WIP alpine build

ci: simplify build-linux job
Split building the release and smoke-testing the release in two, and don't redo some tests that are already done by assert-test-linux. Some benefits: - Lower overall CI time because tests aren't done multiple times. - TinyHCI can run earlier because the build-linux job is finished as soon as the build artifact is ready. It does however have the downside of an extra job, which costs a few seconds to spin up and a few seconds to push and pull the workspace. But even with this, overall CI time is down by a few minutes per workflow run.
2026-07-26 22:58:41 +00:00 · 2021-11-16 21:57:51 +01:00 · 2021-11-16 18:43:32 +01:00 · 2021-11-16 16:23:59 +01:00 · 2021-11-16 11:08:30 +01:00 · 2021-11-16 11:08:30 +01:00
408 changed files with 13372 additions and 4579 deletions
@@ -6,29 +6,6 @@ commands:
      - run:
          name: "Pull submodules"
          command: git submodule update --init
-  apt-dependencies:
-    parameters:
-      llvm:
-        type: string
-    steps:
-      - run:
-          name: "Install apt dependencies"
-          command: |
-            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 \
-                llvm-<<parameters.llvm>>-dev \
-                clang-<<parameters.llvm>> \
-                libclang-<<parameters.llvm>>-dev \
-                lld-<<parameters.llvm>> \
-                gcc-arm-linux-gnueabihf \
-                gcc-aarch64-linux-gnu \
-                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:
@@ -52,6 +29,13 @@ commands:
          command: |
            curl https://wasmtime.dev/install.sh -sSf | bash
            sudo ln -s ~/.wasmtime/bin/wasmtime /usr/local/bin/wasmtime
+  install-cmake:
+    steps:
+      - run:
+          name: "Install CMake"
+          command: |
+            wget https://github.com/Kitware/CMake/releases/download/v3.21.4/cmake-3.21.4-linux-x86_64.tar.gz
+            sudo tar --strip-components=1 -C /usr/local -xf cmake-3.21.4-linux-x86_64.tar.gz
  install-xtensa-toolchain:
    parameters:
      variant:
@@ -79,6 +63,39 @@ commands:
            - llvm-project/clang/include
            - llvm-project/lld/include
            - llvm-project/llvm/include
+  hack-ninja-jobs:
+    steps:
+      - run:
+          name: "Hack Ninja to use less jobs"
+          command: |
+            echo -e '#!/bin/sh\n/usr/bin/ninja -j3 "$@"' > /go/bin/ninja
+            chmod +x /go/bin/ninja
+  build-binaryen-linux:
+    steps:
+      - restore_cache:
+          keys:
+            - binaryen-linux-v1
+      - run:
+          name: "Build Binaryen"
+          command: |
+            make binaryen
+      - save_cache:
+          key: binaryen-linux-v1
+          paths:
+            - build/wasm-opt
+  build-binaryen-linux-stretch:
+    steps:
+      - restore_cache:
+          keys:
+            - binaryen-linux-stretch-v1
+      - run:
+          name: "Build Binaryen"
+          command: |
+            CC=$PWD/llvm-build/bin/clang make binaryen
+      - save_cache:
+          key: binaryen-linux-stretch-v1
+          paths:
+            - build/wasm-opt
  build-wasi-libc:
    steps:
      - restore_cache:
@@ -98,11 +115,23 @@ commands:
    steps:
      - checkout
      - submodules
-      - apt-dependencies:
-          llvm: "<<parameters.llvm>>"
-      - install-node
-      - install-chrome
-      - install-wasmtime
+      - run:
+          name: "Install apt dependencies"
+          command: |
+            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 --no-install-recommends \
+                llvm-<<parameters.llvm>>-dev \
+                clang-<<parameters.llvm>> \
+                libclang-<<parameters.llvm>>-dev \
+                lld-<<parameters.llvm>> \
+                gcc-avr \
+                avr-libc \
+                cmake \
+                ninja-build
+      - hack-ninja-jobs
+      - build-binaryen-linux
      - restore_cache:
          keys:
            - go-cache-v2-{{ checksum "go.mod" }}-{{ .Environment.CIRCLE_PREVIOUS_BUILD_NUM }}
@@ -117,11 +146,8 @@ commands:
          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 XTENSA=0
-      - run: make tinygo-test
-      - run: make wasmtest
      - save_cache:
          key: go-cache-v2-{{ checksum "go.mod" }}-{{ .Environment.CIRCLE_BUILD_NUM }}
          paths:
@@ -136,18 +162,19 @@ commands:
          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 \
+            sudo apt-get install --no-install-recommends \
+                libgnutls30 libssl1.0.2 \
                qemu-system-arm \
                qemu-user \
                gcc-avr \
-                avr-libc
-            sudo apt-get install --no-install-recommends libc6-dev-i386 lib32gcc-6-dev
+                avr-libc \
+                ninja-build \
+                python3
      - install-node
+      - install-chrome
      - install-wasmtime
+      - install-cmake
+      - hack-ninja-jobs
      - install-xtensa-toolchain:
          variant: "linux-amd64"
      - restore_cache:
@@ -163,14 +190,9 @@ commands:
          command: |
            if [ ! -f llvm-build/lib/liblldELF.a ]
            then
-              # fetch LLVM source
+              # fetch LLVM source (may only have headers right now)
              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 "$@"' > /go/bin/ninja
-              chmod +x /go/bin/ninja
              # build!
              make ASSERT=1 llvm-build
              find llvm-build -name CMakeFiles -prune -exec rm -r '{}' \;
@@ -179,11 +201,17 @@ commands:
          key: llvm-build-11-linux-v4-assert
          paths:
            llvm-build
-      - run: make ASSERT=1
+      - build-binaryen-linux-stretch
+      - run:
+          name: "Build TinyGo"
+          command: |
+            make ASSERT=1
+            echo 'export PATH=$(pwd)/build:$PATH' >> $BASH_ENV
      - build-wasi-libc
      - run:
          name: "Test TinyGo"
          command: make ASSERT=1 test
+          no_output_timeout: 20m
          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
@@ -195,62 +223,67 @@ commands:
            - ~/.cache/go-build
            - /go/pkg/mod
      - run: make gen-device -j4
-      - run: make smoketest TINYGO=build/tinygo
+      - run: make smoketest
+      - run: make tinygo-test
+      - run: make wasmtest
  build-linux:
    steps:
-      - checkout
-      - submodules
      - run:
-          name: "Install apt dependencies"
+          name: "Install apk 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"
+            apk add git openssh make g++ gcompat
+      - checkout
+      - run:
+          name: "Install Go"
+          command: |
+            wget https://dl.google.com/go/go1.17.3.linux-amd64.tar.gz
+            tar -C /usr/local -xzf go1.17.3.linux-amd64.tar.gz
+            ln -s /usr/local/go/bin/go /usr/local/bin/go
+      #- submodules
      - restore_cache:
+          name: "Restore Go cache"
          keys:
-            - go-cache-v2-{{ checksum "go.mod" }}-{{ .Environment.CIRCLE_PREVIOUS_BUILD_NUM }}
-            - go-cache-v2-{{ checksum "go.mod" }}
-      - llvm-source-linux
+            - go-cache-alpine-v2-{{ checksum "go.mod" }}-{{ .Environment.CIRCLE_PREVIOUS_BUILD_NUM }}
+            - go-cache-alpine-v2-{{ checksum "go.mod" }}
      - restore_cache:
+          name: "Restore llvm-source cache"
          keys:
-            - llvm-build-11-linux-v4-noassert
+            - llvm-source-11-alpine-v2-x0
+      - run:
+          name: "Fetch LLVM source"
+          command: make llvm-source
+      - save_cache:
+          name: "Save llvm-source cache"
+          key: llvm-source-11-alpine-v2-x0
+          paths:
+            - llvm-project
+            #- llvm-project/clang/lib/Headers
+            #- llvm-project/clang/include
+            #- llvm-project/lld/include
+            #- llvm-project/llvm/include
+      - restore_cache:
+          name: "Restore llvm-build cache"
+          keys:
+            - llvm-build-11-alpine-v1-noassert-x0
      - run:
          name: "Build LLVM"
          command: |
            if [ ! -f llvm-build/lib/liblldELF.a ]
            then
-              # fetch LLVM source
-              rm -rf llvm-project
-              make llvm-source
+              # fetch LLVM source (may only have headers right now)
+              #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
+              apk add cmake samurai python3
              # build!
-              make llvm-build
-              find llvm-build -name CMakeFiles -prune -exec rm -r '{}' \;
+              SAMUFLAGS=-j3 make llvm-build
+              #find llvm-build -name CMakeFiles -prune -exec rm -r '{}' \;
            fi
      - save_cache:
-          key: llvm-build-11-linux-v4-noassert
+          name: "Save llvm-build cache"
+          key: llvm-build-11-alpine-v1-noassert-x0
          paths:
            llvm-build
-      - build-wasi-libc
-      - run:
-          name: "Test TinyGo"
-          command: make test
      - run:
          name: "Install fpm"
          command: |
@@ -262,21 +295,41 @@ commands:
            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
+      - persist_to_workspace:
+          root: /tmp
+          paths:
+            - tinygo.linux-amd64.tar.gz
      - 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 }}
+          name: "Save Go cache"
+          key: go-cache-alpine-v2-{{ checksum "go.mod" }}-{{ .Environment.CIRCLE_BUILD_NUM }}
          paths:
            - ~/.cache/go-build
            - /go/pkg/mod
+  test-linux-build:
+    # Now run the smoke tests for the generated binary.
+    steps:
+      - attach_workspace:
+          at: /tmp/workspace
+      - checkout
+      - run:
+          name: "Install apt dependencies"
+          command: |
+            sudo apt-get update
+            sudo apt-get install --no-install-recommends \
+                gcc-avr \
+                avr-libc
+      - install-xtensa-toolchain:
+          variant: "linux-amd64"
      - run:
          name: "Extract release tarball"
          command: |
            mkdir -p ~/lib
-            tar -C ~/lib -xf /tmp/tinygo.linux-amd64.tar.gz
-            ln -s ~/lib/tinygo/bin/tinygo /go/bin/tinygo
+            tar -C ~/lib -xf /tmp/workspace/tinygo.linux-amd64.tar.gz
+            ln -s ~/lib/tinygo/bin/tinygo ~/go/bin/tinygo
            tinygo version
      - run: make smoketest
  build-macos:
@@ -286,10 +339,10 @@ commands:
      - run:
          name: "Install dependencies"
          command: |
-            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
+            curl https://dl.google.com/go/go1.17.darwin-amd64.tar.gz -o go1.17.darwin-amd64.tar.gz
+            sudo tar -C /usr/local -xzf go1.17.darwin-amd64.tar.gz
            ln -s /usr/local/go/bin/go /usr/local/bin/go
-            HOMEBREW_NO_AUTO_UPDATE=1 brew install qemu
+            HOMEBREW_NO_AUTO_UPDATE=1 brew install qemu cmake ninja
      - install-xtensa-toolchain:
          variant: "macos"
      - restore_cache:
@@ -317,11 +370,9 @@ commands:
          command: |
            if [ ! -f llvm-build/lib/liblldELF.a ]
            then
-              # fetch LLVM source
+              # fetch LLVM source (may only have headers right now)
              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 '{}' \;
@@ -330,6 +381,20 @@ commands:
          key: llvm-build-11-macos-v5
          paths:
            llvm-build
+      - restore_cache:
+          keys:
+            - binaryen-macos-v1
+      - run:
+          name: "Build Binaryen"
+          command: |
+            if [ ! -f build/wasm-opt ]
+            then
+              make binaryen
+            fi
+      - save_cache:
+          key: binaryen-macos-v1
+          paths:
+            - build/wasm-opt
      - restore_cache:
          keys:
            - wasi-libc-sysroot-macos-v4
@@ -343,6 +408,7 @@ commands:
      - run:
          name: "Test TinyGo"
          command: make test
+          no_output_timeout: 20m
      - run:
          name: "Build TinyGo release"
          command: |
@@ -365,18 +431,6 @@ commands:
            - /go/pkg/mod

 jobs:
-  test-llvm10-go113:
-    docker:
-      - image: circleci/golang:1.13-buster
-    steps:
-      - test-linux:
-          llvm: "10"
-  test-llvm10-go114:
-    docker:
-      - image: circleci/golang:1.14-buster
-    steps:
-      - test-linux:
-          llvm: "10"
  test-llvm11-go115:
    docker:
      - image: circleci/golang:1.15-buster
@@ -391,14 +445,19 @@ jobs:
          llvm: "11"
  assert-test-linux:
    docker:
-      - image: circleci/golang:1.14-stretch
+      - image: circleci/golang:1.17-stretch
    steps:
      - assert-test-linux
  build-linux:
    docker:
-      - image: circleci/golang:1.14-stretch
+      - image: alpine:3.14
    steps:
      - build-linux
+  test-linux-build:
+    docker:
+      - image: cimg/go:1.17
+    steps:
+      - test-linux-build
  build-macos:
    macos:
      xcode: "11.1.0" # macOS 10.14
@@ -410,10 +469,11 @@ jobs:
 workflows:
  test-all:
    jobs:
-      - test-llvm10-go113
-      - test-llvm10-go114
-      - test-llvm11-go115
-      - test-llvm11-go116
+      #- test-llvm11-go115
+      #- test-llvm11-go116
      - build-linux
-      - build-macos
-      - assert-test-linux
+      - test-linux-build:
+          requires:
+            - build-linux
+      #- build-macos
+      #- assert-test-linux
@@ -0,0 +1,3 @@
+build/
+llvm-*/
+
@@ -1,7 +1,7 @@
 name: CI for tinygo-dev docker container
 on:
  push:
-    branches: [ dev ]
+    branches: [ dev, fix-docker-llvm-build ]

 jobs:
  push_to_registry:
@@ -0,0 +1,103 @@
+name: Windows
+
+on:
+  #pull_request:
+  push:
+    branches:
+      - dev
+      - release
+
+jobs:
+  build-windows:
+    runs-on: windows-2019
+    steps:
+      - name: Install Go
+        uses: actions/setup-go@v2
+        with:
+          go-version: '1.17'
+      - name: Install QEMU
+        shell: bash
+        run: |
+          choco install qemu --version=2020.06.12
+          echo "C:\Program Files\QEMU" >> $GITHUB_PATH
+      - name: Install Ninja
+        shell: bash
+        run: |
+          choco install ninja
+      - name: Checkout
+        uses: actions/checkout@v2
+        with:
+          submodules: true
+      - name: Cache LLVM source
+        uses: actions/cache@v2
+        id: cache-llvm-source
+        with:
+          key: llvm-source-11-windows-v1
+          path: |
+            llvm-project/clang/lib/Headers
+            llvm-project/clang/include
+            llvm-project/lld/include
+            llvm-project/llvm/include
+      - name: Download LLVM source
+        if: steps.cache-llvm-source.outputs.cache-hit != 'true'
+        run: make llvm-source
+      - name: Cache LLVM build
+        uses: actions/cache@v2
+        id: cache-llvm-build
+        with:
+          key: llvm-build-11-windows-v2
+          path: llvm-build
+      - name: Build LLVM
+        if: steps.cache-llvm-build.outputs.cache-hit != 'true'
+        shell: bash
+        run: |
+          # fetch LLVM source
+          rm -rf llvm-project
+          make llvm-source
+          # build!
+          make llvm-build
+          # Remove unnecessary object files (to reduce cache size).
+          find llvm-build -name CMakeFiles -prune -exec rm -r '{}' \;
+      - name: Cache wasi-libc sysroot
+        uses: actions/cache@v2
+        id: cache-wasi-libc
+        with:
+          key: wasi-libc-sysroot-v1
+          path: lib/wasi-libc/sysroot
+      - name: Build wasi-libc
+        if: steps.cache-wasi-libc.outputs.cache-hit != 'true'
+        run: make wasi-libc
+      - name: Cache Binaryen
+        uses: actions/cache@v2
+        id: cache-binaryen
+        with:
+          key: binaryen-v1
+          path: build/binaryen
+      - name: Build Binaryen
+        if: steps.cache-binaryen.outputs.cache-hit != 'true'
+        run: make binaryen
+      - name: Test TinyGo
+        shell: bash
+        run: make test
+      - name: Build TinyGo release tarball
+        run: make build/release -j4
+      - name: Make release artifact
+        shell: bash
+        working-directory: build/release
+        run: 7z -tzip a release.zip tinygo
+      - name: Publish release artifact
+        # Note: this release artifact is double-zipped, see:
+        # https://github.com/actions/upload-artifact/issues/39
+        # We can essentially pick one of these:
+        # - have a dobule-zipped artifact when downloaded from the UI
+        # - have a very slow artifact upload
+        # We're doing the former here, to keep artifact uploads fast.
+        uses: actions/upload-artifact@v2
+        with:
+          name: release-double-zipped
+          path: build/release/release.zip
+      - name: Smoke tests
+        shell: bash
+        run: make smoketest TINYGO=build/tinygo AVR=0 XTENSA=0
+      - name: Test stdlib packages
+        run: make tinygo-test
@@ -1,4 +1,3 @@
-build
 docs/_build
 src/device/avr/*.go
 src/device/avr/*.ld
@@ -23,3 +23,12 @@
 [submodule "lib/stm32-svd"]
 	path = lib/stm32-svd
 	url = https://github.com/tinygo-org/stm32-svd
+[submodule "lib/musl"]
+	path = lib/musl
+	url = git://git.musl-libc.org/musl
+[submodule "lib/binaryen"]
+	path = lib/binaryen
+	url = https://github.com/WebAssembly/binaryen.git
+[submodule "lib/mingw-w64"]
+	path = lib/mingw-w64
+	url = https://github.com/mingw-w64/mingw-w64.git
@@ -23,7 +23,7 @@ different guide:
 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 (1.13+)
+  * Go (1.15+)
  * Standard build tools (gcc/clang)
  * git
  * CMake
@@ -1,3 +1,73 @@
+0.20.0
+---
+
+* **command line**
+  - add support for Go 1.17
+  - improve Go version detection
+  - add support for the Black Magic Probe (BMP)
+  - add a flag for creating cpu profiles
+* **compiler**
+  - `builder:` list libraries at the end of the linker command
+  - `builder:` strip debug information at link time instead of at compile time
+  - `builder:` add missing error check for `ioutil.TempFile()`
+  - `builder:` simplify running of jobs
+  - `compiler:` move LLVM math builtin support into the compiler
+  - `compiler:` move math aliases from the runtime to the compiler
+  - `compiler:` add aliases for many hashing packages
+  - `compiler:` add `*ssa.MakeSlice` bounds tests
+  - `compiler:` fix max possible slice
+  - `compiler:` add support for new language features of Go 1.17
+  - `compiler:` fix equally named structs in different scopes
+  - `compiler:` avoid zero-sized alloca in channel operations
+  - `interp:` don't ignore array indices for untyped objects
+  - `interp:` keep reverted package initializers in order
+  - `interp:` fix bug in compiler-time/run-time package initializers
+  - `loader:` fix panic in CGo files with syntax errors
+  - `transform:` improve GC stack slot pass to work around a bug
+* **standard library**
+  - `crypto/rand`: switch to `arc4random_buf`
+  - `math:` fix `math.Max` and `math.Min`
+  - `math/big`: fix undefined symbols error
+  - `net:` add MAC address implementation
+  - `os:` implement `os.Executable`
+  - `os:` add `SEEK_SET`, `SEEK_CUR`, and `SEEK_END`
+  - `reflect:` add StructField.IsExported method
+  - `runtime:` reset heapptr to heapStart after preinit()
+  - `runtime:` add `subsections_via_symbols` to assembly files on darwin
+  - `testing:` add subset implementation of Benchmark
+  - `testing:` test testing package using `tinygo test`
+  - `testing:` add support for the `-test.v` flag
+* **targets**
+  - `386:` bump minimum requirement to the Pentium 4
+  - `arm:` switch to Thumb instruction set on ARM
+  - `atsamd:` fix copy-paste error for atsamd21/51 calibTrim block
+  - `baremetal`,`wasm`: support command line params and environment variables
+  - `cortexm:` fix stack overflow because of unaligned stacks
+  - `esp32c3:` add support for the ESP32-C3 from Espressif
+  - `nrf52840:` fix ram size
+  - `nxpmk66f18:` fix a suspicious bitwise operation
+  - `rp2040:` add SPI support
+  - `rp2040:` add I2C support
+  - `rp2040:` add PWM implementation
+  - `rp2040:` add openocd configuration
+  - `stm32:` add support for PortMask* functions for WS2812 support
+  - `unix:` fix time base for time.Now()
+  - `unix:` check for mmap error and act accordingly
+  - `wasm:` override dlmalloc heap implementation from wasi-libc
+  - `wasm:` align heap to 16 bytes
+  - `wasm:` add support for the crypto/rand package
+* **boards**
+  - add `DefaultUART` to adafruit boards
+  - `arduino-mkrwifi1010:` add board definition for Arduino MKR WiFi 1010
+  - `arduino-mkrwifi1010:` fix pin definition of `NINA_RESETN`
+  - `feather-nrf52:` fix pin definition of uart
+  - `feather-rp2040:` add pin name definition
+  - `gameboy-advance:` fix ROM header
+  - `mdbt50qrx-uf2:` add Raytac MDBT50Q-RX Dongle with TinyUF2
+  - `nano-rp2040:` define `NINA_SPI` and fix wifinina pins
+  - `teensy40:` enable hardware UART reconfiguration, fix receive watermark interrupt
+
+
 0.19.0
 ---

@@ -1,8 +1,8 @@
-# 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
+# TinyGo base stage installs the most recent Go 1.17.x, LLVM 11 and the TinyGo compiler itself.
+FROM golang:1.17 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-11 main" >> /etc/apt/sources.list && \
+    echo "deb http://apt.llvm.org/bullseye/ llvm-toolchain-bullseye-11 main" >> /etc/apt/sources.list && \
    apt-get update && \
    apt-get install -y llvm-11-dev libclang-11-dev lld-11 git

@@ -15,7 +15,7 @@ RUN cd /tinygo/ && \
    git submodule sync && \
    git submodule update --init --recursive --force

-COPY ./lib/picolibc-include/* /tinygo/lib/picolibc-include/
+COPY ./lib/picolibc-stdio.c /tinygo/lib/picolibc-stdio.c

 RUN cd /tinygo/ && \
    go install /tinygo/
@@ -29,8 +29,9 @@ COPY --from=tinygo-base /tinygo/targets /tinygo/targets

 RUN cd /tinygo/ && \
    apt-get update && \
-    apt-get install -y make clang-11 libllvm11 lld-11 && \
-    make wasi-libc
+    apt-get install -y make clang-11 libllvm11 lld-11 cmake ninja-build && \
+    mkdir build && \
+    make wasi-libc binaryen

 # tinygo-avr stage installs the needed dependencies to compile TinyGo programs for AVR microcontrollers.
 FROM tinygo-base AS tinygo-avr
@@ -10,9 +10,9 @@ LLD_SRC ?= $(LLVM_PROJECTDIR)/lld

 # 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))
+CLANG ?= $(word 1,$(abspath $(call detect,llvm-build/bin/clang))$(call detect,clang-11)$(call detect,clang))
+LLVM_AR ?= $(word 1,$(abspath $(call detect,llvm-build/bin/llvm-ar))$(call detect,llvm-ar-11)$(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))

 # Go binary and GOROOT to select
 GO ?= go
@@ -38,7 +38,7 @@ 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
+LLVM_COMPONENTS = all-targets analysis asmparser asmprinter bitreader bitwriter codegen core coroutines coverage debuginfodwarf debuginfopdb executionengine frontendopenmp instrumentation interpreter ipo irreader libdriver linker lto mc mcjit objcarcopts option profiledata scalaropts support target windowsmanifest

 ifeq ($(OS),Windows_NT)
    EXE = .exe
@@ -54,6 +54,8 @@ ifeq ($(OS),Windows_NT)
    CGO_LDFLAGS += -static -static-libgcc -static-libstdc++
    CGO_LDFLAGS_EXTRA += -lversion

+    BINARYEN_OPTION += -DCMAKE_EXE_LINKER_FLAGS='-static-libgcc -static-libstdc++'
+
    LIBCLANG_NAME = libclang

 else ifeq ($(shell uname -s),Darwin)
@@ -100,7 +102,7 @@ endif
 clean:
 	@rm -rf build

-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_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/testing src/internal/reflectlite transform
 fmt:
 	@gofmt -l -w $(FMT_PATHS)
 fmt-check:
@@ -124,6 +126,7 @@ build/gen-device-svd: ./tools/gen-device-svd/*.go

 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/
+	./build/gen-device-svd -source=https://github.com/posborne/cmsis-svd/tree/master/data/Espressif -interrupts=software lib/cmsis-svd/data/Espressif/ src/device/esp/
 	GO111MODULE=off $(GO) fmt ./src/device/esp

 gen-device-nrf: build/gen-device-svd
@@ -162,19 +165,25 @@ 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_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)
+	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 $(NINJA_BUILD_TARGETS)
+$(LLVM_BUILDDIR): #$(LLVM_BUILDDIR)/build.ninja
+	cd $(LLVM_BUILDDIR) && ninja llvm-nm #$(NINJA_BUILD_TARGETS)

+# Build Binaryen
+.PHONY: binaryen
+binaryen: build/wasm-opt
+build/wasm-opt:
+	cd lib/binaryen && cmake -G Ninja . -DBUILD_STATIC_LIB=ON $(BINARYEN_OPTION) && ninja
+	cp lib/binaryen/bin/wasm-opt build/wasm-opt

 # 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=$(LLVM_NM)
+	cd lib/wasi-libc && make -j4 WASM_CFLAGS="-O2 -g -DNDEBUG" WASM_CC=$(CLANG) WASM_AR=$(LLVM_AR) WASM_NM=$(LLVM_NM)


 # Build the Go compiler.
@@ -183,23 +192,40 @@ tinygo:
 	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: wasi-libc
-	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GO) test -v -buildmode exe -tags byollvm ./builder ./cgo ./compileopts ./compiler ./interp ./transform .
+	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GO) test -v -timeout=20m -buildmode exe -tags byollvm ./builder ./cgo ./compileopts ./compiler ./interp ./transform .

