We can't always designate a timer thread, so any sleepers must
also perform ubf wakeups. Note: a similar change needs to be
made for rb_thread_fd_select and rb_wait_for_single_fd.
[ruby-core:88088] [Misc #14937] [Bug #5343]
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64111 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
thread_pthread.c relies on ppoll for rb_sigwait_sleep, so ensure
the compatibility wrapper is available for it.
[Bug #14950]
Reported-by: SHIBATA Hiroshi <hsbt@ruby-lang.org>
Reported-by: Greg L <Greg.mpls@gmail.com>
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64110 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
thread_pthread.c relies on ppoll for rb_sigwait_sleep, so ensure
the compatibility wrapper is available for it.
Reported-by: SHIBATA Hiroshi <hsbt@ruby-lang.org>
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64109 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
To reduce resource use and reduce CI failure; remove
timer-thread. Single-threaded Ruby processes (including forked
children) will never see extra thread overhead. This prevents
glibc and jemalloc from going into multi-threaded mode and
initializing locks or causing fragmentation via arena explosion.
The GVL is implements its own wait-queue as a ccan/list to
permit controlling wakeup order. Timeslice under contention is
handled by a designated timer thread (similar to choosing a
"patrol_thread" for current deadlock checking).
There is only one self-pipe, now, as wakeups for timeslice are
done independently using condition variables. This reduces FD
pressure slightly.
Signal handling is handled directly by a Ruby Thread (instead
of timer-thread) by exposing signal self-pipe to callers of
rb_thread_fd_select, native_sleep, rb_wait_for_single_fd, etc...
Acquiring, using, and releasing the self-pipe is exposed via 4
new internal functions:
1) rb_sigwait_fd_get - exclusively acquire timer_thread_pipe.normal[0]
2) rb_sigwait_fd_sleep - sleep and wait for signal (and no other FDs)
3) rb_sigwait_fd_put - release acquired result from rb_sigwait_fd_get
4) rb_sigwait_fd_migrate - migrate signal handling to another thread
after calling rb_sigwait_fd_put.
rb_sigwait_fd_migrate is necessary for waitpid callers because
only one thread can wait on self-pipe at a time, otherwise a
deadlock will occur if threads fight over the self-pipe.
TRAP_INTERRUPT_MASK is now set for the main thread directly in
signal handler via rb_thread_wakeup_timer_thread.
Originally, I wanted to use POSIX timers
(timer_create/timer_settime) for this. Unfortunately, this
proved unfeasible as Mutex#sleep resumes on spurious wakeups and
test/thread/test_cv.rb::test_condvar_timed_wait failed. Using
pthread_sigmask to mask out SIGVTALRM fixed that test, but
test/fiddle/test_function.rb::test_nogvl_poll proved there'd be
some unavoidable (and frequent) incompatibilities from that
approach.
Finally, this allows us to drop thread_destruct_lock and
interrupt current ec directly.
We don't need to rely on vm->thread_destruct_lock or a coherent
vm->running_thread on any platform. Separate timer-thread for
time slice and signal handling is relegated to thread_win32.c,
now.
[ruby-core:88088] [Misc #14937]
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64107 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
If we keep ubf set after unregistering, there is a window for
other threads (including timer thread) to put this thread back
on the ubf_list right away. Entering ubf_list unexpectedly
after GVL acquisition may cause spurious wakeup and trigger
unexpected behavior.
Finally, clear ubf before acquiring GVL, to since ubf is useless
during GVL acquisition anyways and we don't want to waste cycles
in other threads calling ubf for useless work.
[ruby-core:88141] [Bug #14945]
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64083 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
There's no need to resize each rb_fdset_t to match the size of
the biggest one. This can allow some small memory savings if
watching several sets of FDs simultaneously.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64017 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
When do_select is interrupted and raise happens from
RUBY_VM_CHECK_INTS_BLOCKING, the original FD sets we copied
do not get freed, leading to a memory leak. Wrap up all the
FD sets into a Ruby object to ensure the GC can release an
allocations made for rb_fdset_t.
This leak existed since Ruby 2.0.0 (r36430)
[Bug #14929]
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@64007 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
pthread_atfork is not idempotent and repeatedly calling it
causes it to register the same hook repeatedly; leading to
unbound memory growth.
