X-Git-Url: http://git.megacz.com/?a=blobdiff_plain;f=rts%2FPrimOps.cmm;h=272c7050016742ae5159ae13911cfcb4ac16c58f;hb=73e88c2a30cbb2fc9cf8c394c620c0f3edcdd0eb;hp=72090c8fb0f4cd0930d972cb417e04c61ed635a6;hpb=03727aa00ed928ae5531a43bb0ae9081f31eb1a8;p=ghc-hetmet.git diff --git a/rts/PrimOps.cmm b/rts/PrimOps.cmm index 72090c8..272c705 100644 --- a/rts/PrimOps.cmm +++ b/rts/PrimOps.cmm @@ -83,6 +83,9 @@ newByteArrayzh_fast RET_P(p); } +#define BA_ALIGN 16 +#define BA_MASK (BA_ALIGN-1) + newPinnedByteArrayzh_fast { W_ words, payload_words, n, p; @@ -91,24 +94,45 @@ newPinnedByteArrayzh_fast n = R1; payload_words = ROUNDUP_BYTES_TO_WDS(n); - // We want an 8-byte aligned array. allocatePinned() gives us + words = payload_words + ((SIZEOF_StgArrWords + BA_MASK) & ~BA_MASK); + + ("ptr" p) = foreign "C" allocatePinned(words) []; + TICK_ALLOC_PRIM(SIZEOF_StgArrWords,WDS(payload_words),0); + + // This bumps p forwards so that the payload falls on an R2-byte boundary. + p = p + ((-p - SIZEOF_StgArrWords) & BA_MASK); + + SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]); + StgArrWords_words(p) = payload_words; + RET_P(p); +} + +newAlignedPinnedByteArrayzh_fast +{ + W_ words, payload_words, n, p, mask; + + MAYBE_GC(NO_PTRS,newAlignedPinnedByteArrayzh_fast); + n = R1; + + if (R2 > SIZEOF_W) { + mask = R2 - 1; + } else { + mask = 0; + } + + payload_words = ROUNDUP_BYTES_TO_WDS(n); + + // We want an -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; - } + words = payload_words + ((SIZEOF_StgArrWords + mask) & ~mask); ("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); - } + // This bumps p forwards so that the payload falls on an R2-byte boundary. + p = p + ((-p - SIZEOF_StgArrWords) & mask); SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]); StgArrWords_words(p) = payload_words; @@ -297,9 +321,14 @@ 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; @@ -309,12 +338,65 @@ 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; @@ -342,9 +424,18 @@ 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); @@ -1084,7 +1175,7 @@ INFO_TABLE_RET(stg_catch_retry_frame, CATCH_RETRY_FRAME, #if defined(PROFILING) W_ unused1, W_ unused2, #endif - W_ unused3, "ptr" W_ unused4, "ptr" W_ unused5) + W_ unused3, P_ unused4, P_ unused5) { W_ r, frame, trec, outer; @@ -1118,7 +1209,7 @@ INFO_TABLE_RET(stg_atomically_frame, ATOMICALLY_FRAME, #if defined(PROFILING) W_ unused1, W_ unused2, #endif - "ptr" W_ unused3, "ptr" W_ unused4) + P_ unused3, P_ unused4) { W_ frame, trec, valid, next_invariant, q, outer; @@ -1180,7 +1271,7 @@ INFO_TABLE_RET(stg_atomically_waiting_frame, ATOMICALLY_FRAME, #if defined(PROFILING) W_ unused1, W_ unused2, #endif - "ptr" W_ unused3, "ptr" W_ unused4) + P_ unused3, P_ unused4) { W_ frame, trec, valid; @@ -1214,7 +1305,7 @@ INFO_TABLE_RET(stg_catch_stm_frame, CATCH_STM_FRAME, #if defined(PROFILING) W_ unused1, W_ unused2, #endif - "ptr" W_ unused3, "ptr" W_ unused4) + P_ unused3, P_ unused4) { W_ r, frame, trec, outer; frame = Sp; @@ -1584,8 +1675,11 @@ takeMVarzh_fast CurrentTSO) []; } StgTSO__link(CurrentTSO) = stg_END_TSO_QUEUE_closure; - StgTSO_why_blocked(CurrentTSO) = BlockedOnMVar::I16; 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; @@ -1741,8 +1835,11 @@ putMVarzh_fast CurrentTSO) []; } StgTSO__link(CurrentTSO) = stg_END_TSO_QUEUE_closure; - StgTSO_why_blocked(CurrentTSO) = BlockedOnMVar::I16; 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; @@ -2289,6 +2386,26 @@ getApStackValzh_fast RET_NP(ok,val); } +/* ----------------------------------------------------------------------------- + Misc. primitives + -------------------------------------------------------------------------- */ + +// 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; + +#ifdef PROFILING + ccs = StgHeader_ccs(UNTAG(R1)); + foreign "C" fprintCCS_stderr(ccs "ptr") [R2]; +#endif + + R1 = R2; + ENTER(); +} + getSparkzh_fast { W_ spark;