GOOL Interpreter

From Crash Bandicoot Hacking Wiki
Revision as of 16:57, 5 August 2015 by Wikia>Chekwob (Crash 2: Fill in table (mostly unknown))
Jump to navigation Jump to search

In the Crash games, the GOOL Interpreter is responsible for interpreting an object's executable bytecode. This bytecode instructs the interpreter to perform a specific sequence of operations that can conditionally or unconditionally alter specific characteristics of the object (ex. appearance, location, status, etc.), and should ultimately render the object according to its changing characteristics. As a result, the player observes a lively, animated, and interactive enemy, box, collectible item, or whatever other kind of in-game object it may be.

Interpretation

A single interpretation of an object's bytecode consists of the following steps:

  1. Fetch the instruction at the object's program counter location
    • Read the instruction at the object's program counter location
    • Point the object's program counter to the location of the following instruction
  2. Perform the operation specified by the fetched instruction
  3. Continue repeating steps 1 and 2 until a suspending instruction is fetched

(A suspending instruction is typically either a RET, which marks the end of a block, or ANIF/ANIS, which change the object's current frame of animation.)

A single call to the GOOL interpreter routine with some desired object as its first argument results in a single interpretation of that object's bytecode. The GOOL bytecode interpreter routine expects an object that provides the following information:

  • The location in its bytecode at which the interpreter shall begin its interpretation = object's program counter
  • The location in its memory of a stack frame created for the interpretation = object's frame pointer
  • The location in its memory at which the interpreter shall begin pushing the results of/popping the operands for interpreted instructions = object's stack pointer

Prior to an interpretation, each of these fields (program counter, frame pointer, stack pointer) is modified according to the object's current state [descriptor] and the type of interpretation.

Types of interpretation

Many separate interpretations of an object's bytecode can occur within a single frame. The GOOL interpreter routine can be called with the same object from up to 7 distinct locations in a single iteration of the game loop. Each of the 7 calls to the interpreter routine with that object, and thus the ensuing interpretations of its bytecode, occurs under a separate condition. The 7 interpretations are identified based on the conditions under which they occur, and are as follows:

  • One that occurs at a specified rate, i.e. after a specific amount of time has elapsed since its previous occurrence. (Code Block interpretation)
  • One that [always] occurs [for each frame]. (Trans Block interpretation)
  • One that occurs whenever the object is sent an event and its current state specifies the location of an event service routine. (Event Block interpretation)
  • One that occurs whenever the object is sent an event and: its current state does not specify an event service routine or its event service routine failed to return an event, and the event sent maps to an event handler. (Handles Block interpretation)
  • One that occurs whenever the object changes state. (Head Block interpretation)
  • One that occurs whenever the object changes state and its recent Trans Block interpretation was unsuccessful (*Trans Block interpretation)
  • One that occurs when the object minimizes the distance during another object's query to 'find the nearest object', and the event sent for asynchronous handling (from the querying object) maps to an event handler. (*Handles Block interpretation)

Ignoring the details for now, the latter 2 types are just a Trans Block and Handles Block interpretation, respectively, although their calls cross-reference the interpreter routine from different locations. Thus, there are a total of 5 types. These 5 types are more simply viewed as 5 separate threads of execution of the object's bytecode. To demonstrate, here is the expanded game loop routine with only the calls to routines that have a descendant call to the interpreter routine:

 sub_80011FC4(levelid)
 { 
   ...	  
   C) sub_80011DD0 - Load entries and/or create universal game objects 
                     (HUD/display, Crash, Aku Aku, Shadows, Boxes, Fruit)
     1) sub_8001C6C8 - CreateObject() [called once for each object created]
     ...
   GAME LOOP:
   {
     1) Code for handling game pause/start button press
       A) If start pressed...
          i) If pause menu object does not exist....
             1) sub_8001C6C8 - CreateObject() [create pause menu object]
             ...
             else
             1) sub_80024040 - SendEvent() [destroy pause menu object]
             ...
     2) If Crash object does not exist: 
       A) sub_8002E98C - Create HUD and initialize level
          i) sub_8001C6C8 - CreateObject() [create HUD object]
          ii)sub_80026650 - Reinitialize Level
             1) sub_8001C6C8 - CreateObject() [create Crash object]
             ...
     3) Code for loading a new level if necessary
       A) If new level to load
          ...
          ii)sub_80011DD0 - Load entries and/or create universal game objects 
             1) sub_8001C6C8 - CreateObject() [called once for each object created]
             ...
     ...
     5) Spawn all objects in current zone [for those that have their respawn bit set]
       A) sub_80025928 - Spawn objects
          1) sub_8001BCC8 - SpawnObject() [called for each object spawned]
          ...
     ...
     9) Update all objects and create transformed primitives for them
       A) sub_8001D5EC - UpdateObjects()
          1) sub_8001DA0C - UpdateObject() [called for each existing object]
     ...
   } 
 }

Based on the above, consider an arbitrary frame of execution in terms of a single object. CreateObject() or SpawnObject() will not be called more than once [to create that object] in that frame; that is-the object is not created or spawned more than once, or even spawned after it is created and vice-versa. The object may not even be created/spawned in that frame because it either already exists or does not exist since the game has not requested it be created/spawned. If it is an existing object, it will also be updated only once in that frame. For a single object, the above can be reduced the following series of potential calls:

 sub_80011FC4(levelid)
 { 
   ...	  
   GAME LOOP:
   {
     ...
     CreateObject(obj,...) or SpawnObject(obj,...)
     ...
     SendEvent(src,obj,...)
     ...
     UpdateObject(obj)
   } 
 }

What the above leaves out, however, are the number of other locations at which a call to the SendEvent() routine reside. Some are located before the CreateObject or SpawnObject routine calls, some after, some within the UpdateObject routine itself, and even some within the interpreter routine, since there exist several types of GOOL instructions that allow another object to send an event to the recipient object. In fact, there is nothing preventing the object from being the recipient of more than one event sent (from potentially multiple sources) in a single frame. There may be an interpretation for each event sent to the object during that frame, and there is no limit on the number of events that can be sent. Thus, there may be many calls to SendEvent() [with the object as the recipient] in a single frame, but still only at most one call to CreateObject/SpawnObject() and UpdateObject(). Without getting too far off track, it is simply necessary to understand that each of these routines contains (or calls a routine that calls a routine that contains) a call to the interpreter routine:

 sub_80011FC4(levelid)
 { 
   ...	  
   GAME LOOP:
   {
     ...
     CreateObject(obj,...) or SpawnObject(obj,...)
       InitObject(obj, ...)
         ChangeObjectState(obj, ...)
           CreateObjectStackFrame(obj); // *create initial stack frame
           -if there is a head block to interpret for the object
             CreateObjectStackFrame(obj);
             obj->pc = obj->pchead;
             InterpretObject(obj, 0x13, &stateref); // # 1 (head block interpretation)
         
     SendEvent(src,obj,...) // there may be multiple calls to this
        -if there is an event service routine for the object (in its current state)
           CreateObjectStackFrame(obj);
           obj->pc = obj->pcevent;
           InterpretObject(obj, 0x8, &stateref);    // # 2 (event block interpretation)
        -if there is not an event service routine for the object (in its current state)
         or the event service routine failed to return an event
          -if the event maps to an event handler in the event->state/handler map
           CreateObjectStackFrame(obj);
           obj->pc = handlerlocation;
           InterpretObject(obj, 0x3, &stateref);    // # 3 (handles block interpretation)
          -else
           ChangeObjectState(obj,state,...); // *

     UpdateObject(obj)
      -if the object is able to animate
        -if there is a trans block to interpret for the object (in its current state)
           CreateObjectStackFrame(obj);
           obj->pc = obj->pctrans;
           InterpretObject(obj, 0x3, &stateref);   // # 4 (trans block interpretation)
        -if at least n ticks have elapsed, where n is given by the 'wait' operand of the 
         most recently suspending animation type instruction of the object's code block 
           InterpretObject(obj, 0x4, &stateref);   // # 5 (code block interpretation)

   } 
 }

 

Notice that immediately prior to each call [to the interpreter routine], with the exception of #5, a new stack frame is created for the object (which includes pointing its frame pointer and stack pointer, respectively, to the beginning and ending of its new frame), and its program counter is pointed to a distinct location within its bytecode. If each bytecode instruction is represented by a pseudo-instruction, a potential sequence of calls to the interpreter routine in a single frame [for only that object] might be represented by the following:

label  | address | pseudo-instruction

pcevent: 0x34      N  // call A
         0x38      N
         0x3C      S
...
pchead:   0x0      N  // call B
          0x4      N
          0x8      N
          0xC      S
...
pctrans: 0x50      N  // call C
         0x54      N
         0x58      N
         0x5C      N
         0x60      N
         0x64      N
         0x68      S
...
*pccode: 0x70      N  // call D
         0x74      N
         0x78      N
         0x7C      N
         0x80      S
...
pcevent: 0x34      N  // call E
         0x38      N
         0x3C      S
...
pcevent: 0x34      N  // call F
         0x38      N
         0x3C      S

Each call above to the interpreter routine is represented by the sequence of instructions (represented as pseudo-instructions) fetched during its interpretation. An N pseudo-instruction represents an instruction that did not suspend the interpreter, causing the next instruction in sequence to be fetched and interpreted. An S pseudo-instruction represents an instruction that did suspend the interpreter, causing the interpreter routine to end the interpretation/return to its caller. Each pseudo-instruction is listed with the address of the instruction it represents in the object's bytecode [relative to the beginning].

The first instruction fetched in each interpretation is also marked with a label; this label is given the name of the object field that points to its associated instruction. Immediately prior to each interpretation, with the exception of the one marked *pccode, the object's program counter is pointed to the location in this particular field; it is this operation that causes each interpretation to begin at the location it does. This particular object field is also primarily responsible for determining the type of interpretation:

Field Interpretation Type
pctrans Trans Block Interpretation
pcevent Event Block Intepretation
pchead Head Block Interpretation
pc Code Block Interpretation

The pctrans, pcevent, and pchead fields each contain a pointer to/the absolute location of an individual block in the object's bytecode. These fields locate the respective entry points for 3 separate threads of interpretation for the object-a trans thread, an event thread, and a head thread. Whenever the object changes state, its pctrans and pcevent fields are modified according to the state descriptor for its new state; its pchead field is then cleared. (If the object wishes to modify its pchead field, it does so via a MOVC instruction in its bytecode.)

Whenever the object changes state, its program counter (pc field) is also set directly to the location of the code block specified by the pccode field of the state descriptor for its new state. The entirety of the object's stack contents are then unwound, or the object's stack pointer is reset to its initial location, and an initial stack frame is [re]created. This program counter location and stack configuration is then restored after each subsequent non-code block interpretation is completed. (That is-since those interpretations require the program counter value to be replaced with a new location and a new stack frame to be created, after they are completed, the original program counter location and initial stack frame are restored). When the interpretation marked #5 in the above pseudo-code (the code block interpretation) finally occurs, the object's program counter will point to its code block and its stack frame will be the initial frame. The code block interpretation is yet another thread of interpretation for the object-a code thread.

For the sake of explanation, handles blocks are viewed as being equivalent to event blocks, so handles interpretations and threads are simply ignored. The timeline to the right shows an arbitrary frame of execution in terms of the times elapsed for the various interpretations of a single object's bytecode. Towards the beginning of the frame, the interpreter has performed a code block interpretation and a trans block interpretation for the object. Towards the end of the frame, the interpreter has performed an event block interpretation for that object, as it has been sent an event from some [unknown] source. The following timeline extends this timeline [to the right] to 6 arbitrary frames of execution:


The extended timeline makes it clear that this particular object has been configured to have its code block interpretation occur every other frame. It also shows the performance of more than one event block interpretation for that object during the second frame. The following timeline shows the same 6 frames of execution in terms of the times elapsed for all interpretations of each existing object's bytecode; in this particular instance, there are only 2 existing objects.

Stack Frame (incomplete)

A new stack frame is created for an object immediately prior to an interpretation of its bytecode. It is also created when jumping to and linking a bytecode routine via the JAL instruction. An object's initial stack frame is created when it enters its initial state, and is recreated for each subsequent state change (after the unwinding the stack). The initial stack frame is used for the object's code block interpretation. At any given point in time, an object's frame pointer and stack pointer, respectively, locate the beginning and end of its current stack frame.

Format

A stack frame has the following format:

Offset Field Size Value
-0x4 x a **Argument a 4 bytes *
0x0 Preserved Interpreter Mode Flags 4 bytes 0xFFFF for initial frame; * otherwise
0x4 Preserved Program Counter 4 bytes *
0x8 Preserved Frame Pointer Relative Offset 2 bytes f
0xA Preserved Stack Pointer Relative Offset 2 bytes s
0xC Frame Data * x 4 bytes *

(**Arguments come before frames and are not actually 'part' of them)

Initial Stack Frame

When the object is first created/spawned, or when it changes state, its program counter is pointed to the code block with offset pccode (given by the state descriptor for its initial or new state), its frame pointer is cleared, and its stack pointer is reset to its initial location:

The object's initial stack frame is then [re]created by pushing the following sequence of values to the stack:

  • 0x0000FFFF - Interpreter Mode Flags. Not an actual interpreter mode flags value but rather an indicator that an unwinding of this frame shall not restore interpreter mode flags.
  • obj->pc - Program Counter. At this point, it points to the code block for the object's initial/new state.
  • (((unsigned long)obj->fp - (unsigned long)&obj->self) << 16) | ((unsigned long)obj->sp - (unsigned long)&obj->self)) Frame Pointer and Stack Pointer relative offsets. At this point, frame pointer (obj->fp) is 0 and stack pointer points to its initial location; this combined value should be of the form 0x0000****.

The object's frame pointer is then pointed to the beginning of that frame. With these values having been pushed to its stack, the object's stack pointer now points to the end of that frame. Thus, immediately following the instantiation or state change, the object's current stack frame is its initial stack frame.

During the object's code block interpretation, this frame is then extended with additional data, including stack operands and/or the results of interpreted instructions.

Non-Initial Stack Frame

After an object's initial stack frame is created, but before its next code block interpretation occurs, several other non-code block interpretations may be performed. Immediately prior to performing any of these interpretations, a new stack frame is created on top of the initial stack frame, and then the current program counter value (which contains the code block location) is replaced with the location of the non-code block. This frame is created by pushing the following sequence of values to the stack:

  • modeflags - Interpreter Mode Flags
  • obj->pc - Program Counter. At this point, it points to the code block for the object's initial/new state.
  • (((unsigned long)obj->fp - (unsigned long)&obj->self) << 16) | ((unsigned long)obj->sp - (unsigned long)&obj->self)) Frame Pointer and Stack Pointer relative offsets. At this point, frame pointer points to the beginning of the initial frame, and stack pointer points to the end of the initial frame.

The object's frame pointer is then pointed to the beginning of that frame. With these values having been pushed to its stack, the object's stack pointer now points to the end of that frame. Thus, immediately following the creation of the frame, it is the object's current stack frame-that is, the initial stack frame is no longer the current stack frame.

When the following interpretation finishes, the object's program counter is restored to the code block location preserved in the current frame, and the preserved frame pointer and stack pointer offsets are used to restore the initial stack frame as the current frame.

UNFINISHED

Interpreter Mode Flags

TBD

GOOL Instructions

Crash 1

The GOOL Interpreter in Crash 1 recognizes a total of 58 different GOOL instructions: 

  • 6 Arithmetic Instructions (incl SHA*) 
  • 4 Non-bitwise Logical/Logical Comparison Instructions 
  • 4 Bitwise Logical Instructions 
  • 5 Comparison Instructions 
  • 5 Data Transfer Instructions 
  • 8 Mathematical Function Instructions 
  • 4 Control Flow Instructions 
  • 19 Miscellaneous/Multi-Purpose Instructions 

GOOL Instruction Tables

Main Instruction Table

The GOOL instruction table below lists the opcode, mnemonic/name, format, operands (based on explicit and/or implicit specification of source origin and/or destination target), general operation (represented with c-style statement), and a description for each GOOL instruction.  

Opcode Name Encoding/Format Explicit GOOL ops

in

Explicit IMM. ops

in

Implicit STACK ops

in

Operation Implicit STACK out Explicit

GOOL ops out

Description
0/0x00 ADD 00000000RRRRRRRRRRRRLLLLLLLLLLLL L,R O = L + R O add
1/0x01 SUB 00000001RRRRRRRRRRRRLLLLLLLLLLLL L,R O = L - R O subtract
2/0x02 MUL 00000010RRRRRRRRRRRRLLLLLLLLLLLL L,R O = L * R O multiply
3/0x03 DIV 00000011RRRRRRRRRRRRLLLLLLLLLLLL L,R O = L / R O divide
4/0X04 CEQ 00000100RRRRRRRRRRRRLLLLLLLLLLLL L,R O = (L == R),

O = ((L ^ R) == 0)

O check if equal
5/0x05 ANDL 00000101RRRRRRRRRRRRLLLLLLLLLLLL L,R O = (L && R),

O = (L ? (R > 0) : 0)

O logical and
6/0x06 ORL 00000110RRRRRRRRRRRRLLLLLLLLLLLL L,R O = (L || R) O
logical or
7/0x07 ANDB 00000111RRRRRRRRRRRRLLLLLLLLLLLL L,R O = L & R O bitwise and
8/0x08 ORB 00001000RRRRRRRRRRRRLLLLLLLLLLLL L,R O = L | R O bitwise or
9/0x09 SLT 00001001RRRRRRRRRRRRLLLLLLLLLLLL L,R O = L < R O set less than
10/0x0A SLE 00001010RRRRRRRRRRRRLLLLLLLLLLLL L,R O = (L <= R) O set less than or equal
11/0x0B SGT 00001011RRRRRRRRRRRRLLLLLLLLLLLL L,R O = L > R O set greater than
12/0x0C SGE 00001100RRRRRRRRRRRRLLLLLLLLLLLL L,R O = (L >= R) O set greater than or equal
13/0x0D MOD 00001101RRRRRRRRRRRRLLLLLLLLLLLL L,R O = L % R O modulo
14/0x0E XOR 00001110RRRRRRRRRRRRLLLLLLLLLLLL L,R O = L ^ R O exclusive or
15/0x0F TST 00001111RRRRRRRRRRRRLLLLLLLLLLLL L,R O = (((L & R) ^ R) == 0) O test bit
16/0x10 RND 00010000AAAAAAAAAAAABBBBBBBBBBBB A,B O = B+(rand() % (A - B)) O random
17/0x11 MOVE 00010001SSSSSSSSSSSSDDDDDDDDDDDD S D = S [D] move data
18/0x12 NOTL 00010010SSSSSSSSSSSSDDDDDDDDDDDD S D = (S == 0) D logical not
19/0x13 PATH 00010011AAAAAAAAAAAABBBBBBBBBBBB (A),B R = 0x100 [R,A] varies P B path progress
20/0x14 LEA 00010100SSSSSSSSSSSSDDDDDDDDDDDD S D = &S D load effective address
21/0x15 SHA 00010101RRRRRRRRRRRRLLLLLLLLLLLL L,R O = ((R < 0) ? (L >> -R)

: (L << R))

O arithmetic shift
22/0x16 PSHV 00010110AAAAAAAAAAAABBBBBBBBBBBB [A,[B]] [arg_buf = A]

I = A; J = B;

[I,[J]] push value to stack
23/0x17 NOTB 00010111SSSSSSSSSSSSDDDDDDDDDDDD D = ~S D bitwise not
24/0x18 MOVC 000110000000RRRRRRIIIIIIIIIIIIII R,C see docs O move code pointer
25/0x19 ABS 00011001SSSSSSSSSSSSDDDDDDDDDDDD S D = (S < 0) ? -S: S D absolute value
26/0x1A PAD 00011010000TDDDDSSPPBBBBBBBBBBBB B,P,S,D,T O = testctrls(instr,0) O test controller buttons
27/0x1B SPD 00011011VVVVVVVVVVVVBBBBBBBBBBBB V,B S = B + ((V*gvel) >> 10) S calculate speed
28/0x1C MSC 00011100PPPPSSSSSLLLXXXXXXXXXXXX X P,S,L various; see docs *** *** multi-purpose
29/0x1D PRS 00011101PPPPPPPPPPPPDDDDDDDDDDDD P,D large calc; see docs O driven sine wave
30/0x1E SSAW 00011110DDDDDDDDDDDDPPPPPPPPPPPP M,P O = (M + frameCount) % P O synchronized saw wave
31/0x1F RGL 00011111000000000000IIIIIIIIIIII I O = globals[I >> 8] O read global variable
32/0x20 WGL 00100000SSSSSSSSSSSSIIIIIIIIIIII I,S globals[I << 8] = *S write global variable
33/0x21 ANGD 00100001RRRRRRRRRRRRLLLLLLLLLLLL L,R O = angdist(L,R) angle between
34/0x22 APCH 00100010RRRRRRRRRRRRLLLLLLLLLLLL L,(R) S = 0x100 [R,S] O = approach(L,R,S) O approach a value
35/0x23 CVMR 00100011000IIIIIILLL000000000000 I,L O = obj.link[L].colors[I] color vector or matrix read
36/0x24 CVMW 00100011000IIIIIILLLCCCCCCCCCCCC C I,L obj.link[L].colors[I] = C color vector or matrix write
37/0x25 ROT 00100101RRRRRRRRRRRRLLLLLLLLLLLL L,(R) [R,S] O = rotate(L,R,S,0) O approach an angle
38/0x26 PSHP 001001100AAAAAAAAAAABBBBBBBBBBBB [A,[B]] I = &A; J = &B; [I,[J]] push pointer to stack
39/0X27 ANID 00100111FFFFFFFFFFFFDDDDDDDDDDDD F D=&obj.global.anim[F>>5] D set animation
128/0x80 DBG 10000000RRRRRRRRRRRRLLLLLLLLLLLL L,R debug print ops (beta only)
129/0x81 NOP 10000001000000000000000000000000 no operation
130/0x82 CFL 10000010TTCCRRRRRRIIIIIIIIIIIIII T,C,R,I general control flow
BRA 100000100000RRRRRRVVVVIIIIIIIIII R,V,I branch
BNEZ 100000100001RRRRRRVVVVIIIIIIIIII R,V,I branch not equal zero
BEQZ 100000100010RRRRRRVVVVIIIIIIIIII R,V,I branch equal zero
CST 100000100100RRRRRRSSSSSSSSSSSSSS R,S change state
CSNZ 100000100101RRRRRRSSSSSSSSSSSSSS R,S change state not zero
CSEZ 100000100110RRRRRRSSSSSSSSSSSSSS R,S change state equal zero
RET 100000101000RRRRRRIIIIIIIIIIIIII R,I return
131/0x83 ANIS 10000011HHTTTTTTSSSSSSSSSFFFFFFF F,S,T,H W change animation sequence
132/0x84 ANIF 10000100HHTTTTTTFFFFFFFFFFFFFFFF F T,H W change animation frame
133/0x85 VECA 10000101CCCTTTBBBAAAVVVVVVVVVVVV V T,A,B,C * ** multi-purpose vector calcs
134/0x86 JAL 10000110VVVV000000IIIIIIIIIIIIII I,V jump and link
135/0x87 EVNT 10000111LLLAAARRRRRREEEEEEEEEEEE E L,A,R send an event
136/0x88 RSTT 10001000TTCCRRRRRR************** R,C,T,* **
state return guard = true variant
137/0x89 RSTF 10001001TTCCRRRRRR************** R,C,T,* ** state return guard = false variant
138/0x8A CHLD 10001010AAAATTTTTTTTSSSSSSCCCCCC C,T,S,A [C],

arg[0 to A]

spawn children objects
139/0x8B NTRY 10001011TTTTTTTTTTTTEEEEEEEEEEEE T,E multi-purpose page operation
140/0x8C SNDA 10001100AAAAAAAAAAAABBBBBBBBBBBB A,B adjust audio levels
141/0x8D SNDB 10001101VVVVRRRRRRSSSSSSSSSSSSS S V,R
play sound
effect
142/0x8E VECB 10001110CCCTTTBBBAAAVVVVVVVVVVVV V T,A,B,C
multi-purpose vector calcs
143/0x8F EVNB 10001111LLLAAARRRRRREEEEEEEEEEEE E L,A,R
broadcast an event
144/0x90 EVNU 10010000LLLAAARRRRRREEEEEEEEEEEE E L,A,R

send event unknown variant
145/0x91 CHLF 10100000AAAATTTTTTTTSSSSSSCCCCCC C,T,S,A [C],

arg[0 to A]


spawn children objects;

no replacement if obj pool full

Note that a few instructions can not be represented with a single mnemonic; operations for these instructions are not only determined by their opcode but also their primary operation subtype and possibly secondary operation subtype fields. These instructions have their own tables, which are given in the following sections.

State Return Instruction Table
Opcode Name Encoding/Format Explicit

IMM. ops

in

Description
136/0x88 RSTT 10001000TTCCRRRRRR************** R,C,T,* state return guard = true variant
RST 100010000100RRRRRRSSSSSSSSSSSSSS R,S state return guard = true
RSNT 100010000101RRRRRRSSSSSSSSSSSSSS R,S state return if nonzero guard = true
RSZT 100010000110RRRRRRSSSSSSSSSSSSSS R,S state return if equal zero guard = true
RSCT 100010000111RRRRRRSSSSSSSSSSSSSS R,S state return eval prev cond guard = true
RNT 100010001000RRRRRRxxxxxxxxxxxxxx R null return guard = true
RNNT 100010001001RRRRRRxxxxxxxxxxxxxx R null return if nonzero guard = true
RNZT 100010001010RRRRRRxxxxxxxxxxxxxx R null return if equal zero guard = true
RNCT 100010001011RRRRRRxxxxxxxxxxxxxx R null return eval prev cond guard = true
GDT 100010001100RRRRRRxxxxxxxxxxxxxx R guard = true
GNT 100010001101RRRRRRxxxxxxxxxxxxxx R if nonzero guard = true
GZT 100010001110RRRRRRxxxxxxxxxxxxxx R if equal zero guard = true
GCT 100010001111RRRRRRxxxxxxxxxxxxxx R eval prev cond guard = true
GBNT 100010000001RRRRRRVVVVIIIIIIIIII R,V,I if nonzero guard = true else branch
GBZT 100010000010RRRRRRVVVVIIIIIIIIII R,V,I if equal zero guard = true else branch
137/0x89 RSTF 10001001TTCCRRRRRR************** R,C,T,* state return guard = false variant
RSF 100010000100RRRRRRSSSSSSSSSSSSSS R,S state return guard = false
RSNF 100010000101RRRRRRSSSSSSSSSSSSSS R,S state return if nonzero guard = false
RSZF 100010000110RRRRRRSSSSSSSSSSSSSS R,S state return if equal zero guard = false
RSCF 100010000111RRRRRRSSSSSSSSSSSSSS R,S state return eval prev cond guard = false
RNF 100010001000RRRRRRxxxxxxxxxxxxxx R null return guard = false
RNNF 100010001001RRRRRRxxxxxxxxxxxxxx R null return if nonzero guard = false
RNZF 100010001010RRRRRRxxxxxxxxxxxxxx R null return if equal zero guard = false
RNCF 100010001011RRRRRRxxxxxxxxxxxxxx R null return eval prev cond guard = false
GDF 100010001100RRRRRRxxxxxxxxxxxxxx R guard = false
GNF 100010001101RRRRRRxxxxxxxxxxxxxx R if nonzero guard = false
GZF 100010001110RRRRRRxxxxxxxxxxxxxx R if equal zero guard = false
GCF 100010001111RRRRRRxxxxxxxxxxxxxx R eval prev cond guard = false
GBNF 100010000001RRRRRRVVVVIIIIIIIIII R,V,I if nonzero guard = false else branch
GBZF 100010000010RRRRRRVVVVIIIIIIIIII R,V,I if equal zero guard = false else branch
Operand Name Table

Specification Format

A AAAAAAAAAAAA AAAA AAA (EVNT/EVNU/EVNB) AAA (VECA/VECB)
Name Value A Argument Count Argument Count Vector A Index
Specification Format B BBBBBBBBBBBB BBBBBBBBBBBB (PAD) BBBBBBBBBBBB (SPD) BBB
Name Value B Controller Buttons Base Speed Vector B Index
Specification Format C CCCCCCCCCCCC CCCCCC CCC CC
Name Color Value Spawn Count Vector C Index Conditional Check Type
Specification Format D DDDDDDDDDDDD DDDDDDDDDDDD (PRS) DDDD
Name Destination Wave Phase Directional Buttons
Specification Format E EEEEEEEEEEEE (NTRY) EEEEEEEEEEEE (EVNT/EVNU/EVNB)
Name Entry Event
Specification Format F FFFFFFFFFFFFFFFF FFFFFFFFFFFF FFFFFFF
Name Animation Frame Animation Descriptor Offset Animation Frame

Specification Format

H HH
Name Horizontal Flip

Specification Format

I IIIIIIIIIIIIII IIIIIIIIIIII IIIIIIIIII IIIIII
Name Immediate Code Location Global Variable Index Immediate Branch Offset Color Index

Specification Format

L LLLLLLLLLLLL LLL
Name Left Operand Object Link Index

Specification Format

P PPPPPPPPPPPP PPPP PP
Name Wave Period Primary Operation Subtype Primary Check Type

Specification Format

R RRRRRRRRRRRR RRRRRR
Name Right Operand Object Register Index

Specification Format

S SSSSSSSSSSSSSS SSSSSSSSSSSS SSSSSSSSS SSSSSS
Name State Source Animation Sequence Index Subtype

Specification Format

SSSSS SS
Name Secondary Operation Subtype Secondary Check Type

Specification Format

T TTTTTTTTTTTT TTTTTTTT TTTTTT TTT
Name Operation Subtype (Object) Type Time Operation Subtype

Specification Format

TT T
Name Operation Subtype Truth Invert Toggle

Specification Format

V VVVVVVVVVVVV VVVV VVVV (SNDB)
Name Velocity Variable Count Volume

Crash 2

The GOOL Interpreter in Crash 2 recognizes a total of 79 GOOL instructions, including two duplicates. Many of the instructions are unknown. The basic arithmetic and logical operations are preserved the same as in Crash 1, but most others have been changed.

Crash 2 GOOL code may also contain sections of machine code for the MIPS processor architecture (used by the playstation).

GOOLv2 Instruction Table

The GOOL instruction table below lists the opcode, mnemonic/name, format, operands (based on explicit and/or implicit specification of source origin and/or destination target), general operation (represented with c-style statement), and a description for each GOOL instruction.

Opcode Name Encoding/Format Explicit GOOL ops

in

Explicit IMM. ops

in

Implicit STACK ops

in

Operation Implicit STACK out Explicit

GOOL ops out

Description
0/0x00 ADD 00000000RRRRRRRRRRRRLLLLLLLLLLLL L,R O = L + R O add
1/0x01 SUB 00000001RRRRRRRRRRRRLLLLLLLLLLLL L,R O = L - R O subtract
2/0x02 MUL 00000010RRRRRRRRRRRRLLLLLLLLLLLL L,R O = L * R O multiply
3/0x03 DIV 00000011RRRRRRRRRRRRLLLLLLLLLLLL L,R O = L / R O divide
4/0X04 CEQ 00000100RRRRRRRRRRRRLLLLLLLLLLLL L,R O = (L == R),

O = ((L ^ R) == 0)

O check if equal
5/0x05 ANDL 00000101RRRRRRRRRRRRLLLLLLLLLLLL L,R O = (L && R),

O = (L ? (R > 0) : 0)

O logical and
6/0x06 ORL 00000110RRRRRRRRRRRRLLLLLLLLLLLL L,R O = (L || R) O
logical or
7/0x07 ANDB 00000111RRRRRRRRRRRRLLLLLLLLLLLL L,R O = L & R O bitwise and
8/0x08 ORB 00001000RRRRRRRRRRRRLLLLLLLLLLLL L,R O = L | R O bitwise or
9/0x09 SLT 00001001RRRRRRRRRRRRLLLLLLLLLLLL L,R O = L < R O set less than
10/0x0A SLE 00001010RRRRRRRRRRRRLLLLLLLLLLLL L,R O = (L <= R) O set less than or equal
11/0x0B SGT 00001011RRRRRRRRRRRRLLLLLLLLLLLL L,R O = L > R O set greater than
12/0x0C SGE 00001100RRRRRRRRRRRRLLLLLLLLLLLL L,R O = (L >= R) O set greater than or equal
13/0x0D MOD 00001101RRRRRRRRRRRRLLLLLLLLLLLL L,R O = L % R O modulo
14/0x0E XOR 00001110RRRRRRRRRRRRLLLLLLLLLLLL L,R O = L ^ R O exclusive or
15/0x0F TST 00001111RRRRRRRRRRRRLLLLLLLLLLLL L,R O = (((L & R) ^ R) == 0) O test bit
16/0x10 RND 00010000AAAAAAAAAAAABBBBBBBBBBBB A,B O = B+(rand() % (A - B)) O random
17/0x11 MOVE 00010001SSSSSSSSSSSSDDDDDDDDDDDD S D = S [D] move data
18/0x12 NOTL 00010010SSSSSSSSSSSSDDDDDDDDDDDD S D = (S == 0) D logical not
19/0x13 PATH 00010011AAAAAAAAAAAABBBBBBBBBBBB (A),B R = 0x100 [R,A] varies P B path progress (unconfirmed for c2)
20/0x14 LEA 00010100SSSSSSSSSSSSDDDDDDDDDDDD S D = &S D load effective address (unconfirmed for c2)
21/0x15 SHA 00010101RRRRRRRRRRRRLLLLLLLLLLLL L,R O = ((R < 0) ? (L >> -R)

: (L << R))

O arithmetic shift
22/0x16 PSHV 00010110AAAAAAAAAAAABBBBBBBBBBBB [A,[B]] [arg_buf = A]

I = A; J = B;

[I,[J]] push value to stack (unconfirmed for c2)
23/0x17 NOTB 00010111SSSSSSSSSSSSDDDDDDDDDDDD D = ~S D bitwise not
24/0x18 MOVC 000110000000RRRRRRIIIIIIIIIIIIII R,C see docs O move code pointer (unconfirmed for c2)
25/0x19 ABS 00011001SSSSSSSSSSSSDDDDDDDDDDDD S D = (S < 0) ? -S: S D absolute value
26/0x1A ? 00011010???????????????????????? unknown
27/0x1B ? 00011011???????????????????????? unknown
28/0x1C ? 00011100???????????????????????? unknown
29/0x1D ? 00011101???????????????????????? unknown
30/0x1E ? 00011110???????????????????????? unknown
31/0x1F ? 00011111???????????????????????? unknown
32/0x20 ? 00100000???????????????????????? unknown
33/0x21 ? 00100001???????????????????????? unknown
34/0x22 ? 00100010???????????????????????? unknown
35/0x23 ? 00100011???????????????????????? unknown
36/0x24 ? 00100100???????????????????????? unknown
37/0x25 ? 00100101???????????????????????? unknown
38/0x26 ? 00100110???????????????????????? unknown
39/0x27 ? 00100111???????????????????????? unknown
40/0x28 ? 00101000???????????????????????? unknown
41/0x29 ? 00101001???????????????????????? unknown
42/0x2A ? 00101010???????????????????????? unknown
43/0x2B ? 00101011???????????????????????? unknown
44/0x2C ? 00101100???????????????????????? unknown
45/0x2D ? 00101101???????????????????????? unknown
46/0x2E ? 00101110???????????????????????? unknown
47/0x2F ? 00101111???????????????????????? unknown
48/0x30 ? 00110000???????????????????????? unknown
49/0x31 ? 00110001???????????????????????? unknown
50/0x32 ? 00110010???????????????????????? unknown
51/0x33 ? 00110011???????????????????????? unknown
52/0x34 ? 00110100???????????????????????? unknown
53/0x35 ? 00110101???????????????????????? unknown
54/0x36 ? 00110110???????????????????????? unknown
55/0x37 ? 00110111???????????????????????? unknown
56/0x38 ? 00111000???????????????????????? unknown
57/0x39 ? 00111001???????????????????????? unknown
58/0x3A ? 00111010???????????????????????? unknown
59/0x3B ? 00111011???????????????????????? unknown
60/0x3C ? 00111100???????????????????????? unknown
61/0x3D ? 00111101???????????????????????? unknown
62/0x3E ? 00111110???????????????????????? unknown
63/0x3F ? 00111111???????????????????????? unknown
64/0x40 ? 01000000???????????????????????? unknown
65/0x41 ? 01000001???????????????????????? unknown
66/0x42 ? 01000010???????????????????????? unknown
67/0x43 ? 01000011???????????????????????? unknown
68/0x44 ? 01000100???????????????????????? unknown
69/0x45 ? 01000101???????????????????????? unknown
70/0x46 ? 01000110???????????????????????? unknown
71/0x47 ? 01000111???????????????????????? unknown
72/0x48 ? 01001000???????????????????????? unknown
73/0x49 ? 01001001???????????????????????? unknown
74/0x4A ? 01001010???????????????????????? unknown
75/0x4B ? 01001011???????????????????????? unknown
76/0x4C ? 01001100???????????????????????? unknown
77/0x4D ? 01001101???????????????????????? unknown
78/0x4E ? 01001110???????????????????????? unknown

GOOL Instruction Operands

...

  • In/Source type operands
    • Explicitly specified (i.e. origin = determined by operand's corresponding bitfield from the instruction)
      • GOOL in/source operand
      • Immediate operand
    • Implicitly specified (i.e. origin = top of object's stack; implicitly popped from top of object's stack)
      • Stack in/source operand
  • Out/Destination type operands
    • Explicitly specified (i.e. target = determined by operand's corresponding bitfield from the instruction)
      •  GOOL out/destination operand
    • Implicitly specified (i.e. target = top of object's stack; implicitly pushed to top of object's stack)
      • Stack out/destination operand/result*

*technically not an 'operand', but rather a result of the instruction's overall operation; it is implied that instructions that use [an] operand(s) of this type always pushes it/them (i.e the result(s)) to the top of the object's stack.

Explicit Operands

A GOOL instruction's explicit operands are explicitly specified in its lower 24 bits (operand portion). Each explicitly specified operand is either an immediate operand, or a GOOL operand.

Immediate Operands

An immediate operand directly specifies its value in its corresponding bitfield (within the operand portion) of the instruction. The number of immediate operands and their positions/bit-lengths varies between instructions.

Immediate Operands are in operands (source operands).

GOOL Operands

Conversely, a GOOL operand is specified as a reference to its actual value or destination location. Each GOOL operand is 12 bits in length; a GOOL instruction may require the specification of at most 2 GOOL operands, within the upper and/or lower halves of its operand portion, respectively. The 12 bit GOOL reference that specifies a GOOL operand encodes a [32 bit] pointer to its corresponding source value and/or destination location. Operations for instructions that require GOOL operand specification first translate the corresponding GOOL reference(s) to yield their decoded pointer(s) before accessing pointed data.

See the GOOL Reference Translation section for more information.

Depending on how a GOOL operand is used by its instruction, it may be an in type (source) operand, an out type (destination) operand, or both (inout operand). If an instruction only reads the [input] value located by its translated GOOL operand (a pointer), it is an in type (source) operand. If an instruction only writes a result/value to the location of its translated GOOL operand, it is an out type (destination) operand. If an operand meets both of these criteria it is an inout operand.

A GOOL operand can be translated to yield, among several other types of pointers, a pointer to any of the object's fields; for example, this pointer could then be used by the instruction operation to read the value at its [implied] source location and/or to write a value to its implied destination location, deeming the operand a source GOOL operand and/or a destination GOOL operand, respectively. A GOOL operand can also be translated to yield a pointer to the value at the top of the objects stack for, before either decrementing or incrementing its stack pointer, a location that can be dereferenced for reading or writing, effectively popping from or pushing to its stack [based on whether it is a GOOL source operand or a GOOL destination operand]. In these cases, the instruction operand is said to be explicitly popped from or pushed onto the objects stack. 

Implicit Operands

Implicit In Operands

Some GOOL instructions may require the implicit specification of one or more (implicit) in operands. In operands are implicitly specified prior to an instruction's interpretation by pushing their value(s)-which must remain the topmost values at interpretation-to the object's stack. It is implied that these so-called implicit operands will exist on the stack for the instruction to pop and use as input.

Implicit Out Operands

For most GOOL instructions, the net effect of their operation involves an implicit push of one or more results to the object's stack. That is-it is implied or expected that these instructions will ultimately push their results to the object's stack.


....

out type operands - By the operation of some specific type of instruction that requires some 

out-type operand to be specified, data will ultimately be written to [the unsigned long of memory 

at] that operand's 'translated location'. Such data is an expected result of that specific type of 

[instruction's] operation. By the operation of an instruction with specific type/opcode that must 

specify multiple out-type operands, its corresponding expected result(s) will ultimately be written 

at the respective translated location(s) for each specified operand. 


GOOL Reference Translation

The following table lists the encoding format, potential type name, and a descriptive type name for each of the possible GOOL reference types:

Format Potential Type Name Descriptive Type Name
00RRRRRRRRRR Ireg Ref object local executable data pool reference
01RRRRRRRRRR Pool Ref object external executable data pool reference
100IIIIIIIII Int Ref constant mult of 16  (range: -8192 - 8176)
1010IIIIFFFF Frac Ref constant mult of 256 (range: -32768 - 32512)
10110SSSSSSS Stack Ref object frame argument/local variable reference [peek stack @ location relative to fp]
101111100000 Null Ref null reference (translates to 0)
101111110000 Sp-Double Ref double stack pop reference (translates to 1)
110LLLMMMMMM Reg Ref object link memory reference
111MMMMMMMMM (M != 0x1F) Var Ref object field/memory reference
111000011111 Stack Pop/Push stack pop/push reference

Potential GOOL Reference type names have been deduced based on what appears to be an unreferenced array of GOOL interpreter debug strings at 0x103B8.

UNFINISHED

Given an instruction with specific type/opcode whose operation expects the [explicit] specification 

of at least one GOOL operand (i.e. at least one operand listed in table under explicit GOOL ops in 

and/or explicit GOOL ops out for that instruction type/opcode), and with a specific one of its GOOL 

operands specified as a particular sequence of bits in its corresponding bitfield (12 bit region) 

within the operand portion of the instruction, the following can be used to determine, in that 

instruction's operation, the specific steps taken for the necessary translation of that operand to 

its pointer of a specific type. When specified as some particular sequences of bits, the operand's 

translation can result in a different operation based on whether the instruction table lists that 

operand [for the accompanying instruction type/opcode] as an 'in' or an 'out' type GOOL operand. 


i.) 111000011111 - object stack [pop/push]


in  - in-type operands of this bit sequence will return a pointer to the top of the object's stack/stack 

pointer prior to popping the value there/decrementing the object's stack pointer. A stack pop will 

result from the translation's decrementing of the object's stack pointer and the operation's 

dereferencing of the returned/translated pointer for a read; this popped value is then used by the 

operation as the operand's actual/translated value. 


out - out-type operands of this bit sequence will return a pointer to the top of the object's stack/stack 

pointer prior to its extension/incrementing the object's stack pointer. A stack push will result 

from the translation's incrementing of the object's stack pointer and the operation's dereferencing 

of the returned/translated pointer for a write. The data written at the location of a returned 

pointer/pushed to the stack by an instruction's operation is an expected result of the specific type 

of operation. 


-- 


ii.) 111AAAAAAAAA - object field/memory (A != 0x1F) 


A = object field index (domain = 0    to 0x1FF; 

                         range = 0x60 to 0x860, R=4N) 


A significant amount of an object's fields lie within its 'process' structure; this structure is 

located at a 0x60 byte offset from the beginning of the object's structure. Note that the last of 

the fields in the object's process structure also includes its array of/address space for its 

[local] memory where its stack contents are stored. 


Operands of the bit sequence 111AAAAAAAAA will return a pointer to the 'A'th object field in 

relation to an object's process structure; because object fields are 4 bytes/an unsigned long in 

length and an object's process structure is located at a 0x60 byte offset from the beginning of its 

structure, then the formula: 


  0x60 + (A * 4) 


gives the byte offset of the actual field pointed to by the translation's returned pointer, and 

consequently the actual field referred to by such an operand. The object structure definition listed 

in previous sections includes byte offsets for all object fields within the object structure. 


Notice that if A [bits] is specified as A = 0x1F, then the bit sequence will have the form 

111000011111; this is, however, the bit sequence whose translation shall instead be processed for a 

stack pop/push. If, however, such a sequence were not checked and ultimately processed for a stack 

pop/push prior to passing the test for the bit sequence 111AAAAAAAAA with A = 0x1F, then the byte 

offset determined by translation of such an operand- 0x60 + (A * 4) = 0x60 + (0x1F * 4) = 0xDC = 

byte offset of object's 'stack pointer' field -would refer to the object's stack pointer field. As 

this is arguably a favourable design choice, it should not be possible to directly modify an 

object's stack pointer field via translation of specified operands-only indirectly, should the 

translation of an such an instruction's GOOL operand when specified in the form 111000011111 

[111AAAAAAAAA with A = 0x1F] modify the object's stack pointer as a result of the appropriate push 

and/or pop. 


in - in-type operands of this bit sequence will return a pointer to the corresponding object field; 

this value is then used by the operation as the operand's actual/translated value. 


out - out-type operands of this bit sequence will return a pointer to the corresponding object 

field. Data will ultimately be written to this location as an expected result of the specific type 

of operation. 


--


iii.) 110BBBAAAAAA


A = [link] object field index (domain =     0 to  0x3F;

                               range  =  0x60 to 0x15C, R=4N)


B = object link index         (domain =     0 to 7;

                               range  =  0x60 to 0x7C,  R=4N)  

Operands of the bit sequence 110BBBAAAAAA will return a pointer to the 'A'th' object field in 

relation to the object's 'B'th' link object's process structure. An object's 'link objects' are 

referred to by the 8 object pointers at the beginning of its process structure; the first of these 

pointers is the object's 'self link', the second a pointer to its parent object, the third a pointer 

to its sibling object, and so on. The A bits in the sequence specify the offset of a field in 

unsigned longs from the link object's process structure; the B bits in the sequence specify the 

index of one of the object's 8 link objects that contains the process structure which the returned 

object field pointer's location will be calculated relative to. Thus, translation of such a bit 

sequence will return a pointer to any of an object's link object's process structure's fields. 


in - in-type operands of this bit sequence will return a pointer to the corresponding object field; 

this value is then used by the operation as the operand's actual/translated value. 


out - out-type operands of this bit sequence will return a pointer to the corresponding object 

field. Data will ultimately be written to this location, where such data is an expected result of 

some specific type of [instruction's] operation. For an instruction that can specify multiple 

out-type operands (with specific type/opcode), by the instruction's operation, its corresponding 

expected result(s) will ultimately be written at the respective translated location(s) for each 

operand specified in this format/with this bit sequence. 


--

iv.) 00AAAAAAAAAA - object global executable static data


A = object executable static data offset (domain =   0 to 0x3FF;

                                          range  =   0 to 0xFFC, R=4N)


A GOOL executable's static data (GOOL_data) item is essentially a table of static, predefined 

'magic' constants that can be used by GOOL operations as the translated values of operands with this 

bit sequence. For example, it includes the EIDs of any entries, preconfigured bitfields, values 

corresponding to specific bits to test, audio sample parameters (volume, pan, etc), specific speeds 

or velocities, and other constants utilized by the operations of instructions in the executable's 

GOOL_code item. It is necessary that a translation can yield these constants for the values of some 

GOOL operands since, as it will be seen, the only other types of translation that can yield a 

specific constant are restricted to yielding constants in certain ranges. 


[Also, in the case that the bit sequence types for the latter translation can be used in operands, 

if the operations for instructions [with the operands] translate them to the same constant [i.e. the 

same bit sequence is specified for the operands in each instruction] and utilize that constant in 

the same fashion, then if an equal change in those operations is desired as some modification during 

development it would generally involve changing the specified sequence for -each- operand in each 

instruction. Then, it is beneficial to keep a table of constants, since the translations for 

different instructions' operands that yield a reference the same constant in the table will, when 

changing the single unsigned long/constant in the table, reference the same -changed- constant in 

the table.] 


Operands of the bit sequence 00AAAAAAAAAA will return a pointer to the static, predefined constant 

at the offset specified by 'A bits' within/relative to the object's global GOOL executable entry's 

static data [item]; that is-a translation of this type will return a pointer to the constant at an 

offset of 'A' unsigned longs relative to the third (static data) 'item' (unsigned long array) within 

the entry structure referred to by the objects 'global' field. 


in - in-type operands of this bit sequence will return a pointer to the corresponding location in 

the object's global executable [entry's] static data [item]; the unsigned long/constant at this 

location is then used by the operation as the operand's actual/translated value. 


out - out-type operands of this bit sequence will return a pointer to the corresponding location in 

the object's global executable [entry's] static data [item]; the operation ultimately replaces the 

constant at this location with its respective unsigned long result (designated for the translated 

location of that operand.) A write to the executable's static data should generally be avoided 

unless a global change to that constant, and therefore an equal change in behavior for each object 

that is an instance of that executable, is desired. 


--


Note the distinction from the object's 'global' GOOL entry from it's 'external' GOOL entry. The 

global entry may or may not contain the byte code [item] that will be interpreted in the object's 

current state; it contains the corresponding state definition that specifies the GOOL executable 

entry which contains the actual GOOL byte code item with the instructions that will be interpreted. 

In the object's current state, its 'external' entry will refer to that specified entry. 


--

v.) 01AAAAAAAAAA - object external executable static data


A = object executable static data offset (domain =   0 to 0x3FF;

                                          range  =   0 to 0xFFC, R=4N)


Operands of the bit sequence 01AAAAAAAAAA will return a pointer to the constant (unsigned long) at 

the offset specified by 'A bits' within/relative to the object's external GOOL executable entry's 

static data [item]; that is-a translation of this type will return a pointer to the constant at an 

offset of 'A' unsigned longs relative to the third (static data) 'item' (unsigned long array) within 

the entry structure referred to by the objects 'external' field. 


in - in-type operands of this bit sequence will return a pointer to the corresponding location in 

the object's external executable [entry's] static data [item]; the unsigned long/constant at this 

location is then used by the operation as the operand's actual /translated value. 


out - out-type operands of this bit sequence will return a pointer to the corresponding location in 

the object's external executable [entry's] static data [item]; the operation ultimately replaces the 

constant at this location with its respective unsigned long result (designated for the translated 

location of that operand.) A write to the executable's static data should generally be avoided 

unless a global change to that constant, and therefore an equal change in behavior for each object 

that is an instance of that executable, is desired. 

--

vi.) 100AAAAAAAAA - constant mult of 16 


A = constant that will be multiplied by 16 (domain:      0 to  0x1FF)

                                           ( range:  -8192 to   8176; R =   16N)

                                      << 4 (       -0x2000 to 0x1FF0; R = 0x10N)

Operands of the bit sequence 100AAAAAAAAA will return a pointer to a value that is written as one of 

two double-buffered signed long 'constants' in their 2 respective buffers located at offsets 0x40 

and 0x44 in scratch memory, as referenced by the pointer at memory location 0x56480 [gp$(0x44)]. 

Translation for operands of this bit sequence will first write the value (16*A)-that is, 16 times 

the value specified by 'A bits'-to the 'currently active constant buffer'. The currently active 

constant buffer is indicated by either a 0 or a 1 in the unsigned long at memory location 0x56484 

[gp$(0x48)], which refers to either the buffer at offset 0x40 or 0x44, respectively, in scratch 

memory. The currently inactive constant buffer at any instant is always the buffer that is NOT the 

currently active constant buffer. The translation then proceeds by changing the currently inactive 

constant buffer to the currently active constant buffer (that is, if 0 previously indicated the 

currently active constant buffer, then this is changed to a 1; conversely, if 1, it is changed to a 

0). The translation finally returns a pointer to the (16*A) signed long value that had been written 

to the previously active buffer, at which point is the currently inactive buffer. 

The double-buffering design allows the sufficient specification of a maximum of 2 (constant) GOOL 

operands [with this bit sequence] per instruction, since each of the 2 buffers will keep track of 1 

and therefore an overall of 1+1=2 [translated] constants to be used by operations of instructions 

with the corresponding operands. 


A can be specified as a negative constant with the appropriate 2's complement format: 


A = 1NNNNNNNN (-N): V = 11111111111111111111NNNNNNNN0000 (16 * -N)

A = 0NNNNNNNN ( N): V = 00000000000000000000NNNNNNNN0000 (16 *  N)


Where V indicates the binary signed long that will be written to the appropriate constant buffer for 

the respective specified formats of A. 


in - in-type operands of this bit sequence will return a pointer to the [currently inactive constant 

buffer after storing the] appropriate constant value (A * 16) [in the active constant buffer and 

changing the active constant buffer to the currently inactive constant buffer]; the unsigned 

long/instruction at this location is then used by the operation as the operand's actual/ translated 

value. 


out-type operands of this bit sequence are useless: they will return a pointer to the currently 

inactive constant buffer after storing the appropriate constant value (A * 16) in the active 

constant buffer and changing the active constant buffer to the currently inactive constant buffer, 

and that stored constant value in the [then currently] inactive constant buffer will ultimately be 

overwritten by the corresponding expected result of the operation. However, since the data in the 

constant buffer can be only accessed by translation of operands with this bit sequence, which 

involves the prior overwriting of the active constant buffer with (A*16), then that result can never 

be accessed [without first being overwritten]. 


--


vii.) 1010AAAAAAAA - constant mult of 256 


A = constant that will be multiplied by 256 (domain:      0 to   0xFF)

                                            ( range: -32768 to  32512; R =   256N)

                                       << 8 (       -0x8000 to 0x7F00; R = 0x100N)

Operands of the bit sequence 1010AAAAAAAA are translated almost identically to those of bit sequence 

type vi; the only difference is in the calculation for the constant written to the respective 

buffer: 


A = 1NNNNNNN (-N): V = 1111111111111111NNNNNNNN00000000 (256 * -N)

A = 0NNNNNNN ( N): V = 0000000000000000NNNNNNNN00000000 (256 *  N)


Where V indicates the binary signed long that will be written to the appropriate constant buffer for 

the respective specified formats of A. 


in - in-type operands of this bit sequence will return a pointer to the [currently inactive constant 

buffer after storing the] appropriate constant value (A * 256) [in the active constant buffer and 

changing the active constant buffer to the currently inactive constant buffer]; the unsigned 

long/instruction at this location is then used by the operation as the operand's actual/ translated 

value. 


out-type operands of this bit sequence are useless for the same reasons those of sequence type vi 

are. 


--


viii.) 10110AAAAAAA - object frame argument/local variable [peek stack @ location relative to fp]


A = argument/local variable offset, relative to object FP (domain =      0 to 0x7F)

                                                          (range  = -0x100 to 0xFC, -40 args to +39 local vars)


Immediately prior to creating a new stack frame for the object, whether it be the initial stack 

frame [recreated due to a state change], the frame for a non-code thread, or the frame created for a 

linked subroutine, a number of 'arguments' [for the new frame] can be pushed to the object's stack 

to occupy the locations that will immediately precede that of the new frame. And, after the frame is 

created, with further interpretation, the results of operations are pushed to the object's stack to 

extend that frame; usually, these results will have a particular, expected order on the object's 

stack based on the expected results of the operations for the interpreted sequence of instructions. 

Then, conceptually, these results are said to be the contents of 'local variables' within that stack 

frame. 


Operands of the bit sequence 10110AAAAAAA will return a pointer to the [unsigned long/data at the] 

location on the object's stack of the data that is 'A' unsigned longs/units of data (4*A bytes) 

relative to the object's current frame pointer. Typically, A is specified as a negative value/offset 

when a pointer to some argument is desired; this is because the arguments are located -before- the 

start of the stack frame. A is specified as a positive offset (>3) when a pointer to some local 

variable (i.e. operation result) is desired. 


A can be specified as a negative offset with the appropriate 2's complement format: 


A = 1NNNNNN (-N): O = 111111111111111111111111NNNNNN00 (4 * -N)  (argument offset)

A = 0NNNNNN ( N): O = 000000000000000000000000NNNNNN00 (4 *  N)  (local variable offset)


in - in-type operands of this bit sequence will return a pointer to the [unsigned long/data at the] 

location on the object's stack of the data that is 'A' unsigned longs/units of data (4*A bytes) 

relative to the object's current frame pointer. The value at this location is then used by the 

operation as the operand's actual/translated value. 


out - out-type operands of this bit sequence will return a pointer to the [unsigned long/data at 

the] location on the object's stack of the data that is 'A' unsigned longs/units of data (4*A bytes) 

relative to the object's current frame pointer. Data will ultimately be written to this location as 

an expected result of the specific type of operation, overwriting either the argument or local 

variable [at that location]. 

--


ix. 10111110**** - 'false' constant (translates to 0)

   (101111100000 = 0xBE0)

Operands of the bit sequence 10111110**** (where * indicates the possibility of either a 1 or 0) 

will, rather than return a pointer to some memory location, return the value 0 (i.e. invalid pointer 

to memory location 0). When specified, operands of this bit sequence will cause the operations of 

most instructions to return without writing/storing any results or making any calculations. 


For a few other instructions, operands of the sequence will cause the operations to 'implicitly' pop 

a number of additional operands from the object's stack. For the few other instructions, prior to [a 

potentially worthless] translation of some operand, if the operand has been specified in the format 

0xBE0 (i.e. the usual specification for this sequence), its translation will be skipped. Then, in 

exchange for the lack of a translated value for that operand, its value, and possibly the values for 

a number of additional operands [depends on the particular instruction's operation], will instead be 

popped off of the object's stack. The additional operands will then be factored in an alternative, 

extended calculation for that instruction by the operation. Usually, rather than having to 

alternatively perform some extended calculation, when the operand is -not- specified in this format, 

the operation -sets- default values for each of the 'additional operands'. The same specific 

calculations will be performed in the operation for the instruction, factoring in the same operand 

values [including those of the 'additional operands'], but the default values set for the additional 

operands can be 'overridden' with the specification of this bit sequence for the operand that would 

cause an implicit popping of their values from the object's stack. 


in  - in-type operands of this bit sequence will return a null pointer (pointer to memory location 

0). If specified, when translated by the operation for most instructions, the remainder of the 

operation will be skipped. For few other instructions the translation for an operand of this 

specific sequence will be skipped; this is followed with a pop of the actual value for that operand 

and a potential pop of the [overriding] values for an additional number of operands that could 

otherwise not be explicitly or implicitly specified for the resolution of their values, which 

therefore are otherwise directly set to some instruction-specific defaults. 


out-type operands of this bit sequence will return a null pointer; a trap will result with the 

attempt to access the memory at that location (0), which lies in the kernel code segment. Out-type 

operands should therefore not be specified with this bit sequence. 


--


x.  10111111**** - 'true' constant (translates to 1)

   (101111110000 = 0xBF0)

Operands of the bit sequence 10111111**** (where * indicates the possibility of either a 1 or 0) 

will, rather than return a pointer to some memory location, return the value 1 (i.e. invalid pointer 

to memory location 1). When specified, operands of this bit sequence will cause the operations of 

most instructions to trap the system; this is due to their attempt to access the memory at the 

operands translated 'location' (1), which lies in the kernel code segment. 


For a few other instructions, operands of the sequence will cause the operations to alternatively, 

implicitly pop a number of additional operands from the object's stack, just like described for 

those of sequence type ix. However, the difference in sequence type x from type ix is that only the 

operations for a completely different, disjoint set of instructions exhibits this behavior-that is, 

operations for only some instructions will behave this way when the operand is specified as 0xBE0; 

conversely, operations for only some other instructions will behave this way when the operand is 

specified as 0xBF0. 


To better understand this behavior, consider the instruction PRW: when its explicitly specified 

operand J is specified with any bit sequence type other than type xi or x, the instruction's 

operation will appropriately translate that operand to resolve its actual value; its operand I 

(which at that point cannot be directly or indirectly specified by the instruction itself) will then 

take on a default value of 0x100. The operation will then perform all the appropriate calculations 

with those operand values, ultimately storing one result to the translated location of operand X and 

pushing another result P to the object's stack. However, when its explicitly specified operand J is 

specified with bit sequence type x (i.e. BF0), the instruction's operation will instead: skip 

translation of operand J, implicitly pop the value for operand I off of the object's stack, and 

resolve operand J's value by additionally [implicitly] popping it off the object's stack. The 

operation will then perform all the appropriate calculations with -those- operand values. 


Thus, the '[]' brackets around an explicit GOOL operand for an instruction in the table indicate 

that, when specified in one format of either sequence type ix or x (i.e. 0xBE0 or 0xBF0) but not the 

other, the instruction's operation performs differently: that same operand and any additional 

operands if '[]' bracketed and listed under implicit STACK operands (for the instruction in the 

table) will instead have their values popped off of the object's stack in the order of their 

listing. When the explicitly specified operand is -not- of bit sequence type ix or x, the 

alternative is to resolve its value by translation, and the values of any 'additional' operands that 

would have otherwise been popped off the object's stack are instead set to the corresponding 

'default' values as indicated under explicit constant operands for the instruction in the table. 


in - in-type operands of this bit sequence for a few instructions will skip their translation; this 

is followed with a pop of the actual value for that operand and a potential pop of the [overriding] 

values for an additional number of operands that could otherwise not be explicitly or implicitly 

specified for the resolution of their values, needing to be directly set to some instruction- 

specific defaults. When translation is not skipped, as in the operations of most other instructions, 

this bit sequence will return an invalid pointer (pointer to memory location 1). When specified, 

operands of this bit sequence will cause the operations of most instructions to trap the system; 

this is due to their attempt to access the memory at the operands translated 'location' (1), which 

lies in the kernel code segment. 


out-type operands of this bit sequence will return an invalid pointer; a trap will result with the 

attempt to access the memory at that location (1), which lies in the kernel code segment. Out-type 

operands should therefore not be specified with this bit sequence. 


--


...