update submodules for GHC.HetMet.GArrow -> Control.GArrow renaming
[ghc-hetmet.git] / rts / PrimOps.cmm
index 029b2b7..e17c6fb 100644 (file)
@@ -1,6 +1,6 @@
 /* -----------------------------------------------------------------------------
  *
- * (c) The GHC Team, 1998-2004
+ * (c) The GHC Team, 1998-2011
  *
  * Out-of-line primitive operations
  *
@@ -34,7 +34,7 @@ import pthread_mutex_unlock;
 import base_ControlziExceptionziBase_nestedAtomically_closure;
 import EnterCriticalSection;
 import LeaveCriticalSection;
-import ghczmprim_GHCziBool_False_closure;
+import ghczmprim_GHCziTypes_False_closure;
 #if !defined(mingw32_HOST_OS)
 import sm_mutex;
 #endif
@@ -109,6 +109,11 @@ stg_newAlignedPinnedByteArrayzh
     n = R1;
     alignment = R2;
 
+    /* we always supply at least word-aligned memory, so there's no
+       need to allow extra space for alignment if the requirement is less
+       than a word.  This also prevents mischief with alignment == 0. */
+    if (alignment <= SIZEOF_W) { alignment = 1; }
+
     bytes = n;
 
     /* payload_words is what we will tell the profiler we had to allocate */
@@ -207,6 +212,7 @@ stg_unsafeThawArrayzh
   }
 }
 
+
 /* -----------------------------------------------------------------------------
    MutVar primitives
    -------------------------------------------------------------------------- */
@@ -225,6 +231,25 @@ stg_newMutVarzh
     RET_P(mv);
 }
 
+stg_casMutVarzh
+ /* MutVar# s a -> a -> a -> State# s -> (# State#, Int#, a #) */
+{
+    W_ mv, old, new, h;
+
+    mv  = R1;
+    old = R2;
+    new = R3;
+
+    (h) = foreign "C" cas(mv + SIZEOF_StgHeader + OFFSET_StgMutVar_var,
+                          old, new) [];
+    if (h != old) {
+        RET_NP(1,h);
+    } else {
+        RET_NP(0,h);
+    }
+}
+
+
 stg_atomicModifyMutVarzh
 {
     W_ mv, f, z, x, y, r, h;
@@ -626,14 +651,9 @@ stg_threadStatuszh
     W_ tso;
     W_ why_blocked;
     W_ what_next;
-    W_ ret;
+    W_ ret, cap, locked;
 
     tso = R1;
-    loop:
-      if (TO_W_(StgTSO_what_next(tso)) == ThreadRelocated) {
-          tso = StgTSO__link(tso);
-          goto loop;
-      }
 
     what_next   = TO_W_(StgTSO_what_next(tso));
     why_blocked = TO_W_(StgTSO_why_blocked(tso));
@@ -651,7 +671,16 @@ stg_threadStatuszh
             ret = why_blocked;
         }
     }
-    RET_N(ret);
+
+    cap = TO_W_(Capability_no(StgTSO_cap(tso)));
+
+    if ((TO_W_(StgTSO_flags(tso)) & TSO_LOCKED) != 0) {
+        locked = 1;
+    } else {
+        locked = 0;
+    }
+
+    RET_NNN(ret,cap,locked);
 }
 
 /* -----------------------------------------------------------------------------
@@ -934,9 +963,9 @@ stg_retryzh
 
   // Find the enclosing ATOMICALLY_FRAME or CATCH_RETRY_FRAME
 retry_pop_stack:
-  StgTSO_sp(CurrentTSO) = Sp;
-  (frame_type) = foreign "C" findRetryFrameHelper(CurrentTSO "ptr") [];
-  Sp = StgTSO_sp(CurrentTSO);
+  SAVE_THREAD_STATE();
+  (frame_type) = foreign "C" findRetryFrameHelper(MyCapability(), CurrentTSO "ptr") [];
+  LOAD_THREAD_STATE();
   frame = Sp;
   trec = StgTSO_trec(CurrentTSO);
   outer  = StgTRecHeader_enclosing_trec(trec);
@@ -1133,13 +1162,17 @@ stg_newMVarzh
 }
 
 
-#define PerformTake(tso, value)                                \
-    W_[StgTSO_sp(tso) + WDS(1)] = value;               \
-    W_[StgTSO_sp(tso) + WDS(0)] = stg_gc_unpt_r1_info;
+#define PerformTake(stack, value)               \
+    W_ sp;                                      \
+    sp = StgStack_sp(stack);                    \
+    W_[sp + WDS(1)] = value;                    \
+    W_[sp + WDS(0)] = stg_gc_unpt_r1_info;
 
-#define PerformPut(tso,lval)                   \
-    StgTSO_sp(tso) = StgTSO_sp(tso) + WDS(3);  \
-    lval = W_[StgTSO_sp(tso) - WDS(1)];
+#define PerformPut(stack,lval)                  \
+    W_ sp;                                      \
+    sp = StgStack_sp(stack) + WDS(3);           \
+    StgStack_sp(stack) = sp;                    \
+    lval = W_[sp - WDS(1)];
 
 stg_takeMVarzh
 {
@@ -1219,24 +1252,20 @@ loop:
         StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
     }
 
-loop2:
-    if (TO_W_(StgTSO_what_next(tso)) == ThreadRelocated) {
-        tso = StgTSO__link(tso);
-        goto loop2;
-    }
-
     ASSERT(StgTSO_why_blocked(tso) == BlockedOnMVar::I16);
     ASSERT(StgTSO_block_info(tso) == mvar);
 
     // actually perform the putMVar for the thread that we just woke up
-    PerformPut(tso,StgMVar_value(mvar));
+    W_ stack;
+    stack = StgTSO_stackobj(tso);
+    PerformPut(stack, StgMVar_value(mvar));
 
     // indicate that the MVar operation has now completed.
     StgTSO__link(tso) = stg_END_TSO_QUEUE_closure;
     
     // no need to mark the TSO dirty, we have only written END_TSO_QUEUE.
 
-    foreign "C" tryWakeupThread_(MyCapability() "ptr", tso) [];
+    foreign "C" tryWakeupThread(MyCapability() "ptr", tso) [];
     
     unlockClosure(mvar, stg_MVAR_DIRTY_info);
     RET_P(val);
@@ -1298,27 +1327,23 @@ loop:
         StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
     }
 
-loop2:
-    if (TO_W_(StgTSO_what_next(tso)) == ThreadRelocated) {
-        tso = StgTSO__link(tso);
-        goto loop2;
-    }
-
     ASSERT(StgTSO_why_blocked(tso) == BlockedOnMVar::I16);
     ASSERT(StgTSO_block_info(tso) == mvar);
 
     // actually perform the putMVar for the thread that we just woke up
-    PerformPut(tso,StgMVar_value(mvar));
+    W_ stack;
+    stack = StgTSO_stackobj(tso);
+    PerformPut(stack, StgMVar_value(mvar));
 
     // indicate that the MVar operation has now completed.
     StgTSO__link(tso) = stg_END_TSO_QUEUE_closure;
     
     // no need to mark the TSO dirty, we have only written END_TSO_QUEUE.
 
-    foreign "C" tryWakeupThread_(MyCapability() "ptr", tso) [];
+    foreign "C" tryWakeupThread(MyCapability() "ptr", tso) [];
     
     unlockClosure(mvar, stg_MVAR_DIRTY_info);
-    RET_P(val);
+    RET_NP(1,val);
 }
 
 
@@ -1390,26 +1415,22 @@ loop:
         StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
     }
 
-loop2:
-    if (TO_W_(StgTSO_what_next(tso)) == ThreadRelocated) {
-        tso = StgTSO__link(tso);
-        goto loop2;
-    }
-
     ASSERT(StgTSO_why_blocked(tso) == BlockedOnMVar::I16);
     ASSERT(StgTSO_block_info(tso) == mvar);
 
     // actually perform the takeMVar
-    PerformTake(tso, val);
+    W_ stack;
+    stack = StgTSO_stackobj(tso);
+    PerformTake(stack, val);
 
     // indicate that the MVar operation has now completed.
     StgTSO__link(tso) = stg_END_TSO_QUEUE_closure;
-    
-    if (TO_W_(StgTSO_dirty(tso)) == 0) {
-        foreign "C" dirty_TSO(MyCapability() "ptr", tso "ptr") [];
+
+    if (TO_W_(StgStack_dirty(stack)) == 0) {
+        foreign "C" dirty_STACK(MyCapability() "ptr", stack "ptr") [];
     }
     
-    foreign "C" tryWakeupThread_(MyCapability() "ptr", tso) [];
+    foreign "C" tryWakeupThread(MyCapability() "ptr", tso) [];
 
     unlockClosure(mvar, stg_MVAR_DIRTY_info);
     jump %ENTRY_CODE(Sp(0));
@@ -1447,7 +1468,7 @@ loop:
        /* No further takes, the MVar is now full. */
        StgMVar_value(mvar) = val;
        unlockClosure(mvar, stg_MVAR_DIRTY_info);
-       jump %ENTRY_CODE(Sp(0));
+        RET_N(1);
     }
     if (StgHeader_info(q) == stg_IND_info ||
         StgHeader_info(q) == stg_MSG_NULL_info) {
@@ -1463,29 +1484,25 @@ loop:
         StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
     }
 
-loop2:
-    if (TO_W_(StgTSO_what_next(tso)) == ThreadRelocated) {
-        tso = StgTSO__link(tso);
-        goto loop2;
-    }
-
     ASSERT(StgTSO_why_blocked(tso) == BlockedOnMVar::I16);
     ASSERT(StgTSO_block_info(tso) == mvar);
 
     // actually perform the takeMVar
-    PerformTake(tso, val);
+    W_ stack;
+    stack = StgTSO_stackobj(tso);
+    PerformTake(stack, val);
 
     // indicate that the MVar operation has now completed.
     StgTSO__link(tso) = stg_END_TSO_QUEUE_closure;
     
-    if (TO_W_(StgTSO_dirty(tso)) == 0) {
-        foreign "C" dirty_TSO(MyCapability() "ptr", tso "ptr") [];
+    if (TO_W_(StgStack_dirty(stack)) == 0) {
+        foreign "C" dirty_STACK(MyCapability() "ptr", stack "ptr") [];
     }
     
-    foreign "C" tryWakeupThread_(MyCapability() "ptr", tso) [];
+    foreign "C" tryWakeupThread(MyCapability() "ptr", tso) [];
 
     unlockClosure(mvar, stg_MVAR_DIRTY_info);
-    jump %ENTRY_CODE(Sp(0));
+    RET_N(1);
 }
 
 
@@ -2009,17 +2026,28 @@ stg_getSparkzh
    W_ spark;
 
 #ifndef THREADED_RTS
-   RET_NP(0,ghczmprim_GHCziBool_False_closure);
+   RET_NP(0,ghczmprim_GHCziTypes_False_closure);
 #else
    (spark) = foreign "C" findSpark(MyCapability());
    if (spark != 0) {
       RET_NP(1,spark);
    } else {
-      RET_NP(0,ghczmprim_GHCziBool_False_closure);
+      RET_NP(0,ghczmprim_GHCziTypes_False_closure);
    }
 #endif
 }
 
+stg_numSparkszh
+{
+  W_ n;
+#ifdef THREADED_RTS
+  (n) = foreign "C" dequeElements(Capability_sparks(MyCapability()));
+#else
+  n = 0;
+#endif
+  RET_N(n);
+}
+
 stg_traceEventzh
 {
    W_ msg;
@@ -2036,7 +2064,20 @@ stg_traceEventzh
    // We should go through the macro HASKELLEVENT_USER_MSG_ENABLED from
    // RtsProbes.h, but that header file includes unistd.h, which doesn't
    // work in Cmm
+#if !defined(solaris2_TARGET_OS)
    (enabled) = foreign "C" __dtrace_isenabled$HaskellEvent$user__msg$v1() [];
+#else
+   // Solaris' DTrace can't handle the
+   //     __dtrace_isenabled$HaskellEvent$user__msg$v1
+   // call above. This call is just for testing whether the user__msg
+   // probe is enabled, and is here for just performance optimization.
+   // Since preparation for the probe is not that complex I disable usage of
+   // this test above for Solaris and enable the probe usage manually
+   // here. Please note that this does not mean that the probe will be
+   // used during the runtime! You still need to enable it by consumption
+   // in your dtrace script as you do with any other probe.
+   enabled = 1;
+#endif
    if (enabled != 0) {
      foreign "C" dtraceUserMsgWrapper(MyCapability() "ptr", msg "ptr") [];
    }