WIP: precise GC

dumb -> leaking:
  make it more clear what this "GC" does: leak everything.
marksweep -> conservative:
  "marksweep" is too generic, use "conservative" to differentiate
  between future garbage collectors: precise marksweep / mark-compact /
  refcounting.

.circleci/config.yml

@@ -82,6 +82,7 @@ commands:
       - run: tinygo build -size short -o test.elf -target=circuitplay-express examples/blinky1
       - run: tinygo build -size short -o test.elf -target=stm32f4disco        examples/blinky1
       - run: tinygo build -size short -o test.elf -target=stm32f4disco        examples/blinky2
+      - run: tinygo build -size short -o test.elf -target=circuitplay-express examples/i2s
       - run: tinygo build             -o wasm.wasm -target=wasm               examples/wasm/export
       - run: tinygo build             -o wasm.wasm -target=wasm               examples/wasm/main
   test-linux:

Makefile

@@ -32,7 +32,7 @@ CGO_LDFLAGS=-L$(LLVM_BUILDDIR)/lib $(CLANG_LIBS) $(LLD_LIBS) $(shell $(LLVM_BUIL
 clean:
 	@rm -rf build
 
-FMT_PATHS = ./*.go compiler interp ir loader src/device/arm src/examples src/machine src/os src/reflect src/runtime src/sync src/syscall
+FMT_PATHS = ./*.go cgo compiler interp ir loader src/device/arm src/examples src/machine src/os src/reflect src/runtime src/sync src/syscall
 fmt:
 	@gofmt -l -w $(FMT_PATHS)
 fmt-check:

cgo/cgo.go

@@ -1,7 +1,15 @@
-package loader
+// Package cgo implements CGo by modifying a loaded AST. It does this by parsing
+// the `import "C"` statements found in the source code with libclang and
+// generating stub function and global declarations.
+//
+// There are a few advantages to modifying the AST directly instead of doing CGo
+// as a preprocessing step, with the main advantage being that debug information
+// is kept intact as much as possible.
+package cgo
 
 // This file extracts the `import "C"` statement from the source and modifies
-// the AST for Cgo. It does not use libclang directly (see libclang.go).
+// the AST for CGo. It does not use libclang directly: see libclang.go for the C
+// source file parsing.
 
 import (
 	"go/ast"
@@ -13,27 +21,38 @@ import (
 	"golang.org/x/tools/go/ast/astutil"
 )
 
-// fileInfo holds all Cgo-related information of a given *ast.File.
+// cgoPackage holds all CGo-related information of a package.
-type fileInfo struct {
+type cgoPackage struct {
-	*ast.File
+	generated       *ast.File
-	*Package
+	generatedPos    token.Pos
-	filename        string
+	errors          []error
-	functions       map[string]*functionInfo
+	dir             string
-	globals         map[string]*globalInfo
+	fset            *token.FileSet
-	typedefs        map[string]*typedefInfo
+	tokenFiles      map[string]*token.File
-	elaboratedTypes map[string]ast.Expr
-	importCPos      token.Pos
 	missingSymbols  map[string]struct{}
+	constants       map[string]constantInfo
+	functions       map[string]*functionInfo
+	globals         map[string]globalInfo
+	typedefs        map[string]*typedefInfo
+	elaboratedTypes map[string]*elaboratedTypeInfo
 }
 
-// functionInfo stores some information about a Cgo function found by libclang
+// constantInfo stores some information about a CGo constant found by libclang
+// and declared in the Go AST.
+type constantInfo struct {
+	expr *ast.BasicLit
+	pos  token.Pos
+}
+
+// functionInfo stores some information about a CGo function found by libclang
 // and declared in the AST.
 type functionInfo struct {
 	args    []paramInfo
 	results *ast.FieldList
+	pos     token.Pos
 }
 
-// paramInfo is a parameter of a Cgo function (see functionInfo).
+// paramInfo is a parameter of a CGo function (see functionInfo).
 type paramInfo struct {
 	name     string
 	typeExpr ast.Expr
@@ -42,11 +61,20 @@ type paramInfo struct {
 // typedefInfo contains information about a single typedef in C.
 type typedefInfo struct {
 	typeExpr ast.Expr
+	pos      token.Pos
+}
+
+// elaboratedTypeInfo contains some information about an elaborated type
+// (struct, union) found in the C AST.
+type elaboratedTypeInfo struct {
+	typeExpr ast.Expr
+	pos      token.Pos
 }
 
 // globalInfo contains information about a declared global variable in C.
 type globalInfo struct {
 	typeExpr ast.Expr
+	pos      token.Pos
 }
 
 // cgoAliases list type aliases between Go and C, for types that are equivalent
@@ -63,9 +91,9 @@ var cgoAliases = map[string]string{
 	"C.uintptr_t": "uintptr",
 }
 
-// cgoBuiltinAliases are handled specially because they only exist on the Go
+// builtinAliases are handled specially because they only exist on the Go side
-// side of CGo, not on the CGo (they're prefixed with "_Cgo_" there).
+// of CGo, not on the CGo side (they're prefixed with "_Cgo_" there).
-var cgoBuiltinAliases = map[string]struct{}{
+var builtinAliases = map[string]struct{}{
 	"char":      struct{}{},
 	"schar":     struct{}{},
 	"uchar":     struct{}{},
@@ -96,27 +124,66 @@ typedef long long           _Cgo_longlong;
 typedef unsigned long long  _Cgo_ulonglong;
 `
 
-// processCgo extracts the `import "C"` statement from the AST, parses the
+// Process extracts `import "C"` statements from the AST, parses the comment
-// comment with libclang, and modifies the AST to use this information.
+// with libclang, and modifies the AST to use this information. It returns a
-func (p *Package) processCgo(filename string, f *ast.File, cflags []string) []error {
+// newly created *ast.File that should be added to the list of to-be-parsed
-	info := &fileInfo{
+// files. If there is one or more error, it returns these in the []error slice
-		File:            f,
+// but still modifies the AST.
-		Package:         p,
+func Process(files []*ast.File, dir string, fset *token.FileSet, cflags []string) (*ast.File, []error) {
-		filename:        filename,
+	p := &cgoPackage{
-		functions:       map[string]*functionInfo{},
+		dir:             dir,
-		globals:         map[string]*globalInfo{},
+		fset:            fset,
-		typedefs:        map[string]*typedefInfo{},
+		tokenFiles:      map[string]*token.File{},
-		elaboratedTypes: map[string]ast.Expr{},
 		missingSymbols:  map[string]struct{}{},
+		constants:       map[string]constantInfo{},
+		functions:       map[string]*functionInfo{},
+		globals:         map[string]globalInfo{},
+		typedefs:        map[string]*typedefInfo{},
+		elaboratedTypes: map[string]*elaboratedTypeInfo{},
+	}
+
+	// Add a new location for the following file.
+	generatedTokenPos := p.fset.AddFile(dir+"/!cgo.go", -1, 0)
+	generatedTokenPos.SetLines([]int{0})
+	p.generatedPos = generatedTokenPos.Pos(0)
+
+	// Construct a new in-memory AST for CGo declarations of this package.
+	unsafeImport := &ast.ImportSpec{
+		Path: &ast.BasicLit{
+			ValuePos: p.generatedPos,
+			Kind:     token.STRING,
+			Value:    "\"unsafe\"",
+		},
+		EndPos: p.generatedPos,
+	}
+	p.generated = &ast.File{
+		Package: p.generatedPos,
+		Name: &ast.Ident{
+			NamePos: p.generatedPos,
+			Name:    files[0].Name.Name,
+		},
+		Decls: []ast.Decl{
+			&ast.GenDecl{
+				TokPos: p.generatedPos,
+				Tok:    token.IMPORT,
+				Specs: []ast.Spec{
+					unsafeImport,
+				},
+			},
+		},
+		Imports: []*ast.ImportSpec{unsafeImport},
 	}
 
 	// Find all C.* symbols.
-	f = astutil.Apply(f, info.findMissingCGoNames, nil).(*ast.File)
+	for _, f := range files {
-	for name := range cgoBuiltinAliases {
+		astutil.Apply(f, p.findMissingCGoNames, nil)
-		info.missingSymbols["_Cgo_"+name] = struct{}{}
+	}
+	for name := range builtinAliases {
+		p.missingSymbols["_Cgo_"+name] = struct{}{}
 	}
 
 	// Find `import "C"` statements in the file.
+	for _, f := range files {
 		for i := 0; i < len(f.Decls); i++ {
 			decl := f.Decls[i]
 			genDecl, ok := decl.(*ast.GenDecl)
@@ -139,15 +206,12 @@ func (p *Package) processCgo(filename string, f *ast.File, cflags []string) []er
 			}
 			cgoComment := genDecl.Doc.Text()
 
-		// Stored for later use by generated functions, to use a somewhat sane
+			pos := genDecl.Pos()
-		// source location.
+			if genDecl.Doc != nil {
-		info.importCPos = spec.Path.ValuePos
+				pos = genDecl.Doc.Pos()
-
-		pos := info.fset.PositionFor(genDecl.Doc.Pos(), true)
-		errs := info.parseFragment(cgoComment+cgoTypes, cflags, pos.Filename, pos.Line)
-		if errs != nil {
-			return errs
 			}
+			position := fset.PositionFor(pos, true)
+			p.parseFragment(cgoComment+cgoTypes, cflags, position.Filename, position.Line)
 
 			// Remove this import declaration.
 			f.Decls = append(f.Decls[:i], f.Decls[i+1:]...)
@@ -155,44 +219,51 @@ func (p *Package) processCgo(filename string, f *ast.File, cflags []string) []er
 		}
 
 		// Print the AST, for debugging.
-	//ast.Print(p.fset, f)
+		//ast.Print(fset, f)
+	}
 
 	// Declare functions found by libclang.
-	info.addFuncDecls()
+	p.addFuncDecls()
 
 	// Declare stub function pointer values found by libclang.
-	info.addFuncPtrDecls()
+	p.addFuncPtrDecls()
 
 	// Declare globals found by libclang.
-	info.addVarDecls()
+	p.addConstDecls()
+
+	// Declare globals found by libclang.
+	p.addVarDecls()
 
 	// Forward C types to Go types (like C.uint32_t -> uint32).
-	info.addTypeAliases()
+	p.addTypeAliases()
 
 	// Add type declarations for C types, declared using typedef in C.
-	info.addTypedefs()
+	p.addTypedefs()
 
 	// Add elaborated types for C structs and unions.
-	info.addElaboratedTypes()
+	p.addElaboratedTypes()
 
 	// Patch the AST to use the declared types and functions.
-	f = astutil.Apply(f, info.walker, nil).(*ast.File)
+	for _, f := range files {
+		astutil.Apply(f, p.walker, nil)
+	}
 
-	return nil
+	// Print the newly generated in-memory AST, for debugging.
+	//ast.Print(fset, p.generated)
+
+	return p.generated, p.errors
 }
 
 // addFuncDecls adds the C function declarations found by libclang in the
 // comment above the `import "C"` statement.
-func (info *fileInfo) addFuncDecls() {
+func (p *cgoPackage) addFuncDecls() {
-	// TODO: replace all uses of importCPos with the real locations from
+	names := make([]string, 0, len(p.functions))
-	// libclang.
+	for name := range p.functions {
-	names := make([]string, 0, len(info.functions))
-	for name := range info.functions {
 		names = append(names, name)
 	}
 	sort.Strings(names)
 	for _, name := range names {
-		fn := info.functions[name]
+		fn := p.functions[name]
 		obj := &ast.Object{
 			Kind: ast.Fun,
 			Name: "C." + name,
@@ -200,16 +271,16 @@ func (info *fileInfo) addFuncDecls() {
 		args := make([]*ast.Field, len(fn.args))
 		decl := &ast.FuncDecl{
 			Name: &ast.Ident{
-				NamePos: info.importCPos,
+				NamePos: fn.pos,
 				Name:    "C." + name,
 				Obj:     obj,
 			},
 			Type: &ast.FuncType{
-				Func: info.importCPos,
+				Func: fn.pos,
 				Params: &ast.FieldList{
-					Opening: info.importCPos,
+					Opening: fn.pos,
 					List:    args,
-					Closing: info.importCPos,
+					Closing: fn.pos,
 				},
 				Results: fn.results,
 			},
@@ -219,7 +290,7 @@ func (info *fileInfo) addFuncDecls() {
 			args[i] = &ast.Field{
 				Names: []*ast.Ident{
 					&ast.Ident{
-						NamePos: info.importCPos,
+						NamePos: fn.pos,
 						Name:    arg.name,
 						Obj: &ast.Object{
 							Kind: ast.Var,
@@ -231,7 +302,7 @@ func (info *fileInfo) addFuncDecls() {
 				Type: arg.typeExpr,
 			}
 		}
-		info.Decls = append(info.Decls, decl)
+		p.generated.Decls = append(p.generated.Decls, decl)
 	}
 }
 
@@ -244,39 +315,40 @@ func (info *fileInfo) addFuncDecls() {
 //         C.mul unsafe.Pointer
 //         // ...
 //     )
-func (info *fileInfo) addFuncPtrDecls() {
+func (p *cgoPackage) addFuncPtrDecls() {
-	if len(info.functions) == 0 {
+	if len(p.functions) == 0 {
 		return
 	}
 	gen := &ast.GenDecl{
-		TokPos: info.importCPos,
+		TokPos: token.NoPos,
 		Tok:    token.VAR,
-		Lparen: info.importCPos,
+		Lparen: token.NoPos,
-		Rparen: info.importCPos,
+		Rparen: token.NoPos,
 	}
-	names := make([]string, 0, len(info.functions))
+	names := make([]string, 0, len(p.functions))
-	for name := range info.functions {
+	for name := range p.functions {
 		names = append(names, name)
 	}
 	sort.Strings(names)
 	for _, name := range names {
+		fn := p.functions[name]
 		obj := &ast.Object{
 			Kind: ast.Typ,
 			Name: "C." + name + "$funcaddr",
 		}
 		valueSpec := &ast.ValueSpec{
 			Names: []*ast.Ident{&ast.Ident{
-				NamePos: info.importCPos,
+				NamePos: fn.pos,
 				Name:    "C." + name + "$funcaddr",
 				Obj:     obj,
 			}},
 			Type: &ast.SelectorExpr{
 				X: &ast.Ident{
-					NamePos: info.importCPos,
+					NamePos: fn.pos,
 					Name:    "unsafe",
 				},
 				Sel: &ast.Ident{
-					NamePos: info.importCPos,
+					NamePos: fn.pos,
 					Name:    "Pointer",
 				},
 			},
@@ -284,7 +356,50 @@ func (info *fileInfo) addFuncPtrDecls() {
 		obj.Decl = valueSpec
 		gen.Specs = append(gen.Specs, valueSpec)
 	}
-	info.Decls = append(info.Decls, gen)
+	p.generated.Decls = append(p.generated.Decls, gen)
+}
+
+// addConstDecls declares external C constants in the Go source.
+// It adds code like the following to the AST:
+//
+//     const (
+//         C.CONST_INT = 5
+//         C.CONST_FLOAT = 5.8
+//         // ...
+//     )
+func (p *cgoPackage) addConstDecls() {
+	if len(p.constants) == 0 {
+		return
+	}
+	gen := &ast.GenDecl{
+		TokPos: token.NoPos,
+		Tok:    token.CONST,
+		Lparen: token.NoPos,
+		Rparen: token.NoPos,
+	}
+	names := make([]string, 0, len(p.constants))
+	for name := range p.constants {
+		names = append(names, name)
+	}
+	sort.Strings(names)
+	for _, name := range names {
+		constVal := p.constants[name]
+		obj := &ast.Object{
+			Kind: ast.Con,
+			Name: "C." + name,
+		}
+		valueSpec := &ast.ValueSpec{
+			Names: []*ast.Ident{&ast.Ident{
+				NamePos: constVal.pos,
+				Name:    "C." + name,
+				Obj:     obj,
+			}},
+			Values: []ast.Expr{constVal.expr},
+		}
+		obj.Decl = valueSpec
+		gen.Specs = append(gen.Specs, valueSpec)
+	}
+	p.generated.Decls = append(p.generated.Decls, gen)
 }
 
 // addVarDecls declares external C globals in the Go source.
@@ -295,30 +410,30 @@ func (info *fileInfo) addFuncPtrDecls() {
 //         C.globalBool bool
 //         // ...
 //     )
-func (info *fileInfo) addVarDecls() {
+func (p *cgoPackage) addVarDecls() {
-	if len(info.globals) == 0 {
+	if len(p.globals) == 0 {
 		return
 	}
 	gen := &ast.GenDecl{
-		TokPos: info.importCPos,
+		TokPos: token.NoPos,
 		Tok:    token.VAR,
-		Lparen: info.importCPos,
+		Lparen: token.NoPos,
-		Rparen: info.importCPos,
+		Rparen: token.NoPos,
 	}
-	names := make([]string, 0, len(info.globals))
+	names := make([]string, 0, len(p.globals))
-	for name := range info.globals {
+	for name := range p.globals {
 		names = append(names, name)
 	}
 	sort.Strings(names)
 	for _, name := range names {
-		global := info.globals[name]
+		global := p.globals[name]
 		obj := &ast.Object{
-			Kind: ast.Typ,
+			Kind: ast.Var,
 			Name: "C." + name,
 		}
 		valueSpec := &ast.ValueSpec{
 			Names: []*ast.Ident{&ast.Ident{
-				NamePos: info.importCPos,
+				NamePos: global.pos,
 				Name:    "C." + name,
 				Obj:     obj,
 			}},
@@ -327,7 +442,7 @@ func (info *fileInfo) addVarDecls() {
 		obj.Decl = valueSpec
 		gen.Specs = append(gen.Specs, valueSpec)
 	}
-	info.Decls = append(info.Decls, gen)
+	p.generated.Decls = append(p.generated.Decls, gen)
 }
 
 // addTypeAliases aliases some built-in Go types with their equivalent C types.
@@ -338,17 +453,17 @@ func (info *fileInfo) addVarDecls() {
 //         C.int16_t = int16
 //         // ...
 //     )
-func (info *fileInfo) addTypeAliases() {
+func (p *cgoPackage) addTypeAliases() {
 	aliasKeys := make([]string, 0, len(cgoAliases))
 	for key := range cgoAliases {
 		aliasKeys = append(aliasKeys, key)
 	}
 	sort.Strings(aliasKeys)
 	gen := &ast.GenDecl{
-		TokPos: info.importCPos,
+		TokPos: token.NoPos,
 		Tok:    token.TYPE,
-		Lparen: info.importCPos,
+		Lparen: token.NoPos,
-		Rparen: info.importCPos,
+		Rparen: token.NoPos,
 	}
 	for _, typeName := range aliasKeys {
 		goTypeName := cgoAliases[typeName]
@@ -358,37 +473,37 @@ func (info *fileInfo) addTypeAliases() {
 		}
 		typeSpec := &ast.TypeSpec{
 			Name: &ast.Ident{
-				NamePos: info.importCPos,
+				NamePos: token.NoPos,
 				Name:    typeName,
 				Obj:     obj,
 			},
-			Assign: info.importCPos,
+			Assign: p.generatedPos,
 			Type: &ast.Ident{
-				NamePos: info.importCPos,
+				NamePos: token.NoPos,
 				Name:    goTypeName,
 			},
 		}
 		obj.Decl = typeSpec
 		gen.Specs = append(gen.Specs, typeSpec)
 	}
-	info.Decls = append(info.Decls, gen)
+	p.generated.Decls = append(p.generated.Decls, gen)
 }
 
