X-Git-Url: http://git.megacz.com/?a=blobdiff_plain;f=rts%2FPrimOps.cmm;h=121102c8e17f1a97dab5e546d89820bd25832166;hb=de75026f5a48d3d052135a973ab4dff76c5b20f5;hp=bfb0b8607883ee809a9c83f44d00d5b697eccdf1;hpb=98cb2efd8279ec48eee1be37dad263e16552fafb;p=ghc-hetmet.git diff --git a/rts/PrimOps.cmm b/rts/PrimOps.cmm index bfb0b86..121102c 100644 --- a/rts/PrimOps.cmm +++ b/rts/PrimOps.cmm @@ -27,6 +27,33 @@ #include "Cmm.h" +#ifdef __PIC__ +#ifndef mingw32_HOST_OS +import __gmpz_init; +import __gmpz_add; +import __gmpz_sub; +import __gmpz_mul; +import __gmpz_gcd; +import __gmpn_gcd_1; +import __gmpn_cmp; +import __gmpz_tdiv_q; +import __gmpz_tdiv_r; +import __gmpz_tdiv_qr; +import __gmpz_fdiv_qr; +import __gmpz_divexact; +import __gmpz_and; +import __gmpz_xor; +import __gmpz_ior; +import __gmpz_com; +#endif +import pthread_mutex_lock; +import pthread_mutex_unlock; +#endif +import base_ControlziExceptionziBase_nestedAtomically_closure; +import EnterCriticalSection; +import LeaveCriticalSection; +import ghczmprim_GHCziBool_False_closure; + /*----------------------------------------------------------------------------- Array Primitives @@ -49,39 +76,72 @@ newByteArrayzh_fast n = R1; payload_words = ROUNDUP_BYTES_TO_WDS(n); words = BYTES_TO_WDS(SIZEOF_StgArrWords) + payload_words; - "ptr" p = foreign "C" allocateLocal(MyCapability() "ptr",words) []; + ("ptr" p) = foreign "C" allocateLocal(MyCapability() "ptr",words) []; TICK_ALLOC_PRIM(SIZEOF_StgArrWords,WDS(payload_words),0); SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]); StgArrWords_words(p) = payload_words; RET_P(p); } +#define BA_ALIGN 16 +#define BA_MASK (BA_ALIGN-1) + newPinnedByteArrayzh_fast { - W_ words, payload_words, n, p; + W_ words, bytes, payload_words, p; MAYBE_GC(NO_PTRS,newPinnedByteArrayzh_fast); - n = R1; - payload_words = ROUNDUP_BYTES_TO_WDS(n); + bytes = R1; + /* payload_words is what we will tell the profiler we had to allocate */ + payload_words = ROUNDUP_BYTES_TO_WDS(bytes); + /* When we actually allocate memory, we need to allow space for the + header: */ + bytes = bytes + SIZEOF_StgArrWords; + /* And we want to align to BA_ALIGN bytes, so we need to allow space + to shift up to BA_ALIGN - 1 bytes: */ + bytes = bytes + BA_ALIGN - 1; + /* Now we convert to a number of words: */ + words = ROUNDUP_BYTES_TO_WDS(bytes); + + ("ptr" p) = foreign "C" allocatePinned(words) []; + TICK_ALLOC_PRIM(SIZEOF_StgArrWords,WDS(payload_words),0); - // We want an 8-byte aligned array. allocatePinned() gives us - // 8-byte aligned memory by default, but we want to align the - // *goods* inside the ArrWords object, so we have to check the - // size of the ArrWords header and adjust our size accordingly. - words = BYTES_TO_WDS(SIZEOF_StgArrWords) + payload_words; - if ((SIZEOF_StgArrWords & 7) != 0) { - words = words + 1; - } + /* Now we need to move p forward so that the payload is aligned + to BA_ALIGN bytes: */ + p = p + ((-p - SIZEOF_StgArrWords) & BA_MASK); - "ptr" p = foreign "C" allocatePinned(words) []; + SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]); + StgArrWords_words(p) = payload_words; + RET_P(p); +} + +newAlignedPinnedByteArrayzh_fast +{ + W_ words, bytes, payload_words, p, alignment; + + MAYBE_GC(NO_PTRS,newAlignedPinnedByteArrayzh_fast); + bytes = R1; + alignment = R2; + + /* payload_words is what we will tell the profiler we had to allocate */ + payload_words = ROUNDUP_BYTES_TO_WDS(bytes); + + /* When we actually allocate memory, we need to allow space for the + header: */ + bytes = bytes + SIZEOF_StgArrWords; + /* And we want to align to bytes, so we need to allow space + to shift up to bytes: */ + bytes = bytes + alignment - 1; + /* Now we convert to a number of words: */ + words = ROUNDUP_BYTES_TO_WDS(bytes); + + ("ptr" p) = foreign "C" allocatePinned(words) []; TICK_ALLOC_PRIM(SIZEOF_StgArrWords,WDS(payload_words),0); - // Again, if the ArrWords header isn't a multiple of 8 bytes, we - // have to push the object forward one word so that the goods - // fall on an 8-byte boundary. - if ((SIZEOF_StgArrWords & 7) != 0) { - p = p + WDS(1); - } + /* Now we need to move p forward so that the payload is aligned + to bytes. Note that we are assuming that + is a power of 2, which is technically not guaranteed */ + p = p + ((-p - SIZEOF_StgArrWords) & (alignment - 1)); SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]); StgArrWords_words(p) = payload_words; @@ -97,7 +157,7 @@ newArrayzh_fast MAYBE_GC(R2_PTR,newArrayzh_fast); words = BYTES_TO_WDS(SIZEOF_StgMutArrPtrs) + n; - "ptr" arr = foreign "C" allocateLocal(MyCapability() "ptr",words) [R2]; + ("ptr" arr) = foreign "C" allocateLocal(MyCapability() "ptr",words) [R2]; TICK_ALLOC_PRIM(SIZEOF_StgMutArrPtrs, WDS(n), 0); SET_HDR(arr, stg_MUT_ARR_PTRS_DIRTY_info, W_[CCCS]); @@ -167,7 +227,7 @@ newMutVarzh_fast atomicModifyMutVarzh_fast { - W_ mv, z, x, y, r; + W_ mv, f, z, x, y, r, h; /* Args: R1 :: MutVar#, R2 :: a -> (a,b) */ /* If x is the current contents of the MutVar#, then @@ -206,19 +266,15 @@ atomicModifyMutVarzh_fast HP_CHK_GEN_TICKY(SIZE, R1_PTR & R2_PTR, atomicModifyMutVarzh_fast); -#if defined(THREADED_RTS) - foreign "C" ACQUIRE_LOCK(atomic_modify_mutvar_mutex "ptr") [R1,R2]; -#endif - - x = StgMutVar_var(R1); + mv = R1; + f = R2; TICK_ALLOC_THUNK_2(); CCCS_ALLOC(THUNK_2_SIZE); z = Hp - THUNK_2_SIZE + WDS(1); SET_HDR(z, stg_ap_2_upd_info, W_[CCCS]); LDV_RECORD_CREATE(z); - StgThunk_payload(z,0) = R2; - StgThunk_payload(z,1) = x; + StgThunk_payload(z,0) = f; TICK_ALLOC_THUNK_1(); CCCS_ALLOC(THUNK_1_SIZE); @@ -227,9 +283,6 @@ atomicModifyMutVarzh_fast LDV_RECORD_CREATE(y); StgThunk_payload(y,0) = z; - StgMutVar_var(R1) = y; - foreign "C" dirty_MUT_VAR(BaseReg "ptr", R1 "ptr") [R1]; - TICK_ALLOC_THUNK_1(); CCCS_ALLOC(THUNK_1_SIZE); r = y - THUNK_1_SIZE; @@ -237,10 +290,20 @@ atomicModifyMutVarzh_fast LDV_RECORD_CREATE(r); StgThunk_payload(r,0) = z; -#if defined(THREADED_RTS) - foreign "C" RELEASE_LOCK(atomic_modify_mutvar_mutex "ptr") []; + retry: + x = StgMutVar_var(mv); + StgThunk_payload(z,1) = x; +#ifdef THREADED_RTS + (h) = foreign "C" cas(mv + SIZEOF_StgHeader + OFFSET_StgMutVar_var, x, y) []; + if (h != x) { goto retry; } +#else + StgMutVar_var(mv) = y; #endif + if (GET_INFO(mv) == stg_MUT_VAR_CLEAN_info) { + foreign "C" dirty_MUT_VAR(BaseReg "ptr", mv "ptr") []; + } + RET_P(r); } @@ -267,9 +330,14 @@ mkWeakzh_fast w = Hp - SIZEOF_StgWeak + WDS(1); SET_HDR(w, stg_WEAK_info, W_[CCCS]); - StgWeak_key(w) = R1; - StgWeak_value(w) = R2; - StgWeak_finalizer(w) = R3; + // We don't care about cfinalizer here. + // Should StgWeak_cfinalizer(w) be stg_NO_FINALIZER_closure or + // something else? + + StgWeak_key(w) = R1; + StgWeak_value(w) = R2; + StgWeak_finalizer(w) = R3; + StgWeak_cfinalizer(w) = stg_NO_FINALIZER_closure; StgWeak_link(w) = W_[weak_ptr_list]; W_[weak_ptr_list] = w; @@ -279,12 +347,65 @@ mkWeakzh_fast RET_P(w); } +mkWeakForeignEnvzh_fast +{ + /* R1 = key + R2 = value + R3 = finalizer + R4 = pointer + R5 = has environment (0 or 1) + R6 = environment + */ + W_ w, payload_words, words, p; + + W_ key, val, fptr, ptr, flag, eptr; + + key = R1; + val = R2; + fptr = R3; + ptr = R4; + flag = R5; + eptr = R6; + + ALLOC_PRIM( SIZEOF_StgWeak, R1_PTR & R2_PTR & R3_PTR, mkWeakForeignEnvzh_fast ); + + w = Hp - SIZEOF_StgWeak + WDS(1); + SET_HDR(w, stg_WEAK_info, W_[CCCS]); + + payload_words = 4; + words = BYTES_TO_WDS(SIZEOF_StgArrWords) + payload_words; + ("ptr" p) = foreign "C" allocateLocal(MyCapability() "ptr", words) []; + + TICK_ALLOC_PRIM(SIZEOF_StgArrWords,WDS(payload_words),0); + SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]); + + StgArrWords_words(p) = payload_words; + StgArrWords_payload(p,0) = fptr; + StgArrWords_payload(p,1) = ptr; + StgArrWords_payload(p,2) = eptr; + StgArrWords_payload(p,3) = flag; + + // We don't care about the value here. + // Should StgWeak_value(w) be stg_NO_FINALIZER_closure or something else? + + StgWeak_key(w) = key; + StgWeak_value(w) = val; + StgWeak_finalizer(w) = stg_NO_FINALIZER_closure; + StgWeak_cfinalizer(w) = p; + + StgWeak_link(w) = W_[weak_ptr_list]; + W_[weak_ptr_list] = w; + + IF_DEBUG(weak, foreign "C" debugBelch(stg_weak_msg,w) []); + + RET_P(w); +} finalizzeWeakzh_fast { /* R1 = weak ptr */ - W_ w, f; + W_ w, f, arr; w = R1; @@ -312,9 +433,18 @@ finalizzeWeakzh_fast SET_INFO(w,stg_DEAD_WEAK_info); LDV_RECORD_CREATE(w); - f = StgWeak_finalizer(w); + f = StgWeak_finalizer(w); + arr = StgWeak_cfinalizer(w); + StgDeadWeak_link(w) = StgWeak_link(w); + if (arr != stg_NO_FINALIZER_closure) { + foreign "C" runCFinalizer(StgArrWords_payload(arr,0), + StgArrWords_payload(arr,1), + StgArrWords_payload(arr,2), + StgArrWords_payload(arr,3)) []; + } + /* return the finalizer */ if (f == stg_NO_FINALIZER_closure) { RET_NP(0,stg_NO_FINALIZER_closure); @@ -426,11 +556,11 @@ int64ToIntegerzh_fast hi = TO_W_(val >> 32); lo = TO_W_(val); - if ( hi != 0 && hi != 0xFFFFFFFF ) { - words_needed = 2; - } else { + if ( hi == 0 || (hi == 0xFFFFFFFF && lo != 0) ) { // minimum is one word words_needed = 1; + } else { + words_needed = 2; } ALLOC_PRIM( SIZEOF_StgArrWords + WDS(words_needed), @@ -676,7 +806,7 @@ gcdIntzh_fast FETCH_MP_TEMP(mp_tmp_w); W_[mp_tmp_w] = R1; - r = foreign "C" __gmpn_gcd_1(mp_tmp_w "ptr", 1, R2) []; + (r) = foreign "C" __gmpn_gcd_1(mp_tmp_w "ptr", 1, R2) []; R1 = r; /* Result parked in R1, return via info-pointer at TOS */ @@ -687,7 +817,9 @@ gcdIntzh_fast gcdIntegerIntzh_fast { /* R1 = s1; R2 = d1; R3 = the int */ - R1 = foreign "C" __gmpn_gcd_1( BYTE_ARR_CTS(R2) "ptr", R1, R3) []; + W_ s1; + (s1) = foreign "C" __gmpn_gcd_1( BYTE_ARR_CTS(R2) "ptr", R1, R3) []; + R1 = s1; /* Result parked in R1, return via info-pointer at TOS */ jump %ENTRY_CODE(Sp(0)); @@ -768,7 +900,7 @@ cmpIntegerzh_fast up = BYTE_ARR_CTS(R2); vp = BYTE_ARR_CTS(R4); - cmp = foreign "C" __gmpn_cmp(up "ptr", vp "ptr", size) []; + (cmp) = foreign "C" __gmpn_cmp(up "ptr", vp "ptr", size) []; if (cmp == 0 :: CInt) { R1 = 0; @@ -848,6 +980,23 @@ decodeFloatzh_fast RET_NNP(W_[mp_tmp_w], TO_W_(MP_INT__mp_size(mp_tmp1)), p); } +decodeFloatzuIntzh_fast +{ + W_ p; + F_ arg; + FETCH_MP_TEMP(mp_tmp1); + FETCH_MP_TEMP(mp_tmp_w); + + /* arguments: F1 = Float# */ + arg = F1; + + /* Perform the operation */ + foreign "C" __decodeFloat_Int(mp_tmp1 "ptr", mp_tmp_w "ptr", arg) []; + + /* returns: (Int# (mantissa), Int# (exponent)) */ + RET_NN(W_[mp_tmp1], W_[mp_tmp_w]); +} + #define DOUBLE_MANTISSA_SIZE SIZEOF_DOUBLE #define ARR_SIZE (SIZEOF_StgArrWords + DOUBLE_MANTISSA_SIZE) @@ -877,6 +1026,28 @@ decodeDoublezh_fast RET_NNP(W_[mp_tmp_w], TO_W_(MP_INT__mp_size(mp_tmp1)), p); } +decodeDoublezu2Intzh_fast +{ + D_ arg; + W_ p; + FETCH_MP_TEMP(mp_tmp1); + FETCH_MP_TEMP(mp_tmp2); + FETCH_MP_TEMP(mp_result1); + FETCH_MP_TEMP(mp_result2); + + /* arguments: D1 = Double# */ + arg = D1; + + /* Perform the operation */ + foreign "C" __decodeDouble_2Int(mp_tmp1 "ptr", mp_tmp2 "ptr", + mp_result1 "ptr", mp_result2 "ptr", + arg) []; + + /* returns: + (Int# (mant sign), Word# (mant high), Word# (mant low), Int# (expn)) */ + RET_NNNN(W_[mp_tmp1], W_[mp_tmp2], W_[mp_result1], W_[mp_result2]); +} + /* ----------------------------------------------------------------------------- * Concurrency primitives * -------------------------------------------------------------------------- */ @@ -891,13 +1062,20 @@ forkzh_fast W_ threadid; closure = R1; - "ptr" threadid = foreign "C" createIOThread( MyCapability() "ptr", + ("ptr" threadid) = foreign "C" createIOThread( MyCapability() "ptr", RtsFlags_GcFlags_initialStkSize(RtsFlags), closure "ptr") []; + + /* start blocked if the current thread is blocked */ + StgTSO_flags(threadid) = + StgTSO_flags(threadid) | (StgTSO_flags(CurrentTSO) & + (TSO_BLOCKEX::I32 | TSO_INTERRUPTIBLE::I32)); + foreign "C" scheduleThread(MyCapability() "ptr", threadid "ptr") []; - // switch at the earliest opportunity - CInt[context_switch] = 1 :: CInt; + // context switch soon, but not immediately: we don't want every + // forkIO to force a context-switch. + Capability_context_switch(MyCapability()) = 1 :: CInt; RET_P(threadid); } @@ -914,13 +1092,20 @@ forkOnzh_fast cpu = R1; closure = R2; - "ptr" threadid = foreign "C" createIOThread( MyCapability() "ptr", + ("ptr" threadid) = foreign "C" createIOThread( MyCapability() "ptr", RtsFlags_GcFlags_initialStkSize(RtsFlags), closure "ptr") []; + + /* start blocked if the current thread is blocked */ + StgTSO_flags(threadid) = + StgTSO_flags(threadid) | (StgTSO_flags(CurrentTSO) & + (TSO_BLOCKEX::I32 | TSO_INTERRUPTIBLE::I32)); + foreign "C" scheduleThreadOn(MyCapability() "ptr", cpu, threadid "ptr") []; - // switch at the earliest opportunity - CInt[context_switch] = 1 :: CInt; + // context switch soon, but not immediately: we don't want every + // forkIO to force a context-switch. + Capability_context_switch(MyCapability()) = 1 :: CInt; RET_P(threadid); } @@ -951,78 +1136,73 @@ isCurrentThreadBoundzh_fast { /* no args */ W_ r; - r = foreign "C" isThreadBound(CurrentTSO) []; + (r) = foreign "C" isThreadBound(CurrentTSO) []; RET_N(r); } +threadStatuszh_fast +{ + /* args: R1 :: ThreadId# */ + W_ tso; + W_ why_blocked; + W_ what_next; + W_ ret; + + 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)); + // Note: these two reads are not atomic, so they might end up + // being inconsistent. It doesn't matter, since we + // only return one or the other. If we wanted to return the + // contents of block_info too, then we'd have to do some synchronisation. + + if (what_next == ThreadComplete) { + ret = 16; // NB. magic, matches up with GHC.Conc.threadStatus + } else { + if (what_next == ThreadKilled) { + ret = 17; + } else { + ret = why_blocked; + } + } + RET_N(ret); +} /* ----------------------------------------------------------------------------- * TVar primitives * -------------------------------------------------------------------------- */ -#ifdef REG_R1 #define SP_OFF 0 -#define IF_NOT_REG_R1(x) -#else -#define SP_OFF 1 -#define IF_NOT_REG_R1(x) x -#endif // Catch retry frame ------------------------------------------------------------ -#define CATCH_RETRY_FRAME_ERROR(label) \ - label { foreign "C" barf("catch_retry_frame incorrectly entered!"); } - -CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_0_ret) -CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_1_ret) -CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_2_ret) -CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_3_ret) -CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_4_ret) -CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_5_ret) -CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_6_ret) -CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_7_ret) - -#if MAX_VECTORED_RTN > 8 -#error MAX_VECTORED_RTN has changed: please modify stg_catch_retry_frame too. -#endif - +INFO_TABLE_RET(stg_catch_retry_frame, CATCH_RETRY_FRAME, #if defined(PROFILING) -#define CATCH_RETRY_FRAME_BITMAP 7 -#define CATCH_RETRY_FRAME_WORDS 5 -#else -#define CATCH_RETRY_FRAME_BITMAP 1 -#define CATCH_RETRY_FRAME_WORDS 3 + W_ unused1, W_ unused2, #endif - -INFO_TABLE_RET(stg_catch_retry_frame, - CATCH_RETRY_FRAME_WORDS, CATCH_RETRY_FRAME_BITMAP, - CATCH_RETRY_FRAME, - stg_catch_retry_frame_0_ret, - stg_catch_retry_frame_1_ret, - stg_catch_retry_frame_2_ret, - stg_catch_retry_frame_3_ret, - stg_catch_retry_frame_4_ret, - stg_catch_retry_frame_5_ret, - stg_catch_retry_frame_6_ret, - stg_catch_retry_frame_7_ret) + W_ unused3, P_ unused4, P_ unused5) { W_ r, frame, trec, outer; - IF_NOT_REG_R1(W_ rval; rval = Sp(0); Sp_adj(1); ) frame = Sp; trec = StgTSO_trec(CurrentTSO); - "ptr" outer = foreign "C" stmGetEnclosingTRec(trec "ptr") []; - r = foreign "C" stmCommitNestedTransaction(MyCapability() "ptr", trec "ptr") []; + ("ptr" outer) = foreign "C" stmGetEnclosingTRec(trec "ptr") []; + (r) = foreign "C" stmCommitNestedTransaction(MyCapability() "ptr", trec "ptr") []; if (r != 0) { /* Succeeded (either first branch or second branch) */ StgTSO_trec(CurrentTSO) = outer; Sp = Sp + SIZEOF_StgCatchRetryFrame; - IF_NOT_REG_R1(Sp_adj(-1); Sp(0) = rval;) jump %ENTRY_CODE(Sp(SP_OFF)); } else { /* Did not commit: re-execute */ W_ new_trec; - "ptr" new_trec = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") []; + ("ptr" new_trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") []; StgTSO_trec(CurrentTSO) = new_trec; if (StgCatchRetryFrame_running_alt_code(frame) != 0::I32) { R1 = StgCatchRetryFrame_alt_code(frame); @@ -1034,56 +1214,23 @@ INFO_TABLE_RET(stg_catch_retry_frame, } -// Atomically frame ------------------------------------------------------------- - - -#define ATOMICALLY_FRAME_ERROR(label) \ - label { foreign "C" barf("atomically_frame incorrectly entered!"); } - -ATOMICALLY_FRAME_ERROR(stg_atomically_frame_0_ret) -ATOMICALLY_FRAME_ERROR(stg_atomically_frame_1_ret) -ATOMICALLY_FRAME_ERROR(stg_atomically_frame_2_ret) -ATOMICALLY_FRAME_ERROR(stg_atomically_frame_3_ret) -ATOMICALLY_FRAME_ERROR(stg_atomically_frame_4_ret) -ATOMICALLY_FRAME_ERROR(stg_atomically_frame_5_ret) -ATOMICALLY_FRAME_ERROR(stg_atomically_frame_6_ret) -ATOMICALLY_FRAME_ERROR(stg_atomically_frame_7_ret) - -#if MAX_VECTORED_RTN > 8 -#error MAX_VECTORED_RTN has changed: please modify stg_atomically_frame too. -#endif +// Atomically frame ------------------------------------------------------------ +INFO_TABLE_RET(stg_atomically_frame, ATOMICALLY_FRAME, #if defined(PROFILING) -#define ATOMICALLY_FRAME_BITMAP 3 -#define ATOMICALLY_FRAME_WORDS 4 -#else -#define ATOMICALLY_FRAME_BITMAP 0 -#define ATOMICALLY_FRAME_WORDS 2 + W_ unused1, W_ unused2, #endif - - -INFO_TABLE_RET(stg_atomically_frame, - ATOMICALLY_FRAME_WORDS, ATOMICALLY_FRAME_BITMAP, - ATOMICALLY_FRAME, - stg_atomically_frame_0_ret, - stg_atomically_frame_1_ret, - stg_atomically_frame_2_ret, - stg_atomically_frame_3_ret, - stg_atomically_frame_4_ret, - stg_atomically_frame_5_ret, - stg_atomically_frame_6_ret, - stg_atomically_frame_7_ret) + P_ unused3, P_ unused4) { W_ frame, trec, valid, next_invariant, q, outer; - IF_NOT_REG_R1(W_ rval; rval = Sp(0); Sp_adj(1); ) frame = Sp; trec = StgTSO_trec(CurrentTSO); - "ptr" outer = foreign "C" stmGetEnclosingTRec(trec "ptr") []; + ("ptr" outer) = foreign "C" stmGetEnclosingTRec(trec "ptr") []; if (outer == NO_TREC) { /* First time back at the atomically frame -- pick up invariants */ - "ptr" q = foreign "C" stmGetInvariantsToCheck(MyCapability() "ptr", trec "ptr") []; + ("ptr" q) = foreign "C" stmGetInvariantsToCheck(MyCapability() "ptr", trec "ptr") []; StgAtomicallyFrame_next_invariant_to_check(frame) = q; } else { @@ -1104,7 +1251,7 @@ INFO_TABLE_RET(stg_atomically_frame, if (q != END_INVARIANT_CHECK_QUEUE) { /* We can't commit yet: another invariant to check */ - "ptr" trec = foreign "C" stmStartTransaction(MyCapability() "ptr", trec "ptr") []; + ("ptr" trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", trec "ptr") []; StgTSO_trec(CurrentTSO) = trec; next_invariant = StgInvariantCheckQueue_invariant(q); @@ -1114,16 +1261,15 @@ INFO_TABLE_RET(stg_atomically_frame, } else { /* We've got no more invariants to check, try to commit */ - valid = foreign "C" stmCommitTransaction(MyCapability() "ptr", trec "ptr") []; + (valid) = foreign "C" stmCommitTransaction(MyCapability() "ptr", trec "ptr") []; if (valid != 0) { /* Transaction was valid: commit succeeded */ StgTSO_trec(CurrentTSO) = NO_TREC; Sp = Sp + SIZEOF_StgAtomicallyFrame; - IF_NOT_REG_R1(Sp_adj(-1); Sp(0) = rval;) jump %ENTRY_CODE(Sp(SP_OFF)); } else { /* Transaction was not valid: try again */ - "ptr" trec = foreign "C" stmStartTransaction(MyCapability() "ptr", NO_TREC "ptr") []; + ("ptr" trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", NO_TREC "ptr") []; StgTSO_trec(CurrentTSO) = trec; StgAtomicallyFrame_next_invariant_to_check(frame) = END_INVARIANT_CHECK_QUEUE; R1 = StgAtomicallyFrame_code(frame); @@ -1132,34 +1278,24 @@ INFO_TABLE_RET(stg_atomically_frame, } } -INFO_TABLE_RET(stg_atomically_waiting_frame, - ATOMICALLY_FRAME_WORDS, ATOMICALLY_FRAME_BITMAP, - ATOMICALLY_FRAME, - stg_atomically_frame_0_ret, - stg_atomically_frame_1_ret, - stg_atomically_frame_2_ret, - stg_atomically_frame_3_ret, - stg_atomically_frame_4_ret, - stg_atomically_frame_5_ret, - stg_atomically_frame_6_ret, - stg_atomically_frame_7_ret) +INFO_TABLE_RET(stg_atomically_waiting_frame, ATOMICALLY_FRAME, +#if defined(PROFILING) + W_ unused1, W_ unused2, +#endif + P_ unused3, P_ unused4) { W_ frame, trec, valid; - IF_NOT_REG_R1(W_ rval; rval = Sp(0); Sp_adj(1); ) frame = Sp; /* The TSO is currently waiting: should we stop waiting? */ - valid = foreign "C" stmReWait(MyCapability() "ptr", CurrentTSO "ptr") []; + (valid) = foreign "C" stmReWait(MyCapability() "ptr", CurrentTSO "ptr") []; if (valid != 0) { /* Previous attempt is still valid: no point trying again yet */ - IF_NOT_REG_R1(Sp_adj(-2); - Sp(1) = stg_NO_FINALIZER_closure; - Sp(0) = stg_ut_1_0_unreg_info;) jump stg_block_noregs; } else { /* Previous attempt is no longer valid: try again */ - "ptr" trec = foreign "C" stmStartTransaction(MyCapability() "ptr", NO_TREC "ptr") []; + ("ptr" trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", NO_TREC "ptr") []; StgTSO_trec(CurrentTSO) = trec; StgHeader_info(frame) = stg_atomically_frame_info; R1 = StgAtomicallyFrame_code(frame); @@ -1169,75 +1305,38 @@ INFO_TABLE_RET(stg_atomically_waiting_frame, // STM catch frame -------------------------------------------------------------- -#define CATCH_STM_FRAME_ENTRY_TEMPLATE(label,ret) \ - label \ - { \ - IF_NOT_REG_R1(W_ rval; rval = Sp(0); Sp_adj(1); ) \ - W_ r, frame, trec, outer; \ - frame = Sp; \ - trec = StgTSO_trec(CurrentTSO); \ - "ptr" outer = foreign "C" stmGetEnclosingTRec(trec "ptr") []; \ - r = foreign "C" stmCommitNestedTransaction(MyCapability() "ptr", trec "ptr") []; \ - if (r != 0) { \ - /* Commit succeeded */ \ - StgTSO_trec(CurrentTSO) = outer; \ - Sp = Sp + SIZEOF_StgCatchSTMFrame; \ - IF_NOT_REG_R1(Sp_adj(-1); Sp(0) = rval;) \ - jump ret; \ - } else { \ - /* Commit failed */ \ - W_ new_trec; \ - "ptr" new_trec = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") []; \ - StgTSO_trec(CurrentTSO) = new_trec; \ - R1 = StgCatchSTMFrame_code(frame); \ - jump stg_ap_v_fast; \ - } \ - } - -#ifdef REG_R1 #define SP_OFF 0 -#else -#define SP_OFF 1 -#endif - -CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_0_ret,%RET_VEC(Sp(SP_OFF),0)) -CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_1_ret,%RET_VEC(Sp(SP_OFF),1)) -CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_2_ret,%RET_VEC(Sp(SP_OFF),2)) -CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_3_ret,%RET_VEC(Sp(SP_OFF),3)) -CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_4_ret,%RET_VEC(Sp(SP_OFF),4)) -CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_5_ret,%RET_VEC(Sp(SP_OFF),5)) -CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_6_ret,%RET_VEC(Sp(SP_OFF),6)) -CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_7_ret,%RET_VEC(Sp(SP_OFF),7)) - -#if MAX_VECTORED_RTN > 8 -#error MAX_VECTORED_RTN has changed: please modify stg_catch_stm_frame too. -#endif - -#if defined(PROFILING) -#define CATCH_STM_FRAME_BITMAP 3 -#define CATCH_STM_FRAME_WORDS 4 -#else -#define CATCH_STM_FRAME_BITMAP 0 -#define CATCH_STM_FRAME_WORDS 2 -#endif /* Catch frames are very similar to update frames, but when entering * one we just pop the frame off the stack and perform the correct * kind of return to the activation record underneath us on the stack. */ -INFO_TABLE_RET(stg_catch_stm_frame, - CATCH_STM_FRAME_WORDS, CATCH_STM_FRAME_BITMAP, - CATCH_STM_FRAME, - stg_catch_stm_frame_0_ret, - stg_catch_stm_frame_1_ret, - stg_catch_stm_frame_2_ret, - stg_catch_stm_frame_3_ret, - stg_catch_stm_frame_4_ret, - stg_catch_stm_frame_5_ret, - stg_catch_stm_frame_6_ret, - stg_catch_stm_frame_7_ret) -CATCH_STM_FRAME_ENTRY_TEMPLATE(,%ENTRY_CODE(Sp(SP_OFF))) +INFO_TABLE_RET(stg_catch_stm_frame, CATCH_STM_FRAME, +#if defined(PROFILING) + W_ unused1, W_ unused2, +#endif + P_ unused3, P_ unused4) + { + W_ r, frame, trec, outer; + frame = Sp; + trec = StgTSO_trec(CurrentTSO); + ("ptr" outer) = foreign "C" stmGetEnclosingTRec(trec "ptr") []; + (r) = foreign "C" stmCommitNestedTransaction(MyCapability() "ptr", trec "ptr") []; + if (r != 0) { + /* Commit succeeded */ + StgTSO_trec(CurrentTSO) = outer; + Sp = Sp + SIZEOF_StgCatchSTMFrame; + jump Sp(SP_OFF); + } else { + /* Commit failed */ + W_ new_trec; + ("ptr" new_trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") []; + StgTSO_trec(CurrentTSO) = new_trec; + R1 = StgCatchSTMFrame_code(frame); + jump stg_ap_v_fast; + } + } // Primop definition ------------------------------------------------------------ @@ -1258,7 +1357,7 @@ atomicallyzh_fast /* Nested transactions are not allowed; raise an exception */ if (old_trec != NO_TREC) { - R1 = base_GHCziIOBase_NestedAtomically_closure; + R1 = base_ControlziExceptionziBase_nestedAtomically_closure; jump raisezh_fast; } @@ -1271,7 +1370,7 @@ atomicallyzh_fast StgAtomicallyFrame_next_invariant_to_check(frame) = END_INVARIANT_CHECK_QUEUE; /* Start the memory transcation */ - "ptr" new_trec = foreign "C" stmStartTransaction(MyCapability() "ptr", old_trec "ptr") [R1]; + ("ptr" new_trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", old_trec "ptr") [R1]; StgTSO_trec(CurrentTSO) = new_trec; /* Apply R1 to the realworld token */ @@ -1299,7 +1398,7 @@ catchSTMzh_fast W_ cur_trec; W_ new_trec; cur_trec = StgTSO_trec(CurrentTSO); - "ptr" new_trec = foreign "C" stmStartTransaction(MyCapability() "ptr", cur_trec "ptr"); + ("ptr" new_trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", cur_trec "ptr"); StgTSO_trec(CurrentTSO) = new_trec; /* Apply R1 to the realworld token */ @@ -1322,7 +1421,7 @@ catchRetryzh_fast /* Start a nested transaction within which to run the first code */ trec = StgTSO_trec(CurrentTSO); - "ptr" new_trec = foreign "C" stmStartTransaction(MyCapability() "ptr", trec "ptr") [R1,R2]; + ("ptr" new_trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", trec "ptr") [R1,R2]; StgTSO_trec(CurrentTSO) = new_trec; /* Set up the catch-retry frame */ @@ -1352,11 +1451,11 @@ retryzh_fast // Find the enclosing ATOMICALLY_FRAME or CATCH_RETRY_FRAME retry_pop_stack: StgTSO_sp(CurrentTSO) = Sp; - frame_type = foreign "C" findRetryFrameHelper(CurrentTSO "ptr") []; + (frame_type) = foreign "C" findRetryFrameHelper(CurrentTSO "ptr") []; Sp = StgTSO_sp(CurrentTSO); frame = Sp; trec = StgTSO_trec(CurrentTSO); - "ptr" outer = foreign "C" stmGetEnclosingTRec(trec "ptr") []; + ("ptr" outer) = foreign "C" stmGetEnclosingTRec(trec "ptr") []; if (frame_type == CATCH_RETRY_FRAME) { // The retry reaches a CATCH_RETRY_FRAME before the atomic frame @@ -1366,7 +1465,7 @@ retry_pop_stack: foreign "C" stmFreeAbortedTRec(MyCapability() "ptr", trec "ptr") []; if (!StgCatchRetryFrame_running_alt_code(frame) != 0::I32) { // Retry in the first branch: try the alternative - "ptr" trec = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") []; + ("ptr" trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") []; StgTSO_trec(CurrentTSO) = trec; StgCatchRetryFrame_running_alt_code(frame) = 1 :: CInt; // true; R1 = StgCatchRetryFrame_alt_code(frame); @@ -1388,25 +1487,22 @@ retry_pop_stack: foreign "C" stmAbortTransaction(MyCapability() "ptr", trec "ptr") []; foreign "C" stmFreeAbortedTRec(MyCapability() "ptr", trec "ptr") []; trec = outer; - StgTSO_trec(CurrentTSO) = trec; - "ptr" outer = foreign "C" stmGetEnclosingTRec(trec "ptr") []; + StgTSO_trec(CurrentTSO) = trec; + ("ptr" outer) = foreign "C" stmGetEnclosingTRec(trec "ptr") []; } ASSERT(outer == NO_TREC); - r = foreign "C" stmWait(MyCapability() "ptr", CurrentTSO "ptr", trec "ptr") []; + (r) = foreign "C" stmWait(MyCapability() "ptr", CurrentTSO "ptr", trec "ptr") []; if (r != 0) { // Transaction was valid: stmWait put us on the TVars' queues, we now block StgHeader_info(frame) = stg_atomically_waiting_frame_info; Sp = frame; // Fix up the stack in the unregisterised case: the return convention is different. - IF_NOT_REG_R1(Sp_adj(-2); - Sp(1) = stg_NO_FINALIZER_closure; - Sp(0) = stg_ut_1_0_unreg_info;) R3 = trec; // passing to stmWaitUnblock() jump stg_block_stmwait; } else { // Transaction was not valid: retry immediately - "ptr" trec = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") []; + ("ptr" trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") []; StgTSO_trec(CurrentTSO) = trec; R1 = StgAtomicallyFrame_code(frame); Sp = frame; @@ -1441,7 +1537,7 @@ newTVarzh_fast MAYBE_GC (R1_PTR, newTVarzh_fast); new_value = R1; - "ptr" tv = foreign "C" stmNewTVar(MyCapability() "ptr", new_value "ptr") []; + ("ptr" tv) = foreign "C" stmNewTVar(MyCapability() "ptr", new_value "ptr") []; RET_P(tv); } @@ -1457,11 +1553,22 @@ readTVarzh_fast MAYBE_GC (R1_PTR, readTVarzh_fast); // Call to stmReadTVar may allocate trec = StgTSO_trec(CurrentTSO); tvar = R1; - "ptr" result = foreign "C" stmReadTVar(MyCapability() "ptr", trec "ptr", tvar "ptr") []; + ("ptr" result) = foreign "C" stmReadTVar(MyCapability() "ptr", trec "ptr", tvar "ptr") []; RET_P(result); } +readTVarIOzh_fast +{ + W_ result; + +again: + result = StgTVar_current_value(R1); + if (%INFO_PTR(result) == stg_TREC_HEADER_info) { + goto again; + } + RET_P(result); +} writeTVarzh_fast { @@ -1518,7 +1625,7 @@ isEmptyMVarzh_fast { /* args: R1 = MVar closure */ - if (GET_INFO(R1) == stg_EMPTY_MVAR_info) { + if (StgMVar_value(R1) == stg_END_TSO_QUEUE_closure) { RET_N(1); } else { RET_N(0); @@ -1533,7 +1640,8 @@ newMVarzh_fast ALLOC_PRIM ( SIZEOF_StgMVar, NO_PTRS, newMVarzh_fast ); mvar = Hp - SIZEOF_StgMVar + WDS(1); - SET_HDR(mvar,stg_EMPTY_MVAR_info,W_[CCCS]); + SET_HDR(mvar,stg_MVAR_DIRTY_info,W_[CCCS]); + // MVARs start dirty: generation 0 has no mutable list StgMVar_head(mvar) = stg_END_TSO_QUEUE_closure; StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure; StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure; @@ -1541,16 +1649,9 @@ newMVarzh_fast } -/* If R1 isn't available, pass it on the stack */ -#ifdef REG_R1 #define PerformTake(tso, value) \ W_[StgTSO_sp(tso) + WDS(1)] = value; \ W_[StgTSO_sp(tso) + WDS(0)] = stg_gc_unpt_r1_info; -#else -#define PerformTake(tso, value) \ - W_[StgTSO_sp(tso) + WDS(1)] = value; \ - W_[StgTSO_sp(tso) + WDS(0)] = stg_ut_1_0_unreg_info; -#endif #define PerformPut(tso,lval) \ StgTSO_sp(tso) = StgTSO_sp(tso) + WDS(3); \ @@ -1564,25 +1665,35 @@ takeMVarzh_fast mvar = R1; #if defined(THREADED_RTS) - "ptr" info = foreign "C" lockClosure(mvar "ptr") []; + ("ptr" info) = foreign "C" lockClosure(mvar "ptr") []; #else info = GET_INFO(mvar); #endif + + if (info == stg_MVAR_CLEAN_info) { + foreign "C" dirty_MVAR(BaseReg "ptr", mvar "ptr") []; + } /* If the MVar is empty, put ourselves on its blocking queue, * and wait until we're woken up. */ - if (info == stg_EMPTY_MVAR_info) { + if (StgMVar_value(mvar) == stg_END_TSO_QUEUE_closure) { if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) { StgMVar_head(mvar) = CurrentTSO; } else { - StgTSO_link(StgMVar_tail(mvar)) = CurrentTSO; + foreign "C" setTSOLink(MyCapability() "ptr", + StgMVar_tail(mvar) "ptr", + CurrentTSO) []; } - StgTSO_link(CurrentTSO) = stg_END_TSO_QUEUE_closure; - StgTSO_why_blocked(CurrentTSO) = BlockedOnMVar::I16; + StgTSO__link(CurrentTSO) = stg_END_TSO_QUEUE_closure; StgTSO_block_info(CurrentTSO) = mvar; + // write barrier for throwTo(), which looks at block_info + // if why_blocked==BlockedOnMVar. + prim %write_barrier() []; + StgTSO_why_blocked(CurrentTSO) = BlockedOnMVar::I16; StgMVar_tail(mvar) = CurrentTSO; + R1 = mvar; jump stg_block_takemvar; } @@ -1599,24 +1710,23 @@ takeMVarzh_fast /* actually perform the putMVar for the thread that we just woke up */ tso = StgMVar_head(mvar); PerformPut(tso,StgMVar_value(mvar)); - dirtyTSO(tso); -#if defined(GRAN) || defined(PAR) - /* ToDo: check 2nd arg (mvar) is right */ - "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar),mvar) []; - StgMVar_head(mvar) = tso; -#else - "ptr" tso = foreign "C" unblockOne(MyCapability() "ptr", - StgMVar_head(mvar) "ptr") []; + if (TO_W_(StgTSO_flags(tso)) & TSO_DIRTY == 0) { + foreign "C" dirty_TSO(MyCapability() "ptr", tso "ptr") []; + } + + ("ptr" tso) = foreign "C" unblockOne_(MyCapability() "ptr", + StgMVar_head(mvar) "ptr", 1) []; StgMVar_head(mvar) = tso; -#endif if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) { StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure; } #if defined(THREADED_RTS) - foreign "C" unlockClosure(mvar "ptr", stg_FULL_MVAR_info) []; + unlockClosure(mvar, stg_MVAR_DIRTY_info); +#else + SET_INFO(mvar,stg_MVAR_DIRTY_info); #endif RET_P(val); } @@ -1626,9 +1736,9 @@ takeMVarzh_fast StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure; #if defined(THREADED_RTS) - foreign "C" unlockClosure(mvar "ptr", stg_EMPTY_MVAR_info) []; + unlockClosure(mvar, stg_MVAR_DIRTY_info); #else - SET_INFO(mvar,stg_EMPTY_MVAR_info); + SET_INFO(mvar,stg_MVAR_DIRTY_info); #endif RET_P(val); @@ -1645,14 +1755,14 @@ tryTakeMVarzh_fast mvar = R1; #if defined(THREADED_RTS) - "ptr" info = foreign "C" lockClosure(mvar "ptr") []; + ("ptr" info) = foreign "C" lockClosure(mvar "ptr") []; #else info = GET_INFO(mvar); #endif - if (info == stg_EMPTY_MVAR_info) { + if (StgMVar_value(mvar) == stg_END_TSO_QUEUE_closure) { #if defined(THREADED_RTS) - foreign "C" unlockClosure(mvar "ptr", stg_EMPTY_MVAR_info) []; + unlockClosure(mvar, info); #endif /* HACK: we need a pointer to pass back, * so we abuse NO_FINALIZER_closure @@ -1660,6 +1770,10 @@ tryTakeMVarzh_fast RET_NP(0, stg_NO_FINALIZER_closure); } + if (info == stg_MVAR_CLEAN_info) { + foreign "C" dirty_MVAR(BaseReg "ptr", mvar "ptr"); + } + /* we got the value... */ val = StgMVar_value(mvar); @@ -1673,23 +1787,21 @@ tryTakeMVarzh_fast /* actually perform the putMVar for the thread that we just woke up */ tso = StgMVar_head(mvar); PerformPut(tso,StgMVar_value(mvar)); - dirtyTSO(tso); + if (TO_W_(StgTSO_flags(tso)) & TSO_DIRTY == 0) { + foreign "C" dirty_TSO(MyCapability() "ptr", tso "ptr") []; + } -#if defined(GRAN) || defined(PAR) - /* ToDo: check 2nd arg (mvar) is right */ - "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr", mvar "ptr") []; - StgMVar_head(mvar) = tso; -#else - "ptr" tso = foreign "C" unblockOne(MyCapability() "ptr", - StgMVar_head(mvar) "ptr") []; + ("ptr" tso) = foreign "C" unblockOne_(MyCapability() "ptr", + StgMVar_head(mvar) "ptr", 1) []; StgMVar_head(mvar) = tso; -#endif if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) { StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure; } #if defined(THREADED_RTS) - foreign "C" unlockClosure(mvar "ptr", stg_FULL_MVAR_info) []; + unlockClosure(mvar, stg_MVAR_DIRTY_info); +#else + SET_INFO(mvar,stg_MVAR_DIRTY_info); #endif } else @@ -1697,9 +1809,9 @@ tryTakeMVarzh_fast /* No further putMVars, MVar is now empty */ StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure; #if defined(THREADED_RTS) - foreign "C" unlockClosure(mvar "ptr", stg_EMPTY_MVAR_info) []; + unlockClosure(mvar, stg_MVAR_DIRTY_info); #else - SET_INFO(mvar,stg_EMPTY_MVAR_info); + SET_INFO(mvar,stg_MVAR_DIRTY_info); #endif } @@ -1709,28 +1821,40 @@ tryTakeMVarzh_fast putMVarzh_fast { - W_ mvar, info, tso; + W_ mvar, val, info, tso; /* args: R1 = MVar, R2 = value */ mvar = R1; + val = R2; #if defined(THREADED_RTS) - "ptr" info = foreign "C" lockClosure(mvar "ptr") [R2]; + ("ptr" info) = foreign "C" lockClosure(mvar "ptr") []; #else info = GET_INFO(mvar); #endif - if (info == stg_FULL_MVAR_info) { + if (info == stg_MVAR_CLEAN_info) { + foreign "C" dirty_MVAR(BaseReg "ptr", mvar "ptr"); + } + + if (StgMVar_value(mvar) != stg_END_TSO_QUEUE_closure) { if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) { StgMVar_head(mvar) = CurrentTSO; } else { - StgTSO_link(StgMVar_tail(mvar)) = CurrentTSO; + foreign "C" setTSOLink(MyCapability() "ptr", + StgMVar_tail(mvar) "ptr", + CurrentTSO) []; } - StgTSO_link(CurrentTSO) = stg_END_TSO_QUEUE_closure; - StgTSO_why_blocked(CurrentTSO) = BlockedOnMVar::I16; + StgTSO__link(CurrentTSO) = stg_END_TSO_QUEUE_closure; StgTSO_block_info(CurrentTSO) = mvar; + // write barrier for throwTo(), which looks at block_info + // if why_blocked==BlockedOnMVar. + prim %write_barrier() []; + StgTSO_why_blocked(CurrentTSO) = BlockedOnMVar::I16; StgMVar_tail(mvar) = CurrentTSO; + R1 = mvar; + R2 = val; jump stg_block_putmvar; } @@ -1742,36 +1866,35 @@ putMVarzh_fast /* actually perform the takeMVar */ tso = StgMVar_head(mvar); - PerformTake(tso, R2); - dirtyTSO(tso); + PerformTake(tso, val); + if (TO_W_(StgTSO_flags(tso)) & TSO_DIRTY == 0) { + foreign "C" dirty_TSO(MyCapability() "ptr", tso "ptr") []; + } -#if defined(GRAN) || defined(PAR) - /* ToDo: check 2nd arg (mvar) is right */ - "ptr" tso = foreign "C" unblockOne(MyCapability() "ptr", StgMVar_head(mvar) "ptr",mvar "ptr") []; + ("ptr" tso) = foreign "C" unblockOne_(MyCapability() "ptr", + StgMVar_head(mvar) "ptr", 1) []; StgMVar_head(mvar) = tso; -#else - "ptr" tso = foreign "C" unblockOne(MyCapability() "ptr", StgMVar_head(mvar) "ptr") []; - StgMVar_head(mvar) = tso; -#endif if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) { StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure; } #if defined(THREADED_RTS) - foreign "C" unlockClosure(mvar "ptr", stg_EMPTY_MVAR_info) []; + unlockClosure(mvar, stg_MVAR_DIRTY_info); +#else + SET_INFO(mvar,stg_MVAR_DIRTY_info); #endif jump %ENTRY_CODE(Sp(0)); } else { /* No further takes, the MVar is now full. */ - StgMVar_value(mvar) = R2; + StgMVar_value(mvar) = val; #if defined(THREADED_RTS) - foreign "C" unlockClosure(mvar "ptr", stg_FULL_MVAR_info) []; + unlockClosure(mvar, stg_MVAR_DIRTY_info); #else - SET_INFO(mvar,stg_FULL_MVAR_info); + SET_INFO(mvar,stg_MVAR_DIRTY_info); #endif jump %ENTRY_CODE(Sp(0)); } @@ -1788,18 +1911,22 @@ tryPutMVarzh_fast mvar = R1; #if defined(THREADED_RTS) - "ptr" info = foreign "C" lockClosure(mvar "ptr") [R2]; + ("ptr" info) = foreign "C" lockClosure(mvar "ptr") [R2]; #else info = GET_INFO(mvar); #endif - if (info == stg_FULL_MVAR_info) { + if (StgMVar_value(mvar) != stg_END_TSO_QUEUE_closure) { #if defined(THREADED_RTS) - foreign "C" unlockClosure(mvar "ptr", stg_FULL_MVAR_info) []; + unlockClosure(mvar, info); #endif RET_N(0); } + if (info == stg_MVAR_CLEAN_info) { + foreign "C" dirty_MVAR(BaseReg "ptr", mvar "ptr"); + } + if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure) { /* There are takeMVar(s) waiting: wake up the first one @@ -1809,23 +1936,22 @@ tryPutMVarzh_fast /* actually perform the takeMVar */ tso = StgMVar_head(mvar); PerformTake(tso, R2); - dirtyTSO(tso); + if (TO_W_(StgTSO_flags(tso)) & TSO_DIRTY == 0) { + foreign "C" dirty_TSO(MyCapability() "ptr", tso "ptr") []; + } -#if defined(GRAN) || defined(PAR) - /* ToDo: check 2nd arg (mvar) is right */ - "ptr" tso = foreign "C" unblockOne(MyCapability() "ptr", StgMVar_head(mvar) "ptr",mvar "ptr") []; + ("ptr" tso) = foreign "C" unblockOne_(MyCapability() "ptr", + StgMVar_head(mvar) "ptr", 1) []; StgMVar_head(mvar) = tso; -#else - "ptr" tso = foreign "C" unblockOne(MyCapability() "ptr", StgMVar_head(mvar) "ptr") []; - StgMVar_head(mvar) = tso; -#endif if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) { StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure; } #if defined(THREADED_RTS) - foreign "C" unlockClosure(mvar "ptr", stg_EMPTY_MVAR_info) []; + unlockClosure(mvar, stg_MVAR_DIRTY_info); +#else + SET_INFO(mvar,stg_MVAR_DIRTY_info); #endif } else @@ -1834,9 +1960,9 @@ tryPutMVarzh_fast StgMVar_value(mvar) = R2; #if defined(THREADED_RTS) - foreign "C" unlockClosure(mvar "ptr", stg_FULL_MVAR_info) []; + unlockClosure(mvar, stg_MVAR_DIRTY_info); #else - SET_INFO(mvar,stg_FULL_MVAR_info); + SET_INFO(mvar,stg_MVAR_DIRTY_info); #endif } @@ -1855,7 +1981,7 @@ makeStableNamezh_fast ALLOC_PRIM( SIZEOF_StgStableName, R1_PTR, makeStableNamezh_fast ); - index = foreign "C" lookupStableName(R1 "ptr") []; + (index) = foreign "C" lookupStableName(R1 "ptr") []; /* Is there already a StableName for this heap object? * stable_ptr_table is a pointer to an array of snEntry structs. @@ -1878,7 +2004,7 @@ makeStablePtrzh_fast /* Args: R1 = a */ W_ sp; MAYBE_GC(R1_PTR, makeStablePtrzh_fast); - "ptr" sp = foreign "C" getStablePtr(R1 "ptr") []; + ("ptr" sp) = foreign "C" getStablePtr(R1 "ptr") []; RET_N(sp); } @@ -1900,17 +2026,17 @@ newBCOzh_fast /* R1 = instrs R2 = literals R3 = ptrs - R4 = itbls - R5 = arity - R6 = bitmap array + R4 = arity + R5 = bitmap array */ W_ bco, bitmap_arr, bytes, words; - bitmap_arr = R6; + bitmap_arr = R5; + words = BYTES_TO_WDS(SIZEOF_StgBCO) + StgArrWords_words(bitmap_arr); bytes = WDS(words); - ALLOC_PRIM( bytes, R1_PTR&R2_PTR&R3_PTR&R4_PTR&R6_PTR, newBCOzh_fast ); + ALLOC_PRIM( bytes, R1_PTR&R2_PTR&R3_PTR&R5_PTR, newBCOzh_fast ); bco = Hp - bytes + WDS(1); SET_HDR(bco, stg_BCO_info, W_[CCCS]); @@ -1918,8 +2044,7 @@ newBCOzh_fast StgBCO_instrs(bco) = R1; StgBCO_literals(bco) = R2; StgBCO_ptrs(bco) = R3; - StgBCO_itbls(bco) = R4; - StgBCO_arity(bco) = HALF_W_(R5); + StgBCO_arity(bco) = HALF_W_(R4); StgBCO_size(bco) = HALF_W_(words); // Copy the arity/bitmap info into the BCO @@ -1961,6 +2086,70 @@ mkApUpd0zh_fast RET_P(ap); } +unpackClosurezh_fast +{ +/* args: R1 = closure to analyze */ +// TODO: Consider the absence of ptrs or nonptrs as a special case ? + + W_ info, ptrs, nptrs, p, ptrs_arr, nptrs_arr; + info = %GET_STD_INFO(UNTAG(R1)); + + // Some closures have non-standard layout, so we omit those here. + W_ type; + type = TO_W_(%INFO_TYPE(info)); + switch [0 .. N_CLOSURE_TYPES] type { + case THUNK_SELECTOR : { + ptrs = 1; + nptrs = 0; + goto out; + } + case THUNK, THUNK_1_0, THUNK_0_1, THUNK_2_0, THUNK_1_1, + THUNK_0_2, THUNK_STATIC, AP, PAP, AP_STACK, BCO : { + ptrs = 0; + nptrs = 0; + goto out; + } + default: { + ptrs = TO_W_(%INFO_PTRS(info)); + nptrs = TO_W_(%INFO_NPTRS(info)); + goto out; + }} +out: + + W_ ptrs_arr_sz, nptrs_arr_sz; + nptrs_arr_sz = SIZEOF_StgArrWords + WDS(nptrs); + ptrs_arr_sz = SIZEOF_StgMutArrPtrs + WDS(ptrs); + + ALLOC_PRIM (ptrs_arr_sz + nptrs_arr_sz, R1_PTR, unpackClosurezh_fast); + + W_ clos; + clos = UNTAG(R1); + + ptrs_arr = Hp - nptrs_arr_sz - ptrs_arr_sz + WDS(1); + nptrs_arr = Hp - nptrs_arr_sz + WDS(1); + + SET_HDR(ptrs_arr, stg_MUT_ARR_PTRS_FROZEN_info, W_[CCCS]); + StgMutArrPtrs_ptrs(ptrs_arr) = ptrs; + p = 0; +for: + if(p < ptrs) { + W_[ptrs_arr + SIZEOF_StgMutArrPtrs + WDS(p)] = StgClosure_payload(clos,p); + p = p + 1; + goto for; + } + + SET_HDR(nptrs_arr, stg_ARR_WORDS_info, W_[CCCS]); + StgArrWords_words(nptrs_arr) = nptrs; + p = 0; +for2: + if(p < nptrs) { + W_[BYTE_ARR_CTS(nptrs_arr) + WDS(p)] = StgClosure_payload(clos, p+ptrs); + p = p + 1; + goto for2; + } + RET_NPP(info, ptrs_arr, nptrs_arr); +} + /* ----------------------------------------------------------------------------- Thread I/O blocking primitives -------------------------------------------------------------------------- */ @@ -1969,11 +2158,11 @@ mkApUpd0zh_fast * macro in Schedule.h). */ #define APPEND_TO_BLOCKED_QUEUE(tso) \ - ASSERT(StgTSO_link(tso) == END_TSO_QUEUE); \ + ASSERT(StgTSO__link(tso) == END_TSO_QUEUE); \ if (W_[blocked_queue_hd] == END_TSO_QUEUE) { \ W_[blocked_queue_hd] = tso; \ } else { \ - StgTSO_link(W_[blocked_queue_tl]) = tso; \ + foreign "C" setTSOLink(MyCapability() "ptr", W_[blocked_queue_tl] "ptr", tso) []; \ } \ W_[blocked_queue_tl] = tso; @@ -1981,7 +2170,7 @@ waitReadzh_fast { /* args: R1 */ #ifdef THREADED_RTS - foreign "C" barf("waitRead# on threaded RTS"); + foreign "C" barf("waitRead# on threaded RTS") never returns; #else ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16); @@ -1998,7 +2187,7 @@ waitWritezh_fast { /* args: R1 */ #ifdef THREADED_RTS - foreign "C" barf("waitWrite# on threaded RTS"); + foreign "C" barf("waitWrite# on threaded RTS") never returns; #else ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16); @@ -2023,7 +2212,7 @@ delayzh_fast #endif #ifdef THREADED_RTS - foreign "C" barf("delay# on threaded RTS"); + foreign "C" barf("delay# on threaded RTS") never returns; #else /* args: R1 (microsecond delay amount) */ @@ -2033,9 +2222,9 @@ delayzh_fast #ifdef mingw32_HOST_OS /* could probably allocate this on the heap instead */ - "ptr" ares = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult, + ("ptr" ares) = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult, stg_delayzh_malloc_str); - reqID = foreign "C" addDelayRequest(R1); + (reqID) = foreign "C" addDelayRequest(R1); StgAsyncIOResult_reqID(ares) = reqID; StgAsyncIOResult_len(ares) = 0; StgAsyncIOResult_errCode(ares) = 0; @@ -2053,8 +2242,12 @@ delayzh_fast W_ time; W_ divisor; - time = foreign "C" getourtimeofday() [R1]; - divisor = TO_W_(RtsFlags_MiscFlags_tickInterval(RtsFlags))*1000; + (time) = foreign "C" getourtimeofday() [R1]; + divisor = TO_W_(RtsFlags_MiscFlags_tickInterval(RtsFlags)); + if (divisor == 0) { + divisor = 50; + } + divisor = divisor * 1000; target = ((R1 + divisor - 1) / divisor) /* divide rounding up */ + time + 1; /* Add 1 as getourtimeofday rounds down */ StgTSO_block_info(CurrentTSO) = target; @@ -2065,15 +2258,15 @@ delayzh_fast while: if (t != END_TSO_QUEUE && StgTSO_block_info(t) < target) { prev = t; - t = StgTSO_link(t); + t = StgTSO__link(t); goto while; } - StgTSO_link(CurrentTSO) = t; + StgTSO__link(CurrentTSO) = t; if (prev == NULL) { W_[sleeping_queue] = CurrentTSO; } else { - StgTSO_link(prev) = CurrentTSO; + foreign "C" setTSOLink(MyCapability() "ptr", prev "ptr", CurrentTSO) []; } jump stg_block_noregs; #endif @@ -2089,7 +2282,7 @@ asyncReadzh_fast CInt reqID; #ifdef THREADED_RTS - foreign "C" barf("asyncRead# on threaded RTS"); + foreign "C" barf("asyncRead# on threaded RTS") never returns; #else /* args: R1 = fd, R2 = isSock, R3 = len, R4 = buf */ @@ -2097,10 +2290,10 @@ asyncReadzh_fast StgTSO_why_blocked(CurrentTSO) = BlockedOnRead::I16; /* could probably allocate this on the heap instead */ - "ptr" ares = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult, + ("ptr" ares) = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult, stg_asyncReadzh_malloc_str) [R1,R2,R3,R4]; - reqID = foreign "C" addIORequest(R1, 0/*FALSE*/,R2,R3,R4 "ptr") []; + (reqID) = foreign "C" addIORequest(R1, 0/*FALSE*/,R2,R3,R4 "ptr") []; StgAsyncIOResult_reqID(ares) = reqID; StgAsyncIOResult_len(ares) = 0; StgAsyncIOResult_errCode(ares) = 0; @@ -2117,17 +2310,17 @@ asyncWritezh_fast CInt reqID; #ifdef THREADED_RTS - foreign "C" barf("asyncWrite# on threaded RTS"); + foreign "C" barf("asyncWrite# on threaded RTS") never returns; #else /* args: R1 = fd, R2 = isSock, R3 = len, R4 = buf */ ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16); StgTSO_why_blocked(CurrentTSO) = BlockedOnWrite::I16; - "ptr" ares = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult, + ("ptr" ares) = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult, stg_asyncWritezh_malloc_str) [R1,R2,R3,R4]; - reqID = foreign "C" addIORequest(R1, 1/*TRUE*/,R2,R3,R4 "ptr") []; + (reqID) = foreign "C" addIORequest(R1, 1/*TRUE*/,R2,R3,R4 "ptr") []; StgAsyncIOResult_reqID(ares) = reqID; StgAsyncIOResult_len(ares) = 0; @@ -2145,7 +2338,7 @@ asyncDoProczh_fast CInt reqID; #ifdef THREADED_RTS - foreign "C" barf("asyncDoProc# on threaded RTS"); + foreign "C" barf("asyncDoProc# on threaded RTS") never returns; #else /* args: R1 = proc, R2 = param */ @@ -2153,10 +2346,10 @@ asyncDoProczh_fast StgTSO_why_blocked(CurrentTSO) = BlockedOnDoProc::I16; /* could probably allocate this on the heap instead */ - "ptr" ares = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult, + ("ptr" ares) = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult, stg_asyncDoProczh_malloc_str) [R1,R2]; - reqID = foreign "C" addDoProcRequest(R1 "ptr",R2 "ptr") []; + (reqID) = foreign "C" addDoProcRequest(R1 "ptr",R2 "ptr") []; StgAsyncIOResult_reqID(ares) = reqID; StgAsyncIOResult_len(ares) = 0; StgAsyncIOResult_errCode(ares) = 0; @@ -2167,20 +2360,75 @@ asyncDoProczh_fast } #endif +// noDuplicate# tries to ensure that none of the thunks under +// evaluation by the current thread are also under evaluation by +// another thread. It relies on *both* threads doing noDuplicate#; +// the second one will get blocked if they are duplicating some work. +noDuplicatezh_fast +{ + SAVE_THREAD_STATE(); + ASSERT(StgTSO_what_next(CurrentTSO) == ThreadRunGHC::I16); + foreign "C" threadPaused (MyCapability() "ptr", CurrentTSO "ptr") []; + + if (StgTSO_what_next(CurrentTSO) == ThreadKilled::I16) { + jump stg_threadFinished; + } else { + LOAD_THREAD_STATE(); + ASSERT(StgTSO_what_next(CurrentTSO) == ThreadRunGHC::I16); + jump %ENTRY_CODE(Sp(0)); + } +} + +getApStackValzh_fast +{ + W_ ap_stack, offset, val, ok; + + /* args: R1 = AP_STACK, R2 = offset */ + ap_stack = R1; + offset = R2; + + if (%INFO_PTR(ap_stack) == stg_AP_STACK_info) { + ok = 1; + val = StgAP_STACK_payload(ap_stack,offset); + } else { + ok = 0; + val = R1; + } + RET_NP(ok,val); +} + /* ----------------------------------------------------------------------------- - ** temporary ** + Misc. primitives + -------------------------------------------------------------------------- */ - classes CCallable and CReturnable don't really exist, but the - compiler insists on generating dictionaries containing references - to GHC_ZcCCallable_static_info etc., so we provide dummy symbols - for these. Some C compilers can't cope with zero-length static arrays, - so we have to make these one element long. - --------------------------------------------------------------------------- */ +// Write the cost center stack of the first argument on stderr; return +// the second. Possibly only makes sense for already evaluated +// things? +traceCcszh_fast +{ + W_ ccs; -section "rodata" { - GHC_ZCCCallable_static_info: W_ 0; +#ifdef PROFILING + ccs = StgHeader_ccs(UNTAG(R1)); + foreign "C" fprintCCS_stderr(ccs "ptr") [R2]; +#endif + + R1 = R2; + ENTER(); } -section "rodata" { - GHC_ZCCReturnable_static_info: W_ 0; +getSparkzh_fast +{ + W_ spark; + +#ifndef THREADED_RTS + RET_NP(0,ghczmprim_GHCziBool_False_closure); +#else + (spark) = foreign "C" findSpark(MyCapability()); + if (spark != 0) { + RET_NP(1,spark); + } else { + RET_NP(0,ghczmprim_GHCziBool_False_closure); + } +#endif }