| Age | Commit message (Collapse) | Author |
|---|
|
* bjorn/compiler/fix-unsafe-type-inference/OTP-15838:
Fix unsafe negative type inference
|
|
The type optimizer pass (`beam_ssa_type`) could make unsafe
negative inferences. That is, incorrectly infer that a variable
could *not* have a particular type.
This bug was found when adding another optimization. It is not
clear how write a failing test case without that added optimization.
|
|
* bjorn/compiler/fix-receive-patch/ERL-950/OTP-15832:
Eliminate compiler crash when compiling complex receive statements
|
|
* bjorn/compiler/fix-freeze/ERL-948/OTP-15828:
Fix non-terminating compilation
|
|
Certain complex receive statements would result in an internal
compiler failure. That would happen when the compiler would fail
to find the common exit block following a receive. See the added
test case for an example.
https://bugs.erlang.org/browse/ERL-950
|
|
The compiler would not terminate while compiling the following code:
foo(<<N:32>>, Tuple, NewValue) ->
_ = element(N, Tuple),
setelement(N, Tuple, NewValue).
The type analysis pass would attempt to construct a huge list when
attempting analyse the type of `Tuple` after the call to
`setelement/3`.
https://bugs.erlang.org/browse/ERL-948
|
|
Code such as the following would crash the compiler in OTP 22:
[some_atom = fun some_function/1]
The reason is that the fun would be copied (used both in the match
operation and as a value in the list), and the copy of the fun would
create two wrapper functions with the same name for calling
some_function/1. In OTP 21, the duplicate functions happened not to
cause any harm (one of the wrappers functions would be unused and
ultimately be removed by beam_clean). In OTP 22, the new beam_ssa_type
pass would be confused by the multiple definitions of the wrapper
function.
|
|
* john/compiler/fix-missing-match-reposition/ERL-923:
compiler: Propagate match context position on fail path
|
|
|
|
|
|
`beam_asm` would encode `{literal,[]}`, `{literal,erlang}`, and
`{literal,42}` in a less efficient way than the equivalent values
`nil`, `{atom,erlang}`, and `{integer,42}`. That would increase the
size of BEAM files and could increase the loaded code size. It would
probably not harm performance, because `literal` was only used this
way in code that generates `badmatch` and `case_clause` exceptions.
|
|
|
|
|
|
The core_parse.hrl can be used in applications with the
+warn_untyped_record compiler flag.
|
|
|
|
* bjorn/compiler/cuddle-with-tests:
Verify the highest opcode for the r21 test suites
Add test_lib:highest_opcode/1
sys_core_fold: Simplify case_expand_var/2
beam_validator: Remove uncovered lines in lists_mod_return_type/3
Cover return type determination of lists functions
|
|
Validation could fail when a function that never returned was used
in a try block (see attached test case). It's possible to solve
this without disabling the optimization as the generated code is
sound, but I'm not comfortable making such a large change this
close to the OTP 22 release.
|
|
5239eb0c62a9 stopped storing patterns that a variable has matched.
The only tuples that are stored in the type database are tuples
that have been previously constructed.
Therefore, the code in sys_core_fold:case_expand_var/2 and friends
that converts a tuple pattern to a tuple construction can be simplified
to used the stored tuple directly.
|
|
Don't bother infering the return types for lists function that
beam_ssa_type does not handle.
|
|
* bjorn/hipe-compilation/OTP-15596:
HiPE: Don't fail the compilation for unimplemented instructions
|
|
|
|
|
|
For reasons better explained in the source code, ssa_opt_float
skips optimizing inside guards but it failed to do so
consistently; while the pass never processed guard blocks, it was
still possible to erroneously defer error checking to a guard
block, crashing the compiler once it realized its state was
invalid.
|
|
This ensures that unreachable branches are properly ignored on
repeated checks (although tuple type subtraction isn't complete
yet).
|
|
Type subtraction never resulted in the 'none' type, even when it
was obvious that it should. Once that was fixed it became apparent
that inequality checks also fell into the same subtraction trap
that the type pass warned about in a comment.
This then led to another funny problem with select_val, consider
the following code:
{bif,'>=',{f,0},[{x,0},{integer,1}],{x,0}}.
{select_val,{x,0},{f,70},{list,[{atom,false},{f,69},
{atom,true},{f,68}]}}.
The validator knows that '>=' can only return a boolean, so once it
has subtracted 'false' and 'true' it killed the state because all
all valid branches had been taken, so validation would crash once
it tried to branch off the fail label.
|
|
|
|
|
|
|
|
The current type conflict resolution works well for the example
case in the comment, but doesn't handle branched code properly,
consider the following:
{label,2}.
{test,is_tagged_tuple,{f,ignored},[{x,0},3,{atom,r}]}.
{allocate_zero,2,1}.
{move,{x,0},{y,0}}.
%% {y,0} is known to be {r, _, _} now.
{get_tuple_element,{x,0},2,{x,0}}.
{'try',{y,1},{f,3}}.
%% ... snip ...
{jump,{f,5}}.
{label,3}.
{try_case,{y,1}}.
%% {x,0} is the error class (an atom), {x,1} is the error term.
{test,is_eq_exact,{f,ignored},[{x,0},{y,0}]}.
%% ... since tuples and atoms can't meet, the type of {y,0} is
%% now {atom,[]} because the current code assumes the type
%% we're updating with.
{move,{x,1},{x,0}}.
{jump,{f,5}}.
{label,5}.
%% ... joining tuple (block 2) and atom (block 3) means 'term',
%% so the get_tuple_element instruction fails to validate
%% despite this being unrechable from block 3.
{test_heap,3,1}.
{get_tuple_element,{y,0},1,{x,1}}.
{put_tuple2,{x,0},{list,[{x,1},{x,0}]}}.
{deallocate,2}.
return.
This commit kills the state on type conflicts, making unreachable
instructions truly unreachable.
|
|
While complex_test made certain branching instructions a lot easier
to read, we're still using `branch_state` for many others which is
hard to read and makes it impossible to "abort" branches on type
conflicts.
This commit replaces nearly all uses of `branch_state` with a
general branching mechanism, improving readability and paving the
way for proper type conflict resolution.
|
|
The element type can not be extracted before the tuple type has
been updated.
|
|
|
|
Move size=all binary clause pruning to v3_kernel
|
|
Tune BEAM instructions for the new compiler (part 1)
|
|
Optimize the beam_ssa_dead sub pass
|
|
The advantage of moving it up is that it reduces the
size of the code emitted by v3_kernel, speeding
v3_kernel itself and beam_kernel_to_ssa pass.
|
|
Prior to this patch, v3_kernel would do multiple
passes on the clauses to group them. This commit
unrolls those passes, making v3_kernel up to 10%
faster in those cases.
|
|
|
|
This is cleaner and slightly faster.
|
|
The general complexity of the shortcut sub pass of `beam_ssa_dead` is
quadratic, but those optimizations will reduce the constant factor
somewhat.
|
|
Refactor the code to avoid putting any variable from a skippable block
into the set of unset variables. Keeping the set of unset variables as
small as possible will make beam_ssa_dead almost twice as fast when
compiling lib/unicode/tokenizer.ex in elixir.
|
|
Consider this code:
foo(X) ->
case X of
{ok,A} -> A;
error -> X
end.
The `is_tagged_tuple` instruction would not be used
because not all instructions in the tuple matching
sequence had the same failure label:
function t:foo(_0) {
0:
@ssa_bool:7 = bif:is_tuple _0
br @ssa_bool:7, label 8, label 4
8:
@ssa_arity = bif:tuple_size _0
@ssa_bool:9 = bif:'=:=' @ssa_arity, literal 2
br @ssa_bool:9, label 6, label 3
6:
_4 = get_tuple_element _0, literal 0
@ssa_bool = bif:'=:=' _4, literal ok
br @ssa_bool, label 5, label 3
5:
_3 = get_tuple_element _0, literal 1
ret _3
4:
@ssa_bool:11 = bif:'=:=' _0, literal error
br @ssa_bool:11, label 10, label 3
10:
ret _0
3:
_2 = put_tuple literal case_clause, _0
%% t.erl:5
@ssa_ret:12 = call remote (literal erlang):(literal error)/1, _2
ret @ssa_ret:12
}
Enhance the ssa_opt_record optimization to use
`is_tagged_tuple` even if all failure labels are not the
same:
function t:foo(_0) {
0:
@ssa_bool:7 = bif:is_tuple _0
br @ssa_bool:7, label 8, label 4
8:
@ssa_bool:9 = is_tagged_tuple _0, literal 2, literal ok
br @ssa_bool:9, label 6, label 3
6:
_3 = get_tuple_element _0, literal 1
ret _3
4:
@ssa_bool:11 = bif:'=:=' _0, literal error
br @ssa_bool:11, label 10, label 3
10:
ret _0
3:
_2 = put_tuple literal case_clause, _0
%% t.erl:5
@ssa_ret:12 = call remote (literal erlang):(literal error)/1, _2
ret @ssa_ret:12
}
The tuple test will be repeated, but since four instructions
are replaced by two instructions, the code will still be faster
and smaller.
|
|
|
|
|
|
We used to cheat by checking if it were possible to meet the Given
and Required types, which caught the most common problems but
potentially let tuple element conflicts pass through.
This was a compromise to let the thing "work" while we were
refactoring the validator, but we can be a lot stricter now that
its type tracking capabilities approach those of the type
optimization pass.
|
|
Building terms with fragile contents is okay because the GC is
disabled during loop_rec, and the resulting term won't be reachable
from the root set afterwards.
ERL-862
|
|
|
|
This is possible now that we track types on a per-value basis, and
no longer need to care whether the source tuple's register has been
clobbered by the time we infer the type.
|
|
|
|
This is a rather subtle but important distinction. While tracking
types on a per-register basis is fairly effective, it forces us to
track which registers alias each other, and makes it tricky to infer
types over large blocks of code as instruction arguments may have
been clobbered between definition and inference.
Tracking types on a per-value basis makes us immune to these
problems.
|
