V8 Crankshaft outperforms DartVM
Comparing this:
http://4.bp.blogspot.com/-umhKFVMan2s/T0wUSwJ0-eI/AAAAAAAAAYA/wAVWOlteJSw/s1600/graph1.png
with Dart's Performance:
https://www.dartlang.org/performance
It seems that V8 CrankShaft outperforms the DartVM, and that's CrankShaft, apparently TurboFan is supposed to outperform CrankShaft too... What do you guys think?
20 comments
[ 4.9 ms ] story [ 74.2 ms ] threadDart and JavaScript are both completely dynamically typed. The Dart VM and V8 both use inline caches to remember what types are flowing to what call sites. Both optimize monomorphic call sites the best and megamorphic call sites the worst. I know of no "blacklist" in the Dart VM.
V8 VM optimized for that (type dynamics).
Dart VM has a lot problem with performance with this dynamics because Dart VM is a truly object-oriented VM and each value has its own type (including a user defined types and a lot types in the Dart SDK, also internal witt hidden implementation, and all their members).
About "blacklists" and "whitelists".
===================================
Method recognizer and [while/black]list
VM: Improve performance of method recognizer and unify the it with the intrinsifier.
https://codereview.chromium.org/468793004/
===================================
https://github.com/dart-lang/bleeding_edge/blob/master/dart/...
INLINE_WHITE_LIST(SET_IS_ALWAYS_INLINE);
INLINE_BLACK_LIST(SET_IS_NEVER_INLINE);
SET_IS_ALWAYS_INLINE(class_name, function_name, dest, fp)
SET_IS_NEVER_INLINE(class_name, function_name, dest, fp)
===================================
https://github.com/dart-lang/bleeding_edge/blob/master/dart/...
===================================
A lot of (problematic) method in these list. Dart VM code generators and optimizers is unable to optimize usage of them by the Dart VM intelligence.
Hundreds of methods in these lists.
OTHER_RECOGNIZED_LIST(V) over 100 methods
CORE_LIB_INTRINSIC_LIST(V) over 50 methods
CORE_INTEGER_LIB_INTRINSIC_LIST(V) over 30 methods
MATH_LIB_INTRINSIC_LIST(V) 2 methods
TYPED_DATA_LIB_INTRINSIC_LIST(V) over 30 methods
GRAPH_TYPED_DATA_INTRINSICS_LIST(V) 9 methods
GRAPH_CORE_INTRINSICS_LIST(V) 10 methods
INLINE_WHITE_LIST(V) over 60 methods
INLINE_BLACK_LIST(V) 8 methods
POLYMORPHIC_TARGET_LIST(V) about 20 methods
Most of them cannot be optimized due the wrong architecture (which states that the type system is unsound).
Most of them cannot be optimized due the other problems in architecture.
P.S.
All mentions about them are hardcoded (added in special lists by humans) because VM has a problems with performance because it cannot optimize code in some cases due the fact that the normal running mode in Dart VM is an (unchecked) production mode.
Normal mode when (types system unsound) means that Dart VM cannot to guarantee that it has possibility correctly recognize a members (of internally implemented types) that are critical for performance.
This claim is substantially false. Where are you deriving this "Dart VM has a lot problem..." from? (feel free to respond in Russian). If you actually knew how V8 is implemented you would know that all values in the V8 have something called a hidden class (internally in the sources it's called map) and V8 optimizes based on identity of those hidden classes. This is not that different from the Dart - classes are just not hidden in Dart.
> A lot of (problematic) method in these list. Dart VM code generators and optimizers is unable to optimize usage of them by the Dart VM intelligence.
You are misunderstanding/misrepresenting why these lists exist. Let's talk about them one by one.
1. INTRINSICS Most methods on the intrinsic list don't even have pure-Dart bodies (so there is nothing to optimized with "Dart VM intelligence") and they are on the list for two purposes:
- provide fast hand written assembly implementation, that is more efficient version written in C++; - tell optimizing compiler how to lower these instructions into IR operations.
Without them being on the list there is no way optimizing compiler can optimize anything - because it would only see a call to some native runtime function with unknown semantics. How is it supposed to understand that _TypedList._getFloat64 is actually a pair of a CheckArrayBound and LoadIndexed<Double>() instructions?
Some of these intrinsified methods do have Dart bodies - but we still provide a hand-written assembly implementation for them. I think these days it's limited to methods of the Bigint class. This is the same in any arbitrary width integer library (e.g. look into GMP sources) - you can write it in C++ but people still write these things in assembly because even C++ can't always produce the best code for this.