 TEST_PACKAGES = \
+	compress/bzip2 \
 	container/heap \
 	container/list \
 	container/ring \
 	crypto/des \
+	crypto/dsa \
+	crypto/md5 \
+	crypto/rc4 \
+	crypto/sha1 \
+	crypto/sha256 \
+	crypto/sha512 \
 	encoding \
 	encoding/ascii85 \
 	encoding/base32 \
 	encoding/hex \
+	hash \
 	hash/adler32 \
 	hash/fnv \
 	hash/crc64 \
+	html \
+	index/suffixarray \
+	internal/itoa \
 	math \
 	math/cmplx \
+	net/mail \
+	reflect \
+	testing \
+	testing/iotest \
 	text/scanner \
+	unicode \
+	unicode/utf16 \
 	unicode/utf8 \

 # Test known-working standard library packages.
@@ -243,17 +269,17 @@ smoketest:
 	$(TINYGO) build -size short -o test.hex -target=pca10040            examples/test
 	@$(MD5SUM) test.hex
 	# test simulated boards on play.tinygo.org
-	$(TINYGO) build             -o test.wasm -tags=arduino              examples/blinky1
+	$(TINYGO) build -size short -o test.wasm -tags=arduino              examples/blinky1
 	@$(MD5SUM) test.wasm
-	$(TINYGO) build             -o test.wasm -tags=hifive1b             examples/blinky1
+	$(TINYGO) build -size short -o test.wasm -tags=hifive1b             examples/blinky1
 	@$(MD5SUM) test.wasm
-	$(TINYGO) build             -o test.wasm -tags=reelboard            examples/blinky1
+	$(TINYGO) build -size short -o test.wasm -tags=reelboard            examples/blinky1
 	@$(MD5SUM) test.wasm
-	$(TINYGO) build             -o test.wasm -tags=pca10040             examples/blinky2
+	$(TINYGO) build -size short -o test.wasm -tags=pca10040             examples/blinky2
 	@$(MD5SUM) test.wasm
-	$(TINYGO) build             -o test.wasm -tags=pca10056             examples/blinky2
+	$(TINYGO) build -size short -o test.wasm -tags=pca10056             examples/blinky2
 	@$(MD5SUM) test.wasm
-	$(TINYGO) build             -o test.wasm -tags=circuitplay_express  examples/blinky1
+	$(TINYGO) build -size short -o test.wasm -tags=circuitplay_express  examples/blinky1
 	@$(MD5SUM) test.wasm
 	# test all targets/boards
 	$(TINYGO) build -size short -o test.hex -target=pca10040-s132v6     examples/blinky1
@@ -358,6 +384,8 @@ smoketest:
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=arduino-nano33      examples/blinky1
 	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=arduino-mkrwifi1010 examples/blinky1
+	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=pico                examples/blinky1
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=nano-33-ble         examples/blinky1
@@ -424,15 +452,19 @@ ifneq ($(XTENSA), 0)
 	@$(MD5SUM) test.bin
 	$(TINYGO) build -size short -o test.bin -target=nodemcu             examples/blinky1
 	@$(MD5SUM) test.bin
+	$(TINYGO) build -size short -o test.bin -target m5stack-core2       examples/serial
+	@$(MD5SUM) test.bin
 endif
+	$(TINYGO) build -size short -o test.bin -target=esp32c3           	examples/serial
+	@$(MD5SUM) test.bin
 	$(TINYGO) build -size short -o test.hex -target=hifive1b            examples/blinky1
 	@$(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
+	$(TINYGO) build -size short -o wasm.wasm -target=wasm               examples/wasm/export
+	$(TINYGO) build -size short -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
@@ -444,46 +476,87 @@ endif
 	@$(MD5SUM) test.nro
 	$(TINYGO) build -size short -o test.hex -target=pca10040 -opt=0     ./testdata/stdlib.go
 	@$(MD5SUM) test.hex
+	GOOS=linux GOARCH=arm $(TINYGO) build -size short -o test.elf       ./testdata/cgo
+	GOOS=windows GOARCH=amd64 $(TINYGO) build         -o test.exe       ./testdata/cgo
+ifneq ($(OS),Windows_NT)
+	# TODO: this does not yet work on Windows. Somehow, unused functions are
+	# not garbage collected.
+	$(TINYGO) build -o test.elf -gc=leaking -scheduler=none examples/serial
+endif


 wasmtest:
 	$(GO) test ./tests/wasm

-build/release: tinygo gen-device wasi-libc
+build/release: tinygo gen-device wasi-libc binaryen
 	@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/mingw-w64/mingw-w64-crt/lib-common
+	@mkdir -p build/release/tinygo/lib/mingw-w64/mingw-w64-headers/defaults
+	@mkdir -p build/release/tinygo/lib/musl/arch
+	@mkdir -p build/release/tinygo/lib/musl/crt
+	@mkdir -p build/release/tinygo/lib/musl/src
 	@mkdir -p build/release/tinygo/lib/nrfx
 	@mkdir -p build/release/tinygo/lib/picolibc/newlib/libc
+	@mkdir -p build/release/tinygo/lib/picolibc/newlib/libm
 	@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
+	@mkdir -p build/release/tinygo/pkg/armv6m-unknown-unknown-eabi
+	@mkdir -p build/release/tinygo/pkg/armv7m-unknown-unknown-eabi
+	@mkdir -p build/release/tinygo/pkg/armv7em-unknown-unknown-eabi
 	@echo copying source files
 	@cp -p  build/tinygo$(EXE)           build/release/tinygo/bin
+	@cp -p  build/wasm-opt$(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
 	@cp -rp lib/compiler-rt/lib/builtins build/release/tinygo/lib/compiler-rt/lib
 	@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/musl/arch/aarch64        build/release/tinygo/lib/musl/arch
+	@cp -rp lib/musl/arch/arm            build/release/tinygo/lib/musl/arch
+	@cp -rp lib/musl/arch/generic        build/release/tinygo/lib/musl/arch
+	@cp -rp lib/musl/arch/i386           build/release/tinygo/lib/musl/arch
+	@cp -rp lib/musl/arch/x86_64         build/release/tinygo/lib/musl/arch
+	@cp -rp lib/musl/crt/crt1.c          build/release/tinygo/lib/musl/crt
+	@cp -rp lib/musl/COPYRIGHT           build/release/tinygo/lib/musl
+	@cp -rp lib/musl/include             build/release/tinygo/lib/musl
+	@cp -rp lib/musl/src/env             build/release/tinygo/lib/musl/src
+	@cp -rp lib/musl/src/errno           build/release/tinygo/lib/musl/src
+	@cp -rp lib/musl/src/exit            build/release/tinygo/lib/musl/src
+	@cp -rp lib/musl/src/include         build/release/tinygo/lib/musl/src
+	@cp -rp lib/musl/src/internal        build/release/tinygo/lib/musl/src
+	@cp -rp lib/musl/src/malloc          build/release/tinygo/lib/musl/src
+	@cp -rp lib/musl/src/mman            build/release/tinygo/lib/musl/src
+	@cp -rp lib/musl/src/signal          build/release/tinygo/lib/musl/src
+	@cp -rp lib/musl/src/stdio           build/release/tinygo/lib/musl/src
+	@cp -rp lib/musl/src/string          build/release/tinygo/lib/musl/src
+	@cp -rp lib/musl/src/thread          build/release/tinygo/lib/musl/src
+	@cp -rp lib/musl/src/time            build/release/tinygo/lib/musl/src
+	@cp -rp lib/musl/src/unistd          build/release/tinygo/lib/musl/src
+	@cp -rp lib/mingw-w64/mingw-w64-crt/def-include                 build/release/tinygo/lib/mingw-w64/mingw-w64-crt
+	@cp -rp lib/mingw-w64/mingw-w64-crt/lib-common/api-ms-win-crt-* build/release/tinygo/lib/mingw-w64/mingw-w64-crt/lib-common
+	@cp -rp lib/mingw-w64/mingw-w64-crt/lib-common/kernel32.def.in  build/release/tinygo/lib/mingw-w64/mingw-w64-crt/lib-common
+	@cp -rp lib/mingw-w64/mingw-w64-headers/crt/                    build/release/tinygo/lib/mingw-w64/mingw-w64-headers
+	@cp -rp lib/mingw-w64/mingw-w64-headers/defaults/include        build/release/tinygo/lib/mingw-w64/mingw-w64-headers/defaults
 	@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/picolibc/newlib/libm/common      build/release/tinygo/lib/picolibc/newlib/libm
+	@cp -rp lib/picolibc-stdio.c         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-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
+	./build/tinygo build-library -target=armv6m-unknown-unknown-eabi  -o build/release/tinygo/pkg/armv6m-unknown-unknown-eabi/compiler-rt compiler-rt
+	./build/tinygo build-library -target=armv7m-unknown-unknown-eabi  -o build/release/tinygo/pkg/armv7m-unknown-unknown-eabi/compiler-rt compiler-rt
+	./build/tinygo build-library -target=armv7em-unknown-unknown-eabi -o build/release/tinygo/pkg/armv7em-unknown-unknown-eabi/compiler-rt compiler-rt
+	./build/tinygo build-library -target=armv6m-unknown-unknown-eabi  -o build/release/tinygo/pkg/armv6m-unknown-unknown-eabi/picolibc picolibc
+	./build/tinygo build-library -target=armv7m-unknown-unknown-eabi  -o build/release/tinygo/pkg/armv7m-unknown-unknown-eabi/picolibc picolibc
+	./build/tinygo build-library -target=armv7em-unknown-unknown-eabi -o build/release/tinygo/pkg/armv7em-unknown-unknown-eabi/picolibc picolibc

 release: build/release
 	tar -czf build/release.tar.gz -C build/release tinygo
@@ -43,7 +43,7 @@ See the [getting started instructions](https://tinygo.org/getting-started/) for

 You can compile TinyGo programs for microcontrollers, WebAssembly and Linux.

-The following 68 microcontroller boards are currently supported:
+The following 71 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)
@@ -69,6 +69,7 @@ The following 68 microcontroller boards are currently supported:
 * [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 MKR WiFi 1010](https://store.arduino.cc/usa/mkr-wifi-1010)
 * [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)
@@ -83,6 +84,7 @@ The following 68 microcontroller boards are currently supported:
 * [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)
+* [M5Stack Core2](https://shop.m5stack.com/products/m5stack-core2-esp32-iot-development-kit)
 * [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)
@@ -101,6 +103,7 @@ The following 68 microcontroller boards are currently supported:
 * [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/)
+* [Raytac MDBT50Q-RX Dongle (with TinyUF2 bootloader)](https://www.adafruit.com/product/5199)
 * [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)
@@ -1,88 +0,0 @@
-# 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
@@ -0,0 +1,2 @@
+*
+!.gitignore
@@ -3,8 +3,10 @@ package builder
 import (
 	"bytes"
 	"debug/elf"
+	"debug/pe"
 	"encoding/binary"
 	"errors"
+	"fmt"
 	"io"
 	"os"
 	"path/filepath"
@@ -17,17 +19,12 @@ import (
 // given as a parameter. It is equivalent to the following command:
 //
 //     ar -rcs <archivePath> <objs...>
-func makeArchive(archivePath string, objs []string) error {
+func makeArchive(arfile *os.File, 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()
+	err := arwriter.WriteGlobalHeader()
 	if err != nil {
-		return &os.PathError{"write ar header", archivePath, err}
+		return &os.PathError{Op: "write ar header", Path: arfile.Name(), Err: err}
 	}

 	// Open all object files and read the symbols for the symbol table.
@@ -35,43 +32,55 @@ func makeArchive(archivePath string, objs []string) error {
 		name      string // symbol name
 		fileIndex int    // index into objfiles
 	}{}
-	objfiles := make([]struct {
-		file          *os.File
-		archiveOffset int32
-	}, len(objs))
+	archiveOffsets := make([]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 dbg, err := elf.NewFile(objfile); err == nil {
+			symbols, err := dbg.Symbols()
+			if err != nil {
+				return err
 			}
-			if elf.ST_TYPE(symbol.Info) != elf.STT_FUNC {
-				// Not a function.
-				// TODO: perhaps globals variables should also be included?
-				continue
+			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 && elf.ST_TYPE(symbol.Info) != elf.STT_OBJECT {
+					// Not a function.
+					continue
+				}
+				// Include in archive.
+				symbolTable = append(symbolTable, struct {
+					name      string
+					fileIndex int
+				}{symbol.Name, i})
 			}
-			// Include in archive.
-			symbolTable = append(symbolTable, struct {
-				name      string
-				fileIndex int
-			}{symbol.Name, i})
+		} else if dbg, err := pe.NewFile(objfile); err == nil {
+			for _, symbol := range dbg.Symbols {
+				if symbol.StorageClass != 2 {
+					continue
+				}
+				if symbol.SectionNumber == 0 {
+					continue
+				}
+				symbolTable = append(symbolTable, struct {
+					name      string
+					fileIndex int
+				}{symbol.Name, i})
+			}
+		} else {
+			return fmt.Errorf("failed to open file %s as ELF or PE/COFF: %w", objpath, err)
 		}
+
+		// Close file, to avoid issues with too many open files (especially on
+		// MacOS X).
+		objfile.Close()
 	}

 	// Create the symbol table buffer.
@@ -125,7 +134,14 @@ func makeArchive(archivePath string, objs []string) error {
 	}

 	// Add all object files to the archive.
-	for i, objfile := range objfiles {
+	var copyBuf bytes.Buffer
+	for i, objpath := range objs {
+		objfile, err := os.Open(objpath)
+		if err != nil {
+			return err
+		}
+		defer objfile.Close()
+
 		// 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)
@@ -133,17 +149,17 @@ func makeArchive(archivePath string, objs []string) error {
 			return err
 		}
 		if int64(int32(offset)) != offset {
-			return errors.New("large archives (4GB+) not supported: " + archivePath)
+			return errors.New("large archives (4GB+) not supported: " + arfile.Name())
 		}
-		objfiles[i].archiveOffset = int32(offset)
+		archiveOffsets[i] = int32(offset)

 		// Write the file header.
-		st, err := objfile.file.Stat()
+		st, err := objfile.Stat()
 		if err != nil {
 			return err
 		}
 		err = arwriter.WriteHeader(&ar.Header{
-			Name:    filepath.Base(objfile.file.Name()),
+			Name:    filepath.Base(objfile.Name()),
 			ModTime: time.Unix(0, 0),
 			Uid:     0,
 			Gid:     0,
@@ -155,22 +171,30 @@ func makeArchive(archivePath string, objs []string) error {
 		}

 		// Copy the file contents into the archive.
-		n, err := io.Copy(arwriter, objfile.file)
+		// First load all contents into a buffer, then write it all in one go to
+		// the archive file. This is a bit complicated, but is necessary because
+		// io.Copy can't deal with files that are of an odd size.
+		copyBuf.Reset()
+		n, err := io.Copy(&copyBuf, objfile)
 		if err != nil {
-			return err
+			return fmt.Errorf("could not copy object file into ar file: %w", err)
 		}
 		if n != st.Size() {
-			return errors.New("file modified during ar creation: " + archivePath)
+			return errors.New("file modified during ar creation: " + arfile.Name())
+		}
+		_, err = arwriter.Write(copyBuf.Bytes())
+		if err != nil {
+			return fmt.Errorf("could not copy object file into ar file: %w", err)
 		}

 		// File is not needed anymore.
-		objfile.file.Close()
+		objfile.Close()
 	}

 	// Create symbol indices.
 	indicesBuf := &bytes.Buffer{}
 	for _, sym := range symbolTable {
-		err = binary.Write(indicesBuf, binary.BigEndian, objfiles[sym.fileIndex].archiveOffset)
+		err = binary.Write(indicesBuf, binary.BigEndian, archiveOffsets[sym.fileIndex])
 		if err != nil {
 			return err
 		}
@@ -16,11 +16,14 @@ import (
 	"io/ioutil"
 	"math/bits"
 	"os"
+	"os/exec"
 	"path/filepath"
 	"runtime"
 	"sort"
+	"strconv"
 	"strings"

+	"github.com/tinygo-org/tinygo/cgo"
 	"github.com/tinygo-org/tinygo/compileopts"
 	"github.com/tinygo-org/tinygo/compiler"
 	"github.com/tinygo-org/tinygo/goenv"
@@ -60,6 +63,7 @@ type BuildResult struct {
 // implementation of an imported package changes.
 type packageAction struct {
 	ImportPath       string
+	CGoVersion       int // cgo.Version
 	CompilerVersion  int // compiler.Version
 	InterpVersion    int // interp.Version
 	LLVMVersion      string
@@ -86,6 +90,44 @@ func Build(pkgName, outpath string, config *compileopts.Config, action func(Buil
 	}
 	defer os.RemoveAll(dir)

+	// Check for a libc dependency.
+	// As a side effect, this also creates the headers for the given libc, if
+	// the libc needs them.
+	root := goenv.Get("TINYGOROOT")
+	var libcDependencies []*compileJob
+	switch config.Target.Libc {
+	case "musl":
+		job, err := Musl.load(config, dir)
+		if err != nil {
+			return err
+		}
+		libcDependencies = append(libcDependencies, dummyCompileJob(filepath.Join(filepath.Dir(job.result), "crt1.o")))
+		libcDependencies = append(libcDependencies, job)
+	case "picolibc":
+		libcJob, err := Picolibc.load(config, dir)
+		if err != nil {
+			return err
+		}
+		libcDependencies = append(libcDependencies, libcJob)
+	case "wasi-libc":
+		path := filepath.Join(root, "lib/wasi-libc/sysroot/lib/wasm32-wasi/libc.a")
+		if _, err := os.Stat(path); os.IsNotExist(err) {
+			return errors.New("could not find wasi-libc, perhaps you need to run `make wasi-libc`?")
+		}
+		libcDependencies = append(libcDependencies, dummyCompileJob(path))
+	case "mingw-w64":
+		_, err := MinGW.load(config, dir)
+		if err != nil {
+			return err
+		}
+		libcDependencies = append(libcDependencies, makeMinGWExtraLibs(dir)...)
+	case "":
+		// no library specified, so nothing to do
+	default:
+		return fmt.Errorf("unknown libc: %s", config.Target.Libc)
+	}
+
+	optLevel, sizeLevel, _ := config.OptLevels()
 	compilerConfig := &compiler.Config{
 		Triple:          config.Triple(),
 		CPU:             config.CPU(),
@@ -94,14 +136,14 @@ func Build(pkgName, outpath string, config *compileopts.Config, action func(Buil
 		GOARCH:          config.GOARCH(),
 		CodeModel:       config.CodeModel(),
 		RelocationModel: config.RelocationModel(),
+		SizeLevel:       sizeLevel,

 		Scheduler:          config.Scheduler(),
 		FuncImplementation: config.FuncImplementation(),
 		AutomaticStackSize: config.AutomaticStackSize(),
 		DefaultStackSize:   config.Target.DefaultStackSize,
 		NeedsStackObjects:  config.NeedsStackObjects(),
-		Debug:              config.Debug(),
-		LLVMFeatures:       config.LLVMFeatures(),
+		Debug:              true,
 	}

 	// Load the target machine, which is the LLVM object that contains all
@@ -124,11 +166,6 @@ func Build(pkgName, outpath string, config *compileopts.Config, action func(Buil
 		return err
 	}

-	// The slice of jobs that orchestrates most of the build.
-	// This is somewhat like an in-memory Makefile with each job being a
-	// Makefile target.
-	var jobs []*compileJob
-
 	// Create the *ssa.Program. This does not yet build the entire SSA of the
 	// program so it's pretty fast and doesn't need to be parallelized.
 	program := lprogram.LoadSSA()
@@ -138,7 +175,6 @@ func Build(pkgName, outpath string, config *compileopts.Config, action func(Buil
 	var packageJobs []*compileJob
 	packageBitcodePaths := make(map[string]string)
 	packageActionIDs := make(map[string]string)
-	optLevel, sizeLevel, _ := config.OptLevels()
 	for _, pkg := range lprogram.Sorted() {
 		pkg := pkg // necessary to avoid a race condition

@@ -152,6 +188,7 @@ func Build(pkgName, outpath string, config *compileopts.Config, action func(Buil
 		// the parameters for the build.
 		actionID := packageAction{
 			ImportPath:       pkg.ImportPath,
+			CGoVersion:       cgo.Version,
 			CompilerVersion:  compiler.Version,
 			InterpVersion:    interp.Version,
 			LLVMVersion:      llvm.Version,
@@ -212,6 +249,50 @@ func Build(pkgName, outpath string, config *compileopts.Config, action func(Buil
 					return errors.New("verification error after compiling package " + pkg.ImportPath)
 				}

+				// Load bitcode of CGo headers and join the modules together.
+				// This may seem vulnerable to cache problems, but this is not
+				// the case: the Go code that was just compiled already tracks
+				// all C files that are read and hashes them.
+				// These headers could be compiled in parallel but the benefit
+				// is so small that it's probably not worth parallelizing.
+				// Packages are compiled independently anyway.
+				for _, cgoHeader := range pkg.CGoHeaders {
+					// Store the header text in a temporary file.
+					f, err := ioutil.TempFile(dir, "cgosnippet-*.c")
+					if err != nil {
+						return err
+					}
+					_, err = f.Write([]byte(cgoHeader))
+					if err != nil {
+						return err
+					}
+					f.Close()
+
+					// Compile the code (if there is any) to bitcode.
+					flags := append([]string{"-c", "-emit-llvm", "-o", f.Name() + ".bc", f.Name()}, pkg.CFlags...)
+					if config.Options.PrintCommands != nil {
+						config.Options.PrintCommands("clang", flags...)
+					}
+					err = runCCompiler(flags...)
+					if err != nil {
+						return &commandError{"failed to build CGo header", "", err}
+					}
+
+					// Load and link the bitcode.
+					// This makes it possible to optimize the functions defined
+					// in the header together with the Go code. In particular,
+					// this allows inlining. It also ensures there is only one
+					// file per package to cache.
+					headerMod, err := mod.Context().ParseBitcodeFile(f.Name() + ".bc")
+					if err != nil {
+						return fmt.Errorf("failed to load bitcode file: %w", err)
+					}
+					err = llvm.LinkModules(mod, headerMod)
+					if err != nil {
+						return fmt.Errorf("failed to link module: %w", err)
+					}
+				}
+
 				// Erase all globals that are part of the undefinedGlobals list.
 				// This list comes from the -ldflags="-X pkg.foo=val" option.
 				// Instead of setting the value directly in the AST (which would
@@ -246,16 +327,6 @@ func Build(pkgName, outpath string, config *compileopts.Config, action func(Buil
 					return errors.New("verification error after interpreting " + pkgInit.Name())
 				}

-				if sizeLevel >= 2 {
-					// Set the "optsize" attribute to make slightly smaller
-					// binaries at the cost of some performance.
-					kind := llvm.AttributeKindID("optsize")
-					attr := mod.Context().CreateEnumAttribute(kind, 0)
-					for fn := mod.FirstFunction(); !fn.IsNil(); fn = llvm.NextFunction(fn) {
-						fn.AddFunctionAttr(attr)
-					}
-				}
-
 				// Run function passes for each function in the module.
 				// These passes are intended to be run on each function right
 				// after they're created to reduce IR size (and maybe also for
@@ -311,7 +382,6 @@ func Build(pkgName, outpath string, config *compileopts.Config, action func(Buil
 				return os.Rename(f.Name(), bitcodePath)
 			},
 		}
-		jobs = append(jobs, job)
 		packageJobs = append(packageJobs, job)
 	}

@@ -342,6 +412,7 @@ func Build(pkgName, outpath string, config *compileopts.Config, action func(Buil
 			llvmInitFn := mod.NamedFunction("runtime.initAll")
 			llvmInitFn.SetLinkage(llvm.InternalLinkage)
 			llvmInitFn.SetUnnamedAddr(true)
+			transform.AddStandardAttributes(llvmInitFn, config)
 			llvmInitFn.Param(0).SetName("context")
 			llvmInitFn.Param(1).SetName("parentHandle")
 			block := mod.Context().AddBasicBlock(llvmInitFn, "entry")
@@ -402,14 +473,13 @@ func Build(pkgName, outpath string, config *compileopts.Config, action func(Buil
 			return nil
 		},
 	}
-	jobs = append(jobs, programJob)

 	// Check whether we only need to create an object file.
 	// If so, we don't need to link anything and will be finished quickly.
 	outext := filepath.Ext(outpath)
 	if outext == ".o" || outext == ".bc" || outext == ".ll" {
 		// Run jobs to produce the LLVM module.
-		err := runJobs(jobs)
+		err := runJobs(programJob, config.Options.Parallelism)
 		if err != nil {
 			return err
 		}
@@ -450,50 +520,26 @@ func Build(pkgName, outpath string, config *compileopts.Config, action func(Buil
 			return ioutil.WriteFile(objfile, llvmBuf.Bytes(), 0666)
 		},
 	}
-	jobs = append(jobs, outputObjectFileJob)

 	// Prepare link command.
 	linkerDependencies := []*compileJob{outputObjectFileJob}
 	executable := filepath.Join(dir, "main")
+	if config.GOOS() == "windows" {
+		executable += ".exe"
+	}
 	tmppath := executable // final file
 	ldflags := append(config.LDFlags(), "-o", executable)

 	// Add compiler-rt dependency if needed. Usually this is a simple load from
 	// a cache.
 	if config.Target.RTLib == "compiler-rt" {
-		job, err := CompilerRT.load(config.Triple(), config.CPU(), dir)
+		job, err := CompilerRT.load(config, dir)
 		if err != nil {
 			return err
 		}
-		jobs = append(jobs, job.dependencies...)
-		jobs = append(jobs, job)
 		linkerDependencies = append(linkerDependencies, job)
 	}

-	// Add libc dependency if needed.
-	root := goenv.Get("TINYGOROOT")
-	switch config.Target.Libc {
-	case "picolibc":
-		job, err := Picolibc.load(config.Triple(), config.CPU(), dir)
-		if err != nil {
-			return err
-		}
-		// The library needs to be compiled (cache miss).
-		jobs = append(jobs, job.dependencies...)
-		jobs = append(jobs, job)
-		linkerDependencies = append(linkerDependencies, job)
-	case "wasi-libc":
-		path := filepath.Join(root, "lib/wasi-libc/sysroot/lib/wasm32-wasi/libc.a")
-		if _, err := os.Stat(path); os.IsNotExist(err) {
-			return errors.New("could not find wasi-libc, perhaps you need to run `make wasi-libc`?")
-		}
-		ldflags = append(ldflags, path)
-	case "":
-		// no library specified, so nothing to do
-	default:
-		return fmt.Errorf("unknown libc: %s", config.Target.Libc)
-	}
-
 	// Add jobs to compile extra files. These files are in C or assembly and
 	// contain things like the interrupt vector table and low level operations
 	// such as stack switching.
@@ -507,7 +553,6 @@ func Build(pkgName, outpath string, config *compileopts.Config, action func(Buil
 				return err
 			},
 		}
-		jobs = append(jobs, job)
 		linkerDependencies = append(linkerDependencies, job)
 	}

@@ -526,7 +571,6 @@ func Build(pkgName, outpath string, config *compileopts.Config, action func(Buil
 					return err
 				},
 			}
-			jobs = append(jobs, job)
 			linkerDependencies = append(linkerDependencies, job)
 		}
 	}
@@ -537,9 +581,48 @@ func Build(pkgName, outpath string, config *compileopts.Config, action func(Buil
 		ldflags = append(ldflags, lprogram.LDFlags...)
 	}

+	// Add libc dependencies, if they exist.
+	linkerDependencies = append(linkerDependencies, libcDependencies...)
+
+	// Strip debug information with -no-debug.
+	if !config.Debug() {
+		for _, tag := range config.BuildTags() {
+			if tag == "baremetal" {
+				// Don't use -no-debug on baremetal targets. It makes no sense:
+				// the debug information isn't flashed to the device anyway.
+				return fmt.Errorf("stripping debug information is unnecessary for baremetal targets")
+			}
+		}
+		if config.Target.Linker == "wasm-ld" {
+			// Don't just strip debug information, also compress relocations
+			// while we're at it. Relocations can only be compressed when debug
+			// information is stripped.
+			ldflags = append(ldflags, "--strip-debug", "--compress-relocations")
+		} else if config.Target.Linker == "ld.lld" {
+			// ld.lld is also used on Linux.
+			ldflags = append(ldflags, "--strip-debug")
+		} else {
+			switch config.GOOS() {
+			case "linux":
+				// Either real linux or an embedded system (like AVR) that
+				// pretends to be Linux. It's a ELF linker wrapped by GCC in any
+				// case (not ld.lld - that case is handled above).
+				ldflags = append(ldflags, "-Wl,--strip-debug")
+			case "darwin":
+				// MacOS (darwin) doesn't have a linker flag to strip debug
+				// information. Apple expects you to use the strip command
+				// instead.
+				return errors.New("cannot remove debug information: MacOS doesn't suppor this linker flag")
+			default:
+				// Other OSes may have different flags.
+				return errors.New("cannot remove debug information: unknown OS: " + config.GOOS())
+			}
+		}
+	}
+
 	// Create a linker job, which links all object files together and does some
 	// extra stuff that can only be done after linking.
-	jobs = append(jobs, &compileJob{
+	linkJob := &compileJob{
 		description:  "link",
 		dependencies: linkerDependencies,
 		run: func(job *compileJob) error {
@@ -586,22 +669,62 @@ func Build(pkgName, outpath string, config *compileopts.Config, action func(Buil
 				}
 			}

