From 2820bbc269bb989052163544f2764bc4287254e4 Mon Sep 17 00:00:00 2001 From: gingerBill Date: Fri, 5 Jan 2024 13:38:30 +0000 Subject: Add `@(entry_point_only)` for procedures --- src/entity.cpp | 1 + 1 file changed, 1 insertion(+) (limited to 'src/entity.cpp') diff --git a/src/entity.cpp b/src/entity.cpp index ce27da3f2..d0b3cf139 100644 --- a/src/entity.cpp +++ b/src/entity.cpp @@ -250,6 +250,7 @@ struct Entity { bool is_export : 1; bool generated_from_polymorphic : 1; bool target_feature_disabled : 1; + bool entry_point_only : 1; String target_feature; } Procedure; struct { -- cgit v1.2.3 From aff8f06e3cb1bdb9c3ffee98a68675c843df78fd Mon Sep 17 00:00:00 2001 From: gingerBill Date: Sun, 7 Jan 2024 19:56:00 +0000 Subject: Add frontend stuff instrumentation tooling //+no-instrumentation @(no_instrumentation) @(instrumentation_enter) @(instrumentation_exit) --- src/check_decl.cpp | 66 ++++++++++++++++++++++++++++++++++++++++++++++++++++++ src/checker.cpp | 44 ++++++++++++++++++++++++++++++++++++ src/checker.hpp | 35 ++++++++++++++++++++--------- src/entity.cpp | 1 + src/parser.cpp | 4 +++- src/parser.hpp | 2 ++ 6 files changed, 140 insertions(+), 12 deletions(-) (limited to 'src/entity.cpp') diff --git a/src/check_decl.cpp b/src/check_decl.cpp index 8a3f89877..c69d8185e 100644 --- a/src/check_decl.cpp +++ b/src/check_decl.cpp @@ -909,6 +909,72 @@ gb_internal void check_proc_decl(CheckerContext *ctx, Entity *e, DeclInfo *d) { e->Procedure.entry_point_only = ac.entry_point_only; e->Procedure.is_export = ac.is_export; + + bool no_instrumentation = false; + if (pl->body == nullptr) { + no_instrumentation = true; + if (ac.no_instrumentation != Instrumentation_Default) { + error(e->token, "@(no_instrumentation) is not allowed on foreign procedures"); + } + } else { + if (e->file) { + no_instrumentation = (e->file->flags & AstFile_NoInstrumentation) != 0; + } + + switch (ac.no_instrumentation) { + case Instrumentation_Enabled: no_instrumentation = false; break; + case Instrumentation_Default: break; + case Instrumentation_Disabled: no_instrumentation = true; break; + } + } + e->Procedure.no_instrumentation = no_instrumentation; + + auto const is_valid_instrumentation_call = [](Type *type) -> bool { + if (type == nullptr || type->kind != Type_Proc) { + return false; + } + if (type->Proc.calling_convention != ProcCC_CDecl) { + return false; + } + if (type->Proc.result_count != 0) { + return false; + } + if (type->Proc.param_count != 2) { + return false; + } + Type *p0 = type->Proc.params->Tuple.variables[0]->type; + Type *p1 = type->Proc.params->Tuple.variables[1]->type; + return is_type_rawptr(p0) && is_type_rawptr(p1); + }; + + if (ac.instrumentation_enter && ac.instrumentation_exit) { + error(e->token, "A procedure cannot be marked with both @(instrumentation_enter) and @(instrumentation_exit)"); + } else if (ac.instrumentation_enter) { + if (!is_valid_instrumentation_call(e->type)) { + gbString s = type_to_string(e->type); + error(e->token, "@(instrumentation_enter) procedures must have the type 'proc \"c\" (rawptr, rawptr)', got %s", s); + gb_string_free(s); + } + MUTEX_GUARD(&ctx->info->instrumentation_mutex); + if (ctx->info->instrumentation_enter_entity != nullptr) { + error(e->token, "@(instrumentation_enter) has already been set"); + } else { + ctx->info->instrumentation_enter_entity = e; + } + } else if (ac.instrumentation_exit) { + if (!is_valid_instrumentation_call(e->type)) { + gbString s = type_to_string(e->type); + error(e->token, "@(instrumentation_exit) procedures must have the type 'proc \"c\" (rawptr, rawptr)', got %s", s); + gb_string_free(s); + } + MUTEX_GUARD(&ctx->info->instrumentation_mutex); + if (ctx->info->instrumentation_exit_entity != nullptr) { + error(e->token, "@(instrumentation_exit) has already been set"); + } else { + ctx->info->instrumentation_exit_entity = e; + } + } + e->deprecated_message = ac.deprecated_message; e->warning_message = ac.warning_message; ac.link_name = handle_link_name(ctx, e->token, ac.link_name, ac.link_prefix); diff --git a/src/checker.cpp b/src/checker.cpp index d25acb15e..5e46e87fe 100644 --- a/src/checker.cpp +++ b/src/checker.cpp @@ -2581,6 +2581,9 @@ gb_internal void generate_minimum_dependency_set(Checker *c, Entity *start) { str_lit("multi_pointer_slice_expr_error"), ); + add_dependency_to_set(c, c->info.instrumentation_enter_entity); + add_dependency_to_set(c, c->info.instrumentation_exit_entity); + generate_minimum_dependency_set_internal(c, start); @@ -3414,8 +3417,38 @@ gb_internal DECL_ATTRIBUTE_PROC(proc_decl_attribute) { } return true; } else if (name == "entry_point_only") { + if (value != nullptr) { + error(value, "'%.*s' expects no parameter", LIT(name)); + } ac->entry_point_only = true; return true; + } else if (name == "no_instrumentation") { + ExactValue ev = check_decl_attribute_value(c, value); + if (ev.kind == ExactValue_Invalid) { + ac->no_instrumentation = Instrumentation_Disabled; + } else if (ev.kind == ExactValue_Bool) { + if (ev.value_bool) { + ac->no_instrumentation = Instrumentation_Disabled; + } else { + ac->no_instrumentation = Instrumentation_Enabled; + } + } else { + error(value, "Expected either a boolean or no parameter for '%.*s'", LIT(name)); + return false; + } + return true; + } else if (name == "instrumentation_enter") { + if (value != nullptr) { + error(value, "'%.*s' expects no parameter", LIT(name)); + } + ac->instrumentation_enter = true; + return true; + } else if (name == "instrumentation_exit") { + if (value != nullptr) { + error(value, "'%.*s' expects no parameter", LIT(name)); + } + ac->instrumentation_exit = true; + return true; } return false; } @@ -6216,6 +6249,17 @@ gb_internal void check_parsed_files(Checker *c) { GB_ASSERT(c->info.entity_queue.count.load(std::memory_order_relaxed) == 0); GB_ASSERT(c->info.definition_queue.count.load(std::memory_order_relaxed) == 0); + TIME_SECTION("check instrumentation calls"); + { + if ((c->info.instrumentation_enter_entity != nullptr) ^ + (c->info.instrumentation_exit_entity != nullptr)) { + Entity *e = c->info.instrumentation_enter_entity; + if (!e) e = c->info.instrumentation_exit_entity; + error(e->token, "Both @(instrumentation_enter) and @(instrumentation_exit) must be defined"); + } + } + + TIME_SECTION("add untyped expression values"); // Add untyped expression values for (UntypedExprInfo u = {}; mpsc_dequeue(&c->global_untyped_queue, &u); /**/) { diff --git a/src/checker.hpp b/src/checker.hpp index 7c399e50f..9da0f2950 100644 --- a/src/checker.hpp +++ b/src/checker.hpp @@ -103,6 +103,12 @@ struct DeferredProcedure { }; +enum InstrumentationFlag : i32 { + Instrumentation_Enabled = -1, + Instrumentation_Default = 0, + Instrumentation_Disabled = +1, +}; + struct AttributeContext { String link_name; String link_prefix; @@ -113,20 +119,23 @@ struct AttributeContext { String deprecated_message; String warning_message; DeferredProcedure deferred_procedure; - bool is_export : 1; - bool is_static : 1; - bool require_results : 1; - bool require_declaration : 1; - bool has_disabled_proc : 1; - bool disabled_proc : 1; - bool test : 1; - bool init : 1; - bool fini : 1; - bool set_cold : 1; - bool entry_point_only : 1; + bool is_export : 1; + bool is_static : 1; + bool require_results : 1; + bool require_declaration : 1; + bool has_disabled_proc : 1; + bool disabled_proc : 1; + bool test : 1; + bool init : 1; + bool fini : 1; + bool set_cold : 1; + bool entry_point_only : 1; + bool instrumentation_enter : 1; + bool instrumentation_exit : 1; u32 optimization_mode; // ProcedureOptimizationMode i64 foreign_import_priority_index; String extra_linker_flags; + InstrumentationFlag no_instrumentation; String objc_class; String objc_name; @@ -403,6 +412,10 @@ struct CheckerInfo { BlockingMutex all_procedures_mutex; Array all_procedures; + + BlockingMutex instrumentation_mutex; + Entity *instrumentation_enter_entity; + Entity *instrumentation_exit_entity; }; struct CheckerContext { diff --git a/src/entity.cpp b/src/entity.cpp index d0b3cf139..0539386d0 100644 --- a/src/entity.cpp +++ b/src/entity.cpp @@ -251,6 +251,7 @@ struct Entity { bool generated_from_polymorphic : 1; bool target_feature_disabled : 1; bool entry_point_only : 1; + bool no_instrumentation : 1; String target_feature; } Procedure; struct { diff --git a/src/parser.cpp b/src/parser.cpp index c0498b425..2671054df 100644 --- a/src/parser.cpp +++ b/src/parser.cpp @@ -5919,7 +5919,7 @@ gb_internal bool parse_file(Parser *p, AstFile *f) { f->vet_flags = parse_vet_tag(tok, lc); f->vet_flags_set = true; } else if (string_starts_with(lc, str_lit("+ignore"))) { - return false; + return false; } else if (string_starts_with(lc, str_lit("+private"))) { f->flags |= AstFile_IsPrivatePkg; String command = string_trim_starts_with(lc, str_lit("+private ")); @@ -5941,6 +5941,8 @@ gb_internal bool parse_file(Parser *p, AstFile *f) { } else { f->flags |= AstFile_IsLazy; } + } else if (lc == "+no-instrumentation") { + f->flags |= AstFile_NoInstrumentation; } else { warning(tok, "Ignoring unknown tag '%.*s'", LIT(lc)); } diff --git a/src/parser.hpp b/src/parser.hpp index bce818652..cc1836ef3 100644 --- a/src/parser.hpp +++ b/src/parser.hpp @@ -76,6 +76,8 @@ enum AstFileFlag : u32 { AstFile_IsTest = 1<<3, AstFile_IsLazy = 1<<4, + + AstFile_NoInstrumentation = 1<<5, }; enum AstDelayQueueKind { -- cgit v1.2.3 From f4782157d3b586a88983d3b770c2c0eeb17946d1 Mon Sep 17 00:00:00 2001 From: gingerBill Date: Sun, 7 Jan 2024 21:34:44 +0000 Subject: Implement instrumentation pass --- core/runtime/core.odin | 1 + core/runtime/entry_unix.odin | 1 + core/runtime/entry_wasm.odin | 1 + core/runtime/entry_windows.odin | 1 + core/runtime/error_checks.odin | 24 ++++++----- core/runtime/procs_windows_amd64.odin | 1 + core/runtime/procs_windows_i386.odin | 1 + src/check_decl.cpp | 26 ++++++++---- src/entity.cpp | 2 +- src/llvm_backend.cpp | 2 - src/llvm_backend.hpp | 2 +- src/llvm_backend_general.cpp | 13 ++++++ src/llvm_backend_opt.cpp | 77 +++++++++++++++++++++++++++++++++++ src/llvm_backend_proc.cpp | 12 ++++++ 14 files changed, 143 insertions(+), 21 deletions(-) (limited to 'src/entity.cpp') diff --git a/core/runtime/core.odin b/core/runtime/core.odin index 2d176f909..e12d9a43d 100644 --- a/core/runtime/core.odin +++ b/core/runtime/core.odin @@ -18,6 +18,7 @@ // This could change at a later date if the all these data structures are // implemented within the compiler rather than in this "preload" file // +//+no-instrumentation package runtime import "core:intrinsics" diff --git a/core/runtime/entry_unix.odin b/core/runtime/entry_unix.odin index 78e545c22..f494a509e 100644 --- a/core/runtime/entry_unix.odin +++ b/core/runtime/entry_unix.odin @@ -1,5 +1,6 @@ //+private //+build linux, darwin, freebsd, openbsd +//+no-instrumentation package runtime import "core:intrinsics" diff --git a/core/runtime/entry_wasm.odin b/core/runtime/entry_wasm.odin index 235d5611b..e7f3f156f 100644 --- a/core/runtime/entry_wasm.odin +++ b/core/runtime/entry_wasm.odin @@ -1,5 +1,6 @@ //+private //+build wasm32, wasm64p32 +//+no-instrumentation package runtime import "core:intrinsics" diff --git a/core/runtime/entry_windows.odin b/core/runtime/entry_windows.odin index a315c1209..277daecca 100644 --- a/core/runtime/entry_windows.odin +++ b/core/runtime/entry_windows.odin @@ -1,5 +1,6 @@ //+private //+build windows +//+no-instrumentation package runtime import "core:intrinsics" diff --git a/core/runtime/error_checks.odin b/core/runtime/error_checks.odin index 9d484979a..ea6333c29 100644 --- a/core/runtime/error_checks.odin +++ b/core/runtime/error_checks.odin @@ -1,5 +1,6 @@ package runtime +@(no_instrumentation) bounds_trap :: proc "contextless" () -> ! { when ODIN_OS == .Windows { windows_trap_array_bounds() @@ -8,6 +9,7 @@ bounds_trap :: proc "contextless" () -> ! { } } +@(no_instrumentation) type_assertion_trap :: proc "contextless" () -> ! { when ODIN_OS == .Windows { windows_trap_type_assertion() @@ -21,7 +23,7 @@ bounds_check_error :: proc "contextless" (file: string, line, column: i32, index if uint(index) < uint(count) { return } - @(cold) + @(cold, no_instrumentation) handle_error :: proc "contextless" (file: string, line, column: i32, index, count: int) -> ! { print_caller_location(Source_Code_Location{file, line, column, ""}) print_string(" Index ") @@ -34,6 +36,7 @@ bounds_check_error :: proc "contextless" (file: string, line, column: i32, index handle_error(file, line, column, index, count) } +@(no_instrumentation) slice_handle_error :: proc "contextless" (file: string, line, column: i32, lo, hi: int, len: int) -> ! { print_caller_location(Source_Code_Location{file, line, column, ""}) print_string(" Invalid slice indices ") @@ -46,6 +49,7 @@ slice_handle_error :: proc "contextless" (file: string, line, column: i32, lo, h bounds_trap() } +@(no_instrumentation) multi_pointer_slice_handle_error :: proc "contextless" (file: string, line, column: i32, lo, hi: int) -> ! { print_caller_location(Source_Code_Location{file, line, column, ""}) print_string(" Invalid slice indices ") @@ -82,7 +86,7 @@ dynamic_array_expr_error :: proc "contextless" (file: string, line, column: i32, if 0 <= low && low <= high && high <= max { return } - @(cold) + @(cold, no_instrumentation) handle_error :: proc "contextless" (file: string, line, column: i32, low, high, max: int) -> ! { print_caller_location(Source_Code_Location{file, line, column, ""}) print_string(" Invalid dynamic array indices ") @@ -103,7 +107,7 @@ matrix_bounds_check_error :: proc "contextless" (file: string, line, column: i32 uint(column_index) < uint(column_count) { return } - @(cold) + @(cold, no_instrumentation) handle_error :: proc "contextless" (file: string, line, column: i32, row_index, column_index, row_count, column_count: int) -> ! { print_caller_location(Source_Code_Location{file, line, column, ""}) print_string(" Matrix indices [") @@ -127,7 +131,7 @@ when ODIN_NO_RTTI { if ok { return } - @(cold) + @(cold, no_instrumentation) handle_error :: proc "contextless" (file: string, line, column: i32) -> ! { print_caller_location(Source_Code_Location{file, line, column, ""}) print_string(" Invalid type assertion\n") @@ -140,7 +144,7 @@ when ODIN_NO_RTTI { if ok { return } - @(cold) + @(cold, no_instrumentation) handle_error :: proc "contextless" (file: string, line, column: i32) -> ! { print_caller_location(Source_Code_Location{file, line, column, ""}) print_string(" Invalid type assertion\n") @@ -153,7 +157,7 @@ when ODIN_NO_RTTI { if ok { return } - @(cold) + @(cold, no_instrumentation) handle_error :: proc "contextless" (file: string, line, column: i32, from, to: typeid) -> ! { print_caller_location(Source_Code_Location{file, line, column, ""}) print_string(" Invalid type assertion from ") @@ -198,7 +202,7 @@ when ODIN_NO_RTTI { return id } - @(cold) + @(cold, no_instrumentation) handle_error :: proc "contextless" (file: string, line, column: i32, from, to: typeid, from_data: rawptr) -> ! { actual := variant_type(from, from_data) @@ -224,7 +228,7 @@ make_slice_error_loc :: #force_inline proc "contextless" (loc := #caller_locatio if 0 <= len { return } - @(cold) + @(cold, no_instrumentation) handle_error :: proc "contextless" (loc: Source_Code_Location, len: int) -> ! { print_caller_location(loc) print_string(" Invalid slice length for make: ") @@ -239,7 +243,7 @@ make_dynamic_array_error_loc :: #force_inline proc "contextless" (loc := #caller if 0 <= len && len <= cap { return } - @(cold) + @(cold, no_instrumentation) handle_error :: proc "contextless" (loc: Source_Code_Location, len, cap: int) -> ! { print_caller_location(loc) print_string(" Invalid dynamic array parameters for make: ") @@ -256,7 +260,7 @@ make_map_expr_error_loc :: #force_inline proc "contextless" (loc := #caller_loca if 0 <= cap { return } - @(cold) + @(cold, no_instrumentation) handle_error :: proc "contextless" (loc: Source_Code_Location, cap: int) -> ! { print_caller_location(loc) print_string(" Invalid map capacity for make: ") diff --git a/core/runtime/procs_windows_amd64.odin b/core/runtime/procs_windows_amd64.odin index e430357be..a30985d5c 100644 --- a/core/runtime/procs_windows_amd64.odin +++ b/core/runtime/procs_windows_amd64.odin @@ -1,4 +1,5 @@ //+private +//+no-instrumentation package runtime foreign import kernel32 "system:Kernel32.lib" diff --git a/core/runtime/procs_windows_i386.odin b/core/runtime/procs_windows_i386.odin index f810197f1..4f606da8f 100644 --- a/core/runtime/procs_windows_i386.odin +++ b/core/runtime/procs_windows_i386.odin @@ -1,4 +1,5 @@ //+private +//+no-instrumentation package runtime @require foreign import "system:int64.lib" diff --git a/src/check_decl.cpp b/src/check_decl.cpp index c69d8185e..a530945f9 100644 --- a/src/check_decl.cpp +++ b/src/check_decl.cpp @@ -910,24 +910,24 @@ gb_internal void check_proc_decl(CheckerContext *ctx, Entity *e, DeclInfo *d) { e->Procedure.entry_point_only = ac.entry_point_only; e->Procedure.is_export = ac.is_export; - bool no_instrumentation = false; + bool has_instrumentation = false; if (pl->body == nullptr) { - no_instrumentation = true; + has_instrumentation = false; if (ac.no_instrumentation != Instrumentation_Default) { error(e->token, "@(no_instrumentation) is not allowed on foreign procedures"); } } else { - if (e->file) { - no_instrumentation = (e->file->flags & AstFile_NoInstrumentation) != 0; + AstFile *file = e->token.pos.file_id ? global_files[e->token.pos.file_id] : nullptr; + if (file) { + has_instrumentation = (file->flags & AstFile_NoInstrumentation) == 0; } switch (ac.no_instrumentation) { - case Instrumentation_Enabled: no_instrumentation = false; break; + case Instrumentation_Enabled: has_instrumentation = true; break; case Instrumentation_Default: break; - case Instrumentation_Disabled: no_instrumentation = true; break; + case Instrumentation_Disabled: has_instrumentation = false; break; } } - e->Procedure.no_instrumentation = no_instrumentation; auto const is_valid_instrumentation_call = [](Type *type) -> bool { if (type == nullptr || type->kind != Type_Proc) { @@ -949,6 +949,9 @@ gb_internal void check_proc_decl(CheckerContext *ctx, Entity *e, DeclInfo *d) { if (ac.instrumentation_enter && ac.instrumentation_exit) { error(e->token, "A procedure cannot be marked with both @(instrumentation_enter) and @(instrumentation_exit)"); + + has_instrumentation = false; + e->flags |= EntityFlag_Require; } else if (ac.instrumentation_enter) { if (!is_valid_instrumentation_call(e->type)) { gbString s = type_to_string(e->type); @@ -961,6 +964,9 @@ gb_internal void check_proc_decl(CheckerContext *ctx, Entity *e, DeclInfo *d) { } else { ctx->info->instrumentation_enter_entity = e; } + + has_instrumentation = false; + e->flags |= EntityFlag_Require; } else if (ac.instrumentation_exit) { if (!is_valid_instrumentation_call(e->type)) { gbString s = type_to_string(e->type); @@ -973,8 +979,14 @@ gb_internal void check_proc_decl(CheckerContext *ctx, Entity *e, DeclInfo *d) { } else { ctx->info->instrumentation_exit_entity = e; } + + has_instrumentation = false; + e->flags |= EntityFlag_Require; } + e->Procedure.has_instrumentation = has_instrumentation; + + e->deprecated_message = ac.deprecated_message; e->warning_message = ac.warning_message; ac.link_name = handle_link_name(ctx, e->token, ac.link_name, ac.link_prefix); diff --git a/src/entity.cpp b/src/entity.cpp index 0539386d0..e6c46d37e 100644 --- a/src/entity.cpp +++ b/src/entity.cpp @@ -251,7 +251,7 @@ struct Entity { bool generated_from_polymorphic : 1; bool target_feature_disabled : 1; bool entry_point_only : 1; - bool no_instrumentation : 1; + bool has_instrumentation : 1; String target_feature; } Procedure; struct { diff --git a/src/llvm_backend.cpp b/src/llvm_backend.cpp index b90fd8495..0175d039e 100644 --- a/src/llvm_backend.cpp +++ b/src/llvm_backend.cpp @@ -1497,8 +1497,6 @@ gb_internal WORKER_TASK_PROC(lb_llvm_module_pass_worker_proc) { auto passes = array_make(heap_allocator(), 0, 64); defer (array_free(&passes)); - - LLVMPassBuilderOptionsRef pb_options = LLVMCreatePassBuilderOptions(); defer (LLVMDisposePassBuilderOptions(pb_options)); diff --git a/src/llvm_backend.hpp b/src/llvm_backend.hpp index 4e193bcea..b645d66d6 100644 --- a/src/llvm_backend.hpp +++ b/src/llvm_backend.hpp @@ -563,7 +563,7 @@ gb_internal LLVMTypeRef OdinLLVMGetVectorElementType(LLVMTypeRef type); gb_internal String lb_filepath_ll_for_module(lbModule *m); - +gb_internal LLVMTypeRef lb_type_internal_for_procedures_raw(lbModule *m, Type *type); gb_internal LLVMTypeRef llvm_array_type(LLVMTypeRef ElementType, uint64_t ElementCount) { #if LB_USE_NEW_PASS_SYSTEM diff --git a/src/llvm_backend_general.cpp b/src/llvm_backend_general.cpp index 54327cc54..f0f5327c6 100644 --- a/src/llvm_backend_general.cpp +++ b/src/llvm_backend_general.cpp @@ -2348,6 +2348,15 @@ gb_internal LLVMAttributeRef lb_create_enum_attribute(LLVMContextRef ctx, char c return LLVMCreateEnumAttribute(ctx, kind, value); } +gb_internal LLVMAttributeRef lb_create_string_attribute(LLVMContextRef ctx, String const &key, String const &value) { + LLVMAttributeRef attr = LLVMCreateStringAttribute( + ctx, + cast(char const *)key.text, cast(unsigned)key.len, + cast(char const *)value.text, cast(unsigned)value.len); + return attr; +} + + gb_internal void lb_add_proc_attribute_at_index(lbProcedure *p, isize index, char const *name, u64 value) { LLVMAttributeRef attr = lb_create_enum_attribute(p->module->ctx, name, value); GB_ASSERT(attr != nullptr); @@ -2361,6 +2370,10 @@ gb_internal void lb_add_proc_attribute_at_index(lbProcedure *p, isize index, cha gb_internal void lb_add_attribute_to_proc(lbModule *m, LLVMValueRef proc_value, char const *name, u64 value=0) { LLVMAddAttributeAtIndex(proc_value, LLVMAttributeIndex_FunctionIndex, lb_create_enum_attribute(m->ctx, name, value)); } +gb_internal void lb_add_attribute_to_proc_with_string(lbModule *m, LLVMValueRef proc_value, String const &name, String const &value) { + LLVMAttributeRef attr = lb_create_string_attribute(m->ctx, name, value); + LLVMAddAttributeAtIndex(proc_value, LLVMAttributeIndex_FunctionIndex, attr); +} diff --git a/src/llvm_backend_opt.cpp b/src/llvm_backend_opt.cpp index 2e03b7974..d5487d259 100644 --- a/src/llvm_backend_opt.cpp +++ b/src/llvm_backend_opt.cpp @@ -380,6 +380,80 @@ gb_internal void lb_run_remove_dead_instruction_pass(lbProcedure *p) { } } +gb_internal LLVMValueRef lb_run_instrumentation_pass_insert_call(lbProcedure *p, Entity *entity, LLVMBuilderRef dummy_builder) { + lbModule *m = p->module; + + lbValue cc = lb_find_procedure_value_from_entity(m, entity); + + LLVMValueRef args[2] = {}; + args[0] = p->value; + + LLVMValueRef returnaddress_args[1] = {}; + + returnaddress_args[0] = LLVMConstInt(LLVMInt32TypeInContext(m->ctx), 0, false); + + char const *instrinsic_name = "llvm.returnaddress"; + unsigned id = LLVMLookupIntrinsicID(instrinsic_name, gb_strlen(instrinsic_name)); + GB_ASSERT_MSG(id != 0, "Unable to find %s", instrinsic_name); + LLVMValueRef ip = LLVMGetIntrinsicDeclaration(m->mod, id, nullptr, 0); + LLVMTypeRef call_type = LLVMIntrinsicGetType(m->ctx, id, nullptr, 0); + args[1] = LLVMBuildCall2(dummy_builder, call_type, ip, returnaddress_args, gb_count_of(returnaddress_args), ""); + + LLVMTypeRef fnp = lb_type_internal_for_procedures_raw(p->module, entity->type); + return LLVMBuildCall2(dummy_builder, fnp, cc.value, args, 2, ""); +} + + +gb_internal void lb_run_instrumentation_pass(lbProcedure *p) { + lbModule *m = p->module; + Entity *enter = m->info->instrumentation_enter_entity; + Entity *exit = m->info->instrumentation_exit_entity; + if (enter == nullptr || exit == nullptr) { + return; + } + if (!(p->entity && + p->entity->kind == Entity_Procedure && + p->entity->Procedure.has_instrumentation)) { + return; + } + +#define LLVM_V_NAME(x) x, cast(unsigned)(gb_count_of(x)-1) + + LLVMBuilderRef dummy_builder = LLVMCreateBuilderInContext(m->ctx); + defer (LLVMDisposeBuilder(dummy_builder)); + + LLVMBasicBlockRef entry_bb = p->entry_block->block; + LLVMPositionBuilder(dummy_builder, entry_bb, LLVMGetFirstInstruction(entry_bb)); + lb_run_instrumentation_pass_insert_call(p, enter, dummy_builder); + LLVMRemoveStringAttributeAtIndex(p->value, LLVMAttributeIndex_FunctionIndex, LLVM_V_NAME("instrument-function-entry")); + + unsigned bb_count = LLVMCountBasicBlocks(p->value); + LLVMBasicBlockRef *bbs = gb_alloc_array(temporary_allocator(), LLVMBasicBlockRef, bb_count); + LLVMGetBasicBlocks(p->value, bbs); + for (unsigned i = 0; i < bb_count; i++) { + LLVMBasicBlockRef bb = bbs[i]; + LLVMValueRef terminator = LLVMGetBasicBlockTerminator(bb); + if (terminator == nullptr || + !LLVMIsAReturnInst(terminator)) { + continue; + } + + // TODO(bill): getTerminatingMustTailCall() + // If T is preceded by a musttail call, that's the real terminator. + // if (CallInst *CI = BB.getTerminatingMustTailCall()) + // T = CI; + + + LLVMPositionBuilderBefore(dummy_builder, terminator); + lb_run_instrumentation_pass_insert_call(p, exit, dummy_builder); + } + + LLVMRemoveStringAttributeAtIndex(p->value, LLVMAttributeIndex_FunctionIndex, LLVM_V_NAME("instrument-function-exit")); + +#undef LLVM_V_NAME +} + + gb_internal void lb_run_function_pass_manager(LLVMPassManagerRef fpm, lbProcedure *p, lbFunctionPassManagerKind pass_manager_kind) { if (p == nullptr) { @@ -401,6 +475,7 @@ gb_internal void lb_run_function_pass_manager(LLVMPassManagerRef fpm, lbProcedur } break; } + lb_run_instrumentation_pass(p); LLVMRunFunctionPassManager(fpm, p->value); } @@ -552,3 +627,5 @@ gb_internal void lb_run_remove_unused_globals_pass(lbModule *m) { } } } + + diff --git a/src/llvm_backend_proc.cpp b/src/llvm_backend_proc.cpp index d4eae84bc..09bebd0cf 100644 --- a/src/llvm_backend_proc.cpp +++ b/src/llvm_backend_proc.cpp @@ -329,6 +329,18 @@ gb_internal lbProcedure *lb_create_procedure(lbModule *m, Entity *entity, bool i } } + if (p->body && entity->Procedure.has_instrumentation) { + Entity *instrumentation_enter = m->info->instrumentation_enter_entity; + Entity *instrumentation_exit = m->info->instrumentation_exit_entity; + if (instrumentation_enter && instrumentation_exit) { + String enter = lb_get_entity_name(m, instrumentation_enter); + String exit = lb_get_entity_name(m, instrumentation_exit); + + lb_add_attribute_to_proc_with_string(m, p->value, make_string_c("instrument-function-entry"), enter); + lb_add_attribute_to_proc_with_string(m, p->value, make_string_c("instrument-function-exit"), exit); + } + } + lbValue proc_value = {p->value, p->type}; lb_add_entity(m, entity, proc_value); lb_add_member(m, p->name, proc_value); -- cgit v1.2.3 From 5a84a0822596fac47dd35bf1c2f1d9bb60bbe5c1 Mon Sep 17 00:00:00 2001 From: gingerBill Date: Thu, 22 Feb 2024 17:24:42 +0000 Subject: Add general support for `bit_field`s --- base/runtime/internal.odin | 22 ++++++++++++ core/fmt/fmt.odin | 10 ++++-- src/check_expr.cpp | 86 ++++++++++++++++++++++++++++++++++++++------ src/check_stmt.cpp | 10 ++++++ src/check_type.cpp | 21 +++++++++-- src/checker.hpp | 1 + src/entity.cpp | 2 ++ src/llvm_backend.hpp | 8 +++++ src/llvm_backend_expr.cpp | 16 +++++++++ src/llvm_backend_general.cpp | 51 ++++++++++++++++++++++++-- src/parser.hpp | 1 + src/types.cpp | 34 ++++++++++++++++++ 12 files changed, 245 insertions(+), 17 deletions(-) (limited to 'src/entity.cpp') diff --git a/base/runtime/internal.odin b/base/runtime/internal.odin index 691f76ff1..62bee8620 100644 --- a/base/runtime/internal.odin +++ b/base/runtime/internal.odin @@ -1034,3 +1034,25 @@ fixdfti :: proc(a: u64) -> i128 { } } + + + +__write_bits :: proc "contextless" (dst, src: [^]byte, offset: uintptr, size: uintptr) { + for i in 0..value.kind == ExactValue_Integer) { gbString b = type_to_string(type); i64 sz = type_size_of(type); + i64 bit_size = 8*sz; + bool size_changed = false; + if (max_bit_size > 0) { + size_changed = (bit_size != max_bit_size); + bit_size = gb_min(bit_size, max_bit_size); + } BigInt *bi = &o->value.value_integer; if (is_type_unsigned(type)) { if (big_int_is_neg(bi)) { @@ -2083,25 +2089,36 @@ gb_internal bool check_integer_exceed_suggestion(CheckerContext *c, Operand *o, } else { BigInt one = big_int_make_u64(1); BigInt max_size = big_int_make_u64(1); - BigInt bits = big_int_make_i64(8*sz); + BigInt bits = big_int_make_i64(bit_size); big_int_shl_eq(&max_size, &bits); big_int_sub_eq(&max_size, &one); String max_size_str = big_int_to_string(temporary_allocator(), &max_size); - error_line("\tThe maximum value that can be represented by '%s' is '%.*s'\n", b, LIT(max_size_str)); + + if (size_changed) { + error_line("\tThe maximum value that can be represented with that bit_field's field of '%s | %u' is '%.*s'\n", b, bit_size, LIT(max_size_str)); + } else { + error_line("\tThe maximum value that can be represented by '%s' is '%.*s'\n", b, LIT(max_size_str)); + } } } else { BigInt zero = big_int_make_u64(0); BigInt one = big_int_make_u64(1); BigInt max_size = big_int_make_u64(1); - BigInt bits = big_int_make_i64(8*sz - 1); + BigInt bits = big_int_make_i64(bit_size - 1); big_int_shl_eq(&max_size, &bits); + + String max_size_str = {}; if (big_int_is_neg(bi)) { big_int_neg(&max_size, &max_size); - String max_size_str = big_int_to_string(temporary_allocator(), &max_size); - error_line("\tThe minimum value that can be represented by '%s' is '%.*s'\n", b, LIT(max_size_str)); + max_size_str = big_int_to_string(temporary_allocator(), &max_size); } else { big_int_sub_eq(&max_size, &one); - String max_size_str = big_int_to_string(temporary_allocator(), &max_size); + max_size_str = big_int_to_string(temporary_allocator(), &max_size); + } + + if (size_changed) { + error_line("\tThe maximum value that can be represented with that bit_field's field of '%s | %u' is '%.*s'\n", b, bit_size, LIT(max_size_str)); + } else { error_line("\tThe maximum value that can be represented by '%s' is '%.*s'\n", b, LIT(max_size_str)); } } @@ -2112,7 +2129,7 @@ gb_internal bool check_integer_exceed_suggestion(CheckerContext *c, Operand *o, } return false; } -gb_internal void check_assignment_error_suggestion(CheckerContext *c, Operand *o, Type *type) { +gb_internal void check_assignment_error_suggestion(CheckerContext *c, Operand *o, Type *type, i64 max_bit_size) { gbString a = expr_to_string(o->expr); gbString b = type_to_string(type); defer( @@ -2143,7 +2160,7 @@ gb_internal void check_assignment_error_suggestion(CheckerContext *c, Operand *o error_line("\t whereas slices in general are assumed to be mutable.\n"); } else if (is_type_u8_slice(src) && are_types_identical(dst, t_string) && o->mode != Addressing_Constant) { error_line("\tSuggestion: the expression may be casted to %s\n", b); - } else if (check_integer_exceed_suggestion(c, o, type)) { + } else if (check_integer_exceed_suggestion(c, o, type, max_bit_size)) { return; } } @@ -2217,13 +2234,18 @@ gb_internal bool check_is_expressible(CheckerContext *ctx, Operand *o, Type *typ if (!is_type_integer(o->type) && is_type_integer(type)) { error(o->expr, "'%s' truncated to '%s', got %s", a, b, s); } else { + i64 max_bit_size = 0; + if (ctx->bit_field_bit_size) { + max_bit_size = ctx->bit_field_bit_size; + } + if (are_types_identical(o->type, type)) { error(o->expr, "Numeric value '%s' from '%s' cannot be represented by '%s'", s, a, b); } else { error(o->expr, "Cannot convert numeric value '%s' from '%s' to '%s' from '%s'", s, a, b, c); } - check_assignment_error_suggestion(ctx, o, type); + check_assignment_error_suggestion(ctx, o, type, max_bit_size); } } else { error(o->expr, "Cannot convert '%s' to '%s' from '%s', got %s", a, b, c, s); @@ -2234,6 +2256,11 @@ gb_internal bool check_is_expressible(CheckerContext *ctx, Operand *o, Type *typ } gb_internal bool check_is_not_addressable(CheckerContext *c, Operand *o) { + if (o->expr && o->expr->kind == Ast_SelectorExpr) { + if (o->expr->SelectorExpr.is_bit_field) { + return true; + } + } if (o->mode == Addressing_OptionalOk) { Ast *expr = unselector_expr(o->expr); if (expr->kind != Ast_TypeAssertion) { @@ -2306,6 +2333,8 @@ gb_internal void check_unary_expr(CheckerContext *c, Operand *o, Token op, Ast * Entity *e = entity_of_node(ue->expr); if (e != nullptr && (e->flags & EntityFlag_Param) != 0) { error(op, "Cannot take the pointer address of '%s' which is a procedure parameter", str); + } else if (e != nullptr && (e->flags & EntityFlag_BitFieldField) != 0) { + error(op, "Cannot take the pointer address of '%s' which is a bit_field's field", str); } else { switch (o->mode) { case Addressing_Constant: @@ -5067,6 +5096,11 @@ gb_internal Entity *check_selector(CheckerContext *c, Operand *operand, Ast *nod operand->type = entity->type; operand->expr = node; + if (entity->flags & EntityFlag_BitFieldField) { + add_package_dependency(c, "runtime", "__write_bits"); + add_package_dependency(c, "runtime", "__read_bits"); + } + switch (entity->kind) { case Entity_Constant: operand->value = entity->Constant.value; @@ -5080,6 +5114,9 @@ gb_internal Entity *check_selector(CheckerContext *c, Operand *operand, Ast *nod } break; case Entity_Variable: + if (sel.is_bit_field) { + se->is_bit_field = true; + } if (sel.indirect) { operand->mode = Addressing_Variable; } else if (operand->mode == Addressing_Context) { @@ -11115,6 +11152,33 @@ gb_internal gbString write_expr_to_string(gbString str, Ast *node, bool shorthan case_end; + case_ast_node(f, BitFieldField, node); + str = write_expr_to_string(str, f->name, shorthand); + str = gb_string_appendc(str, ": "); + str = write_expr_to_string(str, f->type, shorthand); + str = gb_string_appendc(str, " | "); + str = write_expr_to_string(str, f->bit_size, shorthand); + case_end; + case_ast_node(bf, BitFieldType, node); + str = gb_string_appendc(str, "bit_field "); + if (!shorthand) { + str = write_expr_to_string(str, bf->backing_type, shorthand); + } + str = gb_string_appendc(str, " {"); + if (shorthand) { + str = gb_string_appendc(str, "..."); + } else { + for_array(i, bf->fields) { + if (i > 0) { + str = gb_string_appendc(str, ", "); + } + str = write_expr_to_string(str, bf->fields[i], false); + } + return str; + } + str = gb_string_appendc(str, "}"); + case_end; + case_ast_node(ia, InlineAsmExpr, node); str = gb_string_appendc(str, "asm("); for_array(i, ia->param_types) { diff --git a/src/check_stmt.cpp b/src/check_stmt.cpp index 6897701d6..a7dd9743b 100644 --- a/src/check_stmt.cpp +++ b/src/check_stmt.cpp @@ -485,7 +485,17 @@ gb_internal Type *check_assignment_variable(CheckerContext *ctx, Operand *lhs, O } } + Entity *lhs_e = entity_of_node(lhs->expr); + u8 prev_bit_field_bit_size = ctx->bit_field_bit_size; + if (lhs_e && lhs_e->kind == Entity_Variable && lhs_e->Variable.bit_field_bit_size) { + // HACK NOTE(bill): This is a bit of a hack, but it will work fine for this use case + ctx->bit_field_bit_size = lhs_e->Variable.bit_field_bit_size; + } + check_assignment(ctx, rhs, assignment_type, str_lit("assignment")); + + ctx->bit_field_bit_size = prev_bit_field_bit_size; + if (rhs->mode == Addressing_Invalid) { return nullptr; } diff --git a/src/check_type.cpp b/src/check_type.cpp index 8afac2fc5..8c746a2f7 100644 --- a/src/check_type.cpp +++ b/src/check_type.cpp @@ -1035,11 +1035,19 @@ gb_internal void check_bit_field_type(CheckerContext *ctx, Type *bit_field_type, error(f->bit_size, "A bit_field's specified bit size cannot exceed 64 bits, got %lld", cast(long long)bit_size_i64); bit_size_i64 = 64; } + i64 sz = 8*type_size_of(type); + if (bit_size_i64 > sz) { + error(f->bit_size, "A bit_field's specified bit size cannot exceed its type, got %lld, expect <=%lld", cast(long long)bit_size_i64, cast(long long)sz); + bit_size_i64 = sz; + } + bit_size_u8 = cast(u8)bit_size_i64; Entity *e = alloc_entity_field(ctx->scope, f->name->Ident.token, type, false, field_src_index); e->Variable.docs = docs; e->Variable.comment = comment; + e->Variable.bit_field_bit_size = bit_size_u8; + e->flags |= EntityFlag_BitFieldField; add_entity(ctx, ctx->scope, nullptr, e); array_add(&fields, e); @@ -1050,6 +1058,14 @@ gb_internal void check_bit_field_type(CheckerContext *ctx, Type *bit_field_type, GB_ASSERT(fields.count <= bf->fields.count); + auto bit_offsets = slice_make(permanent_allocator(), fields.count); + i64 curr_offset = 0; + for_array(i, bit_sizes) { + bit_offsets[i] = curr_offset; + curr_offset += cast(i64)bit_sizes[i]; + } + + if (total_bit_size > maximum_bit_size) { gbString s = type_to_string(backing_type); error(node, "The numbers required %llu exceeds the backing type's (%s) bit size %llu", @@ -1059,8 +1075,9 @@ gb_internal void check_bit_field_type(CheckerContext *ctx, Type *bit_field_type, gb_string_free(s); } - bit_field_type->BitField.fields = slice_from_array(fields); - bit_field_type->BitField.bit_sizes = slice_from_array(bit_sizes); + bit_field_type->BitField.fields = slice_from_array(fields); + bit_field_type->BitField.bit_sizes = slice_from_array(bit_sizes); + bit_field_type->BitField.bit_offsets = bit_offsets; } gb_internal bool is_type_valid_bit_set_range(Type *t) { diff --git a/src/checker.hpp b/src/checker.hpp index 9aee82257..066d6bb4a 100644 --- a/src/checker.hpp +++ b/src/checker.hpp @@ -475,6 +475,7 @@ struct CheckerContext { bool hide_polymorphic_errors; bool in_polymorphic_specialization; bool allow_arrow_right_selector_expr; + u8 bit_field_bit_size; Scope * polymorphic_scope; Ast *assignment_lhs_hint; diff --git a/src/entity.cpp b/src/entity.cpp index e6c46d37e..916c2b2bd 100644 --- a/src/entity.cpp +++ b/src/entity.cpp @@ -43,6 +43,7 @@ enum EntityFlag : u64 { EntityFlag_NoAlias = 1ull<<9, EntityFlag_TypeField = 1ull<<10, EntityFlag_Value = 1ull<<11, + EntityFlag_BitFieldField = 1ull<<12, @@ -212,6 +213,7 @@ struct Entity { Ast *init_expr; // only used for some variables within procedure bodies i32 field_index; i32 field_group_index; + u8 bit_field_bit_size; ParameterValue param_value; diff --git a/src/llvm_backend.hpp b/src/llvm_backend.hpp index 026454c81..00d1b7a21 100644 --- a/src/llvm_backend.hpp +++ b/src/llvm_backend.hpp @@ -84,6 +84,8 @@ enum lbAddrKind { lbAddr_Swizzle, lbAddr_SwizzleLarge, + + lbAddr_BitField, }; struct lbAddr { @@ -118,6 +120,12 @@ struct lbAddr { Type *type; Slice indices; } swizzle_large; + struct { + Type *type; + i64 index; + i64 bit_offset; + i64 bit_size; + } bitfield; }; }; diff --git a/src/llvm_backend_expr.cpp b/src/llvm_backend_expr.cpp index 0c06c8c1b..6bef21822 100644 --- a/src/llvm_backend_expr.cpp +++ b/src/llvm_backend_expr.cpp @@ -4627,6 +4627,22 @@ gb_internal lbAddr lb_build_addr_internal(lbProcedure *p, Ast *expr) { Selection sel = lookup_field(type, selector, false); GB_ASSERT(sel.entity != nullptr); + if (sel.is_bit_field) { + lbAddr addr = lb_build_addr(p, se->expr); + Type *bf_type = base_type(type_deref(lb_addr_type(addr))); + GB_ASSERT(bf_type->kind == Type_BitField); + + lbValue a = lb_addr_get_ptr(p, addr); + Selection sub_sel = sel; + sub_sel.index.count -= 1; + i32 index = sel.index[sel.index.count-1]; + + Entity *f = bf_type->BitField.fields[index]; + u8 bit_size = bf_type->BitField.bit_sizes[index]; + i64 bit_offset = bf_type->BitField.bit_offsets[index]; + + return lb_addr_bit_field(a, f->type, index, bit_offset, bit_size); + } if (sel.pseudo_field) { GB_ASSERT(sel.entity->kind == Entity_Procedure || sel.entity->kind == Entity_ProcGroup); Entity *e = entity_of_node(sel_node); diff --git a/src/llvm_backend_general.cpp b/src/llvm_backend_general.cpp index 2102420f8..4ff8482a7 100644 --- a/src/llvm_backend_general.cpp +++ b/src/llvm_backend_general.cpp @@ -451,6 +451,20 @@ gb_internal lbAddr lb_addr_swizzle_large(lbValue addr, Type *array_type, Slice(temporary_allocator(), 4); + args[0] = dst; + args[1] = lb_address_from_load_or_generate_local(p, value); + args[2] = lb_const_int(p->module, t_uintptr, addr.bitfield.bit_offset); + args[3] = lb_const_int(p->module, t_uintptr, addr.bitfield.bit_size); + lb_emit_runtime_call(p, "__write_bits", args); + return; + } else if (addr.kind == lbAddr_RelativePointer) { Type *rel_ptr = base_type(lb_addr_type(addr)); GB_ASSERT(rel_ptr->kind == Type_RelativePointer || rel_ptr->kind == Type_RelativeMultiPointer); @@ -1074,8 +1098,31 @@ gb_internal lbValue lb_emit_load(lbProcedure *p, lbValue value) { gb_internal lbValue lb_addr_load(lbProcedure *p, lbAddr const &addr) { GB_ASSERT(addr.addr.value != nullptr); + if (addr.kind == lbAddr_BitField) { + lbAddr dst = lb_add_local_generated(p, addr.bitfield.type, true); + lbValue src = addr.addr; - if (addr.kind == lbAddr_RelativePointer) { + auto args = array_make(temporary_allocator(), 4); + args[0] = dst.addr; + args[1] = src; + args[2] = lb_const_int(p->module, t_uintptr, addr.bitfield.bit_offset); + args[3] = lb_const_int(p->module, t_uintptr, addr.bitfield.bit_size); + lb_emit_runtime_call(p, "__read_bits", args); + + lbValue r = lb_addr_load(p, dst); + + if (!is_type_unsigned(core_type(addr.bitfield.type))) { + // Sign extension + // m := 1<<(bit_size-1) + // r = (r XOR m) - m + Type *t = addr.bitfield.type; + lbValue m = lb_const_int(p->module, t, 1ull<<(addr.bitfield.bit_size-1)); + r = lb_emit_arith(p, Token_Xor, r, m, t); + r = lb_emit_arith(p, Token_Sub, r, m, t); + } + + return r; + } else if (addr.kind == lbAddr_RelativePointer) { Type *rel_ptr = base_type(lb_addr_type(addr)); Type *base_integer = nullptr; Type *pointer_type = nullptr; diff --git a/src/parser.hpp b/src/parser.hpp index ff77c88c7..1f4ec8726 100644 --- a/src/parser.hpp +++ b/src/parser.hpp @@ -429,6 +429,7 @@ AST_KIND(_ExprBegin, "", bool) \ Ast *expr, *selector; \ u8 swizzle_count; /*maximum of 4 components, if set, count >= 2*/ \ u8 swizzle_indices; /*2 bits per component*/ \ + bool is_bit_field; \ }) \ AST_KIND(ImplicitSelectorExpr, "implicit selector expression", struct { Token token; Ast *selector; }) \ AST_KIND(SelectorCallExpr, "selector call expression", struct { \ diff --git a/src/types.cpp b/src/types.cpp index 1c28e6583..be4b8944b 100644 --- a/src/types.cpp +++ b/src/types.cpp @@ -287,6 +287,7 @@ struct TypeProc { Type * backing_type; \ Slice fields; \ Slice bit_sizes; \ + Slice bit_offsets; \ Ast * node; \ }) \ TYPE_KIND(SoaPointer, struct { Type *elem; }) @@ -408,6 +409,7 @@ struct Selection { bool indirect; // Set if there was a pointer deref anywhere down the line u8 swizzle_count; // maximum components = 4 u8 swizzle_indices; // 2 bits per component, representing which swizzle index + bool is_bit_field; bool pseudo_field; }; gb_global Selection const empty_selection = {0}; @@ -3187,6 +3189,21 @@ gb_internal Selection lookup_field_with_selection(Type *type_, String field_name else if (field_name == "a") mapped_field_name = str_lit("w"); return lookup_field_with_selection(type, mapped_field_name, is_type, sel, allow_blank_ident); } + } else if (type->kind == Type_BitField) { + for_array(i, type->BitField.fields) { + Entity *f = type->BitField.fields[i]; + if (f->kind != Entity_Variable || (f->flags & EntityFlag_Field) == 0) { + continue; + } + String str = f->token.string; + if (field_name == str) { + selection_add_index(&sel, i); // HACK(bill): Leaky memory + sel.entity = f; + sel.is_bit_field = true; + return sel; + } + } + } else if (type->kind == Type_Basic) { switch (type->Basic.kind) { case Basic_any: { @@ -4551,6 +4568,23 @@ gb_internal gbString write_type_to_string(gbString str, Type *type, bool shortha str = gb_string_appendc(str, gb_bprintf("matrix[%d, %d]", cast(int)type->Matrix.row_count, cast(int)type->Matrix.column_count)); str = write_type_to_string(str, type->Matrix.elem); break; + + case Type_BitField: + str = gb_string_appendc(str, "bit_field "); + str = write_type_to_string(str, type->BitField.backing_type); + str = gb_string_appendc(str, " {"); + for (isize i = 0; i < type->BitField.fields.count; i++) { + Entity *f = type->BitField.fields[i]; + if (i > 0) { + str = gb_string_appendc(str, ", "); + } + str = gb_string_append_length(str, f->token.string.text, f->token.string.len); + str = gb_string_appendc(str, ": "); + str = write_type_to_string(str, f->type); + str = gb_string_append_fmt(str, " | %u", type->BitField.bit_sizes[i]); + } + str = gb_string_appendc(str, " }"); + break; } return str; -- cgit v1.2.3 From 29e5f94c2a0d666eed93a1013f895f3c86d6373f Mon Sep 17 00:00:00 2001 From: gingerBill Date: Thu, 21 Mar 2024 11:52:48 +0000 Subject: Add `#no_broadcast` procedure parameter to disallow automatic array programming broadcasting on procedure arguments --- base/runtime/core_builtin.odin | 12 +++---- core/image/common.odin | 36 ++++++++++---------- core/math/big/prime.odin | 2 +- core/odin/ast/ast.odin | 16 +++++---- core/thread/thread.odin | 2 +- src/check_expr.cpp | 76 ++++++++++++++++++++++++++---------------- src/check_type.cpp | 16 ++++++++- src/entity.cpp | 1 + src/parser.cpp | 17 +++++----- src/parser.hpp | 3 +- vendor/raylib/raymath.odin | 4 +-- 11 files changed, 113 insertions(+), 72 deletions(-) (limited to 'src/entity.cpp') diff --git a/base/runtime/core_builtin.odin b/base/runtime/core_builtin.odin index 401dcb857..fba2e1328 100644 --- a/base/runtime/core_builtin.odin +++ b/base/runtime/core_builtin.odin @@ -447,12 +447,12 @@ _append_elem :: #force_inline proc(array: ^$T/[dynamic]$E, arg: E, should_zero: } @builtin -append_elem :: proc(array: ^$T/[dynamic]$E, arg: E, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error { +append_elem :: proc(array: ^$T/[dynamic]$E, #no_broadcast arg: E, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error { return _append_elem(array, arg, true, loc=loc) } @builtin -non_zero_append_elem :: proc(array: ^$T/[dynamic]$E, arg: E, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error { +non_zero_append_elem :: proc(array: ^$T/[dynamic]$E, #no_broadcast arg: E, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error { return _append_elem(array, arg, false, loc=loc) } @@ -496,12 +496,12 @@ _append_elems :: #force_inline proc(array: ^$T/[dynamic]$E, should_zero: bool, l } @builtin -append_elems :: proc(array: ^$T/[dynamic]$E, args: ..E, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error { +append_elems :: proc(array: ^$T/[dynamic]$E, #no_broadcast args: ..E, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error { return _append_elems(array, true, loc, ..args) } @builtin -non_zero_append_elems :: proc(array: ^$T/[dynamic]$E, args: ..E, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error { +non_zero_append_elems :: proc(array: ^$T/[dynamic]$E, #no_broadcast args: ..E, loc := #caller_location) -> (n: int, err: Allocator_Error) #optional_allocator_error { return _append_elems(array, false, loc, ..args) } @@ -556,7 +556,7 @@ append_nothing :: proc(array: ^$T/[dynamic]$E, loc := #caller_location) -> (n: i @builtin -inject_at_elem :: proc(array: ^$T/[dynamic]$E, index: int, arg: E, loc := #caller_location) -> (ok: bool, err: Allocator_Error) #no_bounds_check #optional_allocator_error { +inject_at_elem :: proc(array: ^$T/[dynamic]$E, index: int, #no_broadcast arg: E, loc := #caller_location) -> (ok: bool, err: Allocator_Error) #no_bounds_check #optional_allocator_error { if array == nil { return } @@ -574,7 +574,7 @@ inject_at_elem :: proc(array: ^$T/[dynamic]$E, index: int, arg: E, loc := #calle } @builtin -inject_at_elems :: proc(array: ^$T/[dynamic]$E, index: int, args: ..E, loc := #caller_location) -> (ok: bool, err: Allocator_Error) #no_bounds_check #optional_allocator_error { +inject_at_elems :: proc(array: ^$T/[dynamic]$E, index: int, #no_broadcast args: ..E, loc := #caller_location) -> (ok: bool, err: Allocator_Error) #no_bounds_check #optional_allocator_error { if array == nil { return } diff --git a/core/image/common.odin b/core/image/common.odin index c7507a85f..b576a9521 100644 --- a/core/image/common.odin +++ b/core/image/common.odin @@ -651,7 +651,7 @@ alpha_add_if_missing :: proc(img: ^Image, alpha_key := Alpha_Key{}, allocator := // We have keyed alpha. o: GA_Pixel for p in inp { - if p == key.r { + if p.r == key.r { o = GA_Pixel{0, key.g} } else { o = GA_Pixel{p.r, 255} @@ -710,7 +710,7 @@ alpha_add_if_missing :: proc(img: ^Image, alpha_key := Alpha_Key{}, allocator := // We have keyed alpha. o: GA_Pixel_16 for p in inp { - if p == key.r { + if p.r == key.r { o = GA_Pixel_16{0, key.g} } else { o = GA_Pixel_16{p.r, 65535} @@ -842,11 +842,11 @@ alpha_drop_if_present :: proc(img: ^Image, options := Options{}, alpha_key := Al bg := G_Pixel{} if temp_bg, temp_bg_ok := img.background.