rb_hash_bulk_insert is added to official C API in r63488. It's no longer
exported only for MJIT.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63506 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
Calling the .eql? and .hash methods during a Hash operation can
result in a thread switch or a signal handler to run: allowing
one execution context to rebuild the hash table while another is
still reading or writing the table. This results in a
use-after-free bug affecting the thread_safe-0.3.6 test suite
and likely other bugs.
This bug did not affect users of commonly keys (String, Symbol,
Fixnum) as those are optimized to avoid method dispatch
for .eql? and .hash methods.
A separate version of this change needs to be ported to Ruby 2.3.x
which had a different implementation of st.c but was affected
by the same bug.
* st.c: Add comment about table rebuilding during comparison.
(DO_PTR_EQUAL_CHECK): New macro.
(REBUILT_TABLE_ENTRY_IND, REBUILT_TABLE_BIN_IND): New macros.
(find_entry, find_table_entry_ind, find_table_bin_ind): Use new
macros. Return the rebuild flag.
(find_table_bin_ptr_and_reserve): Ditto.
(st_lookup, st_get_key, st_insert, st_insert2): Retry the
operation if the table was rebuilt.
(st_general_delete, st_shift, st_update, st_general_foreach):
Ditto.
(st_rehash_linear, st_rehash_indexed): Use DO_PTR_EQUAL_CHECK.
Return the rebuild flag.
(st_rehash): Retry the operation if the table was rebuilt.
[ruby-core:85510] [Ruby trunk Bug#14357]
Thanks to Vit Ondruch for reporting the bug.
From: Vladimir Makarov <vmakarov@redhat.com>
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@62396 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
These casts are guarded. Must be safe to assume alignments.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@61829 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
This reverts commit r61309
Because it was unstable on mswin CI.
[ruby-dev:50370][Bug #14203]
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@61341 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
* st.c (_st_table_list, _st_table_pool): symbols beginning with an
underscore and a lower letter are preserved by the standard.
* st.c (get_st_table): protoized.
* st.c (st_insert_generic): adjust local variable type to an
argument argc.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@61338 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
When generate Hash object, the heap area of st_table will be always allocated in internally
and seems it take a time.
To improve performance of creating Hash object,
this patch will reduce count of allocating heap areas for st_table by reuse them.
Performance of creating Hash literal -> 1.53 times faster.
[Fix GH-1766] [ruby-core:84008] [Feature #14146]
### Environment
* OS : macOS 10.13.1
* CPU : 1.4 GHz Intel Core i7
* Compiler : Apple LLVM version 9.0.0 (clang-900.0.39)
### Before
$ ./miniruby -v -I. -I../benchmark-ips/lib ~/tmp/bench/literal.rb
ruby 2.5.0dev (2017-11-28 hash 60926) [x86_64-darwin17]
Warming up --------------------------------------
Hash literal 51.544k i/100ms
Calculating -------------------------------------
Hash literal 869.132k (± 1.1%) i/s - 4.381M in 5.041574s
### After
$ ./miniruby -v -I. -I../benchmark-ips/lib ~/tmp/bench/literal.rb
ruby 2.5.0dev (2017-11-28 hash 60926) [x86_64-darwin17]
Warming up --------------------------------------
Hash literal 63.068k i/100ms
Calculating -------------------------------------
Hash literal 1.328M (± 2.3%) i/s - 6.685M in 5.037861s
### Test code
require 'benchmark/ips'
Benchmark.ips do |x|
x.report "Hash literal" do |loop|
count = 0
while count < loop
hash = {foo: 12, bar: 34, baz: 56}
count += 1
end
end
end
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@61309 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
* st.c (st_insert2): should manage num_entries when the key is
undefined, as well as st_insert().
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@59748 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
* st.c: split assertion conditions.
* st.c (rb_hash_bulk_insert): the number of arguments should be
even.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@59747 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
Specialized routine for small linear-probling hash instances to
boost creation of such things [Bug #13861]
Signed-off-by: Urabe, Shyouhei <shyouhei@ruby-lang.org>
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@59746 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
These caused numerous CI failures I haven't been able to
reproduce [ruby-core:82102]
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@59364 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
This gives the newhash VM instruction the same string reuse
capabilities as rb_hash_aset.
* st.c (str_key): new wrapper function to call rb_fstring_existing
(rb_hash_bulk_insert): use str_key
* test/ruby/test_optimization.rb (test_hash_reuse_fstring):
ensure key reuse for newhash instructions
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@59355 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
* st.c (rb_hash_bulk_insert): new API to bulk insert entries
into a hash. Given arguments are first inserted into the
table at once, then reindexed. This is faster than inserting
things using rb_hash_aset() one by one.
This arrangement (rb_ prefixed function placed in st.c) is
unavoidable because it both touches table internal and write
barrier at once.
* internal.h: delcare the new function.
* hash.c (rb_hash_s_create): use the new function.
* vm.c (core_hash_merge): ditto.
* insns.def (newhash): ditto.
* test/ruby/test_hash.rb: more coverage on hash creation.
* test/ruby/test_literal.rb: ditto.
-----------------------------------------------------------
benchmark results:
minimum results in each 7 measurements.
Execution time (sec)
name before after
loop_whileloop2 0.136 0.137
vm2_bighash* 1.249 0.623
Speedup ratio: compare with the result of `before' (greater is better)
name after
loop_whileloop2 0.996
vm2_bighash* 2.004
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@58492 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
Previous implementation had an issues:
- macros murmur1 assumes murmur_step takes rotation value
as a second argument
- but murmur_step second argument is "next block"
- this makes st_hash_uint and st_hash_end to not mix high bits of
hash value into lower bits
- this leads to pure hash behavior on doubles and mixing hashes using
st_hash_uint.
It didn't matter when bins amount were prime numbers, but it hurts
when bins are powers of two.
Mistake were created cause of attempt to co-exist Murmur1 and Murmur2
in a same code.
Change it to single hash-function implementation.
- block function is in a spirit of Murmur functions,
but handles inter-block dependency a bit better (imho).
- final block is read in bit more optimal way on CPU with unaligned word access,
- final block is mixed in simple way,
- finalizer is taken from MurmurHash3 (it makes most of magic :) )
(64bit finalizer is taken from
http://zimbry.blogspot.ru/2011/09/better-bit-mixing-improving-on.html)
Also remove ST_USE_FNV1: it lacks implementation of many functions,
and looks to be abandoned
Author: Sokolov Yura aka funny_falcon <funny.falcon@gmail.com>
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@57134 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
* st.h (struct st_hash_type): Remove strong_hash.
(struct st_table): Remove inside_rebuild_p and curr_hash.
* st.c (do_hash): Use type->hash instead of curr_hash.
(make_tab_empty): Remove setting up curr_hash.
(st_init_table_with_size): Remove setting up inside_rebuild_p.
(rebuild_table): Remove clearing inside_rebuild_p.
(reset_entry_hashes, HIT_THRESHOULD_FOR_STRONG_HASH): Remove code
recognizing a denial attack and switching to strong hash.
* hash.c (rb_dbl_long_hash, rb_objid_hash, rb_ident_hash): Use
rb_hash_start to randomize the hash.
(str_seed): Remove.
(any_hash): Remove strong_p and use always rb_str_hash for
strings.
(any_hash_weak, rb_any_hash_weak): Remove.
(st_hash_type objhash): Remove rb_any_hash_weak.
based on the patch by Vladimir N Makarov <vmakarov@redhat.com> at
[ruby-core:78490]. [Bug #13002]
* test/ruby/test_hash.rb (test_wrapper): objects other than special
constants should be able to be wrapped.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@56992 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
From: Vladimir Makarov <vmakarov@redhat.com>
By Vladimir's estimation, this requires at least 64 GB of memory
to reproduce this bug due to the hash sizes required. So there
is no new test case (and I am unable to test it, myself).
* st.c (get_bins_num): avoid out-of-bounds on shift by using correct type
[ruby-core:78139] [Bug #12939]
* st.c (get_allocated_entries): ditto
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@56793 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
[Feature #12142]
See header of st.c for improvment details.
You can see all of code history here:
<https://github.com/vnmakarov/ruby/tree/hash_tables_with_open_addressing>
This improvement is discussed at
<https://bugs.ruby-lang.org/issues/12142>
with many people, especially with Yura Sokolov.
* st.c: improve st_table.
* include/ruby/st.h: ditto.
* internal.h, numeric.c, hash.c (rb_dbl_long_hash): extract a function.
* ext/-test-/st/foreach/foreach.c: catch up this change.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@56650 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
and define CONSTFUNC and PUREFUNC if available.
Note that I don't add those options as default because
it still shows many false-positive (it seems not to consider
longjmp).
* vm_eval.c (stack_check): get rb_thread_t* as an argument
to avoid duplicate call of GET_THREAD().
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@54952 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
* st.c (EQUAL, st_delete_safe): fix arguments order to compare
function, searching key is the first and stored key is the
second always.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@51364 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
variable used). I think that these are wrong, but should shut them
up.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@51105 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
This improves the bm_vm2_bighash benchmark significantly by
removing branches during insert, but slows down anything
requiring iteration with the more complex loop termination
checking.
Speedup ratio of 1.10 - 1.20 is typical for the vm2_bighash
benchmark.
v3 - st_head calculates list_head address in two steps
to avoid a bug in old gcc 4.4 (Debian 4.4.7-2)
bug which incorrectly warned with:
warning: dereferencing pointer ‘({anonymous})’ does break
strict-aliasing rules
* include/ruby/st.h (struct st_table): hide struct list_head
* st.c (struct st_table_entry): adjust struct
(head, tail): remove shortcut macros
(st_head): new wrapper function
(st_init_table_with_size): adjust to new struct and API
(st_clear): ditto
(add_direct): ditto
(unpack_entries): ditto
(rehash): ditto
(st_copy): ditto
(remove_entry): ditto
(st_shift): ditto
(st_foreach_check): ditto
(st_foreach): ditto
(get_keys): ditto
(get_values): ditto
(st_values_check): ditto
(st_reverse_foreach_check): ditto (unused)
(st_reverse_foreach): ditto (unused)
[ruby-core:69726] [Misc #10278]
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@51064 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
This reverts commit r51044
Still getting failure notices from ko1's CI machine.
ref: g3qkqn
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@51045 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
I suspect the build failures with r51034 ("st.c: use ccan linked-list")
on ko1's CI machine was due to me forgetting to update common.mk :x
Lets see what happens when I only include ccan/list/list.h
Will followup with appropriate reverts or reinstating the
rest of r51034 along with a common.mk update as I watch CI
build.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@51041 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
Reduces object size slightly on x86-64:
text data bss dec hex filename
2782359 22400 71880 2876639 2be4df ruby.orig
2781831 22400 71880 2876111 2be2cf ruby.pow2
And on 32-bit x86:
text data bss dec hex filename
2814751 12100 30552 2857403 2b99bb ruby.orig
2814051 12100 30552 2856703 2b96ff ruby.pow2
This is not a performance-critical function, but the
smaller icache footprint seems worth it.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@47767 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
* st.c (st_update): old_key is uninitialized by jump to the label
unpacked.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@46725 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
* st.c (st_update): remove equality checks, callers should ensure
the equality, otherwise the behavior is undefined.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@46723 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
k0kubun
normal
shyouhei
watson1978
nobu
naruse
ko1
usa
akr
nari