Adding extra files for testing multiline, globbing, etc.

These two heaps conflict with each other, so that if any function uses
the dlmalloc heap implementation it will eventually result in memory
corruption.

This commit fixes this by implementing all heap-related functions. This
overrides the functions that are implemented in wasi-libc. That's why
all of them are implemented (even if they just panic): to make sure no
program accidentally uses the wrong one.

.circleci/config.yml

@@ -14,23 +14,21 @@ commands:
       - run:
           name: "Install apt dependencies"
           command: |
-            echo 'deb http://apt.llvm.org/stretch/ llvm-toolchain-stretch<<parameters.llvm>> main' | sudo tee /etc/apt/sources.list.d/llvm.list
+            echo 'deb https://apt.llvm.org/buster/ llvm-toolchain-buster-<<parameters.llvm>> main' | sudo tee /etc/apt/sources.list.d/llvm.list
             wget -O - https://apt.llvm.org/llvm-snapshot.gpg.key|sudo apt-key add -
             sudo apt-get update
             sudo apt-get install \
-                python3 \
-                llvm<<parameters.llvm>>-dev \
-                clang<<parameters.llvm>> \
-                libclang<<parameters.llvm>>-dev \
-                lld<<parameters.llvm>> \
+                llvm-<<parameters.llvm>>-dev \
+                clang-<<parameters.llvm>> \
+                libclang-<<parameters.llvm>>-dev \
+                lld-<<parameters.llvm>> \
                 gcc-arm-linux-gnueabihf \
-                libc6-dev-armel-cross \
                 gcc-aarch64-linux-gnu \
-                libc6-dev-arm64-cross \
                 qemu-system-arm \
                 qemu-user \
                 gcc-avr \
                 avr-libc
+            sudo apt-get install --no-install-recommends libc6-dev-i386 lib32gcc-8-dev
   install-node:
     steps:
       - run:
@@ -40,52 +38,106 @@ commands:
             sudo tar -C /usr/local -xf node-v10.15.1-linux-x64.tar.xz
             sudo ln -s /usr/local/node-v10.15.1-linux-x64/bin/node /usr/bin/node
             rm node-v10.15.1-linux-x64.tar.xz
+  install-chrome:
+    steps:
+      - run:
+          name: "Install Chrome"
+          command: |
+            wget https://dl.google.com/linux/direct/google-chrome-stable_current_amd64.deb
+            sudo apt install ./google-chrome-stable_current_amd64.deb
+  install-wasmtime:
+    steps:
+      - run:
+          name: "Install wasmtime"
+          command: |
+            curl https://wasmtime.dev/install.sh -sSf | bash
+            sudo ln -s ~/.wasmtime/bin/wasmtime /usr/local/bin/wasmtime
+  install-xtensa-toolchain:
+    parameters:
+      variant:
+        type: string
+    steps:
+      - run:
+          name: "Install Xtensa toolchain"
+          command: |
+            curl -L https://github.com/espressif/crosstool-NG/releases/download/esp-2020r2/xtensa-esp32-elf-gcc8_2_0-esp-2020r2-<<parameters.variant>>.tar.gz -o xtensa-esp32-elf-gcc8_2_0-esp-2020r2-<<parameters.variant>>.tar.gz
+            sudo tar -C /usr/local -xf xtensa-esp32-elf-gcc8_2_0-esp-2020r2-<<parameters.variant>>.tar.gz
+            sudo ln -s /usr/local/xtensa-esp32-elf/bin/xtensa-esp32-elf-ld /usr/local/bin/xtensa-esp32-elf-ld
+            rm xtensa-esp32-elf-gcc8_2_0-esp-2020r2-<<parameters.variant>>.tar.gz
   llvm-source-linux:
     steps:
       - restore_cache:
           keys:
-            - llvm-source-8-v5
+            - llvm-source-11-v2
       - run:
           name: "Fetch LLVM source"
           command: make llvm-source
       - save_cache:
-          key: llvm-source-8-v5
+          key: llvm-source-11-v2
           paths:
-            - llvm-project
+            - llvm-project/clang/lib/Headers
+            - llvm-project/clang/include
+            - llvm-project/lld/include
+            - llvm-project/llvm/include
+  build-wasi-libc:
+    steps:
+      - restore_cache:
+          keys:
+            - wasi-libc-sysroot-v4
+      - run:
+          name: "Build wasi-libc"
+          command: make wasi-libc
+      - save_cache:
+          key: wasi-libc-sysroot-v4
+          paths:
+            - lib/wasi-libc/sysroot
   test-linux:
+    parameters:
+      llvm:
+        type: string
     steps:
       - checkout
       - submodules
       - apt-dependencies:
-          llvm: "-8"
+          llvm: "<<parameters.llvm>>"
       - install-node
+      - install-chrome
+      - install-wasmtime
       - restore_cache:
           keys:
             - go-cache-v2-{{ checksum "go.mod" }}-{{ .Environment.CIRCLE_PREVIOUS_BUILD_NUM }}
             - go-cache-v2-{{ checksum "go.mod" }}
       - llvm-source-linux
-      - run: go install .
-      - run: go test -v ./transform .
+      - run: go install -tags=llvm<<parameters.llvm>> .
+      - restore_cache:
+          keys:
+            - wasi-libc-sysroot-systemclang-v3
+      - run: make wasi-libc
+      - save_cache:
+          key: wasi-libc-sysroot-systemclang-v3
+          paths:
+            - lib/wasi-libc/sysroot
+      - run: go test -v -tags=llvm<<parameters.llvm>> ./cgo ./compileopts ./compiler ./interp ./transform .
       - run: make gen-device -j4
-      - run: make smoketest RISCV=0
+      - run: make smoketest XTENSA=0
+      - run: make tinygo-test
+      - run: make wasmtest
       - save_cache:
           key: go-cache-v2-{{ checksum "go.mod" }}-{{ .Environment.CIRCLE_BUILD_NUM }}
           paths:
             - ~/.cache/go-build
-            - ~/.cache/tinygo
             - /go/pkg/mod
       - run: make fmt-check
-  build-linux:
+  assert-test-linux:
     steps:
       - checkout
       - submodules
       - run:
           name: "Install apt dependencies"
           command: |
+            sudo apt-get update
             sudo apt-get install \
-                python3 \
                 gcc-arm-linux-gnueabihf \
-                binutils-arm-none-eabi \
                 libc6-dev-armel-cross \
                 gcc-aarch64-linux-gnu \
                 libc6-dev-arm64-cross \
@@ -93,7 +145,11 @@ commands:
                 qemu-user \
                 gcc-avr \
                 avr-libc
+            sudo apt-get install --no-install-recommends libc6-dev-i386 lib32gcc-6-dev
       - install-node
+      - install-wasmtime
+      - install-xtensa-toolchain:
+          variant: "linux-amd64"
       - restore_cache:
           keys:
             - go-cache-v2-{{ checksum "go.mod" }}-{{ .Environment.CIRCLE_PREVIOUS_BUILD_NUM }}
@@ -101,48 +157,119 @@ commands:
       - llvm-source-linux
       - restore_cache:
           keys:
-            - llvm-build-8-linux-v7
+            - llvm-build-11-linux-v4-assert
       - run:
           name: "Build LLVM"
           command: |
             if [ ! -f llvm-build/lib/liblldELF.a ]
             then
+              # fetch LLVM source
+              rm -rf llvm-project
+              make llvm-source
               # install dependencies
-              sudo apt-get install cmake clang ninja-build
-              # make build faster
-              export CC=clang
-              export CXX=clang++
+              sudo apt-get install cmake ninja-build
+              # hack ninja to use less jobs
+              echo -e '#!/bin/sh\n/usr/bin/ninja -j3 "$@"' > /go/bin/ninja
+              chmod +x /go/bin/ninja
+              # build!
+              make ASSERT=1 llvm-build
+              find llvm-build -name CMakeFiles -prune -exec rm -r '{}' \;
+            fi
+      - save_cache:
+          key: llvm-build-11-linux-v4-assert
+          paths:
+            llvm-build
+      - run: make ASSERT=1
+      - build-wasi-libc
+      - run:
+          name: "Test TinyGo"
+          command: make ASSERT=1 test
+          environment:
+            # Note: -p=2 limits parallelism to two jobs at a time, which is
+            # necessary to keep memory consumption down and avoid OOM (for a
+            # 2CPU/4GB executor).
+            GOFLAGS: -p=2
+      - save_cache:
+          key: go-cache-v2-{{ checksum "go.mod" }}-{{ .Environment.CIRCLE_BUILD_NUM }}
+          paths:
+            - ~/.cache/go-build
+            - /go/pkg/mod
+      - run: make gen-device -j4
+      - run: make smoketest TINYGO=build/tinygo
+  build-linux:
+    steps:
+      - checkout
+      - submodules
+      - run:
+          name: "Install apt dependencies"
+          command: |
+            sudo apt-get update
+            sudo apt-get install \
+                gcc-arm-linux-gnueabihf \
+                libc6-dev-armel-cross \
+                gcc-aarch64-linux-gnu \
+                libc6-dev-arm64-cross \
+                qemu-system-arm \
+                qemu-user \
+                gcc-avr \
+                avr-libc
+            sudo apt-get install --no-install-recommends libc6-dev-i386 lib32gcc-6-dev
+      - install-node
+      - install-wasmtime
+      - install-xtensa-toolchain:
+          variant: "linux-amd64"
+      - restore_cache:
+          keys:
+            - go-cache-v2-{{ checksum "go.mod" }}-{{ .Environment.CIRCLE_PREVIOUS_BUILD_NUM }}
+            - go-cache-v2-{{ checksum "go.mod" }}
+      - llvm-source-linux
+      - restore_cache:
+          keys:
+            - llvm-build-11-linux-v4-noassert
+      - run:
+          name: "Build LLVM"
+          command: |
+            if [ ! -f llvm-build/lib/liblldELF.a ]
+            then
+              # fetch LLVM source
+              rm -rf llvm-project
+              make llvm-source
+              # install dependencies
+              sudo apt-get install cmake ninja-build
               # hack ninja to use less jobs
               echo -e '#!/bin/sh\n/usr/bin/ninja -j3 "$@"' > /go/bin/ninja
               chmod +x /go/bin/ninja
               # build!
               make llvm-build
+              find llvm-build -name CMakeFiles -prune -exec rm -r '{}' \;
             fi
       - save_cache:
-          key: llvm-build-8-linux-v7
+          key: llvm-build-11-linux-v4-noassert
           paths:
             llvm-build
-      - run:
-          name: "Create LLVM symlinks"
-          command: |
-            ln -s $PWD/llvm-build/bin/clang-8 /go/bin/clang-8
-            ln -s $PWD/llvm-build/bin/ld.lld  /go/bin/ld.lld-8
-            ln -s $PWD/llvm-build/bin/wasm-ld /go/bin/wasm-ld-8
+      - build-wasi-libc
       - run:
           name: "Test TinyGo"
           command: make test
+      - run:
+          name: "Install fpm"
+          command: |
+            sudo apt-get install ruby ruby-dev
+            sudo gem install --no-document fpm
       - run:
           name: "Build TinyGo release"
           command: |
-            make release -j3
+            make release deb -j3
             cp -p build/release.tar.gz /tmp/tinygo.linux-amd64.tar.gz
+            cp -p build/release.deb    /tmp/tinygo_amd64.deb
       - store_artifacts:
           path: /tmp/tinygo.linux-amd64.tar.gz
+      - store_artifacts:
+          path: /tmp/tinygo_amd64.deb
       - save_cache:
           key: go-cache-v2-{{ checksum "go.mod" }}-{{ .Environment.CIRCLE_BUILD_NUM }}
           paths:
             - ~/.cache/go-build
-            - ~/.cache/tinygo
             - /go/pkg/mod
       - run:
           name: "Extract release tarball"
@@ -151,11 +278,6 @@ commands:
             tar -C ~/lib -xf /tmp/tinygo.linux-amd64.tar.gz
             ln -s ~/lib/tinygo/bin/tinygo /go/bin/tinygo
             tinygo version
-      - run:
-          name: "Download SiFive GNU toolchain"
-          command: |
-            curl -O https://static.dev.sifive.com/dev-tools/riscv64-unknown-elf-gcc-8.2.0-2019.05.3-x86_64-linux-ubuntu14.tar.gz
-            sudo tar -C /usr/local --strip-components=1 -xf riscv64-unknown-elf-gcc-8.2.0-2019.05.3-x86_64-linux-ubuntu14.tar.gz
       - run: make smoketest
   build-macos:
     steps:
@@ -164,45 +286,60 @@ commands:
       - run:
           name: "Install dependencies"
           command: |
-            curl https://dl.google.com/go/go1.12.5.darwin-amd64.tar.gz -o go1.12.5.darwin-amd64.tar.gz
-            sudo tar -C /usr/local -xzf go1.12.5.darwin-amd64.tar.gz
+            curl https://dl.google.com/go/go1.16.darwin-amd64.tar.gz -o go1.16.darwin-amd64.tar.gz
+            sudo tar -C /usr/local -xzf go1.16.darwin-amd64.tar.gz
             ln -s /usr/local/go/bin/go /usr/local/bin/go
             HOMEBREW_NO_AUTO_UPDATE=1 brew install qemu
+      - install-xtensa-toolchain:
+          variant: "macos"
       - restore_cache:
           keys:
-            - go-cache-macos-v2-{{ checksum "go.mod" }}-{{ .Environment.CIRCLE_PREVIOUS_BUILD_NUM }}
-            - go-cache-macos-v2-{{ checksum "go.mod" }}
+            - go-cache-macos-v3-{{ checksum "go.mod" }}-{{ .Environment.CIRCLE_PREVIOUS_BUILD_NUM }}
+            - go-cache-macos-v3-{{ checksum "go.mod" }}
       - restore_cache:
           keys:
-            - llvm-source-8-macos-v5
+            - llvm-source-11-macos-v3
       - run:
           name: "Fetch LLVM source"
           command: make llvm-source
       - save_cache:
-          key: llvm-source-8-macos-v5
+          key: llvm-source-11-macos-v3
           paths:
-            - llvm-project
+            - llvm-project/clang/lib/Headers
+            - llvm-project/clang/include
+            - llvm-project/lld/include
+            - llvm-project/llvm/include
       - restore_cache:
           keys:
-            - llvm-build-8-macos-v6
+            - llvm-build-11-macos-v5
       - run:
           name: "Build LLVM"
           command: |
             if [ ! -f llvm-build/lib/liblldELF.a ]
             then
+              # fetch LLVM source
+              rm -rf llvm-project
+              make llvm-source
               # install dependencies
               HOMEBREW_NO_AUTO_UPDATE=1 brew install cmake ninja
               # build!
               make llvm-build
+              find llvm-build -name CMakeFiles -prune -exec rm -r '{}' \;
             fi
       - save_cache:
-          key: llvm-build-8-macos-v6
+          key: llvm-build-11-macos-v5
           paths:
             llvm-build
+      - restore_cache:
+          keys:
+            - wasi-libc-sysroot-macos-v4
       - run:
-          name: "Create LLVM symlinks"
-          command: |
-            ln -s $PWD/llvm-build/bin/clang-8 /usr/local/bin/clang-8
+          name: "Build wasi-libc"
+          command: make wasi-libc
+      - save_cache:
+          key: wasi-libc-sysroot-macos-v4
+          paths:
+            - lib/wasi-libc/sysroot
       - run:
           name: "Test TinyGo"
           command: make test
@@ -220,49 +357,63 @@ commands:
             tar -C /usr/local/opt -xf /tmp/tinygo.darwin-amd64.tar.gz
             ln -s /usr/local/opt/tinygo/bin/tinygo /usr/local/bin/tinygo
             tinygo version
-      - run:
-          name: "Download SiFive GNU toolchain"
-          command: |
-            curl -O https://static.dev.sifive.com/dev-tools/riscv64-unknown-elf-gcc-8.2.0-2019.05.3-x86_64-apple-darwin.tar.gz
-            sudo tar -C /usr/local --strip-components=1 -xf riscv64-unknown-elf-gcc-8.2.0-2019.05.3-x86_64-apple-darwin.tar.gz
       - run: make smoketest AVR=0
       - save_cache:
-          key: go-cache-macos-v2-{{ checksum "go.mod" }}-{{ .Environment.CIRCLE_BUILD_NUM }}
+          key: go-cache-macos-v3-{{ checksum "go.mod" }}-{{ .Environment.CIRCLE_BUILD_NUM }}
           paths:
             - ~/.cache/go-build
-            - ~/.cache/tinygo
             - /go/pkg/mod
 
-
 jobs:
-  test-llvm8-go111:
+  test-llvm10-go113:
     docker:
-      - image: circleci/golang:1.11-stretch
+      - image: circleci/golang:1.13-buster
     steps:
-      - test-linux
-  test-llvm8-go112:
+      - test-linux:
+          llvm: "10"
+  test-llvm10-go114:
     docker:
-      - image: circleci/golang:1.12-stretch
+      - image: circleci/golang:1.14-buster
     steps:
-      - test-linux
+      - test-linux:
+          llvm: "10"
+  test-llvm11-go115:
+    docker:
+      - image: circleci/golang:1.15-buster
+    steps:
+      - test-linux:
+          llvm: "11"
+  test-llvm11-go116:
+    docker:
+      - image: circleci/golang:1.16-buster
+    steps:
+      - test-linux:
+          llvm: "11"
+  assert-test-linux:
+    docker:
+      - image: circleci/golang:1.14-stretch
+    steps:
+      - assert-test-linux
   build-linux:
     docker:
-      - image: circleci/golang:1.12-stretch
+      - image: circleci/golang:1.14-stretch
     steps:
       - build-linux
   build-macos:
     macos:
-      xcode: "10.1.0"
+      xcode: "11.1.0" # macOS 10.14
     steps:
       - build-macos
 
 
 
-
 workflows:
   test-all:
     jobs:
-      - test-llvm8-go111
-      - test-llvm8-go112
+      - test-llvm10-go113
+      - test-llvm10-go114
+      - test-llvm11-go115
+      - test-llvm11-go116
       - build-linux
       - build-macos
+      - assert-test-linux

.github/workflows/build-tinygo-dev-docker.yml

@@ -0,0 +1,45 @@
+name: CI for tinygo-dev docker container
+on:
+  push:
+    branches: [ dev ]
+
+jobs:
+  push_to_registry:
+    name: Push Docker image to GHCR/Docker Hub
+    runs-on: ubuntu-latest
+    permissions:
+      packages: write
+      contents: read
+    steps:
+      - name: Check out the repo
+        uses: actions/checkout@v2
+        with:
+          submodules: recursive
+      - name: Docker meta
+        id: meta
+        uses: docker/metadata-action@v3
+        with:
+          images: |
+            tinygo/tinygo-dev
+            ghcr.io/${{ github.repository }}/tinygo-dev
+          tags: |
+            type=sha,format=long
+            type=raw,value=latest
+      - name: Log in to Docker Hub
+        uses: docker/login-action@v1
+        with:
+          username: ${{ secrets.DOCKER_HUB_USERNAME }}
+          password: ${{ secrets.DOCKER_HUB_ACCESS_TOKEN }}
+      - name: Log in to Github Container Registry
+        uses: docker/login-action@v1
+        with:
+          registry: ghcr.io
+          username: ${{ github.actor }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+      - name: Build and push
+        uses: docker/build-push-action@v2
+        with:
+          context: .
+          push: true
+          tags: ${{ steps.meta.outputs.tags }}
+          labels: ${{ steps.meta.outputs.labels }}

.gitignore

@@ -3,14 +3,28 @@ docs/_build
 src/device/avr/*.go
 src/device/avr/*.ld
 src/device/avr/*.s
+src/device/esp/*.go
 src/device/nrf/*.go
 src/device/nrf/*.s
+src/device/nxp/*.go
+src/device/nxp/*.s
 src/device/sam/*.go
 src/device/sam/*.s
 src/device/sifive/*.go
 src/device/sifive/*.s
 src/device/stm32/*.go
 src/device/stm32/*.s
+src/device/kendryte/*.go
+src/device/kendryte/*.s
+src/device/rp/*.go
+src/device/rp/*.s
 vendor
-llvm
 llvm-build
+llvm-project
+
+# Ignore files generated by smoketest
+test.gba
+test.hex
+test.nro
+test.wasm
+wasm.wasm

.gitmodules

@@ -14,3 +14,12 @@
 	path = lib/compiler-rt
 	url = https://github.com/llvm-mirror/compiler-rt.git
 	branch = release_80
+[submodule "lib/wasi-libc"]
+	path = lib/wasi-libc
+	url = https://github.com/CraneStation/wasi-libc
+[submodule "lib/picolibc"]
+	path = lib/picolibc
+	url = https://github.com/keith-packard/picolibc.git
+[submodule "lib/stm32-svd"]
+	path = lib/stm32-svd
+	url = https://github.com/tinygo-org/stm32-svd

BUILDING.md

@@ -10,12 +10,20 @@ This guide describes how to statically link TinyGo against LLVM, libclang and
 lld so that the binary can be easily moved between systems. It also shows how to
 build a release tarball that includes this binary and all necessary extra files.
 
+**Note**: this documentation describes how to build a statically linked release
+tarball. If you want to develop TinyGo, you will probably want to follow a
+different guide:
+
+  * [Linux](https://tinygo.org/getting-started/linux/#source-install)
+  * [macOS](https://tinygo.org/getting-started/macos/#source-install)
+  * [Windows](https://tinygo.org/getting-started/windows/#source-install)
+
 ## Dependencies
 
 LLVM, Clang and LLD are quite light on dependencies, requiring only standard
 build tools to be built. Go is of course necessary to build TinyGo itself.
 
-  * Go (1.11+)
+  * Go (1.13+)
   * Standard build tools (gcc/clang)
   * git
   * CMake
@@ -35,11 +43,6 @@ You can also store LLVM outside of the TinyGo root directory by setting the
 `LLVM_BUILDDIR`, `CLANG_SRC` and `LLD_SRC` make variables, but that is not
 covered by this guide.
 
-TinyGo uses Go modules, so if you clone TinyGo inside your GOPATH (and are using
-Go below 1.13), make sure that Go modules are enabled:
-
-    export GO111MODULE=on
-
 ## Build LLVM, Clang, LLD
 
 Before starting the build, you may want to set the following environment
@@ -84,6 +87,10 @@ Now that we have a working static build, it's time to make a release tarball:
 
     make release
 
+If you did not clone the repository with the `--recursive` option, you will get errors until you initialize the project submodules:
+
+    git submodule update --init
+
 The release tarball is stored in build/release.tar.gz, and can be extracted with
 the following command (for example in ~/lib):
 

CHANGELOG.md

@@ -1,3 +1,784 @@
+0.19.0
+---
+
+* **command line**
+  - don't consider compile-only tests as failing
+  - add -test flag for `tinygo list`
+  - escape commands while printing them with the -x flag
+  - make flash-command portable and safer to use
+  - use `extended-remote` instead of `remote` in GDB
+  - detect specific serial port IDs based on USB vid/pid
+  - add a flag to the command line to select the serial implementation
+* **compiler**
+  - `cgo`: improve constant parser
+  - `compiler`: support chained interrupt handlers
+  - `compiler`: add support for running a builtin in a goroutine
+  - `compiler`: do not emit nil checks for loading closure variables
+  - `compiler`: skip context parameter when starting regular goroutine
+  - `compiler`: refactor method names
+  - `compiler`: add function and global section pragmas
+  - `compiler`: implement `syscall.rawSyscallNoError` in inline assembly
+  - `interp`: ignore inline assembly in markExternal
+  - `interp`: fix a bug in pointer cast workaround
+  - `loader`: fix testing a main package
+* **standard library**
+  - `crypto/rand`: replace this package with a TinyGo version
+  - `machine`: make USBCDC global a pointer
+  - `machine`: make UART objects pointer receivers
+  - `machine`: define Serial as the default output
+  - `net`: add initial support for net.IP
+  - `net`: add more net compatibility
+  - `os`: add stub for os.ReadDir
+  - `os`: add FileMode constants from Go 1.16
+  - `os`: add stubs required for net/http
+  - `os`: implement process related functions
+  - `reflect`: implement AppendSlice
+  - `reflect`: add stubs required for net/http
+  - `runtime`: make task.Data a 64-bit integer to avoid overflow
+  - `runtime`: expose memory stats
+  - `sync`: implement NewCond
+  - `syscall`: fix int type in libc version
+* **targets**
+  - `cortexm`: do not disable interrupts on abort
+  - `cortexm`: bump default stack size to 2048 bytes
+  - `nrf`: avoid heap allocation in waitForEvent
+  - `nrf`: don't trigger a heap allocation in SPI.Transfer
+  - `nrf52840`: add support for flashing with the BOSSA tool
+  - `rp2040`: add support for GPIO input
+  - `rp2040`: add basic support for ADC
+  - `rp2040`: gpio and adc pin definitions
+  - `rp2040`: implement UART
+  - `rp2040`: patch elf to checksum 2nd stage boot
+  - `stm32`: add PWM for most chips
+  - `stm32`: add support for pin interrupts
+  - `stm32f103`: add support for PinInputPullup / PinInputPulldown
+  - `wasi`: remove wasm build tag
+* **boards**
+  - `feather-rp2040`: add support for this board
+  - `feather-nrf52840-sense`: add board definition for this board
+  - `pca10059`: support flashing from Windows
+  - `nano-rp2040`: add this board
+  - `nano-33-ble`: add support for this board
+  - `pico`: add the Raspberry Pi Pico board with the new RP2040 chip
+  - `qtpy`: add pin for neopixels
+  - all: add definition for ws2812 for supported boards
+
+
+0.18.0
+---
+
+* **command line**
+  - drop support for Go 1.11 and 1.12
+  - throw an error when no target is specified on Windows
+  - improve error messages in `getDefaultPort()`, support for multiple ports
+  - remove `-cflags` and `-ldflags` flags
+  - implement `-ldflags="-X ..."`
+  - add `-print-allocs` flag that lets you print all heap allocations
+  - openocd commands in tinygo command line
+  - add `-llvm-features` parameter
+  - match `go test` output
+  - discover USB ports only, this will ignore f.ex. bluetooth
+  - use physicmal path instead of cached GOROOT in function getGoroot
+  - add goroot for snap installs
+* **compiler**
+  - `builder`: add support for `-opt=0`
+  - `builder`, `compiler`: compile and cache packages in parallel
+  - `builder`: run interp per package
+  - `builder`: cache C and assembly file outputs
+  - `builder`: add support for `-x` flag to print commands
+  - `builder`: add optsize attribute while building the package
+  - `builder`: run function passes per package
+  - `builder`: hard code Clang compiler
+  - `compiler`: do not use `llvm.GlobalContext()`
+  - `compiler`: remove SimpleDCE pass
+  - `compiler`: do not emit nil checks for `*ssa.Alloc` instructions
+  - `compiler`: merge `runtime.typecodeID` and runtime.typeInInterface
+  - `compiler`: do not check for impossible type asserts
+  - `compiler`: fix use of global context: `llvm.Int32Type()`
+  - `compiler`: add interface IR test
+  - `compiler`: fix lack of method name in interface matching
+  - `compiler`: fix "fragment covers entire variable" bug
+  - `compiler`: optimize string literals and globals
+  - `compiler`: decouple func lowering from interface type codes
+  - `compiler`: add function attributes to some runtime calls
+  - `compiler`: improve position information in error messages
+  - `cgo`: add support for CFLAGS in .c files
+  - `interp`: support GEP on fixed (MMIO) addresses
+  - `interp`: handle `(reflect.Type).Elem()`
+  - `interp`: add support for runtime.interfaceMethod
+  - `interp`: make toLLVMValue return an error instead of panicking
+  - `interp`: add support for switch statement
+  - `interp`: fix phi instruction
+  - `interp`: remove map support
+  - `interp`: support extractvalue/insertvalue with multiple operands
+  - `transform`: optimize string comparisons against ""
+  - `transform`: optimize `reflect.Type` `Implements()` method
+  - `transform`: fix bug in interface lowering when signatures are renamed
+  - `transform`: don't rely on struct name of `runtime.typecodeID`
+  - `transform`: use IPSCCP pass instead of the constant propagation pass
+  - `transform`: fix func lowering assertion failure
+  - `transform`: do not lower zero-sized alloc to alloca
+  - `transform`: split interface and reflect lowering
+* **standard library**
+  - `runtime`: add dummy debug package
+  - `machine`: fix data shift/mask in newUSBSetup
+  - `machine`: make `machine.I2C0` and similar objects pointers
+  - `machine`: unify usbcdc code
+  - `machine`: refactor PWM support
+  - `machine`: avoid heap allocations in USB code
+  - `reflect`: let `reflect.Type` be of interface type
+  - `reflect`: implement a number of stub functions
+  - `reflect`: check for access in the `Interface` method call
+  - `reflect`: fix `AssignableTo` and `Implements` methods
+  - `reflect`: implement `Value.CanAddr`
+  - `reflect`: implement `Sizeof` and `Alignof` for func values
+  - `reflect`: implement `New` function
+  - `runtime`: implement command line arguments in hosted environments
+  - `runtime`: implement environment variables for Linux
+  - `runtime`: improve timers on nrf, and samd chips
+* **targets**
+  - all: use -Qunused-arguments only for assembly files
+  - `atmega1280`: add PWM support
+  - `attiny`: remove dummy UART
+  - `atsamd21`: improve SPI
+  - `atsamd51`: fix PWM support in atsamd51p20
+  - `atsamd5x`: improve SPI
+  - `atsamd51`, `atsame5x`: unify samd51 and same5x
+  - `atsamd51`, `atsamd21`: fix `ADC.Get()` value at 8bit and 10bit
+  - `atsame5x`: add support for CAN
+  - `avr`: remove I2C stubs from attiny support
+  - `cortexm`: check for `arm-none-eabi-gdb` and `gdb-multiarch` commands
+  - `cortexm`: add `__isr_vector` symbol
+  - `cortexm`: disable FPU on Cortex-M4
+  - `cortexm`: clean up Cortex-M target files
+  - `fe310`: fix SPI read
+  - `gameboy-advance`: Fix RGBA color interpretation
+  - `nrf52833`: add PWM support
+  - `stm32l0`: use unified UART logic
+  - `stm32`: move f103 (bluepill) to common i2c code
+  - `stm32`: separate altfunc selection for UART Tx/Rx
+  - `stm32`: i2c implementation for F7, L5 and L4 MCUs
+  - `stm32`: make SPI CLK fast to fix data issue
+  - `stm32`: support SPI on L4 series
+  - `unix`: avoid possible heap allocation with `-opt=0`
+  - `unix`: use conservative GC by default
+  - `unix`: use the tasks scheduler instead of coroutines
+  - `wasi`: upgrade WASI version to wasi_snapshot_preview1
+  - `wasi`: darwin: support basic file io based on libc
+  - `wasm`: only export explicitly exported functions
+  - `wasm`: use WASI ABI for exit function
+  - `wasm`: scan globals conservatively
+* **boards**
+  - `arduino-mega1280`: add support for the Arduino Mega 1280
+  - `arduino-nano-new`: Add Arduino Nano w/ New Bootloader target
+  - `atsame54-xpro`: add initial support this board
+  - `feather-m4-can`: add initial support for this board
+  - `grandcentral-m4`: add board support for Adafruit Grand Central M4 (SAMD51)
+  - `lgt92`: update to new UART structure
+  - `microbit`: remove LED constant
+  - `microbit-v2`: add support for S113 SoftDevice
+  - `nucleol432`: add support for this board
+  - `nucleo-l031k6`: add this board
+  - `pca10059`: initial support for this board
+  - `qtpy`: fix msd-volume-name
+  - `qtpy`: fix i2c setting
+  - `teensy40`: move txBuffer allocation to UART declaration
+  - `teensy40`: add UART0 as alias for UART1
+
+
+0.17.0
+
+---
+* **command line**
+  - switch to LLVM 11 for static builds
+  - support gdb debugging with AVR
+  - add support for additional openocd commands
+  - add `-x` flag to print commands
+  - use LLVM 11 by default when linking LLVM dynamically
+  - update go-llvm to use LLVM 11 on macOS
+  - bump go.bug.st/serial to version 1.1.2
+  - do not build LLVM with libxml to work around a bugo on macOS
+  - add support for Go 1.16
+  - support gdb daemonization on Windows
+  - remove support for LLVM 9, to fix CI
+  - kill OpenOCD if it does not exit with a regular quit signal
+  - support `-ocd-output` on Windows
+* **compiler**
+  - `builder`: parallelize most of the build
+  - `builder`: remove unused cacheKey parameter
+  - `builder`: add -mcpu flag while building libraries
+  - `builder`: wait for running jobs to finish
+  - `cgo`: add support for variadic functions
+  - `compiler`: fix undefined behavior in wordpack
+  - `compiler`: fix incorrect "exported function" panic
+  - `compiler`: fix non-int integer constants (fixing a crash)
+  - `compiler`: refactor and add tests
+  - `compiler`: emit a nil check when slicing an array pointer
+  - `compiler`: saturate float-to-int conversions
+  - `compiler`: test float to int conversions and fix upper-bound calculation
+  - `compiler`: support all kinds of deferred builtins
+  - `compiler`: remove ir package
+  - `compiler`: remove unnecessary main.main call workaround
+  - `compiler`: move the setting of attributes to getFunction
+  - `compiler`: create runtime types lazily when needed
+  - `compiler`: move settings to a separate Config struct
+  - `compiler`: work around an ARM backend bug in LLVM
+  - `interp`: rewrite entire package
+  - `interp`: fix alignment of untyped globals
+  - `loader`: use name "main" for the main package
+  - `loader`: support imports from vendor directories
+  - `stacksize`: add support for DW_CFA_offset_extended
+  - `transform`: show better error message in coroutines lowering
+* **standard library**
+  - `machine`: accept configuration struct for ADC parameters
+  - `machine`: make I2C.Configure signature consistent
+  - `reflect`: implement PtrTo
+  - `runtime`: refactor to simplify stack switching
+  - `runtime`: put metadata at the top end of the heap
+* **targets**
+  - `atsam`: add a length check to findPinPadMapping
+  - `atsam`: improve USBCDC
+  - `atsam`: avoid infinite loop when USBCDC is disconnected
+  - `avr`: add SPI support for Atmega based chips
+  - `avr`: use Clang for compiling C and assembly files
+  - `esp32`: implement task based scheduler
+  - `esp32`: enable the FPU
+  - `esp8266`: implement task based scheduler
+  - `esp`: add compiler-rt library
+  - `esp`: add picolibc
+  - `nrf`: refactor code a bit to reduce duplication
+  - `nrf`: use SPIM peripheral instead of the legacy SPI peripheral
+  - `nrf`: update nrfx submodule to latest commit
+  - `nrf52840`: ensure that USB CDC interface is only initialized once
+  - `nrf52840`: improve USBCDC
+  - `stm32`: use stm32-rs SVDs which are of much higher quality
+  - `stm32`: harmonization of UART logic
+  - `stm32`: replace I2C addressable interface with simpler type
+  - `stm32`: fix i2c and add stm32f407 i2c
+  - `stm32`: revert change that adds support for channels in interrupts
+  - `wasm`: implement a growable heap
+  - `wasm`: fix typo in wasm_exec.js, syscall/js.valueLoadString()
+  - `wasm`: Namespaced Wasm Imports so they don't conflict across modules, or reserved LLVM IR
+  - `wasi`: support env variables based on libc
+  - `wasi`: specify wasi-libc in a different way, to improve error message
+* **boards**
+  - `matrixportal-m4`: add support for board Adafruit Matrix Portal M4
+  - `mkr1000`: add this board
+  - `nucleo-f722ze`: add this board
+  - `clue`: correct volume name and add alias for release version of Adafruit Clue board
+  - `p1am-100`: add support for the P1AM-100 (similar to Arduino MKR)
+  - `microbit-v2`: add initial support based on work done by @alankrantas thank you!
+  - `lgt92`: support for STM32L0 MCUs and Dragino LGT92 device
+  - `nicenano`: nice!nano board support
+  - `circuitplay-bluefruit`: correct internal I2C pin mapping
+  - `clue`: correct for lack of low frequency crystal
+  - `digispark`: split off attiny85 target
+  - `nucleo-l552ze`: implementation with CLOCK, LED, and UART
+  - `nrf52840-mdk-usb-dongle`: add this board
+
+0.16.0
+---
+
+* **command-line**
+  - add initial support for LLVM 11
+  - make lib64 clang include path check more robust
+  - `build`: improve support for GOARCH=386 and add tests
+  - `gdb`: add support for qemu-user targets
+  - `test`: support non-host tests
+  - `test`: add support for -c and -o flags
+  - `test`: implement some benchmark stubs
+* **compiler**
+  - `builder`: improve detection of clang on Fedora
+  - `compiler`: fix floating point comparison bugs
+  - `compiler`: implement negate for complex numbers
+  - `loader`: fix linkname in test binaries
+  - `transform`: add missing return pointer restore for regular coroutine tail
+    calls
+* **standard library**
+  - `machine`: switch default frequency to 4MHz
+  - `machine`: clarify caller's responsibility in `SetInterrupt`
+  - `os`: add `LookupEnv()` stub
+  - `reflect`: implement `Swapper`
+  - `runtime`: fix UTF-8 decoding
+  - `runtime`: gc: use raw stack access whenever possible
+  - `runtime`: use dedicated printfloat32
+  - `runtime`: allow ranging over a nil map
+  - `runtime`: avoid device/nxp dependency in HardFault handler
+  - `testing`: implement dummy Helper method
+  - `testing`: add Run method
+* **targets**
+  - `arm64`: add support for SVCall intrinsic
+  - `atsamd51`: avoid panic when configuring SPI with SDI=NoPin
+  - `avr`: properly support the `.rodata` section
+  - `esp8266`: implement `Pin.Get` function
+  - `nintendoswitch`: fix crash when printing long lines (> 120)
+  - `nintendoswitch`: add env parser and removed unused stuff
+  - `nrf`: add I2C error checking
+  - `nrf`: give more flexibility in picking SPI speeds
+  - `nrf`: fix nrf52832 flash size
+  - `stm32f103`: support wakeups from interrupts
+  - `stm32f405`: add SPI support
+  - `stm32f405`: add I2C support
+  - `wasi`: add support for this target
+  - `wasi`: use 'generic' ABI by default
+  - `wasi`: remove --no-threads flag from wasm-ld
+  - `wasm`: add instanceof support for WebAssembly
+  - `wasm`: use fixed length buffer for putchar
+* **boards**
+  - `d1mini`: add this ESP8266 based board
+  - `esp32`: use board definitions instead of chip names
+  - `qtpy`: add board definition for Adafruit QTPy
+  - `teensy40`: add this board
+
+0.15.0
+---
+
+* **command-line**
+  - add cached GOROOT to info subcommand
+  - embed git-hash in tinygo-dev executable
+  - implement tinygo targets to list usable targets
+  - use simpler file copy instead of file renaming to avoid issues on nrf52840 UF2 bootloaders
+  - use ToSlash() to specify program path
+  - support flashing esp32/esp8266 directly from tinygo
+  - when flashing call PortReset only on other than openocd
+* **compiler**
+  - `compileopts`: add support for custom binary formats
+  - `compiler`: improve display of goroutine wrappers
+  - `interp`: don't panic in the Store method
+  - `interp`: replace some panics with error messages
+  - `interp`: show error line in first line of the traceback
+  - `loader`: be more robust when creating the cached GOROOT
+  - `loader`: rewrite/refactor much of the code to use go list directly
+  - `loader`: use ioutil.TempDir to create a temporary directory
+  - `stacksize`: deal with DW_CFA_advance_loc1
+* **standard library**
+  - `runtime`: use waitForEvents when appropriate
+* **wasm**
+  - `wasm`: Remove --no-threads from wasm-ld calls.
+  - `wasm`: update wasi-libc dependency
+* **targets**
+  - `arduino-mega2560`: fix flashing on Windows
+  - `arm`: automatically determine stack sizes
+  - `arm64`: make dynamic loader structs and constants private
+  - `avr`: configure emulator in board files
+  - `cortexm`: fix stack size calculation with interrupts
+  - `flash`: add openocd settings to atsamd21 / atsamd51
+  - `flash`: add openocd settings to nrf5
+  - `microbit`: reelboard: flash using OpenOCD when needed
+  - `nintendoswitch`: Add dynamic loader for runtime loading PIE sections
+  - `nintendoswitch`: fix import cycle on dynamic_arm64.go
+  - `nintendoswitch`: Fix invalid memory read / write in print calls
+  - `nintendoswitch`: simplified assembly code
+  - `nintendoswitch`: support outputting .nro files directly
+* **boards**
+  - `arduino-zero`: Adding support for the Arduino Zero (#1365)
+  - `atsamd2x`: fix BAUD value
+  - `atsamd5x`: fix BAUD value
+  - `bluepill`: Enable stm32's USART2 for the board and map it to UART1 tinygo's device
+  - `device/atsamd51x`: add all remaining bitfield values for PCHCTRLm Mapping
+  - `esp32`: add libgcc ROM functions to linker script
+  - `esp32`: add SPI support
+  - `esp32`: add support for basic GPIO
+  - `esp32`: add support for the Espressif ESP32 chip
+  - `esp32`: configure the I/O matrix for GPIO pins
+  - `esp32`: export machine.PortMask* for bitbanging implementations
+  - `esp8266`: add support for this chip
+  - `machine/atsamd51x,runtime/atsamd51x`: fixes needed for full support for all PWM pins. Also adds some useful constants to clarify peripheral clock usage
+  - `machine/itsybitsy-nrf52840`: add support for Adafruit Itsybitsy nrf52840 (#1243)
+  - `machine/stm32f4`: refactor common code and add new build tag stm32f4 (#1332)
+  - `nrf`: add SoftDevice support for the Circuit Playground Bluefruit
+  - `nrf`: call sd_app_evt_wait when the SoftDevice is enabled
+  - `nrf52840`: add build tags for SoftDevice support
+  - `nrf52840`: use higher priority for USB-CDC code
+  - `runtime/atsamd51x`: use PCHCTRL_GCLK_SERCOMX_SLOW for setting clocks on all SERCOM ports
+  - `stm32f405`: add basic UART handler
+  - `stm32f405`: add STM32F405 machine/runtime, and new board/target feather-stm32f405
+* **build**
+  - `all`: run test binaries in the correct directory
+  - `build`: Fix arch release job
+  - `ci`: run `tinygo test` for known-working packages
+  - `ci`: set git-fetch-depth to 1
+  - `docker`: fix the problem with the wasm build (#1357)
+  - `Makefile`: check whether submodules have been downloaded in some common cases
+* **docs**
+  - add ESP32, ESP8266, and Adafruit Feather STM32F405 to list of supported boards
+
+0.14.1
+---
+* **command-line**
+  - support for Go 1.15
+* **compiler**
+  - loader:  work around Windows symlink limitation
+
+0.14.0
+---
+* **command-line**
+  - fix `getDefaultPort()` on non-English Windows locales
+  - compileopts: improve error reporting of unsupported flags
+  - fix test subcommand
+  - use auto-retry to locate MSD for UF2 and HEX flashing
+  - fix touchSerialPortAt1200bps on Windows
+  - support package names with backslashes on Windows
+* **compiler**
+  - fix a few crashes due to named types
+  - add support for atomic operations
+  - move the channel blocked list onto the stack
+  - fix -gc=none
+  - fix named string to `[]byte` slice conversion
+  - implement func value and builtin defers
+  - add proper parameter names to runtime.initAll, to fix a panic
+  - builder: fix picolibc include path
+  - builder: use newer version of gohex
+  - builder: try to determine stack size information at compile time
+  - builder: remove -opt=0
+  - interp: fix sync/atomic.Value load/store methods
+  - loader: add Go module support
+  - transform: fix debug information in func lowering pass
+  - transform: do not special-case zero or one implementations of a method call
+  - transform: introduce check for method calls on nil interfaces
+  - transform: gc: track 0-index GEPs to fix miscompilation
+* **cgo**
+  - Add LDFlags support
+* **standard library**
+  - extend stdlib to allow import of more packages
+  - replace master/slave terminology with appropriate alternatives (MOSI->SDO
+    etc)
+  - `internal/bytealg`: reimplement bytealg in pure Go
+  - `internal/task`: fix nil panic in (*internal/task.Stack).Pop
+  - `os`: add Args and stub it with mock data
+  - `os`: implement virtual filesystem support
+  - `reflect`: add Cap and Len support for map and chan
+  - `runtime`: fix return address in scheduler on RISC-V
+  - `runtime`: avoid recursion in printuint64 function
+  - `runtime`: replace ReadRegister with AsmFull inline assembly
+  - `runtime`: fix compilation errors when using gc.extalloc
+  - `runtime`: add cap and len support for chans
+  - `runtime`: refactor time handling (improving accuracy)
+  - `runtime`: make channels work in interrupts
+  - `runtime/interrupt`: add cross-chip disable/restore interrupt support
+  - `sync`: implement `sync.Cond`
+  - `sync`: add WaitGroup
+* **targets**
+  - `arm`: allow nesting in DisableInterrupts and EnableInterrupts
+  - `arm`: make FPU configuraton consistent
+  - `arm`: do not mask fault handlers in critical sections
+  - `atmega2560`: fix pin mapping for pins D2, D5 and the L port
+  - `atsamd`: return an error when an incorrect PWM pin is used
+  - `atsamd`: add support for pin change interrupts
+  - `atsamd`: add DAC support
+  - `atsamd21`: add more ADC pins
+  - `atsamd51`: fix ROM / RAM size on atsamd51j20
+  - `atsamd51`: add more pins
+  - `atsamd51`: add more ADC pins
+  - `atsamd51`: add pin change interrupt settings
+  - `atsamd51`: extend pinPadMapping
+  - `arduino-nano33`: use (U)SB flag to ensure that device can be found when
+     not on default port
+  - `arduino-nano33`: remove (d)ebug flag to reduce console noise when flashing
+  - `avr`: use standard pin numbering
+  - `avr`: unify GPIO pin/port code
+  - `avr`: add support for PinInputPullup
+  - `avr`: work around codegen bug in LLVM 10
+  - `avr`: fix target triple
+  - `fe310`: remove extra println left in by mistake
+  - `feather-nrf52840`: add support for the Feather nRF52840
+  - `maixbit`: add board definition and dummy runtime
+  - `nintendoswitch`: Add experimental Nintendo Switch support without CRT
+  - `nrf`: expose the RAM base address
+  - `nrf`: add support for pin change interrupts
+  - `nrf`: add microbit-s110v8 target
+  - `nrf`: fix bug in SPI.Tx
+  - `nrf`: support debugging the PCA10056
+  - `pygamer`: add Adafruit PyGamer suport
+  - `riscv`: fix interrupt configuration bug
+  - `riscv`: disable linker relaxations during gp init
+  - `stm32f4disco`: add new target with ST-Link v2.1 debugger
+  - `teensy36`: add Teensy 3.6 support
+  - `wasm`: fix event handling
+  - `wasm`: add --no-demangle linker option
+  - `wioterminal`: add support for the Seeed Wio Terminal
+  - `xiao`: add support for the Seeed XIAO
+
+0.13.1
+---
+* **standard library**
+  - `runtime`: do not put scheduler and GC code in the same section
+  - `runtime`: copy stack scan assembly for GBA
+* **boards**
+  - `gameboy-advance`: always use ARM mode instead of Thumb mode
+
+
+0.13.0
+---
+* **command line**
+  - use `gdb-multiarch` for debugging Cortex-M chips
+  - support `tinygo run` with simavr
+  - support LLVM 10
+  - support Go 1.14
+  - retry 3 times when attempting to do a 1200-baud reset
+* **compiler**
+  - mark the `abort` function as noreturn
+  - fix deferred calls to exported functions
+  - add debug info for local variables
+  - check for channel size limit
+  - refactor coroutine lowering
+  - add `dereferenceable_or_null` attribute to pointer parameters
+  - do not perform nil checking when indexing slices and on `unsafe.Pointer`
+  - remove `runtime.isnil` hack
+  - use LLVM builtins for runtime `memcpy`/`memmove`/`memzero` functions
+  - implement spec-compliant shifts on negative/overflow
+  - support anonymous type asserts
+  - track pointer result of string concatenation for GC
+  - track PHI nodes for GC
+  - add debug info to goroutine start wrappers
+  - optimize comparing interface values against nil
+  - fix miscompilation when deferring an interface call
+  - builder: include picolibc for most baremetal targets
+  - builder: run tools (clang, lld) as separate processes
+  - builder: use `-fshort-enums` consistently
+  - interp: add support for constant type asserts
+  - interp: better support for interface operations
+  - interp: include backtrace with error
+  - transform: do not track const globals for GC
+  - transform: replace panics with source locations
+  - transform: fix error in interface lowering pass
+  - transform: make coroutine lowering deterministic
+  - transform: fix miscompilation in func lowering
+* **cgo**
+  - make `-I` and `-L` paths absolute
+* **standard library**
+  - `machine`: set the USB VID and PID to the manufacturer values
+  - `machine`: correct USB CDC composite descriptors
+  - `machine`: move `errors.New` calls to globals
+  - `runtime`: support operations on nil maps
+  - `runtime`: fix copy builtin return value on AVR
+  - `runtime`: refactor goroutines
+  - `runtime`: support `-scheduler=none` on most platforms
+  - `runtime`: run package initialization in the main goroutine
+  - `runtime`: export `malloc` / `free` for use from C
+  - `runtime`: add garbage collector that uses an external allocator
+  - `runtime`: scan callee-saved registers while marking the stack
+  - `runtime`: remove recursion from conservative GC
+  - `runtime`: fix blocking select on nil channel
+  - `runtime/volatile`: include `ReplaceBits` method
+  - `sync`: implement trivial `sync.Map`
+* **targets**
+  - `arm`: use `-fomit-frame-pointer`
+  - `atmega1284`: support this chip for testing purposes
+  - `atsamd51`: make QSPI available on all boards
+  - `atsamd51`: add support for ADC1
+  - `atsamd51`: use new interrupt registration in UART code
+  - `attiny`: clean up pin definitions
+  - `avr`: use the correct RAM start address
+  - `avr`: pass the correct `-mmcu` flag to the linker
+  - `avr`: add support for tasks scheduler (disabled by default)
+  - `avr`: fix linker problem with overlapping program/data areas
+  - `nrf`: fix typo in pin configuration options
+  - `nrf`: add lib/nrfx/mdk to include dirs
+  - `nrf52840`: implement USB-CDC
+  - `riscv`: implement VirtIO target and add RISC-V integration test
+  - `riscv`: add I2C support for the HiFive1 rev B board
+  - `stm32`: refactor GPIO pin handling
+  - `stm32`: refactor UART code
+  - `stm32f4`: add SPI
+  - `wasm`: support Go 1.14 (breaking previous versions)
+  - `wasm`: support `syscall/js.CopyBytesToJS`
+  - `wasm`: sync polyfills from Go 1.14.
+* **boards**
+  - `arduino-mega2560`: add the Arduino Mega 2560
+  - `clue-alpha`: add the Adafruit CLUE Alpha
+  - `gameboy-advance`: enable debugging with GDB
+  - `particle-argon`: add the Particle Argon board
+  - `particle-boron`: add the Particle Boron board
+  - `particle-xenon`: add the Particle Xenon board
+  - `reelboard`: add `reelboard-s140v7` SoftDevice target
+
+0.12.0
+---
+* **command line**
+  - add initial FreeBSD support
+  - remove getting a serial port in gdb subcommand
+  - add support for debugging through JLinkGDBServer
+  - fix CGo when cross compiling
+  - remove default port check for Digispark as micronucleus communicates directly using HID
+  - differentiate between various serial/USB error messages
+* **builder**
+  - improve detection of Clang headers
+* **compiler**
+  - fix assertion on empty interface
+  - don't crash when encountering `types.Invalid`
+  - revise defer to use heap allocations when running a variable number of times
+  - improve error messages for failed imports
+  - improve "function redeclared" error
+  - add globaldce pass to start of optimization pipeline
+  - add support for debugging globals
+  - implement RISC-V CSR operations as intrinsics
+  - add support for CGO_ENABLED environment variable
+  - do not emit debug info for extern globals (bugfix)
+  - add support for interrupts
+  - implement maps for arbitrary keys
+  - interp: error location for "unknown GEP" error
+  - wasm-abi: create temporary allocas in the entry block
+* **cgo**
+  - add support for symbols in `#define`
+  - fix a bug in number tokenization
+* **standard library**
+  - `machine`: avoid bytes package in USB logic
+  - `runtime`: fix external address declarations
+  - `runtime`: provide implementation for `internal/bytealg.IndexByte`
+* **targets**
+  - `atsamd51`: fix volatile usage
+  - `atsamd51`: fix ADC, updating to 12-bits precision
+  - `atsamd51`: refactor SPI pin configuration to only look at pin numbers
+  - `atsamd51`: switch UART to use new pin configuration
+  - `atsamd51`: fix obvious bug in I2C code
+  - `atsamd51`: use only the necessary UART interrupts
+  - `atsamd51`: refactor I2C pin handling to auto-detect pin mode
+  - `avr`: use a garbage collector
+  - `fe310`: use CLINT peripheral for timekeeping
+  - `fe310`: add support for PLIC interrupts
+  - `fe310`: implement UART receive interrupts
+  - `riscv`: support sleeping in QEMU
+  - `riscv`: add bare-bones interrupt support
+  - `riscv`: print exception PC and code
+  - `wasm`: implement memcpy and memset
+  - `wasm`: include wasi-libc
+  - `wasm`: use wasi ABI for basic startup/stdout
+* **boards**
+  - `arduino`: make avrdude command line compatible with Windows
+  - `arduino-nano`: add this board
+  - `arduino-nano33`: fix UART1 and UART2
+  - `circuitplay-bluefruit`: add this board
+  - `digispark`: add clock speed and pin mappings
+  - `gameboy-advance`: include compiler-rt in build
+  - `gameboy-advance`: implement interrupt handler
+  - `hifive1b`: add support for gdb subcommand
+  - `pyportal`: add this board
+  - `pyportal`: remove manual SPI pin mapping as now handled by default
+
+
+0.11.0
+---
+* **command line**
+  - add support for QEMU in `gdb` subcommand
+  - use builtin Clang when building statically, dropping the clang-9 dependency
+  - search for default serial port on both macOS and Linux
+  - windows: support `tinygo flash` directly by using win32 wmi
+* **compiler**
+  - add location information to the IR checker
+  - make reflection sidetables constant globals
+  - improve error locations in goroutine lowering
+  - interp: improve support for maps with string keys
+  - interp: add runtime fallback for mapassign operations
+* **standard library**
+  - `machine`: add support for `SPI.Tx()` on play.tinygo.org
+  - `machine`: rename `CPU_FREQUENCY` to `CPUFrequency()`
+* **targets**
+  - `adafruit-pybadge`: add Adafruit Pybadge
+  - `arduino-nano33`: allow simulation on play.tinygo.org
+  - `arduino-nano33`: fix default SPI pin numbers to be D13/D11/D12
+  - `circuitplay-express`: allow simulation on play.tinygo.org
+  - `hifive1-qemu`: add target for testing RISC-V bare metal in QEMU
+  - `riscv`: fix heap corruption due to changes in LLVM 9
+  - `riscv`: add support for compiler-rt
+  - `qemu`: rename to `cortex-m-qemu`
+
+0.10.0
+---
+* **command line**
+  - halt GDB after flashing with `gdb` subcommand
+  - fix a crash when using `-ocd-output`
+  - add `info` subcommand
+  - add `-programmer` flag
+* **builder**
+  - macos: use llvm@8 instead of just llvm in paths
+  - add `linkerscript` key to target JSON files
+  - write a symbol table when writing out the compiler-rt lib
+  - make Clang header detection more robust
+  - switch to LLVM 9
+* **compiler**
+  - fix interface miscompilation with reflect
+  - fix miscompile of static goroutine calls to closures
+  - fix `todo: store` panic
+  - fix incorrect starting value for optimized allocations in a loop
+  - optimize coroutines on non-Cortex-M targets
+  - fix crash for programs which have heap allocations but never hit the GC
+  - add support for async interface calls
+  - fix inserting non-const values in a const global
+  - interp: improve error reporting
+  - interp: implement comparing ptrtoint to 0
+* **cgo**
+  - improve diagnostics
+  - implement the constant parser (for `#define`) as a real parser
+  - rename reserved field names such as `type`
+  - avoid `"unsafe" imported but not used` error
+  - include all enums in the CGo Go AST
+  - add support for nested structs and unions
+  - implement `#cgo CFLAGS`
+* **standard library**
+  - `reflect`: add implementation of array alignment
+  - `runtime`: improve scheduler performance when no goroutines are queued
+  - `runtime`: add blocking select
+  - `runtime`: implement interface equality in non-trivial cases
+  - `runtime`: add AdjustTimeOffset to update current time
+  - `runtime`: only implement CountString for required platforms
+  - `runtime`: use MSP/PSP registers for scheduling on Cortex-M
+* **targets**
+  - `arm`: add system timer registers
+  - `atmega`: add port C GPIO support
+  - `atsamd21`: correct handling of pins >= 32
+  - `atsamd21`: i2s initialization fixes
+  - `atsamd51`: fix clock init code
+  - `atsamd51`: correct initialization for RTC
+  - `atsamd51`: fix pin function selection
+  - `atsamd51`: pin method cleanup
+  - `atsamd51`: allow setting pin mode for each of the SPI pins
+  - `atsamd51`: correct channel init and pin map for ADC based on ItsyBitsy-M4
+  - `feather-m4`: add Adafruit Feather M4 board
+  - `hifive1b`: add support for SPI1
+  - `hifive1b`: fix compiling in simulation
+  - `linux`: fix time on arm32
+  - `metro-m4`: add support for Adafruit Metro M4 Express Airlift board
+  - `metro-m4`: fixes for UART2
+  - `pinetime-devkit0`: add support for the PineTime dev kit
+  - `x9pro`: add support for this smartwatch
+  - `pca10040-s132v6`: add support for SoftDevice
+  - `pca10056-s140v7`: add support for SoftDevice
+  - `arduino-nano33`: added SPI1 connected to NINA-W102 chip on Arduino Nano 33 IOT
+
+0.9.0
+---
+* **command line**
+  - implement 1200-baud UART bootloader reset when flashing boards that support
+    it
+  - flash using mass-storage device for boards that support it
+  - implement `tinygo env`
+  - add support for Windows (but not yet producing Windows binaries)
+  - add Go version to `tinygo env`
+  - update SVD files for up-to-date peripheral interfaces
+* **compiler**
+  - add `//go:align` pragma
+  - fix bug related to type aliases
+  - add support for buffered channels
+  - remove incorrect reflect optimization
+  - implement copying slices in init interpretation
+  - add support for constant indices with a named type
+  - add support for recursive types like linked lists
+  - fix miscompile of function nil panics
+  - fix bug related to goroutines
+* **standard library**
+  - `machine`: do not check for nil slices in `SPI.Tx`
+  - `reflectlite`: add support for Go 1.13
+  - `runtime`: implement `internal/bytealg.CountString`
+  - `sync`: properly handle nil `New` func in `sync.Pool`
+* **targets**
+  - `arduino`: fix .bss section initialization
+  - `fe310`: implement `Pin.Get`
+  - `gameboy-advance`: support directly outputting .gba files
+  - `samd`: reduce code size by avoiding reflection
+  - `samd21`: do not hardcode pin numbers for peripherals
+  - `stm32f103`: avoid issue with `time.Sleep` less than 200µs
+
 0.8.0
 ---
 * **command line**

CODE-OF-CONDUCT.md

@@ -0,0 +1,76 @@
+# Contributor Covenant Code of Conduct
+
+## Our Pledge
+
+In the interest of fostering an open and welcoming environment, we as
+contributors and maintainers pledge to make participation in our project and
+our community a harassment-free experience for everyone, regardless of age, body
+size, disability, ethnicity, sex characteristics, gender identity and expression,
+level of experience, education, socio-economic status, nationality, personal
+appearance, race, religion, or sexual identity and orientation.
+
+## Our Standards
+
+Examples of behavior that contributes to creating a positive environment
+include:
+
+* Using welcoming and inclusive language
+* Being respectful of differing viewpoints and experiences
+* Gracefully accepting constructive criticism
+* Focusing on what is best for the community
+* Showing empathy towards other community members
+
+Examples of unacceptable behavior by participants include:
+
+* The use of sexualized language or imagery and unwelcome sexual attention or
+  advances
+* Trolling, insulting/derogatory comments, and personal or political attacks
+* Public or private harassment
+* Publishing others' private information, such as a physical or electronic
+  address, without explicit permission
+* Other conduct which could reasonably be considered inappropriate in a
+  professional setting
+
+## Our Responsibilities
+
+Project maintainers are responsible for clarifying the standards of acceptable
+behavior and are expected to take appropriate and fair corrective action in
+response to any instances of unacceptable behavior.
+
+Project maintainers have the right and responsibility to remove, edit, or
+reject comments, commits, code, wiki edits, issues, and other contributions
+that are not aligned to this Code of Conduct, or to ban temporarily or
+permanently any contributor for other behaviors that they deem inappropriate,
+threatening, offensive, or harmful.
+
+## Scope
+
+This Code of Conduct applies within all project spaces, and it also applies when
+an individual is representing the project or its community in public spaces.
+Examples of representing a project or community include using an official
+project e-mail address, posting via an official social media account, or acting
+as an appointed representative at an online or offline event. Representation of
+a project may be further defined and clarified by project maintainers.
+
+## Enforcement
+
+Instances of abusive, harassing, or otherwise unacceptable behavior may be
+reported by contacting the project team at [conduct@tinygo.org](mailto:conduct@tinygo.org). All
+complaints will be reviewed and investigated and will result in a response that
+is deemed necessary and appropriate to the circumstances. The project team is
+obligated to maintain confidentiality with regard to the reporter of an incident.
+Further details of specific enforcement policies may be posted separately.
+
+Project maintainers who do not follow or enforce the Code of Conduct in good
+faith may face temporary or permanent repercussions as determined by other
+members of the project's leadership.
+
+## Attribution
+
+This Code of Conduct is adapted from the [Contributor Covenant][homepage], version 1.4,
+available at https://www.contributor-covenant.org/version/1/4/code-of-conduct.html
+
+[homepage]: https://www.contributor-covenant.org
+
+For answers to common questions about this code of conduct, see
+https://www.contributor-covenant.org/faq

CONTRIBUTING.md

@@ -32,7 +32,7 @@ Microcontrollers have lots of peripherals (I2C, SPI, ADC, etc.) and many don't h
 
 ## How to use our Github repository
 
-The `master` branch of this repo will always have the latest released version of TinyGo. All of the active development work for the next release will take place in the `dev` branch. TinyGo will use semantic versioning and will create a tag/release for each release.
+The `release` branch of this repo will always have the latest released version of TinyGo. All of the active development work for the next release will take place in the `dev` branch. TinyGo will use semantic versioning and will create a tag/release for each release.
 
 Here is how to contribute back some code or documentation:
 

CONTRIBUTORS

@@ -15,3 +15,4 @@ Ayke van Laethem <aykevanlaethem@gmail.com>
 Daniel Esteban <conejo@conejo.me>
 Loon, LLC.
 Ron Evans <ron@hybridgroup.com>
+Nia Weiss <niaow1234@gmail.com>

Dockerfile

@@ -1,10 +1,10 @@
-# TinyGo base stage installs Go 1.12, LLVM 8 and the TinyGo compiler itself.
-FROM golang:1.12 AS tinygo-base
+# TinyGo base stage installs the most recent Go 1.16.x, LLVM 11 and the TinyGo compiler itself.
+FROM golang:1.16 AS tinygo-base
 
 RUN wget -O- https://apt.llvm.org/llvm-snapshot.gpg.key| apt-key add - && \
-    echo "deb http://apt.llvm.org/buster/ llvm-toolchain-buster-8 main" >> /etc/apt/sources.list && \
+    echo "deb http://apt.llvm.org/buster/ llvm-toolchain-buster-11 main" >> /etc/apt/sources.list && \
     apt-get update && \
-    apt-get install -y llvm-8-dev libclang-8-dev git
+    apt-get install -y llvm-11-dev libclang-11-dev lld-11 git
 
 COPY . /tinygo
 
@@ -12,8 +12,11 @@ COPY . /tinygo
 # after copying the tinygo directory in the previous step.
 RUN cd /tinygo/ && \
     rm -rf ./lib/* && \
+    git submodule sync && \
     git submodule update --init --recursive --force
 
+COPY ./lib/picolibc-include/* /tinygo/lib/picolibc-include/
+
 RUN cd /tinygo/ && \
     go install /tinygo/
 
@@ -24,10 +27,10 @@ COPY --from=tinygo-base /go/bin/tinygo /go/bin/tinygo
 COPY --from=tinygo-base /tinygo/src /tinygo/src
 COPY --from=tinygo-base /tinygo/targets /tinygo/targets
 
-RUN wget -O- https://apt.llvm.org/llvm-snapshot.gpg.key| apt-key add - && \
-    echo "deb http://apt.llvm.org/buster/ llvm-toolchain-buster-8 main" >> /etc/apt/sources.list && \
+RUN cd /tinygo/ && \
     apt-get update && \
-    apt-get install -y libllvm8 lld-8
+    apt-get install -y make clang-11 libllvm11 lld-11 && \
+    make wasi-libc
 
 # tinygo-avr stage installs the needed dependencies to compile TinyGo programs for AVR microcontrollers.
 FROM tinygo-base AS tinygo-avr
@@ -41,9 +44,8 @@ COPY --from=tinygo-base /tinygo/lib /tinygo/lib
 
 RUN cd /tinygo/ && \
     apt-get update && \
-    apt-get install -y apt-utils python3 make binutils-avr gcc-avr avr-libc && \
+    apt-get install -y apt-utils make binutils-avr gcc-avr avr-libc && \
     make gen-device-avr && \
-    apt-get remove -y python3 && \
     apt-get autoremove -y && \
     apt-get clean
 
@@ -59,11 +61,8 @@ COPY --from=tinygo-base /tinygo/lib /tinygo/lib
 
 RUN cd /tinygo/ && \
     apt-get update && \
-    apt-get install -y apt-utils python3 make clang-8 && \
-    make gen-device-nrf && make gen-device-stm32 && \
-    apt-get remove -y python3 && \
-    apt-get autoremove -y && \
-    apt-get clean
+    apt-get install -y apt-utils make clang-11 && \
+    make gen-device-nrf && make gen-device-stm32
 
 # tinygo-all stage installs the needed dependencies to compile TinyGo programs for all platforms.
 FROM tinygo-wasm AS tinygo-all
@@ -74,10 +73,7 @@ COPY --from=tinygo-base /tinygo/lib /tinygo/lib
 
 RUN cd /tinygo/ && \
     apt-get update && \
-    apt-get install -y apt-utils python3 make clang-8 binutils-avr gcc-avr avr-libc && \
-    make gen-device && \
-    apt-get remove -y python3 && \
-    apt-get autoremove -y && \
-    apt-get clean
+    apt-get install -y apt-utils make clang-11 binutils-avr gcc-avr avr-libc && \
+    make gen-device
 
 CMD ["tinygo"]

LICENSE

@@ -1,7 +1,10 @@
-Copyright (c) 2018-2019 TinyGo Authors. All rights reserved.
+Copyright (c) 2018-2021 TinyGo Authors. All rights reserved.
 
 TinyGo includes portions of the Go standard library.
-Copyright (c) 2009-2019 The Go Authors. All rights reserved.
+Copyright (c) 2009-2021 The Go Authors. All rights reserved.
+
+TinyGo includes portions of LLVM, which is under the Apache License v2.0 with
+LLVM Exceptions. See https://llvm.org/LICENSE.txt for license information.
 
 Redistribution and use in source and binary forms, with or without
 modification, are permitted provided that the following conditions are

Makefile

@@ -1,153 +1,467 @@
 
 # aliases
 all: tinygo
-tinygo: build/tinygo
 
 # Default build and source directories, as created by `make llvm-build`.
 LLVM_BUILDDIR ?= llvm-build
-CLANG_SRC ?= llvm-project/clang
-LLD_SRC ?= llvm-project/lld
+LLVM_PROJECTDIR ?= llvm-project
+CLANG_SRC ?= $(LLVM_PROJECTDIR)/clang
+LLD_SRC ?= $(LLVM_PROJECTDIR)/lld
 
-.PHONY: all tinygo build/tinygo test $(LLVM_BUILDDIR) llvm-source clean fmt gen-device gen-device-nrf gen-device-avr
+# Try to autodetect LLVM build tools.
+detect = $(shell command -v $(1) 2> /dev/null && echo $(1))
+CLANG ?= $(word 1,$(abspath $(call detect,llvm-build/bin/clang))$(call detect,clang-11)$(call detect,clang-10)$(call detect,clang))
+LLVM_AR ?= $(word 1,$(abspath $(call detect,llvm-build/bin/llvm-ar))$(call detect,llvm-ar-11)$(call detect,llvm-ar-10)$(call detect,llvm-ar))
+LLVM_NM ?= $(word 1,$(abspath $(call detect,llvm-build/bin/llvm-nm))$(call detect,llvm-nm-11)$(call detect,llvm-nm-10)$(call detect,llvm-nm))
 
-LLVM_COMPONENTS = all-targets analysis asmparser asmprinter bitreader bitwriter codegen core coroutines debuginfodwarf executionengine instrumentation interpreter ipo irreader linker lto mc mcjit objcarcopts option profiledata scalaropts support target
+# Go binary and GOROOT to select
+GO ?= go
+export GOROOT = $(shell $(GO) env GOROOT)
 
-UNAME_S := $(shell uname -s)
-ifeq ($(UNAME_S),Linux)
+# md5sum binary
+MD5SUM = md5sum
+
+# tinygo binary for tests
+TINYGO ?= $(word 1,$(call detect,tinygo)$(call detect,build/tinygo))
+
+# Use CCACHE for LLVM if possible
+ifneq (, $(shell command -v ccache 2> /dev/null))
+    LLVM_OPTION += '-DLLVM_CCACHE_BUILD=ON'
+endif
+
+# Allow enabling LLVM assertions
+ifeq (1, $(ASSERT))
+    LLVM_OPTION += '-DLLVM_ENABLE_ASSERTIONS=ON'
+else
+    LLVM_OPTION += '-DLLVM_ENABLE_ASSERTIONS=OFF'
+endif
+
+.PHONY: all tinygo test $(LLVM_BUILDDIR) llvm-source clean fmt gen-device gen-device-nrf gen-device-nxp gen-device-avr gen-device-rp
+
+LLVM_COMPONENTS = all-targets analysis asmparser asmprinter bitreader bitwriter codegen core coroutines coverage debuginfodwarf executionengine frontendopenmp instrumentation interpreter ipo irreader linker lto mc mcjit objcarcopts option profiledata scalaropts support target
+
+ifeq ($(OS),Windows_NT)
+    EXE = .exe
+    START_GROUP = -Wl,--start-group
+    END_GROUP = -Wl,--end-group
+
+    # LLVM compiled using MinGW on Windows appears to have problems with threads.
+    # Without this flag, linking results in errors like these:
+    #     libLLVMSupport.a(Threading.cpp.obj):Threading.cpp:(.text+0x55): undefined reference to `std::thread::hardware_concurrency()'
+    LLVM_OPTION += -DLLVM_ENABLE_THREADS=OFF -DLLVM_ENABLE_PIC=OFF
+
+    CGO_CPPFLAGS += -DCINDEX_NO_EXPORTS
+    CGO_LDFLAGS += -static -static-libgcc -static-libstdc++
+    CGO_LDFLAGS_EXTRA += -lversion
+
+    LIBCLANG_NAME = libclang
+
+else ifeq ($(shell uname -s),Darwin)
+    MD5SUM = md5
+    LIBCLANG_NAME = clang
+else ifeq ($(shell uname -s),FreeBSD)
+    MD5SUM = md5
+    LIBCLANG_NAME = clang
+    START_GROUP = -Wl,--start-group
+    END_GROUP = -Wl,--end-group
+else
+    LIBCLANG_NAME = clang
     START_GROUP = -Wl,--start-group
     END_GROUP = -Wl,--end-group
 endif
 
-CLANG_LIBS = $(START_GROUP) $(abspath $(LLVM_BUILDDIR))/lib/libclang.a -lclangAnalysis -lclangARCMigrate -lclangAST -lclangASTMatchers -lclangBasic -lclangCodeGen -lclangCrossTU -lclangDriver -lclangDynamicASTMatchers -lclangEdit -lclangFormat -lclangFrontend -lclangFrontendTool -lclangHandleCXX -lclangHandleLLVM -lclangIndex -lclangLex -lclangParse -lclangRewrite -lclangRewriteFrontend -lclangSema -lclangSerialization -lclangStaticAnalyzerCheckers -lclangStaticAnalyzerCore -lclangStaticAnalyzerFrontend -lclangTooling -lclangToolingASTDiff -lclangToolingCore -lclangToolingInclusions $(END_GROUP) -lstdc++
+# Libraries that should be linked in for the statically linked Clang.
+CLANG_LIB_NAMES = clangAnalysis clangAST clangASTMatchers clangBasic clangCodeGen clangCrossTU clangDriver clangDynamicASTMatchers clangEdit clangFormat clangFrontend clangFrontendTool clangHandleCXX clangHandleLLVM clangIndex clangLex clangParse clangRewrite clangRewriteFrontend clangSema clangSerialization clangTooling clangToolingASTDiff clangToolingCore clangToolingInclusions
+CLANG_LIBS = $(START_GROUP) $(addprefix -l,$(CLANG_LIB_NAMES)) $(END_GROUP) -lstdc++
 
-LLD_LIBS = $(START_GROUP) -llldCOFF -llldCommon -llldCore -llldDriver -llldELF -llldMachO -llldMinGW -llldReaderWriter -llldWasm -llldYAML $(END_GROUP)
+# Libraries that should be linked in for the statically linked LLD.
+LLD_LIB_NAMES = lldCOFF lldCommon lldCore lldDriver lldELF lldMachO lldMinGW lldReaderWriter lldWasm lldYAML
+LLD_LIBS = $(START_GROUP) $(addprefix -l,$(LLD_LIB_NAMES)) $(END_GROUP)
 
+# Other libraries that are needed to link TinyGo.
+EXTRA_LIB_NAMES = LLVMInterpreter
+
+# These build targets appear to be the only ones necessary to build all TinyGo
+# dependencies. Only building a subset significantly speeds up rebuilding LLVM.
+# The Makefile rules convert a name like lldELF to lib/liblldELF.a to match the
+# library path (for ninja).
+# This list also includes a few tools that are necessary as part of the full
+# TinyGo build.
+NINJA_BUILD_TARGETS = clang llvm-config llvm-ar llvm-nm $(addprefix lib/lib,$(addsuffix .a,$(LIBCLANG_NAME) $(CLANG_LIB_NAMES) $(LLD_LIB_NAMES) $(EXTRA_LIB_NAMES)))
 
 # For static linking.
-ifneq ("$(wildcard $(LLVM_BUILDDIR)/bin/llvm-config)","")
-    CGO_CPPFLAGS=$(shell $(LLVM_BUILDDIR)/bin/llvm-config --cppflags) -I$(abspath $(CLANG_SRC))/include -I$(abspath $(LLD_SRC))/include
-    CGO_CXXFLAGS=-std=c++11
-    CGO_LDFLAGS=-L$(LLVM_BUILDDIR)/lib $(CLANG_LIBS) $(LLD_LIBS) $(shell $(LLVM_BUILDDIR)/bin/llvm-config --ldflags --libs --system-libs $(LLVM_COMPONENTS))
+ifneq ("$(wildcard $(LLVM_BUILDDIR)/bin/llvm-config*)","")
+    CGO_CPPFLAGS+=$(shell $(LLVM_BUILDDIR)/bin/llvm-config --cppflags) -I$(abspath $(LLVM_BUILDDIR))/tools/clang/include -I$(abspath $(CLANG_SRC))/include -I$(abspath $(LLD_SRC))/include
+    CGO_CXXFLAGS=-std=c++14
+    CGO_LDFLAGS+=$(abspath $(LLVM_BUILDDIR))/lib/lib$(LIBCLANG_NAME).a -L$(abspath $(LLVM_BUILDDIR)/lib) $(CLANG_LIBS) $(LLD_LIBS) $(shell $(LLVM_BUILDDIR)/bin/llvm-config --ldflags --libs --system-libs $(LLVM_COMPONENTS)) -lstdc++ $(CGO_LDFLAGS_EXTRA)
 endif
 
 
 clean:
 	@rm -rf build
 
-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_PATHS = ./*.go builder cgo compiler interp loader src/device/arm src/examples src/machine src/os src/reflect src/runtime src/sync src/syscall src/internal/reflectlite transform
 fmt:
 	@gofmt -l -w $(FMT_PATHS)
 fmt-check:
 	@unformatted=$$(gofmt -l $(FMT_PATHS)); [ -z "$$unformatted" ] && exit 0; echo "Unformatted:"; for fn in $$unformatted; do echo "  $$fn"; done; exit 1
 
 
-gen-device: gen-device-avr gen-device-nrf gen-device-sam gen-device-sifive gen-device-stm32
+gen-device: gen-device-avr gen-device-esp gen-device-nrf gen-device-sam gen-device-sifive gen-device-kendryte gen-device-nxp gen-device-rp
+ifneq ($(STM32), 0)
+gen-device: gen-device-stm32
+endif
 
 gen-device-avr:
-	./tools/gen-device-avr.py lib/avr/packs/atmega src/device/avr/
-	./tools/gen-device-avr.py lib/avr/packs/tiny src/device/avr/
-	go fmt ./src/device/avr
+	@if [ ! -e lib/avr/README.md ]; then echo "Submodules have not been downloaded. Please download them using:\n  git submodule update --init"; exit 1; fi
+	$(GO) build -o ./build/gen-device-avr ./tools/gen-device-avr/
+	./build/gen-device-avr lib/avr/packs/atmega src/device/avr/
+	./build/gen-device-avr lib/avr/packs/tiny src/device/avr/
+	@GO111MODULE=off $(GO) fmt ./src/device/avr
 
-gen-device-nrf:
-	./tools/gen-device-svd.py lib/nrfx/mdk/ src/device/nrf/ --source=https://github.com/NordicSemiconductor/nrfx/tree/master/mdk
-	go fmt ./src/device/nrf
+build/gen-device-svd: ./tools/gen-device-svd/*.go
+	$(GO) build -o $@ ./tools/gen-device-svd/
 
-gen-device-sam:
-	./tools/gen-device-svd.py lib/cmsis-svd/data/Atmel/ src/device/sam/ --source=https://github.com/posborne/cmsis-svd/tree/master/data/Atmel
-	go fmt ./src/device/sam
+gen-device-esp: build/gen-device-svd
+	./build/gen-device-svd -source=https://github.com/posborne/cmsis-svd/tree/master/data/Espressif-Community -interrupts=software lib/cmsis-svd/data/Espressif-Community/ src/device/esp/
+	GO111MODULE=off $(GO) fmt ./src/device/esp
 
-gen-device-sifive:
-	./tools/gen-device-svd.py lib/cmsis-svd/data/SiFive-Community/ src/device/sifive/ --source=https://github.com/AdaCore/svd2ada/tree/master/CMSIS-SVD/SiFive-Community
-	go fmt ./src/device/sifive
+gen-device-nrf: build/gen-device-svd
+	./build/gen-device-svd -source=https://github.com/NordicSemiconductor/nrfx/tree/master/mdk lib/nrfx/mdk/ src/device/nrf/
+	GO111MODULE=off $(GO) fmt ./src/device/nrf
 
-gen-device-stm32:
-	./tools/gen-device-svd.py lib/cmsis-svd/data/STMicro/ src/device/stm32/ --source=https://github.com/posborne/cmsis-svd/tree/master/data/STMicro
-	go fmt ./src/device/stm32
+gen-device-nxp: build/gen-device-svd
+	./build/gen-device-svd -source=https://github.com/posborne/cmsis-svd/tree/master/data/NXP lib/cmsis-svd/data/NXP/ src/device/nxp/
+	GO111MODULE=off $(GO) fmt ./src/device/nxp
 
+gen-device-sam: build/gen-device-svd
+	./build/gen-device-svd -source=https://github.com/posborne/cmsis-svd/tree/master/data/Atmel lib/cmsis-svd/data/Atmel/ src/device/sam/
+	GO111MODULE=off $(GO) fmt ./src/device/sam
+
+gen-device-sifive: build/gen-device-svd
+	./build/gen-device-svd -source=https://github.com/posborne/cmsis-svd/tree/master/data/SiFive-Community -interrupts=software lib/cmsis-svd/data/SiFive-Community/ src/device/sifive/
+	GO111MODULE=off $(GO) fmt ./src/device/sifive
+
+gen-device-kendryte: build/gen-device-svd
+	./build/gen-device-svd -source=https://github.com/posborne/cmsis-svd/tree/master/data/Kendryte-Community -interrupts=software lib/cmsis-svd/data/Kendryte-Community/ src/device/kendryte/
+	GO111MODULE=off $(GO) fmt ./src/device/kendryte
+
+gen-device-stm32: build/gen-device-svd
+	./build/gen-device-svd -source=https://github.com/tinygo-org/stm32-svd lib/stm32-svd/svd src/device/stm32/
+	GO111MODULE=off $(GO) fmt ./src/device/stm32
+
+gen-device-rp: build/gen-device-svd
+	./build/gen-device-svd -source=https://github.com/posborne/cmsis-svd/tree/master/data/RaspberryPi lib/cmsis-svd/data/RaspberryPi/ src/device/rp/
+	GO111MODULE=off $(GO) fmt ./src/device/rp
 
 # Get LLVM sources.
-llvm-project/README.md:
-	git clone -b release/8.x https://github.com/llvm/llvm-project
-llvm-source: llvm-project/README.md
+$(LLVM_PROJECTDIR)/llvm:
+	git clone -b xtensa_release_11.0.0 --depth=1 https://github.com/tinygo-org/llvm-project $(LLVM_PROJECTDIR)
+llvm-source: $(LLVM_PROJECTDIR)/llvm
 
 # Configure LLVM.
 TINYGO_SOURCE_DIR=$(shell pwd)
 $(LLVM_BUILDDIR)/build.ninja: llvm-source
-	mkdir -p $(LLVM_BUILDDIR); cd $(LLVM_BUILDDIR); cmake -G Ninja $(TINYGO_SOURCE_DIR)/llvm-project/llvm "-DLLVM_TARGETS_TO_BUILD=X86;ARM;AArch64;WebAssembly" "-DLLVM_EXPERIMENTAL_TARGETS_TO_BUILD=AVR;RISCV" -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_ASSERTIONS=OFF -DLIBCLANG_BUILD_STATIC=ON -DLLVM_ENABLE_TERMINFO=OFF -DLLVM_ENABLE_ZLIB=OFF -DLLVM_ENABLE_PROJECTS="clang;lld" -DLLVM_TOOL_CLANG_TOOLS_EXTRA_BUILD=OFF
+	mkdir -p $(LLVM_BUILDDIR); cd $(LLVM_BUILDDIR); cmake -G Ninja $(TINYGO_SOURCE_DIR)/$(LLVM_PROJECTDIR)/llvm "-DLLVM_TARGETS_TO_BUILD=X86;ARM;AArch64;RISCV;WebAssembly" "-DLLVM_EXPERIMENTAL_TARGETS_TO_BUILD=AVR;Xtensa" -DCMAKE_BUILD_TYPE=Release -DLIBCLANG_BUILD_STATIC=ON -DLLVM_ENABLE_TERMINFO=OFF -DLLVM_ENABLE_ZLIB=OFF -DLLVM_ENABLE_LIBEDIT=OFF -DLLVM_ENABLE_Z3_SOLVER=OFF -DLLVM_ENABLE_OCAMLDOC=OFF -DLLVM_ENABLE_LIBXML2=OFF -DLLVM_ENABLE_PROJECTS="clang;lld" -DLLVM_TOOL_CLANG_TOOLS_EXTRA_BUILD=OFF -DCLANG_ENABLE_STATIC_ANALYZER=OFF -DCLANG_ENABLE_ARCMT=OFF $(LLVM_OPTION)
 
 # Build LLVM.
 $(LLVM_BUILDDIR): $(LLVM_BUILDDIR)/build.ninja
-	cd $(LLVM_BUILDDIR); ninja
+	cd $(LLVM_BUILDDIR); ninja $(NINJA_BUILD_TARGETS)
+
+
+# Build wasi-libc sysroot
+.PHONY: wasi-libc
+wasi-libc: lib/wasi-libc/sysroot/lib/wasm32-wasi/libc.a
+lib/wasi-libc/sysroot/lib/wasm32-wasi/libc.a:
+	@if [ ! -e lib/wasi-libc/Makefile ]; then echo "Submodules have not been downloaded. Please download them using:\n  git submodule update --init"; exit 1; fi
+	cd lib/wasi-libc && make -j4 WASM_CC=$(CLANG) WASM_AR=$(LLVM_AR) WASM_NM=$(LLVM_NM)
 
 
 # Build the Go compiler.
-build/tinygo:
+tinygo:
 	@if [ ! -f "$(LLVM_BUILDDIR)/bin/llvm-config" ]; then echo "Fetch and build LLVM first by running:"; echo "  make llvm-source"; echo "  make $(LLVM_BUILDDIR)"; exit 1; fi
-	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" go build -o build/tinygo -tags byollvm .
+	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GO) build -buildmode exe -o build/tinygo$(EXE) -tags byollvm -ldflags="-X main.gitSha1=`git rev-parse --short HEAD`" .
 
