The vm1_ prefix and vm2_ had had special meaning until
820ad9cb1d and
12068aa4e9. AFAIK there's no special
meaning in vm3_ prefix.
As they have confused people (like "In `benchmark` what is difference
between `vm1_`, `vm2_` and `vm3_`"), I'd like to remove the obsoleted
prefix as we obviated that two years ago.
for VM_METHOD_TYPE_CFUNC.
This has been known to decrease optcarrot fps:
```
$ benchmark-driver -v --rbenv 'before --jit;after --jit' benchmark.yml --repeat-count=24 --output=all
before --jit: ruby 2.8.0dev (2020-04-13T16:25:13Z master fb40495cd9) +JIT [x86_64-linux]
after --jit: ruby 2.8.0dev (2020-04-13T23:23:11Z mjit-inline-c bdcd06d159) +JIT [x86_64-linux]
Calculating -------------------------------------
before --jit after --jit
Optcarrot Lan_Master.nes 66.38132676191719 67.41369177299630 fps
69.42728743772243 68.90327567263054
72.16028300263211 69.62605130880686
72.46631319102777 70.48818243767207
73.37078877002490 70.79522887347566
73.69422431217367 70.99021920193194
74.01471487018695 74.69931965402584
75.48685183295630 74.86714575949016
75.54445264507932 75.97864419721677
77.28089738169756 76.48908637569581
78.04183397891302 76.54320932488021
78.36807984096562 76.59407262898067
78.92898762543574 77.31316743361343
78.93576483233765 77.97153484180480
79.13754917503078 77.98478782102325
79.62648945850653 78.02263322726446
79.86334213878064 78.26333724045934
80.05100635898518 78.60056756355614
80.26186843769584 78.91082645644468
80.34205717020330 79.01226659142263
80.62286066044338 79.32733939423721
80.95883033058557 79.63793060542024
80.97376819251613 79.73108936622778
81.23050939202896 80.18280109433088
```
and I deleted this capability in an early stage of YARV-MJIT development:
0ab130feee
I suspect either of the following things could be the cause:
* Directly calling vm_call_cfunc requires more optimization effort in GCC,
resulting in 30ms-ish compilation time increase for such methods and
decreasing the number of methods compiled in a benchmarked period.
* Code size increase => icache miss hit
These hypotheses could be verified by some methodologies. However, I'd
like to introduce this regardless of the result because this blocks
inlining C method's definition.
I may revert this commit when I give up to implement inlining C method
definition, which requires this change.
Microbenchmark-wise, this gives slight performance improvement:
```
$ benchmark-driver -v --rbenv 'before --jit;after --jit' benchmark/mjit_send_cfunc.yml --repeat-count=4
before --jit: ruby 2.8.0dev (2020-04-13T16:25:13Z master fb40495cd9) +JIT [x86_64-linux]
after --jit: ruby 2.8.0dev (2020-04-13T23:23:11Z mjit-inline-c bdcd06d159) +JIT [x86_64-linux]
Calculating -------------------------------------
before --jit after --jit
mjit_send_cfunc 41.961M 56.489M i/s - 100.000M times in 2.383143s 1.770244s
Comparison:
mjit_send_cfunc
after --jit: 56489372.5 i/s
before --jit: 41961388.1 i/s - 1.35x slower
```
Previously, passing a keyword splat to a method always allocated
a hash on the caller side, and accepting arbitrary keywords in
a method allocated a separate hash on the callee side. Passing
explicit keywords to a method that accepted a keyword splat
did not allocate a hash on the caller side, but resulted in two
hashes allocated on the callee side.
This commit makes passing a single keyword splat to a method not
allocate a hash on the caller side. Passing multiple keyword
splats or a mix of explicit keywords and a keyword splat still
generates a hash on the caller side. On the callee side,
if arbitrary keywords are not accepted, it does not allocate a
hash. If arbitrary keywords are accepted, it will allocate a
hash, but this commit uses a callinfo flag to indicate whether
the caller already allocated a hash, and if so, the callee can
use the passed hash without duplicating it. So this commit
should make it so that a maximum of a single hash is allocated
during method calls.
To set the callinfo flag appropriately, method call argument
compilation checks if only a single keyword splat is given.
If only one keyword splat is given, the VM_CALL_KW_SPLAT_MUT
callinfo flag is not set, since in that case the keyword
splat is passed directly and not mutable. If more than one
splat is used, a new hash needs to be generated on the caller
side, and in that case the callinfo flag is set, indicating
the keyword splat is mutable by the callee.
In compile_hash, used for both hash and keyword argument
compilation, if compiling keyword arguments and only a
single keyword splat is used, pass the argument directly.
On the caller side, in vm_args.c, the callinfo flag needs to
be recognized and handled. Because the keyword splat
argument may not be a hash, it needs to be converted to a
hash first if not. Then, unless the callinfo flag is set,
the hash needs to be duplicated. The temporary copy of the
callinfo flag, kw_flag, is updated if a hash was duplicated,
to prevent the need to duplicate it again. If we are
converting to a hash or duplicating a hash, we need to update
the argument array, which can including duplicating the
positional splat array if one was passed. CALLER_SETUP_ARG
and a couple other places needs to be modified to handle
similar issues for other types of calls.
This includes fairly comprehensive tests for different ways
keywords are handled internally, checking that you get equal
results but that keyword splats on the caller side result in
distinct objects for keyword rest parameters.
Included are benchmarks for keyword argument calls.