+			// Run wasm-opt if necessary.
+			if config.Scheduler() == "asyncify" {
+				var optLevel, shrinkLevel int
+				switch config.Options.Opt {
+				case "none", "0":
+				case "1":
+					optLevel = 1
+				case "2":
+					optLevel = 2
+				case "s":
+					optLevel = 2
+					shrinkLevel = 1
+				case "z":
+					optLevel = 2
+					shrinkLevel = 2
+				default:
+					return fmt.Errorf("unknown opt level: %q", config.Options.Opt)
+				}
+				cmd := exec.Command(goenv.Get("WASMOPT"), "--asyncify", "-g",
+					"--optimize-level", strconv.Itoa(optLevel),
+					"--shrink-level", strconv.Itoa(shrinkLevel),
+					executable, "--output", executable)
+				cmd.Stdout = os.Stdout
+				cmd.Stderr = os.Stderr
+
+				err := cmd.Run()
+				if err != nil {
+					return fmt.Errorf("wasm-opt failed: %w", err)
+				}
+			}
+
+			// Print code size if requested.
 			if config.Options.PrintSizes == "short" || config.Options.PrintSizes == "full" {
-				sizes, err := loadProgramSize(executable)
+				packagePathMap := make(map[string]string, len(lprogram.Packages))
+				for _, pkg := range lprogram.Sorted() {
+					packagePathMap[pkg.OriginalDir()] = pkg.Pkg.Path()
+				}
+				sizes, err := loadProgramSize(executable, packagePathMap)
 				if err != nil {
 					return err
 				}
 				if config.Options.PrintSizes == "short" {
 					fmt.Printf("   code    data     bss |   flash     ram\n")
-					fmt.Printf("%7d %7d %7d | %7d %7d\n", sizes.Code, sizes.Data, sizes.BSS, sizes.Code+sizes.Data, sizes.Data+sizes.BSS)
+					fmt.Printf("%7d %7d %7d | %7d %7d\n", sizes.Code+sizes.ROData, sizes.Data, sizes.BSS, sizes.Flash(), sizes.RAM())
 				} else {
+					if !config.Debug() {
+						fmt.Println("warning: data incomplete, remove the -no-debug flag for more detail")
+					}
 					fmt.Printf("   code  rodata    data     bss |   flash     ram | package\n")
+					fmt.Printf("------------------------------- | --------------- | -------\n")
 					for _, name := range sizes.sortedPackageNames() {
 						pkgSize := sizes.Packages[name]
 						fmt.Printf("%7d %7d %7d %7d | %7d %7d | %s\n", pkgSize.Code, pkgSize.ROData, pkgSize.Data, pkgSize.BSS, pkgSize.Flash(), pkgSize.RAM(), name)
 					}
-					fmt.Printf("%7d %7d %7d %7d | %7d %7d | (sum)\n", sizes.Sum.Code, sizes.Sum.ROData, sizes.Sum.Data, sizes.Sum.BSS, sizes.Sum.Flash(), sizes.Sum.RAM())
-					fmt.Printf("%7d       - %7d %7d | %7d %7d | (all)\n", sizes.Code, sizes.Data, sizes.BSS, sizes.Code+sizes.Data, sizes.Data+sizes.BSS)
+					fmt.Printf("------------------------------- | --------------- | -------\n")
+					fmt.Printf("%7d %7d %7d %7d | %7d %7d | total\n", sizes.Code, sizes.ROData, sizes.Data, sizes.BSS, sizes.Code+sizes.ROData+sizes.Data, sizes.Data+sizes.BSS)
 				}
 			}

@@ -612,12 +735,12 @@ func Build(pkgName, outpath string, config *compileopts.Config, action func(Buil

 			return nil
 		},
-	})
+	}

 	// Run all jobs to compile and link the program.
 	// Do this now (instead of after elf-to-hex and similar conversions) as it
 	// is simpler and cannot be parallelized.
-	err = runJobs(jobs)
+	err = runJobs(linkJob, config.Options.Parallelism)
 	if err != nil {
 		return err
 	}
@@ -642,7 +765,7 @@ func Build(pkgName, outpath string, config *compileopts.Config, action func(Buil
 		if err != nil {
 			return err
 		}
-	case "esp32", "esp8266":
+	case "esp32", "esp32c3", "esp8266":
 		// Special format for the ESP family of chips (parsed by the ROM
 		// bootloader).
 		tmppath = filepath.Join(dir, "main"+outext)
@@ -709,7 +832,7 @@ func optimizeProgram(mod llvm.Module, config *compileopts.Config) error {
 	// stack-allocated values.
 	// Use -wasm-abi=generic to disable this behaviour.
 	if config.WasmAbi() == "js" {
-		err := transform.ExternalInt64AsPtr(mod)
+		err := transform.ExternalInt64AsPtr(mod, config)
 		if err != nil {
 			return err
 		}
@@ -953,15 +1076,19 @@ func modifyStackSizes(executable string, stackSizeLoads []string, stackSizes map
 		if fn.stackSizeType == stacksize.Bounded {
 			stackSize := uint32(fn.stackSize)

-			// Adding 4 for the stack canary. Even though the size may be
-			// automatically determined, stack overflow checking is still
-			// important as the stack size cannot be determined for all
-			// goroutines.
-			stackSize += 4
-
 			// Add stack size used by interrupts.
 			switch fileHeader.Machine {
 			case elf.EM_ARM:
+				if stackSize%8 != 0 {
+					// If the stack isn't a multiple of 8, it means the leaf
+					// function with the biggest stack depth doesn't have an aligned
+					// stack. If the STKALIGN flag is set (which it is by default)
+					// the interrupt controller will forcibly align the stack before
+					// storing in-use registers. This will thus overwrite one word
+					// past the end of the stack (off-by-one).
+					stackSize += 4
+				}
+
 				// On Cortex-M (assumed here), this stack size is 8 words or 32
 				// bytes. This is only to store the registers that the interrupt
 				// may modify, the interrupt will switch to the interrupt stack
@@ -969,6 +1096,14 @@ func modifyStackSizes(executable string, stackSizeLoads []string, stackSizes map
 				// Some background:
 				// https://interrupt.memfault.com/blog/cortex-m-rtos-context-switching
 				stackSize += 32
+
+				// Adding 4 for the stack canary, and another 4 to keep the
+				// stack aligned. Even though the size may be automatically
+				// determined, stack overflow checking is still important as the
+				// stack size cannot be determined for all goroutines.
+				stackSize += 8
+			default:
+				return fmt.Errorf("unknown architecture: %s", fileHeader.Machine.String())
 			}

 			// Finally write the stack size to the binary.
@@ -1,105 +0,0 @@
-package builder
-
-import (
-	"io"
-	"os"
-	"path/filepath"
-	"time"
-
-	"github.com/tinygo-org/tinygo/goenv"
-)
-
-// Return the newest timestamp of all the file paths passed in. Used to check
-// for stale caches.
-func cacheTimestamp(paths []string) (time.Time, error) {
-	var timestamp time.Time
-	for _, path := range paths {
-		st, err := os.Stat(path)
-		if err != nil {
-			return time.Time{}, err
-		}
-		if timestamp.IsZero() {
-			timestamp = st.ModTime()
-		} else if timestamp.Before(st.ModTime()) {
-			timestamp = st.ModTime()
-		}
-	}
-	return timestamp, nil
-}
-
-// 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.
-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
-	} else if err != nil {
-		return "", err // cannot stat cache file
-	}
-
-	sourceTimestamp, err := cacheTimestamp(sourceFiles)
-	if err != nil {
-		return "", err // cannot stat source files
-	}
-
-	if cacheStat.ModTime().After(sourceTimestamp) {
-		return cachepath, nil
-	} else {
-		os.Remove(cachepath)
-		// stale cache
-		return "", nil
-	}
-}
-
-// Store the file located at tmppath in the cache with the given name. The
-// tmppath may or may not be gone afterwards.
-func cacheStore(tmppath, name string, sourceFiles []string) (string, error) {
-	// get the last modified time
-	if len(sourceFiles) == 0 {
-		panic("cache: no source files")
-	}
-
-	// TODO: check the config key
-
-	dir := goenv.Get("GOCACHE")
-	err := os.MkdirAll(dir, 0777)
-	if err != nil {
-		return "", err
-	}
-	cachepath := filepath.Join(dir, name)
-	err = copyFile(tmppath, cachepath)
-	if err != nil {
-		return "", err
-	}
-	return cachepath, nil
-}
-
-// 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
-	}
-	defer inf.Close()
-	outpath := dst + ".tmp"
-	outf, err := os.Create(outpath)
-	if err != nil {
-		return err
-	}
-
-	_, err = io.Copy(outf, inf)
-	if err != nil {
-		os.Remove(outpath)
-		return err
-	}
-
-	err = outf.Close()
-	if err != nil {
-		return err
-	}
-
-	return os.Rename(dst+".tmp", dst)
-}
@@ -0,0 +1,163 @@
+package builder
+
+import (
+	"fmt"
+	"io/ioutil"
+	"os"
+	"path/filepath"
+	"testing"
+
+	"github.com/tinygo-org/tinygo/compileopts"
+	"github.com/tinygo-org/tinygo/goenv"
+	"tinygo.org/x/go-llvm"
+)
+
+// Test whether the Clang generated "target-cpu" and "target-features"
+// attributes match the CPU and Features property in TinyGo target files.
+func TestClangAttributes(t *testing.T) {
+	var targetNames = []string{
+		// Please keep this list sorted!
+		"atmega328p",
+		"atmega1280",
+		"atmega1284p",
+		"atmega2560",
+		"attiny85",
+		"cortex-m0",
+		"cortex-m0plus",
+		"cortex-m3",
+		//"cortex-m33", // TODO: broken in LLVM 11, fixed in https://reviews.llvm.org/D90305
+		"cortex-m4",
+		"cortex-m7",
+		"esp32c3",
+		"fe310",
+		"gameboy-advance",
+		"k210",
+		"nintendoswitch",
+		"riscv-qemu",
+		"wasi",
+		"wasm",
+	}
+	if hasBuiltinTools {
+		// hasBuiltinTools is set when TinyGo is statically linked with LLVM,
+		// which also implies it was built with Xtensa support.
+		targetNames = append(targetNames, "esp32", "esp8266")
+	}
+	for _, targetName := range targetNames {
+		targetName := targetName
+		t.Run(targetName, func(t *testing.T) {
+			testClangAttributes(t, &compileopts.Options{Target: targetName})
+		})
+	}
+
+	for _, options := range []*compileopts.Options{
+		{GOOS: "linux", GOARCH: "386"},
+		{GOOS: "linux", GOARCH: "amd64"},
+		{GOOS: "linux", GOARCH: "arm", GOARM: "5"},
+		{GOOS: "linux", GOARCH: "arm", GOARM: "6"},
+		{GOOS: "linux", GOARCH: "arm", GOARM: "7"},
+		{GOOS: "linux", GOARCH: "arm64"},
+		{GOOS: "darwin", GOARCH: "amd64"},
+		{GOOS: "darwin", GOARCH: "arm64"},
+		{GOOS: "windows", GOARCH: "amd64"},
+	} {
+		t.Run("GOOS="+options.GOOS+",GOARCH="+options.GOARCH, func(t *testing.T) {
+			testClangAttributes(t, options)
+		})
+	}
+}
+
+func testClangAttributes(t *testing.T, options *compileopts.Options) {
+	testDir := t.TempDir()
+	clangHeaderPath := getClangHeaderPath(goenv.Get("TINYGOROOT"))
+
+	ctx := llvm.NewContext()
+	defer ctx.Dispose()
+
+	target, err := compileopts.LoadTarget(options)
+	if err != nil {
+		t.Fatalf("could not load target: %s", err)
+	}
+	config := compileopts.Config{
+		Options:      options,
+		Target:       target,
+		ClangHeaders: clangHeaderPath,
+	}
+
+	// Create a very simple C input file.
+	srcpath := filepath.Join(testDir, "test.c")
+	err = ioutil.WriteFile(srcpath, []byte("int add(int a, int b) { return a + b; }"), 0o666)
+	if err != nil {
+		t.Fatalf("could not write target file %s: %s", srcpath, err)
+	}
+
+	// Compile this file using Clang.
+	outpath := filepath.Join(testDir, "test.bc")
+	flags := append([]string{"-c", "-emit-llvm", "-o", outpath, srcpath}, config.CFlags()...)
+	if config.GOOS() == "darwin" {
+		// Silence some warnings that happen when testing GOOS=darwin on
+		// something other than MacOS.
+		flags = append(flags, "-Wno-missing-sysroot", "-Wno-incompatible-sysroot")
+	}
+	err = runCCompiler(flags...)
+	if err != nil {
+		t.Fatalf("failed to compile %s: %s", srcpath, err)
+	}
+
+	// Read the resulting LLVM bitcode.
+	mod, err := ctx.ParseBitcodeFile(outpath)
+	if err != nil {
+		t.Fatalf("could not parse bitcode file %s: %s", outpath, err)
+	}
+	defer mod.Dispose()
+
+	// Check whether the LLVM target matches.
+	if mod.Target() != config.Triple() {
+		t.Errorf("target has LLVM triple %#v but Clang makes it LLVM triple %#v", config.Triple(), mod.Target())
+	}
+
+	// Check the "target-cpu" and "target-features" string attribute of the add
+	// function.
+	add := mod.NamedFunction("add")
+	var cpu, features string
+	cpuAttr := add.GetStringAttributeAtIndex(-1, "target-cpu")
+	featuresAttr := add.GetStringAttributeAtIndex(-1, "target-features")
+	if !cpuAttr.IsNil() {
+		cpu = cpuAttr.GetStringValue()
+	}
+	if !featuresAttr.IsNil() {
+		features = featuresAttr.GetStringValue()
+	}
+	if cpu != config.CPU() {
+		t.Errorf("target has CPU %#v but Clang makes it CPU %#v", config.CPU(), cpu)
+	}
+	if features != config.Features() {
+		if llvm.Version != "11.0.0" {
+			// This needs to be removed once we switch to LLVM 12.
+			// LLVM 11.0.0 uses a different "target-features" string than LLVM
+			// 11.1.0 for Thumb targets. The Xtensa fork is still based on LLVM
+			// 11.0.0, so we need to skip this check on that version.
+			t.Errorf("target has LLVM features %#v but Clang makes it %#v", config.Features(), features)
+		}
+	}
+}
+
+// This TestMain is necessary because TinyGo may also be invoked to run certain
+// LLVM tools in a separate process. Not capturing these invocations would lead
+// to recursive tests.
+func TestMain(m *testing.M) {
+	if len(os.Args) >= 2 {
+		switch os.Args[1] {
+		case "clang", "ld.lld", "wasm-ld":
+			// Invoke a specific tool.
+			err := RunTool(os.Args[1], os.Args[2:]...)
+			if err != nil {
+				fmt.Fprintln(os.Stderr, err)
+				os.Exit(1)
+			}
+			os.Exit(0)
+		}
+	}
+
+	// Run normal tests.
+	os.Exit(m.Run())
+}
@@ -157,10 +157,12 @@ var aeabiBuiltins = []string{
 //
 // 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"} },
+	name: "compiler-rt",
+	cflags: func(target, headerPath string) []string {
+		return []string{"-Werror", "-Wall", "-std=c11", "-nostdlibinc"}
+	},
 	sourceDir: "lib/compiler-rt/lib/builtins",
-	sources: func(target string) []string {
+	librarySources: func(target string) []string {
 		builtins := append([]string{}, genericBuiltins...) // copy genericBuiltins
 		if strings.HasPrefix(target, "arm") || strings.HasPrefix(target, "thumb") {
 			builtins = append(builtins, aeabiBuiltins...)
@@ -155,6 +155,10 @@ func compileAndCacheCFile(abspath, tmpdir string, cflags []string, printCommands

 	// Write dependencies file.
 	f, err := ioutil.TempFile(filepath.Dir(depfileCachePath), depfileName)
+	if err != nil {
+		return "", err
+	}
+
 	buf, err = json.MarshalIndent(dependencySlice, "", "\t")
 	if err != nil {
 		panic(err) // shouldn't happen
@@ -21,6 +21,7 @@ func init() {
 	commands["clang"] = []string{"clang-" + llvmMajor}
 	commands["ld.lld"] = []string{"ld.lld-" + llvmMajor, "ld.lld"}
 	commands["wasm-ld"] = []string{"wasm-ld-" + llvmMajor, "wasm-ld"}
+	commands["lldb"] = []string{"lldb-" + llvmMajor, "lldb"}
 	// Add the path to a Homebrew-installed LLVM for ease of use (no need to
 	// manually set $PATH).
 	if runtime.GOOS == "darwin" {
@@ -28,6 +29,7 @@ func init() {
 		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")
+		commands["lldb"] = append(commands["lldb"], prefix+"lldb")
 	}
 	// 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
@@ -36,6 +38,7 @@ func init() {
 		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")
+		commands["lldb"] = append(commands["lldb"], "C:\\Program Files\\LLVM\\bin\\lldb.exe")
 	}
 	// Add the path to LLVM installed from ports.
 	if runtime.GOOS == "freebsd" {
@@ -43,23 +46,34 @@ func init() {
 		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")
+		commands["lldb"] = append(commands["lldb"], prefix+"lldb")
 	}
 }

-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()
+// LookupCommand looks up the executable name for a given LLVM tool such as
+// clang or wasm-ld. It returns the (relative) command that can be used to
+// invoke the tool or an error if it could not be found.
+func LookupCommand(name string) (string, error) {
+	for _, cmdName := range commands[name] {
+		_, err := exec.LookPath(cmdName)
 		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
+			if errors.Unwrap(err) == exec.ErrNotFound {
 				continue
 			}
-			return err
+			return cmdName, err
 		}
-		return nil
+		return cmdName, nil
 	}
-	return errors.New("none of these commands were found in your $PATH: " + strings.Join(cmdNames, " "))
+	return "", errors.New("%#v: none of these commands were found in your $PATH: " + strings.Join(commands[name], " "))
+}
+
+func execCommand(name string, args ...string) error {
+	name, err := LookupCommand(name)
+	if err != nil {
+		return err
+	}
+	cmd := exec.Command(name, args...)
+	cmd.Stdout = os.Stdout
+	cmd.Stderr = os.Stderr
+	return cmd.Run()
 }
@@ -13,7 +13,7 @@ import (
 // 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)
+	spec, err := compileopts.LoadTarget(options)
 	if err != nil {
 		return nil, err
 	}
@@ -33,8 +33,8 @@ func NewConfig(options *compileopts.Options) (*compileopts.Config, error) {
 	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)
+	if major != 1 || minor < 15 || minor > 17 {
+		return nil, fmt.Errorf("requires go version 1.15 through 1.17, got go%d.%d", major, minor)
 	}

 	clangHeaderPath := getClangHeaderPath(goenv.Get("TINYGOROOT"))
@@ -78,11 +78,21 @@ func makeESPFirmareImage(infile, outfile, format string) error {
 	// An added benefit is that we don't need to check for errors all the time.
 	outf := &bytes.Buffer{}

+	// Chip IDs. Source:
+	// https://github.com/espressif/esp-idf/blob/v4.3/components/bootloader_support/include/esp_app_format.h#L22
+	chip_id := map[string]uint16{
+		"esp32":   0x0000,
+		"esp32c3": 0x0005,
+	}[format]
+
 	// Image header.
 	switch format {
-	case "esp32":
+	case "esp32", "esp32c3":
 		// Header format:
-		// https://github.com/espressif/esp-idf/blob/8fbb63c2/components/bootloader_support/include/esp_image_format.h#L58
+		// https://github.com/espressif/esp-idf/blob/v4.3/components/bootloader_support/include/esp_app_format.h#L71
+		// Note: not adding a SHA256 hash as the binary is modified by
+		// esptool.py while flashing and therefore the hash won't be valid
+		// anymore.
 		binary.Write(outf, binary.LittleEndian, struct {
 			magic          uint8
 			segment_count  uint8
@@ -91,15 +101,18 @@ func makeESPFirmareImage(infile, outfile, format string) error {
 			entry_addr     uint32
 			wp_pin         uint8
 			spi_pin_drv    [3]uint8
-			reserved       [11]uint8
+			chip_id        uint16
+			min_chip_rev   uint8
+			reserved       [8]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
+			spi_mode:       2,    // ESP_IMAGE_SPI_MODE_DIO
+			spi_speed_size: 0x1f, // ESP_IMAGE_SPI_SPEED_80M, ESP_IMAGE_FLASH_SIZE_2MB
 			entry_addr:     uint32(inf.Entry),
 			wp_pin:         0xEE, // disable WP pin
+			chip_id:        chip_id,
 			hash_appended:  true, // add a SHA256 hash
 		})
 	case "esp8266":
@@ -142,7 +155,7 @@ func makeESPFirmareImage(infile, outfile, format string) error {
 	outf.Write(make([]byte, 15-outf.Len()%16))
 	outf.WriteByte(checksum)

-	if format == "esp32" {
+	if format != "esp8266" {
 		// SHA256 hash (to protect against image corruption, not for security).
 		hash := sha256.Sum256(outf.Bytes())
 		outf.Write(hash[:])
@@ -65,19 +65,45 @@ func (job *compileJob) readyToRun() bool {
 	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 {
+// runJobs runs the indicated job and all its dependencies. For every job, all
+// the dependencies are run first. It returns the error of the first job that
+// fails.
+// It runs all jobs in the order of the dependencies slice, depth-first.
+// Therefore, if some jobs are preferred to run before others, they should be
+// ordered as such in the job dependencies.
+func runJobs(job *compileJob, parallelism int) error {
+	if parallelism == 0 {
+		// Have a default, if the parallelism isn't set. This is useful for
+		// tests.
+		parallelism = runtime.NumCPU()
+	}
+	if parallelism < 1 {
+		return fmt.Errorf("-p flag must be at least 1, provided -p=%d", parallelism)
+	}
+
+	// Create a slice of jobs to run, where all dependencies are run in order.
+	jobs := []*compileJob{}
+	addedJobs := map[*compileJob]struct{}{}
+	var addJobs func(*compileJob)
+	addJobs = func(job *compileJob) {
+		if _, ok := addedJobs[job]; ok {
+			return
+		}
+		for _, dep := range job.dependencies {
+			addJobs(dep)
+		}
+		jobs = append(jobs, job)
+		addedJobs[job] = struct{}{}
+	}
+	addJobs(job)
+
 	// 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++ {
+	for i := 0; i < parallelism; i++ {
 		if jobRunnerDebug {
 			fmt.Println("## starting worker", i)
 		}
@@ -92,7 +118,7 @@ func runJobs(jobs []*compileJob) error {
 	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() {
+		if numRunningJobs < parallelism {
 			jobToRun := nextJob(jobs)
 			if jobToRun != nil {
 				// Start job.
@@ -1,10 +1,12 @@
 package builder

 import (
+	"io/ioutil"
 	"os"
 	"path/filepath"
 	"strings"

+	"github.com/tinygo-org/tinygo/compileopts"
 	"github.com/tinygo-org/tinygo/goenv"
 )

@@ -14,72 +16,87 @@ type Library struct {
 	// The library name, such as compiler-rt or picolibc.
 	name string

-	cflags func() []string
+	// makeHeaders creates a header include dir for the library
+	makeHeaders func(target, includeDir string) error
+
+	// cflags returns the C flags specific to this library
+	cflags func(target, headerPath string) []string

 	// The source directory, relative to TINYGOROOT.
 	sourceDir string

 	// The source files, relative to sourceDir.
-	sources func(target string) []string
-}
+	librarySources 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
+	// The source code for the crt1.o file, relative to sourceDir.
+	crt1Source string
 }

 // 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)
+// The resulting directory may be stored in the provided tmpdir, which is
+// expected to be removed after the Load call.
+func (l *Library) Load(config *compileopts.Config, tmpdir string) (dir string, err error) {
+	job, err := l.load(config, tmpdir)
 	if err != nil {
 		return "", err
 	}
-	jobs := append([]*compileJob{job}, job.dependencies...)
-	err = runJobs(jobs)
-	return job.result, err
+	err = runJobs(job, config.Options.Parallelism)
+	return filepath.Dir(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.
+// cached. The path is stored as job.result but is only valid after the job has
+// 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 {
+// As a side effect, this call creates the library header files if they didn't
+// exist yet.
+func (l *Library) load(config *compileopts.Config, tmpdir string) (job *compileJob, err error) {
+	outdir, precompiled := config.LibcPath(l.name)
+	archiveFilePath := filepath.Join(outdir, "lib.a")
+	if precompiled {
 		// 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"
+		return dummyCompileJob(archiveFilePath), nil
 	}

 	// 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
+	if _, err := os.Stat(archiveFilePath); err == nil {
+		return dummyCompileJob(archiveFilePath), nil
 	}
 	// Cache miss, build it now.

