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

@@ -47,6 +47,18 @@ commands:
           command: |
             curl https://raw.githubusercontent.com/golang/dep/master/install.sh | sh
             dep ensure --vendor-only
+  llvm-source-linux:
+    steps:
+      - restore_cache:
+          keys:
+            - llvm-source-8-v2
+      - run:
+          name: "Fetch LLVM source"
+          command: make llvm-source
+      - save_cache:
+          key: llvm-source-8-v2
+          paths:
+            - llvm
   smoketest:
     steps:
       - smoketest-no-avr
@@ -70,6 +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:
@@ -86,6 +99,7 @@ commands:
           keys:
             - go-cache-{{ checksum "Gopkg.lock" }}-{{ .Environment.CIRCLE_PREVIOUS_BUILD_NUM }}
             - go-cache-{{ checksum "Gopkg.lock" }}
+      - llvm-source-linux
       - dep
       - run: go install .
       - run: go test -v
@@ -105,7 +119,6 @@ commands:
           name: "Install apt dependencies"
           command: |
             sudo apt-get install \
-                libtinfo-dev \
                 python3 \
                 gcc-arm-linux-gnueabihf \
                 binutils-arm-none-eabi \
@@ -121,19 +134,10 @@ commands:
           keys:
             - go-cache-{{ checksum "Gopkg.lock" }}-{{ .Environment.CIRCLE_PREVIOUS_BUILD_NUM }}
             - go-cache-{{ checksum "Gopkg.lock" }}
+      - llvm-source-linux
       - restore_cache:
           keys:
-            - llvm-source-8-v2
+            - llvm-build-8-linux-v4
-      - run:
-          name: "Fetch LLVM source"
-          command: make llvm-source
-      - save_cache:
-          key: llvm-source-8-v2
-          paths:
-            - llvm
-      - restore_cache:
-          keys:
-            - llvm-build-8-v2
       - run:
           name: "Build LLVM"
           command: |
@@ -151,7 +155,7 @@ commands:
               make llvm-build
             fi
       - save_cache:
-          key: llvm-build-8-v2
+          key: llvm-build-8-linux-v4
           paths:
             llvm-build
       - run:
@@ -204,7 +208,7 @@ commands:
             - llvm
       - restore_cache:
           keys:
-            - llvm-build-8-macos-v2
+            - llvm-build-8-macos-v3
       - run:
           name: "Build LLVM"
           command: |
@@ -216,7 +220,7 @@ commands:
               make llvm-build
             fi
       - save_cache:
-          key: llvm-build-8-macos-v2
+          key: llvm-build-8-macos-v3
           paths:
             llvm-build
       - run:

.gitignore

@@ -10,3 +10,5 @@ src/device/stm32/*.s
 src/device/sam/*.go
 src/device/sam/*.s
 vendor
+llvm
+llvm-build

Dockerfile

@@ -10,8 +10,11 @@ RUN wget -O- https://raw.githubusercontent.com/golang/dep/master/install.sh | sh
 
 COPY . /go/src/github.com/tinygo-org/tinygo
 
+# remove submodules directories and re-init them to fix any hard-coded paths
+# after copying the tinygo directory in the previous step.
 RUN cd /go/src/github.com/tinygo-org/tinygo/ && \
-    git submodule update --init
+    rm -rf ./lib/* && \
+    git submodule update --init --recursive --force
 
 RUN cd /go/src/github.com/tinygo-org/tinygo/ && \
     dep ensure --vendor-only && \

LICENSE

@@ -1,5 +1,8 @@
 Copyright (c) 2018-2019 TinyGo Authors. All rights reserved.
 
+TinyGo includes portions of the Go standard library.
+Copyright (c) 2009-2019 The Go Authors. All rights reserved.
+
 Redistribution and use in source and binary forms, with or without
 modification, are permitted provided that the following conditions are
 met:

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:
@@ -69,7 +69,7 @@ llvm-source: llvm/README.txt llvm/tools/clang/README.txt llvm/tools/lld/README.m
 
 # Configure LLVM.
 llvm-build/build.ninja: llvm-source
-	mkdir -p llvm-build; cd llvm-build; cmake -G Ninja ../llvm "-DLLVM_TARGETS_TO_BUILD=X86;ARM;AArch64;WebAssembly" "-DLLVM_EXPERIMENTAL_TARGETS_TO_BUILD=AVR" -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_ASSERTIONS=OFF -DLIBCLANG_BUILD_STATIC=ON
+	mkdir -p llvm-build; cd llvm-build; cmake -G Ninja ../llvm "-DLLVM_TARGETS_TO_BUILD=X86;ARM;AArch64;WebAssembly" "-DLLVM_EXPERIMENTAL_TARGETS_TO_BUILD=AVR" -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_ASSERTIONS=OFF -DLIBCLANG_BUILD_STATIC=ON -DLLVM_ENABLE_TERMINFO=OFF -DLLVM_ENABLE_ZLIB=OFF
 
 # Build LLVM.
 llvm-build: llvm-build/build.ninja
@@ -86,13 +86,16 @@ test:
 
 release: build/tinygo gen-device
 	@mkdir -p build/release/tinygo/bin
+	@mkdir -p build/release/tinygo/lib/clang/include
 	@mkdir -p build/release/tinygo/lib/CMSIS/CMSIS
 	@mkdir -p build/release/tinygo/lib/compiler-rt/lib
 	@mkdir -p build/release/tinygo/lib/nrfx
 	@mkdir -p build/release/tinygo/pkg/armv6m-none-eabi
 	@mkdir -p build/release/tinygo/pkg/armv7m-none-eabi
 	@mkdir -p build/release/tinygo/pkg/armv7em-none-eabi
+	@echo copying source files
 	@cp -p  build/tinygo                 build/release/tinygo/bin
+	@cp -p $(abspath $(CLANG_SRC))/lib/Headers/*.h build/release/tinygo/lib/clang/include
 	@cp -rp lib/CMSIS/CMSIS/Include      build/release/tinygo/lib/CMSIS/CMSIS
 	@cp -rp lib/CMSIS/README.md          build/release/tinygo/lib/CMSIS
 	@cp -rp lib/compiler-rt/lib/builtins build/release/tinygo/lib/compiler-rt/lib

buildcache.go

@@ -9,9 +9,11 @@ import (
 
 // Get the cache directory, usually ~/.cache/tinygo
 func cacheDir() string {
-	home := getHomeDir()
+	dir, err := os.UserCacheDir()
-	dir := filepath.Join(home, ".cache", "tinygo")
+	if err != nil {
-	return dir
+		panic("could not find cache dir: " + err.Error())
+	}
+	return filepath.Join(dir, "tinygo")
 }
 
 // Return the newest timestamp of all the file paths passed in. Used to check

builtins.go

@@ -5,7 +5,6 @@ import (
 	"io"
 	"io/ioutil"
 	"os"
-	"os/exec"
 	"path/filepath"
 	"strings"
 	"time"
@@ -192,13 +191,13 @@ func loadBuiltins(target string) (path string, err error) {
 		srcs[i] = filepath.Join(builtinsDir, name)
 	}
 
-	if path, err := cacheLoad(outfile, commands["clang"], srcs); path != "" || err != nil {
+	if path, err := cacheLoad(outfile, commands["clang"][0], srcs); path != "" || err != nil {
 		return path, err
 	}
 
 	var cachepath string
 	err = compileBuiltins(target, func(path string) error {
-		path, err := cacheStore(path, outfile, commands["clang"], srcs)
+		path, err := cacheStore(path, outfile, commands["clang"][0], srcs)
 		cachepath = path
 		return err
 	})
@@ -240,11 +239,7 @@ func compileBuiltins(target string, callback func(path string) error) error {
 		// Note: -fdebug-prefix-map is necessary to make the output archive
 		// reproducible. Otherwise the temporary directory is stored in the
 		// archive itself, which varies each run.
-		cmd := exec.Command(commands["clang"], "-c", "-Oz", "-g", "-Werror", "-Wall", "-std=c11", "-fshort-enums", "-nostdlibinc", "-ffunction-sections", "-fdata-sections", "--target="+target, "-fdebug-prefix-map="+dir+"="+remapDir, "-o", objpath, srcpath)
+		err := execCommand(commands["clang"], "-c", "-Oz", "-g", "-Werror", "-Wall", "-std=c11", "-fshort-enums", "-nostdlibinc", "-ffunction-sections", "-fdata-sections", "--target="+target, "-fdebug-prefix-map="+dir+"="+remapDir, "-o", objpath, srcpath)
-		cmd.Stdout = os.Stdout
-		cmd.Stderr = os.Stderr
-		cmd.Dir = dir
-		err = cmd.Run()
 		if err != nil {
 			return &commandError{"failed to build", srcpath, err}
 		}
@@ -294,5 +289,6 @@ func compileBuiltins(target string, callback func(path string) error) error {
 
 	// Give the caller the resulting file. The callback must copy the file,
 	// because after it returns the temporary directory will be removed.
+	arfile.Close()
 	return callback(arpath)
 }

cgo/cgo.go

@@ -0,0 +1,636 @@
+// 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 for the C
+// source file parsing.
+
+import (
+	"go/ast"
+	"go/token"
+	"sort"
+	"strconv"
+	"strings"
+
+	"golang.org/x/tools/go/ast/astutil"
+)
+
+// cgoPackage holds all CGo-related information of a package.
+type cgoPackage struct {
+	generated       *ast.File
+	generatedPos    token.Pos
+	errors          []error
+	dir             string
+	fset            *token.FileSet
+	tokenFiles      map[string]*token.File
+	missingSymbols  map[string]struct{}
+	constants       map[string]constantInfo
+	functions       map[string]*functionInfo
+	globals         map[string]globalInfo
+	typedefs        map[string]*typedefInfo
+	elaboratedTypes map[string]*elaboratedTypeInfo
+}
+
+// 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).
+type paramInfo struct {
+	name     string
+	typeExpr ast.Expr
+}
+
+// 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
+// in both languages. See addTypeAliases.
+var cgoAliases = map[string]string{
+	"C.int8_t":    "int8",
+	"C.int16_t":   "int16",
+	"C.int32_t":   "int32",
+	"C.int64_t":   "int64",
+	"C.uint8_t":   "uint8",
+	"C.uint16_t":  "uint16",
+	"C.uint32_t":  "uint32",
+	"C.uint64_t":  "uint64",
+	"C.uintptr_t": "uintptr",
+}
+
+// builtinAliases are handled specially because they only exist on the Go side
+// of CGo, not on the CGo side (they're prefixed with "_Cgo_" there).
+var builtinAliases = map[string]struct{}{
+	"char":      struct{}{},
+	"schar":     struct{}{},
+	"uchar":     struct{}{},
+	"short":     struct{}{},
+	"ushort":    struct{}{},
+	"int":       struct{}{},
+	"uint":      struct{}{},
+	"long":      struct{}{},
+	"ulong":     struct{}{},
+	"longlong":  struct{}{},
+	"ulonglong": struct{}{},
+}
+
+// cgoTypes lists some C types with ambiguous sizes that must be retrieved
+// somehow from C. This is done by adding some typedefs to get the size of each
+// type.
+const cgoTypes = `
+typedef char                _Cgo_char;
+typedef signed char         _Cgo_schar;
+typedef unsigned char       _Cgo_uchar;
+typedef short               _Cgo_short;
+typedef unsigned short      _Cgo_ushort;
+typedef int                 _Cgo_int;
+typedef unsigned int        _Cgo_uint;
+typedef long                _Cgo_long;
+typedef unsigned long       _Cgo_ulong;
+typedef long long           _Cgo_longlong;
+typedef unsigned long long  _Cgo_ulonglong;
+`
+
+// Process extracts `import "C"` statements from the AST, parses the comment
+// with libclang, and modifies the AST to use this information. It returns a
+// newly created *ast.File that should be added to the list of to-be-parsed
+// files. If there is one or more error, it returns these in the []error slice
+// but still modifies the AST.
+func Process(files []*ast.File, dir string, fset *token.FileSet, cflags []string) (*ast.File, []error) {
+	p := &cgoPackage{
+		dir:             dir,
+		fset:            fset,
+		tokenFiles:      map[string]*token.File{},
+		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.
+	for _, f := range files {
+		astutil.Apply(f, p.findMissingCGoNames, nil)
+	}
+	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)
+			if !ok {
+				continue
+			}
+			if len(genDecl.Specs) != 1 {
+				continue
+			}
+			spec, ok := genDecl.Specs[0].(*ast.ImportSpec)
+			if !ok {
+				continue
+			}
+			path, err := strconv.Unquote(spec.Path.Value)
+			if err != nil {
+				panic("could not parse import path: " + err.Error())
+			}
+			if path != "C" {
+				continue
+			}
+			cgoComment := genDecl.Doc.Text()
+
+			pos := genDecl.Pos()
+			if genDecl.Doc != nil {
+				pos = genDecl.Doc.Pos()
+			}
+			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:]...)
+			i--
+		}
+
+		// Print the AST, for debugging.
+		//ast.Print(fset, f)
+	}
+
+	// Declare functions found by libclang.
+	p.addFuncDecls()
+
+	// Declare stub function pointer values found by libclang.
+	p.addFuncPtrDecls()
+
+	// Declare globals found by libclang.
+	p.addConstDecls()
+
+	// Declare globals found by libclang.
+	p.addVarDecls()
+
+	// Forward C types to Go types (like C.uint32_t -> uint32).
+	p.addTypeAliases()
+
+	// Add type declarations for C types, declared using typedef in C.
+	p.addTypedefs()
+
+	// Add elaborated types for C structs and unions.
+	p.addElaboratedTypes()
+
+	// Patch the AST to use the declared types and functions.
+	for _, f := range files {
+		astutil.Apply(f, p.walker, 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 (p *cgoPackage) addFuncDecls() {
+	names := make([]string, 0, len(p.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.Fun,
+			Name: "C." + name,
+		}
+		args := make([]*ast.Field, len(fn.args))
+		decl := &ast.FuncDecl{
+			Name: &ast.Ident{
+				NamePos: fn.pos,
+				Name:    "C." + name,
+				Obj:     obj,
+			},
+			Type: &ast.FuncType{
+				Func: fn.pos,
+				Params: &ast.FieldList{
+					Opening: fn.pos,
+					List:    args,
+					Closing: fn.pos,
+				},
+				Results: fn.results,
+			},
+		}
+		obj.Decl = decl
+		for i, arg := range fn.args {
+			args[i] = &ast.Field{
+				Names: []*ast.Ident{
+					&ast.Ident{
+						NamePos: fn.pos,
+						Name:    arg.name,
+						Obj: &ast.Object{
+							Kind: ast.Var,
+							Name: arg.name,
+							Decl: decl,
+						},
+					},
+				},
+				Type: arg.typeExpr,
+			}
+		}
+		p.generated.Decls = append(p.generated.Decls, decl)
+	}
+}
+
+// addFuncPtrDecls creates stub declarations of function pointer values. These
+// values will later be replaced with the real values in the compiler.
+// It adds code like the following to the AST:
+//
+//     var (
+//         C.add unsafe.Pointer
+//         C.mul unsafe.Pointer
+//         // ...
+//     )
+func (p *cgoPackage) addFuncPtrDecls() {
+	if len(p.functions) == 0 {
+		return
+	}
+	gen := &ast.GenDecl{
+		TokPos: token.NoPos,
+		Tok:    token.VAR,
+		Lparen: token.NoPos,
+		Rparen: token.NoPos,
+	}
+	names := make([]string, 0, len(p.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 + "$funcaddr",
+		}
+		valueSpec := &ast.ValueSpec{
+			Names: []*ast.Ident{&ast.Ident{
+				NamePos: fn.pos,
+				Name:    "C." + name + "$funcaddr",
+				Obj:     obj,
+			}},
+			Type: &ast.SelectorExpr{
+				X: &ast.Ident{
+					NamePos: fn.pos,
+					Name:    "unsafe",
+				},
+				Sel: &ast.Ident{
+					NamePos: fn.pos,
+					Name:    "Pointer",
+				},
+			},
+		}
+		obj.Decl = valueSpec
+		gen.Specs = append(gen.Specs, valueSpec)
+	}
+	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.
+// It adds code like the following to the AST:
+//
+//     var (
+//         C.globalInt  int
+//         C.globalBool bool
+//         // ...
+//     )
+func (p *cgoPackage) addVarDecls() {
+	if len(p.globals) == 0 {
+		return
+	}
+	gen := &ast.GenDecl{
+		TokPos: token.NoPos,
+		Tok:    token.VAR,
+		Lparen: token.NoPos,
+		Rparen: token.NoPos,
+	}
+	names := make([]string, 0, len(p.globals))
+	for name := range p.globals {
+		names = append(names, name)
+	}
+	sort.Strings(names)
+	for _, name := range names {
+		global := p.globals[name]
+		obj := &ast.Object{
+			Kind: ast.Var,
+			Name: "C." + name,
+		}
+		valueSpec := &ast.ValueSpec{
+			Names: []*ast.Ident{&ast.Ident{
+				NamePos: global.pos,
+				Name:    "C." + name,
+				Obj:     obj,
+			}},
+			Type: global.typeExpr,
+		}
+		obj.Decl = valueSpec
+		gen.Specs = append(gen.Specs, valueSpec)
+	}
+	p.generated.Decls = append(p.generated.Decls, gen)
+}
+
+// addTypeAliases aliases some built-in Go types with their equivalent C types.
+// It adds code like the following to the AST:
+//
+//     type (
+//         C.int8_t  = int8
+//         C.int16_t = int16
+//         // ...
+//     )
+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: token.NoPos,
+		Tok:    token.TYPE,
+		Lparen: token.NoPos,
+		Rparen: token.NoPos,
+	}
+	for _, typeName := range aliasKeys {
+		goTypeName := cgoAliases[typeName]
+		obj := &ast.Object{
+			Kind: ast.Typ,
+			Name: typeName,
+		}
+		typeSpec := &ast.TypeSpec{
+			Name: &ast.Ident{
+				NamePos: token.NoPos,
+				Name:    typeName,
+				Obj:     obj,
+			},
+			Assign: p.generatedPos,
+			Type: &ast.Ident{
+				NamePos: token.NoPos,
+				Name:    goTypeName,
+			},
+		}
+		obj.Decl = typeSpec
+		gen.Specs = append(gen.Specs, typeSpec)
+	}
+	p.generated.Decls = append(p.generated.Decls, gen)
+}
+
+func (p *cgoPackage) addTypedefs() {
+	if len(p.typedefs) == 0 {
+		return
+	}
+	gen := &ast.GenDecl{
+		TokPos: token.NoPos,
+		Tok:    token.TYPE,
+	}
+	names := make([]string, 0, len(p.typedefs))
+	for name := range p.typedefs {
+		names = append(names, name)
+	}
+	sort.Strings(names)
+	for _, name := range names {
+		typedef := p.typedefs[name]
+		typeName := "C." + name
+		isAlias := true
+		if strings.HasPrefix(name, "_Cgo_") {
+			typeName = "C." + name[len("_Cgo_"):]
+			isAlias = false // C.short etc. should not be aliased to the equivalent Go type (not portable)
+		}
+		if _, ok := cgoAliases[typeName]; ok {
+			// This is a type that also exists in Go (defined in stdint.h).
+			continue
+		}
+		obj := &ast.Object{
+			Kind: ast.Typ,
+			Name: typeName,
+		}
+		typeSpec := &ast.TypeSpec{
+			Name: &ast.Ident{
+				NamePos: typedef.pos,
+				Name:    typeName,
+				Obj:     obj,
+			},
+			Type: typedef.typeExpr,
+		}
+		if isAlias {
+			typeSpec.Assign = typedef.pos
+		}
+		obj.Decl = typeSpec
+		gen.Specs = append(gen.Specs, typeSpec)
+	}
+	p.generated.Decls = append(p.generated.Decls, gen)
+}
+
+// addElaboratedTypes adds C elaborated types as aliases. These are the "struct
+// foo" or "union foo" types, often used in a typedef.
+//
+// See also:
+// https://en.cppreference.com/w/cpp/language/elaborated_type_specifier
+func (p *cgoPackage) addElaboratedTypes() {
+	if len(p.elaboratedTypes) == 0 {
+		return
+	}
+	gen := &ast.GenDecl{
+		TokPos: token.NoPos,
+		Tok:    token.TYPE,
+	}
+	names := make([]string, 0, len(p.elaboratedTypes))
+	for name := range p.elaboratedTypes {
+		names = append(names, name)
+	}
+	sort.Strings(names)
+	for _, name := range names {
+		typ := p.elaboratedTypes[name]
+		typeName := "C." + name
+		obj := &ast.Object{
+			Kind: ast.Typ,
+			Name: typeName,
+		}
+		typeSpec := &ast.TypeSpec{
+			Name: &ast.Ident{
+				NamePos: typ.pos,
+				Name:    typeName,
+				Obj:     obj,
+			},
+			Type: typ.typeExpr,
+		}
+		obj.Decl = typeSpec
+		gen.Specs = append(gen.Specs, typeSpec)
+	}
+	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 (p *cgoPackage) findMissingCGoNames(cursor *astutil.Cursor) bool {
+	switch node := cursor.Node().(type) {
+	case *ast.SelectorExpr:
+		x, ok := node.X.(*ast.Ident)
+		if !ok {
+			return true
+		}
+		if x.Name == "C" {
+			name := node.Sel.Name
+			if _, ok := builtinAliases[name]; ok {
+				name = "_Cgo_" + name
+			}
+			p.missingSymbols[name] = struct{}{}
+		}
+	}
+	return true
+}
+
+// walker replaces all "C".<something> expressions to literal "C.<something>"
+// 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 (p *cgoPackage) walker(cursor *astutil.Cursor) bool {
+	switch node := cursor.Node().(type) {
+	case *ast.CallExpr:
+		fun, ok := node.Fun.(*ast.SelectorExpr)
+		if !ok {
+			return true
+		}
+		x, ok := fun.X.(*ast.Ident)
+		if !ok {
+			return true
+		}
+		if _, ok := p.functions[fun.Sel.Name]; ok && x.Name == "C" {
+			node.Fun = &ast.Ident{
+				NamePos: x.NamePos,
+				Name:    "C." + fun.Sel.Name,
+			}
+		}
+	case *ast.SelectorExpr:
+		x, ok := node.X.(*ast.Ident)
+		if !ok {
+			return true
+		}
+		if x.Name == "C" {
+			name := "C." + node.Sel.Name
+			if _, ok := p.functions[node.Sel.Name]; ok {
+				name += "$funcaddr"
+			}
+			cursor.Replace(&ast.Ident{
+				NamePos: x.NamePos,
+				Name:    name,
+			})
+		}
+	}
+	return true
+}