Ruby already has a (confusing-named) internal API for to call
in the forked child process: rb_thread_atfork
Call the MJIT child_after_fork hook inside that to prevent
unbound growth with the following loop:
loop do
RubyVM::MJIT.pause
RubyVM::MJIT.resume
end
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63884 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
Round up non-zero <1ms timeouts to 1ms and use INT_MAX instead
of infinite (-1) for extremely large timeouts. All of our
ppoll/select callers are able to handle spurious wakeups,
anyways.
This avoids excessive CPU usage and busy waits with short
timeouts to rb_wait_for_single_fd.
CPU usage with the following script is significantly reduced
for systems with "#undef HAVE_PPOLL":
require 'io/wait'
r, w = IO.pipe
Thread.new { loop { r.wait_readable(0.000001) } }.join(5)
exit!(0)
Low-resolution in poll(2) still sucks, though...
Note: I don't see the value in making a similar change to
time_timeval of a <1us sleep is attempted because of GVL
release and syscall latency.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63867 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
Rename it to "ruby_ppoll" so it looks more obvious in debuggers.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63866 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
Reading win32/win32.c waitpid implementation, maybe waitpid(-1, ...)
on that platform will never conflict with mjit use of waitpid.
In any case, I've added WAITPID_USE_SIGCHLD macro to vm_core.h
so it can be easy for Linux/BSD users to test (hopefully!)
win32-compatible code.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63855 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
The change is unstable on Windows. Please re-commit it when it correctly
supports Windows.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63852 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
Use a global SIGCHLD handler to guard all callers of rb_waitpid.
To work safely with multi-threaded programs, we introduce a
VM-wide waitpid_lock to be acquired BEFORE fork/vfork spawns the
process. This is to be combined with the new ruby_waitpid_locked
function used by mjit.c in a non-Ruby thread.
Ruby-level SIGCHLD handlers registered with Signal.trap(:CHLD)
continues to work as before and there should be no regressions
in any existing use cases.
Splitting the wait queues for PID > 0 and groups (PID <= 0)
ensures we favor PID > 0 callers.
The disabling of SIGCHLD in rb_f_system is longer necessary,
as we use deferred signal handling and no longer make ANY
blocking waitpid syscalls in other threads which could "beat"
the waitpid call made by rb_f_system.
We prevent SIGCHLD from firing in normal Ruby Threads and only
enable it in the timer-thread, to prevent spurious wakeups
from in test/-ext-/gvl/test_last_thread.rb with MJIT enabled.
I've tried to guard as much of the code for RUBY_SIGCHLD==0
using C "if" statements rather than CPP "#if" so to reduce
the likelyhood of portability problems as the compiler will
see more code.
We also work to suppress false-positives from
Process.wait(-1, Process::WNOHANG) to quiets warnings from
spec/ruby/core/process/wait2_spec.rb with MJIT enabled.
Lastly, we must implement rb_grantpt for ext/pty. We need a
MJIT-compatible way of supporting grantpt(3) which may spawn
the `pt_chown' binary and call waitpid(2) on it.
[ruby-core:87605] [Ruby trunk Bug#14867]
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63758 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
No point in wasting cycles updating the timespec when not
checking on spurious wakeups.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63719 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
I can't seem to reproduce the maybe-uninitialized warning on
gcc 7 or 8 on Debian sid (7.3.0-16 / 8-20180425-1 r259628),
so the guard from r62305 is dropped.
* thread.c (timeout_prepare): hoist out from do_select
(do_select): ditto
(rb_wait_for_single_fd): use timeout_prepare
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63672 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
Same thing as https://bugs.ruby-lang.org/issues/14798
My easily-confused mind gets function call ordering confused
easily:
sleep_forever(..., TRUE, FALSE);
sleep_forever(..., FALSE, TRUE);
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63647 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
We must not leak EINTR to users in case a signal hits a
ppoll/select caller right when (or just before) the timeout
expires. In other words, the timeout should take precedence
over the -1 result from ppoll or select.
We also try one more time in case of EINTR with a zero timeout,
since technically the syscall finished before timing out if
it returns EINTR.
Regression appeared in r62457
("thread.c (update_timespec): use timespec_update_expire",
commit e6bf0128ad)
and is not in any stable release of Ruby.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63462 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
This allows native_sleep to use less stack (80 -> 64 bytes on
x86-64) for GVL_UNLOCK_BEGIN/END. For future APIs, we will pass
`ec` or `th` around anyways, so the BLOCKING_REGION change
should be beneficial in the future.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63448 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
FreeBSD 11.0+ supports ppoll, so we may use it after accounting
for portability differences in how it treats POLLOUT vs POLLHUP
events as mutually exclusive (as documented in the FreeBSD
poll(2) manpage).
