Optimizations in Namer and Typer #5875
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
https://scala-ci.typesafe.com/job/scala-2.12.x-validate-test/4897/consoleText |
|
/rebuild |
adriaanm
suggested changes
May 2, 2017
| @@ -588,9 +590,12 @@ trait Implicits { | |||
| // We can only rule out a subtype relationship if the left hand | |||
| // side is a class, else we may not know enough. | |||
| case tr1 @ TypeRef(_, sym1, _) if sym1.isClass => | |||
| def hasMember(tp: Type, name: Name) = { | |||
There was a problem hiding this comment.
Could we narrow this helper method's scope a bit? typeRefHasNoMember? I fear a future refactoring could pull this out and miss that it's only valid under certain assumptions.
| def pop(): Unit = () | ||
| def inMode(mode: Mode, tree: Tree): Tree = tree | ||
| } | ||
| class CheckingCheckDead() extends CheckDead { |
There was a problem hiding this comment.
final?
| // for use with silent type checking to when we can't have results with undetermined type params | ||
| // note that this captures the context var | ||
| val isMonoContext = (_: Any) => context.undetparams.isEmpty | ||
| private var _checkDead: CheckDead = _ |
| @@ -196,11 +197,13 @@ abstract class SymbolTable extends macros.Universe | |||
| final def pushPhase(ph: Phase): Phase = { | |||
| val current = phase | |||
| phase = ph | |||
| phStack ::= ph | |||
| phStack(phStackIndex) = ph | |||
There was a problem hiding this comment.
bounds check, just in case?
There was a problem hiding this comment.
I was happy to let an ArrayIndexOutOfBoundException occur in that case rather than adding explicit code here to handle it.
There was a problem hiding this comment.
The alternative would be to use an ArrayBuffer, which doesn't require a hard limit to the depth of recursive atPhase-s. But I was being a bit paranoid with performance.
| @@ -171,11 +171,12 @@ abstract class SymbolTable extends macros.Universe | |||
| final val NoRunId = 0 | |||
|
|
|||
| // sigh, this has to be public or enteringPhase doesn't inline. | |||
| var phStack: List[Phase] = Nil | |||
| var phStack: Array[Phase] = new Array(128) | |||
There was a problem hiding this comment.
[nitpick] Only store phase name?
There was a problem hiding this comment.
Storage wise, there isn't a difference, both are just already allocated Object refs.
If we store the phase name, we have to eagerly call the virtual Phase.name in phase_=, even though under normal circumstances we don't use the result.
In general, It would be good to avoid retaining Phase-s as fields in Global so they are eligible for GC after the Run that spawned them has been collected. But we know we'll end up with an empty phase stack here at the end of the run, so it doesn't make a practical difference.
Non local returns aren't eliminated after inlined in 2.11 or 2.12
```
⚡ scala
Welcome to Scala 2.12.1 (Java HotSpot(TM) 64-Bit Server VM, Java 1.8.0_112).
Type in expressions for evaluation. Or try :help.
scala> @inlune def foo(a: => Any) = if ("".isEmpty) a else ""
<console>:11: error: not found: type inlune
@inlune def foo(a: => Any) = if ("".isEmpty) a else ""
^
scala> @inline def foo(a: => Any) = if ("".isEmpty) a else ""
foo: (a: => Any)Any
scala> class InlineReturn { def test: Any = foo(return "") }
defined class InlineReturn
scala> :javap -c InlineReturn#test
public java.lang.Object test();
Code:
0: new #4 // class java/lang/Object
3: dup
4: invokespecial #32 // Method java/lang/Object."<init>":()V
7: astore_1
8: getstatic #36 // Field $line4/$read$$iw$$iw$.MODULE$:L$line4/$read$$iw$$iw$;
11: aload_1
12: invokedynamic #59, 0 // InvokeDynamic #0:apply:(Ljava/lang/Object;)Lscala/Function0;
17: invokevirtual #63 // Method $line4/$read$$iw$$iw$.foo:(Lscala/Function0;)Ljava/lang/Object;
20: goto 44
23: astore_2
24: aload_2
25: invokevirtual #66 // Method scala/runtime/NonLocalReturnControl.key:()Ljava/lang/Object;
28: aload_1
29: if_acmpne 39
32: aload_2
33: invokevirtual #69 // Method scala/runtime/NonLocalReturnControl.value:()Ljava/lang/Object;
36: goto 41
39: aload_2
40: athrow
41: goto 44
44: areturn
Exception table:
from to target type
8 20 23 Class scala/runtime/NonLocalReturnControl
```
```
⚡ ~/scala/2.11.8/bin/scala
Welcome to Scala 2.11.8 (Java HotSpot(TM) 64-Bit Server VM, Java 1.8.0_112).