cgo/libclang.go

@@ -0,0 +1,624 @@
+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.
+
+import (
+	"fmt"
+	"go/ast"
+	"go/scanner"
+	"go/token"
+	"path/filepath"
+	"strconv"
+	"strings"
+	"unsafe"
+)
+
+/*
+#include <clang-c/Index.h> // if this fails, install libclang-8-dev
+#include <stdlib.h>
+#include <stdint.h>
+
+// This struct should be ABI-compatible on all platforms (uintptr_t has the same
+// alignment etc. as void*) but does not include void* pointers that are not
+// always real pointers.
+// The Go garbage collector assumes that all non-nil pointer-typed integers are
+// actually pointers. This is not always true, as data[1] often contains 0x1,
+// which is clearly not a valid pointer. Usually the GC won't catch this issue,
+// but occasionally it will leading to a crash with a vague error message.
+typedef struct {
+	enum CXCursorKind kind;
+	int xdata;
+	uintptr_t data[3];
+} GoCXCursor;
+
+// Forwarding functions. They are implemented in libclang_stubs.c and forward to
+// the real functions without doing anything else, thus they are entirely
+// compatible with the versions without tinygo_ prefix. The only difference is
+// the CXCursor type, which has been replaced with GoCXCursor.
+GoCXCursor tinygo_clang_getTranslationUnitCursor(CXTranslationUnit tu);
+unsigned tinygo_clang_visitChildren(GoCXCursor parent, CXCursorVisitor visitor, CXClientData client_data);
+CXString tinygo_clang_getCursorSpelling(GoCXCursor c);
+enum CXCursorKind tinygo_clang_getCursorKind(GoCXCursor c);
+CXType tinygo_clang_getCursorType(GoCXCursor c);
+GoCXCursor tinygo_clang_getTypeDeclaration(CXType t);
+CXType tinygo_clang_getTypedefDeclUnderlyingType(GoCXCursor c);
+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);
+int tinygo_clang_struct_visitor(GoCXCursor c, GoCXCursor parent, CXClientData client_data);
+*/
+import "C"
+
+// storedRefs stores references to types, used for clang_visitChildren.
+var storedRefs refMap
+
+var diagnosticSeverity = [...]string{
+	C.CXDiagnostic_Ignored: "ignored",
+	C.CXDiagnostic_Note:    "note",
+	C.CXDiagnostic_Warning: "warning",
+	C.CXDiagnostic_Error:   "error",
+	C.CXDiagnostic_Fatal:   "fatal",
+}
+
+func (p *cgoPackage) parseFragment(fragment string, cflags []string, posFilename string, posLine int) {
+	index := C.clang_createIndex(0, 0)
+	defer C.clang_disposeIndex(index)
+
+	// pretend to be a .c file
+	filenameC := C.CString(posFilename + "!cgo.c")
+	defer C.free(unsafe.Pointer(filenameC))
+
+	// fix up error locations
+	fragment = fmt.Sprintf("# %d %#v\n", posLine+1, posFilename) + fragment
+
+	fragmentC := C.CString(fragment)
+	defer C.free(unsafe.Pointer(fragmentC))
+
+	unsavedFile := C.struct_CXUnsavedFile{
+		Filename: filenameC,
+		Length:   C.ulong(len(fragment)),
+		Contents: fragmentC,
+	}
+
+	// convert Go slice of strings to C array of strings.
+	cmdargsC := C.malloc(C.size_t(len(cflags)) * C.size_t(unsafe.Sizeof(uintptr(0))))
+	defer C.free(cmdargsC)
+	cmdargs := (*[1 << 16]*C.char)(cmdargsC)
+	for i, cflag := range cflags {
+		s := C.CString(cflag)
+		cmdargs[i] = s
+		defer C.free(unsafe.Pointer(s))
+	}
+
+	var unit C.CXTranslationUnit
+	errCode := C.clang_parseTranslationUnit2(
+		index,
+		filenameC,
+		(**C.char)(cmdargsC), C.int(len(cflags)), // command line args
+		&unsavedFile, 1, // unsaved files
+		C.CXTranslationUnit_DetailedPreprocessingRecord,
+		&unit)
+	if errCode != 0 {
+		panic("loader: failed to parse source with libclang")
+	}
+	defer C.clang_disposeTranslationUnit(unit)
+
+	if numDiagnostics := int(C.clang_getNumDiagnostics(unit)); numDiagnostics != 0 {
+		addDiagnostic := func(diagnostic C.CXDiagnostic) {
+			spelling := getString(C.clang_getDiagnosticSpelling(diagnostic))
+			severity := diagnosticSeverity[C.clang_getDiagnosticSeverity(diagnostic)]
+			location := C.clang_getDiagnosticLocation(diagnostic)
+			var libclangFilename C.CXString
+			var line C.unsigned
+			var column C.unsigned
+			C.clang_getPresumedLocation(location, &libclangFilename, &line, &column)
+			filename := getString(libclangFilename)
+			if filepath.IsAbs(filename) {
+				// Relative paths for readability, like other Go parser errors.
+				relpath, err := filepath.Rel(p.dir, filename)
+				if err == nil {
+					filename = relpath
+				}
+			}
+			p.errors = append(p.errors, &scanner.Error{
+				Pos: token.Position{
+					Filename: filename,
+					Offset:   0, // not provided by clang_getPresumedLocation
+					Line:     int(line),
+					Column:   int(column),
+				},
+				Msg: severity + ": " + spelling,
+			})
+		}
+		for i := 0; i < numDiagnostics; i++ {
+			diagnostic := C.clang_getDiagnostic(unit, C.uint(i))
+			addDiagnostic(diagnostic)
+
+			// Child diagnostics (like notes on redefinitions).
+			diagnostics := C.clang_getChildDiagnostics(diagnostic)
+			for j := 0; j < int(C.clang_getNumDiagnosticsInSet(diagnostics)); j++ {
+				addDiagnostic(C.clang_getDiagnosticInSet(diagnostics, C.uint(j)))
+			}
+		}
+		return
+	}
+
+	ref := storedRefs.Put(p)
+	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))
+}
+
+//export tinygo_clang_globals_visitor
+func tinygo_clang_globals_visitor(c, parent C.GoCXCursor, client_data C.CXClientData) C.int {
+	p := storedRefs.Get(unsafe.Pointer(client_data)).(*cgoPackage)
+	kind := C.tinygo_clang_getCursorKind(c)
+	pos := p.getCursorPosition(c)
+	switch kind {
+	case C.CXCursor_FunctionDecl:
+		name := getString(C.tinygo_clang_getCursorSpelling(c))
+		if _, required := p.missingSymbols[name]; !required {
+			return C.CXChildVisit_Continue
+		}
+		cursorType := C.tinygo_clang_getCursorType(c)
+		if C.clang_isFunctionTypeVariadic(cursorType) != 0 {
+			return C.CXChildVisit_Continue // not supported
+		}
+		numArgs := int(C.tinygo_clang_Cursor_getNumArguments(c))
+		fn := &functionInfo{
+			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))
+			argType := C.clang_getArgType(cursorType, C.uint(i))
+			if argName == "" {
+				argName = "$" + strconv.Itoa(i)
+			}
+			fn.args = append(fn.args, paramInfo{
+				name:     argName,
+				typeExpr: p.makeASTType(argType, pos),
+			})
+		}
+		resultType := C.tinygo_clang_getCursorResultType(c)
+		if resultType.kind != C.CXType_Void {
+			fn.results = &ast.FieldList{
+				List: []*ast.Field{
+					&ast.Field{
+						Type: p.makeASTType(resultType, pos),
+					},
+				},
+			}
+		}
+	case C.CXCursor_StructDecl:
+		typ := C.tinygo_clang_getCursorType(c)
+		name := getString(C.tinygo_clang_getCursorSpelling(c))
+		if _, required := p.missingSymbols["struct_"+name]; !required {
+			return C.CXChildVisit_Continue
+		}
+		p.makeASTType(typ, pos)
+	case C.CXCursor_TypedefDecl:
+		typedefType := C.tinygo_clang_getCursorType(c)
+		name := getString(C.clang_getTypedefName(typedefType))
+		if _, required := p.missingSymbols[name]; !required {
+			return C.CXChildVisit_Continue
+		}
+		p.makeASTType(typedefType, pos)
+	case C.CXCursor_VarDecl:
+		name := getString(C.tinygo_clang_getCursorSpelling(c))
+		if _, required := p.missingSymbols[name]; !required {
+			return C.CXChildVisit_Continue
+		}
+		cursorType := C.tinygo_clang_getCursorType(c)
+		p.globals[name] = globalInfo{
+			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
+}
+
+func getString(clangString C.CXString) (s string) {
+	rawString := C.clang_getCString(clangString)
+	s = C.GoString(rawString)
+	C.clang_disposeString(clangString)
+	return
+}
+
+// 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 (p *cgoPackage) getCursorPosition(cursor C.GoCXCursor) token.Pos {
+	location := C.tinygo_clang_getCursorLocation(cursor)
+	var file C.CXFile
+	var line C.unsigned
+	var column C.unsigned
+	var offset C.unsigned
+	C.clang_getExpansionLocation(location, &file, &line, &column, &offset)
+	if line == 0 || file == nil {
+		// Invalid token.
+		return token.NoPos
+	}
+	filename := getString(C.clang_getFileName(file))
+	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)
+		var size C.size_t
+		sourcePtr := C.clang_getFileContents(tu, file, &size)
+		source := ((*[1 << 28]byte)(unsafe.Pointer(sourcePtr)))[:size:size]
+		lines := []int{0}
+		for i := 0; i < len(source)-1; i++ {
+			if source[i] == '\n' {
+				lines = append(lines, i+1)
+			}
+		}
+		f := p.fset.AddFile(filename, -1, int(size))
+		f.SetLines(lines)
+		p.tokenFiles[filename] = f
+	}
+	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 (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:
+		typeName = "C.char"
+	case C.CXType_SChar:
+		typeName = "C.schar"
+	case C.CXType_UChar:
+		typeName = "C.uchar"
+	case C.CXType_Short:
+		typeName = "C.short"
+	case C.CXType_UShort:
+		typeName = "C.ushort"
+	case C.CXType_Int:
+		typeName = "C.int"
+	case C.CXType_UInt:
+		typeName = "C.uint"
+	case C.CXType_Long:
+		typeName = "C.long"
+	case C.CXType_ULong:
+		typeName = "C.ulong"
+	case C.CXType_LongLong:
+		typeName = "C.longlong"
+	case C.CXType_ULongLong:
+		typeName = "C.ulonglong"
+	case C.CXType_Bool:
+		typeName = "bool"
+	case C.CXType_Float, C.CXType_Double, C.CXType_LongDouble:
+		switch C.clang_Type_getSizeOf(typ) {
+		case 4:
+			typeName = "float32"
+		case 8:
+			typeName = "float64"
+		default:
+			// Don't do anything, rely on the fallback code to show a somewhat
+			// sensible error message like "undeclared name: C.long double".
+		}
+	case C.CXType_Complex:
+		switch C.clang_Type_getSizeOf(typ) {
+		case 8:
+			typeName = "complex64"
+		case 16:
+			typeName = "complex128"
+		}
+	case C.CXType_Pointer:
+		pointeeType := C.clang_getPointeeType(typ)
+		if pointeeType.kind == C.CXType_Void {
+			// void* type is translated to Go as unsafe.Pointer
+			return &ast.SelectorExpr{
+				X: &ast.Ident{
+					NamePos: pos,
+					Name:    "unsafe",
+				},
+				Sel: &ast.Ident{
+					NamePos: pos,
+					Name:    "Pointer",
+				},
+			}
+		}
+		return &ast.StarExpr{
+			Star: pos,
+			X:    p.makeASTType(pointeeType, pos),
+		}
+	case C.CXType_ConstantArray:
+		return &ast.ArrayType{
+			Lbrack: pos,
+			Len: &ast.BasicLit{
+				ValuePos: pos,
+				Kind:     token.INT,
+				Value:    strconv.FormatInt(int64(C.clang_getArraySize(typ)), 10),
+			},
+			Elt: p.makeASTType(C.clang_getElementType(typ), pos),
+		}
+	case C.CXType_FunctionProto:
+		// Be compatible with gc, which uses the *[0]byte type for function
+		// pointer types.
+		// Return type [0]byte because this is a function type, not a pointer to
+		// this function type.
+		return &ast.ArrayType{
+			Lbrack: pos,
+			Len: &ast.BasicLit{
+				ValuePos: pos,
+				Kind:     token.INT,
+				Value:    "0",
+			},
+			Elt: &ast.Ident{
+				NamePos: pos,
+				Name:    "byte",
+			},
+		}
+	case C.CXType_Typedef:
+		name := getString(C.clang_getTypedefName(typ))
+		if _, ok := p.typedefs[name]; !ok {
+			p.typedefs[name] = nil // don't recurse
+			c := C.tinygo_clang_getTypeDeclaration(typ)
+			underlyingType := C.tinygo_clang_getTypedefDeclUnderlyingType(c)
+			expr := p.makeASTType(underlyingType, pos)
+			if strings.HasPrefix(name, "_Cgo_") {
+				expr := expr.(*ast.Ident)
+				typeSize := C.clang_Type_getSizeOf(underlyingType)
+				switch expr.Name {
+				case "C.char":
+					if typeSize != 1 {
+						// This happens for some very special purpose architectures
+						// (DSPs etc.) that are not currently targeted.
+						// https://www.embecosm.com/2017/04/18/non-8-bit-char-support-in-clang-and-llvm/
+						panic("unknown char width")
+					}
+					switch underlyingType.kind {
+					case C.CXType_Char_S:
+						expr.Name = "int8"
+					case C.CXType_Char_U:
+						expr.Name = "uint8"
+					}
+				case "C.schar", "C.short", "C.int", "C.long", "C.longlong":
+					switch typeSize {
+					case 1:
+						expr.Name = "int8"
+					case 2:
+						expr.Name = "int16"
+					case 4:
+						expr.Name = "int32"
+					case 8:
+						expr.Name = "int64"
+					}
+				case "C.uchar", "C.ushort", "C.uint", "C.ulong", "C.ulonglong":
+					switch typeSize {
+					case 1:
+						expr.Name = "uint8"
+					case 2:
+						expr.Name = "uint16"
+					case 4:
+						expr.Name = "uint32"
+					case 8:
+						expr.Name = "uint64"
+					}
+				}
+			}
+			p.typedefs[name] = &typedefInfo{
+				typeExpr: expr,
+				pos:      pos,
+			}
+		}
+		return &ast.Ident{
+			NamePos: pos,
+			Name:    "C." + name,
+		}
+	case C.CXType_Elaborated:
+		underlying := C.clang_Type_getNamedType(typ)
+		switch underlying.kind {
+		case C.CXType_Record:
+			return p.makeASTType(underlying, pos)
+		default:
+			panic("unknown elaborated type")
+		}
+	case C.CXType_Record:
+		cursor := C.tinygo_clang_getTypeDeclaration(typ)
+		name := getString(C.tinygo_clang_getCursorSpelling(cursor))
+		var cgoName string
+		switch C.tinygo_clang_getCursorKind(cursor) {
+		case C.CXCursor_StructDecl:
+			cgoName = "struct_" + name
+		case C.CXCursor_UnionDecl:
+			cgoName = "union_" + name
+		default:
+			panic("unknown record declaration")
+		}
+		if _, ok := p.elaboratedTypes[cgoName]; !ok {
+			p.elaboratedTypes[cgoName] = nil // predeclare (to avoid endless recursion)
+			fieldList := &ast.FieldList{
+				Opening: pos,
+				Closing: pos,
+			}
+			ref := storedRefs.Put(struct {
+				fieldList *ast.FieldList
+				pkg       *cgoPackage
+			}{fieldList, p})
+			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:
+				p.elaboratedTypes[cgoName] = &elaboratedTypeInfo{
+					typeExpr: &ast.StructType{
+						Struct: pos,
+						Fields: fieldList,
+					},
+					pos: pos,
+				}
+			case C.CXCursor_UnionDecl:
+				if len(fieldList.List) > 1 {
+					// Insert a special field at the front (of zero width) as a
+					// marker that this is struct is actually a union. This is done
+					// by giving the field a name that cannot be expressed directly
+					// in Go.
+					// Other parts of the compiler look at the first element in a
+					// struct (of size > 2) to know whether this is a union.
+					// Note that we don't have to insert it for single-element
+					// unions as they're basically equivalent to a struct.
+					unionMarker := &ast.Field{
+						Type: &ast.StructType{
+							Struct: pos,
+						},
+					}
+					unionMarker.Names = []*ast.Ident{
+						&ast.Ident{
+							NamePos: pos,
+							Name:    "C union",
+							Obj: &ast.Object{
+								Kind: ast.Var,
+								Name: "C union",
+								Decl: unionMarker,
+							},
+						},
+					}
+					fieldList.List = append([]*ast.Field{unionMarker}, fieldList.List...)
+				}
+				p.elaboratedTypes[cgoName] = &elaboratedTypeInfo{
+					typeExpr: &ast.StructType{
+						Struct: pos,
+						Fields: fieldList,
+					},
+					pos: pos,
+				}
+			default:
+				panic("unreachable")
+			}
+		}
+		return &ast.Ident{
+			NamePos: pos,
+			Name:    "C." + cgoName,
+		}
+	}
+	if typeName == "" {
+		// Fallback, probably incorrect but at least the error points to an odd
+		// type name.
+		typeName = "C." + getString(C.clang_getTypeSpelling(typ))
+	}
+	return &ast.Ident{
+		NamePos: pos,
+		Name:    typeName,
+	}
+}
+
+//export tinygo_clang_struct_visitor
+func tinygo_clang_struct_visitor(c, parent C.GoCXCursor, client_data C.CXClientData) C.int {
+	passed := storedRefs.Get(unsafe.Pointer(client_data)).(struct {
+		fieldList *ast.FieldList
+		pkg       *cgoPackage
+	})
+	fieldList := passed.fieldList
+	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: p.makeASTType(typ, p.getCursorPosition(c)),
+	}
+	field.Names = []*ast.Ident{
+		&ast.Ident{
+			NamePos: p.getCursorPosition(c),
+			Name:    name,
+			Obj: &ast.Object{
+				Kind: ast.Var,
+				Name: name,
+				Decl: field,
+			},
+		},
+	}
+	fieldList.List = append(fieldList.List, field)
+	return C.CXChildVisit_Continue
+}

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

@@ -44,3 +44,15 @@ int tinygo_clang_Cursor_getNumArguments(CXCursor c) {
 CXCursor tinygo_clang_Cursor_getArgument(CXCursor c, unsigned i) {
 	return clang_Cursor_getArgument(c, i);
 }
+
+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)

commands.go

@@ -1,10 +1,33 @@
-// +build !darwin
-
 package main
 
-// commands used by the compilation process might have different file names on Linux than those used on macOS.
+import (
-var commands = map[string]string{
+	"errors"
-	"clang":   "clang-8",
+	"os"
-	"ld.lld":  "ld.lld-8",
+	"os/exec"
-	"wasm-ld": "wasm-ld-8",
+	"strings"
+)
+
+// Commands used by the compilation process might have different file names
+// across operating systems and distributions.
+var commands = map[string][]string{
+	"clang":   {"clang-8"},
+	"ld.lld":  {"ld.lld-8", "ld.lld"},
+	"wasm-ld": {"wasm-ld-8", "wasm-ld"},
+}
+
+func execCommand(cmdNames []string, args ...string) error {
+	for _, cmdName := range cmdNames {
+		cmd := exec.Command(cmdName, args...)
+		cmd.Stdout = os.Stdout
+		cmd.Stderr = os.Stderr
+		err := cmd.Run()
+		if err != nil {
+			if err, ok := err.(*exec.Error); ok && err.Err == exec.ErrNotFound {
+				// this command was not found, try the next
+				continue
+			}
+		}
+		return nil
+	}
+	return errors.New("none of these commands were found in your $PATH: " + strings.Join(cmdNames, " "))
 }

commands_macos.go

@@ -1,10 +0,0 @@
-// +build darwin
-
-package main
-
-// commands used by the compilation process might have different file names on macOS than those used on Linux.
-var commands = map[string]string{
-	"clang":   "clang-8",
-	"ld.lld":  "ld.lld",
-	"wasm-ld": "wasm-ld",
-}

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 {
@@ -123,10 +163,7 @@ func (c *Compiler) collapseFormalParamInternal(t llvm.Type, fields []llvm.Value)
 	switch t.TypeKind() {
 	case llvm.StructTypeKind:
 		if len(c.flattenAggregateType(t)) <= MaxFieldsPerParam {
-			value, err := c.getZeroValue(t)
+			value := c.getZeroValue(t)
-			if err != nil {
-				panic("could not get zero value of struct: " + err.Error())
-			}
 			for i, subtyp := range t.StructElementTypes() {
 				structField, remaining := c.collapseFormalParamInternal(subtyp, fields)
 				fields = remaining

compiler/channel.go

@@ -12,16 +12,6 @@ import (
 
 // emitMakeChan returns a new channel value for the given channel type.
 func (c *Compiler) emitMakeChan(expr *ssa.MakeChan) (llvm.Value, error) {
-	valueType, err := c.getLLVMType(expr.Type().(*types.Chan).Elem())
-	if err != nil {
-		return llvm.Value{}, err
-	}
-	if c.targetData.TypeAllocSize(valueType) > c.targetData.TypeAllocSize(c.intType) {
-		// Values bigger than int overflow the data part of the coroutine.
-		// TODO: make the coroutine data part big enough to hold these bigger
-		// values.
-		return llvm.Value{}, c.makeError(expr.Pos(), "todo: channel with values bigger than int")
-	}
 	chanType := c.mod.GetTypeByName("runtime.channel")
 	size := c.targetData.TypeAllocSize(chanType)
 	sizeValue := llvm.ConstInt(c.uintptrType, size, false)
@@ -32,66 +22,64 @@ func (c *Compiler) emitMakeChan(expr *ssa.MakeChan) (llvm.Value, error) {
 
 // emitChanSend emits a pseudo chan send operation. It is lowered to the actual
 // channel send operation during goroutine lowering.
-func (c *Compiler) emitChanSend(frame *Frame, instr *ssa.Send) error {
+func (c *Compiler) emitChanSend(frame *Frame, instr *ssa.Send) {
-	valueType, err := c.getLLVMType(instr.Chan.Type().(*types.Chan).Elem())
+	valueType := c.getLLVMType(instr.X.Type())
-	if err != nil {
+	ch := c.getValue(frame, instr.Chan)
-		return err
+	chanValue := c.getValue(frame, instr.X)
-	}
-	ch, err := c.parseExpr(frame, instr.Chan)
-	if err != nil {
-		return err
-	}
-	chanValue, err := c.parseExpr(frame, instr.X)
-	if err != nil {
-		return err
-	}
 	valueSize := llvm.ConstInt(c.uintptrType, c.targetData.TypeAllocSize(chanValue.Type()), false)
+	coroutine := c.createRuntimeCall("getCoroutine", nil, "")
+
+	// store value-to-send
+	c.builder.SetInsertPointBefore(coroutine.InstructionParent().Parent().EntryBasicBlock().FirstInstruction())
 	valueAlloca := c.builder.CreateAlloca(valueType, "chan.value")
+	c.builder.SetInsertPointBefore(coroutine)
+	c.builder.SetInsertPointAtEnd(coroutine.InstructionParent())
 	c.builder.CreateStore(chanValue, valueAlloca)
 	valueAllocaCast := c.builder.CreateBitCast(valueAlloca, c.i8ptrType, "chan.value.i8ptr")
-	c.createRuntimeCall("chanSendStub", []llvm.Value{llvm.Undef(c.i8ptrType), ch, valueAllocaCast, valueSize}, "")
+
-	return nil
+	// Do the send.
+	c.createRuntimeCall("chanSend", []llvm.Value{coroutine, ch, valueAllocaCast, valueSize}, "")
+
+	// Make sure CoroSplit includes the alloca in the coroutine frame.
+	// This is a bit dirty, but it works (at least in LLVM 8).
+	valueSizeI64 := llvm.ConstInt(c.ctx.Int64Type(), c.targetData.TypeAllocSize(chanValue.Type()), false)
+	c.builder.CreateCall(c.getLifetimeEndFunc(), []llvm.Value{valueSizeI64, valueAllocaCast}, "")
 }
 
