| Age | Commit message (Collapse) | Author |
|---|
|
In a select_val instruction, values associated with a label
which is the same as the failure label can be removed. We
already do this optimization in beam_clean, but it is better
do this sort of optimization before the beam_jump pass.
Also rewrite a select_val instruction with a single value
to is_eq_exact instruction followed by a jump instruction.
|
|
We can rewrite more instances of select_val to is_boolean because
it is not necessary that a particular label follows the select_val.
|
|
|
|
|
|
When matching a binary literal as in:
<<"abc">> = Bin
the compiler will produce a sequence of three instructions
(some details in the instructions removed for simplicity):
bs_start_match2 Fail BinReg CtxtReg
bs_match_string Fail CtxtReg "abc"
bs_test_tail2 Fail CtxtReg 0
The sequence can be replaced with:
is_eq_exact Fail BinReg "abc"
|
|
The actual bs_match_string instruction has four operands:
bs_match_string {f,Lbl} Ctxt NumBits {string,ListOfBytes}
However, v3_codegen emits a more compact representation where
the bits to match are packaged in a bitstring:
bs_match_string {f,Lbl} Ctxt Bitstring
Currently, beam_clean:clean_labels/1 will rewrite the compact
representation to the final representation. That is unfortunate
since clean_labels/1 is called by beam_dead, which means that
the less compact representation will be introduced long before
it is actually needed by beam_asm. It will also complicate any
optimizations that we might want to do.
Move the rewriting of bs_match_string from beam_clean:clean_labels/1
to the beam_z pass, which is the last pass executed before
beam_validator and beam_asm.
|
|
|
|
|
|
Optimize away 'not' in sys_core_fold instead of in beam_block
and beam_dead, as we can do a better job in sys_core_fold.
I modified the test suite temporarily to never turn off Core Erlang
modifications and looked at the coverage. With the new optimizations
active in sys_core_fold, the code in beam_block and beam_dead did not
find a single 'not' that it could optimize. That proves that the new
optimization is at least as good as the old one. Manually, I could
also verify that the new optimization would optimize some variations
of 'not' that the old one would not handle.
|
|
|
|
|
|
While we are, clean up the comments and rearrange the code for
clarity. Also add a test to cover the last uncovered line in
beam_dead.erl.
|
|
Better optimizations with less code.
|
|
The BEAM compiler translates code such as:
is_hex_digit(D) when $0 =< D, D =< $9 -> true;
is_hex_digit(D) when $a =< D, D =< $z -> true;
is_hex_digit(D) when $A =< D, D =< $Z -> true;
is_hex_digit(_) -> false.
to something like this:
L0: test is_ge L1 {x,0} 48
test is_ge L1 57 {x,0}
move true {x,0}
return.
L1: test is_ge L2 {x,0} 97
test is_ge L2 122 {x,0}
move true {x,0}
return
L2: test is_ge L3 {x,0} 65
test is_ge L3 90 {x,0}
move true {x,0}
return
L3: move false {x,0}
return
We can see that tests will be repeated even if they cannot possibly
succeed. For instance, if we pass in {x,0} equal to 32, the first
test that {x,0} is greater than or equal to 48 at L0 will fail.
The control will transfer to L1, where it will be tested whether
{x,0} is greater than 97. That test will fail and control
will pass to L2, where again the test will fail.
The compiler can do better by short-circuiting repeating tests:
L0: test is_ge L3 {x,0} 48
test is_ge L1 57 {x,0}
move true {x,0}
return.
L1: test is_ge L2 {x,0} 97
test is_ge L3 122 {x,0}
move true {x,0}
return
L2: test is_ge L3 {x,0} 65
test is_ge L3 90 {x,0}
move true {x,0}
return
L3: move false {x,0}
return
|
|
|
|
Eliminate some code bloat.
|
|
The bs_match_string instruction is used to speed up matching of
binary literals. For example, given this source code:
foo1(<<1,2,3>>) -> ok.
The matching part of the code will look like:
{test,bs_start_match2,{f,1},1,[{x,0},0],{x,0}}.
{test,bs_match_string,{f,3},[{x,0},24,{string,[1,2,3]}]}.
{test,bs_test_tail2,{f,3},[{x,0},0]}.
Nice. However, if we do a simple change to the source code:
foo2(<<1,2,3>>) -> ok;
foo2(<<>>) -> error.
the resulting matching code will look like (sligthly simplified):
{test,bs_start_match2,{f,4},1,[{x,0},0],{x,0}}.
{test,bs_get_integer2,{f,7},1,[{x,0},{integer,8},1,Flags],{x,1}}.
{test,is_eq_exact,{f,8},[{x,1},{integer,1}]}.
{test,bs_match_string,{f,6},[{x,0},16,{string,[2,3]}]}.
{test,bs_test_tail2,{f,6},[{x,0},0]}.
{move,{atom,ok},{x,0}}.
return.
{label,6}.
{bs_restore2,{x,0},{atom,start}}.
{label,7}.
{test,bs_test_tail2,{f,8},[{x,0},0]}.
That is, matching of the first byte is not combined into the
bs_match_string instruction that follows.
Fix this problem by allowing a bs_match_string instruction to be
used if all clauses will match either the same integer literal or
the empty binary.
|
|
To facilitate debugging of compiler bugs, teach the compiler the
'no_dead' option. Since the beam_dead pass used to do the necessary
splitting of basic blocks to expose all labels, we must move that
splitting into a separate pass that is always run.
|
|
|
|
|
|
There is never any empty blocks when beam_dead is invoked.
Even if there were, they will be removed a little bit later in
forward/4.
|
|
In the optimization of binary matching, it seems that two clauses
cannot never be reached. Removing the clauses is safe, since that
would only mean that an opportunity for an optimization is lost
|
|
Because the code generator (v3_codegen) would not include the
same value more than once in a select_val/3 instruction and
because a label can only be referenced by one select_val/3
instruction, there is no way that the correct value could already
be in the gb_tree. (Even if it could happen, this change is
safe because only opportunity for an optimization would be missed;
incorrect code would not be generated.)
|
|
Since the optimizations in forward/4 already depends on some
assumptions on how code is generated anyway, document the
assumptions in a comment and remove the uncoverable code.
|
|
* bg/compiler:
beam_peep: Remove optimization already done by beam_dead
beam_dead: Combine is_eq_exact instructions into select_val instructions
Evaluate is_record/3 at compile-time using type information
Evaluate element/2 at compile-time using type information
erl_expand_records: Replace is_record() with matching
OTP-8668 bg/compiler
The compiler optimizes record operations better.
|
|
Combine a sequence of chained is_eq_exact instructions into
a select_val instruction.
|
|
|
