| Age | Commit message (Collapse) | Author |
|---|
|
62473daf introduced an unsafe optimization in the loader.
See the comments in the test case for an explanation of
the problem.
|
|
The perf_counter is a very very cheap and high resolution timer
that can be used to timestamp system events. It does not have
monoticity guarantees, but should on most OS's expose a monotonous
time.
A special instruction has been created for this counter to further
speed up fetching it.
OTP-12908
|
|
* egil/pd-opt-get/OTP-13167:
erts: Add i_get_hash instruction
erts: Use internal hash for process dictionaries
|
|
Calculate hashvalue in load-time for constant process dictionary gets.
|
|
The combination is_non_empty_list followed by get_list is extremly
common (but not in estone_SUITE, which is why it has not been noticed
before). Therefore it is worthwile to introduce a combined
instruction.
|
|
Not pre-fetching in conditional instructions (instructions that use
-fail_action) seems to improve performance slightly.
The reason for that is that conditional instructions may jump to the
failure label, wasting the pre-fetched instruction. Another reason
is that that many conditional instructions do a function call, and
the register pressure is therefore high. Avoiding the pre-fetch
may reduce the register pressure and pontentially result in more
efficient code.
|
|
It is currently only possible to pack up to 4 operands. However,
the move_window4 instrucion has 5 operands and move_window5 and
move3 instrucations have 6 operands.
Teach beam_makeops to pack instructions with 5 or 6 operands.
Also rewrite the move_window instructions in beam_emu.c to macros
to allow their operands to get packed.
|
|
Update transformations to ensure that the move_call_ext_last
and move_call_ext_last are used.
|
|
Sequences of three move instructionst that effectively swap the
contents of two registers are fairly common. We can replace them
with a swap_temp/3 instruction. The third operand is the temporary
register to be used for swapping, since the temporary register
may actually be used.
If swap_temp/3 instruction is followed by a call, the temporary
register will often (but not always) be killed by the call. If
it is killed, we can replace the swap_temp/3 instruction with a
slightly cheaper swap/2 instruction.
|
|
Currently, move2/2 does the two moves sequentially to ensure
that the instruction will always work correctly.
We can do better than that. If the two move instructions have
any registers in common, we can introduce simpler and slightly
more efficient instructions to handle those cases:
move_shift/3
move_dup/3
For the remaining cases when the the move instructions
have no common registers, the move2/4 instruction can perform
the moves in parallel which is probably slightly more efficient.
For clarity's sake, we will remain the instruction to move2_par/4.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
The i_fetch instruction fetches two operands and places them
in the tmp_arg1 and tmp_arg2 variables. The next instruction
(such as i_plus) does not have to handle different types of
operands, but can get get them simply from the tmp_arg*
variables. Thus, i_fetch was introduced as a way to temper
a potentail combinatorial explosion.
Unfortunately, clang will generate terrible code because of
the tmp_arg1 and tmp_arg2 variables being live across multiple
instructions. Note that Clang has no way to predict the control
flow from one instruction to another. Clang must assume that
any instruction can jump to any other instruction. Somehow GCC
manages to cope with this situation much better.
Therefore, to improve the quality of the code generated by clang, we
must eliminate all uses of the tmp_arg1 and tmp_arg2 variables. This
commit eliminates the use of i_fetch in combination with the
arithmetic and logical instructions.
While we are touching the code for the bsr and bsl instructions,
also move the tmp_big[] array from top scope of process main into
the block that encloses the bsr and bsl instructions.
|
|
The 'r' type is now mandatory. That means in order to handle
both of the following instructions:
move x(0) y(7)
move x(1) y(7)
we would need to define two specific operations in ops.tab:
move r y
move x y
We want to make 'r' operands optional. That is, if we have
only this specific instruction:
move x y
it will match both of the following instructions:
move x(0) y(7)
move x(1) y(7)
Make 'r' optional allows us to save code space when we don't
want to make handling of x(0) a special case, but we can still
use 'r' to optimize commonly used instructions.
|
|
Consider the try_case_end instruction:
try_case_end s
The 's' operand type means that the operand can either be a
literal of one of the types atom, integer, or empty list, or
a register. That worked well before R12. In R12 additional
types of literals where introduced. Because of way the
overloading was done, an 's' operand cannot handle the
new types of literals. Therefore, code such as the following
is necessary in ops.tab to avoid giving an 's' operand a
literal:
try_case_end Literal=q => move Literal x | try_case_end x
While this work, it is error-prone in that it is easy to
forget to add that kind of rule. It would also be complicated
in case we wanted to introduce a new kind of addition operator
such as:
i_plus jssd
Since there are two 's' operands, two scratch registers and
two 'move' instructions would be needed.
Therefore, we'll need to find a smarter way to find tag
register operands. We will overload the pid and port tags
for X and Y register, respectively. That works because pids
and port are immediate values (fit in one word), and there
are no literals for pids and ports.
|
|
|
|
Seen on SSL application where substraction with x registers were prevalent:
* i_minus specialization on x registers
* i_plus specialization on x registers
|
|
Common pattern seen in SSL:
move y x | move r x -> move2
move r x | move y x -> move2
Common pattern seen in SSL and Compiler:
move x r | move x x -> move2
|
|
* i_rem specialization on x registers
|
|
* i_band specialization on x registers and constants
|
|
Move an entire region of x registers to the stack.
This reduces the dispatch pressure of move instructions.
