X-Git-Url: http://git.megacz.com/?a=blobdiff_plain;f=ghc%2Frts%2FPrimOps.hc;h=0196e213f88cb9f5b100ae24c487dc1d8de2453b;hb=97906cfcc30dd591e840921d336fdabeb1b8a315;hp=364e20ab6f8150168ced04fcbf5656ba0d081c06;hpb=52c078343e70a170441fed71480ba8569475c4cd;p=ghc-hetmet.git diff --git a/ghc/rts/PrimOps.hc b/ghc/rts/PrimOps.hc index 364e20a..0196e21 100644 --- a/ghc/rts/PrimOps.hc +++ b/ghc/rts/PrimOps.hc @@ -1,5 +1,5 @@ /* ----------------------------------------------------------------------------- - * $Id: PrimOps.hc,v 1.83 2001/08/08 10:50:37 simonmar Exp $ + * $Id: PrimOps.hc,v 1.100 2002/07/17 09:21:50 simonmar Exp $ * * (c) The GHC Team, 1998-2000 * @@ -18,11 +18,16 @@ #include "Storage.h" #include "BlockAlloc.h" /* tmp */ #include "StablePriv.h" -#include "HeapStackCheck.h" #include "StgRun.h" #include "Itimer.h" #include "Prelude.h" +#ifdef HAVE_SYS_TYPES_H +# include +#endif + +#include + /* ** temporary ** classes CCallable and CReturnable don't really exist, but the @@ -245,40 +250,57 @@ StgWord GHC_ZCCReturnable_static_info[1]; #define BYTES_TO_STGWORDS(n) ((n) + sizeof(W_) - 1)/sizeof(W_) -FN_(newByteArrayzh_fast) \ - { \ - W_ size, stuff_size, n; \ - StgArrWords* p; \ - FB_ \ - MAYBE_GC(NO_PTRS,newByteArrayzh_fast); \ - n = R1.w; \ - stuff_size = BYTES_TO_STGWORDS(n); \ - size = sizeofW(StgArrWords)+ stuff_size; \ - p = (StgArrWords *)RET_STGCALL1(P_,allocate,size); \ - TICK_ALLOC_PRIM(sizeofW(StgArrWords),stuff_size,0); \ - SET_HDR(p, &stg_ARR_WORDS_info, CCCS); \ - p->words = stuff_size; \ - TICK_RET_UNBOXED_TUP(1) \ - RET_P(p); \ - FE_ \ +FN_(newByteArrayzh_fast) + { + W_ size, stuff_size, n; + StgArrWords* p; + FB_ + MAYBE_GC(NO_PTRS,newByteArrayzh_fast); + n = R1.w; + stuff_size = BYTES_TO_STGWORDS(n); + size = sizeofW(StgArrWords)+ stuff_size; + p = (StgArrWords *)RET_STGCALL1(P_,allocate,size); + TICK_ALLOC_PRIM(sizeofW(StgArrWords),stuff_size,0); + SET_HDR(p, &stg_ARR_WORDS_info, CCCS); + p->words = stuff_size; + TICK_RET_UNBOXED_TUP(1) + RET_P(p); + FE_ } -FN_(newPinnedByteArrayzh_fast) \ - { \ - W_ size, stuff_size, n; \ - StgArrWords* p; \ - FB_ \ - MAYBE_GC(NO_PTRS,newPinnedByteArrayzh_fast); \ - n = R1.w; \ - stuff_size = BYTES_TO_STGWORDS(n); \ - size = sizeofW(StgArrWords)+ stuff_size; \ - p = (StgArrWords *)RET_STGCALL1(P_,allocatePinned,size); \ - TICK_ALLOC_PRIM(sizeofW(StgArrWords),stuff_size,0); \ - SET_HDR(p, &stg_ARR_WORDS_info, CCCS); \ - p->words = stuff_size; \ - TICK_RET_UNBOXED_TUP(1) \ - RET_P(p); \ - FE_ \ +FN_(newPinnedByteArrayzh_fast) + { + W_ size, stuff_size, n; + StgArrWords* p; + FB_ + MAYBE_GC(NO_PTRS,newPinnedByteArrayzh_fast); + n = R1.w; + stuff_size = BYTES_TO_STGWORDS(n); + + // 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. + size = sizeofW(StgArrWords)+ stuff_size; + if ((sizeof(StgArrWords) & 7) != 0) { + size++; + } + + p = (StgArrWords *)RET_STGCALL1(P_,allocatePinned,size); + TICK_ALLOC_PRIM(sizeofW(StgArrWords),stuff_size,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) { + ((StgPtr)p)++; + } + + SET_HDR(p, &stg_ARR_WORDS_info, CCCS); + p->words = stuff_size; + TICK_RET_UNBOXED_TUP(1) + RET_P(p); + FE_ } FN_(newArrayzh_fast) @@ -330,7 +352,6 @@ FN_(newMutVarzh_fast) /* ----------------------------------------------------------------------------- Foreign Object Primitives - -------------------------------------------------------------------------- */ FN_(mkForeignObjzh_fast) @@ -421,7 +442,25 @@ FN_(finalizzeWeakzh_fast) } /* kill it */ +#ifdef PROFILING + // @LDV profiling + // A weak pointer is inherently used, so we do not need to call + // LDV_recordDead_FILL_SLOP_DYNAMIC(): + // LDV_recordDead_FILL_SLOP_DYNAMIC((StgClosure *)w); + // or, LDV_recordDead(): + // LDV_recordDead((StgClosure *)w, sizeofW(StgWeak) - sizeofW(StgProfHeader)); + // Furthermore, when PROFILING is turned on, dead weak pointers are exactly as + // large as weak pointers, so there is no need to fill the slop, either. + // See stg_DEAD_WEAK_info in StgMiscClosures.hc. +#endif + // + // Todo: maybe use SET_HDR() and remove LDV_recordCreate()? + // w->header.info = &stg_DEAD_WEAK_info; +#ifdef PROFILING + // @LDV profiling + LDV_recordCreate((StgClosure *)w); +#endif f = ((StgWeak *)w)->finalizer; w->link = ((StgWeak *)w)->link; @@ -434,6 +473,25 @@ FN_(finalizzeWeakzh_fast) FE_ } +FN_(deRefWeakzh_fast) +{ + /* R1.p = weak ptr */ + StgWeak* w; + I_ code; + P_ val; + FB_ + w = (StgWeak*)R1.p; + if (w->header.info == &stg_WEAK_info) { + code = 1; + val = (P_)((StgWeak *)w)->value; + } else { + code = 0; + val = (P_)w; + } + RET_NP(code,val); + FE_ +} + /* ----------------------------------------------------------------------------- Arbitrary-precision Integer operations. -------------------------------------------------------------------------- */ @@ -734,6 +792,161 @@ GMP_TAKE1_RET1(complementIntegerzh_fast, mpz_com); GMP_TAKE2_RET2(quotRemIntegerzh_fast, mpz_tdiv_qr); GMP_TAKE2_RET2(divModIntegerzh_fast, mpz_fdiv_qr); + +FN_(gcdIntzh_fast) +{ + /* R1 = the first Int#; R2 = the second Int# */ + mp_limb_t aa; + I_ r; + FB_ + aa = (mp_limb_t)(R1.i); + r = RET_STGCALL3(StgInt, mpn_gcd_1, (mp_limb_t *)(&aa), 1, (mp_limb_t)(R2.i)); + + R1.i = r; + /* Result parked in R1, return via info-pointer at TOS */ + JMP_(ENTRY_CODE(Sp[0])); + FE_ +} + +FN_(gcdIntegerIntzh_fast) +{ + /* R1 = s1; R2 = d1; R3 = the int */ + I_ r; + FB_ + r = RET_STGCALL3(StgInt,mpn_gcd_1,(mp_limb_t *)(BYTE_ARR_CTS(R2.p)), R1.i, R3.i); + + R1.i = r; + /* Result parked in R1, return via info-pointer at TOS */ + JMP_(ENTRY_CODE(Sp[0])); + FE_ +} + +FN_(cmpIntegerIntzh_fast) +{ + /* R1 = s1; R2 = d1; R3 = the int */ + I_ usize; + I_ vsize; + I_ v_digit; + mp_limb_t u_digit; + FB_ + + usize = R1.i; + vsize = 0; + v_digit = R3.i; + + // paraphrased from mpz_cmp_si() in the GMP sources + if (v_digit > 0) { + vsize = 1; + } else if (v_digit < 0) { + vsize = -1; + v_digit = -v_digit; + } + + if (usize != vsize) { + R1.i = usize - vsize; JMP_(ENTRY_CODE(Sp[0])); + } + + if (usize == 0) { + R1.i = 0; JMP_(ENTRY_CODE(Sp[0])); + } + + u_digit = *(mp_limb_t *)(BYTE_ARR_CTS(R2.p)); + + if (u_digit == (mp_limb_t) (unsigned long) v_digit) { + R1.i = 0; JMP_(ENTRY_CODE(Sp[0])); + } + + if (u_digit > (mp_limb_t) (unsigned long) v_digit) { + R1.i = usize; + } else { + R1.i = -usize; + } + + JMP_(ENTRY_CODE(Sp[0])); + FE_ +} + +FN_(cmpIntegerzh_fast) +{ + /* R1 = s1; R2 = d1; R3 = s2; R4 = d2 */ + I_ usize; + I_ vsize; + I_ size; + StgPtr up, vp; + int cmp; + FB_ + + // paraphrased from mpz_cmp() in the GMP sources + usize = R1.i; + vsize = R3.i; + + if (usize != vsize) { + R1.i = usize - vsize; JMP_(ENTRY_CODE(Sp[0])); + } + + if (usize == 0) { + R1.i = 0; JMP_(ENTRY_CODE(Sp[0])); + } + + size = abs(usize); + + up = BYTE_ARR_CTS(R2.p); + vp = BYTE_ARR_CTS(R4.p); + + cmp = RET_STGCALL3(I_, mpn_cmp, (mp_limb_t *)up, (mp_limb_t *)vp, size); + + if (cmp == 0) { + R1.i = 0; JMP_(ENTRY_CODE(Sp[0])); + } + + if ((cmp < 0) == (usize < 0)) { + R1.i = 1; + } else { + R1.i = (-1); + } + /* Result parked in R1, return via info-pointer at TOS */ + JMP_(ENTRY_CODE(Sp[0])); + FE_ +} + +FN_(integer2Intzh_fast) +{ + /* R1 = s; R2 = d */ + I_ r, s; + FB_ + s = R1.i; + if (s == 0) + r = 0; + else { + r = ((mp_limb_t *) (BYTE_ARR_CTS(R2.p)))[0]; + if (s < 0) r = -r; + } + /* Result parked in R1, return via info-pointer at TOS */ + R1.i = r; + JMP_(ENTRY_CODE(Sp[0])); + FE_ +} + +FN_(integer2Wordzh_fast) +{ + /* R1 = s; R2 = d */ + I_ s; + W_ r; + FB_ + s = R1.i; + if (s == 0) + r = 0; + else { + r = ((mp_limb_t *) (BYTE_ARR_CTS(R2.p)))[0]; + if (s < 0) r = -r; + } + /* Result parked in R1, return via info-pointer at TOS */ + R1.w = r; + JMP_(ENTRY_CODE(Sp[0])); + FE_ +} + + FN_(decodeFloatzh_fast) { MP_INT mantissa; @@ -809,13 +1022,28 @@ FN_(forkzh_fast) /* create it right now, return ThreadID in R1 */ R1.t = RET_STGCALL2(StgTSO *, createIOThread, - RtsFlags.GcFlags.initialStkSize, R1.cl); + RtsFlags.GcFlags.initialStkSize, R1.cl); STGCALL1(scheduleThread, R1.t); /* switch at the earliest opportunity */ context_switch = 1; + RET_P(R1.t); + FE_ +} + +FN_(forkProcesszh_fast) +{ + pid_t pid; + + FB_ + /* args: none */ + /* result: Pid */ + + R1.i = RET_STGCALL1(StgInt, forkProcess, CurrentTSO); + JMP_(ENTRY_CODE(Sp[0])); + FE_ } @@ -826,6 +1054,28 @@ FN_(yieldzh_fast) FE_ } +FN_(myThreadIdzh_fast) +{ + /* no args. */ + FB_ + RET_P((P_)CurrentTSO); + FE_ +} + +FN_(labelThreadzh_fast) +{ + FB_ + /* args: + R1.p = ThreadId# + R2.p = Addr# */ +#ifdef DEBUG + STGCALL2(labelThread,(StgTSO *)R1.p,(char *)R2.p); +#endif + JMP_(ENTRY_CODE(Sp[0])); + FE_ +} + + /* ----------------------------------------------------------------------------- * MVar primitives * @@ -858,6 +1108,17 @@ FN_(yieldzh_fast) * * -------------------------------------------------------------------------- */ +FN_(isEmptyMVarzh_fast) +{ + /* args: R1 = MVar closure */ + I_ r; + FB_ + r = (I_)((GET_INFO((StgMVar*)(R1.p))) == &stg_EMPTY_MVAR_info); + RET_N(r); + FE_ +} + + FN_(newMVarzh_fast) { StgMVar *mvar; @@ -880,14 +1141,23 @@ FN_(newMVarzh_fast) FE_ } -#define PerformTake(tso, value) ({ \ - (tso)->sp[1] = (W_)value; \ - (tso)->sp[0] = (W_)&stg_gc_unpt_r1_ret_info; \ +/* If R1 isn't available, pass it on the stack */ +#ifdef REG_R1 +#define PerformTake(tso, value) ({ \ + (tso)->sp[1] = (W_)value; \ + (tso)->sp[0] = (W_)&stg_gc_unpt_r1_info; \ + }) +#else +#define PerformTake(tso, value) ({ \ + (tso)->sp[1] = (W_)value; \ + (tso)->sp[0] = (W_)&stg_ut_1_0_unreg_info; \ }) +#endif + #define PerformPut(tso) ({ \ StgClosure *val = (StgClosure *)(tso)->sp[2]; \ - (tso)->sp[2] = (W_)&stg_gc_noregs_ret_info; \ + (tso)->sp[2] = (W_)&stg_gc_noregs_info; \ (tso)->sp += 2; \ val; \ }) @@ -1026,19 +1296,18 @@ FN_(tryTakeMVarzh_fast) /* unlock in the SMP case */ SET_INFO(mvar,&stg_FULL_MVAR_info); #endif - TICK_RET_UNBOXED_TUP(1); - RET_P(val); } else { /* No further putMVars, MVar is now empty */ + mvar->value = (StgClosure *)&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. */ SET_INFO(mvar,&stg_EMPTY_MVAR_info); - mvar->value = (StgClosure *)&stg_END_TSO_QUEUE_closure; - TICK_RET_UNBOXED_TUP(1); - RET_P(val); } + + TICK_RET_UNBOXED_TUP(1); + RET_NP((I_)1, val); FE_ } @@ -1201,6 +1470,30 @@ FN_(makeStableNamezh_fast) RET_P(sn_obj); } + +FN_(makeStablePtrzh_fast) +{ + /* Args: R1 = a */ + StgStablePtr sp; + FB_ + MAYBE_GC(R1_PTR, makeStablePtrzh_fast); + sp = RET_STGCALL1(StgStablePtr,getStablePtr,R1.p); + RET_N(sp); + FE_ +} + +FN_(deRefStablePtrzh_fast) +{ + /* Args: R1 = the stable ptr */ + P_ r; + StgStablePtr sp; + FB_ + sp = (StgStablePtr)R1.w; + r = stable_ptr_table[(StgWord)sp].addr; + RET_P(r); + FE_ +} + /* ----------------------------------------------------------------------------- Bytecode object primitives ------------------------------------------------------------------------- */