diff options
Diffstat (limited to 'erts/emulator/beam')
39 files changed, 3466 insertions, 446 deletions
diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c index b413f0e859..1ae413d46e 100644 --- a/erts/emulator/beam/beam_emu.c +++ b/erts/emulator/beam/beam_emu.c @@ -31,6 +31,7 @@ #include "big.h" #include "beam_load.h" #include "erl_binary.h" +#include "erl_map.h" #include "erl_bits.h" #include "dist.h" #include "beam_bp.h" @@ -217,6 +218,7 @@ BeamInstr beam_continue_exit[1]; BeamInstr* em_call_error_handler; BeamInstr* em_apply_bif; +BeamInstr* em_call_nif; /* NOTE These should be the only variables containing trace instructions. @@ -700,6 +702,19 @@ extern int count_instructions; Fail; \ } +#define IsMap(Src, Fail) if (is_not_map(Src)) { Fail; } + +#define HasMapField(Src, Key, Fail) if (has_not_map_field(Src, Key)) { Fail; } + +#define GetMapElement(Src, Key, Dst, Fail) \ + do { \ + Eterm _res = get_map_element(Src, Key); \ + if (is_non_value(_res)) { \ + Fail; \ + } \ + Dst = _res; \ + } while (0) + #define IsFunction(X, Action) \ do { \ if ( !(is_any_fun(X)) ) { \ @@ -943,7 +958,13 @@ static BeamInstr* apply_fun(Process* p, Eterm fun, Eterm args, Eterm* reg) NOINLINE; static Eterm new_fun(Process* p, Eterm* reg, ErlFunEntry* fe, int num_free) NOINLINE; - +static Eterm new_map(Process* p, Eterm* reg, BeamInstr* I) NOINLINE; +static Eterm update_map_assoc(Process* p, Eterm* reg, + Eterm map, BeamInstr* I) NOINLINE; +static Eterm update_map_exact(Process* p, Eterm* reg, + Eterm map, BeamInstr* I) NOINLINE; +static int has_not_map_field(Eterm map, Eterm key); +static Eterm get_map_element(Eterm map, Eterm key); /* * Functions not directly called by process_main(). OK to inline. @@ -2322,6 +2343,55 @@ void process_main(void) Goto(*I); } + OpCase(new_map_jdII): { + Eterm res; + + x(0) = r(0); + SWAPOUT; + res = new_map(c_p, reg, I); + SWAPIN; + r(0) = x(0); + StoreResult(res, Arg(1)); + Next(4+Arg(3)); + } + + OpCase(update_map_assoc_jsdII): { + Eterm res; + Eterm map; + + GetArg1(1, map); + x(0) = r(0); + SWAPOUT; + res = update_map_assoc(c_p, reg, map, I); + SWAPIN; + if (is_value(res)) { + r(0) = x(0); + StoreResult(res, Arg(2)); + Next(5+Arg(4)); + } else { + goto badarg; + } + } + + OpCase(update_map_exact_jsdII): { + Eterm res; + Eterm map; + + GetArg1(1, map); + x(0) = r(0); + SWAPOUT; + res = update_map_exact(c_p, reg, map, I); + SWAPIN; + if (is_value(res)) { + r(0) = x(0); + StoreResult(res, Arg(2)); + Next(5+Arg(4)); + } else { + goto badarg; + } + } + + /* * All guards with zero arguments have special instructions: * self/0 @@ -3323,6 +3393,13 @@ void process_main(void) reg[0] = r(0); nif_bif_result = (*fp)(&env, bif_nif_arity, reg); erts_post_nif(&env); +#ifdef ERTS_DIRTY_SCHEDULERS + if (is_non_value(nif_bif_result) && c_p->freason == TRAP) { + Export* ep = (Export*) c_p->psd->data[ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT]; + ep->code[0] = I[-3]; + ep->code[1] = I[-2]; + } +#endif } ASSERT(!ERTS_PROC_IS_EXITING(c_p) || is_non_value(nif_bif_result)); PROCESS_MAIN_CHK_LOCKS(c_p); @@ -4964,6 +5041,7 @@ void process_main(void) em_call_error_handler = OpCode(call_error_handler); em_apply_bif = OpCode(apply_bif); + em_call_nif = OpCode(call_nif); beam_apply[0] = (BeamInstr) OpCode(i_apply); beam_apply[1] = (BeamInstr) OpCode(normal_exit); @@ -5022,6 +5100,8 @@ translate_gc_bif(void* gcf) return bit_size_1; } else if (gcf == erts_gc_byte_size_1) { return byte_size_1; + } else if (gcf == erts_gc_map_size_1) { + return map_size_1; } else if (gcf == erts_gc_abs_1) { return abs_1; } else if (gcf == erts_gc_float_1) { @@ -6218,6 +6298,397 @@ new_fun(Process* p, Eterm* reg, ErlFunEntry* fe, int num_free) return make_fun(funp); } +static int has_not_map_field(Eterm map, Eterm key) +{ + map_t* mp; + Eterm* keys; + Uint i; + Uint n; + + mp = (map_t *)map_val(map); + keys = map_get_keys(mp); + n = map_get_size(mp); + if (is_immed(key)) { + for (i = 0; i < n; i++) { + if (keys[i] == key) { + return 0; + } + } + } else { + for (i = 0; i < n; i++) { + if (EQ(keys[i], key)) { + return 0; + } + } + } + return 1; +} + +static Eterm get_map_element(Eterm map, Eterm key) +{ + map_t *mp; + Eterm* ks, *vs; + Uint i; + Uint n; + + mp = (map_t *)map_val(map); + ks = map_get_keys(mp); + vs = map_get_values(mp); + n = map_get_size(mp); + if (is_immed(key)) { + for (i = 0; i < n; i++) { + if (ks[i] == key) { + return vs[i]; + } + } + } else { + for (i = 0; i < n; i++) { + if (EQ(ks[i], key)) { + return vs[i]; + } + } + } + return THE_NON_VALUE; +} + +#define GET_TERM(term, dest) \ +do { \ + Eterm src = (Eterm)(term); \ + switch (src & _TAG_IMMED1_MASK) { \ + case (R_REG_DEF << _TAG_PRIMARY_SIZE) | TAG_PRIMARY_HEADER: \ + dest = x(0); \ + break; \ + case (X_REG_DEF << _TAG_PRIMARY_SIZE) | TAG_PRIMARY_HEADER: \ + dest = x(src >> _TAG_IMMED1_SIZE); \ + break; \ + case (Y_REG_DEF << _TAG_PRIMARY_SIZE) | TAG_PRIMARY_HEADER: \ + dest = y(src >> _TAG_IMMED1_SIZE); \ + break; \ + default: \ + dest = src; \ + break; \ + } \ +} while(0) + + +static Eterm +new_map(Process* p, Eterm* reg, BeamInstr* I) +{ + Uint n = Arg(3); + Uint i; + Uint need = n + 1 /* hdr */ + 1 /*size*/ + 1 /* ptr */ + 1 /* arity */; + Eterm keys; + Eterm *mhp,*thp; + Eterm *E; + BeamInstr *ptr; + map_t *mp; + + if (HeapWordsLeft(p) < need) { + erts_garbage_collect(p, need, reg, Arg(2)); + } + + thp = p->htop; + mhp = thp + 1 + n/2; + E = p->stop; + ptr = &Arg(4); + keys = make_tuple(thp); + *thp++ = make_arityval(n/2); + + mp = (map_t *)mhp; mhp += MAP_HEADER_SIZE; + mp->thing_word = MAP_HEADER; + mp->size = n/2; + mp->keys = keys; + + for (i = 0; i < n/2; i++) { + GET_TERM(*ptr++, *thp++); + GET_TERM(*ptr++, *mhp++); + } + p->htop = mhp; + return make_map(mp); +} + +static Eterm +update_map_assoc(Process* p, Eterm* reg, Eterm map, BeamInstr* I) +{ + Uint n; + Uint num_old; + Uint num_updates; + Uint need; + map_t *old_mp, *mp; + Eterm res; + Eterm* hp; + Eterm* E; + Eterm* old_keys; + Eterm* old_vals; + BeamInstr* new_p; + Eterm new_key; + Eterm* kp; + + if (is_not_map(map)) { + return THE_NON_VALUE; + } + + old_mp = (map_t *) map_val(map); + num_old = map_get_size(old_mp); + + /* + * If the old map is empty, create a new map. + */ + + if (num_old == 0) { + return new_map(p, reg, I+1); + } + + /* + * Allocate heap space for the worst case (i.e. all keys in the + * update list are new). + */ + + num_updates = Arg(4) / 2; + need = 2*(num_old+num_updates) + 1 + MAP_HEADER_SIZE; + if (HeapWordsLeft(p) < need) { + Uint live = Arg(3); + reg[live] = map; + erts_garbage_collect(p, need, reg, live+1); + map = reg[live]; + old_mp = (map_t *)map_val(map); + } + + /* + * Build the skeleton for the map, ready to be filled in. + * + * +-----------------------------------+ + * | (Space for aritvyal for keys) | <-----------+ + * +-----------------------------------+ | + * | (Space for key 1) | | <-- kp + * +-----------------------------------+ | + * . | + * . | + * . | + * +-----------------------------------+ | + * | (Space for last key) | | + * +-----------------------------------+ | + * | MAP_HEADER | | + * +-----------------------------------+ | + * | (Space for number of keys/values) | | + * +-----------------------------------+ | + * | Boxed tuple pointer >----------------+ + * +-----------------------------------+ + * | (Space for value 1) | <-- hp + * +-----------------------------------+ + */ + + E = p->stop; + kp = p->htop + 1; /* Point to first key */ + hp = kp + num_old + num_updates; + + res = make_map(hp); + mp = (map_t *)hp; + hp += MAP_HEADER_SIZE; + mp->thing_word = MAP_HEADER; + mp->keys = make_tuple(kp-1); + + old_vals = map_get_values(old_mp); + old_keys = map_get_keys(old_mp); + + new_p = &Arg(5); + GET_TERM(*new_p, new_key); + n = num_updates; + + /* + * Fill in keys and values, until we run out of either updates + * or old values and keys. + */ + + for (;;) { + Eterm key; + Sint c; + + ASSERT(kp < (Eterm *)mp); + key = *old_keys; + if ((c = CMP_TERM(key, new_key)) < 0) { + /* Copy old key and value */ + *kp++ = key; + *hp++ = *old_vals; + old_keys++, old_vals++, num_old--; + } else { /* Replace or insert new */ + GET_TERM(new_p[1], *hp++); + if (c > 0) { /* If new new key */ + *kp++ = new_key; + } else { /* If replacement */ + *kp++ = key; + old_keys++, old_vals++, num_old--; + } + n--; + if (n == 0) { + break; + } else { + new_p += 2; + GET_TERM(*new_p, new_key); + } + } + if (num_old == 0) { + break; + } + } + + /* + * At this point, we have run out of either old keys and values, + * or the update list. In other words, at least of one n and + * num_old must be zero. + */ + + if (n > 0) { + /* + * All old keys and values have been copied, but there + * are still new keys and values in the update list that + * must be copied. + */ + ASSERT(num_old == 0); + while (n-- > 0) { + GET_TERM(new_p[0], *kp++); + GET_TERM(new_p[1], *hp++); + new_p += 2; + } + } else { + /* + * All updates are now done. We may still have old + * keys and values that we must copy. + */ + ASSERT(n == 0); + while (num_old-- > 0) { + ASSERT(kp < (Eterm *)mp); + *kp++ = *old_keys++; + *hp++ = *old_vals++; + } + } + + /* + * Calculate how many values that are unused at the end of the + * key tuple and fill it out with a bignum header. + */ + if ((n = (Eterm *)mp - kp) > 0) { + *kp = make_pos_bignum_header(n-1); + } + + /* + * Fill in the size of the map in both the key tuple and in the map. + */ + + n = kp - p->htop - 1; /* Actual number of keys/values */ + *p->htop = make_arityval(n); + mp->size = n; + p->htop = hp; + return res; +} + +/* + * Update values for keys that already exist in the map. + */ + +static Eterm +update_map_exact(Process* p, Eterm* reg, Eterm map, BeamInstr* I) +{ + Uint n; + Uint i; + Uint num_old; + Uint need; + map_t *old_mp, *mp; + Eterm res; + Eterm* hp; + Eterm* E; + Eterm* old_keys; + Eterm* old_vals; + BeamInstr* new_p; + Eterm new_key; + + if (is_not_map(map)) { + return THE_NON_VALUE; + } + + old_mp = (map_t *) map_val(map); + num_old = map_get_size(old_mp); + + /* + * If the old map is empty, create a new map. + */ + + if (num_old == 0) { + return THE_NON_VALUE; + } + + /* + * Allocate the exact heap space needed. + */ + + need = num_old + MAP_HEADER_SIZE; + if (HeapWordsLeft(p) < need) { + Uint live = Arg(3); + reg[live] = map; + erts_garbage_collect(p, need, reg, live+1); + map = reg[live]; + old_mp = (map_t *)map_val(map); + } + + /* + * Update map, keeping the old key tuple. + */ + + hp = p->htop; + E = p->stop; + + old_vals = map_get_values(old_mp); + old_keys = map_get_keys(old_mp); + + res = make_map(hp); + mp = (map_t *)hp; + hp += MAP_HEADER_SIZE; + mp->thing_word = MAP_HEADER; + mp->size = num_old; + mp->keys = old_mp->keys; + + /* Get array of key/value pairs to be updated */ + new_p = &Arg(5); + GET_TERM(*new_p, new_key); + + /* Update all values */ + n = Arg(4) / 2; /* Number of values to be updated */ + ASSERT(n > 0); + for (i = 0; i < num_old; i++) { + if (!EQ(*old_keys, new_key)) { + /* Not same keys */ + *hp++ = *old_vals; + } else { + GET_TERM(new_p[1], *hp); + hp++; + n--; + if (n == 0) { + /* + * All updates done. Copy remaining values + * and return the result. + */ + for (i++, old_vals++; i < num_old; i++) { + *hp++ = *old_vals++; + } + ASSERT(hp == p->htop + need); + p->htop = hp; + return res; + } else { + new_p += 2; + GET_TERM(*new_p, new_key); + } + } + old_vals++, old_keys++; + } + + /* + * Updates left. That means that at least one the keys in the + * update list did not previously exist. + */ + ASSERT(hp == p->htop + need); + return THE_NON_VALUE; +} +#undef GET_TERM int catchlevel(Process *p) { diff --git a/erts/emulator/beam/beam_load.c b/erts/emulator/beam/beam_load.c index 938fd8f2c9..b589d1c930 100644 --- a/erts/emulator/beam/beam_load.c +++ b/erts/emulator/beam/beam_load.c @@ -3783,6 +3783,8 @@ gen_guard_bif1(LoaderState* stp, GenOpArg Fail, GenOpArg Live, GenOpArg Bif, op->a[1].val = (BeamInstr) (void *) erts_gc_bit_size_1; } else if (bf == byte_size_1) { op->a[1].val = (BeamInstr) (void *) erts_gc_byte_size_1; + } else if (bf == map_size_1) { + op->a[1].val = (BeamInstr) (void *) erts_gc_map_size_1; } else if (bf == abs_1) { op->a[1].val = (BeamInstr) (void *) erts_gc_abs_1; } else if (bf == float_1) { @@ -4376,6 +4378,7 @@ transform_engine(LoaderState* st) Uint* restart; /* Where to restart if current match fails. */ GenOpArg def_vars[TE_MAX_VARS]; /* Default buffer for variables. */ GenOpArg* var = def_vars; + int num_vars = 0; int i; /* General index. */ Uint mask; GenOp* instr; @@ -4578,9 +4581,9 @@ transform_engine(LoaderState* st) { int n = *pc++; int formal_arity = gen_opc[instr->op].arity; - int num_vars = n + (instr->arity - formal_arity); int j = formal_arity; + num_vars = n + (instr->arity - formal_arity); var = erts_alloc(ERTS_ALC_T_LOADER_TMP, num_vars * sizeof(GenOpArg)); for (i = 0; i < n; i++) { @@ -4592,7 +4595,6 @@ transform_engine(LoaderState* st) } break; #endif - case TOP_next_arg: ap++; break; @@ -4680,6 +4682,20 @@ transform_engine(LoaderState* st) instr->a[ap].val = var[i].val; ap++; break; +#if defined(TOP_store_rest_args) + case TOP_store_rest_args: + { + int n = *pc++; + int num_extra = num_vars - n; + + ASSERT(n <= num_vars); + GENOP_ARITY(instr, instr->arity+num_extra); + memcpy(instr->a, instr->def_args, ap*sizeof(GenOpArg)); + memcpy(instr->a+ap, var+n, num_extra*sizeof(GenOpArg)); + ap += num_extra; + } + break; +#endif case TOP_try_me_else: restart = pc + 1; restart += *pc++; diff --git a/erts/emulator/beam/beam_load.h b/erts/emulator/beam/beam_load.h index 65a8f26d7c..bd22b0c4de 100644 --- a/erts/emulator/beam/beam_load.h +++ b/erts/emulator/beam/beam_load.h @@ -49,6 +49,7 @@ extern void** beam_ops; extern BeamInstr beam_debug_apply[]; extern BeamInstr* em_call_error_handler; extern BeamInstr* em_apply_bif; +extern BeamInstr* em_call_nif; /* * The following variables keep a sorted list of address ranges for diff --git a/erts/emulator/beam/bif.c b/erts/emulator/beam/bif.c index 61c1abedb5..9c4801041f 100644 --- a/erts/emulator/beam/bif.c +++ b/erts/emulator/beam/bif.c @@ -4615,6 +4615,17 @@ BIF_RETTYPE bump_reductions_1(BIF_ALIST_1) BIF_RET2(am_true, reds); } +BIF_RETTYPE erts_internal_cmp_term_2(BIF_ALIST_2) { + int res = CMP_TERM(BIF_ARG_1,BIF_ARG_2); + + /* ensure -1, 0, 1 result */ + if (res < 0) { + BIF_RET(make_small(-1)); + } else if (res > 0) { + BIF_RET(make_small(1)); + } + BIF_RET(make_small(0)); +} /* * Processes doing yield on return in a bif ends up in bif_return_trap(). */ diff --git a/erts/emulator/beam/bif.tab b/erts/emulator/beam/bif.tab index 3ec534f0bc..2d888862bf 100644 --- a/erts/emulator/beam/bif.tab +++ b/erts/emulator/beam/bif.tab @@ -578,6 +578,23 @@ bif os:unsetenv/1 bif re:inspect/2 +ubif erlang:is_map/1 +ubif erlang:map_size/1 +bif maps:to_list/1 +bif maps:find/2 +bif maps:get/2 +bif maps:from_list/1 +bif maps:is_key/2 +bif maps:keys/1 +bif maps:merge/2 +bif maps:new/0 +bif maps:put/3 +bif maps:remove/2 +bif maps:update/3 +bif maps:values/1 + +bif erts_internal:cmp_term/2 + # # Obsolete # diff --git a/erts/emulator/beam/big.c b/erts/emulator/beam/big.c index 2b27b111d8..41a041eba6 100644 --- a/erts/emulator/beam/big.c +++ b/erts/emulator/beam/big.c @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 1996-2013. All Rights Reserved. + * Copyright Ericsson AB 1996-2014. All Rights Reserved. * * The contents of this file are subject to the Erlang Public License, * Version 1.1, (the "License"); you may not use this file except in @@ -1603,9 +1603,11 @@ big_to_double(Wterm x, double* resp) /* * Logic has been copied from erl_bif_guard.c and slightly * modified to use a static instead of dynamic heap + * + * HALFWORD: Return relative term with 'heap' as base. */ Eterm -double_to_big(double x, Eterm *heap) +double_to_big(double x, Eterm *heap, Uint hsz) { int is_negative; int ds; @@ -1633,9 +1635,10 @@ double_to_big(double x, Eterm *heap) sz = BIG_NEED_SIZE(ds); /* number of words including arity */ hp = heap; - res = make_big(hp); + res = make_big_rel(hp, heap); xp = (ErtsDigit*) (hp + 1); + ASSERT(ds < hsz); for (i = ds - 1; i >= 0; i--) { ErtsDigit d; diff --git a/erts/emulator/beam/big.h b/erts/emulator/beam/big.h index 1a7b14170f..d80111822e 100644 --- a/erts/emulator/beam/big.h +++ b/erts/emulator/beam/big.h @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 1996-2013. All Rights Reserved. + * Copyright Ericsson AB 1996-2014. All Rights Reserved. * * The contents of this file are subject to the Erlang Public License, * Version 1.1, (the "License"); you may not use this file except in @@ -141,7 +141,7 @@ Eterm big_lshift(Eterm, Sint, Eterm*); int big_comp (Wterm, Wterm); int big_ucomp (Eterm, Eterm); int big_to_double(Wterm x, double* resp); -Eterm double_to_big(double, Eterm*); +Eterm double_to_big(double, Eterm*, Uint hsz); Eterm small_to_big(Sint, Eterm*); Eterm uint_to_big(Uint, Eterm*); Eterm uword_to_big(UWord, Eterm*); diff --git a/erts/emulator/beam/copy.c b/erts/emulator/beam/copy.c index 23c0fca6aa..3a987e213b 100644 --- a/erts/emulator/beam/copy.c +++ b/erts/emulator/beam/copy.c @@ -27,6 +27,7 @@ #include "erl_process.h" #include "erl_gc.h" #include "big.h" +#include "erl_map.h" #include "erl_binary.h" #include "erl_bits.h" #include "dtrace-wrapper.h" @@ -150,6 +151,24 @@ Uint size_object(Eterm obj) goto pop_next; } break; + case MAP_SUBTAG: + { + Uint n; + map_t *mp; + mp = (map_t*)map_val_rel(obj,base); + ptr = (Eterm *)mp; + n = map_get_size(mp) + 1; + sum += n + 2; + ptr += 2; /* hdr + size words */ + while (n--) { + obj = *ptr++; + if (!IS_CONST(obj)) { + ESTACK_PUSH(s, obj); + } + } + goto pop_next; + } + break; case BIN_MATCHSTATE_SUBTAG: erl_exit(ERTS_ABORT_EXIT, "size_object: matchstate term not allowed"); @@ -318,6 +337,15 @@ Eterm copy_struct(Eterm obj, Uint sz, Eterm** hpp, ErlOffHeap* off_heap) } } break; + case MAP_SUBTAG: + { + i = map_get_size(objp) + 3; + *argp = make_map_rel(htop, dst_base); + while (i--) { + *htop++ = *objp++; + } + } + break; case REFC_BINARY_SUBTAG: { ProcBin* pb; @@ -537,6 +565,10 @@ Eterm copy_shallow(Eterm* ptr, Uint sz, Eterm** hpp, ErlOffHeap* off_heap) } goto off_heap_common; + case MAP_SUBTAG: + *hp++ = *tp++; + sz--; + break; case EXTERNAL_PID_SUBTAG: case EXTERNAL_PORT_SUBTAG: case EXTERNAL_REF_SUBTAG: diff --git a/erts/emulator/beam/erl_alloc.c b/erts/emulator/beam/erl_alloc.c index 8094c6ee2e..64b3e01ed4 100644 --- a/erts/emulator/beam/erl_alloc.c +++ b/erts/emulator/beam/erl_alloc.c @@ -242,32 +242,13 @@ do { \ sys_memcpy((void *) (IP), (void *) &aui__, sizeof(struct au_init)); \ } while (0) -#if ERTS_ALC_DEFAULT_ACUL \ - || ERTS_ALC_DEFAULT_ACUL_LL_ALLOC \ - || ERTS_ALC_DEFAULT_ACUL_EHEAP_ALLOC - -static ERTS_INLINE void -set_default_acul(struct au_init *ip, int acul) -{ - ip->thr_spec = 1; - ip->atype = AOFIRSTFIT; - ip->init.aoff.flavor = AOFF_BF; - ip->init.util.acul = acul; -} - -#endif - static void set_default_sl_alloc_opts(struct au_init *ip) { SET_DEFAULT_ALLOC_OPTS(ip); ip->enable = AU_ALLOC_DEFAULT_ENABLE(1); -#if ERTS_ALC_DEFAULT_ACUL - set_default_acul(ip, ERTS_ALC_DEFAULT_ACUL); -#else ip->thr_spec = 1; ip->atype = GOODFIT; -#endif ip->init.util.name_prefix = "sl_"; ip->init.util.alloc_no = ERTS_ALC_A_SHORT_LIVED; #ifndef SMALL_MEMORY @@ -281,7 +262,7 @@ set_default_sl_alloc_opts(struct au_init *ip) ip->init.util.force = 1; ip->init.util.low_mem = 1; #endif - + ip->init.util.acul = ERTS_ALC_DEFAULT_ACUL; } static void @@ -289,12 +270,8 @@ set_default_std_alloc_opts(struct au_init *ip) { SET_DEFAULT_ALLOC_OPTS(ip); ip->enable = AU_ALLOC_DEFAULT_ENABLE(1); -#if ERTS_ALC_DEFAULT_ACUL - set_default_acul(ip, ERTS_ALC_DEFAULT_ACUL); -#else ip->thr_spec = 1; ip->atype = BESTFIT; -#endif ip->init.util.name_prefix = "std_"; ip->init.util.alloc_no = ERTS_ALC_A_STANDARD; #ifndef SMALL_MEMORY @@ -303,6 +280,7 @@ set_default_std_alloc_opts(struct au_init *ip) ip->init.util.mmbcs = 32*1024; /* Main carrier size */ #endif ip->init.util.ts = ERTS_ALC_MTA_STANDARD; + ip->init.util.acul = ERTS_ALC_DEFAULT_ACUL; } static void @@ -310,13 +288,9 @@ set_default_ll_alloc_opts(struct au_init *ip) { SET_DEFAULT_ALLOC_OPTS(ip); ip->enable = AU_ALLOC_DEFAULT_ENABLE(1); -#if ERTS_ALC_DEFAULT_ACUL_LL_ALLOC - set_default_acul(ip, ERTS_ALC_DEFAULT_ACUL_LL_ALLOC); -#else ip->thr_spec = 0; ip->atype = BESTFIT; ip->init.bf.ao = 1; -#endif ip->init.util.ramv = 0; ip->init.util.mmsbc = 0; ip->init.util.sbct = ~((UWord) 0); @@ -332,6 +306,7 @@ set_default_ll_alloc_opts(struct au_init *ip) ip->init.util.rsbcst = 0; ip->init.util.rsbcmt = 0; ip->init.util.rmbcmt = 0; + ip->init.util.acul = ERTS_ALC_DEFAULT_ACUL_LL_ALLOC; } static void @@ -363,12 +338,8 @@ set_default_eheap_alloc_opts(struct au_init *ip) { SET_DEFAULT_ALLOC_OPTS(ip); ip->enable = AU_ALLOC_DEFAULT_ENABLE(1); -#if ERTS_ALC_DEFAULT_ACUL_EHEAP_ALLOC - set_default_acul(ip, ERTS_ALC_DEFAULT_ACUL_EHEAP_ALLOC); -#else ip->thr_spec = 1; ip->atype = GOODFIT; -#endif ip->init.util.name_prefix = "eheap_"; ip->init.util.alloc_no = ERTS_ALC_A_EHEAP; #ifndef SMALL_MEMORY @@ -382,6 +353,7 @@ set_default_eheap_alloc_opts(struct au_init *ip) ip->init.util.force = 1; ip->init.util.low_mem = 1; #endif + ip->init.util.acul = ERTS_ALC_DEFAULT_ACUL_EHEAP_ALLOC; } static void @@ -389,12 +361,8 @@ set_default_binary_alloc_opts(struct au_init *ip) { SET_DEFAULT_ALLOC_OPTS(ip); ip->enable = AU_ALLOC_DEFAULT_ENABLE(1); -#if ERTS_ALC_DEFAULT_ACUL - set_default_acul(ip, ERTS_ALC_DEFAULT_ACUL); -#else ip->thr_spec = 1; ip->atype = BESTFIT; -#endif ip->init.util.name_prefix = "binary_"; ip->init.util.alloc_no = ERTS_ALC_A_BINARY; #ifndef SMALL_MEMORY @@ -403,6 +371,7 @@ set_default_binary_alloc_opts(struct au_init *ip) ip->init.util.mmbcs = 32*1024; /* Main carrier size */ #endif ip->init.util.ts = ERTS_ALC_MTA_BINARY; + ip->init.util.acul = ERTS_ALC_DEFAULT_ACUL; } static void @@ -410,12 +379,8 @@ set_default_ets_alloc_opts(struct au_init *ip) { SET_DEFAULT_ALLOC_OPTS(ip); ip->enable = AU_ALLOC_DEFAULT_ENABLE(1); -#if ERTS_ALC_DEFAULT_ACUL - set_default_acul(ip, ERTS_ALC_DEFAULT_ACUL); -#else ip->thr_spec = 1; ip->atype = BESTFIT; -#endif ip->init.util.name_prefix = "ets_"; ip->init.util.alloc_no = ERTS_ALC_A_ETS; #ifndef SMALL_MEMORY @@ -424,6 +389,7 @@ set_default_ets_alloc_opts(struct au_init *ip) ip->init.util.mmbcs = 32*1024; /* Main carrier size */ #endif ip->init.util.ts = ERTS_ALC_MTA_ETS; + ip->init.util.acul = ERTS_ALC_DEFAULT_ACUL; } static void @@ -431,12 +397,8 @@ set_default_driver_alloc_opts(struct au_init *ip) { SET_DEFAULT_ALLOC_OPTS(ip); ip->enable = AU_ALLOC_DEFAULT_ENABLE(1); -#if ERTS_ALC_DEFAULT_ACUL - set_default_acul(ip, ERTS_ALC_DEFAULT_ACUL); -#else ip->thr_spec = 1; ip->atype = BESTFIT; -#endif ip->init.util.name_prefix = "driver_"; ip->init.util.alloc_no = ERTS_ALC_A_DRIVER; #ifndef SMALL_MEMORY @@ -445,6 +407,7 @@ set_default_driver_alloc_opts(struct au_init *ip) ip->init.util.mmbcs = 32*1024; /* Main carrier size */ #endif ip->init.util.ts = ERTS_ALC_MTA_DRIVER; + ip->init.util.acul = ERTS_ALC_DEFAULT_ACUL; } static void @@ -453,12 +416,8 @@ set_default_fix_alloc_opts(struct au_init *ip, { SET_DEFAULT_ALLOC_OPTS(ip); ip->enable = AU_ALLOC_DEFAULT_ENABLE(1); -#if ERTS_ALC_DEFAULT_ACUL - set_default_acul(ip, ERTS_ALC_DEFAULT_ACUL); -#else ip->thr_spec = 1; ip->atype = BESTFIT; -#endif ip->init.bf.ao = 1; ip->init.util.name_prefix = "fix_"; ip->init.util.fix_type_size = fix_type_sizes; @@ -469,6 +428,7 @@ set_default_fix_alloc_opts(struct au_init *ip, ip->init.util.mmbcs = 128*1024; /* Main carrier size */ #endif ip->init.util.ts = ERTS_ALC_MTA_FIXED_SIZE; + ip->init.util.acul = ERTS_ALC_DEFAULT_ACUL; } #ifdef ERTS_SMP @@ -560,26 +520,25 @@ strategy_support_carrier_migration(struct au_init *auip) } static ERTS_INLINE void -check_disable_carrier_migration(struct au_init *auip) -{ - if (!strategy_support_carrier_migration(auip) || !auip->thr_spec) - auip->init.util.acul = 0; -} - -static ERTS_INLINE void -ensure_carrier_migration_support(struct au_init *auip) +adjust_carrier_migration_support(struct au_init *auip) { - auip->thr_spec = 1; /* Need thread preferred */ +#ifdef ERTS_SMP + if (auip->init.util.acul) { + auip->thr_spec = -1; /* Need thread preferred */ - /* - * If strategy cannot handle carrier migration, - * default to a strategy that can... - */ - if (!strategy_support_carrier_migration(auip)) { - /* Default to aoffcbf */ - auip->atype = AOFIRSTFIT; - auip->init.aoff.flavor = AOFF_BF; + /* + * If strategy cannot handle carrier migration, + * default to a strategy that can... + */ + if (!strategy_support_carrier_migration(auip)) { + /* Default to aoffcbf */ + auip->atype = AOFIRSTFIT; + auip->init.aoff.flavor = AOFF_BF; + } } +#else + auip->init.util.acul = 0; +#endif } void @@ -670,8 +629,21 @@ erts_alloc_init(int *argc, char **argv, ErtsAllocInitOpts *eaiop) init.fix_alloc.thr_spec = 0; #endif + /* Make adjustments for carrier migration support */ + init.temp_alloc.init.util.acul = 0; + adjust_carrier_migration_support(&init.sl_alloc); + adjust_carrier_migration_support(&init.std_alloc); + adjust_carrier_migration_support(&init.ll_alloc); + adjust_carrier_migration_support(&init.eheap_alloc); + adjust_carrier_migration_support(&init.binary_alloc); + adjust_carrier_migration_support(&init.ets_alloc); + adjust_carrier_migration_support(&init.driver_alloc); + adjust_carrier_migration_support(&init.fix_alloc); + if (init.erts_alloc_config) { /* Adjust flags that erts_alloc_config won't like */ + + /* No thread specific instances */ init.temp_alloc.thr_spec = 0; init.sl_alloc.thr_spec = 0; init.std_alloc.thr_spec = 0; @@ -680,18 +652,19 @@ erts_alloc_init(int *argc, char **argv, ErtsAllocInitOpts *eaiop) init.binary_alloc.thr_spec = 0; init.ets_alloc.thr_spec = 0; init.driver_alloc.thr_spec = 0; - init.fix_alloc.thr_spec = 0; - } - - check_disable_carrier_migration(&init.sl_alloc); - check_disable_carrier_migration(&init.std_alloc); - check_disable_carrier_migration(&init.ll_alloc); - check_disable_carrier_migration(&init.eheap_alloc); - check_disable_carrier_migration(&init.binary_alloc); - check_disable_carrier_migration(&init.ets_alloc); - check_disable_carrier_migration(&init.driver_alloc); - check_disable_carrier_migration(&init.fix_alloc); + init.fix_alloc.thr_spec = 0; + /* No carrier migration */ + init.temp_alloc.init.util.acul = 0; + init.sl_alloc.init.util.acul = 0; + init.std_alloc.init.util.acul = 0; + init.ll_alloc.init.util.acul = 0; + init.eheap_alloc.init.util.acul = 0; + init.binary_alloc.init.util.acul = 0; + init.ets_alloc.init.util.acul = 0; + init.driver_alloc.init.util.acul = 0; + init.fix_alloc.init.util.acul = 0; + } #ifdef ERTS_SMP /* Only temp_alloc can use thread specific interface */ @@ -1290,8 +1263,6 @@ handle_au_arg(struct au_init *auip, break; } } - ensure_carrier_migration_support(auip); - auip->init.util.acul = get_acul_value(auip, sub_param + 4, argv, ip); } else if(has_prefix("asbcst", sub_param)) { @@ -1328,7 +1299,8 @@ handle_au_arg(struct au_init *auip, else { bad_value(param, sub_param + 1, alg); } - check_disable_carrier_migration(auip); + if (!strategy_support_carrier_migration(auip)) + auip->init.util.acul = 0; } else goto bad_switch; @@ -1409,7 +1381,7 @@ handle_au_arg(struct au_init *auip, } else if (res == 0) { auip->thr_spec = 0; - check_disable_carrier_migration(auip); + auip->init.util.acul = 0; break; } goto bad_switch; @@ -1607,7 +1579,7 @@ handle_args(int *argc, char **argv, erts_alc_hndl_args_init_t *init) for (a = 0; a < aui_sz; a++) { aui[a]->thr_spec = 0; - check_disable_carrier_migration(aui[a]); + aui[a]->init.util.acul = 0; aui[a]->init.util.ramv = 0; aui[a]->init.util.lmbcs = 5*1024*1024; } @@ -1754,6 +1726,9 @@ erts_alloc_register_scheduler(void *vesdp) int ix = (int) esdp->no; int aix; +#ifdef ERTS_DIRTY_SCHEDULERS + ASSERT(!ERTS_SCHEDULER_IS_DIRTY(esdp)); +#endif for (aix = ERTS_ALC_A_MIN; aix <= ERTS_ALC_A_MAX; aix++) { ErtsAllocatorThrSpec_t *tspec = &erts_allctr_thr_spec[aix]; esdp->alloc_data.deallctr[aix] = NULL; diff --git a/erts/emulator/beam/erl_async.c b/erts/emulator/beam/erl_async.c index e6d72f569b..f0cec1c53c 100644 --- a/erts/emulator/beam/erl_async.c +++ b/erts/emulator/beam/erl_async.c @@ -602,7 +602,7 @@ unsigned int driver_async_port_key(ErlDrvPort port) ** return values: ** 0 completed ** -1 error -** N handle value (used with async_cancel) +** N handle value ** arguments: ** ix driver index ** key pointer to secedule queue (NULL means round robin) @@ -687,23 +687,3 @@ long driver_async(ErlDrvPort ix, unsigned int* key, return id; } - -int driver_async_cancel(unsigned int id) -{ - /* - * Not supported anymore. Always fail (which is backward - * compatible). - * - * This functionality could be implemented again. However, - * it is (and always has been) completely useless since - * it doesn't give you any guarantees whatsoever. The user - * needs to (and always have had to) synchronize in his/her - * own code in order to get any guarantees. - */ - return 0; -} - - - - - diff --git a/erts/emulator/beam/erl_bif_guard.c b/erts/emulator/beam/erl_bif_guard.c index a715756c15..bbd8aa31d9 100644 --- a/erts/emulator/beam/erl_bif_guard.c +++ b/erts/emulator/beam/erl_bif_guard.c @@ -33,6 +33,7 @@ #include "bif.h" #include "big.h" #include "erl_binary.h" +#include "erl_map.h" static Eterm gc_double_to_integer(Process* p, double x, Eterm* reg, Uint live); @@ -455,6 +456,28 @@ Eterm erts_gc_byte_size_1(Process* p, Eterm* reg, Uint live) } } +Eterm erts_gc_map_size_1(Process* p, Eterm* reg, Uint live) +{ + Eterm arg = reg[live]; + if (is_map(arg)) { + map_t *mp = (map_t*)map_val(arg); + Uint size = map_get_size(mp); + if (IS_USMALL(0, size)) { + return make_small(size); + } else { + Eterm* hp; + if (ERTS_NEED_GC(p, BIG_UINT_HEAP_SIZE)) { + erts_garbage_collect(p, BIG_UINT_HEAP_SIZE, reg, live); + } + hp = p->htop; + p->htop += BIG_UINT_HEAP_SIZE; + return uint_to_big(size, hp); + } + } else { + BIF_ERROR(p, BADARG); + } +} + Eterm erts_gc_abs_1(Process* p, Eterm* reg, Uint live) { Eterm arg; diff --git a/erts/emulator/beam/erl_bif_info.c b/erts/emulator/beam/erl_bif_info.c index e0b654cb22..f25b4dbae5 100755 --- a/erts/emulator/beam/erl_bif_info.c +++ b/erts/emulator/beam/erl_bif_info.c @@ -90,6 +90,9 @@ static char erts_system_version[] = ("Erlang/OTP " ERLANG_OTP_RELEASE " [smp:%beu:%beu]" #endif #ifdef USE_THREADS +#ifdef ERTS_DIRTY_SCHEDULERS + " [ds:%beu:%beu:%beu]" +#endif " [async-threads:%d]" #endif #ifdef HIPE @@ -312,7 +315,13 @@ erts_print_system_version(int to, void *arg, Process *c_p) char *ocp = otp_correction_package; #ifdef ERTS_SMP Uint total, online, active; - (void) erts_schedulers_state(&total, &online, &active, 0); +#ifdef ERTS_DIRTY_SCHEDULERS + Uint dirty_cpu, dirty_cpu_onln, dirty_io; + + (void) erts_schedulers_state(&total, &online, &active, &dirty_cpu, &dirty_cpu_onln, &dirty_io, 0); +#else + (void) erts_schedulers_state(&total, &online, &active, NULL, NULL, NULL, 0); +#endif #endif for (i = 0; i < sizeof(otp_correction_package)-4; i++) { if (ocp[i] == '-' && ocp[i+1] == 'r' && ocp[i+2] == 'c') @@ -330,6 +339,9 @@ erts_print_system_version(int to, void *arg, Process *c_p) rc_str #ifdef ERTS_SMP , total, online +#ifdef ERTS_DIRTY_SCHEDULERS + , dirty_cpu, dirty_cpu_onln, dirty_io +#endif #endif #ifdef USE_THREADS , erts_async_max_threads @@ -2477,6 +2489,9 @@ BIF_RETTYPE system_info_1(BIF_ALIST_1) switch (erts_schedulers_state(&total, &online, &active, + NULL, + NULL, + NULL, 1)) { case ERTS_SCHDLR_SSPND_DONE: { Eterm *hp = HAlloc(BIF_P, 4); @@ -2500,7 +2515,7 @@ BIF_RETTYPE system_info_1(BIF_ALIST_1) BIF_RET(make_small(1)); #else Uint total, online, active; - switch (erts_schedulers_state(&total, &online, &active, 1)) { + switch (erts_schedulers_state(&total, &online, &active, NULL, NULL, NULL, 1)) { case ERTS_SCHDLR_SSPND_DONE: BIF_RET(make_small(online)); case ERTS_SCHDLR_SSPND_YIELD_RESTART: @@ -2517,7 +2532,7 @@ BIF_RETTYPE system_info_1(BIF_ALIST_1) BIF_RET(make_small(1)); #else Uint total, online, active; - switch (erts_schedulers_state(&total, &online, &active, 1)) { + switch (erts_schedulers_state(&total, &online, &active, NULL, NULL, NULL, 1)) { case ERTS_SCHDLR_SSPND_DONE: BIF_RET(make_small(active)); case ERTS_SCHDLR_SSPND_YIELD_RESTART: @@ -2529,6 +2544,20 @@ BIF_RETTYPE system_info_1(BIF_ALIST_1) BIF_ERROR(BIF_P, EXC_INTERNAL_ERROR); } #endif +#if defined(ERTS_SMP) && defined(ERTS_DIRTY_SCHEDULERS) + } else if (ERTS_IS_ATOM_STR("dirty_cpu_schedulers", BIF_ARG_1)) { + Uint dirty_cpu; + erts_schedulers_state(NULL, NULL, NULL, &dirty_cpu, NULL, NULL, 1); + BIF_RET(make_small(dirty_cpu)); + } else if (ERTS_IS_ATOM_STR("dirty_cpu_schedulers_online", BIF_ARG_1)) { + Uint dirty_cpu_onln; + erts_schedulers_state(NULL, NULL, NULL, NULL, &dirty_cpu_onln, NULL, 1); + BIF_RET(make_small(dirty_cpu_onln)); + } else if (ERTS_IS_ATOM_STR("dirty_io_schedulers", BIF_ARG_1)) { + Uint dirty_io; + erts_schedulers_state(NULL, NULL, NULL, NULL, NULL, &dirty_io, 1); + BIF_RET(make_small(dirty_io)); +#endif } else if (ERTS_IS_ATOM_STR("run_queues", BIF_ARG_1)) { res = make_small(erts_no_run_queues); BIF_RET(res); diff --git a/erts/emulator/beam/erl_bif_op.c b/erts/emulator/beam/erl_bif_op.c index adac0052d6..37dd6457db 100644 --- a/erts/emulator/beam/erl_bif_op.c +++ b/erts/emulator/beam/erl_bif_op.c @@ -36,6 +36,7 @@ #include "dist.h" #include "erl_version.h" #include "erl_binary.h" +#include "erl_map.h" BIF_RETTYPE and_2(BIF_ALIST_2) { @@ -321,7 +322,10 @@ BIF_RETTYPE is_record_3(BIF_ALIST_3) BIF_RET(am_false); } - - - - +BIF_RETTYPE is_map_1(BIF_ALIST_1) +{ + if (is_map(BIF_ARG_1)) { + BIF_RET(am_true); + } + BIF_RET(am_false); +} diff --git a/erts/emulator/beam/erl_db_util.c b/erts/emulator/beam/erl_db_util.c index ef3749a2c4..3986ccd4d3 100644 --- a/erts/emulator/beam/erl_db_util.c +++ b/erts/emulator/beam/erl_db_util.c @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 1998-2013. All Rights Reserved. + * Copyright Ericsson AB 1998-2014. All Rights Reserved. * * The contents of this file are subject to the Erlang Public License, * Version 1.1, (the "License"); you may not use this file except in @@ -35,6 +35,7 @@ #include "bif.h" #include "big.h" #include "erl_binary.h" +#include "erl_map.h" #include "erl_thr_progress.h" #include "erl_db_util.h" @@ -565,6 +566,12 @@ static DMCGuardBif guard_tab[] = DBIF_ALL }, { + am_is_map, + &is_map_1, + 1, + DBIF_ALL + }, + { am_is_binary, &is_binary_1, 1, @@ -631,6 +638,12 @@ static DMCGuardBif guard_tab[] = DBIF_ALL }, { + am_map_size, + &map_size_1, + 1, + DBIF_ALL + }, + { am_bit_size, &bit_size_1, 1, @@ -1838,7 +1851,7 @@ restart: ep = termp; break; case matchArrayBind: /* When the array size is unknown. */ - ASSERT(termp); + ASSERT(termp || arity==0); n = *pc++; variables[n].term = dpm_array_to_list(psp, termp, arity); break; diff --git a/erts/emulator/beam/erl_debug.c b/erts/emulator/beam/erl_debug.c index 873a9860da..50bdc79506 100644 --- a/erts/emulator/beam/erl_debug.c +++ b/erts/emulator/beam/erl_debug.c @@ -29,6 +29,7 @@ #include "bif.h" #include "beam_catches.h" #include "erl_debug.h" +#include "erl_map.h" #define WITHIN(ptr, x, y) ((x) <= (ptr) && (ptr) < (y)) diff --git a/erts/emulator/beam/erl_driver.h b/erts/emulator/beam/erl_driver.h index 5cffae92be..ab9ee63104 100644 --- a/erts/emulator/beam/erl_driver.h +++ b/erts/emulator/beam/erl_driver.h @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 1999-2013. All Rights Reserved. + * Copyright Ericsson AB 1999-2014. All Rights Reserved. * * The contents of this file are subject to the Erlang Public License, * Version 1.1, (the "License"); you may not use this file except in @@ -132,8 +132,8 @@ typedef struct { #define DO_WRITE ERL_DRV_WRITE #define ERL_DRV_EXTENDED_MARKER (0xfeeeeeed) -#define ERL_DRV_EXTENDED_MAJOR_VERSION 2 -#define ERL_DRV_EXTENDED_MINOR_VERSION 2 +#define ERL_DRV_EXTENDED_MAJOR_VERSION 3 +#define ERL_DRV_EXTENDED_MINOR_VERSION 0 /* * The emulator will refuse to load a driver with different major @@ -605,6 +605,8 @@ EXTERN int null_func(void); #define ERL_DRV_INT64 ((ErlDrvTermData) 15) /* ErlDrvSInt64 * */ #define ERL_DRV_UINT64 ((ErlDrvTermData) 16) /* ErlDrvUInt64 * */ +#define ERL_DRV_MAP ((ErlDrvTermData) 17) /* ErlDrvUInt */ + #ifndef ERL_DRIVER_TYPES_ONLY /* make terms for driver_output_term and driver_send_term */ @@ -657,12 +659,6 @@ EXTERN long driver_async(ErlDrvPort ix, void* async_data, void (*async_free)(void*)); -/* - * driver_async_cancel() is deprecated. It is scheduled for removal - * in OTP-R16. For more information see the erl_driver(3) documentation. - */ -EXTERN int driver_async_cancel(unsigned int key) ERL_DRV_DEPRECATED_FUNC; - /* Locks the driver in the machine "forever", there is no unlock function. Note that this is almost never useful, as an open port towards the driver locks it until the port is closed, why unexpected diff --git a/erts/emulator/beam/erl_gc.c b/erts/emulator/beam/erl_gc.c index 8ff6f9a3b9..2022f70cbb 100644 --- a/erts/emulator/beam/erl_gc.c +++ b/erts/emulator/beam/erl_gc.c @@ -28,6 +28,7 @@ #include "beam_catches.h" #include "erl_binary.h" #include "erl_bits.h" +#include "erl_map.h" #include "error.h" #include "big.h" #include "erl_gc.h" @@ -1157,7 +1158,7 @@ do_minor(Process *p, Uint new_sz, Eterm* objv, int nobj) old_htop = sweep_one_area(OLD_HTOP(p), old_htop, heap, heap_size); } OLD_HTOP(p) = old_htop; - HIGH_WATER(p) = (HEAP_START(p) != HIGH_WATER(p)) ? n_heap : n_htop; + HIGH_WATER(p) = n_htop; if (MSO(p).first) { sweep_off_heap(p, 0); diff --git a/erts/emulator/beam/erl_gc.h b/erts/emulator/beam/erl_gc.h index 1801df359a..5203dda263 100644 --- a/erts/emulator/beam/erl_gc.h +++ b/erts/emulator/beam/erl_gc.h @@ -20,6 +20,8 @@ #ifndef __ERL_GC_H__ #define __ERL_GC_H__ +#include "erl_map.h" + /* GC declarations shared by beam/erl_gc.c and hipe/hipe_gc.c */ #if defined(DEBUG) && !ERTS_GLB_INLINE_INCL_FUNC_DEF @@ -42,23 +44,24 @@ do { \ HTOP += 2; /* update tospace htop */ \ } while(0) -#define MOVE_BOXED(PTR,HDR,HTOP,ORIG) \ -do { \ - Eterm gval; \ - Sint nelts; \ - \ - ASSERT(is_header(HDR)); \ - gval = make_boxed(HTOP); \ - *ORIG = gval; \ - *HTOP++ = HDR; \ - *PTR++ = gval; \ - nelts = header_arity(HDR); \ - switch ((HDR) & _HEADER_SUBTAG_MASK) { \ - case SUB_BINARY_SUBTAG: nelts++; break; \ - case FUN_SUBTAG: nelts+=((ErlFunThing*)(PTR-1))->num_free+1; break; \ - } \ - while (nelts--) \ - *HTOP++ = *PTR++; \ +#define MOVE_BOXED(PTR,HDR,HTOP,ORIG) \ +do { \ + Eterm gval; \ + Sint nelts; \ + \ + ASSERT(is_header(HDR)); \ + nelts = header_arity(HDR); \ + switch ((HDR) & _HEADER_SUBTAG_MASK) { \ + case SUB_BINARY_SUBTAG: nelts++; break; \ + case MAP_SUBTAG: nelts+=map_get_size(PTR) + 1; break; \ + case FUN_SUBTAG: nelts+=((ErlFunThing*)(PTR))->num_free+1; break; \ + } \ + gval = make_boxed(HTOP); \ + *ORIG = gval; \ + *HTOP++ = HDR; \ + *PTR++ = gval; \ + while (nelts--) *HTOP++ = *PTR++; \ + \ } while(0) #define in_area(ptr,start,nbytes) \ diff --git a/erts/emulator/beam/erl_init.c b/erts/emulator/beam/erl_init.c index 19088fd913..c17256f466 100644 --- a/erts/emulator/beam/erl_init.c +++ b/erts/emulator/beam/erl_init.c @@ -484,98 +484,107 @@ void erts_usage(void) /* erts_fprintf(stderr, "-# number set the number of items to be used in traces etc\n"); */ - erts_fprintf(stderr, "-a size suggested stack size in kilo words for threads\n"); - erts_fprintf(stderr, " in the async-thread pool, valid range is [%d-%d]\n", + erts_fprintf(stderr, "-a size suggested stack size in kilo words for threads\n"); + erts_fprintf(stderr, " in the async-thread pool, valid range is [%d-%d]\n", ERTS_ASYNC_THREAD_MIN_STACK_SIZE, ERTS_ASYNC_THREAD_MAX_STACK_SIZE); - erts_fprintf(stderr, "-A number set number of threads in async thread pool,\n"); - erts_fprintf(stderr, " valid range is [0-%d]\n", + erts_fprintf(stderr, "-A number set number of threads in async thread pool,\n"); + erts_fprintf(stderr, " valid range is [0-%d]\n", ERTS_MAX_NO_OF_ASYNC_THREADS); - erts_fprintf(stderr, "-B[c|d|i] c to have Ctrl-c interrupt the Erlang shell,\n"); - erts_fprintf(stderr, " d (or no extra option) to disable the break\n"); - erts_fprintf(stderr, " handler, i to ignore break signals\n"); + erts_fprintf(stderr, "-B[c|d|i] c to have Ctrl-c interrupt the Erlang shell,\n"); + erts_fprintf(stderr, " d (or no extra option) to disable the break\n"); + erts_fprintf(stderr, " handler, i to ignore break signals\n"); /* erts_fprintf(stderr, "-b func set the boot function (default boot)\n"); */ - erts_fprintf(stderr, "-c disable continuous date/time correction with\n"); - erts_fprintf(stderr, " respect to uptime\n"); + erts_fprintf(stderr, "-c disable continuous date/time correction with\n"); + erts_fprintf(stderr, " respect to uptime\n"); - erts_fprintf(stderr, "-d don't write a crash dump for internally detected errors\n"); - erts_fprintf(stderr, " (halt(String) will still produce a crash dump)\n"); - erts_fprintf(stderr, "-fn[u|a|l] Control how filenames are interpreted\n"); - erts_fprintf(stderr, "-hms size set minimum heap size in words (default %d)\n", + erts_fprintf(stderr, "-d don't write a crash dump for internally detected errors\n"); + erts_fprintf(stderr, " (halt(String) will still produce a crash dump)\n"); + erts_fprintf(stderr, "-fn[u|a|l] Control how filenames are interpreted\n"); + erts_fprintf(stderr, "-hms size set minimum heap size in words (default %d)\n", H_DEFAULT_SIZE); - erts_fprintf(stderr, "-hmbs size set minimum binary virtual heap size in words (default %d)\n", + erts_fprintf(stderr, "-hmbs size set minimum binary virtual heap size in words (default %d)\n", VH_DEFAULT_SIZE); /* erts_fprintf(stderr, "-i module set the boot module (default init)\n"); */ - erts_fprintf(stderr, "-K boolean enable or disable kernel poll\n"); - erts_fprintf(stderr, "-n[s|a|d] Control behavior of signals to ports\n"); - erts_fprintf(stderr, " Note that this flag is deprecated!