| Age | Commit message (Collapse) | Author |
|---|
|
`sys_core_fold` has an optimization of repeated pattern matching.
For example, when a record is matched the first time, the pattern
is remembered. When the same record is matched again, the matching
does not need to be repeated, but the variables bound in the first
matching can be re-used.
It turns out that that there is a name capture problem when the old
inliner is used. The old inliner is used when explicitly inling
certain functions, and by the compiler test suites for testing the
compiler.
The name capture problem could be eliminated by more aggressive
variable renaming when inlining.
But, fortunately, given the new SSA passes, this optimization is no
longer as essential as it used to be. Removing the optimization
turns out to be mostly benefical, leading to a smaller stack
frame in many cases.
Also remove the optimizations of `element/2`, `is_record/3`, and
`setelement/3` from `sys_core_fold`. Because matched patterns are no
longer remembered, those optimizations can very rarely be applied any
more. (Those same optimizations are already done in `beam_ssa_type`.)
|
|
Fix a bug where the number of live registers in a `bs_get_tail`
instruction was too low.
Consider this example:
-export([bs_get_tail/2]).
bs_get_tail(Bin, Config) ->
bs_get_tail_1(Bin, 0, 0, Config).
bs_get_tail_1(<<_:32, Rest/binary>>, Z1, Z2, F1) ->
{Rest,Z1,Z2,F1}.
`beam_validator` would emit the following diagnostics:
t: function bs_get_tail_1/4+2:
Internal consistency check failed - please report this bug.
Instruction: {func_info,{atom,t},{atom,bs_get_tail_1},4}
Error: {uninitialized_reg,{x,3}}:
Here is the part of the code that generates the `function_clause`
exception before the optimization:
{test_heap,6,4}.
{put_list,{x,3},nil,{x,2}}.
{put_list,{integer,0},{x,2},{x,2}}.
{put_list,{integer,0},{x,2},{x,2}}.
{bs_set_position,{x,1},{x,0}}.
{bs_get_tail,{x,1},{x,0},3}. %3 live registers.
{test_heap,2,3}.
{put_list,{x,0},{x,2},{x,1}}.
{move,{atom,function_clause},{x,0}}.
{line,[{location,"t.erl",8}]}.
{call_ext_only,2,{extfunc,erlang,error,2}}.
The `bs_get_tail` instruction expects that 3 registers will be live
at this point. `beam_except` rewrites the code like this:
{bs_set_position,{x,1},{x,0}}.
{bs_get_tail,{x,1},{x,0},3}. %Still 3. Too low.
{move,{integer,0},{x,1}}.
{move,{integer,0},{x,2}}.
{jump,{f,3}}.
Now the number of live registers in `bs_get_tail` is too low,
because the `{x,3}` register will become undefined.
This commit adds code to update the number of live registers
in the `bs_get_tail` instruction, producing this code:
{bs_set_position,{x,1},{x,0}}.
{bs_get_tail,{x,1},{x,0},4}. %Adjusted to 4.
{move,{integer,0},{x,1}}.
{move,{integer,0},{x,2}}.
{jump,{f,3}}.
|
|
There is an optimization for reducing the number of instructions needed
to generate a `function_clause`. After the latest improvements of the
type optimization pass, that optimization is not always applied.
Here is an example:
-export([foo/3]).
foo(X, Y, Z) ->
bar(a, X, Y, Z).
bar(a, X, Y, Z) when is_tuple(X) ->
{X,Y,Z}.
Note that the compiler internally adds a clause to each function to
generate a `function_clause` exception. Thus:
bar(a, X, Y, Z) when is_tuple(X) ->
{X,Y,Z};
bar(A1, A2, A3, A4) ->
erlang:error(function_clause, [A1,A2,A3,A4]).
Optimizations will rewrite the code basically like this:
bar(_, X, Y, Z) when is_tuple(X) ->
{X,Y,Z};
bar(_, A2, A3, A4) ->
erlang:error(function_clause, [a,A2,A3,A4]).
Note the `a` as the first element of the list of arguments. It
will prevent the optimization of the `function_clause` exception.
The BEAM code for `bar/4` looks like this:
{function, bar, 4, 4}.
{label,3}.
{line,[{location,"t.erl",8}]}.
{func_info,{atom,t},{atom,bar},4}.
{label,4}.
