#ifndef RUBY_SHAPE_H #define RUBY_SHAPE_H #include "internal/gc.h" #include "internal/struct.h" typedef uint8_t attr_index_t; typedef uint32_t shape_id_t; #define SHAPE_ID_NUM_BITS 32 #define SHAPE_ID_OFFSET_NUM_BITS 19 STATIC_ASSERT(shape_id_num_bits, SHAPE_ID_NUM_BITS == sizeof(shape_id_t) * CHAR_BIT); #define SHAPE_BUFFER_SIZE (1 << SHAPE_ID_OFFSET_NUM_BITS) #define SHAPE_ID_OFFSET_MASK (SHAPE_BUFFER_SIZE - 1) #define SHAPE_ID_HEAP_INDEX_BITS 4 #define SHAPE_ID_HEAP_INDEX_MAX (((attr_index_t)1 << SHAPE_ID_HEAP_INDEX_BITS) - 1) #define SHAPE_ID_HEAP_INDEX_OFFSET SHAPE_ID_OFFSET_NUM_BITS #define SHAPE_ID_FL_USHIFT (SHAPE_ID_OFFSET_NUM_BITS + SHAPE_ID_HEAP_INDEX_BITS) // shape_id_t bits: // 0-18 SHAPE_ID_OFFSET_MASK // index in rb_shape_tree.shape_list. Allow to access `rb_shape_t *`. // This is the part that describe how fields are laid out in memory. // 19-22 SHAPE_ID_HEAP_INDEX_MASK // index in rb_shape_tree.capacities. Allow to access slot size. // Currently always 0 except for T_OBJECT. // 23 SHAPE_ID_FL_FROZEN // Whether the object is frozen or not. // 24 SHAPE_ID_FL_HAS_OBJECT_ID // Whether the object has an `SHAPE_OBJ_ID` transition. // 25 SHAPE_ID_FL_COMPLEX // The object is backed by a `st_table`. enum shape_id_fl_type { #define RBIMPL_SHAPE_ID_FL(n) (1<<(SHAPE_ID_FL_USHIFT+n)) SHAPE_ID_HEAP_INDEX_MASK = ((1 << SHAPE_ID_HEAP_INDEX_BITS) - 1) << SHAPE_ID_HEAP_INDEX_OFFSET, SHAPE_ID_FL_COMPLEX = RBIMPL_SHAPE_ID_FL(0), SHAPE_ID_FL_FROZEN = RBIMPL_SHAPE_ID_FL(1), SHAPE_ID_FL_HAS_OBJECT_ID = RBIMPL_SHAPE_ID_FL(2), SHAPE_ID_FL_NON_CANONICAL_MASK = SHAPE_ID_FL_FROZEN | SHAPE_ID_FL_HAS_OBJECT_ID, SHAPE_ID_FLAGS_MASK = SHAPE_ID_HEAP_INDEX_MASK | SHAPE_ID_FL_NON_CANONICAL_MASK | SHAPE_ID_FL_COMPLEX, #undef RBIMPL_SHAPE_ID_FL }; // This mask allows to check if a shape_id contains any ivar. // It relies on ROOT_SHAPE_WITH_OBJ_ID==1. enum shape_id_mask { SHAPE_ID_HAS_IVAR_MASK = SHAPE_ID_FL_COMPLEX | (SHAPE_ID_OFFSET_MASK - 1), }; // The interpreter doesn't care about frozen status, slot size or object id when reading ivars. // So we normalize shape_id by clearing these bits to improve cache hits. // JITs however might care about some of it. #define SHAPE_ID_READ_ONLY_MASK (~(SHAPE_ID_FL_FROZEN | SHAPE_ID_HEAP_INDEX_MASK | SHAPE_ID_FL_HAS_OBJECT_ID)) // For write it's the same idea, but here we do care about frozen status. #define SHAPE_ID_WRITE_MASK (~(SHAPE_ID_HEAP_INDEX_MASK | SHAPE_ID_FL_HAS_OBJECT_ID)) typedef uint32_t redblack_id_t; #define SHAPE_FLAG_SHIFT ((SIZEOF_VALUE * CHAR_BIT) - SHAPE_ID_NUM_BITS) #define SHAPE_FLAG_MASK (((VALUE)-1) >> SHAPE_ID_NUM_BITS) #define SHAPE_MAX_VARIATIONS 8 #define INVALID_SHAPE_ID (SHAPE_BUFFER_SIZE - 1) #define ATTR_INDEX_NOT_SET ((attr_index_t)-1) #define ROOT_SHAPE_ID 0x0 #define ROOT_SHAPE_WITH_OBJ_ID 0x1 #define