/* -----------------------------------------------------------------------------
- * $Id: PrimOps.h,v 1.4 1999/01/14 11:57:48 sewardj Exp $
+ * $Id: PrimOps.h,v 1.107 2003/11/12 17:27:01 sof Exp $
+ *
+ * (c) The GHC Team, 1998-2000
*
* Macros for primitive operations in STG-ish C code.
*
* ---------------------------------------------------------------------------*/
+/* As of 5 Dec 01, this file no longer implements the primops, since they are
+ translated into standard C in compiler/absCSyn/AbsCUtils during the absC
+ flattening pass. Only {add,sub,mul}IntCzh remain untranslated. Most of
+ what is here is now EXTFUN_RTS declarations for the out-of-line primop
+ implementations which live in compiler/rts/PrimOps.hc.
+*/
+
#ifndef PRIMOPS_H
#define PRIMOPS_H
-/* -----------------------------------------------------------------------------
- Comparison PrimOps.
- -------------------------------------------------------------------------- */
-
-#define gtCharZh(r,a,b) r=(I_)((a)> (b))
-#define geCharZh(r,a,b) r=(I_)((a)>=(b))
-#define eqCharZh(r,a,b) r=(I_)((a)==(b))
-#define neCharZh(r,a,b) r=(I_)((a)!=(b))
-#define ltCharZh(r,a,b) r=(I_)((a)< (b))
-#define leCharZh(r,a,b) r=(I_)((a)<=(b))
-
-/* Int comparisons: >#, >=# etc */
-#define ZgZh(r,a,b) r=(I_)((a) >(b))
-#define ZgZeZh(r,a,b) r=(I_)((a)>=(b))
-#define ZeZeZh(r,a,b) r=(I_)((a)==(b))
-#define ZdZeZh(r,a,b) r=(I_)((a)!=(b))
-#define ZlZh(r,a,b) r=(I_)((a) <(b))
-#define ZlZeZh(r,a,b) r=(I_)((a)<=(b))
-
-#define gtWordZh(r,a,b) r=(I_)((a) >(b))
-#define geWordZh(r,a,b) r=(I_)((a)>=(b))
-#define eqWordZh(r,a,b) r=(I_)((a)==(b))
-#define neWordZh(r,a,b) r=(I_)((a)!=(b))
-#define ltWordZh(r,a,b) r=(I_)((a) <(b))
-#define leWordZh(r,a,b) r=(I_)((a)<=(b))
-
-#define gtAddrZh(r,a,b) r=(I_)((a) >(b))
-#define geAddrZh(r,a,b) r=(I_)((a)>=(b))
-#define eqAddrZh(r,a,b) r=(I_)((a)==(b))
-#define neAddrZh(r,a,b) r=(I_)((a)!=(b))
-#define ltAddrZh(r,a,b) r=(I_)((a) <(b))
-#define leAddrZh(r,a,b) r=(I_)((a)<=(b))
-
-#define gtFloatZh(r,a,b) r=(I_)((a)> (b))
-#define geFloatZh(r,a,b) r=(I_)((a)>=(b))
-#define eqFloatZh(r,a,b) r=(I_)((a)==(b))
-#define neFloatZh(r,a,b) r=(I_)((a)!=(b))
-#define ltFloatZh(r,a,b) r=(I_)((a)< (b))
-#define leFloatZh(r,a,b) r=(I_)((a)<=(b))
-
-/* Double comparisons: >##, >=#@ etc */
-#define ZgZhZh(r,a,b) r=(I_)((a) >(b))
-#define ZgZeZhZh(r,a,b) r=(I_)((a)>=(b))
-#define ZeZeZhZh(r,a,b) r=(I_)((a)==(b))
-#define ZdZeZhZh(r,a,b) r=(I_)((a)!=(b))
-#define ZlZhZh(r,a,b) r=(I_)((a) <(b))
-#define ZlZeZhZh(r,a,b) r=(I_)((a)<=(b))
-
-/* used by returning comparison primops, defined in Prims.hc. */
-extern const StgClosure *PrelBase_Bool_closure_tbl[];
+#include "MachDeps.h"
-/* -----------------------------------------------------------------------------
- Char# PrimOps.
- -------------------------------------------------------------------------- */
+#if WORD_SIZE_IN_BITS < 32
+#error GHC C backend requires 32+-bit words
+#endif
-#define ordZh(r,a) r=(I_)((W_) (a))
-#define chrZh(r,a) r=(StgChar)((W_)(a))
/* -----------------------------------------------------------------------------
- Int# PrimOps.
- -------------------------------------------------------------------------- */
-
-I_ stg_div (I_ a, I_ b);
-
-#define ZpZh(r,a,b) r=(a)+(b)
-#define ZmZh(r,a,b) r=(a)-(b)
-#define ZtZh(r,a,b) r=(a)*(b)
-#define quotIntZh(r,a,b) r=(a)/(b)
-#define ZdZh(r,a,b) r=ULTRASAFESTGCALL2(I_,(void *, I_, I_),stg_div,(a),(b))
-#define remIntZh(r,a,b) r=(a)%(b)
-#define negateIntZh(r,a) r=-(a)
+ * Int operations with carry.
+ * -------------------------------------------------------------------------- */
-/* The following operations are the standard add,subtract and multiply
- * except that they return a carry if the operation overflows.
+/* Multiply with overflow checking.
*
- * They are all defined in terms of 32-bit integers and use the GCC
- * 'long long' extension to get a 64-bit result. We'd like to use
- * 64-bit integers on 64-bit architectures, but it seems that gcc's
- * 'long long' type is set at 64-bits even on a 64-bit machine.
+ * This is tricky - the usual sign rules for add/subtract don't apply.
+ *
+ * On 32-bit machines we use gcc's 'long long' types, finding
+ * overflow with some careful bit-twiddling.
+ *
+ * On 64-bit machines where gcc's 'long long' type is also 64-bits,
+ * we use a crude approximation, testing whether either operand is
+ * larger than 32-bits; if neither is, then we go ahead with the
+ * multiplication.
+ *
+ * Return non-zero if there is any possibility that the signed multiply
+ * of a and b might overflow. Return zero only if you are absolutely sure
+ * that it won't overflow. If in doubt, return non-zero.
*/
+#if SIZEOF_VOID_P == 4
+
#ifdef WORDS_BIGENDIAN
-#define C 0
-#define R 1
+#define RTS_CARRY_IDX__ 0
+#define RTS_REM_IDX__ 1
#else
-#define C 1
-#define R 0
+#define RTS_CARRY_IDX__ 1
+#define RTS_REM_IDX__ 0
#endif
typedef union {
StgInt32 i[2];
} long_long_u ;
-#define addWithCarryZh(r,c,a,b) \
-{ long_long_u z; \
- z.l = a + b; \
- r = z.i[R]; \
- c = z.i[C]; \
-}
-
-
-
-#define subWithCarryZh(r,c,a,b) \
-{ long_long_u z; \
- z.l = a + b; \
- r = z.i[R]; \
- c = z.i[C]; \
-}
-
-#define mulWithCarryZh(r,c,a,b) \
-{ long_long_u z; \
- z.l = a * b; \
- r = z.i[R]; \
- c = z.i[C]; \
-}
-
-/* -----------------------------------------------------------------------------
- Word PrimOps.
- -------------------------------------------------------------------------- */
-
-#define quotWordZh(r,a,b) r=((W_)a)/((W_)b)
-#define remWordZh(r,a,b) r=((W_)a)%((W_)b)
-
-#define andZh(r,a,b) r=(a)&(b)
-#define orZh(r,a,b) r=(a)|(b)
-#define xorZh(r,a,b) r=(a)^(b)
-#define notZh(r,a) r=~(a)
-
-#define shiftLZh(r,a,b) r=(a)<<(b)
-#define shiftRLZh(r,a,b) r=(a)>>(b)
-#define iShiftLZh(r,a,b) r=(a)<<(b)
-/* Right shifting of signed quantities is not portable in C, so
- the behaviour you'll get from using these primops depends
- on the whatever your C compiler is doing. ToDo: fix/document. -- sof 8/98
-*/
-#define iShiftRAZh(r,a,b) r=(a)>>(b)
-#define iShiftRLZh(r,a,b) r=(a)>>(b)
-
-#define int2WordZh(r,a) r=(W_)(a)
-#define word2IntZh(r,a) r=(I_)(a)
-
-/* -----------------------------------------------------------------------------
- Addr PrimOps.
- -------------------------------------------------------------------------- */
+#define mulIntMayOflo(a,b) \
+({ \
+ StgInt32 r, c; \
+ long_long_u z; \
+ z.l = (StgInt64)a * (StgInt64)b; \
+ r = z.i[RTS_REM_IDX__]; \
+ c = z.i[RTS_CARRY_IDX__]; \
+ if (c == 0 || c == -1) { \
+ c = ((StgWord)((a^b) ^ r)) \
+ >> (BITS_IN (I_) - 1); \
+ } \
+ c; \
+})
+
+/* Careful: the carry calculation above is extremely delicate. Make sure
+ * you test it thoroughly after changing it.
+ */
-#define int2AddrZh(r,a) r=(A_)(a)
-#define addr2IntZh(r,a) r=(I_)(a)
+#else
-#define indexCharOffAddrZh(r,a,i) r= ((C_ *)(a))[i]
-#define indexIntOffAddrZh(r,a,i) r= ((I_ *)(a))[i]
-#define indexAddrOffAddrZh(r,a,i) r= ((PP_)(a))[i]
-#define indexFloatOffAddrZh(r,a,i) r= PK_FLT((P_) (((StgFloat *)(a)) + i))
-#define indexDoubleOffAddrZh(r,a,i) r= PK_DBL((P_) (((StgDouble *)(a)) + i))
-#define indexStablePtrOffAddrZh(r,a,i) r= ((StgStablePtr *)(a))[i]
-#ifdef SUPPORT_LONG_LONGS
-#define indexInt64OffAddrZh(r,a,i) r= ((LI_ *)(a))[i]
-#define indexWord64OffAddrZh(r,a,i) r= ((LW_ *)(a))[i]
+#define HALF_INT (((I_)1) << (BITS_IN (I_) / 2))
+
+#define stg_abs(a) (((I_)(a)) < 0 ? -((I_)(a)) : ((I_)(a)))
+
+#define mulIntMayOflo(a,b) \
+({ \
+ I_ c; \
+ if (stg_abs(a) >= HALF_INT || \
+ stg_abs(b) >= HALF_INT) { \
+ c = 1; \
+ } else { \
+ c = 0; \
+ } \
+ c; \
+})
#endif
-#define writeCharOffAddrZh(a,i,v) ((C_ *)(a))[i] = (v)
-#define writeIntOffAddrZh(a,i,v) ((I_ *)(a))[i] = (v)
-#define writeWordOffAddrZh(a,i,v) ((W_ *)(a))[i] = (v)
-#define writeAddrOffAddrZh(a,i,v) ((PP_)(a))[i] = (v)
-#define writeForeignObjOffAddrZh(a,i,v) ((PP_)(a))[i] = ForeignObj_CLOSURE_DATA(v)
-#define writeFloatOffAddrZh(a,i,v) ASSIGN_FLT((P_) (((StgFloat *)(a)) + i),v)
-#define writeDoubleOffAddrZh(a,i,v) ASSIGN_DBL((P_) (((StgDouble *)(a)) + i),v)
-#define writeStablePtrOffAddrZh(a,i,v) ((StgStablePtr *)(a))[i] = (v)
-#ifdef SUPPORT_LONG_LONGS
-#define writeInt64OffAddrZh(a,i,v) ((LI_ *)(a))[i] = (v)
-#define writeWord64OffAddrZh(a,i,v) ((LW_ *)(a))[i] = (v)
-#endif
-
-/* -----------------------------------------------------------------------------
- Float PrimOps.
- -------------------------------------------------------------------------- */
-
-#define plusFloatZh(r,a,b) r=(a)+(b)
-#define minusFloatZh(r,a,b) r=(a)-(b)
-#define timesFloatZh(r,a,b) r=(a)*(b)
-#define divideFloatZh(r,a,b) r=(a)/(b)
-#define negateFloatZh(r,a) r=-(a)
-
-#define int2FloatZh(r,a) r=(StgFloat)(a)
-#define float2IntZh(r,a) r=(I_)(a)
-
-#define expFloatZh(r,a) r=(StgFloat) RET_PRIM_STGCALL1(StgDouble,exp,a)
-#define logFloatZh(r,a) r=(StgFloat) RET_PRIM_STGCALL1(StgDouble,log,a)
-#define sqrtFloatZh(r,a) r=(StgFloat) RET_PRIM_STGCALL1(StgDouble,sqrt,a)
-#define sinFloatZh(r,a) r=(StgFloat) RET_PRIM_STGCALL1(StgDouble,sin,a)
-#define cosFloatZh(r,a) r=(StgFloat) RET_PRIM_STGCALL1(StgDouble,cos,a)
-#define tanFloatZh(r,a) r=(StgFloat) RET_PRIM_STGCALL1(StgDouble,tan,a)
-#define asinFloatZh(r,a) r=(StgFloat) RET_PRIM_STGCALL1(StgDouble,asin,a)
-#define acosFloatZh(r,a) r=(StgFloat) RET_PRIM_STGCALL1(StgDouble,acos,a)
-#define atanFloatZh(r,a) r=(StgFloat) RET_PRIM_STGCALL1(StgDouble,atan,a)
-#define sinhFloatZh(r,a) r=(StgFloat) RET_PRIM_STGCALL1(StgDouble,sinh,a)
-#define coshFloatZh(r,a) r=(StgFloat) RET_PRIM_STGCALL1(StgDouble,cosh,a)
-#define tanhFloatZh(r,a) r=(StgFloat) RET_PRIM_STGCALL1(StgDouble,tanh,a)
-#define powerFloatZh(r,a,b) r=(StgFloat) RET_PRIM_STGCALL2(StgDouble,pow,a,b)
-
-/* -----------------------------------------------------------------------------
- Double PrimOps.
- -------------------------------------------------------------------------- */
-
-#define ZpZhZh(r,a,b) r=(a)+(b)
-#define ZmZhZh(r,a,b) r=(a)-(b)
-#define ZtZhZh(r,a,b) r=(a)*(b)
-#define ZdZhZh(r,a,b) r=(a)/(b)
-#define negateDoubleZh(r,a) r=-(a)
-
-#define int2DoubleZh(r,a) r=(StgDouble)(a)
-#define double2IntZh(r,a) r=(I_)(a)
-
-#define float2DoubleZh(r,a) r=(StgDouble)(a)
-#define double2FloatZh(r,a) r=(StgFloat)(a)
-
-#define expDoubleZh(r,a) r=(StgDouble) RET_PRIM_STGCALL1(StgDouble,exp,a)
-#define logDoubleZh(r,a) r=(StgDouble) RET_PRIM_STGCALL1(StgDouble,log,a)
-#define sqrtDoubleZh(r,a) r=(StgDouble) RET_PRIM_STGCALL1(StgDouble,sqrt,a)
-#define sinDoubleZh(r,a) r=(StgDouble) RET_PRIM_STGCALL1(StgDouble,sin,a)
-#define cosDoubleZh(r,a) r=(StgDouble) RET_PRIM_STGCALL1(StgDouble,cos,a)
-#define tanDoubleZh(r,a) r=(StgDouble) RET_PRIM_STGCALL1(StgDouble,tan,a)
-#define asinDoubleZh(r,a) r=(StgDouble) RET_PRIM_STGCALL1(StgDouble,asin,a)
-#define acosDoubleZh(r,a) r=(StgDouble) RET_PRIM_STGCALL1(StgDouble,acos,a)
-#define atanDoubleZh(r,a) r=(StgDouble) RET_PRIM_STGCALL1(StgDouble,atan,a)
-#define sinhDoubleZh(r,a) r=(StgDouble) RET_PRIM_STGCALL1(StgDouble,sinh,a)
-#define coshDoubleZh(r,a) r=(StgDouble) RET_PRIM_STGCALL1(StgDouble,cosh,a)
-#define tanhDoubleZh(r,a) r=(StgDouble) RET_PRIM_STGCALL1(StgDouble,tanh,a)
-/* Power: **## */
-#define ZtZtZhZh(r,a,b) r=(StgDouble) RET_PRIM_STGCALL2(StgDouble,pow,a,b)
/* -----------------------------------------------------------------------------
Integer PrimOps.
-------------------------------------------------------------------------- */
-/* We can do integer2Int and cmpInteger inline, since they don't need
- * to allocate any memory.
- */
-
-#define integer2IntZh(r, aa,sa,da) \
-{ MP_INT arg; \
- \
- arg._mp_alloc = (aa); \
- arg._mp_size = (sa); \
- arg._mp_d = (unsigned long int *) (BYTE_ARR_CTS(da)); \
- \
- (r) = RET_PRIM_STGCALL1(I_,mpz_get_si,&arg); \
-}
-
-#define integer2WordZh(r, aa,sa,da) \
-{ MP_INT arg; \
- \
- arg._mp_alloc = (aa); \
- arg._mp_size = (sa); \
- arg._mp_d = (unsigned long int *) (BYTE_ARR_CTS(da)); \
- \
- (r) = RET_PRIM_STGCALL1(I_,mpz_get_ui,&arg); \
-}
+/* NOTE: gcdIntzh and gcdIntegerIntzh work only for positive inputs! */
-#define cmpIntegerZh(r, a1,s1,d1, a2,s2,d2) \
-{ MP_INT arg1; \
- MP_INT arg2; \
- \
- arg1._mp_alloc= (a1); \
- arg1._mp_size = (s1); \
- arg1._mp_d = (unsigned long int *) (BYTE_ARR_CTS(d1)); \
- arg2._mp_alloc= (a2); \
- arg2._mp_size = (s2); \
- arg2._mp_d = (unsigned long int *) (BYTE_ARR_CTS(d2)); \
- \
- (r) = RET_PRIM_STGCALL2(I_,mpz_cmp,&arg1,&arg2); \
-}
-
-/* A glorious hack: calling mpz_neg would entail allocation and
- * copying, but by looking at what mpz_neg actually does, we can
- * derive a better version:
- */
-
-#define negateIntegerZh(ra, rs, rd, a, s, d) \
-{ \
- (ra) = (a); \
- (rs) = -(s); \
- (rd) = d; \
-}
-
-/* The rest are all out-of-line: -------- */
+/* Some of these are out-of-line: -------- */
/* Integer arithmetic */
-EF_(plusIntegerZh_fast);
-EF_(minusIntegerZh_fast);
-EF_(timesIntegerZh_fast);
-EF_(gcdIntegerZh_fast);
-EF_(quotRemIntegerZh_fast);
-EF_(divModIntegerZh_fast);
+EXTFUN_RTS(plusIntegerzh_fast);
+EXTFUN_RTS(minusIntegerzh_fast);
+EXTFUN_RTS(timesIntegerzh_fast);
+EXTFUN_RTS(gcdIntegerzh_fast);
+EXTFUN_RTS(quotRemIntegerzh_fast);
+EXTFUN_RTS(quotIntegerzh_fast);
+EXTFUN_RTS(remIntegerzh_fast);
+EXTFUN_RTS(divExactIntegerzh_fast);
+EXTFUN_RTS(divModIntegerzh_fast);
+
+EXTFUN_RTS(cmpIntegerIntzh_fast);
+EXTFUN_RTS(cmpIntegerzh_fast);
+EXTFUN_RTS(integer2Intzh_fast);
+EXTFUN_RTS(integer2Wordzh_fast);
+EXTFUN_RTS(gcdIntegerIntzh_fast);
+EXTFUN_RTS(gcdIntzh_fast);
/* Conversions */
-EF_(int2IntegerZh_fast);
-EF_(word2IntegerZh_fast);
-EF_(addr2IntegerZh_fast);
+EXTFUN_RTS(int2Integerzh_fast);
+EXTFUN_RTS(word2Integerzh_fast);
-/* Floating-point encodings/decodings */
-EF_(encodeFloatZh_fast);
-EF_(decodeFloatZh_fast);
+/* Floating-point decodings */
+EXTFUN_RTS(decodeFloatzh_fast);
+EXTFUN_RTS(decodeDoublezh_fast);
+
+/* Bit operations */
+EXTFUN_RTS(andIntegerzh_fast);
+EXTFUN_RTS(orIntegerzh_fast);
+EXTFUN_RTS(xorIntegerzh_fast);
+EXTFUN_RTS(complementIntegerzh_fast);
-EF_(encodeDoubleZh_fast);
-EF_(decodeDoubleZh_fast);
/* -----------------------------------------------------------------------------
Word64 PrimOps.
#ifdef SUPPORT_LONG_LONGS
-#define integerToWord64Zh(r, aa,sa,da) \
-{ unsigned long int* d; \
- StgNat64 res; \
- \
- d = (unsigned long int *) (BYTE_ARR_CTS(da)); \
- if ( (aa) == 0 ) { \
- res = (LW_)0; \
- } else if ( (aa) == 1) { \
- res = (LW_)d[0]; \
- } else { \
- res = (LW_)d[0] + (LW_)d[1] * 0x100000000ULL; \
- } \
- (r) = res; \
-}
-
-#define integerToInt64Zh(r, aa,sa,da) \
-{ unsigned long int* d; \
- StgInt64 res; \
- \
- d = (unsigned long int *) (BYTE_ARR_CTS(da)); \
- if ( (aa) == 0 ) { \
- res = (LI_)0; \
- } else if ( (aa) == 1) { \
- res = (LI_)d[0]; \
- } else { \
- res = (LI_)d[0] + (LI_)d[1] * 0x100000000LL; \
- } \
- (r) = res; \
-}
-
/* Conversions */
-EF_(int64ToIntegerZh_fast);
-EF_(word64ToIntegerZh_fast);
+EXTFUN_RTS(int64ToIntegerzh_fast);
+EXTFUN_RTS(word64ToIntegerzh_fast);
-/* The rest are (way!) out of line, implemented via C entry points.
- */
-I_ stg_gtWord64 (StgNat64, StgNat64);
-I_ stg_geWord64 (StgNat64, StgNat64);
-I_ stg_eqWord64 (StgNat64, StgNat64);
-I_ stg_neWord64 (StgNat64, StgNat64);
-I_ stg_ltWord64 (StgNat64, StgNat64);
-I_ stg_leWord64 (StgNat64, StgNat64);
-
-I_ stg_gtInt64 (StgInt64, StgInt64);
-I_ stg_geInt64 (StgInt64, StgInt64);
-I_ stg_eqInt64 (StgInt64, StgInt64);
-I_ stg_neInt64 (StgInt64, StgInt64);
-I_ stg_ltInt64 (StgInt64, StgInt64);
-I_ stg_leInt64 (StgInt64, StgInt64);
-
-LW_ stg_remWord64 (StgNat64, StgNat64);
-LW_ stg_quotWord64 (StgNat64, StgNat64);
-
-LI_ stg_remInt64 (StgInt64, StgInt64);
-LI_ stg_quotInt64 (StgInt64, StgInt64);
-LI_ stg_negateInt64 (StgInt64);
-LI_ stg_plusInt64 (StgInt64, StgInt64);
-LI_ stg_minusInt64 (StgInt64, StgInt64);
-LI_ stg_timesInt64 (StgInt64, StgInt64);
-
-LW_ stg_and64 (StgNat64, StgNat64);
-LW_ stg_or64 (StgNat64, StgNat64);
-LW_ stg_xor64 (StgNat64, StgNat64);
-LW_ stg_not64 (StgNat64);
-
-LW_ stg_shiftL64 (StgNat64, StgInt);
-LW_ stg_shiftRL64 (StgNat64, StgInt);
-LI_ stg_iShiftL64 (StgInt64, StgInt);
-LI_ stg_iShiftRL64 (StgInt64, StgInt);
-LI_ stg_iShiftRA64 (StgInt64, StgInt);
-
-LI_ stg_intToInt64 (StgInt);
-I_ stg_int64ToInt (StgInt64);
-LW_ stg_int64ToWord64 (StgInt64);
-
-LW_ stg_wordToWord64 (StgWord);
-W_ stg_word64ToWord (StgNat64);
-LI_ stg_word64ToInt64 (StgNat64);
#endif
/* -----------------------------------------------------------------------------
#ifdef DEBUG
#define BYTE_ARR_CTS(a) \
- ({ ASSERT((GET_INFO(a) == &ARR_WORDS_info) \
- || (GET_INFO(a) == &MUT_ARR_WORDS_info)); \
+ ({ ASSERT(GET_INFO((StgArrWords *)(a)) == &stg_ARR_WORDS_info); \
REAL_BYTE_ARR_CTS(a); })
#define PTRS_ARR_CTS(a) \
- ({ ASSERT((GET_INFO(a) == &ARR_PTRS_info) \
- || (GET_INFO(a) == &MUT_ARR_PTRS_info)); \
+ ({ ASSERT((GET_INFO((StgMutArrPtrs *)(a)) == &stg_MUT_ARR_PTRS_FROZEN_info) \
+ || (GET_INFO((StgMutArrPtrs *)(a)) == &stg_MUT_ARR_PTRS_info)); \
REAL_PTRS_ARR_CTS(a); })
#else
#define BYTE_ARR_CTS(a) REAL_BYTE_ARR_CTS(a)
#define PTRS_ARR_CTS(a) REAL_PTRS_ARR_CTS(a)
#endif
+
extern I_ genSymZh(void);
extern I_ resetGenSymZh(void);
-/*--- everything except new*Array is done inline: */
-
-#define sameMutableArrayZh(r,a,b) r=(I_)((a)==(b))
-#define sameMutableByteArrayZh(r,a,b) r=(I_)((a)==(b))
-
-#define readArrayZh(r,a,i) r=((PP_) PTRS_ARR_CTS(a))[(i)]
-
-#define readCharArrayZh(r,a,i) indexCharOffAddrZh(r,BYTE_ARR_CTS(a),i)
-#define readIntArrayZh(r,a,i) indexIntOffAddrZh(r,BYTE_ARR_CTS(a),i)
-#define readWordArrayZh(r,a,i) indexWordOffAddrZh(r,BYTE_ARR_CTS(a),i)
-#define readAddrArrayZh(r,a,i) indexAddrOffAddrZh(r,BYTE_ARR_CTS(a),i)
-#define readFloatArrayZh(r,a,i) indexFloatOffAddrZh(r,BYTE_ARR_CTS(a),i)
-#define readDoubleArrayZh(r,a,i) indexDoubleOffAddrZh(r,BYTE_ARR_CTS(a),i)
-#define readStablePtrArrayZh(r,a,i) indexStablePtrOffAddrZh(r,BYTE_ARR_CTS(a),i)
-#ifdef SUPPORT_LONG_LONGS
-#define readInt64ArrayZh(r,a,i) indexInt64OffAddrZh(r,BYTE_ARR_CTS(a),i)
-#define readWord64ArrayZh(r,a,i) indexWord64OffAddrZh(r,BYTE_ARR_CTS(a),i)
-#endif
-
-/* result ("r") arg ignored in write macros! */
-#define writeArrayZh(a,i,v) ((PP_) PTRS_ARR_CTS(a))[(i)]=(v)
-
-#define writeCharArrayZh(a,i,v) ((C_ *)(BYTE_ARR_CTS(a)))[i] = (v)
-#define writeIntArrayZh(a,i,v) ((I_ *)(BYTE_ARR_CTS(a)))[i] = (v)
-#define writeWordArrayZh(a,i,v) ((W_ *)(BYTE_ARR_CTS(a)))[i] = (v)
-#define writeAddrArrayZh(a,i,v) ((PP_)(BYTE_ARR_CTS(a)))[i] = (v)
-#define writeFloatArrayZh(a,i,v) \
- ASSIGN_FLT((P_) (((StgFloat *)(BYTE_ARR_CTS(a))) + i),v)
-#define writeDoubleArrayZh(a,i,v) \
- ASSIGN_DBL((P_) (((StgDouble *)(BYTE_ARR_CTS(a))) + i),v)
-#define writeStablePtrArrayZh(a,i,v) ((StgStablePtr *)(BYTE_ARR_CTS(a)))[i] = (v)
-#ifdef SUPPORT_LONG_LONGS
-#define writeInt64ArrayZh(a,i,v) ((LI_ *)(BYTE_ARR_CTS(a)))[i] = (v)
-#define writeWord64ArrayZh(a,i,v) ((LW_ *)(BYTE_ARR_CTS(a)))[i] = (v)
-#endif
-
-#define indexArrayZh(r,a,i) r=((PP_) PTRS_ARR_CTS(a))[(i)]
+/*--- Almost everything in line. */
-#define indexCharArrayZh(r,a,i) indexCharOffAddrZh(r,BYTE_ARR_CTS(a),i)
-#define indexIntArrayZh(r,a,i) indexIntOffAddrZh(r,BYTE_ARR_CTS(a),i)
-#define indexWordArrayZh(r,a,i) indexWordOffAddrZh(r,BYTE_ARR_CTS(a),i)
-#define indexAddrArrayZh(r,a,i) indexAddrOffAddrZh(r,BYTE_ARR_CTS(a),i)
-#define indexFloatArrayZh(r,a,i) indexFloatOffAddrZh(r,BYTE_ARR_CTS(a),i)
-#define indexDoubleArrayZh(r,a,i) indexDoubleOffAddrZh(r,BYTE_ARR_CTS(a),i)
-#define indexStablePtrArrayZh(r,a,i) indexStablePtrOffAddrZh(r,BYTE_ARR_CTS(a),i)
-#ifdef SUPPORT_LONG_LONGS
-#define indexInt64ArrayZh(r,a,i) indexInt64OffAddrZh(r,BYTE_ARR_CTS(a),i)
-#define indexWord64ArrayZh(r,a,i) indexWord64OffAddrZh(r,BYTE_ARR_CTS(a),i)
-#endif
-
-#define indexCharOffForeignObjZh(r,fo,i) indexCharOffAddrZh(r,ForeignObj_CLOSURE_DATA(fo),i)
-#define indexIntOffForeignObjZh(r,fo,i) indexIntOffAddrZh(r,ForeignObj_CLOSURE_DATA(fo),i)
-#define indexWordOffForeignObjZh(r,fo,i) indexWordOffAddrZh(r,ForeignObj_CLOSURE_DATA(fo),i)
-#define indexAddrOffForeignObjZh(r,fo,i) indexAddrOffAddrZh(r,ForeignObj_CLOSURE_DATA(fo),i)
-#define indexFloatOffForeignObjZh(r,fo,i) indexFloatOffAddrZh(r,ForeignObj_CLOSURE_DATA(fo),i)
-#define indexDoubleOffForeignObjZh(r,fo,i) indexDoubleOffAddrZh(r,ForeignObj_CLOSURE_DATA(fo),i)
-#define indexStablePtrOffForeignObjZh(r,fo,i) indexStablePtrOffAddrZh(r,ForeignObj_CLOSURE_DATA(fo),i)
-#ifdef SUPPORT_LONG_LONGS
-#define indexInt64OffForeignObjZh(r,fo,i) indexInt64OffAddrZh(r,ForeignObj_CLOSURE_DATA(fo),i)
-#define indexWord64OffForeignObjZh(r,fo,i) indexWord64OffAddrZh(r,ForeignObj_CLOSURE_DATA(fo),i)
-#endif
-
-#define indexCharOffAddrZh(r,a,i) r= ((C_ *)(a))[i]
-#define indexIntOffAddrZh(r,a,i) r= ((I_ *)(a))[i]
-#define indexWordOffAddrZh(r,a,i) r= ((W_ *)(a))[i]
-#define indexAddrOffAddrZh(r,a,i) r= ((PP_)(a))[i]
-#define indexFloatOffAddrZh(r,a,i) r= PK_FLT((P_) (((StgFloat *)(a)) + i))
-#define indexDoubleOffAddrZh(r,a,i) r= PK_DBL((P_) (((StgDouble *)(a)) + i))
-#ifdef SUPPORT_LONG_LONGS
-#define indexInt64OffAddrZh(r,a,i) r= ((LI_ *)(a))[i]
-#define indexWord64OffAddrZh(r,a,i) r= ((LW_ *)(a))[i]
-#endif
-
-/* Freezing arrays-of-ptrs requires changing an info table, for the
- benefit of the generational collector. It needs to scavenge mutable
- objects, even if they are in old space. When they become immutable,
- they can be removed from this scavenge list. */
-
-#define unsafeFreezeArrayZh(r,a) \
- { \
- SET_INFO((StgClosure *)a,&MUT_ARR_PTRS_FROZEN_info); \
- r = a; \
- }
-
-#define unsafeFreezeByteArrayZh(r,a) r=(a)
-
-#define sizeofByteArrayZh(r,a) \
- r = (((StgArrWords *)(a))->words * sizeof(W_))
-#define sizeofMutableByteArrayZh(r,a) \
- r = (((StgArrWords *)(a))->words * sizeof(W_))
-
-/* and the out-of-line ones... */
-
-EF_(newCharArrayZh_fast);
-EF_(newIntArrayZh_fast);
-EF_(newWordArrayZh_fast);
-EF_(newAddrArrayZh_fast);
-EF_(newFloatArrayZh_fast);
-EF_(newDoubleArrayZh_fast);
-EF_(newStablePtrArrayZh_fast);
-EF_(newArrayZh_fast);
-
-/* encoding and decoding of floats/doubles. */
-
-/* We only support IEEE floating point format */
-#include "ieee-flpt.h"
-
-#if FLOATS_AS_DOUBLES /* i.e. 64-bit machines */
-#define encodeFloatZh(r, aa,sa,da, expon) encodeDoubleZh(r, aa,sa,da, expon)
-#else
-#define encodeFloatZh(r, aa,sa,da, expon) \
-{ MP_INT arg; \
- /* Does not allocate memory */ \
- \
- arg._mp_alloc = aa; \
- arg._mp_size = sa; \
- arg._mp_d = (unsigned long int *) (BYTE_ARR_CTS(da)); \
- \
- r = RET_PRIM_STGCALL2(StgFloat, __encodeFloat,&arg,(expon));\
-}
-#endif /* FLOATS_AS_DOUBLES */
-
-#define encodeDoubleZh(r, aa,sa,da, expon) \
-{ MP_INT arg; \
- /* Does not allocate memory */ \
- \
- arg._mp_alloc = aa; \
- arg._mp_size = sa; \
- arg._mp_d = (unsigned long int *) (BYTE_ARR_CTS(da)); \
- \
- r = RET_PRIM_STGCALL2(StgDouble, __encodeDouble,&arg,(expon));\
-}
+EXTFUN_RTS(unsafeThawArrayzh_fast);
+EXTFUN_RTS(newByteArrayzh_fast);
+EXTFUN_RTS(newPinnedByteArrayzh_fast);
+EXTFUN_RTS(newArrayzh_fast);
/* The decode operations are out-of-line because they need to allocate
* a byte array.
*/
-
-#ifdef FLOATS_AS_DOUBLES
-#define decodeFloatZh_fast decodeDoubleZh_fast
-#else
-EF_(decodeFloatZh_fast);
-#endif
-EF_(decodeDoubleZh_fast);
+/* We only support IEEE floating point formats. */
+#include "ieee-flpt.h"
+EXTFUN_RTS(decodeFloatzh_fast);
+EXTFUN_RTS(decodeDoublezh_fast);
/* grimy low-level support functions defined in StgPrimFloat.c */
-
-extern StgDouble __encodeDouble (MP_INT *s, I_ e);
-extern StgFloat __encodeFloat (MP_INT *s, I_ e);
+extern StgDouble __encodeDouble (I_ size, StgByteArray arr, I_ e);
+extern StgDouble __int_encodeDouble (I_ j, I_ e);
+extern StgFloat __encodeFloat (I_ size, StgByteArray arr, I_ e);
+extern StgFloat __int_encodeFloat (I_ j, I_ e);
extern void __decodeDouble (MP_INT *man, I_ *_exp, StgDouble dbl);
extern void __decodeFloat (MP_INT *man, I_ *_exp, StgFloat flt);
extern StgInt isDoubleNaN(StgDouble d);
extern StgInt isFloatDenormalized(StgFloat f);
extern StgInt isFloatNegativeZero(StgFloat f);
+
/* -----------------------------------------------------------------------------
Mutable variables
newMutVar is out of line.
-------------------------------------------------------------------------- */
-EF_(newMutVarZh_fast);
-
-#define readMutVarZh(r,a) r=(P_)(((StgMutVar *)(a))->var)
-#define writeMutVarZh(a,v) (P_)(((StgMutVar *)(a))->var)=(v)
-#define sameMutVarZh(r,a,b) r=(I_)((a)==(b))
+EXTFUN_RTS(newMutVarzh_fast);
+EXTFUN_RTS(atomicModifyMutVarzh_fast);
/* -----------------------------------------------------------------------------
MVar PrimOps.
All out of line, because they either allocate or may block.
-------------------------------------------------------------------------- */
-#define sameMVarZh(r,a,b) r=(I_)((a)==(b))
-EF_(newMVarZh_fast);
-EF_(takeMVarZh_fast);
-EF_(putMVarZh_fast);
+EXTFUN_RTS(isEmptyMVarzh_fast);
+EXTFUN_RTS(newMVarzh_fast);
+EXTFUN_RTS(takeMVarzh_fast);
+EXTFUN_RTS(putMVarzh_fast);
+EXTFUN_RTS(tryTakeMVarzh_fast);
+EXTFUN_RTS(tryPutMVarzh_fast);
+
/* -----------------------------------------------------------------------------
Delay/Wait PrimOps
-------------------------------------------------------------------------- */
-/* Hmm, I'll think about these later. */
+EXTFUN_RTS(waitReadzh_fast);
+EXTFUN_RTS(waitWritezh_fast);
+EXTFUN_RTS(delayzh_fast);
+#ifdef mingw32_TARGET_OS
+EXTFUN_RTS(asyncReadzh_fast);
+EXTFUN_RTS(asyncWritezh_fast);
+EXTFUN_RTS(asyncDoProczh_fast);
+#endif
+
/* -----------------------------------------------------------------------------
Primitive I/O, error-handling PrimOps
-------------------------------------------------------------------------- */
-EF_(catchZh_fast);
-EF_(raiseZh_fast);
+EXTFUN_RTS(catchzh_fast);
+EXTFUN_RTS(raisezh_fast);
+EXTFUN_RTS(raiseIOzh_fast);
-extern void stg_exit(I_ n) __attribute__ ((noreturn));
+extern void stg_exit(int n) GNU_ATTRIBUTE(__noreturn__);
/* -----------------------------------------------------------------------------
- Stable Pointer PrimOps.
+ Stable Name / Stable Pointer PrimOps
-------------------------------------------------------------------------- */
-#ifndef PAR
-
-extern StgPtr *stable_ptr_table;
-extern StgPtr *stable_ptr_free;
-#define deRefStablePtrZh(r,sp) (r=stable_ptr_table[(sp)])
-#define eqStablePtrZh(r,sp1,sp2) (r=(sp1==sp2))
-
-#define freeStablePointer(stable_ptr) \
- { \
- stable_ptr_table[stable_ptr] = (P_)stable_ptr_free; \
- stable_ptr_free = &stable_ptr_table[stable_ptr]; \
- }
-
-EF_(makeStablePtrZh_fast);
-
-#else /* PAR */
-#define deRefStablePtrZh(ri,sp) \
-do { \
- fflush(stdout); \
- fprintf(stderr, "deRefStablePtr#: no stable pointer support.\n");\
- stg_exit(EXIT_FAILURE); \
-} while(0)
-
-#define eqStablePtrZh(ri,sp1,sp2) \
-do { \
- fflush(stdout); \
- fprintf(stderr, "eqStablePtr#: no stable pointer support.\n"); \
- stg_exit(EXIT_FAILURE); \
-} while(0)
-
-#define makeStablePtrZh(stablePtr,liveness,unstablePtr) \
-do { \
- fflush(stdout); \
- fprintf(stderr, "makeStablePtr#: no stable pointer support.\n");\
- EXIT(EXIT_FAILURE); \
-} while(0)
-
-#define freeStablePtrZh(stablePtr,liveness,unstablePtr) \
-do { \
- fflush(stdout); \
- fprintf(stderr, "makeStablePtr#: no stable pointer support.\n");\
- EXIT(EXIT_FAILURE); \
-} while(0)
-#endif
+EXTFUN_RTS(makeStableNamezh_fast);
+EXTFUN_RTS(makeStablePtrzh_fast);
+EXTFUN_RTS(deRefStablePtrzh_fast);
/* -----------------------------------------------------------------------------
- Parallel PrimOps.
+ Concurrency/Exception PrimOps.
-------------------------------------------------------------------------- */
-EF_(forkZh_fast);
-EF_(killThreadZh_fast);
-EF_(seqZh_fast);
+EXTFUN_RTS(forkzh_fast);
+EXTFUN_RTS(yieldzh_fast);
+EXTFUN_RTS(killThreadzh_fast);
+EXTFUN_RTS(seqzh_fast);
+EXTFUN_RTS(blockAsyncExceptionszh_fast);
+EXTFUN_RTS(unblockAsyncExceptionszh_fast);
+EXTFUN_RTS(myThreadIdzh_fast);
+EXTFUN_RTS(labelThreadzh_fast);
+EXTFUN_RTS(isCurrentThreadBoundzh_fast);
+
+extern int cmp_thread(StgPtr tso1, StgPtr tso2);
+extern int rts_getThreadId(StgPtr tso);
+extern int forkOS_createThread ( HsStablePtr entry );
-/* Hmm, I'll think about these later. */
/* -----------------------------------------------------------------------------
- Pointer equality
+ Weak Pointer PrimOps.
-------------------------------------------------------------------------- */
-/* warning: extremely non-referentially transparent, need to hide in
- an appropriate monad.
+EXTFUN_RTS(mkWeakzh_fast);
+EXTFUN_RTS(finalizzeWeakzh_fast);
+EXTFUN_RTS(deRefWeakzh_fast);
- ToDo: follow indirections.
-*/
-
-#define reallyUnsafePtrEqualityZh(r,a,b) r=((StgPtr)(a) == (StgPtr)(b))
/* -----------------------------------------------------------------------------
- Weak Pointer PrimOps.
+ Foreign Object PrimOps.
-------------------------------------------------------------------------- */
-#ifndef PAR
+EXTFUN_RTS(mkForeignObjzh_fast);
-EF_(mkWeakZh_fast);
-EF_(deRefWeakZh_fast);
-#define sameWeakZh(w1,w2) ((w1)==(w2))
+/* -----------------------------------------------------------------------------
+ Constructor tags
+ -------------------------------------------------------------------------- */
+
+/*
+ * This macro is only used when compiling unregisterised code (see
+ * AbsCUtils.dsCOpStmt for motivation & the Story).
+ */
+#ifndef TABLES_NEXT_TO_CODE
+# define dataToTagzh(r,a) r=(GET_TAG(((StgClosure *)a)->header.info))
#endif
/* -----------------------------------------------------------------------------
- Foreign Object PrimOps.
+ BCOs and BCO linkery
-------------------------------------------------------------------------- */
-#ifndef PAR
+EXTFUN_RTS(newBCOzh_fast);
+EXTFUN_RTS(mkApUpd0zh_fast);
+
+/* ------------------------------------------------------------------------
+ Parallel PrimOps
+
+ A par in the Haskell code is ultimately translated to a parzh macro
+ (with a case wrapped around it to guarantee that the macro is actually
+ executed; see compiler/prelude/PrimOps.lhs)
+ In GUM and SMP we only add a pointer to the spark pool.
+ In GranSim we call an RTS fct, forwarding additional parameters which
+ supply info on granularity of the computation, size of the result value
+ and the degree of parallelism in the sparked expression.
+ ---------------------------------------------------------------------- */
+
+#if defined(GRAN)
+//@cindex _par_
+#define parzh(r,node) parAny(r,node,1,0,0,0,0,0)
+
+//@cindex _parAt_
+#define parAtzh(r,node,where,identifier,gran_info,size_info,par_info,rest) \
+ parAT(r,node,where,identifier,gran_info,size_info,par_info,rest,1)
+
+//@cindex _parAtAbs_
+#define parAtAbszh(r,node,proc,identifier,gran_info,size_info,par_info,rest) \
+ parAT(r,node,proc,identifier,gran_info,size_info,par_info,rest,2)
+
+//@cindex _parAtRel_
+#define parAtRelzh(r,node,proc,identifier,gran_info,size_info,par_info,rest) \
+ parAT(r,node,proc,identifier,gran_info,size_info,par_info,rest,3)
+
+//@cindex _parAtForNow_
+#define parAtForNowzh(r,node,where,identifier,gran_info,size_info,par_info,rest) \
+ parAT(r,node,where,identifier,gran_info,size_info,par_info,rest,0)
+
+#define parAT(r,node,where,identifier,gran_info,size_info,par_info,rest,local) \
+{ \
+ if (closure_SHOULD_SPARK((StgClosure*)node)) { \
+ rtsSparkQ result; \
+ PEs p; \
+ \
+ STGCALL6(newSpark, node,identifier,gran_info,size_info,par_info,local); \
+ switch (local) { \
+ case 2: p = where; /* parAtAbs means absolute PE no. expected */ \
+ break; \
+ case 3: p = CurrentProc+where; /* parAtRel means rel PE no. expected */\
+ break; \
+ default: p = where_is(where); /* parAt means closure expected */ \
+ break; \
+ } \
+ /* update GranSim state according to this spark */ \
+ STGCALL3(GranSimSparkAtAbs, result, (I_)p, identifier); \
+ } \
+}
-#define ForeignObj_CLOSURE_DATA(c) (((StgForeignObj *)c)->data)
+//@cindex _parLocal_
+#define parLocalzh(r,node,identifier,gran_info,size_info,par_info,rest) \
+ parAny(r,node,rest,identifier,gran_info,size_info,par_info,1)
+
+//@cindex _parGlobal_
+#define parGlobalzh(r,node,identifier,gran_info,size_info,par_info,rest) \
+ parAny(r,node,rest,identifier,gran_info,size_info,par_info,0)
+
+#define parAny(r,node,rest,identifier,gran_info,size_info,par_info,local) \
+{ \
+ if (closure_SHOULD_SPARK((StgClosure*)node)) { \
+ rtsSpark *result; \
+ result = RET_STGCALL6(rtsSpark*, newSpark, \
+ node,identifier,gran_info,size_info,par_info,local);\
+ STGCALL1(add_to_spark_queue,result); \
+ STGCALL2(GranSimSpark, local,(P_)node); \
+ } \
+}
-EF_(makeForeignObjZh_fast);
+#define copyablezh(r,node) \
+ /* copyable not yet implemented!! */
-#define writeForeignObjZh(res,datum) \
- (ForeignObj_CLOSURE_DATA(res) = (P_)(datum))
+#define noFollowzh(r,node) \
+ /* noFollow not yet implemented!! */
-#define eqForeignObj(f1,f2) ((f1)==(f2))
+#elif defined(SMP) || defined(PAR)
+#define parzh(r,node) \
+{ \
+ extern unsigned int context_switch; \
+ if (closure_SHOULD_SPARK((StgClosure *)node) && \
+ SparkTl < SparkLim) { \
+ *SparkTl++ = (StgClosure *)(node); \
+ } \
+ r = context_switch = 1; \
+}
+#else /* !GRAN && !SMP && !PAR */
+#define parzh(r,node) r = 1
#endif
/* -----------------------------------------------------------------------------
- Signal processing. Not really primops, but called directly from
- Haskell.
+ ForeignObj - the C backend still needs this.
-------------------------------------------------------------------------- */
+#define ForeignObj_CLOSURE_DATA(c) (((StgForeignObj *)c)->data)
-#define STG_SIG_DFL (-1)
-#define STG_SIG_IGN (-2)
-#define STG_SIG_ERR (-3)
-#define STG_SIG_HAN (-4)
-
-extern StgInt sig_install (StgInt, StgInt, StgStablePtr, sigset_t *);
-#define stg_sig_default(sig,mask) sig_install(sig,STG_SIG_DFL,0,(sigset_t *)mask)
-#define stg_sig_ignore(sig,mask) sig_install(sig,STG_SIG_IGN,0,(sigset_t *)mask)
-#define stg_sig_catch(sig,ptr,mask) sig_install(sig,STG_SIG_HAN,ptr,(sigset_t *)mask)
-#endif PRIMOPS_H
+#endif /* PRIMOPS_H */