(RGB_Pixel_16); temp_bg_ok { // Background is RGB 16-bit, take just the red channel's topmost byte. - bg = u8(temp_bg.r >> 8) + bg.r = u8(temp_bg.r >> 8) } for p in inp { - out[0] = bg if p == key else p + out[0] = bg if p.r == key else p out = out[1:] } @@ -865,8 +865,8 @@ alpha_drop_if_present :: proc(img: ^Image, options := Options{}, alpha_key := Al for p in inp { a := f32(p.g) / 255.0 c := ((1.0 - a) * bg + a * f32(p.r)) - out[0] = u8(c) - out = out[1:] + out[0].r = u8(c) + out = out[1:] } } else if .alpha_premultiply in options { @@ -874,14 +874,14 @@ alpha_drop_if_present :: proc(img: ^Image, options := Options{}, alpha_key := Al for p in inp { a := f32(p.g) / 255.0 c := f32(p.r) * a - out[0] = u8(c) - out = out[1:] + out[0].r = u8(c) + out = out[1:] } } else { // Just drop alpha on the floor. for p in inp { - out[0] = p.r - out = out[1:] + out[0].r = p.r + out = out[1:] } } @@ -951,11 +951,11 @@ alpha_drop_if_present :: proc(img: ^Image, options := Options{}, alpha_key := Al bg := G_Pixel_16{} if temp_bg, temp_bg_ok := img.background.(RGB_Pixel_16); temp_bg_ok { // Background is RGB 16-bit, take just the red channel. - bg = temp_bg.r + bg.r = temp_bg.r } for p in inp { - out[0] = bg if p == key else p + out[0] = bg if p.r == key else p out = out[1:] } @@ -974,8 +974,8 @@ alpha_drop_if_present :: proc(img: ^Image, options := Options{}, alpha_key := Al for p in inp { a := f32(p.g) / 65535.0 c := ((1.0 - a) * bg + a * f32(p.r)) - out[0] = u16(c) - out = out[1:] + out[0].r = u16(c) + out = out[1:] } } else if .alpha_premultiply in options { @@ -983,14 +983,14 @@ alpha_drop_if_present :: proc(img: ^Image, options := Options{}, alpha_key := Al for p in inp { a := f32(p.g) / 65535.0 c := f32(p.r) * a - out[0] = u16(c) - out = out[1:] + out[0].r = u16(c) + out = out[1:] } } else { // Just drop alpha on the floor. for p in inp { - out[0] = p.r - out = out[1:] + out[0].r = p.r + out = out[1:] } } diff --git a/core/math/big/prime.odin b/core/math/big/prime.odin index cb0b08dbb..b02b7cb4e 100644 --- a/core/math/big/prime.odin +++ b/core/math/big/prime.odin @@ -1112,7 +1112,7 @@ internal_int_prime_next_prime :: proc(a: ^Int, trials: int, bbs_style: bool, all Generate the restable. */ for x := 1; x < _PRIME_TAB_SIZE; x += 1 { - res_tab = internal_mod(a, _private_prime_table[x]) or_return + res_tab = cast(type_of(res_tab))(internal_mod(a, _private_prime_table[x]) or_return) } for { diff --git a/core/odin/ast/ast.odin b/core/odin/ast/ast.odin index f6bcbab4e..3e215e0f2 100644 --- a/core/odin/ast/ast.odin +++ b/core/odin/ast/ast.odin @@ -597,6 +597,7 @@ Field_Flag :: enum { Any_Int, Subtype, By_Ptr, + No_Broadcast, Results, Tags, @@ -616,6 +617,7 @@ field_flag_strings := [Field_Flag]string{ .Any_Int = "#any_int", .Subtype = "#subtype", .By_Ptr = "#by_ptr", + .No_Broadcast ="#no_broadcast", .Results = "results", .Tags = "field tag", @@ -624,12 +626,13 @@ field_flag_strings := [Field_Flag]string{ } field_hash_flag_strings := []struct{key: string, flag: Field_Flag}{ - {"no_alias", .No_Alias}, - {"c_vararg", .C_Vararg}, - {"const", .Const}, - {"any_int", .Any_Int}, - {"subtype", .Subtype}, - {"by_ptr", .By_Ptr}, + {"no_alias", .No_Alias}, + {"c_vararg", .C_Vararg}, + {"const", .Const}, + {"any_int", .Any_Int}, + {"subtype", .Subtype}, + {"by_ptr", .By_Ptr}, + {"no_broadcast", .No_Broadcast}, } @@ -650,6 +653,7 @@ Field_Flags_Signature :: Field_Flags{ .Const, .Any_Int, .By_Ptr, + .No_Broadcast, .Default_Parameters, } diff --git a/core/thread/thread.odin b/core/thread/thread.odin index 1c473bd1d..55f73d106 100644 --- a/core/thread/thread.odin +++ b/core/thread/thread.odin @@ -163,7 +163,7 @@ create_and_start_with_data :: proc(data: rawptr, fn: proc(data: rawptr), init_co t := create(thread_proc, priority) t.data = rawptr(fn) t.user_index = 1 - t.user_args = data + t.user_args[0] = data if self_cleanup { t.flags += {.Self_Cleanup} } diff --git a/src/check_expr.cpp b/src/check_expr.cpp index 8fb2cf36b..51fe3fc8a 100644 --- a/src/check_expr.cpp +++ b/src/check_expr.cpp @@ -623,7 +623,7 @@ gb_internal bool check_cast_internal(CheckerContext *c, Operand *x, Type *type); #define MAXIMUM_TYPE_DISTANCE 10 -gb_internal i64 check_distance_between_types(CheckerContext *c, Operand *operand, Type *type) { +gb_internal i64 check_distance_between_types(CheckerContext *c, Operand *operand, Type *type, bool allow_array_programming) { if (c == nullptr) { GB_ASSERT(operand->mode == Addressing_Value); GB_ASSERT(is_type_typed(operand->type)); @@ -832,7 +832,7 @@ gb_internal i64 check_distance_between_types(CheckerContext *c, Operand *operand if (dst->Union.variants.count == 1) { Type *vt = dst->Union.variants[0]; - i64 score = check_distance_between_types(c, operand, vt); + i64 score = check_distance_between_types(c, operand, vt, allow_array_programming); if (score >= 0) { return score+2; } @@ -840,7 +840,7 @@ gb_internal i64 check_distance_between_types(CheckerContext *c, Operand *operand i64 prev_lowest_score = -1; i64 lowest_score = -1; for (Type *vt : dst->Union.variants) { - i64 score = check_distance_between_types(c, operand, vt); + i64 score = check_distance_between_types(c, operand, vt, allow_array_programming); if (score >= 0) { if (lowest_score < 0) { lowest_score = score; @@ -863,14 +863,14 @@ gb_internal i64 check_distance_between_types(CheckerContext *c, Operand *operand } if (is_type_relative_pointer(dst)) { - i64 score = check_distance_between_types(c, operand, dst->RelativePointer.pointer_type); + i64 score = check_distance_between_types(c, operand, dst->RelativePointer.pointer_type, allow_array_programming); if (score >= 0) { return score+2; } } if (is_type_relative_multi_pointer(dst)) { - i64 score = check_distance_between_types(c, operand, dst->RelativeMultiPointer.pointer_type); + i64 score = check_distance_between_types(c, operand, dst->RelativeMultiPointer.pointer_type, allow_array_programming); if (score >= 0) { return score+2; } @@ -896,19 +896,21 @@ gb_internal i64 check_distance_between_types(CheckerContext *c, Operand *operand } } - if (is_type_array(dst)) { - Type *elem = base_array_type(dst); - i64 distance = check_distance_between_types(c, operand, elem); - if (distance >= 0) { - return distance + 6; + if (allow_array_programming) { + if (is_type_array(dst)) { + Type *elem = base_array_type(dst); + i64 distance = check_distance_between_types(c, operand, elem, allow_array_programming); + if (distance >= 0) { + return distance + 6; + } } - } - if (is_type_simd_vector(dst)) { - Type *dst_elem = base_array_type(dst); - i64 distance = check_distance_between_types(c, operand, dst_elem); - if (distance >= 0) { - return distance + 6; + if (is_type_simd_vector(dst)) { + Type *dst_elem = base_array_type(dst); + i64 distance = check_distance_between_types(c, operand, dst_elem, allow_array_programming); + if (distance >= 0) { + return distance + 6; + } } } @@ -918,7 +920,7 @@ gb_internal i64 check_distance_between_types(CheckerContext *c, Operand *operand } if (dst->Matrix.row_count == dst->Matrix.column_count) { Type *dst_elem = base_array_type(dst); - i64 distance = check_distance_between_types(c, operand, dst_elem); + i64 distance = check_distance_between_types(c, operand, dst_elem, allow_array_programming); if (distance >= 0) { return distance + 7; } @@ -966,9 +968,9 @@ gb_internal i64 assign_score_function(i64 distance, bool is_variadic=false) { } -gb_internal bool check_is_assignable_to_with_score(CheckerContext *c, Operand *operand, Type *type, i64 *score_, bool is_variadic=false) { +gb_internal bool check_is_assignable_to_with_score(CheckerContext *c, Operand *operand, Type *type, i64 *score_, bool is_variadic=false, bool allow_array_programming=true) { i64 score = 0; - i64 distance = check_distance_between_types(c, operand, type); + i64 distance = check_distance_between_types(c, operand, type, allow_array_programming); bool ok = distance >= 0; if (ok) { score = assign_score_function(distance, is_variadic); @@ -978,9 +980,9 @@ gb_internal bool check_is_assignable_to_with_score(CheckerContext *c, Operand *o } -gb_internal bool check_is_assignable_to(CheckerContext *c, Operand *operand, Type *type) { +gb_internal bool check_is_assignable_to(CheckerContext *c, Operand *operand, Type *type, bool allow_array_programming=true) { i64 score = 0; - return check_is_assignable_to_with_score(c, operand, type, &score); + return check_is_assignable_to_with_score(c, operand, type, &score, /*is_variadic*/false, allow_array_programming); } gb_internal bool internal_check_is_assignable_to(Type *src, Type *dst) { @@ -3142,6 +3144,14 @@ gb_internal bool check_is_castable_to(CheckerContext *c, Operand *operand, Type return true; } + + if (is_type_array(dst)) { + Type *elem = base_array_type(dst); + if (check_is_castable_to(c, operand, elem)) { + return true; + } + } + if (is_type_simd_vector(src) && is_type_simd_vector(dst)) { if (src->SimdVector.count != dst->SimdVector.count) { return false; @@ -5853,15 +5863,20 @@ gb_internal CallArgumentError check_call_arguments_internal(CheckerContext *c, A } } - auto eval_param_and_score = [](CheckerContext *c, Operand *o, Type *param_type, CallArgumentError &err, bool param_is_variadic, Entity *e, bool show_error) -> i64 { + auto eval_param_and_score = [](CheckerContext *c, Operand *o, Type *param_type, CallArgumentError &err, bool param_is_variadic, Entity *e, bool show_error, bool allow_array_programming) -> i64 { i64 s = 0; - if (!check_is_assignable_to_with_score(c, o, param_type, &s, param_is_variadic)) { + if (!check_is_assignable_to_with_score(c, o, param_type, &s, param_is_variadic, allow_array_programming)) { bool ok = false; - if (e && e->flags & EntityFlag_AnyInt) { + if (e && (e->flags & EntityFlag_AnyInt)) { if (is_type_integer(param_type)) { ok = check_is_castable_to(c, o, param_type); } } + if (!allow_array_programming && check_is_assignable_to_with_score(c, o, param_type, nullptr, param_is_variadic, !allow_array_programming)) { + if (show_error) { + error(o->expr, "'#no_broadcast' disallows automatic broadcasting a value across all elements of an array-like type in a procedure argument"); + } + } if (ok) { s = assign_score_function(MAXIMUM_TYPE_DISTANCE); } else { @@ -5878,7 +5893,6 @@ gb_internal CallArgumentError check_call_arguments_internal(CheckerContext *c, A } err = CallArgumentError_WrongTypes; } - } else if (show_error) { check_assignment(c, o, param_type, str_lit("procedure argument")); } @@ -5963,12 +5977,14 @@ gb_internal CallArgumentError check_call_arguments_internal(CheckerContext *c, A if (param_is_variadic) { continue; } - score += eval_param_and_score(c, o, e->type, err, param_is_variadic, e, show_error); + bool allow_array_programming = !(e && (e->flags & EntityFlag_NoBroadcast)); + score += eval_param_and_score(c, o, e->type, err, param_is_variadic, e, show_error, allow_array_programming); } } if (variadic) { - Type *slice = pt->params->Tuple.variables[pt->variadic_index]->type; + Entity *var_entity = pt->params->Tuple.variables[pt->variadic_index]; + Type *slice = var_entity->type; GB_ASSERT(is_type_slice(slice)); Type *elem = base_type(slice)->Slice.elem; Type *t = elem; @@ -5994,7 +6010,8 @@ gb_internal CallArgumentError check_call_arguments_internal(CheckerContext *c, A return CallArgumentError_MultipleVariadicExpand; } } - score += eval_param_and_score(c, o, t, err, true, nullptr, show_error); + bool allow_array_programming = !(var_entity && (var_entity->flags & EntityFlag_NoBroadcast)); + score += eval_param_and_score(c, o, t, err, true, nullptr, show_error, allow_array_programming); } } @@ -11148,6 +11165,9 @@ gb_internal gbString write_expr_to_string(gbString str, Ast *node, bool shorthan if (f->flags&FieldFlag_any_int) { str = gb_string_appendc(str, "#any_int "); } + if (f->flags&FieldFlag_no_broadcast) { + str = gb_string_appendc(str, "#no_broadcast "); + } if (f->flags&FieldFlag_const) { str = gb_string_appendc(str, "#const "); } diff --git a/src/check_type.cpp b/src/check_type.cpp index 3fe633892..96885bd27 100644 --- a/src/check_type.cpp +++ b/src/check_type.cpp @@ -1869,6 +1869,10 @@ gb_internal Type *check_get_params(CheckerContext *ctx, Scope *scope, Ast *_para error(name, "'#any_int' can only be applied to variable fields"); p->flags &= ~FieldFlag_any_int; } + if (p->flags&FieldFlag_no_broadcast) { + error(name, "'#no_broadcast' can only be applied to variable fields"); + p->flags &= ~FieldFlag_no_broadcast; + } if (p->flags&FieldFlag_by_ptr) { error(name, "'#by_ptr' can only be applied to variable fields"); p->flags &= ~FieldFlag_by_ptr; @@ -1926,7 +1930,13 @@ gb_internal Type *check_get_params(CheckerContext *ctx, Scope *scope, Ast *_para } } } - if (type != t_invalid && !check_is_assignable_to(ctx, &op, type)) { + + bool allow_array_programming = true; + if (p->flags&FieldFlag_no_broadcast) { + allow_array_programming = false; + } + + if (type != t_invalid && !check_is_assignable_to(ctx, &op, type, allow_array_programming)) { bool ok = true; if (p->flags&FieldFlag_any_int) { if ((!is_type_integer(op.type) && !is_type_enum(op.type)) || (!is_type_integer(type) && !is_type_enum(type))) { @@ -2002,6 +2012,10 @@ gb_internal Type *check_get_params(CheckerContext *ctx, Scope *scope, Ast *_para if (p->flags&FieldFlag_no_alias) { param->flags |= EntityFlag_NoAlias; } + if (p->flags&FieldFlag_no_broadcast) { + param->flags |= EntityFlag_NoBroadcast; + } + if (p->flags&FieldFlag_any_int) { if (!is_type_integer(param->type) && !is_type_enum(param->type)) { gbString str = type_to_string(param->type); diff --git a/src/entity.cpp b/src/entity.cpp index 916c2b2bd..9161ea733 100644 --- a/src/entity.cpp +++ b/src/entity.cpp @@ -61,6 +61,7 @@ enum EntityFlag : u64 { EntityFlag_CVarArg = 1ull<<22, + EntityFlag_NoBroadcast = 1ull<<23, EntityFlag_AnyInt = 1ull<<24, EntityFlag_Disabled = 1ull<<25, diff --git a/src/parser.cpp b/src/parser.cpp index 6a7be8a7c..b6b62461f 100644 --- a/src/parser.cpp +++ b/src/parser.cpp @@ -3898,14 +3898,15 @@ struct ParseFieldPrefixMapping { FieldFlag flag; }; -gb_global ParseFieldPrefixMapping parse_field_prefix_mappings[] = { - {str_lit("using"), Token_using, FieldFlag_using}, - {str_lit("no_alias"), Token_Hash, FieldFlag_no_alias}, - {str_lit("c_vararg"), Token_Hash, FieldFlag_c_vararg}, - {str_lit("const"), Token_Hash, FieldFlag_const}, - {str_lit("any_int"), Token_Hash, FieldFlag_any_int}, - {str_lit("subtype"), Token_Hash, FieldFlag_subtype}, - {str_lit("by_ptr"), Token_Hash, FieldFlag_by_ptr}, +gb_global ParseFieldPrefixMapping const parse_field_prefix_mappings[] = { + {str_lit("using"), Token_using, FieldFlag_using}, + {str_lit("no_alias"), Token_Hash, FieldFlag_no_alias}, + {str_lit("c_vararg"), Token_Hash, FieldFlag_c_vararg}, + {str_lit("const"), Token_Hash, FieldFlag_const}, + {str_lit("any_int"), Token_Hash, FieldFlag_any_int}, + {str_lit("subtype"), Token_Hash, FieldFlag_subtype}, + {str_lit("by_ptr"), Token_Hash, FieldFlag_by_ptr}, + {str_lit("no_broadcast"), Token_Hash, FieldFlag_no_broadcast}, }; diff --git a/src/parser.hpp b/src/parser.hpp index ff3c5eb34..5820275c8 100644 --- a/src/parser.hpp +++ b/src/parser.hpp @@ -326,6 +326,7 @@ enum FieldFlag : u32 { FieldFlag_any_int = 1<<6, FieldFlag_subtype = 1<<7, FieldFlag_by_ptr = 1<<8, + FieldFlag_no_broadcast = 1<<9, // disallow array programming // Internal use by the parser only FieldFlag_Tags = 1<<10, @@ -336,7 +337,7 @@ enum FieldFlag : u32 { FieldFlag_Invalid = 1u<<31, // Parameter List Restrictions - FieldFlag_Signature = FieldFlag_ellipsis|FieldFlag_using|FieldFlag_no_alias|FieldFlag_c_vararg|FieldFlag_const|FieldFlag_any_int|FieldFlag_by_ptr, + FieldFlag_Signature = FieldFlag_ellipsis|FieldFlag_using|FieldFlag_no_alias|FieldFlag_c_vararg|FieldFlag_const|FieldFlag_any_int|FieldFlag_by_ptr|FieldFlag_no_broadcast, FieldFlag_Struct = FieldFlag_using|FieldFlag_subtype|FieldFlag_Tags, }; diff --git a/vendor/raylib/raymath.odin b/vendor/raylib/raymath.odin index beeda7989..9682ffe4f 100644 --- a/vendor/raylib/raymath.odin +++ b/vendor/raylib/raymath.odin @@ -159,7 +159,7 @@ Vector2Transform :: proc "c" (v: Vector2, m: Matrix) -> Vector2 { // Calculate linear interpolation between two vectors @(require_results, deprecated="Prefer = linalg.lerp(v1, v2, amount)") Vector2Lerp :: proc "c" (v1, v2: Vector2, amount: f32) -> Vector2 { - return linalg.lerp(v1, v2, amount) + return linalg.lerp(v1, v2, Vector2(amount)) } // Calculate reflected vector to normal @(require_results, deprecated="Prefer = linalg.reflect(v, normal)") @@ -405,7 +405,7 @@ Vector3Transform :: proc "c" (v: Vector3, m: Matrix) -> Vector3 { // Calculate linear interpolation between two vectors @(require_results, deprecated="Prefer = linalg.lerp(v1, v2, amount)") Vector3Lerp :: proc "c" (v1, v2: Vector3, amount: f32) -> Vector3 { - return linalg.lerp(v1, v2, amount) + return linalg.lerp(v1, v2, Vector3(amount)) } // Calculate reflected vector to normal @(require_results, deprecated="Prefer = linalg.reflect(v, normal)") -- cgit v1.2.3 From 006ea11c569ce0724a844103059b47de797dbd85 Mon Sep 17 00:00:00 2001 From: gingerBill Date: Thu, 21 Mar 2024 11:58:32 +0000 Subject: Update doc-format for `#by_ptr` and `#no_broadcast` parameters --- core/odin/doc-format/doc_format.odin | 16 +++++++++------- src/docs_format.cpp | 18 ++++++++++-------- src/docs_writer.cpp | 12 +++++++----- src/entity.cpp | 1 - 4 files changed, 26 insertions(+), 21 deletions(-) (limited to 'src/entity.cpp') diff --git a/core/odin/doc-format/doc_format.odin b/core/odin/doc-format/doc_format.odin index 5636b1059..664e73d4d 100644 --- a/core/odin/doc-format/doc_format.odin +++ b/core/odin/doc-format/doc_format.odin @@ -102,13 +102,15 @@ Entity_Flag :: enum u32le { Foreign = 0, Export = 1, - Param_Using = 2, // using - Param_Const = 3, // #const - Param_Auto_Cast = 4, // auto_cast - Param_Ellipsis = 5, // Variadic parameter - Param_CVararg = 6, // #c_vararg - Param_No_Alias = 7, // #no_alias - Param_Any_Int = 8, // #any_int + Param_Using = 2, // using + Param_Const = 3, // #const + Param_Auto_Cast = 4, // auto_cast + Param_Ellipsis = 5, // Variadic parameter + Param_CVararg = 6, // #c_vararg + Param_No_Alias = 7, // #no_alias + Param_Any_Int = 8, // #any_int + Param_By_Ptr = 9, // #by_ptr + Param_No_Broadcast = 10, // #no_broadcast Bit_Field_Field = 19, diff --git a/src/docs_format.cpp b/src/docs_format.cpp index 779190e62..ca6ecb5c2 100644 --- a/src/docs_format.cpp +++ b/src/docs_format.cpp @@ -15,7 +15,7 @@ struct OdinDocVersionType { #define OdinDocVersionType_Major 0 #define OdinDocVersionType_Minor 3 -#define OdinDocVersionType_Patch 0 +#define OdinDocVersionType_Patch 1 struct OdinDocHeaderBase { u8 magic[8]; @@ -163,13 +163,15 @@ enum OdinDocEntityFlag : u64 { OdinDocEntityFlag_Foreign = 1ull<<0, OdinDocEntityFlag_Export = 1ull<<1, - OdinDocEntityFlag_Param_Using = 1ull<<2, - OdinDocEntityFlag_Param_Const = 1ull<<3, - OdinDocEntityFlag_Param_AutoCast = 1ull<<4, - OdinDocEntityFlag_Param_Ellipsis = 1ull<<5, - OdinDocEntityFlag_Param_CVararg = 1ull<<6, - OdinDocEntityFlag_Param_NoAlias = 1ull<<7, - OdinDocEntityFlag_Param_AnyInt = 1ull<<8, + OdinDocEntityFlag_Param_Using = 1ull<<2, + OdinDocEntityFlag_Param_Const = 1ull<<3, + OdinDocEntityFlag_Param_AutoCast = 1ull<<4, + OdinDocEntityFlag_Param_Ellipsis = 1ull<<5, + OdinDocEntityFlag_Param_CVararg = 1ull<<6, + OdinDocEntityFlag_Param_NoAlias = 1ull<<7, + OdinDocEntityFlag_Param_AnyInt = 1ull<<8, + OdinDocEntityFlag_Param_ByPtr = 1ull<<9, + OdinDocEntityFlag_Param_NoBroadcast = 1ull<<10, OdinDocEntityFlag_BitField_Field = 1ull<<19, diff --git a/src/docs_writer.cpp b/src/docs_writer.cpp index 824445ed5..9ced78d33 100644 --- a/src/docs_writer.cpp +++ b/src/docs_writer.cpp @@ -925,11 +925,13 @@ gb_internal OdinDocEntityIndex odin_doc_add_entity(OdinDocWriter *w, Entity *e) break; } - if (e->flags & EntityFlag_Using) { flags |= OdinDocEntityFlag_Param_Using; } - if (e->flags & EntityFlag_ConstInput) { flags |= OdinDocEntityFlag_Param_Const; } - if (e->flags & EntityFlag_Ellipsis) { flags |= OdinDocEntityFlag_Param_Ellipsis; } - if (e->flags & EntityFlag_NoAlias) { flags |= OdinDocEntityFlag_Param_NoAlias; } - if (e->flags & EntityFlag_AnyInt) { flags |= OdinDocEntityFlag_Param_AnyInt; } + if (e->flags & EntityFlag_Using) { flags |= OdinDocEntityFlag_Param_Using; } + if (e->flags & EntityFlag_ConstInput) { flags |= OdinDocEntityFlag_Param_Const; } + if (e->flags & EntityFlag_Ellipsis) { flags |= OdinDocEntityFlag_Param_Ellipsis; } + if (e->flags & EntityFlag_NoAlias) { flags |= OdinDocEntityFlag_Param_NoAlias; } + if (e->flags & EntityFlag_AnyInt) { flags |= OdinDocEntityFlag_Param_AnyInt; } + if (e->flags & EntityFlag_ByPtr) { flags |= OdinDocEntityFlag_Param_ByPtr; } + if (e->flags & EntityFlag_NoBroadcast) { flags |= OdinDocEntityFlag_Param_NoBroadcast; } if (e->scope && (e->scope->flags & (ScopeFlag_File|ScopeFlag_Pkg)) && !is_entity_exported(e)) { flags |= OdinDocEntityFlag_Private; diff --git a/src/entity.cpp b/src/entity.cpp index 9161ea733..a160313b4 100644 --- a/src/entity.cpp +++ b/src/entity.cpp @@ -60,7 +60,6 @@ enum EntityFlag : u64 { EntityFlag_ProcBodyChecked = 1ull<<21, EntityFlag_CVarArg = 1ull<<22, - EntityFlag_NoBroadcast = 1ull<<23, EntityFlag_AnyInt = 1ull<<24, -- cgit v1.2.3 From 624b870f2827b330c0e5c8aa887c61cfe3c7b33f Mon Sep 17 00:00:00 2001 From: gingerBill Date: Sat, 23 Mar 2024 14:58:10 +0000 Subject: Add some basic escape analysis errors for `return &x` --- src/check_stmt.cpp | 95 +++++++++++++++++++++++++----------------------------- src/checker.cpp | 1 + src/entity.cpp | 23 +++++++++++++ 3 files changed, 68 insertions(+), 51 deletions(-) (limited to 'src/entity.cpp') diff --git a/src/check_stmt.cpp b/src/check_stmt.cpp index 8a876eb01..04b7359a8 100644 --- a/src/check_stmt.cpp +++ b/src/check_stmt.cpp @@ -1465,25 +1465,6 @@ gb_internal bool check_stmt_internal_builtin_proc_id(Ast *expr, BuiltinProcId *i return id != BuiltinProc_Invalid; } -gb_internal bool check_expr_is_stack_variable(Ast *expr) { - /* - expr = unparen_expr(expr); - Entity *e = entity_of_node(expr); - if (e && e->kind == Entity_Variable) { - if (e->flags & (EntityFlag_Static|EntityFlag_Using|EntityFlag_ImplicitReference|EntityFlag_ForValue)) { - // okay - } else if (e->Variable.thread_local_model.len != 0) { - // okay - } else if (e->scope) { - if ((e->scope->flags & (ScopeFlag_Global|ScopeFlag_File|ScopeFlag_Type)) == 0) { - return true; - } - } - } - */ - return false; -} - gb_internal void check_range_stmt(CheckerContext *ctx, Ast *node, u32 mod_flags) { ast_node(rs, RangeStmt, node); @@ -2297,29 +2278,6 @@ gb_internal void check_return_stmt(CheckerContext *ctx, Ast *node) { if (is_type_untyped(o->type)) { update_untyped_expr_type(ctx, o->expr, e->type, true); } - - - // NOTE(bill): This is very basic escape analysis - // This needs to be improved tremendously, and a lot of it done during the - // middle-end (or LLVM side) to improve checks and error messages - Ast *expr = unparen_expr(o->expr); - if (expr->kind == Ast_UnaryExpr && expr->UnaryExpr.op.kind == Token_And) { - Ast *x = unparen_expr(expr->UnaryExpr.expr); - if (x->kind == Ast_CompoundLit) { - error(expr, "Cannot return the address to a stack value from a procedure"); - } else if (x->kind == Ast_IndexExpr) { - Ast *array = x->IndexExpr.expr; - if (is_type_array_like(type_of_expr(array)) && check_expr_is_stack_variable(array)) { - gbString t = type_to_string(type_of_expr(array)); - error(expr, "Cannot return the address to an element of stack variable from a procedure, of type %s", t); - gb_string_free(t); - } - } else { - if (check_expr_is_stack_variable(x)) { - error(expr, "Cannot return the address to a stack variable from a procedure"); - } - } - } } } @@ -2327,16 +2285,51 @@ gb_internal void check_return_stmt(CheckerContext *ctx, Ast *node) { if (o.expr == nullptr) { continue; } - if (o.expr->kind != Ast_CompoundLit || !is_type_slice(o.type)) { - continue; - } - ast_node(cl, CompoundLit, o.expr); - if (cl->elems.count == 0) { - continue; + Ast *expr = unparen_expr(o.expr); + + auto unsafe_return_error = [](Operand const &o, char const *msg, Type *extra_type=nullptr) { + gbString s = expr_to_string(o.expr); + if (extra_type) { + gbString t = type_to_string(extra_type); + error(o.expr, "It is unsafe to return %s ('%s') of type ('%s') from a procedure, as it uses the current stack frame's memory", msg, s, t); + gb_string_free(t); + } else { + error(o.expr, "It is unsafe to return %s ('%s') from a procedure, as it uses the current stack frame's memory", msg, s); + } + gb_string_free(s); + }; + + + // NOTE(bill): This is very basic escape analysis + // This needs to be improved tremendously, and a lot of it done during the + // middle-end (or LLVM side) to improve checks and error messages + if (expr->kind == Ast_CompoundLit && is_type_slice(o.type)) { + ast_node(cl, CompoundLit, expr); + if (cl->elems.count == 0) { + continue; + } + unsafe_return_error(o, "a compound literal of a slice"); + } else if (expr->kind == Ast_UnaryExpr && expr->UnaryExpr.op.kind == Token_And) { + Ast *x = unparen_expr(expr->UnaryExpr.expr); + Entity *e = entity_of_node(x); + if (is_entity_local_variable(e)) { + unsafe_return_error(o, "the address of a local variable"); + } else if(x->kind == Ast_CompoundLit) { + unsafe_return_error(o, "the address of a compound literal"); + } else if (x->kind == Ast_IndexExpr) { + Entity *f = entity_of_node(x->IndexExpr.expr); + if (is_type_array_like(f->type) || is_type_matrix(f->type)) { + if (is_entity_local_variable(f)) { + unsafe_return_error(o, "the address of an indexed variable", f->type); + } + } + } else if (x->kind == Ast_MatrixIndexExpr) { + Entity *f = entity_of_node(x->MatrixIndexExpr.expr); + if (is_entity_local_variable(f)) { + unsafe_return_error(o, "the address of an indexed variable", f->type); + } + } } - gbString s = type_to_string(o.type); - error(o.expr, "It is unsafe to return a compound literal of a slice ('%s') with elements from a procedure, as the contents of the slice uses the current stack frame's memory", s); - gb_string_free(s); } } diff --git a/src/checker.cpp b/src/checker.cpp index 6a1bce573..bf6a84588 100644 --- a/src/checker.cpp +++ b/src/checker.cpp @@ -4050,6 +4050,7 @@ gb_internal void check_collect_value_decl(CheckerContext *c, Ast *decl) { Entity *e = alloc_entity_variable(c->scope, name->Ident.token, nullptr); e->identifier = name; e->file = c->file; + e->Variable.is_global = true; if (entity_visibility_kind != EntityVisiblity_Public) { e->flags |= EntityFlag_NotExported; diff --git a/src/entity.cpp b/src/entity.cpp index a160313b4..6cea0930f 100644 --- a/src/entity.cpp +++ b/src/entity.cpp @@ -229,6 +229,7 @@ struct Entity { CommentGroup *comment; bool is_foreign; bool is_export; + bool is_global; } Variable; struct { Type * type_parameter_specialization; @@ -480,3 +481,25 @@ gb_internal Entity *strip_entity_wrapping(Ast *expr) { Entity *e = entity_from_expr(expr); return strip_entity_wrapping(e); } + + +gb_internal bool is_entity_local_variable(Entity *e) { + if (e == nullptr) { + return false; + } + if (e->kind != Entity_Variable) { + return false; + } + if (e->Variable.is_global) { + return false; + } + if (e->scope == nullptr) { + return true; + } + if (e->flags & (EntityFlag_ForValue|EntityFlag_SwitchValue)) { + return false; + } + + return ((e->scope->flags &~ ScopeFlag_ContextDefined) == 0) || + (e->scope->flags & ScopeFlag_Proc) != 0; +} \ No newline at end of file -- cgit v1.2.3 From 45d7a670ce689b5be046e023102871566cac9b7b Mon Sep 17 00:00:00 2001 From: gingerBill Date: Thu, 11 Apr 2024 15:34:34 +0100 Subject: Fix `@(static)` error message bug --- src/entity.cpp | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'src/entity.cpp') diff --git a/src/entity.cpp b/src/entity.cpp index 6cea0930f..a12e1d0a6 100644 --- a/src/entity.cpp +++ b/src/entity.cpp @@ -496,7 +496,7 @@ gb_internal bool is_entity_local_variable(Entity *e) { if (e->scope == nullptr) { return true; } - if (e->flags & (EntityFlag_ForValue|EntityFlag_SwitchValue)) { + if (e->flags & (EntityFlag_ForValue|EntityFlag_SwitchValue|EntityFlag_Static)) { return false; } -- cgit v1.2.3 From 25f1d0906d2b5a8276c3832783970a798c12cc6c Mon Sep 17 00:00:00 2001 From: Laytan Laats Date: Wed, 1 May 2024 22:12:37 +0200 Subject: compiler: improve target features support --- .gitignore | 4 +- base/intrinsics/intrinsics.odin | 10 +- core/simd/x86/sse3.odin | 4 +- core/simd/x86/sse41.odin | 4 +- core/sync/futex_wasm.odin | 44 ++- misc/featuregen/README.md | 28 ++ misc/featuregen/featuregen.cpp | 37 +++ misc/featuregen/featuregen.py | 116 ++++++++ src/build_settings.cpp | 617 ++++++++++++++++++++++++++++++++++------ src/check_builtin.cpp | 46 ++- src/check_decl.cpp | 49 +++- src/check_expr.cpp | 68 ++++- src/checker_builtin_procs.hpp | 4 + src/entity.cpp | 4 +- src/llvm_backend.cpp | 104 +++---- src/llvm_backend_proc.cpp | 27 +- src/llvm_backend_utility.cpp | 3 +- src/main.cpp | 79 ++++- src/types.cpp | 22 ++ 19 files changed, 1066 insertions(+), 204 deletions(-) create mode 100644 misc/featuregen/README.md create mode 100644 misc/featuregen/featuregen.cpp create mode 100644 misc/featuregen/featuregen.py (limited to 'src/entity.cpp') diff --git a/.gitignore b/.gitignore index f6c3927a2..2b6b5281a 100644 --- a/.gitignore +++ b/.gitignore @@ -322,4 +322,6 @@ build.sh !core/debug/ # RAD debugger project file -*.raddbg \ No newline at end of file +*.raddbg + +misc/featuregen/featuregen diff --git a/base/intrinsics/intrinsics.odin b/base/intrinsics/intrinsics.odin index dca33bfd9..d887f8dcc 100644 --- a/base/intrinsics/intrinsics.odin +++ b/base/intrinsics/intrinsics.odin @@ -282,6 +282,12 @@ simd_reverse :: proc(a: #simd[N]T) -> #simd[N]T --- simd_rotate_left :: proc(a: #simd[N]T, $offset: int) -> #simd[N]T --- simd_rotate_right :: proc(a: #simd[N]T, $offset: int) -> #simd[N]T --- +// Checks if the current target supports the given target features. +// +// Takes a constant comma-seperated string (eg: "sha512,sse4.1"), or a procedure type which has either +// `@(require_target_feature)` or `@(enable_target_feature)` as its input and returns a boolean indicating +// if all listed features are supported. +has_target_feature :: proc($test: $T) -> bool where type_is_string(T) || type_is_proc(T) --- // WASM targets only wasm_memory_grow :: proc(index, delta: uintptr) -> int --- @@ -293,9 +299,9 @@ wasm_memory_size :: proc(index: uintptr) -> int --- // 0 - indicates that the thread blocked and then was woken up // 1 - the loaded value from `ptr` did not match `expected`, the thread did not block // 2 - the thread blocked, but the timeout -@(enable_target_feature="atomics") +@(require_target_feature="atomics") wasm_memory_atomic_wait32 :: proc(ptr: ^u32, expected: u32, timeout_ns: i64) -> u32 --- -@(enable_target_feature="atomics") +@(require_target_feature="atomics") wasm_memory_atomic_notify32 :: proc(ptr: ^u32, waiters: u32) -> (waiters_woken_up: u32) --- // x86 Targets (i386, amd64) diff --git a/core/simd/x86/sse3.odin b/core/simd/x86/sse3.odin index cf5f3b2fa..ca19c3954 100644 --- a/core/simd/x86/sse3.odin +++ b/core/simd/x86/sse3.odin @@ -36,7 +36,7 @@ _mm_lddqu_si128 :: #force_inline proc "c" (mem_addr: ^__m128i) -> __m128i { _mm_movedup_pd :: #force_inline proc "c" (a: __m128d) -> __m128d { return simd.shuffle(a, a, 0, 0) } -@(require_results, enable_target_feature="sse3") +@(require_results, enable_target_feature="sse2,sse3") _mm_loaddup_pd :: #force_inline proc "c" (mem_addr: [^]f64) -> __m128d { return _mm_load1_pd(mem_addr) } @@ -65,4 +65,4 @@ foreign _ { hsubps :: proc(a, b: __m128) -> __m128 --- @(link_name = "llvm.x86.sse3.ldu.dq") lddqu :: proc(mem_addr: rawptr) -> i8x16 --- -} \ No newline at end of file +} diff --git a/core/simd/x86/sse41.odin b/core/simd/x86/sse41.odin index 8c306ba4c..0b9c5986f 100644 --- a/core/simd/x86/sse41.odin +++ b/core/simd/x86/sse41.odin @@ -268,7 +268,7 @@ _mm_testnzc_si128 :: #force_inline proc "c" (a: __m128i, mask: __m128i) -> i32 { _mm_test_all_zeros :: #force_inline proc "c" (a: __m128i, mask: __m128i) -> i32 { return _mm_testz_si128(a, mask) } -@(require_results, enable_target_feature="sse4.1") +@(require_results, enable_target_feature="sse2,sse4.1") _mm_test_all_ones :: #force_inline proc "c" (a: __m128i) -> i32 { return _mm_testc_si128(a, _mm_cmpeq_epi32(a, a)) } @@ -349,4 +349,4 @@ foreign _ { ptestc :: proc(a, mask: i64x2) -> i32 --- @(link_name = "llvm.x86.sse41.ptestnzc") ptestnzc :: proc(a, mask: i64x2) -> i32 --- -} \ No newline at end of file +} diff --git a/core/sync/futex_wasm.odin b/core/sync/futex_wasm.odin index de1013364..de88e8198 100644 --- a/core/sync/futex_wasm.odin +++ b/core/sync/futex_wasm.odin @@ -5,31 +5,49 @@ package sync import "base:intrinsics" import "core:time" +// NOTE: because `core:sync` is in the dependency chain of a lot of the core packages (mostly through `core:mem`) +// without actually calling into it much, I opted for a runtime panic instead of a compile error here. + _futex_wait :: proc "contextless" (f: ^Futex, expected: u32) -> bool { - s := intrinsics.wasm_memory_atomic_wait32((^u32)(f), expected, -1) - return s != 0 + when !intrinsics.has_target_feature("atomics") { + _panic("usage of `core:sync` requires the `-target-feature:\"atomics\"` or a `-microarch` that supports it") + } else { + s := intrinsics.wasm_memory_atomic_wait32((^u32)(f), expected, -1) + return s != 0 + } } _futex_wait_with_timeout :: proc "contextless" (f: ^Futex, expected: u32, duration: time.Duration) -> bool { - s := intrinsics.wasm_memory_atomic_wait32((^u32)(f), expected, i64(duration)) - return s != 0 - + when !intrinsics.has_target_feature("atomics") { + _panic("usage of `core:sync` requires the `-target-feature:\"atomics\"` or a `-microarch` that supports it") + } else { + s := intrinsics.wasm_memory_atomic_wait32((^u32)(f), expected, i64(duration)) + return s != 0 + } } _futex_signal :: proc "contextless" (f: ^Futex) { - loop: for { - s := intrinsics.wasm_memory_atomic_notify32((^u32)(f), 1) - if s >= 1 { - return + when !intrinsics.has_target_feature("atomics") { + _panic("usage of `core:sync` requires the `-target-feature:\"atomics\"` or a `-microarch` that supports it") + } else { + loop: for { + s := intrinsics.wasm_memory_atomic_notify32((^u32)(f), 1) + if s >= 1 { + return + } } } } _futex_broadcast :: proc "contextless" (f: ^Futex) { - loop: for { - s := intrinsics.wasm_memory_atomic_notify32((^u32)(f), ~u32(0)) - if s >= 0 { - return + when !intrinsics.has_target_feature("atomics") { + _panic("usage of `core:sync` requires the `-target-feature:\"atomics\"` or a `-microarch` that supports it") + } else { + loop: for { + s := intrinsics.wasm_memory_atomic_notify32((^u32)(f), ~u32(0)) + if s >= 0 { + return + } } } } diff --git a/misc/featuregen/README.md b/misc/featuregen/README.md new file mode 100644 index 000000000..22a798cca --- /dev/null +++ b/misc/featuregen/README.md @@ -0,0 +1,28 @@ +# Featuregen + +This directory contains a python and CPP script that generates the needed information +for features regarding microarchitecture and target features of the compiler. + +It is not pretty! But LLVM has no way to query this information with their C API. + +It generates these globals (intended for `src/build_settings.cpp`: + +- `target_microarch_list`: an array of strings indexed by the architecture, each string is a comma-seperated list of microarchitectures available on that architecture +- `target_features_list`: an array of strings indexed by the architecture, each string is a comma-seperated list of target features available on that architecture +- `target_microarch_counts`: an array of ints indexed by the architecture, each int represents the amount of microarchitectures available on that target, intended for easier iteration of the next global +- `microarch_features_list`: an array of a tuple like struct where the first string is a microarchitecture and the second is a comma-seperated list of all features that are enabled by default for it + +In order to get the default features for a microarchitecture there is a small CPP program that takes +a target triple and microarchitecture and spits out the default features, this is then parsed by the python script. + +This should be ran each time we update LLVM to stay in sync. + +If there are minor differences (like the Odin user using LLVM 14 and this table being generated on LLVM 17) it +does not impact much at all, the only thing it will do is make LLVM print a message that the feature is ignored (if it was added between 14 and 17 in this case). + +## Usage + +1. Make sure the table of architectures at the top of the python script is up-to-date (the triple can be any valid triple for the architecture) +1. `./build.sh` +1. `python3 featuregen.py` +1. Copy the output into `src/build_settings.cpp` diff --git a/misc/featuregen/featuregen.cpp b/misc/featuregen/featuregen.cpp new file mode 100644 index 000000000..a1d00ab31 --- /dev/null +++ b/misc/featuregen/featuregen.cpp @@ -0,0 +1,37 @@ +#include +#include +#include +#include +#include +#include + +// Dumps the default set of supported features for the given microarch. +int main(int argc, char **argv) { + if (argc < 3) { + llvm::errs() << "Error: first arg should be triple, second should be microarch\n"; + return 1; + } + + llvm::InitializeAllTargets(); + llvm::InitializeAllTargetMCs(); + + std::string error; + const llvm::Target* target = llvm::TargetRegistry::lookupTarget(argv[1], error); + + if (!target) { + llvm::errs() << "Error: " << error << "\n"; + return 1; + } + + auto STI = target->createMCSubtargetInfo(argv[1], argv[2], ""); + + std::string plus = "+"; + llvm::ArrayRef features = STI->getAllProcessorFeatures(); + for (const auto& feature : features) { + if (STI->checkFeatures(plus + feature.Key)) { + llvm::outs() << feature.Key << "\n"; + } + } + + return 0; +} diff --git a/misc/featuregen/featuregen.py b/misc/featuregen/featuregen.py new file mode 100644 index 000000000..da4cc68f5 --- /dev/null +++ b/misc/featuregen/featuregen.py @@ -0,0 +1,116 @@ +import subprocess +import tempfile +import os +import sys + +archs = [ + ("amd64", "linux_amd64", "x86_64-pc-linux-gnu", [], []), + ("i386", "linux_i386", "i386-pc-linux-gnu", [], []), + ("arm32", "linux_arm32", "arm-linux-gnu", [], []), + ("arm64", "linux_arm64", "aarch64-linux-elf", [], []), + ("wasm32", "js_wasm32", "wasm32-js-js", [], []), + ("wasm64p32", "js_wasm64p32","wasm32-js-js", [], []), +]; + +SEEKING_CPUS = 0 +PARSING_CPUS = 1 +PARSING_FEATURES = 2 + +with tempfile.NamedTemporaryFile(suffix=".odin", delete=True) as temp_file: + temp_file.write(b"package main\n") + + for arch, target, triple, cpus, features in archs: + cmd = ["odin", "build", temp_file.name, "-file", "-build-mode:llvm", "-out:temp", "-target-features:\"help\"", f"-target:\"{target}\""] + process = subprocess.Popen(cmd, stderr=subprocess.PIPE, text=True) + + state = SEEKING_CPUS + for line in process.stderr: + + if state == SEEKING_CPUS: + if line == "Available CPUs for this target:\n": + state = PARSING_CPUS + + elif state == PARSING_CPUS: + if line == "Available features for this target:\n": + state = PARSING_FEATURES + continue + + parts = line.split(" -", maxsplit=1) + if len(parts) < 2: + continue + + cpu = parts[0].strip() + cpus.append(cpu) + + elif state == PARSING_FEATURES: + if line == "\n" and len(features) > 0: + break + + parts = line.split(" -", maxsplit=1) + if len(parts) < 2: + continue + + feature = parts[0].strip() + features.append(feature) + + process.wait() + if process.returncode != 0: + print(f"odin build returned with non-zero exit code {process.returncode}") + sys.exit(1) + + os.remove("temp.ll") + +def print_default_features(triple, microarch): + cmd = ["./featuregen", triple, microarch] + process = subprocess.Popen(cmd, stdout=subprocess.PIPE, text=True) + first = True + for line in process.stdout: + print("" if first else ",", line.strip(), sep="", end="") + first = False + process.wait() + if process.returncode != 0: + print(f"featuregen returned with non-zero exit code {process.returncode}") + sys.exit(1) + +print("// Generated with the featuregen script in `misc/featuregen`") +print("gb_global String target_microarch_list[TargetArch_COUNT] = {") +print("\t// TargetArch_Invalid:") +print('\tstr_lit(""),') +for arch, target, triple, cpus, features in archs: + print(f"\t// TargetArch_{arch}:") + print(f'\tstr_lit("{','.join(cpus)}"),') +print("};") + +print("") + +print("// Generated with the featuregen script in `misc/featuregen`") +print("gb_global String target_features_list[TargetArch_COUNT] = {") +print("\t// TargetArch_Invalid:") +print('\tstr_lit(""),') +for arch, target, triple, cpus, features in archs: + print(f"\t// TargetArch_{arch}:") + print(f'\tstr_lit("{','.join(features)}"),') +print("};") + +print("") + +print("// Generated with the featuregen script in `misc/featuregen`") +print("gb_global int target_microarch_counts[TargetArch_COUNT] = {") +print("\t// TargetArch_Invalid:") +print("\t0,") +for arch, target, triple, cpus, feature in archs: + print(f"\t// TargetArch_{arch}:") + print(f"\t{len(cpus)},") +print("};") + +print("") + +print("// Generated with the featuregen script in `misc/featuregen`") +print("gb_global MicroarchFeatureList microarch_features_list[] = {") +for arch, target, triple, cpus, features in archs: + print(f"\t// TargetArch_{arch}:") + for cpu in cpus: + print(f'\t{{ str_lit("{cpu}"), str_lit("', end="") + print_default_features(triple, cpu) + print('") },') +print("};") diff --git a/src/build_settings.cpp b/src/build_settings.cpp index 8509394ff..8ad03b1b9 100644 --- a/src/build_settings.cpp +++ b/src/build_settings.cpp @@ -71,6 +71,11 @@ enum Windows_Subsystem : u8 { Windows_Subsystem_COUNT, }; +struct MicroarchFeatureList { + String microarch; + String features; +}; + gb_global String target_os_names[TargetOs_COUNT] = { str_lit(""), str_lit("windows"), @@ -97,21 +102,467 @@ gb_global String target_arch_names[TargetArch_COUNT] = { str_lit("wasm64p32"), }; +// Generated with the featuregen script in `misc/featuregen` gb_global String target_microarch_list[TargetArch_COUNT] = { - // TargetArch_Invalid, - str_lit("Invalid!"), - // TargetArch_amd64, - str_lit("alderlake,amdfam10,athlon-fx,athlon64,athlon64-sse3,atom_sse4_2,atom_sse4_2_movbe,barcelona,bdver1,bdver2,bdver3,bdver4,broadwell,btver1,btver2,cannonlake,cascadelake,cooperlake,core-avx-i,core-avx2,core2,core_2_duo_sse4_1,core_2_duo_ssse3,core_2nd_gen_avx,core_3rd_gen_avx,core_4th_gen_avx,core_4th_gen_avx_tsx,core_5th_gen_avx,core_5th_gen_avx_tsx,core_aes_pclmulqdq,core_i7_sse4_2,corei7,corei7-avx,generic,goldmont,goldmont-plus,goldmont_plus,grandridge,graniterapids,graniterapids-d,graniterapids_d,haswell,icelake-client,icelake-server,icelake_client,icelake_server,ivybridge,k8,k8-sse3,knl,knm,meteorlake,mic_avx512,native,nehalem,nocona,opteron,opteron-sse3,penryn,raptorlake,rocketlake,sandybridge,sapphirerapids,sierraforest,silvermont,skx,skylake,skylake-avx512,skylake_avx512,slm,tigerlake,tremont,westmere,x86-64,x86-64-v2,x86-64-v3,x86-64-v4,znver1,znver2,znver3,znver4"), - // TargetArch_i386, - str_lit("athlon,athlon-4,athlon-mp,athlon-tbird,athlon-xp,atom,bonnell,c3,c3-2,generic,geode,i386,i486,i586,i686,k6,k6-2,k6-3,lakemont,native,pentium,pentium-m,pentium-mmx,pentium2,pentium3,pentium3m,pentium4,pentium4m,pentium_4,pentium_4_sse3,pentium_ii,pentium_iii,pentium_iii_no_xmm_regs,pentium_m,pentium_mmx,pentium_pro,pentiumpro,prescott,winchip-c6,winchip2,yonah"), - // TargetArch_arm32, - str_lit("arm1020e,arm1020t,arm1022e,arm10e,arm10tdmi,arm1136j-s,arm1136jf-s,arm1156t2-s,arm1156t2f-s,arm1176jz-s,arm1176jzf-s,arm710t,arm720t,arm7tdmi,arm7tdmi-s,arm8,arm810,arm9,arm920,arm920t,arm922t,arm926ej-s,arm940t,arm946e-s,arm966e-s,arm968e-s,arm9e,arm9tdmi,cortex-a12,cortex-a15,cortex-a17,cortex-a32,cortex-a35,cortex-a5,cortex-a53,cortex-a55,cortex-a57,cortex-a7,cortex-a710,cortex-a72,cortex-a73,cortex-a75,cortex-a76,cortex-a76ae,cortex-a77,cortex-a78,cortex-a78c,cortex-a8,cortex-a9,cortex-m0,cortex-m0plus,cortex-m1,cortex-m23,cortex-m3,cortex-m33,cortex-m35p,cortex-m4,cortex-m55,cortex-m7,cortex-m85,cortex-r4,cortex-r4f,cortex-r5,cortex-r52,cortex-r7,cortex-r8,cortex-x1,cortex-x1c,cyclone,ep9312,exynos-m3,exynos-m4,exynos-m5,generic,iwmmxt,krait,kryo,mpcore,mpcorenovfp,native,neoverse-n1,neoverse-n2,neoverse-v1,sc000,sc300,strongarm,strongarm110,strongarm1100,strongarm1110,swift,xscale"), - // TargetArch_arm64, - str_lit("a64fx,ampere1,ampere1a,apple-a10,apple-a11,apple-a12,apple-a13,apple-a14,apple-a15,apple-a16,apple-a7,apple-a8,apple-a9,apple-latest,apple-m1,apple-m2,apple-s4,apple-s5,carmel,cortex-a34,cortex-a35,cortex-a510,cortex-a53,cortex-a55,cortex-a57,cortex-a65,cortex-a65ae,cortex-a710,cortex-a715,cortex-a72,cortex-a73,cortex-a75,cortex-a76,cortex-a76ae,cortex-a77,cortex-a78,cortex-a78c,cortex-r82,cortex-x1,cortex-x1c,cortex-x2,cortex-x3,cyclone,exynos-m3,exynos-m4,exynos-m5,falkor,generic,kryo,native,neoverse-512tvb,neoverse-e1,neoverse-n1,neoverse-n2,neoverse-v1,neoverse-v2,saphira,thunderx,thunderx2t99,thunderx3t110,thunderxt81,thunderxt83,thunderxt88,tsv110"), - // TargetArch_wasm32, - str_lit("generic"), - // TargetArch_wasm64p32, - str_lit("generic"), + // TargetArch_Invalid: + str_lit(""), + // TargetArch_amd64: + str_lit("alderlake,amdfam10,athlon,athlon-4,athlon-fx,athlon-mp,athlon-tbird,athlon-xp,athlon64,athlon64-sse3,atom,atom_sse4_2,atom_sse4_2_movbe,barcelona,bdver1,bdver2,bdver3,bdver4,bonnell,broadwell,btver1,btver2,c3,c3-2,cannonlake,cascadelake,cooperlake,core-avx-i,core-avx2,core2,core_2_duo_sse4_1,core_2_duo_ssse3,core_2nd_gen_avx,core_3rd_gen_avx,core_4th_gen_avx,core_4th_gen_avx_tsx,core_5th_gen_avx,core_5th_gen_avx_tsx,core_aes_pclmulqdq,core_i7_sse4_2,corei7,corei7-avx,emeraldrapids,generic,geode,goldmont,goldmont-plus,goldmont_plus,grandridge,graniterapids,graniterapids-d,graniterapids_d,haswell,i386,i486,i586,i686,icelake-client,icelake-server,icelake_client,icelake_server,ivybridge,k6,k6-2,k6-3,k8,k8-sse3,knl,knm,lakemont,meteorlake,mic_avx512,nehalem,nocona,opteron,opteron-sse3,penryn,pentium,pentium-m,pentium-mmx,pentium2,pentium3,pentium3m,pentium4,pentium4m,pentium_4,pentium_4_sse3,pentium_ii,pentium_iii,pentium_iii_no_xmm_regs,pentium_m,pentium_mmx,pentium_pro,pentiumpro,prescott,raptorlake,rocketlake,sandybridge,sapphirerapids,sierraforest,silvermont,skx,skylake,skylake-avx512,skylake_avx512,slm,tigerlake,tremont,westmere,winchip-c6,winchip2,x86-64,x86-64-v2,x86-64-v3,x86-64-v4,yonah,znver1,znver2,znver3,znver4"), + // TargetArch_i386: + str_lit("alderlake,amdfam10,athlon,athlon-4,athlon-fx,athlon-mp,athlon-tbird,athlon-xp,athlon64,athlon64-sse3,atom,atom_sse4_2,atom_sse4_2_movbe,barcelona,bdver1,bdver2,bdver3,bdver4,bonnell,broadwell,btver1,btver2,c3,c3-2,cannonlake,cascadelake,cooperlake,core-avx-i,core-avx2,core2,core_2_duo_sse4_1,core_2_duo_ssse3,core_2nd_gen_avx,core_3rd_gen_avx,core_4th_gen_avx,core_4th_gen_avx_tsx,core_5th_gen_avx,core_5th_gen_avx_tsx,core_aes_pclmulqdq,core_i7_sse4_2,corei7,corei7-avx,emeraldrapids,generic,geode,goldmont,goldmont-plus,goldmont_plus,grandridge,graniterapids,graniterapids-d,graniterapids_d,haswell,i386,i486,i586,i686,icelake-client,icelake-server,icelake_client,icelake_server,ivybridge,k6,k6-2,k6-3,k8,k8-sse3,knl,knm,lakemont,meteorlake,mic_avx512,nehalem,nocona,opteron,opteron-sse3,penryn,pentium,pentium-m,pentium-mmx,pentium2,pentium3,pentium3m,pentium4,pentium4m,pentium_4,pentium_4_sse3,pentium_ii,pentium_iii,pentium_iii_no_xmm_regs,pentium_m,pentium_mmx,pentium_pro,pentiumpro,prescott,raptorlake,rocketlake,sandybridge,sapphirerapids,sierraforest,silvermont,skx,skylake,skylake-avx512,skylake_avx512,slm,tigerlake,tremont,westmere,winchip-c6,winchip2,x86-64,x86-64-v2,x86-64-v3,x86-64-v4,yonah,znver1,znver2,znver3,znver4"), + // TargetArch_arm32: + str_lit("arm1020e,arm1020t,arm1022e,arm10e,arm10tdmi,arm1136j-s,arm1136jf-s,arm1156t2-s,arm1156t2f-s,arm1176jz-s,arm1176jzf-s,arm710t,arm720t,arm7tdmi,arm7tdmi-s,arm8,arm810,arm9,arm920,arm920t,arm922t,arm926ej-s,arm940t,arm946e-s,arm966e-s,arm968e-s,arm9e,arm9tdmi,cortex-a12,cortex-a15,cortex-a17,cortex-a32,cortex-a35,cortex-a5,cortex-a53,cortex-a55,cortex-a57,cortex-a7,cortex-a710,cortex-a72,cortex-a73,cortex-a75,cortex-a76,cortex-a76ae,cortex-a77,cortex-a78,cortex-a78c,cortex-a8,cortex-a9,cortex-m0,cortex-m0plus,cortex-m1,cortex-m23,cortex-m3,cortex-m33,cortex-m35p,cortex-m4,cortex-m55,cortex-m7,cortex-m85,cortex-r4,cortex-r4f,cortex-r5,cortex-r52,cortex-r7,cortex-r8,cortex-x1,cortex-x1c,cyclone,ep9312,exynos-m3,exynos-m4,exynos-m5,generic,iwmmxt,krait,kryo,mpcore,mpcorenovfp,neoverse-n1,neoverse-n2,neoverse-v1,sc000,sc300,strongarm,strongarm110,strongarm1100,strongarm1110,swift,xscale"), + // TargetArch_arm64: + str_lit("a64fx,ampere1,ampere1a,apple-a10,apple-a11,apple-a12,apple-a13,apple-a14,apple-a15,apple-a16,apple-a7,apple-a8,apple-a9,apple-latest,apple-m1,apple-m2,apple-s4,apple-s5,carmel,cortex-a34,cortex-a35,cortex-a510,cortex-a53,cortex-a55,cortex-a57,cortex-a65,cortex-a65ae,cortex-a710,cortex-a715,cortex-a72,cortex-a73,cortex-a75,cortex-a76,cortex-a76ae,cortex-a77,cortex-a78,cortex-a78c,cortex-r82,cortex-x1,cortex-x1c,cortex-x2,cortex-x3,cyclone,exynos-m3,exynos-m4,exynos-m5,falkor,generic,kryo,neoverse-512tvb,neoverse-e1,neoverse-n1,neoverse-n2,neoverse-v1,neoverse-v2,saphira,thunderx,thunderx2t99,thunderx3t110,thunderxt81,thunderxt83,thunderxt88,tsv110"), + // TargetArch_wasm32: + str_lit("bleeding-edge,generic,mvp"), + // TargetArch_wasm64p32: + str_lit("bleeding-edge,generic,mvp"), +}; + +// Generated with the featuregen script in `misc/featuregen` +gb_global String target_features_list[TargetArch_COUNT] = { + // TargetArch_Invalid: + str_lit(""), + // TargetArch_amd64: + str_lit("16bit-mode,32bit-mode,3dnow,3dnowa,64bit,64bit-mode,adx,aes,allow-light-256-bit,amx-bf16,amx-complex,amx-fp16,amx-int8,amx-tile,avx,avx2,avx512bf16,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512er,avx512f,avx512fp16,avx512ifma,avx512pf,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vp2intersect,avx512vpopcntdq,avxifma,avxneconvert,avxvnni,avxvnniint16,avxvnniint8,bmi,bmi2,branchfusion,cldemote,clflushopt,clwb,clzero,cmov,cmpccxadd,crc32,cx16,cx8,enqcmd,ermsb,f16c,false-deps-getmant,false-deps-lzcnt-tzcnt,false-deps-mulc,false-deps-mullq,false-deps-perm,false-deps-popcnt,false-deps-range,fast-11bytenop,fast-15bytenop,fast-7bytenop,fast-bextr,fast-gather,fast-hops,fast-lzcnt,fast-movbe,fast-scalar-fsqrt,fast-scalar-shift-masks,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fast-vector-shift-masks,faster-shift-than-shuffle,fma,fma4,fsgsbase,fsrm,fxsr,gfni,harden-sls-ijmp,harden-sls-ret,hreset,idivl-to-divb,idivq-to-divl,invpcid,kl,lea-sp,lea-uses-ag,lvi-cfi,lvi-load-hardening,lwp,lzcnt,macrofusion,mmx,movbe,movdir64b,movdiri,mwaitx,no-bypass-delay,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pad-short-functions,pclmul,pconfig,pku,popcnt,prefer-128-bit,prefer-256-bit,prefer-mask-registers,prefer-movmsk-over-vtest,prefer-no-gather,prefer-no-scatter,prefetchi,prefetchwt1,prfchw,ptwrite,raoint,rdpid,rdpru,rdrnd,rdseed,retpoline,retpoline-external-thunk,retpoline-indirect-branches,retpoline-indirect-calls,rtm,sahf,sbb-dep-breaking,serialize,seses,sgx,sha,sha512,shstk,slow-3ops-lea,slow-incdec,slow-lea,slow-pmaddwd,slow-pmulld,slow-shld,slow-two-mem-ops,slow-unaligned-mem-16,slow-unaligned-mem-32,sm3,sm4,soft-float,sse,sse-unaligned-mem,sse2,sse3,sse4.1,sse4.2,sse4a,ssse3,tagged-globals,tbm,tsxldtrk,tuning-fast-imm-vector-shift,uintr,use-glm-div-sqrt-costs,use-slm-arith-costs,vaes,vpclmulqdq,vzeroupper,waitpkg,wbnoinvd,widekl,x87,xop,xsave,xsavec,xsaveopt,xsaves"), + // TargetArch_i386: + str_lit("16bit-mode,32bit-mode,3dnow,3dnowa,64bit,64bit-mode,adx,aes,allow-light-256-bit,amx-bf16,amx-complex,amx-fp16,amx-int8,amx-tile,avx,avx2,avx512bf16,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512er,avx512f,avx512fp16,avx512ifma,avx512pf,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vp2intersect,avx512vpopcntdq,avxifma,avxneconvert,avxvnni,avxvnniint16,avxvnniint8,bmi,bmi2,branchfusion,cldemote,clflushopt,clwb,clzero,cmov,cmpccxadd,crc32,cx16,cx8,enqcmd,ermsb,f16c,false-deps-getmant,false-deps-lzcnt-tzcnt,false-deps-mulc,false-deps-mullq,false-deps-perm,false-deps-popcnt,false-deps-range,fast-11bytenop,fast-15bytenop,fast-7bytenop,fast-bextr,fast-gather,fast-hops,fast-lzcnt,fast-movbe,fast-scalar-fsqrt,fast-scalar-shift-masks,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fast-vector-shift-masks,faster-shift-than-shuffle,fma,fma4,fsgsbase,fsrm,fxsr,gfni,harden-sls-ijmp,harden-sls-ret,hreset,idivl-to-divb,idivq-to-divl,invpcid,kl,lea-sp,lea-uses-ag,lvi-cfi,lvi-load-hardening,lwp,lzcnt,macrofusion,mmx,movbe,movdir64b,movdiri,mwaitx,no-bypass-delay,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pad-short-functions,pclmul,pconfig,pku,popcnt,prefer-128-bit,prefer-256-bit,prefer-mask-registers,prefer-movmsk-over-vtest,prefer-no-gather,prefer-no-scatter,prefetchi,prefetchwt1,prfchw,ptwrite,raoint,rdpid,rdpru,rdrnd,rdseed,retpoline,retpoline-external-thunk,retpoline-indirect-branches,retpoline-indirect-calls,rtm,sahf,sbb-dep-breaking,serialize,seses,sgx,sha,sha512,shstk,slow-3ops-lea,slow-incdec,slow-lea,slow-pmaddwd,slow-pmulld,slow-shld,slow-two-mem-ops,slow-unaligned-mem-16,slow-unaligned-mem-32,sm3,sm4,soft-float,sse,sse-unaligned-mem,sse2,sse3,sse4.1,sse4.2,sse4a,ssse3,tagged-globals,tbm,tsxldtrk,tuning-fast-imm-vector-shift,uintr,use-glm-div-sqrt-costs,use-slm-arith-costs,vaes,vpclmulqdq,vzeroupper,waitpkg,wbnoinvd,widekl,x87,xop,xsave,xsavec,xsaveopt,xsaves"), + // TargetArch_arm32: + str_lit("32bit,8msecext,a12,a15,a17,a32,a35,a5,a53,a55,a57,a7,a72,a73,a75,a76,a77,a78c,a8,a9,aapcs-frame-chain,aapcs-frame-chain-leaf,aclass,acquire-release,aes,armv4,armv4t,armv5t,armv5te,armv5tej,armv6,armv6-m,armv6j,armv6k,armv6kz,armv6s-m,armv6t2,armv7-a,armv7-m,armv7-r,armv7e-m,armv7k,armv7s,armv7ve,armv8-a,armv8-m.base,armv8-m.main,armv8-r,armv8.1-a,armv8.1-m.main,armv8.2-a,armv8.3-a,armv8.4-a,armv8.5-a,armv8.6-a,armv8.7-a,armv8.8-a,armv8.9-a,armv9-a,armv9.1-a,armv9.2-a,armv9.3-a,armv9.4-a,atomics-32,avoid-movs-shop,avoid-partial-cpsr,bf16,big-endian-instructions,cde,cdecp0,cdecp1,cdecp2,cdecp3,cdecp4,cdecp5,cdecp6,cdecp7,cheap-predicable-cpsr,clrbhb,cortex-a710,cortex-a78,cortex-x1,cortex-x1c,crc,crypto,d32,db,dfb,disable-postra-scheduler,dont-widen-vmovs,dotprod,dsp,execute-only,expand-fp-mlx,exynos,fix-cmse-cve-2021-35465,fix-cortex-a57-aes-1742098,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp16fml,fp64,fpao,fpregs,fpregs16,fpregs64,fullfp16,fuse-aes,fuse-literals,harden-sls-blr,harden-sls-nocomdat,harden-sls-retbr,hwdiv,hwdiv-arm,i8mm,iwmmxt,iwmmxt2,krait,kryo,lob,long-calls,loop-align,m3,m7,mclass,mp,muxed-units,mve,mve.fp,mve1beat,mve2beat,mve4beat,nacl-trap,neon,neon-fpmovs,neonfp,neoverse-v1,no-branch-predictor,no-bti-at-return-twice,no-movt,no-neg-immediates,noarm,nonpipelined-vfp,pacbti,perfmon,prefer-ishst,prefer-vmovsr,prof-unpr,r4,r5,r52,r7,ras,rclass,read-tp-tpidrprw,read-tp-tpidruro,read-tp-tpidrurw,reserve-r9,ret-addr-stack,sb,sha2,slow-fp-brcc,slow-load-D-subreg,slow-odd-reg,slow-vdup32,slow-vgetlni32,slowfpvfmx,slowfpvmlx,soft-float,splat-vfp-neon,strict-align,swift,thumb-mode,thumb2,trustzone,use-mipipeliner,use-misched,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8.1a,v8.1m.main,v8.2a,v8.3a,v8.4a,v8.5a,v8.6a,v8.7a,v8.8a,v8.9a,v8m,v8m.main,v9.1a,v9.2a,v9.3a,v9.4a,v9a,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization,vldn-align,vmlx-forwarding,vmlx-hazards,wide-stride-vfp,xscale,zcz"), + // TargetArch_arm64: + str_lit("CONTEXTIDREL2,a35,a510,a53,a55,a57,a64fx,a65,a710,a715,a72,a73,a75,a76,a77,a78,a78c,aes,aggressive-fma,all,alternate-sextload-cvt-f32-pattern,altnzcv,am,ampere1,ampere1a,amvs,apple-a10,apple-a11,apple-a12,apple-a13,apple-a14,apple-a15,apple-a16,apple-a7,apple-a7-sysreg,arith-bcc-fusion,arith-cbz-fusion,ascend-store-address,b16b16,balance-fp-ops,bf16,brbe,bti,call-saved-x10,call-saved-x11,call-saved-x12,call-saved-x13,call-saved-x14,call-saved-x15,call-saved-x18,call-saved-x8,call-saved-x9,carmel,ccdp,ccidx,ccpp,chk,clrbhb,cmp-bcc-fusion,complxnum,cortex-r82,cortex-x1,cortex-x2,cortex-x3,crc,crypto,cssc,custom-cheap-as-move,d128,disable-latency-sched-heuristic,dit,dotprod,ecv,el2vmsa,el3,enable-select-opt,ete,exynos-cheap-as-move,exynosm3,exynosm4,f32mm,f64mm,falkor,fgt,fix-cortex-a53-835769,flagm,fmv,force-32bit-jump-tables,fp-armv8,fp16fml,fptoint,fullfp16,fuse-address,fuse-addsub-2reg-const1,fuse-adrp-add,fuse-aes,fuse-arith-logic,fuse-crypto-eor,fuse-csel,fuse-literals,gcs,harden-sls-blr,harden-sls-nocomdat,harden-sls-retbr,hbc,hcx,i8mm,ite,jsconv,kryo,lor,ls64,lse,lse128,lse2,lsl-fast,mec,mops,mpam,mte,neon,neoverse512tvb,neoversee1,neoversen1,neoversen2,neoversev1,neoversev2,nmi,no-bti-at-return-twice,no-neg-immediates,no-sve-fp-ld1r,no-zcz-fp,nv,outline-atomics,pan,pan-rwv,pauth,perfmon,predictable-select-expensive,predres,prfm-slc-target,rand,ras,rasv2,rcpc,rcpc-immo,rcpc3,rdm,reserve-x1,reserve-x10,reserve-x11,reserve-x12,reserve-x13,reserve-x14,reserve-x15,reserve-x18,reserve-x2,reserve-x20,reserve-x21,reserve-x22,reserve-x23,reserve-x24,reserve-x25,reserve-x26,reserve-x27,reserve-x28,reserve-x3,reserve-x30,reserve-x4,reserve-x5,reserve-x6,reserve-x7,reserve-x9,rme,saphira,sb,sel2,sha2,sha3,slow-misaligned-128store,slow-paired-128,slow-strqro-store,sm4,sme,sme-f16f16,sme-f64f64,sme-i16i64,sme2,sme2p1,spe,spe-eef,specres2,specrestrict,ssbs,strict-align,sve,sve2,sve2-aes,sve2-bitperm,sve2-sha3,sve2-sm4,sve2p1,tagged-globals,the,thunderx,thunderx2t99,thunderx3t110,thunderxt81,thunderxt83,thunderxt88,tlb-rmi,tme,tpidr-el1,tpidr-el2,tpidr-el3,tpidrro-el0,tracev8.4,trbe,tsv110,uaops,use-experimental-zeroing-pseudos,use-postra-scheduler,use-reciprocal-square-root,use-scalar-inc-vl,v8.1a,v8.2a,v8.3a,v8.4a,v8.5a,v8.6a,v8.7a,v8.8a,v8.9a,v8a,v8r,v9.1a,v9.2a,v9.3a,v9.4a,v9a,vh,wfxt,xs,zcm,zcz,zcz-fp-workaround,zcz-gp"), + // TargetArch_wasm32: + str_lit("atomics,bulk-memory,exception-handling,extended-const,multivalue,mutable-globals,nontrapping-fptoint,reference-types,relaxed-simd,sign-ext,simd128,tail-call"), + // TargetArch_wasm64p32: + str_lit("atomics,bulk-memory,exception-handling,extended-const,multivalue,mutable-globals,nontrapping-fptoint,reference-types,relaxed-simd,sign-ext,simd128,tail-call"), +}; + +// Generated with the featuregen script in `misc/featuregen` +gb_global int target_microarch_counts[TargetArch_COUNT] = { + // TargetArch_Invalid: + 0, + // TargetArch_amd64: + 120, + // TargetArch_i386: + 120, + // TargetArch_arm32: + 90, + // TargetArch_arm64: + 63, + // TargetArch_wasm32: + 3, + // TargetArch_wasm64p32: + 3, +}; + +// Generated with the featuregen script in `misc/featuregen` +gb_global MicroarchFeatureList microarch_features_list[] = { + // TargetArch_amd64: + { str_lit("alderlake"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,avx,avx2,avxvnni,bmi,bmi2,cldemote,clflushopt,clwb,cmov,crc32,cx16,cx8,f16c,false-deps-perm,false-deps-popcnt,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fxsr,gfni,hreset,idivq-to-divl,invpcid,kl,lzcnt,macrofusion,mmx,movbe,movdir64b,movdiri,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pconfig,pku,popcnt,prefer-movmsk-over-vtest,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,serialize,sha,shstk,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vaes,vpclmulqdq,vzeroupper,waitpkg,widekl,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("amdfam10"), str_lit("3dnow,3dnowa,64bit,64bit-mode,cmov,cx16,cx8,fast-scalar-shift-masks,fxsr,lzcnt,mmx,nopl,popcnt,prfchw,sahf,sbb-dep-breaking,slow-shld,sse,sse2,sse3,sse4a,vzeroupper,x87") }, + { str_lit("athlon"), str_lit("3dnow,3dnowa,64bit-mode,cmov,cx8,mmx,nopl,slow-shld,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("athlon-4"), str_lit("3dnow,3dnowa,64bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-shld,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("athlon-fx"), str_lit("3dnow,3dnowa,64bit,64bit-mode,cmov,cx8,fast-scalar-shift-masks,fxsr,mmx,nopl,sbb-dep-breaking,slow-shld,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("athlon-mp"), str_lit("3dnow,3dnowa,64bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-shld,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("athlon-tbird"), str_lit("3dnow,3dnowa,64bit-mode,cmov,cx8,mmx,nopl,slow-shld,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("athlon-xp"), str_lit("3dnow,3dnowa,64bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-shld,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("athlon64"), str_lit("3dnow,3dnowa,64bit,64bit-mode,cmov,cx8,fast-scalar-shift-masks,fxsr,mmx,nopl,sbb-dep-breaking,slow-shld,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("athlon64-sse3"), str_lit("3dnow,3dnowa,64bit,64bit-mode,cmov,cx16,cx8,fast-scalar-shift-masks,fxsr,mmx,nopl,sbb-dep-breaking,slow-shld,slow-unaligned-mem-16,sse,sse2,sse3,vzeroupper,x87") }, + { str_lit("atom"), str_lit("64bit,64bit-mode,cmov,cx16,cx8,fxsr,idivl-to-divb,idivq-to-divl,lea-sp,lea-uses-ag,mmx,movbe,no-bypass-delay,nopl,pad-short-functions,sahf,slow-two-mem-ops,slow-unaligned-mem-16,sse,sse2,sse3,ssse3,vzeroupper,x87") }, + { str_lit("atom_sse4_2"), str_lit("64bit,64bit-mode,cmov,crc32,cx16,cx8,false-deps-popcnt,fast-7bytenop,fast-movbe,fxsr,idivq-to-divl,mmx,movbe,no-bypass-delay,nopl,pclmul,popcnt,prfchw,rdrnd,sahf,slow-incdec,slow-lea,slow-pmulld,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,use-slm-arith-costs,vzeroupper,x87") }, + { str_lit("atom_sse4_2_movbe"), str_lit("64bit,64bit-mode,aes,clflushopt,cmov,crc32,cx16,cx8,false-deps-popcnt,fast-7bytenop,fast-movbe,fsgsbase,fxsr,idivq-to-divl,mmx,movbe,no-bypass-delay,nopl,pclmul,popcnt,prfchw,rdrnd,rdseed,sahf,sha,slow-incdec,slow-lea,slow-pmulld,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,use-slm-arith-costs,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("barcelona"), str_lit("3dnow,3dnowa,64bit,64bit-mode,cmov,cx16,cx8,fast-scalar-shift-masks,fxsr,lzcnt,mmx,nopl,popcnt,prfchw,sahf,sbb-dep-breaking,slow-shld,sse,sse2,sse3,sse4a,vzeroupper,x87") }, + { str_lit("bdver1"), str_lit("64bit,64bit-mode,aes,avx,branchfusion,cmov,crc32,cx16,cx8,fast-11bytenop,fast-scalar-shift-masks,fma4,fxsr,lwp,lzcnt,mmx,nopl,pclmul,popcnt,prfchw,sahf,sbb-dep-breaking,slow-shld,sse,sse2,sse3,sse4.1,sse4.2,sse4a,ssse3,vzeroupper,x87,xop,xsave") }, + { str_lit("bdver2"), str_lit("64bit,64bit-mode,aes,avx,bmi,branchfusion,cmov,crc32,cx16,cx8,f16c,fast-11bytenop,fast-bextr,fast-movbe,fast-scalar-shift-masks,fma,fma4,fxsr,lwp,lzcnt,mmx,nopl,pclmul,popcnt,prfchw,sahf,sbb-dep-breaking,slow-shld,sse,sse2,sse3,sse4.1,sse4.2,sse4a,ssse3,tbm,vzeroupper,x87,xop,xsave") }, + { str_lit("bdver3"), str_lit("64bit,64bit-mode,aes,avx,bmi,branchfusion,cmov,crc32,cx16,cx8,f16c,fast-11bytenop,fast-bextr,fast-movbe,fast-scalar-shift-masks,fma,fma4,fsgsbase,fxsr,lwp,lzcnt,mmx,nopl,pclmul,popcnt,prfchw,sahf,sbb-dep-breaking,slow-shld,sse,sse2,sse3,sse4.1,sse4.2,sse4a,ssse3,tbm,vzeroupper,x87,xop,xsave,xsaveopt") }, + { str_lit("bdver4"), str_lit("64bit,64bit-mode,aes,avx,avx2,bmi,bmi2,branchfusion,cmov,crc32,cx16,cx8,f16c,fast-11bytenop,fast-bextr,fast-movbe,fast-scalar-shift-masks,fma,fma4,fsgsbase,fxsr,lwp,lzcnt,mmx,movbe,mwaitx,nopl,pclmul,popcnt,prfchw,rdrnd,sahf,sbb-dep-breaking,slow-shld,sse,sse2,sse3,sse4.1,sse4.2,sse4a,ssse3,tbm,vzeroupper,x87,xop,xsave,xsaveopt") }, + { str_lit("bonnell"), str_lit("64bit,64bit-mode,cmov,cx16,cx8,fxsr,idivl-to-divb,idivq-to-divl,lea-sp,lea-uses-ag,mmx,movbe,no-bypass-delay,nopl,pad-short-functions,sahf,slow-two-mem-ops,slow-unaligned-mem-16,sse,sse2,sse3,ssse3,vzeroupper,x87") }, + { str_lit("broadwell"), str_lit("64bit,64bit-mode,adx,allow-light-256-bit,avx,avx2,bmi,bmi2,cmov,crc32,cx16,cx8,ermsb,f16c,false-deps-lzcnt-tzcnt,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,popcnt,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("btver1"), str_lit("64bit,64bit-mode,cmov,cx16,cx8,fast-15bytenop,fast-scalar-shift-masks,fast-vector-shift-masks,fxsr,lzcnt,mmx,nopl,popcnt,prfchw,sahf,sbb-dep-breaking,slow-shld,sse,sse2,sse3,sse4a,ssse3,vzeroupper,x87") }, + { str_lit("btver2"), str_lit("64bit,64bit-mode,aes,avx,bmi,cmov,crc32,cx16,cx8,f16c,fast-15bytenop,fast-bextr,fast-hops,fast-lzcnt,fast-movbe,fast-scalar-shift-masks,fast-vector-shift-masks,fxsr,lzcnt,mmx,movbe,nopl,pclmul,popcnt,prfchw,sahf,sbb-dep-breaking,slow-shld,sse,sse2,sse3,sse4.1,sse4.2,sse4a,ssse3,x87,xsave,xsaveopt") }, + { str_lit("c3"), str_lit("3dnow,64bit-mode,mmx,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("c3-2"), str_lit("64bit-mode,cmov,cx8,fxsr,mmx,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("cannonlake"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,avx,avx2,avx512bw,avx512cd,avx512dq,avx512f,avx512ifma,avx512vbmi,avx512vl,bmi,bmi2,clflushopt,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pku,popcnt,prefer-256-bit,prfchw,rdrnd,rdseed,sahf,sha,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("cascadelake"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,avx,avx2,avx512bw,avx512cd,avx512dq,avx512f,avx512vl,avx512vnni,bmi,bmi2,clflushopt,clwb,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,false-deps-popcnt,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,faster-shift-than-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pku,popcnt,prefer-256-bit,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("cooperlake"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,avx,avx2,avx512bf16,avx512bw,avx512cd,avx512dq,avx512f,avx512vl,avx512vnni,bmi,bmi2,clflushopt,clwb,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,false-deps-popcnt,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,faster-shift-than-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pku,popcnt,prefer-256-bit,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("core-avx-i"), str_lit("64bit,64bit-mode,avx,cmov,crc32,cx16,cx8,f16c,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fsgsbase,fxsr,idivq-to-divl,macrofusion,mmx,no-bypass-delay-mov,nopl,pclmul,popcnt,rdrnd,sahf,slow-3ops-lea,slow-unaligned-mem-32,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("core-avx2"), str_lit("64bit,64bit-mode,allow-light-256-bit,avx,avx2,bmi,bmi2,cmov,crc32,cx16,cx8,ermsb,f16c,false-deps-lzcnt-tzcnt,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,popcnt,rdrnd,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("core2"), str_lit("64bit,64bit-mode,cmov,cx16,cx8,fxsr,macrofusion,mmx,nopl,sahf,slow-unaligned-mem-16,sse,sse2,sse3,ssse3,vzeroupper,x87") }, + { str_lit("core_2_duo_sse4_1"), str_lit("64bit,64bit-mode,cmov,cx16,cx8,fxsr,macrofusion,mmx,nopl,sahf,slow-unaligned-mem-16,sse,sse2,sse3,sse4.1,ssse3,vzeroupper,x87") }, + { str_lit("core_2_duo_ssse3"), str_lit("64bit,64bit-mode,cmov,cx16,cx8,fxsr,macrofusion,mmx,nopl,sahf,slow-unaligned-mem-16,sse,sse2,sse3,ssse3,vzeroupper,x87") }, + { str_lit("core_2nd_gen_avx"), str_lit("64bit,64bit-mode,avx,cmov,crc32,cx16,cx8,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fxsr,idivq-to-divl,macrofusion,mmx,no-bypass-delay-mov,nopl,pclmul,popcnt,sahf,slow-3ops-lea,slow-unaligned-mem-32,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("core_3rd_gen_avx"), str_lit("64bit,64bit-mode,avx,cmov,crc32,cx16,cx8,f16c,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fsgsbase,fxsr,idivq-to-divl,macrofusion,mmx,no-bypass-delay-mov,nopl,pclmul,popcnt,rdrnd,sahf,slow-3ops-lea,slow-unaligned-mem-32,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("core_4th_gen_avx"), str_lit("64bit,64bit-mode,allow-light-256-bit,avx,avx2,bmi,bmi2,cmov,crc32,cx16,cx8,ermsb,f16c,false-deps-lzcnt-tzcnt,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,popcnt,rdrnd,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("core_4th_gen_avx_tsx"), str_lit("64bit,64bit-mode,allow-light-256-bit,avx,avx2,bmi,bmi2,cmov,crc32,cx16,cx8,ermsb,f16c,false-deps-lzcnt-tzcnt,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,popcnt,rdrnd,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("core_5th_gen_avx"), str_lit("64bit,64bit-mode,adx,allow-light-256-bit,avx,avx2,bmi,bmi2,cmov,crc32,cx16,cx8,ermsb,f16c,false-deps-lzcnt-tzcnt,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,popcnt,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("core_5th_gen_avx_tsx"), str_lit("64bit,64bit-mode,adx,allow-light-256-bit,avx,avx2,bmi,bmi2,cmov,crc32,cx16,cx8,ermsb,f16c,false-deps-lzcnt-tzcnt,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,popcnt,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("core_aes_pclmulqdq"), str_lit("64bit,64bit-mode,cmov,crc32,cx16,cx8,fxsr,macrofusion,mmx,no-bypass-delay-mov,nopl,pclmul,popcnt,sahf,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87") }, + { str_lit("core_i7_sse4_2"), str_lit("64bit,64bit-mode,cmov,crc32,cx16,cx8,fxsr,macrofusion,mmx,no-bypass-delay-mov,nopl,popcnt,sahf,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87") }, + { str_lit("corei7"), str_lit("64bit,64bit-mode,cmov,crc32,cx16,cx8,fxsr,macrofusion,mmx,no-bypass-delay-mov,nopl,popcnt,sahf,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87") }, + { str_lit("corei7-avx"), str_lit("64bit,64bit-mode,avx,cmov,crc32,cx16,cx8,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fxsr,idivq-to-divl,macrofusion,mmx,no-bypass-delay-mov,nopl,pclmul,popcnt,sahf,slow-3ops-lea,slow-unaligned-mem-32,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("emeraldrapids"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,amx-bf16,amx-int8,amx-tile,avx,avx2,avx512bf16,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512fp16,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,avxvnni,bmi,bmi2,cldemote,clflushopt,clwb,cmov,crc32,cx16,cx8,enqcmd,ermsb,evex512,f16c,false-deps-getmant,false-deps-mulc,false-deps-mullq,false-deps-perm,false-deps-range,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,movdir64b,movdiri,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pconfig,pku,popcnt,prefer-256-bit,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,serialize,sha,shstk,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tsxldtrk,tuning-fast-imm-vector-shift,uintr,vaes,vpclmulqdq,vzeroupper,waitpkg,wbnoinvd,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("generic"), str_lit("64bit,64bit-mode,cx8,fast-15bytenop,fast-scalar-fsqrt,idivq-to-divl,macrofusion,slow-3ops-lea,sse,sse2,vzeroupper,x87") }, + { str_lit("geode"), str_lit("3dnow,3dnowa,64bit-mode,cx8,mmx,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("goldmont"), str_lit("64bit,64bit-mode,aes,clflushopt,cmov,crc32,cx16,cx8,false-deps-popcnt,fast-movbe,fsgsbase,fxsr,mmx,movbe,no-bypass-delay,nopl,pclmul,popcnt,prfchw,rdrnd,rdseed,sahf,sha,slow-incdec,slow-lea,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,use-glm-div-sqrt-costs,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("goldmont-plus"), str_lit("64bit,64bit-mode,aes,clflushopt,cmov,crc32,cx16,cx8,fast-movbe,fsgsbase,fxsr,mmx,movbe,no-bypass-delay,nopl,pclmul,popcnt,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,sha,slow-incdec,slow-lea,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,use-glm-div-sqrt-costs,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("goldmont_plus"), str_lit("64bit,64bit-mode,aes,clflushopt,cmov,crc32,cx16,cx8,fast-movbe,fsgsbase,fxsr,mmx,movbe,no-bypass-delay,nopl,pclmul,popcnt,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,sha,slow-incdec,slow-lea,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,use-glm-div-sqrt-costs,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("grandridge"), str_lit("64bit,64bit-mode,adx,aes,avx,avx2,avxifma,avxneconvert,avxvnni,avxvnniint8,bmi,bmi2,cldemote,clflushopt,clwb,cmov,cmpccxadd,crc32,cx16,cx8,enqcmd,f16c,fast-movbe,fma,fsgsbase,fxsr,gfni,hreset,invpcid,kl,lzcnt,mmx,movbe,movdir64b,movdiri,no-bypass-delay,nopl,pclmul,pconfig,pku,popcnt,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,serialize,sha,shstk,slow-incdec,slow-lea,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,uintr,use-glm-div-sqrt-costs,vaes,vpclmulqdq,vzeroupper,waitpkg,widekl,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("graniterapids"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,amx-bf16,amx-fp16,amx-int8,amx-tile,avx,avx2,avx512bf16,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512fp16,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,avxvnni,bmi,bmi2,cldemote,clflushopt,clwb,cmov,crc32,cx16,cx8,enqcmd,ermsb,evex512,f16c,false-deps-getmant,false-deps-mulc,false-deps-mullq,false-deps-perm,false-deps-range,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,movdir64b,movdiri,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pconfig,pku,popcnt,prefer-256-bit,prefetchi,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,serialize,sha,shstk,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tsxldtrk,tuning-fast-imm-vector-shift,uintr,vaes,vpclmulqdq,vzeroupper,waitpkg,wbnoinvd,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("graniterapids-d"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,amx-bf16,amx-complex,amx-fp16,amx-int8,amx-tile,avx,avx2,avx512bf16,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512fp16,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,avxvnni,bmi,bmi2,cldemote,clflushopt,clwb,cmov,crc32,cx16,cx8,enqcmd,ermsb,evex512,f16c,false-deps-getmant,false-deps-mulc,false-deps-mullq,false-deps-perm,false-deps-range,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,movdir64b,movdiri,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pconfig,pku,popcnt,prefer-256-bit,prefetchi,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,serialize,sha,shstk,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tsxldtrk,tuning-fast-imm-vector-shift,uintr,vaes,vpclmulqdq,vzeroupper,waitpkg,wbnoinvd,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("graniterapids_d"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,amx-bf16,amx-complex,amx-fp16,amx-int8,amx-tile,avx,avx2,avx512bf16,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512fp16,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,avxvnni,bmi,bmi2,cldemote,clflushopt,clwb,cmov,crc32,cx16,cx8,enqcmd,ermsb,evex512,f16c,false-deps-getmant,false-deps-mulc,false-deps-mullq,false-deps-perm,false-deps-range,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,movdir64b,movdiri,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pconfig,pku,popcnt,prefer-256-bit,prefetchi,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,serialize,sha,shstk,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tsxldtrk,tuning-fast-imm-vector-shift,uintr,vaes,vpclmulqdq,vzeroupper,waitpkg,wbnoinvd,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("haswell"), str_lit("64bit,64bit-mode,allow-light-256-bit,avx,avx2,bmi,bmi2,cmov,crc32,cx16,cx8,ermsb,f16c,false-deps-lzcnt-tzcnt,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,popcnt,rdrnd,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("i386"), str_lit("64bit-mode,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("i486"), str_lit("64bit-mode,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("i586"), str_lit("64bit-mode,cx8,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("i686"), str_lit("64bit-mode,cmov,cx8,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("icelake-client"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,avx,avx2,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,bmi,bmi2,clflushopt,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pku,popcnt,prefer-256-bit,prfchw,rdpid,rdrnd,rdseed,sahf,sha,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vaes,vpclmulqdq,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("icelake-server"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,avx,avx2,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,bmi,bmi2,clflushopt,clwb,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pconfig,pku,popcnt,prefer-256-bit,prfchw,rdpid,rdrnd,rdseed,sahf,sha,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vaes,vpclmulqdq,vzeroupper,wbnoinvd,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("icelake_client"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,avx,avx2,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,bmi,bmi2,clflushopt,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pku,popcnt,prefer-256-bit,prfchw,rdpid,rdrnd,rdseed,sahf,sha,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vaes,vpclmulqdq,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("icelake_server"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,avx,avx2,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,bmi,bmi2,clflushopt,clwb,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pconfig,pku,popcnt,prefer-256-bit,prfchw,rdpid,rdrnd,rdseed,sahf,sha,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vaes,vpclmulqdq,vzeroupper,wbnoinvd,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("ivybridge"), str_lit("64bit,64bit-mode,avx,cmov,crc32,cx16,cx8,f16c,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fsgsbase,fxsr,idivq-to-divl,macrofusion,mmx,no-bypass-delay-mov,nopl,pclmul,popcnt,rdrnd,sahf,slow-3ops-lea,slow-unaligned-mem-32,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("k6"), str_lit("64bit-mode,cx8,mmx,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("k6-2"), str_lit("3dnow,64bit-mode,cx8,mmx,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("k6-3"), str_lit("3dnow,64bit-mode,cx8,mmx,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("k8"), str_lit("3dnow,3dnowa,64bit,64bit-mode,cmov,cx8,fast-scalar-shift-masks,fxsr,mmx,nopl,sbb-dep-breaking,slow-shld,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("k8-sse3"), str_lit("3dnow,3dnowa,64bit,64bit-mode,cmov,cx16,cx8,fast-scalar-shift-masks,fxsr,mmx,nopl,sbb-dep-breaking,slow-shld,slow-unaligned-mem-16,sse,sse2,sse3,vzeroupper,x87") }, + { str_lit("knl"), str_lit("64bit,64bit-mode,adx,aes,avx,avx2,avx512cd,avx512er,avx512f,avx512pf,bmi,bmi2,cmov,crc32,cx16,cx8,evex512,f16c,fast-gather,fast-movbe,fma,fsgsbase,fxsr,idivq-to-divl,lzcnt,mmx,movbe,nopl,pclmul,popcnt,prefer-mask-registers,prefetchwt1,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,slow-incdec,slow-pmaddwd,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,x87,xsave,xsaveopt") }, + { str_lit("knm"), str_lit("64bit,64bit-mode,adx,aes,avx,avx2,avx512cd,avx512er,avx512f,avx512pf,avx512vpopcntdq,bmi,bmi2,cmov,crc32,cx16,cx8,evex512,f16c,fast-gather,fast-movbe,fma,fsgsbase,fxsr,idivq-to-divl,lzcnt,mmx,movbe,nopl,pclmul,popcnt,prefer-mask-registers,prefetchwt1,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,slow-incdec,slow-pmaddwd,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,x87,xsave,xsaveopt") }, + { str_lit("lakemont"), str_lit("64bit-mode,cx8,slow-unaligned-mem-16,sse,sse2,vzeroupper") }, + { str_lit("meteorlake"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,avx,avx2,avxvnni,bmi,bmi2,cldemote,clflushopt,clwb,cmov,crc32,cx16,cx8,f16c,false-deps-perm,false-deps-popcnt,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fxsr,gfni,hreset,idivq-to-divl,invpcid,kl,lzcnt,macrofusion,mmx,movbe,movdir64b,movdiri,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pconfig,pku,popcnt,prefer-movmsk-over-vtest,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,serialize,sha,shstk,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vaes,vpclmulqdq,vzeroupper,waitpkg,widekl,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("mic_avx512"), str_lit("64bit,64bit-mode,adx,aes,avx,avx2,avx512cd,avx512er,avx512f,avx512pf,bmi,bmi2,cmov,crc32,cx16,cx8,evex512,f16c,fast-gather,fast-movbe,fma,fsgsbase,fxsr,idivq-to-divl,lzcnt,mmx,movbe,nopl,pclmul,popcnt,prefer-mask-registers,prefetchwt1,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,slow-incdec,slow-pmaddwd,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,x87,xsave,xsaveopt") }, + { str_lit("nehalem"), str_lit("64bit,64bit-mode,cmov,crc32,cx16,cx8,fxsr,macrofusion,mmx,no-bypass-delay-mov,nopl,popcnt,sahf,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87") }, + { str_lit("nocona"), str_lit("64bit,64bit-mode,cmov,cx16,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,sse3,vzeroupper,x87") }, + { str_lit("opteron"), str_lit("3dnow,3dnowa,64bit,64bit-mode,cmov,cx8,fast-scalar-shift-masks,fxsr,mmx,nopl,sbb-dep-breaking,slow-shld,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("opteron-sse3"), str_lit("3dnow,3dnowa,64bit,64bit-mode,cmov,cx16,cx8,fast-scalar-shift-masks,fxsr,mmx,nopl,sbb-dep-breaking,slow-shld,slow-unaligned-mem-16,sse,sse2,sse3,vzeroupper,x87") }, + { str_lit("penryn"), str_lit("64bit,64bit-mode,cmov,cx16,cx8,fxsr,macrofusion,mmx,nopl,sahf,slow-unaligned-mem-16,sse,sse2,sse3,sse4.1,ssse3,vzeroupper,x87") }, + { str_lit("pentium"), str_lit("64bit-mode,cx8,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("pentium-m"), str_lit("64bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("pentium-mmx"), str_lit("64bit-mode,cx8,mmx,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("pentium2"), str_lit("64bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("pentium3"), str_lit("64bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("pentium3m"), str_lit("64bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("pentium4"), str_lit("64bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("pentium4m"), str_lit("64bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("pentium_4"), str_lit("64bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("pentium_4_sse3"), str_lit("64bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,sse3,vzeroupper,x87") }, + { str_lit("pentium_ii"), str_lit("64bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("pentium_iii"), str_lit("64bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("pentium_iii_no_xmm_regs"), str_lit("64bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("pentium_m"), str_lit("64bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("pentium_mmx"), str_lit("64bit-mode,cx8,mmx,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("pentium_pro"), str_lit("64bit-mode,cmov,cx8,nopl,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("pentiumpro"), str_lit("64bit-mode,cmov,cx8,nopl,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("prescott"), str_lit("64bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,sse3,vzeroupper,x87") }, + { str_lit("raptorlake"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,avx,avx2,avxvnni,bmi,bmi2,cldemote,clflushopt,clwb,cmov,crc32,cx16,cx8,f16c,false-deps-perm,false-deps-popcnt,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fxsr,gfni,hreset,idivq-to-divl,invpcid,kl,lzcnt,macrofusion,mmx,movbe,movdir64b,movdiri,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pconfig,pku,popcnt,prefer-movmsk-over-vtest,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,serialize,sha,shstk,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vaes,vpclmulqdq,vzeroupper,waitpkg,widekl,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("rocketlake"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,avx,avx2,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,bmi,bmi2,clflushopt,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pku,popcnt,prefer-256-bit,prfchw,rdpid,rdrnd,rdseed,sahf,sha,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vaes,vpclmulqdq,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("sandybridge"), str_lit("64bit,64bit-mode,avx,cmov,crc32,cx16,cx8,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fxsr,idivq-to-divl,macrofusion,mmx,no-bypass-delay-mov,nopl,pclmul,popcnt,sahf,slow-3ops-lea,slow-unaligned-mem-32,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("sapphirerapids"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,amx-bf16,amx-int8,amx-tile,avx,avx2,avx512bf16,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512fp16,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,avxvnni,bmi,bmi2,cldemote,clflushopt,clwb,cmov,crc32,cx16,cx8,enqcmd,ermsb,evex512,f16c,false-deps-getmant,false-deps-mulc,false-deps-mullq,false-deps-perm,false-deps-range,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,movdir64b,movdiri,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pconfig,pku,popcnt,prefer-256-bit,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,serialize,sha,shstk,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tsxldtrk,tuning-fast-imm-vector-shift,uintr,vaes,vpclmulqdq,vzeroupper,waitpkg,wbnoinvd,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("sierraforest"), str_lit("64bit,64bit-mode,adx,aes,avx,avx2,avxifma,avxneconvert,avxvnni,avxvnniint8,bmi,bmi2,cldemote,clflushopt,clwb,cmov,cmpccxadd,crc32,cx16,cx8,enqcmd,f16c,fast-movbe,fma,fsgsbase,fxsr,gfni,hreset,invpcid,kl,lzcnt,mmx,movbe,movdir64b,movdiri,no-bypass-delay,nopl,pclmul,pconfig,pku,popcnt,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,serialize,sha,shstk,slow-incdec,slow-lea,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,uintr,use-glm-div-sqrt-costs,vaes,vpclmulqdq,vzeroupper,waitpkg,widekl,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("silvermont"), str_lit("64bit,64bit-mode,cmov,crc32,cx16,cx8,false-deps-popcnt,fast-7bytenop,fast-movbe,fxsr,idivq-to-divl,mmx,movbe,no-bypass-delay,nopl,pclmul,popcnt,prfchw,rdrnd,sahf,slow-incdec,slow-lea,slow-pmulld,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,use-slm-arith-costs,vzeroupper,x87") }, + { str_lit("skx"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,avx,avx2,avx512bw,avx512cd,avx512dq,avx512f,avx512vl,bmi,bmi2,clflushopt,clwb,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,false-deps-popcnt,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,faster-shift-than-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pku,popcnt,prefer-256-bit,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("skylake"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,avx,avx2,bmi,bmi2,clflushopt,cmov,crc32,cx16,cx8,ermsb,f16c,false-deps-popcnt,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,popcnt,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("skylake-avx512"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,avx,avx2,avx512bw,avx512cd,avx512dq,avx512f,avx512vl,bmi,bmi2,clflushopt,clwb,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,false-deps-popcnt,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,faster-shift-than-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pku,popcnt,prefer-256-bit,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("skylake_avx512"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,avx,avx2,avx512bw,avx512cd,avx512dq,avx512f,avx512vl,bmi,bmi2,clflushopt,clwb,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,false-deps-popcnt,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,faster-shift-than-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pku,popcnt,prefer-256-bit,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("slm"), str_lit("64bit,64bit-mode,cmov,crc32,cx16,cx8,false-deps-popcnt,fast-7bytenop,fast-movbe,fxsr,idivq-to-divl,mmx,movbe,no-bypass-delay,nopl,pclmul,popcnt,prfchw,rdrnd,sahf,slow-incdec,slow-lea,slow-pmulld,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,use-slm-arith-costs,vzeroupper,x87") }, + { str_lit("tigerlake"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,avx,avx2,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vp2intersect,avx512vpopcntdq,bmi,bmi2,clflushopt,clwb,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,movdir64b,movdiri,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pku,popcnt,prefer-256-bit,prfchw,rdpid,rdrnd,rdseed,sahf,sha,shstk,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vaes,vpclmulqdq,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("tremont"), str_lit("64bit,64bit-mode,aes,clflushopt,clwb,cmov,crc32,cx16,cx8,fast-movbe,fsgsbase,fxsr,gfni,mmx,movbe,no-bypass-delay,nopl,pclmul,popcnt,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,sha,slow-incdec,slow-lea,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,use-glm-div-sqrt-costs,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("westmere"), str_lit("64bit,64bit-mode,cmov,crc32,cx16,cx8,fxsr,macrofusion,mmx,no-bypass-delay-mov,nopl,pclmul,popcnt,sahf,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87") }, + { str_lit("winchip-c6"), str_lit("64bit-mode,mmx,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("winchip2"), str_lit("3dnow,64bit-mode,mmx,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("x86-64"), str_lit("64bit,64bit-mode,cmov,cx8,fxsr,idivq-to-divl,macrofusion,mmx,nopl,slow-3ops-lea,slow-incdec,sse,sse2,vzeroupper,x87") }, + { str_lit("x86-64-v2"), str_lit("64bit,64bit-mode,cmov,crc32,cx16,cx8,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fxsr,idivq-to-divl,macrofusion,mmx,nopl,popcnt,sahf,slow-3ops-lea,slow-unaligned-mem-32,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87") }, + { str_lit("x86-64-v3"), str_lit("64bit,64bit-mode,allow-light-256-bit,avx,avx2,bmi,bmi2,cmov,crc32,cx16,cx8,f16c,false-deps-lzcnt-tzcnt,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fma,fxsr,idivq-to-divl,lzcnt,macrofusion,mmx,movbe,nopl,popcnt,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave") }, + { str_lit("x86-64-v4"), str_lit("64bit,64bit-mode,allow-light-256-bit,avx,avx2,avx512bw,avx512cd,avx512dq,avx512f,avx512vl,bmi,bmi2,cmov,crc32,cx16,cx8,evex512,f16c,false-deps-popcnt,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fxsr,idivq-to-divl,lzcnt,macrofusion,mmx,movbe,nopl,popcnt,prefer-256-bit,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave") }, + { str_lit("yonah"), str_lit("64bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,sse3,vzeroupper,x87") }, + { str_lit("znver1"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,avx,avx2,bmi,bmi2,branchfusion,clflushopt,clzero,cmov,crc32,cx16,cx8,f16c,fast-15bytenop,fast-bextr,fast-lzcnt,fast-movbe,fast-scalar-fsqrt,fast-scalar-shift-masks,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fxsr,lzcnt,mmx,movbe,mwaitx,nopl,pclmul,popcnt,prfchw,rdrnd,rdseed,sahf,sbb-dep-breaking,sha,slow-shld,sse,sse2,sse3,sse4.1,sse4.2,sse4a,ssse3,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("znver2"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,avx,avx2,bmi,bmi2,branchfusion,clflushopt,clwb,clzero,cmov,crc32,cx16,cx8,f16c,fast-15bytenop,fast-bextr,fast-lzcnt,fast-movbe,fast-scalar-fsqrt,fast-scalar-shift-masks,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fxsr,lzcnt,mmx,movbe,mwaitx,nopl,pclmul,popcnt,prfchw,rdpid,rdpru,rdrnd,rdseed,sahf,sbb-dep-breaking,sha,slow-shld,sse,sse2,sse3,sse4.1,sse4.2,sse4a,ssse3,vzeroupper,wbnoinvd,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("znver3"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,avx,avx2,bmi,bmi2,branchfusion,clflushopt,clwb,clzero,cmov,crc32,cx16,cx8,f16c,fast-15bytenop,fast-bextr,fast-lzcnt,fast-movbe,fast-scalar-fsqrt,fast-scalar-shift-masks,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,invpcid,lzcnt,macrofusion,mmx,movbe,mwaitx,nopl,pclmul,pku,popcnt,prfchw,rdpid,rdpru,rdrnd,rdseed,sahf,sbb-dep-breaking,sha,slow-shld,sse,sse2,sse3,sse4.1,sse4.2,sse4a,ssse3,vaes,vpclmulqdq,vzeroupper,wbnoinvd,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("znver4"), str_lit("64bit,64bit-mode,adx,aes,allow-light-256-bit,avx,avx2,avx512bf16,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,bmi,bmi2,branchfusion,clflushopt,clwb,clzero,cmov,crc32,cx16,cx8,evex512,f16c,fast-15bytenop,fast-bextr,fast-lzcnt,fast-movbe,fast-scalar-fsqrt,fast-scalar-shift-masks,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,invpcid,lzcnt,macrofusion,mmx,movbe,mwaitx,nopl,pclmul,pku,popcnt,prfchw,rdpid,rdpru,rdrnd,rdseed,sahf,sbb-dep-breaking,sha,shstk,slow-shld,sse,sse2,sse3,sse4.1,sse4.2,sse4a,ssse3,vaes,vpclmulqdq,vzeroupper,wbnoinvd,x87,xsave,xsavec,xsaveopt,xsaves") }, + // TargetArch_i386: + { str_lit("alderlake"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,avx,avx2,avxvnni,bmi,bmi2,cldemote,clflushopt,clwb,cmov,crc32,cx16,cx8,f16c,false-deps-perm,false-deps-popcnt,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fxsr,gfni,hreset,idivq-to-divl,invpcid,kl,lzcnt,macrofusion,mmx,movbe,movdir64b,movdiri,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pconfig,pku,popcnt,prefer-movmsk-over-vtest,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,serialize,sha,shstk,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vaes,vpclmulqdq,vzeroupper,waitpkg,widekl,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("amdfam10"), str_lit("32bit-mode,3dnow,3dnowa,64bit,cmov,cx16,cx8,fast-scalar-shift-masks,fxsr,lzcnt,mmx,nopl,popcnt,prfchw,sahf,sbb-dep-breaking,slow-shld,sse,sse2,sse3,sse4a,vzeroupper,x87") }, + { str_lit("athlon"), str_lit("32bit-mode,3dnow,3dnowa,cmov,cx8,mmx,nopl,slow-shld,slow-unaligned-mem-16,vzeroupper,x87") }, + { str_lit("athlon-4"), str_lit("32bit-mode,3dnow,3dnowa,cmov,cx8,fxsr,mmx,nopl,slow-shld,slow-unaligned-mem-16,sse,vzeroupper,x87") }, + { str_lit("athlon-fx"), str_lit("32bit-mode,3dnow,3dnowa,64bit,cmov,cx8,fast-scalar-shift-masks,fxsr,mmx,nopl,sbb-dep-breaking,slow-shld,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("athlon-mp"), str_lit("32bit-mode,3dnow,3dnowa,cmov,cx8,fxsr,mmx,nopl,slow-shld,slow-unaligned-mem-16,sse,vzeroupper,x87") }, + { str_lit("athlon-tbird"), str_lit("32bit-mode,3dnow,3dnowa,cmov,cx8,mmx,nopl,slow-shld,slow-unaligned-mem-16,vzeroupper,x87") }, + { str_lit("athlon-xp"), str_lit("32bit-mode,3dnow,3dnowa,cmov,cx8,fxsr,mmx,nopl,slow-shld,slow-unaligned-mem-16,sse,vzeroupper,x87") }, + { str_lit("athlon64"), str_lit("32bit-mode,3dnow,3dnowa,64bit,cmov,cx8,fast-scalar-shift-masks,fxsr,mmx,nopl,sbb-dep-breaking,slow-shld,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("athlon64-sse3"), str_lit("32bit-mode,3dnow,3dnowa,64bit,cmov,cx16,cx8,fast-scalar-shift-masks,fxsr,mmx,nopl,sbb-dep-breaking,slow-shld,slow-unaligned-mem-16,sse,sse2,sse3,vzeroupper,x87") }, + { str_lit("atom"), str_lit("32bit-mode,64bit,cmov,cx16,cx8,fxsr,idivl-to-divb,idivq-to-divl,lea-sp,lea-uses-ag,mmx,movbe,no-bypass-delay,nopl,pad-short-functions,sahf,slow-two-mem-ops,slow-unaligned-mem-16,sse,sse2,sse3,ssse3,vzeroupper,x87") }, + { str_lit("atom_sse4_2"), str_lit("32bit-mode,64bit,cmov,crc32,cx16,cx8,false-deps-popcnt,fast-7bytenop,fast-movbe,fxsr,idivq-to-divl,mmx,movbe,no-bypass-delay,nopl,pclmul,popcnt,prfchw,rdrnd,sahf,slow-incdec,slow-lea,slow-pmulld,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,use-slm-arith-costs,vzeroupper,x87") }, + { str_lit("atom_sse4_2_movbe"), str_lit("32bit-mode,64bit,aes,clflushopt,cmov,crc32,cx16,cx8,false-deps-popcnt,fast-7bytenop,fast-movbe,fsgsbase,fxsr,idivq-to-divl,mmx,movbe,no-bypass-delay,nopl,pclmul,popcnt,prfchw,rdrnd,rdseed,sahf,sha,slow-incdec,slow-lea,slow-pmulld,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,use-slm-arith-costs,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("barcelona"), str_lit("32bit-mode,3dnow,3dnowa,64bit,cmov,cx16,cx8,fast-scalar-shift-masks,fxsr,lzcnt,mmx,nopl,popcnt,prfchw,sahf,sbb-dep-breaking,slow-shld,sse,sse2,sse3,sse4a,vzeroupper,x87") }, + { str_lit("bdver1"), str_lit("32bit-mode,64bit,aes,avx,branchfusion,cmov,crc32,cx16,cx8,fast-11bytenop,fast-scalar-shift-masks,fma4,fxsr,lwp,lzcnt,mmx,nopl,pclmul,popcnt,prfchw,sahf,sbb-dep-breaking,slow-shld,sse,sse2,sse3,sse4.1,sse4.2,sse4a,ssse3,vzeroupper,x87,xop,xsave") }, + { str_lit("bdver2"), str_lit("32bit-mode,64bit,aes,avx,bmi,branchfusion,cmov,crc32,cx16,cx8,f16c,fast-11bytenop,fast-bextr,fast-movbe,fast-scalar-shift-masks,fma,fma4,fxsr,lwp,lzcnt,mmx,nopl,pclmul,popcnt,prfchw,sahf,sbb-dep-breaking,slow-shld,sse,sse2,sse3,sse4.1,sse4.2,sse4a,ssse3,tbm,vzeroupper,x87,xop,xsave") }, + { str_lit("bdver3"), str_lit("32bit-mode,64bit,aes,avx,bmi,branchfusion,cmov,crc32,cx16,cx8,f16c,fast-11bytenop,fast-bextr,fast-movbe,fast-scalar-shift-masks,fma,fma4,fsgsbase,fxsr,lwp,lzcnt,mmx,nopl,pclmul,popcnt,prfchw,sahf,sbb-dep-breaking,slow-shld,sse,sse2,sse3,sse4.1,sse4.2,sse4a,ssse3,tbm,vzeroupper,x87,xop,xsave,xsaveopt") }, + { str_lit("bdver4"), str_lit("32bit-mode,64bit,aes,avx,avx2,bmi,bmi2,branchfusion,cmov,crc32,cx16,cx8,f16c,fast-11bytenop,fast-bextr,fast-movbe,fast-scalar-shift-masks,fma,fma4,fsgsbase,fxsr,lwp,lzcnt,mmx,movbe,mwaitx,nopl,pclmul,popcnt,prfchw,rdrnd,sahf,sbb-dep-breaking,slow-shld,sse,sse2,sse3,sse4.1,sse4.2,sse4a,ssse3,tbm,vzeroupper,x87,xop,xsave,xsaveopt") }, + { str_lit("bonnell"), str_lit("32bit-mode,64bit,cmov,cx16,cx8,fxsr,idivl-to-divb,idivq-to-divl,lea-sp,lea-uses-ag,mmx,movbe,no-bypass-delay,nopl,pad-short-functions,sahf,slow-two-mem-ops,slow-unaligned-mem-16,sse,sse2,sse3,ssse3,vzeroupper,x87") }, + { str_lit("broadwell"), str_lit("32bit-mode,64bit,adx,allow-light-256-bit,avx,avx2,bmi,bmi2,cmov,crc32,cx16,cx8,ermsb,f16c,false-deps-lzcnt-tzcnt,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,popcnt,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("btver1"), str_lit("32bit-mode,64bit,cmov,cx16,cx8,fast-15bytenop,fast-scalar-shift-masks,fast-vector-shift-masks,fxsr,lzcnt,mmx,nopl,popcnt,prfchw,sahf,sbb-dep-breaking,slow-shld,sse,sse2,sse3,sse4a,ssse3,vzeroupper,x87") }, + { str_lit("btver2"), str_lit("32bit-mode,64bit,aes,avx,bmi,cmov,crc32,cx16,cx8,f16c,fast-15bytenop,fast-bextr,fast-hops,fast-lzcnt,fast-movbe,fast-scalar-shift-masks,fast-vector-shift-masks,fxsr,lzcnt,mmx,movbe,nopl,pclmul,popcnt,prfchw,sahf,sbb-dep-breaking,slow-shld,sse,sse2,sse3,sse4.1,sse4.2,sse4a,ssse3,x87,xsave,xsaveopt") }, + { str_lit("c3"), str_lit("32bit-mode,3dnow,mmx,slow-unaligned-mem-16,vzeroupper,x87") }, + { str_lit("c3-2"), str_lit("32bit-mode,cmov,cx8,fxsr,mmx,slow-unaligned-mem-16,sse,vzeroupper,x87") }, + { str_lit("cannonlake"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,avx,avx2,avx512bw,avx512cd,avx512dq,avx512f,avx512ifma,avx512vbmi,avx512vl,bmi,bmi2,clflushopt,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pku,popcnt,prefer-256-bit,prfchw,rdrnd,rdseed,sahf,sha,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("cascadelake"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,avx,avx2,avx512bw,avx512cd,avx512dq,avx512f,avx512vl,avx512vnni,bmi,bmi2,clflushopt,clwb,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,false-deps-popcnt,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,faster-shift-than-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pku,popcnt,prefer-256-bit,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("cooperlake"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,avx,avx2,avx512bf16,avx512bw,avx512cd,avx512dq,avx512f,avx512vl,avx512vnni,bmi,bmi2,clflushopt,clwb,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,false-deps-popcnt,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,faster-shift-than-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pku,popcnt,prefer-256-bit,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("core-avx-i"), str_lit("32bit-mode,64bit,avx,cmov,crc32,cx16,cx8,f16c,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fsgsbase,fxsr,idivq-to-divl,macrofusion,mmx,no-bypass-delay-mov,nopl,pclmul,popcnt,rdrnd,sahf,slow-3ops-lea,slow-unaligned-mem-32,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("core-avx2"), str_lit("32bit-mode,64bit,allow-light-256-bit,avx,avx2,bmi,bmi2,cmov,crc32,cx16,cx8,ermsb,f16c,false-deps-lzcnt-tzcnt,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,popcnt,rdrnd,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("core2"), str_lit("32bit-mode,64bit,cmov,cx16,cx8,fxsr,macrofusion,mmx,nopl,sahf,slow-unaligned-mem-16,sse,sse2,sse3,ssse3,vzeroupper,x87") }, + { str_lit("core_2_duo_sse4_1"), str_lit("32bit-mode,64bit,cmov,cx16,cx8,fxsr,macrofusion,mmx,nopl,sahf,slow-unaligned-mem-16,sse,sse2,sse3,sse4.1,ssse3,vzeroupper,x87") }, + { str_lit("core_2_duo_ssse3"), str_lit("32bit-mode,64bit,cmov,cx16,cx8,fxsr,macrofusion,mmx,nopl,sahf,slow-unaligned-mem-16,sse,sse2,sse3,ssse3,vzeroupper,x87") }, + { str_lit("core_2nd_gen_avx"), str_lit("32bit-mode,64bit,avx,cmov,crc32,cx16,cx8,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fxsr,idivq-to-divl,macrofusion,mmx,no-bypass-delay-mov,nopl,pclmul,popcnt,sahf,slow-3ops-lea,slow-unaligned-mem-32,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("core_3rd_gen_avx"), str_lit("32bit-mode,64bit,avx,cmov,crc32,cx16,cx8,f16c,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fsgsbase,fxsr,idivq-to-divl,macrofusion,mmx,no-bypass-delay-mov,nopl,pclmul,popcnt,rdrnd,sahf,slow-3ops-lea,slow-unaligned-mem-32,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("core_4th_gen_avx"), str_lit("32bit-mode,64bit,allow-light-256-bit,avx,avx2,bmi,bmi2,cmov,crc32,cx16,cx8,ermsb,f16c,false-deps-lzcnt-tzcnt,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,popcnt,rdrnd,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("core_4th_gen_avx_tsx"), str_lit("32bit-mode,64bit,allow-light-256-bit,avx,avx2,bmi,bmi2,cmov,crc32,cx16,cx8,ermsb,f16c,false-deps-lzcnt-tzcnt,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,popcnt,rdrnd,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("core_5th_gen_avx"), str_lit("32bit-mode,64bit,adx,allow-light-256-bit,avx,avx2,bmi,bmi2,cmov,crc32,cx16,cx8,ermsb,f16c,false-deps-lzcnt-tzcnt,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,popcnt,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("core_5th_gen_avx_tsx"), str_lit("32bit-mode,64bit,adx,allow-light-256-bit,avx,avx2,bmi,bmi2,cmov,crc32,cx16,cx8,ermsb,f16c,false-deps-lzcnt-tzcnt,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,popcnt,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("core_aes_pclmulqdq"), str_lit("32bit-mode,64bit,cmov,crc32,cx16,cx8,fxsr,macrofusion,mmx,no-bypass-delay-mov,nopl,pclmul,popcnt,sahf,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87") }, + { str_lit("core_i7_sse4_2"), str_lit("32bit-mode,64bit,cmov,crc32,cx16,cx8,fxsr,macrofusion,mmx,no-bypass-delay-mov,nopl,popcnt,sahf,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87") }, + { str_lit("corei7"), str_lit("32bit-mode,64bit,cmov,crc32,cx16,cx8,fxsr,macrofusion,mmx,no-bypass-delay-mov,nopl,popcnt,sahf,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87") }, + { str_lit("corei7-avx"), str_lit("32bit-mode,64bit,avx,cmov,crc32,cx16,cx8,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fxsr,idivq-to-divl,macrofusion,mmx,no-bypass-delay-mov,nopl,pclmul,popcnt,sahf,slow-3ops-lea,slow-unaligned-mem-32,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("emeraldrapids"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,amx-bf16,amx-int8,amx-tile,avx,avx2,avx512bf16,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512fp16,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,avxvnni,bmi,bmi2,cldemote,clflushopt,clwb,cmov,crc32,cx16,cx8,enqcmd,ermsb,evex512,f16c,false-deps-getmant,false-deps-mulc,false-deps-mullq,false-deps-perm,false-deps-range,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,movdir64b,movdiri,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pconfig,pku,popcnt,prefer-256-bit,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,serialize,sha,shstk,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tsxldtrk,tuning-fast-imm-vector-shift,uintr,vaes,vpclmulqdq,vzeroupper,waitpkg,wbnoinvd,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("generic"), str_lit("32bit-mode,64bit,cx8,fast-15bytenop,fast-scalar-fsqrt,idivq-to-divl,macrofusion,slow-3ops-lea,vzeroupper,x87") }, + { str_lit("geode"), str_lit("32bit-mode,3dnow,3dnowa,cx8,mmx,slow-unaligned-mem-16,vzeroupper,x87") }, + { str_lit("goldmont"), str_lit("32bit-mode,64bit,aes,clflushopt,cmov,crc32,cx16,cx8,false-deps-popcnt,fast-movbe,fsgsbase,fxsr,mmx,movbe,no-bypass-delay,nopl,pclmul,popcnt,prfchw,rdrnd,rdseed,sahf,sha,slow-incdec,slow-lea,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,use-glm-div-sqrt-costs,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("goldmont-plus"), str_lit("32bit-mode,64bit,aes,clflushopt,cmov,crc32,cx16,cx8,fast-movbe,fsgsbase,fxsr,mmx,movbe,no-bypass-delay,nopl,pclmul,popcnt,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,sha,slow-incdec,slow-lea,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,use-glm-div-sqrt-costs,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("goldmont_plus"), str_lit("32bit-mode,64bit,aes,clflushopt,cmov,crc32,cx16,cx8,fast-movbe,fsgsbase,fxsr,mmx,movbe,no-bypass-delay,nopl,pclmul,popcnt,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,sha,slow-incdec,slow-lea,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,use-glm-div-sqrt-costs,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("grandridge"), str_lit("32bit-mode,64bit,adx,aes,avx,avx2,avxifma,avxneconvert,avxvnni,avxvnniint8,bmi,bmi2,cldemote,clflushopt,clwb,cmov,cmpccxadd,crc32,cx16,cx8,enqcmd,f16c,fast-movbe,fma,fsgsbase,fxsr,gfni,hreset,invpcid,kl,lzcnt,mmx,movbe,movdir64b,movdiri,no-bypass-delay,nopl,pclmul,pconfig,pku,popcnt,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,serialize,sha,shstk,slow-incdec,slow-lea,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,uintr,use-glm-div-sqrt-costs,vaes,vpclmulqdq,vzeroupper,waitpkg,widekl,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("graniterapids"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,amx-bf16,amx-fp16,amx-int8,amx-tile,avx,avx2,avx512bf16,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512fp16,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,avxvnni,bmi,bmi2,cldemote,clflushopt,clwb,cmov,crc32,cx16,cx8,enqcmd,ermsb,evex512,f16c,false-deps-getmant,false-deps-mulc,false-deps-mullq,false-deps-perm,false-deps-range,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,movdir64b,movdiri,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pconfig,pku,popcnt,prefer-256-bit,prefetchi,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,serialize,sha,shstk,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tsxldtrk,tuning-fast-imm-vector-shift,uintr,vaes,vpclmulqdq,vzeroupper,waitpkg,wbnoinvd,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("graniterapids-d"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,amx-bf16,amx-complex,amx-fp16,amx-int8,amx-tile,avx,avx2,avx512bf16,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512fp16,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,avxvnni,bmi,bmi2,cldemote,clflushopt,clwb,cmov,crc32,cx16,cx8,enqcmd,ermsb,evex512,f16c,false-deps-getmant,false-deps-mulc,false-deps-mullq,false-deps-perm,false-deps-range,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,movdir64b,movdiri,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pconfig,pku,popcnt,prefer-256-bit,prefetchi,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,serialize,sha,shstk,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tsxldtrk,tuning-fast-imm-vector-shift,uintr,vaes,vpclmulqdq,vzeroupper,waitpkg,wbnoinvd,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("graniterapids_d"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,amx-bf16,amx-complex,amx-fp16,amx-int8,amx-tile,avx,avx2,avx512bf16,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512fp16,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,avxvnni,bmi,bmi2,cldemote,clflushopt,clwb,cmov,crc32,cx16,cx8,enqcmd,ermsb,evex512,f16c,false-deps-getmant,false-deps-mulc,false-deps-mullq,false-deps-perm,false-deps-range,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,movdir64b,movdiri,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pconfig,pku,popcnt,prefer-256-bit,prefetchi,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,serialize,sha,shstk,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tsxldtrk,tuning-fast-imm-vector-shift,uintr,vaes,vpclmulqdq,vzeroupper,waitpkg,wbnoinvd,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("haswell"), str_lit("32bit-mode,64bit,allow-light-256-bit,avx,avx2,bmi,bmi2,cmov,crc32,cx16,cx8,ermsb,f16c,false-deps-lzcnt-tzcnt,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,popcnt,rdrnd,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("i386"), str_lit("32bit-mode,slow-unaligned-mem-16,vzeroupper,x87") }, + { str_lit("i486"), str_lit("32bit-mode,slow-unaligned-mem-16,vzeroupper,x87") }, + { str_lit("i586"), str_lit("32bit-mode,cx8,slow-unaligned-mem-16,vzeroupper,x87") }, + { str_lit("i686"), str_lit("32bit-mode,cmov,cx8,slow-unaligned-mem-16,vzeroupper,x87") }, + { str_lit("icelake-client"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,avx,avx2,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,bmi,bmi2,clflushopt,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pku,popcnt,prefer-256-bit,prfchw,rdpid,rdrnd,rdseed,sahf,sha,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vaes,vpclmulqdq,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("icelake-server"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,avx,avx2,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,bmi,bmi2,clflushopt,clwb,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pconfig,pku,popcnt,prefer-256-bit,prfchw,rdpid,rdrnd,rdseed,sahf,sha,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vaes,vpclmulqdq,vzeroupper,wbnoinvd,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("icelake_client"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,avx,avx2,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,bmi,bmi2,clflushopt,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pku,popcnt,prefer-256-bit,prfchw,rdpid,rdrnd,rdseed,sahf,sha,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vaes,vpclmulqdq,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("icelake_server"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,avx,avx2,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,bmi,bmi2,clflushopt,clwb,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pconfig,pku,popcnt,prefer-256-bit,prfchw,rdpid,rdrnd,rdseed,sahf,sha,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vaes,vpclmulqdq,vzeroupper,wbnoinvd,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("ivybridge"), str_lit("32bit-mode,64bit,avx,cmov,crc32,cx16,cx8,f16c,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fsgsbase,fxsr,idivq-to-divl,macrofusion,mmx,no-bypass-delay-mov,nopl,pclmul,popcnt,rdrnd,sahf,slow-3ops-lea,slow-unaligned-mem-32,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("k6"), str_lit("32bit-mode,cx8,mmx,slow-unaligned-mem-16,vzeroupper,x87") }, + { str_lit("k6-2"), str_lit("32bit-mode,3dnow,cx8,mmx,slow-unaligned-mem-16,vzeroupper,x87") }, + { str_lit("k6-3"), str_lit("32bit-mode,3dnow,cx8,mmx,slow-unaligned-mem-16,vzeroupper,x87") }, + { str_lit("k8"), str_lit("32bit-mode,3dnow,3dnowa,64bit,cmov,cx8,fast-scalar-shift-masks,fxsr,mmx,nopl,sbb-dep-breaking,slow-shld,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("k8-sse3"), str_lit("32bit-mode,3dnow,3dnowa,64bit,cmov,cx16,cx8,fast-scalar-shift-masks,fxsr,mmx,nopl,sbb-dep-breaking,slow-shld,slow-unaligned-mem-16,sse,sse2,sse3,vzeroupper,x87") }, + { str_lit("knl"), str_lit("32bit-mode,64bit,adx,aes,avx,avx2,avx512cd,avx512er,avx512f,avx512pf,bmi,bmi2,cmov,crc32,cx16,cx8,evex512,f16c,fast-gather,fast-movbe,fma,fsgsbase,fxsr,idivq-to-divl,lzcnt,mmx,movbe,nopl,pclmul,popcnt,prefer-mask-registers,prefetchwt1,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,slow-incdec,slow-pmaddwd,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,x87,xsave,xsaveopt") }, + { str_lit("knm"), str_lit("32bit-mode,64bit,adx,aes,avx,avx2,avx512cd,avx512er,avx512f,avx512pf,avx512vpopcntdq,bmi,bmi2,cmov,crc32,cx16,cx8,evex512,f16c,fast-gather,fast-movbe,fma,fsgsbase,fxsr,idivq-to-divl,lzcnt,mmx,movbe,nopl,pclmul,popcnt,prefer-mask-registers,prefetchwt1,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,slow-incdec,slow-pmaddwd,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,x87,xsave,xsaveopt") }, + { str_lit("lakemont"), str_lit("32bit-mode,cx8,slow-unaligned-mem-16,vzeroupper") }, + { str_lit("meteorlake"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,avx,avx2,avxvnni,bmi,bmi2,cldemote,clflushopt,clwb,cmov,crc32,cx16,cx8,f16c,false-deps-perm,false-deps-popcnt,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fxsr,gfni,hreset,idivq-to-divl,invpcid,kl,lzcnt,macrofusion,mmx,movbe,movdir64b,movdiri,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pconfig,pku,popcnt,prefer-movmsk-over-vtest,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,serialize,sha,shstk,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vaes,vpclmulqdq,vzeroupper,waitpkg,widekl,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("mic_avx512"), str_lit("32bit-mode,64bit,adx,aes,avx,avx2,avx512cd,avx512er,avx512f,avx512pf,bmi,bmi2,cmov,crc32,cx16,cx8,evex512,f16c,fast-gather,fast-movbe,fma,fsgsbase,fxsr,idivq-to-divl,lzcnt,mmx,movbe,nopl,pclmul,popcnt,prefer-mask-registers,prefetchwt1,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,slow-incdec,slow-pmaddwd,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,x87,xsave,xsaveopt") }, + { str_lit("nehalem"), str_lit("32bit-mode,64bit,cmov,crc32,cx16,cx8,fxsr,macrofusion,mmx,no-bypass-delay-mov,nopl,popcnt,sahf,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87") }, + { str_lit("nocona"), str_lit("32bit-mode,64bit,cmov,cx16,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,sse3,vzeroupper,x87") }, + { str_lit("opteron"), str_lit("32bit-mode,3dnow,3dnowa,64bit,cmov,cx8,fast-scalar-shift-masks,fxsr,mmx,nopl,sbb-dep-breaking,slow-shld,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("opteron-sse3"), str_lit("32bit-mode,3dnow,3dnowa,64bit,cmov,cx16,cx8,fast-scalar-shift-masks,fxsr,mmx,nopl,sbb-dep-breaking,slow-shld,slow-unaligned-mem-16,sse,sse2,sse3,vzeroupper,x87") }, + { str_lit("penryn"), str_lit("32bit-mode,64bit,cmov,cx16,cx8,fxsr,macrofusion,mmx,nopl,sahf,slow-unaligned-mem-16,sse,sse2,sse3,sse4.1,ssse3,vzeroupper,x87") }, + { str_lit("pentium"), str_lit("32bit-mode,cx8,slow-unaligned-mem-16,vzeroupper,x87") }, + { str_lit("pentium-m"), str_lit("32bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("pentium-mmx"), str_lit("32bit-mode,cx8,mmx,slow-unaligned-mem-16,vzeroupper,x87") }, + { str_lit("pentium2"), str_lit("32bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,vzeroupper,x87") }, + { str_lit("pentium3"), str_lit("32bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,vzeroupper,x87") }, + { str_lit("pentium3m"), str_lit("32bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,vzeroupper,x87") }, + { str_lit("pentium4"), str_lit("32bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("pentium4m"), str_lit("32bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("pentium_4"), str_lit("32bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("pentium_4_sse3"), str_lit("32bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,sse3,vzeroupper,x87") }, + { str_lit("pentium_ii"), str_lit("32bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,vzeroupper,x87") }, + { str_lit("pentium_iii"), str_lit("32bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,vzeroupper,x87") }, + { str_lit("pentium_iii_no_xmm_regs"), str_lit("32bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,vzeroupper,x87") }, + { str_lit("pentium_m"), str_lit("32bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,vzeroupper,x87") }, + { str_lit("pentium_mmx"), str_lit("32bit-mode,cx8,mmx,slow-unaligned-mem-16,vzeroupper,x87") }, + { str_lit("pentium_pro"), str_lit("32bit-mode,cmov,cx8,nopl,slow-unaligned-mem-16,vzeroupper,x87") }, + { str_lit("pentiumpro"), str_lit("32bit-mode,cmov,cx8,nopl,slow-unaligned-mem-16,vzeroupper,x87") }, + { str_lit("prescott"), str_lit("32bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,sse3,vzeroupper,x87") }, + { str_lit("raptorlake"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,avx,avx2,avxvnni,bmi,bmi2,cldemote,clflushopt,clwb,cmov,crc32,cx16,cx8,f16c,false-deps-perm,false-deps-popcnt,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fxsr,gfni,hreset,idivq-to-divl,invpcid,kl,lzcnt,macrofusion,mmx,movbe,movdir64b,movdiri,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pconfig,pku,popcnt,prefer-movmsk-over-vtest,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,serialize,sha,shstk,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vaes,vpclmulqdq,vzeroupper,waitpkg,widekl,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("rocketlake"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,avx,avx2,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,bmi,bmi2,clflushopt,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pku,popcnt,prefer-256-bit,prfchw,rdpid,rdrnd,rdseed,sahf,sha,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vaes,vpclmulqdq,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("sandybridge"), str_lit("32bit-mode,64bit,avx,cmov,crc32,cx16,cx8,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fxsr,idivq-to-divl,macrofusion,mmx,no-bypass-delay-mov,nopl,pclmul,popcnt,sahf,slow-3ops-lea,slow-unaligned-mem-32,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsaveopt") }, + { str_lit("sapphirerapids"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,amx-bf16,amx-int8,amx-tile,avx,avx2,avx512bf16,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512fp16,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,avxvnni,bmi,bmi2,cldemote,clflushopt,clwb,cmov,crc32,cx16,cx8,enqcmd,ermsb,evex512,f16c,false-deps-getmant,false-deps-mulc,false-deps-mullq,false-deps-perm,false-deps-range,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,movdir64b,movdiri,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pconfig,pku,popcnt,prefer-256-bit,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,serialize,sha,shstk,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tsxldtrk,tuning-fast-imm-vector-shift,uintr,vaes,vpclmulqdq,vzeroupper,waitpkg,wbnoinvd,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("sierraforest"), str_lit("32bit-mode,64bit,adx,aes,avx,avx2,avxifma,avxneconvert,avxvnni,avxvnniint8,bmi,bmi2,cldemote,clflushopt,clwb,cmov,cmpccxadd,crc32,cx16,cx8,enqcmd,f16c,fast-movbe,fma,fsgsbase,fxsr,gfni,hreset,invpcid,kl,lzcnt,mmx,movbe,movdir64b,movdiri,no-bypass-delay,nopl,pclmul,pconfig,pku,popcnt,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,serialize,sha,shstk,slow-incdec,slow-lea,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,uintr,use-glm-div-sqrt-costs,vaes,vpclmulqdq,vzeroupper,waitpkg,widekl,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("silvermont"), str_lit("32bit-mode,64bit,cmov,crc32,cx16,cx8,false-deps-popcnt,fast-7bytenop,fast-movbe,fxsr,idivq-to-divl,mmx,movbe,no-bypass-delay,nopl,pclmul,popcnt,prfchw,rdrnd,sahf,slow-incdec,slow-lea,slow-pmulld,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,use-slm-arith-costs,vzeroupper,x87") }, + { str_lit("skx"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,avx,avx2,avx512bw,avx512cd,avx512dq,avx512f,avx512vl,bmi,bmi2,clflushopt,clwb,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,false-deps-popcnt,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,faster-shift-than-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pku,popcnt,prefer-256-bit,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("skylake"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,avx,avx2,bmi,bmi2,clflushopt,cmov,crc32,cx16,cx8,ermsb,f16c,false-deps-popcnt,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,popcnt,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("skylake-avx512"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,avx,avx2,avx512bw,avx512cd,avx512dq,avx512f,avx512vl,bmi,bmi2,clflushopt,clwb,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,false-deps-popcnt,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,faster-shift-than-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pku,popcnt,prefer-256-bit,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("skylake_avx512"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,avx,avx2,avx512bw,avx512cd,avx512dq,avx512f,avx512vl,bmi,bmi2,clflushopt,clwb,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,false-deps-popcnt,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,faster-shift-than-shuffle,fma,fsgsbase,fxsr,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pku,popcnt,prefer-256-bit,prfchw,rdrnd,rdseed,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("slm"), str_lit("32bit-mode,64bit,cmov,crc32,cx16,cx8,false-deps-popcnt,fast-7bytenop,fast-movbe,fxsr,idivq-to-divl,mmx,movbe,no-bypass-delay,nopl,pclmul,popcnt,prfchw,rdrnd,sahf,slow-incdec,slow-lea,slow-pmulld,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,use-slm-arith-costs,vzeroupper,x87") }, + { str_lit("tigerlake"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,avx,avx2,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vp2intersect,avx512vpopcntdq,bmi,bmi2,clflushopt,clwb,cmov,crc32,cx16,cx8,ermsb,evex512,f16c,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,idivq-to-divl,invpcid,lzcnt,macrofusion,mmx,movbe,movdir64b,movdiri,no-bypass-delay-blend,no-bypass-delay-mov,no-bypass-delay-shuffle,nopl,pclmul,pku,popcnt,prefer-256-bit,prfchw,rdpid,rdrnd,rdseed,sahf,sha,shstk,sse,sse2,sse3,sse4.1,sse4.2,ssse3,tuning-fast-imm-vector-shift,vaes,vpclmulqdq,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("tremont"), str_lit("32bit-mode,64bit,aes,clflushopt,clwb,cmov,crc32,cx16,cx8,fast-movbe,fsgsbase,fxsr,gfni,mmx,movbe,no-bypass-delay,nopl,pclmul,popcnt,prfchw,ptwrite,rdpid,rdrnd,rdseed,sahf,sha,slow-incdec,slow-lea,slow-two-mem-ops,sse,sse2,sse3,sse4.1,sse4.2,ssse3,use-glm-div-sqrt-costs,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("westmere"), str_lit("32bit-mode,64bit,cmov,crc32,cx16,cx8,fxsr,macrofusion,mmx,no-bypass-delay-mov,nopl,pclmul,popcnt,sahf,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87") }, + { str_lit("winchip-c6"), str_lit("32bit-mode,mmx,slow-unaligned-mem-16,vzeroupper,x87") }, + { str_lit("winchip2"), str_lit("32bit-mode,3dnow,mmx,slow-unaligned-mem-16,vzeroupper,x87") }, + { str_lit("x86-64"), str_lit("32bit-mode,64bit,cmov,cx8,fxsr,idivq-to-divl,macrofusion,mmx,nopl,slow-3ops-lea,slow-incdec,sse,sse2,vzeroupper,x87") }, + { str_lit("x86-64-v2"), str_lit("32bit-mode,64bit,cmov,crc32,cx16,cx8,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fxsr,idivq-to-divl,macrofusion,mmx,nopl,popcnt,sahf,slow-3ops-lea,slow-unaligned-mem-32,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87") }, + { str_lit("x86-64-v3"), str_lit("32bit-mode,64bit,allow-light-256-bit,avx,avx2,bmi,bmi2,cmov,crc32,cx16,cx8,f16c,false-deps-lzcnt-tzcnt,false-deps-popcnt,fast-15bytenop,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fma,fxsr,idivq-to-divl,lzcnt,macrofusion,mmx,movbe,nopl,popcnt,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave") }, + { str_lit("x86-64-v4"), str_lit("32bit-mode,64bit,allow-light-256-bit,avx,avx2,avx512bw,avx512cd,avx512dq,avx512f,avx512vl,bmi,bmi2,cmov,crc32,cx16,cx8,evex512,f16c,false-deps-popcnt,fast-15bytenop,fast-gather,fast-scalar-fsqrt,fast-shld-rotate,fast-variable-crosslane-shuffle,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fxsr,idivq-to-divl,lzcnt,macrofusion,mmx,movbe,nopl,popcnt,prefer-256-bit,sahf,slow-3ops-lea,sse,sse2,sse3,sse4.1,sse4.2,ssse3,vzeroupper,x87,xsave") }, + { str_lit("yonah"), str_lit("32bit-mode,cmov,cx8,fxsr,mmx,nopl,slow-unaligned-mem-16,sse,sse2,sse3,vzeroupper,x87") }, + { str_lit("znver1"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,avx,avx2,bmi,bmi2,branchfusion,clflushopt,clzero,cmov,crc32,cx16,cx8,f16c,fast-15bytenop,fast-bextr,fast-lzcnt,fast-movbe,fast-scalar-fsqrt,fast-scalar-shift-masks,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fxsr,lzcnt,mmx,movbe,mwaitx,nopl,pclmul,popcnt,prfchw,rdrnd,rdseed,sahf,sbb-dep-breaking,sha,slow-shld,sse,sse2,sse3,sse4.1,sse4.2,sse4a,ssse3,vzeroupper,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("znver2"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,avx,avx2,bmi,bmi2,branchfusion,clflushopt,clwb,clzero,cmov,crc32,cx16,cx8,f16c,fast-15bytenop,fast-bextr,fast-lzcnt,fast-movbe,fast-scalar-fsqrt,fast-scalar-shift-masks,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fxsr,lzcnt,mmx,movbe,mwaitx,nopl,pclmul,popcnt,prfchw,rdpid,rdpru,rdrnd,rdseed,sahf,sbb-dep-breaking,sha,slow-shld,sse,sse2,sse3,sse4.1,sse4.2,sse4a,ssse3,vzeroupper,wbnoinvd,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("znver3"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,avx,avx2,bmi,bmi2,branchfusion,clflushopt,clwb,clzero,cmov,crc32,cx16,cx8,f16c,fast-15bytenop,fast-bextr,fast-lzcnt,fast-movbe,fast-scalar-fsqrt,fast-scalar-shift-masks,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,invpcid,lzcnt,macrofusion,mmx,movbe,mwaitx,nopl,pclmul,pku,popcnt,prfchw,rdpid,rdpru,rdrnd,rdseed,sahf,sbb-dep-breaking,sha,slow-shld,sse,sse2,sse3,sse4.1,sse4.2,sse4a,ssse3,vaes,vpclmulqdq,vzeroupper,wbnoinvd,x87,xsave,xsavec,xsaveopt,xsaves") }, + { str_lit("znver4"), str_lit("32bit-mode,64bit,adx,aes,allow-light-256-bit,avx,avx2,avx512bf16,avx512bitalg,avx512bw,avx512cd,avx512dq,avx512f,avx512ifma,avx512vbmi,avx512vbmi2,avx512vl,avx512vnni,avx512vpopcntdq,bmi,bmi2,branchfusion,clflushopt,clwb,clzero,cmov,crc32,cx16,cx8,evex512,f16c,fast-15bytenop,fast-bextr,fast-lzcnt,fast-movbe,fast-scalar-fsqrt,fast-scalar-shift-masks,fast-variable-perlane-shuffle,fast-vector-fsqrt,fma,fsgsbase,fsrm,fxsr,gfni,invpcid,lzcnt,macrofusion,mmx,movbe,mwaitx,nopl,pclmul,pku,popcnt,prfchw,rdpid,rdpru,rdrnd,rdseed,sahf,sbb-dep-breaking,sha,shstk,slow-shld,sse,sse2,sse3,sse4.1,sse4.2,sse4a,ssse3,vaes,vpclmulqdq,vzeroupper,wbnoinvd,x87,xsave,xsavec,xsaveopt,xsaves") }, + // TargetArch_arm32: + { str_lit("arm1020e"), str_lit("armv5te,v4t,v5t,v5te") }, + { str_lit("arm1020t"), str_lit("armv5t,v4t,v5t") }, + { str_lit("arm1022e"), str_lit("armv5te,v4t,v5t,v5te") }, + { str_lit("arm10e"), str_lit("armv5te,v4t,v5t,v5te") }, + { str_lit("arm10tdmi"), str_lit("armv5t,v4t,v5t") }, + { str_lit("arm1136j-s"), str_lit("armv6,dsp,v4t,v5t,v5te,v6") }, + { str_lit("arm1136jf-s"), str_lit("armv6,dsp,fp64,fpregs,fpregs64,slowfpvmlx,v4t,v5t,v5te,v6,vfp2,vfp2sp") }, + { str_lit("arm1156t2-s"), str_lit("armv6t2,dsp,thumb2,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v8m") }, + { str_lit("arm1156t2f-s"), str_lit("armv6t2,dsp,fp64,fpregs,fpregs64,slowfpvmlx,thumb2,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v8m,vfp2,vfp2sp") }, + { str_lit("arm1176jz-s"), str_lit("armv6kz,trustzone,v4t,v5t,v5te,v6,v6k") }, + { str_lit("arm1176jzf-s"), str_lit("armv6kz,fp64,fpregs,fpregs64,slowfpvmlx,trustzone,v4t,v5t,v5te,v6,v6k,vfp2,vfp2sp") }, + { str_lit("arm710t"), str_lit("armv4t,v4t") }, + { str_lit("arm720t"), str_lit("armv4t,v4t") }, + { str_lit("arm7tdmi"), str_lit("armv4t,v4t") }, + { str_lit("arm7tdmi-s"), str_lit("armv4t,v4t") }, + { str_lit("arm8"), str_lit("armv4") }, + { str_lit("arm810"), str_lit("armv4") }, + { str_lit("arm9"), str_lit("armv4t,v4t") }, + { str_lit("arm920"), str_lit("armv4t,v4t") }, + { str_lit("arm920t"), str_lit("armv4t,v4t") }, + { str_lit("arm922t"), str_lit("armv4t,v4t") }, + { str_lit("arm926ej-s"), str_lit("armv5te,v4t,v5t,v5te") }, + { str_lit("arm940t"), str_lit("armv4t,v4t") }, + { str_lit("arm946e-s"), str_lit("armv5te,v4t,v5t,v5te") }, + { str_lit("arm966e-s"), str_lit("armv5te,v4t,v5t,v5te") }, + { str_lit("arm968e-s"), str_lit("armv5te,v4t,v5t,v5te") }, + { str_lit("arm9e"), str_lit("armv5te,v4t,v5t,v5te") }, + { str_lit("arm9tdmi"), str_lit("armv4t,v4t") }, + { str_lit("cortex-a12"), str_lit("a12,aclass,armv7-a,avoid-partial-cpsr,d32,db,dsp,fp16,fp64,fpregs,fpregs64,hwdiv,hwdiv-arm,mp,neon,perfmon,ret-addr-stack,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization,vmlx-forwarding") }, + { str_lit("cortex-a15"), str_lit("a15,aclass,armv7-a,avoid-partial-cpsr,d32,db,dont-widen-vmovs,dsp,fp16,fp64,fpregs,fpregs64,hwdiv,hwdiv-arm,mp,muxed-units,neon,perfmon,ret-addr-stack,splat-vfp-neon,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization,vldn-align") }, + { str_lit("cortex-a17"), str_lit("a17,aclass,armv7-a,avoid-partial-cpsr,d32,db,dsp,fp16,fp64,fpregs,fpregs64,hwdiv,hwdiv-arm,mp,neon,perfmon,ret-addr-stack,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization,vmlx-forwarding") }, + { str_lit("cortex-a32"), str_lit("aclass,acquire-release,aes,armv8-a,crc,crypto,d32,db,dsp,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs64,hwdiv,hwdiv-arm,mp,neon,perfmon,sha2,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("cortex-a35"), str_lit("a35,aclass,acquire-release,aes,armv8-a,crc,crypto,d32,db,dsp,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs64,hwdiv,hwdiv-arm,mp,neon,perfmon,sha2,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("cortex-a5"), str_lit("a5,aclass,armv7-a,d32,db,dsp,fp16,fp64,fpregs,fpregs64,mp,neon,perfmon,ret-addr-stack,slow-fp-brcc,slowfpvfmx,slowfpvmlx,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,vmlx-forwarding") }, + { str_lit("cortex-a53"), str_lit("a53,aclass,acquire-release,aes,armv8-a,crc,crypto,d32,db,dsp,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpao,fpregs,fpregs64,hwdiv,hwdiv-arm,mp,neon,perfmon,sha2,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("cortex-a55"), str_lit("a55,aclass,acquire-release,aes,armv8.2-a,crc,crypto,d32,db,dotprod,dsp,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs64,hwdiv,hwdiv-arm,mp,neon,perfmon,ras,sha2,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8.1a,v8.2a,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("cortex-a57"), str_lit("a57,aclass,acquire-release,aes,armv8-a,avoid-partial-cpsr,cheap-predicable-cpsr,crc,crypto,d32,db,dsp,fix-cortex-a57-aes-1742098,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpao,fpregs,fpregs64,hwdiv,hwdiv-arm,mp,neon,perfmon,sha2,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("cortex-a7"), str_lit("a7,aclass,armv7-a,d32,db,dsp,fp16,fp64,fpregs,fpregs64,hwdiv,hwdiv-arm,mp,neon,perfmon,ret-addr-stack,slow-fp-brcc,slowfpvfmx,slowfpvmlx,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization,vmlx-forwarding,vmlx-hazards") }, + { str_lit("cortex-a710"), str_lit("aclass,acquire-release,armv9-a,bf16,cortex-a710,crc,d32,db,dotprod,dsp,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp16fml,fp64,fpregs,fpregs16,fpregs64,fullfp16,hwdiv,hwdiv-arm,i8mm,mp,neon,perfmon,ras,sb,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8.1a,v8.2a,v8.3a,v8.4a,v8.5a,v8m,v9a,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("cortex-a72"), str_lit("a72,aclass,acquire-release,aes,armv8-a,crc,crypto,d32,db,dsp,fix-cortex-a57-aes-1742098,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs64,hwdiv,hwdiv-arm,mp,neon,perfmon,sha2,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("cortex-a73"), str_lit("a73,aclass,acquire-release,aes,armv8-a,crc,crypto,d32,db,dsp,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs64,hwdiv,hwdiv-arm,mp,neon,perfmon,sha2,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("cortex-a75"), str_lit("a75,aclass,acquire-release,aes,armv8.2-a,crc,crypto,d32,db,dotprod,dsp,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs64,hwdiv,hwdiv-arm,mp,neon,perfmon,ras,sha2,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8.1a,v8.2a,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("cortex-a76"), str_lit("a76,aclass,acquire-release,aes,armv8.2-a,crc,crypto,d32,db,dotprod,dsp,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs16,fpregs64,fullfp16,hwdiv,hwdiv-arm,mp,neon,perfmon,ras,sha2,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8.1a,v8.2a,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("cortex-a76ae"), str_lit("a76,aclass,acquire-release,aes,armv8.2-a,crc,crypto,d32,db,dotprod,dsp,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs16,fpregs64,fullfp16,hwdiv,hwdiv-arm,mp,neon,perfmon,ras,sha2,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8.1a,v8.2a,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("cortex-a77"), str_lit("a77,aclass,acquire-release,aes,armv8.2-a,crc,crypto,d32,db,dotprod,dsp,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs16,fpregs64,fullfp16,hwdiv,hwdiv-arm,mp,neon,perfmon,ras,sha2,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8.1a,v8.2a,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("cortex-a78"), str_lit("aclass,acquire-release,aes,armv8.2-a,cortex-a78,crc,crypto,d32,db,dotprod,dsp,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs16,fpregs64,fullfp16,hwdiv,hwdiv-arm,mp,neon,perfmon,ras,sha2,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8.1a,v8.2a,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("cortex-a78c"), str_lit("a78c,aclass,acquire-release,aes,armv8.2-a,crc,crypto,d32,db,dotprod,dsp,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs16,fpregs64,fullfp16,hwdiv,hwdiv-arm,mp,neon,perfmon,ras,sha2,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8.1a,v8.2a,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("cortex-a8"), str_lit("a8,aclass,armv7-a,d32,db,dsp,fp64,fpregs,fpregs64,neon,nonpipelined-vfp,perfmon,ret-addr-stack,slow-fp-brcc,slowfpvfmx,slowfpvmlx,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vmlx-forwarding,vmlx-hazards") }, + { str_lit("cortex-a9"), str_lit("a9,aclass,armv7-a,avoid-partial-cpsr,d32,db,dsp,expand-fp-mlx,fp16,fp64,fpregs,fpregs64,mp,muxed-units,neon,neon-fpmovs,perfmon,prefer-vmovsr,ret-addr-stack,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vldn-align,vmlx-forwarding,vmlx-hazards") }, + { str_lit("cortex-m0"), str_lit("armv6-m,db,mclass,no-branch-predictor,noarm,strict-align,thumb-mode,v4t,v5t,v5te,v6,v6m") }, + { str_lit("cortex-m0plus"), str_lit("armv6-m,db,mclass,no-branch-predictor,noarm,strict-align,thumb-mode,v4t,v5t,v5te,v6,v6m") }, + { str_lit("cortex-m1"), str_lit("armv6-m,db,mclass,no-branch-predictor,noarm,strict-align,thumb-mode,v4t,v5t,v5te,v6,v6m") }, + { str_lit("cortex-m23"), str_lit("8msecext,acquire-release,armv8-m.base,db,hwdiv,mclass,no-branch-predictor,no-movt,noarm,strict-align,thumb-mode,v4t,v5t,v5te,v6,v6m,v7clrex,v8m") }, + { str_lit("cortex-m3"), str_lit("armv7-m,db,hwdiv,loop-align,m3,mclass,no-branch-predictor,noarm,thumb-mode,thumb2,use-misched,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8m") }, + { str_lit("cortex-m33"), str_lit("8msecext,acquire-release,armv8-m.main,db,dsp,fix-cmse-cve-2021-35465,fp-armv8d16sp,fp16,fpregs,hwdiv,loop-align,mclass,no-branch-predictor,noarm,slowfpvfmx,slowfpvmlx,thumb-mode,thumb2,use-misched,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8m,v8m.main,vfp2sp,vfp3d16sp,vfp4d16sp") }, + { str_lit("cortex-m35p"), str_lit("8msecext,acquire-release,armv8-m.main,db,dsp,fix-cmse-cve-2021-35465,fp-armv8d16sp,fp16,fpregs,hwdiv,loop-align,mclass,no-branch-predictor,noarm,slowfpvfmx,slowfpvmlx,thumb-mode,thumb2,use-misched,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8m,v8m.main,vfp2sp,vfp3d16sp,vfp4d16sp") }, + { str_lit("cortex-m4"), str_lit("armv7e-m,db,dsp,fp16,fpregs,hwdiv,loop-align,mclass,no-branch-predictor,noarm,slowfpvfmx,slowfpvmlx,thumb-mode,thumb2,use-misched,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8m,vfp2sp,vfp3d16sp,vfp4d16sp") }, + { str_lit("cortex-m55"), str_lit("8msecext,acquire-release,armv8.1-m.main,db,dsp,fix-cmse-cve-2021-35465,fp-armv8d16,fp-armv8d16sp,fp16,fp64,fpregs,fpregs16,fpregs64,fullfp16,hwdiv,lob,loop-align,mclass,mve,mve.fp,no-branch-predictor,noarm,ras,slowfpvmlx,thumb-mode,thumb2,use-misched,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8.1m.main,v8m,v8m.main,vfp2,vfp2sp,vfp3d16,vfp3d16sp,vfp4d16,vfp4d16sp") }, + { str_lit("cortex-m7"), str_lit("armv7e-m,db,dsp,fp-armv8d16,fp-armv8d16sp,fp16,fp64,fpregs,fpregs64,hwdiv,m7,mclass,noarm,thumb-mode,thumb2,use-mipipeliner,use-misched,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8m,vfp2,vfp2sp,vfp3d16,vfp3d16sp,vfp4d16,vfp4d16sp") }, + { str_lit("cortex-m85"), str_lit("8msecext,acquire-release,armv8.1-m.main,db,dsp,fp-armv8d16,fp-armv8d16sp,fp16,fp64,fpregs,fpregs16,fpregs64,fullfp16,hwdiv,lob,mclass,mve,mve.fp,noarm,pacbti,ras,thumb-mode,thumb2,use-misched,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8.1m.main,v8m,v8m.main,vfp2,vfp2sp,vfp3d16,vfp3d16sp,vfp4d16,vfp4d16sp") }, + { str_lit("cortex-r4"), str_lit("armv7-r,avoid-partial-cpsr,db,dsp,hwdiv,perfmon,r4,rclass,ret-addr-stack,thumb2,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8m") }, + { str_lit("cortex-r4f"), str_lit("armv7-r,avoid-partial-cpsr,db,dsp,fp64,fpregs,fpregs64,hwdiv,perfmon,r4,rclass,ret-addr-stack,slow-fp-brcc,slowfpvfmx,slowfpvmlx,thumb2,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8m,vfp2,vfp2sp,vfp3d16,vfp3d16sp") }, + { str_lit("cortex-r5"), str_lit("armv7-r,avoid-partial-cpsr,db,dsp,fp64,fpregs,fpregs64,hwdiv,hwdiv-arm,perfmon,r5,rclass,ret-addr-stack,slow-fp-brcc,slowfpvfmx,slowfpvmlx,thumb2,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8m,vfp2,vfp2sp,vfp3d16,vfp3d16sp") }, + { str_lit("cortex-r52"), str_lit("acquire-release,armv8-r,crc,d32,db,dfb,dsp,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpao,fpregs,fpregs64,hwdiv,hwdiv-arm,mp,neon,perfmon,r52,rclass,thumb2,use-misched,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("cortex-r7"), str_lit("armv7-r,avoid-partial-cpsr,db,dsp,fp16,fp64,fpregs,fpregs64,hwdiv,hwdiv-arm,mp,perfmon,r7,rclass,ret-addr-stack,slow-fp-brcc,slowfpvfmx,slowfpvmlx,thumb2,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8m,vfp2,vfp2sp,vfp3d16,vfp3d16sp") }, + { str_lit("cortex-r8"), str_lit("armv7-r,avoid-partial-cpsr,db,dsp,fp16,fp64,fpregs,fpregs64,hwdiv,hwdiv-arm,mp,perfmon,rclass,ret-addr-stack,slow-fp-brcc,slowfpvfmx,slowfpvmlx,thumb2,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8m,vfp2,vfp2sp,vfp3d16,vfp3d16sp") }, + { str_lit("cortex-x1"), str_lit("aclass,acquire-release,aes,armv8.2-a,cortex-x1,crc,crypto,d32,db,dotprod,dsp,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs16,fpregs64,fullfp16,hwdiv,hwdiv-arm,mp,neon,perfmon,ras,sha2,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8.1a,v8.2a,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("cortex-x1c"), str_lit("aclass,acquire-release,aes,armv8.2-a,cortex-x1c,crc,crypto,d32,db,dotprod,dsp,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs16,fpregs64,fullfp16,hwdiv,hwdiv-arm,mp,neon,perfmon,ras,sha2,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8.1a,v8.2a,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("cyclone"), str_lit("aclass,acquire-release,aes,armv8-a,avoid-movs-shop,avoid-partial-cpsr,crc,crypto,d32,db,disable-postra-scheduler,dsp,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs64,hwdiv,hwdiv-arm,mp,neon,neonfp,perfmon,ret-addr-stack,sha2,slowfpvfmx,slowfpvmlx,swift,thumb2,trustzone,use-misched,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization,zcz") }, + { str_lit("ep9312"), str_lit("armv4t,v4t") }, + { str_lit("exynos-m3"), str_lit("aclass,acquire-release,aes,armv8-a,crc,crypto,d32,db,dont-widen-vmovs,dsp,expand-fp-mlx,exynos,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs64,fuse-aes,fuse-literals,hwdiv,hwdiv-arm,mp,neon,perfmon,prof-unpr,ret-addr-stack,sha2,slow-fp-brcc,slow-vdup32,slow-vgetlni32,slowfpvfmx,slowfpvmlx,splat-vfp-neon,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization,wide-stride-vfp,zcz") }, + { str_lit("exynos-m4"), str_lit("aclass,acquire-release,aes,armv8.2-a,crc,crypto,d32,db,dont-widen-vmovs,dotprod,dsp,expand-fp-mlx,exynos,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs16,fpregs64,fullfp16,fuse-aes,fuse-literals,hwdiv,hwdiv-arm,mp,neon,perfmon,prof-unpr,ras,ret-addr-stack,sha2,slow-fp-brcc,slow-vdup32,slow-vgetlni32,slowfpvfmx,slowfpvmlx,splat-vfp-neon,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8.1a,v8.2a,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization,wide-stride-vfp,zcz") }, + { str_lit("exynos-m5"), str_lit("aclass,acquire-release,aes,armv8.2-a,crc,crypto,d32,db,dont-widen-vmovs,dotprod,dsp,expand-fp-mlx,exynos,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs16,fpregs64,fullfp16,fuse-aes,fuse-literals,hwdiv,hwdiv-arm,mp,neon,perfmon,prof-unpr,ras,ret-addr-stack,sha2,slow-fp-brcc,slow-vdup32,slow-vgetlni32,slowfpvfmx,slowfpvmlx,splat-vfp-neon,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8.1a,v8.2a,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization,wide-stride-vfp,zcz") }, + { str_lit("generic"), str_lit("") }, + { str_lit("iwmmxt"), str_lit("armv5te,v4t,v5t,v5te") }, + { str_lit("krait"), str_lit("aclass,armv7-a,avoid-partial-cpsr,d32,db,dsp,fp16,fp64,fpregs,fpregs64,hwdiv,hwdiv-arm,krait,muxed-units,neon,perfmon,ret-addr-stack,thumb2,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,vldn-align,vmlx-forwarding") }, + { str_lit("kryo"), str_lit("aclass,acquire-release,aes,armv8-a,crc,crypto,d32,db,dsp,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs64,hwdiv,hwdiv-arm,kryo,mp,neon,perfmon,sha2,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("mpcore"), str_lit("armv6k,fp64,fpregs,fpregs64,slowfpvmlx,v4t,v5t,v5te,v6,v6k,vfp2,vfp2sp") }, + { str_lit("mpcorenovfp"), str_lit("armv6k,v4t,v5t,v5te,v6,v6k") }, + { str_lit("neoverse-n1"), str_lit("aclass,acquire-release,aes,armv8.2-a,crc,crypto,d32,db,dotprod,dsp,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs64,hwdiv,hwdiv-arm,mp,neon,perfmon,ras,sha2,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8.1a,v8.2a,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("neoverse-n2"), str_lit("aclass,acquire-release,armv9-a,bf16,crc,d32,db,dotprod,dsp,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs64,hwdiv,hwdiv-arm,i8mm,mp,neon,perfmon,ras,sb,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8.1a,v8.2a,v8.3a,v8.4a,v8.5a,v8m,v9a,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("neoverse-v1"), str_lit("aclass,acquire-release,aes,armv8.4-a,bf16,crc,crypto,d32,db,dotprod,dsp,fp-armv8,fp-armv8d16,fp-armv8d16sp,fp-armv8sp,fp16,fp64,fpregs,fpregs16,fpregs64,fullfp16,hwdiv,hwdiv-arm,i8mm,mp,neon,perfmon,ras,sha2,thumb2,trustzone,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8,v8.1a,v8.2a,v8.3a,v8.4a,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,virtualization") }, + { str_lit("sc000"), str_lit("armv6-m,db,mclass,no-branch-predictor,noarm,strict-align,thumb-mode,v4t,v5t,v5te,v6,v6m") }, + { str_lit("sc300"), str_lit("armv7-m,db,hwdiv,m3,mclass,no-branch-predictor,noarm,thumb-mode,thumb2,use-misched,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8m") }, + { str_lit("strongarm"), str_lit("armv4") }, + { str_lit("strongarm110"), str_lit("armv4") }, + { str_lit("strongarm1100"), str_lit("armv4") }, + { str_lit("strongarm1110"), str_lit("armv4") }, + { str_lit("swift"), str_lit("aclass,armv7-a,avoid-movs-shop,avoid-partial-cpsr,d32,db,disable-postra-scheduler,dsp,fp16,fp64,fpregs,fpregs64,hwdiv,hwdiv-arm,mp,neon,neonfp,perfmon,prefer-ishst,prof-unpr,ret-addr-stack,slow-load-D-subreg,slow-odd-reg,slow-vdup32,slow-vgetlni32,slowfpvfmx,slowfpvmlx,swift,thumb2,use-misched,v4t,v5t,v5te,v6,v6k,v6m,v6t2,v7,v7clrex,v8m,vfp2,vfp2sp,vfp3,vfp3d16,vfp3d16sp,vfp3sp,vfp4,vfp4d16,vfp4d16sp,vfp4sp,vmlx-hazards,wide-stride-vfp") }, + { str_lit("xscale"), str_lit("armv5te,v4t,v5t,v5te") }, + // TargetArch_arm64: + { str_lit("a64fx"), str_lit("CONTEXTIDREL2,a64fx,aggressive-fma,arith-bcc-fusion,ccpp,complxnum,crc,el2vmsa,el3,fp-armv8,fullfp16,lor,lse,neon,pan,pan-rwv,perfmon,predictable-select-expensive,ras,rdm,sha2,store-pair-suppress,sve,uaops,use-postra-scheduler,v8.1a,v8.2a,v8a,vh") }, + { str_lit("ampere1"), str_lit("CONTEXTIDREL2,addr-lsl-fast,aes,aggressive-fma,altnzcv,alu-lsl-fast,am,ampere1,amvs,arith-bcc-fusion,bf16,bti,ccdp,ccidx,ccpp,cmp-bcc-fusion,complxnum,crc,dit,dotprod,ecv,el2vmsa,el3,fgt,flagm,fp-armv8,fptoint,fuse-address,fuse-aes,fuse-literals,i8mm,jsconv,ldp-aligned-only,lor,lse,lse2,mpam,neon,nv,pan,pan-rwv,pauth,perfmon,predres,rand,ras,rcpc,rcpc-immo,rdm,sb,sel2,sha2,sha3,specrestrict,ssbs,store-pair-suppress,stp-aligned-only,tlb-rmi,tracev8.4,uaops,use-postra-scheduler,v8.1a,v8.2a,v8.3a,v8.4a,v8.5a,v8.6a,v8a,vh") }, + { str_lit("ampere1a"), str_lit("CONTEXTIDREL2,addr-lsl-fast,aes,aggressive-fma,altnzcv,alu-lsl-fast,am,ampere1a,amvs,arith-bcc-fusion,bf16,bti,ccdp,ccidx,ccpp,cmp-bcc-fusion,complxnum,crc,dit,dotprod,ecv,el2vmsa,el3,fgt,flagm,fp-armv8,fptoint,fuse-address,fuse-aes,fuse-literals,i8mm,jsconv,ldp-aligned-only,lor,lse,lse2,mpam,mte,neon,nv,pan,pan-rwv,pauth,perfmon,predres,rand,ras,rcpc,rcpc-immo,rdm,sb,sel2,sha2,sha3,sm4,specrestrict,ssbs,store-pair-suppress,stp-aligned-only,tlb-rmi,tracev8.4,uaops,use-postra-scheduler,v8.1a,v8.2a,v8.3a,v8.4a,v8.5a,v8.6a,v8a,vh") }, + { str_lit("apple-a10"), str_lit("CONTEXTIDREL2,aes,alternate-sextload-cvt-f32-pattern,apple-a10,arith-bcc-fusion,arith-cbz-fusion,crc,crypto,disable-latency-sched-heuristic,el2vmsa,el3,fp-armv8,fuse-aes,fuse-crypto-eor,lor,neon,pan,perfmon,rdm,sha2,store-pair-suppress,v8a,vh,zcm,zcz,zcz-gp") }, + { str_lit("apple-a11"), str_lit("CONTEXTIDREL2,aes,alternate-sextload-cvt-f32-pattern,apple-a11,arith-bcc-fusion,arith-cbz-fusion,ccpp,crc,crypto,disable-latency-sched-heuristic,el2vmsa,el3,fp-armv8,fullfp16,fuse-aes,fuse-crypto-eor,lor,lse,neon,pan,pan-rwv,perfmon,ras,rdm,sha2,store-pair-suppress,uaops,v8.1a,v8.2a,v8a,vh,zcm,zcz,zcz-gp") }, + { str_lit("apple-a12"), str_lit("CONTEXTIDREL2,aes,alternate-sextload-cvt-f32-pattern,apple-a12,arith-bcc-fusion,arith-cbz-fusion,ccidx,ccpp,complxnum,crc,crypto,disable-latency-sched-heuristic,el2vmsa,el3,fp-armv8,fullfp16,fuse-aes,fuse-crypto-eor,jsconv,lor,lse,neon,pan,pan-rwv,pauth,perfmon,ras,rcpc,rdm,sha2,store-pair-suppress,uaops,v8.1a,v8.2a,v8.3a,v8a,vh,zcm,zcz,zcz-gp") }, + { str_lit("apple-a13"), str_lit("CONTEXTIDREL2,aes,alternate-sextload-cvt-f32-pattern,am,apple-a13,arith-bcc-fusion,arith-cbz-fusion,ccidx,ccpp,complxnum,crc,crypto,disable-latency-sched-heuristic,dit,dotprod,el2vmsa,el3,flagm,fp-armv8,fp16fml,fullfp16,fuse-aes,fuse-crypto-eor,jsconv,lor,lse,lse2,mpam,neon,nv,pan,pan-rwv,pauth,perfmon,ras,rcpc,rcpc-immo,rdm,sel2,sha2,sha3,store-pair-suppress,tlb-rmi,tracev8.4,uaops,v8.1a,v8.2a,v8.3a,v8.4a,v8a,vh,zcm,zcz,zcz-gp") }, + { str_lit("apple-a14"), str_lit("CONTEXTIDREL2,aes,aggressive-fma,alternate-sextload-cvt-f32-pattern,altnzcv,am,apple-a14,arith-bcc-fusion,arith-cbz-fusion,ccdp,ccidx,ccpp,complxnum,crc,crypto,disable-latency-sched-heuristic,dit,dotprod,el2vmsa,el3,flagm,fp-armv8,fp16fml,fptoint,fullfp16,fuse-address,fuse-adrp-add,fuse-aes,fuse-arith-logic,fuse-crypto-eor,fuse-csel,fuse-literals,jsconv,lor,lse,lse2,mpam,neon,nv,pan,pan-rwv,pauth,perfmon,predres,ras,rcpc,rcpc-immo,rdm,sb,sel2,sha2,sha3,specrestrict,ssbs,store-pair-suppress,tlb-rmi,tracev8.4,uaops,v8.1a,v8.2a,v8.3a,v8.4a,v8a,vh,zcm,zcz,zcz-gp") }, + { str_lit("apple-a15"), str_lit("CONTEXTIDREL2,aes,alternate-sextload-cvt-f32-pattern,altnzcv,am,amvs,apple-a15,arith-bcc-fusion,arith-cbz-fusion,bf16,bti,ccdp,ccidx,ccpp,complxnum,crc,crypto,disable-latency-sched-heuristic,dit,dotprod,ecv,el2vmsa,el3,fgt,flagm,fp-armv8,fp16fml,fptoint,fullfp16,fuse-address,fuse-aes,fuse-arith-logic,fuse-crypto-eor,fuse-csel,fuse-literals,i8mm,jsconv,lor,lse,lse2,mpam,neon,nv,pan,pan-rwv,pauth,perfmon,predres,ras,rcpc,rcpc-immo,rdm,sb,sel2,sha2,sha3,specrestrict,ssbs,store-pair-suppress,tlb-rmi,tracev8.4,uaops,v8.1a,v8.2a,v8.3a,v8.4a,v8.5a,v8.6a,v8a,vh,zcm,zcz,zcz-gp") }, + { str_lit("apple-a16"), str_lit("CONTEXTIDREL2,aes,alternate-sextload-cvt-f32-pattern,altnzcv,am,amvs,apple-a16,arith-bcc-fusion,arith-cbz-fusion,bf16,bti,ccdp,ccidx,ccpp,complxnum,crc,crypto,disable-latency-sched-heuristic,dit,dotprod,ecv,el2vmsa,el3,fgt,flagm,fp-armv8,fp16fml,fptoint,fullfp16,fuse-address,fuse-aes,fuse-arith-logic,fuse-crypto-eor,fuse-csel,fuse-literals,hcx,i8mm,jsconv,lor,lse,lse2,mpam,neon,nv,pan,pan-rwv,pauth,perfmon,predres,ras,rcpc,rcpc-immo,rdm,sb,sel2,sha2,sha3,specrestrict,ssbs,store-pair-suppress,tlb-rmi,tracev8.4,uaops,v8.1a,v8.2a,v8.3a,v8.4a,v8.5a,v8.6a,v8a,vh,zcm,zcz,zcz-gp") }, + { str_lit("apple-a7"), str_lit("aes,alternate-sextload-cvt-f32-pattern,apple-a7,apple-a7-sysreg,arith-bcc-fusion,arith-cbz-fusion,crypto,disable-latency-sched-heuristic,el2vmsa,el3,fp-armv8,fuse-aes,fuse-crypto-eor,neon,perfmon,sha2,store-pair-suppress,v8a,zcm,zcz,zcz-fp-workaround,zcz-gp") }, + { str_lit("apple-a8"), str_lit("aes,alternate-sextload-cvt-f32-pattern,apple-a7,apple-a7-sysreg,arith-bcc-fusion,arith-cbz-fusion,crypto,disable-latency-sched-heuristic,el2vmsa,el3,fp-armv8,fuse-aes,fuse-crypto-eor,neon,perfmon,sha2,store-pair-suppress,v8a,zcm,zcz,zcz-fp-workaround,zcz-gp") }, + { str_lit("apple-a9"), str_lit("aes,alternate-sextload-cvt-f32-pattern,apple-a7,apple-a7-sysreg,arith-bcc-fusion,arith-cbz-fusion,crypto,disable-latency-sched-heuristic,el2vmsa,el3,fp-armv8,fuse-aes,fuse-crypto-eor,neon,perfmon,sha2,store-pair-suppress,v8a,zcm,zcz,zcz-fp-workaround,zcz-gp") }, + { str_lit("apple-latest"), str_lit("CONTEXTIDREL2,aes,alternate-sextload-cvt-f32-pattern,altnzcv,am,amvs,apple-a16,arith-bcc-fusion,arith-cbz-fusion,bf16,bti,ccdp,ccidx,ccpp,complxnum,crc,crypto,disable-latency-sched-heuristic,dit,dotprod,ecv,el2vmsa,el3,fgt,flagm,fp-armv8,fp16fml,fptoint,fullfp16,fuse-address,fuse-aes,fuse-arith-logic,fuse-crypto-eor,fuse-csel,fuse-literals,hcx,i8mm,jsconv,lor,lse,lse2,mpam,neon,nv,pan,pan-rwv,pauth,perfmon,predres,ras,rcpc,rcpc-immo,rdm,sb,sel2,sha2,sha3,specrestrict,ssbs,store-pair-suppress,tlb-rmi,tracev8.4,uaops,v8.1a,v8.2a,v8.3a,v8.4a,v8.5a,v8.6a,v8a,vh,zcm,zcz,zcz-gp") }, + { str_lit("apple-m1"), str_lit("CONTEXTIDREL2,aes,aggressive-fma,alternate-sextload-cvt-f32-pattern,altnzcv,am,apple-a14,arith-bcc-fusion,arith-cbz-fusion,ccdp,ccidx,ccpp,complxnum,crc,crypto,disable-latency-sched-heuristic,dit,dotprod,el2vmsa,el3,flagm,fp-armv8,fp16fml,fptoint,fullfp16,fuse-address,fuse-adrp-add,fuse-aes,fuse-arith-logic,fuse-crypto-eor,fuse-csel,fuse-literals,jsconv,lor,lse,lse2,mpam,neon,nv,pan,pan-rwv,pauth,perfmon,predres,ras,rcpc,rcpc-immo,rdm,sb,sel2,sha2,sha3,specrestrict,ssbs,store-pair-suppress,tlb-rmi,tracev8.4,uaops,v8.1a,v8.2a,v8.3a,v8.4a,v8a,vh,zcm,zcz,zcz-gp") }, + { str_lit("apple-m2"), str_lit("CONTEXTIDREL2,aes,alternate-sextload-cvt-f32-pattern,altnzcv,am,amvs,apple-a15,arith-bcc-fusion,arith-cbz-fusion,bf16,bti,ccdp,ccidx,ccpp,complxnum,crc,crypto,disable-latency-sched-heuristic,dit,dotprod,ecv,el2vmsa,el3,fgt,flagm,fp-armv8,fp16fml,fptoint,fullfp16,fuse-address,fuse-aes,fuse-arith-logic,fuse-crypto-eor,fuse-csel,fuse-literals,i8mm,jsconv,lor,lse,lse2,mpam,neon,nv,pan,pan-rwv,pauth,perfmon,predres,ras,rcpc,rcpc-immo,rdm,sb,sel2,sha2,sha3,specrestrict,ssbs,store-pair-suppress,tlb-rmi,tracev8.4,uaops,v8.1a,v8.2a,v8.3a,v8.4a,v8.5a,v8.6a,v8a,vh,zcm,zcz,zcz-gp") }, + { str_lit("apple-s4"), str_lit("CONTEXTIDREL2,aes,alternate-sextload-cvt-f32-pattern,apple-a12,arith-bcc-fusion,arith-cbz-fusion,ccidx,ccpp,complxnum,crc,crypto,disable-latency-sched-heuristic,el2vmsa,el3,fp-armv8,fullfp16,fuse-aes,fuse-crypto-eor,jsconv,lor,lse,neon,pan,pan-rwv,pauth,perfmon,ras,rcpc,rdm,sha2,store-pair-suppress,uaops,v8.1a,v8.2a,v8.3a,v8a,vh,zcm,zcz,zcz-gp") }, + { str_lit("apple-s5"), str_lit("CONTEXTIDREL2,aes,alternate-sextload-cvt-f32-pattern,apple-a12,arith-bcc-fusion,arith-cbz-fusion,ccidx,ccpp,complxnum,crc,crypto,disable-latency-sched-heuristic,el2vmsa,el3,fp-armv8,fullfp16,fuse-aes,fuse-crypto-eor,jsconv,lor,lse,neon,pan,pan-rwv,pauth,perfmon,ras,rcpc,rdm,sha2,store-pair-suppress,uaops,v8.1a,v8.2a,v8.3a,v8a,vh,zcm,zcz,zcz-gp") }, + { str_lit("carmel"), str_lit("CONTEXTIDREL2,aes,carmel,ccpp,crc,crypto,el2vmsa,el3,fp-armv8,fullfp16,lor,lse,neon,pan,pan-rwv,ras,rdm,sha2,uaops,v8.1a,v8.2a,v8a,vh") }, + { str_lit("cortex-a34"), str_lit("a35,aes,crc,crypto,el2vmsa,el3,fp-armv8,neon,perfmon,sha2,v8a") }, + { str_lit("cortex-a35"), str_lit("a35,aes,crc,crypto,el2vmsa,el3,fp-armv8,neon,perfmon,sha2,v8a") }, + { str_lit("cortex-a510"), str_lit("CONTEXTIDREL2,a510,altnzcv,am,bf16,bti,ccdp,ccidx,ccpp,complxnum,crc,dit,dotprod,el2vmsa,el3,ete,flagm,fp-armv8,fp16fml,fptoint,fullfp16,fuse-adrp-add,fuse-aes,i8mm,jsconv,lor,lse,lse2,mec,mpam,mte,neon,nv,pan,pan-rwv,pauth,perfmon,predres,ras,rcpc,rcpc-immo,rdm,rme,sb,sel2,specrestrict,ssbs,sve,sve2,sve2-bitperm,tlb-rmi,tracev8.4,trbe,uaops,use-postra-scheduler,use-scalar-inc-vl,v8.1a,v8.2a,v8.3a,v8.4a,v8.5a,v8a,v9a,vh") }, + { str_lit("cortex-a53"), str_lit("a53,aes,balance-fp-ops,crc,crypto,el2vmsa,el3,fp-armv8,fuse-adrp-add,fuse-aes,neon,perfmon,sha2,use-postra-scheduler,v8a") }, + { str_lit("cortex-a55"), str_lit("CONTEXTIDREL2,a55,aes,ccpp,crc,crypto,dotprod,el2vmsa,el3,fp-armv8,fullfp16,fuse-address,fuse-adrp-add,fuse-aes,lor,lse,neon,pan,pan-rwv,perfmon,ras,rcpc,rdm,sha2,uaops,use-postra-scheduler,v8.1a,v8.2a,v8a,vh") }, + { str_lit("cortex-a57"), str_lit("a57,aes,balance-fp-ops,crc,crypto,el2vmsa,el3,enable-select-opt,fp-armv8,fuse-adrp-add,fuse-aes,fuse-literals,neon,perfmon,predictable-select-expensive,sha2,use-postra-scheduler,v8a") }, + { str_lit("cortex-a65"), str_lit("CONTEXTIDREL2,a65,aes,ccpp,crc,crypto,dotprod,el2vmsa,el3,enable-select-opt,fp-armv8,fullfp16,fuse-address,fuse-adrp-add,fuse-aes,fuse-literals,lor,lse,neon,pan,pan-rwv,perfmon,predictable-select-expensive,ras,rcpc,rdm,sha2,ssbs,uaops,v8.1a,v8.2a,v8a,vh") }, + { str_lit("cortex-a65ae"), str_lit("CONTEXTIDREL2,a65,aes,ccpp,crc,crypto,dotprod,el2vmsa,el3,enable-select-opt,fp-armv8,fullfp16,fuse-address,fuse-adrp-add,fuse-aes,fuse-literals,lor,lse,neon,pan,pan-rwv,perfmon,predictable-select-expensive,ras,rcpc,rdm,sha2,ssbs,uaops,v8.1a,v8.2a,v8a,vh") }, + { str_lit("cortex-a710"), str_lit("CONTEXTIDREL2,a710,addr-lsl-fast,altnzcv,alu-lsl-fast,am,bf16,bti,ccdp,ccidx,ccpp,cmp-bcc-fusion,complxnum,crc,dit,dotprod,el2vmsa,el3,enable-select-opt,ete,flagm,fp-armv8,fp16fml,fptoint,fullfp16,fuse-adrp-add,fuse-aes,i8mm,jsconv,lor,lse,lse2,mec,mpam,mte,neon,nv,pan,pan-rwv,pauth,perfmon,predictable-select-expensive,predres,ras,rcpc,rcpc-immo,rdm,rme,sb,sel2,specrestrict,ssbs,sve,sve2,sve2-bitperm,tlb-rmi,tracev8.4,trbe,uaops,use-postra-scheduler,use-scalar-inc-vl,v8.1a,v8.2a,v8.3a,v8.4a,v8.5a,v8a,v9a,vh") }, + { str_lit("cortex-a715"), str_lit("CONTEXTIDREL2,a715,addr-lsl-fast,altnzcv,alu-lsl-fast,am,bf16,bti,ccdp,ccidx,ccpp,cmp-bcc-fusion,complxnum,crc,dit,dotprod,el2vmsa,el3,enable-select-opt,ete,flagm,fp-armv8,fp16fml,fptoint,fullfp16,fuse-adrp-add,fuse-aes,i8mm,jsconv,lor,lse,lse2,mec,mpam,mte,neon,nv,pan,pan-rwv,pauth,perfmon,predictable-select-expensive,predres,ras,rcpc,rcpc-immo,rdm,rme,sb,sel2,spe,specrestrict,ssbs,sve,sve2,sve2-bitperm,tlb-rmi,tracev8.4,trbe,uaops,use-postra-scheduler,use-scalar-inc-vl,v8.1a,v8.2a,v8.3a,v8.4a,v8.5a,v8a,v9a,vh") }, + { str_lit("cortex-a72"), str_lit("a72,aes,crc,crypto,el2vmsa,el3,enable-select-opt,fp-armv8,fuse-adrp-add,fuse-aes,fuse-literals,neon,perfmon,predictable-select-expensive,sha2,v8a") }, + { str_lit("cortex-a73"), str_lit("a73,aes,crc,crypto,el2vmsa,el3,enable-select-opt,fp-armv8,fuse-adrp-add,fuse-aes,neon,perfmon,predictable-select-expensive,sha2,v8a") }, + { str_lit("cortex-a75"), str_lit("CONTEXTIDREL2,a75,aes,ccpp,crc,crypto,dotprod,el2vmsa,el3,enable-select-opt,fp-armv8,fullfp16,fuse-adrp-add,fuse-aes,lor,lse,neon,pan,pan-rwv,perfmon,predictable-select-expensive,ras,rcpc,rdm,sha2,uaops,v8.1a,v8.2a,v8a,vh") }, + { str_lit("cortex-a76"), str_lit("CONTEXTIDREL2,a76,addr-lsl-fast,aes,alu-lsl-fast,ccpp,crc,crypto,dotprod,el2vmsa,el3,enable-select-opt,fp-armv8,fullfp16,fuse-adrp-add,fuse-aes,lor,lse,neon,pan,pan-rwv,perfmon,predictable-select-expensive,ras,rcpc,rdm,sha2,ssbs,uaops,v8.1a,v8.2a,v8a,vh") }, + { str_lit("cortex-a76ae"), str_lit("CONTEXTIDREL2,a76,addr-lsl-fast,aes,alu-lsl-fast,ccpp,crc,crypto,dotprod,el2vmsa,el3,enable-select-opt,fp-armv8,fullfp16,fuse-adrp-add,fuse-aes,lor,lse,neon,pan,pan-rwv,perfmon,predictable-select-expensive,ras,rcpc,rdm,sha2,ssbs,uaops,v8.1a,v8.2a,v8a,vh") }, + { str_lit("cortex-a77"), str_lit("CONTEXTIDREL2,a77,addr-lsl-fast,aes,alu-lsl-fast,ccpp,cmp-bcc-fusion,crc,crypto,dotprod,el2vmsa,el3,enable-select-opt,fp-armv8,fullfp16,fuse-adrp-add,fuse-aes,lor,lse,neon,pan,pan-rwv,perfmon,predictable-select-expensive,ras,rcpc,rdm,sha2,ssbs,uaops,v8.1a,v8.2a,v8a,vh") }, + { str_lit("cortex-a78"), str_lit("CONTEXTIDREL2,a78,addr-lsl-fast,aes,alu-lsl-fast,ccpp,cmp-bcc-fusion,crc,crypto,dotprod,el2vmsa,el3,enable-select-opt,fp-armv8,fullfp16,fuse-adrp-add,fuse-aes,lor,lse,neon,pan,pan-rwv,perfmon,predictable-select-expensive,ras,rcpc,rdm,sha2,spe,ssbs,uaops,use-postra-scheduler,v8.1a,v8.2a,v8a,vh") }, + { str_lit("cortex-a78c"), str_lit("CONTEXTIDREL2,a78c,addr-lsl-fast,aes,alu-lsl-fast,ccpp,cmp-bcc-fusion,crc,crypto,dotprod,el2vmsa,el3,enable-select-opt,flagm,fp-armv8,fullfp16,fuse-adrp-add,fuse-aes,lor,lse,neon,pan,pan-rwv,pauth,perfmon,predictable-select-expensive,ras,rcpc,rdm,sha2,spe,ssbs,uaops,use-postra-scheduler,v8.1a,v8.2a,v8a,vh") }, + { str_lit("cortex-r82"), str_lit("CONTEXTIDREL2,ccidx,ccpp,complxnum,cortex-r82,crc,dit,dotprod,flagm,fp-armv8,fp16fml,fullfp16,jsconv,lse,neon,pan,pan-rwv,pauth,perfmon,predres,ras,rcpc,rcpc-immo,rdm,sb,sel2,specrestrict,ssbs,tlb-rmi,tracev8.4,uaops,use-postra-scheduler,v8r") }, + { str_lit("cortex-x1"), str_lit("CONTEXTIDREL2,addr-lsl-fast,aes,alu-lsl-fast,ccpp,cmp-bcc-fusion,cortex-x1,crc,crypto,dotprod,el2vmsa,el3,enable-select-opt,fp-armv8,fullfp16,fuse-adrp-add,fuse-aes,lor,lse,neon,pan,pan-rwv,perfmon,predictable-select-expensive,ras,rcpc,rdm,sha2,spe,ssbs,uaops,use-postra-scheduler,v8.1a,v8.2a,v8a,vh") }, + { str_lit("cortex-x1c"), str_lit("CONTEXTIDREL2,addr-lsl-fast,aes,alu-lsl-fast,ccpp,cmp-bcc-fusion,cortex-x1,crc,crypto,dotprod,el2vmsa,el3,enable-select-opt,flagm,fp-armv8,fullfp16,fuse-adrp-add,fuse-aes,lor,lse,lse2,neon,pan,pan-rwv,pauth,perfmon,predictable-select-expensive,ras,rcpc,rcpc-immo,rdm,sha2,spe,ssbs,uaops,use-postra-scheduler,v8.1a,v8.2a,v8a,vh") }, + { str_lit("cortex-x2"), str_lit("CONTEXTIDREL2,addr-lsl-fast,altnzcv,alu-lsl-fast,am,bf16,bti,ccdp,ccidx,ccpp,cmp-bcc-fusion,complxnum,cortex-x2,crc,dit,dotprod,el2vmsa,el3,enable-select-opt,ete,flagm,fp-armv8,fp16fml,fptoint,fullfp16,fuse-adrp-add,fuse-aes,i8mm,jsconv,lor,lse,lse2,mec,mpam,mte,neon,nv,pan,pan-rwv,pauth,perfmon,predictable-select-expensive,predres,ras,rcpc,rcpc-immo,rdm,rme,sb,sel2,specrestrict,ssbs,sve,sve2,sve2-bitperm,tlb-rmi,tracev8.4,trbe,uaops,use-postra-scheduler,use-scalar-inc-vl,v8.1a,v8.2a,v8.3a,v8.4a,v8.5a,v8a,v9a,vh") }, + { str_lit("cortex-x3"), str_lit("CONTEXTIDREL2,addr-lsl-fast,altnzcv,alu-lsl-fast,am,bf16,bti,ccdp,ccidx,ccpp,complxnum,cortex-x3,crc,dit,dotprod,el2vmsa,el3,enable-select-opt,ete,flagm,fp-armv8,fp16fml,fptoint,fullfp16,fuse-adrp-add,fuse-aes,i8mm,jsconv,lor,lse,lse2,mec,mpam,mte,neon,nv,pan,pan-rwv,pauth,perfmon,predictable-select-expensive,predres,ras,rcpc,rcpc-immo,rdm,rme,sb,sel2,spe,specrestrict,ssbs,sve,sve2,sve2-bitperm,tlb-rmi,tracev8.4,trbe,uaops,use-postra-scheduler,use-scalar-inc-vl,v8.1a,v8.2a,v8.3a,v8.4a,v8.5a,v8a,v9a,vh") }, + { str_lit("cyclone"), str_lit("aes,alternate-sextload-cvt-f32-pattern,apple-a7,apple-a7-sysreg,arith-bcc-fusion,arith-cbz-fusion,crypto,disable-latency-sched-heuristic,el2vmsa,el3,fp-armv8,fuse-aes,fuse-crypto-eor,neon,perfmon,sha2,store-pair-suppress,v8a,zcm,zcz,zcz-fp-workaround,zcz-gp") }, + { str_lit("exynos-m3"), str_lit("addr-lsl-fast,aes,alu-lsl-fast,crc,crypto,el2vmsa,el3,exynos-cheap-as-move,exynosm3,force-32bit-jump-tables,fp-armv8,fuse-address,fuse-adrp-add,fuse-aes,fuse-csel,fuse-literals,neon,perfmon,predictable-select-expensive,sha2,store-pair-suppress,use-postra-scheduler,v8a") }, + { str_lit("exynos-m4"), str_lit("CONTEXTIDREL2,addr-lsl-fast,aes,alu-lsl-fast,arith-bcc-fusion,arith-cbz-fusion,ccpp,crc,crypto,dotprod,el2vmsa,el3,exynos-cheap-as-move,exynosm4,force-32bit-jump-tables,fp-armv8,fullfp16,fuse-address,fuse-adrp-add,fuse-aes,fuse-arith-logic,fuse-csel,fuse-literals,lor,lse,neon,pan,pan-rwv,perfmon,ras,rdm,sha2,store-pair-suppress,uaops,use-postra-scheduler,v8.1a,v8.2a,v8a,vh,zcz,zcz-gp") }, + { str_lit("exynos-m5"), str_lit("CONTEXTIDREL2,addr-lsl-fast,aes,alu-lsl-fast,arith-bcc-fusion,arith-cbz-fusion,ccpp,crc,crypto,dotprod,el2vmsa,el3,exynos-cheap-as-move,exynosm4,force-32bit-jump-tables,fp-armv8,fullfp16,fuse-address,fuse-adrp-add,fuse-aes,fuse-arith-logic,fuse-csel,fuse-literals,lor,lse,neon,pan,pan-rwv,perfmon,ras,rdm,sha2,store-pair-suppress,uaops,use-postra-scheduler,v8.1a,v8.2a,v8a,vh,zcz,zcz-gp") }, + { str_lit("falkor"), str_lit("addr-lsl-fast,aes,alu-lsl-fast,crc,crypto,el2vmsa,el3,falkor,fp-armv8,neon,perfmon,predictable-select-expensive,rdm,sha2,slow-strqro-store,store-pair-suppress,use-postra-scheduler,v8a,zcz,zcz-gp") }, + { str_lit("generic"), str_lit("enable-select-opt,ete,fp-armv8,fuse-adrp-add,fuse-aes,neon,trbe,use-postra-scheduler") }, + { str_lit("kryo"), str_lit("addr-lsl-fast,aes,alu-lsl-fast,crc,crypto,el2vmsa,el3,fp-armv8,kryo,neon,perfmon,predictable-select-expensive,sha2,store-pair-suppress,use-postra-scheduler,v8a,zcz,zcz-gp") }, + { str_lit("neoverse-512tvb"), str_lit("CONTEXTIDREL2,addr-lsl-fast,aes,alu-lsl-fast,am,bf16,ccdp,ccidx,ccpp,complxnum,crc,crypto,dit,dotprod,el2vmsa,el3,enable-select-opt,flagm,fp-armv8,fp16fml,fullfp16,fuse-adrp-add,fuse-aes,i8mm,jsconv,lor,lse,lse2,mpam,neon,neoverse512tvb,nv,pan,pan-rwv,pauth,perfmon,predictable-select-expensive,rand,ras,rcpc,rcpc-immo,rdm,sel2,sha2,spe,ssbs,sve,tlb-rmi,tracev8.4,uaops,use-postra-scheduler,v8.1a,v8.2a,v8.3a,v8.4a,v8a,vh") }, + { str_lit("neoverse-e1"), str_lit("CONTEXTIDREL2,aes,ccpp,crc,crypto,dotprod,el2vmsa,el3,fp-armv8,fullfp16,fuse-adrp-add,fuse-aes,lor,lse,neon,neoversee1,pan,pan-rwv,perfmon,ras,rcpc,rdm,sha2,ssbs,uaops,use-postra-scheduler,v8.1a,v8.2a,v8a,vh") }, + { str_lit("neoverse-n1"), str_lit("CONTEXTIDREL2,addr-lsl-fast,aes,alu-lsl-fast,ccpp,crc,crypto,dotprod,el2vmsa,el3,enable-select-opt,fp-armv8,fullfp16,fuse-adrp-add,fuse-aes,lor,lse,neon,neoversen1,pan,pan-rwv,perfmon,predictable-select-expensive,ras,rcpc,rdm,sha2,spe,ssbs,uaops,use-postra-scheduler,v8.1a,v8.2a,v8a,vh") }, + { str_lit("neoverse-n2"), str_lit("CONTEXTIDREL2,addr-lsl-fast,altnzcv,alu-lsl-fast,am,bf16,bti,ccdp,ccidx,ccpp,complxnum,crc,dit,dotprod,el2vmsa,el3,enable-select-opt,ete,flagm,fp-armv8,fptoint,fullfp16,fuse-adrp-add,fuse-aes,i8mm,jsconv,lor,lse,lse2,mec,mpam,mte,neon,neoversen2,nv,pan,pan-rwv,pauth,perfmon,predictable-select-expensive,predres,ras,rcpc,rcpc-immo,rdm,rme,sb,sel2,specrestrict,ssbs,sve,sve2,sve2-bitperm,tlb-rmi,tracev8.4,trbe,uaops,use-postra-scheduler,use-scalar-inc-vl,v8.1a,v8.2a,v8.3a,v8.4a,v8.5a,v8a,v9a,vh") }, + { str_lit("neoverse-v1"), str_lit("CONTEXTIDREL2,addr-lsl-fast,aes,alu-lsl-fast,am,bf16,ccdp,ccidx,ccpp,complxnum,crc,crypto,dit,dotprod,el2vmsa,el3,enable-select-opt,flagm,fp-armv8,fp16fml,fullfp16,fuse-adrp-add,fuse-aes,i8mm,jsconv,lor,lse,lse2,mpam,neon,neoversev1,no-sve-fp-ld1r,nv,pan,pan-rwv,pauth,perfmon,predictable-select-expensive,rand,ras,rcpc,rcpc-immo,rdm,sel2,sha2,spe,ssbs,sve,tlb-rmi,tracev8.4,uaops,use-postra-scheduler,v8.1a,v8.2a,v8.3a,v8.4a,v8a,vh") }, + { str_lit("neoverse-v2"), str_lit("CONTEXTIDREL2,addr-lsl-fast,altnzcv,alu-lsl-fast,am,bf16,bti,ccdp,ccidx,ccpp,complxnum,crc,dit,dotprod,el2vmsa,el3,enable-select-opt,ete,flagm,fp-armv8,fp16fml,fptoint,fullfp16,fuse-adrp-add,fuse-aes,i8mm,jsconv,lor,lse,lse2,mec,mpam,mte,neon,neoversev2,nv,pan,pan-rwv,pauth,perfmon,predictable-select-expensive,predres,rand,ras,rcpc,rcpc-immo,rdm,rme,sb,sel2,spe,specrestrict,ssbs,sve,sve2,sve2-bitperm,tlb-rmi,tracev8.4,trbe,uaops,use-postra-scheduler,use-scalar-inc-vl,v8.1a,v8.2a,v8.3a,v8.4a,v8.5a,v8a,v9a,vh") }, + { str_lit("saphira"), str_lit("CONTEXTIDREL2,addr-lsl-fast,aes,alu-lsl-fast,am,ccidx,ccpp,complxnum,crc,crypto,dit,dotprod,el2vmsa,el3,flagm,fp-armv8,jsconv,lor,lse,lse2,mpam,neon,nv,pan,pan-rwv,pauth,perfmon,predictable-select-expensive,ras,rcpc,rcpc-immo,rdm,saphira,sel2,sha2,spe,store-pair-suppress,tlb-rmi,tracev8.4,uaops,use-postra-scheduler,v8.1a,v8.2a,v8.3a,v8.4a,v8a,vh,zcz,zcz-gp") }, + { str_lit("thunderx"), str_lit("aes,crc,crypto,el2vmsa,el3,fp-armv8,neon,perfmon,predictable-select-expensive,sha2,store-pair-suppress,thunderx,use-postra-scheduler,v8a") }, + { str_lit("thunderx2t99"), str_lit("CONTEXTIDREL2,aes,aggressive-fma,arith-bcc-fusion,crc,crypto,el2vmsa,el3,fp-armv8,lor,lse,neon,pan,predictable-select-expensive,rdm,sha2,store-pair-suppress,thunderx2t99,use-postra-scheduler,v8.1a,v8a,vh") }, + { str_lit("thunderx3t110"), str_lit("CONTEXTIDREL2,aes,aggressive-fma,arith-bcc-fusion,balance-fp-ops,ccidx,ccpp,complxnum,crc,crypto,el2vmsa,el3,fp-armv8,jsconv,lor,lse,neon,pan,pan-rwv,pauth,perfmon,predictable-select-expensive,ras,rcpc,rdm,sha2,store-pair-suppress,strict-align,thunderx3t110,uaops,use-postra-scheduler,v8.1a,v8.2a,v8.3a,v8a,vh") }, + { str_lit("thunderxt81"), str_lit("aes,crc,crypto,el2vmsa,el3,fp-armv8,neon,perfmon,predictable-select-expensive,sha2,store-pair-suppress,thunderxt81,use-postra-scheduler,v8a") }, + { str_lit("thunderxt83"), str_lit("aes,crc,crypto,el2vmsa,el3,fp-armv8,neon,perfmon,predictable-select-expensive,sha2,store-pair-suppress,thunderxt83,use-postra-scheduler,v8a") }, + { str_lit("thunderxt88"), str_lit("aes,crc,crypto,el2vmsa,el3,fp-armv8,neon,perfmon,predictable-select-expensive,sha2,store-pair-suppress,thunderxt88,use-postra-scheduler,v8a") }, + { str_lit("tsv110"), str_lit("CONTEXTIDREL2,aes,ccpp,complxnum,crc,crypto,dotprod,el2vmsa,el3,fp-armv8,fp16fml,fullfp16,fuse-aes,jsconv,lor,lse,neon,pan,pan-rwv,perfmon,ras,rdm,sha2,spe,store-pair-suppress,tsv110,uaops,use-postra-scheduler,v8.1a,v8.2a,v8a,vh") }, + // TargetArch_wasm32: + { str_lit("bleeding-edge"), str_lit("atomics,bulk-memory,mutable-globals,nontrapping-fptoint,sign-ext,simd128,tail-call") }, + { str_lit("generic"), str_lit("mutable-globals,sign-ext") }, + { str_lit("mvp"), str_lit("") }, + // TargetArch_wasm64p32: + { str_lit("bleeding-edge"), str_lit("atomics,bulk-memory,mutable-globals,nontrapping-fptoint,sign-ext,simd128,tail-call") }, + { str_lit("generic"), str_lit("mutable-globals,sign-ext") }, + { str_lit("mvp"), str_lit("") }, }; gb_global String target_endian_names[TargetEndian_COUNT] = { @@ -443,9 +894,9 @@ struct BuildContext { PtrMap defined_values; - BlockingMutex target_features_mutex; StringSet target_features_set; String target_features_string; + bool strict_target_features; String minimum_os_version_string; bool minimum_os_version_string_given; @@ -1596,48 +2047,53 @@ gb_internal void init_build_context(TargetMetrics *cross_target, Subtarget subta #include "microsoft_craziness.h" #endif +// NOTE: the target feature and microarch lists are all sorted, so if it turns out to be slow (I don't think it will) +// a binary search is possible. -gb_internal Array split_by_comma(String const &list) { - isize n = 1; - for (isize i = 0; i < list.len; i++) { - if (list.text[i] == ',') { - n++; +gb_internal bool check_single_target_feature_is_valid(String const &feature_list, String const &feature) { + String_Iterator it = {feature_list, 0}; + for (;;) { + String str = string_split_iterator(&it, ','); + if (str == "") break; + if (str == feature) { + return true; } } - auto res = array_make(heap_allocator(), n); - String s = list; - for (isize i = 0; i < n; i++) { - isize m = string_index_byte(s, ','); - if (m < 0) { - res[i] = s; - break; + return false; +} + +gb_internal bool check_target_feature_is_valid(String const &feature, TargetArchKind arch, String *invalid) { + String feature_list = target_features_list[arch]; + String_Iterator it = {feature, 0}; + for (;;) { + String str = string_split_iterator(&it, ','); + if (str == "") break; + if (!check_single_target_feature_is_valid(feature_list, str)) { + if (invalid) *invalid = str; + return false; } - res[i] = substring(s, 0, m); - s = substring(s, m+1, s.len); } - return res; -} -gb_internal bool check_target_feature_is_valid(TokenPos pos, String const &feature) { - // TODO(bill): check_target_feature_is_valid return true; } -gb_internal bool check_target_feature_is_enabled(TokenPos pos, String const &target_feature_list) { - BuildContext *bc = &build_context; - mutex_lock(&bc->target_features_mutex); - defer (mutex_unlock(&bc->target_features_mutex)); - - auto items = split_by_comma(target_feature_list); - array_free(&items); - for (String const &item : items) { - if (!check_target_feature_is_valid(pos, item)) { - error(pos, "Target feature '%.*s' is not valid", LIT(item)); - return false; +gb_internal bool check_target_feature_is_valid_globally(String const &feature, String *invalid) { + String_Iterator it = {feature, 0}; + for (;;) { + String str = string_split_iterator(&it, ','); + if (str == "") break; + + bool valid = false; + for (int arch = TargetArch_Invalid; arch < TargetArch_COUNT; arch += 1) { + if (check_target_feature_is_valid(str, cast(TargetArchKind)arch, invalid)) { + valid = true; + break; + } } - if (!string_set_exists(&bc->target_features_set, item)) { - error(pos, "Target feature '%.*s' is not enabled", LIT(item)); + + if (!valid) { + if (invalid) *invalid = str; return false; } } @@ -1645,54 +2101,35 @@ gb_internal bool check_target_feature_is_enabled(TokenPos pos, String const &tar return true; } -gb_internal void enable_target_feature(TokenPos pos, String const &target_feature_list) { - BuildContext *bc = &build_context; - mutex_lock(&bc->target_features_mutex); - defer (mutex_unlock(&bc->target_features_mutex)); - - auto items = split_by_comma(target_feature_list); - for (String const &item : items) { - if (!check_target_feature_is_valid(pos, item)) { - error(pos, "Target feature '%.*s' is not valid", LIT(item)); - continue; - } - - string_set_add(&bc->target_features_set, item); - } - array_free(&items); +gb_internal bool check_target_feature_is_valid_for_target_arch(String const &feature, String *invalid) { + return check_target_feature_is_valid(feature, build_context.metrics.arch, invalid); } - -gb_internal char const *target_features_set_to_cstring(gbAllocator allocator, bool with_quotes, bool with_plus) { - isize len = 0; - isize i = 0; - for (String const &feature : build_context.target_features_set) { - if (i != 0) { - len += 1; +gb_internal bool check_target_feature_is_enabled(String const &feature, String *not_enabled) { + String_Iterator it = {feature, 0}; + for (;;) { + String str = string_split_iterator(&it, ','); + if (str == "") break; + if (!string_set_exists(&build_context.target_features_set, str)) { + if (not_enabled) *not_enabled = str; + return false; } - len += feature.len; - if (with_quotes) len += 2; - if (with_plus) len += 1; - i += 1; } - char *features = gb_alloc_array(allocator, char, len+1); - len = 0; - i = 0; - for (String const &feature : build_context.target_features_set) { - if (i != 0) { - features[len++] = ','; - } - if (with_quotes) features[len++] = '"'; - if (with_plus) features[len++] = '+'; - gb_memmove(features + len, feature.text, feature.len); - len += feature.len; - if (with_quotes) features[len++] = '"'; - i += 1; - } - features[len++] = 0; + return true; +} - return features; +gb_internal bool check_target_feature_is_superset_of(String const &superset, String const &of, String *missing) { + String_Iterator it = {of, 0}; + for (;;) { + String str = string_split_iterator(&it, ','); + if (str == "") break; + if (!check_single_target_feature_is_valid(superset, str)) { + if (missing) *missing = str; + return false; + } + } + return true; } // NOTE(Jeroen): Set/create the output and other paths and report an error as appropriate. @@ -1983,10 +2420,6 @@ gb_internal bool init_build_paths(String init_filename) { } } - if (bc->target_features_string.len != 0) { - enable_target_feature({}, bc->target_features_string); - } - return true; } diff --git a/src/check_builtin.cpp b/src/check_builtin.cpp index c3c217ec7..825fc6448 100644 --- a/src/check_builtin.cpp +++ b/src/check_builtin.cpp @@ -1719,6 +1719,7 @@ gb_internal bool check_builtin_procedure(CheckerContext *c, Operand *operand, As case BuiltinProc_objc_register_selector: case BuiltinProc_objc_register_class: case BuiltinProc_atomic_type_is_lock_free: + case BuiltinProc_has_target_feature: // NOTE(bill): The first arg may be a Type, this will be checked case by case break; @@ -3663,6 +3664,41 @@ gb_internal bool check_builtin_procedure(CheckerContext *c, Operand *operand, As break; } + case BuiltinProc_has_target_feature: { + String features = str_lit(""); + + check_expr_or_type(c, operand, ce->args[0]); + + if (is_type_string(operand->type) && operand->mode == Addressing_Constant) { + GB_ASSERT(operand->value.kind == ExactValue_String); + features = operand->value.value_string; + } else { + Type *pt = base_type(operand->type); + if (pt->kind == Type_Proc) { + if (pt->Proc.require_target_feature.len != 0) { + GB_ASSERT(pt->Proc.enable_target_feature.len == 0); + features = pt->Proc.require_target_feature; + } else if (pt->Proc.enable_target_feature.len != 0) { + features = pt->Proc.enable_target_feature; + } else { + error(ce->args[0], "Expected the procedure type given to '%.*s' to have @(require_target_feature=\"...\") or @(enable_target_feature=\"...\")", LIT(builtin_name)); + } + } else { + error(ce->args[0], "Expected a constant string or procedure type for '%.*s'", LIT(builtin_name)); + } + } + + String invalid; + if (!check_target_feature_is_valid_globally(features, &invalid)) { + error(ce->args[0], "Target feature '%.*s' is not a valid target feature", LIT(invalid)); + } + + operand->value = exact_value_bool(check_target_feature_is_enabled(features, nullptr)); + operand->mode = Addressing_Constant; + operand->type = t_untyped_bool; + break; + } + case BuiltinProc_soa_struct: { Operand x = {}; Operand y = {}; @@ -6014,7 +6050,10 @@ gb_internal bool check_builtin_procedure(CheckerContext *c, Operand *operand, As return false; } - enable_target_feature({}, str_lit("atomics")); + if (!check_target_feature_is_enabled(str_lit("atomics"), nullptr)) { + error(call, "'%.*s' requires target feature 'atomics' to be enabled, enable it with -target-features:\"atomics\" or choose a different -microarch", LIT(builtin_name)); + return false; + } Operand ptr = {}; Operand expected = {}; @@ -6068,7 +6107,10 @@ gb_internal bool check_builtin_procedure(CheckerContext *c, Operand *operand, As return false; } - enable_target_feature({}, str_lit("atomics")); + if (!check_target_feature_is_enabled(str_lit("atomics"), nullptr)) { + error(call, "'%.*s' requires target feature 'atomics' to be enabled, enable it with -target-features:\"atomics\" or choose a different -microarch", LIT(builtin_name)); + return false; + } Operand ptr = {}; Operand waiters = {}; diff --git a/src/check_decl.cpp b/src/check_decl.cpp index 952a877a4..5b9486873 100644 --- a/src/check_decl.cpp +++ b/src/check_decl.cpp @@ -886,17 +886,37 @@ gb_internal void check_proc_decl(CheckerContext *ctx, Entity *e, DeclInfo *d) { check_objc_methods(ctx, e, ac); - if (ac.require_target_feature.len != 0 && ac.enable_target_feature.len != 0) { - error(e->token, "Attributes @(require_target_feature=...) and @(enable_target_feature=...) cannot be used together"); - } else if (ac.require_target_feature.len != 0) { - if (check_target_feature_is_enabled(e->token.pos, ac.require_target_feature)) { - e->Procedure.target_feature = ac.require_target_feature; - } else { - e->Procedure.target_feature_disabled = true; + { + if (ac.require_target_feature.len != 0 && ac.enable_target_feature.len != 0) { + error(e->token, "A procedure cannot have both @(require_target_feature=\"...\") and @(enable_target_feature=\"...\")"); + } + + if (build_context.strict_target_features && ac.enable_target_feature.len != 0) { + ac.require_target_feature = ac.enable_target_feature; + ac.enable_target_feature.len = 0; + } + + if (ac.require_target_feature.len != 0) { + pt->require_target_feature = ac.require_target_feature; + String invalid; + if (!check_target_feature_is_valid_globally(ac.require_target_feature, &invalid)) { + error(e->token, "Required target feature '%.*s' is not a valid target feature", LIT(invalid)); + } else if (!check_target_feature_is_enabled(ac.require_target_feature, nullptr)) { + e->flags |= EntityFlag_Disabled; + } + } else if (ac.enable_target_feature.len != 0) { + + // NOTE: disallow wasm, features on that arch are always global to the module. + if (is_arch_wasm()) { + error(e->token, "@(enable_target_feature=\"...\") is not allowed on wasm, features for wasm must be declared globally"); + } + + pt->enable_target_feature = ac.enable_target_feature; + String invalid; + if (!check_target_feature_is_valid_globally(ac.enable_target_feature, &invalid)) { + error(e->token, "Procedure enabled target feature '%.*s' is not a valid target feature", LIT(invalid)); + } } - } else if (ac.enable_target_feature.len != 0) { - enable_target_feature(e->token.pos, ac.enable_target_feature); - e->Procedure.target_feature = ac.enable_target_feature; } switch (e->Procedure.optimization_mode) { @@ -1370,6 +1390,10 @@ gb_internal void check_proc_group_decl(CheckerContext *ctx, Entity *pg_entity, D continue; } + if (p->flags & EntityFlag_Disabled) { + continue; + } + String name = p->token.string; for (isize k = j+1; k < pge->entities.count; k++) { @@ -1387,6 +1411,10 @@ gb_internal void check_proc_group_decl(CheckerContext *ctx, Entity *pg_entity, D ERROR_BLOCK(); + if (q->flags & EntityFlag_Disabled) { + continue; + } + ProcTypeOverloadKind kind = are_proc_types_overload_safe(p->type, q->type); bool both_have_where_clauses = false; if (p->decl_info->proc_lit != nullptr && q->decl_info->proc_lit != nullptr) { @@ -1423,6 +1451,7 @@ gb_internal void check_proc_group_decl(CheckerContext *ctx, Entity *pg_entity, D break; case ProcOverload_ParamCount: case ProcOverload_ParamTypes: + case ProcOverload_TargetFeatures: // This is okay :) break; diff --git a/src/check_expr.cpp b/src/check_expr.cpp index 06d0a8b12..490c9aae7 100644 --- a/src/check_expr.cpp +++ b/src/check_expr.cpp @@ -6526,12 +6526,17 @@ gb_internal CallArgumentData check_call_arguments_proc_group(CheckerContext *c, array_add(&proc_entities, proc); } + int max_matched_features = 0; gbString expr_name = expr_to_string(operand->expr); defer (gb_string_free(expr_name)); for_array(i, procs) { Entity *p = procs[i]; + if (p->flags & EntityFlag_Disabled) { + continue; + } + Type *pt = base_type(p->type); if (pt != nullptr && is_type_proc(pt)) { CallArgumentData data = {}; @@ -6562,11 +6567,24 @@ gb_internal CallArgumentData check_call_arguments_proc_group(CheckerContext *c, item.score += assign_score_function(1); } + max_matched_features = gb_max(max_matched_features, matched_target_features(&pt->Proc)); + item.index = index; array_add(&valids, item); } } + if (max_matched_features > 0) { + for_array(i, valids) { + Entity *p = procs[valids[i].index]; + Type *t = base_type(p->type); + GB_ASSERT(t->kind == Type_Proc); + + int matched = matched_target_features(&t->Proc); + valids[i].score += assign_score_function(max_matched_features-matched); + } + } + if (valids.count > 1) { array_sort(valids, valid_index_and_score_cmp); i64 best_score = valids[0].score; @@ -6708,7 +6726,11 @@ gb_internal CallArgumentData check_call_arguments_proc_group(CheckerContext *c, ERROR_BLOCK(); error(operand->expr, "Ambiguous procedure group call '%s' that match with the given arguments", expr_name); - print_argument_types(); + if (positional_operands.count == 0 && named_operands.count == 0) { + error_line("\tNo given arguments\n"); + } else { + print_argument_types(); + } for (auto const &valid : valids) { Entity *proc = proc_entities[valid.index]; @@ -7553,8 +7575,11 @@ gb_internal ExprKind check_call_expr(CheckerContext *c, Operand *operand, Ast *c } } + bool is_call_inlined = false; + switch (inlining) { case ProcInlining_inline: + is_call_inlined = true; if (proc != nullptr) { Entity *e = entity_from_expr(proc); if (e != nullptr && e->kind == Entity_Procedure) { @@ -7570,6 +7595,47 @@ gb_internal ExprKind check_call_expr(CheckerContext *c, Operand *operand, Ast *c break; case ProcInlining_no_inline: break; + case ProcInlining_none: + if (proc != nullptr) { + Entity *e = entity_from_expr(proc); + if (e != nullptr && e->kind == Entity_Procedure) { + DeclInfo *decl = e->decl_info; + if (decl->proc_lit) { + ast_node(pl, ProcLit, decl->proc_lit); + if (pl->inlining == ProcInlining_inline) { + is_call_inlined = true; + } + } + } + } + } + + { + String invalid; + if (pt->kind == Type_Proc && pt->Proc.require_target_feature.len != 0) { + if (!check_target_feature_is_valid_for_target_arch(pt->Proc.require_target_feature, &invalid)) { + error(call, "Called procedure requires target feature '%.*s' which is invalid for the build target", LIT(invalid)); + } else if (!check_target_feature_is_enabled(pt->Proc.require_target_feature, &invalid)) { + error(call, "Calling this procedure requires target feature '%.*s' to be enabled", LIT(invalid)); + } + } + + if (pt->kind == Type_Proc && pt->Proc.enable_target_feature.len != 0) { + if (!check_target_feature_is_valid_for_target_arch(pt->Proc.enable_target_feature, &invalid)) { + error(call, "Called procedure enables target feature '%.*s' which is invalid for the build target", LIT(invalid)); + } + + // NOTE: Due to restrictions in LLVM you can not inline calls with a superset of features. + if (is_call_inlined) { + GB_ASSERT(c->curr_proc_decl); + GB_ASSERT(c->curr_proc_decl->entity); + GB_ASSERT(c->curr_proc_decl->entity->type->kind == Type_Proc); + String scope_features = c->curr_proc_decl->entity->type->Proc.enable_target_feature; + if (!check_target_feature_is_superset_of(scope_features, pt->Proc.enable_target_feature, &invalid)) { + error(call, "Inlined procedure enables target feature '%.*s', this requires the calling procedure to at least enable the same feature", LIT(invalid)); + } + } + } } operand->expr = call; diff --git a/src/checker_builtin_procs.hpp b/src/checker_builtin_procs.hpp index c15ec7137..8419c6568 100644 --- a/src/checker_builtin_procs.hpp +++ b/src/checker_builtin_procs.hpp @@ -44,6 +44,8 @@ enum BuiltinProcId { // "Intrinsics" BuiltinProc_is_package_imported, + BuiltinProc_has_target_feature, + BuiltinProc_transpose, BuiltinProc_outer_product, BuiltinProc_hadamard_product, @@ -354,6 +356,8 @@ gb_global BuiltinProc builtin_procs[BuiltinProc_COUNT] = { // "Intrinsics" {STR_LIT("is_package_imported"), 1, false, Expr_Expr, BuiltinProcPkg_intrinsics}, + {STR_LIT("has_target_feature"), 1, false, Expr_Expr, BuiltinProcPkg_intrinsics}, + {STR_LIT("transpose"), 1, false, Expr_Expr, BuiltinProcPkg_intrinsics}, {STR_LIT("outer_product"), 2, false, Expr_Expr, BuiltinProcPkg_intrinsics}, {STR_LIT("hadamard_product"), 2, false, Expr_Expr, BuiltinProcPkg_intrinsics}, diff --git a/src/entity.cpp b/src/entity.cpp index a12e1d0a6..d76d5f441 100644 --- a/src/entity.cpp +++ b/src/entity.cpp @@ -252,10 +252,8 @@ struct Entity { bool is_foreign : 1; bool is_export : 1; bool generated_from_polymorphic : 1; - bool target_feature_disabled : 1; bool entry_point_only : 1; bool has_instrumentation : 1; - String target_feature; } Procedure; struct { Array entities; @@ -502,4 +500,4 @@ gb_internal bool is_entity_local_variable(Entity *e) { return ((e->scope->flags &~ ScopeFlag_ContextDefined) == 0) || (e->scope->flags & ScopeFlag_Proc) != 0; -} \ No newline at end of file +} diff --git a/src/llvm_backend.cpp b/src/llvm_backend.cpp index 4b94cf020..fad130b99 100644 --- a/src/llvm_backend.cpp +++ b/src/llvm_backend.cpp @@ -41,6 +41,37 @@ String get_default_microarchitecture() { return default_march; } +String get_final_microarchitecture() { + BuildContext *bc = &build_context; + + String microarch = bc->microarch; + if (microarch.len == 0) { + microarch = get_default_microarchitecture(); + } else if (microarch == str_lit("native")) { + microarch = make_string_c(LLVMGetHostCPUName()); + } + return microarch; +} + +gb_internal String get_default_features() { + BuildContext *bc = &build_context; + + int off = 0; + for (int i = 0; i < bc->metrics.arch; i += 1) { + off += target_microarch_counts[i]; + } + + String microarch = get_final_microarchitecture(); + for (int i = off; i < off+target_microarch_counts[bc->metrics.arch]; i += 1) { + if (microarch_features_list[i].microarch == microarch) { + return microarch_features_list[i].features; + } + } + + GB_PANIC("unknown microarch"); + return {}; +} + gb_internal void lb_add_foreign_library_path(lbModule *m, Entity *e) { if (e == nullptr) { return; @@ -2468,69 +2499,24 @@ gb_internal bool lb_generate_code(lbGenerator *gen) { code_mode = LLVMCodeModelKernel; } - String host_cpu_name = copy_string(permanent_allocator(), make_string_c(LLVMGetHostCPUName())); - String llvm_cpu = get_default_microarchitecture(); - char const *llvm_features = ""; - if (build_context.microarch.len != 0) { - if (build_context.microarch == "native") { - llvm_cpu = host_cpu_name; - } else { - llvm_cpu = copy_string(permanent_allocator(), build_context.microarch); - } - if (llvm_cpu == host_cpu_name) { - llvm_features = LLVMGetHostCPUFeatures(); + String llvm_cpu = get_final_microarchitecture(); + + gbString llvm_features = gb_string_make(temporary_allocator(), ""); + String_Iterator it = {build_context.target_features_string, 0}; + bool first = true; + for (;;) { + String str = string_split_iterator(&it, ','); + if (str == "") break; + if (!first) { + llvm_features = gb_string_appendc(llvm_features, ","); } - } + first = false; - // NOTE(Jeroen): Uncomment to get the list of supported microarchitectures. - /* - if (build_context.microarch == "?") { - string_set_add(&build_context.target_features_set, str_lit("+cpuhelp")); + llvm_features = gb_string_appendc(llvm_features, "+"); + llvm_features = gb_string_append_length(llvm_features, str.text, str.len); } - */ - if (build_context.target_features_set.entries.count != 0) { - // Prefix all of the features with a `+`, because we are - // enabling additional features. - char const *additional_features = target_features_set_to_cstring(permanent_allocator(), false, true); - - String f_string = make_string_c(llvm_features); - String a_string = make_string_c(additional_features); - isize f_len = f_string.len; - - if (f_len == 0) { - // The common case is that llvm_features is empty, so - // the target_features_set additions can be used as is. - llvm_features = additional_features; - } else { - // The user probably specified `-microarch:native`, so - // llvm_features is populated by LLVM's idea of what - // the host CPU supports. - // - // As far as I can tell, (which is barely better than - // wild guessing), a bitset is formed by parsing the - // string left to right. - // - // So, llvm_features + ',' + additonal_features, will - // makes the target_features_set override llvm_features. - - char *tmp = gb_alloc_array(permanent_allocator(), char, f_len + 1 + a_string.len + 1); - isize len = 0; - - // tmp = f_string - gb_memmove(tmp, f_string.text, f_string.len); - len += f_string.len; - // tmp += ',' - tmp[len++] = ','; - // tmp += a_string - gb_memmove(tmp + len, a_string.text, a_string.len); - len += a_string.len; - // tmp += NUL - tmp[len++] = 0; - - llvm_features = tmp; - } - } + debugf("CPU: %.*s, Features: %s\n", LIT(llvm_cpu), llvm_features); // GB_ASSERT_MSG(LLVMTargetHasAsmBackend(target)); diff --git a/src/llvm_backend_proc.cpp b/src/llvm_backend_proc.cpp index f73698d34..898c9ac31 100644 --- a/src/llvm_backend_proc.cpp +++ b/src/llvm_backend_proc.cpp @@ -177,17 +177,24 @@ gb_internal lbProcedure *lb_create_procedure(lbModule *m, Entity *entity, bool i break; } - if (!entity->Procedure.target_feature_disabled && - entity->Procedure.target_feature.len != 0) { - auto features = split_by_comma(entity->Procedure.target_feature); - for_array(i, features) { - String feature = features[i]; - LLVMAttributeRef ref = LLVMCreateStringAttribute( - m->ctx, - cast(char const *)feature.text, cast(unsigned)feature.len, - "", 0); - LLVMAddAttributeAtIndex(p->value, LLVMAttributeIndex_FunctionIndex, ref); + if (pt->Proc.enable_target_feature.len != 0) { + gbString feature_str = gb_string_make(temporary_allocator(), ""); + + String_Iterator it = {pt->Proc.enable_target_feature, 0}; + bool first = true; + for (;;) { + String str = string_split_iterator(&it, ','); + if (str == "") break; + if (!first) { + feature_str = gb_string_appendc(feature_str, ","); + } + first = false; + + feature_str = gb_string_appendc(feature_str, "+"); + feature_str = gb_string_append_length(feature_str, str.text, str.len); } + + lb_add_attribute_to_proc_with_string(m, p->value, make_string_c("target-features"), make_string_c(feature_str)); } if (entity->flags & EntityFlag_Cold) { diff --git a/src/llvm_backend_utility.cpp b/src/llvm_backend_utility.cpp index c01ab0692..db99ebc99 100644 --- a/src/llvm_backend_utility.cpp +++ b/src/llvm_backend_utility.cpp @@ -1708,7 +1708,8 @@ gb_internal lbValue lb_emit_mul_add(lbProcedure *p, lbValue a, lbValue b, lbValu if (is_possible) { switch (build_context.metrics.arch) { case TargetArch_amd64: - if (type_size_of(t) == 2) { + // NOTE: using the intrinsic when not supported causes slow codegen (See #2928). + if (type_size_of(t) == 2 || !check_target_feature_is_enabled(str_lit("fma"), nullptr)) { is_possible = false; } break; diff --git a/src/main.cpp b/src/main.cpp index ee7de7f81..93685acb9 100644 --- a/src/main.cpp +++ b/src/main.cpp @@ -272,6 +272,7 @@ enum BuildFlagKind { BuildFlag_ExtraAssemblerFlags, BuildFlag_Microarch, BuildFlag_TargetFeatures, + BuildFlag_StrictTargetFeatures, BuildFlag_MinimumOSVersion, BuildFlag_NoThreadLocal, @@ -467,6 +468,7 @@ gb_internal bool parse_build_flags(Array args) { add_flag(&build_flags, BuildFlag_ExtraAssemblerFlags, str_lit("extra-assembler-flags"), BuildFlagParam_String, Command__does_build); add_flag(&build_flags, BuildFlag_Microarch, str_lit("microarch"), BuildFlagParam_String, Command__does_build); add_flag(&build_flags, BuildFlag_TargetFeatures, str_lit("target-features"), BuildFlagParam_String, Command__does_build); + add_flag(&build_flags, BuildFlag_StrictTargetFeatures, str_lit("strict-target-features"), BuildFlagParam_None, Command__does_build); add_flag(&build_flags, BuildFlag_MinimumOSVersion, str_lit("minimum-os-version"), BuildFlagParam_String, Command__does_build); add_flag(&build_flags, BuildFlag_RelocMode, str_lit("reloc-mode"), BuildFlagParam_String, Command__does_build); @@ -1083,6 +1085,9 @@ gb_internal bool parse_build_flags(Array args) { string_to_lower(&build_context.target_features_string); break; } + case BuildFlag_StrictTargetFeatures: + build_context.strict_target_features = true; + break; case BuildFlag_MinimumOSVersion: { GB_ASSERT(value.kind == ExactValue_String); build_context.minimum_os_version_string = value.value_string; @@ -1981,7 +1986,20 @@ gb_internal void print_show_help(String const arg0, String const &command) { print_usage_line(2, "Examples:"); print_usage_line(3, "-microarch:sandybridge"); print_usage_line(3, "-microarch:native"); - print_usage_line(3, "-microarch:? for a list"); + print_usage_line(3, "-microarch:\"?\" for a list"); + print_usage_line(0, ""); + + print_usage_line(1, "-target-features:"); + print_usage_line(2, "Specifies CPU features to enable on top of the enabled features implied by -microarch."); + print_usage_line(2, "Examples:"); + print_usage_line(3, "-target-features:atomics"); + print_usage_line(3, "-target-features:\"sse2,aes\""); + print_usage_line(3, "-target-features:\"?\" for a list"); + print_usage_line(0, ""); + + print_usage_line(1, "-strict-target-features"); + print_usage_line(2, "Makes @(enable_target_features=\"...\") behave the same way as @(require_target_features=\"...\")."); + print_usage_line(2, "This enforces that all generated code uses features supported by the combination of -target, -microarch, and -target-features."); print_usage_line(0, ""); print_usage_line(1, "-reloc-mode:"); @@ -2663,7 +2681,7 @@ int main(int arg_count, char const **arg_ptr) { // Check chosen microarchitecture. If not found or ?, print list. bool print_microarch_list = true; - if (build_context.microarch.len == 0) { + if (build_context.microarch.len == 0 || build_context.microarch == str_lit("native")) { // Autodetect, no need to print list. print_microarch_list = false; } else { @@ -2680,6 +2698,11 @@ int main(int arg_count, char const **arg_ptr) { } } + // Set and check build paths... + if (!init_build_paths(init_filename)) { + return 1; + } + String default_march = get_default_microarchitecture(); if (print_microarch_list) { if (build_context.microarch != "?") { @@ -2703,13 +2726,57 @@ int main(int arg_count, char const **arg_ptr) { return 0; } - // Set and check build paths... - if (!init_build_paths(init_filename)) { - return 1; + String march = get_final_microarchitecture(); + String default_features = get_default_features(); + { + String_Iterator it = {default_features, 0}; + for (;;) { + String str = string_split_iterator(&it, ','); + if (str == "") break; + string_set_add(&build_context.target_features_set, str); + } + } + + if (build_context.target_features_string.len != 0) { + String_Iterator target_it = {build_context.target_features_string, 0}; + for (;;) { + String item = string_split_iterator(&target_it, ','); + if (item == "") break; + + String invalid; + if (!check_target_feature_is_valid_for_target_arch(item, &invalid) && item != str_lit("help")) { + if (item != str_lit("?")) { + gb_printf_err("Unkown target feature '%.*s'.\n", LIT(invalid)); + } + gb_printf("Possible -target-features for target %.*s are:\n", LIT(target_arch_names[build_context.metrics.arch])); + gb_printf("\n"); + + String feature_list = target_features_list[build_context.metrics.arch]; + String_Iterator it = {feature_list, 0}; + for (;;) { + String str = string_split_iterator(&it, ','); + if (str == "") break; + if (check_single_target_feature_is_valid(default_features, str)) { + if (has_ansi_terminal_colours()) { + gb_printf("\t%.*s\x1b[38;5;244m (implied by target microarch %.*s)\x1b[0m\n", LIT(str), LIT(march)); + } else { + gb_printf("\t%.*s (implied by current microarch %.*s)\n", LIT(str), LIT(march)); + } + } else { + gb_printf("\t%.*s\n", LIT(str)); + } + } + + return 1; + } + + string_set_add(&build_context.target_features_set, item); + } } if (build_context.show_debug_messages) { - debugf("Selected microarch: %.*s\n", LIT(default_march)); + debugf("Selected microarch: %.*s\n", LIT(march)); + debugf("Default microarch features: %.*s\n", LIT(default_features)); for_array(i, build_context.build_paths) { String build_path = path_to_string(heap_allocator(), build_context.build_paths[i]); debugf("build_paths[%ld]: %.*s\n", i, LIT(build_path)); diff --git a/src/types.cpp b/src/types.cpp index 18cb12ea1..3ec05059f 100644 --- a/src/types.cpp +++ b/src/types.cpp @@ -184,6 +184,8 @@ struct TypeProc { isize specialization_count; ProcCallingConvention calling_convention; i32 variadic_index; + String require_target_feature; + String enable_target_feature; // TODO(bill): Make this a flag set rather than bools bool variadic; bool require_results; @@ -2991,7 +2993,22 @@ gb_internal Type *union_tag_type(Type *u) { return t_uint; } +gb_internal int matched_target_features(TypeProc *t) { + if (t->require_target_feature.len == 0) { + return 0; + } + int matches = 0; + String_Iterator it = {t->require_target_feature, 0}; + for (;;) { + String str = string_split_iterator(&it, ','); + if (str == "") break; + if (check_target_feature_is_valid_for_target_arch(str, nullptr)) { + matches += 1; + } + } + return matches; +} enum ProcTypeOverloadKind { ProcOverload_Identical, // The types are identical @@ -3003,6 +3020,7 @@ enum ProcTypeOverloadKind { ProcOverload_ResultCount, ProcOverload_ResultTypes, ProcOverload_Polymorphic, + ProcOverload_TargetFeatures, ProcOverload_NotProcedure, @@ -3060,6 +3078,10 @@ gb_internal ProcTypeOverloadKind are_proc_types_overload_safe(Type *x, Type *y) } } + if (matched_target_features(&px) != matched_target_features(&py)) { + return ProcOverload_TargetFeatures; + } + if (px.params != nullptr && py.params != nullptr) { Entity *ex = px.params->Tuple.variables[0]; Entity *ey = py.params->Tuple.variables[0]; -- cgit v1.2.3 From d85c8f0b2c5989f7d14b02c9023060990d241111 Mon Sep 17 00:00:00 2001 From: gingerBill Date: Thu, 9 May 2024 10:58:57 +0100 Subject: Fix #3555 --- src/check_decl.cpp | 11 +++++++++++ src/check_type.cpp | 1 + src/entity.cpp | 1 + src/types.cpp | 4 ++++ 4 files changed, 17 insertions(+) (limited to 'src/entity.cpp') diff --git a/src/check_decl.cpp b/src/check_decl.cpp index 5b9486873..441c8000d 100644 --- a/src/check_decl.cpp +++ b/src/check_decl.cpp @@ -1619,6 +1619,17 @@ gb_internal bool check_proc_body(CheckerContext *ctx_, Token token, DeclInfo *de if (e->kind != Entity_Variable) { continue; } + if (is_type_polymorphic(e->type)) { + gbString s = type_to_string(e->type); + char const *msg = "Unspecialized polymorphic types are not allowed in procedure parameters, got %s"; + if (e->Variable.type_expr) { + error(e->Variable.type_expr, msg, s); + } else { + error(e->token, msg, s); + } + gb_string_free(s); + } + if (!(e->flags & EntityFlag_Using)) { continue; } diff --git a/src/check_type.cpp b/src/check_type.cpp index c119ce6b5..3d11b5012 100644 --- a/src/check_type.cpp +++ b/src/check_type.cpp @@ -2076,6 +2076,7 @@ gb_internal Type *check_get_params(CheckerContext *ctx, Scope *scope, Ast *_para param = alloc_entity_param(scope, name->Ident.token, type, is_using, true); param->Variable.param_value = param_value; param->Variable.field_group_index = field_group_index; + param->Variable.type_expr = type_expr; } } if (p->flags&FieldFlag_no_alias) { diff --git a/src/entity.cpp b/src/entity.cpp index d76d5f441..60ca208ec 100644 --- a/src/entity.cpp +++ b/src/entity.cpp @@ -210,6 +210,7 @@ struct Entity { CommentGroup *comment; } Constant; struct { + Ast *type_expr; // only used for some variables within procedure bodies Ast *init_expr; // only used for some variables within procedure bodies i32 field_index; i32 field_group_index; diff --git a/src/types.cpp b/src/types.cpp index 3ec05059f..30e009086 100644 --- a/src/types.cpp +++ b/src/types.cpp @@ -3267,6 +3267,10 @@ gb_internal Selection lookup_field_with_selection(Type *type_, String field_name } } + if (is_type_polymorphic(type)) { + // NOTE(bill): A polymorphic struct has no fields, this only hits in the case of an error + return sel; + } wait_signal_until_available(&type->Struct.fields_wait_signal); isize field_count = type->Struct.fields.count; if (field_count != 0) for_array(i, type->Struct.fields) { -- cgit v1.2.3 From a344bc4c0e672d9740f5777dae862723fe269973 Mon Sep 17 00:00:00 2001 From: gingerBill Date: Thu, 16 May 2024 14:39:16 +0100 Subject: Remove the old switch/for semantics entirely and enforce `switch &x in y` --- src/check_expr.cpp | 46 +++++++++++++--------------------------------- src/check_stmt.cpp | 30 ++---------------------------- src/entity.cpp | 2 -- 3 files changed, 15 insertions(+), 63 deletions(-) (limited to 'src/entity.cpp') diff --git a/src/check_expr.cpp b/src/check_expr.cpp index 6a293a97e..c44232dab 100644 --- a/src/check_expr.cpp +++ b/src/check_expr.cpp @@ -2440,32 +2440,6 @@ gb_internal bool check_is_not_addressable(CheckerContext *c, Operand *o) { return o->mode != Addressing_Variable && o->mode != Addressing_SoaVariable; } -gb_internal void check_old_for_or_switch_value_usage(Ast *expr) { - Entity *e = entity_of_node(expr); - if (e != nullptr && (e->flags & EntityFlag_OldForOrSwitchValue) != 0) { - GB_ASSERT(e->kind == Entity_Variable); - - ERROR_BLOCK(); - - if ((e->flags & EntityFlag_ForValue) != 0) { - Type *parent_type = type_deref(e->Variable.for_loop_parent_type); - - error(expr, "Assuming a for-in defined value is addressable as the iterable is passed by value has been disallowed."); - - if (is_type_map(parent_type)) { - error_line("\tSuggestion: Prefer doing 'for key, &%.*s in ...'\n", LIT(e->token.string)); - } else { - error_line("\tSuggestion: Prefer doing 'for &%.*s in ...'\n", LIT(e->token.string)); - } - } else { - GB_ASSERT((e->flags & EntityFlag_SwitchValue) != 0); - - error(expr, "Assuming a switch-in defined value is addressable as the iterable is passed by value has been disallowed."); - error_line("\tSuggestion: Prefer doing 'switch &%.*s in ...'\n", LIT(e->token.string)); - } - } -} - gb_internal void check_unary_expr(CheckerContext *c, Operand *o, Token op, Ast *node) { switch (op.kind) { case Token_And: { // Pointer address @@ -2493,7 +2467,10 @@ gb_internal void check_unary_expr(CheckerContext *c, Operand *o, Token op, Ast * { ERROR_BLOCK(); error(op, "Cannot take the pointer address of '%s'", str); - if (e != nullptr && (e->flags & EntityFlag_ForValue) != 0) { + if (e == nullptr) { + break; + } + if ((e->flags & EntityFlag_ForValue) != 0) { Type *parent_type = type_deref(e->Variable.for_loop_parent_type); if (parent_type != nullptr && is_type_string(parent_type)) { @@ -2503,9 +2480,17 @@ gb_internal void check_unary_expr(CheckerContext *c, Operand *o, Token op, Ast * } else { error_line("\tSuggestion: Did you want to pass the iterable value to the for statement by pointer to get addressable semantics?\n"); } + + if (is_type_map(parent_type)) { + error_line("\t Prefer doing 'for key, &%.*s in ...'\n", LIT(e->token.string)); + } else { + error_line("\t Prefer doing 'for &%.*s in ...'\n", LIT(e->token.string)); + } } - if (e != nullptr && (e->flags & EntityFlag_SwitchValue) != 0) { + if ((e->flags & EntityFlag_SwitchValue) != 0) { error_line("\tSuggestion: Did you want to pass the value to the switch statement by pointer to get addressable semantics?\n"); + + error_line("\t Prefer doing 'switch &%.*s in ...'\n", LIT(e->token.string)); } } break; @@ -2527,11 +2512,6 @@ gb_internal void check_unary_expr(CheckerContext *c, Operand *o, Token op, Ast * o->type = alloc_type_pointer(o->type); } } else { - if (ast_node_expect(node, Ast_UnaryExpr)) { - ast_node(ue, UnaryExpr, node); - check_old_for_or_switch_value_usage(ue->expr); - } - o->type = alloc_type_pointer(o->type); } diff --git a/src/check_stmt.cpp b/src/check_stmt.cpp index 875503874..0267bdf80 100644 --- a/src/check_stmt.cpp +++ b/src/check_stmt.cpp @@ -501,7 +501,6 @@ gb_internal Type *check_assignment_variable(CheckerContext *ctx, Operand *lhs, O return nullptr; case Addressing_Variable: - check_old_for_or_switch_value_usage(lhs->expr); break; case Addressing_MapIndex: { @@ -523,9 +522,8 @@ gb_internal Type *check_assignment_variable(CheckerContext *ctx, Operand *lhs, O break; } - case Addressing_Context: { + case Addressing_Context: break; - } case Addressing_SoaVariable: break; @@ -1328,7 +1326,6 @@ gb_internal void check_type_switch_stmt(CheckerContext *ctx, Ast *node, u32 mod_ } } - bool is_ptr = is_type_pointer(x.type); // NOTE(bill): Check for multiple defaults Ast *first_default = nullptr; @@ -1447,15 +1444,6 @@ gb_internal void check_type_switch_stmt(CheckerContext *ctx, Ast *node, u32 mod_ } bool is_reference = is_addressed; - bool old_style = false; - - if (!is_reference && - is_ptr && - cc->list.count == 1 && - case_type != nullptr) { - is_reference = true; - old_style = true; - } if (cc->list.count > 1 || saw_nil) { case_type = nullptr; @@ -1477,9 +1465,6 @@ gb_internal void check_type_switch_stmt(CheckerContext *ctx, Ast *node, u32 mod_ if (!is_reference) { tag_var->flags |= EntityFlag_Value; } - if (old_style) { - tag_var->flags |= EntityFlag_OldForOrSwitchValue; - } add_entity(ctx, ctx->scope, lhs, tag_var); add_entity_use(ctx, lhs, tag_var); add_implicit_entity(ctx, stmt, tag_var); @@ -1618,7 +1603,6 @@ gb_internal void check_range_stmt(CheckerContext *ctx, Ast *node, u32 mod_flags) auto entities = array_make(temporary_allocator(), 0, 2); bool is_map = false; bool is_bit_set = false; - bool use_by_reference_for_value = false; bool is_soa = false; bool is_reverse = rs->reverse; @@ -1679,7 +1663,6 @@ gb_internal void check_range_stmt(CheckerContext *ctx, Ast *node, u32 mod_flags) } } } - bool is_ptr = is_type_pointer(operand.type); Type *t = base_type(type_deref(operand.type)); switch (t->kind) { @@ -1719,32 +1702,27 @@ gb_internal void check_range_stmt(CheckerContext *ctx, Ast *node, u32 mod_flags) break; case Type_EnumeratedArray: - if (is_ptr) use_by_reference_for_value = true; array_add(&vals, t->EnumeratedArray.elem); array_add(&vals, t->EnumeratedArray.index); break; case Type_Array: - if (is_ptr) use_by_reference_for_value = true; - if (!is_ptr) is_possibly_addressable = operand.mode == Addressing_Variable; + is_possibly_addressable = operand.mode == Addressing_Variable; array_add(&vals, t->Array.elem); array_add(&vals, t_int); break; case Type_DynamicArray: - if (is_ptr) use_by_reference_for_value = true; array_add(&vals, t->DynamicArray.elem); array_add(&vals, t_int); break; case Type_Slice: - if (is_ptr) use_by_reference_for_value = true; array_add(&vals, t->Slice.elem); array_add(&vals, t_int); break; case Type_Map: - if (is_ptr) use_by_reference_for_value = true; is_map = true; array_add(&vals, t->Map.key); array_add(&vals, t->Map.value); @@ -1817,7 +1795,6 @@ gb_internal void check_range_stmt(CheckerContext *ctx, Ast *node, u32 mod_flags) case Type_Struct: if (t->Struct.soa_kind != StructSoa_None) { is_soa = true; - if (is_ptr) use_by_reference_for_value = true; array_add(&vals, t->Struct.soa_elem); array_add(&vals, t_int); } @@ -1894,9 +1871,6 @@ gb_internal void check_range_stmt(CheckerContext *ctx, Ast *node, u32 mod_flags) char const *idx_name = is_map ? "key" : is_bit_set ? "element" : "index"; error(token, "The %s variable '%.*s' cannot be made addressable", idx_name, LIT(str)); } - } else if (i == addressable_index && use_by_reference_for_value) { - entity->flags |= EntityFlag_OldForOrSwitchValue; - entity->flags &= ~EntityFlag_Value; } if (is_soa) { if (i == 0) { diff --git a/src/entity.cpp b/src/entity.cpp index 60ca208ec..8a7417006 100644 --- a/src/entity.cpp +++ b/src/entity.cpp @@ -85,8 +85,6 @@ enum EntityFlag : u64 { EntityFlag_Require = 1ull<<50, EntityFlag_ByPtr = 1ull<<51, // enforce parameter is passed by pointer - EntityFlag_OldForOrSwitchValue = 1ull<<52, - EntityFlag_Overridden = 1ull<<63, }; -- cgit v1.2.3 From a1b8749e74639875467cba56b0ab02c342870338 Mon Sep 17 00:00:00 2001 From: gingerBill Date: Tue, 28 May 2024 00:23:23 +0100 Subject: Delay checking foreign import paths until after global scope is checked --- src/checker.cpp | 168 +++++++++++++++++++++++++++++--------------------------- src/checker.hpp | 3 + src/entity.cpp | 1 + src/parser.cpp | 28 +++++++++- src/parser.hpp | 1 + 5 files changed, 118 insertions(+), 83 deletions(-) (limited to 'src/entity.cpp') diff --git a/src/checker.cpp b/src/checker.cpp index 4e5aed2bf..1ded6ea6e 100644 --- a/src/checker.cpp +++ b/src/checker.cpp @@ -1283,6 +1283,7 @@ gb_internal void init_checker_info(CheckerInfo *i) { mpsc_init(&i->definition_queue, a); //); // 1<<20); mpsc_init(&i->required_global_variable_queue, a); // 1<<10); mpsc_init(&i->required_foreign_imports_through_force_queue, a); // 1<<10); + mpsc_init(&i->foreign_imports_to_check_fullpaths, a); // 1<<10); mpsc_init(&i->intrinsics_entry_point_usage, a); // 1<<10); // just waste some memory here, even if it probably never used string_map_init(&i->load_directory_cache); @@ -1307,6 +1308,7 @@ gb_internal void destroy_checker_info(CheckerInfo *i) { mpsc_destroy(&i->definition_queue); mpsc_destroy(&i->required_global_variable_queue); mpsc_destroy(&i->required_foreign_imports_through_force_queue); + mpsc_destroy(&i->foreign_imports_to_check_fullpaths); map_destroy(&i->objc_msgSend_types); string_map_destroy(&i->load_file_cache); @@ -4874,105 +4876,112 @@ gb_internal DECL_ATTRIBUTE_PROC(foreign_import_decl_attribute) { return false; } -gb_internal void check_add_foreign_import_decl(CheckerContext *ctx, Ast *decl) { - if (decl->state_flags & StateFlag_BeenHandled) return; - decl->state_flags |= StateFlag_BeenHandled; +gb_internal void check_foreign_import_fullpaths(Checker *c) { + CheckerContext ctx = make_checker_context(c); - ast_node(fl, ForeignImportDecl, decl); + UntypedExprInfoMap untyped = {}; + defer (map_destroy(&untyped)); - Scope *parent_scope = ctx->scope; - GB_ASSERT(parent_scope->flags&ScopeFlag_File); + for (Entity *e = nullptr; mpsc_dequeue(&c->info.foreign_imports_to_check_fullpaths, &e); /**/) { + GB_ASSERT(e != nullptr); + GB_ASSERT(e->kind == Entity_LibraryName); + Ast *decl = e->LibraryName.decl; + ast_node(fl, ForeignImportDecl, decl); - String base_dir = dir_from_path(decl->file()->fullpath); + AstFile *f = decl->file(); - auto fullpaths = array_make(permanent_allocator(), 0, fl->filepaths.count); + reset_checker_context(&ctx, f, &untyped); + ctx.collect_delayed_decls = false; - for (Ast *fp_node : fl->filepaths) { - Operand op = {}; - check_expr(ctx, &op, fp_node); - if (op.mode != Addressing_Constant && op.value.kind != ExactValue_String) { - gbString s = expr_to_string(op.expr); - error(fp_node, "Expected a constant string value, got '%s'", s); - gb_string_free(s); - continue; - } - if (!is_type_string(op.type)) { - gbString s = type_to_string(op.type); - error(fp_node, "Expected a constant string value, got value of type '%s'", s); - gb_string_free(s); - continue; - } + GB_ASSERT(ctx.scope == e->scope); - String file_str = op.value.value_string; - file_str = string_trim_whitespace(file_str); + if (fl->fullpaths.count == 0) { + String base_dir = dir_from_path(decl->file()->fullpath); - String fullpath = file_str; - if (allow_check_foreign_filepath()) { - String foreign_path = {}; - bool ok = determine_path_from_string(nullptr, decl, base_dir, file_str, &foreign_path); - gb_unused(ok); - fullpath = foreign_path; - } - array_add(&fullpaths, fullpath); - } - fl->fullpaths = slice_from_array(fullpaths); + auto fullpaths = array_make(permanent_allocator(), 0, fl->filepaths.count); + for (Ast *fp_node : fl->filepaths) { + Operand op = {}; + check_expr(&ctx, &op, fp_node); + if (op.mode != Addressing_Constant && op.value.kind != ExactValue_String) { + gbString s = expr_to_string(op.expr); + error(fp_node, "Expected a constant string value, got '%s'", s); + gb_string_free(s); + continue; + } + if (!is_type_string(op.type)) { + gbString s = type_to_string(op.type); + error(fp_node, "Expected a constant string value, got value of type '%s'", s); + gb_string_free(s); + continue; + } - if (fl->fullpaths.count == 0) { - return; - } - String fullpath = fl->fullpaths[0]; - String library_name = path_to_entity_name(fl->library_name.string, fullpath); - if (is_blank_ident(library_name)) { - error(fl->token, "File name, %.*s, cannot be as a library name as it is not a valid identifier", LIT(fl->library_name.string)); - return; - } + String file_str = op.value.value_string; + file_str = string_trim_whitespace(file_str); - for (String const &path : fl->fullpaths) { - String ext = path_extension(path); - if (str_eq_ignore_case(ext, ".c") || - str_eq_ignore_case(ext, ".cpp") || - str_eq_ignore_case(ext, ".cxx") || - str_eq_ignore_case(ext, ".h") || - str_eq_ignore_case(ext, ".hpp") || - str_eq_ignore_case(ext, ".hxx") || - false - ) { - error(fl->token, "With 'foreign import', you cannot import a %.*s file directory, you must precompile the library and link against that", LIT(ext)); - break; + String fullpath = file_str; + if (allow_check_foreign_filepath()) { + String foreign_path = {}; + bool ok = determine_path_from_string(nullptr, decl, base_dir, file_str, &foreign_path, /*use error not syntax_error*/true); + if (ok) { + fullpath = foreign_path; + } + } + array_add(&fullpaths, fullpath); + } + fl->fullpaths = slice_from_array(fullpaths); + } + + for (String const &path : fl->fullpaths) { + String ext = path_extension(path); + if (str_eq_ignore_case(ext, ".c") || + str_eq_ignore_case(ext, ".cpp") || + str_eq_ignore_case(ext, ".cxx") || + str_eq_ignore_case(ext, ".h") || + str_eq_ignore_case(ext, ".hpp") || + str_eq_ignore_case(ext, ".hxx") || + false + ) { + error(fl->token, "With 'foreign import', you cannot import a %.*s file/directory, you must precompile the library and link against that", LIT(ext)); + break; + } } + + add_untyped_expressions(ctx.info, &untyped); + + e->LibraryName.paths = fl->fullpaths; } +} + +gb_internal void check_add_foreign_import_decl(CheckerContext *ctx, Ast *decl) { + if (decl->state_flags & StateFlag_BeenHandled) return; + decl->state_flags |= StateFlag_BeenHandled; + ast_node(fl, ForeignImportDecl, decl); - // if (fl->collection_name != "system") { - // char *c_str = gb_alloc_array(heap_allocator(), char, fullpath.len+1); - // defer (gb_free(heap_allocator(), c_str)); - // gb_memmove(c_str, fullpath.text, fullpath.len); - // c_str[fullpath.len] = '\0'; + Scope *parent_scope = ctx->scope; + GB_ASSERT(parent_scope->flags&ScopeFlag_File); - // gbFile f = {}; - // gbFileError file_err = gb_file_open(&f, c_str); - // defer (gb_file_close(&f)); + String library_name = fl->library_name.string; + if (library_name.len == 0 && fl->fullpaths.count != 0) { + String fullpath = fl->fullpaths[0]; + library_name = path_to_entity_name(fl->library_name.string, fullpath); + } + if (library_name.len == 0 || is_blank_ident(library_name)) { + error(fl->token, "File name, '%.*s', cannot be as a library name as it is not a valid identifier", LIT(library_name)); + return; + } - // switch (file_err) { - // case gbFileError_Invalid: - // error(decl, "Invalid file or cannot be found ('%.*s')", LIT(fullpath)); - // return; - // case gbFileError_NotExists: - // error(decl, "File cannot be found ('%.*s')", LIT(fullpath)); - // return; - // } - // } GB_ASSERT(fl->library_name.pos.line != 0); fl->library_name.string = library_name; Entity *e = alloc_entity_library_name(parent_scope, fl->library_name, t_invalid, fl->fullpaths, library_name); + e->LibraryName.decl = decl; add_entity_flags_from_file(ctx, e, parent_scope); add_entity(ctx, parent_scope, nullptr, e); - AttributeContext ac = {}; check_decl_attributes(ctx, fl->attributes, foreign_import_decl_attribute, &ac); if (ac.require_declaration) { @@ -4987,12 +4996,8 @@ gb_internal void check_add_foreign_import_decl(CheckerContext *ctx, Ast *decl) { e->LibraryName.extra_linker_flags = extra_linker_flags; } - if (has_asm_extension(fullpath)) { - if (build_context.metrics.arch != TargetArch_amd64 && build_context.metrics.os != TargetOs_darwin) { - error(decl, "Assembly files are not yet supported on this platform: %.*s_%.*s", - LIT(target_os_names[build_context.metrics.os]), LIT(target_arch_names[build_context.metrics.arch])); - } - } + mpsc_enqueue(&ctx->info->foreign_imports_to_check_fullpaths, e); + } // Returns true if a new package is present @@ -6354,6 +6359,9 @@ gb_internal void check_parsed_files(Checker *c) { TIME_SECTION("check procedure bodies"); check_procedure_bodies(c); + TIME_SECTION("check foreign import fullpaths"); + check_foreign_import_fullpaths(c); + TIME_SECTION("add entities from procedure bodies"); check_merge_queues_into_arrays(c); diff --git a/src/checker.hpp b/src/checker.hpp index 2ade9312e..6ae7b90e2 100644 --- a/src/checker.hpp +++ b/src/checker.hpp @@ -414,6 +414,7 @@ struct CheckerInfo { MPSCQueue entity_queue; MPSCQueue required_global_variable_queue; MPSCQueue required_foreign_imports_through_force_queue; + MPSCQueue foreign_imports_to_check_fullpaths; MPSCQueue intrinsics_entry_point_usage; @@ -434,6 +435,8 @@ struct CheckerInfo { BlockingMutex load_directory_mutex; StringMap load_directory_cache; PtrMap load_directory_map; // Key: Ast_CallExpr * + + }; struct CheckerContext { diff --git a/src/entity.cpp b/src/entity.cpp index 8a7417006..1461b96d7 100644 --- a/src/entity.cpp +++ b/src/entity.cpp @@ -266,6 +266,7 @@ struct Entity { Scope *scope; } ImportName; struct { + Ast *decl; Slice paths; String name; i64 priority_index; diff --git a/src/parser.cpp b/src/parser.cpp index be0d68177..7e72f3c21 100644 --- a/src/parser.cpp +++ b/src/parser.cpp @@ -1285,13 +1285,15 @@ gb_internal Ast *ast_import_decl(AstFile *f, Token token, Token relpath, Token i } gb_internal Ast *ast_foreign_import_decl(AstFile *f, Token token, Array filepaths, Token library_name, - CommentGroup *docs, CommentGroup *comment) { + bool multiple_filepaths, + CommentGroup *docs, CommentGroup *comment) { Ast *result = alloc_ast_node(f, Ast_ForeignImportDecl); result->ForeignImportDecl.token = token; result->ForeignImportDecl.filepaths = slice_from_array(filepaths); result->ForeignImportDecl.library_name = library_name; result->ForeignImportDecl.docs = docs; result->ForeignImportDecl.comment = comment; + result->ForeignImportDecl.multiple_filepaths = multiple_filepaths; result->ForeignImportDecl.attributes.allocator = ast_allocator(f); return result; @@ -4882,8 +4884,11 @@ gb_internal Ast *parse_foreign_decl(AstFile *f) { if (is_blank_ident(lib_name)) { syntax_error(lib_name, "Illegal foreign import name: '_'"); } + bool multiple_filepaths = false; + Array filepaths = {}; if (allow_token(f, Token_OpenBrace)) { + multiple_filepaths = true; array_init(&filepaths, ast_allocator(f)); while (f->curr_token.kind != Token_CloseBrace && @@ -4912,7 +4917,7 @@ gb_internal Ast *parse_foreign_decl(AstFile *f) { syntax_error(lib_name, "You cannot use foreign import within a procedure. This must be done at the file scope"); s = ast_bad_decl(f, lib_name, ast_token(filepaths[0])); } else { - s = ast_foreign_import_decl(f, token, filepaths, lib_name, docs, f->line_comment); + s = ast_foreign_import_decl(f, token, filepaths, lib_name, multiple_filepaths, docs, f->line_comment); } expect_semicolon(f); return s; @@ -5859,7 +5864,24 @@ gb_internal void parse_setup_file_decls(Parser *p, AstFile *f, String const &bas syntax_error(decls[i], "No foreign paths found"); decls[i] = ast_bad_decl(f, ast_token(fl->filepaths[0]), ast_end_token(fl->filepaths[fl->filepaths.count-1])); goto end; - + } else if (!fl->multiple_filepaths && + fl->filepaths.count == 1) { + Ast *fp = fl->filepaths[0]; + GB_ASSERT(fp->kind == Ast_BasicLit); + Token fp_token = fp->BasicLit.token; + String file_str = string_trim_whitespace(string_value_from_token(f, fp_token)); + String fullpath = file_str; + if (allow_check_foreign_filepath()) { + String foreign_path = {}; + bool ok = determine_path_from_string(&p->file_decl_mutex, node, base_dir, file_str, &foreign_path); + if (!ok) { + decls[i] = ast_bad_decl(f, fp_token, fp_token); + goto end; + } + fullpath = foreign_path; + } + fl->fullpaths = slice_make(permanent_allocator(), 1); + fl->fullpaths[0] = fullpath; } } else if (node->kind == Ast_WhenStmt) { diff --git a/src/parser.hpp b/src/parser.hpp index 1e07cfd59..0e411d9ac 100644 --- a/src/parser.hpp +++ b/src/parser.hpp @@ -632,6 +632,7 @@ AST_KIND(_DeclBegin, "", bool) \ AST_KIND(ForeignImportDecl, "foreign import declaration", struct { \ Token token; \ Slice filepaths; \ + bool multiple_filepaths; \ Token library_name; \ String collection_name; \ Slice fullpaths; \ -- cgit v1.2.3 From 66acbb7fed88ed9132dfc8107865e0ac27ed3ac8 Mon Sep 17 00:00:00 2001 From: gingerBill Date: Thu, 30 May 2024 21:48:23 +0100 Subject: Add `@(link_suffix=)` --- src/check_decl.cpp | 13 +++++++------ src/check_stmt.cpp | 4 ++-- src/checker.cpp | 46 ++++++++++++++++++++++++++++++++++++++++++++++ src/checker.hpp | 5 ++++- src/entity.cpp | 2 ++ 5 files changed, 61 insertions(+), 9 deletions(-) (limited to 'src/entity.cpp') diff --git a/src/check_decl.cpp b/src/check_decl.cpp index 1ec366ae7..44b06d712 100644 --- a/src/check_decl.cpp +++ b/src/check_decl.cpp @@ -724,15 +724,16 @@ gb_internal Entity *init_entity_foreign_library(CheckerContext *ctx, Entity *e) return nullptr; } -gb_internal String handle_link_name(CheckerContext *ctx, Token token, String link_name, String link_prefix) { +gb_internal String handle_link_name(CheckerContext *ctx, Token token, String link_name, String link_prefix, String link_suffix) { if (link_prefix.len > 0) { if (link_name.len > 0) { error(token, "'link_name' and 'link_prefix' cannot be used together"); } else { - isize len = link_prefix.len + token.string.len; + isize len = link_prefix.len + token.string.len + link_suffix.len; u8 *name = gb_alloc_array(permanent_allocator(), u8, len+1); gb_memmove(name, &link_prefix[0], link_prefix.len); gb_memmove(name+link_prefix.len, &token.string[0], token.string.len); + gb_memmove(name+link_prefix.len+token.string.len, link_suffix.text, link_suffix.len); name[len] = 0; link_name = make_string(name, len); @@ -862,7 +863,7 @@ gb_internal void check_proc_decl(CheckerContext *ctx, Entity *e, DeclInfo *d) { } TypeProc *pt = &proc_type->Proc; - AttributeContext ac = make_attribute_context(e->Procedure.link_prefix); + AttributeContext ac = make_attribute_context(e->Procedure.link_prefix, e->Procedure.link_suffix); if (d != nullptr) { check_decl_attributes(ctx, d->attributes, proc_decl_attribute, &ac); @@ -1015,7 +1016,7 @@ gb_internal void check_proc_decl(CheckerContext *ctx, Entity *e, DeclInfo *d) { e->deprecated_message = ac.deprecated_message; e->warning_message = ac.warning_message; - ac.link_name = handle_link_name(ctx, e->token, ac.link_name, ac.link_prefix); + ac.link_name = handle_link_name(ctx, e->token, ac.link_name, ac.link_prefix,ac.link_suffix); if (ac.has_disabled_proc) { if (ac.disabled_proc) { e->flags |= EntityFlag_Disabled; @@ -1223,7 +1224,7 @@ gb_internal void check_global_variable_decl(CheckerContext *ctx, Entity *&e, Ast } e->flags |= EntityFlag_Visited; - AttributeContext ac = make_attribute_context(e->Variable.link_prefix); + AttributeContext ac = make_attribute_context(e->Variable.link_prefix, e->Variable.link_suffix); ac.init_expr_list_count = init_expr != nullptr ? 1 : 0; DeclInfo *decl = decl_info_of_entity(e); @@ -1244,7 +1245,7 @@ gb_internal void check_global_variable_decl(CheckerContext *ctx, Entity *&e, Ast if (ac.is_static) { error(e->token, "@(static) is not supported for global variables, nor required"); } - ac.link_name = handle_link_name(ctx, e->token, ac.link_name, ac.link_prefix); + ac.link_name = handle_link_name(ctx, e->token, ac.link_name, ac.link_prefix, ac.link_suffix); if (is_arch_wasm() && e->Variable.thread_local_model.len != 0) { e->Variable.thread_local_model.len = 0; diff --git a/src/check_stmt.cpp b/src/check_stmt.cpp index 866cdb5a1..2c37bced0 100644 --- a/src/check_stmt.cpp +++ b/src/check_stmt.cpp @@ -2020,7 +2020,7 @@ gb_internal void check_value_decl_stmt(CheckerContext *ctx, Ast *node, u32 mod_f // TODO NOTE(bill): This technically checks things multple times - AttributeContext ac = make_attribute_context(ctx->foreign_context.link_prefix); + AttributeContext ac = make_attribute_context(ctx->foreign_context.link_prefix, ctx->foreign_context.link_suffix); check_decl_attributes(ctx, vd->attributes, var_decl_attribute, &ac); for (isize i = 0; i < entity_count; i++) { @@ -2037,7 +2037,7 @@ gb_internal void check_value_decl_stmt(CheckerContext *ctx, Ast *node, u32 mod_f e->type = init_type; e->state = EntityState_Resolved; } - ac.link_name = handle_link_name(ctx, e->token, ac.link_name, ac.link_prefix); + ac.link_name = handle_link_name(ctx, e->token, ac.link_name, ac.link_prefix, ac.link_suffix); if (ac.link_name.len > 0) { e->Variable.link_name = ac.link_name; diff --git a/src/checker.cpp b/src/checker.cpp index 1ded6ea6e..ec58b9d8e 100644 --- a/src/checker.cpp +++ b/src/checker.cpp @@ -3127,6 +3127,18 @@ gb_internal DECL_ATTRIBUTE_PROC(foreign_block_decl_attribute) { error(elem, "Expected a string value for '%.*s'", LIT(name)); } return true; + } else if (name == "link_suffix") { + if (ev.kind == ExactValue_String) { + String link_suffix = ev.value_string; + if (!is_foreign_name_valid(link_suffix)) { + error(elem, "Invalid link suffix: '%.*s'", LIT(link_suffix)); + } else { + c->foreign_context.link_suffix = link_suffix; + } + } else { + error(elem, "Expected a string value for '%.*s'", LIT(name)); + } + return true; } else if (name == "private") { EntityVisiblityKind kind = EntityVisiblity_PrivateToPackage; if (ev.kind == ExactValue_Invalid) { @@ -3421,6 +3433,18 @@ gb_internal DECL_ATTRIBUTE_PROC(proc_decl_attribute) { error(elem, "Expected a string value for '%.*s'", LIT(name)); } return true; + } else if (name == "link_suffix") { + ExactValue ev = check_decl_attribute_value(c, value); + + if (ev.kind == ExactValue_String) { + ac->link_suffix = ev.value_string; + if (!is_foreign_name_valid(ac->link_suffix)) { + error(elem, "Invalid link suffix: %.*s", LIT(ac->link_suffix)); + } + } else { + error(elem, "Expected a string value for '%.*s'", LIT(name)); + } + return true; } else if (name == "deprecated") { ExactValue ev = check_decl_attribute_value(c, value); @@ -3702,6 +3726,17 @@ gb_internal DECL_ATTRIBUTE_PROC(var_decl_attribute) { error(elem, "Expected a string value for '%.*s'", LIT(name)); } return true; + } else if (name == "link_suffix") { + ExactValue ev = check_decl_attribute_value(c, value); + if (ev.kind == ExactValue_String) { + ac->link_suffix = ev.value_string; + if (!is_foreign_name_valid(ac->link_suffix)) { + error(elem, "Invalid link suffix: %.*s", LIT(ac->link_suffix)); + } + } else { + error(elem, "Expected a string value for '%.*s'", LIT(name)); + } + return true; } else if (name == "link_section") { ExactValue ev = check_decl_attribute_value(c, value); if (ev.kind == ExactValue_String) { @@ -3733,6 +3768,7 @@ gb_internal DECL_ATTRIBUTE_PROC(const_decl_attribute) { name == "linkage" || name == "link_name" || name == "link_prefix" || + name == "link_suffix" || false) { error(elem, "@(%.*s) is not supported for compile time constant value declarations", LIT(name)); return true; @@ -3775,8 +3811,10 @@ gb_internal void check_decl_attributes(CheckerContext *c, Array const &at if (attributes.count == 0) return; String original_link_prefix = {}; + String original_link_suffix = {}; if (ac) { original_link_prefix = ac->link_prefix; + original_link_suffix = ac->link_suffix; } StringSet set = {}; @@ -3851,6 +3889,12 @@ gb_internal void check_decl_attributes(CheckerContext *c, Array const &at ac->link_prefix.len = 0; } } + if (ac->link_suffix.text == original_link_suffix.text) { + if (ac->link_name.len > 0) { + ac->link_suffix.text = nullptr; + ac->link_suffix.len = 0; + } + } } } @@ -4145,6 +4189,7 @@ gb_internal void check_collect_value_decl(CheckerContext *c, Ast *decl) { e->Variable.foreign_library_ident = fl; e->Variable.link_prefix = c->foreign_context.link_prefix; + e->Variable.link_suffix = c->foreign_context.link_suffix; } Ast *init_expr = value; @@ -4219,6 +4264,7 @@ gb_internal void check_collect_value_decl(CheckerContext *c, Ast *decl) { } } e->Procedure.link_prefix = c->foreign_context.link_prefix; + e->Procedure.link_suffix = c->foreign_context.link_suffix; GB_ASSERT(cc != ProcCC_Invalid); pl->type->ProcType.calling_convention = cc; diff --git a/src/checker.hpp b/src/checker.hpp index 6ae7b90e2..539b72b2d 100644 --- a/src/checker.hpp +++ b/src/checker.hpp @@ -112,6 +112,7 @@ enum InstrumentationFlag : i32 { struct AttributeContext { String link_name; String link_prefix; + String link_suffix; String link_section; String linkage; isize init_expr_list_count; @@ -146,9 +147,10 @@ struct AttributeContext { String enable_target_feature; // will be enabled for the procedure only }; -gb_internal gb_inline AttributeContext make_attribute_context(String link_prefix) { +gb_internal gb_inline AttributeContext make_attribute_context(String link_prefix, String link_suffix) { AttributeContext ac = {}; ac.link_prefix = link_prefix; + ac.link_suffix = link_suffix; return ac; } @@ -302,6 +304,7 @@ struct ForeignContext { Ast * curr_library; ProcCallingConvention default_cc; String link_prefix; + String link_suffix; EntityVisiblityKind visibility_kind; }; diff --git a/src/entity.cpp b/src/entity.cpp index 1461b96d7..e4fc66dac 100644 --- a/src/entity.cpp +++ b/src/entity.cpp @@ -223,6 +223,7 @@ struct Entity { Ast * foreign_library_ident; String link_name; String link_prefix; + String link_suffix; String link_section; CommentGroup *docs; CommentGroup *comment; @@ -243,6 +244,7 @@ struct Entity { Ast * foreign_library_ident; String link_name; String link_prefix; + String link_suffix; DeferredProcedure deferred_procedure; struct GenProcsData *gen_procs; -- cgit v1.2.3 From 9ef43fc782159893b7af139f9d9be3aec3108ecd Mon Sep 17 00:00:00 2001 From: gingerBill Date: Thu, 6 Jun 2024 15:16:34 +0100 Subject: Add `@(rodata)` --- src/check_decl.cpp | 6 ++++++ src/check_stmt.cpp | 6 ++++++ src/checker.cpp | 6 ++++++ src/checker.hpp | 1 + src/entity.cpp | 1 + src/llvm_backend.cpp | 20 ++++++++++++++++++-- 6 files changed, 38 insertions(+), 2 deletions(-) (limited to 'src/entity.cpp') diff --git a/src/check_decl.cpp b/src/check_decl.cpp index f2afce59c..43947836b 100644 --- a/src/check_decl.cpp +++ b/src/check_decl.cpp @@ -1264,6 +1264,9 @@ gb_internal void check_global_variable_decl(CheckerContext *ctx, Entity *&e, Ast if (ac.is_static) { error(e->token, "@(static) is not supported for global variables, nor required"); } + if (ac.rodata) { + e->Variable.is_rodata = true; + } ac.link_name = handle_link_name(ctx, e->token, ac.link_name, ac.link_prefix, ac.link_suffix); if (is_arch_wasm() && e->Variable.thread_local_model.len != 0) { @@ -1350,6 +1353,9 @@ gb_internal void check_global_variable_decl(CheckerContext *ctx, Entity *&e, Ast Operand o = {}; check_expr_with_type_hint(ctx, &o, init_expr, e->type); check_init_variable(ctx, e, &o, str_lit("variable declaration")); + if (e->Variable.is_rodata && o.mode != Addressing_Constant) { + error(o.expr, "Variables declared with @(rodata) must have constant initialization"); + } check_rtti_type_disallowed(e->token, e->type, "A variable declaration is using a type, %s, which has been disallowed"); } diff --git a/src/check_stmt.cpp b/src/check_stmt.cpp index 2c37bced0..fc443a7b5 100644 --- a/src/check_stmt.cpp +++ b/src/check_stmt.cpp @@ -501,6 +501,9 @@ gb_internal Type *check_assignment_variable(CheckerContext *ctx, Operand *lhs, O return nullptr; case Addressing_Variable: + if (e && e->kind == Entity_Variable && e->Variable.is_rodata) { + error(lhs->expr, "Assignment to variable '%.*s' marked as @(rodata) is not allowed", LIT(e->token.string)); + } break; case Addressing_MapIndex: { @@ -2055,6 +2058,9 @@ gb_internal void check_value_decl_stmt(CheckerContext *ctx, Ast *node, u32 mod_f } } } + if (ac.rodata) { + error(e->token, "Only global variables can have @(rodata) applied"); + } if (ac.thread_local_model != "") { String name = e->token.string; if (name == "_") { diff --git a/src/checker.cpp b/src/checker.cpp index 97e685d33..8a58bb425 100644 --- a/src/checker.cpp +++ b/src/checker.cpp @@ -3628,6 +3628,12 @@ gb_internal DECL_ATTRIBUTE_PROC(var_decl_attribute) { } ac->is_static = true; return true; + } else if (name == "rodata") { + if (value != nullptr) { + error(elem, "'rodata' does not have any parameters"); + } + ac->rodata = true; + return true; } else if (name == "thread_local") { ExactValue ev = check_decl_attribute_value(c, value); if (ac->init_expr_list_count > 0) { diff --git a/src/checker.hpp b/src/checker.hpp index e793540e3..2ac4c8e7a 100644 --- a/src/checker.hpp +++ b/src/checker.hpp @@ -133,6 +133,7 @@ struct AttributeContext { bool entry_point_only : 1; bool instrumentation_enter : 1; bool instrumentation_exit : 1; + bool rodata : 1; u32 optimization_mode; // ProcedureOptimizationMode i64 foreign_import_priority_index; String extra_linker_flags; diff --git a/src/entity.cpp b/src/entity.cpp index e4fc66dac..7f484e308 100644 --- a/src/entity.cpp +++ b/src/entity.cpp @@ -230,6 +230,7 @@ struct Entity { bool is_foreign; bool is_export; bool is_global; + bool is_rodata; } Variable; struct { Type * type_parameter_specialization; diff --git a/src/llvm_backend.cpp b/src/llvm_backend.cpp index 03c17a8bb..5dc6d94d5 100644 --- a/src/llvm_backend.cpp +++ b/src/llvm_backend.cpp @@ -1160,6 +1160,10 @@ gb_internal lbProcedure *lb_create_startup_runtime(lbModule *main_module, lbProc if (is_type_untyped_nil(init.type)) { LLVMSetInitializer(var.var.value, LLVMConstNull(global_type)); var.is_initialized = true; + + if (e->Variable.is_rodata) { + LLVMSetGlobalConstant(var.var.value, true); + } continue; } GB_PANIC("Invalid init value, got %s", expr_to_string(init_expr)); @@ -1174,6 +1178,10 @@ gb_internal lbProcedure *lb_create_startup_runtime(lbModule *main_module, lbProc } LLVMSetInitializer(var.var.value, init.value); var.is_initialized = true; + + if (e->Variable.is_rodata) { + LLVMSetGlobalConstant(var.var.value, true); + } continue; } } else { @@ -1206,8 +1214,9 @@ gb_internal lbProcedure *lb_create_startup_runtime(lbModule *main_module, lbProc var.is_initialized = true; } + + } - CheckerInfo *info = main_module->gen->info; for (Entity *e : info->init_procedures) { @@ -3210,14 +3219,21 @@ gb_internal bool lb_generate_code(lbGenerator *gen) { lbValue init = lb_const_value(m, tav.type, v); LLVMSetInitializer(g.value, init.value); var.is_initialized = true; + if (e->kind == Entity_Variable && e->Variable.is_rodata) { + LLVMSetGlobalConstant(g.value, true); + } } } } if (!var.is_initialized && is_type_untyped_nil(tav.type)) { var.is_initialized = true; + if (e->kind == Entity_Variable && e->Variable.is_rodata) { + LLVMSetGlobalConstant(g.value, true); + } } + } else if (e->kind == Entity_Variable && e->Variable.is_rodata) { + LLVMSetGlobalConstant(g.value, true); } - array_add(&global_variables, var); lb_add_entity(m, e, g); -- cgit v1.2.3 From 0b02c67cdf316d55232f0433d51f6ea781d74a22 Mon Sep 17 00:00:00 2001 From: gingerBill Date: Tue, 11 Jun 2024 12:19:52 +0100 Subject: Minor clean up for backend --- src/check_decl.cpp | 9 ++- src/entity.cpp | 1 + src/llvm_backend.cpp | 130 +++++++++++++++++++++++++------------------ src/llvm_backend.hpp | 3 +- src/llvm_backend_general.cpp | 3 +- 5 files changed, 89 insertions(+), 57 deletions(-) (limited to 'src/entity.cpp') diff --git a/src/check_decl.cpp b/src/check_decl.cpp index 13b14149a..a5c9119ea 100644 --- a/src/check_decl.cpp +++ b/src/check_decl.cpp @@ -1142,7 +1142,14 @@ gb_internal void check_proc_decl(CheckerContext *ctx, Entity *e, DeclInfo *d) { } if (ac.link_name.len > 0) { - e->Procedure.link_name = ac.link_name; + String ln = ac.link_name; + e->Procedure.link_name = ln; + if (ln == "memcpy" || + ln == "memmove" || + ln == "mem_copy" || + ln == "mem_copy_non_overlapping") { + e->Procedure.is_memcpy_like = true; + } } if (ac.deferred_procedure.entity != nullptr) { diff --git a/src/entity.cpp b/src/entity.cpp index 7f484e308..8f55c1faf 100644 --- a/src/entity.cpp +++ b/src/entity.cpp @@ -256,6 +256,7 @@ struct Entity { bool generated_from_polymorphic : 1; bool entry_point_only : 1; bool has_instrumentation : 1; + bool is_memcpy_like : 1; } Procedure; struct { Array entities; diff --git a/src/llvm_backend.cpp b/src/llvm_backend.cpp index 01680ffa9..81de2c224 100644 --- a/src/llvm_backend.cpp +++ b/src/llvm_backend.cpp @@ -1300,18 +1300,14 @@ gb_internal lbProcedure *lb_create_cleanup_runtime(lbModule *main_module) { // C gb_internal WORKER_TASK_PROC(lb_generate_procedures_and_types_per_module) { lbModule *m = cast(lbModule *)data; - for (Entity *e : m->global_procedures_and_types_to_create) { - if (e->kind == Entity_TypeName) { - (void)lb_get_entity_name(m, e); - lb_type(m, e->type); - } + for (Entity *e : m->global_types_to_create) { + (void)lb_get_entity_name(m, e); + (void)lb_type(m, e->type); } - for (Entity *e : m->global_procedures_and_types_to_create) { - if (e->kind == Entity_Procedure) { - (void)lb_get_entity_name(m, e); - array_add(&m->procedures_to_generate, lb_create_procedure(m, e)); - } + for (Entity *e : m->global_procedures_to_create) { + (void)lb_get_entity_name(m, e); + array_add(&m->procedures_to_generate, lb_create_procedure(m, e)); } return 0; } @@ -1365,16 +1361,24 @@ gb_internal void lb_create_global_procedures_and_types(lbGenerator *gen, Checker m = lb_module_of_entity(gen, e); } - array_add(&m->global_procedures_and_types_to_create, e); + if (e->kind == Entity_Procedure) { + array_add(&m->global_procedures_to_create, e); + } else if (e->kind == Entity_TypeName) { + array_add(&m->global_types_to_create, e); + } } - for (auto const &entry : gen->modules) { - lbModule *m = entry.value; - if (do_threading) { + if (do_threading) { + for (auto const &entry : gen->modules) { + lbModule *m = entry.value; thread_pool_add_task(lb_generate_procedures_and_types_per_module, m); - } else { + } + } else { + for (auto const &entry : gen->modules) { + lbModule *m = entry.value; lb_generate_procedures_and_types_per_module(m); } + } thread_pool_wait(); @@ -2405,16 +2409,19 @@ gb_internal WORKER_TASK_PROC(lb_generate_procedures_worker_proc) { } gb_internal void lb_generate_procedures(lbGenerator *gen, bool do_threading) { - for (auto const &entry : gen->modules) { - lbModule *m = entry.value; - if (do_threading) { + if (do_threading) { + for (auto const &entry : gen->modules) { + lbModule *m = entry.value; thread_pool_add_task(lb_generate_procedures_worker_proc, m); - } else { + } + + thread_pool_wait(); + } else { + for (auto const &entry : gen->modules) { + lbModule *m = entry.value; lb_generate_procedures_worker_proc(m); } } - - thread_pool_wait(); } gb_internal WORKER_TASK_PROC(lb_generate_missing_procedures_to_check_worker_proc) { @@ -2428,17 +2435,20 @@ gb_internal WORKER_TASK_PROC(lb_generate_missing_procedures_to_check_worker_proc } gb_internal void lb_generate_missing_procedures(lbGenerator *gen, bool do_threading) { - for (auto const &entry : gen->modules) { - lbModule *m = entry.value; - // NOTE(bill): procedures may be added during generation - if (do_threading) { + if (do_threading) { + for (auto const &entry : gen->modules) { + lbModule *m = entry.value; + // NOTE(bill): procedures may be added during generation thread_pool_add_task(lb_generate_missing_procedures_to_check_worker_proc, m); - } else { + } + thread_pool_wait(); + } else { + for (auto const &entry : gen->modules) { + lbModule *m = entry.value; + // NOTE(bill): procedures may be added during generation lb_generate_missing_procedures_to_check_worker_proc(m); } } - - thread_pool_wait(); } gb_internal void lb_debug_info_complete_types_and_finalize(lbGenerator *gen) { @@ -2451,32 +2461,45 @@ gb_internal void lb_debug_info_complete_types_and_finalize(lbGenerator *gen) { } gb_internal void lb_llvm_function_passes(lbGenerator *gen, bool do_threading) { - for (auto const &entry : gen->modules) { - lbModule *m = entry.value; - if (do_threading) { + if (do_threading) { + for (auto const &entry : gen->modules) { + lbModule *m = entry.value; thread_pool_add_task(lb_llvm_function_pass_per_module, m); - } else { + } + thread_pool_wait(); + } else { + for (auto const &entry : gen->modules) { + lbModule *m = entry.value; lb_llvm_function_pass_per_module(m); } } - thread_pool_wait(); } gb_internal void lb_llvm_module_passes(lbGenerator *gen, bool do_threading) { - for (auto const &entry : gen->modules) { - lbModule *m = entry.value; - auto wd = gb_alloc_item(permanent_allocator(), lbLLVMModulePassWorkerData); - wd->m = m; - wd->target_machine = m->target_machine; + if (do_threading) { + for (auto const &entry : gen->modules) { + lbModule *m = entry.value; + auto wd = gb_alloc_item(permanent_allocator(), lbLLVMModulePassWorkerData); + wd->m = m; + wd->target_machine = m->target_machine; - if (do_threading) { - thread_pool_add_task(lb_llvm_module_pass_worker_proc, wd); - } else { + if (do_threading) { + thread_pool_add_task(lb_llvm_module_pass_worker_proc, wd); + } else { + lb_llvm_module_pass_worker_proc(wd); + } + } + thread_pool_wait(); + } else { + for (auto const &entry : gen->modules) { + lbModule *m = entry.value; + auto wd = gb_alloc_item(permanent_allocator(), lbLLVMModulePassWorkerData); + wd->m = m; + wd->target_machine = m->target_machine; lb_llvm_module_pass_worker_proc(wd); } } - thread_pool_wait(); } gb_internal String lb_filepath_ll_for_module(lbModule *m) { @@ -2556,17 +2579,21 @@ gb_internal String lb_filepath_obj_for_module(lbModule *m) { gb_internal bool lb_llvm_module_verification(lbGenerator *gen, bool do_threading) { - for (auto const &entry : gen->modules) { - lbModule *m = entry.value; - if (do_threading) { + if (do_threading) { + for (auto const &entry : gen->modules) { + lbModule *m = entry.value; thread_pool_add_task(lb_llvm_module_verification_worker_proc, m); - } else { + } + thread_pool_wait(); + + } else { + for (auto const &entry : gen->modules) { + lbModule *m = entry.value; if (lb_llvm_module_verification_worker_proc(m)) { return false; } } } - thread_pool_wait(); return true; } @@ -2808,13 +2835,8 @@ gb_internal void lb_generate_procedure(lbModule *m, lbProcedure *p) { lb_end_procedure(p); // Add Flags - if (p->body != nullptr) { - if (p->name == "memcpy" || p->name == "memmove" || - p->name == "runtime.mem_copy" || p->name == "mem_copy_non_overlapping" || - string_starts_with(p->name, str_lit("llvm.memcpy")) || - string_starts_with(p->name, str_lit("llvm.memmove"))) { - p->flags |= lbProcedureFlag_WithoutMemcpyPass; - } + if (p->entity && p->entity->kind == Entity_Procedure && p->entity->Procedure.is_memcpy_like) { + p->flags |= lbProcedureFlag_WithoutMemcpyPass; } lb_verify_function(m, p, true); diff --git a/src/llvm_backend.hpp b/src/llvm_backend.hpp index 9f7bc8843..447e93d42 100644 --- a/src/llvm_backend.hpp +++ b/src/llvm_backend.hpp @@ -181,7 +181,8 @@ struct lbModule { std::atomic nested_type_name_guid; Array procedures_to_generate; - Array global_procedures_and_types_to_create; + Array global_procedures_to_create; + Array global_types_to_create; lbProcedure *curr_procedure; diff --git a/src/llvm_backend_general.cpp b/src/llvm_backend_general.cpp index ea98fc60a..03d0f8b32 100644 --- a/src/llvm_backend_general.cpp +++ b/src/llvm_backend_general.cpp @@ -78,7 +78,8 @@ gb_internal void lb_init_module(lbModule *m, Checker *c) { array_init(&m->procedures_to_generate, a, 0, c->info.all_procedures.count); map_init(&m->procedure_values, c->info.all_procedures.count*2); } - array_init(&m->global_procedures_and_types_to_create, a, 0, 1024); + array_init(&m->global_procedures_to_create, a, 0, 1024); + array_init(&m->global_types_to_create, a, 0, 1024); array_init(&m->missing_procedures_to_check, a, 0, 16); map_init(&m->debug_values); -- cgit v1.2.3 From 2a219fa8305f2d246d5b3ae5b2b1e25c75bb4332 Mon Sep 17 00:00:00 2001 From: gingerBill Date: Mon, 8 Jul 2024 15:13:40 +0100 Subject: Correct `-use-separate-module` behaviour --- src/entity.cpp | 3 +++ src/llvm_backend.cpp | 2 +- src/llvm_backend_general.cpp | 3 ++- src/main.cpp | 7 ++++++- src/timings.cpp | 7 ++++--- 5 files changed, 16 insertions(+), 6 deletions(-) (limited to 'src/entity.cpp') diff --git a/src/entity.cpp b/src/entity.cpp index 8f55c1faf..9dee9cbe7 100644 --- a/src/entity.cpp +++ b/src/entity.cpp @@ -338,6 +338,9 @@ gb_internal Entity *alloc_entity(EntityKind kind, Scope *scope, Token token, Typ entity->token = token; entity->type = type; entity->id = 1 + global_entity_id.fetch_add(1); + if (token.pos.file_id) { + entity->file = thread_safe_get_ast_file_from_id(token.pos.file_id); + } return entity; } diff --git a/src/llvm_backend.cpp b/src/llvm_backend.cpp index 8c82d7117..9818c5435 100644 --- a/src/llvm_backend.cpp +++ b/src/llvm_backend.cpp @@ -3439,7 +3439,7 @@ gb_internal bool lb_generate_code(lbGenerator *gen) { gbString label_object_generation = gb_string_make(heap_allocator(), "LLVM Object Generation"); if (gen->modules.count > 1) { - label_object_generation = gb_string_append_fmt(label_object_generation, " (%d modules)", gen->modules.count); + label_object_generation = gb_string_append_fmt(label_object_generation, " (%td modules)", gen->modules.count); } TIME_SECTION_WITH_LEN(label_object_generation, gb_string_length(label_object_generation)); diff --git a/src/llvm_backend_general.cpp b/src/llvm_backend_general.cpp index f65bc9ac7..77f78a24c 100644 --- a/src/llvm_backend_general.cpp +++ b/src/llvm_backend_general.cpp @@ -118,6 +118,7 @@ gb_internal bool lb_init_generator(lbGenerator *gen, Checker *c) { map_init(&gen->anonymous_proc_lits, 1024); if (USE_SEPARATE_MODULES) { + bool module_per_file = build_context.module_per_file && build_context.optimization_level <= 0; for (auto const &entry : gen->info->packages) { AstPackage *pkg = entry.value; auto m = gb_alloc_item(permanent_allocator(), lbModule); @@ -125,7 +126,7 @@ gb_internal bool lb_init_generator(lbGenerator *gen, Checker *c) { m->gen = gen; map_set(&gen->modules, cast(void *)pkg, m); lb_init_module(m, c); - if (build_context.module_per_file) { + if (module_per_file) { // NOTE(bill): Probably per file is not a good idea, so leave this for later for (AstFile *file : pkg->files) { auto m = gb_alloc_item(permanent_allocator(), lbModule); diff --git a/src/main.cpp b/src/main.cpp index 8ea0ae62c..89ce93b18 100644 --- a/src/main.cpp +++ b/src/main.cpp @@ -3304,11 +3304,16 @@ int main(int arg_count, char const **arg_ptr) { } else #endif { - MAIN_TIME_SECTION("LLVM API Code Gen"); lbGenerator *gen = gb_alloc_item(permanent_allocator(), lbGenerator); if (!lb_init_generator(gen, checker)) { return 1; } + + gbString label_code_gen = gb_string_make(heap_allocator(), "LLVM API Code Gen"); + if (gen->modules.count > 1) { + label_code_gen = gb_string_append_fmt(label_code_gen, " ( %4td modules )", gen->modules.count); + } + MAIN_TIME_SECTION_WITH_LEN(label_code_gen, gb_string_length(label_code_gen)); if (lb_generate_code(gen)) { switch (build_context.build_mode) { case BuildMode_Executable: diff --git a/src/timings.cpp b/src/timings.cpp index 712e804cb..3f8402b36 100644 --- a/src/timings.cpp +++ b/src/timings.cpp @@ -146,9 +146,10 @@ gb_internal f64 time_stamp_as_us(TimeStamp const &ts, u64 freq) { return 1000000.0*time_stamp_as_s(ts, freq); } -#define MAIN_TIME_SECTION(str) do { debugf("[Section] %s\n", str); timings_start_section(&global_timings, str_lit(str)); } while (0) -#define TIME_SECTION(str) do { debugf("[Section] %s\n", str); if (build_context.show_more_timings) timings_start_section(&global_timings, str_lit(str)); } while (0) -#define TIME_SECTION_WITH_LEN(str, len) do { debugf("[Section] %s\n", str); if (build_context.show_more_timings) timings_start_section(&global_timings, make_string((u8 *)str, len)); } while (0) +#define MAIN_TIME_SECTION(str) do { debugf("[Section] %s\n", str); timings_start_section(&global_timings, str_lit(str)); } while (0) +#define MAIN_TIME_SECTION_WITH_LEN(str, len) do { debugf("[Section] %s\n", str); timings_start_section(&global_timings, make_string((u8 *)str, len)); } while (0) +#define TIME_SECTION(str) do { debugf("[Section] %s\n", str); if (build_context.show_more_timings) timings_start_section(&global_timings, str_lit(str)); } while (0) +#define TIME_SECTION_WITH_LEN(str, len) do { debugf("[Section] %s\n", str); if (build_context.show_more_timings) timings_start_section(&global_timings, make_string((u8 *)str, len)); } while (0) enum TimingUnit { -- cgit v1.2.3 From 1a20b78633038614635da99b5e634015d4ce7d6e Mon Sep 17 00:00:00 2001 From: Laytan Laats Date: Mon, 8 Jul 2024 21:06:57 +0200 Subject: remove misleading `@(optimization_mode)` values and make "none" inhibit optimizations --- src/check_decl.cpp | 1 - src/checker.cpp | 12 ++++++------ src/entity.cpp | 4 +--- src/llvm_backend.cpp | 17 +++++------------ src/llvm_backend.hpp | 5 ----- src/llvm_backend_general.cpp | 6 ++++++ src/llvm_backend_proc.cpp | 8 +------- 7 files changed, 19 insertions(+), 34 deletions(-) (limited to 'src/entity.cpp') diff --git a/src/check_decl.cpp b/src/check_decl.cpp index 84c3d9ecc..7d81d102d 100644 --- a/src/check_decl.cpp +++ b/src/check_decl.cpp @@ -1030,7 +1030,6 @@ gb_internal void check_proc_decl(CheckerContext *ctx, Entity *e, DeclInfo *d) { switch (e->Procedure.optimization_mode) { case ProcedureOptimizationMode_None: - case ProcedureOptimizationMode_Minimal: if (pl->inlining == ProcInlining_inline) { error(e->token, "#force_inline cannot be used in conjunction with the attribute 'optimization_mode' with neither \"none\" nor \"minimal\""); } diff --git a/src/checker.cpp b/src/checker.cpp index 18b9db6ef..8756cce1a 100644 --- a/src/checker.cpp +++ b/src/checker.cpp @@ -3544,19 +3544,19 @@ gb_internal DECL_ATTRIBUTE_PROC(proc_decl_attribute) { String mode = ev.value_string; if (mode == "none") { ac->optimization_mode = ProcedureOptimizationMode_None; + } else if (mode == "favor_size") { + ac->optimization_mode = ProcedureOptimizationMode_FavorSize; } else if (mode == "minimal") { - ac->optimization_mode = ProcedureOptimizationMode_Minimal; + error(elem, "Invalid optimization_mode 'minimal' for '%.*s', mode has been removed due to confusion, but 'none' has the same behaviour", LIT(name)); } else if (mode == "size") { - ac->optimization_mode = ProcedureOptimizationMode_Size; + error(elem, "Invalid optimization_mode 'size' for '%.*s', mode has been removed due to confusion, but 'favor_size' has the same behaviour", LIT(name)); } else if (mode == "speed") { - ac->optimization_mode = ProcedureOptimizationMode_Speed; + error(elem, "Invalid optimization_mode 'speed' for '%.*s', mode has been removed due to confusion, but 'favor_size' has the same behaviour", LIT(name)); } else { ERROR_BLOCK(); error(elem, "Invalid optimization_mode for '%.*s'. Valid modes:", LIT(name)); error_line("\tnone\n"); - error_line("\tminimal\n"); - error_line("\tsize\n"); - error_line("\tspeed\n"); + error_line("\tfavor_size\n"); } } else { error(elem, "Expected a string for '%.*s'", LIT(name)); diff --git a/src/entity.cpp b/src/entity.cpp index 8f55c1faf..62a190437 100644 --- a/src/entity.cpp +++ b/src/entity.cpp @@ -133,9 +133,7 @@ enum EntityConstantFlags : u32 { enum ProcedureOptimizationMode : u8 { ProcedureOptimizationMode_Default, ProcedureOptimizationMode_None, - ProcedureOptimizationMode_Minimal, - ProcedureOptimizationMode_Size, - ProcedureOptimizationMode_Speed, + ProcedureOptimizationMode_FavorSize, }; diff --git a/src/llvm_backend.cpp b/src/llvm_backend.cpp index 8c82d7117..cc9254269 100644 --- a/src/llvm_backend.cpp +++ b/src/llvm_backend.cpp @@ -1486,10 +1486,6 @@ gb_internal WORKER_TASK_PROC(lb_llvm_function_pass_per_module) { lb_populate_function_pass_manager(m, m->function_pass_managers[lbFunctionPassManager_default], false, build_context.optimization_level); lb_populate_function_pass_manager(m, m->function_pass_managers[lbFunctionPassManager_default_without_memcpy], true, build_context.optimization_level); lb_populate_function_pass_manager_specific(m, m->function_pass_managers[lbFunctionPassManager_none], -1); - lb_populate_function_pass_manager_specific(m, m->function_pass_managers[lbFunctionPassManager_minimal], 0); - lb_populate_function_pass_manager_specific(m, m->function_pass_managers[lbFunctionPassManager_size], 1); - lb_populate_function_pass_manager_specific(m, m->function_pass_managers[lbFunctionPassManager_speed], 2); - lb_populate_function_pass_manager_specific(m, m->function_pass_managers[lbFunctionPassManager_aggressive], 3); for (i32 i = 0; i < lbFunctionPassManager_COUNT; i++) { LLVMFinalizeFunctionPassManager(m->function_pass_managers[i]); @@ -1513,15 +1509,12 @@ gb_internal WORKER_TASK_PROC(lb_llvm_function_pass_per_module) { if (p->entity && p->entity->kind == Entity_Procedure) { switch (p->entity->Procedure.optimization_mode) { case ProcedureOptimizationMode_None: - case ProcedureOptimizationMode_Minimal: - pass_manager_kind = lbFunctionPassManager_minimal; + pass_manager_kind = lbFunctionPassManager_none; + GB_ASSERT(lb_proc_has_attribute(p->module, p->value, "optnone")); + GB_ASSERT(lb_proc_has_attribute(p->module, p->value, "noinline")); break; - case ProcedureOptimizationMode_Size: - pass_manager_kind = lbFunctionPassManager_size; - lb_add_attribute_to_proc(p->module, p->value, "optsize"); - break; - case ProcedureOptimizationMode_Speed: - pass_manager_kind = lbFunctionPassManager_speed; + case ProcedureOptimizationMode_FavorSize: + GB_ASSERT(lb_proc_has_attribute(p->module, p->value, "optsize")); break; } } diff --git a/src/llvm_backend.hpp b/src/llvm_backend.hpp index 447e93d42..022f47857 100644 --- a/src/llvm_backend.hpp +++ b/src/llvm_backend.hpp @@ -134,11 +134,6 @@ enum lbFunctionPassManagerKind { lbFunctionPassManager_default, lbFunctionPassManager_default_without_memcpy, lbFunctionPassManager_none, - lbFunctionPassManager_minimal, - lbFunctionPassManager_size, - lbFunctionPassManager_speed, - lbFunctionPassManager_aggressive, - lbFunctionPassManager_COUNT }; diff --git a/src/llvm_backend_general.cpp b/src/llvm_backend_general.cpp index f65bc9ac7..f8a5e1ebf 100644 --- a/src/llvm_backend_general.cpp +++ b/src/llvm_backend_general.cpp @@ -2522,6 +2522,12 @@ gb_internal void lb_add_proc_attribute_at_index(lbProcedure *p, isize index, cha gb_internal void lb_add_attribute_to_proc(lbModule *m, LLVMValueRef proc_value, char const *name, u64 value=0) { LLVMAddAttributeAtIndex(proc_value, LLVMAttributeIndex_FunctionIndex, lb_create_enum_attribute(m->ctx, name, value)); } + +gb_internal bool lb_proc_has_attribute(lbModule *m, LLVMValueRef proc_value, char const *name) { + LLVMAttributeRef ref = LLVMGetEnumAttributeAtIndex(proc_value, LLVMAttributeIndex_FunctionIndex, LLVMGetEnumAttributeKindForName(name, gb_strlen(name))); + return ref != nullptr; +} + gb_internal void lb_add_attribute_to_proc_with_string(lbModule *m, LLVMValueRef proc_value, String const &name, String const &value) { LLVMAttributeRef attr = lb_create_string_attribute(m->ctx, name, value); LLVMAddAttributeAtIndex(proc_value, LLVMAttributeIndex_FunctionIndex, attr); diff --git a/src/llvm_backend_proc.cpp b/src/llvm_backend_proc.cpp index 68ba4f74e..610c34de2 100644 --- a/src/llvm_backend_proc.cpp +++ b/src/llvm_backend_proc.cpp @@ -163,16 +163,10 @@ gb_internal lbProcedure *lb_create_procedure(lbModule *m, Entity *entity, bool i switch (entity->Procedure.optimization_mode) { case ProcedureOptimizationMode_None: - break; - case ProcedureOptimizationMode_Minimal: lb_add_attribute_to_proc(m, p->value, "optnone"); lb_add_attribute_to_proc(m, p->value, "noinline"); break; - case ProcedureOptimizationMode_Size: - lb_add_attribute_to_proc(m, p->value, "optsize"); - break; - case ProcedureOptimizationMode_Speed: - // TODO(bill): handle this correctly + case ProcedureOptimizationMode_FavorSize: lb_add_attribute_to_proc(m, p->value, "optsize"); break; } -- cgit v1.2.3 From edc793d7c123a38826860ef72684308902a7012c Mon Sep 17 00:00:00 2001 From: gingerBill Date: Sun, 14 Jul 2024 11:39:05 +0100 Subject: Add `#no_capture args: ..T` to reuse the backing array stack memory --- core/fmt/fmt.odin | 58 +++++++++++++++++++++++------------------------ core/fmt/fmt_js.odin | 24 ++++++++++---------- core/fmt/fmt_os.odin | 24 ++++++++++---------- src/check_expr.cpp | 17 ++++++++++++++ src/check_type.cpp | 17 ++++++++++++++ src/checker.cpp | 1 + src/checker.hpp | 7 ++++++ src/entity.cpp | 2 +- src/llvm_backend.hpp | 7 ++++++ src/llvm_backend_proc.cpp | 27 +++++++++++++++++++++- src/parser.cpp | 1 + src/parser.hpp | 8 +++++-- 12 files changed, 136 insertions(+), 57 deletions(-) (limited to 'src/entity.cpp') diff --git a/core/fmt/fmt.odin b/core/fmt/fmt.odin index 234f4afbd..e56211346 100644 --- a/core/fmt/fmt.odin +++ b/core/fmt/fmt.odin @@ -125,7 +125,7 @@ register_user_formatter :: proc(id: typeid, formatter: User_Formatter) -> Regist // Returns: A formatted string. // @(require_results) -aprint :: proc(args: ..any, sep := " ", allocator := context.allocator) -> string { +aprint :: proc(#no_capture args: ..any, sep := " ", allocator := context.allocator) -> string { str: strings.Builder strings.builder_init(&str, allocator) return sbprint(&str, ..args, sep=sep) @@ -141,7 +141,7 @@ aprint :: proc(args: ..any, sep := " ", allocator := context.allocator) -> strin // Returns: A formatted string with a newline character at the end. // @(require_results) -aprintln :: proc(args: ..any, sep := " ", allocator := context.allocator) -> string { +aprintln :: proc(#no_capture args: ..any, sep := " ", allocator := context.allocator) -> string { str: strings.Builder strings.builder_init(&str, allocator) return sbprintln(&str, ..args, sep=sep) @@ -158,7 +158,7 @@ aprintln :: proc(args: ..any, sep := " ", allocator := context.allocator) -> str // Returns: A formatted string. The returned string must be freed accordingly. // @(require_results) -aprintf :: proc(fmt: string, args: ..any, allocator := context.allocator, newline := false) -> string { +aprintf :: proc(fmt: string, #no_capture args: ..any, allocator := context.allocator, newline := false) -> string { str: strings.Builder strings.builder_init(&str, allocator) return sbprintf(&str, fmt, ..args, newline=newline) @@ -174,7 +174,7 @@ aprintf :: proc(fmt: string, args: ..any, allocator := context.allocator, newlin // Returns: A formatted string. The returned string must be freed accordingly. // @(require_results) -aprintfln :: proc(fmt: string, args: ..any, allocator := context.allocator) -> string { +aprintfln :: proc(fmt: string, #no_capture args: ..any, allocator := context.allocator) -> string { return aprintf(fmt, ..args, allocator=allocator, newline=true) } // Creates a formatted string @@ -188,7 +188,7 @@ aprintfln :: proc(fmt: string, args: ..any, allocator := context.allocator) -> s // Returns: A formatted string. // @(require_results) -tprint :: proc(args: ..any, sep := " ") -> string { +tprint :: proc(#no_capture args: ..any, sep := " ") -> string { str: strings.Builder strings.builder_init(&str, context.temp_allocator) return sbprint(&str, ..args, sep=sep) @@ -204,7 +204,7 @@ tprint :: proc(args: ..any, sep := " ") -> string { // Returns: A formatted string with a newline character at the end. // @(require_results) -tprintln :: proc(args: ..any, sep := " ") -> string { +tprintln :: proc(#no_capture args: ..any, sep := " ") -> string { str: strings.Builder strings.builder_init(&str, context.temp_allocator) return sbprintln(&str, ..args, sep=sep) @@ -221,7 +221,7 @@ tprintln :: proc(args: ..any, sep := " ") -> string { // Returns: A formatted string. // @(require_results) -tprintf :: proc(fmt: string, args: ..any, newline := false) -> string { +tprintf :: proc(fmt: string, #no_capture args: ..any, newline := false) -> string { str: strings.Builder strings.builder_init(&str, context.temp_allocator) return sbprintf(&str, fmt, ..args, newline=newline) @@ -237,7 +237,7 @@ tprintf :: proc(fmt: string, args: ..any, newline := false) -> string { // Returns: A formatted string. // @(require_results) -tprintfln :: proc(fmt: string, args: ..any) -> string { +tprintfln :: proc(fmt: string, #no_capture args: ..any) -> string { return tprintf(fmt, ..args, newline=true) } // Creates a formatted string using a supplied buffer as the backing array. Writes into the buffer. @@ -249,7 +249,7 @@ tprintfln :: proc(fmt: string, args: ..any) -> string { // // Returns: A formatted string // -bprint :: proc(buf: []byte, args: ..any, sep := " ") -> string { +bprint :: proc(buf: []byte, #no_capture args: ..any, sep := " ") -> string { sb := strings.builder_from_bytes(buf) return sbprint(&sb, ..args, sep=sep) } @@ -262,7 +262,7 @@ bprint :: proc(buf: []byte, args: ..any, sep := " ") -> string { // // Returns: A formatted string with a newline character at the end // -bprintln :: proc(buf: []byte, args: ..any, sep := " ") -> string { +bprintln :: proc(buf: []byte, #no_capture args: ..any, sep := " ") -> string { sb := strings.builder_from_bytes(buf) return sbprintln(&sb, ..args, sep=sep) } @@ -276,7 +276,7 @@ bprintln :: proc(buf: []byte, args: ..any, sep := " ") -> string { // // Returns: A formatted string // -bprintf :: proc(buf: []byte, fmt: string, args: ..any, newline := false) -> string { +bprintf :: proc(buf: []byte, fmt: string, #no_capture args: ..any, newline := false) -> string { sb := strings.builder_from_bytes(buf) return sbprintf(&sb, fmt, ..args, newline=newline) } @@ -289,7 +289,7 @@ bprintf :: proc(buf: []byte, fmt: string, args: ..any, newline := false) -> stri // // Returns: A formatted string // -bprintfln :: proc(buf: []byte, fmt: string, args: ..any) -> string { +bprintfln :: proc(buf: []byte, fmt: string, #no_capture args: ..any) -> string { return bprintf(buf, fmt, ..args, newline=true) } // Runtime assertion with a formatted message @@ -301,14 +301,14 @@ bprintfln :: proc(buf: []byte, fmt: string, args: ..any) -> string { // - loc: The location of the caller // @(disabled=ODIN_DISABLE_ASSERT) -assertf :: proc(condition: bool, fmt: string, args: ..any, loc := #caller_location) { +assertf :: proc(condition: bool, fmt: string, #no_capture args: ..any, loc := #caller_location) { if !condition { // NOTE(dragos): We are using the same trick as in builtin.assert // to improve performance to make the CPU not // execute speculatively, making it about an order of // magnitude faster @(cold) - internal :: proc(loc: runtime.Source_Code_Location, fmt: string, args: ..any) { + internal :: proc(loc: runtime.Source_Code_Location, fmt: string, #no_capture args: ..any) { p := context.assertion_failure_proc if p == nil { p = runtime.default_assertion_failure_proc @@ -326,7 +326,7 @@ assertf :: proc(condition: bool, fmt: string, args: ..any, loc := #caller_locati // - args: A variadic list of arguments to be formatted // - loc: The location of the caller // -panicf :: proc(fmt: string, args: ..any, loc := #caller_location) -> ! { +panicf :: proc(fmt: string, #no_capture args: ..any, loc := #caller_location) -> ! { p := context.assertion_failure_proc if p == nil { p = runtime.default_assertion_failure_proc @@ -346,7 +346,7 @@ panicf :: proc(fmt: string, args: ..any, loc := #caller_location) -> ! { // Returns: A formatted C string // @(require_results) -caprintf :: proc(format: string, args: ..any, newline := false) -> cstring { +caprintf :: proc(format: string, #no_capture args: ..any, newline := false) -> cstring { str: strings.Builder strings.builder_init(&str) sbprintf(&str, format, ..args, newline=newline) @@ -365,7 +365,7 @@ caprintf :: proc(format: string, args: ..any, newline := false) -> cstring { // Returns: A formatted C string // @(require_results) -caprintfln :: proc(format: string, args: ..any) -> cstring { +caprintfln :: proc(format: string, #no_capture args: ..any) -> cstring { return caprintf(format, ..args, newline=true) } // Creates a formatted C string @@ -379,7 +379,7 @@ caprintfln :: proc(format: string, args: ..any) -> cstring { // Returns: A formatted C string. // @(require_results) -ctprint :: proc(args: ..any, sep := " ") -> cstring { +ctprint :: proc(#no_capture args: ..any, sep := " ") -> cstring { str: strings.Builder strings.builder_init(&str, context.temp_allocator) sbprint(&str, ..args, sep=sep) @@ -399,7 +399,7 @@ ctprint :: proc(args: ..any, sep := " ") -> cstring { // Returns: A formatted C string // @(require_results) -ctprintf :: proc(format: string, args: ..any, newline := false) -> cstring { +ctprintf :: proc(format: string, #no_capture args: ..any, newline := false) -> cstring { str: strings.Builder strings.builder_init(&str, context.temp_allocator) sbprintf(&str, format, ..args, newline=newline) @@ -418,7 +418,7 @@ ctprintf :: proc(format: string, args: ..any, newline := false) -> cstring { // Returns: A formatted C string // @(require_results) -ctprintfln :: proc(format: string, args: ..any) -> cstring { +ctprintfln :: proc(format: string, #no_capture args: ..any) -> cstring { return ctprintf(format, ..args, newline=true) } // Formats using the default print settings and writes to the given strings.Builder @@ -430,7 +430,7 @@ ctprintfln :: proc(format: string, args: ..any) -> cstring { // // Returns: A formatted string // -sbprint :: proc(buf: ^strings.Builder, args: ..any, sep := " ") -> string { +sbprint :: proc(buf: ^strings.Builder, #no_capture args: ..any, sep := " ") -> string { wprint(strings.to_writer(buf), ..args, sep=sep, flush=true) return strings.to_string(buf^) } @@ -443,7 +443,7 @@ sbprint :: proc(buf: ^strings.Builder, args: ..any, sep := " ") -> string { // // Returns: The resulting formatted string // -sbprintln :: proc(buf: ^strings.Builder, args: ..any, sep := " ") -> string { +sbprintln :: proc(buf: ^strings.Builder, #no_capture args: ..any, sep := " ") -> string { wprintln(strings.to_writer(buf), ..args, sep=sep, flush=true) return strings.to_string(buf^) } @@ -457,7 +457,7 @@ sbprintln :: proc(buf: ^strings.Builder, args: ..any, sep := " ") -> string { // // Returns: The resulting formatted string // -sbprintf :: proc(buf: ^strings.Builder, fmt: string, args: ..any, newline := false) -> string { +sbprintf :: proc(buf: ^strings.Builder, fmt: string, #no_capture args: ..any, newline := false) -> string { wprintf(strings.to_writer(buf), fmt, ..args, flush=true, newline=newline) return strings.to_string(buf^) } @@ -469,7 +469,7 @@ sbprintf :: proc(buf: ^strings.Builder, fmt: string, args: ..any, newline := fal // // Returns: A formatted string // -sbprintfln :: proc(buf: ^strings.Builder, format: string, args: ..any) -> string { +sbprintfln :: proc(buf: ^strings.Builder, format: string, #no_capture args: ..any) -> string { return sbprintf(buf, format, ..args, newline=true) } // Formats and writes to an io.Writer using the default print settings @@ -481,7 +481,7 @@ sbprintfln :: proc(buf: ^strings.Builder, format: string, args: ..any) -> string // // Returns: The number of bytes written // -wprint :: proc(w: io.Writer, args: ..any, sep := " ", flush := true) -> int { +wprint :: proc(w: io.Writer, #no_capture args: ..any, sep := " ", flush := true) -> int { fi: Info fi.writer = w @@ -522,7 +522,7 @@ wprint :: proc(w: io.Writer, args: ..any, sep := " ", flush := true) -> int { // // Returns: The number of bytes written // -wprintln :: proc(w: io.Writer, args: ..any, sep := " ", flush := true) -> int { +wprintln :: proc(w: io.Writer, #no_capture args: ..any, sep := " ", flush := true) -> int { fi: Info fi.writer = w @@ -549,11 +549,11 @@ wprintln :: proc(w: io.Writer, args: ..any, sep := " ", flush := true) -> int { // // Returns: The number of bytes written // -wprintf :: proc(w: io.Writer, fmt: string, args: ..any, flush := true, newline := false) -> int { +wprintf :: proc(w: io.Writer, fmt: string, #no_capture args: ..any, flush := true, newline := false) -> int { MAX_CHECKED_ARGS :: 64 assert(len(args) <= MAX_CHECKED_ARGS, "number of args > 64 is unsupported") - parse_options :: proc(fi: ^Info, fmt: string, index, end: int, unused_args: ^bit_set[0 ..< MAX_CHECKED_ARGS], args: ..any) -> int { + parse_options :: proc(fi: ^Info, fmt: string, index, end: int, unused_args: ^bit_set[0 ..