Also introduce a move2 specialization for some common move patterns:
move r y | move x y -> move2 : As above, moving regions to the stack
move x r | move x y -> move2 : A seemingly common pattern
|
|
|
|
* i_is_lt for r, x registers and constants
* i_is_ge for x registers and constants
* i_is_exact_eq for r and x registers
|
|
See the previous commit for justification and use cases.
|
|
Let the loader pre-compute the hash value when a single, literal key
is matched as in:
#{<<"some_key">>:=V} = Map
In my measurements, this optimization resulted in a 30 percent
speedup for short binary keys.
Unfortunately, this optimizization makes no difference for small
maps with less than 32 keys, since the hash value is not used.
Still, there are the following use cases:
* A map used instead of a record with more than 32 entries. I have
seen some applications with huge records.
* Lookup in JSON dictionaries represented as maps.
The hash value will only be used when the map is a hash map
(currently, that means at least 32 entries).
|
|
In the i_get_map_element/4 instruction, for literal keys other than
atoms, the key would be put into x[0] instead of used directly in the
instruction. The reason is that the original implementation of maps
only supported atom keys.
|
|
The map instructions require that the keys in the instructions
are sorted (for flatmaps). But that is an implementation detail
that should not exposed outside of the BEAM virtual machine.
Therefore, make the sorting of the keys the responsibility of
the loader and not the compiler.
Also note that the sort order for maps with numeric keys or keys
with numeric components has changed in OTP 18. That means that
code compiled for OTP 17 that operated on maps with map keys
might not work in OTP 18 without the sorting in the loader
(although it is unlikely to be an issue in practice).
|
|
The has_map_fields instruction is infrequently used. Thus there
is no need to have the fastest possible implementation; it is
better to have an implementation that reduces the code size in
the already big process_main() function.
We can transform has_map_fields to a get_map_elements instruction,
targeting the same unused x[0] register for all keys. That
instruction will only be marginally slower than existing
implementation.
|
|
The compiler will only emit is_map/1 instructions with literal
argument if optimization is turned off. Therefore, the only
reason for this commit is cleanliness.
|
|
The new_map instruction cannot fail, and thus needs no fail label.
|
|
A put_map_assoc instruction with an empty source map should be
converted to a simpler new_map instruction. The transformation
didn't happen because an empty source map is no longer represented
as a NIL term (as it was in the beginning before map literals
were implemented).
|
|
Using the exact operator (':=') is only allowed when an existing map
is being updated. Thus the following causes a compilation error:
#{k:=v}
Therefore there is no need to support the put_map_exact instruction
without a source map.
|
|
For searching a key in an array we use linear search in arrays
up to 10 elements.
Selecting on tuple arity will always use linear search.
Instead of using two different instructions we assume selecting on
different tuple arities are always few in numbers.
|
|
|
|
Move src to a register if it is a literal.
|
|
|
|
Map source may be anything, not only registers.
|
|
|
|
|
|
To make it possible to build the entire OTP system, also define
dummys for the instructions in ops.tab.
|
|
A process requesting a system task to be executed in the context of
another process will be notified by a message when the task has
executed. This message will be on the form:
{RequestType, RequestId, Pid, Result}.
A process requesting a system task to be executed can set priority
on the system task. The requester typically set the same priority
on the task as its own process priority, and by this avoiding
priority inversion. A request for execution of a system task is
made by calling the statically linked in NIF
erts_internal:request_system_task(Pid, Prio, Request). This is an
undocumented ERTS internal function that should remain so. It
should *only* be called from BIF implementations.
Currently defined system tasks are:
* garbage_collect
* check_process_code
Further system tasks can and will be implemented in the future.
The erlang:garbage_collect/[1,2] and erlang:check_process_code/[2,3]
BIFs are now implemented using system tasks. Both the
'garbage_collect' and the 'check_process_code' operations perform
or may perform garbage_collections. By doing these via the
system task functionality all garbage collect operations in the
system will be performed solely in the context of the process
being garbage collected. This makes it possible to later implement
functionality for disabling garbage collection of a process over
context switches.
Newly introduced BIFs:
* erlang:garbage_collect/2 - The new second argument is an option
list. Introduced option:
* {async, RequestId} - making it possible for users to issue
asynchronous garbage collect requests.
* erlang:check_process_code/3 - The new third argument is an
option list. Introduced options:
* {async, RequestId} - making it possible for users to issue
asynchronous check process code requests.
* {allow_gc, boolean()} - making it possible to issue requests
that aren't allowed to garbage collect (operation will abort
if gc should be needed).
These options have been introduced as a preparation for
parallelization of check_process_code operations when the
code_server is about to purge a module.
|
|
|
|
The loader failed to load non-optimized BEAM code generated from:
element(2, not_a_tuple)
Commit ece4c17d2288a3161c995 introduced such code into
core_fold_SUITE, leading to core_fold_no_opt_SUITE and
core_fold_post_opt_SUITE failing to load.
|
|
Calls to erlang:set_trace_pattern/3 will no longer block all
other schedulers.
We will still go to single-scheduler mode when new code is loaded
for a module that is traced, or when loading code when there is a
default trace pattern set. That is not impossible to fix, but that
requires much closer cooperation between tracing BIFs and the loader
BIFs.
|
|
Change the data structures for breakpoints to make it possible
(in a future commit) to manage breakpoints without taking down the
system to single-scheduling mode.
The current "breakpoint wheel" data structure (a circular,
double-linked list of breakpoints) was invented before the
SMP emulator. To support it in the SMP emulator, there is essentially
one breakpoint wheel per scheduler. As more breakpoint types have
been added, the implementation has become messy and hard to understand
and maintain.
Therefore, the time for a rewrite has come. Use one struct to hold
all breakpoint data for a breakpoint in a function. Use a flag field
to indicate what different type of break actions that are enabled.
|
|
|