Type in expressions for evaluation. Or try :help.
scala> @inline def foo(a: => Any) = if ("".isEmpty) a else ""
foo: (a: => Any)Any
scala> class InlineReturn { def test: Any = foo(return "") }
defined class InlineReturn
scala> :javap -c InlineReturn#test
public java.lang.Object test();
Code:
0: new #4 // class java/lang/Object
3: dup
4: invokespecial #13 // Method java/lang/Object."<init>":()V
7: astore_1
8: getstatic #19 // Field .MODULE$:L;
11: new #21 // class InlineReturn$$anonfun$test$1
14: dup
15: aload_0
16: aload_1
17: invokespecial #24 // Method InlineReturn$$anonfun$test$1."<init>":(LInlineReturn;Ljava/lang/Object;)V
20: invokevirtual #28 // Method .foo:(Lscala/Function0;)Ljava/lang/Object;
23: goto 39
26: astore_2
27: aload_2
28: invokevirtual #31 // Method scala/runtime/NonLocalReturnControl.key:()Ljava/lang/Object;
31: aload_1
32: if_acmpne 40
35: aload_2
36: invokevirtual #34 // Method scala/runtime/NonLocalReturnControl.value:()Ljava/lang/Object;
39: areturn
40: aload_2
41: athrow
Exception table:
from to target type
8 26 26 Class scala/runtime/NonLocalReturnControl
scala> :quit
```
We can avoid the call altogether in some cases.
Rather than doing this generically with a traversal in `atPos`. This method is hot enough to be as frugal as possible.
As it turns out, all our fast track macros are defined as `def foo = macro ???`, rather than `def foo = ???`. We can assume that the MACRO flag is authoritative.
When using the primary constructor paramater trees as prototypes for the parameter trees of the apply method, restrict tree duplication and reset attrs to the primary constructor, rather than the entire case class body.
We're only interested in whether or not a candidate implicit type has a member with a given name, which we can determine cheaply by checking if any ancestor has a so-named decl. We avoid both the as-seen-froms that FindMember uses to differentiate overriding and overloading, and creation of overloaded symbols.
…bility Avoids lots of calls to `isFunctionSymbol`.
If we're already in the requested phase, avoid the bookkeeping in atPhaseStack.
Keep the phase stack (which is just used for logging) in a pre-allocated array, rather than in a List.
Use an iterator to inspect annotations of owners, rather than creating a list of owners.
Contrary to the comment, it wasn't capturing the the context var at typer initialization, which would not have been correct anyway. The inlining here is motivated by avoiding allocation of lambda in every Typer. Even when we're using this value in eta expansion, SilentResult.filter will actually be inlined to avoid the need for a lambda at all.
As per review suggestions
|
I've resolved the merge conflict by dropping the commit "Avoid overhead of dead code checking unless |
adriaanm
approved these changes
May 25, 2017
retronym
added a commit
to retronym/scala
that referenced
this pull request
Apr 26, 2018
More of the same as scala#5875 The change to default getters reduced tree churn by 6x in a real world project.
retronym
added a commit
to retronym/scala
that referenced
this pull request
May 2, 2018
More of the same as scala#5875 The change to default getters reduced tree churn by 6x in a real world project.
retronym
added a commit
to retronym/scala
that referenced
this pull request
May 2, 2018
More of the same as scala#5875 The change to default getters reduced tree churn by 6x in a real world project.
retronym
added a commit
to retronym/scala
that referenced
this pull request
May 2, 2018
More of the same as scala#5875 The change to default getters reduced tree churn by 6x in a real world project.
retronym
added a commit
to retronym/scala
that referenced
this pull request
May 2, 2018
More of the same as scala#5875 The change to default getters reduced tree churn by 6x in a real world project.
retronym
added a commit
to retronym/scala
that referenced
this pull request
May 3, 2018
More of the same as scala#5875 The change to default getters reduced tree churn by 6x in a real world project.
retronym
added a commit
to retronym/scala
that referenced
this pull request
May 4, 2018
More of the same as scala#5875 The change to default getters reduced tree churn by 6x in a real world project.
retronym
added a commit
to retronym/scala
that referenced
this pull request
May 4, 2018
More of the same as scala#5875 The change to default getters reduced tree churn by 6x in a real world project.
retronym
added a commit
to retronym/scala
that referenced
this pull request
May 8, 2018
More of the same as scala#5875 The change to default getters reduced tree churn by 6x in a real world project.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
These amount to a 3% performance improvement.