Redefer function bodies - #1585

pleath · 2016-09-15T19:14:13Z

Redefer function bodies that are not currently being executed and are eligible for deferred parsing (e.g., not arrow functions, not functions-in-block). Define a 'force' mode in which all eligible functions are redeferred on GC, as well as a 'stress' mode in which all candidates are redeferred on each stack probe.

pleath · 2016-10-22T01:43:22Z

Merged with OOP JIT. That was fun. :)

pleath · 2016-10-22T19:55:22Z

Fully synched.

rajatd · 2016-11-07T21:32:06Z


    Js::FunctionEntryPointInfo *entryPointInfo = function->GetFunctionEntryPointInfo();
-    uint8 callsCount = entryPointInfo->callsCount;
+    uint32 callsCount = entryPointInfo->callsCount;


As @LouisLaf mentioned in #1757, we should have callsCount be MIN of entryPointInfo->callsCount and 255

rajatd · 2016-11-07T21:33:10Z

-        entryPointInfo->totalJittedLoopIterations = UINT8_MAX;
-    }
-    uint8 totalJittedLoopIterations = (uint8)entryPointInfo->totalJittedLoopIterations;
+    uint32 totalJittedLoopIterations = (uint8)entryPointInfo->totalJittedLoopIterations;


We should retain the previous rejit behaviour in keeping with @LouisLaf 's suggestion in #1757

pleath · 2016-11-08T01:13:15Z

Reverted to treating call counts as saturating uint8's in bailout code. (Thanks, @rajatd.)

rajatd · 2016-11-10T00:34:25Z

+    uint gcSinceLastRedeferral;
+    uint gcSinceCallCountsCollected;
+
+    static const uint InitialRedeferralDelay = 5;


InitialRedeferralDelay [](start = 22, length = 22)

should we have these as configurable values so we can tune redeferral in the future if need be?

That's a good idea.

rajatd · 2016-11-10T02:01:23Z

@@ -2062,19 +2080,6 @@ ThreadContext::ExecuteRecyclerCollectionFunction(Recycler * recycler, Collection

    BOOL ret = FALSE;



RedeferFunctionBodies [](start = 23, length = 21)

Correct me if I'm wrong, but I think we should also pass the result of this->GetRedeferralalCollectionInterval() to ScriptContext::RedeferFunctionBodies. Right now, the result of ThreadContext::GetRedeferralInactiveThreshold is being used in FunctionBody::DoRedeferFunction to determine if a function has remained inactive for long enough that it could be redeferred.

Thanks for explaining offline. No action needed for this comment.

rajatd · 2016-11-10T02:03:04Z

+            uint inactiveCount;
+            auto fn = [&](){ inactiveCount = 0xFFFFFFFF; };
+            inactiveCount = UInt32Math::Mul(this->GetInactiveCount(), this->GetCompileCount(), fn);
+            if (inactiveCount < inactiveThreshold)


inactiveCount < inactiveThreshold [](start = 16, length = 33)

The idea here is that if a function was parsed after having been redeferred, make it less likely to be deferred again, right? If so, shouldn't we multiply the threshold by the compileCount?

rajatd · 2016-11-10T02:06:38Z

-                (Attributes)(this->GetAttributes() & ~(Attributes::DeferredDeserialize | Attributes::DeferredParse)),
-                Js::FunctionBody::FunctionBodyFlags::Flags_HasNoExplicitReturnValue
-#ifdef PERF_COUNTERS
-                , false /* is function from deferred deserialized proxy */


Is this assert no longer valid? If so, can we just remove it?

rajatd · 2016-11-10T02:09:24Z

-            static_cast<uint16>(simpleJitLimit) - callCount - 1;
+            static_cast<uint16>(simpleJitLimit) - static_cast<uint16>(callCount) - 1;
    }



Is this just a bug fix?

