diff options
Diffstat (limited to 'erts/emulator/beam')
70 files changed, 8139 insertions, 2331 deletions
diff --git a/erts/emulator/beam/atom.names b/erts/emulator/beam/atom.names index 5d06a32941..c097866c7e 100644 --- a/erts/emulator/beam/atom.names +++ b/erts/emulator/beam/atom.names @@ -198,6 +198,7 @@ atom dotall atom driver atom driver_options atom dsend +atom dsend_continue_trap atom dunlink atom duplicate_bag atom dupnames @@ -279,7 +280,6 @@ atom http httph https http_response http_request http_header http_eoh http_error atom id atom if_clause atom ignore -atom imports atom in atom in_exiting atom inactive @@ -335,6 +335,7 @@ atom max atom maximum atom max_tables max_processes atom mbuf_size +atom md5 atom memory atom memory_internal atom memory_types diff --git a/erts/emulator/beam/beam_debug.c b/erts/emulator/beam/beam_debug.c index a3cd08834f..6bb987985d 100644 --- a/erts/emulator/beam/beam_debug.c +++ b/erts/emulator/beam/beam_debug.c @@ -579,9 +579,29 @@ print_op(int to, void *to_arg, int op, int size, BeamInstr* addr) unpacked = ap; ap = addr + size; switch (op) { - case op_i_select_val_rfI: - case op_i_select_val_xfI: - case op_i_select_val_yfI: + case op_i_select_val_lins_rfI: + case op_i_select_val_lins_xfI: + case op_i_select_val_lins_yfI: + { + int n = ap[-1]; + int ix = n; + + while (ix--) { + erts_print(to, to_arg, "%T ", (Eterm) ap[0]); + ap++; + size++; + } + ix = n; + while (ix--) { + erts_print(to, to_arg, "f(" HEXF ") ", (Eterm) ap[0]); + ap++; + size++; + } + } + break; + case op_i_select_val_bins_rfI: + case op_i_select_val_bins_xfI: + case op_i_select_val_bins_yfI: { int n = ap[-1]; @@ -598,13 +618,19 @@ print_op(int to, void *to_arg, int op, int size, BeamInstr* addr) case op_i_select_tuple_arity_yfI: { int n = ap[-1]; + int ix = n; - while (n > 0) { + while (ix--) { Uint arity = arityval(ap[0]); - erts_print(to, to_arg, " {%d} f(" HEXF ")", arity, ap[1]); - ap += 2; - size += 2; - n--; + erts_print(to, to_arg, "{%d} ", arity, ap[1]); + ap++; + size++; + } + ix = n; + while (ix--) { + erts_print(to, to_arg, "f(" HEXF ") ", ap[0]); + ap++; + size++; } } break; diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c index 8bfb7d2ad2..6526e87e4c 100644 --- a/erts/emulator/beam/beam_emu.c +++ b/erts/emulator/beam/beam_emu.c @@ -241,10 +241,6 @@ BeamInstr beam_return_time_trace[1]; /* OpCode(i_return_time_trace) */ void** beam_ops; #endif -#ifndef ERTS_SMP /* Not supported with smp emulator */ -extern int count_instructions; -#endif - #define SWAPIN \ HTOP = HEAP_TOP(c_p); \ E = c_p->stop @@ -703,7 +699,7 @@ extern int count_instructions; Fail; \ } -#define IsMap(Src, Fail) if (is_not_map(Src)) { Fail; } +#define IsMap(Src, Fail) if (!is_map(Src)) { Fail; } #define HasMapField(Src, Key, Fail) if (has_not_map_field(Src, Key)) { Fail; } @@ -1163,14 +1159,15 @@ void process_main(void) Eterm (*arith_func)(Process* p, Eterm* reg, Uint live); -#ifndef NO_JUMP_TABLE - static void* opcodes[] = { DEFINE_OPCODES }; #ifdef ERTS_OPCODE_COUNTER_SUPPORT static void* counting_opcodes[] = { DEFINE_COUNTING_OPCODES }; -#endif +#else +#ifndef NO_JUMP_TABLE + static void* opcodes[] = { DEFINE_OPCODES }; #else int Go; #endif +#endif Uint temp_bits; /* Temporary used by BsSkipBits2 & BsGetInteger2 */ @@ -2141,19 +2138,18 @@ void process_main(void) NextPF(0, next); } - { Eterm select_val2; - OpCase(i_select_tuple_arity2_yfAfAf): + OpCase(i_select_tuple_arity2_yfAAff): select_val2 = yb(Arg(0)); goto do_select_tuple_arity2; - OpCase(i_select_tuple_arity2_xfAfAf): + OpCase(i_select_tuple_arity2_xfAAff): select_val2 = xb(Arg(0)); goto do_select_tuple_arity2; - OpCase(i_select_tuple_arity2_rfAfAf): + OpCase(i_select_tuple_arity2_rfAAff): select_val2 = r(0); I--; @@ -2164,22 +2160,22 @@ void process_main(void) select_val2 = *tuple_val(select_val2); goto do_select_val2; - OpCase(i_select_val2_yfcfcf): + OpCase(i_select_val2_yfccff): select_val2 = yb(Arg(0)); goto do_select_val2; - OpCase(i_select_val2_xfcfcf): + OpCase(i_select_val2_xfccff): select_val2 = xb(Arg(0)); goto do_select_val2; - OpCase(i_select_val2_rfcfcf): + OpCase(i_select_val2_rfccff): select_val2 = r(0); I--; do_select_val2: if (select_val2 == Arg(2)) { - I += 2; - } else if (select_val2 == Arg(4)) { + I += 3; + } else if (select_val2 == Arg(3)) { I += 4; } @@ -2206,20 +2202,50 @@ void process_main(void) do_select_tuple_arity: if (is_tuple(select_val)) { select_val = *tuple_val(select_val); - goto do_binary_search; + goto do_linear_search; + } + SET_I((BeamInstr *) Arg(1)); + Goto(*I); + + OpCase(i_select_val_lins_xfI): + select_val = xb(Arg(0)); + goto do_linear_search; + + OpCase(i_select_val_lins_yfI): + select_val = yb(Arg(0)); + goto do_linear_search; + + OpCase(i_select_val_lins_rfI): + select_val = r(0); + I--; + + do_linear_search: { + BeamInstr *vs = &Arg(3); + int ix = 0; + + for(;;) { + if (vs[ix+0] >= select_val) { ix += 0; break; } + if (vs[ix+1] >= select_val) { ix += 1; break; } + ix += 2; } + + if (vs[ix] == select_val) { + I += ix + Arg(2) + 2; + } + SET_I((BeamInstr *) Arg(1)); Goto(*I); + } - OpCase(i_select_val_xfI): + OpCase(i_select_val_bins_xfI): select_val = xb(Arg(0)); goto do_binary_search; - OpCase(i_select_val_yfI): + OpCase(i_select_val_bins_yfI): select_val = yb(Arg(0)); goto do_binary_search; - OpCase(i_select_val_rfI): + OpCase(i_select_val_bins_rfI): select_val = r(0); I--; @@ -2366,7 +2392,7 @@ void process_main(void) } OpCase(i_has_map_fields_fsI): { - map_t* mp; + flatmap_t* mp; Eterm map; Eterm field; Eterm *ks; @@ -2374,22 +2400,34 @@ void process_main(void) Uint sz,n; GetArg1(1, map); + n = (Uint)Arg(2); + fs = &Arg(3); /* pattern fields */ - /* this instruction assumes Arg1 is a map, - * i.e. that it follows a test is_map if needed. - */ + /* get term from field? */ + if (is_hashmap(map)) { + Uint32 hx; + while(n--) { + field = *fs++; + hx = hashmap_make_hash(field); + if (!erts_hashmap_get(hx,field,map)) { + SET_I((BeamInstr *) Arg(0)); + goto has_map_fields_fail; + } + } + goto has_map_fields_ok; + } + + ASSERT(is_flatmap(map)); - mp = (map_t *)map_val(map); - sz = map_get_size(mp); + mp = (flatmap_t *)flatmap_val(map); + sz = flatmap_get_size(mp); if (sz == 0) { SET_I((BeamInstr *) Arg(0)); goto has_map_fields_fail; } - ks = map_get_keys(mp); - n = (Uint)Arg(2); - fs = &Arg(3); /* pattern fields */ + ks = flatmap_get_keys(mp); ASSERT(n>0); @@ -2407,7 +2445,7 @@ void process_main(void) SET_I((BeamInstr *) Arg(0)); goto has_map_fields_fail; } - +has_map_fields_ok: I += 4 + Arg(2); has_map_fields_fail: ASSERT(VALID_INSTR(*I)); @@ -2434,12 +2472,8 @@ do { \ OpCase(i_get_map_elements_fsI): { Eterm map; - map_t *mp; - Eterm field; - Eterm *ks; - Eterm *vs; BeamInstr *fs; - Uint sz,n; + Uint sz, n; GetArg1(1, map); @@ -2447,36 +2481,56 @@ do { \ * i.e. that it follows a test is_map if needed. */ - mp = (map_t *)map_val(map); - sz = map_get_size(mp); - - if (sz == 0) { - SET_I((BeamInstr *) Arg(0)); - goto get_map_elements_fail; - } - n = (Uint)Arg(2) / 2; fs = &Arg(3); /* pattern fields and target registers */ - ks = map_get_keys(mp); - vs = map_get_values(mp); - while(sz) { - field = (Eterm)*fs; - if (EQ(field,*ks)) { - PUT_TERM_REG(*vs, fs[1]); - n--; + if (is_flatmap(map)) { + flatmap_t *mp; + Eterm *ks; + Eterm *vs; + + mp = (flatmap_t *)flatmap_val(map); + sz = flatmap_get_size(mp); + + if (sz == 0) { + SET_I((BeamInstr *) Arg(0)); + goto get_map_elements_fail; + } + + ks = flatmap_get_keys(mp); + vs = flatmap_get_values(mp); + + while(sz) { + if (EQ((Eterm)*fs,*ks)) { + PUT_TERM_REG(*vs, fs[1]); + n--; + fs += 2; + /* no more values to fetch, we are done */ + if (n == 0) break; + } + ks++; sz--; + vs++; + } + + if (n) { + SET_I((BeamInstr *) Arg(0)); + goto get_map_elements_fail; + } + } else { + const Eterm *v; + Uint32 hx; + ASSERT(is_hashmap(map)); + while(n--) { + hx = hashmap_make_hash((Eterm)*fs); + if ((v = erts_hashmap_get(hx,(Eterm)*fs, map)) == NULL) { + SET_I((BeamInstr *) Arg(0)); + goto get_map_elements_fail; + } + PUT_TERM_REG(*v, fs[1]); fs += 2; - /* no more values to fetch, we are done */ - if (n == 0) break; } - ks++; sz--; - vs++; } - if (n) { - SET_I((BeamInstr *) Arg(0)); - goto get_map_elements_fail; - } I += 4 + Arg(2); get_map_elements_fail: @@ -2775,6 +2829,7 @@ get_map_elements_fail: } PreFetch(1, next); ASSERT(!ERTS_PROC_IS_EXITING(c_p)); + ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p); reg[0] = r(0); result = (*bf)(c_p, reg, I); ASSERT(!ERTS_PROC_IS_EXITING(c_p) || is_non_value(result)); @@ -3556,7 +3611,7 @@ get_map_elements_fail: vbf = (BifFunction) Arg(0); PROCESS_MAIN_CHK_LOCKS(c_p); bif_nif_arity = I[-1]; - ASSERT(bif_nif_arity <= 3); + ASSERT(bif_nif_arity <= 4); ERTS_SMP_UNREQ_PROC_MAIN_LOCK(c_p); reg[0] = r(0); { @@ -3782,8 +3837,6 @@ get_map_elements_fail: * Allocate the binary struct itself. */ bptr = erts_bin_nrml_alloc(num_bytes); - bptr->flags = 0; - bptr->orig_size = num_bytes; erts_refc_init(&bptr->refc, 1); erts_current_bin = (byte *) bptr->orig_bytes; @@ -3883,8 +3936,6 @@ get_map_elements_fail: * Allocate the binary struct itself. */ bptr = erts_bin_nrml_alloc(tmp_arg1); - bptr->flags = 0; - bptr->orig_size = tmp_arg1; erts_refc_init(&bptr->refc, 1); erts_current_bin = (byte *) bptr->orig_bytes; @@ -4987,14 +5038,14 @@ get_map_elements_fail: * ... remainder of original BEAM code */ ASSERT(I[-5] == (Uint) OpCode(i_func_info_IaaI)); - c_p->hipe.ncallee = (void(*)(void)) I[-4]; + c_p->hipe.u.ncallee = (void(*)(void)) I[-4]; cmd = HIPE_MODE_SWITCH_CMD_CALL | (I[-1] << 8); ++hipe_trap_count; goto L_hipe_mode_switch; } OpCase(hipe_trap_call_closure): { ASSERT(I[-5] == (Uint) OpCode(i_func_info_IaaI)); - c_p->hipe.ncallee = (void(*)(void)) I[-4]; + c_p->hipe.u.ncallee = (void(*)(void)) I[-4]; cmd = HIPE_MODE_SWITCH_CMD_CALL_CLOSURE | (I[-1] << 8); ++hipe_trap_count; goto L_hipe_mode_switch; @@ -5028,7 +5079,10 @@ get_map_elements_fail: case HIPE_MODE_SWITCH_RES_RETURN: ASSERT(is_value(reg[0])); MoveReturn(reg[0], r(0)); - case HIPE_MODE_SWITCH_RES_CALL: + case HIPE_MODE_SWITCH_RES_CALL_EXPORTED: + c_p->i = c_p->hipe.u.callee_exp->addressv[erts_active_code_ix()]; + /*fall through*/ + case HIPE_MODE_SWITCH_RES_CALL_BEAM: SET_I(c_p->i); r(0) = reg[0]; Dispatch(); @@ -5145,22 +5199,16 @@ get_map_elements_fail: #ifndef NO_JUMP_TABLE #ifdef ERTS_OPCODE_COUNTER_SUPPORT - /* Are tables correctly generated by beam_makeops? */ - ASSERT(sizeof(counting_opcodes) == sizeof(opcodes)); - - if (count_instructions) { + ERTS_CT_ASSERT(sizeof(counting_opcodes) == sizeof(opcodes)); #ifdef DEBUG - counting_opcodes[op_catch_end_y] = LabelAddr(lb_catch_end_y); + counting_opcodes[op_catch_end_y] = LabelAddr(lb_catch_end_y); #endif - counting_opcodes[op_i_func_info_IaaI] = LabelAddr(lb_i_func_info_IaaI); - beam_ops = counting_opcodes; - } - else -#endif /* #ifndef ERTS_OPCODE_COUNTER_SUPPORT */ - { - beam_ops = opcodes; - } + counting_opcodes[op_i_func_info_IaaI] = LabelAddr(lb_i_func_info_IaaI); + beam_ops = counting_opcodes; +#else /* #ifndef ERTS_OPCODE_COUNTER_SUPPORT */ + beam_ops = opcodes; +#endif /* ERTS_OPCODE_COUNTER_SUPPORT */ #endif /* NO_JUMP_TABLE */ em_call_error_handler = OpCode(call_error_handler); @@ -6424,55 +6472,69 @@ new_fun(Process* p, Eterm* reg, ErlFunEntry* fe, int num_free) 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; + Uint32 hx; + if (is_flatmap(map)) { + flatmap_t* mp; + Eterm* keys; + Uint i; + Uint n; + + mp = (flatmap_t *)flatmap_val(map); + keys = flatmap_get_keys(mp); + n = flatmap_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; + } else { + for (i = 0; i < n; i++) { + if (EQ(keys[i], key)) { + return 0; + } } } + return 1; } - return 1; + ASSERT(is_hashmap(map)); + hx = hashmap_make_hash(key); + return erts_hashmap_get(hx,key,map) ? 0 : 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]; + Uint32 hx; + const Eterm *vs; + if (is_flatmap(map)) { + flatmap_t *mp; + Eterm *ks; + Uint i; + Uint n; + + mp = (flatmap_t *)flatmap_val(map); + ks = flatmap_get_keys(mp); + vs = flatmap_get_values(mp); + n = flatmap_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]; + } else { + for (i = 0; i < n; i++) { + if (EQ(ks[i], key)) { + return vs[i]; + } } } + return THE_NON_VALUE; } - return THE_NON_VALUE; + ASSERT(is_hashmap(map)); + hx = hashmap_make_hash(key); + vs = erts_hashmap_get(hx,key,map); + return vs ? *vs : THE_NON_VALUE; } #define GET_TERM(term, dest) \ @@ -6505,7 +6567,30 @@ new_map(Process* p, Eterm* reg, BeamInstr* I) Eterm *mhp,*thp; Eterm *E; BeamInstr *ptr; - map_t *mp; + flatmap_t *mp; + ErtsHeapFactory factory; + + ptr = &Arg(4); + + if (n > 2*MAP_SMALL_MAP_LIMIT) { + if (HeapWordsLeft(p) < n) { + erts_garbage_collect(p, n, reg, Arg(2)); + } + + mhp = p->htop; + thp = p->htop; + E = p->stop; + + for (i = 0; i < n/2; i++) { + GET_TERM(*ptr++, *mhp++); + GET_TERM(*ptr++, *mhp++); + } + + p->htop = mhp; + + factory.p = p; + return erts_hashmap_from_array(&factory, thp, n/2, 0); + } if (HeapWordsLeft(p) < need) { erts_garbage_collect(p, need, reg, Arg(2)); @@ -6514,11 +6599,10 @@ new_map(Process* p, Eterm* reg, BeamInstr* I) 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 = (flatmap_t *)mhp; mhp += MAP_HEADER_SIZE; mp->thing_word = MAP_HEADER; mp->size = n/2; mp->keys = keys; @@ -6528,7 +6612,7 @@ new_map(Process* p, Eterm* reg, BeamInstr* I) GET_TERM(*ptr++, *mhp++); } p->htop = mhp; - return make_map(mp); + return make_flatmap(mp); } static Eterm @@ -6538,7 +6622,7 @@ update_map_assoc(Process* p, Eterm* reg, Eterm map, BeamInstr* I) Uint num_old; Uint num_updates; Uint need; - map_t *old_mp, *mp; + flatmap_t *old_mp, *mp; Eterm res; Eterm* hp; Eterm* E; @@ -6548,12 +6632,44 @@ update_map_assoc(Process* p, Eterm* reg, Eterm map, BeamInstr* I) Eterm new_key; Eterm* kp; - if (is_not_map(map)) { - return THE_NON_VALUE; + new_p = &Arg(5); + num_updates = Arg(4) / 2; + + if (is_not_flatmap(map)) { + Uint32 hx; + Eterm val; + + /* apparently the compiler does not emit is_map instructions, + * bad compiler */ + + if (is_not_hashmap(map)) + return THE_NON_VALUE; + + res = map; + E = p->stop; + while(num_updates--) { + /* assoc can't fail */ + GET_TERM(new_p[0], new_key); + GET_TERM(new_p[1], val); + hx = hashmap_make_hash(new_key); + + res = erts_hashmap_insert(p, hx, new_key, val, res, 0); + if (p->mbuf) { + Uint live = Arg(3); + reg[live] = res; + erts_garbage_collect(p, 0, reg, live+1); + res = reg[live]; + } + + E = p->stop; + + new_p += 2; + } + return res; } - old_mp = (map_t *) map_val(map); - num_old = map_get_size(old_mp); + old_mp = (flatmap_t *) flatmap_val(map); + num_old = flatmap_get_size(old_mp); /* * If the old map is empty, create a new map. @@ -6568,14 +6684,13 @@ update_map_assoc(Process* p, Eterm* reg, Eterm map, BeamInstr* I) * 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); + old_mp = (flatmap_t *)flatmap_val(map); } /* @@ -6606,16 +6721,15 @@ update_map_assoc(Process* p, Eterm* reg, Eterm map, BeamInstr* I) kp = p->htop + 1; /* Point to first key */ hp = kp + num_old + num_updates; - res = make_map(hp); - mp = (map_t *)hp; + res = make_flatmap(hp); + mp = (flatmap_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); + old_vals = flatmap_get_values(old_mp); + old_keys = flatmap_get_keys(old_mp); - new_p = &Arg(5); GET_TERM(*new_p, new_key); n = num_updates; @@ -6701,8 +6815,19 @@ update_map_assoc(Process* p, Eterm* reg, Eterm map, BeamInstr* I) n = kp - p->htop - 1; /* Actual number of keys/values */ *p->htop = make_arityval(n); + p->htop = hp; mp->size = n; - p->htop = hp; + + /* The expensive case, need to build a hashmap */ + if (n > MAP_SMALL_MAP_LIMIT) { + res = erts_hashmap_from_ks_and_vs(p,flatmap_get_keys(mp),flatmap_get_values(mp),n); + if (p->mbuf) { + Uint live = Arg(3); + reg[live] = res; + erts_garbage_collect(p, 0, reg, live+1); + res = reg[live]; + } + } return res; } @@ -6717,7 +6842,7 @@ update_map_exact(Process* p, Eterm* reg, Eterm map, BeamInstr* I) Uint i; Uint num_old; Uint need; - map_t *old_mp, *mp; + flatmap_t *old_mp, *mp; Eterm res; Eterm* hp; Eterm* E; @@ -6726,12 +6851,48 @@ update_map_exact(Process* p, Eterm* reg, Eterm map, BeamInstr* I) BeamInstr* new_p; Eterm new_key; - if (is_not_map(map)) { - return THE_NON_VALUE; + new_p = &Arg(5); + n = Arg(4) / 2; /* Number of values to be updated */ + ASSERT(n > 0); + + if (is_not_flatmap(map)) { + Uint32 hx; + Eterm val; + + /* apparently the compiler does not emit is_map instructions, + * bad compiler */ + + if (is_not_hashmap(map)) + return THE_NON_VALUE; + + res = map; + E = p->stop; + while(n--) { + /* assoc can't fail */ + GET_TERM(new_p[0], new_key); + GET_TERM(new_p[1], val); + hx = hashmap_make_hash(new_key); + + res = erts_hashmap_insert(p, hx, new_key, val, res, 1); + if (is_non_value(res)) + return res; + + if (p->mbuf) { + Uint live = Arg(3); + reg[live] = res; + erts_garbage_collect(p, 0, reg, live+1); + res = reg[live]; + } + + E = p->stop; + + new_p += 2; + } + return res; } - old_mp = (map_t *) map_val(map); - num_old = map_get_size(old_mp); + old_mp = (flatmap_t *) flatmap_val(map); + num_old = flatmap_get_size(old_mp); /* * If the old map is empty, create a new map. @@ -6751,7 +6912,7 @@ update_map_exact(Process* p, Eterm* reg, Eterm map, BeamInstr* I) reg[live] = map; erts_garbage_collect(p, need, reg, live+1); map = reg[live]; - old_mp = (map_t *)map_val(map); + old_mp = (flatmap_t *)flatmap_val(map); } /* @@ -6761,23 +6922,20 @@ update_map_exact(Process* p, Eterm* reg, Eterm map, BeamInstr* I) hp = p->htop; E = p->stop; - old_vals = map_get_values(old_mp); - old_keys = map_get_keys(old_mp); + old_vals = flatmap_get_values(old_mp); + old_keys = flatmap_get_keys(old_mp); - res = make_map(hp); - mp = (map_t *)hp; + res = make_flatmap(hp); + mp = (flatmap_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 */ diff --git a/erts/emulator/beam/beam_load.c b/erts/emulator/beam/beam_load.c index cfc6146b0a..02689e5b19 100644 --- a/erts/emulator/beam/beam_load.c +++ b/erts/emulator/beam/beam_load.c @@ -245,7 +245,7 @@ typedef struct { /* * This structure contains all information about the module being loaded. */ - +#define MD5_SIZE 16 typedef struct LoaderState { /* * The current logical file within the binary. @@ -292,7 +292,7 @@ typedef struct LoaderState { StringPatch* string_patches; /* Linked list of position into string table to patch. */ BeamInstr catches; /* Linked list of catch_yf instructions. */ unsigned loaded_size; /* Final size of code when loaded. */ - byte mod_md5[16]; /* MD5 for module code. */ + byte mod_md5[MD5_SIZE]; /* MD5 for module code. */ int may_load_nif; /* true if NIFs may later be loaded for this module */ int on_load; /* Index in the code for the on_load function * (or 0 if there is no on_load function) @@ -528,6 +528,7 @@ static Eterm exported_from_module(Process* p, Eterm mod); static Eterm functions_in_module(Process* p, Eterm mod); static Eterm attributes_for_module(Process* p, Eterm mod); static Eterm compilation_info_for_module(Process* p, Eterm mod); +static Eterm md5_of_module(Process* p, Eterm mod); static Eterm native_addresses(Process* p, Eterm mod); int patch_funentries(Eterm Patchlist); int patch(Eterm Addresses, Uint fe); @@ -648,6 +649,7 @@ erts_prepare_loading(Binary* magic, Process *c_p, Eterm group_leader, stp->code[MI_COMPILE_PTR] = 0; stp->code[MI_COMPILE_SIZE] = 0; stp->code[MI_COMPILE_SIZE_ON_HEAP] = 0; + stp->code[MI_MD5_PTR] = 0; /* * Read the atom table. @@ -3319,9 +3321,10 @@ gen_select_tuple_arity(LoaderState* stp, GenOpArg S, GenOpArg Fail, { GenOp* op; + GenOpArg *tmp; int arity = Size.val + 3; int size = Size.val / 2; - int i; + int i, j, align = 0; /* * Verify the validity of the list. @@ -3336,9 +3339,37 @@ gen_select_tuple_arity(LoaderState* stp, GenOpArg S, GenOpArg Fail, } /* + * Use a special-cased instruction if there are only two values. + */ + if (size == 2) { + NEW_GENOP(stp, op); + op->next = NULL; + op->op = genop_i_select_tuple_arity2_6; + GENOP_ARITY(op, arity - 1); + op->a[0] = S; + op->a[1] = Fail; + op->a[2].type = TAG_u; + op->a[2].val = Rest[0].val; + op->a[3].type = TAG_u; + op->a[3].val = Rest[2].val; + op->a[4] = Rest[1]; + op->a[5] = Rest[3]; + + return op; + } + + /* * Generate the generic instruction. + * Assumption: + * Few different tuple arities to select on (fewer than 20). + * Use linear scan approach. */ + align = 1; + + arity += 2*align; + size += align; + NEW_GENOP(stp, op); op->next = NULL; op->op = genop_i_select_tuple_arity_3; @@ -3346,39 +3377,36 @@ gen_select_tuple_arity(LoaderState* stp, GenOpArg S, GenOpArg Fail, op->a[0] = S; op->a[1] = Fail; op->a[2].type = TAG_u; - op->a[2].val = Size.val / 2; - for (i = 0; i < Size.val; i += 2) { - op->a[i+3].type = TAG_v; - op->a[i+3].val = make_arityval(Rest[i].val); - op->a[i+4] = Rest[i+1]; - } + op->a[2].val = size; - /* - * Sort the values to make them useful for a binary search. - */ + tmp = (GenOpArg *) erts_alloc(ERTS_ALC_T_LOADER_TMP, sizeof(GenOpArg)*(arity-2*align)); - qsort(op->a+3, size, 2*sizeof(GenOpArg), - (int (*)(const void *, const void *)) genopargcompare); -#ifdef DEBUG - for (i = 3; i < arity-2; i += 2) { - ASSERT(op->a[i].val < op->a[i+2].val); + for (i = 3; i < arity - 2*align; i+=2) { + tmp[i-3].type = TAG_v; + tmp[i-3].val = make_arityval(Rest[i-3].val); + tmp[i-2] = Rest[i-2]; } -#endif /* - * Use a special-cased instruction if there are only two values. + * Sort the values to make them useful for a sentinel search */ - if (size == 2) { - op->op = genop_i_select_tuple_arity2_6; - op->arity--; - op->a[2].type = TAG_u; - op->a[2].val = arityval(op->a[3].val); - op->a[3] = op->a[4]; - op->a[4].type = TAG_u; - op->a[4].val = arityval(op->a[5].val); - op->a[5] = op->a[6]; + + qsort(tmp, size - align, 2*sizeof(GenOpArg), + (int (*)(const void *, const void *)) genopargcompare); + + j = 3; + for (i = 3; i < arity - 2*align; i += 2) { + op->a[j] = tmp[i-3]; + op->a[j + size] = tmp[i-2]; + j++; } + erts_free(ERTS_ALC_T_LOADER_TMP, (void *) tmp); + + op->a[j].type = TAG_u; + op->a[j].val = ~((BeamInstr)0); + op->a[j+size] = Fail; + return op; } @@ -3600,45 +3628,109 @@ gen_select_val(LoaderState* stp, GenOpArg S, GenOpArg Fail, GenOpArg Size, GenOpArg* Rest) { GenOp* op; + GenOpArg *tmp; int arity = Size.val + 3; int size = Size.val / 2; - int i; + int i, j, align = 0; + + if (size == 2) { + + /* + * Use a special-cased instruction if there are only two values. + */ + + NEW_GENOP(stp, op); + op->next = NULL; + op->op = genop_i_select_val2_6; + GENOP_ARITY(op, arity - 1); + op->a[0] = S; + op->a[1] = Fail; + op->a[2] = Rest[0]; + op->a[3] = Rest[2]; + op->a[4] = Rest[1]; + op->a[5] = Rest[3]; + + return op; + + } else if (size > 10) { + + /* binary search instruction */ + + NEW_GENOP(stp, op); + op->next = NULL; + op->op = genop_i_select_val_bins_3; + GENOP_ARITY(op, arity); + op->a[0] = S; + op->a[1] = Fail; + op->a[2].type = TAG_u; + op->a[2].val = size; + for (i = 3; i < arity; i++) { + op->a[i] = Rest[i-3]; + } + + /* + * Sort the values to make them useful for a binary search. + */ + + qsort(op->a+3, size, 2*sizeof(GenOpArg), + (int (*)(const void *, const void *)) genopargcompare); +#ifdef DEBUG + for (i = 3; i < arity-2; i += 2) { + ASSERT(op->a[i].val < op->a[i+2].val); + } +#endif + return op; + } + + /* linear search instruction */ + + align = 1; + + arity += 2*align; + size += align; NEW_GENOP(stp, op); op->next = NULL; - op->op = genop_i_select_val_3; + op->op = genop_i_select_val_lins_3; GENOP_ARITY(op, arity); op->a[0] = S; op->a[1] = Fail; op->a[2].type = TAG_u; op->a[2].val = size; - for (i = 3; i < arity; i++) { - op->a[i] = Rest[i-3]; + + tmp = (GenOpArg *) erts_alloc(ERTS_ALC_T_LOADER_TMP, sizeof(GenOpArg)*(arity-2*align)); + + for (i = 3; i < arity - 2*align; i++) { + tmp[i-3] = Rest[i-3]; } /* - * Sort the values to make them useful for a binary search. + * Sort the values to make them useful for a sentinel search */ - qsort(op->a+3, size, 2*sizeof(GenOpArg), - (int (*)(const void *, const void *)) genopargcompare); -#ifdef DEBUG - for (i = 3; i < arity-2; i += 2) { - ASSERT(op->a[i].val < op->a[i+2].val); + qsort(tmp, size - align, 2*sizeof(GenOpArg), + (int (*)(const void *, const void *)) genopargcompare); + + j = 3; + for (i = 3; i < arity - 2*align; i += 2) { + op->a[j] = tmp[i-3]; + op->a[j+size] = tmp[i-2]; + j++; } -#endif - /* - * Use a special-cased instruction if there are only two values. - */ - if (size == 2) { - op->op = genop_i_select_val2_6; - op->arity--; - op->a[2] = op->a[3]; - op->a[3] = op->a[4]; - op->a[4] = op->a[5]; - op->a[5] = op->a[6]; + erts_free(ERTS_ALC_T_LOADER_TMP, (void *) tmp); + + /* add sentinel */ + + op->a[j].type = TAG_u; + op->a[j].val = ~((BeamInstr)0); + op->a[j+size] = Fail; + +#ifdef DEBUG + for (i = 0; i < size - 1; i++) { + ASSERT(op->a[i+3].val <= op->a[i+4].val); } +#endif return op; } @@ -4042,7 +4134,7 @@ freeze_code(LoaderState* stp) } size = (stp->ci * sizeof(BeamInstr)) + (stp->total_literal_size * sizeof(Eterm)) + - strtab_size + attr_size + compile_size + line_size; + strtab_size + attr_size + compile_size + MD5_SIZE + line_size; /* * Move the code to its final location. @@ -4251,11 +4343,20 @@ freeze_code(LoaderState* stp) code[MI_COMPILE_SIZE_ON_HEAP] = decoded_size; } CHKBLK(ERTS_ALC_T_CODE,code); + { + byte* md5_sum = str_table + strtab_size + attr_size + compile_size; + CHKBLK(ERTS_ALC_T_CODE,code); + sys_memcpy(md5_sum, stp->mod_md5, MD5_SIZE); + CHKBLK(ERTS_ALC_T_CODE,code); + code[MI_MD5_PTR] = (BeamInstr) md5_sum; + CHKBLK(ERTS_ALC_T_CODE,code); + } + CHKBLK(ERTS_ALC_T_CODE,code); /* * Make sure that we have not overflowed the allocated code space. */ - ASSERT(str_table + strtab_size + attr_size + compile_size == + ASSERT(str_table + strtab_size + attr_size + compile_size + MD5_SIZE == ((byte *) code) + size); /* @@ -4971,7 +5072,8 @@ get_tag_and_value(LoaderState* stp, Uint len_code, arity = count/sizeof(Eterm); *result = new_literal(stp, &hp, arity+1); - (void) bytes_to_big(bigbuf, count, neg, hp); + if (is_nil(bytes_to_big(bigbuf, count, neg, hp))) + goto load_error; if (bigbuf != default_buf) { erts_free(ERTS_ALC_T_LOADER_TMP, (void *) bigbuf); @@ -5107,10 +5209,11 @@ erts_module_info_0(Process* p, Eterm module) hp += 3; \ list = CONS(hp, tup, list) + BUILD_INFO(am_md5); BUILD_INFO(am_compile); BUILD_INFO(am_attributes); - BUILD_INFO(am_imports); BUILD_INFO(am_exports); + BUILD_INFO(am_module); #undef BUILD_INFO return list; } @@ -5120,8 +5223,8 @@ erts_module_info_1(Process* p, Eterm module, Eterm what) { if (what == am_module) { return module; - } else if (what == am_imports) { - return NIL; + } else if (what == am_md5) { + return md5_of_module(p, module); } else if (what == am_exports) { return exported_from_module(p, module); } else if (what == am_functions) { @@ -5310,7 +5413,7 @@ attributes_for_module(Process* p, /* Process whose heap to use. */ Eterm result = NIL; Eterm* end; - if (is_not_atom(mod) || (is_not_list(result) && is_not_nil(result))) { + if (is_not_atom(mod)) { return THE_NON_VALUE; } @@ -5349,7 +5452,7 @@ compilation_info_for_module(Process* p, /* Process whose heap to use. */ Eterm result = NIL; Eterm* end; - if (is_not_atom(mod) || (is_not_list(result) && is_not_nil(result))) { + if (is_not_atom(mod)) { return THE_NON_VALUE; } @@ -5372,6 +5475,33 @@ compilation_info_for_module(Process* p, /* Process whose heap to use. */ } /* + * Returns the MD5 checksum for a module + * + * Returns a tagged term, or 0 on error. + */ + +Eterm +md5_of_module(Process* p, /* Process whose heap to use. */ + Eterm mod) /* Tagged atom for module. */ +{ + Module* modp; + BeamInstr* code; + Eterm res = NIL; + + if (is_not_atom(mod)) { + return THE_NON_VALUE; + } + + modp = erts_get_module(mod, erts_active_code_ix()); + if (modp == NULL) { + return THE_NON_VALUE; + } + code = modp->curr.code; + res = new_binary(p, (byte *) code[MI_MD5_PTR], MD5_SIZE); + return res; +} + +/* * Build a single {M,F,A,Loction} item to be part of * a stack trace. */ @@ -5547,7 +5677,7 @@ code_module_md5_1(BIF_ALIST_1) res = am_undefined; goto done; } - res = new_binary(p, stp->mod_md5, sizeof(stp->mod_md5)); + res = new_binary(p, stp->mod_md5, MD5_SIZE); done: erts_free_aligned_binary_bytes(temp_alloc); @@ -5943,6 +6073,7 @@ erts_make_stub_module(Process* p, Eterm Mod, Eterm Beam, Eterm Info) code[MI_LITERALS_END] = 0; code[MI_LITERALS_OFF_HEAP] = 0; code[MI_ON_LOAD_FUNCTION_PTR] = 0; + code[MI_MD5_PTR] = 0; ci = MI_FUNCTIONS + n + 1; /* diff --git a/erts/emulator/beam/beam_load.h b/erts/emulator/beam/beam_load.h index bd22b0c4de..0e3ca0bdb0 100644 --- a/erts/emulator/beam/beam_load.h +++ b/erts/emulator/beam/beam_load.h @@ -91,7 +91,6 @@ extern Uint erts_total_code_size; #define MI_LITERALS_END 8 #define MI_LITERALS_OFF_HEAP 9 - /* * Pointer to the on_load function (or NULL if none). */ @@ -103,6 +102,11 @@ extern Uint erts_total_code_size; #define MI_LINE_TABLE 11 /* + * Pointer to the module MD5 sum (16 bytes) + */ +#define MI_MD5_PTR 12 + +/* * Start of function pointer table. This table contains pointers to * all functions in the module plus an additional pointer just beyond * the end of the last function. @@ -111,7 +115,7 @@ extern Uint erts_total_code_size; * this table. */ -#define MI_FUNCTIONS 12 +#define MI_FUNCTIONS 13 /* * Layout of the line table. diff --git a/erts/emulator/beam/benchmark.c b/erts/emulator/beam/benchmark.c index 8613131176..b16fe6b271 100644 --- a/erts/emulator/beam/benchmark.c +++ b/erts/emulator/beam/benchmark.c @@ -37,37 +37,9 @@ unsigned long long major_gc; #ifdef BM_TIMERS -#if (defined(__i386__) || defined(__x86_64__)) && USE_PERFCTR - -#include "libperfctr.h" -struct vperfctr *system_clock; -double cpu_khz; -BM_NEW_TIMER(start); - -static double get_hrvtime(void) -{ - unsigned long long ticks; - double milli_seconds; - - ticks = vperfctr_read_tsc(system_clock); - milli_seconds = (double)ticks / cpu_khz; - return milli_seconds; -} - -static void stop_hrvtime(void) -{ - if(system_clock) - { - vperfctr_stop(system_clock); - vperfctr_close(system_clock); - system_clock = NULL; - } -} - -#else /* not perfctr, asuming Solaris */ +/* assuming Solaris */ #include <time.h> BM_TIMER_T system_clock; -#endif unsigned long local_pause_times[MAX_PAUSE_TIME]; unsigned long pause_times[MAX_PAUSE_TIME]; @@ -117,40 +89,6 @@ unsigned long long message_sizes[1000]; void init_benchmarking() { #ifdef BM_TIMERS -#if (defined(__i386__) || defined(__x86_64__)) && USE_PERFCTR - /* pass `--with-perfctr=/path/to/perfctr' when configuring */ - struct perfctr_info info; - struct vperfctr_control control; - int i; - - system_clock = vperfctr_open(); - if (system_clock != NULL) - { - if (vperfctr_info(system_clock,&info) >= 0) - { - cpu_khz = (double)info.cpu_khz; - if (info.cpu_features & PERFCTR_FEATURE_RDTSC) - { - memset(&control,0,sizeof control); - control.cpu_control.tsc_on = 1; - } - } - if (vperfctr_control(system_clock,&control) < 0) - { - vperfctr_close(system_clock); - system_clock = NULL; - } - } - - for (i = 0; i < 1000; i++) - { - BM_START_TIMER(system); - BM_STOP_TIMER(system); - } - - timer_time = system_time / 1000; - start_time = 0; -#else int i; for (i = 0; i < 1000; i++) { @@ -158,7 +96,6 @@ void init_benchmarking() BM_STOP_TIMER(system); } timer_time = system_time / 1000; -#endif for (i = 0; i < MAX_PAUSE_TIME; i++) { local_pause_times[i] = 0; diff --git a/erts/emulator/beam/benchmark.h b/erts/emulator/beam/benchmark.h index 766edaac42..904564a96b 100644 --- a/erts/emulator/beam/benchmark.h +++ b/erts/emulator/beam/benchmark.h @@ -37,10 +37,7 @@ /* BM_TIMERS keeps track of the time spent in diferent parts of the * system. It only measures accual active time, not time spent in idle - * mode. These timers requires hardware support. For Linux, use the - * package perfctr from user.it.uu.se/~mikpe/linux/perfctr. If this - * package is not specified when configuring the system - * (--with-perfctr=PATH), the Solaris hrtime_t will be used. + * mode. Currently, the Solaris hrtime_t will be used. * To add new timers look below. */ #define BM_TIMERS @@ -142,38 +139,7 @@ extern unsigned long long major_gc; * meassure (send time in shared heap for instance). */ -#if (defined(__i386__) || defined(__x86_64__)) && USE_PERFCTR -#include "libperfctr.h" - -#define BM_TIMER_T double - -extern struct vperfctr *system_clock; -extern double cpu_khz; -extern BM_TIMER_T start_time; - -#define BM_START_TIMER(t) start_time = \ - (BM_TIMER_T)vperfctr_read_tsc(system_clock) / \ - cpu_khz; - -#define BM_STOP_TIMER(t) do { \ - BM_TIMER_T tmp = ((BM_TIMER_T)vperfctr_read_tsc(system_clock) / cpu_khz); \ - tmp -= (start_time + timer_time); \ - t##_time += (tmp > 0 ? tmp : 0); \ -} while(0) - -#define BM_TIME_PRINTER(str,time) do { \ - int min,sec,milli,micro; \ - BM_TIMER_T tmp = (time) * 1000; \ - micro = (uint)(tmp - ((int)(tmp / 1000)) * 1000); \ - tmp /= 1000; \ - milli = (uint)(tmp - ((int)(tmp / 1000)) * 1000); \ - tmp /= 1000; \ - sec = (uint)(tmp - ((int)(tmp / 60)) * 60); \ - min = (uint)tmp / 60; \ - erts_fprintf(file,str": %d:%02d.%03d %03d\n",min,sec,milli,micro); \ -} while(0) - -#else /* !USE_PERFCTR (Assuming Solaris) */ +/* (Assuming Solaris) */ #define BM_TIMER_T hrtime_t #define BM_START_TIMER(t) system_clock = sys_gethrtime() @@ -196,7 +162,6 @@ extern BM_TIMER_T start_time; } while(0) extern BM_TIMER_T system_clock; -#endif /* USE_PERFCTR */ extern BM_TIMER_T timer_time; extern BM_TIMER_T system_time; diff --git a/erts/emulator/beam/bif.c b/erts/emulator/beam/bif.c index 42dd160e38..49996e7f0b 100644 --- a/erts/emulator/beam/bif.c +++ b/erts/emulator/beam/bif.c @@ -28,7 +28,9 @@ #include "global.h" #include "erl_process.h" #include "error.h" +#define ERL_WANT_HIPE_BIF_WRAPPER__ #include "bif.h" +#undef ERL_WANT_HIPE_BIF_WRAPPER__ #include "big.h" #include "dist.h" #include "erl_version.h" @@ -46,7 +48,7 @@ static Export* set_cpu_topology_trap = NULL; static Export* await_proc_exit_trap = NULL; static Export* await_port_send_result_trap = NULL; Export* erts_format_cpu_topology_trap = NULL; - +static Export dsend_continue_trap_export; static Export *await_sched_wall_time_mod_trap; static erts_smp_atomic32_t sched_wall_time; @@ -1777,6 +1779,8 @@ BIF_RETTYPE whereis_1(BIF_ALIST_1) * erlang:'!'/2 */ +HIPE_WRAPPER_BIF_DISABLE_GC(ebif_bang, 2) + BIF_RETTYPE ebif_bang_2(BIF_ALIST_2) { @@ -1795,34 +1799,36 @@ ebif_bang_2(BIF_ALIST_2) #define SEND_USER_ERROR (-5) #define SEND_INTERNAL_ERROR (-6) #define SEND_AWAIT_RESULT (-7) +#define SEND_YIELD_CONTINUE (-8) + -Sint do_send(Process *p, Eterm to, Eterm msg, int suspend, Eterm *refp); +Sint do_send(Process *p, Eterm to, Eterm msg, Eterm *refp, ErtsSendContext*); static Sint remote_send(Process *p, DistEntry *dep, - Eterm to, Eterm full_to, Eterm msg, int suspend) + Eterm to, Eterm full_to, Eterm msg, + ErtsSendContext* ctx) { Sint res; int code; - ErtsDSigData dsd; ASSERT(is_atom(to) || is_external_pid(to)); - code = erts_dsig_prepare(&dsd, dep, p, ERTS_DSP_NO_LOCK, !suspend); + code = erts_dsig_prepare(&ctx->dsd, dep, p, ERTS_DSP_NO_LOCK, !ctx->suspend); switch (code) { case ERTS_DSIG_PREP_NOT_ALIVE: case ERTS_DSIG_PREP_NOT_CONNECTED: res = SEND_TRAP; break; case ERTS_DSIG_PREP_WOULD_SUSPEND: - ASSERT(!suspend); + ASSERT(!ctx->suspend); res = SEND_YIELD; break; case ERTS_DSIG_PREP_CONNECTED: { if (is_atom(to)) - code = erts_dsig_send_reg_msg(&dsd, to, msg); + code = erts_dsig_send_reg_msg(to, msg, ctx); else - code = erts_dsig_send_msg(&dsd, to, msg); + code = erts_dsig_send_msg(to, msg, ctx); /* * Note that reductions have been bumped on calling * process by erts_dsig_send_reg_msg() or @@ -1830,6 +1836,8 @@ static Sint remote_send(Process *p, DistEntry *dep, */ if (code == ERTS_DSIG_SEND_YIELD) res = SEND_YIELD_RETURN; + else if (code == ERTS_DSIG_SEND_CONTINUE) + res = SEND_YIELD_CONTINUE; else res = 0; break; @@ -1850,7 +1858,8 @@ static Sint remote_send(Process *p, DistEntry *dep, } Sint -do_send(Process *p, Eterm to, Eterm msg, int suspend, Eterm *refp) { +do_send(Process *p, Eterm to, Eterm msg, Eterm *refp, ErtsSendContext* ctx) +{ Eterm portid; Port *pt; Process* rp; @@ -1883,7 +1892,7 @@ do_send(Process *p, Eterm to, Eterm msg, int suspend, Eterm *refp) { #endif return 0; } - return remote_send(p, dep, to, to, msg, suspend); + return remote_send(p, dep, to, to, msg, ctx); } else if (is_atom(to)) { Eterm id = erts_whereis_name_to_id(p, to); @@ -1940,7 +1949,7 @@ do_send(Process *p, Eterm to, Eterm msg, int suspend, Eterm *refp) { ret_val = 0; if (pt) { - int ps_flags = suspend ? 0 : ERTS_PORT_SIG_FLG_NOSUSPEND; + int ps_flags = ctx->suspend ? 0 : ERTS_PORT_SIG_FLG_NOSUSPEND; *refp = NIL; switch (erts_port_command(p, ps_flags, pt, msg, refp)) { @@ -1949,12 +1958,12 @@ do_send(Process *p, Eterm to, Eterm msg, int suspend, Eterm *refp) { return SEND_USER_ERROR; case ERTS_PORT_OP_BUSY: /* Nothing has been sent */ - if (suspend) + if (ctx->suspend) erts_suspend(p, ERTS_PROC_LOCK_MAIN, pt); return SEND_YIELD; case ERTS_PORT_OP_BUSY_SCHEDULED: /* Message was sent */ - if (suspend) { + if (ctx->suspend) { erts_suspend(p, ERTS_PROC_LOCK_MAIN, pt); ret_val = SEND_YIELD_RETURN; break; @@ -2034,9 +2043,14 @@ do_send(Process *p, Eterm to, Eterm msg, int suspend, Eterm *refp) { return 0; } - ret = remote_send(p, dep, tp[1], to, msg, suspend); - if (dep) - erts_deref_dist_entry(dep); + ret = remote_send(p, dep, tp[1], to, msg, ctx); + if (ret != SEND_YIELD_CONTINUE) { + if (dep) { + erts_deref_dist_entry(dep); + } + } else { + ctx->dep_to_deref = dep; + } return ret; } else { if (IS_TRACED(p)) /* XXX Is this really neccessary ??? */ @@ -2067,9 +2081,11 @@ do_send(Process *p, Eterm to, Eterm msg, int suspend, Eterm *refp) { } } +HIPE_WRAPPER_BIF_DISABLE_GC(send, 3) BIF_RETTYPE send_3(BIF_ALIST_3) { + BIF_RETTYPE retval; Eterm ref; Process *p = BIF_P; Eterm to = BIF_ARG_1; @@ -2077,34 +2093,44 @@ BIF_RETTYPE send_3(BIF_ALIST_3) Eterm opts = BIF_ARG_3; int connect = !0; - int suspend = !0; Eterm l = opts; Sint result; - + DeclareTypedTmpHeap(ErtsSendContext, ctx, BIF_P); + UseTmpHeap(sizeof(ErtsSendContext)/sizeof(Eterm), BIF_P); + + ctx->suspend = !0; + ctx->dep_to_deref = NULL; + ctx->return_term = am_ok; + ctx->dss.reds = (Sint) (ERTS_BIF_REDS_LEFT(p) * TERM_TO_BINARY_LOOP_FACTOR); + ctx->dss.phase = ERTS_DSIG_SEND_PHASE_INIT; + while (is_list(l)) { if (CAR(list_val(l)) == am_noconnect) { connect = 0; } else if (CAR(list_val(l)) == am_nosuspend) { - suspend = 0; + ctx->suspend = 0; } else { - BIF_ERROR(p, BADARG); + ERTS_BIF_PREP_ERROR(retval, p, BADARG); + goto done; } l = CDR(list_val(l)); } if(!is_nil(l)) { - BIF_ERROR(p, BADARG); + ERTS_BIF_PREP_ERROR(retval, p, BADARG); + goto done; } #ifdef DEBUG ref = NIL; #endif - result = do_send(p, to, msg, suspend, &ref); + result = do_send(p, to, msg, &ref, ctx); if (result > 0) { ERTS_VBUMP_REDS(p, result); if (ERTS_IS_PROC_OUT_OF_REDS(p)) goto yield_return; - BIF_RET(am_ok); + ERTS_BIF_PREP_RET(retval, am_ok); + goto done; } switch (result) { @@ -2112,68 +2138,127 @@ BIF_RETTYPE send_3(BIF_ALIST_3) /* May need to yield even though we do not bump reds here... */ if (ERTS_IS_PROC_OUT_OF_REDS(p)) goto yield_return; - BIF_RET(am_ok); + ERTS_BIF_PREP_RET(retval, am_ok); break; case SEND_TRAP: if (connect) { - BIF_TRAP3(dsend3_trap, p, to, msg, opts); + ERTS_BIF_PREP_TRAP3(retval, dsend3_trap, p, to, msg, opts); } else { - BIF_RET(am_noconnect); + ERTS_BIF_PREP_RET(retval, am_noconnect); } break; case SEND_YIELD: - if (suspend) { - ERTS_BIF_YIELD3(bif_export[BIF_send_3], p, to, msg, opts); + if (ctx->suspend) { + ERTS_BIF_PREP_YIELD3(retval, + bif_export[BIF_send_3], p, to, msg, opts); } else { - BIF_RET(am_nosuspend); + ERTS_BIF_PREP_RET(retval, am_nosuspend); } break; case SEND_YIELD_RETURN: - if (!suspend) - BIF_RET(am_nosuspend); + if (!ctx->suspend) { + ERTS_BIF_PREP_RET(retval, am_nosuspend); + break; + } yield_return: - ERTS_BIF_YIELD_RETURN(p, am_ok); + ERTS_BIF_PREP_YIELD_RETURN(retval, p, am_ok); + break; case SEND_AWAIT_RESULT: ASSERT(is_internal_ref(ref)); - BIF_TRAP3(await_port_send_result_trap, p, ref, am_nosuspend, am_ok); + ERTS_BIF_PREP_TRAP3(retval, await_port_send_result_trap, p, ref, am_nosuspend, am_ok); + break; case SEND_BADARG: - BIF_ERROR(p, BADARG); + ERTS_BIF_PREP_ERROR(retval, p, BADARG); break; case SEND_USER_ERROR: - BIF_ERROR(p, EXC_ERROR); + ERTS_BIF_PREP_ERROR(retval, p, EXC_ERROR); break; case SEND_INTERNAL_ERROR: - BIF_ERROR(p, EXC_INTERNAL_ERROR); + ERTS_BIF_PREP_ERROR(retval, p, EXC_INTERNAL_ERROR); + break; + case SEND_YIELD_CONTINUE: + BUMP_ALL_REDS(p); + erts_set_gc_state(p, 0); + ERTS_BIF_PREP_TRAP1(retval, &dsend_continue_trap_export, p, + erts_dsend_export_trap_context(p, ctx)); break; default: - ASSERT(! "Illegal send result"); + erl_exit(ERTS_ABORT_EXIT, "send_3 invalid result %d\n", (int)result); break; } - ASSERT(! "Can not arrive here"); - BIF_ERROR(p, BADARG); + +done: + UnUseTmpHeap(sizeof(ErtsSendContext)/sizeof(Eterm), BIF_P); + return retval; } +HIPE_WRAPPER_BIF_DISABLE_GC(send, 2) + BIF_RETTYPE send_2(BIF_ALIST_2) { return erl_send(BIF_P, BIF_ARG_1, BIF_ARG_2); } +static BIF_RETTYPE dsend_continue_trap_1(BIF_ALIST_1) +{ + Binary* bin = ((ProcBin*) binary_val(BIF_ARG_1))->val; + ErtsSendContext* ctx = (ErtsSendContext*) ERTS_MAGIC_BIN_DATA(bin); + Sint initial_reds = (Sint) (ERTS_BIF_REDS_LEFT(BIF_P) * TERM_TO_BINARY_LOOP_FACTOR); + int result; + + ASSERT(ERTS_MAGIC_BIN_DESTRUCTOR(bin) == erts_dsend_context_dtor); + + ctx->dss.reds = initial_reds; + result = erts_dsig_send(&ctx->dsd, &ctx->dss); + + switch (result) { + case ERTS_DSIG_SEND_OK: + erts_set_gc_state(BIF_P, 1); + BIF_RET(ctx->return_term); + break; + case ERTS_DSIG_SEND_YIELD: /*SEND_YIELD_RETURN*/ + erts_set_gc_state(BIF_P, 1); + if (!ctx->suspend) + BIF_RET(am_nosuspend); + ERTS_BIF_YIELD_RETURN(BIF_P, ctx->return_term); + + case ERTS_DSIG_SEND_CONTINUE: { /*SEND_YIELD_CONTINUE*/ + BUMP_ALL_REDS(BIF_P); + BIF_TRAP1(&dsend_continue_trap_export, BIF_P, BIF_ARG_1); + } + default: + erl_exit(ERTS_ABORT_EXIT, "dsend_continue_trap invalid result %d\n", (int)result); + break; + } + ASSERT(! "Can not arrive here"); + BIF_ERROR(BIF_P, BADARG); +} + Eterm erl_send(Process *p, Eterm to, Eterm msg) { + Eterm retval; Eterm ref; Sint result; + DeclareTypedTmpHeap(ErtsSendContext, ctx, p); + UseTmpHeap(sizeof(ErtsSendContext)/sizeof(Eterm), p); #ifdef DEBUG ref = NIL; #endif + ctx->suspend = !0; + ctx->dep_to_deref = NULL; + ctx->return_term = msg; + ctx->dss.reds = (Sint) (ERTS_BIF_REDS_LEFT(p) * TERM_TO_BINARY_LOOP_FACTOR); + ctx->dss.phase = ERTS_DSIG_SEND_PHASE_INIT; - result = do_send(p, to, msg, !0, &ref); + result = do_send(p, to, msg, &ref, ctx); if (result > 0) { ERTS_VBUMP_REDS(p, result); if (ERTS_IS_PROC_OUT_OF_REDS(p)) goto yield_return; - BIF_RET(msg); + ERTS_BIF_PREP_RET(retval, msg); + goto done; } switch (result) { @@ -2181,35 +2266,46 @@ Eterm erl_send(Process *p, Eterm to, Eterm msg) /* May need to yield even though we do not bump reds here... */ if (ERTS_IS_PROC_OUT_OF_REDS(p)) goto yield_return; - BIF_RET(msg); + ERTS_BIF_PREP_RET(retval, msg); break; case SEND_TRAP: - BIF_TRAP2(dsend2_trap, p, to, msg); + ERTS_BIF_PREP_TRAP2(retval, dsend2_trap, p, to, msg); break; case SEND_YIELD: - ERTS_BIF_YIELD2(bif_export[BIF_send_2], p, to, msg); + ERTS_BIF_PREP_YIELD2(retval, bif_export[BIF_send_2], p, to, msg); break; case SEND_YIELD_RETURN: yield_return: - ERTS_BIF_YIELD_RETURN(p, msg); + ERTS_BIF_PREP_YIELD_RETURN(retval, p, msg); + break; case SEND_AWAIT_RESULT: ASSERT(is_internal_ref(ref)); - BIF_TRAP3(await_port_send_result_trap, p, ref, msg, msg); + ERTS_BIF_PREP_TRAP3(retval, + await_port_send_result_trap, p, ref, msg, msg); + break; case SEND_BADARG: - BIF_ERROR(p, BADARG); + ERTS_BIF_PREP_ERROR(retval, p, BADARG); break; case SEND_USER_ERROR: - BIF_ERROR(p, EXC_ERROR); + ERTS_BIF_PREP_ERROR(retval, p, EXC_ERROR); break; case SEND_INTERNAL_ERROR: - BIF_ERROR(p, EXC_INTERNAL_ERROR); + ERTS_BIF_PREP_ERROR(retval, p, EXC_INTERNAL_ERROR); + break; + case SEND_YIELD_CONTINUE: + BUMP_ALL_REDS(p); + erts_set_gc_state(p, 0); + ERTS_BIF_PREP_TRAP1(retval, &dsend_continue_trap_export, p, + erts_dsend_export_trap_context(p, ctx)); break; default: - ASSERT(! "Illegal send result"); + erl_exit(ERTS_ABORT_EXIT, "invalid send result %d\n", (int)result); break; } - ASSERT(! "Can not arrive here"); - BIF_ERROR(p, BADARG); + +done: + UnUseTmpHeap(sizeof(ErtsSendContext)/sizeof(Eterm), p); + return retval; } /**********************************************************************/ @@ -3996,16 +4092,19 @@ BIF_RETTYPE halt_2(BIF_ALIST_2) BIF_RETTYPE function_exported_3(BIF_ALIST_3) { + int arity; if (is_not_atom(BIF_ARG_1) || is_not_atom(BIF_ARG_2) || is_not_small(BIF_ARG_3)) { BIF_ERROR(BIF_P, BADARG); } - if (erts_find_function(BIF_ARG_1, BIF_ARG_2, signed_val(BIF_ARG_3), - erts_active_code_ix()) == NULL) { - BIF_RET(am_false); + arity = signed_val(BIF_ARG_3); + if (erts_find_function(BIF_ARG_1, BIF_ARG_2, arity, + erts_active_code_ix()) != NULL || + erts_is_builtin(BIF_ARG_1, BIF_ARG_2, arity)) { + BIF_RET(am_true); } - BIF_RET(am_true); + BIF_RET(am_false); } /**********************************************************************/ @@ -4816,6 +4915,10 @@ void erts_init_bif(void) #endif , &bif_return_trap); + erts_init_trap_export(&dsend_continue_trap_export, + am_erts_internal, am_dsend_continue_trap, 1, + dsend_continue_trap_1); + flush_monitor_message_trap = erts_export_put(am_erlang, am_flush_monitor_message, 2); diff --git a/erts/emulator/beam/bif.h b/erts/emulator/beam/bif.h index 72c55ccb55..837cb017ac 100644 --- a/erts/emulator/beam/bif.h +++ b/erts/emulator/beam/bif.h @@ -30,10 +30,12 @@ extern Export* erts_format_cpu_topology_trap; #define BIF_ALIST_1 Process* A__p, Eterm* BIF__ARGS #define BIF_ALIST_2 Process* A__p, Eterm* BIF__ARGS #define BIF_ALIST_3 Process* A__p, Eterm* BIF__ARGS +#define BIF_ALIST_4 Process* A__p, Eterm* BIF__ARGS #define BIF_ARG_1 (BIF__ARGS[0]) #define BIF_ARG_2 (BIF__ARGS[1]) #define BIF_ARG_3 (BIF__ARGS[2]) +#define BIF_ARG_4 (BIF__ARGS[3]) #define ERTS_IS_PROC_OUT_OF_REDS(p) \ ((p)->fcalls > 0 \ @@ -465,6 +467,8 @@ erts_bif_prep_await_proc_exit_apply_trap(Process *c_p, Eterm args[], int nargs); +#ifdef ERL_WANT_HIPE_BIF_WRAPPER__ + #ifndef HIPE #define HIPE_WRAPPER_BIF_DISABLE_GC(BIF_NAME, ARITY) @@ -509,6 +513,7 @@ BIF_RETTYPE hipe_wrapper_ ## BIF_NAME ## _ ## ARITY (Process* c_p, \ #endif +#endif /* ERL_WANT_HIPE_BIF_WRAPPER__ */ #include "erl_bif_table.h" diff --git a/erts/emulator/beam/bif.tab b/erts/emulator/beam/bif.tab index e68b8e6274..5a0ee2ffc3 100644 --- a/erts/emulator/beam/bif.tab +++ b/erts/emulator/beam/bif.tab @@ -157,6 +157,8 @@ bif erts_internal:request_system_task/3 bif erts_internal:check_process_code/2 bif erts_internal:map_to_tuple_keys/1 +bif erts_internal:map_type/1 +bif erts_internal:map_hashmap_children/1 # inet_db support bif erlang:port_set_data/2 @@ -194,6 +196,7 @@ bif math:erf/1 bif math:erfc/1 bif math:exp/1 bif math:log/1 +bif math:log2/1 bif math:log10/1 bif math:sqrt/1 bif math:atan2/2 @@ -578,7 +581,7 @@ bif io:printable_range/0 bif os:unsetenv/1 # -# New in R17A +# New in 17.0 # bif re:inspect/2 @@ -600,10 +603,21 @@ bif maps:values/1 bif erts_internal:cmp_term/2 +bif ets:take/2 + # -# New in 17.1. +# New in 17.1 # + bif erlang:fun_info_mfa/1 + +# New in 18.0 +# + +bif erlang:get_keys/0 +bif ets:update_counter/4 +bif erts_debug:map_info/1 + # # Obsolete # diff --git a/erts/emulator/beam/big.c b/erts/emulator/beam/big.c index de7d370938..d1e46e3063 100644 --- a/erts/emulator/beam/big.c +++ b/erts/emulator/beam/big.c @@ -1900,6 +1900,8 @@ Eterm bytes_to_big(byte *xp, dsize_t xsz, int xsgn, Eterm *r) *rwp = d; rwp++; } + if (rsz > BIG_ARITY_MAX) + return NIL; if (xsgn) { *r = make_neg_bignum_header(rsz); } diff --git a/erts/emulator/beam/binary.c b/erts/emulator/beam/binary.c index f50d484576..cc0b3b9b6c 100644 --- a/erts/emulator/beam/binary.c +++ b/erts/emulator/beam/binary.c @@ -26,7 +26,9 @@ #include "global.h" #include "erl_process.h" #include "error.h" +#define ERL_WANT_HIPE_BIF_WRAPPER__ #include "bif.h" +#undef ERL_WANT_HIPE_BIF_WRAPPER__ #include "big.h" #include "erl_binary.h" #include "erl_bits.h" @@ -83,8 +85,6 @@ new_binary(Process *p, byte *buf, Uint len) * Allocate the binary struct itself. */ bptr = erts_bin_nrml_alloc(len); - bptr->flags = 0; - bptr->orig_size = len; erts_refc_init(&bptr->refc, 1); if (buf != NULL) { sys_memcpy(bptr->orig_bytes, buf, len); @@ -122,8 +122,6 @@ Eterm erts_new_mso_binary(Process *p, byte *buf, int len) * Allocate the binary struct itself. */ bptr = erts_bin_nrml_alloc(len); - bptr->flags = 0; - bptr->orig_size = len; erts_refc_init(&bptr->refc, 1); if (buf != NULL) { sys_memcpy(bptr->orig_bytes, buf, len); @@ -177,7 +175,6 @@ erts_realloc_binary(Eterm