-test:
-	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" go test -v -tags byollvm ./transform .
+test: wasi-libc
+	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GO) test -v -buildmode exe -tags byollvm ./builder ./cgo ./compileopts ./compiler ./interp ./transform .
 
+TEST_PACKAGES = \
+	container/heap \
+	container/list \
+	container/ring \
+	crypto/des \
+	encoding \
+	encoding/ascii85 \
+	encoding/base32 \
+	encoding/hex \
+	hash/adler32 \
+	hash/fnv \
+	hash/crc64 \
+	math \
+	math/cmplx \
+	text/scanner \
+	unicode/utf8 \
+
+# Test known-working standard library packages.
+# TODO: parallelize, and only show failing tests (no implied -v flag).
+.PHONY: tinygo-test
 tinygo-test:
-	cd tests/tinygotest && tinygo test
+	$(TINYGO) test $(TEST_PACKAGES)
 
 .PHONY: smoketest
 smoketest:
-	# test all examples
-	tinygo build -size short -o test.elf -target=pca10040            examples/blinky1
-	tinygo build -size short -o test.elf -target=pca10040            examples/adc
-	tinygo build -size short -o test.elf -target=pca10040            examples/blinkm
-	tinygo build -size short -o test.elf -target=pca10040            examples/blinky2
-	tinygo build -size short -o test.elf -target=pca10040            examples/button
-	tinygo build -size short -o test.elf -target=pca10040            examples/button2
-	tinygo build -size short -o test.elf -target=pca10040            examples/echo
-	tinygo build -size short -o test.elf -target=circuitplay-express examples/i2s
-	tinygo build -size short -o test.elf -target=pca10040            examples/mcp3008
-	tinygo build -size short -o test.elf -target=microbit            examples/microbit-blink
-	tinygo build -size short -o test.elf -target=pca10040            examples/pwm
-	tinygo build -size short -o test.elf -target=pca10040            examples/serial
-	tinygo build -size short -o test.elf -target=pca10040            examples/test
+	$(TINYGO) version
+	# test all examples (except pwm)
+	$(TINYGO) build -size short -o test.hex -target=pca10040            examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pca10040            examples/adc
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pca10040            examples/blinkm
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pca10040            examples/blinky2
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pca10040            examples/button
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pca10040            examples/button2
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pca10040            examples/echo
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=circuitplay-express examples/i2s
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pca10040            examples/mcp3008
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pca10040            examples/memstats
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=microbit            examples/microbit-blink
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pca10040            examples/pininterrupt
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pca10040            examples/serial
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pca10040            examples/systick
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pca10040            examples/test
+	@$(MD5SUM) test.hex
+	# test simulated boards on play.tinygo.org
+	$(TINYGO) build             -o test.wasm -tags=arduino              examples/blinky1
+	@$(MD5SUM) test.wasm
+	$(TINYGO) build             -o test.wasm -tags=hifive1b             examples/blinky1
+	@$(MD5SUM) test.wasm
+	$(TINYGO) build             -o test.wasm -tags=reelboard            examples/blinky1
+	@$(MD5SUM) test.wasm
+	$(TINYGO) build             -o test.wasm -tags=pca10040             examples/blinky2
+	@$(MD5SUM) test.wasm
+	$(TINYGO) build             -o test.wasm -tags=pca10056             examples/blinky2
+	@$(MD5SUM) test.wasm
+	$(TINYGO) build             -o test.wasm -tags=circuitplay_express  examples/blinky1
+	@$(MD5SUM) test.wasm
 	# test all targets/boards
-	tinygo build             -o test.wasm -tags=pca10040             examples/blinky2
-	tinygo build -size short -o test.elf -target=microbit            examples/echo
-	tinygo build -size short -o test.elf -target=nrf52840-mdk        examples/blinky1
-	tinygo build -size short -o test.elf -target=pca10031            examples/blinky1
-	tinygo build -size short -o test.elf -target=bluepill            examples/blinky1
-	tinygo build -size short -o test.elf -target=reelboard           examples/blinky1
-	tinygo build -size short -o test.elf -target=reelboard           examples/blinky2
-	tinygo build -size short -o test.elf -target=pca10056            examples/blinky1
-	tinygo build -size short -o test.elf -target=pca10056            examples/blinky2
-	tinygo build -size short -o test.elf -target=itsybitsy-m0        examples/blinky1
-	tinygo build -size short -o test.elf -target=feather-m0          examples/blinky1
-	tinygo build -size short -o test.elf -target=trinket-m0          examples/blinky1
-	tinygo build -size short -o test.elf -target=circuitplay-express examples/blinky1
-	tinygo build -size short -o test.elf -target=stm32f4disco        examples/blinky1
-	tinygo build -size short -o test.elf -target=stm32f4disco        examples/blinky2
-	tinygo build -size short -o test.elf -target=circuitplay-express examples/i2s
-	tinygo build -size short -o test.elf -target=gameboy-advance     examples/gba-display
-	tinygo build -size short -o test.elf -target=itsybitsy-m4        examples/blinky1
-	tinygo build -size short -o test.elf -target=nucleo-f103rb       examples/blinky1
+	$(TINYGO) build -size short -o test.hex -target=pca10040-s132v6     examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=microbit            examples/echo
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=microbit-s110v8     examples/echo
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=microbit-v2         examples/microbit-blink
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=microbit-v2-s113v7  examples/microbit-blink
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=nrf52840-mdk        examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pca10031            examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=reelboard           examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=reelboard           examples/blinky2
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pca10056            examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pca10056            examples/blinky2
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pca10059            examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pca10059            examples/blinky2
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=itsybitsy-m0        examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=feather-m0          examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=trinket-m0          examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=circuitplay-express examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=circuitplay-bluefruit examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=circuitplay-express examples/i2s
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=clue-alpha          examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.gba -target=gameboy-advance     examples/gba-display
+	@$(MD5SUM) test.gba
+	$(TINYGO) build -size short -o test.hex -target=grandcentral-m4     examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=itsybitsy-m4        examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=feather-m4          examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pybadge             examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=metro-m4-airlift    examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pyportal            examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=particle-argon      examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=particle-boron      examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=particle-xenon      examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pinetime-devkit0    examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=x9pro               examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pca10056-s140v7     examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=reelboard-s140v7    examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=wioterminal         examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pygamer             examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=xiao                examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=circuitplay-express examples/dac
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pyportal            examples/dac
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=feather-nrf52840  	examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=feather-nrf52840-sense examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=itsybitsy-nrf52840  examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=qtpy                examples/serial
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=teensy40            examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=teensy36            examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=p1am-100            examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=atsame54-xpro       examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=atsame54-xpro       examples/can
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=feather-m4-can      examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=feather-m4-can      examples/caninterrupt
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=arduino-nano33      examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pico                examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=nano-33-ble         examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=nano-rp2040         examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=feather-rp2040 		examples/blinky1
+	@$(MD5SUM) test.hex
+	# test pwm
+	$(TINYGO) build -size short -o test.hex -target=itsybitsy-m0        examples/pwm
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=itsybitsy-m4        examples/pwm
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=feather-m4          examples/pwm
+	@$(MD5SUM) test.hex
+ifneq ($(STM32), 0)
+	$(TINYGO) build -size short -o test.hex -target=bluepill            examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=feather-stm32f405   examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=lgt92               examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=nucleo-f103rb       examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=nucleo-f722ze       examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=nucleo-l031k6       examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=nucleo-l432kc       examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=nucleo-l552ze       examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=stm32f4disco        examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=stm32f4disco        examples/blinky2
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=stm32f4disco-1      examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=stm32f4disco-1      examples/pwm
+	@$(MD5SUM) test.hex
+endif
 ifneq ($(AVR), 0)
-	tinygo build -size short -o test.elf -target=arduino             examples/blinky1
-	tinygo build -size short -o test.elf -target=digispark           examples/blinky1
+	$(TINYGO) build -size short -o test.hex -target=atmega1284p         examples/serial
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=arduino             examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=arduino             examples/pwm
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=arduino -scheduler=tasks  examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=arduino-mega1280    examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=arduino-mega1280    examples/pwm
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=arduino-nano        examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=digispark           examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=digispark -gc=leaking examples/blinky1
+	@$(MD5SUM) test.hex
 endif
-ifneq ($(RISCV), 0)
-	tinygo build -size short -o test.elf -target=hifive1b            examples/blinky1
+ifneq ($(XTENSA), 0)
+	$(TINYGO) build -size short -o test.bin -target=esp32-mini32      	examples/blinky1
+	@$(MD5SUM) test.bin
+	$(TINYGO) build -size short -o test.bin -target=nodemcu             examples/blinky1
+	@$(MD5SUM) test.bin
 endif
-	tinygo build             -o wasm.wasm -target=wasm               examples/wasm/export
-	tinygo build             -o wasm.wasm -target=wasm               examples/wasm/main
+	$(TINYGO) build -size short -o test.hex -target=hifive1b            examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=hifive1-qemu        examples/serial
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=maixbit             examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build             -o wasm.wasm -target=wasm               examples/wasm/export
+	$(TINYGO) build             -o wasm.wasm -target=wasm               examples/wasm/main
+	# test various compiler flags
+	$(TINYGO) build -size short -o test.hex -target=pca10040 -gc=none -scheduler=none examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pca10040 -opt=1     examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pca10040 -serial=none examples/echo
+	@$(MD5SUM) test.hex
+	$(TINYGO) build             -o test.nro -target=nintendoswitch      examples/serial
+	@$(MD5SUM) test.nro
+	$(TINYGO) build -size short -o test.hex -target=pca10040 -opt=0     ./testdata/stdlib.go
+	@$(MD5SUM) test.hex
 
-release: build/tinygo gen-device
+
+wasmtest:
+	$(GO) test ./tests/wasm
+
+build/release: tinygo gen-device wasi-libc
 	@mkdir -p build/release/tinygo/bin
 	@mkdir -p build/release/tinygo/lib/clang/include
 	@mkdir -p build/release/tinygo/lib/CMSIS/CMSIS
 	@mkdir -p build/release/tinygo/lib/compiler-rt/lib
 	@mkdir -p build/release/tinygo/lib/nrfx
+	@mkdir -p build/release/tinygo/lib/picolibc/newlib/libc
+	@mkdir -p build/release/tinygo/lib/wasi-libc
 	@mkdir -p build/release/tinygo/pkg/armv6m-none-eabi
 	@mkdir -p build/release/tinygo/pkg/armv7m-none-eabi
 	@mkdir -p build/release/tinygo/pkg/armv7em-none-eabi
 	@echo copying source files
-	@cp -p  build/tinygo                 build/release/tinygo/bin
+	@cp -p  build/tinygo$(EXE)           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
@@ -155,9 +469,28 @@ release: build/tinygo gen-device
 	@cp -rp lib/compiler-rt/LICENSE.TXT  build/release/tinygo/lib/compiler-rt
 	@cp -rp lib/compiler-rt/README.txt   build/release/tinygo/lib/compiler-rt
 	@cp -rp lib/nrfx/*                   build/release/tinygo/lib/nrfx
+	@cp -rp lib/picolibc/newlib/libc/ctype       build/release/tinygo/lib/picolibc/newlib/libc
+	@cp -rp lib/picolibc/newlib/libc/include     build/release/tinygo/lib/picolibc/newlib/libc
+	@cp -rp lib/picolibc/newlib/libc/locale      build/release/tinygo/lib/picolibc/newlib/libc
+	@cp -rp lib/picolibc/newlib/libc/string      build/release/tinygo/lib/picolibc/newlib/libc
+	@cp -rp lib/picolibc/newlib/libc/tinystdio   build/release/tinygo/lib/picolibc/newlib/libc
+	@cp -rp lib/picolibc-include         build/release/tinygo/lib
+	@cp -rp lib/wasi-libc/sysroot        build/release/tinygo/lib/wasi-libc/sysroot
 	@cp -rp src                          build/release/tinygo/src
 	@cp -rp targets                      build/release/tinygo/targets
-	./build/tinygo build-builtins -target=armv6m-none-eabi  -o build/release/tinygo/pkg/armv6m-none-eabi/compiler-rt.a
-	./build/tinygo build-builtins -target=armv7m-none-eabi  -o build/release/tinygo/pkg/armv7m-none-eabi/compiler-rt.a
-	./build/tinygo build-builtins -target=armv7em-none-eabi -o build/release/tinygo/pkg/armv7em-none-eabi/compiler-rt.a
+	./build/tinygo build-library -target=armv6m-none-eabi  -o build/release/tinygo/pkg/armv6m-none-eabi/compiler-rt.a compiler-rt
+	./build/tinygo build-library -target=armv7m-none-eabi  -o build/release/tinygo/pkg/armv7m-none-eabi/compiler-rt.a compiler-rt
+	./build/tinygo build-library -target=armv7em-none-eabi -o build/release/tinygo/pkg/armv7em-none-eabi/compiler-rt.a compiler-rt
+	./build/tinygo build-library -target=armv6m-none-eabi  -o build/release/tinygo/pkg/armv6m-none-eabi/picolibc.a picolibc
+	./build/tinygo build-library -target=armv7m-none-eabi  -o build/release/tinygo/pkg/armv7m-none-eabi/picolibc.a picolibc
+	./build/tinygo build-library -target=armv7em-none-eabi -o build/release/tinygo/pkg/armv7em-none-eabi/picolibc.a picolibc
+
+release: build/release
 	tar -czf build/release.tar.gz -C build/release tinygo
+
+deb: build/release
+	@mkdir -p build/release-deb/usr/local/bin
+	@mkdir -p build/release-deb/usr/local/lib
+	cp -ar build/release/tinygo build/release-deb/usr/local/lib/tinygo
+	ln -sf ../lib/tinygo/bin/tinygo build/release-deb/usr/local/bin/tinygo
+	fpm -f -s dir -t deb -n tinygo -v $(shell grep "const Version = " goenv/version.go | awk '{print $$NF}') -m '@tinygo-org' --description='TinyGo is a Go compiler for small places.' --license='BSD 3-Clause' --url=https://tinygo.org/ --deb-changelog CHANGELOG.md -p build/release.deb -C ./build/release-deb

README.md

@@ -1,6 +1,6 @@
 # TinyGo - Go compiler for small places
 
-[![CircleCI](https://circleci.com/gh/tinygo-org/tinygo/tree/dev.svg?style=svg)](https://circleci.com/gh/tinygo-org/tinygo/tree/dev)
+[![CircleCI](https://circleci.com/gh/tinygo-org/tinygo/tree/dev.svg?style=svg)](https://circleci.com/gh/tinygo-org/tinygo/tree/dev) [![Build Status](https://dev.azure.com/tinygo/tinygo/_apis/build/status/tinygo-CI?branchName=dev)](https://dev.azure.com/tinygo/tinygo/_build/latest?definitionId=1&branchName=dev)
 
 TinyGo is a Go compiler intended for use in small places such as microcontrollers, WebAssembly (Wasm), and command-line tools.
 
@@ -43,26 +43,76 @@ See the [getting started instructions](https://tinygo.org/getting-started/) for
 
 You can compile TinyGo programs for microcontrollers, WebAssembly and Linux.
 
-The following 18 microcontroller boards are currently supported:
+The following 68 microcontroller boards are currently supported:
 
+* [Adafruit Circuit Playground Bluefruit](https://www.adafruit.com/product/4333)
 * [Adafruit Circuit Playground Express](https://www.adafruit.com/product/3333)
+* [Adafruit CLUE](https://www.adafruit.com/product/4500)
 * [Adafruit Feather M0](https://www.adafruit.com/product/2772)
+* [Adafruit Feather M4](https://www.adafruit.com/product/3857)
+* [Adafruit Feather M4 CAN](https://www.adafruit.com/product/4759)
+* [Adafruit Feather nRF52840 Express](https://www.adafruit.com/product/4062)
+* [Adafruit Feather nRF52840 Sense](https://www.adafruit.com/product/4516)
+* [Adafruit Feather RP2040](https://www.adafruit.com/product/4884)
+* [Adafruit Feather STM32F405 Express](https://www.adafruit.com/product/4382)
+* [Adafruit Grand Central M4](https://www.adafruit.com/product/4064)
 * [Adafruit ItsyBitsy M0](https://www.adafruit.com/product/3727)
 * [Adafruit ItsyBitsy M4](https://www.adafruit.com/product/3800)
+* [Adafruit ItsyBitsy nRF52840](https://www.adafruit.com/product/4481)
+* [Adafruit Matrix Portal M4](https://www.adafruit.com/product/4745)
+* [Adafruit Metro M4 Express Airlift](https://www.adafruit.com/product/4000)
+* [Adafruit PyBadge](https://www.adafruit.com/product/4200)
+* [Adafruit PyGamer](https://www.adafruit.com/product/4242)
+* [Adafruit PyPortal](https://www.adafruit.com/product/4116)
+* [Adafruit QT Py](https://www.adafruit.com/product/4600)
 * [Adafruit Trinket M0](https://www.adafruit.com/product/3500)
-* [Arduino Nano33 IoT](https://store.arduino.cc/nano-33-iot)
+* [Arduino Mega 1280](https://www.arduino.cc/en/Main/arduinoBoardMega/)
+* [Arduino Mega 2560](https://store.arduino.cc/arduino-mega-2560-rev3)
+* [Arduino MKR1000](https://store.arduino.cc/arduino-mkr1000-wifi)
+* [Arduino Nano](https://store.arduino.cc/arduino-nano)
+* [Arduino Nano 33 BLE](https://store.arduino.cc/nano-33-ble)
+* [Arduino Nano 33 BLE Sense](https://store.arduino.cc/nano-33-ble-sense)
+* [Arduino Nano 33 IoT](https://store.arduino.cc/nano-33-iot)
+* [Arduino Nano RP2040 Connect](https://store.arduino.cc/nano-rp2040-connect)
 * [Arduino Uno](https://store.arduino.cc/arduino-uno-rev3)
+* [Arduino Zero](https://store.arduino.cc/usa/arduino-zero)
 * [BBC micro:bit](https://microbit.org/)
-* [Game Boy Advance](https://en.wikipedia.org/wiki/Game_Boy_Advance)
-* [ST Micro STM32F103XX "Bluepill"](http://wiki.stm32duino.com/index.php?title=Blue_Pill)
-* [ST Micro "Nucleo F103RB"](https://www.st.com/en/evaluation-tools/nucleo-f103rb.html)
-* [ST Micro STM32F407 "Discovery"](https://www.st.com/en/evaluation-tools/stm32f4discovery.html)
+* [BBC micro:bit v2](https://microbit.org/new-microbit/)
 * [Digispark](http://digistump.com/products/1)
+* [Dragino LoRaWAN GPS Tracker LGT-92](http://www.dragino.com/products/lora-lorawan-end-node/item/142-lgt-92.html)
+* [ESP32](https://www.espressif.com/en/products/socs/esp32)
+* [ESP8266](https://www.espressif.com/en/products/socs/esp8266)
+* [Game Boy Advance](https://en.wikipedia.org/wiki/Game_Boy_Advance)
+* [Makerdiary nRF52840-MDK](https://wiki.makerdiary.com/nrf52840-mdk/)
+* [Makerdiary nRF52840-MDK USB Dongle](https://wiki.makerdiary.com/nrf52840-mdk-usb-dongle/)
+* [Microchip SAM E54 Xplained Pro](https://www.microchip.com/developmenttools/productdetails/atsame54-xpro)
+* [nice!nano](https://docs.nicekeyboards.com/#/nice!nano/)
+* [Nintendo Switch](https://www.nintendo.com/switch/)
 * [Nordic Semiconductor PCA10031](https://www.nordicsemi.com/eng/Products/nRF51-Dongle)
 * [Nordic Semiconductor PCA10040](https://www.nordicsemi.com/eng/Products/Bluetooth-low-energy/nRF52-DK)
 * [Nordic Semiconductor PCA10056](https://www.nordicsemi.com/Software-and-Tools/Development-Kits/nRF52840-DK)
-* [Makerdiary nRF52840-MDK](https://wiki.makerdiary.com/nrf52840-mdk/)
+* [Nordic Semiconductor pca10059](https://www.nordicsemi.com/Software-and-tools/Development-Kits/nRF52840-Dongle)
+* [Particle Argon](https://docs.particle.io/datasheets/wi-fi/argon-datasheet/)
+* [Particle Boron](https://docs.particle.io/datasheets/cellular/boron-datasheet/)
+* [Particle Xenon](https://docs.particle.io/datasheets/discontinued/xenon-datasheet/)
 * [Phytec reel board](https://www.phytec.eu/product-eu/internet-of-things/reelboard/)
+* [PineTime DevKit](https://www.pine64.org/pinetime/)
+* [PJRC Teensy 3.6](https://www.pjrc.com/store/teensy36.html)
+* [PJRC Teensy 4.0](https://www.pjrc.com/store/teensy40.html)
+* [ProductivityOpen P1AM-100](https://facts-engineering.github.io/modules/P1AM-100/P1AM-100.html)
+* [Raspberry Pi Pico](https://www.raspberrypi.org/products/raspberry-pi-pico/)
+* [Seeed Wio Terminal](https://www.seeedstudio.com/Wio-Terminal-p-4509.html)
+* [Seeed Seeeduino XIAO](https://www.seeedstudio.com/Seeeduino-XIAO-Arduino-Microcontroller-SAMD21-Cortex-M0+-p-4426.html)
+* [Seeed Sipeed MAix BiT](https://www.seeedstudio.com/Sipeed-MAix-BiT-for-RISC-V-AI-IoT-p-2872.html)
+* [SiFIve HiFive1](https://www.sifive.com/boards/hifive1)
+* [ST Micro "Nucleo" F103RB](https://www.st.com/en/evaluation-tools/nucleo-f103rb.html)
+* [ST Micro "Nucleo" F722ZE](https://www.st.com/en/evaluation-tools/nucleo-f722ze.html)
+* [ST Micro "Nucleo" L031K6](https://www.st.com/ja/evaluation-tools/nucleo-l031k6.html)
+* [ST Micro "Nucleo" L432KC](https://www.st.com/ja/evaluation-tools/nucleo-l432kc.html)
+* [ST Micro "Nucleo" L552ZE](https://www.st.com/en/evaluation-tools/nucleo-l552ze-q.html)
+* [ST Micro STM32F103XX "Bluepill"](https://stm32-base.org/boards/STM32F103C8T6-Blue-Pill)
+* [ST Micro STM32F407 "Discovery"](https://www.st.com/en/evaluation-tools/stm32f4discovery.html)
+* [X9 Pro smartwatch](https://github.com/curtpw/nRF5x-device-reverse-engineering/tree/master/X9-nrf52832-activity-tracker/)
 
 For more information, see [this list of boards](https://tinygo.org/microcontrollers/). Pull requests for additional support are welcome!
 
@@ -121,3 +171,7 @@ The original reasoning was: if [Python](https://micropython.org/) can run on mic
 ## License
 
 This project is licensed under the BSD 3-clause license, just like the [Go project](https://golang.org/LICENSE) itself.
+
+Some code has been copied from the LLVM project and is therefore licensed under [a variant of the Apache 2.0 license](http://releases.llvm.org/11.0.0/LICENSE.TXT). This has been clearly indicated in the header of these files.
+
+Some code has been copied and/or ported from Paul Stoffregen's Teensy libraries and is therefore licensed under PJRC's license. This has been clearly indicated in the header of these files.

azure-pipelines.yml

@@ -0,0 +1,88 @@
+# Avoid lengthy LLVM rebuilds on each newly pushed branch. Pull requests will
+# be built anyway.
+trigger:
+- release
+- dev
+
+jobs:
+- job: Build
+  timeoutInMinutes: 240 # 4h
+  pool:
+    vmImage: 'VS2017-Win2016'
+  steps:
+    - task: GoTool@0
+      inputs:
+        version: '1.16'
+    - checkout: self
+      fetchDepth: 1
+    - task: Cache@2
+      displayName: Cache LLVM source
+      inputs:
+        key: llvm-source-11-windows-v1
+        path: llvm-project
+    - task: Bash@3
+      displayName: Download LLVM source
+      inputs:
+        targetType: inline
+        script: make llvm-source
+    - task: CacheBeta@0
+      displayName: Cache LLVM build
+      inputs:
+        key: llvm-build-11-windows-v5
+        path: llvm-build
+    - task: Bash@3
+      displayName: Build LLVM
+      inputs:
+        targetType: inline
+        script: |
+          if [ ! -f llvm-build/lib/liblldELF.a ]
+          then
+            # install dependencies
+            choco install ninja
+            # hack ninja to use fewer jobs
+            echo -e 'C:\\ProgramData\\Chocolatey\\bin\\ninja -j4 %*' > /usr/bin/ninja.bat
+            # build!
+            make llvm-build
+          fi
+    - task: Bash@3
+      displayName: Install QEMU
+      inputs:
+        targetType: inline
+        script: choco install qemu --version=2020.06.12
+    - task: CacheBeta@0
+      displayName: Cache wasi-libc sysroot
+      inputs:
+        key: wasi-libc-sysroot-v4
+        path: lib/wasi-libc/sysroot
+    - task: Bash@3
+      displayName: Build wasi-libc
+      inputs:
+        targetType: inline
+        script: PATH=/usr/bin:$PATH make wasi-libc
+    - task: Bash@3
+      displayName: Test TinyGo
+      inputs:
+        targetType: inline
+        script: |
+          export PATH="$PATH:./llvm-build/bin:/c/Program Files/qemu"
+          unset GOROOT
+          make test
+    - task: Bash@3
+      displayName: Build TinyGo release tarball
+      inputs:
+        targetType: inline
+        script: |
+          export PATH="$PATH:./llvm-build/bin:/c/Program Files/qemu"
+          unset GOROOT
+          make build/release -j4
+    - publish: $(System.DefaultWorkingDirectory)/build/release/tinygo
+      displayName: Publish zip as artifact
+      artifact: tinygo
+    - task: Bash@3
+      displayName: Smoke tests
+      inputs:
+        targetType: inline
+        script: |
+          export PATH="$PATH:./llvm-build/bin:/c/Program Files/qemu"
+          unset GOROOT
+          make smoketest TINYGO=build/tinygo AVR=0 XTENSA=0

builder/ar.go

@@ -0,0 +1,182 @@
+package builder
+
+import (
+	"bytes"
+	"debug/elf"
+	"encoding/binary"
+	"errors"
+	"io"
+	"os"
+	"path/filepath"
+	"time"
+
+	"github.com/blakesmith/ar"
+)
+
+// makeArchive creates an arcive for static linking from a list of object files
+// given as a parameter. It is equivalent to the following command:
+//
+//     ar -rcs <archivePath> <objs...>
+func makeArchive(archivePath string, objs []string) error {
+	// Open the archive file.
+	arfile, err := os.Create(archivePath)
+	if err != nil {
+		return err
+	}
+	defer arfile.Close()
+	arwriter := ar.NewWriter(arfile)
+	err = arwriter.WriteGlobalHeader()
+	if err != nil {
+		return &os.PathError{"write ar header", archivePath, err}
+	}
+
+	// Open all object files and read the symbols for the symbol table.
+	symbolTable := []struct {
+		name      string // symbol name
+		fileIndex int    // index into objfiles
+	}{}
+	objfiles := make([]struct {
+		file          *os.File
+		archiveOffset int32
+	}, len(objs))
+	for i, objpath := range objs {
+		objfile, err := os.Open(objpath)
+		if err != nil {
+			return err
+		}
+		objfiles[i].file = objfile
+
+		// Read the symbols and add them to the symbol table.
+		dbg, err := elf.NewFile(objfile)
+		if err != nil {
+			return err
+		}
+		symbols, err := dbg.Symbols()
+		if err != nil {
+			return err
+		}
+		for _, symbol := range symbols {
+			bind := elf.ST_BIND(symbol.Info)
+			if bind != elf.STB_GLOBAL && bind != elf.STB_WEAK {
+				// Don't include local symbols (STB_LOCAL).
+				continue
+			}
+			if elf.ST_TYPE(symbol.Info) != elf.STT_FUNC {
+				// Not a function.
+				// TODO: perhaps globals variables should also be included?
+				continue
+			}
+			// Include in archive.
+			symbolTable = append(symbolTable, struct {
+				name      string
+				fileIndex int
+			}{symbol.Name, i})
+		}
+	}
+
+	// Create the symbol table buffer.
+	// For some (sparse) details on the file format:
+	// https://en.wikipedia.org/wiki/Ar_(Unix)#System_V_(or_GNU)_variant
+	buf := &bytes.Buffer{}
+	binary.Write(buf, binary.BigEndian, int32(len(symbolTable)))
+	for range symbolTable {
+		// This is a placeholder index, it will be updated after all files have
+		// been written to the archive (see the end of this function).
+		err = binary.Write(buf, binary.BigEndian, int32(0))
+		if err != nil {
+			return err
+		}
+	}
+	for _, sym := range symbolTable {
+		_, err := buf.Write([]byte(sym.name + "\x00"))
+		if err != nil {
+			return err
+		}
+	}
+	for buf.Len()%2 != 0 {
+		// The symbol table must be aligned.
+		// This appears to be required by lld.
+		buf.WriteByte(0)
+	}
+
+	// Write the symbol table.
+	err = arwriter.WriteHeader(&ar.Header{
+		Name:    "/",
+		ModTime: time.Unix(0, 0),
+		Uid:     0,
+		Gid:     0,
+		Mode:    0,
+		Size:    int64(buf.Len()),
+	})
+	if err != nil {
+		return err
+	}
+
+	// Keep track of the start of the symbol table.
+	symbolTableStart, err := arfile.Seek(0, os.SEEK_CUR)
+	if err != nil {
+		return err
+	}
+
+	// Write symbol table contents.
+	_, err = arfile.Write(buf.Bytes())
+	if err != nil {
+		return err
+	}
+
+	// Add all object files to the archive.
+	for i, objfile := range objfiles {
+		// Store the start index, for when we'll update the symbol table with
+		// the correct file start indices.
+		offset, err := arfile.Seek(0, os.SEEK_CUR)
+		if err != nil {
+			return err
+		}
+		if int64(int32(offset)) != offset {
+			return errors.New("large archives (4GB+) not supported: " + archivePath)
+		}
+		objfiles[i].archiveOffset = int32(offset)
+
+		// Write the file header.
+		st, err := objfile.file.Stat()
+		if err != nil {
+			return err
+		}
+		err = arwriter.WriteHeader(&ar.Header{
+			Name:    filepath.Base(objfile.file.Name()),
+			ModTime: time.Unix(0, 0),
+			Uid:     0,
+			Gid:     0,
+			Mode:    0644,
+			Size:    st.Size(),
+		})
+		if err != nil {
+			return err
+		}
+
+		// Copy the file contents into the archive.
+		n, err := io.Copy(arwriter, objfile.file)
+		if err != nil {
+			return err
+		}
+		if n != st.Size() {
+			return errors.New("file modified during ar creation: " + archivePath)
+		}
+
+		// File is not needed anymore.
+		objfile.file.Close()
+	}
+
+	// Create symbol indices.
+	indicesBuf := &bytes.Buffer{}
+	for _, sym := range symbolTable {
+		err = binary.Write(indicesBuf, binary.BigEndian, objfiles[sym.fileIndex].archiveOffset)
+		if err != nil {
+			return err
+		}
+	}
+
+	// Overwrite placeholder indices.
+	_, err = arfile.WriteAt(indicesBuf.Bytes(), symbolTableStart+4)
+	return err
+}

builder/buildcache.go

@@ -1,20 +1,13 @@
-package main
+package builder
 
 import (
 	"io"
 	"os"
 	"path/filepath"
 	"time"
-)
 
-// Get the cache directory, usually ~/.cache/tinygo
-func cacheDir() string {
-	dir, err := os.UserCacheDir()
-	if err != nil {
-		panic("could not find cache dir: " + err.Error())
-	}
-	return filepath.Join(dir, "tinygo")
-}
+	"github.com/tinygo-org/tinygo/goenv"
+)
 
 // Return the newest timestamp of all the file paths passed in. Used to check
 // for stale caches.
@@ -37,12 +30,8 @@ func cacheTimestamp(paths []string) (time.Time, error) {
 // Try to load a given file from the cache. Return "", nil if no cached file can
 // be found (or the file is stale), return the absolute path if there is a cache
 // and return an error on I/O errors.
-//
-// TODO: the configKey is currently ignored. It is supposed to be used as extra
-// data for the cache key, like the compiler version and arguments.
-func cacheLoad(name, configKey string, sourceFiles []string) (string, error) {
-	dir := cacheDir()
-	cachepath := filepath.Join(dir, name)
+func cacheLoad(name string, sourceFiles []string) (string, error) {
+	cachepath := filepath.Join(goenv.Get("GOCACHE"), name)
 	cacheStat, err := os.Stat(cachepath)
 	if os.IsNotExist(err) {
 		return "", nil // does not exist
@@ -66,9 +55,7 @@ func cacheLoad(name, configKey string, sourceFiles []string) (string, error) {
 
 // Store the file located at tmppath in the cache with the given name. The
 // tmppath may or may not be gone afterwards.
-//
-// Note: the configKey is ignored, see cacheLoad.
-func cacheStore(tmppath, name, configKey string, sourceFiles []string) (string, error) {
+func cacheStore(tmppath, name string, sourceFiles []string) (string, error) {
 	// get the last modified time
 	if len(sourceFiles) == 0 {
 		panic("cache: no source files")
@@ -76,30 +63,22 @@ func cacheStore(tmppath, name, configKey string, sourceFiles []string) (string,
 
 	// TODO: check the config key
 
-	dir := cacheDir()
+	dir := goenv.Get("GOCACHE")
 	err := os.MkdirAll(dir, 0777)
 	if err != nil {
 		return "", err
 	}
 	cachepath := filepath.Join(dir, name)
-	err = moveFile(tmppath, cachepath)
+	err = copyFile(tmppath, cachepath)
 	if err != nil {
 		return "", err
 	}
 	return cachepath, nil
 }
 
-// moveFile renames the file from src to dst. If renaming doesn't work (for
-// example, the rename crosses a filesystem boundary), the file is copied and
-// the old file is removed.
-func moveFile(src, dst string) error {
-	err := os.Rename(src, dst)
-	if err == nil {
-		// Success!
-		return nil
-	}
-	// Failed to move, probably a different filesystem.
-	// Do a copy + remove.
+// copyFile copies the given file from src to dst. It can copy over
+// a possibly already existing file at the destination.
+func copyFile(src, dst string) error {
 	inf, err := os.Open(src)
 	if err != nil {
 		return err
@@ -117,10 +96,10 @@ func moveFile(src, dst string) error {
 		return err
 	}
 
-	err = os.Rename(dst+".tmp", dst)
+	err = outf.Close()
 	if err != nil {
 		return err
 	}
 
-	return outf.Close()
+	return os.Rename(dst+".tmp", dst)
 }

builder/builtins.go

@@ -0,0 +1,170 @@
+package builder
+
+import (
+	"strings"
+)
+
+// These are the GENERIC_SOURCES according to CMakeList.txt.
+var genericBuiltins = []string{
+	"absvdi2.c",
+	"absvsi2.c",
+	"absvti2.c",
+	"adddf3.c",
+	"addsf3.c",
+	"addtf3.c",
+	"addvdi3.c",
+	"addvsi3.c",
+	"addvti3.c",
+	"apple_versioning.c",
+	"ashldi3.c",
+	"ashlti3.c",
+	"ashrdi3.c",
+	"ashrti3.c",
+	"bswapdi2.c",
+	"bswapsi2.c",
+	"clzdi2.c",
+	"clzsi2.c",
+	"clzti2.c",
+	"cmpdi2.c",
+	"cmpti2.c",
+	"comparedf2.c",
+	"comparesf2.c",
+	"ctzdi2.c",
+	"ctzsi2.c",
+	"ctzti2.c",
+	"divdc3.c",
+	"divdf3.c",
+	"divdi3.c",
+	"divmoddi4.c",
+	"divmodsi4.c",
+	"divsc3.c",
+	"divsf3.c",
+	"divsi3.c",
+	"divtc3.c",
+	"divti3.c",
+	"divtf3.c",
+	"extendsfdf2.c",
+	"extendhfsf2.c",
+	"ffsdi2.c",
+	"ffssi2.c",
+	"ffsti2.c",
+	"fixdfdi.c",
+	"fixdfsi.c",
+	"fixdfti.c",
+	"fixsfdi.c",
+	"fixsfsi.c",
+	"fixsfti.c",
+	"fixunsdfdi.c",
+	"fixunsdfsi.c",
+	"fixunsdfti.c",
+	"fixunssfdi.c",
+	"fixunssfsi.c",
+	"fixunssfti.c",
+	"floatdidf.c",
+	"floatdisf.c",
+	"floatsidf.c",
+	"floatsisf.c",
+	"floattidf.c",
+	"floattisf.c",
+	"floatundidf.c",
+	"floatundisf.c",
+	"floatunsidf.c",
+	"floatunsisf.c",
+	"floatuntidf.c",
+	"floatuntisf.c",
+	//"int_util.c",
+	"lshrdi3.c",
+	"lshrti3.c",
+	"moddi3.c",
+	"modsi3.c",
+	"modti3.c",
+	"muldc3.c",
+	"muldf3.c",
+	"muldi3.c",
+	"mulodi4.c",
+	"mulosi4.c",
+	"muloti4.c",
+	"mulsc3.c",
+	"mulsf3.c",
+	"multi3.c",
+	"multf3.c",
+	"mulvdi3.c",
+	"mulvsi3.c",
+	"mulvti3.c",
+	"negdf2.c",
+	"negdi2.c",
+	"negsf2.c",
+	"negti2.c",
+	"negvdi2.c",
+	"negvsi2.c",
+	"negvti2.c",
+	"os_version_check.c",
+	"paritydi2.c",
+	"paritysi2.c",
+	"parityti2.c",
+	"popcountdi2.c",
+	"popcountsi2.c",
+	"popcountti2.c",
+	"powidf2.c",
+	"powisf2.c",
+	"powitf2.c",
+	"subdf3.c",
+	"subsf3.c",
+	"subvdi3.c",
+	"subvsi3.c",
+	"subvti3.c",
+	"subtf3.c",
+	"trampoline_setup.c",
+	"truncdfhf2.c",
+	"truncdfsf2.c",
+	"truncsfhf2.c",
+	"ucmpdi2.c",
+	"ucmpti2.c",
+	"udivdi3.c",
+	"udivmoddi4.c",
+	"udivmodsi4.c",
+	"udivmodti4.c",
+	"udivsi3.c",
+	"udivti3.c",
+	"umoddi3.c",
+	"umodsi3.c",
+	"umodti3.c",
+}
+
+var aeabiBuiltins = []string{
+	"arm/aeabi_cdcmp.S",
+	"arm/aeabi_cdcmpeq_check_nan.c",
+	"arm/aeabi_cfcmp.S",
+	"arm/aeabi_cfcmpeq_check_nan.c",
+	"arm/aeabi_dcmp.S",
+	"arm/aeabi_div0.c",
+	"arm/aeabi_drsub.c",
+	"arm/aeabi_fcmp.S",
+	"arm/aeabi_frsub.c",
+	"arm/aeabi_idivmod.S",
+	"arm/aeabi_ldivmod.S",
+	"arm/aeabi_memcmp.S",
+	"arm/aeabi_memcpy.S",
+	"arm/aeabi_memmove.S",
+	"arm/aeabi_memset.S",
+	"arm/aeabi_uidivmod.S",
+	"arm/aeabi_uldivmod.S",
+}
+
+// CompilerRT is a library with symbols required by programs compiled with LLVM.
+// These symbols are for operations that cannot be emitted with a single
+// instruction or a short sequence of instructions for that target.
+//
+// For more information, see: https://compiler-rt.llvm.org/
+var CompilerRT = Library{
+	name:      "compiler-rt",
+	cflags:    func() []string { return []string{"-Werror", "-Wall", "-std=c11", "-nostdlibinc"} },
+	sourceDir: "lib/compiler-rt/lib/builtins",
+	sources: func(target string) []string {
+		builtins := append([]string{}, genericBuiltins...) // copy genericBuiltins
+		if strings.HasPrefix(target, "arm") || strings.HasPrefix(target, "thumb") {
+			builtins = append(builtins, aeabiBuiltins...)
+		}
+		return builtins
+	},
+}

builder/cc.go

@@ -0,0 +1,309 @@
+package builder
+
+// This file implements a wrapper around the C compiler (Clang) which uses a
+// build cache.
+
+import (
+	"crypto/sha512"
+	"encoding/hex"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"io"
+	"io/ioutil"
+	"os"
+	"path/filepath"
+	"sort"
+	"strings"
+	"unicode"
+
+	"github.com/tinygo-org/tinygo/goenv"
+	"tinygo.org/x/go-llvm"
+)
+
+// compileAndCacheCFile compiles a C or assembly file using a build cache.
+// Compiling the same file again (if nothing changed, including included header
+// files) the output is loaded from the build cache instead.
+//
+// Its operation is a bit complex (more complex than Go package build caching)
+// because the list of file dependencies is only known after the file is
+// compiled. However, luckily compilers have a flag to write a list of file
+// dependencies in Makefile syntax which can be used for caching.
+//
+// Because of this complexity, every file has in fact two cached build outputs:
+// the file itself, and the list of dependencies. Its operation is as follows:
+//
+//   depfile = hash(path, compiler, cflags, ...)
+//   if depfile exists:
+//     outfile = hash of all files and depfile name
+//     if outfile exists:
+//       # cache hit
+//       return outfile
+//   # cache miss
+//   tmpfile = compile file
+//   read dependencies (side effect of compile)
+//   write depfile
+//   outfile = hash of all files and depfile name
+//   rename tmpfile to outfile
+//
+// There are a few edge cases that are not handled:
+// - If a file is added to an include path, that file may be included instead of
+//   some other file. This would be fixed by also including lookup failures in the
+//   dependencies file, but I'm not aware of a compiler which does that.
+// - The Makefile syntax that compilers output has issues, see readDepFile for
+//   details.
+// - A header file may be changed to add/remove an include. This invalidates the
+//   depfile but without invalidating its name. For this reason, the depfile is
+//   written on each new compilation (even when it seems unnecessary). However, it
+//   could in rare cases lead to a stale file fetched from the cache.
+func compileAndCacheCFile(abspath, tmpdir string, cflags []string, printCommands func(string, ...string)) (string, error) {
+	// Hash input file.
+	fileHash, err := hashFile(abspath)
+	if err != nil {
+		return "", err
+	}
+
+	// Create cache key for the dependencies file.
+	buf, err := json.Marshal(struct {
+		Path        string
+		Hash        string
+		Flags       []string
+		LLVMVersion string
+	}{
+		Path:        abspath,
+		Hash:        fileHash,
+		Flags:       cflags,
+		LLVMVersion: llvm.Version,
+	})
+	if err != nil {
+		panic(err) // shouldn't happen
+	}
+	depfileNameHashBuf := sha512.Sum512_224(buf)
+	depfileNameHash := hex.EncodeToString(depfileNameHashBuf[:])
+
+	// Load dependencies file, if possible.
+	depfileName := "dep-" + depfileNameHash + ".json"
+	depfileCachePath := filepath.Join(goenv.Get("GOCACHE"), depfileName)
+	depfileBuf, err := ioutil.ReadFile(depfileCachePath)
+	var dependencies []string // sorted list of dependency paths
+	if err == nil {
+		// There is a dependency file, that's great!
+		// Parse it first.
+		err := json.Unmarshal(depfileBuf, &dependencies)
+		if err != nil {
+			return "", fmt.Errorf("could not parse dependencies JSON: %w", err)
+		}
+
+		// Obtain hashes of all the files listed as a dependency.
+		outpath, err := makeCFileCachePath(dependencies, depfileNameHash)
+		if err == nil {
+			if _, err := os.Stat(outpath); err == nil {
+				return outpath, nil
+			} else if !os.IsNotExist(err) {
+				return "", err
+			}
+		}
+	} else if !os.IsNotExist(err) {
+		// expected either nil or IsNotExist
+		return "", err
+	}
+
+	objTmpFile, err := ioutil.TempFile(goenv.Get("GOCACHE"), "tmp-*.o")
+	if err != nil {
+		return "", err
+	}
+	objTmpFile.Close()
+	depTmpFile, err := ioutil.TempFile(tmpdir, "dep-*.d")
+	if err != nil {
+		return "", err
+	}
+	depTmpFile.Close()
+	flags := append([]string{}, cflags...)                                   // copy cflags
+	flags = append(flags, "-MD", "-MV", "-MTdeps", "-MF", depTmpFile.Name()) // autogenerate dependencies
+	flags = append(flags, "-c", "-o", objTmpFile.Name(), abspath)
+	if strings.ToLower(filepath.Ext(abspath)) == ".s" {
+		// If this is an assembly file (.s or .S, lowercase or uppercase), then
+		// we'll need to add -Qunused-arguments because many parameters are
+		// relevant to C, not assembly. And with -Werror, having meaningless
+		// flags (for the assembler) is a compiler error.
+		flags = append(flags, "-Qunused-arguments")
+	}
+	if printCommands != nil {
+		printCommands("clang", flags...)
+	}
+	err = runCCompiler(flags...)
+	if err != nil {
+		return "", &commandError{"failed to build", abspath, err}
+	}
+
+	// Create sorted and uniqued slice of dependencies.
+	dependencyPaths, err := readDepFile(depTmpFile.Name())
+	if err != nil {
+		return "", err
+	}
+	dependencyPaths = append(dependencyPaths, abspath) // necessary for .s files
+	dependencySet := make(map[string]struct{}, len(dependencyPaths))
+	var dependencySlice []string
+	for _, path := range dependencyPaths {
+		if _, ok := dependencySet[path]; ok {
+			continue
+		}
+		dependencySet[path] = struct{}{}
+		dependencySlice = append(dependencySlice, path)
+	}
+	sort.Strings(dependencySlice)
+
+	// Write dependencies file.
+	f, err := ioutil.TempFile(filepath.Dir(depfileCachePath), depfileName)
+	buf, err = json.MarshalIndent(dependencySlice, "", "\t")
+	if err != nil {
+		panic(err) // shouldn't happen
+	}
+	_, err = f.Write(buf)
+	if err != nil {
+		return "", err
+	}
+	err = f.Close()
+	if err != nil {
+		return "", err
+	}
+	err = os.Rename(f.Name(), depfileCachePath)
+	if err != nil {
+		return "", err
+	}
+
+	// Move temporary object file to final location.
+	outpath, err := makeCFileCachePath(dependencySlice, depfileNameHash)
+	if err != nil {
+		return "", err
+	}
+	err = os.Rename(objTmpFile.Name(), outpath)
+	if err != nil {
+		return "", err
+	}
+
+	return outpath, nil
+}
+
+// Create a cache path (a path in GOCACHE) to store the output of a compiler
+// job. This path is based on the dep file name (which is a hash of metadata
+// including compiler flags) and the hash of all input files in the paths slice.
+func makeCFileCachePath(paths []string, depfileNameHash string) (string, error) {
+	// Hash all input files.
+	fileHashes := make(map[string]string, len(paths))
+	for _, path := range paths {
+		hash, err := hashFile(path)
+		if err != nil {
+			return "", err
+		}
+		fileHashes[path] = hash
+	}
+
+	// Calculate a cache key based on the above hashes.
+	buf, err := json.Marshal(struct {
+		DepfileHash string
+		FileHashes  map[string]string
+	}{
+		DepfileHash: depfileNameHash,
+		FileHashes:  fileHashes,
+	})
+	if err != nil {
+		panic(err) // shouldn't happen
+	}
+	outFileNameBuf := sha512.Sum512_224(buf)
+	cacheKey := hex.EncodeToString(outFileNameBuf[:])
+
+	outpath := filepath.Join(goenv.Get("GOCACHE"), "obj-"+cacheKey+".o")
+	return outpath, nil
+}
+
+// hashFile hashes the given file path and returns the hash as a hex string.
+func hashFile(path string) (string, error) {
+	f, err := os.Open(path)
+	if err != nil {
+		return "", fmt.Errorf("failed to hash file: %w", err)
+	}
+	defer f.Close()
+	fileHasher := sha512.New512_224()
+	_, err = io.Copy(fileHasher, f)
+	if err != nil {
+		return "", fmt.Errorf("failed to hash file: %w", err)
+	}
+	return hex.EncodeToString(fileHasher.Sum(nil)), nil
+}
+
+// readDepFile reads a dependency file in NMake (Visual Studio make) format. The
+// file is assumed to have a single target named deps.
+//
+// There are roughly three make syntax variants:
+// - BSD make, which doesn't support any escaping. This means that many special
+//   characters are not supported in file names.
+// - GNU make, which supports escaping using a backslash but when it fails to
+//   find a file it tries to fall back with the literal path name (to match BSD
+//   make).
+// - NMake (Visual Studio) and Jom, which simply quote the string if there are
+//   any weird characters.
+// Clang supports two variants: a format that's a compromise between BSD and GNU
+// make (and is buggy to match GCC which is equally buggy), and NMake/Jom, which
+// is at least somewhat sane. This last format isn't perfect either: it does not
+// correctly handle filenames with quote marks in them. Those are generally not
+// allowed on Windows, but of course can be used on POSIX like systems. Still,
+// it's the most sane of any of the formats so readDepFile will use that format.
+func readDepFile(filename string) ([]string, error) {
+	buf, err := ioutil.ReadFile(filename)
+	if err != nil {
+		return nil, err
+	}
+	if len(buf) == 0 {
+		return nil, nil
+	}
+	return parseDepFile(string(buf))
+}
+
+func parseDepFile(s string) ([]string, error) {
+	// This function makes no attempt at parsing anything other than Clang -MD
+	// -MV output.
+
+	// For Windows: replace CRLF with LF to make the logic below simpler.
+	s = strings.ReplaceAll(s, "\r\n", "\n")
+
+	// Collapse all lines ending in a backslash. These backslashes are really
+	// just a way to continue a line without making very long lines.
+	s = strings.ReplaceAll(s, "\\\n", " ")
+
+	// Only use the first line, which is expected to begin with "deps:".
+	line := strings.SplitN(s, "\n", 2)[0]
+	if !strings.HasPrefix(line, "deps:") {
+		return nil, errors.New("readDepFile: expected 'deps:' prefix")
+	}
+	line = strings.TrimSpace(line[len("deps:"):])
+
+	var deps []string
+	for line != "" {
+		if line[0] == '"' {
+			// File path is quoted. Path ends with double quote.
+			// This does not handle double quotes in path names, which is a
+			// problem on non-Windows systems.
+			line = line[1:]
+			end := strings.IndexByte(line, '"')
+			if end < 0 {
+				return nil, errors.New("readDepFile: path is incorrectly quoted")
+			}
+			dep := line[:end]
+			line = strings.TrimSpace(line[end+1:])
+			deps = append(deps, dep)
+		} else {
+			// File path is not quoted. Path ends in space or EOL.
+			end := strings.IndexFunc(line, unicode.IsSpace)
+			if end < 0 {
+				// last dependency
+				deps = append(deps, line)
+				break
+			}
+			dep := line[:end]
+			line = strings.TrimSpace(line[end:])
+			deps = append(deps, dep)
+		}
+	}
+	return deps, nil
+}

builder/cc1as.cpp

@@ -0,0 +1,513 @@
+// +build byollvm
+
+//===-- cc1as.cpp - Clang Assembler  --------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// This is the entry point to the clang -cc1as functionality, which implements
+// the direct interface to the LLVM MC based assembler.
+//
+//===----------------------------------------------------------------------===//
+
+#include "clang/Basic/Diagnostic.h"
+#include "clang/Basic/DiagnosticOptions.h"
+#include "clang/Driver/DriverDiagnostic.h"
+#include "clang/Driver/Options.h"
+#include "clang/Frontend/FrontendDiagnostic.h"
+#include "clang/Frontend/TextDiagnosticPrinter.h"
+#include "clang/Frontend/Utils.h"
+#include "llvm/ADT/STLExtras.h"
+#include "llvm/ADT/StringSwitch.h"
+#include "llvm/ADT/Triple.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/MC/MCAsmBackend.h"
+#include "llvm/MC/MCAsmInfo.h"
+#include "llvm/MC/MCCodeEmitter.h"
+#include "llvm/MC/MCContext.h"
+#include "llvm/MC/MCInstrInfo.h"
+#include "llvm/MC/MCObjectFileInfo.h"
+#include "llvm/MC/MCObjectWriter.h"
+#include "llvm/MC/MCParser/MCAsmParser.h"
+#include "llvm/MC/MCParser/MCTargetAsmParser.h"
+#include "llvm/MC/MCRegisterInfo.h"
+#include "llvm/MC/MCSectionMachO.h"
+#include "llvm/MC/MCStreamer.h"
+#include "llvm/MC/MCSubtargetInfo.h"
+#include "llvm/MC/MCTargetOptions.h"
+#include "llvm/Option/Arg.h"
+#include "llvm/Option/ArgList.h"
+#include "llvm/Option/OptTable.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/FileSystem.h"
+#include "llvm/Support/FormattedStream.h"
+#include "llvm/Support/Host.h"
+#include "llvm/Support/MemoryBuffer.h"
+#include "llvm/Support/Path.h"
+#include "llvm/Support/Process.h"
+#include "llvm/Support/Signals.h"
+#include "llvm/Support/SourceMgr.h"
+#include "llvm/Support/TargetRegistry.h"
+#include "llvm/Support/TargetSelect.h"
+#include "llvm/Support/Timer.h"
+#include "llvm/Support/raw_ostream.h"
+#include <memory>
+#include <system_error>
+using namespace clang;
+using namespace clang::driver;
+using namespace clang::driver::options;
+using namespace llvm;
+using namespace llvm::opt;
+
+#include "cc1as.h"
+
+bool AssemblerInvocation::CreateFromArgs(AssemblerInvocation &Opts,
+                                         ArrayRef<const char *> Argv,
+                                         DiagnosticsEngine &Diags) {
+  bool Success = true;
+
+  // Parse the arguments.
+  const OptTable &OptTbl = getDriverOptTable();
+
+  const unsigned IncludedFlagsBitmask = options::CC1AsOption;
+  unsigned MissingArgIndex, MissingArgCount;
+  InputArgList Args = OptTbl.ParseArgs(Argv, MissingArgIndex, MissingArgCount,
+                                       IncludedFlagsBitmask);
+
+  // Check for missing argument error.
+  if (MissingArgCount) {
+    Diags.Report(diag::err_drv_missing_argument)
+        << Args.getArgString(MissingArgIndex) << MissingArgCount;
+    Success = false;
+  }
+
+  // Issue errors on unknown arguments.
+  for (const Arg *A : Args.filtered(OPT_UNKNOWN)) {
+    auto ArgString = A->getAsString(Args);
+    std::string Nearest;
+    if (OptTbl.findNearest(ArgString, Nearest, IncludedFlagsBitmask) > 1)
+      Diags.Report(diag::err_drv_unknown_argument) << ArgString;
+    else
+      Diags.Report(diag::err_drv_unknown_argument_with_suggestion)
+          << ArgString << Nearest;
+    Success = false;
+  }
+
+  // Construct the invocation.
+
+  // Target Options
+  Opts.Triple = llvm::Triple::normalize(Args.getLastArgValue(OPT_triple));
+  Opts.CPU = std::string(Args.getLastArgValue(OPT_target_cpu));
+  Opts.Features = Args.getAllArgValues(OPT_target_feature);
+
+  // Use the default target triple if unspecified.
+  if (Opts.Triple.empty())
+    Opts.Triple = llvm::sys::getDefaultTargetTriple();
+
+  // Language Options
+  Opts.IncludePaths = Args.getAllArgValues(OPT_I);
+  Opts.NoInitialTextSection = Args.hasArg(OPT_n);
+  Opts.SaveTemporaryLabels = Args.hasArg(OPT_msave_temp_labels);
+  // Any DebugInfoKind implies GenDwarfForAssembly.
+  Opts.GenDwarfForAssembly = Args.hasArg(OPT_debug_info_kind_EQ);
+
+  if (const Arg *A = Args.getLastArg(OPT_compress_debug_sections,
+                                     OPT_compress_debug_sections_EQ)) {
+    if (A->getOption().getID() == OPT_compress_debug_sections) {
+      // TODO: be more clever about the compression type auto-detection
+      Opts.CompressDebugSections = llvm::DebugCompressionType::GNU;
+    } else {
+      Opts.CompressDebugSections =
+          llvm::StringSwitch<llvm::DebugCompressionType>(A->getValue())
+              .Case("none", llvm::DebugCompressionType::None)
+              .Case("zlib", llvm::DebugCompressionType::Z)
+              .Case("zlib-gnu", llvm::DebugCompressionType::GNU)
+              .Default(llvm::DebugCompressionType::None);
+    }
+  }
+
+  Opts.RelaxELFRelocations = Args.hasArg(OPT_mrelax_relocations);
+  Opts.DwarfVersion = getLastArgIntValue(Args, OPT_dwarf_version_EQ, 2, Diags);
+  Opts.DwarfDebugFlags =
+      std::string(Args.getLastArgValue(OPT_dwarf_debug_flags));
+  Opts.DwarfDebugProducer =
+      std::string(Args.getLastArgValue(OPT_dwarf_debug_producer));
+  Opts.DebugCompilationDir =
+      std::string(Args.getLastArgValue(OPT_fdebug_compilation_dir));
+  Opts.MainFileName = std::string(Args.getLastArgValue(OPT_main_file_name));
+
+  for (const auto &Arg : Args.getAllArgValues(OPT_fdebug_prefix_map_EQ)) {
+    auto Split = StringRef(Arg).split('=');
+    Opts.DebugPrefixMap.insert(
+        {std::string(Split.first), std::string(Split.second)});
+  }
+
+  // Frontend Options
+  if (Args.hasArg(OPT_INPUT)) {
+    bool First = true;
+    for (const Arg *A : Args.filtered(OPT_INPUT)) {
+      if (First) {
+        Opts.InputFile = A->getValue();
+        First = false;
+      } else {
+        Diags.Report(diag::err_drv_unknown_argument) << A->getAsString(Args);
+        Success = false;
+      }
+    }
+  }
+  Opts.LLVMArgs = Args.getAllArgValues(OPT_mllvm);
+  Opts.OutputPath = std::string(Args.getLastArgValue(OPT_o));
+  Opts.SplitDwarfOutput =
+      std::string(Args.getLastArgValue(OPT_split_dwarf_output));
+  if (Arg *A = Args.getLastArg(OPT_filetype)) {
+    StringRef Name = A->getValue();
+    unsigned OutputType = StringSwitch<unsigned>(Name)
+      .Case("asm", FT_Asm)
+      .Case("null", FT_Null)
+      .Case("obj", FT_Obj)
+      .Default(~0U);
+    if (OutputType == ~0U) {
+      Diags.Report(diag::err_drv_invalid_value) << A->getAsString(Args) << Name;
+      Success = false;
+    } else
+      Opts.OutputType = FileType(OutputType);
+  }
+  Opts.ShowHelp = Args.hasArg(OPT_help);
+  Opts.ShowVersion = Args.hasArg(OPT_version);
+
+  // Transliterate Options
+  Opts.OutputAsmVariant =
+      getLastArgIntValue(Args, OPT_output_asm_variant, 0, Diags);
+  Opts.ShowEncoding = Args.hasArg(OPT_show_encoding);
+  Opts.ShowInst = Args.hasArg(OPT_show_inst);
+
+  // Assemble Options
+  Opts.RelaxAll = Args.hasArg(OPT_mrelax_all);
+  Opts.NoExecStack = Args.hasArg(OPT_mno_exec_stack);
+  Opts.FatalWarnings = Args.hasArg(OPT_massembler_fatal_warnings);
+  Opts.NoWarn = Args.hasArg(OPT_massembler_no_warn);
+  Opts.RelocationModel =
+      std::string(Args.getLastArgValue(OPT_mrelocation_model, "pic"));
+  Opts.TargetABI = std::string(Args.getLastArgValue(OPT_target_abi));
+  Opts.IncrementalLinkerCompatible =
+      Args.hasArg(OPT_mincremental_linker_compatible);
+  Opts.SymbolDefs = Args.getAllArgValues(OPT_defsym);
+
+  // EmbedBitcode Option. If -fembed-bitcode is enabled, set the flag.
+  // EmbedBitcode behaves the same for all embed options for assembly files.
+  if (auto *A = Args.getLastArg(OPT_fembed_bitcode_EQ)) {
+    Opts.EmbedBitcode = llvm::StringSwitch<unsigned>(A->getValue())
+                            .Case("all", 1)
+                            .Case("bitcode", 1)
+                            .Case("marker", 1)
+                            .Default(0);
+  }
+
+  return Success;
+}
+
+static std::unique_ptr<raw_fd_ostream>
+getOutputStream(StringRef Path, DiagnosticsEngine &Diags, bool Binary) {
+  // Make sure that the Out file gets unlinked from the disk if we get a
+  // SIGINT.
+  if (Path != "-")
+    sys::RemoveFileOnSignal(Path);
+
+  std::error_code EC;
+  auto Out = std::make_unique<raw_fd_ostream>(
+      Path, EC, (Binary ? sys::fs::OF_None : sys::fs::OF_Text));
+  if (EC) {
+    Diags.Report(diag::err_fe_unable_to_open_output) << Path << EC.message();
+    return nullptr;
+  }
+
+  return Out;
+}
+
+bool ExecuteAssembler(AssemblerInvocation &Opts, DiagnosticsEngine &Diags) {
+  // Get the target specific parser.
+  std::string Error;
+  const Target *TheTarget = TargetRegistry::lookupTarget(Opts.Triple, Error);
+  if (!TheTarget)
+    return Diags.Report(diag::err_target_unknown_triple) << Opts.Triple;
+
+  ErrorOr<std::unique_ptr<MemoryBuffer>> Buffer =
+      MemoryBuffer::getFileOrSTDIN(Opts.InputFile);
+
+  if (std::error_code EC = Buffer.getError()) {
+    Error = EC.message();
+    return Diags.Report(diag::err_fe_error_reading) << Opts.InputFile;
+  }
+
+  SourceMgr SrcMgr;
+
+  // Tell SrcMgr about this buffer, which is what the parser will pick up.
+  unsigned BufferIndex = SrcMgr.AddNewSourceBuffer(std::move(*Buffer), SMLoc());
+
+  // Record the location of the include directories so that the lexer can find
+  // it later.
+  SrcMgr.setIncludeDirs(Opts.IncludePaths);
+
+  std::unique_ptr<MCRegisterInfo> MRI(TheTarget->createMCRegInfo(Opts.Triple));
+  assert(MRI && "Unable to create target register info!");
+
+  MCTargetOptions MCOptions;
+  std::unique_ptr<MCAsmInfo> MAI(
+      TheTarget->createMCAsmInfo(*MRI, Opts.Triple, MCOptions));
+  assert(MAI && "Unable to create target asm info!");
+
+  // Ensure MCAsmInfo initialization occurs before any use, otherwise sections
+  // may be created with a combination of default and explicit settings.
+  MAI->setCompressDebugSections(Opts.CompressDebugSections);
+
+  MAI->setRelaxELFRelocations(Opts.RelaxELFRelocations);
+
+  bool IsBinary = Opts.OutputType == AssemblerInvocation::FT_Obj;
+  if (Opts.OutputPath.empty())
+    Opts.OutputPath = "-";
+  std::unique_ptr<raw_fd_ostream> FDOS =
+      getOutputStream(Opts.OutputPath, Diags, IsBinary);
+  if (!FDOS)
+    return true;
+  std::unique_ptr<raw_fd_ostream> DwoOS;
+  if (!Opts.SplitDwarfOutput.empty())
+    DwoOS = getOutputStream(Opts.SplitDwarfOutput, Diags, IsBinary);
+
+  // FIXME: This is not pretty. MCContext has a ptr to MCObjectFileInfo and
+  // MCObjectFileInfo needs a MCContext reference in order to initialize itself.
+  std::unique_ptr<MCObjectFileInfo> MOFI(new MCObjectFileInfo());
+
+  MCContext Ctx(MAI.get(), MRI.get(), MOFI.get(), &SrcMgr, &MCOptions);
+
+  bool PIC = false;
+  if (Opts.RelocationModel == "static") {
+    PIC = false;
+  } else if (Opts.RelocationModel == "pic") {
+    PIC = true;
+  } else {
+    assert(Opts.RelocationModel == "dynamic-no-pic" &&
+           "Invalid PIC model!");
+    PIC = false;
+  }
+
+  MOFI->InitMCObjectFileInfo(Triple(Opts.Triple), PIC, Ctx);
+  if (Opts.SaveTemporaryLabels)
+    Ctx.setAllowTemporaryLabels(false);
+  if (Opts.GenDwarfForAssembly)
+    Ctx.setGenDwarfForAssembly(true);
+  if (!Opts.DwarfDebugFlags.empty())
+    Ctx.setDwarfDebugFlags(StringRef(Opts.DwarfDebugFlags));
+  if (!Opts.DwarfDebugProducer.empty())
+    Ctx.setDwarfDebugProducer(StringRef(Opts.DwarfDebugProducer));
+  if (!Opts.DebugCompilationDir.empty())
+    Ctx.setCompilationDir(Opts.DebugCompilationDir);
+  else {
+    // If no compilation dir is set, try to use the current directory.
+    SmallString<128> CWD;
+    if (!sys::fs::current_path(CWD))
+      Ctx.setCompilationDir(CWD);
+  }
+  if (!Opts.DebugPrefixMap.empty())
+    for (const auto &KV : Opts.DebugPrefixMap)
+      Ctx.addDebugPrefixMapEntry(KV.first, KV.second);
+  if (!Opts.MainFileName.empty())
+    Ctx.setMainFileName(StringRef(Opts.MainFileName));
+  Ctx.setDwarfVersion(Opts.DwarfVersion);
+  if (Opts.GenDwarfForAssembly)
+    Ctx.setGenDwarfRootFile(Opts.InputFile,
+                            SrcMgr.getMemoryBuffer(BufferIndex)->getBuffer());
+
+  // Build up the feature string from the target feature list.
+  std::string FS = llvm::join(Opts.Features, ",");
+
+  std::unique_ptr<MCStreamer> Str;
+
+  std::unique_ptr<MCInstrInfo> MCII(TheTarget->createMCInstrInfo());
+  std::unique_ptr<MCSubtargetInfo> STI(
+      TheTarget->createMCSubtargetInfo(Opts.Triple, Opts.CPU, FS));
+
+  raw_pwrite_stream *Out = FDOS.get();
+  std::unique_ptr<buffer_ostream> BOS;
+
+  MCOptions.MCNoWarn = Opts.NoWarn;
+  MCOptions.MCFatalWarnings = Opts.FatalWarnings;
+  MCOptions.ABIName = Opts.TargetABI;
+
+  // FIXME: There is a bit of code duplication with addPassesToEmitFile.
+  if (Opts.OutputType == AssemblerInvocation::FT_Asm) {
+    MCInstPrinter *IP = TheTarget->createMCInstPrinter(
+        llvm::Triple(Opts.Triple), Opts.OutputAsmVariant, *MAI, *MCII, *MRI);
+
+    std::unique_ptr<MCCodeEmitter> CE;
+    if (Opts.ShowEncoding)
+      CE.reset(TheTarget->createMCCodeEmitter(*MCII, *MRI, Ctx));
+    std::unique_ptr<MCAsmBackend> MAB(
+        TheTarget->createMCAsmBackend(*STI, *MRI, MCOptions));
+
+    auto FOut = std::make_unique<formatted_raw_ostream>(*Out);
+    Str.reset(TheTarget->createAsmStreamer(
+        Ctx, std::move(FOut), /*asmverbose*/ true,
+        /*useDwarfDirectory*/ true, IP, std::move(CE), std::move(MAB),
+        Opts.ShowInst));
+  } else if (Opts.OutputType == AssemblerInvocation::FT_Null) {
+    Str.reset(createNullStreamer(Ctx));
+  } else {
+    assert(Opts.OutputType == AssemblerInvocation::FT_Obj &&
+           "Invalid file type!");
+    if (!FDOS->supportsSeeking()) {
+      BOS = std::make_unique<buffer_ostream>(*FDOS);
+      Out = BOS.get();
+    }
+
+    std::unique_ptr<MCCodeEmitter> CE(
+        TheTarget->createMCCodeEmitter(*MCII, *MRI, Ctx));
+    std::unique_ptr<MCAsmBackend> MAB(
+        TheTarget->createMCAsmBackend(*STI, *MRI, MCOptions));
+    std::unique_ptr<MCObjectWriter> OW =
+        DwoOS ? MAB->createDwoObjectWriter(*Out, *DwoOS)
+              : MAB->createObjectWriter(*Out);
+
+    Triple T(Opts.Triple);
+    Str.reset(TheTarget->createMCObjectStreamer(
+        T, Ctx, std::move(MAB), std::move(OW), std::move(CE), *STI,
+        Opts.RelaxAll, Opts.IncrementalLinkerCompatible,
+        /*DWARFMustBeAtTheEnd*/ true));
+    Str.get()->InitSections(Opts.NoExecStack);
+  }
+
+  // When -fembed-bitcode is passed to clang_as, a 1-byte marker
+  // is emitted in __LLVM,__asm section if the object file is MachO format.
+  if (Opts.EmbedBitcode && Ctx.getObjectFileInfo()->getObjectFileType() ==
+                               MCObjectFileInfo::IsMachO) {
+    MCSection *AsmLabel = Ctx.getMachOSection(
+        "__LLVM", "__asm", MachO::S_REGULAR, 4, SectionKind::getReadOnly());
+    Str.get()->SwitchSection(AsmLabel);
+    Str.get()->emitZeros(1);
+  }
+
+  // Assembly to object compilation should leverage assembly info.
+  Str->setUseAssemblerInfoForParsing(true);
+
+  bool Failed = false;
+
+  std::unique_ptr<MCAsmParser> Parser(
+      createMCAsmParser(SrcMgr, Ctx, *Str.get(), *MAI));
+
+  // FIXME: init MCTargetOptions from sanitizer flags here.
+  std::unique_ptr<MCTargetAsmParser> TAP(
+      TheTarget->createMCAsmParser(*STI, *Parser, *MCII, MCOptions));
+  if (!TAP)
+    Failed = Diags.Report(diag::err_target_unknown_triple) << Opts.Triple;
+
+  // Set values for symbols, if any.
+  for (auto &S : Opts.SymbolDefs) {
+    auto Pair = StringRef(S).split('=');
+    auto Sym = Pair.first;
+    auto Val = Pair.second;
+    int64_t Value;
+    // We have already error checked this in the driver.
+    Val.getAsInteger(0, Value);
+    Ctx.setSymbolValue(Parser->getStreamer(), Sym, Value);
+  }
+
+  if (!Failed) {
+    Parser->setTargetParser(*TAP.get());
+    Failed = Parser->Run(Opts.NoInitialTextSection);
+  }
+
+  // Close Streamer first.
+  // It might have a reference to the output stream.
+  Str.reset();
+  // Close the output stream early.
+  BOS.reset();
+  FDOS.reset();
+
+  // Delete output file if there were errors.
+  if (Failed) {
+    if (Opts.OutputPath != "-")
+      sys::fs::remove(Opts.OutputPath);
+    if (!Opts.SplitDwarfOutput.empty() && Opts.SplitDwarfOutput != "-")
+      sys::fs::remove(Opts.SplitDwarfOutput);
+  }
+
+  return Failed;
+}
+
+static void LLVMErrorHandler(void *UserData, const std::string &Message,
+                             bool GenCrashDiag) {
+  DiagnosticsEngine &Diags = *static_cast<DiagnosticsEngine*>(UserData);
+
+  Diags.Report(diag::err_fe_error_backend) << Message;
+
+  // We cannot recover from llvm errors.
+  sys::Process::Exit(1);
+}
+
+int cc1as_main(ArrayRef<const char *> Argv, const char *Argv0, void *MainAddr) {
+  // Initialize targets and assembly printers/parsers.
+  InitializeAllTargetInfos();
+  InitializeAllTargetMCs();
+  InitializeAllAsmParsers();
+
+  // Construct our diagnostic client.
+  IntrusiveRefCntPtr<DiagnosticOptions> DiagOpts = new DiagnosticOptions();
+  TextDiagnosticPrinter *DiagClient
+    = new TextDiagnosticPrinter(errs(), &*DiagOpts);
+  DiagClient->setPrefix("clang -cc1as");
+  IntrusiveRefCntPtr<DiagnosticIDs> DiagID(new DiagnosticIDs());
+  DiagnosticsEngine Diags(DiagID, &*DiagOpts, DiagClient);
+
+  // Set an error handler, so that any LLVM backend diagnostics go through our
+  // error handler.
+  ScopedFatalErrorHandler FatalErrorHandler
+    (LLVMErrorHandler, static_cast<void*>(&Diags));
+
+  // Parse the arguments.
+  AssemblerInvocation Asm;
+  if (!AssemblerInvocation::CreateFromArgs(Asm, Argv, Diags))
+    return 1;
+
+  if (Asm.ShowHelp) {
+    getDriverOptTable().PrintHelp(
+        llvm::outs(), "clang -cc1as [options] file...",
+        "Clang Integrated Assembler",
+        /*Include=*/driver::options::CC1AsOption, /*Exclude=*/0,
+        /*ShowAllAliases=*/false);
+    return 0;
+  }
+
+  // Honor -version.
+  //
+  // FIXME: Use a better -version message?
+  if (Asm.ShowVersion) {
+    llvm::cl::PrintVersionMessage();
+    return 0;
+  }
+
+  // Honor -mllvm.
+  //
+  // FIXME: Remove this, one day.
+  if (!Asm.LLVMArgs.empty()) {
+    unsigned NumArgs = Asm.LLVMArgs.size();
+    auto Args = std::make_unique<const char*[]>(NumArgs + 2);
+    Args[0] = "clang (LLVM option parsing)";
+    for (unsigned i = 0; i != NumArgs; ++i)
+      Args[i + 1] = Asm.LLVMArgs[i].c_str();
+    Args[NumArgs + 1] = nullptr;
+    llvm::cl::ParseCommandLineOptions(NumArgs + 1, Args.get());
+  }
+
+  // Execute the invocation, unless there were parsing errors.
+  bool Failed = Diags.hasErrorOccurred() || ExecuteAssembler(Asm, Diags);
+
+  // If any timers were active but haven't been destroyed yet, print their
+  // results now.
+  TimerGroup::printAll(errs());
+  TimerGroup::clearAll();
+
+  return !!Failed;
+}

builder/cc1as.h

@@ -0,0 +1,120 @@
+//===-- cc1as.h - Clang Assembler  ----------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// This is the entry point to the clang -cc1as functionality, which implements
+// the direct interface to the LLVM MC based assembler.
+//
+//===----------------------------------------------------------------------===//
+
+/// Helper class for representing a single invocation of the assembler.
+struct AssemblerInvocation {
+  /// @name Target Options
+  /// @{
+
+  /// The name of the target triple to assemble for.
+  std::string Triple;
+
+  /// If given, the name of the target CPU to determine which instructions
+  /// are legal.
+  std::string CPU;
+
+  /// The list of target specific features to enable or disable -- this should
+  /// be a list of strings starting with '+' or '-'.
+  std::vector<std::string> Features;
+
+  /// The list of symbol definitions.
+  std::vector<std::string> SymbolDefs;
+
+  /// @}
+  /// @name Language Options
+  /// @{
+
+  std::vector<std::string> IncludePaths;
+  unsigned NoInitialTextSection : 1;
+  unsigned SaveTemporaryLabels : 1;
+  unsigned GenDwarfForAssembly : 1;
+  unsigned RelaxELFRelocations : 1;
+  unsigned DwarfVersion;
+  std::string DwarfDebugFlags;
+  std::string DwarfDebugProducer;
+  std::string DebugCompilationDir;
+  std::map<const std::string, const std::string> DebugPrefixMap;
+  llvm::DebugCompressionType CompressDebugSections =
+      llvm::DebugCompressionType::None;
+  std::string MainFileName;
+  std::string SplitDwarfOutput;
+
+  /// @}
+  /// @name Frontend Options
+  /// @{
+
+  std::string InputFile;
+  std::vector<std::string> LLVMArgs;
+  std::string OutputPath;
+  enum FileType {
+    FT_Asm,  ///< Assembly (.s) output, transliterate mode.
+    FT_Null, ///< No output, for timing purposes.
+    FT_Obj   ///< Object file output.
+  };
+  FileType OutputType;
+  unsigned ShowHelp : 1;
+  unsigned ShowVersion : 1;
+
+  /// @}
+  /// @name Transliterate Options
+  /// @{
+
+  unsigned OutputAsmVariant;
+  unsigned ShowEncoding : 1;
+  unsigned ShowInst : 1;
+
+  /// @}
+  /// @name Assembler Options
+  /// @{
+
+  unsigned RelaxAll : 1;
+  unsigned NoExecStack : 1;
+  unsigned FatalWarnings : 1;
+  unsigned NoWarn : 1;
+  unsigned IncrementalLinkerCompatible : 1;
+  unsigned EmbedBitcode : 1;
+
+  /// The name of the relocation model to use.
+  std::string RelocationModel;
+
+  /// The ABI targeted by the backend. Specified using -target-abi. Empty
+  /// otherwise.
+  std::string TargetABI;
+
+  /// @}
+
+public:
+  AssemblerInvocation() {
+    Triple = "";
+    NoInitialTextSection = 0;
+    InputFile = "-";
+    OutputPath = "-";
+    OutputType = FT_Asm;
+    OutputAsmVariant = 0;
+    ShowInst = 0;
+    ShowEncoding = 0;
+    RelaxAll = 0;
+    NoExecStack = 0;
+    FatalWarnings = 0;
+    NoWarn = 0;
+    IncrementalLinkerCompatible = 0;
+    DwarfVersion = 0;
+    EmbedBitcode = 0;
+  }
+
+  static bool CreateFromArgs(AssemblerInvocation &Res,
+                             ArrayRef<const char *> Argv,
+                             DiagnosticsEngine &Diags);
+};
+
+bool ExecuteAssembler(AssemblerInvocation &Opts, DiagnosticsEngine &Diags);

builder/cc_test.go

@@ -0,0 +1,33 @@
+package builder
+
+import (
+	"reflect"
+	"testing"
+)
+
+func TestSplitDepFile(t *testing.T) {
+	for i, tc := range []struct {
+		in  string
+		out []string
+	}{
+		{`deps: foo bar`, []string{"foo", "bar"}},
+		{`deps: foo "bar"`, []string{"foo", "bar"}},
+		{`deps: "foo" bar`, []string{"foo", "bar"}},
+		{`deps: "foo bar"`, []string{"foo bar"}},
+		{`deps: "foo bar" `, []string{"foo bar"}},
+		{"deps: foo\nbar", []string{"foo"}},
+		{"deps: foo \\\nbar", []string{"foo", "bar"}},
+		{"deps: foo\\bar \\\nbaz", []string{"foo\\bar", "baz"}},
+		{"deps: foo\\bar \\\r\n baz", []string{"foo\\bar", "baz"}}, // Windows uses CRLF line endings
+	} {
+		out, err := parseDepFile(tc.in)
+		if err != nil {
+			t.Errorf("test #%d failed: %v", i, err)
+			continue
+		}
+		if !reflect.DeepEqual(out, tc.out) {
+			t.Errorf("test #%d failed: expected %#v but got %#v", i, tc.out, out)
+			continue
+		}
+	}
+}

builder/clang.cpp

@@ -0,0 +1,98 @@
+// +build byollvm
+
+#include <clang/Basic/DiagnosticOptions.h>
+#include <clang/CodeGen/CodeGenAction.h>
+#include <clang/Driver/Compilation.h>
+#include <clang/Driver/Driver.h>
+#include <clang/Frontend/CompilerInstance.h>
+#include <clang/Frontend/CompilerInvocation.h>
+#include <clang/Frontend/FrontendDiagnostic.h>
+#include <clang/Frontend/TextDiagnosticPrinter.h>
+#include <clang/FrontendTool/Utils.h>
+#include <llvm/ADT/IntrusiveRefCntPtr.h>
+#include <llvm/Option/Option.h>
+#include <llvm/Support/Host.h>
+
+using namespace llvm;
+using namespace clang;
+
+#include "cc1as.h"
+
+// This file provides C wrappers for the builtin tools cc1 and cc1as
+// provided by Clang, and calls them as the driver would call them.
+
+extern "C" {
+
+bool tinygo_clang_driver(int argc, char **argv) {
+	std::vector<const char*> args(argv, argv + argc);
+
+	// The compiler invocation needs a DiagnosticsEngine so it can report problems
+	llvm::IntrusiveRefCntPtr<clang::DiagnosticOptions> DiagOpts = new clang::DiagnosticOptions();
+	clang::TextDiagnosticPrinter DiagnosticPrinter(llvm::errs(), &*DiagOpts);
+	clang::DiagnosticsEngine Diags(llvm::IntrusiveRefCntPtr<clang::DiagnosticIDs>(new clang::DiagnosticIDs()), &*DiagOpts, &DiagnosticPrinter, false);
+
+	// Create the clang driver
+	clang::driver::Driver TheDriver(args[0], llvm::sys::getDefaultTargetTriple(), Diags);
+
+	// Create the set of actions to perform
+	std::unique_ptr<clang::driver::Compilation> C(TheDriver.BuildCompilation(args));
+	if (!C) {
+		return false;
+	}
+	const clang::driver::JobList &Jobs = C->getJobs();
+
+	// There may be more than one job, for example for .S files
+	// (preprocessor + assembler).
+	for (auto Cmd : Jobs) {
+		// Select the tool: cc1 or cc1as.
+		const llvm::opt::ArgStringList &CCArgs = Cmd.getArguments();
+
+		if (strcmp(*CCArgs.data(), "-cc1") == 0) {
+			// This is the C frontend.
+			// Initialize a compiler invocation object from the clang (-cc1) arguments.
+			std::unique_ptr<clang::CompilerInstance> Clang(new clang::CompilerInstance());
+			bool success = clang::CompilerInvocation::CreateFromArgs(
+					Clang->getInvocation(),
+					CCArgs,
+					Diags);
+			if (!success) {
+				return false;
+			}
+
+			// Create the actual diagnostics engine.
+			Clang->createDiagnostics();
+			if (!Clang->hasDiagnostics()) {
+				return false;
+			}
+
+			// Execute the frontend actions.
+			success = ExecuteCompilerInvocation(Clang.get());
+			if (!success) {
+				return false;
+			}
+
+		} else if (strcmp(*CCArgs.data(), "-cc1as") == 0) {
+			// This is the assembler frontend. Parse the arguments.
+			AssemblerInvocation Asm;
+			ArrayRef<const char *> Argv = llvm::ArrayRef<const char*>(CCArgs);
+			if (!AssemblerInvocation::CreateFromArgs(Asm, Argv.slice(1), Diags))
+				return false;
+
+			// Execute the invocation, unless there were parsing errors.
+			bool failed = Diags.hasErrorOccurred() || ExecuteAssembler(Asm, Diags);
+			if (failed) {
+				return false;
+			}
+
+		} else {
+			// Unknown tool, print the tool and exit.
+			fprintf(stderr, "unknown tool: %s\n", *CCArgs.data());
+			return false;
+		}
+	}
+
+	// Commands executed successfully.
+	return true;
+}
+
+} // extern "C"

builder/commands.go

@@ -0,0 +1,65 @@
+package builder
+
+import (
+	"errors"
+	"os"
+	"os/exec"
+	"runtime"
+	"strings"
+
+	"tinygo.org/x/go-llvm"
+)
+
+// Commands lists command alternatives for various operating systems. These
+// commands may have a slightly different name across operating systems and
+// distributions or may not even exist in $PATH, in which case absolute paths
+// may be used.
+var commands = map[string][]string{}
+
+func init() {
+	llvmMajor := strings.Split(llvm.Version, ".")[0]
+	commands["clang"] = []string{"clang-" + llvmMajor}
+	commands["ld.lld"] = []string{"ld.lld-" + llvmMajor, "ld.lld"}
+	commands["wasm-ld"] = []string{"wasm-ld-" + llvmMajor, "wasm-ld"}
+	// Add the path to a Homebrew-installed LLVM for ease of use (no need to
+	// manually set $PATH).
+	if runtime.GOOS == "darwin" {
+		prefix := "/usr/local/opt/llvm@" + llvmMajor + "/bin/"
+		commands["clang"] = append(commands["clang"], prefix+"clang-"+llvmMajor)
+		commands["ld.lld"] = append(commands["ld.lld"], prefix+"ld.lld")
+		commands["wasm-ld"] = append(commands["wasm-ld"], prefix+"wasm-ld")
+	}
+	// Add the path for when LLVM was installed with the installer from
+	// llvm.org, which by default doesn't add LLVM to the $PATH environment
+	// variable.
+	if runtime.GOOS == "windows" {
+		commands["clang"] = append(commands["clang"], "clang", "C:\\Program Files\\LLVM\\bin\\clang.exe")
+		commands["ld.lld"] = append(commands["ld.lld"], "lld", "C:\\Program Files\\LLVM\\bin\\lld.exe")
+		commands["wasm-ld"] = append(commands["wasm-ld"], "C:\\Program Files\\LLVM\\bin\\wasm-ld.exe")
+	}
+	// Add the path to LLVM installed from ports.
+	if runtime.GOOS == "freebsd" {
+		prefix := "/usr/local/llvm" + llvmMajor + "/bin/"
+		commands["clang"] = append(commands["clang"], prefix+"clang-"+llvmMajor)
+		commands["ld.lld"] = append(commands["ld.lld"], prefix+"ld.lld")
+		commands["wasm-ld"] = append(commands["wasm-ld"], prefix+"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 || err.Err.Error() == "file does not exist") {
+				// this command was not found, try the next
+				continue
+			}
+			return err
+		}
+		return nil
+	}
+	return errors.New("none of these commands were found in your $PATH: " + strings.Join(cmdNames, " "))
+}

builder/config.go

@@ -0,0 +1,49 @@
+package builder
+
+import (
+	"errors"
+	"fmt"
+
+	"github.com/tinygo-org/tinygo/compileopts"
+	"github.com/tinygo-org/tinygo/goenv"
+)
+
+// NewConfig builds a new Config object from a set of compiler options. It also
+// loads some information from the environment while doing that. For example, it
+// uses the currently active GOPATH (from the goenv package) to determine the Go
+// version to use.
+func NewConfig(options *compileopts.Options) (*compileopts.Config, error) {
+	spec, err := compileopts.LoadTarget(options.Target)
+	if err != nil {
+		return nil, err
+	}
+
+	if options.OpenOCDCommands != nil {
+		// Override the OpenOCDCommands from the target spec if specified on
+		// the command-line
+		spec.OpenOCDCommands = options.OpenOCDCommands
+	}
+
+	goroot := goenv.Get("GOROOT")
+	if goroot == "" {
+		return nil, errors.New("cannot locate $GOROOT, please set it manually")
+	}
+
+	major, minor, err := goenv.GetGorootVersion(goroot)
+	if err != nil {
+		return nil, fmt.Errorf("could not read version from GOROOT (%v): %v", goroot, err)
+	}
+	if major != 1 || minor < 13 || minor > 16 {
+		return nil, fmt.Errorf("requires go version 1.13 through 1.16, got go%d.%d", major, minor)
+	}
+
+	clangHeaderPath := getClangHeaderPath(goenv.Get("TINYGOROOT"))
+
+	return &compileopts.Config{
+		Options:        options,
+		Target:         spec,
+		GoMinorVersion: minor,
+		ClangHeaders:   clangHeaderPath,
+		TestConfig:     options.TestConfig,
+	}, nil
+}

builder/elfpatch.go

@@ -0,0 +1,57 @@
+package builder
+
+import (
+	"debug/elf"
+	"fmt"
+	"os"
+)
+
+func getElfSectionData(executable string, sectionName string) ([]byte, elf.FileHeader, error) {
+	elfFile, err := elf.Open(executable)
+	if err != nil {
+		return nil, elf.FileHeader{}, err
+	}
+	defer elfFile.Close()
+
+	section := elfFile.Section(sectionName)
+	if section == nil {
+		return nil, elf.FileHeader{}, fmt.Errorf("could not find %s section", sectionName)
+	}
+
+	data, err := section.Data()
+
+	return data, elfFile.FileHeader, err
+}
+
+func replaceElfSection(executable string, sectionName string, data []byte) error {
+	fp, err := os.OpenFile(executable, os.O_RDWR, 0)
+	if err != nil {
+		return err
+	}
+	defer fp.Close()
+
+	elfFile, err := elf.Open(executable)
+	if err != nil {
+		return err
+	}
+	defer elfFile.Close()
+
+	section := elfFile.Section(sectionName)
+	if section == nil {
+		return fmt.Errorf("could not find %s section", sectionName)
+	}
+
+	// Implicitly check for compressed sections
+	if section.Size != section.FileSize {
+		return fmt.Errorf("expected section %s to have identical size and file size, got %d and %d", sectionName, section.Size, section.FileSize)
+	}
+
+	// Only permit complete replacement of section
+	if section.Size != uint64(len(data)) {
+		return fmt.Errorf("expected section %s to have size %d, was actually %d", sectionName, len(data), section.Size)
+	}
+
+	// Write the replacement section data
+	_, err = fp.WriteAt(data, int64(section.Offset))
+	return err
+}

builder/env.go

@@ -0,0 +1,89 @@
+package builder
+
+import (
+	"io/ioutil"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"sort"
+	"strings"
+
+	"tinygo.org/x/go-llvm"
+)
+
+// getClangHeaderPath returns the path to the built-in Clang headers. It tries
+// multiple locations, which should make it find the directory when installed in
+// various ways.
+func getClangHeaderPath(TINYGOROOT string) string {
+	// Check whether we're running from the source directory.
+	path := filepath.Join(TINYGOROOT, "llvm-project", "clang", "lib", "Headers")
+	if _, err := os.Stat(path); !os.IsNotExist(err) {
+		return path
+	}
+
+	// Check whether we're running from the installation directory.
+	path = filepath.Join(TINYGOROOT, "lib", "clang", "include")
+	if _, err := os.Stat(path); !os.IsNotExist(err) {
+		return path
+	}
+
+	// It looks like we are built with a system-installed LLVM. Do a last
+	// attempt: try to use Clang headers relative to the clang binary.
+	llvmMajor := strings.Split(llvm.Version, ".")[0]
+	for _, cmdName := range commands["clang"] {
+		binpath, err := exec.LookPath(cmdName)
+		if err == nil {
+			// This should be the command that will also be used by
+			// execCommand. To avoid inconsistencies, make sure we use the
+			// headers relative to this command.
+			binpath, err = filepath.EvalSymlinks(binpath)
+			if err != nil {
+				// Unexpected.
+				return ""
+			}
+			// Example executable:
+			//     /usr/lib/llvm-9/bin/clang
+			// Example include path:
+			//     /usr/lib/llvm-9/lib64/clang/9.0.1/include/
+			llvmRoot := filepath.Dir(filepath.Dir(binpath))
+			clangVersionRoot := filepath.Join(llvmRoot, "lib64", "clang")
+			dirs64, err64 := ioutil.ReadDir(clangVersionRoot)
+			// Example include path:
+			//     /usr/lib/llvm-9/lib/clang/9.0.1/include/
+			clangVersionRoot = filepath.Join(llvmRoot, "lib", "clang")
+			dirs32, err32 := ioutil.ReadDir(clangVersionRoot)
+			if err64 != nil && err32 != nil {
+				// Unexpected.
+				continue
+			}
+			dirnames := make([]string, len(dirs64)+len(dirs32))
+			dirCount := 0
+			for _, d := range dirs32 {
+				name := d.Name()
+				if name == llvmMajor || strings.HasPrefix(name, llvmMajor+".") {
+					dirnames[dirCount] = filepath.Join(llvmRoot, "lib", "clang", name)
+					dirCount++
+				}
+			}
+			for _, d := range dirs64 {
+				name := d.Name()
+				if name == llvmMajor || strings.HasPrefix(name, llvmMajor+".") {
+					dirnames[dirCount] = filepath.Join(llvmRoot, "lib64", "clang", name)
+					dirCount++
+				}
+			}
+			sort.Strings(dirnames)
+			// Check for the highest version first.
+			for i := dirCount - 1; i >= 0; i-- {
+				path := filepath.Join(dirnames[i], "include")
+				_, err := os.Stat(filepath.Join(path, "stdint.h"))
+				if err == nil {
+					return path
+				}
+			}
+		}
+	}
+
+	// Could not find it.
+	return ""
+}

builder/error.go

@@ -0,0 +1,39 @@
+package builder
+
+// 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 the first error, to conform to the error interface. Clients should
+	// really do a type-assertion on *MultiError.
+	return e.Errs[0].Error()
+}
+
+// newMultiError returns a *MultiError if there is more than one error, or
+// returns that error directly when there is only one. Passing an empty slice
+// will lead to a panic.
+func newMultiError(errs []error) error {
+	switch len(errs) {
+	case 0:
+		panic("attempted to create empty MultiError")
+	case 1:
+		return errs[0]
+	default:
+		return &MultiError{errs}
+	}
+}
+
+// commandError is an error type to wrap os/exec.Command errors. This provides
+// some more information regarding what went wrong while running a command.
+type commandError struct {
+	Msg  string
+	File string
+	Err  error
+}
+
+func (e *commandError) Error() string {
+	return e.Msg + " " + e.File + ": " + e.Err.Error()
+}

builder/esp.go

@@ -0,0 +1,153 @@
+package builder
+
+// This file implements support for writing ESP image files. These image files
+// are read by the ROM bootloader so have to be in a particular format.
+//
+// In the future, it may be necessary to implement support for other image
+// formats, such as the ESP8266 image formats (again, used by the ROM bootloader
+// to load the firmware).
+
+import (
+	"bytes"
+	"crypto/sha256"
+	"debug/elf"
+	"encoding/binary"
+	"fmt"
+	"io/ioutil"
+	"sort"
+)
+
+type espImageSegment struct {
+	addr uint32
+	data []byte
+}
+
+// makeESPFirmare converts an input ELF file to an image file for an ESP32 or
+// ESP8266 chip. This is a special purpose image format just for the ESP chip
+// family, and is parsed by the on-chip mask ROM bootloader.
+//
+// The following documentation has been used:
+// https://github.com/espressif/esptool/wiki/Firmware-Image-Format
+// https://github.com/espressif/esp-idf/blob/8fbb63c2a701c22ccf4ce249f43aded73e134a34/components/bootloader_support/include/esp_image_format.h#L58
+// https://github.com/espressif/esptool/blob/master/esptool.py
+func makeESPFirmareImage(infile, outfile, format string) error {
+	inf, err := elf.Open(infile)
+	if err != nil {
+		return err
+	}
+	defer inf.Close()
+
+	// Load all segments to be written to the image. These are actually ELF
+	// sections, not true ELF segments (similar to how esptool does it).
+	var segments []*espImageSegment
+	for _, section := range inf.Sections {
+		if section.Type != elf.SHT_PROGBITS || section.Size == 0 || section.Flags&elf.SHF_ALLOC == 0 {
+			continue
+		}
+		data, err := section.Data()
+		if err != nil {
+			return fmt.Errorf("failed to read section data: %w", err)
+		}
+		for len(data)%4 != 0 {
+			// Align segment to 4 bytes.
+			data = append(data, 0)
+		}
+		if uint64(uint32(section.Addr)) != section.Addr {
+			return fmt.Errorf("section address too big: 0x%x", section.Addr)
+		}
+		segments = append(segments, &espImageSegment{
+			addr: uint32(section.Addr),
+			data: data,
+		})
+	}
+
+	// Sort the segments by address. This is what esptool does too.
+	sort.SliceStable(segments, func(i, j int) bool { return segments[i].addr < segments[j].addr })
+
+	// Calculate checksum over the segment data. This is used in the image
+	// footer.
+	checksum := uint8(0xef)
+	for _, segment := range segments {
+		for _, b := range segment.data {
+			checksum ^= b
+		}
+	}
+
+	// Write first to an in-memory buffer, primarily so that we can easily
+	// calculate a hash over the entire image.
+	// An added benefit is that we don't need to check for errors all the time.
+	outf := &bytes.Buffer{}
+
+	// Image header.
+	switch format {
+	case "esp32":
+		// Header format:
+		// https://github.com/espressif/esp-idf/blob/8fbb63c2/components/bootloader_support/include/esp_image_format.h#L58
+		binary.Write(outf, binary.LittleEndian, struct {
+			magic          uint8
+			segment_count  uint8
+			spi_mode       uint8
+			spi_speed_size uint8
+			entry_addr     uint32
+			wp_pin         uint8
+			spi_pin_drv    [3]uint8
+			reserved       [11]uint8
+			hash_appended  bool
+		}{
+			magic:          0xE9,
+			segment_count:  byte(len(segments)),
+			spi_mode:       0, // irrelevant, replaced by esptool when flashing
+			spi_speed_size: 0, // spi_speed, spi_size: replaced by esptool when flashing
+			entry_addr:     uint32(inf.Entry),
+			wp_pin:         0xEE, // disable WP pin
+			hash_appended:  true, // add a SHA256 hash
+		})
+	case "esp8266":
+		// Header format:
+		// https://github.com/espressif/esptool/wiki/Firmware-Image-Format
+		// Basically a truncated version of the ESP32 header.
+		binary.Write(outf, binary.LittleEndian, struct {
+			magic          uint8
+			segment_count  uint8
+			spi_mode       uint8
+			spi_speed_size uint8
+			entry_addr     uint32
+		}{
+			magic:          0xE9,
+			segment_count:  byte(len(segments)),
+			spi_mode:       0,    // irrelevant, replaced by esptool when flashing
+			spi_speed_size: 0x20, // spi_speed, spi_size: replaced by esptool when flashing
+			entry_addr:     uint32(inf.Entry),
+		})
+	default:
+		return fmt.Errorf("builder: unknown binary format %#v, expected esp32 or esp8266", format)
+	}
+
+	// Write all segments to the image.
+	// https://github.com/espressif/esptool/wiki/Firmware-Image-Format#segment
+	for _, segment := range segments {
+		binary.Write(outf, binary.LittleEndian, struct {
+			addr   uint32
+			length uint32
+		}{
+			addr:   segment.addr,
+			length: uint32(len(segment.data)),
+		})
+		outf.Write(segment.data)
+	}
+
+	// Footer, including checksum.
+	// The entire image size must be a multiple of 16, so pad the image to one
+	// byte less than that before writing the checksum.
+	outf.Write(make([]byte, 15-outf.Len()%16))
+	outf.WriteByte(checksum)
+
+	if format == "esp32" {
+		// SHA256 hash (to protect against image corruption, not for security).
+		hash := sha256.Sum256(outf.Bytes())
+		outf.Write(hash[:])
+	}
+
+	// Write the image to the output file.
+	return ioutil.WriteFile(outfile, outf.Bytes(), 0666)
+}

builder/jobs.go

@@ -0,0 +1,173 @@
+package builder
+
+// This file implements a job runner for the compiler, which runs jobs in
+// parallel while taking care of dependencies.
+
+import (
+	"fmt"
+	"runtime"
+	"time"
+)
+
+// Set to true to enable logging in the job runner. This may help to debug
+// concurrency or performance issues.
+const jobRunnerDebug = false
+
+type jobState uint8
+
+const (
+	jobStateQueued   jobState = iota // not yet running
+	jobStateRunning                  // running
+	jobStateFinished                 // finished running
+)
+
+// compileJob is a single compiler job, comparable to a single Makefile target.
+// It is used to orchestrate various compiler tasks that can be run in parallel
+// but that have dependencies and thus have limitations in how they can be run.
+type compileJob struct {
+	description  string // description, only used for logging
+	dependencies []*compileJob
+	result       string // result (path)
+	run          func(*compileJob) (err error)
+	state        jobState
+	err          error         // error if finished
+	duration     time.Duration // how long it took to run this job (only set after finishing)
+}
+
+// dummyCompileJob returns a new *compileJob that produces an output without
+// doing anything. This can be useful where a *compileJob producing an output is
+// expected but nothing needs to be done, for example for a load from a cache.
+func dummyCompileJob(result string) *compileJob {
+	return &compileJob{
+		description: "<dummy>",
+		result:      result,
+	}
+}
+
+// readyToRun returns whether this job is ready to run: it is itself not yet
+// started and all dependencies are finished.
+func (job *compileJob) readyToRun() bool {
+	if job.state != jobStateQueued {
+		// Already running or finished, so shouldn't be run again.
+		return false
+	}
+
+	// Check dependencies.
+	for _, dep := range job.dependencies {
+		if dep.state != jobStateFinished {
+			// A dependency is not finished, so this job has to wait until it
+			// is.
+			return false
+		}
+	}
+
+	// All conditions are satisfied.
+	return true
+}
+
+// runJobs runs all the jobs indicated in the jobs slice and returns the error
+// of the first job that fails to run.
+// It runs all jobs in the order of the slice, as long as all dependencies have
+// already run. Therefore, if some jobs are preferred to run before others, they
+// should be ordered as such in this slice.
+func runJobs(jobs []*compileJob) error {
+	// Create channels to communicate with the workers.
+	doneChan := make(chan *compileJob)
+	workerChan := make(chan *compileJob)
+	defer close(workerChan)
+
+	// Start a number of workers.
+	for i := 0; i < runtime.NumCPU(); i++ {
+		if jobRunnerDebug {
+			fmt.Println("## starting worker", i)
+		}
+		go jobWorker(workerChan, doneChan)
+	}
+
+	// Send each job in the jobs slice to a worker, taking care of job
+	// dependencies.
+	numRunningJobs := 0
+	var totalTime time.Duration
+	start := time.Now()
+	for {
+		// If there are free workers, try starting a new job (if one is
+		// available). If it succeeds, try again to fill the entire worker pool.
+		if numRunningJobs < runtime.NumCPU() {
+			jobToRun := nextJob(jobs)
+			if jobToRun != nil {
+				// Start job.
+				if jobRunnerDebug {
+					fmt.Println("## start:   ", jobToRun.description)
+				}
+				jobToRun.state = jobStateRunning
+				workerChan <- jobToRun
+				numRunningJobs++
+				continue
+			}
+		}
+
+		// When there are no jobs running, all jobs in the jobs slice must have
+		// been finished. Therefore, the work is done.
+		if numRunningJobs == 0 {
+			break
+		}
+
+		// Wait until a job is finished.
+		job := <-doneChan
+		job.state = jobStateFinished
+		numRunningJobs--
+		totalTime += job.duration
+		if jobRunnerDebug {
+			fmt.Println("## finished:", job.description, "(time "+job.duration.String()+")")
+		}
+		if job.err != nil {
+			// Wait for running jobs to finish.
+			for numRunningJobs != 0 {
+				<-doneChan
+				numRunningJobs--
+			}
+			// Return error of first failing job.
+			return job.err
+		}
+	}
+
+	// Some statistics, if debugging.
+	if jobRunnerDebug {
+		// Total duration of running all jobs.
+		duration := time.Since(start)
+		fmt.Println("## total:   ", duration)
+
+		// The individual time of each job combined. On a multicore system, this
+		// should be lower than the total above.
+		fmt.Println("## job sum: ", totalTime)
+	}
+
+	return nil
+}
+
+// nextJob returns the first ready-to-run job.
+// This is an implementation detail of runJobs.
+func nextJob(jobs []*compileJob) *compileJob {
+	for _, job := range jobs {
+		if job.readyToRun() {
+			return job
+		}
+	}
+	return nil
+}
+
+// jobWorker is the goroutine that runs received jobs.
+// This is an implementation detail of runJobs.
+func jobWorker(workerChan, doneChan chan *compileJob) {
+	for job := range workerChan {
+		start := time.Now()
+		if job.run != nil {
+			err := job.run(job)
+			if err != nil {
+				job.err = err
+			}
+		}
+		job.duration = time.Since(start)
+		doneChan <- job
+	}
+}

builder/library.go

@@ -0,0 +1,149 @@
+package builder
+
+import (
+	"os"
+	"path/filepath"
+	"strings"
+
+	"github.com/tinygo-org/tinygo/goenv"
+)
+
+// Library is a container for information about a single C library, such as a
+// compiler runtime or libc.
+type Library struct {
+	// The library name, such as compiler-rt or picolibc.
+	name string
+
+	cflags func() []string
+
+	// The source directory, relative to TINYGOROOT.
+	sourceDir string
+
+	// The source files, relative to sourceDir.
+	sources func(target string) []string
+}
+
+// fullPath returns the full path to the source directory.
+func (l *Library) fullPath() string {
+	return filepath.Join(goenv.Get("TINYGOROOT"), l.sourceDir)
+}
+
+// sourcePaths returns a slice with the full paths to the source files.
+func (l *Library) sourcePaths(target string) []string {
+	sources := l.sources(target)
+	paths := make([]string, len(sources))
+	for i, name := range sources {
+		paths[i] = filepath.Join(l.fullPath(), name)
+	}
+	return paths
+}
+
+// Load the library archive, possibly generating and caching it if needed.
+// The resulting file is stored in the provided tmpdir, which is expected to be
+// removed after the Load call.
+func (l *Library) Load(target, tmpdir string) (path string, err error) {
+	job, err := l.load(target, "", tmpdir)
+	if err != nil {
+		return "", err
+	}
+	jobs := append([]*compileJob{job}, job.dependencies...)
+	err = runJobs(jobs)
+	return job.result, err
+}
+
+// load returns a compile job to build this library file for the given target
+// and CPU. It may return a dummy compileJob if the library build is already
+// cached. The path is stored as job.result but is only valid if the job and
+// job.dependencies have been run.
+// The provided tmpdir will be used to store intermediary files and possibly the
+// output archive file, it is expected to be removed after use.
+func (l *Library) load(target, cpu, tmpdir string) (job *compileJob, err error) {
+	// Try to load a precompiled library.
+	precompiledPath := filepath.Join(goenv.Get("TINYGOROOT"), "pkg", target, l.name+".a")
+	if _, err := os.Stat(precompiledPath); err == nil {
+		// Found a precompiled library for this OS/architecture. Return the path
+		// directly.
+		return dummyCompileJob(precompiledPath), nil
+	}
+
+	var outfile string
+	if cpu != "" {
+		outfile = l.name + "-" + target + "-" + cpu + ".a"
+	} else {
+		outfile = l.name + "-" + target + ".a"
+	}
+
+	// Try to fetch this library from the cache.
+	if path, err := cacheLoad(outfile, l.sourcePaths(target)); path != "" || err != nil {
+		// Cache hit.
+		return dummyCompileJob(path), nil
+	}
+	// Cache miss, build it now.
+
+	remapDir := filepath.Join(os.TempDir(), "tinygo-"+l.name)
+	dir := filepath.Join(tmpdir, "build-lib-"+l.name)
+	err = os.Mkdir(dir, 0777)
+	if err != nil {
+		return nil, err
+	}
+
+	// Precalculate the flags to the compiler invocation.
+	// 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.
+	args := append(l.cflags(), "-c", "-Oz", "-g", "-ffunction-sections", "-fdata-sections", "-Wno-macro-redefined", "--target="+target, "-fdebug-prefix-map="+dir+"="+remapDir)
+	if cpu != "" {
+		args = append(args, "-mcpu="+cpu)
+	}
+	if strings.HasPrefix(target, "arm") || strings.HasPrefix(target, "thumb") {
+		args = append(args, "-fshort-enums", "-fomit-frame-pointer", "-mfloat-abi=soft")
+	}
+	if strings.HasPrefix(target, "riscv32-") {
+		args = append(args, "-march=rv32imac", "-mabi=ilp32", "-fforce-enable-int128")
+	}
+	if strings.HasPrefix(target, "riscv64-") {
+		args = append(args, "-march=rv64gc", "-mabi=lp64")
+	}
+
+	// Create job to put all the object files in a single archive. This archive
+	// file is the (static) library file.
+	var objs []string
+	arpath := filepath.Join(dir, l.name+".a")
+	job = &compileJob{
+		description: "ar " + l.name + ".a",
+		result:      arpath,
+		run: func(*compileJob) error {
+			// Create an archive of all object files.
+			err := makeArchive(arpath, objs)
+			if err != nil {
+				return err
+			}
+			// Store this archive in the cache.
+			_, err = cacheStore(arpath, outfile, l.sourcePaths(target))
+			return err
+		},
+	}
+
+	// Create jobs to compile all sources. These jobs are depended upon by the
+	// archive job above, so must be run first.
+	for _, srcpath := range l.sourcePaths(target) {
+		srcpath := srcpath // avoid concurrency issues by redefining inside the loop
+		objpath := filepath.Join(dir, filepath.Base(srcpath)+".o")
+		objs = append(objs, objpath)
+		job.dependencies = append(job.dependencies, &compileJob{
+			description: "compile " + srcpath,
+			run: func(*compileJob) error {
+				var compileArgs []string
+				compileArgs = append(compileArgs, args...)
+				compileArgs = append(compileArgs, "-o", objpath, srcpath)
+				err := runCCompiler(compileArgs...)
+				if err != nil {
+					return &commandError{"failed to build", srcpath, err}
+				}
+				return nil
+			},
+		})
+	}
+
+	return job, nil
+}

builder/lld.cpp

@@ -8,12 +8,12 @@ extern "C" {
 
 bool tinygo_link_elf(int argc, char **argv) {
 	std::vector<const char*> args(argv, argv + argc);
-	return lld::elf::link(args, false);
+	return lld::elf::link(args, false, llvm::outs(), llvm::errs());
 }
 
 bool tinygo_link_wasm(int argc, char **argv) {
 	std::vector<const char*> args(argv, argv + argc);
-	return lld::wasm::link(args, false);
+	return lld::wasm::link(args, false, llvm::outs(), llvm::errs());
 }
 
 } // external "C"

builder/nrfutil.go

@@ -0,0 +1,27 @@
+package builder
+
+import (
+	"fmt"
+	"io/ioutil"
+	"os/exec"
+
+	"github.com/tinygo-org/tinygo/compileopts"
+)
+
+// https://infocenter.nordicsemi.com/index.jsp?topic=%2Fug_nrfutil%2FUG%2Fnrfutil%2Fnrfutil_intro.html
+
+func makeDFUFirmwareImage(options *compileopts.Options, infile, outfile string) error {
+	cmdLine := []string{"nrfutil", "pkg", "generate", "--hw-version", "52", "--sd-req", "0x0", "--debug-mode", "--application", infile, outfile}
+
+	if options.PrintCommands != nil {
+		options.PrintCommands(cmdLine[0], cmdLine[1:]...)
+	}
+
+	cmd := exec.Command(cmdLine[0], cmdLine[1:]...)
+	cmd.Stdout = ioutil.Discard
+	err := cmd.Run()
+	if err != nil {
+		return fmt.Errorf("could not run nrfutil pkg generate: %w", err)
+	}
+	return nil
+}

builder/objcopy.go

@@ -1,40 +1,43 @@
-package main
+package builder
 
 import (
 	"debug/elf"
 	"io/ioutil"
 	"os"
-	"path/filepath"
 	"sort"
 
 	"github.com/marcinbor85/gohex"
 )
 
-// ObjcopyError is an error returned by functions that act like objcopy.
-type ObjcopyError struct {
+// maxPadBytes is the maximum allowed bytes to be padded in a rom extraction
+// this value is currently defined by Nintendo Switch Page Alignment (4096 bytes)
+const maxPadBytes = 4095
+
+// objcopyError is an error returned by functions that act like objcopy.
+type objcopyError struct {
 	Op  string
 	Err error
 }
 
-func (e ObjcopyError) Error() string {
+func (e objcopyError) Error() string {
 	if e.Err == nil {
 		return e.Op
 	}
 	return e.Op + ": " + e.Err.Error()
 }
 
-type ProgSlice []*elf.Prog
+type progSlice []*elf.Prog
 
-func (s ProgSlice) Len() int           { return len(s) }
-func (s ProgSlice) Less(i, j int) bool { return s[i].Paddr < s[j].Paddr }
-func (s ProgSlice) Swap(i, j int)      { s[i], s[j] = s[j], s[i] }
+func (s progSlice) Len() int           { return len(s) }
+func (s progSlice) Less(i, j int) bool { return s[i].Paddr < s[j].Paddr }
+func (s progSlice) Swap(i, j int)      { s[i], s[j] = s[j], s[i] }
 
-// ExtractROM extracts a firmware image and the first load address from the
+// extractROM extracts a firmware image and the first load address from the
 // given ELF file. It tries to emulate the behavior of objcopy.
-func ExtractROM(path string) (uint64, []byte, error) {
+func extractROM(path string) (uint64, []byte, error) {
 	f, err := elf.Open(path)
 	if err != nil {
-		return 0, nil, ObjcopyError{"failed to open ELF file to extract text segment", err}
+		return 0, nil, objcopyError{"failed to open ELF file to extract text segment", err}
 	}
 	defer f.Close()
 
@@ -56,26 +59,37 @@ func ExtractROM(path string) (uint64, []byte, error) {
 		}
 	}
 
-	progs := make(ProgSlice, 0, 2)
+	progs := make(progSlice, 0, 2)
 	for _, prog := range f.Progs {
-		if prog.Type != elf.PT_LOAD || prog.Filesz == 0 {
+		if prog.Type != elf.PT_LOAD || prog.Filesz == 0 || prog.Off == 0 {
 			continue
 		}
 		progs = append(progs, prog)
 	}
 	if len(progs) == 0 {
-		return 0, nil, ObjcopyError{"file does not contain ROM segments: " + path, nil}
+		return 0, nil, objcopyError{"file does not contain ROM segments: " + path, nil}
 	}
 	sort.Sort(progs)
 
 	var rom []byte
 	for _, prog := range progs {
+		romEnd := progs[0].Paddr + uint64(len(rom))
+		if prog.Paddr > romEnd && prog.Paddr < romEnd+16 {
+			// Sometimes, the linker seems to insert a bit of padding between
+			// segments. Simply zero-fill these parts.
+			rom = append(rom, make([]byte, prog.Paddr-romEnd)...)
+		}
 		if prog.Paddr != progs[0].Paddr+uint64(len(rom)) {
-			return 0, nil, ObjcopyError{"ROM segments are non-contiguous: " + path, nil}
+			diff := prog.Paddr - (progs[0].Paddr + uint64(len(rom)))
+			if diff > maxPadBytes {
+				return 0, nil, objcopyError{"ROM segments are non-contiguous: " + path, nil}
+			}
+			// Pad the difference
+			rom = append(rom, make([]byte, diff)...)
 		}
 		data, err := ioutil.ReadAll(prog.Open())
 		if err != nil {
-			return 0, nil, ObjcopyError{"failed to extract segment from ELF file: " + path, err}
+			return 0, nil, objcopyError{"failed to extract segment from ELF file: " + path, err}
 		}
 		rom = append(rom, data...)
 	}
@@ -91,9 +105,9 @@ func ExtractROM(path string) (uint64, []byte, error) {
 	}
 }
 
-// Objcopy converts an ELF file to a different (simpler) output file format:
+// objcopy converts an ELF file to a different (simpler) output file format:
 // .bin or .hex. It extracts only the .text section.
-func Objcopy(infile, outfile string) error {
+func objcopy(infile, outfile, binaryFormat string) error {
 	f, err := os.OpenFile(outfile, os.O_RDWR|os.O_CREATE|os.O_TRUNC, 0666)
 	if err != nil {
 		return err
@@ -101,26 +115,26 @@ func Objcopy(infile, outfile string) error {
 	defer f.Close()
 
 	// Read the .text segment.
-	addr, data, err := ExtractROM(infile)
+	addr, data, err := extractROM(infile)
 	if err != nil {
 		return err
 	}
 
 	// Write to the file, in the correct format.
-	switch filepath.Ext(outfile) {
-	case ".bin":
-		// The address is not stored in a .bin file (therefore you
-		// should use .hex files in most cases).
-		_, err := f.Write(data)
-		return err
-	case ".hex":
+	switch binaryFormat {
+	case "hex":
+		// Intel hex file, includes the firmware start address.
 		mem := gohex.NewMemory()
 		err := mem.AddBinary(uint32(addr), data)
 		if err != nil {
-			return ObjcopyError{"failed to create .hex file", err}
+			return objcopyError{"failed to create .hex file", err}
 		}
-		mem.DumpIntelHex(f, 16) // TODO: handle error
-		return nil
+		return mem.DumpIntelHex(f, 16)
+	case "bin":
+		// The start address is not stored in raw firmware files (therefore you
+		// should use .hex files in most cases).
+		_, err := f.Write(data)
+		return err
 	default:
 		panic("unreachable")
 	}

builder/picolibc.go

@@ -0,0 +1,127 @@
+package builder
+
+import (
+	"path/filepath"
+
+	"github.com/tinygo-org/tinygo/goenv"
+)
+
+// Picolibc is a C library for bare metal embedded devices. It was originally
+// based on newlib.
+var Picolibc = Library{
+	name: "picolibc",
+	cflags: func() []string {
+		picolibcDir := filepath.Join(goenv.Get("TINYGOROOT"), "lib/picolibc/newlib/libc")
+		return []string{"-Werror", "-Wall", "-std=gnu11", "-D_COMPILING_NEWLIB", "-nostdlibinc", "-Xclang", "-internal-isystem", "-Xclang", picolibcDir + "/include", "-I" + picolibcDir + "/tinystdio", "-I" + goenv.Get("TINYGOROOT") + "/lib/picolibc-include"}
+	},
+	sourceDir: "lib/picolibc/newlib/libc",
+	sources: func(target string) []string {
+		return picolibcSources
+	},
+}
+
+var picolibcSources = []string{
+	"string/bcmp.c",
+	"string/bcopy.c",
+	"string/bzero.c",
+	"string/explicit_bzero.c",
+	"string/ffsl.c",
+	"string/ffsll.c",
+	"string/fls.c",
+	"string/flsl.c",
+	"string/flsll.c",
+	"string/gnu_basename.c",
+	"string/index.c",
+	"string/memccpy.c",
+	"string/memchr.c",
+	"string/memcmp.c",
+	"string/memcpy.c",
+	"string/memmem.c",
+	"string/memmove.c",
+	"string/mempcpy.c",
+	"string/memrchr.c",
+	"string/memset.c",
+	"string/rawmemchr.c",
+	"string/rindex.c",
+	"string/stpcpy.c",
+	"string/stpncpy.c",
+	"string/strcasecmp.c",
+	"string/strcasecmp_l.c",
+	"string/strcasestr.c",
+	"string/strcat.c",
+	"string/strchr.c",
+	"string/strchrnul.c",
+	"string/strcmp.c",
+	"string/strcoll.c",
+	"string/strcoll_l.c",
+	"string/strcpy.c",
+	"string/strcspn.c",
+	"string/strdup.c",
+	"string/strerror.c",
+	"string/strerror_r.c",
+	"string/strlcat.c",
+	"string/strlcpy.c",
+	"string/strlen.c",
+	"string/strlwr.c",
+	"string/strncasecmp.c",
+	"string/strncasecmp_l.c",
+	"string/strncat.c",
+	"string/strncmp.c",
+	"string/strncpy.c",
+	"string/strndup.c",
+	"string/strnlen.c",
+	"string/strnstr.c",
+	"string/strpbrk.c",
+	"string/strrchr.c",
+	"string/strsep.c",
+	"string/strsignal.c",
+	"string/strspn.c",
+	"string/strstr.c",
+	"string/strtok.c",
+	"string/strtok_r.c",
+	"string/strupr.c",
+	"string/strverscmp.c",
+	"string/strxfrm.c",
+	"string/strxfrm_l.c",
+	"string/swab.c",
+	"string/timingsafe_bcmp.c",
+	"string/timingsafe_memcmp.c",
+	"string/u_strerr.c",
+	"string/wcpcpy.c",
+	"string/wcpncpy.c",
+	"string/wcscasecmp.c",
+	"string/wcscasecmp_l.c",
+	"string/wcscat.c",
+	"string/wcschr.c",
+	"string/wcscmp.c",
+	"string/wcscoll.c",
+	"string/wcscoll_l.c",
+	"string/wcscpy.c",
+	"string/wcscspn.c",
+	"string/wcsdup.c",
+	"string/wcslcat.c",
+	"string/wcslcpy.c",
+	"string/wcslen.c",
+	"string/wcsncasecmp.c",
+	"string/wcsncasecmp_l.c",
+	"string/wcsncat.c",
+	"string/wcsncmp.c",
+	"string/wcsncpy.c",
+	"string/wcsnlen.c",
+	"string/wcspbrk.c",
+	"string/wcsrchr.c",
+	"string/wcsspn.c",
+	"string/wcsstr.c",
+	"string/wcstok.c",
+	"string/wcswidth.c",
+	"string/wcsxfrm.c",
+	"string/wcsxfrm_l.c",
+	"string/wcwidth.c",
+	"string/wmemchr.c",
+	"string/wmemcmp.c",
+	"string/wmemcpy.c",
+	"string/wmemmove.c",
+	"string/wmempcpy.c",
+	"string/wmemset.c",
+	"string/xpg_strerror_r.c",
+}

builder/sizes.go

@@ -1,4 +1,4 @@
-package main
+package builder
 
 import (
 	"debug/elf"
@@ -6,18 +6,18 @@ import (
 	"strings"
 )
 
-// Statistics about code size in a program.
-type ProgramSize struct {
-	Packages map[string]*PackageSize
-	Sum      *PackageSize
+// programSize contains size statistics per package of a compiled program.
+type programSize struct {
+	Packages map[string]*packageSize
+	Sum      *packageSize
 	Code     uint64
 	Data     uint64
 	BSS      uint64
 }
 
-// Return the list of package names (ProgramSize.Packages) sorted
-// alphabetically.
-func (ps *ProgramSize) SortedPackageNames() []string {
+// sortedPackageNames returns the list of package names (ProgramSize.Packages)
+// sorted alphabetically.
+func (ps *programSize) sortedPackageNames() []string {
 	names := make([]string, 0, len(ps.Packages))
 	for name := range ps.Packages {
 		names = append(names, name)
@@ -26,8 +26,9 @@ func (ps *ProgramSize) SortedPackageNames() []string {
 	return names
 }
 
-// The size of a package, calculated from the linked object file.
-type PackageSize struct {
+// packageSize contains the size of a package, calculated from the linked object
+// file.
+type packageSize struct {
 	Code   uint64
 	ROData uint64
 	Data   uint64
@@ -35,12 +36,12 @@ type PackageSize struct {
 }
 
 // Flash usage in regular microcontrollers.
-func (ps *PackageSize) Flash() uint64 {
+func (ps *packageSize) Flash() uint64 {
 	return ps.Code + ps.ROData + ps.Data
 }
 
 // Static RAM usage in regular microcontrollers.
-func (ps *PackageSize) RAM() uint64 {
+func (ps *packageSize) RAM() uint64 {
 	return ps.Data + ps.BSS
 }
 
@@ -64,8 +65,9 @@ func (l symbolList) Swap(i, j int) {
 	l[i], l[j] = l[j], l[i]
 }
 
-// Calculate program/data size breakdown of each package for a given ELF file.
-func Sizes(path string) (*ProgramSize, error) {
+// loadProgramSize calculate a program/data size breakdown of each package for a
+// given ELF file.
+func loadProgramSize(path string) (*programSize, error) {
 	file, err := elf.Open(path)
 	if err != nil {
 		return nil, err
@@ -82,7 +84,18 @@ func Sizes(path string) (*ProgramSize, error) {
 		if section.Type != elf.SHT_PROGBITS && section.Type != elf.SHT_NOBITS {
 			continue
 		}
-		if section.Type == elf.SHT_NOBITS {
+		if section.Name == ".stack" {
+			// HACK: this works around a bug in ld.lld from LLVM 10. The linker
+			// marks sections with no input symbols (such as is the case for the
+			// .stack section) as SHT_PROGBITS instead of SHT_NOBITS. While it
+			// doesn't affect the generated binaries (.hex and .bin), it does
+			// affect the reported size.
+			// https://bugs.llvm.org/show_bug.cgi?id=45336
+			// https://reviews.llvm.org/D76981
+			// It has been merged in master, but it has not (yet) been
+			// backported to the LLVM 10 release branch.
+			sumBSS += section.Size
+		} else if section.Type == elf.SHT_NOBITS {
 			sumBSS += section.Size
 		} else if section.Flags&elf.SHF_EXECINSTR != 0 {
 			sumCode += section.Size
@@ -115,7 +128,7 @@ func Sizes(path string) (*ProgramSize, error) {
 	}
 	sort.Sort(symbolList(symbols))
 
-	sizes := map[string]*PackageSize{}
+	sizes := map[string]*packageSize{}
 	var lastSymbolValue uint64
 	for _, symbol := range symbols {
 		symType := elf.ST_TYPE(symbol.Info)
@@ -129,7 +142,7 @@ func Sizes(path string) (*ProgramSize, error) {
 		}
 		pkgSize := sizes[pkgName]
 		if pkgSize == nil {
-			pkgSize = &PackageSize{}
+			pkgSize = &packageSize{}
 			sizes[pkgName] = pkgSize
 		}
 		if lastSymbolValue != symbol.Value || lastSymbolValue == 0 {
@@ -148,7 +161,7 @@ func Sizes(path string) (*ProgramSize, error) {
 		lastSymbolValue = symbol.Value
 	}
 
-	sum := &PackageSize{}
+	sum := &packageSize{}
 	for _, pkg := range sizes {
 		sum.Code += pkg.Code
 		sum.ROData += pkg.ROData
@@ -156,5 +169,5 @@ func Sizes(path string) (*ProgramSize, error) {
 		sum.BSS += pkg.BSS
 	}
 
-	return &ProgramSize{Packages: sizes, Code: sumCode, Data: sumData, BSS: sumBSS, Sum: sum}, nil
+	return &programSize{Packages: sizes, Code: sumCode, Data: sumData, BSS: sumBSS, Sum: sum}, nil
 }

builder/tools-builtin.go

@@ -0,0 +1,54 @@
+// +build byollvm
+
+package builder
+
+import (
+	"errors"
+	"unsafe"
+)
+
+/*
+#cgo CXXFLAGS: -fno-rtti
+#include <stdbool.h>
+#include <stdlib.h>
+bool tinygo_clang_driver(int argc, char **argv);
+bool tinygo_link_elf(int argc, char **argv);
+bool tinygo_link_wasm(int argc, char **argv);
+*/
+import "C"
+
+const hasBuiltinTools = true
+
+// RunTool runs the given tool (such as clang).
+//
+// This version actually runs the tools because TinyGo was compiled while
+// linking statically with LLVM (with the byollvm build tag).
+func RunTool(tool string, args ...string) error {
+	args = append([]string{"tinygo:" + tool}, args...)
+
+	var cflag *C.char
+	buf := C.calloc(C.size_t(len(args)), C.size_t(unsafe.Sizeof(cflag)))
+	defer C.free(buf)
+	cflags := (*[1 << 10]*C.char)(unsafe.Pointer(buf))[:len(args):len(args)]
+	for i, flag := range args {
+		cflag := C.CString(flag)
+		cflags[i] = cflag
+		defer C.free(unsafe.Pointer(cflag))
+	}
+
+	var ok C.bool
+	switch tool {
+	case "clang":
+		ok = C.tinygo_clang_driver(C.int(len(args)), (**C.char)(buf))
+	case "ld.lld":
+		ok = C.tinygo_link_elf(C.int(len(args)), (**C.char)(buf))
+	case "wasm-ld":
+		ok = C.tinygo_link_wasm(C.int(len(args)), (**C.char)(buf))
+	default:
+		return errors.New("unknown tool: " + tool)
+	}
+	if !ok {
+		return errors.New("failed to run tool: " + tool)
+	}
+	return nil
+}

builder/tools-external.go

@@ -0,0 +1,15 @@
+// +build !byollvm
+
+package builder
+
+import "errors"
+
+const hasBuiltinTools = false
+
+// RunTool runs the given tool (such as clang).
+//
+// This version doesn't actually run the tool: TinyGo has not been compiled by
+// statically linking to LLVM.
+func RunTool(tool string, args ...string) error {
+	return errors.New("cannot run tool: " + tool)
+}

builder/tools.go

@@ -0,0 +1,50 @@
+package builder
+
+import (
+	"errors"
+	"os"
+	"os/exec"
+
+	"github.com/tinygo-org/tinygo/goenv"
+)
+
+// runCCompiler invokes a C compiler with the given arguments.
+func runCCompiler(flags ...string) error {
+	if hasBuiltinTools {
+		// Compile this with the internal Clang compiler.
+		headerPath := getClangHeaderPath(goenv.Get("TINYGOROOT"))
+		if headerPath == "" {
+			return errors.New("could not locate Clang headers")
+		}
+		flags = append(flags, "-I"+headerPath)
+		cmd := exec.Command(os.Args[0], append([]string{"clang"}, flags...)...)
+		cmd.Stdout = os.Stdout
+		cmd.Stderr = os.Stderr
+		return cmd.Run()
+	}
+
+	// Compile this with an external invocation of the Clang compiler.
+	return execCommand(commands["clang"], flags...)
+}
+
+// link invokes a linker with the given name and flags.
+func link(linker string, flags ...string) error {
+	if hasBuiltinTools && (linker == "ld.lld" || linker == "wasm-ld") {
+		// Run command with internal linker.
+		cmd := exec.Command(os.Args[0], append([]string{linker}, flags...)...)
+		cmd.Stdout = os.Stdout
+		cmd.Stderr = os.Stderr
+		return cmd.Run()
+	}
+
+	// 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
+	cmd.Dir = goenv.Get("TINYGOROOT")
+	return cmd.Run()
+}

builder/uf2.go

@@ -1,35 +1,43 @@
-// Converts firmware files from BIN to UF2 format before flashing.
+package builder
+
+// This file converts firmware files from BIN to UF2 format before flashing.
 //
 // For more information about the UF2 firmware file format, please see:
 // https://github.com/Microsoft/uf2
 //
 //
-package main
 
 import (
 	"bytes"
 	"encoding/binary"
 	"io/ioutil"
+	"strconv"
 )
 
-// ConvertELFFileToUF2File converts an ELF file to a UF2 file.
-func ConvertELFFileToUF2File(infile, outfile string) error {
+// convertELFFileToUF2File converts an ELF file to a UF2 file.
+func convertELFFileToUF2File(infile, outfile string, uf2FamilyID string) error {
 	// Read the .text segment.
-	targetAddress, data, err := ExtractROM(infile)
+	targetAddress, data, err := extractROM(infile)
 	if err != nil {
 		return err
 	}
 
-	output, _ := ConvertBinToUF2(data, uint32(targetAddress))
+	output, _, err := convertBinToUF2(data, uint32(targetAddress), uf2FamilyID)
+	if err != nil {
+		return err
+	}
 	return ioutil.WriteFile(outfile, output, 0644)
 }
 
-// ConvertBinToUF2 converts the binary bytes in input to UF2 formatted data.
-func ConvertBinToUF2(input []byte, targetAddr uint32) ([]byte, int) {
+// convertBinToUF2 converts the binary bytes in input to UF2 formatted data.
+func convertBinToUF2(input []byte, targetAddr uint32, uf2FamilyID string) ([]byte, int, error) {
 	blocks := split(input, 256)
 	output := make([]byte, 0)
 
-	bl := NewUF2Block(targetAddr)
+	bl, err := newUF2Block(targetAddr, uf2FamilyID)
+	if err != nil {
+		return nil, 0, err
+	}
 	bl.SetNumBlocks(len(blocks))
 
 	for i := 0; i < len(blocks); i++ {
@@ -40,7 +48,7 @@ func ConvertBinToUF2(input []byte, targetAddr uint32) ([]byte, int) {
 		bl.IncrementAddress(bl.payloadSize)
 	}
 
-	return output, len(blocks)
+	return output, len(blocks), nil
 }
 
 const (
@@ -49,8 +57,8 @@ const (
 	uf2MagicEnd    = 0x0AB16F30 // Ditto
 )
 
-// UF2Block is the structure used for each UF2 code block sent to device.
-type UF2Block struct {
+// uf2Block is the structure used for each UF2 code block sent to device.
+type uf2Block struct {
 	magicStart0 uint32
 	magicStart1 uint32
 	flags       uint32
@@ -63,21 +71,35 @@ type UF2Block struct {
 	magicEnd    uint32
 }
 
-// NewUF2Block returns a new UF2Block struct that has been correctly populated
-func NewUF2Block(targetAddr uint32) *UF2Block {
-	return &UF2Block{magicStart0: uf2MagicStart0,
+// newUF2Block returns a new uf2Block struct that has been correctly populated
+func newUF2Block(targetAddr uint32, uf2FamilyID string) (*uf2Block, error) {
+	var flags uint32
+	var familyID uint32
+	if uf2FamilyID != "" {
+		flags |= flagFamilyIDPresent
+		v, err := strconv.ParseUint(uf2FamilyID, 0, 32)
+		if err != nil {
+			return nil, err
+		}
+		familyID = uint32(v)
+	}
+	return &uf2Block{magicStart0: uf2MagicStart0,
 		magicStart1: uf2MagicStart1,
 		magicEnd:    uf2MagicEnd,
 		targetAddr:  targetAddr,
-		flags:       0x0,
-		familyID:    0x0,
+		flags:       flags,
+		familyID:    familyID,
 		payloadSize: 256,
 		data:        make([]byte, 476),
-	}
+	}, nil
 }
 
-// Bytes converts the UF2Block to a slice of bytes that can be written to file.
-func (b *UF2Block) Bytes() []byte {
+const (
+	flagFamilyIDPresent = 0x00002000
+)
+
+// Bytes converts the uf2Block to a slice of bytes that can be written to file.
+func (b *uf2Block) Bytes() []byte {
 	buf := bytes.NewBuffer(make([]byte, 0, 512))
 	binary.Write(buf, binary.LittleEndian, b.magicStart0)
 	binary.Write(buf, binary.LittleEndian, b.magicStart1)
@@ -94,23 +116,23 @@ func (b *UF2Block) Bytes() []byte {
 }
 
 // IncrementAddress moves the target address pointer forward by count bytes.
-func (b *UF2Block) IncrementAddress(count uint32) {
+func (b *uf2Block) IncrementAddress(count uint32) {
 	b.targetAddr += b.payloadSize
 }
 
 // SetData sets the data to be used for the current block.
-func (b *UF2Block) SetData(d []byte) {
+func (b *uf2Block) SetData(d []byte) {
 	b.data = make([]byte, 476)
 	copy(b.data[:], d)
 }
 
 // SetBlockNo sets the current block number to be used.
-func (b *UF2Block) SetBlockNo(bn int) {
+func (b *uf2Block) SetBlockNo(bn int) {
 	b.blockNo = uint32(bn)
 }
 
 // SetNumBlocks sets the total number of blocks for this UF2 file.
-func (b *UF2Block) SetNumBlocks(total int) {
+func (b *uf2Block) SetNumBlocks(total int) {
 	b.numBlocks = uint32(total)
 }
 

builtins.go

@@ -1,294 +0,0 @@
-package main
-
-import (
-	"errors"
-	"io"
-	"io/ioutil"
-	"os"
-	"path/filepath"
-	"strings"
-	"time"
-
-	"github.com/blakesmith/ar"
-)
-
-// These are the GENERIC_SOURCES according to CMakeList.txt.
-var genericBuiltins = []string{
-	"absvdi2.c",
-	"absvsi2.c",
-	"absvti2.c",
-	"adddf3.c",
-	"addsf3.c",
-	"addtf3.c",
-	"addvdi3.c",
-	"addvsi3.c",
-	"addvti3.c",
-	"apple_versioning.c",
-	"ashldi3.c",
-	"ashlti3.c",
-	"ashrdi3.c",
-	"ashrti3.c",
-	"bswapdi2.c",
-	"bswapsi2.c",
-	"clzdi2.c",
-	"clzsi2.c",
-	"clzti2.c",
-	"cmpdi2.c",
-	"cmpti2.c",
-	"comparedf2.c",
-	"comparesf2.c",
-	"ctzdi2.c",
-	"ctzsi2.c",
-	"ctzti2.c",
-	"divdc3.c",
-	"divdf3.c",
-	"divdi3.c",
-	"divmoddi4.c",
-	"divmodsi4.c",
-	"divsc3.c",
-	"divsf3.c",
-	"divsi3.c",
-	"divtc3.c",
-	"divti3.c",
-	"divtf3.c",
-	"extendsfdf2.c",
-	"extendhfsf2.c",
-	"ffsdi2.c",
-	"ffssi2.c",
-	"ffsti2.c",
-	"fixdfdi.c",
-	"fixdfsi.c",
-	"fixdfti.c",
-	"fixsfdi.c",
-	"fixsfsi.c",
-	"fixsfti.c",
-	"fixunsdfdi.c",
-	"fixunsdfsi.c",
-	"fixunsdfti.c",
-	"fixunssfdi.c",
-	"fixunssfsi.c",
-	"fixunssfti.c",
-	"floatdidf.c",
-	"floatdisf.c",
-	"floatsidf.c",
-	"floatsisf.c",
-	"floattidf.c",
-	"floattisf.c",
-	"floatundidf.c",
-	"floatundisf.c",
-	"floatunsidf.c",
-	"floatunsisf.c",
-	"floatuntidf.c",
-	"floatuntisf.c",
-	//"int_util.c",
-	"lshrdi3.c",
-	"lshrti3.c",
-	"moddi3.c",
-	"modsi3.c",
-	"modti3.c",
-	"muldc3.c",
-	"muldf3.c",
-	"muldi3.c",
-	"mulodi4.c",
-	"mulosi4.c",
-	"muloti4.c",
-	"mulsc3.c",
-	"mulsf3.c",
-	"multi3.c",
-	"multf3.c",
-	"mulvdi3.c",
-	"mulvsi3.c",
-	"mulvti3.c",
-	"negdf2.c",
-	"negdi2.c",
-	"negsf2.c",
-	"negti2.c",
-	"negvdi2.c",
-	"negvsi2.c",
-	"negvti2.c",
-	"os_version_check.c",
-	"paritydi2.c",
-	"paritysi2.c",
-	"parityti2.c",
-	"popcountdi2.c",
-	"popcountsi2.c",
-	"popcountti2.c",
-	"powidf2.c",
-	"powisf2.c",
-	"powitf2.c",
-	"subdf3.c",
-	"subsf3.c",
-	"subvdi3.c",
-	"subvsi3.c",
-	"subvti3.c",
-	"subtf3.c",
-	"trampoline_setup.c",
-	"truncdfhf2.c",
-	"truncdfsf2.c",
-	"truncsfhf2.c",
-	"ucmpdi2.c",
-	"ucmpti2.c",
-	"udivdi3.c",
-	"udivmoddi4.c",
-	"udivmodsi4.c",
-	"udivmodti4.c",
-	"udivsi3.c",
-	"udivti3.c",
-	"umoddi3.c",
-	"umodsi3.c",
-	"umodti3.c",
-}
-
-var aeabiBuiltins = []string{
-	"arm/aeabi_cdcmp.S",
-	"arm/aeabi_cdcmpeq_check_nan.c",
-	"arm/aeabi_cfcmp.S",
-	"arm/aeabi_cfcmpeq_check_nan.c",
-	"arm/aeabi_dcmp.S",
-	"arm/aeabi_div0.c",
-	"arm/aeabi_drsub.c",
-	"arm/aeabi_fcmp.S",
-	"arm/aeabi_frsub.c",
-	"arm/aeabi_idivmod.S",
-	"arm/aeabi_ldivmod.S",
-	"arm/aeabi_memcmp.S",
-	"arm/aeabi_memcpy.S",
-	"arm/aeabi_memmove.S",
-	"arm/aeabi_memset.S",
-	"arm/aeabi_uidivmod.S",
-	"arm/aeabi_uldivmod.S",
-}
-
-func builtinFiles(target string) []string {
-	builtins := append([]string{}, genericBuiltins...) // copy genericBuiltins
-	if strings.HasPrefix(target, "arm") {
-		builtins = append(builtins, aeabiBuiltins...)
-	}
-	return builtins
-}
-
-// builtinsDir returns the directory where the sources for compiler-rt are kept.
-func builtinsDir() string {
-	return filepath.Join(sourceDir(), "lib", "compiler-rt", "lib", "builtins")
-}
-
-// Get the builtins archive, possibly generating it as needed.
-func loadBuiltins(target string) (path string, err error) {
-	// Try to load a precompiled compiler-rt library.
-	precompiledPath := filepath.Join(sourceDir(), "pkg", target, "compiler-rt.a")
-	if _, err := os.Stat(precompiledPath); err == nil {
-		// Found a precompiled compiler-rt for this OS/architecture. Return the
-		// path directly.
-		return precompiledPath, nil
-	}
-
-	outfile := "librt-" + target + ".a"
-	builtinsDir := builtinsDir()
-
-	builtins := builtinFiles(target)
-	srcs := make([]string, len(builtins))
-	for i, name := range builtins {
-		srcs[i] = filepath.Join(builtinsDir, name)
-	}
-
-	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"][0], srcs)
-		cachepath = path
-		return err
-	})
-	return cachepath, err
-}
-
-// compileBuiltins compiles builtins from compiler-rt into a static library.
-// When it succeeds, it will call the callback with the resulting path. The path
-// will be removed after callback returns. If callback returns an error, this is
-// passed through to the return value of this function.
-func compileBuiltins(target string, callback func(path string) error) error {
-	builtinsDir := builtinsDir()
-
-	builtins := builtinFiles(target)
-	srcs := make([]string, len(builtins))
-	for i, name := range builtins {
-		srcs[i] = filepath.Join(builtinsDir, name)
-	}
-
-	dirPrefix := "tinygo-builtins"
-	remapDir := filepath.Join(os.TempDir(), dirPrefix)
-	dir, err := ioutil.TempDir(os.TempDir(), dirPrefix)
-	if err != nil {
-		return err
-	}
-	defer os.RemoveAll(dir)
-
-	// Compile all builtins.
-	// TODO: use builtins optimized for a given target if available.
-	objs := make([]string, 0, len(builtins))
-	for _, name := range builtins {
-		objname := name
-		if strings.LastIndexByte(objname, '/') >= 0 {
-			objname = objname[strings.LastIndexByte(objname, '/'):]
-		}
-		objpath := filepath.Join(dir, objname+".o")
-		objs = append(objs, objpath)
-		srcpath := filepath.Join(builtinsDir, name)
-		// 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.
-		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)
-		if err != nil {
-			return &commandError{"failed to build", srcpath, err}
-		}
-	}
-
-	// Put all builtins in an archive to link as a static library.
-	// Note: this does not create a symbol index, but ld.lld doesn't seem to
-	// care.
-	arpath := filepath.Join(dir, "librt.a")
-	arfile, err := os.Create(arpath)
-	if err != nil {
-		return err
-	}
-	defer arfile.Close()
-	arwriter := ar.NewWriter(arfile)
-	err = arwriter.WriteGlobalHeader()
-	if err != nil {
-		return &os.PathError{"write ar header", arpath, err}
-	}
-	for _, objpath := range objs {
-		name := filepath.Base(objpath)
-		objfile, err := os.Open(objpath)
-		if err != nil {
-			return err
-		}
-		defer objfile.Close()
-		st, err := objfile.Stat()
-		if err != nil {
-			return err
-		}
-		arwriter.WriteHeader(&ar.Header{
-			Name:    name,
-			ModTime: time.Unix(0, 0),
-			Uid:     0,
-			Gid:     0,
-			Mode:    0644,
-			Size:    st.Size(),
-		})
-		n, err := io.Copy(arwriter, objfile)
-		if err != nil {
-			return err
-		}
-		if n != st.Size() {
-			return errors.New("file modified during ar creation: " + arpath)
-		}
-	}
-
-	// 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

@@ -12,12 +12,16 @@ package cgo
 // source file parsing.
 
 import (
+	"fmt"
 	"go/ast"
+	"go/scanner"
 	"go/token"
+	"path/filepath"
 	"sort"
 	"strconv"
 	"strings"
 
+	"github.com/google/shlex"
 	"golang.org/x/tools/go/ast/astutil"
 )
 
@@ -37,21 +41,25 @@ type cgoPackage struct {
 	elaboratedTypes map[string]*elaboratedTypeInfo
 	enums           map[string]enumInfo
 	anonStructNum   int
+	cflags          []string // CFlags from #cgo lines
+	ldflags         []string // LDFlags from #cgo lines
+	visitedFiles    map[string][]byte
 }
 
 // constantInfo stores some information about a CGo constant found by libclang
 // and declared in the Go AST.
 type constantInfo struct {
-	expr *ast.BasicLit
+	expr ast.Expr
 	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
+	args     []paramInfo
+	results  *ast.FieldList
+	pos      token.Pos
+	variadic bool
 }
 
 // paramInfo is a parameter of a CGo function (see functionInfo).
@@ -69,9 +77,11 @@ type typedefInfo struct {
 // elaboratedTypeInfo contains some information about an elaborated type
 // (struct, union) found in the C AST.
 type elaboratedTypeInfo struct {
-	typeExpr  *ast.StructType
-	pos       token.Pos
-	bitfields []bitfieldInfo
+	typeExpr   *ast.StructType
+	pos        token.Pos
+	bitfields  []bitfieldInfo
+	unionSize  int64 // union size in bytes, nonzero when union getters/setters should be created
+	unionAlign int64 // union alignment in bytes
 }
 
 // bitfieldInfo contains information about a single bitfield in a struct. It
@@ -131,6 +141,7 @@ var builtinAliases = map[string]struct{}{
 // somehow from C. This is done by adding some typedefs to get the size of each
 // type.
 const cgoTypes = `
+# 1 "<cgo>"
 typedef char                _Cgo_char;
 typedef signed char         _Cgo_schar;
 typedef unsigned char       _Cgo_uchar;
@@ -147,9 +158,10 @@ 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) {
+// files, the CFLAGS and LDFLAGS found in #cgo lines, and a map of file hashes
+// of the accessed C header 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, []string, []string, map[string][]byte, []error) {
 	p := &cgoPackage{
 		dir:             dir,
 		fset:            fset,
@@ -161,6 +173,7 @@ func Process(files []*ast.File, dir string, fset *token.FileSet, cflags []string
 		typedefs:        map[string]*typedefInfo{},
 		elaboratedTypes: map[string]*elaboratedTypeInfo{},
 		enums:           map[string]enumInfo{},
+		visitedFiles:    map[string][]byte{},
 	}
 
 	// Add a new location for the following file.
@@ -168,6 +181,18 @@ func Process(files []*ast.File, dir string, fset *token.FileSet, cflags []string
 	generatedTokenPos.SetLines([]int{0})
 	p.generatedPos = generatedTokenPos.Pos(0)
 
+	// Find the absolute path for this package.
+	packagePath, err := filepath.Abs(fset.File(files[0].Pos()).Name())
+	if err != nil {
+		return nil, nil, nil, nil, []error{
+			scanner.Error{
+				Pos: fset.Position(files[0].Pos()),
+				Msg: "cgo: cannot find absolute path: " + err.Error(), // TODO: wrap this error
+			},
+		}
+	}
+	packagePath = filepath.Dir(packagePath)
+
 	// Construct a new in-memory AST for CGo declarations of this package.
 	unsafeImport := &ast.ImportSpec{
 		Path: &ast.BasicLit{
@@ -184,6 +209,7 @@ func Process(files []*ast.File, dir string, fset *token.FileSet, cflags []string
 			Name:    files[0].Name.Name,
 		},
 		Decls: []ast.Decl{
+			// import "unsafe"
 			&ast.GenDecl{
 				TokPos: p.generatedPos,
 				Tok:    token.IMPORT,
@@ -191,6 +217,32 @@ func Process(files []*ast.File, dir string, fset *token.FileSet, cflags []string
 					unsafeImport,
 				},
 			},
+			// var _ unsafe.Pointer
+			// This avoids type errors when the unsafe package is never used.
+			&ast.GenDecl{
+				Tok: token.VAR,
+				Specs: []ast.Spec{
+					&ast.ValueSpec{
+						Names: []*ast.Ident{
+							&ast.Ident{
+								Name: "_",
+								Obj: &ast.Object{
+									Kind: ast.Var,
+									Name: "_",
+								},
+							},
+						},
+						Type: &ast.SelectorExpr{
+							X: &ast.Ident{
+								Name: "unsafe",
+							},
+							Sel: &ast.Ident{
+								Name: "Pointer",
+							},
+						},
+					},
+				},
+			},
 		},
 		Imports: []*ast.ImportSpec{unsafeImport},
 	}
@@ -204,6 +256,7 @@ func Process(files []*ast.File, dir string, fset *token.FileSet, cflags []string
 	}
 
 	// Find `import "C"` statements in the file.
+	var statements []*ast.GenDecl
 	for _, f := range files {
 		for i := 0; i < len(f.Decls); i++ {
 			decl := f.Decls[i]
@@ -220,19 +273,16 @@ func Process(files []*ast.File, dir string, fset *token.FileSet, cflags []string
 			}
 			path, err := strconv.Unquote(spec.Path.Value)
 			if err != nil {
+				// This should not happen. An import path that is not properly
+				// quoted should not exist in a correct AST.
 				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)
+			// Found a CGo statement.
+			statements = append(statements, genDecl)
 
 			// Remove this import declaration.
 			f.Decls = append(f.Decls[:i], f.Decls[i+1:]...)
@@ -243,6 +293,114 @@ func Process(files []*ast.File, dir string, fset *token.FileSet, cflags []string
 		//ast.Print(fset, f)
 	}
 
+	// Find all #cgo lines.
+	for _, genDecl := range statements {
+		if genDecl.Doc == nil {
+			continue
+		}
+		for _, comment := range genDecl.Doc.List {
+			for {
+				// Extract the #cgo line, and replace it with spaces.
+				// Replacing with spaces makes sure that error locations are
+				// still correct, while not interfering with parsing in any way.
+				lineStart := strings.Index(comment.Text, "#cgo ")
+				if lineStart < 0 {
+					break
+				}
+				lineLen := strings.IndexByte(comment.Text[lineStart:], '\n')
+				if lineLen < 0 {
+					lineLen = len(comment.Text) - lineStart
+				}
+				lineEnd := lineStart + lineLen
+				line := comment.Text[lineStart:lineEnd]
+				spaces := make([]byte, len(line))
+				for i := range spaces {
+					spaces[i] = ' '
+				}
+				lenBefore := len(comment.Text)
+				comment.Text = comment.Text[:lineStart] + string(spaces) + comment.Text[lineEnd:]
+				if len(comment.Text) != lenBefore {
+					println(lenBefore, len(comment.Text))
+					panic("length of preamble changed!")
+				}
+
+				// Get the text before the colon in the #cgo directive.
+				colon := strings.IndexByte(line, ':')
+				if colon < 0 {
+					p.addErrorAfter(comment.Slash, comment.Text[:lineStart], "missing colon in #cgo line")
+					continue
+				}
+
+				// Extract the fields before the colon. These fields are a list
+				// of build tags and the C environment variable.
+				fields := strings.Fields(line[4:colon])
+				if len(fields) == 0 {
+					p.addErrorAfter(comment.Slash, comment.Text[:lineStart+colon-1], "invalid #cgo line")
+					continue
+				}
+
+				if len(fields) > 1 {
+					p.addErrorAfter(comment.Slash, comment.Text[:lineStart+5], "not implemented: build constraints in #cgo line")
+					continue
+				}
+
+				name := fields[len(fields)-1]
+				value := line[colon+1:]
+				switch name {
+				case "CFLAGS":
+					flags, err := shlex.Split(value)
+					if err != nil {
+						// TODO: find the exact location where the error happened.
+						p.addErrorAfter(comment.Slash, comment.Text[:lineStart+colon+1], "failed to parse flags in #cgo line: "+err.Error())
+						continue
+					}
+					if err := checkCompilerFlags(name, flags); err != nil {
+						p.addErrorAfter(comment.Slash, comment.Text[:lineStart+colon+1], err.Error())
+						continue
+					}
+					makePathsAbsolute(flags, packagePath)
+					p.cflags = append(p.cflags, flags...)
+				case "LDFLAGS":
+					flags, err := shlex.Split(value)
+					if err != nil {
+						// TODO: find the exact location where the error happened.
+						p.addErrorAfter(comment.Slash, comment.Text[:lineStart+colon+1], "failed to parse flags in #cgo line: "+err.Error())
+						continue
+					}
+					if err := checkLinkerFlags(name, flags); err != nil {
+						p.addErrorAfter(comment.Slash, comment.Text[:lineStart+colon+1], err.Error())
+						continue
+					}
+					makePathsAbsolute(flags, packagePath)
+					p.ldflags = append(p.ldflags, flags...)
+				default:
+					startPos := strings.LastIndex(line[4:colon], name) + 4
+					p.addErrorAfter(comment.Slash, comment.Text[:lineStart+startPos], "invalid #cgo line: "+name)
+					continue
+				}
+			}
+		}
+	}
+
+	// Define CFlags that will be used while parsing the package.
+	// Disable _FORTIFY_SOURCE as it causes problems on macOS.
+	// Note that it is only disabled for memcpy (etc) calls made from Go, which
+	// have better alternatives anyway.
+	cflagsForCGo := append([]string{"-D_FORTIFY_SOURCE=0"}, cflags...)
+	cflagsForCGo = append(cflagsForCGo, p.cflags...)
+
+	// Process all CGo imports.
+	for _, genDecl := range statements {
+		cgoComment := genDecl.Doc.Text()
+
+		pos := genDecl.Pos()
+		if genDecl.Doc != nil {
+			pos = genDecl.Doc.Pos()
+		}
+		position := fset.PositionFor(pos, true)
+		p.parseFragment(cgoComment+cgoTypes, cflagsForCGo, position.Filename, position.Line)
+	}
+
 	// Declare functions found by libclang.
 	p.addFuncDecls()
 
@@ -275,7 +433,31 @@ func Process(files []*ast.File, dir string, fset *token.FileSet, cflags []string
 	// Print the newly generated in-memory AST, for debugging.
 	//ast.Print(fset, p.generated)
 
-	return p.generated, p.errors
+	return p.generated, p.cflags, p.ldflags, p.visitedFiles, p.errors
+}
+
+// makePathsAbsolute converts some common path compiler flags (-I, -L) from
+// relative flags into absolute flags, if they are relative. This is necessary
+// because the C compiler is usually not invoked from the package path.
+func makePathsAbsolute(args []string, packagePath string) {
+	nextIsPath := false
+	for i, arg := range args {
+		if nextIsPath {
+			if !filepath.IsAbs(arg) {
+				args[i] = filepath.Join(packagePath, arg)
+			}
+		}
+		if arg == "-I" || arg == "-L" {
+			nextIsPath = true
+			continue
+		}
+		if strings.HasPrefix(arg, "-I") || strings.HasPrefix(arg, "-L") {
+			path := arg[2:]
+			if !filepath.IsAbs(path) {
+				args[i] = arg[:2] + filepath.Join(packagePath, path)
+			}
+		}
+	}
 }
 
 // addFuncDecls adds the C function declarations found by libclang in the
@@ -309,6 +491,16 @@ func (p *cgoPackage) addFuncDecls() {
 				Results: fn.results,
 			},
 		}
+		if fn.variadic {
+			decl.Doc = &ast.CommentGroup{
+				List: []*ast.Comment{
+					&ast.Comment{
+						Slash: fn.pos,
+						Text:  "//go:variadic",
+					},
+				},
+			}
+		}
 		obj.Decl = decl
 		for i, arg := range fn.args {
 			args[i] = &ast.Field{
@@ -343,18 +535,18 @@ 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 {
+		gen := &ast.GenDecl{
+			TokPos: token.NoPos,
+			Tok:    token.VAR,
+			Lparen: token.NoPos,
+			Rparen: token.NoPos,
+		}
 		fn := p.functions[name]
 		obj := &ast.Object{
 			Kind: ast.Typ,
@@ -379,34 +571,32 @@ func (p *cgoPackage) addFuncPtrDecls() {
 		}
 		obj.Decl = valueSpec
 		gen.Specs = append(gen.Specs, valueSpec)
+		p.generated.Decls = append(p.generated.Decls, gen)
 	}
-	p.generated.Decls = append(p.generated.Decls, gen)
 }
 
 // addConstDecls declares external C constants in the Go source.
 // It adds code like the following to the AST:
 //
-//     const (
-//         C.CONST_INT = 5
-//         C.CONST_FLOAT = 5.8
-//         // ...
-//     )
+//     const C.CONST_INT = 5
+//     const 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 {
+		gen := &ast.GenDecl{
+			TokPos: token.NoPos,
+			Tok:    token.CONST,
+			Lparen: token.NoPos,
+			Rparen: token.NoPos,
+		}
 		constVal := p.constants[name]
 		obj := &ast.Object{
 			Kind: ast.Con,
@@ -422,34 +612,32 @@ func (p *cgoPackage) addConstDecls() {
 		}
 		obj.Decl = valueSpec
 		gen.Specs = append(gen.Specs, valueSpec)
+		p.generated.Decls = append(p.generated.Decls, gen)
 	}
-	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
-//         // ...
-//     )
+//     var C.globalInt  int
+//     var 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 {
+		gen := &ast.GenDecl{
+			TokPos: token.NoPos,
+			Tok:    token.VAR,
+			Lparen: token.NoPos,
+			Rparen: token.NoPos,
+		}
 		global := p.globals[name]
 		obj := &ast.Object{
 			Kind: ast.Var,
@@ -465,31 +653,29 @@ func (p *cgoPackage) addVarDecls() {
 		}
 		obj.Decl = valueSpec
 		gen.Specs = append(gen.Specs, valueSpec)
+		p.generated.Decls = append(p.generated.Decls, gen)
 	}
-	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
-//         // ...
-//     )
+//     type C.int8_t  = int8
+//     type 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 {
+		gen := &ast.GenDecl{
+			TokPos: token.NoPos,
+			Tok:    token.TYPE,
+			Lparen: token.NoPos,
+			Rparen: token.NoPos,
+		}
 		goTypeName := cgoAliases[typeName]
 		obj := &ast.Object{
 			Kind: ast.Typ,
@@ -509,24 +695,24 @@ func (p *cgoPackage) addTypeAliases() {
 		}
 		obj.Decl = typeSpec
 		gen.Specs = append(gen.Specs, typeSpec)
+		p.generated.Decls = append(p.generated.Decls, gen)
 	}
-	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 {
+		gen := &ast.GenDecl{
+			TokPos: token.NoPos,
+			Tok:    token.TYPE,
+		}
 		typedef := p.typedefs[name]
 		typeName := "C." + name
 		isAlias := true
@@ -555,8 +741,8 @@ func (p *cgoPackage) addTypedefs() {
 		}
 		obj.Decl = typeSpec
 		gen.Specs = append(gen.Specs, typeSpec)
+		p.generated.Decls = append(p.generated.Decls, gen)
 	}
-	p.generated.Decls = append(p.generated.Decls, gen)
 }
 
 // addElaboratedTypes adds C elaborated types as aliases. These are the "struct
@@ -568,29 +754,47 @@ 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 {
+		gen := &ast.GenDecl{
+			TokPos: token.NoPos,
+			Tok:    token.TYPE,
+		}
 		typ := p.elaboratedTypes[name]
 		typeName := "C." + name
 		obj := &ast.Object{
 			Kind: ast.Typ,
 			Name: typeName,
 		}
+		typeExpr := typ.typeExpr
+		if typ.unionSize != 0 {
+			// Create getters/setters.
+			for _, field := range typ.typeExpr.Fields.List {
+				if len(field.Names) != 1 {
+					p.addError(typ.pos, fmt.Sprintf("union must have field with a single name, it has %d names", len(field.Names)))
+					continue
+				}
+				p.createUnionAccessor(field, typeName)
+			}
+			// Convert to a single-field struct type.
+			typeExpr = p.makeUnionField(typ)
+			if typeExpr == nil {
+				// There was an error, that was already added to the list of
+				// errors.
+				continue
+			}
+		}
 		typeSpec := &ast.TypeSpec{
 			Name: &ast.Ident{
 				NamePos: typ.pos,
 				Name:    typeName,
 				Obj:     obj,
 			},
-			Type: typ.typeExpr,
+			Type: typeExpr,
 		}
 		obj.Decl = typeSpec
 		gen.Specs = append(gen.Specs, typeSpec)
@@ -599,8 +803,185 @@ func (p *cgoPackage) addElaboratedTypes() {
 			p.createBitfieldGetter(bitfield, typeName)
 			p.createBitfieldSetter(bitfield, typeName)
 		}
+		p.generated.Decls = append(p.generated.Decls, gen)
 	}
-	p.generated.Decls = append(p.generated.Decls, gen)
+}
+
+// makeUnionField creates a new struct from an existing *elaboratedTypeInfo,
+// that has just a single field that must be accessed through special accessors.
+// It returns nil when there is an error. In case of an error, that error has
+// already been added to the list of errors using p.addError.
+func (p *cgoPackage) makeUnionField(typ *elaboratedTypeInfo) *ast.StructType {
+	unionFieldTypeName, ok := map[int64]string{
+		1: "uint8",
+		2: "uint16",
+		4: "uint32",
+		8: "uint64",
+	}[typ.unionAlign]
+	if !ok {
+		p.addError(typ.typeExpr.Struct, fmt.Sprintf("expected union alignment to be one of 1, 2, 4, or 8, but got %d", typ.unionAlign))
+		return nil
+	}
+	var unionFieldType ast.Expr = &ast.Ident{
+		NamePos: token.NoPos,
+		Name:    unionFieldTypeName,
+	}
+	if typ.unionSize != typ.unionAlign {
+		// A plain struct{uintX} isn't enough, we have to make a
+		// struct{[N]uintX} to make the union big enough.
+		if typ.unionSize/typ.unionAlign*typ.unionAlign != typ.unionSize {
+			p.addError(typ.typeExpr.Struct, fmt.Sprintf("union alignment (%d) must be a multiple of union alignment (%d)", typ.unionSize, typ.unionAlign))
+			return nil
+		}
+		unionFieldType = &ast.ArrayType{
+			Len: &ast.BasicLit{
+				Kind:  token.INT,
+				Value: strconv.FormatInt(typ.unionSize/typ.unionAlign, 10),
+			},
+			Elt: unionFieldType,
+		}
+	}
+	return &ast.StructType{
+		Struct: typ.typeExpr.Struct,
+		Fields: &ast.FieldList{
+			Opening: typ.typeExpr.Fields.Opening,
+			List: []*ast.Field{&ast.Field{
+				Names: []*ast.Ident{
+					&ast.Ident{
+						NamePos: typ.typeExpr.Fields.Opening,
+						Name:    "$union",
+					},
+				},
+				Type: unionFieldType,
+			}},
+			Closing: typ.typeExpr.Fields.Closing,
+		},
+	}
+}
+
+// createUnionAccessor creates a function that returns a typed pointer to a
+// union field for each field in a union. For example:
+//
+//     func (union *C.union_1) unionfield_d() *float64 {
+//         return (*float64)(unsafe.Pointer(&union.$union))
+//     }
+//
+// Where C.union_1 is defined as:
+//
+//     type C.union_1 struct{
+//         $union uint64
+//     }
+//
+// The returned pointer can be used to get or set the field, or get the pointer
+// to a subfield.
+func (p *cgoPackage) createUnionAccessor(field *ast.Field, typeName string) {
+	if len(field.Names) != 1 {
+		panic("number of names in union field must be exactly 1")
+	}
+	fieldName := field.Names[0]
+	pos := fieldName.NamePos
+
+	// The method receiver.
+	receiver := &ast.SelectorExpr{
+		X: &ast.Ident{
+			NamePos: pos,
+			Name:    "union",
+			Obj:     nil,
+		},
+		Sel: &ast.Ident{
+			NamePos: pos,
+			Name:    "$union",
+		},
+	}
+
+	// Get the address of the $union field.
+	receiverPtr := &ast.UnaryExpr{
+		Op: token.AND,
+		X:  receiver,
+	}
+
+	// Cast to unsafe.Pointer.
+	sourcePointer := &ast.CallExpr{
+		Fun: &ast.SelectorExpr{
+			X:   &ast.Ident{Name: "unsafe"},
+			Sel: &ast.Ident{Name: "Pointer"},
+		},
+		Args: []ast.Expr{receiverPtr},
+	}
+
+	// Cast to the target pointer type.
+	targetPointer := &ast.CallExpr{
+		Lparen: pos,
+		Fun: &ast.ParenExpr{
+			Lparen: pos,
+			X: &ast.StarExpr{
+				X: field.Type,
+			},
+			Rparen: pos,
+		},
+		Args:   []ast.Expr{sourcePointer},
+		Rparen: pos,
+	}
+
+	// Create the accessor function.
+	accessor := &ast.FuncDecl{
+		Recv: &ast.FieldList{
+			Opening: pos,
+			List: []*ast.Field{
+				&ast.Field{
+					Names: []*ast.Ident{
+						&ast.Ident{
+							NamePos: pos,
+							Name:    "union",
+						},
+					},
+					Type: &ast.StarExpr{
+						Star: pos,
+						X: &ast.Ident{
+							NamePos: pos,
+							Name:    typeName,
+							Obj:     nil,
+						},
+					},
+				},
+			},
+			Closing: pos,
+		},
+		Name: &ast.Ident{
+			NamePos: pos,
+			Name:    "unionfield_" + fieldName.Name,
+		},
+		Type: &ast.FuncType{
+			Func: pos,
+			Params: &ast.FieldList{
+				Opening: pos,
+				Closing: pos,
+			},
+			Results: &ast.FieldList{
+				List: []*ast.Field{
+					&ast.Field{
+						Type: &ast.StarExpr{
+							Star: pos,
+							X:    field.Type,
+						},
+					},
+				},
+			},
+		},
+		Body: &ast.BlockStmt{
+			Lbrace: pos,
+			List: []ast.Stmt{
+				&ast.ReturnStmt{
+					Return: pos,
+					Results: []ast.Expr{
+						targetPointer,
+					},
+				},
+			},
+			Rbrace: pos,
+		},
+	}
+	p.generated.Decls = append(p.generated.Decls, accessor)
 }
 
 // createBitfieldGetter creates a bitfield getter function like the following:
@@ -905,16 +1286,16 @@ func (p *cgoPackage) addEnumTypes() {
 	if len(p.enums) == 0 {
 		return
 	}
-	gen := &ast.GenDecl{
-		TokPos: token.NoPos,
-		Tok:    token.TYPE,
-	}
 	names := make([]string, 0, len(p.enums))
 	for name := range p.enums {
 		names = append(names, name)
 	}
 	sort.Strings(names)
 	for _, name := range names {
+		gen := &ast.GenDecl{
+			TokPos: token.NoPos,
+			Tok:    token.TYPE,
+		}
 		typ := p.enums[name]
 		typeName := "C.enum_" + name
 		obj := &ast.Object{
@@ -931,8 +1312,8 @@ func (p *cgoPackage) addEnumTypes() {
 		}
 		obj.Decl = typeSpec
 		gen.Specs = append(gen.Specs, typeSpec)
+		p.generated.Decls = append(p.generated.Decls, gen)
 	}
-	p.generated.Decls = append(p.generated.Decls, gen)
 }
 
 // findMissingCGoNames traverses the AST and finds all C.something names. Only
@@ -995,3 +1376,29 @@ func (p *cgoPackage) walker(cursor *astutil.Cursor) bool {
 	}
 	return true
 }
+
+// renameFieldKeywords renames all reserved words in Go to some other field name
+// with a "_" prefix. For example, it renames `type` to `_type`.
+//
+// See: https://golang.org/cmd/cgo/#hdr-Go_references_to_C
+func renameFieldKeywords(fieldList *ast.FieldList) {
+	renameFieldName(fieldList, "type")
+}
+
+// renameFieldName renames a given field name to a name with a "_" prepended. It
+// makes sure to do the same thing for any field sharing the same name.
+func renameFieldName(fieldList *ast.FieldList, name string) {
+	var ident *ast.Ident
+	for _, f := range fieldList.List {
+		for _, n := range f.Names {
+			if n.Name == name {
+				ident = n
+			}
+		}
+	}
+	if ident == nil {
+		return
+	}
+	renameFieldName(fieldList, "_"+name)
+	ident.Name = "_" + ident.Name
+}

cgo/cgo_test.go

@@ -0,0 +1,160 @@
+package cgo
+
+import (
+	"bytes"
+	"flag"
+	"fmt"
+	"go/ast"
+	"go/build"
+	"go/format"
+	"go/parser"
+	"go/token"
+	"go/types"
+	"io/ioutil"
+	"path/filepath"
+	"regexp"
+	"runtime"
+	"strings"
+	"testing"
+)
+
+// Pass -update to go test to update the output of the test files.
+var flagUpdate = flag.Bool("update", false, "Update images based on test output.")
+
+// normalizeResult normalizes Go source code that comes out of tests across
+// platforms and Go versions.
+func normalizeResult(result string) string {
+	actual := strings.ReplaceAll(result, "\r\n", "\n")
+
+	// Make sure all functions are wrapped, even those that would otherwise be
+	// single-line functions. This is necessary because Go 1.14 changed the way
+	// such functions are wrapped and it's important to have consistent test
+	// results.
+	re := regexp.MustCompile(`func \((.+)\)( .*?) +{ (.+) }`)
+	actual = re.ReplaceAllString(actual, "func ($1)$2 {\n\t$3\n}")
+
+	return actual
+}
+
+func TestCGo(t *testing.T) {
+	var cflags = []string{"--target=armv6m-none-eabi"}
+
+	for _, name := range []string{"basic", "errors", "types", "flags", "const"} {
+		name := name // avoid a race condition
+		t.Run(name, func(t *testing.T) {
+			// Skip tests that require specific Go version.
+			if name == "errors" {
+				ok := false
+				for _, version := range build.Default.ReleaseTags {
+					if version == "go1.16" {
+						ok = true
+						break
+					}
+				}
+				if !ok {
+					t.Skip("Results for errors test are only valid for Go 1.16+")
+				}
+			}
+
+			// Read the AST in memory.
+			path := filepath.Join("testdata", name+".go")
+			fset := token.NewFileSet()
+			f, err := parser.ParseFile(fset, path, nil, parser.ParseComments)
+			if err != nil {
+				t.Fatal("could not parse Go source file:", err)
+			}
+
+			// Process the AST with CGo.
+			cgoAST, _, _, _, cgoErrors := Process([]*ast.File{f}, "testdata", fset, cflags)
+
+			// Check the AST for type errors.
+			var typecheckErrors []error
+			config := types.Config{
+				Error: func(err error) {
+					typecheckErrors = append(typecheckErrors, err)
+				},
+				Importer: simpleImporter{},
+				Sizes:    types.SizesFor("gccgo", "arm"),
+			}
+			_, err = config.Check("", fset, []*ast.File{f, cgoAST}, nil)
+			if err != nil && len(typecheckErrors) == 0 {
+				// Only report errors when no type errors are found (an
+				// unexpected condition).
+				t.Error(err)
+			}
+
+			// Store the (formatted) output in a buffer. Format it, so it
+			// becomes easier to read (and will hopefully change less with CGo
+			// changes).
+			buf := &bytes.Buffer{}
+			if len(cgoErrors) != 0 {
+				buf.WriteString("// CGo errors:\n")
+				for _, err := range cgoErrors {
+					buf.WriteString(formatDiagnostic(err))
+				}
+				buf.WriteString("\n")
+			}
+			if len(typecheckErrors) != 0 {
+				buf.WriteString("// Type checking errors after CGo processing:\n")
+				for _, err := range typecheckErrors {
+					buf.WriteString(formatDiagnostic(err))
+				}
+				buf.WriteString("\n")
+			}
+			err = format.Node(buf, fset, cgoAST)
+			if err != nil {
+				t.Errorf("could not write out CGo AST: %v", err)
+			}
+			actual := normalizeResult(string(buf.Bytes()))
+
+			// Read the file with the expected output, to compare against.
+			outfile := filepath.Join("testdata", name+".out.go")
+			expectedBytes, err := ioutil.ReadFile(outfile)
+			if err != nil {
+				t.Fatalf("could not read expected output: %v", err)
+			}
+			expected := strings.ReplaceAll(string(expectedBytes), "\r\n", "\n")
+
+			// Check whether the output is as expected.
+			if expected != actual {
+				// It is not. Test failed.
+				if *flagUpdate {
+					// Update the file with the expected data.
+					err := ioutil.WriteFile(outfile, []byte(actual), 0666)
+					if err != nil {
+						t.Error("could not write updated output file:", err)
+					}
+					return
+				}
+				t.Errorf("output did not match:\n%s", string(actual))
+			}
+		})
+	}
+}
+
+// simpleImporter implements the types.Importer interface, but only allows
+// importing the unsafe package.
+type simpleImporter struct {
+}
+
+// Import implements the Importer interface. For testing usage only: it only
+// supports importing the unsafe package.
+func (i simpleImporter) Import(path string) (*types.Package, error) {
+	switch path {
+	case "unsafe":
+		return types.Unsafe, nil
+	default:
+		return nil, fmt.Errorf("importer not implemented for package %s", path)
+	}
+}
+
+// formatDiagnostics formats the error message to be an indented comment. It
+// also fixes Windows path name issues (backward slashes).
+func formatDiagnostic(err error) string {
+	msg := err.Error()
+	if runtime.GOOS == "windows" {
+		// Fix Windows path slashes.
+		msg = strings.ReplaceAll(msg, "testdata\\", "testdata/")
+	}
+	return "//     " + msg + "\n"
+}

cgo/const.go

@@ -0,0 +1,308 @@
+package cgo
+
+// This file implements a parser of a subset of the C language, just enough to
+// parse common #define statements to Go constant expressions.
+
+import (
+	"fmt"
+	"go/ast"
+	"go/scanner"
+	"go/token"
+	"strings"
+)
+
+var (
+	prefixParseFns map[token.Token]func(*tokenizer) (ast.Expr, *scanner.Error)
+	precedences    = map[token.Token]int{
+		token.ADD: precedenceAdd,
+		token.SUB: precedenceAdd,
+		token.MUL: precedenceMul,
+		token.QUO: precedenceMul,
+		token.REM: precedenceMul,
+	}
+)
+
+const (
+	precedenceLowest = iota + 1
+	precedenceAdd
+	precedenceMul
+	precedencePrefix
+)
+
+func init() {
+	// This must be done in an init function to avoid an initialization order
+	// failure.
+	prefixParseFns = map[token.Token]func(*tokenizer) (ast.Expr, *scanner.Error){
+		token.IDENT:  parseIdent,
+		token.INT:    parseBasicLit,
+		token.FLOAT:  parseBasicLit,
+		token.STRING: parseBasicLit,
+		token.CHAR:   parseBasicLit,
+		token.LPAREN: parseParenExpr,
+		token.SUB:    parseUnaryExpr,
+	}
+}
+
+// parseConst parses the given string as a C constant.
+func parseConst(pos token.Pos, fset *token.FileSet, value string) (ast.Expr, *scanner.Error) {
+	t := newTokenizer(pos, fset, value)
+	expr, err := parseConstExpr(t, precedenceLowest)
+	t.Next()
+	if t.curToken != token.EOF {
+		return nil, &scanner.Error{
+			Pos: t.fset.Position(t.curPos),
+			Msg: "unexpected token " + t.curToken.String() + ", expected end of expression",
+		}
+	}
+	return expr, err
+}
+
+// parseConstExpr parses a stream of C tokens to a Go expression.
+func parseConstExpr(t *tokenizer, precedence int) (ast.Expr, *scanner.Error) {
+	if t.curToken == token.EOF {
+		return nil, &scanner.Error{
+			Pos: t.fset.Position(t.curPos),
+			Msg: "empty constant",
+		}
+	}
+	prefix := prefixParseFns[t.curToken]
+	if prefix == nil {
+		return nil, &scanner.Error{
+			Pos: t.fset.Position(t.curPos),
+			Msg: fmt.Sprintf("unexpected token %s", t.curToken),
+		}
+	}
+	leftExpr, err := prefix(t)
+
+	for t.peekToken != token.EOF && precedence < precedences[t.peekToken] {
+		switch t.peekToken {
+		case token.ADD, token.SUB, token.MUL, token.QUO, token.REM:
+			t.Next()
+			leftExpr, err = parseBinaryExpr(t, leftExpr)
+		}
+	}
+
+	return leftExpr, err
+}
+
+func parseIdent(t *tokenizer) (ast.Expr, *scanner.Error) {
+	return &ast.Ident{
+		NamePos: t.curPos,
+		Name:    "C." + t.curValue,
+	}, nil
+}
+
+func parseBasicLit(t *tokenizer) (ast.Expr, *scanner.Error) {
+	return &ast.BasicLit{
+		ValuePos: t.curPos,
+		Kind:     t.curToken,
+		Value:    t.curValue,
+	}, nil
+}
+
+func parseParenExpr(t *tokenizer) (ast.Expr, *scanner.Error) {
+	lparen := t.curPos
+	t.Next()
+	x, err := parseConstExpr(t, precedenceLowest)
+	if err != nil {
+		return nil, err
+	}
+	t.Next()
+	if t.curToken != token.RPAREN {
+		return nil, unexpectedToken(t, token.RPAREN)
+	}
+	expr := &ast.ParenExpr{
+		Lparen: lparen,
+		X:      x,
+		Rparen: t.curPos,
+	}
+	return expr, nil
+}
+
+func parseBinaryExpr(t *tokenizer, left ast.Expr) (ast.Expr, *scanner.Error) {
+	expression := &ast.BinaryExpr{
+		X:     left,
+		Op:    t.curToken,
+		OpPos: t.curPos,
+	}
+	precedence := precedences[t.curToken]
+	t.Next()
+	right, err := parseConstExpr(t, precedence)
+	expression.Y = right
+	return expression, err
+}
+
+func parseUnaryExpr(t *tokenizer) (ast.Expr, *scanner.Error) {
+	expression := &ast.UnaryExpr{
+		OpPos: t.curPos,
+		Op:    t.curToken,
+	}
+	t.Next()
+	x, err := parseConstExpr(t, precedencePrefix)
+	expression.X = x
+	return expression, err
+}
+
+// unexpectedToken returns an error of the form "unexpected token FOO, expected
+// BAR".
+func unexpectedToken(t *tokenizer, expected token.Token) *scanner.Error {
+	return &scanner.Error{
+		Pos: t.fset.Position(t.curPos),
+		Msg: fmt.Sprintf("unexpected token %s, expected %s", t.curToken, expected),
+	}
+}
+
+// tokenizer reads C source code and converts it to Go tokens.
+type tokenizer struct {
+	curPos, peekPos     token.Pos
+	fset                *token.FileSet
+	curToken, peekToken token.Token
+	curValue, peekValue string
+	buf                 string
+}
+
+// newTokenizer initializes a new tokenizer, positioned at the first token in
+// the string.
+func newTokenizer(start token.Pos, fset *token.FileSet, buf string) *tokenizer {
+	t := &tokenizer{
+		peekPos:   start,
+		fset:      fset,
+		buf:       buf,
+		peekToken: token.ILLEGAL,
+	}
+	// Parse the first two tokens (cur and peek).
+	t.Next()
+	t.Next()
+	return t
+}
+
+// Next consumes the next token in the stream. There is no return value, read
+// the next token from the pos, token and value properties.
+func (t *tokenizer) Next() {
+	// The previous peek is now the current token.
+	t.curPos = t.peekPos
+	t.curToken = t.peekToken
+	t.curValue = t.peekValue
+
+	// Parse the next peek token.
+	t.peekPos += token.Pos(len(t.curValue))
+	for {
+		if len(t.buf) == 0 {
+			t.peekToken = token.EOF
+			return
+		}
+		c := t.buf[0]
+		switch {
+		case c == ' ' || c == '\f' || c == '\n' || c == '\r' || c == '\t' || c == '\v':
+			// Skip whitespace.
+			// Based on this source, not sure whether it represents C whitespace:
+			// https://en.cppreference.com/w/cpp/string/byte/isspace
+			t.peekPos++
+			t.buf = t.buf[1:]
+		case c == '(' || c == ')' || c == '+' || c == '-' || c == '*' || c == '/' || c == '%':
+			// Single-character tokens.
+			// TODO: ++ (increment) and -- (decrement) operators.
+			switch c {
+			case '(':
+				t.peekToken = token.LPAREN
+			case ')':
+				t.peekToken = token.RPAREN
+			case '+':
+				t.peekToken = token.ADD
+			case '-':
+				t.peekToken = token.SUB
+			case '*':
+				t.peekToken = token.MUL
+			case '/':
+				t.peekToken = token.QUO
+			case '%':
+				t.peekToken = token.REM
+			}
+			t.peekValue = t.buf[:1]
+			t.buf = t.buf[1:]
+			return
+		case c >= '0' && c <= '9':
+			// Numeric constant (int, float, etc.).
+			// Find the last non-numeric character.
+			tokenLen := len(t.buf)
+			hasDot := false
+			for i, c := range t.buf {
+				if c == '.' {
+					hasDot = true
+				}
+				if c >= '0' && c <= '9' || c == '.' || c == '_' || c >= 'a' && c <= 'z' || c >= 'A' && c <= 'Z' {
+					tokenLen = i + 1
+				} else {
+					break
+				}
+			}
+			t.peekValue = t.buf[:tokenLen]
+			t.buf = t.buf[tokenLen:]
+			if hasDot {
+				// Integer constants are more complicated than this but this is
+				// a close approximation.
+				// https://en.cppreference.com/w/cpp/language/integer_literal
+				t.peekToken = token.FLOAT
+				t.peekValue = strings.TrimRight(t.peekValue, "f")
+			} else {
+				t.peekToken = token.INT
+				t.peekValue = strings.TrimRight(t.peekValue, "uUlL")
+			}
+			return
+		case c >= 'A' && c <= 'Z' || c >= 'a' && c <= 'z' || c == '_':
+			// Identifier. Find all remaining tokens that are part of this
+			// identifier.
+			tokenLen := len(t.buf)
+			for i, c := range t.buf {
+				if c >= '0' && c <= '9' || c >= 'A' && c <= 'Z' || c >= 'a' && c <= 'z' || c == '_' {
+					tokenLen = i + 1
+				} else {
+					break
+				}
+			}
+			t.peekValue = t.buf[:tokenLen]
+			t.buf = t.buf[tokenLen:]
+			t.peekToken = token.IDENT
+			return
+		case c == '"':
+			// String constant. Find the first '"' character that is not
+			// preceded by a backslash.
+			escape := false
+			tokenLen := len(t.buf)
+			for i, c := range t.buf {
+				if i != 0 && c == '"' && !escape {
+					tokenLen = i + 1
+					break
+				}
+				if !escape {
+					escape = c == '\\'
+				}
+			}
+			t.peekToken = token.STRING
+			t.peekValue = t.buf[:tokenLen]
+			t.buf = t.buf[tokenLen:]
+			return
+		case c == '\'':
+			// Char (rune) constant. Find the first '\'' character that is not
+			// preceded by a backslash.
+			escape := false
+			tokenLen := len(t.buf)
+			for i, c := range t.buf {
+				if i != 0 && c == '\'' && !escape {
+					tokenLen = i + 1
+					break
+				}
+				if !escape {
+					escape = c == '\\'
+				}
+			}
+			t.peekToken = token.CHAR
+			t.peekValue = t.buf[:tokenLen]
+			t.buf = t.buf[tokenLen:]
+			return
+		default:
+			t.peekToken = token.ILLEGAL
+			return
+		}
+	}
+}

cgo/const_test.go

@@ -0,0 +1,77 @@
+package cgo
+
+import (
+	"bytes"
+	"go/format"
+	"go/token"
+	"strings"
+	"testing"
+)
+
+func TestParseConst(t *testing.T) {
+	// Test converting a C constant to a Go constant.
+	for _, tc := range []struct {
+		C  string
+		Go string
+	}{
+		{`5`, `5`},
+		{`(5)`, `(5)`},
+		{`(((5)))`, `(5)`},
+		{`)`, `error: 1:1: unexpected token )`},
+		{`5)`, `error: 1:2: unexpected token ), expected end of expression`},
+		{"  \t)", `error: 1:4: unexpected token )`},
+		{`5.8f`, `5.8`},
+		{`foo`, `C.foo`},
+		{``, `error: 1:1: empty constant`}, // empty constants not allowed in Go
+		{`"foo"`, `"foo"`},
+		{`"a\\n"`, `"a\\n"`},
+		{`"a\n"`, `"a\n"`},
+		{`"a\""`, `"a\""`},
+		{`'a'`, `'a'`},
+		{`0b10`, `0b10`},
+		{`0x1234_5678`, `0x1234_5678`},
+		{`5 5`, `error: 1:3: unexpected token INT, expected end of expression`}, // test for a bugfix
+		// Binary operators.
+		{`5+5`, `5 + 5`},
+		{`5-5`, `5 - 5`},
+		{`5*5`, `5 * 5`},
+		{`5/5`, `5 / 5`},
+		{`5%5`, `5 % 5`},
+		{`(5/5)`, `(5 / 5)`},
+		{`1 - 2`, `1 - 2`},
+		{`1 - 2 + 3`, `1 - 2 + 3`},
+		{`1 - 2 * 3`, `1 - 2*3`},
+		{`(1 - 2) * 3`, `(1 - 2) * 3`},
+		{`1 * 2 - 3`, `1*2 - 3`},
+		{`1 * (2 - 3)`, `1 * (2 - 3)`},
+		// Unary operators.
+		{`-5`, `-5`},
+		{`-5-2`, `-5 - 2`},
+		{`5 - - 2`, `5 - -2`},
+	} {
+		fset := token.NewFileSet()
+		startPos := fset.AddFile("", -1, 1000).Pos(0)
+		expr, err := parseConst(startPos, fset, tc.C)
+		s := "<invalid>"
+		if err != nil {
+			if !strings.HasPrefix(tc.Go, "error: ") {
+				t.Errorf("expected value %#v for C constant %#v but got error %#v", tc.Go, tc.C, err.Error())
+				continue
+			}
+			s = "error: " + err.Error()
+		} else if expr != nil {
+			// Serialize the Go constant to a string, for more readable test
+			// cases.
+			buf := &bytes.Buffer{}
+			err := format.Node(buf, fset, expr)
+			if err != nil {
+				t.Errorf("could not format expr from C constant %#v: %v", tc.C, err)
+				continue
+			}
+			s = buf.String()
+		}
+		if s != tc.Go {
+			t.Errorf("C constant %#v was parsed to %#v while expecting %#v", tc.C, s, tc.Go)
+		}
+	}
+}

cgo/libclang.go

@@ -4,6 +4,7 @@ package cgo
 // modification. It does not touch the AST itself.
 
 import (
+	"crypto/sha512"
 	"fmt"
 	"go/ast"
 	"go/scanner"
@@ -15,7 +16,7 @@ import (
 )
 
 /*
-#include <clang-c/Index.h> // if this fails, install libclang-8-dev
+#include <clang-c/Index.h> // if this fails, install libclang-10-dev
 #include <stdlib.h>
 #include <stdint.h>
 
@@ -56,6 +57,7 @@ unsigned tinygo_clang_Cursor_isBitField(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);
 int tinygo_clang_enum_visitor(GoCXCursor c, GoCXCursor parent, CXClientData client_data);
+void tinygo_clang_inclusion_visitor(CXFile included_file, CXSourceLocation *inclusion_stack, unsigned include_len, CXClientData client_data);
 */
 import "C"
 
@@ -109,36 +111,19 @@ func (p *cgoPackage) parseFragment(fragment string, cflags []string, posFilename
 		C.CXTranslationUnit_DetailedPreprocessingRecord,
 		&unit)
 	if errCode != 0 {
-		panic("loader: failed to parse source with libclang")
+		// This is probably a bug in the usage of libclang.
+		panic("cgo: failed to parse source with libclang")
 	}
 	defer C.clang_disposeTranslationUnit(unit)
 
+	// Report parser and type errors.
 	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,
-			})
+			pos := p.getClangLocationPosition(location, unit)
+			p.addError(pos, severity+": "+spelling)
 		}
 		for i := 0; i < numDiagnostics; i++ {
 			diagnostic := C.clang_getDiagnostic(unit, C.uint(i))
@@ -150,13 +135,38 @@ func (p *cgoPackage) parseFragment(fragment string, cflags []string, posFilename
 				addDiagnostic(C.clang_getDiagnosticInSet(diagnostics, C.uint(j)))
 			}
 		}
-		return
 	}
 
+	// Extract information required by CGo.
 	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))
+
+	// Determine files read during CGo processing, for caching.
+	inclusionCallback := func(includedFile C.CXFile) {
+		// Get full file path.
+		path := getString(C.clang_getFileName(includedFile))
+
+		// Get contents of file (that should be in-memory).
+		size := C.size_t(0)
+		rawData := C.clang_getFileContents(unit, includedFile, &size)
+		if rawData == nil {
+			// Sanity check. This should (hopefully) never trigger.
+			panic("libclang: file contents was not loaded")
+		}
+		data := (*[1 << 24]byte)(unsafe.Pointer(rawData))[:size]
+
+		// Hash the contents if it isn't hashed yet.
+		if _, ok := p.visitedFiles[path]; !ok {
+			// already stored
+			sum := sha512.Sum512_224(data)
+			p.visitedFiles[path] = sum[:]
+		}
+	}
+	inclusionCallbackRef := storedRefs.Put(inclusionCallback)
+	defer storedRefs.Remove(inclusionCallbackRef)
+	C.clang_getInclusions(unit, C.CXInclusionVisitor(C.tinygo_clang_inclusion_visitor), C.CXClientData(inclusionCallbackRef))
 }
 
 //export tinygo_clang_globals_visitor
@@ -171,12 +181,10 @@ func tinygo_clang_globals_visitor(c, parent C.GoCXCursor, client_data C.CXClient
 			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,
+			pos:      pos,
+			variadic: C.clang_isFunctionTypeVariadic(cursorType) != 0,
 		}
 		p.functions[name] = fn
 		for i := 0; i < numArgs; i++ {
@@ -237,14 +245,17 @@ func tinygo_clang_globals_visitor(c, parent C.GoCXCursor, client_data C.CXClient
 		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")
+			p.addError(pos, "internal error: could not find file where macro is defined")
+			break
 		}
 		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")
+			p.addError(pos, "internal error: expected start and end location of a macro to be in the same file")
+			break
 		}
 		if startOffset > endOffset {
-			panic("startOffset > endOffset")
+			p.addError(pos, "internal error: start offset of macro is after end offset")
+			break
 		}
 
 		// read file contents and extract the relevant byte range
@@ -252,59 +263,29 @@ func tinygo_clang_globals_visitor(c, parent C.GoCXCursor, client_data C.CXClient
 		var size C.size_t
 		sourcePtr := C.clang_getFileContents(tu, file, &size)
 		if endOffset >= C.uint(size) {
-			panic("endOffset lies after end of file")
+			p.addError(pos, "internal error: end offset of macro lies after end of file")
+			break
 		}
 		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))
+			p.addError(pos, fmt.Sprintf("internal error: expected macro value to start with %#v, got %#v", name, source))
+			break
 		}
-		value := strings.TrimSpace(source[len(name):])
-		for len(value) != 0 && value[0] == '(' && value[len(value)-1] == ')' {
-			value = strings.TrimSpace(value[1 : len(value)-1])
+		value := source[len(name):]
+		// Try to convert this #define into a Go constant expression.
+		expr, scannerError := parseConst(pos+token.Pos(len(name)), p.fset, value)
+		if scannerError != nil {
+			p.errors = append(p.errors, *scannerError)
 		}
-		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
+		if expr != nil {
+			// Parsing was successful.
+			p.constants[name] = constantInfo{expr, pos}
 		}
+	case C.CXCursor_EnumDecl:
+		// Visit all enums, because the fields may be used even when the enum
+		// type itself is not.
+		typ := C.tinygo_clang_getCursorType(c)
+		p.makeASTType(typ, pos)
 	}
 	return C.CXChildVisit_Continue
 }
@@ -316,11 +297,16 @@ func getString(clangString C.CXString) (s string) {
 	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.
+// getCursorPosition returns a usable token.Pos from a libclang cursor.
 func (p *cgoPackage) getCursorPosition(cursor C.GoCXCursor) token.Pos {
-	location := C.tinygo_clang_getCursorLocation(cursor)
+	return p.getClangLocationPosition(C.tinygo_clang_getCursorLocation(cursor), C.tinygo_clang_Cursor_getTranslationUnit(cursor))
+}
+
+// getClangLocationPosition returns a usable token.Pos based on a libclang
+// location and translation unit. 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) getClangLocationPosition(location C.CXSourceLocation, tu C.CXTranslationUnit) token.Pos {
 	var file C.CXFile
 	var line C.unsigned
 	var column C.unsigned
@@ -334,7 +320,6 @@ func (p *cgoPackage) getCursorPosition(cursor C.GoCXCursor) token.Pos {
 	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]
@@ -348,7 +333,62 @@ func (p *cgoPackage) getCursorPosition(cursor C.GoCXCursor) token.Pos {
 		f.SetLines(lines)
 		p.tokenFiles[filename] = f
 	}
-	return p.tokenFiles[filename].Pos(int(offset))
+	positionFile := p.tokenFiles[filename]
+
+	// Check for alternative line/column information (set with a line directive).
+	var filename2String C.CXString
+	var line2 C.unsigned
+	var column2 C.unsigned
+	C.clang_getPresumedLocation(location, &filename2String, &line2, &column2)
+	filename2 := getString(filename2String)
+	if filename2 != filename || line2 != line || column2 != column {
+		// The location was changed with a preprocessor directive.
+		// TODO: this only works for locations that are added in order. Adding
+		// line/column info to a file that already has line/column info after
+		// the given offset is ignored.
+		positionFile.AddLineColumnInfo(int(offset), filename2, int(line2), int(column2))
+	}
+
+	return positionFile.Pos(int(offset))
+}
+
+// addError is a utility function to add an error to the list of errors. It will
+// convert the token position to a line/column position first, and call
+// addErrorAt.
+func (p *cgoPackage) addError(pos token.Pos, msg string) {
+	p.addErrorAt(p.fset.PositionFor(pos, true), msg)
+}
+
+// addErrorAfter is like addError, but adds the text `after` to the source
+// location.
+func (p *cgoPackage) addErrorAfter(pos token.Pos, after, msg string) {
+	position := p.fset.PositionFor(pos, true)
+	lines := strings.Split(after, "\n")
+	if len(lines) != 1 {
+		// Adjust lines.
+		// For why we can't just do pos+token.Pos(len(after)), see:
+		// https://github.com/golang/go/issues/35803
+		position.Line += len(lines) - 1
+		position.Column = len(lines[len(lines)-1]) + 1
+	} else {
+		position.Column += len(after)
+	}
+	p.addErrorAt(position, msg)
+}
+
+// addErrorAt is a utility function to add an error to the list of errors.
+func (p *cgoPackage) addErrorAt(position token.Position, msg string) {
+	if filepath.IsAbs(position.Filename) {
+		// Relative paths for readability, like other Go parser errors.
+		relpath, err := filepath.Rel(p.dir, position.Filename)
+		if err == nil {
+			position.Filename = relpath
+		}
+	}
+	p.errors = append(p.errors, scanner.Error{
+		Pos: position,
+		Msg: msg,
+	})
 }
 
 // makeASTType return the ast.Expr for the given libclang type. In other words,
@@ -459,7 +499,7 @@ func (p *cgoPackage) makeASTType(typ C.CXType, pos token.Pos) ast.Expr {
 						// 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")
+						p.addError(pos, fmt.Sprintf("unknown char width: %d", typeSize))
 					}
 					switch underlyingType.kind {
 					case C.CXType_Char_S:
@@ -508,48 +548,44 @@ func (p *cgoPackage) makeASTType(typ C.CXType, pos token.Pos) ast.Expr {
 		case C.CXType_Enum:
 			return p.makeASTType(underlying, pos)
 		default:
-			panic("unknown elaborated type")
+			typeKindSpelling := getString(C.clang_getTypeKindSpelling(underlying.kind))
+			p.addError(pos, fmt.Sprintf("unknown elaborated type (libclang type kind %s)", typeKindSpelling))
+			typeName = "<unknown>"
 		}
 	case C.CXType_Record:
 		cursor := C.tinygo_clang_getTypeDeclaration(typ)
 		name := getString(C.tinygo_clang_getCursorSpelling(cursor))
+		var cgoRecordPrefix string
+		switch C.tinygo_clang_getCursorKind(cursor) {
+		case C.CXCursor_StructDecl:
+			cgoRecordPrefix = "struct_"
+		case C.CXCursor_UnionDecl:
+			cgoRecordPrefix = "union_"
+		default:
+			// makeASTRecordType will create an appropriate error.
+			cgoRecordPrefix = "record_"
+		}
 		if name == "" {
 			// Anonymous record, probably inside a typedef.
-			typeExpr, bitfieldList := p.makeASTRecordType(cursor, pos)
-			if bitfieldList != nil {
-				// This struct has bitfields, so we have to declare it as a
-				// named type (for bitfield getters/setters to work).
+			typeInfo := p.makeASTRecordType(cursor, pos)
+			if typeInfo.bitfields != nil || typeInfo.unionSize != 0 {
+				// This record is a union or is a struct with bitfields, so we
+				// have to declare it as a named type (for getters/setters to
+				// work).
 				p.anonStructNum++
-				cgoName := "struct_" + strconv.Itoa(p.anonStructNum)
-				p.elaboratedTypes[cgoName] = &elaboratedTypeInfo{
-					typeExpr:  typeExpr,
-					pos:       pos,
-					bitfields: bitfieldList,
-				}
+				cgoName := cgoRecordPrefix + strconv.Itoa(p.anonStructNum)
+				p.elaboratedTypes[cgoName] = typeInfo
 				return &ast.Ident{
 					NamePos: pos,
 					Name:    "C." + cgoName,
 				}
 			}
-			return typeExpr
+			return typeInfo.typeExpr
 		} else {
-			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")
-			}
+			cgoName := cgoRecordPrefix + name
 			if _, ok := p.elaboratedTypes[cgoName]; !ok {
 				p.elaboratedTypes[cgoName] = nil // predeclare (to avoid endless recursion)
-				typeExpr, bitfieldList := p.makeASTRecordType(cursor, pos)
-				p.elaboratedTypes[cgoName] = &elaboratedTypeInfo{
-					typeExpr:  typeExpr,
-					pos:       pos,
-					bitfields: bitfieldList,
-				}
+				p.elaboratedTypes[cgoName] = p.makeASTRecordType(cursor, pos)
 			}
 			return &ast.Ident{
 				NamePos: pos,
@@ -585,14 +621,10 @@ func (p *cgoPackage) makeASTType(typ C.CXType, pos token.Pos) ast.Expr {
 	}
 	if typeName == "" {
 		// Report this as an error.
-		spelling := getString(C.clang_getTypeSpelling(typ))
-		p.errors = append(p.errors, scanner.Error{
-			Pos: p.fset.PositionFor(pos, true),
-			Msg: fmt.Sprintf("unknown C type: %v (libclang type kind %d)", spelling, typ.kind),
-		})
-		// Fallback, probably incorrect but at least the error points to an odd
-		// type name.
-		typeName = "C." + spelling
+		typeSpelling := getString(C.clang_getTypeSpelling(typ))
+		typeKindSpelling := getString(C.clang_getTypeKindSpelling(typ.kind))
+		p.addError(pos, fmt.Sprintf("unknown C type: %v (libclang type kind %s)", typeSpelling, typeKindSpelling))
+		typeName = "C.<unknown>"
 	}
 	return &ast.Ident{
 		NamePos: pos,
@@ -601,9 +633,8 @@ func (p *cgoPackage) makeASTType(typ C.CXType, pos token.Pos) ast.Expr {
 }
 
 // makeASTRecordType parses a C record (struct or union) and translates it into
-// a Go struct type. Unions are implemented by setting the first field to a
-// zero-lengt "C union" field, which cannot be written in Go directly.
-func (p *cgoPackage) makeASTRecordType(cursor C.GoCXCursor, pos token.Pos) (*ast.StructType, []bitfieldInfo) {
+// a Go struct type.
+func (p *cgoPackage) makeASTRecordType(cursor C.GoCXCursor, pos token.Pos) *elaboratedTypeInfo {
 	fieldList := &ast.FieldList{
 		Opening: pos,
 		Closing: pos,
@@ -620,53 +651,53 @@ func (p *cgoPackage) makeASTRecordType(cursor C.GoCXCursor, pos token.Pos) (*ast
 	}{fieldList, p, &inBitfield, &bitfieldNum, &bitfieldList})
 	defer storedRefs.Remove(ref)
 	C.tinygo_clang_visitChildren(cursor, C.CXCursorVisitor(C.tinygo_clang_struct_visitor), C.CXClientData(ref))
+	renameFieldKeywords(fieldList)
 	switch C.tinygo_clang_getCursorKind(cursor) {
 	case C.CXCursor_StructDecl:
-		return &ast.StructType{
-			Struct: pos,
-			Fields: fieldList,
-		}, bitfieldList
+		return &elaboratedTypeInfo{
+			typeExpr: &ast.StructType{
+				Struct: pos,
+				Fields: fieldList,
+			},
+			pos:       pos,
+			bitfields: bitfieldList,
+		}
 	case C.CXCursor_UnionDecl:
+		typeInfo := &elaboratedTypeInfo{
+			typeExpr: &ast.StructType{
+				Struct: pos,
+				Fields: fieldList,
+			},
+			pos:       pos,
+			bitfields: bitfieldList,
+		}
+		if len(fieldList.List) <= 1 {
+			// Useless union, treat it as a regular struct.
+			return typeInfo
+		}
 		if bitfieldList != nil {
 			// This is valid C... but please don't do this.
-			p.errors = append(p.errors, scanner.Error{
-				Pos: p.fset.PositionFor(pos, true),
-				Msg: fmt.Sprintf("bitfield in a union is not supported"),
-			})
+			p.addError(pos, "bitfield in a union is not supported")
 		}
-		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...)
+		typ := C.tinygo_clang_getCursorType(cursor)
+		alignInBytes := int64(C.clang_Type_getAlignOf(typ))
+		sizeInBytes := int64(C.clang_Type_getSizeOf(typ))
+		if sizeInBytes == 0 {
+			p.addError(pos, "zero-length union is not supported")
 		}
-		return &ast.StructType{
-			Struct: pos,
-			Fields: fieldList,
-		}, bitfieldList
+		typeInfo.unionSize = sizeInBytes
+		typeInfo.unionAlign = alignInBytes
+		return typeInfo
 	default:
-		panic("unknown record declaration")
+		cursorKind := C.tinygo_clang_getCursorKind(cursor)
+		cursorKindSpelling := getString(C.clang_getCursorKindSpelling(cursorKind))
+		p.addError(pos, fmt.Sprintf("expected StructDecl or UnionDecl, not %s", cursorKindSpelling))
+		return &elaboratedTypeInfo{
+			typeExpr: &ast.StructType{
+				Struct: pos,
+			},
+			pos: pos,
+		}
 	}
 }
 
@@ -684,8 +715,17 @@ func tinygo_clang_struct_visitor(c, parent C.GoCXCursor, client_data C.CXClientD
 	inBitfield := passed.inBitfield
 	bitfieldNum := passed.bitfieldNum
 	bitfieldList := passed.bitfieldList
-	if C.tinygo_clang_getCursorKind(c) != C.CXCursor_FieldDecl {
-		panic("expected field inside cursor")
+	pos := p.getCursorPosition(c)
+	switch cursorKind := C.tinygo_clang_getCursorKind(c); cursorKind {
+	case C.CXCursor_FieldDecl:
+		// Expected. This is a regular field.
+	case C.CXCursor_StructDecl, C.CXCursor_UnionDecl:
+		// Ignore. The next field will be the struct/union itself.
+		return C.CXChildVisit_Continue
+	default:
+		cursorKindSpelling := getString(C.clang_getCursorKindSpelling(cursorKind))
+		p.addError(pos, fmt.Sprintf("expected FieldDecl in struct or union, not %s", cursorKindSpelling))
+		return C.CXChildVisit_Continue
 	}
 	name := getString(C.tinygo_clang_getCursorSpelling(c))
 	if name == "" {
@@ -694,7 +734,6 @@ func tinygo_clang_struct_visitor(c, parent C.GoCXCursor, client_data C.CXClientD
 		return C.CXChildVisit_Continue
 	}
 	typ := C.tinygo_clang_getCursorType(c)
-	pos := p.getCursorPosition(c)
 	field := &ast.Field{
 		Type: p.makeASTType(typ, p.getCursorPosition(c)),
 	}
@@ -703,7 +742,8 @@ func tinygo_clang_struct_visitor(c, parent C.GoCXCursor, client_data C.CXClientD
 	bitfieldOffset := offsetof % alignOf
 	if bitfieldOffset != 0 {
 		if C.tinygo_clang_Cursor_isBitField(c) != 1 {
-			panic("expected a bitfield")
+			p.addError(pos, "expected a bitfield")
+			return C.CXChildVisit_Continue
 		}
 		if !*inBitfield {
 			*bitfieldNum++
@@ -761,3 +801,9 @@ func tinygo_clang_enum_visitor(c, parent C.GoCXCursor, client_data C.CXClientDat
 	}
 	return C.CXChildVisit_Continue
 }
+
+//export tinygo_clang_inclusion_visitor
+func tinygo_clang_inclusion_visitor(includedFile C.CXFile, inclusionStack *C.CXSourceLocation, includeLen C.unsigned, clientData C.CXClientData) {
+	callback := storedRefs.Get(unsafe.Pointer(clientData)).(func(C.CXFile))
+	callback(includedFile)
+}

cgo/libclang_config.go

@@ -1,11 +1,14 @@
 // +build !byollvm
+// +build !llvm10
 
 package cgo
 
 /*
-#cgo linux  CFLAGS: -I/usr/lib/llvm-8/include
-#cgo darwin CFLAGS: -I/usr/local/opt/llvm/include
-#cgo linux  LDFLAGS: -L/usr/lib/llvm-8/lib -lclang
-#cgo darwin LDFLAGS: -L/usr/local/opt/llvm/lib -lclang -lffi
+#cgo linux   CFLAGS:  -I/usr/lib/llvm-11/include
+#cgo darwin  CFLAGS:  -I/usr/local/opt/llvm@11/include
+#cgo freebsd CFLAGS:  -I/usr/local/llvm11/include
+#cgo linux   LDFLAGS: -L/usr/lib/llvm-11/lib -lclang
+#cgo darwin  LDFLAGS: -L/usr/local/opt/llvm@11/lib -lclang -lffi
+#cgo freebsd LDFLAGS: -L/usr/local/llvm11/lib -lclang
 */
 import "C"

cgo/libclang_config_llvm10.go

@@ -0,0 +1,14 @@
+// +build !byollvm
+// +build llvm10
+
+package cgo
+
+/*
+#cgo linux  CFLAGS: -I/usr/lib/llvm-10/include
+#cgo darwin CFLAGS: -I/usr/local/opt/llvm@10/include
+#cgo freebsd CFLAGS: -I/usr/local/llvm10/include
+#cgo linux  LDFLAGS: -L/usr/lib/llvm-10/lib -lclang
+#cgo darwin LDFLAGS: -L/usr/local/opt/llvm@10/lib -lclang -lffi
+#cgo freebsd LDFLAGS: -L/usr/local/llvm10/lib -lclang
+*/
+import "C"

cgo/libclang_stubs.c

@@ -3,7 +3,7 @@
 // are slightly different from the ones defined in libclang.go, but they
 // should be ABI compatible.
 
-#include <clang-c/Index.h> // if this fails, install libclang-8-dev
+#include <clang-c/Index.h> // if this fails, install libclang-10-dev
 
 CXCursor tinygo_clang_getTranslationUnitCursor(CXTranslationUnit tu) {
 	return clang_getTranslationUnitCursor(tu);

cgo/security.go

@@ -0,0 +1,301 @@
+// Copyright 2018 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.
+
+// This file has been copied from the Go 1.13 release tree.
+
+// Checking of compiler and linker flags.
+// We must avoid flags like -fplugin=, which can allow
+// arbitrary code execution during the build.
+// Do not make changes here without carefully
+// considering the implications.
+// (That's why the code is isolated in a file named security.go.)
+//
+// Note that -Wl,foo means split foo on commas and pass to
+// the linker, so that -Wl,-foo,bar means pass -foo bar to
+// the linker. Similarly -Wa,foo for the assembler and so on.
+// If any of these are permitted, the wildcard portion must
+// disallow commas.
+//
+// Note also that GNU binutils accept any argument @foo
+// as meaning "read more flags from the file foo", so we must
+// guard against any command-line argument beginning with @,
+// even things like "-I @foo".
+// We use safeArg (which is even more conservative)
+// to reject these.
+//
+// Even worse, gcc -I@foo (one arg) turns into cc1 -I @foo (two args),
+// so although gcc doesn't expand the @foo, cc1 will.
+// So out of paranoia, we reject @ at the beginning of every
+// flag argument that might be split into its own argument.
+
+package cgo
+
+import (
+	"fmt"
+	"os"
+	"regexp"
+	"strings"
+	"unicode/utf8"
+)
+
+var re = regexp.MustCompile
+
+var validCompilerFlags = []*regexp.Regexp{
+	re(`-D([A-Za-z_].*)`),
+	re(`-F([^@\-].*)`),
+	re(`-I([^@\-].*)`),
+	re(`-O`),
+	re(`-O([^@\-].*)`),
+	re(`-W`),
+	re(`-W([^@,]+)`), // -Wall but not -Wa,-foo.
+	re(`-Wa,-mbig-obj`),
+	re(`-Wp,-D([A-Za-z_].*)`),
+	re(`-ansi`),
+	re(`-f(no-)?asynchronous-unwind-tables`),
+	re(`-f(no-)?blocks`),
+	re(`-f(no-)builtin-[a-zA-Z0-9_]*`),
+	re(`-f(no-)?common`),
+	re(`-f(no-)?constant-cfstrings`),
+	re(`-fdiagnostics-show-note-include-stack`),
+	re(`-f(no-)?eliminate-unused-debug-types`),
+	re(`-f(no-)?exceptions`),
+	re(`-f(no-)?fast-math`),
+	re(`-f(no-)?inline-functions`),
+	re(`-finput-charset=([^@\-].*)`),
+	re(`-f(no-)?fat-lto-objects`),
+	re(`-f(no-)?keep-inline-dllexport`),
+	re(`-f(no-)?lto`),
+	re(`-fmacro-backtrace-limit=(.+)`),
+	re(`-fmessage-length=(.+)`),
+	re(`-f(no-)?modules`),
+	re(`-f(no-)?objc-arc`),
+	re(`-f(no-)?objc-nonfragile-abi`),
+	re(`-f(no-)?objc-legacy-dispatch`),
+	re(`-f(no-)?omit-frame-pointer`),
+	re(`-f(no-)?openmp(-simd)?`),
+	re(`-f(no-)?permissive`),
+	re(`-f(no-)?(pic|PIC|pie|PIE)`),
+	re(`-f(no-)?plt`),
+	re(`-f(no-)?rtti`),
+	re(`-f(no-)?split-stack`),
+	re(`-f(no-)?stack-(.+)`),
+	re(`-f(no-)?strict-aliasing`),
+	re(`-f(un)signed-char`),
+	re(`-f(no-)?use-linker-plugin`), // safe if -B is not used; we don't permit -B
+	re(`-f(no-)?visibility-inlines-hidden`),
+	re(`-fsanitize=(.+)`),
+	re(`-ftemplate-depth-(.+)`),
+	re(`-fvisibility=(.+)`),
+	re(`-g([^@\-].*)?`),
+	re(`-m32`),
+	re(`-m64`),
+	re(`-m(abi|arch|cpu|fpu|tune)=([^@\-].*)`),
+	re(`-m(no-)?v?aes`),
+	re(`-marm`),
+	re(`-m(no-)?avx[0-9a-z]*`),
+	re(`-mfloat-abi=([^@\-].*)`),
+	re(`-mfpmath=[0-9a-z,+]*`),
+	re(`-m(no-)?avx[0-9a-z.]*`),
+	re(`-m(no-)?ms-bitfields`),
+	re(`-m(no-)?stack-(.+)`),
+	re(`-mmacosx-(.+)`),
+	re(`-mios-simulator-version-min=(.+)`),
+	re(`-miphoneos-version-min=(.+)`),
+	re(`-mtvos-simulator-version-min=(.+)`),
+	re(`-mtvos-version-min=(.+)`),
+	re(`-mwatchos-simulator-version-min=(.+)`),
+	re(`-mwatchos-version-min=(.+)`),
+	re(`-mnop-fun-dllimport`),
+	re(`-m(no-)?sse[0-9.]*`),
+	re(`-m(no-)?ssse3`),
+	re(`-mthumb(-interwork)?`),
+	re(`-mthreads`),
+	re(`-mwindows`),
+	re(`--param=ssp-buffer-size=[0-9]*`),
+	re(`-pedantic(-errors)?`),
+	re(`-pipe`),
+	re(`-pthread`),
+	re(`-?-std=([^@\-].*)`),
+	re(`-?-stdlib=([^@\-].*)`),
+	re(`--sysroot=([^@\-].*)`),
+	re(`-w`),
+	re(`-x([^@\-].*)`),
+	re(`-v`),
+}
+
+var validCompilerFlagsWithNextArg = []string{
+	"-arch",
+	"-D",
+	"-I",
+	"-framework",
+	"-isysroot",
+	"-isystem",
+	"--sysroot",
+	"-target",
+	"-x",
+}
+
+var validLinkerFlags = []*regexp.Regexp{
+	re(`-F([^@\-].*)`),
+	re(`-l([^@\-].*)`),
+	re(`-L([^@\-].*)`),
+	re(`-O`),
+	re(`-O([^@\-].*)`),
+	re(`-f(no-)?(pic|PIC|pie|PIE)`),
+	re(`-f(no-)?openmp(-simd)?`),
+	re(`-fsanitize=([^@\-].*)`),
+	re(`-flat_namespace`),
+	re(`-g([^@\-].*)?`),
+	re(`-headerpad_max_install_names`),
+	re(`-m(abi|arch|cpu|fpu|tune)=([^@\-].*)`),
+	re(`-mfloat-abi=([^@\-].*)`),
+	re(`-mmacosx-(.+)`),
+	re(`-mios-simulator-version-min=(.+)`),
+	re(`-miphoneos-version-min=(.+)`),
+	re(`-mthreads`),
+	re(`-mwindows`),
+	re(`-(pic|PIC|pie|PIE)`),
+	re(`-pthread`),
+	re(`-rdynamic`),
+	re(`-shared`),
+	re(`-?-static([-a-z0-9+]*)`),
+	re(`-?-stdlib=([^@\-].*)`),
+	re(`-v`),
+
+	// Note that any wildcards in -Wl need to exclude comma,
+	// since -Wl splits its argument at commas and passes
+	// them all to the linker uninterpreted. Allowing comma
+	// in a wildcard would allow tunnelling arbitrary additional
+	// linker arguments through one of these.
+	re(`-Wl,--(no-)?allow-multiple-definition`),
+	re(`-Wl,--(no-)?allow-shlib-undefined`),
+	re(`-Wl,--(no-)?as-needed`),
+	re(`-Wl,-Bdynamic`),
+	re(`-Wl,-berok`),
+	re(`-Wl,-Bstatic`),
+	re(`-WL,-O([^@,\-][^,]*)?`),
+	re(`-Wl,-d[ny]`),
+	re(`-Wl,--disable-new-dtags`),
+	re(`-Wl,-e[=,][a-zA-Z0-9]*`),
+	re(`-Wl,--enable-new-dtags`),
+	re(`-Wl,--end-group`),
+	re(`-Wl,--(no-)?export-dynamic`),
+	re(`-Wl,-framework,[^,@\-][^,]+`),
+	re(`-Wl,-headerpad_max_install_names`),
+	re(`-Wl,--no-undefined`),
+	re(`-Wl,-R([^@\-][^,@]*$)`),
+	re(`-Wl,--just-symbols[=,]([^,@\-][^,@]+)`),
+	re(`-Wl,-rpath(-link)?[=,]([^,@\-][^,]+)`),
+	re(`-Wl,-s`),
+	re(`-Wl,-search_paths_first`),
+	re(`-Wl,-sectcreate,([^,@\-][^,]+),([^,@\-][^,]+),([^,@\-][^,]+)`),
+	re(`-Wl,--start-group`),
+	re(`-Wl,-?-static`),
+	re(`-Wl,-?-subsystem,(native|windows|console|posix|xbox)`),
+	re(`-Wl,-syslibroot[=,]([^,@\-][^,]+)`),
+	re(`-Wl,-undefined[=,]([^,@\-][^,]+)`),
+	re(`-Wl,-?-unresolved-symbols=[^,]+`),
+	re(`-Wl,--(no-)?warn-([^,]+)`),
+	re(`-Wl,-z,(no)?execstack`),
+	re(`-Wl,-z,relro`),
+
+	re(`[a-zA-Z0-9_/].*\.(a|o|obj|dll|dylib|so)`), // direct linker inputs: x.o or libfoo.so (but not -foo.o or @foo.o)
+	re(`\./.*\.(a|o|obj|dll|dylib|so)`),
+}
+
+var validLinkerFlagsWithNextArg = []string{
+	"-arch",
+	"-F",
+	"-l",
+	"-L",
+	"-framework",
+	"-isysroot",
+	"--sysroot",
+	"-target",
+	"-Wl,-framework",
+	"-Wl,-rpath",
+	"-Wl,-R",
+	"-Wl,--just-symbols",
+	"-Wl,-undefined",
+}
+
+func checkCompilerFlags(name string, list []string) error {
+	return checkFlags(name, list, validCompilerFlags, validCompilerFlagsWithNextArg)
+}
+
+func checkLinkerFlags(name string, list []string) error {
+	return checkFlags(name, list, validLinkerFlags, validLinkerFlagsWithNextArg)
+}
+
+func checkFlags(name string, list []string, valid []*regexp.Regexp, validNext []string) error {
+	// Let users override rules with $CGO_CFLAGS_ALLOW, $CGO_CFLAGS_DISALLOW, etc.
+	var (
+		allow    *regexp.Regexp
+		disallow *regexp.Regexp
+	)
+	if env := os.Getenv("CGO_" + name + "_ALLOW"); env != "" {
+		r, err := regexp.Compile(env)
+		if err != nil {
+			return fmt.Errorf("parsing $CGO_%s_ALLOW: %v", name, err)
+		}
+		allow = r
+	}
+	if env := os.Getenv("CGO_" + name + "_DISALLOW"); env != "" {
+		r, err := regexp.Compile(env)
+		if err != nil {
+			return fmt.Errorf("parsing $CGO_%s_DISALLOW: %v", name, err)
+		}
+		disallow = r
+	}
+
+Args:
+	for i := 0; i < len(list); i++ {
+		arg := list[i]
+		if disallow != nil && disallow.FindString(arg) == arg {
+			goto Bad
+		}
+		if allow != nil && allow.FindString(arg) == arg {
+			continue Args
+		}
+		for _, re := range valid {
+			if re.FindString(arg) == arg { // must be complete match
+				continue Args
+			}
+		}
+		for _, x := range validNext {
+			if arg == x {
+				if i+1 < len(list) && safeArg(list[i+1]) {
+					i++
+					continue Args
+				}
+
+				// Permit -Wl,-framework -Wl,name.
+				if i+1 < len(list) &&
+					strings.HasPrefix(arg, "-Wl,") &&
+					strings.HasPrefix(list[i+1], "-Wl,") &&
+					safeArg(list[i+1][4:]) &&
+					!strings.Contains(list[i+1][4:], ",") {
+					i++
+					continue Args
+				}
+
+				if i+1 < len(list) {
+					return fmt.Errorf("invalid flag: %s %s (see https://golang.org/s/invalidflag)", arg, list[i+1])
+				}
+				return fmt.Errorf("invalid flag: %s without argument (see https://golang.org/s/invalidflag)", arg)
+			}
+		}
+	Bad:
+		return fmt.Errorf("invalid flag: %s", arg)
+	}
+	return nil
+}
+
+func safeArg(name string) bool {
+	if name == "" {
+		return false
+	}
+	c := name[0]
+	return '0' <= c && c <= '9' || 'A' <= c && c <= 'Z' || 'a' <= c && c <= 'z' || c == '.' || c == '_' || c == '/' || c >= utf8.RuneSelf
+}

cgo/security_test.go

@@ -0,0 +1,260 @@
+// Copyright 2018 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.
+
+// This file has been copied from the Go 1.13 release tree.
+
+package cgo
+
+import (
+	"os"
+	"testing"
+)
+
+var goodCompilerFlags = [][]string{
+	{"-DFOO"},
+	{"-Dfoo=bar"},
+	{"-F/Qt"},
+	{"-I/"},
+	{"-I/etc/passwd"},
+	{"-I."},
+	{"-O"},
+	{"-O2"},
+	{"-Osmall"},
+	{"-W"},
+	{"-Wall"},
+	{"-fobjc-arc"},
+	{"-fno-objc-arc"},
+	{"-fomit-frame-pointer"},
+	{"-fno-omit-frame-pointer"},
+	{"-fpic"},
+	{"-fno-pic"},
+	{"-fPIC"},
+	{"-fno-PIC"},
+	{"-fpie"},
+	{"-fno-pie"},
+	{"-fPIE"},
+	{"-fno-PIE"},
+	{"-fsplit-stack"},
+	{"-fno-split-stack"},
+	{"-fstack-xxx"},
+	{"-fno-stack-xxx"},
+	{"-fsanitize=hands"},
+	{"-g"},
+	{"-ggdb"},
+	{"-march=souza"},
+	{"-mcpu=123"},
+	{"-mfpu=123"},
+	{"-mtune=happybirthday"},
+	{"-mstack-overflow"},
+	{"-mno-stack-overflow"},
+	{"-mmacosx-version"},
+	{"-mnop-fun-dllimport"},
+	{"-pthread"},
+	{"-std=c99"},
+	{"-xc"},
+	{"-D", "FOO"},
+	{"-D", "foo=bar"},
+	{"-I", "."},
+	{"-I", "/etc/passwd"},
+	{"-I", "世界"},
+	{"-framework", "Chocolate"},
+	{"-x", "c"},
+	{"-v"},
+}
+
+var badCompilerFlags = [][]string{
+	{"-D@X"},
+	{"-D-X"},
+	{"-F@dir"},
+	{"-F-dir"},
+	{"-I@dir"},
+	{"-I-dir"},
+	{"-O@1"},
+	{"-Wa,-foo"},
+	{"-W@foo"},
+	{"-g@gdb"},
+	{"-g-gdb"},
+	{"-march=@dawn"},
+	{"-march=-dawn"},
+	{"-std=@c99"},
+	{"-std=-c99"},
+	{"-x@c"},
+	{"-x-c"},
+	{"-D", "@foo"},
+	{"-D", "-foo"},
+	{"-I", "@foo"},
+	{"-I", "-foo"},
+	{"-framework", "-Caffeine"},
+	{"-framework", "@Home"},
+	{"-x", "--c"},
+	{"-x", "@obj"},
+}
+
+func TestCheckCompilerFlags(t *testing.T) {
+	for _, f := range goodCompilerFlags {
+		if err := checkCompilerFlags("test", f); err != nil {
+			t.Errorf("unexpected error for %q: %v", f, err)
+		}
+	}
+	for _, f := range badCompilerFlags {
+		if err := checkCompilerFlags("test", f); err == nil {
+			t.Errorf("missing error for %q", f)
+		}
+	}
+}
+
+var goodLinkerFlags = [][]string{
+	{"-Fbar"},
+	{"-lbar"},
+	{"-Lbar"},
+	{"-fpic"},
+	{"-fno-pic"},
+	{"-fPIC"},
+	{"-fno-PIC"},
+	{"-fpie"},
+	{"-fno-pie"},
+	{"-fPIE"},
+	{"-fno-PIE"},
+	{"-fsanitize=hands"},
+	{"-g"},
+	{"-ggdb"},
+	{"-march=souza"},
+	{"-mcpu=123"},
+	{"-mfpu=123"},
+	{"-mtune=happybirthday"},
+	{"-pic"},
+	{"-pthread"},
+	{"-Wl,-rpath,foo"},
+	{"-Wl,-rpath,$ORIGIN/foo"},
+	{"-Wl,-R", "/foo"},
+	{"-Wl,-R", "foo"},
+	{"-Wl,-R,foo"},
+	{"-Wl,--just-symbols=foo"},
+	{"-Wl,--just-symbols,foo"},
+	{"-Wl,--warn-error"},
+	{"-Wl,--no-warn-error"},
+	{"foo.so"},
+	{"_世界.dll"},
+	{"./x.o"},
+	{"libcgosotest.dylib"},
+	{"-F", "framework"},
+	{"-l", "."},
+	{"-l", "/etc/passwd"},
+	{"-l", "世界"},
+	{"-L", "framework"},
+	{"-framework", "Chocolate"},
+	{"-v"},
+	{"-Wl,-framework", "-Wl,Chocolate"},
+	{"-Wl,-framework,Chocolate"},
+	{"-Wl,-unresolved-symbols=ignore-all"},
+}
+
+var badLinkerFlags = [][]string{
+	{"-DFOO"},
+	{"-Dfoo=bar"},
+	{"-W"},
+	{"-Wall"},
+	{"-fobjc-arc"},
+	{"-fno-objc-arc"},
+	{"-fomit-frame-pointer"},
+	{"-fno-omit-frame-pointer"},
+	{"-fsplit-stack"},
+	{"-fno-split-stack"},
+	{"-fstack-xxx"},
+	{"-fno-stack-xxx"},
+	{"-mstack-overflow"},
+	{"-mno-stack-overflow"},
+	{"-mnop-fun-dllimport"},
+	{"-std=c99"},
+	{"-xc"},
+	{"-D", "FOO"},
+	{"-D", "foo=bar"},
+	{"-I", "FOO"},
+	{"-L", "@foo"},
+	{"-L", "-foo"},
+	{"-x", "c"},
+	{"-D@X"},
+	{"-D-X"},
+	{"-I@dir"},
+	{"-I-dir"},
+	{"-O@1"},
+	{"-Wa,-foo"},
+	{"-W@foo"},
+	{"-g@gdb"},
+	{"-g-gdb"},
+	{"-march=@dawn"},
+	{"-march=-dawn"},
+	{"-std=@c99"},
+	{"-std=-c99"},
+	{"-x@c"},
+	{"-x-c"},
+	{"-D", "@foo"},
+	{"-D", "-foo"},
+	{"-I", "@foo"},
+	{"-I", "-foo"},
+	{"-l", "@foo"},
+	{"-l", "-foo"},
+	{"-framework", "-Caffeine"},
+	{"-framework", "@Home"},
+	{"-Wl,-framework,-Caffeine"},
+	{"-Wl,-framework", "-Wl,@Home"},
+	{"-Wl,-framework", "@Home"},
+	{"-Wl,-framework,Chocolate,@Home"},
+	{"-x", "--c"},
+	{"-x", "@obj"},
+	{"-Wl,-rpath,@foo"},
+	{"-Wl,-R,foo,bar"},
+	{"-Wl,-R,@foo"},
+	{"-Wl,--just-symbols,@foo"},
+	{"../x.o"},
+}
+
+func TestCheckLinkerFlags(t *testing.T) {
+	for _, f := range goodLinkerFlags {
+		if err := checkLinkerFlags("test", f); err != nil {
+			t.Errorf("unexpected error for %q: %v", f, err)
+		}
+	}
+	for _, f := range badLinkerFlags {
+		if err := checkLinkerFlags("test", f); err == nil {
+			t.Errorf("missing error for %q", f)
+		}
+	}
+}
+
+func TestCheckFlagAllowDisallow(t *testing.T) {
+	if err := checkCompilerFlags("TEST", []string{"-disallow"}); err == nil {
+		t.Fatalf("missing error for -disallow")
+	}
+	os.Setenv("CGO_TEST_ALLOW", "-disallo")
+	if err := checkCompilerFlags("TEST", []string{"-disallow"}); err == nil {
+		t.Fatalf("missing error for -disallow with CGO_TEST_ALLOW=-disallo")
+	}
+	os.Setenv("CGO_TEST_ALLOW", "-disallow")
+	if err := checkCompilerFlags("TEST", []string{"-disallow"}); err != nil {
+		t.Fatalf("unexpected error for -disallow with CGO_TEST_ALLOW=-disallow: %v", err)
+	}
+	os.Unsetenv("CGO_TEST_ALLOW")
+
+	if err := checkCompilerFlags("TEST", []string{"-Wall"}); err != nil {
+		t.Fatalf("unexpected error for -Wall: %v", err)
+	}
+	os.Setenv("CGO_TEST_DISALLOW", "-Wall")
+	if err := checkCompilerFlags("TEST", []string{"-Wall"}); err == nil {
+		t.Fatalf("missing error for -Wall with CGO_TEST_DISALLOW=-Wall")
+	}
+	os.Setenv("CGO_TEST_ALLOW", "-Wall") // disallow wins
+	if err := checkCompilerFlags("TEST", []string{"-Wall"}); err == nil {
+		t.Fatalf("missing error for -Wall with CGO_TEST_DISALLOW=-Wall and CGO_TEST_ALLOW=-Wall")
+	}
+
+	os.Setenv("CGO_TEST_ALLOW", "-fplugin.*")
+	os.Setenv("CGO_TEST_DISALLOW", "-fplugin=lint.so")
+	if err := checkCompilerFlags("TEST", []string{"-fplugin=faster.so"}); err != nil {
+		t.Fatalf("unexpected error for -fplugin=faster.so: %v", err)
+	}
+	if err := checkCompilerFlags("TEST", []string{"-fplugin=lint.so"}); err == nil {
+		t.Fatalf("missing error for -fplugin=lint.so: %v", err)
+	}
+}

cgo/testdata/basic.go

@@ -0,0 +1,3 @@
+package main
+
+import "C"

cgo/testdata/basic.out.go

@@ -0,0 +1,26 @@
+package main
+
+import "unsafe"
+
+var _ unsafe.Pointer
+
+type C.int16_t = int16
+type C.int32_t = int32
+type C.int64_t = int64
+type C.int8_t = int8
+type C.uint16_t = uint16
+type C.uint32_t = uint32
+type C.uint64_t = uint64
+type C.uint8_t = uint8
+type C.uintptr_t = uintptr
+type C.char uint8
+type C.int int32
+type C.long int32
+type C.longlong int64
+type C.schar int8
+type C.short int16
+type C.uchar uint8
+type C.uint uint32
+type C.ulong uint32
+type C.ulonglong uint64
+type C.ushort uint16

cgo/testdata/const.go

@@ -0,0 +1,12 @@
+package main
+
+/*
+#define foo 3
+#define bar foo
+*/
+import "C"
+
+const (
+	Foo = C.foo
+	Bar = C.bar
+)

cgo/testdata/const.out.go

@@ -0,0 +1,29 @@
+package main
+
+import "unsafe"
+
+var _ unsafe.Pointer
+
+const C.bar = C.foo
+const C.foo = 3
+
+type C.int16_t = int16
+type C.int32_t = int32
+type C.int64_t = int64
+type C.int8_t = int8
+type C.uint16_t = uint16
+type C.uint32_t = uint32
+type C.uint64_t = uint64
+type C.uint8_t = uint8
+type C.uintptr_t = uintptr
+type C.char uint8
+type C.int int32
+type C.long int32
+type C.longlong int64
+type C.schar int8
+type C.short int16
+type C.uchar uint8
+type C.uint uint32
+type C.ulong uint32
+type C.ulonglong uint64
+type C.ushort uint16

cgo/testdata/errors.go

@@ -0,0 +1,33 @@
+package main
+
+/*
+#warning some warning
+
+typedef struct {
+	int x;
+	int y;
+} point_t;
+
+typedef someType noType; // undefined type
+
+#define SOME_CONST_1 5) // invalid const syntax
+#define SOME_CONST_2 6) // const not used (so no error)
+#define SOME_CONST_3 1234 // const too large for byte
+*/
+import "C"
+
+// Make sure that errors for the following lines won't change with future
+// additions to the CGo preamble.
+//line errors.go:100
+var (
+	// constant too large
+	_ C.uint8_t = 2 << 10
+
+	// z member does not exist
+	_ C.point_t = C.point_t{z: 3}
+
+	// constant has syntax error
+	_ = C.SOME_CONST_1
+
+	_ byte = C.SOME_CONST_3
+)

cgo/testdata/errors.out.go

@@ -0,0 +1,43 @@
+// CGo errors:
+//     testdata/errors.go:4:2: warning: some warning
+//     testdata/errors.go:11:9: error: unknown type name 'someType'
+//     testdata/errors.go:13:23: unexpected token ), expected end of expression
+
+// Type checking errors after CGo processing:
+//     testdata/errors.go:102: cannot use 2 << 10 (untyped int constant 2048) as uint8 value in variable declaration (overflows)
+//     testdata/errors.go:105: unknown field z in struct literal
+//     testdata/errors.go:108: undeclared name: C.SOME_CONST_1
+//     testdata/errors.go:110: cannot use C.SOME_CONST_3 (untyped int constant 1234) as byte value in variable declaration (overflows)
+
+package main
+
+import "unsafe"
+
+var _ unsafe.Pointer
+
+const C.SOME_CONST_3 = 1234
+
+type C.int16_t = int16
+type C.int32_t = int32
+type C.int64_t = int64
+type C.int8_t = int8
+type C.uint16_t = uint16
+type C.uint32_t = uint32
+type C.uint64_t = uint64
+type C.uint8_t = uint8
+type C.uintptr_t = uintptr
+type C.char uint8
+type C.int int32
+type C.long int32
+type C.longlong int64
+type C.schar int8
+type C.short int16
+type C.uchar uint8
+type C.uint uint32
+type C.ulong uint32
+type C.ulonglong uint64
+type C.ushort uint16
+type C.point_t = struct {
+	x C.int
+	y C.int
+}

cgo/testdata/flags.go

@@ -0,0 +1,37 @@
+package main
+
+/*
+// this name doesn't exist
+#cgo  NOFLAGS: -foo
+
+// unknown flag
+#cgo CFLAGS: -fdoes-not-exist -DNOTDEFINED
+
+#cgo CFLAGS: -DFOO
+
+#cgo CFLAGS: -Iinclude
+#include "foo.h"
+
+#if defined(FOO)
+#define BAR 3
+#else
+#define BAR 5
+#endif
+
+#if defined(NOTDEFINED)
+#warning flag must not be defined
+#endif
+
+// Check Compiler flags
+#cgo LDFLAGS: -lc
+
+// This flag is not valid ldflags
+#cgo LDFLAGS: -does-not-exists
+
+*/
+import "C"
+
+var (
+	_ = C.BAR
+	_ = C.FOO_H
+)

cgo/testdata/flags.out.go

@@ -0,0 +1,34 @@
+// CGo errors:
+//     testdata/flags.go:5:7: invalid #cgo line: NOFLAGS
+//     testdata/flags.go:8:13: invalid flag: -fdoes-not-exist
+//     testdata/flags.go:29:14: invalid flag: -does-not-exists
+
+package main
+
+import "unsafe"
+
+var _ unsafe.Pointer
+
+const C.BAR = 3
+const C.FOO_H = 1
+
+type C.int16_t = int16
+type C.int32_t = int32
+type C.int64_t = int64
+type C.int8_t = int8
+type C.uint16_t = uint16
+type C.uint32_t = uint32
+type C.uint64_t = uint64
+type C.uint8_t = uint8
+type C.uintptr_t = uintptr
+type C.char uint8
+type C.int int32
+type C.long int32
+type C.longlong int64
+type C.schar int8
+type C.short int16
+type C.uchar uint8
+type C.uint uint32
+type C.ulong uint32
+type C.ulonglong uint64
+type C.ushort uint16

cgo/testdata/include/foo.h

@@ -0,0 +1 @@
+#define FOO_H 1

cgo/testdata/types.go

@@ -0,0 +1,175 @@
+package main
+
+/*
+// Simple typedef.
+typedef int myint;
+
+// Structs, with or without name.
+typedef struct {
+	int x;
+	int y;
+} point2d_t;
+typedef struct point3d {
+	int x;
+	int y;
+	int z;
+} point3d_t;
+
+// Structs with reserved field names.
+struct type1 {
+	// All these fields should be renamed.
+	int type;
+	int _type;
+	int __type;
+};
+struct type2 {
+	// This field should not be renamed.
+	int _type;
+};
+
+// Unions.
+typedef union {
+	// Union should be treated as a struct.
+	int i;
+} union1_t;
+typedef union {
+	// Union must contain a single field and have special getters/setters.
+	int    i;
+	double d;
+	short  s;
+} union3_t;
+typedef union union2d {
+	int i;
+	double d[2];
+} union2d_t;
+typedef union {
+	unsigned char arr[10];
+} unionarray_t;
+
+// Nested structs and unions.
+typedef struct {
+	point2d_t begin;
+	point2d_t end;
+	int       tag;
+	union {
+		point2d_t area;
+		point3d_t solid;
+	} coord;
+} struct_nested_t;
+typedef union {
+	point3d_t    point;
+	unionarray_t array;
+	union3_t     thing;
+} union_nested_t;
+
+// Enums. These define constant numbers. All these constants must be given the
+// correct number.
+typedef enum option {
+	optionA,
+	optionB,
+	optionC = -5,
+	optionD,
+	optionE = 10,
+	optionF,
+	optionG,
+} option_t;
+enum unused {
+	unused1 = 5,
+};
+
+// Anonymous enum.
+typedef enum {
+	option2A = 20,
+} option2_t;
+
+// Various types that are usually translated directly to Go types, but storing
+// them in a struct reveals them.
+typedef struct {
+	float  f;
+	double d;
+	int    *ptr;
+} types_t;
+
+// Arrays.
+typedef int myIntArray[10];
+
+// Bitfields.
+typedef struct {
+	unsigned char start;
+	unsigned char a : 5;
+	unsigned char b : 1;
+	unsigned char c : 2;
+	unsigned char :0; // new field
+	unsigned char d : 6;
+	unsigned char e : 3;
+	// Note that C++ allows bitfields bigger than the underlying type.
+} bitfield_t;
+
+// Function signatures.
+void variadic0();
+void variadic2(int x, int y, ...);
+*/
+import "C"
+
+var (
+	// Simple typedefs.
+	_ C.myint
+
+	// Structs.
+	_ C.point2d_t
+	_ C.point3d_t
+	_ C.struct_point3d
+
+	// Structs with reserved field names.
+	_ C.struct_type1
+	_ C.struct_type2
+
+	// Unions.
+	_ C.union1_t
+	_ C.union3_t
+	_ C.union2d_t
+	_ C.unionarray_t
+
+	// Nested structs and unions.
+	_ C.struct_nested_t
+	_ C.union_nested_t
+
+	// Enums (anonymous and named).
+	_ C.option_t
+	_ C.enum_option
+	_ C.option2_t
+
+	// Various types.
+	_ C.types_t
+
+	// Arrays.
+	_ C.myIntArray
+)
+
+// Test bitfield accesses.
+func accessBitfields() {
+	var x C.bitfield_t
+	x.start = 3
+	x.set_bitfield_a(4)
+	x.set_bitfield_b(1)
+	x.set_bitfield_c(2)
+	x.d = 10
+	x.e = 5
+	var _ C.uchar = x.bitfield_a()
+}
+
+// Test union accesses.
+func accessUnion() {
+	var union1 C.union1_t
+	union1.i = 5
+
+	var union2d C.union2d_t
+	var _ *C.int = union2d.unionfield_i()
+	var _ *[2]float64 = union2d.unionfield_d()
+}
+
+// Test function signatures.
+func accessFunctions() {
+	C.variadic0()
+	C.variadic2(3, 5)
+}

cgo/testdata/types.out.go

@@ -0,0 +1,150 @@
+package main
+
+import "unsafe"
+
+var _ unsafe.Pointer
+
+func C.variadic0() //go:variadic
+func C.variadic2(x C.int, y C.int) //go:variadic
+var C.variadic0$funcaddr unsafe.Pointer
+var C.variadic2$funcaddr unsafe.Pointer
+
+const C.option2A = 20
+const C.optionA = 0
+const C.optionB = 1
+const C.optionC = -5
+const C.optionD = -4
+const C.optionE = 10
+const C.optionF = 11
+const C.optionG = 12
+const C.unused1 = 5
+
+type C.int16_t = int16
+type C.int32_t = int32
+type C.int64_t = int64
+type C.int8_t = int8
+type C.uint16_t = uint16
+type C.uint32_t = uint32
+type C.uint64_t = uint64
+type C.uint8_t = uint8
+type C.uintptr_t = uintptr
+type C.char uint8
+type C.int int32
+type C.long int32
+type C.longlong int64
+type C.schar int8
+type C.short int16
+type C.uchar uint8
+type C.uint uint32
+type C.ulong uint32
+type C.ulonglong uint64
+type C.ushort uint16
+type C.bitfield_t = C.struct_4
+type C.myIntArray = [10]C.int
+type C.myint = C.int
+type C.option2_t = C.uint
+type C.option_t = C.enum_option
+type C.point2d_t = struct {
+	x C.int
+	y C.int
+}
+type C.point3d_t = C.struct_point3d
+type C.struct_nested_t = struct {
+	begin C.point2d_t
+	end   C.point2d_t
+	tag   C.int
+
+	coord C.union_2
+}
+type C.types_t = struct {
+	f   float32
+	d   float64
+	ptr *C.int
+}
+type C.union1_t = struct{ i C.int }
+type C.union2d_t = C.union_union2d
+type C.union3_t = C.union_1
+type C.union_nested_t = C.union_3
+type C.unionarray_t = struct{ arr [10]C.uchar }
+
+func (s *C.struct_4) bitfield_a() C.uchar {
+	return s.__bitfield_1 & 0x1f
+}
+func (s *C.struct_4) set_bitfield_a(value C.uchar) {
+	s.__bitfield_1 = s.__bitfield_1&^0x1f | value&0x1f<<0
+}
+func (s *C.struct_4) bitfield_b() C.uchar {
+	return s.__bitfield_1 >> 5 & 0x1
+}
+func (s *C.struct_4) set_bitfield_b(value C.uchar) {
+	s.__bitfield_1 = s.__bitfield_1&^0x20 | value&0x1<<5
+}
+func (s *C.struct_4) bitfield_c() C.uchar {
+	return s.__bitfield_1 >> 6
+}
+func (s *C.struct_4) set_bitfield_c(value C.uchar,
+
+) { s.__bitfield_1 = s.__bitfield_1&0x3f | value<<6 }
+
+type C.struct_4 struct {
+	start        C.uchar
+	__bitfield_1 C.uchar
+
+	d C.uchar
+	e C.uchar
+}
+type C.struct_point3d struct {
+	x C.int
+	y C.int
+	z C.int
+}
+type C.struct_type1 struct {
+	_type   C.int
+	__type  C.int
+	___type C.int
+}
+type C.struct_type2 struct{ _type C.int }
+
+func (union *C.union_1) unionfield_i() *C.int {
+	return (*C.int)(unsafe.Pointer(&union.$union))
+}
+func (union *C.union_1) unionfield_d() *float64 {
+	return (*float64)(unsafe.Pointer(&union.$union))
+}
+func (union *C.union_1) unionfield_s() *C.short {
+	return (*C.short)(unsafe.Pointer(&union.$union))
+}
+
+type C.union_1 struct{ $union uint64 }
+
+func (union *C.union_2) unionfield_area() *C.point2d_t {
+	return (*C.point2d_t)(unsafe.Pointer(&union.$union))
+}
+func (union *C.union_2) unionfield_solid() *C.point3d_t {
+	return (*C.point3d_t)(unsafe.Pointer(&union.$union))
+}
+
+type C.union_2 struct{ $union [3]uint32 }
+
+func (union *C.union_3) unionfield_point() *C.point3d_t {
+	return (*C.point3d_t)(unsafe.Pointer(&union.$union))
+}
+func (union *C.union_3) unionfield_array() *C.unionarray_t {
+	return (*C.unionarray_t)(unsafe.Pointer(&union.$union))
+}
+func (union *C.union_3) unionfield_thing() *C.union3_t {
+	return (*C.union3_t)(unsafe.Pointer(&union.$union))
+}
+
+type C.union_3 struct{ $union [2]uint64 }
+
+func (union *C.union_union2d) unionfield_i() *C.int {
+	return (*C.int)(unsafe.Pointer(&union.$union))
+}
+func (union *C.union_union2d) unionfield_d() *[2]float64 {
+	return (*[2]float64)(unsafe.Pointer(&union.$union))
+}
+
+type C.union_union2d struct{ $union [2]uint64 }
+type C.enum_option C.int
+type C.enum_unused C.uint

commands.go

@@ -1,45 +0,0 @@
-package main
-
-import (
-	"errors"
-	"os"
-	"os/exec"
-	"runtime"
-	"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 init() {
-	// Add the path to a Homebrew-installed LLVM 8 for ease of use (no need to
-	// manually set $PATH).
-	if runtime.GOOS == "darwin" {
-		commands["clang"] = append(commands["clang"], "/usr/local/opt/llvm/bin/clang-8")
-		commands["ld.lld"] = append(commands["ld.lld"], "/usr/local/opt/llvm/bin/ld.lld")
-		commands["wasm-ld"] = append(commands["wasm-ld"], "/usr/local/opt/llvm/bin/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 err
-		}
-		return nil
-	}
-	return errors.New("none of these commands were found in your $PATH: " + strings.Join(cmdNames, " "))
-}

compileopts/config.go

@@ -0,0 +1,373 @@
+// Package compileopts contains the configuration for a single to-be-built
+// binary.
+package compileopts
+
+import (
+	"errors"
+	"fmt"
+	"path/filepath"
+	"regexp"
+	"strings"
+
+	"github.com/tinygo-org/tinygo/goenv"
+)
+
+// Config keeps all configuration affecting the build in a single struct.
+type Config struct {
+	Options        *Options
+	Target         *TargetSpec
+	GoMinorVersion int
+	ClangHeaders   string // Clang built-in header include path
+	TestConfig     TestConfig
+}
+
+// Triple returns the LLVM target triple, like armv6m-none-eabi.
+func (c *Config) Triple() string {
+	return c.Target.Triple
+}
+
+// CPU returns the LLVM CPU name, like atmega328p or arm7tdmi. It may return an
+// empty string if the CPU name is not known.
+func (c *Config) CPU() string {
+	return c.Target.CPU
+}
+
+// Features returns a list of features this CPU supports. For example, for a
+// RISC-V processor, that could be ["+a", "+c", "+m"]. For many targets, an
+// empty list will be returned.
+func (c *Config) Features() []string {
+	return c.Target.Features
+}
+
+// GOOS returns the GOOS of the target. This might not always be the actual OS:
+// for example, bare-metal targets will usually pretend to be linux to get the
+// standard library to compile.
+func (c *Config) GOOS() string {
+	return c.Target.GOOS
+}
+
+// GOARCH returns the GOARCH of the target. This might not always be the actual
+// archtecture: for example, the AVR target is not supported by the Go standard
+// library so such targets will usually pretend to be linux/arm.
+func (c *Config) GOARCH() string {
+	return c.Target.GOARCH
+}
+
+// BuildTags returns the complete list of build tags used during this build.
+func (c *Config) BuildTags() []string {
+	tags := append(c.Target.BuildTags, []string{"tinygo", "gc." + c.GC(), "scheduler." + c.Scheduler(), "serial." + c.Serial()}...)
+	for i := 1; i <= c.GoMinorVersion; i++ {
+		tags = append(tags, fmt.Sprintf("go1.%d", i))
+	}
+	if extraTags := strings.Fields(c.Options.Tags); len(extraTags) != 0 {
+		tags = append(tags, extraTags...)
+	}
+	return tags
+}
+
+// CgoEnabled returns true if (and only if) CGo is enabled. It is true by
+// default and false if CGO_ENABLED is set to "0".
+func (c *Config) CgoEnabled() bool {
+	return goenv.Get("CGO_ENABLED") == "1"
+}
+
+// GC returns the garbage collection strategy in use on this platform. Valid
+// values are "none", "leaking", "extalloc", and "conservative".
+func (c *Config) GC() string {
+	if c.Options.GC != "" {
+		return c.Options.GC
+	}
+	if c.Target.GC != "" {
+		return c.Target.GC
+	}
+	return "conservative"
+}
+
+// NeedsStackObjects returns true if the compiler should insert stack objects
+// that can be traced by the garbage collector.
+func (c *Config) NeedsStackObjects() bool {
+	switch c.GC() {
+	case "conservative", "extalloc":
+		for _, tag := range c.BuildTags() {
+			if tag == "tinygo.wasm" {
+				return true
+			}
+		}
+
+		return false
+	default:
+		return false
+	}
+}
+
+// Scheduler returns the scheduler implementation. Valid values are "none",
+//"coroutines" and "tasks".
+func (c *Config) Scheduler() string {
+	if c.Options.Scheduler != "" {
+		return c.Options.Scheduler
+	}
+	if c.Target.Scheduler != "" {
+		return c.Target.Scheduler
+	}
+	// Fall back to coroutines, which are supported everywhere.
+	return "coroutines"
+}
+
+// Serial returns the serial implementation for this build configuration: uart,
+// usb (meaning USB-CDC), or none.
+func (c *Config) Serial() string {
+	if c.Options.Serial != "" {
+		return c.Options.Serial
+	}
+	if c.Target.Serial != "" {
+		return c.Target.Serial
+	}
+	return "none"
+}
+
+// OptLevels returns the optimization level (0-2), size level (0-2), and inliner
+// threshold as used in the LLVM optimization pipeline.
+func (c *Config) OptLevels() (optLevel, sizeLevel int, inlinerThreshold uint) {
+	switch c.Options.Opt {
+	case "none", "0":
+		return 0, 0, 0 // -O0
+	case "1":
+		return 1, 0, 0 // -O1
+	case "2":
+		return 2, 0, 225 // -O2
+	case "s":
+		return 2, 1, 225 // -Os
+	case "z":
+		return 2, 2, 5 // -Oz, default
+	default:
+		// This is not shown to the user: valid choices are already checked as
+		// part of Options.Verify(). It is here as a sanity check.
+		panic("unknown optimization level: -opt=" + c.Options.Opt)
+	}
+}
+
+// FuncImplementation picks an appropriate func value implementation for the
+// target.
+func (c *Config) FuncImplementation() string {
+	switch c.Scheduler() {
+	case "tasks":
+		// A func value is implemented as a pair of pointers:
+		//     {context, function pointer}
+		// where the context may be a pointer to a heap-allocated struct
+		// containing the free variables, or it may be undef if the function
+		// being pointed to doesn't need a context. The function pointer is a
+		// regular function pointer.
+		return "doubleword"
+	case "none", "coroutines":
+		// As "doubleword", but with the function pointer replaced by a unique
+		// ID per function signature. Function values are called by using a
+		// switch statement and choosing which function to call.
+		// Pick the switch implementation with the coroutines scheduler, as it
+		// allows the use of blocking inside a function that is used as a func
+		// value.
+		return "switch"
+	default:
+		panic("unknown scheduler type")
+	}
+}
+
+// PanicStrategy returns the panic strategy selected for this target. Valid
+// values are "print" (print the panic value, then exit) or "trap" (issue a trap
+// instruction).
+func (c *Config) PanicStrategy() string {
+	return c.Options.PanicStrategy
+}
+
+// AutomaticStackSize returns whether goroutine stack sizes should be determined
+// automatically at compile time, if possible. If it is false, no attempt is
+// made.
+func (c *Config) AutomaticStackSize() bool {
+	if c.Target.AutoStackSize != nil && c.Scheduler() == "tasks" {
+		return *c.Target.AutoStackSize
+	}
+	return false
+}
+
+// RP2040BootPatch returns whether the RP2040 boot patch should be applied that
+// calculates and patches in the checksum for the 2nd stage bootloader.
+func (c *Config) RP2040BootPatch() bool {
+	if c.Target.RP2040BootPatch != nil {
+		return *c.Target.RP2040BootPatch
+	}
+	return false
+}
+
+// CFlags returns the flags to pass to the C compiler. This is necessary for CGo
+// preprocessing.
+func (c *Config) CFlags() []string {
+	var cflags []string
+	for _, flag := range c.Target.CFlags {
+		cflags = append(cflags, strings.ReplaceAll(flag, "{root}", goenv.Get("TINYGOROOT")))
+	}
+	if c.Target.Libc == "picolibc" {
+		root := goenv.Get("TINYGOROOT")
+		cflags = append(cflags, "-nostdlibinc", "-Xclang", "-internal-isystem", "-Xclang", filepath.Join(root, "lib", "picolibc", "newlib", "libc", "include"))
+		cflags = append(cflags, "-I"+filepath.Join(root, "lib/picolibc-include"))
+	}
+	if c.Debug() {
+		cflags = append(cflags, "-g")
+	}
+	return cflags
+}
+
+// LDFlags returns the flags to pass to the linker. A few more flags are needed
+// (like the one for the compiler runtime), but this represents the majority of
+// the flags.
+func (c *Config) LDFlags() []string {
+	root := goenv.Get("TINYGOROOT")
+	// Merge and adjust LDFlags.
+	var ldflags []string
+	for _, flag := range c.Target.LDFlags {
+		ldflags = append(ldflags, strings.ReplaceAll(flag, "{root}", root))
+	}
+	ldflags = append(ldflags, "-L", root)
+	if c.Target.LinkerScript != "" {
+		ldflags = append(ldflags, "-T", c.Target.LinkerScript)
+	}
+	return ldflags
+}
+
+// ExtraFiles returns the list of extra files to be built and linked with the
+// executable. This can include extra C and assembly files.
+func (c *Config) ExtraFiles() []string {
+	return c.Target.ExtraFiles
+}
+
+// DumpSSA returns whether to dump Go SSA while compiling (-dumpssa flag). Only
+// enable this for debugging.
+func (c *Config) DumpSSA() bool {
+	return c.Options.DumpSSA
+}
+
+// VerifyIR returns whether to run extra checks on the IR. This is normally
+// disabled but enabled during testing.
+func (c *Config) VerifyIR() bool {
+	return c.Options.VerifyIR
+}
+
+// Debug returns whether to add debug symbols to the IR, for debugging with GDB
+// and similar.
+func (c *Config) Debug() bool {
+	return c.Options.Debug
+}
+
+// BinaryFormat returns an appropriate binary format, based on the file
+// extension and the configured binary format in the target JSON file.
+func (c *Config) BinaryFormat(ext string) string {
+	switch ext {
+	case ".bin", ".gba", ".nro":
+		// The simplest format possible: dump everything in a raw binary file.
+		if c.Target.BinaryFormat != "" {
+			return c.Target.BinaryFormat
+		}
+		return "bin"
+	case ".hex":
+		// Similar to bin, but includes the start address and is thus usually a
+		// better format.
+		return "hex"
+	case ".uf2":
+		// Special purpose firmware format, mainly used on Adafruit boards.
+		// More information:
+		// https://github.com/Microsoft/uf2
+		return "uf2"
+	case ".zip":
+		if c.Target.BinaryFormat != "" {
+			return c.Target.BinaryFormat
+		}
+		return "zip"
+	default:
+		// Use the ELF format for unrecognized file formats.
+		return "elf"
+	}
+}
+
+// Programmer returns the flash method and OpenOCD interface name given a
+// particular configuration. It may either be all configured in the target JSON
+// file or be modified using the -programmmer command-line option.
+func (c *Config) Programmer() (method, openocdInterface string) {
+	switch c.Options.Programmer {
+	case "":
+		// No configuration supplied.
+		return c.Target.FlashMethod, c.Target.OpenOCDInterface
+	case "openocd", "msd", "command":
+		// The -programmer flag only specifies the flash method.
+		return c.Options.Programmer, c.Target.OpenOCDInterface
+	default:
+		// The -programmer flag specifies something else, assume it specifies
+		// the OpenOCD interface name.
+		return "openocd", c.Options.Programmer
+	}
+}
+
+// OpenOCDConfiguration returns a list of command line arguments to OpenOCD.
+// This list of command-line arguments is based on the various OpenOCD-related
+// flags in the target specification.
+func (c *Config) OpenOCDConfiguration() (args []string, err error) {
+	_, openocdInterface := c.Programmer()
+	if openocdInterface == "" {
+		return nil, errors.New("OpenOCD programmer not set")
+	}
+	if !regexp.MustCompile("^[\\p{L}0-9_-]+$").MatchString(openocdInterface) {
+		return nil, fmt.Errorf("OpenOCD programmer has an invalid name: %#v", openocdInterface)
+	}
+	if c.Target.OpenOCDTarget == "" {
+		return nil, errors.New("OpenOCD chip not set")
+	}
+	if !regexp.MustCompile("^[\\p{L}0-9_-]+$").MatchString(c.Target.OpenOCDTarget) {
+		return nil, fmt.Errorf("OpenOCD target has an invalid name: %#v", c.Target.OpenOCDTarget)
+	}
+	if c.Target.OpenOCDTransport != "" && c.Target.OpenOCDTransport != "swd" {
+		return nil, fmt.Errorf("unknown OpenOCD transport: %#v", c.Target.OpenOCDTransport)
+	}
+	args = []string{"-f", "interface/" + openocdInterface + ".cfg"}
+	for _, cmd := range c.Target.OpenOCDCommands {
+		args = append(args, "-c", cmd)
+	}
+	if c.Target.OpenOCDTransport != "" {
+		args = append(args, "-c", "transport select "+c.Target.OpenOCDTransport)
+	}
+	args = append(args, "-f", "target/"+c.Target.OpenOCDTarget+".cfg")
+	return args, nil
+}
+
+// CodeModel returns the code model used on this platform.
+func (c *Config) CodeModel() string {
+	if c.Target.CodeModel != "" {
+		return c.Target.CodeModel
+	}
+
+	return "default"
+}
+
+// RelocationModel returns the relocation model in use on this platform. Valid
+// values are "static", "pic", "dynamicnopic".
+func (c *Config) RelocationModel() string {
+	if c.Target.RelocationModel != "" {
+		return c.Target.RelocationModel
+	}
+
+	return "static"
+}
+
+// WasmAbi returns the WASM ABI which is specified in the target JSON file, and
+// the value is overridden by `-wasm-abi` flag if it is provided
+func (c *Config) WasmAbi() string {
+	if c.Options.WasmAbi != "" {
+		return c.Options.WasmAbi
+	}
+	return c.Target.WasmAbi
+}
+
+func (c *Config) LLVMFeatures() string {
+	return c.Options.LLVMFeatures
+}
+
+type TestConfig struct {
+	CompileTestBinary bool
+	// TODO: Filter the test functions to run, include verbose flag, etc
+}

compileopts/options.go

@@ -0,0 +1,107 @@
+package compileopts
+
+import (
+	"fmt"
+	"regexp"
+	"strings"
+)
+
+var (
+	validGCOptions            = []string{"none", "leaking", "extalloc", "conservative"}
+	validSchedulerOptions     = []string{"none", "tasks", "coroutines"}
+	validSerialOptions        = []string{"none", "uart", "usb"}
+	validPrintSizeOptions     = []string{"none", "short", "full"}
+	validPanicStrategyOptions = []string{"print", "trap"}
+	validOptOptions           = []string{"none", "0", "1", "2", "s", "z"}
+)
+
+// Options contains extra options to give to the compiler. These options are
+// usually passed from the command line.
+type Options struct {
+	Target          string
+	Opt             string
+	GC              string
+	PanicStrategy   string
+	Scheduler       string
+	Serial          string
+	PrintIR         bool
+	DumpSSA         bool
+	VerifyIR        bool
+	PrintCommands   func(cmd string, args ...string)
+	Debug           bool
+	PrintSizes      string
+	PrintAllocs     *regexp.Regexp // regexp string
+	PrintStacks     bool
+	Tags            string
+	WasmAbi         string
+	GlobalValues    map[string]map[string]string // map[pkgpath]map[varname]value
+	TestConfig      TestConfig
+	Programmer      string
+	OpenOCDCommands []string
+	LLVMFeatures    string
+}
+
+// Verify performs a validation on the given options, raising an error if options are not valid.
+func (o *Options) Verify() error {
+	if o.GC != "" {
+		valid := isInArray(validGCOptions, o.GC)
+		if !valid {
+			return fmt.Errorf(`invalid gc option '%s': valid values are %s`,
+				o.GC,
+				strings.Join(validGCOptions, ", "))
+		}
+	}
+
+	if o.Scheduler != "" {
+		valid := isInArray(validSchedulerOptions, o.Scheduler)
+		if !valid {
+			return fmt.Errorf(`invalid scheduler option '%s': valid values are %s`,
+				o.Scheduler,
+				strings.Join(validSchedulerOptions, ", "))
+		}
+	}
+
+	if o.Serial != "" {
+		valid := isInArray(validSerialOptions, o.Serial)
+		if !valid {
+			return fmt.Errorf(`invalid serial option '%s': valid values are %s`,
+				o.Serial,
+				strings.Join(validSerialOptions, ", "))
+		}
+	}
+
+	if o.PrintSizes != "" {
+		valid := isInArray(validPrintSizeOptions, o.PrintSizes)
+		if !valid {
+			return fmt.Errorf(`invalid size option '%s': valid values are %s`,
+				o.PrintSizes,
+				strings.Join(validPrintSizeOptions, ", "))
+		}
+	}
+
+	if o.PanicStrategy != "" {
+		valid := isInArray(validPanicStrategyOptions, o.PanicStrategy)
+		if !valid {
+			return fmt.Errorf(`invalid panic option '%s': valid values are %s`,
+				o.PanicStrategy,
+				strings.Join(validPanicStrategyOptions, ", "))
+		}
+	}
+
+	if o.Opt != "" {
+		if !isInArray(validOptOptions, o.Opt) {
+			return fmt.Errorf("invalid -opt=%s: valid values are %s", o.Opt, strings.Join(validOptOptions, ", "))
+		}
+	}
+
+	return nil
+}
+
+func isInArray(arr []string, item string) bool {
+	for _, i := range arr {
+		if i == item {
+			return true
+		}
+	}
+	return false
+}

compileopts/options_test.go

@@ -0,0 +1,138 @@
+package compileopts_test
+
+import (
+	"errors"
+	"testing"
+
+	"github.com/tinygo-org/tinygo/compileopts"
+)
+
+func TestVerifyOptions(t *testing.T) {
+
+	expectedGCError := errors.New(`invalid gc option 'incorrect': valid values are none, leaking, extalloc, conservative`)
+	expectedSchedulerError := errors.New(`invalid scheduler option 'incorrect': valid values are none, tasks, coroutines`)
+	expectedPrintSizeError := errors.New(`invalid size option 'incorrect': valid values are none, short, full`)
+	expectedPanicStrategyError := errors.New(`invalid panic option 'incorrect': valid values are print, trap`)
+
+	testCases := []struct {
+		name          string
+		opts          compileopts.Options
+		expectedError error
+	}{
+		{
+			name: "OptionsEmpty",
+			opts: compileopts.Options{},
+		},
+		{
+			name: "InvalidGCOption",
+			opts: compileopts.Options{
+				GC: "incorrect",
+			},
+			expectedError: expectedGCError,
+		},
+		{
+			name: "GCOptionNone",
+			opts: compileopts.Options{
+				GC: "none",
+			},
+		},
+		{
+			name: "GCOptionLeaking",
+			opts: compileopts.Options{
+				GC: "leaking",
+			},
+		},
+		{
+			name: "GCOptionExtalloc",
+			opts: compileopts.Options{
+				GC: "extalloc",
+			},
+		},
+		{
+			name: "GCOptionConservative",
+			opts: compileopts.Options{
+				GC: "conservative",
+			},
+		},
+		{
+			name: "InvalidSchedulerOption",
+			opts: compileopts.Options{
+				Scheduler: "incorrect",
+			},
+			expectedError: expectedSchedulerError,
+		},
+		{
+			name: "SchedulerOptionNone",
+			opts: compileopts.Options{
+				Scheduler: "none",
+			},
+		},
+		{
+			name: "SchedulerOptionTasks",
+			opts: compileopts.Options{
+				Scheduler: "tasks",
+			},
+		},
+		{
+			name: "SchedulerOptionCoroutines",
+			opts: compileopts.Options{
+				Scheduler: "coroutines",
+			},
+		},
+		{
+			name: "InvalidPrintSizeOption",
+			opts: compileopts.Options{
+				PrintSizes: "incorrect",
+			},
+			expectedError: expectedPrintSizeError,
+		},
+		{
+			name: "PrintSizeOptionNone",
+			opts: compileopts.Options{
+				PrintSizes: "none",
+			},
+		},
+		{
+			name: "PrintSizeOptionShort",
+			opts: compileopts.Options{
+				PrintSizes: "short",
+			},
+		},
+		{
+			name: "PrintSizeOptionFull",
+			opts: compileopts.Options{
+				PrintSizes: "full",
+			},
+		},
+		{
+			name: "InvalidPanicOption",
+			opts: compileopts.Options{
+				PanicStrategy: "incorrect",
+			},
+			expectedError: expectedPanicStrategyError,
+		},
+		{
+			name: "PanicOptionPrint",
+			opts: compileopts.Options{
+				PanicStrategy: "print",
+			},
+		},
+		{
+			name: "PanicOptionTrap",
+			opts: compileopts.Options{
+				PanicStrategy: "trap",
+			},
+		},
+	}
+
+	for _, tc := range testCases {
+		t.Run(tc.name, func(t *testing.T) {
+			err := tc.opts.Verify()
+			if tc.expectedError != err {
+				if tc.expectedError.Error() != err.Error() {
+					t.Errorf("expected %v, got %v", tc.expectedError, err)
+				}
+			}
+		})
+	}
+}

compileopts/target.go

@@ -0,0 +1,299 @@
+package compileopts
+
+// This file loads a target specification from a JSON file.
+
+import (
+	"encoding/json"
+	"errors"
+	"io"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"reflect"
+	"runtime"
+	"strings"
+
+	"github.com/tinygo-org/tinygo/goenv"
+)
+
+// Target specification for a given target. Used for bare metal targets.
+//
+// The target specification is mostly inspired by Rust:
+// https://doc.rust-lang.org/nightly/nightly-rustc/rustc_target/spec/struct.TargetOptions.html
+// https://github.com/shepmaster/rust-arduino-blink-led-no-core-with-cargo/blob/master/blink/arduino.json
+type TargetSpec struct {
+	Inherits         []string `json:"inherits"`
+	Triple           string   `json:"llvm-target"`
+	CPU              string   `json:"cpu"`
+	Features         []string `json:"features"`
+	GOOS             string   `json:"goos"`
+	GOARCH           string   `json:"goarch"`
+	BuildTags        []string `json:"build-tags"`
+	GC               string   `json:"gc"`
+	Scheduler        string   `json:"scheduler"`
+	Serial           string   `json:"serial"` // which serial output to use (uart, usb, none)
+	Linker           string   `json:"linker"`
+	RTLib            string   `json:"rtlib"` // compiler runtime library (libgcc, compiler-rt)
+	Libc             string   `json:"libc"`
+	AutoStackSize    *bool    `json:"automatic-stack-size"` // Determine stack size automatically at compile time.
+	DefaultStackSize uint64   `json:"default-stack-size"`   // Default stack size if the size couldn't be determined at compile time.
+	CFlags           []string `json:"cflags"`
+	LDFlags          []string `json:"ldflags"`
+	LinkerScript     string   `json:"linkerscript"`
+	ExtraFiles       []string `json:"extra-files"`
+	RP2040BootPatch  *bool    `json:"rp2040-boot-patch"`        // Patch RP2040 2nd stage bootloader checksum
+	Emulator         []string `json:"emulator" override:"copy"` // inherited Emulator must not be append
+	FlashCommand     string   `json:"flash-command"`
+	GDB              []string `json:"gdb"`
+	PortReset        string   `json:"flash-1200-bps-reset"`
+	SerialPort       []string `json:"serial-port"` // serial port IDs in the form "acm:vid:pid" or "usb:vid:pid"
+	FlashMethod      string   `json:"flash-method"`
+	FlashVolume      string   `json:"msd-volume-name"`
+	FlashFilename    string   `json:"msd-firmware-name"`
+	UF2FamilyID      string   `json:"uf2-family-id"`
+	BinaryFormat     string   `json:"binary-format"`
+	OpenOCDInterface string   `json:"openocd-interface"`
+	OpenOCDTarget    string   `json:"openocd-target"`
+	OpenOCDTransport string   `json:"openocd-transport"`
+	OpenOCDCommands  []string `json:"openocd-commands"`
+	JLinkDevice      string   `json:"jlink-device"`
+	CodeModel        string   `json:"code-model"`
+	RelocationModel  string   `json:"relocation-model"`
+	WasmAbi          string   `json:"wasm-abi"`
+}
+
+// overrideProperties overrides all properties that are set in child into itself using reflection.
+func (spec *TargetSpec) overrideProperties(child *TargetSpec) {
+	specType := reflect.TypeOf(spec).Elem()
+	specValue := reflect.ValueOf(spec).Elem()
+	childValue := reflect.ValueOf(child).Elem()
+
+	for i := 0; i < specType.NumField(); i++ {
+		field := specType.Field(i)
+		src := childValue.Field(i)
+		dst := specValue.Field(i)
+
+		switch kind := field.Type.Kind(); kind {
+		case reflect.String: // for strings, just copy the field of child to spec if not empty
+			if src.Len() > 0 {
+				dst.Set(src)
+			}
+		case reflect.Uint, reflect.Uint32, reflect.Uint64: // for Uint, copy if not zero
+			if src.Uint() != 0 {
+				dst.Set(src)
+			}
+		case reflect.Ptr: // for pointers, copy if not nil
+			if !src.IsNil() {
+				dst.Set(src)
+			}
+		case reflect.Slice: // for slices...
+			if src.Len() > 0 { // ... if not empty ...
+				switch tag := field.Tag.Get("override"); tag {
+				case "copy":
+					// copy the field of child to spec
+					dst.Set(src)
+				case "append", "":
+					// or append the field of child to spec
+					dst.Set(reflect.AppendSlice(src, dst))
+				default:
+					panic("override mode must be 'copy' or 'append' (default). I don't know how to '" + tag + "'.")
+				}
+			}
+		default:
+			panic("unknown field type : " + kind.String())
+		}
+	}
+}
+
+// load reads a target specification from the JSON in the given io.Reader. It
+// may load more targets specified using the "inherits" property.
+func (spec *TargetSpec) load(r io.Reader) error {
+	err := json.NewDecoder(r).Decode(spec)
+	if err != nil {
+		return err
+	}
+
+	return nil
+}
+
+// loadFromGivenStr loads the TargetSpec from the given string that could be:
+// - targets/ directory inside the compiler sources
+// - a relative or absolute path to custom (project specific) target specification .json file;
+//   the Inherits[] could contain the files from target folder (ex. stm32f4disco)
+//   as well as path to custom files (ex. myAwesomeProject.json)
+func (spec *TargetSpec) loadFromGivenStr(str string) error {
+	path := ""
+	if strings.HasSuffix(str, ".json") {
+		path, _ = filepath.Abs(str)
+	} else {
+		path = filepath.Join(goenv.Get("TINYGOROOT"), "targets", strings.ToLower(str)+".json")
+	}
+	fp, err := os.Open(path)
+	if err != nil {
+		return err
+	}
+	defer fp.Close()
+	return spec.load(fp)
+}
+
+// resolveInherits loads inherited targets, recursively.
+func (spec *TargetSpec) resolveInherits() error {
+	// First create a new spec with all the inherited properties.
+	newSpec := &TargetSpec{}
+	for _, name := range spec.Inherits {
+		subtarget := &TargetSpec{}
+		err := subtarget.loadFromGivenStr(name)
+		if err != nil {
+			return err
+		}
+		err = subtarget.resolveInherits()
+		if err != nil {
+			return err
+		}
+		newSpec.overrideProperties(subtarget)
+	}
+
+	// When all properties are loaded, make sure they are properly inherited.
+	newSpec.overrideProperties(spec)
+	*spec = *newSpec
+
+	return nil
+}
+
+// Load a target specification.
+func LoadTarget(target string) (*TargetSpec, error) {
+	if target == "" {
+		// Configure based on GOOS/GOARCH environment variables (falling back to
+		// runtime.GOOS/runtime.GOARCH), and generate a LLVM target based on it.
+		goos := goenv.Get("GOOS")
+		goarch := goenv.Get("GOARCH")
+		llvmos := goos
+		llvmarch := map[string]string{
+			"386":   "i386",
+			"amd64": "x86_64",
+			"arm64": "aarch64",
+		}[goarch]
+		if llvmarch == "" {
+			llvmarch = goarch
+		}
+		target = llvmarch + "--" + llvmos
+		if goarch == "arm" {
+			target += "-gnueabihf"
+		}
+		return defaultTarget(goos, goarch, target)
+	}
+
+	// See whether there is a target specification for this target (e.g.
+	// Arduino).
+	spec := &TargetSpec{}
+	err := spec.loadFromGivenStr(target)
+	if err == nil {
+		// Successfully loaded this target from a built-in .json file. Make sure
+		// it includes all parents as specified in the "inherits" key.
+		err = spec.resolveInherits()
+		if err != nil {
+			return nil, err
+		}
+		return spec, nil
+	} else if !os.IsNotExist(err) {
+		// Expected a 'file not found' error, got something else. Report it as
+		// an error.
+		return nil, err
+	} else {
+		// Load target from given triple, ignore GOOS/GOARCH environment
+		// variables.
+		tripleSplit := strings.Split(target, "-")
+		if len(tripleSplit) < 3 {
+			return nil, errors.New("expected a full LLVM target or a custom target in -target flag")
+		}
+		if tripleSplit[0] == "arm" {
+			// LLVM and Clang have a different idea of what "arm" means, so
+			// upgrade to a slightly more modern ARM. In fact, when you pass
+			// --target=arm--linux-gnueabihf to Clang, it will convert that
+			// internally to armv7-unknown-linux-gnueabihf. Changing the
+			// architecture to armv7 will keep things consistent.
+			tripleSplit[0] = "armv7"
+		}
+		goos := tripleSplit[2]
+		if strings.HasPrefix(goos, "darwin") {
+			goos = "darwin"
+		}
+		goarch := map[string]string{ // map from LLVM arch to Go arch
+			"i386":    "386",
+			"i686":    "386",
+			"x86_64":  "amd64",
+			"aarch64": "arm64",
+			"armv7":   "arm",
+		}[tripleSplit[0]]
+		if goarch == "" {
+			goarch = tripleSplit[0]
+		}
+		return defaultTarget(goos, goarch, strings.Join(tripleSplit, "-"))
+	}
+}
+
+// WindowsBuildNotSupportedErr is being thrown, when goos is windows and no target has been specified.
+var WindowsBuildNotSupportedErr = errors.New("Building Windows binaries is currently not supported. Try specifying a different target")
+
+func defaultTarget(goos, goarch, triple string) (*TargetSpec, error) {
+	if goos == "windows" {
+		return nil, WindowsBuildNotSupportedErr
+	}
+	// No target spec available. Use the default one, useful on most systems
+	// with a regular OS.
+	spec := TargetSpec{
+		Triple:           triple,
+		GOOS:             goos,
+		GOARCH:           goarch,
+		BuildTags:        []string{goos, goarch},
+		Scheduler:        "tasks",
+		Linker:           "cc",
+		DefaultStackSize: 1024 * 64, // 64kB
+		CFlags:           []string{"--target=" + triple},
+		GDB:              []string{"gdb"},
+		PortReset:        "false",
+	}
+	if goos == "darwin" {
+		spec.CFlags = append(spec.CFlags, "-isysroot", "/Library/Developer/CommandLineTools/SDKs/MacOSX.sdk")
+		spec.LDFlags = append(spec.LDFlags, "-Wl,-dead_strip")
+	} else {
+		spec.LDFlags = append(spec.LDFlags, "-no-pie", "-Wl,--gc-sections") // WARNING: clang < 5.0 requires -nopie
+	}
+	if goarch != "wasm" {
+		spec.ExtraFiles = append(spec.ExtraFiles, "src/runtime/gc_"+goarch+".S")
+		spec.ExtraFiles = append(spec.ExtraFiles, "src/internal/task/task_stack_"+goarch+".S")
+	}
+	if goarch != runtime.GOARCH {
+		// Some educated guesses as to how to invoke helper programs.
+		spec.GDB = []string{"gdb-multiarch"}
+		if goarch == "arm" && goos == "linux" {
+			spec.CFlags = append(spec.CFlags, "--sysroot=/usr/arm-linux-gnueabihf")
+			spec.Linker = "arm-linux-gnueabihf-gcc"
+			spec.Emulator = []string{"qemu-arm", "-L", "/usr/arm-linux-gnueabihf"}
+		}
+		if goarch == "arm64" && goos == "linux" {
+			spec.CFlags = append(spec.CFlags, "--sysroot=/usr/aarch64-linux-gnu")
+			spec.Linker = "aarch64-linux-gnu-gcc"
+			spec.Emulator = []string{"qemu-aarch64", "-L", "/usr/aarch64-linux-gnu"}
+		}
+		if goarch == "386" && runtime.GOARCH == "amd64" {
+			spec.CFlags = append(spec.CFlags, "-m32")
+			spec.LDFlags = append(spec.LDFlags, "-m32")
+		}
+	}
+	return &spec, nil
+}
+
+// LookupGDB looks up a gdb executable.
+func (spec *TargetSpec) LookupGDB() (string, error) {
+	if len(spec.GDB) == 0 {
+		return "", errors.New("gdb not configured in the target specification")
+	}
+	for _, d := range spec.GDB {
+		_, err := exec.LookPath(d)
+		if err == nil {
+			return d, nil
+		}
+	}
+	return "", errors.New("no gdb found configured in the target specification (" + strings.Join(spec.GDB, ", ") + ")")
+}

compileopts/target_test.go

@@ -0,0 +1,86 @@
+package compileopts
+
+import (
+	"reflect"
+	"testing"
+)
+
+func TestLoadTarget(t *testing.T) {
+	_, err := LoadTarget("arduino")
+	if err != nil {
+		t.Error("LoadTarget test failed:", err)
+	}
+
+	_, err = LoadTarget("notexist")
+	if err == nil {
+		t.Error("LoadTarget should have failed with non existing target")
+	}
+
+	if err.Error() != "expected a full LLVM target or a custom target in -target flag" {
+		t.Error("LoadTarget failed for wrong reason:", err)
+	}
+}
+
+func TestOverrideProperties(t *testing.T) {
+	baseAutoStackSize := true
+	base := &TargetSpec{
+		GOOS:             "baseGoos",
+		CPU:              "baseCpu",
+		Features:         []string{"bf1", "bf2"},
+		BuildTags:        []string{"bt1", "bt2"},
+		Emulator:         []string{"be1", "be2"},
+		DefaultStackSize: 42,
+		AutoStackSize:    &baseAutoStackSize,
+	}
+	childAutoStackSize := false
+	child := &TargetSpec{
+		GOOS:             "",
+		CPU:              "chlidCpu",
+		Features:         []string{"cf1", "cf2"},
+		Emulator:         []string{"ce1", "ce2"},
+		AutoStackSize:    &childAutoStackSize,
+		DefaultStackSize: 64,
+	}
+
+	base.overrideProperties(child)
+
+	if base.GOOS != "baseGoos" {
+		t.Errorf("Overriding failed : got %v", base.GOOS)
+	}
+	if base.CPU != "chlidCpu" {
+		t.Errorf("Overriding failed : got %v", base.CPU)
+	}
+	if !reflect.DeepEqual(base.Features, []string{"cf1", "cf2", "bf1", "bf2"}) {
+		t.Errorf("Overriding failed : got %v", base.Features)
+	}
+	if !reflect.DeepEqual(base.BuildTags, []string{"bt1", "bt2"}) {
+		t.Errorf("Overriding failed : got %v", base.BuildTags)
+	}
+	if !reflect.DeepEqual(base.Emulator, []string{"ce1", "ce2"}) {
+		t.Errorf("Overriding failed : got %v", base.Emulator)
+	}
+	if *base.AutoStackSize != false {
+		t.Errorf("Overriding failed : got %v", base.AutoStackSize)
+	}
+	if base.DefaultStackSize != 64 {
+		t.Errorf("Overriding failed : got %v", base.DefaultStackSize)
+	}
+
+	baseAutoStackSize = true
+	base = &TargetSpec{
+		AutoStackSize:    &baseAutoStackSize,
+		DefaultStackSize: 42,
+	}
+	child = &TargetSpec{
+		AutoStackSize:    nil,
+		DefaultStackSize: 0,
+	}
+	base.overrideProperties(child)
+	if *base.AutoStackSize != true {
+		t.Errorf("Overriding failed : got %v", base.AutoStackSize)
+	}
+	if base.DefaultStackSize != 42 {
+		t.Errorf("Overriding failed : got %v", base.DefaultStackSize)
+	}
+
+}

compiler/asserts.go

@@ -4,16 +4,19 @@ package compiler
 // required by the Go programming language.
 
 import (
+	"fmt"
+	"go/token"
 	"go/types"
 
+	"golang.org/x/tools/go/ssa"
 	"tinygo.org/x/go-llvm"
 )
 
-// emitLookupBoundsCheck emits a bounds check before doing a lookup into a
+// createLookupBoundsCheck emits a bounds check before doing a lookup into a
 // slice. This is required by the Go language spec: an index out of bounds must
 // cause a panic.
-func (c *Compiler) emitLookupBoundsCheck(frame *Frame, arrayLen, index llvm.Value, indexType types.Type) {
-	if frame.fn.IsNoBounds() {
+func (b *builder) createLookupBoundsCheck(arrayLen, index llvm.Value, indexType types.Type) {
+	if b.info.nobounds {
 		// The //go:nobounds pragma was added to the function to avoid bounds
 		// checking.
 		return
@@ -22,42 +25,30 @@ func (c *Compiler) emitLookupBoundsCheck(frame *Frame, arrayLen, index llvm.Valu
 	if index.Type().IntTypeWidth() < arrayLen.Type().IntTypeWidth() {
 		// Sometimes, the index can be e.g. an uint8 or int8, and we have to
 		// correctly extend that type.
-		if indexType.(*types.Basic).Info()&types.IsUnsigned == 0 {
-			index = c.builder.CreateZExt(index, arrayLen.Type(), "")
+		if indexType.Underlying().(*types.Basic).Info()&types.IsUnsigned == 0 {
+			index = b.CreateZExt(index, arrayLen.Type(), "")
 		} else {
-			index = c.builder.CreateSExt(index, arrayLen.Type(), "")
+			index = b.CreateSExt(index, arrayLen.Type(), "")
 		}
 	} else if index.Type().IntTypeWidth() > arrayLen.Type().IntTypeWidth() {
 		// The index is bigger than the array length type, so extend it.
-		arrayLen = c.builder.CreateZExt(arrayLen, index.Type(), "")
+		arrayLen = b.CreateZExt(arrayLen, index.Type(), "")
 	}
 
-	faultBlock := c.ctx.AddBasicBlock(frame.fn.LLVMFn, "lookup.outofbounds")
-	nextBlock := c.ctx.AddBasicBlock(frame.fn.LLVMFn, "lookup.next")
-	frame.blockExits[frame.currentBlock] = nextBlock // adjust outgoing block for phi nodes
-
 	// Now do the bounds check: index >= arrayLen
-	outOfBounds := c.builder.CreateICmp(llvm.IntUGE, index, arrayLen, "")
-	c.builder.CreateCondBr(outOfBounds, faultBlock, nextBlock)
-
-	// Fail: this is a nil pointer, exit with a panic.
-	c.builder.SetInsertPointAtEnd(faultBlock)
-	c.createRuntimeCall("lookupPanic", nil, "")
-	c.builder.CreateUnreachable()
-
-	// Ok: this is a valid pointer.
-	c.builder.SetInsertPointAtEnd(nextBlock)
+	outOfBounds := b.CreateICmp(llvm.IntUGE, index, arrayLen, "")
+	b.createRuntimeAssert(outOfBounds, "lookup", "lookupPanic")
 }
 
-// emitSliceBoundsCheck emits a bounds check before a slicing operation to make
+// createSliceBoundsCheck emits a bounds check before a slicing operation to make
 // sure it is within bounds.
 //
 // This function is both used for slicing a slice (low and high have their
 // normal meaning) and for creating a new slice, where 'capacity' means the
 // biggest possible slice capacity, 'low' means len and 'high' means cap. The
 // logic is the same in both cases.
-func (c *Compiler) emitSliceBoundsCheck(frame *Frame, capacity, low, high, max llvm.Value, lowType, highType, maxType *types.Basic) {
-	if frame.fn.IsNoBounds() {
+func (b *builder) createSliceBoundsCheck(capacity, low, high, max llvm.Value, lowType, highType, maxType *types.Basic) {
+	if b.info.nobounds {
 		// The //go:nobounds pragma was added to the function to avoid bounds
 		// checking.
 		return
@@ -75,92 +66,171 @@ func (c *Compiler) emitSliceBoundsCheck(frame *Frame, capacity, low, high, max l
 		capacityType = max.Type()
 	}
 	if capacityType != capacity.Type() {
-		capacity = c.builder.CreateZExt(capacity, capacityType, "")
+		capacity = b.CreateZExt(capacity, capacityType, "")
 	}
 
 	// Extend low and high to be the same size as capacity.
 	if low.Type().IntTypeWidth() < capacityType.IntTypeWidth() {
 		if lowType.Info()&types.IsUnsigned != 0 {
-			low = c.builder.CreateZExt(low, capacityType, "")
+			low = b.CreateZExt(low, capacityType, "")
 		} else {
-			low = c.builder.CreateSExt(low, capacityType, "")
+			low = b.CreateSExt(low, capacityType, "")
 		}
 	}
 	if high.Type().IntTypeWidth() < capacityType.IntTypeWidth() {
 		if highType.Info()&types.IsUnsigned != 0 {
-			high = c.builder.CreateZExt(high, capacityType, "")
+			high = b.CreateZExt(high, capacityType, "")
 		} else {
-			high = c.builder.CreateSExt(high, capacityType, "")
+			high = b.CreateSExt(high, capacityType, "")
 		}
 	}
 	if max.Type().IntTypeWidth() < capacityType.IntTypeWidth() {
 		if maxType.Info()&types.IsUnsigned != 0 {
-			max = c.builder.CreateZExt(max, capacityType, "")
+			max = b.CreateZExt(max, capacityType, "")
 		} else {
-			max = c.builder.CreateSExt(max, capacityType, "")
+			max = b.CreateSExt(max, capacityType, "")
 		}
 	}
 
-	faultBlock := c.ctx.AddBasicBlock(frame.fn.LLVMFn, "slice.outofbounds")
-	nextBlock := c.ctx.AddBasicBlock(frame.fn.LLVMFn, "slice.next")
-	frame.blockExits[frame.currentBlock] = nextBlock // adjust outgoing block for phi nodes
-
 	// Now do the bounds check: low > high || high > capacity
-	outOfBounds1 := c.builder.CreateICmp(llvm.IntUGT, low, high, "slice.lowhigh")
-	outOfBounds2 := c.builder.CreateICmp(llvm.IntUGT, high, max, "slice.highmax")
-	outOfBounds3 := c.builder.CreateICmp(llvm.IntUGT, max, capacity, "slice.maxcap")
-	outOfBounds := c.builder.CreateOr(outOfBounds1, outOfBounds2, "slice.lowmax")
-	outOfBounds = c.builder.CreateOr(outOfBounds, outOfBounds3, "slice.lowcap")
-	c.builder.CreateCondBr(outOfBounds, faultBlock, nextBlock)
-
-	// Fail: this is a nil pointer, exit with a panic.
-	c.builder.SetInsertPointAtEnd(faultBlock)
-	c.createRuntimeCall("slicePanic", nil, "")
-	c.builder.CreateUnreachable()
-
-	// Ok: this is a valid pointer.
-	c.builder.SetInsertPointAtEnd(nextBlock)
+	outOfBounds1 := b.CreateICmp(llvm.IntUGT, low, high, "slice.lowhigh")
+	outOfBounds2 := b.CreateICmp(llvm.IntUGT, high, max, "slice.highmax")
+	outOfBounds3 := b.CreateICmp(llvm.IntUGT, max, capacity, "slice.maxcap")
+	outOfBounds := b.CreateOr(outOfBounds1, outOfBounds2, "slice.lowmax")
+	outOfBounds = b.CreateOr(outOfBounds, outOfBounds3, "slice.lowcap")
+	b.createRuntimeAssert(outOfBounds, "slice", "slicePanic")
 }
 
-// emitNilCheck checks whether the given pointer is nil, and panics if it is. It
-// has no effect in well-behaved programs, but makes sure no uncaught nil
+// createChanBoundsCheck creates a bounds check before creating a new channel to
+// check that the value is not too big for runtime.chanMake.
+func (b *builder) createChanBoundsCheck(elementSize uint64, bufSize llvm.Value, bufSizeType *types.Basic, pos token.Pos) {
+	if b.info.nobounds {
+		// The //go:nobounds pragma was added to the function to avoid bounds
+		// checking.
+		return
+	}
+
+	// Check whether the bufSize parameter must be cast to a wider integer for
+	// comparison.
+	if bufSize.Type().IntTypeWidth() < b.uintptrType.IntTypeWidth() {
+		if bufSizeType.Info()&types.IsUnsigned != 0 {
+			// Unsigned, so zero-extend to uint type.
+			bufSizeType = types.Typ[types.Uint]
+			bufSize = b.CreateZExt(bufSize, b.intType, "")
+		} else {
+			// Signed, so sign-extend to int type.
+			bufSizeType = types.Typ[types.Int]
+			bufSize = b.CreateSExt(bufSize, b.intType, "")
+		}
+	}
+
+	// Calculate (^uintptr(0)) >> 1, which is the max value that fits in an
+	// uintptr if uintptrs were signed.
+	maxBufSize := llvm.ConstLShr(llvm.ConstNot(llvm.ConstInt(b.uintptrType, 0, false)), llvm.ConstInt(b.uintptrType, 1, false))
+	if elementSize > maxBufSize.ZExtValue() {
+		b.addError(pos, fmt.Sprintf("channel element type is too big (%v bytes)", elementSize))
+		return
+	}
+	// Avoid divide-by-zero.
+	if elementSize == 0 {
+		elementSize = 1
+	}
+	// Make the maxBufSize actually the maximum allowed value (in number of
+	// elements in the channel buffer).
+	maxBufSize = llvm.ConstUDiv(maxBufSize, llvm.ConstInt(b.uintptrType, elementSize, false))
+
+	// Make sure maxBufSize has the same type as bufSize.
+	if maxBufSize.Type() != bufSize.Type() {
+		maxBufSize = llvm.ConstZExt(maxBufSize, bufSize.Type())
+	}
+
+	// Do the check for a too large (or negative) buffer size.
+	bufSizeTooBig := b.CreateICmp(llvm.IntUGE, bufSize, maxBufSize, "")
+	b.createRuntimeAssert(bufSizeTooBig, "chan", "chanMakePanic")
+}
+
+// createNilCheck checks whether the given pointer is nil, and panics if it is.
+// It has no effect in well-behaved programs, but makes sure no uncaught nil
 // pointer dereferences exist in valid Go code.
-func (c *Compiler) emitNilCheck(frame *Frame, ptr llvm.Value, blockPrefix string) {
+func (b *builder) createNilCheck(inst ssa.Value, ptr llvm.Value, blockPrefix string) {
 	// Check whether we need to emit this check at all.
 	if !ptr.IsAGlobalValue().IsNil() {
 		return
 	}
 
-	// Check whether this is a nil pointer.
-	faultBlock := c.ctx.AddBasicBlock(frame.fn.LLVMFn, blockPrefix+".nil")
-	nextBlock := c.ctx.AddBasicBlock(frame.fn.LLVMFn, blockPrefix+".next")
-	frame.blockExits[frame.currentBlock] = nextBlock // adjust outgoing block for phi nodes
+	switch inst := inst.(type) {
+	case *ssa.Alloc:
+		// An alloc is never nil.
+		return
+	case *ssa.FreeVar:
+		// A free variable is allocated in a parent function and is thus never
+		// nil.
+		return
+	case *ssa.IndexAddr:
+		// This pointer is the result of an index operation into a slice or
+		// array. Such slices/arrays are already bounds checked so the pointer
+		// must be a valid (non-nil) pointer. No nil checking is necessary.
+		return
+	case *ssa.Convert:
+		// This is a pointer that comes from a conversion from unsafe.Pointer.
+		// Don't do nil checking because this is unsafe code and the code should
+		// know what it is doing.
+		// Note: all *ssa.Convert instructions that result in a pointer must
+		// come from unsafe.Pointer. Testing here for unsafe.Pointer to be sure.
+		if inst.X.Type() == types.Typ[types.UnsafePointer] {
+			return
+		}
+	}
 
 	// Compare against nil.
-	var isnil llvm.Value
-	if ptr.Type().PointerAddressSpace() == 0 {
-		// Do the nil check using the isnil builtin, which marks the parameter
-		// as nocapture.
-		// The reason it has to go through a builtin, is that a regular icmp
-		// instruction may capture the pointer in LLVM semantics, see
-		// https://reviews.llvm.org/D60047 for details. Pointer capturing
-		// unfortunately breaks escape analysis, so we use this trick to let the
-		// functionattr pass know that this pointer doesn't really escape.
-		ptr = c.builder.CreateBitCast(ptr, c.i8ptrType, "")
-		isnil = c.createRuntimeCall("isnil", []llvm.Value{ptr}, "")
-	} else {
-		// Do the nil check using a regular icmp. This can happen with function
-		// pointers on AVR, which don't benefit from escape analysis anyway.
-		nilptr := llvm.ConstPointerNull(ptr.Type())
-		isnil = c.builder.CreateICmp(llvm.IntEQ, ptr, nilptr, "")
-	}
-	c.builder.CreateCondBr(isnil, faultBlock, nextBlock)
+	// We previously used a hack to make sure this wouldn't break escape
+	// analysis, but this is not necessary anymore since
+	// https://reviews.llvm.org/D60047 has been merged.
+	nilptr := llvm.ConstPointerNull(ptr.Type())
+	isnil := b.CreateICmp(llvm.IntEQ, ptr, nilptr, "")
 
-	// Fail: this is a nil pointer, exit with a panic.
-	c.builder.SetInsertPointAtEnd(faultBlock)
-	c.createRuntimeCall("nilPanic", nil, "")
-	c.builder.CreateUnreachable()
-
-	// Ok: this is a valid pointer.
-	c.builder.SetInsertPointAtEnd(nextBlock)
+	// Emit the nil check in IR.
+	b.createRuntimeAssert(isnil, blockPrefix, "nilPanic")
+}
+
+// createNegativeShiftCheck creates an assertion that panics if the given shift value is negative.
+// This function assumes that the shift value is signed.
+func (b *builder) createNegativeShiftCheck(shift llvm.Value) {
+	if b.info.nobounds {
+		// Function disabled bounds checking - skip shift check.
+		return
+	}
+
+	// isNegative = shift < 0
+	isNegative := b.CreateICmp(llvm.IntSLT, shift, llvm.ConstInt(shift.Type(), 0, false), "")
+	b.createRuntimeAssert(isNegative, "shift", "negativeShiftPanic")
+}
+
+// createRuntimeAssert is a common function to create a new branch on an assert
+// bool, calling an assert func if the assert value is true (1).
+func (b *builder) createRuntimeAssert(assert llvm.Value, blockPrefix, assertFunc string) {
+	// Check whether we can resolve this check at compile time.
+	if !assert.IsAConstantInt().IsNil() {
+		val := assert.ZExtValue()
+		if val == 0 {
+			// Everything is constant so the check does not have to be emitted
+			// in IR. This avoids emitting some redundant IR.
+			return
+		}
+	}
+
+	faultBlock := b.ctx.AddBasicBlock(b.llvmFn, blockPrefix+".throw")
+	nextBlock := b.ctx.AddBasicBlock(b.llvmFn, blockPrefix+".next")
+	b.blockExits[b.currentBlock] = nextBlock // adjust outgoing block for phi nodes
+
+	// Now branch to the out-of-bounds or the regular block.
+	b.CreateCondBr(assert, faultBlock, nextBlock)
+
+	// Fail: the assert triggered so panic.
+	b.SetInsertPointAtEnd(faultBlock)
+	b.createRuntimeCall(assertFunc, nil, "")
+	b.CreateUnreachable()
+
+	// Ok: assert didn't trigger so continue normally.
+	b.SetInsertPointAtEnd(nextBlock)
 }

compiler/atomic.go

@@ -0,0 +1,84 @@
+package compiler
+
+import (
+	"strings"
+
+	"golang.org/x/tools/go/ssa"
+	"tinygo.org/x/go-llvm"
+)
+
+// createAtomicOp lowers an atomic library call by lowering it as an LLVM atomic
+// operation. It returns the result of the operation and true if the call could
+// be lowered inline, and false otherwise.
+func (b *builder) createAtomicOp(call *ssa.CallCommon) (llvm.Value, bool) {
+	name := call.Value.(*ssa.Function).Name()
+	switch name {
+	case "AddInt32", "AddInt64", "AddUint32", "AddUint64", "AddUintptr":
+		ptr := b.getValue(call.Args[0])
+		val := b.getValue(call.Args[1])
+		oldVal := b.CreateAtomicRMW(llvm.AtomicRMWBinOpAdd, ptr, val, llvm.AtomicOrderingSequentiallyConsistent, true)
+		// Return the new value, not the original value returned by atomicrmw.
+		return b.CreateAdd(oldVal, val, ""), true
+	case "SwapInt32", "SwapInt64", "SwapUint32", "SwapUint64", "SwapUintptr", "SwapPointer":
+		ptr := b.getValue(call.Args[0])
+		val := b.getValue(call.Args[1])
+		isPointer := val.Type().TypeKind() == llvm.PointerTypeKind
+		if isPointer {
+			// atomicrmw only supports integers, so cast to an integer.
+			val = b.CreatePtrToInt(val, b.uintptrType, "")
+			ptr = b.CreateBitCast(ptr, llvm.PointerType(val.Type(), 0), "")
+		}
+		oldVal := b.CreateAtomicRMW(llvm.AtomicRMWBinOpXchg, ptr, val, llvm.AtomicOrderingSequentiallyConsistent, true)
+		if isPointer {
+			oldVal = b.CreateIntToPtr(oldVal, b.i8ptrType, "")
+		}
+		return oldVal, true
+	case "CompareAndSwapInt32", "CompareAndSwapInt64", "CompareAndSwapUint32", "CompareAndSwapUint64", "CompareAndSwapUintptr", "CompareAndSwapPointer":
+		ptr := b.getValue(call.Args[0])
+		old := b.getValue(call.Args[1])
+		newVal := b.getValue(call.Args[2])
+		if strings.HasSuffix(name, "64") {
+			arch := strings.Split(b.Triple, "-")[0]
+			if strings.HasPrefix(arch, "arm") && strings.HasSuffix(arch, "m") {
+				// Work around a bug in LLVM, at least LLVM 11:
+				// https://reviews.llvm.org/D95891
+				// Check for armv6m, armv7, armv7em, and perhaps others.
+				// See also: https://gcc.gnu.org/onlinedocs/gcc/_005f_005fsync-Builtins.html
+				compareAndSwap := b.mod.NamedFunction("__sync_val_compare_and_swap_8")
+				if compareAndSwap.IsNil() {
+					// Declare the function if it isn't already declared.
+					i64Type := b.ctx.Int64Type()
+					fnType := llvm.FunctionType(i64Type, []llvm.Type{llvm.PointerType(i64Type, 0), i64Type, i64Type}, false)
+					compareAndSwap = llvm.AddFunction(b.mod, "__sync_val_compare_and_swap_8", fnType)
+				}
+				actualOldValue := b.CreateCall(compareAndSwap, []llvm.Value{ptr, old, newVal}, "")
+				// The __sync_val_compare_and_swap_8 function returns the old
+				// value. However, we shouldn't return the old value, we should
+				// return whether the compare/exchange was successful. This is
+				// easily done by comparing the returned (actual) old value with
+				// the expected old value passed to
+				// __sync_val_compare_and_swap_8.
+				swapped := b.CreateICmp(llvm.IntEQ, old, actualOldValue, "")
+				return swapped, true
+			}
+		}
+		tuple := b.CreateAtomicCmpXchg(ptr, old, newVal, llvm.AtomicOrderingSequentiallyConsistent, llvm.AtomicOrderingSequentiallyConsistent, true)
+		swapped := b.CreateExtractValue(tuple, 1, "")
+		return swapped, true
+	case "LoadInt32", "LoadInt64", "LoadUint32", "LoadUint64", "LoadUintptr", "LoadPointer":
+		ptr := b.getValue(call.Args[0])
+		val := b.CreateLoad(ptr, "")
+		val.SetOrdering(llvm.AtomicOrderingSequentiallyConsistent)
+		val.SetAlignment(b.targetData.PrefTypeAlignment(val.Type())) // required
+		return val, true
+	case "StoreInt32", "StoreInt64", "StoreUint32", "StoreUint64", "StoreUintptr", "StorePointer":
+		ptr := b.getValue(call.Args[0])
+		val := b.getValue(call.Args[1])
+		store := b.CreateStore(val, ptr)
+		store.SetOrdering(llvm.AtomicOrderingSequentiallyConsistent)
+		store.SetAlignment(b.targetData.PrefTypeAlignment(val.Type())) // required
+		return store, true
+	default:
+		return llvm.Value{}, false
+	}
+}

compiler/calls.go

@@ -1,6 +1,9 @@
 package compiler
 
 import (
+	"go/types"
+	"strconv"
+
 	"golang.org/x/tools/go/ssa"
 	"tinygo.org/x/go-llvm"
 )
@@ -10,59 +13,80 @@ import (
 
 // The maximum number of arguments that can be expanded from a single struct. If
 // a struct contains more fields, it is passed as a struct without expanding.
-const MaxFieldsPerParam = 3
+const maxFieldsPerParam = 3
 
-// Shortcut: create a call to runtime.<fnName> with the given arguments.
-func (c *Compiler) createRuntimeCall(fnName string, args []llvm.Value, name string) llvm.Value {
-	runtimePkg := c.ir.Program.ImportedPackage("runtime")
-	member := runtimePkg.Members[fnName]
-	if member == nil {
-		panic("trying to call runtime." + fnName)
-	}
-	fn := c.ir.GetFunction(member.(*ssa.Function))
-	if !fn.IsExported() {
-		args = append(args, llvm.Undef(c.i8ptrType))            // unused context parameter
-		args = append(args, llvm.ConstPointerNull(c.i8ptrType)) // coroutine handle
-	}
-	return c.createCall(fn.LLVMFn, args, name)
+// paramInfo contains some information collected about a function parameter,
+// useful while declaring or defining a function.
+type paramInfo struct {
+	llvmType llvm.Type
+	name     string // name, possibly with suffixes for e.g. struct fields
+	flags    paramFlags
 }
 
-// Create a call to the given function with the arguments possibly expanded.
-func (c *Compiler) createCall(fn llvm.Value, args []llvm.Value, name string) llvm.Value {
+// paramFlags identifies parameter attributes for flags. Most importantly, it
+// determines which parameters are dereferenceable_or_null and which aren't.
+type paramFlags uint8
+
+const (
+	// Parameter may have the deferenceable_or_null attribute. This attribute
+	// cannot be applied to unsafe.Pointer and to the data pointer of slices.
+	paramIsDeferenceableOrNull = 1 << iota
+)
+
+// createCall creates a new call to runtime.<fnName> with the given arguments.
+func (b *builder) createRuntimeCall(fnName string, args []llvm.Value, name string) llvm.Value {
+	fn := b.program.ImportedPackage("runtime").Members[fnName].(*ssa.Function)
+	llvmFn := b.getFunction(fn)
+	if llvmFn.IsNil() {
+		panic("trying to call non-existent function: " + fn.RelString(nil))
+	}
+	args = append(args, llvm.Undef(b.i8ptrType))            // unused context parameter
+	args = append(args, llvm.ConstPointerNull(b.i8ptrType)) // coroutine handle
+	return b.createCall(llvmFn, args, name)
+}
+
+// createCall creates a call to the given function with the arguments possibly
+// expanded.
+func (b *builder) createCall(fn llvm.Value, args []llvm.Value, name string) llvm.Value {
 	expanded := make([]llvm.Value, 0, len(args))
 	for _, arg := range args {
-		fragments := c.expandFormalParam(arg)
+		fragments := b.expandFormalParam(arg)
 		expanded = append(expanded, fragments...)
 	}
-	return c.builder.CreateCall(fn, expanded, name)
+	return b.CreateCall(fn, expanded, name)
 }
 
 // Expand an argument type to a list that can be used in a function call
 // parameter list.
-func (c *Compiler) expandFormalParamType(t llvm.Type) []llvm.Type {
+func (c *compilerContext) expandFormalParamType(t llvm.Type, name string, goType types.Type) []paramInfo {
 	switch t.TypeKind() {
 	case llvm.StructTypeKind:
-		fields := c.flattenAggregateType(t)
-		if len(fields) <= MaxFieldsPerParam {
-			return fields
+		fieldInfos := c.flattenAggregateType(t, name, goType)
+		if len(fieldInfos) <= maxFieldsPerParam {
+			return fieldInfos
 		} else {
 			// failed to lower
-			return []llvm.Type{t}
 		}
-	default:
-		// TODO: split small arrays
-		return []llvm.Type{t}
+	}
+	// TODO: split small arrays
+	return []paramInfo{
+		{
+			llvmType: t,
+			name:     name,
+			flags:    getTypeFlags(goType),
+		},
 	}
 }
 
-// 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 {
+// expandFormalParamOffsets returns a list of offsets from the start of an
+// object of type t after it would have been split up by expandFormalParam. This
+// is useful for debug information, where it is necessary to know the offset
+// from the start of the combined object.
+func (b *builder) expandFormalParamOffsets(t llvm.Type) []uint64 {
 	switch t.TypeKind() {
 	case llvm.StructTypeKind:
-		fields := c.flattenAggregateTypeOffsets(t)
-		if len(fields) <= MaxFieldsPerParam {
+		fields := b.flattenAggregateTypeOffsets(t)
+		if len(fields) <= maxFieldsPerParam {
 			return fields
 		} else {
 			// failed to lower
@@ -74,14 +98,17 @@ func (c *Compiler) expandFormalParamOffsets(t llvm.Type) []uint64 {
 	}
 }
 
-// Equivalent of expandFormalParamType for parameter values.
-func (c *Compiler) expandFormalParam(v llvm.Value) []llvm.Value {
+// expandFormalParam splits a formal param value into pieces, so it can be
+// passed directly as part of a function call. For example, it splits up small
+// structs into individual fields. It is the equivalent of expandFormalParamType
+// for parameter values.
+func (b *builder) expandFormalParam(v llvm.Value) []llvm.Value {
 	switch v.Type().TypeKind() {
 	case llvm.StructTypeKind:
-		fieldTypes := c.flattenAggregateType(v.Type())
-		if len(fieldTypes) <= MaxFieldsPerParam {
-			fields := c.flattenAggregate(v)
-			if len(fields) != len(fieldTypes) {
+		fieldInfos := b.flattenAggregateType(v.Type(), "", nil)
+		if len(fieldInfos) <= maxFieldsPerParam {
+			fields := b.flattenAggregate(v)
+			if len(fields) != len(fieldInfos) {
 				panic("type and value param lowering don't match")
 			}
 			return fields
@@ -97,28 +124,108 @@ func (c *Compiler) expandFormalParam(v llvm.Value) []llvm.Value {
 
 // Try to flatten a struct type to a list of types. Returns a 1-element slice
 // with the passed in type if this is not possible.
-func (c *Compiler) flattenAggregateType(t llvm.Type) []llvm.Type {
+func (c *compilerContext) flattenAggregateType(t llvm.Type, name string, goType types.Type) []paramInfo {
+	typeFlags := getTypeFlags(goType)
 	switch t.TypeKind() {
 	case llvm.StructTypeKind:
-		fields := make([]llvm.Type, 0, t.StructElementTypesCount())
-		for _, subfield := range t.StructElementTypes() {
-			subfields := c.flattenAggregateType(subfield)
-			fields = append(fields, subfields...)
+		var paramInfos []paramInfo
+		for i, subfield := range t.StructElementTypes() {
+			if c.targetData.TypeAllocSize(subfield) == 0 {
+				continue
+			}
+			suffix := strconv.Itoa(i)
+			if goType != nil {
+				// Try to come up with a good suffix for this struct field,
+				// depending on which Go type it's based on.
+				switch goType := goType.Underlying().(type) {
+				case *types.Interface:
+					suffix = []string{"typecode", "value"}[i]
+				case *types.Slice:
+					suffix = []string{"data", "len", "cap"}[i]
+				case *types.Struct:
+					suffix = goType.Field(i).Name()
+				case *types.Basic:
+					switch goType.Kind() {
+					case types.Complex64, types.Complex128:
+						suffix = []string{"r", "i"}[i]
+					case types.String:
+						suffix = []string{"data", "len"}[i]
+					}
+				case *types.Signature:
+					suffix = []string{"context", "funcptr"}[i]
+				}
+			}
+			subInfos := c.flattenAggregateType(subfield, name+"."+suffix, extractSubfield(goType, i))
+			for i := range subInfos {
+				subInfos[i].flags |= typeFlags
+			}
+			paramInfos = append(paramInfos, subInfos...)
 		}
-		return fields
+		return paramInfos
 	default:
-		return []llvm.Type{t}
+		return []paramInfo{
+			{
+				llvmType: t,
+				name:     name,
+				flags:    typeFlags,
+			},
+		}
 	}
 }
 
-// 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 {
+// getTypeFlags returns the type flags for a given type. It will not recurse
+// into sub-types (such as in structs).
+func getTypeFlags(t types.Type) paramFlags {
+	if t == nil {
+		return 0
+	}
+	switch t.Underlying().(type) {
+	case *types.Pointer:
+		// Pointers in Go must either point to an object or be nil.
+		return paramIsDeferenceableOrNull
+	case *types.Chan, *types.Map:
+		// Channels and maps are implemented as pointers pointing to some
+		// object, and follow the same rules as *types.Pointer.
+		return paramIsDeferenceableOrNull
+	default:
+		return 0
+	}
+}
+
+// extractSubfield extracts a field from a struct, or returns null if this is
+// not a struct and thus no subfield can be obtained.
+func extractSubfield(t types.Type, field int) types.Type {
+	if t == nil {
+		return nil
+	}
+	switch t := t.Underlying().(type) {
+	case *types.Struct:
+		return t.Field(field).Type()
+	case *types.Interface, *types.Slice, *types.Basic, *types.Signature:
+		// These Go types are (sometimes) implemented as LLVM structs but can't
+		// really be split further up in Go (with the possible exception of
+		// complex numbers).
+		return nil
+	default:
+		// This should be unreachable.
+		panic("cannot split subfield: " + t.String())
+	}
+}
+
+// flattenAggregateTypeOffset returns the offsets from the start of an object of
+// type t if this object were flattened like in flattenAggregate. Used together
+// with flattenAggregate to know the start indices of each value in the
+// non-flattened object.
+//
+// Note: this is an implementation detail, use expandFormalParamOffsets instead.
+func (c *compilerContext) flattenAggregateTypeOffsets(t llvm.Type) []uint64 {
 	switch t.TypeKind() {
 	case llvm.StructTypeKind:
-		fields := make([]uint64, 0, t.StructElementTypesCount())
+		var fields []uint64
 		for fieldIndex, field := range t.StructElementTypes() {
+			if c.targetData.TypeAllocSize(field) == 0 {
+				continue
+			}
 			suboffsets := c.flattenAggregateTypeOffsets(field)
 			offset := c.targetData.ElementOffset(t, fieldIndex)
 			for i := range suboffsets {
@@ -132,15 +239,18 @@ func (c *Compiler) flattenAggregateTypeOffsets(t llvm.Type) []uint64 {
 	}
 }
 
-// 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 {
+// flattenAggregate breaks down a struct into its elementary values for argument
+// passing. It is the value equivalent of flattenAggregateType
+func (b *builder) flattenAggregate(v llvm.Value) []llvm.Value {
 	switch v.Type().TypeKind() {
 	case llvm.StructTypeKind:
-		fields := make([]llvm.Value, 0, v.Type().StructElementTypesCount())
-		for i := range v.Type().StructElementTypes() {
-			subfield := c.builder.CreateExtractValue(v, i, "")
-			subfields := c.flattenAggregate(subfield)
+		var fields []llvm.Value
+		for i, field := range v.Type().StructElementTypes() {
+			if b.targetData.TypeAllocSize(field) == 0 {
+				continue
+			}
+			subfield := b.CreateExtractValue(v, i, "")
+			subfields := b.flattenAggregate(subfield)
 			fields = append(fields, subfields...)
 		}
 		return fields
@@ -149,25 +259,32 @@ func (c *Compiler) flattenAggregate(v llvm.Value) []llvm.Value {
 	}
 }
 
-// Collapse a list of fields into its original value.
-func (c *Compiler) collapseFormalParam(t llvm.Type, fields []llvm.Value) llvm.Value {
-	param, remaining := c.collapseFormalParamInternal(t, fields)
+// collapseFormalParam combines an aggregate object back into the original
+// value. This is used to join multiple LLVM parameters into a single Go value
+// in the function entry block.
+func (b *builder) collapseFormalParam(t llvm.Type, fields []llvm.Value) llvm.Value {
+	param, remaining := b.collapseFormalParamInternal(t, fields)
 	if len(remaining) != 0 {
 		panic("failed to expand back all fields")
 	}
 	return param
 }
 
-// Returns (value, remainingFields). Used by collapseFormalParam.
-func (c *Compiler) collapseFormalParamInternal(t llvm.Type, fields []llvm.Value) (llvm.Value, []llvm.Value) {
+// collapseFormalParamInternal is an implementation detail of
+// collapseFormalParam: it works by recursing until there are no fields left.
+func (b *builder) collapseFormalParamInternal(t llvm.Type, fields []llvm.Value) (llvm.Value, []llvm.Value) {
 	switch t.TypeKind() {
 	case llvm.StructTypeKind:
-		if len(c.flattenAggregateType(t)) <= MaxFieldsPerParam {
+		flattened := b.flattenAggregateType(t, "", nil)
+		if len(flattened) <= maxFieldsPerParam {
 			value := llvm.ConstNull(t)
 			for i, subtyp := range t.StructElementTypes() {
-				structField, remaining := c.collapseFormalParamInternal(subtyp, fields)
+				if b.targetData.TypeAllocSize(subtyp) == 0 {
+					continue
+				}
+				structField, remaining := b.collapseFormalParamInternal(subtyp, fields)
 				fields = remaining
-				value = c.builder.CreateInsertValue(value, structField, i, "")
+				value = b.CreateInsertValue(value, structField, i, "")
 			}
 			return value, fields
 		} else {

compiler/channel.go

@@ -4,98 +4,95 @@ package compiler
 // or pseudo-operations that are lowered during goroutine lowering.
 
 import (
-	"fmt"
 	"go/types"
 
+	"github.com/tinygo-org/tinygo/compiler/llvmutil"
 	"golang.org/x/tools/go/ssa"
 	"tinygo.org/x/go-llvm"
 )
 
-// emitMakeChan returns a new channel value for the given channel type.
-func (c *Compiler) emitMakeChan(expr *ssa.MakeChan) (llvm.Value, error) {
-	chanType := c.getLLVMType(expr.Type())
-	size := c.targetData.TypeAllocSize(chanType.ElementType())
-	sizeValue := llvm.ConstInt(c.uintptrType, size, false)
-	ptr := c.createRuntimeCall("alloc", []llvm.Value{sizeValue}, "chan.alloc")
-	ptr = c.builder.CreateBitCast(ptr, chanType, "chan")
-	// Set the elementSize field
-	elementSizePtr := c.builder.CreateGEP(ptr, []llvm.Value{
-		llvm.ConstInt(c.ctx.Int32Type(), 0, false),
-		llvm.ConstInt(c.ctx.Int32Type(), 0, false),
-	}, "")
-	elementSize := c.targetData.TypeAllocSize(c.getLLVMType(expr.Type().(*types.Chan).Elem()))
-	if elementSize > 0xffff {
-		return ptr, c.makeError(expr.Pos(), fmt.Sprintf("element size is %d bytes, which is bigger than the maximum of %d bytes", elementSize, 0xffff))
+func (b *builder) createMakeChan(expr *ssa.MakeChan) llvm.Value {
+	elementSize := b.targetData.TypeAllocSize(b.getLLVMType(expr.Type().Underlying().(*types.Chan).Elem()))
+	elementSizeValue := llvm.ConstInt(b.uintptrType, elementSize, false)
+	bufSize := b.getValue(expr.Size)
+	b.createChanBoundsCheck(elementSize, bufSize, expr.Size.Type().Underlying().(*types.Basic), expr.Pos())
+	if bufSize.Type().IntTypeWidth() < b.uintptrType.IntTypeWidth() {
+		bufSize = b.CreateZExt(bufSize, b.uintptrType, "")
+	} else if bufSize.Type().IntTypeWidth() > b.uintptrType.IntTypeWidth() {
+		bufSize = b.CreateTrunc(bufSize, b.uintptrType, "")
 	}
-	elementSizeValue := llvm.ConstInt(c.ctx.Int16Type(), elementSize, false)
-	c.builder.CreateStore(elementSizeValue, elementSizePtr)
-	return ptr, nil
+	return b.createRuntimeCall("chanMake", []llvm.Value{elementSizeValue, bufSize}, "")
 }
 
-// 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) {
-	ch := c.getValue(frame, instr.Chan)
-	chanValue := c.getValue(frame, instr.X)
+// createChanSend emits a pseudo chan send operation. It is lowered to the
+// actual channel send operation during goroutine lowering.
+func (b *builder) createChanSend(instr *ssa.Send) {
+	ch := b.getValue(instr.Chan)
+	chanValue := b.getValue(instr.X)
 
 	// store value-to-send
-	valueType := c.getLLVMType(instr.X.Type())
-	valueAlloca, valueAllocaCast, valueAllocaSize := c.createTemporaryAlloca(valueType, "chan.value")
-	c.builder.CreateStore(chanValue, valueAlloca)
+	valueType := b.getLLVMType(instr.X.Type())
+	valueAlloca, valueAllocaCast, valueAllocaSize := b.createTemporaryAlloca(valueType, "chan.value")
+	b.CreateStore(chanValue, valueAlloca)
+
+	// Allocate blockedlist buffer.
+	channelBlockedList := b.mod.GetTypeByName("runtime.channelBlockedList")
+	channelBlockedListAlloca, channelBlockedListAllocaCast, channelBlockedListAllocaSize := b.createTemporaryAlloca(channelBlockedList, "chan.blockedList")
 
 	// Do the send.
-	coroutine := c.createRuntimeCall("getCoroutine", nil, "")
-	c.createRuntimeCall("chanSend", []llvm.Value{coroutine, ch, valueAllocaCast}, "")
+	b.createRuntimeCall("chanSend", []llvm.Value{ch, valueAllocaCast, channelBlockedListAlloca}, "")
 
-	// End the lifetime of the alloca.
+	// End the lifetime of the allocas.
 	// This also works around a bug in CoroSplit, at least in LLVM 8:
 	// https://bugs.llvm.org/show_bug.cgi?id=41742
-	c.emitLifetimeEnd(valueAllocaCast, valueAllocaSize)
+	b.emitLifetimeEnd(channelBlockedListAllocaCast, channelBlockedListAllocaSize)
+	b.emitLifetimeEnd(valueAllocaCast, valueAllocaSize)
 }
 
-// emitChanRecv emits a pseudo chan receive operation. It is lowered to the
+// createChanRecv 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 {
-	valueType := c.getLLVMType(unop.X.Type().(*types.Chan).Elem())
-	ch := c.getValue(frame, unop.X)
+func (b *builder) createChanRecv(unop *ssa.UnOp) llvm.Value {
+	valueType := b.getLLVMType(unop.X.Type().Underlying().(*types.Chan).Elem())
+	ch := b.getValue(unop.X)
 
 	// Allocate memory to receive into.
-	valueAlloca, valueAllocaCast, valueAllocaSize := c.createTemporaryAlloca(valueType, "chan.value")
+	valueAlloca, valueAllocaCast, valueAllocaSize := b.createTemporaryAlloca(valueType, "chan.value")
+
+	// Allocate blockedlist buffer.
+	channelBlockedList := b.mod.GetTypeByName("runtime.channelBlockedList")
+	channelBlockedListAlloca, channelBlockedListAllocaCast, channelBlockedListAllocaSize := b.createTemporaryAlloca(channelBlockedList, "chan.blockedList")
 
 	// Do the receive.
-	coroutine := c.createRuntimeCall("getCoroutine", nil, "")
-	c.createRuntimeCall("chanRecv", []llvm.Value{coroutine, ch, valueAllocaCast}, "")
-	received := c.builder.CreateLoad(valueAlloca, "chan.received")
-	c.emitLifetimeEnd(valueAllocaCast, valueAllocaSize)
+	commaOk := b.createRuntimeCall("chanRecv", []llvm.Value{ch, valueAllocaCast, channelBlockedListAlloca}, "")
+	received := b.CreateLoad(valueAlloca, "chan.received")
+	b.emitLifetimeEnd(channelBlockedListAllocaCast, channelBlockedListAllocaSize)
+	b.emitLifetimeEnd(valueAllocaCast, valueAllocaSize)
 
 	if unop.CommaOk {
-		commaOk := c.createRuntimeCall("getTaskStateData", []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, "")
+		tuple := llvm.Undef(b.ctx.StructType([]llvm.Type{valueType, b.ctx.Int1Type()}, false))
+		tuple = b.CreateInsertValue(tuple, received, 0, "")
+		tuple = b.CreateInsertValue(tuple, commaOk, 1, "")
 		return tuple
 	} else {
 		return received
 	}
 }
 
-// emitChanClose closes the given channel.
-func (c *Compiler) emitChanClose(frame *Frame, param ssa.Value) {
-	ch := c.getValue(frame, param)
-	c.createRuntimeCall("chanClose", []llvm.Value{ch}, "")
+// createChanClose closes the given channel.
+func (b *builder) createChanClose(ch llvm.Value) {
+	b.createRuntimeCall("chanClose", []llvm.Value{ch}, "")
 }
 
-// emitSelect emits all IR necessary for a select statements. That's a
+// createSelect emits all IR necessary for a select statements. That's a
 // non-trivial amount of code because select is very complex to implement.
-func (c *Compiler) emitSelect(frame *Frame, expr *ssa.Select) llvm.Value {
+func (b *builder) createSelect(expr *ssa.Select) llvm.Value {
 	if len(expr.States) == 0 {
 		// Shortcuts for some simple selects.
-		llvmType := c.getLLVMType(expr.Type())
+		llvmType := b.getLLVMType(expr.Type())
 		if expr.Blocking {
 			// Blocks forever:
 			//     select {}
-			c.createRuntimeCall("deadlock", nil, "")
+			b.createRuntimeCall("deadlock", nil, "")
 			return llvm.Undef(llvmType)
 		} else {
 			// No-op:
@@ -103,7 +100,7 @@ func (c *Compiler) emitSelect(frame *Frame, expr *ssa.Select) llvm.Value {
 			//     default:
 			//     }
 			retval := llvm.Undef(llvmType)
-			retval = c.builder.CreateInsertValue(retval, llvm.ConstInt(c.intType, 0xffffffffffffffff, true), 0, "")
+			retval = b.CreateInsertValue(retval, llvm.ConstInt(b.intType, 0xffffffffffffffff, true), 0, "")
 			return retval // {-1, false}
 		}
 	}
@@ -121,30 +118,30 @@ func (c *Compiler) emitSelect(frame *Frame, expr *ssa.Select) llvm.Value {
 	recvbufAlign := 0
 	hasReceives := false
 	var selectStates []llvm.Value
-	chanSelectStateType := c.getLLVMRuntimeType("chanSelectState")
+	chanSelectStateType := b.getLLVMRuntimeType("chanSelectState")
 	for _, state := range expr.States {
-		ch := c.getValue(frame, state.Chan)
+		ch := b.getValue(state.Chan)
 		selectState := llvm.ConstNull(chanSelectStateType)
-		selectState = c.builder.CreateInsertValue(selectState, ch, 0, "")
+		selectState = b.CreateInsertValue(selectState, ch, 0, "")
 		switch state.Dir {
 		case types.RecvOnly:
 			// Make sure the receive buffer is big enough and has the correct alignment.
-			llvmType := c.getLLVMType(state.Chan.Type().(*types.Chan).Elem())
-			if size := c.targetData.TypeAllocSize(llvmType); size > recvbufSize {
+			llvmType := b.getLLVMType(state.Chan.Type().Underlying().(*types.Chan).Elem())
+			if size := b.targetData.TypeAllocSize(llvmType); size > recvbufSize {
 				recvbufSize = size
 			}
-			if align := c.targetData.ABITypeAlignment(llvmType); align > recvbufAlign {
+			if align := b.targetData.ABITypeAlignment(llvmType); align > recvbufAlign {
 				recvbufAlign = align
 			}
 			hasReceives = true
 		case types.SendOnly:
 			// Store this value in an alloca and put a pointer to this alloca
 			// in the send state.
-			sendValue := c.getValue(frame, state.Send)
-			alloca := c.createEntryBlockAlloca(sendValue.Type(), "select.send.value")
-			c.builder.CreateStore(sendValue, alloca)
-			ptr := c.builder.CreateBitCast(alloca, c.i8ptrType, "")
-			selectState = c.builder.CreateInsertValue(selectState, ptr, 1, "")
+			sendValue := b.getValue(state.Send)
+			alloca := llvmutil.CreateEntryBlockAlloca(b.Builder, sendValue.Type(), "select.send.value")
+			b.CreateStore(sendValue, alloca)
+			ptr := b.CreateBitCast(alloca, b.i8ptrType, "")
+			selectState = b.CreateInsertValue(selectState, ptr, 1, "")
 		default:
 			panic("unreachable")
 		}
@@ -152,57 +149,74 @@ func (c *Compiler) emitSelect(frame *Frame, expr *ssa.Select) llvm.Value {
 	}
 
 	// Create a receive buffer, where the received value will be stored.
-	recvbuf := llvm.Undef(c.i8ptrType)
+	recvbuf := llvm.Undef(b.i8ptrType)
 	if hasReceives {
-		allocaType := llvm.ArrayType(c.ctx.Int8Type(), int(recvbufSize))
-		recvbufAlloca := c.builder.CreateAlloca(allocaType, "select.recvbuf.alloca")
+		allocaType := llvm.ArrayType(b.ctx.Int8Type(), int(recvbufSize))
+		recvbufAlloca, _, _ := b.createTemporaryAlloca(allocaType, "select.recvbuf.alloca")
 		recvbufAlloca.SetAlignment(recvbufAlign)
-		recvbuf = c.builder.CreateGEP(recvbufAlloca, []llvm.Value{
-			llvm.ConstInt(c.ctx.Int32Type(), 0, false),
-			llvm.ConstInt(c.ctx.Int32Type(), 0, false),
+		recvbuf = b.CreateGEP(recvbufAlloca, []llvm.Value{
+			llvm.ConstInt(b.ctx.Int32Type(), 0, false),
+			llvm.ConstInt(b.ctx.Int32Type(), 0, false),
 		}, "select.recvbuf")
 	}
 
 	// Create the states slice (allocated on the stack).
 	statesAllocaType := llvm.ArrayType(chanSelectStateType, len(selectStates))
-	statesAlloca := c.builder.CreateAlloca(statesAllocaType, "select.states.alloca")
+	statesAlloca, statesI8, statesSize := b.createTemporaryAlloca(statesAllocaType, "select.states.alloca")
 	for i, state := range selectStates {
 		// Set each slice element to the appropriate channel.
-		gep := c.builder.CreateGEP(statesAlloca, []llvm.Value{
-			llvm.ConstInt(c.ctx.Int32Type(), 0, false),
-			llvm.ConstInt(c.ctx.Int32Type(), uint64(i), false),
+		gep := b.CreateGEP(statesAlloca, []llvm.Value{
+			llvm.ConstInt(b.ctx.Int32Type(), 0, false),
+			llvm.ConstInt(b.ctx.Int32Type(), uint64(i), false),
 		}, "")
-		c.builder.CreateStore(state, gep)
+		b.CreateStore(state, gep)
 	}
-	statesPtr := c.builder.CreateGEP(statesAlloca, []llvm.Value{
-		llvm.ConstInt(c.ctx.Int32Type(), 0, false),
-		llvm.ConstInt(c.ctx.Int32Type(), 0, false),
+	statesPtr := b.CreateGEP(statesAlloca, []llvm.Value{
+		llvm.ConstInt(b.ctx.Int32Type(), 0, false),
+		llvm.ConstInt(b.ctx.Int32Type(), 0, false),
 	}, "select.states")
-	statesLen := llvm.ConstInt(c.uintptrType, uint64(len(selectStates)), false)
-
-	// Convert the 'blocking' flag on this select into a LLVM value.
-	blockingInt := uint64(0)
-	if expr.Blocking {
-		blockingInt = 1
-	}
-	blockingValue := llvm.ConstInt(c.ctx.Int1Type(), blockingInt, false)
+	statesLen := llvm.ConstInt(b.uintptrType, uint64(len(selectStates)), false)
 
 	// Do the select in the runtime.
-	results := c.createRuntimeCall("chanSelect", []llvm.Value{
-		recvbuf,
-		statesPtr, statesLen, statesLen, // []chanSelectState
-		blockingValue,
-	}, "")
+	var results llvm.Value
+	if expr.Blocking {
+		// Stack-allocate operation structures.
+		// If these were simply created as a slice, they would heap-allocate.
+		chBlockAllocaType := llvm.ArrayType(b.getLLVMRuntimeType("channelBlockedList"), len(selectStates))
+		chBlockAlloca, chBlockAllocaPtr, chBlockSize := b.createTemporaryAlloca(chBlockAllocaType, "select.block.alloca")
+		chBlockLen := llvm.ConstInt(b.uintptrType, uint64(len(selectStates)), false)
+		chBlockPtr := b.CreateGEP(chBlockAlloca, []llvm.Value{
+			llvm.ConstInt(b.ctx.Int32Type(), 0, false),
+			llvm.ConstInt(b.ctx.Int32Type(), 0, false),
+		}, "select.block")
+
+		results = b.createRuntimeCall("chanSelect", []llvm.Value{
+			recvbuf,
+			statesPtr, statesLen, statesLen, // []chanSelectState
+			chBlockPtr, chBlockLen, chBlockLen, // []channelBlockList
+		}, "select.result")
+
+		// Terminate the lifetime of the operation structures.
+		b.emitLifetimeEnd(chBlockAllocaPtr, chBlockSize)
+	} else {
+		results = b.createRuntimeCall("tryChanSelect", []llvm.Value{
+			recvbuf,
+			statesPtr, statesLen, statesLen, // []chanSelectState
+		}, "select.result")
+	}
+
+	// Terminate the lifetime of the states alloca.
+	b.emitLifetimeEnd(statesI8, statesSize)
 
 	// The result value does not include all the possible received values,
 	// because we can't load them in advance. Instead, the *ssa.Extract
 	// instruction will treat a *ssa.Select specially and load it there inline.
 	// Store the receive alloca in a sidetable until we hit this extract
 	// instruction.
-	if frame.selectRecvBuf == nil {
-		frame.selectRecvBuf = make(map[*ssa.Select]llvm.Value)
+	if b.selectRecvBuf == nil {
+		b.selectRecvBuf = make(map[*ssa.Select]llvm.Value)
 	}
-	frame.selectRecvBuf[expr] = recvbuf
+	b.selectRecvBuf[expr] = recvbuf
 
 	return results
 }
@@ -211,28 +225,28 @@ func (c *Compiler) emitSelect(frame *Frame, expr *ssa.Select) llvm.Value {
 // when extracting a value from a select statement (*ssa.Select). Because
 // *ssa.Select cannot load all values in advance, it does this later in the
 // *ssa.Extract expression.
-func (c *Compiler) getChanSelectResult(frame *Frame, expr *ssa.Extract) llvm.Value {
+func (b *builder) getChanSelectResult(expr *ssa.Extract) llvm.Value {
 	if expr.Index == 0 {
 		// index
-		value := c.getValue(frame, expr.Tuple)
-		index := c.builder.CreateExtractValue(value, expr.Index, "")
-		if index.Type().IntTypeWidth() < c.intType.IntTypeWidth() {
-			index = c.builder.CreateSExt(index, c.intType, "")
+		value := b.getValue(expr.Tuple)
+		index := b.CreateExtractValue(value, expr.Index, "")
+		if index.Type().IntTypeWidth() < b.intType.IntTypeWidth() {
+			index = b.CreateSExt(index, b.intType, "")
 		}
 		return index
 	} else if expr.Index == 1 {
 		// comma-ok
-		value := c.getValue(frame, expr.Tuple)
-		return c.builder.CreateExtractValue(value, expr.Index, "")
+		value := b.getValue(expr.Tuple)
+		return b.CreateExtractValue(value, expr.Index, "")
 	} else {
 		// Select statements are (index, ok, ...) where ... is a number of
 		// received values, depending on how many receive statements there
 		// are. They are all combined into one alloca (because only one
 		// receive can proceed at a time) so we'll get that alloca, bitcast
 		// it to the correct type, and dereference it.
-		recvbuf := frame.selectRecvBuf[expr.Tuple.(*ssa.Select)]
-		typ := llvm.PointerType(c.getLLVMType(expr.Type()), 0)
-		ptr := c.builder.CreateBitCast(recvbuf, typ, "")
-		return c.builder.CreateLoad(ptr, "")
+		recvbuf := b.selectRecvBuf[expr.Tuple.(*ssa.Select)]
+		typ := llvm.PointerType(b.getLLVMType(expr.Type()), 0)
+		ptr := b.CreateBitCast(recvbuf, typ, "")
+		return b.CreateLoad(ptr, "")
 	}
 }

compiler/check.go

@@ -1,128 +0,0 @@
-package compiler
-
-// This file implements a set of sanity checks for the IR that is generated.
-// It can catch some mistakes that LLVM's verifier cannot.
-
-import (
-	"errors"
-	"fmt"
-
-	"tinygo.org/x/go-llvm"
-)
-
-func (c *Compiler) checkType(t llvm.Type, checked map[llvm.Type]struct{}, specials map[llvm.TypeKind]llvm.Type) {
-	if t.IsNil() {
-		panic(t)
-	}
-
-	// prevent infinite recursion for self-referential types
-	if _, ok := checked[t]; ok {
-		return
-	}
-	checked[t] = struct{}{}
-
-	// check for any context mismatches
-	switch {
-	case t.Context() == c.ctx:
-		// this is correct
-	case t.Context() == llvm.GlobalContext():
-		// somewhere we accidentally used the global context instead of a real context
-		panic(fmt.Errorf("type %q uses global context", t.String()))
-	default:
-		// we used some other context by accident
-		panic(fmt.Errorf("type %q uses context %v instead of the main context %v", t.Context(), c.ctx))
-	}
-
-	// if this is a composite type, check the components of the type
-	switch t.TypeKind() {
-	case llvm.VoidTypeKind, llvm.LabelTypeKind, llvm.TokenTypeKind, llvm.MetadataTypeKind:
-		// there should only be one of any of these
-		if s, ok := specials[t.TypeKind()]; !ok {
-			specials[t.TypeKind()] = t
-		} else if s != t {
-			panic(fmt.Errorf("duplicate special type %q: %v and %v", t.TypeKind().String(), t, s))
-		}
-	case llvm.FloatTypeKind, llvm.DoubleTypeKind, llvm.X86_FP80TypeKind, llvm.FP128TypeKind, llvm.PPC_FP128TypeKind:
-		// floating point numbers are primitives - nothing to recurse
-	case llvm.IntegerTypeKind:
-		// integers are primitives - nothing to recurse
-	case llvm.FunctionTypeKind:
-		// check arguments and return(s)
-		for _, v := range t.ParamTypes() {
-			c.checkType(v, checked, specials)
-		}
-		c.checkType(t.ReturnType(), checked, specials)
-	case llvm.StructTypeKind:
-		// check all elements
-		for _, v := range t.StructElementTypes() {
-			c.checkType(v, checked, specials)
-		}
-	case llvm.ArrayTypeKind:
-		// check element type
-		c.checkType(t.ElementType(), checked, specials)
-	case llvm.PointerTypeKind:
-		// check underlying type
-		c.checkType(t.ElementType(), checked, specials)
-	case llvm.VectorTypeKind:
-		// check element type
-		c.checkType(t.ElementType(), checked, specials)
-	}
-}
-
-func (c *Compiler) checkValue(v llvm.Value, types map[llvm.Type]struct{}, specials map[llvm.TypeKind]llvm.Type) {
-	// check type
-	c.checkType(v.Type(), types, specials)
-}
-
-func (c *Compiler) checkInstruction(inst llvm.Value, types map[llvm.Type]struct{}, specials map[llvm.TypeKind]llvm.Type) {
-	// check value properties
-	c.checkValue(inst, types, specials)
-
-	// check operands
-	for i := 0; i < inst.OperandsCount(); i++ {
-		c.checkValue(inst.Operand(i), types, specials)
-	}
-}
-
-func (c *Compiler) checkBasicBlock(bb llvm.BasicBlock, types map[llvm.Type]struct{}, specials map[llvm.TypeKind]llvm.Type) {
-	// check basic block value and type
-	c.checkValue(bb.AsValue(), types, specials)
-
-	// check instructions
-	for inst := bb.FirstInstruction(); !inst.IsNil(); inst = llvm.NextInstruction(inst) {
-		c.checkInstruction(inst, types, specials)
-	}
-}
-
-func (c *Compiler) checkFunction(fn llvm.Value, types map[llvm.Type]struct{}, specials map[llvm.TypeKind]llvm.Type) {
-	// check function value and type
-	c.checkValue(fn, types, specials)
-
-	// check basic blocks
-	for bb := fn.FirstBasicBlock(); !bb.IsNil(); bb = llvm.NextBasicBlock(bb) {
-		c.checkBasicBlock(bb, types, specials)
-	}
-}
-
-func (c *Compiler) checkModule() {
-	// check for any context mismatches
-	switch {
-	case c.mod.Context() == c.ctx:
-		// this is correct
-	case c.mod.Context() == llvm.GlobalContext():
-		// somewhere we accidentally used the global context instead of a real context
-		panic(errors.New("module uses global context"))
-	default:
-		// we used some other context by accident
-		panic(fmt.Errorf("module uses context %v instead of the main context %v", c.mod.Context(), c.ctx))
-	}
-
-	types := map[llvm.Type]struct{}{}
-	specials := map[llvm.TypeKind]llvm.Type{}
-	for fn := c.mod.FirstFunction(); !fn.IsNil(); fn = llvm.NextFunction(fn) {
-		c.checkFunction(fn, types, specials)
-	}
-	for g := c.mod.FirstGlobal(); !g.IsNil(); g = llvm.NextGlobal(g) {
-		c.checkValue(g, types, specials)
-	}
-}

compiler/compiler_test.go

@@ -0,0 +1,186 @@
+package compiler
+
+import (
+	"flag"
+	"go/types"
+	"io/ioutil"
+	"strconv"
+	"strings"
+	"testing"
+
+	"github.com/tinygo-org/tinygo/compileopts"
+	"github.com/tinygo-org/tinygo/loader"
+	"tinygo.org/x/go-llvm"
+)
+
+// Pass -update to go test to update the output of the test files.
+var flagUpdate = flag.Bool("update", false, "update tests based on test output")
+
+// Basic tests for the compiler. Build some Go files and compare the output with
+// the expected LLVM IR for regression testing.
+func TestCompiler(t *testing.T) {
+	t.Parallel()
+
+	// Check LLVM version.
+	llvmMajor, err := strconv.Atoi(strings.SplitN(llvm.Version, ".", 2)[0])
+	if err != nil {
+		t.Fatal("could not parse LLVM version:", llvm.Version)
+	}
+	if llvmMajor < 11 {
+		// It is likely this version needs to be bumped in the future.
+		// The goal is to at least test the LLVM version that's used by default
+		// in TinyGo and (if possible without too many workarounds) also some
+		// previous versions.
+		t.Skip("compiler tests require LLVM 11 or above, got LLVM ", llvm.Version)
+	}
+
+	tests := []struct {
+		file   string
+		target string
+	}{
+		{"basic.go", ""},
+		{"pointer.go", ""},
+		{"slice.go", ""},
+		{"string.go", ""},
+		{"float.go", ""},
+		{"interface.go", ""},
+		{"func.go", ""},
+		{"pragma.go", ""},
+		{"goroutine.go", "wasm"},
+		{"goroutine.go", "cortex-m-qemu"},
+	}
+
+	for _, tc := range tests {
+		name := tc.file
+		targetString := "wasm"
+		if tc.target != "" {
+			targetString = tc.target
+			name = tc.file + "-" + tc.target
+		}
+
+		t.Run(name, func(t *testing.T) {
+			target, err := compileopts.LoadTarget(targetString)
+			if err != nil {
+				t.Fatal("failed to load target:", err)
+			}
+			config := &compileopts.Config{
+				Options: &compileopts.Options{},
+				Target:  target,
+			}
+			compilerConfig := &Config{
+				Triple:             config.Triple(),
+				GOOS:               config.GOOS(),
+				GOARCH:             config.GOARCH(),
+				CodeModel:          config.CodeModel(),
+				RelocationModel:    config.RelocationModel(),
+				Scheduler:          config.Scheduler(),
+				FuncImplementation: config.FuncImplementation(),
+				AutomaticStackSize: config.AutomaticStackSize(),
+			}
+			machine, err := NewTargetMachine(compilerConfig)
+			if err != nil {
+				t.Fatal("failed to create target machine:", err)
+			}
+
+			// Load entire program AST into memory.
+			lprogram, err := loader.Load(config, []string{"./testdata/" + tc.file}, config.ClangHeaders, types.Config{
+				Sizes: Sizes(machine),
+			})
+			if err != nil {
+				t.Fatal("failed to create target machine:", err)
+			}
+			err = lprogram.Parse()
+			if err != nil {
+				t.Fatalf("could not parse test case %s: %s", tc.file, err)
+			}
+
+			// Compile AST to IR.
+			program := lprogram.LoadSSA()
+			pkg := lprogram.MainPkg()
+			mod, errs := CompilePackage(tc.file, pkg, program.Package(pkg.Pkg), machine, compilerConfig, false)
+			if errs != nil {
+				for _, err := range errs {
+					t.Error(err)
+				}
+				return
+			}
+
+			err = llvm.VerifyModule(mod, llvm.PrintMessageAction)
+			if err != nil {
+				t.Error(err)
+			}
+
+			// Optimize IR a little.
+			funcPasses := llvm.NewFunctionPassManagerForModule(mod)
+			defer funcPasses.Dispose()
+			funcPasses.AddInstructionCombiningPass()
+			funcPasses.InitializeFunc()
+			for fn := mod.FirstFunction(); !fn.IsNil(); fn = llvm.NextFunction(fn) {
+				funcPasses.RunFunc(fn)
+			}
+			funcPasses.FinalizeFunc()
+
+			outFilePrefix := tc.file[:len(tc.file)-3]
+			if tc.target != "" {
+				outFilePrefix += "-" + tc.target
+			}
+			outPath := "./testdata/" + outFilePrefix + ".ll"
+
+			// Update test if needed. Do not check the result.
+			if *flagUpdate {
+				err := ioutil.WriteFile(outPath, []byte(mod.String()), 0666)
+				if err != nil {
+					t.Error("failed to write updated output file:", err)
+				}
+				return
+			}
+
+			expected, err := ioutil.ReadFile(outPath)
+			if err != nil {
+				t.Fatal("failed to read golden file:", err)
+			}
+
+			if !fuzzyEqualIR(mod.String(), string(expected)) {
+				t.Errorf("output does not match expected output:\n%s", mod.String())
+			}
+		})
+	}
+}
+
+// fuzzyEqualIR returns true if the two LLVM IR strings passed in are roughly
+// equal. That means, only relevant lines are compared (excluding comments
+// etc.).
+func fuzzyEqualIR(s1, s2 string) bool {
+	lines1 := filterIrrelevantIRLines(strings.Split(s1, "\n"))
+	lines2 := filterIrrelevantIRLines(strings.Split(s2, "\n"))
+	if len(lines1) != len(lines2) {
+		return false
+	}
+	for i, line1 := range lines1 {
+		line2 := lines2[i]
+		if line1 != line2 {
+			return false
+		}
+	}
+
+	return true
+}
+
+// filterIrrelevantIRLines removes lines from the input slice of strings that
+// are not relevant in comparing IR. For example, empty lines and comments are
+// stripped out.
+func filterIrrelevantIRLines(lines []string) []string {
+	var out []string
+	for _, line := range lines {
+		line = strings.Split(line, ";")[0]    // strip out comments/info
+		line = strings.TrimRight(line, "\r ") // drop '\r' on Windows and remove trailing spaces from comments
+		if line == "" {
+			continue
+		}
+		if strings.HasPrefix(line, "source_filename = ") {
+			continue
+		}
+		out = append(out, line)
+	}
+	return out
+}

compiler/defer.go

@@ -14,7 +14,9 @@ package compiler
 //     frames.
 
 import (
-	"github.com/tinygo-org/tinygo/ir"
+	"go/types"
+
+	"github.com/tinygo-org/tinygo/compiler/llvmutil"
 	"golang.org/x/tools/go/ssa"
 	"tinygo.org/x/go-llvm"
 )
@@ -22,65 +24,100 @@ import (
 // deferInitFunc sets up this function for future deferred calls. It must be
 // called from within the entry block when this function contains deferred
 // calls.
-func (c *Compiler) deferInitFunc(frame *Frame) {
+func (b *builder) deferInitFunc() {
 	// Some setup.
-	frame.deferFuncs = make(map[*ir.Function]int)
-	frame.deferInvokeFuncs = make(map[string]int)
-	frame.deferClosureFuncs = make(map[*ir.Function]int)
+	b.deferFuncs = make(map[*ssa.Function]int)
+	b.deferInvokeFuncs = make(map[string]int)
+	b.deferClosureFuncs = make(map[*ssa.Function]int)
+	b.deferExprFuncs = make(map[ssa.Value]int)
+	b.deferBuiltinFuncs = make(map[ssa.Value]deferBuiltin)
 
 	// Create defer list pointer.
-	deferType := llvm.PointerType(c.getLLVMRuntimeType("_defer"), 0)
-	frame.deferPtr = c.builder.CreateAlloca(deferType, "deferPtr")
-	c.builder.CreateStore(llvm.ConstPointerNull(deferType), frame.deferPtr)
+	deferType := llvm.PointerType(b.getLLVMRuntimeType("_defer"), 0)
+	b.deferPtr = b.CreateAlloca(deferType, "deferPtr")
+	b.CreateStore(llvm.ConstPointerNull(deferType), b.deferPtr)
 }
 
-// emitDefer emits a single defer instruction, to be run when this function
+// isInLoop checks if there is a path from a basic block to itself.
+func isInLoop(start *ssa.BasicBlock) bool {
+	// Use a breadth-first search to scan backwards through the block graph.
+	queue := []*ssa.BasicBlock{start}
+	checked := map[*ssa.BasicBlock]struct{}{}
+
+	for len(queue) > 0 {
+		// pop a block off of the queue
+		block := queue[len(queue)-1]
+		queue = queue[:len(queue)-1]
+
+		// Search through predecessors.
+		// Searching backwards means that this is pretty fast when the block is close to the start of the function.
+		// Defers are often placed near the start of the function.
+		for _, pred := range block.Preds {
+			if pred == start {
+				// cycle found
+				return true
+			}
+
+			if _, ok := checked[pred]; ok {
+				// block already checked
+				continue
+			}
+
+			// add to queue and checked map
+			queue = append(queue, pred)
+			checked[pred] = struct{}{}
+		}
+	}
+
+	return false
+}
+
+// createDefer emits a single defer instruction, to be run when this function
 // returns.
-func (c *Compiler) emitDefer(frame *Frame, instr *ssa.Defer) {
+func (b *builder) createDefer(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")
+	next := b.CreateLoad(b.deferPtr, "defer.next")
 
 	var values []llvm.Value
-	valueTypes := []llvm.Type{c.uintptrType, next.Type()}
+	valueTypes := []llvm.Type{b.uintptrType, next.Type()}
 	if instr.Call.IsInvoke() {
 		// Method call on an interface.
 
 		// Get callback type number.
 		methodName := instr.Call.Method.FullName()
-		if _, ok := frame.deferInvokeFuncs[methodName]; !ok {
-			frame.deferInvokeFuncs[methodName] = len(frame.allDeferFuncs)
-			frame.allDeferFuncs = append(frame.allDeferFuncs, &instr.Call)
+		if _, ok := b.deferInvokeFuncs[methodName]; !ok {
+			b.deferInvokeFuncs[methodName] = len(b.allDeferFuncs)
+			b.allDeferFuncs = append(b.allDeferFuncs, &instr.Call)
 		}
-		callback := llvm.ConstInt(c.uintptrType, uint64(frame.deferInvokeFuncs[methodName]), false)
+		callback := llvm.ConstInt(b.uintptrType, uint64(b.deferInvokeFuncs[methodName]), false)
 
 		// Collect all values to be put in the struct (starting with
 		// runtime._defer fields, followed by the call parameters).
-		itf := c.getValue(frame, instr.Call.Value) // interface
-		receiverValue := c.builder.CreateExtractValue(itf, 1, "invoke.func.receiver")
-		values = []llvm.Value{callback, next, receiverValue}
-		valueTypes = append(valueTypes, c.i8ptrType)
+		itf := b.getValue(instr.Call.Value) // interface
+		typecode := b.CreateExtractValue(itf, 0, "invoke.func.typecode")
+		receiverValue := b.CreateExtractValue(itf, 1, "invoke.func.receiver")
+		values = []llvm.Value{callback, next, typecode, receiverValue}
+		valueTypes = append(valueTypes, b.uintptrType, b.i8ptrType)
 		for _, arg := range instr.Call.Args {
-			val := c.getValue(frame, arg)
+			val := b.getValue(arg)
 			values = append(values, val)
 			valueTypes = append(valueTypes, val.Type())
 		}
 
 	} else if callee, ok := instr.Call.Value.(*ssa.Function); ok {
 		// Regular function call.
-		fn := c.ir.GetFunction(callee)
-
-		if _, ok := frame.deferFuncs[fn]; !ok {
-			frame.deferFuncs[fn] = len(frame.allDeferFuncs)
-			frame.allDeferFuncs = append(frame.allDeferFuncs, fn)
+		if _, ok := b.deferFuncs[callee]; !ok {
+			b.deferFuncs[callee] = len(b.allDeferFuncs)
+			b.allDeferFuncs = append(b.allDeferFuncs, callee)
 		}
-		callback := llvm.ConstInt(c.uintptrType, uint64(frame.deferFuncs[fn]), false)
+		callback := llvm.ConstInt(b.uintptrType, uint64(b.deferFuncs[callee]), false)
 
 		// Collect all values to be put in the struct (starting with
 		// runtime._defer fields).
 		values = []llvm.Value{callback, next}
 		for _, param := range instr.Call.Args {
-			llvmParam := c.getValue(frame, param)
+			llvmParam := b.getValue(param)
 			values = append(values, llvmParam)
 			valueTypes = append(valueTypes, llvmParam.Type())
 		}
@@ -92,55 +129,109 @@ func (c *Compiler) emitDefer(frame *Frame, instr *ssa.Defer) {
 		// pointer.
 		// TODO: ignore this closure entirely and put pointers to the free
 		// variables directly in the defer struct, avoiding a memory allocation.
-		closure := c.getValue(frame, instr.Call.Value)
-		context := c.builder.CreateExtractValue(closure, 0, "")
+		closure := b.getValue(instr.Call.Value)
+		context := b.CreateExtractValue(closure, 0, "")
 
 		// Get the callback number.
-		fn := c.ir.GetFunction(makeClosure.Fn.(*ssa.Function))
-		if _, ok := frame.deferClosureFuncs[fn]; !ok {
-			frame.deferClosureFuncs[fn] = len(frame.allDeferFuncs)
-			frame.allDeferFuncs = append(frame.allDeferFuncs, makeClosure)
+		fn := makeClosure.Fn.(*ssa.Function)
+		if _, ok := b.deferClosureFuncs[fn]; !ok {
+			b.deferClosureFuncs[fn] = len(b.allDeferFuncs)
+			b.allDeferFuncs = append(b.allDeferFuncs, makeClosure)
 		}
-		callback := llvm.ConstInt(c.uintptrType, uint64(frame.deferClosureFuncs[fn]), false)
+		callback := llvm.ConstInt(b.uintptrType, uint64(b.deferClosureFuncs[fn]), false)
 
 		// Collect all values to be put in the struct (starting with
 		// runtime._defer fields, followed by all parameters including the
 		// context pointer).
 		values = []llvm.Value{callback, next}
 		for _, param := range instr.Call.Args {
-			llvmParam := c.getValue(frame, param)
+			llvmParam := b.getValue(param)
 			values = append(values, llvmParam)
 			valueTypes = append(valueTypes, llvmParam.Type())
 		}
 		values = append(values, context)
 		valueTypes = append(valueTypes, context.Type())
 
+	} else if builtin, ok := instr.Call.Value.(*ssa.Builtin); ok {
+		var argTypes []types.Type
+		var argValues []llvm.Value
+		for _, arg := range instr.Call.Args {
+			argTypes = append(argTypes, arg.Type())
+			argValues = append(argValues, b.getValue(arg))
+		}
+
+		if _, ok := b.deferBuiltinFuncs[instr.Call.Value]; !ok {
+			b.deferBuiltinFuncs[instr.Call.Value] = deferBuiltin{
+				callName: builtin.Name(),
+				pos:      builtin.Pos(),
+				argTypes: argTypes,
+				callback: len(b.allDeferFuncs),
+			}
+			b.allDeferFuncs = append(b.allDeferFuncs, instr.Call.Value)
+		}
+		callback := llvm.ConstInt(b.uintptrType, uint64(b.deferBuiltinFuncs[instr.Call.Value].callback), false)
+
+		// Collect all values to be put in the struct (starting with
+		// runtime._defer fields).
+		values = []llvm.Value{callback, next}
+		for _, param := range argValues {
+			values = append(values, param)
+			valueTypes = append(valueTypes, param.Type())
+		}
+
 	} else {
-		c.addError(instr.Pos(), "todo: defer on uncommon function call type")
-		return
+		funcValue := b.getValue(instr.Call.Value)
+
+		if _, ok := b.deferExprFuncs[instr.Call.Value]; !ok {
+			b.deferExprFuncs[instr.Call.Value] = len(b.allDeferFuncs)
+			b.allDeferFuncs = append(b.allDeferFuncs, &instr.Call)
+		}
+
+		callback := llvm.ConstInt(b.uintptrType, uint64(b.deferExprFuncs[instr.Call.Value]), false)
+
+		// Collect all values to be put in the struct (starting with
+		// runtime._defer fields, followed by all parameters including the
+		// context pointer).
+		values = []llvm.Value{callback, next, funcValue}
+		valueTypes = append(valueTypes, funcValue.Type())
+		for _, param := range instr.Call.Args {
+			llvmParam := b.getValue(param)
+			values = append(values, llvmParam)
+			valueTypes = append(valueTypes, llvmParam.Type())
+		}
 	}
 
 	// Make a struct out of the collected values to put in the defer frame.
-	deferFrameType := c.ctx.StructType(valueTypes, false)
+	deferFrameType := b.ctx.StructType(valueTypes, false)
 	deferFrame := llvm.ConstNull(deferFrameType)
 	for i, value := range values {
-		deferFrame = c.builder.CreateInsertValue(deferFrame, value, i, "")
+		deferFrame = b.CreateInsertValue(deferFrame, value, i, "")
 	}
 
-	// Put this struct in an alloca.
-	alloca := c.builder.CreateAlloca(deferFrameType, "defer.alloca")
-	c.builder.CreateStore(deferFrame, alloca)
-	if c.needsStackObjects() {
-		c.trackPointer(alloca)
+	// Put this struct in an allocation.
+	var alloca llvm.Value
+	if !isInLoop(instr.Block()) {
+		// This can safely use a stack allocation.
+		alloca = llvmutil.CreateEntryBlockAlloca(b.Builder, deferFrameType, "defer.alloca")
+	} else {
+		// This may be hit a variable number of times, so use a heap allocation.
+		size := b.targetData.TypeAllocSize(deferFrameType)
+		sizeValue := llvm.ConstInt(b.uintptrType, size, false)
+		allocCall := b.createRuntimeCall("alloc", []llvm.Value{sizeValue}, "defer.alloc.call")
+		alloca = b.CreateBitCast(allocCall, llvm.PointerType(deferFrameType, 0), "defer.alloc")
 	}
+	if b.NeedsStackObjects {
+		b.trackPointer(alloca)
+	}
+	b.CreateStore(deferFrame, alloca)
 
 	// 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)
+	allocaCast := b.CreateBitCast(alloca, next.Type(), "defer.alloca.cast")
+	b.CreateStore(allocaCast, b.deferPtr)
 }
 
-// emitRunDefers emits code to run all deferred functions.
-func (c *Compiler) emitRunDefers(frame *Frame) {
+// createRunDefers emits code to run all deferred functions.
+func (b *builder) createRunDefers() {
 	// Add a loop like the following:
 	//     for stack != nil {
 	//         _stack := stack
@@ -157,151 +248,208 @@ func (c *Compiler) emitRunDefers(frame *Frame) {
 	//     }
 
 	// Create loop.
-	loophead := c.ctx.AddBasicBlock(frame.fn.LLVMFn, "rundefers.loophead")
-	loop := c.ctx.AddBasicBlock(frame.fn.LLVMFn, "rundefers.loop")
-	unreachable := c.ctx.AddBasicBlock(frame.fn.LLVMFn, "rundefers.default")
-	end := c.ctx.AddBasicBlock(frame.fn.LLVMFn, "rundefers.end")
-	c.builder.CreateBr(loophead)
+	loophead := b.ctx.AddBasicBlock(b.llvmFn, "rundefers.loophead")
+	loop := b.ctx.AddBasicBlock(b.llvmFn, "rundefers.loop")
+	unreachable := b.ctx.AddBasicBlock(b.llvmFn, "rundefers.default")
+	end := b.ctx.AddBasicBlock(b.llvmFn, "rundefers.end")
+	b.CreateBr(loophead)
 
 	// Create loop head:
 	//     for stack != nil {
-	c.builder.SetInsertPointAtEnd(loophead)
-	deferData := c.builder.CreateLoad(frame.deferPtr, "")
-	stackIsNil := c.builder.CreateICmp(llvm.IntEQ, deferData, llvm.ConstPointerNull(deferData.Type()), "stackIsNil")
-	c.builder.CreateCondBr(stackIsNil, end, loop)
+	b.SetInsertPointAtEnd(loophead)
+	deferData := b.CreateLoad(b.deferPtr, "")
+	stackIsNil := b.CreateICmp(llvm.IntEQ, deferData, llvm.ConstPointerNull(deferData.Type()), "stackIsNil")
+	b.CreateCondBr(stackIsNil, end, loop)
 
 	// Create loop body:
 	//     _stack := stack
 	//     stack = stack.next
 	//     switch stack.callback {
-	c.builder.SetInsertPointAtEnd(loop)
-	nextStackGEP := c.builder.CreateInBoundsGEP(deferData, []llvm.Value{
-		llvm.ConstInt(c.ctx.Int32Type(), 0, false),
-		llvm.ConstInt(c.ctx.Int32Type(), 1, false), // .next field
+	b.SetInsertPointAtEnd(loop)
+	nextStackGEP := b.CreateInBoundsGEP(deferData, []llvm.Value{
+		llvm.ConstInt(b.ctx.Int32Type(), 0, false),
+		llvm.ConstInt(b.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.CreateInBoundsGEP(deferData, []llvm.Value{
-		llvm.ConstInt(c.ctx.Int32Type(), 0, false),
-		llvm.ConstInt(c.ctx.Int32Type(), 0, false), // .callback field
+	nextStack := b.CreateLoad(nextStackGEP, "stack.next")
+	b.CreateStore(nextStack, b.deferPtr)
+	gep := b.CreateInBoundsGEP(deferData, []llvm.Value{
+		llvm.ConstInt(b.ctx.Int32Type(), 0, false),
+		llvm.ConstInt(b.ctx.Int32Type(), 0, false), // .callback field
 	}, "callback.gep")
-	callback := c.builder.CreateLoad(gep, "callback")
-	sw := c.builder.CreateSwitch(callback, unreachable, len(frame.allDeferFuncs))
+	callback := b.CreateLoad(gep, "callback")
+	sw := b.CreateSwitch(callback, unreachable, len(b.allDeferFuncs))
 
-	for i, callback := range frame.allDeferFuncs {
+	for i, callback := range b.allDeferFuncs {
 		// Create switch case, for example:
 		//     case 0:
 		//         // run first deferred call
-		block := c.ctx.AddBasicBlock(frame.fn.LLVMFn, "rundefers.callback")
-		sw.AddCase(llvm.ConstInt(c.uintptrType, uint64(i), false), block)
-		c.builder.SetInsertPointAtEnd(block)
+		block := b.ctx.AddBasicBlock(b.llvmFn, "rundefers.callback")
+		sw.AddCase(llvm.ConstInt(b.uintptrType, uint64(i), false), block)
+		b.SetInsertPointAtEnd(block)
 		switch callback := callback.(type) {
 		case *ssa.CallCommon:
-			// Call on an interface value.
-			if !callback.IsInvoke() {
-				panic("expected an invoke call, not a direct call")
-			}
+			// Call on an value or interface value.
 
 			// Get the real defer struct type and cast to it.
-			valueTypes := []llvm.Type{c.uintptrType, llvm.PointerType(c.getLLVMRuntimeType("_defer"), 0), c.i8ptrType}
-			for _, arg := range callback.Args {
-				valueTypes = append(valueTypes, c.getLLVMType(arg.Type()))
+			valueTypes := []llvm.Type{b.uintptrType, llvm.PointerType(b.getLLVMRuntimeType("_defer"), 0)}
+
+			if !callback.IsInvoke() {
+				//Expect funcValue to be passed through the defer frame.
+				valueTypes = append(valueTypes, b.getFuncType(callback.Signature()))
+			} else {
+				//Expect typecode
+				valueTypes = append(valueTypes, b.uintptrType, b.i8ptrType)
 			}
-			deferFrameType := c.ctx.StructType(valueTypes, false)
-			deferFramePtr := c.builder.CreateBitCast(deferData, llvm.PointerType(deferFrameType, 0), "deferFrame")
+
+			for _, arg := range callback.Args {
+				valueTypes = append(valueTypes, b.getLLVMType(arg.Type()))
+			}
+
+			deferFrameType := b.ctx.StructType(valueTypes, false)
+			deferFramePtr := b.CreateBitCast(deferData, llvm.PointerType(deferFrameType, 0), "deferFrame")
 
 			// Extract the params from the struct (including receiver).
 			forwardParams := []llvm.Value{}
-			zero := llvm.ConstInt(c.ctx.Int32Type(), 0, false)
+			zero := llvm.ConstInt(b.ctx.Int32Type(), 0, false)
 			for i := 2; i < len(valueTypes); i++ {
-				gep := c.builder.CreateInBoundsGEP(deferFramePtr, []llvm.Value{zero, llvm.ConstInt(c.ctx.Int32Type(), uint64(i), false)}, "gep")
-				forwardParam := c.builder.CreateLoad(gep, "param")
+				gep := b.CreateInBoundsGEP(deferFramePtr, []llvm.Value{zero, llvm.ConstInt(b.ctx.Int32Type(), uint64(i), false)}, "gep")
+				forwardParam := b.CreateLoad(gep, "param")
 				forwardParams = append(forwardParams, forwardParam)
 			}
 
-			// Add the context parameter. An interface call cannot also be a
-			// closure but we have to supply the parameter anyway for platforms
-			// with a strict calling convention.
-			forwardParams = append(forwardParams, llvm.Undef(c.i8ptrType))
+			var fnPtr llvm.Value
+
+			if !callback.IsInvoke() {
+				// Isolate the func value.
+				funcValue := forwardParams[0]
+				forwardParams = forwardParams[1:]
+
+				//Get function pointer and context
+				fp, context := b.decodeFuncValue(funcValue, callback.Signature())
+				fnPtr = fp
+
+				//Pass context
+				forwardParams = append(forwardParams, context)
+			} else {
+				// Isolate the typecode.
+				typecode := forwardParams[0]
+				forwardParams = forwardParams[1:]
+				fnPtr = b.getInvokePtr(callback, typecode)
+
+				// Add the context parameter. An interface call cannot also be a
+				// closure but we have to supply the parameter anyway for platforms
+				// with a strict calling convention.
+				forwardParams = append(forwardParams, llvm.Undef(b.i8ptrType))
+			}
 
 			// Parent coroutine handle.
-			forwardParams = append(forwardParams, llvm.Undef(c.i8ptrType))
+			forwardParams = append(forwardParams, llvm.Undef(b.i8ptrType))
 
-			fnPtr, _ := c.getInvokeCall(frame, callback)
-			c.createCall(fnPtr, forwardParams, "")
+			b.createCall(fnPtr, forwardParams, "")
 
-		case *ir.Function:
+		case *ssa.Function:
 			// Direct call.
 
 			// Get the real defer struct type and cast to it.
-			valueTypes := []llvm.Type{c.uintptrType, llvm.PointerType(c.getLLVMRuntimeType("_defer"), 0)}
-			for _, param := range callback.Params {
-				valueTypes = append(valueTypes, c.getLLVMType(param.Type()))
+			valueTypes := []llvm.Type{b.uintptrType, llvm.PointerType(b.getLLVMRuntimeType("_defer"), 0)}
+			for _, param := range getParams(callback.Signature) {
+				valueTypes = append(valueTypes, b.getLLVMType(param.Type()))
 			}
-			deferFrameType := c.ctx.StructType(valueTypes, false)
-			deferFramePtr := c.builder.CreateBitCast(deferData, llvm.PointerType(deferFrameType, 0), "deferFrame")
+			deferFrameType := b.ctx.StructType(valueTypes, false)
+			deferFramePtr := b.CreateBitCast(deferData, llvm.PointerType(deferFrameType, 0), "deferFrame")
 
 			// Extract the params from the struct.
 			forwardParams := []llvm.Value{}
-			zero := llvm.ConstInt(c.ctx.Int32Type(), 0, false)
-			for i := range callback.Params {
-				gep := c.builder.CreateInBoundsGEP(deferFramePtr, []llvm.Value{zero, llvm.ConstInt(c.ctx.Int32Type(), uint64(i+2), false)}, "gep")
-				forwardParam := c.builder.CreateLoad(gep, "param")
+			zero := llvm.ConstInt(b.ctx.Int32Type(), 0, false)
+			for i := range getParams(callback.Signature) {
+				gep := b.CreateInBoundsGEP(deferFramePtr, []llvm.Value{zero, llvm.ConstInt(b.ctx.Int32Type(), uint64(i+2), false)}, "gep")
+				forwardParam := b.CreateLoad(gep, "param")
 				forwardParams = append(forwardParams, forwardParam)
 			}
 
-			// Add the context parameter. We know it is ignored by the receiving
-			// function, but we have to pass one anyway.
-			forwardParams = append(forwardParams, llvm.Undef(c.i8ptrType))
+			// Plain TinyGo functions add some extra parameters to implement async functionality and function recievers.
+			// These parameters should not be supplied when calling into an external C/ASM function.
+			if !b.getFunctionInfo(callback).exported {
+				// Add the context parameter. We know it is ignored by the receiving
+				// function, but we have to pass one anyway.
+				forwardParams = append(forwardParams, llvm.Undef(b.i8ptrType))
 
-			// Parent coroutine handle.
-			forwardParams = append(forwardParams, llvm.Undef(c.i8ptrType))
+				// Parent coroutine handle.
+				forwardParams = append(forwardParams, llvm.Undef(b.i8ptrType))
+			}
 
 			// Call real function.
-			c.createCall(callback.LLVMFn, forwardParams, "")
+			b.createCall(b.getFunction(callback), forwardParams, "")
 
 		case *ssa.MakeClosure:
 			// Get the real defer struct type and cast to it.
-			fn := c.ir.GetFunction(callback.Fn.(*ssa.Function))
-			valueTypes := []llvm.Type{c.uintptrType, llvm.PointerType(c.getLLVMRuntimeType("_defer"), 0)}
+			fn := callback.Fn.(*ssa.Function)
+			valueTypes := []llvm.Type{b.uintptrType, llvm.PointerType(b.getLLVMRuntimeType("_defer"), 0)}
 			params := fn.Signature.Params()
 			for i := 0; i < params.Len(); i++ {
-				valueTypes = append(valueTypes, c.getLLVMType(params.At(i).Type()))
+				valueTypes = append(valueTypes, b.getLLVMType(params.At(i).Type()))
 			}
-			valueTypes = append(valueTypes, c.i8ptrType) // closure
-			deferFrameType := c.ctx.StructType(valueTypes, false)
-			deferFramePtr := c.builder.CreateBitCast(deferData, llvm.PointerType(deferFrameType, 0), "deferFrame")
+			valueTypes = append(valueTypes, b.i8ptrType) // closure
+			deferFrameType := b.ctx.StructType(valueTypes, false)
+			deferFramePtr := b.CreateBitCast(deferData, llvm.PointerType(deferFrameType, 0), "deferFrame")
 
 			// Extract the params from the struct.
 			forwardParams := []llvm.Value{}
-			zero := llvm.ConstInt(c.ctx.Int32Type(), 0, false)
+			zero := llvm.ConstInt(b.ctx.Int32Type(), 0, false)
 			for i := 2; i < len(valueTypes); i++ {
-				gep := c.builder.CreateInBoundsGEP(deferFramePtr, []llvm.Value{zero, llvm.ConstInt(c.ctx.Int32Type(), uint64(i), false)}, "")
-				forwardParam := c.builder.CreateLoad(gep, "param")
+				gep := b.CreateInBoundsGEP(deferFramePtr, []llvm.Value{zero, llvm.ConstInt(b.ctx.Int32Type(), uint64(i), false)}, "")
+				forwardParam := b.CreateLoad(gep, "param")
 				forwardParams = append(forwardParams, forwardParam)
 			}
 
 			// Parent coroutine handle.
-			forwardParams = append(forwardParams, llvm.Undef(c.i8ptrType))
+			forwardParams = append(forwardParams, llvm.Undef(b.i8ptrType))
 
 			// Call deferred function.
-			c.createCall(fn.LLVMFn, forwardParams, "")
+			b.createCall(b.getFunction(fn), forwardParams, "")
+		case *ssa.Builtin:
+			db := b.deferBuiltinFuncs[callback]
 
+			//Get parameter types
+			valueTypes := []llvm.Type{b.uintptrType, llvm.PointerType(b.getLLVMRuntimeType("_defer"), 0)}
+
+			//Get signature from call results
+			params := callback.Type().Underlying().(*types.Signature).Params()
+			for i := 0; i < params.Len(); i++ {
+				valueTypes = append(valueTypes, b.getLLVMType(params.At(i).Type()))
+			}
+
+			deferFrameType := b.ctx.StructType(valueTypes, false)
+			deferFramePtr := b.CreateBitCast(deferData, llvm.PointerType(deferFrameType, 0), "deferFrame")
+
+			// Extract the params from the struct.
+			var argValues []llvm.Value
+			zero := llvm.ConstInt(b.ctx.Int32Type(), 0, false)
+			for i := 0; i < params.Len(); i++ {
+				gep := b.CreateInBoundsGEP(deferFramePtr, []llvm.Value{zero, llvm.ConstInt(b.ctx.Int32Type(), uint64(i+2), false)}, "gep")
+				forwardParam := b.CreateLoad(gep, "param")
+				argValues = append(argValues, forwardParam)
+			}
+
+			_, err := b.createBuiltin(db.argTypes, argValues, db.callName, db.pos)
+			if err != nil {
+				b.diagnostics = append(b.diagnostics, err)
+			}
 		default:
 			panic("unknown deferred function type")
 		}
 
 		// Branch back to the start of the loop.
-		c.builder.CreateBr(loophead)
+		b.CreateBr(loophead)
 	}
 
 	// Create default unreachable block:
 	//     default:
 	//         unreachable
 	//     }
-	c.builder.SetInsertPointAtEnd(unreachable)
-	c.builder.CreateUnreachable()
+	b.SetInsertPointAtEnd(unreachable)
+	b.CreateUnreachable()
 
 	// End of loop.
-	c.builder.SetInsertPointAtEnd(end)
+	b.SetInsertPointAtEnd(end)
 }

compiler/embed.go

@@ -0,0 +1,70 @@
+package compiler
+
+import (
+	"fmt"
+	"strconv"
+	"strings"
+	"unicode"
+	"unicode/utf8"
+)
+
+// parseGoEmbed parses the text following "//go:embed" to extract the glob patterns.
+// It accepts unquoted space-separated patterns as well as double-quoted and back-quoted Go strings.
+// Copied from https://github.com/golang/go/blob/219fe9d547d09d3de1b024c6c8b8314dd0bf12e4/src/cmd/compile/internal/noder/noder.go#L1717-L1776
+func parseGoEmbed(args string) ([]string, error) {
+	var list []string
+	for args = strings.TrimSpace(args); args != ""; args = strings.TrimSpace(args) {
+		var path string
+	Switch:
+		switch args[0] {
+		default:
+			i := len(args)
+			for j, c := range args {
+				if unicode.IsSpace(c) {
+					i = j
+					break
+				}
+			}
+			path = args[:i]
+			args = args[i:]
+
+		case '`':
+			i := strings.Index(args[1:], "`")
+			if i < 0 {
+				return nil, fmt.Errorf("invalid quoted string in //go:embed: %s", args)
+			}
+			path = args[1 : 1+i]
+			args = args[1+i+1:]
+
+		case '"':
+			i := 1
+			for ; i < len(args); i++ {
+				if args[i] == '\\' {
+					i++
+					continue
+				}
+				if args[i] == '"' {
+					q, err := strconv.Unquote(args[:i+1])
+					if err != nil {
+						return nil, fmt.Errorf("invalid quoted string in //go:embed: %s", args[:i+1])
+					}
+					path = q
+					args = args[i+1:]
+					break Switch
+				}
+			}
+			if i >= len(args) {
+				return nil, fmt.Errorf("invalid quoted string in //go:embed: %s", args)
+			}
+		}
+
+		if args != "" {
+			r, _ := utf8.DecodeRuneInString(args)
+			if !unicode.IsSpace(r) {
+				return nil, fmt.Errorf("invalid quoted string in //go:embed: %s", args)
+			}
+		}
+		list = append(list, path)
+	}
+	return list, nil
+}

compiler/errors.go

@@ -1,18 +1,64 @@
 package compiler
 
+// This file contains some utility functions related to error handling.
+
 import (
+	"go/scanner"
 	"go/token"
 	"go/types"
+	"path/filepath"
+
+	"tinygo.org/x/go-llvm"
 )
 
-func (c *Compiler) makeError(pos token.Pos, msg string) types.Error {
+// makeError makes it easy to create an error from a token.Pos with a message.
+func (c *compilerContext) makeError(pos token.Pos, msg string) types.Error {
 	return types.Error{
-		Fset: c.ir.Program.Fset,
+		Fset: c.program.Fset,
 		Pos:  pos,
 		Msg:  msg,
 	}
 }
 
-func (c *Compiler) addError(pos token.Pos, msg string) {
+func (c *compilerContext) addError(pos token.Pos, msg string) {
 	c.diagnostics = append(c.diagnostics, c.makeError(pos, msg))
 }
+
+// errorAt returns an error value at the location of the instruction.
+// The location information may not be complete as it depends on debug
+// information in the IR.
+func errorAt(inst llvm.Value, msg string) scanner.Error {
+	return scanner.Error{
+		Pos: getPosition(inst),
+		Msg: msg,
+	}
+}
+
+// getPosition returns the position information for the given value, as far as
+// it is available.
+func getPosition(val llvm.Value) token.Position {
+	if !val.IsAInstruction().IsNil() {
+		loc := val.InstructionDebugLoc()
+		if loc.IsNil() {
+			return token.Position{}
+		}
+		file := loc.LocationScope().ScopeFile()
+		return token.Position{
+			Filename: filepath.Join(file.FileDirectory(), file.FileFilename()),
+			Line:     int(loc.LocationLine()),
+			Column:   int(loc.LocationColumn()),
+		}
+	} else if !val.IsAFunction().IsNil() {
+		loc := val.Subprogram()
+		if loc.IsNil() {
+			return token.Position{}
+		}
+		file := loc.ScopeFile()
+		return token.Position{
+			Filename: filepath.Join(file.FileDirectory(), file.FileFilename()),
+			Line:     int(loc.SubprogramLine()),
+		}
+	} else {
+		return token.Position{}
+	}
+}

compiler/func-lowering.go

@@ -1,285 +0,0 @@
-package compiler
-
-// This file lowers func values into their final form. This is necessary for
-// funcValueSwitch, which needs full program analysis.
-
-import (
-	"sort"
-	"strconv"
-
-	"tinygo.org/x/go-llvm"
-)
-
-// funcSignatureInfo keeps information about a single signature and its uses.
-type funcSignatureInfo struct {
-	sig                     llvm.Value   // *uint8 to identify the signature
-	funcValueWithSignatures []llvm.Value // slice of runtime.funcValueWithSignature
-}
-
-// funcWithUses keeps information about a single function used as func value and
-// the assigned function ID. More commonly used functions are assigned a lower
-// ID.
-type funcWithUses struct {
-	funcPtr  llvm.Value
-	useCount int // how often this function is used in a func value
-	id       int // assigned ID
-}
-
-// Slice to sort functions by their use counts, or else their name if they're
-// used equally often.
-type funcWithUsesList []*funcWithUses
-
-func (l funcWithUsesList) Len() int { return len(l) }
-func (l funcWithUsesList) Less(i, j int) bool {
-	if l[i].useCount != l[j].useCount {
-		// return the reverse: we want the highest use counts sorted first
-		return l[i].useCount > l[j].useCount
-	}
-	iName := l[i].funcPtr.Name()
-	jName := l[j].funcPtr.Name()
-	return iName < jName
-}
-func (l funcWithUsesList) Swap(i, j int) {
-	l[i], l[j] = l[j], l[i]
-}
-
-// LowerFuncValue lowers the runtime.funcValueWithSignature type and
-// runtime.getFuncPtr function to their final form.
-func (c *Compiler) LowerFuncValues() {
-	if c.funcImplementation() != funcValueSwitch {
-		return
-	}
-
-	// Find all func values used in the program with their signatures.
-	funcValueWithSignaturePtr := llvm.PointerType(c.getLLVMRuntimeType("funcValueWithSignature"), 0)
-	signatures := map[string]*funcSignatureInfo{}
-	for global := c.mod.FirstGlobal(); !global.IsNil(); global = llvm.NextGlobal(global) {
-		if global.Type() != funcValueWithSignaturePtr {
-			continue
-		}
-		sig := llvm.ConstExtractValue(global.Initializer(), []uint32{1})
-		name := sig.Name()
-		if info, ok := signatures[name]; ok {
-			info.funcValueWithSignatures = append(info.funcValueWithSignatures, global)
-		} else {
-			signatures[name] = &funcSignatureInfo{
-				sig:                     sig,
-				funcValueWithSignatures: []llvm.Value{global},
-			}
-		}
-	}
-
-	// Sort the signatures, for deterministic execution.
-	names := make([]string, 0, len(signatures))
-	for name := range signatures {
-		names = append(names, name)
-	}
-	sort.Strings(names)
-
-	for _, name := range names {
-		info := signatures[name]
-		functions := make(funcWithUsesList, len(info.funcValueWithSignatures))
-		for i, use := range info.funcValueWithSignatures {
-			var useCount int
-			for _, use2 := range getUses(use) {
-				useCount += len(getUses(use2))
-			}
-			functions[i] = &funcWithUses{
-				funcPtr:  llvm.ConstExtractValue(use.Initializer(), []uint32{0}).Operand(0),
-				useCount: useCount,
-			}
-		}
-		sort.Sort(functions)
-
-		for i, fn := range functions {
-			fn.id = i + 1
-			for _, ptrtoint := range getUses(fn.funcPtr) {
-				if ptrtoint.IsAConstantExpr().IsNil() || ptrtoint.Opcode() != llvm.PtrToInt {
-					continue
-				}
-				for _, funcValueWithSignatureConstant := range getUses(ptrtoint) {
-					for _, funcValueWithSignatureGlobal := range getUses(funcValueWithSignatureConstant) {
-						for _, use := range getUses(funcValueWithSignatureGlobal) {
-							if ptrtoint.IsAConstantExpr().IsNil() || ptrtoint.Opcode() != llvm.PtrToInt {
-								panic("expected const ptrtoint")
-							}
-							use.ReplaceAllUsesWith(llvm.ConstInt(c.uintptrType, uint64(fn.id), false))
-						}
-					}
-				}
-			}
-		}
-
-		for _, getFuncPtrCall := range getUses(info.sig) {
-			if getFuncPtrCall.IsACallInst().IsNil() {
-				continue
-			}
-			if getFuncPtrCall.CalledValue().Name() != "runtime.getFuncPtr" {
-				panic("expected all call uses to be runtime.getFuncPtr")
-			}
-			funcID := getFuncPtrCall.Operand(1)
-			switch len(functions) {
-			case 0:
-				// There are no functions used in a func value that implement
-				// this signature. The only possible value is a nil value.
-				for _, inttoptr := range getUses(getFuncPtrCall) {
-					if inttoptr.IsAIntToPtrInst().IsNil() {
-						panic("expected inttoptr")
-					}
-					nilptr := llvm.ConstPointerNull(inttoptr.Type())
-					inttoptr.ReplaceAllUsesWith(nilptr)
-					inttoptr.EraseFromParentAsInstruction()
-				}
-				getFuncPtrCall.EraseFromParentAsInstruction()
-			case 1:
-				// There is exactly one function with this signature that is
-				// used in a func value. The func value itself can be either nil
-				// or this one function.
-				c.builder.SetInsertPointBefore(getFuncPtrCall)
-				zero := llvm.ConstInt(c.uintptrType, 0, false)
-				isnil := c.builder.CreateICmp(llvm.IntEQ, funcID, zero, "")
-				funcPtrNil := llvm.ConstPointerNull(functions[0].funcPtr.Type())
-				funcPtr := c.builder.CreateSelect(isnil, funcPtrNil, functions[0].funcPtr, "")
-				for _, inttoptr := range getUses(getFuncPtrCall) {
-					if inttoptr.IsAIntToPtrInst().IsNil() {
-						panic("expected inttoptr")
-					}
-					inttoptr.ReplaceAllUsesWith(funcPtr)
-					inttoptr.EraseFromParentAsInstruction()
-				}
-				getFuncPtrCall.EraseFromParentAsInstruction()
-			default:
-				// There are multiple functions used in a func value that
-				// implement this signature.
-				// What we'll do is transform the following:
-				//     rawPtr := runtime.getFuncPtr(func.ptr)
-				//     if rawPtr == nil {
-				//         runtime.nilPanic()
-				//     }
-				//     result := rawPtr(...args, func.context)
-				// into this:
-				//     if false {
-				//         runtime.nilPanic()
-				//     }
-				//     var result // Phi
-				//     switch fn.id {
-				//     case 0:
-				//         runtime.nilPanic()
-				//     case 1:
-				//         result = call first implementation...
-				//     case 2:
-				//         result = call second implementation...
-				//     default:
-				//         unreachable
-				//     }
-
-				// Remove some casts, checks, and the old call which we're going
-				// to replace.
-				for _, callIntPtr := range getUses(getFuncPtrCall) {
-					if !callIntPtr.IsACallInst().IsNil() && callIntPtr.CalledValue().Name() == "runtime.makeGoroutine" {
-						for _, inttoptr := range getUses(callIntPtr) {
-							if inttoptr.IsAIntToPtrInst().IsNil() {
-								panic("expected a inttoptr")
-							}
-							for _, use := range getUses(inttoptr) {
-								c.addFuncLoweringSwitch(funcID, use, c.emitStartGoroutine, functions)
-								use.EraseFromParentAsInstruction()
-							}
-							inttoptr.EraseFromParentAsInstruction()
-						}
-						callIntPtr.EraseFromParentAsInstruction()
-						continue
-					}
-					if callIntPtr.IsAIntToPtrInst().IsNil() {
-						panic("expected inttoptr")
-					}
-					for _, ptrUse := range getUses(callIntPtr) {
-						if !ptrUse.IsABitCastInst().IsNil() {
-							for _, bitcastUse := range getUses(ptrUse) {
-								if bitcastUse.IsACallInst().IsNil() || bitcastUse.CalledValue().IsAFunction().IsNil() {
-									panic("expected a call instruction")
-								}
-								switch bitcastUse.CalledValue().Name() {
-								case "runtime.isnil":
-									bitcastUse.ReplaceAllUsesWith(llvm.ConstInt(c.ctx.Int1Type(), 0, false))
-									bitcastUse.EraseFromParentAsInstruction()
-								default:
-									panic("expected a call to runtime.isnil")
-								}
-							}
-						} else if !ptrUse.IsACallInst().IsNil() && ptrUse.CalledValue() == callIntPtr {
-							c.addFuncLoweringSwitch(funcID, ptrUse, func(funcPtr llvm.Value, params []llvm.Value) llvm.Value {
-								return c.builder.CreateCall(funcPtr, params, "")
-							}, functions)
-						} else {
-							panic("unexpected getFuncPtrCall")
-						}
-						ptrUse.EraseFromParentAsInstruction()
-					}
-					callIntPtr.EraseFromParentAsInstruction()
-				}
-				getFuncPtrCall.EraseFromParentAsInstruction()
-			}
-		}
-	}
-}
-
-// addFuncLoweringSwitch creates a new switch on a function ID and inserts calls
-// to the newly created direct calls. The funcID is the number to switch on,
-// call is the call instruction to replace, and createCall is the callback that
-// actually creates the new call. By changing createCall to something other than
-// c.builder.CreateCall, instead of calling a function it can start a new
-// goroutine for example.
-func (c *Compiler) addFuncLoweringSwitch(funcID, call llvm.Value, createCall func(funcPtr llvm.Value, params []llvm.Value) llvm.Value, functions funcWithUsesList) {
-	// The block that cannot be reached with correct funcValues (to help the
-	// optimizer).
-	c.builder.SetInsertPointBefore(call)
-	defaultBlock := c.ctx.AddBasicBlock(call.InstructionParent().Parent(), "func.default")
-	c.builder.SetInsertPointAtEnd(defaultBlock)
-	c.builder.CreateUnreachable()
-
-	// Create the switch.
-	c.builder.SetInsertPointBefore(call)
-	sw := c.builder.CreateSwitch(funcID, defaultBlock, len(functions)+1)
-
-	// Split right after the switch. We will need to insert a few basic blocks
-	// in this gap.
-	nextBlock := c.splitBasicBlock(sw, llvm.NextBasicBlock(sw.InstructionParent()), "func.next")
-
-	// The 0 case, which is actually a nil check.
-	nilBlock := c.ctx.InsertBasicBlock(nextBlock, "func.nil")
-	c.builder.SetInsertPointAtEnd(nilBlock)
-	c.createRuntimeCall("nilPanic", nil, "")
-	c.builder.CreateUnreachable()
-	sw.AddCase(llvm.ConstInt(c.uintptrType, 0, false), nilBlock)
-
-	// Gather the list of parameters for every call we're going to make.
-	callParams := make([]llvm.Value, call.OperandsCount()-1)
-	for i := range callParams {
-		callParams[i] = call.Operand(i)
-	}
-
-	// If the call produces a value, we need to get it using a PHI
-	// node.
-	phiBlocks := make([]llvm.BasicBlock, len(functions))
-	phiValues := make([]llvm.Value, len(functions))
-	for i, fn := range functions {
-		// Insert a switch case.
-		bb := c.ctx.InsertBasicBlock(nextBlock, "func.call"+strconv.Itoa(fn.id))
-		c.builder.SetInsertPointAtEnd(bb)
-		result := createCall(fn.funcPtr, callParams)
-		c.builder.CreateBr(nextBlock)
-		sw.AddCase(llvm.ConstInt(c.uintptrType, uint64(fn.id), false), bb)
-		phiBlocks[i] = bb
-		phiValues[i] = result
-	}
-	// Create the PHI node so that the call result flows into the
-	// next block (after the split). This is only necessary when the
-	// call produced a value.
-	if call.Type().TypeKind() != llvm.VoidTypeKind {
-		c.builder.SetInsertPointBefore(nextBlock.FirstInstruction())
-		phi := c.builder.CreatePHI(call.Type(), "")
-		phi.AddIncoming(phiValues, phiBlocks)
-		call.ReplaceAllUsesWith(phi)
-	}
-}

compiler/func.go

@@ -10,62 +10,33 @@ import (
 	"tinygo.org/x/go-llvm"
 )
 
-type funcValueImplementation int
-
-const (
-	funcValueNone funcValueImplementation = iota
-
-	// A func value is implemented as a pair of pointers:
-	//     {context, function pointer}
-	// where the context may be a pointer to a heap-allocated struct containing
-	// the free variables, or it may be undef if the function being pointed to
-	// doesn't need a context. The function pointer is a regular function
-	// pointer.
-	funcValueDoubleword
-
-	// As funcValueDoubleword, but with the function pointer replaced by a
-	// unique ID per function signature. Function values are called by using a
-	// switch statement and choosing which function to call.
-	funcValueSwitch
-)
-
-// funcImplementation picks an appropriate func value implementation for the
-// target.
-func (c *Compiler) funcImplementation() funcValueImplementation {
-	// Always pick the switch implementation, as it allows the use of blocking
-	// inside a function that is used as a func value.
-	switch c.selectScheduler() {
-	case "coroutines":
-		return funcValueSwitch
-	case "tasks":
-		return funcValueDoubleword
-	default:
-		panic("unknown scheduler type")
-	}
+// createFuncValue creates a function value from a raw function pointer with no
+// context.
+func (b *builder) createFuncValue(funcPtr, context llvm.Value, sig *types.Signature) llvm.Value {
+	return b.compilerContext.createFuncValue(b.Builder, funcPtr, context, sig)
 }
 
 // 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 {
+func (c *compilerContext) createFuncValue(builder llvm.Builder, funcPtr, context llvm.Value, sig *types.Signature) llvm.Value {
 	var funcValueScalar llvm.Value
-	switch c.funcImplementation() {
-	case funcValueDoubleword:
+	switch c.FuncImplementation {
+	case "doubleword":
 		// Closure is: {context, function pointer}
 		funcValueScalar = funcPtr
-	case funcValueSwitch:
-		sigGlobal := c.getTypeCode(sig)
+	case "switch":
 		funcValueWithSignatureGlobalName := funcPtr.Name() + "$withSignature"
 		funcValueWithSignatureGlobal := c.mod.NamedGlobal(funcValueWithSignatureGlobalName)
 		if funcValueWithSignatureGlobal.IsNil() {
 			funcValueWithSignatureType := c.getLLVMRuntimeType("funcValueWithSignature")
 			funcValueWithSignature := llvm.ConstNamedStruct(funcValueWithSignatureType, []llvm.Value{
 				llvm.ConstPtrToInt(funcPtr, c.uintptrType),
-				sigGlobal,
+				c.getFuncSignatureID(sig),
 			})
 			funcValueWithSignatureGlobal = llvm.AddGlobal(c.mod, funcValueWithSignatureType, funcValueWithSignatureGlobalName)
 			funcValueWithSignatureGlobal.SetInitializer(funcValueWithSignature)
 			funcValueWithSignatureGlobal.SetGlobalConstant(true)
-			funcValueWithSignatureGlobal.SetLinkage(llvm.InternalLinkage)
+			funcValueWithSignatureGlobal.SetLinkage(llvm.LinkOnceODRLinkage)
 		}
 		funcValueScalar = llvm.ConstPtrToInt(funcValueWithSignatureGlobal, c.uintptrType)
 	default:
@@ -73,35 +44,48 @@ func (c *Compiler) createFuncValue(funcPtr, context llvm.Value, sig *types.Signa
 	}
 	funcValueType := c.getFuncType(sig)
 	funcValue := llvm.Undef(funcValueType)
-	funcValue = c.builder.CreateInsertValue(funcValue, context, 0, "")
-	funcValue = c.builder.CreateInsertValue(funcValue, funcValueScalar, 1, "")
+	funcValue = builder.CreateInsertValue(funcValue, context, 0, "")
+	funcValue = builder.CreateInsertValue(funcValue, funcValueScalar, 1, "")
 	return funcValue
 }
 
+// getFuncSignatureID returns a new external global for a given signature. This
+// global reference is not real, it is only used during func lowering to assign
+// signature types to functions and will then be removed.
+func (c *compilerContext) getFuncSignatureID(sig *types.Signature) llvm.Value {
+	sigGlobalName := "reflect/types.funcid:" + getTypeCodeName(sig)
+	sigGlobal := c.mod.NamedGlobal(sigGlobalName)
+	if sigGlobal.IsNil() {
+		sigGlobal = llvm.AddGlobal(c.mod, c.ctx.Int8Type(), sigGlobalName)
+		sigGlobal.SetGlobalConstant(true)
+	}
+	return sigGlobal
+}
+
 // extractFuncScalar returns some scalar that can be used in comparisons. It is
 // a cheap operation.
-func (c *Compiler) extractFuncScalar(funcValue llvm.Value) llvm.Value {
-	return c.builder.CreateExtractValue(funcValue, 1, "")
+func (b *builder) extractFuncScalar(funcValue llvm.Value) llvm.Value {
+	return b.CreateExtractValue(funcValue, 1, "")
 }
 
 // extractFuncContext extracts the context pointer from this function value. It
 // is a cheap operation.
-func (c *Compiler) extractFuncContext(funcValue llvm.Value) llvm.Value {
-	return c.builder.CreateExtractValue(funcValue, 0, "")
+func (b *builder) extractFuncContext(funcValue llvm.Value) llvm.Value {
+	return b.CreateExtractValue(funcValue, 0, "")
 }
 
 // decodeFuncValue extracts the context and the function pointer from this func
 // value. This may be an expensive operation.
-func (c *Compiler) decodeFuncValue(funcValue llvm.Value, sig *types.Signature) (funcPtr, context llvm.Value) {
-	context = c.builder.CreateExtractValue(funcValue, 0, "")
-	switch c.funcImplementation() {
-	case funcValueDoubleword:
-		funcPtr = c.builder.CreateExtractValue(funcValue, 1, "")
-	case funcValueSwitch:
-		llvmSig := c.getRawFuncType(sig)
-		sigGlobal := c.getTypeCode(sig)
-		funcPtr = c.createRuntimeCall("getFuncPtr", []llvm.Value{funcValue, sigGlobal}, "")
-		funcPtr = c.builder.CreateIntToPtr(funcPtr, llvmSig, "")
+func (b *builder) decodeFuncValue(funcValue llvm.Value, sig *types.Signature) (funcPtr, context llvm.Value) {
+	context = b.CreateExtractValue(funcValue, 0, "")
+	switch b.FuncImplementation {
+	case "doubleword":
+		funcPtr = b.CreateExtractValue(funcValue, 1, "")
+	case "switch":
+		llvmSig := b.getRawFuncType(sig)
+		sigGlobal := b.getFuncSignatureID(sig)
+		funcPtr = b.createRuntimeCall("getFuncPtr", []llvm.Value{funcValue, sigGlobal}, "")
+		funcPtr = b.CreateIntToPtr(funcPtr, llvmSig, "")
 	default:
 		panic("unimplemented func value variant")
 	}
@@ -109,12 +93,12 @@ 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 {
-	switch c.funcImplementation() {
-	case funcValueDoubleword:
+func (c *compilerContext) getFuncType(typ *types.Signature) llvm.Type {
+	switch c.FuncImplementation {
+	case "doubleword":
 		rawPtr := c.getRawFuncType(typ)
 		return c.ctx.StructType([]llvm.Type{c.i8ptrType, rawPtr}, false)
-	case funcValueSwitch:
+	case "switch":
 		return c.getLLVMRuntimeType("funcValue")
 	default:
 		panic("unimplemented func value variant")
@@ -122,7 +106,7 @@ func (c *Compiler) getFuncType(typ *types.Signature) llvm.Type {
 }
 
 // getRawFuncType returns a LLVM function pointer type for a given signature.
-func (c *Compiler) getRawFuncType(typ *types.Signature) llvm.Type {
+func (c *compilerContext) getRawFuncType(typ *types.Signature) llvm.Type {
 	// Get the return type.
 	var returnType llvm.Type
 	switch typ.Results().Len() {
@@ -152,11 +136,15 @@ func (c *Compiler) getRawFuncType(typ *types.Signature) llvm.Type {
 			// The receiver is not an interface, but a i8* type.
 			recv = c.i8ptrType
 		}
-		paramTypes = append(paramTypes, c.expandFormalParamType(recv)...)
+		for _, info := range c.expandFormalParamType(recv, "", nil) {
+			paramTypes = append(paramTypes, info.llvmType)
+		}
 	}
 	for i := 0; i < typ.Params().Len(); i++ {
 		subType := c.getLLVMType(typ.Params().At(i).Type())
-		paramTypes = append(paramTypes, c.expandFormalParamType(subType)...)
+		for _, info := range c.expandFormalParamType(subType, "", nil) {
+			paramTypes = append(paramTypes, info.llvmType)
+		}
 	}
 	// All functions take these parameters at the end.
 	paramTypes = append(paramTypes, c.i8ptrType) // context
@@ -168,24 +156,24 @@ func (c *Compiler) getRawFuncType(typ *types.Signature) llvm.Type {
 
 // parseMakeClosure makes a function value (with context) from the given
 // closure expression.
-func (c *Compiler) parseMakeClosure(frame *Frame, expr *ssa.MakeClosure) (llvm.Value, error) {
+func (b *builder) parseMakeClosure(expr *ssa.MakeClosure) (llvm.Value, error) {
 	if len(expr.Bindings) == 0 {
 		panic("unexpected: MakeClosure without bound variables")
 	}
-	f := c.ir.GetFunction(expr.Fn.(*ssa.Function))
+	f := expr.Fn.(*ssa.Function)
 
 	// Collect all bound variables.
 	boundVars := make([]llvm.Value, len(expr.Bindings))
 	for i, binding := range expr.Bindings {
 		// The context stores the bound variables.
-		llvmBoundVar := c.getValue(frame, binding)
+		llvmBoundVar := b.getValue(binding)
 		boundVars[i] = llvmBoundVar
 	}
 
 	// Store the bound variables in a single object, allocating it on the heap
 	// if necessary.
-	context := c.emitPointerPack(boundVars)
+	context := b.emitPointerPack(boundVars)
 
 	// Create the closure.
-	return c.createFuncValue(f.LLVMFn, context, f.Signature), nil
+	return b.createFuncValue(b.getFunction(f), context, f.Signature), nil
 }

compiler/gc.go

@@ -5,76 +5,66 @@ package compiler
 
 import (
 	"go/token"
-	"math/big"
 
 	"golang.org/x/tools/go/ssa"
 	"tinygo.org/x/go-llvm"
 )
 
-// needsStackObjects returns true if the compiler should insert stack objects
-// that can be traced by the garbage collector.
-func (c *Compiler) needsStackObjects() bool {
-	if c.selectGC() != "conservative" {
-		return false
-	}
-	for _, tag := range c.BuildTags {
-		if tag == "baremetal" {
-			return false
-		}
-	}
-
-	return true
-}
-
 // trackExpr inserts pointer tracking intrinsics for the GC if the expression is
 // one of the expressions that need this.
-func (c *Compiler) trackExpr(frame *Frame, expr ssa.Value, value llvm.Value) {
+func (b *builder) trackExpr(expr ssa.Value, value llvm.Value) {
 	// There are uses of this expression, Make sure the pointers
 	// are tracked during GC.
 	switch expr := expr.(type) {
 	case *ssa.Alloc, *ssa.MakeChan, *ssa.MakeMap:
 		// These values are always of pointer type in IR.
-		c.trackPointer(value)
+		b.trackPointer(value)
 	case *ssa.Call, *ssa.Convert, *ssa.MakeClosure, *ssa.MakeInterface, *ssa.MakeSlice, *ssa.Next:
 		if !value.IsNil() {
-			c.trackValue(value)
+			b.trackValue(value)
 		}
 	case *ssa.Select:
-		if alloca, ok := frame.selectRecvBuf[expr]; ok {
+		if alloca, ok := b.selectRecvBuf[expr]; ok {
 			if alloca.IsAUndefValue().IsNil() {
-				c.trackPointer(alloca)
+				b.trackPointer(alloca)
 			}
 		}
 	case *ssa.UnOp:
 		switch expr.Op {
 		case token.MUL:
 			// Pointer dereference.
-			c.trackValue(value)
+			b.trackValue(value)
 		case token.ARROW:
 			// Channel receive operator.
 			// It's not necessary to look at commaOk here, because in that
 			// case it's just an aggregate and trackValue will extract the
 			// pointer in there (if there is one).
-			c.trackValue(value)
+			b.trackValue(value)
+		}
+	case *ssa.BinOp:
+		switch expr.Op {
+		case token.ADD:
+			// String concatenation.
+			b.trackValue(value)
 		}
 	}
 }
 
 // trackValue locates pointers in a value (possibly an aggregate) and tracks the
 // individual pointers
-func (c *Compiler) trackValue(value llvm.Value) {
+func (b *builder) trackValue(value llvm.Value) {
 	typ := value.Type()
 	switch typ.TypeKind() {
 	case llvm.PointerTypeKind:
-		c.trackPointer(value)
+		b.trackPointer(value)
 	case llvm.StructTypeKind:
 		if !typeHasPointers(typ) {
 			return
 		}
 		numElements := typ.StructElementTypesCount()
 		for i := 0; i < numElements; i++ {
-			subValue := c.builder.CreateExtractValue(value, i, "")
-			c.trackValue(subValue)
+			subValue := b.CreateExtractValue(value, i, "")
+			b.trackValue(subValue)
 		}
 	case llvm.ArrayTypeKind:
 		if !typeHasPointers(typ) {
@@ -82,19 +72,19 @@ func (c *Compiler) trackValue(value llvm.Value) {
 		}
 		numElements := typ.ArrayLength()
 		for i := 0; i < numElements; i++ {
-			subValue := c.builder.CreateExtractValue(value, i, "")
-			c.trackValue(subValue)
+			subValue := b.CreateExtractValue(value, i, "")
+			b.trackValue(subValue)
 		}
 	}
 }
 
 // trackPointer creates a call to runtime.trackPointer, bitcasting the poitner
 // first if needed. The input value must be of LLVM pointer type.
-func (c *Compiler) trackPointer(value llvm.Value) {
-	if value.Type() != c.i8ptrType {
-		value = c.builder.CreateBitCast(value, c.i8ptrType, "")
+func (b *builder) trackPointer(value llvm.Value) {
+	if value.Type() != b.i8ptrType {
+		value = b.CreateBitCast(value, b.i8ptrType, "")
 	}
-	c.createRuntimeCall("trackPointer", []llvm.Value{value}, "")
+	b.createRuntimeCall("trackPointer", []llvm.Value{value}, "")
 }
 
 // typeHasPointers returns whether this type is a pointer or contains pointers.
@@ -120,352 +110,3 @@ func typeHasPointers(t llvm.Type) bool {
 		return false
 	}
 }
-
-// makeGCStackSlots converts all calls to runtime.trackPointer to explicit
-// stores to stack slots that are scannable by the GC.
-func (c *Compiler) makeGCStackSlots() bool {
-	// Check whether there are allocations at all.
-	alloc := c.mod.NamedFunction("runtime.alloc")
-	if alloc.IsNil() {
-		// Nothing to. Make sure all remaining bits and pieces for stack
-		// chains are neutralized.
-		for _, call := range getUses(c.mod.NamedFunction("runtime.trackPointer")) {
-			call.EraseFromParentAsInstruction()
-		}
-		stackChainStart := c.mod.NamedGlobal("runtime.stackChainStart")
-		if !stackChainStart.IsNil() {
-			stackChainStart.SetInitializer(llvm.ConstNull(stackChainStart.Type().ElementType()))
-			stackChainStart.SetGlobalConstant(true)
-		}
-	}
-
-	trackPointer := c.mod.NamedFunction("runtime.trackPointer")
-	if trackPointer.IsNil() || trackPointer.FirstUse().IsNil() {
-		return false // nothing to do
-	}
-
-	// Look at *all* functions to see whether they are free of function pointer
-	// calls.
-	// This takes less than 5ms for ~100kB of WebAssembly but would perhaps be
-	// faster when written in C++ (to avoid the CGo overhead).
-	funcsWithFPCall := map[llvm.Value]struct{}{}
-	n := 0
-	for fn := c.mod.FirstFunction(); !fn.IsNil(); fn = llvm.NextFunction(fn) {
-		n++
-		if _, ok := funcsWithFPCall[fn]; ok {
-			continue // already found
-		}
-		done := false
-		for bb := fn.FirstBasicBlock(); !bb.IsNil() && !done; bb = llvm.NextBasicBlock(bb) {
-			for call := bb.FirstInstruction(); !call.IsNil() && !done; call = llvm.NextInstruction(call) {
-				if call.IsACallInst().IsNil() {
-					continue // only looking at calls
-				}
-				called := call.CalledValue()
-				if !called.IsAFunction().IsNil() {
-					continue // only looking for function pointers
-				}
-				funcsWithFPCall[fn] = struct{}{}
-				markParentFunctions(funcsWithFPCall, fn)
-				done = true
-			}
-		}
-	}
-
-	// Determine which functions need stack objects. Many leaf functions don't
-	// need it: it only causes overhead for them.
-	// Actually, in one test it was only able to eliminate stack object from 12%
-	// of functions that had a call to runtime.trackPointer (8 out of 68
-	// functions), so this optimization is not as big as it may seem.
-	allocatingFunctions := map[llvm.Value]struct{}{} // set of allocating functions
-
-	// Work from runtime.alloc and trace all parents to check which functions do
-	// a heap allocation (and thus which functions do not).
-	markParentFunctions(allocatingFunctions, alloc)
-
-	// Also trace all functions that call a function pointer.
-	for fn := range funcsWithFPCall {
-		// Assume that functions that call a function pointer do a heap
-		// allocation as a conservative guess because the called function might
-		// do a heap allocation.
-		allocatingFunctions[fn] = struct{}{}
-		markParentFunctions(allocatingFunctions, fn)
-	}
-
-	// Collect some variables used below in the loop.
-	stackChainStart := c.mod.NamedGlobal("runtime.stackChainStart")
-	if stackChainStart.IsNil() {
-		panic("stack chain start not found!")
-	}
-	stackChainStartType := stackChainStart.Type().ElementType()
-	stackChainStart.SetInitializer(llvm.ConstNull(stackChainStartType))
-
-	// Iterate until runtime.trackPointer has no uses left.
-	for use := trackPointer.FirstUse(); !use.IsNil(); use = trackPointer.FirstUse() {
-		// Pick the first use of runtime.trackPointer.
-		call := use.User()
-		if call.IsACallInst().IsNil() {
-			panic("expected runtime.trackPointer use to be a call")
-		}
-
-		// Pick the parent function.
-		fn := call.InstructionParent().Parent()
-
-		if _, ok := allocatingFunctions[fn]; !ok {
-			// This function nor any of the functions it calls (recursively)
-			// allocate anything from the heap, so it will not trigger a garbage
-			// collection cycle. Thus, it does not need to track local pointer
-			// values.
-			// This is a useful optimization but not as big as you might guess,
-			// as described above (it avoids stack objects for ~12% of
-			// functions).
-			call.EraseFromParentAsInstruction()
-			continue
-		}
-
-		// Find all calls to runtime.trackPointer in this function.
-		var calls []llvm.Value
-		var returns []llvm.Value
-		for bb := fn.FirstBasicBlock(); !bb.IsNil(); bb = llvm.NextBasicBlock(bb) {
-			for inst := bb.FirstInstruction(); !inst.IsNil(); inst = llvm.NextInstruction(inst) {
-				switch inst.InstructionOpcode() {
-				case llvm.Call:
-					if inst.CalledValue() == trackPointer {
-						calls = append(calls, inst)
-					}
-				case llvm.Ret:
-					returns = append(returns, inst)
-				}
-			}
-		}
-
-		// Determine what to do with each call.
-		var allocas, pointers []llvm.Value
-		for _, call := range calls {
-			ptr := call.Operand(0)
-			call.EraseFromParentAsInstruction()
-			if ptr.IsAInstruction().IsNil() {
-				continue
-			}
-
-			// Some trivial optimizations.
-			if ptr.IsAInstruction().IsNil() {
-				continue
-			}
-			switch ptr.InstructionOpcode() {
-			case llvm.PHI, llvm.GetElementPtr:
-				// These values do not create new values: the values already
-				// existed locally in this function so must have been tracked
-				// already.
-				continue
-			case llvm.ExtractValue, llvm.BitCast:
-				// These instructions do not create new values, but their
-				// original value may not be tracked. So keep tracking them for
-				// now.
-				// With more analysis, it should be possible to optimize a
-				// significant chunk of these away.
-			case llvm.Call, llvm.Load, llvm.IntToPtr:
-				// These create new values so must be stored locally. But
-				// perhaps some of these can be fused when they actually refer
-				// to the same value.
-			default:
-				// Ambiguous. These instructions are uncommon, but perhaps could
-				// be optimized if needed.
-			}
-
-			if !ptr.IsAAllocaInst().IsNil() {
-				if typeHasPointers(ptr.Type().ElementType()) {
-					allocas = append(allocas, ptr)
-				}
-			} else {
-				pointers = append(pointers, ptr)
-			}
-		}
-
-		if len(allocas) == 0 && len(pointers) == 0 {
-			// This function does not need to keep track of stack pointers.
-			continue
-		}
-
-		// Determine the type of the required stack slot.
-		fields := []llvm.Type{
-			stackChainStartType, // Pointer to parent frame.
-			c.uintptrType,       // Number of elements in this frame.
-		}
-		for _, alloca := range allocas {
-			fields = append(fields, alloca.Type().ElementType())
-		}
-		for _, ptr := range pointers {
-			fields = append(fields, ptr.Type())
-		}
-		stackObjectType := c.ctx.StructType(fields, false)
-
-		// Create the stack object at the function entry.
-		c.builder.SetInsertPointBefore(fn.EntryBasicBlock().FirstInstruction())
-		stackObject := c.builder.CreateAlloca(stackObjectType, "gc.stackobject")
-		initialStackObject := llvm.ConstNull(stackObjectType)
-		numSlots := (c.targetData.TypeAllocSize(stackObjectType) - c.targetData.TypeAllocSize(c.i8ptrType)*2) / uint64(c.targetData.ABITypeAlignment(c.uintptrType))
-		numSlotsValue := llvm.ConstInt(c.uintptrType, numSlots, false)
-		initialStackObject = llvm.ConstInsertValue(initialStackObject, numSlotsValue, []uint32{1})
-		c.builder.CreateStore(initialStackObject, stackObject)
-
-		// Update stack start.
-		parent := c.builder.CreateLoad(stackChainStart, "")
-		gep := c.builder.CreateGEP(stackObject, []llvm.Value{
-			llvm.ConstInt(c.ctx.Int32Type(), 0, false),
-			llvm.ConstInt(c.ctx.Int32Type(), 0, false),
-		}, "")
-		c.builder.CreateStore(parent, gep)
-		stackObjectCast := c.builder.CreateBitCast(stackObject, stackChainStartType, "")
-		c.builder.CreateStore(stackObjectCast, stackChainStart)
-
-		// Replace all independent allocas with GEPs in the stack object.
-		for i, alloca := range allocas {
-			gep := c.builder.CreateGEP(stackObject, []llvm.Value{
-				llvm.ConstInt(c.ctx.Int32Type(), 0, false),
-				llvm.ConstInt(c.ctx.Int32Type(), uint64(2+i), false),
-			}, "")
-			alloca.ReplaceAllUsesWith(gep)
-			alloca.EraseFromParentAsInstruction()
-		}
-
-		// Do a store to the stack object after each new pointer that is created.
-		for i, ptr := range pointers {
-			c.builder.SetInsertPointBefore(llvm.NextInstruction(ptr))
-			gep := c.builder.CreateGEP(stackObject, []llvm.Value{
-				llvm.ConstInt(c.ctx.Int32Type(), 0, false),
-				llvm.ConstInt(c.ctx.Int32Type(), uint64(2+len(allocas)+i), false),
-			}, "")
-			c.builder.CreateStore(ptr, gep)
-		}
-
-		// Make sure this stack object is popped from the linked list of stack
-		// objects at return.
-		for _, ret := range returns {
-			c.builder.SetInsertPointBefore(ret)
-			c.builder.CreateStore(parent, stackChainStart)
-		}
-	}
-
-	return true
-}
-
-func (c *Compiler) addGlobalsBitmap() bool {
-	if c.mod.NamedGlobal("runtime.trackedGlobalsStart").IsNil() {
-		return false // 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(c.ctx.Int8Type(), bitmapValues)
-	bitmapNew := llvm.AddGlobal(c.mod, bitmapArray.Type(), "runtime.trackedGlobalsBitmap.tmp")
-	bitmapOld := c.mod.NamedGlobal("runtime.trackedGlobalsBitmap")
-	bitmapOld.ReplaceAllUsesWith(llvm.ConstBitCast(bitmapNew, bitmapOld.Type()))
-	bitmapNew.SetInitializer(bitmapArray)
-	bitmapNew.SetName("runtime.trackedGlobalsBitmap")
-
-	return true // the IR was changed
-}
-
-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)
-	}
-}
-
-// markParentFunctions traverses all parent function calls (recursively) and
-// adds them to the set of marked functions. It only considers function calls:
-// any other uses of such a function is ignored.
-func markParentFunctions(marked map[llvm.Value]struct{}, fn llvm.Value) {
-	worklist := []llvm.Value{fn}
-	for len(worklist) != 0 {
-		fn := worklist[len(worklist)-1]
-		worklist = worklist[:len(worklist)-1]
-		for _, use := range getUses(fn) {
-			if use.IsACallInst().IsNil() || use.CalledValue() != fn {
-				// Not the parent function.
-				continue
-			}
-			parent := use.InstructionParent().Parent()
-			if _, ok := marked[parent]; !ok {
-				marked[parent] = struct{}{}
-				worklist = append(worklist, parent)
-			}
-		}
-	}
-}

compiler/goroutine-lowering.go

@@ -1,673 +0,0 @@
-package compiler
-
-// This file implements lowering for the goroutine scheduler. There are two
-// scheduler implementations, one based on tasks (like RTOSes and the main Go
-// runtime) and one based on a coroutine compiler transformation. The task based
-// implementation requires very little work from the compiler but is not very
-// portable (in particular, it is very hard if not impossible to support on
-// WebAssembly). The coroutine based one requires a lot of work by the compiler
-// to implement, but can run virtually anywhere with a single scheduler
-// implementation.
-//
-// The below description is for the coroutine based scheduler.
-//
-// This file lowers goroutine pseudo-functions into coroutines scheduled by a
-// scheduler at runtime. It uses coroutine support in LLVM for this
-// transformation: https://llvm.org/docs/Coroutines.html
-//
-// For example, take the following code:
-//
-//     func main() {
-//         go foo()
-//         time.Sleep(2 * time.Second)
-//         println("some other operation")
-//         i := bar()
-//         println("done", *i)
-//     }
-//
-//     func foo() {
-//         for {
-//             println("foo!")
-//             time.Sleep(time.Second)
-//         }
-//     }
-//
-//     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
-// pseudo-Go code:
-//
-//     func main() {
-//         fn := runtime.makeGoroutine(foo)
-//         fn()
-//         time.Sleep(2 * time.Second)
-//         println("some other operation")
-//         i := bar() // imagine an 'await' keyword in front of this call
-//         println("done", *i)
-//     }
-//
-//     func foo() {
-//         for {
-//             println("foo!")
-//             time.Sleep(time.Second)
-//         }
-//     }
-//
-//     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
-// async/await style pseudocode:
-//
-//     func main(parent) {
-//         hdl := llvm.makeCoroutine()
-//         foo(nil)                                // do not pass the parent coroutine: this is an independent goroutine
-//         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.setTaskStatePtr(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", *i)
-//         runtime.activateTask(parent)            // re-activate the parent (nop, there is no parent)
-//     }
-//
-//     func foo(parent) {
-//         hdl := llvm.makeCoroutine()
-//         for {
-//             println("foo!")
-//             runtime.sleepTask(hdl, time.Second) // ask the scheduler to re-activate this coroutine at the right time
-//             llvm.suspend(hdl)                   // suspend point
-//         }
-//     }
-//
-//     func bar(parent) {
-//         hdl := llvm.makeCoroutine()
-//         runtime.sleepTask(hdl, time.Second) // ask the scheduler to re-activate this coroutine at the right time
-//         llvm.suspend(hdl)                   // suspend point
-//         println("blocking operation completed)
-//         runtime.activateTask(parent)        // re-activate the parent coroutine before returning
-//     }
-//
-// The real LLVM code is more complicated, but this is the general idea.
-//
-// The LLVM coroutine passes will then process this file further transforming
-// these three functions into coroutines. Most of the actual work is done by the
-// scheduler, which runs in the background scheduling all coroutines.
-
-import (
-	"errors"
-	"strings"
-
-	"tinygo.org/x/go-llvm"
-)
-
-type asyncFunc struct {
-	taskHandle       llvm.Value
-	cleanupBlock     llvm.BasicBlock
-	suspendBlock     llvm.BasicBlock
-	unreachableBlock llvm.BasicBlock
-}
-
-// LowerGoroutines performs some IR transformations necessary to support
-// goroutines. It does something different based on whether it uses the
-// coroutine or the tasks implementation of goroutines, and whether goroutines
-// are necessary at all.
-func (c *Compiler) LowerGoroutines() error {
-	switch c.selectScheduler() {
-	case "coroutines":
-		return c.lowerCoroutines()
-	case "tasks":
-		return c.lowerTasks()
-	default:
-		panic("unknown scheduler type")
-	}
-}
-
-// lowerTasks starts the main goroutine and then runs the scheduler.
-// This is enough compiler-level transformation for the task-based scheduler.
-func (c *Compiler) lowerTasks() error {
-	uses := getUses(c.mod.NamedFunction("runtime.callMain"))
-	if len(uses) != 1 || uses[0].IsACallInst().IsNil() {
-		panic("expected exactly 1 call of runtime.callMain, check the entry point")
-	}
-	mainCall := uses[0]
-
-	realMain := c.mod.NamedFunction(c.ir.MainPkg().Pkg.Path() + ".main")
-	if len(getUses(c.mod.NamedFunction("runtime.startGoroutine"))) != 0 {
-		// Program needs a scheduler. Start main.main as a goroutine and start
-		// the scheduler.
-		realMainWrapper := c.createGoroutineStartWrapper(realMain)
-		c.builder.SetInsertPointBefore(mainCall)
-		zero := llvm.ConstInt(c.uintptrType, 0, false)
-		c.createRuntimeCall("startGoroutine", []llvm.Value{realMainWrapper, zero}, "")
-		c.createRuntimeCall("scheduler", nil, "")
-		sleep := c.mod.NamedFunction("time.Sleep")
-		if !sleep.IsNil() {
-			sleep.ReplaceAllUsesWith(c.mod.NamedFunction("runtime.sleepCurrentTask"))
-		}
-	} else {
-		// Program doesn't need a scheduler. Call main.main directly.
-		c.builder.SetInsertPointBefore(mainCall)
-		params := []llvm.Value{
-			llvm.Undef(c.i8ptrType), // unused context parameter
-			llvm.Undef(c.i8ptrType), // unused coroutine handle
-		}
-		c.createCall(realMain, params, "")
-		// runtime.Goexit isn't needed so let it be optimized away by
-		// globalopt.
-		c.mod.NamedFunction("runtime.Goexit").SetLinkage(llvm.InternalLinkage)
-	}
-	mainCall.EraseFromParentAsInstruction()
-
-	// main.main was set to external linkage during IR construction. Set it to
-	// internal linkage to enable interprocedural optimizations.
-	realMain.SetLinkage(llvm.InternalLinkage)
-
-	return nil
-}
-
-// lowerCoroutines transforms the IR into one where all blocking functions are
-// turned into goroutines and blocking calls into await calls. It also makes
-// sure that the first coroutine is started and the coroutine scheduler will be
-// run.
-func (c *Compiler) lowerCoroutines() error {
-	needsScheduler, err := c.markAsyncFunctions()
-	if err != nil {
-		return err
-	}
-
-	uses := getUses(c.mod.NamedFunction("runtime.callMain"))
-	if len(uses) != 1 || uses[0].IsACallInst().IsNil() {
-		panic("expected exactly 1 call of runtime.callMain, check the entry point")
-	}
-	mainCall := uses[0]
-
-	// Replace call of runtime.callMain() with a real call to main.main(),
-	// optionally followed by a call to runtime.scheduler().
-	c.builder.SetInsertPointBefore(mainCall)
-	realMain := c.mod.NamedFunction(c.ir.MainPkg().Pkg.Path() + ".main")
-	c.builder.CreateCall(realMain, []llvm.Value{llvm.Undef(c.i8ptrType), llvm.ConstPointerNull(c.i8ptrType)}, "")
-	if needsScheduler {
-		c.createRuntimeCall("scheduler", nil, "")
-	}
-	mainCall.EraseFromParentAsInstruction()
-
-	if !needsScheduler {
-		go_scheduler := c.mod.NamedFunction("go_scheduler")
-		if !go_scheduler.IsNil() {
-			// This is the WebAssembly backend.
-			// There is no need to export the go_scheduler function, but it is
-			// still exported. Make sure it is optimized away.
-			go_scheduler.SetLinkage(llvm.InternalLinkage)
-		}
-	}
-
-	// main.main was set to external linkage during IR construction. Set it to
-	// internal linkage to enable interprocedural optimizations.
-	realMain.SetLinkage(llvm.InternalLinkage)
-
-	return nil
-}
-
-// markAsyncFunctions does the bulk of the work of lowering goroutines. It
-// determines whether a scheduler is needed, and if it is, it transforms
-// blocking operations into goroutines and blocking calls into await calls.
-//
-// It does the following operations:
-//    * Find all blocking functions.
-//    * Determine whether a scheduler is necessary. If not, it skips the
-//      following operations.
-//    * Transform call instructions into await calls.
-//    * Transform return instructions into final suspends.
-//    * Set up the coroutine frames for async functions.
-//    * Transform blocking calls into their async equivalents.
-func (c *Compiler) markAsyncFunctions() (needsScheduler bool, err error) {
-	var worklist []llvm.Value
-
-	sleep := c.mod.NamedFunction("time.Sleep")
-	if !sleep.IsNil() {
-		worklist = append(worklist, sleep)
-	}
-	deadlock := c.mod.NamedFunction("runtime.deadlock")
-	if !deadlock.IsNil() {
-		worklist = append(worklist, deadlock)
-	}
-	chanSend := c.mod.NamedFunction("runtime.chanSend")
-	if !chanSend.IsNil() {
-		worklist = append(worklist, chanSend)
-	}
-	chanRecv := c.mod.NamedFunction("runtime.chanRecv")
-	if !chanRecv.IsNil() {
-		worklist = append(worklist, chanRecv)
-	}
-
-	if len(worklist) == 0 {
-		// There are no blocking operations, so no need to transform anything.
-		return false, c.lowerMakeGoroutineCalls()
-	}
-
-	// Find all async functions.
-	// Keep reducing this worklist by marking a function as recursively async
-	// from the worklist and pushing all its parents that are non-async.
-	// This is somewhat similar to a worklist in a mark-sweep garbage collector:
-	// the work items are then grey objects.
-	asyncFuncs := make(map[llvm.Value]*asyncFunc)
-	asyncList := make([]llvm.Value, 0, 4)
-	for len(worklist) != 0 {
-		// Pick the topmost.
-		f := worklist[len(worklist)-1]
-		worklist = worklist[:len(worklist)-1]
-		if _, ok := asyncFuncs[f]; ok {
-			continue // already processed
-		}
-		// Add to set of async functions.
-		asyncFuncs[f] = &asyncFunc{}
-		asyncList = append(asyncList, f)
-
-		// Add all callees to the worklist.
-		for _, use := range getUses(f) {
-			if use.IsConstant() && use.Opcode() == llvm.PtrToInt {
-				for _, call := range getUses(use) {
-					if call.IsACallInst().IsNil() || call.CalledValue().Name() != "runtime.makeGoroutine" {
-						return false, errors.New("async function " + f.Name() + " incorrectly used in ptrtoint, expected runtime.makeGoroutine")
-					}
-				}
-				// This is a go statement. Do not mark the parent as async, as
-				// starting a goroutine is not a blocking operation.
-				continue
-			}
-			if use.IsConstant() && use.Opcode() == llvm.BitCast {
-				// Not sure why this const bitcast is here but as long as it
-				// has no uses it can be ignored, I guess?
-				// I think it was created for the runtime.isnil check but
-				// somehow wasn't removed when all these checks are removed.
-				if len(getUses(use)) == 0 {
-					continue
-				}
-			}
-			if use.IsACallInst().IsNil() {
-				// Not a call instruction. Maybe a store to a global? In any
-				// case, this requires support for async calls across function
-				// pointers which is not yet supported.
-				return false, errors.New("async function " + f.Name() + " used as function pointer")
-			}
-			parent := use.InstructionParent().Parent()
-			for i := 0; i < use.OperandsCount()-1; i++ {
-				if use.Operand(i) == f {
-					return false, errors.New("async function " + f.Name() + " used as function pointer in " + parent.Name())
-				}
-			}
-			worklist = append(worklist, parent)
-		}
-	}
-
-	// Check whether a scheduler is needed.
-	makeGoroutine := c.mod.NamedFunction("runtime.makeGoroutine")
-	if c.GOOS == "js" && strings.HasPrefix(c.Triple, "wasm") {
-		// JavaScript always needs a scheduler, as in general no blocking
-		// operations are possible. Blocking operations block the browser UI,
-		// which is very bad.
-		needsScheduler = true
-	} else {
-		// Only use a scheduler when an async goroutine is started. When the
-		// goroutine is not async (does not do any blocking operation), no
-		// scheduler is necessary as it can be called directly.
-		for _, use := range getUses(makeGoroutine) {
-			// Input param must be const ptrtoint of function.
-			ptrtoint := use.Operand(0)
-			if !ptrtoint.IsConstant() || ptrtoint.Opcode() != llvm.PtrToInt {
-				panic("expected const ptrtoint operand of runtime.makeGoroutine")
-			}
-			goroutine := ptrtoint.Operand(0)
-			if _, ok := asyncFuncs[goroutine]; ok {
-				needsScheduler = true
-				break
-			}
-		}
-	}
-
-	if !needsScheduler {
-		// No scheduler is needed. Do not transform all functions here.
-		// However, make sure that all go calls (which are all non-async) are
-		// transformed into regular calls.
-		return false, c.lowerMakeGoroutineCalls()
-	}
-
-	// Create a few LLVM intrinsics for coroutine support.
-
-	coroIdType := llvm.FunctionType(c.ctx.TokenType(), []llvm.Type{c.ctx.Int32Type(), c.i8ptrType, c.i8ptrType, c.i8ptrType}, false)
-	coroIdFunc := llvm.AddFunction(c.mod, "llvm.coro.id", coroIdType)
-
-	coroSizeType := llvm.FunctionType(c.ctx.Int32Type(), nil, false)
-	coroSizeFunc := llvm.AddFunction(c.mod, "llvm.coro.size.i32", coroSizeType)
-
-	coroBeginType := llvm.FunctionType(c.i8ptrType, []llvm.Type{c.ctx.TokenType(), c.i8ptrType}, false)
-	coroBeginFunc := llvm.AddFunction(c.mod, "llvm.coro.begin", coroBeginType)
-
-	coroSuspendType := llvm.FunctionType(c.ctx.Int8Type(), []llvm.Type{c.ctx.TokenType(), c.ctx.Int1Type()}, false)
-	coroSuspendFunc := llvm.AddFunction(c.mod, "llvm.coro.suspend", coroSuspendType)
-
-	coroEndType := llvm.FunctionType(c.ctx.Int1Type(), []llvm.Type{c.i8ptrType, c.ctx.Int1Type()}, false)
-	coroEndFunc := llvm.AddFunction(c.mod, "llvm.coro.end", coroEndType)
-
-	coroFreeType := llvm.FunctionType(c.i8ptrType, []llvm.Type{c.ctx.TokenType(), c.i8ptrType}, false)
-	coroFreeFunc := llvm.AddFunction(c.mod, "llvm.coro.free", coroFreeType)
-
-	// Transform all async functions into coroutines.
-	for _, f := range asyncList {
-		if f == sleep || f == deadlock || f == chanSend || f == chanRecv {
-			continue
-		}
-
-		frame := asyncFuncs[f]
-		frame.cleanupBlock = c.ctx.AddBasicBlock(f, "task.cleanup")
-		frame.suspendBlock = c.ctx.AddBasicBlock(f, "task.suspend")
-		frame.unreachableBlock = c.ctx.AddBasicBlock(f, "task.unreachable")
-
-		// Scan for async calls and return instructions that need to have
-		// suspend points inserted.
-		var asyncCalls []llvm.Value
-		var returns []llvm.Value
-		for bb := f.EntryBasicBlock(); !bb.IsNil(); bb = llvm.NextBasicBlock(bb) {
-			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 == deadlock || callee == chanSend || callee == chanRecv {
-						continue
-					}
-					asyncCalls = append(asyncCalls, inst)
-				} else if !inst.IsAReturnInst().IsNil() {
-					returns = append(returns, inst)
-				}
-			}
-		}
-
-		// Coroutine setup.
-		c.builder.SetInsertPointBefore(f.EntryBasicBlock().FirstInstruction())
-		taskState := c.builder.CreateAlloca(c.getLLVMRuntimeType("taskState"), "task.state")
-		stateI8 := c.builder.CreateBitCast(taskState, c.i8ptrType, "task.state.i8")
-		id := c.builder.CreateCall(coroIdFunc, []llvm.Value{
-			llvm.ConstInt(c.ctx.Int32Type(), 0, false),
-			stateI8,
-			llvm.ConstNull(c.i8ptrType),
-			llvm.ConstNull(c.i8ptrType),
-		}, "task.token")
-		size := c.builder.CreateCall(coroSizeFunc, nil, "task.size")
-		if c.targetData.TypeAllocSize(size.Type()) > c.targetData.TypeAllocSize(c.uintptrType) {
-			size = c.builder.CreateTrunc(size, c.uintptrType, "task.size.uintptr")
-		} else if c.targetData.TypeAllocSize(size.Type()) < c.targetData.TypeAllocSize(c.uintptrType) {
-			size = c.builder.CreateZExt(size, c.uintptrType, "task.size.uintptr")
-		}
-		data := c.createRuntimeCall("alloc", []llvm.Value{size}, "task.data")
-		if c.needsStackObjects() {
-			c.trackPointer(data)
-		}
-		frame.taskHandle = c.builder.CreateCall(coroBeginFunc, []llvm.Value{id, data}, "task.handle")
-
-		// Modify async calls so this function suspends right after the child
-		// returns, because the child is probably not finished yet. Wait until
-		// the child reactivates the parent.
-		for _, inst := range asyncCalls {
-			inst.SetOperand(inst.OperandsCount()-2, frame.taskHandle)
-
-			// Split this basic block.
-			await := c.splitBasicBlock(inst, llvm.NextBasicBlock(c.builder.GetInsertBlock()), "task.await")
-
-			// Allocate space for the return value.
-			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("setTaskStatePtr", []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),
-			}, "")
-			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 {
-			// These properties were added by the functionattrs pass. 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)
-			}
-
-			c.builder.SetInsertPointBefore(inst)
-
-			var parentHandle llvm.Value
-			if f.Linkage() == llvm.ExternalLinkage {
-				// Exported function.
-				// Note that getTaskStatePtr will panic if it is called with
-				// a nil pointer, so blocking exported functions that try to
-				// return anything will not work.
-				parentHandle = llvm.ConstPointerNull(c.i8ptrType)
-			} else {
-				parentHandle = f.LastParam()
-				if parentHandle.IsNil() || parentHandle.Name() != "parentHandle" {
-					// sanity check
-					panic("trying to make exported function async: " + f.Name())
-				}
-			}
-
-			// 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("getTaskStatePtr", []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.
-			// It would look like this is unnecessary, but if this
-			// suspend point is left out, it leads to undefined
-			// behavior somehow (with the unreachable instruction).
-			continuePoint := c.builder.CreateCall(coroSuspendFunc, []llvm.Value{
-				llvm.ConstNull(c.ctx.TokenType()),
-				llvm.ConstInt(c.ctx.Int1Type(), 0, false),
-			}, "ret")
-			sw := c.builder.CreateSwitch(continuePoint, frame.suspendBlock, 2)
-			sw.AddCase(llvm.ConstInt(c.ctx.Int8Type(), 0, false), frame.unreachableBlock)
-			sw.AddCase(llvm.ConstInt(c.ctx.Int8Type(), 1, false), frame.cleanupBlock)
-			inst.EraseFromParentAsInstruction()
-		}
-
-		// Coroutine cleanup. Free resources associated with this coroutine.
-		c.builder.SetInsertPointAtEnd(frame.cleanupBlock)
-		mem := c.builder.CreateCall(coroFreeFunc, []llvm.Value{id, frame.taskHandle}, "task.data.free")
-		c.createRuntimeCall("free", []llvm.Value{mem}, "")
-		c.builder.CreateBr(frame.suspendBlock)
-
-		// Coroutine suspend. A call to llvm.coro.suspend() will branch here.
-		c.builder.SetInsertPointAtEnd(frame.suspendBlock)
-		c.builder.CreateCall(coroEndFunc, []llvm.Value{frame.taskHandle, llvm.ConstInt(c.ctx.Int1Type(), 0, false)}, "unused")
-		returnType := f.Type().ElementType().ReturnType()
-		if returnType.TypeKind() == llvm.VoidTypeKind {
-			c.builder.CreateRetVoid()
-		} else {
-			c.builder.CreateRet(llvm.Undef(returnType))
-		}
-
-		// Coroutine exit. All final suspends (return instructions) will branch
-		// here.
-		c.builder.SetInsertPointAtEnd(frame.unreachableBlock)
-		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.
-		frame := asyncFuncs[sleepCall.InstructionParent().Parent()]
-		duration := sleepCall.Operand(0)
-
-		// Set task state to TASK_STATE_SLEEP and set the duration.
-		c.builder.SetInsertPointBefore(sleepCall)
-		c.createRuntimeCall("sleepTask", []llvm.Value{frame.taskHandle, duration}, "")
-
-		// Yield to scheduler.
-		continuePoint := c.builder.CreateCall(coroSuspendFunc, []llvm.Value{
-			llvm.ConstNull(c.ctx.TokenType()),
-			llvm.ConstInt(c.ctx.Int1Type(), 0, false),
-		}, "")
-		wakeup := c.splitBasicBlock(sleepCall, llvm.NextBasicBlock(c.builder.GetInsertBlock()), "task.wakeup")
-		c.builder.SetInsertPointBefore(sleepCall)
-		sw := c.builder.CreateSwitch(continuePoint, frame.suspendBlock, 2)
-		sw.AddCase(llvm.ConstInt(c.ctx.Int8Type(), 0, false), wakeup)
-		sw.AddCase(llvm.ConstInt(c.ctx.Int8Type(), 1, false), frame.cleanupBlock)
-		sleepCall.EraseFromParentAsInstruction()
-	}
-
-	// Transform calls to runtime.deadlock into coroutine suspends (without
-	// resume).
-	for _, deadlockCall := range getUses(deadlock) {
-		// deadlockCall must be a call instruction.
-		frame := asyncFuncs[deadlockCall.InstructionParent().Parent()]
-
-		// Exit coroutine.
-		c.builder.SetInsertPointBefore(deadlockCall)
-		continuePoint := c.builder.CreateCall(coroSuspendFunc, []llvm.Value{
-			llvm.ConstNull(c.ctx.TokenType()),
-			llvm.ConstInt(c.ctx.Int1Type(), 0, false),
-		}, "")
-		c.splitBasicBlock(deadlockCall, llvm.NextBasicBlock(c.builder.GetInsertBlock()), "task.wakeup.dead")
-		c.builder.SetInsertPointBefore(deadlockCall)
-		sw := c.builder.CreateSwitch(continuePoint, frame.suspendBlock, 2)
-		sw.AddCase(llvm.ConstInt(c.ctx.Int8Type(), 0, false), frame.unreachableBlock)
-		sw.AddCase(llvm.ConstInt(c.ctx.Int8Type(), 1, false), frame.cleanupBlock)
-		deadlockCall.EraseFromParentAsInstruction()
-	}
-
-	// Transform calls to runtime.chanSend into channel send operations.
-	for _, sendOp := range getUses(chanSend) {
-		// sendOp must be a call instruction.
-		frame := asyncFuncs[sendOp.InstructionParent().Parent()]
-
-		// Yield to scheduler.
-		c.builder.SetInsertPointBefore(llvm.NextInstruction(sendOp))
-		continuePoint := c.builder.CreateCall(coroSuspendFunc, []llvm.Value{
-			llvm.ConstNull(c.ctx.TokenType()),
-			llvm.ConstInt(c.ctx.Int1Type(), 0, false),
-		}, "")
-		sw := c.builder.CreateSwitch(continuePoint, frame.suspendBlock, 2)
-		wakeup := c.splitBasicBlock(sw, llvm.NextBasicBlock(c.builder.GetInsertBlock()), "task.sent")
-		sw.AddCase(llvm.ConstInt(c.ctx.Int8Type(), 0, false), wakeup)
-		sw.AddCase(llvm.ConstInt(c.ctx.Int8Type(), 1, false), frame.cleanupBlock)
-	}
-
-	// Transform calls to runtime.chanRecv into channel receive operations.
-	for _, recvOp := range getUses(chanRecv) {
-		// recvOp must be a call instruction.
-		frame := asyncFuncs[recvOp.InstructionParent().Parent()]
-
-		// Yield to scheduler.
-		c.builder.SetInsertPointBefore(llvm.NextInstruction(recvOp))
-		continuePoint := c.builder.CreateCall(coroSuspendFunc, []llvm.Value{
-			llvm.ConstNull(c.ctx.TokenType()),
-			llvm.ConstInt(c.ctx.Int1Type(), 0, false),
-		}, "")
-		sw := c.builder.CreateSwitch(continuePoint, frame.suspendBlock, 2)
-		wakeup := c.splitBasicBlock(sw, llvm.NextBasicBlock(c.builder.GetInsertBlock()), "task.received")
-		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)
-	}
-
-	return true, c.lowerMakeGoroutineCalls()
-}
-
-// Lower runtime.makeGoroutine calls to regular call instructions. This is done
-// after the regular goroutine transformations. The started goroutines are
-// either non-blocking (in which case they can be called directly) or blocking,
-// in which case they will ask the scheduler themselves to be rescheduled.
-func (c *Compiler) lowerMakeGoroutineCalls() error {
-	// The following Go code:
-	//   go startedGoroutine()
-	//
-	// Is translated to the following during IR construction, to preserve the
-	// fact that this function should be called as a new goroutine.
-	//   %0 = call i8* @runtime.makeGoroutine(i8* bitcast (void (i8*, i8*)* @main.startedGoroutine to i8*), i8* undef, i8* null)
-	//   %1 = bitcast i8* %0 to void (i8*, i8*)*
-	//   call void %1(i8* undef, i8* undef)
-	//
-	// This function rewrites it to a direct call:
-	//   call void @main.startedGoroutine(i8* undef, i8* null)
-
-	makeGoroutine := c.mod.NamedFunction("runtime.makeGoroutine")
-	for _, goroutine := range getUses(makeGoroutine) {
-		ptrtointIn := goroutine.Operand(0)
-		origFunc := ptrtointIn.Operand(0)
-		uses := getUses(goroutine)
-		if len(uses) != 1 || uses[0].IsAIntToPtrInst().IsNil() {
-			return errors.New("expected exactly 1 inttoptr use of runtime.makeGoroutine")
-		}
-		inttoptrOut := uses[0]
-		uses = getUses(inttoptrOut)
-		if len(uses) != 1 || uses[0].IsACallInst().IsNil() {
-			return errors.New("expected exactly 1 call use of runtime.makeGoroutine bitcast")
-		}
-		realCall := uses[0]
-
-		// Create call instruction.
-		var params []llvm.Value
-		for i := 0; i < realCall.OperandsCount()-1; i++ {
-			params = append(params, realCall.Operand(i))
-		}
-		params[len(params)-1] = llvm.ConstPointerNull(c.i8ptrType) // parent coroutine handle (must be nil)
-		c.builder.SetInsertPointBefore(realCall)
-		c.builder.CreateCall(origFunc, params, "")
-		realCall.EraseFromParentAsInstruction()
-		inttoptrOut.EraseFromParentAsInstruction()
-		goroutine.EraseFromParentAsInstruction()
-	}
-
-	return nil
-}

compiler/goroutine.go

@@ -3,32 +3,137 @@ package compiler
 // This file implements the 'go' keyword to start a new goroutine. See
 // goroutine-lowering.go for more details.
 
-import "tinygo.org/x/go-llvm"
+import (
+	"go/token"
+	"go/types"
 
-// emitStartGoroutine starts a new goroutine with the provided function pointer
-// and parameters.
-//
-// Because a go statement doesn't return anything, return undef.
-func (c *Compiler) emitStartGoroutine(funcPtr llvm.Value, params []llvm.Value) llvm.Value {
-	switch c.selectScheduler() {
-	case "tasks":
-		paramBundle := c.emitPointerPack(params)
-		paramBundle = c.builder.CreatePtrToInt(paramBundle, c.uintptrType, "")
+	"github.com/tinygo-org/tinygo/compiler/llvmutil"
+	"golang.org/x/tools/go/ssa"
+	"tinygo.org/x/go-llvm"
+)
 
-		calleeValue := c.createGoroutineStartWrapper(funcPtr)
-		c.createRuntimeCall("startGoroutine", []llvm.Value{calleeValue, paramBundle}, "")
+// createGo emits code to start a new goroutine.
+func (b *builder) createGo(instr *ssa.Go) {
+	// Get all function parameters to pass to the goroutine.
+	var params []llvm.Value
+	for _, param := range instr.Call.Args {
+		params = append(params, b.getValue(param))
+	}
+
+	var prefix string
+	var funcPtr llvm.Value
+	hasContext := false
+	if callee := instr.Call.StaticCallee(); callee != nil {
+		// Static callee is known. This makes it easier to start a new
+		// goroutine.
+		var context llvm.Value
+		switch value := instr.Call.Value.(type) {
+		case *ssa.Function:
+			// Goroutine call is regular function call. No context is necessary.
+			if b.Scheduler == "coroutines" {
+				// The context parameter is assumed to be always present in the
+				// coroutines scheduler.
+				context = llvm.Undef(b.i8ptrType)
+			}
+		case *ssa.MakeClosure:
+			// A goroutine call on a func value, but the callee is trivial to find. For
+			// example: immediately applied functions.
+			funcValue := b.getValue(value)
+			context = b.extractFuncContext(funcValue)
+		default:
+			panic("StaticCallee returned an unexpected value")
+		}
+		if !context.IsNil() {
+			params = append(params, context) // context parameter
+			hasContext = true
+		}
+		funcPtr = b.getFunction(callee)
+	} else if builtin, ok := instr.Call.Value.(*ssa.Builtin); ok {
+		// We cheat. None of the builtins do any long or blocking operation, so
+		// we might as well run these builtins right away without the program
+		// noticing the difference.
+		// Possible exceptions:
+		//   - copy: this is a possibly long operation, but not a blocking
+		//     operation. Semantically it makes no difference to run it right
+		//     away (not in a goroutine). However, in practice it makes no sense
+		//     to run copy in a goroutine as there is no way to (safely) know
+		//     when it is finished.
+		//   - panic: the error message would appear in the parent goroutine.
+		//     But because `go panic("err")` would halt the program anyway
+		//     (there is no recover), panicking right away would give the same
+		//     behavior as creating a goroutine, switching the scheduler to that
+		//     goroutine, and panicking there. So this optimization seems
+		//     correct.
+		//   - recover: because it runs in a new goroutine, it is never a
+		//     deferred function. Thus this is a no-op.
+		if builtin.Name() == "recover" {
+			// This is a no-op, even in a deferred function:
+			//   go recover()
+			return
+		}
+		var argTypes []types.Type
+		var argValues []llvm.Value
+		for _, arg := range instr.Call.Args {
+			argTypes = append(argTypes, arg.Type())
+			argValues = append(argValues, b.getValue(arg))
+		}
+		b.createBuiltin(argTypes, argValues, builtin.Name(), instr.Pos())
+		return
+	} else if !instr.Call.IsInvoke() {
+		// This is a function pointer.
+		// At the moment, two extra params are passed to the newly started
+		// goroutine:
+		//   * The function context, for closures.
+		//   * The function pointer (for tasks).
+		var context llvm.Value
+		funcPtr, context = b.decodeFuncValue(b.getValue(instr.Call.Value), instr.Call.Value.Type().Underlying().(*types.Signature))
+		params = append(params, context) // context parameter
+		hasContext = true
+		switch b.Scheduler {
+		case "none", "coroutines":
+			// There are no additional parameters needed for the goroutine start operation.
+		case "tasks":
+			// Add the function pointer as a parameter to start the goroutine.
+			params = append(params, funcPtr)
+		default:
+			panic("unknown scheduler type")
+		}
+		prefix = b.fn.RelString(nil)
+	} else {
+		b.addError(instr.Pos(), "todo: go on interface call")
+		return
+	}
+
+	paramBundle := b.emitPointerPack(params)
+	var callee, stackSize llvm.Value
+	switch b.Scheduler {
+	case "none", "tasks":
+		callee = b.createGoroutineStartWrapper(funcPtr, prefix, hasContext, instr.Pos())
+		if b.AutomaticStackSize {
+			// The stack size is not known until after linking. Call a dummy
+			// function that will be replaced with a load from a special ELF
+			// section that contains the stack size (and is modified after
+			// linking).
+			stackSizeFn := b.getFunction(b.program.ImportedPackage("internal/task").Members["getGoroutineStackSize"].(*ssa.Function))
+			stackSize = b.createCall(stackSizeFn, []llvm.Value{callee, llvm.Undef(b.i8ptrType), llvm.Undef(b.i8ptrType)}, "stacksize")
+		} else {
+			// The stack size is fixed at compile time. By emitting it here as a
+			// constant, it can be optimized.
+			if b.Scheduler == "tasks" && b.DefaultStackSize == 0 {
+				b.addError(instr.Pos(), "default stack size for goroutines is not set")
+			}
+			stackSize = llvm.ConstInt(b.uintptrType, b.DefaultStackSize, false)
+		}
 	case "coroutines":
-		// We roundtrip through runtime.makeGoroutine as a signal (to find these
-		// calls) and to break any optimizations LLVM will try to do: they are
-		// invalid if we called this as a regular function to be updated later.
-		calleeValue := c.builder.CreatePtrToInt(funcPtr, c.uintptrType, "")
-		calleeValue = c.createRuntimeCall("makeGoroutine", []llvm.Value{calleeValue}, "")
-		calleeValue = c.builder.CreateIntToPtr(calleeValue, funcPtr.Type(), "")
-		c.createCall(calleeValue, params, "")
+		callee = b.CreatePtrToInt(funcPtr, b.uintptrType, "")
+		// There is no goroutine stack size: coroutines are used instead of
+		// stacks.
+		stackSize = llvm.Undef(b.uintptrType)
 	default:
 		panic("unreachable")
 	}
-	return llvm.Undef(funcPtr.Type().ElementType().ReturnType())
+	start := b.getFunction(b.program.ImportedPackage("internal/task").Members["start"].(*ssa.Function))
+	b.createCall(start, []llvm.Value{callee, paramBundle, stackSize, llvm.Undef(b.i8ptrType), llvm.ConstPointerNull(b.i8ptrType)}, "")
 }
 
 // createGoroutineStartWrapper creates a wrapper for the task-based
@@ -49,36 +154,70 @@ func (c *Compiler) emitStartGoroutine(funcPtr llvm.Value, params []llvm.Value) l
 // allows a single (pointer) argument to the newly started goroutine. Also, it
 // ignores the return value because newly started goroutines do not have a
 // return value.
-func (c *Compiler) createGoroutineStartWrapper(fn llvm.Value) llvm.Value {
+//
+// The hasContext parameter indicates whether the context parameter (the second
+// to last parameter of the function) is used for this wrapper. If hasContext is
+// false, the parameter bundle is assumed to have no context parameter and undef
+// is passed instead.
+func (c *compilerContext) createGoroutineStartWrapper(fn llvm.Value, prefix string, hasContext bool, pos token.Pos) llvm.Value {
 	var wrapper llvm.Value
 
+	builder := c.ctx.NewBuilder()
+	defer builder.Dispose()
+
 	if !fn.IsAFunction().IsNil() {
 		// See whether this wrapper has already been created. If so, return it.
 		name := fn.Name()
 		wrapper = c.mod.NamedFunction(name + "$gowrapper")
 		if !wrapper.IsNil() {
-			return c.builder.CreatePtrToInt(wrapper, c.uintptrType, "")
+			return llvm.ConstPtrToInt(wrapper, c.uintptrType)
 		}
 
-		// Save the current position in the IR builder.
-		currentBlock := c.builder.GetInsertBlock()
-		defer c.builder.SetInsertPointAtEnd(currentBlock)
-
 		// Create the wrapper.
 		wrapperType := llvm.FunctionType(c.ctx.VoidType(), []llvm.Type{c.i8ptrType}, false)
 		wrapper = llvm.AddFunction(c.mod, name+"$gowrapper", wrapperType)
-		wrapper.SetLinkage(llvm.PrivateLinkage)
+		wrapper.SetLinkage(llvm.LinkOnceODRLinkage)
 		wrapper.SetUnnamedAddr(true)
+		wrapper.AddAttributeAtIndex(-1, c.ctx.CreateStringAttribute("tinygo-gowrapper", name))
 		entry := c.ctx.AddBasicBlock(wrapper, "entry")
-		c.builder.SetInsertPointAtEnd(entry)
+		builder.SetInsertPointAtEnd(entry)
+
+		if c.Debug {
+			pos := c.program.Fset.Position(pos)
+			diFuncType := c.dibuilder.CreateSubroutineType(llvm.DISubroutineType{
+				File:       c.getDIFile(pos.Filename),
+				Parameters: nil, // do not show parameters in debugger
+				Flags:      0,   // ?
+			})
+			difunc := c.dibuilder.CreateFunction(c.getDIFile(pos.Filename), llvm.DIFunction{
+				Name:         "<goroutine wrapper>",
+				File:         c.getDIFile(pos.Filename),
+				Line:         pos.Line,
+				Type:         diFuncType,
+				LocalToUnit:  true,
+				IsDefinition: true,
+				ScopeLine:    0,
+				Flags:        llvm.FlagPrototyped,
+				Optimized:    true,
+			})
+			wrapper.SetSubprogram(difunc)
+			builder.SetCurrentDebugLocation(uint(pos.Line), uint(pos.Column), difunc, llvm.Metadata{})
+		}
 
 		// Create the list of params for the call.
 		paramTypes := fn.Type().ElementType().ParamTypes()
-		params := c.emitPointerUnpack(wrapper.Param(0), paramTypes[:len(paramTypes)-2])
-		params = append(params, llvm.Undef(c.i8ptrType), llvm.ConstPointerNull(c.i8ptrType))
+		paramTypes = paramTypes[:len(paramTypes)-1] // strip parentHandle parameter
+		if !hasContext {
+			paramTypes = paramTypes[:len(paramTypes)-1] // strip context parameter
+		}
+		params := llvmutil.EmitPointerUnpack(builder, c.mod, wrapper.Param(0), paramTypes)
+		if !hasContext {
+			params = append(params, llvm.Undef(c.i8ptrType)) // add dummy context parameter
+		}
+		params = append(params, llvm.Undef(c.i8ptrType)) // add dummy parentHandle parameter
 
 		// Create the call.
-		c.builder.CreateCall(fn, params, "")
+		builder.CreateCall(fn, params, "")
 
 	} else {
 		// For a function pointer like this:
@@ -98,38 +237,60 @@ func (c *Compiler) createGoroutineStartWrapper(fn llvm.Value) llvm.Value {
 		// With a bit of luck, identical wrapper functions like these can be
 		// merged into one.
 
-		// Save the current position in the IR builder.
-		currentBlock := c.builder.GetInsertBlock()
-		defer c.builder.SetInsertPointAtEnd(currentBlock)
-
 		// Create the wrapper.
 		wrapperType := llvm.FunctionType(c.ctx.VoidType(), []llvm.Type{c.i8ptrType}, false)
-		wrapper = llvm.AddFunction(c.mod, ".gowrapper", wrapperType)
-		wrapper.SetLinkage(llvm.InternalLinkage)
+		wrapper = llvm.AddFunction(c.mod, prefix+".gowrapper", wrapperType)
+		wrapper.SetLinkage(llvm.LinkOnceODRLinkage)
 		wrapper.SetUnnamedAddr(true)
+		wrapper.AddAttributeAtIndex(-1, c.ctx.CreateStringAttribute("tinygo-gowrapper", ""))
 		entry := c.ctx.AddBasicBlock(wrapper, "entry")
-		c.builder.SetInsertPointAtEnd(entry)
+		builder.SetInsertPointAtEnd(entry)
+
+		if c.Debug {
+			pos := c.program.Fset.Position(pos)
+			diFuncType := c.dibuilder.CreateSubroutineType(llvm.DISubroutineType{
+				File:       c.getDIFile(pos.Filename),
+				Parameters: nil, // do not show parameters in debugger
+				Flags:      0,   // ?
+			})
+			difunc := c.dibuilder.CreateFunction(c.getDIFile(pos.Filename), llvm.DIFunction{
+				Name:         "<goroutine wrapper>",
+				File:         c.getDIFile(pos.Filename),
+				Line:         pos.Line,
+				Type:         diFuncType,
+				LocalToUnit:  true,
+				IsDefinition: true,
+				ScopeLine:    0,
+				Flags:        llvm.FlagPrototyped,
+				Optimized:    true,
+			})
+			wrapper.SetSubprogram(difunc)
+			builder.SetCurrentDebugLocation(uint(pos.Line), uint(pos.Column), difunc, llvm.Metadata{})
+		}
 
 		// Get the list of parameters, with the extra parameters at the end.
 		paramTypes := fn.Type().ElementType().ParamTypes()
 		paramTypes[len(paramTypes)-1] = fn.Type() // the last element is the function pointer
-		params := c.emitPointerUnpack(wrapper.Param(0), paramTypes)
+		params := llvmutil.EmitPointerUnpack(builder, c.mod, wrapper.Param(0), paramTypes)
 
 		// Get the function pointer.
 		fnPtr := params[len(params)-1]
 
-		// Ignore the last param, which isn't used anymore.
-		// TODO: avoid this extra "parent handle" parameter in most functions.
+		// The last parameter in the packed object has somewhat of a dual role.
+		// Inside the parameter bundle it's the function pointer, stored right
+		// after the context pointer. But in the IR call instruction, it's the
+		// parentHandle function that's always undef outside of the coroutines
+		// scheduler. Thus, make the parameter undef here.
 		params[len(params)-1] = llvm.Undef(c.i8ptrType)
 
 		// Create the call.
-		c.builder.CreateCall(fnPtr, params, "")
+		builder.CreateCall(fnPtr, params, "")
 	}
 
 	// Finish the function. Every basic block must end in a terminator, and
 	// because goroutines never return a value we can simply return void.
-	c.builder.CreateRetVoid()
+	builder.CreateRetVoid()
 
 	// Return a ptrtoint of the wrapper, not the function itself.
-	return c.builder.CreatePtrToInt(wrapper, c.uintptrType, "")
+	return builder.CreatePtrToInt(wrapper, c.uintptrType, "")
 }

compiler/inlineasm.go

@@ -13,27 +13,6 @@ import (
 	"tinygo.org/x/go-llvm"
 )
 
-// This is a compiler builtin, which reads the given register by name:
-//
-//     func ReadRegister(name string) uintptr
-//
-// The register name must be a constant, for example "sp".
-func (c *Compiler) emitReadRegister(name string, args []ssa.Value) (llvm.Value, error) {
-	fnType := llvm.FunctionType(c.uintptrType, []llvm.Type{}, false)
-	regname := constant.StringVal(args[0].(*ssa.Const).Value)
-	var asm string
-	switch name {
-	case "device/arm.ReadRegister":
-		asm = "mov $0, " + regname
-	case "device/riscv.ReadRegister":
-		asm = "mv $0, " + regname
-	default:
-		panic("unknown architecture")
-	}
-	target := llvm.InlineAsm(fnType, asm, "=r", false, false, 0)
-	return c.builder.CreateCall(target, nil, ""), nil
-}
-
 // This is a compiler builtin, which emits a piece of inline assembly with no
 // operands or return values. It is useful for trivial instructions, like wfi in
 // ARM or sleep in AVR.
@@ -41,18 +20,18 @@ func (c *Compiler) emitReadRegister(name string, args []ssa.Value) (llvm.Value,
 //     func Asm(asm string)
 //
 // The provided assembly must be a constant.
-func (c *Compiler) emitAsm(args []ssa.Value) (llvm.Value, error) {
+func (b *builder) createInlineAsm(args []ssa.Value) (llvm.Value, error) {
 	// Magic function: insert inline assembly instead of calling it.
-	fnType := llvm.FunctionType(c.ctx.VoidType(), []llvm.Type{}, false)
+	fnType := llvm.FunctionType(b.ctx.VoidType(), []llvm.Type{}, false)
 	asm := constant.StringVal(args[0].(*ssa.Const).Value)
 	target := llvm.InlineAsm(fnType, asm, "", true, false, 0)
-	return c.builder.CreateCall(target, nil, ""), nil
+	return b.CreateCall(target, nil, ""), nil
 }
 
 // This is a compiler builtin, which allows assembly to be called in a flexible
 // way.
 //
-//     func AsmFull(asm string, regs map[string]interface{})
+//     func AsmFull(asm string, regs map[string]interface{}) uintptr
 //
 // The asm parameter must be a constant string. The regs parameter must be
 // provided immediately. For example:
@@ -63,27 +42,27 @@ func (c *Compiler) emitAsm(args []ssa.Value) (llvm.Value, error) {
 //             "value":  1
 //             "result": &dest,
 //         })
-func (c *Compiler) emitAsmFull(frame *Frame, instr *ssa.CallCommon) (llvm.Value, error) {
+func (b *builder) createInlineAsmFull(instr *ssa.CallCommon) (llvm.Value, error) {
 	asmString := constant.StringVal(instr.Args[0].(*ssa.Const).Value)
 	registers := map[string]llvm.Value{}
-	registerMap := instr.Args[1].(*ssa.MakeMap)
-	for _, r := range *registerMap.Referrers() {
-		switch r := r.(type) {
-		case *ssa.DebugRef:
-			// ignore
-		case *ssa.MapUpdate:
-			if r.Block() != registerMap.Block() {
-				return llvm.Value{}, c.makeError(instr.Pos(), "register value map must be created in the same basic block")
+	if registerMap, ok := instr.Args[1].(*ssa.MakeMap); ok {
+		for _, r := range *registerMap.Referrers() {
+			switch r := r.(type) {
+			case *ssa.DebugRef:
+				// ignore
+			case *ssa.MapUpdate:
+				if r.Block() != registerMap.Block() {
+					return llvm.Value{}, b.makeError(instr.Pos(), "register value map must be created in the same basic block")
+				}
+				key := constant.StringVal(r.Key.(*ssa.Const).Value)
+				registers[key] = b.getValue(r.Value.(*ssa.MakeInterface).X)
+			case *ssa.Call:
+				if r.Common() == instr {
+					break
+				}
+			default:
+				return llvm.Value{}, b.makeError(instr.Pos(), "don't know how to handle argument to inline assembly: "+r.String())
 			}
-			key := constant.StringVal(r.Key.(*ssa.Const).Value)
-			//println("value:", r.Value.(*ssa.MakeInterface).X.String())
-			registers[key] = c.getValue(frame, r.Value.(*ssa.MakeInterface).X)
-		case *ssa.Call:
-			if r.Common() == instr {
-				break
-			}
-		default:
-			return llvm.Value{}, c.makeError(instr.Pos(), "don't know how to handle argument to inline assembly: "+r.String())
 		}
 	}
 	// TODO: handle dollar signs in asm string
@@ -92,13 +71,22 @@ func (c *Compiler) emitAsmFull(frame *Frame, instr *ssa.CallCommon) (llvm.Value,
 	argTypes := []llvm.Type{}
 	args := []llvm.Value{}
 	constraints := []string{}
+	hasOutput := false
+	asmString = regexp.MustCompile("\\{\\}").ReplaceAllStringFunc(asmString, func(s string) string {
+		hasOutput = true
+		return "$0"
+	})
+	if hasOutput {
+		constraints = append(constraints, "=&r")
+		registerNumbers[""] = 0
+	}
 	asmString = regexp.MustCompile("\\{[a-zA-Z]+\\}").ReplaceAllStringFunc(asmString, func(s string) string {
 		// TODO: skip strings like {r4} etc. that look like ARM push/pop
 		// instructions.
 		name := s[1 : len(s)-1]
 		if _, ok := registers[name]; !ok {
 			if err == nil {
-				err = c.makeError(instr.Pos(), "unknown register name: "+name)
+				err = b.makeError(instr.Pos(), "unknown register name: "+name)
 			}
 			return s
 		}
@@ -112,7 +100,7 @@ func (c *Compiler) emitAsmFull(frame *Frame, instr *ssa.CallCommon) (llvm.Value,
 			case llvm.PointerTypeKind:
 				constraints = append(constraints, "*m")
 			default:
-				err = c.makeError(instr.Pos(), "unknown type in inline assembly for value: "+name)
+				err = b.makeError(instr.Pos(), "unknown type in inline assembly for value: "+name)
 				return s
 			}
 		}
@@ -121,9 +109,21 @@ func (c *Compiler) emitAsmFull(frame *Frame, instr *ssa.CallCommon) (llvm.Value,
 	if err != nil {
 		return llvm.Value{}, err
 	}
-	fnType := llvm.FunctionType(c.ctx.VoidType(), argTypes, false)
+	var outputType llvm.Type
+	if hasOutput {
+		outputType = b.uintptrType
+	} else {
+		outputType = b.ctx.VoidType()
+	}
+	fnType := llvm.FunctionType(outputType, argTypes, false)
 	target := llvm.InlineAsm(fnType, asmString, strings.Join(constraints, ","), true, false, 0)
-	return c.builder.CreateCall(target, args, ""), nil
+	result := b.CreateCall(target, args, "")
+	if hasOutput {
+		return result, nil
+	} else {
+		// Make sure we return something valid.
+		return llvm.ConstInt(b.uintptrType, 0, false), nil
+	}
 }
 
 // This is a compiler builtin which emits an inline SVCall instruction. It can
@@ -137,7 +137,7 @@ func (c *Compiler) emitAsmFull(frame *Frame, instr *ssa.CallCommon) (llvm.Value,
 //
 // The num parameter must be a constant. All other parameters may be any scalar
 // value supported by LLVM inline assembly.
-func (c *Compiler) emitSVCall(frame *Frame, args []ssa.Value) (llvm.Value, error) {
+func (b *builder) emitSVCall(args []ssa.Value) (llvm.Value, error) {
 	num, _ := constant.Uint64Val(args[0].(*ssa.Const).Value)
 	llvmArgs := []llvm.Value{}
 	argTypes := []llvm.Type{}
@@ -150,7 +150,7 @@ func (c *Compiler) emitSVCall(frame *Frame, args []ssa.Value) (llvm.Value, error
 		} else {
 			constraints += ",{r" + strconv.Itoa(i) + "}"
 		}
-		llvmValue := c.getValue(frame, arg)
+		llvmValue := b.getValue(arg)
 		llvmArgs = append(llvmArgs, llvmValue)
 		argTypes = append(argTypes, llvmValue.Type())
 	}
@@ -158,7 +158,90 @@ func (c *Compiler) emitSVCall(frame *Frame, args []ssa.Value) (llvm.Value, error
 	// clobbered. r0 is used as an output register so doesn't have to be
 	// marked as clobbered.
 	constraints += ",~{r1},~{r2},~{r3}"
-	fnType := llvm.FunctionType(c.uintptrType, argTypes, false)
+	fnType := llvm.FunctionType(b.uintptrType, argTypes, false)
 	target := llvm.InlineAsm(fnType, asm, constraints, true, false, 0)
-	return c.builder.CreateCall(target, llvmArgs, ""), nil
+	return b.CreateCall(target, llvmArgs, ""), nil
+}
+
+// This is a compiler builtin which emits an inline SVCall instruction. It can
+// be one of:
+//
+//     func SVCall0(num uintptr) uintptr
+//     func SVCall1(num uintptr, a1 interface{}) uintptr
+//     func SVCall2(num uintptr, a1, a2 interface{}) uintptr
+//     func SVCall3(num uintptr, a1, a2, a3 interface{}) uintptr
+//     func SVCall4(num uintptr, a1, a2, a3, a4 interface{}) uintptr
+//
+// The num parameter must be a constant. All other parameters may be any scalar
+// value supported by LLVM inline assembly.
+// Same as emitSVCall but for AArch64
+func (b *builder) emitSV64Call(args []ssa.Value) (llvm.Value, error) {
+	num, _ := constant.Uint64Val(args[0].(*ssa.Const).Value)
+	llvmArgs := []llvm.Value{}
+	argTypes := []llvm.Type{}
+	asm := "svc #" + strconv.FormatUint(num, 10)
+	constraints := "={x0}"
+	for i, arg := range args[1:] {
+		arg = arg.(*ssa.MakeInterface).X
+		if i == 0 {
+			constraints += ",0"
+		} else {
+			constraints += ",{x" + strconv.Itoa(i) + "}"
+		}
+		llvmValue := b.getValue(arg)
+		llvmArgs = append(llvmArgs, llvmValue)
+		argTypes = append(argTypes, llvmValue.Type())
+	}
+	// Implement the ARM64 calling convention by marking x1-x7 as
+	// clobbered. x0 is used as an output register so doesn't have to be
+	// marked as clobbered.
+	constraints += ",~{x1},~{x2},~{x3},~{x4},~{x5},~{x6},~{x7}"
+	fnType := llvm.FunctionType(b.uintptrType, argTypes, false)
+	target := llvm.InlineAsm(fnType, asm, constraints, true, false, 0)
+	return b.CreateCall(target, llvmArgs, ""), nil
+}
+
+// This is a compiler builtin which emits CSR instructions. It can be one of:
+//
+//     func (csr CSR) Get() uintptr
+//     func (csr CSR) Set(uintptr)
+//     func (csr CSR) SetBits(uintptr) uintptr
+//     func (csr CSR) ClearBits(uintptr) uintptr
+//
+// The csr parameter (method receiver) must be a constant. Other parameter can
+// be any value.
+func (b *builder) emitCSROperation(call *ssa.CallCommon) (llvm.Value, error) {
+	csrConst, ok := call.Args[0].(*ssa.Const)
+	if !ok {
+		return llvm.Value{}, b.makeError(call.Pos(), "CSR must be constant")
+	}
+	csr := csrConst.Uint64()
+	switch name := call.StaticCallee().Name(); name {
+	case "Get":
+		// Note that this instruction may have side effects, and thus must be
+		// marked as such.
+		fnType := llvm.FunctionType(b.uintptrType, nil, false)
+		asm := fmt.Sprintf("csrr $0, %d", csr)
+		target := llvm.InlineAsm(fnType, asm, "=r", true, false, 0)
+		return b.CreateCall(target, nil, ""), nil
+	case "Set":
+		fnType := llvm.FunctionType(b.ctx.VoidType(), []llvm.Type{b.uintptrType}, false)
+		asm := fmt.Sprintf("csrw %d, $0", csr)
+		target := llvm.InlineAsm(fnType, asm, "r", true, false, 0)
+		return b.CreateCall(target, []llvm.Value{b.getValue(call.Args[1])}, ""), nil
+	case "SetBits":
+		// Note: it may be possible to optimize this to csrrsi in many cases.
+		fnType := llvm.FunctionType(b.uintptrType, []llvm.Type{b.uintptrType}, false)
+		asm := fmt.Sprintf("csrrs $0, %d, $1", csr)
+		target := llvm.InlineAsm(fnType, asm, "=r,r", true, false, 0)
+		return b.CreateCall(target, []llvm.Value{b.getValue(call.Args[1])}, ""), nil
+	case "ClearBits":
+		// Note: it may be possible to optimize this to csrrci in many cases.
+		fnType := llvm.FunctionType(b.uintptrType, []llvm.Type{b.uintptrType}, false)
+		asm := fmt.Sprintf("csrrc $0, %d, $1", csr)
+		target := llvm.InlineAsm(fnType, asm, "=r,r", true, false, 0)
+		return b.CreateCall(target, []llvm.Value{b.getValue(call.Args[1])}, ""), nil
+	default:
+		return llvm.Value{}, b.makeError(call.Pos(), "unknown CSR operation: "+name)
+	}
 }

compiler/interface.go

@@ -11,40 +11,30 @@ import (
 	"strconv"
 	"strings"
 
-	"github.com/tinygo-org/tinygo/ir"
 	"golang.org/x/tools/go/ssa"
 	"tinygo.org/x/go-llvm"
 )
 
-// parseMakeInterface emits the LLVM IR for the *ssa.MakeInterface instruction.
+// createMakeInterface emits the LLVM IR for the *ssa.MakeInterface instruction.
 // It tries to put the type in the interface value, but if that's not possible,
 // it will do an allocation of the right size and put that in the interface
 // 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 {
-	itfValue := c.emitPointerPack([]llvm.Value{val})
-	itfTypeCodeGlobal := c.getTypeCode(typ)
-	itfMethodSetGlobal := c.getTypeMethodSet(typ)
-	itfConcreteTypeGlobal := c.mod.NamedGlobal("typeInInterface:" + itfTypeCodeGlobal.Name())
-	if itfConcreteTypeGlobal.IsNil() {
-		typeInInterface := c.getLLVMRuntimeType("typeInInterface")
-		itfConcreteTypeGlobal = llvm.AddGlobal(c.mod, typeInInterface, "typeInInterface:"+itfTypeCodeGlobal.Name())
-		itfConcreteTypeGlobal.SetInitializer(llvm.ConstNamedStruct(typeInInterface, []llvm.Value{itfTypeCodeGlobal, itfMethodSetGlobal}))
-		itfConcreteTypeGlobal.SetGlobalConstant(true)
-		itfConcreteTypeGlobal.SetLinkage(llvm.PrivateLinkage)
-	}
-	itfTypeCode := c.builder.CreatePtrToInt(itfConcreteTypeGlobal, c.uintptrType, "")
-	itf := llvm.Undef(c.getLLVMRuntimeType("_interface"))
-	itf = c.builder.CreateInsertValue(itf, itfTypeCode, 0, "")
-	itf = c.builder.CreateInsertValue(itf, itfValue, 1, "")
+// An interface value is a {typecode, value} tuple named runtime._interface.
+func (b *builder) createMakeInterface(val llvm.Value, typ types.Type, pos token.Pos) llvm.Value {
+	itfValue := b.emitPointerPack([]llvm.Value{val})
+	itfTypeCodeGlobal := b.getTypeCode(typ)
+	itfTypeCode := b.CreatePtrToInt(itfTypeCodeGlobal, b.uintptrType, "")
+	itf := llvm.Undef(b.getLLVMRuntimeType("_interface"))
+	itf = b.CreateInsertValue(itf, itfTypeCode, 0, "")
+	itf = b.CreateInsertValue(itf, itfValue, 1, "")
 	return itf
 }
 
 // getTypeCode returns a reference to a type code.
 // It returns a pointer to an external global which should be replaced with the
 // real type in the interface lowering pass.
-func (c *Compiler) getTypeCode(typ types.Type) llvm.Value {
+func (c *compilerContext) getTypeCode(typ types.Type) llvm.Value {
 	globalName := "reflect/types.type:" + getTypeCodeName(typ)
 	global := c.mod.NamedGlobal(globalName)
 	if global.IsNil() {
@@ -55,6 +45,8 @@ func (c *Compiler) getTypeCode(typ types.Type) llvm.Value {
 		// reflect lowering simpler.
 		var references llvm.Value
 		var length int64
+		var methodSet llvm.Value
+		var ptrTo llvm.Value
 		switch typ := typ.(type) {
 		case *types.Named:
 			references = c.getTypeCode(typ.Underlying())
@@ -71,18 +63,32 @@ func (c *Compiler) getTypeCode(typ types.Type) llvm.Value {
 			// Take a pointer to the typecodeID of the first field (if it exists).
 			structGlobal := c.makeStructTypeFields(typ)
 			references = llvm.ConstBitCast(structGlobal, global.Type())
+		case *types.Interface:
+			methodSetGlobal := c.getInterfaceMethodSet(typ)
+			references = llvm.ConstBitCast(methodSetGlobal, global.Type())
 		}
+		if _, ok := typ.Underlying().(*types.Interface); !ok {
+			methodSet = c.getTypeMethodSet(typ)
+		}
+		if _, ok := typ.Underlying().(*types.Pointer); !ok {
+			ptrTo = c.getTypeCode(types.NewPointer(typ))
+		}
+		globalValue := llvm.ConstNull(global.Type().ElementType())
 		if !references.IsNil() {
-			// Set the 'references' field of the runtime.typecodeID struct.
-			globalValue := llvm.ConstNull(global.Type().ElementType())
 			globalValue = llvm.ConstInsertValue(globalValue, references, []uint32{0})
-			if length != 0 {
-				lengthValue := llvm.ConstInt(c.uintptrType, uint64(length), false)
-				globalValue = llvm.ConstInsertValue(globalValue, lengthValue, []uint32{1})
-			}
-			global.SetInitializer(globalValue)
-			global.SetLinkage(llvm.PrivateLinkage)
 		}
+		if length != 0 {
+			lengthValue := llvm.ConstInt(c.uintptrType, uint64(length), false)
+			globalValue = llvm.ConstInsertValue(globalValue, lengthValue, []uint32{1})
+		}
+		if !methodSet.IsNil() {
+			globalValue = llvm.ConstInsertValue(globalValue, methodSet, []uint32{2})
+		}
+		if !ptrTo.IsNil() {
+			globalValue = llvm.ConstInsertValue(globalValue, ptrTo, []uint32{3})
+		}
+		global.SetInitializer(globalValue)
+		global.SetLinkage(llvm.LinkOnceODRLinkage)
 		global.SetGlobalConstant(true)
 	}
 	return global
@@ -91,7 +97,7 @@ func (c *Compiler) getTypeCode(typ types.Type) llvm.Value {
 // makeStructTypeFields creates a new global that stores all type information
 // related to this struct type, and returns the resulting global. This global is
 // actually an array of all the fields in the structs.
-func (c *Compiler) makeStructTypeFields(typ *types.Struct) llvm.Value {
+func (c *compilerContext) makeStructTypeFields(typ *types.Struct) llvm.Value {
 	// The global is an array of runtime.structField structs.
 	runtimeStructField := c.getLLVMRuntimeType("structField")
 	structGlobalType := llvm.ArrayType(runtimeStructField, typ.NumFields())
@@ -104,8 +110,8 @@ func (c *Compiler) makeStructTypeFields(typ *types.Struct) llvm.Value {
 		fieldName.SetLinkage(llvm.PrivateLinkage)
 		fieldName.SetUnnamedAddr(true)
 		fieldName = llvm.ConstGEP(fieldName, []llvm.Value{
-			llvm.ConstInt(llvm.Int32Type(), 0, false),
-			llvm.ConstInt(llvm.Int32Type(), 0, false),
+			llvm.ConstInt(c.ctx.Int32Type(), 0, false),
+			llvm.ConstInt(c.ctx.Int32Type(), 0, false),
 		})
 		fieldGlobalValue = llvm.ConstInsertValue(fieldGlobalValue, fieldName, []uint32{1})
 		if typ.Tag(i) != "" {
@@ -113,8 +119,8 @@ func (c *Compiler) makeStructTypeFields(typ *types.Struct) llvm.Value {
 			fieldTag.SetLinkage(llvm.PrivateLinkage)
 			fieldTag.SetUnnamedAddr(true)
 			fieldTag = llvm.ConstGEP(fieldTag, []llvm.Value{
-				llvm.ConstInt(llvm.Int32Type(), 0, false),
-				llvm.ConstInt(llvm.Int32Type(), 0, false),
+				llvm.ConstInt(c.ctx.Int32Type(), 0, false),
+				llvm.ConstInt(c.ctx.Int32Type(), 0, false),
 			})
 			fieldGlobalValue = llvm.ConstInsertValue(fieldGlobalValue, fieldTag, []uint32{2})
 		}
@@ -187,7 +193,7 @@ func getTypeCodeName(t types.Type) string {
 	case *types.Interface:
 		methods := make([]string, t.NumMethods())
 		for i := 0; i < t.NumMethods(); i++ {
-			methods[i] = getTypeCodeName(t.Method(i).Type())
+			methods[i] = t.Method(i).Name() + ":" + getTypeCodeName(t.Method(i).Type())
 		}
 		return "interface:" + "{" + strings.Join(methods, ",") + "}"
 	case *types.Map:
@@ -210,10 +216,6 @@ func getTypeCodeName(t types.Type) string {
 		return "slice:" + getTypeCodeName(t.Elem())
 	case *types.Struct:
 		elems := make([]string, t.NumFields())
-		if t.NumFields() > 2 && t.Field(0).Name() == "C union" {
-			// TODO: report this as a normal error instead of panicking.
-			panic("cgo unions are not allowed in interfaces")
-		}
 		for i := 0; i < t.NumFields(); i++ {
 			embedded := ""
 			if t.Field(i).Embedded() {
@@ -232,7 +234,7 @@ 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 {
+func (c *compilerContext) getTypeMethodSet(typ types.Type) llvm.Value {
 	global := c.mod.NamedGlobal(typ.String() + "$methodset")
 	zero := llvm.ConstInt(c.ctx.Int32Type(), 0, false)
 	if !global.IsNil() {
@@ -240,7 +242,7 @@ func (c *Compiler) getTypeMethodSet(typ types.Type) llvm.Value {
 		return llvm.ConstGEP(global, []llvm.Value{zero, zero})
 	}
 
-	ms := c.ir.Program.MethodSets.MethodSet(typ)
+	ms := c.program.MethodSets.MethodSet(typ)
 	if ms.Len() == 0 {
 		// no methods, so can leave that one out
 		return llvm.ConstPointerNull(llvm.PointerType(c.getLLVMRuntimeType("interfaceMethodInfo"), 0))
@@ -251,15 +253,16 @@ func (c *Compiler) getTypeMethodSet(typ types.Type) llvm.Value {
 	for i := 0; i < ms.Len(); i++ {
 		method := ms.At(i)
 		signatureGlobal := c.getMethodSignature(method.Obj().(*types.Func))
-		f := c.ir.GetFunction(c.ir.Program.MethodValue(method))
-		if f.LLVMFn.IsNil() {
+		fn := c.program.MethodValue(method)
+		llvmFn := c.getFunction(fn)
+		if llvmFn.IsNil() {
 			// compiler error, so panic
-			panic("cannot find function: " + f.LinkName())
+			panic("cannot find function: " + c.getFunctionInfo(fn).linkName)
 		}
-		fn := c.getInterfaceInvokeWrapper(f)
+		wrapper := c.getInterfaceInvokeWrapper(fn, llvmFn)
 		methodInfo := llvm.ConstNamedStruct(interfaceMethodInfoType, []llvm.Value{
 			signatureGlobal,
-			llvm.ConstPtrToInt(fn, c.uintptrType),
+			llvm.ConstPtrToInt(wrapper, c.uintptrType),
 		})
 		methods[i] = methodInfo
 	}
@@ -268,61 +271,77 @@ func (c *Compiler) getTypeMethodSet(typ types.Type) llvm.Value {
 	global = llvm.AddGlobal(c.mod, arrayType, typ.String()+"$methodset")
 	global.SetInitializer(value)
 	global.SetGlobalConstant(true)
-	global.SetLinkage(llvm.PrivateLinkage)
+	global.SetLinkage(llvm.LinkOnceODRLinkage)
 	return llvm.ConstGEP(global, []llvm.Value{zero, zero})
 }
 
 // getInterfaceMethodSet returns a global variable with the method set of the
 // given named interface type. This method set is used by the interface lowering
 // pass.
-func (c *Compiler) getInterfaceMethodSet(typ *types.Named) llvm.Value {
-	global := c.mod.NamedGlobal(typ.String() + "$interface")
+func (c *compilerContext) getInterfaceMethodSet(typ types.Type) llvm.Value {
+	name := typ.String()
+	if _, ok := typ.(*types.Named); !ok {
+		// Anonymous interface.
+		name = "reflect/types.interface:" + name
+	}
+	global := c.mod.NamedGlobal(name + "$interface")
 	zero := llvm.ConstInt(c.ctx.Int32Type(), 0, false)
 	if !global.IsNil() {
 		// method set already exist, return it
 		return llvm.ConstGEP(global, []llvm.Value{zero, zero})
 	}
 
-	// Every method is a *i16 reference indicating the signature of this method.
+	// Every method is a *i8 reference indicating the signature of this method.
 	methods := make([]llvm.Value, typ.Underlying().(*types.Interface).NumMethods())
 	for i := range methods {
 		method := typ.Underlying().(*types.Interface).Method(i)
 		methods[i] = c.getMethodSignature(method)
 	}
 
-	value := llvm.ConstArray(methods[0].Type(), methods)
-	global = llvm.AddGlobal(c.mod, value.Type(), typ.String()+"$interface")
+	value := llvm.ConstArray(c.i8ptrType, methods)
+	global = llvm.AddGlobal(c.mod, value.Type(), name+"$interface")
 	global.SetInitializer(value)
 	global.SetGlobalConstant(true)
-	global.SetLinkage(llvm.PrivateLinkage)
+	global.SetLinkage(llvm.LinkOnceODRLinkage)
 	return llvm.ConstGEP(global, []llvm.Value{zero, zero})
 }
 
 // getMethodSignature returns a global variable which is a reference to an
-// external *i16 indicating the indicating the signature of this method. It is
+// external *i8 indicating the indicating the signature of this method. It is
 // used during the interface lowering pass.
-func (c *Compiler) getMethodSignature(method *types.Func) llvm.Value {
-	signature := ir.MethodSignature(method)
-	signatureGlobal := c.mod.NamedGlobal("func " + signature)
+func (c *compilerContext) getMethodSignature(method *types.Func) llvm.Value {
+	signature := methodSignature(method)
+	var globalName string
+	if token.IsExported(method.Name()) {
+		globalName = "reflect/methods." + signature
+	} else {
+		globalName = method.Type().(*types.Signature).Recv().Pkg().Path() + ".$methods." + signature
+	}
+	signatureGlobal := c.mod.NamedGlobal(globalName)
 	if signatureGlobal.IsNil() {
-		signatureGlobal = llvm.AddGlobal(c.mod, c.ctx.Int8Type(), "func "+signature)
+		// TODO: put something useful in these globals, such as the method
+		// signature. Useful to one day implement reflect.Value.Method(n).
+		signatureGlobal = llvm.AddGlobal(c.mod, c.ctx.Int8Type(), globalName)
+		signatureGlobal.SetInitializer(llvm.ConstInt(c.ctx.Int8Type(), 0, false))
+		signatureGlobal.SetLinkage(llvm.LinkOnceODRLinkage)
 		signatureGlobal.SetGlobalConstant(true)
+		signatureGlobal.SetAlignment(1)
 	}
 	return signatureGlobal
 }
 
-// parseTypeAssert will emit the code for a typeassert, used in if statements
+// createTypeAssert will emit the code for a typeassert, used in if statements
 // and in type switches (Go SSA does not have type switches, only if/else
 // chains). Note that even though the Go SSA does not contain type switches,
 // LLVM will recognize the pattern and make it a real switch in many cases.
 //
 // 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 {
-	itf := c.getValue(frame, expr.X)
-	assertedType := c.getLLVMType(expr.AssertedType)
+func (b *builder) createTypeAssert(expr *ssa.TypeAssert) llvm.Value {
+	itf := b.getValue(expr.X)
+	assertedType := b.getLLVMType(expr.AssertedType)
 
-	actualTypeNum := c.builder.CreateExtractValue(itf, 0, "interface.type")
+	actualTypeNum := b.CreateExtractValue(itf, 0, "interface.type")
 	commaOk := llvm.Value{}
 	if _, ok := expr.AssertedType.Underlying().(*types.Interface); ok {
 		// Type assert on interface type.
@@ -333,15 +352,21 @@ func (c *Compiler) parseTypeAssert(frame *Frame, expr *ssa.TypeAssert) llvm.Valu
 		// the main Go compiler, where the runtime checks whether the type
 		// implements each method of the interface. See:
 		// https://research.swtch.com/interfaces
-		methodSet := c.getInterfaceMethodSet(expr.AssertedType.(*types.Named))
-		commaOk = c.createRuntimeCall("interfaceImplements", []llvm.Value{actualTypeNum, methodSet}, "")
+		methodSet := b.getInterfaceMethodSet(expr.AssertedType)
+		commaOk = b.createRuntimeCall("interfaceImplements", []llvm.Value{actualTypeNum, methodSet}, "")
 
 	} else {
+		globalName := "reflect/types.typeid:" + getTypeCodeName(expr.AssertedType)
+		assertedTypeCodeGlobal := b.mod.NamedGlobal(globalName)
+		if assertedTypeCodeGlobal.IsNil() {
+			// Create a new typecode global.
+			assertedTypeCodeGlobal = llvm.AddGlobal(b.mod, b.ctx.Int8Type(), globalName)
+			assertedTypeCodeGlobal.SetGlobalConstant(true)
+		}
 		// Type assert on concrete type.
 		// Call runtime.typeAssert, which will be lowered to a simple icmp or
 		// const false in the interface lowering pass.
-		assertedTypeCodeGlobal := c.getTypeCode(expr.AssertedType)
-		commaOk = c.createRuntimeCall("typeAssert", []llvm.Value{actualTypeNum, assertedTypeCodeGlobal}, "typecode")
+		commaOk = b.createRuntimeCall("typeAssert", []llvm.Value{actualTypeNum, assertedTypeCodeGlobal}, "typecode")
 	}
 
 	// Add 2 new basic blocks (that should get optimized away): one for the
@@ -355,15 +380,15 @@ func (c *Compiler) parseTypeAssert(frame *Frame, expr *ssa.TypeAssert) llvm.Valu
 	// typeassert should return a zero value, not an incorrectly casted
 	// value.
 
-	prevBlock := c.builder.GetInsertBlock()
-	okBlock := c.ctx.AddBasicBlock(frame.fn.LLVMFn, "typeassert.ok")
-	nextBlock := c.ctx.AddBasicBlock(frame.fn.LLVMFn, "typeassert.next")
-	frame.blockExits[frame.currentBlock] = nextBlock // adjust outgoing block for phi nodes
-	c.builder.CreateCondBr(commaOk, okBlock, nextBlock)
+	prevBlock := b.GetInsertBlock()
+	okBlock := b.ctx.AddBasicBlock(b.llvmFn, "typeassert.ok")
+	nextBlock := b.ctx.AddBasicBlock(b.llvmFn, "typeassert.next")
+	b.blockExits[b.currentBlock] = nextBlock // adjust outgoing block for phi nodes
+	b.CreateCondBr(commaOk, okBlock, nextBlock)
 
 	// Retrieve the value from the interface if the type assert was
 	// successful.
-	c.builder.SetInsertPointAtEnd(okBlock)
+	b.SetInsertPointAtEnd(okBlock)
 	var valueOk llvm.Value
 	if _, ok := expr.AssertedType.Underlying().(*types.Interface); ok {
 		// Type assert on interface type. Easy: just return the same
@@ -372,75 +397,71 @@ func (c *Compiler) parseTypeAssert(frame *Frame, expr *ssa.TypeAssert) llvm.Valu
 	} else {
 		// 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")
-		valueOk = c.emitPointerUnpack(valuePtr, []llvm.Type{assertedType})[0]
+		valuePtr := b.CreateExtractValue(itf, 1, "typeassert.value.ptr")
+		valueOk = b.emitPointerUnpack(valuePtr, []llvm.Type{assertedType})[0]
 	}
-	c.builder.CreateBr(nextBlock)
+	b.CreateBr(nextBlock)
 
 	// Continue after the if statement.
-	c.builder.SetInsertPointAtEnd(nextBlock)
-	phi := c.builder.CreatePHI(assertedType, "typeassert.value")
+	b.SetInsertPointAtEnd(nextBlock)
+	phi := b.CreatePHI(assertedType, "typeassert.value")
 	phi.AddIncoming([]llvm.Value{llvm.ConstNull(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
+		tuple := b.ctx.ConstStruct([]llvm.Value{llvm.Undef(assertedType), llvm.Undef(b.ctx.Int1Type())}, false) // create empty tuple
+		tuple = b.CreateInsertValue(tuple, phi, 0, "")                                                          // insert value
+		tuple = b.CreateInsertValue(tuple, commaOk, 1, "")                                                      // insert 'comma ok' boolean
 		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}, "")
+		b.createRuntimeCall("interfaceTypeAssert", []llvm.Value{commaOk}, "")
 		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) {
-	// Call an interface method with dynamic dispatch.
-	itf := c.getValue(frame, instr.Value) // interface
-
-	llvmFnType := c.getRawFuncType(instr.Method.Type().(*types.Signature))
-
-	typecode := c.builder.CreateExtractValue(itf, 0, "invoke.typecode")
+// getInvokePtr creates an interface function pointer lookup for the specified invoke instruction, using a specified typecode.
+func (b *builder) getInvokePtr(instr *ssa.CallCommon, typecode llvm.Value) llvm.Value {
+	llvmFnType := b.getRawFuncType(instr.Method.Type().(*types.Signature))
 	values := []llvm.Value{
 		typecode,
-		c.getInterfaceMethodSet(instr.Value.Type().(*types.Named)),
-		c.getMethodSignature(instr.Method),
+		b.getInterfaceMethodSet(instr.Value.Type()),
+		b.getMethodSignature(instr.Method),
 	}
-	fn := c.createRuntimeCall("interfaceMethod", values, "invoke.func")
-	fnCast := c.builder.CreateIntToPtr(fn, llvmFnType, "invoke.func.cast")
-	receiverValue := c.builder.CreateExtractValue(itf, 1, "invoke.func.receiver")
+	fn := b.createRuntimeCall("interfaceMethod", values, "invoke.func")
+	return b.CreateIntToPtr(fn, llvmFnType, "invoke.func.cast")
+}
+
+// getInvokeCall creates and returns the function pointer and parameters of an
+// interface call.
+func (b *builder) getInvokeCall(instr *ssa.CallCommon) (llvm.Value, []llvm.Value) {
+	// Call an interface method with dynamic dispatch.
+	itf := b.getValue(instr.Value) // interface
+
+	typecode := b.CreateExtractValue(itf, 0, "invoke.typecode")
+	fnCast := b.getInvokePtr(instr, typecode)
+	receiverValue := b.CreateExtractValue(itf, 1, "invoke.func.receiver")
 
 	args := []llvm.Value{receiverValue}
 	for _, arg := range instr.Args {
-		args = append(args, c.getValue(frame, arg))
+		args = append(args, b.getValue(arg))
 	}
 	// Add the context parameter. An interface call never takes a context but we
 	// have to supply the parameter anyway.
-	args = append(args, llvm.Undef(c.i8ptrType))
+	args = append(args, llvm.Undef(b.i8ptrType))
 	// Add the parent goroutine handle.
-	args = append(args, llvm.Undef(c.i8ptrType))
+	args = append(args, llvm.Undef(b.i8ptrType))
 
 	return fnCast, args
 }
 
-// interfaceInvokeWrapper keeps some state between getInterfaceInvokeWrapper and
-// createInterfaceInvokeWrapper. The former is called during IR construction
-// itself and the latter is called when finishing up the IR.
-type interfaceInvokeWrapper struct {
-	fn           *ir.Function
-	wrapper      llvm.Value
-	receiverType llvm.Type
-}
-
-// Wrap an interface method function pointer. The wrapper takes in a pointer to
-// 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 {
-	wrapperName := f.LinkName() + "$invoke"
+// getInterfaceInvokeWrapper returns a wrapper for the given method so it can be
+// invoked from an interface. The wrapper takes in a pointer to the underlying
+// value, dereferences or unpacks it if necessary, and calls the real method.
+// If the method to wrap has a pointer receiver, no wrapping is necessary and
+// the function is returned directly.
+func (c *compilerContext) getInterfaceInvokeWrapper(fn *ssa.Function, llvmFn llvm.Value) llvm.Value {
+	wrapperName := llvmFn.Name() + "$invoke"
 	wrapper := c.mod.NamedFunction(wrapperName)
 	if !wrapper.IsNil() {
 		// Wrapper already created. Return it directly.
@@ -448,8 +469,11 @@ func (c *Compiler) getInterfaceInvokeWrapper(f *ir.Function) llvm.Value {
 	}
 
 	// Get the expanded receiver type.
-	receiverType := c.getLLVMType(f.Params[0].Type())
-	expandedReceiverType := c.expandFormalParamType(receiverType)
+	receiverType := c.getLLVMType(fn.Signature.Recv().Type())
+	var expandedReceiverType []llvm.Type
+	for _, info := range c.expandFormalParamType(receiverType, "", nil) {
+		expandedReceiverType = append(expandedReceiverType, info.llvmType)
+	}
 
 	// Does this method even need any wrapping?
 	if len(expandedReceiverType) == 1 && receiverType.TypeKind() == llvm.PointerTypeKind {
@@ -457,49 +481,93 @@ func (c *Compiler) getInterfaceInvokeWrapper(f *ir.Function) llvm.Value {
 		// 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
+		return llvmFn
 	}
 
 	// create wrapper function
-	fnType := f.LLVMFn.Type().ElementType()
+	fnType := llvmFn.Type().ElementType()
 	paramTypes := append([]llvm.Type{c.i8ptrType}, fnType.ParamTypes()[len(expandedReceiverType):]...)
 	wrapFnType := llvm.FunctionType(fnType.ReturnType(), paramTypes, false)
 	wrapper = llvm.AddFunction(c.mod, wrapperName, wrapFnType)
-	c.interfaceInvokeWrappers = append(c.interfaceInvokeWrappers, interfaceInvokeWrapper{
-		fn:           f,
-		wrapper:      wrapper,
-		receiverType: receiverType,
-	})
-	return wrapper
-}
+	wrapper.LastParam().SetName("parentHandle")
 
-// createInterfaceInvokeWrapper finishes the work of getInterfaceInvokeWrapper,
-// see that function for details.
-func (c *Compiler) createInterfaceInvokeWrapper(state interfaceInvokeWrapper) {
-	wrapper := state.wrapper
-	fn := state.fn
-	receiverType := state.receiverType
-	wrapper.SetLinkage(llvm.InternalLinkage)
+	wrapper.SetLinkage(llvm.LinkOnceODRLinkage)
 	wrapper.SetUnnamedAddr(true)
 
+	// Create a new builder just to create this wrapper.
+	b := builder{
+		compilerContext: c,
+		Builder:         c.ctx.NewBuilder(),
+	}
+	defer b.Builder.Dispose()
+
 	// add debug info if needed
 	if c.Debug {
-		pos := c.ir.Program.Fset.Position(fn.Pos())
+		pos := c.program.Fset.Position(fn.Pos())
 		difunc := c.attachDebugInfoRaw(fn, wrapper, "$invoke", pos.Filename, pos.Line)
-		c.builder.SetCurrentDebugLocation(uint(pos.Line), uint(pos.Column), difunc, llvm.Metadata{})
+		b.SetCurrentDebugLocation(uint(pos.Line), uint(pos.Column), difunc, llvm.Metadata{})
 	}
 
 	// set up IR builder
-	block := c.ctx.AddBasicBlock(wrapper, "entry")
-	c.builder.SetInsertPointAtEnd(block)
+	block := b.ctx.AddBasicBlock(wrapper, "entry")
+	b.SetInsertPointAtEnd(block)
 
-	receiverValue := c.emitPointerUnpack(wrapper.Param(0), []llvm.Type{receiverType})[0]
-	params := append(c.expandFormalParam(receiverValue), wrapper.Params()[1:]...)
-	if fn.LLVMFn.Type().ElementType().ReturnType().TypeKind() == llvm.VoidTypeKind {
-		c.builder.CreateCall(fn.LLVMFn, params, "")
-		c.builder.CreateRetVoid()
+	receiverValue := b.emitPointerUnpack(wrapper.Param(0), []llvm.Type{receiverType})[0]
+	params := append(b.expandFormalParam(receiverValue), wrapper.Params()[1:]...)
+	if llvmFn.Type().ElementType().ReturnType().TypeKind() == llvm.VoidTypeKind {
+		b.CreateCall(llvmFn, params, "")
+		b.CreateRetVoid()
 	} else {
-		ret := c.builder.CreateCall(fn.LLVMFn, params, "ret")
-		c.builder.CreateRet(ret)
+		ret := b.CreateCall(llvmFn, params, "ret")
+		b.CreateRet(ret)
 	}
+
+	return wrapper
+}
+
+// methodSignature creates a readable version of a method signature (including
+// the function name, excluding the receiver name). This string is used
+// internally to match interfaces and to call the correct method on an
+// interface. Examples:
+//
+//     String() string
+//     Read([]byte) (int, error)
+func methodSignature(method *types.Func) string {
+	return method.Name() + signature(method.Type().(*types.Signature))
+}
+
+// Make a readable version of a function (pointer) signature.
+// Examples:
+//
+//     () string
+//     (string, int) (int, error)
+func signature(sig *types.Signature) string {
+	s := ""
+	if sig.Params().Len() == 0 {
+		s += "()"
+	} else {
+		s += "("
+		for i := 0; i < sig.Params().Len(); i++ {
+			if i > 0 {
+				s += ", "
+			}
+			s += sig.Params().At(i).Type().String()
+		}
+		s += ")"
+	}
+	if sig.Results().Len() == 0 {
+		// keep as-is
+	} else if sig.Results().Len() == 1 {
+		s += " " + sig.Results().At(0).Type().String()
+	} else {
+		s += " ("
+		for i := 0; i < sig.Results().Len(); i++ {
+			if i > 0 {
+				s += ", "
+			}
+			s += sig.Results().At(i).Type().String()
+		}
+		s += ")"
+	}
+	return s
 }

compiler/interrupt.go

@@ -0,0 +1,89 @@
+package compiler
+
+import (
+	"strconv"
+	"strings"
+
+	"golang.org/x/tools/go/ssa"
+	"tinygo.org/x/go-llvm"
+)
+
+// createInterruptGlobal creates a new runtime/interrupt.Interrupt struct that
+// will be lowered to a real interrupt during interrupt lowering.
+//
+// This two-stage approach allows unused interrupts to be optimized away if
+// necessary.
+func (b *builder) createInterruptGlobal(instr *ssa.CallCommon) (llvm.Value, error) {
+	// Get the interrupt number, which must be a compile-time constant.
+	id, ok := instr.Args[0].(*ssa.Const)
+	if !ok {
+		return llvm.Value{}, b.makeError(instr.Pos(), "interrupt ID is not a constant")
+	}
+
+	// Get the func value, which also must be a compile time constant.
+	// Note that bound functions are allowed if the function has a pointer
+	// receiver and is a global. This is rather strict but still allows for
+	// idiomatic Go code.
+	funcValue := b.getValue(instr.Args[1])
+	if funcValue.IsAConstant().IsNil() {
+		// Try to determine the cause of the non-constantness for a nice error
+		// message.
+		switch instr.Args[1].(type) {
+		case *ssa.MakeClosure:
+			// This may also be a bound method.
+			return llvm.Value{}, b.makeError(instr.Pos(), "closures are not supported in interrupt.New")
+		}
+		// Fall back to a generic error.
+		return llvm.Value{}, b.makeError(instr.Pos(), "interrupt function must be constant")
+	}
+
+	// Create a new global of type runtime/interrupt.handle. Globals of this
+	// type are lowered in the interrupt lowering pass.
+	globalType := b.program.ImportedPackage("runtime/interrupt").Type("handle").Type()
+	globalLLVMType := b.getLLVMType(globalType)
+	globalName := b.fn.Package().Pkg.Path() + "$interrupt" + strconv.FormatInt(id.Int64(), 10)
+	global := llvm.AddGlobal(b.mod, globalLLVMType, globalName)
+	global.SetVisibility(llvm.HiddenVisibility)
+	global.SetGlobalConstant(true)
+	global.SetUnnamedAddr(true)
+	initializer := llvm.ConstNull(globalLLVMType)
+	initializer = llvm.ConstInsertValue(initializer, funcValue, []uint32{0})
+	initializer = llvm.ConstInsertValue(initializer, llvm.ConstInt(b.intType, uint64(id.Int64()), true), []uint32{1, 0})
+	global.SetInitializer(initializer)
+
+	// Add debug info to the interrupt global.
+	if b.Debug {
+		pos := b.program.Fset.Position(instr.Pos())
+		diglobal := b.dibuilder.CreateGlobalVariableExpression(b.getDIFile(pos.Filename), llvm.DIGlobalVariableExpression{
+			Name:        "interrupt" + strconv.FormatInt(id.Int64(), 10),
+			LinkageName: globalName,
+			File:        b.getDIFile(pos.Filename),
+			Line:        pos.Line,
+			Type:        b.getDIType(globalType),
+			Expr:        b.dibuilder.CreateExpression(nil),
+			LocalToUnit: false,
+		})
+		global.AddMetadata(0, diglobal)
+	}
+
+	// Create the runtime/interrupt.Interrupt type. It is a struct with a single
+	// member of type int.
+	num := llvm.ConstPtrToInt(global, b.intType)
+	interrupt := llvm.ConstNamedStruct(b.mod.GetTypeByName("runtime/interrupt.Interrupt"), []llvm.Value{num})
+
+	// Add dummy "use" call for AVR, because interrupts may be used even though
+	// they are never referenced again. This is unlike Cortex-M or the RISC-V
+	// PLIC where each interrupt must be enabled using the interrupt number, and
+	// thus keeps the Interrupt object alive.
+	// This call is removed during interrupt lowering.
+	if strings.HasPrefix(b.Triple, "avr") {
+		useFn := b.mod.NamedFunction("runtime/interrupt.use")
+		if useFn.IsNil() {
+			useFnType := llvm.FunctionType(b.ctx.VoidType(), []llvm.Type{interrupt.Type()}, false)
+			useFn = llvm.AddFunction(b.mod, "runtime/interrupt.use", useFnType)
+		}
+		b.CreateCall(useFn, []llvm.Value{interrupt}, "")
+	}
+
+	return interrupt, nil
+}

compiler/intrinsics.go

@@ -0,0 +1,50 @@
+package compiler
+
+// This file contains helper functions to create calls to LLVM intrinsics.
+
+import (
+	"strconv"
+
+	"golang.org/x/tools/go/ssa"
+	"tinygo.org/x/go-llvm"
+)
+
+// createMemoryCopyCall creates a call to a builtin LLVM memcpy or memmove
+// function, declaring this function if needed. These calls are treated
+// specially by optimization passes possibly resulting in better generated code,
+// and will otherwise be lowered to regular libc memcpy/memmove calls.
+func (b *builder) createMemoryCopyCall(fn *ssa.Function, args []ssa.Value) (llvm.Value, error) {
+	fnName := "llvm." + fn.Name() + ".p0i8.p0i8.i" + strconv.Itoa(b.uintptrType.IntTypeWidth())
+	llvmFn := b.mod.NamedFunction(fnName)
+	if llvmFn.IsNil() {
+		fnType := llvm.FunctionType(b.ctx.VoidType(), []llvm.Type{b.i8ptrType, b.i8ptrType, b.uintptrType, b.ctx.Int1Type()}, false)
+		llvmFn = llvm.AddFunction(b.mod, fnName, fnType)
+	}
+	var params []llvm.Value
+	for _, param := range args {
+		params = append(params, b.getValue(param))
+	}
+	params = append(params, llvm.ConstInt(b.ctx.Int1Type(), 0, false))
+	b.CreateCall(llvmFn, params, "")
+	return llvm.Value{}, nil
+}
+
+// createMemoryZeroCall creates calls to llvm.memset.* to zero a block of
+// memory, declaring the function if needed. These calls will be lowered to
+// regular libc memset calls if they aren't optimized out in a different way.
+func (b *builder) createMemoryZeroCall(args []ssa.Value) (llvm.Value, error) {
+	fnName := "llvm.memset.p0i8.i" + strconv.Itoa(b.uintptrType.IntTypeWidth())
+	llvmFn := b.mod.NamedFunction(fnName)
+	if llvmFn.IsNil() {
+		fnType := llvm.FunctionType(b.ctx.VoidType(), []llvm.Type{b.i8ptrType, b.ctx.Int8Type(), b.uintptrType, b.ctx.Int1Type()}, false)
+		llvmFn = llvm.AddFunction(b.mod, fnName, fnType)
+	}
+	params := []llvm.Value{
+		b.getValue(args[0]),
+		llvm.ConstInt(b.ctx.Int8Type(), 0, false),
+		b.getValue(args[1]),
+		llvm.ConstInt(b.ctx.Int1Type(), 0, false),
+	}
+	b.CreateCall(llvmFn, params, "")
+	return llvm.Value{}, nil
+}

compiler/ircheck/check.go

@@ -0,0 +1,194 @@
+// Package ircheck implements a checker for LLVM IR, that goes a bit further
+// than the regular LLVM IR verifier. Note that it checks different things, so
+// this is not a replacement for the LLVM verifier but does catch things that
+// the LLVM verifier doesn't catch.
+package ircheck
+
+import (
+	"errors"
+	"fmt"
+
+	"tinygo.org/x/go-llvm"
+)
+
+type checker struct {
+	ctx llvm.Context
+}
+
+func (c *checker) checkType(t llvm.Type, checked map[llvm.Type]struct{}, specials map[llvm.TypeKind]llvm.Type) error {
+	// prevent infinite recursion for self-referential types
+	if _, ok := checked[t]; ok {
+		return nil
+	}
+	checked[t] = struct{}{}
+
+	// check for any context mismatches
+	switch {
+	case t.Context() == c.ctx:
+		// this is correct
+	case t.Context() == llvm.GlobalContext():
+		// somewhere we accidentally used the global context instead of a real context
+		return fmt.Errorf("type %q uses global context", t.String())
+	default:
+		// we used some other context by accident
+		return fmt.Errorf("type %q uses context %v instead of the main context %v", t.Context(), c.ctx)
+	}
+
+	// if this is a composite type, check the components of the type
+	switch t.TypeKind() {
+	case llvm.VoidTypeKind, llvm.LabelTypeKind, llvm.TokenTypeKind, llvm.MetadataTypeKind:
+		// there should only be one of any of these
+		if s, ok := specials[t.TypeKind()]; !ok {
+			specials[t.TypeKind()] = t
+		} else if s != t {
+			return fmt.Errorf("duplicate special type %q: %v and %v", t.TypeKind().String(), t, s)
+		}
+	case llvm.FloatTypeKind, llvm.DoubleTypeKind, llvm.X86_FP80TypeKind, llvm.FP128TypeKind, llvm.PPC_FP128TypeKind:
+		// floating point numbers are primitives - nothing to recurse
+	case llvm.IntegerTypeKind:
+		// integers are primitives - nothing to recurse
+	case llvm.FunctionTypeKind:
+		// check arguments and return(s)
+		for i, v := range t.ParamTypes() {
+			if err := c.checkType(v, checked, specials); err != nil {
+				return fmt.Errorf("failed to verify argument %d of type %s: %s", i, t.String(), err.Error())
+			}
+		}
+		if err := c.checkType(t.ReturnType(), checked, specials); err != nil {
+			return fmt.Errorf("failed to verify return type of type %s: %s", t.String(), err.Error())
+		}
+	case llvm.StructTypeKind:
+		// check all elements
+		for i, v := range t.StructElementTypes() {
+			if err := c.checkType(v, checked, specials); err != nil {
+				return fmt.Errorf("failed to verify type of field %d of struct type %s: %s", i, t.String(), err.Error())
+			}
+		}
+	case llvm.ArrayTypeKind:
+		// check element type
+		if err := c.checkType(t.ElementType(), checked, specials); err != nil {
+			return fmt.Errorf("failed to verify element type of array type %s: %s", t.String(), err.Error())
+		}
+	case llvm.PointerTypeKind:
+		// check underlying type
+		if err := c.checkType(t.ElementType(), checked, specials); err != nil {
+			return fmt.Errorf("failed to verify underlying type of pointer type %s: %s", t.String(), err.Error())
+		}
+	case llvm.VectorTypeKind:
+		// check element type
+		if err := c.checkType(t.ElementType(), checked, specials); err != nil {
+			return fmt.Errorf("failed to verify element type of vector type %s: %s", t.String(), err.Error())
+		}
+	default:
+		return fmt.Errorf("unrecognized kind %q of type %s", t.TypeKind(), t.String())
+	}
+
+	return nil
+}
+
+func (c *checker) checkValue(v llvm.Value, types map[llvm.Type]struct{}, specials map[llvm.TypeKind]llvm.Type) error {
+	// check type
+	if err := c.checkType(v.Type(), types, specials); err != nil {
+		return fmt.Errorf("failed to verify type of value: %s", err.Error())
+	}
+
+	// check if this is an undefined void
+	if v.IsUndef() && v.Type().TypeKind() == llvm.VoidTypeKind {
+		return errors.New("encountered undefined void value")
+	}
+
+	return nil
+}
+
+func (c *checker) checkInstruction(inst llvm.Value, types map[llvm.Type]struct{}, specials map[llvm.TypeKind]llvm.Type) error {
+	// check value properties
+	if err := c.checkValue(inst, types, specials); err != nil {
+		return errorAt(inst, err.Error())
+	}
+
+	// The alloca instruction can be present in every basic block. However,
+	// allocas in basic blocks other than the entry basic block have a number of
+	// problems:
+	//   * They are hard to optimize, leading to potential missed optimizations.
+	//   * They may cause stack overflows in loops that would otherwise be
+	//     innocent.
+	//   * They cause extra code to be generated, because it requires the use of
+	//     a frame pointer.
+	//   * Perhaps most importantly, the coroutine lowering pass of LLVM (as of
+	//     LLVM 9) cannot deal with these allocas:
+	//     https://llvm.org/docs/Coroutines.html
+	// Therefore, alloca instructions should be limited to the entry block.
+	if !inst.IsAAllocaInst().IsNil() {
+		if inst.InstructionParent() != inst.InstructionParent().Parent().EntryBasicBlock() {
+			return errorAt(inst, "internal error: non-static alloca")
+		}
+	}
+
+	// check operands
+	for i := 0; i < inst.OperandsCount(); i++ {
+		if err := c.checkValue(inst.Operand(i), types, specials); err != nil {
+			return errorAt(inst, fmt.Sprintf("failed to validate operand %d of instruction %q: %s", i, inst.Name(), err.Error()))
+		}
+	}
+
+	return nil
+}
+
+func (c *checker) checkBasicBlock(bb llvm.BasicBlock, types map[llvm.Type]struct{}, specials map[llvm.TypeKind]llvm.Type) []error {
+	// check basic block value and type
+	var errs []error
+	if err := c.checkValue(bb.AsValue(), types, specials); err != nil {
+		errs = append(errs, errorAt(bb.Parent(), fmt.Sprintf("failed to validate value of basic block %s: %v", bb.AsValue().Name(), err)))
+	}
+
+	// check instructions
+	for inst := bb.FirstInstruction(); !inst.IsNil(); inst = llvm.NextInstruction(inst) {
+		if err := c.checkInstruction(inst, types, specials); err != nil {
+			errs = append(errs, err)
+		}
+	}
+
+	return errs
+}
+
+func (c *checker) checkFunction(fn llvm.Value, types map[llvm.Type]struct{}, specials map[llvm.TypeKind]llvm.Type) []error {
+	// check function value and type
+	var errs []error
+	if err := c.checkValue(fn, types, specials); err != nil {
+		errs = append(errs, fmt.Errorf("failed to validate value of function %s: %s", fn.Name(), err.Error()))
+	}
+
+	// check basic blocks
+	for bb := fn.FirstBasicBlock(); !bb.IsNil(); bb = llvm.NextBasicBlock(bb) {
+		errs = append(errs, c.checkBasicBlock(bb, types, specials)...)
+	}
+
+	return errs
+}
+
+// Module checks the given module and returns a slice of error, if there are
+// any.
+func Module(mod llvm.Module) []error {
+	// check for any context mismatches
+	var errs []error
+	c := checker{
+		ctx: mod.Context(),
+	}
+	if c.ctx == llvm.GlobalContext() {
+		// somewhere we accidentally used the global context instead of a real context
+		errs = append(errs, errors.New("module uses global context"))
+	}
+
+	types := map[llvm.Type]struct{}{}
+	specials := map[llvm.TypeKind]llvm.Type{}
+	for fn := mod.FirstFunction(); !fn.IsNil(); fn = llvm.NextFunction(fn) {
+		errs = append(errs, c.checkFunction(fn, types, specials)...)
+	}
+	for g := mod.FirstGlobal(); !g.IsNil(); g = llvm.NextGlobal(g) {
+		if err := c.checkValue(g, types, specials); err != nil {
+			errs = append(errs, fmt.Errorf("failed to verify global %s of module: %s", g.Name(), err.Error()))
+		}
+	}
+
+	return errs
+}

compiler/ircheck/errors.go

@@ -0,0 +1,48 @@
+package ircheck
+
+import (
+	"go/scanner"
+	"go/token"
+	"path/filepath"
+
+	"tinygo.org/x/go-llvm"
+)
+
+// errorAt returns an error value at the location of the instruction.
+// The location information may not be complete as it depends on debug
+// information in the IR.
+func errorAt(inst llvm.Value, msg string) scanner.Error {
+	return scanner.Error{
+		Pos: getPosition(inst),
+		Msg: msg,
+	}
+}
+
+// getPosition returns the position information for the given value, as far as
+// it is available.
+func getPosition(val llvm.Value) token.Position {
+	if !val.IsAInstruction().IsNil() {
+		loc := val.InstructionDebugLoc()
+		if loc.IsNil() {
+			return token.Position{}
+		}
+		file := loc.LocationScope().ScopeFile()
+		return token.Position{
+			Filename: filepath.Join(file.FileDirectory(), file.FileFilename()),
+			Line:     int(loc.LocationLine()),
+			Column:   int(loc.LocationColumn()),
+		}
+	} else if !val.IsAFunction().IsNil() {
+		loc := val.Subprogram()
+		if loc.IsNil() {
+			return token.Position{}
+		}
+		file := loc.ScopeFile()
+		return token.Position{
+			Filename: filepath.Join(file.FileDirectory(), file.FileFilename()),
+			Line:     int(loc.SubprogramLine()),
+		}
+	} else {
+		return token.Position{}
+	}
+}

compiler/llvm.go

@@ -1,8 +1,7 @@
 package compiler
 
 import (
-	"reflect"
-
+	"github.com/tinygo-org/tinygo/compiler/llvmutil"
 	"tinygo.org/x/go-llvm"
 )
 
