diff options
| author | gingerBill <ginger.bill.22@gmail.com> | 2016-07-30 00:09:30 +0100 |
|---|---|---|
| committer | gingerBill <ginger.bill.22@gmail.com> | 2016-07-30 00:17:13 +0100 |
| commit | 776dc0e8f1aa506ae0096c78ff10565e56c175e7 (patch) | |
| tree | 946508a869196a3bf1d005e0ddffd182a1786a1d /src/codegen | |
| parent | 32ab8fcf99df786c264ca566799b022c66cca34b (diff) | |
Restart LLVM IR SSA generation
This is the third go and I'm going for it!
Diffstat (limited to 'src/codegen')
| -rw-r--r-- | src/codegen/codegen.cpp | 318 | ||||
| -rw-r--r-- | src/codegen/print.cpp | 401 | ||||
| -rw-r--r-- | src/codegen/ssa.cpp | 746 | ||||
| -rw-r--r-- | src/codegen/value.cpp | 0 |
4 files changed, 1465 insertions, 0 deletions
diff --git a/src/codegen/codegen.cpp b/src/codegen/codegen.cpp new file mode 100644 index 000000000..a28d0d170 --- /dev/null +++ b/src/codegen/codegen.cpp @@ -0,0 +1,318 @@ +#include "ssa.cpp" +#include "print.cpp" + +struct ssaGen { + ssaModule module; + gbFile output_file; +}; + +b32 ssa_gen_init(ssaGen *s, Checker *c) { + if (c->error_collector.count != 0) + return false; + + gb_for_array(i, c->parser->files) { + AstFile *f = &c->parser->files[i]; + if (f->error_collector.count != 0) + return false; + if (f->tokenizer.error_count != 0) + return false; + } + + ssa_module_init(&s->module, c); + + gbFileError err = gb_file_create(&s->output_file, "../examples/test.ll"); + if (err != gbFileError_None) + return false; + + return true; +} + +void ssa_gen_destroy(ssaGen *s) { + ssa_module_destroy(&s->module); + gb_file_close(&s->output_file); +} + +void ssa_gen_code(ssaGen *s) { + ssaModule *m = &s->module; + CheckerInfo *info = m->info; + gbAllocator a = m->allocator; + gb_for_array(i, info->entities.entries) { + auto *entry = &info->entities.entries[i]; + Entity *e = cast(Entity *)cast(uintptr)entry->key; + DeclarationInfo *decl = entry->value; + + String name = e->token.string; + + switch (e->kind) { + case Entity_TypeName: { + ssaValue *t = ssa_make_value_type_name(a, e); + map_set(&m->members, hash_string(name), t); + } break; + + case Entity_Variable: { + ssaValue *g = ssa_make_value_global(a, e, NULL); + map_set(&m->values, hash_pointer(e), g); + map_set(&m->members, hash_string(name), g); + } break; + + case Entity_Procedure: { + ssaValue *p = ssa_make_value_procedure(a, e, decl, m); + map_set(&m->values, hash_pointer(e), p); + map_set(&m->members, hash_string(name), p); + } break; + } + } + + gb_for_array(i, m->members.entries) { + auto *entry = &m->members.entries[i]; + ssaValue *v = entry->value; + if (v->kind == ssaValue_Procedure) + ssa_build_procedure(v); + } + + ssa_print_llvm_ir(&s->output_file, &s->module); +} + + + + + +#if 0 +#include "type.cpp" +#include "ir.cpp" + +struct Codegen { + Checker *checker; + gbFile file; + gbAllocator allocator; + + irModule module; + + ErrorCollector error_collector; +}; + +b32 init_codegen(Codegen *c, Checker *checker) { + c->checker = checker; + + if (c->error_collector.count != 0) + return false; + for (isize i = 0; i < gb_array_count(checker->parser->files); i++) { + AstFile *f = &checker->parser->files[i]; + if (f->error_collector.count != 0) + return false; + if (f->tokenizer.error_count != 0) + return false; + } + + c->allocator = gb_heap_allocator(); + + ir_module_init(&c->module, c->checker); + + return true; +} + +void destroy_codegen(Codegen *c) { + ir_module_destroy(&c->module); +} + +b32 is_blank_identifier(AstNode *identifier) { + if (identifier->kind == AstNode_Identifier) { + return are_strings_equal(identifier->identifier.token.string, make_string("_")); + } + return false; +} + + +irValue *ir_add_basic_block(gbAllocator a, irValue *p, String label) { + irValue *b = ir_make_value_basic_block(a, gb_array_count(p->procedure.blocks), label, p); + gb_array_append(p->procedure.blocks, b); + return b; +} + +irValue *ir_emit_from_block(irValue *b, irInstruction *i) { + GB_ASSERT(b->kind == irValue_BasicBlock); + i->block = b; + gb_array_append(b->basic_block.instructions, i); + return ir_make_value_instruction(gb_heap_allocator(), i); +} + + +irValue *ir_emit(irValue *p, irInstruction *i) { + GB_ASSERT(p->kind == irValue_Procedure); + return ir_emit_from_block(p->procedure.curr_block, i); +} + + +irInstruction *ir_add_local(irValue *p, Type *type, TokenPos pos) { + irInstruction *i = ir_alloc_instruction(gb_heap_allocator(), irInstruction_Alloca); + i->reg.type = type; + i->reg.pos = pos; + gb_array_append(p->procedure.locals, ir_emit(p, i)); + return i; +} + +irInstruction *ir_add_named_local(irValue *p, Entity *e) { + irInstruction *i = ir_add_local(p, e->type, e->token.pos); + i->alloca.label = e->token.string; + // map_set(&p->procedure.variables, hash_pointer(e), ); + return i; +} + +irInstruction *ir_add_local_for_identifier(irValue *p, AstNode *i) { + GB_ASSERT(p->kind == irValue_Procedure); + GB_ASSERT(i->kind == AstNode_Identifier); + auto *found = map_get(&p->procedure.module->checker->definitions, hash_pointer(i)); + return ir_add_named_local(p, *found); +} + + +void ir_build_variable_declaration(irValue *p, AstNode *d) { + GB_ASSERT(p->kind == irValue_Procedure); + auto *vd = &d->variable_declaration; + + if (vd->name_count == vd->value_count) { + AstNode *name = vd->name_list; + AstNode *value = vd->value_list; + for (; + name != NULL && value != NULL; + name = name->next, value = value->next) { + if (!is_blank_identifier(name)) { + ir_add_local_for_identifier(p, name); + } + // auto lvalue = build_address(p, name, false); + // build_assignment(p, lvalue, value, true, NULL); + } + } else if (vd->value_count == 0) { + AstNode *name = vd->name_list; + for (; + name != NULL; + name = name->next) { + if (!is_blank_identifier(name)) { + + } + + // build_assignment(p, ) + } + } else { + // TODO(bill): Tuple + } + +} + + +void ir_build_expression(irValue *p, AstNode *e) { + GB_ASSERT(p->kind == irValue_Procedure); + +} + + +void ir_build_statement(irValue *p, AstNode *s); + +void ir_build_statement_list(irValue *p, AstNode *list) { + GB_ASSERT(p->kind == irValue_Procedure); + for (AstNode *item = list; item != NULL; item = item->next) { + ir_build_statement(p, item); + } +} + +void ir_build_statement(irValue *p, AstNode *s) { + GB_ASSERT(p->kind == irValue_Procedure); + + switch (s->kind) { + case AstNode_EmptyStatement: + break; + + case AstNode_VariableDeclaration: { + auto *vd = &s->variable_declaration; + if (vd->kind == Declaration_Mutable) { + ir_build_variable_declaration(p, s); + } + } break; + + + case AstNode_ExpressionStatement: + ir_build_expression(p, s->expression_statement.expression); + break; + + case AstNode_BlockStatement: + ir_build_statement_list(p, s->block_statement.list); + break; + } + +} + + + + + +void ir_begin_procedure_body(irValue *p) { + gbAllocator a = gb_heap_allocator(); + p->procedure.curr_block = ir_add_basic_block(a, p, make_string("entry")); + map_init(&p->procedure.variables, a); +} + +void ir_end_procedure_body(irValue *p) { + p->procedure.curr_block = NULL; + map_destroy(&p->procedure.variables); +} + + +void ir_build_procedure(irModule *m, irValue *p) { + if (p->procedure.blocks != NULL) + return; + AstNode *proc_type = NULL; + AstNode *body = NULL; + switch (p->procedure.node->kind) { + case AstNode_ProcedureDeclaration: + proc_type = p->procedure.node->procedure_declaration.procedure_type; + body = p->procedure.node->procedure_declaration.body; + break; + case AstNode_ProcedureLiteral: + proc_type = p->procedure.node->procedure_literal.type; + body = p->procedure.node->procedure_literal.body; + break; + default: + return; + } + + if (body == NULL) { + // NOTE(bill): External procedure + return; + } + + defer (gb_printf("build procedure %.*s\n", LIT(p->procedure.token.string))); + + + ir_begin_procedure_body(p); + ir_build_statement(p, body); + ir_end_procedure_body(p); +} + +void ir_build_proc_decl(irModule *m, AstNode *decl) { + GB_ASSERT(decl != NULL); + auto *pd = &decl->procedure_declaration; + if (is_blank_identifier(pd->name)) + return; + + Entity *e = entity_of_identifier(m->checker, pd->name); + irValue *p = *map_get(&m->values, hash_pointer(e)); + ir_build_procedure(m, p); +} + + + +void generate_code(Codegen *c) { + gbAllocator a = gb_heap_allocator(); + + ir_module_create(&c->module); + + for (isize i = 0; i < gb_array_count(c->module.values.entries); i++) { + irValue *v = c->module.values.entries[i].value; + switch (v->kind) { + case irValue_Procedure: + ir_build_proc_decl(&c->module, v->procedure.node); + break; + } + } +} +#endif diff --git a/src/codegen/print.cpp b/src/codegen/print.cpp new file mode 100644 index 000000000..6ce039e0b --- /dev/null +++ b/src/codegen/print.cpp @@ -0,0 +1,401 @@ +void ssa_fprintf(gbFile *f, char *fmt, ...) { + va_list va; + va_start(va, fmt); + gb_fprintf_va(f, fmt, va); +#if 1 + gb_printf_va(fmt, va); +#endif + va_end(va); +} + + +b32 ssa_valid_char(u8 c) { + if (gb_char_is_alphanumeric(c)) + return true; + + switch (c) { + case '$': + case '-': + case '.': + case '_': + return true; + } + + return false; +} + +void ssa_print_escape_string(gbFile *f, String name) { + isize extra = 0; + for (isize i = 0; i < name.len; i++) { + u8 c = name.text[i]; + if (!ssa_valid_char(c)) + extra += 2; + } + + if (extra == 0) { + ssa_fprintf(f, "%.*s", LIT(name)); + return; + } + + char hex_table[] = "0123456789ABCDEF"; + isize buf_len = name.len + extra; + u8 *buf = gb_alloc_array(gb_heap_allocator(), u8, buf_len); + defer (gb_free(gb_heap_allocator(), buf)); + + isize j = 0; + for (isize i = 0; i < name.len; i++) { + u8 c = name.text[i]; + if (ssa_valid_char(c)) { + buf[j++] = c; + } else { + buf[j] = '\\'; + buf[j+1] = hex_table[c >> 4]; + buf[j+2] = hex_table[c & 0x0f]; + j += 3; + } + } + + gb_file_write(f, buf, buf_len); +} + + + +void ssa_print_encoded_local(gbFile *f, String name) { + ssa_fprintf(f, "%%"); + ssa_print_escape_string(f, name); +} + +void ssa_print_encoded_global(gbFile *f, String name) { + ssa_fprintf(f, "@"); + ssa_print_escape_string(f, name); +} + + +void ssa_print_type(gbFile *f, BaseTypeSizes s, Type *t) { + i64 word_bits = 8*s.word_size; + GB_ASSERT_NOT_NULL(t); + t = default_type(t); + + switch (t->kind) { + case Type_Basic: + switch (t->basic.kind) { + case Basic_bool: ssa_fprintf(f, "i1"); break; + case Basic_i8: ssa_fprintf(f, "i8"); break; + case Basic_i16: ssa_fprintf(f, "i16"); break; + case Basic_i32: ssa_fprintf(f, "i32"); break; + case Basic_i64: ssa_fprintf(f, "i64"); break; + case Basic_u8: ssa_fprintf(f, "i8"); break; + case Basic_u16: ssa_fprintf(f, "i16"); break; + case Basic_u32: ssa_fprintf(f, "i32"); break; + case Basic_u64: ssa_fprintf(f, "i64"); break; + case Basic_f32: ssa_fprintf(f, "float"); break; + case Basic_f64: ssa_fprintf(f, "double"); break; + case Basic_rawptr: ssa_fprintf(f, "void*"); break; + case Basic_string: ssa_fprintf(f, "{i8*, i%lld}", word_bits); break; + case Basic_int: ssa_fprintf(f, "i%lld", word_bits); break; + case Basic_uint: ssa_fprintf(f, "i%lld", word_bits); break; + } + break; + case Type_Array: + ssa_fprintf(f, "[%lld x ", t->array.count); + ssa_print_type(f, s, t->array.element); + ssa_fprintf(f, "]"); + break; + case Type_Slice: + ssa_fprintf(f, "{"); + ssa_print_type(f, s, t->slice.element); + ssa_fprintf(f, "*, %lld, %lld}", word_bits, word_bits); + break; + case Type_Structure: + ssa_fprintf(f, "{"); + for (isize i = 0; i < t->structure.field_count; i++) { + if (i > 0) ssa_fprintf(f, ", "); + ssa_print_type(f, s, t->structure.fields[i]->type); + } + ssa_fprintf(f, "}"); + break; + case Type_Pointer: + ssa_print_type(f, s, t->pointer.element); + ssa_fprintf(f, "*"); + break; + case Type_Named: + ssa_print_encoded_local(f, t->named.name); + break; + case Type_Alias: + ssa_print_type(f, s, t->alias.base); + break; + case Type_Tuple: + if (t->tuple.variable_count == 1) { + ssa_print_type(f, s, t->tuple.variables[0]->type); + } else { + ssa_fprintf(f, "{"); + for (isize i = 0; i < t->tuple.variable_count; i++) { + if (i > 0) ssa_fprintf(f, ", "); + ssa_print_type(f, s, t->tuple.variables[i]->type); + } + ssa_fprintf(f, "}"); + } + break; + case Type_Procedure: + if (t->procedure.result_count == 0) + ssa_fprintf(f, "void"); + else + ssa_print_type(f, s, t->procedure.results); + ssa_fprintf(f, " ("); + for (isize i = 0; i < t->procedure.param_count; i++) { + if (i > 0) ssa_fprintf(f, ", "); + ssa_print_type(f, s, &t->procedure.params[i]); + } + ssa_fprintf(f, ")"); + break; + } +} + +void ssa_print_exact_value(gbFile *f, ssaModule *m, ExactValue value, Type *type) { + switch (value.kind) { + case ExactValue_Bool: + ssa_fprintf(f, (value.value_bool ? "true" : "false")); + break; + case ExactValue_String: { + ssa_fprintf(f, "{"); + ssa_print_type(f, m->sizes, &basic_types[Basic_i8]); + ssa_fprintf(f, "* \""); + // TODO(bill): Make unquote string function + String unquoted = value.value_string; + unquoted.text++; + unquoted.len -= 2; + ssa_print_escape_string(f, unquoted); + ssa_fprintf(f, "\", "); + ssa_print_type(f, m->sizes, &basic_types[Basic_int]); + ssa_fprintf(f, " %td}", value.value_string.len); + } break; + case ExactValue_Integer: + ssa_fprintf(f, "%lld", value.value_integer); + break; + case ExactValue_Float: { + u64 u = 0; + if (is_type_float(type) && type->basic.kind == Basic_f32) { + // IMPORTANT NOTE(bill): LLVM requires all floating point constants to be + // a 64 bit number if bits_of(float type) <= 64. + // To overcome this problem, fill the "bottom" 32 bits with zeros + // https://groups.google.com/forum/#!topic/llvm-dev/IlqV3TbSk6M + f32 fp = cast(f32)value.value_float; + u = *cast(u32 *)&fp; + u <<= 32; + + } else { + u = *cast(u64 *)&value.value_float; + } + ssa_fprintf(f, "0x%llx", u); + } break; + case ExactValue_Pointer: + if (value.value_float == NULL) { + ssa_fprintf(f, "null"); + } else { + GB_PANIC("TODO(bill): ExactValue_Pointer"); + } + break; + default: + GB_PANIC("Invalid ExactValue"); + break; + } +} + + +void ssa_print_value(gbFile *f, ssaModule *m, ssaValue *value, Type *type_hint) { + if (value == NULL) { + ssa_fprintf(f, "!!!NULL_VALUE"); + return; + } + switch (value->kind) { + case ssaValue_TypeName: + ssa_print_encoded_local(f, value->type_name->token.string); + break; + case ssaValue_Global: + ssa_print_encoded_global(f, value->global.entity->token.string); + break; + case ssaValue_Procedure: + ssa_print_encoded_global(f, value->procedure.entity->token.string); + break; + case ssaValue_Constant: { + ssa_print_exact_value(f, m, value->constant.value, type_hint); + } break; + case ssaValue_Instruction: + ssa_fprintf(f, "%%%d", value->id); + break; + } +} + +void ssa_print_instruction(gbFile *f, ssaModule *m, ssaValue *value) { + GB_ASSERT(value->kind == ssaValue_Instruction); + + ssa_fprintf(f, "\t"); + + ssaInstruction *instr = &value->instruction; + switch (instr->kind) { + case ssaInstruction_Local: { + Type *type = instr->local.entity->type; + ssa_fprintf(f, "%%%d = alloca ", value->id); + ssa_print_type(f, m->sizes, type); + ssa_fprintf(f, ", align %lld\n", type_align_of(m->sizes, gb_heap_allocator(), type)); + ssa_fprintf(f, "\tstore "); + ssa_print_type(f, m->sizes, type); + ssa_fprintf(f, " zeroinitializer, "); + ssa_print_type(f, m->sizes, type); + ssa_fprintf(f, "* %%%d\n", value->id); + } break; + + case ssaInstruction_Store: { + Type *type = ssa_value_type(instr->store.address); + ssa_fprintf(f, "store "); + ssa_print_type(f, m->sizes, type); + ssa_fprintf(f, " "); + ssa_print_value(f, m, instr->store.value, type); + ssa_fprintf(f, ", "); + ssa_print_type(f, m->sizes, type); + ssa_fprintf(f, "* "); + ssa_print_value(f, m, instr->store.address, type); + ssa_fprintf(f, "\n"); + } break; + + case ssaInstruction_Load: { + Type *type = ssa_value_type(instr->load.address); + ssa_fprintf(f, "%%%d = load ", value->id); + ssa_print_type(f, m->sizes, type); + ssa_fprintf(f, ", "); + ssa_print_type(f, m->sizes, type); + ssa_fprintf(f, "* "); + ssa_print_value(f, m, instr->load.address, type); + ssa_fprintf(f, "\n"); + } break; + + + case ssaInstruction_BinaryOp: { + ssaBinaryOp *bo = &value->instruction.binary_op; + Type *type = ssa_value_type(bo->left); + + if (is_type_float(type)) + ssa_fprintf(f, "f"); + + switch (bo->op.kind) { + case Token_Add: ssa_fprintf(f, "add"); break; + case Token_Sub: ssa_fprintf(f, "sub"); break; + case Token_And: ssa_fprintf(f, "and"); break; + case Token_Or: ssa_fprintf(f, "or"); break; + case Token_Xor: ssa_fprintf(f, "xor"); break; + + case Token_AndNot: GB_PANIC("TODO(bill): print Token_AndNot"); + + case Token_Mul: ssa_fprintf(f, "mul"); break; + + default: { + if (!is_type_float(type)) { + if (is_type_unsigned(type)) + ssa_fprintf(f, "u"); + else + ssa_fprintf(f, "s"); + } + + switch (bo->op.kind) { + case Token_Quo: ssa_fprintf(f, "div"); break; + case Token_Mod: ssa_fprintf(f, "rem"); break; + } + } break; + } + + ssa_fprintf(f, " "); + ssa_print_type(f, m->sizes, type); + ssa_fprintf(f, " "); + ssa_print_value(f, m, bo->left, type); + ssa_fprintf(f, ", "); + ssa_print_value(f, m, bo->right, type); + ssa_fprintf(f, "\n"); + + } break; + + default: + ssa_fprintf(f, "; <unknown instr> %d\n", instr->kind); + break; + } +} + +void ssa_print_llvm_ir(gbFile *f, ssaModule *m) { + gb_printf("-- Printing LLVM-IR -- \n\n"); + gb_for_array(member_index, m->members.entries) { + auto *entry = &m->members.entries[member_index]; + ssaValue *v = entry->value; + switch (v->kind) { + case ssaValue_TypeName: { + ssa_print_encoded_local(f, v->type_name->token.string); + ssa_fprintf(f, " = type "); + ssa_print_type(f, m->sizes, get_base_type(v->type_name->type)); + ssa_fprintf(f, "\n"); + } break; + + case ssaValue_Global: { + ssaGlobal *g = &v->global; + ssa_print_encoded_global(f, g->entity->token.string); + ssa_fprintf(f, " = global "); + ssa_print_type(f, m->sizes, get_base_type(g->entity->type)); + ssa_fprintf(f, " "); + ssa_print_value(f, m, g->value, g->entity->type); + ssa_fprintf(f, ", align %td\n", type_align_of(m->sizes, gb_heap_allocator(), g->entity->type)); + } break; + + case ssaValue_Procedure: { + ssaProcedure *proc = &v->procedure; + if (proc->body == NULL) { + ssa_fprintf(f, "declare "); + } else { + ssa_fprintf(f, "define "); + } + + auto *proc_type = &proc->entity->type->procedure; + + if (proc_type->result_count == 0) { + ssa_fprintf(f, "void"); + } else { + ssa_print_type(f, m->sizes, proc_type->results); + } + + ssa_fprintf(f, " "); + + ssa_print_encoded_global(f, proc->name); + ssa_fprintf(f, "("); + + if (proc_type->param_count > 0) { + auto *params = &proc_type->params->tuple; + for (isize i = 0; i < params->variable_count; i++) { + Entity *e = params->variables[i]; + if (i > 0) + ssa_fprintf(f, ", "); + ssa_print_type(f, m->sizes, e->type); + ssa_fprintf(f, " %%%.*s", LIT(e->token.string)); + } + } + + ssa_fprintf(f, ") "); + + if (proc->body == NULL) { + ssa_fprintf(f, "\n"); + } else { + ssa_fprintf(f, "{\n"); + gb_for_array(i, proc->blocks) { + ssaBlock *block = &proc->blocks[i]->block; + ssa_fprintf(f, "%.*s:\n", LIT(block->label)); + gb_for_array(j, block->instructions) { + ssaValue *value = block->instructions[j]; + ssa_print_instruction(f, m, value); + } + } + + if (proc_type->result_count == 0) { + ssa_fprintf(f, "\tret void\n"); + } + ssa_fprintf(f, "}\n"); + } + + } break; + } + } +} diff --git a/src/codegen/ssa.cpp b/src/codegen/ssa.cpp new file mode 100644 index 000000000..d17264f64 --- /dev/null +++ b/src/codegen/ssa.cpp @@ -0,0 +1,746 @@ +struct ssaModule; +struct ssaProcedure; +struct ssaBlock; +struct ssaValue; + + +struct ssaModule { + CheckerInfo *info; + BaseTypeSizes sizes; + gbAllocator allocator; + + Map<ssaValue *> values; // Key: Entity * + Map<ssaValue *> members; // Key: String + i32 global_string_index; +}; + + +struct ssaBlock { + i32 id; + AstNode *node; + Scope *scope; + String label; + ssaProcedure *parent; + + gbArray(ssaValue *) instructions; + gbArray(ssaValue *) values; +}; + +struct ssaProcedure { + ssaModule *module; + String name; + Entity *entity; + DeclarationInfo *decl; + AstNode *type_expr; + AstNode *body; + + gbArray(ssaValue *) blocks; + ssaBlock *curr_block; +}; + + +struct ssaLocal { + Entity *entity; +}; +struct ssaGlobal { + b32 generated; + Entity *entity; + ssaValue *value; +}; +struct ssaStore { + ssaValue *address; + ssaValue *value; +}; +struct ssaLoad { + ssaValue *address; +}; +struct ssaBinaryOp { + Token op; + ssaValue *left, *right; +}; +struct ssaGetElementPtr { + ssaValue *address; + Type *result_type; + Type *element_type; + gbArray(ssaValue *) indices; +}; + +enum ssaInstructionKind { + ssaInstruction_Invalid, + + ssaInstruction_Local, + ssaInstruction_Store, + ssaInstruction_Load, + ssaInstruction_GetElementPtr, + + ssaInstruction_BinaryOp, + + ssaInstruction_Count, +}; + +struct ssaInstruction { + ssaInstructionKind kind; + + ssaBlock *parent; + Type *type; + TokenPos pos; + + union { + ssaLocal local; + ssaStore store; + ssaLoad load; + ssaGetElementPtr get_element_ptr; + + ssaBinaryOp binary_op; + }; +}; + + +enum ssaValueKind { + ssaValue_Invalid, + + ssaValue_TypeName, + ssaValue_Global, + ssaValue_Procedure, + ssaValue_Constant, + + ssaValue_Block, + ssaValue_Instruction, + + ssaValue_Count, +}; + +struct ssaValue { + ssaValueKind kind; + i32 id; + + union { + Entity * type_name; + ssaGlobal global; + ssaProcedure procedure; + TypeAndValue constant; + ssaBlock block; + ssaInstruction instruction; + }; +}; + +enum ssaLvalueKind { + ssaLvalue_Blank, + ssaLvalue_Address, + + ssaLvalue_Count, +}; + +struct ssaLvalue { + ssaLvalueKind kind; + union { + struct {} blank; + struct { + ssaValue *value; + AstNode *expr; + } address; + }; +}; + +void ssa_module_init(ssaModule *m, Checker *c) { + m->allocator = gb_heap_allocator(); + m->info = &c->info; + m->sizes = c->sizes; + + map_init(&m->values, m->allocator); + map_init(&m->members, m->allocator); +} + +void ssa_module_destroy(ssaModule *m) { + map_destroy(&m->values); + map_destroy(&m->members); +} + +void ssa_module_add_value(ssaModule *m, Entity *e, ssaValue *v) { + map_set(&m->values, hash_pointer(e), v); +} + + +Type *ssa_value_type(ssaValue *value); + +Type *ssa_instruction_type(ssaInstruction *instr) { + switch (instr->kind) { + case ssaInstruction_Local: + return instr->local.entity->type; + case ssaInstruction_Store: + return ssa_value_type(instr->store.address); + case ssaInstruction_Load: + return ssa_value_type(instr->load.address); + } + return NULL; +} + +Type *ssa_value_type(ssaValue *value) { + switch (value->kind) { + case ssaValue_TypeName: + return value->type_name->type; + case ssaValue_Global: + return value->global.entity->type; + case ssaValue_Procedure: + return value->procedure.entity->type; + case ssaValue_Constant: + return value->constant.type; + case ssaValue_Instruction: + return ssa_instruction_type(&value->instruction); + } + return NULL; +} + + + + + + + + + + + +ssaValue *ssa_alloc_value(gbAllocator a, ssaValueKind kind) { + ssaValue *v = gb_alloc_item(a, ssaValue); + v->kind = kind; + return v; +} + +ssaValue *ssa_alloc_instruction(gbAllocator a, ssaInstructionKind kind) { + ssaValue *v = ssa_alloc_value(a, ssaValue_Instruction); + v->instruction.kind = kind; + return v; +} + +ssaValue *ssa_make_value_type_name(gbAllocator a, Entity *e) { + ssaValue *v = ssa_alloc_value(a, ssaValue_TypeName); + v->type_name = e; + return v; +} + +ssaValue *ssa_make_value_global(gbAllocator a, Entity *e, ssaValue *value) { + ssaValue *v = ssa_alloc_value(a, ssaValue_Global); + v->global.entity = e; + v->global.value = value; + return v; +} + + + +ssaValue *ssa_make_instruction_local(ssaProcedure *p, Entity *e) { + ssaValue *v = ssa_alloc_instruction(p->module->allocator, ssaInstruction_Local); + ssaInstruction *i = &v->instruction; + i->local.entity = e; + if (p->curr_block) { + gb_array_append(p->curr_block->values, v); + } + ssa_module_add_value(p->module, e, v); + return v; +} + + +ssaValue *ssa_make_instruction_store(ssaProcedure *p, ssaValue *address, ssaValue *value) { + ssaValue *v = ssa_alloc_instruction(p->module->allocator, ssaInstruction_Store); + ssaInstruction *i = &v->instruction; + i->store.address = address; + i->store.value = value; + if (p->curr_block) { + gb_array_append(p->curr_block->values, v); + } + return v; +} + +ssaValue *ssa_make_instruction_load(ssaProcedure *p, ssaValue *address) { + ssaValue *v = ssa_alloc_instruction(p->module->allocator, ssaInstruction_Load); + ssaInstruction *i = &v->instruction; + i->load.address = address; + if (p->curr_block) { + gb_array_append(p->curr_block->values, v); + } + return v; +} + +ssaValue *ssa_make_instruction_get_element_ptr(ssaProcedure *p, ssaValue *address) { + ssaValue *v = ssa_alloc_instruction(p->module->allocator, ssaInstruction_GetElementPtr); + ssaInstruction *i = &v->instruction; + i->get_element_ptr.address = address; + if (p->curr_block) { + gb_array_append(p->curr_block->values, v); + } + return v; +} + +ssaValue *ssa_make_instruction_binary_op(ssaProcedure *p, Token op, ssaValue *left, ssaValue *right) { + ssaValue *v = ssa_alloc_instruction(p->module->allocator, ssaInstruction_BinaryOp); + ssaInstruction *i = &v->instruction; + i->binary_op.op = op; + i->binary_op.left = left; + i->binary_op.right = right; + if (p->curr_block) { + gb_array_append(p->curr_block->values, v); + } + return v; +} + + +ssaValue *ssa_make_value_constant(gbAllocator a, Type *type, ExactValue value) { + ssaValue *v = ssa_alloc_value(a, ssaValue_Constant); + v->constant.type = type; + v->constant.value = value; + return v; +} + +ssaValue *ssa_make_value_procedure(gbAllocator a, Entity *e, DeclarationInfo *decl, ssaModule *m) { + ssaValue *v = ssa_alloc_value(a, ssaValue_Procedure); + v->procedure.module = m; + v->procedure.entity = e; + v->procedure.decl = decl; + v->procedure.name = e->token.string; + return v; +} + +ssaValue *ssa_make_value_block(gbAllocator a, ssaProcedure *proc, AstNode *node, Scope *scope, String label) { + ssaValue *v = ssa_alloc_value(a, ssaValue_Block); + v->block.label = label; + v->block.node = node; + v->block.scope = scope; + v->block.parent = proc; + + gb_array_init(v->block.instructions, gb_heap_allocator()); + gb_array_init(v->block.values, gb_heap_allocator()); + + return v; +} + + + +ssaValue *ssa_add_global_string(ssaProcedure *proc, ExactValue value) { + GB_ASSERT(value.kind == ExactValue_String); + gbAllocator a = gb_heap_allocator(); + + + isize max_len = 4+8+1; + u8 *str = cast(u8 *)gb_alloc_array(a, u8, max_len); + isize len = gb_snprintf(cast(char *)str, max_len, ".str%x", proc->module->global_string_index); + proc->module->global_string_index++; + + String name = make_string(str, len-1); + Token token = {Token_String}; + token.string = name; + Type *type = &basic_types[Basic_string]; + Entity *entity = make_entity_constant(a, NULL, token, type, value); + ssaValue *v = ssa_make_value_constant(a, type, value); + + + ssaValue *g = ssa_make_value_global(a, entity, v); + g->global.generated = true; + + map_set(&proc->module->values, hash_pointer(entity), g); + map_set(&proc->module->members, hash_string(name), g); + + return g; +} + + +ssaValue *ssa_add_block(ssaProcedure *proc, AstNode *node, String label) { + gbAllocator a = proc->module->allocator; + Scope *scope = NULL; + Scope **found = map_get(&proc->module->info->scopes, hash_pointer(node)); + if (found) scope = *found; + ssaValue *block = ssa_make_value_block(a, proc, node, scope, label); + gb_array_append(proc->blocks, block); + return block; +} + + + +void ssa_begin_procedure_body(ssaProcedure *proc) { + gb_array_init(proc->blocks, gb_heap_allocator()); + ssaValue *b = ssa_add_block(proc, proc->body, make_string("entry")); + proc->curr_block = &b->block; +} + + +void ssa_end_procedure_body(ssaProcedure *proc) { + +// Number registers + i32 reg_id = 0; + gb_for_array(i, proc->blocks) { + ssaBlock *b = &proc->blocks[i]->block; + gb_for_array(j, b->instructions) { + ssaValue *value = b->instructions[j]; + ssaInstruction *instr = &value->instruction; + if (instr->kind == ssaInstruction_Store) { + continue; + } + value->id = reg_id; + reg_id++; + } + } +} + + +b32 ssa_is_blank_identifier(AstNode *i) { + GB_ASSERT(i->kind == AstNode_Identifier); + return are_strings_equal(i->identifier.token.string, make_string("_")); +} + + +ssaValue *ssa_block_emit(ssaBlock *b, ssaValue *instr) { + instr->instruction.parent = b; + gb_array_append(b->instructions, instr); + return instr; +} + +ssaValue *ssa_emit(ssaProcedure *proc, ssaValue *instr) { + return ssa_block_emit(proc->curr_block, instr); +} + + +ssaValue *ssa_add_local(ssaProcedure *proc, Entity *e) { + ssaValue *instr = ssa_make_instruction_local(proc, e); + ssa_emit(proc, instr); + return instr; +} + +ssaValue *ssa_add_local_for_identifier(ssaProcedure *proc, AstNode *name) { + Entity **found = map_get(&proc->module->info->definitions, hash_pointer(name)); + if (found) { + return ssa_add_local(proc, *found); + } + return NULL; +} + + +ssaValue *ssa_emit_store(ssaProcedure *p, ssaValue *address, ssaValue *value) { + ssaValue *store = ssa_make_instruction_store(p, address, value); + ssa_emit(p, store); + return store; +} + +ssaValue *ssa_emit_load(ssaProcedure *p, ssaValue *address) { + ssaValue *v = ssa_make_instruction_load(p, address); + ssa_emit(p, v); + return v; +} + +ssaValue *ssa_lvalue_store(ssaLvalue lval, ssaProcedure *p, ssaValue *value) { + switch (lval.kind) { + case ssaLvalue_Address: + return ssa_emit_store(p, lval.address.value, value); + } + GB_PANIC("Illegal lvalue store"); + return NULL; +} + +ssaValue *ssa_lvalue_load(ssaLvalue lval, ssaProcedure *p) { + switch (lval.kind) { + case ssaLvalue_Address: + return ssa_emit_load(p, lval.address.value); + } + GB_PANIC("Illegal lvalue load"); + return NULL; +} + + +ssaValue *ssa_lvalue_address(ssaLvalue lval, ssaProcedure *p) { + switch (lval.kind) { + case ssaLvalue_Address: + return lval.address.value; + } + return NULL; +} + +ssaValue *ssa_build_expression(ssaProcedure *proc, AstNode *expr); +ssaValue *ssa_build_single_expression(ssaProcedure *proc, AstNode *expr, TypeAndValue *tv); + +ssaValue *ssa_emit_conversion(ssaProcedure *proc, ssaValue *value, Type *a_type) { + Type *b_type = ssa_value_type(value); + if (are_types_identical(a_type, b_type)) + return value; + + Type *a = get_base_type(a_type); + Type *b = get_base_type(b_type); + + if (value->kind == ssaValue_Constant) { + if (a->kind == Type_Basic) + return ssa_make_value_constant(proc->module->allocator, a_type, value->constant.value); + } + + GB_PANIC("TODO(bill): ssa_emit_conversion"); + + return NULL; +} + +ssaValue *ssa_emit_arith(ssaProcedure *proc, Token op, ssaValue *left, ssaValue *right, Type *type) { + switch (op.kind) { + case Token_Add: + case Token_Sub: + case Token_Mul: + case Token_Quo: + case Token_Mod: + case Token_And: + case Token_Or: + case Token_Xor: + case Token_AndNot: + left = ssa_emit_conversion(proc, left, type); + right = ssa_emit_conversion(proc, right, type); + break; + } + + ssaValue *v = ssa_make_instruction_binary_op(proc, op, left, right); + return ssa_emit(proc, v); +} + +ssaValue *ssa_build_single_expression(ssaProcedure *proc, AstNode *expr, TypeAndValue *tv) { + switch (expr->kind) { + case AstNode_Identifier: { + Entity *e = *map_get(&proc->module->info->uses, hash_pointer(expr)); + if (e->kind == Entity_Builtin) { + // TODO(bill): Entity_Builtin + return NULL; + } + + auto *found = map_get(&proc->module->values, hash_pointer(e)); + if (found) { + return ssa_emit_load(proc, *found); + } + } break; + + case AstNode_UnaryExpression: { + auto *ue = &expr->unary_expression; + switch (ue->op.kind) { + case Token_Add: + return ssa_build_expression(proc, ue->operand); + case Token_Sub: + return NULL; + case Token_Xor: + return NULL; + case Token_Not: + return NULL; + case Token_Pointer: + return NULL; + } + } break; + + case AstNode_BinaryExpression: { + auto *be = &expr->binary_expression; + switch (be->op.kind) { + case Token_Add: + case Token_Sub: + case Token_Mul: + case Token_Quo: + case Token_Mod: + case Token_And: + case Token_Or: + case Token_Xor: + case Token_AndNot: + return ssa_emit_arith(proc, be->op, + ssa_build_expression(proc, be->left), + ssa_build_expression(proc, be->right), + tv->type); + } + } break; + case AstNode_ProcedureLiteral: + break; + case AstNode_CastExpression: + break; + case AstNode_CallExpression: + break; + case AstNode_SliceExpression: + break; + case AstNode_IndexExpression: + break; + case AstNode_SelectorExpression: + break; + } + + GB_PANIC("Unexpected expression"); + return NULL; +} + + +ssaValue *ssa_build_expression(ssaProcedure *proc, AstNode *expr) { + expr = unparen_expression(expr); + + TypeAndValue *tv = map_get(&proc->module->info->types, hash_pointer(expr)); + if (tv) { + if (tv->value.kind != ExactValue_Invalid) { + if (tv->value.kind == ExactValue_String) { + ssaValue *global_str = ssa_add_global_string(proc, tv->value); + return ssa_emit_load(proc, global_str); + } + return ssa_make_value_constant(proc->module->allocator, tv->type, tv->value); + } + + ssaValue *value = NULL; + if (tv->mode == Addressing_Variable) { + gb_printf("!Addressable!\n"); + // TODO(bill): Addressing_Variable + } else { + value = ssa_build_single_expression(proc, expr, tv); + } + + return value; + } + return NULL; +} + + +ssaLvalue ssa_build_address(ssaProcedure *proc, AstNode *expr) { + switch (expr->kind) { + case AstNode_Identifier: { + if (!ssa_is_blank_identifier(expr)) { + Entity *e = entity_of_identifier(proc->module->info, expr); + + ssaLvalue val = {ssaLvalue_Address}; + val.address.expr = expr; + ssaValue **found = map_get(&proc->module->values, hash_pointer(e)); + if (found) { + val.address.value = *found; + } + return val; + } + } break; + + case AstNode_ParenExpression: + return ssa_build_address(proc, unparen_expression(expr)); + + case AstNode_DereferenceExpression: { + ssaLvalue val = {ssaLvalue_Address}; + val.address.value = ssa_build_expression(proc, expr); + val.address.expr = expr; + return val; + } break; + + case AstNode_SelectorExpression: + break; + + case AstNode_IndexExpression: + break; + + // TODO(bill): Others address + } + + ssaLvalue blank = {ssaLvalue_Blank}; + return blank; +} + +void ssa_build_statement(ssaProcedure *proc, AstNode *s); + +void ssa_build_statement_list(ssaProcedure *proc, AstNode *list) { + for (AstNode *stmt = list ; stmt != NULL; stmt = stmt->next) + ssa_build_statement(proc, stmt); +} + +void ssa_build_statement(ssaProcedure *proc, AstNode *s) { + switch (s->kind) { + case AstNode_EmptyStatement: + break; + case AstNode_VariableDeclaration: { + auto *vd = &s->variable_declaration; + if (vd->kind == Declaration_Mutable) { + if (vd->name_count == vd->value_count) { // 1:1 assigment + + } else if (vd->value_count == 0) { // declared and zero-initialized + for (AstNode *name = vd->name_list; name != NULL; name = name->next) { + if (!ssa_is_blank_identifier(name)) { + // TODO(bill): add local + ssa_add_local_for_identifier(proc, name); + } + } + } else { // Tuple(s) + GB_PANIC("TODO(bill): tuple assignment variable declaration"); + } + } + } break; + + case AstNode_AssignStatement: { + auto *assign = &s->assign_statement; + switch (assign->op.kind) { + case Token_Eq: { + gbArray(ssaLvalue) lvals; + gb_array_init(lvals, gb_heap_allocator()); + defer (gb_array_free(lvals)); + + for (AstNode *lhs = assign->lhs_list; + lhs != NULL; + lhs = lhs->next) { + ssaLvalue lval = {}; + if (!ssa_is_blank_identifier(lhs)) { + lval = ssa_build_address(proc, lhs); + } + gb_array_append(lvals, lval); + } + + if (assign->lhs_count == assign->rhs_count) { + if (assign->lhs_count == 1) { + AstNode *lhs = assign->lhs_list; + AstNode *rhs = assign->rhs_list; + ssaValue *init = ssa_build_expression(proc, rhs); + ssa_lvalue_store(lvals[0], proc, init); + } else { + GB_PANIC("TODO(bill): parallel assignment"); + } + } else { + GB_PANIC("TODO(bill): tuple assignment"); + } + + } break; + + default: // +=, -=, etc + break; + } + } break; + + case AstNode_ExpressionStatement: + ssa_build_expression(proc, s->expression_statement.expression); + break; + + case AstNode_BlockStatement: + ssa_build_statement_list(proc, s->block_statement.list); + break; + } +} + + + +void ssa_build_procedure(ssaValue *value) { + ssaProcedure *proc = &value->procedure; + + gb_printf("Building %.*s: %.*s\n", LIT(entity_strings[proc->entity->kind]), LIT(proc->name)); + + + AstNode *proc_decl = proc->decl->proc_decl; + switch (proc_decl->kind) { + case AstNode_ProcedureDeclaration: + proc->type_expr = proc_decl->procedure_declaration.type; + proc->body = proc_decl->procedure_declaration.body; + break; + default: + return; + } + + if (proc->body == NULL) { + // TODO(bill): External procedure + return; + } + + + ssa_begin_procedure_body(proc); + ssa_build_statement(proc, proc->body); + ssa_end_procedure_body(proc); + +} + + + + + + + + + + + diff --git a/src/codegen/value.cpp b/src/codegen/value.cpp new file mode 100644 index 000000000..e69de29bb --- /dev/null +++ b/src/codegen/value.cpp |