X-Git-Url: http://git.megacz.com/?a=blobdiff_plain;f=ghc%2Frts%2FPrimOps.cmm;h=f1c214e3043cd1c09dc1c1c2b4a0a0c66ac60a2a;hb=28a464a75e14cece5db40f2765a29348273ff2d2;hp=7b1c8aad1bbbd5dcd4c46deaeeb657498f0357ab;hpb=ab3d1f285cef784138d99e70f7359ea6e67f6c25;p=ghc-hetmet.git diff --git a/ghc/rts/PrimOps.cmm b/ghc/rts/PrimOps.cmm index 7b1c8aa..f1c214e 100644 --- a/ghc/rts/PrimOps.cmm +++ b/ghc/rts/PrimOps.cmm @@ -49,7 +49,7 @@ newByteArrayzh_fast n = R1; payload_words = ROUNDUP_BYTES_TO_WDS(n); words = BYTES_TO_WDS(SIZEOF_StgArrWords) + payload_words; - "ptr" p = foreign "C" allocateLocal(BaseReg "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; @@ -73,7 +73,7 @@ newPinnedByteArrayzh_fast words = words + 1; } - "ptr" p = foreign "C" allocatePinned(words); + "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 @@ -97,10 +97,10 @@ newArrayzh_fast MAYBE_GC(R2_PTR,newArrayzh_fast); words = BYTES_TO_WDS(SIZEOF_StgMutArrPtrs) + n; - "ptr" arr = foreign "C" allocateLocal(BaseReg "ptr",words); + "ptr" arr = foreign "C" allocateLocal(MyCapability() "ptr",words) [R2]; TICK_ALLOC_PRIM(SIZEOF_StgMutArrPtrs, WDS(n), 0); - SET_HDR(arr, stg_MUT_ARR_PTRS_info, W_[CCCS]); + SET_HDR(arr, stg_MUT_ARR_PTRS_DIRTY_info, W_[CCCS]); StgMutArrPtrs_ptrs(arr) = n; // Initialise all elements of the the array with the value in R2 @@ -126,21 +126,25 @@ unsafeThawArrayzh_fast // the mutable list is not easy, because the mut_list is only singly-linked). // // So that we can tell whether a MUT_ARR_PTRS_FROZEN is on the mutable list, - // when we freeze it we set the info ptr to be MUT_ARR_PTRS_FROZEN0 to indicate - // that it is still on the mutable list. - + // when we freeze it we set the info ptr to be MUT_ARR_PTRS_FROZEN0 + // to indicate that it is still on the mutable list. + // // So, when we thaw a MUT_ARR_PTRS_FROZEN, we must cope with two cases: // either it is on a mut_list, or it isn't. We adopt the convention that - // the mut_link field is NULL if it isn't on a mut_list, and the GC - // maintains this invariant. + // the closure type is MUT_ARR_PTRS_FROZEN0 if it is on the mutable list, + // and MUT_ARR_PTRS_FROZEN otherwise. In fact it wouldn't matter if + // we put it on the mutable list more than once, but it would get scavenged + // multiple times during GC, which would be unnecessarily slow. // - if (%INFO_TYPE(%GET_STD_INFO(R1)) != HALF_W_(MUT_ARR_PTRS_FROZEN0)) { - foreign "C" recordMutableLock(R1 "ptr"); + if (StgHeader_info(R1) != stg_MUT_ARR_PTRS_FROZEN0_info) { + SET_INFO(R1,stg_MUT_ARR_PTRS_DIRTY_info); + foreign "C" recordMutableLock(R1 "ptr") [R1]; + // must be done after SET_INFO, because it ASSERTs closure_MUTABLE() + RET_P(R1); + } else { + SET_INFO(R1,stg_MUT_ARR_PTRS_DIRTY_info); + RET_P(R1); } - - SET_INFO(R1,stg_MUT_ARR_PTRS_info); - - RET_P(R1); } /* ----------------------------------------------------------------------------- @@ -155,7 +159,7 @@ newMutVarzh_fast ALLOC_PRIM( SIZEOF_StgMutVar, R1_PTR, newMutVarzh_fast); mv = Hp - SIZEOF_StgMutVar + WDS(1); - SET_HDR(mv,stg_MUT_VAR_info,W_[CCCS]); + SET_HDR(mv,stg_MUT_VAR_DIRTY_info,W_[CCCS]); StgMutVar_var(mv) = R1; RET_P(mv); @@ -202,8 +206,8 @@ atomicModifyMutVarzh_fast HP_CHK_GEN_TICKY(SIZE, R1_PTR & R2_PTR, atomicModifyMutVarzh_fast); -#if defined(SMP) - foreign "C" ACQUIRE_LOCK(sm_mutex "ptr"); +#if defined(THREADED_RTS) + foreign "C" ACQUIRE_LOCK(atomic_modify_mutvar_mutex "ptr") [R1,R2]; #endif x = StgMutVar_var(R1); @@ -224,6 +228,7 @@ atomicModifyMutVarzh_fast 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); @@ -232,34 +237,14 @@ atomicModifyMutVarzh_fast LDV_RECORD_CREATE(r); StgThunk_payload(r,0) = z; -#if defined(SMP) - foreign "C" RELEASE_LOCK(sm_mutex "ptr"); +#if defined(THREADED_RTS) + foreign "C" RELEASE_LOCK(atomic_modify_mutvar_mutex "ptr") []; #endif RET_P(r); } /* ----------------------------------------------------------------------------- - Foreign Object Primitives - -------------------------------------------------------------------------- */ - -mkForeignObjzh_fast -{ - /* R1 = ptr to foreign object, - */ - W_ result; - - ALLOC_PRIM( SIZEOF_StgForeignObj, NO_PTRS, mkForeignObjzh_fast); - - result = Hp - SIZEOF_StgForeignObj + WDS(1); - SET_HDR(result,stg_FOREIGN_info,W_[CCCS]); - StgForeignObj_data(result) = R1; - - /* returns (# s#, ForeignObj# #) */ - RET_P(result); -} - -/* ----------------------------------------------------------------------------- Weak Pointer Primitives -------------------------------------------------------------------------- */ @@ -289,7 +274,7 @@ mkWeakzh_fast StgWeak_link(w) = W_[weak_ptr_list]; W_[weak_ptr_list] = w; - IF_DEBUG(weak, foreign "C" debugBelch(stg_weak_msg,w)); + IF_DEBUG(weak, foreign "C" debugBelch(stg_weak_msg,w) []); RET_P(w); } @@ -527,7 +512,7 @@ word64ToIntegerzh_fast /* ToDo: this is shockingly inefficient */ -#ifndef SMP +#ifndef THREADED_RTS section "bss" { mp_tmp1: bits8 [SIZEOF_MP_INT]; @@ -549,7 +534,7 @@ section "bss" { } #endif -#ifdef SMP +#ifdef THREADED_RTS #define FETCH_MP_TEMP(X) \ W_ X; \ X = BaseReg + (OFFSET_StgRegTable_r ## X); @@ -582,10 +567,10 @@ name \ MP_INT__mp_size(mp_tmp2) = (s2); \ MP_INT__mp_d(mp_tmp2) = BYTE_ARR_CTS(d2); \ \ - foreign "C" mpz_init(mp_result1 "ptr"); \ + foreign "C" mpz_init(mp_result1 "ptr") []; \ \ /* Perform the operation */ \ - foreign "C" mp_fun(mp_result1 "ptr",mp_tmp1 "ptr",mp_tmp2 "ptr"); \ + foreign "C" mp_fun(mp_result1 "ptr",mp_tmp1 "ptr",mp_tmp2 "ptr") []; \ \ RET_NP(TO_W_(MP_INT__mp_size(mp_result1)), \ MP_INT__mp_d(mp_result1) - SIZEOF_StgArrWords); \ @@ -609,10 +594,10 @@ name \ MP_INT__mp_size(mp_tmp1) = (s1); \ MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(d1); \ \ - foreign "C" mpz_init(mp_result1 "ptr"); \ + foreign "C" mpz_init(mp_result1 "ptr") []; \ \ /* Perform the operation */ \ - foreign "C" mp_fun(mp_result1 "ptr",mp_tmp1 "ptr"); \ + foreign "C" mp_fun(mp_result1 "ptr",mp_tmp1 "ptr") []; \ \ RET_NP(TO_W_(MP_INT__mp_size(mp_result1)), \ MP_INT__mp_d(mp_result1) - SIZEOF_StgArrWords); \ @@ -643,11 +628,11 @@ name MP_INT__mp_size(mp_tmp2) = (s2); \ MP_INT__mp_d(mp_tmp2) = BYTE_ARR_CTS(d2); \ \ - foreign "C" mpz_init(mp_result1 "ptr"); \ - foreign "C" mpz_init(mp_result2 "ptr"); \ + foreign "C" mpz_init(mp_result1 "ptr") []; \ + foreign "C" mpz_init(mp_result2 "ptr") []; \ \ /* Perform the operation */ \ - foreign "C" mp_fun(mp_result1 "ptr",mp_result2 "ptr",mp_tmp1 "ptr",mp_tmp2 "ptr"); \ + foreign "C" mp_fun(mp_result1 "ptr",mp_result2 "ptr",mp_tmp1 "ptr",mp_tmp2 "ptr") []; \ \ RET_NPNP(TO_W_(MP_INT__mp_size(mp_result1)), \ MP_INT__mp_d(mp_result1) - SIZEOF_StgArrWords, \ @@ -670,7 +655,7 @@ GMP_TAKE1_RET1(complementIntegerzh_fast, mpz_com) GMP_TAKE2_RET2(quotRemIntegerzh_fast, mpz_tdiv_qr) GMP_TAKE2_RET2(divModIntegerzh_fast, mpz_fdiv_qr) -#ifndef SMP +#ifndef THREADED_RTS section "bss" { mp_tmp_w: W_; // NB. mp_tmp_w is really an here mp_limb_t } @@ -683,7 +668,7 @@ gcdIntzh_fast FETCH_MP_TEMP(mp_tmp_w); W_[mp_tmp_w] = R1; - r = foreign "C" mpn_gcd_1(mp_tmp_w "ptr", 1, R2); + r = foreign "C" mpn_gcd_1(mp_tmp_w "ptr", 1, R2) []; R1 = r; /* Result parked in R1, return via info-pointer at TOS */ @@ -694,7 +679,7 @@ gcdIntzh_fast gcdIntegerIntzh_fast { /* R1 = s1; R2 = d1; R3 = the int */ - R1 = foreign "C" mpn_gcd_1( BYTE_ARR_CTS(R2) "ptr", R1, R3); + R1 = foreign "C" mpn_gcd_1( BYTE_ARR_CTS(R2) "ptr", R1, R3) []; /* Result parked in R1, return via info-pointer at TOS */ jump %ENTRY_CODE(Sp(0)); @@ -775,7 +760,7 @@ cmpIntegerzh_fast up = BYTE_ARR_CTS(R2); vp = BYTE_ARR_CTS(R4); - cmp = foreign "C" mpn_cmp(up "ptr", vp "ptr", size); + cmp = foreign "C" mpn_cmp(up "ptr", vp "ptr", size) []; if (cmp == 0 :: CInt) { R1 = 0; @@ -849,7 +834,7 @@ decodeFloatzh_fast MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(p); /* Perform the operation */ - foreign "C" __decodeFloat(mp_tmp1 "ptr",mp_tmp_w "ptr" ,arg); + foreign "C" __decodeFloat(mp_tmp1 "ptr",mp_tmp_w "ptr" ,arg) []; /* returns: (Int# (expn), Int#, ByteArray#) */ RET_NNP(W_[mp_tmp_w], TO_W_(MP_INT__mp_size(mp_tmp1)), p); @@ -878,7 +863,7 @@ decodeDoublezh_fast MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(p); /* Perform the operation */ - foreign "C" __decodeDouble(mp_tmp1 "ptr", mp_tmp_w "ptr",arg); + foreign "C" __decodeDouble(mp_tmp1 "ptr", mp_tmp_w "ptr",arg) []; /* returns: (Int# (expn), Int#, ByteArray#) */ RET_NNP(W_[mp_tmp_w], TO_W_(MP_INT__mp_size(mp_tmp1)), p); @@ -891,18 +876,45 @@ decodeDoublezh_fast forkzh_fast { /* args: R1 = closure to spark */ - + MAYBE_GC(R1_PTR, forkzh_fast); - // create it right now, return ThreadID in R1 - "ptr" R1 = foreign "C" createIOThread( RtsFlags_GcFlags_initialStkSize(RtsFlags), - R1 "ptr"); - foreign "C" scheduleThread(R1 "ptr"); + W_ closure; + W_ threadid; + closure = R1; + + "ptr" threadid = foreign "C" createIOThread( MyCapability() "ptr", + RtsFlags_GcFlags_initialStkSize(RtsFlags), + closure "ptr") []; + foreign "C" scheduleThread(MyCapability() "ptr", threadid "ptr") []; + + // switch at the earliest opportunity + CInt[context_switch] = 1 :: CInt; + + RET_P(threadid); +} + +forkOnzh_fast +{ + /* args: R1 = cpu, R2 = closure to spark */ + + MAYBE_GC(R2_PTR, forkOnzh_fast); + + W_ cpu; + W_ closure; + W_ threadid; + cpu = R1; + closure = R2; + + "ptr" threadid = foreign "C" createIOThread( MyCapability() "ptr", + RtsFlags_GcFlags_initialStkSize(RtsFlags), + closure "ptr") []; + foreign "C" scheduleThreadOn(MyCapability() "ptr", cpu, threadid "ptr") []; // switch at the earliest opportunity CInt[context_switch] = 1 :: CInt; - RET_P(R1); + RET_P(threadid); } yieldzh_fast @@ -922,7 +934,7 @@ labelThreadzh_fast R1 = ThreadId# R2 = Addr# */ #ifdef DEBUG - foreign "C" labelThread(R1 "ptr", R2 "ptr"); + foreign "C" labelThread(R1 "ptr", R2 "ptr") []; #endif jump %ENTRY_CODE(Sp(0)); } @@ -931,7 +943,7 @@ isCurrentThreadBoundzh_fast { /* no args */ W_ r; - r = foreign "C" isThreadBound(CurrentTSO); + r = foreign "C" isThreadBound(CurrentTSO) []; RET_N(r); } @@ -991,8 +1003,8 @@ INFO_TABLE_RET(stg_catch_retry_frame, frame = Sp; trec = StgTSO_trec(CurrentTSO); - "ptr" outer = foreign "C" stmGetEnclosingTRec(trec "ptr"); - r = foreign "C" stmCommitNestedTransaction(BaseReg "ptr", trec "ptr"); + "ptr" outer = foreign "C" stmGetEnclosingTRec(trec "ptr") []; + r = foreign "C" stmCommitNestedTransaction(MyCapability() "ptr", trec "ptr") []; if (r) { /* Succeeded (either first branch or second branch) */ StgTSO_trec(CurrentTSO) = outer; @@ -1002,7 +1014,7 @@ INFO_TABLE_RET(stg_catch_retry_frame, } else { /* Did not commit: retry */ W_ new_trec; - "ptr" new_trec = foreign "C" stmStartTransaction(BaseReg "ptr", outer "ptr"); + "ptr" new_trec = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") []; StgTSO_trec(CurrentTSO) = new_trec; if (StgCatchRetryFrame_running_alt_code(frame)) { R1 = StgCatchRetryFrame_alt_code(frame); @@ -1010,8 +1022,7 @@ INFO_TABLE_RET(stg_catch_retry_frame, R1 = StgCatchRetryFrame_first_code(frame); StgCatchRetryFrame_first_code_trec(frame) = new_trec; } - Sp_adj(-1); - jump RET_LBL(stg_ap_v); + jump stg_ap_v_fast; } } @@ -1036,11 +1047,11 @@ ATOMICALLY_FRAME_ERROR(stg_atomically_frame_7_ret) #endif #if defined(PROFILING) -#define ATOMICALLY_FRAME_BITMAP 7 -#define ATOMICALLY_FRAME_WORDS 4 +#define ATOMICALLY_FRAME_BITMAP 3 +#define ATOMICALLY_FRAME_WORDS 3 #else -#define ATOMICALLY_FRAME_BITMAP 1 -#define ATOMICALLY_FRAME_WORDS 2 +#define ATOMICALLY_FRAME_BITMAP 0 +#define ATOMICALLY_FRAME_WORDS 1 #endif @@ -1056,50 +1067,64 @@ INFO_TABLE_RET(stg_atomically_frame, stg_atomically_frame_6_ret, stg_atomically_frame_7_ret) { - W_ frame, trec, valid; - IF_NOT_REG_R1(W_ rval; rval = Sp(0); Sp_adj(1); ) + W_ frame, trec, valid; + IF_NOT_REG_R1(W_ rval; rval = Sp(0); Sp_adj(1); ) - frame = Sp; - trec = StgTSO_trec(CurrentTSO); - if (StgAtomicallyFrame_waiting(frame)) { - /* The TSO is currently waiting: should we stop waiting? */ - valid = foreign "C" stmReWait(CurrentTSO "ptr"); - if (valid) { - /* Previous attempt is still valid: no point trying again yet */ + frame = Sp; + trec = StgTSO_trec(CurrentTSO); + + /* The TSO is not currently waiting: try to commit the transaction */ + valid = foreign "C" stmCommitTransaction(MyCapability() "ptr", trec "ptr") []; + if (valid) { + /* 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") []; + StgTSO_trec(CurrentTSO) = trec; + R1 = StgAtomicallyFrame_code(frame); + jump stg_ap_v_fast; + } +} + +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) +{ + 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") []; + if (valid) { + /* 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(BaseReg "ptr", NO_TREC "ptr"); - StgTSO_trec(CurrentTSO) = trec; - StgAtomicallyFrame_waiting(frame) = 0 :: CInt; /* false; */ - R1 = StgAtomicallyFrame_code(frame); - Sp_adj(-1); - jump RET_LBL(stg_ap_v); - } - } else { - /* The TSO is not currently waiting: try to commit the transaction */ - valid = foreign "C" stmCommitTransaction(BaseReg "ptr", trec "ptr"); - if (valid) { - /* 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(BaseReg "ptr", NO_TREC "ptr"); - StgTSO_trec(CurrentTSO) = trec; - R1 = StgAtomicallyFrame_code(frame); - Sp_adj(-1); - jump RET_LBL(stg_ap_v); - } - } + jump stg_block_noregs; + } else { + /* Previous attempt is no longer valid: try again */ + "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); + jump stg_ap_v_fast; + } } - // STM catch frame -------------------------------------------------------------- #define CATCH_STM_FRAME_ENTRY_TEMPLATE(label,ret) \ @@ -1171,22 +1196,27 @@ atomicallyzh_fast /* Args: R1 = m :: STM a */ STK_CHK_GEN(SIZEOF_StgAtomicallyFrame + WDS(1), R1_PTR, atomicallyzh_fast); + old_trec = StgTSO_trec(CurrentTSO); + + /* Nested transactions are not allowed; raise an exception */ + if (old_trec != NO_TREC) { + R1 = GHCziIOBase_NestedAtomically_closure; + jump raisezh_fast; + } + /* Set up the atomically frame */ Sp = Sp - SIZEOF_StgAtomicallyFrame; frame = Sp; SET_HDR(frame,stg_atomically_frame_info, W_[CCCS]); - StgAtomicallyFrame_waiting(frame) = 0 :: CInt; // False StgAtomicallyFrame_code(frame) = R1; /* Start the memory transcation */ - old_trec = StgTSO_trec(CurrentTSO); - "ptr" new_trec = foreign "C" stmStartTransaction(BaseReg "ptr", old_trec "ptr"); + "ptr" new_trec = foreign "C" stmStartTransaction(MyCapability() "ptr", old_trec "ptr") [R1]; StgTSO_trec(CurrentTSO) = new_trec; /* Apply R1 to the realworld token */ - Sp_adj(-1); - jump RET_LBL(stg_ap_v); + jump stg_ap_v_fast; } @@ -1206,8 +1236,7 @@ catchSTMzh_fast StgCatchSTMFrame_handler(frame) = R2; /* Apply R1 to the realworld token */ - Sp_adj(-1); - jump RET_LBL(stg_ap_v); + jump stg_ap_v_fast; } @@ -1226,7 +1255,7 @@ catchRetryzh_fast /* Start a nested transaction within which to run the first code */ trec = StgTSO_trec(CurrentTSO); - "ptr" new_trec = foreign "C" stmStartTransaction(BaseReg "ptr", trec "ptr"); + "ptr" new_trec = foreign "C" stmStartTransaction(MyCapability() "ptr", trec "ptr") [R1,R2]; StgTSO_trec(CurrentTSO) = new_trec; /* Set up the catch-retry frame */ @@ -1240,8 +1269,7 @@ catchRetryzh_fast StgCatchRetryFrame_first_code_trec(frame) = new_trec; /* Apply R1 to the realworld token */ - Sp_adj(-1); - jump RET_LBL(stg_ap_v); + jump stg_ap_v_fast; } @@ -1258,9 +1286,9 @@ retryzh_fast // Find the enclosing ATOMICALLY_FRAME or CATCH_RETRY_FRAME retry_pop_stack: trec = StgTSO_trec(CurrentTSO); - "ptr" outer = foreign "C" stmGetEnclosingTRec(trec "ptr"); + "ptr" outer = foreign "C" stmGetEnclosingTRec(trec "ptr") []; StgTSO_sp(CurrentTSO) = Sp; - frame_type = foreign "C" findRetryFrameHelper(CurrentTSO "ptr"); + frame_type = foreign "C" findRetryFrameHelper(CurrentTSO "ptr") []; Sp = StgTSO_sp(CurrentTSO); frame = Sp; @@ -1269,19 +1297,20 @@ retry_pop_stack: ASSERT(outer != NO_TREC); if (!StgCatchRetryFrame_running_alt_code(frame)) { // Retry in the first code: try the alternative - "ptr" trec = foreign "C" stmStartTransaction(BaseReg "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); - Sp_adj(-1); - jump RET_LBL(stg_ap_v); + jump stg_ap_v_fast; } else { // Retry in the alternative code: propagate W_ other_trec; other_trec = StgCatchRetryFrame_first_code_trec(frame); - r = foreign "C" stmCommitNestedTransaction(BaseReg "ptr", other_trec "ptr"); + r = foreign "C" stmCommitNestedTransaction(MyCapability() "ptr", other_trec "ptr") []; if (r) { - r = foreign "C" stmCommitNestedTransaction(BaseReg "ptr", trec "ptr"); + r = foreign "C" stmCommitNestedTransaction(MyCapability() "ptr", trec "ptr") []; + } else { + foreign "C" stmAbortTransaction(MyCapability() "ptr", trec "ptr") []; } if (r) { // Merge between siblings succeeded: commit it back to enclosing transaction @@ -1291,13 +1320,12 @@ retry_pop_stack: goto retry_pop_stack; } else { // Merge failed: we musn't propagate the retry. Try both paths again. - "ptr" trec = foreign "C" stmStartTransaction(BaseReg "ptr", outer "ptr"); + "ptr" trec = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") []; StgCatchRetryFrame_first_code_trec(frame) = trec; StgCatchRetryFrame_running_alt_code(frame) = 0 :: CInt; // false; StgTSO_trec(CurrentTSO) = trec; R1 = StgCatchRetryFrame_first_code(frame); - Sp_adj(-1); - jump RET_LBL(stg_ap_v); + jump stg_ap_v_fast; } } } @@ -1305,24 +1333,24 @@ retry_pop_stack: // We've reached the ATOMICALLY_FRAME: attempt to wait ASSERT(frame_type == ATOMICALLY_FRAME); ASSERT(outer == NO_TREC); - r = foreign "C" stmWait(BaseReg "ptr", CurrentTSO "ptr", trec "ptr"); + r = foreign "C" stmWait(MyCapability() "ptr", CurrentTSO "ptr", trec "ptr") []; if (r) { // Transaction was valid: stmWait put us on the TVars' queues, we now block - StgAtomicallyFrame_waiting(frame) = 1 :: CInt; // true + 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;) - jump stg_block_noregs; + R3 = trec; // passing to stmWaitUnblock() + jump stg_block_stmwait; } else { // Transaction was not valid: retry immediately - "ptr" trec = foreign "C" stmStartTransaction(BaseReg "ptr", outer "ptr"); + "ptr" trec = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") []; StgTSO_trec(CurrentTSO) = trec; R1 = StgAtomicallyFrame_code(frame); Sp = frame; - Sp_adj(-1); - jump RET_LBL(stg_ap_v); + jump stg_ap_v_fast; } } @@ -1336,7 +1364,7 @@ newTVarzh_fast MAYBE_GC (R1_PTR, newTVarzh_fast); new_value = R1; - tv = foreign "C" stmNewTVar(BaseReg "ptr", new_value "ptr"); + "ptr" tv = foreign "C" stmNewTVar(MyCapability() "ptr", new_value "ptr") []; RET_P(tv); } @@ -1352,7 +1380,7 @@ readTVarzh_fast MAYBE_GC (R1_PTR, readTVarzh_fast); // Call to stmReadTVar may allocate trec = StgTSO_trec(CurrentTSO); tvar = R1; - "ptr" result = foreign "C" stmReadTVar(BaseReg "ptr", trec "ptr", tvar "ptr"); + "ptr" result = foreign "C" stmReadTVar(MyCapability() "ptr", trec "ptr", tvar "ptr") []; RET_P(result); } @@ -1371,7 +1399,7 @@ writeTVarzh_fast trec = StgTSO_trec(CurrentTSO); tvar = R1; new_value = R2; - foreign "C" stmWriteTVar(BaseReg "ptr", trec "ptr", tvar "ptr", new_value "ptr"); + foreign "C" stmWriteTVar(MyCapability() "ptr", trec "ptr", tvar "ptr", new_value "ptr") []; jump %ENTRY_CODE(Sp(0)); } @@ -1458,8 +1486,8 @@ takeMVarzh_fast /* args: R1 = MVar closure */ mvar = R1; -#if defined(SMP) - "ptr" info = foreign "C" lockClosure(mvar "ptr"); +#if defined(THREADED_RTS) + "ptr" info = foreign "C" lockClosure(mvar "ptr") []; #else info = GET_INFO(mvar); #endif @@ -1494,13 +1522,15 @@ takeMVarzh_fast /* actually perform the putMVar for the thread that we just woke up */ tso = StgMVar_head(mvar); PerformPut(tso,StgMVar_value(mvar)); + foreign "C" dirtyTSO(tso "ptr") []; #if defined(GRAN) || defined(PAR) /* ToDo: check 2nd arg (mvar) is right */ - "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar),mvar); + "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar),mvar) []; StgMVar_head(mvar) = tso; #else - "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr"); + "ptr" tso = foreign "C" unblockOne(MyCapability() "ptr", + StgMVar_head(mvar) "ptr") []; StgMVar_head(mvar) = tso; #endif @@ -1508,8 +1538,8 @@ takeMVarzh_fast StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure; } -#if defined(SMP) - foreign "C" unlockClosure(mvar "ptr", stg_FULL_MVAR_info); +#if defined(THREADED_RTS) + foreign "C" unlockClosure(mvar "ptr", stg_FULL_MVAR_info) []; #endif RET_P(val); } @@ -1518,8 +1548,8 @@ takeMVarzh_fast /* No further putMVars, MVar is now empty */ StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure; -#if defined(SMP) - foreign "C" unlockClosure(mvar "ptr", stg_EMPTY_MVAR_info); +#if defined(THREADED_RTS) + foreign "C" unlockClosure(mvar "ptr", stg_EMPTY_MVAR_info) []; #else SET_INFO(mvar,stg_EMPTY_MVAR_info); #endif @@ -1537,15 +1567,15 @@ tryTakeMVarzh_fast mvar = R1; -#if defined(SMP) - "ptr" info = foreign "C" lockClosure(mvar "ptr"); +#if defined(THREADED_RTS) + "ptr" info = foreign "C" lockClosure(mvar "ptr") []; #else info = GET_INFO(mvar); #endif if (info == stg_EMPTY_MVAR_info) { -#if defined(SMP) - foreign "C" unlockClosure(mvar "ptr", stg_EMPTY_MVAR_info); +#if defined(THREADED_RTS) + foreign "C" unlockClosure(mvar "ptr", stg_EMPTY_MVAR_info) []; #endif /* HACK: we need a pointer to pass back, * so we abuse NO_FINALIZER_closure @@ -1566,29 +1596,31 @@ tryTakeMVarzh_fast /* actually perform the putMVar for the thread that we just woke up */ tso = StgMVar_head(mvar); PerformPut(tso,StgMVar_value(mvar)); + foreign "C" dirtyTSO(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"); + "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr", mvar "ptr") []; StgMVar_head(mvar) = tso; #else - "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr"); + "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(SMP) - foreign "C" unlockClosure(mvar "ptr", stg_FULL_MVAR_info); +#if defined(THREADED_RTS) + foreign "C" unlockClosure(mvar "ptr", stg_FULL_MVAR_info) []; #endif } else { /* No further putMVars, MVar is now empty */ StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure; -#if defined(SMP) - foreign "C" unlockClosure(mvar "ptr", stg_EMPTY_MVAR_info); +#if defined(THREADED_RTS) + foreign "C" unlockClosure(mvar "ptr", stg_EMPTY_MVAR_info) []; #else SET_INFO(mvar,stg_EMPTY_MVAR_info); #endif @@ -1605,8 +1637,8 @@ putMVarzh_fast /* args: R1 = MVar, R2 = value */ mvar = R1; -#if defined(SMP) - "ptr" info = foreign "C" lockClosure(mvar "ptr"); +#if defined(THREADED_RTS) + "ptr" info = foreign "C" lockClosure(mvar "ptr") [R2]; #else info = GET_INFO(mvar); #endif @@ -1634,13 +1666,14 @@ putMVarzh_fast /* actually perform the takeMVar */ tso = StgMVar_head(mvar); PerformTake(tso, R2); + foreign "C" dirtyTSO(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"); + "ptr" tso = foreign "C" unblockOne(MyCapability() "ptr", StgMVar_head(mvar) "ptr",mvar "ptr") []; StgMVar_head(mvar) = tso; #else - "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr"); + "ptr" tso = foreign "C" unblockOne(MyCapability() "ptr", StgMVar_head(mvar) "ptr") []; StgMVar_head(mvar) = tso; #endif @@ -1648,8 +1681,8 @@ putMVarzh_fast StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure; } -#if defined(SMP) - foreign "C" unlockClosure(mvar "ptr", stg_EMPTY_MVAR_info); +#if defined(THREADED_RTS) + foreign "C" unlockClosure(mvar "ptr", stg_EMPTY_MVAR_info) []; #endif jump %ENTRY_CODE(Sp(0)); } @@ -1658,8 +1691,8 @@ putMVarzh_fast /* No further takes, the MVar is now full. */ StgMVar_value(mvar) = R2; -#if defined(SMP) - foreign "C" unlockClosure(mvar "ptr", stg_FULL_MVAR_info); +#if defined(THREADED_RTS) + foreign "C" unlockClosure(mvar "ptr", stg_FULL_MVAR_info) []; #else SET_INFO(mvar,stg_FULL_MVAR_info); #endif @@ -1677,15 +1710,15 @@ tryPutMVarzh_fast /* args: R1 = MVar, R2 = value */ mvar = R1; -#if defined(SMP) - "ptr" info = foreign "C" lockClosure(mvar "ptr"); +#if defined(THREADED_RTS) + "ptr" info = foreign "C" lockClosure(mvar "ptr") [R2]; #else info = GET_INFO(mvar); #endif if (info == stg_FULL_MVAR_info) { -#if defined(SMP) - foreign "C" unlockClosure(mvar "ptr", stg_FULL_MVAR_info); +#if defined(THREADED_RTS) + foreign "C" unlockClosure(mvar "ptr", stg_FULL_MVAR_info) []; #endif RET_N(0); } @@ -1699,13 +1732,14 @@ tryPutMVarzh_fast /* actually perform the takeMVar */ tso = StgMVar_head(mvar); PerformTake(tso, R2); + foreign "C" dirtyTSO(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"); + "ptr" tso = foreign "C" unblockOne(MyCapability() "ptr", StgMVar_head(mvar) "ptr",mvar "ptr") []; StgMVar_head(mvar) = tso; #else - "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr"); + "ptr" tso = foreign "C" unblockOne(MyCapability() "ptr", StgMVar_head(mvar) "ptr") []; StgMVar_head(mvar) = tso; #endif @@ -1713,24 +1747,23 @@ tryPutMVarzh_fast StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure; } -#if defined(SMP) - foreign "C" unlockClosure(mvar "ptr", stg_EMPTY_MVAR_info); +#if defined(THREADED_RTS) + foreign "C" unlockClosure(mvar "ptr", stg_EMPTY_MVAR_info) []; #endif - jump %ENTRY_CODE(Sp(0)); } else { /* No further takes, the MVar is now full. */ StgMVar_value(mvar) = R2; -#if defined(SMP) - foreign "C" unlockClosure(mvar "ptr", stg_FULL_MVAR_info); +#if defined(THREADED_RTS) + foreign "C" unlockClosure(mvar "ptr", stg_FULL_MVAR_info) []; #else SET_INFO(mvar,stg_FULL_MVAR_info); #endif - jump %ENTRY_CODE(Sp(0)); } + RET_N(1); /* ToDo: yield afterward for better communication performance? */ } @@ -1745,7 +1778,7 @@ 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. @@ -1768,7 +1801,7 @@ 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); } @@ -1872,7 +1905,7 @@ waitReadzh_fast /* args: R1 */ #ifdef THREADED_RTS foreign "C" barf("waitRead# on threaded RTS"); -#endif +#else ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16); StgTSO_why_blocked(CurrentTSO) = BlockedOnRead::I16; @@ -1881,6 +1914,7 @@ waitReadzh_fast // threaded RTS anyway. APPEND_TO_BLOCKED_QUEUE(CurrentTSO); jump stg_block_noregs; +#endif } waitWritezh_fast @@ -1888,7 +1922,7 @@ waitWritezh_fast /* args: R1 */ #ifdef THREADED_RTS foreign "C" barf("waitWrite# on threaded RTS"); -#endif +#else ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16); StgTSO_why_blocked(CurrentTSO) = BlockedOnWrite::I16; @@ -1897,6 +1931,7 @@ waitWritezh_fast // threaded RTS anyway. APPEND_TO_BLOCKED_QUEUE(CurrentTSO); jump stg_block_noregs; +#endif } @@ -1912,7 +1947,7 @@ delayzh_fast #ifdef THREADED_RTS foreign "C" barf("delay# on threaded RTS"); -#endif +#else /* args: R1 (microsecond delay amount) */ ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16); @@ -1962,6 +1997,7 @@ while: } jump stg_block_noregs; #endif +#endif /* !THREADED_RTS */ } @@ -1974,7 +2010,7 @@ asyncReadzh_fast #ifdef THREADED_RTS foreign "C" barf("asyncRead# on threaded RTS"); -#endif +#else /* args: R1 = fd, R2 = isSock, R3 = len, R4 = buf */ ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16); @@ -1982,14 +2018,16 @@ asyncReadzh_fast /* could probably allocate this on the heap instead */ "ptr" ares = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult, - stg_asyncReadzh_malloc_str); - reqID = foreign "C" addIORequest(R1, 0/*FALSE*/,R2,R3,R4 "ptr"); + stg_asyncReadzh_malloc_str) + [R1,R2,R3,R4]; + reqID = foreign "C" addIORequest(R1, 0/*FALSE*/,R2,R3,R4 "ptr") []; StgAsyncIOResult_reqID(ares) = reqID; StgAsyncIOResult_len(ares) = 0; StgAsyncIOResult_errCode(ares) = 0; StgTSO_block_info(CurrentTSO) = ares; APPEND_TO_BLOCKED_QUEUE(CurrentTSO); jump stg_block_async; +#endif } STRING(stg_asyncWritezh_malloc_str, "asyncWritezh_fast") @@ -2000,15 +2038,16 @@ asyncWritezh_fast #ifdef THREADED_RTS foreign "C" barf("asyncWrite# on threaded RTS"); -#endif +#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, - stg_asyncWritezh_malloc_str); - reqID = foreign "C" addIORequest(R1, 1/*TRUE*/,R2,R3,R4 "ptr"); + stg_asyncWritezh_malloc_str) + [R1,R2,R3,R4]; + reqID = foreign "C" addIORequest(R1, 1/*TRUE*/,R2,R3,R4 "ptr") []; StgAsyncIOResult_reqID(ares) = reqID; StgAsyncIOResult_len(ares) = 0; @@ -2016,6 +2055,7 @@ asyncWritezh_fast StgTSO_block_info(CurrentTSO) = ares; APPEND_TO_BLOCKED_QUEUE(CurrentTSO); jump stg_block_async; +#endif } STRING(stg_asyncDoProczh_malloc_str, "asyncDoProczh_fast") @@ -2024,20 +2064,26 @@ asyncDoProczh_fast W_ ares; CInt reqID; +#ifdef THREADED_RTS + foreign "C" barf("asyncDoProc# on threaded RTS"); +#else + /* args: R1 = proc, R2 = param */ ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16); StgTSO_why_blocked(CurrentTSO) = BlockedOnDoProc::I16; /* could probably allocate this on the heap instead */ "ptr" ares = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult, - stg_asyncDoProczh_malloc_str); - reqID = foreign "C" addDoProcRequest(R1 "ptr",R2 "ptr"); + stg_asyncDoProczh_malloc_str) + [R1,R2]; + reqID = foreign "C" addDoProcRequest(R1 "ptr",R2 "ptr") []; StgAsyncIOResult_reqID(ares) = reqID; StgAsyncIOResult_len(ares) = 0; StgAsyncIOResult_errCode(ares) = 0; StgTSO_block_info(CurrentTSO) = ares; APPEND_TO_BLOCKED_QUEUE(CurrentTSO); jump stg_block_async; +#endif } #endif