bin, size_t size) } else { /* REFC */ ProcBin* pb = (ProcBin *) bval; Binary* newbin = erts_bin_realloc(pb->val, size); - newbin->orig_size = size; pb->val = newbin; pb->size = size; pb->bytes = (byte*) newbin->orig_bytes; diff --git a/erts/emulator/beam/break.c b/erts/emulator/beam/break.c index 08265b590d..e2fa572546 100644 --- a/erts/emulator/beam/break.c +++ b/erts/emulator/beam/break.c @@ -181,7 +181,6 @@ static void doit_print_monitor(ErtsMonitor *mon, void *vpcontext) ASSERT(is_node_name_atom(mon->pid)); erts_print(to, to_arg, "%s{to,{%T,%T},%T}", prefix, mon->name, mon->pid, mon->ref); - erts_print(to, to_arg,"}"); } else if (is_atom(mon->name)){ /* local by name */ erts_print(to, to_arg, "%s{to,{%T,%T},%T}", prefix, mon->name, erts_this_dist_entry->sysname, mon->ref); @@ -210,25 +209,12 @@ print_process_info(int to, void *to_arg, Process *p) erts_print(to, to_arg, "State: "); state = erts_smp_atomic32_read_acqb(&p->state); - if (state & ERTS_PSFLG_FREE) - erts_print(to, to_arg, "Non Existing\n"); /* Should never happen */ - else if (state & ERTS_PSFLG_EXITING) - erts_print(to, to_arg, "Exiting\n"); - else if (state & ERTS_PSFLG_GC) { - garbing = 1; - running = 1; - erts_print(to, to_arg, "Garbing\n"); - } - else if (state & ERTS_PSFLG_SUSPENDED) - erts_print(to, to_arg, "Suspended\n"); - else if (state & ERTS_PSFLG_RUNNING) { - running = 1; - erts_print(to, to_arg, "Running\n"); - } - else if (state & ERTS_PSFLG_ACTIVE) - erts_print(to, to_arg, "Scheduled\n"); - else - erts_print(to, to_arg, "Waiting\n"); + erts_dump_process_state(to, to_arg, state); + if (state & ERTS_PSFLG_GC) { + garbing = 1; + running = 1; + } else if (state & ERTS_PSFLG_RUNNING) + running = 1; /* * If the process is registered as a global process, display the @@ -352,6 +338,10 @@ print_process_info(int to, void *to_arg, Process *p) #endif erts_stack_dump(to, to_arg, p); } + + /* Display all states */ + erts_print(to, to_arg, "Internal State: "); + erts_dump_extended_process_state(to, to_arg, state); } static void @@ -681,27 +671,39 @@ erl_crash_dump_v(char *file, int line, char* fmt, va_list args) char* dumpname; int secs; int env_erl_crash_dump_seconds_set = 1; + int i; if (ERTS_SOMEONE_IS_CRASH_DUMPING) return; #ifdef ERTS_SMP + /* Order all managed threads to block, this has to be done + first to guarantee that this is the only thread to generate + crash dump. */ + erts_thr_progress_fatal_error_block(&tpd_buf); + +#ifdef ERTS_THR_HAVE_SIG_FUNCS /* - * Wait for all managed threads to block. If all threads haven't blocked - * after a minute, we go anyway and hope for the best... - * - * We do not release system again. We expect an exit() or abort() after - * dump has been written. + * We suspend all scheduler threads so that we can dump some + * data about the currently running processes and scheduler data. + * We have to be very very careful when doing this as the schedulers + * could be anywhere. */ - erts_thr_progress_fatal_error_block(60000, &tpd_buf); - /* Either worked or not... */ + for (i = 0; i < erts_no_schedulers; i++) { + erts_tid_t tid = ERTS_SCHEDULER_IX(i)->tid; + if (!erts_equal_tids(tid,erts_thr_self())) + sys_thr_suspend(tid); + } + +#endif /* Allow us to pass certain places without locking... */ erts_smp_atomic32_set_mb(&erts_writing_erl_crash_dump, 1); erts_smp_tsd_set(erts_is_crash_dumping_key, (void *) 1); -#else + +#else /* !ERTS_SMP */ erts_writing_erl_crash_dump = 1; -#endif +#endif /* ERTS_SMP */ envsz = sizeof(env); /* ERL_CRASH_DUMP_SECONDS not set @@ -754,11 +756,12 @@ erl_crash_dump_v(char *file, int line, char* fmt, va_list args) dumpname = "erl_crash.dump"; else dumpname = &dumpnamebuf[0]; + + erts_fprintf(stderr,"\nCrash dump is being written to: %s...", dumpname); fd = open(dumpname,O_WRONLY | O_CREAT | O_TRUNC,0640); - if (fd < 0) + if (fd < 0) return; /* Can't create the crash dump, skip it */ - time(&now); erts_fdprintf(fd, "=erl_crash_dump:0.3\n%s", ctime(&now)); @@ -772,9 +775,74 @@ erl_crash_dump_v(char *file, int line, char* fmt, va_list args) erts_fdprintf(fd, "System version: "); erts_print_system_version(fd, NULL, NULL); erts_fdprintf(fd, "%s\n", "Compiled: " ERLANG_COMPILE_DATE); + erts_fdprintf(fd, "Taints: "); erts_print_nif_taints(fd, NULL); erts_fdprintf(fd, "Atoms: %d\n", atom_table_size()); + +#ifdef USE_THREADS + /* We want to note which thread it was that called erl_exit */ + if (erts_get_scheduler_data()) { + erts_fdprintf(fd, "Calling Thread: scheduler:%d\n", + erts_get_scheduler_data()->no); + } else { + if (!erts_thr_getname(erts_thr_self(), dumpnamebuf, MAXPATHLEN)) + erts_fdprintf(fd, "Calling Thread: %s\n", dumpnamebuf); + else + erts_fdprintf(fd, "Calling Thread: %p\n", erts_thr_self()); + } +#else + erts_fdprintf(fd, "Calling Thread: scheduler:1\n"); +#endif + +#if defined(ERTS_HAVE_TRY_CATCH) + + /* + * erts_print_scheduler_info is not guaranteed to be safe to call + * here for all schedulers as we may have suspended a scheduler + * in the middle of updating the STACK_TOP and STACK_START + * variables and thus when scanning the stack we could get + * segmentation faults. We protect against this very unlikely + * scenario by using the ERTS_SYS_TRY_CATCH. + */ + for (i = 0; i < erts_no_schedulers; i++) { + ERTS_SYS_TRY_CATCH( + erts_print_scheduler_info(fd, NULL, ERTS_SCHEDULER_IX(i)), + erts_fdprintf(fd, "** crashed **\n")); + } +#endif + +#ifdef ERTS_SMP + +#if defined(ERTS_THR_HAVE_SIG_FUNCS) + + /* We resume all schedulers so that we are in a known safe state + when we write the rest of the crash dump */ + for (i = 0; i < erts_no_schedulers; i++) { + erts_tid_t tid = ERTS_SCHEDULER_IX(i)->tid; + if (!erts_equal_tids(tid,erts_thr_self())) + sys_thr_resume(tid); + } +#endif + + /* + * Wait for all managed threads to block. If all threads haven't blocked + * after a minute, we go anyway and hope for the best... + * + * We do not release system again. We expect an exit() or abort() after + * dump has been written. + */ + erts_thr_progress_fatal_error_wait(60000); + /* Either worked or not... */ +#endif + +#ifndef ERTS_HAVE_TRY_CATCH + /* This is safe to call here, as all schedulers are blocked */ + for (i = 0; i < erts_no_schedulers; i++) { + erts_print_scheduler_info(fd, NULL, ERTS_SCHEDULER_IX(i)); + } +#endif + info(fd, NULL); /* General system info */ if (erts_ptab_initialized(&erts_proc)) process_info(fd, NULL); /* Info about each process and port */ @@ -804,7 +872,7 @@ erl_crash_dump_v(char *file, int line, char* fmt, va_list args) erts_fdprintf(fd, "=end\n"); close(fd); - erts_fprintf(stderr,"\nCrash dump was written to: %s\n", dumpname); + erts_fprintf(stderr,"done\n"); } void diff --git a/erts/emulator/beam/copy.c b/erts/emulator/beam/copy.c index 50548850eb..027b85b079 100644 --- a/erts/emulator/beam/copy.c +++ b/erts/emulator/beam/copy.c @@ -21,6 +21,8 @@ # include "config.h" #endif +#define ERL_WANT_GC_INTERNALS__ + #include "sys.h" #include "erl_vm.h" #include "global.h" @@ -125,6 +127,35 @@ Uint size_object(Eterm obj) obj = *bptr; break; } + case HASHMAP_SUBTAG: + switch (MAP_HEADER_TYPE(hdr)) { + case MAP_HEADER_TAG_HAMT_HEAD_BITMAP : + case MAP_HEADER_TAG_HAMT_HEAD_ARRAY : + case MAP_HEADER_TAG_HAMT_NODE_BITMAP : + { + Eterm *head; + Uint sz; + head = hashmap_val_rel(obj, base); + sz = hashmap_bitcount(MAP_HEADER_VAL(hdr)); + sum += 1 + sz + header_arity(hdr); + head += 1 + header_arity(hdr); + + if (sz == 0) { + goto pop_next; + } + while(sz-- > 1) { + obj = head[sz]; + if (!IS_CONST(obj)) { + ESTACK_PUSH(s, obj); + } + } + obj = head[0]; + } + break; + default: + erl_exit(ERTS_ABORT_EXIT, "size_object: bad hashmap type %d\n", MAP_HEADER_TYPE(hdr)); + } + break; case SUB_BINARY_SUBTAG: { Eterm real_bin; @@ -155,10 +186,10 @@ Uint size_object(Eterm obj) case MAP_SUBTAG: { Uint n; - map_t *mp; - mp = (map_t*)map_val_rel(obj,base); + flatmap_t *mp; + mp = (flatmap_t*)flatmap_val_rel(obj,base); ptr = (Eterm *)mp; - n = map_get_size(mp) + 1; + n = flatmap_get_size(mp) + 1; sum += n + 2; ptr += 2; /* hdr + size words */ while (n--) { @@ -340,8 +371,8 @@ 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); + i = flatmap_get_size(objp) + 3; + *argp = make_flatmap_rel(htop, dst_base); while (i--) { *htop++ = *objp++; } @@ -457,7 +488,7 @@ Eterm copy_struct(Eterm obj, Uint sz, Eterm** hpp, ErlOffHeap* off_heap) { ExternalThing *etp = (ExternalThing *) htop; - i = thing_arityval(hdr) + 1; + i = thing_arityval(hdr) + 1; tp = htop; while (i--) { @@ -471,6 +502,21 @@ Eterm copy_struct(Eterm obj, Uint sz, Eterm** hpp, ErlOffHeap* off_heap) *argp = make_external_rel(tp, dst_base); } break; + case HASHMAP_SUBTAG: + tp = htop; + switch (MAP_HEADER_TYPE(hdr)) { + case MAP_HEADER_TAG_HAMT_HEAD_BITMAP : + case MAP_HEADER_TAG_HAMT_HEAD_ARRAY : + *htop++ = *objp++; + case MAP_HEADER_TAG_HAMT_NODE_BITMAP : + i = 1 + hashmap_bitcount(MAP_HEADER_VAL(hdr)); + while (i--) { *htop++ = *objp++; } + *argp = make_hashmap_rel(tp, dst_base); + break; + default: + erl_exit(ERTS_ABORT_EXIT, "copy_struct: bad hashmap type %d\n", MAP_HEADER_TYPE(hdr)); + } + break; case BIN_MATCHSTATE_SUBTAG: erl_exit(ERTS_ABORT_EXIT, "copy_struct: matchstate term not allowed"); diff --git a/erts/emulator/beam/dist.c b/erts/emulator/beam/dist.c index ec07ddcd9c..bfecac1612 100644 --- a/erts/emulator/beam/dist.c +++ b/erts/emulator/beam/dist.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 @@ -119,7 +119,7 @@ Export* dmonitor_p_trap = NULL; /* forward declarations */ static void clear_dist_entry(DistEntry*); -static int dsig_send(ErtsDSigData *, Eterm, Eterm, int); +static int dsig_send_ctl(ErtsDSigData* dsdp, Eterm ctl, int force_busy); static void send_nodes_mon_msgs(Process *, Eterm, Eterm, Eterm, Eterm); static void init_nodes_monitors(void); @@ -622,9 +622,7 @@ alloc_dist_obuf(Uint size) ErtsDistOutputBuf *obuf; Uint obuf_size = sizeof(ErtsDistOutputBuf)+sizeof(byte)*(size-1); Binary *bin = erts_bin_drv_alloc(obuf_size); - bin->flags = BIN_FLAG_DRV; erts_refc_init(&bin->refc, 1); - bin->orig_size = (SWord) obuf_size; obuf = (ErtsDistOutputBuf *) &bin->orig_bytes[0]; #ifdef DEBUG obuf->dbg_pattern = ERTS_DIST_OUTPUT_BUF_DBG_PATTERN; @@ -709,6 +707,55 @@ static void clear_dist_entry(DistEntry *dep) } } +void erts_dsend_context_dtor(Binary* ctx_bin) +{ + ErtsSendContext* ctx = ERTS_MAGIC_BIN_DATA(ctx_bin); + switch (ctx->dss.phase) { + case ERTS_DSIG_SEND_PHASE_MSG_SIZE: + DESTROY_SAVED_ESTACK(&ctx->dss.u.sc.estack); + break; + case ERTS_DSIG_SEND_PHASE_MSG_ENCODE: + DESTROY_SAVED_WSTACK(&ctx->dss.u.ec.wstack); + break; + default:; + } + if (ctx->dss.phase >= ERTS_DSIG_SEND_PHASE_ALLOC && ctx->dss.obuf) { + free_dist_obuf(ctx->dss.obuf); + } + if (ctx->dep_to_deref) + erts_deref_dist_entry(ctx->dep_to_deref); +} + +Eterm erts_dsend_export_trap_context(Process* p, ErtsSendContext* ctx) +{ + struct exported_ctx { + ErtsSendContext ctx; + ErtsAtomCacheMap acm; + }; + Binary* ctx_bin = erts_create_magic_binary(sizeof(struct exported_ctx), + erts_dsend_context_dtor); + struct exported_ctx* dst = ERTS_MAGIC_BIN_DATA(ctx_bin); + Uint ctl_size = !HALFWORD_HEAP ? 0 : (arityval(ctx->ctl_heap[0]) + 1); + Eterm* hp = HAlloc(p, ctl_size + PROC_BIN_SIZE); + + sys_memcpy(&dst->ctx, ctx, sizeof(ErtsSendContext)); + ASSERT(ctx->dss.ctl == make_tuple(ctx->ctl_heap)); +#if !HALFWORD_HEAP + dst->ctx.dss.ctl = make_tuple(dst->ctx.ctl_heap); +#else + /* Must put control tuple in low mem */ + sys_memcpy(hp, ctx->ctl_heap, ctl_size*sizeof(Eterm)); + dst->ctx.dss.ctl = make_tuple(hp); + hp += ctl_size; +#endif + if (ctx->dss.acmp) { + sys_memcpy(&dst->acm, ctx->dss.acmp, sizeof(ErtsAtomCacheMap)); + dst->ctx.dss.acmp = &dst->acm; + } + return erts_mk_magic_binary_term(&hp, &MSO(p), ctx_bin); +} + + /* * The erts_dsig_send_*() functions implemented below, sends asynchronous * distributed signals to other Erlang nodes. Before sending a distributed @@ -731,7 +778,7 @@ erts_dsig_send_link(ErtsDSigData *dsdp, Eterm local, Eterm remote) int res; UseTmpHeapNoproc(4); - res = dsig_send(dsdp, ctl, THE_NON_VALUE, 0); + res = dsig_send_ctl(dsdp, ctl, 0); UnUseTmpHeapNoproc(4); return res; } @@ -744,7 +791,7 @@ erts_dsig_send_unlink(ErtsDSigData *dsdp, Eterm local, Eterm remote) int res; UseTmpHeapNoproc(4); - res = dsig_send(dsdp, ctl, THE_NON_VALUE, 0); + res = dsig_send_ctl(dsdp, ctl, 0); UnUseTmpHeapNoproc(4); return res; } @@ -772,7 +819,7 @@ erts_dsig_send_m_exit(ErtsDSigData *dsdp, Eterm watcher, Eterm watched, erts_smp_de_links_unlock(dsdp->dep); #endif - res = dsig_send(dsdp, ctl, THE_NON_VALUE, 1); + res = dsig_send_ctl(dsdp, ctl, 1); UnUseTmpHeapNoproc(6); return res; } @@ -793,7 +840,7 @@ erts_dsig_send_monitor(ErtsDSigData *dsdp, Eterm watcher, Eterm watched, make_small(DOP_MONITOR_P), watcher, watched, ref); - res = dsig_send(dsdp, ctl, THE_NON_VALUE, 0); + res = dsig_send_ctl(dsdp, ctl, 0); UnUseTmpHeapNoproc(5); return res; } @@ -815,18 +862,17 @@ erts_dsig_send_demonitor(ErtsDSigData *dsdp, Eterm watcher, make_small(DOP_DEMONITOR_P), watcher, watched, ref); - res = dsig_send(dsdp, ctl, THE_NON_VALUE, force); + res = dsig_send_ctl(dsdp, ctl, force); UnUseTmpHeapNoproc(5); return res; } int -erts_dsig_send_msg(ErtsDSigData *dsdp, Eterm remote, Eterm message) +erts_dsig_send_msg(Eterm remote, Eterm message, ErtsSendContext* ctx) { Eterm ctl; - DeclareTmpHeapNoproc(ctl_heap,5); Eterm token = NIL; - Process *sender = dsdp->proc; + Process *sender = ctx->dsd.proc; int res; #ifdef USE_VM_PROBES Sint tok_label = 0; @@ -838,8 +884,7 @@ erts_dsig_send_msg(ErtsDSigData *dsdp, Eterm remote, Eterm message) DTRACE_CHARBUF(receiver_name, 64); #endif - UseTmpHeapNoproc(5); - if (SEQ_TRACE_TOKEN(sender) != NIL + if (SEQ_TRACE_TOKEN(sender) != NIL #ifdef USE_VM_PROBES && SEQ_TRACE_TOKEN(sender) != am_have_dt_utag #endif @@ -852,7 +897,7 @@ erts_dsig_send_msg(ErtsDSigData *dsdp, Eterm remote, Eterm message) *node_name = *sender_name = *receiver_name = '\0'; if (DTRACE_ENABLED(message_send) || DTRACE_ENABLED(message_send_remote)) { erts_snprintf(node_name, sizeof(DTRACE_CHARBUF_NAME(node_name)), - "%T", dsdp->dep->sysname); + "%T", ctx->dsd.dep->sysname); erts_snprintf(sender_name, sizeof(DTRACE_CHARBUF_NAME(sender_name)), "%T", sender->common.id); erts_snprintf(receiver_name, sizeof(DTRACE_CHARBUF_NAME(receiver_name)), @@ -867,26 +912,28 @@ erts_dsig_send_msg(ErtsDSigData *dsdp, Eterm remote, Eterm message) #endif if (token != NIL) - ctl = TUPLE4(&ctl_heap[0], + ctl = TUPLE4(&ctx->ctl_heap[0], make_small(DOP_SEND_TT), am_Cookie, remote, token); else - ctl = TUPLE3(&ctl_heap[0], make_small(DOP_SEND), am_Cookie, remote); + ctl = TUPLE3(&ctx->ctl_heap[0], make_small(DOP_SEND), am_Cookie, remote); DTRACE6(message_send, sender_name, receiver_name, msize, tok_label, tok_lastcnt, tok_serial); DTRACE7(message_send_remote, sender_name, node_name, receiver_name, msize, tok_label, tok_lastcnt, tok_serial); - res = dsig_send(dsdp, ctl, message, 0); - UnUseTmpHeapNoproc(5); + ctx->dss.ctl = ctl; + ctx->dss.msg = message; + ctx->dss.force_busy = 0; + res = erts_dsig_send(&ctx->dsd, &ctx->dss); return res; } int -erts_dsig_send_reg_msg(ErtsDSigData *dsdp, Eterm remote_name, Eterm message) +erts_dsig_send_reg_msg(Eterm remote_name, Eterm message, + ErtsSendContext* ctx) { Eterm ctl; - DeclareTmpHeapNoproc(ctl_heap,6); Eterm token = NIL; - Process *sender = dsdp->proc; + Process *sender = ctx->dsd.proc; int res; #ifdef USE_VM_PROBES Sint tok_label = 0; @@ -898,7 +945,6 @@ erts_dsig_send_reg_msg(ErtsDSigData *dsdp, Eterm remote_name, Eterm message) DTRACE_CHARBUF(receiver_name, 128); #endif - UseTmpHeapNoproc(6); if (SEQ_TRACE_TOKEN(sender) != NIL #ifdef USE_VM_PROBES && SEQ_TRACE_TOKEN(sender) != am_have_dt_utag @@ -912,7 +958,7 @@ erts_dsig_send_reg_msg(ErtsDSigData *dsdp, Eterm remote_name, Eterm message) *node_name = *sender_name = *receiver_name = '\0'; if (DTRACE_ENABLED(message_send) || DTRACE_ENABLED(message_send_remote)) { erts_snprintf(node_name, sizeof(DTRACE_CHARBUF_NAME(node_name)), - "%T", dsdp->dep->sysname); + "%T", ctx->dsd.dep->sysname); erts_snprintf(sender_name, sizeof(DTRACE_CHARBUF_NAME(sender_name)), "%T", sender->common.id); erts_snprintf(receiver_name, sizeof(DTRACE_CHARBUF_NAME(receiver_name)), @@ -927,17 +973,19 @@ erts_dsig_send_reg_msg(ErtsDSigData *dsdp, Eterm remote_name, Eterm message) #endif if (token != NIL) - ctl = TUPLE5(&ctl_heap[0], make_small(DOP_REG_SEND_TT), + ctl = TUPLE5(&ctx->ctl_heap[0], make_small(DOP_REG_SEND_TT), sender->common.id, am_Cookie, remote_name, token); else - ctl = TUPLE4(&ctl_heap[0], make_small(DOP_REG_SEND), + ctl = TUPLE4(&ctx->ctl_heap[0], make_small(DOP_REG_SEND), sender->common.id, am_Cookie, remote_name); DTRACE6(message_send, sender_name, receiver_name, msize, tok_label, tok_lastcnt, tok_serial); DTRACE7(message_send_remote, sender_name, node_name, receiver_name, msize, tok_label, tok_lastcnt, tok_serial); - res = dsig_send(dsdp, ctl, message, 0); - UnUseTmpHeapNoproc(6); + ctx->dss.ctl = ctl; + ctx->dss.msg = message; + ctx->dss.force_busy = 0; + res = erts_dsig_send(&ctx->dsd, &ctx->dss); return res; } @@ -994,7 +1042,7 @@ erts_dsig_send_exit_tt(ErtsDSigData *dsdp, Eterm local, Eterm remote, DTRACE7(process_exit_signal_remote, sender_name, node_name, remote_name, reason_str, tok_label, tok_lastcnt, tok_serial); /* forced, i.e ignore busy */ - res = dsig_send(dsdp, ctl, THE_NON_VALUE, 1); + res = dsig_send_ctl(dsdp, ctl, 1); UnUseTmpHeapNoproc(6); return res; } @@ -1010,7 +1058,7 @@ erts_dsig_send_exit(ErtsDSigData *dsdp, Eterm local, Eterm remote, Eterm reason) ctl = TUPLE4(&ctl_heap[0], make_small(DOP_EXIT), local, remote, reason); /* forced, i.e ignore busy */ - res = dsig_send(dsdp, ctl, THE_NON_VALUE, 1); + res = dsig_send_ctl(dsdp, ctl, 1); UnUseTmpHeapNoproc(5); return res; } @@ -1026,7 +1074,7 @@ erts_dsig_send_exit2(ErtsDSigData *dsdp, Eterm local, Eterm remote, Eterm reason ctl = TUPLE4(&ctl_heap[0], make_small(DOP_EXIT2), local, remote, reason); - res = dsig_send(dsdp, ctl, THE_NON_VALUE, 0); + res = dsig_send_ctl(dsdp, ctl, 0); UnUseTmpHeapNoproc(5); return res; } @@ -1043,7 +1091,7 @@ erts_dsig_send_group_leader(ErtsDSigData *dsdp, Eterm leader, Eterm remote) ctl = TUPLE3(&ctl_heap[0], make_small(DOP_GROUP_LEADER), leader, remote); - res = dsig_send(dsdp, ctl, THE_NON_VALUE, 0); + res = dsig_send_ctl(dsdp, ctl, 0); UnUseTmpHeapNoproc(4); return res; } @@ -1693,194 +1741,235 @@ int erts_net_message(Port *prt, return -1; } -static int -dsig_send(ErtsDSigData *dsdp, Eterm ctl, Eterm msg, int force_busy) +static int dsig_send_ctl(ErtsDSigData* dsdp, Eterm ctl, int force_busy) { + struct erts_dsig_send_context ctx; + int ret; + ctx.ctl = ctl; + ctx.msg = THE_NON_VALUE; + ctx.force_busy = force_busy; + ctx.phase = ERTS_DSIG_SEND_PHASE_INIT; +#ifdef DEBUG + ctx.reds = 1; /* provoke assert below (no reduction count without msg) */ +#endif + ret = erts_dsig_send(dsdp, &ctx); + ASSERT(ret != ERTS_DSIG_SEND_CONTINUE); + return ret; +} + +int +erts_dsig_send(ErtsDSigData *dsdp, struct erts_dsig_send_context* ctx) +{ + int retval; + Sint initial_reds = ctx->reds; Eterm cid; - int suspended = 0; - int resume = 0; - Uint32 pass_through_size; - Uint data_size, dhdr_ext_size; - ErtsAtomCacheMap *acmp; - ErtsDistOutputBuf *obuf; - DistEntry *dep = dsdp->dep; - Uint32 flags = dep->flags; - Process *c_p = dsdp->proc; - if (!c_p || dsdp->no_suspend) - force_busy = 1; + while (1) { + switch (ctx->phase) { + case ERTS_DSIG_SEND_PHASE_INIT: + ctx->flags = dsdp->dep->flags; + ctx->c_p = dsdp->proc; - ERTS_SMP_LC_ASSERT(!c_p - || (ERTS_PROC_LOCK_MAIN - == erts_proc_lc_my_proc_locks(c_p))); + if (!ctx->c_p || dsdp->no_suspend) + ctx->force_busy = 1; - if (!erts_is_alive) - return ERTS_DSIG_SEND_OK; + ERTS_SMP_LC_ASSERT(!ctx->c_p + || (ERTS_PROC_LOCK_MAIN + == erts_proc_lc_my_proc_locks(ctx->c_p))); - if (flags & DFLAG_DIST_HDR_ATOM_CACHE) { - acmp = erts_get_atom_cache_map(c_p); - pass_through_size = 0; - } - else { - acmp = NULL; - pass_through_size = 1; - } + if (!erts_is_alive) + return ERTS_DSIG_SEND_OK; -#ifdef ERTS_DIST_MSG_DBG - erts_fprintf(stderr, ">>%s CTL: %T\n", pass_through_size ? "P" : " ", ctl); - if (is_value(msg)) - erts_fprintf(stderr, " MSG: %T\n", msg); -#endif + if (ctx->flags & DFLAG_DIST_HDR_ATOM_CACHE) { + ctx->acmp = erts_get_atom_cache_map(ctx->c_p); + ctx->pass_through_size = 0; + } + else { + ctx->acmp = NULL; + ctx->pass_through_size = 1; + } - data_size = pass_through_size; - erts_reset_atom_cache_map(acmp); - data_size += erts_encode_dist_ext_size(ctl, flags, acmp); - if (is_value(msg)) - data_size += erts_encode_dist_ext_size(msg, flags, acmp); - erts_finalize_atom_cache_map(acmp, flags); + #ifdef ERTS_DIST_MSG_DBG + erts_fprintf(stderr, ">>%s CTL: %T\n", ctx->pass_through_size ? "P" : " ", ctx->ctl); + if (is_value(msg)) + erts_fprintf(stderr, " MSG: %T\n", msg); + #endif + + ctx->data_size = ctx->pass_through_size; + erts_reset_atom_cache_map(ctx->acmp); + erts_encode_dist_ext_size(ctx->ctl, ctx->flags, ctx->acmp, &ctx->data_size); + + if (is_value(ctx->msg)) { + ctx->u.sc.estack.start = NULL; + ctx->u.sc.flags = ctx->flags; + ctx->u.sc.level = 0; + ctx->phase = ERTS_DSIG_SEND_PHASE_MSG_SIZE; + } else { + ctx->phase = ERTS_DSIG_SEND_PHASE_ALLOC; + } + break; - dhdr_ext_size = erts_encode_ext_dist_header_size(acmp); - data_size += dhdr_ext_size; + case ERTS_DSIG_SEND_PHASE_MSG_SIZE: + if (erts_encode_dist_ext_size_int(ctx->msg, ctx, &ctx->data_size)) { + retval = ERTS_DSIG_SEND_CONTINUE; + goto done; + } - obuf = alloc_dist_obuf(data_size); - obuf->ext_endp = &obuf->data[0] + pass_through_size + dhdr_ext_size; + ctx->phase = ERTS_DSIG_SEND_PHASE_ALLOC; + case ERTS_DSIG_SEND_PHASE_ALLOC: + erts_finalize_atom_cache_map(ctx->acmp, ctx->flags); + + ctx->dhdr_ext_size = erts_encode_ext_dist_header_size(ctx->acmp); + ctx->data_size += ctx->dhdr_ext_size; + + ctx->obuf = alloc_dist_obuf(ctx->data_size); + ctx->obuf->ext_endp = &ctx->obuf->data[0] + ctx->pass_through_size + ctx->dhdr_ext_size; + + /* Encode internal version of dist header */ + ctx->obuf->extp = erts_encode_ext_dist_header_setup(ctx->obuf->ext_endp, ctx->acmp); + /* Encode control message */ + erts_encode_dist_ext(ctx->ctl, &ctx->obuf->ext_endp, ctx->flags, ctx->acmp, NULL, NULL); + if (is_value(ctx->msg)) { + ctx->u.ec.flags = ctx->flags; + ctx->u.ec.level = 0; + ctx->u.ec.wstack.wstart = NULL; + ctx->phase = ERTS_DSIG_SEND_PHASE_MSG_ENCODE; + } else { + ctx->phase = ERTS_DSIG_SEND_PHASE_FIN; + } + break; - /* Encode internal version of dist header */ - obuf->extp = erts_encode_ext_dist_header_setup(obuf->ext_endp, acmp); - /* Encode control message */ - erts_encode_dist_ext(ctl, &obuf->ext_endp, flags, acmp); - if (is_value(msg)) { - /* Encode message */ - erts_encode_dist_ext(msg, &obuf->ext_endp, flags, acmp); - } + case ERTS_DSIG_SEND_PHASE_MSG_ENCODE: + if (erts_encode_dist_ext(ctx->msg, &ctx->obuf->ext_endp, ctx->flags, ctx->acmp, &ctx->u.ec, &ctx->reds)) { + retval = ERTS_DSIG_SEND_CONTINUE; + goto done; + } - ASSERT(obuf->extp < obuf->ext_endp); - ASSERT(&obuf->data[0] <= obuf->extp - pass_through_size); - ASSERT(obuf->ext_endp <= &obuf->data[0] + data_size); + ctx->phase = ERTS_DSIG_SEND_PHASE_FIN; + case ERTS_DSIG_SEND_PHASE_FIN: { + DistEntry *dep = dsdp->dep; + int suspended = 0; + int resume = 0; - data_size = obuf->ext_endp - obuf->extp; + ASSERT(ctx->obuf->extp < ctx->obuf->ext_endp); + ASSERT(&ctx->obuf->data[0] <= ctx->obuf->extp - ctx->pass_through_size); + ASSERT(ctx->obuf->ext_endp <= &ctx->obuf->data[0] + ctx->data_size); - /* - * Signal encoded; now verify that the connection still exists, - * and if so enqueue the signal and schedule it for send. - */ - obuf->next = NULL; - erts_smp_de_rlock(dep); - cid = dep->cid; - if (cid != dsdp->cid - || dep->connection_id != dsdp->connection_id - || dep->status & ERTS_DE_SFLG_EXITING) { - /* Not the same connection as when we started; drop message... */ - erts_smp_de_runlock(dep); - free_dist_obuf(obuf); - } - else { - ErtsProcList *plp = NULL; - erts_smp_mtx_lock(&dep->qlock); - dep->qsize += size_obuf(obuf); - if (dep->qsize >= erts_dist_buf_busy_limit) - dep->qflgs |= ERTS_DE_QFLG_BUSY; - if (!force_busy && (dep->qflgs & ERTS_DE_QFLG_BUSY)) { - erts_smp_mtx_unlock(&dep->qlock); + ctx->data_size = ctx->obuf->ext_endp - ctx->obuf->extp; - plp = erts_proclist_create(c_p); - erts_suspend(c_p, ERTS_PROC_LOCK_MAIN, NULL); - suspended = 1; - erts_smp_mtx_lock(&dep->qlock); - } + /* + * Signal encoded; now verify that the connection still exists, + * and if so enqueue the signal and schedule it for send. + */ + ctx->obuf->next = NULL; + erts_smp_de_rlock(dep); + cid = dep->cid; + if (cid != dsdp->cid + || dep->connection_id != dsdp->connection_id + || dep->status & ERTS_DE_SFLG_EXITING) { + /* Not the same connection as when we started; drop message... */ + erts_smp_de_runlock(dep); + free_dist_obuf(ctx->obuf); + } + else { + ErtsProcList *plp = NULL; + erts_smp_mtx_lock(&dep->qlock); + dep->qsize += size_obuf(ctx->obuf); + if (dep->qsize >= erts_dist_buf_busy_limit) + dep->qflgs |= ERTS_DE_QFLG_BUSY; + if (!ctx->force_busy && (dep->qflgs & ERTS_DE_QFLG_BUSY)) { + erts_smp_mtx_unlock(&dep->qlock); + + plp = erts_proclist_create(ctx->c_p); + erts_suspend(ctx->c_p, ERTS_PROC_LOCK_MAIN, NULL); + suspended = 1; + erts_smp_mtx_lock(&dep->qlock); + } - /* Enqueue obuf on dist entry */ - if (dep->out_queue.last) - dep->out_queue.last->next = obuf; - else - dep->out_queue.first = obuf; - dep->out_queue.last = obuf; + /* Enqueue obuf on dist entry */ + if (dep->out_queue.last) + dep->out_queue.last->next = ctx->obuf; + else + dep->out_queue.first = ctx->obuf; + dep->out_queue.last = ctx->obuf; + + if (!ctx->force_busy) { + if (!(dep->qflgs & ERTS_DE_QFLG_BUSY)) { + if (suspended) + resume = 1; /* was busy when we started, but isn't now */ + #ifdef USE_VM_PROBES + if (resume && DTRACE_ENABLED(dist_port_not_busy)) { + DTRACE_CHARBUF(port_str, 64); + DTRACE_CHARBUF(remote_str, 64); + + erts_snprintf(port_str, sizeof(DTRACE_CHARBUF_NAME(port_str)), + "%T", cid); + erts_snprintf(remote_str, sizeof(DTRACE_CHARBUF_NAME(remote_str)), + "%T", dep->sysname); + DTRACE3(dist_port_not_busy, erts_this_node_sysname, + port_str, remote_str); + } + #endif + } + else { + /* Enqueue suspended process on dist entry */ + ASSERT(plp); + erts_proclist_store_last(&dep->suspended, plp); + } + } - if (!force_busy) { - if (!(dep->qflgs & ERTS_DE_QFLG_BUSY)) { - if (suspended) - resume = 1; /* was busy when we started, but isn't now */ -#ifdef USE_VM_PROBES - if (resume && DTRACE_ENABLED(dist_port_not_busy)) { - DTRACE_CHARBUF(port_str, 64); - DTRACE_CHARBUF(remote_str, 64); - - erts_snprintf(port_str, sizeof(DTRACE_CHARBUF_NAME(port_str)), - "%T", cid); - erts_snprintf(remote_str, sizeof(DTRACE_CHARBUF_NAME(remote_str)), - "%T", dep->sysname); - DTRACE3(dist_port_not_busy, erts_this_node_sysname, - port_str, remote_str); - } -#endif + erts_smp_mtx_unlock(&dep->qlock); + erts_schedule_dist_command(NULL, dep); + erts_smp_de_runlock(dep); + + if (resume) { + erts_resume(ctx->c_p, ERTS_PROC_LOCK_MAIN); + erts_proclist_destroy(plp); + /* + * Note that the calling process still have to yield as if it + * suspended. If not, the calling process could later be + * erroneously scheduled when it shouldn't be. + */ + } } - else { - /* Enqueue suspended process on dist entry */ - ASSERT(plp); - erts_proclist_store_last(&dep->suspended, plp); + ctx->obuf = NULL; + + if (suspended) { + #ifdef USE_VM_PROBES + if (!resume && DTRACE_ENABLED(dist_port_busy)) { + DTRACE_CHARBUF(port_str, 64); + DTRACE_CHARBUF(remote_str, 64); + DTRACE_CHARBUF(pid_str, 16); + + erts_snprintf(port_str, sizeof(DTRACE_CHARBUF_NAME(port_str)), "%T", cid); + erts_snprintf(remote_str, sizeof(DTRACE_CHARBUF_NAME(remote_str)), + "%T", dep->sysname); + erts_snprintf(pid_str, sizeof(DTRACE_CHARBUF_NAME(pid_str)), + "%T", ctx->c_p->common.id); + DTRACE4(dist_port_busy, erts_this_node_sysname, + port_str, remote_str, pid_str); + } + #endif + if (!resume && erts_system_monitor_flags.busy_dist_port) + monitor_generic(ctx->c_p, am_busy_dist_port, cid); + retval = ERTS_DSIG_SEND_YIELD; + } else { + retval = ERTS_DSIG_SEND_OK; } + goto done; } - - erts_smp_mtx_unlock(&dep->qlock); - erts_schedule_dist_command(NULL, dep); - erts_smp_de_runlock(dep); - - if (resume) { - erts_resume(c_p, ERTS_PROC_LOCK_MAIN); - erts_proclist_destroy(plp); - /* - * Note that the calling process still have to yield as if it - * suspended. If not, the calling process could later be - * erroneously scheduled when it shouldn't be. - */ + default: + erl_exit(ERTS_ABORT_EXIT, "dsig_send invalid phase (%d)\n", (int)ctx->phase); } } - if (c_p) { - int reds; - /* - * Bump reductions on calling process. - * - * This is the reduction cost: Always a base cost of 8 reductions - * plus 16 reductions per kilobyte generated external data. - */ - - data_size >>= (10-4); -#if defined(ARCH_64) && !HALFWORD_HEAP - data_size &= 0x003fffffffffffff; -#elif defined(ARCH_32) || HALFWORD_HEAP - data_size &= 0x003fffff; -#else -# error "Ohh come on ... !?!" -#endif - reds = 8 + ((int) data_size > 1000000 ? 1000000 : (int) data_size); - BUMP_REDS(c_p, reds); - } - - if (suspended) { -#ifdef USE_VM_PROBES - if (!resume && DTRACE_ENABLED(dist_port_busy)) { - DTRACE_CHARBUF(port_str, 64); - DTRACE_CHARBUF(remote_str, 64); - DTRACE_CHARBUF(pid_str, 16); - - erts_snprintf(port_str, sizeof(DTRACE_CHARBUF_NAME(port_str)), "%T", cid); - erts_snprintf(remote_str, sizeof(DTRACE_CHARBUF_NAME(remote_str)), - "%T", dep->sysname); - erts_snprintf(pid_str, sizeof(DTRACE_CHARBUF_NAME(pid_str)), - "%T", c_p->common.id); - DTRACE4(dist_port_busy, erts_this_node_sysname, - port_str, remote_str, pid_str); - } -#endif - if (!resume && erts_system_monitor_flags.busy_dist_port) - monitor_generic(c_p, am_busy_dist_port, cid); - return ERTS_DSIG_SEND_YIELD; +done: + if (ctx->msg && ctx->c_p) { + BUMP_REDS(ctx->c_p, (initial_reds - ctx->reds) / TERM_TO_BINARY_LOOP_FACTOR); } - return ERTS_DSIG_SEND_OK; + return retval; } - static Uint dist_port_command(Port *prt, ErtsDistOutputBuf *obuf) { diff --git a/erts/emulator/beam/dist.h b/erts/emulator/beam/dist.h index f32b999198..2a2ba0c83f 100644 --- a/erts/emulator/beam/dist.h +++ b/erts/emulator/beam/dist.h @@ -22,6 +22,7 @@ #include "erl_process.h" #include "erl_node_tables.h" +#include "zlib.h" #define DFLAG_PUBLISHED 0x01 #define DFLAG_ATOM_CACHE 0x02 @@ -264,17 +265,105 @@ erts_destroy_dist_link(ErtsDistLinkData *dldp) #endif + + +/* Define for testing */ +/* #define EXTREME_TTB_TRAPPING 1 */ + +#ifndef EXTREME_TTB_TRAPPING +#define TERM_TO_BINARY_LOOP_FACTOR 32 +#else +#define TERM_TO_BINARY_LOOP_FACTOR 1 +#endif + +typedef enum { TTBSize, TTBEncode, TTBCompress } TTBState; +typedef struct TTBSizeContext_ { + Uint flags; + int level; + Uint result; + Eterm obj; + ErtsEStack estack; +} TTBSizeContext; + +typedef struct TTBEncodeContext_ { + Uint flags; + int level; + byte* ep; + Eterm obj; + ErtsWStack wstack; + Binary *result_bin; +} TTBEncodeContext; + +typedef struct { + Uint real_size; + Uint dest_len; + byte *dbytes; + Binary *result_bin; + Binary *destination_bin; + z_stream stream; +} TTBCompressContext; + +typedef struct { + int alive; + TTBState state; + union { + TTBSizeContext sc; + TTBEncodeContext ec; + TTBCompressContext cc; + } s; +} TTBContext; + +enum erts_dsig_send_phase { + ERTS_DSIG_SEND_PHASE_INIT, + ERTS_DSIG_SEND_PHASE_MSG_SIZE, + ERTS_DSIG_SEND_PHASE_ALLOC, + ERTS_DSIG_SEND_PHASE_MSG_ENCODE, + ERTS_DSIG_SEND_PHASE_FIN +}; + +struct erts_dsig_send_context { + enum erts_dsig_send_phase phase; + Sint reds; + + Eterm ctl; + Eterm msg; + int force_busy; + Uint32 pass_through_size; + Uint data_size, dhdr_ext_size; + ErtsAtomCacheMap *acmp; + ErtsDistOutputBuf *obuf; + Uint32 flags; + Process *c_p; + union { + TTBSizeContext sc; + TTBEncodeContext ec; + }u; +}; + +typedef struct { + int suspend; + + Eterm ctl_heap[6]; + ErtsDSigData dsd; + DistEntry* dep_to_deref; + struct erts_dsig_send_context dss; + + Eterm return_term; +}ErtsSendContext; + + /* * erts_dsig_send_* return values. */ #define ERTS_DSIG_SEND_OK 0 #define ERTS_DSIG_SEND_YIELD 1 +#define ERTS_DSIG_SEND_CONTINUE 2 extern int erts_dsig_send_link(ErtsDSigData *, Eterm, Eterm); -extern int erts_dsig_send_msg(ErtsDSigData *, Eterm, Eterm); +extern int erts_dsig_send_msg(Eterm, Eterm, ErtsSendContext*); extern int erts_dsig_send_exit_tt(ErtsDSigData *, Eterm, Eterm, Eterm, Eterm); extern int erts_dsig_send_unlink(ErtsDSigData *, Eterm, Eterm); -extern int erts_dsig_send_reg_msg(ErtsDSigData *, Eterm, Eterm); +extern int erts_dsig_send_reg_msg(Eterm, Eterm, ErtsSendContext*); extern int erts_dsig_send_group_leader(ErtsDSigData *, Eterm, Eterm); extern int erts_dsig_send_exit(ErtsDSigData *, Eterm, Eterm, Eterm); extern int erts_dsig_send_exit2(ErtsDSigData *, Eterm, Eterm, Eterm); @@ -282,6 +371,10 @@ extern int erts_dsig_send_demonitor(ErtsDSigData *, Eterm, Eterm, Eterm, int); extern int erts_dsig_send_monitor(ErtsDSigData *, Eterm, Eterm, Eterm); extern int erts_dsig_send_m_exit(ErtsDSigData *, Eterm, Eterm, Eterm, Eterm); +extern int erts_dsig_send(ErtsDSigData *dsdp, struct erts_dsig_send_context* ctx); +extern void erts_dsend_context_dtor(Binary*); +extern Eterm erts_dsend_export_trap_context(Process* p, ErtsSendContext* ctx); + extern int erts_dist_command(Port *prt, int reds); extern void erts_dist_port_not_busy(Port *prt); extern void erts_kill_dist_connection(DistEntry *dep, Uint32); diff --git a/erts/emulator/beam/erl_alloc.c b/erts/emulator/beam/erl_alloc.c index 90cd227fae..f2bceff4eb 100644 --- a/erts/emulator/beam/erl_alloc.c +++ b/erts/emulator/beam/erl_alloc.c @@ -3939,7 +3939,7 @@ static Uint install_debug_functions(void) { int i; - ASSERT(sizeof(erts_allctrs) == sizeof(real_allctrs)); + ERTS_CT_ASSERT(sizeof(erts_allctrs) == sizeof(real_allctrs)); sys_memcpy((void *)real_allctrs,(void *)erts_allctrs,sizeof(erts_allctrs)); diff --git a/erts/emulator/beam/erl_alloc.types b/erts/emulator/beam/erl_alloc.types index 21434eb117..61def65235 100644 --- a/erts/emulator/beam/erl_alloc.types +++ b/erts/emulator/beam/erl_alloc.types @@ -415,6 +415,8 @@ type CS_PROG_PATH LONG_LIVED SYSTEM cs_prog_path type ENVIRONMENT TEMPORARY SYSTEM environment type PUTENV_STR SYSTEM SYSTEM putenv_string type PRT_REP_EXIT STANDARD SYSTEM port_report_exit +type SYS_BLOCKING STANDARD SYSTEM sys_blocking +type SYS_WRITE_BUF TEMPORARY SYSTEM sys_write_buf +endif diff --git a/erts/emulator/beam/erl_alloc_util.c b/erts/emulator/beam/erl_alloc_util.c index e3172dc4fb..2f277690e4 100644 --- a/erts/emulator/beam/erl_alloc_util.c +++ b/erts/emulator/beam/erl_alloc_util.c @@ -1442,7 +1442,7 @@ get_pref_allctr(void *extra) pref_ix = ERTS_ALC_GET_THR_IX(); - ASSERT(sizeof(UWord) == sizeof(Allctr_t *)); + ERTS_CT_ASSERT(sizeof(UWord) == sizeof(Allctr_t *)); ASSERT(0 <= pref_ix && pref_ix < tspec->size); return tspec->allctr[pref_ix]; @@ -1861,7 +1861,7 @@ handle_delayed_dealloc(Allctr_t *allctr, * if this carrier is pulled from dc_list by cpool_fetch() */ ERTS_ALC_CPOOL_ASSERT(FBLK_TO_MBC(blk) != crr); - ERTS_ALC_CPOOL_ASSERT(sizeof(ErtsAllctrDDBlock_t) == sizeof(void*)); + ERTS_CT_ASSERT(sizeof(ErtsAllctrDDBlock_t) == sizeof(void*)); #ifdef MBC_ABLK_OFFSET_BITS blk->u.carrier = crr; #else @@ -5942,7 +5942,7 @@ erts_alcu_init(AlcUInit_t *init) erts_atomic_init_nob(&sentinel->prev, (erts_aint_t) sentinel); } #endif - ASSERT(SBC_BLK_SZ_MASK == MBC_FBLK_SZ_MASK); /* see BLK_SZ */ + ERTS_CT_ASSERT(SBC_BLK_SZ_MASK == MBC_FBLK_SZ_MASK); /* see BLK_SZ */ #if HAVE_ERTS_MSEG ASSERT(erts_mseg_unit_size() == ERTS_SACRR_UNIT_SZ); max_mseg_carriers = init->mmc; diff --git a/erts/emulator/beam/erl_async.c b/erts/emulator/beam/erl_async.c index decae6b2ca..bc06d41720 100644 --- a/erts/emulator/beam/erl_async.c +++ b/erts/emulator/beam/erl_async.c @@ -176,7 +176,7 @@ erts_init_async(void) ErtsThrQInit_t qinit = ERTS_THR_Q_INIT_DEFAULT; #endif erts_thr_opts_t thr_opts = ERTS_THR_OPTS_DEFAULT_INITER; - char *ptr; + char *ptr, thr_name[16]; size_t tot_size = 0; int i; @@ -227,23 +227,16 @@ erts_init_async(void) thr_opts.suggested_stack_size = erts_async_thread_suggested_stack_size; -#ifdef ETHR_HAVE_THREAD_NAMES - thr_opts.name = malloc(sizeof(char)*(strlen("async_XXXX")+1)); -#endif + thr_opts.name = thr_name; for (i = 0; i < erts_async_max_threads; i++) { ErtsAsyncQ *aq = async_q(i); -#ifdef ETHR_HAVE_THREAD_NAMES - sprintf(thr_opts.name, "async_%d", i+1); -#endif + erts_snprintf(thr_opts.name, 16, "async_%d", i+1); erts_thr_create(&aq->thr_id, async_main, (void*) aq, &thr_opts); } -#ifdef ETHR_HAVE_THREAD_NAMES - free(thr_opts.name); -#endif /* Wait for async threads to initialize... */ erts_mtx_lock(&async->init.data.mtx); diff --git a/erts/emulator/beam/erl_bif_binary.c b/erts/emulator/beam/erl_bif_binary.c index 3bf78adce7..bd0d7c71cc 100644 --- a/erts/emulator/beam/erl_bif_binary.c +++ b/erts/emulator/beam/erl_bif_binary.c @@ -32,7 +32,9 @@ #include "global.h" #include "erl_process.h" #include "error.h" +#define ERL_WANT_HIPE_BIF_WRAPPER__ #include "bif.h" +#undef ERL_WANT_HIPE_BIF_WRAPPER__ #include "big.h" #include "erl_binary.h" #include "erl_bits.h" @@ -2424,8 +2426,6 @@ static BIF_RETTYPE do_binary_copy(Process *p, Eterm bin, Eterm en) } cbs->result = erts_bin_nrml_alloc(target_size); /* Always offheap if trapping */ - cbs->result->flags = 0; - cbs->result->orig_size = target_size; erts_refc_init(&(cbs->result->refc), 1); t = (byte *) cbs->result->orig_bytes; /* No offset or anything */ pos = 0; diff --git a/erts/emulator/beam/erl_bif_guard.c b/erts/emulator/beam/erl_bif_guard.c index bbd8aa31d9..e7d84ebda1 100644 --- a/erts/emulator/beam/erl_bif_guard.c +++ b/erts/emulator/beam/erl_bif_guard.c @@ -459,23 +459,25 @@ 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); - } + Eterm* hp; + Uint size; + if (is_flatmap(arg)) { + flatmap_t *mp = (flatmap_t*)flatmap_val(arg); + size = flatmap_get_size(mp); + } else if (is_hashmap(arg)) { + size = hashmap_size(arg); } else { BIF_ERROR(p, BADARG); } + if (IS_USMALL(0, size)) { + return make_small(size); + } + 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); } Eterm erts_gc_abs_1(Process* p, Eterm* reg, Uint live) diff --git a/erts/emulator/beam/erl_bif_info.c b/erts/emulator/beam/erl_bif_info.c index b90362d82c..d750e34be3 100644 --- a/erts/emulator/beam/erl_bif_info.c +++ b/erts/emulator/beam/erl_bif_info.c @@ -116,6 +116,9 @@ static char erts_system_version[] = ("Erlang/OTP " ERLANG_OTP_RELEASE #ifdef ERTS_ENABLE_LOCK_COUNT " [lock-counting]" #endif +#ifdef ERTS_OPCODE_COUNTER_SUPPORT + " [instruction-counting]" +#endif #ifdef PURIFY " [purify-compiled]" #endif @@ -2301,7 +2304,7 @@ BIF_RETTYPE system_info_1(BIF_ALIST_1) for (i = num_instructions-1; i >= 0; i--) { res = erts_bld_cons(hpp, hszp, erts_bld_tuple(hpp, hszp, 2, - erts_atom_put(opc[i].name, + erts_atom_put((byte *)opc[i].name, strlen(opc[i].name), ERTS_ATOM_ENC_LATIN1, 1), @@ -3897,6 +3900,13 @@ BIF_RETTYPE erts_debug_set_internal_state_2(BIF_ALIST_2) } } } + else if (ERTS_IS_ATOM_STR("broken_halt", BIF_ARG_1)) { + /* Ugly ugly code used by bif_SUITE:erlang_halt/1 */ +#if defined(ERTS_HAVE_TRY_CATCH) + erts_get_scheduler_data()->run_queue = NULL; +#endif + erl_exit(ERTS_DUMP_EXIT, "%T", BIF_ARG_2); + } } BIF_ERROR(BIF_P, BADARG); diff --git a/erts/emulator/beam/erl_bif_trace.c b/erts/emulator/beam/erl_bif_trace.c index 06fbbea123..f5e582b1c5 100644 --- a/erts/emulator/beam/erl_bif_trace.c +++ b/erts/emulator/beam/erl_bif_trace.c @@ -651,7 +651,7 @@ Eterm trace_3(BIF_ALIST_3) if (pid_spec == am_all) { if (on) { if (!erts_cpu_timestamp) { -#ifdef HAVE_CLOCK_GETTIME +#ifdef HAVE_CLOCK_GETTIME_CPU_TIME /* Perhaps clock_gettime was found during config on a different machine than this. We check @@ -678,7 +678,7 @@ Eterm trace_3(BIF_ALIST_3) if (erts_start_now_cpu() < 0) { goto error; } -#endif /* HAVE_CLOCK_GETTIME */ +#endif /* HAVE_CLOCK_GETTIME_CPU_TIME */ erts_cpu_timestamp = !0; } } diff --git a/erts/emulator/beam/erl_binary.h b/erts/emulator/beam/erl_binary.h index 06dfeb1260..8d264d166e 100644 --- a/erts/emulator/beam/erl_binary.h +++ b/erts/emulator/beam/erl_binary.h @@ -231,41 +231,58 @@ erts_free_aligned_binary_bytes(byte* buf) # define CHICKEN_PAD (sizeof(void*) - 1) #endif +/* Caller must initialize 'refc' +*/ ERTS_GLB_INLINE Binary * erts_bin_drv_alloc_fnf(Uint size) { Uint bsize = ERTS_SIZEOF_Binary(size) + CHICKEN_PAD; - void *res; + Binary *res; + if (bsize < size) /* overflow */ return NULL; res = erts_alloc_fnf(ERTS_ALC_T_DRV_BINARY, bsize); ERTS_CHK_BIN_ALIGNMENT(res); - return (Binary *) res; + if (res) { + res->orig_size = size; + res->flags = BIN_FLAG_DRV; + } + return res; } +/* Caller must initialize 'refc' +*/ ERTS_GLB_INLINE Binary * erts_bin_drv_alloc(Uint size) { Uint bsize = ERTS_SIZEOF_Binary(size) + CHICKEN_PAD; - void *res; + Binary *res; + if (bsize < size) /* overflow */ erts_alloc_enomem(ERTS_ALC_T_DRV_BINARY, size); res = erts_alloc(ERTS_ALC_T_DRV_BINARY, bsize); ERTS_CHK_BIN_ALIGNMENT(res); - return (Binary *) res; + res->orig_size = size; + res->flags = BIN_FLAG_DRV; + return res; } +/* Caller must initialize 'refc' +*/ ERTS_GLB_INLINE Binary * erts_bin_nrml_alloc(Uint size) { Uint bsize = ERTS_SIZEOF_Binary(size) + CHICKEN_PAD; - void *res; + Binary *res; + if (bsize < size) /* overflow */ erts_alloc_enomem(ERTS_ALC_T_BINARY, size); res = erts_alloc(ERTS_ALC_T_BINARY, bsize); ERTS_CHK_BIN_ALIGNMENT(res); - return (Binary *) res; + res->orig_size = size; + res->flags = 0; + return res; } ERTS_GLB_INLINE Binary * @@ -280,6 +297,8 @@ erts_bin_realloc_fnf(Binary *bp, Uint size) return NULL; nbp = erts_realloc_fnf(type, (void *) bp, bsize); ERTS_CHK_BIN_ALIGNMENT(nbp); + if (nbp) + nbp->orig_size = size; return nbp; } @@ -297,6 +316,7 @@ erts_bin_realloc(Binary *bp, Uint size) if (!nbp) erts_realloc_enomem(type, bp, bsize); ERTS_CHK_BIN_ALIGNMENT(nbp); + nbp->orig_size = size; return nbp; } @@ -329,4 +349,4 @@ erts_create_magic_binary(Uint size, void (*destructor)(Binary *)) #endif /* #if ERTS_GLB_INLINE_INCL_FUNC_DEF */ -#endif +#endif /* !__ERL_BINARY_H */ diff --git a/erts/emulator/beam/erl_bits.c b/erts/emulator/beam/erl_bits.c index 73765772c8..5cc0a23dc9 100644 --- a/erts/emulator/beam/erl_bits.c +++ b/erts/emulator/beam/erl_bits.c @@ -403,7 +403,10 @@ erts_bs_get_integer_2(Process *p, Uint num_bits, unsigned flags, ErlBinMatchBuff words_needed = 1+WSIZE(bytes); hp = HeapOnlyAlloc(p, words_needed); res = bytes_to_big(LSB, bytes, sgn, hp); - if (is_small(res)) { + if (is_nil(res)) { + p->htop = hp; + res = THE_NON_VALUE; + } else if (is_small(res)) { p->htop = hp; } else if ((actual = bignum_header_arity(*hp)+1) < words_needed) { p->htop = hp + actual; @@ -1299,7 +1302,6 @@ erts_bs_append(Process* c_p, Eterm* reg, Uint live, Eterm build_size_term, if (binp->orig_size < pb->size) { Uint new_size = 2*pb->size; binp = erts_bin_realloc(binp, new_size); - binp->orig_size = new_size; pb->val = binp; pb->bytes = (byte *) binp->orig_bytes; } @@ -1371,8 +1373,6 @@ erts_bs_append(Process* c_p, Eterm* reg, Uint live, Eterm build_size_term, * Allocate the binary data struct itself. */ bptr = erts_bin_nrml_alloc(bin_size); - bptr->flags = 0; - bptr->orig_size = bin_size; erts_refc_init(&bptr->refc, 1); erts_current_bin = (byte *) bptr->orig_bytes; @@ -1475,7 +1475,6 @@ erts_bs_private_append(Process* p, Eterm bin, Eterm build_size_term, Uint unit) * is safe to reallocate it. */ binp = erts_bin_realloc(binp, new_size); - binp->orig_size = new_size; pb->val = binp; pb->bytes = (byte *) binp->orig_bytes; } else { @@ -1488,8 +1487,6 @@ erts_bs_private_append(Process* p, Eterm bin, Eterm build_size_term, Uint unit) * binary and copy the contents of the old binary into it. */ Binary* bptr = erts_bin_nrml_alloc(new_size); - bptr->flags = 0; - bptr->orig_size = new_size; erts_refc_init(&bptr->refc, 1); sys_memcpy(bptr->orig_bytes, binp->orig_bytes, binp->orig_size); pb->flags |= PB_IS_WRITABLE | PB_ACTIVE_WRITER; @@ -1537,8 +1534,6 @@ erts_bs_init_writable(Process* p, Eterm sz) * Allocate the binary data struct itself. */ bptr = erts_bin_nrml_alloc(bin_size); - bptr->flags = 0; - bptr->orig_size = bin_size; erts_refc_init(&bptr->refc, 1); /* @@ -1585,9 +1580,7 @@ erts_emasculate_writable_binary(ProcBin* pb) /* Our allocators are 8 byte aligned, i.e., shrinking with less than 8 bytes will have no real effect */ if (unused >= 8) { - Uint new_size = pb->size; binp = erts_bin_realloc(binp, pb->size); - binp->orig_size = new_size; pb->val = binp; pb->bytes = (byte *) binp->orig_bytes; } diff --git a/erts/emulator/beam/erl_db.c b/erts/emulator/beam/erl_db.c index 8f246ffa07..fff892ae54 100644 --- a/erts/emulator/beam/erl_db.c +++ b/erts/emulator/beam/erl_db.c @@ -753,6 +753,31 @@ BIF_RETTYPE ets_prev_2(BIF_ALIST_2) BIF_RET(ret); } +/* +** take(Tab, Key) +*/ +BIF_RETTYPE ets_take_2(BIF_ALIST_2) +{ + DbTable* tb; +#ifdef DEBUG + int cret; +#endif + Eterm ret; + CHECK_TABLES(); + + tb = db_get_table(BIF_P, BIF_ARG_1, DB_WRITE, LCK_WRITE_REC); + if (!tb) { + BIF_ERROR(BIF_P, BADARG); + } +#ifdef DEBUG + cret = +#endif + tb->common.meth->db_take(BIF_P, tb, BIF_ARG_2, &ret); + ASSERT(cret == DB_ERROR_NONE); + db_unlock(tb, LCK_WRITE_REC); + BIF_RET(ret); +} + /* ** update_element(Tab, Key, {Pos, Value}) ** update_element(Tab, Key, [{Pos, Value}]) @@ -780,7 +805,7 @@ BIF_RETTYPE ets_update_element_3(BIF_ALIST_3) list = BIF_ARG_3; } - if (!tb->common.meth->db_lookup_dbterm(tb, BIF_ARG_2, &handle)) { + if (!tb->common.meth->db_lookup_dbterm(BIF_P, tb, BIF_ARG_2, THE_NON_VALUE, &handle)) { cret = DB_ERROR_BADKEY; goto bail_out; } @@ -819,7 +844,7 @@ BIF_RETTYPE ets_update_element_3(BIF_ALIST_3) } finalize: - tb->common.meth->db_finalize_dbterm(&handle); + tb->common.meth->db_finalize_dbterm(cret, &handle); bail_out: UnUseTmpHeap(2,BIF_P); @@ -838,14 +863,8 @@ bail_out: } } -/* -** update_counter(Tab, Key, Incr) -** update_counter(Tab, Key, {Upop}) -** update_counter(Tab, Key, [{Upop}]) -** Upop = {Pos,Incr} | {Pos,Incr,Threshold,WarpTo} -** Returns new value(s) (integer or [integer]) -*/ -BIF_RETTYPE ets_update_counter_3(BIF_ALIST_3) +static BIF_RETTYPE +do_update_counter(Process *p, Eterm arg1, Eterm arg2, Eterm arg3, Eterm arg4) { DbTable* tb; int cret = DB_ERROR_BADITEM; @@ -855,7 +874,7 @@ BIF_RETTYPE ets_update_counter_3(BIF_ALIST_3) Eterm* ret_list_currp = NULL; Eterm* ret_list_prevp = NULL; Eterm iter; - DeclareTmpHeap(cell,5,BIF_P); + DeclareTmpHeap(cell, 5, p); Eterm *tuple = cell+2; DbUpdateHandle handle; Uint halloc_size = 0; /* overestimated heap usage */ @@ -863,28 +882,29 @@ BIF_RETTYPE ets_update_counter_3(BIF_ALIST_3) Eterm* hstart; Eterm* hend; - if ((tb = db_get_table(BIF_P, BIF_ARG_1, DB_WRITE, LCK_WRITE_REC)) == NULL) { - BIF_ERROR(BIF_P, BADARG); + if ((tb = db_get_table(p, arg1, DB_WRITE, LCK_WRITE_REC)) == NULL) { + BIF_ERROR(p, BADARG); } - UseTmpHeap(5,BIF_P); + UseTmpHeap(5, p); if (!