< MAX_CHECKED_ARGS], #no_capture args: ..any) -> int { i := index // Prefix @@ -809,7 +809,7 @@ wprintf :: proc(w: io.Writer, fmt: string, args: ..any, flush := true, newline : // // Returns: The number of bytes written. // -wprintfln :: proc(w: io.Writer, format: string, args: ..any, flush := true) -> int { +wprintfln :: proc(w: io.Writer, format: string, #no_capture args: ..any, flush := true) -> int { return wprintf(w, format, ..args, flush=flush, newline=true) } // Writes a ^runtime.Type_Info value to an io.Writer diff --git a/core/fmt/fmt_js.odin b/core/fmt/fmt_js.odin index acf218eb5..4389b8d87 100644 --- a/core/fmt/fmt_js.odin +++ b/core/fmt/fmt_js.odin @@ -43,7 +43,7 @@ fd_to_writer :: proc(fd: os.Handle, loc := #caller_location) -> io.Writer { } // fprint formats using the default print settings and writes to fd -fprint :: proc(fd: os.Handle, args: ..any, sep := " ", flush := true, loc := #caller_location) -> int { +fprint :: proc(fd: os.Handle, #no_capture args: ..any, sep := " ", flush := true, loc := #caller_location) -> int { buf: [1024]byte b: bufio.Writer defer bufio.writer_flush(&b) @@ -54,7 +54,7 @@ fprint :: proc(fd: os.Handle, args: ..any, sep := " ", flush := true, loc := #ca } // fprintln formats using the default print settings and writes to fd -fprintln :: proc(fd: os.Handle, args: ..any, sep := " ", flush := true, loc := #caller_location) -> int { +fprintln :: proc(fd: os.Handle, #no_capture args: ..any, sep := " ", flush := true, loc := #caller_location) -> int { buf: [1024]byte b: bufio.Writer defer bufio.writer_flush(&b) @@ -66,7 +66,7 @@ fprintln :: proc(fd: os.Handle, args: ..any, sep := " ", flush := true, loc := # } // fprintf formats according to the specified format string and writes to fd -fprintf :: proc(fd: os.Handle, fmt: string, args: ..any, flush := true, newline := false, loc := #caller_location) -> int { +fprintf :: proc(fd: os.Handle, fmt: string, #no_capture args: ..any, flush := true, newline := false, loc := #caller_location) -> int { buf: [1024]byte b: bufio.Writer defer bufio.writer_flush(&b) @@ -78,24 +78,24 @@ fprintf :: proc(fd: os.Handle, fmt: string, args: ..any, flush := true, newline } // fprintfln formats according to the specified format string and writes to fd, followed by a newline. -fprintfln :: proc(fd: os.Handle, fmt: string, args: ..any, flush := true, loc := #caller_location) -> int { +fprintfln :: proc(fd: os.Handle, fmt: string, #no_capture args: ..any, flush := true, loc := #caller_location) -> int { return fprintf(fd, fmt, ..args, flush=flush, newline=true, loc=loc) } // print formats using the default print settings and writes to stdout -print :: proc(args: ..any, sep := " ", flush := true) -> int { return wprint(w=stdout, args=args, sep=sep, flush=flush) } +print :: proc(#no_capture args: ..any, sep := " ", flush := true) -> int { return wprint(w=stdout, args=args, sep=sep, flush=flush) } // println formats using the default print settings and writes to stdout -println :: proc(args: ..any, sep := " ", flush := true) -> int { return wprintln(w=stdout, args=args, sep=sep, flush=flush) } +println :: proc(#no_capture args: ..any, sep := " ", flush := true) -> int { return wprintln(w=stdout, args=args, sep=sep, flush=flush) } // printf formats according to the specififed format string and writes to stdout -printf :: proc(fmt: string, args: ..any, flush := true) -> int { return wprintf(stdout, fmt, ..args, flush=flush) } +printf :: proc(fmt: string, #no_capture args: ..any, flush := true) -> int { return wprintf(stdout, fmt, ..args, flush=flush) } // printfln formats according to the specified format string and writes to stdout, followed by a newline. -printfln :: proc(fmt: string, args: ..any, flush := true) -> int { return wprintf(stdout, fmt, ..args, flush=flush, newline=true) } +printfln :: proc(fmt: string, #no_capture args: ..any, flush := true) -> int { return wprintf(stdout, fmt, ..args, flush=flush, newline=true) } // eprint formats using the default print settings and writes to stderr -eprint :: proc(args: ..any, sep := " ", flush := true) -> int { return wprint(w=stderr, args=args, sep=sep, flush=flush) } +eprint :: proc(#no_capture args: ..any, sep := " ", flush := true) -> int { return wprint(w=stderr, args=args, sep=sep, flush=flush) } // eprintln formats using the default print settings and writes to stderr -eprintln :: proc(args: ..any, sep := " ", flush := true) -> int { return wprintln(w=stderr, args=args, sep=sep, flush=flush) } +eprintln :: proc(#no_capture args: ..any, sep := " ", flush := true) -> int { return wprintln(w=stderr, args=args, sep=sep, flush=flush) } // eprintf formats according to the specififed format string and writes to stderr -eprintf :: proc(fmt: string, args: ..any, flush := true) -> int { return wprintf(stderr, fmt, ..args, flush=flush) } +eprintf :: proc(fmt: string, #no_capture args: ..any, flush := true) -> int { return wprintf(stderr, fmt, ..args, flush=flush) } // eprintfln formats according to the specified format string and writes to stderr, followed by a newline. -eprintfln :: proc(fmt: string, args: ..any, flush := true) -> int { return wprintf(stdout, fmt, ..args, flush=flush, newline=true) } +eprintfln :: proc(fmt: string, #no_capture args: ..any, flush := true) -> int { return wprintf(stdout, fmt, ..args, flush=flush, newline=true) } diff --git a/core/fmt/fmt_os.odin b/core/fmt/fmt_os.odin index 9de0d43be..538f7a08b 100644 --- a/core/fmt/fmt_os.odin +++ b/core/fmt/fmt_os.odin @@ -9,7 +9,7 @@ import "core:io" import "core:bufio" // fprint formats using the default print settings and writes to fd -fprint :: proc(fd: os.Handle, args: ..any, sep := " ", flush := true) -> int { +fprint :: proc(fd: os.Handle, #no_capture args: ..any, sep := " ", flush := true) -> int { buf: [1024]byte b: bufio.Writer defer bufio.writer_flush(&b) @@ -20,7 +20,7 @@ fprint :: proc(fd: os.Handle, args: ..any, sep := " ", flush := true) -> int { } // fprintln formats using the default print settings and writes to fd -fprintln :: proc(fd: os.Handle, args: ..any, sep := " ", flush := true) -> int { +fprintln :: proc(fd: os.Handle, #no_capture args: ..any, sep := " ", flush := true) -> int { buf: [1024]byte b: bufio.Writer defer bufio.writer_flush(&b) @@ -31,7 +31,7 @@ fprintln :: proc(fd: os.Handle, args: ..any, sep := " ", flush := true) -> int { return wprintln(w, ..args, sep=sep, flush=flush) } // fprintf formats according to the specified format string and writes to fd -fprintf :: proc(fd: os.Handle, fmt: string, args: ..any, flush := true, newline := false) -> int { +fprintf :: proc(fd: os.Handle, fmt: string, #no_capture args: ..any, flush := true, newline := false) -> int { buf: [1024]byte b: bufio.Writer defer bufio.writer_flush(&b) @@ -42,7 +42,7 @@ fprintf :: proc(fd: os.Handle, fmt: string, args: ..any, flush := true, newline return wprintf(w, fmt, ..args, flush=flush, newline=newline) } // fprintfln formats according to the specified format string and writes to fd, followed by a newline. -fprintfln :: proc(fd: os.Handle, fmt: string, args: ..any, flush := true) -> int { +fprintfln :: proc(fd: os.Handle, fmt: string, #no_capture args: ..any, flush := true) -> int { return fprintf(fd, fmt, ..args, flush=flush, newline=true) } fprint_type :: proc(fd: os.Handle, info: ^runtime.Type_Info, flush := true) -> (n: int, err: io.Error) { @@ -67,19 +67,19 @@ fprint_typeid :: proc(fd: os.Handle, id: typeid, flush := true) -> (n: int, err: } // print formats using the default print settings and writes to os.stdout -print :: proc(args: ..any, sep := " ", flush := true) -> int { return fprint(os.stdout, ..args, sep=sep, flush=flush) } +print :: proc(#no_capture args: ..any, sep := " ", flush := true) -> int { return fprint(os.stdout, ..args, sep=sep, flush=flush) } // println formats using the default print settings and writes to os.stdout -println :: proc(args: ..any, sep := " ", flush := true) -> int { return fprintln(os.stdout, ..args, sep=sep, flush=flush) } +println :: proc(#no_capture args: ..any, sep := " ", flush := true) -> int { return fprintln(os.stdout, ..args, sep=sep, flush=flush) } // printf formats according to the specified format string and writes to os.stdout -printf :: proc(fmt: string, args: ..any, flush := true) -> int { return fprintf(os.stdout, fmt, ..args, flush=flush) } +printf :: proc(fmt: string, #no_capture args: ..any, flush := true) -> int { return fprintf(os.stdout, fmt, ..args, flush=flush) } // printfln formats according to the specified format string and writes to os.stdout, followed by a newline. -printfln :: proc(fmt: string, args: ..any, flush := true) -> int { return fprintf(os.stdout, fmt, ..args, flush=flush, newline=true) } +printfln :: proc(fmt: string, #no_capture args: ..any, flush := true) -> int { return fprintf(os.stdout, fmt, ..args, flush=flush, newline=true) } // eprint formats using the default print settings and writes to os.stderr -eprint :: proc(args: ..any, sep := " ", flush := true) -> int { return fprint(os.stderr, ..args, sep=sep, flush=flush) } +eprint :: proc(#no_capture args: ..any, sep := " ", flush := true) -> int { return fprint(os.stderr, ..args, sep=sep, flush=flush) } // eprintln formats using the default print settings and writes to os.stderr -eprintln :: proc(args: ..any, sep := " ", flush := true) -> int { return fprintln(os.stderr, ..args, sep=sep, flush=flush) } +eprintln :: proc(#no_capture args: ..any, sep := " ", flush := true) -> int { return fprintln(os.stderr, ..args, sep=sep, flush=flush) } // eprintf formats according to the specified format string and writes to os.stderr -eprintf :: proc(fmt: string, args: ..any, flush := true) -> int { return fprintf(os.stderr, fmt, ..args, flush=flush) } +eprintf :: proc(fmt: string, #no_capture args: ..any, flush := true) -> int { return fprintf(os.stderr, fmt, ..args, flush=flush) } // eprintfln formats according to the specified format string and writes to os.stderr, followed by a newline. -eprintfln :: proc(fmt: string, args: ..any, flush := true) -> int { return fprintf(os.stderr, fmt, ..args, flush=flush, newline=true) } +eprintfln :: proc(fmt: string, #no_capture args: ..any, flush := true) -> int { return fprintf(os.stderr, fmt, ..args, flush=flush, newline=true) } diff --git a/src/check_expr.cpp b/src/check_expr.cpp index 12acca0cb..645d8ac5a 100644 --- a/src/check_expr.cpp +++ b/src/check_expr.cpp @@ -6033,6 +6033,23 @@ gb_internal CallArgumentError check_call_arguments_internal(CheckerContext *c, A Entity *vt = pt->params->Tuple.variables[pt->variadic_index]; o.type = vt->type; + + // NOTE(bill, 2024-07-14): minimize the stack usage for variadic parameter that use `#no_capture` + // on the variadic parameter + if (c->decl && (vt->flags & EntityFlag_NoCapture)) { + bool found = false; + for (NoCaptureData &nc : c->decl->no_captures) { + if (are_types_identical(vt->type, nc.slice_type)) { + nc.max_count = gb_max(nc.max_count, variadic_operands.count); + found = true; + break; + } + } + if (!found) { + array_add(&c->decl->no_captures, NoCaptureData{vt->type, variadic_operands.count}); + } + } + } else { dummy_argument_count += 1; o.type = t_untyped_nil; diff --git a/src/check_type.cpp b/src/check_type.cpp index dd8559114..d1c9bb381 100644 --- a/src/check_type.cpp +++ b/src/check_type.cpp @@ -1953,6 +1953,10 @@ gb_internal Type *check_get_params(CheckerContext *ctx, Scope *scope, Ast *_para error(name, "'#by_ptr' can only be applied to variable fields"); p->flags &= ~FieldFlag_by_ptr; } + if (p->flags&FieldFlag_no_capture) { + error(name, "'#no_capture' can only be applied to variable variadic fields"); + p->flags &= ~FieldFlag_no_capture; + } param = alloc_entity_type_name(scope, name->Ident.token, type, EntityState_Resolved); param->TypeName.is_type_alias = true; @@ -2054,6 +2058,15 @@ gb_internal Type *check_get_params(CheckerContext *ctx, Scope *scope, Ast *_para p->flags &= ~FieldFlag_by_ptr; // Remove the flag } } + if (p->flags&FieldFlag_no_capture) { + if (!(is_variadic && variadic_index == variables.count)) { + error(name, "'#no_capture' can only be applied to a variadic parameter"); + p->flags &= ~FieldFlag_no_capture; + } else if (p->flags & FieldFlag_c_vararg) { + error(name, "'#no_capture' cannot be applied to a #c_vararg parameter"); + p->flags &= ~FieldFlag_no_capture; + } + } if (is_poly_name) { if (p->flags&FieldFlag_no_alias) { @@ -2115,6 +2128,10 @@ gb_internal Type *check_get_params(CheckerContext *ctx, Scope *scope, Ast *_para if (p->flags&FieldFlag_by_ptr) { param->flags |= EntityFlag_ByPtr; } + if (p->flags&FieldFlag_no_capture) { + param->flags |= EntityFlag_NoCapture; + } + param->state = EntityState_Resolved; // NOTE(bill): This should have be resolved whilst determining it add_entity(ctx, scope, name, param); diff --git a/src/checker.cpp b/src/checker.cpp index 8756cce1a..abacc13cb 100644 --- a/src/checker.cpp +++ b/src/checker.cpp @@ -184,6 +184,7 @@ gb_internal void init_decl_info(DeclInfo *d, Scope *scope, DeclInfo *parent) { ptr_set_init(&d->deps, 0); ptr_set_init(&d->type_info_deps, 0); d->labels.allocator = heap_allocator(); + d->no_captures.allocator = heap_allocator(); } gb_internal DeclInfo *make_decl_info(Scope *scope, DeclInfo *parent) { diff --git a/src/checker.hpp b/src/checker.hpp index 781737140..17722f6b6 100644 --- a/src/checker.hpp +++ b/src/checker.hpp @@ -181,6 +181,11 @@ char const *ProcCheckedState_strings[ProcCheckedState_COUNT] { "Checked", }; +struct NoCaptureData { + Type *slice_type; // ..elem_type + isize max_count; +}; + // DeclInfo is used to store information of certain declarations to allow for "any order" usage struct DeclInfo { DeclInfo * parent; // NOTE(bill): only used for procedure literals at the moment @@ -219,6 +224,8 @@ struct DeclInfo { Array labels; + Array no_captures; + // NOTE(bill): this is to prevent a race condition since these procedure literals can be created anywhere at any time struct lbModule *code_gen_module; }; diff --git a/src/entity.cpp b/src/entity.cpp index 41d84e0f7..db6ffdd52 100644 --- a/src/entity.cpp +++ b/src/entity.cpp @@ -45,7 +45,7 @@ enum EntityFlag : u64 { EntityFlag_Value = 1ull<<11, EntityFlag_BitFieldField = 1ull<<12, - + EntityFlag_NoCapture = 1ull<<13, // #no_capture EntityFlag_PolyConst = 1ull<<15, EntityFlag_NotExported = 1ull<<16, diff --git a/src/llvm_backend.hpp b/src/llvm_backend.hpp index 005358734..dd1041702 100644 --- a/src/llvm_backend.hpp +++ b/src/llvm_backend.hpp @@ -296,6 +296,11 @@ enum lbProcedureFlag : u32 { lbProcedureFlag_DebugAllocaCopy = 1<<1, }; +struct lbNoCaptureData { + Type *slice_type; + lbAddr base_array; +}; + struct lbProcedure { u32 flags; u16 state_flags; @@ -336,6 +341,8 @@ struct lbProcedure { bool in_multi_assignment; Array raw_input_parameters; + Array no_captures; + LLVMValueRef temp_callee_return_struct_memory; Ast *curr_stmt; diff --git a/src/llvm_backend_proc.cpp b/src/llvm_backend_proc.cpp index 610c34de2..ec244e185 100644 --- a/src/llvm_backend_proc.cpp +++ b/src/llvm_backend_proc.cpp @@ -517,6 +517,7 @@ gb_internal void lb_begin_procedure_body(lbProcedure *p) { lb_start_block(p, p->entry_block); map_init(&p->direct_parameters); + p->no_captures.allocator = heap_allocator(); GB_ASSERT(p->type != nullptr); @@ -3450,8 +3451,32 @@ gb_internal lbValue lb_build_call_expr_internal(lbProcedure *p, Ast *expr) { } isize slice_len = var_args.count; if (slice_len > 0) { + lbAddr base_array = {}; + if (e->flags & EntityFlag_NoCapture) { + for (lbNoCaptureData const &nc : p->no_captures) { + if (are_types_identical(nc.slice_type, slice_type)) { + base_array = nc.base_array; + break; + } + } + DeclInfo *d = decl_info_of_entity(p->entity); + if (d != nullptr && base_array.addr.value == nullptr) { + for (NoCaptureData const &nc : d->no_captures) { + if (are_types_identical(nc.slice_type, slice_type)) { + base_array = lb_add_local_generated(p, alloc_type_array(elem_type, nc.max_count), true); + array_add(&p->no_captures, lbNoCaptureData{slice_type, base_array}); + break; + } + } + } + } + + if (base_array.addr.value == nullptr) { + base_array = lb_add_local_generated(p, alloc_type_array(elem_type, slice_len), true); + } + GB_ASSERT(base_array.addr.value != nullptr); + lbAddr slice = lb_add_local_generated(p, slice_type, true); - lbAddr base_array = lb_add_local_generated(p, alloc_type_array(elem_type, slice_len), true); for (isize i = 0; i < var_args.count; i++) { lbValue addr = lb_emit_array_epi(p, base_array.addr, cast(i32)i); diff --git a/src/parser.cpp b/src/parser.cpp index 9ce3d563d..a6a146cfd 100644 --- a/src/parser.cpp +++ b/src/parser.cpp @@ -4014,6 +4014,7 @@ struct ParseFieldPrefixMapping { gb_global ParseFieldPrefixMapping const parse_field_prefix_mappings[] = { {str_lit("using"), Token_using, FieldFlag_using}, {str_lit("no_alias"), Token_Hash, FieldFlag_no_alias}, + {str_lit("no_capture"), Token_Hash, FieldFlag_no_capture}, {str_lit("c_vararg"), Token_Hash, FieldFlag_c_vararg}, {str_lit("const"), Token_Hash, FieldFlag_const}, {str_lit("any_int"), Token_Hash, FieldFlag_any_int}, diff --git a/src/parser.hpp b/src/parser.hpp index 86b3393af..15176f789 100644 --- a/src/parser.hpp +++ b/src/parser.hpp @@ -330,9 +330,10 @@ enum FieldFlag : u32 { FieldFlag_subtype = 1<<7, FieldFlag_by_ptr = 1<<8, FieldFlag_no_broadcast = 1<<9, // disallow array programming + FieldFlag_no_capture = 1<<10, // Internal use by the parser only - FieldFlag_Tags = 1<<10, + FieldFlag_Tags = 1<<11, FieldFlag_Results = 1<<16, @@ -340,7 +341,10 @@ enum FieldFlag : u32 { FieldFlag_Invalid = 1u<<31, // Parameter List Restrictions - FieldFlag_Signature = FieldFlag_ellipsis|FieldFlag_using|FieldFlag_no_alias|FieldFlag_c_vararg|FieldFlag_const|FieldFlag_any_int|FieldFlag_by_ptr|FieldFlag_no_broadcast, + FieldFlag_Signature = FieldFlag_ellipsis|FieldFlag_using|FieldFlag_no_alias|FieldFlag_c_vararg| + FieldFlag_const|FieldFlag_any_int|FieldFlag_by_ptr|FieldFlag_no_broadcast| + FieldFlag_no_capture, + FieldFlag_Struct = FieldFlag_using|FieldFlag_subtype|FieldFlag_Tags, }; -- cgit v1.2.3 From 603efa860a5631f1708f6761d753146b6d47b4ba Mon Sep 17 00:00:00 2001 From: Laytan Laats Date: Sat, 14 Sep 2024 21:43:25 +0200 Subject: add '#caller_expression' --- base/runtime/core_builtin.odin | 4 +-- core/odin/parser/parser.odin | 10 +++++++ core/testing/testing.odin | 12 +++++--- src/check_builtin.cpp | 16 +++++++++++ src/check_expr.cpp | 7 ++++- src/check_type.cpp | 32 ++++++++++++++++++++++ src/entity.cpp | 1 + src/llvm_backend.hpp | 2 +- src/llvm_backend_proc.cpp | 62 ++++++++++++++++++++++++++++++++++++++---- 9 files changed, 133 insertions(+), 13 deletions(-) (limited to 'src/entity.cpp') diff --git a/base/runtime/core_builtin.odin b/base/runtime/core_builtin.odin index 8157afe09..67d249d11 100644 --- a/base/runtime/core_builtin.odin +++ b/base/runtime/core_builtin.odin @@ -913,7 +913,7 @@ card :: proc "contextless" (s: $S/bit_set[$E; $U]) -> int { @builtin @(disabled=ODIN_DISABLE_ASSERT) -assert :: proc(condition: bool, message := "", loc := #caller_location) { +assert :: proc(condition: bool, message := #caller_expression(condition), loc := #caller_location) { if !condition { // NOTE(bill): This is wrapped in a procedure call // to improve performance to make the CPU not @@ -952,7 +952,7 @@ unimplemented :: proc(message := "", loc := #caller_location) -> ! { @builtin @(disabled=ODIN_DISABLE_ASSERT) -assert_contextless :: proc "contextless" (condition: bool, message := "", loc := #caller_location) { +assert_contextless :: proc "contextless" (condition: bool, message := #caller_expression(condition), loc := #caller_location) { if !condition { // NOTE(bill): This is wrapped in a procedure call // to improve performance to make the CPU not diff --git a/core/odin/parser/parser.odin b/core/odin/parser/parser.odin index aab59c29d..6f42c17db 100644 --- a/core/odin/parser/parser.odin +++ b/core/odin/parser/parser.odin @@ -2277,6 +2277,16 @@ parse_operand :: proc(p: ^Parser, lhs: bool) -> ^ast.Expr { bd.name = name.text return bd + case "caller_expression": + bd := ast.new(ast.Basic_Directive, tok.pos, end_pos(name)) + bd.tok = tok + bd.name = name.text + + if peek_token_kind(p, .Open_Paren) { + return parse_call_expr(p, bd) + } + return bd + case "location", "exists", "load", "load_directory", "load_hash", "hash", "assert", "panic", "defined", "config": bd := ast.new(ast.Basic_Directive, tok.pos, end_pos(name)) bd.tok = tok diff --git a/core/testing/testing.odin b/core/testing/testing.odin index d5e7c6830..09bf6dc0e 100644 --- a/core/testing/testing.odin +++ b/core/testing/testing.odin @@ -105,9 +105,13 @@ cleanup :: proc(t: ^T, procedure: proc(rawptr), user_data: rawptr) { append(&t.cleanups, Internal_Cleanup{procedure, user_data, context}) } -expect :: proc(t: ^T, ok: bool, msg: string = "", loc := #caller_location) -> bool { +expect :: proc(t: ^T, ok: bool, msg := "", expr := #caller_expression(ok), loc := #caller_location) -> bool { if !ok { - log.error(msg, location=loc) + if msg == "" { + log.errorf("expected %v to be true", expr, location=loc) + } else { + log.error(msg, location=loc) + } } return ok } @@ -119,10 +123,10 @@ expectf :: proc(t: ^T, ok: bool, format: string, args: ..any, loc := #caller_loc return ok } -expect_value :: proc(t: ^T, value, expected: $T, loc := #caller_location) -> bool where intrinsics.type_is_comparable(T) { +expect_value :: proc(t: ^T, value, expected: $T, loc := #caller_location, value_expr := #caller_expression(value)) -> bool where intrinsics.type_is_comparable(T) { ok := value == expected || reflect.is_nil(value) && reflect.is_nil(expected) if !ok { - log.errorf("expected %v, got %v", expected, value, location=loc) + log.errorf("expected %v to be %v, got %v", value_expr, expected, value, location=loc) } return ok } diff --git a/src/check_builtin.cpp b/src/check_builtin.cpp index 888aa074d..909eb668e 100644 --- a/src/check_builtin.cpp +++ b/src/check_builtin.cpp @@ -1632,6 +1632,22 @@ gb_internal bool check_builtin_procedure_directive(CheckerContext *c, Operand *o operand->type = t_source_code_location; operand->mode = Addressing_Value; + } else if (name == "caller_expression") { + if (ce->args.count > 1) { + error(ce->args[0], "'#caller_expression' expects either 0 or 1 arguments, got %td", ce->args.count); + } + if (ce->args.count > 0) { + Ast *arg = ce->args[0]; + Operand o = {}; + Entity *e = check_ident(c, &o, arg, nullptr, nullptr, true); + if (e == nullptr || (e->flags & EntityFlag_Param) == 0) { + error(ce->args[0], "'#caller_expression' expected a valid earlier parameter name"); + } + arg->Ident.entity = e; + } + + operand->type = t_string; + operand->mode = Addressing_Value; } else if (name == "exists") { if (ce->args.count != 1) { error(ce->close, "'#exists' expects 1 argument, got %td", ce->args.count); diff --git a/src/check_expr.cpp b/src/check_expr.cpp index 7f82fb58a..6776094bf 100644 --- a/src/check_expr.cpp +++ b/src/check_expr.cpp @@ -7807,7 +7807,8 @@ gb_internal ExprKind check_call_expr(CheckerContext *c, Operand *operand, Ast *c name == "load" || name == "load_directory" || name == "load_hash" || - name == "hash" + name == "hash" || + name == "caller_expression" ) { operand->mode = Addressing_Builtin; operand->builtin_id = BuiltinProc_DIRECTIVE; @@ -8725,6 +8726,10 @@ gb_internal ExprKind check_basic_directive_expr(CheckerContext *c, Operand *o, A error(node, "#caller_location may only be used as a default argument parameter"); o->type = t_source_code_location; o->mode = Addressing_Value; + } else if (name == "caller_expression") { + error(node, "#caller_expression may only be used as a default argument parameter"); + o->type = t_string; + o->mode = Addressing_Value; } else { if (name == "location") { init_core_source_code_location(c->checker); diff --git a/src/check_type.cpp b/src/check_type.cpp index 3767f7666..f0e0acb9b 100644 --- a/src/check_type.cpp +++ b/src/check_type.cpp @@ -1605,6 +1605,25 @@ gb_internal bool is_expr_from_a_parameter(CheckerContext *ctx, Ast *expr) { return false; } +gb_internal bool is_caller_expression(Ast *expr) { + if (expr->kind == Ast_BasicDirective && expr->BasicDirective.name.string == "caller_expression") { + return true; + } + + Ast *call = unparen_expr(expr); + if (call->kind != Ast_CallExpr) { + return false; + } + + ast_node(ce, CallExpr, call); + if (ce->proc->kind != Ast_BasicDirective) { + return false; + } + + ast_node(bd, BasicDirective, ce->proc); + String name = bd->name.string; + return name == "caller_expression"; +} gb_internal ParameterValue handle_parameter_value(CheckerContext *ctx, Type *in_type, Type **out_type_, Ast *expr, bool allow_caller_location) { ParameterValue param_value = {}; @@ -1626,7 +1645,19 @@ gb_internal ParameterValue handle_parameter_value(CheckerContext *ctx, Type *in_ if (in_type) { check_assignment(ctx, &o, in_type, str_lit("parameter value")); } + } else if (is_caller_expression(expr)) { + if (expr->kind != Ast_BasicDirective) { + check_builtin_procedure_directive(ctx, &o, expr, t_string); + } + + param_value.kind = ParameterValue_Expression; + o.type = t_string; + o.mode = Addressing_Value; + o.expr = expr; + if (in_type) { + check_assignment(ctx, &o, in_type, str_lit("parameter value")); + } } else { if (in_type) { check_expr_with_type_hint(ctx, &o, expr, in_type); @@ -1858,6 +1889,7 @@ gb_internal Type *check_get_params(CheckerContext *ctx, Scope *scope, Ast *_para case ParameterValue_Nil: break; case ParameterValue_Location: + case ParameterValue_Expression: case ParameterValue_Value: gbString str = type_to_string(type); error(params[i], "A default value for a parameter must not be a polymorphic constant type, got %s", str); diff --git a/src/entity.cpp b/src/entity.cpp index db6ffdd52..0c4a20df4 100644 --- a/src/entity.cpp +++ b/src/entity.cpp @@ -104,6 +104,7 @@ enum ParameterValueKind { ParameterValue_Constant, ParameterValue_Nil, ParameterValue_Location, + ParameterValue_Expression, ParameterValue_Value, }; diff --git a/src/llvm_backend.hpp b/src/llvm_backend.hpp index 29d2ccfe6..68f95cb03 100644 --- a/src/llvm_backend.hpp +++ b/src/llvm_backend.hpp @@ -528,7 +528,7 @@ gb_internal lbAddr lb_store_range_stmt_val(lbProcedure *p, Ast *stmt_val, lbValu gb_internal lbValue lb_emit_source_code_location_const(lbProcedure *p, String const &procedure, TokenPos const &pos); gb_internal lbValue lb_const_source_code_location_const(lbModule *m, String const &procedure, TokenPos const &pos); -gb_internal lbValue lb_handle_param_value(lbProcedure *p, Type *parameter_type, ParameterValue const ¶m_value, TokenPos const &pos); +gb_internal lbValue lb_handle_param_value(lbProcedure *p, Type *parameter_type, ParameterValue const ¶m_value, TypeProc *procedure_type, Ast *call_expression); gb_internal lbValue lb_equal_proc_for_type(lbModule *m, Type *type); gb_internal lbValue lb_hasher_proc_for_type(lbModule *m, Type *type); diff --git a/src/llvm_backend_proc.cpp b/src/llvm_backend_proc.cpp index e850d3364..d84599eb0 100644 --- a/src/llvm_backend_proc.cpp +++ b/src/llvm_backend_proc.cpp @@ -699,7 +699,9 @@ gb_internal void lb_begin_procedure_body(lbProcedure *p) { } if (e->Variable.param_value.kind != ParameterValue_Invalid) { - lbValue c = lb_handle_param_value(p, e->type, e->Variable.param_value, e->token.pos); + GB_ASSERT(e->Variable.param_value.kind != ParameterValue_Location); + GB_ASSERT(e->Variable.param_value.kind != ParameterValue_Expression); + lbValue c = lb_handle_param_value(p, e->type, e->Variable.param_value, nullptr, nullptr); lb_addr_store(p, res, c); } @@ -3420,7 +3422,7 @@ gb_internal lbValue lb_build_builtin_proc(lbProcedure *p, Ast *expr, TypeAndValu } -gb_internal lbValue lb_handle_param_value(lbProcedure *p, Type *parameter_type, ParameterValue const ¶m_value, TokenPos const &pos) { +gb_internal lbValue lb_handle_param_value(lbProcedure *p, Type *parameter_type, ParameterValue const ¶m_value, TypeProc *procedure_type, Ast* call_expression) { switch (param_value.kind) { case ParameterValue_Constant: if (is_type_constant_type(parameter_type)) { @@ -3446,8 +3448,60 @@ gb_internal lbValue lb_handle_param_value(lbProcedure *p, Type *parameter_type, if (p->entity != nullptr) { proc_name = p->entity->token.string; } + + ast_node(ce, CallExpr, call_expression); + TokenPos pos = ast_token(ce->proc).pos; + return lb_emit_source_code_location_as_global(p, proc_name, pos); } + case ParameterValue_Expression: + { + Ast *orig = param_value.original_ast_expr; + if (orig->kind == Ast_BasicDirective) { + gbString expr = expr_to_string(call_expression, temporary_allocator()); + return lb_const_string(p->module, make_string_c(expr)); + } + + isize param_idx = -1; + String param_str = {0}; + { + Ast *call = unparen_expr(orig); + GB_ASSERT(call->kind == Ast_CallExpr); + ast_node(ce, CallExpr, call); + GB_ASSERT(ce->proc->kind == Ast_BasicDirective); + GB_ASSERT(ce->args.count == 1); + Ast *target = ce->args[0]; + GB_ASSERT(target->kind == Ast_Ident); + String target_str = target->Ident.token.string; + + param_idx = lookup_procedure_parameter(procedure_type, target_str); + param_str = target_str; + } + GB_ASSERT(param_idx >= 0); + + + Ast *target_expr = nullptr; + ast_node(ce, CallExpr, call_expression); + + if (ce->split_args->positional.count > param_idx) { + target_expr = ce->split_args->positional[param_idx]; + } + + for_array(i, ce->split_args->named) { + Ast *arg = ce->split_args->named[i]; + ast_node(fv, FieldValue, arg); + GB_ASSERT(fv->field->kind == Ast_Ident); + String name = fv->field->Ident.token.string; + if (name == param_str) { + target_expr = fv->value; + break; + } + } + + gbString expr = expr_to_string(target_expr, temporary_allocator()); + return lb_const_string(p->module, make_string_c(expr)); + } + case ParameterValue_Value: return lb_build_expr(p, param_value.ast_value); } @@ -3739,8 +3793,6 @@ gb_internal lbValue lb_build_call_expr_internal(lbProcedure *p, Ast *expr) { } } - TokenPos pos = ast_token(ce->proc).pos; - if (pt->params != nullptr) { isize min_count = pt->params->Tuple.variables.count; @@ -3764,7 +3816,7 @@ gb_internal lbValue lb_build_call_expr_internal(lbProcedure *p, Ast *expr) { args[arg_index] = lb_const_nil(p->module, e->type); break; case Entity_Variable: - args[arg_index] = lb_handle_param_value(p, e->type, e->Variable.param_value, pos); + args[arg_index] = lb_handle_param_value(p, e->type, e->Variable.param_value, pt, expr); break; case Entity_Constant: -- cgit v1.2.3 From 7b334d2bd9e881b450fb19e394d6d71840a62cf9 Mon Sep 17 00:00:00 2001 From: gingerBill Date: Wed, 1 Jan 2025 17:26:15 +0000 Subject: Add `#branch_location` --- src/check_expr.cpp | 12 +++++++ src/entity.cpp | 1 + src/llvm_backend.hpp | 7 +++- src/llvm_backend_expr.cpp | 12 +++++-- src/llvm_backend_proc.cpp | 6 +++- src/llvm_backend_stmt.cpp | 81 ++++++++++++++++++++++++++++---------------- src/llvm_backend_utility.cpp | 8 ++--- 7 files changed, 89 insertions(+), 38 deletions(-) (limited to 'src/entity.cpp') diff --git a/src/check_expr.cpp b/src/check_expr.cpp index cc9483187..fba9b8dad 100644 --- a/src/check_expr.cpp +++ b/src/check_expr.cpp @@ -8725,6 +8725,18 @@ gb_internal ExprKind check_basic_directive_expr(CheckerContext *c, Operand *o, A error(node, "#caller_expression may only be used as a default argument parameter"); o->type = t_string; o->mode = Addressing_Value; + } else if (name == "branch_location") { + if (!c->in_defer) { + error(node, "#branch_location may only be used within a 'defer' statement"); + } else if (c->curr_proc_decl) { + Entity *e = c->curr_proc_decl->entity; + if (e != nullptr) { + GB_ASSERT(e->kind == Entity_Procedure); + e->Procedure.uses_branch_location = true; + } + } + o->type = t_source_code_location; + o->mode = Addressing_Value; } else { if (name == "location") { init_core_source_code_location(c->checker); diff --git a/src/entity.cpp b/src/entity.cpp index 0c4a20df4..802b381f9 100644 --- a/src/entity.cpp +++ b/src/entity.cpp @@ -256,6 +256,7 @@ struct Entity { bool entry_point_only : 1; bool has_instrumentation : 1; bool is_memcpy_like : 1; + bool uses_branch_location : 1; } Procedure; struct { Array entities; diff --git a/src/llvm_backend.hpp b/src/llvm_backend.hpp index e84ffd1cd..8ca11bf28 100644 --- a/src/llvm_backend.hpp +++ b/src/llvm_backend.hpp @@ -359,6 +359,10 @@ struct lbProcedure { bool in_multi_assignment; Array raw_input_parameters; + bool uses_branch_location; + TokenPos branch_location_pos; + TokenPos curr_token_pos; + Array variadic_reuses; lbAddr variadic_reuse_base_array_ptr; @@ -444,7 +448,8 @@ gb_internal lbValue lb_emit_matrix_ev(lbProcedure *p, lbValue s, isize row, isiz gb_internal lbValue lb_emit_arith(lbProcedure *p, TokenKind op, lbValue lhs, lbValue rhs, Type *type); gb_internal lbValue lb_emit_byte_swap(lbProcedure *p, lbValue value, Type *end_type); -gb_internal void lb_emit_defer_stmts(lbProcedure *p, lbDeferExitKind kind, lbBlock *block); +gb_internal void lb_emit_defer_stmts(lbProcedure *p, lbDeferExitKind kind, lbBlock *block, TokenPos pos); +gb_internal void lb_emit_defer_stmts(lbProcedure *p, lbDeferExitKind kind, lbBlock *block, Ast *node); gb_internal lbValue lb_emit_transmute(lbProcedure *p, lbValue value, Type *t); gb_internal lbValue lb_emit_comp(lbProcedure *p, TokenKind op_kind, lbValue left, lbValue right); gb_internal lbValue lb_emit_call(lbProcedure *p, lbValue value, Array const &args, ProcInlining inlining = ProcInlining_none); diff --git a/src/llvm_backend_expr.cpp b/src/llvm_backend_expr.cpp index 9c325e088..3b238bcd8 100644 --- a/src/llvm_backend_expr.cpp +++ b/src/llvm_backend_expr.cpp @@ -3502,7 +3502,13 @@ gb_internal lbValue lb_build_expr_internal(lbProcedure *p, Ast *expr) { case_ast_node(bd, BasicDirective, expr); TokenPos pos = bd->token.pos; - GB_PANIC("Non-constant basic literal %s - %.*s", token_pos_to_string(pos), LIT(bd->name.string)); + String name = bd->name.string; + if (name == "branch_location") { + GB_ASSERT(p->uses_branch_location); + String proc_name = p->entity->token.string; + return lb_emit_source_code_location_as_global(p, proc_name, p->branch_location_pos); + } + GB_PANIC("Non-constant basic literal %s - %.*s", token_pos_to_string(pos), LIT(name)); case_end; case_ast_node(i, Implicit, expr); @@ -3668,7 +3674,7 @@ gb_internal lbValue lb_build_expr_internal(lbProcedure *p, Ast *expr) { lb_emit_if(p, lb_emit_try_has_value(p, rhs), then, else_); lb_start_block(p, else_); - lb_emit_defer_stmts(p, lbDeferExit_Branch, block); + lb_emit_defer_stmts(p, lbDeferExit_Branch, block, expr); lb_emit_jump(p, block); lb_start_block(p, then); @@ -5493,7 +5499,7 @@ gb_internal lbAddr lb_build_addr_internal(lbProcedure *p, Ast *expr) { lb_emit_if(p, lb_emit_try_has_value(p, rhs), then, else_); lb_start_block(p, else_); - lb_emit_defer_stmts(p, lbDeferExit_Branch, block); + lb_emit_defer_stmts(p, lbDeferExit_Branch, block, expr); lb_emit_jump(p, block); lb_start_block(p, then); diff --git a/src/llvm_backend_proc.cpp b/src/llvm_backend_proc.cpp index 712e13592..7e44a0046 100644 --- a/src/llvm_backend_proc.cpp +++ b/src/llvm_backend_proc.cpp @@ -125,6 +125,10 @@ gb_internal lbProcedure *lb_create_procedure(lbModule *m, Entity *entity, bool i // map_init(&p->selector_addr, 0); // map_init(&p->tuple_fix_map, 0); + if (p->entity != nullptr && p->entity->Procedure.uses_branch_location) { + p->uses_branch_location = true; + } + if (p->is_foreign) { lb_add_foreign_library_path(p->module, entity->Procedure.foreign_library); } @@ -757,7 +761,7 @@ gb_internal void lb_end_procedure_body(lbProcedure *p) { if (p->type->Proc.result_count == 0) { instr = LLVMGetLastInstruction(p->curr_block->block); if (!lb_is_instr_terminating(instr)) { - lb_emit_defer_stmts(p, lbDeferExit_Return, nullptr); + lb_emit_defer_stmts(p, lbDeferExit_Return, nullptr, p->body); lb_set_debug_position_to_procedure_end(p); LLVMBuildRetVoid(p->builder); } diff --git a/src/llvm_backend_stmt.cpp b/src/llvm_backend_stmt.cpp index 9a5f25712..a2f0d2f4a 100644 --- a/src/llvm_backend_stmt.cpp +++ b/src/llvm_backend_stmt.cpp @@ -208,8 +208,8 @@ gb_internal void lb_open_scope(lbProcedure *p, Scope *s) { } -gb_internal void lb_close_scope(lbProcedure *p, lbDeferExitKind kind, lbBlock *block, bool pop_stack=true) { - lb_emit_defer_stmts(p, kind, block); +gb_internal void lb_close_scope(lbProcedure *p, lbDeferExitKind kind, lbBlock *block, Ast *node, bool pop_stack=true) { + lb_emit_defer_stmts(p, kind, block, node); GB_ASSERT(p->scope_index > 0); // NOTE(bill): Remove `context`s made in that scope @@ -721,7 +721,7 @@ gb_internal void lb_build_range_interval(lbProcedure *p, AstBinaryExpr *node, lb_build_stmt(p, rs->body); - lb_close_scope(p, lbDeferExit_Default, nullptr); + lb_close_scope(p, lbDeferExit_Default, nullptr, node->left); lb_pop_target_list(p); if (check != nullptr) { @@ -854,7 +854,7 @@ gb_internal void lb_build_range_tuple(lbProcedure *p, AstRangeStmt *rs, Scope *s lb_build_stmt(p, rs->body); - lb_close_scope(p, lbDeferExit_Default, nullptr); + lb_close_scope(p, lbDeferExit_Default, nullptr, rs->body); lb_pop_target_list(p); lb_emit_jump(p, loop); lb_start_block(p, done); @@ -976,7 +976,7 @@ gb_internal void lb_build_range_stmt_struct_soa(lbProcedure *p, AstRangeStmt *rs lb_build_stmt(p, rs->body); - lb_close_scope(p, lbDeferExit_Default, nullptr); + lb_close_scope(p, lbDeferExit_Default, nullptr, rs->body); lb_pop_target_list(p); lb_emit_jump(p, loop); lb_start_block(p, done); @@ -1192,7 +1192,7 @@ gb_internal void lb_build_range_stmt(lbProcedure *p, AstRangeStmt *rs, Scope *sc lb_build_stmt(p, rs->body); - lb_close_scope(p, lbDeferExit_Default, nullptr); + lb_close_scope(p, lbDeferExit_Default, nullptr, rs->body); lb_pop_target_list(p); lb_emit_jump(p, loop); lb_start_block(p, done); @@ -1363,7 +1363,7 @@ gb_internal void lb_build_unroll_range_stmt(lbProcedure *p, AstUnrollRangeStmt * } - lb_close_scope(p, lbDeferExit_Default, nullptr); + lb_close_scope(p, lbDeferExit_Default, nullptr, rs->body); } gb_internal bool lb_switch_stmt_can_be_trivial_jump_table(AstSwitchStmt *ss, bool *default_found_) { @@ -1433,6 +1433,7 @@ gb_internal void lb_build_switch_stmt(lbProcedure *p, AstSwitchStmt *ss, Scope * ast_node(body, BlockStmt, ss->body); isize case_count = body->stmts.count; + Ast *default_clause = nullptr; Slice default_stmts = {}; lbBlock *default_fall = nullptr; lbBlock *default_block = nullptr; @@ -1482,6 +1483,7 @@ gb_internal void lb_build_switch_stmt(lbProcedure *p, AstSwitchStmt *ss, Scope * if (cc->list.count == 0) { // default case + default_clause = clause; default_stmts = cc->stmts; default_fall = fall; if (switch_instr == nullptr) { @@ -1552,7 +1554,7 @@ gb_internal void lb_build_switch_stmt(lbProcedure *p, AstSwitchStmt *ss, Scope * lb_push_target_list(p, ss->label, done, nullptr, fall); lb_open_scope(p, body->scope); lb_build_stmt_list(p, cc->stmts); - lb_close_scope(p, lbDeferExit_Default, body); + lb_close_scope(p, lbDeferExit_Default, body, clause); lb_pop_target_list(p); lb_emit_jump(p, done); @@ -1570,13 +1572,13 @@ gb_internal void lb_build_switch_stmt(lbProcedure *p, AstSwitchStmt *ss, Scope * lb_push_target_list(p, ss->label, done, nullptr, default_fall); lb_open_scope(p, default_block->scope); lb_build_stmt_list(p, default_stmts); - lb_close_scope(p, lbDeferExit_Default, default_block); + lb_close_scope(p, lbDeferExit_Default, default_block, default_clause); lb_pop_target_list(p); } lb_emit_jump(p, done); lb_start_block(p, done); - lb_close_scope(p, lbDeferExit_Default, done); + lb_close_scope(p, lbDeferExit_Default, done, ss->body); } gb_internal void lb_store_type_case_implicit(lbProcedure *p, Ast *clause, lbValue value, bool is_default_case) { @@ -1627,7 +1629,7 @@ gb_internal void lb_type_case_body(lbProcedure *p, Ast *label, Ast *clause, lbBl lb_push_target_list(p, label, done, nullptr, nullptr); lb_build_stmt_list(p, cc->stmts); - lb_close_scope(p, lbDeferExit_Default, body); + lb_close_scope(p, lbDeferExit_Default, body, clause); lb_pop_target_list(p); lb_emit_jump(p, done); @@ -1835,7 +1837,7 @@ gb_internal void lb_build_type_switch_stmt(lbProcedure *p, AstTypeSwitchStmt *ss lb_emit_jump(p, done); lb_start_block(p, done); - lb_close_scope(p, lbDeferExit_Default, done); + lb_close_scope(p, lbDeferExit_Default, done, ss->body); } @@ -1959,7 +1961,7 @@ gb_internal void lb_build_assignment(lbProcedure *p, Array &lvals, Slice p->in_multi_assignment = prev_in_assignment; } -gb_internal void lb_build_return_stmt_internal(lbProcedure *p, lbValue res) { +gb_internal void lb_build_return_stmt_internal(lbProcedure *p, lbValue res, TokenPos pos) { lbFunctionType *ft = lb_get_function_type(p->module, p->type); bool return_by_pointer = ft->ret.kind == lbArg_Indirect; bool split_returns = ft->multiple_return_original_type != nullptr; @@ -1982,7 +1984,7 @@ gb_internal void lb_build_return_stmt_internal(lbProcedure *p, lbValue res) { LLVMBuildStore(p->builder, LLVMConstNull(p->abi_function_type->ret.type), p->return_ptr.addr.value); } - lb_emit_defer_stmts(p, lbDeferExit_Return, nullptr); + lb_emit_defer_stmts(p, lbDeferExit_Return, nullptr, pos); // Check for terminator in the defer stmts LLVMValueRef instr = LLVMGetLastInstruction(p->curr_block->block); @@ -2012,7 +2014,7 @@ gb_internal void lb_build_return_stmt_internal(lbProcedure *p, lbValue res) { ret_val = OdinLLVMBuildTransmute(p, ret_val, ret_type); } - lb_emit_defer_stmts(p, lbDeferExit_Return, nullptr); + lb_emit_defer_stmts(p, lbDeferExit_Return, nullptr, pos); // Check for terminator in the defer stmts LLVMValueRef instr = LLVMGetLastInstruction(p->curr_block->block); @@ -2021,7 +2023,7 @@ gb_internal void lb_build_return_stmt_internal(lbProcedure *p, lbValue res) { } } } -gb_internal void lb_build_return_stmt(lbProcedure *p, Slice const &return_results) { +gb_internal void lb_build_return_stmt(lbProcedure *p, Slice const &return_results, TokenPos pos) { lb_ensure_abi_function_type(p->module, p); isize return_count = p->type->Proc.result_count; @@ -2029,7 +2031,7 @@ gb_internal void lb_build_return_stmt(lbProcedure *p, Slice const &return if (return_count == 0) { // No return values - lb_emit_defer_stmts(p, lbDeferExit_Return, nullptr); + lb_emit_defer_stmts(p, lbDeferExit_Return, nullptr, pos); // Check for terminator in the defer stmts LLVMValueRef instr = LLVMGetLastInstruction(p->curr_block->block); @@ -2138,11 +2140,11 @@ gb_internal void lb_build_return_stmt(lbProcedure *p, Slice const &return GB_ASSERT(result_values.count-1 == result_eps.count); lb_addr_store(p, p->return_ptr, result_values[result_values.count-1]); - lb_emit_defer_stmts(p, lbDeferExit_Return, nullptr); + lb_emit_defer_stmts(p, lbDeferExit_Return, nullptr, pos); LLVMBuildRetVoid(p->builder); return; } else { - return lb_build_return_stmt_internal(p, result_values[result_values.count-1]); + return lb_build_return_stmt_internal(p, result_values[result_values.count-1], pos); } } else { @@ -2169,7 +2171,7 @@ gb_internal void lb_build_return_stmt(lbProcedure *p, Slice const &return } if (return_by_pointer) { - lb_emit_defer_stmts(p, lbDeferExit_Return, nullptr); + lb_emit_defer_stmts(p, lbDeferExit_Return, nullptr, pos); LLVMBuildRetVoid(p->builder); return; } @@ -2177,13 +2179,13 @@ gb_internal void lb_build_return_stmt(lbProcedure *p, Slice const &return res = lb_emit_load(p, res); } } - lb_build_return_stmt_internal(p, res); + lb_build_return_stmt_internal(p, res, pos); } gb_internal void lb_build_if_stmt(lbProcedure *p, Ast *node) { ast_node(is, IfStmt, node); lb_open_scope(p, is->scope); // Scope #1 - defer (lb_close_scope(p, lbDeferExit_Default, nullptr)); + defer (lb_close_scope(p, lbDeferExit_Default, nullptr, node)); lbBlock *then = lb_create_block(p, "if.then"); lbBlock *done = lb_create_block(p, "if.done"); @@ -2234,7 +2236,7 @@ gb_internal void lb_build_if_stmt(lbProcedure *p, Ast *node) { lb_open_scope(p, scope_of_node(is->else_stmt)); lb_build_stmt(p, is->else_stmt); - lb_close_scope(p, lbDeferExit_Default, nullptr); + lb_close_scope(p, lbDeferExit_Default, nullptr, is->else_stmt); } lb_emit_jump(p, done); @@ -2251,7 +2253,7 @@ gb_internal void lb_build_if_stmt(lbProcedure *p, Ast *node) { lb_open_scope(p, scope_of_node(is->else_stmt)); lb_build_stmt(p, is->else_stmt); - lb_close_scope(p, lbDeferExit_Default, nullptr); + lb_close_scope(p, lbDeferExit_Default, nullptr, is->else_stmt); lb_emit_jump(p, done); } @@ -2322,7 +2324,7 @@ gb_internal void lb_build_for_stmt(lbProcedure *p, Ast *node) { } lb_start_block(p, done); - lb_close_scope(p, lbDeferExit_Default, nullptr); + lb_close_scope(p, lbDeferExit_Default, nullptr, node); } gb_internal void lb_build_assign_stmt_array(lbProcedure *p, TokenKind op, lbAddr const &lhs, lbValue const &value) { @@ -2588,7 +2590,7 @@ gb_internal void lb_build_stmt(lbProcedure *p, Ast *node) { lb_open_scope(p, bs->scope); lb_build_stmt_list(p, bs->stmts); - lb_close_scope(p, lbDeferExit_Default, nullptr); + lb_close_scope(p, lbDeferExit_Default, nullptr, node); if (done != nullptr) { lb_emit_jump(p, done); @@ -2702,7 +2704,7 @@ gb_internal void lb_build_stmt(lbProcedure *p, Ast *node) { case_end; case_ast_node(rs, ReturnStmt, node); - lb_build_return_stmt(p, rs->results); + lb_build_return_stmt(p, rs->results, ast_token(node).pos); case_end; case_ast_node(is, IfStmt, node); @@ -2755,7 +2757,7 @@ gb_internal void lb_build_stmt(lbProcedure *p, Ast *node) { } } if (block != nullptr) { - lb_emit_defer_stmts(p, lbDeferExit_Branch, block); + lb_emit_defer_stmts(p, lbDeferExit_Branch, block, node); } lb_emit_jump(p, block); lb_start_block(p, lb_create_block(p, "unreachable")); @@ -2795,7 +2797,13 @@ gb_internal void lb_build_defer_stmt(lbProcedure *p, lbDefer const &d) { } } -gb_internal void lb_emit_defer_stmts(lbProcedure *p, lbDeferExitKind kind, lbBlock *block) { +gb_internal void lb_emit_defer_stmts(lbProcedure *p, lbDeferExitKind kind, lbBlock *block, TokenPos pos) { + TokenPos prev_token_pos = p->branch_location_pos; + if (p->uses_branch_location) { + p->branch_location_pos = pos; + } + defer (p->branch_location_pos = prev_token_pos); + isize count = p->defer_stmts.count; isize i = count; while (i --> 0) { @@ -2822,6 +2830,21 @@ gb_internal void lb_emit_defer_stmts(lbProcedure *p, lbDeferExitKind kind, lbBlo } } +gb_internal void lb_emit_defer_stmts(lbProcedure *p, lbDeferExitKind kind, lbBlock *block, Ast *node) { + TokenPos pos = {}; + if (node) { + if (node->kind == Ast_BlockStmt) { + pos = ast_end_token(node).pos; + } else if (node->kind == Ast_CaseClause) { + pos = ast_end_token(node).pos; + } else { + pos = ast_token(node).pos; + } + } + return lb_emit_defer_stmts(p, kind, block, pos); +} + + gb_internal void lb_add_defer_node(lbProcedure *p, isize scope_index, Ast *stmt) { Type *pt = base_type(p->type); GB_ASSERT(pt->kind == Type_Proc); diff --git a/src/llvm_backend_utility.cpp b/src/llvm_backend_utility.cpp index a2a0ba4cc..3e4393a8f 100644 --- a/src/llvm_backend_utility.cpp +++ b/src/llvm_backend_utility.cpp @@ -476,8 +476,8 @@ gb_internal lbValue lb_emit_or_else(lbProcedure *p, Ast *arg, Ast *else_expr, Ty } } -gb_internal void lb_build_return_stmt(lbProcedure *p, Slice const &return_results); -gb_internal void lb_build_return_stmt_internal(lbProcedure *p, lbValue res); +gb_internal void lb_build_return_stmt(lbProcedure *p, Slice const &return_results, TokenPos pos); +gb_internal void lb_build_return_stmt_internal(lbProcedure *p, lbValue res, TokenPos pos); gb_internal lbValue lb_emit_or_return(lbProcedure *p, Ast *arg, TypeAndValue const &tv) { lbValue lhs = {}; @@ -506,10 +506,10 @@ gb_internal lbValue lb_emit_or_return(lbProcedure *p, Ast *arg, TypeAndValue con lbValue found = map_must_get(&p->module->values, end_entity); lb_emit_store(p, found, rhs); - lb_build_return_stmt(p, {}); + lb_build_return_stmt(p, {}, ast_token(arg).pos); } else { GB_ASSERT(tuple->variables.count == 1); - lb_build_return_stmt_internal(p, rhs); + lb_build_return_stmt_internal(p, rhs, ast_token(arg).pos); } } lb_start_block(p, continue_block); -- cgit v1.2.3