For waiting on high-numbered single FDs, this should put
FreeBSD on equal footing with Linux and should allow cheaper
FD readiness checking with sleepy GC in the future.
* thread.c (USE_POLL, POLLERR_SET): define for FreeBSD 11.0+
(rb_wait_for_single_fd): return all requested events on POLLERR_SET
io.c (USE_POLL): define for FreeBSD 11.0+
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63427 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
While we cannot use LIST_HEAD since r63312, we can at
least use list_head_init to make our code more readable.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63314 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
Address of a variable whose storage duration is `auto` is _not_ a
compile time constant, according to ISO 9899 section 6.4.
LIST_HEAD takes such thing. You can't use it to declare local
variables.
Interestingly, address of a static variable _is_ a compile time
constant. So a declaration like `static LIST_HEAD..` is
completely legal even in C90.
In C99 and newer, this is not a constraint violation.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63312 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
It is unsafe to release GVL and call rb_notify_fd_close after
close(2) on any given FD. FDs (file descriptor) may be recycled
in other threads immediately after close() to point to a different
file description. Note the distinction between "file description"
and "file descriptor".
th-1 | th-2
-------------------------------+---------------------------------------
io_close_fptr |
rb_notify_fd_close(fd) |
fptr_finalize_flush |
close(fd) |
rb_thread_schedule |
| fd reused (via pipe/open/socket/etc)
rb_notify_fd_close(fd) |
| sees "stream closed" exception
| for DIFFERENT file description
* thread.c (rb_thread_io_blocking_region): adjust comment for list_del
* thread.c (rb_notify_fd_close): give busy list to caller
* thread.c (rb_thread_fd_close): loop on busy list
* io.c (io_close_fptr): do not call rb_thread_fd_close on invalid FD
* io.c (io_reopen): use rb_thread_fd_close
Fixes: r57422 ("io.c: close before wait")
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63216 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
Instead of maintaining linked-lists to store all
rb_queue/rb_szqueue/rb_condvar structs; store only a fork_gen
serial number to simplify management of these items.
This reduces initialization costs and avoids the up-front cost
of resetting all Queue/SizedQueue/ConditionVariable objects at
fork while saving 8 bytes per-structure on 64-bit. There are no
savings on 32-bit.
* thread.c (rb_thread_atfork_internal): remove rb_thread_sync_reset_all call
* thread_sync.c (rb_thread_sync_reset_all): remove
* thread_sync.c (queue_live): remove
* thread_sync.c (queue_free): remove
* thread_sync.c (struct rb_queue): s/live/fork_gen/
* thread_sync.c (queue_data_type): use default free
* thread_sync.c (queue_alloc): remove list_add
* thread_sync.c (queue_fork_check): new function
* thread_sync.c (queue_ptr): call queue_fork_check
* thread_sync.c (szqueue_free): remove
* thread_sync.c (szqueue_data_type): use default free
* thread_sync.c (szqueue_alloc): remove list_add
* thread_sync.c (szqueue_ptr): check fork_gen via queue_fork_check
* thread_sync.c (struct rb_condvar): s/live/fork_gen/
* thread_sync.c (condvar_free): remove
* thread_sync.c (cv_data_type): use default free
* thread_sync.c (condvar_ptr): check fork_gen
* thread_sync.c (condvar_alloc): remove list_add
[ruby-core:86316] [Bug #14634]
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63215 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
Instead of allocating and registering the altstack in different
places, do it together to reduce code and improve readability.
When thread cache is enabled, storing altstack in rb_thread_t
is wasteful and we may reuse altstack in the same pthread.
This also lets us clearly allow use of xmalloc to allow GC to
recover from ENOMEM.
[ruby-core:85621] [Feature #14487]
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63213 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
This is fairly non-intrusive bugfix to prevent children
from trying to reach into thread stacks of the parent.
I will probably reuse this idea and redo r62934, too
(same bug).
* vm_core.h (typedef struct rb_vm_struct): add fork_gen counter
* thread.c (rb_thread_atfork_internal): increment fork_gen
* variable.c (struct autoload_data_i): store fork_gen
* variable.c (check_autoload_data): remove (replaced with get_...)
