/Articles/478657/ struct example1 { u32 m1 : 1; u32 m2 : 4; u32 m3 : 7; };

4/7/2023
Read more from EllipticEllipsis
Published on HackMD