2. INLINING_WHITELIST These are the methods that that we know are usually beneficial to inline into the caller's code even if other budget constraints are preventing it. Inlining is a very hard problem in compiler construction - it's impossible to always make optimal decisions given compilation time constraints of the JIT. Even AOT compilers need hints for this, and if you take a random JIT you will discover that it most likely has a similar whitelist built in. Essentially this white list is here precisely because we know that we can optimize this methods very well once they are inlined (which is complete opposite of "cannot be optimized" that you invoke).
3. INLINING_BLACKLIST Similar to whitelist it contains functions that we know are not beneficial to inline. Limited to Bigint class actually now, which is a very special case as explained above.
4. POLYMORPHIC_TARGET_LIST this is utility list that improves the structure of the graph that polymorphic inliner builds (it tells inliner that resulting code would not benefit from merging different branches of the polymorphic call even if they share the same target). None of the methods on this list have an actual Dart body.
To summarize all these lists exist because they encode knowledge that is impossible for Dart VM to get because those properties are essentially algorithmically undecidable.
https://github.com/dart-lang/bleeding_edge/blob/master/dart/...
===============================
class _IntegerImplementation extends _Num {
}===============================
The same method in CORE_INTEGER_LIB_INTRINSIC_LIST
https://github.com/dart-lang/bleeding_edge/blob/master/dart/...
===============================
V(_IntegerImplementation, _bitAndFromInteger, \ Integer_bitAndFromInteger, 504496713)
===============================
If Dart VM can use "patch class" declarations then why this "@inline" should be considered as a bad?
Maybe your answer in that the "Dart VM ignored all kind of annotations (including type annotations) why it should use this approach".
P.S.
This is a not a direct problem of dynamism (ignore annotations even if they hard coded in source).
Problem in that the Dart VM does not want rely on annotations (dynamism is better, determine everything at runtime or from magic lists).
Annotation like that is not enough for intrinsics. You would still need to provide some way of detecting in the compilation pipeline and intrinsifier that method you are looking at is indeed `Integer_bitAndFromInteger`. In this particular case you could look at the name of the native, but that does not work with those intrinsified methods that have Dart bodies. In those cases it would have to be @intrinsic("name of the intrinsic").
Right now we can write C++ code like this:
If we start to relying on strings - this code will become less readable (or alternatively you would have to map strings into C++ enumeration which would require a list quite list one of the above).To be honest aesthetically I like annotations, so I would like to use them to replace these lists, but that does not really solve anything or improve much. Even more: there is really no fundamental difference between having an annotation and a list like that, so I don't really understand what bothers you about these lists.
> (dynamism is better, determine everything at runtime or from magic lists)
Annotation would be no less magic (and not available to the user code anyways).
This approach does not allow me to use annotations (or other way) for specifying that my own "native" methods in "native extension" are required some attention.
You use for such attention these lists.
Eg.
I have the following methods (via macro defs) in my "native extension":
=========================
=========================These methods are very fast (in C++) but in fact Dart VM executes them (native methods) very SLOOOOOOOOOOOWWWWWWWWWWWW...
See my question on stack overflow:
========================= Why native wrapped functions in Dart are such heavyweight in comparison with “DEFINE NATIVE ENTRY” functions that are very lightweight?
http://stackoverflow.com/questions/21363429/why-native-wrapp... =========================
- Dart developers does not like annotation
- Dart VM does not uses annotation
- Dart VM does not uses type annotation
- Dart VM executes custom "native (which in fact are very fast)" methods very slow
This is what is means for me that Dart VM is not a very well balanced for high performance.
P.S.
This is why "I bothers you about these lists".
Because they are in some cases the only way to improve performance.
Can't you just expose the data you are reading as an external typed data array to the Dart code? That would remove any need for those methods.
> - Dart developers does not like annotation
I like annotations!
> - Dart VM executes custom "native (which in fact are very fast)" methods very slow
This concern is very valid: it is true that transition between Dart code and native methods is too heavyweight and as a developer you have no way to fix it yourself. We had plans to fix it eventually - but they never been very high on the list of things.
Did you file a bug for the slowness of your native extension?
I work on integrating into Dart ecosysem "foreign functions interface" (ffi).