@@ -24,178 +23,47 @@ 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
-// at the end of the current block.
-func (c *Compiler) createEntryBlockAlloca(t llvm.Type, name string) llvm.Value {
-	currentBlock := c.builder.GetInsertBlock()
-	entryBlock := currentBlock.Parent().EntryBasicBlock()
-	if entryBlock.FirstInstruction().IsNil() {
-		c.builder.SetInsertPointAtEnd(entryBlock)
-	} else {
-		c.builder.SetInsertPointBefore(entryBlock.FirstInstruction())
-	}
-	alloca := c.builder.CreateAlloca(t, name)
-	c.builder.SetInsertPointAtEnd(currentBlock)
-	return alloca
-}
-
 // createTemporaryAlloca creates a new alloca in the entry block and adds
-// lifetime start infromation in the IR signalling that the alloca won't be used
+// lifetime start information in 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 using emitLifetimeEnd after you're done with it.
-func (c *Compiler) createTemporaryAlloca(t llvm.Type, name string) (alloca, bitcast, size llvm.Value) {
-	alloca = c.createEntryBlockAlloca(t, name)
-	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
+func (b *builder) createTemporaryAlloca(t llvm.Type, name string) (alloca, bitcast, size llvm.Value) {
+	return llvmutil.CreateTemporaryAlloca(b.Builder, b.mod, t, name)
 }
 
 // emitLifetimeEnd signals the end of an (alloca) lifetime by calling the
 // llvm.lifetime.end intrinsic. It is commonly used together with
 // createTemporaryAlloca.
-func (c *Compiler) emitLifetimeEnd(ptr, size llvm.Value) {
-	c.builder.CreateCall(c.getLifetimeEndFunc(), []llvm.Value{size, ptr}, "")
+func (b *builder) emitLifetimeEnd(ptr, size llvm.Value) {
+	llvmutil.EmitLifetimeEnd(b.Builder, b.mod, ptr, size)
 }
 
-// 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
+// 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 (b *builder) emitPointerPack(values []llvm.Value) llvm.Value {
+	return llvmutil.EmitPointerPack(b.Builder, b.mod, b.NeedsStackObjects, values)
 }
 
-// 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.
-func (c *Compiler) splitBasicBlock(afterInst llvm.Value, insertAfter llvm.BasicBlock, name string) llvm.BasicBlock {
-	oldBlock := afterInst.InstructionParent()
-	newBlock := c.ctx.InsertBasicBlock(insertAfter, name)
-	var nextInstructions []llvm.Value // values to move
-
-	// Collect to-be-moved instructions.
-	inst := afterInst
-	for {
-		inst = llvm.NextInstruction(inst)
-		if inst.IsNil() {
-			break
-		}
-		nextInstructions = append(nextInstructions, inst)
-	}
-
-	// Move instructions.
-	c.builder.SetInsertPointAtEnd(newBlock)
-	for _, inst := range nextInstructions {
-		inst.RemoveFromParentAsInstruction()
-		c.builder.Insert(inst)
-	}
-
-	// Find PHI nodes to update.
-	var phiNodes []llvm.Value // PHI nodes to update
-	for bb := insertAfter.Parent().FirstBasicBlock(); !bb.IsNil(); bb = llvm.NextBasicBlock(bb) {
-		for inst := bb.FirstInstruction(); !inst.IsNil(); inst = llvm.NextInstruction(inst) {
-			if inst.IsAPHINode().IsNil() {
-				continue
-			}
-			needsUpdate := false
-			incomingCount := inst.IncomingCount()
-			for i := 0; i < incomingCount; i++ {
-				if inst.IncomingBlock(i) == oldBlock {
-					needsUpdate = true
-					break
-				}
-			}
-			if !needsUpdate {
-				// PHI node has no incoming edge from the old block.
-				continue
-			}
-			phiNodes = append(phiNodes, inst)
-		}
-	}
-
-	// Update PHI nodes.
-	for _, phi := range phiNodes {
-		c.builder.SetInsertPointBefore(phi)
-		newPhi := c.builder.CreatePHI(phi.Type(), "")
-		incomingCount := phi.IncomingCount()
-		incomingVals := make([]llvm.Value, incomingCount)
-		incomingBlocks := make([]llvm.BasicBlock, incomingCount)
-		for i := 0; i < incomingCount; i++ {
-			value := phi.IncomingValue(i)
-			block := phi.IncomingBlock(i)
-			if block == oldBlock {
-				block = newBlock
-			}
-			incomingVals[i] = value
-			incomingBlocks[i] = block
-		}
-		newPhi.AddIncoming(incomingVals, incomingBlocks)
-		phi.ReplaceAllUsesWith(newPhi)
-		phi.EraseFromParentAsInstruction()
-	}
-
-	return newBlock
+// emitPointerUnpack extracts a list of values packed using emitPointerPack.
+func (b *builder) emitPointerUnpack(ptr llvm.Value, valueTypes []llvm.Type) []llvm.Value {
+	return llvmutil.EmitPointerUnpack(b.Builder, b.mod, ptr, valueTypes)
 }
 
 // makeGlobalArray creates a new LLVM global with the given name and integers as
 // contents, and returns the global.
 // Note that it is left with the default linkage etc., you should set
 // linkage/constant/etc properties yourself.
-func (c *Compiler) makeGlobalArray(bufItf interface{}, name string, elementType llvm.Type) llvm.Value {
-	buf := reflect.ValueOf(bufItf)
-	globalType := llvm.ArrayType(elementType, buf.Len())
+func (c *compilerContext) makeGlobalArray(buf []byte, name string, elementType llvm.Type) llvm.Value {
+	globalType := llvm.ArrayType(elementType, len(buf))
 	global := llvm.AddGlobal(c.mod, globalType, name)
 	value := llvm.Undef(globalType)
-	for i := 0; i < buf.Len(); i++ {
-		ch := buf.Index(i).Uint()
+	for i := 0; i < len(buf); i++ {
+		ch := uint64(buf[i])
 		value = llvm.ConstInsertValue(value, llvm.ConstInt(elementType, ch, false), []uint32{uint32(i)})
 	}
 	global.SetInitializer(value)
 	return global
 }
-
-// getGlobalBytes returns the slice contained in the array of the provided
-// global. It can recover the bytes originally created using makeGlobalArray, if
-// makeGlobalArray was given a byte slice.
-func getGlobalBytes(global llvm.Value) []byte {
-	value := global.Initializer()
-	buf := make([]byte, value.Type().ArrayLength())
-	for i := range buf {
-		buf[i] = byte(llvm.ConstExtractValue(value, []uint32{uint32(i)}).ZExtValue())
-	}
-	return buf
-}
-
-// replaceGlobalByteWithArray replaces a global integer type in the module with
-// an integer array, using a GEP to make the types match. It is a convenience
-// function used for creating reflection sidetables, for example.
-func (c *Compiler) replaceGlobalIntWithArray(name string, buf interface{}) llvm.Value {
-	oldGlobal := c.mod.NamedGlobal(name)
-	global := c.makeGlobalArray(buf, name+".tmp", oldGlobal.Type().ElementType())
-	gep := llvm.ConstGEP(global, []llvm.Value{
-		llvm.ConstInt(c.ctx.Int32Type(), 0, false),
-		llvm.ConstInt(c.ctx.Int32Type(), 0, false),
-	})
-	oldGlobal.ReplaceAllUsesWith(gep)
-	oldGlobal.EraseFromParentAsGlobal()
-	global.SetName(name)
-	return global
-}

compiler/llvmutil/llvm.go

@@ -0,0 +1,168 @@
+// Package llvmutil contains utility functions used across multiple compiler
+// packages. For example, they may be used by both the compiler pacakge and
+// transformation packages.
+//
+// Normally, utility packages are avoided. However, in this case, the utility
+// functions are non-trivial and hard to get right. Copying them to multiple
+// places would be a big risk if only one of them is updated.
+package llvmutil
+
+import "tinygo.org/x/go-llvm"
+
+// CreateEntryBlockAlloca creates a new alloca in the entry block, even though
+// the IR builder is located elsewhere. It assumes that the insert point is
+// at the end of the current block.
+func CreateEntryBlockAlloca(builder llvm.Builder, t llvm.Type, name string) llvm.Value {
+	currentBlock := builder.GetInsertBlock()
+	entryBlock := currentBlock.Parent().EntryBasicBlock()
+	if entryBlock.FirstInstruction().IsNil() {
+		builder.SetInsertPointAtEnd(entryBlock)
+	} else {
+		builder.SetInsertPointBefore(entryBlock.FirstInstruction())
+	}
+	alloca := builder.CreateAlloca(t, name)
+	builder.SetInsertPointAtEnd(currentBlock)
+	return alloca
+}
+
+// CreateTemporaryAlloca creates a new alloca in the entry block and adds
+// lifetime start infromation in 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 using emitLifetimeEnd after you're done with it.
+func CreateTemporaryAlloca(builder llvm.Builder, mod llvm.Module, t llvm.Type, name string) (alloca, bitcast, size llvm.Value) {
+	ctx := t.Context()
+	targetData := llvm.NewTargetData(mod.DataLayout())
+	i8ptrType := llvm.PointerType(ctx.Int8Type(), 0)
+	alloca = CreateEntryBlockAlloca(builder, t, name)
+	bitcast = builder.CreateBitCast(alloca, i8ptrType, name+".bitcast")
+	size = llvm.ConstInt(ctx.Int64Type(), targetData.TypeAllocSize(t), false)
+	builder.CreateCall(getLifetimeStartFunc(mod), []llvm.Value{size, bitcast}, "")
+	return
+}
+
+// CreateInstructionAlloca creates an alloca in the entry block, and places lifetime control intrinsics around the instruction
+func CreateInstructionAlloca(builder llvm.Builder, mod llvm.Module, t llvm.Type, inst llvm.Value, name string) llvm.Value {
+	ctx := mod.Context()
+	targetData := llvm.NewTargetData(mod.DataLayout())
+	i8ptrType := llvm.PointerType(ctx.Int8Type(), 0)
+
+	alloca := CreateEntryBlockAlloca(builder, t, name)
+	builder.SetInsertPointBefore(inst)
+	bitcast := builder.CreateBitCast(alloca, i8ptrType, name+".bitcast")
+	size := llvm.ConstInt(ctx.Int64Type(), targetData.TypeAllocSize(t), false)
+	builder.CreateCall(getLifetimeStartFunc(mod), []llvm.Value{size, bitcast}, "")
+	if next := llvm.NextInstruction(inst); !next.IsNil() {
+		builder.SetInsertPointBefore(next)
+	} else {
+		builder.SetInsertPointAtEnd(inst.InstructionParent())
+	}
+	builder.CreateCall(getLifetimeEndFunc(mod), []llvm.Value{size, bitcast}, "")
+	return alloca
+}
+
+// EmitLifetimeEnd signals the end of an (alloca) lifetime by calling the
+// llvm.lifetime.end intrinsic. It is commonly used together with
+// createTemporaryAlloca.
+func EmitLifetimeEnd(builder llvm.Builder, mod llvm.Module, ptr, size llvm.Value) {
+	builder.CreateCall(getLifetimeEndFunc(mod), []llvm.Value{size, ptr}, "")
+}
+
+// getLifetimeStartFunc returns the llvm.lifetime.start intrinsic and creates it
+// first if it doesn't exist yet.
+func getLifetimeStartFunc(mod llvm.Module) llvm.Value {
+	fn := mod.NamedFunction("llvm.lifetime.start.p0i8")
+	ctx := mod.Context()
+	i8ptrType := llvm.PointerType(ctx.Int8Type(), 0)
+	if fn.IsNil() {
+		fnType := llvm.FunctionType(ctx.VoidType(), []llvm.Type{ctx.Int64Type(), i8ptrType}, false)
+		fn = llvm.AddFunction(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 getLifetimeEndFunc(mod llvm.Module) llvm.Value {
+	fn := mod.NamedFunction("llvm.lifetime.end.p0i8")
+	ctx := mod.Context()
+	i8ptrType := llvm.PointerType(ctx.Int8Type(), 0)
+	if fn.IsNil() {
+		fnType := llvm.FunctionType(ctx.VoidType(), []llvm.Type{ctx.Int64Type(), i8ptrType}, false)
+		fn = llvm.AddFunction(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.
+func SplitBasicBlock(builder llvm.Builder, afterInst llvm.Value, insertAfter llvm.BasicBlock, name string) llvm.BasicBlock {
+	oldBlock := afterInst.InstructionParent()
+	newBlock := afterInst.Type().Context().InsertBasicBlock(insertAfter, name)
+	var nextInstructions []llvm.Value // values to move
+
+	// Collect to-be-moved instructions.
+	inst := afterInst
+	for {
+		inst = llvm.NextInstruction(inst)
+		if inst.IsNil() {
+			break
+		}
+		nextInstructions = append(nextInstructions, inst)
+	}
+
+	// Move instructions.
+	builder.SetInsertPointAtEnd(newBlock)
+	for _, inst := range nextInstructions {
+		inst.RemoveFromParentAsInstruction()
+		builder.Insert(inst)
+	}
+
+	// Find PHI nodes to update.
+	var phiNodes []llvm.Value // PHI nodes to update
+	for bb := insertAfter.Parent().FirstBasicBlock(); !bb.IsNil(); bb = llvm.NextBasicBlock(bb) {
+		for inst := bb.FirstInstruction(); !inst.IsNil(); inst = llvm.NextInstruction(inst) {
+			if inst.IsAPHINode().IsNil() {
+				continue
+			}
+			needsUpdate := false
+			incomingCount := inst.IncomingCount()
+			for i := 0; i < incomingCount; i++ {
+				if inst.IncomingBlock(i) == oldBlock {
+					needsUpdate = true
+					break
+				}
+			}
+			if !needsUpdate {
+				// PHI node has no incoming edge from the old block.
+				continue
+			}
+			phiNodes = append(phiNodes, inst)
+		}
+	}
+
+	// Update PHI nodes.
+	for _, phi := range phiNodes {
+		builder.SetInsertPointBefore(phi)
+		newPhi := builder.CreatePHI(phi.Type(), "")
+		incomingCount := phi.IncomingCount()
+		incomingVals := make([]llvm.Value, incomingCount)
+		incomingBlocks := make([]llvm.BasicBlock, incomingCount)
+		for i := 0; i < incomingCount; i++ {
+			value := phi.IncomingValue(i)
+			block := phi.IncomingBlock(i)
+			if block == oldBlock {
+				block = newBlock
+			}
+			incomingVals[i] = value
+			incomingBlocks[i] = block
+		}
+		newPhi.AddIncoming(incomingVals, incomingBlocks)
+		phi.ReplaceAllUsesWith(newPhi)
+		phi.EraseFromParentAsInstruction()
+	}
+
+	return newBlock
+}

compiler/llvmutil/wordpack.go

@@ -0,0 +1,182 @@
+package llvmutil
+
+// 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.
+// If the values are all constants, they are be stored in a constant global and deduplicated.
+func EmitPointerPack(builder llvm.Builder, mod llvm.Module, needsStackObjects bool, values []llvm.Value) llvm.Value {
+	ctx := mod.Context()
+	targetData := llvm.NewTargetData(mod.DataLayout())
+	i8ptrType := llvm.PointerType(mod.Context().Int8Type(), 0)
+	uintptrType := ctx.IntType(llvm.NewTargetData(mod.DataLayout()).PointerSize() * 8)
+
+	valueTypes := make([]llvm.Type, len(values))
+	for i, value := range values {
+		valueTypes[i] = value.Type()
+	}
+	packedType := ctx.StructType(valueTypes, false)
+
+	// Allocate memory for the packed data.
+	size := targetData.TypeAllocSize(packedType)
+	if size == 0 {
+		return llvm.ConstPointerNull(i8ptrType)
+	} else if len(values) == 1 && values[0].Type().TypeKind() == llvm.PointerTypeKind {
+		return builder.CreateBitCast(values[0], i8ptrType, "pack.ptr")
+	} else if size <= targetData.TypeAllocSize(i8ptrType) {
+		// 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 builder.CreateIntToPtr(values[0], i8ptrType, "pack.int")
+		}
+
+		// Because packedType is a struct and we have to cast it to a *i8, store
+		// it in a *i8 alloca first and load the *i8 value from there. This is
+		// effectively a bitcast.
+		packedAlloc, _, _ := CreateTemporaryAlloca(builder, mod, i8ptrType, "")
+
+		if size < targetData.TypeAllocSize(i8ptrType) {
+			// The alloca is bigger than the value that will be stored in it.
+			// To avoid having some bits undefined, zero the alloca first.
+			// Hopefully this will get optimized away.
+			builder.CreateStore(llvm.ConstNull(i8ptrType), packedAlloc)
+		}
+
+		// Store all values in the alloca.
+		packedAllocCast := builder.CreateBitCast(packedAlloc, llvm.PointerType(packedType, 0), "")
+		for i, value := range values {
+			indices := []llvm.Value{
+				llvm.ConstInt(ctx.Int32Type(), 0, false),
+				llvm.ConstInt(ctx.Int32Type(), uint64(i), false),
+			}
+			gep := builder.CreateInBoundsGEP(packedAllocCast, indices, "")
+			builder.CreateStore(value, gep)
+		}
+
+		// Load value (the *i8) from the alloca.
+		result := builder.CreateLoad(packedAlloc, "")
+
+		// End the lifetime of the alloca, to help the optimizer.
+		packedPtr := builder.CreateBitCast(packedAlloc, i8ptrType, "")
+		packedSize := llvm.ConstInt(ctx.Int64Type(), targetData.TypeAllocSize(packedAlloc.Type()), false)
+		EmitLifetimeEnd(builder, mod, packedPtr, packedSize)
+
+		return result
+	} else {
+		// Check if the values are all constants.
+		constant := true
+		for _, v := range values {
+			if !v.IsConstant() {
+				constant = false
+				break
+			}
+		}
+
+		if constant {
+			// The data is known at compile time, so store it in a constant global.
+			// The global address is marked as unnamed, which allows LLVM to merge duplicates.
+			funcName := builder.GetInsertBlock().Parent().Name()
+			global := llvm.AddGlobal(mod, packedType, funcName+"$pack")
+			global.SetInitializer(ctx.ConstStruct(values, false))
+			global.SetGlobalConstant(true)
+			global.SetUnnamedAddr(true)
+			global.SetLinkage(llvm.PrivateLinkage)
+			return llvm.ConstBitCast(global, i8ptrType)
+		}
+
+		// Packed data is bigger than a pointer, so allocate it on the heap.
+		sizeValue := llvm.ConstInt(uintptrType, size, false)
+		alloc := mod.NamedFunction("runtime.alloc")
+		packedHeapAlloc := builder.CreateCall(alloc, []llvm.Value{
+			sizeValue,
+			llvm.Undef(i8ptrType),            // unused context parameter
+			llvm.ConstPointerNull(i8ptrType), // coroutine handle
+		}, "")
+		if needsStackObjects {
+			trackPointer := mod.NamedFunction("runtime.trackPointer")
+			builder.CreateCall(trackPointer, []llvm.Value{
+				packedHeapAlloc,
+				llvm.Undef(i8ptrType),            // unused context parameter
+				llvm.ConstPointerNull(i8ptrType), // coroutine handle
+			}, "")
+		}
+		packedAlloc := builder.CreateBitCast(packedHeapAlloc, llvm.PointerType(packedType, 0), "")
+
+		// Store all values in the heap pointer.
+		for i, value := range values {
+			indices := []llvm.Value{
+				llvm.ConstInt(ctx.Int32Type(), 0, false),
+				llvm.ConstInt(ctx.Int32Type(), uint64(i), false),
+			}
+			gep := builder.CreateInBoundsGEP(packedAlloc, indices, "")
+			builder.CreateStore(value, gep)
+		}
+
+		// Return the original heap allocation pointer, which already is an *i8.
+		return packedHeapAlloc
+	}
+}
+
+// EmitPointerUnpack extracts a list of values packed using EmitPointerPack.
+func EmitPointerUnpack(builder llvm.Builder, mod llvm.Module, ptr llvm.Value, valueTypes []llvm.Type) []llvm.Value {
+	ctx := mod.Context()
+	targetData := llvm.NewTargetData(mod.DataLayout())
+	i8ptrType := llvm.PointerType(mod.Context().Int8Type(), 0)
+	uintptrType := ctx.IntType(llvm.NewTargetData(mod.DataLayout()).PointerSize() * 8)
+
+	packedType := ctx.StructType(valueTypes, false)
+
+	// Get a correctly-typed pointer to the packed data.
+	var packedAlloc, packedRawAlloc llvm.Value
+	size := 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{builder.CreateBitCast(ptr, valueTypes[0], "unpack.ptr")}
+	} else if size <= targetData.TypeAllocSize(i8ptrType) {
+		// Packed data stored directly in pointer.
+		if len(valueTypes) == 1 && valueTypes[0].TypeKind() == llvm.IntegerTypeKind {
+			// Keep this cast in SSA form.
+			return []llvm.Value{builder.CreatePtrToInt(ptr, valueTypes[0], "unpack.int")}
+		}
+		// Fallback: load it using an alloca.
+		packedRawAlloc, _, _ = CreateTemporaryAlloca(builder, mod, llvm.PointerType(i8ptrType, 0), "unpack.raw.alloc")
+		packedRawValue := builder.CreateBitCast(ptr, llvm.PointerType(i8ptrType, 0), "unpack.raw.value")
+		builder.CreateStore(packedRawValue, packedRawAlloc)
+		packedAlloc = 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 = 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 targetData.TypeAllocSize(valueType) == 0 {
+			// This value has length zero, so there's nothing to load.
+			values[i] = llvm.ConstNull(valueType)
+			continue
+		}
+		indices := []llvm.Value{
+			llvm.ConstInt(ctx.Int32Type(), 0, false),
+			llvm.ConstInt(ctx.Int32Type(), uint64(i), false),
+		}
+		gep := builder.CreateInBoundsGEP(packedAlloc, indices, "")
+		values[i] = builder.CreateLoad(gep, "")
+	}
+	if !packedRawAlloc.IsNil() {
+		allocPtr := builder.CreateBitCast(packedRawAlloc, i8ptrType, "")
+		allocSize := llvm.ConstInt(ctx.Int64Type(), targetData.TypeAllocSize(uintptrType), false)
+		EmitLifetimeEnd(builder, mod, allocPtr, allocSize)
+	}
+	return values
+}

compiler/map.go

@@ -6,115 +6,167 @@ import (
 	"go/token"
 	"go/types"
 
+	"golang.org/x/tools/go/ssa"
 	"tinygo.org/x/go-llvm"
 )
 
-func (c *Compiler) emitMapLookup(keyType, valueType types.Type, m, key llvm.Value, commaOk bool, pos token.Pos) (llvm.Value, error) {
-	llvmValueType := c.getLLVMType(valueType)
+// createMakeMap creates a new map object (runtime.hashmap) by allocating and
+// initializing an appropriately sized object.
+func (b *builder) createMakeMap(expr *ssa.MakeMap) (llvm.Value, error) {
+	mapType := expr.Type().Underlying().(*types.Map)
+	keyType := mapType.Key().Underlying()
+	llvmValueType := b.getLLVMType(mapType.Elem().Underlying())
+	var llvmKeyType llvm.Type
+	if t, ok := keyType.(*types.Basic); ok && t.Info()&types.IsString != 0 {
+		// String keys.
+		llvmKeyType = b.getLLVMType(keyType)
+	} else if hashmapIsBinaryKey(keyType) {
+		// Trivially comparable keys.
+		llvmKeyType = b.getLLVMType(keyType)
+	} else {
+		// All other keys. Implemented as map[interface{}]valueType for ease of
+		// implementation.
+		llvmKeyType = b.getLLVMRuntimeType("_interface")
+	}
+	keySize := b.targetData.TypeAllocSize(llvmKeyType)
+	valueSize := b.targetData.TypeAllocSize(llvmValueType)
+	llvmKeySize := llvm.ConstInt(b.ctx.Int8Type(), keySize, false)
+	llvmValueSize := llvm.ConstInt(b.ctx.Int8Type(), valueSize, false)
+	sizeHint := llvm.ConstInt(b.uintptrType, 8, false)
+	if expr.Reserve != nil {
+		sizeHint = b.getValue(expr.Reserve)
+		var err error
+		sizeHint, err = b.createConvert(expr.Reserve.Type(), types.Typ[types.Uintptr], sizeHint, expr.Pos())
+		if err != nil {
+			return llvm.Value{}, err
+		}
+	}
+	hashmap := b.createRuntimeCall("hashmapMake", []llvm.Value{llvmKeySize, llvmValueSize, sizeHint}, "")
+	return hashmap, nil
+}
+
+// createMapLookup returns the value in a map. It calls a runtime function
+// depending on the map key type to load the map value and its comma-ok value.
+func (b *builder) createMapLookup(keyType, valueType types.Type, m, key llvm.Value, commaOk bool, pos token.Pos) (llvm.Value, error) {
+	llvmValueType := b.getLLVMType(valueType)
 
 	// Allocate the memory for the resulting type. Do not zero this memory: it
 	// will be zeroed by the hashmap get implementation if the key is not
 	// present in the map.
-	mapValueAlloca, mapValuePtr, mapValueSize := c.createTemporaryAlloca(llvmValueType, "hashmap.value")
+	mapValueAlloca, mapValuePtr, mapValueAllocaSize := b.createTemporaryAlloca(llvmValueType, "hashmap.value")
+
+	// We need the map size (with type uintptr) to pass to the hashmap*Get
+	// functions. This is necessary because those *Get functions are valid on
+	// nil maps, and they'll need to zero the value pointer by that number of
+	// bytes.
+	mapValueSize := mapValueAllocaSize
+	if mapValueSize.Type().IntTypeWidth() > b.uintptrType.IntTypeWidth() {
+		mapValueSize = llvm.ConstTrunc(mapValueSize, b.uintptrType)
+	}
 
 	// Do the lookup. How it is done depends on the key type.
 	var commaOkValue llvm.Value
+	keyType = keyType.Underlying()
 	if t, ok := keyType.(*types.Basic); ok && t.Info()&types.IsString != 0 {
 		// key is a string
-		params := []llvm.Value{m, key, mapValuePtr}
-		commaOkValue = c.createRuntimeCall("hashmapStringGet", params, "")
+		params := []llvm.Value{m, key, mapValuePtr, mapValueSize}
+		commaOkValue = b.createRuntimeCall("hashmapStringGet", params, "")
 	} else if hashmapIsBinaryKey(keyType) {
 		// key can be compared with runtime.memequal
 		// Store the key in an alloca, in the entry block to avoid dynamic stack
 		// growth.
-		mapKeyAlloca, mapKeyPtr, mapKeySize := c.createTemporaryAlloca(key.Type(), "hashmap.key")
-		c.builder.CreateStore(key, mapKeyAlloca)
+		mapKeyAlloca, mapKeyPtr, mapKeySize := b.createTemporaryAlloca(key.Type(), "hashmap.key")
+		b.CreateStore(key, mapKeyAlloca)
 		// Fetch the value from the hashmap.
-		params := []llvm.Value{m, mapKeyPtr, mapValuePtr}
-		commaOkValue = c.createRuntimeCall("hashmapBinaryGet", params, "")
-		c.emitLifetimeEnd(mapKeyPtr, mapKeySize)
+		params := []llvm.Value{m, mapKeyPtr, mapValuePtr, mapValueSize}
+		commaOkValue = b.createRuntimeCall("hashmapBinaryGet", params, "")
+		b.emitLifetimeEnd(mapKeyPtr, mapKeySize)
 	} else {
-		// Not trivially comparable using memcmp.
-		return llvm.Value{}, c.makeError(pos, "only strings, bools, ints, pointers or structs of bools/ints are supported as map keys, but got: "+keyType.String())
+		// Not trivially comparable using memcmp. Make it an interface instead.
+		itfKey := key
+		if _, ok := keyType.(*types.Interface); !ok {
+			// Not already an interface, so convert it to an interface now.
+			itfKey = b.createMakeInterface(key, keyType, pos)
+		}
+		params := []llvm.Value{m, itfKey, mapValuePtr, mapValueSize}
+		commaOkValue = b.createRuntimeCall("hashmapInterfaceGet", params, "")
 	}
 
 	// 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.emitLifetimeEnd(mapValuePtr, mapValueSize)
+	mapValue := b.CreateLoad(mapValueAlloca, "")
+	b.emitLifetimeEnd(mapValuePtr, mapValueAllocaSize)
 
 	if commaOk {
-		tuple := llvm.Undef(c.ctx.StructType([]llvm.Type{llvmValueType, c.ctx.Int1Type()}, false))
-		tuple = c.builder.CreateInsertValue(tuple, mapValue, 0, "")
-		tuple = c.builder.CreateInsertValue(tuple, commaOkValue, 1, "")
+		tuple := llvm.Undef(b.ctx.StructType([]llvm.Type{llvmValueType, b.ctx.Int1Type()}, false))
+		tuple = b.CreateInsertValue(tuple, mapValue, 0, "")
+		tuple = b.CreateInsertValue(tuple, commaOkValue, 1, "")
 		return tuple, nil
 	} else {
 		return mapValue, nil
 	}
 }
 
-func (c *Compiler) emitMapUpdate(keyType types.Type, m, key, value llvm.Value, pos token.Pos) {
-	valueAlloca, valuePtr, valueSize := c.createTemporaryAlloca(value.Type(), "hashmap.value")
-	c.builder.CreateStore(value, valueAlloca)
+// createMapUpdate updates a map key to a given value, by creating an
+// appropriate runtime call.
+func (b *builder) createMapUpdate(keyType types.Type, m, key, value llvm.Value, pos token.Pos) {
+	valueAlloca, valuePtr, valueSize := b.createTemporaryAlloca(value.Type(), "hashmap.value")
+	b.CreateStore(value, valueAlloca)
 	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, "")
+		b.createRuntimeCall("hashmapStringSet", params, "")
 	} else if hashmapIsBinaryKey(keyType) {
 		// key can be compared with runtime.memequal
-		keyAlloca, keyPtr, keySize := c.createTemporaryAlloca(key.Type(), "hashmap.key")
-		c.builder.CreateStore(key, keyAlloca)
+		keyAlloca, keyPtr, keySize := b.createTemporaryAlloca(key.Type(), "hashmap.key")
+		b.CreateStore(key, keyAlloca)
 		params := []llvm.Value{m, keyPtr, valuePtr}
-		c.createRuntimeCall("hashmapBinarySet", params, "")
-		c.emitLifetimeEnd(keyPtr, keySize)
+		b.createRuntimeCall("hashmapBinarySet", params, "")
+		b.emitLifetimeEnd(keyPtr, keySize)
 	} else {
-		c.addError(pos, "only strings, bools, ints, pointers or structs of bools/ints are supported as map keys, but got: "+keyType.String())
+		// Key is not trivially comparable, so compare it as an interface instead.
+		itfKey := key
+		if _, ok := keyType.(*types.Interface); !ok {
+			// Not already an interface, so convert it to an interface first.
+			itfKey = b.createMakeInterface(key, keyType, pos)
+		}
+		params := []llvm.Value{m, itfKey, valuePtr}
+		b.createRuntimeCall("hashmapInterfaceSet", params, "")
 	}
-	c.emitLifetimeEnd(valuePtr, valueSize)
+	b.emitLifetimeEnd(valuePtr, valueSize)
 }
 
-func (c *Compiler) emitMapDelete(keyType types.Type, m, key llvm.Value, pos token.Pos) error {
+// createMapDelete deletes a key from a map by calling the appropriate runtime
+// function. It is the implementation of the Go delete() builtin.
+func (b *builder) createMapDelete(keyType types.Type, m, key llvm.Value, pos token.Pos) error {
 	keyType = keyType.Underlying()
 	if t, ok := keyType.(*types.Basic); ok && t.Info()&types.IsString != 0 {
 		// key is a string
 		params := []llvm.Value{m, key}
-		c.createRuntimeCall("hashmapStringDelete", params, "")
+		b.createRuntimeCall("hashmapStringDelete", params, "")
 		return nil
 	} else if hashmapIsBinaryKey(keyType) {
-		keyAlloca, keyPtr, keySize := c.createTemporaryAlloca(key.Type(), "hashmap.key")
-		c.builder.CreateStore(key, keyAlloca)
+		keyAlloca, keyPtr, keySize := b.createTemporaryAlloca(key.Type(), "hashmap.key")
+		b.CreateStore(key, keyAlloca)
 		params := []llvm.Value{m, keyPtr}
-		c.createRuntimeCall("hashmapBinaryDelete", params, "")
-		c.emitLifetimeEnd(keyPtr, keySize)
+		b.createRuntimeCall("hashmapBinaryDelete", params, "")
+		b.emitLifetimeEnd(keyPtr, keySize)
 		return nil
 	} else {
-		return c.makeError(pos, "only strings, bools, ints, pointers or structs of bools/ints are supported as map keys, but got: "+keyType.String())
+		// Key is not trivially comparable, so compare it as an interface
+		// instead.
+		itfKey := key
+		if _, ok := keyType.(*types.Interface); !ok {
+			// Not already an interface, so convert it to an interface first.
+			itfKey = b.createMakeInterface(key, keyType, pos)
+		}
+		params := []llvm.Value{m, itfKey}
+		b.createRuntimeCall("hashmapInterfaceDelete", params, "")
+		return nil
 	}
 }
 
-// Get FNV-1a hash of this string.
-//
-// https://en.wikipedia.org/wiki/Fowler%E2%80%93Noll%E2%80%93Vo_hash_function#FNV-1a_hash
-func hashmapHash(data []byte) uint32 {
-	var result uint32 = 2166136261 // FNV offset basis
-	for _, c := range data {
-		result ^= uint32(c)
-		result *= 16777619 // FNV prime
-	}
-	return result
-}
-
-// Get the topmost 8 bits of the hash, without using a special value (like 0).
-func hashmapTopHash(hash uint32) uint8 {
-	tophash := uint8(hash >> 24)
-	if tophash < 1 {
-		// 0 means empty slot, so make it bigger.
-		tophash += 1
-	}
-	return tophash
-}
-
 // Returns true if this key type does not contain strings, interfaces etc., so
 // can be compared with runtime.memequal.
 func hashmapIsBinaryKey(keyType types.Type) bool {

compiler/optimizer.go

@@ -1,260 +0,0 @@
-package compiler
-
-import (
-	"errors"
-
-	"github.com/tinygo-org/tinygo/transform"
-	"tinygo.org/x/go-llvm"
-)
-
-// Run the LLVM optimizer over the module.
-// The inliner can be disabled (if necessary) by passing 0 to the inlinerThreshold.
-func (c *Compiler) Optimize(optLevel, sizeLevel int, inlinerThreshold uint) error {
-	builder := llvm.NewPassManagerBuilder()
-	defer builder.Dispose()
-	builder.SetOptLevel(optLevel)
-	builder.SetSizeLevel(sizeLevel)
-	if inlinerThreshold != 0 {
-		builder.UseInlinerWithThreshold(inlinerThreshold)
-	}
-	builder.AddCoroutinePassesToExtensionPoints()
-
-	if c.PanicStrategy == "trap" {
-		c.replacePanicsWithTrap() // -panic=trap
-	}
-
-	// run a check of all of our code
-	if c.VerifyIR {
-		c.checkModule()
-	}
-
-	// Run function passes for each function.
-	funcPasses := llvm.NewFunctionPassManagerForModule(c.mod)
-	defer funcPasses.Dispose()
-	builder.PopulateFunc(funcPasses)
-	funcPasses.InitializeFunc()
-	for fn := c.mod.FirstFunction(); !fn.IsNil(); fn = llvm.NextFunction(fn) {
-		funcPasses.RunFunc(fn)
-	}
-	funcPasses.FinalizeFunc()
-
-	if optLevel > 0 {
-		// Run some preparatory passes for the Go optimizer.
-		goPasses := llvm.NewPassManager()
-		defer goPasses.Dispose()
-		goPasses.AddGlobalOptimizerPass()
-		goPasses.AddConstantPropagationPass()
-		goPasses.AddAggressiveDCEPass()
-		goPasses.AddFunctionAttrsPass()
-		goPasses.Run(c.mod)
-
-		// Run Go-specific optimization passes.
-		transform.OptimizeMaps(c.mod)
-		c.OptimizeStringToBytes()
-		transform.OptimizeAllocs(c.mod)
-		c.LowerInterfaces()
-		c.LowerFuncValues()
-
-		// After interfaces are lowered, there are many more opportunities for
-		// interprocedural optimizations. To get them to work, function
-		// attributes have to be updated first.
-		goPasses.Run(c.mod)
-
-		// Run TinyGo-specific interprocedural optimizations.
-		transform.OptimizeAllocs(c.mod)
-		c.OptimizeStringToBytes()
-
-		// Lower runtime.isnil calls to regular nil comparisons.
-		isnil := c.mod.NamedFunction("runtime.isnil")
-		if !isnil.IsNil() {
-			for _, use := range getUses(isnil) {
-				c.builder.SetInsertPointBefore(use)
-				ptr := use.Operand(0)
-				if !ptr.IsABitCastInst().IsNil() {
-					ptr = ptr.Operand(0)
-				}
-				nilptr := llvm.ConstPointerNull(ptr.Type())
-				icmp := c.builder.CreateICmp(llvm.IntEQ, ptr, nilptr, "")
-				use.ReplaceAllUsesWith(icmp)
-				use.EraseFromParentAsInstruction()
-			}
-		}
-
-		err := c.LowerGoroutines()
-		if err != nil {
-			return err
-		}
-	} else {
-		// Must be run at any optimization level.
-		c.LowerInterfaces()
-		c.LowerFuncValues()
-		err := c.LowerGoroutines()
-		if err != nil {
-			return err
-		}
-	}
-	if err := c.Verify(); err != nil {
-		return errors.New("optimizations caused a verification failure")
-	}
-
-	if sizeLevel >= 2 {
-		// Set the "optsize" attribute to make slightly smaller binaries at the
-		// cost of some performance.
-		kind := llvm.AttributeKindID("optsize")
-		attr := c.ctx.CreateEnumAttribute(kind, 0)
-		for fn := c.mod.FirstFunction(); !fn.IsNil(); fn = llvm.NextFunction(fn) {
-			fn.AddFunctionAttr(attr)
-		}
-	}
-
-	// After TinyGo-specific transforms have finished, undo exporting these functions.
-	for _, name := range functionsUsedInTransforms {
-		fn := c.mod.NamedFunction(name)
-		if fn.IsNil() {
-			continue
-		}
-		fn.SetLinkage(llvm.InternalLinkage)
-	}
-
-	// Run function passes again, because without it, llvm.coro.size.i32()
-	// doesn't get lowered.
-	for fn := c.mod.FirstFunction(); !fn.IsNil(); fn = llvm.NextFunction(fn) {
-		funcPasses.RunFunc(fn)
-	}
-	funcPasses.FinalizeFunc()
-
-	// Run module passes.
-	modPasses := llvm.NewPassManager()
-	defer modPasses.Dispose()
-	builder.Populate(modPasses)
-	modPasses.Run(c.mod)
-
-	hasGCPass := c.addGlobalsBitmap()
-	hasGCPass = c.makeGCStackSlots() || hasGCPass
-	if hasGCPass {
-		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, "")
-		}
-	}
-}
-
-// Transform runtime.stringToBytes(...) calls into const []byte slices whenever
-// possible. This optimizes the following pattern:
-//     w.Write([]byte("foo"))
-// where Write does not store to the slice.
-func (c *Compiler) OptimizeStringToBytes() {
-	stringToBytes := c.mod.NamedFunction("runtime.stringToBytes")
-	if stringToBytes.IsNil() {
-		// nothing to optimize
-		return
-	}
-
-	for _, call := range getUses(stringToBytes) {
-		strptr := call.Operand(0)
-		strlen := call.Operand(1)
-
-		// strptr is always constant because strings are always constant.
-
-		convertedAllUses := true
-		for _, use := range getUses(call) {
-			nilValue := llvm.Value{}
-			if use.IsAExtractValueInst() == nilValue {
-				convertedAllUses = false
-				continue
-			}
-			switch use.Type().TypeKind() {
-			case llvm.IntegerTypeKind:
-				// A length (len or cap). Propagate the length value.
-				use.ReplaceAllUsesWith(strlen)
-				use.EraseFromParentAsInstruction()
-			case llvm.PointerTypeKind:
-				// The string pointer itself.
-				if !c.isReadOnly(use) {
-					convertedAllUses = false
-					continue
-				}
-				use.ReplaceAllUsesWith(strptr)
-				use.EraseFromParentAsInstruction()
-			default:
-				// should not happen
-				panic("unknown return type of runtime.stringToBytes: " + use.Type().String())
-			}
-		}
-		if convertedAllUses {
-			// Call to runtime.stringToBytes can be eliminated: both the input
-			// and the output is constant.
-			call.EraseFromParentAsInstruction()
-		}
-	}
-}
-
-// Check whether the given value (which is of pointer type) is never stored to.
-func (c *Compiler) isReadOnly(value llvm.Value) bool {
-	uses := getUses(value)
-	for _, use := range uses {
-		nilValue := llvm.Value{}
-		if use.IsAGetElementPtrInst() != nilValue {
-			if !c.isReadOnly(use) {
-				return false
-			}
-		} else if use.IsACallInst() != nilValue {
-			if !c.hasFlag(use, value, "readonly") {
-				return false
-			}
-		} else {
-			// Unknown instruction, might not be readonly.
-			return false
-		}
-	}
-	return true
-}
-
-// Check whether all uses of this param as parameter to the call have the given
-// flag. In most cases, there will only be one use but a function could take the
-// same parameter twice, in which case both must have the flag.
-// A flag can be any enum flag, like "readonly".
-func (c *Compiler) hasFlag(call, param llvm.Value, kind string) bool {
-	fn := call.CalledValue()
-	nilValue := llvm.Value{}
-	if fn.IsAFunction() == nilValue {
-		// This is not a function but something else, like a function pointer.
-		return false
-	}
-	kindID := llvm.AttributeKindID(kind)
-	for i := 0; i < fn.ParamsCount(); i++ {
-		if call.Operand(i) != param {
-			// This is not the parameter we're checking.
-			continue
-		}
-		index := i + 1 // param attributes start at 1
-		attr := fn.GetEnumAttributeAtIndex(index, kindID)
-		nilAttribute := llvm.Attribute{}
-		if attr == nilAttribute {
-			// At least one parameter doesn't have the flag (there may be
-			// multiple).
-			return false
-		}
-	}
-	return true
-}

compiler/sizes.go

@@ -10,13 +10,13 @@ import (
 // The original license can be found here:
 //     https://golang.org/LICENSE
 
-type StdSizes struct {
+type stdSizes struct {
 	IntSize  int64
 	PtrSize  int64
 	MaxAlign int64
 }
 
-func (s *StdSizes) Alignof(T types.Type) int64 {
+func (s *stdSizes) Alignof(T types.Type) int64 {
 	// For arrays and structs, alignment is defined in terms
 	// of alignment of the elements and fields, respectively.
 	switch t := T.Underlying().(type) {
@@ -61,14 +61,8 @@ func (s *StdSizes) Alignof(T types.Type) int64 {
 	return a
 }
 
-func (s *StdSizes) Offsetsof(fields []*types.Var) []int64 {
+func (s *stdSizes) Offsetsof(fields []*types.Var) []int64 {
 	offsets := make([]int64, len(fields))
-	if len(fields) > 1 && fields[0].Name() == "C union" {
-		// This struct contains the magic "C union" field which indicates that
-		// this is actually a union from CGo.
-		// All fields in the union start at 0 so return that.
-		return offsets // all fields are still set to 0
-	}
 	var o int64
 	for i, f := range fields {
 		a := s.Alignof(f.Type())
@@ -95,7 +89,7 @@ var basicSizes = [...]byte{
 	types.Complex128: 16,
 }
 
-func (s *StdSizes) Sizeof(T types.Type) int64 {
+func (s *stdSizes) Sizeof(T types.Type) int64 {
 	switch t := T.Underlying().(type) {
 	case *types.Basic:
 		k := t.Kind()
@@ -116,6 +110,9 @@ func (s *StdSizes) Sizeof(T types.Type) int64 {
 		if k == types.UnsafePointer {
 			return s.PtrSize
 		}
+		if k == types.Invalid {
+			return 0 // only relevant when there is a type error somewhere
+		}
 		panic("unknown basic type: " + t.String())
 	case *types.Array:
 		n := t.Len()
@@ -143,33 +140,19 @@ func (s *StdSizes) Sizeof(T types.Type) int64 {
 				maxAlign = al
 			}
 		}
-		if fields[0].Name() == "C union" {
-			// Magic field that indicates this is a CGo union and not a struct.
-			// The size is the biggest element, aligned to the element with the
-			// biggest alignment. This is not necessarily the same, for example
-			// in the following union:
-			//     union { int32_t l; int16_t s[3] }
-			maxSize := int64(0)
-			for _, field := range fields[1:] {
-				si := s.Sizeof(field.Type())
-				if si > maxSize {
-					maxSize = si
-				}
-			}
-			return align(maxSize, maxAlign)
-		} else {
-			// This is a regular struct.
-			// Pick the size that fits this struct and add some alignment. Some
-			// structs have some extra padding at the end which should also be
-			// taken care of:
-			//     struct { int32 n; byte b }
-			offsets := s.Offsetsof(fields)
-			return align(offsets[n-1]+s.Sizeof(fields[n-1].Type()), maxAlign)
-		}
+		// Pick the size that fits this struct and add some alignment. Some
+		// structs have some extra padding at the end which should also be taken
+		// care of:
+		//     struct { int32 n; byte b }
+		offsets := s.Offsetsof(fields)
+		return align(offsets[n-1]+s.Sizeof(fields[n-1].Type()), maxAlign)
 	case *types.Interface:
 		return s.PtrSize * 2
 	case *types.Pointer:
 		return s.PtrSize
+	case *types.Signature:
+		// Func values in TinyGo are two words in size.
+		return s.PtrSize * 2
 	default:
 		panic("unknown type: " + t.String())
 	}

compiler/symbol.go

@@ -7,6 +7,7 @@ import (
 	"go/ast"
 	"go/token"
 	"go/types"
+	"strconv"
 	"strings"
 
 	"github.com/tinygo-org/tinygo/loader"
@@ -14,35 +15,343 @@ import (
 	"tinygo.org/x/go-llvm"
 )
 
+// functionInfo contains some information about a function or method. In
+// particular, it contains information obtained from pragmas.
+//
+// The linkName value contains a valid link name, even if //go:linkname is not
+// present.
+type functionInfo struct {
+	module     string     // go:wasm-module
+	importName string     // go:linkname, go:export - The name the developer assigns
+	linkName   string     // go:linkname, go:export - The name that we map for the particular module -> importName
+	section    string     // go:section - object file section name
+	exported   bool       // go:export, CGo
+	nobounds   bool       // go:nobounds
+	variadic   bool       // go:variadic (CGo only)
+	inline     inlineType // go:inline
+}
+
+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
+
+	// Don't inline, just like the GCC noinline attribute. Signalled using
+	// //go:noinline.
+	inlineNone
+)
+
+// getFunction returns the LLVM function for the given *ssa.Function, creating
+// it if needed. It can later be filled with compilerContext.createFunction().
+func (c *compilerContext) getFunction(fn *ssa.Function) llvm.Value {
+	info := c.getFunctionInfo(fn)
+	llvmFn := c.mod.NamedFunction(info.linkName)
+	if !llvmFn.IsNil() {
+		return llvmFn
+	}
+
+	var retType llvm.Type
+	if fn.Signature.Results() == nil {
+		retType = c.ctx.VoidType()
+	} else if fn.Signature.Results().Len() == 1 {
+		retType = c.getLLVMType(fn.Signature.Results().At(0).Type())
+	} else {
+		results := make([]llvm.Type, 0, fn.Signature.Results().Len())
+		for i := 0; i < fn.Signature.Results().Len(); i++ {
+			results = append(results, c.getLLVMType(fn.Signature.Results().At(i).Type()))
+		}
+		retType = c.ctx.StructType(results, false)
+	}
+
+	var paramInfos []paramInfo
+	for _, param := range getParams(fn.Signature) {
+		paramType := c.getLLVMType(param.Type())
+		paramFragmentInfos := c.expandFormalParamType(paramType, param.Name(), param.Type())
+		paramInfos = append(paramInfos, paramFragmentInfos...)
+	}
+
+	// Add an extra parameter as the function context. This context is used in
+	// closures and bound methods, but should be optimized away when not used.
+	if !info.exported {
+		paramInfos = append(paramInfos, paramInfo{llvmType: c.i8ptrType, name: "context", flags: 0})
+		paramInfos = append(paramInfos, paramInfo{llvmType: c.i8ptrType, name: "parentHandle", flags: 0})
+	}
+
+	var paramTypes []llvm.Type
+	for _, info := range paramInfos {
+		paramTypes = append(paramTypes, info.llvmType)
+	}
+
+	fnType := llvm.FunctionType(retType, paramTypes, info.variadic)
+	llvmFn = llvm.AddFunction(c.mod, info.linkName, fnType)
+	if strings.HasPrefix(c.Triple, "wasm") {
+		// C functions without prototypes like this:
+		//   void foo();
+		// are actually variadic functions. However, it appears that it has been
+		// decided in WebAssembly that such prototype-less functions are not
+		// allowed in WebAssembly.
+		// In C, this can only happen when there are zero parameters, hence this
+		// check here. For more information:
+		// https://reviews.llvm.org/D48443
+		// https://github.com/WebAssembly/tool-conventions/issues/16
+		if info.variadic && len(fn.Params) == 0 {
+			attr := c.ctx.CreateStringAttribute("no-prototype", "")
+			llvmFn.AddFunctionAttr(attr)
+		}
+	}
+
+	dereferenceableOrNullKind := llvm.AttributeKindID("dereferenceable_or_null")
+	for i, info := range paramInfos {
+		if info.flags&paramIsDeferenceableOrNull == 0 {
+			continue
+		}
+		if info.llvmType.TypeKind() == llvm.PointerTypeKind {
+			el := info.llvmType.ElementType()
+			size := c.targetData.TypeAllocSize(el)
+			if size == 0 {
+				// dereferenceable_or_null(0) appears to be illegal in LLVM.
+				continue
+			}
+			dereferenceableOrNull := c.ctx.CreateEnumAttribute(dereferenceableOrNullKind, size)
+			llvmFn.AddAttributeAtIndex(i+1, dereferenceableOrNull)
+		}
+	}
+
+	// Set a number of function or parameter attributes, depending on the
+	// function. These functions are runtime functions that are known to have
+	// certain attributes that might not be inferred by the compiler.
+	switch info.linkName {
+	case "abort":
+		// On *nix systems, the "abort" functuion in libc is used to handle fatal panics.
+		// Mark it as noreturn so LLVM can optimize away code.
+		llvmFn.AddFunctionAttr(c.ctx.CreateEnumAttribute(llvm.AttributeKindID("noreturn"), 0))
+	case "runtime.alloc":
+		// Tell the optimizer that runtime.alloc is an allocator, meaning that it
+		// returns values that are never null and never alias to an existing value.
+		for _, attrName := range []string{"noalias", "nonnull"} {
+			llvmFn.AddAttributeAtIndex(0, c.ctx.CreateEnumAttribute(llvm.AttributeKindID(attrName), 0))
+		}
+	case "runtime.sliceAppend":
+		// Appending a slice will only read the to-be-appended slice, it won't
+		// be modified.
+		llvmFn.AddAttributeAtIndex(2, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("nocapture"), 0))
+		llvmFn.AddAttributeAtIndex(2, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("readonly"), 0))
+	case "runtime.sliceCopy":
+		// Copying a slice won't capture any of the parameters.
+		llvmFn.AddAttributeAtIndex(1, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("writeonly"), 0))
+		llvmFn.AddAttributeAtIndex(1, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("nocapture"), 0))
+		llvmFn.AddAttributeAtIndex(2, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("readonly"), 0))
+		llvmFn.AddAttributeAtIndex(2, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("nocapture"), 0))
+	case "runtime.trackPointer":
+		// This function is necessary for tracking pointers on the stack in a
+		// portable way (see gc_stack_portable.go). Indicate to the optimizer
+		// that the only thing we'll do is read the pointer.
+		llvmFn.AddAttributeAtIndex(1, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("nocapture"), 0))
+		llvmFn.AddAttributeAtIndex(1, c.ctx.CreateEnumAttribute(llvm.AttributeKindID("readonly"), 0))
+	}
+
+	// External/exported functions may not retain pointer values.
+	// https://golang.org/cmd/cgo/#hdr-Passing_pointers
+	if info.exported {
+		// Set the wasm-import-module attribute if the function's module is set.
+		if info.module != "" {
+
+			// We need to add the wasm-import-module and the wasm-import-name
+			wasmImportModuleAttr := c.ctx.CreateStringAttribute("wasm-import-module", info.module)
+			llvmFn.AddFunctionAttr(wasmImportModuleAttr)
+
+			// Add the Wasm Import Name, if we are a named wasm import
+			if info.importName != "" {
+				wasmImportNameAttr := c.ctx.CreateStringAttribute("wasm-import-name", info.importName)
+				llvmFn.AddFunctionAttr(wasmImportNameAttr)
+			}
+		}
+		nocaptureKind := llvm.AttributeKindID("nocapture")
+		nocapture := c.ctx.CreateEnumAttribute(nocaptureKind, 0)
+		for i, typ := range paramTypes {
+			if typ.TypeKind() == llvm.PointerTypeKind {
+				llvmFn.AddAttributeAtIndex(i+1, nocapture)
+			}
+		}
+	}
+
+	// Synthetic functions are functions that do not appear in the source code,
+	// they are artificially constructed. Usually they are wrapper functions
+	// that are not referenced anywhere except in a SSA call instruction so
+	// should be created right away.
+	// The exception is the package initializer, which does appear in the
+	// *ssa.Package members and so shouldn't be created here.
+	if fn.Synthetic != "" && fn.Synthetic != "package initializer" {
+		irbuilder := c.ctx.NewBuilder()
+		b := newBuilder(c, irbuilder, fn)
+		b.createFunction()
+		irbuilder.Dispose()
+		llvmFn.SetLinkage(llvm.LinkOnceODRLinkage)
+		llvmFn.SetUnnamedAddr(true)
+	}
+
+	return llvmFn
+}
+
+// getFunctionInfo returns information about a function that is not directly
+// present in *ssa.Function, such as the link name and whether it should be
+// exported.
+func (c *compilerContext) getFunctionInfo(f *ssa.Function) functionInfo {
+	info := functionInfo{}
+	if strings.HasPrefix(f.Name(), "C.") {
+		// Created by CGo: such a name cannot be created by regular C code.
+		info.linkName = f.Name()[2:]
+		info.exported = true
+	} else {
+		// Pick the default linkName.
+		info.linkName = f.RelString(nil)
+	}
+	// Check for //go: pragmas, which may change the link name (among others).
+	info.parsePragmas(f)
+	return info
+}
+
+// parsePragmas is used by getFunctionInfo to parse function pragmas such as
+// //export or //go:noinline.
+func (info *functionInfo) parsePragmas(f *ssa.Function) {
+	if f.Syntax() == nil {
+		return
+	}
+	if decl, ok := f.Syntax().(*ast.FuncDecl); ok && decl.Doc != nil {
+
+		// Our importName for a wasm module (if we are compiling to wasm), or llvm link name
+		var importName string
+
+		for _, comment := range decl.Doc.List {
+			text := comment.Text
+			if strings.HasPrefix(text, "//export ") {
+				// Rewrite '//export' to '//go:export' for compatibility with
+				// gc.
+				text = "//go:" + text[2:]
+			}
+			if !strings.HasPrefix(text, "//go:") {
+				continue
+			}
+			parts := strings.Fields(text)
+			switch parts[0] {
+			case "//go:export":
+				if len(parts) != 2 {
+					continue
+				}
+
+				importName = parts[1]
+				info.exported = true
+			case "//go:wasm-module":
+				// Alternative comment for setting the import module.
+				if len(parts) != 2 {
+					continue
+				}
+				info.module = parts[1]
+			case "//go:inline":
+				info.inline = inlineHint
+			case "//go:noinline":
+				info.inline = inlineNone
+			case "//go:linkname":
+				if len(parts) != 3 || parts[1] != f.Name() {
+					continue
+				}
+				// Only enable go:linkname when the package imports "unsafe".
+				// This is a slightly looser requirement than what gc uses: gc
+				// requires the file to import "unsafe", not the package as a
+				// whole.
+				if hasUnsafeImport(f.Pkg.Pkg) {
+					info.linkName = parts[2]
+				}
+			case "//go:section":
+				if len(parts) == 2 && hasUnsafeImport(f.Pkg.Pkg) {
+					info.section = parts[1]
+				}
+			case "//go:nobounds":
+				// Skip bounds checking in this function. Useful for some
+				// runtime functions.
+				// This is somewhat dangerous and thus only imported in packages
+				// that import unsafe.
+				if hasUnsafeImport(f.Pkg.Pkg) {
+					info.nobounds = true
+				}
+			case "//go:variadic":
+				// The //go:variadic pragma is emitted by the CGo preprocessing
+				// pass for C variadic functions. This includes both explicit
+				// (with ...) and implicit (no parameters in signature)
+				// functions.
+				if strings.HasPrefix(f.Name(), "C.") {
+					// This prefix cannot naturally be created, it must have
+					// been created as a result of CGo preprocessing.
+					info.variadic = true
+				}
+			}
+		}
+
+		// Set the importName for our exported function if we have one
+		if importName != "" {
+			if info.module == "" {
+				info.linkName = importName
+			} else {
+				// WebAssembly import
+				info.importName = importName
+			}
+		}
+
+	}
+}
+
+// getParams returns the function parameters, including the receiver at the
+// start. This is an alternative to the Params member of *ssa.Function, which is
+// not yet populated when the package has not yet been built.
+func getParams(sig *types.Signature) []*types.Var {
+	params := []*types.Var{}
+	if sig.Recv() != nil {
+		params = append(params, sig.Recv())
+	}
+	for i := 0; i < sig.Params().Len(); i++ {
+		params = append(params, sig.Params().At(i))
+	}
+	return params
+}
+
 // globalInfo contains some information about a specific global. By default,
 // linkName is equal to .RelString(nil) on a global and extern is false, but for
 // some symbols this is different (due to //go:extern for example).
 type globalInfo struct {
 	linkName string // go:extern
 	extern   bool   // go:extern
+	align    int    // go:align
+	section  string // go:section
+	embeds   string // go:embed
 }
 
 // loadASTComments loads comments on globals from the AST, for use later in the
 // program. In particular, they are required for //go:extern pragmas on globals.
-func (c *Compiler) loadASTComments(lprogram *loader.Program) {
-	c.astComments = map[string]*ast.CommentGroup{}
-	for _, pkgInfo := range lprogram.Sorted() {
-		for _, file := range pkgInfo.Files {
-			for _, decl := range file.Decls {
-				switch decl := decl.(type) {
-				case *ast.GenDecl:
-					switch decl.Tok {
-					case token.VAR:
-						if len(decl.Specs) != 1 {
-							continue
-						}
-						for _, spec := range decl.Specs {
-							switch spec := spec.(type) {
-							case *ast.ValueSpec: // decl.Tok == token.VAR
-								for _, name := range spec.Names {
-									id := pkgInfo.Pkg.Path() + "." + name.Name
-									c.astComments[id] = decl.Doc
-								}
+func (c *compilerContext) loadASTComments(pkg *loader.Package) {
+	for _, file := range pkg.Files {
+		for _, decl := range file.Decls {
+			switch decl := decl.(type) {
+			case *ast.GenDecl:
+				switch decl.Tok {
+				case token.VAR:
+					if len(decl.Specs) != 1 {
+						continue
+					}
+					for _, spec := range decl.Specs {
+						switch spec := spec.(type) {
+						case *ast.ValueSpec: // decl.Tok == token.VAR
+							for _, name := range spec.Names {
+								id := pkg.Pkg.Path() + "." + name.Name
+								c.astComments[id] = decl.Doc
 							}
 						}
 					}
@@ -54,22 +363,51 @@ func (c *Compiler) loadASTComments(lprogram *loader.Program) {
 
 // getGlobal returns a LLVM IR global value for a Go SSA global. It is added to
 // the LLVM IR if it has not been added already.
-func (c *Compiler) getGlobal(g *ssa.Global) llvm.Value {
+func (c *compilerContext) getGlobal(g *ssa.Global) llvm.Value {
 	info := c.getGlobalInfo(g)
 	llvmGlobal := c.mod.NamedGlobal(info.linkName)
 	if llvmGlobal.IsNil() {
-		llvmType := c.getLLVMType(g.Type().(*types.Pointer).Elem())
+		typ := g.Type().(*types.Pointer).Elem()
+		llvmType := c.getLLVMType(typ)
 		llvmGlobal = llvm.AddGlobal(c.mod, llvmType, info.linkName)
-		if !info.extern {
-			llvmGlobal.SetInitializer(llvm.ConstNull(llvmType))
-			llvmGlobal.SetLinkage(llvm.InternalLinkage)
+
+		// Set alignment from the //go:align comment.
+		var alignInBits uint32
+		alignment := c.targetData.ABITypeAlignment(llvmType)
+		if info.align > alignment {
+			alignment = info.align
+		}
+		if alignment <= 0 || alignment&(alignment-1) != 0 {
+			// Check for power-of-two (or 0).
+			// See: https://stackoverflow.com/a/108360
+			c.addError(g.Pos(), "global variable alignment must be a positive power of two")
+		} else {
+			// Set the alignment only when it is a power of two.
+			alignInBits = uint32(alignment) ^ uint32(alignment-1)
+			llvmGlobal.SetAlignment(alignment)
+		}
+
+		if c.Debug && !info.extern {
+			// Add debug info.
+			pos := c.program.Fset.Position(g.Pos())
+			diglobal := c.dibuilder.CreateGlobalVariableExpression(c.difiles[pos.Filename], llvm.DIGlobalVariableExpression{
+				Name:        g.RelString(nil),
+				LinkageName: info.linkName,
+				File:        c.getDIFile(pos.Filename),
+				Line:        pos.Line,
+				Type:        c.getDIType(typ),
+				LocalToUnit: false,
+				Expr:        c.dibuilder.CreateExpression(nil),
+				AlignInBits: alignInBits,
+			})
+			llvmGlobal.AddMetadata(0, diglobal)
 		}
 	}
 	return llvmGlobal
 }
 
 // getGlobalInfo returns some information about a specific global.
-func (c *Compiler) getGlobalInfo(g *ssa.Global) globalInfo {
+func (c *compilerContext) getGlobalInfo(g *ssa.Global) globalInfo {
 	info := globalInfo{}
 	if strings.HasPrefix(g.Name(), "C.") {
 		// Created by CGo: such a name cannot be created by regular C code.
@@ -102,6 +440,38 @@ func (info *globalInfo) parsePragmas(doc *ast.CommentGroup) {
 			if len(parts) == 2 {
 				info.linkName = parts[1]
 			}
+		case "//go:align":
+			align, err := strconv.Atoi(parts[1])
+			if err == nil {
+				info.align = align
+			}
+		case "//go:section":
+			if len(parts) == 2 {
+				info.section = parts[1]
+			}
+		case "//go:embed":
+			raw := strings.TrimPrefix(comment.Text, "//go:embed")
+			info.embeds += " " + raw
 		}
 	}
 }
+
+// Get all methods of a type.
+func getAllMethods(prog *ssa.Program, typ types.Type) []*types.Selection {
+	ms := prog.MethodSets.MethodSet(typ)
+	methods := make([]*types.Selection, ms.Len())
+	for i := 0; i < ms.Len(); i++ {
+		methods[i] = ms.At(i)
+	}
+	return methods
+}
+
+// Return true if this package imports "unsafe", false otherwise.
+func hasUnsafeImport(pkg *types.Package) bool {
+	for _, imp := range pkg.Imports() {
+		if imp == types.Unsafe {
+			return true
+		}
+	}
+	return false
+}

compiler/syscall.go

@@ -10,14 +10,14 @@ import (
 	"tinygo.org/x/go-llvm"
 )
 
-// emitSyscall emits an inline system call instruction, depending on the target
-// OS/arch.
-func (c *Compiler) emitSyscall(frame *Frame, call *ssa.CallCommon) (llvm.Value, error) {
-	num := c.getValue(frame, call.Args[0])
-	var syscallResult llvm.Value
+// createRawSyscall creates a system call with the provided system call number
+// and returns the result as a single integer (the system call result). The
+// result is not further interpreted.
+func (b *builder) createRawSyscall(call *ssa.CallCommon) (llvm.Value, error) {
+	num := b.getValue(call.Args[0])
 	switch {
-	case c.GOARCH == "amd64":
-		if c.GOOS == "darwin" {
+	case b.GOARCH == "amd64":
+		if b.GOOS == "darwin" {
 			// Darwin adds this magic number to system call numbers:
 			//
 			// > Syscall classes for 64-bit system call entry.
@@ -28,13 +28,13 @@ func (c *Compiler) emitSyscall(frame *Frame, call *ssa.CallCommon) (llvm.Value,
 			// > All system classes enter the kernel via the syscall instruction.
 			//
 			// Source: https://opensource.apple.com/source/xnu/xnu-792.13.8/osfmk/mach/i386/syscall_sw.h
-			num = c.builder.CreateOr(num, llvm.ConstInt(c.uintptrType, 0x2000000, false), "")
+			num = b.CreateOr(num, llvm.ConstInt(b.uintptrType, 0x2000000, false), "")
 		}
 		// Sources:
 		//   https://stackoverflow.com/a/2538212
 		//   https://en.wikibooks.org/wiki/X86_Assembly/Interfacing_with_Linux#syscall
 		args := []llvm.Value{num}
-		argTypes := []llvm.Type{c.uintptrType}
+		argTypes := []llvm.Type{b.uintptrType}
 		// Constraints will look something like:
 		//   "={rax},0,{rdi},{rsi},{rdx},{r10},{r8},{r9},~{rcx},~{r11}"
 		constraints := "={rax},0"
@@ -50,21 +50,21 @@ func (c *Compiler) emitSyscall(frame *Frame, call *ssa.CallCommon) (llvm.Value,
 				"{r12}",
 				"{r13}",
 			}[i]
-			llvmValue := c.getValue(frame, arg)
+			llvmValue := b.getValue(arg)
 			args = append(args, llvmValue)
 			argTypes = append(argTypes, llvmValue.Type())
 		}
 		constraints += ",~{rcx},~{r11}"
-		fnType := llvm.FunctionType(c.uintptrType, argTypes, false)
+		fnType := llvm.FunctionType(b.uintptrType, argTypes, false)
 		target := llvm.InlineAsm(fnType, "syscall", constraints, true, false, llvm.InlineAsmDialectIntel)
-		syscallResult = c.builder.CreateCall(target, args, "")
-	case c.GOARCH == "386" && c.GOOS == "linux":
+		return b.CreateCall(target, args, ""), nil
+	case b.GOARCH == "386" && b.GOOS == "linux":
 		// Sources:
 		//   syscall(2) man page
 		//   https://stackoverflow.com/a/2538212
 		//   https://en.wikibooks.org/wiki/X86_Assembly/Interfacing_with_Linux#int_0x80
 		args := []llvm.Value{num}
-		argTypes := []llvm.Type{c.uintptrType}
+		argTypes := []llvm.Type{b.uintptrType}
 		// Constraints will look something like:
 		//   "={eax},0,{ebx},{ecx},{edx},{esi},{edi},{ebp}"
 		constraints := "={eax},0"
@@ -77,14 +77,14 @@ func (c *Compiler) emitSyscall(frame *Frame, call *ssa.CallCommon) (llvm.Value,
 				"{edi}",
 				"{ebp}",
 			}[i]
-			llvmValue := c.getValue(frame, arg)
+			llvmValue := b.getValue(arg)
 			args = append(args, llvmValue)
 			argTypes = append(argTypes, llvmValue.Type())
 		}
-		fnType := llvm.FunctionType(c.uintptrType, argTypes, false)
+		fnType := llvm.FunctionType(b.uintptrType, argTypes, false)
 		target := llvm.InlineAsm(fnType, "int 0x80", constraints, true, false, llvm.InlineAsmDialectIntel)
-		syscallResult = c.builder.CreateCall(target, args, "")
-	case c.GOARCH == "arm" && c.GOOS == "linux":
+		return b.CreateCall(target, args, ""), nil
+	case b.GOARCH == "arm" && b.GOOS == "linux":
 		// Implement the EABI system call convention for Linux.
 		// Source: syscall(2) man page.
 		args := []llvm.Value{}
@@ -102,21 +102,21 @@ func (c *Compiler) emitSyscall(frame *Frame, call *ssa.CallCommon) (llvm.Value,
 				"{r5}",
 				"{r6}",
 			}[i]
-			llvmValue := c.getValue(frame, arg)
+			llvmValue := b.getValue(arg)
 			args = append(args, llvmValue)
 			argTypes = append(argTypes, llvmValue.Type())
 		}
 		args = append(args, num)
-		argTypes = append(argTypes, c.uintptrType)
+		argTypes = append(argTypes, b.uintptrType)
 		constraints += ",{r7}" // syscall number
 		for i := len(call.Args) - 1; i < 4; i++ {
 			// r0-r3 get clobbered after the syscall returns
 			constraints += ",~{r" + strconv.Itoa(i) + "}"
 		}
-		fnType := llvm.FunctionType(c.uintptrType, argTypes, false)
+		fnType := llvm.FunctionType(b.uintptrType, argTypes, false)
 		target := llvm.InlineAsm(fnType, "svc #0", constraints, true, false, 0)
-		syscallResult = c.builder.CreateCall(target, args, "")
-	case c.GOARCH == "arm64" && c.GOOS == "linux":
+		return b.CreateCall(target, args, ""), nil
+	case b.GOARCH == "arm64" && b.GOOS == "linux":
 		// Source: syscall(2) man page.
 		args := []llvm.Value{}
 		argTypes := []llvm.Type{}
@@ -132,12 +132,12 @@ func (c *Compiler) emitSyscall(frame *Frame, call *ssa.CallCommon) (llvm.Value,
 				"{x4}",
 				"{x5}",
 			}[i]
-			llvmValue := c.getValue(frame, arg)
+			llvmValue := b.getValue(arg)
 			args = append(args, llvmValue)
 			argTypes = append(argTypes, llvmValue.Type())
 		}
 		args = append(args, num)
-		argTypes = append(argTypes, c.uintptrType)
+		argTypes = append(argTypes, b.uintptrType)
 		constraints += ",{x8}" // syscall number
 		for i := len(call.Args) - 1; i < 8; i++ {
 			// x0-x7 may get clobbered during the syscall following the aarch64
@@ -145,14 +145,23 @@ func (c *Compiler) emitSyscall(frame *Frame, call *ssa.CallCommon) (llvm.Value,
 			constraints += ",~{x" + strconv.Itoa(i) + "}"
 		}
 		constraints += ",~{x16},~{x17}" // scratch registers
-		fnType := llvm.FunctionType(c.uintptrType, argTypes, false)
+		fnType := llvm.FunctionType(b.uintptrType, argTypes, false)
 		target := llvm.InlineAsm(fnType, "svc #0", constraints, true, false, 0)
-		syscallResult = c.builder.CreateCall(target, args, "")
+		return b.CreateCall(target, args, ""), nil
 	default:
-		return llvm.Value{}, c.makeError(call.Pos(), "unknown GOOS/GOARCH for syscall: "+c.GOOS+"/"+c.GOARCH)
+		return llvm.Value{}, b.makeError(call.Pos(), "unknown GOOS/GOARCH for syscall: "+b.GOOS+"/"+b.GOARCH)
 	}
-	switch c.GOOS {
-	case "linux":
+}
+
+// createSyscall emits instructions for the syscall.Syscall* family of
+// functions, depending on the target OS/arch.
+func (b *builder) createSyscall(call *ssa.CallCommon) (llvm.Value, error) {
+	syscallResult, err := b.createRawSyscall(call)
+	if err != nil {
+		return syscallResult, err
+	}
+	switch b.GOOS {
+	case "linux", "freebsd":
 		// Return values: r0, r1 uintptr, err Errno
 		// Pseudocode:
 		//     var err uintptr
@@ -160,15 +169,15 @@ func (c *Compiler) emitSyscall(frame *Frame, call *ssa.CallCommon) (llvm.Value,
 		//         err = -syscallResult
 		//     }
 		//     return syscallResult, 0, err
-		zero := llvm.ConstInt(c.uintptrType, 0, false)
-		inrange1 := c.builder.CreateICmp(llvm.IntSLT, syscallResult, llvm.ConstInt(c.uintptrType, 0, false), "")
-		inrange2 := c.builder.CreateICmp(llvm.IntSGT, syscallResult, llvm.ConstInt(c.uintptrType, 0xfffffffffffff000, true), "") // -4096
-		hasError := c.builder.CreateAnd(inrange1, inrange2, "")
-		errResult := c.builder.CreateSelect(hasError, c.builder.CreateSub(zero, syscallResult, ""), zero, "syscallError")
-		retval := llvm.Undef(c.ctx.StructType([]llvm.Type{c.uintptrType, c.uintptrType, c.uintptrType}, false))
-		retval = c.builder.CreateInsertValue(retval, syscallResult, 0, "")
-		retval = c.builder.CreateInsertValue(retval, zero, 1, "")
-		retval = c.builder.CreateInsertValue(retval, errResult, 2, "")
+		zero := llvm.ConstInt(b.uintptrType, 0, false)
+		inrange1 := b.CreateICmp(llvm.IntSLT, syscallResult, llvm.ConstInt(b.uintptrType, 0, false), "")
+		inrange2 := b.CreateICmp(llvm.IntSGT, syscallResult, llvm.ConstInt(b.uintptrType, 0xfffffffffffff000, true), "") // -4096
+		hasError := b.CreateAnd(inrange1, inrange2, "")
+		errResult := b.CreateSelect(hasError, b.CreateSub(zero, syscallResult, ""), zero, "syscallError")
+		retval := llvm.Undef(b.ctx.StructType([]llvm.Type{b.uintptrType, b.uintptrType, b.uintptrType}, false))
+		retval = b.CreateInsertValue(retval, syscallResult, 0, "")
+		retval = b.CreateInsertValue(retval, zero, 1, "")
+		retval = b.CreateInsertValue(retval, errResult, 2, "")
 		return retval, nil
 	case "darwin":
 		// Return values: r0, r1 uintptr, err Errno
@@ -178,15 +187,28 @@ func (c *Compiler) emitSyscall(frame *Frame, call *ssa.CallCommon) (llvm.Value,
 		//         err = syscallResult
 		//     }
 		//     return syscallResult, 0, err
-		zero := llvm.ConstInt(c.uintptrType, 0, false)
-		hasError := c.builder.CreateICmp(llvm.IntNE, syscallResult, llvm.ConstInt(c.uintptrType, 0, false), "")
-		errResult := c.builder.CreateSelect(hasError, syscallResult, zero, "syscallError")
-		retval := llvm.Undef(c.ctx.StructType([]llvm.Type{c.uintptrType, c.uintptrType, c.uintptrType}, false))
-		retval = c.builder.CreateInsertValue(retval, syscallResult, 0, "")
-		retval = c.builder.CreateInsertValue(retval, zero, 1, "")
-		retval = c.builder.CreateInsertValue(retval, errResult, 2, "")
+		zero := llvm.ConstInt(b.uintptrType, 0, false)
+		hasError := b.CreateICmp(llvm.IntNE, syscallResult, llvm.ConstInt(b.uintptrType, 0, false), "")
+		errResult := b.CreateSelect(hasError, syscallResult, zero, "syscallError")
+		retval := llvm.Undef(b.ctx.StructType([]llvm.Type{b.uintptrType, b.uintptrType, b.uintptrType}, false))
+		retval = b.CreateInsertValue(retval, syscallResult, 0, "")
+		retval = b.CreateInsertValue(retval, zero, 1, "")
+		retval = b.CreateInsertValue(retval, errResult, 2, "")
 		return retval, nil
 	default:
-		return llvm.Value{}, c.makeError(call.Pos(), "unknown GOOS/GOARCH for syscall: "+c.GOOS+"/"+c.GOARCH)
+		return llvm.Value{}, b.makeError(call.Pos(), "unknown GOOS/GOARCH for syscall: "+b.GOOS+"/"+b.GOARCH)
 	}
 }
+
+// createRawSyscallNoError emits instructions for the Linux-specific
+// syscall.rawSyscallNoError function.
+func (b *builder) createRawSyscallNoError(call *ssa.CallCommon) (llvm.Value, error) {
+	syscallResult, err := b.createRawSyscall(call)
+	if err != nil {
+		return syscallResult, err
+	}
+	retval := llvm.ConstNull(b.ctx.StructType([]llvm.Type{b.uintptrType, b.uintptrType}, false))
+	retval = b.CreateInsertValue(retval, syscallResult, 0, "")
+	retval = b.CreateInsertValue(retval, llvm.ConstInt(b.uintptrType, 0, false), 1, "")
+	return retval, nil
+}

compiler/testdata/basic.go

@@ -0,0 +1,57 @@
+package main
+
+// Basic tests that don't need to be split into a separate file.
+
+func addInt(x, y int) int {
+	return x + y
+}
+
+func equalInt(x, y int) bool {
+	return x == y
+}
+
+func floatEQ(x, y float32) bool {
+	return x == y
+}
+
+func floatNE(x, y float32) bool {
+	return x != y
+}
+
+func floatLower(x, y float32) bool {
+	return x < y
+}
+
+func floatLowerEqual(x, y float32) bool {
+	return x <= y
+}
+
+func floatGreater(x, y float32) bool {
+	return x > y
+}
+
+func floatGreaterEqual(x, y float32) bool {
+	return x >= y
+}
+
+func complexReal(x complex64) float32 {
+	return real(x)
+}
+
+func complexImag(x complex64) float32 {
+	return imag(x)
+}
+
+func complexAdd(x, y complex64) complex64 {
+	return x + y
+}
+
+func complexSub(x, y complex64) complex64 {
+	return x - y
+}
+
+func complexMul(x, y complex64) complex64 {
+	return x * y
+}
+
+// TODO: complexDiv (requires runtime call)

compiler/testdata/basic.ll

@@ -0,0 +1,100 @@
+; ModuleID = 'basic.go'
+source_filename = "basic.go"
+target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
+target triple = "wasm32--wasi"
+
+declare noalias nonnull i8* @runtime.alloc(i32, i8*, i8*)
+
+define hidden void @main.init(i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  ret void
+}
+
+define hidden i32 @main.addInt(i32 %x, i32 %y, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %0 = add i32 %x, %y
+  ret i32 %0
+}
+
+define hidden i1 @main.equalInt(i32 %x, i32 %y, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %0 = icmp eq i32 %x, %y
+  ret i1 %0
+}
+
+define hidden i1 @main.floatEQ(float %x, float %y, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %0 = fcmp oeq float %x, %y
+  ret i1 %0
+}
+
+define hidden i1 @main.floatNE(float %x, float %y, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %0 = fcmp une float %x, %y
+  ret i1 %0
+}
+
+define hidden i1 @main.floatLower(float %x, float %y, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %0 = fcmp olt float %x, %y
+  ret i1 %0
+}
+
+define hidden i1 @main.floatLowerEqual(float %x, float %y, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %0 = fcmp ole float %x, %y
+  ret i1 %0
+}
+
+define hidden i1 @main.floatGreater(float %x, float %y, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %0 = fcmp ogt float %x, %y
+  ret i1 %0
+}
+
+define hidden i1 @main.floatGreaterEqual(float %x, float %y, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %0 = fcmp oge float %x, %y
+  ret i1 %0
+}
+
+define hidden float @main.complexReal(float %x.r, float %x.i, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  ret float %x.r
+}
+
+define hidden float @main.complexImag(float %x.r, float %x.i, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  ret float %x.i
+}
+
+define hidden { float, float } @main.complexAdd(float %x.r, float %x.i, float %y.r, float %y.i, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %0 = fadd float %x.r, %y.r
+  %1 = fadd float %x.i, %y.i
+  %2 = insertvalue { float, float } undef, float %0, 0
+  %3 = insertvalue { float, float } %2, float %1, 1
+  ret { float, float } %3
+}
+
+define hidden { float, float } @main.complexSub(float %x.r, float %x.i, float %y.r, float %y.i, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %0 = fsub float %x.r, %y.r
+  %1 = fsub float %x.i, %y.i
+  %2 = insertvalue { float, float } undef, float %0, 0
+  %3 = insertvalue { float, float } %2, float %1, 1
+  ret { float, float } %3
+}
+
+define hidden { float, float } @main.complexMul(float %x.r, float %x.i, float %y.r, float %y.i, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %0 = fmul float %x.r, %y.r
+  %1 = fmul float %x.i, %y.i
+  %2 = fsub float %0, %1
+  %3 = fmul float %x.r, %y.i
+  %4 = fmul float %x.i, %y.r
+  %5 = fadd float %3, %4
+  %6 = insertvalue { float, float } undef, float %2, 0
+  %7 = insertvalue { float, float } %6, float %5, 1
+  ret { float, float } %7
+}

compiler/testdata/float.go

@@ -0,0 +1,39 @@
+package main
+
+// Test converting floats to ints.
+
+func f32tou32(v float32) uint32 {
+	return uint32(v)
+}
+
+func maxu32f() float32 {
+	return float32(^uint32(0))
+}
+
+func maxu32tof32() uint32 {
+	f := float32(^uint32(0))
+	return uint32(f)
+}
+
+func inftoi32() (uint32, uint32, int32, int32) {
+	inf := 1.0
+	inf /= 0.0
+
+	return uint32(inf), uint32(-inf), int32(inf), int32(-inf)
+}
+
+func u32tof32tou32(v uint32) uint32 {
+	return uint32(float32(v))
+}
+
+func f32tou32tof32(v float32) float32 {
+	return float32(uint32(v))
+}
+
+func f32tou8(v float32) uint8 {
+	return uint8(v)
+}
+
+func f32toi8(v float32) int8 {
+	return int8(v)
+}