holey-bytes/spec.md

HoleyBytes ISA Specification

Bytecode format

• All numbers are encoded little-endian
• There is 60 registers (0 – 59), they are represented by a byte
• Immediate values are 64 bit

Instruction encoding

• Instruction parameters are packed (no alignment)
• [opcode, …parameters…]

Instruction parameter types

• R = Register
• I = Immediate

Name	Size
RRRR	32 bits
RRR	24 bits
RRI	80 bits
RR	16 bits
RI	72 bits
I	64 bits
N	0 bits

Instructions

• #n: register in parameter n
• imm #n: for immediate in parameter n
• P ← V: Set register P to value V
• [x]: Address x

No-op

• N type

Opcode	Name	Action
0	NOP	Do nothing

Integer binary ops.

• RRR type
• #0 ← #1 <op> #2

Opcode	Name	Action
1	ADD	Wrapping addition
2	SUB	Wrapping subtraction
3	MUL	Wrapping multiplication
4	AND	Bitand
5	OR	Bitor
6	XOR	Bitxor
7	SL	Unsigned left bitshift
8	SR	Unsigned right bitshift
9	SRS	Signed right bitshift

Comparsion

Opcode	Name	Action
10	CMP	Signed comparsion
11	CMPU	Unsigned comparsion

Comparsion table

#1 op #2	Result
<	-1
=	0
>	1

Division-remainder

• Type RRRR
• In case of #3 is zero, the resulting value is all-ones
• #0 ← #2 ÷ #3
• #1 ← #2 % #3

Opcode	Name	Action
12	DIR	Divide and remainder combinated

Negations

• Type RR
• #0 ← #1 <op> #2

Opcode	Name	Action
13	NEG	Bit negation
14	NOT	Logical negation

Integer immediate binary ops.

• Type RRI
• #0 ← #1 <op> imm #2

Opcode	Name	Action
18	ADDI	Wrapping addition
19	MULI	Wrapping subtraction
20	ANDI	Bitand
21	ORI	Bitor
22	XORI	Bitxor
23	SLI	Unsigned left bitshift
24	SRI	Unsigned right bitshift
25	SRSI	Signed right bitshift

Comparsion

• Comparsion is the same as when RRR type

Opcode	Name	Action
26	CMPI	Signed comparsion
27	CMPUI	Unsigned comparsion

Register value set / copy

Copy

• Type RR
• #0 ← #1

Opcode	Name	Action
28	CP	Copy

Load immediate

• Type RI
• #0 ← #1

Opcode	Name	Action
29	LI	Load immediate

Memory operations

• Type RRI

Load

• #0 ← [#1 + imm #2]

Opcode	Name	Action
30	LB	Load byte (8 bits)
31	LD	Load doublet (16 bits)
32	LQ	Load quadlet (32 bits)
33	LO	Load octlet (64 bits)

Store

• [#1 + imm #2] ← #0

Opcode	Name	Action
34	SB	Store byte (8 bits)
35	SD	Store doublet (16 bits)
36	SQ	Store quadlet (32 bits)
37	SO	Store octlet (64 bits)

Control flow

Unconditional jump

• Type RI

Opcode	Name	Action
38	JMP	Jump at #0 + imm #1

Conditional jumps

• Type RRI
• Jump at imm #2 if #0 <op> #1

Opcode	Name	Comparsion
39	JEQ	=
40	JNE	≠
41	JLT	< (signed)
42	JGT	> (signed)
43	JLTU	< (unsigned)
44	JGTU	> (unsigned)

Environment call

• Type N

Opcode	Name	Action
45	ECALL	Cause an trap to the host environment

Floating point operations

• Type RRR
• #0 ← #1 <op> #2

Opcode	Name	Action
46	ADDF	Addition
47	MULF	Multiplication

Division-remainder

• Type RRRR

Opcode	Name	Action
48	DIRF	Same flow applies as for integer DIR

Floating point immediate operations

• Type RRI
• #0 ← #1 <op> imm #2

Opcode	Name	Action
49	ADDFI	Addition
50	MULFI	Multiplication

Registers

• There is 59 registers + one zero register (with index 0)
• Reading from zero register yields zero
• Writing to zero register is a no-op

Memory

• Addresses are 64 bit
• Memory implementation is arbitrary
• In case of accessing invalid address:
  • Program shall trap (LoadAccessEx, StoreAccessEx) with parameter of accessed address
  • Value of register when trapped is undefined

Recommendations

• Leave address 0x0 as invalid
• If paging used:
  • Leave first page invalid
  • Pages should be at least 4 KiB

Program execution

• The way of program execution is implementation defined
• The order of instruction is arbitrary, as long all observable effects are applied in the program's order

Program validation

• Invalid program should cause runtime error:
  • The form of error is arbitrary. Can be a trap or an interpreter-specified error
  • It shall not be handleable from within the program
• Executing invalid opcode should trap
• Program can be validaded either before execution or when executing

Traps

Name	Parameters	Cause
LoadAccessEx	Accessed address	Loading invalid address
StoreAccessEx	Accessed address	Storing to invalid address
InvalidOpcode	Loaded opcode	Executing invalid opcode
Ecall	None	Ecall instruction

Assembly

HoleyBytes assembly format is not defined, this is just a weak description of hbasm syntax.

• Opcode names correspond to specified opcode names, lowercase (nop)
• Parameters are separated by comma (addi r0, r0, 1)
• Instructions are separated by either line feed or semicolon
• Registers are represented by r followed by the number (r10)
• Labels are defined by label name followed with colon (loop:)
• Labels are references simply by their name (print)
• Immediates are entered plainly. Negative numbers supported.