 // emitChanRecv emits a pseudo chan receive operation. It is lowered to the
 // actual channel receive operation during goroutine lowering.
-func (c *Compiler) emitChanRecv(frame *Frame, unop *ssa.UnOp) (llvm.Value, error) {
+func (c *Compiler) emitChanRecv(frame *Frame, unop *ssa.UnOp) llvm.Value {
-	valueType, err := c.getLLVMType(unop.X.Type().(*types.Chan).Elem())
+	valueType := c.getLLVMType(unop.X.Type().(*types.Chan).Elem())
-	if err != nil {
-		return llvm.Value{}, err
-	}
 	valueSize := llvm.ConstInt(c.uintptrType, c.targetData.TypeAllocSize(valueType), false)
-	ch, err := c.parseExpr(frame, unop.X)
+	ch := c.getValue(frame, unop.X)
-	if err != nil {
+	coroutine := c.createRuntimeCall("getCoroutine", nil, "")
-		return llvm.Value{}, err
+
-	}
+	// Allocate memory to receive into.
+	c.builder.SetInsertPointBefore(coroutine.InstructionParent().Parent().EntryBasicBlock().FirstInstruction())
 	valueAlloca := c.builder.CreateAlloca(valueType, "chan.value")
+	c.builder.SetInsertPointBefore(coroutine)
+	c.builder.SetInsertPointAtEnd(coroutine.InstructionParent())
 	valueAllocaCast := c.builder.CreateBitCast(valueAlloca, c.i8ptrType, "chan.value.i8ptr")
-	valueOk := c.builder.CreateAlloca(c.ctx.Int1Type(), "chan.comma-ok.alloca")
+
-	c.createRuntimeCall("chanRecvStub", []llvm.Value{llvm.Undef(c.i8ptrType), ch, valueAllocaCast, valueOk, valueSize}, "")
+	// Do the receive.
+	c.createRuntimeCall("chanRecv", []llvm.Value{coroutine, ch, valueAllocaCast, valueSize}, "")
 	received := c.builder.CreateLoad(valueAlloca, "chan.received")
 	if unop.CommaOk {
-		commaOk := c.builder.CreateLoad(valueOk, "chan.comma-ok")
+		commaOk := c.createRuntimeCall("getTaskPromiseData", []llvm.Value{coroutine}, "chan.commaOk.wide")
+		commaOk = c.builder.CreateTrunc(commaOk, c.ctx.Int1Type(), "chan.commaOk")
 		tuple := llvm.Undef(c.ctx.StructType([]llvm.Type{valueType, c.ctx.Int1Type()}, false))
 		tuple = c.builder.CreateInsertValue(tuple, received, 0, "")
 		tuple = c.builder.CreateInsertValue(tuple, commaOk, 1, "")
-		return tuple, nil
+		return tuple
 	} else {
-		return received, nil
+		return received
 	}
 }
 
 // emitChanClose closes the given channel.
-func (c *Compiler) emitChanClose(frame *Frame, param ssa.Value) error {
+func (c *Compiler) emitChanClose(frame *Frame, param ssa.Value) {
-	valueType, err := c.getLLVMType(param.Type().(*types.Chan).Elem())
+	valueType := c.getLLVMType(param.Type().(*types.Chan).Elem())
 	valueSize := llvm.ConstInt(c.uintptrType, c.targetData.TypeAllocSize(valueType), false)
-	if err != nil {
+	ch := c.getValue(frame, param)
-		return err
-	}
-	ch, err := c.parseExpr(frame, param)
-	if err != nil {
-		return err
-	}
 	c.createRuntimeCall("chanClose", []llvm.Value{ch, valueSize}, "")
-	return nil
 }

compiler/defer.go

@@ -36,7 +36,7 @@ func (c *Compiler) deferInitFunc(frame *Frame) {
 
 // emitDefer emits a single defer instruction, to be run when this function
 // returns.
-func (c *Compiler) emitDefer(frame *Frame, instr *ssa.Defer) error {
+func (c *Compiler) emitDefer(frame *Frame, instr *ssa.Defer) {
 	// The pointer to the previous defer struct, which we will replace to
 	// make a linked list.
 	next := c.builder.CreateLoad(frame.deferPtr, "defer.next")
@@ -56,18 +56,12 @@ func (c *Compiler) emitDefer(frame *Frame, instr *ssa.Defer) error {
 
 		// Collect all values to be put in the struct (starting with
 		// runtime._defer fields, followed by the call parameters).
-		itf, err := c.parseExpr(frame, instr.Call.Value) // interface
+		itf := c.getValue(frame, instr.Call.Value) // interface
-		if err != nil {
-			return err
-		}
 		receiverValue := c.builder.CreateExtractValue(itf, 1, "invoke.func.receiver")
 		values = []llvm.Value{callback, next, receiverValue}
 		valueTypes = append(valueTypes, c.i8ptrType)
 		for _, arg := range instr.Call.Args {
-			val, err := c.parseExpr(frame, arg)
+			val := c.getValue(frame, arg)
-			if err != nil {
-				return err
-			}
 			values = append(values, val)
 			valueTypes = append(valueTypes, val.Type())
 		}
@@ -86,10 +80,7 @@ func (c *Compiler) emitDefer(frame *Frame, instr *ssa.Defer) error {
 		// runtime._defer fields).
 		values = []llvm.Value{callback, next}
 		for _, param := range instr.Call.Args {
-			llvmParam, err := c.parseExpr(frame, param)
+			llvmParam := c.getValue(frame, param)
-			if err != nil {
-				return err
-			}
 			values = append(values, llvmParam)
 			valueTypes = append(valueTypes, llvmParam.Type())
 		}
@@ -101,10 +92,7 @@ func (c *Compiler) emitDefer(frame *Frame, instr *ssa.Defer) error {
 		// pointer.
 		// TODO: ignore this closure entirely and put pointers to the free
 		// variables directly in the defer struct, avoiding a memory allocation.
-		closure, err := c.parseExpr(frame, instr.Call.Value)
+		closure := c.getValue(frame, instr.Call.Value)
-		if err != nil {
-			return err
-		}
 		context := c.builder.CreateExtractValue(closure, 0, "")
 
 		// Get the callback number.
@@ -120,10 +108,7 @@ func (c *Compiler) emitDefer(frame *Frame, instr *ssa.Defer) error {
 		// context pointer).
 		values = []llvm.Value{callback, next}
 		for _, param := range instr.Call.Args {
-			llvmParam, err := c.parseExpr(frame, param)
+			llvmParam := c.getValue(frame, param)
-			if err != nil {
-				return err
-			}
 			values = append(values, llvmParam)
 			valueTypes = append(valueTypes, llvmParam.Type())
 		}
@@ -131,15 +116,13 @@ func (c *Compiler) emitDefer(frame *Frame, instr *ssa.Defer) error {
 		valueTypes = append(valueTypes, context.Type())
 
 	} else {
-		return c.makeError(instr.Pos(), "todo: defer on uncommon function call type")
+		c.addError(instr.Pos(), "todo: defer on uncommon function call type")
+		return
 	}
 
 	// Make a struct out of the collected values to put in the defer frame.
 	deferFrameType := c.ctx.StructType(valueTypes, false)
-	deferFrame, err := c.getZeroValue(deferFrameType)
+	deferFrame := c.getZeroValue(deferFrameType)
-	if err != nil {
-		return err
-	}
 	for i, value := range values {
 		deferFrame = c.builder.CreateInsertValue(deferFrame, value, i, "")
 	}
@@ -151,11 +134,10 @@ func (c *Compiler) emitDefer(frame *Frame, instr *ssa.Defer) error {
 	// Push it on top of the linked list by replacing deferPtr.
 	allocaCast := c.builder.CreateBitCast(alloca, next.Type(), "defer.alloca.cast")
 	c.builder.CreateStore(allocaCast, frame.deferPtr)
-	return nil
 }
 
 // emitRunDefers emits code to run all deferred functions.
-func (c *Compiler) emitRunDefers(frame *Frame) error {
+func (c *Compiler) emitRunDefers(frame *Frame) {
 	// Add a loop like the following:
 	//     for stack != nil {
 	//         _stack := stack
@@ -190,13 +172,13 @@ func (c *Compiler) emitRunDefers(frame *Frame) error {
 	//     stack = stack.next
 	//     switch stack.callback {
 	c.builder.SetInsertPointAtEnd(loop)
-	nextStackGEP := c.builder.CreateGEP(deferData, []llvm.Value{
+	nextStackGEP := c.builder.CreateInBoundsGEP(deferData, []llvm.Value{
 		llvm.ConstInt(c.ctx.Int32Type(), 0, false),
 		llvm.ConstInt(c.ctx.Int32Type(), 1, false), // .next field
 	}, "stack.next.gep")
 	nextStack := c.builder.CreateLoad(nextStackGEP, "stack.next")
 	c.builder.CreateStore(nextStack, frame.deferPtr)
-	gep := c.builder.CreateGEP(deferData, []llvm.Value{
+	gep := c.builder.CreateInBoundsGEP(deferData, []llvm.Value{
 		llvm.ConstInt(c.ctx.Int32Type(), 0, false),
 		llvm.ConstInt(c.ctx.Int32Type(), 0, false), // .callback field
 	}, "callback.gep")
@@ -220,11 +202,7 @@ func (c *Compiler) emitRunDefers(frame *Frame) error {
 			// Get the real defer struct type and cast to it.
 			valueTypes := []llvm.Type{c.uintptrType, llvm.PointerType(c.mod.GetTypeByName("runtime._defer"), 0), c.i8ptrType}
 			for _, arg := range callback.Args {
-				llvmType, err := c.getLLVMType(arg.Type())
+				valueTypes = append(valueTypes, c.getLLVMType(arg.Type()))
-				if err != nil {
-					return err
-				}
-				valueTypes = append(valueTypes, llvmType)
 			}
 			deferFrameType := c.ctx.StructType(valueTypes, false)
 			deferFramePtr := c.builder.CreateBitCast(deferData, llvm.PointerType(deferFrameType, 0), "deferFrame")
@@ -233,7 +211,7 @@ func (c *Compiler) emitRunDefers(frame *Frame) error {
 			forwardParams := []llvm.Value{}
 			zero := llvm.ConstInt(c.ctx.Int32Type(), 0, false)
 			for i := 2; i < len(valueTypes); i++ {
-				gep := c.builder.CreateGEP(deferFramePtr, []llvm.Value{zero, llvm.ConstInt(c.ctx.Int32Type(), uint64(i), false)}, "gep")
+				gep := c.builder.CreateInBoundsGEP(deferFramePtr, []llvm.Value{zero, llvm.ConstInt(c.ctx.Int32Type(), uint64(i), false)}, "gep")
 				forwardParam := c.builder.CreateLoad(gep, "param")
 				forwardParams = append(forwardParams, forwardParam)
 			}
@@ -246,10 +224,7 @@ func (c *Compiler) emitRunDefers(frame *Frame) error {
 			// Parent coroutine handle.
 			forwardParams = append(forwardParams, llvm.Undef(c.i8ptrType))
 
-			fnPtr, _, err := c.getInvokeCall(frame, callback)
+			fnPtr, _ := c.getInvokeCall(frame, callback)
-			if err != nil {
-				return err
-			}
 			c.createCall(fnPtr, forwardParams, "")
 
 		case *ir.Function:
@@ -258,11 +233,7 @@ func (c *Compiler) emitRunDefers(frame *Frame) error {
 			// Get the real defer struct type and cast to it.
 			valueTypes := []llvm.Type{c.uintptrType, llvm.PointerType(c.mod.GetTypeByName("runtime._defer"), 0)}
 			for _, param := range callback.Params {
-				llvmType, err := c.getLLVMType(param.Type())
+				valueTypes = append(valueTypes, c.getLLVMType(param.Type()))
-				if err != nil {
-					return err
-				}
-				valueTypes = append(valueTypes, llvmType)
 			}
 			deferFrameType := c.ctx.StructType(valueTypes, false)
 			deferFramePtr := c.builder.CreateBitCast(deferData, llvm.PointerType(deferFrameType, 0), "deferFrame")
@@ -271,7 +242,7 @@ func (c *Compiler) emitRunDefers(frame *Frame) error {
 			forwardParams := []llvm.Value{}
 			zero := llvm.ConstInt(c.ctx.Int32Type(), 0, false)
 			for i := range callback.Params {
-				gep := c.builder.CreateGEP(deferFramePtr, []llvm.Value{zero, llvm.ConstInt(c.ctx.Int32Type(), uint64(i+2), false)}, "gep")
+				gep := c.builder.CreateInBoundsGEP(deferFramePtr, []llvm.Value{zero, llvm.ConstInt(c.ctx.Int32Type(), uint64(i+2), false)}, "gep")
 				forwardParam := c.builder.CreateLoad(gep, "param")
 				forwardParams = append(forwardParams, forwardParam)
 			}
@@ -292,11 +263,7 @@ func (c *Compiler) emitRunDefers(frame *Frame) error {
 			valueTypes := []llvm.Type{c.uintptrType, llvm.PointerType(c.mod.GetTypeByName("runtime._defer"), 0)}
 			params := fn.Signature.Params()
 			for i := 0; i < params.Len(); i++ {
-				llvmType, err := c.getLLVMType(params.At(i).Type())
+				valueTypes = append(valueTypes, c.getLLVMType(params.At(i).Type()))
-				if err != nil {
-					return err
-				}
-				valueTypes = append(valueTypes, llvmType)
 			}
 			valueTypes = append(valueTypes, c.i8ptrType) // closure
 			deferFrameType := c.ctx.StructType(valueTypes, false)
@@ -306,7 +273,7 @@ func (c *Compiler) emitRunDefers(frame *Frame) error {
 			forwardParams := []llvm.Value{}
 			zero := llvm.ConstInt(c.ctx.Int32Type(), 0, false)
 			for i := 2; i < len(valueTypes); i++ {
-				gep := c.builder.CreateGEP(deferFramePtr, []llvm.Value{zero, llvm.ConstInt(c.ctx.Int32Type(), uint64(i), false)}, "")
+				gep := c.builder.CreateInBoundsGEP(deferFramePtr, []llvm.Value{zero, llvm.ConstInt(c.ctx.Int32Type(), uint64(i), false)}, "")
 				forwardParam := c.builder.CreateLoad(gep, "param")
 				forwardParams = append(forwardParams, forwardParam)
 			}
@@ -334,5 +301,4 @@ func (c *Compiler) emitRunDefers(frame *Frame) error {
 
 	// End of loop.
 	c.builder.SetInsertPointAtEnd(end)
-	return nil
 }

compiler/errors.go

@@ -12,3 +12,7 @@ func (c *Compiler) makeError(pos token.Pos, msg string) types.Error {
 		Msg:  msg,
 	}
 }
+
+func (c *Compiler) addError(pos token.Pos, msg string) {
+	c.diagnostics = append(c.diagnostics, c.makeError(pos, msg))
+}

compiler/func.go

@@ -41,7 +41,7 @@ func (c *Compiler) funcImplementation() funcValueImplementation {
 
 // createFuncValue creates a function value from a raw function pointer with no
 // context.
-func (c *Compiler) createFuncValue(funcPtr, context llvm.Value, sig *types.Signature) (llvm.Value, error) {
+func (c *Compiler) createFuncValue(funcPtr, context llvm.Value, sig *types.Signature) llvm.Value {
 	var funcValueScalar llvm.Value
 	switch c.funcImplementation() {
 	case funcValueDoubleword:
@@ -66,14 +66,11 @@ func (c *Compiler) createFuncValue(funcPtr, context llvm.Value, sig *types.Signa
 	default:
 		panic("unimplemented func value variant")
 	}
-	funcValueType, err := c.getFuncType(sig)
+	funcValueType := c.getFuncType(sig)
-	if err != nil {
-		return llvm.Value{}, err
-	}
 	funcValue := llvm.Undef(funcValueType)
 	funcValue = c.builder.CreateInsertValue(funcValue, context, 0, "")
 	funcValue = c.builder.CreateInsertValue(funcValue, funcValueScalar, 1, "")
-	return funcValue, nil
+	return funcValue
 }
 
 // getFuncSignature returns a global for identification of a particular function
@@ -112,10 +109,7 @@ func (c *Compiler) decodeFuncValue(funcValue llvm.Value, sig *types.Signature) (
 	case funcValueDoubleword:
 		funcPtr = c.builder.CreateExtractValue(funcValue, 1, "")
 	case funcValueSwitch:
-		llvmSig, err := c.getRawFuncType(sig)
+		llvmSig := c.getRawFuncType(sig)
-		if err != nil {
-			return llvm.Value{}, llvm.Value{}, err
-		}
 		sigGlobal := c.getFuncSignature(sig)
 		funcPtr = c.createRuntimeCall("getFuncPtr", []llvm.Value{funcValue, sigGlobal}, "")
 		funcPtr = c.builder.CreateIntToPtr(funcPtr, llvmSig, "")
@@ -126,25 +120,21 @@ func (c *Compiler) decodeFuncValue(funcValue llvm.Value, sig *types.Signature) (
 }
 
 // getFuncType returns the type of a func value given a signature.
-func (c *Compiler) getFuncType(typ *types.Signature) (llvm.Type, error) {
+func (c *Compiler) getFuncType(typ *types.Signature) llvm.Type {
 	switch c.funcImplementation() {
 	case funcValueDoubleword:
-		rawPtr, err := c.getRawFuncType(typ)
+		rawPtr := c.getRawFuncType(typ)
-		if err != nil {
+		return c.ctx.StructType([]llvm.Type{c.i8ptrType, rawPtr}, false)
-			return llvm.Type{}, err
-		}
-		return c.ctx.StructType([]llvm.Type{c.i8ptrType, rawPtr}, false), nil
 	case funcValueSwitch:
-		return c.mod.GetTypeByName("runtime.funcValue"), nil
+		return c.mod.GetTypeByName("runtime.funcValue")
 	default:
 		panic("unimplemented func value variant")
 	}
 }
 
 // getRawFuncType returns a LLVM function pointer type for a given signature.
-func (c *Compiler) getRawFuncType(typ *types.Signature) (llvm.Type, error) {
+func (c *Compiler) getRawFuncType(typ *types.Signature) llvm.Type {
 	// Get the return type.
-	var err error
 	var returnType llvm.Type
 	switch typ.Results().Len() {
 	case 0:
@@ -152,21 +142,14 @@ func (c *Compiler) getRawFuncType(typ *types.Signature) (llvm.Type, error) {
 		returnType = c.ctx.VoidType()
 	case 1:
 		// Just one return value.
-		returnType, err = c.getLLVMType(typ.Results().At(0).Type())
+		returnType = c.getLLVMType(typ.Results().At(0).Type())
-		if err != nil {
-			return llvm.Type{}, err
-		}
 	default:
 		// Multiple return values. Put them together in a struct.
 		// This appears to be the common way to handle multiple return values in
 		// LLVM.
 		members := make([]llvm.Type, typ.Results().Len())
 		for i := 0; i < typ.Results().Len(); i++ {
-			returnType, err := c.getLLVMType(typ.Results().At(i).Type())
+			members[i] = c.getLLVMType(typ.Results().At(i).Type())
-			if err != nil {
-				return llvm.Type{}, err
-			}
-			members[i] = returnType
 		}
 		returnType = c.ctx.StructType(members, false)
 	}
@@ -174,10 +157,7 @@ func (c *Compiler) getRawFuncType(typ *types.Signature) (llvm.Type, error) {
 	// Get the parameter types.
 	var paramTypes []llvm.Type
 	if typ.Recv() != nil {
-		recv, err := c.getLLVMType(typ.Recv().Type())
+		recv := c.getLLVMType(typ.Recv().Type())
-		if err != nil {
-			return llvm.Type{}, err
-		}
 		if recv.StructName() == "runtime._interface" {
 			// This is a call on an interface, not a concrete type.
 			// The receiver is not an interface, but a i8* type.
@@ -186,10 +166,7 @@ func (c *Compiler) getRawFuncType(typ *types.Signature) (llvm.Type, error) {
 		paramTypes = append(paramTypes, c.expandFormalParamType(recv)...)
 	}
 	for i := 0; i < typ.Params().Len(); i++ {
-		subType, err := c.getLLVMType(typ.Params().At(i).Type())
+		subType := c.getLLVMType(typ.Params().At(i).Type())
-		if err != nil {
-			return llvm.Type{}, err
-		}
 		paramTypes = append(paramTypes, c.expandFormalParamType(subType)...)
 	}
 	// All functions take these parameters at the end.
@@ -197,7 +174,7 @@ func (c *Compiler) getRawFuncType(typ *types.Signature) (llvm.Type, error) {
 	paramTypes = append(paramTypes, c.i8ptrType) // parent coroutine
 
 	// Make a func type out of the signature.
-	return llvm.PointerType(llvm.FunctionType(returnType, paramTypes, false), c.funcPtrAddrSpace), nil
+	return llvm.PointerType(llvm.FunctionType(returnType, paramTypes, false), c.funcPtrAddrSpace)
 }
 
 // parseMakeClosure makes a function value (with context) from the given
@@ -209,57 +186,17 @@ func (c *Compiler) parseMakeClosure(frame *Frame, expr *ssa.MakeClosure) (llvm.V
 	f := c.ir.GetFunction(expr.Fn.(*ssa.Function))
 
 	// Collect all bound variables.
-	boundVars := make([]llvm.Value, 0, len(expr.Bindings))
+	boundVars := make([]llvm.Value, len(expr.Bindings))
-	boundVarTypes := make([]llvm.Type, 0, len(expr.Bindings))
+	for i, binding := range expr.Bindings {
-	for _, binding := range expr.Bindings {
 		// The context stores the bound variables.
-		llvmBoundVar, err := c.parseExpr(frame, binding)
+		llvmBoundVar := c.getValue(frame, binding)
-		if err != nil {
+		boundVars[i] = llvmBoundVar
-			return llvm.Value{}, err
-		}
-		boundVars = append(boundVars, llvmBoundVar)
-		boundVarTypes = append(boundVarTypes, llvmBoundVar.Type())
-	}
-	contextType := c.ctx.StructType(boundVarTypes, false)
-
-	// Allocate memory for the context.
-	contextAlloc := llvm.Value{}
-	contextHeapAlloc := llvm.Value{}
-	if c.targetData.TypeAllocSize(contextType) <= c.targetData.TypeAllocSize(c.i8ptrType) {
-		// Context fits in a pointer - e.g. when it is a pointer. Store it
-		// directly in the stack after a convert.
-		// Because contextType is a struct and we have to cast it to a *i8,
-		// store it in an alloca first for bitcasting (store+bitcast+load).
-		contextAlloc = c.builder.CreateAlloca(contextType, "")
-	} else {
-		// Context is bigger than a pointer, so allocate it on the heap.
-		size := c.targetData.TypeAllocSize(contextType)
-		sizeValue := llvm.ConstInt(c.uintptrType, size, false)
-		contextHeapAlloc = c.createRuntimeCall("alloc", []llvm.Value{sizeValue}, "")
-		contextAlloc = c.builder.CreateBitCast(contextHeapAlloc, llvm.PointerType(contextType, 0), "")
 	}
 
-	// Store all bound variables in the alloca or heap pointer.
+	// Store the bound variables in a single object, allocating it on the heap
-	for i, boundVar := range boundVars {
+	// if necessary.
-		indices := []llvm.Value{
+	context := c.emitPointerPack(boundVars)
-			llvm.ConstInt(c.ctx.Int32Type(), 0, false),
-			llvm.ConstInt(c.ctx.Int32Type(), uint64(i), false),
-		}
-		gep := c.builder.CreateInBoundsGEP(contextAlloc, indices, "")
-		c.builder.CreateStore(boundVar, gep)
-	}
-
-	context := llvm.Value{}
-	if c.targetData.TypeAllocSize(contextType) <= c.targetData.TypeAllocSize(c.i8ptrType) {
-		// Load value (as *i8) from the alloca.
-		contextAlloc = c.builder.CreateBitCast(contextAlloc, llvm.PointerType(c.i8ptrType, 0), "")
-		context = c.builder.CreateLoad(contextAlloc, "")
-	} else {
-		// Get the original heap allocation pointer, which already is an
-		// *i8.
-		context = contextHeapAlloc
-	}
 
 	// Create the closure.
-	return c.createFuncValue(f.LLVMFn, context, f.Signature)
+	return c.createFuncValue(f.LLVMFn, context, f.Signature), 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/goroutine-lowering.go