-func (info *fileInfo) addTypedefs() {
+func (p *cgoPackage) addTypedefs() {
-	if len(info.typedefs) == 0 {
+	if len(p.typedefs) == 0 {
 		return
 	}
 	gen := &ast.GenDecl{
-		TokPos: info.importCPos,
+		TokPos: token.NoPos,
 		Tok:    token.TYPE,
 	}
-	names := make([]string, 0, len(info.typedefs))
+	names := make([]string, 0, len(p.typedefs))
-	for name := range info.typedefs {
+	for name := range p.typedefs {
 		names = append(names, name)
 	}
 	sort.Strings(names)
 	for _, name := range names {
-		typedef := info.typedefs[name]
+		typedef := p.typedefs[name]
 		typeName := "C." + name
 		isAlias := true
 		if strings.HasPrefix(name, "_Cgo_") {
@@ -405,19 +520,19 @@ func (info *fileInfo) addTypedefs() {
 		}
 		typeSpec := &ast.TypeSpec{
 			Name: &ast.Ident{
-				NamePos: info.importCPos,
+				NamePos: typedef.pos,
 				Name:    typeName,
 				Obj:     obj,
 			},
 			Type: typedef.typeExpr,
 		}
 		if isAlias {
-			typeSpec.Assign = info.importCPos
+			typeSpec.Assign = typedef.pos
 		}
 		obj.Decl = typeSpec
 		gen.Specs = append(gen.Specs, typeSpec)
 	}
-	info.Decls = append(info.Decls, gen)
+	p.generated.Decls = append(p.generated.Decls, gen)
 }
 
 // addElaboratedTypes adds C elaborated types as aliases. These are the "struct
@@ -425,21 +540,21 @@ func (info *fileInfo) addTypedefs() {
 //
 // See also:
 // https://en.cppreference.com/w/cpp/language/elaborated_type_specifier
-func (info *fileInfo) addElaboratedTypes() {
+func (p *cgoPackage) addElaboratedTypes() {
-	if len(info.elaboratedTypes) == 0 {
+	if len(p.elaboratedTypes) == 0 {
 		return
 	}
 	gen := &ast.GenDecl{
-		TokPos: info.importCPos,
+		TokPos: token.NoPos,
 		Tok:    token.TYPE,
 	}
-	names := make([]string, 0, len(info.elaboratedTypes))
+	names := make([]string, 0, len(p.elaboratedTypes))
-	for name := range info.elaboratedTypes {
+	for name := range p.elaboratedTypes {
 		names = append(names, name)
 	}
 	sort.Strings(names)
 	for _, name := range names {
-		typ := info.elaboratedTypes[name]
+		typ := p.elaboratedTypes[name]
 		typeName := "C." + name
 		obj := &ast.Object{
 			Kind: ast.Typ,
@@ -447,22 +562,22 @@ func (info *fileInfo) addElaboratedTypes() {
 		}
 		typeSpec := &ast.TypeSpec{
 			Name: &ast.Ident{
-				NamePos: info.importCPos,
+				NamePos: typ.pos,
 				Name:    typeName,
 				Obj:     obj,
 			},
-			Type: typ,
+			Type: typ.typeExpr,
 		}
 		obj.Decl = typeSpec
 		gen.Specs = append(gen.Specs, typeSpec)
 	}
-	info.Decls = append(info.Decls, gen)
+	p.generated.Decls = append(p.generated.Decls, gen)
 }
 
 // findMissingCGoNames traverses the AST and finds all C.something names. Only
 // these symbols are extracted from the parsed C AST and converted to the Go
 // equivalent.
-func (info *fileInfo) findMissingCGoNames(cursor *astutil.Cursor) bool {
+func (p *cgoPackage) findMissingCGoNames(cursor *astutil.Cursor) bool {
 	switch node := cursor.Node().(type) {
 	case *ast.SelectorExpr:
 		x, ok := node.X.(*ast.Ident)
@@ -471,10 +586,10 @@ func (info *fileInfo) findMissingCGoNames(cursor *astutil.Cursor) bool {
 		}
 		if x.Name == "C" {
 			name := node.Sel.Name
-			if _, ok := cgoBuiltinAliases[name]; ok {
+			if _, ok := builtinAliases[name]; ok {
 				name = "_Cgo_" + name
 			}
-			info.missingSymbols[name] = struct{}{}
+			p.missingSymbols[name] = struct{}{}
 		}
 	}
 	return true
@@ -484,7 +599,7 @@ func (info *fileInfo) findMissingCGoNames(cursor *astutil.Cursor) bool {
 // expressions. Such expressions are impossible to write in Go (a dot cannot be
 // used in the middle of a name) so in practice all C identifiers live in a
 // separate namespace (no _Cgo_ hacks like in gc).
-func (info *fileInfo) walker(cursor *astutil.Cursor) bool {
+func (p *cgoPackage) walker(cursor *astutil.Cursor) bool {
 	switch node := cursor.Node().(type) {
 	case *ast.CallExpr:
 		fun, ok := node.Fun.(*ast.SelectorExpr)
@@ -495,7 +610,7 @@ func (info *fileInfo) walker(cursor *astutil.Cursor) bool {
 		if !ok {
 			return true
 		}
-		if _, ok := info.functions[fun.Sel.Name]; ok && x.Name == "C" {
+		if _, ok := p.functions[fun.Sel.Name]; ok && x.Name == "C" {
 			node.Fun = &ast.Ident{
 				NamePos: x.NamePos,
 				Name:    "C." + fun.Sel.Name,
@@ -508,7 +623,7 @@ func (info *fileInfo) walker(cursor *astutil.Cursor) bool {
 		}
 		if x.Name == "C" {
 			name := "C." + node.Sel.Name
-			if _, ok := info.functions[node.Sel.Name]; ok {
+			if _, ok := p.functions[node.Sel.Name]; ok {
 				name += "$funcaddr"
 			}
 			cursor.Replace(&ast.Ident{

cgo/libclang.go

@@ -1,4 +1,4 @@
-package loader
+package cgo
 
 // This file parses a fragment of C with libclang and stores the result for AST
 // modification. It does not touch the AST itself.
@@ -47,6 +47,7 @@ CXType tinygo_clang_getCursorResultType(GoCXCursor c);
 int tinygo_clang_Cursor_getNumArguments(GoCXCursor c);
 GoCXCursor tinygo_clang_Cursor_getArgument(GoCXCursor c, unsigned i);
 CXSourceLocation tinygo_clang_getCursorLocation(GoCXCursor c);
+CXSourceRange tinygo_clang_getCursorExtent(GoCXCursor c);
 CXTranslationUnit tinygo_clang_Cursor_getTranslationUnit(GoCXCursor c);
 
 int tinygo_clang_globals_visitor(GoCXCursor c, GoCXCursor parent, CXClientData client_data);
@@ -54,8 +55,8 @@ int tinygo_clang_struct_visitor(GoCXCursor c, GoCXCursor parent, CXClientData cl
 */
 import "C"
 
-// refMap stores references to types, used for clang_visitChildren.
+// storedRefs stores references to types, used for clang_visitChildren.
-var refMap RefMap
+var storedRefs refMap
 
 var diagnosticSeverity = [...]string{
 	C.CXDiagnostic_Ignored: "ignored",
@@ -65,7 +66,7 @@ var diagnosticSeverity = [...]string{
 	C.CXDiagnostic_Fatal:   "fatal",
 }
 
-func (info *fileInfo) parseFragment(fragment string, cflags []string, posFilename string, posLine int) []error {
+func (p *cgoPackage) parseFragment(fragment string, cflags []string, posFilename string, posLine int) {
 	index := C.clang_createIndex(0, 0)
 	defer C.clang_disposeIndex(index)
 
@@ -101,7 +102,7 @@ func (info *fileInfo) parseFragment(fragment string, cflags []string, posFilenam
 		filenameC,
 		(**C.char)(cmdargsC), C.int(len(cflags)), // command line args
 		&unsavedFile, 1, // unsaved files
-		C.CXTranslationUnit_None,
+		C.CXTranslationUnit_DetailedPreprocessingRecord,
 		&unit)
 	if errCode != 0 {
 		panic("loader: failed to parse source with libclang")
@@ -109,7 +110,6 @@ func (info *fileInfo) parseFragment(fragment string, cflags []string, posFilenam
 	defer C.clang_disposeTranslationUnit(unit)
 
 	if numDiagnostics := int(C.clang_getNumDiagnostics(unit)); numDiagnostics != 0 {
-		errs := []error{}
 		addDiagnostic := func(diagnostic C.CXDiagnostic) {
 			spelling := getString(C.clang_getDiagnosticSpelling(diagnostic))
 			severity := diagnosticSeverity[C.clang_getDiagnosticSeverity(diagnostic)]
@@ -121,12 +121,12 @@ func (info *fileInfo) parseFragment(fragment string, cflags []string, posFilenam
 			filename := getString(libclangFilename)
 			if filepath.IsAbs(filename) {
 				// Relative paths for readability, like other Go parser errors.
-				relpath, err := filepath.Rel(info.Program.Dir, filename)
+				relpath, err := filepath.Rel(p.dir, filename)
 				if err == nil {
 					filename = relpath
 				}
 			}
-			errs = append(errs, &scanner.Error{
+			p.errors = append(p.errors, &scanner.Error{
 				Pos: token.Position{
 					Filename: filename,
 					Offset:   0, // not provided by clang_getPresumedLocation
@@ -146,26 +146,24 @@ func (info *fileInfo) parseFragment(fragment string, cflags []string, posFilenam
 				addDiagnostic(C.clang_getDiagnosticInSet(diagnostics, C.uint(j)))
 			}
 		}
-		return errs
+		return
 	}
 
-	ref := refMap.Put(info)
+	ref := storedRefs.Put(p)
-	defer refMap.Remove(ref)
+	defer storedRefs.Remove(ref)
 	cursor := C.tinygo_clang_getTranslationUnitCursor(unit)
 	C.tinygo_clang_visitChildren(cursor, C.CXCursorVisitor(C.tinygo_clang_globals_visitor), C.CXClientData(ref))
-
-	return nil
 }
 
 //export tinygo_clang_globals_visitor
 func tinygo_clang_globals_visitor(c, parent C.GoCXCursor, client_data C.CXClientData) C.int {
-	info := refMap.Get(unsafe.Pointer(client_data)).(*fileInfo)
+	p := storedRefs.Get(unsafe.Pointer(client_data)).(*cgoPackage)
 	kind := C.tinygo_clang_getCursorKind(c)
-	pos := info.getCursorPosition(c)
+	pos := p.getCursorPosition(c)
 	switch kind {
 	case C.CXCursor_FunctionDecl:
 		name := getString(C.tinygo_clang_getCursorSpelling(c))
-		if _, required := info.missingSymbols[name]; !required {
+		if _, required := p.missingSymbols[name]; !required {
 			return C.CXChildVisit_Continue
 		}
 		cursorType := C.tinygo_clang_getCursorType(c)
@@ -173,8 +171,10 @@ func tinygo_clang_globals_visitor(c, parent C.GoCXCursor, client_data C.CXClient
 			return C.CXChildVisit_Continue // not supported
 		}
 		numArgs := int(C.tinygo_clang_Cursor_getNumArguments(c))
-		fn := &functionInfo{}
+		fn := &functionInfo{
-		info.functions[name] = fn
+			pos: pos,
+		}
+		p.functions[name] = fn
 		for i := 0; i < numArgs; i++ {
 			arg := C.tinygo_clang_Cursor_getArgument(c, C.uint(i))
 			argName := getString(C.tinygo_clang_getCursorSpelling(arg))
@@ -184,7 +184,7 @@ func tinygo_clang_globals_visitor(c, parent C.GoCXCursor, client_data C.CXClient
 			}
 			fn.args = append(fn.args, paramInfo{
 				name:     argName,
-				typeExpr: info.makeASTType(argType, pos),
+				typeExpr: p.makeASTType(argType, pos),
 			})
 		}
 		resultType := C.tinygo_clang_getCursorResultType(c)
@@ -192,7 +192,7 @@ func tinygo_clang_globals_visitor(c, parent C.GoCXCursor, client_data C.CXClient
 			fn.results = &ast.FieldList{
 				List: []*ast.Field{
 					&ast.Field{
-						Type: info.makeASTType(resultType, pos),
+						Type: p.makeASTType(resultType, pos),
 					},
 				},
 			}
@@ -200,25 +200,106 @@ func tinygo_clang_globals_visitor(c, parent C.GoCXCursor, client_data C.CXClient
 	case C.CXCursor_StructDecl:
 		typ := C.tinygo_clang_getCursorType(c)
 		name := getString(C.tinygo_clang_getCursorSpelling(c))
-		if _, required := info.missingSymbols["struct_"+name]; !required {
+		if _, required := p.missingSymbols["struct_"+name]; !required {
 			return C.CXChildVisit_Continue
 		}
-		info.makeASTType(typ, pos)
+		p.makeASTType(typ, pos)
 	case C.CXCursor_TypedefDecl:
 		typedefType := C.tinygo_clang_getCursorType(c)
 		name := getString(C.clang_getTypedefName(typedefType))
-		if _, required := info.missingSymbols[name]; !required {
+		if _, required := p.missingSymbols[name]; !required {
 			return C.CXChildVisit_Continue
 		}
-		info.makeASTType(typedefType, pos)
+		p.makeASTType(typedefType, pos)
 	case C.CXCursor_VarDecl:
 		name := getString(C.tinygo_clang_getCursorSpelling(c))
-		if _, required := info.missingSymbols[name]; !required {
+		if _, required := p.missingSymbols[name]; !required {
 			return C.CXChildVisit_Continue
 		}
 		cursorType := C.tinygo_clang_getCursorType(c)
-		info.globals[name] = &globalInfo{
+		p.globals[name] = globalInfo{
-			typeExpr: info.makeASTType(cursorType, pos),
+			typeExpr: p.makeASTType(cursorType, pos),
+			pos:      pos,
+		}
+	case C.CXCursor_MacroDefinition:
+		name := getString(C.tinygo_clang_getCursorSpelling(c))
+		if _, required := p.missingSymbols[name]; !required {
+			return C.CXChildVisit_Continue
+		}
+		sourceRange := C.tinygo_clang_getCursorExtent(c)
+		start := C.clang_getRangeStart(sourceRange)
+		end := C.clang_getRangeEnd(sourceRange)
+		var file, endFile C.CXFile
+		var startOffset, endOffset C.unsigned
+		C.clang_getExpansionLocation(start, &file, nil, nil, &startOffset)
+		if file == nil {
+			panic("could not find file where macro is defined")
+		}
+		C.clang_getExpansionLocation(end, &endFile, nil, nil, &endOffset)
+		if file != endFile {
+			panic("expected start and end location of a #define to be in the same file")
+		}
+		if startOffset > endOffset {
+			panic("startOffset > endOffset")
+		}
+
+		// read file contents and extract the relevant byte range
+		tu := C.tinygo_clang_Cursor_getTranslationUnit(c)
+		var size C.size_t
+		sourcePtr := C.clang_getFileContents(tu, file, &size)
+		if endOffset >= C.uint(size) {
+			panic("endOffset lies after end of file")
+		}
+		source := string(((*[1 << 28]byte)(unsafe.Pointer(sourcePtr)))[startOffset:endOffset:endOffset])
+		if !strings.HasPrefix(source, name) {
+			panic(fmt.Sprintf("expected #define value to start with %#v, got %#v", name, source))
+		}
+		value := strings.TrimSpace(source[len(name):])
+		for len(value) != 0 && value[0] == '(' && value[len(value)-1] == ')' {
+			value = strings.TrimSpace(value[1 : len(value)-1])
+		}
+		if len(value) == 0 {
+			// Pretend it doesn't exist at all.
+			return C.CXChildVisit_Continue
+		}
+		// For information about integer literals:
+		// https://en.cppreference.com/w/cpp/language/integer_literal
+		if value[0] == '"' {
+			// string constant
+			p.constants[name] = constantInfo{&ast.BasicLit{pos, token.STRING, value}, pos}
+			return C.CXChildVisit_Continue
+		}
+		if value[0] == '\'' {
+			// char constant
+			p.constants[name] = constantInfo{&ast.BasicLit{pos, token.CHAR, value}, pos}
+			return C.CXChildVisit_Continue
+		}
+		// assume it's a number (int or float)
+		value = strings.Replace(value, "'", "", -1) // remove ' chars
+		value = strings.TrimRight(value, "lu")      // remove llu suffixes etc.
+		// find the first non-number
+		nonnum := byte(0)
+		for i := 0; i < len(value); i++ {
+			if value[i] < '0' || value[i] > '9' {
+				nonnum = value[i]
+				break
+			}
+		}
+		// determine number type based on the first non-number
+		switch nonnum {
+		case 0:
+			// no non-number found, must be an integer
+			p.constants[name] = constantInfo{&ast.BasicLit{pos, token.INT, value}, pos}
+		case 'x', 'X':
+			// hex integer constant
+			// TODO: may also be a floating point number per C++17.
+			p.constants[name] = constantInfo{&ast.BasicLit{pos, token.INT, value}, pos}
+		case '.', 'e':
+			// float constant
+			value = strings.TrimRight(value, "fFlL")
+			p.constants[name] = constantInfo{&ast.BasicLit{pos, token.FLOAT, value}, pos}
+		default:
+			// unknown type, ignore
 		}
 	}
 	return C.CXChildVisit_Continue
@@ -234,7 +315,7 @@ func getString(clangString C.CXString) (s string) {
 // getCursorPosition returns a usable token.Pos from a libclang cursor. If the
 // file for this cursor has not been seen before, it is read from libclang
 // (which already has the file in memory) and added to the token.FileSet.
-func (info *fileInfo) getCursorPosition(cursor C.GoCXCursor) token.Pos {
+func (p *cgoPackage) getCursorPosition(cursor C.GoCXCursor) token.Pos {
 	location := C.tinygo_clang_getCursorLocation(cursor)
 	var file C.CXFile
 	var line C.unsigned
@@ -246,7 +327,7 @@ func (info *fileInfo) getCursorPosition(cursor C.GoCXCursor) token.Pos {
 		return token.NoPos
 	}
 	filename := getString(C.clang_getFileName(file))
-	if _, ok := info.tokenFiles[filename]; !ok {
+	if _, ok := p.tokenFiles[filename]; !ok {
 		// File has not been seen before in this package, add line information
 		// now by reading the file from libclang.
 		tu := C.tinygo_clang_Cursor_getTranslationUnit(cursor)
@@ -259,16 +340,16 @@ func (info *fileInfo) getCursorPosition(cursor C.GoCXCursor) token.Pos {
 				lines = append(lines, i+1)
 			}
 		}
-		f := info.fset.AddFile(filename, -1, int(size))
+		f := p.fset.AddFile(filename, -1, int(size))
 		f.SetLines(lines)
-		info.tokenFiles[filename] = f
+		p.tokenFiles[filename] = f
 	}
-	return info.tokenFiles[filename].Pos(int(offset))
+	return p.tokenFiles[filename].Pos(int(offset))
 }
 
