From 9b26bb2e6a1e32e17102550b481c6909549b87e5 Mon Sep 17 00:00:00 2001 From: gingerBill Date: Mon, 17 Feb 2025 13:10:38 +0000 Subject: Begin work on hash types --- src/llvm_backend.cpp | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-) (limited to 'src/llvm_backend.cpp') diff --git a/src/llvm_backend.cpp b/src/llvm_backend.cpp index 0896ea8c7..8cb480dd4 100644 --- a/src/llvm_backend.cpp +++ b/src/llvm_backend.cpp @@ -24,7 +24,7 @@ #include "llvm_backend_stmt.cpp" #include "llvm_backend_proc.cpp" -String get_default_microarchitecture() { +gb_internal String get_default_microarchitecture() { String default_march = str_lit("generic"); if (build_context.metrics.arch == TargetArch_amd64) { // NOTE(bill): x86-64-v2 is more than enough for everyone @@ -47,7 +47,7 @@ String get_default_microarchitecture() { return default_march; } -String get_final_microarchitecture() { +gb_internal String get_final_microarchitecture() { BuildContext *bc = &build_context; String microarch = bc->microarch; @@ -3182,7 +3182,8 @@ gb_internal bool lb_generate_code(lbGenerator *gen) { isize count = 0; isize offsets_extra = 0; - for (Type *t : m->info->type_info_types) { + for (auto const &tt : m->info->type_info_types) { + Type *t = tt.type; isize index = lb_type_info_index(m->info, t, false); if (index < 0) { continue; -- cgit v1.2.3 From 4a29d9bb845050c483e537c7a0d6b2889af0f7bc Mon Sep 17 00:00:00 2001 From: gingerBill Date: Mon, 17 Feb 2025 16:29:42 +0000 Subject: Simplify type info table construction --- src/checker.cpp | 162 ++++++++++++++++++++++++++---------------- src/checker.hpp | 17 +++-- src/llvm_backend.cpp | 4 +- src/llvm_backend_type.cpp | 25 +++---- src/name_canonicalization.cpp | 23 ++++-- 5 files changed, 146 insertions(+), 85 deletions(-) (limited to 'src/llvm_backend.cpp') diff --git a/src/checker.cpp b/src/checker.cpp index 6ceb31489..1c7126e9a 100644 --- a/src/checker.cpp +++ b/src/checker.cpp @@ -828,9 +828,15 @@ gb_internal void add_dependency(CheckerInfo *info, DeclInfo *d, Entity *e) { rw_mutex_unlock(&d->deps_mutex); } gb_internal void add_type_info_dependency(CheckerInfo *info, DeclInfo *d, Type *type) { - if (d == nullptr) { + if (d == nullptr || type == nullptr) { return; } + if (type->kind == Type_Named) { + Entity *e = type->Named.type_name; + if (e->TypeName.is_type_alias) { + type = type->Named.base; + } + } rw_mutex_lock(&d->type_info_deps_mutex); ptr_set_add(&d->type_info_deps, type); rw_mutex_unlock(&d->type_info_deps_mutex); @@ -1361,9 +1367,12 @@ gb_internal void init_checker_info(CheckerInfo *i) { string_map_init(&i->foreigns); // map_init(&i->gen_procs); map_init(&i->gen_types); + array_init(&i->type_info_types, a); - map_init(&i->type_info_map); - type_set_init(&i->type_info_set); + type_set_init(&i->min_dep_type_info_set); + map_init(&i->minimum_dependency_type_info_index_map); + + // map_init(&i->type_info_map); string_map_init(&i->files); string_map_init(&i->packages); array_init(&i->variable_init_order, a); @@ -1396,9 +1405,11 @@ gb_internal void destroy_checker_info(CheckerInfo *i) { string_map_destroy(&i->foreigns); // map_destroy(&i->gen_procs); map_destroy(&i->gen_types); + array_free(&i->type_info_types); - map_destroy(&i->type_info_map); - type_set_destroy(&i->type_info_set); + type_set_destroy(&i->min_dep_type_info_set); + map_destroy(&i->minimum_dependency_type_info_index_map); + string_map_destroy(&i->files); string_map_destroy(&i->packages); array_free(&i->variable_init_order); @@ -1632,41 +1643,36 @@ gb_internal void check_remove_expr_info(CheckerContext *c, Ast *e) { } } - -gb_internal isize type_info_index(CheckerInfo *info, Type *type, bool error_on_failure) { - type = default_type(type); - if (type == t_llvm_bool) { - type = t_bool; - } - - mutex_lock(&info->type_info_mutex); +gb_internal isize type_info_index(CheckerInfo *info, TypeInfoPair pair, bool error_on_failure) { + mutex_lock(&info->minimum_dependency_type_info_mutex); isize entry_index = -1; - isize *found_entry_index = map_get(&info->type_info_map, type); + uintptr hash = cast(uintptr)pair.hash; + isize *found_entry_index = map_get(&info->minimum_dependency_type_info_index_map, hash); if (found_entry_index) { entry_index = *found_entry_index; } - if (entry_index < 0) { - // NOTE(bill): Do manual linear search - for (auto const &e : info->type_info_map) { - if (are_types_identical_unique_tuples(e.key, type)) { - entry_index = e.value; - // NOTE(bill): Add it to the search map - map_set(&info->type_info_map, type, entry_index); - break; - } - } - } - - mutex_unlock(&info->type_info_mutex); + mutex_unlock(&info->minimum_dependency_type_info_mutex); if (error_on_failure && entry_index < 0) { - compiler_error("Type_Info for '%s' could not be found", type_to_string(type)); + compiler_error("Type_Info for '%s' could not be found", type_to_string(pair.type)); } return entry_index; } +gb_internal isize type_info_index(CheckerInfo *info, Type *type, bool error_on_failure) { + type = default_type(type); + if (type == t_llvm_bool) { + type = t_bool; + } + + u64 hash = type_hash_canonical_type(type); + return type_info_index(info, {type, hash}, error_on_failure); +} + + + gb_internal void add_untyped(CheckerContext *c, Ast *expr, AddressingMode mode, Type *type, ExactValue const &value) { if (expr == nullptr) { return; @@ -2018,8 +2024,12 @@ gb_internal void add_type_info_type_internal(CheckerContext *c, Type *t) { } add_type_info_dependency(c->info, c->decl, t); - +#if 0 MUTEX_GUARD_BLOCK(&c->info->type_info_mutex) { + if (type_set_update(&c->info->type_info_set, t)) { + // return; + } + auto found = map_get(&c->info->type_info_map, t); if (found != nullptr) { // Types have already been added @@ -2238,6 +2248,7 @@ gb_internal void add_type_info_type_internal(CheckerContext *c, Type *t) { GB_PANIC("Unhandled type: %*.s %d", LIT(type_strings[bt->kind]), bt->kind); break; } +#endif } @@ -2295,11 +2306,10 @@ gb_internal void add_min_dep_type_info(Checker *c, Type *t) { return; } - if (type_set_update(&c->info.type_info_set, t)) { - // return; + if (type_set_update(&c->info.min_dep_type_info_set, t)) { + return; } - // Add nested types if (t->kind == Type_Named) { // NOTE(bill): Just in case @@ -2697,7 +2707,6 @@ gb_internal void generate_minimum_dependency_set(Checker *c, Entity *start) { isize min_dep_set_cap = next_pow2_isize(entity_count*4); // empirically determined factor ptr_set_init(&c->info.minimum_dependency_set, min_dep_set_cap); - map_init(&c->info.minimum_dependency_type_info_set); #define FORCE_ADD_RUNTIME_ENTITIES(condition, ...) do { \ if (condition) { \ @@ -6720,39 +6729,70 @@ gb_internal void check_parsed_files(Checker *c) { add_type_and_value(&c->builtin_ctx, u.expr, u.info->mode, u.info->type, u.info->value); } - TIME_SECTION("check for type hash collisions"); + TIME_SECTION("initilize type info array"); { - PtrSet found = {}; - ptr_set_init(&found, c->info.type_info_types.count); - defer (ptr_set_destroy(&found)); - for (auto const &tt : c->info.type_info_types) { - if (ptr_set_update(&found, cast(uintptr)tt.hash)) { - Type *other_type = nullptr; - for (auto const &other : c->info.type_info_types) { - if (&tt == &other) { - continue; - } - if (cast(uintptr)other.hash == cast(uintptr)tt.hash && - !are_types_identical(tt.type, other.type)) { - other_type = other.type; - break; - } - } - if (other_type != nullptr) { - String ts = type_to_canonical_string(temporary_allocator(), tt.type); - String os = type_to_canonical_string(temporary_allocator(), other_type); - if (ts != os) { - compiler_error("%s found type hash collision with %s (hash = %llu)\n" - "%s vs %s\n", - type_to_string(tt.type), type_to_string(other_type), cast(unsigned long long)tt.hash, - temp_canonical_string(tt.type), - temp_canonical_string(other_type) - ); + for (auto const &tt : c->info.min_dep_type_info_set) { + array_add(&c->info.type_info_types, tt); + } + array_sort(c->info.type_info_types, type_info_pair_cmp); + + map_reserve(&c->info.minimum_dependency_type_info_index_map, c->info.type_info_types.count); + + for_array(i, c->info.type_info_types) { + auto const &tt = c->info.type_info_types[i]; + bool exists = map_set_if_not_previously_exists(&c->info.minimum_dependency_type_info_index_map, cast(uintptr)tt.hash, i); + if (exists) { + for (auto const &entry : c->info.minimum_dependency_type_info_index_map) { + if (entry.key == cast(uintptr)tt.hash) { + auto const &other = c->info.type_info_types[entry.value]; + if (!are_types_identical_unique_tuples(tt.type, other.type)) { + gbString t = temp_canonical_string(tt.type); + gbString o = temp_canonical_string(other.type); + GB_PANIC("%s (%s) %llu vs %s (%s) %llu", + type_to_string(tt.type, false), t, cast(unsigned long long)tt.hash, + type_to_string(other.type, false), o, cast(unsigned long long)other.hash); + } } } } } - } + + GB_ASSERT(c->info.minimum_dependency_type_info_index_map.count <= c->info.type_info_types.count); + } + + // TIME_SECTION("check for type hash collisions"); + // { + // PtrSet found = {}; + // ptr_set_init(&found, c->info.type_info_types.count); + // defer (ptr_set_destroy(&found)); + // for (auto const &tt : c->info.type_info_types) { + // if (ptr_set_update(&found, cast(uintptr)tt.hash)) { + // Type *other_type = nullptr; + // for (auto const &other : c->info.type_info_types) { + // if (&tt == &other) { + // continue; + // } + // if (cast(uintptr)other.hash == cast(uintptr)tt.hash && + // !are_types_identical(tt.type, other.type)) { + // other_type = other.type; + // break; + // } + // } + // if (other_type != nullptr) { + // String ts = type_to_canonical_string(temporary_allocator(), tt.type); + // String os = type_to_canonical_string(temporary_allocator(), other_type); + // if (ts != os) { + // compiler_error("%s found type hash collision with %s (hash = %llu)\n" + // "%s vs %s\n", + // type_to_string(tt.type), type_to_string(other_type), cast(unsigned long long)tt.hash, + // temp_canonical_string(tt.type), + // temp_canonical_string(other_type) + // ); + // } + // } + // } + // } + // } diff --git a/src/checker.hpp b/src/checker.hpp index 725c1ccf5..52676d4ee 100644 --- a/src/checker.hpp +++ b/src/checker.hpp @@ -222,7 +222,7 @@ struct DeclInfo { PtrSet deps; RwMutex type_info_deps_mutex; - PtrSet type_info_deps; + PtrSet type_info_deps; // TODO(bill): Use TypeSet BlockingMutex type_and_value_mutex; @@ -444,8 +444,10 @@ struct CheckerInfo { Scope * init_scope; Entity * entry_point; PtrSet minimum_dependency_set; - PtrMap minimum_dependency_type_info_set; - + BlockingMutex minimum_dependency_type_info_mutex; + PtrMap minimum_dependency_type_info_index_map; + TypeSet min_dep_type_info_set; + Array type_info_types; // sorted after filled Array testing_procedures; @@ -473,10 +475,10 @@ struct CheckerInfo { BlockingMutex gen_types_mutex; PtrMap gen_types; - BlockingMutex type_info_mutex; // NOT recursive - Array type_info_types; - PtrMap type_info_map; - TypeSet type_info_set; + // BlockingMutex type_info_mutex; // NOT recursive + // Array type_info_types; + // PtrMap type_info_map; + // TypeSet type_info_set; BlockingMutex foreign_mutex; // NOT recursive StringMap foreigns; @@ -595,6 +597,7 @@ gb_internal DeclInfo * decl_info_of_entity (Entity * e); gb_internal AstFile * ast_file_of_filename (CheckerInfo *i, String filename); // IMPORTANT: Only to use once checking is done gb_internal isize type_info_index (CheckerInfo *i, Type *type, bool error_on_failure); +gb_internal isize type_info_index (CheckerInfo *info, TypeInfoPair pair, bool error_on_failure); // Will return nullptr if not found gb_internal Entity *entity_of_node(Ast *expr); diff --git a/src/llvm_backend.cpp b/src/llvm_backend.cpp index 8cb480dd4..908117501 100644 --- a/src/llvm_backend.cpp +++ b/src/llvm_backend.cpp @@ -3154,7 +3154,9 @@ gb_internal bool lb_generate_code(lbGenerator *gen) { lbModule *m = default_module; { // Add type info data - isize max_type_info_count = info->minimum_dependency_type_info_set.count+1; + GB_ASSERT_MSG(info->minimum_dependency_type_info_index_map.count == info->type_info_types.count, "%tu vs %tu", info->minimum_dependency_type_info_index_map.count, info->type_info_types.count); + + isize max_type_info_count = info->minimum_dependency_type_info_index_map.count+1; Type *t = alloc_type_array(t_type_info_ptr, max_type_info_count); // IMPORTANT NOTE(bill): As LLVM does not have a union type, an array of unions cannot be initialized diff --git a/src/llvm_backend_type.cpp b/src/llvm_backend_type.cpp index 6f9f94fbd..8e0f15f35 100644 --- a/src/llvm_backend_type.cpp +++ b/src/llvm_backend_type.cpp @@ -1,16 +1,12 @@ -gb_internal isize lb_type_info_index(CheckerInfo *info, Type *type, bool err_on_not_found=true) { - auto *set = &info->minimum_dependency_type_info_set; - isize index = type_info_index(info, type, err_on_not_found); + +gb_internal isize lb_type_info_index(CheckerInfo *info, TypeInfoPair pair, bool err_on_not_found=true) { + isize index = type_info_index(info, pair, err_on_not_found); if (index >= 0) { - auto *found = map_get(set, index+1); - if (found) { - GB_ASSERT(*found >= 0); - return *found + 1; - } + return index+1; } if (err_on_not_found) { - gb_printf_err("NOT FOUND lb_type_info_index:\n\t%s\n\t@ index %td\n\tmax count: %u\nFound:\n", type_to_string(type), index, set->count); - for (auto const &entry : *set) { + gb_printf_err("NOT FOUND lb_type_info_index:\n\t%s\n\t@ index %td\n\tmax count: %u\nFound:\n", type_to_string(pair.type), index, info->minimum_dependency_type_info_index_map.count); + for (auto const &entry : info->minimum_dependency_type_info_index_map) { isize type_info_index = entry.key; gb_printf_err("\t%s\n", type_to_string(info->type_info_types[type_info_index].type)); } @@ -19,6 +15,10 @@ gb_internal isize lb_type_info_index(CheckerInfo *info, Type *type, bool err_on_ return -1; } +gb_internal isize lb_type_info_index(CheckerInfo *info, Type *type, bool err_on_not_found=true) { + return lb_type_info_index(info, {type, type_hash_canonical_type(type)}, err_on_not_found); +} + gb_internal u64 lb_typeid_kind(lbModule *m, Type *type, u64 id=0) { GB_ASSERT(!build_context.no_rtti); @@ -280,12 +280,13 @@ gb_internal void lb_setup_type_info_data_giant_array(lbModule *m, i64 global_typ LLVMTypeRef *modified_types = lb_setup_modified_types_for_type_info(m, global_type_info_data_entity_count); defer (gb_free(heap_allocator(), modified_types)); for_array(type_info_type_index, info->type_info_types) { - Type *t = info->type_info_types[type_info_type_index].type; + auto const &tt = info->type_info_types[type_info_type_index]; + Type *t = tt.type; if (t == nullptr || t == t_invalid) { continue; } - isize entry_index = lb_type_info_index(info, t, false); + isize entry_index = lb_type_info_index(info, tt, false); if (entry_index <= 0) { continue; } diff --git a/src/name_canonicalization.cpp b/src/name_canonicalization.cpp index b83441b19..de35312da 100644 --- a/src/name_canonicalization.cpp +++ b/src/name_canonicalization.cpp @@ -48,6 +48,15 @@ gb_internal gbString temp_canonical_string(Type *type); struct TypeInfoPair; struct TypeSet; +gb_internal GB_COMPARE_PROC(type_info_pair_cmp) { + TypeInfoPair *x = cast(TypeInfoPair *)a; + TypeInfoPair *y = cast(TypeInfoPair *)b; + if (x->hash == y->hash) { + return 0; + } + return x->hash < y->hash ? -1 : +1; +} + static constexpr u64 TYPE_SET_TOMBSTONE = ~(u64)(0ull); gb_internal void type_set_init (TypeSet *s, isize capacity); @@ -136,7 +145,7 @@ gb_internal isize type_set__find(TypeSet *s, TypeInfoPair pair) { usize hash_index = cast(usize)hash & mask; for (usize i = 0; i < s->capacity; i++) { Type *key = s->keys[hash_index].type; - if (are_types_identical(key, pair.type)) { + if (are_types_identical_unique_tuples(key, pair.type)) { return hash_index; } else if (key == 0) { return -1; @@ -154,7 +163,7 @@ gb_internal isize type_set__find(TypeSet *s, Type *ptr) { usize hash_index = cast(usize)hash & mask; for (usize i = 0; i < s->capacity; i++) { Type *key = s->keys[hash_index].type; - if (are_types_identical(key, ptr)) { + if (are_types_identical_unique_tuples(key, ptr)) { return hash_index; } else if (key == 0) { return -1; @@ -224,7 +233,7 @@ gb_internal bool type_set_update(TypeSet *s, TypeInfoPair pair) { // returns tru GB_ASSERT(hash_index < s->capacity); for (usize i = 0; i < s->capacity; i++) { TypeInfoPair *key = &s->keys[hash_index]; - GB_ASSERT(!are_types_identical(key->type, pair.type)); + GB_ASSERT(!are_types_identical_unique_tuples(key->type, pair.type)); if (key->hash == TYPE_SET_TOMBSTONE || key->hash == 0) { *key = pair; s->count++; @@ -274,6 +283,9 @@ gb_internal gbString write_canonical_params(gbString w, Type *params) { if (i > 0) { w = gb_string_appendc(w, CANONICAL_PARAM_SEPARATOR); } + w = gb_string_append_length(w, v->token.string.text, v->token.string.len); + w = gb_string_appendc(w, CANONICAL_TYPE_SEPARATOR); + if (v->kind == Entity_Variable) { if (v->flags&EntityFlag_CVarArg) { w = gb_string_appendc(w, CANONICAL_PARAM_C_VARARG); @@ -466,14 +478,17 @@ write_base_name: switch (e->kind) { case Entity_TypeName: { + Type *params = nullptr; Entity *parent = type_get_polymorphic_parent(e->type, ¶ms); - if (parent) { + if (parent && (parent->token.string == e->token.string)) { w = gb_string_append_length(w, parent->token.string.text, parent->token.string.len); w = write_canonical_params(w, params); } else { w = gb_string_append_length(w, e->token.string.text, e->token.string.len); } + gb_unused(parent); + } // Handle parapoly stuff here? return w; -- cgit v1.2.3