@@ -10,8 +10,8 @@ package compiler
 //         go foo()
 //         time.Sleep(2 * time.Second)
 //         println("some other operation")
-//         bar()
+//         i := bar()
-//         println("done")
+//         println("done", *i)
 //     }
 //
 //     func foo() {
@@ -21,9 +21,10 @@ package compiler
 //         }
 //     }
 //
-//     func bar() {
+//     func bar() *int {
 //         time.Sleep(time.Second)
 //         println("blocking operation completed)
+//         return new(int)
 //     }
 //
 // It is transformed by the IR generator in compiler.go into the following
@@ -34,8 +35,8 @@ package compiler
 //         fn()
 //         time.Sleep(2 * time.Second)
 //         println("some other operation")
-//         bar() // imagine an 'await' keyword in front of this call
+//         i := bar() // imagine an 'await' keyword in front of this call
-//         println("done")
+//         println("done", *i)
 //     }
 //
 //     func foo() {
@@ -45,9 +46,10 @@ package compiler
 //         }
 //     }
 //
-//     func bar() {
+//     func bar() *int {
 //         time.Sleep(time.Second)
 //         println("blocking operation completed)
+//         return new(int)
 //     }
 //
 // The pass in this file transforms this code even further, to the following
@@ -59,9 +61,11 @@ package compiler
 //         runtime.sleepTask(hdl, 2 * time.Second) // ask the scheduler to re-activate this coroutine at the right time
 //         llvm.suspend(hdl)                       // suspend point
 //         println("some other operation")
+//         var i *int                              // allocate space on the stack for the return value
+//         runtime.setTaskPromisePtr(hdl, &i)      // store return value alloca in our coroutine promise
 //         bar(hdl)                                // await, pass a continuation (hdl) to bar
 //         llvm.suspend(hdl)                       // suspend point, wait for the callee to re-activate
-//         println("done")
+//         println("done", *i)
 //         runtime.activateTask(parent)            // re-activate the parent (nop, there is no parent)
 //     }
 //
@@ -142,10 +146,9 @@ func (c *Compiler) LowerGoroutines() error {
 	realMain.SetLinkage(llvm.InternalLinkage)
 	c.mod.NamedFunction("runtime.alloc").SetLinkage(llvm.InternalLinkage)
 	c.mod.NamedFunction("runtime.free").SetLinkage(llvm.InternalLinkage)
-	c.mod.NamedFunction("runtime.chanSend").SetLinkage(llvm.InternalLinkage)
-	c.mod.NamedFunction("runtime.chanRecv").SetLinkage(llvm.InternalLinkage)
 	c.mod.NamedFunction("runtime.sleepTask").SetLinkage(llvm.InternalLinkage)
-	c.mod.NamedFunction("runtime.activateTask").SetLinkage(llvm.InternalLinkage)
+	c.mod.NamedFunction("runtime.setTaskPromisePtr").SetLinkage(llvm.InternalLinkage)
+	c.mod.NamedFunction("runtime.getTaskPromisePtr").SetLinkage(llvm.InternalLinkage)
 	c.mod.NamedFunction("runtime.scheduler").SetLinkage(llvm.InternalLinkage)
 
 	return nil
@@ -174,13 +177,13 @@ func (c *Compiler) markAsyncFunctions() (needsScheduler bool, err error) {
 	if !deadlockStub.IsNil() {
 		worklist = append(worklist, deadlockStub)
 	}
-	chanSendStub := c.mod.NamedFunction("runtime.chanSendStub")
+	chanSend := c.mod.NamedFunction("runtime.chanSend")
-	if !chanSendStub.IsNil() {
+	if !chanSend.IsNil() {
-		worklist = append(worklist, chanSendStub)
+		worklist = append(worklist, chanSend)
 	}
-	chanRecvStub := c.mod.NamedFunction("runtime.chanRecvStub")
+	chanRecv := c.mod.NamedFunction("runtime.chanRecv")
-	if !chanRecvStub.IsNil() {
+	if !chanRecv.IsNil() {
-		worklist = append(worklist, chanRecvStub)
+		worklist = append(worklist, chanRecv)
 	}
 
 	if len(worklist) == 0 {
@@ -278,9 +281,6 @@ func (c *Compiler) markAsyncFunctions() (needsScheduler bool, err error) {
 	coroBeginType := llvm.FunctionType(c.i8ptrType, []llvm.Type{c.ctx.TokenType(), c.i8ptrType}, false)
 	coroBeginFunc := llvm.AddFunction(c.mod, "llvm.coro.begin", coroBeginType)
 
-	coroPromiseType := llvm.FunctionType(c.i8ptrType, []llvm.Type{c.i8ptrType, c.ctx.Int32Type(), c.ctx.Int1Type()}, false)
-	coroPromiseFunc := llvm.AddFunction(c.mod, "llvm.coro.promise", coroPromiseType)
-
 	coroSuspendType := llvm.FunctionType(c.ctx.Int8Type(), []llvm.Type{c.ctx.TokenType(), c.ctx.Int1Type()}, false)
 	coroSuspendFunc := llvm.AddFunction(c.mod, "llvm.coro.suspend", coroSuspendType)
 
@@ -292,7 +292,7 @@ func (c *Compiler) markAsyncFunctions() (needsScheduler bool, err error) {
 
 	// Transform all async functions into coroutines.
 	for _, f := range asyncList {
-		if f == sleep || f == deadlockStub || f == chanSendStub || f == chanRecvStub {
+		if f == sleep || f == deadlockStub || f == chanSend || f == chanRecv {
 			continue
 		}
 
@@ -309,7 +309,7 @@ func (c *Compiler) markAsyncFunctions() (needsScheduler bool, err error) {
 			for inst := bb.FirstInstruction(); !inst.IsNil(); inst = llvm.NextInstruction(inst) {
 				if !inst.IsACallInst().IsNil() {
 					callee := inst.CalledValue()
-					if _, ok := asyncFuncs[callee]; !ok || callee == sleep || callee == deadlockStub || callee == chanSendStub || callee == chanRecvStub {
+					if _, ok := asyncFuncs[callee]; !ok || callee == sleep || callee == deadlockStub || callee == chanSend || callee == chanRecv {
 						continue
 					}
 					asyncCalls = append(asyncCalls, inst)
@@ -347,10 +347,18 @@ func (c *Compiler) markAsyncFunctions() (needsScheduler bool, err error) {
 			// Split this basic block.
 			await := c.splitBasicBlock(inst, llvm.NextBasicBlock(c.builder.GetInsertBlock()), "task.await")
 
-			// Set task state to TASK_STATE_CALL.
+			// Allocate space for the return value.
-			c.builder.SetInsertPointAtEnd(inst.InstructionParent())
+			var retvalAlloca llvm.Value
+			if inst.Type().TypeKind() != llvm.VoidTypeKind {
+				c.builder.SetInsertPointBefore(inst.InstructionParent().Parent().EntryBasicBlock().FirstInstruction())
+				retvalAlloca = c.builder.CreateAlloca(inst.Type(), "coro.retvalAlloca")
+				c.builder.SetInsertPointBefore(inst)
+				data := c.builder.CreateBitCast(retvalAlloca, c.i8ptrType, "")
+				c.createRuntimeCall("setTaskPromisePtr", []llvm.Value{frame.taskHandle, data}, "")
+			}
 
 			// Suspend.
+			c.builder.SetInsertPointAtEnd(inst.InstructionParent())
 			continuePoint := c.builder.CreateCall(coroSuspendFunc, []llvm.Value{
 				llvm.ConstNull(c.ctx.TokenType()),
 				llvm.ConstInt(c.ctx.Int1Type(), 0, false),
@@ -358,27 +366,49 @@ func (c *Compiler) markAsyncFunctions() (needsScheduler bool, err error) {
 			sw := c.builder.CreateSwitch(continuePoint, frame.suspendBlock, 2)
 			sw.AddCase(llvm.ConstInt(c.ctx.Int8Type(), 0, false), await)
 			sw.AddCase(llvm.ConstInt(c.ctx.Int8Type(), 1, false), frame.cleanupBlock)
+
+			if inst.Type().TypeKind() != llvm.VoidTypeKind {
+				// Load the return value from the alloca. The callee has
+				// written the return value to it.
+				c.builder.SetInsertPointBefore(await.FirstInstruction())
+				retval := c.builder.CreateLoad(retvalAlloca, "coro.retval")
+				inst.ReplaceAllUsesWith(retval)
+			}
 		}
 
 		// Replace return instructions with suspend points that should
 		// reactivate the parent coroutine.
 		for _, inst := range returns {
-			if inst.OperandsCount() == 0 {
+			// These properties were added by the functionattrs pass. Remove
-				// These properties were added by the functionattrs pass.
+			// them, because now we start using the parameter.
-				// Remove them, because now we start using the parameter.
 			// https://llvm.org/docs/Passes.html#functionattrs-deduce-function-attributes
 			for _, kind := range []string{"nocapture", "readnone"} {
 				kindID := llvm.AttributeKindID(kind)
 				f.RemoveEnumAttributeAtIndex(f.ParamsCount(), kindID)
 			}
 
-				// Reactivate the parent coroutine. This adds it back to
-				// the run queue, so it is started again by the
-				// scheduler when possible (possibly right after the
-				// following suspend).
 			c.builder.SetInsertPointBefore(inst)
 
 			parentHandle := f.LastParam()
+
+			// Store return values.
+			switch inst.OperandsCount() {
+			case 0:
+				// Nothing to return.
+			case 1:
+				// Return this value by writing to the pointer stored in the
+				// parent handle. The parent coroutine has made an alloca that
+				// we can write to to store our return value.
+				returnValuePtr := c.createRuntimeCall("getTaskPromisePtr", []llvm.Value{parentHandle}, "coro.parentData")
+				alloca := c.builder.CreateBitCast(returnValuePtr, llvm.PointerType(inst.Operand(0).Type(), 0), "coro.parentAlloca")
+				c.builder.CreateStore(inst.Operand(0), alloca)
+			default:
+				panic("unreachable")
+			}
+
+			// Reactivate the parent coroutine. This adds it back to the run
+			// queue, so it is started again by the scheduler when possible
+			// (possibly right after the following suspend).
 			c.createRuntimeCall("activateTask", []llvm.Value{parentHandle}, "")
 
 			// Suspend this coroutine.
@@ -393,9 +423,6 @@ func (c *Compiler) markAsyncFunctions() (needsScheduler bool, err error) {
 			sw.AddCase(llvm.ConstInt(c.ctx.Int8Type(), 0, false), frame.unreachableBlock)
 			sw.AddCase(llvm.ConstInt(c.ctx.Int8Type(), 1, false), frame.cleanupBlock)
 			inst.EraseFromParentAsInstruction()
-			} else {
-				panic("todo: return value from coroutine")
-			}
 		}
 
 		// Coroutine cleanup. Free resources associated with this coroutine.
@@ -420,6 +447,14 @@ func (c *Compiler) markAsyncFunctions() (needsScheduler bool, err error) {
 		c.builder.CreateUnreachable()
 	}
 
+	// Replace calls to runtime.getCoroutineCall with the coroutine of this
+	// frame.
+	for _, getCoroutineCall := range getUses(c.mod.NamedFunction("runtime.getCoroutine")) {
+		frame := asyncFuncs[getCoroutineCall.InstructionParent().Parent()]
+		getCoroutineCall.ReplaceAllUsesWith(frame.taskHandle)
+		getCoroutineCall.EraseFromParentAsInstruction()
+	}
+
 	// Transform calls to time.Sleep() into coroutine suspend points.
 	for _, sleepCall := range getUses(sleep) {
 		// sleepCall must be a call instruction.
@@ -463,37 +498,11 @@ func (c *Compiler) markAsyncFunctions() (needsScheduler bool, err error) {
 		deadlockCall.EraseFromParentAsInstruction()
 	}
 
-	// Transform calls to runtime.chanSendStub into channel send operations.
+	// Transform calls to runtime.chanSend into channel send operations.
-	for _, sendOp := range getUses(chanSendStub) {
+	for _, sendOp := range getUses(chanSend) {
 		// sendOp must be a call instruction.
 		frame := asyncFuncs[sendOp.InstructionParent().Parent()]
 
-		// Send the value over the channel, or block.
-		sendOp.SetOperand(0, frame.taskHandle)
-		sendOp.SetOperand(sendOp.OperandsCount()-1, c.mod.NamedFunction("runtime.chanSend"))
-
-		// Use taskState.data to store the value to send:
-		//     *(*valueType)(&coroutine.promise().data) = valueToSend
-		//     runtime.chanSend(coroutine, ch)
-		bitcast := sendOp.Operand(2)
-		valueAlloca := bitcast.Operand(0)
-		c.builder.SetInsertPointBefore(valueAlloca)
-		promiseType := c.mod.GetTypeByName("runtime.taskState")
-		promiseRaw := c.builder.CreateCall(coroPromiseFunc, []llvm.Value{
-			frame.taskHandle,
-			llvm.ConstInt(c.ctx.Int32Type(), uint64(c.targetData.PrefTypeAlignment(promiseType)), false),
-			llvm.ConstInt(c.ctx.Int1Type(), 0, false),
-		}, "task.promise.raw")
-		promise := c.builder.CreateBitCast(promiseRaw, llvm.PointerType(promiseType, 0), "task.promise")
-		dataPtr := c.builder.CreateGEP(promise, []llvm.Value{
-			llvm.ConstInt(c.ctx.Int32Type(), 0, false),
-			llvm.ConstInt(c.ctx.Int32Type(), 2, false),
-		}, "task.promise.data")
-		sendOp.SetOperand(2, llvm.Undef(c.i8ptrType))
-		valueAlloca.ReplaceAllUsesWith(c.builder.CreateBitCast(dataPtr, valueAlloca.Type(), ""))
-		bitcast.EraseFromParentAsInstruction()
-		valueAlloca.EraseFromParentAsInstruction()
-
 		// Yield to scheduler.
 		c.builder.SetInsertPointBefore(llvm.NextInstruction(sendOp))
 		continuePoint := c.builder.CreateCall(coroSuspendFunc, []llvm.Value{
@@ -506,21 +515,11 @@ func (c *Compiler) markAsyncFunctions() (needsScheduler bool, err error) {
 		sw.AddCase(llvm.ConstInt(c.ctx.Int8Type(), 1, false), frame.cleanupBlock)
 	}
 
-	// Transform calls to runtime.chanRecvStub into channel receive operations.
+	// Transform calls to runtime.chanRecv into channel receive operations.
-	for _, recvOp := range getUses(chanRecvStub) {
+	for _, recvOp := range getUses(chanRecv) {
 		// recvOp must be a call instruction.
 		frame := asyncFuncs[recvOp.InstructionParent().Parent()]
 
-		bitcast := recvOp.Operand(2)
-		commaOk := recvOp.Operand(3)
-		valueAlloca := bitcast.Operand(0)
-
-		// Receive the value over the channel, or block.
-		recvOp.SetOperand(0, frame.taskHandle)
-		recvOp.SetOperand(recvOp.OperandsCount()-1, c.mod.NamedFunction("runtime.chanRecv"))
-		recvOp.SetOperand(2, llvm.Undef(c.i8ptrType))
-		bitcast.EraseFromParentAsInstruction()
-
 		// Yield to scheduler.
 		c.builder.SetInsertPointBefore(llvm.NextInstruction(recvOp))
 		continuePoint := c.builder.CreateCall(coroSuspendFunc, []llvm.Value{
@@ -532,32 +531,6 @@ func (c *Compiler) markAsyncFunctions() (needsScheduler bool, err error) {
 		c.builder.SetInsertPointAtEnd(recvOp.InstructionParent())
 		sw.AddCase(llvm.ConstInt(c.ctx.Int8Type(), 0, false), wakeup)
 		sw.AddCase(llvm.ConstInt(c.ctx.Int8Type(), 1, false), frame.cleanupBlock)
-
-		// The value to receive is stored in taskState.data:
-		//     runtime.chanRecv(coroutine, ch)
-		//     promise := coroutine.promise()
-		//     valueReceived := *(*valueType)(&promise.data)
-		//     ok := promise.commaOk
-		c.builder.SetInsertPointBefore(wakeup.FirstInstruction())
-		promiseType := c.mod.GetTypeByName("runtime.taskState")
-		promiseRaw := c.builder.CreateCall(coroPromiseFunc, []llvm.Value{
-			frame.taskHandle,
-			llvm.ConstInt(c.ctx.Int32Type(), uint64(c.targetData.PrefTypeAlignment(promiseType)), false),
-			llvm.ConstInt(c.ctx.Int1Type(), 0, false),
-		}, "task.promise.raw")
-		promise := c.builder.CreateBitCast(promiseRaw, llvm.PointerType(promiseType, 0), "task.promise")
-		dataPtr := c.builder.CreateGEP(promise, []llvm.Value{
-			llvm.ConstInt(c.ctx.Int32Type(), 0, false),
-			llvm.ConstInt(c.ctx.Int32Type(), 2, false),
-		}, "task.promise.data")
-		valueAlloca.ReplaceAllUsesWith(c.builder.CreateBitCast(dataPtr, valueAlloca.Type(), ""))
-		valueAlloca.EraseFromParentAsInstruction()
-		commaOkPtr := c.builder.CreateGEP(promise, []llvm.Value{
-			llvm.ConstInt(c.ctx.Int32Type(), 0, false),
-			llvm.ConstInt(c.ctx.Int32Type(), 1, false),
-		}, "task.promise.comma-ok")
-		commaOk.ReplaceAllUsesWith(commaOkPtr)
-		recvOp.SetOperand(3, llvm.Undef(commaOk.Type()))
 	}
 
 	return true, c.lowerMakeGoroutineCalls()

compiler/inlineasm.go

@@ -68,11 +68,7 @@ func (c *Compiler) emitAsmFull(frame *Frame, instr *ssa.CallCommon) (llvm.Value,
 			}
 			key := constant.StringVal(r.Key.(*ssa.Const).Value)
 			//println("value:", r.Value.(*ssa.MakeInterface).X.String())
-			value, err := c.parseExpr(frame, r.Value.(*ssa.MakeInterface).X)
+			registers[key] = c.getValue(frame, r.Value.(*ssa.MakeInterface).X)
-			if err != nil {
-				return llvm.Value{}, err
-			}
-			registers[key] = value
 		case *ssa.Call:
 			if r.Common() == instr {
 				break
@@ -145,10 +141,7 @@ func (c *Compiler) emitSVCall(frame *Frame, args []ssa.Value) (llvm.Value, error
 		} else {
 			constraints += ",{r" + strconv.Itoa(i) + "}"
 		}
-		llvmValue, err := c.parseExpr(frame, arg)
+		llvmValue := c.getValue(frame, arg)
-		if err != nil {
-			return llvm.Value{}, err
-		}
 		llvmArgs = append(llvmArgs, llvmValue)
 		argTypes = append(argTypes, llvmValue.Type())
 	}

compiler/interface-lowering.go

@@ -527,11 +527,40 @@ func (p *lowerInterfacesPass) replaceInvokeWithCall(use llvm.Value, typ *typeInf
 	}
 	inttoptr := inttoptrs[0]
 	function := typ.getMethod(signature).function
-	if inttoptr.Type() != function.Type() {
+	if inttoptr.Type() == function.Type() {
-		p.builder.SetInsertPointBefore(use)
+		// Easy case: the types are the same. Simply replace the inttoptr
-		function = p.builder.CreateBitCast(function, inttoptr.Type(), "")
+		// result (which is directly called) with the actual function.
-	}
 		inttoptr.ReplaceAllUsesWith(function)
+	} else {
+		// Harder case: the type is not actually the same. Go through each call
+		// (of which there should be only one), extract the receiver params for
+		// this call and replace the call with a direct call to the target
+		// function.
+		for _, call := range getUses(inttoptr) {
+			if call.IsACallInst().IsNil() || call.CalledValue() != inttoptr {
+				panic("expected the inttoptr to be called as a method, this is not a method call")
+			}
+			operands := make([]llvm.Value, call.OperandsCount()-1)
+			for i := range operands {
+				operands[i] = call.Operand(i)
+			}
+			paramTypes := function.Type().ElementType().ParamTypes()
+			receiverParamTypes := paramTypes[:len(paramTypes)-(len(operands)-1)]
+			methodParamTypes := paramTypes[len(paramTypes)-(len(operands)-1):]
+			for i, methodParamType := range methodParamTypes {
+				if methodParamType != operands[i+1].Type() {
+					panic("expected method call param type and function param type to be the same")
+				}
+			}
+			p.builder.SetInsertPointBefore(call)
+			receiverParams := p.emitPointerUnpack(operands[0], receiverParamTypes)
+			result := p.builder.CreateCall(function, append(receiverParams, operands[1:]...), "")
+			if result.Type().TypeKind() != llvm.VoidTypeKind {
+				call.ReplaceAllUsesWith(result)
+			}
+			call.EraseFromParentAsInstruction()
+		}
+	}
 	inttoptr.EraseFromParentAsInstruction()
 	use.EraseFromParentAsInstruction()
 }

compiler/interface.go

@@ -22,43 +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 {
-	var itfValue llvm.Value
+	itfValue := c.emitPointerPack([]llvm.Value{val})
-	size := c.targetData.TypeAllocSize(val.Type())
-	if size > c.targetData.TypeAllocSize(c.i8ptrType) {
-		// Allocate on the heap and put a pointer in the interface.
-		// TODO: escape analysis.
-		sizeValue := llvm.ConstInt(c.uintptrType, size, false)
-		alloc := c.createRuntimeCall("alloc", []llvm.Value{sizeValue}, "makeinterface.alloc")
-		itfValueCast := c.builder.CreateBitCast(alloc, llvm.PointerType(val.Type(), 0), "makeinterface.cast.value")
-		c.builder.CreateStore(val, itfValueCast)
-		itfValue = c.builder.CreateBitCast(itfValueCast, c.i8ptrType, "makeinterface.cast.i8ptr")
-	} else if size == 0 {
-		itfValue = llvm.ConstPointerNull(c.i8ptrType)
-	} else {
-		// Directly place the value in the interface.
-		switch val.Type().TypeKind() {
-		case llvm.IntegerTypeKind:
-			itfValue = c.builder.CreateIntToPtr(val, c.i8ptrType, "makeinterface.cast.int")
-		case llvm.PointerTypeKind:
-			itfValue = c.builder.CreateBitCast(val, c.i8ptrType, "makeinterface.cast.ptr")
-		case llvm.StructTypeKind, llvm.FloatTypeKind, llvm.DoubleTypeKind:
-			// A bitcast would be useful here, but bitcast doesn't allow
-			// aggregate types. So we'll bitcast it using an alloca.
-			// Hopefully this will get optimized away.
-			mem := c.builder.CreateAlloca(c.i8ptrType, "makeinterface.cast.struct")
-			memStructPtr := c.builder.CreateBitCast(mem, llvm.PointerType(val.Type(), 0), "makeinterface.cast.struct.cast")
-			c.builder.CreateStore(val, memStructPtr)
-			itfValue = c.builder.CreateLoad(mem, "makeinterface.cast.load")
-		default:
-			return llvm.Value{}, c.makeError(pos, "todo: makeinterface: cast small type to i8*")
-		}
-	}
 	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")
@@ -71,7 +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.
@@ -185,18 +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())
@@ -209,10 +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),
@@ -225,7 +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
@@ -274,19 +238,9 @@ func (c *Compiler) getMethodSignature(method *types.Func) llvm.Value {
 //
 // Type asserts on concrete types are trivial: just compare type numbers. Type
 // asserts on interfaces are more difficult, see the comments in the function.
-func (c *Compiler) parseTypeAssert(frame *Frame, expr *ssa.TypeAssert) (llvm.Value, error) {
+func (c *Compiler) parseTypeAssert(frame *Frame, expr *ssa.TypeAssert) llvm.Value {
-	itf, err := c.parseExpr(frame, expr.X)
+	itf := c.getValue(frame, expr.X)
-	if err != nil {
+	assertedType := c.getLLVMType(expr.AssertedType)
-		return llvm.Value{}, err
-	}
-	assertedType, err := c.getLLVMType(expr.AssertedType)
-	if err != nil {
-		return llvm.Value{}, err
-	}
-	valueNil, err := c.getZeroValue(assertedType)
-	if err != nil {
-		return llvm.Value{}, err
-	}
 
 	actualTypeNum := c.builder.CreateExtractValue(itf, 0, "interface.type")
 	commaOk := llvm.Value{}
@@ -339,67 +293,35 @@ func (c *Compiler) parseTypeAssert(frame *Frame, expr *ssa.TypeAssert) (llvm.Val
 		// Type assert on concrete type. Extract the underlying type from
 		// the interface (but only after checking it matches).
 		valuePtr := c.builder.CreateExtractValue(itf, 1, "typeassert.value.ptr")
-		size := c.targetData.TypeAllocSize(assertedType)
+		valueOk = c.emitPointerUnpack(valuePtr, []llvm.Type{assertedType})[0]
-		if size > c.targetData.TypeAllocSize(c.i8ptrType) {
-			// Value was stored in an allocated buffer, load it from there.
-			valuePtrCast := c.builder.CreateBitCast(valuePtr, llvm.PointerType(assertedType, 0), "")
-			valueOk = c.builder.CreateLoad(valuePtrCast, "typeassert.value.ok")
-		} else if size == 0 {
-			valueOk, err = c.getZeroValue(assertedType)
-			if err != nil {
-				return llvm.Value{}, err
-			}
-		} else {
-			// Value was stored directly in the interface.
-			switch assertedType.TypeKind() {
-			case llvm.IntegerTypeKind:
-				valueOk = c.builder.CreatePtrToInt(valuePtr, assertedType, "typeassert.value.ok")
-			case llvm.PointerTypeKind:
-				valueOk = c.builder.CreateBitCast(valuePtr, assertedType, "typeassert.value.ok")
-			default: // struct, float, etc.
-				// A bitcast would be useful here, but bitcast doesn't allow
-				// aggregate types. So we'll bitcast it using an alloca.
-				// Hopefully this will get optimized away.
-				mem := c.builder.CreateAlloca(c.i8ptrType, "")
-				c.builder.CreateStore(valuePtr, mem)
-				memCast := c.builder.CreateBitCast(mem, llvm.PointerType(assertedType, 0), "")
-				valueOk = c.builder.CreateLoad(memCast, "typeassert.value.ok")
-			}
-		}
 	}
 	c.builder.CreateBr(nextBlock)
 
