Appendix A: BEAM Instructions

Here we will go through most of the instructions in the BEAM generic instruction set in detail. In the next section we list all instructions with a brief explanation generated from the documentaion in the code (see lib/compiler/src/genop.tab).

Functions and Labels

label Lbl

Instruction number 1 in the generic instruction set is not really an instruction at all. It is just a module local label giving a name, or actually a number to the current position in the code.

Each label potentially marks the beginning of a basic block since it is a potential destination of a jump.

func_info Module Function Arity

The code for each function starts with a func_info instruction. This instruction is used for generating a function clause error, and the execution of the code in the function actually starts at the label following the func_info instruction.

Imagine a function with a guard:

id(I) when is_integer(I) -> I.

The Beam code for this function might look like:

{function, id, 1, 4}.
  {label,3}.
    {func_info,{atom,test1},{atom,id},1}.
  {label,4}.
    {test,is_integer,{f,3},[{x,0}]}.
    return.

Here the meta information {function, id, 1, 4} tells us that execution of the id/1 function will start at label 4. At label 4 we do an is_integer on x0 and if we fail we jump to label 3 (f3) which points to the func_info instruction, which will generate a function clause exception. Otherwise we just fall through and return the argument (x0).

Test instructions

Type tests

The type test instructions (is_\* Lbl Argument) checks whether the argument is of the given type and if not jumps to the label Lbl. The beam disassembler wraps all these instructions in a test instruction. E.g.:

    {test,is_integer,{f,3},[{x,0}]}.

The current type test instructions are is_integer, is_float, is_number, is_atom, is_pid, is_reference, is_port, is_nil, is_binary, is_list, is_nonempty_list, is_function, is_function2, is_boolean, is_bitstr, and is_tuple.

And then there is also one type test instruction of Arity 3: test_arity Lbl Arg Arity. This instruction tests that the arity of the argument (assumed to be a tuple) is of Arity. This instruction is usually preceded by an is_tuple instruction.

Comparisons

The comparison instructions (is_\* Lbl Arg1 Arg2) compares the two arguments according to the instructions and jumps to Lbl if the comparison fails.

The comparison instructions are: : is_lt, is_ge, is_eq, is_ne, is_eq_exact, and is_ne_exact.

Remember that all Erlang terms are ordered so these instructions can compare any two terms. You can for example test if the atom self is less than the pid returned by self(). (It is.)

Note that for numbers the comparison is done on the Erlang type number, see [CH-TypeSystem]. That is, for a mixed float and integer comparison the number of lower precision is converted to the other type before comparison. For example on my system 1 and 0.1 compares as equal, as well as 9999999999999999 and 1.0e16. Comparing floating point numbers is always risk and best avoided, the result may wary depending on the underlying hardware.

If you want to make sure that the integer 1 and the floating point number 1.0 are compared different you can use is_eq_exact and is_ne_exact. This corresponds to the Erlang operators =:= and =/=.

Function Calls

In this chapter we will summarize what the different call instructions does. For a thorough description of how function calls work see [CH-Calls].

call Arity Label

Does a call to the function of arity Arity in the same module at label Label. First count down the reductions and if needed do a context switch.

For all local calls the label is the second label of the function where the code starts. It is assumed that the preceding instruction at that label is func_info in order to get the MFA if a context switch is needed.

call_last Arity Label

Do a tail recursive call the function of arity Arity in the same module at label Label. First count down the reductions and if needed do a context switch.

call_only Arity Label Deallocate

Deallocate Deallocate words of stack, then do a tail recursive call to the function of arity Arity in the same module at label Label First count down the reductions and if needed do a context switch.

call_ext Arity Destination

Does an external call to the function of arity Arity given by Destination. Destination is usually of the form {extfunc, Module, Function, Arity}. First count down the reductions and if needed do a context switch.

call_ext_only Arity Destination

Does a tail recursive external call to the function of arity Arity given by Destination. Destination is usually of the form {extfunc, Module, Function, Arity}. First count down the reductions and if needed do a context switch.

call_ext_last Arity Destination Deallocate

Deallocate Deallocate words of stack, then do a tail recursive external call to the function of arity Arity given by Destination. Destination is usually of the form {extfunc, Module, Function, Arity}. First count down the reductions and if needed do a context switch.

bif0 Bif Reg, bif1 Lbl Bif Arg Reg, bif2 Lbl Bif Arg1 Arg2 Reg

Call the bif Bif with the given arguments, and store the result in Reg. If the bif fails jump to Lbl. No zero arity bif can fail an thus those calls doesn’t take a fail label.

gc_bif1-3 Lbl Live Bif Arg1-3 Reg

Call the bif Bif with the given arguments, and store the result in Reg. If the bif fails jump to Lbl. Store the arguments in x(Live), x(Live+1) and x(live+2).

call_fun Arity

The instruction call_fun assumes that the arguments are placed in the argument registers and that the fun (the pointer to the closure) is placed in the last argument register.

That is, for a zero arity call, the closure is in x0. For a arity 1 call x0 contains the argument and x1 contains the closure.

apply/1

TODO

apply_last/2

TODO

Stack (and Heap) Management

The stack and the heap of an Erlang process on Beam share the same memory area see [CH-Processes] and [CH-Memory] for a full discussion. The stack grows toward lower addresses and the heap toward higher addresses. Beam will do a garbage collection if more space than what is available is needed on either the stack or the heap.

A leaf function: A leaf function is a function which doesn’t call any other function.
A non leaf function: A non leaf function is a function which may call another function.

These instructions are also used by non leaf functions for setting up and tearing down the stack frame for the current instruction. That is, on entry to the function the continuation pointer (CP) is saved on the stack, and on exit it is read back from the stack.

A function skeleton for a leaf function looks like this:

{function, Name, Arity, StartLabel}.
  {label,L1}.
    {func_info,{atom,Module},{atom,Name},Arity}.
  {label,L2}.
    ...
    return.

A function skeleton for a non leaf function looks like this:

{function, Name, Arity, StartLabel}.
  {label,L1}.
    {func_info,{atom,Module},{atom,Name},Arity}.
  {label,L2}.
    {allocate,Need,Live}.

    ...
    call ...
    ...

    {deallocate,Need}.
    return.

allocate StackNeed Live

Save the continuation pointer (CP) and allocate space for StackNeed extra words on the stack. If a GC is needed during allocation save Live number of X registers. E.g. if Live is 2 then registers X0 and X1 are saved.

When allocating on the stack, the stack pointer (E) is decreased.

Example 1. Allocate 1 0

       Before           After
         | xxx |            | xxx |
    E -> | xxx |            | xxx |
         |     |            | ??? | caller save slot
           ...         E -> | CP  |
           ...                ...
 HTOP -> |     |    HTOP -> |     |
         | xxx |            | xxx |

allocate_heap StackNeed HeapNeed Live

Save the continuation pointer (CP) and allocate space for StackNeed extra words on the stack. Ensure that there also is space for HeapNeed words on the heap. If a GC is needed during allocation save Live number of X registers.

Note that the heap pointer (HTOP) is not changed until the actual heap allocation takes place.

allocate_zero StackNeed Live

This instruction works the same way as allocate, but it also clears out the allocated stack slots with NIL.

Example 2. allocate_zero 1 0

       Before           After
         | xxx |            | xxx |
    E -> | xxx |            | xxx |
         |     |            | NIL | caller save slot
           ...         E -> | CP  |
           ...                ...
 HTOP -> |     |    HTOP -> |     |
         | xxx |            | xxx |

allocate_heap_zero StackNeed HeapNeed Live

The allocate_heap_zero instruction works as the allocate_heap instruction, but it also clears out the allocated stack slots with NIL.

test_heap HeapNeed Live

The test_heap instruction ensures there is space for HeapNeed words on the heap. If a GC is needed save Live number of X registers.

init N

The init instruction clears N stack words. By writing NIL to them.

deallocate N

The deallocate instruction is the opposite of the allocate instruction, it restores the continuation pointer and deallocates N+1 stack words.

return

The return instructions jumps to the address in the continuation pointer (CP).

trim N Remaining

Removes N words of stack usage, while keeping the continuation pointer on the top of the stack. (The argument Remaining is to the best of my knowledge unused.)

Trim 2
       Before           After
         | ??? |            | ??? |
         | xxx |       E -> | CP  |
         | xxx |            | ... |
    E -> | CP  |            | ... |
         |     |            | ... |
           ...                ...
 HTOP -> |     |    HTOP -> |     |
         | xxx |            | xxx |

Moving, extracting, modifying.

move Source Destination

Move the source Source (a litteral or a register) to the destination register Destination.

get_list Source Head Tail

Get the head and tail (or car and cdr) parts of a list (a cons cell) from Source and put them into the registers Head and Tail.

get_tuple_element Source Element Destination

Get element number Element from the tuple in Source and put it in the destination register Destination.

set_tuple_element NewElement Tuple Position

Update the element at postition Position of the tuple Tuple with the new element NewElement.

Building terms.

put_list/3

TODO ==== put_tuple/2 TODO ==== put/1

make_fun2/1

TODO

Binary Syntax

bs_put_integer/5

TODO ==== bs_put_binary/5 TODO ==== bs_put_float/5 TODO ==== bs_put_string/2 TODO ==== bs_init2/6 TODO ==== bs_add/5 TODO ==== bs_start_match2/5 TODO ==== bs_get_integer2/7 TODO ==== bs_get_float2/7 TODO ==== bs_get_binary2/7 TODO ==== bs_skip_bits2/5 TODO ==== bs_test_tail2/3 TODO ==== bs_save2/2 TODO ==== bs_restore2/2 TODO ==== bs_context_to_binary/1 TODO ==== bs_test_unit/3 TODO ==== bs_match_string/4 TODO ==== bs_init_writable/0 TODO ==== bs_append/8 TODO ==== bs_private_append/6 TODO ==== bs_init_bits/6 TODO ==== bs_get_utf8/5 TODO ==== bs_skip_utf8/4 TODO ==== bs_get_utf16/5 TODO ==== bs_skip_utf16/4 TODO ==== bs_get_utf32/5 TODO ==== bs_skip_utf32/4 TODO ==== bs_utf8_size/3 TODO ==== bs_put_utf8/3 TODO ==== bs_utf16_size/3 TODO ==== bs_put_utf16/3 TODO ==== bs_put_utf32/3 TODO

Floating Point Arithmetic

fclearerror/0

TODO ==== fcheckerror/1 TODO ==== fmove/2 TODO ==== fconv/2 TODO ==== fadd/4 TODO ==== fsub/4 TODO ==== fmul/4 TODO ==== fdiv/4 TODO ==== fnegate/3 TODO

Pattern Matching

select_val

TODO ==== select_arity_val TODO ==== jump TODO === Exception handling ==== catch/2 TODO ==== catch_end/1 TODO ==== badmatch/1 TODO ==== if_end/0 TODO ==== case_end/1 TODO

Meta instructions

on_load

TODO ==== line TODO

opcodes_doc.asciidoc

Specific Instructions

Argument types

Type	Explanation
t	A term, e.g. [{foo,bar}]
I	An integer e.g. 42
x	A register, e.g. 5
y	A stack slot, e.g. 1
c	A constant (atom,nil,small int) // Pid?
a	An atom, e.g. 'foo'
f	A label, i.e. a code address
s	Either a literal, a register or a stack slot
d	Either a register or a stack slot
r	A register R0
P	A positive (unsigned) integer literal
j	An optional code label
e	A reference to an export table entry
l	A floating-point register