(tb->common.status & (DB_SET | DB_ORDERED_SET))) { goto bail_out; } - if (is_integer(BIF_ARG_3)) { /* Incr */ - upop_list = CONS(cell, TUPLE2(tuple, make_small(tb->common.keypos+1), - BIF_ARG_3), NIL); + if (is_integer(arg3)) { /* Incr */ + upop_list = CONS(cell, + TUPLE2(tuple, make_small(tb->common.keypos+1), arg3), + NIL); } - else if (is_tuple(BIF_ARG_3)) { /* {Upop} */ - upop_list = CONS(cell, BIF_ARG_3, NIL); + else if (is_tuple(arg3)) { /* {Upop} */ + upop_list = CONS(cell, arg3, NIL); } else { /* [{Upop}] (probably) */ - upop_list = BIF_ARG_3; + upop_list = arg3; ret_list_prevp = &ret; } - if (!tb->common.meth->db_lookup_dbterm(tb, BIF_ARG_2, &handle)) { + if (!tb->common.meth->db_lookup_dbterm(p, tb, arg2, arg4, &handle)) { goto bail_out; /* key not found */ } @@ -957,13 +977,13 @@ BIF_RETTYPE ets_update_counter_3(BIF_ALIST_3) if (ret_list_prevp) { /* Prepare to return a list */ ret = NIL; halloc_size += list_size; - hstart = HAlloc(BIF_P, halloc_size); + hstart = HAlloc(p, halloc_size); ret_list_currp = hstart; htop = hstart + list_size; hend = hstart + halloc_size; } else { - hstart = htop = HAlloc(BIF_P, halloc_size); + hstart = htop = HAlloc(p, halloc_size); } hend = hstart + halloc_size; @@ -1010,26 +1030,54 @@ BIF_RETTYPE ets_update_counter_3(BIF_ALIST_3) (is_list(ret) && (list_val(ret)+list_size)==ret_list_currp)); ASSERT(htop <= hend); - HRelease(BIF_P,hend,htop); + HRelease(p, hend, htop); finalize: - tb->common.meth->db_finalize_dbterm(&handle); + tb->common.meth->db_finalize_dbterm(cret, &handle); bail_out: - UnUseTmpHeap(5,BIF_P); + UnUseTmpHeap(5, p); db_unlock(tb, LCK_WRITE_REC); switch (cret) { case DB_ERROR_NONE: BIF_RET(ret); case DB_ERROR_SYSRES: - BIF_ERROR(BIF_P, SYSTEM_LIMIT); + BIF_ERROR(p, SYSTEM_LIMIT); default: - BIF_ERROR(BIF_P, BADARG); + BIF_ERROR(p, BADARG); break; } } +/* +** update_counter(Tab, Key, Incr) +** update_counter(Tab, Key, Upop) +** update_counter(Tab, Key, [{Upop}]) +** Upop = {Pos,Incr} | {Pos,Incr,Threshold,WarpTo} +** Returns new value(s) (integer or [integer]) +*/ +BIF_RETTYPE ets_update_counter_3(BIF_ALIST_3) +{ + return do_update_counter(BIF_P, BIF_ARG_1, BIF_ARG_2, BIF_ARG_3, THE_NON_VALUE); +} + +/* +** update_counter(Tab, Key, Incr, Default) +** update_counter(Tab, Key, Upop, Default) +** update_counter(Tab, Key, [{Upop}], Default) +** Upop = {Pos,Incr} | {Pos,Incr,Threshold,WarpTo} +** Returns new value(s) (integer or [integer]) +*/ +BIF_RETTYPE ets_update_counter_4(BIF_ALIST_4) +{ + if (is_not_tuple(BIF_ARG_4)) { + BIF_ERROR(BIF_P, BADARG); + } + return do_update_counter(BIF_P, BIF_ARG_1, BIF_ARG_2, BIF_ARG_3, BIF_ARG_4); +} + + /* ** The put BIF */ @@ -2643,7 +2691,9 @@ BIF_RETTYPE ets_match_object_3(BIF_ALIST_3) BIF_RETTYPE ets_info_1(BIF_ALIST_1) { static Eterm fields[] = {am_protection, am_keypos, am_type, am_named_table, - am_node, am_size, am_name, am_heir, am_owner, am_memory, am_compressed}; + am_node, am_size, am_name, am_heir, am_owner, am_memory, am_compressed, + am_write_concurrency, + am_read_concurrency}; Eterm results[sizeof(fields)/sizeof(Eterm)]; DbTable* tb; Eterm res; @@ -3670,6 +3720,10 @@ static Eterm table_info(Process* p, DbTable* tb, Eterm What) ret = am_protected; else if (tb->common.status & DB_PUBLIC) ret = am_public; + } else if (What == am_write_concurrency) { + ret = tb->common.status & DB_FINE_LOCKED ? am_true : am_false; + } else if (What == am_read_concurrency) { + ret = tb->common.status & DB_FREQ_READ ? am_true : am_false; } else if (What == am_name) { ret = tb->common.the_name; } else if (What == am_keypos) { @@ -3752,7 +3806,7 @@ static Eterm table_info(Process* p, DbTable* tb, Eterm What) avg, std_dev_real, std_dev_exp, make_small(stats.min_chain_len), make_small(stats.max_chain_len), - make_small(db_kept_items_hash(&tb->hash))); + make_small(stats.kept_items)); } else { ret = am_false; @@ -3774,6 +3828,11 @@ static void print_table(int to, void *to_arg, int show, DbTable* tb) + sizeof(Uint) - 1) / sizeof(Uint))); + erts_print(to, to_arg, "Type: %T\n", table_info(NULL, tb, am_type)); + erts_print(to, to_arg, "Protection: %T\n", table_info(NULL, tb, am_protection)); + erts_print(to, to_arg, "Compressed: %T\n", table_info(NULL, tb, am_compressed)); + erts_print(to, to_arg, "Write Concurrency: %T\n", table_info(NULL, tb, am_write_concurrency)); + erts_print(to, to_arg, "Read Concurrency: %T\n", table_info(NULL, tb, am_read_concurrency)); } void db_info(int to, void *to_arg, int show) /* Called by break handler */ diff --git a/erts/emulator/beam/erl_db_hash.c b/erts/emulator/beam/erl_db_hash.c index 06dac8f161..045c8ae135 100644 --- a/erts/emulator/beam/erl_db_hash.c +++ b/erts/emulator/beam/erl_db_hash.c @@ -174,7 +174,7 @@ static ERTS_INLINE void add_fixed_deletion(DbTableHash* tb, int ix) /* optimised version of make_hash (normal case? atomic key) */ #define MAKE_HASH(term) \ ((is_atom(term) ? (atom_tab(atom_val(term))->slot.bucket.hvalue) : \ - make_hash2(term)) % MAX_HASH) + make_internal_hash(term)) % MAX_HASH) #ifdef ERTS_SMP # define DB_HASH_LOCK_MASK (DB_HASH_LOCK_CNT-1) @@ -382,7 +382,7 @@ static HashDbTerm* search_list(DbTableHash* tb, Eterm key, static void shrink(DbTableHash* tb, int nactive); static void grow(DbTableHash* tb, int nactive); static Eterm build_term_list(Process* p, HashDbTerm* ptr1, HashDbTerm* ptr2, - DbTableHash*); + Uint sz, DbTableHash*); static int analyze_pattern(DbTableHash *tb, Eterm pattern, struct mp_info *mpi); @@ -426,6 +426,7 @@ static int db_select_count_continue_hash(Process *p, DbTable *tbl, static int db_select_delete_continue_hash(Process *p, DbTable *tbl, Eterm continuation, Eterm *ret); +static int db_take_hash(Process *, DbTable *, Eterm, Eterm *); static void db_print_hash(int to, void *to_arg, int show, @@ -443,8 +444,11 @@ static int db_delete_all_objects_hash(Process* p, DbTable* tbl); #ifdef HARDDEBUG static void db_check_table_hash(DbTableHash *tb); #endif -static int db_lookup_dbterm_hash(DbTable *tbl, Eterm key, DbUpdateHandle* handle); -static void db_finalize_dbterm_hash(DbUpdateHandle* handle); +static int +db_lookup_dbterm_hash(Process *p, DbTable *tbl, Eterm key, Eterm obj, + DbUpdateHandle* handle); +static void +db_finalize_dbterm_hash(int cret, DbUpdateHandle* handle); static ERTS_INLINE void try_shrink(DbTableHash* tb) { @@ -536,6 +540,7 @@ DbTableMethod db_hash = db_select_delete_continue_hash, db_select_count_hash, db_select_count_continue_hash, + db_take_hash, db_delete_all_objects_hash, db_free_table_hash, db_free_table_continue_hash, @@ -646,25 +651,6 @@ restart: /* ToDo: Maybe try grow/shrink the table as well */ } -/* Only used by tests -*/ -Uint db_kept_items_hash(DbTableHash *tb) -{ - Uint kept_items = 0; - Uint ix = 0; - erts_smp_rwmtx_t* lck = RLOCK_HASH(tb,ix); - HashDbTerm* b; - do { - for (b = BUCKET(tb, ix); b != NULL; b = b->next) { - if (b->hvalue == INVALID_HASH) { - ++kept_items; - } - } - ix = next_slot(tb, ix, &lck); - }while (ix); - return kept_items; -} - int db_create_hash(Process *p, DbTable *tbl) { DbTableHash *tb = &tbl->hash; @@ -879,34 +865,49 @@ Ldone: return ret; } +static Eterm +get_term_list(Process *p, DbTableHash *tb, Eterm key, HashValue hval, + HashDbTerm *b1, HashDbTerm **bend) +{ + HashDbTerm* b2 = b1->next; + Eterm copy; + Uint sz = b1->dbterm.size + 2; + + if (tb->common.status & (DB_BAG | DB_DUPLICATE_BAG)) { + while (b2 && has_key(tb, b2, key, hval)) { + if (b2->hvalue != INVALID_HASH) + sz += b2->dbterm.size + 2; + + b2 = b2->next; + } + } + copy = build_term_list(p, b1, b2, sz, tb); + CHECK_TABLES(); + if (bend) { + *bend = b2; + } + return copy; +} + int db_get_hash(Process *p, DbTable *tbl, Eterm key, Eterm *ret) { DbTableHash *tb = &tbl->hash; HashValue hval; int ix; - HashDbTerm* b1; + HashDbTerm* b; erts_smp_rwmtx_t* lck; hval = MAKE_HASH(key); lck = RLOCK_HASH(tb,hval); ix = hash_to_ix(tb, hval); - b1 = BUCKET(tb, ix); - - while(b1 != 0) { - if (has_live_key(tb,b1,key,hval)) { - HashDbTerm* b2 = b1->next; - Eterm copy; + b = BUCKET(tb, ix); - if (tb->common.status & (DB_BAG | DB_DUPLICATE_BAG)) { - while(b2 != NULL && has_key(tb,b2,key,hval)) - b2 = b2->next; - } - copy = build_term_list(p, b1, b2, tb); - CHECK_TABLES(); - *ret = copy; + while(b != 0) { + if (has_live_key(tb, b, key, hval)) { + *ret = get_term_list(p, tb, key, hval, b, NULL); goto done; } - b1 = b1->next; + b = b->next; } *ret = NIL; done: @@ -1240,7 +1241,7 @@ static int db_slot_hash(Process *p, DbTable *tbl, Eterm slot_term, Eterm *ret) lck = RLOCK_HASH(tb, slot); nactive = NACTIVE(tb); if (slot < nactive) { - *ret = build_term_list(p, BUCKET(tb, slot), 0, tb); + *ret = build_term_list(p, BUCKET(tb, slot), NULL, 0, tb); retval = DB_ERROR_NONE; } else if (slot == nactive) { @@ -2069,6 +2070,46 @@ trap: } +static int db_take_hash(Process *p, DbTable *tbl, Eterm key, Eterm *ret) +{ + DbTableHash *tb = &tbl->hash; + HashDbTerm **bp, *b; + HashValue hval = MAKE_HASH(key); + erts_smp_rwmtx_t *lck = WLOCK_HASH(tb, hval); + int ix = hash_to_ix(tb, hval); + int nitems_diff = 0; + + *ret = NIL; + for (bp = &BUCKET(tb, ix), b = *bp; b; bp = &b->next, b = b->next) { + if (has_live_key(tb, b, key, hval)) { + HashDbTerm *bend; + + *ret = get_term_list(p, tb, key, hval, b, &bend); + while (b != bend) { + --nitems_diff; + if (nitems_diff == -1 && IS_FIXED(tb)) { + /* Pseudo remove (no need to keep several of same key) */ + add_fixed_deletion(tb, ix); + bp = &b->next; + b->hvalue = INVALID_HASH; + b = b->next; + } else { + *bp = b->next; + free_term(tb, b); + b = *bp; + } + } + break; + } + } + WUNLOCK_HASH(lck); + if (nitems_diff) { + erts_smp_atomic_add_nob(&tb->common.nitems, nitems_diff); + try_shrink(tb); + } + return DB_ERROR_NONE; +} + /* ** Other interface routines (not directly coupled to one bif) */ @@ -2104,10 +2145,38 @@ int db_mark_all_deleted_hash(DbTable *tbl) static void db_print_hash(int to, void *to_arg, int show, DbTable *tbl) { DbTableHash *tb = &tbl->hash; + DbHashStats stats; int i; erts_print(to, to_arg, "Buckets: %d\n", NACTIVE(tb)); - + +#ifdef ERTS_SMP + i = tbl->common.is_thread_safe; + /* If crash dumping we set table to thread safe in order to + avoid taking any locks */ + if (ERTS_IS_CRASH_DUMPING) + tbl->common.is_thread_safe = 1; + + db_calc_stats_hash(&tbl->hash, &stats); + + tbl->common.is_thread_safe = i; +#else + db_calc_stats_hash(&tbl->hash, &stats); +#endif + + erts_print(to, to_arg, "Chain Length Avg: %f\n", stats.avg_chain_len); + erts_print(to, to_arg, "Chain Length Max: %d\n", stats.max_chain_len); + erts_print(to, to_arg, "Chain Length Min: %d\n", stats.min_chain_len); + erts_print(to, to_arg, "Chain Length Std Dev: %f\n", + stats.std_dev_chain_len); + erts_print(to, to_arg, "Chain Length Expected Std Dev: %f\n", + stats.std_dev_expected); + + if (IS_FIXED(tb)) + erts_print(to, to_arg, "Fixed: %d\n", stats.kept_items); + else + erts_print(to, to_arg, "Fixed: false\n"); + if (show) { for (i = 0; i < NACTIVE(tb); i++) { HashDbTerm* list = BUCKET(tb,i); @@ -2483,23 +2552,23 @@ static int free_seg(DbTableHash *tb, int free_records) ** Copy terms from ptr1 until ptr2 ** works for ptr1 == ptr2 == 0 => [] ** or ptr2 == 0 +** sz is either precalculated heap size or 0 if not known */ static Eterm build_term_list(Process* p, HashDbTerm* ptr1, HashDbTerm* ptr2, - DbTableHash* tb) + Uint sz, DbTableHash* tb) { - int sz = 0; HashDbTerm* ptr; Eterm list = NIL; Eterm copy; Eterm *hp, *hend; - ptr = ptr1; - while(ptr != ptr2) { - - if (ptr->hvalue != INVALID_HASH) - sz += ptr->dbterm.size + 2; - - ptr = ptr->next; + if (!sz) { + ptr = ptr1; + while(ptr != ptr2) { + if (ptr->hvalue != INVALID_HASH) + sz += ptr->dbterm.size + 2; + ptr = ptr->next; + } } hp = HAlloc(p, sz); @@ -2730,59 +2799,129 @@ static HashDbTerm* next(DbTableHash *tb, Uint *iptr, erts_smp_rwmtx_t** lck_ptr, return NULL; } -static int db_lookup_dbterm_hash(DbTable *tbl, Eterm key, DbUpdateHandle* handle) +static int +db_lookup_dbterm_hash(Process *p, DbTable *tbl, Eterm key, Eterm obj, + DbUpdateHandle* handle) { DbTableHash *tb = &tbl->hash; - HashDbTerm* b; - HashDbTerm** prevp; - int ix; HashValue hval; + HashDbTerm **bp, *b; erts_smp_rwmtx_t* lck; + int flags = 0; + + ASSERT(tb->common.status & DB_SET); hval = MAKE_HASH(key); - lck = WLOCK_HASH(tb,hval); - ix = hash_to_ix(tb, hval); - prevp = &BUCKET(tb, ix); - b = *prevp; + lck = WLOCK_HASH(tb, hval); + bp = &BUCKET(tb, hash_to_ix(tb, hval)); + b = *bp; - while (b != 0) { - if (has_live_key(tb,b,key,hval)) { - handle->tb = tbl; - handle->bp = (void**) prevp; - handle->dbterm = &b->dbterm; - handle->mustResize = 0; - handle->new_size = b->dbterm.size; - #if HALFWORD_HEAP - handle->abs_vec = NULL; - #endif - handle->lck = lck; - /* KEEP hval WLOCKED, db_finalize_dbterm_hash will WUNLOCK */ - return 1; - } - prevp = &b->next; - b = *prevp; + for (;;) { + if (b == NULL) { + break; + } + if (has_key(tb, b, key, hval)) { + if (b->hvalue != INVALID_HASH) { + goto Ldone; + } + break; + } + bp = &b->next; + b = *bp; } - WUNLOCK_HASH(lck); - return 0; + + if (obj == THE_NON_VALUE) { + WUNLOCK_HASH(lck); + return 0; + } + + { + Eterm *objp = tuple_val(obj); + int arity = arityval(*objp); + Eterm *htop, *hend; + + ASSERT(arity >= tb->common.keypos); + htop = HAlloc(p, arity + 1); + hend = htop + arity + 1; + sys_memcpy(htop, objp, sizeof(Eterm) * (arity + 1)); + htop[tb->common.keypos] = key; + obj = make_tuple(htop); + + if (b == NULL) { + HashDbTerm *q = new_dbterm(tb, obj); + + q->hvalue = hval; + q->next = NULL; + *bp = b = q; + + { + int nitems = erts_smp_atomic_inc_read_nob(&tb->common.nitems); + int nactive = NACTIVE(tb); + + if (nitems > nactive * (CHAIN_LEN + 1) && !IS_FIXED(tb)) { + grow(tb, nactive); + } + } + } else { + HashDbTerm *q, *next = b->next; + + ASSERT(b->hvalue == INVALID_HASH); + q = replace_dbterm(tb, b, obj); + q->next = next; + q->hvalue = hval; + *bp = b = q; + erts_smp_atomic_inc_nob(&tb->common.nitems); + } + + HRelease(p, hend, htop); + flags |= DB_NEW_OBJECT; + } + +Ldone: + handle->tb = tbl; + handle->bp = (void **)bp; + handle->dbterm = &b->dbterm; + handle->flags = flags; + handle->new_size = b->dbterm.size; +#if HALFWORD_HEAP + handle->abs_vec = NULL; +#endif + handle->lck = lck; + return 1; } /* Must be called after call to db_lookup_dbterm */ -static void db_finalize_dbterm_hash(DbUpdateHandle* handle) +static void +db_finalize_dbterm_hash(int cret, DbUpdateHandle* handle) { DbTable* tbl = handle->tb; - HashDbTerm* oldp = (HashDbTerm*) *(handle->bp); + DbTableHash *tb = &tbl->hash; + HashDbTerm **bp = (HashDbTerm **) handle->bp; + HashDbTerm *b = *bp; erts_smp_rwmtx_t* lck = (erts_smp_rwmtx_t*) handle->lck; - ERTS_SMP_LC_ASSERT(IS_HASH_WLOCKED(&tbl->hash,lck)); /* locked by db_lookup_dbterm_hash */ + ERTS_SMP_LC_ASSERT(IS_HASH_WLOCKED(tb, lck)); /* locked by db_lookup_dbterm_hash */ - ASSERT((&oldp->dbterm == handle->dbterm) == !(tbl->common.compress && handle->mustResize)); + ASSERT((&b->dbterm == handle->dbterm) == !(tb->common.compress && handle->flags & DB_MUST_RESIZE)); + + if (handle->flags & DB_NEW_OBJECT && cret != DB_ERROR_NONE) { + if (IS_FIXED(tb)) { + add_fixed_deletion(tb, hash_to_ix(tb, b->hvalue)); + b->hvalue = INVALID_HASH; + } else { + *bp = b->next; + free_term(tb, b); + } - if (handle->mustResize) { + WUNLOCK_HASH(lck); + erts_smp_atomic_dec_nob(&tb->common.nitems); + try_shrink(tb); + } else if (handle->flags & DB_MUST_RESIZE) { db_finalize_resize(handle, offsetof(HashDbTerm,dbterm)); WUNLOCK_HASH(lck); - free_term(&tbl->hash, oldp); + free_term(tb, b); } else { WUNLOCK_HASH(lck); @@ -2833,6 +2972,7 @@ void db_calc_stats_hash(DbTableHash* tb, DbHashStats* stats) erts_smp_rwmtx_t* lck; int sum = 0; int sq_sum = 0; + int kept_items = 0; int ix; int len; @@ -2844,6 +2984,8 @@ void db_calc_stats_hash(DbTableHash* tb, DbHashStats* stats) len = 0; for (b = BUCKET(tb,ix); b!=NULL; b=b->next) { len++; + if (b->hvalue == INVALID_HASH) + ++kept_items; } sum += len; sq_sum += len*len; @@ -2855,7 +2997,8 @@ void db_calc_stats_hash(DbTableHash* tb, DbHashStats* stats) stats->std_dev_chain_len = sqrt((sq_sum - stats->avg_chain_len*sum) / NACTIVE(tb)); /* Expected standard deviation from a good uniform hash function, ie binomial distribution (not taking the linear hashing into acount) */ - stats->std_dev_expected = sqrt(stats->avg_chain_len * (1 - 1.0/NACTIVE(tb))); + stats->std_dev_expected = sqrt(stats->avg_chain_len * (1 - 1.0/NACTIVE(tb))); + stats->kept_items = kept_items; } #ifdef HARDDEBUG diff --git a/erts/emulator/beam/erl_db_hash.h b/erts/emulator/beam/erl_db_hash.h index 908cec11d4..f12cd363b0 100644 --- a/erts/emulator/beam/erl_db_hash.h +++ b/erts/emulator/beam/erl_db_hash.h @@ -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 @@ -42,7 +42,7 @@ typedef struct hash_db_term { typedef struct db_table_hash_fine_locks { union { erts_smp_rwmtx_t lck; - byte _cache_line_alignment[64]; + byte _cache_line_alignment[ERTS_ALC_CACHE_LINE_ALIGN_SIZE(sizeof(erts_smp_rwmtx_t))]; }lck_vec[DB_HASH_LOCK_CNT]; } DbTableHashFineLocks; @@ -104,6 +104,7 @@ typedef struct { float std_dev_expected; int max_chain_len; int min_chain_len; + int kept_items; }DbHashStats; void db_calc_stats_hash(DbTableHash* tb, DbHashStats*); diff --git a/erts/emulator/beam/erl_db_tree.c b/erts/emulator/beam/erl_db_tree.c index a62a83a928..577da35b75 100644 --- a/erts/emulator/beam/erl_db_tree.c +++ b/erts/emulator/beam/erl_db_tree.c @@ -383,6 +383,7 @@ static int db_select_delete_tree(Process *p, DbTable *tbl, Eterm pattern, Eterm *ret); static int db_select_delete_continue_tree(Process *p, DbTable *tbl, Eterm continuation, Eterm *ret); +static int db_take_tree(Process *, DbTable *, Eterm, Eterm *); static void db_print_tree(int to, void *to_arg, int show, DbTable *tbl); static int db_free_table_tree(DbTable *tbl); @@ -398,8 +399,11 @@ static int db_delete_all_objects_tree(Process* p, DbTable* tbl); #ifdef HARDDEBUG static void db_check_table_tree(DbTable *tbl); #endif -static int db_lookup_dbterm_tree(DbTable *, Eterm key, DbUpdateHandle*); -static void db_finalize_dbterm_tree(DbUpdateHandle*); +static int +db_lookup_dbterm_tree(Process *, DbTable *, Eterm key, Eterm obj, + DbUpdateHandle*); +static void +db_finalize_dbterm_tree(int cret, DbUpdateHandle *); /* ** Static variables @@ -431,6 +435,7 @@ DbTableMethod db_tree = db_select_delete_continue_tree, db_select_count_tree, db_select_count_continue_tree, + db_take_tree, db_delete_all_objects_tree, db_free_table_tree, db_free_table_continue_tree, @@ -1722,6 +1727,28 @@ static int db_select_delete_tree(Process *p, DbTable *tbl, } +static int db_take_tree(Process *p, DbTable *tbl, Eterm key, Eterm *ret) +{ + DbTableTree *tb = &tbl->tree; + TreeDbTerm *this; + + *ret = NIL; + this = linkout_tree(tb, key, NULL); + if (this) { + Eterm copy, *hp, *hend; + + hp = HAlloc(p, this->dbterm.size + 2); + hend = hp + this->dbterm.size + 2; + copy = db_copy_object_from_ets(&tb->common, + &this->dbterm, &hp, &MSO(p)); + *ret = CONS(hp, copy, NIL); + hp += 2; + HRelease(p, hend, hp); + free_term(tb, this); + } + return DB_ERROR_NONE; +} + /* ** Other interface routines (not directly coupled to one bif) */ @@ -2522,16 +2549,43 @@ static TreeDbTerm **find_node2(DbTableTree *tb, Eterm key) return this; } -static int db_lookup_dbterm_tree(DbTable *tbl, Eterm key, DbUpdateHandle* handle) +static int +db_lookup_dbterm_tree(Process *p, DbTable *tbl, Eterm key, Eterm obj, + DbUpdateHandle* handle) { DbTableTree *tb = &tbl->tree; TreeDbTerm **pp = find_node2(tb, key); - - if (pp == NULL) return 0; + int flags = 0; + + if (pp == NULL) { + if (obj == THE_NON_VALUE) { + return 0; + } else { + Eterm *objp = tuple_val(obj); + int arity = arityval(*objp); + Eterm *htop, *hend; + + ASSERT(arity >= tb->common.keypos); + htop = HAlloc(p, arity + 1); + hend = htop + arity + 1; + sys_memcpy(htop, objp, sizeof(Eterm) * (arity + 1)); + htop[tb->common.keypos] = key; + obj = make_tuple(htop); + + if (db_put_tree(tbl, obj, 1) != DB_ERROR_NONE) { + return 0; + } + + pp = find_node2(tb, key); + ASSERT(pp != NULL); + HRelease(p, hend, htop); + flags |= DB_NEW_OBJECT; + } + } handle->tb = tbl; handle->dbterm = &(*pp)->dbterm; - handle->mustResize = 0; + handle->flags = flags; handle->bp = (void**) pp; handle->new_size = (*pp)->dbterm.size; #if HALFWORD_HEAP @@ -2540,15 +2594,21 @@ static int db_lookup_dbterm_tree(DbTable *tbl, Eterm key, DbUpdateHandle* handle return 1; } -static void db_finalize_dbterm_tree(DbUpdateHandle* handle) +static void +db_finalize_dbterm_tree(int cret, DbUpdateHandle *handle) { - if (handle->mustResize) { - TreeDbTerm* oldp = (TreeDbTerm*) *handle->bp; + DbTable *tbl = handle->tb; + DbTableTree *tb = &tbl->tree; + TreeDbTerm *bp = (TreeDbTerm *) *handle->bp; + if (handle->flags & DB_NEW_OBJECT && cret != DB_ERROR_NONE) { + Eterm ret; + db_erase_tree(tbl, GETKEY(tb, bp->dbterm.tpl), &ret); + } else if (handle->flags & DB_MUST_RESIZE) { db_finalize_resize(handle, offsetof(TreeDbTerm,dbterm)); - reset_static_stack(&handle->tb->tree); + reset_static_stack(tb); - free_term(&handle->tb->tree, oldp); + free_term(tb, bp); } #ifdef DEBUG handle->dbterm = 0; diff --git a/erts/emulator/beam/erl_db_util.c b/erts/emulator/beam/erl_db_util.c index 3927615e04..9d699d4b22 100644 --- a/erts/emulator/beam/erl_db_util.c +++ b/erts/emulator/beam/erl_db_util.c @@ -198,11 +198,6 @@ set_match_trace(Process *tracee_p, Eterm fail_term, Eterm tracer, return ret; } - -/* Type checking... */ - -#define BOXED_IS_TUPLE(Boxed) is_arity_value(*boxed_val((Boxed))) - /* ** ** Types and enum's (compiled matches) @@ -218,7 +213,9 @@ typedef enum { matchTuple, matchPushT, matchPushL, + matchPushM, matchPop, + matchSwap, matchBind, matchCmp, matchEqBin, @@ -227,11 +224,15 @@ typedef enum { matchEqRef, matchEq, matchList, + matchMap, + matchKey, matchSkip, matchPushC, matchConsA, /* Car is below Cdr */ matchConsB, /* Cdr is below Car (unusual) */ matchMkTuple, + matchMkFlatMap, + matchMkHashMap, matchCall0, matchCall1, matchCall2, @@ -856,6 +857,13 @@ static int match_compact(ErlHeapFragment *expr, DMCErrInfo *err_info); static Uint my_size_object(Eterm t); static Eterm my_copy_struct(Eterm t, Eterm **hp, ErlOffHeap* off_heap); +/* Guard subroutines */ +static void +dmc_rearrange_constants(DMCContext *context, DMC_STACK_TYPE(UWord) *text, + int textpos, Eterm *p, Uint nelems); +static DMCRet +dmc_array(DMCContext *context, DMCHeap *heap, DMC_STACK_TYPE(UWord) *text, + Eterm *p, Uint nelems, int *constant); /* Guard compilation */ static void do_emit_constant(DMCContext *context, DMC_STACK_TYPE(UWord) *text, Eterm t); @@ -869,6 +877,9 @@ static DMCRet dmc_tuple(DMCContext *context, DMC_STACK_TYPE(UWord) *text, Eterm t, int *constant); +static DMCRet +dmc_map(DMCContext *context, DMCHeap *heap, DMC_STACK_TYPE(UWord) *text, + Eterm t, int *constant); static DMCRet dmc_variable(DMCContext *context, DMCHeap *heap, DMC_STACK_TYPE(UWord) *text, @@ -888,12 +899,14 @@ static DMCRet compile_guard_expr(DMCContext *context, DMCHeap *heap, DMC_STACK_TYPE(UWord) *text, Eterm t); -/* match expression subroutine */ +/* match expression subroutines */ static DMCRet dmc_one_term(DMCContext *context, DMCHeap *heap, DMC_STACK_TYPE(Eterm) *stack, DMC_STACK_TYPE(UWord) *text, Eterm c); +static Eterm +dmc_private_copy(DMCContext *context, Eterm c); #ifdef DMC_DEBUG @@ -1364,7 +1377,112 @@ restart: for (;;) { switch (t & _TAG_PRIMARY_MASK) { case TAG_PRIMARY_BOXED: - if (!BOXED_IS_TUPLE(t)) { + if (is_flatmap(t)) { + num_iters = flatmap_get_size(flatmap_val(t)); + if (!structure_checked) { + DMC_PUSH(text, matchMap); + DMC_PUSH(text, num_iters); + } + structure_checked = 0; + for (i = 0; i < num_iters; ++i) { + Eterm key = flatmap_get_keys(flatmap_val(t))[i]; + if (db_is_variable(key) >= 0) { + if (context.err_info) { + add_dmc_err(context.err_info, + "Variable found in map key.", + -1, 0UL, dmcError); + } + goto error; + } else if (key == am_Underscore) { + if (context.err_info) { + add_dmc_err(context.err_info, + "Underscore found in map key.", + -1, 0UL, dmcError); + } + goto error; + } + DMC_PUSH(text, matchKey); + DMC_PUSH(text, dmc_private_copy(&context, key)); + { + int old_stack = ++(context.stack_used); + Eterm value = flatmap_get_values(flatmap_val(t))[i]; + res = dmc_one_term(&context, &heap, &stack, &text, + value); + ASSERT(res != retFail); + if (res == retRestart) { + goto restart; + } + if (old_stack != context.stack_used) { + ASSERT(old_stack + 1 == context.stack_used); + DMC_PUSH(text, matchSwap); + } + if (context.stack_used > context.stack_need) { + context.stack_need = context.stack_used; + } + DMC_PUSH(text, matchPop); + --(context.stack_used); + } + } + break; + } + if (is_hashmap(t)) { + DECLARE_WSTACK(wstack); + Eterm *kv; + num_iters = hashmap_size(t); + if (!structure_checked) { + DMC_PUSH(text, matchMap); + DMC_PUSH(text, num_iters); + } + structure_checked = 0; + + hashmap_iterator_init(&wstack, t, 0); + + while ((kv=hashmap_iterator_next(&wstack)) != NULL) { + Eterm key = CAR(kv); + Eterm value = CDR(kv); + if (db_is_variable(key) >= 0) { + if (context.err_info) { + add_dmc_err(context.err_info, + "Variable found in map key.", + -1, 0UL, dmcError); + } + DESTROY_WSTACK(wstack); + goto error; + } else if (key == am_Underscore) { + if (context.err_info) { + add_dmc_err(context.err_info, + "Underscore found in map key.", + -1, 0UL, dmcError); + } + DESTROY_WSTACK(wstack); + goto error; + } + DMC_PUSH(text, matchKey); + DMC_PUSH(text, dmc_private_copy(&context, key)); + { + int old_stack = ++(context.stack_used); + res = dmc_one_term(&context, &heap, &stack, &text, + value); + ASSERT(res != retFail); + if (res == retRestart) { + DESTROY_WSTACK(wstack); + goto restart; + } + if (old_stack != context.stack_used) { + ASSERT(old_stack + 1 == context.stack_used); + DMC_PUSH(text, matchSwap); + } + if (context.stack_used > context.stack_need) { + context.stack_need = context.stack_used; + } + DMC_PUSH(text, matchPop); + --(context.stack_used); + } + } + DESTROY_WSTACK(wstack); + break; + } + if (!is_tuple(t)) { goto simple_term; } num_iters = arityval(*tuple_val(t)); @@ -1715,10 +1833,8 @@ Eterm db_prog_match(Process *c_p, Binary *bprog, Uint32 *return_flags) { MatchProg *prog = Binary2MatchProg(bprog); - Eterm *ep; - Eterm *tp; + const Eterm *ep, *tp, **sp; Eterm t; - Eterm **sp; Eterm *esp; MatchVariable* variables; BeamInstr *cp; @@ -1808,7 +1924,7 @@ Eterm db_prog_match(Process *c_p, Binary *bprog, restart: ep = &term; esp = (Eterm*)((char*)mpsp->u.heap + prog->stack_offset); - sp = (Eterm **) esp; + sp = (const Eterm **)esp; ret = am_true; do_catch = 0; fail_label = -1; @@ -1887,9 +2003,57 @@ restart: *sp++ = list_val_rel(*ep,base); ++ep; break; + case matchMap: + if (!is_map_rel(*ep, base)) { + FAIL(); + } + n = *pc++; + if (is_flatmap_rel(*ep,base)) { + if (flatmap_get_size(flatmap_val_rel(*ep, base)) < n) { + FAIL(); + } + } else { + ASSERT(is_hashmap_rel(*ep,base)); + if (hashmap_size_rel(*ep, base) < n) { + FAIL(); + } + } + ep = flatmap_val_rel(*ep, base); + break; + case matchPushM: + if (!is_map_rel(*ep, base)) { + FAIL(); + } + n = *pc++; + if (is_flatmap_rel(*ep,base)) { + if (flatmap_get_size(flatmap_val_rel(*ep, base)) < n) { + FAIL(); + } + } else { + ASSERT(is_hashmap_rel(*ep,base)); + if (hashmap_size_rel(*ep, base) < n) { + FAIL(); + } + } + *sp++ = flatmap_val_rel(*ep++, base); + break; + case matchKey: + t = (Eterm) *pc++; + tp = erts_maps_get_rel(t, make_flatmap_rel(ep, base), base); + if (!tp) { + FAIL(); + } + *sp++ = ep; + ep = tp; + break; case matchPop: ep = *(--sp); break; + case matchSwap: + tp = sp[-1]; + sp[-1] = sp[-2]; + sp[-2] = tp; + break; case matchBind: n = *pc++; variables[n].term = *ep++; @@ -1987,6 +2151,38 @@ restart: } *esp++ = t; break; + case matchMkFlatMap: + n = *pc++; + ehp = HAllocX(build_proc, 1 + MAP_HEADER_SIZE + n, HEAP_XTRA); + t = *ehp++ = *--esp; + { + flatmap_t *m = (flatmap_t *)ehp; + m->thing_word = MAP_HEADER; + m->size = n; + m->keys = t; + } + t = make_flatmap(ehp); + ehp += MAP_HEADER_SIZE; + while (n--) { + *ehp++ = *--esp; + } + *esp++ = t; + break; + case matchMkHashMap: + n = *pc++; + esp -= 2*n; + ehp = HAllocX(build_proc, 2*n, HEAP_XTRA); + { + ErtsHeapFactory factory; + Uint ix; + factory.p = build_proc; + for (ix = 0; ix < 2*n; ix++){ + ehp[ix] = esp[ix]; + } + t = erts_hashmap_from_array(&factory, ehp, n, 0); + } + *esp++ = t; + break; case matchCall0: bif = (Eterm (*)(Process*, ...)) *pc++; t = (*bif)(build_proc, bif_args); @@ -2601,10 +2797,10 @@ Wterm db_do_read_element(DbUpdateHandle* handle, Sint position) } ASSERT(((DbTableCommon*)handle->tb)->compress); - ASSERT(!handle->mustResize); + ASSERT(!(handle->flags & DB_MUST_RESIZE)); handle->dbterm = db_alloc_tmp_uncompressed(&handle->tb->common, handle->dbterm); - handle->mustResize = 1; + handle->flags |= DB_MUST_RESIZE; return handle->dbterm->tpl[position]; } @@ -2637,11 +2833,11 @@ void db_do_update_element(DbUpdateHandle* handle, #endif return; } - if (!handle->mustResize) { + if (!(handle->flags & DB_MUST_RESIZE)) { if (handle->tb->common.compress) { handle->dbterm = db_alloc_tmp_uncompressed(&handle->tb->common, handle->dbterm); - handle->mustResize = 1; + handle->flags |= DB_MUST_RESIZE; oldval = handle->dbterm->tpl[position]; #if HALFWORD_HEAP old_base = NULL; @@ -2701,7 +2897,7 @@ both_size_set: /* write new value in old dbterm, finalize will make a flat copy */ handle->dbterm->tpl[position] = newval; - handle->mustResize = 1; + handle->flags |= DB_MUST_RESIZE; #if HALFWORD_HEAP if (old_base && newval_sz > 0) { @@ -3156,34 +3352,45 @@ int db_is_variable(Eterm obj) /* return 1 if obj contains a variable or underscore */ /* return 0 if obj is fully ground */ -int db_has_variable(Eterm obj) -{ - switch(obj & _TAG_PRIMARY_MASK) { - case TAG_PRIMARY_LIST: { - while (is_list(obj)) { - if (db_has_variable(CAR(list_val(obj)))) +int db_has_variable(Eterm node) { + DECLARE_ESTACK(s); + + ESTACK_PUSH(s,node); + while (!ESTACK_ISEMPTY(s)) { + node = ESTACK_POP(s); + switch(node & _TAG_PRIMARY_MASK) { + case TAG_PRIMARY_LIST: + while (is_list(node)) { + ESTACK_PUSH(s,CAR(list_val(node))); + node = CDR(list_val(node)); + } + ESTACK_PUSH(s,node); /* Non wellformed list or [] */ + break; + case TAG_PRIMARY_BOXED: + if (is_tuple(node)) { + Eterm *tuple = tuple_val(node); + int arity = arityval(*tuple); + while(arity--) { + ESTACK_PUSH(s,*(++tuple)); + } + } else if (is_flatmap(node)) { + Eterm *values = flatmap_get_values(flatmap_val(node)); + Uint size = flatmap_get_size(flatmap_val(node)); + ESTACK_PUSH(s, ((flatmap_t *) flatmap_val(node))->keys); + while (size--) { + ESTACK_PUSH(s, *(values++)); + } + } + break; + case TAG_PRIMARY_IMMED1: + if (node == am_Underscore || db_is_variable(node) >= 0) { + DESTROY_ESTACK(s); return 1; - obj = CDR(list_val(obj)); - } - return(db_has_variable(obj)); /* Non wellformed list or [] */ - } - case TAG_PRIMARY_BOXED: - if (!BOXED_IS_TUPLE(obj)) { - return 0; - } else { - Eterm *tuple = tuple_val(obj); - int arity = arityval(*tuple++); - while(arity--) { - if (db_has_variable(*tuple)) - return 1; - tuple++; } - return(0); + break; } - case TAG_PRIMARY_IMMED1: - if (obj == am_Underscore || db_is_variable(obj) >= 0) - return 1; } + DESTROY_ESTACK(s); return 0; } @@ -3243,11 +3450,9 @@ static DMCRet dmc_one_term(DMCContext *context, { Sint n; Eterm *hp; - ErlHeapFragment *tmp_mb; Uint sz, sz2, sz3; Uint i, j; - switch (c & _TAG_PRIMARY_MASK) { case TAG_PRIMARY_IMMED1: if ((n = db_is_variable(c)) >= 0) { /* variable */ @@ -3334,6 +3539,20 @@ static DMCRet dmc_one_term(DMCContext *context, DMC_PUSH(*text, n); DMC_PUSH(*stack, c); break; + case (_TAG_HEADER_MAP >> _TAG_PRIMARY_SIZE): + n = flatmap_get_size(flatmap_val(c)); + DMC_PUSH(*text, matchPushM); + ++(context->stack_used); + DMC_PUSH(*text, n); + DMC_PUSH(*stack, c); + break; + case (_TAG_HEADER_HASHMAP >> _TAG_PRIMARY_SIZE): + n = hashmap_size(c); + DMC_PUSH(*text, matchPushM); + ++(context->stack_used); + DMC_PUSH(*text, n); + DMC_PUSH(*stack, c); + break; case (_TAG_HEADER_REF >> _TAG_PRIMARY_SIZE): { Eterm* ref_val = internal_ref_val(c); @@ -3415,16 +3634,8 @@ static DMCRet dmc_one_term(DMCContext *context, #endif break; default: /* BINARY, FUN, VECTOR, or EXTERNAL */ - /* - ** Make a private copy... - */ - n = size_object(c); - tmp_mb = new_message_buffer(n); - hp = tmp_mb->mem; DMC_PUSH(*text, matchEqBin); - DMC_PUSH(*text, copy_struct(c, n, &hp, &(tmp_mb->off_heap))); - tmp_mb->next = context->save; - context->save = tmp_mb; + DMC_PUSH(*text, dmc_private_copy(context, c)); break; } break; @@ -3437,6 +3648,22 @@ static DMCRet dmc_one_term(DMCContext *context, } /* +** Make a private copy of a term in a context. +*/ + +static Eterm +dmc_private_copy(DMCContext *context, Eterm c) +{ + Uint n = size_object(c); + ErlHeapFragment *tmp_mb = new_message_buffer(n); + Eterm *hp = tmp_mb->mem; + Eterm copy = copy_struct(c, n, &hp, &(tmp_mb->off_heap)); + tmp_mb->next = context->save; + context->save = tmp_mb; + return copy; +} + +/* ** Match guard compilation */ @@ -3527,57 +3754,78 @@ static DMCRet dmc_list(DMCContext *context, return retOk; } -static DMCRet dmc_tuple(DMCContext *context, - DMCHeap *heap, - DMC_STACK_TYPE(UWord) *text, - Eterm t, - int *constant) +static void +dmc_rearrange_constants(DMCContext *context, DMC_STACK_TYPE(UWord) *text, + int textpos, Eterm *p, Uint nelems) { DMC_STACK_TYPE(UWord) instr_save; + Uint i; + + DMC_INIT_STACK(instr_save); + while (DMC_STACK_NUM(*text) > textpos) { + DMC_PUSH(instr_save, DMC_POP(*text)); + } + for (i = nelems; i--;) { + do_emit_constant(context, text, p[i]); + } + while(!DMC_EMPTY(instr_save)) { + DMC_PUSH(*text, DMC_POP(instr_save)); + } + DMC_FREE(instr_save); +} + +static DMCRet +dmc_array(DMCContext *context, DMCHeap *heap, DMC_STACK_TYPE(UWord) *text, + Eterm *p, Uint nelems, int *constant) +{ int all_constant = 1; int textpos = DMC_STACK_NUM(*text); - Eterm *p = tuple_val(t); - Uint nelems = arityval(*p); Uint i; - int c; - DMCRet ret; /* - ** We remember where we started to layout code, + ** We remember where we started to layout code, ** assume all is constant and back up and restart if not so. - ** The tuple should be laid out with the last element first, - ** so we can memcpy the tuple to the eheap. + ** The array should be laid out with the last element first, + ** so we can memcpy it to the eheap. */ - for (i = nelems; i > 0; --i) { - if ((ret = dmc_expr(context, heap, text, p[i], &c)) != retOk) - return ret; - if (!c && all_constant) { - all_constant = 0; - if (i < nelems) { - Uint j; + for (i = nelems; i--;) { + DMCRet ret; + int c; - /* - * Oops, we need to relayout the constants. - * Save the already laid out instructions. - */ - DMC_INIT_STACK(instr_save); - while (DMC_STACK_NUM(*text) > textpos) - DMC_PUSH(instr_save, DMC_POP(*text)); - for (j = nelems; j > i; --j) - do_emit_constant(context, text, p[j]); - while(!DMC_EMPTY(instr_save)) - DMC_PUSH(*text, DMC_POP(instr_save)); - DMC_FREE(instr_save); - } - } else if (c && !all_constant) { - /* push a constant */ - do_emit_constant(context, text, p[i]); - } + ret = dmc_expr(context, heap, text, p[i], &c); + if (ret != retOk) { + return ret; + } + if (!c && all_constant) { + all_constant = 0; + if (i < nelems - 1) { + dmc_rearrange_constants(context, text, textpos, + p + i + 1, nelems - i - 1); + } + } else if (c && !all_constant) { + do_emit_constant(context, text, p[i]); + } + } + *constant = all_constant; + return retOk; +} + +static DMCRet +dmc_tuple(DMCContext *context, DMCHeap *heap, DMC_STACK_TYPE(UWord) *text, + Eterm t, int *constant) +{ + int all_constant; + Eterm *p = tuple_val(t); + Uint nelems = arityval(*p); + DMCRet ret; + + ret = dmc_array(context, heap, text, p + 1, nelems, &all_constant); + if (ret != retOk) { + return ret; } - if (all_constant) { - *constant = 1; - return retOk; + *constant = 1; + return retOk; } DMC_PUSH(*text, matchMkTuple); DMC_PUSH(*text, nelems); @@ -3586,6 +3834,93 @@ static DMCRet dmc_tuple(DMCContext *context, return retOk; } +static DMCRet +dmc_map(DMCContext *context, DMCHeap *heap, DMC_STACK_TYPE(UWord) *text, + Eterm t, int *constant) +{ + int nelems; + int constant_values; + DMCRet ret; + if (is_flatmap(t)) { + flatmap_t *m = (flatmap_t *)flatmap_val(t); + Eterm *values = flatmap_get_values(m); + + nelems = flatmap_get_size(m); + ret = dmc_array(context, heap, text, values, nelems, &constant_values); + + if (ret != retOk) { + return ret; + } + if (constant_values) { + *constant = 1; + return retOk; + } + DMC_PUSH(*text, matchPushC); + DMC_PUSH(*text, dmc_private_copy(context, m->keys)); + if (++context->stack_used > context->stack_need) { + context->stack_need = context->stack_used; + } + DMC_PUSH(*text, matchMkFlatMap); + DMC_PUSH(*text, nelems); + context->stack_used -= nelems; + *constant = 0; + return retOk; + } else { + DECLARE_WSTACK(wstack); + Eterm *kv; + int c; + + ASSERT(is_hashmap(t)); + + hashmap_iterator_init(&wstack, t, 1); + constant_values = 1; + nelems = hashmap_size(t); + + while ((kv=hashmap_iterator_prev(&wstack)) != NULL) { + if ((ret = dmc_expr(context, heap, text, CDR(kv), &c)) != retOk) { + DESTROY_WSTACK(wstack); + return ret; + } + if (!c) + constant_values = 0; + } + + if (constant_values) { + *constant = 1; + DESTROY_WSTACK(wstack); + return retOk; + } + + *constant = 0; + + hashmap_iterator_init(&wstack, t, 1); + + while ((kv=hashmap_iterator_prev(&wstack)) != NULL) { + /* push key */ + if ((ret = dmc_expr(context, heap, text, CAR(kv), &c)) != retOk) { + DESTROY_WSTACK(wstack); + return ret; + } + if (c) { + do_emit_constant(context, text, CAR(kv)); + } + /* push value */ + if ((ret = dmc_expr(context, heap, text, CDR(kv), &c)) != retOk) { + DESTROY_WSTACK(wstack); + return ret; + } + if (c) { + do_emit_constant(context, text, CDR(kv)); + } + } + DMC_PUSH(*text, matchMkHashMap); + DMC_PUSH(*text, nelems); + context->stack_used -= nelems; + DESTROY_WSTACK(wstack); + return retOk; + } +} + static DMCRet dmc_whole_expression(DMCContext *context, DMCHeap *heap, DMC_STACK_TYPE(UWord) *text, @@ -4580,7 +4915,10 @@ static DMCRet dmc_expr(DMCContext *context, return ret; break; case TAG_PRIMARY_BOXED: - if (!BOXED_IS_TUPLE(t)) { + if (is_map(t)) { + return dmc_map(context, heap, text, t, constant); + } + if (!is_tuple(t)) { goto simple_term; } p = tuple_val(t); @@ -4855,7 +5193,7 @@ static Eterm my_copy_struct(Eterm t, Eterm **hp, ErlOffHeap* off_heap) *hp += 2; break; case TAG_PRIMARY_BOXED: - if (BOXED_IS_TUPLE(t)) { + if (is_tuple(t)) { if (arityval(*tuple_val(t)) == 1 && is_tuple(a = tuple_val(t)[1])) { Uint i,n; @@ -5126,6 +5464,18 @@ void db_match_dis(Binary *bp) ++t; erts_printf("Tuple\t%beu\n", n); break; + case matchMap: + ++t; + n = *t; + ++t; + erts_printf("Map\t%beu\n", n); + break; + case matchKey: + ++t; + p = (Eterm) *t; + ++t; + erts_printf("Key\t%p (%T)\n", t, p); + break; case matchPushT: ++t; n = *t; @@ -5136,10 +5486,20 @@ void db_match_dis(Binary *bp) ++t; erts_printf("PushL\n"); break; + case matchPushM: + ++t; + n = *t; + ++t; + erts_printf("PushM\t%beu\n", n); + break; case matchPop: ++t; erts_printf("Pop\n"); break; + case matchSwap: + ++t; + erts_printf("Swap\n"); + break; case matchBind: ++t; n = *t; @@ -5252,6 +5612,18 @@ void db_match_dis(Binary *bp) ++t; erts_printf("MkTuple\t%beu\n", n); break; + case matchMkFlatMap: + ++t; + n = *t; + ++t; + erts_printf("MkFlatMap\t%beu\n", n); + break; + case matchMkHashMap: + ++t; + n = *t; + ++t; + erts_printf("MkHashMap\t%beu\n", n); + break; case matchOr: ++t; n = *t; diff --git a/erts/emulator/beam/erl_db_util.h b/erts/emulator/beam/erl_db_util.h index 328b19dfc9..ca206c7f58 100644 --- a/erts/emulator/beam/erl_db_util.h +++ b/erts/emulator/beam/erl_db_util.h @@ -76,6 +76,9 @@ typedef struct db_term { union db_table; typedef union db_table DbTable; +#define DB_MUST_RESIZE 1 +#define DB_NEW_OBJECT 2 + /* Info about a database entry while it's being updated * (by update_counter or update_element) */ @@ -84,7 +87,7 @@ typedef struct { DbTerm* dbterm; void** bp; /* {Hash|Tree}DbTerm** */ Uint new_size; - int mustResize; + int flags; void* lck; #if HALFWORD_HEAP unsigned char* abs_vec; /* [i] true if dbterm->tpl[i] is absolute Eterm */ @@ -165,6 +168,7 @@ typedef struct db_table_method DbTable* tb, /* [in out] */ Eterm continuation, Eterm* ret); + int (*db_take)(Process *, DbTable *, Eterm, Eterm *); int (*db_delete_all_objects)(Process* p, DbTable* db /* [in out] */ ); @@ -182,15 +186,14 @@ typedef struct db_table_method void *arg); void (*db_check_table)(DbTable* tb); - /* Lookup a dbterm for updating. Return false if not found. - */ - int (*db_lookup_dbterm)(DbTable*, Eterm key, - DbUpdateHandle* handle); /* [out] */ + /* Lookup a dbterm for updating. Return false if not found. */ + int (*db_lookup_dbterm)(Process *, DbTable *, Eterm key, Eterm obj, + DbUpdateHandle* handle); - /* Must be called for each db_lookup_dbterm that returned true, - ** even if dbterm was not updated. - */ - void (*db_finalize_dbterm)(DbUpdateHandle* handle); + /* Must be called for each db_lookup_dbterm that returned true, even if + ** dbterm was not updated. If the handle was of a new object and cret is + ** not DB_ERROR_NONE, the object is removed from the table. */ + void (*db_finalize_dbterm)(int cret, DbUpdateHandle* handle); } DbTableMethod; diff --git a/erts/emulator/beam/erl_driver.h b/erts/emulator/beam/erl_driver.h index f9938fc66c..e498ac70ec 100644 --- a/erts/emulator/beam/erl_driver.h +++ b/erts/emulator/beam/erl_driver.h @@ -133,7 +133,7 @@ typedef struct { #define ERL_DRV_EXTENDED_MARKER (0xfeeeeeed) #define ERL_DRV_EXTENDED_MAJOR_VERSION 3 -#define ERL_DRV_EXTENDED_MINOR_VERSION 1 +#define ERL_DRV_EXTENDED_MINOR_VERSION 2 /* * The emulator will refuse to load a driver with a major version @@ -361,6 +361,9 @@ typedef struct erl_drv_entry { /* Called on behalf of driver_select when it is safe to release 'event'. A typical unix driver would call close(event) */ + void (*emergency_close)(ErlDrvData drv_data); + /* called when the port is closed abruptly. + specifically when erl_crash_dump is called. */ /* When adding entries here, dont forget to pad in obsolete/driver.h */ } ErlDrvEntry; diff --git a/erts/emulator/beam/erl_drv_thread.c b/erts/emulator/beam/erl_drv_thread.c index 147249f751..31b05d22af 100644 --- a/erts/emulator/beam/erl_drv_thread.c +++ b/erts/emulator/beam/erl_drv_thread.c @@ -601,17 +601,14 @@ erl_drv_thread_create(char *name, #ifdef USE_THREADS int res; struct ErlDrvTid_ *dtid; - ethr_thr_opts ethr_opts; + ethr_thr_opts ethr_opts = ETHR_THR_OPTS_DEFAULT_INITER; ethr_thr_opts *use_opts; - ethr_thr_opts def_ethr_opts = ETHR_THR_OPTS_DEFAULT_INITER; if (!opts) use_opts = NULL; else { - sys_memcpy((void *) ðr_opts, - (void *) &def_ethr_opts, - sizeof(ethr_thr_opts)); ethr_opts.suggested_stack_size = opts->suggested_stack_size; + ethr_opts.name = name; use_opts = ðr_opts; } diff --git a/erts/emulator/beam/erl_gc.c b/erts/emulator/beam/erl_gc.c index 0db42d4325..d1a7ee113b 100644 --- a/erts/emulator/beam/erl_gc.c +++ b/erts/emulator/beam/erl_gc.c @@ -20,6 +20,8 @@ # include "config.h" #endif +#define ERL_WANT_GC_INTERNALS__ + #include "sys.h" #include "erl_vm.h" #include "global.h" @@ -173,15 +175,15 @@ erts_init_gc(void) int i = 0, ix; Sint max_heap_size = 0; - ASSERT(offsetof(ProcBin,thing_word) == offsetof(struct erl_off_heap_header,thing_word)); - ASSERT(offsetof(ProcBin,thing_word) == offsetof(ErlFunThing,thing_word)); - ASSERT(offsetof(ProcBin,thing_word) == offsetof(ExternalThing,header)); - ASSERT(offsetof(ProcBin,size) == offsetof(struct erl_off_heap_header,size)); - ASSERT(offsetof(ProcBin,size) == offsetof(ErlSubBin,size)); - ASSERT(offsetof(ProcBin,size) == offsetof(ErlHeapBin,size)); - ASSERT(offsetof(ProcBin,next) == offsetof(struct erl_off_heap_header,next)); - ASSERT(offsetof(ProcBin,next) == offsetof(ErlFunThing,next)); - ASSERT(offsetof(ProcBin,next) == offsetof(ExternalThing,next)); + ERTS_CT_ASSERT(offsetof(ProcBin,thing_word) == offsetof(struct erl_off_heap_header,thing_word)); + ERTS_CT_ASSERT(offsetof(ProcBin,thing_word) == offsetof(ErlFunThing,thing_word)); + ERTS_CT_ASSERT(offsetof(ProcBin,thing_word) == offsetof(ExternalThing,header)); + ERTS_CT_ASSERT(offsetof(ProcBin,size) == offsetof(struct erl_off_heap_header,size)); + ERTS_CT_ASSERT(offsetof(ProcBin,size) == offsetof(ErlSubBin,size)); + ERTS_CT_ASSERT(offsetof(ProcBin,size) == offsetof(ErlHeapBin,size)); + ERTS_CT_ASSERT(offsetof(ProcBin,next) == offsetof(struct erl_off_heap_header,next)); + ERTS_CT_ASSERT(offsetof(ProcBin,next) == offsetof(ErlFunThing,next)); + ERTS_CT_ASSERT(offsetof(ProcBin,next) == offsetof(ExternalThing,next)); erts_test_long_gc_sleep = 0; @@ -2400,7 +2402,6 @@ sweep_off_heap(Process *p, int fullsweep) } pb->val = erts_bin_realloc(pb->val, new_size); - pb->val->orig_size = new_size; pb->bytes = (byte *) pb->val->orig_bytes; } } diff --git a/erts/emulator/beam/erl_gc.h b/erts/emulator/beam/erl_gc.h index 5203dda263..8afcb060a1 100644 --- a/erts/emulator/beam/erl_gc.h +++ b/erts/emulator/beam/erl_gc.h @@ -20,10 +20,12 @@ #ifndef __ERL_GC_H__ #define __ERL_GC_H__ -#include "erl_map.h" +#if defined(ERL_WANT_GC_INTERNALS__) || defined(ERTS_DO_INCL_GLB_INLINE_FUNC_DEF) /* GC declarations shared by beam/erl_gc.c and hipe/hipe_gc.c */ +#include "erl_map.h" + #if defined(DEBUG) && !ERTS_GLB_INLINE_INCL_FUNC_DEF # define HARDDEBUG 1 #endif @@ -53,7 +55,9 @@ do { \ 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 MAP_SUBTAG: nelts+=flatmap_get_size(PTR) + 1; break; \ + case HASHMAP_SUBTAG: nelts=hashmap_bitcount(MAP_HEADER_VAL(HDR)); \ + nelts += is_hashmap_header_head(HDR) ? 