 	// Continue after the if statement.
 	c.builder.SetInsertPointAtEnd(nextBlock)
 	phi := c.builder.CreatePHI(assertedType, "typeassert.value")
-	phi.AddIncoming([]llvm.Value{valueNil, valueOk}, []llvm.BasicBlock{prevBlock, okBlock})
+	phi.AddIncoming([]llvm.Value{c.getZeroValue(assertedType), valueOk}, []llvm.BasicBlock{prevBlock, okBlock})
 
 	if expr.CommaOk {
 		tuple := c.ctx.ConstStruct([]llvm.Value{llvm.Undef(assertedType), llvm.Undef(c.ctx.Int1Type())}, false) // create empty tuple
 		tuple = c.builder.CreateInsertValue(tuple, phi, 0, "")                                                  // insert value
 		tuple = c.builder.CreateInsertValue(tuple, commaOk, 1, "")                                              // insert 'comma ok' boolean
-		return tuple, nil
+		return tuple
 	} else {
 		// This is kind of dirty as the branch above becomes mostly useless,
 		// but hopefully this gets optimized away.
 		c.createRuntimeCall("interfaceTypeAssert", []llvm.Value{commaOk}, "")
-		return phi, nil
+		return phi
 	}
 }
 
 // getInvokeCall creates and returns the function pointer and parameters of an
 // interface call. It can be used in a call or defer instruction.
-func (c *Compiler) getInvokeCall(frame *Frame, instr *ssa.CallCommon) (llvm.Value, []llvm.Value, error) {
+func (c *Compiler) getInvokeCall(frame *Frame, instr *ssa.CallCommon) (llvm.Value, []llvm.Value) {
 	// Call an interface method with dynamic dispatch.
-	itf, err := c.parseExpr(frame, instr.Value) // interface
+	itf := c.getValue(frame, instr.Value) // interface
-	if err != nil {
-		return llvm.Value{}, nil, err
-	}
 
-	llvmFnType, err := c.getRawFuncType(instr.Method.Type().(*types.Signature))
+	llvmFnType := c.getRawFuncType(instr.Method.Type().(*types.Signature))
-	if err != nil {
-		return llvm.Value{}, nil, err
-	}
 
 	typecode := c.builder.CreateExtractValue(itf, 0, "invoke.typecode")
 	values := []llvm.Value{
@@ -413,11 +335,7 @@ func (c *Compiler) getInvokeCall(frame *Frame, instr *ssa.CallCommon) (llvm.Valu
 
 	args := []llvm.Value{receiverValue}
 	for _, arg := range instr.Args {
-		val, err := c.parseExpr(frame, arg)
+		args = append(args, c.getValue(frame, arg))
-		if err != nil {
-			return llvm.Value{}, nil, err
-		}
-		args = append(args, val)
 	}
 	// Add the context parameter. An interface call never takes a context but we
 	// have to supply the parameter anyway.
@@ -425,7 +343,7 @@ func (c *Compiler) getInvokeCall(frame *Frame, instr *ssa.CallCommon) (llvm.Valu
 	// Add the parent goroutine handle.
 	args = append(args, llvm.Undef(c.i8ptrType))
 
-	return fnCast, args, nil
+	return fnCast, args
 }
 
 // interfaceInvokeWrapper keeps some state between getInterfaceInvokeWrapper and
@@ -441,19 +359,16 @@ 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.
-	receiverType, err := c.getLLVMType(f.Params[0].Type())
+	receiverType := c.getLLVMType(f.Params[0].Type())
-	if err != nil {
-		return llvm.Value{}, err
-	}
 	expandedReceiverType := c.expandFormalParamType(receiverType)
 
 	// Does this method even need any wrapping?
@@ -462,7 +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
@@ -475,12 +390,12 @@ 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,
 // see that function for details.
-func (c *Compiler) createInterfaceInvokeWrapper(state interfaceInvokeWrapper) error {
+func (c *Compiler) createInterfaceInvokeWrapper(state interfaceInvokeWrapper) {
 	wrapper := state.wrapper
 	fn := state.fn
 	receiverType := state.receiverType
@@ -490,10 +405,7 @@ func (c *Compiler) createInterfaceInvokeWrapper(state interfaceInvokeWrapper) er
 	// add debug info if needed
 	if c.Debug {
 		pos := c.ir.Program.Fset.Position(fn.Pos())
-		difunc, err := c.attachDebugInfoRaw(fn, wrapper, "$invoke", pos.Filename, pos.Line)
+		difunc := c.attachDebugInfoRaw(fn, wrapper, "$invoke", pos.Filename, pos.Line)
-		if err != nil {
-			return err
-		}
 		c.builder.SetCurrentDebugLocation(uint(pos.Line), uint(pos.Column), difunc, llvm.Metadata{})
 	}
 
@@ -501,28 +413,7 @@ func (c *Compiler) createInterfaceInvokeWrapper(state interfaceInvokeWrapper) er
 	block := c.ctx.AddBasicBlock(wrapper, "entry")
 	c.builder.SetInsertPointAtEnd(block)
 
-	var receiverPtr llvm.Value
+	receiverValue := c.emitPointerUnpack(wrapper.Param(0), []llvm.Type{receiverType})[0]
-	if c.targetData.TypeAllocSize(receiverType) > c.targetData.TypeAllocSize(c.i8ptrType) {
-		// The receiver is passed in using a pointer. We have to load it here
-		// and pass it by value to the real function.
-
-		// Load the underlying value.
-		receiverPtrType := llvm.PointerType(receiverType, 0)
-		receiverPtr = c.builder.CreateBitCast(wrapper.Param(0), receiverPtrType, "receiver.ptr")
-	} else {
-		// The value is stored in the interface, but it is of type struct which
-		// is expanded to multiple parameters (e.g. {i8, i8}). So we have to
-		// receive the struct as parameter, expand it, and pass it on to the
-		// real function.
-
-		// Cast the passed-in i8* to the struct value (using an alloca) and
-		// extract its values.
-		alloca := c.builder.CreateAlloca(c.i8ptrType, "receiver.alloca")
-		c.builder.CreateStore(wrapper.Param(0), alloca)
-		receiverPtr = c.builder.CreateBitCast(alloca, llvm.PointerType(receiverType, 0), "receiver.ptr")
-	}
-
-	receiverValue := c.builder.CreateLoad(receiverPtr, "receiver")
 	params := append(c.expandFormalParam(receiverValue), wrapper.Params()[1:]...)
 	if fn.LLVMFn.Type().ElementType().ReturnType().TypeKind() == llvm.VoidTypeKind {
 		c.builder.CreateCall(fn.LLVMFn, params, "")
@@ -531,6 +422,4 @@ func (c *Compiler) createInterfaceInvokeWrapper(state interfaceInvokeWrapper) er
 		ret := c.builder.CreateCall(fn.LLVMFn, params, "ret")
 		c.builder.CreateRet(ret)
 	}
-
-	return nil
 }

compiler/llvm.go

@@ -22,6 +22,47 @@ 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 {
+	fn := c.mod.NamedFunction("llvm.lifetime.end.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.end.p0i8", fnType)
+	}
+	return fn
+}
+
 // splitBasicBlock splits a LLVM basic block into two parts. All instructions
 // after afterInst are moved into a new basic block (created right after the
 // current one) with the given name.

compiler/map.go

@@ -10,12 +10,14 @@ import (
 )
 
 func (c *Compiler) emitMapLookup(keyType, valueType types.Type, m, key llvm.Value, commaOk bool, pos token.Pos) (llvm.Value, error) {
-	llvmValueType, err := c.getLLVMType(valueType)
+	llvmValueType := c.getLLVMType(valueType)
-	if err != nil {
+
-		return llvm.Value{}, err
+	// 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
-	mapValueAlloca := c.builder.CreateAlloca(llvmValueType, "hashmap.value")
+	// present in the map.
-	mapValuePtr := c.builder.CreateBitCast(mapValueAlloca, c.i8ptrType, "hashmap.valueptr")
+	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
@@ -23,15 +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, "")
@@ -42,27 +53,25 @@ 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) error {
+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
 		params := []llvm.Value{m, key, valuePtr}
 		c.createRuntimeCall("hashmapStringSet", params, "")
-		return nil
 	} 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, "")
-		return nil
+		c.builder.CreateCall(c.getLifetimeEndFunc(), []llvm.Value{keySize, keyPtr}, "")
 	} else {
-		return c.makeError(pos, "only strings, bools, ints or structs of bools/ints are supported as map keys, but got: "+keyType.String())
+		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 {
@@ -73,11 +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

@@ -18,6 +18,10 @@ func (c *Compiler) Optimize(optLevel, sizeLevel int, inlinerThreshold uint) erro
 	}
 	builder.AddCoroutinePassesToExtensionPoints()
 
+	if c.PanicStrategy == "trap" {
+		c.replacePanicsWithTrap() // -panic=trap
+	}
+
 	// Run function passes for each function.
 	funcPasses := llvm.NewFunctionPassManagerForModule(c.mod)
 	defer funcPasses.Dispose()
@@ -33,6 +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()
@@ -110,9 +115,35 @@ 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
 }
 
+// Replace panic calls with calls to llvm.trap, to reduce code size. This is the
+// -panic=trap intrinsic.
+func (c *Compiler) replacePanicsWithTrap() {
+	trap := c.mod.NamedFunction("llvm.trap")
+	for _, name := range []string{"runtime._panic", "runtime.runtimePanic"} {
+		fn := c.mod.NamedFunction(name)
+		if fn.IsNil() {
+			continue
+		}
+		for _, use := range getUses(fn) {
+			if use.IsACallInst().IsNil() || use.CalledValue() != fn {
+				panic("expected use of a panic function to be a call")
+			}
+			c.builder.SetInsertPointBefore(use)
+			c.builder.CreateCall(trap, nil, "")
+		}
+	}
+}
+
 // Eliminate created but not used maps.
 //
 // In the future, this should statically allocate created but never modified
@@ -249,7 +280,7 @@ func (c *Compiler) OptimizeAllocs() {
 			sizeInWords := (size + uint64(alignment) - 1) / uint64(alignment)
 			allocaType := llvm.ArrayType(c.ctx.IntType(alignment*8), int(sizeInWords))
 			alloca := c.builder.CreateAlloca(allocaType, "stackalloc.alloca")
-			zero, _ := c.getZeroValue(alloca.Type().ElementType())
+			zero := c.getZeroValue(alloca.Type().ElementType())
 			c.builder.CreateStore(zero, alloca)
 			stackalloc := c.builder.CreateBitCast(alloca, bitcast.Type(), "stackalloc")
 			bitcast.ReplaceAllUsesWith(stackalloc)
@@ -284,18 +315,6 @@ func (c *Compiler) doesEscape(value llvm.Value) bool {
 				return true
 			}
 		} else if use.IsACallInst() != nilValue {
-			// Call only escapes when the (pointer) parameter is not marked
-			// "nocapture". This flag means that the parameter does not escape
-			// the give function.
-			if use.CalledValue().IsAFunction() != nilValue {
-				if use.CalledValue().IsDeclaration() {
-					// Kind of dirty: assume external functions don't let
-					// pointers escape.
-					// TODO: introduce //go:noescape that sets the 'nocapture'
-					// flag on each input parameter.
-					continue
-				}
-			}
 			if !c.hasFlag(use, value, "nocapture") {
 				return true
 			}

compiler/sizes.go

@@ -113,6 +113,9 @@ func (s *StdSizes) Sizeof(T types.Type) int64 {
 		if k == types.Uintptr {
 			return s.PtrSize
 		}
+		if k == types.UnsafePointer {
+			return s.PtrSize
+		}
 		panic("unknown basic type: " + t.String())
 	case *types.Array:
 		n := t.Len()

compiler/syscall.go

@@ -51,10 +51,7 @@ func (c *Compiler) emitSyscall(frame *Frame, call *ssa.CallCommon) (llvm.Value,
 				"{r12}",
 				"{r13}",
 			}[i]
-			llvmValue, err := c.parseExpr(frame, arg)
+			llvmValue := c.getValue(frame, arg)
-			if err != nil {
-				return llvm.Value{}, err
-			}
 			args = append(args, llvmValue)
 			argTypes = append(argTypes, llvmValue.Type())
 		}
@@ -80,10 +77,7 @@ func (c *Compiler) emitSyscall(frame *Frame, call *ssa.CallCommon) (llvm.Value,
 				"{r5}",
 				"{r6}",
 			}[i]
-			llvmValue, err := c.parseExpr(frame, arg)
+			llvmValue := c.getValue(frame, arg)
-			if err != nil {
-				return llvm.Value{}, err
-			}
 			args = append(args, llvmValue)
 			argTypes = append(argTypes, llvmValue.Type())
 		}
@@ -113,10 +107,7 @@ func (c *Compiler) emitSyscall(frame *Frame, call *ssa.CallCommon) (llvm.Value,
 				"{x4}",
 				"{x5}",
 			}[i]
-			llvmValue, err := c.parseExpr(frame, arg)
+			llvmValue := c.getValue(frame, arg)
-			if err != nil {
-				return llvm.Value{}, err
-			}
 			args = append(args, llvmValue)
 			argTypes = append(argTypes, llvmValue.Type())
 		}

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

@@ -0,0 +1,108 @@
+package compiler
+
+// This file contains utility functions to pack and unpack sets of values. It
+// can take in a list of values and tries to store it efficiently in the pointer
+// itself if possible and legal.
+
+import (
+	"tinygo.org/x/go-llvm"
+)
+
+// emitPointerPack packs the list of values into a single pointer value using
+// bitcasts, or else allocates a value on the heap if it cannot be packed in the
+// pointer value directly. It returns the pointer with the packed data.
+func (c *Compiler) emitPointerPack(values []llvm.Value) llvm.Value {
+	valueTypes := make([]llvm.Type, len(values))
+	for i, value := range values {
+		valueTypes[i] = value.Type()
+	}
+	packedType := c.ctx.StructType(valueTypes, false)
+
+	// Allocate memory for the packed data.
+	var packedAlloc, packedHeapAlloc llvm.Value
+	size := c.targetData.TypeAllocSize(packedType)
+	if size == 0 {
+		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) && !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 {
+			// Try to keep this cast in SSA form.
+			return c.builder.CreateIntToPtr(values[0], c.i8ptrType, "pack.int")
+		}
+		// Because packedType is a struct and we have to cast it to a *i8, store
+		// it in an alloca first for bitcasting (store+bitcast+load).
+		packedAlloc = c.builder.CreateAlloca(packedType, "")
+	} else {
+		// Packed data is bigger than a pointer, so allocate it on the heap.
+		sizeValue := llvm.ConstInt(c.uintptrType, size, false)
+		packedHeapAlloc = c.createRuntimeCall("alloc", []llvm.Value{sizeValue}, "")
+		packedAlloc = c.builder.CreateBitCast(packedHeapAlloc, llvm.PointerType(packedType, 0), "")
+	}
+	// Store all values in the alloca or heap pointer.
+	for i, value := range values {
+		indices := []llvm.Value{
+			llvm.ConstInt(c.ctx.Int32Type(), 0, false),
+			llvm.ConstInt(c.ctx.Int32Type(), uint64(i), false),
+		}
+		gep := c.builder.CreateInBoundsGEP(packedAlloc, indices, "")
+		c.builder.CreateStore(value, gep)
+	}
+
+	if packedHeapAlloc.IsNil() {
+		// Load value (as *i8) from the alloca.
+		packedAlloc = c.builder.CreateBitCast(packedAlloc, llvm.PointerType(c.i8ptrType, 0), "")
+		return c.builder.CreateLoad(packedAlloc, "")
+	} else {
+		// Get the original heap allocation pointer, which already is an *i8.
+		return c.builder.CreateBitCast(packedAlloc, c.i8ptrType, "")
+	}
+}
+
+// emitPointerUnpack extracts a list of values packed using emitPointerPack.
+func (c *Compiler) emitPointerUnpack(ptr llvm.Value, valueTypes []llvm.Type) []llvm.Value {
+	packedType := c.ctx.StructType(valueTypes, false)
+
+	// Get a correctly-typed pointer to the packed data.
+	var packedAlloc llvm.Value
+	size := c.targetData.TypeAllocSize(packedType)
+	if size == 0 {
+		// No data to unpack.
+	} 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) && !c.gcIsPrecise() {
+		// Packed data stored directly in pointer.
+		if len(valueTypes) == 1 && valueTypes[0].TypeKind() == llvm.IntegerTypeKind {
+			// Keep this cast in SSA form.
+			return []llvm.Value{c.builder.CreatePtrToInt(ptr, valueTypes[0], "unpack.int")}
+		}
+		// Fallback: load it using an alloca.
+		packedRawAlloc := c.builder.CreateAlloca(llvm.PointerType(c.i8ptrType, 0), "unpack.raw.alloc")
+		packedRawValue := c.builder.CreateBitCast(ptr, llvm.PointerType(c.i8ptrType, 0), "unpack.raw.value")
+		c.builder.CreateStore(packedRawValue, packedRawAlloc)
+		packedAlloc = c.builder.CreateBitCast(packedRawAlloc, llvm.PointerType(packedType, 0), "unpack.alloc")
+	} else {
+		// Packed data stored on the heap. Bitcast the passed-in pointer to the
+		// correct pointer type.
+		packedAlloc = c.builder.CreateBitCast(ptr, llvm.PointerType(packedType, 0), "unpack.raw.ptr")
+	}
+	// Load each value from the packed data.
+	values := make([]llvm.Value, len(valueTypes))
+	for i, valueType := range valueTypes {
+		if c.targetData.TypeAllocSize(valueType) == 0 {
+			// This value has length zero, so there's nothing to load.
+			values[i] = c.getZeroValue(valueType)
+			continue
+		}
+		indices := []llvm.Value{
+			llvm.ConstInt(c.ctx.Int32Type(), 0, false),
+			llvm.ConstInt(c.ctx.Int32Type(), uint64(i), false),
+		}
+		gep := c.builder.CreateInBoundsGEP(packedAlloc, indices, "")
+		values[i] = c.builder.CreateLoad(gep, "")
+	}
+	return values
+}

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 != "" {

linker-builtin.go

@@ -24,7 +24,7 @@ import "C"
 // This version uses the built-in linker when trying to use lld.
 func Link(linker string, flags ...string) error {
 	switch linker {
-	case "ld.lld", commands["ld.lld"]:
+	case "ld.lld":
 		flags = append([]string{"tinygo:" + linker}, flags...)
 		var cflag *C.char
 		buf := C.calloc(C.size_t(len(flags)), C.size_t(unsafe.Sizeof(cflag)))
@@ -39,7 +39,7 @@ func Link(linker string, flags ...string) error {
 			return errors.New("failed to link using built-in ld.lld")
 		}
 		return nil
-	case "wasm-ld", commands["wasm-ld"]:
+	case "wasm-ld":
 		flags = append([]string{"tinygo:" + linker}, flags...)
 		var cflag *C.char
 		buf := C.calloc(C.size_t(len(flags)), C.size_t(unsafe.Sizeof(cflag)))
@@ -57,6 +57,9 @@ func Link(linker string, flags ...string) error {
 		return nil
 	default:
 		// Fall back to external command.
+		if cmdNames, ok := commands[linker]; ok {
+			return execCommand(cmdNames, flags...)
+		}
 		cmd := exec.Command(linker, flags...)
 		cmd.Stdout = os.Stdout
 		cmd.Stderr = os.Stderr

linker-external.go

@@ -14,6 +14,9 @@ import (
 //
 // This version always runs the linker as an external command.
 func Link(linker string, flags ...string) error {
+	if cmdNames, ok := commands[linker]; ok {
+		return execCommand(cmdNames, flags...)
+	}
 	cmd := exec.Command(linker, flags...)
 	cmd.Stdout = os.Stdout
 	cmd.Stderr = os.Stderr

