Age | Commit message (Collapse) | Author |
|
Consider the following function:
function({function,Name,Arity,CLabel,Is0}, Lc0) ->
try
%% Optimize the code for the function.
catch
Class:Error:Stack ->
io:format("Function: ~w/~w\n", [Name,Arity]),
erlang:raise(Class, Error, Stack)
end.
The stacktrace is retrieved, but it is only used in the call
to erlang:raise/3. There is no need to build a stacktrace
in this function. We can avoid the building if we introduce
an instruction called raw_raise/3 that works exactly like
the erlang:raise/3 BIF except that its third argument must
be a raw stacktrace.
|
|
Argument order can prevent the delayed sub binary creation.
Here is an example directly from the Efficiency Guide:
non_opt_eq([H|T1], <<H,T2/binary>>) ->
non_opt_eq(T1, T2);
non_opt_eq([_|_], <<_,_/binary>>) ->
false;
non_opt_eq([], <<>>) ->
true.
When compiling with the bin_opt_info option, there will be a
suggestion to change the argument order.
It turns out sys_core_bsm can itself change the order, not the
order of the arguments of themselves, but the order in which
the arguments are matched. Here is how it can be rewritten in
pseudo Core Erlang code:
non_opt_eq(Arg1, Arg2) ->
case < Arg2,Arg1 > of
< <<H1,T2/binary>>, [H2|T1] > when H1 =:= H2 ->
non_opt_eq(T1, T2);
< <<_,T2/binary>ffff>, [_|T1] > ->
false;
< <<>>, [] >> ->
true
end.
When rewritten like this, the bs_start_match2 instruction will be
the first instruction in the function and it will be possible to
store the match context in the same register as the binary
({x,1} in this case) and to delay the creation of sub binaries.
The switching of matching order also enables many other simplifications
in sys_core_bsm, since there is no longer any need to pass the position
of the pattern as an argument.
We will update the Efficiency Guide in a separate branch before the
release of OTP 21.
|
|
Run beam_block a second time
|
|
Clean up and improve sys_core_fold optimizations
|
|
Running beam_block again after the other optimizations have run will
give it more opportunities for optimizations. In particular, more
allocate_zero/2 instructions can be turned into allocate/2
instructions, and more get_tuple_element/3 instructions can store the
retrieved value into the correct register at once.
Out of a sample of about 700 modules in OTP, 64 modules were improved
by this commit.
|
|
If possible, when adding move/2 instructions, try to insert
them into a block. That could potentially allow them to
be optimized.
|
|
beam_utils:live_opt/1 is currently only run early (from
beam_block). Prepare it to be run after beam_split when
instructions with failure labels have been taken out of
blocks.
While we are it, also improve check_liveness/3. That will
improve the optimizations in beam_record (replacing tuple
matching instructions with an is_tagged_tuple instruction).
|
|
Since the select_val instruction never transfer directly to the next
instruction, the incoming live registers should be ignored. This
bug have not caused any problems yet, but it will in the future
if we are to run the liveness optimizations again after
the optimizations in beam_dead and beam_jump.
|
|
In a guard, reorder two consecutive calls to the element/2 BIF that
access the same tuple and have the same failure label so that highest
index is fetched first. That will allow the second element/2 to be
replace with the slightly cheaper get_tuple_element/3 instruction.
|
|
Consider a 'case' that exports variables and whose return
value is ignored:
foo(N) ->
case N of
1 ->
Res = one;
2 ->
Res = two
end,
{ok,Res}.
That code will be translated to the following Core Erlang code:
'foo'/1 =
fun (_@c0) ->
let <_@c5,Res> =
case _@c0 of
<1> when 'true' ->
<'one','one'>
<2> when 'true' ->
<'two','two'>
<_@c3> when 'true' ->
primop 'match_fail'({'case_clause',_@c3})
end
in
{'ok',Res}
The exported variables has been rewritten to explicit return
values. Note that the original return value from the 'case' is bound to
the variable _@c5, which is unused.
The corresponding BEAM assembly code looks like this:
{function, foo, 1, 2}.
{label,1}.
{line,[...]}.
{func_info,{atom,t},{atom,foo},1}.
{label,2}.
{test,is_integer,{f,6},[{x,0}]}.
{select_val,{x,0},{f,6},{list,[{integer,2},{f,3},{integer,1},{f,4}]}}.
{label,3}.
{move,{atom,two},{x,1}}.
{move,{atom,two},{x,0}}.
{jump,{f,5}}.
{label,4}.
{move,{atom,one},{x,1}}.
{move,{atom,one},{x,0}}.
{label,5}.
{test_heap,3,2}.
{put_tuple,2,{x,0}}.
{put,{atom,ok}}.
{put,{x,1}}.
return.
{label,6}.
{line,[...]}.
{case_end,{x,0}}.
Because of the test_heap instruction following label 5, the assignment
to {x,0} cannot be optimized away by the passes that optimize BEAM assembly
code.
Refactor the optimizations of 'let' in sys_core_fold to eliminate the
unused variable. Thus:
'foo'/1 =
fun (_@c0) ->
let <Res> =
case _@c0 of
<1> when 'true' ->
'one'
<2> when 'true' ->
'two'
<_@c3> when 'true' ->
primop 'match_fail'({'case_clause',_@c3})
end
in
{'ok',Res}
The resulting BEAM code will look like:
{function, foo, 1, 2}.
{label,1}.
{line,[...]}.
{func_info,{atom,t},{atom,foo},1}.
{label,2}.
{test,is_integer,{f,6},[{x,0}]}.
{select_val,{x,0},{f,6},{list,[{integer,2},{f,3},{integer,1},{f,4}]}}.
{label,3}.
{move,{atom,two},{x,0}}.
{jump,{f,5}}.
{label,4}.
{move,{atom,one},{x,0}}.
{label,5}.
{test_heap,3,1}.
{put_tuple,2,{x,1}}.
{put,{atom,ok}}.
{put,{x,0}}.
{move,{x,1},{x,0}}.
return.
{label,6}.
{line,[...]}.
{case_end,{x,0}}.
|
|
Improve handling of #c_seq{}, making sure to simplify a #c_seq{}
as much as possible. With that improvement, we can remove some
special-case code from opt_simple_let_2/6.
|
|
|
|
|
|
The annotations in the optimizing passes currently looks like this:
{'%live',NumRegistersUsed,RegistersUsedBitmap}
{'%def',RegistersDefinedBitmap}
(NumRegistersUsed is no longer used.)
When I attempted to extend some optimizations, I found that I had to
add additional clauses to tolerate/handle both types of
annotations. That problem would only get worse if any more annotations
are added in the future.
To simplify annotation handling, this commit wraps both types of
annotations in a {'%anno',_} tuple:
{'%anno',{used,RegistersUsedBitmap}}
{'%anno',{def,RegistersDefinedBitmap}}
The '%live' annotation has been renamed to 'used' to make it somewhat
clearer what it means, and the unused NumRegistersUsed part of the
old annotation has been removed.
Alternatives considered: My first attempt was to wrap the annotation
in a 'set' tuple so that there would only be 'set' tuples in a block.
For example:
{set,[],[],{anno,{live,RegistersUsedBitmap}}}
It was not as convenient as expected. Annotations often need to be
handled specially from other instructions in a block. When they are
wrapped in a 'set' tuple, they can very easily be handled incorrectly
or passed on to the next pass. That causes subtle errors or worse
code, and it can be difficult to debug.
Therefore, my conclusion is that annotations should be distinct from
other instructions, to make it obvious when one have missed to handle
an annotation.
|
|
jhogberg/john/compiler/reintroduce-tuple-arity-optimizations/OTP-14857
Reintroduce the tuple arity optimizations removed in PR #1673
|
|
Turn more allocate_zero instructions into allocate instructions.
|
|
An 'allocate' or 'allocate_zero' instruction should not be
shortly followed by a 'test_heap' instruction. For example,
we don't want this type of code:
{allocate_zero,3,4}.
{line,...}.
{test_heap,7,4}.
{bif,element,{f,0},...,...}.
