LIR

In Nanojit, LIR is the source language for compilation to machine code. LIR stands for low-level intermediate representation. It is a typed assembly language.

The LIR instruction set is best learnt by reading nanojit/LIRopcode.tbl. Code for manipulating LIR is in nanojit/LIR.h.

Nanojit LIR instruction cheat sheet.Also in pdf format see attachment.

Category	OP Code	Code Name	Return Type	Featured	Description
Miscellaneous	0	start	void		start of a fragment
	1	regfence	void		A register fence causes no code to be generated, but it affects register allocation so that no registers are live when it is reached.
	2	skip	void		links code chunks.
	3	parami	integer	32 bit	load an int parameter (register or stack location).
	4	paramq	quad	64 bit	load a quad parameter (register or stack location).
	5	allocp	pointer		allocate stack space (result is an address)
	6	reti	void		return an int
	7	retq	void	64 bit	return a quad
	8	retd	void		return a double
	9	livei	void		extend live range of an int
	10	liveq	void	64 bit	extend live range of a quad
	11	lived	void		extend live range of a double
	12	file	void		source filename for debug symbols
	13	line	void		source line number for debug symbols
	14	comment	void		a comment shown, on its own line, in LIR dumps
	15	not in use
Load	16	not in use
	17	ldc2i	integer		load char and sign-extend to an int
	18	lds2i	integer		load short and sign-extend to an int
	19	lduc2ui	integer		load unsigned char and zero-extend to an unsigned int
	20	ldus2ui	integer		load unsigned short and zero-extend to an unsigned int
	21	ldi	integer		load int
	22	ldq	quad	64 bit	load quad
	23	ldd	double		load double
	24	ldf2d	double		load float and extend to a double
Store	25	sti2c	void		store int truncated to char
	26	sti2s	void		store int truncated to short
	27	sti	void		store int
	28	stq	void	64 bit	store quad
	29	std	void		store double
	30	std2f	void		store double as a float (losing precision)
	31	not in use
Call	32	not in use
	33	callv	void		call subroutine that returns void
	34	calli	integer		call subroutine that returns an int
	35	callq	quad	64 bit	call subroutine that returns a quad
	36	calld	double		call subroutine that returns a double
Branch	37	j	void		jump always
	38	jt	void		jump if true
	39	jf	void		jump if false
	40	jtbl	void		jump to address in table
	41	label	void		a jump target (no machine code is emitted for this)
	42	not in use
Guards	43	x	void		exit always
	44	xt	void		exit if true
	45	xf	void		exit if false
	46	xtbl	void		exit via indirect jump
	47	xbarrier	void		A LIR_xbarrier cause no code to be generated, but it acts like a never-taken guard in that it inhibits certain optimisations, such as dead stack store elimination.
	48	not in use
Immediate	49	immi			int immediate
	50	immq		64 bit	quad immediate
	51	immd			double immediate
	52	not in use
Comparisons	53	eqi	integer		int equality
	54	lti	integer		signed int less-than
	55	gti	integer		signed int greater-than
	56	lei	integer		signed int less-than-or-equal
	57	gei	integer		signed int greater-than-or-equal
	58	ltui	integer		unsigned int less-than
	59	gtui	integer		unsigned int greater-than
	60	leui	integer		unsigned int less-than-or-equal
	61	geui	integer		unsigned int greater-than-or-equal
	62	not in use
	63	eqq	integer	64 bit	quad equality
	64	ltq	integer	64 bit	signed quad less-than
	65	gtq	integer	64 bit	signed quad greater-than
	66	leq	integer	64 bit	signed quad less-than-or-equal
	67	geq	integer	64 bit	signed quad greater-than-or-equal
	68	ltuq	integer	64 bit	unsigned quad less-than
	69	gtuq	integer	64 bit	unsigned quad greater-than
