[project @ 2002-12-05 23:49:43 by mthomas]
[ghc-hetmet.git] / ghc / rts / HeapStackCheck.hc
index cf85fbe..f3882fe 100644 (file)
@@ -1,5 +1,5 @@
 /* -----------------------------------------------------------------------------
- * $Id: HeapStackCheck.hc,v 1.7 1999/05/13 17:31:10 simonm Exp $
+ * $Id: HeapStackCheck.hc,v 1.26 2002/03/02 17:43:44 sof Exp $
  *
  * (c) The GHC Team, 1998-1999
  *
@@ -7,11 +7,11 @@
  *
  * ---------------------------------------------------------------------------*/
 
+#include "Stg.h"
 #include "Rts.h"
 #include "Storage.h"           /* for CurrentTSO */
 #include "StgRun.h"    /* for StgReturn and register saving */
 #include "Schedule.h"   /* for context_switch */
-#include "HeapStackCheck.h"
 
 /* Stack/Heap Check Failure
  * ------------------------
@@ -20,7 +20,7 @@
  *
  *    - If the context_switch flag is set, indicating that there are more
  *      threads waiting to run, we yield to the scheduler 
- *     (return ThreadYeilding).
+ *     (return ThreadYielding).
  *
  *    - If Hp > HpLim, we've had a heap check failure.  This means we've
  *     come to the end of the current heap block, so we try to chain
  * ThreadRunGHC thread.
  */
 
-
 #define GC_GENERIC                                     \
   if (Hp > HpLim) {                                    \
-    if (ExtendNursery(Hp,HpLim)) {                     \
+    Hp -= HpAlloc;                                     \
+    if (HpAlloc <= BLOCK_SIZE_W && ExtendNursery(Hp,HpLim)) {\
        if (context_switch) {                           \
            R1.i = ThreadYielding;                      \
        } else {                                        \
     R1.i = StackOverflow;                              \
   }                                                    \
   SaveThreadState();                                   \
-  CurrentTSO->whatNext = ThreadRunGHC;                 \
+  CurrentTSO->what_next = ThreadRunGHC;                        \
   JMP_(StgReturn);
 
 #define GC_ENTER                                       \
   if (Hp > HpLim) {                                    \
-    if (ExtendNursery(Hp,HpLim)) {                     \
+    Hp -= HpAlloc;                                     \
+    if (HpAlloc <= BLOCK_SIZE_W && ExtendNursery(Hp,HpLim)) {\
        if (context_switch) {                           \
            R1.i = ThreadYielding;                      \
        } else {                                        \
     R1.i = StackOverflow;                              \
   }                                                    \
   SaveThreadState();                                   \
-  CurrentTSO->whatNext = ThreadEnterGHC;               \
+  CurrentTSO->what_next = ThreadEnterGHC;              \
   JMP_(StgReturn);
 
 #define HP_GENERIC                     \
   SaveThreadState();                   \
-  CurrentTSO->whatNext = ThreadRunGHC; \
+  CurrentTSO->what_next = ThreadRunGHC;        \
   R1.i = HeapOverflow;                 \
   JMP_(StgReturn);
 
 #define STK_GENERIC                    \
   SaveThreadState();                   \
-  CurrentTSO->whatNext = ThreadRunGHC; \
+  CurrentTSO->what_next = ThreadRunGHC;        \
   R1.i = StackOverflow;                        \
   JMP_(StgReturn);
 
 #define YIELD_GENERIC                  \
   SaveThreadState();                   \
-  CurrentTSO->whatNext = ThreadRunGHC; \
+  CurrentTSO->what_next = ThreadRunGHC;        \
   R1.i = ThreadYielding;               \
   JMP_(StgReturn);
 
-#define YIELD_TO_HUGS                    \
+#define YIELD_TO_INTERPRETER             \
   SaveThreadState();                     \
-  CurrentTSO->whatNext = ThreadEnterHugs; \
+  CurrentTSO->what_next = ThreadEnterInterp; \
   R1.i = ThreadYielding;                 \
   JMP_(StgReturn);
 
 #define BLOCK_GENERIC                  \
   SaveThreadState();                   \
-  CurrentTSO->whatNext = ThreadRunGHC; \
+  CurrentTSO->what_next = ThreadRunGHC;        \
   R1.i = ThreadBlocked;                        \
   JMP_(StgReturn);
 
 #define BLOCK_ENTER                    \
   SaveThreadState();                   \
-  CurrentTSO->whatNext = ThreadEnterGHC;\
+  CurrentTSO->what_next = ThreadEnterGHC;\
   R1.i = ThreadBlocked;                        \
   JMP_(StgReturn);
 
@@ -150,7 +151,7 @@ EXTFUN(stg_gc_entertop)
    There are canned sequences for 'n' pointer values in registers.
    -------------------------------------------------------------------------- */
 
-EXTFUN(stg_gc_enter_1)
+EXTFUN(__stg_gc_enter_1)
 {
   FB_
   Sp -= 1;
@@ -159,6 +160,18 @@ EXTFUN(stg_gc_enter_1)
   FE_
 }
 
+EXTFUN(stg_gc_enter_1_hponly)
+{
+  FB_
+  Sp -= 1;
+  Sp[0] = R1.w;
+  R1.i = HeapOverflow;
+  SaveThreadState();
+  CurrentTSO->what_next = ThreadEnterGHC;
+  JMP_(StgReturn);
+  FE_
+}
+
 /*- 2 Regs--------------------------------------------------------------------*/
 
 EXTFUN(stg_gc_enter_2)
@@ -264,6 +277,334 @@ EXTFUN(stg_gc_enter_8)
   FE_
 }
 
+#if defined(GRAN)
+/*
+  ToDo: merge the block and yield macros, calling something like BLOCK(N)
+        at the end;
+*/
+
+/* 
+   Should we actually ever do a yield in such a case?? -- HWL
+*/
+EXTFUN(gran_yield_0)
+{
+  FB_
+  SaveThreadState();                                   
+  CurrentTSO->what_next = ThreadEnterGHC;              
+  R1.i = ThreadYielding;
+  JMP_(StgReturn);
+  FE_
+}
+
+EXTFUN(gran_yield_1)
+{
+  FB_
+  Sp -= 1;
+  Sp[0] = R1.w;
+  SaveThreadState();                                   
+  CurrentTSO->what_next = ThreadEnterGHC;              
+  R1.i = ThreadYielding;
+  JMP_(StgReturn);
+  FE_
+}
+
+/*- 2 Regs--------------------------------------------------------------------*/
+
+EXTFUN(gran_yield_2)
+{
+  FB_
+  Sp -= 2;
+  Sp[1] = R2.w;
+  Sp[0] = R1.w;
+  SaveThreadState();                                   
+  CurrentTSO->what_next = ThreadEnterGHC;              
+  R1.i = ThreadYielding;
+  JMP_(StgReturn);
+  FE_
+}
+
+/*- 3 Regs -------------------------------------------------------------------*/
+
+EXTFUN(gran_yield_3)
+{
+  FB_
+  Sp -= 3;
+  Sp[2] = R3.w;
+  Sp[1] = R2.w;
+  Sp[0] = R1.w;
+  SaveThreadState();                                   
+  CurrentTSO->what_next = ThreadEnterGHC;              
+  R1.i = ThreadYielding;
+  JMP_(StgReturn);
+  FE_
+}
+
+/*- 4 Regs -------------------------------------------------------------------*/
+
+EXTFUN(gran_yield_4)
+{
+  FB_
+  Sp -= 4;
+  Sp[3] = R4.w;
+  Sp[2] = R3.w;
+  Sp[1] = R2.w;
+  Sp[0] = R1.w;
+  SaveThreadState();                                   
+  CurrentTSO->what_next = ThreadEnterGHC;              
+  R1.i = ThreadYielding;
+  JMP_(StgReturn);
+  FE_
+}
+
+/*- 5 Regs -------------------------------------------------------------------*/
+
+EXTFUN(gran_yield_5)
+{
+  FB_
+  Sp -= 5;
+  Sp[4] = R5.w;
+  Sp[3] = R4.w;
+  Sp[2] = R3.w;
+  Sp[1] = R2.w;
+  Sp[0] = R1.w;
+  SaveThreadState();                                   
+  CurrentTSO->what_next = ThreadEnterGHC;              
+  R1.i = ThreadYielding;
+  JMP_(StgReturn);
+  FE_
+}
+
+/*- 6 Regs -------------------------------------------------------------------*/
+
+EXTFUN(gran_yield_6)
+{
+  FB_
+  Sp -= 6;
+  Sp[5] = R6.w;
+  Sp[4] = R5.w;
+  Sp[3] = R4.w;
+  Sp[2] = R3.w;
+  Sp[1] = R2.w;
+  Sp[0] = R1.w;
+  SaveThreadState();                                   
+  CurrentTSO->what_next = ThreadEnterGHC;              
+  R1.i = ThreadYielding;
+  JMP_(StgReturn);
+  FE_
+}
+
+/*- 7 Regs -------------------------------------------------------------------*/
+
+EXTFUN(gran_yield_7)
+{
+  FB_
+  Sp -= 7;
+  Sp[6] = R7.w;
+  Sp[5] = R6.w;
+  Sp[4] = R5.w;
+  Sp[3] = R4.w;
+  Sp[2] = R3.w;
+  Sp[1] = R2.w;
+  Sp[0] = R1.w;
+  SaveThreadState();                                   
+  CurrentTSO->what_next = ThreadEnterGHC;              
+  R1.i = ThreadYielding;
+  JMP_(StgReturn);
+  FE_
+}
+
+/*- 8 Regs -------------------------------------------------------------------*/
+
+EXTFUN(gran_yield_8)
+{
+  FB_
+  Sp -= 8;
+  Sp[7] = R8.w;
+  Sp[6] = R7.w;
+  Sp[5] = R6.w;
+  Sp[4] = R5.w;
+  Sp[3] = R4.w;
+  Sp[2] = R3.w;
+  Sp[1] = R2.w;
+  Sp[0] = R1.w;
+  SaveThreadState();                                   
+  CurrentTSO->what_next = ThreadEnterGHC;              
+  R1.i = ThreadYielding;
+  JMP_(StgReturn);
+  FE_
+}
+
+// the same routines but with a block rather than a yield
+
+EXTFUN(gran_block_1)
+{
+  FB_
+  Sp -= 1;
+  Sp[0] = R1.w;
+  SaveThreadState();                                   
+  CurrentTSO->what_next = ThreadEnterGHC;              
+  R1.i = ThreadBlocked;
+  JMP_(StgReturn);
+  FE_
+}
+
+/*- 2 Regs--------------------------------------------------------------------*/
+
+EXTFUN(gran_block_2)
+{
+  FB_
+  Sp -= 2;
+  Sp[1] = R2.w;
+  Sp[0] = R1.w;
+  SaveThreadState();                                   
+  CurrentTSO->what_next = ThreadEnterGHC;              
+  R1.i = ThreadBlocked;
+  JMP_(StgReturn);
+  FE_
+}
+
+/*- 3 Regs -------------------------------------------------------------------*/
+
+EXTFUN(gran_block_3)
+{
+  FB_
+  Sp -= 3;
+  Sp[2] = R3.w;
+  Sp[1] = R2.w;
+  Sp[0] = R1.w;
+  SaveThreadState();                                   
+  CurrentTSO->what_next = ThreadEnterGHC;              
+  R1.i = ThreadBlocked;
+  JMP_(StgReturn);
+  FE_
+}
+
+/*- 4 Regs -------------------------------------------------------------------*/
+
+EXTFUN(gran_block_4)
+{
+  FB_
+  Sp -= 4;
+  Sp[3] = R4.w;
+  Sp[2] = R3.w;
+  Sp[1] = R2.w;
+  Sp[0] = R1.w;
+  SaveThreadState();                                   
+  CurrentTSO->what_next = ThreadEnterGHC;              
+  R1.i = ThreadBlocked;
+  JMP_(StgReturn);
+  FE_
+}
+
+/*- 5 Regs -------------------------------------------------------------------*/
+
+EXTFUN(gran_block_5)
+{
+  FB_
+  Sp -= 5;
+  Sp[4] = R5.w;
+  Sp[3] = R4.w;
+  Sp[2] = R3.w;
+  Sp[1] = R2.w;
+  Sp[0] = R1.w;
+  SaveThreadState();                                   
+  CurrentTSO->what_next = ThreadEnterGHC;              
+  R1.i = ThreadBlocked;
+  JMP_(StgReturn);
+  FE_
+}
+
+/*- 6 Regs -------------------------------------------------------------------*/
+
+EXTFUN(gran_block_6)
+{
+  FB_
+  Sp -= 6;
+  Sp[5] = R6.w;
+  Sp[4] = R5.w;
+  Sp[3] = R4.w;
+  Sp[2] = R3.w;
+  Sp[1] = R2.w;
+  Sp[0] = R1.w;
+  SaveThreadState();                                   
+  CurrentTSO->what_next = ThreadEnterGHC;              
+  R1.i = ThreadBlocked;
+  JMP_(StgReturn);
+  FE_
+}
+
+/*- 7 Regs -------------------------------------------------------------------*/
+
+EXTFUN(gran_block_7)
+{
+  FB_
+  Sp -= 7;
+  Sp[6] = R7.w;
+  Sp[5] = R6.w;
+  Sp[4] = R5.w;
+  Sp[3] = R4.w;
+  Sp[2] = R3.w;
+  Sp[1] = R2.w;
+  Sp[0] = R1.w;
+  SaveThreadState();                                   
+  CurrentTSO->what_next = ThreadEnterGHC;              
+  R1.i = ThreadBlocked;
+  JMP_(StgReturn);
+  FE_
+}
+
+/*- 8 Regs -------------------------------------------------------------------*/
+
+EXTFUN(gran_block_8)
+{
+  FB_
+  Sp -= 8;
+  Sp[7] = R8.w;
+  Sp[6] = R7.w;
+  Sp[5] = R6.w;
+  Sp[4] = R5.w;
+  Sp[3] = R4.w;
+  Sp[2] = R3.w;
+  Sp[1] = R2.w;
+  Sp[0] = R1.w;
+  SaveThreadState();                                   
+  CurrentTSO->what_next = ThreadEnterGHC;              
+  R1.i = ThreadBlocked;
+  JMP_(StgReturn);
+  FE_
+}
+
+#endif
+
+#if 0 && defined(PAR)
+
+/*
+  Similar to stg_block_1 (called via StgMacro BLOCK_NP) but separates the
+  saving of the thread state from the actual jump via an StgReturn.
+  We need this separation because we call RTS routines in blocking entry codes
+  before jumping back into the RTS (see parallel/FetchMe.hc).
+*/
+
+EXTFUN(par_block_1_no_jump)
+{
+  FB_
+  Sp -= 1;
+  Sp[0] = R1.w;
+  SaveThreadState();                                   
+  FE_
+}
+
+EXTFUN(par_jump)
+{
+  FB_
+  CurrentTSO->what_next = ThreadEnterGHC;              
+  R1.i = ThreadBlocked;
+  JMP_(StgReturn);
+  FE_
+}
+
+#endif
+
 /* -----------------------------------------------------------------------------
    For a case expression on a polymorphic or function-typed object, if
    the default branch (there can only be one branch) of the case fails
@@ -293,22 +634,39 @@ EXTFUN(stg_gc_seq_1)
    cases are covered below.
    -------------------------------------------------------------------------- */
 
-/*-- No regsiters live, return address already on the stack: ---------------- */
+/*-- No registers live (probably a void return) ----------------------------- */
+
+/* If we change the policy for thread startup to *not* remove the
+ * return address from the stack, we can get rid of this little
+ * function/info table...  
+ */
+INFO_TABLE_SRT_BITMAP(stg_gc_noregs_info, stg_gc_noregs_ret, 0/*BITMAP*/, 
+                     0/*SRT*/, 0/*SRT_OFF*/, 0/*SRT_LEN*/, 
+                     RET_SMALL,, EF_, 0, 0);
+
+EXTFUN(stg_gc_noregs_ret)
+{
+  FB_
+  JMP_(ENTRY_CODE(Sp[0]));
+  FE_
+}
 
 EXTFUN(stg_gc_noregs)
 {
   FB_
+  Sp -= 1;
+  Sp[0] = (W_)&stg_gc_noregs_info;
   GC_GENERIC
   FE_
 }
 
 /*-- R1 is boxed/unpointed -------------------------------------------------- */
 
-INFO_TABLE_SRT_BITMAP(stg_gc_unpt_r1_info, stg_gc_unpt_r1_entry, 0/*BITMAP*/, 
+INFO_TABLE_SRT_BITMAP(stg_gc_unpt_r1_info, stg_gc_unpt_r1_ret, 0/*BITMAP*/, 
                      0/*SRT*/, 0/*SRT_OFF*/, 0/*SRT_LEN*/, 
                      RET_SMALL,, EF_, 0, 0);
 
-EXTFUN(stg_gc_unpt_r1_entry)
+EXTFUN(stg_gc_unpt_r1_ret)
 {
   FB_
   R1.w = Sp[0];
@@ -327,14 +685,28 @@ EXTFUN(stg_gc_unpt_r1)
   FE_
 }
 
+/*-- Unboxed tuple return (unregisterised build only)------------------ */
+
+INFO_TABLE_SRT_BITMAP(stg_ut_1_0_unreg_info, stg_ut_1_0_unreg_ret, 0/*BITMAP*/, 
+                     0/*SRT*/, 0/*SRT_OFF*/, 0/*SRT_LEN*/, 
+                     RET_SMALL,, EF_, 0, 0);
+
+EXTFUN(stg_ut_1_0_unreg_ret)
+{
+  FB_
+  /* R1 is on the stack (*Sp) */
+  JMP_(ENTRY_CODE(Sp[1]));
+  FE_
+}
+
 /*-- R1 is unboxed -------------------------------------------------- */
 
-INFO_TABLE_SRT_BITMAP(stg_gc_unbx_r1_info, stg_gc_unbx_r1_entry, 1/*BITMAP*/,
+INFO_TABLE_SRT_BITMAP(stg_gc_unbx_r1_info, stg_gc_unbx_r1_ret, 1/*BITMAP*/,
                      0/*SRT*/, 0/*SRT_OFF*/, 0/*SRT_LEN*/, 
                      RET_SMALL,, EF_, 0, 0);
 /* the 1 is a bitmap - i.e. 1 non-pointer word on the stack. */
 
-EXTFUN(stg_gc_unbx_r1_entry)
+EXTFUN(stg_gc_unbx_r1_ret)
 {
   FB_
   R1.w = Sp[0];
@@ -355,11 +727,11 @@ EXTFUN(stg_gc_unbx_r1)
 
 /*-- F1 contains a float ------------------------------------------------- */
 
-INFO_TABLE_SRT_BITMAP(stg_gc_f1_info, stg_gc_f1_entry, 1/*BITMAP*/,
+INFO_TABLE_SRT_BITMAP(stg_gc_f1_info, stg_gc_f1_ret, 1/*BITMAP*/,
                      0/*SRT*/, 0/*SRT_OFF*/, 0/*SRT_LEN*/, 
                      RET_SMALL,, EF_, 0, 0);
 
-EXTFUN(stg_gc_f1_entry)
+EXTFUN(stg_gc_f1_ret)
 {
   FB_
   F1 = PK_FLT(Sp);
@@ -388,11 +760,11 @@ EXTFUN(stg_gc_f1)
 #  define DBL_BITMAP 3
 #endif 
 
-INFO_TABLE_SRT_BITMAP(stg_gc_d1_info, stg_gc_d1_entry, DBL_BITMAP,
+INFO_TABLE_SRT_BITMAP(stg_gc_d1_info, stg_gc_d1_ret, DBL_BITMAP,
                      0/*SRT*/, 0/*SRT_OFF*/, 0/*SRT_LEN*/, 
                      RET_SMALL,, EF_, 0, 0);
 
-EXTFUN(stg_gc_d1_entry)
+EXTFUN(stg_gc_d1_ret)
 {
   FB_
   D1 = PK_DBL(Sp);
@@ -411,6 +783,40 @@ EXTFUN(stg_gc_d1)
   FE_
 }
 
+
+/*-- L1 contains an int64 ------------------------------------------------- */
+
+/* we support int64s of either 1 or 2 words in size */
+
+#if SIZEOF_VOID_P == 8
+#  define LLI_BITMAP 1
+#else
+#  define LLI_BITMAP 3
+#endif 
+
+INFO_TABLE_SRT_BITMAP(stg_gc_l1_info, stg_gc_l1_ret, LLI_BITMAP,
+                     0/*SRT*/, 0/*SRT_OFF*/, 0/*SRT_LEN*/, 
+                     RET_SMALL,, EF_, 0, 0);
+
+EXTFUN(stg_gc_l1_ret)
+{
+  FB_
+  L1 = PK_Int64(Sp);
+  Sp += sizeofW(StgWord64);
+  JMP_(ENTRY_CODE(Sp[0]));
+  FE_
+}
+
+EXTFUN(stg_gc_l1)
+{
+  FB_
+  Sp -= 1 + sizeofW(StgWord64);
+  ASSIGN_Int64(Sp+1,L1);
+  Sp[0] = (W_)&stg_gc_l1_info;
+  GC_GENERIC
+  FE_
+}
+
 /* -----------------------------------------------------------------------------
    Heap checks for unboxed tuple case alternatives
 
@@ -441,11 +847,11 @@ EXTFUN(stg_gc_d1)
 
 /*---- R1 contains a pointer: ------ */
 
-INFO_TABLE_SRT_BITMAP(stg_gc_ut_1_0_info, stg_gc_ut_1_0_entry, 1/*BITMAP*/, 
+INFO_TABLE_SRT_BITMAP(stg_gc_ut_1_0_info, stg_gc_ut_1_0_ret, 1/*BITMAP*/, 
                      0/*SRT*/, 0/*SRT_OFF*/, 0/*SRT_LEN*/, 
                      RET_SMALL,, EF_, 0, 0);
 
-EXTFUN(stg_gc_ut_1_0_entry)
+EXTFUN(stg_gc_ut_1_0_ret)
 {
   FB_
   R1.w = Sp[1];
@@ -467,11 +873,11 @@ EXTFUN(stg_gc_ut_1_0)
 
 /*---- R1 contains a non-pointer: ------ */
 
-INFO_TABLE_SRT_BITMAP(stg_gc_ut_0_1_info, stg_gc_ut_0_1_entry, 3/*BITMAP*/, 
+INFO_TABLE_SRT_BITMAP(stg_gc_ut_0_1_info, stg_gc_ut_0_1_ret, 3/*BITMAP*/, 
                      0/*SRT*/, 0/*SRT_OFF*/, 0/*SRT_LEN*/, 
                      RET_SMALL,, EF_, 0, 0);
 
-EXTFUN(stg_gc_ut_0_1_entry)
+EXTFUN(stg_gc_ut_0_1_ret)
 {
   FB_
   R1.w = Sp[1];
@@ -522,7 +928,7 @@ EXTFUN(stg_gc_ut_0_1)
 
 /*- 0 Regs -------------------------------------------------------------------*/
 
-EXTFUN(stg_chk_0)
+EXTFUN(__stg_chk_0)
 {
   FB_
   Sp -= 1;
@@ -533,7 +939,7 @@ EXTFUN(stg_chk_0)
 
 /*- 1 Reg --------------------------------------------------------------------*/
 
-EXTFUN(stg_chk_1)
+EXTFUN(__stg_chk_1)
 {
   FB_
   Sp -= 2;
@@ -810,31 +1216,20 @@ FN_(stg_gen_yield)
   FE_
 }
 
-INFO_TABLE_SRT_BITMAP(stg_yield_noregs_info, stg_yield_noregs_ret, 0/*BITMAP*/, 
-                     0/*SRT*/, 0/*SRT_OFF*/, 0/*SRT_LEN*/, 
-                     RET_SMALL,, EF_, 0, 0);
-
-FN_(stg_yield_noregs_ret)
-{
-  FB_
-  JMP_(ENTRY_CODE(Sp[0]));
-  FE_
-}
-
 FN_(stg_yield_noregs)
 {
   FB_
   Sp--;
-  Sp[0] = (W_)&stg_yield_noregs_info;
+  Sp[0] = (W_)&stg_gc_noregs_info;
   YIELD_GENERIC;
   FE_
 }
 
-FN_(stg_yield_to_Hugs)
+FN_(stg_yield_to_interpreter)
 {
   FB_
   /* No need to save everything - no live registers */
-  YIELD_TO_HUGS
+  YIELD_TO_INTERPRETER
   FE_
 }
 
@@ -850,6 +1245,15 @@ FN_(stg_gen_block)
   FE_
 }
 
+FN_(stg_block_noregs)
+{
+  FB_
+  Sp--;
+  Sp[0] = (W_)&stg_gc_noregs_info;
+  BLOCK_GENERIC;
+  FE_
+}
+
 FN_(stg_block_1)
 {
   FB_
@@ -858,3 +1262,71 @@ FN_(stg_block_1)
   BLOCK_ENTER;
   FE_
 }
+
+/* -----------------------------------------------------------------------------
+ * takeMVar/putMVar-specific blocks
+ *
+ * Stack layout for a thread blocked in takeMVar:
+ *      
+ *       ret. addr
+ *       ptr to MVar   (R1)
+ *       stg_block_takemvar_info
+ *
+ * Stack layout for a thread blocked in putMVar:
+ *      
+ *       ret. addr
+ *       ptr to Value  (R2)
+ *       ptr to MVar   (R1)
+ *       stg_block_putmvar_info
+ *
+ * See PrimOps.hc for a description of the workings of take/putMVar.
+ * 
+ * -------------------------------------------------------------------------- */
+
+INFO_TABLE_SRT_BITMAP(stg_block_takemvar_info,  stg_block_takemvar_ret,
+                     0/*BITMAP*/, 0/*SRT*/, 0/*SRT_OFF*/, 0/*SRT_LEN*/, 
+                     RET_SMALL,, IF_, 0, 0);
+
+IF_(stg_block_takemvar_ret)
+{
+  FB_
+  R1.w = Sp[0];
+  Sp++;
+  JMP_(takeMVarzh_fast);
+  FE_
+}
+
+FN_(stg_block_takemvar)
+{
+  FB_
+  Sp -= 2;
+  Sp[1] = R1.w;
+  Sp[0] = (W_)&stg_block_takemvar_info;
+  BLOCK_GENERIC;
+  FE_
+}
+
+INFO_TABLE_SRT_BITMAP(stg_block_putmvar_info,  stg_block_putmvar_ret,
+                     0/*BITMAP*/, 0/*SRT*/, 0/*SRT_OFF*/, 0/*SRT_LEN*/, 
+                     RET_SMALL,, IF_, 0, 0);
+
+IF_(stg_block_putmvar_ret)
+{
+  FB_
+  R2.w = Sp[1];
+  R1.w = Sp[0];
+  Sp += 2;
+  JMP_(putMVarzh_fast);
+  FE_
+}
+
+FN_(stg_block_putmvar)
+{
+  FB_
+  Sp -= 3;
+  Sp[2] = R2.w;
+  Sp[1] = R1.w;
+  Sp[0] = (W_)&stg_block_putmvar_info;
+  BLOCK_GENERIC;
+  FE_
+}