List of all BEAM Instructions

Instruction	Arguments	Explanation
allocate	t t
allocate_heap	t I t
deallocate	I
init	y
init2	y y
init3	y y y
i_trim	I
test_heap	I t
allocate_zero	t t
allocate_heap_zero	t I t
i_select_val	r f I
i_select_val	x f I
i_select_val	y f I
i_select_val2	r f c f c f
i_select_val2	x f c f c f
i_select_val2	y f c f c f
i_jump_on_val	rxy f I I
i_jump_on_val_zero	rxy f I
i_select_tuple_arity	r f I
i_select_tuple_arity	x f I
i_select_tuple_arity	y f I
i_select_tuple_arity2	r f A f A f
i_select_tuple_arity2	x f A f A f
i_select_tuple_arity2	y f A f A f
i_func_info	I a a I
return
get_list	rxy rxy rxy
catch	y f
catch_end	y
try_end	y
try_case_end	s
set_tuple_element	s d P
i_get_tuple_element	rxy P rxy
is_number	f rxy
jump	f
case_end	rxy
badmatch	rxy
if_end
raise	s s
badarg	j
system_limit	j
move_jump	f ncxy
move_x1	c
move_x2	c
move2	x y x y
move2	y x y x
move2	x x x x
move	rxync rxy
recv_mark	f
i_recv_set	f
remove_message
timeout
timeout_locked
i_loop_rec	f r
loop_rec_end	f
wait	f
wait_locked	f
wait_unlocked	f
i_wait_timeout	f I
i_wait_timeout	f s
i_wait_timeout_locked	f I
i_wait_timeout_locked	f s
i_wait_error
i_wait_error_locked
send
i_is_eq_exact_immed	f rxy c
i_is_ne_exact_immed	f rxy c
i_is_eq_exact_literal	f rxy c
i_is_ne_exact_literal	f rxy c
i_is_eq_exact	f
i_is_ne_exact	f
i_is_lt	f
i_is_ge	f
i_is_eq	f
i_is_ne	f
i_put_tuple	rxy I
put_list	s s d
i_fetch	s s
move_return	xcn r
move_deallocate_return	xycn r Q
deallocate_return	Q
test_heap_1_put_list	I y
is_tuple_of_arity	f rxy A
is_tuple	f rxy
test_arity	f rxy A
extract_next_element	xy
extract_next_element2	xy
extract_next_element3	xy
is_integer_allocate	f rx I I
is_integer	f rxy
is_list	f rxy
is_nonempty_list_allocate	f rx I t
is_nonempty_list_test_heap	f r I t
is_nonempty_list	f rxy
is_atom	f rxy
is_float	f rxy
is_nil	f rxy
is_bitstring	f rxy
is_reference	f rxy
is_pid	f rxy
is_port	f rxy
is_boolean	f rxy
is_function2	f s s
allocate_init	t I y
i_apply
i_apply_last	P
i_apply_only
apply	I
apply_last	I P
i_apply_fun
i_apply_fun_last	P
i_apply_fun_only
i_hibernate
call_bif0	e
call_bif1	e
call_bif2	e
call_bif3	e
i_get	s d
self	rxy
node	rxy
i_fast_element	rxy j I d
i_element	rxy j s d
bif1	f b s d
bif1_body	b s d
i_bif2	f b d
i_bif2_body	b d
i_move_call	c r f
i_move_call_last	f P c r
i_move_call_only	f c r
move_call	xy r f
move_call_last	xy r f Q
move_call_only	x r f
i_call	f
i_call_last	f P
i_call_only	f
i_call_ext	e
i_call_ext_last	e P
i_call_ext_only	e
i_move_call_ext	c r e
i_move_call_ext_last	e P c r
i_move_call_ext_only	e c r
i_call_fun	I
i_call_fun_last	I P
i_make_fun	I t
is_function	f rxy
i_bs_start_match2	rxy f I I d
i_bs_save2	rx I
i_bs_restore2	rx I
i_bs_match_string	rx f I I
i_bs_get_integer_small_imm	rx I f I d
i_bs_get_integer_imm	rx I I f I d
i_bs_get_integer	f I I d
i_bs_get_integer_8	rx f d
i_bs_get_integer_16	rx f d
i_bs_get_integer_32	rx f I d
i_bs_get_binary_imm2	f rx I I I d
i_bs_get_binary2	f rx I s I d
i_bs_get_binary_all2	f rx I I d
i_bs_get_binary_all_reuse	rx f I
i_bs_get_float2	f rx I s I d
i_bs_skip_bits2_imm2	f rx I
i_bs_skip_bits2	f rx rxy I
i_bs_skip_bits_all2	f rx I
bs_test_zero_tail2	f rx
bs_test_tail_imm2	f rx I
bs_test_unit	f rx I
bs_test_unit8	f rx
bs_context_to_binary	rxy
i_bs_get_utf8	rx f d
i_bs_get_utf16	rx f I d
i_bs_validate_unicode_retract	j
i_bs_init_fail	rxy j I d
i_bs_init_fail_heap	I j I d
i_bs_init	I I d
i_bs_init_heap	I I I d
i_bs_init_heap_bin	I I d
i_bs_init_heap_bin_heap	I I I d
i_bs_init_bits_fail	rxy j I d
i_bs_init_bits_fail_heap	I j I d
i_bs_init_bits	I I d
i_bs_init_bits_heap	I I I d
i_bs_add	j I d
i_bs_init_writable
i_bs_append	j I I I d
i_bs_private_append	j I d
i_new_bs_put_integer	j s I s
i_new_bs_put_integer_imm	j I I s
i_bs_utf8_size	s d
i_bs_utf16_size	s d
i_bs_put_utf8	j s
i_bs_put_utf16	j I s
i_bs_validate_unicode	j s
i_new_bs_put_float	j s I s
i_new_bs_put_float_imm	j I I s
i_new_bs_put_binary	j s I s
i_new_bs_put_binary_imm	j I s
i_new_bs_put_binary_all	j s I
bs_put_string	I I
fmove	qdl ld
fconv	d l
i_fadd	l l l
i_fsub	l l l
i_fmul	l l l
i_fdiv	l l l
i_fnegate	l l l
i_fcheckerror
fclearerror
i_increment	rxy I I d
i_plus	j I d
i_minus	j I d
i_times	j I d
i_m_div	j I d
i_int_div	j I d
i_rem	j I d
i_bsl	j I d
i_bsr	j I d
i_band	j I d
i_bor	j I d
i_bxor	j I d
i_int_bnot	j s I d
i_gc_bif1	j I s I d
i_gc_bif2	j I I d
i_gc_bif3	j I s I d
int_code_end
label	L
line	I

Files

ap-beam_instructions.asciidoc

Latest commit

History

ap-beam_instructions.asciidoc

File metadata and controls

Appendix A: BEAM Instructions

Functions and Labels

label Lbl

func_info Module Function Arity

Test instructions

Type tests

Comparisons

Function Calls

call Arity Label

call_last Arity Label

call_only Arity Label Deallocate

call_ext Arity Destination

call_ext_only Arity Destination

call_ext_last Arity Destination Deallocate

bif0 Bif Reg, bif1 Lbl Bif Arg Reg, bif2 Lbl Bif Arg1 Arg2 Reg

gc_bif1-3 Lbl Live Bif Arg1-3 Reg

call_fun Arity

apply/1

apply_last/2

Stack (and Heap) Management

allocate StackNeed Live

allocate_heap StackNeed HeapNeed Live

allocate_zero StackNeed Live

allocate_heap_zero StackNeed HeapNeed Live

test_heap HeapNeed Live

init N

deallocate N

return

trim N Remaining

Moving, extracting, modifying.

move Source Destination

get_list Source Head Tail

get_tuple_element Source Element Destination

set_tuple_element NewElement Tuple Position

Building terms.

put_list/3

make_fun2/1

Binary Syntax

bs_put_integer/5

Floating Point Arithmetic

fclearerror/0

Pattern Matching

select_val

Meta instructions

on_load

Specific Instructions