* variable.c (get_autoload_data): check fork_gen when retrieving
* variable.c (check_autoload_required): use get_autoload_data
* variable.c (rb_autoloading_value): ditto
* variable.c (rb_autoload_p): ditto
* variable.c (current_autoload_data): ditto
* variable.c (autoload_reset): reset fork_gen, adjust indent
* variable.c (rb_autoload_load): set fork_gen when setting state
* test/ruby/test_autoload.rb (test_autoload_fork): new test
[ruby-core:86410] [Bug #14634]
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63210 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
* cont.c (root_fiber_alloc): call `ConvertThreadToFiber()` here.
`rb_fiber_t` for root_fiber is allocated before running Threads.
Fiber objects wrapping this rb_fiber_t for root_fiber are created
when root Fiber object is required explicitly (for example, Fiber
switching and so on). We can put calling `ConvertThreadToFiber()`.
In other words, we can pending `ConvertThreadToFiber()`
until Fiber objects are created.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63090 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
* cont.c (rb_threadptr_root_fiber_setup): divide into two functions:
* rb_threadptr_root_fiber_setup_by_parent(): called by the parent thread.
* rb_threadptr_root_fiber_setup_by_child(): called by the created thread.
`rb_threadptr_root_fiber_setup()` is called by the parent thread and
set fib->fib_handle by ConvertThreadToFiber() on the parent thread on
Windows enveironment.
This means that root_fib->fib_handle of child thread is initialized
with parent thread's Fiber handle. Furthermore, second call of
`ConvertThreadToFiber()` for the same thread fails.
This patch solves this weird situateion. However, maybe we can make more
clean code.
* thread.c (thread_start_func_2): call
`rb_threadptr_root_fiber_setup_by_child()` at thread initialize routine.
* vm.c (th_init): call `rb_threadptr_root_fiber_setup_by_parent()`.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63073 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
rb_ensure is insufficient cleanup for fork and we must
reinitialize all waitqueues in the child process.
Unfortunately this increases the footprint of ConditionVariable,
Queue and SizedQueue by 8 bytes on 32-bit (16 bytes on 64-bit).
[ruby-core:86316] [Bug #14634]
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@62934 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
* thread.c (unblock_function_set): check interrupts just once
during raising exceptions, as they are deferred since r16651.
[ruby-core:85939] [Bug #14577]
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@62673 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
We must not maintain references to threads in the parent process
in any mutexes held by the child process.
* thread_sync.c (rb_mutex_cleanup_keeping_mutexes): new function
* thread.c (rb_thread_atfork): cleanup keeping mutexes
[ruby-core:85940] [Bug #14578]
Fixes: r58604 (commit 3586c9e087)
("reduce rb_mutex_t size from 160 to 80 bytes on 64-bit")
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@62668 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
thread.c already includes vm_core.h where USE_SIGALTSTACK is
defined, #include it explicitly (eval_intern.h already includes
it)
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@62473 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
Users may subtract and round into negative values when using
Thread#join, so clamp the timeout to zero to avoid infinite/long
timeouts.
Note: other methods such as Kernel#sleep and IO.select will
raise on negative values, but Thread#join is an outlier *shrug*
This restores Ruby 2.5 (and earlier) behavior.
Fixes: r62182 (commit c915390b95)
("thread.c: avoid FP for Thread#join")
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@62462 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
Using:
strace ruby -e 'Thread.new { sleep }.join(Float::INFINITY)'
Will show a difference in futex() syscall args (not that I'd
ever advocate Float::INFINITY as a Thread#join arg :P)
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@62461 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
No need to waste cycles updating timespecs if there's no expiry.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@62458 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
This hopefully improves readability when comparing timespecs.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@62456 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
Naming the constant timespec as "end" should make it more
apparent is is an absolute time. Update callers, too.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@62455 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
* thread.c (timeval_for): cast to int32_t instead of suseconds_t,
which is not defined non-POSIX platforms.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@62276 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
* thread.c (timeval_for): tv_usec is suseconds_t which may be
smaller than long.