{'%',{type_info,{x,0},{atom,a}}}.
{test,is_tuple,{f,5},[{x,1}]}.
{test_heap,4,4}.
{put_tuple2,{x,0},{list,[{x,1},{x,2},{x,3}]}}.
return.
{label,5}.
{test_heap,8,4}.
{put_list,{x,3},nil,{x,0}}.
{put_list,{x,2},{x,0},{x,0}}.
{put_list,{x,1},{x,0},{x,0}}.
{put_list,{atom,a},{x,0},{x,1}}.
{move,{atom,function_clause},{x,0}}.
{line,[{location,"t.erl",8}]}.
{call_ext,2,{extfunc,erlang,error,2}}.
The code after label 5 is the clause that generates the
`function_clause` exception.
This commit generalizes the optimization so that it can be applied for
this function:
{function, bar, 4, 4}.
{label,3}.
{line,[{location,"t.erl",8}]}.
{func_info,{atom,t},{atom,bar},4}.
{label,4}.
{'%',{type_info,{x,0},{atom,a}}}.
{test,is_tuple,{f,5},[{x,1}]}.
{test_heap,4,4}.
{put_tuple2,{x,0},{list,[{x,1},{x,2},{x,3}]}}.
return.
{label,5}.
{move,{atom,a},{x,0}}.
{jump,{f,3}}.
For this particular function, it would be safe to omit the
`move` instruction before the `{jump,{f,3}}` instruction, but
it would not be safe in general to omit `move` instructions.
|
|
The translation from `error(function_clause, Args)` to a jump
to the `func_info` label is not safe if there is a stack frame.
|
|
The `beam_except` pass replaces some calls to `erlang:error/1` or
`erlang:error/2` with specialized instructions in order to reduce the
size of the BEAM code.
In functions that do binary matching, `beam_except` would fail to
translate the instructions that generate a `function_clause`
exception. Here is an example:
bsum(<<H:16,T/binary>>, Acc) ->
bsum(T, Acc+H);
bsum(<<>>, Acc) ->
Acc.
The BEAM code that generates the `function_clause` exception looks
like this:
{label,4}.
{test_heap,2,3}.
{put_list,{x,1},nil,{x,1}}.
%% The following two instructions prevents the translation.
{bs_set_position,{x,2},{x,0}}.
{bs_get_tail,{x,2},{x,0},3}.
{test_heap,2,2}.
{put_list,{x,0},{x,1},{x,1}}.
{move,{atom,function_clause},{x,0}}.
{line,...}.
{call_ext,2,{extfunc,erlang,error,2}}.
Make the translation of `function_clause` exceptions smarter, so
that the following code will be produced:
{label,4}.
{bs_set_position,{x,2},{x,0}}.
{bs_get_tail,{x,2},{x,0},3}.
{jump,{f,1}}. %Jump to `func_info` instruction.
(This issue was noticed when looking at the code generated by
https://github.com/tomas-abrahamsson/gpb.)
|
|
Sometimes when building a tuple, there is no way to avoid an
extra `move` instruction. Consider this code:
make_tuple(A) -> {ok,A}.
The corresponding BEAM code looks like this:
{test_heap,3,1}.
{put_tuple,2,{x,1}}.
{put,{atom,ok}}.
{put,{x,0}}.
{move,{x,1},{x,0}}.
return.
To avoid overwriting the source register `{x,0}`, a `move`
instruction is necessary.
The problem doesn't exist when building a list:
%% build_list(A) -> [A].
{test_heap,2,1}.
{put_list,{x,0},nil,{x,0}}.
return.
Introduce a new `put_tuple2` instruction that builds a tuple in a
single instruction, so that the `move` instruction can be eliminated:
%% make_tuple(A) -> {ok,A}.
{test_heap,3,1}.
{put_tuple2,{x,0},{list,[{atom,ok},{x,0}]}}.
return.
Note that the BEAM loader already combines `put_tuple` and `put`
instructions into an internal instruction similar to `put_tuple2`.
Therefore the introduction of the new instruction will not speed up
execution of tuple building itself, but it will be less work for
the loader to load the new instruction.
|
|
Don't match exact BEAM instructions when trying to recognize
function_clause exceptions that should be replaced with a jump to the
func_info instruction. Instead, do a symbolic evaluation of the list
building code and see if the result is a list of the argument
registers.
While at it, also teach fix_block_1/2 to completely remove a test_heap
instruction before an exception generation instruction.
|
|
|
|
|
|
|
|
A negative allocation could be calculated if a block had multiple
allocations. Make sure to process the block in the right order
so that the correct allocation is processed. Also add an assertion.
This bug was often not noticed because beam_type usually silently
recalculates the allocation amount in test_heap/2 instructions.
|
|
|
|
|
|
Code such as:
bar(X) ->
case {X+1} of
1 -> ok
end.
would crash the beam_except pass of the compiler.
The reason for the crash is that the '+' operator would add a line/1
instruction that the beam_except pass was not prepared to handle.
Reported-by: Erik Søe Sørensen
|
|
Commits 53bd4974a101 and 726f6e4c7afe simplified the handling of
match_fail (used to generated exceptions such as 'function_clause')
by first rewriting them to a call to erlang/error{1,2} and later
rewriting them to specialized BEAM instructions (to reduce the
code size).
There was one flaw, though, which only was exposed when more
aggressive optimizations were added in c3b60f86c622. Here is an
example to explain it:
t(V) ->
fun(get) -> V end.
The following BEAM code will be initially generated for the fun:
{function, '-t/1-fun-0-', 2, 5}.
{label,1}.
{line,[{location,"t.erl",5}]}.
{func_info,{atom,t},{atom,'-t/1-fun-0-'},2}.
{label,2}.
{test,is_eq_exact,{f,2},[{x,0},{atom,get}]}.
{move,{x,1},{x,0}}.
return.
{label,2}.
{test_heap,2,1}.
{put_list,{x,0},nil,{x,1}}.
{move,{atom,function_clause},{x,0}}.
{line,[{location,"t.erl",5}]}.
{call_ext_only,2,{extfunc,erlang,error,2}}.
Translating back to Erlang code, that would be roughly:
'-t/1-fun-0-'(get, V) -> V;
'-t/1-fun-0-'(Arg1, _) -> erlang:error(function_clause, [Arg1]).
Note that the second argument (the free variable V) is not included
in the call to erlang:error/2.
The beam_except pass will simplify the code to:
{function, '-t/1-fun-0-', 2, 8}.
{label,1}.
{line,[{location,"t.erl",5}]}.
{func_info,{atom,t},{atom,'-t/1-fun-0-'},2}.
{label,2}.
{test,is_eq_exact,{f,1},[{x,0},{atom,get}]}.
{move,{x,1},{x,0}}.
return.
The code has been shortened by jumping to the func_info/3 instruction.
Translating back to Erlang:
'-t/1-fun-0-'(get, V) -> V;
'-t/1-fun-0-'(Arg1, Arg2) -> erlang:error(function_clause, [Arg1,Arg2]).
it is clear that both arguments are now included in the
'function_clause' exception, even though the initially generated
code only included the first argument.
That is no problem in this particular case, but for some more complex
funs, optimizing the first version based on variable usage could make
the second version unsafe.
I rejected the following potential solutions:
- Including the free arguments in the call to erlang:error/2:
'-t/1-fun-0-'(get, V) -> V;
'-t/1-fun-0-'(Arg1, Arg2) -> erlang:error(function_clause, [Arg1,Arg2]).
Unfortunately, that is tricky. The free variables are only known
after the second pass in v3_kernel when variable usage has been
calculated. We would need to add a third pass (only for funs) that
would the free arguments to the second argument for erlang:error/2
*and* update the variable usage information.
- Calling beam_except earlier, from within beam_block before any
optimizations based on variable usages are done. But means that the
problem could reappear in some other form in the future when other
updates are done to the code generator and/or optimization passes.
The solution I have chosen is to modify beam_except to only replace
a call to erlang:error(function_class, Args) if the length of Args
is the same as the arity in the func_info/3 instruction. The code
will be slightly larger. Also, the free variables for funs and list
comprehensions will no longer be included in the function_clause
exception (that could be less confusing, but it also means less
information during debugging).
|
|
|
|
Somewhat reduce the code bloat by eliminating special cases.
|
|
In order to save space, rewrite suitable calls to erlang:error/{1,2}
to special BEAM instructions.
This code is probably longer than the code taken out of v3_life and
v3_codegen in the previous commit, but it is much easier to
understand and maintain since the BEAM assembler format is better
understood than the v3_life format.
|