Just silencing any possible complaint about overflow.

rajatd · 2016-11-10T02:09:55Z

+        else if (!canAllocInPreReservedHeapPageSegment && isAnyJittedCode)
+        {
+            *isAllJITCodeInPreReservedRegion = false;
+        }


not part of your change, right?

Well, it's a real issue that my change exposed, probably by causing more re-jitting.

rajatd · 2016-11-10T22:32:55Z

LGTM apart from the comments.

pleath · 2016-11-10T23:27:20Z

@dotnet-bot please test Windows x86_release

eligible for deferred parsing (e.g., not arrow functions, not functions-in-block). Define a force mode in which all eligible functions are redeferred on GC, as well as a 'stress' mode in which all candidates are redeferred on each stack probe.

Merge pull request #1585 from pleath:redefer Redefer function bodies that are not currently being executed and are eligible for deferred parsing (e.g., not arrow functions, not functions-in-block). Define a 'force' mode in which all eligible functions are redeferred on GC, as well as a 'stress' mode in which all candidates are redeferred on each stack probe.

msftclas added the cla-not-required label Sep 15, 2016

pleath force-pushed the redefer branch from b12c9f8 to f15f3c4 Compare September 28, 2016 20:30

pleath force-pushed the redefer branch from f15f3c4 to 69448f4 Compare October 22, 2016 01:42

pleath force-pushed the redefer branch from 69448f4 to 85818fb Compare October 22, 2016 19:54

This was referenced Oct 22, 2016

Revise redeferral heuristic #1786

Closed

Functional improvements and fixes for redeferral. #1736

Closed

Initial redeferral heuristic #1674

Closed

dilijev mentioned this pull request Oct 25, 2016

Restructure FunctionBody hierarchy for re-deferral #1278

Closed

pleath force-pushed the redefer branch 2 times, most recently from ea45851 to d6ac8c3 Compare October 26, 2016 18:43

This was referenced Oct 26, 2016

Minor functional fixes for redeferral. #1818

Closed

Fix a functional redeferral issue involving inlinees. #1815

Closed

pleath force-pushed the redefer branch 10 times, most recently from 154d897 to 317462a Compare October 31, 2016 16:54

pleath force-pushed the redefer branch from 317462a to 5b8e839 Compare November 4, 2016 22:40

This was referenced Nov 4, 2016

Redeferral fixes #1915

Closed

Couple of redeferral fixes #1905

Closed

Redeferral allocation savings #1845

Closed

pleath force-pushed the redefer branch from 5b8e839 to 225f48a Compare November 7, 2016 18:58

rajatd reviewed Nov 7, 2016

View reviewed changes

pleath force-pushed the redefer branch 2 times, most recently from 189c0f3 to 0826835 Compare November 8, 2016 01:12

pleath mentioned this pull request Nov 8, 2016

Unpin redeferred functions, etc. #1929

Closed

rajatd reviewed Nov 10, 2016

View reviewed changes

pleath changed the title ~~Redefer function bodies (experimental)~~ Redefer function bodies Nov 10, 2016

pleath force-pushed the redefer branch from 0826835 to 746730d Compare November 10, 2016 23:12

pleath force-pushed the redefer branch 2 times, most recently from b5b2104 to 223b4a9 Compare November 11, 2016 01:28

pleath force-pushed the redefer branch from 223b4a9 to a384d2e Compare November 11, 2016 02:22

chakrabot merged commit a384d2e into chakra-core:master Nov 11, 2016

		@@ -2062,19 +2080,6 @@ ThreadContext::ExecuteRecyclerCollectionFunction(Recycler * recycler, Collection

		BOOL ret = FALSE;

Uh oh!

Conversation

pleath commented Sep 15, 2016 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

pleath commented Oct 22, 2016

Uh oh!

pleath commented Oct 22, 2016

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

pleath commented Nov 8, 2016

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

rajatd commented Nov 10, 2016

Uh oh!

pleath commented Nov 10, 2016

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

pleath commented Sep 15, 2016 •

edited

Loading