* thread_pthread.c (native_cond_timeout): ret is now used in
CLOCK_MONOTONIC case only.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@62275 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
This results in fewer conversion on common modern systems with
support for clock_gettime, pthread_cond_timedwait and ppoll.
gettimeofday is declared obsolete by POSIX.1-2008, so it is yet
another reason to move away from it. This also appears to result
in the reduction of compatibility code required for dealing
with inconsistent implementations of "struct timeval".tv_sec
In the future, this will also result in fewer conversions for
kqueue and pselect if we elect to use them.
[ruby-core:85416] [Feature #14452]
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@62272 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
which has been developed by Takashi Kokubun <takashikkbn@gmail> as
YARV-MJIT. Many of its bugs are fixed by wanabe <s.wanabe@gmail.com>.
This JIT compiler is designed to be a safe migration path to introduce
JIT compiler to MRI. So this commit does not include any bytecode
changes or dynamic instruction modifications, which are done in original
MJIT.
This commit even strips off some aggressive optimizations from
YARV-MJIT, and thus it's slower than YARV-MJIT too. But it's still
fairly faster than Ruby 2.5 in some benchmarks (attached below).
Note that this JIT compiler passes `make test`, `make test-all`, `make
test-spec` without JIT, and even with JIT. Not only it's perfectly safe
with JIT disabled because it does not replace VM instructions unlike
MJIT, but also with JIT enabled it stably runs Ruby applications
including Rails applications.
I'm expecting this version as just "initial" JIT compiler. I have many
optimization ideas which are skipped for initial merging, and you may
easily replace this JIT compiler with a faster one by just replacing
mjit_compile.c. `mjit_compile` interface is designed for the purpose.
common.mk: update dependencies for mjit_compile.c.
internal.h: declare `rb_vm_insn_addr2insn` for MJIT.
vm.c: exclude some definitions if `-DMJIT_HEADER` is provided to
compiler. This avoids to include some functions which take a long time
to compile, e.g. vm_exec_core. Some of the purpose is achieved in
transform_mjit_header.rb (see `IGNORED_FUNCTIONS`) but others are
manually resolved for now. Load mjit_helper.h for MJIT header.
mjit_helper.h: New. This is a file used only by JIT-ed code. I'll
refactor `mjit_call_cfunc` later.
vm_eval.c: add some #ifdef switches to skip compiling some functions
like Init_vm_eval.
win32/mkexports.rb: export thread/ec functions, which are used by MJIT.
include/ruby/defines.h: add MJIT_FUNC_EXPORTED macro alis to clarify
that a function is exported only for MJIT.
array.c: export a function used by MJIT.
bignum.c: ditto.
class.c: ditto.
compile.c: ditto.
error.c: ditto.
gc.c: ditto.
hash.c: ditto.
iseq.c: ditto.
numeric.c: ditto.
object.c: ditto.
proc.c: ditto.
re.c: ditto.
st.c: ditto.
string.c: ditto.
thread.c: ditto.
variable.c: ditto.
vm_backtrace.c: ditto.
vm_insnhelper.c: ditto.
vm_method.c: ditto.
I would like to improve maintainability of function exports, but I
believe this way is acceptable as initial merging if we clarify the
new exports are for MJIT (so that we can use them as TODO list to fix)
and add unit tests to detect unresolved symbols.
I'll add unit tests of JIT compilations in succeeding commits.
Author: Takashi Kokubun <takashikkbn@gmail.com>
Contributor: wanabe <s.wanabe@gmail.com>
Part of [Feature #14235]
---
* Known issues
* Code generated by gcc is faster than clang. The benchmark may be worse
in macOS. Following benchmark result is provided by gcc w/ Linux.
* Performance is decreased when Google Chrome is running
* JIT can work on MinGW, but it doesn't improve performance at least
in short running benchmark.
* Currently it doesn't perform well with Rails. We'll try to fix this
before release.