\n"); - erts_fprintf(stderr, "-M<X> <Y> memory allocator switches,\n"); - erts_fprintf(stderr, " see the erts_alloc(3) documentation for more info.\n"); - erts_fprintf(stderr, "-pc <set> Control what characters are considered printable (default latin1)\n"); - erts_fprintf(stderr, "-P number set maximum number of processes on this node,\n"); - erts_fprintf(stderr, " valid range is [%d-%d]\n", + erts_fprintf(stderr, "-K boolean enable or disable kernel poll\n"); + erts_fprintf(stderr, "-n[s|a|d] Control behavior of signals to ports\n"); + erts_fprintf(stderr, " Note that this flag is deprecated!\n"); + erts_fprintf(stderr, "-M<X> <Y> memory allocator switches,\n"); + erts_fprintf(stderr, " see the erts_alloc(3) documentation for more info.\n"); + erts_fprintf(stderr, "-pc <set> Control what characters are considered printable (default latin1)\n"); + erts_fprintf(stderr, "-P number set maximum number of processes on this node,\n"); + erts_fprintf(stderr, " valid range is [%d-%d]\n", ERTS_MIN_PROCESSES, ERTS_MAX_PROCESSES); - erts_fprintf(stderr, "-Q number set maximum number of ports on this node,\n"); - erts_fprintf(stderr, " valid range is [%d-%d]\n", + erts_fprintf(stderr, "-Q number set maximum number of ports on this node,\n"); + erts_fprintf(stderr, " valid range is [%d-%d]\n", ERTS_MIN_PORTS, ERTS_MAX_PORTS); - erts_fprintf(stderr, "-R number set compatibility release number,\n"); - erts_fprintf(stderr, " valid range [%d-%d]\n", + erts_fprintf(stderr, "-R number set compatibility release number,\n"); + erts_fprintf(stderr, " valid range [%d-%d]\n", this_rel-2, this_rel); - erts_fprintf(stderr, "-r force ets memory block to be moved on realloc\n"); - erts_fprintf(stderr, "-rg amount set reader groups limit\n"); - erts_fprintf(stderr, "-sbt type set scheduler bind type, valid types are:\n"); - erts_fprintf(stderr, "-stbt type u|ns|ts|ps|s|nnts|nnps|tnnps|db\n"); - erts_fprintf(stderr, "-sbwt val set scheduler busy wait threshold, valid values are:\n"); - erts_fprintf(stderr, " none|very_short|short|medium|long|very_long.\n"); - erts_fprintf(stderr, "-scl bool enable/disable compaction of scheduler load,\n"); - erts_fprintf(stderr, " see the erl(1) documentation for more info.\n"); - erts_fprintf(stderr, "-sct cput set cpu topology,\n"); - erts_fprintf(stderr, " see the erl(1) documentation for more info.\n"); + erts_fprintf(stderr, "-r force ets memory block to be moved on realloc\n"); + erts_fprintf(stderr, "-rg amount set reader groups limit\n"); + erts_fprintf(stderr, "-sbt type set scheduler bind type, valid types are:\n"); + erts_fprintf(stderr, "-stbt type u|ns|ts|ps|s|nnts|nnps|tnnps|db\n"); + erts_fprintf(stderr, "-sbwt val set scheduler busy wait threshold, valid values are:\n"); + erts_fprintf(stderr, " none|very_short|short|medium|long|very_long.\n"); + erts_fprintf(stderr, "-scl bool enable/disable compaction of scheduler load,\n"); + erts_fprintf(stderr, " see the erl(1) documentation for more info.\n"); + erts_fprintf(stderr, "-sct cput set cpu topology,\n"); + erts_fprintf(stderr, " see the erl(1) documentation for more info.\n"); #if ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT_OPT - erts_fprintf(stderr, "-sub bool enable/disable scheduler utilization balancing,\n"); + erts_fprintf(stderr, "-sub bool enable/disable scheduler utilization balancing,\n"); #else - erts_fprintf(stderr, "-sub false disable scheduler utilization balancing,\n"); + erts_fprintf(stderr, "-sub false disable scheduler utilization balancing,\n"); #endif - erts_fprintf(stderr, " see the erl(1) documentation for more info.\n"); - erts_fprintf(stderr, "-sws val set scheduler wakeup strategy, valid values are:\n"); - erts_fprintf(stderr, " default|legacy.\n"); - erts_fprintf(stderr, "-swct val set scheduler wake cleanup threshold, valid values are:\n"); - erts_fprintf(stderr, " very_lazy|lazy|medium|eager|very_eager.\n"); - erts_fprintf(stderr, "-swt val set scheduler wakeup threshold, valid values are:\n"); - erts_fprintf(stderr, " very_low|low|medium|high|very_high.\n"); - erts_fprintf(stderr, "-sss size suggested stack size in kilo words for scheduler threads,\n"); - erts_fprintf(stderr, " valid range is [%d-%d]\n", + erts_fprintf(stderr, " see the erl(1) documentation for more info.\n"); + erts_fprintf(stderr, "-sws val set scheduler wakeup strategy, valid values are:\n"); + erts_fprintf(stderr, " default|legacy.\n"); + erts_fprintf(stderr, "-swct val set scheduler wake cleanup threshold, valid values are:\n"); + erts_fprintf(stderr, " very_lazy|lazy|medium|eager|very_eager.\n"); + erts_fprintf(stderr, "-swt val set scheduler wakeup threshold, valid values are:\n"); + erts_fprintf(stderr, " very_low|low|medium|high|very_high.\n"); + erts_fprintf(stderr, "-sss size suggested stack size in kilo words for scheduler threads,\n"); + erts_fprintf(stderr, " valid range is [%d-%d]\n", ERTS_SCHED_THREAD_MIN_STACK_SIZE, ERTS_SCHED_THREAD_MAX_STACK_SIZE); - erts_fprintf(stderr, "-spp Bool set port parallelism scheduling hint\n"); - erts_fprintf(stderr, "-S n1:n2 set number of schedulers (n1), and number of\n"); - erts_fprintf(stderr, " schedulers online (n2), maximum for both\n"); - erts_fprintf(stderr, " numbers is %d\n", + erts_fprintf(stderr, "-spp Bool set port parallelism scheduling hint\n"); + erts_fprintf(stderr, "-S n1:n2 set number of schedulers (n1), and number of\n"); + erts_fprintf(stderr, " schedulers online (n2), maximum for both\n"); + erts_fprintf(stderr, " numbers is %d\n", ERTS_MAX_NO_OF_SCHEDULERS); - erts_fprintf(stderr, "-SP p1:p2 specify schedulers (p1) and schedulers online (p2)\n"); - erts_fprintf(stderr, " as percentages of logical processors configured and logical\n"); - erts_fprintf(stderr, " processors available, respectively\n"); - erts_fprintf(stderr, "-t size set the maximum number of atoms the " - "emulator can handle\n"); - erts_fprintf(stderr, " valid range is [%d-%d]\n", + erts_fprintf(stderr, "-SP p1:p2 specify schedulers (p1) and schedulers online (p2)\n"); + erts_fprintf(stderr, " as percentages of logical processors configured and logical\n"); + erts_fprintf(stderr, " processors available, respectively\n"); +#ifdef ERTS_DIRTY_SCHEDULERS + erts_fprintf(stderr, "-SDcpu n1:n2 set number of dirty CPU schedulers (n1), and number of\n"); + erts_fprintf(stderr, " dirty CPU schedulers online (n2), valid range for both\n"); + erts_fprintf(stderr, " numbers is [1-%d], and n2 must be less than or equal to n1\n", + ERTS_MAX_NO_OF_DIRTY_CPU_SCHEDULERS); + erts_fprintf(stderr, "-SDPcpu p1:p2 specify dirty CPU schedulers (p1) and dirty CPU schedulers\n"); + erts_fprintf(stderr, " online (p2) as percentages of logical processors configured\n"); + erts_fprintf(stderr, " and logical processors available, respectively\n"); + erts_fprintf(stderr, "-SDio n set number of dirty I/O schedulers, valid range is [0-%d]\n", + ERTS_MAX_NO_OF_DIRTY_IO_SCHEDULERS); +#endif + erts_fprintf(stderr, "-t size set the maximum number of atoms the emulator can handle\n"); + erts_fprintf(stderr, " valid range is [%d-%d]\n", MIN_ATOM_TABLE_SIZE, MAX_ATOM_TABLE_SIZE); - erts_fprintf(stderr, "-T number set modified timing level,\n"); - erts_fprintf(stderr, " valid range is [0-%d]\n", + erts_fprintf(stderr, "-T number set modified timing level, valid range is [0-%d]\n", ERTS_MODIFIED_TIMING_LEVELS-1); - erts_fprintf(stderr, "-V print Erlang version\n"); + erts_fprintf(stderr, "-V print Erlang version\n"); - erts_fprintf(stderr, "-v turn on chatty mode (GCs will be reported etc)\n"); + erts_fprintf(stderr, "-v turn on chatty mode (GCs will be reported etc)\n"); - erts_fprintf(stderr, "-W<i|w> set error logger warnings mapping,\n"); - erts_fprintf(stderr, " see error_logger documentation for details\n"); - erts_fprintf(stderr, "-zdbbl size set the distribution buffer busy limit in kilobytes\n"); - erts_fprintf(stderr, " valid range is [1-%d]\n", INT_MAX/1024); + erts_fprintf(stderr, "-W<i|w> set error logger warnings mapping,\n"); + erts_fprintf(stderr, " see error_logger documentation for details\n"); + erts_fprintf(stderr, "-zdbbl size set the distribution buffer busy limit in kilobytes\n"); + erts_fprintf(stderr, " valid range is [1-%d]\n", INT_MAX/1024); erts_fprintf(stderr, "\n"); erts_fprintf(stderr, "Note that if the emulator is started with erlexec (typically\n"); erts_fprintf(stderr, "from the erl script), these flags should be specified with +.\n"); @@ -643,6 +652,13 @@ early_init(int *argc, char **argv) /* int schdlrs_percentage = 100; int schdlrs_onln_percentage = 100; int max_main_threads; +#ifdef ERTS_DIRTY_SCHEDULERS + int dirty_cpu_scheds; + int dirty_cpu_scheds_online; + int dirty_cpu_scheds_pctg = 100; + int dirty_cpu_scheds_onln_pctg = 100; + int dirty_io_scheds; +#endif int max_reader_groups; int reader_groups; char envbuf[21]; /* enough for any 64-bit integer */ @@ -718,6 +734,12 @@ early_init(int *argc, char **argv) /* schdlrs = no_schedulers; schdlrs_onln = no_schedulers_online; +#ifdef ERTS_DIRTY_SCHEDULERS + dirty_cpu_scheds = no_schedulers; + dirty_cpu_scheds_online = no_schedulers_online; + dirty_io_scheds = 10; +#endif + envbufsz = sizeof(envbuf); /* erts_sys_getenv(_raw)() not initialized yet; need erts_sys_getenv__() */ @@ -808,7 +830,121 @@ early_init(int *argc, char **argv) /* VERBOSE(DEBUG_SYSTEM, ("using %d:%d scheduler percentages\n", schdlrs_percentage, schdlrs_onln_percentage)); - } else { + } +#ifdef ERTS_DIRTY_SCHEDULERS + else if (argv[i][2] == 'D') { + char *arg; + char *type = argv[i]+3; + if (strcmp(type, "Pcpu") == 0) { + int ptot, ponln; + arg = get_arg(argv[i]+7, argv[i+1], &i); + switch (sscanf(arg, "%d:%d", &ptot, &ponln)) { + case 0: + switch (sscanf(arg, ":%d", &ponln)) { + case 1: + if (ponln < 0) + goto bad_SDPcpu; + ptot = 100; + goto chk_SDPcpu; + default: + goto bad_SDPcpu; + } + case 1: + if (ptot < 0) + goto bad_SDPcpu; + ponln = ptot < 100 ? ptot : 100; + goto chk_SDPcpu; + case 2: + if (ptot < 0 || ponln < 0) + goto bad_SDPcpu; + chk_SDPcpu: + dirty_cpu_scheds_pctg = ptot; + dirty_cpu_scheds_onln_pctg = ponln; + break; + default: + bad_SDPcpu: + erts_fprintf(stderr, + "bad dirty CPU schedulers percentage specifier %s\n", + arg); + erts_usage(); + break; + } + VERBOSE(DEBUG_SYSTEM, + ("using %d:%d dirty CPU scheduler percentages\n", + dirty_cpu_scheds_pctg, dirty_cpu_scheds_onln_pctg)); + } else if (strcmp(type, "cpu") == 0) { + int tot, onln; + arg = get_arg(argv[i]+6, argv[i+1], &i); + switch (sscanf(arg, "%d:%d", &tot, &onln)) { + case 0: + switch (sscanf(arg, ":%d", &onln)) { + case 1: + tot = no_schedulers; + goto chk_SDcpu; + default: + goto bad_SDcpu; + } + case 1: + onln = tot < dirty_cpu_scheds_online ? + tot : dirty_cpu_scheds_online; + case 2: + chk_SDcpu: + if (tot > 0) + dirty_cpu_scheds = tot; + else + dirty_cpu_scheds = no_schedulers + tot; + if (onln > 0) + dirty_cpu_scheds_online = onln; + else + dirty_cpu_scheds_online = no_schedulers_online + onln; + if (dirty_cpu_scheds < 1 || + ERTS_MAX_NO_OF_DIRTY_CPU_SCHEDULERS < dirty_cpu_scheds) { + erts_fprintf(stderr, + "bad amount of dirty CPU schedulers %d\n", + tot); + erts_usage(); + } + if (dirty_cpu_scheds_online < 1 || + dirty_cpu_scheds < dirty_cpu_scheds_online) { + erts_fprintf(stderr, + "bad amount of dirty CPU schedulers online %d " + "(total amount of dirty CPU schedulers %d)\n", + dirty_cpu_scheds_online, dirty_cpu_scheds); + erts_usage(); + } + break; + default: + bad_SDcpu: + erts_fprintf(stderr, + "bad amount of dirty CPU schedulers %s\n", + arg); + erts_usage(); + break; + } + VERBOSE(DEBUG_SYSTEM, + ("using %d:%d dirty CPU scheduler(s)\n", tot, onln)); + } else if (strcmp(type, "io") == 0) { + arg = get_arg(argv[i]+5, argv[i+1], &i); + dirty_io_scheds = atoi(arg); + if (dirty_io_scheds < 0 || + dirty_io_scheds > ERTS_MAX_NO_OF_DIRTY_IO_SCHEDULERS) { + erts_fprintf(stderr, + "bad number of dirty I/O schedulers %s\n", + arg); + erts_usage(); + } + VERBOSE(DEBUG_SYSTEM, + ("using %d dirty I/O scheduler(s)\n", dirty_io_scheds)); + } else { + erts_fprintf(stderr, + "bad or missing dirty scheduler specifier: %s\n", + argv[i]); + erts_usage(); + break; + } + } +#endif + else { int tot, onln; char *arg = get_arg(argv[i]+2, argv[i+1], &i); switch (sscanf(arg, "%d:%d", &tot, &onln)) { @@ -895,6 +1031,17 @@ early_init(int *argc, char **argv) /* (void)schdlrs_percentage; (void)schdlrs_onln_percentage; #endif +#ifdef ERTS_DIRTY_SCHEDULERS + /* apply any dirty scheduler precentages */ + if (dirty_cpu_scheds_pctg != 100 || dirty_cpu_scheds_onln_pctg != 100) { + dirty_cpu_scheds = dirty_cpu_scheds * dirty_cpu_scheds_pctg / 100; + dirty_cpu_scheds_online = dirty_cpu_scheds_online * dirty_cpu_scheds_onln_pctg / 100; + } + if (dirty_cpu_scheds > schdlrs) + dirty_cpu_scheds = schdlrs; + if (dirty_cpu_scheds_online > schdlrs_onln) + dirty_cpu_scheds_online = schdlrs_onln; +#endif } #ifndef USE_THREADS @@ -907,6 +1054,11 @@ early_init(int *argc, char **argv) /* erts_no_schedulers = (Uint) no_schedulers; #endif +#ifdef ERTS_DIRTY_SCHEDULERS + erts_no_dirty_cpu_schedulers = dirty_cpu_scheds; + erts_no_dirty_cpu_schedulers_online = dirty_cpu_scheds_online; + erts_no_dirty_io_schedulers = dirty_io_scheds; +#endif erts_early_init_scheduling(no_schedulers); alloc_opts.ncpu = ncpu; @@ -924,10 +1076,18 @@ early_init(int *argc, char **argv) /* * * * Unmanaged threads that need to register: * ** Async threads (see erl_async.c) + * ** Dirty scheduler threads */ erts_thr_progress_init(no_schedulers, no_schedulers+2, - erts_async_max_threads); +#ifndef ERTS_DIRTY_SCHEDULERS + erts_async_max_threads +#else + erts_async_max_threads + + erts_no_dirty_cpu_schedulers + + erts_no_dirty_io_schedulers +#endif + ); #endif erts_thr_q_init(); erts_init_utils(); @@ -1392,7 +1552,15 @@ erl_start(int argc, char **argv) break; case 'S' : /* Was handled in early_init() just read past it */ - if (argv[i][2] == 'P') + if (argv[i][2] == 'D') { + char* type = argv[i]+3; + if (strcmp(type, "Pcpu") == 0) + (void) get_arg(argv[i]+7, argv[i+1], &i); + if (strcmp(type, "cpu") == 0) + (void) get_arg(argv[i]+6, argv[i+1], &i); + else if (strcmp(type, "io") == 0) + (void) get_arg(argv[i]+5, argv[i+1], &i); + } else if (argv[i][2] == 'P') (void) get_arg(argv[i]+3, argv[i+1], &i); else (void) get_arg(argv[i]+2, argv[i+1], &i); diff --git a/erts/emulator/beam/erl_lock_check.c b/erts/emulator/beam/erl_lock_check.c index 0dd83fa6ed..a8ff94ac89 100644 --- a/erts/emulator/beam/erl_lock_check.c +++ b/erts/emulator/beam/erl_lock_check.c @@ -123,6 +123,9 @@ static erts_lc_lock_order_t erts_lock_order[] = { { "schdlr_sspnd", NULL }, { "migration_info_update", NULL }, { "run_queue", "address" }, +#ifdef ERTS_DIRTY_SCHEDULERS + { "dirty_run_queue_sleep_list", "address" }, +#endif { "process_table", NULL }, { "cpu_info", NULL }, { "pollset", "address" }, diff --git a/erts/emulator/beam/erl_map.c b/erts/emulator/beam/erl_map.c new file mode 100644 index 0000000000..2fff7f9390 --- /dev/null +++ b/erts/emulator/beam/erl_map.c @@ -0,0 +1,819 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 2014. All Rights Reserved. + * + * The contents of this file are subject to the Erlang Public License, + * Version 1.1, (the "License"); you may not use this file except in + * compliance with the License. You should have received a copy of the + * Erlang Public License along with this software. If not, it can be + * retrieved online at http://www.erlang.org/. + * + * Software distributed under the License is distributed on an "AS IS" + * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See + * the License for the specific language governing rights and limitations + * under the License. + * + * %CopyrightEnd% + * + * Author: Björn-Egil Dahlberg + */ + +#ifdef HAVE_CONFIG_H +# include "config.h" +#endif + +#include "sys.h" +#include "erl_vm.h" +#include "global.h" +#include "erl_process.h" +#include "error.h" +#include "bif.h" + +#include "erl_map.h" + +/* BIFs + * + * DONE: + * - erlang:is_map/1 + * - erlang:map_size/1 + * + * - maps:find/2 + * - maps:from_list/1 + * - maps:get/2 + * - maps:is_key/2 + * - maps:keys/1 + * - maps:merge/2 + * - maps:new/0 + * - maps:put/3 + * - maps:remove/2 + * - maps:to_list/1 + * - maps:update/3 + * - maps:values/1 + * + * TODO: + * - maps:foldl/3 + * - maps:foldr/3 + * - maps:map/3 + * - maps:size/1 + * - maps:without/2 + * + */ + +/* erlang:map_size/1 + * the corresponding instruction is implemented in: + * beam/erl_bif_guard.c + */ + +BIF_RETTYPE map_size_1(BIF_ALIST_1) { + if (is_map(BIF_ARG_1)) { + Eterm *hp; + Uint hsz = 0; + map_t *mp = (map_t*)map_val(BIF_ARG_1); + Uint n = map_get_size(mp); + + erts_bld_uint(NULL, &hsz, n); + hp = HAlloc(BIF_P, hsz); + BIF_RET(erts_bld_uint(&hp, NULL, n)); + } + + BIF_ERROR(BIF_P, BADARG); +} + +/* maps:to_list/1 + */ + +BIF_RETTYPE maps_to_list_1(BIF_ALIST_1) { + if (is_map(BIF_ARG_1)) { + Uint n; + Eterm* hp; + Eterm *ks,*vs, res, tup; + map_t *mp = (map_t*)map_val(BIF_ARG_1); + + ks = map_get_keys(mp); + vs = map_get_values(mp); + n = map_get_size(mp); + hp = HAlloc(BIF_P, (2 + 3) * n); + res = NIL; + + while(n--) { + tup = TUPLE2(hp, ks[n], vs[n]); hp += 3; + res = CONS(hp, tup, res); hp += 2; + } + + BIF_RET(res); + } + + BIF_ERROR(BIF_P, BADARG); +} + +/* maps:find/2 + * return value if key *matches* a key in the map + */ + +int erts_maps_find(Eterm key, Eterm map, Eterm *value) { + + Eterm *ks,*vs; + map_t *mp; + Uint n,i; + + mp = (map_t*)map_val(map); + n = map_get_size(mp); + ks = map_get_keys(mp); + vs = map_get_values(mp); + + for( i = 0; i < n; i++) { + if (EQ(ks[i], key)) { + *value = vs[i]; + return 1; + } + } + return 0; +} + +BIF_RETTYPE maps_find_2(BIF_ALIST_2) { + if (is_map(BIF_ARG_2)) { + Eterm *hp, value,res; + + if (erts_maps_find(BIF_ARG_1, BIF_ARG_2, &value)) { + hp = HAlloc(BIF_P, 3); + res = make_tuple(hp); + *hp++ = make_arityval(2); + *hp++ = am_ok; + *hp++ = value; + BIF_RET(res); + } + + BIF_RET(am_error); + } + BIF_ERROR(BIF_P, BADARG); +} +/* maps:get/2 + * return value if key *matches* a key in the map + * exception bad_key if none matches + */ + + +int erts_maps_get(Eterm key, Eterm map, Eterm *value) { + Eterm *ks,*vs; + map_t *mp; + Uint n,i; + + mp = (map_t*)map_val(map); + n = map_get_size(mp); + + if (n == 0) + return 0; + + ks = map_get_keys(mp); + vs = map_get_values(mp); + + if (is_immed(key)) { + for( i = 0; i < n; i++) { + if (ks[i] == key) { + *value = vs[i]; + return 1; + } + } + } + + for( i = 0; i < n; i++) { + if (EQ(ks[i], key)) { + *value = vs[i]; + return 1; + } + } + return 0; +} + +BIF_RETTYPE maps_get_2(BIF_ALIST_2) { + if (is_map(BIF_ARG_2)) { + Eterm *hp; + Eterm value, error; + char *s_error; + + if (erts_maps_get(BIF_ARG_1, BIF_ARG_2, &value)) { + BIF_RET(value); + } + + s_error = "bad_key"; + error = am_atom_put(s_error, sys_strlen(s_error)); + + hp = HAlloc(BIF_P, 3); + BIF_P->fvalue = TUPLE2(hp, error, BIF_ARG_1); + BIF_ERROR(BIF_P, EXC_ERROR_2); + } + BIF_ERROR(BIF_P, BADARG); +} + +/* maps:from_list/1 + * List may be unsorted [{K,V}] + */ + +BIF_RETTYPE maps_from_list_1(BIF_ALIST_1) { + Eterm *kv, item = BIF_ARG_1; + Eterm *hp, *thp,*vs, *ks, keys, res; + map_t *mp; + Uint size = 0, unused_size = 0; + Sint c = 0; + Sint idx = 0; + + if (is_list(item) || is_nil(item)) { + + /* Calculate size and check validity */ + + while(is_list(item)) { + res = CAR(list_val(item)); + if (is_not_tuple(res)) + goto error; + + kv = tuple_val(res); + if (*kv != make_arityval(2)) + goto error; + + size++; + item = CDR(list_val(item)); + } + + if (is_not_nil(item)) + goto error; + + hp = HAlloc(BIF_P, 3 + 1 + (2 * size)); + thp = hp; + keys = make_tuple(hp); + *hp++ = make_arityval(size); + ks = hp; + hp += size; + mp = (map_t*)hp; + res = make_map(mp); + hp += MAP_HEADER_SIZE; + vs = hp; + + mp->thing_word = MAP_HEADER; + mp->size = size; /* set later, might shrink*/ + mp->keys = keys; + + if (size == 0) + BIF_RET(res); + + item = BIF_ARG_1; + + /* first entry */ + kv = tuple_val(CAR(list_val(item))); + ks[0] = kv[1]; + vs[0] = kv[2]; + size = 1; + item = CDR(list_val(item)); + + /* insert sort key/value pairs */ + while(is_list(item)) { + + kv = tuple_val(CAR(list_val(item))); + + /* compare ks backwards + * idx represent word index to be written (hole position). + * We cannot copy the elements when searching since we might + * have an equal key. So we search for just the index first =( + * + * It is perhaps faster to move the values in the first pass. + * Check for uniqueness during insert phase and then have a + * second phace compacting the map if duplicates are found + * during insert. .. or do someother sort .. shell-sort perhaps. + */ + + idx = size; + + while(idx > 0 && (c = CMP_TERM(kv[1],ks[idx-1])) < 0) { idx--; } + + if (c == 0) { + /* last compare was equal, + * i.e. we have to release memory + * and overwrite that key/value + */ + ks[idx-1] = kv[1]; + vs[idx-1] = kv[2]; + unused_size++; + } else { + Uint i = size; + while(i > idx) { + ks[i] = ks[i-1]; + vs[i] = vs[i-1]; + i--; + } + ks[idx] = kv[1]; + vs[idx] = kv[2]; + size++; + } + item = CDR(list_val(item)); + } + + if (unused_size) { + /* the key tuple is embedded in the heap + * write a bignum to clear it. + */ + /* release values as normal since they are on the top of the heap */ + + ks[size] = make_pos_bignum_header(unused_size - 1); + HRelease(BIF_P, vs + size + unused_size, vs + size); + } + + *thp = make_arityval(size); + mp->size = size; + BIF_RET(res); + } + +error: + + BIF_ERROR(BIF_P, BADARG); +} + +/* maps:is_key/2 + */ + +BIF_RETTYPE maps_is_key_2(BIF_ALIST_2) { + if (is_map(BIF_ARG_2)) { + Eterm *ks, key; + map_t *mp; + Uint n,i; + + mp = (map_t*)map_val(BIF_ARG_2); + key = BIF_ARG_1; + n = map_get_size(mp); + ks = map_get_keys(mp); + + if (n == 0) + BIF_RET(am_false); + + if (is_immed(key)) { + for( i = 0; i < n; i++) { + if (ks[i] == key) { + BIF_RET(am_true); + } + } + } + + for( i = 0; i < n; i++) { + if (EQ(ks[i], key)) { + BIF_RET(am_true); + } + } + BIF_RET(am_false); + } + BIF_ERROR(BIF_P, BADARG); +} + +/* maps:keys/1 + */ + +BIF_RETTYPE maps_keys_1(BIF_ALIST_1) { + if (is_map(BIF_ARG_1)) { + Eterm *hp, *ks, res = NIL; + map_t *mp; + Uint n; + + mp = (map_t*)map_val(BIF_ARG_1); + n = map_get_size(mp); + + if (n == 0) + BIF_RET(res); + + hp = HAlloc(BIF_P, (2 * n)); + ks = map_get_keys(mp); + + while(n--) { + res = CONS(hp, ks[n], res); hp += 2; + } + + BIF_RET(res); + } + BIF_ERROR(BIF_P, BADARG); +} +/* maps:merge/2 + */ + +BIF_RETTYPE maps_merge_2(BIF_ALIST_2) { + if (is_map(BIF_ARG_1) && is_map(BIF_ARG_2)) { + Eterm *hp,*thp; + Eterm tup; + Eterm *ks,*vs,*ks1,*vs1,*ks2,*vs2; + map_t *mp1,*mp2,*mp_new; + Uint n1,n2,i1,i2,need,unused_size=0; + int c = 0; + + mp1 = (map_t*)map_val(BIF_ARG_1); + mp2 = (map_t*)map_val(BIF_ARG_2); + n1 = map_get_size(mp1); + n2 = map_get_size(mp2); + + need = MAP_HEADER_SIZE + 1 + 2*(n1 + n2); + + hp = HAlloc(BIF_P, need); + thp = hp; + tup = make_tuple(thp); + ks = hp + 1; hp += 1 + n1 + n2; + mp_new = (map_t*)hp; hp += MAP_HEADER_SIZE; + vs = hp; hp += n1 + n2; + + mp_new->thing_word = MAP_HEADER; + mp_new->size = 0; + mp_new->keys = tup; + + i1 = 0; i2 = 0; + ks1 = map_get_keys(mp1); + vs1 = map_get_values(mp1); + ks2 = map_get_keys(mp2); + vs2 = map_get_values(mp2); + + while(i1 < n1 && i2 < n2) { + c = CMP_TERM(ks1[i1],ks2[i2]); + if ( c == 0) { + /* use righthand side arguments map value, + * but advance both maps */ + *ks++ = ks2[i2]; + *vs++ = vs2[i2]; + i1++, i2++, unused_size++; + } else if ( c < 0) { + *ks++ = ks1[i1]; + *vs++ = vs1[i1]; + i1++; + } else { + *ks++ = ks2[i2]; + *vs++ = vs2[i2]; + i2++; + } + } + + /* copy remaining */ + while (i1 < n1) { + *ks++ = ks1[i1]; + *vs++ = vs1[i1]; + i1++; + } + + while (i2 < n2) { + *ks++ = ks2[i2]; + *vs++ = vs2[i2]; + i2++; + } + + if (unused_size) { + /* the key tuple is embedded in the heap, write a bignum to clear it. + * + * release values as normal since they are on the top of the heap + * size = n1 + n1 - unused_size + */ + + *ks = make_pos_bignum_header(unused_size - 1); + HRelease(BIF_P, vs + unused_size, vs); + } + + mp_new->size = n1 + n2 - unused_size; + *thp = make_arityval(n1 + n2 - unused_size); + + BIF_RET(make_map(mp_new)); + } + BIF_ERROR(BIF_P, BADARG); +} +/* maps:new/2 + */ + +BIF_RETTYPE maps_new_0(BIF_ALIST_0) { + Eterm* hp; + Eterm tup; + map_t *mp; + + hp = HAlloc(BIF_P, (MAP_HEADER_SIZE + 1)); + tup = make_tuple(hp); + *hp++ = make_arityval(0); + + mp = (map_t*)hp; + mp->thing_word = MAP_HEADER; + mp->size = 0; + mp->keys = tup; + + BIF_RET(make_map(mp)); +} + +/* maps:put/3 + */ + +Eterm erts_maps_put(Process *p, Eterm key, Eterm value, Eterm map) { + Sint n,i; + Sint c = 0; + Eterm* hp, *shp; + Eterm *ks,*vs, res, tup; + map_t *mp = (map_t*)map_val(map); + + n = map_get_size(mp); + + if (n == 0) { + hp = HAlloc(p, MAP_HEADER_SIZE + 1 + 2); + tup = make_tuple(hp); + *hp++ = make_arityval(1); + *hp++ = key; + res = make_map(hp); + *hp++ = MAP_HEADER; + *hp++ = 1; + *hp++ = tup; + *hp++ = value; + + return res; + } + + ks = map_get_keys(mp); + vs = map_get_values(mp); + + /* only allocate for values, + * assume key-tuple will be intact + */ + + hp = HAlloc(p, MAP_HEADER_SIZE + n); + shp = hp; /* save hp, used if optimistic update fails */ + res = make_map(hp); + *hp++ = MAP_HEADER; + *hp++ = n; + *hp++ = mp->keys; + + if (is_immed(key)) { + for( i = 0; i < n; i ++) { + if (ks[i] == key) { + *hp++ = value; + vs++; + c = 1; + } else { + *hp++ = *vs++; + } + } + } else { + for( i = 0; i < n; i ++) { + if (EQ(ks[i], key)) { + *hp++ = value; + vs++; + c = 1; + } else { + *hp++ = *vs++; + } + } + } + + if (c) + return res; + + /* need to make a new tuple, + * use old hp since it needs to be recreated anyway. + */ + tup = make_tuple(shp); + *shp++ = make_arityval(n+1); + + hp = HAlloc(p, 3 + n + 1); + res = make_map(hp); + *hp++ = MAP_HEADER; + *hp++ = n + 1; + *hp++ = tup; + + ks = map_get_keys(mp); + vs = map_get_values(mp); + + ASSERT(n >= 0); + + /* copy map in order */ + while (n && ((c = CMP_TERM(*ks, key)) < 0)) { + *shp++ = *ks++; + *hp++ = *vs++; + n--; + } + + *shp++ = key; + *hp++ = value; + + ASSERT(n >= 0); + + while(n--) { + *shp++ = *ks++; + *hp++ = *vs++; + } + /* we have one word remaining + * this will work out fine once we get the size word + * in the header. + */ + *shp = make_pos_bignum_header(0); + return res; +} + +BIF_RETTYPE maps_put_3(BIF_ALIST_3) { + if (is_map(BIF_ARG_3)) { + BIF_RET(erts_maps_put(BIF_P, BIF_ARG_1, BIF_ARG_2, BIF_ARG_3)); + } + BIF_ERROR(BIF_P, BADARG); +} + +/* maps:remove/3 + */ + +int erts_maps_remove(Process *p, Eterm key, Eterm map, Eterm *res) { + Sint n; + Uint need; + Eterm *hp_start; + Eterm *thp, *mhp; + Eterm *ks, *vs, tup; + map_t *mp = (map_t*)map_val(map); + + n = map_get_size(mp); + + if (n == 0) { + *res = map; + return 1; + } + + ks = map_get_keys(mp); + vs = map_get_values(mp); + + /* Assume key exists. + * Release allocated if it didn't. + * Allocate key tuple first. + */ + + need = n + 1 - 1 + 3 + n - 1; /* tuple - 1 + map - 1 */ + hp_start = HAlloc(p, need); + thp = hp_start; + mhp = thp + n; /* offset with tuple heap size */ + + tup = make_tuple(thp); + *thp++ = make_arityval(n - 1); + + *res = make_map(mhp); + *mhp++ = MAP_HEADER; + *mhp++ = n - 1; + *mhp++ = tup; + + if (is_immed(key)) { + while(n--) { + if (*ks == key) { + goto found_key; + } else { + *mhp++ = *vs++; + *thp++ = *ks++; + } + } + } else { + while(n--) { + if (EQ(*ks, key)) { + goto found_key; + } else { + *mhp++ = *vs++; + *thp++ = *ks++; + } + } + } + + /* Not found, remove allocated memory + * and return previous map. + */ + HRelease(p, hp_start + need, hp_start); + + *res = map; + return 1; + +found_key: + /* Copy rest of keys and values */ + if (n) { + sys_memcpy(mhp, vs+1, n*sizeof(Eterm)); + sys_memcpy(thp, ks+1, n*sizeof(Eterm)); + } + return 1; +} + +BIF_RETTYPE maps_remove_2(BIF_ALIST_2) { + if (is_map(BIF_ARG_2)) { + Eterm res; + if (erts_maps_remove(BIF_P, BIF_ARG_1, BIF_ARG_2, &res)) { + BIF_RET(res); + } + } + BIF_ERROR(BIF_P, BADARG); +} + +/* maps:update/3 + */ + +int erts_maps_update(Process *p, Eterm key, Eterm value, Eterm map, Eterm *res) { + Sint n,i; + Eterm* hp,*shp; + Eterm *ks,*vs; + map_t *mp = (map_t*)map_val(map); + + if ((n = map_get_size(mp)) == 0) { + return 0; + } + + ks = map_get_keys(mp); + vs = map_get_values(mp); + + /* only allocate for values, + * assume key-tuple will be intact + */ + + hp = HAlloc(p, MAP_HEADER_SIZE + n); + shp = hp; + *hp++ = MAP_HEADER; + *hp++ = n; + *hp++ = mp->keys; + + if (is_immed(key)) { + for( i = 0; i < n; i ++) { + if (ks[i] == key) { + goto found_key; + } else { + *hp++ = *vs++; + } + } + } else { + for( i = 0; i < n; i ++) { + if (EQ(ks[i], key)) { + goto found_key; + } else { + *hp++ = *vs++; + } + } + } + + HRelease(p, shp + MAP_HEADER_SIZE + n, shp); + return 0; + +found_key: + *hp++ = value; + vs++; + if (++i < n) + sys_memcpy(hp, vs, (n - i)*sizeof(Eterm)); + *res = make_map(shp); + return 1; +} + +BIF_RETTYPE maps_update_3(BIF_ALIST_3) { + if (is_map(BIF_ARG_3)) { + Eterm res; + if (erts_maps_update(BIF_P, BIF_ARG_1, BIF_ARG_2, BIF_ARG_3, &res)) { + BIF_RET(res); + } + } + BIF_ERROR(BIF_P, BADARG); +} + + +/* maps:values/1 + */ + +BIF_RETTYPE maps_values_1(BIF_ALIST_1) { + if (is_map(BIF_ARG_1)) { + Eterm *hp, *vs, res = NIL; + map_t *mp; + Uint n; + + mp = (map_t*)map_val(BIF_ARG_1); + n = map_get_size(mp); + + if (n == 0) + BIF_RET(res); + + hp = HAlloc(BIF_P, (2 * n)); + vs = map_get_values(mp); + + while(n--) { + res = CONS(hp, vs[n], res); hp += 2; + } + + BIF_RET(res); + } + BIF_ERROR(BIF_P, BADARG); +} + +int erts_validate_and_sort_map(map_t* mp) +{ + Eterm *ks = map_get_keys(mp); + Eterm *vs = map_get_values(mp); + Uint sz = map_get_size(mp); + Uint ix,jx; + Eterm tmp; + int c; + + /* sort */ + + for (ix = 1; ix < sz; ix++) { + jx = ix; + while( jx > 0 && (c = CMP_TERM(ks[jx],ks[jx-1])) <= 0 ) { + /* identical key -> error */ + if (c == 0) return 0; + + tmp = ks[jx]; + ks[jx] = ks[jx - 1]; + ks[jx - 1] = tmp; + + tmp = vs[jx]; + vs[jx] = vs[jx - 1]; + vs[jx - 1] = tmp; + + jx--; + } + } + return 1; +} diff --git a/erts/emulator/beam/erl_map.h b/erts/emulator/beam/erl_map.h new file mode 100644 index 0000000000..cfacb2ec28 --- /dev/null +++ b/erts/emulator/beam/erl_map.h @@ -0,0 +1,72 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 2014. All Rights Reserved. + * + * The contents of this file are subject to the Erlang Public License, + * Version 1.1, (the "License"); you may not use this file except in + * compliance with the License. You should have received a copy of the + * Erlang Public License along with this software. If not, it can be + * retrieved online at http://www.erlang.org/. + * + * Software distributed under the License is distributed on an "AS IS" + * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See + * the License for the specific language governing rights and limitations + * under the License. + * + * %CopyrightEnd% + */ + + +#ifndef __ERL_MAP_H__ +#define __ERL_MAP_H__ + +#include "sys.h" +/* MAP */ + +typedef struct map_s { + Eterm thing_word; + Uint size; + Eterm keys; /* tuple */ +} map_t; +/* map node + * + * ----------- + * Eterm THING + * Uint size + * Eterm Keys -> {K1, K2, K3, ..., Kn} where n = size + * ---- + * Eterm V1 + * ... + * Eterm Vn, where n = size + * ----------- + */ + + + +/* erl_term.h stuff */ +#define make_map(x) make_boxed((Eterm*)(x)) +#define make_map_rel(x, BASE) make_boxed_rel((Eterm*)(x),(BASE)) +#define is_map(x) (is_boxed((x)) && is_map_header(*boxed_val((x)))) +#define is_map_rel(RTERM,BASE) is_map(rterm2wterm(RTERM,BASE)) +#define is_not_map(x) (!is_map((x))) +#define is_map_header(x) (((x) & (_TAG_HEADER_MASK)) == _TAG_HEADER_MAP) +#define header_is_map(x) ((((x) & (_HEADER_SUBTAG_MASK)) == MAP_SUBTAG)) +#define map_val(x) (_unchecked_boxed_val((x))) +#define map_val_rel(RTERM, BASE) map_val(rterm2wterm(RTERM, BASE)) + +#define map_get_values(x) (((Eterm *)(x)) + 3) +#define map_get_keys(x) (((Eterm *)tuple_val(((map_t *)(x))->keys)) + 1) +#define map_get_size(x) (((map_t*)(x))->size) + +#define MAP_HEADER _make_header(1,_TAG_HEADER_MAP) +#define MAP_HEADER_SIZE (sizeof(map_t) / sizeof(Eterm)) + +Eterm erts_maps_put(Process *p, Eterm key, Eterm value, Eterm map); +int erts_maps_update(Process *p, Eterm key, Eterm value, Eterm map, Eterm *res); +int erts_maps_find(Eterm key, Eterm map, Eterm *value); +int erts_maps_get(Eterm key, Eterm map, Eterm *value); +int erts_maps_remove(Process *p, Eterm key, Eterm map, Eterm *res); +int erts_validate_and_sort_map(map_t* map); +#endif + diff --git a/erts/emulator/beam/erl_nif.c b/erts/emulator/beam/erl_nif.c index dc285b3cf7..c35f1fc2c6 100644 --- a/erts/emulator/beam/erl_nif.c +++ b/erts/emulator/beam/erl_nif.c @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 2009-2013. All Rights Reserved. + * Copyright Ericsson AB 2009-2014. All Rights Reserved. * * The contents of this file are subject to the Erlang Public License, * Version 1.1, (the "License"); you may not use this file except in @@ -31,9 +31,11 @@ #include "bif.h" #include "error.h" #include "big.h" +#include "erl_map.h" #include "beam_bp.h" #include "erl_thr_progress.h" #include "dtrace-wrapper.h" +#include "erl_process.h" #if defined(USE_DYNAMIC_TRACE) && (defined(USE_DTRACE) || defined(USE_SYSTEMTAP)) #define HAVE_USE_DTRACE 1 #endif @@ -1451,6 +1453,347 @@ int enif_consume_timeslice(ErlNifEnv* env, int percent) return ERTS_BIF_REDS_LEFT(env->proc) == 0; } +#ifdef ERTS_DIRTY_SCHEDULERS + +static void +alloc_proc_psd(Process* proc, Export **ep) +{ + int i; + if (!*ep) { + *ep = erts_alloc(ERTS_ALC_T_PSD, sizeof(Export)); + sys_memset((void*) *ep, 0, sizeof(Export)); + for (i=0; i<ERTS_NUM_CODE_IX; i++) { + (*ep)->addressv[i] = &(*ep)->code[3]; + } + (*ep)->code[3] = (BeamInstr) em_call_nif; + } + (void) ERTS_PROC_SET_DIRTY_SCHED_TRAP_EXPORT(proc, ERTS_PROC_LOCK_MAIN, *ep); +} + +static ERL_NIF_TERM +execute_dirty_nif_finalizer(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) +{ + Eterm* reg = ERTS_PROC_GET_SCHDATA(env->proc)->x_reg_array; + ERL_NIF_TERM result = (ERL_NIF_TERM) reg[0]; + typedef ERL_NIF_TERM (*FinalizerFP)(ErlNifEnv*, ERL_NIF_TERM); + FinalizerFP fp; +#if HAVE_INT64 && SIZEOF_LONG != 8 + ASSERT(sizeof(fp) <= sizeof(ErlNifUInt64)); + enif_get_uint64(env, reg[1], (ErlNifUInt64 *) &fp); +#else + ASSERT(sizeof(fp) <= sizeof(unsigned long)); + enif_get_ulong(env, reg[1], (unsigned long *) &fp); +#endif + return (*fp)(env, result); +} + +#endif /* ERTS_DIRTY_SCHEDULERS */ + +#ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT + +ERL_NIF_TERM +enif_schedule_dirty_nif(ErlNifEnv* env, int flags, + ERL_NIF_TERM (*fp)(ErlNifEnv*, int, const ERL_NIF_TERM[]), + int argc, const ERL_NIF_TERM argv[]) +{ +#ifdef USE_THREADS + erts_aint32_t state, n, a; + Process* proc = env->proc; + Eterm* reg = ERTS_PROC_GET_SCHDATA(proc)->x_reg_array; + Export* ep = NULL; + int i; + + int chkflgs = (flags & (ERL_NIF_DIRTY_JOB_IO_BOUND|ERL_NIF_DIRTY_JOB_CPU_BOUND)); + if (chkflgs != ERL_NIF_DIRTY_JOB_IO_BOUND && chkflgs != ERL_NIF_DIRTY_JOB_CPU_BOUND) + return enif_make_badarg(env); + + a = erts_smp_atomic32_read_acqb(&proc->state); + while (1) { + n = state = a; + if (chkflgs == ERL_NIF_DIRTY_JOB_CPU_BOUND) + n |= ERTS_PSFLG_DIRTY_CPU_PROC; + else + n |= ERTS_PSFLG_DIRTY_IO_PROC; + a = erts_smp_atomic32_cmpxchg_mb(&proc->state, n, state); + if (a == state) + break; + } + if (!(ep = ERTS_PROC_GET_DIRTY_SCHED_TRAP_EXPORT(proc))) + alloc_proc_psd(proc, &ep); + ERTS_VBUMP_ALL_REDS(proc); + ep->code[2] = argc; + for (i = 0; i < argc; i++) { + reg[i] = (Eterm) argv[i]; + } + proc->i = (BeamInstr*) ep->addressv[0]; + ep->code[4] = (BeamInstr) fp; + proc->freason = TRAP; + + return THE_NON_VALUE; +#else + return (*fp)(env, argc, argv); +#endif +} + +ERL_NIF_TERM +enif_schedule_dirty_nif_finalizer(ErlNifEnv* env, ERL_NIF_TERM result, + ERL_NIF_TERM (*fp)(ErlNifEnv*, ERL_NIF_TERM)) +{ +#ifdef USE_THREADS + erts_aint32_t state, n, a; + Process* proc = env->proc; + Eterm* reg = ERTS_PROC_GET_SCHDATA(proc)->x_reg_array; + Export* ep; + + a = erts_smp_atomic32_read_acqb(&proc->state); + while (1) { + n = state = a; + if (!(n & (ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q|ERTS_PSFLG_DIRTY_IO_PROC_IN_Q))) + break; + n &= ~(ERTS_PSFLG_DIRTY_CPU_PROC|ERTS_PSFLG_DIRTY_IO_PROC + |ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q|ERTS_PSFLG_DIRTY_IO_PROC_IN_Q); + a = erts_smp_atomic32_cmpxchg_mb(&proc->state, n, state); + if (a == state) + break; + } + if (!(ep = ERTS_PROC_GET_DIRTY_SCHED_TRAP_EXPORT(proc))) + alloc_proc_psd(proc, &ep); + ERTS_VBUMP_ALL_REDS(proc); + ep->code[2] = 2; + reg[0] = (Eterm) result; +#if HAVE_INT64 && SIZEOF_LONG != 8 + ASSERT(sizeof(fp) <= sizeof(ErlNifUInt64)); + reg[1] = (Eterm) enif_make_uint64(env, (ErlNifUInt64) fp); +#else + ASSERT(sizeof(fp) <= sizeof(unsigned long)); + reg[1] = (Eterm) enif_make_ulong(env, (unsigned long) fp); +#endif + proc->i = (BeamInstr*) ep->addressv[0]; + ep->code[4] = (BeamInstr) execute_dirty_nif_finalizer; + proc->freason = TRAP; + + return THE_NON_VALUE; +#else + return (*fp)(env, result); +#endif +} + +/* A simple finalizer that just returns its result argument */ +ERL_NIF_TERM +enif_dirty_nif_finalizer(ErlNifEnv* env, ERL_NIF_TERM result) +{ + return result; +} + +int +enif_is_on_dirty_scheduler(ErlNifEnv* env) +{ + return ERTS_SCHEDULER_IS_DIRTY(env->proc->scheduler_data); +} + +int +enif_have_dirty_schedulers() +{ +#ifdef USE_THREADS + return 1; +#else + return 0; +#endif +} + +#endif /* ERL_NIF_DIRTY_SCHEDULER_SUPPORT */ + +/* Maps */ + +int enif_is_map(ErlNifEnv* env, ERL_NIF_TERM term) +{ + return is_map(term); +} + +int enif_get_map_size(ErlNifEnv* env, ERL_NIF_TERM term, size_t *size) +{ + if (is_map(term)) { + map_t *mp; + mp = (map_t*)map_val(term); + *size = map_get_size(mp); + return 1; + } + return 0; +} + +ERL_NIF_TERM enif_make_new_map(ErlNifEnv* env) +{ + Eterm* hp = alloc_heap(env,MAP_HEADER_SIZE+1); + Eterm tup; + map_t *mp; + + tup = make_tuple(hp); + *hp++ = make_arityval(0); + mp = (map_t*)hp; + mp->thing_word = MAP_HEADER; + mp->size = 0; + mp->keys = tup; + + return make_map(mp); +} + +int enif_make_map_put(ErlNifEnv* env, + Eterm map_in, + Eterm key, + Eterm value, + Eterm *map_out) +{ + if (is_not_map(map_in)) { + return 0; + } + flush_env(env); + *map_out = erts_maps_put(env->proc, key, value, map_in); + cache_env(env); + return 1; +} + +int enif_get_map_value(ErlNifEnv* env, + Eterm map, + Eterm key, + Eterm *value) +{ + if (is_not_map(map)) { + return 0; + } + return erts_maps_get(key, map, value); +} + +int enif_make_map_update(ErlNifEnv* env, + Eterm map_in, + Eterm key, + Eterm value, + Eterm *map_out) +{ + int res; + if (is_not_map(map_in)) { + return 0; + } + + flush_env(env); + res = erts_maps_update(env->proc, key, value, map_in, map_out); + cache_env(env); + return res; +} + +int enif_make_map_remove(ErlNifEnv* env, + Eterm map_in, + Eterm key, + Eterm *map_out) +{ + int res; + if (is_not_map(map_in)) { + return 0; + } + flush_env(env); + res = erts_maps_remove(env->proc, key, map_in, map_out); + cache_env(env); + return res; +} + +int enif_map_iterator_create(ErlNifEnv *env, + Eterm map, + ErlNifMapIterator *iter, + ErlNifMapIteratorEntry entry) +{ + if (is_map(map)) { + map_t *mp = (map_t*)map_val(map); + size_t offset; + + switch (entry) { + case ERL_NIF_MAP_ITERATOR_HEAD: offset = 0; break; + case ERL_NIF_MAP_ITERATOR_TAIL: offset = map_get_size(mp) - 1; break; + default: goto error; + } + + /* empty maps are ok but will leave the iterator + * in bad shape. + */ + + iter->map = map; + iter->ks = ((Eterm *)map_get_keys(mp)) + offset; + iter->vs = ((Eterm *)map_get_values(mp)) + offset; + iter->t_limit = map_get_size(mp) + 1; + iter->idx = offset + 1; + + return 1; + } + +error: +#ifdef DEBUG + iter->map = THE_NON_VALUE; +#endif + return 0; +} + +void enif_map_iterator_destroy(ErlNifEnv *env, ErlNifMapIterator *iter) +{ + /* not used */ +#ifdef DEBUG + iter->map = THE_NON_VALUE; +#endif + +} + +int enif_map_iterator_is_tail(ErlNifEnv *env, ErlNifMapIterator *iter) +{ + ASSERT(iter && is_map(iter->map)); + ASSERT(iter->idx >= 0 && (iter->idx <= map_get_size(map_val(iter->map)) + 1)); + return (iter->t_limit == 1 || iter->idx == iter->t_limit); +} + +int enif_map_iterator_is_head(ErlNifEnv *env, ErlNifMapIterator *iter) +{ + ASSERT(iter && is_map(iter->map)); + ASSERT(iter->idx >= 0 && (iter->idx <= map_get_size(map_val(iter->map)) + 1)); + return (iter->t_limit == 1 || iter->idx == 0); +} + + +int enif_map_iterator_next(ErlNifEnv *env, ErlNifMapIterator *iter) +{ + ASSERT(iter && is_map(iter->map)); + if (iter->idx < iter->t_limit) { + iter->idx++; + iter->ks++; + iter->vs++; + } + return (iter->idx != iter->t_limit); +} + +int enif_map_iterator_prev(ErlNifEnv *env, ErlNifMapIterator *iter) +{ + ASSERT(iter && is_map(iter->map)); + if (iter->idx > 0) { + iter->idx--; + iter->ks--; + iter->vs--; + } + return (iter->idx > 0); +} + +int enif_map_iterator_get_pair(ErlNifEnv *env, + ErlNifMapIterator *iter, + Eterm *key, + Eterm *value) +{ + ASSERT(iter && is_map(iter->map)); + if (iter->idx > 0 && iter->idx < iter->t_limit) { + ASSERT(iter->ks >= map_get_keys(map_val(iter->map)) && + iter->ks < (map_get_keys(map_val(iter->map)) + map_get_size(map_val(iter->map)))); + ASSERT(iter->vs >= map_get_values(map_val(iter->map)) && + iter->vs < (map_get_values(map_val(iter->map)) + map_get_size(map_val(iter->map)))); + *key = *(iter->ks); + *value = *(iter->vs); + return 1; + } + return 0; +} + /*************************************************************************** ** load_nif/2 ** ***************************************************************************/ @@ -1647,7 +1990,8 @@ BIF_RETTYPE load_nif_2(BIF_ALIST_2) ret = load_nif_error(BIF_P, bad_lib, "Library init-call unsuccessful"); } else if (entry->major != ERL_NIF_MAJOR_VERSION - || entry->minor > ERL_NIF_MINOR_VERSION) { + || entry->minor > ERL_NIF_MINOR_VERSION + || (entry->major==2 && entry->minor == 5)) { /* experimental maps */ ret = load_nif_error(BIF_P, bad_lib, "Library version (%d.%d) not compatible (with %d.%d).", entry->major, entry->minor, ERL_NIF_MAJOR_VERSION, ERL_NIF_MINOR_VERSION); diff --git a/erts/emulator/beam/erl_nif.h b/erts/emulator/beam/erl_nif.h index 5f4dc21d5c..7613446f64 100644 --- a/erts/emulator/beam/erl_nif.h +++ b/erts/emulator/beam/erl_nif.h @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 2009-2013. All Rights Reserved. + * Copyright Ericsson AB 2009-2014. All Rights Reserved. * * The contents of this file are subject to the Erlang Public License, * Version 1.1, (the "License"); you may not use this file except in @@ -23,7 +23,11 @@ #ifndef __ERL_NIF_H__ #define __ERL_NIF_H__ +#ifdef HAVE_CONFIG_H +# include "config.h" +#endif +#include "erl_native_features_config.h" #include "erl_drv_nif.h" /* Version history: @@ -34,9 +38,13 @@ ** 2.2: R14B03 enif_is_exception ** 2.3: R15 enif_make_reverse_list, enif_is_number ** 2.4: R16 enif_consume_timeslice +** 2.5: First experimental maps API additions (libs of this version is not compatible with any other VM) +** 2.5: R17 Maps API additions +** 2.6: R17 with maps +** R17 dirty schedulers */ #define ERL_NIF_MAJOR_VERSION 2 -#define ERL_NIF_MINOR_VERSION 4 +#define ERL_NIF_MINOR_VERSION 6 #include <stdlib.h> @@ -95,6 +103,8 @@ typedef unsigned long long ERL_NIF_TERM; # endif #endif +typedef ERL_NIF_TERM ERL_NIF_UINT; + struct enif_environment_t; typedef struct enif_environment_t ErlNifEnv; @@ -159,6 +169,29 @@ typedef int ErlNifTSDKey; typedef ErlDrvThreadOpts ErlNifThreadOpts; +#ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT +typedef enum +{ + ERL_NIF_DIRTY_JOB_CPU_BOUND = 1, + ERL_NIF_DIRTY_JOB_IO_BOUND = 2 +}ErlNifDirtyTaskFlags; +#endif + +typedef struct /* All fields all internal and may change */ +{ + ERL_NIF_TERM map; + ERL_NIF_UINT t_limit; + ERL_NIF_UINT idx; + ERL_NIF_TERM *ks; + ERL_NIF_TERM *vs; + void* __spare__[2]; /* for future additions to be ABI compatible (same struct size) */ +} ErlNifMapIterator; + +typedef enum { + ERL_NIF_MAP_ITERATOR_HEAD = 1, + ERL_NIF_MAP_ITERATOR_TAIL = 2 +} ErlNifMapIteratorEntry; + #if (defined(__WIN32__) || defined(_WIN32) || defined(_WIN32_)) # define ERL_NIF_API_FUNC_DECL(RET_TYPE, NAME, ARGS) RET_TYPE (*NAME) ARGS typedef struct { diff --git a/erts/emulator/beam/erl_nif_api_funcs.h b/erts/emulator/beam/erl_nif_api_funcs.h index 2f841645e1..d7c554e60b 100644 --- a/erts/emulator/beam/erl_nif_api_funcs.h +++ b/erts/emulator/beam/erl_nif_api_funcs.h @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 2009-2013. All Rights Reserved. + * Copyright Ericsson AB 2009-2014. All Rights Reserved. * * The contents of this file are subject to the Erlang Public License, * Version 1.1, (the "License"); you may not use this file except in @@ -141,6 +141,29 @@ ERL_NIF_API_FUNC_DECL(int,enif_is_number,(ErlNifEnv*, ERL_NIF_TERM term)); ERL_NIF_API_FUNC_DECL(void*,enif_dlopen,(const char* lib, void (*err_handler)(void*,const char*), void* err_arg)); ERL_NIF_API_FUNC_DECL(void*,enif_dlsym,(void* handle, const char* symbol, void (*err_handler)(void*,const char*), void* err_arg)); ERL_NIF_API_FUNC_DECL(int,enif_consume_timeslice,(ErlNifEnv*, int percent)); +#ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT +ERL_NIF_API_FUNC_DECL(ERL_NIF_TERM,enif_schedule_dirty_nif,(ErlNifEnv*,int,ERL_NIF_TERM (*)(ErlNifEnv*,int,const ERL_NIF_TERM[]),int,const ERL_NIF_TERM[])); +ERL_NIF_API_FUNC_DECL(ERL_NIF_TERM,enif_schedule_dirty_nif_finalizer,(ErlNifEnv*,ERL_NIF_TERM,ERL_NIF_TERM (*)(ErlNifEnv*,ERL_NIF_TERM))); +ERL_NIF_API_FUNC_DECL(ERL_NIF_TERM,enif_dirty_nif_finalizer,(ErlNifEnv*,ERL_NIF_TERM)); +ERL_NIF_API_FUNC_DECL(int,enif_is_on_dirty_scheduler,(ErlNifEnv*)); +ERL_NIF_API_FUNC_DECL(int,enif_have_dirty_schedulers,(void)); +#endif + +ERL_NIF_API_FUNC_DECL(int, enif_is_map, (ErlNifEnv* env, ERL_NIF_TERM term)); +ERL_NIF_API_FUNC_DECL(int, enif_get_map_size, (ErlNifEnv* env, ERL_NIF_TERM term, size_t *size)); +ERL_NIF_API_FUNC_DECL(ERL_NIF_TERM, enif_make_new_map, (ErlNifEnv* env)); +ERL_NIF_API_FUNC_DECL(int, enif_make_map_put, (ErlNifEnv* env, ERL_NIF_TERM map_in, ERL_NIF_TERM key, ERL_NIF_TERM value, ERL_NIF_TERM* map_out)); +ERL_NIF_API_FUNC_DECL(int, enif_get_map_value, (ErlNifEnv* env, ERL_NIF_TERM map, ERL_NIF_TERM key, ERL_NIF_TERM* value)); +ERL_NIF_API_FUNC_DECL(int, enif_make_map_update, (ErlNifEnv* env, ERL_NIF_TERM map_in, ERL_NIF_TERM key, ERL_NIF_TERM value, ERL_NIF_TERM* map_out)); +ERL_NIF_API_FUNC_DECL(int, enif_make_map_remove, (ErlNifEnv* env, ERL_NIF_TERM map_in, ERL_NIF_TERM key, ERL_NIF_TERM* map_out)); +ERL_NIF_API_FUNC_DECL(int, enif_map_iterator_create, (ErlNifEnv *env, ERL_NIF_TERM map, ErlNifMapIterator *iter, ErlNifMapIteratorEntry entry)); +ERL_NIF_API_FUNC_DECL(void, enif_map_iterator_destroy, (ErlNifEnv *env, ErlNifMapIterator *iter)); +ERL_NIF_API_FUNC_DECL(int, enif_map_iterator_is_head, (ErlNifEnv *env, ErlNifMapIterator *iter)); +ERL_NIF_API_FUNC_DECL(int, enif_map_iterator_is_tail, (ErlNifEnv *env, ErlNifMapIterator *iter)); +ERL_NIF_API_FUNC_DECL(int, enif_map_iterator_next, (ErlNifEnv *env, ErlNifMapIterator *iter)); +ERL_NIF_API_FUNC_DECL(int, enif_map_iterator_prev, (ErlNifEnv *env, ErlNifMapIterator *iter)); +ERL_NIF_API_FUNC_DECL(int, enif_map_iterator_get_pair, (ErlNifEnv *env, ErlNifMapIterator *iter, ERL_NIF_TERM *key, ERL_NIF_TERM *value)); + /* ** Add new entries here to keep compatibility on Windows!!! @@ -266,6 +289,28 @@ ERL_NIF_API_FUNC_DECL(int,enif_consume_timeslice,(ErlNifEnv*, int percent)); # define enif_dlopen ERL_NIF_API_FUNC_MACRO(enif_dlopen) # define enif_dlsym ERL_NIF_API_FUNC_MACRO(enif_dlsym) # define enif_consume_timeslice ERL_NIF_API_FUNC_MACRO(enif_consume_timeslice) +#ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT +# define enif_schedule_dirty_nif ERL_NIF_API_FUNC_MACRO(enif_schedule_dirty_nif) +# define enif_schedule_dirty_nif_finalizer ERL_NIF_API_FUNC_MACRO(enif_schedule_dirty_nif_finalizer) +# define enif_dirty_nif_finalizer ERL_NIF_API_FUNC_MACRO(enif_dirty_nif_finalizer) +# define enif_is_on_dirty_scheduler ERL_NIF_API_FUNC_MACRO(enif_is_on_dirty_scheduler) +# define enif_have_dirty_schedulers ERL_NIF_API_FUNC_MACRO(enif_have_dirty_schedulers) +#endif + +# define enif_is_map ERL_NIF_API_FUNC_MACRO(enif_is_map) +# define enif_get_map_size ERL_NIF_API_FUNC_MACRO(enif_get_map_size) +# define enif_make_new_map ERL_NIF_API_FUNC_MACRO(enif_make_new_map) +# define enif_make_map_put ERL_NIF_API_FUNC_MACRO(enif_make_map_put) +# define enif_get_map_value ERL_NIF_API_FUNC_MACRO(enif_get_map_value) +# define enif_make_map_update ERL_NIF_API_FUNC_MACRO(enif_make_map_update) +# define enif_make_map_remove ERL_NIF_API_FUNC_MACRO(enif_make_map_remove) +# define enif_map_iterator_create ERL_NIF_API_FUNC_MACRO(enif_map_iterator_create) +# define enif_map_iterator_destroy ERL_NIF_API_FUNC_MACRO(enif_map_iterator_destroy) +# define enif_map_iterator_is_head ERL_NIF_API_FUNC_MACRO(enif_map_iterator_is_head) +# define enif_map_iterator_is_tail ERL_NIF_API_FUNC_MACRO(enif_map_iterator_is_tail) +# define enif_map_iterator_next ERL_NIF_API_FUNC_MACRO(enif_map_iterator_next) +# define enif_map_iterator_prev ERL_NIF_API_FUNC_MACRO(enif_map_iterator_prev) +# define enif_map_iterator_get_pair ERL_NIF_API_FUNC_MACRO(enif_map_iterator_get_pair) /* ** Add new entries here diff --git a/erts/emulator/beam/erl_port_task.c b/erts/emulator/beam/erl_port_task.c index 547a42beb2..d4108067d0 100644 --- a/erts/emulator/beam/erl_port_task.c +++ b/erts/emulator/beam/erl_port_task.c @@ -877,6 +877,11 @@ enqueue_port(ErtsRunQueue *runq, Port *pp) ASSERT(runq->ports.start && runq->ports.end); erts_smp_inc_runq_len(runq, &runq->ports.info, ERTS_PORT_PRIO_LEVEL); + +#ifdef ERTS_SMP + if (runq->halt_in_progress) + erts_non_empty_runq(runq); +#endif } static ERTS_INLINE Port * diff --git a/erts/emulator/beam/erl_printf_term.c b/erts/emulator/beam/erl_printf_term.c index 436147749e..d18760dc43 100644 --- a/erts/emulator/beam/erl_printf_term.c +++ b/erts/emulator/beam/erl_printf_term.c @@ -24,6 +24,7 @@ #include "erl_printf_term.h" #include "sys.h" #include "big.h" +#include "erl_map.h" #define PRINT_CHAR(CNT, FN, ARG, C) \ do { \ @@ -216,14 +217,15 @@ static int print_atom_name(fmtfn_t fn, void* arg, Eterm atom, long *dcount) } -#define PRT_BAR ((Eterm) 0) -#define PRT_COMMA ((Eterm) 1) -#define PRT_CLOSE_LIST ((Eterm) 2) -#define PRT_CLOSE_TUPLE ((Eterm) 3) -#define PRT_TERM ((Eterm) 4) -#define PRT_ONE_CONS ((Eterm) 5) -#define PRT_PATCH_FUN_SIZE ((Eterm) 6) -#define PRT_LAST_ARRAY_ELEMENT ((Eterm) 7) /* Note! Must be last... */ +#define PRT_BAR ((Eterm) 0) +#define PRT_COMMA ((Eterm) 1) +#define PRT_CLOSE_LIST ((Eterm) 2) +#define PRT_CLOSE_TUPLE ((Eterm) 3) +#define PRT_ASSOC ((Eterm) 4) +#define PRT_TERM ((Eterm) 5) +#define PRT_ONE_CONS ((Eterm) 6) +#define PRT_PATCH_FUN_SIZE ((Eterm) 7) +#define PRT_LAST_ARRAY_ELEMENT ((Eterm) 8) /* Note! Must be last... */ static int print_term(fmtfn_t fn, void* arg, Eterm obj, long *dcount, @@ -260,6 +262,9 @@ print_term(fmtfn_t fn, void* arg, Eterm obj, long *dcount, case PRT_CLOSE_TUPLE: PRINT_CHAR(res, fn, arg, '}'); goto L_outer_loop; + case PRT_ASSOC: + PRINT_STRING(res, fn, arg, "=>"); + goto L_outer_loop; default: popped.word = WSTACK_POP(s); @@ -483,6 +488,37 @@ print_term(fmtfn_t fn, void* arg, Eterm obj, long *dcount, PRINT_CHAR(res, fn, arg, '>'); } break; + case MAP_DEF: + { + Uint n; + Eterm *ks, *vs; + map_t *mp = (map_t *)map_val(wobj); + n = map_get_size(mp); + ks = map_get_keys(mp); + vs = map_get_values(mp); + + PRINT_CHAR(res, fn, arg, '#'); + PRINT_CHAR(res, fn, arg, '{'); + WSTACK_PUSH(s, PRT_CLOSE_TUPLE); + if (n > 0) { + n--; + WSTACK_PUSH(s, vs[n]); + WSTACK_PUSH(s, PRT_TERM); + WSTACK_PUSH(s, PRT_ASSOC); + WSTACK_PUSH(s, ks[n]); + WSTACK_PUSH(s, PRT_TERM); + + while (n--) { + WSTACK_PUSH(s, PRT_COMMA); + WSTACK_PUSH(s, vs[n]); + WSTACK_PUSH(s, PRT_TERM); + WSTACK_PUSH(s, PRT_ASSOC); + WSTACK_PUSH(s, ks[n]); + WSTACK_PUSH(s, PRT_TERM); + } + } + } + break; default: PRINT_STRING(res, fn, arg, "<unknown:"); PRINT_POINTER(res, fn, arg, wobj); diff --git a/erts/emulator/beam/erl_process.c b/erts/emulator/beam/erl_process.c index 2f383f4c01..937881212a 100644 --- a/erts/emulator/beam/erl_process.c +++ b/erts/emulator/beam/erl_process.c @@ -146,6 +146,11 @@ extern BeamInstr beam_continue_exit[]; int erts_sched_compact_load; int erts_sched_balance_util = 0; Uint erts_no_schedulers; +#ifdef ERTS_DIRTY_SCHEDULERS +Uint erts_no_dirty_cpu_schedulers; +Uint erts_no_dirty_cpu_schedulers_online; +Uint erts_no_dirty_io_schedulers; +#endif #define ERTS_THR_PRGR_LATER_CLEANUP_OP_THRESHOLD_VERY_LAZY (4*1024*1024) #define ERTS_THR_PRGR_LATER_CLEANUP_OP_THRESHOLD_LAZY (512*1024) @@ -259,6 +264,10 @@ ErtsAlignedRunQueue *erts_aligned_run_queues; Uint erts_no_run_queues; ErtsAlignedSchedulerData *erts_aligned_scheduler_data; +#ifdef ERTS_DIRTY_SCHEDULERS +ErtsAlignedSchedulerData *erts_aligned_dirty_cpu_scheduler_data; +ErtsAlignedSchedulerData *erts_aligned_dirty_io_scheduler_data; +#endif typedef union { ErtsSchedulerSleepInfo ssi; @@ -266,6 +275,12 @@ typedef union { } ErtsAlignedSchedulerSleepInfo; static ErtsAlignedSchedulerSleepInfo *aligned_sched_sleep_info; +#ifdef ERTS_DIRTY_SCHEDULERS +#ifdef ERTS_SMP +static ErtsAlignedSchedulerSleepInfo *aligned_dirty_cpu_sched_sleep_info; +static ErtsAlignedSchedulerSleepInfo *aligned_dirty_io_sched_sleep_info; +#endif +#endif static Uint last_reductions; static Uint last_exact_reductions; @@ -332,6 +347,16 @@ ERTS_SCHED_PREF_QUICK_ALLOC_IMPL(proclist, (ASSERT(-1 <= ((int) (IX)) \ && ((int) (IX)) < ((int) erts_no_schedulers)), \ &aligned_sched_sleep_info[(IX)].ssi) +#ifdef ERTS_DIRTY_SCHEDULERS +#define ERTS_DIRTY_CPU_SCHED_SLEEP_INFO_IX(IX) \ + (ASSERT(0 <= ((int) (IX)) \ + && ((int) (IX)) < ((int) erts_no_dirty_cpu_schedulers)), \ + &aligned_dirty_cpu_sched_sleep_info[(IX)].ssi) +#define ERTS_DIRTY_IO_SCHED_SLEEP_INFO_IX(IX) \ + (ASSERT(0 <= ((int) (IX)) \ + && ((int) (IX)) < ((int) erts_no_dirty_io_schedulers)), \ + &aligned_dirty_io_sched_sleep_info[(IX)].ssi) +#endif #define ERTS_FOREACH_RUNQ(RQVAR, DO) \ do { \ @@ -519,6 +544,13 @@ erts_pre_init_process(void) erts_psd_required_locks[ERTS_PSD_DELAYED_GC_TASK_QS].set_locks = ERTS_PSD_DELAYED_GC_TASK_QS_SET_LOCKS; +#ifdef ERTS_DIRTY_SCHEDULERS + erts_psd_required_locks[ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT].get_locks + = ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT_GET_LOCKS; + erts_psd_required_locks[ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT].set_locks + = ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT_SET_LOCKS; +#endif + /* Check that we have locks for all entries */ for (ix = 0; ix < ERTS_PSD_SIZE; ix++) { ERTS_SMP_LC_ASSERT(erts_psd_required_locks[ix].get_locks); @@ -931,7 +963,9 @@ reply_sched_wall_time(void *vswtrp) ErlHeapFragment *bp = NULL; ASSERT(esdp); - +#ifdef ERTS_DIRTY_SCHEDULERS + ASSERT(!ERTS_SCHEDULER_IS_DIRTY(esdp)); +#endif if (swtrp->set) { if (!swtrp->enable && esdp->sched_wall_time.enabled) { esdp->sched_wall_time.need = erts_sched_balance_util; @@ -1015,6 +1049,9 @@ erts_sched_wall_time_request(Process *c_p, int set, int enable) if (!set && !esdp->sched_wall_time.enabled) return THE_NON_VALUE; +#ifdef ERTS_DIRTY_SCHEDULERS + ASSERT(!ERTS_SCHEDULER_IS_DIRTY(esdp)); +#endif swtrp = swtreq_alloc(); ref = erts_make_ref(c_p); @@ -1492,6 +1529,9 @@ erts_schedule_multi_misc_aux_work(int ignore_self, if (ignore_self) { ErtsSchedulerData *esdp = erts_get_scheduler_data(); +#ifdef ERTS_DIRTY_SCHEDULERS + ASSERT(!ERTS_SCHEDULER_IS_DIRTY(esdp)); +#endif if (esdp) self = (int) esdp->no; } @@ -1601,7 +1641,7 @@ void erts_alloc_notify_delayed_dealloc(int ix) { ErtsSchedulerData *esdp = erts_get_scheduler_data(); - if (esdp) + if (esdp && !ERTS_SCHEDULER_IS_DIRTY(esdp)) schedule_aux_work_wakeup(&esdp->aux_work_data, ix, ERTS_SSI_AUX_WORK_DD); @@ -1615,6 +1655,10 @@ erts_alloc_ensure_handle_delayed_dealloc_call(int ix) { #ifdef DEBUG ErtsSchedulerData *esdp = erts_get_scheduler_data(); +#ifdef ERTS_DIRTY_SCHEDULERS + if (esdp && ERTS_SCHEDULER_IS_DIRTY(esdp)) + return; +#endif ASSERT(!esdp || ix == (int) esdp->no); #endif set_aux_work_flags_wakeup_nob(ERTS_SCHED_SLEEP_INFO_IX(ix-1), @@ -2220,6 +2264,9 @@ static ERTS_INLINE void sched_active_sys(Uint no, ErtsRunQueue *rq) { ERTS_SMP_LC_ASSERT(erts_smp_lc_runq_is_locked(rq)); +#ifdef ERTS_DIRTY_SCHEDULERS + ASSERT(!ERTS_RUNQ_IX_IS_DIRTY(rq->ix)); +#endif ASSERT(rq->waiting < 0); rq->waiting *= -1; rq->waiting--; @@ -2276,6 +2323,9 @@ static ERTS_INLINE void sched_change_waiting_sys_to_waiting(Uint no, ErtsRunQueue *rq) { ERTS_SMP_LC_ASSERT(erts_smp_lc_runq_is_locked(rq)); +#ifdef ERTS_DIRTY_SCHEDULERS + ASSERT(!ERTS_RUNQ_IX_IS_DIRTY(rq->ix)); +#endif ASSERT(rq->waiting < 0); rq->waiting *= -1; } @@ -2291,7 +2341,7 @@ sched_waiting(Uint no, ErtsRunQueue *rq) else rq->waiting++; rq->woken = 0; - if (erts_system_profile_flags.scheduler) + if (!ERTS_RUNQ_IX_IS_DIRTY(rq->ix) && erts_system_profile_flags.scheduler) profile_scheduler(make_small(no), am_inactive); } @@ -2303,7 +2353,7 @@ sched_active(Uint no, ErtsRunQueue *rq) rq->waiting++; else rq->waiting--; - if (erts_system_profile_flags.scheduler) + if (!ERTS_RUNQ_IX_IS_DIRTY(rq->ix) && erts_system_profile_flags.scheduler) profile_scheduler(make_small(no), am_active); } @@ -2317,7 +2367,7 @@ ongoing_multi_scheduling_block(void) static ERTS_INLINE void empty_runq_aux(ErtsRunQueue *rq, Uint32 old_flags) { - if (old_flags & ERTS_RUNQ_FLG_NONEMPTY) { + if (!ERTS_RUNQ_IX_IS_DIRTY(rq->ix) && old_flags & ERTS_RUNQ_FLG_NONEMPTY) { #ifdef DEBUG erts_aint32_t empty = erts_smp_atomic32_read_nob(&no_empty_run_queues); /* @@ -2357,7 +2407,7 @@ static ERTS_INLINE void non_empty_runq(ErtsRunQueue *rq) { Uint32 old_flags = ERTS_RUNQ_FLGS_SET(rq, ERTS_RUNQ_FLG_NONEMPTY); - if (!(old_flags & ERTS_RUNQ_FLG_NONEMPTY)) { + if (!ERTS_RUNQ_IX_IS_DIRTY(rq->ix) && (!(old_flags & ERTS_RUNQ_FLG_NONEMPTY))) { #ifdef DEBUG erts_aint32_t empty = erts_smp_atomic32_read_nob(&no_empty_run_queues); /* @@ -2602,18 +2652,37 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) ERTS_SMP_LC_ASSERT(erts_smp_lc_runq_is_locked(rq)); +#ifdef ERTS_DIRTY_SCHEDULERS + if (ERTS_RUNQ_IX_IS_DIRTY(rq->ix)) + erts_smp_spin_lock(&rq->sleepers.lock); +#endif flgs = sched_prep_spin_wait(ssi); if (flgs & ERTS_SSI_FLG_SUSPENDED) { /* Go suspend instead... */ +#ifdef ERTS_DIRTY_SCHEDULERS + if (ERTS_RUNQ_IX_IS_DIRTY(rq->ix)) + erts_smp_spin_unlock(&rq->sleepers.lock); +#endif return; } +#ifdef ERTS_DIRTY_SCHEDULERS + if (ERTS_RUNQ_IX_IS_DIRTY(rq->ix)) { + ssi->prev = NULL; + ssi->next = rq->sleepers.list; + if (rq->sleepers.list) + rq->sleepers.list->prev = ssi; + rq->sleepers.list = ssi; + erts_smp_spin_unlock(&rq->sleepers.lock); + } +#endif + /* * If all schedulers are waiting, one of them *should* * be waiting in erl_sys_schedule() */ - if (!prepare_for_sys_schedule()) { + if (ERTS_SCHEDULER_IS_DIRTY(esdp) || !prepare_for_sys_schedule()) { sched_waiting(esdp->no, rq); @@ -2623,12 +2692,13 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) tse_wait: - if (thr_prgr_active != working) + if (!ERTS_SCHEDULER_IS_DIRTY(esdp) && thr_prgr_active != working) sched_wall_time_change(esdp, thr_prgr_active); while (1) { - aux_work = erts_atomic32_read_acqb(&ssi->aux_work); + aux_work = ERTS_SCHEDULER_IS_DIRTY(esdp) ? 0 : + erts_atomic32_read_acqb(&ssi->aux_work); if (aux_work) { if (!thr_prgr_active) { erts_thr_progress_active(esdp, thr_prgr_active = 1); @@ -2642,11 +2712,13 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) if (aux_work) flgs = erts_smp_atomic32_read_acqb(&ssi->flags); else { - if (thr_prgr_active) { - erts_thr_progress_active(esdp, thr_prgr_active = 0); - sched_wall_time_change(esdp, 0); + if (!ERTS_SCHEDULER_IS_DIRTY(esdp)) { + if (thr_prgr_active) { + erts_thr_progress_active(esdp, thr_prgr_active = 0); + sched_wall_time_change(esdp, 0); + } + erts_thr_progress_prepare_wait(esdp); } - erts_thr_progress_prepare_wait(esdp); flgs = sched_spin_wait(ssi, spincount); if (flgs & ERTS_SSI_FLG_SLEEPING) { @@ -2661,7 +2733,8 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) } while (res == EINTR); } } - erts_thr_progress_finalize_wait(esdp); + if (!ERTS_SCHEDULER_IS_DIRTY(esdp)) + erts_thr_progress_finalize_wait(esdp); } if (!(flgs & ERTS_SSI_FLG_WAITING)) { @@ -2682,7 +2755,7 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) if (flgs & ~ERTS_SSI_FLG_SUSPENDED) erts_smp_atomic32_read_band_nob(&ssi->flags, ERTS_SSI_FLG_SUSPENDED); - if (!thr_prgr_active) { + if (!ERTS_SCHEDULER_IS_DIRTY(esdp) && !thr_prgr_active) { erts_thr_progress_active(esdp, thr_prgr_active = 1); sched_wall_time_change(esdp, 1); } @@ -2699,6 +2772,9 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) erts_smp_atomic32_set_relb(&function_calls, 0); *fcalls = 0; +#ifdef ERTS_DIRTY_SCHEDULERS + ASSERT(!ERTS_SCHEDULER_IS_DIRTY(esdp)); +#endif sched_waiting_sys(esdp->no, rq); @@ -2912,6 +2988,34 @@ wake_scheduler(ErtsRunQueue *rq) erts_sched_finish_poke(ssi, flgs); } +#ifdef ERTS_DIRTY_SCHEDULERS +static void +wake_dirty_scheduler(ErtsRunQueue *rq) +{ + ErtsSchedulerSleepInfo *ssi; + ErtsSchedulerSleepList *sl; + + ASSERT(ERTS_RUNQ_IX_IS_DIRTY(rq->ix)); + + sl = &rq->sleepers; + erts_smp_spin_lock(&sl->lock); + ssi = sl->list; + if (!ssi) + erts_smp_spin_unlock(&sl->lock); + else { + sl->list = NULL; + erts_smp_spin_unlock(&sl->lock); + + ERTS_THR_MEMORY_BARRIER; + do { + ErtsSchedulerSleepInfo *wake_ssi = ssi; + ssi = ssi->next; + erts_sched_finish_poke(wake_ssi, ssi_flags_set_wake(wake_ssi)); + } while (ssi); + } +} +#endif + #define ERTS_NO_USED_RUNQS_SHIFT 16 #define ERTS_NO_RUNQS_MASK 0xffff @@ -3047,8 +3151,14 @@ static ERTS_INLINE void smp_notify_inc_runq(ErtsRunQueue *runq) { #ifdef ERTS_SMP - if (runq) - wake_scheduler(runq); + if (runq) { +#ifdef ERTS_DIRTY_SCHEDULERS + if (ERTS_RUNQ_IX_IS_DIRTY(runq->ix)) + wake_dirty_scheduler(runq); + else +#endif + wake_scheduler(runq); + } #endif } @@ -4899,7 +5009,10 @@ erts_sched_set_wake_cleanup_threshold(char *str) static void init_aux_work_data(ErtsAuxWorkData *awdp, ErtsSchedulerData *esdp, char *dawwp) { - awdp->sched_id = esdp ? (int) esdp->no : 0; + if (!esdp || ERTS_SCHEDULER_IS_DIRTY(esdp)) + awdp->sched_id = 0; + else + awdp->sched_id = (int) esdp->no; awdp->esdp = esdp; awdp->ssi = esdp ? esdp->ssi : NULL; #ifdef ERTS_SMP @@ -4939,14 +5052,71 @@ init_aux_work_data(ErtsAuxWorkData *awdp, ErtsSchedulerData *esdp, char *dawwp) #endif } +static void +init_scheduler_data(ErtsSchedulerData* esdp, int num, + ErtsSchedulerSleepInfo* ssi, + ErtsRunQueue* runq, + char** daww_ptr, size_t daww_sz) +{ +#ifdef ERTS_SMP + erts_bits_init_state(&esdp->erl_bits_state); + esdp->match_pseudo_process = NULL; + esdp->free_process = NULL; +#endif + esdp->x_reg_array = + erts_alloc_permanent_cache_aligned(ERTS_ALC_T_BEAM_REGISTER, + ERTS_X_REGS_ALLOCATED * + sizeof(Eterm)); + esdp->f_reg_array = + erts_alloc_permanent_cache_aligned(ERTS_ALC_T_BEAM_REGISTER, + MAX_REG * sizeof(FloatDef)); +#if !HEAP_ON_C_STACK + esdp->num_tmp_heap_used = 0; +#endif +#ifdef ERTS_DIRTY_SCHEDULERS + if (ERTS_RUNQ_IX_IS_DIRTY(runq->ix)) { + esdp->no = 0; + esdp->dirty_no = (Uint) num; + } + else { + esdp->no = (Uint) num; + esdp->dirty_no = 0; + } +#else + esdp->no = (Uint) num; +#endif + esdp->ssi = ssi; + esdp->current_process = NULL; + esdp->current_port = NULL; + + esdp->virtual_reds = 0; + esdp->cpu_id = -1; + + erts_init_atom_cache_map(&esdp->atom_cache_map); + + esdp->run_queue = runq; + esdp->run_queue->scheduler = esdp; + + if (daww_ptr) { + init_aux_work_data(&esdp->aux_work_data, esdp, *daww_ptr); +#ifdef ERTS_SMP + *daww_ptr += daww_sz; +#endif + } + + esdp->reductions = 0; + + init_sched_wall_time(&esdp->sched_wall_time); + erts_port_task_handle_init(&esdp->nosuspend_port_task_handle); +} + void erts_init_scheduling(int no_schedulers, int no_schedulers_online) { int ix, n, no_ssi; char *daww_ptr; -#ifdef ERTS_SMP size_t daww_sz; -#endif + size_t size_runqs; init_misc_op_list_alloc(); init_proc_sys_task_queues_alloc(); @@ -4967,19 +5137,26 @@ erts_init_scheduling(int no_schedulers, int no_schedulers_online) /* Create and initialize run queues */ n = no_schedulers; - - erts_aligned_run_queues = - erts_alloc_permanent_cache_aligned(ERTS_ALC_T_RUNQS, - sizeof(ErtsAlignedRunQueue) * n); + size_runqs = sizeof(ErtsAlignedRunQueue) * (n + ERTS_NUM_DIRTY_RUNQS); + erts_aligned_run_queues = + erts_alloc_permanent_cache_aligned(ERTS_ALC_T_RUNQS, size_runqs); #ifdef ERTS_SMP +#ifdef ERTS_DIRTY_SCHEDULERS + erts_aligned_run_queues += ERTS_NUM_DIRTY_RUNQS; +#endif erts_smp_atomic32_init_nob(&no_empty_run_queues, 0); #endif erts_no_run_queues = n; - for (ix = 0; ix < n; ix++) { + for (ix = -(ERTS_NUM_DIRTY_RUNQS); ix < n; ix++) { int pix, rix; +#ifdef ERTS_DIRTY_SCHEDULERS + ErtsRunQueue *rq = ERTS_RUNQ_IX_IS_DIRTY(ix) ? + ERTS_DIRTY_RUNQ_IX(ix) : ERTS_RUNQ_IX(ix); +#else ErtsRunQueue *rq = ERTS_RUNQ_IX(ix); +#endif rq->ix = ix; @@ -4990,6 +5167,15 @@ erts_init_scheduling(int no_schedulers, int no_schedulers_online) erts_smp_mtx_init_x(&rq->mtx, "run_queue", make_small(ix + 1)); erts_smp_cnd_init(&rq->cnd); +#ifdef ERTS_DIRTY_SCHEDULERS +#ifdef ERTS_SMP + if (ERTS_RUNQ_IX_IS_DIRTY(ix)) { + erts_smp_spinlock_init(&rq->sleepers.lock, "dirty_run_queue_sleep_list"); + rq->sleepers.list = NULL; + } +#endif +#endif + rq->waiting = 0; rq->woken = 0; ERTS_RUNQ_FLGS_INIT(rq, ERTS_RUNQ_FLG_NONEMPTY); @@ -5076,6 +5262,29 @@ erts_init_scheduling(int no_schedulers, int no_schedulers_online) #ifdef ERTS_SMP aligned_sched_sleep_info++; + +#ifdef ERTS_DIRTY_SCHEDULERS + aligned_dirty_cpu_sched_sleep_info = + erts_alloc_permanent_cache_aligned( + ERTS_ALC_T_SCHDLR_SLP_INFO, + erts_no_dirty_cpu_schedulers*sizeof(ErtsAlignedSchedulerSleepInfo)); + for (ix = 0; ix < erts_no_dirty_cpu_schedulers; ix++) { + ErtsSchedulerSleepInfo *ssi = &aligned_dirty_cpu_sched_sleep_info[ix].ssi; + erts_smp_atomic32_init_nob(&ssi->flags, 0); + ssi->event = NULL; /* initialized in sched_thread_func */ + erts_atomic32_init_nob(&ssi->aux_work, 0); + } + aligned_dirty_io_sched_sleep_info = + erts_alloc_permanent_cache_aligned( + ERTS_ALC_T_SCHDLR_SLP_INFO, + erts_no_dirty_io_schedulers*sizeof(ErtsAlignedSchedulerSleepInfo)); + for (ix = 0; ix < erts_no_dirty_io_schedulers; ix++) { + ErtsSchedulerSleepInfo *ssi = &aligned_dirty_io_sched_sleep_info[ix].ssi; + erts_smp_atomic32_init_nob(&ssi->flags, 0); + ssi->event = NULL; /* initialized in sched_thread_func */ + erts_atomic32_init_nob(&ssi->aux_work, 0); + } +#endif #endif /* Create and initialize scheduler specific data */ @@ -5086,6 +5295,7 @@ erts_init_scheduling(int no_schedulers, int no_schedulers_online) daww_ptr = erts_alloc_permanent_cache_aligned(ERTS_ALC_T_SCHDLR_DATA, daww_sz*n); #else + daww_sz = 0; daww_ptr = NULL; #endif @@ -5095,46 +5305,32 @@ erts_init_scheduling(int no_schedulers, int no_schedulers_online) for (ix = 0; ix < n; ix++) { ErtsSchedulerData *esdp = ERTS_SCHEDULER_IX(ix); -#ifdef ERTS_SMP - erts_bits_init_state(&esdp->erl_bits_state); - esdp->match_pseudo_process = NULL; - esdp->free_process = NULL; -#endif - esdp->x_reg_array = - erts_alloc_permanent_cache_aligned(ERTS_ALC_T_BEAM_REGISTER, - ERTS_X_REGS_ALLOCATED * - sizeof(Eterm)); - esdp->f_reg_array = - erts_alloc_permanent_cache_aligned(ERTS_ALC_T_BEAM_REGISTER, - MAX_REG * sizeof(FloatDef)); -#if !HEAP_ON_C_STACK - esdp->num_tmp_heap_used = 0; -#endif - esdp->no = (Uint) ix+1; - esdp->ssi = ERTS_SCHED_SLEEP_INFO_IX(ix); - esdp->current_process = NULL; - esdp->current_port = NULL; - - esdp->virtual_reds = 0; - esdp->cpu_id = -1; - - erts_init_atom_cache_map(&esdp->atom_cache_map); - - esdp->run_queue = ERTS_RUNQ_IX(ix); - esdp->run_queue->scheduler = esdp; + init_scheduler_data(esdp, ix+1, ERTS_SCHED_SLEEP_INFO_IX(ix), + ERTS_RUNQ_IX(ix), &daww_ptr, daww_sz); + } - init_aux_work_data(&esdp->aux_work_data, esdp, daww_ptr); +#ifdef ERTS_DIRTY_SCHEDULERS #ifdef ERTS_SMP - daww_ptr += daww_sz; -#endif - - esdp->reductions = 0; - - init_sched_wall_time(&esdp->sched_wall_time); - - erts_port_task_handle_init(&esdp->nosuspend_port_task_handle); - + erts_aligned_dirty_cpu_scheduler_data = + erts_alloc_permanent_cache_aligned( + ERTS_ALC_T_SCHDLR_DATA, + erts_no_dirty_cpu_schedulers*sizeof(ErtsAlignedSchedulerData)); + for (ix = 0; ix < erts_no_dirty_cpu_schedulers; ix++) { + ErtsSchedulerData *esdp = ERTS_DIRTY_CPU_SCHEDULER_IX(ix); + init_scheduler_data(esdp, ix+1, ERTS_DIRTY_CPU_SCHED_SLEEP_INFO_IX(ix), + ERTS_DIRTY_CPU_RUNQ, NULL, 0); + } + erts_aligned_dirty_io_scheduler_data = + erts_alloc_permanent_cache_aligned( + ERTS_ALC_T_SCHDLR_DATA, + erts_no_dirty_io_schedulers*sizeof(ErtsAlignedSchedulerData)); + for (ix = 0; ix < erts_no_dirty_io_schedulers; ix++) { + ErtsSchedulerData *esdp = ERTS_DIRTY_IO_SCHEDULER_IX(ix); + init_scheduler_data(esdp, ix+1, ERTS_DIRTY_IO_SCHED_SLEEP_INFO_IX(ix), + ERTS_DIRTY_IO_RUNQ, NULL, 0); } +#endif +#endif init_misc_aux_work(); #if !HALFWORD_HEAP @@ -5198,6 +5394,10 @@ erts_init_scheduling(int no_schedulers, int no_schedulers_online) #endif } erts_no_schedulers = 1; +#ifdef ERTS_DIRTY_SCHEDULERS + erts_no_dirty_cpu_schedulers = 0; + erts_no_dirty_io_schedulers = 0; +#endif #endif erts_smp_atomic32_init_nob(&function_calls, 0); @@ -5310,28 +5510,39 @@ check_enqueue_in_prio_queue(erts_aint32_t *prq_prio_p, *prq_prio_p = aprio; - max_qbit = (actual >> ERTS_PSFLGS_IN_PRQ_MASK_OFFSET) & ERTS_PSFLGS_QMASK; - max_qbit |= 1 << ERTS_PSFLGS_QMASK_BITS; - max_qbit &= -max_qbit; - /* - * max_qbit now either contain bit set for highest prio queue or a bit - * out of range (which will have a value larger than valid range). - */ +#ifdef ERTS_DIRTY_SCHEDULERS + if (!(actual & (ERTS_PSFLG_DIRTY_CPU_PROC|ERTS_PSFLG_DIRTY_IO_PROC))) { +#endif + max_qbit = (actual >> ERTS_PSFLGS_IN_PRQ_MASK_OFFSET) & ERTS_PSFLGS_QMASK; + max_qbit |= 1 << ERTS_PSFLGS_QMASK_BITS; + max_qbit &= -max_qbit; + /* + * max_qbit now either contain bit set for highest prio queue or a bit + * out of range (which will have a value larger than valid range). + */ - if (qbit >= max_qbit) - return 0; /* Already queued in higher or equal prio */ + if (qbit >= max_qbit) + return 0; /* Already queued in higher or equal prio */ - /* Need to enqueue (if already enqueued, it is in lower prio) */ - *newp |= qbit << ERTS_PSFLGS_IN_PRQ_MASK_OFFSET; + /* Need to enqueue (if already enqueued, it is in lower prio) */ + *newp |= qbit << ERTS_PSFLGS_IN_PRQ_MASK_OFFSET; - if ((actual & (ERTS_PSFLG_IN_RUNQ|ERTS_PSFLGS_USR_PRIO_MASK)) - != (aprio << ERTS_PSFLGS_USR_PRIO_OFFSET)) { - /* - * Process struct already enqueued, or actual prio not - * equal to user prio, i.e., enqueue using proxy. - */ - return -1; + if ((actual & (ERTS_PSFLG_IN_RUNQ|ERTS_PSFLGS_USR_PRIO_MASK)) + != (aprio << ERTS_PSFLGS_USR_PRIO_OFFSET)) { + /* + * Process struct already enqueued, or actual prio not + * equal to user prio, i.e., enqueue using proxy. + */ + return -1; + } +#ifdef ERTS_DIRTY_SCHEDULERS + } else { + if (actual & ERTS_PSFLG_DIRTY_CPU_PROC) + *newp |= ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q; + else + *newp |= ERTS_PSFLG_DIRTY_IO_PROC_IN_Q; } +#endif /* * Enqueue using process struct. @@ -5387,10 +5598,21 @@ schedule_out_process(ErtsRunQueue *c_rq, erts_aint32_t state, Process *p, Proces } else { Process *sched_p; - ErtsRunQueue *runq = erts_get_runq_proc(p); + ErtsRunQueue *runq; ASSERT(!(n & ERTS_PSFLG_SUSPENDED) || (n & ERTS_PSFLG_ACTIVE_SYS)); +#ifdef ERTS_DIRTY_SCHEDULERS +#ifdef ERTS_SMP + if (ERTS_PSFLG_DIRTY_CPU_PROC & a) + runq = ERTS_DIRTY_CPU_RUNQ; + else if (ERTS_PSFLG_DIRTY_IO_PROC & a) + runq = ERTS_DIRTY_IO_RUNQ; + else +#endif +#endif + runq = erts_get_runq_proc(p); + if (enqueue < 0) sched_p = make_proxy_proc(proxy, p, enq_prio); else { @@ -5400,7 +5622,11 @@ schedule_out_process(ErtsRunQueue *c_rq, erts_aint32_t state, Process *p, Proces } #ifdef ERTS_SMP - if (!(ERTS_PSFLG_BOUND & n)) { + if (!(ERTS_PSFLG_BOUND & n) +#ifdef ERTS_DIRTY_SCHEDULERS + && !(n & (ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q|ERTS_PSFLG_DIRTY_IO_PROC_IN_Q)) +#endif + ) { ErtsRunQueue *new_runq = erts_check_emigration_need(runq, enq_prio); if (new_runq) { RUNQ_SET_RQ(&sched_p->run_queue, new_runq); @@ -5827,6 +6053,9 @@ suspend_scheduler(ErtsSchedulerData *esdp) * Regardless of why a scheduler is suspended, it ends up here. */ +#ifdef ERTS_DIRTY_SCHEDULERS + ASSERT(!ERTS_SCHEDULER_IS_DIRTY(esdp)); +#endif ASSERT(no != 1); evacuate_run_queue(esdp->run_queue, &sbp); @@ -5994,22 +6223,44 @@ ErtsSchedSuspendResult erts_schedulers_state(Uint *total, Uint *online, Uint *active, + Uint *dirty_cpu, + Uint *dirty_cpu_online, + Uint *dirty_io, int yield_allowed) { - int res; + int res = ERTS_SCHDLR_SSPND_EINVAL; erts_aint32_t changing; - erts_smp_mtx_lock(&schdlr_sspnd.mtx); - changing = erts_smp_atomic32_read_nob(&schdlr_sspnd.changing); - if (yield_allowed && (changing & ~ERTS_SCHDLR_SSPND_CHNG_WAITER)) - res = ERTS_SCHDLR_SSPND_YIELD_RESTART; - else { - *active = *online = schdlr_sspnd.online; - if (ongoing_multi_scheduling_block()) - *active = 1; - res = ERTS_SCHDLR_SSPND_DONE; + if (total) { + ASSERT(online); + ASSERT(active); + erts_smp_mtx_lock(&schdlr_sspnd.mtx); + changing = erts_smp_atomic32_read_nob(&schdlr_sspnd.changing); + if (yield_allowed && (changing & ~ERTS_SCHDLR_SSPND_CHNG_WAITER)) + res = ERTS_SCHDLR_SSPND_YIELD_RESTART; + else { + *active = *online = schdlr_sspnd.online; + if (ongoing_multi_scheduling_block()) + *active = 1; + res = ERTS_SCHDLR_SSPND_DONE; + } + erts_smp_mtx_unlock(&schdlr_sspnd.mtx); + *total = erts_no_schedulers; } - erts_smp_mtx_unlock(&schdlr_sspnd.mtx); - *total = erts_no_schedulers; +#ifdef ERTS_DIRTY_SCHEDULERS + if (dirty_cpu) + *dirty_cpu = erts_no_dirty_cpu_schedulers; + if (dirty_cpu_online) + *dirty_cpu_online = erts_no_dirty_cpu_schedulers_online; + if (dirty_io) + *dirty_io = erts_no_dirty_io_schedulers; +#else + if (dirty_cpu) + *dirty_cpu = 0; + if (dirty_cpu_online) + *dirty_cpu_online = 0; + if (dirty_io) + *dirty_io = 0; +#endif return res; } @@ -6101,6 +6352,10 @@ erts_set_schedulers_online(Process *p, ErtsRunQueue *rq = ERTS_RUNQ_IX(ix); wake_scheduler(rq); } +#ifdef ERTS_DIRTY_SCHEDULERS + wake_dirty_scheduler(ERTS_DIRTY_CPU_RUNQ); + wake_dirty_scheduler(ERTS_DIRTY_IO_RUNQ); +#endif } } @@ -6448,6 +6703,92 @@ sched_thread_func(void *vesdp) return NULL; } +#ifdef ERTS_DIRTY_SCHEDULERS +#ifdef ERTS_SMP +static void* +sched_dirty_cpu_thread_func(void *vesdp) +{ + ErtsThrPrgrCallbacks callbacks; + ErtsSchedulerData *esdp = vesdp; + Uint no = esdp->dirty_no; + ERTS_DIRTY_CPU_SCHED_SLEEP_INFO_IX(no-1)->event = erts_tse_fetch(); + callbacks.arg = (void *) esdp->ssi; + callbacks.wakeup = thr_prgr_wakeup; + callbacks.prepare_wait = NULL; + callbacks.wait = NULL; + callbacks.finalize_wait = NULL; + + erts_thr_progress_register_unmanaged_thread(&callbacks); +#ifdef ERTS_ENABLE_LOCK_CHECK + { + char buf[31]; + erts_snprintf(&buf[0], 31, "dirty cpu scheduler %beu", no); + erts_lc_set_thread_name(&buf[0]); + } +#endif + erts_tsd_set(sched_data_key, vesdp); +#if ERTS_USE_ASYNC_READY_Q + esdp->aux_work_data.async_ready.queue = NULL; +#endif + + erts_proc_lock_prepare_proc_lock_waiter(); + +#ifdef HIPE + hipe_thread_signal_init(); +#endif + erts_thread_init_float(); + + process_main(); + /* No schedulers should *ever* terminate */ + erl_exit(ERTS_ABORT_EXIT, + "Dirty CPU scheduler thread number %beu terminated\n", + no); + return NULL; +} + +static void* +sched_dirty_io_thread_func(void *vesdp) +{ + ErtsThrPrgrCallbacks callbacks; + ErtsSchedulerData *esdp = vesdp; + Uint no = esdp->dirty_no; + ERTS_DIRTY_IO_SCHED_SLEEP_INFO_IX(no-1)->event = erts_tse_fetch(); + callbacks.arg = (void *) esdp->ssi; + callbacks.wakeup = thr_prgr_wakeup; + callbacks.prepare_wait = NULL; + callbacks.wait = NULL; + callbacks.finalize_wait = NULL; + + erts_thr_progress_register_unmanaged_thread(&callbacks); +#ifdef ERTS_ENABLE_LOCK_CHECK + { + char buf[31]; + erts_snprintf(&buf[0], 31, "dirty io scheduler %beu", no); + erts_lc_set_thread_name(&buf[0]); + } +#endif + erts_tsd_set(sched_data_key, vesdp); +#if ERTS_USE_ASYNC_READY_Q + esdp->aux_work_data.async_ready.queue = NULL; +#endif + + erts_proc_lock_prepare_proc_lock_waiter(); + +#ifdef HIPE + hipe_thread_signal_init(); +#endif + erts_thread_init_float(); + + process_main(); + /* No schedulers should *ever* terminate */ + erl_exit(ERTS_ABORT_EXIT, + "Dirty I/O scheduler thread number %beu terminated\n", + no); + return NULL; +} +#endif +#endif + static ethr_tid aux_tid; void @@ -6498,6 +6839,26 @@ erts_start_schedulers(void) erts_no_schedulers = actual; +#ifdef ERTS_DIRTY_SCHEDULERS +#ifdef ERTS_SMP + { + int ix; + for (ix = 0; ix < erts_no_dirty_cpu_schedulers; ix++) { + ErtsSchedulerData *esdp = ERTS_DIRTY_CPU_SCHEDULER_IX(ix); + res = ethr_thr_create(&esdp->tid,sched_dirty_cpu_thread_func,(void*)esdp,&opts); + if (res != 0) + erl_exit(1, "Failed to create dirty cpu scheduler thread %d\n", ix); + } + for (ix = 0; ix < erts_no_dirty_io_schedulers; ix++) { + ErtsSchedulerData *esdp = ERTS_DIRTY_IO_SCHEDULER_IX(ix); + res = ethr_thr_create(&esdp->tid,sched_dirty_io_thread_func,(void*)esdp,&opts); + if (res != 0) + erl_exit(1, "Failed to create dirty io scheduler thread %d\n", ix); + } + } +#endif +#endif + ERTS_THR_MEMORY_BARRIER; res = ethr_thr_create(&aux_tid, aux_thread, NULL, &opts); @@ -7486,7 +7847,8 @@ Process *schedule(Process *p, int calls) input_reductions = INPUT_REDUCTIONS; } - ERTS_SMP_LC_ASSERT(!erts_thr_progress_is_blocking()); + ERTS_SMP_LC_ASSERT(ERTS_SCHEDULER_IS_DIRTY(erts_get_scheduler_data()) + || !erts_thr_progress_is_blocking()); /* * Clean up after the process being scheduled out. @@ -7597,42 +7959,40 @@ Process *schedule(Process *p, int calls) } - ERTS_SMP_LC_ASSERT(!erts_thr_progress_is_blocking()); + ERTS_SMP_LC_ASSERT(ERTS_SCHEDULER_IS_DIRTY(esdp) + || !erts_thr_progress_is_blocking()); check_activities_to_run: { #ifdef ERTS_SMP ErtsMigrationPaths *mps; ErtsMigrationPath *mp; - -#ifdef ERTS_SMP - { - ErtsProcList *pnd_xtrs = rq->procs.pending_exiters; - if (erts_proclist_fetch(&pnd_xtrs, NULL)) { - rq->procs.pending_exiters = NULL; - erts_smp_runq_unlock(rq); - handle_pending_exiters(pnd_xtrs); - erts_smp_runq_lock(rq); - } - + ErtsProcList *pnd_xtrs = rq->procs.pending_exiters; + if (erts_proclist_fetch(&pnd_xtrs, NULL)) { + rq->procs.pending_exiters = NULL; + erts_smp_runq_unlock(rq); + handle_pending_exiters(pnd_xtrs); + erts_smp_runq_lock(rq); } -#endif - if (rq->check_balance_reds <= 0) - check_balance(rq); + if (!ERTS_SCHEDULER_IS_DIRTY(esdp)) { + if (rq->check_balance_reds <= 0) + check_balance(rq); - ERTS_SMP_LC_ASSERT(!erts_thr_progress_is_blocking()); - ERTS_SMP_LC_ASSERT(erts_smp_lc_runq_is_locked(rq)); + ERTS_SMP_LC_ASSERT(!erts_thr_progress_is_blocking()); - mps = erts_get_migration_paths_managed(); - mp = &mps->mpath[rq->ix]; + mps = erts_get_migration_paths_managed(); + mp = &mps->mpath[rq->ix]; - if (mp->flags & ERTS_RUNQ_FLGS_IMMIGRATE_QMASK) - immigrate(rq, mp); + if (mp->flags & ERTS_RUNQ_FLGS_IMMIGRATE_QMASK) + immigrate(rq, mp); + } + ERTS_SMP_LC_ASSERT(erts_smp_lc_runq_is_locked(rq)); continue_check_activities_to_run: flags = ERTS_RUNQ_FLGS_GET_NOB(rq); continue_check_activities_to_run_known_flags: - ASSERT(flags & ERTS_RUNQ_FLG_NONEMPTY); + ASSERT(ERTS_SCHEDULER_IS_DIRTY(esdp) + || flags & ERTS_RUNQ_FLG_NONEMPTY); if (flags & (ERTS_RUNQ_FLG_CHK_CPU_BIND|ERTS_RUNQ_FLG_SUSPENDED)) { @@ -7647,7 +8007,7 @@ Process *schedule(Process *p, int calls) } } - { + if (!ERTS_SCHEDULER_IS_DIRTY(esdp)) { erts_aint32_t aux_work; int leader_update = erts_thr_progress_update(esdp); aux_work = erts_atomic32_read_acqb(&esdp->ssi->aux_work); @@ -7659,9 +8019,9 @@ Process *schedule(Process *p, int calls) handle_aux_work(&esdp->aux_work_data, aux_work, 0); erts_smp_runq_lock(rq); } - } - ERTS_SMP_LC_ASSERT(!erts_thr_progress_is_blocking()); + ERTS_SMP_LC_ASSERT(!erts_thr_progress_is_blocking()); + } ERTS_SMP_LC_ASSERT(erts_smp_lc_runq_is_locked(rq)); #else /* ERTS_SMP */ @@ -7690,7 +8050,7 @@ Process *schedule(Process *p, int calls) if (flags & ERTS_RUNQ_FLG_INACTIVE) empty_runq(rq); else { - if (try_steal_task(rq)) + if (!ERTS_RUNQ_IX_IS_DIRTY(rq->ix) && try_steal_task(rq)) goto continue_check_activities_to_run; empty_runq(rq); @@ -7702,6 +8062,7 @@ Process *schedule(Process *p, int calls) flags = ERTS_RUNQ_FLGS_GET_NOB(rq); if (flags & ERTS_RUNQ_FLG_SUSPENDED) { non_empty_runq(rq); + flags |= ERTS_RUNQ_FLG_NONEMPTY; goto continue_check_activities_to_run_known_flags; } } @@ -7715,7 +8076,8 @@ Process *schedule(Process *p, int calls) goto check_activities_to_run; } - else if (fcalls > input_reductions && prepare_for_sys_schedule()) { + else if (!ERTS_SCHEDULER_IS_DIRTY(esdp) && + (fcalls > input_reductions && prepare_for_sys_schedule())) { /* * Schedule system-level activities. */ @@ -7819,6 +8181,14 @@ Process *schedule(Process *p, int calls) psflg_band_mask = ~(((erts_aint32_t) 1) << (ERTS_PSFLGS_GET_PRQ_PRIO(state) + ERTS_PSFLGS_IN_PRQ_MASK_OFFSET)); +#ifdef ERTS_DIRTY_SCHEDULERS + /* if a non-dirty scheduler picks up a process marked as already being + in a dirty run queue, just drop it and go get another process */ + if (state & (ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q|ERTS_PSFLG_DIRTY_IO_PROC_IN_Q) && + !ERTS_SCHEDULER_IS_DIRTY(esdp)) + goto pick_next_process; +#endif + if (!