External typed data is not suitable in many cases:
- Does not provide physical storage address to use code (binary interop requires that)
- I don't need it because it heavyweight if use it with C pointers and references
- It has limit in size (specified in Dart VM on max array length)
- I need malloc and typed data does not help here in any way
- I access data (by physical address) allocated externally
>> Did you file a bug for the slowness of your native extension?
This is useless. A long time ago we already did not come to a consensus.
Truly efficient FFI has to be part of the VM, you can't implement efficient FFI from outside.
> - I access data (by physical address) allocated externally
Do you really mean "physical address" here?
External data is created precisely to efficiently access data allocated externally, because it allows direct raw access to a given region of memory (sans bounds checking).
Evil of the "Truly efficient FFI" in that the it in many cases does not exists as such.
Benefit of "NOT Truly efficient FFI" in that the it in many cases it may be considered as acceptable in many cases (and exist in reality).
>> Do you really mean "physical address" here?
See here.
https://gist.github.com/mezoni/f85b04bf19dab8333abf
>> External data is created precisely to efficiently access data allocated externally, because it allows direct raw access to a given region of memory (sans bounds checking).
Serach in code "Unsafe." and you understand what means "physical address" as a base and offset.
Eg.
https://gist.github.com/mezoni/f85b04bf19dab8333abfMagic of annotations in that the they are only annotates but a real magic goes elsewhere...
They can be ignored and this is not a magic because they are only a metadata (data about data).
I think that Dart VM and Dart developers does not like annotations because they thinking that VM cannot use them effectively (even if they written by humans).
But Dart VM are more complex than V8 VM and a ways that uses developers to tell VM about how specify metadata about critical methods is not a very good.
========================
Are there attributes that affect how the CLR optimizes during a JIT compile?
http://stackoverflow.com/questions/2367094/are-there-attribu...
========================
Also
========================
MethodImplOptions Enumeration
http://msdn.microsoft.com/ru-ru/library/system.runtime.compi...
- Unmanaged
- ForwardRef
- PreserveSig
- InternalCall
- Synchronized
- NoInlining
- AggressiveInlining
- NoOptimization
========================
This is from:
"The System.Runtime.CompilerServices namespace provides functionality for compiler writers who use managed code to specify attributes in metadata that affect the run-time behavior of the common language runtime."
The fact that Dart is not a multi-layer VM does not means that it are simple this is good for optimizations.
Do you use such a layer as a "Well structured intermediate language" in Dart VM?
The fear is actually completely opposite: people would misuse these annotations without understanding what they do and then complain that Dart VM "is slow".
As I said: I'd love to convert these lists to annotations. Don't tell anyone but we actually have inlining annotations[1] which we use when writing tests.
> Do you use such a layer as a "Well structured intermediate language" in Dart VM?
I don't understand the question. Dart VM uses intermediate representation when compiling, which is an implementation detail (unlike CLR IL - which well specified input to the VM essentially)
[1] https://code.google.com/p/dart/source/browse/branches/bleedi...)
>> I don't understand the question. Dart VM uses intermediate representation when compiling, which is an implementation detail (unlike CLR IL - which well specified input to the VM essentially)
I think that optimizing code easy if it represent in suitable form (intermediate format) for optimizations.
Dart VM does have any specifications about how it represent code and data in such form (intermediate language).
Intermediate language in this case means nothing more than the an intermediate language itself (with documentation) but not a some clases written in C++ language in VM project.
Does you have it?
Your approach very bad.
It does not help in the case where other professionals in this area can help you make VM better and effective.
"Using asm.js" is not some magic that just makes all JavaScript faster. For example all those benchmarks from http://dartlang.org/performance are written in a normal JavaScript not in what is known as "asm.js".
Now TurboFan does not even really "use asm.js". It respects "use asm" directive to decide which functions to optimize and enables a few optimizations on such functions. Other than that it's a completely generic JavaScript JIT compiler, which does not even performs a complete asm.js module type verification pass (something that OdinMonkey does).
Furthermore if you try enabling TurboFan on a non-asm.js code then you'll discover that right now TurobFan is lacking infrastructure for adaptive optimizations which is an absolute requirement for compiling normal JavaScript (as opposed to asm.js compliant code which is limted to arithmetic, typed arrays manipulation and essentially statically typed) into efficient machine code. It will run this normal JavaScript code - but it will not necessarily run it fast.
That said I am looking forward for TF with those adaptive optimizations implemented. It's a bold project with a lot of promised value for V8. Though I would not expect anything as impressive as "6x over Crankshaft across the board".