X-Git-Url: http://git.megacz.com/?a=blobdiff_plain;f=ghc%2Frts%2FPrimOps.cmm;h=23bc22efd13e3bb33e60829ad54cefc5acc12660;hb=080c9600d0ca63cc4125c6d2957ed8336b311c76;hp=d4a54c65286f415321465d7b808625b4008ebea3;hpb=aa07427aca52bcf2d76b8967aa27fe23fc8803bc;p=ghc-hetmet.git diff --git a/ghc/rts/PrimOps.cmm b/ghc/rts/PrimOps.cmm index d4a54c6..23bc22e 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" allocate(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" allocate(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 @@ -118,8 +118,6 @@ newArrayzh_fast unsafeThawArrayzh_fast { - SET_INFO(R1,stg_MUT_ARR_PTRS_info); - // SUBTLETY TO DO WITH THE OLD GEN MUTABLE LIST // // A MUT_ARR_PTRS lives on the mutable list, but a MUT_ARR_PTRS_FROZEN @@ -127,16 +125,26 @@ unsafeThawArrayzh_fast // it on the mutable list for the GC to remove (removing something from // 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. + // // 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 (StgMutClosure_mut_link(R1) == NULL) { - foreign "C" recordMutable(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); } - - RET_P(R1); } /* ----------------------------------------------------------------------------- @@ -151,7 +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); @@ -179,18 +187,18 @@ atomicModifyMutVarzh_fast */ #if MIN_UPD_SIZE > 1 -#define THUNK_1_SIZE (SIZEOF_StgHeader + WDS(MIN_UPD_SIZE)) +#define THUNK_1_SIZE (SIZEOF_StgThunkHeader + WDS(MIN_UPD_SIZE)) #define TICK_ALLOC_THUNK_1() TICK_ALLOC_UP_THK(WDS(1),WDS(MIN_UPD_SIZE-1)) #else -#define THUNK_1_SIZE (SIZEOF_StgHeader + WDS(1)) +#define THUNK_1_SIZE (SIZEOF_StgThunkHeader + WDS(1)) #define TICK_ALLOC_THUNK_1() TICK_ALLOC_UP_THK(WDS(1),0) #endif #if MIN_UPD_SIZE > 2 -#define THUNK_2_SIZE (SIZEOF_StgHeader + WDS(MIN_UPD_SIZE)) +#define THUNK_2_SIZE (SIZEOF_StgThunkHeader + WDS(MIN_UPD_SIZE)) #define TICK_ALLOC_THUNK_2() TICK_ALLOC_UP_THK(WDS(2),WDS(MIN_UPD_SIZE-2)) #else -#define THUNK_2_SIZE (SIZEOF_StgHeader + WDS(2)) +#define THUNK_2_SIZE (SIZEOF_StgThunkHeader + WDS(2)) #define TICK_ALLOC_THUNK_2() TICK_ALLOC_UP_THK(WDS(2),0) #endif @@ -198,6 +206,10 @@ 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); TICK_ALLOC_THUNK_2(); @@ -205,46 +217,31 @@ atomicModifyMutVarzh_fast z = Hp - THUNK_2_SIZE + WDS(1); SET_HDR(z, stg_ap_2_upd_info, W_[CCCS]); LDV_RECORD_CREATE(z); - StgClosure_payload(z,0) = R2; - StgClosure_payload(z,1) = x; + StgThunk_payload(z,0) = R2; + StgThunk_payload(z,1) = x; TICK_ALLOC_THUNK_1(); CCCS_ALLOC(THUNK_1_SIZE); y = z - THUNK_1_SIZE; SET_HDR(y, stg_sel_0_upd_info, W_[CCCS]); LDV_RECORD_CREATE(y); - StgClosure_payload(y,0) = z; + 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; SET_HDR(r, stg_sel_1_upd_info, W_[CCCS]); LDV_RECORD_CREATE(r); - StgClosure_payload(r,0) = z; - - RET_P(r); -} - -/* ----------------------------------------------------------------------------- - Foreign Object Primitives - -------------------------------------------------------------------------- */ - -mkForeignObjzh_fast -{ - /* R1 = ptr to foreign object, - */ - W_ result; - - ALLOC_PRIM( SIZEOF_StgForeignObj, NO_PTRS, mkForeignObjzh_fast); + StgThunk_payload(r,0) = z; - result = Hp - SIZEOF_StgForeignObj + WDS(1); - SET_HDR(result,stg_FOREIGN_info,W_[CCCS]); - StgForeignObj_data(result) = R1; +#if defined(THREADED_RTS) + foreign "C" RELEASE_LOCK(atomic_modify_mutvar_mutex "ptr") []; +#endif - /* returns (# s#, ForeignObj# #) */ - RET_P(result); + RET_P(r); } /* ----------------------------------------------------------------------------- @@ -277,7 +274,7 @@ mkWeakzh_fast StgWeak_link(w) = W_[weak_ptr_list]; W_[weak_ptr_list] = w; - IF_DEBUG(weak, foreign "C" fprintf(stderr,stg_weak_msg,w)); + IF_DEBUG(weak, foreign "C" debugBelch(stg_weak_msg,w) []); RET_P(w); } @@ -316,6 +313,7 @@ finalizzeWeakzh_fast LDV_RECORD_CREATE(w); f = StgWeak_finalizer(w); + StgDeadWeak_link(w) = StgWeak_link(w); /* return the finalizer */ if (f == stg_NO_FINALIZER_closure) { @@ -514,6 +512,7 @@ word64ToIntegerzh_fast /* ToDo: this is shockingly inefficient */ +#ifndef THREADED_RTS section "bss" { mp_tmp1: bits8 [SIZEOF_MP_INT]; @@ -525,98 +524,120 @@ section "bss" { } section "bss" { - result1: + mp_result1: bits8 [SIZEOF_MP_INT]; } section "bss" { - result2: + mp_result2: bits8 [SIZEOF_MP_INT]; } +#endif + +#ifdef THREADED_RTS +#define FETCH_MP_TEMP(X) \ +W_ X; \ +X = BaseReg + (OFFSET_StgRegTable_r ## X); +#else +#define FETCH_MP_TEMP(X) /* Nothing */ +#endif -#define GMP_TAKE2_RET1(name,mp_fun) \ -name \ -{ \ - W_ s1, s2, d1, d2; \ - \ - /* call doYouWantToGC() */ \ - MAYBE_GC(R2_PTR & R4_PTR, name); \ - \ - s1 = R1; \ - d1 = R2; \ - s2 = R3; \ - d2 = R4; \ - \ - MP_INT__mp_alloc(mp_tmp1) = StgArrWords_words(d1); \ - MP_INT__mp_size(mp_tmp1) = (s1); \ - MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(d1); \ - MP_INT__mp_alloc(mp_tmp2) = StgArrWords_words(d2); \ - MP_INT__mp_size(mp_tmp2) = (s2); \ - MP_INT__mp_d(mp_tmp2) = BYTE_ARR_CTS(d2); \ - \ - foreign "C" mpz_init(result1); \ - \ - /* Perform the operation */ \ - foreign "C" mp_fun(result1,mp_tmp1,mp_tmp2); \ - \ - RET_NP(MP_INT__mp_size(result1), \ - MP_INT__mp_d(result1) - SIZEOF_StgArrWords); \ +#define GMP_TAKE2_RET1(name,mp_fun) \ +name \ +{ \ + CInt s1, s2; \ + W_ d1, d2; \ + FETCH_MP_TEMP(mp_tmp1); \ + FETCH_MP_TEMP(mp_tmp2); \ + FETCH_MP_TEMP(mp_result1) \ + FETCH_MP_TEMP(mp_result2); \ + \ + /* call doYouWantToGC() */ \ + MAYBE_GC(R2_PTR & R4_PTR, name); \ + \ + s1 = W_TO_INT(R1); \ + d1 = R2; \ + s2 = W_TO_INT(R3); \ + d2 = R4; \ + \ + MP_INT__mp_alloc(mp_tmp1) = W_TO_INT(StgArrWords_words(d1)); \ + MP_INT__mp_size(mp_tmp1) = (s1); \ + MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(d1); \ + MP_INT__mp_alloc(mp_tmp2) = W_TO_INT(StgArrWords_words(d2)); \ + MP_INT__mp_size(mp_tmp2) = (s2); \ + MP_INT__mp_d(mp_tmp2) = BYTE_ARR_CTS(d2); \ + \ + foreign "C" mpz_init(mp_result1 "ptr") []; \ + \ + /* Perform the operation */ \ + 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); \ } -#define GMP_TAKE1_RET1(name,mp_fun) \ -name \ -{ \ - W_ s1, d1; \ - \ - /* call doYouWantToGC() */ \ - MAYBE_GC(R2_PTR, name); \ - \ - d1 = R2; \ - s1 = R1; \ - \ - MP_INT__mp_alloc(mp_tmp1) = StgArrWords_words(d1); \ - MP_INT__mp_size(mp_tmp1) = (s1); \ - MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(d1); \ - \ - foreign "C" mpz_init(result1); \ - \ - /* Perform the operation */ \ - foreign "C" mp_fun(result1,mp_tmp1); \ - \ - RET_NP(MP_INT__mp_size(result1), \ - MP_INT__mp_d(result1) - SIZEOF_StgArrWords); \ +#define GMP_TAKE1_RET1(name,mp_fun) \ +name \ +{ \ + CInt s1; \ + W_ d1; \ + FETCH_MP_TEMP(mp_tmp1); \ + FETCH_MP_TEMP(mp_result1) \ + \ + /* call doYouWantToGC() */ \ + MAYBE_GC(R2_PTR, name); \ + \ + d1 = R2; \ + s1 = W_TO_INT(R1); \ + \ + MP_INT__mp_alloc(mp_tmp1) = W_TO_INT(StgArrWords_words(d1)); \ + MP_INT__mp_size(mp_tmp1) = (s1); \ + MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(d1); \ + \ + foreign "C" mpz_init(mp_result1 "ptr") []; \ + \ + /* Perform the operation */ \ + 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); \ } -#define GMP_TAKE2_RET2(name,mp_fun) \ -name \ -{ \ - W_ s1, s2, d1, d2; \ - \ - /* call doYouWantToGC() */ \ - MAYBE_GC(R2_PTR & R4_PTR, name); \ - \ - s1 = R1; \ - d1 = R2; \ - s2 = R3; \ - d2 = R4; \ - \ - MP_INT__mp_alloc(mp_tmp1) = StgArrWords_words(d1); \ - MP_INT__mp_size(mp_tmp1) = (s1); \ - MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(d1); \ - MP_INT__mp_alloc(mp_tmp2) = StgArrWords_words(d2); \ - MP_INT__mp_size(mp_tmp2) = (s2); \ - MP_INT__mp_d(mp_tmp2) = BYTE_ARR_CTS(d2); \ - \ - foreign "C" mpz_init(result1); \ - foreign "C" mpz_init(result2); \ - \ - /* Perform the operation */ \ - foreign "C" mp_fun(result1,result2,mp_tmp1,mp_tmp2); \ - \ - RET_NPNP(MP_INT__mp_size(result1), \ - MP_INT__mp_d(result1) - SIZEOF_StgArrWords, \ - MP_INT__mp_size(result2), \ - MP_INT__mp_d(result2) - SIZEOF_StgArrWords); \ +#define GMP_TAKE2_RET2(name,mp_fun) \ +name \ +{ \ + CInt s1, s2; \ + W_ d1, d2; \ + FETCH_MP_TEMP(mp_tmp1); \ + FETCH_MP_TEMP(mp_tmp2); \ + FETCH_MP_TEMP(mp_result1) \ + FETCH_MP_TEMP(mp_result2) \ + \ + /* call doYouWantToGC() */ \ + MAYBE_GC(R2_PTR & R4_PTR, name); \ + \ + s1 = W_TO_INT(R1); \ + d1 = R2; \ + s2 = W_TO_INT(R3); \ + d2 = R4; \ + \ + MP_INT__mp_alloc(mp_tmp1) = W_TO_INT(StgArrWords_words(d1)); \ + MP_INT__mp_size(mp_tmp1) = (s1); \ + MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(d1); \ + MP_INT__mp_alloc(mp_tmp2) = W_TO_INT(StgArrWords_words(d2)); \ + 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") []; \ + \ + /* Perform the operation */ \ + 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, \ + TO_W_(MP_INT__mp_size(mp_result2)), \ + MP_INT__mp_d(mp_result2) - SIZEOF_StgArrWords); \ } GMP_TAKE2_RET1(plusIntegerzh_fast, mpz_add) @@ -634,17 +655,20 @@ GMP_TAKE1_RET1(complementIntegerzh_fast, mpz_com) GMP_TAKE2_RET2(quotRemIntegerzh_fast, mpz_tdiv_qr) GMP_TAKE2_RET2(divModIntegerzh_fast, mpz_fdiv_qr) +#ifndef THREADED_RTS section "bss" { - aa: W_; // NB. aa is really an mp_limb_t + mp_tmp_w: W_; // NB. mp_tmp_w is really an here mp_limb_t } +#endif gcdIntzh_fast { /* R1 = the first Int#; R2 = the second Int# */ W_ r; + FETCH_MP_TEMP(mp_tmp_w); - W_[aa] = R1; - r = foreign "C" mpn_gcd_1(aa, 1, R2); + W_[mp_tmp_w] = R1; + r = foreign "C" mpn_gcd_1(mp_tmp_w "ptr", 1, R2) []; R1 = r; /* Result parked in R1, return via info-pointer at TOS */ @@ -655,7 +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)); @@ -736,14 +760,14 @@ 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) { + if (cmp == 0 :: CInt) { R1 = 0; jump %ENTRY_CODE(Sp(0)); } - if (%lt(cmp,0) == %lt(usize,0)) { + if (%lt(cmp,0 :: CInt) == %lt(usize,0)) { R1 = 1; } else { R1 = (-1); @@ -790,14 +814,12 @@ integer2Wordzh_fast jump %ENTRY_CODE(Sp(0)); } -section "bss" { - exponent: W_; -} - decodeFloatzh_fast { W_ p; F_ arg; + FETCH_MP_TEMP(mp_tmp1); + FETCH_MP_TEMP(mp_tmp_w); /* arguments: F1 = Float# */ arg = F1; @@ -812,10 +834,10 @@ decodeFloatzh_fast MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(p); /* Perform the operation */ - foreign "C" __decodeFloat(mp_tmp1,exponent,arg); + foreign "C" __decodeFloat(mp_tmp1 "ptr",mp_tmp_w "ptr" ,arg) []; /* returns: (Int# (expn), Int#, ByteArray#) */ - RET_NNP(W_[exponent], MP_INT__mp_size(mp_tmp1), p); + RET_NNP(W_[mp_tmp_w], TO_W_(MP_INT__mp_size(mp_tmp1)), p); } #define DOUBLE_MANTISSA_SIZE SIZEOF_DOUBLE @@ -825,6 +847,8 @@ decodeDoublezh_fast { D_ arg; W_ p; + FETCH_MP_TEMP(mp_tmp1); + FETCH_MP_TEMP(mp_tmp_w); /* arguments: D1 = Double# */ arg = D1; @@ -839,10 +863,10 @@ decodeDoublezh_fast MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(p); /* Perform the operation */ - foreign "C" __decodeDouble(mp_tmp1,exponent,arg); + foreign "C" __decodeDouble(mp_tmp1 "ptr", mp_tmp_w "ptr",arg) []; /* returns: (Int# (expn), Int#, ByteArray#) */ - RET_NNP(W_[exponent], MP_INT__mp_size(mp_tmp1), p); + RET_NNP(W_[mp_tmp_w], TO_W_(MP_INT__mp_size(mp_tmp1)), p); } /* ----------------------------------------------------------------------------- @@ -856,12 +880,13 @@ forkzh_fast 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"); + "ptr" R1 = foreign "C" createIOThread( MyCapability() "ptr", + RtsFlags_GcFlags_initialStkSize(RtsFlags), + R1 "ptr") [R1]; + foreign "C" scheduleThread(MyCapability() "ptr", R1 "ptr") [R1]; // switch at the earliest opportunity - CInt[context_switch] = 1; + CInt[context_switch] = 1 :: CInt; RET_P(R1); } @@ -883,7 +908,7 @@ 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)); } @@ -892,10 +917,468 @@ isCurrentThreadBoundzh_fast { /* no args */ W_ r; - r = foreign "C" isThreadBound(CurrentTSO); + r = foreign "C" isThreadBound(CurrentTSO) []; RET_N(r); } + +/* ----------------------------------------------------------------------------- + * 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 + +#if defined(PROFILING) +#define CATCH_RETRY_FRAME_BITMAP 7 +#define CATCH_RETRY_FRAME_WORDS 6 +#else +#define CATCH_RETRY_FRAME_BITMAP 1 +#define CATCH_RETRY_FRAME_WORDS 4 +#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_ 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") []; + if (r) { + /* 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: retry */ + W_ new_trec; + "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); + } else { + R1 = StgCatchRetryFrame_first_code(frame); + StgCatchRetryFrame_first_code_trec(frame) = new_trec; + } + jump stg_ap_v_fast; + } +} + + +// 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 + +#if defined(PROFILING) +#define ATOMICALLY_FRAME_BITMAP 3 +#define ATOMICALLY_FRAME_WORDS 3 +#else +#define ATOMICALLY_FRAME_BITMAP 0 +#define ATOMICALLY_FRAME_WORDS 1 +#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) +{ + W_ frame, trec, valid; + IF_NOT_REG_R1(W_ rval; rval = Sp(0); Sp_adj(1); ) + + 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(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) \ + label \ + { \ + IF_NOT_REG_R1(W_ rval; rval = Sp(0); Sp_adj(1); ) \ + Sp = Sp + SIZEOF_StgCatchSTMFrame; \ + IF_NOT_REG_R1(Sp_adj(-1); Sp(0) = rval;) \ + jump ret; \ + } + +#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 3 +#else +#define CATCH_STM_FRAME_BITMAP 0 +#define CATCH_STM_FRAME_WORDS 1 +#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))) + + +// Primop definition ------------------------------------------------------------ + +atomicallyzh_fast +{ + W_ frame; + W_ old_trec; + W_ new_trec; + + // stmStartTransaction may allocate + MAYBE_GC (R1_PTR, 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_code(frame) = R1; + + /* Start the memory transcation */ + "ptr" new_trec = foreign "C" stmStartTransaction(MyCapability() "ptr", old_trec "ptr") [R1]; + StgTSO_trec(CurrentTSO) = new_trec; + + /* Apply R1 to the realworld token */ + jump stg_ap_v_fast; +} + + +catchSTMzh_fast +{ + W_ frame; + + /* Args: R1 :: STM a */ + /* Args: R2 :: Exception -> STM a */ + STK_CHK_GEN(SIZEOF_StgCatchSTMFrame + WDS(1), R1_PTR & R2_PTR, catchSTMzh_fast); + + /* Set up the catch frame */ + Sp = Sp - SIZEOF_StgCatchSTMFrame; + frame = Sp; + + SET_HDR(frame, stg_catch_stm_frame_info, W_[CCCS]); + StgCatchSTMFrame_handler(frame) = R2; + + /* Apply R1 to the realworld token */ + jump stg_ap_v_fast; +} + + +catchRetryzh_fast +{ + W_ frame; + W_ new_trec; + W_ trec; + + // stmStartTransaction may allocate + MAYBE_GC (R1_PTR & R2_PTR, catchRetryzh_fast); + + /* Args: R1 :: STM a */ + /* Args: R2 :: STM a */ + STK_CHK_GEN(SIZEOF_StgCatchRetryFrame + WDS(1), R1_PTR & R2_PTR, 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]; + StgTSO_trec(CurrentTSO) = new_trec; + + /* Set up the catch-retry frame */ + Sp = Sp - SIZEOF_StgCatchRetryFrame; + frame = Sp; + + SET_HDR(frame, stg_catch_retry_frame_info, W_[CCCS]); + StgCatchRetryFrame_running_alt_code(frame) = 0 :: CInt; // false; + StgCatchRetryFrame_first_code(frame) = R1; + StgCatchRetryFrame_alt_code(frame) = R2; + StgCatchRetryFrame_first_code_trec(frame) = new_trec; + + /* Apply R1 to the realworld token */ + jump stg_ap_v_fast; +} + + +retryzh_fast +{ + W_ frame_type; + W_ frame; + W_ trec; + W_ outer; + W_ r; + + MAYBE_GC (NO_PTRS, retryzh_fast); // STM operations may allocate + + // Find the enclosing ATOMICALLY_FRAME or CATCH_RETRY_FRAME +retry_pop_stack: + trec = StgTSO_trec(CurrentTSO); + "ptr" outer = foreign "C" stmGetEnclosingTRec(trec "ptr") []; + StgTSO_sp(CurrentTSO) = Sp; + frame_type = foreign "C" findRetryFrameHelper(CurrentTSO "ptr") []; + Sp = StgTSO_sp(CurrentTSO); + frame = Sp; + + if (frame_type == CATCH_RETRY_FRAME) { + // The retry reaches a CATCH_RETRY_FRAME before the atomic frame + ASSERT(outer != NO_TREC); + if (!StgCatchRetryFrame_running_alt_code(frame)) { + // Retry in the first code: try the alternative + "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); + 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(MyCapability() "ptr", other_trec "ptr") []; + if (r) { + 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 + // and then propagate the retry + StgTSO_trec(CurrentTSO) = outer; + Sp = Sp + SIZEOF_StgCatchRetryFrame; + goto retry_pop_stack; + } else { + // Merge failed: we musn't propagate the retry. Try both paths again. + "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); + jump stg_ap_v_fast; + } + } + } + + // We've reached the ATOMICALLY_FRAME: attempt to wait + ASSERT(frame_type == ATOMICALLY_FRAME); + ASSERT(outer == NO_TREC); + 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 + 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") []; + StgTSO_trec(CurrentTSO) = trec; + R1 = StgAtomicallyFrame_code(frame); + Sp = frame; + jump stg_ap_v_fast; + } +} + + +newTVarzh_fast +{ + W_ tv; + W_ new_value; + + /* Args: R1 = initialisation value */ + + MAYBE_GC (R1_PTR, newTVarzh_fast); + new_value = R1; + "ptr" tv = foreign "C" stmNewTVar(MyCapability() "ptr", new_value "ptr") []; + RET_P(tv); +} + + +readTVarzh_fast +{ + W_ trec; + W_ tvar; + W_ result; + + /* Args: R1 = TVar closure */ + + 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") []; + + RET_P(result); +} + + +writeTVarzh_fast +{ + W_ trec; + W_ tvar; + W_ new_value; + + /* Args: R1 = TVar closure */ + /* R2 = New value */ + + MAYBE_GC (R1_PTR & R2_PTR, writeTVarzh_fast); // Call to stmWriteTVar may allocate + trec = StgTSO_trec(CurrentTSO); + tvar = R1; + new_value = R2; + foreign "C" stmWriteTVar(MyCapability() "ptr", trec "ptr", tvar "ptr", new_value "ptr") []; + + jump %ENTRY_CODE(Sp(0)); +} + + /* ----------------------------------------------------------------------------- * MVar primitives * @@ -970,7 +1453,6 @@ newMVarzh_fast StgTSO_sp(tso) = StgTSO_sp(tso) + WDS(3); \ lval = W_[StgTSO_sp(tso) - WDS(1)]; - takeMVarzh_fast { W_ mvar, val, info, tso; @@ -978,7 +1460,11 @@ takeMVarzh_fast /* args: R1 = MVar closure */ mvar = R1; +#if defined(THREADED_RTS) + "ptr" info = foreign "C" lockClosure(mvar "ptr") []; +#else info = GET_INFO(mvar); +#endif /* If the MVar is empty, put ourselves on its blocking queue, * and wait until we're woken up. @@ -1010,29 +1496,38 @@ 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 + 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) []; +#endif RET_P(val); } else { /* No further putMVars, MVar is now empty */ - - /* do this last... we might have locked the MVar in the SMP case, - * and writing the info pointer will unlock it. - */ - SET_INFO(mvar,stg_EMPTY_MVAR_info); StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure; + +#if defined(THREADED_RTS) + foreign "C" unlockClosure(mvar "ptr", stg_EMPTY_MVAR_info) []; +#else + SET_INFO(mvar,stg_EMPTY_MVAR_info); +#endif + RET_P(val); } } @@ -1046,9 +1541,16 @@ tryTakeMVarzh_fast mvar = R1; +#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(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 */ @@ -1059,6 +1561,7 @@ tryTakeMVarzh_fast val = StgMVar_value(mvar); if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure) { + /* There are putMVar(s) waiting... * wake up the first thread on the queue */ @@ -1067,29 +1570,34 @@ 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(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; - - /* do this last... we might have locked the MVar in the SMP case, - * and writing the info pointer will unlock it. - */ +#if defined(THREADED_RTS) + foreign "C" unlockClosure(mvar "ptr", stg_EMPTY_MVAR_info) []; +#else SET_INFO(mvar,stg_EMPTY_MVAR_info); +#endif } RET_NP(1, val); @@ -1103,7 +1611,11 @@ putMVarzh_fast /* args: R1 = MVar, R2 = value */ mvar = R1; +#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 (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) { @@ -1120,6 +1632,7 @@ putMVarzh_fast } if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure) { + /* There are takeMVar(s) waiting: wake up the first one */ ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16); @@ -1127,13 +1640,14 @@ 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 @@ -1141,14 +1655,21 @@ putMVarzh_fast StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure; } +#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; - /* unlocks the MVar in the SMP case */ + +#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)); } @@ -1163,13 +1684,21 @@ tryPutMVarzh_fast /* args: R1 = MVar, R2 = value */ mvar = R1; +#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(THREADED_RTS) + foreign "C" unlockClosure(mvar "ptr", stg_FULL_MVAR_info) []; +#endif RET_N(0); } if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure) { + /* There are takeMVar(s) waiting: wake up the first one */ ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16); @@ -1177,13 +1706,14 @@ 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 @@ -1191,17 +1721,23 @@ tryPutMVarzh_fast StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure; } - jump %ENTRY_CODE(Sp(0)); +#if defined(THREADED_RTS) + foreign "C" unlockClosure(mvar "ptr", stg_EMPTY_MVAR_info) []; +#endif } else { /* No further takes, the MVar is now full. */ StgMVar_value(mvar) = R2; - /* unlocks the MVar in the SMP case */ + +#if defined(THREADED_RTS) + foreign "C" unlockClosure(mvar "ptr", stg_FULL_MVAR_info) []; +#else SET_INFO(mvar,stg_FULL_MVAR_info); - jump %ENTRY_CODE(Sp(0)); +#endif } + RET_N(1); /* ToDo: yield afterward for better communication performance? */ } @@ -1216,18 +1752,18 @@ 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 an array of snEntry structs. + * stable_ptr_table is a pointer to an array of snEntry structs. */ - if ( snEntry_sn_obj(stable_ptr_table + index*SIZEOF_snEntry) == NULL ) { + if ( snEntry_sn_obj(W_[stable_ptr_table] + index*SIZEOF_snEntry) == NULL ) { sn_obj = Hp - SIZEOF_StgStableName + WDS(1); SET_HDR(sn_obj, stg_STABLE_NAME_info, W_[CCCS]); StgStableName_sn(sn_obj) = index; - snEntry_sn_obj(stable_ptr_table + index*SIZEOF_snEntry) = sn_obj; + snEntry_sn_obj(W_[stable_ptr_table] + index*SIZEOF_snEntry) = sn_obj; } else { - sn_obj = snEntry_sn_obj(stable_ptr_table + index*SIZEOF_snEntry); + sn_obj = snEntry_sn_obj(W_[stable_ptr_table] + index*SIZEOF_snEntry); } RET_P(sn_obj); @@ -1239,7 +1775,7 @@ 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); } @@ -1248,7 +1784,7 @@ deRefStablePtrzh_fast /* Args: R1 = the stable ptr */ W_ r, sp; sp = R1; - r = snEntry_addr(stable_ptr_table + sp*SIZEOF_snEntry); + r = snEntry_addr(W_[stable_ptr_table] + sp*SIZEOF_snEntry); RET_P(r); } @@ -1306,8 +1842,8 @@ mkApUpd0zh_fast // This function is *only* used to wrap zero-arity BCOs in an // updatable wrapper (see ByteCodeLink.lhs). An AP thunk is always // saturated and always points directly to a FUN or BCO. - ASSERT(%INFO_TYPE(%GET_STD_INFO(R1)) == BCO::I16 && - StgBCO_arity(R1) == 0::I16); + ASSERT(%INFO_TYPE(%GET_STD_INFO(R1)) == HALF_W_(BCO) && + StgBCO_arity(R1) == HALF_W_(0)); HP_CHK_GEN_TICKY(SIZEOF_StgAP, R1_PTR, mkApUpd0zh_fast); TICK_ALLOC_UP_THK(0, 0); @@ -1316,7 +1852,7 @@ mkApUpd0zh_fast ap = Hp - SIZEOF_StgAP + WDS(1); SET_HDR(ap, stg_AP_info, W_[CCCS]); - StgAP_n_args(ap) = 0::I16; + StgAP_n_args(ap) = HALF_W_(0); StgAP_fun(ap) = R1; RET_P(ap); @@ -1341,6 +1877,10 @@ mkApUpd0zh_fast waitReadzh_fast { /* args: R1 */ +#ifdef THREADED_RTS + foreign "C" barf("waitRead# on threaded RTS"); +#else + ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16); StgTSO_why_blocked(CurrentTSO) = BlockedOnRead::I16; StgTSO_block_info(CurrentTSO) = R1; @@ -1348,11 +1888,16 @@ waitReadzh_fast // threaded RTS anyway. APPEND_TO_BLOCKED_QUEUE(CurrentTSO); jump stg_block_noregs; +#endif } waitWritezh_fast { /* args: R1 */ +#ifdef THREADED_RTS + foreign "C" barf("waitWrite# on threaded RTS"); +#else + ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16); StgTSO_why_blocked(CurrentTSO) = BlockedOnWrite::I16; StgTSO_block_info(CurrentTSO) = R1; @@ -1360,24 +1905,29 @@ waitWritezh_fast // threaded RTS anyway. APPEND_TO_BLOCKED_QUEUE(CurrentTSO); jump stg_block_noregs; +#endif } STRING(stg_delayzh_malloc_str, "delayzh_fast") delayzh_fast { -#ifdef mingw32_TARGET_OS +#ifdef mingw32_HOST_OS W_ ares; CInt reqID; #else W_ t, prev, target; #endif +#ifdef THREADED_RTS + foreign "C" barf("delay# on threaded RTS"); +#else + /* args: R1 (microsecond delay amount) */ ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16); StgTSO_why_blocked(CurrentTSO) = BlockedOnDelay::I16; -#ifdef mingw32_TARGET_OS +#ifdef mingw32_HOST_OS /* could probably allocate this on the heap instead */ "ptr" ares = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult, @@ -1394,12 +1944,13 @@ delayzh_fast */ StgTSO_why_blocked(CurrentTSO) = BlockedOnDoProc::I16; APPEND_TO_BLOCKED_QUEUE(CurrentTSO); + jump stg_block_async_void; #else - CInt time; + W_ time; time = foreign "C" getourtimeofday(); - target = (R1 / (TICK_MILLISECS*1000)) + TO_W_(time); + target = (R1 / (TICK_MILLISECS*1000)) + time; StgTSO_block_info(CurrentTSO) = target; /* Insert the new thread in the sleeping queue. */ @@ -1418,33 +1969,39 @@ while: } else { StgTSO_link(prev) = CurrentTSO; } -#endif - jump stg_block_noregs; +#endif +#endif /* !THREADED_RTS */ } -#ifdef mingw32_TARGET_OS +#ifdef mingw32_HOST_OS STRING(stg_asyncReadzh_malloc_str, "asyncReadzh_fast") asyncReadzh_fast { W_ ares; CInt reqID; +#ifdef THREADED_RTS + foreign "C" barf("asyncRead# on threaded RTS"); +#else + /* args: R1 = fd, R2 = isSock, R3 = len, R4 = buf */ ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16); StgTSO_why_blocked(CurrentTSO) = BlockedOnRead::I16; /* 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") @@ -1453,13 +2010,18 @@ asyncWritezh_fast W_ ares; CInt reqID; +#ifdef THREADED_RTS + foreign "C" barf("asyncWrite# on threaded RTS"); +#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; @@ -1467,6 +2029,7 @@ asyncWritezh_fast StgTSO_block_info(CurrentTSO) = ares; APPEND_TO_BLOCKED_QUEUE(CurrentTSO); jump stg_block_async; +#endif } STRING(stg_asyncDoProczh_malloc_str, "asyncDoProczh_fast") @@ -1475,20 +2038,26 @@ 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