	70	leuq	integer	64 bit	unsigned quad less-than-or-equal
	71	geuq	integer	64 bit	unsigned quad greater-than-or-equal
	72	not in use
Arithmetic	73	eqd	integer		double equality
	74	ltd	integer		double less-than
	75	gtd	integer		double greater-than
	76	led	integer		double less-than-or-equal
	77	ged	integer		double greater-than-or-equal
	78	negi	integer		negate int
	79	addi	integer		add int
	80	subi	integer		subtract int
	81	muli	integer		multiply int
	82	divi	integer	i386/64	divide int
	83	modi	integer		modulo int - LIR_modi is a hack. It's only used on i386/X64. The operand is the result of a LIR_divi because on i386/X64 div and mod results are computed by the same instruction.
	84	noti	integer		bitwise-NOT int
	85	andi	integer		bitwise-AND int
	86	ori	integer		bitwise-OR int
	87	xori	integer		bitwise-XOR int
	88	lshi	integer		left shift int
	89	rshi	integer		right shift int (>>)
	90	rshui	integer		right shift unsigned int (>>>)
	91	addq	quad	64 bit	add quad
	92	subq	quad	64 bit	subtract quad
	93	andq	quad	64 bit	bitwise-AND quad
	94	orq	quad	64 bit	bitwise-OR quad
	95	xorq	quad	64 bit	bitwise-XOR quad
	96	lshq	quad	64 bit	left shift quad,2nd operand is an int
	97	rshq	quad	64 bit	right shift quad, 2nd operand is an int
	98	rshuq	quad	64 bit	right shift unsigned quad; 2nd operand is an int
	99	negd	double		negate double
	100	addd	double		add double
	101	subd	double		subtract double
	102	muld	double		multiply double
	103	divd	double		divide double
	104	modd	double		modulo double LIR_modd is just a place-holder opcode, ie. the back-ends cannot generate code for it.It's used in TraceMonkey briefly but is always demoted to a LIR_modl or converted to a function call before Nanojit has to do anything serious with it.
	105	cmovi	integer		conditional move int
	106	cmovq	quad	64 bit	conditional move quad
	107	cmovd	double		conditional move double
Conversions	108	i2q	quad	64 bit	sign-extend int to quad
	109	ui2uq	quad	64 bit	zero-extend unsigned int to unsigned quad
	110	q2i	integer	64 bit	truncate quad to int (removes the high 32 bits)
	111	i2d	double		convert int to double
	112	ui2d	double		convert unsigned int to double
	113	d2i	integer		convert double to int (no exceptions raised)
	114	dasq	quad	64 bit	interpret the bits of a double as a quad
	115	qasd	double	64 bit	interpret the bits of a quad as a double
Overflow arithmetic	116	addxovi	integer		add int and exit on overflow
	117	subxovi	integer		subtract int and exit on overflow
	118	mulxovi	integer		multiply int and exit on overflow
	119	addjovi	integer		add int and branch on overflow
	120	subjovi	integer		subtract int and branch on overflow
	121	muljovi	integer		multiply int and branch on overflow
	122	addjovq	quad	64 bit	add quad and branch on overflow
	123	subjovq	quad	64 bit	subtract quad and branch on overflow
SoftFloat	124	dlo2i	integer	SoftFloat	get the low 32 bits of a double as an int
	125	dhi2i	integer	SoftFloat	get the high 32 bits of a double as an int
	126	ii2d	double	SoftFloat	join two ints (1st arg is low bits, 2nd is high)
	127	hcalli	integer		LIR_hcalli is a hack that's only used on 32-bit platforms that use SoftFloat. Its operand is always a LIR_calli, but one that specifies a function that returns a double. It indicates that the double result is returned via two 32-bit integer registers. The result is always used as the second operand of a LIR_ii2d.

Document Tags and Contributors

Contributors to this page: teoli, vivek.gangadharan, nnethercote, paritosh1010, Jorend, Edwsmith

Last updated by: teoli, Feb 29, 2016, 11:29:16 AM