| Age | Commit message (Collapse) | Author |
|---|
|
The expansion of record field updates, when more than one field is
updated, but not a majority of the fields, will create a sequence of
calls to `erlang:setelement(Index, Value, Tuple)` where Tuple in the
first call is the original record tuple, and in the subsequent calls
Tuple is the result of the previous call. Furthermore, all Index
values are constant positive integers, and the first call to
`setelement` will have the greatest index. Thus all the following
calls do not actually need to test at run-time whether Tuple has type
tuple, nor that the index is within the tuple bounds.
Since OTP R7, the `sys_core_dsetel` pass, run as the very last Core
Erlang pass, has optimized this sequence of `setelement` calls to use
a special destructive version of `setelement` (called
`set_tuple_element`) for all but the very first `setelement` in the
sequence.
It turns out that the presence of the `set_tuple_element` in SSA code
is awkward and can prevent or complicate type analysis and aggressive
optimizations.
Therefore, this commit removes the `sys_core_dsetel` pass and
reimplements it for SSA code. The optimization will be done in the
`beam_ssa_pre_codegen` pass (that is, just before code generation and
after running all other SSA code optimization passes).
In most cases, the resulting BEAM code is identical to previous
code. For a few modules, the BEAM code is actually slightly better,
with smaller stack frames.
|
|
|
|
Do the optimizations of bs_put* instructions in beam_ssa_opt
and remove the beam_bs pass. This can lead to a slight improvement
of compilation times.
|
|
Some lines in beam_peep were no longer covered when the sharing optimization
was added to beam_ssa_opt. Also remove some code from beam_peep that no
longer seems possible to cover.
|
|
Share code for semantically equivalent blocks referred to to by `br`
and `switch` instructions.
A similar optimization is done in `beam_jump`, but doing it here as
well is beneficial as it may enable other optimizations. Also, if
there are many semantically equivalent clauses, this optimization can
substanstially decrease compilation times.
|
|
This will help investigation of compiler bugs.
|
|
This commit improves the bit-syntax match optimization pass,
leveraging the new SSA intermediate format to perform much more
aggressive optimizations. Some highlights:
* Watch contexts can be reused even after being passed to a
function or being used in a try block.
* Sub-binaries are no longer eagerly extracted, making it far
easier to keep "happy paths" free from binary creation.
* Trivial wrapper functions no longer disable context reuse.
|
|
Most of the optimizations in beam_dead have been superseded
by the optimizations in beam_ssa_dead.
The forward/1 pass of beam_dead has been moved to beam_jump.
The beam_split pass splits blocks that contain instructions with
non-zero labels. Because there are no optimizations left that optimize
instructions within blocks, beam_block never needs to put such
instructions into blocks in the first place. beam_split also moved
'move' instructions out block to help beam_dead. That is no longer
necessary since beam_dead no longer exists.
|
|
v3_codegen is replaced by three new passes:
* beam_kernel_to_ssa which translates the Kernel Erlang format
to a new SSA-based intermediate format.
* beam_ssa_pre_codegen which prepares the SSA-based format
for code generation, including register allocation. Registers
are allocated using the linear scan algorithm.
* beam_ssa_codegen which generates BEAM assembly code from the
SSA-based format.
It easier and more effective to optimize the SSA-based format before X
and Y registers have been assigned. The current optimization passes
constantly have to make sure no "holes" in the X register assignments
are created (that is, that no X register becomes undefined that an
allocation instruction depends on).
This commit also introduces the following optimizations:
* Replacing of tuple matching of records with the is_tagged_tuple
instruction. (Replacing beam_record.)
* Sinking of get_tuple_element instructions to just before the first
use of the extracted values. As well as potentially avoiding
extracting tuple elements when they are not actually used on all
executions paths, this optimization could also reduce the number
values that will need to be stored in Y registers. (Similar to
beam_reorder, but more effective.)
* Live optimizations, removing the definition of a variable that is
not subsequently used (provided that the operation has no side
effects), as well strength reduction of binary matching by replacing
the extraction of value from a binary with a skip instruction. (Used
to be done by beam_block, beam_utils, and v3_codegen.)
* Removal of redundant bs_restore2 instructions. (Formerly done
by beam_bs.)
* Type-based optimizations across branches. More effective than
the old beam_type pass that only did type-based optimizations in
basic blocks.
* Optimization of floating point instructions. (Formerly done
by beam_type.)
* Optimization of receive statements to introduce recv_mark and
recv_set instructions. More effective with far fewer restrictions
on what instructions are allowed between creating the reference
and entering the receive statement.
* Common subexpression elimination. (Formerly done by beam_block.)
|
|
As a preparation for replacing v3_codegen with a new code generator,
remove unsafe optimization passes. Especially the older compiler
passes have implicit assumptions about how the code is generated.
Remove the optimizations in beam_block (keep the code that creates
blocks) because they are unsafe. beam_block also calls
beam_utils:live_opt/1, which is unsafe.
Remove beam_type because it calls beam_utils:live_opt/1, and also
because it recalculates the number of heaps words and number of live
registers in allocation instructions, thus potentially hiding bugs in
other passes.
Remove beam_receive because it is unsafe.
Remove beam_record because it is the only remaining user
of beam_utils:anno_defs/1.
Remove beam_reorder because it makes much more sense to run it
as an early SSA-based optimization pass.
Remove the now unused functions in beam_utils:
anno_def/1
delete_annos/1
is_killed_block/2
live_opt/1
usage/3
Note that the following test cases will fail because of the
removed optimizations:
compile_SUITE:optimized_guards/1
compile_SUITE:bc_options/1
receive_SUITE:ref_opt/1
|
|
Call test_lib:recompile/1 from init_per_suite/1 instead of
from all/0. That makes it easy to find the log from the
compilation in the log file for the init_per_suite/1 test
case.
|
|
|
|
misc_SUITE:integer_encoding/1 was written to make sure
that big integers were encoding correctly in a reasonable
amount of time. Now that beam_asm will encode big integers
as literals, we can reduce the scope of integer_encode/1.
That will make it significantly faster, especially when
cover is running.
|
|
|
|
The v3_life pass does not do enough to be worth being its own
pass. Essentially it does two things:
* Calculates life-time information starting from the annotations
that v3_kernel provides. That part can be moved into v3_codegen.
* Rewrites the Kernel Erlang records to similar plain tuples
(for example, #k_cons{hd=Hd,tl=Tl} is rewritten to {cons,Hd,Tl}).
That rewriting is not needed and can be eliminated.
|
|
The goal of this pass is to find values that are built from
patterns and generate aliases for those values to remove
pressure from the GC. For example, this code:
example({ok, Val}) ->
{ok, Val}.
shall become:
example({ok, Val} = Tuple) ->
Tuple.
Currently this pass aliases tuple and cons nodes made of literals,
variables and other cons. The tuple/cons may appear anywhere in the
pattern and it will be aliased if used later on.
Notice a tuple/cons made only of literals is not aliased as it may
be part of the literal pool.
|
|
As part of sys_core_fold, variables involved in bit syntax
matching would be annotated when it would be safe for a later
pass to do the delayed sub-binary creation optimization.
An implicit assumption regarding the annotation was that the
code must not be further optimized. That assumption was broken
in 05130e48555891, which introduced a fixpoint iteration
(applying the optimizations until there were no more changes).
That means that a variable could be annotated as safe for
reusing the match context in one iteration, but a later iteration
could rewrite the code in a way that would make the optimization
unsafe.
One way to fix this would be to clear all reuse_for_context
annotations before each iteration. But that would be wasteful.
Instead I chose to fix the problem by moving out the annotation
code to a separate pass (sys_core_bsm) that is run later after
all major optimizations of Core Erlang has been done.
|
|
|
|
|
|
The guard optimizations in v3_kernel has removed the need for
beam_bool.
|
|
|
|
The integer_encoding/1 test is supposed to be run with a tighter
timetrap ensure that encoding of integer in BEAM files is efficient
enough.
|
|
|
|
|
|
Those clause are obsolete and never used by common_test.
|
|
|
|
?config is ugly and not recommended. Use proplists:get_value/2
instead.
|
|
Either rely on the default 30 minutes timetrap, or set the timeout
using the supported methods in common_test.
|
|
As a first step to removing the test_server application as
as its own separate application, change the inclusion of
test_server.hrl to an inclusion of ct.hrl and remove the
inclusion of test_server_line.hrl.
|
|
|
|
|
|
In the future we might want to add more bit syntax optimizations,
but beam_block is already sufficiently complicated. Therefore, move
the bit syntax optimizations out of beam_block into a separate
compiler pass called beam_bs.
|
|
When matching tuples, the pattern matching compiler would generate
code that would fetch all elements of the tuple that will ultimately
be used, *before* testing that (for example) the first element is the
correct record tag. For example:
is_tuple Fail {x,0}
test_arity Fail {x,0} 3
get_tuple_element {x,0} 0 {x,1}
get_tuple_element {x,0} 1 {x,2}
get_tuple_element {x,0} 2 {x,3}
is_eq_exact Fail {x,1} some_tag
If {x,2} and {x,3} are not used at label Fail, we can re-arrange the
code like this:
is_tuple Fail {x,0}
test_arity Fail {x,0} 3
get_tuple_element {x,0} 0 {x,1}
is_eq_exact Fail {x,1} some_tag
get_tuple_element {x,0} 1 {x,2}
get_tuple_element {x,0} 2 {x,3}
Doing that may be beneficial in two ways.
If the branch is taken, we have eliminated the execution of two
unnecessary instructions.
Even if the branch is never or rarely taken, there is the possibility
for more optimizations following the is_eq_exact instructions.
For example, imagine that the code looks like this:
get_tuple_element {x,0} 1 {x,2}
get_tuple_element {x,0} 2 {x,3}
move {x,2} {y,0}
move {x,3} {y,1}
Assuming that {x,2} and {x,3} have no further uses in the code
that follows, that can be rewritten to:
get_tuple_element {x,0} 1 {y,0}
get_tuple_element {x,0} 2 {y,1}
When should we perform this optimization?
At the very latest, it must be done before opt_blocks/1 in
beam_block which does the elimination of unnecessary moves.
Actually, we want do the optimization before the blocks have
been established, since moving instructions out of one block
into another is cumbersome.
Therefore, we will do the optimization in a new pass that is
run before beam_block. A new pass will make debugging easier,
and beam_block already has a fair number of sub passes.
|
|
|
|
The misc_SUITE:integer_encoding/1 test case is annoyingly slow.
Rewrite the encoding of integers in beam_asm to use the
binary:encode_unsigned/1 BIF.
Also tweak the test case itself. Scale the down the maximum
size of the numbers being generated, but also add test of
numbers around boundaries of power of two (which are the numbers
most likely to expose bugs in the encoding).
|
|
The beam_validator catches all exceptions and collect them.
It makes more sense to don't catch 'error' and 'exit' exceptions,
but to just print out the name of the current function and pass
on the exception just as all other compilation passes do. Those
kind of exceptions are the symptoms of the kind of severe but
easily catched bugs that occur during development.
|
|
I have spent too much time lately waiting for 'cover' to finish,
so now its time to optimize the running time of the tests suite
in coverage mode.
Basically, when 'cover' is running, the test suites would not
run any tests in parallel. The reason is that using too many
parallel processes when running 'cover' would be slower than
running them sequentially. But those measurements were made
several years ago, and many improvements have been made to
improve the parallelism of the run-time system.
Experimenting with the test_lib:p_run/2 function, I found that
increasing the number of parallel processes would speed up the
self_compile tests cases in compilation_SUITE. The difference
between using 3 processes or 4 processes was slight, though,
so it seems that we should not use more than 4 processes when
running 'cover'.
We don't want to change test_lib:parallel/0, because there is
no way to limit the number of test cases that will be run in
parallel by common_test. However, there as test suites (such as
andor_SUITE) that don't invoke the compiler at run-time. We can
run the cases in such test suites in parallel even if 'cover'
is running.
|
|
Amend the test suite to call beam_dead as originally intended (and not
beam_block), and modify the input data so that the exception will
occur within the try ... catch block in function/2.
|
|
Run testcases in parallel will make the test suite run slightly
faster. Another reason for this change is that we want more testing
of parallel testcase support in common_test.
|
|
Introduce the mandary beam_a pass that will be run directly after code
generation, and the mandatory beam_z pass that will be run just before
beam_asm. Since these passes surround the optimizations, beam_a can
(for example) do instruction renaming to simplify the optimization
passes and beam_z can undo those renamings.
|
|
|
|
|
|
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.
|
|
|
|
In 3d0f4a3085f11389e5b22d10f96f0cbf08c9337f (an update to conform
with common_test), in all test_lib:recompile(?MODULE) calls, ?MODULE
was changed to the actual name of the module. That would cause
test_lib:recompile/1 to compile the module with the incorrect
compiler options in cloned modules such as record_no_opt_SUITE,
causing worse coverage.
|
|
|
|
|
|
|
|
|
|
|