While the code is safe because 'allocate_zero' has initialized the
stack frame, it is wasteful. Also note that the code would become
unsafe if the 'allocate_zero' instruction were to be replaced with
an 'allocate' instruction.
What we want to see is this:
{allocate_heap_zero,3,7,4}.
{line,...}.
{bif,element,{f,0},...,...}.
|
|
In 21dd6e55877832, beam_utils:combine_heap_needs/2 stopped
wrapping an allocation list in an {alloc,...} tuple. That was
not noticed because the faulty heap need created in beam_block
was discarded by beam_type.
|
|
beam_utils:is_killed/3 could incorrectly indicate that a
register was killed, when in fact it was referenced by
an instruction that did a GC.
|
|
* maint:
beam_validator: Strengthen validation of GC instructions
|
|
beam_validator: Strengthen validation of GC instructions
OTP-14863
|
|
We can safely tell when a test_arity or is_record instruction is
superflous by keeping track of whether the size is exactly known
or not.
|
|
|
|
beam_validator did not verify that the Y registers were initialized
before executing the following instructions that could cause a GC:
bs_append/8
bs_init2/6
bs_init_bits/6
gc_bif1/5
gc_bif2/6
gc_bif3/7
test_heap/2
That means that, for example, an incorrect optimization that replaced
an 'allocate_zero' instruction with an 'allocate' instruction when it
was not safe, would not be rejected by beam_validtor, but would
instead cause a crash or other undefined behavior at runtime.
Also fix a minor bug in beam_type exposed by the stronger checking.
When compiling from .S files, beam_type did not handle the
init/1 instruction and could produce unsafe code.
|
|
The type optimizations for is_record and test_arity checked whether
the arity was equal to the size stored in the type information,
which is incorrect since said size is the *minimum* size of the
tuple (as determined by previous instructions) and not its exact
size.
A future patch to the 'master' branch will restore these
optimizations in a safe manner.
|
|
Handle a few more instructions in beam_utils. That will allow
beam_reorder to reorder more instructions, delaying get_tuple_element
instructions and reducing register shuffling in receive clauses.
|
|
This is an enhancement of the optimization added in 2e5d6201bb044,
where we tried to avoid forcing a stack frame for functions
that don't really need them.
That optimization would not suppress the stack frame for this
function:
f(A) ->
Res = case A of
a -> x;
b -> y
end,
{ok,Res}.
The reason is that internally the compiler would rewrite
the code to something like this:
f(A) ->
Res = case A of
a -> x;
b -> y;
Other -> error({case_clause,Other})
end,
{ok,Res}.
The call to error/1 would force creation of a stack frame,
even though it is not really needed because error/1 causes
an exception.
Handle calls to exit BIFs specially to allow suppressing the
stack frame.
|
|
Delay creation of stack frames
|
|
* bjorn/compiler/coverage:
beam_utils: Refactor combine_alloc_lists() to cover more lines
map_SUITE: Cover beam_utils:bif_to_test/3
beam_disasm: Remove support for obsolete instructions
guard_SUITE: Test is_bitstring/1 and is_map/1 on literals
|
|
v3_codegen currently wraps a stack frame around each clause in
a function (unless the clause is simple without any 'case' or
other complex constructions).
Consider this function:
f({a,X}) ->
A = abs(X),
case A of
0 ->
{result,"0"};
_ ->
{result,integer_to_list(A)}
end;
f(_) ->
error.
The first clause needs a stack frame because there is a function
call to integer_to_list/1 not in the tail position. v3_codegen
currently wraps the entire first clause in stack frame.
We can delay the creation of the stack frame, and create a
stack frame in each arm of the 'case' (if needed):
f({a,X}) ->
A = abs(X),
case A of
0 ->
%% Don't create a stack frame here.
{result,"0"};
_ ->
%% Create a stack frame here.
{result,integer_to_list(A)}
end;
f(_) ->
error.
There are pros and cons of this approach.
The cons are that the code size may increase if there are many
'case' clauses and each needs its own stack frame. The allocation
instructions may also interfere with other optimizations, but
the new optimizations introduced in previous commits will mitigate
most of those issues.
The pros are the following:
* For some clauses in a 'case', there is no need to create any
stack frame at all.
* Often when moving an allocation instruction into a 'case' clause,
the slightly cheaper 'allocate' instruction can be used instead
of 'allocate_zero'. There is also the possibility that the
allocate instruction can be be combined with a 'test_heap'
instruction.
* Each stack frame for each arm of the 'case' will have exactly as
many slots as needed.
|
|
When rewriting tuple matching of the first element of a tuple to an
is_tagged_tuple instruction, the get_tuple_element instruction that
fetches the tag will be left unless the register that is fetched is
subsequently killed.
We can do better than that. If the register is referenced in an
allocating instruction, but its value is never actually used, we
can do one of two things: if the value is known to be defined earlier
(using annotations added by beam_utils:anno_defs/1) the instruction
can be removed altogether; if not, it can be replaced with a
'move nil TagRegister' instruction.
|
|
Use annotations added by beam_utils:anno_defs/1 to move more
allocations upwards in the instruction stream. That in turn
allows us to optimize away more 'move' instructions.
|
|
To avoid having to call both is_killed/3 and is_not_used/3,
add usage/3 to answer both questions in one call.
|
|
Add beam_utils:anno_defs/1 which will add an annotation to the
beginning of each block indicating which X registers that are
defined. Having that information can improve some optimizations.
|
|
|
|
Add -MMD option to erlc
OTP-14830
|
|
There are four uncovered lines in combine_heap_needs/2 and
combine_alloc_lists/2. There is no way to reach starting from
Erlang source code using the standard compiler. However, they
can be reached starting from BEAM assembly code, so we don't
want to remove them.
We could add a test case that covers the lines using assembly
code, but an easier solution is to rewrite the code in a more
generic way using sofs so that the code can be covered with
existing test cases.
|
|
01835845579e9 fixed some problems, but introduced a bug where
is_not_used/3 would report that a register was not used when it
in fact was.
|
|
758712d6294 changed the need_heap/2 function so that it stopped
using its second argument.
Remove the second argument from need_heap(), and update all callers
to similarly remove unused arguments.
|
|
|
|
|
|
Add syntax in try/catch to retrieve the stacktrace directly
|
|
It turns out that we don't need to keep track of locked
variables, because the locked variables are always the same
variables that will be alive after a #k_guard_break{}.
|
|
Remove handling of #k_match{} in bsm_rename_ctx/4.
It can never be reached because bsm_rename_ctx/4 will never recurse
into a block that is not in the scope of a #k_protected{}, and
in a #k_protected{}, #k_match{} is not allowed.
|
|
Put guard_cg_list/6 directly after guard_cg/5.
|
|
The function guard_cg/5 handles constructs found within
the records #k_guard_clause{] and #k_protected{}.
Since #k_guard_clause{} can only contain a #k_protected{},
and #k_protected{} in turn cannot contain a #cg_block{},
the clause for handling #cg_block{} in guard_cg/5 is never
executed and can be removed.
|
|
The variable being added will already be there (added by v3_kernel).
|
|
When converting a comparison BIF (such as '=:=') to a test
instruction, run the other optimizations on the result.
When trying to combine is_eq_exact tests, handle the case
that is_eq_exact is followed by a jump instead of a label
to handle a test that has been newly converted from a BIF.
Taken together, those changes will coalesce more is_eq_exact
instructions into select_val instructions.
|
|
A 'case' or 'if' that does not occur last in a function clause will
always force a stack frame. The reasoning behind this is that in most
uses of 'case' there will be a function call from within the
'case'. When there is a function call, the stack frame is needed both
to save the continuation pointer and to save any X registers that will
need to survive the call.
When there is no function call from a 'case', the resulting stack
frame is annoying. There will be register shuffling, and the existence
of the stack frame may thwart many optimizations (for example, in
beam_dead).
Therefore, add an extra pass to v3_codegen to avoid creating a
stack frame when not needed.
https://bugs.erlang.org/browse/ERL-514
|
|
The compile option makedep_side_effect, erlc -MMD, instructs
the compiler to emit dependencies and continue to compile
as normal.
|