diff --git a/src/codegen/wasm.zig b/src/codegen/wasm.zig index 036243dcca..97703ee42a 100644 --- a/src/codegen/wasm.zig +++ b/src/codegen/wasm.zig @@ -7,136 +7,292 @@ const mem = std.mem; const Module = @import("../Module.zig"); const Decl = Module.Decl; -const Inst = @import("../ir.zig").Inst; +const ir = @import("../ir.zig"); +const Inst = ir.Inst; const Type = @import("../type.zig").Type; const Value = @import("../value.zig").Value; +const Compilation = @import("../Compilation.zig"); -fn genValtype(ty: Type) u8 { +/// Wasm Value, created when generating an instruction +const WValue = union(enum) { + none: void, + /// Index of the local variable + local: u32, + /// A constant instruction + constant: *Inst, + /// Each newly created wasm block have a label + /// in the form of an index. + block_idx: u32, +}; + +pub const ValueTable = std.AutoArrayHashMap(*Inst, WValue); + +/// Using a given Zig type, returns the corresponding wasm value type +fn genValtype(ty: Type) ?u8 { return switch (ty.tag()) { - .u32, .i32 => 0x7F, - .u64, .i64 => 0x7E, .f32 => 0x7D, .f64 => 0x7C, - else => @panic("TODO: Implement more types for wasm."), + .u32, .i32 => 0x7F, + .u64, .i64 => 0x7E, + else => null, }; } -pub fn genFunctype(buf: *ArrayList(u8), decl: *Decl) !void { - const ty = decl.typed_value.most_recent.typed_value.ty; - const writer = buf.writer(); +/// Code represents the `Code` section of wasm that +/// belongs to a function +pub const Code = struct { + /// Reference to the function declaration the code + /// section belongs to + decl: *Decl, + gpa: *mem.Allocator, + /// Table to save `WValue`'s generated by an `Inst` + values: ValueTable, + /// `bytes` contains the wasm instructions that have been emitted + /// this is what will be emitted after codegen to write the wasm binary + bytes: ArrayList(u8), + /// Contains the generated function type bytecode for the current function + func_type_data: ArrayList(u8), + /// The index the next local generated will have + /// NOTE: arguments share the index with locals therefore the first variable + /// will have the index that comes after the last argument's index + local_index: u32 = 0, + /// The index the next argument generated will have + arg_index: u32 = 0, + /// If codegen fails, an error messages will be allocated and saved + /// in `err_msg` + err_msg: *Compilation.ErrorMsg, - // functype magic - try writer.writeByte(0x60); + const InnerError = error{ + OutOfMemory, + CodegenFail, + }; - // param types - try leb.writeULEB128(writer, @intCast(u32, ty.fnParamLen())); - if (ty.fnParamLen() != 0) { - const params = try buf.allocator.alloc(Type, ty.fnParamLen()); - defer buf.allocator.free(params); - ty.fnParamTypes(params); - for (params) |param_type| try writer.writeByte(genValtype(param_type)); + fn fail(self: *Code, src: usize, comptime fmt: []const u8, args: anytype) InnerError { + self.err_msg = try Compilation.ErrorMsg.create(self.gpa, src, fmt, args); + return error.CodegenFail; } - // return type - const return_type = ty.fnReturnType(); - switch (return_type.tag()) { - .void, .noreturn => try leb.writeULEB128(writer, @as(u32, 0)), - else => { + /// Returns the `WValue` for the given `inst` + /// creates a new WValue for constants and returns that instead + fn resolveInst(self: Code, inst: *Inst) !WValue { + if (inst.value()) |_| { + return WValue{ .constant = inst }; + } + + return self.values.get(inst).?; // Instruction does not dominate all uses! + } + + /// Writes the bytecode depending on the given `WValue` in `val` + fn emitWValue(self: *Code, val: WValue) !void { + const writer = self.bytes.writer(); + switch (val) { + .none => unreachable, + .block_idx => unreachable, + // loads the local onto the stack at the given index + .local => |idx| { + // local.set + try writer.writeByte(0x20); + try leb.writeULEB128(writer, idx); + }, + // creates a new constant onto the stack + .constant => |inst| try self.emitConstant(inst.castTag(.constant).?), + } + } + + fn genFunctype(self: *Code) !void { + const ty = self.decl.typed_value.most_recent.typed_value.ty; + const writer = self.func_type_data.writer(); + + // functype magic + try writer.writeByte(0x60); + + // param types + try leb.writeULEB128(writer, @intCast(u32, ty.fnParamLen())); + if (ty.fnParamLen() != 0) { + const params = try self.gpa.alloc(Type, ty.fnParamLen()); + defer self.gpa.free(params); + ty.fnParamTypes(params); + for (params) |param_type| { + const val_type = genValtype(param_type) orelse + return self.fail(self.decl.src(), "TODO: Wasm generate wasm type value for type '{s}'", .{param_type.tag()}); + try writer.writeByte(val_type); + } + } + + // return type + const return_type = ty.fnReturnType(); + switch (return_type.tag()) { + .void, .noreturn => try leb.writeULEB128(writer, @as(u32, 0)), + else => |ret_type| { + try leb.writeULEB128(writer, @as(u32, 1)); + const val_type = genValtype(return_type) orelse + return self.fail(self.decl.src(), "TODO: Wasm generate wasm return type value for type '{s}'", .{ret_type}); + try writer.writeByte(val_type); + }, + } + } + + /// Generates the wasm bytecode for the given `code` + pub fn gen(self: *Code) !void { + assert(self.bytes.items.len == 0); + try self.genFunctype(); + const writer = self.bytes.writer(); + + // Reserve space to write the size after generating the code + try self.bytes.resize(5); + + // Write instructions + // TODO: check for and handle death of instructions + const tv = self.decl.typed_value.most_recent.typed_value; + const mod_fn = tv.val.castTag(.function).?.data; + + var locals = std.ArrayList(u8).init(self.gpa); + defer locals.deinit(); + + for (mod_fn.body.instructions) |inst| { + if (inst.tag != .alloc) continue; + + const alloc: *Inst.NoOp = inst.castTag(.alloc).?; + const elem_type = alloc.base.ty.elemType(); + + const wasm_type = genValtype(elem_type) orelse + return self.fail(inst.src, "TODO: Wasm generate wasm type value for type '{s}'", .{elem_type.tag()}); + + try locals.append(wasm_type); + } + + try leb.writeULEB128(writer, @intCast(u32, locals.items.len)); + + // emit the actual locals amount + for (locals.items) |local| { try leb.writeULEB128(writer, @as(u32, 1)); - try writer.writeByte(genValtype(return_type)); - }, + try leb.writeULEB128(writer, local); // valtype + } + + for (mod_fn.body.instructions) |inst| { + const result = try self.genInst(inst); + + if (result != .none) { + try self.values.putNoClobber(inst, result); + } + } + + // Write 'end' opcode + try writer.writeByte(0x0B); + + // Fill in the size of the generated code to the reserved space at the + // beginning of the buffer. + const size = self.bytes.items.len - 5 + self.decl.fn_link.wasm.?.idx_refs.items.len * 5; + leb.writeUnsignedFixed(5, self.bytes.items[0..5], @intCast(u32, size)); } -} -pub fn genCode(buf: *ArrayList(u8), decl: *Decl) !void { - assert(buf.items.len == 0); - const writer = buf.writer(); - - // Reserve space to write the size after generating the code - try buf.resize(5); - - // Write the size of the locals vec - // TODO: implement locals - try leb.writeULEB128(writer, @as(u32, 0)); - - // Write instructions - // TODO: check for and handle death of instructions - const tv = decl.typed_value.most_recent.typed_value; - const mod_fn = tv.val.castTag(.function).?.data; - for (mod_fn.body.instructions) |inst| try genInst(buf, decl, inst); - - // Write 'end' opcode - try writer.writeByte(0x0B); - - // Fill in the size of the generated code to the reserved space at the - // beginning of the buffer. - const size = buf.items.len - 5 + decl.fn_link.wasm.?.idx_refs.items.len * 5; - leb.writeUnsignedFixed(5, buf.items[0..5], @intCast(u32, size)); -} - -fn genInst(buf: *ArrayList(u8), decl: *Decl, inst: *Inst) !void { - return switch (inst.tag) { - .call => genCall(buf, decl, inst.castTag(.call).?), - .constant => genConstant(buf, decl, inst.castTag(.constant).?), - .dbg_stmt => {}, - .ret => genRet(buf, decl, inst.castTag(.ret).?), - .retvoid => {}, - else => error.TODOImplementMoreWasmCodegen, - }; -} - -fn genConstant(buf: *ArrayList(u8), decl: *Decl, inst: *Inst.Constant) !void { - const writer = buf.writer(); - switch (inst.base.ty.tag()) { - .u32 => { - try writer.writeByte(0x41); // i32.const - try leb.writeILEB128(writer, inst.val.toUnsignedInt()); - }, - .i32 => { - try writer.writeByte(0x41); // i32.const - try leb.writeILEB128(writer, inst.val.toSignedInt()); - }, - .u64 => { - try writer.writeByte(0x42); // i64.const - try leb.writeILEB128(writer, inst.val.toUnsignedInt()); - }, - .i64 => { - try writer.writeByte(0x42); // i64.const - try leb.writeILEB128(writer, inst.val.toSignedInt()); - }, - .f32 => { - try writer.writeByte(0x43); // f32.const - // TODO: enforce LE byte order - try writer.writeAll(mem.asBytes(&inst.val.toFloat(f32))); - }, - .f64 => { - try writer.writeByte(0x44); // f64.const - // TODO: enforce LE byte order - try writer.writeAll(mem.asBytes(&inst.val.toFloat(f64))); - }, - .void => {}, - else => return error.TODOImplementMoreWasmCodegen, + fn genInst(self: *Code, inst: *Inst) !WValue { + return switch (inst.tag) { + .alloc => self.genAlloc(inst.castTag(.alloc).?), + .arg => self.genArg(inst.castTag(.arg).?), + .call => self.genCall(inst.castTag(.call).?), + .constant => unreachable, + .dbg_stmt => WValue.none, + .load => self.genLoad(inst.castTag(.load).?), + .ret => self.genRet(inst.castTag(.ret).?), + .retvoid => WValue.none, + .store => self.genStore(inst.castTag(.store).?), + else => self.fail(inst.src, "TODO: Implement wasm inst: {s}", .{inst.tag}), + }; } -} -fn genRet(buf: *ArrayList(u8), decl: *Decl, inst: *Inst.UnOp) !void { - try genInst(buf, decl, inst.operand); -} + fn genRet(self: *Code, inst: *Inst.UnOp) !WValue { + const operand = try self.resolveInst(inst.operand); + try self.emitWValue(operand); + return WValue.none; + } -fn genCall(buf: *ArrayList(u8), decl: *Decl, inst: *Inst.Call) !void { - const func_inst = inst.func.castTag(.constant).?; - const func = func_inst.val.castTag(.function).?.data; - const target = func.owner_decl; - const target_ty = target.typed_value.most_recent.typed_value.ty; + fn genCall(self: *Code, inst: *Inst.Call) !WValue { + const func_inst = inst.func.castTag(.constant).?; + const func = func_inst.val.castTag(.function).?.data; + const target = func.owner_decl; + const target_ty = target.typed_value.most_recent.typed_value.ty; - if (inst.args.len != 0) return error.TODOImplementMoreWasmCodegen; + for (inst.args) |arg| { + const arg_val = try self.resolveInst(arg); + try self.emitWValue(arg_val); + } - try buf.append(0x10); // call + try self.bytes.append(0x10); // call - // The function index immediate argument will be filled in using this data - // in link.Wasm.flush(). - try decl.fn_link.wasm.?.idx_refs.append(buf.allocator, .{ - .offset = @intCast(u32, buf.items.len), - .decl = target, - }); -} + // The function index immediate argument will be filled in using this data + // in link.Wasm.flush(). + try self.decl.fn_link.wasm.?.idx_refs.append(self.gpa, .{ + .offset = @intCast(u32, self.bytes.items.len), + .decl = target, + }); + + return WValue.none; + } + + fn genAlloc(self: *Code, inst: *Inst.NoOp) !WValue { + defer self.local_index += 1; + return WValue{ .local = self.local_index }; + } + + fn genStore(self: *Code, inst: *Inst.BinOp) !WValue { + const writer = self.bytes.writer(); + + const lhs = try self.resolveInst(inst.lhs); + + const rhs = try self.resolveInst(inst.rhs); + try self.emitWValue(rhs); + + try writer.writeByte(0x21); // local.set + try leb.writeULEB128(writer, lhs.local); + + return WValue.none; + } + + fn genLoad(self: *Code, inst: *Inst.UnOp) !WValue { + const operand = self.resolveInst(inst.operand); + + // ensure index to local + return WValue{ .local = operand.local }; + } + + fn genArg(self: *Code, inst: *Inst.Arg) !WValue { + // arguments share the index with locals + defer self.local_index += 1; + return WValue{ .local = self.local_index }; + } + + fn emitConstant(self: *Code, inst: *Inst.Constant) !void { + const writer = self.bytes.writer(); + switch (inst.base.ty.tag()) { + .u32 => { + try writer.writeByte(0x41); // i32.const + try leb.writeILEB128(writer, inst.val.toUnsignedInt()); + }, + .i32 => { + try writer.writeByte(0x41); // i32.const + try leb.writeILEB128(writer, inst.val.toSignedInt()); + }, + .u64 => { + try writer.writeByte(0x42); // i64.const + try leb.writeILEB128(writer, inst.val.toUnsignedInt()); + }, + .i64 => { + try writer.writeByte(0x42); // i64.const + try leb.writeILEB128(writer, inst.val.toSignedInt()); + }, + .f32 => { + try writer.writeByte(0x43); // f32.const + // TODO: enforce LE byte order + try writer.writeAll(mem.asBytes(&inst.val.toFloat(f32))); + }, + .f64 => { + try writer.writeByte(0x44); // f64.const + // TODO: enforce LE byte order + try writer.writeAll(mem.asBytes(&inst.val.toFloat(f64))); + }, + .void => {}, + else => |ty| return self.fail(inst.base.src, "Wasm TODO: emitConstant for type {s}", .{ty}), + } + } +}; diff --git a/src/link/Wasm.zig b/src/link/Wasm.zig index 576be2d306..c5e86ada3d 100644 --- a/src/link/Wasm.zig +++ b/src/link/Wasm.zig @@ -118,10 +118,29 @@ pub fn updateDecl(self: *Wasm, module: *Module, decl: *Module.Decl) !void { var managed_functype = fn_data.functype.toManaged(self.base.allocator); var managed_code = fn_data.code.toManaged(self.base.allocator); - try codegen.genFunctype(&managed_functype, decl); - try codegen.genCode(&managed_code, decl); - fn_data.functype = managed_functype.toUnmanaged(); - fn_data.code = managed_code.toUnmanaged(); + + var code = codegen.Code{ + .gpa = self.base.allocator, + .values = codegen.ValueTable.init(self.base.allocator), + .bytes = managed_code, + .func_type_data = managed_functype, + .decl = decl, + .err_msg = undefined, + }; + defer code.values.deinit(); + + // generate the 'code' section for the function declaration + code.gen() catch |err| switch (err) { + error.CodegenFail => { + decl.analysis = .codegen_failure; + try module.failed_decls.put(module.gpa, decl, code.err_msg); + return; + }, + else => |e| return err, + }; + + fn_data.functype = code.func_type_data.toUnmanaged(); + fn_data.code = code.bytes.toUnmanaged(); } pub fn updateDeclExports(