From: simonmar <unknown>
Date: Mon, 28 Feb 2005 17:01:23 +0000 (+0000)
Subject: [project @ 2005-02-28 17:01:13 by simonmar]
X-Git-Tag: Initial_conversion_from_CVS_complete~1008
X-Git-Url: http://git.megacz.com/?p=ghc-hetmet.git;a=commitdiff_plain;h=6d194f48d7313fa8daa004b7b3e3a55ffa52f4a9

[project @ 2005-02-28 17:01:13 by simonmar]
Optimisation: the update code only needs to check whether the closure
being updated is a BLACKHOLE_BQ if it has been blackholed; which is
only the case after lazy blackholing has taken place.

So to avoid this unnecessary cost, we have two kinds of update frame:
one which checks for BLACKHOLE_BQ and one that doesn't, and the lazy
blackholing algorithm converts the latter kind into the former as it
replaces thunks with blackholes.
---

diff --git a/ghc/includes/StgMiscClosures.h b/ghc/includes/StgMiscClosures.h
index 16d1483..18e9b63 100644
--- a/ghc/includes/StgMiscClosures.h
+++ b/ghc/includes/StgMiscClosures.h
@@ -37,6 +37,7 @@
 
 /* Stack frames */
 RTS_RET_INFO(stg_upd_frame_info);
+RTS_RET_INFO(stg_bh_upd_frame_info);
 RTS_RET_INFO(stg_noupd_frame_info);
 RTS_RET_INFO(stg_seq_frame_info);
 RTS_RET_INFO(stg_catch_frame_info);
@@ -45,6 +46,7 @@ RTS_RET_INFO(stg_atomically_frame_info);
 RTS_RET_INFO(stg_catch_stm_frame_info);
 
 RTS_ENTRY(stg_upd_frame_ret);
+RTS_ENTRY(stg_bh_upd_frame_ret);
 RTS_ENTRY(stg_seq_frame_ret);
 
 /* Entry code for constructors created by the bytecode interpreter */
diff --git a/ghc/includes/Updates.h b/ghc/includes/Updates.h
index 0845e20..034412e 100644
--- a/ghc/includes/Updates.h
+++ b/ghc/includes/Updates.h
@@ -28,12 +28,16 @@
    UPD_PERM_IND(updclosure,heapptr)
 # define UPD_SPEC_IND(updclosure, ind_info, heapptr, and_then) \
    UPD_PERM_IND(updclosure,heapptr); and_then
+# define NOBH_UPD_SPEC_IND(updclosure, ind_info, heapptr, and_then) \
+   NOBH_UPD_PERM_IND(updclosure,heapptr); and_then
 #else
 #  define SEMI ;
 # define UPD_IND(updclosure, heapptr) \
    UPD_REAL_IND(updclosure,INFO_PTR(stg_IND_info),heapptr,SEMI)
 # define UPD_SPEC_IND(updclosure, ind_info, heapptr, and_then) \
    UPD_REAL_IND(updclosure,ind_info,heapptr,and_then)
+# define NOBH_UPD_SPEC_IND(updclosure, ind_info, heapptr, and_then) \
+   NOBH_UPD_REAL_IND(updclosure,ind_info,heapptr,and_then)
 #endif
 
 /* These macros have to work in both C and C--, so here's the
@@ -71,6 +75,14 @@
 			      and_then);			\
 	BLOCK_END
 
+#define NOBH_UPD_REAL_IND(updclosure, ind_info, heapptr, and_then)	\
+        BLOCK_BEGIN						\
+	updateWithIndirection(GET_INFO(updclosure), ind_info,	\
+			      updclosure,			\
+			      heapptr,				\
+			      and_then);			\
+	BLOCK_END
+
 #if defined(PROFILING) || defined(TICKY_TICKY)
 #define UPD_PERM_IND(updclosure, heapptr)	\
         BLOCK_BEGIN				\
@@ -81,6 +93,13 @@
 				  updclosure,	\
 				  heapptr);	\
 	BLOCK_END
+
+#define NOBH_UPD_PERM_IND(updclosure, heapptr)	\
+        BLOCK_BEGIN				\
+	updateWithPermIndirection(GET_INFO(updclosure),		\
+				  updclosure,	\
+				  heapptr);	\
+	BLOCK_END
 #endif
 
 #if defined(RTS_SUPPORTS_THREADS)
diff --git a/ghc/rts/GC.c b/ghc/rts/GC.c
index a470f32..75f11a2 100644
--- a/ghc/rts/GC.c
+++ b/ghc/rts/GC.c
@@ -4121,6 +4121,11 @@ threadSqueezeStack(StgTSO *tso)
 		    // Todo: maybe use SET_HDR() and remove LDV_RECORD_CREATE()?
 		    SET_INFO(bh,&stg_BLACKHOLE_info);
 
+		    // Set the update frame to stg_bh_upd_info, which
+		    // checks for blackholes (the normal update frame
+		    // doesn't check, for efficiency).
+		    ((StgClosure *)frame)->header.info = &stg_bh_upd_frame_info;
+
 		    // We pretend that bh has just been created.
 		    LDV_RECORD_CREATE(bh);
 		}
diff --git a/ghc/rts/Makefile b/ghc/rts/Makefile
index fe7841c..15cbdd4 100644
--- a/ghc/rts/Makefile
+++ b/ghc/rts/Makefile
@@ -338,14 +338,10 @@ ifneq "$(BootingFromHc)" "YES"
 	$(HC_POST_OPTS)
 
 %.$(way_)hc : %.cmm $(H_FILES)
-	$(HC_PRE_OPTS)
 	$(HC) $(HC_OPTS) -C $< -o $@
-	$(HC_POST_OPTS)
 
 %.$(way_)s : %.cmm $(H_FILES)
-	$(HC_PRE_OPTS)
 	$(HC) $(HC_OPTS) -S $< -o $@
-	$(HC_POST_OPTS)
 endif
 
 #-----------------------------------------------------------------------------
diff --git a/ghc/rts/Updates.cmm b/ghc/rts/Updates.cmm
index 02d1827..bec9b40 100644
--- a/ghc/rts/Updates.cmm
+++ b/ghc/rts/Updates.cmm
@@ -55,7 +55,7 @@
 	  /* ToDo: it might be a PAP, so we should check... */		\
 	  TICK_UPD_CON_IN_NEW(sizeW_fromITBL(%GET_STD_INFO(updatee)));	\
 									\
-	  UPD_SPEC_IND(updatee, ind_info, R1, jump (ret));		\
+	  NOBH_UPD_SPEC_IND(updatee, ind_info, R1, jump (ret));		\
 	}
 
 UPD_FRAME_ENTRY_TEMPLATE(stg_upd_frame_0_ret,stg_IND_0_info,%RET_VEC(Sp(0),0))
@@ -102,6 +102,62 @@ INFO_TABLE_RET( stg_upd_frame,
 	    )
 UPD_FRAME_ENTRY_TEMPLATE(,stg_IND_direct_info,%ENTRY_CODE(Sp(0)))
 
+/* -----------------------------------------------------------------------------
+ * Update frames that check for BLACKHOLE_BQ
+ *
+ * A normal update frame doesn't check whether the closure has been
+ * overwritten with BLACKHOLE_BQ, because it can't until lazy blackholing
+ * has happened.  Lazy blackholing overwrites the update frames pointing
+ * to BLACKHOLE with stg_bh_upd_frame_info.
+ * -------------------------------------------------------------------------- */
+
+#define BH_UPD_FRAME_ENTRY_TEMPLATE(label,ind_info,ret)			\
+        label								\
+	{								\
+          W_ updatee;							\
+									\
+          updatee = StgUpdateFrame_updatee(Sp);				\
+									\
+	  /* remove the update frame from the stack */			\
+	  Sp = Sp + SIZEOF_StgUpdateFrame;				\
+									\
+	  /* ToDo: it might be a PAP, so we should check... */		\
+	  TICK_UPD_CON_IN_NEW(sizeW_fromITBL(%GET_STD_INFO(updatee)));	\
+									\
+	  UPD_SPEC_IND(updatee, ind_info, R1, jump (ret));		\
+	}
+
+BH_UPD_FRAME_ENTRY_TEMPLATE(stg_bh_upd_frame_0_ret,stg_IND_0_info,%RET_VEC(Sp(0),0))
+BH_UPD_FRAME_ENTRY_TEMPLATE(stg_bh_upd_frame_1_ret,stg_IND_1_info,%RET_VEC(Sp(0),1))
+BH_UPD_FRAME_ENTRY_TEMPLATE(stg_bh_upd_frame_2_ret,stg_IND_2_info,%RET_VEC(Sp(0),2))
+BH_UPD_FRAME_ENTRY_TEMPLATE(stg_bh_upd_frame_3_ret,stg_IND_3_info,%RET_VEC(Sp(0),3))
+BH_UPD_FRAME_ENTRY_TEMPLATE(stg_bh_upd_frame_4_ret,stg_IND_4_info,%RET_VEC(Sp(0),4))
+BH_UPD_FRAME_ENTRY_TEMPLATE(stg_bh_upd_frame_5_ret,stg_IND_5_info,%RET_VEC(Sp(0),5))
+BH_UPD_FRAME_ENTRY_TEMPLATE(stg_bh_upd_frame_6_ret,stg_IND_6_info,%RET_VEC(Sp(0),6))
+BH_UPD_FRAME_ENTRY_TEMPLATE(stg_bh_upd_frame_7_ret,stg_IND_7_info,%RET_VEC(Sp(0),7))
+
+#if MAX_VECTORED_RTN > 8
+#error MAX_VECTORED_RTN has changed: please modify stg_bh_upd_frame too.
+#endif
+
+/* this bitmap indicates that the first word of an update frame is a
+ * non-pointer - this is the update frame link.  (for profiling,
+ * there's a cost-centre-stack in there too).
+ */
+
+INFO_TABLE_RET( stg_bh_upd_frame, 
+	    UPD_FRAME_WORDS, UPD_FRAME_BITMAP, UPDATE_FRAME,
+	    stg_bh_upd_frame_0_ret,
+	    stg_bh_upd_frame_1_ret,
+	    stg_bh_upd_frame_2_ret,
+	    stg_bh_upd_frame_3_ret,
+	    stg_bh_upd_frame_4_ret,
+	    stg_bh_upd_frame_5_ret,
+	    stg_bh_upd_frame_6_ret,
+	    stg_bh_upd_frame_7_ret
+	    )
+UPD_FRAME_ENTRY_TEMPLATE(,stg_IND_direct_info,%ENTRY_CODE(Sp(0)))
+
 /*-----------------------------------------------------------------------------
   Seq frames