ROOT_COMPLEX_SHAPE_ID (ROOT_SHAPE_ID | SHAPE_ID_FL_COMPLEX) #define ROOT_COMPLEX_WITH_OBJ_ID (ROOT_SHAPE_WITH_OBJ_ID | SHAPE_ID_FL_COMPLEX | SHAPE_ID_FL_HAS_OBJECT_ID) enum shape_type { SHAPE_ROOT, SHAPE_IVAR, SHAPE_OBJ_ID, }; struct rb_shape { VALUE edges; // id_table from ID (ivar) to next shape ID edge_name; // ID (ivar) for transition from parent to rb_shape redblack_id_t ancestor_index; shape_id_t parent_offset; attr_index_t next_field_index; // Fields are either ivars or internal properties like `object_id` attr_index_t capacity; // Total capacity of the object with this shape enum shape_type type : 8; }; typedef struct rb_shape rb_shape_t; enum shape_flags { SHAPE_FL_FROZEN = 1 << 0, SHAPE_FL_HAS_OBJECT_ID = 1 << 1, SHAPE_FL_COMPLEX = 1 << 2, SHAPE_FL_NON_CANONICAL_MASK = SHAPE_FL_FROZEN | SHAPE_FL_HAS_OBJECT_ID, }; typedef struct { rb_shape_t *shape_list; attr_index_t max_capacity; attr_index_t heaps_count; attr_index_t capacities[SHAPE_ID_HEAP_INDEX_MAX]; } rb_shape_tree_t; RUBY_SYMBOL_EXPORT_BEGIN RUBY_EXTERN rb_shape_tree_t rb_shape_tree; RUBY_SYMBOL_EXPORT_END size_t rb_shapes_cache_size(void); size_t rb_shapes_count(void); static inline shape_id_t RBASIC_SHAPE_ID(VALUE obj) { RUBY_ASSERT(!RB_SPECIAL_CONST_P(obj)); RUBY_ASSERT(!RB_TYPE_P(obj, T_IMEMO) || IMEMO_TYPE_P(obj, imemo_fields)); #if RBASIC_SHAPE_ID_FIELD return (shape_id_t)((RBASIC(obj)->shape_id)); #else return (shape_id_t)((RBASIC(obj)->flags) >> SHAPE_FLAG_SHIFT); #endif } // Same as RBASIC_SHAPE_ID but with flags that have no impact // on reads removed. e.g. Remove FL_FROZEN. static inline shape_id_t RBASIC_SHAPE_ID_FOR_READ(VALUE obj) { return RBASIC_SHAPE_ID(obj) & SHAPE_ID_READ_ONLY_MASK; } #if RUBY_DEBUG bool rb_shape_verify_consistency(VALUE obj, shape_id_t shape_id); #endif static inline void RBASIC_SET_SHAPE_ID_NO_CHECKS(VALUE obj, shape_id_t shape_id) { #if RBASIC_SHAPE_ID_FIELD RBASIC(obj)->shape_id = (VALUE)shape_id; #else // Object shapes are occupying top bits RBASIC(obj)->flags &= SHAPE_FLAG_MASK; RBASIC(obj)->flags |= ((VALUE)(shape_id) << SHAPE_FLAG_SHIFT); #endif } static inline void RBASIC_SET_SHAPE_ID(VALUE obj, shape_id_t shape_id) { RUBY_ASSERT(!RB_SPECIAL_CONST_P(obj)); RUBY_ASSERT(!RB_TYPE_P(obj, T_IMEMO) || IMEMO_TYPE_P(obj, imemo_fields)); RBASIC_SET_SHAPE_ID_NO_CHECKS(obj, shape_id); RUBY_ASSERT(rb_shape_verify_consistency(obj, shape_id)); } static inline shape_id_t RSHAPE_FLAGS(shape_id_t shape_id) { return shape_id & SHAPE_ID_FLAGS_MASK; } static inline shape_id_t RSHAPE_OFFSET(shape_id_t shape_id) { return shape_id & SHAPE_ID_OFFSET_MASK; } static inline rb_shape_t * RSHAPE(shape_id_t shape_id) { shape_id_t offset = RSHAPE_OFFSET(shape_id); RUBY_ASSERT(offset != INVALID_SHAPE_ID); return &rb_shape_tree.shape_list[offset]; } int32_t rb_shape_id_offset(void); RUBY_FUNC_EXPORTED shape_id_t rb_obj_shape_id(VALUE obj); shape_id_t rb_shape_get_next_iv_shape(shape_id_t shape_id, ID id); bool rb_shape_get_iv_index(shape_id_t shape_id, ID id, attr_index_t *value); bool rb_shape_get_iv_index_with_hint(shape_id_t shape_id, ID id, attr_index_t *value, shape_id_t *shape_id_hint); bool rb_shape_find_ivar(shape_id_t shape_id, ID id, shape_id_t *ivar_shape); typedef int rb_shape_foreach_transition_callback(shape_id_t shape_id, void *data); bool rb_shape_foreach_field(shape_id_t shape_id, rb_shape_foreach_transition_callback func, void *data); shape_id_t rb_shape_transition_add_ivar_no_warnings(shape_id_t shape_id, ID id, VALUE klass); shape_id_t rb_shape_object_id(shape_id_t original_shape_id); shape_id_t rb_shape_rebuild(shape_id_t initial_shape_id, shape_id_t dest_shape_id); void rb_shape_copy_fields(VALUE dest, VALUE *dest_buf, shape_id_t dest_shape_id, VALUE *src_buf, shape_id_t src_shape_id); void rb_shape_copy_complex_ivars(VALUE dest, VALUE obj, shape_id_t src_shape_id, st_table *fields_table); static inline bool rb_shape_frozen_p(shape_id_t shape_id) { return shape_id & SHAPE_ID_FL_FROZEN; } static inline bool rb_shape_complex_p(shape_id_t shape_id) { return shape_id & SHAPE_ID_FL_COMPLEX; } static inline bool rb_obj_shape_complex_p(VALUE obj) { return !RB_SPECIAL_CONST_P(obj) && rb_shape_complex_p(RBASIC_SHAPE_ID(obj)); } static inline bool rb_shape_has_object_id(shape_id_t shape_id) { return shape_id & SHAPE_ID_FL_HAS_OBJECT_ID; } static inline bool rb_shape_canonical_p(shape_id_t shape_id) { return !(shape_id & SHAPE_ID_FL_NON_CANONICAL_MASK); } static inline uint8_t rb_shape_heap_index(shape_id_t shape_id) { return (uint8_t)((shape_id & SHAPE_ID_HEAP_INDEX_MASK) >> SHAPE_ID_HEAP_INDEX_OFFSET); } static inline shape_id_t rb_shape_root(size_t heap_id) { shape_id_t heap_index = (shape_id_t)(heap_id + 1); shape_id_t heap_flags = heap_index << SHAPE_ID_HEAP_INDEX_OFFSET; RUBY_ASSERT((heap_flags & SHAPE_ID_HEAP_INDEX_MASK) == heap_flags); RUBY_ASSERT(rb_shape_heap_index(heap_flags) == heap_index); return ROOT_SHAPE_ID | heap_flags; } static inline shape_id_t RSHAPE_PARENT_OFFSET(shape_id_t shape_id) { return RSHAPE(shape_id)->parent_offset; } static inline bool RSHAPE_DIRECT_CHILD_P(shape_id_t parent_offset, shape_id_t child_id) { return RSHAPE_PARENT_OFFSET(child_id) == RSHAPE_OFFSET(parent_offset); } static inline enum shape_type RSHAPE_TYPE(shape_id_t shape_id) { return RSHAPE(shape_id)->type; } static inline bool RSHAPE_TYPE_P(shape_id_t shape_id, enum shape_type type) { return RSHAPE_TYPE(shape_id) == type; } static inline attr_index_t RSHAPE_EMBEDDED_CAPACITY(shape_id_t shape_id) { uint8_t heap_index = rb_shape_heap_index(shape_id); if (heap_index) { return rb_shape_tree.capacities[heap_index - 1]; } return 0; } static inline attr_index_t RSHAPE_CAPACITY(shape_id_t shape_id) { attr_index_t embedded_capacity = RSHAPE_EMBEDDED_CAPACITY(shape_id); if (embedded_capacity > RSHAPE(shape_id)->capacity) { return embedded_capacity; } else { return RSHAPE(shape_id)->capacity; } } static inline attr_index_t RSHAPE_LEN(shape_id_t shape_id) { return RSHAPE(shape_id)->next_field_index; } static inline attr_index_t RSHAPE_INDEX(shape_id_t shape_id) { RUBY_ASSERT(RSHAPE_LEN(shape_id) > 0); return RSHAPE_LEN(shape_id) - 1; } static inline ID RSHAPE_EDGE_NAME(shape_id_t shape_id) { return RSHAPE(shape_id)->edge_name; } static inline uint32_t ROBJECT_FIELDS_CAPACITY(VALUE obj) { RBIMPL_ASSERT_TYPE(obj, RUBY_T_OBJECT); // Asking for capacity doesn't make sense when the object is using // a hash table for storing instance variables RUBY_ASSERT(!rb_obj_shape_complex_p(obj)); return RSHAPE_CAPACITY(RBASIC_SHAPE_ID(obj)); } static inline st_table * ROBJECT_FIELDS_HASH(VALUE obj) { RBIMPL_ASSERT_TYPE(obj, RUBY_T_OBJECT); RUBY_ASSERT(rb_obj_shape_complex_p(obj)); RUBY_ASSERT(FL_TEST_RAW(obj, ROBJECT_HEAP)); return ROBJECT(obj)->as.hash; } static inline void ROBJECT_SET_FIELDS_HASH(VALUE obj, st_table *tbl) { RBIMPL_ASSERT_TYPE(obj, RUBY_T_OBJECT); RUBY_ASSERT(rb_obj_shape_complex_p(obj)); RUBY_ASSERT(FL_TEST_RAW(obj, ROBJECT_HEAP)); ROBJECT(obj)->as.hash = tbl; } static inline uint32_t ROBJECT_FIELDS_COUNT_COMPLEX(VALUE obj) { return (uint32_t)rb_st_table_size(ROBJECT_FIELDS_HASH(obj)); } static inline uint32_t ROBJECT_FIELDS_COUNT_NOT_COMPLEX(VALUE obj) { RBIMPL_ASSERT_TYPE(obj, RUBY_T_OBJECT); RUBY_ASSERT(!rb_obj_shape_complex_p(obj)); return RSHAPE(RBASIC_SHAPE_ID(obj))->next_field_index; } static inline uint32_t ROBJECT_FIELDS_COUNT(VALUE obj) { if (rb_obj_shape_complex_p(obj)) { return ROBJECT_FIELDS_COUNT_COMPLEX(obj); } else { return ROBJECT_FIELDS_COUNT_NOT_COMPLEX(obj); } } static inline uint32_t RBASIC_FIELDS_COUNT(VALUE obj) { return RSHAPE(RBASIC_SHAPE_ID(obj))->next_field_index; } static inline bool rb_obj_shape_has_id(VALUE obj) { return rb_shape_has_object_id(RBASIC_SHAPE_ID(obj)); } static inline bool rb_shape_has_ivars(shape_id_t shape_id) { return shape_id & SHAPE_ID_HAS_IVAR_MASK; } static inline bool rb_obj_shape_has_ivars(VALUE obj) { return rb_shape_has_ivars(RBASIC_SHAPE_ID(obj)); } static inline bool rb_shape_has_fields(shape_id_t shape_id) { return shape_id & (SHAPE_ID_OFFSET_MASK | SHAPE_ID_FL_COMPLEX); } static inline bool rb_obj_shape_has_fields(VALUE obj) { return rb_shape_has_fields(RBASIC_SHAPE_ID(obj)); } static inline bool rb_obj_gen_fields_p(VALUE obj) { switch (TYPE(obj)) { case T_NONE: case T_OBJECT: case T_CLASS: case T_MODULE: case T_IMEMO: return false; default: break; } return rb_obj_shape_has_fields(obj); } static inline bool rb_obj_using_gen_fields_table_p(VALUE obj) { switch (BUILTIN_TYPE(obj)) { case T_DATA: if (RTYPEDDATA_P(obj)) return false; break; case T_STRUCT: if (!FL_TEST_RAW(obj, RSTRUCT_GEN_FIELDS)) return false; break; default: break; } return rb_obj_gen_fields_p(obj); } static inline shape_id_t rb_shape_transition_frozen(shape_id_t shape_id) { return shape_id | SHAPE_ID_FL_FROZEN; } static inline shape_id_t rb_shape_transition_complex(shape_id_t shape_id) { shape_id_t next_shape_id = ROOT_COMPLEX_SHAPE_ID; if (rb_shape_has_object_id(shape_id)) { next_shape_id = ROOT_COMPLEX_WITH_OBJ_ID; } uint8_t heap_index = rb_shape_heap_index(shape_id); if (heap_index) { next_shape_id |= rb_shape_root(heap_index - 1); } RUBY_ASSERT(rb_shape_has_object_id(shape_id) == rb_shape_has_object_id(next_shape_id)); return next_shape_id; } static inline shape_id_t rb_shape_transition_offset(shape_id_t shape_id, shape_id_t offset) { offset = RSHAPE_OFFSET(offset); RUBY_ASSERT(RSHAPE_OFFSET(shape_id) == offset || RSHAPE_DIRECT_CHILD_P(shape_id, offset)); return RSHAPE_FLAGS(shape_id) | offset; } static inline shape_id_t rb_shape_transition_heap(shape_id_t shape_id, size_t heap_index) { return (shape_id & (~SHAPE_ID_HEAP_INDEX_MASK)) | rb_shape_root(heap_index); } shape_id_t rb_shape_transition_object_id(shape_id_t shape_id); static inline shape_id_t rb_obj_shape_transition_frozen(VALUE obj) { RUBY_ASSERT(RB_OBJ_FROZEN(obj)); return rb_shape_transition_frozen(RBASIC_SHAPE_ID(obj)); } static inline shape_id_t rb_obj_shape_transition_complex(VALUE obj) { return rb_shape_transition_complex(RBASIC_SHAPE_ID(obj)); } static inline shape_id_t rb_obj_shape_transition_heap(VALUE obj, size_t heap_index) { return rb_shape_transition_heap(RBASIC_SHAPE_ID(obj), heap_index); } static inline shape_id_t rb_obj_shape_transition_object_id(VALUE obj) { return rb_shape_transition_object_id(RBASIC_SHAPE_ID(obj)); } shape_id_t rb_obj_shape_transition_remove_ivar(VALUE obj, ID id, shape_id_t *removed_shape_id); shape_id_t rb_obj_shape_transition_add_ivar(VALUE obj, ID id); // For ext/objspace RUBY_SYMBOL_EXPORT_BEGIN typedef void each_shape_callback(shape_id_t shape_id, void *data); void rb_shape_each_shape_id(each_shape_callback callback, void *data); size_t rb_shape_memsize(shape_id_t shape); size_t rb_shape_edges_count(shape_id_t shape_id); size_t rb_shape_depth(shape_id_t shape_id); RUBY_SYMBOL_EXPORT_END // Inline cache helpers typedef struct { attr_index_t index; shape_id_t shape_offset; } rb_getivar_cache; union rb_getivar_cache { uint64_t pack; rb_getivar_cache unpack; }; STATIC_ASSERT(rb_getivar_cache_size, sizeof(union rb_getivar_cache) <= sizeof(uint64_t)); #define IVAR_CACHE_INIT ((uint64_t)-1) #define ATTR_INDEX_T_NUM_BITS (sizeof(attr_index_t) * CHAR_BIT) static inline rb_getivar_cache rb_getivar_cache_unpack(uint64_t packed) { union rb_getivar_cache cache = { .pack = packed, }; // Because caches may initialized with all bits set (IVAR_CACHE_INIT), and `shape_offset` if 32bits, // we need to remove any potential extra bits set in the "padding". cache.unpack.shape_offset &= SHAPE_ID_OFFSET_MASK; return cache.unpack; } static inline uint64_t rb_getivar_cache_pack(shape_id_t shape_offset, attr_index_t index) { RUBY_ASSERT(shape_offset == RSHAPE_OFFSET(shape_offset)); RUBY_ASSERT(shape_offset != INVALID_SHAPE_ID); union rb_getivar_cache cache = { .unpack = { .shape_offset = shape_offset, .index = index, }, }; return cache.pack; } typedef struct { attr_index_t index; shape_id_t source_shape_offset; shape_id_t dest_shape_offset; } rb_setivar_cache; static inline rb_setivar_cache rb_setivar_cache_unpack(uint64_t packed) { rb_setivar_cache cache = { .index = (attr_index_t)packed, .source_shape_offset = RSHAPE_OFFSET((shape_id_t)(packed >> ATTR_INDEX_T_NUM_BITS)), .dest_shape_offset = RSHAPE_OFFSET((shape_id_t)(packed >> (ATTR_INDEX_T_NUM_BITS + SHAPE_ID_OFFSET_NUM_BITS))), }; return cache; } static inline uint64_t rb_setivar_cache_pack(shape_id_t shape_offset, shape_id_t dest_shape_offset, attr_index_t index) { RUBY_ASSERT(shape_offset == RSHAPE_OFFSET(shape_offset)); RUBY_ASSERT(dest_shape_offset == RSHAPE_OFFSET(dest_shape_offset)); RUBY_ASSERT(shape_offset == dest_shape_offset || RSHAPE_DIRECT_CHILD_P(shape_offset, dest_shape_offset)); uint64_t packed_cache = (uint64_t)dest_shape_offset << (ATTR_INDEX_T_NUM_BITS + SHAPE_ID_OFFSET_NUM_BITS); packed_cache |= (uint64_t)shape_offset << ATTR_INDEX_T_NUM_BITS; packed_cache |= (uint64_t)index; return packed_cache; } ALWAYS_INLINE(static shape_id_t rb_setivar_cache_revalidate(shape_id_t shape_id, rb_setivar_cache cache)); static shape_id_t rb_setivar_cache_revalidate(shape_id_t shape_id, rb_setivar_cache cache) { RUBY_ASSERT(shape_id != INVALID_SHAPE_ID); RUBY_ASSERT(cache.dest_shape_offset == INVALID_SHAPE_ID || cache.dest_shape_offset == RSHAPE_OFFSET(cache.dest_shape_offset)); shape_id_t normalized_shape_id = shape_id & SHAPE_ID_WRITE_MASK; if (UNLIKELY(normalized_shape_id != cache.source_shape_offset)) { return INVALID_SHAPE_ID; } if (UNLIKELY(cache.index >= RSHAPE_CAPACITY(shape_id))) { // That's still a hit in term of layout, but the object will need to be resized, // so unfortunately we'll have to go through the slow path regardless... return INVALID_SHAPE_ID; } // Cache hit case RUBY_ASSERT(cache.source_shape_offset == cache.dest_shape_offset || RSHAPE_DIRECT_CHILD_P(shape_id, cache.dest_shape_offset)); RUBY_ASSERT(cache.index < RSHAPE_CAPACITY(shape_id)); RUBY_ASSERT(!rb_shape_frozen_p(shape_id)); RUBY_ASSERT(!rb_shape_complex_p(shape_id)); // We use the cached offset, but combined with the current shape flags. return rb_shape_transition_offset(shape_id, cache.dest_shape_offset); } #endif