New implementation of BLACKHOLEs

This replaces the global blackhole_queue with a clever scheme that
enables us to queue up blocked threads on the closure that they are
blocked on, while still avoiding atomic instructions in the common
case.

Advantages:

 - gets rid of a locked global data structure and some tricky GC code
   (replacing it with some per-thread data structures and different
   tricky GC code :)

 - wakeups are more prompt: parallel/concurrent performance should
   benefit.  I haven't seen anything dramatic in the parallel
   benchmarks so far, but a couple of threading benchmarks do improve
   a bit.

 - waking up a thread blocked on a blackhole is now O(1) (e.g. if
   it is the target of throwTo).

 - less sharing and better separation of Capabilities: communication
   is done with messages, the data structures are strictly owned by a
   Capability and cannot be modified except by sending messages.

 - this change will utlimately enable us to do more intelligent
   scheduling when threads block on each other.  This is what started
   off the whole thing, but it isn't done yet (#3838).

I'll be documenting all this on the wiki in due course.

1 files changed, 23 insertions, 5 deletions

diff --git a/compiler/codeGen/CgStackery.lhs b/compiler/codeGen/CgStackery.lhs
index 6683de4c8b..532127a147 100644
--- a/compiler/codeGen/CgStackery.lhs
+++ b/compiler/codeGen/CgStackery.lhs
@@ -17,7 +17,7 @@ module CgStackery (
 	setStackFrame, getStackFrame,
 	mkVirtStkOffsets, mkStkAmodes,
 	freeStackSlots, 
-	pushUpdateFrame, emitPushUpdateFrame,
+	pushUpdateFrame, pushBHUpdateFrame, emitPushUpdateFrame,
     ) where
 
 #include "HsVersions.h"
@@ -265,6 +265,14 @@ to reflect the frame pushed.
 \begin{code}
 pushUpdateFrame :: CmmExpr -> Code -> Code
 pushUpdateFrame updatee code
+  = pushSpecUpdateFrame mkUpdInfoLabel updatee code
+
+pushBHUpdateFrame :: CmmExpr -> Code -> Code
+pushBHUpdateFrame updatee code
+  = pushSpecUpdateFrame mkBHUpdInfoLabel updatee code
+
+pushSpecUpdateFrame :: CLabel -> CmmExpr -> Code -> Code
+pushSpecUpdateFrame lbl updatee code
   = do	{
       when debugIsOn $ do
     	{ EndOfBlockInfo _ sequel <- getEndOfBlockInfo ;
@@ -277,15 +285,25 @@ pushUpdateFrame updatee code
 		-- The location of the lowest-address
 		-- word of the update frame itself
 
-	; setEndOfBlockInfo (EndOfBlockInfo vsp UpdateCode) $
-	    do	{ emitPushUpdateFrame frame_addr updatee
+                -- NB. we used to set the Sequel to 'UpdateCode' so
+                -- that we could jump directly to the update code if
+                -- we know that the next frame on the stack is an
+                -- update frame.  However, the RTS can sometimes
+                -- change an update frame into something else (see
+                -- e.g. Note [upd-black-hole] in rts/sm/Scav.c), so we
+                -- no longer make this assumption.
+	; setEndOfBlockInfo (EndOfBlockInfo vsp OnStack) $
+	    do	{ emitSpecPushUpdateFrame lbl frame_addr updatee
 		; code }
 	}
 
 emitPushUpdateFrame :: CmmExpr -> CmmExpr -> Code
-emitPushUpdateFrame frame_addr updatee = do
+emitPushUpdateFrame = emitSpecPushUpdateFrame mkUpdInfoLabel
+
+emitSpecPushUpdateFrame :: CLabel -> CmmExpr -> CmmExpr -> Code
+emitSpecPushUpdateFrame lbl frame_addr updatee = do
 	stmtsC [  -- Set the info word
-		  CmmStore frame_addr (mkLblExpr mkUpdInfoLabel)
+		  CmmStore frame_addr (mkLblExpr lbl)
 		, -- And the updatee
 		  CmmStore (cmmOffsetB frame_addr off_updatee) updatee ]
 	initUpdFrameProf frame_addr