 // makeASTType return the ast.Expr for the given libclang type. In other words,
 // it converts a libclang type to a type in the Go AST.
-func (info *fileInfo) makeASTType(typ C.CXType, pos token.Pos) ast.Expr {
+func (p *cgoPackage) makeASTType(typ C.CXType, pos token.Pos) ast.Expr {
 	var typeName string
 	switch typ.kind {
 	case C.CXType_Char_S, C.CXType_Char_U:
@@ -329,7 +410,7 @@ func (info *fileInfo) makeASTType(typ C.CXType, pos token.Pos) ast.Expr {
 		}
 		return &ast.StarExpr{
 			Star: pos,
-			X:    info.makeASTType(pointeeType, pos),
+			X:    p.makeASTType(pointeeType, pos),
 		}
 	case C.CXType_ConstantArray:
 		return &ast.ArrayType{
@@ -339,7 +420,7 @@ func (info *fileInfo) makeASTType(typ C.CXType, pos token.Pos) ast.Expr {
 				Kind:     token.INT,
 				Value:    strconv.FormatInt(int64(C.clang_getArraySize(typ)), 10),
 			},
-			Elt: info.makeASTType(C.clang_getElementType(typ), pos),
+			Elt: p.makeASTType(C.clang_getElementType(typ), pos),
 		}
 	case C.CXType_FunctionProto:
 		// Be compatible with gc, which uses the *[0]byte type for function
@@ -360,11 +441,11 @@ func (info *fileInfo) makeASTType(typ C.CXType, pos token.Pos) ast.Expr {
 		}
 	case C.CXType_Typedef:
 		name := getString(C.clang_getTypedefName(typ))
-		if _, ok := info.typedefs[name]; !ok {
+		if _, ok := p.typedefs[name]; !ok {
-			info.typedefs[name] = nil // don't recurse
+			p.typedefs[name] = nil // don't recurse
 			c := C.tinygo_clang_getTypeDeclaration(typ)
 			underlyingType := C.tinygo_clang_getTypedefDeclUnderlyingType(c)
-			expr := info.makeASTType(underlyingType, pos)
+			expr := p.makeASTType(underlyingType, pos)
 			if strings.HasPrefix(name, "_Cgo_") {
 				expr := expr.(*ast.Ident)
 				typeSize := C.clang_Type_getSizeOf(underlyingType)
@@ -406,8 +487,9 @@ func (info *fileInfo) makeASTType(typ C.CXType, pos token.Pos) ast.Expr {
 					}
 				}
 			}
-			info.typedefs[name] = &typedefInfo{
+			p.typedefs[name] = &typedefInfo{
 				typeExpr: expr,
+				pos:      pos,
 			}
 		}
 		return &ast.Ident{
@@ -418,7 +500,7 @@ func (info *fileInfo) makeASTType(typ C.CXType, pos token.Pos) ast.Expr {
 		underlying := C.clang_Type_getNamedType(typ)
 		switch underlying.kind {
 		case C.CXType_Record:
-			return info.makeASTType(underlying, pos)
+			return p.makeASTType(underlying, pos)
 		default:
 			panic("unknown elaborated type")
 		}
@@ -434,23 +516,26 @@ func (info *fileInfo) makeASTType(typ C.CXType, pos token.Pos) ast.Expr {
 		default:
 			panic("unknown record declaration")
 		}
-		if _, ok := info.elaboratedTypes[cgoName]; !ok {
+		if _, ok := p.elaboratedTypes[cgoName]; !ok {
-			info.elaboratedTypes[cgoName] = nil // predeclare (to avoid endless recursion)
+			p.elaboratedTypes[cgoName] = nil // predeclare (to avoid endless recursion)
 			fieldList := &ast.FieldList{
 				Opening: pos,
 				Closing: pos,
 			}
-			ref := refMap.Put(struct {
+			ref := storedRefs.Put(struct {
 				fieldList *ast.FieldList
-				info      *fileInfo
+				pkg       *cgoPackage
-			}{fieldList, info})
+			}{fieldList, p})
-			defer refMap.Remove(ref)
+			defer storedRefs.Remove(ref)
 			C.tinygo_clang_visitChildren(cursor, C.CXCursorVisitor(C.tinygo_clang_struct_visitor), C.CXClientData(ref))
 			switch C.tinygo_clang_getCursorKind(cursor) {
 			case C.CXCursor_StructDecl:
-				info.elaboratedTypes[cgoName] = &ast.StructType{
+				p.elaboratedTypes[cgoName] = &elaboratedTypeInfo{
+					typeExpr: &ast.StructType{
 						Struct: pos,
 						Fields: fieldList,
+					},
+					pos: pos,
 				}
 			case C.CXCursor_UnionDecl:
 				if len(fieldList.List) > 1 {
@@ -480,9 +565,12 @@ func (info *fileInfo) makeASTType(typ C.CXType, pos token.Pos) ast.Expr {
 					}
 					fieldList.List = append([]*ast.Field{unionMarker}, fieldList.List...)
 				}
-				info.elaboratedTypes[cgoName] = &ast.StructType{
+				p.elaboratedTypes[cgoName] = &elaboratedTypeInfo{
+					typeExpr: &ast.StructType{
 						Struct: pos,
 						Fields: fieldList,
+					},
+					pos: pos,
 				}
 			default:
 				panic("unreachable")
@@ -506,23 +594,23 @@ func (info *fileInfo) makeASTType(typ C.CXType, pos token.Pos) ast.Expr {
 
 //export tinygo_clang_struct_visitor
 func tinygo_clang_struct_visitor(c, parent C.GoCXCursor, client_data C.CXClientData) C.int {
-	passed := refMap.Get(unsafe.Pointer(client_data)).(struct {
+	passed := storedRefs.Get(unsafe.Pointer(client_data)).(struct {
 		fieldList *ast.FieldList
-		info      *fileInfo
+		pkg       *cgoPackage
 	})
 	fieldList := passed.fieldList
-	info := passed.info
+	p := passed.pkg
 	if C.tinygo_clang_getCursorKind(c) != C.CXCursor_FieldDecl {
 		panic("expected field inside cursor")
 	}
 	name := getString(C.tinygo_clang_getCursorSpelling(c))
 	typ := C.tinygo_clang_getCursorType(c)
 	field := &ast.Field{
-		Type: info.makeASTType(typ, info.getCursorPosition(c)),
+		Type: p.makeASTType(typ, p.getCursorPosition(c)),
 	}
 	field.Names = []*ast.Ident{
 		&ast.Ident{
-			NamePos: info.getCursorPosition(c),
+			NamePos: p.getCursorPosition(c),
 			Name:    name,
 			Obj: &ast.Object{
 				Kind: ast.Var,

cgo/libclang_config.go

@@ -1,6 +1,6 @@
 // +build !byollvm
 
-package loader
+package cgo
 
 /*
 #cgo linux  CFLAGS: -I/usr/lib/llvm-8/include

cgo/libclang_stubs.c

@@ -49,6 +49,10 @@ CXSourceLocation tinygo_clang_getCursorLocation(CXCursor c) {
 	return clang_getCursorLocation(c);
 }
 
+CXSourceRange tinygo_clang_getCursorExtent(CXCursor c) {
+	return clang_getCursorExtent(c);
+}
+
 CXTranslationUnit tinygo_clang_Cursor_getTranslationUnit(CXCursor c) {
 	return clang_Cursor_getTranslationUnit(c);
 }

cgo/sync.go

@@ -1,4 +1,4 @@
-package loader
+package cgo
 
 import (
 	"sync"
@@ -8,17 +8,17 @@ import (
 // #include <stdlib.h>
 import "C"
 
-// RefMap is a convenient way to store opaque references that can be passed to
+// refMap is a convenient way to store opaque references that can be passed to
 // C. It is useful if an API uses function pointers and you cannot pass a Go
 // pointer but only a C pointer.
-type RefMap struct {
+type refMap struct {
 	refs map[unsafe.Pointer]interface{}
 	lock sync.Mutex
 }
 
 // Put stores a value in the map. It can later be retrieved using Get. It must
 // be removed using Remove to avoid memory leaks.
-func (m *RefMap) Put(v interface{}) unsafe.Pointer {
+func (m *refMap) Put(v interface{}) unsafe.Pointer {
 	m.lock.Lock()
 	defer m.lock.Unlock()
 	if m.refs == nil {
@@ -31,14 +31,14 @@ func (m *RefMap) Put(v interface{}) unsafe.Pointer {
 
 // Get returns a stored value previously inserted with Put. Use the same
 // reference as you got from Put.
-func (m *RefMap) Get(ref unsafe.Pointer) interface{} {
+func (m *refMap) Get(ref unsafe.Pointer) interface{} {
 	m.lock.Lock()
 	defer m.lock.Unlock()
 	return m.refs[ref]
 }
 
 // Remove deletes a single reference from the map.
-func (m *RefMap) Remove(ref unsafe.Pointer) {
+func (m *refMap) Remove(ref unsafe.Pointer) {
 	m.lock.Lock()
 	defer m.lock.Unlock()
 	delete(m.refs, ref)

compiler/calls.go

@@ -55,6 +55,25 @@ func (c *Compiler) expandFormalParamType(t llvm.Type) []llvm.Type {
 	}
 }
 
+// Expand an argument type to a list of offsets from the start of the object.
+// Used together with expandFormalParam to get the offset of each value from the
+// start of the non-expanded value.
+func (c *Compiler) expandFormalParamOffsets(t llvm.Type) []uint64 {
+	switch t.TypeKind() {
+	case llvm.StructTypeKind:
+		fields := c.flattenAggregateTypeOffsets(t)
+		if len(fields) <= MaxFieldsPerParam {
+			return fields
+		} else {
+			// failed to lower
+			return []uint64{0}
+		}
+	default:
+		// TODO: split small arrays
+		return []uint64{0}
+	}
+}
+
 // Equivalent of expandFormalParamType for parameter values.
 func (c *Compiler) expandFormalParam(v llvm.Value) []llvm.Value {
 	switch v.Type().TypeKind() {
@@ -92,6 +111,27 @@ func (c *Compiler) flattenAggregateType(t llvm.Type) []llvm.Type {
 	}
 }
 
+// Return the offsets from the start of the object if this object type were
+// flattened like in flattenAggregate. Used together with flattenAggregate to
+// know the start indices of each value in the non-flattened object.
+func (c *Compiler) flattenAggregateTypeOffsets(t llvm.Type) []uint64 {
+	switch t.TypeKind() {
+	case llvm.StructTypeKind:
+		fields := make([]uint64, 0, t.StructElementTypesCount())
+		for fieldIndex, field := range t.StructElementTypes() {
+			suboffsets := c.flattenAggregateTypeOffsets(field)
+			offset := c.targetData.ElementOffset(t, fieldIndex)
+			for i := range suboffsets {
+				suboffsets[i] += offset
+			}
+			fields = append(fields, suboffsets...)
+		}
+		return fields
+	default:
+		return []uint64{0}
+	}
+}
+
 // Break down a struct into its elementary types for argument passing. The value
 // equivalent of flattenAggregateType
 func (c *Compiler) flattenAggregate(v llvm.Value) []llvm.Value {

compiler/compiler.go

@@ -52,7 +52,6 @@ type Compiler struct {
 	dibuilder               *llvm.DIBuilder
 	cu                      llvm.Metadata
 	difiles                 map[string]llvm.Metadata
-	ditypes                 map[string]llvm.Metadata
 	machine                 llvm.TargetMachine
 	targetData              llvm.TargetData
 	intType                 llvm.Type
@@ -96,7 +95,6 @@ func NewCompiler(pkgName string, config Config) (*Compiler, error) {
 	c := &Compiler{
 		Config:  config,
 		difiles: make(map[string]llvm.Metadata),
-		ditypes: make(map[string]llvm.Metadata),
 	}
 
 	target, err := llvm.GetTargetFromTriple(config.Triple)
@@ -153,11 +151,15 @@ func (c *Compiler) TargetData() llvm.TargetData {
 func (c *Compiler) selectGC() string {
 	gc := c.GC
 	if gc == "" {
-		gc = "dumb"
+		gc = "leaking"
 	}
 	return gc
 }
 
+func (c *Compiler) gcIsPrecise() bool {
+	return c.GC == "precise"
+}
+
 // Compile the given package path or .go file path. Return an error when this
 // fails (in any stage).
 func (c *Compiler) Compile(mainPath string) []error {
@@ -197,7 +199,7 @@ func (c *Compiler) Compile(mainPath string) []error {
 		},
 		ShouldOverlay: func(path string) bool {
 			switch path {
-			case "machine", "os", "reflect", "runtime", "sync":
+			case "machine", "os", "reflect", "runtime", "runtime/volatile", "sync":
 				return true
 			default:
 				if strings.HasPrefix(path, "device/") || strings.HasPrefix(path, "examples/") {
@@ -252,7 +254,7 @@ func (c *Compiler) Compile(mainPath string) []error {
 	// Initialize debug information.
 	if c.Debug {
 		c.cu = c.dibuilder.CreateCompileUnit(llvm.DICompileUnit{
-			Language:  llvm.DW_LANG_Go,
+			Language:  0xb, // DW_LANG_C99 (0xc, off-by-one?)
 			File:      mainPath,
 			Dir:       "",
 			Producer:  "TinyGo",
@@ -387,6 +389,13 @@ func (c *Compiler) Compile(mainPath string) []error {
 				llvm.ConstInt(c.ctx.Int32Type(), 3, false).ConstantAsMetadata(), // DWARF version
 			}),
 		)
+		c.mod.AddNamedMetadataOperand("llvm.module.flags",
+			c.ctx.MDNode([]llvm.Metadata{
+				llvm.ConstInt(c.ctx.Int32Type(), 1, false).ConstantAsMetadata(),
+				llvm.GlobalContext().MDString("Dwarf Version"),
+				llvm.ConstInt(c.ctx.Int32Type(), 4, false).ConstantAsMetadata(),
+			}),
+		)
 		c.dibuilder.Finalize()
 	}
 
@@ -550,15 +559,23 @@ func isPointer(typ types.Type) bool {
 
 // Get the DWARF type for this Go type.
 func (c *Compiler) getDIType(typ types.Type) llvm.Metadata {
-	name := typ.String()
-	if dityp, ok := c.ditypes[name]; ok {
-		return dityp
-	} else {
 	llvmType := c.getLLVMType(typ)
 	sizeInBytes := c.targetData.TypeAllocSize(llvmType)
-		var encoding llvm.DwarfTypeEncoding
 	switch typ := typ.(type) {
+	case *types.Array:
+		return c.dibuilder.CreateArrayType(llvm.DIArrayType{
+			SizeInBits:  sizeInBytes * 8,
+			AlignInBits: uint32(c.targetData.ABITypeAlignment(llvmType)) * 8,
+			ElementType: c.getDIType(typ.Elem()),
+			Subscripts: []llvm.DISubrange{
+				llvm.DISubrange{
+					Lo:    0,
+					Count: typ.Len(),
+				},
+			},
+		})
 	case *types.Basic:
+		var encoding llvm.DwarfTypeEncoding
 		if typ.Info()&types.IsBoolean != 0 {
 			encoding = llvm.DW_ATE_boolean
 		} else if typ.Info()&types.IsFloat != 0 {
@@ -570,19 +587,138 @@ func (c *Compiler) getDIType(typ types.Type) llvm.Metadata {
 		} else if typ.Info()&types.IsInteger != 0 {
 			encoding = llvm.DW_ATE_signed
 		} else if typ.Kind() == types.UnsafePointer {
-				encoding = llvm.DW_ATE_address
+			return c.dibuilder.CreatePointerType(llvm.DIPointerType{
+				Name:         "unsafe.Pointer",
+				SizeInBits:   c.targetData.TypeAllocSize(llvmType) * 8,
+				AlignInBits:  uint32(c.targetData.ABITypeAlignment(llvmType)) * 8,
+				AddressSpace: 0,
+			})
+		} else if typ.Info()&types.IsString != 0 {
+			return c.dibuilder.CreateStructType(llvm.Metadata{}, llvm.DIStructType{
+				Name:        "string",
+				SizeInBits:  sizeInBytes * 8,
+				AlignInBits: uint32(c.targetData.ABITypeAlignment(llvmType)) * 8,
+				Elements: []llvm.Metadata{
+					c.dibuilder.CreateMemberType(llvm.Metadata{}, llvm.DIMemberType{
+						Name:         "ptr",
+						SizeInBits:   c.targetData.TypeAllocSize(c.i8ptrType) * 8,
+						AlignInBits:  uint32(c.targetData.ABITypeAlignment(c.i8ptrType)) * 8,
+						OffsetInBits: 0,
+						Type:         c.getDIType(types.NewPointer(types.Typ[types.Byte])),
+					}),
+					c.dibuilder.CreateMemberType(llvm.Metadata{}, llvm.DIMemberType{
+						Name:         "len",
+						SizeInBits:   c.targetData.TypeAllocSize(c.uintptrType) * 8,
+						AlignInBits:  uint32(c.targetData.ABITypeAlignment(c.uintptrType)) * 8,
+						OffsetInBits: c.targetData.ElementOffset(llvmType, 1) * 8,
+						Type:         c.getDIType(types.Typ[types.Uintptr]),
+					}),
+				},
+			})
+		} else {
+			panic("unknown basic type")
 		}
-		case *types.Pointer:
+		return c.dibuilder.CreateBasicType(llvm.DIBasicType{
-			encoding = llvm.DW_ATE_address
+			Name:       typ.String(),
-		}
-		// TODO: other types
-		dityp = c.dibuilder.CreateBasicType(llvm.DIBasicType{
-			Name:       name,
 			SizeInBits: sizeInBytes * 8,
 			Encoding:   encoding,
 		})
-		c.ditypes[name] = dityp
+	case *types.Chan:
-		return dityp
+		return c.getDIType(types.NewPointer(c.ir.Program.ImportedPackage("runtime").Members["channel"].(*ssa.Type).Type()))
+	case *types.Interface:
+		return c.getDIType(c.ir.Program.ImportedPackage("runtime").Members["_interface"].(*ssa.Type).Type())
+	case *types.Map:
+		return c.getDIType(types.NewPointer(c.ir.Program.ImportedPackage("runtime").Members["hashmap"].(*ssa.Type).Type()))
+	case *types.Named:
+		return c.dibuilder.CreateTypedef(llvm.DITypedef{
+			Type: c.getDIType(typ.Underlying()),
+			Name: typ.String(),
+		})
+	case *types.Pointer:
+		return c.dibuilder.CreatePointerType(llvm.DIPointerType{
+			Pointee:      c.getDIType(typ.Elem()),
+			SizeInBits:   c.targetData.TypeAllocSize(llvmType) * 8,
+			AlignInBits:  uint32(c.targetData.ABITypeAlignment(llvmType)) * 8,
+			AddressSpace: 0,
+		})
+	case *types.Signature:
+		// actually a closure
+		fields := llvmType.StructElementTypes()
+		return c.dibuilder.CreateStructType(llvm.Metadata{}, llvm.DIStructType{
+			SizeInBits:  sizeInBytes * 8,
+			AlignInBits: uint32(c.targetData.ABITypeAlignment(llvmType)) * 8,
+			Elements: []llvm.Metadata{
+				c.dibuilder.CreateMemberType(llvm.Metadata{}, llvm.DIMemberType{
+					Name:         "context",
+					SizeInBits:   c.targetData.TypeAllocSize(fields[1]) * 8,
+					AlignInBits:  uint32(c.targetData.ABITypeAlignment(fields[1])) * 8,
+					OffsetInBits: 0,
+					Type:         c.getDIType(types.Typ[types.UnsafePointer]),
+				}),
+				c.dibuilder.CreateMemberType(llvm.Metadata{}, llvm.DIMemberType{
+					Name:         "fn",
+					SizeInBits:   c.targetData.TypeAllocSize(fields[0]) * 8,
+					AlignInBits:  uint32(c.targetData.ABITypeAlignment(fields[0])) * 8,
+					OffsetInBits: c.targetData.ElementOffset(llvmType, 1) * 8,
+					Type:         c.getDIType(types.Typ[types.UnsafePointer]),
+				}),
+			},
+		})
+	case *types.Slice:
+		fields := llvmType.StructElementTypes()
+		return c.dibuilder.CreateStructType(llvm.Metadata{}, llvm.DIStructType{
+			Name:        typ.String(),
+			SizeInBits:  sizeInBytes * 8,
+			AlignInBits: uint32(c.targetData.ABITypeAlignment(llvmType)) * 8,
+			Elements: []llvm.Metadata{
+				c.dibuilder.CreateMemberType(llvm.Metadata{}, llvm.DIMemberType{
+					Name:         "ptr",
+					SizeInBits:   c.targetData.TypeAllocSize(fields[0]) * 8,
+					AlignInBits:  uint32(c.targetData.ABITypeAlignment(fields[0])) * 8,
+					OffsetInBits: 0,
+					Type:         c.getDIType(types.NewPointer(typ.Elem())),
+				}),
+				c.dibuilder.CreateMemberType(llvm.Metadata{}, llvm.DIMemberType{
+					Name:         "len",
+					SizeInBits:   c.targetData.TypeAllocSize(c.uintptrType) * 8,
+					AlignInBits:  uint32(c.targetData.ABITypeAlignment(c.uintptrType)) * 8,
+					OffsetInBits: c.targetData.ElementOffset(llvmType, 1) * 8,
+					Type:         c.getDIType(types.Typ[types.Uintptr]),
+				}),
+				c.dibuilder.CreateMemberType(llvm.Metadata{}, llvm.DIMemberType{
+					Name:         "cap",
+					SizeInBits:   c.targetData.TypeAllocSize(c.uintptrType) * 8,
+					AlignInBits:  uint32(c.targetData.ABITypeAlignment(c.uintptrType)) * 8,
+					OffsetInBits: c.targetData.ElementOffset(llvmType, 2) * 8,
+					Type:         c.getDIType(types.Typ[types.Uintptr]),
+				}),
+			},
+		})
+	case *types.Struct:
+		elements := make([]llvm.Metadata, typ.NumFields())
+		for i := range elements {
+			field := typ.Field(i)
+			fieldType := field.Type()
+			if _, ok := fieldType.Underlying().(*types.Pointer); ok {
+				// XXX hack to avoid recursive types
+				fieldType = types.Typ[types.UnsafePointer]
+			}
+			llvmField := c.getLLVMType(fieldType)
+			elements[i] = c.dibuilder.CreateMemberType(llvm.Metadata{}, llvm.DIMemberType{
+				Name:         field.Name(),
+				SizeInBits:   c.targetData.TypeAllocSize(llvmField) * 8,
+				AlignInBits:  uint32(c.targetData.ABITypeAlignment(llvmField)) * 8,
+				OffsetInBits: c.targetData.ElementOffset(llvmType, i) * 8,
+				Type:         c.getDIType(fieldType),
+			})
+		}
+		return c.dibuilder.CreateStructType(llvm.Metadata{}, llvm.DIStructType{
+			SizeInBits:  sizeInBytes * 8,
+			AlignInBits: uint32(c.targetData.ABITypeAlignment(llvmType)) * 8,
+			Elements:    elements,
+		})
+	default:
+		panic("unknown type while generating DWARF debug type: " + typ.String())
 	}
 }
 
@@ -696,6 +832,14 @@ func (c *Compiler) parseFunc(frame *Frame) {
 		frame.fn.LLVMFn.SetFunctionCallConv(85) // CallingConv::AVR_SIGNAL
 	}
 
+	// Some functions have a pragma controlling the inlining level.
+	switch frame.fn.Inline() {
+	case ir.InlineHint:
+		// Add LLVM inline hint to functions with //go:inline pragma.
+		inline := c.ctx.CreateEnumAttribute(llvm.AttributeKindID("inlinehint"), 0)
+		frame.fn.LLVMFn.AddFunctionAttr(inline)
+	}
+
 	// Add debug info, if needed.
 	if c.Debug {
 		if frame.fn.Synthetic == "package initializer" {
@@ -733,15 +877,30 @@ func (c *Compiler) parseFunc(frame *Frame) {
 		// Add debug information to this parameter (if available)
 		if c.Debug && frame.fn.Syntax() != nil {
 			pos := c.ir.Program.Fset.Position(frame.fn.Syntax().Pos())
-			c.dibuilder.CreateParameterVariable(frame.difunc, llvm.DIParameterVariable{
+			diType := c.getDIType(param.Type())
+			dbgParam := c.dibuilder.CreateParameterVariable(frame.difunc, llvm.DIParameterVariable{
 				Name:           param.Name(),
 				File:           c.difiles[pos.Filename],
 				Line:           pos.Line,
-				Type:           c.getDIType(param.Type()),
+				Type:           diType,
 				AlwaysPreserve: true,
 				ArgNo:          i + 1,
 			})
-			// TODO: set the value of this parameter.
+			loc := c.builder.GetCurrentDebugLocation()
+			if len(fields) == 1 {
+				expr := c.dibuilder.CreateExpression(nil)
+				c.dibuilder.InsertValueAtEnd(fields[0], dbgParam, expr, loc, entryBlock)
+			} else {
+				fieldOffsets := c.expandFormalParamOffsets(llvmType)
+				for i, field := range fields {
+					expr := c.dibuilder.CreateExpression([]int64{
+						0x1000,                     // DW_OP_LLVM_fragment
+						int64(fieldOffsets[i]) * 8, // offset in bits
+						int64(c.targetData.TypeAllocSize(field.Type())) * 8, // size in bits
+					})
+					c.dibuilder.InsertValueAtEnd(field, dbgParam, expr, loc, entryBlock)
+				}
+			}
 		}
 	}
 
@@ -907,6 +1066,7 @@ func (c *Compiler) parseInstr(frame *Frame, instr ssa.Instruction) {
 	case *ssa.Store:
 		llvmAddr := c.getValue(frame, instr.Addr)
 		llvmVal := c.getValue(frame, instr.Val)
+		c.emitNilCheck(frame, llvmAddr, "store")
 		if c.targetData.TypeAllocSize(llvmVal.Type()) == 0 {
 			// nothing to store
 			return
@@ -1117,17 +1277,22 @@ func (c *Compiler) parseCall(frame *Frame, instr *ssa.CallCommon) (llvm.Value, e
 
 	// Try to call the function directly for trivially static calls.
 	if fn := instr.StaticCallee(); fn != nil {
-		switch fn.RelString(nil) {
+		name := fn.RelString(nil)
-		case "device/arm.ReadRegister":
+		switch {
+		case name == "device/arm.ReadRegister":
 			return c.emitReadRegister(instr.Args)
-		case "device/arm.Asm", "device/avr.Asm":
+		case name == "device/arm.Asm" || name == "device/avr.Asm":
 			return c.emitAsm(instr.Args)
-		case "device/arm.AsmFull", "device/avr.AsmFull":
+		case name == "device/arm.AsmFull" || name == "device/avr.AsmFull":
 			return c.emitAsmFull(frame, instr)
-		case "device/arm.SVCall0", "device/arm.SVCall1", "device/arm.SVCall2", "device/arm.SVCall3", "device/arm.SVCall4":
+		case strings.HasPrefix(name, "device/arm.SVCall"):
 			return c.emitSVCall(frame, instr.Args)
-		case "syscall.Syscall", "syscall.Syscall6", "syscall.Syscall9":
+		case strings.HasPrefix(name, "syscall.Syscall"):
 			return c.emitSyscall(frame, instr)
+		case strings.HasPrefix(name, "runtime/volatile.Load"):
+			return c.emitVolatileLoad(frame, instr)
+		case strings.HasPrefix(name, "runtime/volatile.Store"):
+			return c.emitVolatileStore(frame, instr)
 		}
 
 		targetFunc := c.ir.GetFunction(fn)
@@ -1416,7 +1581,7 @@ func (c *Compiler) parseExpr(frame *Frame, expr ssa.Value) (llvm.Value, error) {
 		return c.parseMakeClosure(frame, expr)
 	case *ssa.MakeInterface:
 		val := c.getValue(frame, expr.X)
-		return c.parseMakeInterface(val, expr.X.Type(), expr.Pos())
+		return c.parseMakeInterface(val, expr.X.Type(), expr.Pos()), nil
 	case *ssa.MakeMap:
 		mapType := expr.Type().Underlying().(*types.Map)
 		llvmKeyType := c.getLLVMType(mapType.Key().Underlying())
@@ -1425,7 +1590,16 @@ func (c *Compiler) parseExpr(frame *Frame, expr ssa.Value) (llvm.Value, error) {
 		valueSize := c.targetData.TypeAllocSize(llvmValueType)
 		llvmKeySize := llvm.ConstInt(c.ctx.Int8Type(), keySize, false)
 		llvmValueSize := llvm.ConstInt(c.ctx.Int8Type(), valueSize, false)
-		hashmap := c.createRuntimeCall("hashmapMake", []llvm.Value{llvmKeySize, llvmValueSize}, "")
+		sizeHint := llvm.ConstInt(c.uintptrType, 8, false)
+		if expr.Reserve != nil {
+			sizeHint = c.getValue(frame, expr.Reserve)
+			var err error
+			sizeHint, err = c.parseConvert(expr.Reserve.Type(), types.Typ[types.Uintptr], sizeHint, expr.Pos())
+			if err != nil {
+				return llvm.Value{}, err
+			}
+		}
+		hashmap := c.createRuntimeCall("hashmapMake", []llvm.Value{llvmKeySize, llvmValueSize, sizeHint}, "")
 		return hashmap, nil
 	case *ssa.MakeSlice:
 		sliceLen := c.getValue(frame, expr.Len)
@@ -1944,10 +2118,12 @@ func (c *Compiler) parseBinOp(op token.Token, typ types.Type, x, y llvm.Value, p
 		default:
 			return llvm.Value{}, c.makeError(pos, "binop on interface: "+op.String())
 		}
-	case *types.Map, *types.Pointer:
+	case *types.Chan, *types.Map, *types.Pointer:
 		// Maps are in general not comparable, but can be compared against nil
 		// (which is a nil pointer). This means they can be trivially compared
 		// by treating them as a pointer.
+		// Channels behave as pointers in that they are equal as long as they
+		// are created with the same call to make or if both are nil.
 		switch op {
 		case token.EQL: // ==
 			return c.builder.CreateICmp(llvm.IntEQ, x, y, ""), nil

compiler/gc-precise.go

@@ -0,0 +1,104 @@
+package compiler
+
+import (
+	"math/big"
+
+	"tinygo.org/x/go-llvm"
+)
+
+func (c *Compiler) addGlobalsBitmap() {
+	if c.mod.NamedGlobal("runtime.trackedGlobalsStart").IsNil() {
+		return // nothing to do: no GC in use
+	}
+
+	var trackedGlobals []llvm.Value
+	var trackedGlobalTypes []llvm.Type
+	for global := c.mod.FirstGlobal(); !global.IsNil(); global = llvm.NextGlobal(global) {
+		if global.IsDeclaration() {
+			continue
+		}
+		typ := global.Type().ElementType()
+		ptrs := c.getPointerBitmap(typ, global.Name())
+		if ptrs.BitLen() == 0 {
+			continue
+		}
+		trackedGlobals = append(trackedGlobals, global)
+		trackedGlobalTypes = append(trackedGlobalTypes, typ)
+	}
+
+	//
+	globalsBundleType := c.ctx.StructType(trackedGlobalTypes, false)
+	globalsBundle := llvm.AddGlobal(c.mod, globalsBundleType, "tinygo.trackedGlobals")
+	globalsBundle.SetLinkage(llvm.InternalLinkage)
+	globalsBundle.SetUnnamedAddr(true)
+	initializer := llvm.Undef(globalsBundleType)
+	for i, global := range trackedGlobals {
+		initializer = llvm.ConstInsertValue(initializer, global.Initializer(), []uint32{uint32(i)})
+		gep := llvm.ConstGEP(globalsBundle, []llvm.Value{
+			llvm.ConstInt(c.ctx.Int32Type(), 0, false),
+			llvm.ConstInt(c.ctx.Int32Type(), uint64(i), false),
+		})
+		global.ReplaceAllUsesWith(gep)
+		global.EraseFromParentAsGlobal()
+	}
+	globalsBundle.SetInitializer(initializer)
+
+	trackedGlobalsStart := llvm.ConstPtrToInt(globalsBundle, c.uintptrType)
+	c.mod.NamedGlobal("runtime.trackedGlobalsStart").SetInitializer(trackedGlobalsStart)
+
+	alignment := c.targetData.PrefTypeAlignment(c.i8ptrType)
+	trackedGlobalsLength := llvm.ConstInt(c.uintptrType, c.targetData.TypeAllocSize(globalsBundleType)/uint64(alignment), false)
+	c.mod.NamedGlobal("runtime.trackedGlobalsLength").SetInitializer(trackedGlobalsLength)
+
+	bitmapBytes := c.getPointerBitmap(globalsBundleType, "globals bundle").Bytes()
+	bitmapValues := make([]llvm.Value, len(bitmapBytes))
+	for i, b := range bitmapBytes {
+		bitmapValues[len(bitmapBytes)-i-1] = llvm.ConstInt(c.ctx.Int8Type(), uint64(b), false)
+	}
+	bitmapArray := llvm.ConstArray(llvm.ArrayType(c.ctx.Int8Type(), len(bitmapBytes)), bitmapValues)
+	bitmapNew := llvm.AddGlobal(c.mod, bitmapArray.Type(), "runtime.trackedGlobalsBitmap.tmp")
+	bitmapOld := c.mod.NamedGlobal("runtime.trackedGlobalsBitmap")
+	bitmapOld.ReplaceAllUsesWith(bitmapNew)
+	bitmapNew.SetInitializer(bitmapArray)
+	bitmapNew.SetName("runtime.trackedGlobalsBitmap")
+}
+
+func (c *Compiler) getPointerBitmap(typ llvm.Type, name string) *big.Int {
+	alignment := c.targetData.PrefTypeAlignment(c.i8ptrType)
+	switch typ.TypeKind() {
+	case llvm.IntegerTypeKind, llvm.FloatTypeKind, llvm.DoubleTypeKind:
+		return big.NewInt(0)
+	case llvm.PointerTypeKind:
+		return big.NewInt(1)
+	case llvm.StructTypeKind:
+		ptrs := big.NewInt(0)
+		for i, subtyp := range typ.StructElementTypes() {
+			subptrs := c.getPointerBitmap(subtyp, name)
+			if subptrs.BitLen() == 0 {
+				continue
+			}
+			offset := c.targetData.ElementOffset(typ, i)
+			if offset%uint64(alignment) != 0 {
+				panic("precise GC: global contains unaligned pointer: " + name)
+			}
+			subptrs.Lsh(subptrs, uint(offset)/uint(alignment))
+			ptrs.Or(ptrs, subptrs)
+		}
+		return ptrs
+	case llvm.ArrayTypeKind:
+		subtyp := typ.ElementType()
+		subptrs := c.getPointerBitmap(subtyp, name)
+		ptrs := big.NewInt(0)
+		if subptrs.BitLen() == 0 {
+			return ptrs
+		}
+		elementSize := c.targetData.TypeAllocSize(subtyp)
+		for i := 0; i < typ.ArrayLength(); i++ {
+			ptrs.Lsh(ptrs, uint(elementSize)/uint(alignment))
+			ptrs.Or(ptrs, subptrs)
+		}
+		return ptrs
+	default:
+		panic("unknown type kind of global: " + name)
+	}
+}

compiler/interface.go

@@ -22,13 +22,10 @@ import (
 // value field.
 //
 // An interface value is a {typecode, value} tuple, or {i16, i8*} to be exact.
-func (c *Compiler) parseMakeInterface(val llvm.Value, typ types.Type, pos token.Pos) (llvm.Value, error) {
+func (c *Compiler) parseMakeInterface(val llvm.Value, typ types.Type, pos token.Pos) llvm.Value {
 	itfValue := c.emitPointerPack([]llvm.Value{val})
 	itfTypeCodeGlobal := c.getTypeCode(typ)
-	itfMethodSetGlobal, err := c.getTypeMethodSet(typ)
+	itfMethodSetGlobal := c.getTypeMethodSet(typ)
-	if err != nil {
-		return llvm.Value{}, nil
-	}
 	itfConcreteTypeGlobal := c.mod.NamedGlobal("typeInInterface:" + itfTypeCodeGlobal.Name())
 	if itfConcreteTypeGlobal.IsNil() {
 		typeInInterface := c.mod.GetTypeByName("runtime.typeInInterface")
@@ -41,7 +38,7 @@ func (c *Compiler) parseMakeInterface(val llvm.Value, typ types.Type, pos token.
 	itf := llvm.Undef(c.mod.GetTypeByName("runtime._interface"))
 	itf = c.builder.CreateInsertValue(itf, itfTypeCode, 0, "")
 	itf = c.builder.CreateInsertValue(itf, itfValue, 1, "")
-	return itf, nil
+	return itf
 }
 
 // getTypeCode returns a reference to a type code.
@@ -155,18 +152,18 @@ func getTypeCodeName(t types.Type) string {
 
 // getTypeMethodSet returns a reference (GEP) to a global method set. This
 // method set should be unreferenced after the interface lowering pass.
-func (c *Compiler) getTypeMethodSet(typ types.Type) (llvm.Value, error) {
+func (c *Compiler) getTypeMethodSet(typ types.Type) llvm.Value {
 	global := c.mod.NamedGlobal(typ.String() + "$methodset")
 	zero := llvm.ConstInt(c.ctx.Int32Type(), 0, false)
 	if !global.IsNil() {
 		// the method set already exists
-		return llvm.ConstGEP(global, []llvm.Value{zero, zero}), nil
+		return llvm.ConstGEP(global, []llvm.Value{zero, zero})
 	}
 
 	ms := c.ir.Program.MethodSets.MethodSet(typ)
 	if ms.Len() == 0 {
 		// no methods, so can leave that one out
-		return llvm.ConstPointerNull(llvm.PointerType(c.mod.GetTypeByName("runtime.interfaceMethodInfo"), 0)), nil
+		return llvm.ConstPointerNull(llvm.PointerType(c.mod.GetTypeByName("runtime.interfaceMethodInfo"), 0))
 	}
 
 	methods := make([]llvm.Value, ms.Len())
@@ -179,10 +176,7 @@ func (c *Compiler) getTypeMethodSet(typ types.Type) (llvm.Value, error) {
 			// compiler error, so panic
 			panic("cannot find function: " + f.LinkName())
 		}
-		fn, err := c.getInterfaceInvokeWrapper(f)
+		fn := c.getInterfaceInvokeWrapper(f)
-		if err != nil {
-			return llvm.Value{}, err
-		}
 		methodInfo := llvm.ConstNamedStruct(interfaceMethodInfoType, []llvm.Value{
 			signatureGlobal,
 			llvm.ConstPtrToInt(fn, c.uintptrType),
@@ -195,7 +189,7 @@ func (c *Compiler) getTypeMethodSet(typ types.Type) (llvm.Value, error) {
 	global.SetInitializer(value)
 	global.SetGlobalConstant(true)
 	global.SetLinkage(llvm.PrivateLinkage)
-	return llvm.ConstGEP(global, []llvm.Value{zero, zero}), nil
+	return llvm.ConstGEP(global, []llvm.Value{zero, zero})
 }
 
 // getInterfaceMethodSet returns a global variable with the method set of the
@@ -365,12 +359,12 @@ type interfaceInvokeWrapper struct {
 // the underlying value, dereferences it, and calls the real method. This
 // wrapper is only needed when the interface value actually doesn't fit in a
 // pointer and a pointer to the value must be created.
-func (c *Compiler) getInterfaceInvokeWrapper(f *ir.Function) (llvm.Value, error) {
+func (c *Compiler) getInterfaceInvokeWrapper(f *ir.Function) llvm.Value {
 	wrapperName := f.LinkName() + "$invoke"
 	wrapper := c.mod.NamedFunction(wrapperName)
 	if !wrapper.IsNil() {
 		// Wrapper already created. Return it directly.
-		return wrapper, nil
+		return wrapper
 	}
 
 	// Get the expanded receiver type.
@@ -383,7 +377,7 @@ func (c *Compiler) getInterfaceInvokeWrapper(f *ir.Function) (llvm.Value, error)
 		// Casting a function signature to a different signature and calling it
 		// with a receiver pointer bitcasted to *i8 (as done in calls on an
 		// interface) is hopefully a safe (defined) operation.
-		return f.LLVMFn, nil
+		return f.LLVMFn
 	}
 
 	// create wrapper function
@@ -396,7 +390,7 @@ func (c *Compiler) getInterfaceInvokeWrapper(f *ir.Function) (llvm.Value, error)
 		wrapper:      wrapper,
 		receiverType: receiverType,
 	})
-	return wrapper, nil
+	return wrapper
 }
 
 // createInterfaceInvokeWrapper finishes the work of getInterfaceInvokeWrapper,

compiler/llvm.go

@@ -22,6 +22,36 @@ func getUses(value llvm.Value) []llvm.Value {
 	return uses
 }
 
+// createEntryBlockAlloca creates a new alloca in the entry block, even though
+// the IR builder is located elsewhere. It assumes that the insert point is
+// after the last instruction in the current block. Also, it adds lifetime
+// information to the IR signalling that the alloca won't be used before this
+// point.
+//
+// This is useful for creating temporary allocas for intrinsics. Don't forget to
+// end the lifetime after you're done with it.
+func (c *Compiler) createEntryBlockAlloca(t llvm.Type, name string) (alloca, bitcast, size llvm.Value) {
+	currentBlock := c.builder.GetInsertBlock()
+	c.builder.SetInsertPointBefore(currentBlock.Parent().EntryBasicBlock().FirstInstruction())
+	alloca = c.builder.CreateAlloca(t, name)
+	c.builder.SetInsertPointAtEnd(currentBlock)
+	bitcast = c.builder.CreateBitCast(alloca, c.i8ptrType, name+".bitcast")
+	size = llvm.ConstInt(c.ctx.Int64Type(), c.targetData.TypeAllocSize(t), false)
+	c.builder.CreateCall(c.getLifetimeStartFunc(), []llvm.Value{size, bitcast}, "")
+	return
+}
+
+// getLifetimeStartFunc returns the llvm.lifetime.start intrinsic and creates it
+// first if it doesn't exist yet.
+func (c *Compiler) getLifetimeStartFunc() llvm.Value {
+	fn := c.mod.NamedFunction("llvm.lifetime.start.p0i8")
+	if fn.IsNil() {
+		fnType := llvm.FunctionType(c.ctx.VoidType(), []llvm.Type{c.ctx.Int64Type(), c.i8ptrType}, false)
+		fn = llvm.AddFunction(c.mod, "llvm.lifetime.start.p0i8", fnType)
+	}
+	return fn
+}
+
 // getLifetimeEndFunc returns the llvm.lifetime.end intrinsic and creates it
 // first if it doesn't exist yet.
 func (c *Compiler) getLifetimeEndFunc() llvm.Value {

compiler/map.go

@@ -11,8 +11,13 @@ import (
 
 func (c *Compiler) emitMapLookup(keyType, valueType types.Type, m, key llvm.Value, commaOk bool, pos token.Pos) (llvm.Value, error) {
 	llvmValueType := c.getLLVMType(valueType)
-	mapValueAlloca := c.builder.CreateAlloca(llvmValueType, "hashmap.value")
+
-	mapValuePtr := c.builder.CreateBitCast(mapValueAlloca, c.i8ptrType, "hashmap.valueptr")
+	// Allocate the memory for the resulting type. Do not zero this memory: it
+	// will be zeroed by the hashmap get implementation if the key is not
+	// present in the map.
+	mapValueAlloca, mapValuePtr, mapValueSize := c.createEntryBlockAlloca(llvmValueType, "hashmap.value")
+
+	// Do the lookup. How it is done depends on the key type.
 	var commaOkValue llvm.Value
 	if t, ok := keyType.(*types.Basic); ok && t.Info()&types.IsString != 0 {
 		// key is a string
@@ -20,15 +25,24 @@ func (c *Compiler) emitMapLookup(keyType, valueType types.Type, m, key llvm.Valu
 		commaOkValue = c.createRuntimeCall("hashmapStringGet", params, "")
 	} else if hashmapIsBinaryKey(keyType) {
 		// key can be compared with runtime.memequal
-		keyAlloca := c.builder.CreateAlloca(key.Type(), "hashmap.key")
+		// Store the key in an alloca, in the entry block to avoid dynamic stack
-		c.builder.CreateStore(key, keyAlloca)
+		// growth.
-		keyPtr := c.builder.CreateBitCast(keyAlloca, c.i8ptrType, "hashmap.keyptr")
+		mapKeyAlloca, mapKeyPtr, mapKeySize := c.createEntryBlockAlloca(key.Type(), "hashmap.key")
-		params := []llvm.Value{m, keyPtr, mapValuePtr}
+		c.builder.CreateStore(key, mapKeyAlloca)
+		// Fetch the value from the hashmap.
+		params := []llvm.Value{m, mapKeyPtr, mapValuePtr}
 		commaOkValue = c.createRuntimeCall("hashmapBinaryGet", params, "")
+		c.builder.CreateCall(c.getLifetimeEndFunc(), []llvm.Value{mapKeySize, mapKeyPtr}, "")
 	} else {
+		// Not trivially comparable using memcmp.
 		return llvm.Value{}, c.makeError(pos, "only strings, bools, ints or structs of bools/ints are supported as map keys, but got: "+keyType.String())
 	}
+
+	// Load the resulting value from the hashmap. The value is set to the zero
+	// value if the key doesn't exist in the hashmap.
 	mapValue := c.builder.CreateLoad(mapValueAlloca, "")
+	c.builder.CreateCall(c.getLifetimeEndFunc(), []llvm.Value{mapValueSize, mapValuePtr}, "")
+
 	if commaOk {
 		tuple := llvm.Undef(c.ctx.StructType([]llvm.Type{llvmValueType, c.ctx.Int1Type()}, false))
 		tuple = c.builder.CreateInsertValue(tuple, mapValue, 0, "")
@@ -40,9 +54,8 @@ func (c *Compiler) emitMapLookup(keyType, valueType types.Type, m, key llvm.Valu
 }
 
 func (c *Compiler) emitMapUpdate(keyType types.Type, m, key, value llvm.Value, pos token.Pos) {
-	valueAlloca := c.builder.CreateAlloca(value.Type(), "hashmap.value")
+	valueAlloca, valuePtr, valueSize := c.createEntryBlockAlloca(value.Type(), "hashmap.value")
 	c.builder.CreateStore(value, valueAlloca)
-	valuePtr := c.builder.CreateBitCast(valueAlloca, c.i8ptrType, "hashmap.valueptr")
 	keyType = keyType.Underlying()
 	if t, ok := keyType.(*types.Basic); ok && t.Info()&types.IsString != 0 {
 		// key is a string
@@ -50,14 +63,15 @@ func (c *Compiler) emitMapUpdate(keyType types.Type, m, key, value llvm.Value, p
 		c.createRuntimeCall("hashmapStringSet", params, "")
 	} else if hashmapIsBinaryKey(keyType) {
 		// key can be compared with runtime.memequal
-		keyAlloca := c.builder.CreateAlloca(key.Type(), "hashmap.key")
+		keyAlloca, keyPtr, keySize := c.createEntryBlockAlloca(key.Type(), "hashmap.key")
 		c.builder.CreateStore(key, keyAlloca)
-		keyPtr := c.builder.CreateBitCast(keyAlloca, c.i8ptrType, "hashmap.keyptr")
 		params := []llvm.Value{m, keyPtr, valuePtr}
 		c.createRuntimeCall("hashmapBinarySet", params, "")
+		c.builder.CreateCall(c.getLifetimeEndFunc(), []llvm.Value{keySize, keyPtr}, "")
 	} else {
 		c.addError(pos, "only strings, bools, ints or structs of bools/ints are supported as map keys, but got: "+keyType.String())
 	}
+	c.builder.CreateCall(c.getLifetimeEndFunc(), []llvm.Value{valueSize, valuePtr}, "")
 }
 
 func (c *Compiler) emitMapDelete(keyType types.Type, m, key llvm.Value, pos token.Pos) error {
@@ -68,11 +82,11 @@ func (c *Compiler) emitMapDelete(keyType types.Type, m, key llvm.Value, pos toke
 		c.createRuntimeCall("hashmapStringDelete", params, "")
 		return nil
 	} else if hashmapIsBinaryKey(keyType) {
-		keyAlloca := c.builder.CreateAlloca(key.Type(), "hashmap.key")
+		keyAlloca, keyPtr, keySize := c.createEntryBlockAlloca(key.Type(), "hashmap.key")
 		c.builder.CreateStore(key, keyAlloca)
-		keyPtr := c.builder.CreateBitCast(keyAlloca, c.i8ptrType, "hashmap.keyptr")
 		params := []llvm.Value{m, keyPtr}
 		c.createRuntimeCall("hashmapBinaryDelete", params, "")
+		c.builder.CreateCall(c.getLifetimeEndFunc(), []llvm.Value{keySize, keyPtr}, "")
 		return nil
 	} else {
 		return c.makeError(pos, "only strings, bools, ints or structs of bools/ints are supported as map keys, but got: "+keyType.String())

compiler/optimizer.go

@@ -37,6 +37,7 @@ func (c *Compiler) Optimize(optLevel, sizeLevel int, inlinerThreshold uint) erro
 		goPasses := llvm.NewPassManager()
 		defer goPasses.Dispose()
 		goPasses.AddGlobalOptimizerPass()
+		goPasses.AddGlobalDCEPass()
 		goPasses.AddConstantPropagationPass()
 		goPasses.AddAggressiveDCEPass()
 		goPasses.AddFunctionAttrsPass()
@@ -114,6 +115,13 @@ func (c *Compiler) Optimize(optLevel, sizeLevel int, inlinerThreshold uint) erro
 	builder.Populate(modPasses)
 	modPasses.Run(c.mod)
 
+	if c.gcIsPrecise() {
+		c.addGlobalsBitmap()
+		if err := c.Verify(); err != nil {
+			return errors.New("GC pass caused a verification failure")
+		}
+	}
+
 	return nil
 }
 

compiler/volatile.go

@@ -0,0 +1,26 @@
+package compiler
+
+// This file implements volatile loads/stores in runtime/volatile.LoadT and
+// runtime/volatile.StoreT as compiler builtins.
+
+import (
+	"golang.org/x/tools/go/ssa"
+	"tinygo.org/x/go-llvm"
+)
+
+func (c *Compiler) emitVolatileLoad(frame *Frame, instr *ssa.CallCommon) (llvm.Value, error) {
+	addr := c.getValue(frame, instr.Args[0])
+	c.emitNilCheck(frame, addr, "deref")
+	val := c.builder.CreateLoad(addr, "")
+	val.SetVolatile(true)
+	return val, nil
+}
+
+func (c *Compiler) emitVolatileStore(frame *Frame, instr *ssa.CallCommon) (llvm.Value, error) {
+	addr := c.getValue(frame, instr.Args[0])
+	val := c.getValue(frame, instr.Args[1])
+	c.emitNilCheck(frame, addr, "deref")
+	store := c.builder.CreateStore(val, addr)
+	store.SetVolatile(true)
+	return llvm.Value{}, nil
+}

compiler/wordpack.go

@@ -25,7 +25,7 @@ func (c *Compiler) emitPointerPack(values []llvm.Value) llvm.Value {
 		return llvm.ConstPointerNull(c.i8ptrType)
 	} else if len(values) == 1 && values[0].Type().TypeKind() == llvm.PointerTypeKind {
 		return c.builder.CreateBitCast(values[0], c.i8ptrType, "pack.ptr")
-	} else if size <= c.targetData.TypeAllocSize(c.i8ptrType) {
+	} else if size <= c.targetData.TypeAllocSize(c.i8ptrType) && !c.gcIsPrecise() {
 		// Packed data fits in a pointer, so store it directly inside the
 		// pointer.
 		if len(values) == 1 && values[0].Type().TypeKind() == llvm.IntegerTypeKind {
@@ -57,7 +57,7 @@ func (c *Compiler) emitPointerPack(values []llvm.Value) llvm.Value {
 		return c.builder.CreateLoad(packedAlloc, "")
 	} else {
 		// Get the original heap allocation pointer, which already is an *i8.
-		return packedHeapAlloc
+		return c.builder.CreateBitCast(packedAlloc, c.i8ptrType, "")
 	}
 }
 
@@ -73,7 +73,7 @@ func (c *Compiler) emitPointerUnpack(ptr llvm.Value, valueTypes []llvm.Type) []l
 	} else if len(valueTypes) == 1 && valueTypes[0].TypeKind() == llvm.PointerTypeKind {
 		// A single pointer is always stored directly.
 		return []llvm.Value{c.builder.CreateBitCast(ptr, valueTypes[0], "unpack.ptr")}
-	} else if size <= c.targetData.TypeAllocSize(c.i8ptrType) {
+	} else if size <= c.targetData.TypeAllocSize(c.i8ptrType) && !c.gcIsPrecise() {
 		// Packed data stored directly in pointer.
 		if len(valueTypes) == 1 && valueTypes[0].TypeKind() == llvm.IntegerTypeKind {
 			// Keep this cast in SSA form.

interp/frame.go

@@ -347,6 +347,8 @@ func (fr *frame) evalBasicBlock(bb, incoming llvm.BasicBlock, indent string) (re
 				fr.locals[inst] = &LocalValue{fr.Eval, llvm.ConstInt(fr.Mod.Context().Int1Type(), implements, false)}
 			case callee.Name() == "runtime.nanotime":
 				fr.locals[inst] = &LocalValue{fr.Eval, llvm.ConstInt(fr.Mod.Context().Int64Type(), 0, false)}
+			case callee.Name() == "llvm.dbg.value":
+				// do nothing
 			case strings.HasPrefix(callee.Name(), "runtime.print") || callee.Name() == "runtime._panic":
 				// This are all print instructions, which necessarily have side
 				// effects but no results.

interp/interp.go

@@ -18,7 +18,6 @@ type Eval struct {
 	TargetData      llvm.TargetData
 	Debug           bool
 	builder         llvm.Builder
-	dibuilder       *llvm.DIBuilder
 	dirtyGlobals    map[llvm.Value]struct{}
 	sideEffectFuncs map[llvm.Value]*sideEffectResult // cache of side effect scan results
 }
@@ -38,7 +37,6 @@ func Run(mod llvm.Module, targetData llvm.TargetData, debug bool) error {
 		dirtyGlobals: map[llvm.Value]struct{}{},
 	}
 	e.builder = mod.Context().NewBuilder()
-	e.dibuilder = llvm.NewDIBuilder(mod)
 
 	initAll := mod.NamedFunction(name)
 	bb := initAll.EntryBasicBlock()
@@ -49,7 +47,6 @@ func Run(mod llvm.Module, targetData llvm.TargetData, debug bool) error {
 	e.builder.SetInsertPointBefore(bb.FirstInstruction())
 	dummy := e.builder.CreateAlloca(e.Mod.Context().Int8Type(), "dummy")
 	e.builder.SetInsertPointBefore(dummy)
-	e.builder.SetInstDebugLocation(bb.FirstInstruction())
 	var initCalls []llvm.Value
 	for inst := bb.FirstInstruction(); !inst.IsNil(); inst = llvm.NextInstruction(inst) {
 		if inst == dummy {

interp/scan.go

@@ -35,6 +35,8 @@ func (e *Eval) hasSideEffects(fn llvm.Value) *sideEffectResult {
 		return &sideEffectResult{severity: sideEffectLimited}
 	case "runtime.interfaceImplements":
 		return &sideEffectResult{severity: sideEffectNone}
+	case "llvm.dbg.value":
+		return &sideEffectResult{severity: sideEffectNone}
 	}
 	if e.sideEffectFuncs == nil {
 		e.sideEffectFuncs = make(map[llvm.Value]*sideEffectResult)

ir/ir.go

@@ -38,6 +38,7 @@ type Function struct {
 	nobounds  bool       // go:nobounds
 	flag      bool       // used by dead code elimination
 	interrupt bool       // go:interrupt
+	inline    InlineType // go:inline
 }
 
 // Global variable, possibly constant.
@@ -69,7 +70,22 @@ type Interface struct {
 	Type *types.Interface
 }
 
-// Create and intialize a new *Program from a *ssa.Program.
+type InlineType int
+
+// How much to inline.
+const (
+	// Default behavior. The compiler decides for itself whether any given
+	// function will be inlined. Whether any function is inlined depends on the
+	// optimization level.
+	InlineDefault InlineType = iota
+
+	// Inline hint, just like the C inline keyword (signalled using
+	// //go:inline). The compiler will be more likely to inline this function,
+	// but it is not a guarantee.
+	InlineHint
+)
+
+// Create and initialize a new *Program from a *ssa.Program.
 func NewProgram(lprogram *loader.Program, mainPath string) *Program {
 	comments := map[string]*ast.CommentGroup{}
 	for _, pkgInfo := range lprogram.Sorted() {
@@ -279,6 +295,8 @@ func (f *Function) parsePragmas() {
 				}
 				f.linkName = parts[1]
 				f.exported = true
+			case "//go:inline":
+				f.inline = InlineHint
 			case "//go:interrupt":
 				if len(parts) != 2 {
 					continue
@@ -332,6 +350,11 @@ func (f *Function) IsInterrupt() bool {
 	return f.interrupt
 }
 
+// Return the inline directive of this function.
+func (f *Function) Inline() InlineType {
+	return f.inline
+}
+
 // Return the link name for this function.
 func (f *Function) LinkName() string {
 	if f.linkName != "" {

loader/loader.go

@@ -10,6 +10,8 @@ import (
 	"os"
 	"path/filepath"
 	"sort"
+
+	"github.com/tinygo-org/tinygo/cgo"
 )
 
 // Program holds all packages and some metadata about the program as a whole.
@@ -33,7 +35,6 @@ type Package struct {
 	Imports   map[string]*Package
 	Importing bool
 	Files     []*ast.File
-	tokenFiles map[string]*token.File
 	Pkg       *types.Package
 	types.Info
 }
@@ -107,7 +108,6 @@ func (p *Program) newPackage(pkg *build.Package) *Package {
 			Scopes:     make(map[ast.Node]*types.Scope),
 			Selections: make(map[*ast.SelectorExpr]*types.Selection),
 		},
-		tokenFiles: map[string]*token.File{},
 	}
 }
 
@@ -295,16 +295,6 @@ func (p *Package) parseFiles() ([]*ast.File, error) {
 		}
 		files = append(files, f)
 	}
-	clangIncludes := ""
-	if len(p.CgoFiles) != 0 {
-		if _, err := os.Stat(filepath.Join(p.TINYGOROOT, "llvm", "tools", "clang", "lib", "Headers")); !os.IsNotExist(err) {
-			// Running from the source directory.
-			clangIncludes = filepath.Join(p.TINYGOROOT, "llvm", "tools", "clang", "lib", "Headers")
-		} else {
-			// Running from the installation directory.
-			clangIncludes = filepath.Join(p.TINYGOROOT, "lib", "clang", "include")
-		}
-	}
 	for _, file := range p.CgoFiles {
 		path := filepath.Join(p.Package.Dir, file)
 		f, err := p.parseFile(path, parser.ParseComments)
@@ -312,12 +302,22 @@ func (p *Package) parseFiles() ([]*ast.File, error) {
 			fileErrs = append(fileErrs, err)
 			continue
 		}
-		errs := p.processCgo(path, f, append(p.CFlags, "-I"+p.Package.Dir, "-I"+clangIncludes))
+		files = append(files, f)
+	}
+	if len(p.CgoFiles) != 0 {
+		clangIncludes := ""
+		if _, err := os.Stat(filepath.Join(p.TINYGOROOT, "llvm", "tools", "clang", "lib", "Headers")); !os.IsNotExist(err) {
+			// Running from the source directory.
+			clangIncludes = filepath.Join(p.TINYGOROOT, "llvm", "tools", "clang", "lib", "Headers")
+		} else {
+			// Running from the installation directory.
+			clangIncludes = filepath.Join(p.TINYGOROOT, "lib", "clang", "include")
+		}
+		generated, errs := cgo.Process(files, p.Program.Dir, p.fset, append(p.CFlags, "-I"+p.Package.Dir, "-I"+clangIncludes))
 		if errs != nil {
 			fileErrs = append(fileErrs, errs...)
-			continue
 		}
-		files = append(files, f)
+		files = append(files, generated)
 	}
 	if len(fileErrs) != 0 {
 		return nil, Errors{p, fileErrs}
@@ -346,7 +346,7 @@ func (p *Package) importRecursively() error {
 	p.Importing = true
 	for _, to := range p.Package.Imports {
 		if to == "C" {
-			// Do Cgo processing in a later stage.
+			// Do CGo processing in a later stage.
 			continue
 		}
 		if _, ok := p.Imports[to]; ok {

main.go

@@ -537,7 +537,7 @@ func handleCompilerError(err error) {
 func main() {
 	outpath := flag.String("o", "", "output filename")
 	opt := flag.String("opt", "z", "optimization level: 0, 1, 2, s, z")
-	gc := flag.String("gc", "", "garbage collector to use (none, dumb, marksweep)")
+	gc := flag.String("gc", "", "garbage collector to use (none, leaking, conservative)")
 	panicStrategy := flag.String("panic", "print", "panic strategy (abort, trap)")
 	printIR := flag.Bool("printir", false, "print LLVM IR")
 	dumpSSA := flag.Bool("dumpssa", false, "dump internal Go SSA")

main_test.go

@@ -58,6 +58,9 @@ func TestCompiler(t *testing.T) {
 
 	t.Log("running tests for emulated cortex-m3...")
 	for _, path := range matches {
+		if path == "testdata/reflect.go" {
+			continue
+		}
 		t.Run(path, func(t *testing.T) {
 			runTest(path, tmpdir, "qemu", t)
 		})

src/examples/i2s/i2s.go

@@ -0,0 +1,25 @@
+// Example using the i2s hardware interface on the Adafruit Circuit Playground Express
+// to read data from the onboard MEMS microphone.
+//
+package main
+
+import (
+	"machine"
+)
+
+func main() {
+	machine.I2S0.Configure(machine.I2SConfig{
+		Mode:        machine.I2SModePDM,
+		ClockSource: machine.I2SClockSourceExternal,
+		Stereo:      true,
+	})
+
+	data := make([]uint32, 64)
+
+	for {
+		// get the next group of samples
+		machine.I2S0.Read(data)
+
+		println("data", data[0], data[1], data[2], data[4], "...")
+	}
+}

src/machine/board_circuitplay_express.go

@@ -99,3 +99,15 @@ const (
 var (
 	SPI0 = SPI{Bus: sam.SERCOM3_SPI}
 )
+
+// I2S pins
+const (
+	I2S_SCK_PIN = PA10
+	I2S_SD_PIN  = PA08
+	I2S_WS_PIN  = 0xff // no WS, instead uses SCK to sync
+)
+
+// I2S on the Circuit Playground Express.
+var (
+	I2S0 = I2S{Bus: sam.I2S}
+)

src/machine/board_itsybitsy-m0.go

@@ -73,3 +73,15 @@ const (
 var (
 	SPI0 = SPI{Bus: sam.SERCOM4_SPI}
 )
+
+// I2S pins
+const (
+	I2S_SCK_PIN = PA10
+	I2S_SD_PIN  = PA08
+	I2S_WS_PIN  = 0xff // TODO: figure out what this is on ItsyBitsy M0.
+)
+
+// I2S on the ItsyBitsy M0.
+var (
+	I2S0 = I2S{Bus: sam.I2S}
+)

src/machine/i2s.go

@@ -0,0 +1,54 @@
+// +build sam
+
+// This is the definition for I2S bus functions.
+// Actual implementations if available for any given hardware
+// are to be found in its the board definition.
+//
+// For more info about I2S, see: https://en.wikipedia.org/wiki/I%C2%B2S
+//
+
+package machine
+
+type I2SMode uint8
+type I2SStandard uint8
+type I2SClockSource uint8
+type I2SDataFormat uint8
+
+const (
+	I2SModeMaster I2SMode = iota
+	I2SModeSlave
+	I2SModePDM
+)
+
+const (
+	I2StandardPhilips I2SStandard = iota
+	I2SStandardMSB
+	I2SStandardLSB
+)
+
+const (
+	I2SClockSourceInternal I2SClockSource = iota
+	I2SClockSourceExternal
+)
+
+const (
+	I2SDataFormatDefault I2SDataFormat = 0
+	I2SDataFormat8bit                  = 8
+	I2SDataFormat16bit                 = 16
+	I2SDataFormat24bit                 = 24
+	I2SDataFormat32bit                 = 32
+)
+
+// All fields are optional and may not be required or used on a particular platform.
+type I2SConfig struct {
+	SCK               uint8
+	WS                uint8
+	SD                uint8
+	Mode              I2SMode
+	Standard          I2SStandard
+	ClockSource       I2SClockSource
+	DataFormat        I2SDataFormat
+	AudioFrequency    uint32
+	MasterClockOutput bool
+	Stereo            bool
+}

src/machine/machine_atmega.go

@@ -10,15 +10,15 @@ import (
 func (p GPIO) Configure(config GPIOConfig) {
 	if config.Mode == GPIO_OUTPUT { // set output bit
 		if p.Pin < 8 {
-			*avr.DDRD |= 1 << p.Pin
+			avr.DDRD.SetBits(1 << p.Pin)
 		} else {
-			*avr.DDRB |= 1 << (p.Pin - 8)
+			avr.DDRB.SetBits(1 << (p.Pin - 8))
 		}
 	} else { // configure input: clear output bit
 		if p.Pin < 8 {
-			*avr.DDRD &^= 1 << p.Pin
+			avr.DDRD.ClearBits(1 << p.Pin)
 		} else {
-			*avr.DDRB &^= 1 << (p.Pin - 8)
+			avr.DDRB.ClearBits(1 << (p.Pin - 8))
 		}
 	}
 }
@@ -26,15 +26,15 @@ func (p GPIO) Configure(config GPIOConfig) {
 // Get returns the current value of a GPIO pin.
 func (p GPIO) Get() bool {
 	if p.Pin < 8 {
-		val := *avr.PIND & (1 << p.Pin)
+		val := avr.PIND.Get() & (1 << p.Pin)
 		return (val > 0)
 	} else {
-		val := *avr.PINB & (1 << (p.Pin - 8))
+		val := avr.PINB.Get() & (1 << (p.Pin - 8))
 		return (val > 0)
 	}
 }
 
-func (p GPIO) getPortMask() (*avr.RegValue, uint8) {
+func (p GPIO) getPortMask() (*avr.Register8, uint8) {
 	if p.Pin < 8 {
 		return avr.PORTD, 1 << p.Pin
 	} else {
@@ -45,64 +45,64 @@ func (p GPIO) getPortMask() (*avr.RegValue, uint8) {
 // InitPWM initializes the registers needed for PWM.
 func InitPWM() {
 	// use waveform generation
-	*avr.TCCR0A |= avr.TCCR0A_WGM00
+	avr.TCCR0A.SetBits(avr.TCCR0A_WGM00)
 
 	// set timer 0 prescale factor to 64
-	*avr.TCCR0B |= avr.TCCR0B_CS01 | avr.TCCR0B_CS00
+	avr.TCCR0B.SetBits(avr.TCCR0B_CS01 | avr.TCCR0B_CS00)
 
 	// set timer 1 prescale factor to 64
-	*avr.TCCR1B |= avr.TCCR1B_CS11
+	avr.TCCR1B.SetBits(avr.TCCR1B_CS11)
 
 	// put timer 1 in 8-bit phase correct pwm mode
-	*avr.TCCR1A |= avr.TCCR1A_WGM10
+	avr.TCCR1A.SetBits(avr.TCCR1A_WGM10)
 
 	// set timer 2 prescale factor to 64
-	*avr.TCCR2B |= avr.TCCR2B_CS22
+	avr.TCCR2B.SetBits(avr.TCCR2B_CS22)
 
 	// configure timer 2 for phase correct pwm (8-bit)
-	*avr.TCCR2A |= avr.TCCR2A_WGM20
+	avr.TCCR2A.SetBits(avr.TCCR2A_WGM20)
 }
 
 // Configure configures a PWM pin for output.
 func (pwm PWM) Configure() {
 	if pwm.Pin < 8 {
-		*avr.DDRD |= 1 << pwm.Pin
+		avr.DDRD.SetBits(1 << pwm.Pin)
 	} else {
-		*avr.DDRB |= 1 << (pwm.Pin - 8)
+		avr.DDRB.SetBits(1 << (pwm.Pin - 8))
 	}
 }
 
 // Set turns on the duty cycle for a PWM pin using the provided value. On the AVR this is normally a
 // 8-bit value ranging from 0 to 255.
 func (pwm PWM) Set(value uint16) {
-	value8 := value >> 8
+	value8 := uint8(value >> 8)
 	switch pwm.Pin {
 	case 3:
 		// connect pwm to pin on timer 2, channel B
-		*avr.TCCR2A |= avr.TCCR2A_COM2B1
+		avr.TCCR2A.SetBits(avr.TCCR2A_COM2B1)
-		*avr.OCR2B = avr.RegValue(value8) // set pwm duty
+		avr.OCR2B.Set(value8) // set pwm duty
 	case 5:
 		// connect pwm to pin on timer 0, channel B
-		*avr.TCCR0A |= avr.TCCR0A_COM0B1
+		avr.TCCR0A.SetBits(avr.TCCR0A_COM0B1)
-		*avr.OCR0B = avr.RegValue(value8) // set pwm duty
+		avr.OCR0B.Set(value8) // set pwm duty
 	case 6:
 		// connect pwm to pin on timer 0, channel A
-		*avr.TCCR0A |= avr.TCCR0A_COM0A1
+		avr.TCCR0A.SetBits(avr.TCCR0A_COM0A1)
-		*avr.OCR0A = avr.RegValue(value8) // set pwm duty
+		avr.OCR0A.Set(value8) // set pwm duty
 	case 9:
 		// connect pwm to pin on timer 1, channel A
-		*avr.TCCR1A |= avr.TCCR1A_COM1A1
+		avr.TCCR1A.SetBits(avr.TCCR1A_COM1A1)
 		// this is a 16-bit value, but we only currently allow the low order bits to be set
-		*avr.OCR1AL = avr.RegValue(value8) // set pwm duty
+		avr.OCR1AL.Set(value8) // set pwm duty
 	case 10:
 		// connect pwm to pin on timer 1, channel B
-		*avr.TCCR1A |= avr.TCCR1A_COM1B1
+		avr.TCCR1A.SetBits(avr.TCCR1A_COM1B1)
 		// this is a 16-bit value, but we only currently allow the low order bits to be set
-		*avr.OCR1BL = avr.RegValue(value8) // set pwm duty
+		avr.OCR1BL.Set(value8) // set pwm duty
 	case 11:
 		// connect pwm to pin on timer 2, channel A
-		*avr.TCCR2A |= avr.TCCR2A_COM2A1
+		avr.TCCR2A.SetBits(avr.TCCR2A_COM2A1)
-		*avr.OCR2A = avr.RegValue(value8) // set pwm duty
+		avr.OCR2A.Set(value8) // set pwm duty
 	default:
 		panic("Invalid PWM pin")
 	}
@@ -121,19 +121,19 @@ func (i2c I2C) Configure(config I2CConfig) {
 	}
 
 	// Activate internal pullups for twi.
-	*avr.PORTC |= (avr.DIDR0_ADC4D | avr.DIDR0_ADC5D)
+	avr.PORTC.SetBits((avr.DIDR0_ADC4D | avr.DIDR0_ADC5D))
 
 	// Initialize twi prescaler and bit rate.
-	*avr.TWSR |= (avr.TWSR_TWPS0 | avr.TWSR_TWPS1)
+	avr.TWSR.SetBits((avr.TWSR_TWPS0 | avr.TWSR_TWPS1))
 
 	// twi bit rate formula from atmega128 manual pg. 204:
 	// SCL Frequency = CPU Clock Frequency / (16 + (2 * TWBR))
 	// NOTE: TWBR should be 10 or higher for master mode.
 	// It is 72 for a 16mhz board with 100kHz TWI
-	*avr.TWBR = avr.RegValue(((CPU_FREQUENCY / config.Frequency) - 16) / 2)
+	avr.TWBR.Set(uint8(((CPU_FREQUENCY / config.Frequency) - 16) / 2))
 
 	// Enable twi module.
-	*avr.TWCR = avr.TWCR_TWEN
+	avr.TWCR.Set(avr.TWCR_TWEN)
 }
 
 // Tx does a single I2C transaction at the specified address.
@@ -162,10 +162,10 @@ func (i2c I2C) Tx(addr uint16, w, r []byte) error {
 // start starts an I2C communication session.
 func (i2c I2C) start(address uint8, write bool) {
 	// Clear TWI interrupt flag, put start condition on SDA, and enable TWI.
-	*avr.TWCR = (avr.TWCR_TWINT | avr.TWCR_TWSTA | avr.TWCR_TWEN)
+	avr.TWCR.Set((avr.TWCR_TWINT | avr.TWCR_TWSTA | avr.TWCR_TWEN))
 
 	// Wait till start condition is transmitted.
-	for (*avr.TWCR & avr.TWCR_TWINT) == 0 {
+	for (avr.TWCR.Get() & avr.TWCR_TWINT) == 0 {
 	}
 
 	// Write 7-bit shifted peripheral address.
@@ -179,36 +179,36 @@ func (i2c I2C) start(address uint8, write bool) {
 // stop ends an I2C communication session.
 func (i2c I2C) stop() {
 	// Send stop condition.
-	*avr.TWCR = (avr.TWCR_TWEN | avr.TWCR_TWINT | avr.TWCR_TWSTO)
+	avr.TWCR.Set(avr.TWCR_TWEN | avr.TWCR_TWINT | avr.TWCR_TWSTO)
 
 	// Wait for stop condition to be executed on bus.
-	for (*avr.TWCR & avr.TWCR_TWSTO) == 0 {
+	for (avr.TWCR.Get() & avr.TWCR_TWSTO) == 0 {
 	}
 }
 
 // writeByte writes a single byte to the I2C bus.
 func (i2c I2C) writeByte(data byte) {
 	// Write data to register.
-	*avr.TWDR = avr.RegValue(data)
+	avr.TWDR.Set(data)
 
 	// Clear TWI interrupt flag and enable TWI.
-	*avr.TWCR = (avr.TWCR_TWEN | avr.TWCR_TWINT)
+	avr.TWCR.Set(avr.TWCR_TWEN | avr.TWCR_TWINT)
 
 	// Wait till data is transmitted.
-	for (*avr.TWCR & avr.TWCR_TWINT) == 0 {
+	for (avr.TWCR.Get() & avr.TWCR_TWINT) == 0 {
 	}
 }
 
 // readByte reads a single byte from the I2C bus.
 func (i2c I2C) readByte() byte {
 	// Clear TWI interrupt flag and enable TWI.
-	*avr.TWCR = (avr.TWCR_TWEN | avr.TWCR_TWINT | avr.TWCR_TWEA)
+	avr.TWCR.Set(avr.TWCR_TWEN | avr.TWCR_TWINT | avr.TWCR_TWEA)
 
 	// Wait till read request is transmitted.
-	for (*avr.TWCR & avr.TWCR_TWINT) == 0 {
+	for (avr.TWCR.Get() & avr.TWCR_TWINT) == 0 {
 	}
 
-	return byte(*avr.TWDR)
+	return byte(avr.TWDR.Get())
 }
 
 // UART on the AVR.
@@ -226,32 +226,32 @@ func (uart UART) Configure(config UARTConfig) {
 	// https://www.microchip.com/webdoc/AVRLibcReferenceManual/FAQ_1faq_wrong_baud_rate.html
 	// ((F_CPU + UART_BAUD_RATE * 8L) / (UART_BAUD_RATE * 16L) - 1)
 	ps := ((CPU_FREQUENCY+config.BaudRate*8)/(config.BaudRate*16) - 1)
-	*avr.UBRR0H = avr.RegValue(ps >> 8)
+	avr.UBRR0H.Set(uint8(ps >> 8))
-	*avr.UBRR0L = avr.RegValue(ps & 0xff)
+	avr.UBRR0L.Set(uint8(ps & 0xff))
 
 	// enable RX, TX and RX interrupt
-	*avr.UCSR0B = avr.UCSR0B_RXEN0 | avr.UCSR0B_TXEN0 | avr.UCSR0B_RXCIE0
+	avr.UCSR0B.Set(avr.UCSR0B_RXEN0 | avr.UCSR0B_TXEN0 | avr.UCSR0B_RXCIE0)
 
 	// 8-bits data
-	*avr.UCSR0C = avr.UCSR0C_UCSZ01 | avr.UCSR0C_UCSZ00
+	avr.UCSR0C.Set(avr.UCSR0C_UCSZ01 | avr.UCSR0C_UCSZ00)
 }
 
 // WriteByte writes a byte of data to the UART.
 func (uart UART) WriteByte(c byte) error {
 	// Wait until UART buffer is not busy.
-	for (*avr.UCSR0A & avr.UCSR0A_UDRE0) == 0 {
+	for (avr.UCSR0A.Get() & avr.UCSR0A_UDRE0) == 0 {
 	}
-	*avr.UDR0 = avr.RegValue(c) // send char
+	avr.UDR0.Set(c) // send char
 	return nil
 }
 
 //go:interrupt USART_RX_vect
 func handleUSART_RX() {
 	// Read register to clear it.
-	data := *avr.UDR0
+	data := avr.UDR0.Get()
 
 	// Ensure no error.
-	if (*avr.UCSR0A & (avr.UCSR0A_FE0 | avr.UCSR0A_DOR0 | avr.UCSR0A_UPE0)) == 0 {
+	if (avr.UCSR0A.Get() & (avr.UCSR0A_FE0 | avr.UCSR0A_DOR0 | avr.UCSR0A_UPE0)) == 0 {
 		// Put data from UDR register into buffer.
 		UART0.Receive(byte(data))
 	}

src/machine/machine_atsamd21.go

@@ -808,6 +808,235 @@ func (i2c I2C) readByte() byte {
 	return byte(i2c.Bus.DATA)
 }
 