loader/cgo.go

@@ -1,458 +0,0 @@
-package loader
-
-// 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).
-
-import (
-	"go/ast"
-	"go/token"
-	"sort"
-	"strconv"
-	"strings"
-
-	"golang.org/x/tools/go/ast/astutil"
-)
-
-// fileInfo holds all Cgo-related information of a given *ast.File.
-type fileInfo struct {
-	*ast.File
-	*Package
-	filename        string
-	functions       map[string]*functionInfo
-	globals         map[string]*globalInfo
-	typedefs        map[string]*typedefInfo
-	elaboratedTypes map[string]ast.Expr
-	importCPos      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
-}
-
-// paramInfo is a parameter of a Cgo function (see functionInfo).
-type paramInfo struct {
-	name     string
-	typeExpr ast.Expr
-}
-
-// typedefInfo contains information about a single typedef in C.
-type typedefInfo struct {
-	typeExpr ast.Expr
-}
-
-// globalInfo contains information about a declared global variable in C.
-type globalInfo struct {
-	typeExpr ast.Expr
-}
-
-// cgoAliases list type aliases between Go and C, for types that are equivalent
-// in both languages. See addTypeAliases.
-var cgoAliases = map[string]string{
-	"C.int8_t":    "int8",
-	"C.int16_t":   "int16",
-	"C.int32_t":   "int32",
-	"C.int64_t":   "int64",
-	"C.uint8_t":   "uint8",
-	"C.uint16_t":  "uint16",
-	"C.uint32_t":  "uint32",
-	"C.uint64_t":  "uint64",
-	"C.uintptr_t": "uintptr",
-}
-
-// cgoTypes lists some C types with ambiguous sizes that must be retrieved
-// somehow from C. This is done by adding some typedefs to get the size of each
-// type.
-const cgoTypes = `
-typedef signed char         _Cgo_schar;
-typedef unsigned char       _Cgo_uchar;
-typedef short               _Cgo_short;
-typedef unsigned short      _Cgo_ushort;
-typedef int                 _Cgo_int;
-typedef unsigned int        _Cgo_uint;
-typedef long                _Cgo_long;
-typedef unsigned long       _Cgo_ulong;
-typedef long long           _Cgo_longlong;
-typedef unsigned long long  _Cgo_ulonglong;
-`
-
-// processCgo extracts the `import "C"` statement from the AST, parses the
-// comment with libclang, and modifies the AST to use this information.
-func (p *Package) processCgo(filename string, f *ast.File, cflags []string) []error {
-	info := &fileInfo{
-		File:            f,
-		Package:         p,
-		filename:        filename,
-		functions:       map[string]*functionInfo{},
-		globals:         map[string]*globalInfo{},
-		typedefs:        map[string]*typedefInfo{},
-		elaboratedTypes: map[string]ast.Expr{},
-	}
-
-	// Find `import "C"` statements in the file.
-	for i := 0; i < len(f.Decls); i++ {
-		decl := f.Decls[i]
-		genDecl, ok := decl.(*ast.GenDecl)
-		if !ok {
-			continue
-		}
-		if len(genDecl.Specs) != 1 {
-			continue
-		}
-		spec, ok := genDecl.Specs[0].(*ast.ImportSpec)
-		if !ok {
-			continue
-		}
-		path, err := strconv.Unquote(spec.Path.Value)
-		if err != nil {
-			panic("could not parse import path: " + err.Error())
-		}
-		if path != "C" {
-			continue
-		}
-		cgoComment := genDecl.Doc.Text()
-
-		// Stored for later use by generated functions, to use a somewhat sane
-		// source location.
-		info.importCPos = spec.Path.ValuePos
-
-		pos := info.fset.PositionFor(genDecl.Doc.Pos(), true)
-		errs := info.parseFragment(cgoComment+cgoTypes, cflags, pos.Filename, pos.Line)
-		if errs != nil {
-			return errs
-		}
-
-		// Remove this import declaration.
-		f.Decls = append(f.Decls[:i], f.Decls[i+1:]...)
-		i--
-	}
-
-	// Print the AST, for debugging.
-	//ast.Print(p.fset, f)
-
-	// Declare functions found by libclang.
-	info.addFuncDecls()
-
-	// Declare stub function pointer values found by libclang.
-	info.addFuncPtrDecls()
-
-	// Declare globals found by libclang.
-	info.addVarDecls()
-
-	// Forward C types to Go types (like C.uint32_t -> uint32).
-	info.addTypeAliases()
-
-	// Add type declarations for C types, declared using typedef in C.
-	info.addTypedefs()
-
-	// Add elaborated types for C structs and unions.
-	info.addElaboratedTypes()
-
-	// Patch the AST to use the declared types and functions.
-	f = astutil.Apply(f, info.walker, nil).(*ast.File)
-
-	return nil
-}
-
-// addFuncDecls adds the C function declarations found by libclang in the
-// comment above the `import "C"` statement.
-func (info *fileInfo) addFuncDecls() {
-	// TODO: replace all uses of importCPos with the real locations from
-	// libclang.
-	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]
-		obj := &ast.Object{
-			Kind: ast.Fun,
-			Name: "C." + name,
-		}
-		args := make([]*ast.Field, len(fn.args))
-		decl := &ast.FuncDecl{
-			Name: &ast.Ident{
-				NamePos: info.importCPos,
-				Name:    "C." + name,
-				Obj:     obj,
-			},
-			Type: &ast.FuncType{
-				Func: info.importCPos,
-				Params: &ast.FieldList{
-					Opening: info.importCPos,
-					List:    args,
-					Closing: info.importCPos,
-				},
-				Results: fn.results,
-			},
-		}
-		obj.Decl = decl
-		for i, arg := range fn.args {
-			args[i] = &ast.Field{
-				Names: []*ast.Ident{
-					&ast.Ident{
-						NamePos: info.importCPos,
-						Name:    arg.name,
-						Obj: &ast.Object{
-							Kind: ast.Var,
-							Name: arg.name,
-							Decl: decl,
-						},
-					},
-				},
-				Type: arg.typeExpr,
-			}
-		}
-		info.Decls = append(info.Decls, decl)
-	}
-}
-
-// addFuncPtrDecls creates stub declarations of function pointer values. These
-// values will later be replaced with the real values in the compiler.
-// It adds code like the following to the AST:
-//
-//     var (
-//         C.add unsafe.Pointer
-//         C.mul unsafe.Pointer
-//         // ...
-//     )
-func (info *fileInfo) addFuncPtrDecls() {
-	gen := &ast.GenDecl{
-		TokPos: info.importCPos,
-		Tok:    token.VAR,
-		Lparen: info.importCPos,
-		Rparen: info.importCPos,
-	}
-	names := make([]string, 0, len(info.functions))
-	for name := range info.functions {
-		names = append(names, name)
-	}
-	sort.Strings(names)
-	for _, name := range names {
-		obj := &ast.Object{
-			Kind: ast.Typ,
-			Name: "C." + name + "$funcaddr",
-		}
-		valueSpec := &ast.ValueSpec{
-			Names: []*ast.Ident{&ast.Ident{
-				NamePos: info.importCPos,
-				Name:    "C." + name + "$funcaddr",
-				Obj:     obj,
-			}},
-			Type: &ast.SelectorExpr{
-				X: &ast.Ident{
-					NamePos: info.importCPos,
-					Name:    "unsafe",
-				},
-				Sel: &ast.Ident{
-					NamePos: info.importCPos,
-					Name:    "Pointer",
-				},
-			},
-		}
-		obj.Decl = valueSpec
-		gen.Specs = append(gen.Specs, valueSpec)
-	}
-	info.Decls = append(info.Decls, gen)
-}
-
-// addVarDecls declares external C globals in the Go source.
-// It adds code like the following to the AST:
-//
-//     var (
-//         C.globalInt  int
-//         C.globalBool bool
-//         // ...
-//     )
-func (info *fileInfo) addVarDecls() {
-	gen := &ast.GenDecl{
-		TokPos: info.importCPos,
-		Tok:    token.VAR,
-		Lparen: info.importCPos,
-		Rparen: info.importCPos,
-	}
-	names := make([]string, 0, len(info.globals))
-	for name := range info.globals {
-		names = append(names, name)
-	}
-	sort.Strings(names)
-	for _, name := range names {
-		global := info.globals[name]
-		obj := &ast.Object{
-			Kind: ast.Typ,
-			Name: "C." + name,
-		}
-		valueSpec := &ast.ValueSpec{
-			Names: []*ast.Ident{&ast.Ident{
-				NamePos: info.importCPos,
-				Name:    "C." + name,
-				Obj:     obj,
-			}},
-			Type: global.typeExpr,
-		}
-		obj.Decl = valueSpec
-		gen.Specs = append(gen.Specs, valueSpec)
-	}
-	info.Decls = append(info.Decls, gen)
-}
-
-// addTypeAliases aliases some built-in Go types with their equivalent C types.
-// It adds code like the following to the AST:
-//
-//     type (
-//         C.int8_t  = int8
-//         C.int16_t = int16
-//         // ...
-//     )
-func (info *fileInfo) addTypeAliases() {
-	aliasKeys := make([]string, 0, len(cgoAliases))
-	for key := range cgoAliases {
-		aliasKeys = append(aliasKeys, key)
-	}
-	sort.Strings(aliasKeys)
-	gen := &ast.GenDecl{
-		TokPos: info.importCPos,
-		Tok:    token.TYPE,
-		Lparen: info.importCPos,
-		Rparen: info.importCPos,
-	}
-	for _, typeName := range aliasKeys {
-		goTypeName := cgoAliases[typeName]
-		obj := &ast.Object{
-			Kind: ast.Typ,
-			Name: typeName,
-		}
-		typeSpec := &ast.TypeSpec{
-			Name: &ast.Ident{
-				NamePos: info.importCPos,
-				Name:    typeName,
-				Obj:     obj,
-			},
-			Assign: info.importCPos,
-			Type: &ast.Ident{
-				NamePos: info.importCPos,
-				Name:    goTypeName,
-			},
-		}
-		obj.Decl = typeSpec
-		gen.Specs = append(gen.Specs, typeSpec)
-	}
-	info.Decls = append(info.Decls, gen)
-}
-
-func (info *fileInfo) addTypedefs() {
-	gen := &ast.GenDecl{
-		TokPos: info.importCPos,
-		Tok:    token.TYPE,
-	}
-	names := make([]string, 0, len(info.typedefs))
-	for name := range info.typedefs {
-		names = append(names, name)
-	}
-	sort.Strings(names)
-	for _, name := range names {
-		typedef := info.typedefs[name]
-		typeName := "C." + name
-		if strings.HasPrefix(name, "_Cgo_") {
-			typeName = "C." + name[len("_Cgo_"):]
-		}
-		if _, ok := cgoAliases[typeName]; ok {
-			// This is a type that also exists in Go (defined in stdint.h).
-			continue
-		}
-		obj := &ast.Object{
-			Kind: ast.Typ,
-			Name: typeName,
-		}
-		typeSpec := &ast.TypeSpec{
-			Name: &ast.Ident{
-				NamePos: info.importCPos,
-				Name:    typeName,
-				Obj:     obj,
-			},
-			Type: typedef.typeExpr,
-		}
-		obj.Decl = typeSpec
-		gen.Specs = append(gen.Specs, typeSpec)
-	}
-	info.Decls = append(info.Decls, gen)
-}
-
-// addElaboratedTypes adds C elaborated types as aliases. These are the "struct
-// foo" or "union foo" types, often used in a typedef.
-//
-// See also:
-// https://en.cppreference.com/w/cpp/language/elaborated_type_specifier
-func (info *fileInfo) addElaboratedTypes() {
-	gen := &ast.GenDecl{
-		TokPos: info.importCPos,
-		Tok:    token.TYPE,
-	}
-	names := make([]string, 0, len(info.elaboratedTypes))
-	for name := range info.elaboratedTypes {
-		names = append(names, name)
-	}
-	sort.Strings(names)
-	for _, name := range names {
-		typ := info.elaboratedTypes[name]
-		typeName := "C.struct_" + name
-		obj := &ast.Object{
-			Kind: ast.Typ,
-			Name: typeName,
-		}
-		typeSpec := &ast.TypeSpec{
-			Name: &ast.Ident{
-				NamePos: info.importCPos,
-				Name:    typeName,
-				Obj:     obj,
-			},
-			Type: typ,
-		}
-		obj.Decl = typeSpec
-		gen.Specs = append(gen.Specs, typeSpec)
-	}
-	info.Decls = append(info.Decls, gen)
-}
-
-// walker replaces all "C".<something> expressions to literal "C.<something>"
-// 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 {
-	switch node := cursor.Node().(type) {
-	case *ast.CallExpr:
-		fun, ok := node.Fun.(*ast.SelectorExpr)
-		if !ok {
-			return true
-		}
-		x, ok := fun.X.(*ast.Ident)
-		if !ok {
-			return true
-		}
-		if _, ok := info.functions[fun.Sel.Name]; ok && x.Name == "C" {
-			node.Fun = &ast.Ident{
-				NamePos: x.NamePos,
-				Name:    "C." + fun.Sel.Name,
-			}
-		}
-	case *ast.SelectorExpr:
-		x, ok := node.X.(*ast.Ident)
-		if !ok {
-			return true
-		}
-		if x.Name == "C" {
-			name := "C." + node.Sel.Name
-			if _, ok := info.functions[node.Sel.Name]; ok {
-				name += "$funcaddr"
-			}
-			cursor.Replace(&ast.Ident{
-				NamePos: x.NamePos,
-				Name:    name,
-			})
-		}
-	}
-	return true
-}

loader/libclang.go

@@ -1,446 +0,0 @@
-package loader
-
-// This file parses a fragment of C with libclang and stores the result for AST
-// modification. It does not touch the AST itself.
-
-import (
-	"go/ast"
-	"go/scanner"
-	"go/token"
-	"path/filepath"
-	"strconv"
-	"strings"
-	"unsafe"
-)
-
-/*
-#include <clang-c/Index.h> // if this fails, install libclang-8-dev
-#include <stdlib.h>
-#include <stdint.h>
-
-// This struct should be ABI-compatible on all platforms (uintptr_t has the same
-// alignment etc. as void*) but does not include void* pointers that are not
-// always real pointers.
-// The Go garbage collector assumes that all non-nil pointer-typed integers are
-// actually pointers. This is not always true, as data[1] often contains 0x1,
-// which is clearly not a valid pointer. Usually the GC won't catch this issue,
-// but occasionally it will leading to a crash with a vague error message.
-typedef struct {
-	enum CXCursorKind kind;
-	int xdata;
-	uintptr_t data[3];
-} GoCXCursor;
-
-// Forwarding functions. They are implemented in libclang_stubs.c and forward to
-// the real functions without doing anything else, thus they are entirely
-// compatible with the versions without tinygo_ prefix. The only difference is
-// the CXCursor type, which has been replaced with GoCXCursor.
-GoCXCursor tinygo_clang_getTranslationUnitCursor(CXTranslationUnit tu);
-unsigned tinygo_clang_visitChildren(GoCXCursor parent, CXCursorVisitor visitor, CXClientData client_data);
-CXString tinygo_clang_getCursorSpelling(GoCXCursor c);
-enum CXCursorKind tinygo_clang_getCursorKind(GoCXCursor c);
-CXType tinygo_clang_getCursorType(GoCXCursor c);
-GoCXCursor tinygo_clang_getTypeDeclaration(CXType t);
-CXType tinygo_clang_getTypedefDeclUnderlyingType(GoCXCursor c);
-CXType tinygo_clang_getCursorResultType(GoCXCursor c);
-int tinygo_clang_Cursor_getNumArguments(GoCXCursor c);
-GoCXCursor tinygo_clang_Cursor_getArgument(GoCXCursor c, unsigned i);
-
-int tinygo_clang_globals_visitor(GoCXCursor c, GoCXCursor parent, CXClientData client_data);
-int tinygo_clang_struct_visitor(GoCXCursor c, GoCXCursor parent, CXClientData client_data);
-*/
-import "C"
-
-// refMap stores references to types, used for clang_visitChildren.
-var refMap RefMap
-
-var diagnosticSeverity = [...]string{
-	C.CXDiagnostic_Ignored: "ignored",
-	C.CXDiagnostic_Note:    "note",
-	C.CXDiagnostic_Warning: "warning",
-	C.CXDiagnostic_Error:   "error",
-	C.CXDiagnostic_Fatal:   "fatal",
-}
-
-func (info *fileInfo) parseFragment(fragment string, cflags []string, posFilename string, posLine int) []error {
-	index := C.clang_createIndex(0, 0)
-	defer C.clang_disposeIndex(index)
-
-	filenameC := C.CString(posFilename + "!cgo.c")
-	defer C.free(unsafe.Pointer(filenameC))
-
-	fragmentC := C.CString(fragment)
-	defer C.free(unsafe.Pointer(fragmentC))
-
-	unsavedFile := C.struct_CXUnsavedFile{
-		Filename: filenameC,
-		Length:   C.ulong(len(fragment)),
-		Contents: fragmentC,
-	}
-
-	// convert Go slice of strings to C array of strings.
-	cmdargsC := C.malloc(C.size_t(len(cflags)) * C.size_t(unsafe.Sizeof(uintptr(0))))
-	defer C.free(cmdargsC)
-	cmdargs := (*[1 << 16]*C.char)(cmdargsC)
-	for i, cflag := range cflags {
-		s := C.CString(cflag)
-		cmdargs[i] = s
-		defer C.free(unsafe.Pointer(s))
-	}
-
-	var unit C.CXTranslationUnit
-	errCode := C.clang_parseTranslationUnit2(
-		index,
-		filenameC,
-		(**C.char)(cmdargsC), C.int(len(cflags)), // command line args
-		&unsavedFile, 1, // unsaved files
-		C.CXTranslationUnit_None,
-		&unit)
-	if errCode != 0 {
-		panic("loader: failed to parse source with libclang")
-	}
-	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)]
-			location := C.clang_getDiagnosticLocation(diagnostic)
-			var file C.CXFile
-			var line C.unsigned
-			var column C.unsigned
-			var offset C.unsigned
-			C.clang_getExpansionLocation(location, &file, &line, &column, &offset)
-			filename := getString(C.clang_getFileName(file))
-			if filename == posFilename+"!cgo.c" {
-				// Adjust errors from the `import "C"` snippet.
-				// Note: doesn't adjust filenames inside the error message
-				// itself.
-				filename = posFilename
-				line += C.uint(posLine)
-				offset = 0 // hard to calculate
-			} else if filepath.IsAbs(filename) {
-				// Relative paths for readability, like other Go parser errors.
-				relpath, err := filepath.Rel(info.Program.Dir, filename)
-				if err == nil {
-					filename = relpath
-				}
-			}
-			errs = append(errs, &scanner.Error{
-				Pos: token.Position{
-					Filename: filename,
-					Offset:   int(offset),
-					Line:     int(line),
-					Column:   int(column),
-				},
-				Msg: severity + ": " + spelling,
-			})
-		}
-		for i := 0; i < numDiagnostics; i++ {
-			diagnostic := C.clang_getDiagnostic(unit, C.uint(i))
-			addDiagnostic(diagnostic)
-
-			// Child diagnostics (like notes on redefinitions).
-			diagnostics := C.clang_getChildDiagnostics(diagnostic)
-			for j := 0; j < int(C.clang_getNumDiagnosticsInSet(diagnostics)); j++ {
-				addDiagnostic(C.clang_getDiagnosticInSet(diagnostics, C.uint(j)))
-			}
-		}
-		return errs
-	}
-
-	ref := refMap.Put(info)
-	defer refMap.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)
-	kind := C.tinygo_clang_getCursorKind(c)
-	switch kind {
-	case C.CXCursor_FunctionDecl:
-		name := getString(C.tinygo_clang_getCursorSpelling(c))
-		cursorType := C.tinygo_clang_getCursorType(c)
-		if C.clang_isFunctionTypeVariadic(cursorType) != 0 {
-			return C.CXChildVisit_Continue // not supported
-		}
-		numArgs := int(C.tinygo_clang_Cursor_getNumArguments(c))
-		fn := &functionInfo{}
-		info.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))
-			argType := C.clang_getArgType(cursorType, C.uint(i))
-			if argName == "" {
-				argName = "$" + strconv.Itoa(i)
-			}
-			fn.args = append(fn.args, paramInfo{
-				name:     argName,
-				typeExpr: info.makeASTType(argType),
-			})
-		}
-		resultType := C.tinygo_clang_getCursorResultType(c)
-		if resultType.kind != C.CXType_Void {
-			fn.results = &ast.FieldList{
-				List: []*ast.Field{
-					&ast.Field{
-						Type: info.makeASTType(resultType),
-					},
-				},
-			}
-		}
-	case C.CXCursor_TypedefDecl:
-		typedefType := C.tinygo_clang_getCursorType(c)
-		name := getString(C.clang_getTypedefName(typedefType))
-		underlyingType := C.tinygo_clang_getTypedefDeclUnderlyingType(c)
-		expr := info.makeASTType(underlyingType)
-		if strings.HasPrefix(name, "_Cgo_") {
-			expr := expr.(*ast.Ident)
-			typeSize := C.clang_Type_getSizeOf(underlyingType)
-			switch expr.Name {
-			// TODO: plain char (may be signed or unsigned)
-			case "C.schar", "C.short", "C.int", "C.long", "C.longlong":
-				switch typeSize {
-				case 1:
-					expr.Name = "int8"
-				case 2:
-					expr.Name = "int16"
-				case 4:
-					expr.Name = "int32"
-				case 8:
-					expr.Name = "int64"
-				}
-			case "C.uchar", "C.ushort", "C.uint", "C.ulong", "C.ulonglong":
-				switch typeSize {
-				case 1:
-					expr.Name = "uint8"
-				case 2:
-					expr.Name = "uint16"
-				case 4:
-					expr.Name = "uint32"
-				case 8:
-					expr.Name = "uint64"
-				}
-			}
-		}
-		info.typedefs[name] = &typedefInfo{
-			typeExpr: expr,
-		}
-	case C.CXCursor_VarDecl:
-		name := getString(C.tinygo_clang_getCursorSpelling(c))
-		cursorType := C.tinygo_clang_getCursorType(c)
-		info.globals[name] = &globalInfo{
-			typeExpr: info.makeASTType(cursorType),
-		}
-	}
-	return C.CXChildVisit_Continue
-}
-
-func getString(clangString C.CXString) (s string) {
-	rawString := C.clang_getCString(clangString)
-	s = C.GoString(rawString)
-	C.clang_disposeString(clangString)
-	return
-}
-
-// 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) ast.Expr {
-	var typeName string
-	switch typ.kind {
-	case C.CXType_SChar:
-		typeName = "C.schar"
-	case C.CXType_UChar:
-		typeName = "C.uchar"
-	case C.CXType_Short:
-		typeName = "C.short"
-	case C.CXType_UShort:
-		typeName = "C.ushort"
-	case C.CXType_Int:
-		typeName = "C.int"
-	case C.CXType_UInt:
-		typeName = "C.uint"
-	case C.CXType_Long:
-		typeName = "C.long"
-	case C.CXType_ULong:
-		typeName = "C.ulong"
-	case C.CXType_LongLong:
-		typeName = "C.longlong"
-	case C.CXType_ULongLong:
-		typeName = "C.ulonglong"
-	case C.CXType_Bool:
-		typeName = "bool"
-	case C.CXType_Float, C.CXType_Double, C.CXType_LongDouble:
-		switch C.clang_Type_getSizeOf(typ) {
-		case 4:
-			typeName = "float32"
-		case 8:
-			typeName = "float64"
-		default:
-			// Don't do anything, rely on the fallback code to show a somewhat
-			// sensible error message like "undeclared name: C.long double".
-		}
-	case C.CXType_Complex:
-		switch C.clang_Type_getSizeOf(typ) {
-		case 8:
-			typeName = "complex64"
-		case 16:
-			typeName = "complex128"
-		}
-	case C.CXType_Pointer:
-		return &ast.StarExpr{
-			Star: info.importCPos,
-			X:    info.makeASTType(C.clang_getPointeeType(typ)),
-		}
-	case C.CXType_ConstantArray:
-		return &ast.ArrayType{
-			Lbrack: info.importCPos,
-			Len: &ast.BasicLit{
-				ValuePos: info.importCPos,
-				Kind:     token.INT,
-				Value:    strconv.FormatInt(int64(C.clang_getArraySize(typ)), 10),
-			},
-			Elt: info.makeASTType(C.clang_getElementType(typ)),
-		}
-	case C.CXType_FunctionProto:
-		// Be compatible with gc, which uses the *[0]byte type for function
-		// pointer types.
-		// Return type [0]byte because this is a function type, not a pointer to
-		// this function type.
-		return &ast.ArrayType{
-			Lbrack: info.importCPos,
-			Len: &ast.BasicLit{
-				ValuePos: info.importCPos,
-				Kind:     token.INT,
-				Value:    "0",
-			},
-			Elt: &ast.Ident{
-				NamePos: info.importCPos,
-				Name:    "byte",
-			},
-		}
-	case C.CXType_Typedef:
-		typedefName := getString(C.clang_getTypedefName(typ))
-		return &ast.Ident{
-			NamePos: info.importCPos,
-			Name:    "C." + typedefName,
-		}
-	case C.CXType_Elaborated:
-		underlying := C.clang_Type_getNamedType(typ)
-		switch underlying.kind {
-		case C.CXType_Record:
-			cursor := C.tinygo_clang_getTypeDeclaration(typ)
-			name := getString(C.tinygo_clang_getCursorSpelling(cursor))
-			// It is possible that this is a recursive definition, for example
-			// in linked lists (structs contain a pointer to the next element
-			// of the same type). If the name exists in info.elaboratedTypes,
-			// it is being processed, although it may not be fully defined yet.
-			if _, ok := info.elaboratedTypes[name]; !ok {
-				info.elaboratedTypes[name] = nil // predeclare (to avoid endless recursion)
-				info.elaboratedTypes[name] = info.makeASTType(underlying)
-			}
-			return &ast.Ident{
-				NamePos: info.importCPos,
-				Name:    "C.struct_" + name,
-			}
-		default:
-			panic("unknown elaborated type")
-		}
-	case C.CXType_Record:
-		cursor := C.tinygo_clang_getTypeDeclaration(typ)
-		fieldList := &ast.FieldList{
-			Opening: info.importCPos,
-			Closing: info.importCPos,
-		}
-		ref := refMap.Put(struct {
-			fieldList *ast.FieldList
-			info      *fileInfo
-		}{fieldList, info})
-		defer refMap.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:
-			return &ast.StructType{
-				Struct: info.importCPos,
-				Fields: fieldList,
-			}
-		case C.CXCursor_UnionDecl:
-			if len(fieldList.List) > 1 {
-				// Insert a special field at the front (of zero width) as a
-				// marker that this is struct is actually a union. This is done
-				// by giving the field a name that cannot be expressed directly
-				// in Go.
-				// Other parts of the compiler look at the first element in a
-				// struct (of size > 2) to know whether this is a union.
-				// Note that we don't have to insert it for single-element
-				// unions as they're basically equivalent to a struct.
-				unionMarker := &ast.Field{
-					Type: &ast.StructType{
-						Struct: info.importCPos,
-					},
-				}
-				unionMarker.Names = []*ast.Ident{
-					&ast.Ident{
-						NamePos: info.importCPos,
-						Name:    "C union",
-						Obj: &ast.Object{
-							Kind: ast.Var,
-							Name: "C union",
-							Decl: unionMarker,
-						},
-					},
-				}
-				fieldList.List = append([]*ast.Field{unionMarker}, fieldList.List...)
-			}
-			return &ast.StructType{
-				Struct: info.importCPos,
-				Fields: fieldList,
-			}
-		}
-	}
-	if typeName == "" {
-		// Fallback, probably incorrect but at least the error points to an odd
-		// type name.
-		typeName = "C." + getString(C.clang_getTypeSpelling(typ))
-	}
-	return &ast.Ident{
-		NamePos: info.importCPos,
-		Name:    typeName,
-	}
-}
-
-//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 {
-		fieldList *ast.FieldList
-		info      *fileInfo
-	})
-	fieldList := passed.fieldList
-	info := passed.info
-	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),
-	}
-	field.Names = []*ast.Ident{
-		&ast.Ident{
-			NamePos: info.importCPos,
-			Name:    name,
-			Obj: &ast.Object{
-				Kind: ast.Var,
-				Name: name,
-				Decl: field,
-			},
-		},
-	}
-	fieldList.List = append(fieldList.List, field)
-	return C.CXChildVisit_Continue
-}

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.
@@ -22,6 +24,7 @@ type Program struct {
 	fset          *token.FileSet
 	TypeChecker   types.Config
 	Dir           string // current working directory (for error reporting)
+	TINYGOROOT    string // root of the TinyGo installation or root of the source code
 	CFlags        []string
 }
 
@@ -299,12 +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))
+		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}
@@ -333,7 +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

@@ -33,9 +33,20 @@ func (e *commandError) Error() string {
 	return e.Msg + " " + e.File + ": " + e.Err.Error()
 }
 
+// multiError is a list of multiple errors (actually: diagnostics) returned
+// during LLVM IR generation.
+type multiError struct {
+	Errs []error
+}
+
+func (e *multiError) Error() string {
+	return e.Errs[0].Error()
+}
+
 type BuildConfig struct {
 	opt           string
 	gc            string
+	panicStrategy string
 	printIR       bool
 	dumpSSA       bool
 	debug         bool
@@ -51,21 +62,37 @@ func Compile(pkgName, outpath string, spec *TargetSpec, config *BuildConfig, act
 		config.gc = spec.GC
 	}
 
-	// Append command line passed CFlags and LDFlags
+	root := sourceDir()
-	spec.CFlags = append(spec.CFlags, config.cFlags...)
-	spec.LDFlags = append(spec.LDFlags, config.ldFlags...)
 
+	// Merge and adjust CFlags.
+	cflags := append([]string{}, config.cFlags...)
+	for _, flag := range spec.CFlags {
+		cflags = append(cflags, strings.Replace(flag, "{root}", root, -1))
+	}
+
+	// Merge and adjust LDFlags.
+	ldflags := append([]string{}, config.ldFlags...)
+	for _, flag := range spec.LDFlags {
+		ldflags = append(ldflags, strings.Replace(flag, "{root}", root, -1))
+	}
+
+	goroot := getGoroot()
+	if goroot == "" {
+		return errors.New("cannot locate $GOROOT, please set it manually")
+	}
 	compilerConfig := compiler.Config{
 		Triple:        spec.Triple,
 		CPU:           spec.CPU,
 		GOOS:          spec.GOOS,
 		GOARCH:        spec.GOARCH,
 		GC:            config.gc,
-		CFlags:    spec.CFlags,
+		PanicStrategy: config.panicStrategy,
-		LDFlags:   spec.LDFlags,
+		CFlags:        cflags,
+		LDFlags:       ldflags,
 		Debug:         config.debug,
 		DumpSSA:       config.dumpSSA,
-		RootDir:   sourceDir(),
+		TINYGOROOT:    root,
+		GOROOT:        goroot,
 		GOPATH:        getGopath(),
 		BuildTags:     spec.BuildTags,
 	}
@@ -75,12 +102,15 @@ func Compile(pkgName, outpath string, spec *TargetSpec, config *BuildConfig, act
 	}
 
 	// Compile Go code to IR.
-	err = c.Compile(pkgName)
+	errs := c.Compile(pkgName)
-	if err != nil {
+	if len(errs) != 0 {
-		return err
+		if len(errs) == 1 {
+			return errs[0]
+		}
+		return &multiError{errs}
 	}
 	if config.printIR {
-		fmt.Println("Generated LLVM IR:")
+		fmt.Println("; Generated LLVM IR:")
 		fmt.Println(c.IR())
 	}
 	if err := c.Verify(); err != nil {
@@ -191,19 +221,20 @@ func Compile(pkgName, outpath string, spec *TargetSpec, config *BuildConfig, act
 		// Prepare link command.
 		executable := filepath.Join(dir, "main")
 		tmppath := executable // final file
-		ldflags := append(spec.LDFlags, "-o", executable, objfile, "-L", sourceDir())
+		ldflags := append(ldflags, "-o", executable, objfile, "-L", root)
 		if spec.RTLib == "compiler-rt" {
 			ldflags = append(ldflags, librt)
 		}
 
 		// Compile extra files.
 		for i, path := range spec.ExtraFiles {
+			abspath := filepath.Join(root, path)
 			outpath := filepath.Join(dir, "extra-"+strconv.Itoa(i)+"-"+filepath.Base(path)+".o")
-			cmd := exec.Command(spec.Compiler, append(spec.CFlags, "-c", "-o", outpath, path)...)
+			cmdNames := []string{spec.Compiler}
-			cmd.Stdout = os.Stdout
+			if names, ok := commands[spec.Compiler]; ok {
-			cmd.Stderr = os.Stderr
+				cmdNames = names
-			cmd.Dir = sourceDir()
+			}
-			err := cmd.Run()
+			err := execCommand(cmdNames, append(cflags, "-c", "-o", outpath, abspath)...)
 			if err != nil {
 				return &commandError{"failed to build", path, err}
 			}
@@ -215,11 +246,11 @@ func Compile(pkgName, outpath string, spec *TargetSpec, config *BuildConfig, act
 			for _, file := range pkg.CFiles {
 				path := filepath.Join(pkg.Package.Dir, file)
 				outpath := filepath.Join(dir, "pkg"+strconv.Itoa(i)+"-"+file+".o")
-				cmd := exec.Command(spec.Compiler, append(spec.CFlags, "-c", "-o", outpath, path)...)
+				cmdNames := []string{spec.Compiler}
-				cmd.Stdout = os.Stdout
+				if names, ok := commands[spec.Compiler]; ok {
-				cmd.Stderr = os.Stderr
+					cmdNames = names
-				cmd.Dir = sourceDir()
+				}
-				err := cmd.Run()
+				err := execCommand(cmdNames, append(cflags, "-c", "-o", outpath, path)...)
 				if err != nil {
 					return &commandError{"failed to build", path, err}
 				}
@@ -228,11 +259,7 @@ func Compile(pkgName, outpath string, spec *TargetSpec, config *BuildConfig, act
 		}
 
 		// Link the object files together.
-		if linker, ok := commands[spec.Linker]; ok {
-			err = Link(linker, ldflags...)
-		} else {
 		err = Link(spec.Linker, ldflags...)
-		}
 		if err != nil {
 			return &commandError{"failed to link", executable, err}
 		}
@@ -496,6 +523,10 @@ func handleCompilerError(err error) {
 			for _, err := range errLoader.Errs {
 				fmt.Fprintln(os.Stderr, err)
 			}
+		} else if errMulti, ok := err.(*multiError); ok {
+			for _, err := range errMulti.Errs {
+				fmt.Fprintln(os.Stderr, err)
+			}
 		} else {
 			fmt.Fprintln(os.Stderr, "error:", err)
 		}
@@ -506,7 +537,8 @@ 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")
 	target := flag.String("target", "", "LLVM target")
@@ -529,6 +561,7 @@ func main() {
 	config := &BuildConfig{
 		opt:           *opt,
 		gc:            *gc,
+		panicStrategy: *panicStrategy,
 		printIR:       *printIR,
 		dumpSSA:       *dumpSSA,
 		debug:         !*nodebug,
@@ -544,6 +577,12 @@ func main() {
 		config.ldFlags = strings.Split(*ldFlags, " ")
 	}
 
+	if *panicStrategy != "print" && *panicStrategy != "trap" {
+		fmt.Fprintln(os.Stderr, "Panic strategy must be either print or trap.")
+		usage()
+		os.Exit(1)
+	}
+
 	os.Setenv("CC", "clang -target="+*target)
 
 	switch command {

main_test.go

@@ -13,17 +13,19 @@ import (
 	"runtime"
 	"sort"
 	"testing"
+
+	"github.com/tinygo-org/tinygo/loader"
 )
 
 const TESTDATA = "testdata"
 
 func TestCompiler(t *testing.T) {
-	matches, err := filepath.Glob(TESTDATA + "/*.go")
+	matches, err := filepath.Glob(filepath.Join(TESTDATA, "*.go"))
 	if err != nil {
 		t.Fatal("could not read test files:", err)
 	}
 
-	dirMatches, err := filepath.Glob(TESTDATA + "/*/main.go")
+	dirMatches, err := filepath.Glob(filepath.Join(TESTDATA, "*", "main.go"))
 	if err != nil {
 		t.Fatal("could not read test packages:", err)
 	}
@@ -56,6 +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)
 		})
@@ -64,7 +69,7 @@ func TestCompiler(t *testing.T) {
 	if runtime.GOOS == "linux" {
 		t.Log("running tests for linux/arm...")
 		for _, path := range matches {
-			if path == "testdata/cgo/" {
+			if path == filepath.Join("testdata", "cgo")+string(filepath.Separator) {
 				continue // TODO: improve CGo
 			}
 			t.Run(path, func(t *testing.T) {
@@ -74,7 +79,7 @@ func TestCompiler(t *testing.T) {
 
 		t.Log("running tests for linux/arm64...")
 		for _, path := range matches {
-			if path == "testdata/cgo/" {
+			if path == filepath.Join("testdata", "cgo")+string(filepath.Separator) {
 				continue // TODO: improve CGo
 			}
 			t.Run(path, func(t *testing.T) {
@@ -84,7 +89,7 @@ func TestCompiler(t *testing.T) {
 
 		t.Log("running tests for WebAssembly...")
 		for _, path := range matches {
-			if path == "testdata/gc.go" {
+			if path == filepath.Join("testdata", "gc.go") {
 				continue // known to fail
 			}
 			t.Run(path, func(t *testing.T) {
@@ -97,7 +102,7 @@ func TestCompiler(t *testing.T) {
 func runTest(path, tmpdir string, target string, t *testing.T) {
 	// Get the expected output for this test.
 	txtpath := path[:len(path)-3] + ".txt"
-	if path[len(path)-1] == '/' {
+	if path[len(path)-1] == os.PathSeparator {
 		txtpath = path + "out.txt"
 	}
 	f, err := os.Open(txtpath)
@@ -121,7 +126,13 @@ func runTest(path, tmpdir string, target string, t *testing.T) {
 	binary := filepath.Join(tmpdir, "test")
 	err = Build("./"+path, binary, target, config)
 	if err != nil {
+		if errLoader, ok := err.(loader.Errors); ok {
+			for _, err := range errLoader.Errs {
 				t.Log("failed to build:", err)
+			}
+		} else {
+			t.Log("failed to build:", err)
+		}
 		t.Fail()
 		return
 	}

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/examples/microbit-blink/microbit-blink.go

@@ -1,4 +1,4 @@
-// blink program for the BBC micro:bit that uses the entire LED matrix
+// blink program for the BBC micro:bit
 package main
 
 import (
@@ -6,14 +6,19 @@ import (
 	"time"
 )
 
+// The LED matrix in the micro:bit is a multiplexed display: https://en.wikipedia.org/wiki/Multiplexed_display
+// Driver for easier control: https://github.com/tinygo-org/drivers/tree/master/microbitmatrix
 func main() {
-	machine.InitLEDMatrix()
+	ledrow := machine.GPIO{machine.LED_ROW_1}
-
+	ledrow.Configure(machine.GPIOConfig{Mode: machine.GPIO_OUTPUT})
+	ledcol := machine.GPIO{machine.LED_COL_1}
+	ledcol.Configure(machine.GPIOConfig{Mode: machine.GPIO_OUTPUT})
+	ledcol.Low()
 	for {
-		machine.ClearLEDMatrix()
+		ledrow.Low()
 		time.Sleep(time.Millisecond * 500)
 
-		machine.SetEntireLEDMatrixOn()
+		ledrow.High()
 		time.Sleep(time.Millisecond * 500)
 	}
 }

src/examples/wasm/README.md

@@ -1,6 +1,6 @@
 # TinyGo WebAssembly examples
 
-The examples here show two different ways of using WebAssembly with TinyGo;
+The examples here show two different ways of using WebAssembly with TinyGo:
 
 1. Defining and exporting functions via the `//go:export <name>` directive. See
 [the export folder](./export) for an example of this.
@@ -31,23 +31,31 @@ $ make main
 
 ## Running
 
-Start the local webserver:
+Start the local web server:
 
 ```bash
 $ go run main.go
 Serving ./html on http://localhost:8080
 ```
 
-`fmt.Println` prints to the browser console.
+Use your web browser to visit http://localhost:8080.
+
+* The wasm "export" example displays a simple math equation using HTML, with
+the result calculated dynamically using WebAssembly.  Changing any of the
+values on the left hand side triggers the exported wasm `update` function to
+recalculate the result.
+* The wasm "main" example uses `println` to write to your browser JavaScript
+console. You may need to open the browser development tools console to see it.
 
 ## How it works
 
-Execution of the contents require a few JS helper functions which are called
+Execution of the contents require a few JavaScript helper functions which are
-from WebAssembly. We have defined these in
+called from WebAssembly.
-[wasm_exec.js](../../../targets/wasm_exec.js). It is based on
+
-`$GOROOT/misc/wasm/wasm_exec.js` from the standard library, but is slightly
+We have defined these in [wasm_exec.js](../../../targets/wasm_exec.js).  It is
-different. Ensure you are using the same version of `wasm_exec.js` as the
+based on `$GOROOT/misc/wasm/wasm_exec.js` from the standard library, but is
-version of `tinygo` you are using to compile.
+slightly different. Ensure you are using the same version of `wasm_exec.js` as
+the version of `tinygo` you are using to compile.
 
 The general steps required to run the WebAssembly file in the browser includes
 loading it into JavaScript with `WebAssembly.instantiateStreaming`, or
@@ -80,8 +88,9 @@ If you have used explicit exports, you can call them by invoking them under the
 `wasm.exports` namespace. See the [`export`](./export/wasm.js) directory for an
 example of this.
 
-In addition to this piece of JavaScript, it is important that the file is served
+In addition to the JavaScript, it is important the wasm file is served with the
-with the correct `Content-Type` header set.
+[`Content-Type`](https://developer.mozilla.org/en-US/docs/Web/HTTP/Headers/Content-Type)
+header set to `application/wasm`.  Without it, most browsers won't run it.
 
 ```go
 package main
@@ -98,14 +107,25 @@ func main() {
 	fs := http.FileServer(http.Dir(dir))
 	log.Print("Serving " + dir + " on http://localhost:8080")
 	http.ListenAndServe(":8080", http.HandlerFunc(func(resp http.ResponseWriter, req *http.Request) {
+		resp.Header().Add("Cache-Control", "no-cache")
 		if strings.HasSuffix(req.URL.Path, ".wasm") {
 			resp.Header().Set("content-type", "application/wasm")
 		}
-
 		fs.ServeHTTP(resp, req)
-	}))
+	}))}
-}
 ```
 
-This simple server serves anything inside the `./html` directory on port `8080`,
+This simple server serves anything inside the `./html` directory on port
-setting any `*.wasm` files `Content-Type` header appropriately.
+`8080`, setting any `*.wasm` files `Content-Type` header appropriately.
+
+For development purposes (**only!**), it also sets the `Cache-Control` header
+so your browser doesn't cache the files.  This is useful while developing, to
+ensure your browser displays the newest wasm when you recompile.
+
+In a production environment you **probably wouldn't** want to set the
+`Cache-Control` header like this.  Caching is generally beneficial for end
+users.
+
+Further information on the `Cache-Control` header can be found here:
+
+* https://developer.mozilla.org/en-US/docs/Web/HTTP/Headers/Cache-Control

src/examples/wasm/main/main.go

@@ -1,9 +1,5 @@
 package main
 
-import (
-	"fmt"
-)
-
 func main() {
-	fmt.Println("Hello world!")
+	println("Hello world!")
 }

src/examples/wasm/server.go

@@ -12,10 +12,10 @@ func main() {
 	fs := http.FileServer(http.Dir(dir))
 	log.Print("Serving " + dir + " on http://localhost:8080")
 	http.ListenAndServe(":8080", http.HandlerFunc(func(resp http.ResponseWriter, req *http.Request) {
+		resp.Header().Add("Cache-Control", "no-cache")
 		if strings.HasSuffix(req.URL.Path, ".wasm") {
 			resp.Header().Set("content-type", "application/wasm")
 		}
-
 		fs.ServeHTTP(resp, req)
 	}))
 }

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/board_microbit.go

@@ -2,11 +2,6 @@
 
 package machine
 
-import (
-	"device/nrf"
-	"errors"
-)
-
 // The micro:bit does not have a 32kHz crystal on board.
 const HasLowFrequencyCrystal = false
 
@@ -79,65 +74,3 @@ const (
 	LED_ROW_2 = 14
 	LED_ROW_3 = 15
 )
-
-// matrixSettings has the legs of the LED grid in the form {row, column} for each LED position.
-var matrixSettings = [5][5][2]uint8{
-	{{LED_ROW_1, LED_COL_1}, {LED_ROW_2, LED_COL_4}, {LED_ROW_1, LED_COL_2}, {LED_ROW_2, LED_COL_5}, {LED_ROW_1, LED_COL_3}},
-	{{LED_ROW_3, LED_COL_4}, {LED_ROW_3, LED_COL_5}, {LED_ROW_3, LED_COL_6}, {LED_ROW_3, LED_COL_7}, {LED_ROW_3, LED_COL_8}},
-	{{LED_ROW_2, LED_COL_2}, {LED_ROW_1, LED_COL_9}, {LED_ROW_2, LED_COL_3}, {LED_ROW_3, LED_COL_9}, {LED_ROW_2, LED_COL_1}},
-	{{LED_ROW_1, LED_COL_8}, {LED_ROW_1, LED_COL_7}, {LED_ROW_1, LED_COL_6}, {LED_ROW_1, LED_COL_5}, {LED_ROW_1, LED_COL_4}},
-	{{LED_ROW_3, LED_COL_3}, {LED_ROW_2, LED_COL_7}, {LED_ROW_3, LED_COL_1}, {LED_ROW_2, LED_COL_6}, {LED_ROW_3, LED_COL_2}}}
-
-// InitLEDMatrix initializes the LED matrix, by setting all of the row/col pins to output
-// then calling ClearLEDMatrix.
-func InitLEDMatrix() {
-	set := 0
-	for i := LED_COL_1; i <= LED_ROW_3; i++ {
-		set |= 1 << uint8(i)
-	}
-	nrf.GPIO.DIRSET = nrf.RegValue(set)
-	ClearLEDMatrix()
-}
-
-// ClearLEDMatrix clears the entire LED matrix.
-func ClearLEDMatrix() {
-	set := 0
-	for i := LED_COL_1; i <= LED_COL_9; i++ {
-		set |= 1 << uint8(i)
-	}
-	nrf.GPIO.OUTSET = nrf.RegValue(set)
-	nrf.GPIO.OUTCLR = (1 << LED_ROW_1) | (1 << LED_ROW_2) | (1 << LED_ROW_3)
-}
-
-// SetLEDMatrix turns on a single LED on the LED matrix.
-// Currently limited to a single LED at a time, it will clear the matrix before setting it.
-func SetLEDMatrix(x, y uint8) error {
-	if x > 4 || y > 4 {
-		return errors.New("Invalid LED matrix row or column")
-	}
-
-	// Clear matrix
-	ClearLEDMatrix()
-
-	nrf.GPIO.OUTSET = (1 << matrixSettings[y][x][0])
-	nrf.GPIO.OUTCLR = (1 << matrixSettings[y][x][1])
-
-	return nil
-}
-
-// SetEntireLEDMatrixOn turns on all of the LEDs on the LED matrix.
-func SetEntireLEDMatrixOn() error {
-	set := 0
-	for i := LED_ROW_1; i <= LED_ROW_3; i++ {
-		set |= 1 << uint8(i)
-	}
-	nrf.GPIO.OUTSET = nrf.RegValue(set)
-
-	set = 0
-	for i := LED_COL_1; i <= LED_COL_9; i++ {
-		set |= 1 << uint8(i)
-	}
-	nrf.GPIO.OUTCLR = nrf.RegValue(set)
-
-	return nil
-}

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

@@ -343,6 +343,12 @@ func (a ADC) Get() uint16 {
 	sam.ADC.INPUTCTRL |= sam.RegValue(ch << sam.ADC_INPUTCTRL_MUXPOS_Pos)
 	waitADCSync()
 
+	// Select internal ground for ADC input
+	sam.ADC.INPUTCTRL &^= sam.ADC_INPUTCTRL_MUXNEG_Msk
+	waitADCSync()
+	sam.ADC.INPUTCTRL |= sam.RegValue(sam.ADC_INPUTCTRL_MUXNEG_GND << sam.ADC_INPUTCTRL_MUXNEG_Pos)
+	waitADCSync()
+
 	// Enable ADC
 	sam.ADC.CTRLA |= sam.ADC_CTRLA_ENABLE
 	waitADCSync()
@@ -368,7 +374,7 @@ func (a ADC) Get() uint16 {
 	sam.ADC.CTRLA &^= sam.ADC_CTRLA_ENABLE
 	waitADCSync()
 
-	return uint16(val)
+	return uint16(val) << 4 // scales from 12 to 16-bit result
 }
 
 func (a ADC) getADCChannel() uint8 {
@@ -802,6 +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/chan.go

@@ -39,33 +39,27 @@ const (
 	chanStateClosed
 )
 
-func chanSendStub(caller *coroutine, ch *channel, _ unsafe.Pointer, size uintptr)
-func chanRecvStub(caller *coroutine, ch *channel, _ unsafe.Pointer, _ *bool, size uintptr)
 func deadlockStub()
 