Brief results when compiled without optimization:
def kw(a: 1) a end
def kws(**kw) kw end
h = {a: 1}
kw(a: 1) # about same
kw(**h) # 2.37x faster
kws(a: 1) # 1.30x faster
kws(**h) # 2.19x faster
kw(a: 1, **h) # 1.03x slower
kw(**h, **h) # about same
kws(a: 1, **h) # 1.16x faster
kws(**h, **h) # 1.14x faster
Instead of searching twice to extract and to delete, extract and
delete the found position at the first search.
This makes faster nearly twice, for regexps and strings.
| |compare-ruby|built-ruby|
|:-------------|-----------:|---------:|
|regexp-short | 2.143M| 3.918M|
|regexp-long | 105.162k| 205.410k|
|string-short | 3.789M| 7.964M|
|string-long | 1.301M| 2.457M|
I noticed that some files in rubygems were executable, and I could think
of no reason why they should be.
In general, I think ruby files should never have the executable bit set
unless they include a shebang, so I run the following command over the
whole repo:
```bash
find . -name '*.rb' -type f -executable -exec bash -c 'grep -L "^#!" $1 || chmod -x $1' _ {} \;
```
* Stop making a redundant hash copy in Hash#dup
It was making a copy of the hash without rehashing, then created an
extra copy of the hash to do the rehashing. Since rehashing creates
a new copy already, this change just uses that rehashing to make
the copy.
[Bug #16121]
* Remove redundant Check_Type after to_hash
* Fix freeing and clearing destination hash in Hash#initialize_copy
The code was assuming the state of the destination hash based on the
source hash for clearing any existing table on it. If these don't match,
then that can cause the old table to be leaked. This can be seen by
compiling hash.c with `#define HASH_DEBUG 1` and running the following
script, which will crash from a debug assertion.
```ruby
h = 9.times.map { |i| [i, i] }.to_h
h.send(:initialize_copy, {})
```
* Remove dead code paths in rb_hash_initialize_copy
Given that `RHASH_ST_TABLE_P(h)` is defined as `(!RHASH_AR_TABLE_P(h))`
it shouldn't be possible for a hash to be neither of these, so there
is no need for the removed `else if` blocks.
* Share implementation between Hash#replace and Hash#initialize_copy
This also fixes key rehashing for small hashes backed by an array
table for Hash#replace. This used to be done consistently in ruby
2.5.x, but stopped being done for small arrays in ruby 2.6.x.
This also bring optimization improvements that were done for
Hash#initialize_copy to Hash#replace.
* Add the Hash#dup benchmark
I noticed that in case of cache misshit, re-calculated cc->me can
be the same method entry than the pevious one. That is an okay
situation but can't we partially reuse the cache, because cc->call
should still be valid then?
One thing that has to be special-cased is when the method entry
gets amended by some refinements. That happens behind-the-scene
of call cache mechanism. We have to check if cc->me->def points to
the previously saved one.
Calculating -------------------------------------
trunk ours
vm2_poly_same_method 1.534M 2.025M i/s - 6.000M times in 3.910203s 2.962752s
Comparison:
vm2_poly_same_method
ours: 2025143.9 i/s
trunk: 1534447.2 i/s - 1.32x slower
This approach is simpler than the previous approach which tries to
emulate realpath(3). It also performs much better on both Linux and
OpenBSD on the included benchmarks.
By using realpath(3), we can better integrate with system security
features such as OpenBSD's unveil(2) system call.
This does not use realpath(3) on Windows even if it exists, as the
approach for checking for absolute paths does not work for drive
letters. This can be fixed without too much difficultly, though until
Windows defines realpath(3), there is no need to do so.
For File.realdirpath, where the last element of the path is not
required to exist, fallback to the previous approach, as realpath(3)
on most operating systems requires the whole path be valid (per POSIX),
and the operating systems where this isn't true either plan to conform
to POSIX or may change to conform to POSIX in the future.
glibc realpath(3) does not handle /path/to/file.rb/../other_file.rb
paths, returning ENOTDIR in that case. Fallback to the previous code
if realpath(3) returns ENOTDIR.
glibc doesn't like realpath(3) usage for paths like /dev/fd/5,
returning ENOENT even though the path may appear to exist in the
filesystem. If ENOENT is returned and the path exists, then fall
back to the default approach.
and switch-case branches.
Buffer allocation optimization using `ALLOCA_N` would be the main
benefit of patch. It eliminates the O(N) buffer extensions.
It also reduces the number of branches using escape table like
https://mattn.kaoriya.net/software/lang/c/20160817011915.htm.
Closes: https://github.com/ruby/ruby/pull/2226
Co-authored-by: Nobuyoshi Nakada <nobu@ruby-lang.org>
Co-authored-by: Yasuhiro MATSUMOTO <mattn.jp@gmail.com>
and switch-case branches.
Buffer allocation optimization using `ALLOCA_N` would be the main
benefit of patch. It eliminates the O(N) buffer extensions.
It also reduces the number of branches using escape table like
https://mattn.kaoriya.net/software/lang/c/20160817011915.htm.
Closes: https://github.com/ruby/ruby/pull/2226
Co-authored-by: Nobuyoshi Nakada <nobu@ruby-lang.org>
Co-authored-by: Yasuhiro MATSUMOTO <mattn.jp@gmail.com>
I heard actually this part would not be a bottleneck for rendering
because writing anything to terminal takes way longer time anyway, but I
thought this benchmark script might be useful for benchmarking Ruby
itself.
卜部昌平
Nobuyoshi Nakada
S.H
Takashi Kokubun
Jeremy Evans
Kazuhiro NISHIYAMA
Lourens Naudé
David Rodríguez
Dylan Thacker-Smith
Masato Ohba
Yaw Boakye
Yusuke Endoh
Aaron Patterson
Samuel Williams
Hiroshi SHIBATA
nobu
eregon