---
* Benchmark reslts
Benchmarked with:
Intel 4.0GHz i7-4790K with 16GB memory under x86-64 Ubuntu 8 Cores
- 2.0.0-p0: Ruby 2.0.0-p0
- r62186: Ruby trunk (early 2.6.0), before MJIT changes
- JIT off: On this commit, but without `--jit` option
- JIT on: On this commit, and with `--jit` option
** Optcarrot fps
Benchmark: https://github.com/mame/optcarrot
| |2.0.0-p0 |r62186 |JIT off |JIT on |
|:--------|:--------|:--------|:--------|:--------|
|fps |37.32 |51.46 |51.31 |58.88 |
|vs 2.0.0 |1.00x |1.38x |1.37x |1.58x |
** MJIT benchmarks
Benchmark: https://github.com/benchmark-driver/mjit-benchmarks
(Original: https://github.com/vnmakarov/ruby/tree/rtl_mjit_branch/MJIT-benchmarks)
| |2.0.0-p0 |r62186 |JIT off |JIT on |
|:----------|:--------|:--------|:--------|:--------|
|aread |1.00 |1.09 |1.07 |2.19 |
|aref |1.00 |1.13 |1.11 |2.22 |
|aset |1.00 |1.50 |1.45 |2.64 |
|awrite |1.00 |1.17 |1.13 |2.20 |
|call |1.00 |1.29 |1.26 |2.02 |
|const2 |1.00 |1.10 |1.10 |2.19 |
|const |1.00 |1.11 |1.10 |2.19 |
|fannk |1.00 |1.04 |1.02 |1.00 |
|fib |1.00 |1.32 |1.31 |1.84 |
|ivread |1.00 |1.13 |1.12 |2.43 |
|ivwrite |1.00 |1.23 |1.21 |2.40 |
|mandelbrot |1.00 |1.13 |1.16 |1.28 |
|meteor |1.00 |2.97 |2.92 |3.17 |
|nbody |1.00 |1.17 |1.15 |1.49 |
|nest-ntimes|1.00 |1.22 |1.20 |1.39 |
|nest-while |1.00 |1.10 |1.10 |1.37 |
|norm |1.00 |1.18 |1.16 |1.24 |
|nsvb |1.00 |1.16 |1.16 |1.17 |
|red-black |1.00 |1.02 |0.99 |1.12 |
|sieve |1.00 |1.30 |1.28 |1.62 |
|trees |1.00 |1.14 |1.13 |1.19 |
|while |1.00 |1.12 |1.11 |2.41 |
** Discourse's script/bench.rb
Benchmark: https://github.com/discourse/discourse/blob/v1.8.7/script/bench.rb
NOTE: Rails performance was somehow a little degraded with JIT for now.
We should fix this.
(At least I know opt_aref is performing badly in JIT and I have an idea
to fix it. Please wait for the fix.)
*** JIT off
Your Results: (note for timings- percentile is first, duration is second in millisecs)
categories_admin:
50: 17
75: 18
90: 22
99: 29
home_admin:
50: 21
75: 21
90: 27
99: 40
topic_admin:
50: 17
75: 18
90: 22
99: 32
categories:
50: 35
75: 41
90: 43
99: 77
home:
50: 39
75: 46
90: 49
99: 95
topic:
50: 46
75: 52
90: 56
99: 101
*** JIT on
Your Results: (note for timings- percentile is first, duration is second in millisecs)
categories_admin:
50: 19
75: 21
90: 25
99: 33
home_admin:
50: 24
75: 26
90: 30
99: 35
topic_admin:
50: 19
75: 20
90: 25
99: 30
categories:
50: 40
75: 44
90: 48
99: 76
home:
50: 42
75: 48
90: 51
99: 89
topic:
50: 49
75: 55
90: 58
99: 99
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@62197 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
that allows to JIT-compile Ruby methods by generating C code and
using C compiler. See the first comment of mjit.c to know what this
file does.
mjit.c is authored by Vladimir Makarov <vmakarov@redhat.com>.
After he invented great method JIT infrastructure for MRI as MJIT,
Lars Kanis <lars@greiz-reinsdorf.de> sent the patch to support MinGW
in MJIT. In addition to merging it, I ported pthread to Windows native
threads. Now this MJIT infrastructure can be compiled on Visual Studio.
This commit simplifies mjit.c to decrease code at initial merge. For
example, this commit does not provide multiple JIT threads support.
We can resurrect them later if we really want them, but I wanted to minimize
diff to make it easier to review this patch.
`/tmp/_mjitXXX` file is renamed to `/tmp/_ruby_mjitXXX` because non-Ruby
developers may not know the name "mjit" and the file name should make
sure it's from Ruby and not from some harmful programs. TODO: it may be
better to store this to some temporary directory which Ruby is already using
by Tempfile, if it's not bad for performance.
mjit.h: New. It has `mjit_exec` interface similar to `vm_exec`, which is
for triggering MJIT. This drops interface for AOT compared to the original
MJIT.
Makefile.in: define macros to let MJIT know the path of MJIT header.
Probably we can refactor this to reduce the number of macros (TODO).
win32/Makefile.sub: ditto.
common.mk: compile mjit.o and mjit_compile.o. Unlike original MJIT, this
commit separates MJIT infrastructure and JIT compiler code as independent
object files. As initial patch is NOT going to have ultra-fast JIT compiler,
it's likely to replace JIT compiler, e.g. original MJIT's compiler or some
future JIT impelementations which are not public now.
inits.c: define MJIT module. This is added because `MJIT.enabled?` was
necessary for testing.
test/lib/zombie_hunter.rb: skip if `MJIT.enabled?`. Obviously this
wouldn't work with current code when JIT is enabled.
test/ruby/test_io.rb: skip this too. This would make no sense with MJIT.
ruby.c: define MJIT CLI options. As major difference from original MJIT,
"-j:l"/"--jit:llvm" are renamed to "--jit-cc" because I want to support
not only gcc/clang but also cl.exe (Visual Studio) in the future. But it
takes only "--jit-cc=gcc", "--jit-cc=clang" for now. And only long "--jit"
options are allowed since some Ruby committers preferred it at Ruby
developers Meeting on January, and some of options are renamed.
This file also triggers to initialize MJIT thread and variables.
eval.c: finalize MJIT worker thread and variables.
test/ruby/test_rubyoptions.rb: fix number of CLI options for --jit.
thread_pthread.c: change for pthread abstraction in MJIT. Prefix rb_ for
functions which are used by other files.
thread_win32.c: ditto, for Windows. Those pthread porting is one of major
works that YARV-MJIT created, which is my fork of MJIT, in Feature 14235.
thread.c: follow rb_ prefix changes
vm.c: trigger MJIT call on VM invocation. Also trigger `mjit_mark` to avoid
SEGV by race between JIT and GC of ISeq. The improvement was provided by
wanabe <s.wanabe@gmail.com>.
In JIT compiler I created and am going to add in my next commit, I found
that having `mjit_exec` after `vm_loop_start:` is harmful because the
JIT-ed function doesn't proceed other ISeqs on RESTORE_REGS of leave insn.
Executing non-FINISH frame is unexpected for my JIT compiler and
`exception_handler` triggers executions of such ISeqs. So `mjit_exec`
here should be executed only when it directly comes from `vm_exec` call.
`RubyVM::MJIT` module and `.enabled?` method is added so that we can skip
some tests which don't expect JIT threads or compiler file descriptors.
vm_insnhelper.h: trigger MJIT on method calls during VM execution.
vm_core.h: add fields required for mjit.c. `bp` must be `cfp[6]` because
rb_control_frame_struct is likely to be casted to another struct. The
last position is the safest place to add the new field.
vm_insnhelper.c: save initial value of cfp->ep as cfp->bp. This is an
optimization which are done in both MJIT and YARV-MJIT. So this change
is added in this commit. Calculating bp from ep is a little heavy work,
so bp is kind of cache for it.
iseq.c: notify ISeq GC to MJIT. We should know which iseq in MJIT queue
is GCed to avoid SEGV. TODO: unload some GCed units in some safe way.
gc.c: add hooks so that MJIT can wait GC, and vice versa. Simultaneous
JIT and GC executions may cause SEGV and so we should synchronize them.
cont.c: save continuation information in MJIT worker. As MJIT shouldn't
unload JIT-ed code which is being used, MJIT wants to know full list of
saved execution contexts for continuation and detect ISeqs in use.
mjit_compile.c: added empty JIT compiler so that you can reuse this commit
to build your own JIT compiler. This commit tries to compile ISeqs but
all of them are considered as not supported in this commit. So you can't
use JIT compiler in this commit yet while we added --jit option now.
Patch author: Vladimir Makarov <vmakarov@redhat.com>.
Contributors:
Takashi Kokubun <takashikkbn@gmail.com>.
wanabe <s.wanabe@gmail.com>.
Lars Kanis <lars@greiz-reinsdorf.de>.
Part of Feature 12589 and 14235.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@62189 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
FP arithmetic can lose precision in some cases leading to
premature wakeup and wasting CPU cycles.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@62183 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
FP arithmetic can lose precision in some cases leading to
premature wakeup and wasting CPU cycles.
Convert to use timeval_* functions for now.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@62182 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