(state & ERTS_PSFLG_PROXY)) psflg_band_mask &= ~ERTS_PSFLG_IN_RUNQ; else { @@ -7901,6 +8271,10 @@ Process *schedule(Process *p, int calls) (UWord) esdp->no); int migrated = old && old != esdp->no; +#ifdef ERTS_DIRTY_SCHEDULERS + ASSERT(!ERTS_SCHEDULER_IS_DIRTY(esdp)); +#endif + prio = (int) ERTS_PSFLGS_GET_USR_PRIO(state); erts_smp_spin_lock(&erts_sched_stat.lock); @@ -9711,8 +10085,12 @@ save_pending_exiter(Process *p) non_empty_runq(rq); erts_smp_runq_unlock(rq); - - wake_scheduler(rq); +#ifdef ERTS_DIRTY_SCHEDULERS + if (ERTS_RUNQ_IX_IS_DIRTY(rq->ix)) + wake_dirty_scheduler(rq); + else +#endif + wake_scheduler(rq); } #endif diff --git a/erts/emulator/beam/erl_process.h b/erts/emulator/beam/erl_process.h index 6155f99b85..dcb9251d0d 100644 --- a/erts/emulator/beam/erl_process.h +++ b/erts/emulator/beam/erl_process.h @@ -74,6 +74,10 @@ struct ErtsNodesMonitor_; #define ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT 0 #define ERTS_MAX_NO_OF_SCHEDULERS 1024 +#ifdef ERTS_DIRTY_SCHEDULERS +#define ERTS_MAX_NO_OF_DIRTY_CPU_SCHEDULERS ERTS_MAX_NO_OF_SCHEDULERS +#define ERTS_MAX_NO_OF_DIRTY_IO_SCHEDULERS ERTS_MAX_NO_OF_SCHEDULERS +#endif #define ERTS_DEFAULT_MAX_PROCESSES (1 << 18) @@ -103,6 +107,11 @@ extern Export exp_send, exp_receive, exp_timeout; extern int erts_sched_compact_load; extern int erts_sched_balance_util; extern Uint erts_no_schedulers; +#ifdef ERTS_DIRTY_SCHEDULERS +extern Uint erts_no_dirty_cpu_schedulers; +extern Uint erts_no_dirty_cpu_schedulers_online; +extern Uint erts_no_dirty_io_schedulers; +#endif extern Uint erts_no_run_queues; extern int erts_sched_thread_suggested_stack_size; #define ERTS_SCHED_THREAD_MIN_STACK_SIZE 4 /* Kilo words */ @@ -275,6 +284,13 @@ typedef enum { typedef struct ErtsSchedulerSleepInfo_ ErtsSchedulerSleepInfo; +#ifdef ERTS_DIRTY_SCHEDULERS +typedef struct { + erts_smp_spinlock_t lock; + ErtsSchedulerSleepInfo *list; +} ErtsSchedulerSleepList; +#endif + struct ErtsSchedulerSleepInfo_ { #ifdef ERTS_SMP ErtsSchedulerSleepInfo *next; @@ -387,6 +403,12 @@ struct ErtsRunQueue_ { erts_smp_mtx_t mtx; erts_smp_cnd_t cnd; +#ifdef ERTS_DIRTY_SCHEDULERS +#ifdef ERTS_SMP + ErtsSchedulerSleepList sleepers; +#endif +#endif + ErtsSchedulerData *scheduler; int waiting; /* < 0 in sys schedule; > 0 on cnd variable */ int woken; @@ -542,12 +564,14 @@ struct ErtsSchedulerData_ { Eterm tmp_heap[TMP_HEAP_SIZE]; int num_tmp_heap_used; Eterm beam_emu_tmp_heap[BEAM_EMU_TMP_HEAP_SIZE]; - Eterm cmp_tmp_heap[CMP_TMP_HEAP_SIZE]; Eterm erl_arith_tmp_heap[ERL_ARITH_TMP_HEAP_SIZE]; #endif ErtsSchedulerSleepInfo *ssi; Process *current_process; - Uint no; /* Scheduler number */ + Uint no; /* Scheduler number for normal schedulers */ +#ifdef ERTS_DIRTY_SCHEDULERS + Uint dirty_no; /* Scheduler number for dirty schedulers */ +#endif Port *current_port; ErtsRunQueue *run_queue; int virtual_reds; @@ -574,6 +598,10 @@ typedef union { } ErtsAlignedSchedulerData; extern ErtsAlignedSchedulerData *erts_aligned_scheduler_data; +#ifdef ERTS_DIRTY_SCHEDULERS +extern ErtsAlignedSchedulerData *erts_aligned_dirty_cpu_scheduler_data; +extern ErtsAlignedSchedulerData *erts_aligned_dirty_io_scheduler_data; +#endif #ifndef ERTS_SMP extern ErtsSchedulerData *erts_scheduler_data; @@ -685,8 +713,13 @@ erts_smp_reset_max_len(ErtsRunQueue *rq, ErtsRunQueueInfo *rqi) #define ERTS_PSD_DIST_ENTRY 3 #define ERTS_PSD_CALL_TIME_BP 4 #define ERTS_PSD_DELAYED_GC_TASK_QS 5 +#ifdef ERTS_DIRTY_SCHEDULERS +#define ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT 6 +#define ERTS_PSD_SIZE 7 +#else #define ERTS_PSD_SIZE 6 +#endif typedef struct { void *data[ERTS_PSD_SIZE]; @@ -713,6 +746,11 @@ typedef struct { #define ERTS_PSD_DELAYED_GC_TASK_QS_GET_LOCKS ERTS_PROC_LOCK_MAIN #define ERTS_PSD_DELAYED_GC_TASK_QS_SET_LOCKS ERTS_PROC_LOCK_MAIN +#ifdef ERTS_DIRTY_SCHEDULERS +#define ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT_GET_LOCKS ERTS_PROC_LOCK_MAIN +#define ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT_SET_LOCKS ERTS_PROC_LOCK_MAIN +#endif + typedef struct { ErtsProcLocks get_locks; ErtsProcLocks set_locks; @@ -1026,6 +1064,12 @@ void erts_check_for_holes(Process* p); #define ERTS_PSFLG_RUNNING_SYS ERTS_PSFLG_BIT(15) #define ERTS_PSFLG_PROXY ERTS_PSFLG_BIT(16) #define ERTS_PSFLG_DELAYED_SYS ERTS_PSFLG_BIT(17) +#ifdef ERTS_DIRTY_SCHEDULERS +#define ERTS_PSFLG_DIRTY_CPU_PROC ERTS_PSFLG_BIT(18) +#define ERTS_PSFLG_DIRTY_IO_PROC ERTS_PSFLG_BIT(19) +#define ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q ERTS_PSFLG_BIT(20) +#define ERTS_PSFLG_DIRTY_IO_PROC_IN_Q ERTS_PSFLG_BIT(21) +#endif #define ERTS_PSFLGS_IN_PRQ_MASK (ERTS_PSFLG_IN_PRQ_MAX \ | ERTS_PSFLG_IN_PRQ_HIGH \ @@ -1231,12 +1275,46 @@ extern struct erts_system_profile_flags_t erts_system_profile_flags; (p)->flags &= ~F_TIMO; \ } while (0) +#if defined(ERTS_DIRTY_SCHEDULERS) && defined(ERTS_SMP) +#define ERTS_NUM_DIRTY_RUNQS 2 +#else +#define ERTS_NUM_DIRTY_RUNQS 0 +#endif + #define ERTS_RUNQ_IX(IX) \ (ASSERT(0 <= (IX) && (IX) < erts_no_run_queues), \ &erts_aligned_run_queues[(IX)].runq) +#ifdef ERTS_DIRTY_SCHEDULERS +#define ERTS_RUNQ_IX_IS_DIRTY(IX) \ + (-(ERTS_NUM_DIRTY_RUNQS) <= (IX) && (IX) < 0) +#define ERTS_DIRTY_RUNQ_IX(IX) \ + (ASSERT(ERTS_RUNQ_IX_IS_DIRTY(IX)), \ + &erts_aligned_run_queues[(IX)].runq) +#define ERTS_DIRTY_CPU_RUNQ (&erts_aligned_run_queues[-1].runq) +#define ERTS_DIRTY_IO_RUNQ (&erts_aligned_run_queues[-2].runq) +#else +#define ERTS_RUNQ_IX_IS_DIRTY(IX) 0 +#endif #define ERTS_SCHEDULER_IX(IX) \ (ASSERT(0 <= (IX) && (IX) < erts_no_schedulers), \ &erts_aligned_scheduler_data[(IX)].esd) +#ifdef ERTS_DIRTY_SCHEDULERS +#define ERTS_DIRTY_CPU_SCHEDULER_IX(IX) \ + (ASSERT(0 <= (IX) && (IX) < erts_no_dirty_cpu_schedulers), \ + &erts_aligned_dirty_cpu_scheduler_data[(IX)].esd) +#define ERTS_DIRTY_IO_SCHEDULER_IX(IX) \ + (ASSERT(0 <= (IX) && (IX) < erts_no_dirty_io_schedulers), \ + &erts_aligned_dirty_io_scheduler_data[(IX)].esd) +#ifdef ERTS_SMP +#define ERTS_SCHEDULER_IS_DIRTY(ESDP) \ + ((ESDP)->dirty_no != 0) +#else +#define ERTS_SCHEDULER_IS_DIRTY(ESDP) 0 +#endif +#else +#define ERTS_RUNQ_IX_IS_DIRTY(IX) 0 +#define ERTS_SCHEDULER_IS_DIRTY(ESDP) 0 +#endif void erts_pre_init_process(void); void erts_late_init_process(void); @@ -1439,9 +1517,11 @@ int erts_dbg_check_halloc_lock(Process *p); void erts_dbg_multi_scheduling_return_trap(Process *, Eterm); #endif int erts_get_max_no_executing_schedulers(void); -#ifdef ERTS_SMP +#if defined(ERTS_SMP) || defined(ERTS_DIRTY_SCHEDULERS) ErtsSchedSuspendResult -erts_schedulers_state(Uint *, Uint *, Uint *, int); +erts_schedulers_state(Uint *, Uint *, Uint *, Uint *, Uint *, Uint *, int); +#endif +#ifdef ERTS_SMP ErtsSchedSuspendResult erts_set_schedulers_online(Process *p, ErtsProcLocks plocks, @@ -1559,7 +1639,7 @@ do { \ ErtsSchedulerData *esdp__ = ((P) \ ? ERTS_PROC_GET_SCHDATA((Process *) (P)) \ : erts_get_scheduler_data()); \ - if (esdp__) \ + if (esdp__ && !ERTS_SCHEDULER_IS_DIRTY(esdp__)) \ esdp__->verify_unused_temp_alloc( \ esdp__->verify_unused_temp_alloc_data); \ } while (0) @@ -1694,6 +1774,13 @@ erts_psd_set(Process *p, ErtsProcLocks plocks, int ix, void *data) #define ERTS_PROC_SET_DELAYED_GC_TASK_QS(P, L, PBT) \ ((ErtsProcSysTaskQs *) erts_psd_set((P), (L), ERTS_PSD_DELAYED_GC_TASK_QS, (void *) (PBT))) +#ifdef ERTS_DIRTY_SCHEDULERS +#define ERTS_PROC_GET_DIRTY_SCHED_TRAP_EXPORT(P) \ + ((Export *) erts_psd_get((P), ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT)) +#define ERTS_PROC_SET_DIRTY_SCHED_TRAP_EXPORT(P, L, DSTE) \ + ((Export *) erts_psd_set((P), (L), ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT, (void *) (DSTE))) +#endif + ERTS_GLB_INLINE Eterm erts_proc_get_error_handler(Process *p); ERTS_GLB_INLINE Eterm erts_proc_set_error_handler(Process *p, @@ -1887,7 +1974,12 @@ Uint erts_get_scheduler_id(void) { #ifdef ERTS_SMP ErtsSchedulerData *esdp = erts_get_scheduler_data(); - return esdp ? esdp->no : (Uint) 0; +#ifdef ERTS_DIRTY_SCHEDULERS + if (esdp && ERTS_SCHEDULER_IS_DIRTY(esdp)) + return 0; + else +#endif + return esdp ? esdp->no : (Uint) 0; #else return erts_get_scheduler_data() ? (Uint) 1 : (Uint) 0; #endif diff --git a/erts/emulator/beam/erl_term.c b/erts/emulator/beam/erl_term.c index 2f206ffbec..28cbe7004f 100644 --- a/erts/emulator/beam/erl_term.c +++ b/erts/emulator/beam/erl_term.c @@ -23,6 +23,7 @@ #include "sys.h" #include "erl_vm.h" #include "global.h" +#include "erl_map.h" #include <stdlib.h> #include <stdio.h> @@ -85,7 +86,10 @@ unsigned tag_val_def(Wterm x) case (_TAG_HEADER_EXTERNAL_PID >> _TAG_PRIMARY_SIZE): return EXTERNAL_PID_DEF; case (_TAG_HEADER_EXTERNAL_PORT >> _TAG_PRIMARY_SIZE): return EXTERNAL_PORT_DEF; case (_TAG_HEADER_EXTERNAL_REF >> _TAG_PRIMARY_SIZE): return EXTERNAL_REF_DEF; - default: return BINARY_DEF; + case (_TAG_HEADER_REFC_BIN >> _TAG_PRIMARY_SIZE): return BINARY_DEF; + case (_TAG_HEADER_HEAP_BIN >> _TAG_PRIMARY_SIZE): return BINARY_DEF; + case (_TAG_HEADER_SUB_BIN >> _TAG_PRIMARY_SIZE): return BINARY_DEF; + case (_TAG_HEADER_MAP >> _TAG_PRIMARY_SIZE): return MAP_DEF; } break; } diff --git a/erts/emulator/beam/erl_term.h b/erts/emulator/beam/erl_term.h index 953edf79ea..f10a3a9d38 100644 --- a/erts/emulator/beam/erl_term.h +++ b/erts/emulator/beam/erl_term.h @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 2000-2013. All Rights Reserved. + * Copyright Ericsson AB 2000-2014. All Rights Reserved. * * The contents of this file are subject to the Erlang Public License, * Version 1.1, (the "License"); you may not use this file except in @@ -135,11 +135,12 @@ struct erl_node_; /* Declared in erl_node_tables.h */ #define REF_SUBTAG (0x4 << _TAG_PRIMARY_SIZE) /* REF */ #define FUN_SUBTAG (0x5 << _TAG_PRIMARY_SIZE) /* FUN */ #define FLOAT_SUBTAG (0x6 << _TAG_PRIMARY_SIZE) /* FLOAT */ -#define EXPORT_SUBTAG (0x7 << _TAG_PRIMARY_SIZE) /* FLOAT */ +#define EXPORT_SUBTAG (0x7 << _TAG_PRIMARY_SIZE) /* FLOAT */ #define _BINARY_XXX_MASK (0x3 << _TAG_PRIMARY_SIZE) #define REFC_BINARY_SUBTAG (0x8 << _TAG_PRIMARY_SIZE) /* BINARY */ #define HEAP_BINARY_SUBTAG (0x9 << _TAG_PRIMARY_SIZE) /* BINARY */ #define SUB_BINARY_SUBTAG (0xA << _TAG_PRIMARY_SIZE) /* BINARY */ +#define MAP_SUBTAG (0xB << _TAG_PRIMARY_SIZE) /* MAP */ #define EXTERNAL_PID_SUBTAG (0xC << _TAG_PRIMARY_SIZE) /* EXTERNAL_PID */ #define EXTERNAL_PORT_SUBTAG (0xD << _TAG_PRIMARY_SIZE) /* EXTERNAL_PORT */ #define EXTERNAL_REF_SUBTAG (0xE << _TAG_PRIMARY_SIZE) /* EXTERNAL_REF */ @@ -155,6 +156,7 @@ struct erl_node_; /* Declared in erl_node_tables.h */ #define _TAG_HEADER_REFC_BIN (TAG_PRIMARY_HEADER|REFC_BINARY_SUBTAG) #define _TAG_HEADER_HEAP_BIN (TAG_PRIMARY_HEADER|HEAP_BINARY_SUBTAG) #define _TAG_HEADER_SUB_BIN (TAG_PRIMARY_HEADER|SUB_BINARY_SUBTAG) +#define _TAG_HEADER_MAP (TAG_PRIMARY_HEADER|MAP_SUBTAG) #define _TAG_HEADER_EXTERNAL_PID (TAG_PRIMARY_HEADER|EXTERNAL_PID_SUBTAG) #define _TAG_HEADER_EXTERNAL_PORT (TAG_PRIMARY_HEADER|EXTERNAL_PORT_SUBTAG) #define _TAG_HEADER_EXTERNAL_REF (TAG_PRIMARY_HEADER|EXTERNAL_REF_SUBTAG) @@ -354,7 +356,10 @@ _ET_DECLARE_CHECKED(Uint,thing_subtag,Eterm) #define is_value(x) ((x) != THE_NON_VALUE) /* binary object access methods */ -#define is_binary_header(x) (((x) & (_TAG_HEADER_MASK-_BINARY_XXX_MASK)) == _TAG_HEADER_REFC_BIN) +#define is_binary_header(x) \ + ((((x) & (_TAG_HEADER_MASK)) == _TAG_HEADER_REFC_BIN) || \ + (((x) & (_TAG_HEADER_MASK)) == _TAG_HEADER_HEAP_BIN) || \ + (((x) & (_TAG_HEADER_MASK)) == _TAG_HEADER_SUB_BIN)) #define make_binary(x) make_boxed((Eterm*)(x)) #define is_binary(x) (is_boxed((x)) && is_binary_header(*boxed_val((x)))) #define is_not_binary(x) (!is_binary((x))) @@ -1064,8 +1069,8 @@ _ET_DECLARE_CHECKED(Uint,y_reg_index,Uint) /* * Backwards compatibility definitions: - * - #define virtal *_DEF constants with values that fit term order: - * number < atom < ref < fun < port < pid < tuple < nil < cons < binary + * - #define virtual *_DEF constants with values that fit term order: + * number < atom < ref < fun < port < pid < tuple < map < nil < cons < binary * - tag_val_def() function generates virtual _DEF tag * - not_eq_tags() and NUMBER_CODE() defined in terms * of the tag_val_def() function @@ -1074,19 +1079,20 @@ _ET_DECLARE_CHECKED(Uint,y_reg_index,Uint) #define BINARY_DEF 0x0 #define LIST_DEF 0x1 #define NIL_DEF 0x2 -#define TUPLE_DEF 0x3 -#define PID_DEF 0x4 -#define EXTERNAL_PID_DEF 0x5 -#define PORT_DEF 0x6 -#define EXTERNAL_PORT_DEF 0x7 -#define EXPORT_DEF 0x8 -#define FUN_DEF 0x9 -#define REF_DEF 0xa -#define EXTERNAL_REF_DEF 0xb -#define ATOM_DEF 0xc -#define FLOAT_DEF 0xd -#define BIG_DEF 0xe -#define SMALL_DEF 0xf +#define MAP_DEF 0x3 +#define TUPLE_DEF 0x4 +#define PID_DEF 0x5 +#define EXTERNAL_PID_DEF 0x6 +#define PORT_DEF 0x7 +#define EXTERNAL_PORT_DEF 0x8 +#define EXPORT_DEF 0x9 +#define FUN_DEF 0xa +#define REF_DEF 0xb +#define EXTERNAL_REF_DEF 0xc +#define ATOM_DEF 0xd +#define FLOAT_DEF 0xe +#define BIG_DEF 0xf +#define SMALL_DEF 0x10 #if ET_DEBUG extern unsigned tag_val_def_debug(Wterm, const char*, unsigned); @@ -1096,8 +1102,8 @@ extern unsigned tag_val_def(Wterm); #endif #define not_eq_tags(X,Y) (tag_val_def((X)) ^ tag_val_def((Y))) -#define NUMBER_CODE(x,y) ((tag_val_def(x) << 4) | tag_val_def(y)) -#define _NUMBER_CODE(TX,TY) ((TX << 4) | TY) +#define NUMBER_CODE(x,y) ((tag_val_def(x) << 5) | tag_val_def(y)) +#define _NUMBER_CODE(TX,TY) ((TX << 5) | TY) #define SMALL_SMALL _NUMBER_CODE(SMALL_DEF,SMALL_DEF) #define SMALL_BIG _NUMBER_CODE(SMALL_DEF,BIG_DEF) #define SMALL_FLOAT _NUMBER_CODE(SMALL_DEF,FLOAT_DEF) @@ -1126,6 +1132,7 @@ extern unsigned tag_val_def(Wterm); #define make_tuple_rel make_boxed_rel #define make_external_rel make_boxed_rel #define make_internal_ref_rel make_boxed_rel +#define make_big_rel make_boxed_rel #define binary_val_rel(RTERM, BASE) binary_val(rterm2wterm(RTERM, BASE)) #define list_val_rel(RTERM, BASE) list_val(rterm2wterm(RTERM, BASE)) diff --git a/erts/emulator/beam/erl_utils.h b/erts/emulator/beam/erl_utils.h index 292d135946..5b81d814c6 100644 --- a/erts/emulator/beam/erl_utils.h +++ b/erts/emulator/beam/erl_utils.h @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 2012-2013. All Rights Reserved. + * Copyright Ericsson AB 2012-2014. All Rights Reserved. * * The contents of this file are subject to the Erlang Public License, * Version 1.1, (the "License"); you may not use this file except in @@ -202,23 +202,37 @@ int eq(Eterm, Eterm); #define EQ(x,y) (((x) == (y)) || (is_not_both_immed((x),(y)) && eq((x),(y)))) #if HALFWORD_HEAP -Sint cmp_rel(Eterm, Eterm*, Eterm, Eterm*); -#define CMP(A,B) cmp_rel(A,NULL,B,NULL) +Sint cmp_rel_opt(Eterm, Eterm*, Eterm, Eterm*, int); +#define cmp_rel(A,A_BASE,B,B_BASE) cmp_rel_opt(A,A_BASE,B,B_BASE,0) +#define cmp_rel_term(A,A_BASE,B,B_BASE) cmp_rel_opt(A,A_BASE,B,B_BASE,1) +#define CMP(A,B) cmp_rel_opt(A,NULL,B,NULL,0) +#define CMP_TERM(A,B) cmp_rel_opt(A,NULL,B,NULL,1) #else -Sint cmp(Eterm, Eterm); -#define cmp_rel(A,A_BASE,B,B_BASE) cmp(A,B) -#define CMP(A,B) cmp(A,B) +Sint cmp(Eterm, Eterm, int); +#define cmp_rel(A,A_BASE,B,B_BASE) cmp(A,B,0) +#define cmp_rel_term(A,A_BASE,B,B_BASE) cmp(A,B,1) +#define CMP(A,B) cmp(A,B,0) +#define CMP_TERM(A,B) cmp(A,B,1) #endif -#define cmp_lt(a,b) (CMP((a),(b)) < 0) -#define cmp_le(a,b) (CMP((a),(b)) <= 0) -#define cmp_eq(a,b) (CMP((a),(b)) == 0) -#define cmp_ne(a,b) (CMP((a),(b)) != 0) -#define cmp_ge(a,b) (CMP((a),(b)) >= 0) -#define cmp_gt(a,b) (CMP((a),(b)) > 0) - -#define CMP_LT(a,b) ((a) != (b) && cmp_lt((a),(b))) -#define CMP_GE(a,b) ((a) == (b) || cmp_ge((a),(b))) -#define CMP_EQ(a,b) ((a) == (b) || cmp_eq((a),(b))) -#define CMP_NE(a,b) ((a) != (b) && cmp_ne((a),(b))) + +#define cmp_lt(a,b) (CMP((a),(b)) < 0) +#define cmp_le(a,b) (CMP((a),(b)) <= 0) +#define cmp_eq(a,b) (CMP((a),(b)) == 0) +#define cmp_ne(a,b) (CMP((a),(b)) != 0) +#define cmp_ge(a,b) (CMP((a),(b)) >= 0) +#define cmp_gt(a,b) (CMP((a),(b)) > 0) + +#define cmp_lt_term(a,b) (CMP_TERM((a),(b)) < 0) +#define cmp_le_term(a,b) (CMP_TERM((a),(b)) <= 0) +#define cmp_ge_term(a,b) (CMP_TERM((a),(b)) >= 0) +#define cmp_gt_term(a,b) (CMP_TERM((a),(b)) > 0) + +#define CMP_LT(a,b) ((a) != (b) && cmp_lt((a),(b))) +#define CMP_GE(a,b) ((a) == (b) || cmp_ge((a),(b))) +#define CMP_EQ(a,b) ((a) == (b) || cmp_eq((a),(b))) +#define CMP_NE(a,b) ((a) != (b) && cmp_ne((a),(b))) + +#define CMP_LT_TERM(a,b) ((a) != (b) && cmp_lt_term((a),(b))) +#define CMP_GE_TERM(a,b) ((a) == (b) || cmp_ge_term((a),(b))) #endif diff --git a/erts/emulator/beam/erl_vm.h b/erts/emulator/beam/erl_vm.h index 337422eead..b7de8208ad 100644 --- a/erts/emulator/beam/erl_vm.h +++ b/erts/emulator/beam/erl_vm.h @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 1996-2013. All Rights Reserved. + * Copyright Ericsson AB 1996-2014. All Rights Reserved. * * The contents of this file are subject to the Erlang Public License, * Version 1.1, (the "License"); you may not use this file except in @@ -46,7 +46,6 @@ heap data on the C stack or if we use the buffers in the scheduler data. */ #define TMP_HEAP_SIZE 128 /* Number of Eterm in the schedulers small heap for transient heap data */ -#define CMP_TMP_HEAP_SIZE 32 /* cmp wants its own tmp-heap... */ #define ERL_ARITH_TMP_HEAP_SIZE 4 /* as does erl_arith... */ #define BEAM_EMU_TMP_HEAP_SIZE 2 /* and beam_emu... */ diff --git a/erts/emulator/beam/external.c b/erts/emulator/beam/external.c index 5e7a5cab6e..9fb2dbd8bf 100644 --- a/erts/emulator/beam/external.c +++ b/erts/emulator/beam/external.c @@ -42,6 +42,7 @@ #include "erl_binary.h" #include "erl_bits.h" #include "erl_zlib.h" +#include "erl_map.h" #ifdef HIPE #include "hipe_mode_switch.h" @@ -1168,6 +1169,7 @@ typedef struct { Eterm* hp_end; int remaining_n; char* remaining_bytes; + Eterm* maps_head; } B2TDecodeContext; typedef struct { @@ -1485,6 +1487,7 @@ static Eterm binary_to_term_int(Process* p, Uint32 flags, Eterm bin, Binary* con ctx->u.dc.hp_start = HAlloc(p, ctx->heap_size); ctx->u.dc.hp = ctx->u.dc.hp_start; ctx->u.dc.hp_end = ctx->u.dc.hp_start + ctx->heap_size; + ctx->u.dc.maps_head = NULL; ctx->state = B2TDecode; /*fall through*/ case B2TDecode: @@ -2555,6 +2558,38 @@ enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, } break; + case MAP_DEF: + { + map_t *mp = (map_t*)map_val(obj); + Uint size = map_get_size(mp); + Eterm *mptr; + + *ep++ = MAP_EXT; + put_int32(size, ep); ep += 4; + + /* Push values first */ + if (size > 0) { + mptr = map_get_values(mp); + for (i = size-1; i >= 1; i--) { + WSTACK_PUSH(s, ENC_TERM); + WSTACK_PUSH(s, (UWord) mptr[i]); + } + + WSTACK_PUSH(s, ENC_TERM); + WSTACK_PUSH(s, (UWord) mptr[0]); + + mptr = map_get_keys(mp); + for (i = size-1; i >= 1; i--) { + WSTACK_PUSH(s, ENC_TERM); + WSTACK_PUSH(s, (UWord) mptr[i]); + } + + obj = mptr[0]; + goto L_jump_start; + } + } + break; + case FLOAT_DEF: GET_DOUBLE(obj, f); if (dflags & DFLAG_NEW_FLOATS) { @@ -2845,6 +2880,7 @@ dec_term(ErtsDistExternal *edep, Eterm** hpp, byte* ep, ErlOffHeap* off_heap, int n; ErtsAtomEncoding char_enc; register Eterm* hp; /* Please don't take the address of hp */ + Eterm *maps_head; /* for validation of maps */ Eterm* next; SWord reds; @@ -2854,6 +2890,7 @@ dec_term(ErtsDistExternal *edep, Eterm** hpp, byte* ep, ErlOffHeap* off_heap, next = ctx->u.dc.next; ep = ctx->u.dc.ep; hpp = &ctx->u.dc.hp; + maps_head = ctx->u.dc.maps_head; if (ctx->state != B2TDecode) { int n_limit = reds; @@ -2934,6 +2971,7 @@ dec_term(ErtsDistExternal *edep, Eterm** hpp, byte* ep, ErlOffHeap* off_heap, reds = ERTS_SWORD_MAX; next = objp; *next = (Eterm) (UWord) NULL; + maps_head = NULL; } hp = *hpp; @@ -3469,6 +3507,65 @@ dec_term_atom_common: break; } break; + case MAP_EXT: + { + map_t *mp; + Uint32 size,n; + Eterm *kptr,*vptr; + Eterm keys; + + size = get_int32(ep); ep += 4; + + keys = make_tuple(hp); + *hp++ = make_arityval(size); + kptr = hp; + hp += size; + + mp = (map_t*)hp; + hp += MAP_HEADER_SIZE; + vptr = hp; + hp += size; + + /* kptr, first word for keys + * vptr, first word for values + */ + + /* + * Use thing_word to link through decoded maps. + * The list of maps is for later validation. + */ + + mp->thing_word = (Eterm) COMPRESS_POINTER(maps_head); + maps_head = (Eterm *) mp; + + mp->size = size; + mp->keys = keys; + *objp = make_map(mp); + + /* We assume the map is wellformed, meaning: + * - ascending key order + * - unique keys + */ + + objp = vptr + size - 1; + n = size; + + while (n-- > 0) { + *objp = (Eterm) COMPRESS_POINTER(next); + next = objp; + objp--; + } + + objp = kptr + size - 1; + n = size; + + while (n-- > 0) { + *objp = (Eterm) COMPRESS_POINTER(next); + next = objp; + objp--; + } + } + break; case NEW_FUN_EXT: { ErlFunThing* funp = (ErlFunThing *) hp; @@ -3678,21 +3775,7 @@ dec_term_atom_common: } default: - error: - /* UNDO: - * Must unlink all off-heap objects that may have been - * linked into the process. - */ - if (hp < *hpp) { /* Sometimes we used hp and sometimes *hpp */ - hp = *hpp; /* the largest must be the freshest */ - } - undo_offheap_in_area(off_heap, hp_saved, hp); - *hpp = hp_saved; - if (ctx) { - ctx->state = B2TDecodeFail; - ctx->reds = reds; - } - return NULL; + goto error; } if (--reds <= 0) { @@ -3701,6 +3784,7 @@ dec_term_atom_common: ctx->u.dc.ep = ep; ctx->u.dc.next = next; ctx->u.dc.hp = hp; + ctx->u.dc.maps_head = maps_head; ctx->reds = 0; return NULL; } @@ -3710,12 +3794,43 @@ dec_term_atom_common: } } } + + /* Iterate through all the maps and check for validity and sort keys + * - done here for when we know it is complete. + */ + + while (maps_head) { + next = (Eterm *)(EXPAND_POINTER(*maps_head)); + *maps_head = MAP_HEADER; + if (!erts_validate_and_sort_map((map_t*)maps_head)) + goto error; + maps_head = next; + } + if (ctx) { ctx->state = B2TDone; ctx->reds = reds; } + *hpp = hp; return ep; + +error: + /* UNDO: + * Must unlink all off-heap objects that may have been + * linked into the process. + */ + if (hp < *hpp) { /* Sometimes we used hp and sometimes *hpp */ + hp = *hpp; /* the largest must be the freshest */ + } + undo_offheap_in_area(off_heap, hp_saved, hp); + *hpp = hp_saved; + if (ctx) { + ctx->state = B2TDecodeFail; + ctx->reds = reds; + } + + return NULL; } /* returns the number of bytes needed to encode an object @@ -3885,6 +4000,46 @@ encode_size_struct_int(TTBSizeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, goto outer_loop; } break; + case MAP_DEF: + { + map_t *mp = (map_t*)map_val(obj); + Uint size = map_get_size(mp); + Uint i; + Eterm *ptr; + + result += 1 + 4; /* tag + 4 bytes size */ + + /* push values first */ + ptr = map_get_values(mp); + i = size; + while(i--) { + if (is_list(*ptr)) { + if ((m = is_string(*ptr)) && (m < MAX_STRING_LEN)) { + result += m + 2 + 1; + } else { + result += 5; + } + } + ESTACK_PUSH(s,*ptr); + ++ptr; + } + + ptr = map_get_keys(mp); + i = size; + while(i--) { + if (is_list(*ptr)) { + if ((m = is_string(*ptr)) && (m < MAX_STRING_LEN)) { + result += m + 2 + 1; + } else { + result += 5; + } + } + ESTACK_PUSH(s,*ptr); + ++ptr; + } + goto outer_loop; + } + break; case FLOAT_DEF: if (dflags & DFLAG_NEW_FLOATS) { result += 9; @@ -4175,6 +4330,13 @@ init_done: ADDTERMS(n); heap_size += n + 1; break; + case MAP_EXT: + CHKSIZE(4); + n = get_int32(ep); + ep += 4; + ADDTERMS(2*n); + heap_size += 3 + n + 1 + n; + break; case STRING_EXT: CHKSIZE(2); n = get_int16(ep); diff --git a/erts/emulator/beam/external.h b/erts/emulator/beam/external.h index 83001b2c7e..bf00958eb1 100644 --- a/erts/emulator/beam/external.h +++ b/erts/emulator/beam/external.h @@ -50,6 +50,7 @@ #define LARGE_BIG_EXT 'o' #define NEW_FUN_EXT 'p' #define EXPORT_EXT 'q' +#define MAP_EXT 't' #define FUN_EXT 'u' #define ATOM_UTF8_EXT 'v' #define SMALL_ATOM_UTF8_EXT 'w' diff --git a/erts/emulator/beam/global.h b/erts/emulator/beam/global.h index c183c519ff..8fcb95d0e2 100755 --- a/erts/emulator/beam/global.h +++ b/erts/emulator/beam/global.h @@ -371,9 +371,9 @@ extern int stackdump_on_exit; */ typedef struct { - UWord* start; - UWord* sp; - UWord* end; + Eterm* start; + Eterm* sp; + Eterm* end; ErtsAlcType_t alloc_type; }ErtsEStack; @@ -384,7 +384,7 @@ void erl_grow_estack(ErtsEStack*, Eterm* def_stack); #define ESTK_DEF_STACK(s) ESTK_CONCAT(s,_default_estack) #define DECLARE_ESTACK(s) \ - UWord ESTK_DEF_STACK(s)[DEF_ESTACK_SIZE]; \ + Eterm ESTK_DEF_STACK(s)[DEF_ESTACK_SIZE]; \ ErtsEStack s = { \ ESTK_DEF_STACK(s), /* start */ \ ESTK_DEF_STACK(s), /* sp */ \ @@ -418,8 +418,8 @@ do {\ if (s.start == ESTK_DEF_STACK(s)) {\ UWord _wsz = ESTACK_COUNT(s);\ (dst)->start = erts_alloc(s.alloc_type,\ - DEF_ESTACK_SIZE * sizeof(UWord));\ - memcpy((dst)->start, s.start,_wsz*sizeof(UWord));\ + DEF_ESTACK_SIZE * sizeof(Eterm));\ + memcpy((dst)->start, s.start,_wsz*sizeof(Eterm));\ (dst)->sp = (dst)->start + _wsz;\ (dst)->end = (dst)->start + DEF_ESTACK_SIZE;\ (dst)->alloc_type = s.alloc_type;\ @@ -495,7 +495,7 @@ typedef struct { #define DEF_WSTACK_SIZE (16) -void erl_grow_wstack(ErtsWStack*, Eterm* def_stack); +void erl_grow_wstack(ErtsWStack*, UWord* def_stack); #define WSTK_CONCAT(a,b) a##b #define WSTK_DEF_STACK(s) WSTK_CONCAT(s,_default_wstack) @@ -998,6 +998,7 @@ Eterm erts_gc_length_1(Process* p, Eterm* reg, Uint live); Eterm erts_gc_size_1(Process* p, Eterm* reg, Uint live); Eterm erts_gc_bit_size_1(Process* p, Eterm* reg, Uint live); Eterm erts_gc_byte_size_1(Process* p, Eterm* reg, Uint live); +Eterm erts_gc_map_size_1(Process* p, Eterm* reg, Uint live); Eterm erts_gc_abs_1(Process* p, Eterm* reg, Uint live); Eterm erts_gc_float_1(Process* p, Eterm* reg, Uint live); Eterm erts_gc_round_1(Process* p, Eterm* reg, Uint live); diff --git a/erts/emulator/beam/io.c b/erts/emulator/beam/io.c index 49af86b36a..3b16cdeb4a 100644 --- a/erts/emulator/beam/io.c +++ b/erts/emulator/beam/io.c @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 1996-2013. All Rights Reserved. + * Copyright Ericsson AB 1996-2014. All Rights Reserved. * * The contents of this file are subject to the Erlang Public License, * Version 1.1, (the "License"); you may not use this file except in @@ -46,6 +46,7 @@ #define ERTS_WANT_EXTERNAL_TAGS #include "external.h" #include "dtrace-wrapper.h" +#include "erl_map.h" extern ErlDrvEntry fd_driver_entry; extern ErlDrvEntry vanilla_driver_entry; @@ -5293,6 +5294,17 @@ driver_deliver_term(Eterm to, ErlDrvTermData* data, int len) depth++; break; } + case ERL_DRV_MAP: { /* int */ + ERTS_DDT_CHK_ENOUGH_ARGS(1); + if ((int) ptr[0] < 0) ERTS_DDT_FAIL; + need += MAP_HEADER_SIZE + 1 + 2*ptr[0]; + depth -= 2*ptr[0]; + if (depth < 0) ERTS_DDT_FAIL; + ptr++; + depth++; + break; + } + default: ERTS_DDT_FAIL; } @@ -5529,6 +5541,36 @@ driver_deliver_term(Eterm to, ErlDrvTermData* data, int len) ptr += 2; break; + case ERL_DRV_MAP: { /* int */ + int size = (int)ptr[0]; + Eterm* tp = hp; + Eterm* vp; + map_t *mp; + + *tp = make_arityval(size); + + hp += 1 + size; + mp = (map_t*)hp; + mp->thing_word = MAP_HEADER; + mp->size = size; + mp->keys = make_tuple(tp); + mess = make_map(mp); + + hp += MAP_HEADER_SIZE + size; /* advance "heap" pointer */ + + tp += size; /* point at last key */ + vp = hp - 1; /* point at last value */ + + while(size--) { + *vp-- = ESTACK_POP(stack); + *tp-- = ESTACK_POP(stack); + } + if (!erts_validate_and_sort_map(mp)) + ERTS_DDT_FAIL; + ptr++; + break; + } + } ESTACK_PUSH(stack, mess); } diff --git a/erts/emulator/beam/ops.tab b/erts/emulator/beam/ops.tab index c29f3f9b1b..9ea3a3a8ea 100644 --- a/erts/emulator/beam/ops.tab +++ b/erts/emulator/beam/ops.tab @@ -1466,6 +1466,75 @@ apply I apply_last I P # +# Map instructions in R17. +# + +# put_map Fail Src Dst Live Size Rest=* => jump Fail +# is_map Fail Src => jump Fail +# has_map_field Fail Src Key => jump Fail +# get_map_element Fail Src Key Dst => jump Fail + +put_map_assoc F n Dst Live Size Rest=* => new_map F Dst Live Size Rest +put_map_exact F n Dst Live Size Rest=* => new_map F Dst Live Size Rest +put_map_assoc F Src Dst Live Size Rest=* => \ + update_map_assoc F Src Dst Live Size Rest +put_map_exact F Src Dst Live Size Rest=* => \ + update_map_exact F Src Dst Live Size Rest + +new_map j d I I +update_map_assoc j s d I I +update_map_exact j s d I I + +is_map Fail cq => jump Fail + +%macro: is_map IsMap -fail_action +is_map f r +is_map f x +is_map f y + +has_map_field Fail Src=rxy Key=arxy => i_has_map_field Fail Src Key +has_map_field Fail Src Key => move Key x | i_has_map_field Fail Src x + +%macro: i_has_map_field HasMapField -fail_action +i_has_map_field f r a +i_has_map_field f x a +i_has_map_field f y a +i_has_map_field f r r +i_has_map_field f x r +i_has_map_field f y r +i_has_map_field f r x +i_has_map_field f x x +i_has_map_field f y x +i_has_map_field f r y +i_has_map_field f x y +i_has_map_field f y y + +get_map_element Fail Src=rxy Key=ax Dst => i_get_map_element Fail Src Key Dst +get_map_element Fail Src=rxy Key=rycq Dst => \ + move Key x | i_get_map_element Fail Src x Dst +get_map_element Fail Src Key Dst => jump Fail + +%macro: i_get_map_element GetMapElement -fail_action +i_get_map_element f r a r +i_get_map_element f x a r +i_get_map_element f y a r +i_get_map_element f r a x +i_get_map_element f x a x +i_get_map_element f y a x +i_get_map_element f r a y +i_get_map_element f x a y +i_get_map_element f y a y +i_get_map_element f r x r +i_get_map_element f x x r +i_get_map_element f y x r +i_get_map_element f r x x +i_get_map_element f x x x +i_get_map_element f y x x +i_get_map_element f r x y +i_get_map_element f x x y +i_get_map_element f y x y + +# # Optimize addition and subtraction of small literals using # the i_increment/4 instruction (in bodies, not in guards). # diff --git a/erts/emulator/beam/utils.c b/erts/emulator/beam/utils.c index 7f8bdcb2ca..bc4a05d385 100644 --- a/erts/emulator/beam/utils.c +++ b/erts/emulator/beam/utils.c @@ -31,6 +31,7 @@ #include "bif.h" #include "erl_binary.h" #include "erl_bits.h" +#include "erl_map.h" #include "packet_parser.h" #include "erl_gc.h" #define ERTS_WANT_DB_INTERNAL__ @@ -734,6 +735,8 @@ erts_bld_atom_2uint_3tup_list(Uint **hpp, Uint *szp, Sint length, #define FUNNY_NUMBER10 268440479 #define FUNNY_NUMBER11 268440577 #define FUNNY_NUMBER12 268440581 +#define FUNNY_NUMBER13 268440593 +#define FUNNY_NUMBER14 268440611 static Uint32 hash_binary_bytes(Eterm bin, Uint sz, Uint32 hash) @@ -785,10 +788,10 @@ Uint32 make_hash(Eterm term_arg) unsigned op; /* Must not collide with the real tag_val_def's: */ -#define MAKE_HASH_TUPLE_OP 0x10 -#define MAKE_HASH_FUN_OP 0x11 -#define MAKE_HASH_CDR_PRE_OP 0x12 -#define MAKE_HASH_CDR_POST_OP 0x13 +#define MAKE_HASH_TUPLE_OP 0x11 +#define MAKE_HASH_TERM_ARRAY_OP 0x12 +#define MAKE_HASH_CDR_PRE_OP 0x13 +#define MAKE_HASH_CDR_POST_OP 0x14 /* ** Convenience macro for calculating a bytewise hash on an unsigned 32 bit @@ -877,7 +880,7 @@ tail_recur: hash = hash*FUNNY_NUMBER2 + funp->fe->old_uniq; if (num_free > 0) { if (num_free > 1) { - WSTACK_PUSH3(stack, (UWord) &funp->env[1], (num_free-1), MAKE_HASH_FUN_OP); + WSTACK_PUSH3(stack, (UWord) &funp->env[1], (num_free-1), MAKE_HASH_TERM_ARRAY_OP); } term = funp->env[0]; goto tail_recur; @@ -967,6 +970,24 @@ tail_recur: hash *= is_neg ? FUNNY_NUMBER4 : FUNNY_NUMBER3; break; } + case MAP_DEF: + { + map_t *mp = (map_t *)map_val(term); + int size = map_get_size(mp); + Eterm *ks = map_get_keys(mp); + Eterm *vs = map_get_values(mp); + + /* Use a prime with size to remedy some of + * the {} and <<>> hash problems */ + hash = hash*FUNNY_NUMBER13 + FUNNY_NUMBER14 + size; + if (size == 0) + break; + + /* push values first */ + WSTACK_PUSH3(stack, (UWord)vs, (UWord) size, MAKE_HASH_TERM_ARRAY_OP); + WSTACK_PUSH3(stack, (UWord)ks, (UWord) size, MAKE_HASH_TERM_ARRAY_OP); + break; + } case TUPLE_DEF: { Eterm* ptr = tuple_val(term); @@ -976,7 +997,7 @@ tail_recur: op = MAKE_HASH_TUPLE_OP; }/*fall through*/ case MAKE_HASH_TUPLE_OP: - case MAKE_HASH_FUN_OP: + case MAKE_HASH_TERM_ARRAY_OP: { Uint i = (Uint) WSTACK_POP(stack); Eterm* ptr = (Eterm*) WSTACK_POP(stack); @@ -1070,9 +1091,11 @@ Uint32 make_hash2(Eterm term) { Uint32 hash; + Uint32 hash_xor_keys = 0; + Uint32 hash_xor_values = 0; DeclareTmpHeapNoproc(tmp_big,2); -/* (HCONST * {2, ..., 14}) mod 2^32 */ +/* (HCONST * {2, ..., 16}) mod 2^32 */ #define HCONST_2 0x3c6ef372UL #define HCONST_3 0xdaa66d2bUL #define HCONST_4 0x78dde6e4UL @@ -1087,6 +1110,11 @@ make_hash2(Eterm term) #define HCONST_13 0x08d12e65UL #define HCONST_14 0xa708a81eUL #define HCONST_15 0x454021d7UL +#define HCONST_16 0xe3779b90UL + +#define HASH_MAP_TAIL (_make_header(1,_TAG_HEADER_REF)) +#define HASH_MAP_KEY (_make_header(2,_TAG_HEADER_REF)) +#define HASH_MAP_VAL (_make_header(3,_TAG_HEADER_REF)) #define UINT32_HASH_2(Expr1, Expr2, AConst) \ do { \ @@ -1182,11 +1210,45 @@ make_hash2(Eterm term) UINT32_HASH(arity, HCONST_9); if (arity == 0) /* Empty tuple */ goto hash2_common; - for (i = arity; i >= 2; i--) { + for (i = arity; i >= 1; i--) { tmp = elem[i]; ESTACK_PUSH(s, tmp); } - term = elem[1]; + goto hash2_common; + } + break; + case MAP_SUBTAG: + { + map_t *mp = (map_t *)map_val(term); + int i; + int size = map_get_size(mp); + Eterm *ks = map_get_keys(mp); + Eterm *vs = map_get_values(mp); + UINT32_HASH(size, HCONST_16); + if (size == 0) { + goto hash2_common; + } + ESTACK_PUSH(s, hash_xor_values); + ESTACK_PUSH(s, hash_xor_keys); + ESTACK_PUSH(s, hash); + ESTACK_PUSH(s, HASH_MAP_TAIL); + hash = 0; + hash_xor_keys = 0; + hash_xor_values = 0; + for (i = size - 1; i >= 0; i--) { + tmp = vs[i]; + ESTACK_PUSH(s, HASH_MAP_VAL); + ESTACK_PUSH(s, tmp); + } + /* We do not want to expose the tuple representation. + * Do not push the keys as a tuple. + */ + for (i = size - 1; i >= 0; i--) { + tmp = ks[i]; + ESTACK_PUSH(s, HASH_MAP_KEY); + ESTACK_PUSH(s, tmp); + } + goto hash2_common; } break; case EXPORT_SUBTAG: @@ -1380,15 +1442,47 @@ make_hash2(Eterm term) default: erl_exit(1, "Invalid tag in make_hash2(0x%X)\n", term); hash2_common: + + /* Uint32 hash always has the hash value of the previous term, + * compounded or otherwise. + */ + if (ESTACK_ISEMPTY(s)) { DESTROY_ESTACK(s); UnUseTmpHeapNoproc(2); return hash; } + term = ESTACK_POP(s); + + switch (term) { + case HASH_MAP_TAIL: { + hash = (Uint32) ESTACK_POP(s); + UINT32_HASH(hash_xor_keys, HCONST_16); + UINT32_HASH(hash_xor_values, HCONST_16); + hash_xor_keys = (Uint32) ESTACK_POP(s); + hash_xor_values = (Uint32) ESTACK_POP(s); + goto hash2_common; + } + case HASH_MAP_KEY: + hash_xor_keys ^= hash; + hash = 0; + goto hash2_common; + case HASH_MAP_VAL: + hash_xor_values ^= hash; + hash = 0; + goto hash2_common; + default: + break; + } } } } + +#undef HASH_MAP_TAIL +#undef HASH_MAP_KEY +#undef HASH_MAP_VAL + #undef UINT32_HASH_2 #undef UINT32_HASH #undef SINT32_HASH @@ -1490,7 +1584,7 @@ tail_recur: hash = hash*FUNNY_NUMBER2 + funp->fe->old_uniq; if (num_free > 0) { if (num_free > 1) { - WSTACK_PUSH3(stack, (UWord) &funp->env[1], (num_free-1), MAKE_HASH_FUN_OP); + WSTACK_PUSH3(stack, (UWord) &funp->env[1], (num_free-1), MAKE_HASH_TERM_ARRAY_OP); } term = funp->env[0]; goto tail_recur; @@ -1603,6 +1697,24 @@ tail_recur: } break; + case MAP_DEF: + { + map_t *mp = (map_t *)map_val(term); + int size = map_get_size(mp); + Eterm *ks = map_get_keys(mp); + Eterm *vs = map_get_values(mp); + + /* Use a prime with size to remedy some of + * the {} and <<>> hash problems */ + hash = hash*FUNNY_NUMBER13 + FUNNY_NUMBER14 + size; + if (size == 0) + break; + + /* push values first */ + WSTACK_PUSH3(stack, (UWord)vs, (UWord) size, MAKE_HASH_TERM_ARRAY_OP); + WSTACK_PUSH3(stack, (UWord)ks, (UWord) size, MAKE_HASH_TERM_ARRAY_OP); + break; + } case TUPLE_DEF: { Eterm* ptr = tuple_val(term); @@ -1612,7 +1724,7 @@ tail_recur: op = MAKE_HASH_TUPLE_OP; }/*fall through*/ case MAKE_HASH_TUPLE_OP: - case MAKE_HASH_FUN_OP: + case MAKE_HASH_TERM_ARRAY_OP: { Uint i = (Uint) WSTACK_POP(stack); Eterm* ptr = (Eterm*) WSTACK_POP(stack); @@ -1640,7 +1752,7 @@ tail_recur: return hash; #undef MAKE_HASH_TUPLE_OP -#undef MAKE_HASH_FUN_OP +#undef MAKE_HASH_TERM_ARRAY_OP #undef MAKE_HASH_CDR_PRE_OP #undef MAKE_HASH_CDR_POST_OP } @@ -2007,6 +2119,18 @@ tailrecur_ne: ++bb; goto term_array; } + case MAP_SUBTAG: + { + aa = map_val_rel(a, a_base); + if (!is_boxed(b) || *boxed_val_rel(b,b_base) != *aa) + goto not_equal; + bb = map_val_rel(b,b_base); + if ((sz = map_get_size((map_t*)aa)) == 0) goto pop_next; + aa += 2; + bb += 2; + sz += 1; /* increment for tuple-keys */ + goto term_array; + } case REFC_BINARY_SUBTAG: case HEAP_BINARY_SUBTAG: case SUB_BINARY_SUBTAG: @@ -2281,7 +2405,7 @@ static int cmpbytes(byte *s1, int l1, byte *s2, int l2) * * According to the Erlang Standard, types are orderered as follows: * numbers < (characters) < atoms < refs < funs < ports < pids < - * tuples < [] < conses < binaries. + * tuples < maps < [] < conses < binaries. * * Note that characters are currently not implemented. * @@ -2301,10 +2425,14 @@ static int cmp_atoms(Eterm a, Eterm b) bb->name+3, bb->len-3); } +/* cmp(Eterm a, Eterm b, int exact) + * exact = 1 -> term-based compare + * exact = 0 -> arith-based compare + */ #if HALFWORD_HEAP -Sint cmp_rel(Eterm a, Eterm* a_base, Eterm b, Eterm* b_base) +Sint cmp_rel_opt(Eterm a, Eterm* a_base, Eterm b, Eterm* b_base, int exact) #else -Sint cmp(Eterm a, Eterm b) +Sint cmp(Eterm a, Eterm b, int exact) #endif { DECLARE_WSTACK(stack); @@ -2464,7 +2592,25 @@ tailrecur_ne: ++aa; ++bb; goto term_array; + case (_TAG_HEADER_MAP >> _TAG_PRIMARY_SIZE) : + if (!is_map_rel(b,b_base)) { + a_tag = MAP_DEF; + goto mixed_types; + } + aa = (Eterm *)map_val_rel(a,a_base); + bb = (Eterm *)map_val_rel(b,b_base); + i = map_get_size((map_t*)aa); + if (i != map_get_size((map_t*)bb)) { + RETURN_NEQ((int)(i - map_get_size((map_t*)bb))); + } + if (i == 0) { + goto pop_next; + } + aa += 2; + bb += 2; + i += 1; /* increment for tuple-keys */ + goto term_array; case (_TAG_HEADER_FLOAT >> _TAG_PRIMARY_SIZE): if (!is_float_rel(b,b_base)) { a_tag = FLOAT_DEF; @@ -2688,11 +2834,6 @@ tailrecur_ne: { FloatDef f1, f2; Eterm big; -#if HEAP_ON_C_STACK - Eterm big_buf[CMP_TMP_HEAP_SIZE]; /* If HEAP_ON_C_STACK */ -#else - Eterm *big_buf = erts_get_scheduler_data()->cmp_tmp_heap; -#endif #if HALFWORD_HEAP Wterm aw = is_immed(a) ? a : rterm2wterm(a,a_base); Wterm bw = is_immed(b) ? b : rterm2wterm(b,b_base); @@ -2703,6 +2844,8 @@ tailrecur_ne: #define MAX_LOSSLESS_FLOAT ((double)((1LL << 53) - 2)) #define MIN_LOSSLESS_FLOAT ((double)(((1LL << 53) - 2)*-1)) #define BIG_ARITY_FLOAT_MAX (1024 / D_EXP) /* arity of max float as a bignum */ + Eterm big_buf[BIG_NEED_SIZE(BIG_ARITY_FLOAT_MAX)]; + b_tag = tag_val_def(bw); switch(_NUMBER_CODE(a_tag, b_tag)) { @@ -2713,13 +2856,15 @@ tailrecur_ne: j = big_sign(aw) ? -1 : 1; break; case SMALL_FLOAT: + if (exact) goto exact_fall_through; GET_DOUBLE(bw, f2); if (f2.fd < MAX_LOSSLESS_FLOAT && f2.fd > MIN_LOSSLESS_FLOAT) { /* Float is within the no loss limit */ f1.fd = signed_val(aw); j = float_comp(f1.fd, f2.fd); + } #if ERTS_SIZEOF_ETERM == 8 - } else if (f2.fd > (double) (MAX_SMALL + 1)) { + else if (f2.fd > (double) (MAX_SMALL + 1)) { /* Float is a positive bignum, i.e. bigger */ j = -1; } else if (f2.fd < (double) (MIN_SMALL - 1)) { @@ -2730,19 +2875,21 @@ tailrecur_ne: j = signed_val(aw) - (Sint) f2.fd; } #else - } else { + else { /* If float is positive it is bigger than small */ j = (f2.fd > 0.0) ? -1 : 1; } #endif /* ERTS_SIZEOF_ETERM == 8 */ break; case FLOAT_BIG: + if (exact) goto exact_fall_through; { Wterm tmp = aw; aw = bw; bw = tmp; }/* fall through */ case BIG_FLOAT: + if (exact) goto exact_fall_through; GET_DOUBLE(bw, f2); if ((f2.fd < (double) (MAX_SMALL + 1)) && (f2.fd > (double) (MIN_SMALL - 1))) { @@ -2764,21 +2911,23 @@ tailrecur_ne: j = float_comp(f1.fd, f2.fd); } } else { - big = double_to_big(f2.fd, big_buf); - j = big_comp(aw, big); + big = double_to_big(f2.fd, big_buf, sizeof(big_buf)/sizeof(Eterm)); + j = big_comp(aw, rterm2wterm(big,big_buf)); } if (_NUMBER_CODE(a_tag, b_tag) == FLOAT_BIG) { j = -j; } break; case FLOAT_SMALL: + if (exact) goto exact_fall_through; GET_DOUBLE(aw, f1); if (f1.fd < MAX_LOSSLESS_FLOAT && f1.fd > MIN_LOSSLESS_FLOAT) { /* Float is within the no loss limit */ f2.fd = signed_val(bw); j = float_comp(f1.fd, f2.fd); + } #if ERTS_SIZEOF_ETERM == 8 - } else if (f1.fd > (double) (MAX_SMALL + 1)) { + else if (f1.fd > (double) (MAX_SMALL + 1)) { /* Float is a positive bignum, i.e. bigger */ j = 1; } else if (f1.fd < (double) (MIN_SMALL - 1)) { @@ -2789,12 +2938,13 @@ tailrecur_ne: j = (Sint) f1.fd - signed_val(bw); } #else - } else { + else { /* If float is positive it is bigger than small */ j = (f1.fd > 0.0) ? 1 : -1; } #endif /* ERTS_SIZEOF_ETERM == 8 */ break; +exact_fall_through: default: j = b_tag - a_tag; } |