1 : 0; break; \ case FUN_SUBTAG: nelts+=((ErlFunThing*)(PTR))->num_free+1; break; \ } \ gval = make_boxed(HTOP); \ @@ -61,14 +65,11 @@ do { \ *HTOP++ = HDR; \ *PTR++ = gval; \ while (nelts--) *HTOP++ = *PTR++; \ - \ } while(0) #define in_area(ptr,start,nbytes) \ ((UWord)((char*)(ptr) - (char*)(start)) < (nbytes)) -extern Uint erts_test_long_gc_sleep; - #if defined(DEBUG) || defined(ERTS_OFFHEAP_DEBUG) int within(Eterm *ptr, Process *p); #endif @@ -97,4 +98,33 @@ ERTS_GLB_INLINE Eterm follow_moved(Eterm term) } #endif +#endif /* ERL_GC_C__ || HIPE_GC_C__ */ + +/* + * Global exported + */ + +extern Uint erts_test_long_gc_sleep; + +typedef struct { + Uint64 reclaimed; + Uint64 garbage_cols; +} ErtsGCInfo; + +void erts_gc_info(ErtsGCInfo *gcip); +void erts_init_gc(void); +int erts_garbage_collect(struct process*, int, Eterm*, int); +void erts_garbage_collect_hibernate(struct process* p); +Eterm erts_gc_after_bif_call(struct process* p, Eterm result, Eterm* regs, Uint arity); +void erts_garbage_collect_literals(struct process* p, Eterm* literals, + Uint lit_size, + struct erl_off_heap_header* oh); +Uint erts_next_heap_size(Uint, Uint); +Eterm erts_heap_sizes(struct process* p); + +void erts_offset_off_heap(struct erl_off_heap*, Sint, Eterm*, Eterm*); +void erts_offset_heap_ptr(Eterm*, Uint, Sint, Eterm*, Eterm*); +void erts_offset_heap(Eterm*, Uint, Sint, Eterm*, Eterm*); +void erts_free_heap_frags(struct process* p); + #endif /* __ERL_GC_H__ */ diff --git a/erts/emulator/beam/erl_init.c b/erts/emulator/beam/erl_init.c index 61f8385efc..fe065e196d 100644 --- a/erts/emulator/beam/erl_init.c +++ b/erts/emulator/beam/erl_init.c @@ -161,9 +161,6 @@ int H_MIN_SIZE; /* The minimum heap grain */ int BIN_VH_MIN_SIZE; /* The minimum binary virtual*/ Uint32 erts_debug_flags; /* Debug flags. */ -#ifdef ERTS_OPCODE_COUNTER_SUPPORT -int count_instructions; -#endif int erts_backtrace_depth; /* How many functions to show in a backtrace * in error codes. */ @@ -194,6 +191,8 @@ int erts_disable_tolerant_timeofday; /* Time correction can be disabled it is int erts_atom_table_size = ATOM_LIMIT; /* Maximum number of atoms */ +int erts_pd_initial_size = 10; + int erts_modified_timing_level; int erts_no_crash_dump = 0; /* Use -d to suppress crash dump. */ @@ -519,6 +518,8 @@ void erts_usage(void) H_DEFAULT_SIZE); erts_fprintf(stderr, "-hmbs size set minimum binary virtual heap size in words (default %d)\n", VH_DEFAULT_SIZE); + erts_fprintf(stderr, "-hpds size initial process dictionary size (default %d)\n", + erts_pd_initial_size); /* erts_fprintf(stderr, "-i module set the boot module (default init)\n"); */ @@ -1408,6 +1409,7 @@ erl_start(int argc, char **argv) * * h|ms - min_heap_size * h|mbs - min_bin_vheap_size + * h|pds - erts_pd_initial_size * */ if (has_prefix("mbs", sub_param)) { @@ -1425,6 +1427,14 @@ erl_start(int argc, char **argv) erts_usage(); } VERBOSE(DEBUG_SYSTEM, ("using minimum heap size %d\n", H_MIN_SIZE)); + } else if (has_prefix("pds", sub_param)) { + arg = get_arg(sub_param+3, argv[i+1], &i); + if ((erts_pd_initial_size = atoi(arg)) <= 0) { + erts_fprintf(stderr, "bad initial process dictionary size %s\n", arg); + erts_usage(); + } + VERBOSE(DEBUG_SYSTEM, ("using initial process dictionary size %d\n", + erts_pd_initial_size)); } else { /* backward compatibility */ arg = get_arg(argv[i]+2, argv[i+1], &i); @@ -1900,11 +1910,6 @@ erl_start(int argc, char **argv) if (argv[i][2] == 0) { /* -c: documented option */ erts_disable_tolerant_timeofday = 1; } -#ifdef ERTS_OPCODE_COUNTER_SUPPORT - else if (argv[i][2] == 'i') { /* -ci: undcoumented option*/ - count_instructions = 1; - } -#endif break; case 'W': arg = get_arg(argv[i]+2, argv[i+1], &i); diff --git a/erts/emulator/beam/erl_instrument.c b/erts/emulator/beam/erl_instrument.c index df7c443387..da85b86c87 100644 --- a/erts/emulator/beam/erl_instrument.c +++ b/erts/emulator/beam/erl_instrument.c @@ -1226,7 +1226,7 @@ erts_instr_init(int stat, int map_stat) mem_anchor = NULL; /* Install instrumentation functions */ - ASSERT(sizeof(erts_allctrs) == sizeof(real_allctrs)); + ERTS_CT_ASSERT(sizeof(erts_allctrs) == sizeof(real_allctrs)); sys_memcpy((void *)real_allctrs,(void *)erts_allctrs,sizeof(erts_allctrs)); diff --git a/erts/emulator/beam/erl_map.c b/erts/emulator/beam/erl_map.c index 5e740aacdd..bd6da0a6f1 100644 --- a/erts/emulator/beam/erl_map.c +++ b/erts/emulator/beam/erl_map.c @@ -16,6 +16,9 @@ * * %CopyrightEnd% * + * hashmaps are an adaption of Rich Hickeys Persistent HashMaps + * which were an adaption of Phil Bagwells - Hash Array Mapped Tries + * * Author: Björn-Egil Dahlberg */ @@ -62,39 +65,78 @@ * - erts_internal:map_to_tuple_keys/1 */ +#ifndef DECL_AM +#define DECL_AM(S) Eterm AM_ ## S = am_atom_put(#S, sizeof(#S) - 1) +#endif + +/* for hashmap_from_list/1 */ +typedef struct { + Uint32 hx; + Uint32 skip; + Uint i; + Eterm val; +} hxnode_t; + + +static Eterm flatmap_merge(Process *p, Eterm nodeA, Eterm nodeB); +static Eterm map_merge_mixed(Process *p, Eterm flat, Eterm tree, int swap_args); +static Eterm hashmap_merge(Process *p, Eterm nodeA, Eterm nodeB); +static Eterm hashmap_to_list(Process *p, Eterm map); +static Eterm hashmap_keys(Process *p, Eterm map); +static Eterm hashmap_values(Process *p, Eterm map); +static Eterm hashmap_delete(Process *p, Uint32 hx, Eterm key, Eterm node); +static Eterm map_from_validated_list(Process *p, Eterm list, Uint size); +static Eterm hashmap_from_validated_list(Process *p, Eterm list, Uint size); +static Eterm hashmap_from_unsorted_array(ErtsHeapFactory*, hxnode_t *hxns, Uint n, int reject_dupkeys); +static Eterm hashmap_from_sorted_unique_array(ErtsHeapFactory*, hxnode_t *hxns, Uint n, int is_root); +static Eterm hashmap_from_chunked_array(ErtsHeapFactory*, hxnode_t *hxns, Uint n, int is_root); +static Eterm hashmap_info(Process *p, Eterm node); +static Eterm hashmap_bld_tuple_uint(Uint **hpp, Uint *szp, Uint n, Uint nums[]); +static int hxnodecmp(hxnode_t* a, hxnode_t* b); +static int hxnodecmpkey(hxnode_t* a, hxnode_t* b); + /* 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)) { + if (is_flatmap(BIF_ARG_1)) { Eterm *hp; Uint hsz = 0; - map_t *mp = (map_t*)map_val(BIF_ARG_1); - Uint n = map_get_size(mp); + flatmap_t *mp = (flatmap_t*)flatmap_val(BIF_ARG_1); + Uint n = flatmap_get_size(mp); erts_bld_uint(NULL, &hsz, n); hp = HAlloc(BIF_P, hsz); BIF_RET(erts_bld_uint(&hp, NULL, n)); + } else if (is_hashmap(BIF_ARG_1)) { + Eterm *head, *hp, res; + Uint size, hsz=0; + + head = hashmap_val(BIF_ARG_1); + size = head[1]; + (void) erts_bld_uint(NULL, &hsz, size); + hp = HAlloc(BIF_P, hsz); + res = erts_bld_uint(&hp, NULL, size); + BIF_RET(res); } BIF_ERROR(BIF_P, BADARG); } -/* maps:to_list/1 - */ +/* maps:to_list/1 */ BIF_RETTYPE maps_to_list_1(BIF_ALIST_1) { - if (is_map(BIF_ARG_1)) { + if (is_flatmap(BIF_ARG_1)) { Uint n; Eterm* hp; Eterm *ks,*vs, res, tup; - map_t *mp = (map_t*)map_val(BIF_ARG_1); + flatmap_t *mp = (flatmap_t*)flatmap_val(BIF_ARG_1); - ks = map_get_keys(mp); - vs = map_get_values(mp); - n = map_get_size(mp); + ks = flatmap_get_keys(mp); + vs = flatmap_get_values(mp); + n = flatmap_get_size(mp); hp = HAlloc(BIF_P, (2 + 3) * n); res = NIL; @@ -104,6 +146,8 @@ BIF_RETTYPE maps_to_list_1(BIF_ALIST_1) { } BIF_RET(res); + } else if (is_hashmap(BIF_ARG_1)) { + return hashmap_to_list(BIF_P, BIF_ARG_1); } BIF_ERROR(BIF_P, BADARG); @@ -113,89 +157,84 @@ BIF_RETTYPE maps_to_list_1(BIF_ALIST_1) { * return value if key *matches* a key in the map */ -int erts_maps_find(Eterm key, Eterm map, Eterm *value) { +const Eterm * +#if HALFWORD_HEAP +erts_maps_get_rel(Eterm key, Eterm map, Eterm *map_base) +#else +erts_maps_get(Eterm key, Eterm map) +#endif +{ + Uint32 hx; + if (is_flatmap_rel(map, map_base)) { + Eterm *ks, *vs; + flatmap_t *mp; + Uint n, i; - Eterm *ks,*vs; - map_t *mp; - Uint n,i; + mp = (flatmap_t *)flatmap_val_rel(map, map_base); + n = flatmap_get_size(mp); - mp = (map_t*)map_val(map); - n = map_get_size(mp); - ks = map_get_keys(mp); - vs = map_get_values(mp); + if (n == 0) { + return NULL; + } - for( i = 0; i < n; i++) { - if (EQ(ks[i], key)) { - *value = vs[i]; - return 1; + ks = (Eterm *)tuple_val_rel(mp->keys, map_base) + 1; + vs = flatmap_get_values(mp); + + if (is_immed(key)) { + for (i = 0; i < n; i++) { + if (ks[i] == key) { + return &vs[i]; + } + } + } + + for (i = 0; i < n; i++) { + if (eq_rel(ks[i], map_base, key, NULL)) { + return &vs[i]; + } } + return NULL; } - return 0; + ASSERT(is_hashmap_rel(map, map_base)); + hx = hashmap_make_hash(key); + + return erts_hashmap_get_rel(hx, key, map, map_base); } BIF_RETTYPE maps_find_2(BIF_ALIST_2) { if (is_map(BIF_ARG_2)) { - Eterm *hp, value,res; + Eterm *hp, res; + const Eterm *value; - if (erts_maps_find(BIF_ARG_1, BIF_ARG_2, &value)) { + value = erts_maps_get(BIF_ARG_1, BIF_ARG_2); + if (value) { hp = HAlloc(BIF_P, 3); res = make_tuple(hp); *hp++ = make_arityval(2); *hp++ = am_ok; - *hp++ = value; + *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; + Eterm error; + const Eterm *value; char *s_error; - if (erts_maps_get(BIF_ARG_1, BIF_ARG_2, &value)) { - BIF_RET(value); + value = erts_maps_get(BIF_ARG_1, BIF_ARG_2); + if (value) { + BIF_RET(*value); } s_error = "bad_key"; @@ -213,13 +252,8 @@ BIF_RETTYPE maps_get_2(BIF_ALIST_2) { */ 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; - + Eterm item = BIF_ARG_1, res, *kv; + Uint size = 0; if (is_list(item) || is_nil(item)) { /* Calculate size and check validity */ @@ -240,368 +274,1091 @@ BIF_RETTYPE maps_from_list_1(BIF_ALIST_1) { if (is_not_nil(item)) goto error; - hp = HAlloc(BIF_P, 3 + 1 + (2 * size)); - thp = hp; + if (size > MAP_SMALL_MAP_LIMIT) { + BIF_RET(hashmap_from_validated_list(BIF_P, BIF_ARG_1, size)); + } else { + BIF_RET(map_from_validated_list(BIF_P, BIF_ARG_1, size)); + } + } + +error: + + BIF_ERROR(BIF_P, BADARG); +} + +static Eterm map_from_validated_list(Process *p, Eterm list, Uint size) { + Eterm *kv, item = list; + Eterm *hp, *thp,*vs, *ks, keys, res; + flatmap_t *mp; + Uint unused_size = 0; + Sint c = 0; + Sint idx = 0; + + + hp = HAlloc(p, 3 + 1 + (2 * size)); + thp = hp; + keys = make_tuple(hp); + *hp++ = make_arityval(size); + ks = hp; + hp += size; + mp = (flatmap_t*)hp; + res = make_flatmap(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) + return res; + + /* 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(p, vs + size + unused_size, vs + size); + } + + *thp = make_arityval(size); + mp->size = size; + return res; +} + +#define swizzle32(D,S) \ + do { \ + (D) = ((S) & 0x0000000f) << 28 | ((S) & 0x000000f0) << 20 \ + | ((S) & 0x00000f00) << 12 | ((S) & 0x0000f000) << 4 \ + | ((S) & 0x000f0000) >> 4 | ((S) & 0x00f00000) >> 12 \ + | ((S) & 0x0f000000) >> 20 | ((S) & 0xf0000000) >> 28; \ + } while(0) + +#define maskval(V,L) (((V) >> ((7 - (L))*4)) & 0xf) +#define cdepth(V1,V2) (hashmap_clz((V1) ^ (V2)) >> 2) + +static Eterm hashmap_from_validated_list(Process *p, Eterm list, Uint size) { + Eterm item = list; + Eterm *hp; + Eterm *kv, res; + Eterm tmp[2]; + Uint32 sw, hx; + Uint ix = 0; + hxnode_t *hxns; + ErtsHeapFactory factory; + + ASSERT(size > 0); + + hp = HAlloc(p, (2 * size)); + + /* create tmp hx values and leaf ptrs */ + hxns = (hxnode_t *)erts_alloc(ERTS_ALC_T_TMP, size * sizeof(hxnode_t)); + + while(is_list(item)) { + res = CAR(list_val(item)); + kv = tuple_val(res); + hx = hashmap_restore_hash(tmp,0,kv[1]); + swizzle32(sw,hx); + hxns[ix].hx = sw; + hxns[ix].val = CONS(hp, kv[1], kv[2]); hp += 2; + hxns[ix].skip = 1; /* will be reassigned in from_array */ + hxns[ix].i = ix; + ix++; + item = CDR(list_val(item)); + } + + factory.p = p; + res = hashmap_from_unsorted_array(&factory, hxns, size, 0); + + erts_free(ERTS_ALC_T_TMP, (void *) hxns); + ERTS_VERIFY_UNUSED_TEMP_ALLOC(p); + + if (hashmap_size(res) <= MAP_SMALL_MAP_LIMIT) { + DECLARE_WSTACK(wstack); + Eterm *kv, *ks, *vs; + flatmap_t *mp; + Eterm keys; + Uint n = hashmap_size(res); + + /* build flat structure */ + hp = HAlloc(p, 3 + 1 + (2 * n)); keys = make_tuple(hp); - *hp++ = make_arityval(size); + *hp++ = make_arityval(n); ks = hp; - hp += size; - mp = (map_t*)hp; - res = make_map(mp); + hp += n; + mp = (flatmap_t*)hp; hp += MAP_HEADER_SIZE; vs = hp; mp->thing_word = MAP_HEADER; - mp->size = size; /* set later, might shrink*/ + mp->size = n; mp->keys = keys; - if (size == 0) - BIF_RET(res); + hashmap_iterator_init(&wstack, res, 0); - item = BIF_ARG_1; + while ((kv=hashmap_iterator_next(&wstack)) != NULL) { + *ks++ = CAR(kv); + *vs++ = CDR(kv); + } - /* first entry */ - kv = tuple_val(CAR(list_val(item))); - ks[0] = kv[1]; - vs[0] = kv[2]; - size = 1; - item = CDR(list_val(item)); + /* it cannot have multiple keys */ + erts_validate_and_sort_flatmap(mp); - /* insert sort key/value pairs */ - while(is_list(item)) { + DESTROY_WSTACK(wstack); + return make_flatmap(mp); + } - 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. - */ + return res; +} - idx = size; +Eterm erts_hashmap_from_array(ErtsHeapFactory* factory, Eterm *leafs, Uint n, + int reject_dupkeys) { + Uint32 sw, hx; + Uint ix; + hxnode_t *hxns; + Eterm res; + + /* create tmp hx values and leaf ptrs */ + hxns = (hxnode_t *)erts_alloc(ERTS_ALC_T_TMP, n * sizeof(hxnode_t)); + + for (ix = 0; ix < n; ix++) { + hx = hashmap_make_hash(*leafs); + swizzle32(sw,hx); + hxns[ix].hx = sw; + hxns[ix].val = make_list(leafs); + hxns[ix].skip = 1; + hxns[ix].i = ix; + leafs += 2; + } - while(idx > 0 && (c = CMP_TERM(kv[1],ks[idx-1])) < 0) { idx--; } + res = hashmap_from_unsorted_array(factory, hxns, n, reject_dupkeys); - 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--; + erts_free(ERTS_ALC_T_TMP, (void *) hxns); + + return res; +} + + +Eterm erts_hashmap_from_ks_and_vs_extra(Process *p, Eterm *ks, Eterm *vs, Uint n, + Eterm key, Eterm value) { + Uint32 sw, hx; + Uint i,sz; + hxnode_t *hxns; + ErtsHeapFactory factory; + Eterm *hp, res; + + sz = (key == THE_NON_VALUE) ? n : (n + 1); + ASSERT(sz > MAP_SMALL_MAP_LIMIT); + hp = HAlloc(p, 2 * sz); + + /* create tmp hx values and leaf ptrs */ + hxns = (hxnode_t *)erts_alloc(ERTS_ALC_T_TMP, sz * sizeof(hxnode_t)); + + for(i = 0; i < n; i++) { + hx = hashmap_make_hash(ks[i]); + swizzle32(sw,hx); + hxns[i].hx = sw; + hxns[i].val = CONS(hp, ks[i], vs[i]); hp += 2; + hxns[i].skip = 1; /* will be reassigned in from_array */ + hxns[i].i = i; + } + + if (key != THE_NON_VALUE) { + hx = hashmap_make_hash(key); + swizzle32(sw,hx); + hxns[i].hx = sw; + hxns[i].val = CONS(hp, key, value); hp += 2; + hxns[i].skip = 1; + hxns[i].i = i; + } + + factory.p = p; + res = hashmap_from_unsorted_array(&factory, hxns, sz, 0); + + erts_free(ERTS_ALC_T_TMP, (void *) hxns); + ERTS_VERIFY_UNUSED_TEMP_ALLOC(p); + + return res; +} + +static Eterm hashmap_from_unsorted_array(ErtsHeapFactory* factory, + hxnode_t *hxns, Uint n, + int reject_dupkeys) { + Uint jx = 0, ix = 0, lx, cx; + Eterm res; + + if (n == 0) { + Eterm *hp; + hp = erts_produce_heap(factory, 2, 0); + hp[0] = MAP_HEADER_HAMT_HEAD_BITMAP(0); + hp[1] = 0; + + return make_hashmap(hp); + } + + /* sort and compact array (remove non-unique entries) */ + qsort(hxns, n, sizeof(hxnode_t), (int (*)(const void *, const void *)) hxnodecmp); + + ix = 0, cx = 0; + while(ix < n - 1) { + if (hxns[ix].hx == hxns[ix+1].hx) { + + /* find region of equal hash values */ + jx = ix + 1; + while(jx < n && hxns[ix].hx == hxns[jx].hx) { jx++; } + /* find all correct keys from region + * (last in list but now hash sorted so we check highest id instead) */ + + /* resort with keys instead of hash value within region */ + + qsort(&hxns[ix], jx - ix, sizeof(hxnode_t), + (int (*)(const void *, const void *)) hxnodecmpkey); + + while(ix < jx) { + lx = ix; + while(ix < jx && EQ(CAR(list_val(hxns[ix].val)), CAR(list_val(hxns[lx].val)))) { + if (reject_dupkeys) + return THE_NON_VALUE; + + if (hxns[ix].i > hxns[lx].i) { + lx = ix; + } + ix++; } - ks[idx] = kv[1]; - vs[idx] = kv[2]; - size++; + hxns[cx].hx = hxns[lx].hx; + hxns[cx].val = hxns[lx].val; + cx++; } - item = CDR(list_val(item)); + ix = jx; + continue; + } + if (ix > cx) { + hxns[cx].hx = hxns[ix].hx; + hxns[cx].val = hxns[ix].val; } + cx++; + ix++; + } - 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 */ + if (ix < n) { + hxns[cx].hx = hxns[ix].hx; + hxns[cx].val = hxns[ix].val; + cx++; + } - ks[size] = make_pos_bignum_header(unused_size - 1); - HRelease(BIF_P, vs + size + unused_size, vs + size); - } + if (cx > 1) { + /* recursive decompose array */ + res = hashmap_from_sorted_unique_array(factory, hxns, cx, 0); + } else { + Eterm *hp; - *thp = make_arityval(size); - mp->size = size; - BIF_RET(res); + /* we only have one item, either because n was 1 or + * because we hade multiples of the same key. + * + * hash value has been swizzled, need to drag it down to get the + * correct slot. */ + + hp = erts_produce_heap(factory, HAMT_HEAD_BITMAP_SZ(1), 0); + hp[0] = MAP_HEADER_HAMT_HEAD_BITMAP(1 << ((hxns[0].hx >> 0x1c) & 0xf)); + hp[1] = 1; + hp[2] = hxns[0].val; + res = make_hashmap(hp); } -error: + return res; +} - BIF_ERROR(BIF_P, BADARG); +static Eterm hashmap_from_sorted_unique_array(ErtsHeapFactory* factory, + hxnode_t *hxns, Uint n, int lvl) { + Eterm res = NIL; + Uint i,ix,jx,elems; + Uint32 sw, hx; + Eterm val; + Eterm th[2]; + hxnode_t *tmp; + + ASSERT(lvl < 32); + ix = 0; + elems = 1; + while (ix < n - 1) { + if (hxns[ix].hx == hxns[ix+1].hx) { + jx = ix + 1; + while (jx < n && hxns[ix].hx == hxns[jx].hx) { jx++; } + tmp = (hxnode_t *)erts_alloc(ERTS_ALC_T_TMP, ((jx - ix)) * sizeof(hxnode_t)); + + for(i = 0; i < jx - ix; i++) { + val = hxns[i + ix].val; + hx = hashmap_restore_hash(th, lvl + 8, CAR(list_val(val))); + swizzle32(sw,hx); + tmp[i].hx = sw; + tmp[i].val = val; + tmp[i].i = i; + tmp[i].skip = 1; + } + + qsort(tmp, jx - ix, sizeof(hxnode_t), (int (*)(const void *, const void *)) hxnodecmp); + + hxns[ix].skip = jx - ix; + hxns[ix].val = hashmap_from_sorted_unique_array(factory, tmp, jx - ix, lvl + 8); + erts_free(ERTS_ALC_T_TMP, (void *) tmp); + ix = jx; + if (ix < n) { elems++; } + continue; + } + hxns[ix].skip = 1; + elems++; + ix++; + } + + res = hashmap_from_chunked_array(factory, hxns, elems, !lvl); + + ERTS_FACTORY_HOLE_CHECK(factory); + + return res; } -/* maps:is_key/2 - */ +#define HALLOC_EXTRA 200 +static Eterm hashmap_from_chunked_array(ErtsHeapFactory *factory, + hxnode_t *hxns, Uint n, int is_root) { + Uint ix, d, dn, dc, slot, elems; + Uint32 v, vp, vn, hdr; + Uint bp, sz; + DECLARE_ESTACK(stack); + Eterm res = NIL, *hp = NULL, *nhp; -BIF_RETTYPE maps_is_key_2(BIF_ALIST_2) { - if (is_map(BIF_ARG_2)) { - Eterm *ks, key; - map_t *mp; - Uint n,i; + ASSERT(n > 1); - mp = (map_t*)map_val(BIF_ARG_2); - key = BIF_ARG_1; - n = map_get_size(mp); - ks = map_get_keys(mp); + /* push initial nodes on the stack, + * this is the starting depth */ - if (n == 0) - BIF_RET(am_false); + ix = 0; + d = 0; + vp = hxns[ix].hx; + v = hxns[ix + hxns[ix].skip].hx; - if (is_immed(key)) { - for( i = 0; i < n; i++) { - if (ks[i] == key) { - BIF_RET(am_true); - } + ASSERT(vp > v); + slot = maskval(vp,d); + + while(slot == maskval(v,d)) { + ESTACK_PUSH(stack, 1 << slot); + d++; + slot = maskval(vp,d); + } + + res = hxns[ix].val; + + if (hxns[ix].skip > 1) { + dc = 7; + /* build collision nodes */ + while (dc > d) { + hp = erts_produce_heap(factory, HAMT_NODE_BITMAP_SZ(1), HALLOC_EXTRA); + hp[0] = MAP_HEADER_HAMT_NODE_BITMAP(1 << maskval(vp,dc)); + hp[1] = res; + res = make_hashmap(hp); + dc--; + } + } + + ESTACK_PUSH(stack, res); + ESTACK_PUSH(stack, 1 << slot); + + /* all of the other nodes .. */ + elems = n - 2; /* remove first and last elements */ + while(elems--) { + hdr = ESTACK_POP(stack); + ix = ix + hxns[ix].skip; + + /* determine if node or subtree should be built by looking + * at the next value. */ + + vn = hxns[ix + hxns[ix].skip].hx; + dn = cdepth(v,vn); + ASSERT(v > vn); + + res = hxns[ix].val; + + if (hxns[ix].skip > 1) { + int wat = (d > dn) ? d : dn; + dc = 7; + /* build collision nodes */ + while (dc > wat) { + hp = erts_produce_heap(factory, HAMT_NODE_BITMAP_SZ(1), HALLOC_EXTRA); + hp[0] = MAP_HEADER_HAMT_NODE_BITMAP(1 << maskval(v,dc)); + hp[1] = res; + res = make_hashmap(hp); + dc--; } } - for( i = 0; i < n; i++) { - if (EQ(ks[i], key)) { - BIF_RET(am_true); + /* next depth is higher (implies collision) */ + if (d < dn) { + /* hdr is the popped one initially */ + while(d < dn) { + slot = maskval(v, d); + bp = 1 << slot; + ESTACK_PUSH(stack, hdr | bp); + d++; + hdr = 0; /* clear hdr for all other collisions */ } + + slot = maskval(v, d); + bp = 1 << slot; + /* no more collisions */ + ESTACK_PUSH(stack,res); + ESTACK_PUSH(stack,bp); + } else if (d == dn) { + /* no collisions at all */ + slot = maskval(v, d); + bp = 1 << slot; + ESTACK_PUSH(stack,res); + ESTACK_PUSH(stack,hdr | bp); + } else { + /* dn < n, we have a drop and we are done + * build nodes and subtree */ + while (dn != d) { + slot = maskval(v, d); + bp = 1 << slot; + /* OR bitposition before sz calculation to handle + * redundant collisions */ + hdr |= bp; + sz = hashmap_bitcount(hdr); + hp = erts_produce_heap(factory, HAMT_NODE_BITMAP_SZ(sz), HALLOC_EXTRA); + nhp = hp; + *hp++ = (hdr == 0xffff) ? MAP_HEADER_HAMT_NODE_ARRAY : MAP_HEADER_HAMT_NODE_BITMAP(hdr); + *hp++ = res; sz--; + while (sz--) { *hp++ = ESTACK_POP(stack); } + ASSERT((hp - nhp) < 18); + res = make_hashmap(nhp); + + /* we need to pop the next hdr and push if we don't need it */ + + hdr = ESTACK_POP(stack); + d--; + } + ESTACK_PUSH(stack, res); + ESTACK_PUSH(stack, hdr); } - BIF_RET(am_false); + + vp = v; + v = vn; + d = dn; + ERTS_FACTORY_HOLE_CHECK(factory); + } + + /* v and vp are reused from above */ + dn = cdepth(vp,v); + ix = ix + hxns[ix].skip; + res = hxns[ix].val; + + if (hxns[ix].skip > 1) { + dc = 7; + /* build collision nodes */ + while (dc > dn) { + hp = erts_produce_heap(factory, HAMT_NODE_BITMAP_SZ(1), HALLOC_EXTRA); + hp[0] = MAP_HEADER_HAMT_NODE_BITMAP(1 << maskval(v,dc)); + hp[1] = res; + res = make_hashmap(hp); + dc--; + } + } + + hdr = ESTACK_POP(stack); + /* pop remaining subtree if any */ + while (dn) { + slot = maskval(v, dn); + bp = 1 << slot; + /* OR bitposition before sz calculation to handle + * redundant collisions */ + hdr |= bp; + sz = hashmap_bitcount(hdr); + hp = erts_produce_heap(factory, HAMT_NODE_BITMAP_SZ(sz), HALLOC_EXTRA); + nhp = hp; + *hp++ = (hdr == 0xffff) ? MAP_HEADER_HAMT_NODE_ARRAY : MAP_HEADER_HAMT_NODE_BITMAP(hdr); + *hp++ = res; sz--; + + while (sz--) { *hp++ = ESTACK_POP(stack); } + res = make_hashmap(nhp); + hdr = ESTACK_POP(stack); + dn--; + } + + /* and finally the root .. */ + + slot = maskval(v, dn); + bp = 1 << slot; + hdr |= bp; + sz = hashmap_bitcount(hdr); + hp = erts_produce_heap(factory, sz + /* hdr + item */ (is_root ? 2 : 1), 0); + nhp = hp; + + if (is_root) { + *hp++ = (hdr == 0xffff) ? MAP_HEADER_HAMT_HEAD_ARRAY : MAP_HEADER_HAMT_HEAD_BITMAP(hdr); + *hp++ = n; + } else { + *hp++ = (hdr == 0xffff) ? MAP_HEADER_HAMT_NODE_ARRAY : MAP_HEADER_HAMT_NODE_BITMAP(hdr); + } + + *hp++ = res; sz--; + while (sz--) { *hp++ = ESTACK_POP(stack); } + + res = make_hashmap(nhp); + + ASSERT(ESTACK_COUNT(stack) == 0); + DESTROY_ESTACK(stack); + ERTS_FACTORY_HOLE_CHECK(factory); + return res; +} +#undef HALLOC_EXTRA + +static int hxnodecmpkey(hxnode_t *a, hxnode_t *b) { + return CMP_TERM(CAR(list_val(a->val)), CAR(list_val(b->val))); +} + +static int hxnodecmp(hxnode_t *a, hxnode_t *b) { + if (a->hx < b->hx) + return 1; + else if (a->hx == b->hx) + return 0; + else + return -1; +} + +/* maps:is_key/2 */ + +BIF_RETTYPE maps_is_key_2(BIF_ALIST_2) { + if (is_map(BIF_ARG_2)) { + BIF_RET(erts_maps_get(BIF_ARG_1, BIF_ARG_2) ? am_true : am_false); } BIF_ERROR(BIF_P, BADARG); } -/* maps:keys/1 - */ +/* maps:keys/1 */ BIF_RETTYPE maps_keys_1(BIF_ALIST_1) { - if (is_map(BIF_ARG_1)) { + if (is_flatmap(BIF_ARG_1)) { Eterm *hp, *ks, res = NIL; - map_t *mp; + flatmap_t *mp; Uint n; - mp = (map_t*)map_val(BIF_ARG_1); - n = map_get_size(mp); + mp = (flatmap_t*)flatmap_val(BIF_ARG_1); + n = flatmap_get_size(mp); if (n == 0) BIF_RET(res); hp = HAlloc(BIF_P, (2 * n)); - ks = map_get_keys(mp); + ks = flatmap_get_keys(mp); while(n--) { res = CONS(hp, ks[n], res); hp += 2; } BIF_RET(res); + } else if (is_hashmap(BIF_ARG_1)) { + BIF_RET(hashmap_keys(BIF_P, BIF_ARG_1)); } BIF_ERROR(BIF_P, BADARG); } -/* maps:merge/2 - */ +/* 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++; - } + if (is_flatmap(BIF_ARG_1)) { + if (is_flatmap(BIF_ARG_2)) { + BIF_RET(flatmap_merge(BIF_P, BIF_ARG_1, BIF_ARG_2)); + } else if (is_hashmap(BIF_ARG_2)) { + /* Will always become a tree */ + BIF_RET(map_merge_mixed(BIF_P, BIF_ARG_1, BIF_ARG_2, 0)); } + } else if (is_hashmap(BIF_ARG_1)) { + if (is_hashmap(BIF_ARG_2)) { + BIF_RET(hashmap_merge(BIF_P, BIF_ARG_1, BIF_ARG_2)); + } else if (is_flatmap(BIF_ARG_2)) { + /* Will always become a tree */ + BIF_RET(map_merge_mixed(BIF_P, BIF_ARG_2, BIF_ARG_1, 1)); + } + } + BIF_ERROR(BIF_P, BADARG); +} + +static Eterm flatmap_merge(Process *p, Eterm nodeA, Eterm nodeB) { + Eterm *hp,*thp; + Eterm tup; + Eterm *ks,*vs,*ks1,*vs1,*ks2,*vs2; + flatmap_t *mp1,*mp2,*mp_new; + Uint n,n1,n2,i1,i2,need,unused_size=0; + int c = 0; + + mp1 = (flatmap_t*)flatmap_val(nodeA); + mp2 = (flatmap_t*)flatmap_val(nodeB); + n1 = flatmap_get_size(mp1); + n2 = flatmap_get_size(mp2); - /* copy remaining */ - while (i1 < n1) { + need = MAP_HEADER_SIZE + 1 + 2*(n1 + n2); + + hp = HAlloc(p, need); + thp = hp; + tup = make_tuple(thp); + ks = hp + 1; hp += 1 + n1 + n2; + mp_new = (flatmap_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 = flatmap_get_keys(mp1); + vs1 = flatmap_get_values(mp1); + ks2 = flatmap_get_keys(mp2); + vs2 = flatmap_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++; - } - - while (i2 < n2) { + } else { *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 - */ + /* copy remaining */ + while (i1 < n1) { + *ks++ = ks1[i1]; + *vs++ = vs1[i1]; + i1++; + } - *ks = make_pos_bignum_header(unused_size - 1); - HRelease(BIF_P, vs + unused_size, vs); - } + while (i2 < n2) { + *ks++ = ks2[i2]; + *vs++ = vs2[i2]; + i2++; + } - mp_new->size = n1 + n2 - unused_size; - *thp = make_arityval(n1 + n2 - unused_size); + 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 + */ - BIF_RET(make_map(mp_new)); + *ks = make_pos_bignum_header(unused_size - 1); + HRelease(p, vs + unused_size, vs); } - BIF_ERROR(BIF_P, BADARG); -} -/* maps:new/2 - */ -BIF_RETTYPE maps_new_0(BIF_ALIST_0) { - Eterm* hp; - Eterm tup; - map_t *mp; + n = n1 + n2 - unused_size; + *thp = make_arityval(n); - hp = HAlloc(BIF_P, (MAP_HEADER_SIZE + 1)); - tup = make_tuple(hp); - *hp++ = make_arityval(0); + /* Reshape map to a hashmap if the map exceeds the limit */ - mp = (map_t*)hp; - mp->thing_word = MAP_HEADER; - mp->size = 0; - mp->keys = tup; + if (n > MAP_SMALL_MAP_LIMIT) { + Uint32 hx,sw; + Uint i; + Eterm res; + hxnode_t *hxns; + ErtsHeapFactory factory; - BIF_RET(make_map(mp)); -} + ks = flatmap_get_keys(mp_new); + vs = flatmap_get_values(mp_new); -/* maps:put/3 - */ + hp = HAlloc(p, 2 * n); -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); + hxns = (hxnode_t *)erts_alloc(ERTS_ALC_T_TMP,n * sizeof(hxnode_t)); - n = map_get_size(mp); + for (i = 0; i < n; i++) { + hx = hashmap_make_hash(ks[i]); + swizzle32(sw,hx); + hxns[i].hx = sw; + hxns[i].val = CONS(hp, ks[i], vs[i]); hp += 2; + hxns[i].skip = 1; + hxns[i].i = i; + } - 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; + factory.p = p; + res = hashmap_from_unsorted_array(&factory, hxns, n, 0); + + erts_free(ERTS_ALC_T_TMP, (void *) hxns); + ERTS_VERIFY_UNUSED_TEMP_ALLOC(p); return res; } - ks = map_get_keys(mp); - vs = map_get_values(mp); + mp_new->size = n; + + return make_flatmap(mp_new); +} + +static Eterm map_merge_mixed(Process *p, Eterm flat, Eterm tree, int swap_args) { + Eterm *ks, *vs, *hp, res; + flatmap_t *mp; + Uint n, i; + hxnode_t *hxns; + Uint32 sw, hx; + ErtsHeapFactory factory; + + /* convert flat to tree */ + + ASSERT(is_flatmap(flat)); + ASSERT(is_hashmap(tree)); + + mp = (flatmap_t*)flatmap_val(flat); + n = flatmap_get_size(mp); - /* only allocate for values, - * assume key-tuple will be intact + ks = flatmap_get_keys(mp); + vs = flatmap_get_values(mp); + + hp = HAlloc(p, 2 * n); + + hxns = (hxnode_t *)erts_alloc(ERTS_ALC_T_TMP, n * sizeof(hxnode_t)); + + for (i = 0; i < n; i++) { + hx = hashmap_make_hash(ks[i]); + swizzle32(sw,hx); + hxns[i].hx = sw; + hxns[i].val = CONS(hp, ks[i], vs[i]); hp += 2; + hxns[i].skip = 1; + hxns[i].i = i; + } + + factory.p = p; + res = hashmap_from_unsorted_array(&factory, hxns, n, 0); + + erts_free(ERTS_ALC_T_TMP, (void *) hxns); + ERTS_VERIFY_UNUSED_TEMP_ALLOC(p); + + return swap_args ? hashmap_merge(p, tree, res) : hashmap_merge(p, res, tree); +} + +#define HALLOC_EXTRA 200 + +static Eterm hashmap_merge(Process *p, Eterm nodeA, Eterm nodeB) { +#define PSTACK_TYPE struct HashmapMergePStackType + struct HashmapMergePStackType { + Eterm *srcA, *srcB; + Uint32 abm, bbm, rbm; /* node bitmaps */ + int keepA; + int ix; + Eterm array[16]; + }; + PSTACK_DECLARE(s, 4); + struct HashmapMergePStackType* sp = PSTACK_PUSH(s); + Eterm *hp, *nhp; + Eterm hdrA, hdrB; + Eterm th[2]; + Uint32 ahx, bhx; + Uint size; /* total key-value counter */ + int keepA = 0; + unsigned lvl = 0; + Eterm res = THE_NON_VALUE; + + /* + * Strategy: Do depth-first traversal of both trees (at the same time) + * and merge each pair of nodes. */ - 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; + { + hashmap_head_t* a = (hashmap_head_t*) hashmap_val(nodeA); + hashmap_head_t* b = (hashmap_head_t*) hashmap_val(nodeB); + size = a->size + b->size; + } + +recurse: + + if (primary_tag(nodeA) == TAG_PRIMARY_BOXED && + primary_tag(nodeB) == TAG_PRIMARY_LIST) { + /* Avoid implementing this combination by switching places */ + Eterm tmp = nodeA; + nodeA = nodeB; + nodeB = tmp; + keepA = !keepA; + } + + switch (primary_tag(nodeA)) { + case TAG_PRIMARY_LIST: { + sp->srcA = list_val(nodeA); + switch (primary_tag(nodeB)) { + case TAG_PRIMARY_LIST: { /* LEAF + LEAF */ + sp->srcB = list_val(nodeB); + + if (EQ(CAR(sp->srcA), CAR(sp->srcB))) { + --size; + res = keepA ? nodeA : nodeB; } else { - *hp++ = *vs++; + ahx = hashmap_restore_hash(th, lvl, CAR(sp->srcA)); + bhx = hashmap_restore_hash(th, lvl, CAR(sp->srcB)); + sp->abm = 1 << hashmap_index(ahx); + sp->bbm = 1 << hashmap_index(bhx); + + sp->srcA = &nodeA; + sp->srcB = &nodeB; } + break; } - } else { - for( i = 0; i < n; i ++) { - if (EQ(ks[i], key)) { - *hp++ = value; - vs++; - c = 1; - } else { - *hp++ = *vs++; + case TAG_PRIMARY_BOXED: { /* LEAF + NODE */ + sp->srcB = boxed_val(nodeB); + ASSERT(is_header(*sp->srcB)); + hdrB = *sp->srcB++; + + ahx = hashmap_restore_hash(th, lvl, CAR(sp->srcA)); + sp->abm = 1 << hashmap_index(ahx); + sp->srcA = &nodeA; + switch(hdrB & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: sp->srcB++; + case HAMT_SUBTAG_NODE_ARRAY: + sp->bbm = 0xffff; + break; + + case HAMT_SUBTAG_HEAD_BITMAP: sp->srcB++; + case HAMT_SUBTAG_NODE_BITMAP: + sp->bbm = MAP_HEADER_VAL(hdrB); + break; + + default: + erl_exit(1, "bad header tag %ld\r\n", *sp->srcB & _HEADER_MAP_SUBTAG_MASK); + break; } + break; } + default: + erl_exit(1, "bad primary tag %ld\r\n", nodeB); + } + break; } + case TAG_PRIMARY_BOXED: { /* NODE + NODE */ + sp->srcA = boxed_val(nodeA); + hdrA = *sp->srcA++; + ASSERT(is_header(hdrA)); + switch (hdrA & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: sp->srcA++; + case HAMT_SUBTAG_NODE_ARRAY: { + ASSERT(primary_tag(nodeB) == TAG_PRIMARY_BOXED); + sp->abm = 0xffff; + sp->srcB = boxed_val(nodeB); + hdrB = *sp->srcB++; + ASSERT(is_header(hdrB)); + switch (hdrB & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: sp->srcB++; + case HAMT_SUBTAG_NODE_ARRAY: + sp->bbm = 0xffff; + break; + case HAMT_SUBTAG_HEAD_BITMAP: sp->srcB++; + case HAMT_SUBTAG_NODE_BITMAP: + sp->bbm = MAP_HEADER_VAL(hdrB); + break; + default: + erl_exit(1, "bad header tag %ld\r\n", *sp->srcB & _HEADER_MAP_SUBTAG_MASK); + } + break; + } + case HAMT_SUBTAG_HEAD_BITMAP: sp->srcA++; + case HAMT_SUBTAG_NODE_BITMAP: { + ASSERT(primary_tag(nodeB) == TAG_PRIMARY_BOXED); + sp->abm = MAP_HEADER_VAL(hdrA); + sp->srcB = boxed_val(nodeB); + hdrB = *sp->srcB++; + ASSERT(is_header(hdrB)); + switch (hdrB & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: sp->srcB++; + case HAMT_SUBTAG_NODE_ARRAY: + sp->bbm = 0xffff; + break; + case HAMT_SUBTAG_HEAD_BITMAP: sp->srcB++; + case HAMT_SUBTAG_NODE_BITMAP: + sp->bbm = MAP_HEADER_VAL(hdrB); + break; - if (c) - return res; + default: + erl_exit(1, "bad header tag %ld\r\n", *sp->srcB & _HEADER_MAP_SUBTAG_MASK); + } + break; + } + default: + erl_exit(1, "bad primary tag %ld\r\n", nodeA); + } + break; + } + default: + erl_exit(1, "bad primary tag %ld\r\n", nodeA); + } - /* 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); + for (;;) { + if (is_value(res)) { /* We have a complete (sub-)tree or leaf */ + if (lvl == 0) + break; - hp = HAlloc(p, 3 + n + 1); - res = make_map(hp); - *hp++ = MAP_HEADER; - *hp++ = n + 1; - *hp++ = tup; + /* Pop from stack and continue build parent node */ + lvl--; + sp = PSTACK_POP(s); + sp->array[sp->ix++] = res; + res = THE_NON_VALUE; + if (sp->rbm) { + sp->srcA++; + sp->srcB++; + keepA = sp->keepA; + } + } else { /* Start build a node */ + sp->ix = 0; + sp->rbm = sp->abm | sp->bbm; + ASSERT(!(sp->rbm == 0 && lvl > 0)); + } + + while (sp->rbm) { + Uint32 next = sp->rbm & (sp->rbm-1); + Uint32 bit = sp->rbm ^ next; + sp->rbm = next; + if (sp->abm & bit) { + if (sp->bbm & bit) { + /* Bit clash. Push and resolve by recursive merge */ + if (sp->rbm) { + sp->keepA = keepA; + } + nodeA = *sp->srcA; + nodeB = *sp->srcB; + lvl++; + sp = PSTACK_PUSH(s); + goto recurse; + } else { + sp->array[sp->ix++] = *sp->srcA++; + } + } else { + ASSERT(sp->bbm & bit); + sp->array[sp->ix++] = *sp->srcB++; + } + } - ks = map_get_keys(mp); - vs = map_get_values(mp); + ASSERT(sp->ix == hashmap_bitcount(sp->abm | sp->bbm)); + if (lvl == 0) { + nhp = HAllocX(p, HAMT_HEAD_BITMAP_SZ(sp->ix), HALLOC_EXTRA); + hp = nhp; + *hp++ = (sp->ix == 16 ? MAP_HEADER_HAMT_HEAD_ARRAY + : MAP_HEADER_HAMT_HEAD_BITMAP(sp->abm | sp->bbm)); + *hp++ = size; + } else { + nhp = HAllocX(p, HAMT_NODE_BITMAP_SZ(sp->ix), HALLOC_EXTRA); + hp = nhp; + *hp++ = (sp->ix == 16 ? make_arityval(16) + : MAP_HEADER_HAMT_NODE_BITMAP(sp->abm | sp->bbm)); + } + memcpy(hp, sp->array, sp->ix * sizeof(Eterm)); + res = make_boxed(nhp); + } + PSTACK_DESTROY(s); + return res; +} - ASSERT(n >= 0); +static int hash_cmp(Uint32 ha, Uint32 hb) +{ + int i; + for (i=0; i<8; i++) { + int cmp = (int)(ha & 0xF) - (int)(hb & 0xF); + if (cmp) + return cmp; + ha >>= 4; + hb >>= 4; + } + return 0; +} - /* copy map in order */ - while (n && ((c = CMP_TERM(*ks, key)) < 0)) { - *shp++ = *ks++; - *hp++ = *vs++; - n--; +int hashmap_key_hash_cmp(Eterm* ap, Eterm* bp) +{ + Eterm th[2]; + unsigned lvl = 0; + + if (ap && bp) { + ASSERT(CMP_TERM(CAR(ap), CAR(bp)) != 0); + for (;;) { + Uint32 ha = hashmap_restore_hash(th, lvl, CAR(ap)); + Uint32 hb = hashmap_restore_hash(th, lvl, CAR(bp)); + int cmp = hash_cmp(ha, hb); + if (cmp) + return cmp; + lvl += 8; + } } + return ap ? -1 : 1; +} - *shp++ = key; - *hp++ = value; +/* maps:new/0 */ - ASSERT(n >= 0); +BIF_RETTYPE maps_new_0(BIF_ALIST_0) { + Eterm* hp; + Eterm tup; + flatmap_t *mp; - 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; + hp = HAlloc(BIF_P, (MAP_HEADER_SIZE + 1)); + tup = make_tuple(hp); + *hp++ = make_arityval(0); + + mp = (flatmap_t*)hp; + mp->thing_word = MAP_HEADER; + mp->size = 0; + mp->keys = tup; + + BIF_RET(make_flatmap(mp)); } +/* maps:put/3 */ + 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)); @@ -609,81 +1366,87 @@ BIF_RETTYPE maps_put_3(BIF_ALIST_3) { BIF_ERROR(BIF_P, BADARG); } -/* maps:remove/3 - */ +/* 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); + Uint32 hx; + if (is_flatmap(map)) { + Sint n; + Uint need; + Eterm *hp_start; + Eterm *thp, *mhp; + Eterm *ks, *vs, tup; + flatmap_t *mp = (flatmap_t*)flatmap_val(map); + + n = flatmap_get_size(mp); + + if (n == 0) { + *res = map; + return 1; + } - n = map_get_size(mp); + ks = flatmap_get_keys(mp); + vs = flatmap_get_values(mp); - if (n == 0) { - *res = map; - return 1; - } + /* Assume key exists. + * Release allocated if it didn't. + * Allocate key tuple first. + */ - ks = map_get_keys(mp); - vs = map_get_values(mp); + 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 */ - /* Assume key exists. - * Release allocated if it didn't. - * Allocate key tuple first. - */ + tup = make_tuple(thp); + *thp++ = make_arityval(n - 1); - 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 */ + *res = make_flatmap(mhp); + *mhp++ = MAP_HEADER; + *mhp++ = n - 1; + *mhp++ = tup; - 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 (1) { - if (*ks == key) { - goto found_key; - } else if (--n) { - *mhp++ = *vs++; - *thp++ = *ks++; - } else - break; - } - } else { - while(1) { - if (EQ(*ks, key)) { - goto found_key; - } else if (--n) { - *mhp++ = *vs++; - *thp++ = *ks++; - } else - break; + if (is_immed(key)) { + while (1) { + if (*ks == key) { + goto found_key; + } else if (--n) { + *mhp++ = *vs++; + *thp++ = *ks++; + } else + break; + } + } else { + while(1) { + if (EQ(*ks, key)) { + goto found_key; + } else if (--n) { + *mhp++ = *vs++; + *thp++ = *ks++; + } else + break; + } } - } - /* Not found, remove allocated memory - * and return previous map. - */ - HRelease(p, hp_start + need, hp_start); + /* Not found, remove allocated memory + * and return previous map. + */ + HRelease(p, hp_start + need, hp_start); - *res = map; - return 1; + *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)); + /* 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; } + ASSERT(is_hashmap(map)); + hx = hashmap_make_hash(key); + *res = hashmap_delete(p, hx, key, map); return 1; } @@ -697,21 +1460,20 @@ BIF_RETTYPE maps_remove_2(BIF_ALIST_2) { BIF_ERROR(BIF_P, BADARG); } -/* maps:update/3 - */ - int erts_maps_update(Process *p, Eterm key, Eterm value, Eterm map, Eterm *res) { + Uint32 hx; + if (is_flatmap(map)) { Sint n,i; Eterm* hp,*shp; Eterm *ks,*vs; - map_t *mp = (map_t*)map_val(map); + flatmap_t *mp = (flatmap_t*)flatmap_val(map); - if ((n = map_get_size(mp)) == 0) { + if ((n = flatmap_get_size(mp)) == 0) { return 0; } - ks = map_get_keys(mp); - vs = map_get_values(mp); + ks = flatmap_get_keys(mp); + vs = flatmap_get_values(mp); /* only allocate for values, * assume key-tuple will be intact @@ -749,10 +1511,147 @@ found_key: vs++; if (++i < n) sys_memcpy(hp, vs, (n - i)*sizeof(Eterm)); - *res = make_map(shp); + *res = make_flatmap(shp); return 1; + } + + ASSERT(is_hashmap(map)); + hx = hashmap_make_hash(key); + *res = erts_hashmap_insert(p, hx, key, value, map, 1); + if (is_value(*res)) + return 1; + + return 0; } +Eterm erts_maps_put(Process *p, Eterm key, Eterm value, Eterm map) { + Uint32 hx; + Eterm res; + if (is_flatmap(map)) { + Sint n,i; + Sint c = 0; + Eterm* hp, *shp; + Eterm *ks, *vs, tup; + flatmap_t *mp = (flatmap_t*)flatmap_val(map); + + n = flatmap_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_flatmap(hp); + *hp++ = MAP_HEADER; + *hp++ = 1; + *hp++ = tup; + *hp++ = value; + + return res; + } + + ks = flatmap_get_keys(mp); + vs = flatmap_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_flatmap(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; + + /* the map will grow */ + + if (n >= MAP_SMALL_MAP_LIMIT) { + HRelease(p, shp + MAP_HEADER_SIZE + n, shp); + ks = flatmap_get_keys(mp); + vs = flatmap_get_values(mp); + + res = erts_hashmap_from_ks_and_vs_extra(p,ks,vs,n,key,value); + + return res; + } + + /* still a small map. 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_flatmap(hp); + *hp++ = MAP_HEADER; + *hp++ = n + 1; + *hp++ = tup; + + ks = flatmap_get_keys(mp); + vs = flatmap_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; + } + ASSERT(is_hashmap(map)); + + hx = hashmap_make_hash(key); + res = erts_hashmap_insert(p, hx, key, value, map, 0); + ASSERT(is_hashmap(res)); + + return res; +} + +/* maps:update/3 */ + BIF_RETTYPE maps_update_3(BIF_ALIST_3) { if (is_map(BIF_ARG_3)) { Eterm res; @@ -764,38 +1663,845 @@ BIF_RETTYPE maps_update_3(BIF_ALIST_3) { } -/* maps:values/1 - */ +/* maps:values/1 */ BIF_RETTYPE maps_values_1(BIF_ALIST_1) { - if (is_map(BIF_ARG_1)) { + if (is_flatmap(BIF_ARG_1)) { Eterm *hp, *vs, res = NIL; - map_t *mp; + flatmap_t *mp; Uint n; - mp = (map_t*)map_val(BIF_ARG_1); - n = map_get_size(mp); + mp = (flatmap_t*)flatmap_val(BIF_ARG_1); + n = flatmap_get_size(mp); if (n == 0) BIF_RET(res); hp = HAlloc(BIF_P, (2 * n)); - vs = map_get_values(mp); + vs = flatmap_get_values(mp); while(n--) { res = CONS(hp, vs[n], res); hp += 2; } BIF_RET(res); + } else if (is_hashmap(BIF_ARG_1)) { + BIF_RET(hashmap_values(BIF_P, BIF_ARG_1)); } BIF_ERROR(BIF_P, BADARG); } -int erts_validate_and_sort_map(map_t* mp) +static Eterm hashmap_to_list(Process *p, Eterm node) { + DECLARE_WSTACK(stack); + Eterm *hp, *kv; + Eterm res = NIL; + + hp = HAlloc(p, hashmap_size(node) * (2 + 3)); + hashmap_iterator_init(&stack, node, 0); + while ((kv=hashmap_iterator_next(&stack)) != NULL) { + Eterm tup = TUPLE2(hp, CAR(kv), CDR(kv)); + hp += 3; + res = CONS(hp, tup, res); + hp += 2; + } + DESTROY_WSTACK(stack); + return res; +} + +void hashmap_iterator_init(ErtsWStack* s, Eterm node, int reverse) { + Eterm hdr = *hashmap_val(node); + Uint sz; + + switch(hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: + sz = 16; + break; + case HAMT_SUBTAG_HEAD_BITMAP: + sz = hashmap_bitcount(MAP_HEADER_VAL(hdr)); + break; + default: + erl_exit(1, "bad header"); + } + + WSTACK_PUSH3((*s), (UWord)THE_NON_VALUE, /* end marker */ + (UWord)(!reverse ? 0 : sz+1), + (UWord)node); +} + +Eterm* hashmap_iterator_next(ErtsWStack* s) { + Eterm node, *ptr, hdr; + Uint32 sz; + Uint idx; + + for (;;) { + ASSERT(!WSTACK_ISEMPTY((*s))); + node = (Eterm) WSTACK_POP((*s)); + if (is_non_value(node)) { + return NULL; + } + idx = (Uint) WSTACK_POP((*s)); + for (;;) { + ASSERT(is_boxed(node)); + ptr = boxed_val(node); + hdr = *ptr; + ASSERT(is_header(hdr)); + switch(hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: + ptr++; + case HAMT_SUBTAG_NODE_ARRAY: + sz = 16; + break; + case HAMT_SUBTAG_HEAD_BITMAP: + ptr++; + case HAMT_SUBTAG_NODE_BITMAP: + sz = hashmap_bitcount(MAP_HEADER_VAL(hdr)); + ASSERT(sz < 17); + break; + default: + erl_exit(1, "bad header"); + } + + idx++; + + if (idx <= sz) { + WSTACK_PUSH2((*s), (UWord)idx, (UWord)node); + + if (is_list(ptr[idx])) { + return list_val(ptr[idx]); + } + ASSERT(is_boxed(ptr[idx])); + node = ptr[idx]; + idx = 0; + } + else + break; /* and pop parent node */ + } + } +} + +Eterm* hashmap_iterator_prev(ErtsWStack* s) { + Eterm node, *ptr, hdr; + Uint32 sz; + Uint idx; + + for (;;) { + ASSERT(!WSTACK_ISEMPTY((*s))); + node = (Eterm) WSTACK_POP((*s)); + if (is_non_value(node)) { + return NULL; + } + idx = (Uint) WSTACK_POP((*s)); + for (;;) { + ASSERT(is_boxed(node)); + ptr = boxed_val(node); + hdr = *ptr; + ASSERT(is_header(hdr)); + switch(hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: + ptr++; + case HAMT_SUBTAG_NODE_ARRAY: + sz = 16; + break; + case HAMT_SUBTAG_HEAD_BITMAP: + ptr++; + case HAMT_SUBTAG_NODE_BITMAP: + sz = hashmap_bitcount(MAP_HEADER_VAL(hdr)); + ASSERT(sz < 17); + break; + default: + erl_exit(1, "bad header"); + } + + if (idx > sz) + idx = sz; + else + idx--; + + if (idx >= 1) { + WSTACK_PUSH2((*s), (UWord)idx, (UWord)node); + + if (is_list(ptr[idx])) { + return list_val(ptr[idx]); + } + ASSERT(is_boxed(ptr[idx])); + node = ptr[idx]; + idx = 17; + } + else + break; /* and pop parent node */ + } + } +} + +const Eterm * +#if HALFWORD_HEAP +erts_hashmap_get_rel(Uint32 hx, Eterm key, Eterm node, Eterm *map_base) +#else +erts_hashmap_get(Uint32 hx, Eterm key, Eterm node) +#endif { - Eterm *ks = map_get_keys(mp); - Eterm *vs = map_get_values(mp); - Uint sz = map_get_size(mp); + Eterm *ptr, hdr; + Uint ix,slot, lvl = 0; + Uint32 hval,bp; + DeclareTmpHeapNoproc(th,2); + UseTmpHeapNoproc(2); + + for (;;) { + switch(primary_tag(node)) { + case TAG_PRIMARY_LIST: /* LEAF NODE [K|V] */ + ptr = list_val(node); + UnUseTmpHeapNoproc(2); + + if (eq_rel(CAR(ptr), map_base, key, NULL)) { + return &(CDR(ptr)); + } + return NULL; + case TAG_PRIMARY_BOXED: + ptr = boxed_val(node); + hdr = *ptr; + ASSERT(is_header(hdr)); + + switch(hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_NODE_ARRAY: + ix = hashmap_index(hx); + hx = hashmap_shift_hash(th,hx,lvl,key); + node = ptr[ix+1]; + break; + case HAMT_SUBTAG_HEAD_ARRAY: + ix = hashmap_index(hx); + hx = hashmap_shift_hash(th,hx,lvl,key); + node = ptr[ix+2]; + break; + case HAMT_SUBTAG_NODE_BITMAP: + hval = MAP_HEADER_VAL(hdr); + ix = hashmap_index(hx); + bp = 1 << ix; + slot = hashmap_bitcount(hval & (bp - 1)); + + /* occupied */ + if (bp & hval) { + hx = hashmap_shift_hash(th,hx,lvl,key); + node = ptr[slot+1]; + break; + } + /* not occupied */ + UnUseTmpHeapNoproc(2); + return NULL; + case HAMT_SUBTAG_HEAD_BITMAP: + hval = MAP_HEADER_VAL(hdr); + ix = hashmap_index(hx); + bp = 1 << ix; + slot = hashmap_bitcount(hval & (bp - 1)); + + /* occupied */ + if (bp & hval) { + hx = hashmap_shift_hash(th,hx,lvl,key); + node = ptr[slot+2]; + break; + } + /* not occupied */ + UnUseTmpHeapNoproc(2); + return NULL; + default: + erl_exit(1, "bad header tag %ld\r\n", hdr & _HEADER_MAP_SUBTAG_MASK); + break; + } + break; + default: + erl_exit(1, "bad primary tag %p\r\n", node); + break; + } + } + UnUseTmpHeapNoproc(2); + return NULL; +} + +Eterm erts_hashmap_insert(Process *p, Uint32 hx, Eterm key, Eterm value, + Eterm map, int is_update) { + Uint size, upsz; + Eterm *hp, res = THE_NON_VALUE; + DECLARE_ESTACK(stack); + if (erts_hashmap_insert_down(hx, key, map, &size, &upsz, &stack, is_update)) { + hp = HAlloc(p, size); + res = erts_hashmap_insert_up(hp, key, value, &upsz, &stack); + } + + DESTROY_ESTACK(stack); + ERTS_VERIFY_UNUSED_TEMP_ALLOC(p); + ERTS_HOLE_CHECK(p); + + return res; +} + + +int erts_hashmap_insert_down(Uint32 hx, Eterm key, Eterm node, Uint *sz, + Uint *update_size, ErtsEStack *sp, int is_update) { + Eterm *ptr; + Eterm hdr, ckey; + Eterm th[2]; + Uint32 ix, cix, bp, hval, chx; + Uint slot, lvl = 0, clvl; + Uint size = 0, n = 0; + + *update_size = 1; + + for (;;) { + switch(primary_tag(node)) { + case TAG_PRIMARY_LIST: /* LEAF NODE [K|V] */ + ptr = list_val(node); + ckey = CAR(ptr); + if (EQ(ckey, key)) { + *update_size = 0; + goto unroll; + } + if (is_update) { + return 0; + } + goto insert_subnodes; + case TAG_PRIMARY_BOXED: + ptr = boxed_val(node); + hdr = *ptr; + ASSERT(is_header(hdr)); + + switch(hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_NODE_ARRAY: + ix = hashmap_index(hx); + hx = hashmap_shift_hash(th,hx,lvl,key); + size += HAMT_NODE_ARRAY_SZ; + ESTACK_PUSH2(*sp, ix, node); + node = ptr[ix+1]; + break; + case HAMT_SUBTAG_HEAD_ARRAY: + ix = hashmap_index(hx); + hx = hashmap_shift_hash(th,hx,lvl,key); + size += HAMT_HEAD_ARRAY_SZ; + ESTACK_PUSH2(*sp, ix, node); + node = ptr[ix+2]; + break; + case HAMT_SUBTAG_NODE_BITMAP: + hval = MAP_HEADER_VAL(hdr); + ix = hashmap_index(hx); + bp = 1 << ix; + slot = hashmap_bitcount(hval & (bp - 1)); + n = hashmap_bitcount(hval); + + ESTACK_PUSH(*sp, n); + ESTACK_PUSH3(*sp, bp, slot, node); + + /* occupied */ + if (bp & hval) { + hx = hashmap_shift_hash(th,hx,lvl,key); + node = ptr[slot+1]; + ASSERT(HAMT_NODE_BITMAP_SZ(n) <= 17); + size += HAMT_NODE_BITMAP_SZ(n); + break; + } + /* not occupied */ + if (is_update) { + return 0; + } + size += HAMT_NODE_BITMAP_SZ(n+1); + goto unroll; + case HAMT_SUBTAG_HEAD_BITMAP: + hval = MAP_HEADER_VAL(hdr); + ix = hashmap_index(hx); + bp = 1 << ix; + slot = hashmap_bitcount(hval & (bp - 1)); + n = hashmap_bitcount(hval); + + ESTACK_PUSH(*sp, n); + ESTACK_PUSH3(*sp, bp, slot, node); + + /* occupied */ + if (bp & hval) { + hx = hashmap_shift_hash(th,hx,lvl,key); + node = ptr[slot+2]; + ASSERT(HAMT_HEAD_BITMAP_SZ(n) <= 18); + size += HAMT_HEAD_BITMAP_SZ(n); + break; + } + /* not occupied */ + if (is_update) { + return 0; + } + size += HAMT_HEAD_BITMAP_SZ(n+1); + goto unroll; + default: + erl_exit(1, "bad header tag %ld\r\n", hdr & _HEADER_MAP_SUBTAG_MASK); + break; + } + break; + default: + erl_exit(1, "bad primary tag %p\r\n", node); + break; + } + } +insert_subnodes: + clvl = lvl; + chx = hashmap_restore_hash(th,clvl,ckey); + size += HAMT_NODE_BITMAP_SZ(2); + ix = hashmap_index(hx); + cix = hashmap_index(chx); + + while (cix == ix) { + ESTACK_PUSH(*sp, 0); + ESTACK_PUSH3(*sp, 1 << ix, 0, MAP_HEADER_HAMT_NODE_BITMAP(0)); + size += HAMT_NODE_BITMAP_SZ(1); + hx = hashmap_shift_hash(th,hx,lvl,key); + chx = hashmap_shift_hash(th,chx,clvl,ckey); + ix = hashmap_index(hx); + cix = hashmap_index(chx); + } + ESTACK_PUSH3(*sp, cix, ix, node); + +unroll: + *sz = size + /* res cons */ 2; + return 1; +} + +Eterm erts_hashmap_insert_up(Eterm *hp, Eterm key, Eterm value, + Uint *update_size, ErtsEStack *sp) { + Eterm node, *ptr, hdr; + Eterm res; + Eterm *nhp = NULL; + Uint32 ix, cix, bp, hval; + Uint slot, n; + /* Needed for halfword */ + DeclareTmpHeapNoproc(fake,1); + UseTmpHeapNoproc(1); + + res = CONS(hp, key, value); hp += 2; + + do { + node = ESTACK_POP(*sp); + switch(primary_tag(node)) { + case TAG_PRIMARY_LIST: + ix = (Uint32) ESTACK_POP(*sp); + cix = (Uint32) ESTACK_POP(*sp); + + nhp = hp; + *hp++ = MAP_HEADER_HAMT_NODE_BITMAP((1 << ix) | (1 << cix)); + if (ix < cix) { + *hp++ = res; + *hp++ = node; + } else { + *hp++ = node; + *hp++ = res; + } + res = make_hashmap(nhp); + break; + case TAG_PRIMARY_HEADER: + /* subnodes, fake it */ + *fake = node; + node = make_boxed(fake); + case TAG_PRIMARY_BOXED: + ptr = boxed_val(node); + hdr = *ptr; + ASSERT(is_header(hdr)); + + switch(hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_NODE_ARRAY: + slot = (Uint) ESTACK_POP(*sp); + nhp = hp; + n = HAMT_NODE_ARRAY_SZ; + while(n--) { *hp++ = *ptr++; } + nhp[slot+1] = res; + res = make_hashmap(nhp); + break; + case HAMT_SUBTAG_HEAD_ARRAY: + slot = (Uint) ESTACK_POP(*sp); + nhp = hp; + n = HAMT_HEAD_ARRAY_SZ - 2; + *hp++ = MAP_HEADER_HAMT_HEAD_ARRAY; ptr++; + *hp++ = (*ptr++) + *update_size; + while(n--) { *hp++ = *ptr++; } + nhp[slot+2] = res; + res = make_hashmap(nhp); + break; + case HAMT_SUBTAG_NODE_BITMAP: + slot = (Uint) ESTACK_POP(*sp); + bp = (Uint32) ESTACK_POP(*sp); + n = (Uint32) ESTACK_POP(*sp); + hval = MAP_HEADER_VAL(hdr); + nhp = hp; + *hp++ = MAP_HEADER_HAMT_NODE_BITMAP(hval | bp); ptr++; + + n -= slot; + while(slot--) { *hp++ = *ptr++; } + *hp++ = res; + if (hval & bp) { ptr++; n--; } + while(n--) { *hp++ = *ptr++; } + + if ((hval | bp) == 0xffff) { + *nhp = make_arityval(16); + } + res = make_hashmap(nhp); + break; + case HAMT_SUBTAG_HEAD_BITMAP: + slot = (Uint) ESTACK_POP(*sp); + bp = (Uint32) ESTACK_POP(*sp); + n = (Uint32) ESTACK_POP(*sp); + hval = MAP_HEADER_VAL(hdr); + nhp = hp; + *hp++ = MAP_HEADER_HAMT_HEAD_BITMAP(hval | bp); ptr++; + *hp++ = (*ptr++) + *update_size; + + n -= slot; + while(slot--) { *hp++ = *ptr++; } + *hp++ = res; + if (hval & bp) { ptr++; n--; } + while(n--) { *hp++ = *ptr++; } + + if ((hval | bp) == 0xffff) { + *nhp = MAP_HEADER_HAMT_HEAD_ARRAY; + } + res = make_hashmap(nhp); + break; + default: + erl_exit(1, "bad header tag %x\r\n", hdr & _HEADER_MAP_SUBTAG_MASK); + break; + } + break; + default: + erl_exit(1, "bad primary tag %x\r\n", primary_tag(node)); + break; + } + + } while(!ESTACK_ISEMPTY(*sp)); + + UnUseTmpHeapNoproc(1); + return res; +} + +static Eterm hashmap_keys(Process* p, Eterm node) { + DECLARE_WSTACK(stack); + hashmap_head_t* root; + Eterm *hp, *kv; + Eterm res = NIL; + + root = (hashmap_head_t*) boxed_val(node); + hp = HAlloc(p, root->size * 2); + hashmap_iterator_init(&stack, node, 0); + while ((kv=hashmap_iterator_next(&stack)) != NULL) { + res = CONS(hp, CAR(kv), res); + hp += 2; + } + DESTROY_WSTACK(stack); + return res; +} + +static Eterm hashmap_values(Process* p, Eterm node) { + DECLARE_WSTACK(stack); + hashmap_head_t* root; + Eterm *hp, *kv; + Eterm res = NIL; + + root = (hashmap_head_t*) boxed_val(node); + hp = HAlloc(p, root->size * 2); + hashmap_iterator_init(&stack, node, 0); + while ((kv=hashmap_iterator_next(&stack)) != NULL) { + res = CONS(hp, CDR(kv), res); + hp += 2; + } + DESTROY_WSTACK(stack); + return res; +} + +static Eterm hashmap_delete(Process *p, Uint32 hx, Eterm key, Eterm map) { + Eterm *hp = NULL, *nhp = NULL, *hp_end = NULL; + Eterm th[2]; + Eterm *ptr; + Eterm hdr, res = map, node = map; + Uint32 ix, bp, hval; + Uint slot, lvl = 0; + Uint size = 0, n = 0; + DECLARE_ESTACK(stack); + + for (;;) { + switch(primary_tag(node)) { + case TAG_PRIMARY_LIST: + if (EQ(CAR(list_val(node)), key)) { + goto unroll; + } + goto not_found; + case TAG_PRIMARY_BOXED: + ptr = boxed_val(node); + hdr = *ptr; + ASSERT(is_header(hdr)); + + switch(hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_NODE_ARRAY: + ix = hashmap_index(hx); + hx = hashmap_shift_hash(th,hx,lvl,key); + size += HAMT_NODE_ARRAY_SZ; + ESTACK_PUSH2(stack, ix, node); + node = ptr[ix+1]; + break; + case HAMT_SUBTAG_HEAD_ARRAY: + ix = hashmap_index(hx); + hx = hashmap_shift_hash(th,hx,lvl,key); + size += HAMT_HEAD_ARRAY_SZ; + ESTACK_PUSH2(stack, ix, node); + node = ptr[ix+2]; + break; + case HAMT_SUBTAG_NODE_BITMAP: + hval = MAP_HEADER_VAL(hdr); + ix = hashmap_index(hx); + bp = 1 << ix; + slot = hashmap_bitcount(hval & (bp - 1)); + n = hashmap_bitcount(hval); + + ESTACK_PUSH(stack, n); + ESTACK_PUSH3(stack, bp, slot, node); + + /* occupied */ + if (bp & hval) { + hx = hashmap_shift_hash(th,hx,lvl,key); + node = ptr[slot+1]; + ASSERT(HAMT_NODE_BITMAP_SZ(n) <= 17); + size += HAMT_NODE_BITMAP_SZ(n); + break; + } + /* not occupied */ + goto not_found; + case HAMT_SUBTAG_HEAD_BITMAP: + hval = MAP_HEADER_VAL(hdr); + ix = hashmap_index(hx); + bp = 1 << ix; + slot = hashmap_bitcount(hval & (bp - 1)); + n = hashmap_bitcount(hval); + + ESTACK_PUSH(stack, n); + ESTACK_PUSH3(stack, bp, slot, node); + + /* occupied */ + if (bp & hval) { + hx = hashmap_shift_hash(th,hx,lvl,key); + node = ptr[slot+2]; + ASSERT(HAMT_HEAD_BITMAP_SZ(n) <= 18); + size += HAMT_HEAD_BITMAP_SZ(n); + break; + } + /* not occupied */ + goto not_found; + default: + erl_exit(1, "bad header tag %ld\r\n", hdr & _HEADER_MAP_SUBTAG_MASK); + break; + } + break; + default: + erl_exit(1, "bad primary tag %p\r\n", node); + break; + } + } + +unroll: + /* the size is bounded and atleast one less than the previous size */ + size -= 1; + n = hashmap_size(map) - 1; + + if (n <= MAP_SMALL_MAP_LIMIT) { + DECLARE_WSTACK(wstack); + Eterm *kv, *ks, *vs; + flatmap_t *mp; + Eterm keys; + + DESTROY_ESTACK(stack); + + /* build flat structure */ + hp = HAlloc(p, 3 + 1 + (2 * n)); + keys = make_tuple(hp); + *hp++ = make_arityval(n); + ks = hp; + hp += n; + mp = (flatmap_t*)hp; + hp += MAP_HEADER_SIZE; + vs = hp; + + mp->thing_word = MAP_HEADER; + mp->size = n; + mp->keys = keys; + + hashmap_iterator_init(&wstack, map, 0); + + while ((kv=hashmap_iterator_next(&wstack)) != NULL) { + if (EQ(CAR(kv),key)) + continue; + *ks++ = CAR(kv); + *vs++ = CDR(kv); + } + + /* it cannot have multiple keys */ + erts_validate_and_sort_flatmap(mp); + + DESTROY_WSTACK(wstack); + return make_flatmap(mp); + } + + hp = HAlloc(p, size); + hp_end = hp + size; + res = THE_NON_VALUE; + + do { + node = ESTACK_POP(stack); + + /* all nodes are things */ + ptr = boxed_val(node); + hdr = *ptr; + ASSERT(is_header(hdr)); + + switch(hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_NODE_ARRAY: + ix = (Uint) ESTACK_POP(stack); + nhp = hp; + if (res == THE_NON_VALUE) { + *hp++ = MAP_HEADER_HAMT_NODE_BITMAP(0xffff ^ (1 << ix)); ptr++; + n = 16; + n -= ix; + while(ix--) { *hp++ = *ptr++; } + ptr++; n--; + while(n--) { *hp++ = *ptr++; } + res = make_hashmap(nhp); + } else { + n = HAMT_NODE_ARRAY_SZ; + while(n--) { *hp++ = *ptr++; } + nhp[ix+1] = res; + res = make_hashmap(nhp); + } + break; + case HAMT_SUBTAG_HEAD_ARRAY: + ix = (Uint) ESTACK_POP(stack); + nhp = hp; + if (res == THE_NON_VALUE) { + n = 16; + n -= ix; + *hp++ = MAP_HEADER_HAMT_HEAD_BITMAP(0xffff ^ (1 << ix)); ptr++; + *hp++ = (*ptr++) - 1; + while(ix--) { *hp++ = *ptr++; } + ptr++; n--; + while(n--) { *hp++ = *ptr++; } + res = make_hashmap(nhp); + } else { + n = 16; + *hp++ = MAP_HEADER_HAMT_HEAD_ARRAY; ptr++; + *hp++ = (*ptr++) - 1; + while(n--) { *hp++ = *ptr++; } + nhp[ix+2] = res; + res = make_hashmap(nhp); + } + break; + case HAMT_SUBTAG_NODE_BITMAP: + slot = (Uint) ESTACK_POP(stack); + bp = (Uint32) ESTACK_POP(stack); + n = (Uint32) ESTACK_POP(stack); + nhp = hp; + + /* bitmap change matrix + * res | none leaf bitmap + * ---------------------------- + * n=1 | remove remove keep + * n=2 | other keep keep + * n>2 | shrink keep keep + * + * other: (remember, n is 2) + * shrink if the other bitmap value is a bitmap node + * remove if the other bitmap value is a leaf + * + * remove: + * this bitmap node is removed, res is moved up in tree (could be none) + * this is a special case of shrink + * + * keep: + * the current path index is still used down in the tree, need to keep it + * copy as usual with the updated res + * + * shrink: + * the current path index is no longer used down in the tree, remove it (shrink) + */ + if (res == THE_NON_VALUE) { + if (n == 1) { + break; + } else if (n == 2) { + if (slot == 0) { + ix = 2; /* off by one 'cause hdr */ + } else { + ix = 1; /* off by one 'cause hdr */ + } + if (primary_tag(ptr[ix]) == TAG_PRIMARY_LIST) { + res = ptr[ix]; + } else { + hval = MAP_HEADER_VAL(hdr); + *hp++ = MAP_HEADER_HAMT_NODE_BITMAP(hval ^ bp); + *hp++ = ptr[ix]; + res = make_hashmap(nhp); + } + } else { + /* n > 2 */ + hval = MAP_HEADER_VAL(hdr); + *hp++ = MAP_HEADER_HAMT_NODE_BITMAP(hval ^ bp); ptr++; + n -= slot; + while(slot--) { *hp++ = *ptr++; } + ptr++; n--; + while(n--) { *hp++ = *ptr++; } + res = make_hashmap(nhp); + } + } else if (primary_tag(res) == TAG_PRIMARY_LIST && n == 1) { + break; + } else { + /* res is bitmap or leaf && n > 1, keep */ + n -= slot; + *hp++ = *ptr++; + while(slot--) { *hp++ = *ptr++; } + *hp++ = res; + ptr++; n--; + while(n--) { *hp++ = *ptr++; } + res = make_hashmap(nhp); + } + break; + case HAMT_SUBTAG_HEAD_BITMAP: + slot = (Uint) ESTACK_POP(stack); + bp = (Uint32) ESTACK_POP(stack); + n = (Uint32) ESTACK_POP(stack); + nhp = hp; + + if (res != THE_NON_VALUE) { + *hp++ = *ptr++; + *hp++ = (*ptr++) - 1; + n -= slot; + while(slot--) { *hp++ = *ptr++; } + *hp++ = res; + ptr++; n--; + while(n--) { *hp++ = *ptr++; } + } else { + hval = MAP_HEADER_VAL(hdr); + *hp++ = MAP_HEADER_HAMT_HEAD_BITMAP(hval ^ bp); ptr++; + *hp++ = (*ptr++) - 1; + n -= slot; + while(slot--) { *hp++ = *ptr++; } + ptr++; n--; + while(n--) { *hp++ = *ptr++; } + } + res = make_hashmap(nhp); + break; + default: + erl_exit(1, "bad header tag %x\r\n", hdr & _HEADER_MAP_SUBTAG_MASK); + break; + } + } while(!ESTACK_ISEMPTY(stack)); + HRelease(p, hp_end, hp); +not_found: + DESTROY_ESTACK(stack); + ERTS_VERIFY_UNUSED_TEMP_ALLOC(p); + ERTS_HOLE_CHECK(p); + return res; +} + + +int erts_validate_and_sort_flatmap(flatmap_t* mp) +{ + Eterm *ks = flatmap_get_keys(mp); + Eterm *vs = flatmap_get_values(mp); + Uint sz = flatmap_get_size(mp); Uint ix,jx; Eterm tmp; int c; @@ -822,6 +2528,50 @@ int erts_validate_and_sort_map(map_t* mp) return 1; } +/* Really rough estimate of sqrt(x) + * Guaranteed not to be less than sqrt(x) + */ +static int int_sqrt_ceiling(Uint x) +{ + int n; + + if (x <= 2) + return x; + + n = erts_fit_in_bits_uint(x-1); + if (n & 1) { + /* Calc: sqrt(2^n) = 2^(n/2) * sqrt(2) ~= 2^(n/2) * 3 / 2 */ + return (1 << (n/2 - 1)) * 3; + } + else { + /* Calc: sqrt(2^n) = 2^(n/2) */ + return 1 << (n / 2); + } +} + +Uint hashmap_over_estimated_heap_size(Uint n) +{ + /* n is nr of key-value pairs. + Average nr of nodes is about n/3. + Standard deviation is about sqrt(n)/3. + Assuming normal probability distribution, + we overestimate nr of nodes by 14 std.devs, which gives a probability + for overrun of 1.0e-49 (same magnitude as a git SHA1 collision). + */ + Uint nodes = (n + int_sqrt_ceiling(n)*14) / 3; + return (n*2 + /* leaf cons cells */ + n + /* leaf list terms */ + nodes*2); /* headers + parent refs */ +} + + +BIF_RETTYPE erts_debug_map_info_1(BIF_ALIST_1) { + if (is_hashmap(BIF_ARG_1)) { + BIF_RET(hashmap_info(BIF_P,BIF_ARG_1)); + } + BIF_ERROR(BIF_P, BADARG); +} + /* * erts_internal:map_to_tuple_keys/1 * @@ -829,9 +2579,233 @@ int erts_validate_and_sort_map(map_t* mp) */ BIF_RETTYPE erts_internal_map_to_tuple_keys_1(BIF_ALIST_1) { - if (is_map(BIF_ARG_1)) { - map_t *mp = (map_t*)map_val(BIF_ARG_1); + if (is_flatmap(BIF_ARG_1)) { + flatmap_t *mp = (flatmap_t*)flatmap_val(BIF_ARG_1); BIF_RET(mp->keys); } BIF_ERROR(BIF_P, BADARG); } + +/* + * erts_internal:map_type/1 + * + * Used in erts_debug:size/1 + */ + +BIF_RETTYPE erts_internal_map_type_1(BIF_ALIST_1) { + DECL_AM(hashmap); + DECL_AM(hashmap_node); + DECL_AM(flatmap); + if (is_flatmap(BIF_ARG_1)) { + BIF_RET(AM_flatmap); + } else if (is_hashmap(BIF_ARG_1)) { + Eterm hdr = *(boxed_val(BIF_ARG_1)); + ASSERT(is_header(hdr)); + switch (hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: + case HAMT_SUBTAG_HEAD_BITMAP: + BIF_RET(AM_hashmap); + case HAMT_SUBTAG_NODE_ARRAY: + case HAMT_SUBTAG_NODE_BITMAP: + BIF_RET(AM_hashmap_node); + default: + erl_exit(1, "bad header"); + } + } + BIF_ERROR(BIF_P, BADARG); +} + +/* + * erts_internal:map_hashmap_children/1 + * + * Used in erts_debug:size/1 + */ + +BIF_RETTYPE erts_internal_map_hashmap_children_1(BIF_ALIST_1) { + if (is_hashmap(BIF_ARG_1)) { + Eterm node = BIF_ARG_1; + Eterm *ptr, hdr, *hp, res = NIL; + Uint sz = 0; + ptr = boxed_val(node); + hdr = *ptr; + + ASSERT(is_header(hdr)); + + switch(hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_NODE_ARRAY: + sz = 16; + ptr += 1; + break; + case HAMT_SUBTAG_NODE_BITMAP: + sz = hashmap_bitcount(MAP_HEADER_VAL(hdr)); + ptr += 1; + break; + case HAMT_SUBTAG_HEAD_BITMAP: + sz = hashmap_bitcount(MAP_HEADER_VAL(hdr)); + ptr += 2; + break; + case HAMT_SUBTAG_HEAD_ARRAY: + sz = 16; + ptr += 2; + break; + default: + erl_exit(1, "bad header\r\n"); + break; + } + ASSERT(sz < 17); + hp = HAlloc(BIF_P, 2*sz); + while(sz--) { res = CONS(hp, *ptr++, res); hp += 2; } + BIF_RET(res); + } + BIF_ERROR(BIF_P, BADARG); +} + + +static Eterm hashmap_info(Process *p, Eterm node) { + Eterm *hp; + Eterm res = NIL, info = NIL; + Eterm *ptr, tup, hdr; + Uint sz; + DECL_AM(depth); + DECL_AM(leafs); + DECL_AM(bitmaps); + DECL_AM(arrays); + Uint nleaf=0, nbitmap=0, narray=0; + Uint bitmap_usage[16], leaf_usage[16]; + Uint lvl = 0, clvl; + DECLARE_ESTACK(stack); + + for (sz = 0; sz < 16; sz++) { + bitmap_usage[sz] = 0; + leaf_usage[sz] = 0; + } + + ptr = boxed_val(node); + ESTACK_PUSH(stack, 0); + ESTACK_PUSH(stack, node); + do { + node = ESTACK_POP(stack); + clvl = ESTACK_POP(stack); + if (lvl < clvl) + lvl = clvl; + switch(primary_tag(node)) { + case TAG_PRIMARY_LIST: + nleaf++; + leaf_usage[clvl] += 1; + break; + case TAG_PRIMARY_BOXED: + ptr = boxed_val(node); + hdr = *ptr; + ASSERT(is_header(hdr)); + switch(hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_NODE_ARRAY: + narray++; + sz = 16; + while(sz--) { + ESTACK_PUSH(stack, clvl + 1); + ESTACK_PUSH(stack, ptr[sz+1]); + } + break; + case HAMT_SUBTAG_NODE_BITMAP: + nbitmap++; + sz = hashmap_bitcount(MAP_HEADER_VAL(hdr)); + ASSERT(sz < 17); + bitmap_usage[sz-1] += 1; + while(sz--) { + ESTACK_PUSH(stack, clvl + 1); + ESTACK_PUSH(stack, ptr[sz+1]); + } + break; + case HAMT_SUBTAG_HEAD_BITMAP: + nbitmap++; + sz = hashmap_bitcount(MAP_HEADER_VAL(hdr)); + bitmap_usage[sz-1] += 1; + while(sz--) { + ESTACK_PUSH(stack, clvl + 1); + ESTACK_PUSH(stack, ptr[sz+2]); + } + break; + case HAMT_SUBTAG_HEAD_ARRAY: + narray++; + sz = 16; + while(sz--) { + ESTACK_PUSH(stack, clvl + 1); + ESTACK_PUSH(stack, ptr[sz+2]); + } + break; + default: + erl_exit(1, "bad header\r\n"); + break; + } + } + } while(!ESTACK_ISEMPTY(stack)); + + + /* size */ + sz = 0; + hashmap_bld_tuple_uint(NULL,&sz,16,leaf_usage); + hashmap_bld_tuple_uint(NULL,&sz,16,bitmap_usage); + + /* alloc */ + hp = HAlloc(p, 2+3 + 3*(2+4) + sz); + + info = hashmap_bld_tuple_uint(&hp,NULL,16,leaf_usage); + tup = TUPLE3(hp, AM_leafs, make_small(nleaf),info); hp += 4; + res = CONS(hp, tup, res); hp += 2; + + info = hashmap_bld_tuple_uint(&hp,NULL,16,bitmap_usage); + tup = TUPLE3(hp, AM_bitmaps, make_small(nbitmap), info); hp += 4; + res = CONS(hp, tup, res); hp += 2; + + tup = TUPLE3(hp, AM_arrays, make_small(narray),NIL); hp += 4; + res = CONS(hp, tup, res); hp += 2; + + tup = TUPLE2(hp, AM_depth, make_small(lvl)); hp += 3; + res = CONS(hp, tup, res); hp += 2; + + DESTROY_ESTACK(stack); + ERTS_HOLE_CHECK(p); + return res; +} + +static Eterm hashmap_bld_tuple_uint(Uint **hpp, Uint *szp, Uint n, Uint nums[]) { + Eterm res = THE_NON_VALUE; + Eterm *ts = (Eterm *)erts_alloc(ERTS_ALC_T_TMP, n * sizeof(Eterm)); + Uint i; + + for (i = 0; i < n; i++) { + ts[i] = erts_bld_uint(hpp, szp, nums[i]); + } + res = erts_bld_tuplev(hpp, szp, n, ts); + erts_free(ERTS_ALC_T_TMP, (void *) ts); + return res; +} + + +/* implementation of builtin emulations */ + +#if !ERTS_AT_LEAST_GCC_VSN__(3, 4, 0) +/* Count leading zeros emulation */ +Uint32 hashmap_clz(Uint32 x) { + Uint32 y; + int n = 32; + y = x >>16; if (y != 0) {n = n -16; x = y;} + y = x >> 8; if (y != 0) {n = n - 8; x = y;} + y = x >> 4; if (y != 0) {n = n - 4; x = y;} + y = x >> 2; if (y != 0) {n = n - 2; x = y;} + y = x >> 1; if (y != 0) return n - 2; + return n - x; +} + +const Uint32 SK5 = 0x55555555, SK3 = 0x33333333; +const Uint32 SKF0 = 0xF0F0F0F, SKFF = 0xFF00FF; + +/* CTPOP emulation */ +Uint32 hashmap_bitcount(Uint32 x) { + x -= ((x >> 1 ) & SK5); + x = (x & SK3 ) + ((x >> 2 ) & SK3 ); + x = (x & SKF0) + ((x >> 4 ) & SKF0); + x += x >> 8; + return (x + (x >> 16)) & 0x3F; +} +#endif diff --git a/erts/emulator/beam/erl_map.h b/erts/emulator/beam/erl_map.h index cfacb2ec28..1333a734a8 100644 --- a/erts/emulator/beam/erl_map.h +++ b/erts/emulator/beam/erl_map.h @@ -22,13 +22,23 @@ #define __ERL_MAP_H__ #include "sys.h" + +/* instrinsic wrappers */ +#if ERTS_AT_LEAST_GCC_VSN__(3, 4, 0) +#define hashmap_clz(x) ((Uint32) __builtin_clz((unsigned int)(x))) +#define hashmap_bitcount(x) ((Uint32) __builtin_popcount((unsigned int) (x))) +#else +Uint32 hashmap_clz(Uint32 x); +Uint32 hashmap_bitcount(Uint32 x); +#endif + /* MAP */ -typedef struct map_s { +typedef struct flatmap_s { Eterm thing_word; Uint size; Eterm keys; /* tuple */ -} map_t; +} flatmap_t; /* map node * * ----------- @@ -42,31 +52,152 @@ typedef struct map_s { * ----------- */ +/* the head-node is a bitmap or array with an untagged size */ + + +#define hashmap_size(x) (((hashmap_head_t*) hashmap_val(x))->size) +#define hashmap_size_rel(RTERM, BASE) hashmap_size(rterm2wterm(RTERM, BASE)) +#define hashmap_make_hash(Key) make_internal_hash(Key) + +#define hashmap_restore_hash(Heap,Lvl,Key) \ + (((Lvl) < 8) ? hashmap_make_hash(Key) >> (4*(Lvl)) : hashmap_make_hash(CONS(Heap, make_small((Lvl)>>3), (Key))) >> (4*((Lvl) & 7))) +#define hashmap_shift_hash(Heap,Hx,Lvl,Key) \ + (((++(Lvl)) & 7) ? (Hx) >> 4 : hashmap_make_hash(CONS(Heap, make_small((Lvl)>>3), Key))) /* 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 make_flatmap(x) make_boxed((Eterm*)(x)) +#define make_flatmap_rel(x, BASE) make_boxed_rel((Eterm*)(x),(BASE)) +#define is_flatmap(x) (is_boxed((x)) && is_flatmap_header(*boxed_val((x)))) +#define is_flatmap_rel(RTERM,BASE) is_flatmap(rterm2wterm(RTERM,BASE)) +#define is_not_flatmap(x) (!is_flatmap((x))) +#define is_flatmap_header(x) (((x) & (_TAG_HEADER_MASK)) == _TAG_HEADER_MAP) +#define header_is_flatmap(x) ((((x) & (_HEADER_SUBTAG_MASK)) == MAP_SUBTAG)) +#define flatmap_val(x) (_unchecked_boxed_val((x))) +#define flatmap_val_rel(RTERM, BASE) flatmap_val(rterm2wterm(RTERM, BASE)) +#define flatmap_get_values(x) (((Eterm *)(x)) + 3) +#define flatmap_get_keys(x) (((Eterm *)tuple_val(((flatmap_t *)(x))->keys)) + 1) +#define flatmap_get_size(x) (((flatmap_t*)(x))->size) + +#ifdef DEBUG +#define MAP_SMALL_MAP_LIMIT (3) +#else +#define MAP_SMALL_MAP_LIMIT (32) +#endif #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); +#define MAP_HEADER_SIZE (sizeof(flatmap_t) / sizeof(Eterm)) + +struct ErtsWStack_; +struct ErtsEStack_; + +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_remove(Process *p, Eterm key, Eterm map, Eterm *res); + +Eterm erts_hashmap_insert(Process *p, Uint32 hx, Eterm key, Eterm value, + Eterm node, int is_update); +int erts_hashmap_insert_down(Uint32 hx, Eterm key, Eterm node, Uint *sz, + Uint *upsz, struct ErtsEStack_ *sp, int is_update); +Eterm erts_hashmap_insert_up(Eterm *hp, Eterm key, Eterm value, + Uint *upsz, struct ErtsEStack_ *sp); + +int erts_validate_and_sort_flatmap(flatmap_t* map); +Uint hashmap_over_estimated_heap_size(Uint n); +void hashmap_iterator_init(struct ErtsWStack_* s, Eterm node, int reverse); +Eterm* hashmap_iterator_next(struct ErtsWStack_* s); +Eterm* hashmap_iterator_prev(struct ErtsWStack_* s); +int hashmap_key_hash_cmp(Eterm* ap, Eterm* bp); +Eterm erts_hashmap_from_array(ErtsHeapFactory*, Eterm *leafs, Uint n, int reject_dupkeys); + +#define erts_hashmap_from_ks_and_vs(P, KS, VS, N) \ + erts_hashmap_from_ks_and_vs_extra((P), (KS), (VS), (N), THE_NON_VALUE, THE_NON_VALUE); + +Eterm erts_hashmap_from_ks_and_vs_extra(Process *p, Eterm *ks, Eterm *vs, Uint n, + Eterm k, Eterm v); + +const Eterm * +#if HALFWORD_HEAP +erts_maps_get_rel(Eterm key, Eterm map, Eterm *map_base); +# define erts_maps_get(A, B) erts_maps_get_rel(A, B, NULL) +#else +erts_maps_get(Eterm key, Eterm map); +# define erts_maps_get_rel(A, B, B_BASE) erts_maps_get(A, B) +#endif + +const Eterm * +#if HALFWORD_HEAP +erts_hashmap_get_rel(Uint32 hx, Eterm key, Eterm node, Eterm *map_base); +# define erts_hashmap_get(Hx, K, M) erts_hashmap_get_rel(Hx, K, M, NULL) +#else +erts_hashmap_get(Uint32 hx, Eterm key, Eterm map); +# define erts_hashmap_get_rel(Hx, K, M, M_BASE) erts_hashmap_get(Hx, K, M) #endif +/* hamt nodes v2.0 + * + * node :: leaf | array | bitmap + * head + */ +typedef struct hashmap_head_s { + Eterm thing_word; + Uint size; + Eterm items[1]; +} hashmap_head_t; + +/* thing_word tagscheme + * Need two bits for map subtags + * + * Original HEADER representation: + * + * aaaaaaaaaaaaaaaa aaaaaaaaaatttt00 arity:26, tag:4 + * + * For maps we have: + * + * vvvvvvvvvvvvvvvv aaaaaaaamm111100 val:16, arity:8, mtype:2 + * + * unsure about trailing zeros + * + * map-tag: + * 00 - flat map tag (non-hamt) -> val:16 = #items + * 01 - map-node bitmap tag -> val:16 = bitmap + * 10 - map-head (array-node) -> val:16 = 0xffff + * 11 - map-head (bitmap-node) -> val:16 = bitmap + */ + +/* erl_map.h stuff */ + +#define is_hashmap_header_head(x) ((MAP_HEADER_TYPE(x) & (0x2))) + +#define MAKE_MAP_HEADER(Type,Arity,Val) \ + (_make_header(((((Uint16)(Val)) << MAP_HEADER_ARITY_SZ) | (Arity)) << MAP_HEADER_TAG_SZ | (Type) , _TAG_HEADER_HASHMAP)) + +#define MAP_HEADER_HAMT_HEAD_ARRAY \ + MAKE_MAP_HEADER(MAP_HEADER_TAG_HAMT_HEAD_ARRAY,0x1,0xffff) + +#define MAP_HEADER_HAMT_HEAD_BITMAP(Bmp) \ + MAKE_MAP_HEADER(MAP_HEADER_TAG_HAMT_HEAD_BITMAP,0x1,Bmp) + +#define MAP_HEADER_HAMT_NODE_ARRAY \ + make_arityval(16) + +#define MAP_HEADER_HAMT_NODE_BITMAP(Bmp) \ + MAKE_MAP_HEADER(MAP_HEADER_TAG_HAMT_NODE_BITMAP,0x0,Bmp) + +#define HAMT_HEAD_EMPTY_SZ (2) +#define HAMT_NODE_ARRAY_SZ (17) +#define HAMT_HEAD_ARRAY_SZ (18) +#define HAMT_NODE_BITMAP_SZ(n) (1 + n) +#define HAMT_HEAD_BITMAP_SZ(n) (2 + n) + +#define _HEADER_MAP_SUBTAG_MASK (0xfc) /* 2 bits maps tag + 4 bits subtag + 2 ignore bits */ +/* SUBTAG_NODE_ARRAY is in fact a tuple with 16 elements */ +#define HAMT_SUBTAG_NODE_ARRAY (((16 << _HEADER_ARITY_OFFS) | ARITYVAL_SUBTAG) & _HEADER_MAP_SUBTAG_MASK) +#define HAMT_SUBTAG_NODE_BITMAP ((MAP_HEADER_TAG_HAMT_NODE_BITMAP << _HEADER_ARITY_OFFS) | HASHMAP_SUBTAG) +#define HAMT_SUBTAG_HEAD_ARRAY ((MAP_HEADER_TAG_HAMT_HEAD_ARRAY << _HEADER_ARITY_OFFS) | HASHMAP_SUBTAG) +#define HAMT_SUBTAG_HEAD_BITMAP ((MAP_HEADER_TAG_HAMT_HEAD_BITMAP << _HEADER_ARITY_OFFS) | HASHMAP_SUBTAG) + +#define hashmap_index(hash) (((Uint32)hash) & 0xf) + + +#endif diff --git a/erts/emulator/beam/erl_math.c b/erts/emulator/beam/erl_math.c index 16d4fdc09c..9b864628db 100644 --- a/erts/emulator/beam/erl_math.c +++ b/erts/emulator/beam/erl_math.c @@ -207,6 +207,24 @@ BIF_RETTYPE math_log_1(BIF_ALIST_1) return math_call_1(BIF_P, log, BIF_ARG_1); } +#ifdef HAVE_LOG2 +static double +log2_wrapper(double x) +{ + return log2(x); +} +#else +static double +log2_wrapper(double x) +{ + return log(x) / 0.6931471805599453; /* log(2.0); */ +} +#endif + +BIF_RETTYPE math_log2_1(BIF_ALIST_1) +{ + return math_call_1(BIF_P, log2_wrapper, BIF_ARG_1); +} BIF_RETTYPE math_log10_1(BIF_ALIST_1) { diff --git a/erts/emulator/beam/erl_message.c b/erts/emulator/beam/erl_message.c index 8870fac7d9..43a03c793e 100644 --- a/erts/emulator/beam/erl_message.c +++ b/erts/emulator/beam/erl_message.c @@ -1146,3 +1146,15 @@ erts_deliver_exit_message(Eterm from, Process *to, ErtsProcLocks *to_locksp, } } +Eterm* erts_produce_heap(ErtsHeapFactory* factory, Uint need, Uint xtra) +{ + Eterm* res; + if (factory->p) { + res = HAllocX(factory->p, need, xtra); + } else { + res = factory->hp; + factory->hp += need; + } + return res; +} + diff --git a/erts/emulator/beam/erl_message.h b/erts/emulator/beam/erl_message.h index 0f3bb8d281..6b8c3cebc7 100644 --- a/erts/emulator/beam/erl_message.h +++ b/erts/emulator/beam/erl_message.h @@ -68,6 +68,21 @@ struct erl_heap_fragment { Eterm mem[1]; /* Data */ }; +typedef struct { + Process* p; + Eterm* hp; +} ErtsHeapFactory; + +Eterm* erts_produce_heap(ErtsHeapFactory*, Uint need, Uint xtra); +#ifdef CHECK_FOR_HOLES +# define ERTS_FACTORY_HOLE_CHECK(f) do { \ + if ((f)->p) erts_check_for_holes((f)->p); \ + } while (0) +#else +# define ERTS_FACTORY_HOLE_CHECK(p) +#endif + + typedef struct erl_mesg { struct erl_mesg* next; /* Next message */ union { diff --git a/erts/emulator/beam/erl_mtrace.c b/erts/emulator/beam/erl_mtrace.c index c8bb126687..fa1bde1c87 100644 --- a/erts/emulator/beam/erl_mtrace.c +++ b/erts/emulator/beam/erl_mtrace.c @@ -627,7 +627,7 @@ erts_mtrace_install_wrapper_functions(void) if (erts_mtrace_enabled) { int i; /* Install trace functions */ - ASSERT(sizeof(erts_allctrs) == sizeof(real_allctrs)); + ERTS_CT_ASSERT(sizeof(erts_allctrs) == sizeof(real_allctrs)); sys_memcpy((void *) real_allctrs, (void *) erts_allctrs, diff --git a/erts/emulator/beam/erl_nif.c b/erts/emulator/beam/erl_nif.c index adc3520ebb..e28365cb1b 100644 --- a/erts/emulator/beam/erl_nif.c +++ b/erts/emulator/beam/erl_nif.c @@ -551,9 +551,7 @@ int enif_alloc_binary(size_t size, ErlNifBinary* bin) if (refbin == NULL) { return 0; /* The NIF must take action */ } - refbin->flags = BIN_FLAG_DRV; /* BUGBUG: Flag? */ erts_refc_init(&refbin->refc, 1); - refbin->orig_size = (SWord) size; bin->size = size; bin->data = (unsigned char*) refbin->orig_bytes; @@ -573,7 +571,6 @@ int enif_realloc_binary(ErlNifBinary* bin, size_t size) if (!newbin) { return 0; } - newbin->orig_size = size; bin->ref_bin = newbin; bin->data = (unsigned char*) newbin->orig_bytes; bin->size = size; @@ -1913,12 +1910,16 @@ int enif_is_map(ErlNifEnv* env, ERL_NIF_TERM 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); + if (is_flatmap(term)) { + flatmap_t *mp; + mp = (flatmap_t*)flatmap_val(term); + *size = flatmap_get_size(mp); return 1; } + else if (is_hashmap(term)) { + *size = hashmap_size(term); + return 1; + } return 0; } @@ -1926,16 +1927,16 @@ ERL_NIF_TERM enif_make_new_map(ErlNifEnv* env) { Eterm* hp = alloc_heap(env,MAP_HEADER_SIZE+1); Eterm tup; - map_t *mp; + flatmap_t *mp; tup = make_tuple(hp); *hp++ = make_arityval(0); - mp = (map_t*)hp; + mp = (flatmap_t*)hp; mp->thing_word = MAP_HEADER; mp->size = 0; mp->keys = tup; - return make_map(mp); + return make_flatmap(mp); } int enif_make_map_put(ErlNifEnv* env, @@ -1944,7 +1945,7 @@ int enif_make_map_put(ErlNifEnv* env, Eterm value, Eterm *map_out) { - if (is_not_map(map_in)) { + if (!is_map(map_in)) { return 0; } flush_env(env); @@ -1958,10 +1959,16 @@ int enif_get_map_value(ErlNifEnv* env, Eterm key, Eterm *value) { - if (is_not_map(map)) { + const Eterm *ret; + if (!is_map(map)) { return 0; } - return erts_maps_get(key, map, value); + ret = erts_maps_get(key, map); + if (ret) { + *value = *ret; + return 1; + } + return 0; } int enif_make_map_update(ErlNifEnv* env, @@ -1971,7 +1978,7 @@ int enif_make_map_update(ErlNifEnv* env, Eterm *map_out) { int res; - if (is_not_map(map_in)) { + if (!is_map(map_in)) { return 0; } @@ -1987,7 +1994,7 @@ int enif_make_map_remove(ErlNifEnv* env, Eterm *map_out) { int res; - if (is_not_map(map_in)) { + if (!is_map(map_in)) { return 0; } flush_env(env); @@ -2001,13 +2008,13 @@ int enif_map_iterator_create(ErlNifEnv *env, ErlNifMapIterator *iter, ErlNifMapIteratorEntry entry) { - if (is_map(map)) { - map_t *mp = (map_t*)map_val(map); + if (is_flatmap(map)) { + flatmap_t *mp = (flatmap_t*)flatmap_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; + case ERL_NIF_MAP_ITERATOR_TAIL: offset = flatmap_get_size(mp) - 1; break; default: goto error; } @@ -2016,14 +2023,37 @@ int enif_map_iterator_create(ErlNifEnv *env, */ 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->u.flat.ks = ((Eterm *)flatmap_get_keys(mp)) + offset; + iter->u.flat.vs = ((Eterm *)flatmap_get_values(mp)) + offset; + iter->size = flatmap_get_size(mp); iter->idx = offset + 1; return 1; } - + else if (is_hashmap(map)) { + iter->map = map; + iter->size = hashmap_size(map); + iter->u.hash.wstack = erts_alloc(ERTS_ALC_T_NIF, sizeof(ErtsDynamicWStack)); + WSTACK_INIT(iter->u.hash.wstack, ERTS_ALC_T_NIF); + + switch (entry) { + case ERL_NIF_MAP_ITERATOR_HEAD: + iter->idx = 1; + hashmap_iterator_init(&iter->u.hash.wstack->ws, map, 0); + iter->u.hash.kv = hashmap_iterator_next(&iter->u.hash.wstack->ws); + break; + case ERL_NIF_MAP_ITERATOR_TAIL: + iter->idx = hashmap_size(map); + hashmap_iterator_init(&iter->u.hash.wstack->ws, map, 1); + iter->u.hash.kv = hashmap_iterator_prev(&iter->u.hash.wstack->ws); + break; + default: + goto error; + } + ASSERT(!!iter->u.hash.kv == (iter->idx >= 1 && + iter->idx <= iter->size)); + return 1; + } error: #ifdef DEBUG iter->map = THE_NON_VALUE; @@ -2033,48 +2063,97 @@ error: void enif_map_iterator_destroy(ErlNifEnv *env, ErlNifMapIterator *iter) { - /* not used */ + if (is_hashmap(iter->map)) { + WSTACK_DESTROY(iter->u.hash.wstack->ws); + erts_free(ERTS_ALC_T_NIF, iter->u.hash.wstack); + } + else + ASSERT(is_flatmap(iter->map)); + #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); + ASSERT(iter); + if (is_flatmap(iter->map)) { + ASSERT(iter->idx >= 0); + ASSERT(iter->idx <= flatmap_get_size(flatmap_val(iter->map)) + 1); + return (iter->size == 0 || iter->idx > iter->size); + } + else { + ASSERT(is_hashmap(iter->map)); + return iter->idx > iter->size; + } } 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); + ASSERT(iter); + if (is_flatmap(iter->map)) { + ASSERT(iter->idx >= 0); + ASSERT(iter->idx <= flatmap_get_size(flatmap_val(iter->map)) + 1); + return (iter->size == 0 || iter->idx == 0); + } + else { + ASSERT(is_hashmap(iter->map)); + return 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++; + ASSERT(iter); + if (is_flatmap(iter->map)) { + if (iter->idx <= iter->size) { + iter->idx++; + iter->u.flat.ks++; + iter->u.flat.vs++; + } + return (iter->idx <= iter->size); + } + else { + ASSERT(is_hashmap(iter->map)); + + if (iter->idx <= hashmap_size(iter->map)) { + if (iter->idx < 1) { + hashmap_iterator_init(&iter->u.hash.wstack->ws, iter->map, 0); + } + iter->u.hash.kv = hashmap_iterator_next(&iter->u.hash.wstack->ws); + iter->idx++; + ASSERT(!!iter->u.hash.kv == (iter->idx <= iter->size)); + } + return iter->idx <= iter->size; } - 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--; + ASSERT(iter); + if (is_flatmap(iter->map)) { + if (iter->idx > 0) { + iter->idx--; + iter->u.flat.ks--; + iter->u.flat.vs--; + } + return iter->idx > 0; + } + else { + ASSERT(is_hashmap(iter->map)); + + if (iter->idx > 0) { + if (iter->idx > iter->size) { + hashmap_iterator_init(&iter->u.hash.wstack->ws, iter->map, 1); + } + iter->u.hash.kv = hashmap_iterator_prev(&iter->u.hash.wstack->ws); + iter->idx--; + ASSERT(!!iter->u.hash.kv == (iter->idx > 0)); + } + return iter->idx > 0; } - return (iter->idx > 0); } int enif_map_iterator_get_pair(ErlNifEnv *env, @@ -2082,15 +2161,25 @@ int enif_map_iterator_get_pair(ErlNifEnv *env, 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; + ASSERT(iter); + if (is_flatmap(iter->map)) { + if (iter->idx > 0 && iter->idx <= iter->size) { + ASSERT(iter->u.flat.ks >= flatmap_get_keys(flatmap_val(iter->map)) && + iter->u.flat.ks < (flatmap_get_keys(flatmap_val(iter->map)) + flatmap_get_size(flatmap_val(iter->map)))); + ASSERT(iter->u.flat.vs >= flatmap_get_values(flatmap_val(iter->map)) && + iter->u.flat.vs < (flatmap_get_values(flatmap_val(iter->map)) + flatmap_get_size(flatmap_val(iter->map)))); + *key = *(iter->u.flat.ks); + *value = *(iter->u.flat.vs); + return 1; + } + } + else { + ASSERT(is_hashmap(iter->map)); + if (iter->idx > 0 && iter->idx <= iter->size) { + *key = CAR(iter->u.hash.kv); + *value = CDR(iter->u.hash.kv); + return 1; + } } return 0; } diff --git a/erts/emulator/beam/erl_nif.h b/erts/emulator/beam/erl_nif.h index 849024453c..9b2b90c82d 100644 --- a/erts/emulator/beam/erl_nif.h +++ b/erts/emulator/beam/erl_nif.h @@ -201,10 +201,18 @@ typedef enum typedef struct /* All fields all internal and may change */ { ERL_NIF_TERM map; - ERL_NIF_UINT t_limit; + ERL_NIF_UINT size; ERL_NIF_UINT idx; - ERL_NIF_TERM *ks; - ERL_NIF_TERM *vs; + union { + struct { + ERL_NIF_TERM *ks; + ERL_NIF_TERM *vs; + }flat; + struct { + struct ErtsDynamicWStack_* wstack; + ERL_NIF_TERM* kv; + }hash; + }u; void* __spare__[2]; /* for future additions to be ABI compatible (same struct size) */ } ErlNifMapIterator; diff --git a/erts/emulator/beam/erl_printf_term.c b/erts/emulator/beam/erl_printf_term.c index d18760dc43..ac5b139f8d 100644 --- a/erts/emulator/beam/erl_printf_term.c +++ b/erts/emulator/beam/erl_printf_term.c @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 2005-2013. All Rights Reserved. + * Copyright Ericsson AB 2005-2014. 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 @@ -25,6 +25,7 @@ #include "sys.h" #include "big.h" #include "erl_map.h" +#include "erl_binary.h" #define PRINT_CHAR(CNT, FN, ARG, C) \ do { \ @@ -138,6 +139,25 @@ is_printable_string(Eterm list, Eterm* base) return 0; } +static int is_printable_ascii(byte* bytep, Uint bytesize, Uint bitoffs) +{ + if (!bitoffs) { + while (bytesize--) { + if (*bytep < ' ' || *bytep >= 127) + return 0; + bytep++; + } + } else { + while (bytesize--) { + byte octet = (bytep[0] << bitoffs) | (bytep[1] >> (8-bitoffs)); + if (octet < ' ' || octet >= 127) + return 0; + bytep++; + } + } + return 1; +} + /* print a atom doing what quoting is necessary */ static int print_atom_name(fmtfn_t fn, void* arg, Eterm atom, long *dcount) { @@ -227,6 +247,17 @@ static int print_atom_name(fmtfn_t fn, void* arg, Eterm atom, long *dcount) #define PRT_PATCH_FUN_SIZE ((Eterm) 7) #define PRT_LAST_ARRAY_ELEMENT ((Eterm) 8) /* Note! Must be last... */ +#if 0 +static char *format_binary(Uint16 x, char *b) { + int z; + b[16] = '\0'; + for (z = 0; z < 16; z++) { + b[15-z] = ((x>>z) & 0x1) ? '1' : '0'; + } + return b; +} +#endif + static int print_term(fmtfn_t fn, void* arg, Eterm obj, long *dcount, Eterm* obj_base) /* ignored if !HALFWORD_HEAP */ @@ -283,13 +314,9 @@ print_term(fmtfn_t fn, void* arg, Eterm obj, long *dcount, tl = CDR(cons); if (is_not_nil(tl)) { if (is_list(tl)) { - WSTACK_PUSH(s, tl); - WSTACK_PUSH(s, PRT_ONE_CONS); - WSTACK_PUSH(s, PRT_COMMA); + WSTACK_PUSH3(s, tl, PRT_ONE_CONS, PRT_COMMA); } else { - WSTACK_PUSH(s, tl); - WSTACK_PUSH(s, PRT_TERM); - WSTACK_PUSH(s, PRT_BAR); + WSTACK_PUSH3(s, tl, PRT_TERM, PRT_BAR); } } } @@ -299,9 +326,7 @@ print_term(fmtfn_t fn, void* arg, Eterm obj, long *dcount, break; default: /* PRT_LAST_ARRAY_ELEMENT+1 and upwards */ obj = *popped.ptr; - WSTACK_PUSH(s, (UWord) (popped.ptr + 1)); - WSTACK_PUSH(s, val-1); - WSTACK_PUSH(s, PRT_COMMA); + WSTACK_PUSH3(s, (UWord) (popped.ptr + 1), val-1, PRT_COMMA); break; } break; @@ -431,8 +456,7 @@ print_term(fmtfn_t fn, void* arg, Eterm obj, long *dcount, WSTACK_PUSH(s,PRT_CLOSE_TUPLE); ++nobj; if (i > 0) { - WSTACK_PUSH(s, (UWord) nobj); - WSTACK_PUSH(s, PRT_LAST_ARRAY_ELEMENT+i-1); + WSTACK_PUSH2(s, (UWord) nobj, PRT_LAST_ARRAY_ELEMENT+i-1); } break; case FLOAT_DEF: { @@ -446,13 +470,65 @@ print_term(fmtfn_t fn, void* arg, Eterm obj, long *dcount, PRINT_STRING(res, fn, arg, "#MatchState"); } else { - ProcBin* pb = (ProcBin *) binary_val(wobj); - if (pb->size == 1) - PRINT_STRING(res, fn, arg, "<<1 byte>>"); - else { + byte* bytep; + Uint bytesize = binary_size_rel(obj,obj_base); + Uint bitoffs; + Uint bitsize; + byte octet; + ERTS_GET_BINARY_BYTES_REL(obj, bytep, bitoffs, bitsize, obj_base); + + if (bitsize || !bytesize + || !is_printable_ascii(bytep, bytesize, bitoffs)) { + int is_first = 1; PRINT_STRING(res, fn, arg, "<<"); - PRINT_UWORD(res, fn, arg, 'u', 0, 1, (ErlPfUWord) pb->size); - PRINT_STRING(res, fn, arg, " bytes>>"); + while (bytesize) { + if (is_first) + is_first = 0; + else + PRINT_CHAR(res, fn, arg, ','); + if (bitoffs) + octet = (bytep[0] << bitoffs) | (bytep[1] >> (8-bitoffs)); + else + octet = bytep[0]; + PRINT_UWORD(res, fn, arg, 'u', 0, 1, octet); + ++bytep; + --bytesize; + } + if (bitsize) { + Uint bits = bitoffs + bitsize; + octet = bytep[0]; + if (bits < 8) + octet >>= 8 - bits; + else if (bits > 8) { + bits -= 8; /* bits in last byte */ + octet <<= bits; + octet |= bytep[1] >> (8 - bits); + } + octet &= (1 << bitsize) - 1; + if (is_first) + is_first = 0; + else + PRINT_CHAR(res, fn, arg, ','); + PRINT_UWORD(res, fn, arg, 'u', 0, 1, octet); + PRINT_CHAR(res, fn, arg, ':'); + PRINT_UWORD(res, fn, arg, 'u', 0, 1, bitsize); + } + PRINT_STRING(res, fn, arg, ">>"); + } + else { + PRINT_STRING(res, fn, arg, "<<\""); + while (bytesize) { + if (bitoffs) + octet = (bytep[0] << bitoffs) | (bytep[1] >> (8-bitoffs)); + else + octet = bytep[0]; + if (octet == '"') + PRINT_CHAR(res, fn, arg, '\\'); + PRINT_CHAR(res, fn, arg, octet); + ++bytep; + --bytesize; + } + PRINT_STRING(res, fn, arg, "\">>"); } } break; @@ -492,33 +568,73 @@ print_term(fmtfn_t fn, void* arg, Eterm obj, long *dcount, { 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); + flatmap_t *mp = (flatmap_t *)flatmap_val(wobj); + n = flatmap_get_size(mp); + ks = flatmap_get_keys(mp); + vs = flatmap_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); - + WSTACK_PUSH5(s, vs[n], PRT_TERM, PRT_ASSOC, ks[n], 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); + WSTACK_PUSH6(s, PRT_COMMA, vs[n], PRT_TERM, PRT_ASSOC, + ks[n], PRT_TERM); } } } break; + case HASHMAP_DEF: + { + Uint n,mapval; + Eterm *head; + head = hashmap_val(wobj); + mapval = MAP_HEADER_VAL(*head); + switch (MAP_HEADER_TYPE(*head)) { + case MAP_HEADER_TAG_HAMT_HEAD_ARRAY: + case MAP_HEADER_TAG_HAMT_HEAD_BITMAP: + PRINT_STRING(res, fn, arg, "#<"); + PRINT_UWORD(res, fn, arg, 'x', 0, 1, mapval); + PRINT_STRING(res, fn, arg, ">{"); + WSTACK_PUSH(s,PRT_CLOSE_TUPLE); + n = hashmap_bitcount(mapval); + ASSERT(n < 17); + head += 2; + if (n > 0) { + n--; + WSTACK_PUSH(s, head[n]); + WSTACK_PUSH(s, PRT_TERM); + while (n--) { + WSTACK_PUSH(s, PRT_COMMA); + WSTACK_PUSH(s, head[n]); + WSTACK_PUSH(s, PRT_TERM); + } + } + break; + case MAP_HEADER_TAG_HAMT_NODE_BITMAP: + n = hashmap_bitcount(mapval); + head++; + PRINT_CHAR(res, fn, arg, '<'); + PRINT_UWORD(res, fn, arg, 'x', 0, 1, mapval); + PRINT_STRING(res, fn, arg, ">{"); + WSTACK_PUSH(s,PRT_CLOSE_TUPLE); + ASSERT(n < 17); + if (n > 0) { + n--; + WSTACK_PUSH(s, head[n]); + WSTACK_PUSH(s, PRT_TERM); + while (n--) { + WSTACK_PUSH(s, PRT_COMMA); + WSTACK_PUSH(s, head[n]); + WSTACK_PUSH(s, PRT_TERM); + } + } + break; + } + } + break; default: PRINT_STRING(res, fn, arg, "<unknown:"); PRINT_POINTER(res, fn, arg, wobj); @@ -528,17 +644,17 @@ print_term(fmtfn_t fn, void* arg, Eterm obj, long *dcount, } L_done: - DESTROY_WSTACK(s); return res; } + int erts_printf_term(fmtfn_t fn, void* arg, ErlPfEterm term, long precision, ErlPfEterm* term_base) { int res; - ASSERT(sizeof(ErlPfEterm) == sizeof(Eterm)); + ERTS_CT_ASSERT(sizeof(ErlPfEterm) == sizeof(Eterm)); res = print_term(fn, arg, (Eterm)term, &precision, (Eterm*)term_base); if (res < 0) diff --git a/erts/emulator/beam/erl_process.c b/erts/emulator/beam/erl_process.c index f84677dea4..ba09ee57c2 100644 --- a/erts/emulator/beam/erl_process.c +++ b/erts/emulator/beam/erl_process.c @@ -152,7 +152,7 @@ extern BeamInstr beam_continue_exit[]; /* Eager check I/O not supported on OSE yet. */ int erts_eager_check_io = 0; #else -int erts_eager_check_io = 0; +int erts_eager_check_io = 1; #endif int erts_sched_compact_load; int erts_sched_balance_util = 0; @@ -457,8 +457,7 @@ do { \ static void exec_misc_ops(ErtsRunQueue *); static void print_function_from_pc(int to, void *to_arg, BeamInstr* x); -static int stack_element_dump(int to, void *to_arg, Process* p, Eterm* sp, - int yreg); +static int stack_element_dump(int to, void *to_arg, Eterm* sp, int yreg); static void aux_work_timeout(void *unused); static void aux_work_timeout_early_init(int no_schedulers); @@ -716,72 +715,24 @@ sched_wall_time_ts(void) #if ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT -#ifdef ARCH_64 - -static ERTS_INLINE Uint64 -aschedtime_read(ErtsAtomicSchedTime *var) -{ - return (Uint64) erts_atomic_read_nob((erts_atomic_t *) var); -} - -static ERTS_INLINE void -aschedtime_set(ErtsAtomicSchedTime *var, Uint64 val) -{ - erts_atomic_set_nob((erts_atomic_t *) var, (erts_aint_t) val); -} - -static ERTS_INLINE void -aschedtime_init(ErtsAtomicSchedTime *var) -{ - erts_atomic_init_nob((erts_atomic_t *) var, (erts_aint_t) 0); -} - -#elif defined(ARCH_32) - static ERTS_INLINE Uint64 aschedtime_read(ErtsAtomicSchedTime *var) { - erts_dw_aint_t dw; - erts_dw_atomic_read_nob((erts_dw_atomic_t *) var, &dw); -#ifdef ETHR_SU_DW_NAINT_T__ - return (Uint64) dw.dw_sint; -#else - { - Uint64 res; - res = (Uint64) ((Uint32) dw.sint[ERTS_DW_AINT_HIGH_WORD]); - res <<= 32; - res |= (Uint64) ((Uint32) dw.sint[ERTS_DW_AINT_LOW_WORD]); - return res; - } -#endif + return (Uint64) erts_atomic64_read_nob((erts_atomic64_t *) var); } static ERTS_INLINE void aschedtime_set(ErtsAtomicSchedTime *var, Uint64 val) { - erts_dw_aint_t dw; -#ifdef ETHR_SU_DW_NAINT_T__ - dw.dw_sint = (ETHR_SU_DW_NAINT_T__) val; -#else - dw.sint[ERTS_DW_AINT_LOW_WORD] = (erts_aint_t) (val & 0xffffffff); - dw.sint[ERTS_DW_AINT_HIGH_WORD] = (erts_aint_t) ((val >> 32) & 0xffffffff); -#endif - erts_dw_atomic_set_nob((erts_dw_atomic_t *) var, &dw); + erts_atomic64_set_nob((erts_atomic64_t *) var, (erts_aint64_t) val); } static ERTS_INLINE void aschedtime_init(ErtsAtomicSchedTime *var) { - erts_dw_aint_t dw; - dw.sint[ERTS_DW_AINT_LOW_WORD] = (erts_aint_t) 0; - dw.sint[ERTS_DW_AINT_HIGH_WORD] = (erts_aint_t) 0; - erts_dw_atomic_init_nob((erts_dw_atomic_t *) var, &dw); + erts_atomic64_init_nob((erts_atomic64_t *) var, (erts_aint64_t) 0); } -#else -# error :-/ -#endif - #define ERTS_GET_AVG_MAX_UNLOCKED_TRY 50 #define ERTS_SCHED_AVG_UTIL_WRITE_MARKER (~((Uint64) 0)) @@ -2366,7 +2317,6 @@ erts_active_schedulers(void) ERTS_ATOMIC_FOREACH_RUNQ(rq, as -= abs(rq->waiting)); - ASSERT(as >= 0); return as; } @@ -7883,23 +7833,17 @@ erts_start_schedulers(void) Uint actual; Uint wanted = erts_no_schedulers; Uint wanted_no_schedulers = erts_no_schedulers; + char name[16]; ethr_thr_opts opts = ETHR_THR_OPTS_DEFAULT_INITER; opts.detached = 1; -#ifdef ETHR_HAVE_THREAD_NAMES - opts.name = malloc(80); - if (!opts.name) { - ERTS_INTERNAL_ERROR("malloc failed to allocate memory!"); - } -#endif + opts.name = name; #ifdef ERTS_SMP if (erts_runq_supervision_interval) { opts.suggested_stack_size = 16; -#ifdef ETHR_HAVE_THREAD_NAMES - sprintf(opts.name, "runq_supervisor"); -#endif + erts_snprintf(opts.name, 16, "runq_supervisor"); erts_atomic_init_nob(&runq_supervisor_sleeping, 0); if (0 != ethr_event_init(&runq_supervision_event)) erl_exit(1, "Failed to create run-queue supervision event\n"); @@ -7926,9 +7870,7 @@ erts_start_schedulers(void) ASSERT(actual == esdp->no - 1); -#ifdef ETHR_HAVE_THREAD_NAMES - sprintf(opts.name, "scheduler_%d", actual + 1); -#endif + erts_snprintf(opts.name, 16, "%lu_scheduler", actual + 1); #ifdef __OSE__ /* This should be done in the bind strategy */ @@ -7950,18 +7892,14 @@ erts_start_schedulers(void) int ix; for (ix = 0; ix < erts_no_dirty_cpu_schedulers; ix++) { ErtsSchedulerData *esdp = ERTS_DIRTY_CPU_SCHEDULER_IX(ix); -#ifdef ETHR_HAVE_THREAD_NAMES - sprintf(opts.name,"dirty_cpu_scheduler_%d", ix + 1); -#endif + erts_snprintf(opts.name, 16, "%d_dirty_cpu_scheduler", ix + 1); 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); -#ifdef ETHR_HAVE_THREAD_NAMES - sprintf(opts.name,"dirty_io_scheduler_%d", ix + 1); -#endif + erts_snprintf(opts.name, 16, "%d_dirty_io_scheduler", ix + 1); 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); @@ -7972,9 +7910,7 @@ erts_start_schedulers(void) ERTS_THR_MEMORY_BARRIER; -#ifdef ETHR_HAVE_THREAD_NAMES - sprintf(opts.name, "aux"); -#endif + erts_snprintf(opts.name, 16, "aux"); #ifdef __OSE__ opts.coreNo = 0; @@ -8000,9 +7936,6 @@ erts_start_schedulers(void) erts_send_error_to_logger_nogl(dsbufp); } -#ifdef ETHR_HAVE_THREAD_NAMES - free(opts.name); -#endif } #endif /* ERTS_SMP */ @@ -10484,7 +10417,7 @@ alloc_process(ErtsRunQueue *rq, erts_aint32_t state) init_arg.run_queue = rq; init_arg.state = state; - ASSERT(((char *) p) == ((char *) &p->common)); + ERTS_CT_ASSERT(offsetof(Process,common) == 0); if (!erts_ptab_new_element(&erts_proc, &p->common, @@ -12203,7 +12136,7 @@ erts_stack_dump(int to, void *to_arg, Process *p) } erts_program_counter_info(to, to_arg, p); for (sp = p->stop; sp < STACK_START(p); sp++) { - yreg = stack_element_dump(to, to_arg, p, sp, yreg); + yreg = stack_element_dump(to, to_arg, sp, yreg); } } @@ -12260,7 +12193,7 @@ print_function_from_pc(int to, void *to_arg, BeamInstr* x) } static int -stack_element_dump(int to, void *to_arg, Process* p, Eterm* sp, int yreg) +stack_element_dump(int to, void *to_arg, Eterm* sp, int yreg) { Eterm x = *sp; @@ -12289,6 +12222,214 @@ stack_element_dump(int to, void *to_arg, Process* p, Eterm* sp, int yreg) } /* + * Print scheduler information + */ +void +erts_print_scheduler_info(int to, void *to_arg, ErtsSchedulerData *esdp) { + int i; + erts_aint32_t flg; + Process *p; + + erts_print(to, to_arg, "=scheduler:%u\n", esdp->no); + +#ifdef ERTS_SMP + flg = erts_smp_atomic32_read_dirty(&esdp->ssi->flags); + erts_print(to, to_arg, "Scheduler Sleep Info Flags: "); + for (i = 0; i < ERTS_SSI_FLGS_MAX && flg; i++) { + erts_aint32_t chk = (1 << i); + if (flg & chk) { + switch (chk) { + case ERTS_SSI_FLG_SLEEPING: + erts_print(to, to_arg, "SLEEPING"); break; + case ERTS_SSI_FLG_POLL_SLEEPING: + erts_print(to, to_arg, "POLL_SLEEPING"); break; + case ERTS_SSI_FLG_TSE_SLEEPING: + erts_print(to, to_arg, "TSE_SLEEPING"); break; + case ERTS_SSI_FLG_WAITING: + erts_print(to, to_arg, "WAITING"); break; + case ERTS_SSI_FLG_SUSPENDED: + erts_print(to, to_arg, "SUSPENDED"); break; + default: + erts_print(to, to_arg, "UNKNOWN(%d)", flg); break; + } + if (flg > chk) + erts_print(to, to_arg, " | "); + flg -= chk; + } + } + erts_print(to, to_arg, "\n"); +#endif + + flg = erts_atomic32_read_dirty(&esdp->ssi->aux_work); + erts_print(to, to_arg, "Scheduler Sleep Info Aux Work: "); + for (i = 0; i < ERTS_SSI_AUX_WORK_MAX && flg; i++) { + erts_aint32_t chk = (1 << i); + if (flg & chk) { + switch (chk) { + case ERTS_SSI_AUX_WORK_DELAYED_AW_WAKEUP: + erts_print(to, to_arg, "DELAYED_AW_WAKEUP"); break; + case ERTS_SSI_AUX_WORK_DD: + erts_print(to, to_arg, "DELAYED_DEALLOC"); break; + case ERTS_SSI_AUX_WORK_DD_THR_PRGR: + erts_print(to, to_arg, "DELAYED_DEALLOC_THR_PRGR"); break; + case ERTS_SSI_AUX_WORK_FIX_ALLOC_DEALLOC: + erts_print(to, to_arg, "FIX_ALLOC_DEALLOC"); break; + case ERTS_SSI_AUX_WORK_FIX_ALLOC_LOWER_LIM: + erts_print(to, to_arg, "FIX_ALLOC_LOWER_LIM"); break; + case ERTS_SSI_AUX_WORK_THR_PRGR_LATER_OP: + erts_print(to, to_arg, "THR_PRGR_LATER_OP"); break; + case ERTS_SSI_AUX_WORK_ASYNC_READY: + erts_print(to, to_arg, "ASYNC_READY"); break; + case ERTS_SSI_AUX_WORK_ASYNC_READY_CLEAN: + erts_print(to, to_arg, "ASYNC_READY_CLEAN"); break; + case ERTS_SSI_AUX_WORK_MISC_THR_PRGR: + erts_print(to, to_arg, "MISC_THR_PRGR"); break; + case ERTS_SSI_AUX_WORK_MISC: + erts_print(to, to_arg, "MISC"); break; + case ERTS_SSI_AUX_WORK_CHECK_CHILDREN: + erts_print(to, to_arg, "CHECK_CHILDREN"); break; + case ERTS_SSI_AUX_WORK_SET_TMO: + erts_print(to, to_arg, "SET_TMO"); break; + case ERTS_SSI_AUX_WORK_MSEG_CACHE_CHECK: + erts_print(to, to_arg, "MSEG_CACHE_CHECK"); break; + case ERTS_SSI_AUX_WORK_REAP_PORTS: + erts_print(to, to_arg, "REAP_PORTS"); break; + default: + erts_print(to, to_arg, "UNKNOWN(%d)", flg); break; + } + if (flg > chk) + erts_print(to, to_arg, " | "); + flg -= chk; + } + } + erts_print(to, to_arg, "\n"); + + erts_print(to, to_arg, "Current Port: "); + if (esdp->current_port) + erts_print(to, to_arg, "%T", esdp->current_port->common.id); + erts_print(to, to_arg, "\n"); + + p = esdp->current_process; + erts_print(to, to_arg, "Current Process: "); + if (esdp->current_process && !