+// I2S on the SAMD21.
+
+// I2S
+type I2S struct {
+	Bus *sam.I2S_Type
+}
+
+// Configure is used to configure the I2S interface. You must call this
+// before you can use the I2S bus.
+func (i2s I2S) Configure(config I2SConfig) {
+	// handle defaults
+	if config.SCK == 0 {
+		config.SCK = I2S_SCK_PIN
+		config.WS = I2S_WS_PIN
+		config.SD = I2S_SD_PIN
+	}
+
+	if config.AudioFrequency == 0 {
+		config.AudioFrequency = 48000
+	}
+
+	if config.DataFormat == I2SDataFormatDefault {
+		if config.Stereo {
+			config.DataFormat = I2SDataFormat16bit
+		} else {
+			config.DataFormat = I2SDataFormat32bit
+		}
+	}
+
+	// Turn on clock for I2S
+	sam.PM.APBCMASK |= sam.PM_APBCMASK_I2S_
+
+	// setting clock rate for sample.
+	division_factor := CPU_FREQUENCY / (config.AudioFrequency * uint32(config.DataFormat))
+
+	// Switch Generic Clock Generator 3 to DFLL48M.
+	sam.GCLK.GENDIV = sam.RegValue((sam.GCLK_CLKCTRL_GEN_GCLK3 << sam.GCLK_GENDIV_ID_Pos) |
+		(division_factor << sam.GCLK_GENDIV_DIV_Pos))
+	waitForSync()
+
+	sam.GCLK.GENCTRL = sam.RegValue((sam.GCLK_CLKCTRL_GEN_GCLK3 << sam.GCLK_GENCTRL_ID_Pos) |
+		(sam.GCLK_GENCTRL_SRC_DFLL48M << sam.GCLK_GENCTRL_SRC_Pos) |
+		sam.GCLK_GENCTRL_IDC |
+		sam.GCLK_GENCTRL_GENEN)
+	waitForSync()
+
+	// Use Generic Clock Generator 3 as source for I2S.
+	sam.GCLK.CLKCTRL = sam.RegValue16((sam.GCLK_CLKCTRL_ID_I2S_0 << sam.GCLK_CLKCTRL_ID_Pos) |
+		(sam.GCLK_CLKCTRL_GEN_GCLK3 << sam.GCLK_CLKCTRL_GEN_Pos) |
+		sam.GCLK_CLKCTRL_CLKEN)
+	waitForSync()
+
+	// reset the device
+	i2s.Bus.CTRLA |= sam.I2S_CTRLA_SWRST
+	for (i2s.Bus.SYNCBUSY & sam.I2S_SYNCBUSY_SWRST) > 0 {
+	}
+
+	// disable device before continuing
+	for (i2s.Bus.SYNCBUSY & sam.I2S_SYNCBUSY_ENABLE) > 0 {
+	}
+	i2s.Bus.CTRLA &^= sam.I2S_CTRLA_ENABLE
+
+	// setup clock
+	if config.ClockSource == I2SClockSourceInternal {
+		// TODO: make sure correct for I2S output
+
+		// set serial clock select pin
+		i2s.Bus.CLKCTRL0 |= sam.I2S_CLKCTRL_SCKSEL
+
+		// set frame select pin
+		i2s.Bus.CLKCTRL0 |= sam.I2S_CLKCTRL_FSSEL
+	} else {
+		// Configure FS generation from SCK clock.
+		i2s.Bus.CLKCTRL0 &^= sam.I2S_CLKCTRL_FSSEL
+	}
+
+	if config.Standard == I2StandardPhilips {
+		// set 1-bit delay
+		i2s.Bus.CLKCTRL0 |= sam.I2S_CLKCTRL_BITDELAY
+	} else {
+		// set 0-bit delay
+		i2s.Bus.CLKCTRL0 &^= sam.I2S_CLKCTRL_BITDELAY
+	}
+
+	// set number of slots.
+	if config.Stereo {
+		i2s.Bus.CLKCTRL0 |= (1 << sam.I2S_CLKCTRL_NBSLOTS_Pos)
+	} else {
+		i2s.Bus.CLKCTRL0 &^= (1 << sam.I2S_CLKCTRL_NBSLOTS_Pos)
+	}
+
+	// set slot size
+	switch config.DataFormat {
+	case I2SDataFormat8bit:
+		i2s.Bus.CLKCTRL0 |= sam.I2S_CLKCTRL_SLOTSIZE_8
+
+	case I2SDataFormat16bit:
+		i2s.Bus.CLKCTRL0 |= sam.I2S_CLKCTRL_SLOTSIZE_16
+
+	case I2SDataFormat24bit:
+		i2s.Bus.CLKCTRL0 |= sam.I2S_CLKCTRL_SLOTSIZE_24
+
+	case I2SDataFormat32bit:
+		i2s.Bus.CLKCTRL0 |= sam.I2S_CLKCTRL_SLOTSIZE_32
+	}
+
+	// configure pin for clock
+	GPIO{config.SCK}.Configure(GPIOConfig{Mode: GPIO_COM})
+
+	// configure pin for WS, if needed
+	if config.WS != 0xff {
+		GPIO{config.WS}.Configure(GPIOConfig{Mode: GPIO_COM})
+	}
+
+	// now set serializer data size.
+	switch config.DataFormat {
+	case I2SDataFormat8bit:
+		i2s.Bus.SERCTRL1 |= sam.I2S_SERCTRL_DATASIZE_8
+
+	case I2SDataFormat16bit:
+		i2s.Bus.SERCTRL1 |= sam.I2S_SERCTRL_DATASIZE_16
+
+	case I2SDataFormat24bit:
+		i2s.Bus.SERCTRL1 |= sam.I2S_SERCTRL_DATASIZE_24
+
+	case I2SDataFormat32bit:
+	case I2SDataFormatDefault:
+		i2s.Bus.SERCTRL1 |= sam.I2S_SERCTRL_DATASIZE_32
+	}
+
+	// set serializer slot adjustment
+	if config.Standard == I2SStandardLSB {
+		// adjust right
+		i2s.Bus.SERCTRL1 &^= sam.I2S_SERCTRL_SLOTADJ
+	} else {
+		// adjust left
+		i2s.Bus.SERCTRL1 |= sam.I2S_SERCTRL_SLOTADJ
+
+		// reverse bit order?
+		i2s.Bus.SERCTRL1 |= sam.I2S_SERCTRL_BITREV
+	}
+
+	// set serializer mode.
+	if config.Mode == I2SModePDM {
+		i2s.Bus.SERCTRL1 |= sam.I2S_SERCTRL_SERMODE_PDM2
+	} else {
+		i2s.Bus.SERCTRL1 |= sam.I2S_SERCTRL_SERMODE_RX
+	}
+
+	// configure data pin
+	GPIO{config.SD}.Configure(GPIOConfig{Mode: GPIO_COM})
+
+	// re-enable
+	i2s.Bus.CTRLA |= sam.I2S_CTRLA_ENABLE
+	for (i2s.Bus.SYNCBUSY & sam.I2S_SYNCBUSY_ENABLE) > 0 {
+	}
+
+	// enable i2s clock
+	i2s.Bus.CTRLA |= sam.I2S_CTRLA_CKEN0
+	for (i2s.Bus.SYNCBUSY & sam.I2S_SYNCBUSY_CKEN0) > 0 {
+	}
+
+	// enable i2s serializer
+	i2s.Bus.CTRLA |= sam.I2S_CTRLA_SEREN1
+	for (i2s.Bus.SYNCBUSY & sam.I2S_SYNCBUSY_SEREN1) > 0 {
+	}
+}
+
+// Read data from the I2S bus into the provided slice.
+// The I2S bus must already have been configured correctly.
+func (i2s I2S) Read(p []uint32) (n int, err error) {
+	i := 0
+	for i = 0; i < len(p); i++ {
+		// Wait until ready
+		for (i2s.Bus.INTFLAG & sam.I2S_INTFLAG_RXRDY1) == 0 {
+		}
+
+		for (i2s.Bus.SYNCBUSY & sam.I2S_SYNCBUSY_DATA1) > 0 {
+		}
+
+		// read data
+		p[i] = uint32(i2s.Bus.DATA1)
+
+		// indicate read complete
+		i2s.Bus.INTFLAG = sam.I2S_INTFLAG_RXRDY1
+	}
+
+	return i, nil
+}
+
+// Write data to the I2S bus from the provided slice.
+// The I2S bus must already have been configured correctly.
+func (i2s I2S) Write(p []uint32) (n int, err error) {
+	i := 0
+	for i = 0; i < len(p); i++ {
+		// Wait until ready
+		for (i2s.Bus.INTFLAG & sam.I2S_INTFLAG_TXRDY1) == 0 {
+		}
+
+		for (i2s.Bus.SYNCBUSY & sam.I2S_SYNCBUSY_DATA1) > 0 {
+		}
+
+		// write data
+		i2s.Bus.DATA1 = sam.RegValue(p[i])
+
+		// indicate write complete
+		i2s.Bus.INTFLAG = sam.I2S_INTFLAG_TXRDY1
+	}
+
+	return i, nil
+}
+
+// Close the I2S bus.
+func (i2s I2S) Close() error {
+	// Sync wait
+	for (i2s.Bus.SYNCBUSY & sam.I2S_SYNCBUSY_ENABLE) > 0 {
+	}
+
+	// disable I2S
+	i2s.Bus.CTRLA &^= sam.I2S_CTRLA_ENABLE
+
+	return nil
+}
+
+func waitForSync() {
+	for (sam.GCLK.STATUS & sam.GCLK_STATUS_SYNCBUSY) > 0 {
+	}
+}
+
 // SPI
 type SPI struct {
 	Bus *sam.SERCOM_SPI_Type

src/machine/machine_attiny.go

@@ -9,19 +9,19 @@ import (
 // Configure sets the pin to input or output.
 func (p GPIO) Configure(config GPIOConfig) {
 	if config.Mode == GPIO_OUTPUT { // set output bit
-		*avr.DDRB |= 1 << p.Pin
+		avr.DDRB.SetBits(1 << p.Pin)
 	} else { // configure input: clear output bit
-		*avr.DDRB &^= 1 << p.Pin
+		avr.DDRB.ClearBits(1 << p.Pin)
 	}
 }
 
-func (p GPIO) getPortMask() (*avr.RegValue, uint8) {
+func (p GPIO) getPortMask() (*avr.Register8, uint8) {
 	return avr.PORTB, 1 << p.Pin
 }
 
 // Get returns the current value of a GPIO pin.
 func (p GPIO) Get() bool {
-	val := *avr.PINB & (1 << p.Pin)
+	val := avr.PINB.Get() & (1 << p.Pin)
 	return (val > 0)
 }
 

src/machine/machine_avr.go

@@ -17,10 +17,10 @@ const (
 func (p GPIO) Set(value bool) {
 	if value { // set bits
 		port, mask := p.PortMaskSet()
-		*port = mask
+		port.Set(mask)
 	} else { // clear bits
 		port, mask := p.PortMaskClear()
-		*port = mask
+		port.Set(mask)
 	}
 }
 
@@ -30,9 +30,9 @@ func (p GPIO) Set(value bool) {
 // Warning: there are no separate pin set/clear registers on the AVR. The
 // returned mask is only valid as long as no other pin in the same port has been
 // changed.
-func (p GPIO) PortMaskSet() (*avr.RegValue, avr.RegValue) {
+func (p GPIO) PortMaskSet() (*avr.Register8, uint8) {
 	port, mask := p.getPortMask()
-	return port, *port | avr.RegValue(mask)
+	return port, port.Get() | mask
 }
 
 // Return the register and mask to disable a given port. This can be used to
@@ -41,18 +41,18 @@ func (p GPIO) PortMaskSet() (*avr.RegValue, avr.RegValue) {
 // Warning: there are no separate pin set/clear registers on the AVR. The
 // returned mask is only valid as long as no other pin in the same port has been
 // changed.
-func (p GPIO) PortMaskClear() (*avr.RegValue, avr.RegValue) {
+func (p GPIO) PortMaskClear() (*avr.Register8, uint8) {
 	port, mask := p.getPortMask()
-	return port, *port &^ avr.RegValue(mask)
+	return port, port.Get() &^ mask
 }
 
 // InitADC initializes the registers needed for ADC.
 func InitADC() {
 	// set a2d prescaler so we are inside the desired 50-200 KHz range at 16MHz.
-	*avr.ADCSRA |= (avr.ADCSRA_ADPS2 | avr.ADCSRA_ADPS1 | avr.ADCSRA_ADPS0)
+	avr.ADCSRA.SetBits(avr.ADCSRA_ADPS2 | avr.ADCSRA_ADPS1 | avr.ADCSRA_ADPS0)
 
 	// enable a2d conversions
-	*avr.ADCSRA |= avr.ADCSRA_ADEN
+	avr.ADCSRA.SetBits(avr.ADCSRA_ADEN)
 }
 
 // Configure configures a ADCPin to be able to be used to read data.
@@ -68,18 +68,16 @@ func (a ADC) Get() uint16 {
 	// set the ADLAR bit (left-adjusted result) to get a value scaled to 16
 	// bits. This has the same effect as shifting the return value left by 6
 	// bits.
-	*avr.ADMUX = avr.RegValue(avr.ADMUX_REFS0 | avr.ADMUX_ADLAR | (a.Pin & 0x07))
+	avr.ADMUX.Set(avr.ADMUX_REFS0 | avr.ADMUX_ADLAR | (a.Pin & 0x07))
 
 	// start the conversion
-	*avr.ADCSRA |= avr.ADCSRA_ADSC
+	avr.ADCSRA.SetBits(avr.ADCSRA_ADSC)
 
 	// ADSC is cleared when the conversion finishes
-	for ok := true; ok; ok = (*avr.ADCSRA & avr.ADCSRA_ADSC) > 0 {
+	for ok := true; ok; ok = (avr.ADCSRA.Get() & avr.ADCSRA_ADSC) > 0 {
 	}
 
-	low := uint16(*avr.ADCL)
+	return uint16(avr.ADCL.Get()) | uint16(avr.ADCH.Get())<<8
-	high := uint16(*avr.ADCH)
-	return uint16(low) | uint16(high<<8)
 }
 
 // I2C on AVR.

src/machine/machine_nrf.go

@@ -185,21 +185,20 @@ func (i2c I2C) Tx(addr uint16, w, r []byte) error {
 		}
 	}
 	if len(r) != 0 {
+		// To trigger suspend task when a byte is received
+		i2c.Bus.SHORTS = nrf.TWI_SHORTS_BB_SUSPEND
 		i2c.Bus.TASKS_STARTRX = 1 // re-start transmission for reading
 		for i := range r {        // read each char
 			if i+1 == len(r) {
-				// The 'stop' signal must be sent before reading back the last
+				// To trigger stop task when last byte is received, set before resume task.
-				// byte, so that it will be sent by the I2C peripheral right
+				i2c.Bus.SHORTS = nrf.TWI_SHORTS_BB_STOP
-				// after the last byte has been read.
+			}
-				r[i] = i2c.readLastByte()
+			i2c.Bus.TASKS_RESUME = 1 // re-start transmission for reading
-			} else {
 			r[i] = i2c.readByte()
 		}
 	}
-	} else {
-		// Nothing to read back. Stop the transmission.
 	i2c.signalStop()
-	}
+	i2c.Bus.SHORTS = nrf.TWI_SHORTS_BB_SUSPEND_Disabled
 	return nil
 }
 
@@ -229,16 +228,6 @@ func (i2c I2C) readByte() byte {
 	return byte(i2c.Bus.RXD)
 }
 
-// readLastByte reads a single byte from the I2C bus, sending a stop signal
-// after it has been read.
-func (i2c I2C) readLastByte() byte {
-	for i2c.Bus.EVENTS_RXDREADY == 0 {
-	}
-	i2c.Bus.EVENTS_RXDREADY = 0
-	i2c.signalStop() // signal 'stop' now, so it is sent when reading RXD
-	return byte(i2c.Bus.RXD)
-}
-
 // SPI on the NRF.
 type SPI struct {
 	Bus *nrf.SPI_Type

src/runtime/gc.go

@@ -1,5 +1,3 @@
-// +build gc.marksweep
-
 package runtime
 
 // This memory manager is a textbook mark/sweep implementation, heavily inspired
@@ -168,42 +166,9 @@ func (b gcBlock) unmark() {
 	}
 }
 
-// Initialize the memory allocator.
-// No memory may be allocated before this is called. That means the runtime and
-// any packages the runtime depends upon may not allocate memory during package
-// initialization.
-func init() {
-	totalSize := heapEnd - heapStart
-
-	// Allocate some memory to keep 2 bits of information about every block.
-	metadataSize := totalSize / (blocksPerStateByte * bytesPerBlock)
-
-	// Align the pool.
-	poolStart = (heapStart + metadataSize + (bytesPerBlock - 1)) &^ (bytesPerBlock - 1)
-	poolEnd := heapEnd &^ (bytesPerBlock - 1)
-	numBlocks := (poolEnd - poolStart) / bytesPerBlock
-	endBlock = gcBlock(numBlocks)
-	if gcDebug {
-		println("heapStart:        ", heapStart)
-		println("heapEnd:          ", heapEnd)
-		println("total size:       ", totalSize)
-		println("metadata size:    ", metadataSize)
-		println("poolStart:        ", poolStart)
-		println("# of blocks:      ", numBlocks)
-		println("# of block states:", metadataSize*blocksPerStateByte)
-	}
-	if gcAsserts && metadataSize*blocksPerStateByte < numBlocks {
-		// sanity check
-		runtimePanic("gc: metadata array is too small")
-	}
-
-	// Set all block states to 'free'.
-	memzero(unsafe.Pointer(heapStart), metadataSize)
-}
-
 // alloc tries to find some free space on the heap, possibly doing a garbage
 // collection cycle if needed. If no space is free, it panics.
-func alloc(size uintptr) unsafe.Pointer {
+func heapAlloc(size uintptr) unsafe.Pointer {
 	if size == 0 {
 		return unsafe.Pointer(&zeroSizedAlloc)
 	}
@@ -275,53 +240,6 @@ func free(ptr unsafe.Pointer) {
 	// TODO: free blocks on request, when the compiler knows they're unused.
 }
 
-// GC performs a garbage collection cycle.
-func GC() {
-	if gcDebug {
-		println("running collection cycle...")
-	}
-
-	// Mark phase: mark all reachable objects, recursively.
-	markRoots(globalsStart, globalsEnd)
-	markRoots(getCurrentStackPointer(), stackTop) // assume a descending stack
-
-	// Sweep phase: free all non-marked objects and unmark marked objects for
-	// the next collection cycle.
-	sweep()
-
-	// Show how much has been sweeped, for debugging.
-	if gcDebug {
-		dumpHeap()
-	}
-}
-
-// markRoots reads all pointers from start to end (exclusive) and if they look
-// like a heap pointer and are unmarked, marks them and scans that object as
-// well (recursively). The start and end parameters must be valid pointers and
-// must be aligned.
-func markRoots(start, end uintptr) {
-	if gcDebug {
-		println("mark from", start, "to", end, int(end-start))
-	}
-
-	for addr := start; addr != end; addr += unsafe.Sizeof(addr) {
-		root := *(*uintptr)(unsafe.Pointer(addr))
-		if looksLikePointer(root) {
-			block := blockFromAddr(root)
-			head := block.findHead()
-			if head.state() != blockStateMark {
-				if gcDebug {
-					println("found unmarked pointer", root, "at address", addr)
-				}
-				head.setState(blockStateMark)
-				next := block.findNext()
-				// TODO: avoid recursion as much as possible
-				markRoots(head.address(), next.address())
-			}
-		}
-	}
-}
-
 // Sweep goes through all memory and frees unmarked memory.
 func sweep() {
 	freeCurrentObject := false
@@ -347,10 +265,8 @@ func sweep() {
 	}
 }
 
-// looksLikePointer returns whether this could be a pointer. Currently, it
+// addressOnHeap returns whether this address points into the heap.
-// simply returns whether it lies anywhere in the heap. Go allows interior
+func addressOnHeap(ptr uintptr) bool {
-// pointers so we can't check alignment or anything like that.
-func looksLikePointer(ptr uintptr) bool {
 	return ptr >= poolStart && ptr < heapEnd
 }
 

src/runtime/gc_conservative.go

@@ -0,0 +1,91 @@
+// +build gc.conservative
+
+package runtime
+
+import (
+	"unsafe"
+)
+
+// Initialize the memory allocator.
+// No memory may be allocated before this is called. That means the runtime and
+// any packages the runtime depends upon may not allocate memory during package
+// initialization.
+func init() {
+	totalSize := heapEnd - heapStart
+
+	// Allocate some memory to keep 2 bits of information about every block.
+	metadataSize := totalSize / (blocksPerStateByte * bytesPerBlock)
+
+	// Align the pool.
+	poolStart = (heapStart + metadataSize + (bytesPerBlock - 1)) &^ (bytesPerBlock - 1)
+	poolEnd := heapEnd &^ (bytesPerBlock - 1)
+	numBlocks := (poolEnd - poolStart) / bytesPerBlock
+	endBlock = gcBlock(numBlocks)
+	if gcDebug {
+		println("heapStart:        ", heapStart)
+		println("heapEnd:          ", heapEnd)
+		println("total size:       ", totalSize)
+		println("metadata size:    ", metadataSize)
+		println("poolStart:        ", poolStart)
+		println("# of blocks:      ", numBlocks)
+		println("# of block states:", metadataSize*blocksPerStateByte)
+	}
+	if gcAsserts && metadataSize*blocksPerStateByte < numBlocks {
+		// sanity check
+		runtimePanic("gc: metadata array is too small")
+	}
+
+	// Set all block states to 'free'.
+	memzero(unsafe.Pointer(heapStart), metadataSize)
+}
+
+func alloc(size uintptr) unsafe.Pointer {
+	return heapAlloc(size)
+}
+
+// GC performs a garbage collection cycle.
+func GC() {
+	if gcDebug {
+		println("running collection cycle...")
+	}
+
+	// Mark phase: mark all reachable objects, recursively.
+	markRoots(globalsStart, globalsEnd)
+	markRoots(getCurrentStackPointer(), stackTop) // assume a descending stack
+
+	// Sweep phase: free all non-marked objects and unmark marked objects for
+	// the next collection cycle.
+	sweep()
+
+	// Show how much has been sweeped, for debugging.
+	if gcDebug {
+		dumpHeap()
+	}
+}
+
+// markRoots reads all pointers from start to end (exclusive) and if they look
+// like a heap pointer and are unmarked, marks them and scans that object as
+// well (recursively). The start and end parameters must be valid pointers and
+// must be aligned.
+func markRoots(start, end uintptr) {
+	if gcDebug {
+		println("mark from", start, "to", end, int(end-start))
+	}
+
+	for addr := start; addr != end; addr += unsafe.Sizeof(addr) {
+		root := *(*uintptr)(unsafe.Pointer(addr))
+		if addressOnHeap(root) {
+			block := blockFromAddr(root)
+			head := block.findHead()
+			if head.state() != blockStateMark {
+				if gcDebug {
+					println("found unmarked pointer", root, "at address", addr)
+				}
+				head.setState(blockStateMark)
+				next := block.findNext()
+				// TODO: avoid recursion as much as possible
+				markRoots(head.address(), next.address())
+			}
+		}
+	}
+}

src/runtime/gc_leaking.go

@@ -1,4 +1,4 @@
-// +build gc.dumb
+// +build gc.leaking
 
 package runtime
 
@@ -30,18 +30,6 @@ func alloc(size uintptr) unsafe.Pointer {
 	return unsafe.Pointer(addr)
 }
 
-func free(ptr unsafe.Pointer) {
-	// Memory is never freed.
-}
-
 func GC() {
 	// No-op.
 }
-
-func KeepAlive(x interface{}) {
-	// Unimplemented. Only required with SetFinalizer().
-}
-
-func SetFinalizer(obj interface{}, finalizer interface{}) {
-	// Unimplemented.
-}

src/runtime/gc_none.go

@@ -12,18 +12,6 @@ import (
 
 func alloc(size uintptr) unsafe.Pointer
 
-func free(ptr unsafe.Pointer) {
-	// Nothing to free when nothing gets allocated.
-}
-
 func GC() {
 	// Unimplemented.
 }
-
-func KeepAlive(x interface{}) {
-	// Unimplemented. Only required with SetFinalizer().
-}
-
-func SetFinalizer(obj interface{}, finalizer interface{}) {
-	// Unimplemented.
-}

src/runtime/gc_precise.go

@@ -0,0 +1,116 @@
+// +build gc.precise
+
+package runtime
+
+import (
+	"unsafe"
+)
+
+//go:extern runtime.trackedGlobalsStart
+var trackedGlobalsStart uintptr
+
+//go:extern runtime.trackedGlobalsLength
+var trackedGlobalsLength uintptr
+
+//go:extern runtime.trackedGlobalsBitmap
+var trackedGlobalsBitmap [0]uint8
+
+// Initialize the memory allocator.
+// No memory may be allocated before this is called. That means the runtime and
+// any packages the runtime depends upon may not allocate memory during package
+// initialization.
+func init() {
+	totalSize := heapEnd - heapStart
+
+	// Allocate some memory to keep 2 bits of information about every block.
+	metadataSize := totalSize / (blocksPerStateByte * bytesPerBlock)
+
+	// Align the pool.
+	poolStart = (heapStart + metadataSize + (bytesPerBlock - 1)) &^ (bytesPerBlock - 1)
+	poolEnd := heapEnd &^ (bytesPerBlock - 1)
+	numBlocks := (poolEnd - poolStart) / bytesPerBlock
+	endBlock = gcBlock(numBlocks)
+	if gcDebug {
+		println("heapStart:        ", heapStart)
+		println("heapEnd:          ", heapEnd)
+		println("total size:       ", totalSize)
+		println("metadata size:    ", metadataSize)
+		println("poolStart:        ", poolStart)
+		println("# of blocks:      ", numBlocks)
+		println("# of block states:", metadataSize*blocksPerStateByte)
+	}
+	if gcAsserts && metadataSize*blocksPerStateByte < numBlocks {
+		// sanity check
+		runtimePanic("gc: metadata array is too small")
+	}
+
+	// Set all block states to 'free'.
+	memzero(unsafe.Pointer(heapStart), metadataSize)
+}
+
+func alloc(size uintptr) unsafe.Pointer {
+	GC()
+	return heapAlloc(size)
+}
+
+// GC performs a garbage collection cycle.
+func GC() {
+	if gcDebug {
+		println("\nrunning collection cycle...")
+	}
+
+	// Mark phase: mark all reachable objects, recursively.
+	markGlobals()
+	markRoots(getCurrentStackPointer(), stackTop) // assume a descending stack
+
+	// Sweep phase: free all non-marked objects and unmark marked objects for
+	// the next collection cycle.
+	sweep()
+
+	// Show how much has been sweeped, for debugging.
+	if gcDebug {
+		dumpHeap()
+	}
+}
+
+//go:nobounds
+func markGlobals() {
+	for i := uintptr(0); i < trackedGlobalsLength; i++ {
+		if trackedGlobalsBitmap[i/8]&(1<<(i%8)) != 0 {
+			addr := trackedGlobalsStart + i*unsafe.Alignof(uintptr(0))
+			root := *(*uintptr)(unsafe.Pointer(addr))
+			markRoot(addr, root)
+		}
+	}
+}
+
+// markRoots reads all pointers from start to end (exclusive) and if they look
+// like a heap pointer and are unmarked, marks them and scans that object as
+// well (recursively). The start and end parameters must be valid pointers and
+// must be aligned.
+func markRoots(start, end uintptr) {
+	if gcDebug {
+		println("mark from", start, "to", end, int(end-start))
+	}
+
+	for addr := start; addr != end; addr += unsafe.Sizeof(addr) {
+		root := *(*uintptr)(unsafe.Pointer(addr))
+		markRoot(addr, root)
+	}
+}
+
+func markRoot(addr, root uintptr) {
+	if addressOnHeap(root) {
+		block := blockFromAddr(root)
+		head := block.findHead()
+		if head.state() != blockStateMark {
+			if gcDebug {
+				println("found unmarked pointer", root, "at address", addr)
+			}
+			head.setState(blockStateMark)
+			next := block.findNext()
+			// TODO: avoid recursion as much as possible
+			markRoots(head.address(), next.address())
+		}
+	}
+}

src/runtime/hashmap.go

@@ -60,14 +60,20 @@ func hashmapTopHash(hash uint32) uint8 {
 }
 
 // Create a new hashmap with the given keySize and valueSize.
-func hashmapMake(keySize, valueSize uint8) *hashmap {
+func hashmapMake(keySize, valueSize uint8, sizeHint uintptr) *hashmap {
+	numBuckets := sizeHint / 8
+	bucketBits := uint8(0)
+	for numBuckets != 0 {
+		numBuckets /= 2
+		bucketBits++
+	}
 	bucketBufSize := unsafe.Sizeof(hashmapBucket{}) + uintptr(keySize)*8 + uintptr(valueSize)*8
-	bucket := alloc(bucketBufSize)
+	buckets := alloc(bucketBufSize * (1 << bucketBits))
 	return &hashmap{
-		buckets:    bucket,
+		buckets:    buckets,
 		keySize:    keySize,
 		valueSize:  valueSize,
-		bucketBits: 0,
+		bucketBits: bucketBits,
 	}
 }
 
@@ -83,13 +89,20 @@ func hashmapLen(m *hashmap) int {
 // Set a specified key to a given value. Grow the map if necessary.
 //go:nobounds
 func hashmapSet(m *hashmap, key unsafe.Pointer, value unsafe.Pointer, hash uint32, keyEqual func(x, y unsafe.Pointer, n uintptr) bool) {
+	tophash := hashmapTopHash(hash)
+
+	if m.buckets == nil {
+		// No bucket was allocated yet, do so now.
+		m.buckets = unsafe.Pointer(hashmapInsertIntoNewBucket(m, key, value, tophash))
+		return
+	}
+
 	numBuckets := uintptr(1) << m.bucketBits
 	bucketNumber := (uintptr(hash) & (numBuckets - 1))
 	bucketSize := unsafe.Sizeof(hashmapBucket{}) + uintptr(m.keySize)*8 + uintptr(m.valueSize)*8
 	bucketAddr := uintptr(m.buckets) + bucketSize*bucketNumber
 	bucket := (*hashmapBucket)(unsafe.Pointer(bucketAddr))
-
+	var lastBucket *hashmapBucket
-	tophash := hashmapTopHash(hash)
 
 	// See whether the key already exists somewhere.
 	var emptySlotKey unsafe.Pointer
@@ -98,9 +111,9 @@ func hashmapSet(m *hashmap, key unsafe.Pointer, value unsafe.Pointer, hash uint3
 	for bucket != nil {
 		for i := uintptr(0); i < 8; i++ {
 			slotKeyOffset := unsafe.Sizeof(hashmapBucket{}) + uintptr(m.keySize)*uintptr(i)
-			slotKey := unsafe.Pointer(bucketAddr + slotKeyOffset)
+			slotKey := unsafe.Pointer(uintptr(unsafe.Pointer(bucket)) + slotKeyOffset)
 			slotValueOffset := unsafe.Sizeof(hashmapBucket{}) + uintptr(m.keySize)*8 + uintptr(m.valueSize)*uintptr(i)
-			slotValue := unsafe.Pointer(bucketAddr + slotValueOffset)
+			slotValue := unsafe.Pointer(uintptr(unsafe.Pointer(bucket)) + slotValueOffset)
 			if bucket.tophash[i] == 0 && emptySlotKey == nil {
 				// Found an empty slot, store it for if we couldn't find an
 				// existing slot.
@@ -109,7 +122,7 @@ func hashmapSet(m *hashmap, key unsafe.Pointer, value unsafe.Pointer, hash uint3
 				emptySlotTophash = &bucket.tophash[i]
 			}
 			if bucket.tophash[i] == tophash {
-				// Could be an existing value that's the same.
+				// Could be an existing key that's the same.
 				if keyEqual(key, slotKey, uintptr(m.keySize)) {
 					// found same key, replace it
 					memcpy(slotValue, value, uintptr(m.valueSize))
@@ -117,16 +130,37 @@ func hashmapSet(m *hashmap, key unsafe.Pointer, value unsafe.Pointer, hash uint3
 				}
 			}
 		}
+		lastBucket = bucket
 		bucket = bucket.next
 	}
-	if emptySlotKey != nil {
+	if emptySlotKey == nil {
+		// Add a new bucket to the bucket chain.
+		// TODO: rebalance if necessary to avoid O(n) insert and lookup time.
+		lastBucket.next = (*hashmapBucket)(hashmapInsertIntoNewBucket(m, key, value, tophash))
+		return
+	}
 	m.count++
 	memcpy(emptySlotKey, key, uintptr(m.keySize))
 	memcpy(emptySlotValue, value, uintptr(m.valueSize))
 	*emptySlotTophash = tophash
-		return
 }
-	panic("todo: hashmap: grow bucket")
+
+// hashmapInsertIntoNewBucket creates a new bucket, inserts the given key and
+// value into the bucket, and returns a pointer to this bucket.
+func hashmapInsertIntoNewBucket(m *hashmap, key, value unsafe.Pointer, tophash uint8) *hashmapBucket {
+	bucketBufSize := unsafe.Sizeof(hashmapBucket{}) + uintptr(m.keySize)*8 + uintptr(m.valueSize)*8
+	bucketBuf := alloc(bucketBufSize)
+	// Insert into the first slot, which is empty as it has just been allocated.
+	slotKeyOffset := unsafe.Sizeof(hashmapBucket{})
+	slotKey := unsafe.Pointer(uintptr(bucketBuf) + slotKeyOffset)
+	slotValueOffset := unsafe.Sizeof(hashmapBucket{}) + uintptr(m.keySize)*8
+	slotValue := unsafe.Pointer(uintptr(bucketBuf) + slotValueOffset)
+	m.count++
+	memcpy(slotKey, key, uintptr(m.keySize))
+	memcpy(slotValue, value, uintptr(m.valueSize))
+	bucket := (*hashmapBucket)(bucketBuf)
+	bucket.tophash[0] = tophash
+	return bucket
 }
 
 // Get the value of a specified key, or zero the value if not found.

src/runtime/runtime_atmega.go

@@ -17,19 +17,19 @@ func sleepWDT(period uint8) {
 	avr.Asm("cli")
 	avr.Asm("wdr")
 	// Start timed sequence.
-	*avr.WDTCSR |= avr.WDTCSR_WDCE | avr.WDTCSR_WDE
+	avr.WDTCSR.SetBits(avr.WDTCSR_WDCE | avr.WDTCSR_WDE)
 	// Enable WDT and set new timeout
-	*avr.WDTCSR = avr.WDTCSR_WDIE | avr.RegValue(period)
+	avr.WDTCSR.SetBits(avr.WDTCSR_WDIE | period)
 	avr.Asm("sei")
 
 	// Set sleep mode to idle and enable sleep mode.
 	// Note: when using something other than idle, the UART won't work
 	// correctly. This needs to be fixed, though, so we can truly sleep.
-	*avr.SMCR = (0 << 1) | avr.SMCR_SE
+	avr.SMCR.Set((0 << 1) | avr.SMCR_SE)
 
 	// go to sleep
 	avr.Asm("sleep")
 
 	// disable sleep
-	*avr.SMCR = 0
+	avr.SMCR.Set(0)
 }

src/runtime/volatile/volatile.go

@@ -0,0 +1,34 @@
+// Package volatile provides definitions for volatile loads and stores. These
+// are implemented as compiler builtins.
+//
+// The load operations load a volatile value. The store operations store to a
+// volatile value. The compiler will emit exactly one load or store operation
+// when possible and will not reorder volatile operations. However, the compiler
+// may move other operations across load/store operations, so make sure that all
+// relevant loads/stores are done in a volatile way if this is a problem.
+//
+// These loads and stores are commonly used to read/write values from memory
+// mapped peripheral devices. They do not provide atomicity, use the sync/atomic
+// package for that.
+//
+// For more details: https://llvm.org/docs/LangRef.html#volatile-memory-accesses
+// and https://blog.regehr.org/archives/28.
+package volatile
+
+// LoadUint8 loads the volatile value *addr.
+func LoadUint8(addr *uint8) (val uint8)
+
+// LoadUint16 loads the volatile value *addr.
+func LoadUint16(addr *uint16) (val uint16)
+
+// LoadUint32 loads the volatile value *addr.
+func LoadUint32(addr *uint32) (val uint32)
+
+// StoreUint8 stores val to the volatile value *addr.
+func StoreUint8(addr *uint8, val uint8)
+
+// StoreUint16 stores val to the volatile value *addr.
+func StoreUint16(addr *uint16, val uint16)
+
+// StoreUint32 stores val to the volatile value *addr.
+func StoreUint32(addr *uint32, val uint32)

targets/cortex-m.json

@@ -3,7 +3,7 @@
 	"goos": "linux",
 	"goarch": "arm",
 	"compiler": "clang",
-	"gc": "marksweep",
+	"gc": "precise",
 	"linker": "ld.lld",
 	"rtlib": "compiler-rt",
 	"cflags": [

testdata/calls.go

@@ -4,6 +4,22 @@ type Thing struct {
 	name string
 }
 
+type ThingOption func(*Thing)
+
+func WithName(name string) ThingOption {
+	return func(t *Thing) {
+		t.name = name
+	}
+}
+
+func NewThing(opts ...ThingOption) *Thing {
+	t := &Thing{}
+	for _, opt := range opts {
+		opt(t)
+	}
+	return t
+}
+
 func (t Thing) String() string {
 	return t.name
 }
@@ -36,6 +52,12 @@ func main() {
 	runFunc(func(i int) {
 		println("inside fp closure:", thing.String(), i)
 	}, 3)
+
+	// functional arguments
+	thingFunctionalArgs1 := NewThing()
+	thingFunctionalArgs1.Print("functional args 1")
+	thingFunctionalArgs2 := NewThing(WithName("named thing"))
+	thingFunctionalArgs2.Print("functional args 2")
 }
 
 func runFunc(f func(int), arg int) {

testdata/calls.txt

@@ -7,3 +7,5 @@ Thing.Print: foo arg: bar
 bound method: foo
 thing inside closure: foo
 inside fp closure: foo 3
+Thing.Print:  arg: functional args 1
+Thing.Print: named thing arg: functional args 2

testdata/cgo/extra.go

@@ -0,0 +1,6 @@
+package main
+
+// Make sure CGo supports multiple files.
+
+// int fortytwo(void);
+import "C"

testdata/cgo/main.go

@@ -7,6 +7,8 @@ int mul(int, int);
 */
 import "C"
 
+import "C"
+
 import "unsafe"
 
 func main() {
@@ -18,6 +20,10 @@ func main() {
 	var y C.longlong = -(1 << 40)
 	println("longlong:", y)
 	println("global:", C.global)
+	println("defined ints:", C.CONST_INT, C.CONST_INT2)
+	println("defined floats:", C.CONST_FLOAT, C.CONST_FLOAT2)
+	println("defined string:", C.CONST_STRING)
+	println("defined char:", C.CONST_CHAR)
 	var ptr C.intPointer
 	var n C.int = 15
 	ptr = C.intPointer(&n)

testdata/cgo/main.h

@@ -10,6 +10,13 @@ int doCallback(int a, int b, binop_t cb);
 typedef int * intPointer;
 void store(int value, int *ptr);
 
+# define CONST_INT 5
+# define CONST_INT2 5llu
+# define CONST_FLOAT 5.8
+# define CONST_FLOAT2 5.8f
+# define CONST_CHAR 'c'
+# define CONST_STRING "defined string"
+
 // this signature should not be included by CGo
 void unusedFunction2(int x, __builtin_va_list args);
 

testdata/cgo/out.txt

@@ -4,6 +4,10 @@ myint: 3 5
 myint size: 2
 longlong: -1099511627776
 global: 3
+defined ints: 5 5
+defined floats: +5.800000e+000 +5.800000e+000
+defined string: defined string
+defined char: 99
 15: 15
 25: 25
 callback 1: 50

testdata/channel.go

@@ -4,7 +4,7 @@ import "time"
 
 func main() {
 	ch := make(chan int)
-	println("len, cap of channel:", len(ch), cap(ch))
+	println("len, cap of channel:", len(ch), cap(ch), ch == nil)
 	go sender(ch)
 
 	n, ok := <-ch

testdata/channel.txt

@@ -1,4 +1,4 @@
-len, cap of channel: 0 0
+len, cap of channel: 0 0 false
 recv from open channel: 1 true
 received num: 2
 received num: 3

testdata/map.go

@@ -47,6 +47,18 @@ func main() {
 	println(testMapArrayKey[arrKey])
 	testMapArrayKey[arrKey] = 5555
 	println(testMapArrayKey[arrKey])
+
+	// test preallocated map
+	squares := make(map[int]int, 200)
+	testBigMap(squares, 100)
+	println("tested preallocated map")
+
+	// test growing maps
+	squares = make(map[int]int, 0)
+	testBigMap(squares, 10)
+	squares = make(map[int]int, 20)
+	testBigMap(squares, 40)
+	println("tested growing of a map")
 }
 
 func readMap(m map[string]int, key string) {
@@ -56,7 +68,27 @@ func readMap(m map[string]int, key string) {
 		println(" ", k, "=", v)
 	}
 }
+
 func lookup(m map[string]int, key string) {
 	value, ok := m[key]
 	println("lookup with comma-ok:", key, value, ok)
 }
+
+func testBigMap(squares map[int]int, n int) {
+	for i := 0; i < n; i++ {
+		if len(squares) != i {
+			println("unexpected length:", len(squares), "at i =", i)
+		}
+		squares[i] = i*i
+		for j := 0; j <= i; j++ {
+			if v, ok := squares[j]; !ok || v != j*j {
+				if !ok {
+					println("key not found in squares map:", j)
+				} else {
+					println("unexpected value read back from squares map:", j, v)
+				}
+				return
+			}
+		}
+	}
+}

testdata/map.txt

@@ -54,3 +54,5 @@ true false 0
 42
 4321
 5555
+tested preallocated map
+tested growing of a map

tools/gen-device-avr.py

@@ -151,12 +151,54 @@ def writeGo(outdir, device):
 // {description}
 package {pkgName}
 
-import "unsafe"
+import (
+	"runtime/volatile"
+	"unsafe"
+)
 
 // Special type that causes loads/stores to be volatile (necessary for
 // memory-mapped registers).
-//go:volatile
+type Register8 struct {{
-type RegValue uint8
+	Reg uint8
+}}
+
+// Get returns the value in the register. It is the volatile equivalent of:
+//
+//     *r.Reg
+//
+//go:inline
+func (r *Register8) Get() uint8 {{
+	return volatile.LoadUint8(&r.Reg)
+}}
+
+// Set updates the register value. It is the volatile equivalent of:
+//
+//     *r.Reg = value
+//
+//go:inline
+func (r *Register8) Set(value uint8) {{
+	volatile.StoreUint8(&r.Reg, value)
+}}
+
+// SetBits reads the register, sets the given bits, and writes it back. It is
+// the volatile equivalent of:
+//
+//     r.Reg |= value
+//
+//go:inline
+func (r *Register8) SetBits(value uint8) {{
+	volatile.StoreUint8(&r.Reg, volatile.LoadUint8(&r.Reg) | value)
+}}
+
+// ClearBits reads the register, clears the given bits, and writes it back. It
+// is the volatile equivalent of:
+//
+//     r.Reg &^= value
+//
+//go:inline
+func (r *Register8) ClearBits(value uint8) {{
+	volatile.StoreUint8(&r.Reg, volatile.LoadUint8(&r.Reg) &^ value)
+}}
 
 // Some information about this device.
 const (
@@ -179,7 +221,7 @@ const (
         out.write('\n\t// {description}\n'.format(**peripheral))
         for register in peripheral['registers']:
             for variant in register['variants']:
-                out.write('\t{name} = (*RegValue)(unsafe.Pointer(uintptr(0x{address:x})))\n'.format(**variant))
+                out.write('\t{name} = (*Register8)(unsafe.Pointer(uintptr(0x{address:x})))\n'.format(**variant))
     out.write(')\n')
 
     for peripheral in device.peripherals: