| Age | Commit message (Collapse) | Author |
|---|
|
Move type-based optimizations from Core Erlang passes to SSA passes
|
|
* maint:
Fix unsafe negative type inference
|
|
* 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.
|
|
Simplify sys_core_fold by removing optimizations by removing the
optimizations that have been obsoleted by the preceding commits.
|
|
Remove is_function/1,2 tests if that are known to never fail.
|
|
|
|
`=:=` is faster than `==`, so when they would return the same result,
we want to replace `==` with `=:=`. There is currently such an optimization
in sys_core_fold, but the optimimization will be more effective if
done in beam_ssa_type because beam_ssa_type has better type information.
|
|
* maint:
Eliminate compiler crash when compiling complex receive statements
|
|
* bjorn/compiler/fix-receive-patch/ERL-950/OTP-15832:
Eliminate compiler crash when compiling complex receive statements
|
|
* maint:
Fix non-terminating compilation
Fix compiler crash when funs were matched
|
|
* bjorn/compiler/fix-freeze/ERL-948/OTP-15828:
Fix non-terminating compilation
|
|
Improve optimization of redundant tests
|
|
Make the swap instruction known to the compiler
|
|
BEAM has had a `swap` instruction for several releases, but it was not
known to the compiler. The loader would translate a sequence of three
`move` instructions to the `swap` instructions, but only when it was
possible to determine that it would be safe.
By making `swap` known to the compiler, it can be applied in more
situations since it is easier for the compiler than for the loader
to ensure that the usage is safe, and the loader shenanigans can be
eliminated.
|
|
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 `beam_ssa_dead` pass is supposed to eliminate tests that are
determined to be redundant based on the outcome of a previous test.
For example, in the following example that repeats a guard test,
the second clause can never be executed:
foo(A) when A >= 42 -> one;
foo(A) when A >= 42 -> two;
foo(_) -> three.
`beam_ssa_dead` should have eliminated the second clause, but
didn't:
{test,is_ge,{f,5},[{x,0},{integer,42}]}.
{move,{atom,one},{x,0}}.
return.
{label,5}.
{test,is_ge,{f,6},[{x,0},{integer,42}]}.
{move,{atom,two},{x,0}}.
return.
{label,6}.
{move,{atom,three},{x,0}}.
return.
Correct the optimization of four different combinations of relational
operations that were too conservate. To ensure the correctness of the
optimization, also add an exahaustive test of all combinations of
relational operations with one variable and one literal. (Also remove
the weak and now redundant coverage tests in `beam_ssa_SUITE`.) With
this correction, the following code will be generated for the example:
{test,is_ge,{f,5},[{x,0},{integer,42}]}.
{move,{atom,one},{x,0}}.
return.
{label,5}.
{move,{atom,three},{x,0}}.
return.
Thanks to Dániel Szoboszlay (@dszoboszlay), whose talk at Code BEAM
STO 2019 made me aware of this missed opportunity for optimization.
|
|
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
|
|
The beam_except pass rewrites certain calls `erlang:error/{1,2}` to
use specialized instructions for common exceptions such as
`{badmatch,Term}`.
Move this optimization to `beam_ssa_pre_codegen` and `beam_ssa_codegen`.
The main reason for this change is that optimization passes operating
on SSA code are easier to maintain than optimization passes working
on BEAM code.
|
|
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.
|
|
|
|
bjorng/bjorn/deprecation-warnings/ERL-904/OTP-15749
Add compiler option for suppressing warnings about removed functions/modules
|
|
An appliction outside of OTP may want to reuse then name of a module
that was previously included in OTP. Therefore, there should be
a way to suppress warnings for removed functions.
|
|
|
|
|
|
|
|
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
|
|
|
|
|
|
* john/compiler/fix-eq-type-infererence-in-validator/ERL-886:
beam_validator: Infer types on both sides of '=:='
|
|
The script would hang if no local hex had been installed previously.
|
|
|
|
Optimize tail-recursive calls of BIFs
OTP-15674
|
|
* maint:
Updated OTP version
Prepare release
|
|
|
|
BEAM currently does not call BIFs at the end of a function in a
tail-recursive way. That is, when calling a BIF at the end of a
function, the BIF is first called, and then the stack frame
is deallocated, and then control is transferred to the caller.
If there is no stack frame when a BIF is called in the tail position,
the loader will emit a sequence of three instructions: first an
instruction that allocates a stack frame and saves the continuation
pointer (`allocate`), then an instruction that calls the BIF
(`call_bif`), and lastly an instruction that deallocates the stack
frame and returns to the caller (`deallocate_return`).
The old compiler would essentially allocate a stack frame for each
clause in a function, so it would not be that common that a BIF was
called in the tail position when there was no stack frame, so the
three-instruction sequence was deemed acceptable.
The new compiler only allocates stack frames when truly needed, so
the three-instruction BIF call sequence has become much more common.
This commit introduces a new `call_bif_only` instruction so that only
one instruction will be needed when calling a BIF in the tail position
when there is no stack frame. This instruction is also used when there
is a stack frame to make it possible to deallocate the stack frame
**before** calling the BIF, which may make a subsequent garbage
collection at the end of the BIF call cheaper (copying less garbage).
The one downside of this change is that the function that called the
BIF will not be included in the stack backtrace (similar to how a
tail-recursive call to an Erlang function will not be included in the
backtrace).
That was the quick summary of the commit. Here comes a detailed look
at how BIF calls are translated by the loader. The first example is a
function that calls `setelement/3` in the tail position:
update_no_stackframe(X) ->
setelement(5, X, new_value).
Here is the BEAM code:
{function, update_no_stackframe, 1, 12}.
{label,11}.
{line,[...]}.
{func_info,{atom,t},{atom,update_no_stackframe},1}.
{label,12}.
{move,{x,0},{x,1}}.
{move,{atom,new_value},{x,2}}.
{move,{integer,5},{x,0}}.
{line,[...]}.
{call_ext_only,3,{extfunc,erlang,setelement,3}}.
Because there is no stack frame, the `call_ext_only` instruction will
be used to call `setelement/3`:
{call_ext_only,3,{extfunc,erlang,setelement,3}}.
The loader will transform this instruction to a three-instruction
sequence:
0000000020BD8130: allocate_tt 0 3
0000000020BD8138: call_bif_e erlang:setelement/3
0000000020BD8148: deallocate_return_Q 0
Using the `call_bif_only` instruction introduced in this commit,
only one instruction is needed:
000000005DC377F0: call_bif_only_e erlang:setelement/3
`call_bif_only` calls the BIF and returns to the caller.
Now let's look at a function that already has a stack frame when
`setelement/3` is called:
update_with_stackframe(X) ->
foobar(X),
setelement(5, X, new_value).
Here is the BEAM code:
{function, update_with_stackframe, 1, 14}.
{label,13}.
{line,[...]}.
{func_info,{atom,t},{atom,update_with_stackframe},1}.
{label,14}.
{allocate,1,1}.
{move,{x,0},{y,0}}.
{line,[...]}.
{call,1,{f,16}}.
{move,{y,0},{x,1}}.
{move,{atom,new_value},{x,2}}.
{move,{integer,5},{x,0}}.
{line,[...]}.
{call_ext_last,3,{extfunc,erlang,setelement,3},1}.
Since there is a stack frame, the `call_ext_last` instruction will be used
to deallocate the stack frame and call the function:
{call_ext_last,3,{extfunc,erlang,setelement,3},1}.
Before this commit, the loader would translate this instruction to:
0000000020BD81B8: call_bif_e erlang:setelement/3
0000000020BD81C8: deallocate_return_Q 1
That is, the BIF is called before deallocating the stack frame and returning
to the calling function.
After this commit, the loader will translate the `call_ext_last` like this:
000000005DC37868: deallocate_Q 1
000000005DC37870: call_bif_only_e erlang:setelement/3
There are still two instructions, but now the stack frame will be
deallocated before calling the BIF, which could make the potential
garbage collection after the BIF call slightly more efficient (copying
less garbage).
We could have introduced a `call_bif_last` instruction, but the code
for calling a BIF is relatively large and there does not seem be a
practical way to share the code between `call_bif` and `call_bif_only`
(since the difference is at the end, after the BIF call). Therefore,
we did not want to clone the BIF calling code yet another time to
make a `call_bif_last` instruction.
|