(ERTS_TRACE_FLAGS(p) & F_SENSITIVE)) { + flg = erts_smp_atomic32_read_dirty(&p->state); + erts_print(to, to_arg, "%T\n", p->common.id); + + erts_print(to, to_arg, "Current Process State: "); + erts_dump_process_state(to, to_arg, flg); + + erts_print(to, to_arg, "Current Process Internal State: "); + erts_dump_extended_process_state(to, to_arg, flg); + + erts_print(to, to_arg, "Current Process Program counter: %p (", p->i); + print_function_from_pc(to, to_arg, p->i); + erts_print(to, to_arg, ")\n"); + erts_print(to, to_arg, "Current Process CP: %p (", p->cp); + print_function_from_pc(to, to_arg, p->cp); + erts_print(to, to_arg, ")\n"); + + /* Getting this stacktrace can segfault if we are very very + unlucky if called while a process is being garbage collected. + Therefore we only call this on other schedulers if we either + have protection against segfaults, or we know that the process + is not garbage collecting. It *should* always be safe to call + on a process owned by us, even if it is currently being garbage + collected. + */ + erts_print(to, to_arg, "Current Process Limited Stack Trace:\n"); + erts_limited_stack_trace(to, to_arg, p); + } else + erts_print(to, to_arg, "\n"); + + for (i = 0; i < ERTS_NO_PROC_PRIO_LEVELS; i++) { + erts_print(to, to_arg, "Run Queue "); + switch (i) { + case PRIORITY_MAX: + erts_print(to, to_arg, "Max "); + break; + case PRIORITY_HIGH: + erts_print(to, to_arg, "High "); + break; + case PRIORITY_NORMAL: + erts_print(to, to_arg, "Normal "); + break; + case PRIORITY_LOW: + erts_print(to, to_arg, "Low "); + break; + default: + erts_print(to, to_arg, "Unknown "); + break; + } + erts_print(to, to_arg, "Length: %d\n", + erts_smp_atomic32_read_dirty(&esdp->run_queue->procs.prio_info[i].len)); + } + erts_print(to, to_arg, "Run Queue Port Length: %d\n", + erts_smp_atomic32_read_dirty(&esdp->run_queue->ports.info.len)); + + flg = erts_smp_atomic32_read_dirty(&esdp->run_queue->flags); + erts_print(to, to_arg, "Run Queue Flags: "); + for (i = 0; i < ERTS_RUNQ_FLG_MAX && flg; i++) { + erts_aint32_t chk = (1 << i); + if (flg & chk) { + switch (chk) { + case (1 << PRIORITY_MAX): + erts_print(to, to_arg, "NONEMPTY_MAX"); break; + case (1 << PRIORITY_HIGH): + erts_print(to, to_arg, "NONEMPTY_HIGH"); break; + case (1 << PRIORITY_NORMAL): + erts_print(to, to_arg, "NONEMPTY_NORMAL"); break; + case (1 << PRIORITY_LOW): + erts_print(to, to_arg, "NONEMPTY_LOW"); break; + case (1 << (PRIORITY_MAX + ERTS_RUNQ_FLGS_EMIGRATE_SHFT)): + erts_print(to, to_arg, "EMIGRATE_MAX"); break; + case (1 << (PRIORITY_HIGH + ERTS_RUNQ_FLGS_EMIGRATE_SHFT)): + erts_print(to, to_arg, "EMIGRATE_HIGH"); break; + case (1 << (PRIORITY_NORMAL + ERTS_RUNQ_FLGS_EMIGRATE_SHFT)): + erts_print(to, to_arg, "EMIGRATE_NORMAL"); break; + case (1 << (PRIORITY_LOW + ERTS_RUNQ_FLGS_EMIGRATE_SHFT)): + erts_print(to, to_arg, "EMIGRATE_LOW"); break; + case (1 << (PRIORITY_MAX + ERTS_RUNQ_FLGS_IMMIGRATE_SHFT)): + erts_print(to, to_arg, "IMMIGRATE_MAX"); break; + case (1 << (PRIORITY_HIGH + ERTS_RUNQ_FLGS_IMMIGRATE_SHFT)): + erts_print(to, to_arg, "IMMIGRATE_HIGH"); break; + case (1 << (PRIORITY_NORMAL + ERTS_RUNQ_FLGS_IMMIGRATE_SHFT)): + erts_print(to, to_arg, "IMMIGRATE_NORMAL"); break; + case (1 << (PRIORITY_LOW + ERTS_RUNQ_FLGS_IMMIGRATE_SHFT)): + erts_print(to, to_arg, "IMMIGRATE_LOW"); break; + case (1 << (PRIORITY_MAX + ERTS_RUNQ_FLGS_EVACUATE_SHFT)): + erts_print(to, to_arg, "EVACUATE_MAX"); break; + case (1 << (PRIORITY_HIGH + ERTS_RUNQ_FLGS_EVACUATE_SHFT)): + erts_print(to, to_arg, "EVACUATE_HIGH"); break; + case (1 << (PRIORITY_NORMAL + ERTS_RUNQ_FLGS_EVACUATE_SHFT)): + erts_print(to, to_arg, "EVACUATE_NORMAL"); break; + case (1 << (PRIORITY_LOW + ERTS_RUNQ_FLGS_EVACUATE_SHFT)): + erts_print(to, to_arg, "EVACUATE_LOW"); break; + case ERTS_RUNQ_FLG_OUT_OF_WORK: + erts_print(to, to_arg, "OUT_OF_WORK"); break; + case ERTS_RUNQ_FLG_HALFTIME_OUT_OF_WORK: + erts_print(to, to_arg, "HALFTIME_OUT_OF_WORK"); break; + case ERTS_RUNQ_FLG_SUSPENDED: + erts_print(to, to_arg, "SUSPENDED"); break; + case ERTS_RUNQ_FLG_CHK_CPU_BIND: + erts_print(to, to_arg, "CHK_CPU_BIND"); break; + case ERTS_RUNQ_FLG_INACTIVE: + erts_print(to, to_arg, "INACTIVE"); break; + case ERTS_RUNQ_FLG_NONEMPTY: + erts_print(to, to_arg, "NONEMPTY"); break; + case ERTS_RUNQ_FLG_PROTECTED: + erts_print(to, to_arg, "PROTECTED"); break; + default: + erts_print(to, to_arg, "UNKNOWN(%d)", flg); break; + } + if (flg > chk) + erts_print(to, to_arg, " | "); + flg -= chk; + } + } + erts_print(to, to_arg, "\n"); +} + +/* * A nice system halt closing all open port goes as follows: * 1) This function schedules the aux work ERTS_SSI_AUX_WORK_REAP_PORTS * on all schedulers, then schedules itself out. diff --git a/erts/emulator/beam/erl_process.h b/erts/emulator/beam/erl_process.h index 27a3a3553b..d12ac792af 100644 --- a/erts/emulator/beam/erl_process.h +++ b/erts/emulator/beam/erl_process.h @@ -58,6 +58,7 @@ typedef struct process Process; #include "external.h" #include "erl_mseg.h" #include "erl_async.h" +#include "erl_gc.h" #ifdef HIPE #include "hipe_process.h" @@ -169,6 +170,8 @@ extern int erts_sched_thread_suggested_stack_size; #define ERTS_RUNQ_FLG_PROTECTED \ (((Uint32) 1) << (ERTS_RUNQ_FLG_BASE2 + 6)) +#define ERTS_RUNQ_FLG_MAX (ERTS_RUNQ_FLG_BASE2 + 7) + #define ERTS_RUNQ_FLGS_MIGRATION_QMASKS \ (ERTS_RUNQ_FLGS_EMIGRATE_QMASK \ | ERTS_RUNQ_FLGS_IMMIGRATE_QMASK \ @@ -251,6 +254,8 @@ typedef enum { #define ERTS_SSI_FLG_WAITING (((erts_aint32_t) 1) << 3) #define ERTS_SSI_FLG_SUSPENDED (((erts_aint32_t) 1) << 4) +#define ERTS_SSI_FLGS_MAX 5 + #define ERTS_SSI_FLGS_SLEEP_TYPE \ (ERTS_SSI_FLG_TSE_SLEEPING|ERTS_SSI_FLG_POLL_SLEEPING) @@ -282,6 +287,8 @@ typedef enum { #define ERTS_SSI_AUX_WORK_MSEG_CACHE_CHECK (((erts_aint32_t) 1) << 12) #define ERTS_SSI_AUX_WORK_REAP_PORTS (((erts_aint32_t) 1) << 13) +#define ERTS_SSI_AUX_WORK_MAX 14 + typedef struct ErtsSchedulerSleepInfo_ ErtsSchedulerSleepInfo; #ifdef ERTS_DIRTY_SCHEDULERS @@ -351,13 +358,7 @@ typedef struct { #undef ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT #define ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT_OPT -#ifdef ARCH_64 -typedef erts_atomic_t ErtsAtomicSchedTime; -#elif defined(ARCH_32) -typedef erts_dw_atomic_t ErtsAtomicSchedTime; -#else -# error :-/ -#endif +typedef erts_atomic64_t ErtsAtomicSchedTime; #if ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT typedef struct { @@ -489,11 +490,6 @@ typedef struct { } ErtsSchedWallTime; typedef struct { - Uint64 reclaimed; - Uint64 garbage_cols; -} ErtsGCInfo; - -typedef struct { int sched; erts_aint32_t aux_work; } ErtsDelayedAuxWorkWakeupJob; @@ -1085,6 +1081,9 @@ void erts_check_for_holes(Process* p); #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) +#define ERTS_PSFLG_MAX (ERTS_PSFLGS_ZERO_BIT_OFFSET + 22) +#else +#define ERTS_PSFLG_MAX (ERTS_PSFLGS_ZERO_BIT_OFFSET + 18) #endif #define ERTS_PSFLGS_IN_PRQ_MASK (ERTS_PSFLG_IN_PRQ_MAX \ @@ -1620,7 +1619,11 @@ void erts_cleanup_empty_process(Process* p); void erts_debug_verify_clean_empty_process(Process* p); #endif void erts_stack_dump(int to, void *to_arg, Process *); +void erts_limited_stack_trace(int to, void *to_arg, Process *); void erts_program_counter_info(int to, void *to_arg, Process *); +void erts_print_scheduler_info(int to, void *to_arg, ErtsSchedulerData *esdp); +void erts_dump_extended_process_state(int to, void *to_arg, erts_aint32_t psflg); +void erts_dump_process_state(int to, void *to_arg, erts_aint32_t psflg); Eterm erts_get_process_priority(Process *p); Eterm erts_set_process_priority(Process *p, Eterm prio); diff --git a/erts/emulator/beam/erl_process_dict.c b/erts/emulator/beam/erl_process_dict.c index 23e5bf737f..00761f2d0e 100644 --- a/erts/emulator/beam/erl_process_dict.c +++ b/erts/emulator/beam/erl_process_dict.c @@ -47,7 +47,7 @@ /* Hash constant macros */ #define MAX_HASH 1342177280UL -#define INITIAL_SIZE 10 +#define INITIAL_SIZE (erts_pd_initial_size) /* Hash utility macros */ #define HASH_RANGE(PDict) ((PDict)->homeSize + (PDict)->splitPosition) @@ -82,6 +82,7 @@ static void pd_hash_erase(Process *p, Eterm id, Eterm *ret); static void pd_hash_erase_all(Process *p); static Eterm pd_hash_get_keys(Process *p, Eterm value); +static Eterm pd_hash_get_all_keys(Process *p, ProcDict *pd); static Eterm pd_hash_get_all(Process *p, ProcDict *pd); static Eterm pd_hash_put(Process *p, Eterm id, Eterm value); @@ -275,6 +276,16 @@ BIF_RETTYPE get_1(BIF_ALIST_1) BIF_RET(ret); } +BIF_RETTYPE get_keys_0(BIF_ALIST_0) +{ + Eterm ret; + + PD_CHECK(BIF_P->dictionary); + ret = pd_hash_get_all_keys(BIF_P,BIF_P->dictionary); + PD_CHECK(BIF_P->dictionary); + BIF_RET(ret); +} + BIF_RETTYPE get_keys_1(BIF_ALIST_1) { Eterm ret; @@ -412,6 +423,47 @@ Eterm erts_pd_hash_get(Process *p, Eterm id) return am_undefined; } +#define PD_GET_TKEY(Dst,Src) \ +do { \ + ASSERT(is_tuple((Src))); \ + ASSERT(arityval(*((Eterm*)tuple_val((Src)))) == 2); \ + (Dst) = ((Eterm*)tuple_val((Src)))[1]; \ +} while(0) + +static Eterm pd_hash_get_all_keys(Process *p, ProcDict *pd) { + Eterm* hp; + Eterm res = NIL; + Eterm tmp, tmp2; + unsigned int i; + unsigned int num; + + if (pd == NULL) { + return res; + } + + num = HASH_RANGE(pd); + hp = HAlloc(p, pd->numElements * 2); + + for (i = 0; i < num; ++i) { + tmp = ARRAY_GET(pd, i); + if (is_boxed(tmp)) { + PD_GET_TKEY(tmp,tmp); + res = CONS(hp, tmp, res); + hp += 2; + } else if (is_list(tmp)) { + while (tmp != NIL) { + tmp2 = TCAR(tmp); + PD_GET_TKEY(tmp2,tmp2); + res = CONS(hp, tmp2, res); + hp += 2; + tmp = TCDR(tmp); + } + } + } + return res; +} +#undef PD_GET_TKEY + static Eterm pd_hash_get_keys(Process *p, Eterm value) { Eterm *hp; diff --git a/erts/emulator/beam/erl_process_dump.c b/erts/emulator/beam/erl_process_dump.c index 2f3cf23b00..36bb6b2f0e 100644 --- a/erts/emulator/beam/erl_process_dump.c +++ b/erts/emulator/beam/erl_process_dump.c @@ -43,8 +43,9 @@ static void dump_process_info(int to, void *to_arg, Process *p); static void dump_element(int to, void *to_arg, Eterm x); static void dump_dist_ext(int to, void *to_arg, ErtsDistExternal *edep); static void dump_element_nl(int to, void *to_arg, Eterm x); -static int stack_element_dump(int to, void *to_arg, Process* p, Eterm* sp, +static int stack_element_dump(int to, void *to_arg, Eterm* sp, int yreg); +static void stack_trace_dump(int to, void *to_arg, Eterm* sp); static void print_function_from_pc(int to, void *to_arg, BeamInstr* x); static void heap_dump(int to, void *to_arg, Eterm x); static void dump_binaries(int to, void *to_arg, Binary* root); @@ -148,7 +149,7 @@ dump_process_info(int to, void *to_arg, Process *p) if ((ERTS_TRACE_FLAGS(p) & F_SENSITIVE) == 0) { erts_print(to, to_arg, "=proc_stack:%T\n", p->common.id); for (sp = p->stop; sp < STACK_START(p); sp++) { - yreg = stack_element_dump(to, to_arg, p, sp, yreg); + yreg = stack_element_dump(to, to_arg, sp, yreg); } erts_print(to, to_arg, "=proc_heap:%T\n", p->common.id); @@ -243,9 +244,65 @@ dump_element_nl(int to, void *to_arg, Eterm x) erts_putc(to, to_arg, '\n'); } +static void +stack_trace_dump(int to, void *to_arg, Eterm *sp) { + Eterm x = *sp; + if (is_CP(x)) { + erts_print(to, to_arg, "%p:", sp); + erts_print(to, to_arg, "SReturn addr 0x%X (", cp_val(x)); + print_function_from_pc(to, to_arg, cp_val(x)); + erts_print(to, to_arg, ")\n"); + } +} + +void +erts_limited_stack_trace(int to, void *to_arg, Process *p) +{ + Eterm* sp; + + + if (ERTS_TRACE_FLAGS(p) & F_SENSITIVE) { + return; + } + + if (STACK_START(p) < STACK_TOP(p)) { + return; + } + + if ((STACK_START(p) - STACK_TOP(p)) < 512) { + if (erts_sys_is_area_readable((char*)STACK_TOP(p), + (char*)STACK_START(p))) + for (sp = STACK_TOP(p); sp < STACK_START(p); sp++) + stack_trace_dump(to, to_arg, sp); + else + erts_print(to, to_arg, "Could not read from stack memory: %p - %p\n", + STACK_TOP(p), STACK_START(p)); + } else { + sp = STACK_TOP(p); + if (erts_sys_is_area_readable((char*)STACK_TOP(p), + (char*)(STACK_TOP(p) + 25))) + for (; sp < (STACK_TOP(p) + 256); sp++) + stack_trace_dump(to, to_arg, sp); + else + erts_print(to, to_arg, "Could not read from stack memory: %p - %p\n", + STACK_TOP(p), STACK_TOP(p) + 256); + + erts_print(to, to_arg, "%p: skipping %d frames\n", + sp, STACK_START(p) - STACK_TOP(p) - 512); + + if (erts_sys_is_area_readable((char*)(STACK_START(p) - 256), + (char*)STACK_START(p))) + for (sp = STACK_START(p) - 256; sp < STACK_START(p); sp++) + stack_trace_dump(to, to_arg, sp); + else + erts_print(to, to_arg, "Could not read from stack memory: %p - %p\n", + STACK_START(p) - 256, STACK_START(p)); + } + +} static int -stack_element_dump(int to, void *to_arg, Process* p, Eterm* sp, int yreg) +stack_element_dump(int to, void *to_arg, Eterm* sp, int yreg) { Eterm x = *sp; @@ -508,3 +565,114 @@ dump_externally(int to, void *to_arg, Eterm term) erts_print(to, to_arg, "%02X", *s++); } } + +void erts_dump_process_state(int to, void *to_arg, erts_aint32_t psflg) { + if (psflg & ERTS_PSFLG_FREE) + erts_print(to, to_arg, "Non Existing\n"); /* Should never happen */ + else if (psflg & ERTS_PSFLG_EXITING) + erts_print(to, to_arg, "Exiting\n"); + else if (psflg & ERTS_PSFLG_GC) { + erts_print(to, to_arg, "Garbing\n"); + } + else if (psflg & ERTS_PSFLG_SUSPENDED) + erts_print(to, to_arg, "Suspended\n"); + else if (psflg & ERTS_PSFLG_RUNNING) { + erts_print(to, to_arg, "Running\n"); + } + else if (psflg & ERTS_PSFLG_ACTIVE) + erts_print(to, to_arg, "Scheduled\n"); + else + erts_print(to, to_arg, "Waiting\n"); +} + +void +erts_dump_extended_process_state(int to, void *to_arg, erts_aint32_t psflg) { + + int i; + + switch (ERTS_PSFLGS_GET_ACT_PRIO(psflg)) { + case PRIORITY_MAX: erts_print(to, to_arg, "ACT_PRIO_MAX | "); break; + case PRIORITY_HIGH: erts_print(to, to_arg, "ACT_PRIO_HIGH | "); break; + case PRIORITY_NORMAL: erts_print(to, to_arg, "ACT_PRIO_NORMAL | "); break; + case PRIORITY_LOW: erts_print(to, to_arg, "ACT_PRIO_LOW | "); break; + } + switch (ERTS_PSFLGS_GET_USR_PRIO(psflg)) { + case PRIORITY_MAX: erts_print(to, to_arg, "USR_PRIO_MAX | "); break; + case PRIORITY_HIGH: erts_print(to, to_arg, "USR_PRIO_HIGH | "); break; + case PRIORITY_NORMAL: erts_print(to, to_arg, "USR_PRIO_NORMAL | "); break; + case PRIORITY_LOW: erts_print(to, to_arg, "USR_PRIO_LOW | "); break; + } + switch (ERTS_PSFLGS_GET_PRQ_PRIO(psflg)) { + case PRIORITY_MAX: erts_print(to, to_arg, "PRQ_PRIO_MAX"); break; + case PRIORITY_HIGH: erts_print(to, to_arg, "PRQ_PRIO_HIGH"); break; + case PRIORITY_NORMAL: erts_print(to, to_arg, "PRQ_PRIO_NORMAL"); break; + case PRIORITY_LOW: erts_print(to, to_arg, "PRQ_PRIO_LOW"); break; + } + + psflg &= ~(ERTS_PSFLGS_ACT_PRIO_MASK | + ERTS_PSFLGS_USR_PRIO_MASK | + ERTS_PSFLGS_PRQ_PRIO_MASK); + + if (psflg) + erts_print(to, to_arg, " | "); + + for (i = 0; i < ERTS_PSFLG_MAX && psflg; i++) { + erts_aint32_t chk = (1 << i); + if (psflg & chk) { + switch (chk) { + case ERTS_PSFLG_IN_PRQ_MAX: + erts_print(to, to_arg, "IN_PRQ_MAX"); break; + case ERTS_PSFLG_IN_PRQ_HIGH: + erts_print(to, to_arg, "IN_PRQ_HIGH"); break; + case ERTS_PSFLG_IN_PRQ_NORMAL: + erts_print(to, to_arg, "IN_PRQ_NORMAL"); break; + case ERTS_PSFLG_IN_PRQ_LOW: + erts_print(to, to_arg, "IN_PRQ_LOW"); break; + case ERTS_PSFLG_FREE: + erts_print(to, to_arg, "FREE"); break; + case ERTS_PSFLG_EXITING: + erts_print(to, to_arg, "EXITING"); break; + case ERTS_PSFLG_PENDING_EXIT: + erts_print(to, to_arg, "PENDING_EXIT"); break; + case ERTS_PSFLG_ACTIVE: + erts_print(to, to_arg, "ACTIVE"); break; + case ERTS_PSFLG_IN_RUNQ: + erts_print(to, to_arg, "IN_RUNQ"); break; + case ERTS_PSFLG_RUNNING: + erts_print(to, to_arg, "RUNNING"); break; + case ERTS_PSFLG_SUSPENDED: + erts_print(to, to_arg, "SUSPENDED"); break; + case ERTS_PSFLG_GC: + erts_print(to, to_arg, "GC"); break; + case ERTS_PSFLG_BOUND: + erts_print(to, to_arg, "BOUND"); break; + case ERTS_PSFLG_TRAP_EXIT: + erts_print(to, to_arg, "TRAP_EXIT"); break; + case ERTS_PSFLG_ACTIVE_SYS: + erts_print(to, to_arg, "ACTIVE_SYS"); break; + case ERTS_PSFLG_RUNNING_SYS: + erts_print(to, to_arg, "RUNNING_SYS"); break; + case ERTS_PSFLG_PROXY: + erts_print(to, to_arg, "PROXY"); break; + case ERTS_PSFLG_DELAYED_SYS: + erts_print(to, to_arg, "DELAYED_SYS"); break; +#ifdef ERTS_DIRTY_SCHEDULERS + case ERTS_PSFLG_DIRTY_CPU_PROC: + erts_print(to, to_arg, "DIRTY_CPU_PROC"); break; + case ERTS_PSFLG_DIRTY_IO_PROC: + erts_print(to, to_arg, "DIRTY_IO_PROC"); break; + case ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q: + erts_print(to, to_arg, "DIRTY_CPU_PROC_IN_Q"); break; + case ERTS_PSFLG_DIRTY_IO_PROC_IN_Q: + erts_print(to, to_arg, "DIRTY_IO_PROC_IN_Q"); break; +#endif + default: + erts_print(to, to_arg, "UNKNOWN(%d)", chk); break; + } + if (psflg > chk) + erts_print(to, to_arg, " | "); + psflg -= chk; + } + } + erts_print(to, to_arg, "\n"); +} diff --git a/erts/emulator/beam/erl_ptab.c b/erts/emulator/beam/erl_ptab.c index eabf016081..02943ee683 100644 --- a/erts/emulator/beam/erl_ptab.c +++ b/erts/emulator/beam/erl_ptab.c @@ -280,124 +280,38 @@ struct ErtsPTabListBifData_ { }; -#ifdef ARCH_32 - -static ERTS_INLINE Uint64 -dw_aint_to_uint64(erts_dw_aint_t *dw) -{ -#ifdef ETHR_SU_DW_NAINT_T__ - return (Uint64) dw->dw_sint; -#else - Uint64 res; - res = (Uint64) ((Uint32) dw->sint[ERTS_DW_AINT_HIGH_WORD]); - res <<= 32; - res |= (Uint64) ((Uint32) dw->sint[ERTS_DW_AINT_LOW_WORD]); - return res; -#endif -} - -static void -unint64_to_dw_aint(erts_dw_aint_t *dw, Uint64 val) -{ -#ifdef ETHR_SU_DW_NAINT_T__ - dw->dw_sint = (ETHR_SU_DW_NAINT_T__) val; -#else - dw->sint[ERTS_DW_AINT_LOW_WORD] = (erts_aint_t) (val & 0xffffffff); - dw->sint[ERTS_DW_AINT_HIGH_WORD] = (erts_aint_t) ((val >> 32) & 0xffffffff); -#endif -} - static ERTS_INLINE void last_data_init_nob(ErtsPTab *ptab, Uint64 val) { - erts_dw_aint_t dw; - unint64_to_dw_aint(&dw, val); - erts_smp_dw_atomic_init_nob(&ptab->vola.tile.last_data, &dw); + erts_smp_atomic64_init_nob(&ptab->vola.tile.last_data, (erts_aint64_t) val); } static ERTS_INLINE void last_data_set_relb(ErtsPTab *ptab, Uint64 val) { - erts_dw_aint_t dw; - unint64_to_dw_aint(&dw, val); - erts_smp_dw_atomic_set_relb(&ptab->vola.tile.last_data, &dw); + erts_smp_atomic64_set_relb(&ptab->vola.tile.last_data, (erts_aint64_t) val); } static ERTS_INLINE Uint64 last_data_read_nob(ErtsPTab *ptab) { - erts_dw_aint_t dw; - erts_smp_dw_atomic_read_nob(&ptab->vola.tile.last_data, &dw); - return dw_aint_to_uint64(&dw); + return (Uint64) erts_smp_atomic64_read_nob(&ptab->vola.tile.last_data); } static ERTS_INLINE Uint64 last_data_read_acqb(ErtsPTab *ptab) { - erts_dw_aint_t dw; - erts_smp_dw_atomic_read_acqb(&ptab->vola.tile.last_data, &dw); - return dw_aint_to_uint64(&dw); + return (Uint64) erts_smp_atomic64_read_acqb(&ptab->vola.tile.last_data); } static ERTS_INLINE Uint64 last_data_cmpxchg_relb(ErtsPTab *ptab, Uint64 new, Uint64 exp) { - erts_dw_aint_t dw_new, dw_xchg; - - unint64_to_dw_aint(&dw_new, new); - unint64_to_dw_aint(&dw_xchg, exp); - - if (erts_smp_dw_atomic_cmpxchg_relb(&ptab->vola.tile.last_data, - &dw_new, - &dw_xchg)) - return exp; - else - return dw_aint_to_uint64(&dw_xchg); -} - -#elif defined(ARCH_64) - -union { - erts_smp_atomic_t pid_data; - char align[ERTS_CACHE_LINE_SIZE]; -} last erts_align_attribute(ERTS_CACHE_LINE_SIZE); - -static ERTS_INLINE void -last_data_init_nob(ErtsPTab *ptab, Uint64 val) -{ - erts_smp_atomic_init_nob(&ptab->vola.tile.last_data, (erts_aint_t) val); + return (Uint64) erts_smp_atomic64_cmpxchg_relb(&ptab->vola.tile.last_data, + (erts_aint64_t) new, + (erts_aint64_t) exp); } -static ERTS_INLINE void -last_data_set_relb(ErtsPTab *ptab, Uint64 val) -{ - erts_smp_atomic_set_relb(&ptab->vola.tile.last_data, (erts_aint_t) val); -} - -static ERTS_INLINE Uint64 -last_data_read_nob(ErtsPTab *ptab) -{ - return (Uint64) erts_smp_atomic_read_nob(&ptab->vola.tile.last_data); -} - -static ERTS_INLINE Uint64 -last_data_read_acqb(ErtsPTab *ptab) -{ - return (Uint64) erts_smp_atomic_read_acqb(&ptab->vola.tile.last_data); -} - -static ERTS_INLINE Uint64 -last_data_cmpxchg_relb(ErtsPTab *ptab, Uint64 new, Uint64 exp) -{ - return (Uint64) erts_smp_atomic_cmpxchg_relb(&ptab->vola.tile.last_data, - (erts_aint_t) new, - (erts_aint_t) exp); -} - -#else -# error "Not 64-bit, nor 32-bit architecture..." -#endif - static ERTS_INLINE int last_data_cmp(Uint64 ld1, Uint64 ld2) { diff --git a/erts/emulator/beam/erl_ptab.h b/erts/emulator/beam/erl_ptab.h index e3e05f14af..876241159b 100644 --- a/erts/emulator/beam/erl_ptab.h +++ b/erts/emulator/beam/erl_ptab.h @@ -88,11 +88,7 @@ typedef struct { } ErtsPTabListData; typedef struct { -#ifdef ARCH_32 - erts_smp_dw_atomic_t last_data; -#else - erts_smp_atomic_t last_data; -#endif + erts_smp_atomic64_t last_data; erts_smp_atomic32_t count; erts_smp_atomic32_t aid_ix; erts_smp_atomic32_t fid_ix; diff --git a/erts/emulator/beam/erl_smp.h b/erts/emulator/beam/erl_smp.h index c38ef47d87..6c40edeb3e 100644 --- a/erts/emulator/beam/erl_smp.h +++ b/erts/emulator/beam/erl_smp.h @@ -60,6 +60,7 @@ typedef erts_tsd_key_t erts_smp_tsd_key_t; #define erts_smp_dw_atomic_t erts_dw_atomic_t #define erts_smp_atomic_t erts_atomic_t #define erts_smp_atomic32_t erts_atomic32_t +#define erts_smp_atomic64_t erts_atomic64_t typedef erts_spinlock_t erts_smp_spinlock_t; typedef erts_rwlock_t erts_smp_rwlock_t; void erts_thr_fatal_error(int, char *); /* implemented in erl_init.c */ @@ -95,6 +96,7 @@ typedef int erts_smp_tsd_key_t; #define erts_smp_dw_atomic_t erts_no_dw_atomic_t #define erts_smp_atomic_t erts_no_atomic_t #define erts_smp_atomic32_t erts_no_atomic32_t +#define erts_smp_atomic64_t erts_no_atomic64_t #if __GNUC__ > 2 typedef struct { } erts_smp_spinlock_t; typedef struct { } erts_smp_rwlock_t; @@ -489,6 +491,116 @@ ERTS_GLB_INLINE void erts_smp_thr_sigwait(const sigset_t *set, int *sig); #define erts_smp_atomic32_set_dirty erts_atomic32_set_dirty #define erts_smp_atomic32_read_dirty erts_atomic32_read_dirty +/* 64-bit atomics */ + +#define erts_smp_atomic64_init_nob erts_atomic64_init_nob +#define erts_smp_atomic64_set_nob erts_atomic64_set_nob +#define erts_smp_atomic64_read_nob erts_atomic64_read_nob +#define erts_smp_atomic64_inc_read_nob erts_atomic64_inc_read_nob +#define erts_smp_atomic64_dec_read_nob erts_atomic64_dec_read_nob +#define erts_smp_atomic64_inc_nob erts_atomic64_inc_nob +#define erts_smp_atomic64_dec_nob erts_atomic64_dec_nob +#define erts_smp_atomic64_add_read_nob erts_atomic64_add_read_nob +#define erts_smp_atomic64_add_nob erts_atomic64_add_nob +#define erts_smp_atomic64_read_bor_nob erts_atomic64_read_bor_nob +#define erts_smp_atomic64_read_band_nob erts_atomic64_read_band_nob +#define erts_smp_atomic64_xchg_nob erts_atomic64_xchg_nob +#define erts_smp_atomic64_cmpxchg_nob erts_atomic64_cmpxchg_nob +#define erts_smp_atomic64_read_bset_nob erts_atomic64_read_bset_nob + +#define erts_smp_atomic64_init_mb erts_atomic64_init_mb +#define erts_smp_atomic64_set_mb erts_atomic64_set_mb +#define erts_smp_atomic64_read_mb erts_atomic64_read_mb +#define erts_smp_atomic64_inc_read_mb erts_atomic64_inc_read_mb +#define erts_smp_atomic64_dec_read_mb erts_atomic64_dec_read_mb +#define erts_smp_atomic64_inc_mb erts_atomic64_inc_mb +#define erts_smp_atomic64_dec_mb erts_atomic64_dec_mb +#define erts_smp_atomic64_add_read_mb erts_atomic64_add_read_mb +#define erts_smp_atomic64_add_mb erts_atomic64_add_mb +#define erts_smp_atomic64_read_bor_mb erts_atomic64_read_bor_mb +#define erts_smp_atomic64_read_band_mb erts_atomic64_read_band_mb +#define erts_smp_atomic64_xchg_mb erts_atomic64_xchg_mb +#define erts_smp_atomic64_cmpxchg_mb erts_atomic64_cmpxchg_mb +#define erts_smp_atomic64_read_bset_mb erts_atomic64_read_bset_mb + +#define erts_smp_atomic64_init_acqb erts_atomic64_init_acqb +#define erts_smp_atomic64_set_acqb erts_atomic64_set_acqb +#define erts_smp_atomic64_read_acqb erts_atomic64_read_acqb +#define erts_smp_atomic64_inc_read_acqb erts_atomic64_inc_read_acqb +#define erts_smp_atomic64_dec_read_acqb erts_atomic64_dec_read_acqb +#define erts_smp_atomic64_inc_acqb erts_atomic64_inc_acqb +#define erts_smp_atomic64_dec_acqb erts_atomic64_dec_acqb +#define erts_smp_atomic64_add_read_acqb erts_atomic64_add_read_acqb +#define erts_smp_atomic64_add_acqb erts_atomic64_add_acqb +#define erts_smp_atomic64_read_bor_acqb erts_atomic64_read_bor_acqb +#define erts_smp_atomic64_read_band_acqb erts_atomic64_read_band_acqb +#define erts_smp_atomic64_xchg_acqb erts_atomic64_xchg_acqb +#define erts_smp_atomic64_cmpxchg_acqb erts_atomic64_cmpxchg_acqb +#define erts_smp_atomic64_read_bset_acqb erts_atomic64_read_bset_acqb + +#define erts_smp_atomic64_init_relb erts_atomic64_init_relb +#define erts_smp_atomic64_set_relb erts_atomic64_set_relb +#define erts_smp_atomic64_read_relb erts_atomic64_read_relb +#define erts_smp_atomic64_inc_read_relb erts_atomic64_inc_read_relb +#define erts_smp_atomic64_dec_read_relb erts_atomic64_dec_read_relb +#define erts_smp_atomic64_inc_relb erts_atomic64_inc_relb +#define erts_smp_atomic64_dec_relb erts_atomic64_dec_relb +#define erts_smp_atomic64_add_read_relb erts_atomic64_add_read_relb +#define erts_smp_atomic64_add_relb erts_atomic64_add_relb +#define erts_smp_atomic64_read_bor_relb erts_atomic64_read_bor_relb +#define erts_smp_atomic64_read_band_relb erts_atomic64_read_band_relb +#define erts_smp_atomic64_xchg_relb erts_atomic64_xchg_relb +#define erts_smp_atomic64_cmpxchg_relb erts_atomic64_cmpxchg_relb +#define erts_smp_atomic64_read_bset_relb erts_atomic64_read_bset_relb + +#define erts_smp_atomic64_init_ddrb erts_atomic64_init_ddrb +#define erts_smp_atomic64_set_ddrb erts_atomic64_set_ddrb +#define erts_smp_atomic64_read_ddrb erts_atomic64_read_ddrb +#define erts_smp_atomic64_inc_read_ddrb erts_atomic64_inc_read_ddrb +#define erts_smp_atomic64_dec_read_ddrb erts_atomic64_dec_read_ddrb +#define erts_smp_atomic64_inc_ddrb erts_atomic64_inc_ddrb +#define erts_smp_atomic64_dec_ddrb erts_atomic64_dec_ddrb +#define erts_smp_atomic64_add_read_ddrb erts_atomic64_add_read_ddrb +#define erts_smp_atomic64_add_ddrb erts_atomic64_add_ddrb +#define erts_smp_atomic64_read_bor_ddrb erts_atomic64_read_bor_ddrb +#define erts_smp_atomic64_read_band_ddrb erts_atomic64_read_band_ddrb +#define erts_smp_atomic64_xchg_ddrb erts_atomic64_xchg_ddrb +#define erts_smp_atomic64_cmpxchg_ddrb erts_atomic64_cmpxchg_ddrb +#define erts_smp_atomic64_read_bset_ddrb erts_atomic64_read_bset_ddrb + +#define erts_smp_atomic64_init_rb erts_atomic64_init_rb +#define erts_smp_atomic64_set_rb erts_atomic64_set_rb +#define erts_smp_atomic64_read_rb erts_atomic64_read_rb +#define erts_smp_atomic64_inc_read_rb erts_atomic64_inc_read_rb +#define erts_smp_atomic64_dec_read_rb erts_atomic64_dec_read_rb +#define erts_smp_atomic64_inc_rb erts_atomic64_inc_rb +#define erts_smp_atomic64_dec_rb erts_atomic64_dec_rb +#define erts_smp_atomic64_add_read_rb erts_atomic64_add_read_rb +#define erts_smp_atomic64_add_rb erts_atomic64_add_rb +#define erts_smp_atomic64_read_bor_rb erts_atomic64_read_bor_rb +#define erts_smp_atomic64_read_band_rb erts_atomic64_read_band_rb +#define erts_smp_atomic64_xchg_rb erts_atomic64_xchg_rb +#define erts_smp_atomic64_cmpxchg_rb erts_atomic64_cmpxchg_rb +#define erts_smp_atomic64_read_bset_rb erts_atomic64_read_bset_rb + +#define erts_smp_atomic64_init_wb erts_atomic64_init_wb +#define erts_smp_atomic64_set_wb erts_atomic64_set_wb +#define erts_smp_atomic64_read_wb erts_atomic64_read_wb +#define erts_smp_atomic64_inc_read_wb erts_atomic64_inc_read_wb +#define erts_smp_atomic64_dec_read_wb erts_atomic64_dec_read_wb +#define erts_smp_atomic64_inc_wb erts_atomic64_inc_wb +#define erts_smp_atomic64_dec_wb erts_atomic64_dec_wb +#define erts_smp_atomic64_add_read_wb erts_atomic64_add_read_wb +#define erts_smp_atomic64_add_wb erts_atomic64_add_wb +#define erts_smp_atomic64_read_bor_wb erts_atomic64_read_bor_wb +#define erts_smp_atomic64_read_band_wb erts_atomic64_read_band_wb +#define erts_smp_atomic64_xchg_wb erts_atomic64_xchg_wb +#define erts_smp_atomic64_cmpxchg_wb erts_atomic64_cmpxchg_wb +#define erts_smp_atomic64_read_bset_wb erts_atomic64_read_bset_wb + +#define erts_smp_atomic64_set_dirty erts_atomic64_set_dirty +#define erts_smp_atomic64_read_dirty erts_atomic64_read_dirty + #else /* !ERTS_SMP */ /* Double word size atomics */ @@ -751,6 +863,116 @@ ERTS_GLB_INLINE void erts_smp_thr_sigwait(const sigset_t *set, int *sig); #define erts_smp_atomic32_set_dirty erts_no_atomic32_set #define erts_smp_atomic32_read_dirty erts_no_atomic32_read +/* 64-bit atomics */ + +#define erts_smp_atomic64_init_nob erts_no_atomic64_set +#define erts_smp_atomic64_set_nob erts_no_atomic64_set +#define erts_smp_atomic64_read_nob erts_no_atomic64_read +#define erts_smp_atomic64_inc_read_nob erts_no_atomic64_inc_read +#define erts_smp_atomic64_dec_read_nob erts_no_atomic64_dec_read +#define erts_smp_atomic64_inc_nob erts_no_atomic64_inc +#define erts_smp_atomic64_dec_nob erts_no_atomic64_dec +#define erts_smp_atomic64_add_read_nob erts_no_atomic64_add_read +#define erts_smp_atomic64_add_nob erts_no_atomic64_add +#define erts_smp_atomic64_read_bor_nob erts_no_atomic64_read_bor +#define erts_smp_atomic64_read_band_nob erts_no_atomic64_read_band +#define erts_smp_atomic64_xchg_nob erts_no_atomic64_xchg +#define erts_smp_atomic64_cmpxchg_nob erts_no_atomic64_cmpxchg +#define erts_smp_atomic64_read_bset_nob erts_no_atomic64_read_bset + +#define erts_smp_atomic64_init_mb erts_no_atomic64_set +#define erts_smp_atomic64_set_mb erts_no_atomic64_set +#define erts_smp_atomic64_read_mb erts_no_atomic64_read +#define erts_smp_atomic64_inc_read_mb erts_no_atomic64_inc_read +#define erts_smp_atomic64_dec_read_mb erts_no_atomic64_dec_read +#define erts_smp_atomic64_inc_mb erts_no_atomic64_inc +#define erts_smp_atomic64_dec_mb erts_no_atomic64_dec +#define erts_smp_atomic64_add_read_mb erts_no_atomic64_add_read +#define erts_smp_atomic64_add_mb erts_no_atomic64_add +#define erts_smp_atomic64_read_bor_mb erts_no_atomic64_read_bor +#define erts_smp_atomic64_read_band_mb erts_no_atomic64_read_band +#define erts_smp_atomic64_xchg_mb erts_no_atomic64_xchg +#define erts_smp_atomic64_cmpxchg_mb erts_no_atomic64_cmpxchg +#define erts_smp_atomic64_read_bset_mb erts_no_atomic64_read_bset + +#define erts_smp_atomic64_init_acqb erts_no_atomic64_set +#define erts_smp_atomic64_set_acqb erts_no_atomic64_set +#define erts_smp_atomic64_read_acqb erts_no_atomic64_read +#define erts_smp_atomic64_inc_read_acqb erts_no_atomic64_inc_read +#define erts_smp_atomic64_dec_read_acqb erts_no_atomic64_dec_read +#define erts_smp_atomic64_inc_acqb erts_no_atomic64_inc +#define erts_smp_atomic64_dec_acqb erts_no_atomic64_dec +#define erts_smp_atomic64_add_read_acqb erts_no_atomic64_add_read +#define erts_smp_atomic64_add_acqb erts_no_atomic64_add +#define erts_smp_atomic64_read_bor_acqb erts_no_atomic64_read_bor +#define erts_smp_atomic64_read_band_acqb erts_no_atomic64_read_band +#define erts_smp_atomic64_xchg_acqb erts_no_atomic64_xchg +#define erts_smp_atomic64_cmpxchg_acqb erts_no_atomic64_cmpxchg +#define erts_smp_atomic64_read_bset_acqb erts_no_atomic64_read_bset + +#define erts_smp_atomic64_init_relb erts_no_atomic64_set +#define erts_smp_atomic64_set_relb erts_no_atomic64_set +#define erts_smp_atomic64_read_relb erts_no_atomic64_read +#define erts_smp_atomic64_inc_read_relb erts_no_atomic64_inc_read +#define erts_smp_atomic64_dec_read_relb erts_no_atomic64_dec_read +#define erts_smp_atomic64_inc_relb erts_no_atomic64_inc +#define erts_smp_atomic64_dec_relb erts_no_atomic64_dec +#define erts_smp_atomic64_add_read_relb erts_no_atomic64_add_read +#define erts_smp_atomic64_add_relb erts_no_atomic64_add +#define erts_smp_atomic64_read_bor_relb erts_no_atomic64_read_bor +#define erts_smp_atomic64_read_band_relb erts_no_atomic64_read_band +#define erts_smp_atomic64_xchg_relb erts_no_atomic64_xchg +#define erts_smp_atomic64_cmpxchg_relb erts_no_atomic64_cmpxchg +#define erts_smp_atomic64_read_bset_relb erts_no_atomic64_read_bset + +#define erts_smp_atomic64_init_ddrb erts_no_atomic64_set +#define erts_smp_atomic64_set_ddrb erts_no_atomic64_set +#define erts_smp_atomic64_read_ddrb erts_no_atomic64_read +#define erts_smp_atomic64_inc_read_ddrb erts_no_atomic64_inc_read +#define erts_smp_atomic64_dec_read_ddrb erts_no_atomic64_dec_read +#define erts_smp_atomic64_inc_ddrb erts_no_atomic64_inc +#define erts_smp_atomic64_dec_ddrb erts_no_atomic64_dec +#define erts_smp_atomic64_add_read_ddrb erts_no_atomic64_add_read +#define erts_smp_atomic64_add_ddrb erts_no_atomic64_add +#define erts_smp_atomic64_read_bor_ddrb erts_no_atomic64_read_bor +#define erts_smp_atomic64_read_band_ddrb erts_no_atomic64_read_band +#define erts_smp_atomic64_xchg_ddrb erts_no_atomic64_xchg +#define erts_smp_atomic64_cmpxchg_ddrb erts_no_atomic64_cmpxchg +#define erts_smp_atomic64_read_bset_ddrb erts_no_atomic64_read_bset + +#define erts_smp_atomic64_init_rb erts_no_atomic64_set +#define erts_smp_atomic64_set_rb erts_no_atomic64_set +#define erts_smp_atomic64_read_rb erts_no_atomic64_read +#define erts_smp_atomic64_inc_read_rb erts_no_atomic64_inc_read +#define erts_smp_atomic64_dec_read_rb erts_no_atomic64_dec_read +#define erts_smp_atomic64_inc_rb erts_no_atomic64_inc +#define erts_smp_atomic64_dec_rb erts_no_atomic64_dec +#define erts_smp_atomic64_add_read_rb erts_no_atomic64_add_read +#define erts_smp_atomic64_add_rb erts_no_atomic64_add +#define erts_smp_atomic64_read_bor_rb erts_no_atomic64_read_bor +#define erts_smp_atomic64_read_band_rb erts_no_atomic64_read_band +#define erts_smp_atomic64_xchg_rb erts_no_atomic64_xchg +#define erts_smp_atomic64_cmpxchg_rb erts_no_atomic64_cmpxchg +#define erts_smp_atomic64_read_bset_rb erts_no_atomic64_read_bset + +#define erts_smp_atomic64_init_wb erts_no_atomic64_set +#define erts_smp_atomic64_set_wb erts_no_atomic64_set +#define erts_smp_atomic64_read_wb erts_no_atomic64_read +#define erts_smp_atomic64_inc_read_wb erts_no_atomic64_inc_read +#define erts_smp_atomic64_dec_read_wb erts_no_atomic64_dec_read +#define erts_smp_atomic64_inc_wb erts_no_atomic64_inc +#define erts_smp_atomic64_dec_wb erts_no_atomic64_dec +#define erts_smp_atomic64_add_read_wb erts_no_atomic64_add_read +#define erts_smp_atomic64_add_wb erts_no_atomic64_add +#define erts_smp_atomic64_read_bor_wb erts_no_atomic64_read_bor +#define erts_smp_atomic64_read_band_wb erts_no_atomic64_read_band +#define erts_smp_atomic64_xchg_wb erts_no_atomic64_xchg +#define erts_smp_atomic64_cmpxchg_wb erts_no_atomic64_cmpxchg +#define erts_smp_atomic64_read_bset_wb erts_no_atomic64_read_bset + +#define erts_smp_atomic64_set_dirty erts_no_atomic64_set +#define erts_smp_atomic64_read_dirty erts_no_atomic64_read + #endif /* !ERTS_SMP */ #if ERTS_GLB_INLINE_INCL_FUNC_DEF diff --git a/erts/emulator/beam/erl_term.c b/erts/emulator/beam/erl_term.c index 28cbe7004f..d6fb88ea61 100644 --- a/erts/emulator/beam/erl_term.c +++ b/erts/emulator/beam/erl_term.c @@ -86,11 +86,13 @@ 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; + case (_TAG_HEADER_MAP >> _TAG_PRIMARY_SIZE): return MAP_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; + case (_TAG_HEADER_HASHMAP >> _TAG_PRIMARY_SIZE): return HASHMAP_DEF; } + break; } case TAG_PRIMARY_IMMED1: { diff --git a/erts/emulator/beam/erl_term.h b/erts/emulator/beam/erl_term.h index 37014ccf94..1625a4ec15 100644 --- a/erts/emulator/beam/erl_term.h +++ b/erts/emulator/beam/erl_term.h @@ -141,6 +141,7 @@ struct erl_node_; /* Declared in erl_node_tables.h */ #define HEAP_BINARY_SUBTAG (0x9 << _TAG_PRIMARY_SIZE) /* BINARY */ #define SUB_BINARY_SUBTAG (0xA << _TAG_PRIMARY_SIZE) /* BINARY */ /* _BINARY_XXX_MASK depends on 0xB being unused */ +#define HASHMAP_SUBTAG (0xB << _TAG_PRIMARY_SIZE) /* HASHMAP */ #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 */ @@ -162,6 +163,7 @@ struct erl_node_; /* Declared in erl_node_tables.h */ #define _TAG_HEADER_EXTERNAL_REF (TAG_PRIMARY_HEADER|EXTERNAL_REF_SUBTAG) #define _TAG_HEADER_BIN_MATCHSTATE (TAG_PRIMARY_HEADER|BIN_MATCHSTATE_SUBTAG) #define _TAG_HEADER_MAP (TAG_PRIMARY_HEADER|MAP_SUBTAG) +#define _TAG_HEADER_HASHMAP (TAG_PRIMARY_HEADER|HASHMAP_SUBTAG) #define _TAG_HEADER_MASK 0x3F @@ -296,9 +298,11 @@ _ET_DECLARE_CHECKED(Uint,atom_val,Eterm) #define atom_val(x) _ET_APPLY(atom_val,(x)) /* header (arityval or thing) access methods */ -#define _make_header(sz,tag) ((Uint)(((sz) << _HEADER_ARITY_OFFS) + (tag))) +#define _make_header(sz,tag) ((Uint)(((Uint)(sz) << _HEADER_ARITY_OFFS) + (tag))) #define is_header(x) (((x) & _TAG_PRIMARY_MASK) == TAG_PRIMARY_HEADER) -#define _unchecked_header_arity(x) ((x) >> _HEADER_ARITY_OFFS) +//#define _unchecked_header_arity(x) ((x) >> _HEADER_ARITY_OFFS) +#define _unchecked_header_arity(x) \ + (is_hashmap_header(x) ? MAP_HEADER_ARITY(x) : ((x) >> _HEADER_ARITY_OFFS)) _ET_DECLARE_CHECKED(Uint,header_arity,Eterm) #define header_arity(x) _ET_APPLY(header_arity,(x)) @@ -361,6 +365,7 @@ _ET_DECLARE_CHECKED(Uint,thing_subtag,Eterm) ((((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))) @@ -990,6 +995,33 @@ _ET_DECLARE_CHECKED(Uint32*,external_ref_data,Wterm) _ET_DECLARE_CHECKED(struct erl_node_*,external_ref_node,Eterm) #define external_ref_node(x) _ET_APPLY(external_ref_node,(x)) +/* maps */ + +#define MAP_HEADER_TAG_SZ (2) +#define MAP_HEADER_ARITY_SZ (8) +#define MAP_HEADER_VAL_SZ (16) + +#define MAP_HEADER_TAG_FLAT (0x0) +#define MAP_HEADER_TAG_HAMT_NODE_BITMAP (0x1) +#define MAP_HEADER_TAG_HAMT_HEAD_ARRAY (0x2) +#define MAP_HEADER_TAG_HAMT_HEAD_BITMAP (0x3) + +#define MAP_HEADER_TYPE(Hdr) (((Hdr) >> (_HEADER_ARITY_OFFS)) & (0x3)) +#define MAP_HEADER_ARITY(Hdr) (((Hdr) >> (_HEADER_ARITY_OFFS + MAP_HEADER_TAG_SZ)) & (0xff)) +#define MAP_HEADER_VAL(Hdr) (((Hdr) >> (_HEADER_ARITY_OFFS + MAP_HEADER_TAG_SZ + MAP_HEADER_ARITY_SZ)) & (0xffff)) + +#define make_hashmap(x) make_boxed((Eterm*)(x)) +#define make_hashmap_rel make_boxed_rel +#define is_hashmap(x) (is_boxed((x)) && is_hashmap_header(*boxed_val((x)))) +#define is_not_hashmap(x) (!is_hashmap(x)) +#define is_hashmap_rel(RTERM,BASE) is_hashmap(rterm2wterm(RTERM,BASE)) +#define is_hashmap_header(x) (((x) & (_TAG_HEADER_MASK)) == _TAG_HEADER_HASHMAP) +#define hashmap_val(x) _unchecked_boxed_val((x)) +#define hashmap_val_rel(RTERM, BASE) hashmap_val(rterm2wterm(RTERM, BASE)) + +#define is_map(x) (is_flatmap(x) || is_hashmap(x)) +#define is_map_rel(x,BASE) (is_flatmap_rel(x,BASE) || is_hashmap_rel(x,BASE)) + /* number tests */ #define is_integer(x) (is_small(x) || is_big(x)) @@ -1081,20 +1113,23 @@ _ET_DECLARE_CHECKED(Uint,y_reg_index,Uint) #define BINARY_DEF 0x0 #define LIST_DEF 0x1 #define NIL_DEF 0x2 -#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 +#define HASHMAP_DEF 0x3 +#define MAP_DEF 0x4 +#define TUPLE_DEF 0x5 +#define PID_DEF 0x6 +#define EXTERNAL_PID_DEF 0x7 +#define PORT_DEF 0x8 +#define EXTERNAL_PORT_DEF 0x9 +#define EXPORT_DEF 0xa +#define FUN_DEF 0xb +#define REF_DEF 0xc +#define EXTERNAL_REF_DEF 0xd +#define ATOM_DEF 0xe +#define FLOAT_DEF 0xf +#define BIG_DEF 0x10 +#define SMALL_DEF 0x11 + +#define FIRST_VACANT_TAG_DEF 0x12 #if ET_DEBUG extern unsigned tag_val_def_debug(Wterm, const char*, unsigned); diff --git a/erts/emulator/beam/erl_thr_progress.c b/erts/emulator/beam/erl_thr_progress.c index 545a0343d0..c2365c5cf7 100644 --- a/erts/emulator/beam/erl_thr_progress.c +++ b/erts/emulator/beam/erl_thr_progress.c @@ -115,70 +115,24 @@ #undef read_nob #define read_nob erts_thr_prgr_read_nob__ -#ifdef ARCH_64 - static ERTS_INLINE void set_mb(ERTS_THR_PRGR_ATOMIC *atmc, ErtsThrPrgrVal val) { - erts_atomic_set_mb(atmc, val); + erts_atomic64_set_mb(atmc, (erts_aint64_t) val); } static ERTS_INLINE void set_nob(ERTS_THR_PRGR_ATOMIC *atmc, ErtsThrPrgrVal val) { - erts_atomic_set_nob(atmc, val); + erts_atomic64_set_nob(atmc, (erts_aint64_t) val); } static ERTS_INLINE void init_nob(ERTS_THR_PRGR_ATOMIC *atmc, ErtsThrPrgrVal val) { - erts_atomic_init_nob(atmc, val); -} - -#else - -#undef dw_aint_to_val -#define dw_aint_to_val erts_thr_prgr_dw_aint_to_val__ - -static void -val_to_dw_aint(erts_dw_aint_t *dw_aint, ErtsThrPrgrVal val) -{ -#ifdef ETHR_SU_DW_NAINT_T__ - dw_aint->dw_sint = (ETHR_SU_DW_NAINT_T__) val; -#else - dw_aint->sint[ERTS_DW_AINT_LOW_WORD] - = (erts_aint_t) (val & 0xffffffff); - dw_aint->sint[ERTS_DW_AINT_HIGH_WORD] - = (erts_aint_t) ((val >> 32) & 0xffffffff); -#endif + erts_atomic64_init_nob(atmc, (erts_aint64_t) val); } -static ERTS_INLINE void -set_mb(ERTS_THR_PRGR_ATOMIC *atmc, ErtsThrPrgrVal val) -{ - erts_dw_aint_t dw_aint; - val_to_dw_aint(&dw_aint, val); - erts_dw_atomic_set_mb(atmc, &dw_aint); -} - -static ERTS_INLINE void -set_nob(ERTS_THR_PRGR_ATOMIC *atmc, ErtsThrPrgrVal val) -{ - erts_dw_aint_t dw_aint; - val_to_dw_aint(&dw_aint, val); - erts_dw_atomic_set_nob(atmc, &dw_aint); -} - -static ERTS_INLINE void -init_nob(ERTS_THR_PRGR_ATOMIC *atmc, ErtsThrPrgrVal val) -{ - erts_dw_aint_t dw_aint; - val_to_dw_aint(&dw_aint, val); - erts_dw_atomic_init_nob(atmc, &dw_aint); -} - -#endif - /* #define ERTS_THR_PROGRESS_STATE_DEBUG */ #ifdef ERTS_THR_PROGRESS_STATE_DEBUG @@ -1381,25 +1335,10 @@ erts_thr_progress_block(void) thr_progress_block(tmp_thr_prgr_data(NULL), 1); } -void -erts_thr_progress_fatal_error_block(SWord timeout, - ErtsThrPrgrData *tmp_tpd_bufp) +int +erts_thr_progress_fatal_error_block(ErtsThrPrgrData *tmp_tpd_bufp) { ErtsThrPrgrData *tpd = perhaps_thr_prgr_data(NULL); - erts_aint32_t bc; - SWord time_left = timeout; - SysTimeval to; - - /* - * Counting poll intervals may give us a too long timeout - * if cpu is busy. If we got tolerant time of day we use it - * to prevent this. - */ - if (!erts_disable_tolerant_timeofday) { - erts_get_timeval(&to); - to.tv_sec += timeout / 1000; - to.tv_sec += timeout % 1000; - } if (!tpd) { /* @@ -1412,9 +1351,26 @@ erts_thr_progress_fatal_error_block(SWord timeout, init_tmp_thr_prgr_data(tpd); } - bc = thr_progress_block(tpd, 0); - if (bc == 0) - return; /* Succefully blocked all managed threads */ + /* Returns number of threads that have not yes been blocked */ + return thr_progress_block(tpd, 0); +} + +void +erts_thr_progress_fatal_error_wait(SWord timeout) { + erts_aint32_t bc; + SWord time_left = timeout; + SysTimeval to; + + /* + * Counting poll intervals may give us a too long timeout + * if cpu is busy. If we got tolerant time of day we use it + * to prevent this. + */ + if (!erts_disable_tolerant_timeofday) { + erts_get_timeval(&to); + to.tv_sec += timeout / 1000; + to.tv_sec += timeout % 1000; + } while (1) { if (erts_milli_sleep(ERTS_THR_PRGR_FTL_ERR_BLCK_POLL_INTERVAL) == 0) diff --git a/erts/emulator/beam/erl_thr_progress.h b/erts/emulator/beam/erl_thr_progress.h index 5f392944c2..cf11c4e114 100644 --- a/erts/emulator/beam/erl_thr_progress.h +++ b/erts/emulator/beam/erl_thr_progress.h @@ -83,8 +83,8 @@ typedef struct { ErtsThrPrgrLeaderState leader_state; } ErtsThrPrgrData; -void erts_thr_progress_fatal_error_block(SWord timeout, - ErtsThrPrgrData *tmp_tpd_bufp); +int erts_thr_progress_fatal_error_block(ErtsThrPrgrData *tmp_tpd_bufp); +void erts_thr_progress_fatal_error_wait(SWord timeout); #endif /* ERTS_SMP */ @@ -115,11 +115,7 @@ struct ErtsThrPrgrLaterOp_ { extern erts_tsd_key_t erts_thr_prgr_data_key__; -#ifdef ARCH_64 -# define ERTS_THR_PRGR_ATOMIC erts_atomic_t -#else /* ARCH_32 */ -# define ERTS_THR_PRGR_ATOMIC erts_dw_atomic_t -#endif +#define ERTS_THR_PRGR_ATOMIC erts_atomic64_t typedef struct { void *arg; @@ -158,10 +154,6 @@ void erts_thr_progress_unmanaged_continue__(int umrefc_ix); void erts_thr_progress_dbg_print_state(void); -#ifdef ARCH_32 -#define ERTS_THR_PRGR_ATOMIC erts_dw_atomic_t -ERTS_GLB_INLINE ErtsThrPrgrVal erts_thr_prgr_dw_aint_to_val__(erts_dw_aint_t *dw_aint); -#endif ERTS_GLB_INLINE ErtsThrPrgrVal erts_thr_prgr_read_nob__(ERTS_THR_PRGR_ATOMIC *atmc); ERTS_GLB_INLINE ErtsThrPrgrVal erts_thr_prgr_read_acqb__(ERTS_THR_PRGR_ATOMIC *atmc); ERTS_GLB_INLINE ErtsThrPrgrVal erts_thr_prgr_read_mb__(ERTS_THR_PRGR_ATOMIC *atmc); @@ -184,68 +176,24 @@ ERTS_GLB_INLINE int erts_thr_progress_has_reached(ErtsThrPrgrVal val); #if ERTS_GLB_INLINE_INCL_FUNC_DEF -#ifdef ARCH_64 - -ERTS_GLB_INLINE ErtsThrPrgrVal -erts_thr_prgr_read_nob__(ERTS_THR_PRGR_ATOMIC *atmc) -{ - return (ErtsThrPrgrVal) erts_atomic_read_nob(atmc); -} - -ERTS_GLB_INLINE ErtsThrPrgrVal -erts_thr_prgr_read_acqb__(ERTS_THR_PRGR_ATOMIC *atmc) -{ - return (ErtsThrPrgrVal) erts_atomic_read_acqb(atmc); -} - -ERTS_GLB_INLINE ErtsThrPrgrVal -erts_thr_prgr_read_mb__(ERTS_THR_PRGR_ATOMIC *atmc) -{ - return (ErtsThrPrgrVal) erts_atomic_read_mb(atmc); -} - -#else /* ARCH_32 */ - -ERTS_GLB_INLINE ErtsThrPrgrVal -erts_thr_prgr_dw_aint_to_val__(erts_dw_aint_t *dw_aint) -{ -#ifdef ETHR_SU_DW_NAINT_T__ - return (ErtsThrPrgrVal) dw_aint->dw_sint; -#else - ErtsThrPrgrVal res; - res = (ErtsThrPrgrVal) ((Uint32) dw_aint->sint[ERTS_DW_AINT_HIGH_WORD]); - res <<= 32; - res |= (ErtsThrPrgrVal) ((Uint32) dw_aint->sint[ERTS_DW_AINT_LOW_WORD]); - return res; -#endif -} - ERTS_GLB_INLINE ErtsThrPrgrVal erts_thr_prgr_read_nob__(ERTS_THR_PRGR_ATOMIC *atmc) { - erts_dw_aint_t dw_aint; - erts_dw_atomic_read_nob(atmc, &dw_aint); - return erts_thr_prgr_dw_aint_to_val__(&dw_aint); + return (ErtsThrPrgrVal) erts_atomic64_read_nob(atmc); } ERTS_GLB_INLINE ErtsThrPrgrVal erts_thr_prgr_read_acqb__(ERTS_THR_PRGR_ATOMIC *atmc) { - erts_dw_aint_t dw_aint; - erts_dw_atomic_read_acqb(atmc, &dw_aint); - return erts_thr_prgr_dw_aint_to_val__(&dw_aint); + return (ErtsThrPrgrVal) erts_atomic64_read_acqb(atmc); } ERTS_GLB_INLINE ErtsThrPrgrVal erts_thr_prgr_read_mb__(ERTS_THR_PRGR_ATOMIC *atmc) { - erts_dw_aint_t dw_aint; - erts_dw_atomic_read_mb(atmc, &dw_aint); - return erts_thr_prgr_dw_aint_to_val__(&dw_aint); + return (ErtsThrPrgrVal) erts_atomic64_read_mb(atmc); } -#endif - ERTS_GLB_INLINE int erts_thr_progress_is_managed_thread(void) { diff --git a/erts/emulator/beam/erl_threads.h b/erts/emulator/beam/erl_threads.h index 80026104db..1fd800d524 100644 --- a/erts/emulator/beam/erl_threads.h +++ b/erts/emulator/beam/erl_threads.h @@ -344,6 +344,16 @@ typedef ethr_ts_event erts_tse_t; #define erts_aint32_t ethr_sint32_t #define erts_atomic32_t ethr_atomic32_t +#if defined(ARCH_32) +# define erts_atomic64_t ethr_dw_atomic_t +# define erts_aint64_t ethr_sint64_t +#elif defined(ARCH_64) +# define erts_atomic64_t ethr_atomic_t +# define erts_aint64_t ethr_sint_t +#else +# error "Not supported architecture" +#endif + #define ERTS_DW_AINT_HIGH_WORD ETHR_DW_SINT_HIGH_WORD #define ERTS_DW_AINT_LOW_WORD ETHR_DW_SINT_LOW_WORD @@ -414,10 +424,12 @@ typedef int erts_tse_t; typedef struct { SWord sint[2]; } erts_dw_aint_t; typedef SWord erts_aint_t; typedef Sint32 erts_aint32_t; +typedef Sint64 erts_aint64_t; #define erts_dw_atomic_t erts_dw_aint_t #define erts_atomic_t erts_aint_t #define erts_atomic32_t erts_aint32_t +#define erts_atomic64_t erts_aint64_t #if __GNUC__ > 2 typedef struct { } erts_spinlock_t; @@ -446,6 +458,7 @@ typedef struct { int gcc_is_buggy; } erts_rwlock_t; #define erts_no_dw_atomic_t erts_dw_aint_t #define erts_no_atomic_t erts_aint_t #define erts_no_atomic32_t erts_aint32_t +#define erts_no_atomic64_t erts_aint64_t #define ERTS_AINT_NULL ((erts_aint_t) NULL) @@ -463,6 +476,7 @@ ERTS_GLB_INLINE void erts_thr_detach(erts_tid_t tid); ERTS_GLB_INLINE void erts_thr_exit(void *res); ERTS_GLB_INLINE void erts_thr_install_exit_handler(void (*exit_handler)(void)); ERTS_GLB_INLINE erts_tid_t erts_thr_self(void); +ERTS_GLB_INLINE int erts_thr_getname(erts_tid_t tid, char *buf, size_t len); ERTS_GLB_INLINE int erts_equal_tids(erts_tid_t x, erts_tid_t y); ERTS_GLB_INLINE void erts_mtx_init_x(erts_mtx_t *mtx, char *name, Eterm extra, int enable_lcnt); @@ -570,6 +584,29 @@ ERTS_GLB_INLINE erts_aint32_t erts_no_atomic32_cmpxchg(erts_no_atomic32_t *xchgp ERTS_GLB_INLINE erts_aint32_t erts_no_atomic32_read_bset(erts_no_atomic32_t *var, erts_aint32_t mask, erts_aint32_t set); +ERTS_GLB_INLINE void erts_no_atomic64_set(erts_no_atomic64_t *var, + erts_aint64_t i); +ERTS_GLB_INLINE erts_aint64_t erts_no_atomic64_read(erts_no_atomic64_t *var); +ERTS_GLB_INLINE erts_aint64_t erts_no_atomic64_inc_read(erts_no_atomic64_t *incp); +ERTS_GLB_INLINE erts_aint64_t erts_no_atomic64_dec_read(erts_no_atomic64_t *decp); +ERTS_GLB_INLINE void erts_no_atomic64_inc(erts_no_atomic64_t *incp); +ERTS_GLB_INLINE void erts_no_atomic64_dec(erts_no_atomic64_t *decp); +ERTS_GLB_INLINE erts_aint64_t erts_no_atomic64_add_read(erts_no_atomic64_t *addp, + erts_aint64_t i); +ERTS_GLB_INLINE void erts_no_atomic64_add(erts_no_atomic64_t *addp, + erts_aint64_t i); +ERTS_GLB_INLINE erts_aint64_t erts_no_atomic64_read_bor(erts_no_atomic64_t *var, + erts_aint64_t mask); +ERTS_GLB_INLINE erts_aint64_t erts_no_atomic64_read_band(erts_no_atomic64_t *var, + erts_aint64_t mask); +ERTS_GLB_INLINE erts_aint64_t erts_no_atomic64_xchg(erts_no_atomic64_t *xchgp, + erts_aint64_t new); +ERTS_GLB_INLINE erts_aint64_t erts_no_atomic64_cmpxchg(erts_no_atomic64_t *xchgp, + erts_aint64_t new, + erts_aint64_t expected); +ERTS_GLB_INLINE erts_aint64_t erts_no_atomic64_read_bset(erts_no_atomic64_t *var, + erts_aint64_t mask, + erts_aint64_t set); ERTS_GLB_INLINE void erts_spinlock_init_x_opt(erts_spinlock_t *lock, char *name, @@ -620,11 +657,17 @@ ERTS_GLB_INLINE void erts_thr_set_main_status(int, int); ERTS_GLB_INLINE int erts_thr_get_main_status(void); ERTS_GLB_INLINE void erts_thr_yield(void); + #ifdef ETHR_HAVE_ETHR_SIG_FUNCS #define ERTS_THR_HAVE_SIG_FUNCS 1 ERTS_GLB_INLINE void erts_thr_sigmask(int how, const sigset_t *set, sigset_t *oset); ERTS_GLB_INLINE void erts_thr_sigwait(const sigset_t *set, int *sig); + +#ifdef USE_THREADS +ERTS_GLB_INLINE void erts_thr_kill(erts_tid_t tid, int sig); +#endif + #endif /* #ifdef HAVE_ETHR_SIG_FUNCS */ #ifdef USE_THREADS @@ -1200,6 +1243,441 @@ erts_atomic32_read_dirty(erts_atomic32_t *var) #endif +/* 64-bit atomics */ + +#if defined(ARCH_64) + +#define erts_atomic64_init_nob ethr_atomic_init +#define erts_atomic64_set_nob ethr_atomic_set +#define erts_atomic64_read_nob ethr_atomic_read +#define erts_atomic64_inc_read_nob ethr_atomic_inc_read +#define erts_atomic64_dec_read_nob ethr_atomic_dec_read +#define erts_atomic64_inc_nob ethr_atomic_inc +#define erts_atomic64_dec_nob ethr_atomic_dec +#define erts_atomic64_add_read_nob ethr_atomic_add_read +#define erts_atomic64_add_nob ethr_atomic_add +#define erts_atomic64_read_bor_nob ethr_atomic_read_bor +#define erts_atomic64_read_band_nob ethr_atomic_read_band +#define erts_atomic64_xchg_nob ethr_atomic_xchg +#define erts_atomic64_cmpxchg_nob ethr_atomic_cmpxchg +#define erts_atomic64_read_bset_nob erts_atomic_read_bset_nob + +#define erts_atomic64_init_mb ethr_atomic_init_mb +#define erts_atomic64_set_mb ethr_atomic_set_mb +#define erts_atomic64_read_mb ethr_atomic_read_mb +#define erts_atomic64_inc_read_mb ethr_atomic_inc_read_mb +#define erts_atomic64_dec_read_mb ethr_atomic_dec_read_mb +#define erts_atomic64_inc_mb ethr_atomic_inc_mb +#define erts_atomic64_dec_mb ethr_atomic_dec_mb +#define erts_atomic64_add_read_mb ethr_atomic_add_read_mb +#define erts_atomic64_add_mb ethr_atomic_add_mb +#define erts_atomic64_read_bor_mb ethr_atomic_read_bor_mb +#define erts_atomic64_read_band_mb ethr_atomic_read_band_mb +#define erts_atomic64_xchg_mb ethr_atomic_xchg_mb +#define erts_atomic64_cmpxchg_mb ethr_atomic_cmpxchg_mb +#define erts_atomic64_read_bset_mb erts_atomic_read_bset_mb + +#define erts_atomic64_init_acqb ethr_atomic_init_acqb +#define erts_atomic64_set_acqb ethr_atomic_set_acqb +#define erts_atomic64_read_acqb ethr_atomic_read_acqb +#define erts_atomic64_inc_read_acqb ethr_atomic_inc_read_acqb +#define erts_atomic64_dec_read_acqb ethr_atomic_dec_read_acqb +#define erts_atomic64_inc_acqb ethr_atomic_inc_acqb +#define erts_atomic64_dec_acqb ethr_atomic_dec_acqb +#define erts_atomic64_add_read_acqb ethr_atomic_add_read_acqb +#define erts_atomic64_add_acqb ethr_atomic_add_acqb +#define erts_atomic64_read_bor_acqb ethr_atomic_read_bor_acqb +#define erts_atomic64_read_band_acqb ethr_atomic_read_band_acqb +#define erts_atomic64_xchg_acqb ethr_atomic_xchg_acqb +#define erts_atomic64_cmpxchg_acqb ethr_atomic_cmpxchg_acqb +#define erts_atomic64_read_bset_acqb erts_atomic_read_bset_acqb + +#define erts_atomic64_init_relb ethr_atomic_init_relb +#define erts_atomic64_set_relb ethr_atomic_set_relb +#define erts_atomic64_read_relb ethr_atomic_read_relb +#define erts_atomic64_inc_read_relb ethr_atomic_inc_read_relb +#define erts_atomic64_dec_read_relb ethr_atomic_dec_read_relb +#define erts_atomic64_inc_relb ethr_atomic_inc_relb +#define erts_atomic64_dec_relb ethr_atomic_dec_relb +#define erts_atomic64_add_read_relb ethr_atomic_add_read_relb +#define erts_atomic64_add_relb ethr_atomic_add_relb +#define erts_atomic64_read_bor_relb ethr_atomic_read_bor_relb +#define erts_atomic64_read_band_relb ethr_atomic_read_band_relb +#define erts_atomic64_xchg_relb ethr_atomic_xchg_relb +#define erts_atomic64_cmpxchg_relb ethr_atomic_cmpxchg_relb +#define erts_atomic64_read_bset_relb erts_atomic_read_bset_relb + +#define erts_atomic64_init_ddrb ethr_atomic_init_ddrb +#define erts_atomic64_set_ddrb ethr_atomic_set_ddrb +#define erts_atomic64_read_ddrb ethr_atomic_read_ddrb +#define erts_atomic64_inc_read_ddrb ethr_atomic_inc_read_ddrb +#define erts_atomic64_dec_read_ddrb ethr_atomic_dec_read_ddrb +#define erts_atomic64_inc_ddrb ethr_atomic_inc_ddrb +#define erts_atomic64_dec_ddrb ethr_atomic_dec_ddrb +#define erts_atomic64_add_read_ddrb ethr_atomic_add_read_ddrb +#define erts_atomic64_add_ddrb ethr_atomic_add_ddrb +#define erts_atomic64_read_bor_ddrb ethr_atomic_read_bor_ddrb +#define erts_atomic64_read_band_ddrb ethr_atomic_read_band_ddrb +#define erts_atomic64_xchg_ddrb ethr_atomic_xchg_ddrb +#define erts_atomic64_cmpxchg_ddrb ethr_atomic_cmpxchg_ddrb +#define erts_atomic64_read_bset_ddrb erts_atomic_read_bset_ddrb + +#define erts_atomic64_init_rb ethr_atomic_init_rb +#define erts_atomic64_set_rb ethr_atomic_set_rb +#define erts_atomic64_read_rb ethr_atomic_read_rb +#define erts_atomic64_inc_read_rb ethr_atomic_inc_read_rb +#define erts_atomic64_dec_read_rb ethr_atomic_dec_read_rb +#define erts_atomic64_inc_rb ethr_atomic_inc_rb +#define erts_atomic64_dec_rb ethr_atomic_dec_rb +#define erts_atomic64_add_read_rb ethr_atomic_add_read_rb +#define erts_atomic64_add_rb ethr_atomic_add_rb +#define erts_atomic64_read_bor_rb ethr_atomic_read_bor_rb +#define erts_atomic64_read_band_rb ethr_atomic_read_band_rb +#define erts_atomic64_xchg_rb ethr_atomic_xchg_rb +#define erts_atomic64_cmpxchg_rb ethr_atomic_cmpxchg_rb +#define erts_atomic64_read_bset_rb erts_atomic_read_bset_rb + +#define erts_atomic64_init_wb ethr_atomic_init_wb +#define erts_atomic64_set_wb ethr_atomic_set_wb +#define erts_atomic64_read_wb ethr_atomic_read_wb +#define erts_atomic64_inc_read_wb ethr_atomic_inc_read_wb +#define erts_atomic64_dec_read_wb ethr_atomic_dec_read_wb +#define erts_atomic64_inc_wb ethr_atomic_inc_wb +#define erts_atomic64_dec_wb ethr_atomic_dec_wb +#define erts_atomic64_add_read_wb ethr_atomic_add_read_wb +#define erts_atomic64_add_wb ethr_atomic_add_wb +#define erts_atomic64_read_bor_wb ethr_atomic_read_bor_wb +#define erts_atomic64_read_band_wb ethr_atomic_read_band_wb +#define erts_atomic64_xchg_wb ethr_atomic_xchg_wb +#define erts_atomic64_cmpxchg_wb ethr_atomic_cmpxchg_wb +#define erts_atomic64_read_bset_wb erts_atomic_read_bset_wb + +#define erts_atomic64_set_dirty erts_atomic_set_dirty +#define erts_atomic64_read_dirty erts_atomic_read_dirty + +#elif defined(ARCH_32) + +#undef ERTS_ATOMIC64_OPS_DECL__ + +#define ERTS_ATOMIC64_OPS_DECL__(BARRIER) \ +ERTS_GLB_INLINE void \ +erts_atomic64_init_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val); \ +ERTS_GLB_INLINE void \ +erts_atomic64_set_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val); \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_read_ ## BARRIER(erts_atomic64_t *var); \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_inc_read_ ## BARRIER(erts_atomic64_t *var); \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_dec_read_ ## BARRIER(erts_atomic64_t *var); \ +ERTS_GLB_INLINE void \ +erts_atomic64_inc_ ## BARRIER(erts_atomic64_t *var); \ +ERTS_GLB_INLINE void \ +erts_atomic64_dec_ ## BARRIER(erts_atomic64_t *var); \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_add_read_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val); \ +ERTS_GLB_INLINE void \ +erts_atomic64_add_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val); \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_read_bor_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val); \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_read_band_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val); \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_xchg_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val); \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_cmpxchg_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t new, \ + erts_aint64_t exp); \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_read_bset_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t mask, \ + erts_aint64_t set) + +ERTS_ATOMIC64_OPS_DECL__(nob); +ERTS_ATOMIC64_OPS_DECL__(mb); +ERTS_ATOMIC64_OPS_DECL__(acqb); +ERTS_ATOMIC64_OPS_DECL__(relb); +ERTS_ATOMIC64_OPS_DECL__(ddrb); +ERTS_ATOMIC64_OPS_DECL__(rb); +ERTS_ATOMIC64_OPS_DECL__(wb); + +#undef ERTS_ATOMIC64_OPS_DECL__ + +ERTS_GLB_INLINE void +erts_atomic64_set_dirty(erts_atomic64_t *var, erts_aint64_t val); +ERTS_GLB_INLINE erts_aint64_t +erts_atomic64_read_dirty(erts_atomic64_t *var); + +#if ERTS_GLB_INLINE_INCL_FUNC_DEF + +/* + * The ethr_dw_atomic_*_nob() functions below + * are here to make it possible for the + * ERTS_ATOMIC64_OPS_IMPL__() to map erts + * barriers to ethread barriers... + */ +static ERTS_INLINE void +ethr_dw_atomic_init_nob(ethr_dw_atomic_t *var, + ethr_dw_sint_t *val) +{ + ethr_dw_atomic_init(var, val); +} + +static ERTS_INLINE void +ethr_dw_atomic_set_nob(ethr_dw_atomic_t *var, + ethr_dw_sint_t *val) +{ + ethr_dw_atomic_set(var, val); +} + +static ERTS_INLINE void +ethr_dw_atomic_read_nob(ethr_dw_atomic_t *var, + ethr_dw_sint_t *val) +{ + ethr_dw_atomic_read(var, val); +} + +static ERTS_INLINE int +ethr_dw_atomic_cmpxchg_nob(ethr_dw_atomic_t *var, + ethr_dw_sint_t *new, + ethr_dw_sint_t *xchg) +{ + return ethr_dw_atomic_cmpxchg(var, new, xchg); +} + +#undef ERTS_ATOMIC64_OPS_IMPL__ +#undef ERTS_ATOMIC64_DW_CMPXCHG_IMPL__ +#undef ERTS_DW_SINT_TO_AINT64__ +#undef ERTS_AINT64_TO_DW_SINT__ + +#ifdef ETHR_SU_DW_NAINT_T__ +#define ERTS_DW_SINT_TO_AINT64__(DW) \ + ((erts_aint64_t) DW.dw_sint) +#define ERTS_AINT64_TO_DW_SINT__(DW, AINT64) \ + (DW.dw_sint = (ETHR_SU_DW_NAINT_T__) AINT64) +#else /* !ETHR_SU_DW_NAINT_T__ */ +#define ERTS_DW_SINT_TO_AINT64__(DW) \ + ((((erts_aint64_t) DW.sint[ETHR_DW_SINT_HIGH_WORD]) << 32) \ + | (((erts_aint64_t) DW.sint[ETHR_DW_SINT_LOW_WORD]) \ + & ((erts_aint64_t) 0xffffffff))) +#define ERTS_AINT64_TO_DW_SINT__(DW, AINT64) \ + do { \ + DW.sint[ETHR_DW_SINT_LOW_WORD] = \ + (ethr_sint_t) (AINT64 & 0xffffffff); \ + DW.sint[ETHR_DW_SINT_HIGH_WORD] = \ + (ethr_sint_t) ((AINT64 >> 32) & 0xffffffff); \ + } while (0) +#endif /* !ETHR_SU_DW_NAINT_T__ */ + +#define ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(CmpXchgOp, \ + AVarP, XchgVar, NewVar, \ + ModificationCode) \ +do { \ + ethr_dw_sint_t dw_xchg__, dw_new__; \ + ethr_dw_atomic_read(AVarP, &dw_xchg__); \ + do { \ + XchgVar = ERTS_DW_SINT_TO_AINT64__(dw_xchg__); \ + { \ + ModificationCode; \ + } \ + ERTS_AINT64_TO_DW_SINT__(dw_new__, NewVar); \ + } while (!CmpXchgOp((AVarP), &dw_new__, &dw_xchg__)); \ +} while (0) + +#define ERTS_ATOMIC64_OPS_IMPL__(BARRIER) \ + \ +ERTS_GLB_INLINE void \ +erts_atomic64_init_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val) \ +{ \ + ethr_dw_sint_t dw; \ + ERTS_AINT64_TO_DW_SINT__(dw, val); \ + ethr_dw_atomic_init_ ## BARRIER(var, &dw); \ +} \ + \ +ERTS_GLB_INLINE void \ +erts_atomic64_set_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val) \ +{ \ + ethr_dw_sint_t dw; \ + ERTS_AINT64_TO_DW_SINT__(dw, val); \ + ethr_dw_atomic_set_ ## BARRIER(var, &dw); \ +} \ + \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_read_ ## BARRIER(erts_atomic64_t *var) \ +{ \ + ethr_dw_sint_t dw; \ + ethr_dw_atomic_read_ ## BARRIER(var, &dw); \ + return ERTS_DW_SINT_TO_AINT64__(dw); \ +} \ + \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_inc_read_ ## BARRIER(erts_atomic64_t *var) \ +{ \ + erts_aint64_t xchg, new; \ + ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(ethr_dw_atomic_cmpxchg_ ## BARRIER, \ + var, xchg, new, \ + new = xchg + 1); \ + return new; \ +} \ + \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_dec_read_ ## BARRIER(erts_atomic64_t *var) \ +{ \ + erts_aint64_t xchg, new; \ + ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(ethr_dw_atomic_cmpxchg_ ## BARRIER, \ + var, xchg, new, \ + new = xchg - 1); \ + return new; \ +} \ + \ +ERTS_GLB_INLINE void \ +erts_atomic64_inc_ ## BARRIER(erts_atomic64_t *var) \ +{ \ + erts_aint64_t xchg, new; \ + ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(ethr_dw_atomic_cmpxchg_ ## BARRIER, \ + var, xchg, new, \ + new = xchg + 1); \ +} \ + \ +ERTS_GLB_INLINE void \ +erts_atomic64_dec_ ## BARRIER(erts_atomic64_t *var) \ +{ \ + erts_aint64_t xchg, new; \ + ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(ethr_dw_atomic_cmpxchg_ ## BARRIER, \ + var, xchg, new, \ + new = xchg - 1); \ +} \ + \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_add_read_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val) \ +{ \ + erts_aint64_t xchg, new; \ + ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(ethr_dw_atomic_cmpxchg_ ## BARRIER, \ + var, xchg, new, \ + new = xchg + val); \ + return new; \ +} \ + \ +ERTS_GLB_INLINE void \ +erts_atomic64_add_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val) \ +{ \ + erts_aint64_t xchg, new; \ + ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(ethr_dw_atomic_cmpxchg_ ## BARRIER, \ + var, xchg, new, \ + new = xchg + val); \ +} \ + \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_read_bor_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val) \ +{ \ + erts_aint64_t xchg, new; \ + ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(ethr_dw_atomic_cmpxchg_ ## BARRIER, \ + var, xchg, new, \ + new = xchg | val); \ + return xchg; \ +} \ + \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_read_band_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val) \ +{ \ + erts_aint64_t xchg, new; \ + ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(ethr_dw_atomic_cmpxchg_ ## BARRIER, \ + var, xchg, new, \ + new = xchg & val); \ + return xchg; \ +} \ + \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_xchg_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val) \ +{ \ + erts_aint64_t xchg, new; \ + ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(ethr_dw_atomic_cmpxchg_ ## BARRIER, \ + var, xchg, new, \ + new = val); \ + return xchg; \ +} \ + \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_cmpxchg_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t new, \ + erts_aint64_t exp) \ +{ \ + ethr_dw_sint_t dw_xchg, dw_new; \ + ERTS_AINT64_TO_DW_SINT__(dw_xchg, exp); \ + ERTS_AINT64_TO_DW_SINT__(dw_new, new); \ + if (ethr_dw_atomic_cmpxchg_ ## BARRIER(var, &dw_new, &dw_xchg)) \ + return exp; \ + return ERTS_DW_SINT_TO_AINT64__(dw_xchg); \ +} \ + \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_read_bset_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t mask, \ + erts_aint64_t set) \ +{ \ + erts_aint64_t xchg, new; \ + ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(ethr_dw_atomic_cmpxchg_ ## BARRIER, \ + var, xchg, new, \ + { \ + new = xchg & ~mask; \ + new |= mask & set; \ + }); \ + return xchg; \ +} + +ERTS_ATOMIC64_OPS_IMPL__(nob) +ERTS_ATOMIC64_OPS_IMPL__(mb) +ERTS_ATOMIC64_OPS_IMPL__(acqb) +ERTS_ATOMIC64_OPS_IMPL__(relb) +ERTS_ATOMIC64_OPS_IMPL__(ddrb) +ERTS_ATOMIC64_OPS_IMPL__(rb) +ERTS_ATOMIC64_OPS_IMPL__(wb) + +#undef ERTS_ATOMIC64_OPS_IMPL__ +#undef ERTS_ATOMIC64_DW_CMPXCHG_IMPL__ + +ERTS_GLB_INLINE void +erts_atomic64_set_dirty(erts_atomic64_t *var, erts_aint64_t val) +{ + ethr_sint_t *sint = ethr_dw_atomic_addr(var); + ethr_dw_sint_t dw; + ERTS_AINT64_TO_DW_SINT__(dw, val); + sint[0] = dw.sint[0]; + sint[1] = dw.sint[1]; +} + +ERTS_GLB_INLINE erts_aint64_t +erts_atomic64_read_dirty(erts_atomic64_t *var) +{ + ethr_sint_t *sint; + ethr_dw_sint_t dw; + sint = ethr_dw_atomic_addr(var); + dw.sint[0] = sint[0]; + dw.sint[1] = sint[1]; + return ERTS_DW_SINT_TO_AINT64__(dw); +} + +#undef ERTS_DW_SINT_TO_AINT64__ +#undef ERTS_AINT64_TO_DW_SINT__ + +#endif /* ERTS_GLB_INLINE_INCL_FUNC_DEF */ + +#endif /* ARCH_32 */ + #else /* !USE_THREADS */ /* Double word size atomics */ @@ -1462,6 +1940,116 @@ erts_atomic32_read_dirty(erts_atomic32_t *var) #define erts_atomic32_set_dirty erts_no_atomic32_set #define erts_atomic32_read_dirty erts_no_atomic32_read +/* 64-bit atomics */ + +#define erts_atomic64_init_nob erts_no_atomic64_set +#define erts_atomic64_set_nob erts_no_atomic64_set +#define erts_atomic64_read_nob erts_no_atomic64_read +#define erts_atomic64_inc_read_nob erts_no_atomic64_inc_read +#define erts_atomic64_dec_read_nob erts_no_atomic64_dec_read +#define erts_atomic64_inc_nob erts_no_atomic64_inc +#define erts_atomic64_dec_nob erts_no_atomic64_dec +#define erts_atomic64_add_read_nob erts_no_atomic64_add_read +#define erts_atomic64_add_nob erts_no_atomic64_add +#define erts_atomic64_read_bor_nob erts_no_atomic64_read_bor +#define erts_atomic64_read_band_nob erts_no_atomic64_read_band +#define erts_atomic64_xchg_nob erts_no_atomic64_xchg +#define erts_atomic64_cmpxchg_nob erts_no_atomic64_cmpxchg +#define erts_atomic64_read_bset_nob erts_no_atomic64_read_bset + +#define erts_atomic64_init_mb erts_no_atomic64_set +#define erts_atomic64_set_mb erts_no_atomic64_set +#define erts_atomic64_read_mb erts_no_atomic64_read +#define erts_atomic64_inc_read_mb erts_no_atomic64_inc_read +#define erts_atomic64_dec_read_mb erts_no_atomic64_dec_read +#define erts_atomic64_inc_mb erts_no_atomic64_inc +#define erts_atomic64_dec_mb erts_no_atomic64_dec +#define erts_atomic64_add_read_mb erts_no_atomic64_add_read +#define erts_atomic64_add_mb erts_no_atomic64_add +#define erts_atomic64_read_bor_mb erts_no_atomic64_read_bor +#define erts_atomic64_read_band_mb erts_no_atomic64_read_band +#define erts_atomic64_xchg_mb erts_no_atomic64_xchg +#define erts_atomic64_cmpxchg_mb erts_no_atomic64_cmpxchg +#define erts_atomic64_read_bset_mb erts_no_atomic64_read_bset + +#define erts_atomic64_init_acqb erts_no_atomic64_set +#define erts_atomic64_set_acqb erts_no_atomic64_set +#define erts_atomic64_read_acqb erts_no_atomic64_read +#define erts_atomic64_inc_read_acqb erts_no_atomic64_inc_read +#define erts_atomic64_dec_read_acqb erts_no_atomic64_dec_read +#define erts_atomic64_inc_acqb erts_no_atomic64_inc +#define erts_atomic64_dec_acqb erts_no_atomic64_dec +#define erts_atomic64_add_read_acqb erts_no_atomic64_add_read +#define erts_atomic64_add_acqb erts_no_atomic64_add +#define erts_atomic64_read_bor_acqb erts_no_atomic64_read_bor +#define erts_atomic64_read_band_acqb erts_no_atomic64_read_band +#define erts_atomic64_xchg_acqb erts_no_atomic64_xchg +#define erts_atomic64_cmpxchg_acqb erts_no_atomic64_cmpxchg +#define erts_atomic64_read_bset_acqb erts_no_atomic64_read_bset + +#define erts_atomic64_init_relb erts_no_atomic64_set +#define erts_atomic64_set_relb erts_no_atomic64_set +#define erts_atomic64_read_relb erts_no_atomic64_read +#define erts_atomic64_inc_read_relb erts_no_atomic64_inc_read +#define erts_atomic64_dec_read_relb erts_no_atomic64_dec_read +#define erts_atomic64_inc_relb erts_no_atomic64_inc +#define erts_atomic64_dec_relb erts_no_atomic64_dec +#define erts_atomic64_add_read_relb erts_no_atomic64_add_read +#define erts_atomic64_add_relb erts_no_atomic64_add +#define erts_atomic64_read_bor_relb erts_no_atomic64_read_bor +#define erts_atomic64_read_band_relb erts_no_atomic64_read_band +#define erts_atomic64_xchg_relb erts_no_atomic64_xchg +#define erts_atomic64_cmpxchg_relb erts_no_atomic64_cmpxchg +#define erts_atomic64_read_bset_relb erts_no_atomic64_read_bset + +#define erts_atomic64_init_ddrb erts_no_atomic64_set +#define erts_atomic64_set_ddrb erts_no_atomic64_set +#define erts_atomic64_read_ddrb erts_no_atomic64_read +#define erts_atomic64_inc_read_ddrb erts_no_atomic64_inc_read +#define erts_atomic64_dec_read_ddrb erts_no_atomic64_dec_read +#define erts_atomic64_inc_ddrb erts_no_atomic64_inc +#define erts_atomic64_dec_ddrb erts_no_atomic64_dec +#define erts_atomic64_add_read_ddrb erts_no_atomic64_add_read +#define erts_atomic64_add_ddrb erts_no_atomic64_add +#define erts_atomic64_read_bor_ddrb erts_no_atomic64_read_bor +#define erts_atomic64_read_band_ddrb erts_no_atomic64_read_band +#define erts_atomic64_xchg_ddrb erts_no_atomic64_xchg +#define erts_atomic64_cmpxchg_ddrb erts_no_atomic64_cmpxchg +#define erts_atomic64_read_bset_ddrb erts_no_atomic64_read_bset + +#define erts_atomic64_init_rb erts_no_atomic64_set +#define erts_atomic64_set_rb erts_no_atomic64_set +#define erts_atomic64_read_rb erts_no_atomic64_read +#define erts_atomic64_inc_read_rb erts_no_atomic64_inc_read +#define erts_atomic64_dec_read_rb erts_no_atomic64_dec_read +#define erts_atomic64_inc_rb erts_no_atomic64_inc +#define erts_atomic64_dec_rb erts_no_atomic64_dec +#define erts_atomic64_add_read_rb erts_no_atomic64_add_read +#define erts_atomic64_add_rb erts_no_atomic64_add +#define erts_atomic64_read_bor_rb erts_no_atomic64_read_bor +#define erts_atomic64_read_band_rb erts_no_atomic64_read_band +#define erts_atomic64_xchg_rb erts_no_atomic64_xchg +#define erts_atomic64_cmpxchg_rb erts_no_atomic64_cmpxchg +#define erts_atomic64_read_bset_rb erts_no_atomic64_read_bset + +#define erts_atomic64_init_wb erts_no_atomic64_set +#define erts_atomic64_set_wb erts_no_atomic64_set +#define erts_atomic64_read_wb erts_no_atomic64_read +#define erts_atomic64_inc_read_wb erts_no_atomic64_inc_read +#define erts_atomic64_dec_read_wb erts_no_atomic64_dec_read +#define erts_atomic64_inc_wb erts_no_atomic64_inc +#define erts_atomic64_dec_wb erts_no_atomic64_dec +#define erts_atomic64_add_read_wb erts_no_atomic64_add_read +#define erts_atomic64_add_wb erts_no_atomic64_add +#define erts_atomic64_read_bor_wb erts_no_atomic64_read_bor +#define erts_atomic64_read_band_wb erts_no_atomic64_read_band +#define erts_atomic64_xchg_wb erts_no_atomic64_xchg +#define erts_atomic64_cmpxchg_wb erts_no_atomic64_cmpxchg +#define erts_atomic64_read_bset_wb erts_no_atomic64_read_bset + +#define erts_atomic64_set_dirty erts_no_atomic64_set +#define erts_atomic64_read_dirty erts_no_atomic64_read + #endif /* !USE_THREADS */ #if ERTS_GLB_INLINE_INCL_FUNC_DEF @@ -1548,6 +2136,16 @@ erts_thr_self(void) #endif } +ERTS_GLB_INLINE int +erts_thr_getname(erts_tid_t tid, char *buf, size_t len) +{ +#ifdef USE_THREADS + return ethr_getname(tid, buf, len); +#else + return -1; +#endif +} + ERTS_GLB_INLINE int erts_equal_tids(erts_tid_t x, erts_tid_t y) @@ -2383,6 +2981,104 @@ erts_no_atomic32_read_bset(erts_no_atomic32_t *var, return old; } +/* atomic64 */ + +ERTS_GLB_INLINE void +erts_no_atomic64_set(erts_no_atomic64_t *var, erts_aint64_t i) +{ + *var = i; +} + +ERTS_GLB_INLINE erts_aint64_t +erts_no_atomic64_read(erts_no_atomic64_t *var) +{ + return *var; +} + +ERTS_GLB_INLINE erts_aint64_t +erts_no_atomic64_inc_read(erts_no_atomic64_t *incp) +{ + return ++(*incp); +} + +ERTS_GLB_INLINE erts_aint64_t +erts_no_atomic64_dec_read(erts_no_atomic64_t *decp) +{ + return --(*decp); +} + +ERTS_GLB_INLINE void +erts_no_atomic64_inc(erts_no_atomic64_t *incp) +{ + ++(*incp); +} + +ERTS_GLB_INLINE void +erts_no_atomic64_dec(erts_no_atomic64_t *decp) +{ + --(*decp); +} + +ERTS_GLB_INLINE erts_aint64_t +erts_no_atomic64_add_read(erts_no_atomic64_t *addp, erts_aint64_t i) +{ + return *addp += i; +} + +ERTS_GLB_INLINE void +erts_no_atomic64_add(erts_no_atomic64_t *addp, erts_aint64_t i) +{ + *addp += i; +} + +ERTS_GLB_INLINE erts_aint64_t +erts_no_atomic64_read_bor(erts_no_atomic64_t *var, erts_aint64_t mask) +{ + erts_aint64_t old; + old = *var; + *var |= mask; + return old; +} + +ERTS_GLB_INLINE erts_aint64_t +erts_no_atomic64_read_band(erts_no_atomic64_t *var, erts_aint64_t mask) +{ + erts_aint64_t old; + old = *var; + *var &= mask; + return old; +} + +ERTS_GLB_INLINE erts_aint64_t +erts_no_atomic64_xchg(erts_no_atomic64_t *xchgp, erts_aint64_t new) +{ + erts_aint64_t old = *xchgp; + *xchgp = new; + return old; +} + +ERTS_GLB_INLINE erts_aint64_t +erts_no_atomic64_cmpxchg(erts_no_atomic64_t *xchgp, + erts_aint64_t new, + erts_aint64_t expected) +{ + erts_aint64_t old = *xchgp; + if (old == expected) + *xchgp = new; + return old; +} + +ERTS_GLB_INLINE erts_aint64_t +erts_no_atomic64_read_bset(erts_no_atomic64_t *var, + erts_aint64_t mask, + erts_aint64_t set) +{ + erts_aint64_t old = *var; + *var &= ~mask; + *var |= (mask & set); + return old; +} + /* spinlock */ ERTS_GLB_INLINE void @@ -2838,6 +3534,15 @@ ERTS_GLB_INLINE void erts_thr_yield(void) #ifdef ETHR_HAVE_ETHR_SIG_FUNCS ERTS_GLB_INLINE void +erts_thr_kill(erts_tid_t tid, int sig) { +#ifdef USE_THREADS + int res = ethr_kill((ethr_tid)tid, sig); + if (res) + erts_thr_fatal_error(res, "killing thread"); +#endif +} + +ERTS_GLB_INLINE void erts_thr_sigmask(int how, const sigset_t *set, sigset_t *oset) { #ifdef USE_THREADS diff --git a/erts/emulator/beam/erl_time.h b/erts/emulator/beam/erl_time.h index 4bbdcaa3e3..7ed1a395ad 100644 --- a/erts/emulator/beam/erl_time.h +++ b/erts/emulator/beam/erl_time.h @@ -107,7 +107,7 @@ ERTS_GLB_INLINE void erts_do_time_add(erts_short_time_t elapsed) /* time_sup */ -#if (defined(HAVE_GETHRVTIME) || defined(HAVE_CLOCK_GETTIME)) +#if (defined(HAVE_GETHRVTIME) || defined(HAVE_CLOCK_GETTIME_CPU_TIME)) # ifndef HAVE_ERTS_NOW_CPU # define HAVE_ERTS_NOW_CPU # ifdef HAVE_GETHRVTIME diff --git a/erts/emulator/beam/erl_trace.c b/erts/emulator/beam/erl_trace.c index ea5c850a30..2f9969b0e7 100644 --- a/erts/emulator/beam/erl_trace.c +++ b/erts/emulator/beam/erl_trace.c @@ -2225,7 +2225,7 @@ trace_gc(Process *p, Eterm what) Eterm* limit; #endif - ASSERT(sizeof(values)/sizeof(*values) == sizeof(tags)/sizeof(Eterm)); + ERTS_CT_ASSERT(sizeof(values)/sizeof(*values) == sizeof(tags)/sizeof(Eterm)); UseTmpHeap(LOCAL_HEAP_SIZE,p); @@ -3492,16 +3492,13 @@ init_sys_msg_dispatcher(void) thr_opts.coreNo = 0; #endif thr_opts.detached = 1; + thr_opts.name = "sys_msg_dispatcher"; init_smq_element_alloc(); sys_message_queue = NULL; sys_message_queue_end = NULL; erts_smp_cnd_init(&smq_cnd); erts_smp_mtx_init(&smq_mtx, "sys_msg_q"); -#ifdef ETHR_HAVE_THREAD_NAMES - thr_opts.name = "sys_msg_dispatcher"; -#endif - erts_smp_thr_create(&sys_msg_dispatcher_tid, sys_msg_dispatcher_func, NULL, diff --git a/erts/emulator/beam/erl_utils.h b/erts/emulator/beam/erl_utils.h index 0807649ea1..7cb8972e29 100644 --- a/erts/emulator/beam/erl_utils.h +++ b/erts/emulator/beam/erl_utils.h @@ -32,11 +32,7 @@ typedef struct { #endif union { Uint64 not_atomic; -#ifdef ARCH_64 - erts_atomic_t atomic; -#else - erts_dw_atomic_t atomic; -#endif + erts_atomic64_t atomic; } counter; } erts_interval_t; @@ -50,9 +46,6 @@ Uint64 erts_ensure_later_interval_nob(erts_interval_t *, Uint64); Uint64 erts_ensure_later_interval_acqb(erts_interval_t *, Uint64); Uint64 erts_smp_ensure_later_interval_nob(erts_interval_t *, Uint64); Uint64 erts_smp_ensure_later_interval_acqb(erts_interval_t *, Uint64); -#ifdef ARCH_32 -ERTS_GLB_INLINE Uint64 erts_interval_dw_aint_to_val__(erts_dw_aint_t *); -#endif ERTS_GLB_INLINE Uint64 erts_current_interval_nob__(erts_interval_t *); ERTS_GLB_INLINE Uint64 erts_current_interval_acqb__(erts_interval_t *); ERTS_GLB_INLINE Uint64 erts_current_interval_nob(erts_interval_t *); @@ -62,46 +55,16 @@ ERTS_GLB_INLINE Uint64 erts_smp_current_interval_acqb(erts_interval_t *); #if ERTS_GLB_INLINE_INCL_FUNC_DEF -#ifdef ARCH_32 - -ERTS_GLB_INLINE Uint64 -erts_interval_dw_aint_to_val__(erts_dw_aint_t *dw) -{ -#ifdef ETHR_SU_DW_NAINT_T__ - return (Uint64) dw->dw_sint; -#else - Uint64 res; - res = (Uint64) ((Uint32) dw->sint[ERTS_DW_AINT_HIGH_WORD]); - res <<= 32; - res |= (Uint64) ((Uint32) dw->sint[ERTS_DW_AINT_LOW_WORD]); - return res; -#endif -} - -#endif - ERTS_GLB_INLINE Uint64 erts_current_interval_nob__(erts_interval_t *icp) { -#ifdef ARCH_64 - return (Uint64) erts_atomic_read_nob(&icp->counter.atomic); -#else - erts_dw_aint_t dw; - erts_dw_atomic_read_nob(&icp->counter.atomic, &dw); - return erts_interval_dw_aint_to_val__(&dw); -#endif + return (Uint64) erts_atomic64_read_nob(&icp->counter.atomic); } ERTS_GLB_INLINE Uint64 erts_current_interval_acqb__(erts_interval_t *icp) { -#ifdef ARCH_64 - return (Uint64) erts_atomic_read_acqb(&icp->counter.atomic); -#else - erts_dw_aint_t dw; - erts_dw_atomic_read_acqb(&icp->counter.atomic, &dw); - return erts_interval_dw_aint_to_val__(&dw); -#endif + return (Uint64) erts_atomic64_read_acqb(&icp->counter.atomic); } ERTS_GLB_INLINE Uint64 @@ -150,12 +113,14 @@ void erts_silence_warn_unused_result(long unused); int erts_fit_in_bits_int64(Sint64); int erts_fit_in_bits_int32(Sint32); +int erts_fit_in_bits_uint(Uint); int erts_list_length(Eterm); int erts_is_builtin(Eterm, Eterm, int); Uint32 make_broken_hash(Eterm); Uint32 block_hash(byte *, unsigned, Uint32); Uint32 make_hash2(Eterm); Uint32 make_hash(Eterm); +Uint32 make_internal_hash(Eterm); void erts_save_emu_args(int argc, char **argv); Eterm erts_get_emu_args(struct process *c_p); diff --git a/erts/emulator/beam/erl_vm.h b/erts/emulator/beam/erl_vm.h index b7de8208ad..3a9fb1e07b 100644 --- a/erts/emulator/beam/erl_vm.h +++ b/erts/emulator/beam/erl_vm.h @@ -20,8 +20,6 @@ #ifndef __ERL_VM_H__ #define __ERL_VM_H__ -/* #define ERTS_OPCODE_COUNTER_SUPPORT */ - /* FORCE_HEAP_FRAGS: * Debug provocation to make HAlloc always create heap fragments (if allowed) * even if there is room on heap. @@ -119,9 +117,9 @@ #if defined(DEBUG) || defined(CHECK_FOR_HOLES) #if HALFWORD_HEAP -# define ERTS_HOLE_MARKER (0xaf5e78ccU) +# define ERTS_HOLE_MARKER (0xdeadbeef) #else -# define ERTS_HOLE_MARKER (((0xaf5e78ccUL << 24) << 8) | 0xaf5e78ccUL) +# define ERTS_HOLE_MARKER (((0xdeadbeef << 24) << 8) | 0xdeadbeef) #endif #endif @@ -174,6 +172,7 @@ extern int H_MIN_SIZE; /* minimum (heap + stack) */ extern int BIN_VH_MIN_SIZE; /* minimum virtual (bin) heap */ extern int erts_atom_table_size;/* Atom table size */ +extern int erts_pd_initial_size;/* Initial Process dictionary table size */ #define ORIG_CREATION 0 #define INTERNAL_CREATION 255 diff --git a/erts/emulator/beam/external.c b/erts/emulator/beam/external.c index ae011a34bd..82c60840e5 100644 --- a/erts/emulator/beam/external.c +++ b/erts/emulator/beam/external.c @@ -36,7 +36,9 @@ #include "erl_process.h" #include "error.h" #include "external.h" +#define ERL_WANT_HIPE_BIF_WRAPPER__ #include "bif.h" +#undef ERL_WANT_HIPE_BIF_WRAPPER__ #include "big.h" #include "dist.h" #include "erl_binary.h" @@ -498,15 +500,37 @@ byte *erts_encode_ext_dist_header_finalize(byte *ext, ErtsAtomCache *cache, Uint return ep; } -Uint erts_encode_dist_ext_size(Eterm term, Uint32 flags, ErtsAtomCacheMap *acmp) +int erts_encode_dist_ext_size(Eterm term, Uint32 flags, ErtsAtomCacheMap *acmp, + Uint* szp) { - Uint sz = 0; + Uint sz; + if (encode_size_struct_int(NULL, acmp, term, flags, NULL, &sz)) { + return -1; + } else { #ifndef ERTS_DEBUG_USE_DIST_SEP - if (!(flags & DFLAG_DIST_HDR_ATOM_CACHE)) + if (!(flags & DFLAG_DIST_HDR_ATOM_CACHE)) #endif - sz++ /* VERSION_MAGIC */; - sz += encode_size_struct2(acmp, term, flags); - return sz; + sz++ /* VERSION_MAGIC */; + + *szp += sz; + return 0; + } +} + +int erts_encode_dist_ext_size_int(Eterm term, struct erts_dsig_send_context* ctx, Uint* szp) +{ + Uint sz; + if (encode_size_struct_int(&ctx->u.sc, ctx->acmp, term, ctx->flags, &ctx->reds, &sz)) { + return -1; + } else { +#ifndef ERTS_DEBUG_USE_DIST_SEP + if (!(ctx->flags & DFLAG_DIST_HDR_ATOM_CACHE)) +#endif + sz++ /* VERSION_MAGIC */; + + *szp += sz; + return 0; + } } Uint erts_encode_ext_size(Eterm term) @@ -527,19 +551,16 @@ Uint erts_encode_ext_size_ets(Eterm term) } -void erts_encode_dist_ext(Eterm term, byte **ext, Uint32 flags, ErtsAtomCacheMap *acmp) +int erts_encode_dist_ext(Eterm term, byte **ext, Uint32 flags, ErtsAtomCacheMap *acmp, + TTBEncodeContext* ctx, Sint* reds) { - byte *ep = *ext; -#ifndef ERTS_DEBUG_USE_DIST_SEP - if (!(flags & DFLAG_DIST_HDR_ATOM_CACHE)) -#endif - *ep++ = VERSION_MAGIC; - ep = enc_term(acmp, term, ep, flags, NULL); - if (!ep) - erl_exit(ERTS_ABORT_EXIT, - "%s:%d:erts_encode_dist_ext(): Internal data structure error\n", - __FILE__, __LINE__); - *ext = ep; + if (!ctx || !ctx->wstack.wstart) { + #ifndef ERTS_DEBUG_USE_DIST_SEP + if (!(flags & DFLAG_DIST_HDR_ATOM_CACHE)) + #endif + *(*ext)++ = VERSION_MAGIC; + } + return enc_term_int(ctx, acmp, term, *ext, flags, NULL, reds, ext); } void erts_encode_ext(Eterm term, byte **ext) @@ -1161,7 +1182,8 @@ typedef struct { Eterm* hp_end; int remaining_n; char* remaining_bytes; - Eterm* maps_head; + Eterm* maps_list; + struct dec_term_hamt_placeholder* hamt_list; } B2TDecodeContext; typedef struct { @@ -1487,7 +1509,8 @@ static BIF_RETTYPE binary_to_term_int(Process* p, Uint32 flags, Eterm bin, Binar 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->u.dc.maps_list = NULL; + ctx->u.dc.hamt_list = NULL; ctx->state = B2TDecode; /*fall through*/ case B2TDecode: @@ -1740,54 +1763,14 @@ erts_term_to_binary(Process* p, Eterm Term, int level, Uint flags) { return erts_term_to_binary_simple(p, Term, size, level, flags); } -/* Define for testing */ -/* #define EXTREME_TTB_TRAPPING 1 */ +/* Define EXTREME_TTB_TRAPPING for testing in dist.h */ #ifndef EXTREME_TTB_TRAPPING -#define TERM_TO_BINARY_LOOP_FACTOR 32 #define TERM_TO_BINARY_COMPRESS_CHUNK (1 << 18) #else -#define TERM_TO_BINARY_LOOP_FACTOR 1 #define TERM_TO_BINARY_COMPRESS_CHUNK 10 #endif - - -typedef enum { TTBSize, TTBEncode, TTBCompress } TTBState; -typedef struct TTBSizeContext_ { - Uint flags; - int level; - Uint result; - Eterm obj; - ErtsEStack estack; -} TTBSizeContext; - -typedef struct TTBEncodeContext_ { - Uint flags; - int level; - byte* ep; - Eterm obj; - ErtsWStack wstack; - Binary *result_bin; -} TTBEncodeContext; - -typedef struct { - Uint real_size; - Uint dest_len; - byte *dbytes; - Binary *result_bin; - Binary *destination_bin; - z_stream stream; -} TTBCompressContext; - -typedef struct { - int alive; - TTBState state; - union { - TTBSizeContext sc; - TTBEncodeContext ec; - TTBCompressContext cc; - } s; -} TTBContext; +#define TERM_TO_BINARY_MEMCPY_FACTOR 8 static void ttb_context_destructor(Binary *context_bin) { @@ -1899,8 +1882,6 @@ static Eterm erts_term_to_binary_int(Process* p, Eterm Term, int level, Uint fla } result_bin = erts_bin_nrml_alloc(size); - result_bin->flags = 0; - result_bin->orig_size = size; erts_refc_init(&result_bin->refc, 0); result_bin->orig_bytes[0] = VERSION_MAGIC; /* Next state immediately, no need to export context */ @@ -1925,7 +1906,6 @@ static Eterm erts_term_to_binary_int(Process* p, Eterm Term, int level, Uint fla } real_size = endp - bytes; result_bin = erts_bin_realloc(context->s.ec.result_bin,real_size); - result_bin->orig_size = real_size; level = context->s.ec.level; BUMP_REDS(p, (initial_reds - reds) / TERM_TO_BINARY_LOOP_FACTOR); if (level == 0 || real_size < 6) { /* We are done */ @@ -1962,8 +1942,6 @@ static Eterm erts_term_to_binary_int(Process* p, Eterm Term, int level, Uint fla context->s.cc.result_bin = result_bin; result_bin = erts_bin_nrml_alloc(real_size); - result_bin->flags = 0; - result_bin->orig_size = real_size; erts_refc_init(&result_bin->refc, 0); result_bin->orig_bytes[0] = VERSION_MAGIC; @@ -2005,7 +1983,6 @@ static Eterm erts_term_to_binary_int(Process* p, Eterm Term, int level, Uint fla erl_zlib_deflate_finish(&(context->s.cc.stream)); result_bin = erts_bin_realloc(context->s.cc.destination_bin, context->s.cc.dest_len+6); - result_bin->orig_size = context->s.cc.dest_len+6; context->s.cc.destination_bin = NULL; pb = (ProcBin *) HAlloc(p, PROC_BIN_SIZE); pb->thing_word = HEADER_PROC_BIN; @@ -2327,8 +2304,10 @@ dec_pid(ErtsDistExternal *edep, Eterm** hpp, byte* ep, ErlOffHeap* off_heap, Ete #define ENC_TERM ((Eterm) 0) #define ENC_ONE_CONS ((Eterm) 1) #define ENC_PATCH_FUN_SIZE ((Eterm) 2) -#define ENC_LAST_ARRAY_ELEMENT ((Eterm) 3) - +#define ENC_BIN_COPY ((Eterm) 3) +#define ENC_MAP_PAIR ((Eterm) 4) +#define ENC_HASHMAP_NODE ((Eterm) 5) +#define ENC_LAST_ARRAY_ELEMENT ((Eterm) 6) static byte* enc_term(ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, Uint32 dflags, @@ -2364,6 +2343,9 @@ enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, WSTACK_RESTORE(s, &ctx->wstack); ep = ctx->ep; obj = ctx->obj; + if (is_non_value(obj)) { + goto outer_loop; + } } } @@ -2387,8 +2369,8 @@ enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, obj = CAR(cons); tl = CDR(cons); - WSTACK_PUSH(s, is_list(tl) ? ENC_ONE_CONS : ENC_TERM); - WSTACK_PUSH(s, tl); + WSTACK_PUSH2(s, (is_list(tl) ? ENC_ONE_CONS : ENC_TERM), + tl); } break; case ENC_PATCH_FUN_SIZE: @@ -2401,6 +2383,46 @@ enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, put_int32(ep - size_p, size_p); } goto outer_loop; + case ENC_BIN_COPY: { + Uint bits = (Uint)obj; + Uint bitoffs = WSTACK_POP(s); + byte* bytes = (byte*) WSTACK_POP(s); + byte* dst = (byte*) WSTACK_POP(s); + if (bits > r * (TERM_TO_BINARY_MEMCPY_FACTOR * 8)) { + Uint n = r * TERM_TO_BINARY_MEMCPY_FACTOR; + WSTACK_PUSH5(s, (UWord)(dst + n), (UWord)(bytes + n), bitoffs, + ENC_BIN_COPY, bits - 8*n); + bits = 8*n; + copy_binary_to_buffer(dst, 0, bytes, bitoffs, bits); + obj = THE_NON_VALUE; + r = 0; /* yield */ + break; + } else { + copy_binary_to_buffer(dst, 0, bytes, bitoffs, bits); + r -= bits / (TERM_TO_BINARY_MEMCPY_FACTOR * 8); + goto outer_loop; + } + } + case ENC_MAP_PAIR: { + Uint pairs_left = obj; + Eterm *vptr = (Eterm*) WSTACK_POP(s); + Eterm *kptr = (Eterm*) WSTACK_POP(s); + + obj = *kptr; + if (--pairs_left > 0) { + WSTACK_PUSH4(s, (UWord)(kptr+1), (UWord)(vptr+1), + ENC_MAP_PAIR, pairs_left); + } + WSTACK_PUSH2(s, ENC_TERM, *vptr); + break; + } + case ENC_HASHMAP_NODE: + if (is_list(obj)) { /* leaf node [K|V] */ + ptr = list_val(obj); + WSTACK_PUSH2(s, ENC_TERM, CDR(ptr)); + obj = CAR(ptr); + } + break; case ENC_LAST_ARRAY_ELEMENT: /* obj is the tuple */ { @@ -2419,17 +2441,16 @@ enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, #else Eterm* ptr = (Eterm *) obj; #endif - WSTACK_PUSH(s, val-1); obj = *ptr++; - WSTACK_PUSH(s, (UWord)ptr); + WSTACK_PUSH2(s, val-1, (UWord)ptr); } break; } L_jump_start: - if (ctx && --r == 0) { - *reds = r; + if (ctx && --r <= 0) { + *reds = 0; ctx->obj = obj; ctx->ep = ep; WSTACK_SAVE(s, &ctx->wstack); @@ -2578,35 +2599,62 @@ enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, ep += 4; } if (i > 0) { - WSTACK_PUSH(s, ENC_LAST_ARRAY_ELEMENT+i-1); - WSTACK_PUSH(s, (UWord)ptr); + WSTACK_PUSH2(s, ENC_LAST_ARRAY_ELEMENT+i-1, (UWord)ptr); } break; case MAP_DEF: { - map_t *mp = (map_t*)map_val(obj); - Uint size = map_get_size(mp); + flatmap_t *mp = (flatmap_t*)flatmap_val(obj); + Uint size = flatmap_get_size(mp); *ep++ = MAP_EXT; put_int32(size, ep); ep += 4; if (size > 0) { - Eterm *kptr = map_get_keys(mp); - Eterm *vptr = map_get_values(mp); - - for (i = size-1; i >= 1; i--) { - WSTACK_PUSH(s, ENC_TERM); - WSTACK_PUSH(s, (UWord) vptr[i]); - WSTACK_PUSH(s, ENC_TERM); - WSTACK_PUSH(s, (UWord) kptr[i]); - } + Eterm *kptr = flatmap_get_keys(mp); + Eterm *vptr = flatmap_get_values(mp); - WSTACK_PUSH(s, ENC_TERM); - WSTACK_PUSH(s, (UWord) vptr[0]); + WSTACK_PUSH4(s, (UWord)kptr, (UWord)vptr, + ENC_MAP_PAIR, size); + } + } + break; - obj = kptr[0]; - goto L_jump_start; + case HASHMAP_DEF: + { + Eterm hdr; + Uint node_sz; + ptr = boxed_val(obj); + hdr = *ptr; + ASSERT(is_header(hdr)); + switch(hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: + *ep++ = MAP_EXT; + ptr++; + put_int32(*ptr, ep); ep += 4; + /*fall through*/ + case HAMT_SUBTAG_NODE_ARRAY: + node_sz = 16; + break; + case HAMT_SUBTAG_HEAD_BITMAP: + *ep++ = MAP_EXT; + ptr++; + put_int32(*ptr, ep); ep += 4; + /*fall through*/ + case HAMT_SUBTAG_NODE_BITMAP: + node_sz = hashmap_bitcount(MAP_HEADER_VAL(hdr)); + ASSERT(node_sz < 17); + break; + default: + erl_exit(1, "bad header\r\n"); + } + + ptr++; + WSTACK_RESERVE(s, node_sz*2); + while(node_sz--) { + WSTACK_FAST_PUSH(s, ENC_HASHMAP_NODE); + WSTACK_FAST_PUSH(s, *ptr++); } } break; @@ -2644,6 +2692,7 @@ enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, Uint bitoffs; Uint bitsize; byte* bytes; + byte* data_dst; ERTS_GET_BINARY_BYTES(obj, bytes, bitoffs, bitsize); if (dflags & DFLAG_INTERNAL_TAGS) { @@ -2689,7 +2738,7 @@ enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, j = binary_size(obj); put_int32(j, ep); ep += 4; - copy_binary_to_buffer(ep, 0, bytes, bitoffs, 8*j); + data_dst = ep; ep += j; } else if (dflags & DFLAG_BIT_BINARIES) { /* Bit-level binary. */ @@ -2699,7 +2748,7 @@ enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, ep += 4; *ep++ = bitsize; ep[j] = 0; /* Zero unused bits at end of binary */ - copy_binary_to_buffer(ep, 0, bytes, bitoffs, 8*j+bitsize); + data_dst = ep; ep += j + 1; } else { /* @@ -2713,11 +2762,18 @@ enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, put_int32((j+1), ep); ep += 4; ep[j] = 0; /* Zero unused bits at end of binary */ - copy_binary_to_buffer(ep, 0, bytes, bitoffs, 8*j+bitsize); + data_dst = ep; ep += j+1; *ep++ = SMALL_INTEGER_EXT; *ep++ = bitsize; } + if (ctx && j > r * TERM_TO_BINARY_MEMCPY_FACTOR) { + WSTACK_PUSH5(s, (UWord)data_dst, (UWord)bytes, bitoffs, + ENC_BIN_COPY, 8*j + bitsize); + } else { + copy_binary_to_buffer(data_dst, 0, bytes, bitoffs, + 8 * j + bitsize); + } } break; case EXPORT_DEF: @@ -2746,13 +2802,12 @@ enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, case FUN_DEF: { ErlFunThing* funp = (ErlFunThing *) fun_val(obj); + int ei; if ((dflags & DFLAG_NEW_FUN_TAGS) != 0) { - int ei; - *ep++ = NEW_FUN_EXT; - WSTACK_PUSH(s, ENC_PATCH_FUN_SIZE); - WSTACK_PUSH(s, (UWord) ep); /* Position for patching in size */ + WSTACK_PUSH2(s, ENC_PATCH_FUN_SIZE, + (UWord) ep); /* Position for patching in size */ ep += 4; *ep = funp->arity; ep += 1; @@ -2766,16 +2821,6 @@ enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, ep = enc_term(acmp, make_small(funp->fe->old_index), ep, dflags, off_heap); ep = enc_term(acmp, make_small(funp->fe->old_uniq), ep, dflags, off_heap); ep = enc_pid(acmp, funp->creator, ep, dflags); - - fun_env: - for (ei = funp->num_free-1; ei > 0; ei--) { - WSTACK_PUSH(s, ENC_TERM); - WSTACK_PUSH(s, (UWord) funp->env[ei]); - } - if (funp->num_free != 0) { - obj = funp->env[0]; - goto L_jump_start; - } } else { /* * Communicating with an obsolete erl_interface or @@ -2807,7 +2852,13 @@ enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, *ep++ = SMALL_TUPLE_EXT; put_int8(funp->num_free, ep); ep += 1; - goto fun_env; + } + for (ei = funp->num_free-1; ei > 0; ei--) { + WSTACK_PUSH2(s, ENC_TERM, (UWord) funp->env[ei]); + } + if (funp->num_free != 0) { + obj = funp->env[0]; + goto L_jump_start; } } break; @@ -2889,9 +2940,19 @@ undo_offheap_in_area(ErlOffHeap* off_heap, Eterm* start, Eterm* end) #endif /* DEBUG */ } +struct dec_term_hamt_placeholder +{ + struct dec_term_hamt_placeholder* next; + Eterm* objp; /* write result here */ + Uint size; /* nr of leafs */ + Eterm leafs[1]; +}; + +#define DEC_TERM_HAMT_PLACEHOLDER_SIZE \ + (offsetof(struct dec_term_hamt_placeholder, leafs) / sizeof(Eterm)) /* Decode term from external format into *objp. -** On failure return NULL and (R13B04) *hpp will be unchanged. +** On failure return NULL and *hpp will be unchanged. */ static byte* dec_term(ErtsDistExternal *edep, Eterm** hpp, byte* ep, ErlOffHeap* off_heap, @@ -2901,7 +2962,8 @@ 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 *maps_list; /* for preprocessing of small maps */ + struct dec_term_hamt_placeholder* hamt_list; /* for preprocessing of big maps */ Eterm* next; SWord reds; @@ -2911,7 +2973,8 @@ 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; + maps_list = ctx->u.dc.maps_list; + hamt_list = ctx->u.dc.hamt_list; if (ctx->state != B2TDecode) { int n_limit = reds; @@ -2992,7 +3055,8 @@ dec_term(ErtsDistExternal *edep, Eterm** hpp, byte* ep, ErlOffHeap* off_heap, reds = ERTS_SWORD_MAX; next = objp; *next = (Eterm) (UWord) NULL; - maps_head = NULL; + maps_list = NULL; + hamt_list = NULL; } hp = *hpp; @@ -3056,6 +3120,8 @@ dec_term(ErtsDistExternal *edep, Eterm** hpp, byte* ep, ErlOffHeap* off_heap, big = make_small(0); } else { big = bytes_to_big(first, n, neg, hp); + if (is_nil(big)) + goto error; if (is_big(big)) { hp += big_arity(big) + 1; } @@ -3387,8 +3453,6 @@ dec_term_atom_common: } else { Binary* dbin = erts_bin_nrml_alloc(n); ProcBin* pb; - dbin->flags = 0; - dbin->orig_size = n; erts_refc_init(&dbin->refc, 1); pb = (ProcBin *) hp; hp += PROC_BIN_SIZE; @@ -3442,8 +3506,6 @@ dec_term_atom_common: Binary* dbin = erts_bin_nrml_alloc(n); ProcBin* pb; - dbin->flags = 0; - dbin->orig_size = n; erts_refc_init(&dbin->refc, 1); pb = (ProcBin *) hp; pb->thing_word = HEADER_PROC_BIN; @@ -3532,46 +3594,67 @@ dec_term_atom_common: 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); - hp += size; - kptr = hp - 1; - - mp = (map_t*)hp; - hp += MAP_HEADER_SIZE; - hp += size; - vptr = hp - 1; - - /* kptr, last word for keys - * vptr, last 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); - - for (n = size; n; n--) { - *vptr = (Eterm) COMPRESS_POINTER(next); - *kptr = (Eterm) COMPRESS_POINTER(vptr); - next = kptr; - vptr--; - kptr--; - } + if (size <= MAP_SMALL_MAP_LIMIT) { + flatmap_t *mp; + + keys = make_tuple(hp); + *hp++ = make_arityval(size); + hp += size; + kptr = hp - 1; + + mp = (flatmap_t*)hp; + hp += MAP_HEADER_SIZE; + hp += size; + vptr = hp - 1; + + /* kptr, last word for keys + * vptr, last 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_list); + maps_list = (Eterm *) mp; + + mp->size = size; + mp->keys = keys; + *objp = make_flatmap(mp); + + for (n = size; n; n--) { + *vptr = (Eterm) COMPRESS_POINTER(next); + *kptr = (Eterm) COMPRESS_POINTER(vptr); + next = kptr; + vptr--; + kptr--; + } + } + else { /* Make hamt */ + struct dec_term_hamt_placeholder* holder = + (struct dec_term_hamt_placeholder*) hp; + + holder->next = hamt_list; + hamt_list = holder; + holder->objp = objp; + holder->size = size; + + hp += DEC_TERM_HAMT_PLACEHOLDER_SIZE; + + for (n = size; n; n--) { + CDR(hp) = (Eterm) COMPRESS_POINTER(next); + CAR(hp) = (Eterm) COMPRESS_POINTER(&CDR(hp)); + next = &CAR(hp); + hp += 2; + } + } } break; case NEW_FUN_EXT: @@ -3794,7 +3877,7 @@ 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->u.dc.maps_list = maps_list; ctx->reds = 0; return NULL; } @@ -3809,12 +3892,40 @@ dec_term_atom_common: * - 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)) + while (maps_list) { + next = (Eterm *)(EXPAND_POINTER(*maps_list)); + *maps_list = MAP_HEADER; + if (!erts_validate_and_sort_flatmap((flatmap_t*)maps_list)) goto error; - maps_head = next; + maps_list = next; + } + + /* Iterate through all the hamts and build tree nodes. + */ + if (hamt_list) { + ErtsHeapFactory factory; + + factory.p = NULL; + factory.hp = hp; + /* We assume heap will suffice (see hashmap_over_estimated_heap_size) */ + + do { + struct dec_term_hamt_placeholder* hamt = hamt_list; + *hamt->objp = erts_hashmap_from_array(&factory, + hamt->leafs, + hamt->size, + 1); + if (is_non_value(*hamt->objp)) + goto error; + + hamt_list = hamt->next; + + /* Yes, we waste a couple of heap words per hamt + for the temporary placeholder */ + *(Eterm*)hamt = make_pos_bignum_header(DEC_TERM_HAMT_PLACEHOLDER_SIZE-1); + } while (hamt_list); + + hp = factory.hp; } if (ctx) { @@ -4012,15 +4123,15 @@ encode_size_struct_int(TTBSizeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, break; case MAP_DEF: { - map_t *mp = (map_t*)map_val(obj); - Uint size = map_get_size(mp); + flatmap_t *mp = (flatmap_t*)flatmap_val(obj); + Uint size = flatmap_get_size(mp); Uint i; Eterm *ptr; result += 1 + 4; /* tag + 4 bytes size */ /* push values first */ - ptr = map_get_values(mp); + ptr = flatmap_get_values(mp); i = size; while(i--) { if (is_list(*ptr)) { @@ -4034,7 +4145,7 @@ encode_size_struct_int(TTBSizeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, ++ptr; } - ptr = map_get_keys(mp); + ptr = flatmap_get_keys(mp); i = size; while(i--) { if (is_list(*ptr)) { @@ -4050,6 +4161,38 @@ encode_size_struct_int(TTBSizeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, goto outer_loop; } break; + + case HASHMAP_DEF: + { + Eterm *ptr; + Eterm hdr; + Uint node_sz; + ptr = boxed_val(obj); + hdr = *ptr; + ASSERT(is_header(hdr)); + switch(hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: ptr++; + case HAMT_SUBTAG_NODE_ARRAY: + node_sz = 16; + break; + case HAMT_SUBTAG_HEAD_BITMAP: ptr++; + case HAMT_SUBTAG_NODE_BITMAP: + node_sz = hashmap_bitcount(MAP_HEADER_VAL(hdr)); + ASSERT(node_sz < 17); + break; + default: + erl_exit(1, "bad header\r\n"); + } + + ptr++; + ESTACK_RESERVE(s, node_sz); + while(node_sz--) { + ESTACK_FAST_PUSH(s, *ptr++); + } + result += 1 + 4; /* tag + 4 bytes size */ + } + + break; case FLOAT_DEF: if (dflags & DFLAG_NEW_FLOATS) { result += 9; @@ -4152,6 +4295,8 @@ encode_size_struct_int(TTBSizeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, return 0; } + + static Sint decoded_size(byte *ep, byte* endp, int internal_tags, B2TContext* ctx) { @@ -4345,7 +4490,11 @@ init_done: n = get_int32(ep); ep += 4; ADDTERMS(2*n); - heap_size += 3 + n + 1 + n; + if (n <= MAP_SMALL_MAP_LIMIT) { + heap_size += 3 + n + 1 + n; + } else { + heap_size += hashmap_over_estimated_heap_size(n); + } break; case STRING_EXT: CHKSIZE(2); diff --git a/erts/emulator/beam/external.h b/erts/emulator/beam/external.h index bf00958eb1..50fcfa04d6 100644 --- a/erts/emulator/beam/external.h +++ b/erts/emulator/beam/external.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 @@ -150,19 +150,23 @@ typedef struct { Uint extsize; } ErtsBinary2TermState; + /* -------------------------------------------------------------------------- */ void erts_init_atom_cache_map(ErtsAtomCacheMap *); void erts_reset_atom_cache_map(ErtsAtomCacheMap *); void erts_destroy_atom_cache_map(ErtsAtomCacheMap *); void erts_finalize_atom_cache_map(ErtsAtomCacheMap *, Uint32); -Uint erts_encode_ext_dist_header_size(ErtsAtomCacheMap *); Uint erts_encode_ext_dist_header_size(ErtsAtomCacheMap *); byte *erts_encode_ext_dist_header_setup(byte *, ErtsAtomCacheMap *); byte *erts_encode_ext_dist_header_finalize(byte *, ErtsAtomCache *, Uint32); -Uint erts_encode_dist_ext_size(Eterm, Uint32, ErtsAtomCacheMap *); -void erts_encode_dist_ext(Eterm, byte **, Uint32, ErtsAtomCacheMap *); +struct erts_dsig_send_context; +int erts_encode_dist_ext_size(Eterm, Uint32, ErtsAtomCacheMap*, Uint* szp); +int erts_encode_dist_ext_size_int(Eterm term, struct erts_dsig_send_context* ctx, Uint* szp); +struct TTBEncodeContext_; +int erts_encode_dist_ext(Eterm, byte **, Uint32, ErtsAtomCacheMap *, + struct TTBEncodeContext_ *, Sint* reds); Uint erts_encode_ext_size(Eterm); Uint erts_encode_ext_size_2(Eterm, unsigned); diff --git a/erts/emulator/beam/global.h b/erts/emulator/beam/global.h index 891046a8b5..42daa2c9ef 100644 --- a/erts/emulator/beam/global.h +++ b/erts/emulator/beam/global.h @@ -41,6 +41,7 @@ #include "error.h" #include "erl_utils.h" #include "erl_port.h" +#include "erl_gc.h" struct enif_environment_t /* ErlNifEnv */ { @@ -160,6 +161,7 @@ struct erts_driver_t_ { void (*ready_async)(ErlDrvData drv_data, ErlDrvThreadData thread_data); /* Might be NULL */ void (*process_exit)(ErlDrvData drv_data, ErlDrvMonitor *monitor); void (*stop_select)(ErlDrvEvent event, void*); /* Might be NULL */ + void (*emergency_close)(ErlDrvData drv_data); /* Might be NULL */ }; extern erts_driver_t *driver_list; @@ -370,16 +372,17 @@ extern int stackdump_on_exit; * DESTROY_ESTACK(Stack) */ -typedef struct { +typedef struct ErtsEStack_ { Eterm* start; Eterm* sp; Eterm* end; + Eterm* edefault; ErtsAlcType_t alloc_type; }ErtsEStack; #define DEF_ESTACK_SIZE (16) -void erl_grow_estack(ErtsEStack*, Eterm* def_stack); +void erl_grow_estack(ErtsEStack*, Uint need); #define ESTK_CONCAT(a,b) a##b #define ESTK_DEF_STACK(s) ESTK_CONCAT(s,_default_estack) @@ -389,22 +392,23 @@ void erl_grow_estack(ErtsEStack*, Eterm* def_stack); ESTK_DEF_STACK(s), /* start */ \ ESTK_DEF_STACK(s), /* sp */ \ ESTK_DEF_STACK(s) + DEF_ESTACK_SIZE, /* end */ \ + ESTK_DEF_STACK(s), /* default */ \ ERTS_ALC_T_ESTACK /* alloc_type */ \ } #define ESTACK_CHANGE_ALLOCATOR(s,t) \ do { \ - if (s.start != ESTK_DEF_STACK(s)) { \ + if ((s).start != ESTK_DEF_STACK(s)) { \ erl_exit(1, "Internal error - trying to change allocator " \ "type of active estack\n"); \ } \ - s.alloc_type = (t); \ + (s).alloc_type = (t); \ } while (0) #define DESTROY_ESTACK(s) \ do { \ - if (s.start != ESTK_DEF_STACK(s)) { \ - erts_free(s.alloc_type, s.start); \ + if ((s).start != ESTK_DEF_STACK(s)) { \ + erts_free((s).alloc_type, (s).start); \ } \ } while(0) @@ -415,16 +419,17 @@ do { \ */ #define ESTACK_SAVE(s,dst)\ do {\ - if (s.start == ESTK_DEF_STACK(s)) {\ + if ((s).start == ESTK_DEF_STACK(s)) {\ UWord _wsz = ESTACK_COUNT(s);\ - (dst)->start = erts_alloc(s.alloc_type,\ + (dst)->start = erts_alloc((s).alloc_type,\ DEF_ESTACK_SIZE * sizeof(Eterm));\ - memcpy((dst)->start, s.start,_wsz*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;\ + (dst)->edefault = NULL;\ + (dst)->alloc_type = (s).alloc_type;\ } else\ - *(dst) = s;\ + *(dst) = (s);\ } while (0) #define DESTROY_SAVED_ESTACK(estack)\ @@ -443,72 +448,114 @@ do {\ */ #define ESTACK_RESTORE(s, src) \ do { \ - ASSERT(s.start == ESTK_DEF_STACK(s)); \ - s = *(src); /* struct copy */ \ + ASSERT((s).start == ESTK_DEF_STACK(s)); \ + (s) = *(src); /* struct copy */ \ (src)->start = NULL; \ - ASSERT(s.sp >= s.start); \ - ASSERT(s.sp <= s.end); \ + ASSERT((s).sp >= (s).start); \ + ASSERT((s).sp <= (s).end); \ } while (0) -#define ESTACK_IS_STATIC(s) (s.start == ESTK_DEF_STACK(s))) +#define ESTACK_IS_STATIC(s) ((s).start == ESTK_DEF_STACK(s)) -#define ESTACK_PUSH(s, x) \ -do { \ - if (s.sp == s.end) { \ - erl_grow_estack(&s, ESTK_DEF_STACK(s)); \ - } \ - *s.sp++ = (x); \ +#define ESTACK_PUSH(s, x) \ +do { \ + if ((s).sp == (s).end) { \ + erl_grow_estack(&(s), 1); \ + } \ + *(s).sp++ = (x); \ } while(0) #define ESTACK_PUSH2(s, x, y) \ do { \ - if (s.sp > s.end - 2) { \ - erl_grow_estack(&s, ESTK_DEF_STACK(s)); \ + if ((s).sp > (s).end - 2) { \ + erl_grow_estack(&(s), 2); \ } \ - *s.sp++ = (x); \ - *s.sp++ = (y); \ + *(s).sp++ = (x); \ + *(s).sp++ = (y); \ } while(0) #define ESTACK_PUSH3(s, x, y, z) \ do { \ - if (s.sp > s.end - 3) { \ - erl_grow_estack(&s, ESTK_DEF_STACK(s)); \ + if ((s).sp > (s).end - 3) { \ + erl_grow_estack(&s, 3); \ } \ - *s.sp++ = (x); \ - *s.sp++ = (y); \ - *s.sp++ = (z); \ + *(s).sp++ = (x); \ + *(s).sp++ = (y); \ + *(s).sp++ = (z); \ +} while(0) + +#define ESTACK_PUSH4(s, E1, E2, E3, E4) \ +do { \ + if ((s).sp > (s).end - 4) { \ + erl_grow_estack(&s, 4); \ + } \ + *(s).sp++ = (E1); \ + *(s).sp++ = (E2); \ + *(s).sp++ = (E3); \ + *(s).sp++ = (E4); \ +} while(0) + +#define ESTACK_RESERVE(s, push_cnt) \ +do { \ + if ((s).sp > (s).end - (push_cnt)) { \ + erl_grow_estack(&(s), (push_cnt)); \ + } \ +} while(0) + +/* Must be preceded by ESTACK_RESERVE */ +#define ESTACK_FAST_PUSH(s, x) \ +do { \ + ASSERT((s).sp < (s).end); \ + *s.sp++ = (x); \ } while(0) -#define ESTACK_COUNT(s) (s.sp - s.start) -#define ESTACK_ISEMPTY(s) (s.sp == s.start) -#define ESTACK_POP(s) (*(--s.sp)) +#define ESTACK_COUNT(s) ((s).sp - (s).start) +#define ESTACK_ISEMPTY(s) ((s).sp == (s).start) +#define ESTACK_POP(s) (*(--(s).sp)) /* * WSTACK: same as ESTACK but with UWord instead of Eterm */ -typedef struct { +typedef struct ErtsWStack_ { UWord* wstart; UWord* wsp; UWord* wend; + UWord* wdefault; ErtsAlcType_t alloc_type; }ErtsWStack; #define DEF_WSTACK_SIZE (16) -void erl_grow_wstack(ErtsWStack*, UWord* def_stack); +void erl_grow_wstack(ErtsWStack*, Uint need); #define WSTK_CONCAT(a,b) a##b #define WSTK_DEF_STACK(s) WSTK_CONCAT(s,_default_wstack) -#define DECLARE_WSTACK(s) \ +#define WSTACK_DECLARE(s) \ UWord WSTK_DEF_STACK(s)[DEF_WSTACK_SIZE]; \ ErtsWStack s = { \ WSTK_DEF_STACK(s), /* wstart */ \ WSTK_DEF_STACK(s), /* wsp */ \ WSTK_DEF_STACK(s) + DEF_WSTACK_SIZE, /* wend */ \ + WSTK_DEF_STACK(s), /* wdflt */ \ ERTS_ALC_T_ESTACK /* alloc_type */ \ } +#define DECLARE_WSTACK WSTACK_DECLARE + +typedef struct ErtsDynamicWStack_ { + UWord default_stack[DEF_WSTACK_SIZE]; + ErtsWStack ws; +}ErtsDynamicWStack; + +#define WSTACK_INIT(dwsp, ALC_TYPE) \ +do { \ + (dwsp)->ws.wstart = (dwsp)->default_stack; \ + (dwsp)->ws.wsp = (dwsp)->default_stack; \ + (dwsp)->ws.wend = (dwsp)->default_stack + DEF_WSTACK_SIZE;\ + (dwsp)->ws.wdefault = (dwsp)->default_stack; \ + (dwsp)->ws.alloc_type = ALC_TYPE; \ +} while (0) #define WSTACK_CHANGE_ALLOCATOR(s,t) \ do { \ @@ -519,13 +566,20 @@ do { \ s.alloc_type = (t); \ } while (0) -#define DESTROY_WSTACK(s) \ +#define WSTACK_DESTROY(s) \ do { \ - if (s.wstart != WSTK_DEF_STACK(s)) { \ + if (s.wstart != s.wdefault) { \ erts_free(s.alloc_type, s.wstart); \ } \ } while(0) +#define DESTROY_WSTACK WSTACK_DESTROY +#define WSTACK_DEBUG(s) \ + do { \ + fprintf(stderr, "wstack size = %ld\r\n", s.wsp - s.wstart); \ + fprintf(stderr, "wstack wstart = %p\r\n", s.wstart); \ + fprintf(stderr, "wstack wsp = %p\r\n", s.wsp); \ + } while(0) /* * Do not free the stack after this, it may have pointers into what @@ -540,6 +594,7 @@ do {\ memcpy((dst)->wstart, s.wstart,_wsz*sizeof(UWord));\ (dst)->wsp = (dst)->wstart + _wsz;\ (dst)->wend = (dst)->wstart + DEF_WSTACK_SIZE;\ + (dst)->wdefault = NULL;\ (dst)->alloc_type = s.alloc_type;\ } else\ *(dst) = s;\ @@ -568,12 +623,12 @@ do { \ ASSERT(s.wsp <= s.wend); \ } while (0) -#define WSTACK_IS_STATIC(s) (s.wstart == WSTK_DEF_STACK(s))) +#define WSTACK_IS_STATIC(s) (s.wstart == WSTK_DEF_STACK(s)) #define WSTACK_PUSH(s, x) \ do { \ if (s.wsp == s.wend) { \ - erl_grow_wstack(&s, WSTK_DEF_STACK(s)); \ + erl_grow_wstack(&s, 1); \ } \ *s.wsp++ = (x); \ } while(0) @@ -581,7 +636,7 @@ do { \ #define WSTACK_PUSH2(s, x, y) \ do { \ if (s.wsp > s.wend - 2) { \ - erl_grow_wstack(&s, WSTK_DEF_STACK(s)); \ + erl_grow_wstack(&s, 2); \ } \ *s.wsp++ = (x); \ *s.wsp++ = (y); \ @@ -589,17 +644,129 @@ do { \ #define WSTACK_PUSH3(s, x, y, z) \ do { \ - if (s.wsp > s.wend - 3) { \ - erl_grow_wstack(&s, WSTK_DEF_STACK(s)); \ + if (s.wsp > s.wend - 3) { \ + erl_grow_wstack(&s, 3); \ } \ *s.wsp++ = (x); \ *s.wsp++ = (y); \ *s.wsp++ = (z); \ } while(0) +#define WSTACK_PUSH4(s, A1, A2, A3, A4) \ +do { \ + if (s.wsp > s.wend - 4) { \ + erl_grow_wstack(&s, 4); \ + } \ + *s.wsp++ = (A1); \ + *s.wsp++ = (A2); \ + *s.wsp++ = (A3); \ + *s.wsp++ = (A4); \ +} while(0) + +#define WSTACK_PUSH5(s, A1, A2, A3, A4, A5) \ +do { \ + if (s.wsp > s.wend - 5) { \ + erl_grow_wstack(&s, 5); \ + } \ + *s.wsp++ = (A1); \ + *s.wsp++ = (A2); \ + *s.wsp++ = (A3); \ + *s.wsp++ = (A4); \ + *s.wsp++ = (A5); \ +} while(0) + +#define WSTACK_PUSH6(s, A1, A2, A3, A4, A5, A6) \ +do { \ + if (s.wsp > s.wend - 6) { \ + erl_grow_wstack(&s, 6); \ + } \ + *s.wsp++ = (A1); \ + *s.wsp++ = (A2); \ + *s.wsp++ = (A3); \ + *s.wsp++ = (A4); \ + *s.wsp++ = (A5); \ + *s.wsp++ = (A6); \ +} while(0) + +#define WSTACK_RESERVE(s, push_cnt) \ +do { \ + if (s.wsp > s.wend - (push_cnt)) { \ + erl_grow_wstack(&s, (push_cnt)); \ + } \ +} while(0) + +/* Must be preceded by WSTACK_RESERVE */ +#define WSTACK_FAST_PUSH(s, x) \ +do { \ + ASSERT(s.wsp < s.wend); \ + *s.wsp++ = (x); \ +} while(0) + #define WSTACK_COUNT(s) (s.wsp - s.wstart) #define WSTACK_ISEMPTY(s) (s.wsp == s.wstart) -#define WSTACK_POP(s) (*(--s.wsp)) +#define WSTACK_POP(s) ((ASSERT(s.wsp > s.wstart)),*(--s.wsp)) + +#define WSTACK_ROLLBACK(s, count) (ASSERT(WSTACK_COUNT(s) >= (count)), \ + s.wsp = s.wstart + (count)) + +/* PSTACK - Stack of any type. + * Usage: + * { + * #define PSTACK_TYPE MyType + * PSTACK_DECLARE(s,16); + * MyType *sp = PSTACK_PUSH(s); + * + * sp->x = .... + * sp->y = .... + * sp = PSTACK_PUSH(s); + * ... + * sp = PSTACK_POP(s); + * if (PSTACK_IS_EMPTY(s)) { + * // sp is invalid when stack is empty after pop + * } + * + * PSTACK_DESTROY(s); + * } + */ + + +typedef struct ErtsPStack_ { + byte* pstart; + byte* psp; + byte* pend; + ErtsAlcType_t alloc_type; +}ErtsPStack; + +void erl_grow_pstack(ErtsPStack* s, void* default_pstack, unsigned need_bytes); +#define PSTK_CONCAT(a,b) a##b +#define PSTK_DEF_STACK(s) PSTK_CONCAT(s,_default_pstack) + +#define PSTACK_DECLARE(s, DEF_PSTACK_SIZE) \ +PSTACK_TYPE PSTK_DEF_STACK(s)[DEF_PSTACK_SIZE]; \ +ErtsPStack s = { (byte*)PSTK_DEF_STACK(s), /* pstart */ \ + (byte*)(PSTK_DEF_STACK(s) - 1), /* psp */ \ + (byte*)(PSTK_DEF_STACK(s) + (DEF_PSTACK_SIZE)), /* pend */\ + ERTS_ALC_T_ESTACK /* alloc_type */ \ +} + +#define PSTACK_DESTROY(s) \ +do { \ + if (s.pstart != (byte*)PSTK_DEF_STACK(s)) { \ + erts_free(s.alloc_type, s.pstart); \ + } \ +} while(0) + +#define PSTACK_IS_EMPTY(s) (s.psp < s.pstart) + +#define PSTACK_TOP(s) (ASSERT(!PSTACK_IS_EMPTY(s)), (PSTACK_TYPE*)(s.psp)) + +#define PSTACK_PUSH(s) \ + (s.psp += sizeof(PSTACK_TYPE), \ + ((s.psp == s.pend) ? erl_grow_pstack(&s, PSTK_DEF_STACK(s), \ + sizeof(PSTACK_TYPE)) : (void)0), \ + ((PSTACK_TYPE*) s.psp)) + +#define PSTACK_POP(s) ((PSTACK_TYPE*) (s.psp -= sizeof(PSTACK_TYPE))) /* binary.c */ @@ -809,23 +976,6 @@ void MD5Init(MD5_CTX *); void MD5Update(MD5_CTX *, unsigned char *, unsigned int); void MD5Final(unsigned char [16], MD5_CTX *); -/* ggc.c */ - -void erts_gc_info(ErtsGCInfo *gcip); -void erts_init_gc(void); -int erts_garbage_collect(Process*, int, Eterm*, int); -void erts_garbage_collect_hibernate(Process* p); -Eterm erts_gc_after_bif_call(Process* p, Eterm result, Eterm* regs, Uint arity); -void erts_garbage_collect_literals(Process* p, Eterm* literals, - Uint lit_size, - struct erl_off_heap_header* oh); -Uint erts_next_heap_size(Uint, Uint); -Eterm erts_heap_sizes(Process* p); - -void erts_offset_off_heap(ErlOffHeap *, Sint, Eterm*, Eterm*); -void erts_offset_heap_ptr(Eterm*, Uint, Sint, Eterm*, Eterm*); -void erts_offset_heap(Eterm*, Uint, Sint, Eterm*, Eterm*); -void erts_free_heap_frags(Process* p); /* io.c */ @@ -852,6 +1002,7 @@ Uint erts_port_ioq_size(Port *pp); void erts_stale_drv_select(Eterm, ErlDrvPort, ErlDrvEvent, int, int); Port *erts_get_heart_port(void); +void erts_emergency_close_ports(void); #if defined(ERTS_SMP) && defined(ERTS_ENABLE_LOCK_COUNT) void erts_lcnt_enable_io_lock_count(int enable); diff --git a/erts/emulator/beam/io.c b/erts/emulator/beam/io.c index 9ae973e108..fb0edbcb1a 100644 --- a/erts/emulator/beam/io.c +++ b/erts/emulator/beam/io.c @@ -391,7 +391,7 @@ static Port *create_port(char *name, /* Set default tracing */ erts_get_default_tracing(&ERTS_TRACE_FLAGS(prt), &ERTS_TRACER_PROC(prt)); - ASSERT(((char *) prt) == ((char *) &prt->common)); + ERTS_CT_ASSERT(offsetof(Port,common) == 0); #if !ERTS_PORT_INIT_INSTR_NEED_ID /* @@ -3153,8 +3153,6 @@ static void deliver_read_message(Port* prt, erts_aint32_t state, Eterm to, Binary* bptr; bptr = erts_bin_nrml_alloc(len); - bptr->flags = 0; - bptr->orig_size = len; erts_refc_init(&bptr->refc, 1); sys_memcpy(bptr->orig_bytes, buf, len); @@ -4087,6 +4085,9 @@ erts_port_control(Process* c_p, size, &resp_bufp, &resp_size); + + control_flags = prt->control_flags; + finalize_imm_drv_call(&try_call_state); if (tmp_alloced) erts_free(ERTS_ALC_T_TMP, bufp); @@ -4094,8 +4095,6 @@ erts_port_control(Process* c_p, return ERTS_PORT_OP_BADARG; } - control_flags = prt->control_flags; - hsz = port_control_result_size(control_flags, resp_bufp, &resp_size, @@ -5351,7 +5350,11 @@ driver_deliver_term(Eterm to, ErlDrvTermData* data, int len) 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]; + if (ptr[0] > MAP_SMALL_MAP_LIMIT) { + need += hashmap_over_estimated_heap_size(ptr[0]); + } else { + need += MAP_HEADER_SIZE + 1 + 2*ptr[0]; + } depth -= 2*ptr[0]; if (depth < 0) ERTS_DDT_FAIL; ptr++; @@ -5506,8 +5509,6 @@ driver_deliver_term(Eterm to, ErlDrvTermData* data, int len) ProcBin* pbp; Binary* bp = erts_bin_nrml_alloc(size); ASSERT(bufp); - bp->flags = 0; - bp->orig_size = (SWord) size; erts_refc_init(&bp->refc, 1); sys_memcpy((void *) bp->orig_bytes, (void *) bufp, size); pbp = (ProcBin *) hp; @@ -5597,31 +5598,52 @@ driver_deliver_term(Eterm to, ErlDrvTermData* data, int len) 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++; + if (size > MAP_SMALL_MAP_LIMIT) { + int ix = 2*size; + ErtsHeapFactory factory; + Eterm* leafs = hp; + + hp += 2*size; + while(ix--) { *--hp = ESTACK_POP(stack); } + + hp += 2*size; + factory.p = NULL; + factory.hp = hp; + /* We assume heap will suffice (see hashmap_over_estimated_heap_size) */ + + mess = erts_hashmap_from_array(&factory, leafs, size, 1); + + if (is_non_value(mess)) + ERTS_DDT_FAIL; + + hp = factory.hp; + } else { + Eterm* tp = hp; + Eterm* vp; + flatmap_t *mp; + + *tp = make_arityval(size); + + hp += 1 + size; + mp = (flatmap_t*)hp; + mp->thing_word = MAP_HEADER; + mp->size = size; + mp->keys = make_tuple(tp); + mess = make_flatmap(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_flatmap(mp)) + ERTS_DDT_FAIL; + } + ptr++; break; } @@ -5999,9 +6021,7 @@ driver_alloc_binary(ErlDrvSizeT size) bin = erts_bin_drv_alloc_fnf((Uint) size); if (!bin) return NULL; /* The driver write must take action */ - bin->flags = BIN_FLAG_DRV; erts_refc_init(&bin->refc, 1); - bin->orig_size = (SWord) size; return Binary2ErlDrvBinary(bin); } @@ -6031,7 +6051,6 @@ ErlDrvBinary* driver_realloc_binary(ErlDrvBinary* bin, ErlDrvSizeT size) if (!newbin) return NULL; - newbin->orig_size = size; return Binary2ErlDrvBinary(newbin); } @@ -6705,7 +6724,7 @@ static void ref_to_driver_monitor(Eterm ref, ErlDrvMonitor *mon) { RefThing *refp; ASSERT(is_internal_ref(ref)); - ASSERT(sizeof(RefThing) <= sizeof(ErlDrvMonitor)); + ERTS_CT_ASSERT(sizeof(RefThing) <= sizeof(ErlDrvMonitor)); refp = ref_thing_ptr(ref); memset(mon,0,sizeof(ErlDrvMonitor)); memcpy(mon,refp,sizeof(RefThing)); @@ -7349,6 +7368,8 @@ no_stop_select_callback(ErlDrvEvent event, void* private) erts_send_error_to_logger_nogl(dsbufp); } +#define IS_DRIVER_VERSION_GE(DE,MAJOR,MINOR) \ + ((DE)->major_version >= (MAJOR) && (DE)->minor_version >= (MINOR)) static int init_driver(erts_driver_t *drv, ErlDrvEntry *de, DE_Handle *handle) @@ -7396,6 +7417,7 @@ init_driver(erts_driver_t *drv, ErlDrvEntry *de, DE_Handle *handle) drv->timeout = de->timeout ? de->timeout : no_timeout_callback; drv->ready_async = de->ready_async; drv->process_exit = de->process_exit; + drv->emergency_close = IS_DRIVER_VERSION_GE(de,3,2) ? de->emergency_close : NULL; if (de->stop_select) drv->stop_select = de->stop_select; else @@ -7414,6 +7436,8 @@ init_driver(erts_driver_t *drv, ErlDrvEntry *de, DE_Handle *handle) } } +#undef IS_DRIVER_VERSION_GE + void erts_destroy_driver(erts_driver_t *drv) { @@ -7557,7 +7581,7 @@ Port *erts_get_heart_port(void) if (!port) continue; /* only examine undead or alive ports */ - if (erts_atomic32_read_nob(&port->state) & ERTS_PORT_SFLGS_DEAD) + if (erts_atomic32_read_nob(&port->state) & ERTS_PORT_SFLGS_INVALID_DRIVER_LOOKUP) continue; /* immediate atom compare */ reg = port->common.u.alive.reg; @@ -7568,3 +7592,23 @@ Port *erts_get_heart_port(void) return NULL; } + +void erts_emergency_close_ports(void) +{ + int ix, max = erts_ptab_max(&erts_port); + + for (ix = 0; ix < max; ix++) { + Port *port = erts_pix2port(ix); + + if (!port) + continue; + /* only examine undead or alive ports */ + if (erts_atomic32_read_nob(&port->state) & ERTS_PORT_SFLGS_INVALID_DRIVER_LOOKUP) + continue; + + /* emergency close socket */ + if (port->drv_ptr->emergency_close) { + port->drv_ptr->emergency_close((ErlDrvData) port->drv_data); + } + } +} diff --git a/erts/emulator/beam/ops.tab b/erts/emulator/beam/ops.tab index 68fcc177ae..d3649080dc 100644 --- a/erts/emulator/beam/ops.tab +++ b/erts/emulator/beam/ops.tab @@ -166,22 +166,26 @@ is_tuple Fail=f S | select_tuple_arity S=d Fail=f Size=u Rest=* => \ select_tuple_arity S=d Fail=f Size=u Rest=* => \ gen_select_tuple_arity(S, Fail, Size, Rest) -i_select_val r f I -i_select_val x f I -i_select_val y f I +i_select_val_bins r f I +i_select_val_bins x f I +i_select_val_bins y f I -i_select_val2 r f c f c f -i_select_val2 x f c f c f -i_select_val2 y f c f c f +i_select_val_lins r f I +i_select_val_lins x f I +i_select_val_lins y f I -i_select_tuple_arity2 r f A f A f -i_select_tuple_arity2 x f A f A f -i_select_tuple_arity2 y f A f A f +i_select_val2 r f c c f f +i_select_val2 x f c c f f +i_select_val2 y f c c f f i_select_tuple_arity r f I i_select_tuple_arity x f I i_select_tuple_arity y f I +i_select_tuple_arity2 r f A A f f +i_select_tuple_arity2 x f A A f f +i_select_tuple_arity2 y f A A f f + i_jump_on_val_zero r f I i_jump_on_val_zero x f I i_jump_on_val_zero y f I diff --git a/erts/emulator/beam/sys.h b/erts/emulator/beam/sys.h index c29d4b3777..d2a8b9e7f4 100644 --- a/erts/emulator/beam/sys.h +++ b/erts/emulator/beam/sys.h @@ -188,6 +188,32 @@ __decl_noreturn void __noreturn erl_assert_error(const char* expr, const char *f # define ASSERT(e) ((void) 1) #endif +/* ERTS_UNDEF can be used to silence false warnings about + * "variable may be used uninitialized" while keeping the variable + * marked as undefined by valgrind. + */ +#ifdef VALGRIND +# define ERTS_UNDEF(V,I) +#else +# define ERTS_UNDEF(V,I) V = I +#endif + +/* + * Compile time assert + * (the actual compiler error msg can be a bit confusing) + */ +#if ERTS_AT_LEAST_GCC_VSN__(3,1,1) +# define ERTS_CT_ASSERT(e) \ + do { \ + enum { compile_time_assert__ = __builtin_choose_expr((e),0,(void)0) }; \ + } while(0) +#else +# define ERTS_CT_ASSERT(e) \ + do { \ + enum { compile_time_assert__ = 1/(e) }; \ + } while (0) +#endif + /* * Microsoft C/C++: We certainly want to use stdarg.h and prototypes. * But MSC doesn't define __STDC__, unless we compile with the -Za @@ -756,6 +782,8 @@ typedef struct { } ErtsCheckIoDebugInfo; int erts_check_io_debug(ErtsCheckIoDebugInfo *ip); +int erts_sys_is_area_readable(char *start, char *stop); + /* xxxP */ #define SYS_DEFAULT_FLOAT_DECIMALS 20 void init_sys_float(void); @@ -784,6 +812,11 @@ int erts_sys_unsetenv(char *key); char *erts_read_env(char *key); void erts_free_read_env(void *value); +#if defined(ERTS_THR_HAVE_SIG_FUNCS) && !defined(ETHR_UNUSABLE_SIGUSRX) +extern void sys_thr_resume(erts_tid_t tid); +extern void sys_thr_suspend(erts_tid_t tid); +#endif + /* utils.c */ /* Options to sys_alloc_opt */ diff --git a/erts/emulator/beam/utils.c b/erts/emulator/beam/utils.c index f20e6e5665..3549e18538 100644 --- a/erts/emulator/beam/utils.c +++ b/erts/emulator/beam/utils.c @@ -49,6 +49,9 @@ #include "beam_bp.h" #include "erl_ptab.h" #include "erl_check_io.h" +#ifdef HIPE +# include "hipe_mode_switch.h" +#endif #undef M_TRIM_THRESHOLD #undef M_TOP_PAD @@ -187,12 +190,18 @@ erts_set_hole_marker(Eterm* ptr, Uint sz) * Helper function for the ESTACK macros defined in global.h. */ void -erl_grow_estack(ErtsEStack* s, Eterm* default_estack) +erl_grow_estack(ErtsEStack* s, Uint need) { Uint old_size = (s->end - s->start); - Uint new_size = old_size * 2; + Uint new_size; Uint sp_offs = s->sp - s->start; - if (s->start != default_estack) { + + if (need < old_size) + new_size = 2*old_size; + else + new_size = ((need / old_size) + 2) * old_size; + + if (s->start != s->edefault) { s->start = erts_realloc(s->alloc_type, s->start, new_size*sizeof(Eterm)); } else { @@ -207,12 +216,18 @@ erl_grow_estack(ErtsEStack* s, Eterm* default_estack) * Helper function for the WSTACK macros defined in global.h. */ void -erl_grow_wstack(ErtsWStack* s, UWord* default_wstack) +erl_grow_wstack(ErtsWStack* s, Uint need) { Uint old_size = (s->wend - s->wstart); - Uint new_size = old_size * 2; + Uint new_size; Uint sp_offs = s->wsp - s->wstart; - if (s->wstart != default_wstack) { + + if (need < old_size) + new_size = 2 * old_size; + else + new_size = ((need / old_size) + 2) * old_size; + + if (s->wstart != s->wdefault) { s->wstart = erts_realloc(s->alloc_type, s->wstart, new_size*sizeof(UWord)); } else { @@ -224,6 +239,32 @@ erl_grow_wstack(ErtsWStack* s, UWord* default_wstack) s->wsp = s->wstart + sp_offs; } +/* + * Helper function for the PSTACK macros defined in global.h. + */ +void +erl_grow_pstack(ErtsPStack* s, void* default_pstack, unsigned need_bytes) +{ + Uint old_size = s->pend - s->pstart; + Uint new_size; + Uint sp_offs = s->psp - s->pstart; + + if (need_bytes < old_size) + new_size = 2 * old_size; + else + new_size = ((need_bytes / old_size) + 2) * old_size; + + if (s->pstart != default_pstack) { + s->pstart = erts_realloc(s->alloc_type, s->pstart, new_size); + } else { + byte* new_ptr = erts_alloc(s->alloc_type, new_size); + sys_memcpy(new_ptr, s->pstart, old_size); + s->pstart = new_ptr; + } + s->pend = s->pstart + new_size; + s->psp = s->pstart + sp_offs; +} + /* CTYPE macros */ #define LATIN1 @@ -311,6 +352,17 @@ int erts_fit_in_bits_int32(Sint32 value) return fit_in_bits((Sint64) (Uint32) value, 4); } +int erts_fit_in_bits_uint(Uint value) +{ +#if ERTS_SIZEOF_ETERM == 4 + return fit_in_bits((Sint64) (Uint32) value, 4); +#elif ERTS_SIZEOF_ETERM == 8 + return fit_in_bits(value, 5); +#else +# error "No way, Jose" +#endif +} + int erts_print(int to, void *arg, char *format, ...) { @@ -707,7 +759,7 @@ erts_bld_atom_2uint_3tup_list(Uint **hpp, Uint *szp, Sint length, ** If N < 0, Y = FUNNY_NUMBER4 else Y = FUNNY_NUMBER3. ** The hash value is Y*h(J) mod 2^32 where h(J) is calculated like ** h(0) = <initial hash> -** h(i) = h(i-i)*X + B(i-1) +** h(i) = h(i-1)*X + B(i-1) ** The above should hold regardless of internal representation. ** Pids are hashed like small numbers but with differrent constants, as are ** ports. @@ -789,10 +841,10 @@ Uint32 make_hash(Eterm term_arg) unsigned op; /* Must not collide with the real tag_val_def's: */ -#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 +#define MAKE_HASH_TUPLE_OP (FIRST_VACANT_TAG_DEF) +#define MAKE_HASH_TERM_ARRAY_OP (FIRST_VACANT_TAG_DEF+1) +#define MAKE_HASH_CDR_PRE_OP (FIRST_VACANT_TAG_DEF+2) +#define MAKE_HASH_CDR_POST_OP (FIRST_VACANT_TAG_DEF+3) /* ** Convenience macro for calculating a bytewise hash on an unsigned 32 bit @@ -972,21 +1024,9 @@ tail_recur: break; } case MAP_DEF: + case HASHMAP_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); + hash = hash*FUNNY_NUMBER13 + FUNNY_NUMBER14 + make_hash2(term); break; } case TUPLE_DEF: @@ -1092,10 +1132,11 @@ Uint32 make_hash2(Eterm term) { Uint32 hash; - Uint32 hash_xor_keys = 0; - Uint32 hash_xor_values = 0; + Uint32 hash_xor_pairs; DeclareTmpHeapNoproc(tmp_big,2); + ERTS_UNDEF(hash_xor_pairs, 0); + /* (HCONST * {2, ..., 16}) mod 2^32 */ #define HCONST_2 0x3c6ef372UL #define HCONST_3 0xdaa66d2bUL @@ -1112,10 +1153,15 @@ make_hash2(Eterm term) #define HCONST_14 0xa708a81eUL #define HCONST_15 0x454021d7UL #define HCONST_16 0xe3779b90UL +#define HCONST_17 0x81af1549UL +#define HCONST_18 0x1fe68f02UL +#define HCONST_19 0xbe1e08bbUL +#define HCONST_20 0x5c558274UL +#define HCONST_21 0xfa8cfc2dUL #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 HASH_MAP_PAIR (_make_header(2,_TAG_HEADER_REF)) +#define HASH_CDR (_make_header(3,_TAG_HEADER_REF)) #define UINT32_HASH_2(Expr1, Expr2, AConst) \ do { \ @@ -1138,6 +1184,13 @@ make_hash2(Eterm term) } while(0) #define IS_SSMALL28(x) (((Uint) (((x) >> (28-1)) + 1)) < 2) + +#ifdef ARCH_64 +# define POINTER_HASH(Ptr, AConst) UINT32_HASH_2((Uint32)(UWord)(Ptr), (((UWord)(Ptr)) >> 32), AConst) +#else +# define POINTER_HASH(Ptr, AConst) UINT32_HASH(Ptr, AConst) +#endif + /* Optimization. Simple cases before declaration of estack. */ if (primary_tag(term) == TAG_PRIMARY_IMMED1) { switch (term & _TAG_IMMED1_MASK) { @@ -1192,9 +1245,9 @@ make_hash2(Eterm term) if (c > 0) UINT32_HASH(sh, HCONST_4); if (is_list(term)) { - term = *ptr; - tmp = *++ptr; - ESTACK_PUSH(s, tmp); + tmp = CDR(ptr); + ESTACK_PUSH(s, tmp); + term = CAR(ptr); } } break; @@ -1211,51 +1264,92 @@ make_hash2(Eterm term) UINT32_HASH(arity, HCONST_9); if (arity == 0) /* Empty tuple */ goto hash2_common; - for (i = arity; i >= 1; i--) { - tmp = elem[i]; - ESTACK_PUSH(s, tmp); + for (i = arity; ; i--) { + term = elem[i]; + if (i == 1) + break; + ESTACK_PUSH(s, term); } - goto hash2_common; } break; case MAP_SUBTAG: { - map_t *mp = (map_t *)map_val(term); + flatmap_t *mp = (flatmap_t *)flatmap_val(term); int i; - int size = map_get_size(mp); - Eterm *ks = map_get_keys(mp); - Eterm *vs = map_get_values(mp); + int size = flatmap_get_size(mp); + Eterm *ks = flatmap_get_keys(mp); + Eterm *vs = flatmap_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); + /* We want a portable hash function that is *independent* of + * the order in which keys and values are encountered. + * We therefore calculate context independent hashes for all . + * key-value pairs and then xor them together. + */ + ESTACK_PUSH(s, hash_xor_pairs); 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. - */ + hash_xor_pairs = 0; for (i = size - 1; i >= 0; i--) { - tmp = ks[i]; - ESTACK_PUSH(s, HASH_MAP_KEY); - ESTACK_PUSH(s, tmp); + ESTACK_PUSH(s, HASH_MAP_PAIR); + ESTACK_PUSH(s, vs[i]); + ESTACK_PUSH(s, ks[i]); } goto hash2_common; } break; + case HASHMAP_SUBTAG: + { + Eterm* ptr = boxed_val(term) + 1; + Uint size; + int i; + switch (hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: + case HAMT_SUBTAG_HEAD_BITMAP: + size = *ptr++; + UINT32_HASH(size, HCONST_16); + if (size == 0) + goto hash2_common; + ESTACK_PUSH(s, hash_xor_pairs); + ESTACK_PUSH(s, hash); + ESTACK_PUSH(s, HASH_MAP_TAIL); + hash = 0; + hash_xor_pairs = 0; + } + switch (hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: + case HAMT_SUBTAG_NODE_ARRAY: + i = 16; + break; + case HAMT_SUBTAG_HEAD_BITMAP: + case HAMT_SUBTAG_NODE_BITMAP: + i = hashmap_bitcount(MAP_HEADER_VAL(hdr)); + break; + default: + erl_exit(1, "bad header"); + } + while (i) { + if (is_list(*ptr)) { + Eterm* cons = list_val(*ptr); + ESTACK_PUSH(s, HASH_MAP_PAIR); + ESTACK_PUSH(s, CDR(cons)); + ESTACK_PUSH(s, CAR(cons)); + } + else { + ASSERT(is_boxed(*ptr)); + ESTACK_PUSH(s, *ptr); + } + i--; ptr++; + } + goto hash2_common; + } + break; case EXPORT_SUBTAG: { Export* ep = *((Export **) (export_val(term) + 1)); - UINT32_HASH_2 (ep->code[2], atom_tab(atom_val(ep->code[0]))->slot.bucket.hvalue, @@ -1270,7 +1364,6 @@ make_hash2(Eterm term) { ErlFunThing* funp = (ErlFunThing *) fun_val(term); Uint num_free = funp->num_free; - UINT32_HASH_2 (num_free, atom_tab(atom_val(funp->fe->module))->slot.bucket.hvalue, @@ -1351,7 +1444,8 @@ make_hash2(Eterm term) do { Uint t; Uint32 x, y; - t = i < n ? BIG_DIGIT(ptr, i++) : 0; + ASSERT(i < n); + t = BIG_DIGIT(ptr, i++); x = t & 0xffffffff; y = t >> 32; UINT32_HASH_2(x, y, con); @@ -1459,20 +1553,397 @@ make_hash2(Eterm term) 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); + UINT32_HASH(hash_xor_pairs, HCONST_19); + hash_xor_pairs = (Uint32) ESTACK_POP(s); goto hash2_common; } - case HASH_MAP_KEY: - hash_xor_keys ^= hash; + case HASH_MAP_PAIR: + hash_xor_pairs ^= hash; hash = 0; goto hash2_common; - case HASH_MAP_VAL: - hash_xor_values ^= hash; + default: + break; + } + } + } + } +} + +/* Term hash function for internal use. + * + * Limitation #1: Is not "portable" in any way between different VM instances. + * + * Limitation #2: The hash value is only valid as long as the term exists + * somewhere in the VM. Why? Because external pids, ports and refs are hashed + * by mixing the node *pointer* value. If a node disappears and later reappears + * with a new ErlNode struct, externals from that node will hash different than + * before. + * + * One IMPORTANT property must hold (for hamt). + * EVERY BIT of the term that is significant for equality (see EQ) + * MUST BE USED AS INPUT FOR THE HASH. Two different terms must always have a + * chance of hashing different when salted: hash([Salt|A]) vs hash([Salt|B]). + * + * This is why we can not use cached hash values for atoms for example. + * + */ + +#define CONST_HASH(AConst) \ +do { /* Lightweight mixing of constant (type info) */ \ + hash ^= AConst; \ + hash = (hash << 17) ^ (hash >> (32-17)); \ +} while (0) + +Uint32 +make_internal_hash(Eterm term) +{ + Uint32 hash; + Uint32 hash_xor_pairs; + + ERTS_UNDEF(hash_xor_pairs, 0); + + /* Optimization. Simple cases before declaration of estack. */ + if (primary_tag(term) == TAG_PRIMARY_IMMED1) { + hash = 0; + #if ERTS_SIZEOF_ETERM == 8 + UINT32_HASH_2((Uint32)term, (Uint32)(term >> 32), HCONST); + #elif ERTS_SIZEOF_ETERM == 4 + UINT32_HASH(term, HCONST); + #else + # error "No you don't" + #endif + return hash; + } + { + Eterm tmp; + DECLARE_ESTACK(s); + + UseTmpHeapNoproc(2); + hash = 0; + for (;;) { + switch (primary_tag(term)) { + case TAG_PRIMARY_LIST: + { + int c = 0; + Uint32 sh = 0; + Eterm* ptr = list_val(term); + while (is_byte(*ptr)) { + /* Optimization for strings. */ + sh = (sh << 8) + unsigned_val(*ptr); + if (c == 3) { + UINT32_HASH(sh, HCONST_4); + c = sh = 0; + } else { + c++; + } + term = CDR(ptr); + if (is_not_list(term)) + break; + ptr = list_val(term); + } + if (c > 0) + UINT32_HASH(sh, HCONST_4); + if (is_list(term)) { + tmp = CDR(ptr); + CONST_HASH(HCONST_17); /* Hash CAR in cons cell */ + ESTACK_PUSH(s, tmp); + if (is_not_list(tmp)) { + ESTACK_PUSH(s, HASH_CDR); + } + term = CAR(ptr); + } + } + break; + case TAG_PRIMARY_BOXED: + { + Eterm hdr = *boxed_val(term); + ASSERT(is_header(hdr)); + switch (hdr & _TAG_HEADER_MASK) { + case ARITYVAL_SUBTAG: + { + int i; + int arity = header_arity(hdr); + Eterm* elem = tuple_val(term); + UINT32_HASH(arity, HCONST_9); + if (arity == 0) /* Empty tuple */ + goto pop_next; + for (i = arity; ; i--) { + term = elem[i]; + if (i == 1) + break; + ESTACK_PUSH(s, term); + } + } + break; + case MAP_SUBTAG: + { + flatmap_t *mp = (flatmap_t *)flatmap_val(term); + int i; + int size = flatmap_get_size(mp); + Eterm *ks = flatmap_get_keys(mp); + Eterm *vs = flatmap_get_values(mp); + UINT32_HASH(size, HCONST_16); + if (size == 0) { + goto pop_next; + } + /* We want a hash function that is *independent* of + * the order in which keys and values are encountered. + * We therefore calculate context independent hashes for all . + * key-value pairs and then xor them together. + */ + ESTACK_PUSH(s, hash_xor_pairs); + ESTACK_PUSH(s, hash); + ESTACK_PUSH(s, HASH_MAP_TAIL); + hash = 0; + hash_xor_pairs = 0; + for (i = size - 1; i >= 0; i--) { + ESTACK_PUSH(s, HASH_MAP_PAIR); + ESTACK_PUSH(s, vs[i]); + ESTACK_PUSH(s, ks[i]); + } + goto pop_next; + } + break; + case HASHMAP_SUBTAG: + { + Eterm* ptr = boxed_val(term) + 1; + Uint size; + int i; + switch (hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: + case HAMT_SUBTAG_HEAD_BITMAP: + size = *ptr++; + UINT32_HASH(size, HCONST_16); + if (size == 0) + goto pop_next; + ESTACK_PUSH(s, hash_xor_pairs); + ESTACK_PUSH(s, hash); + ESTACK_PUSH(s, HASH_MAP_TAIL); + hash = 0; + hash_xor_pairs = 0; + } + switch (hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: + case HAMT_SUBTAG_NODE_ARRAY: + i = 16; + break; + case HAMT_SUBTAG_HEAD_BITMAP: + case HAMT_SUBTAG_NODE_BITMAP: + i = hashmap_bitcount(MAP_HEADER_VAL(hdr)); + break; + default: + erl_exit(1, "bad header"); + } + while (i) { + if (is_list(*ptr)) { + Eterm* cons = list_val(*ptr); + ESTACK_PUSH(s, HASH_MAP_PAIR); + ESTACK_PUSH(s, CDR(cons)); + ESTACK_PUSH(s, CAR(cons)); + } + else { + ASSERT(is_boxed(*ptr)); + ESTACK_PUSH(s, *ptr); + } + i--; ptr++; + } + goto pop_next; + } + break; + case EXPORT_SUBTAG: + { + Export* ep = *((Export **) (export_val(term) + 1)); + /* Assumes Export entries never moves */ + POINTER_HASH(ep, HCONST_14); + goto pop_next; + } + + case FUN_SUBTAG: + { + ErlFunThing* funp = (ErlFunThing *) fun_val(term); + Uint num_free = funp->num_free; + UINT32_HASH_2(num_free, funp->fe->module, HCONST_20); + UINT32_HASH_2(funp->fe->old_index, funp->fe->old_uniq, HCONST_21); + if (num_free == 0) { + goto pop_next; + } else { + Eterm* bptr = funp->env + num_free - 1; + while (num_free-- > 1) { + term = *bptr--; + ESTACK_PUSH(s, term); + } + term = *bptr; + } + } + break; + case REFC_BINARY_SUBTAG: + case HEAP_BINARY_SUBTAG: + case SUB_BINARY_SUBTAG: + { + byte* bptr; + unsigned sz = binary_size(term); + Uint32 con = HCONST_13 + hash; + Uint bitoffs; + Uint bitsize; + + ERTS_GET_BINARY_BYTES(term, bptr, bitoffs, bitsize); + if (sz == 0 && bitsize == 0) { + hash = con; + } else { + if (bitoffs == 0) { + hash = block_hash(bptr, sz, con); + if (bitsize > 0) { + UINT32_HASH_2(bitsize, (bptr[sz] >> (8 - bitsize)), + HCONST_15); + } + } else { + byte* buf = (byte *) erts_alloc(ERTS_ALC_T_TMP, + sz + (bitsize != 0)); + erts_copy_bits(bptr, bitoffs, 1, buf, 0, 1, sz*8+bitsize); + hash = block_hash(buf, sz, con); + if (bitsize > 0) { + UINT32_HASH_2(bitsize, (buf[sz] >> (8 - bitsize)), + HCONST_15); + } + erts_free(ERTS_ALC_T_TMP, (void *) buf); + } + } + goto pop_next; + } + break; + case POS_BIG_SUBTAG: + case NEG_BIG_SUBTAG: + { + Eterm* ptr = big_val(term); + Uint i = 0; + Uint n = BIG_SIZE(ptr); + Uint32 con = BIG_SIGN(ptr) ? HCONST_10 : HCONST_11; +#if D_EXP == 16 + do { + Uint32 x, y; + x = i < n ? BIG_DIGIT(ptr, i++) : 0; + x += (Uint32)(i < n ? BIG_DIGIT(ptr, i++) : 0) << 16; + y = i < n ? BIG_DIGIT(ptr, i++) : 0; + y += (Uint32)(i < n ? BIG_DIGIT(ptr, i++) : 0) << 16; + UINT32_HASH_2(x, y, con); + } while (i < n); +#elif D_EXP == 32 + do { + Uint32 x, y; + x = i < n ? BIG_DIGIT(ptr, i++) : 0; + y = i < n ? BIG_DIGIT(ptr, i++) : 0; + UINT32_HASH_2(x, y, con); + } while (i < n); +#elif D_EXP == 64 + do { + Uint t; + Uint32 x, y; + ASSERT(i < n); + t = BIG_DIGIT(ptr, i++); + x = t & 0xffffffff; + y = t >> 32; + UINT32_HASH_2(x, y, con); + } while (i < n); +#else +#error "unsupported D_EXP size" +#endif + goto pop_next; + } + break; + case REF_SUBTAG: + UINT32_HASH(internal_ref_numbers(term)[0], HCONST_7); + ASSERT(internal_ref_no_of_numbers(term) == 3); + UINT32_HASH_2(internal_ref_numbers(term)[1], + internal_ref_numbers(term)[2], HCONST_8); + goto pop_next; + + case EXTERNAL_REF_SUBTAG: + { + ExternalThing* thing = external_thing_ptr(term); + + ASSERT(external_thing_ref_no_of_numbers(thing) == 3); + /* See limitation #2 */ + #ifdef ARCH_64 + POINTER_HASH(thing->node, HCONST_7); + UINT32_HASH(external_thing_ref_numbers(thing)[0], HCONST_7); + #else + UINT32_HASH_2(thing->node, + external_thing_ref_numbers(thing)[0], HCONST_7); + #endif + UINT32_HASH_2(external_thing_ref_numbers(thing)[1], + external_thing_ref_numbers(thing)[2], HCONST_8); + goto pop_next; + } + case EXTERNAL_PID_SUBTAG: { + ExternalThing* thing = external_thing_ptr(term); + /* See limitation #2 */ + #ifdef ARCH_64 + POINTER_HASH(thing->node, HCONST_5); + UINT32_HASH(thing->data.ui[0], HCONST_5); + #else + UINT32_HASH_2(thing->node, thing->data.ui[0], HCONST_5); + #endif + goto pop_next; + } + case EXTERNAL_PORT_SUBTAG: { + ExternalThing* thing = external_thing_ptr(term); + /* See limitation #2 */ + #ifdef ARCH_64 + POINTER_HASH(thing->node, HCONST_6); + UINT32_HASH(thing->data.ui[0], HCONST_6); + #else + UINT32_HASH_2(thing->node, thing->data.ui[0], HCONST_6); + #endif + goto pop_next; + } + case FLOAT_SUBTAG: + { + FloatDef ff; + GET_DOUBLE(term, ff); + UINT32_HASH_2(ff.fw[0], ff.fw[1], HCONST_12); + goto pop_next; + } + + default: + erl_exit(1, "Invalid tag in make_hash2(0x%X)\n", term); + } + } + break; + case TAG_PRIMARY_IMMED1: + #if ERTS_SIZEOF_ETERM == 8 + UINT32_HASH_2((Uint32)term, (Uint32)(term >> 32), HCONST); + #else + UINT32_HASH(term, HCONST); + #endif + goto pop_next; + + default: + erl_exit(1, "Invalid tag in make_hash2(0x%X)\n", term); + + pop_next: + 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_pairs, HCONST_19); + hash_xor_pairs = (Uint32) ESTACK_POP(s); + goto pop_next; + } + case HASH_MAP_PAIR: + hash_xor_pairs ^= hash; hash = 0; - goto hash2_common; + goto pop_next; + + case HASH_CDR: + CONST_HASH(HCONST_18); /* Hash CDR i cons cell */ + goto pop_next; default: break; } @@ -1480,9 +1951,10 @@ make_hash2(Eterm term) } } +#undef CONST_HASH #undef HASH_MAP_TAIL -#undef HASH_MAP_KEY -#undef HASH_MAP_VAL +#undef HASH_MAP_PAIR +#undef HASH_CDR #undef UINT32_HASH_2 #undef UINT32_HASH @@ -1699,21 +2171,9 @@ tail_recur: break; case MAP_DEF: + case HASHMAP_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); + hash = hash*FUNNY_NUMBER13 + FUNNY_NUMBER14 + make_hash2(term); break; } case TUPLE_DEF: @@ -2122,13 +2582,13 @@ tailrecur_ne: } case MAP_SUBTAG: { - aa = map_val_rel(a, a_base); + aa = flatmap_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); - sz = map_get_size((map_t*)aa); + bb = flatmap_val_rel(b,b_base); + sz = flatmap_get_size((flatmap_t*)aa); - if (sz != map_get_size((map_t*)bb)) goto not_equal; + if (sz != flatmap_get_size((flatmap_t*)bb)) goto not_equal; if (sz == 0) goto pop_next; aa += 2; @@ -2327,6 +2787,32 @@ tailrecur_ne: } break; /* not equal */ } + case HASHMAP_SUBTAG: + { + if (!is_boxed(b) || *boxed_val_rel(b,b_base) != hdr) + goto not_equal; + + aa = hashmap_val_rel(a, a_base) + 1; + bb = hashmap_val_rel(b, b_base) + 1; + switch (hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: + aa++; bb++; + case HAMT_SUBTAG_NODE_ARRAY: + sz = 16; + break; + case HAMT_SUBTAG_HEAD_BITMAP: + aa++; bb++; + case HAMT_SUBTAG_NODE_BITMAP: + sz = hashmap_bitcount(MAP_HEADER_VAL(hdr)); + ASSERT(sz > 0 && sz < 16); + break; + default: + erl_exit(1, "Unknown hashmap subsubtag\n"); + } + goto term_array; + } + default: + ASSERT(!"Unknown boxed subtab in EQ"); } break; } @@ -2450,7 +2936,18 @@ Sint erts_cmp_rel_opt(Eterm a, Eterm* a_base, Eterm b, Eterm* b_base, int exact) Sint erts_cmp(Eterm a, Eterm b, int exact) #endif { - DECLARE_WSTACK(stack); +#define PSTACK_TYPE struct erts_cmp_hashmap_state + struct erts_cmp_hashmap_state { + Sint wstack_rollback; + int was_exact; + Eterm *ap; + Eterm *bp; + Eterm min_key; + Sint cmp_res; /* result so far -1,0,+1 */ + }; + PSTACK_DECLARE(hmap_stack, 1); + WSTACK_DECLARE(stack); + WSTACK_DECLARE(b_stack); /* only used by hashmaps */ Eterm* aa; Eterm* bb; int i; @@ -2466,6 +2963,26 @@ Sint erts_cmp(Eterm a, Eterm b, int exact) Uint32 *anum; Uint32 *bnum; +/* The WSTACK contains naked Eterms and Operations marked with header-tags */ +#define OP_BITS 4 +#define OP_MASK 0xF +#define TERM_ARRAY_OP 0 +#define SWITCH_EXACT_OFF_OP 1 +#define HASHMAP_PHASE1_ARE_KEYS_EQUAL 2 +#define HASHMAP_PHASE1_IS_MIN_KEY 3 +#define HASHMAP_PHASE1_CMP_VALUES 4 +#define HASHMAP_PHASE2_ARE_KEYS_EQUAL 5 +#define HASHMAP_PHASE2_IS_MIN_KEY_A 6 +#define HASHMAP_PHASE2_IS_MIN_KEY_B 7 + + +#define OP_WORD(OP) (((OP) << _TAG_PRIMARY_SIZE) | TAG_PRIMARY_HEADER) +#define TERM_ARRAY_OP_WORD(SZ) OP_WORD(((SZ) << OP_BITS) | TERM_ARRAY_OP) + +#define GET_OP(WORD) (ASSERT(is_header(WORD)), ((WORD) >> _TAG_PRIMARY_SIZE) & OP_MASK) +#define GET_OP_ARG(WORD) (ASSERT(is_header(WORD)), ((WORD) >> (OP_BITS + _TAG_PRIMARY_SIZE))) + + #define RETURN_NEQ(cmp) { j=(cmp); ASSERT(j != 0); goto not_equal; } #define ON_CMP_GOTO(cmp) if ((j=(cmp)) == 0) goto pop_next; else goto not_equal @@ -2481,6 +2998,8 @@ Sint erts_cmp(Eterm a, Eterm b, int exact) } while (0) +bodyrecur: + j = 0; tailrecur: if (is_same(a,a_base,b,b_base)) { /* Equal values or pointers. */ goto pop_next; @@ -2608,24 +3127,83 @@ tailrecur_ne: ++bb; goto term_array; case (_TAG_HEADER_MAP >> _TAG_PRIMARY_SIZE) : - if (!is_map_rel(b,b_base)) { + if (!is_flatmap_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); + aa = (Eterm *)flatmap_val_rel(a,a_base); + bb = (Eterm *)flatmap_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))); + i = flatmap_get_size((flatmap_t*)aa); + if (i != flatmap_get_size((flatmap_t*)bb)) { + RETURN_NEQ((int)(i - flatmap_get_size((flatmap_t*)bb))); } if (i == 0) { goto pop_next; } aa += 2; bb += 2; - i += 1; /* increment for tuple-keys */ - goto term_array; + if (exact) { + i += 1; /* increment for tuple-keys */ + goto term_array; + } + else { + /* Value array */ + WSTACK_PUSH3(stack, (UWord)(bb+1), (UWord)(aa+1), TERM_ARRAY_OP_WORD(i)); + /* Switch back from 'exact' key compare */ + WSTACK_PUSH(stack, OP_WORD(SWITCH_EXACT_OFF_OP)); + /* Now do 'exact' compare of key tuples */ + a = *aa; + b = *bb; + exact = 1; + goto bodyrecur; + } + + case (_TAG_HEADER_HASHMAP >> _TAG_PRIMARY_SIZE) : + { + struct erts_cmp_hashmap_state* sp; + if (!is_hashmap_rel(b,b_base)) { + a_tag = HASHMAP_DEF; + goto mixed_types; + } + i = hashmap_size_rel(a,a_base) - hashmap_size_rel(b,b_base); + if (i) { + RETURN_NEQ(i); + } + if (hashmap_size_rel(a,a_base) == 0) { + goto pop_next; + } + + /* Hashmap compare strategy: + Phase 1. While keys are identical + Do synchronous stepping through leafs of both trees in hash + order. Maintain value compare result of minimal key. + + Phase 2. If key diff was found in phase 1 + Ignore values from now on. + Continue iterate trees by always advancing the one + lagging behind hash-wise. Identical keys are skipped. + A minimal key can only be candidate as tie-breaker if we + have passed that hash value in the other tree (which means + the key did not exist in the other tree). + */ + + sp = PSTACK_PUSH(hmap_stack); + hashmap_iterator_init(&stack, a, 0); + hashmap_iterator_init(&b_stack, b, 0); + sp->ap = hashmap_iterator_next(&stack); + sp->bp = hashmap_iterator_next(&b_stack); + sp->cmp_res = 0; + ASSERT(sp->ap && sp->bp); + + a = CAR(sp->ap); + b = CAR(sp->bp); + sp->was_exact = exact; + exact = 1; + WSTACK_PUSH(stack, OP_WORD(HASHMAP_PHASE1_ARE_KEYS_EQUAL)); + sp->wstack_rollback = WSTACK_COUNT(stack); + goto bodyrecur; + } case (_TAG_HEADER_FLOAT >> _TAG_PRIMARY_SIZE): if (!is_float_rel(b,b_base)) { a_tag = FLOAT_DEF; @@ -2985,8 +3563,7 @@ term_array: /* arrays in 'aa' and 'bb', length in 'i' */ goto not_equal; } } else { - /* (ab)Use TAG_PRIMARY_HEADER to recognize a term_array */ - WSTACK_PUSH3(stack, i, (UWord)bb, (UWord)aa | TAG_PRIMARY_HEADER); + WSTACK_PUSH3(stack, (UWord)bb, (UWord)aa, TERM_ARRAY_OP_WORD(i)); goto tailrecur_ne; } } @@ -2998,22 +3575,179 @@ term_array: /* arrays in 'aa' and 'bb', length in 'i' */ pop_next: if (!WSTACK_ISEMPTY(stack)) { UWord something = WSTACK_POP(stack); - if (primary_tag((Eterm) something) == TAG_PRIMARY_HEADER) { /* a term_array */ - aa = (Eterm*) something; - bb = (Eterm*) WSTACK_POP(stack); - i = WSTACK_POP(stack); - goto term_array; + struct erts_cmp_hashmap_state* sp; + if (primary_tag((Eterm) something) == TAG_PRIMARY_HEADER) { /* an operation */ + switch (GET_OP(something)) { + case TERM_ARRAY_OP: + i = GET_OP_ARG(something); + aa = (Eterm*)WSTACK_POP(stack); + bb = (Eterm*) WSTACK_POP(stack); + goto term_array; + + case SWITCH_EXACT_OFF_OP: + /* Done with exact compare of map keys, switch back */ + ASSERT(exact); + exact = 0; + goto pop_next; + + case HASHMAP_PHASE1_ARE_KEYS_EQUAL: { + sp = PSTACK_TOP(hmap_stack); + if (j) { + /* Key diff found, enter phase 2 */ + if (hashmap_key_hash_cmp(sp->ap, sp->bp) < 0) { + sp->min_key = CAR(sp->ap); + sp->cmp_res = -1; + sp->ap = hashmap_iterator_next(&stack); + } + else { + sp->min_key = CAR(sp->bp); + sp->cmp_res = 1; + sp->bp = hashmap_iterator_next(&b_stack); + } + exact = 1; /* only exact key compares in phase 2 */ + goto case_HASHMAP_PHASE2_LOOP; + } + + /* No key diff found so far, compare values if min key */ + + if (sp->cmp_res) { + a = CAR(sp->ap); + b = sp->min_key; + exact = 1; + WSTACK_PUSH(stack, OP_WORD(HASHMAP_PHASE1_IS_MIN_KEY)); + sp->wstack_rollback = WSTACK_COUNT(stack); + goto bodyrecur; + } + /* no min key-value found yet */ + a = CDR(sp->ap); + b = CDR(sp->bp); + exact = sp->was_exact; + WSTACK_PUSH(stack, OP_WORD(HASHMAP_PHASE1_CMP_VALUES)); + sp->wstack_rollback = WSTACK_COUNT(stack); + goto bodyrecur; + } + case HASHMAP_PHASE1_IS_MIN_KEY: + sp = PSTACK_TOP(hmap_stack); + if (j < 0) { + a = CDR(sp->ap); + b = CDR(sp->bp); + exact = sp->was_exact; + WSTACK_PUSH(stack, OP_WORD(HASHMAP_PHASE1_CMP_VALUES)); + sp->wstack_rollback = WSTACK_COUNT(stack); + goto bodyrecur; + } + goto case_HASHMAP_PHASE1_LOOP; + + case HASHMAP_PHASE1_CMP_VALUES: + sp = PSTACK_TOP(hmap_stack); + if (j) { + sp->cmp_res = j; + sp->min_key = CAR(sp->ap); + } + case_HASHMAP_PHASE1_LOOP: + sp->ap = hashmap_iterator_next(&stack); + sp->bp = hashmap_iterator_next(&b_stack); + if (!sp->ap) { + /* end of maps with identical keys */ + ASSERT(!sp->bp); + j = sp->cmp_res; + exact = sp->was_exact; + (void) PSTACK_POP(hmap_stack); + ON_CMP_GOTO(j); + } + a = CAR(sp->ap); + b = CAR(sp->bp); + exact = 1; + WSTACK_PUSH(stack, OP_WORD(HASHMAP_PHASE1_ARE_KEYS_EQUAL)); + sp->wstack_rollback = WSTACK_COUNT(stack); + goto bodyrecur; + + case_HASHMAP_PHASE2_LOOP: + if (sp->ap && sp->bp) { + a = CAR(sp->ap); + b = CAR(sp->bp); + ASSERT(exact); + WSTACK_PUSH(stack, OP_WORD(HASHMAP_PHASE2_ARE_KEYS_EQUAL)); + sp->wstack_rollback = WSTACK_COUNT(stack); + goto bodyrecur; + } + goto case_HASHMAP_PHASE2_NEXT_STEP; + + case HASHMAP_PHASE2_ARE_KEYS_EQUAL: + sp = PSTACK_TOP(hmap_stack); + if (j == 0) { + /* keys are equal, skip them */ + sp->ap = hashmap_iterator_next(&stack); + sp->bp = hashmap_iterator_next(&b_stack); + goto case_HASHMAP_PHASE2_LOOP; + } + /* fall through */ + case_HASHMAP_PHASE2_NEXT_STEP: + if (sp->ap || sp->bp) { + if (hashmap_key_hash_cmp(sp->ap, sp->bp) < 0) { + ASSERT(sp->ap); + a = CAR(sp->ap); + b = sp->min_key; + ASSERT(exact); + WSTACK_PUSH(stack, OP_WORD(HASHMAP_PHASE2_IS_MIN_KEY_A)); + } + else { /* hash_cmp > 0 */ + ASSERT(sp->bp); + a = CAR(sp->bp); + b = sp->min_key; + ASSERT(exact); + WSTACK_PUSH(stack, OP_WORD(HASHMAP_PHASE2_IS_MIN_KEY_B)); + } + sp->wstack_rollback = WSTACK_COUNT(stack); + goto bodyrecur; + } + /* End of both maps */ + j = sp->cmp_res; + exact = sp->was_exact; + (void) PSTACK_POP(hmap_stack); + ON_CMP_GOTO(j); + + case HASHMAP_PHASE2_IS_MIN_KEY_A: + sp = PSTACK_TOP(hmap_stack); + if (j < 0) { + sp->min_key = CAR(sp->ap); + sp->cmp_res = -1; + } + sp->ap = hashmap_iterator_next(&stack); + goto case_HASHMAP_PHASE2_LOOP; + + case HASHMAP_PHASE2_IS_MIN_KEY_B: + sp = PSTACK_TOP(hmap_stack); + if (j < 0) { + sp->min_key = CAR(sp->bp); + sp->cmp_res = 1; + } + sp->bp = hashmap_iterator_next(&b_stack); + goto case_HASHMAP_PHASE2_LOOP; + + default: + ASSERT(!"Invalid cmp op"); + } /* switch */ } a = (Eterm) something; b = (Eterm) WSTACK_POP(stack); goto tailrecur; } - DESTROY_WSTACK(stack); + ASSERT(PSTACK_IS_EMPTY(hmap_stack)); + PSTACK_DESTROY(hmap_stack); + WSTACK_DESTROY(stack); + WSTACK_DESTROY(b_stack); return 0; not_equal: - DESTROY_WSTACK(stack); + if (!PSTACK_IS_EMPTY(hmap_stack)) { + WSTACK_ROLLBACK(stack, PSTACK_TOP(hmap_stack)->wstack_rollback); + goto pop_next; + } + PSTACK_DESTROY(hmap_stack); + WSTACK_DESTROY(stack); + WSTACK_DESTROY(b_stack); return j; #undef CMP_NODES @@ -4216,19 +4950,7 @@ void erts_silence_warn_unused_result(long unused) void erts_interval_init(erts_interval_t *icp) { -#ifdef ARCH_64 - erts_atomic_init_nob(&icp->counter.atomic, 0); -#else - erts_dw_aint_t dw; -#ifdef ETHR_SU_DW_NAINT_T__ - dw.dw_sint = 0; -#else - dw.sint[ERTS_DW_AINT_HIGH_WORD] = 0; - dw.sint[ERTS_DW_AINT_LOW_WORD] = 0; -#endif - erts_dw_atomic_init_nob(&icp->counter.atomic, &dw); - -#endif + erts_atomic64_init_nob(&icp->counter.atomic, 0); #ifdef DEBUG icp->smp_api = 0; #endif @@ -4250,55 +4972,13 @@ erts_smp_interval_init(erts_interval_t *icp) static ERTS_INLINE Uint64 step_interval_nob(erts_interval_t *icp) { -#ifdef ARCH_64 - return (Uint64) erts_atomic_inc_read_nob(&icp->counter.atomic); -#else - erts_dw_aint_t exp; - - erts_dw_atomic_read_nob(&icp->counter.atomic, &exp); - while (1) { - erts_dw_aint_t new = exp; - -#ifdef ETHR_SU_DW_NAINT_T__ - new.dw_sint++; -#else - new.sint[ERTS_DW_AINT_LOW_WORD]++; - if (new.sint[ERTS_DW_AINT_LOW_WORD] == 0) - new.sint[ERTS_DW_AINT_HIGH_WORD]++; -#endif - - if (erts_dw_atomic_cmpxchg_nob(&icp->counter.atomic, &new, &exp)) - return erts_interval_dw_aint_to_val__(&new); - - } -#endif + return (Uint64) erts_atomic64_inc_read_nob(&icp->counter.atomic); } static ERTS_INLINE Uint64 step_interval_relb(erts_interval_t *icp) { -#ifdef ARCH_64 - return (Uint64) erts_atomic_inc_read_relb(&icp->counter.atomic); -#else - erts_dw_aint_t exp; - - erts_dw_atomic_read_nob(&icp->counter.atomic, &exp); - while (1) { - erts_dw_aint_t new = exp; - -#ifdef ETHR_SU_DW_NAINT_T__ - new.dw_sint++; -#else - new.sint[ERTS_DW_AINT_LOW_WORD]++; - if (new.sint[ERTS_DW_AINT_LOW_WORD] == 0) - new.sint[ERTS_DW_AINT_HIGH_WORD]++; -#endif - - if (erts_dw_atomic_cmpxchg_relb(&icp->counter.atomic, &new, &exp)) - return erts_interval_dw_aint_to_val__(&new); - - } -#endif + return (Uint64) erts_atomic64_inc_read_relb(&icp->counter.atomic); } @@ -4306,38 +4986,10 @@ static ERTS_INLINE Uint64 ensure_later_interval_nob(erts_interval_t *icp, Uint64 ic) { Uint64 curr_ic; -#ifdef ARCH_64 - curr_ic = (Uint64) erts_atomic_read_nob(&icp->counter.atomic); + curr_ic = (Uint64) erts_atomic64_read_nob(&icp->counter.atomic); if (curr_ic > ic) return curr_ic; - return (Uint64) erts_atomic_inc_read_nob(&icp->counter.atomic); -#else - erts_dw_aint_t exp; - - erts_dw_atomic_read_nob(&icp->counter.atomic, &exp); - curr_ic = erts_interval_dw_aint_to_val__(&exp); - if (curr_ic > ic) - return curr_ic; - - while (1) { - erts_dw_aint_t new = exp; - -#ifdef ETHR_SU_DW_NAINT_T__ - new.dw_sint++; -#else - new.sint[ERTS_DW_AINT_LOW_WORD]++; - if (new.sint[ERTS_DW_AINT_LOW_WORD] == 0) - new.sint[ERTS_DW_AINT_HIGH_WORD]++; -#endif - - if (erts_dw_atomic_cmpxchg_nob(&icp->counter.atomic, &new, &exp)) - return erts_interval_dw_aint_to_val__(&new); - - curr_ic = erts_interval_dw_aint_to_val__(&exp); - if (curr_ic > ic) - return curr_ic; - } -#endif + return (Uint64) erts_atomic64_inc_read_nob(&icp->counter.atomic); } @@ -4345,38 +4997,10 @@ static ERTS_INLINE Uint64 ensure_later_interval_acqb(erts_interval_t *icp, Uint64 ic) { Uint64 curr_ic; -#ifdef ARCH_64 - curr_ic = (Uint64) erts_atomic_read_acqb(&icp->counter.atomic); - if (curr_ic > ic) - return curr_ic; - return (Uint64) erts_atomic_inc_read_acqb(&icp->counter.atomic); -#else - erts_dw_aint_t exp; - - erts_dw_atomic_read_acqb(&icp->counter.atomic, &exp); - curr_ic = erts_interval_dw_aint_to_val__(&exp); + curr_ic = (Uint64) erts_atomic64_read_acqb(&icp->counter.atomic); if (curr_ic > ic) return curr_ic; - - while (1) { - erts_dw_aint_t new = exp; - -#ifdef ETHR_SU_DW_NAINT_T__ - new.dw_sint++; -#else - new.sint[ERTS_DW_AINT_LOW_WORD]++; - if (new.sint[ERTS_DW_AINT_LOW_WORD] == 0) - new.sint[ERTS_DW_AINT_HIGH_WORD]++; -#endif - - if (erts_dw_atomic_cmpxchg_acqb(&icp->counter.atomic, &new, &exp)) - return erts_interval_dw_aint_to_val__(&new); - - curr_ic = erts_interval_dw_aint_to_val__(&exp); - if (curr_ic > ic) - return curr_ic; - } -#endif + return (Uint64) erts_atomic64_inc_read_acqb(&icp->counter.atomic); } Uint64 |