 // chanSend sends a single value over the channel. If this operation can
 // complete immediately (there is a goroutine waiting for a value), it sends the
 // value and re-activates both goroutines. If not, it sets itself as waiting on
 // a value.
-//
+func chanSend(sender *coroutine, ch *channel, value unsafe.Pointer, size uintptr) {
-// The unsafe.Pointer value is used during lowering. During IR generation, it
-// points to the to-be-received value. During coroutine lowering, this value is
-// replaced with a read from the coroutine promise.
-func chanSend(sender *coroutine, ch *channel, _ unsafe.Pointer, size uintptr) {
 	if ch == nil {
 		// A nil channel blocks forever. Do not scheduler this goroutine again.
 		return
 	}
 	switch ch.state {
 	case chanStateEmpty:
+		sender.promise().ptr = value
 		ch.state = chanStateSend
 		ch.blocked = sender
 	case chanStateRecv:
 		receiver := ch.blocked
 		receiverPromise := receiver.promise()
-		senderPromise := sender.promise()
+		memcpy(receiverPromise.ptr, value, size)
-		memcpy(unsafe.Pointer(&receiverPromise.data), unsafe.Pointer(&senderPromise.data), size)
+		receiverPromise.data = 1 // commaOk = true
-		receiverPromise.commaOk = true
 		ch.blocked = receiverPromise.next
 		receiverPromise.next = nil
 		activateTask(receiver)
@@ -76,6 +70,7 @@ func chanSend(sender *coroutine, ch *channel, _ unsafe.Pointer, size uintptr) {
 	case chanStateClosed:
 		runtimePanic("send on closed channel")
 	case chanStateSend:
+		sender.promise().ptr = value
 		sender.promise().next = ch.blocked
 		ch.blocked = sender
 	}
@@ -85,11 +80,7 @@ func chanSend(sender *coroutine, ch *channel, _ unsafe.Pointer, size uintptr) {
 // sender, it receives the value immediately and re-activates both coroutines.
 // If not, it sets itself as available for receiving. If the channel is closed,
 // it immediately activates itself with a zero value as the result.
-//
+func chanRecv(receiver *coroutine, ch *channel, value unsafe.Pointer, size uintptr) {
-// The two unnamed values exist to help during lowering. The unsafe.Pointer
-// points to the value, and the *bool points to the comma-ok value. Both are
-// replaced by reads from the coroutine promise.
-func chanRecv(receiver *coroutine, ch *channel, _ unsafe.Pointer, _ *bool, size uintptr) {
 	if ch == nil {
 		// A nil channel blocks forever. Do not scheduler this goroutine again.
 		return
@@ -97,10 +88,9 @@ func chanRecv(receiver *coroutine, ch *channel, _ unsafe.Pointer, _ *bool, size
 	switch ch.state {
 	case chanStateSend:
 		sender := ch.blocked
-		receiverPromise := receiver.promise()
 		senderPromise := sender.promise()
-		memcpy(unsafe.Pointer(&receiverPromise.data), unsafe.Pointer(&senderPromise.data), size)
+		memcpy(value, senderPromise.ptr, size)
-		receiverPromise.commaOk = true
+		receiver.promise().data = 1 // commaOk = true
 		ch.blocked = senderPromise.next
 		senderPromise.next = nil
 		activateTask(receiver)
@@ -109,14 +99,15 @@ func chanRecv(receiver *coroutine, ch *channel, _ unsafe.Pointer, _ *bool, size
 			ch.state = chanStateEmpty
 		}
 	case chanStateEmpty:
+		receiver.promise().ptr = value
 		ch.state = chanStateRecv
 		ch.blocked = receiver
 	case chanStateClosed:
-		receiverPromise := receiver.promise()
+		memzero(value, size)
-		memzero(unsafe.Pointer(&receiverPromise.data), size)
+		receiver.promise().data = 0 // commaOk = false
-		receiverPromise.commaOk = false
 		activateTask(receiver)
 	case chanStateRecv:
+		receiver.promise().ptr = value
 		receiver.promise().next = ch.blocked
 		ch.blocked = receiver
 	}
@@ -142,8 +133,8 @@ func chanClose(ch *channel, size uintptr) {
 	case chanStateRecv:
 		// The receiver must be re-activated with a zero value.
 		receiverPromise := ch.blocked.promise()
-		memzero(unsafe.Pointer(&receiverPromise.data), size)
+		memzero(receiverPromise.ptr, size)
-		receiverPromise.commaOk = false
+		receiverPromise.data = 0 // commaOk = false
 		activateTask(ch.blocked)
 		ch.state = chanStateClosed
 		ch.blocked = nil

src/runtime/complex.go

@@ -0,0 +1,65 @@
+// Copyright 2010 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package runtime
+
+// inf2one returns a signed 1 if f is an infinity and a signed 0 otherwise.
+// The sign of the result is the sign of f.
+func inf2one(f float64) float64 {
+	g := 0.0
+	if isInf(f) {
+		g = 1.0
+	}
+	return copysign(g, f)
+}
+
+func complex64div(n complex64, m complex64) complex64 {
+	return complex64(complex128div(complex128(n), complex128(m)))
+}
+
+func complex128div(n complex128, m complex128) complex128 {
+	var e, f float64 // complex(e, f) = n/m
+
+	// Algorithm for robust complex division as described in
+	// Robert L. Smith: Algorithm 116: Complex division. Commun. ACM 5(8): 435 (1962).
+	if abs(real(m)) >= abs(imag(m)) {
+		ratio := imag(m) / real(m)
+		denom := real(m) + ratio*imag(m)
+		e = (real(n) + imag(n)*ratio) / denom
+		f = (imag(n) - real(n)*ratio) / denom
+	} else {
+		ratio := real(m) / imag(m)
+		denom := imag(m) + ratio*real(m)
+		e = (real(n)*ratio + imag(n)) / denom
+		f = (imag(n)*ratio - real(n)) / denom
+	}
+
+	if isNaN(e) && isNaN(f) {
+		// Correct final result to infinities and zeros if applicable.
+		// Matches C99: ISO/IEC 9899:1999 - G.5.1  Multiplicative operators.
+
+		a, b := real(n), imag(n)
+		c, d := real(m), imag(m)
+
+		switch {
+		case m == 0 && (!isNaN(a) || !isNaN(b)):
+			e = copysign(inf, c) * a
+			f = copysign(inf, c) * b
+
+		case (isInf(a) || isInf(b)) && isFinite(c) && isFinite(d):
+			a = inf2one(a)
+			b = inf2one(b)
+			e = inf * (a*c + b*d)
+			f = inf * (b*c - a*d)
+
+		case (isInf(c) || isInf(d)) && isFinite(a) && isFinite(b):
+			c = inf2one(c)
+			d = inf2one(d)
+			e = 0 * (a*c + b*d)
+			f = 0 * (b*c - a*d)
+		}
+	}
+
+	return complex(e, f)
+}

src/runtime/float.go

@@ -0,0 +1,53 @@
+// Copyright 2017 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package runtime
+
+import "unsafe"
+
+var inf = float64frombits(0x7FF0000000000000)
+
+// isNaN reports whether f is an IEEE 754 ``not-a-number'' value.
+func isNaN(f float64) (is bool) {
+	// IEEE 754 says that only NaNs satisfy f != f.
+	return f != f
+}
+
+// isFinite reports whether f is neither NaN nor an infinity.
+func isFinite(f float64) bool {
+	return !isNaN(f - f)
+}
+
+// isInf reports whether f is an infinity.
+func isInf(f float64) bool {
+	return !isNaN(f) && !isFinite(f)
+}
+
+// Abs returns the absolute value of x.
+//
+// Special cases are:
+//	Abs(±Inf) = +Inf
+//	Abs(NaN) = NaN
+func abs(x float64) float64 {
+	const sign = 1 << 63
+	return float64frombits(float64bits(x) &^ sign)
+}
+
+// copysign returns a value with the magnitude
+// of x and the sign of y.
+func copysign(x, y float64) float64 {
+	const sign = 1 << 63
+	return float64frombits(float64bits(x)&^sign | float64bits(y)&sign)
+}
+
+// Float64bits returns the IEEE 754 binary representation of f.
+func float64bits(f float64) uint64 {
+	return *(*uint64)(unsafe.Pointer(&f))
+}
+
+// Float64frombits returns the floating point number corresponding
+// the IEEE 754 binary representation b.
+func float64frombits(b uint64) float64 {
+	return *(*float64)(unsafe.Pointer(&b))
+}

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/print.go

@@ -4,6 +4,10 @@ import (
 	"unsafe"
 )
 
+type stringer interface {
+	String() string
+}
+
 //go:nobounds
 func printstring(s string) {
 	for i := 0; i < len(s); i++ {
@@ -199,8 +203,44 @@ func printnl() {
 
 func printitf(msg interface{}) {
 	switch msg := msg.(type) {
+	case bool:
+		print(msg)
+	case int:
+		print(msg)
+	case int8:
+		print(msg)
+	case int16:
+		print(msg)
+	case int32:
+		print(msg)
+	case int64:
+		print(msg)
+	case uint:
+		print(msg)
+	case uint8:
+		print(msg)
+	case uint16:
+		print(msg)
+	case uint32:
+		print(msg)
+	case uint64:
+		print(msg)
+	case uintptr:
+		print(msg)
+	case float32:
+		print(msg)
+	case float64:
+		print(msg)
+	case complex64:
+		print(msg)
+	case complex128:
+		print(msg)
 	case string:
 		print(msg)
+	case error:
+		print(msg.Error())
+	case stringer:
+		print(msg.String())
 	default:
 		// cast to underlying type
 		itf := *(*_interface)(unsafe.Pointer(&msg))

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/scheduler.go

@@ -49,12 +49,16 @@ func (t *coroutine) promise() *taskState {
 
 func makeGoroutine(*uint8) *uint8
 
+// Compiler stub to get the current goroutine. Calls to this function are
+// removed in the goroutine lowering pass.
+func getCoroutine() *coroutine
+
 // State/promise of a task. Internally represented as:
 //
 //     {i8* next, i1 commaOk, i32/i64 data}
 type taskState struct {
 	next *coroutine
-	commaOk bool // 'comma-ok' flag for channel receive operation
+	ptr  unsafe.Pointer
 	data uint
 }
 
@@ -107,6 +111,24 @@ func activateTask(task *coroutine) {
 	runqueuePushBack(task)
 }
 
+// getTaskPromisePtr is a helper function to set the current .ptr field of a
+// coroutine promise.
+func setTaskPromisePtr(task *coroutine, value unsafe.Pointer) {
+	task.promise().ptr = value
+}
+
+// getTaskPromisePtr is a helper function to get the current .ptr field from a
+// coroutine promise.
+func getTaskPromisePtr(task *coroutine) unsafe.Pointer {
+	return task.promise().ptr
+}
+
+// getTaskPromiseData is a helper function to get the current .data field of a
+// coroutine promise.
+func getTaskPromiseData(task *coroutine) uint {
+	return task.promise().data
+}
+
 // Add this task to the end of the run queue. May also destroy the task if it's
 // done.
 func runqueuePushBack(t *coroutine) {

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)

target.go

@@ -6,6 +6,7 @@ import (
 	"fmt"
 	"io"
 	"os"
+	"os/exec"
 	"os/user"
 	"path/filepath"
 	"runtime"
@@ -211,7 +212,7 @@ func defaultTarget(goos, goarch, triple string) (*TargetSpec, error) {
 		GOOS:      goos,
 		GOARCH:    goarch,
 		BuildTags: []string{goos, goarch},
-		Compiler:  commands["clang"],
+		Compiler:  "clang",
 		Linker:    "cc",
 		GDB:       "gdb",
 		GDBCmds:   []string{"run"},
@@ -310,3 +311,60 @@ func getHomeDir() string {
 	}
 	return u.HomeDir
 }
+
+// getGoroot returns an appropriate GOROOT from various sources. If it can't be
+// found, it returns an empty string.
+func getGoroot() string {
+	goroot := os.Getenv("GOROOT")
+	if goroot != "" {
+		// An explicitly set GOROOT always has preference.
+		return goroot
+	}
+
+	// Check for the location of the 'go' binary and base GOROOT on that.
+	binpath, err := exec.LookPath("go")
+	if err == nil {
+		binpath, err = filepath.EvalSymlinks(binpath)
+		if err == nil {
+			goroot := filepath.Dir(filepath.Dir(binpath))
+			if isGoroot(goroot) {
+				return goroot
+			}
+		}
+	}
+
+	// Check what GOROOT was at compile time.
+	if isGoroot(runtime.GOROOT()) {
+		return runtime.GOROOT()
+	}
+
+	// Check for some standard locations, as a last resort.
+	var candidates []string
+	switch runtime.GOOS {
+	case "linux":
+		candidates = []string{
+			"/usr/local/go", // manually installed
+			"/usr/lib/go",   // from the distribution
+		}
+	case "darwin":
+		candidates = []string{
+			"/usr/local/go",             // manually installed
+			"/usr/local/opt/go/libexec", // from Homebrew
+		}
+	}
+
+	for _, candidate := range candidates {
+		if isGoroot(candidate) {
+			return candidate
+		}
+	}
+
+	// Can't find GOROOT...
+	return ""
+}
+
+// isGoroot checks whether the given path looks like a GOROOT.
+func isGoroot(goroot string) bool {
+	_, err := os.Stat(filepath.Join(goroot, "src", "runtime", "internal", "sys", "zversion.go"))
+	return err == nil
+}

targets/cortex-m.json

@@ -2,14 +2,17 @@
 	"build-tags": ["cortexm", "linux", "arm"],
 	"goos": "linux",
 	"goarch": "arm",
-	"compiler": "clang-8",
+	"compiler": "clang",
-	"gc": "marksweep",
+	"gc": "precise",
 	"linker": "ld.lld",
 	"rtlib": "compiler-rt",
 	"cflags": [
 		"-Oz",
 		"-mthumb",
+		"-Werror",
 		"-fshort-enums",
+		"-nostdlibinc",
+		"-Wno-macro-redefined",
 		"-fno-exceptions", "-fno-unwind-tables",
 		"-ffunction-sections", "-fdata-sections"
 	],

targets/nrf51.json

@@ -6,7 +6,7 @@
 		"--target=armv6m-none-eabi",
 		"-Qunused-arguments",
 		"-DNRF51",
-		"-Ilib/CMSIS/CMSIS/Include"
+		"-I{root}/lib/CMSIS/CMSIS/Include"
 	],
 	"ldflags": [
 		"-T", "targets/nrf51.ld"

targets/nrf52.json

@@ -7,7 +7,7 @@
 		"-mfloat-abi=soft",
 		"-Qunused-arguments",
 		"-DNRF52832_XXAA",
-		"-Ilib/CMSIS/CMSIS/Include"
+		"-I{root}/lib/CMSIS/CMSIS/Include"
 	],
 	"ldflags": [
 		"-T", "targets/nrf52.ld"

targets/nrf52840.json

@@ -7,7 +7,7 @@
 		"-mfloat-abi=soft",
 		"-Qunused-arguments",
 		"-DNRF52840_XXAA",
-		"-Ilib/CMSIS/CMSIS/Include"
+		"-I{root}/lib/CMSIS/CMSIS/Include"
 	],
 	"ldflags": [
 		"-T", "targets/nrf52840.ld"

targets/wasm.json

@@ -3,10 +3,12 @@
 	"build-tags":    ["js", "wasm"],
 	"goos":          "js",
 	"goarch":        "wasm",
-	"compiler":      "clang-8",
+	"compiler":      "clang",
 	"linker":        "wasm-ld",
 	"cflags": [
 		"--target=wasm32",
+		"-nostdlibinc",
+		"-Wno-macro-redefined",
 		"-Oz"
 	],
 	"ldflags": [

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.c

@@ -1,13 +1,17 @@
 #include "main.h"
 
 int global = 3;
-_Bool globalBool = 1;
+bool globalBool = 1;
-_Bool globalBool2 = 10; // test narrowing
+bool globalBool2 = 10; // test narrowing
 float globalFloat = 3.1;
 double globalDouble = 3.2;
 _Complex float globalComplexFloat = 4.1+3.3i;
 _Complex double globalComplexDouble = 4.2+3.4i;
 _Complex double globalComplexLongDouble = 4.3+3.5i;
+char globalChar = 100;
+void *globalVoidPtrSet = &global;
+void *globalVoidPtrNull;
+int64_t globalInt64 = -(2LL << 40);
 collection_t globalStruct = {256, -123456, 3.14, 88};
 int globalStructSize = sizeof(globalStruct);
 short globalArray[3] = {5, 6, 7};

testdata/cgo/main.go

@@ -7,11 +7,9 @@ int mul(int, int);
 */
 import "C"
 
-import "unsafe"
+import "C"
 
-func (s C.myint) Int() int {
+import "unsafe"
-	return int(s)
-}
 
 func main() {
 	println("fortytwo:", C.fortytwo())
@@ -22,6 +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)
@@ -33,6 +35,10 @@ func main() {
 	cb = C.binop_t(C.mul)
 	println("callback 2:", C.doCallback(20, 30, cb))
 
+	// equivalent types
+	var goInt8 int8 = 5
+	var _ C.int8_t = goInt8
+
 	// more globals
 	println("bool:", C.globalBool, C.globalBool2 == true)
 	println("float:", C.globalFloat)
@@ -40,6 +46,10 @@ func main() {
 	println("complex float:", C.globalComplexFloat)
 	println("complex double:", C.globalComplexDouble)
 	println("complex long double:", C.globalComplexLongDouble)
+	println("char match:", C.globalChar == 100)
+	var voidPtr unsafe.Pointer = C.globalVoidPtrNull
+	println("void* match:", voidPtr == nil, C.globalVoidPtrNull == nil, (*C.int)(C.globalVoidPtrSet) == &C.global)
+	println("int64_t match:", C.globalInt64 == C.int64_t(-(2<<40)))
 
 	// complex types
 	println("struct:", C.int(unsafe.Sizeof(C.globalStruct)) == C.globalStructSize, C.globalStruct.s, C.globalStruct.l, C.globalStruct.f)
@@ -53,10 +63,14 @@ func main() {
 	C.unionSetData(5, 8, 1)
 	println("union global data:", C.globalUnion.data[0], C.globalUnion.data[1], C.globalUnion.data[2])
 	println("union field:", printUnion(C.globalUnion).f)
+	var _ C.union_joined = C.globalUnion
+
+	// elaborated type
+	p := C.struct_point{x: 3, y: 5}
+	println("struct:", p.x, p.y)
 
 	// recursive types, test using a linked list
-	lastElement := &C.list_t{n: 7, next: nil}
+	list := &C.list_t{n: 3, next: &C.struct_list_t{n: 6, next: &C.list_t{n: 7, next: nil}}}
-	list := &C.list_t{n: 3, next: &C.struct_list_t{n: 6, next: (*C.struct_list_t)(lastElement)}}
 	for list != nil {
 		println("n in chain:", list.n)
 		list = (*C.list_t)(list.next)
@@ -66,7 +80,7 @@ func main() {
 func printUnion(union C.joined_t) C.joined_t {
 	println("union local data: ", union.data[0], union.data[1], union.data[2])
 	union.s = -33
-	println("union s method:", union.s.Int(), union.data[0] == 5)
+	println("union s:", union.data[0] == -33)
 	union.f = 6.28
 	println("union f:", union.f)
 	return union

testdata/cgo/main.h

@@ -1,10 +1,25 @@
+#include <stdbool.h>
+#include <stdint.h>
+
 typedef short myint;
+typedef short unusedTypedef;
 int add(int a, int b);
+int unusedFunction(void);
 typedef int (*binop_t) (int, int);
 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);
+
 typedef struct collection {
 	short         s;
 	long          l;
@@ -12,6 +27,11 @@ typedef struct collection {
 	unsigned char c;
 } collection_t;
 
+struct point {
+	int x;
+	int y;
+};
+
 // linked list
 typedef struct list_t {
 	int           n;
@@ -27,15 +47,20 @@ void unionSetShort(short s);
 void unionSetFloat(float f);
 void unionSetData(short f0, short f1, short f2);
 
-// test globals
+// test globals and datatypes
 extern int global;
-extern _Bool globalBool;
+extern int unusedGlobal;
-extern _Bool globalBool2;
+extern bool globalBool;
+extern bool globalBool2;
 extern float globalFloat;
 extern double globalDouble;
 extern _Complex float globalComplexFloat;
 extern _Complex double globalComplexDouble;
 extern _Complex double globalComplexLongDouble;
+extern char globalChar;
+extern void *globalVoidPtrSet;
+extern void *globalVoidPtrNull;
+extern int64_t globalInt64;
 extern collection_t globalStruct;
 extern int globalStructSize;
 extern short globalArray[3];

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
@@ -14,6 +18,9 @@ double: +3.200000e+000
 complex float: (+4.100000e+000+3.300000e+000i)
 complex double: (+4.200000e+000+3.400000e+000i)
 complex long double: (+4.300000e+000+3.500000e+000i)
+char match: true
+void* match: true true true
+int64_t match: true
 struct: true 256 -123456 +3.140000e+000
 array: 5 6 7
 union: true
@@ -21,9 +28,10 @@ union s: 22
 union f: +3.140000e+000
 union global data: 5 8 1
 union local data:  5 8 1
-union s method: -33 false
+union s: true
 union f: +6.280000e+000
 union field: +6.280000e+000
+struct: 3 5
 n in chain: 3
 n in chain: 6
 n in chain: 7

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
@@ -20,6 +20,11 @@ func main() {
 	n, ok = <-ch
 	println("recv from closed channel:", n, ok)
 
+	// Test bigger values
+	ch2 := make(chan complex128)
+	go sendComplex(ch2)
+	println("complex128:", <-ch2)
+
 	// Test multi-sender.
 	ch = make(chan int)
 	go fastsender(ch)
@@ -62,6 +67,10 @@ func sender(ch chan int) {
 	close(ch)
 }
 
+func sendComplex(ch chan complex128) {
+	ch <- 7+10.5i
+}
+
 func fastsender(ch chan int) {
 	ch <- 10
 	ch <- 11

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
@@ -9,6 +9,7 @@ received num: 6
 received num: 7
 received num: 8
 recv from closed channel: 0 false
+complex128: (+7.000000e+000+1.050000e+001i)
 got n: 10
 got n: 11
 got n: 10

testdata/coroutines.go

@@ -15,12 +15,19 @@ func main() {
 	wait()
 	println("end waiting")
 
+	value := delayedValue()
+	println("value produced after some time:", value)
+
 	// Run a non-blocking call in a goroutine. This should be turned into a
 	// regular call, so should be equivalent to calling nowait() without 'go'
 	// prefix.
 	go nowait()
 	time.Sleep(time.Millisecond)
 	println("done with non-blocking goroutine")
+
+	var printer Printer
+	printer = &myPrinter{}
+	printer.Print()
 }
 
 func sub() {
@@ -35,6 +42,23 @@ func wait() {
 	println("  wait end")
 }
 
+func delayedValue() int {
+	time.Sleep(time.Millisecond)
+	return 42
+}
+
 func nowait() {
 	println("non-blocking goroutine")
 }
+
+type Printer interface {
+	Print()
+}
+
+type myPrinter struct{
+}
+
+func (i *myPrinter) Print() {
+	time.Sleep(time.Millisecond)
+	println("async interface method call")
+}

testdata/coroutines.txt

@@ -7,5 +7,7 @@ wait:
   wait start
   wait end
 end waiting
+value produced after some time: 42
 non-blocking goroutine
 done with non-blocking goroutine
+async interface method call

testdata/float.go

@@ -56,4 +56,16 @@ func main() {
 	// cast complex
 	println(complex64(c128))
 	println(complex128(c64))
+
+	// binops on complex numbers
+	c64 = 5+2i
+	println("complex64 add: ", c64 + -3+8i)
+	println("complex64 sub: ", c64 - -3+8i)
+	println("complex64 mul: ", c64 * -3+8i)
+	println("complex64 div: ", c64 / -3+8i)
+	c128 = -5+2i
+	println("complex128 add:", c128 + 2+6i)
+	println("complex128 sub:", c128 - 2+6i)
+	println("complex128 mul:", c128 * 2+6i)
+	println("complex128 div:", c128 / 2+6i)
 }

testdata/float.txt

@@ -23,3 +23,11 @@
 (+2.000000e+000-2.000000e+000i)
 (+6.666667e-001-2.000000e+000i)
 (+6.666667e-001+1.200000e+000i)
+complex64 add:  (+2.000000e+000+1.000000e+001i)
+complex64 sub:  (+8.000000e+000+1.000000e+001i)
+complex64 mul:  (-1.500000e+001+2.000000e+000i)
+complex64 div:  (-1.666667e+000+7.333333e+000i)
+complex128 add: (-3.000000e+000+8.000000e+000i)
+complex128 sub: (-7.000000e+000+8.000000e+000i)
+complex128 mul: (-1.000000e+001+1.000000e+001i)
+complex128 div: (-2.500000e+000+7.000000e+000i)

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: