admin/mruby-mruby
1
0
Fork 0
mirror of https://github.com/mruby/mruby synced 2026-06-08 16:11:16 +00:00
Code Issues Packages Projects Releases Wiki Activity
Files
eedc95460c2ffa545d951fb22e75b560832f08eb
mruby-mruby/doc/internal/opcode.md
T
Yukihiro "Matz" Matsumoto b841bd7439 doc/internal/opcode.md: document OP_GETIDX/OP_SETIDX optimization 
Add notes section explaining the optimization behavior:
- Which functions are used for direct access
- When fallback to method dispatch occurs
- Why subclasses can override []/[]=

Co-authored-by: Claude <noreply@anthropic.com>
2025-12-25 18:01:44 +09:00

164 lines
8.8 KiB
Markdown
Raw Blame History

<!-- summary: About mruby Virtual Machine Instructions -->
# The new bytecode
We will reimplement the VM to use 8bit instruction code. By
bytecode, we mean real byte code. The whole purpose is
reducing the memory consumption of mruby VM.
# Instructions
Instructions are bytes. There can be 256 instructions. Currently, we
have 106 instructions. Instructions can take 0 to 3 operands.
## operands
The size of operands can be either 8bits, 16bits or 24bits.
In the table.1 below, the third field describes the size
of operands.
- B: 8bit
- S: 16bit
- W: 24bit
If the first and second operands are of type `B` (8bits), they may be
extended to 16bits by the operand extension instruction immediately
preceding them.
See also `OP_EXT1`, `OP_EXT2` and `OP_EXT3`.
## table.1 Instruction Table
| No. | Instruction Name | Operand type | Semantics
| --: | ---------------- | ------------ | ---------------
| 0 | `OP_NOP` | `-` | `no operation`
| 1 | `OP_MOVE` | `BB` | `R(a) = R(b)`
| 2 | `OP_LOADL` | `BB` | `R(a) = Pool(b)`
| 3 | `OP_LOADI8` | `BB` | `R(a) = mrb_int(b)`
| 4 | `OP_LOADINEG` | `BB` | `R(a) = mrb_int(-b)`
| 5 | `OP_LOADI__1` | `B` | `R(a) = mrb_int(-1)`
| 6 | `OP_LOADI_0` | `B` | `R(a) = mrb_int(0)`
| 7 | `OP_LOADI_1` | `B` | `R(a) = mrb_int(1)`
| 8 | `OP_LOADI_2` | `B` | `R(a) = mrb_int(2)`
| 9 | `OP_LOADI_3` | `B` | `R(a) = mrb_int(3)`
| 10 | `OP_LOADI_4` | `B` | `R(a) = mrb_int(4)`
| 11 | `OP_LOADI_5` | `B` | `R(a) = mrb_int(5)`
| 12 | `OP_LOADI_6` | `B` | `R(a) = mrb_int(6)`
| 13 | `OP_LOADI_7` | `B` | `R(a) = mrb_int(7)`
| 14 | `OP_LOADI16` | `BS` | `R(a) = mrb_int(b)`
| 15 | `OP_LOADI32` | `BSS` | `R(a) = mrb_int((b<<16)+c)`
| 16 | `OP_LOADSYM` | `BB` | `R(a) = Syms(b)`
| 17 | `OP_LOADNIL` | `B` | `R(a) = nil`
| 18 | `OP_LOADSELF` | `B` | `R(a) = self`
| 19 | `OP_LOADT` | `B` | `R(a) = true`
| 20 | `OP_LOADF` | `B` | `R(a) = false`
| 21 | `OP_GETGV` | `BB` | `R(a) = getglobal(Syms(b))`
| 22 | `OP_SETGV` | `BB` | `setglobal(Syms(b), R(a))`
| 23 | `OP_GETSV` | `BB` | `R(a) = Special[Syms(b)]`
| 24 | `OP_SETSV` | `BB` | `Special[Syms(b)] = R(a)`
| 25 | `OP_GETIV` | `BB` | `R(a) = ivget(Syms(b))`
| 26 | `OP_SETIV` | `BB` | `ivset(Syms(b),R(a))`
| 27 | `OP_GETCV` | `BB` | `R(a) = cvget(Syms(b))`
| 28 | `OP_SETCV` | `BB` | `cvset(Syms(b),R(a))`
| 29 | `OP_GETCONST` | `BB` | `R(a) = constget(Syms(b))`
| 30 | `OP_SETCONST` | `BB` | `constset(Syms(b),R(a))`
| 31 | `OP_GETMCNST` | `BB` | `R(a) = R(a)::Syms(b)`
| 32 | `OP_SETMCNST` | `BB` | `R(a+1)::Syms(b) = R(a)`
| 33 | `OP_GETUPVAR` | `BBB` | `R(a) = uvget(b,c)`
| 34 | `OP_SETUPVAR` | `BBB` | `uvset(b,c,R(a))`
| 35 | `OP_GETIDX` | `B` | `R(a) = R(a)[R(a+1)]`
| 36 | `OP_SETIDX` | `B` | `R(a)[R(a+1)] = R(a+2)`
| 37 | `OP_JMP` | `S` | `pc+=a`
| 38 | `OP_JMPIF` | `BS` | `if R(a) pc+=b`
| 39 | `OP_JMPNOT` | `BS` | `if !R(a) pc+=b`
| 40 | `OP_JMPNIL` | `BS` | `if R(a)==nil pc+=b`
| 41 | `OP_JMPUW` | `S` | `unwind_and_jump_to(a)`
| 42 | `OP_EXCEPT` | `B` | `R(a) = exc`
| 43 | `OP_RESCUE` | `BB` | `R(b) = R(a).isa?(R(b))`
| 44 | `OP_RAISEIF` | `B` | `raise(R(a)) if R(a)`
| 45 | `OP_SSEND` | `BBB` | `R(a) = self.send(Syms(b),R(a+1)..,R(a+n+1):R(a+n+2)..) (c=n\|k<<4)`
| 46 | `OP_SSENDB` | `BBB` | `R(a) = self.send(Syms(b),R(a+1)..,R(a+n+1):R(a+n+2)..,&R(a+n+2k+1))`
| 47 | `OP_SEND` | `BBB` | `R(a) = R(a).send(Syms(b),R(a+1)..,R(a+n+1):R(a+n+2)..) (c=n\|k<<4)`
| 48 | `OP_SENDB` | `BBB` | `R(a) = R(a).send(Syms(b),R(a+1)..,R(a+n+1):R(a+n+2)..,&R(a+n+2k+1))`
| 49 | `OP_CALL` | `-` | `self.call(*, **, &) (But overlay the current call frame; tailcall)`
| 50 | `OP_SUPER` | `BB` | `R(a) = super(R(a+1),... ,R(a+b+1))`
| 51 | `OP_ARGARY` | `BS` | `R(a) = argument array (16=m5:r1:m5:d1:lv4)`
| 52 | `OP_ENTER` | `W` | `arg setup according to flags (23=m5:o5:r1:m5:k5:d1:b1)`
| 53 | `OP_KEY_P` | `BB` | `R(a) = kdict.key?(Syms(b))`
| 54 | `OP_KEYEND` | `-` | `raise unless kdict.empty?`
| 55 | `OP_KARG` | `BB` | `R(a) = kdict[Syms(b)]; kdict.delete(Syms(b))`
| 56 | `OP_RETURN` | `B` | `return R(a) (normal)`
| 57 | `OP_RETURN_BLK` | `B` | `return R(a) (in-block return)`
| 58 | `OP_BREAK` | `B` | `break R(a)`
| 59 | `OP_BLKPUSH` | `BS` | `R(a) = block (16=m5:r1:m5:d1:lv4)`
| 60 | `OP_ADD` | `B` | `R(a) = R(a)+R(a+1)`
| 61 | `OP_ADDI` | `BB` | `R(a) = R(a)+mrb_int(b)`
| 62 | `OP_SUB` | `B` | `R(a) = R(a)-R(a+1)`
| 63 | `OP_SUBI` | `BB` | `R(a) = R(a)-mrb_int(b)`
| 64 | `OP_MUL` | `B` | `R(a) = R(a)*R(a+1)`
| 65 | `OP_DIV` | `B` | `R(a) = R(a)/R(a+1)`
| 66 | `OP_EQ` | `B` | `R(a) = R(a)==R(a+1)`
| 67 | `OP_LT` | `B` | `R(a) = R(a)<R(a+1)`
| 68 | `OP_LE` | `B` | `R(a) = R(a)<=R(a+1)`
| 69 | `OP_GT` | `B` | `R(a) = R(a)>R(a+1)`
| 70 | `OP_GE` | `B` | `R(a) = R(a)>=R(a+1)`
| 71 | `OP_ARRAY` | `BB` | `R(a) = ary_new(R(a),R(a+1)..R(a+b))`
| 72 | `OP_ARRAY2` | `BBB` | `R(a) = ary_new(R(b),R(b+1)..R(b+c))`
| 73 | `OP_ARYCAT` | `B` | `ary_cat(R(a),R(a+1))`
| 74 | `OP_ARYPUSH` | `BB` | `ary_push(R(a),R(a+1)..R(a+b))`
| 75 | `OP_ARYSPLAT` | `B` | `R(a) = ary_splat(R(a))`
| 76 | `OP_AREF` | `BBB` | `R(a) = R(b)[c]`
| 77 | `OP_ASET` | `BBB` | `R(b)[c] = R(a)`
| 78 | `OP_APOST` | `BBB` | `*R(a),R(a+1)..R(a+c) = R(a)[b..]`
| 79 | `OP_INTERN` | `B` | `R(a) = intern(R(a))`
| 80 | `OP_SYMBOL` | `BB` | `R(a) = intern(Pool(b))`
| 81 | `OP_STRING` | `BB` | `R(a) = str_dup(Pool(b))`
| 82 | `OP_STRCAT` | `B` | `str_cat(R(a),R(a+1))`
| 83 | `OP_HASH` | `BB` | `R(a) = hash_new(R(a),R(a+1)..R(a+b*2-1))`
| 84 | `OP_HASHADD` | `BB` | `hash_push(R(a),R(a+1)..R(a+b*2))`
| 85 | `OP_HASHCAT` | `B` | `R(a) = hash_cat(R(a),R(a+1))`
| 86 | `OP_LAMBDA` | `BB` | `R(a) = lambda(Irep(b),OP_L_LAMBDA)`
| 87 | `OP_BLOCK` | `BB` | `R(a) = lambda(Irep(b),OP_L_BLOCK)`
| 88 | `OP_METHOD` | `BB` | `R(a) = lambda(Irep(b),OP_L_METHOD)`
| 89 | `OP_RANGE_INC` | `B` | `R(a) = range_new(R(a),R(a+1),FALSE)`
| 90 | `OP_RANGE_EXC` | `B` | `R(a) = range_new(R(a),R(a+1),TRUE)`
| 91 | `OP_OCLASS` | `B` | `R(a) = ::Object`
| 92 | `OP_CLASS` | `BB` | `R(a) = newclass(R(a),Syms(b),R(a+1))`
| 93 | `OP_MODULE` | `BB` | `R(a) = newmodule(R(a),Syms(b))`
| 94 | `OP_EXEC` | `BB` | `R(a) = blockexec(R(a),Irep(b))`
| 95 | `OP_DEF` | `BB` | `R(a).newmethod(Syms(b),R(a+1)); R(a) = Syms(b)`
| 96 | `OP_ALIAS` | `BB` | `alias_method(target_class,Syms(a),Syms(b))`
| 97 | `OP_UNDEF` | `B` | `undef_method(target_class,Syms(a))`
| 98 | `OP_SCLASS` | `B` | `R(a) = R(a).singleton_class`
| 99 | `OP_TCLASS` | `B` | `R(a) = target_class`
| 100 | `OP_DEBUG` | `BBB` | `print a,b,c`
| 101 | `OP_ERR` | `B` | `raise(LocalJumpError, Pool(a))`
| 102 | `OP_EXT1` | `-` | `make 1st operand (a) 16bit`
| 103 | `OP_EXT2` | `-` | `make 2nd operand (b) 16bit`
| 104 | `OP_EXT3` | `-` | `make 1st and 2nd operands 16bit`
| 105 | `OP_STOP` | `-` | `stop VM`
## Notes
### OP_GETIDX / OP_SETIDX Optimization
These instructions optimize `[]` and `[]=` access for Array, Hash, and String.
**OP_GETIDX** uses direct function calls:
- `Array`: `mrb_ary_entry()` (integer index only)
- `Hash`: `mrb_hash_get()`
- `String`: `mrb_str_aref()` (integer/string/range index)
**OP_SETIDX** uses direct function calls:
- `Array`: `mrb_ary_set()` (integer index only)
- `Hash`: `mrb_hash_set()`
**Fallback to method dispatch** occurs when:
- Object is a subclass (e.g., `MyArray < Array`)
- Object has a singleton class (singleton methods defined)
- Index type is not supported (e.g., non-integer for Array)
This allows subclasses to override `[]`/`[]=` while base classes remain optimized.
Reference in New Issue View Git Blame Copy Permalink