+	// Create the destination directory where the components of this library
+	// (lib.a file, include directory) are placed.
+	outname := filepath.Base(outdir)
+	err = os.MkdirAll(filepath.Join(goenv.Get("GOCACHE"), outname), 0o777)
+	if err != nil {
+		// Could not create directory (and not because it already exists).
+		return nil, err
+	}
+
+	// Make headers if needed.
+	headerPath := filepath.Join(outdir, "include")
+	target := config.Triple()
+	if l.makeHeaders != nil {
+		if _, err = os.Stat(headerPath); err != nil {
+			temporaryHeaderPath, err := ioutil.TempDir(outdir, "include.tmp*")
+			if err != nil {
+				return nil, err
+			}
+			defer os.RemoveAll(temporaryHeaderPath)
+			err = l.makeHeaders(target, temporaryHeaderPath)
+			if err != nil {
+				return nil, err
+			}
+			err = os.Rename(temporaryHeaderPath, headerPath)
+			if err != nil {
+				return nil, err
+			}
+		}
+	}
+
 	remapDir := filepath.Join(os.TempDir(), "tinygo-"+l.name)
 	dir := filepath.Join(tmpdir, "build-lib-"+l.name)
 	err = os.Mkdir(dir, 0777)
@@ -91,9 +108,16 @@ func (l *Library) load(target, cpu, tmpdir string) (job *compileJob, err error)
 	// Note: -fdebug-prefix-map is necessary to make the output archive
 	// reproducible. Otherwise the temporary directory is stored in the archive
 	// itself, which varies each run.
-	args := append(l.cflags(), "-c", "-Oz", "-g", "-ffunction-sections", "-fdata-sections", "-Wno-macro-redefined", "--target="+target, "-fdebug-prefix-map="+dir+"="+remapDir)
+	args := append(l.cflags(target, headerPath), "-c", "-Oz", "-g", "-ffunction-sections", "-fdata-sections", "-Wno-macro-redefined", "--target="+target, "-fdebug-prefix-map="+dir+"="+remapDir)
+	cpu := config.CPU()
 	if cpu != "" {
-		args = append(args, "-mcpu="+cpu)
+		// X86 has deprecated the -mcpu flag, so we need to use -march instead.
+		// However, ARM has not done this.
+		if strings.HasPrefix(target, "i386") || strings.HasPrefix(target, "x86_64") {
+			args = append(args, "-march="+cpu)
+		} else {
+			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")
@@ -108,27 +132,39 @@ func (l *Library) load(target, cpu, tmpdir string) (job *compileJob, err error)
 	// 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,
+		description: "ar " + l.name + "/lib.a",
+		result:      filepath.Join(goenv.Get("GOCACHE"), outname, "lib.a"),
 		run: func(*compileJob) error {
 			// Create an archive of all object files.
-			err := makeArchive(arpath, objs)
+			f, err := ioutil.TempFile(outdir, "libc.a.tmp*")
+			if err != nil {
+				return err
+			}
+			err = makeArchive(f, objs)
+			if err != nil {
+				return err
+			}
+			err = f.Close()
 			if err != nil {
 				return err
 			}
 			// Store this archive in the cache.
-			_, err = cacheStore(arpath, outfile, l.sourcePaths(target))
-			return err
+			return os.Rename(f.Name(), archiveFilePath)
 		},
 	}

 	// 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")
+	for _, path := range l.librarySources(target) {
+		// Strip leading "../" parts off the path.
+		cleanpath := path
+		for strings.HasPrefix(cleanpath, "../") {
+			cleanpath = cleanpath[3:]
+		}
+		srcpath := filepath.Join(goenv.Get("TINYGOROOT"), l.sourceDir, path)
+		objpath := filepath.Join(dir, cleanpath+".o")
+		os.MkdirAll(filepath.Dir(objpath), 0o777)
 		objs = append(objs, objpath)
 		job.dependencies = append(job.dependencies, &compileJob{
 			description: "compile " + srcpath,
@@ -145,5 +181,31 @@ func (l *Library) load(target, cpu, tmpdir string) (job *compileJob, err error)
 		})
 	}

+	// Create crt1.o job, if needed.
+	// Add this as a (fake) dependency to the ar file so it gets compiled.
+	// (It could be done in parallel with creating the ar file, but it probably
+	// won't make much of a difference in speed).
+	if l.crt1Source != "" {
+		srcpath := filepath.Join(goenv.Get("TINYGOROOT"), l.sourceDir, l.crt1Source)
+		job.dependencies = append(job.dependencies, &compileJob{
+			description: "compile " + srcpath,
+			run: func(*compileJob) error {
+				var compileArgs []string
+				compileArgs = append(compileArgs, args...)
+				tmpfile, err := ioutil.TempFile(outdir, "crt1.o.tmp*")
+				if err != nil {
+					return err
+				}
+				tmpfile.Close()
+				compileArgs = append(compileArgs, "-o", tmpfile.Name(), srcpath)
+				err = runCCompiler(compileArgs...)
+				if err != nil {
+					return &commandError{"failed to build", srcpath, err}
+				}
+				return os.Rename(tmpfile.Name(), filepath.Join(outdir, "crt1.o"))
+			},
+		})
+	}
+
 	return job, nil
 }
@@ -11,6 +11,11 @@ bool tinygo_link_elf(int argc, char **argv) {
 	return lld::elf::link(args, false, llvm::outs(), llvm::errs());
 }

+bool tinygo_link_mingw(int argc, char **argv) {
+	std::vector<const char*> args(argv, argv + argc);
+	return lld::mingw::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, llvm::outs(), llvm::errs());
@@ -0,0 +1,91 @@
+package builder
+
+import (
+	"io"
+	"os"
+	"path/filepath"
+	"strings"
+
+	"github.com/tinygo-org/tinygo/goenv"
+)
+
+var MinGW = Library{
+	name: "mingw-w64",
+	makeHeaders: func(target, includeDir string) error {
+		// copy _mingw.h
+		srcDir := filepath.Join(goenv.Get("TINYGOROOT"), "lib", "mingw-w64")
+		outf, err := os.Create(includeDir + "/_mingw.h")
+		if err != nil {
+			return err
+		}
+		defer outf.Close()
+		inf, err := os.Open(srcDir + "/mingw-w64-headers/crt/_mingw.h.in")
+		if err != nil {
+			return err
+		}
+		_, err = io.Copy(outf, inf)
+		return err
+	},
+	cflags: func(target, headerPath string) []string {
+		// No flags necessary because there are no files to compile.
+		return nil
+	},
+	librarySources: func(target string) []string {
+		// We only use the UCRT DLL file. No source files necessary.
+		return nil
+	},
+}
+
+// makeMinGWExtraLibs returns a slice of jobs to import the correct .dll
+// libraries. This is done by converting input .def files to .lib files which
+// can then be linked as usual.
+//
+// TODO: cache the result. At the moment, it costs a few hundred milliseconds to
+// compile these files.
+func makeMinGWExtraLibs(tmpdir string) []*compileJob {
+	var jobs []*compileJob
+	root := goenv.Get("TINYGOROOT")
+	// Normally all the api-ms-win-crt-*.def files are all compiled to a single
+	// .lib file. But to simplify things, we're going to leave them as separate
+	// files.
+	for _, name := range []string{
+		"kernel32.def.in",
+		"api-ms-win-crt-conio-l1-1-0.def",
+		"api-ms-win-crt-convert-l1-1-0.def",
+		"api-ms-win-crt-environment-l1-1-0.def",
+		"api-ms-win-crt-filesystem-l1-1-0.def",
+		"api-ms-win-crt-heap-l1-1-0.def",
+		"api-ms-win-crt-locale-l1-1-0.def",
+		"api-ms-win-crt-math-l1-1-0.def.in",
+		"api-ms-win-crt-multibyte-l1-1-0.def",
+		"api-ms-win-crt-private-l1-1-0.def.in",
+		"api-ms-win-crt-process-l1-1-0.def",
+		"api-ms-win-crt-runtime-l1-1-0.def.in",
+		"api-ms-win-crt-stdio-l1-1-0.def",
+		"api-ms-win-crt-string-l1-1-0.def",
+		"api-ms-win-crt-time-l1-1-0.def",
+		"api-ms-win-crt-utility-l1-1-0.def",
+	} {
+		outpath := filepath.Join(tmpdir, filepath.Base(name)+".lib")
+		inpath := filepath.Join(root, "lib/mingw-w64/mingw-w64-crt/lib-common/"+name)
+		job := &compileJob{
+			description: "create lib file " + inpath,
+			result:      outpath,
+			run: func(job *compileJob) error {
+				defpath := inpath
+				if strings.HasSuffix(inpath, ".in") {
+					// .in files need to be preprocessed by a preprocessor (-E)
+					// first.
+					defpath = outpath + ".def"
+					err := runCCompiler("-E", "-x", "c", "-Wp,-w", "-P", "-DDEF_X64", "-o", defpath, inpath, "-I"+goenv.Get("TINYGOROOT")+"/lib/mingw-w64/mingw-w64-crt/def-include/")
+					if err != nil {
+						return err
+					}
+				}
+				return link("ld.lld", "-m", "i386pep", "-o", outpath, defpath)
+			},
+		}
+		jobs = append(jobs, job)
+	}
+	return jobs
+}
@@ -0,0 +1,162 @@
+package builder
+
+import (
+	"bytes"
+	"fmt"
+	"io/ioutil"
+	"os"
+	"path/filepath"
+	"regexp"
+	"strings"
+
+	"github.com/tinygo-org/tinygo/compileopts"
+	"github.com/tinygo-org/tinygo/goenv"
+)
+
+var Musl = Library{
+	name: "musl",
+	makeHeaders: func(target, includeDir string) error {
+		bits := filepath.Join(includeDir, "bits")
+		err := os.Mkdir(bits, 0777)
+		if err != nil {
+			return err
+		}
+
+		arch := compileopts.MuslArchitecture(target)
+		muslDir := filepath.Join(goenv.Get("TINYGOROOT"), "lib", "musl")
+
+		// Create the file alltypes.h.
+		f, err := os.Create(filepath.Join(bits, "alltypes.h"))
+		if err != nil {
+			return err
+		}
+		infiles := []string{
+			filepath.Join(muslDir, "arch", arch, "bits", "alltypes.h.in"),
+			filepath.Join(muslDir, "include", "alltypes.h.in"),
+		}
+		for _, infile := range infiles {
+			data, err := ioutil.ReadFile(infile)
+			if err != nil {
+				return err
+			}
+			lines := strings.Split(string(data), "\n")
+			for _, line := range lines {
+				if strings.HasPrefix(line, "TYPEDEF ") {
+					matches := regexp.MustCompile(`TYPEDEF (.*) ([^ ]*);`).FindStringSubmatch(line)
+					value := matches[1]
+					name := matches[2]
+					line = fmt.Sprintf("#if defined(__NEED_%s) && !defined(__DEFINED_%s)\ntypedef %s %s;\n#define __DEFINED_%s\n#endif\n", name, name, value, name, name)
+				}
+				if strings.HasPrefix(line, "STRUCT ") {
+					matches := regexp.MustCompile(`STRUCT * ([^ ]*) (.*);`).FindStringSubmatch(line)
+					name := matches[1]
+					value := matches[2]
+					line = fmt.Sprintf("#if defined(__NEED_struct_%s) && !defined(__DEFINED_struct_%s)\nstruct %s %s;\n#define __DEFINED_struct_%s\n#endif\n", name, name, name, value, name)
+				}
+				f.WriteString(line + "\n")
+			}
+		}
+		f.Close()
+
+		// Create the file syscall.h.
+		f, err = os.Create(filepath.Join(bits, "syscall.h"))
+		if err != nil {
+			return err
+		}
+		data, err := ioutil.ReadFile(filepath.Join(muslDir, "arch", arch, "bits", "syscall.h.in"))
+		if err != nil {
+			return err
+		}
+		_, err = f.Write(bytes.ReplaceAll(data, []byte("__NR_"), []byte("SYS_")))
+		if err != nil {
+			return err
+		}
+		f.Close()
+
+		return nil
+	},
+	cflags: func(target, headerPath string) []string {
+		arch := compileopts.MuslArchitecture(target)
+		muslDir := filepath.Join(goenv.Get("TINYGOROOT"), "lib/musl")
+		return []string{
+			"-std=c99",            // same as in musl
+			"-D_XOPEN_SOURCE=700", // same as in musl
+			// Musl triggers some warnings and we don't want to show any
+			// warnings while compiling (only errors or silence), so disable
+			// specific warnings that are triggered in musl.
+			"-Werror",
+			"-Wno-logical-op-parentheses",
+			"-Wno-bitwise-op-parentheses",
+			"-Wno-shift-op-parentheses",
+			"-Wno-ignored-attributes",
+			"-Wno-string-plus-int",
+			"-Qunused-arguments",
+			// Select include dirs. Don't include standard library includes
+			// (that would introduce host dependencies and other complications),
+			// but do include all the include directories expected by musl.
+			"-nostdlibinc",
+			"-I" + muslDir + "/arch/" + arch,
+			"-I" + muslDir + "/arch/generic",
+			"-I" + muslDir + "/src/include",
+			"-I" + muslDir + "/src/internal",
+			"-I" + headerPath,
+			"-I" + muslDir + "/include",
+		}
+	},
+	sourceDir: "lib/musl/src",
+	librarySources: func(target string) []string {
+		arch := compileopts.MuslArchitecture(target)
+		globs := []string{
+			"env/*.c",
+			"errno/*.c",
+			"exit/*.c",
+			"internal/defsysinfo.c",
+			"internal/libc.c",
+			"internal/syscall_ret.c",
+			"internal/vdso.c",
+			"malloc/*.c",
+			"mman/*.c",
+			"signal/*.c",
+			"stdio/*.c",
+			"string/*.c",
+			"thread/" + arch + "/*.s",
+			"thread/" + arch + "/*.c",
+			"thread/*.c",
+			"time/*.c",
+			"unistd/*.c",
+		}
+
+		var sources []string
+		seenSources := map[string]struct{}{}
+		basepath := goenv.Get("TINYGOROOT") + "/lib/musl/src/"
+		for _, pattern := range globs {
+			matches, err := filepath.Glob(basepath + pattern)
+			if err != nil {
+				// From the documentation:
+				// > Glob ignores file system errors such as I/O errors reading
+				// > directories. The only possible returned error is
+				// > ErrBadPattern, when pattern is malformed.
+				// So the only possible error is when the (statically defined)
+				// pattern is wrong. In other words, a programming bug.
+				panic("could not glob source dirs: " + err.Error())
+			}
+			for _, match := range matches {
+				relpath, err := filepath.Rel(basepath, match)
+				if err != nil {
+					// Not sure if this is even possible.
+					panic(err)
+				}
+				// Make sure architecture specific files override generic files.
+				id := strings.ReplaceAll(relpath, "/"+arch+"/", "/")
+				if _, ok := seenSources[id]; ok {
+					// Already seen this file, skipping this (generic) file.
+					continue
+				}
+				seenSources[id] = struct{}{}
+				sources = append(sources, relpath)
+			}
+		}
+		return sources
+	},
+	crt1Source: "../crt/crt1.c", // lib/musl/crt/crt1.c
+}
@@ -1,6 +1,7 @@
 package builder

 import (
+	"os"
 	"path/filepath"

 	"github.com/tinygo-org/tinygo/goenv"
@@ -10,17 +11,134 @@ import (
 // based on newlib.
 var Picolibc = Library{
 	name: "picolibc",
-	cflags: func() []string {
+	makeHeaders: func(target, includeDir string) error {
+		f, err := os.Create(filepath.Join(includeDir, "picolibc.h"))
+		if err != nil {
+			return err
+		}
+		return f.Close()
+	},
+	cflags: func(target, headerPath string) []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"}
+		return []string{
+			"-Werror",
+			"-Wall",
+			"-std=gnu11",
+			"-D_COMPILING_NEWLIB",
+			"-DHAVE_ALIAS_ATTRIBUTE",
+			"-DTINY_STDIO",
+			"-nostdlibinc",
+			"-Xclang", "-internal-isystem", "-Xclang", picolibcDir + "/include",
+			"-I" + picolibcDir + "/tinystdio",
+			"-I" + headerPath,
+		}
 	},
 	sourceDir: "lib/picolibc/newlib/libc",
-	sources: func(target string) []string {
+	librarySources: func(target string) []string {
 		return picolibcSources
 	},
 }

 var picolibcSources = []string{
+	"../../../picolibc-stdio.c",
+
+	"tinystdio/asprintf.c",
+	"tinystdio/atod_engine.c",
+	"tinystdio/atod_ryu.c",
+	"tinystdio/atof_engine.c",
+	"tinystdio/atof_ryu.c",
+	//"tinystdio/atold_engine.c", // have_long_double and not long_double_equals_double
+	"tinystdio/clearerr.c",
+	"tinystdio/compare_exchange.c",
+	"tinystdio/dtoa_data.c",
+	"tinystdio/dtoa_engine.c",
+	"tinystdio/dtoa_ryu.c",
+	"tinystdio/ecvtbuf.c",
+	"tinystdio/ecvt.c",
+	"tinystdio/ecvt_data.c",
+	"tinystdio/ecvtfbuf.c",
+	"tinystdio/ecvtf.c",
+	"tinystdio/ecvtf_data.c",
+	"tinystdio/exchange.c",
+	//"tinystdio/fclose.c", // posix-io
+	"tinystdio/fcvtbuf.c",
+	"tinystdio/fcvt.c",
+	"tinystdio/fcvtfbuf.c",
+	"tinystdio/fcvtf.c",
+	"tinystdio/fdevopen.c",
+	//"tinystdio/fdopen.c", // posix-io
+	"tinystdio/feof.c",
+	"tinystdio/ferror.c",
+	"tinystdio/fflush.c",
+	"tinystdio/fgetc.c",
+	"tinystdio/fgets.c",
+	"tinystdio/fileno.c",
+	"tinystdio/filestrget.c",
+	"tinystdio/filestrputalloc.c",
+	"tinystdio/filestrput.c",
+	//"tinystdio/fopen.c", // posix-io
+	"tinystdio/fprintf.c",
+	"tinystdio/fputc.c",
+	"tinystdio/fputs.c",
+	"tinystdio/fread.c",
+	"tinystdio/fscanf.c",
+	"tinystdio/fseek.c",
+	"tinystdio/ftell.c",
+	"tinystdio/ftoa_data.c",
+	"tinystdio/ftoa_engine.c",
+	"tinystdio/ftoa_ryu.c",
+	"tinystdio/fwrite.c",
+	"tinystdio/gcvtbuf.c",
+	"tinystdio/gcvt.c",
+	"tinystdio/gcvtfbuf.c",
+	"tinystdio/gcvtf.c",
+	"tinystdio/getchar.c",
+	"tinystdio/gets.c",
+	"tinystdio/matchcaseprefix.c",
+	"tinystdio/perror.c",
+	//"tinystdio/posixiob.c", // posix-io
+	//"tinystdio/posixio.c", // posix-io
+	"tinystdio/printf.c",
+	"tinystdio/putchar.c",
+	"tinystdio/puts.c",
+	"tinystdio/ryu_divpow2.c",
+	"tinystdio/ryu_log10.c",
+	"tinystdio/ryu_log2pow5.c",
+	"tinystdio/ryu_pow5bits.c",
+	"tinystdio/ryu_table.c",
+	"tinystdio/ryu_umul128.c",
+	"tinystdio/scanf.c",
+	"tinystdio/setbuf.c",
+	"tinystdio/setvbuf.c",
+	//"tinystdio/sflags.c", // posix-io
+	"tinystdio/snprintf.c",
+	"tinystdio/snprintfd.c",
+	"tinystdio/snprintff.c",
+	"tinystdio/sprintf.c",
+	"tinystdio/sprintfd.c",
+	"tinystdio/sprintff.c",
+	"tinystdio/sscanf.c",
+	"tinystdio/strfromd.c",
+	"tinystdio/strfromf.c",
+	"tinystdio/strtod.c",
+	"tinystdio/strtod_l.c",
+	"tinystdio/strtof.c",
+	//"tinystdio/strtold.c",   // have_long_double and not long_double_equals_double
+	//"tinystdio/strtold_l.c", // have_long_double and not long_double_equals_double
+	"tinystdio/ungetc.c",
+	"tinystdio/vasprintf.c",
+	"tinystdio/vfiprintf.c",
+	"tinystdio/vfiscanf.c",
+	"tinystdio/vfprintf.c",
+	"tinystdio/vfprintff.c",
+	"tinystdio/vfscanf.c",
+	"tinystdio/vfscanff.c",
+	"tinystdio/vprintf.c",
+	"tinystdio/vscanf.c",
+	"tinystdio/vsnprintf.c",
+	"tinystdio/vsprintf.c",
+	"tinystdio/vsscanf.c",
+
 	"string/bcmp.c",
 	"string/bcopy.c",
 	"string/bzero.c",
@@ -1,16 +1,30 @@
 package builder

 import (
+	"bytes"
+	"debug/dwarf"
 	"debug/elf"
+	"encoding/binary"
+	"fmt"
+	"io"
+	"os"
+	"path/filepath"
+	"regexp"
 	"sort"
 	"strings"
+
+	"github.com/aykevl/go-wasm"
+	"github.com/tinygo-org/tinygo/goenv"
 )

+// Set to true to print extra debug logs.
+const sizesDebug = false
+
 // programSize contains size statistics per package of a compiled program.
 type programSize struct {
-	Packages map[string]*packageSize
-	Sum      *packageSize
+	Packages map[string]packageSize
 	Code     uint64
+	ROData   uint64
 	Data     uint64
 	BSS      uint64
 }
@@ -26,6 +40,16 @@ func (ps *programSize) sortedPackageNames() []string {
 	return names
 }

+// Flash usage in regular microcontrollers.
+func (ps *programSize) Flash() uint64 {
+	return ps.Code + ps.ROData + ps.Data
+}
+
+// Static RAM usage in regular microcontrollers.
+func (ps *programSize) RAM() uint64 {
+	return ps.Data + ps.BSS
+}
+
 // packageSize contains the size of a package, calculated from the linked object
 // file.
 type packageSize struct {
@@ -45,129 +69,527 @@ func (ps *packageSize) RAM() uint64 {
 	return ps.Data + ps.BSS
 }

-type symbolList []elf.Symbol
-
-func (l symbolList) Len() int {
-	return len(l)
+// A mapping of a single chunk of code or data to a file path.
+type addressLine struct {
+	Address    uint64
+	Length     uint64 // length of this chunk
+	File       string // file path as stored in DWARF
+	IsVariable bool   // true if this is a variable (or constant), false if it is code
 }

-func (l symbolList) Less(i, j int) bool {
-	bind_i := elf.ST_BIND(l[i].Info)
-	bind_j := elf.ST_BIND(l[j].Info)
-	if l[i].Value == l[j].Value && bind_i != elf.STB_WEAK && bind_j == elf.STB_WEAK {
-		// sort weak symbols after non-weak symbols
-		return true
+// Sections defined in the input file. This struct defines them in a
+// filetype-agnostic way but roughly follow the ELF types (.text, .data, .bss,
+// etc).
+type memorySection struct {
+	Type    memoryType
+	Address uint64
+	Size    uint64
+}
+
+type memoryType int
+
+const (
+	memoryCode memoryType = iota + 1
+	memoryData
+	memoryROData
+	memoryBSS
+	memoryStack
+)
+
+// Regular expressions to match particular symbol names. These are not stored as
+// DWARF variables because they have no mapping to source code global variables.
+var (
+	// Various globals that aren't a variable but nonetheless need to be stored
+	// somewhere:
+	//   alloc:  heap allocations during init interpretation
+	//   pack:   data created when storing a constant in an interface for example
+	//   string: buffer behind strings
+	packageSymbolRegexp = regexp.MustCompile(`\$(alloc|pack|string)(\.[0-9]+)?$`)
+
+	// Reflect sidetables. Created by the reflect lowering pass.
+	// See src/reflect/sidetables.go.
+	reflectDataRegexp = regexp.MustCompile(`^reflect\.[a-zA-Z]+Sidetable$`)
+)
+
+// readProgramSizeFromDWARF reads the source location for each line of code and
+// each variable in the program, as far as this is stored in the DWARF debug
+// information.
+func readProgramSizeFromDWARF(data *dwarf.Data, codeOffset uint64) ([]addressLine, error) {
+	r := data.Reader()
+	var lines []*dwarf.LineFile
+	var addresses []addressLine
+	for {
+		e, err := r.Next()
+		if err != nil {
+			return nil, err
+		}
+		if e == nil {
+			break
+		}
+		switch e.Tag {
+		case dwarf.TagCompileUnit:
+			// Found a compile unit.
+			// We can read the .debug_line section using it, which contains a
+			// mapping for most instructions to their file/line/column - even
+			// for inlined functions!
+			lr, err := data.LineReader(e)
+			if err != nil {
+				return nil, err
+			}
+			lines = lr.Files()
+			var lineEntry = dwarf.LineEntry{
+				EndSequence: true,
+			}
+
+			// Line tables are organized as sequences of line entries until an
+			// end sequence. A single line table can contain multiple such
+			// sequences. The last line entry is an EndSequence to indicate the
+			// end.
+			for {
+				// Read the next .debug_line entry.
+				prevLineEntry := lineEntry
+				err := lr.Next(&lineEntry)
+				if err != nil {
+					if err == io.EOF {
+						break
+					}
+					return nil, err
+				}
+
+				if prevLineEntry.EndSequence && lineEntry.Address == 0 {
+					// Tombstone value. This symbol has been removed, for
+					// example by the --gc-sections linker flag. It is still
+					// here in the debug information because the linker can't
+					// just remove this reference.
+					// Read until the next EndSequence so that this sequence is
+					// skipped.
+					// For more details, see (among others):
+					// https://reviews.llvm.org/D84825
+					for {
+						err := lr.Next(&lineEntry)
+						if err != nil {
+							return nil, err
+						}
+						if lineEntry.EndSequence {
+							break
+						}
+					}
+				}
+
+				if !prevLineEntry.EndSequence {
+					// The chunk describes the code from prevLineEntry to
+					// lineEntry.
+					line := addressLine{
+						Address: prevLineEntry.Address + codeOffset,
+						Length:  lineEntry.Address - prevLineEntry.Address,
+						File:    prevLineEntry.File.Name,
+					}
+					if line.Length != 0 {
+						addresses = append(addresses, line)
+					}
+				}
+			}
+		case dwarf.TagVariable:
+			// Global variable (or constant). Most of these are not actually
+			// stored in the binary, because they have been optimized out. Only
+			// the ones with a location are still present.
+			r.SkipChildren()
+
+			file := e.AttrField(dwarf.AttrDeclFile)
+			location := e.AttrField(dwarf.AttrLocation)
+			globalType := e.AttrField(dwarf.AttrType)
+			if file == nil || location == nil || globalType == nil {
+				// Doesn't contain the requested information.
+				continue
+			}
+
+			// Try to parse the location. While this could in theory be a very
+			// complex expression, usually it's just a DW_OP_addr opcode
+			// followed by an address.
+			locationCode := location.Val.([]uint8)
+			if locationCode[0] != 3 { // DW_OP_addr
+				continue
+			}
+			var addr uint64
+			switch len(locationCode) {
+			case 1 + 2:
+				addr = uint64(binary.LittleEndian.Uint16(locationCode[1:]))
+			case 1 + 4:
+				addr = uint64(binary.LittleEndian.Uint32(locationCode[1:]))
+			case 1 + 8:
+				addr = binary.LittleEndian.Uint64(locationCode[1:])
+			default:
+				continue // unknown address
+			}
+
+			// Parse the type of the global variable, which (importantly)
+			// contains the variable size. We're not interested in the type,
+			// only in the size.
+			typ, err := data.Type(globalType.Val.(dwarf.Offset))
+			if err != nil {
+				return nil, err
+			}
+
+			addresses = append(addresses, addressLine{
+				Address:    addr,
+				Length:     uint64(typ.Size()),
+				File:       lines[file.Val.(int64)].Name,
+				IsVariable: true,
+			})
+		default:
+			r.SkipChildren()
+		}
 	}
-	return l[i].Value < l[j].Value
-}
-
-func (l symbolList) Swap(i, j int) {
-	l[i], l[j] = l[j], l[i]
+	return addresses, nil
 }

 // 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 the file doesn't contain DWARF debug information, the returned program
+// size will still have valid summaries but won't have complete size information
+// per package.
+func loadProgramSize(path string, packagePathMap map[string]string) (*programSize, error) {
+	// Open the binary file.
+	f, err := os.Open(path)
 	if err != nil {
 		return nil, err
 	}
-	defer file.Close()
+	defer f.Close()

-	var sumCode uint64
-	var sumData uint64
-	var sumBSS uint64
-	for _, section := range file.Sections {
-		if section.Flags&elf.SHF_ALLOC == 0 {
-			continue
-		}
-		if section.Type != elf.SHT_PROGBITS && section.Type != elf.SHT_NOBITS {
-			continue
-		}
-		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
-		} else if section.Flags&elf.SHF_WRITE != 0 {
-			sumData += section.Size
-		}
-	}
+	// This stores all chunks of addresses found in the binary.
+	var addresses []addressLine

-	allSymbols, err := file.Symbols()
-	if err != nil {
-		return nil, err
-	}
-	symbols := make([]elf.Symbol, 0, len(allSymbols))
-	for _, symbol := range allSymbols {
-		symType := elf.ST_TYPE(symbol.Info)
-		if symbol.Size == 0 {
-			continue
-		}
-		if symType != elf.STT_FUNC && symType != elf.STT_OBJECT && symType != elf.STT_NOTYPE {
-			continue
-		}
-		if symbol.Section >= elf.SectionIndex(len(file.Sections)) {
-			continue
-		}
-		section := file.Sections[symbol.Section]
-		if section.Flags&elf.SHF_ALLOC == 0 {
-			continue
-		}
-		symbols = append(symbols, symbol)
-	}
-	sort.Sort(symbolList(symbols))
-
-	sizes := map[string]*packageSize{}
-	var lastSymbolValue uint64
-	for _, symbol := range symbols {
-		symType := elf.ST_TYPE(symbol.Info)
-		//bind := elf.ST_BIND(symbol.Info)
-		section := file.Sections[symbol.Section]
-		pkgName := "(bootstrap)"
-		symName := strings.TrimLeft(symbol.Name, "(*")
-		dot := strings.IndexByte(symName, '.')
-		if dot > 0 {
-			pkgName = symName[:dot]
-		}
-		pkgSize := sizes[pkgName]
-		if pkgSize == nil {
-			pkgSize = &packageSize{}
-			sizes[pkgName] = pkgSize
-		}
-		if lastSymbolValue != symbol.Value || lastSymbolValue == 0 {
-			if symType == elf.STT_FUNC {
-				pkgSize.Code += symbol.Size
-			} else if section.Flags&elf.SHF_WRITE != 0 {
-				if section.Type == elf.SHT_NOBITS {
-					pkgSize.BSS += symbol.Size
-				} else {
-					pkgSize.Data += symbol.Size
-				}
-			} else {
-				pkgSize.ROData += symbol.Size
+	// Load the binary file, which could be in a number of file formats.
+	var sections []memorySection
+	if file, err := elf.NewFile(f); err == nil {
+		// Read DWARF information. The error is intentionally ignored.
+		data, _ := file.DWARF()
+		if data != nil {
+			addresses, err = readProgramSizeFromDWARF(data, 0)
+			if err != nil {
+				// However, _do_ report an error here. Something must have gone
+				// wrong while trying to parse DWARF data.
+				return nil, err
 			}
 		}
-		lastSymbolValue = symbol.Value
+
+		// Read the ELF symbols for some more chunks of location information.
+		// Some globals (such as strings) aren't stored in the DWARF debug
+		// information and therefore need to be obtained in a different way.
+		allSymbols, err := file.Symbols()
+		if err != nil {
+			return nil, err
+		}
+		for _, symbol := range allSymbols {
+			symType := elf.ST_TYPE(symbol.Info)
+			if symbol.Size == 0 {
+				continue
+			}
+			if symType != elf.STT_FUNC && symType != elf.STT_OBJECT && symType != elf.STT_NOTYPE {
+				continue
+			}
+			if symbol.Section >= elf.SHN_LORESERVE {
+				// Not a regular section, so skip it.
+				// One example is elf.SHN_ABS, which is used for symbols
+				// declared with an absolute value such as the memset function
+				// on the ESP32 which is defined in the mask ROM.
+				continue
+			}
+			section := file.Sections[symbol.Section]
+			if section.Flags&elf.SHF_ALLOC == 0 {
+				continue
+			}
+			if packageSymbolRegexp.MatchString(symbol.Name) || reflectDataRegexp.MatchString(symbol.Name) {
+				addresses = append(addresses, addressLine{
+					Address:    symbol.Value,
+					Length:     symbol.Size,
+					File:       symbol.Name,
+					IsVariable: true,
+				})
+			}
+		}
+
+		// Load allocated sections.
+		for _, section := range file.Sections {
+			if section.Flags&elf.SHF_ALLOC == 0 {
+				continue
+			}
+			if section.Type == elf.SHT_NOBITS {
+				if section.Name == ".stack" {
+					// TinyGo emits stack sections on microcontroller using the
+					// ".stack" name.
+					// This is a bit ugly, but I don't think there is a way to
+					// mark the stack section in a linker script.
+					sections = append(sections, memorySection{
+						Address: section.Addr,
+						Size:    section.Size,
+						Type:    memoryStack,
+					})
+				} else {
+					// Regular .bss section.
+					sections = append(sections, memorySection{
+						Address: section.Addr,
+						Size:    section.Size,
+						Type:    memoryBSS,
+					})
+				}
+			} else if section.Type == elf.SHT_PROGBITS && section.Flags&elf.SHF_EXECINSTR != 0 {
+				// .text
+				sections = append(sections, memorySection{
+					Address: section.Addr,
+					Size:    section.Size,
+					Type:    memoryCode,
+				})
+			} else if section.Type == elf.SHT_PROGBITS && section.Flags&elf.SHF_WRITE != 0 {
+				// .data
+				sections = append(sections, memorySection{
+					Address: section.Addr,
+					Size:    section.Size,
+					Type:    memoryData,
+				})
+			} else if section.Type == elf.SHT_PROGBITS {
+				// .rodata
+				sections = append(sections, memorySection{
+					Address: section.Addr,
+					Size:    section.Size,
+					Type:    memoryROData,
+				})
+			}
+		}
+	} else if file, err := wasm.Parse(f); err == nil {
+		// File is in WebAssembly format.
+
+		// Put code at a very high address, so that it won't conflict with the
+		// data in the memory section.
+		const codeOffset = 0x8000_0000_0000_0000
+
+		// Read DWARF information. The error is intentionally ignored.
+		data, err := file.DWARF()
+		if data != nil {
+			addresses, err = readProgramSizeFromDWARF(data, codeOffset)
+			if err != nil {
+				// However, _do_ report an error here. Something must have gone
+				// wrong while trying to parse DWARF data.
+				return nil, err
+			}
+		}
+
+		var linearMemorySize uint64
+		for _, section := range file.Sections {
+			switch section := section.(type) {
+			case *wasm.SectionCode:
+				sections = append(sections, memorySection{
+					Address: codeOffset,
+					Size:    uint64(section.Size()),
+					Type:    memoryCode,
+				})
+			case *wasm.SectionMemory:
+				// This value is used when processing *wasm.SectionData (which
+				// always comes after *wasm.SectionMemory).
+				linearMemorySize = uint64(section.Entries[0].Limits.Initial) * 64 * 1024
+			case *wasm.SectionData:
+				// Data sections contain initial values for linear memory.
+				// First load the list of data sections, and sort them by
+				// address for easier processing.
+				var dataSections []memorySection
+				for _, entry := range section.Entries {
+					address, err := wasm.Eval(bytes.NewBuffer(entry.Offset))
+					if err != nil {
+						return nil, fmt.Errorf("could not parse data section address: %w", err)
+					}
+					dataSections = append(dataSections, memorySection{
+						Address: uint64(address[0].(int32)),
+						Size:    uint64(len(entry.Data)),
+						Type:    memoryData,
+					})
+				}
+				sort.Slice(dataSections, func(i, j int) bool {
+					return dataSections[i].Address < dataSections[j].Address
+				})
+
+				// And now add all data sections for linear memory.
+				// Parts that are in the slice of data sections are added as
+				// memoryData, and parts that are not are added as memoryBSS.
+				addr := uint64(0)
+				for _, section := range dataSections {
+					if addr < section.Address {
+						sections = append(sections, memorySection{
+							Address: addr,
+							Size:    section.Address - addr,
+							Type:    memoryBSS,
+						})
+					}
+					if addr > section.Address {
+						// This might be allowed, I'm not sure.
+						// It certainly doesn't make a lot of sense.
+						return nil, fmt.Errorf("overlapping data section")
+					}
+					// addr == section.Address
+					sections = append(sections, section)
+					addr = section.Address + section.Size
+				}
+				if addr < linearMemorySize {
+					sections = append(sections, memorySection{
+						Address: addr,
+						Size:    linearMemorySize - addr,
+						Type:    memoryBSS,
+					})
+				}
+			}
+		}
+	} else {
+		return nil, fmt.Errorf("could not parse file: %w", err)
 	}

-	sum := &packageSize{}
+	// Sort the slice of address chunks by address, so that we can iterate
+	// through it to calculate section sizes.
+	sort.Slice(addresses, func(i, j int) bool {
+		if addresses[i].Address == addresses[j].Address {
+			// Very rarely, there might be duplicate addresses.
+			// If that happens, sort the largest chunks first.
+			return addresses[i].Length > addresses[j].Length
+		}
+		return addresses[i].Address < addresses[j].Address
+	})
+
+	// Now finally determine the binary/RAM size usage per package by going
+	// through each allocated section.
+	sizes := make(map[string]packageSize)
+	for _, section := range sections {
+		switch section.Type {
+		case memoryCode:
+			readSection(section, addresses, func(path string, size uint64, isVariable bool) {
+				field := sizes[path]
+				if isVariable {
+					field.ROData += size
+				} else {
+					field.Code += size
+				}
+				sizes[path] = field
+			}, packagePathMap)
+		case memoryROData:
+			readSection(section, addresses, func(path string, size uint64, isVariable bool) {
+				field := sizes[path]
+				field.ROData += size
+				sizes[path] = field
+			}, packagePathMap)
+		case memoryData:
+			readSection(section, addresses, func(path string, size uint64, isVariable bool) {
+				field := sizes[path]
+				field.Data += size
+				sizes[path] = field
+			}, packagePathMap)
+		case memoryBSS:
+			readSection(section, addresses, func(path string, size uint64, isVariable bool) {
+				field := sizes[path]
+				field.BSS += size
+				sizes[path] = field
+			}, packagePathMap)
+		case memoryStack:
+			// We store the C stack as a pseudo-package.
+			sizes["C stack"] = packageSize{
+				BSS: section.Size,
+			}
+		}
+	}
+
+	// ...and summarize the results.
+	program := &programSize{
+		Packages: sizes,
+	}
 	for _, pkg := range sizes {
-		sum.Code += pkg.Code
-		sum.ROData += pkg.ROData
-		sum.Data += pkg.Data
-		sum.BSS += pkg.BSS
+		program.Code += pkg.Code
+		program.ROData += pkg.ROData
+		program.Data += pkg.Data
+		program.BSS += pkg.BSS
 	}
-
-	return &programSize{Packages: sizes, Code: sumCode, Data: sumData, BSS: sumBSS, Sum: sum}, nil
+	return program, nil
+}
+
+// readSection determines for each byte in this section to which package it
+// belongs. It reports this usage through the addSize callback.
+func readSection(section memorySection, addresses []addressLine, addSize func(string, uint64, bool), packagePathMap map[string]string) {
+	// The addr variable tracks at which address we are while going through this
+	// section. We start at the beginning.
+	addr := section.Address
+	sectionEnd := section.Address + section.Size
+	for _, line := range addresses {
+		if line.Address < section.Address || line.Address+line.Length >= sectionEnd {
+			// Check that this line is entirely within the section.
+			// Don't bother dealing with line entries that cross sections (that
+			// seems rather unlikely anyway).
+			continue
+		}
+		if addr < line.Address {
+			// There is a gap: there is a space between the current and the
+			// previous line entry.
+			addSize("(unknown)", line.Address-addr, false)
+			if sizesDebug {
+				fmt.Printf("%08x..%08x %4d: unknown (gap)\n", addr, line.Address, line.Address-addr)
+			}
+		}
+		if addr > line.Address+line.Length {
+			// The current line is already covered by a previous line entry.
+			// Simply skip it.
+			continue
+		}
+		// At this point, addr falls within the current line (probably at the
+		// start).
+		length := line.Length
+		if addr > line.Address {
+			// There is some overlap: the previous line entry already covered
+			// part of this line entry. So reduce the length to add to the
+			// remaining bit of the line entry.
+			length = line.Length - (addr - line.Address)
+		}
+		// Finally, mark this chunk of memory as used by the given package.
+		addSize(findPackagePath(line.File, packagePathMap), length, line.IsVariable)
+		addr = line.Address + line.Length
+	}
+	if addr < sectionEnd {
+		// There is a gap at the end of the section.
+		addSize("(unknown)", sectionEnd-addr, false)
+		if sizesDebug {
+			fmt.Printf("%08x..%08x %4d: unknown (end)\n", addr, sectionEnd, sectionEnd-addr)
+		}
+	}
+}
+
+// findPackagePath returns the Go package (or a pseudo package) for the given
+// path. It uses some heuristics, for example for some C libraries.
+func findPackagePath(path string, packagePathMap map[string]string) string {
+	// Check whether this path is part of one of the compiled packages.
+	packagePath, ok := packagePathMap[filepath.Dir(path)]
+	if !ok {
+		if strings.HasPrefix(path, filepath.Join(goenv.Get("TINYGOROOT"), "lib")) {
+			// Emit C libraries (in the lib subdirectory of TinyGo) as a single
+			// package, with a "C" prefix. For example: "C compiler-rt" for the
+			// compiler runtime library from LLVM.
+			packagePath = "C " + strings.Split(strings.TrimPrefix(path, filepath.Join(goenv.Get("TINYGOROOT"), "lib")), string(os.PathSeparator))[1]
+		} else if packageSymbolRegexp.MatchString(path) {
+			// Parse symbol names like main$alloc or runtime$string.
+			packagePath = path[:strings.LastIndex(path, "$")]
+		} else if reflectDataRegexp.MatchString(path) {
+			// Parse symbol names like reflect.structTypesSidetable.
+			packagePath = "Go reflect data"
+		} else if path == "<Go interface assert>" {
+			// Interface type assert, generated by the interface lowering pass.
+			packagePath = "Go interface assert"
+		} else if path == "<Go interface method>" {
+			// Interface method wrapper (switch over all concrete types),
+			// generated by the interface lowering pass.
+			packagePath = "Go interface method"
+		} else if path == "<stdin>" {
+			// This can happen when the source code (in Go) doesn't have a
+			// source file and uses "-" as the location. Somewhere this is
+			// converted to "<stdin>".
+			// Convert this back to the "-" string. Eventually, this should be
+			// fixed in the compiler.
+			packagePath = "-"
+		} else {
+			// This is some other path. Not sure what it is, so just emit its directory.
+			packagePath = filepath.Dir(path) // fallback
+		}
+	}
+	return packagePath
 }
@@ -13,6 +13,7 @@ import (
 #include <stdlib.h>
 bool tinygo_clang_driver(int argc, char **argv);
 bool tinygo_link_elf(int argc, char **argv);
+bool tinygo_link_mingw(int argc, char **argv);
 bool tinygo_link_wasm(int argc, char **argv);
 */
 import "C"
@@ -24,6 +25,10 @@ const hasBuiltinTools = true
 // 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 {
+	linker := "elf"
+	if tool == "ld.lld" && len(args) >= 2 && args[0] == "-m" && args[1] == "i386pep" {
+		linker = "mingw"
+	}
 	args = append([]string{"tinygo:" + tool}, args...)

 	var cflag *C.char
@@ -41,7 +46,14 @@ func RunTool(tool string, args ...string) error {
 	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))
+		switch linker {
+		case "elf":
+			ok = C.tinygo_link_elf(C.int(len(args)), (**C.char)(buf))
+		case "mingw":
+			ok = C.tinygo_link_mingw(C.int(len(args)), (**C.char)(buf))
+		default:
+			return errors.New("unknown linker: " + linker)
+		}
 	case "wasm-ld":
 		ok = C.tinygo_link_wasm(C.int(len(args)), (**C.char)(buf))
 	default:
@@ -24,7 +24,7 @@ func runCCompiler(flags ...string) error {
 	}

 	// Compile this with an external invocation of the Clang compiler.
-	return execCommand(commands["clang"], flags...)
+	return execCommand("clang", flags...)
 }

 // link invokes a linker with the given name and flags.
@@ -38,8 +38,8 @@ func link(linker string, flags ...string) error {
 	}

 	// Fall back to external command.
-	if cmdNames, ok := commands[linker]; ok {
-		return execCommand(cmdNames, flags...)
+	if _, ok := commands[linker]; ok {
+		return execCommand(linker, flags...)
 	}

 	cmd := exec.Command(linker, flags...)
@@ -141,11 +141,13 @@ func split(input []byte, limit int) [][]byte {
 	var block []byte
 	output := make([][]byte, 0, len(input)/limit+1)
 	for len(input) >= limit {
+		// add all blocks
 		block, input = input[:limit], input[limit:]
 		output = append(output, block)
 	}
 	if len(input) > 0 {
-		output = append(output, input[:len(input)])
+		// add remaining block (that isn't full sized)
+		output = append(output, input)
 	}
 	return output
 }
@@ -25,12 +25,19 @@ import (
 	"golang.org/x/tools/go/ast/astutil"
 )

+// Version of the cgo package. It must be incremented whenever the cgo package
+// is changed in a way that affects the output so that cached package builds
+// will be invalidated.
+// This version is independent of the TinyGo version number.
+const Version = 1 // last change: run libclang once per Go file
+
 // cgoPackage holds all CGo-related information of a package.
 type cgoPackage struct {
 	generated       *ast.File
 	generatedPos    token.Pos
 	errors          []error
-	dir             string
+	currentDir      string // current working directory
+	packageDir      string // full path to the package to process
 	fset            *token.FileSet
 	tokenFiles      map[string]*token.File
 	missingSymbols  map[string]struct{}
@@ -124,17 +131,17 @@ var cgoAliases = map[string]string{
 // builtinAliases are handled specially because they only exist on the Go side
 // of CGo, not on the CGo side (they're prefixed with "_Cgo_" there).
 var builtinAliases = map[string]struct{}{
-	"char":      struct{}{},
-	"schar":     struct{}{},
-	"uchar":     struct{}{},
-	"short":     struct{}{},
-	"ushort":    struct{}{},
-	"int":       struct{}{},
-	"uint":      struct{}{},
-	"long":      struct{}{},
-	"ulong":     struct{}{},
-	"longlong":  struct{}{},
-	"ulonglong": struct{}{},
+	"char":      {},
+	"schar":     {},
+	"uchar":     {},
+	"short":     {},
+	"ushort":    {},
+	"int":       {},
+	"uint":      {},
+	"long":      {},
+	"ulong":     {},
+	"longlong":  {},
+	"ulonglong": {},
 }

 // cgoTypes lists some C types with ambiguous sizes that must be retrieved
@@ -158,12 +165,13 @@ 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, 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) {
+// files, the CGo header snippets that should be compiled (for inline
+// functions), 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, []string, map[string][]byte, []error) {
 	p := &cgoPackage{
-		dir:             dir,
+		currentDir:      dir,
 		fset:            fset,
 		tokenFiles:      map[string]*token.File{},
 		missingSymbols:  map[string]struct{}{},
@@ -184,14 +192,14 @@ func Process(files []*ast.File, dir string, fset *token.FileSet, cflags []string
 	// 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{
+		return nil, 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)
+	p.packageDir = filepath.Dir(packagePath)

 	// Construct a new in-memory AST for CGo declarations of this package.
 	unsafeImport := &ast.ImportSpec{
@@ -224,7 +232,7 @@ func Process(files []*ast.File, dir string, fset *token.FileSet, cflags []string
 				Specs: []ast.Spec{
 					&ast.ValueSpec{
 						Names: []*ast.Ident{
-							&ast.Ident{
+							{
 								Name: "_",
 								Obj: &ast.Object{
 									Kind: ast.Var,
@@ -255,9 +263,10 @@ func Process(files []*ast.File, dir string, fset *token.FileSet, cflags []string
 		p.missingSymbols["_Cgo_"+name] = struct{}{}
 	}

-	// Find `import "C"` statements in the file.
-	var statements []*ast.GenDecl
-	for _, f := range files {
+	// Find `import "C"` C fragments in the file.
+	cgoHeaders := make([]string, len(files)) // combined CGo header fragment for each file
+	for i, f := range files {
+		var cgoHeader string
 		for i := 0; i < len(f.Decls); i++ {
 			decl := f.Decls[i]
 			genDecl, ok := decl.(*ast.GenDecl)
@@ -281,105 +290,40 @@ func Process(files []*ast.File, dir string, fset *token.FileSet, cflags []string
 				continue
 			}

-			// Found a CGo statement.
-			statements = append(statements, genDecl)
-
 			// Remove this import declaration.
 			f.Decls = append(f.Decls[:i], f.Decls[i+1:]...)
 			i--
-		}

-		// Print the AST, for debugging.
-		//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
-				}
+			if genDecl.Doc == nil {
+				continue
 			}
+
+			// Iterate through all parts of the CGo header. Note that every //
+			// line is a new comment.
+			position := fset.Position(genDecl.Doc.Pos())
+			fragment := fmt.Sprintf("# %d %#v\n", position.Line, position.Filename)
+			for _, comment := range genDecl.Doc.List {
+				// Find all #cgo lines, extract and use their contents, and
+				// replace the lines with spaces (to preserve locations).
+				c := p.parseCGoPreprocessorLines(comment.Text, comment.Slash)
+
+				// Change the comment (which is still in Go syntax, with // and
+				// /* */ ) to a regular string by replacing the start/end
+				// markers of comments with spaces.
+				// It is similar to the Text() method but differs in that it
+				// doesn't strip anything and tries to keep all offsets correct
+				// by adding spaces and newlines where necessary.
+				if c[1] == '/' { /* comment */
+					c = "  " + c[2:]
+				} else { // comment
+					c = "  " + c[2:len(c)-2]
+				}
+				fragment += c + "\n"
+			}
+			cgoHeader += fragment
 		}
+
+		cgoHeaders[i] = cgoHeader
 	}

 	// Define CFlags that will be used while parsing the package.
@@ -389,16 +333,9 @@ func Process(files []*ast.File, dir string, fset *token.FileSet, cflags []string
 	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)
+	// Process CGo imports for each file.
+	for i, f := range files {
+		p.parseFragment(cgoHeaders[i]+cgoTypes, cflagsForCGo, filepath.Base(fset.File(f.Pos()).Name()))
 	}

 	// Declare functions found by libclang.
@@ -433,18 +370,18 @@ 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.cflags, p.ldflags, p.visitedFiles, p.errors
+	return p.generated, cgoHeaders, 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) {
+func (p *cgoPackage) makePathsAbsolute(args []string) {
 	nextIsPath := false
 	for i, arg := range args {
 		if nextIsPath {
 			if !filepath.IsAbs(arg) {
-				args[i] = filepath.Join(packagePath, arg)
+				args[i] = filepath.Join(p.packageDir, arg)
 			}
 		}
 		if arg == "-I" || arg == "-L" {
@@ -454,12 +391,95 @@ func makePathsAbsolute(args []string, packagePath string) {
 		if strings.HasPrefix(arg, "-I") || strings.HasPrefix(arg, "-L") {
 			path := arg[2:]
 			if !filepath.IsAbs(path) {
-				args[i] = arg[:2] + filepath.Join(packagePath, path)
+				args[i] = arg[:2] + filepath.Join(p.packageDir, path)
 			}
 		}
 	}
 }

+// parseCGoPreprocessorLines reads #cgo pseudo-preprocessor lines in the source
+// text (import "C" fragment), stores their information such as CFLAGS, and
+// returns the same text but with those #cgo lines replaced by spaces (to keep
+// position offsets the same).
+func (p *cgoPackage) parseCGoPreprocessorLines(text string, pos token.Pos) string {
+	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(text, "#cgo ")
+		if lineStart < 0 {
+			break
+		}
+		lineLen := strings.IndexByte(text[lineStart:], '\n')
+		if lineLen < 0 {
+			lineLen = len(text) - lineStart
+		}
+		lineEnd := lineStart + lineLen
+		line := text[lineStart:lineEnd]
+		spaces := make([]byte, len(line))
+		for i := range spaces {
+			spaces[i] = ' '
+		}
+		text = text[:lineStart] + string(spaces) + text[lineEnd:]
+
+		// Get the text before the colon in the #cgo directive.
+		colon := strings.IndexByte(line, ':')
+		if colon < 0 {
+			p.addErrorAfter(pos, 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(pos, text[:lineStart+colon-1], "invalid #cgo line")
+			continue
+		}
+
+		if len(fields) > 1 {
+			p.addErrorAfter(pos, 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(pos, text[:lineStart+colon+1], "failed to parse flags in #cgo line: "+err.Error())
+				continue
+			}
+			if err := checkCompilerFlags(name, flags); err != nil {
+				p.addErrorAfter(pos, text[:lineStart+colon+1], err.Error())
+				continue
+			}
+			p.makePathsAbsolute(flags)
+			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(pos, text[:lineStart+colon+1], "failed to parse flags in #cgo line: "+err.Error())
+				continue
+			}
+			if err := checkLinkerFlags(name, flags); err != nil {
+				p.addErrorAfter(pos, text[:lineStart+colon+1], err.Error())
+				continue
+			}
+			p.makePathsAbsolute(flags)
+			p.ldflags = append(p.ldflags, flags...)
+		default:
+			startPos := strings.LastIndex(line[4:colon], name) + 4
+			p.addErrorAfter(pos, text[:lineStart+startPos], "invalid #cgo line: "+name)
+			continue
+		}
+	}
+	return text
+}
+
 // addFuncDecls adds the C function declarations found by libclang in the
 // comment above the `import "C"` statement.
 func (p *cgoPackage) addFuncDecls() {
@@ -494,7 +514,7 @@ func (p *cgoPackage) addFuncDecls() {
 		if fn.variadic {
 			decl.Doc = &ast.CommentGroup{
 				List: []*ast.Comment{
-					&ast.Comment{
+					{
 						Slash: fn.pos,
 						Text:  "//go:variadic",
 					},
@@ -505,7 +525,7 @@ func (p *cgoPackage) addFuncDecls() {
 		for i, arg := range fn.args {
 			args[i] = &ast.Field{
 				Names: []*ast.Ident{
-					&ast.Ident{
+					{
 						NamePos: fn.pos,
 						Name:    arg.name,
 						Obj: &ast.Object{
@@ -553,7 +573,7 @@ func (p *cgoPackage) addFuncPtrDecls() {
 			Name: "C." + name + "$funcaddr",
 		}
 		valueSpec := &ast.ValueSpec{
-			Names: []*ast.Ident{&ast.Ident{
+			Names: []*ast.Ident{{
 				NamePos: fn.pos,
 				Name:    "C." + name + "$funcaddr",
 				Obj:     obj,
@@ -603,7 +623,7 @@ func (p *cgoPackage) addConstDecls() {
 			Name: "C." + name,
 		}
 		valueSpec := &ast.ValueSpec{
-			Names: []*ast.Ident{&ast.Ident{
+			Names: []*ast.Ident{{
 				NamePos: constVal.pos,
 				Name:    "C." + name,
 				Obj:     obj,
@@ -644,7 +664,7 @@ func (p *cgoPackage) addVarDecls() {
 			Name: "C." + name,
 		}
 		valueSpec := &ast.ValueSpec{
-			Names: []*ast.Ident{&ast.Ident{
+			Names: []*ast.Ident{{
 				NamePos: global.pos,
 				Name:    "C." + name,
 				Obj:     obj,
@@ -845,9 +865,9 @@ func (p *cgoPackage) makeUnionField(typ *elaboratedTypeInfo) *ast.StructType {
 		Struct: typ.typeExpr.Struct,
 		Fields: &ast.FieldList{
 			Opening: typ.typeExpr.Fields.Opening,
-			List: []*ast.Field{&ast.Field{
+			List: []*ast.Field{{
 				Names: []*ast.Ident{
-					&ast.Ident{
+					{
 						NamePos: typ.typeExpr.Fields.Opening,
 						Name:    "$union",
 					},
@@ -928,9 +948,9 @@ func (p *cgoPackage) createUnionAccessor(field *ast.Field, typeName string) {
 		Recv: &ast.FieldList{
 			Opening: pos,
 			List: []*ast.Field{
-				&ast.Field{
+				{
 					Names: []*ast.Ident{
-						&ast.Ident{
+						{
 							NamePos: pos,
 							Name:    "union",
 						},
@@ -959,7 +979,7 @@ func (p *cgoPackage) createUnionAccessor(field *ast.Field, typeName string) {
 			},
 			Results: &ast.FieldList{
 				List: []*ast.Field{
-					&ast.Field{
+					{
 						Type: &ast.StarExpr{
 							Star: pos,
 							X:    field.Type,
@@ -1038,9 +1058,9 @@ func (p *cgoPackage) createBitfieldGetter(bitfield bitfieldInfo, typeName string
 		Recv: &ast.FieldList{
 			Opening: bitfield.pos,
 			List: []*ast.Field{
-				&ast.Field{
+				{
 					Names: []*ast.Ident{
-						&ast.Ident{
+						{
 							NamePos: bitfield.pos,
 							Name:    "s",
 							Obj: &ast.Object{
@@ -1074,7 +1094,7 @@ func (p *cgoPackage) createBitfieldGetter(bitfield bitfieldInfo, typeName string
 			},
 			Results: &ast.FieldList{
 				List: []*ast.Field{
-					&ast.Field{
+					{
 						Type: bitfield.field.Type,
 					},
 				},
@@ -1191,9 +1211,9 @@ func (p *cgoPackage) createBitfieldSetter(bitfield bitfieldInfo, typeName string
 		Recv: &ast.FieldList{
 			Opening: bitfield.pos,
 			List: []*ast.Field{
-				&ast.Field{
+				{
 					Names: []*ast.Ident{
-						&ast.Ident{
+						{
 							NamePos: bitfield.pos,
 							Name:    "s",
 							Obj: &ast.Object{
@@ -1224,9 +1244,9 @@ func (p *cgoPackage) createBitfieldSetter(bitfield bitfieldInfo, typeName string
 			Params: &ast.FieldList{
 				Opening: bitfield.pos,
 				List: []*ast.Field{
-					&ast.Field{
+					{
 						Names: []*ast.Ident{
-							&ast.Ident{
+							{
 								NamePos: bitfield.pos,
 								Name:    "value",
 								Obj:     nil,
@@ -12,7 +12,6 @@ import (
 	"go/types"
 	"io/ioutil"
 	"path/filepath"
-	"regexp"
 	"runtime"
 	"strings"
 	"testing"
@@ -25,19 +24,11 @@ var flagUpdate = flag.Bool("update", false, "Update images based on test output.
 // 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"}
+	var cflags = []string{"--target=armv6m-unknown-unknown-eabi"}

 	for _, name := range []string{"basic", "errors", "types", "flags", "const"} {
 		name := name // avoid a race condition
@@ -65,7 +56,7 @@ func TestCGo(t *testing.T) {
 			}

 			// Process the AST with CGo.
-			cgoAST, _, _, _, cgoErrors := Process([]*ast.File{f}, "testdata", fset, cflags)
+			cgoAST, _, _, _, _, cgoErrors := Process([]*ast.File{f}, "testdata", fset, cflags)

 			// Check the AST for type errors.
 			var typecheckErrors []error
@@ -105,7 +96,7 @@ func TestCGo(t *testing.T) {
 			if err != nil {
 				t.Errorf("could not write out CGo AST: %v", err)
 			}
-			actual := normalizeResult(string(buf.Bytes()))
+			actual := normalizeResult(buf.String())

 			// Read the file with the expected output, to compare against.
 			outfile := filepath.Join("testdata", name+".out.go")
@@ -16,7 +16,7 @@ import (
 )

 /*
-#include <clang-c/Index.h> // if this fails, install libclang-10-dev
+#include <clang-c/Index.h> // if this fails, install libclang-11-dev
 #include <stdlib.h>
 #include <stdint.h>

@@ -72,17 +72,14 @@ var diagnosticSeverity = [...]string{
 	C.CXDiagnostic_Fatal:   "fatal",
 }

-func (p *cgoPackage) parseFragment(fragment string, cflags []string, posFilename string, posLine int) {
+func (p *cgoPackage) parseFragment(fragment string, cflags []string, filename string) {
 	index := C.clang_createIndex(0, 0)
 	defer C.clang_disposeIndex(index)

 	// pretend to be a .c file
-	filenameC := C.CString(posFilename + "!cgo.c")
+	filenameC := C.CString(filename + "!cgo.c")
 	defer C.free(unsafe.Pointer(filenameC))

-	// fix up error locations
-	fragment = fmt.Sprintf("# %d %#v\n", posLine+1, posFilename) + fragment
-
 	fragmentC := C.CString(fragment)
 	defer C.free(unsafe.Pointer(fragmentC))

@@ -196,14 +193,14 @@ func tinygo_clang_globals_visitor(c, parent C.GoCXCursor, client_data C.CXClient
 			}
 			fn.args = append(fn.args, paramInfo{
 				name:     argName,
-				typeExpr: p.makeASTType(argType, pos),
+				typeExpr: p.makeDecayingASTType(argType, pos),
 			})
 		}
 		resultType := C.tinygo_clang_getCursorResultType(c)
 		if resultType.kind != C.CXType_Void {
 			fn.results = &ast.FieldList{
 				List: []*ast.Field{
-					&ast.Field{
+					{
 						Type: p.makeASTType(resultType, pos),
 					},
 				},
@@ -380,7 +377,7 @@ func (p *cgoPackage) addErrorAfter(pos token.Pos, after, msg string) {
 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)
+		relpath, err := filepath.Rel(p.currentDir, position.Filename)
 		if err == nil {
 			position.Filename = relpath
 		}
@@ -391,6 +388,41 @@ func (p *cgoPackage) addErrorAt(position token.Position, msg string) {
 	})
 }

+// makeDecayingASTType does the same as makeASTType but takes care of decaying
+// types (arrays in function parameters, etc). It is otherwise identical to
+// makeASTType.
+func (p *cgoPackage) makeDecayingASTType(typ C.CXType, pos token.Pos) ast.Expr {
+	// Strip typedefs, if any.
+	underlyingType := typ
+	if underlyingType.kind == C.CXType_Typedef {
+		c := C.tinygo_clang_getTypeDeclaration(typ)
+		underlyingType = C.tinygo_clang_getTypedefDeclUnderlyingType(c)
+		// TODO: support a chain of typedefs. At the moment, it seems to get
+		// stuck in an endless loop when trying to get to the most underlying
+		// type.
+	}
+	// Check for decaying type. An example would be an array type in a
+	// parameter. This declaration:
+	//   void foo(char buf[6]);
+	// is the same as this one:
+	//   void foo(char *buf);
+	// But this one:
+	//   void bar(char buf[6][4]);
+	// equals this:
+	//   void bar(char *buf[4]);
+	// so not all array dimensions should be stripped, just the first one.
+	// TODO: there are more kinds of decaying types.
+	if underlyingType.kind == C.CXType_ConstantArray {
+		// Apply type decaying.
+		pointeeType := C.clang_getElementType(underlyingType)
+		return &ast.StarExpr{
+			Star: pos,
+			X:    p.makeASTType(pointeeType, pos),
+		}
+	}
+	return p.makeASTType(typ, pos)
+}
+
 // makeASTType return the ast.Expr for the given libclang type. In other words,
 // it converts a libclang type to a type in the Go AST.
 func (p *cgoPackage) makeASTType(typ C.CXType, pos token.Pos) ast.Expr {
@@ -775,7 +807,7 @@ func tinygo_clang_struct_visitor(c, parent C.GoCXCursor, client_data C.CXClientD
 	}
 	*inBitfield = false
 	field.Names = []*ast.Ident{
-		&ast.Ident{
+		{
 			NamePos: pos,
 			Name:    name,
 			Obj: &ast.Object{
@@ -796,8 +828,12 @@ func tinygo_clang_enum_visitor(c, parent C.GoCXCursor, client_data C.CXClientDat
 	pos := p.getCursorPosition(c)
 	value := C.tinygo_clang_getEnumConstantDeclValue(c)
 	p.constants[name] = constantInfo{
-		expr: &ast.BasicLit{pos, token.INT, strconv.FormatInt(int64(value), 10)},
-		pos:  pos,
+		expr: &ast.BasicLit{
+			ValuePos: pos,
+			Kind:     token.INT,
+			Value:    strconv.FormatInt(int64(value), 10),
+		},
+		pos: pos,
 	}
 	return C.CXChildVisit_Continue
 }
@@ -1,5 +1,4 @@
 // +build !byollvm
-// +build !llvm10

 package cgo

@@ -1,14 +0,0 @@
-// +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"
@@ -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-10-dev
+#include <clang-c/Index.h> // if this fails, install libclang-11-dev

 CXCursor tinygo_clang_getTranslationUnitCursor(CXTranslationUnit tu) {
 	return clang_getTranslationUnitCursor(tu);
@@ -14,6 +14,12 @@ typedef someType noType; // undefined type
 #define SOME_CONST_2 6) // const not used (so no error)
 #define SOME_CONST_3 1234 // const too large for byte
 */
+//
+//
+// #define SOME_CONST_4 8) // after some empty lines
+import "C"
+
+// #warning another warning
 import "C"

 // Make sure that errors for the following lines won't change with future
@@ -30,4 +36,6 @@ var (
 	_ = C.SOME_CONST_1

 	_ byte = C.SOME_CONST_3
+
+	_ = C.SOME_CONST_4
 )
@@ -1,13 +1,16 @@
 // CGo errors:
 //     testdata/errors.go:4:2: warning: some warning
 //     testdata/errors.go:11:9: error: unknown type name 'someType'
+//     testdata/errors.go:22:5: warning: another warning
 //     testdata/errors.go:13:23: unexpected token ), expected end of expression
+//     testdata/errors.go:19:26: 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)
+//     testdata/errors.go:112: undeclared name: C.SOME_CONST_4

 package main

@@ -111,6 +111,10 @@ void variadic2(int x, int y, ...);
 */
 import "C"

+// // Test that we can refer from this CGo fragment to the fragment above.
+// typedef myint myint2;
+import "C"
+
 var (
 	// Simple typedefs.
 	_ C.myint
@@ -67,9 +67,7 @@ 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) 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
 }
@@ -105,15 +103,9 @@ type C.struct_type1 struct {
 }
 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.$union))
-}
+func (union *C.union_1) unionfield_i() *C.int   { return (*C.int)(unsafe.Pointer(&union.$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 }

@@ -138,9 +130,7 @@ func (union *C.union_3) unionfield_thing() *C.union3_t {

 type C.union_3 struct{ $union [2]uint64 }

-func (union *C.union_union2d) unionfield_i() *C.int {
-	return (*C.int)(unsafe.Pointer(&union.$union))
-}
+func (union *C.union_union2d) unionfield_i() *C.int { return (*C.int)(unsafe.Pointer(&union.$union)) }
 func (union *C.union_union2d) unionfield_d() *[2]float64 {
 	return (*[2]float64)(unsafe.Pointer(&union.$union))
 }
@@ -5,6 +5,7 @@ package compileopts
 import (
 	"errors"
 	"fmt"
+	"os"
 	"path/filepath"
 	"regexp"
 	"strings"
@@ -21,7 +22,7 @@ type Config struct {
 	TestConfig     TestConfig
 }

-// Triple returns the LLVM target triple, like armv6m-none-eabi.
+// Triple returns the LLVM target triple, like armv6m-unknown-unknown-eabi.
 func (c *Config) Triple() string {
 	return c.Target.Triple
 }
@@ -33,10 +34,16 @@ func (c *Config) CPU() string {
 }

 // 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
+// RISC-V processor, that could be "+a,+c,+m". For many targets, an empty list
+// will be returned.
+func (c *Config) Features() string {
+	if c.Target.Features == "" {
+		return c.Options.LLVMFeatures
+	}
+	if c.Options.LLVMFeatures == "" {
+		return c.Target.Features
+	}
+	return c.Target.Features + "," + c.Options.LLVMFeatures
 }

 // GOOS returns the GOOS of the target. This might not always be the actual OS:
@@ -53,9 +60,15 @@ func (c *Config) GOARCH() string {
 	return c.Target.GOARCH
 }

+// GOARM will return the GOARM environment variable given to the compiler when
+// building a program.
+func (c *Config) GOARM() string {
+	return c.Options.GOARM
+}
+
 // 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()}...)
+	tags := append(c.Target.BuildTags, []string{"tinygo", "math_big_pure_go", "gc." + c.GC(), "scheduler." + c.Scheduler(), "serial." + c.Serial()}...)
 	for i := 1; i <= c.GoMinorVersion; i++ {
 		tags = append(tags, fmt.Sprintf("go1.%d", i))
 	}
@@ -150,7 +163,7 @@ func (c *Config) OptLevels() (optLevel, sizeLevel int, inlinerThreshold uint) {
 // target.
 func (c *Config) FuncImplementation() string {
 	switch c.Scheduler() {
-	case "tasks":
+	case "tasks", "asyncify":
 		// 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
@@ -197,6 +210,39 @@ func (c *Config) RP2040BootPatch() bool {
 	return false
 }

+// MuslArchitecture returns the architecture name as used in musl libc. It is
+// usually the same as the first part of the LLVM triple, but not always.
+func MuslArchitecture(triple string) string {
+	arch := strings.Split(triple, "-")[0]
+	if strings.HasPrefix(arch, "arm") || strings.HasPrefix(arch, "thumb") {
+		arch = "arm"
+	}
+	return arch
+}
+
+// LibcPath returns the path to the libc directory. The libc path will be either
+// a precompiled libc shipped with a TinyGo build, or a libc path in the cache
+// directory (which might not yet be built).
+func (c *Config) LibcPath(name string) (path string, precompiled bool) {
+	// Try to load a precompiled library.
+	precompiledDir := filepath.Join(goenv.Get("TINYGOROOT"), "pkg", c.Triple(), name)
+	if _, err := os.Stat(precompiledDir); err == nil {
+		// Found a precompiled library for this OS/architecture. Return the path
+		// directly.
+		return precompiledDir, true
+	}
+
+	// No precompiled library found. Determine the path name that will be used
+	// in the build cache.
+	var outname string
+	if c.CPU() != "" {
+		outname = name + "-" + c.Triple() + "-" + c.CPU()
+	} else {
+		outname = name + "-" + c.Triple()
+	}
+	return filepath.Join(goenv.Get("GOCACHE"), outname), false
+}
+
 // CFlags returns the flags to pass to the C compiler. This is necessary for CGo
 // preprocessing.
 func (c *Config) CFlags() []string {
@@ -204,13 +250,62 @@ func (c *Config) CFlags() []string {
 	for _, flag := range c.Target.CFlags {
 		cflags = append(cflags, strings.ReplaceAll(flag, "{root}", goenv.Get("TINYGOROOT")))
 	}
-	if c.Target.Libc == "picolibc" {
+	switch c.Target.Libc {
+	case "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"))
+		picolibcDir := filepath.Join(root, "lib", "picolibc", "newlib", "libc")
+		path, _ := c.LibcPath("picolibc")
+		cflags = append(cflags,
+			"--sysroot="+path,
+			"-Xclang", "-internal-isystem", "-Xclang", filepath.Join(picolibcDir, "include"),
+			"-Xclang", "-internal-isystem", "-Xclang", filepath.Join(picolibcDir, "tinystdio"),
+		)
+	case "musl":
+		root := goenv.Get("TINYGOROOT")
+		path, _ := c.LibcPath("musl")
+		arch := MuslArchitecture(c.Triple())
+		cflags = append(cflags,
+			"--sysroot="+path,
+			"-Xclang", "-internal-isystem", "-Xclang", filepath.Join(root, "lib", "musl", "arch", arch),
+			"-Xclang", "-internal-isystem", "-Xclang", filepath.Join(root, "lib", "musl", "include"),
+		)
+	case "wasi-libc":
+		root := goenv.Get("TINYGOROOT")
+		cflags = append(cflags, "--sysroot="+root+"/lib/wasi-libc/sysroot")
+	case "mingw-w64":
+		root := goenv.Get("TINYGOROOT")
+		path, _ := c.LibcPath("mingw-w64")
+		cflags = append(cflags,
+			"--sysroot="+path,
+			"-Xclang", "-internal-isystem", "-Xclang", filepath.Join(root, "lib", "mingw-w64", "mingw-w64-headers", "crt"),
+			"-Xclang", "-internal-isystem", "-Xclang", filepath.Join(root, "lib", "mingw-w64", "mingw-w64-headers", "defaults", "include"),
+			"-D_UCRT",
+		)
+	case "":
+		// No libc specified, nothing to add.
+	default:
+		// Incorrect configuration. This could be handled in a better way, but
+		// usually this will be found by developers (not by TinyGo users).
+		panic("unknown libc: " + c.Target.Libc)
 	}
-	if c.Debug() {
-		cflags = append(cflags, "-g")
+	// Always emit debug information. It is optionally stripped at link time.
+	cflags = append(cflags, "-g")
+	// Use the same optimization level as TinyGo.
+	cflags = append(cflags, "-O"+c.Options.Opt)
+	// Set the LLVM target triple.
+	cflags = append(cflags, "--target="+c.Triple())
+	// Set the -mcpu (or similar) flag.
+	if c.Target.CPU != "" {
+		if c.GOARCH() == "amd64" || c.GOARCH() == "386" {
+			// x86 prefers the -march flag (-mcpu is deprecated there).
+			cflags = append(cflags, "-march="+c.Target.CPU)
+		} else if strings.HasPrefix(c.Triple(), "avr") {
+			// AVR MCUs use -mmcu instead of -mcpu.
+			cflags = append(cflags, "-mmcu="+c.Target.CPU)
+		} else {
+			// The rest just uses -mcpu.
+			cflags = append(cflags, "-mcpu="+c.Target.CPU)
+		}
 	}
 	return cflags
 }
@@ -250,8 +345,9 @@ 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.
+// Debug returns whether debug (DWARF) information should be retained by the
+// linker. By default, debug information is retained but it can be removed with
+// the -no-debug flag.
 func (c *Config) Debug() bool {
 	return c.Options.Debug
 }
@@ -297,6 +393,9 @@ func (c *Config) Programmer() (method, openocdInterface string) {
 	case "openocd", "msd", "command":
 		// The -programmer flag only specifies the flash method.
 		return c.Options.Programmer, c.Target.OpenOCDInterface
+	case "bmp":
+		// The -programmer flag only specifies the flash method.
+		return c.Options.Programmer, ""
 	default:
 		// The -programmer flag specifies something else, assume it specifies
 		// the OpenOCD interface name.
@@ -363,10 +462,6 @@ func (c *Config) WasmAbi() string {
 	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
@@ -8,7 +8,7 @@ import (

 var (
 	validGCOptions            = []string{"none", "leaking", "extalloc", "conservative"}
-	validSchedulerOptions     = []string{"none", "tasks", "coroutines"}
+	validSchedulerOptions     = []string{"none", "tasks", "coroutines", "asyncify"}
 	validSerialOptions        = []string{"none", "uart", "usb"}
 	validPrintSizeOptions     = []string{"none", "short", "full"}
 	validPanicStrategyOptions = []string{"print", "trap"}
@@ -16,8 +16,12 @@ var (
 )

 // Options contains extra options to give to the compiler. These options are
-// usually passed from the command line.
+// usually passed from the command line, but can also be passed in environment
+// variables for example.
 type Options struct {
+	GOOS            string // environment variable
+	GOARCH          string // environment variable
+	GOARM           string // environment variable (only used with GOARCH=arm)
 	Target          string
 	Opt             string
 	GC              string
@@ -28,6 +32,7 @@ type Options struct {
 	DumpSSA         bool
 	VerifyIR        bool
 	PrintCommands   func(cmd string, args ...string)
+	Parallelism     int // -p flag
 	Debug           bool
 	PrintSizes      string
 	PrintAllocs     *regexp.Regexp // regexp string
@@ -10,7 +10,7 @@ import (
 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`)
+	expectedSchedulerError := errors.New(`invalid scheduler option 'incorrect': valid values are none, tasks, coroutines, asyncify`)
 	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`)

@@ -5,6 +5,7 @@ package compileopts
 import (
 	"encoding/json"
 	"errors"
+	"fmt"
 	"io"
 	"os"
 	"os/exec"
@@ -25,7 +26,7 @@ type TargetSpec struct {
 	Inherits         []string `json:"inherits"`
 	Triple           string   `json:"llvm-target"`
 	CPU              string   `json:"cpu"`
-	Features         []string `json:"features"`
+	Features         string   `json:"features"`
 	GOOS             string   `json:"goos"`
 	GOARCH           string   `json:"goarch"`
 	BuildTags        []string `json:"build-tags"`
@@ -94,7 +95,7 @@ func (spec *TargetSpec) overrideProperties(child *TargetSpec) {
 					dst.Set(src)
 				case "append", "":
 					// or append the field of child to spec
-					dst.Set(reflect.AppendSlice(src, dst))
+					dst.Set(reflect.AppendSlice(dst, src))
 				default:
 					panic("override mode must be 'copy' or 'append' (default). I don't know how to '" + tag + "'.")
 				}
@@ -161,84 +162,78 @@ func (spec *TargetSpec) resolveInherits() error {
 }

 // Load a target specification.
-func LoadTarget(target string) (*TargetSpec, error) {
-	if target == "" {
+func LoadTarget(options *Options) (*TargetSpec, error) {
+	if options.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
+		var llvmarch string
+		switch options.GOARCH {
+		case "386":
+			llvmarch = "i386"
+		case "amd64":
+			llvmarch = "x86_64"
+		case "arm64":
+			llvmarch = "aarch64"
+		case "arm":
+			switch options.GOARM {
+			case "5":
+				llvmarch = "armv5"
+			case "6":
+				llvmarch = "armv6"
+			case "7":
+				llvmarch = "armv7"
+			default:
+				return nil, fmt.Errorf("invalid GOARM=%s, must be 5, 6, or 7", options.GOARM)
+			}
+		default:
+			llvmarch = options.GOARCH
 		}
-		target = llvmarch + "--" + llvmos
-		if goarch == "arm" {
+		llvmos := options.GOOS
+		if llvmos == "darwin" {
+			// Use macosx* instead of darwin, otherwise darwin/arm64 will refer
+			// to iOS!
+			llvmos = "macosx10.12.0"
+			if llvmarch == "aarch64" {
+				// Looks like Apple prefers to call this architecture ARM64
+				// instead of AArch64.
+				llvmarch = "arm64"
+			}
+		}
+		// Target triples (which actually have four components, but are called
+		// triples for historical reasons) have the form:
+		//   arch-vendor-os-environment
+		target := llvmarch + "-unknown-" + llvmos
+		if options.GOARCH == "arm" {
 			target += "-gnueabihf"
 		}
-		return defaultTarget(goos, goarch, target)
+		if options.GOOS == "windows" {
+			target += "-gnu"
+		}
+		return defaultTarget(options.GOOS, options.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.
+	err := spec.loadFromGivenStr(options.Target)
+	if err != nil {
 		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, "-"))
 	}
+	// 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
+	}
+
+	if spec.Scheduler == "asyncify" {
+		spec.ExtraFiles = append(spec.ExtraFiles, "src/internal/task/task_asyncify_wasm.S")
+	}
+
+	return spec, nil
 }

-// 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{
@@ -249,36 +244,74 @@ func defaultTarget(goos, goarch, triple string) (*TargetSpec, error) {
 		Scheduler:        "tasks",
 		Linker:           "cc",
 		DefaultStackSize: 1024 * 64, // 64kB
-		CFlags:           []string{"--target=" + triple},
 		GDB:              []string{"gdb"},
 		PortReset:        "false",
 	}
+	switch goarch {
+	case "386":
+		spec.CPU = "pentium4"
+		spec.Features = "+cx8,+fxsr,+mmx,+sse,+sse2,+x87"
+	case "amd64":
+		spec.CPU = "x86-64"
+		spec.Features = "+cx8,+fxsr,+mmx,+sse,+sse2,+x87"
+	case "arm":
+		spec.CPU = "generic"
+		switch strings.Split(triple, "-")[0] {
+		case "armv5":
+			spec.Features = "+armv5t,+strict-align,-thumb-mode"
+		case "armv6":
+			spec.Features = "+armv6,+dsp,+fp64,+strict-align,+vfp2,+vfp2sp,-thumb-mode"
+		case "armv7":
+			spec.Features = "+armv7-a,+dsp,+fp64,+vfp2,+vfp2sp,+vfp3d16,+vfp3d16sp,-thumb-mode"
+		}
+	case "arm64":
+		spec.CPU = "generic"
+		spec.Features = "+neon"
+	}
 	if goos == "darwin" {
 		spec.CFlags = append(spec.CFlags, "-isysroot", "/Library/Developer/CommandLineTools/SDKs/MacOSX.sdk")
 		spec.LDFlags = append(spec.LDFlags, "-Wl,-dead_strip")
+	} else if goos == "linux" {
+		spec.Linker = "ld.lld"
+		spec.RTLib = "compiler-rt"
+		spec.Libc = "musl"
+		spec.LDFlags = append(spec.LDFlags, "--gc-sections")
+	} else if goos == "windows" {
+		spec.Linker = "ld.lld"
+		spec.Libc = "mingw-w64"
+		spec.LDFlags = append(spec.LDFlags,
+			"-m", "i386pep",
+			"-Bdynamic",
+			"--image-base", "0x400000",
+			"--gc-sections",
+			"--no-insert-timestamp",
+		)
 	} 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")
+		suffix := ""
+		if goos == "windows" {
+			// Windows uses a different calling convention from other operating
+			// systems so we need separate assembly files.
+			suffix = "_windows"
+		}
+		spec.ExtraFiles = append(spec.ExtraFiles, "src/runtime/gc_"+goarch+suffix+".S")
+		spec.ExtraFiles = append(spec.ExtraFiles, "src/internal/task/task_stack_"+goarch+suffix+".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"}
+			spec.Emulator = []string{"qemu-arm"}
 		}
 		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"}
+			spec.Emulator = []string{"qemu-aarch64"}
 		}
-		if goarch == "386" && runtime.GOARCH == "amd64" {
-			spec.CFlags = append(spec.CFlags, "-m32")
-			spec.LDFlags = append(spec.LDFlags, "-m32")
+	}
+	if goos != runtime.GOOS {
+		if goos == "windows" {
+			spec.Emulator = []string{"wine"}
 		}
 	}
 	return &spec, nil
@@ -1,22 +1,23 @@
 package compileopts

 import (
+	"os"
 	"reflect"
 	"testing"
 )

 func TestLoadTarget(t *testing.T) {
-	_, err := LoadTarget("arduino")
+	_, err := LoadTarget(&Options{Target: "arduino"})
 	if err != nil {
 		t.Error("LoadTarget test failed:", err)
 	}

-	_, err = LoadTarget("notexist")
+	_, err = LoadTarget(&Options{Target: "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" {
+	if !os.IsNotExist(err) {
 		t.Error("LoadTarget failed for wrong reason:", err)
 	}
 }
@@ -26,7 +27,7 @@ func TestOverrideProperties(t *testing.T) {
 	base := &TargetSpec{
 		GOOS:             "baseGoos",
 		CPU:              "baseCpu",
-		Features:         []string{"bf1", "bf2"},
+		CFlags:           []string{"-base-foo", "-base-bar"},
 		BuildTags:        []string{"bt1", "bt2"},
 		Emulator:         []string{"be1", "be2"},
 		DefaultStackSize: 42,
@@ -36,7 +37,7 @@ func TestOverrideProperties(t *testing.T) {
 	child := &TargetSpec{
 		GOOS:             "",
 		CPU:              "chlidCpu",
-		Features:         []string{"cf1", "cf2"},
+		CFlags:           []string{"-child-foo", "-child-bar"},
 		Emulator:         []string{"ce1", "ce2"},
 		AutoStackSize:    &childAutoStackSize,
 		DefaultStackSize: 64,
@@ -50,8 +51,8 @@ func TestOverrideProperties(t *testing.T) {
 	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.CFlags, []string{"-base-foo", "-base-bar", "-child-foo", "-child-bar"}) {
+		t.Errorf("Overriding failed : got %v", base.CFlags)
 	}
 	if !reflect.DeepEqual(base.BuildTags, []string{"bt1", "bt2"}) {
 		t.Errorf("Overriding failed : got %v", base.BuildTags)
@@ -0,0 +1,102 @@
+package compiler
+
+// This file defines alias functions for functions that are normally defined in
+// Go assembly.
+//
+// The Go toolchain defines many performance critical functions in assembly
+// instead of plain Go. This is a problem for TinyGo as it currently (as of
+// august 2021) is not able to compile these assembly files and even if it
+// could, it would not be able to make use of them for many targets that are
+// supported by TinyGo (baremetal RISC-V, AVR, etc). Therefore, many of these
+// functions are aliased to their generic Go implementation.
+// This results in slower than possible implementations, but at least they are
+// usable.
+
+import "tinygo.org/x/go-llvm"
+
+var stdlibAliases = map[string]string{
+	// crypto packages
+	"crypto/md5.block":         "crypto/md5.blockGeneric",
+	"crypto/sha1.block":        "crypto/sha1.blockGeneric",
+	"crypto/sha1.blockAMD64":   "crypto/sha1.blockGeneric",
+	"crypto/sha256.block":      "crypto/sha256.blockGeneric",
+	"crypto/sha512.blockAMD64": "crypto/sha512.blockGeneric",
+
+	// math package
+	"math.Asin":      "math.asin",
+	"math.Asinh":     "math.asinh",
+	"math.Acos":      "math.acos",
+	"math.Acosh":     "math.acosh",
+	"math.Atan":      "math.atan",
+	"math.Atanh":     "math.atanh",
+	"math.Atan2":     "math.atan2",
+	"math.Cbrt":      "math.cbrt",
+	"math.Ceil":      "math.ceil",
+	"math.archCeil":  "math.ceil",
+	"math.Cos":       "math.cos",
+	"math.Cosh":      "math.cosh",
+	"math.Erf":       "math.erf",
+	"math.Erfc":      "math.erfc",
+	"math.Exp":       "math.exp",
+	"math.archExp":   "math.exp",
+	"math.Expm1":     "math.expm1",
+	"math.Exp2":      "math.exp2",
+	"math.archExp2":  "math.exp2",
+	"math.Floor":     "math.floor",
+	"math.archFloor": "math.floor",
+	"math.Frexp":     "math.frexp",
+	"math.Hypot":     "math.hypot",
+	"math.archHypot": "math.hypot",
+	"math.Ldexp":     "math.ldexp",
+	"math.Log":       "math.log",
+	"math.archLog":   "math.log",
+	"math.Log1p":     "math.log1p",
+	"math.Log10":     "math.log10",
+	"math.Log2":      "math.log2",
+	"math.Max":       "math.max",
+	"math.archMax":   "math.max",
+	"math.Min":       "math.min",
+	"math.archMin":   "math.min",
+	"math.Mod":       "math.mod",
+	"math.Modf":      "math.modf",
+	"math.archModf":  "math.modf",
+	"math.Pow":       "math.pow",
+	"math.Remainder": "math.remainder",
+	"math.Sin":       "math.sin",
+	"math.Sinh":      "math.sinh",
+	"math.Sqrt":      "math.sqrt",
+	"math.archSqrt":  "math.sqrt",
+	"math.Tan":       "math.tan",
+	"math.Tanh":      "math.tanh",
+	"math.Trunc":     "math.trunc",
+	"math.archTrunc": "math.trunc",
+}
+
+// createAlias implements the function (in the builder) as a call to the alias
+// function.
+func (b *builder) createAlias(alias llvm.Value) {
+	b.llvmFn.SetVisibility(llvm.HiddenVisibility)
+	b.llvmFn.SetUnnamedAddr(true)
+
+	if b.Debug {
+		if b.fn.Syntax() != nil {
+			// Create debug info file if present.
+			b.difunc = b.attachDebugInfo(b.fn)
+		}
+		pos := b.program.Fset.Position(b.fn.Pos())
+		b.SetCurrentDebugLocation(uint(pos.Line), uint(pos.Column), b.difunc, llvm.Metadata{})
+	}
+	entryBlock := llvm.AddBasicBlock(b.llvmFn, "entry")
+	b.SetInsertPointAtEnd(entryBlock)
+	if b.llvmFn.Type() != alias.Type() {
+		b.addError(b.fn.Pos(), "alias function should have the same type as aliasee "+alias.Name())
+		b.CreateUnreachable()
+		return
+	}
+	result := b.CreateCall(alias, b.llvmFn.Params(), "")
+	if result.Type().TypeKind() == llvm.VoidTypeKind {
+		b.CreateRetVoid()
+	} else {
+		b.CreateRet(result)
+	}
+}
@@ -15,22 +15,16 @@ import (
 // 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 (b *builder) createLookupBoundsCheck(arrayLen, index llvm.Value, indexType types.Type) {
+// The caller should make sure that index is at least as big as arrayLen.
+func (b *builder) createLookupBoundsCheck(arrayLen, index llvm.Value) {
 	if b.info.nobounds {
 		// The //go:nobounds pragma was added to the function to avoid bounds
 		// checking.
 		return
 	}

-	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.Underlying().(*types.Basic).Info()&types.IsUnsigned == 0 {
-			index = b.CreateZExt(index, arrayLen.Type(), "")
-		} else {
-			index = b.CreateSExt(index, arrayLen.Type(), "")
-		}
-	} else if index.Type().IntTypeWidth() > arrayLen.Type().IntTypeWidth() {
+	// Extend arrayLen if it's too small.
+	if index.Type().IntTypeWidth() > arrayLen.Type().IntTypeWidth() {
 		// The index is bigger than the array length type, so extend it.
 		arrayLen = b.CreateZExt(arrayLen, index.Type(), "")
 	}
@@ -70,27 +64,9 @@ func (b *builder) createSliceBoundsCheck(capacity, low, high, max llvm.Value, lo
 	}

 	// Extend low and high to be the same size as capacity.
-	if low.Type().IntTypeWidth() < capacityType.IntTypeWidth() {
-		if lowType.Info()&types.IsUnsigned != 0 {
-			low = b.CreateZExt(low, capacityType, "")
-		} else {
-			low = b.CreateSExt(low, capacityType, "")
-		}
-	}
-	if high.Type().IntTypeWidth() < capacityType.IntTypeWidth() {
-		if highType.Info()&types.IsUnsigned != 0 {
-			high = b.CreateZExt(high, capacityType, "")
-		} else {
-			high = b.CreateSExt(high, capacityType, "")
-		}
-	}
-	if max.Type().IntTypeWidth() < capacityType.IntTypeWidth() {
-		if maxType.Info()&types.IsUnsigned != 0 {
-			max = b.CreateZExt(max, capacityType, "")
-		} else {
-			max = b.CreateSExt(max, capacityType, "")
-		}
-	}
+	low = b.extendInteger(low, lowType, capacityType)
+	high = b.extendInteger(high, highType, capacityType)
+	max = b.extendInteger(max, maxType, capacityType)

 	// Now do the bounds check: low > high || high > capacity
 	outOfBounds1 := b.CreateICmp(llvm.IntUGT, low, high, "slice.lowhigh")
@@ -101,6 +77,49 @@ func (b *builder) createSliceBoundsCheck(capacity, low, high, max llvm.Value, lo
 	b.createRuntimeAssert(outOfBounds, "slice", "slicePanic")
 }

+// createSliceToArrayPointerCheck adds a check for slice-to-array pointer
+// conversions. This conversion was added in Go 1.17. For details, see:
+// https://tip.golang.org/ref/spec#Conversions_from_slice_to_array_pointer
+func (b *builder) createSliceToArrayPointerCheck(sliceLen llvm.Value, arrayLen int64) {
+	// From the spec:
+	// > If the length of the slice is less than the length of the array, a
+	// > run-time panic occurs.
+	arrayLenValue := llvm.ConstInt(b.uintptrType, uint64(arrayLen), false)
+	isLess := b.CreateICmp(llvm.IntULT, sliceLen, arrayLenValue, "")
+	b.createRuntimeAssert(isLess, "slicetoarray", "sliceToArrayPointerPanic")
+}
+
+// createUnsafeSliceCheck inserts a runtime check used for unsafe.Slice. This
+// function must panic if the ptr/len parameters are invalid.
+func (b *builder) createUnsafeSliceCheck(ptr, len llvm.Value, lenType *types.Basic) {
+	// From the documentation of unsafe.Slice:
+	//   > At run time, if len is negative, or if ptr is nil and len is not
+	//   > zero, a run-time panic occurs.
+	// However, in practice, it is also necessary to check that the length is
+	// not too big that a GEP wouldn't be possible without wrapping the pointer.
+	// These two checks (non-negative and not too big) can be merged into one
+	// using an unsiged greater than.
+
+	// Make sure the len value is at least as big as a uintptr.
+	len = b.extendInteger(len, lenType, b.uintptrType)
+
+	// Determine the maximum slice size, and therefore the maximum value of the
+	// len parameter.
+	maxSize := b.maxSliceSize(ptr.Type().ElementType())
+	maxSizeValue := llvm.ConstInt(len.Type(), maxSize, false)
+
+	// Do the check. By using unsigned greater than for the length check, signed
+	// negative values are also checked (which are very large numbers when
+	// interpreted as signed values).
+	zero := llvm.ConstInt(len.Type(), 0, false)
+	lenOutOfBounds := b.CreateICmp(llvm.IntUGT, len, maxSizeValue, "")
+	ptrIsNil := b.CreateICmp(llvm.IntEQ, ptr, llvm.ConstNull(ptr.Type()), "")
+	lenIsNotZero := b.CreateICmp(llvm.IntNE, len, zero, "")
+	assert := b.CreateAnd(ptrIsNil, lenIsNotZero, "")
+	assert = b.CreateOr(assert, lenOutOfBounds, "")
+	b.createRuntimeAssert(assert, "unsafe.Slice", "unsafeSlicePanic")
+}
+
 // 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) {
@@ -110,19 +129,8 @@ func (b *builder) createChanBoundsCheck(elementSize uint64, bufSize llvm.Value,
 		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, "")
-		}
-	}
+	// Make sure bufSize is at least as big as maxBufSize (an uintptr).
+	bufSize = b.extendInteger(bufSize, bufSizeType, b.uintptrType)

 	// Calculate (^uintptr(0)) >> 1, which is the max value that fits in an
 	// uintptr if uintptrs were signed.
@@ -206,6 +214,19 @@ func (b *builder) createNegativeShiftCheck(shift llvm.Value) {
 	b.createRuntimeAssert(isNegative, "shift", "negativeShiftPanic")
 }

+// createDivideByZeroCheck asserts that y is not zero. If it is, a runtime panic
+// will be emitted. This follows the Go specification which says that a divide
+// by zero must cause a run time panic.
+func (b *builder) createDivideByZeroCheck(y llvm.Value) {
+	if b.info.nobounds {
+		return
+	}
+
+	// isZero = y == 0
+	isZero := b.CreateICmp(llvm.IntEQ, y, llvm.ConstInt(y.Type(), 0, false), "")
+	b.createRuntimeAssert(isZero, "divbyzero", "divideByZeroPanic")
+}
+
 // 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) {
@@ -234,3 +255,19 @@ func (b *builder) createRuntimeAssert(assert llvm.Value, blockPrefix, assertFunc
 	// Ok: assert didn't trigger so continue normally.
 	b.SetInsertPointAtEnd(nextBlock)
 }
+
+// extendInteger extends the value to at least targetType using a zero or sign
+// extend. The resulting value is not truncated: it may still be bigger than
+// targetType.
+func (b *builder) extendInteger(value llvm.Value, valueType types.Type, targetType llvm.Type) llvm.Value {
+	if value.Type().IntTypeWidth() < targetType.IntTypeWidth() {
+		if valueType.Underlying().(*types.Basic).Info()&types.IsUnsigned != 0 {
+			// Unsigned, so zero-extend to the target type.
+			value = b.CreateZExt(value, targetType, "")
+		} else {
+			// Signed, so sign-extend to the target type.
+			value = b.CreateSExt(value, targetType, "")
+		}
+	}
+	return value
+}
@@ -38,11 +38,10 @@ func (b *builder) createAtomicOp(call *ssa.CallCommon) (llvm.Value, bool) {
 		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") {
+			if strings.HasPrefix(b.Triple, "thumb") {
 				// Work around a bug in LLVM, at least LLVM 11:
 				// https://reviews.llvm.org/D95891
-				// Check for armv6m, armv7, armv7em, and perhaps others.
+				// Check for thumbv6m, thumbv7, thumbv7em, 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() {
@@ -63,10 +63,10 @@ func (c *compilerContext) expandFormalParamType(t llvm.Type, name string, goType
 	case llvm.StructTypeKind:
 		fieldInfos := c.flattenAggregateType(t, name, goType)
 		if len(fieldInfos) <= maxFieldsPerParam {
+			// managed to expand this parameter
 			return fieldInfos
-		} else {
-			// failed to lower
 		}
+		// failed to expand this parameter: too many fields
 	}
 	// TODO: split small arrays
 	return []paramInfo{
@@ -32,8 +32,15 @@ func (b *builder) createChanSend(instr *ssa.Send) {

 	// store value-to-send
 	valueType := b.getLLVMType(instr.X.Type())
-	valueAlloca, valueAllocaCast, valueAllocaSize := b.createTemporaryAlloca(valueType, "chan.value")
-	b.CreateStore(chanValue, valueAlloca)
+	isZeroSize := b.targetData.TypeAllocSize(valueType) == 0
+	var valueAlloca, valueAllocaCast, valueAllocaSize llvm.Value
+	if isZeroSize {
+		valueAlloca = llvm.ConstNull(llvm.PointerType(valueType, 0))
+		valueAllocaCast = llvm.ConstNull(b.i8ptrType)
+	} else {
+		valueAlloca, valueAllocaCast, valueAllocaSize = b.createTemporaryAlloca(valueType, "chan.value")
+		b.CreateStore(chanValue, valueAlloca)
+	}

 	// Allocate blockedlist buffer.
 	channelBlockedList := b.mod.GetTypeByName("runtime.channelBlockedList")
@@ -46,7 +53,9 @@ func (b *builder) createChanSend(instr *ssa.Send) {
 	// This also works around a bug in CoroSplit, at least in LLVM 8:
 	// https://bugs.llvm.org/show_bug.cgi?id=41742
 	b.emitLifetimeEnd(channelBlockedListAllocaCast, channelBlockedListAllocaSize)
-	b.emitLifetimeEnd(valueAllocaCast, valueAllocaSize)
+	if !isZeroSize {
+		b.emitLifetimeEnd(valueAllocaCast, valueAllocaSize)
+	}
 }

 // createChanRecv emits a pseudo chan receive operation. It is lowered to the
@@ -56,7 +65,14 @@ func (b *builder) createChanRecv(unop *ssa.UnOp) llvm.Value {
 	ch := b.getValue(unop.X)

 	// Allocate memory to receive into.
-	valueAlloca, valueAllocaCast, valueAllocaSize := b.createTemporaryAlloca(valueType, "chan.value")
+	isZeroSize := b.targetData.TypeAllocSize(valueType) == 0
+	var valueAlloca, valueAllocaCast, valueAllocaSize llvm.Value
+	if isZeroSize {
+		valueAlloca = llvm.ConstNull(llvm.PointerType(valueType, 0))
+		valueAllocaCast = llvm.ConstNull(b.i8ptrType)
+	} else {
+		valueAlloca, valueAllocaCast, valueAllocaSize = b.createTemporaryAlloca(valueType, "chan.value")
+	}

 	// Allocate blockedlist buffer.
 	channelBlockedList := b.mod.GetTypeByName("runtime.channelBlockedList")
@@ -64,9 +80,14 @@ func (b *builder) createChanRecv(unop *ssa.UnOp) llvm.Value {

 	// Do the receive.
 	commaOk := b.createRuntimeCall("chanRecv", []llvm.Value{ch, valueAllocaCast, channelBlockedListAlloca}, "")
-	received := b.CreateLoad(valueAlloca, "chan.received")
+	var received llvm.Value
+	if isZeroSize {
+		received = llvm.ConstNull(valueType)
+	} else {
+		received = b.CreateLoad(valueAlloca, "chan.received")
+		b.emitLifetimeEnd(valueAllocaCast, valueAllocaSize)
+	}
 	b.emitLifetimeEnd(channelBlockedListAllocaCast, channelBlockedListAllocaSize)
-	b.emitLifetimeEnd(valueAllocaCast, valueAllocaSize)

 	if unop.CommaOk {
 		tuple := llvm.Undef(b.ctx.StructType([]llvm.Type{valueType, b.ctx.Int1Type()}, false))
@@ -116,7 +137,6 @@ func (b *builder) createSelect(expr *ssa.Select) llvm.Value {
 	// determine the receive buffer size and alignment.
 	recvbufSize := uint64(0)
 	recvbufAlign := 0
-	hasReceives := false
 	var selectStates []llvm.Value
 	chanSelectStateType := b.getLLVMRuntimeType("chanSelectState")
 	for _, state := range expr.States {
@@ -133,7 +153,6 @@ func (b *builder) createSelect(expr *ssa.Select) llvm.Value {
 			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.
@@ -150,7 +169,7 @@ func (b *builder) createSelect(expr *ssa.Select) llvm.Value {

 	// Create a receive buffer, where the received value will be stored.
 	recvbuf := llvm.Undef(b.i8ptrType)
-	if hasReceives {
+	if recvbufSize != 0 {
 		allocaType := llvm.ArrayType(b.ctx.Int8Type(), int(recvbufSize))
 		recvbufAlloca, _, _ := b.createTemporaryAlloca(allocaType, "select.recvbuf.alloca")
 		recvbufAlloca.SetAlignment(recvbufAlign)
@@ -23,7 +23,7 @@ import (
 // Version of the compiler pacakge. Must be incremented each time the compiler
 // package changes in a way that affects the generated LLVM module.
 // This version is independent of the TinyGo version number.
-const Version = 12 // last change: implement syscall.rawSyscallNoError
+const Version = 25 // last change: add "target-cpu" and "target-features" attributes

 func init() {
 	llvm.InitializeAllTargets()
@@ -33,9 +33,6 @@ func init() {
 	llvm.InitializeAllAsmPrinters()
 }

-// The TinyGo import path.
-const tinygoPath = "github.com/tinygo-org/tinygo"
-
 // Config is the configuration for the compiler. Most settings should be copied
 // directly from compileopts.Config, it recreated here to decouple the compiler
 // package a bit and because it makes caching easier.
@@ -46,11 +43,12 @@ type Config struct {
 	// Target and output information.
 	Triple          string
 	CPU             string
-	Features        []string
+	Features        string
 	GOOS            string
 	GOARCH          string
 	CodeModel       string
 	RelocationModel string
+	SizeLevel       int

 	// Various compiler options that determine how code is generated.
 	Scheduler          string
@@ -59,7 +57,6 @@ type Config struct {
 	DefaultStackSize   uint64
 	NeedsStackObjects  bool
 	Debug              bool // Whether to emit debug information in the LLVM module.
-	LLVMFeatures       string
 }

 // compilerContext contains function-independent data that should still be
@@ -74,15 +71,18 @@ type compilerContext struct {
 	cu               llvm.Metadata
 	difiles          map[string]llvm.Metadata
 	ditypes          map[types.Type]llvm.Metadata
+	llvmTypes        map[types.Type]llvm.Type
 	machine          llvm.TargetMachine
 	targetData       llvm.TargetData
 	intType          llvm.Type
 	i8ptrType        llvm.Type // for convenience
+	rawVoidFuncType  llvm.Type // for convenience
 	funcPtrAddrSpace int
 	uintptrType      llvm.Type
 	program          *ssa.Program
 	diagnostics      []error
 	astComments      map[string]*ast.CommentGroup
+	pkg              *types.Package
 	runtimePkg       *types.Package
 }

@@ -94,6 +94,7 @@ func newCompilerContext(moduleName string, machine llvm.TargetMachine, config *C
 		DumpSSA:     dumpSSA,
 		difiles:     make(map[string]llvm.Metadata),
 		ditypes:     make(map[types.Type]llvm.Metadata),
+		llvmTypes:   make(map[types.Type]llvm.Type),
 		machine:     machine,
 		targetData:  machine.CreateTargetData(),
 		astComments: map[string]*ast.CommentGroup{},
@@ -122,6 +123,7 @@ func newCompilerContext(moduleName string, machine llvm.TargetMachine, config *C
 	dummyFuncType := llvm.FunctionType(c.ctx.VoidType(), nil, false)
 	dummyFunc := llvm.AddFunction(c.mod, "tinygo.dummy", dummyFuncType)
 	c.funcPtrAddrSpace = dummyFunc.Type().PointerAddressSpace()
+	c.rawVoidFuncType = dummyFunc.Type()
 	dummyFunc.EraseFromParentAsFunction()

 	return c
@@ -139,7 +141,6 @@ type builder struct {
 	blockExits        map[*ssa.BasicBlock]llvm.BasicBlock // these are the exit blocks
 	currentBlock      *ssa.BasicBlock
 	phis              []phiNode
-	taskHandle        llvm.Value
 	deferPtr          llvm.Value
 	difunc            llvm.Metadata
 	dilocals          map[*types.Var]llvm.Metadata
@@ -187,12 +188,6 @@ func NewTargetMachine(config *Config) (llvm.TargetMachine, error) {
 		return llvm.TargetMachine{}, err
 	}

-	feat := config.Features
-	if len(config.LLVMFeatures) > 0 {
-		feat = append(feat, config.LLVMFeatures)
-	}
-	features := strings.Join(feat, ",")
-
 	var codeModel llvm.CodeModel
 	var relocationModel llvm.RelocMode

@@ -220,7 +215,7 @@ func NewTargetMachine(config *Config) (llvm.TargetMachine, error) {
 		relocationModel = llvm.RelocDynamicNoPic
 	}

-	machine := target.CreateTargetMachine(config.Triple, config.CPU, features, llvm.CodeGenLevelDefault, relocationModel, codeModel)
+	machine := target.CreateTargetMachine(config.Triple, config.CPU, config.Features, llvm.CodeGenLevelDefault, relocationModel, codeModel)
 	return machine, nil
 }

@@ -231,10 +226,6 @@ func Sizes(machine llvm.TargetMachine) types.Sizes {
 	targetData := machine.CreateTargetData()
 	defer targetData.Dispose()

-	intPtrType := targetData.IntPtrType()
-	if intPtrType.IntTypeWidth()/8 <= 32 {
-	}
-
 	var intWidth int
 	if targetData.PointerSize() <= 4 {
 		// 8, 16, 32 bits targets
@@ -249,13 +240,14 @@ func Sizes(machine llvm.TargetMachine) types.Sizes {
 	return &stdSizes{
 		IntSize:  int64(intWidth / 8),
 		PtrSize:  int64(targetData.PointerSize()),
-		MaxAlign: int64(targetData.PrefTypeAlignment(intPtrType)),
+		MaxAlign: int64(targetData.PrefTypeAlignment(targetData.IntPtrType())),
 	}
 }

 // CompilePackage compiles a single package to a LLVM module.
 func CompilePackage(moduleName string, pkg *loader.Package, ssaPkg *ssa.Package, machine llvm.TargetMachine, config *Config, dumpSSA bool) (llvm.Module, []error) {
 	c := newCompilerContext(moduleName, machine, config, dumpSSA)
+	c.pkg = pkg.Pkg
 	c.runtimePkg = ssaPkg.Prog.ImportedPackage("runtime").Pkg
 	c.program = ssaPkg.Prog

@@ -289,14 +281,14 @@ func CompilePackage(moduleName string, pkg *loader.Package, ssaPkg *ssa.Package,
 	if c.Debug {
 		c.mod.AddNamedMetadataOperand("llvm.module.flags",
 			c.ctx.MDNode([]llvm.Metadata{
-				llvm.ConstInt(c.ctx.Int32Type(), 1, false).ConstantAsMetadata(), // Error on mismatch
+				llvm.ConstInt(c.ctx.Int32Type(), 2, false).ConstantAsMetadata(), // Warning on mismatch
 				c.ctx.MDString("Debug Info Version"),
 				llvm.ConstInt(c.ctx.Int32Type(), 3, false).ConstantAsMetadata(), // DWARF version
 			}),
 		)
 		c.mod.AddNamedMetadataOperand("llvm.module.flags",
 			c.ctx.MDNode([]llvm.Metadata{
-				llvm.ConstInt(c.ctx.Int32Type(), 1, false).ConstantAsMetadata(),
+				llvm.ConstInt(c.ctx.Int32Type(), 7, false).ConstantAsMetadata(), // Max on mismatch
 				c.ctx.MDString("Dwarf Version"),
 				llvm.ConstInt(c.ctx.Int32Type(), 4, false).ConstantAsMetadata(),
 			}),
@@ -315,10 +307,23 @@ func (c *compilerContext) getLLVMRuntimeType(name string) llvm.Type {
 	return c.getLLVMType(typ)
 }

-// getLLVMType creates and returns a LLVM type for a Go type. In the case of
-// named struct types (or Go types implemented as named LLVM structs such as
-// strings) it also creates it first if necessary.
+// getLLVMType returns a LLVM type for a Go type. It doesn't recreate already
+// created types. This is somewhat important for performance, but especially
+// important for named struct types (which should only be created once).
 func (c *compilerContext) getLLVMType(goType types.Type) llvm.Type {
+	// Try to load the LLVM type from the cache.
+	if t, ok := c.llvmTypes[goType]; ok {
+		return t
+	}
+	// Not already created, so adding this type to the cache.
+	llvmType := c.makeLLVMType(goType)
+	c.llvmTypes[goType] = llvmType
+	return llvmType
+}
+
+// makeLLVMType creates a LLVM type for a Go type. Don't call this, use
+// getLLVMType instead.
+func (c *compilerContext) makeLLVMType(goType types.Type) llvm.Type {
 	switch typ := goType.(type) {
 	case *types.Array:
 		elemType := c.getLLVMType(typ.Elem())
@@ -367,12 +372,10 @@ func (c *compilerContext) getLLVMType(goType types.Type) llvm.Type {
 			// LLVM. This is because it is otherwise impossible to create
 			// self-referencing types such as linked lists.
 			llvmName := typ.Obj().Pkg().Path() + "." + typ.Obj().Name()
-			llvmType := c.mod.GetTypeByName(llvmName)
-			if llvmType.IsNil() {
-				llvmType = c.ctx.StructCreateNamed(llvmName)
-				underlying := c.getLLVMType(st)
-				llvmType.StructSetBody(underlying.StructElementTypes(), false)
-			}
+			llvmType := c.ctx.StructCreateNamed(llvmName)
+			c.llvmTypes[goType] = llvmType // avoid infinite recursion
+			underlying := c.getLLVMType(st)
+			llvmType.StructSetBody(underlying.StructElementTypes(), false)
 			return llvmType
 		}
 		return c.getLLVMType(typ.Underlying())
@@ -439,7 +442,7 @@ func (c *compilerContext) createDIType(typ types.Type) llvm.Metadata {
 			AlignInBits: uint32(c.targetData.ABITypeAlignment(llvmType)) * 8,
 			ElementType: c.getDIType(typ.Elem()),
 			Subscripts: []llvm.DISubrange{
-				llvm.DISubrange{
+				{
 					Lo:    0,
 					Count: typ.Len(),
 				},
@@ -768,6 +771,29 @@ func (c *compilerContext) createPackage(irbuilder llvm.Builder, pkg *ssa.Package
 			}
 		}
 	}
+
+	// Add forwarding functions for functions that would otherwise be
+	// implemented in assembly.
+	for _, name := range members {
+		member := pkg.Members[name]
+		switch member := member.(type) {
+		case *ssa.Function:
+			if member.Blocks != nil {
+				continue // external function
+			}
+			info := c.getFunctionInfo(member)
+			if aliasName, ok := stdlibAliases[info.linkName]; ok {
+				alias := c.mod.NamedFunction(aliasName)
+				if alias.IsNil() {
+					// Shouldn't happen, but perhaps best to just ignore.
+					// The error will be a link error, if there is an error.
+					continue
+				}
+				b := newBuilder(c, irbuilder, member)
+				b.createAlias(alias)
+			}
+		}
+	}
 }

 // createFunction builds the LLVM IR implementation for this function. The
@@ -786,6 +812,7 @@ func (b *builder) createFunction() {
 		b.addError(b.fn.Pos(), errValue)
 		return
 	}
+	b.addStandardDefinedAttributes(b.llvmFn)
 	if !b.info.exported {
 		b.llvmFn.SetVisibility(llvm.HiddenVisibility)
 		b.llvmFn.SetUnnamedAddr(true)
@@ -812,6 +839,17 @@ func (b *builder) createFunction() {
 		b.llvmFn.AddFunctionAttr(noinline)
 	}

+	if b.info.interrupt {
+		// Mark this function as an interrupt.
+		// This is necessary on MCUs that don't push caller saved registers when
+		// entering an interrupt, such as on AVR.
+		if strings.HasPrefix(b.Triple, "avr") {
+			b.llvmFn.AddFunctionAttr(b.ctx.CreateStringAttribute("signal", ""))
+		} else {
+			b.addError(b.fn.Pos(), "//go:interrupt not supported on this architecture")
+		}
+	}
+
 	// Add debug info, if needed.
 	if b.Debug {
 		if b.fn.Synthetic == "package initializer" {
@@ -1270,6 +1308,38 @@ func (b *builder) createBuiltin(argTypes []types.Type, argValues []llvm.Value, c
 	case "ssa:wrapnilchk":
 		// TODO: do an actual nil check?
 		return argValues[0], nil
+
+	// Builtins from the unsafe package.
+	case "Add": // unsafe.Add
+		// This is basically just a GEP operation.
+		// Note: the pointer is always of type *i8.
+		ptr := argValues[0]
+		len := argValues[1]
+		return b.CreateGEP(ptr, []llvm.Value{len}, ""), nil
+	case "Slice": // unsafe.Slice
+		// This creates a slice from a pointer and a length.
+		// Note that the exception mentioned in the documentation (if the
+		// pointer and length are nil, the slice is also nil) is trivially
+		// already the case.
+		ptr := argValues[0]
+		len := argValues[1]
+		slice := llvm.Undef(b.ctx.StructType([]llvm.Type{
+			ptr.Type(),
+			b.uintptrType,
+			b.uintptrType,
+		}, false))
+		b.createUnsafeSliceCheck(ptr, len, argTypes[1].Underlying().(*types.Basic))
+		if len.Type().IntTypeWidth() < b.uintptrType.IntTypeWidth() {
+			// Too small, zero-extend len.
+			len = b.CreateZExt(len, b.uintptrType, "")
+		} else if len.Type().IntTypeWidth() > b.uintptrType.IntTypeWidth() {
+			// Too big, truncate len.
+			len = b.CreateTrunc(len, b.uintptrType, "")
+		}
+		slice = b.CreateInsertValue(slice, ptr, 0, "")
+		slice = b.CreateInsertValue(slice, len, 1, "")
+		slice = b.CreateInsertValue(slice, len, 2, "")
+		return slice, nil
 	default:
 		return llvm.Value{}, b.makeError(pos, "todo: builtin: "+callName)
 	}
@@ -1281,9 +1351,9 @@ func (b *builder) createBuiltin(argTypes []types.Type, argValues []llvm.Value, c
 //
 // This is also where compiler intrinsics are implemented.
 func (b *builder) createFunctionCall(instr *ssa.CallCommon) (llvm.Value, error) {
-	if instr.IsInvoke() {
-		fnCast, args := b.getInvokeCall(instr)
-		return b.createCall(fnCast, args, ""), nil
+	var params []llvm.Value
+	for _, param := range instr.Args {
+		params = append(params, b.getValue(param))
 	}

 	// Try to call the function directly for trivially static calls.
@@ -1298,6 +1368,11 @@ func (b *builder) createFunctionCall(instr *ssa.CallCommon) (llvm.Value, error)
 			return b.createMemoryCopyCall(fn, instr.Args)
 		case name == "runtime.memzero":
 			return b.createMemoryZeroCall(instr.Args)
+		case name == "math.Ceil" || name == "math.Floor" || name == "math.Sqrt" || name == "math.Trunc":
+			result, ok := b.createMathOp(instr)
+			if ok {
+				return result, nil
+			}
 		case name == "device.Asm" || name == "device/arm.Asm" || name == "device/arm64.Asm" || name == "device/avr.Asm" || name == "device/riscv.Asm":
 			return b.createInlineAsm(instr.Args)
 		case name == "device.AsmFull" || name == "device/arm.AsmFull" || name == "device/arm64.AsmFull" || name == "device/avr.AsmFull" || name == "device/riscv.AsmFull":
@@ -1349,12 +1424,20 @@ func (b *builder) createFunctionCall(instr *ssa.CallCommon) (llvm.Value, error)
 	} else if call, ok := instr.Value.(*ssa.Builtin); ok {
 		// Builtin function (append, close, delete, etc.).)
 		var argTypes []types.Type
-		var argValues []llvm.Value
 		for _, arg := range instr.Args {
 			argTypes = append(argTypes, arg.Type())
-			argValues = append(argValues, b.getValue(arg))
 		}
-		return b.createBuiltin(argTypes, argValues, call.Name(), instr.Pos())
+		return b.createBuiltin(argTypes, params, call.Name(), instr.Pos())
+	} else if instr.IsInvoke() {
+		// Interface method call (aka invoke call).
+		itf := b.getValue(instr.Value) // interface value (runtime._interface)
+		typecode := b.CreateExtractValue(itf, 0, "invoke.func.typecode")
+		value := b.CreateExtractValue(itf, 1, "invoke.func.value") // receiver
+		// Prefix the params with receiver value and suffix with typecode.
+		params = append([]llvm.Value{value}, params...)
+		params = append(params, typecode)
+		callee = b.getInvokeFunction(instr)
+		context = llvm.Undef(b.i8ptrType)
 	} else {
 		// Function pointer.
 		value := b.getValue(instr.Value)
@@ -1364,11 +1447,6 @@ func (b *builder) createFunctionCall(instr *ssa.CallCommon) (llvm.Value, error)
 		b.createNilCheck(instr.Value, callee, "fpcall")
 	}

-	var params []llvm.Value
-	for _, param := range instr.Args {
-		params = append(params, b.getValue(param))
-	}
-
 	if !exported {
 		// This function takes a context parameter.
 		// Add it to the end of the parameter list.
@@ -1386,7 +1464,7 @@ func (b *builder) createFunctionCall(instr *ssa.CallCommon) (llvm.Value, error)
 func (b *builder) getValue(expr ssa.Value) llvm.Value {
 	switch expr := expr.(type) {
 	case *ssa.Const:
-		return b.createConst(b.info.linkName, expr)
+		return b.createConst(expr)
 	case *ssa.Function:
 		if b.getFunctionInfo(expr).exported {
 			b.addError(expr.Pos(), "cannot use an exported function as value: "+expr.String())
@@ -1411,6 +1489,28 @@ func (b *builder) getValue(expr ssa.Value) llvm.Value {
 	}
 }

+// maxSliceSize determines the maximum size a slice of the given element type
+// can be.
+func (c *compilerContext) maxSliceSize(elementType llvm.Type) uint64 {
+	// Calculate ^uintptr(0), which is the max value that fits in uintptr.
+	maxPointerValue := llvm.ConstNot(llvm.ConstInt(c.uintptrType, 0, false)).ZExtValue()
+	// Calculate (^uint(0))/2, which is the max value that fits in an int.
+	maxIntegerValue := llvm.ConstNot(llvm.ConstInt(c.intType, 0, false)).ZExtValue() / 2
+
+	// Determine the maximum allowed size for a slice. The biggest possible
+	// pointer (starting from 0) would be maxPointerValue*sizeof(elementType) so
+	// divide by the element type to get the real maximum size.
+	maxSize := maxPointerValue / c.targetData.TypeAllocSize(elementType)
+
+	// len(slice) is an int. Make sure the length remains small enough to fit in
+	// an int.
+	if maxSize > maxIntegerValue {
+		maxSize = maxIntegerValue
+	}
+
+	return maxSize
+}
+
 // createExpr translates a Go SSA expression to LLVM IR. This can be zero, one,
 // or multiple LLVM IR instructions and/or runtime calls.
 func (b *builder) createExpr(expr ssa.Value) (llvm.Value, error) {
@@ -1431,7 +1531,8 @@ func (b *builder) createExpr(expr ssa.Value) (llvm.Value, error) {
 				return llvm.Value{}, b.makeError(expr.Pos(), fmt.Sprintf("value is too big (%v bytes)", size))
 			}
 			sizeValue := llvm.ConstInt(b.uintptrType, size, false)
-			buf := b.createRuntimeCall("alloc", []llvm.Value{sizeValue}, expr.Comment)
+			layoutValue := b.createObjectLayout(typ, expr.Pos())
+			buf := b.createRuntimeCall("alloc", []llvm.Value{sizeValue, layoutValue}, expr.Comment)
 			buf = b.CreateBitCast(buf, llvm.PointerType(typ, 0), "")
 			return buf, nil
 		} else {
@@ -1521,10 +1622,14 @@ func (b *builder) createExpr(expr ssa.Value) (llvm.Value, error) {
 		array := b.getValue(expr.X)
 		index := b.getValue(expr.Index)

+		// Extend index to at least uintptr size, because getelementptr assumes
+		// index is a signed integer.
+		index = b.extendInteger(index, expr.Index.Type(), b.uintptrType)
+
 		// Check bounds.
 		arrayLen := expr.X.Type().Underlying().(*types.Array).Len()
 		arrayLenLLVM := llvm.ConstInt(b.uintptrType, uint64(arrayLen), false)
-		b.createLookupBoundsCheck(arrayLenLLVM, index, expr.Index.Type())
+		b.createLookupBoundsCheck(arrayLenLLVM, index)

 		// Can't load directly from array (as index is non-constant), so have to
 		// do it using an alloca+gep+load.
@@ -1565,8 +1670,12 @@ func (b *builder) createExpr(expr ssa.Value) (llvm.Value, error) {
 			return llvm.Value{}, b.makeError(expr.Pos(), "todo: indexaddr: "+ptrTyp.String())
 		}

+		// Make sure index is at least the size of uintptr becuase getelementptr
+		// assumes index is a signed integer.
+		index = b.extendInteger(index, expr.Index.Type(), b.uintptrType)
+
 		// Bounds check.
-		b.createLookupBoundsCheck(buflen, index, expr.Index.Type())
+		b.createLookupBoundsCheck(buflen, index)

 		switch expr.X.Type().Underlying().(type) {
 		case *types.Pointer:
@@ -1590,9 +1699,17 @@ func (b *builder) createExpr(expr ssa.Value) (llvm.Value, error) {
 				panic("lookup on non-string?")
 			}

+			// Sometimes, the index can be e.g. an uint8 or int8, and we have to
+			// correctly extend that type for two reasons:
+			//  1. The lookup bounds check expects an index of at least uintptr
+			//     size.
+			//  2. getelementptr has signed operands, and therefore s[uint8(x)]
+			//     can be lowered as s[int8(x)]. That would be a bug.
+			index = b.extendInteger(index, expr.Index.Type(), b.uintptrType)
+
 			// Bounds check.
 			length := b.CreateExtractValue(value, 1, "len")
-			b.createLookupBoundsCheck(length, index, expr.Index.Type())
+			b.createLookupBoundsCheck(length, index)

 			// Lookup byte
 			buf := b.CreateExtractValue(value, 0, "")
@@ -1624,10 +1741,8 @@ func (b *builder) createExpr(expr ssa.Value) (llvm.Value, error) {
 		elemSize := b.targetData.TypeAllocSize(llvmElemType)
 		elemSizeValue := llvm.ConstInt(b.uintptrType, elemSize, false)

-		// Calculate (^uintptr(0)) >> 1, which is the max value that fits in
-		// uintptr if uintptr were signed.
-		maxSize := llvm.ConstLShr(llvm.ConstNot(llvm.ConstInt(b.uintptrType, 0, false)), llvm.ConstInt(b.uintptrType, 1, false))
-		if elemSize > maxSize.ZExtValue() {
+		maxSize := b.maxSliceSize(llvmElemType)
+		if elemSize > maxSize {
 			// This seems to be checked by the typechecker already, but let's
 			// check it again just to be sure.
 			return llvm.Value{}, b.makeError(expr.Pos(), fmt.Sprintf("slice element type is too big (%v bytes)", elemSize))
@@ -1636,7 +1751,8 @@ func (b *builder) createExpr(expr ssa.Value) (llvm.Value, error) {
 		// Bounds checking.
 		lenType := expr.Len.Type().Underlying().(*types.Basic)
 		capType := expr.Cap.Type().Underlying().(*types.Basic)
-		b.createSliceBoundsCheck(maxSize, sliceLen, sliceCap, sliceCap, lenType, capType, capType)
+		maxSizeValue := llvm.ConstInt(b.uintptrType, maxSize, false)
+		b.createSliceBoundsCheck(maxSizeValue, sliceLen, sliceCap, sliceCap, lenType, capType, capType)

 		// Allocate the backing array.
 		sliceCapCast, err := b.createConvert(expr.Cap.Type(), types.Typ[types.Uintptr], sliceCap, expr.Pos())
@@ -1644,7 +1760,8 @@ func (b *builder) createExpr(expr ssa.Value) (llvm.Value, error) {
 			return llvm.Value{}, err
 		}
 		sliceSize := b.CreateBinOp(llvm.Mul, elemSizeValue, sliceCapCast, "makeslice.cap")
-		slicePtr := b.createRuntimeCall("alloc", []llvm.Value{sliceSize}, "makeslice.buf")
+		layoutValue := b.createObjectLayout(llvmElemType, expr.Pos())
+		slicePtr := b.createRuntimeCall("alloc", []llvm.Value{sliceSize, layoutValue}, "makeslice.buf")
 		slicePtr = b.CreateBitCast(slicePtr, llvm.PointerType(llvmElemType, 0), "makeslice.array")

 		// Extend or truncate if necessary. This is safe as we've already done
@@ -1718,13 +1835,7 @@ func (b *builder) createExpr(expr ssa.Value) (llvm.Value, error) {
 		if expr.Low != nil {
 			lowType = expr.Low.Type().Underlying().(*types.Basic)
 			low = b.getValue(expr.Low)
-			if low.Type().IntTypeWidth() < b.uintptrType.IntTypeWidth() {
-				if lowType.Info()&types.IsUnsigned != 0 {
-					low = b.CreateZExt(low, b.uintptrType, "")
-				} else {
-					low = b.CreateSExt(low, b.uintptrType, "")
-				}
-			}
+			low = b.extendInteger(low, lowType, b.uintptrType)
 		} else {
 			lowType = types.Typ[types.Uintptr]
 			low = llvm.ConstInt(b.uintptrType, 0, false)
@@ -1733,13 +1844,7 @@ func (b *builder) createExpr(expr ssa.Value) (llvm.Value, error) {
 		if expr.High != nil {
 			highType = expr.High.Type().Underlying().(*types.Basic)
 			high = b.getValue(expr.High)
-			if high.Type().IntTypeWidth() < b.uintptrType.IntTypeWidth() {
-				if highType.Info()&types.IsUnsigned != 0 {
-					high = b.CreateZExt(high, b.uintptrType, "")
-				} else {
-					high = b.CreateSExt(high, b.uintptrType, "")
-				}
-			}
+			high = b.extendInteger(high, highType, b.uintptrType)
 		} else {
 			highType = types.Typ[types.Uintptr]
 		}
@@ -1747,13 +1852,7 @@ func (b *builder) createExpr(expr ssa.Value) (llvm.Value, error) {
 		if expr.Max != nil {
 			maxType = expr.Max.Type().Underlying().(*types.Basic)
 			max = b.getValue(expr.Max)
-			if max.Type().IntTypeWidth() < b.uintptrType.IntTypeWidth() {
-				if maxType.Info()&types.IsUnsigned != 0 {
-					max = b.CreateZExt(max, b.uintptrType, "")
-				} else {
-					max = b.CreateSExt(max, b.uintptrType, "")
-				}
-			}
+			max = b.extendInteger(max, maxType, b.uintptrType)
 		} else {
 			maxType = types.Typ[types.Uintptr]
 		}
@@ -1879,6 +1978,17 @@ func (b *builder) createExpr(expr ssa.Value) (llvm.Value, error) {
 		default:
 			return llvm.Value{}, b.makeError(expr.Pos(), "unknown slice type: "+typ.String())
 		}
+	case *ssa.SliceToArrayPointer:
+		// Conversion from a slice to an array pointer, as the name clearly
+		// says. This requires a runtime check to make sure the slice is at
+		// least as big as the array.
+		slice := b.getValue(expr.X)
+		sliceLen := b.CreateExtractValue(slice, 1, "")
+		arrayLen := expr.Type().Underlying().(*types.Pointer).Elem().Underlying().(*types.Array).Len()
+		b.createSliceToArrayPointerCheck(sliceLen, arrayLen)
+		ptr := b.CreateExtractValue(slice, 0, "")
+		ptr = b.CreateBitCast(ptr, b.getLLVMType(expr.Type()), "")
+		return ptr, nil
 	case *ssa.TypeAssert:
 		return b.createTypeAssert(expr), nil
 	case *ssa.UnOp:
@@ -1907,17 +2017,58 @@ func (b *builder) createBinOp(op token.Token, typ, ytyp types.Type, x, y llvm.Va
 				return b.CreateSub(x, y, ""), nil
 			case token.MUL: // *
 				return b.CreateMul(x, y, ""), nil
-			case token.QUO: // /
+			case token.QUO, token.REM: // /, %
+				// Check for a divide by zero. If y is zero, the Go
+				// specification says that a runtime error must be triggered.
+				b.createDivideByZeroCheck(y)
+
 				if signed {
-					return b.CreateSDiv(x, y, ""), nil
+					// Deal with signed division overflow.
+					// The LLVM LangRef says:
+					//
+					//   Overflow also leads to undefined behavior; this is a
+					//   rare case, but can occur, for example, by doing a
+					//   32-bit division of -2147483648 by -1.
+					//
+					// The Go specification however says this about division:
+					//
+					//   The one exception to this rule is that if the dividend
+					//   x is the most negative value for the int type of x, the
+					//   quotient q = x / -1 is equal to x (and r = 0) due to
+					//   two's-complement integer overflow.
+					//
+					// In other words, in the special case that the lowest
+					// possible signed integer is divided by -1, the result of
+					// the division is the same as x (the dividend).
+					// This is implemented by checking for this condition and
+					// changing y to 1 if it occurs, for example for 32-bit
+					// ints:
+					//
+					//   if x == -2147483648 && y == -1 {
+					//       y = 1
+					//   }
+					//
+					// Dividing x by 1 obviously returns x, therefore satisfying
+					// the Go specification without a branch.
+					llvmType := x.Type()
+					minusOne := llvm.ConstSub(llvm.ConstInt(llvmType, 0, false), llvm.ConstInt(llvmType, 1, false))
+					lowestInteger := llvm.ConstInt(x.Type(), 1<<(llvmType.IntTypeWidth()-1), false)
+					yIsMinusOne := b.CreateICmp(llvm.IntEQ, y, minusOne, "")
+					xIsLowestInteger := b.CreateICmp(llvm.IntEQ, x, lowestInteger, "")
+					hasOverflow := b.CreateAnd(yIsMinusOne, xIsLowestInteger, "")
+					y = b.CreateSelect(hasOverflow, llvm.ConstInt(llvmType, 1, true), y, "")
+
+					if op == token.QUO {
+						return b.CreateSDiv(x, y, ""), nil
+					} else {
+						return b.CreateSRem(x, y, ""), nil
+					}
 				} else {
-					return b.CreateUDiv(x, y, ""), nil
-				}
-			case token.REM: // %
-				if signed {
-					return b.CreateSRem(x, y, ""), nil
-				} else {
-					return b.CreateURem(x, y, ""), nil
+					if op == token.QUO {
+						return b.CreateUDiv(x, y, ""), nil
+					} else {
+						return b.CreateURem(x, y, ""), nil
+					}
 				}
 			case token.AND: // &
 				return b.CreateAnd(x, y, ""), nil
@@ -2264,7 +2415,7 @@ func (b *builder) createBinOp(op token.Token, typ, ytyp types.Type, x, y llvm.Va
 }

 // createConst creates a LLVM constant value from a Go constant.
-func (b *builder) createConst(prefix string, expr *ssa.Const) llvm.Value {
+func (b *builder) createConst(expr *ssa.Const) llvm.Value {
 	switch typ := expr.Type().Underlying().(type) {
 	case *types.Basic:
 		llvmType := b.getLLVMType(typ)
@@ -2280,7 +2431,7 @@ func (b *builder) createConst(prefix string, expr *ssa.Const) llvm.Value {
 			strLen := llvm.ConstInt(b.uintptrType, uint64(len(str)), false)
 			var strPtr llvm.Value
 			if str != "" {
-				objname := prefix + "$string"
+				objname := b.pkg.Path() + "$string"
 				global := llvm.AddGlobal(b.mod, llvm.ArrayType(b.ctx.Int8Type(), len(str)), objname)
 				global.SetInitializer(b.ctx.ConstString(str, false))
 				global.SetLinkage(llvm.InternalLinkage)
@@ -2310,15 +2461,15 @@ func (b *builder) createConst(prefix string, expr *ssa.Const) llvm.Value {
 			n, _ := constant.Float64Val(expr.Value)
 			return llvm.ConstFloat(llvmType, n)
 		} else if typ.Kind() == types.Complex64 {
-			r := b.createConst(prefix, ssa.NewConst(constant.Real(expr.Value), types.Typ[types.Float32]))
-			i := b.createConst(prefix, ssa.NewConst(constant.Imag(expr.Value), types.Typ[types.Float32]))
+			r := b.createConst(ssa.NewConst(constant.Real(expr.Value), types.Typ[types.Float32]))
+			i := b.createConst(ssa.NewConst(constant.Imag(expr.Value), types.Typ[types.Float32]))
 			cplx := llvm.Undef(b.ctx.StructType([]llvm.Type{b.ctx.FloatType(), b.ctx.FloatType()}, false))
 			cplx = b.CreateInsertValue(cplx, r, 0, "")
 			cplx = b.CreateInsertValue(cplx, i, 1, "")
 			return cplx
 		} else if typ.Kind() == types.Complex128 {
-			r := b.createConst(prefix, ssa.NewConst(constant.Real(expr.Value), types.Typ[types.Float64]))
-			i := b.createConst(prefix, ssa.NewConst(constant.Imag(expr.Value), types.Typ[types.Float64]))
+			r := b.createConst(ssa.NewConst(constant.Real(expr.Value), types.Typ[types.Float64]))
+			i := b.createConst(ssa.NewConst(constant.Imag(expr.Value), types.Typ[types.Float64]))
 			cplx := llvm.Undef(b.ctx.StructType([]llvm.Type{b.ctx.DoubleType(), b.ctx.DoubleType()}, false))
 			cplx = b.CreateInsertValue(cplx, r, 0, "")
 			cplx = b.CreateInsertValue(cplx, i, 1, "")
@@ -9,6 +9,7 @@ import (
 	"testing"

 	"github.com/tinygo-org/tinygo/compileopts"
+	"github.com/tinygo-org/tinygo/goenv"
 	"github.com/tinygo-org/tinygo/loader"
 	"tinygo.org/x/go-llvm"
 )
@@ -16,6 +17,12 @@ import (
 // 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")

+type testCase struct {
+	file      string
+	target    string
+	scheduler string
+}
+
 // 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) {
@@ -34,20 +41,30 @@ func TestCompiler(t *testing.T) {
 		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"},
+	tests := []testCase{
+		{"basic.go", "", ""},
+		{"pointer.go", "", ""},
+		{"slice.go", "", ""},
+		{"string.go", "", ""},
+		{"float.go", "", ""},
+		{"interface.go", "", ""},
+		{"func.go", "", "coroutines"},
+		{"pragma.go", "", ""},
+		{"goroutine.go", "wasm", "asyncify"},
+		{"goroutine.go", "wasm", "coroutines"},
+		{"goroutine.go", "cortex-m-qemu", "tasks"},
+		{"channel.go", "", ""},
+		{"intrinsics.go", "cortex-m-qemu", ""},
+		{"intrinsics.go", "wasm", ""},
+		{"gc.go", "", ""},
+	}
+
+	_, minor, err := goenv.GetGorootVersion(goenv.Get("GOROOT"))
+	if err != nil {
+		t.Fatal("could not read Go version:", err)
+	}
+	if minor >= 17 {
+		tests = append(tests, testCase{"go1.17.go", "", ""})
 	}

 	for _, tc := range tests {
@@ -55,16 +72,25 @@ func TestCompiler(t *testing.T) {
 		targetString := "wasm"
 		if tc.target != "" {
 			targetString = tc.target
-			name = tc.file + "-" + tc.target
+			name += "-" + tc.target
+		}
+		if tc.scheduler != "" {
+			name += "-" + tc.scheduler
 		}

 		t.Run(name, func(t *testing.T) {
-			target, err := compileopts.LoadTarget(targetString)
+			options := &compileopts.Options{
+				Target: targetString,
+			}
+			target, err := compileopts.LoadTarget(options)
 			if err != nil {
 				t.Fatal("failed to load target:", err)
 			}
+			if tc.scheduler != "" {
+				options.Scheduler = tc.scheduler
+			}
 			config := &compileopts.Config{
-				Options: &compileopts.Options{},
+				Options: options,
 				Target:  target,
 			}
 			compilerConfig := &Config{
@@ -76,6 +102,7 @@ func TestCompiler(t *testing.T) {
 				Scheduler:          config.Scheduler(),
 				FuncImplementation: config.FuncImplementation(),
 				AutomaticStackSize: config.AutomaticStackSize(),
+				DefaultStackSize:   config.Target.DefaultStackSize,
 			}
 			machine, err := NewTargetMachine(compilerConfig)
 			if err != nil {
@@ -124,6 +151,9 @@ func TestCompiler(t *testing.T) {
 			if tc.target != "" {
 				outFilePrefix += "-" + tc.target
 			}
+			if tc.scheduler != "" {
+				outFilePrefix += "-" + tc.scheduler
+			}
 			outPath := "./testdata/" + outFilePrefix + ".ll"

 			// Update test if needed. Do not check the result.
@@ -217,7 +217,8 @@ func (b *builder) createDefer(instr *ssa.Defer) {
 		// 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")
+		nilPtr := llvm.ConstNull(b.i8ptrType)
+		allocCall := b.createRuntimeCall("alloc", []llvm.Value{sizeValue, nilPtr}, "defer.alloc.call")
 		alloca = b.CreateBitCast(allocCall, llvm.PointerType(deferFrameType, 0), "defer.alloc")
 	}
 	if b.NeedsStackObjects {
@@ -331,10 +332,10 @@ func (b *builder) createRunDefers() {
 				//Pass context
 				forwardParams = append(forwardParams, context)
 			} else {
-				// Isolate the typecode.
-				typecode := forwardParams[0]
-				forwardParams = forwardParams[1:]
-				fnPtr = b.getInvokePtr(callback, typecode)
+				// Move typecode from the start to the end of the list of
+				// parameters.
+				forwardParams = append(forwardParams[1:], forwardParams[0])
+				fnPtr = b.getInvokeFunction(callback)

 				// Add the context parameter. An interface call cannot also be a
 				// closure but we have to supply the parameter anyway for platforms
@@ -3,7 +3,6 @@ package compiler
 // This file contains some utility functions related to error handling.

 import (
-	"go/scanner"
 	"go/token"
 	"go/types"
 	"path/filepath"
@@ -20,20 +19,11 @@ func (c *compilerContext) makeError(pos token.Pos, msg string) types.Error {
 	}
 }

+// addError adds a new compiler diagnostic with the given location and message.
 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 {
@@ -23,7 +23,7 @@ func (c *compilerContext) createFuncValue(builder llvm.Builder, funcPtr, context
 	switch c.FuncImplementation {
 	case "doubleword":
 		// Closure is: {context, function pointer}
-		funcValueScalar = funcPtr
+		funcValueScalar = llvm.ConstBitCast(funcPtr, c.rawVoidFuncType)
 	case "switch":
 		funcValueWithSignatureGlobalName := funcPtr.Name() + "$withSignature"
 		funcValueWithSignatureGlobal := c.mod.NamedGlobal(funcValueWithSignatureGlobalName)
@@ -80,8 +80,21 @@ func (b *builder) decodeFuncValue(funcValue llvm.Value, sig *types.Signature) (f
 	context = b.CreateExtractValue(funcValue, 0, "")
 	switch b.FuncImplementation {
 	case "doubleword":
-		funcPtr = b.CreateExtractValue(funcValue, 1, "")
+		bitcast := b.CreateExtractValue(funcValue, 1, "")
+		if !bitcast.IsAConstantExpr().IsNil() && bitcast.Opcode() == llvm.BitCast {
+			funcPtr = bitcast.Operand(0)
+			return
+		}
+		llvmSig := b.getRawFuncType(sig)
+		funcPtr = b.CreateBitCast(bitcast, llvmSig, "")
 	case "switch":
+		if !funcValue.IsAConstant().IsNil() {
+			// If this is a constant func value, the underlying function is
+			// known and can be returned directly.
+			funcValueWithSignatureGlobal := llvm.ConstExtractValue(funcValue, []uint32{1}).Operand(0)
+			funcPtr = llvm.ConstExtractValue(funcValueWithSignatureGlobal.Initializer(), []uint32{0}).Operand(0)
+			return
+		}
 		llvmSig := b.getRawFuncType(sig)
 		sigGlobal := b.getFuncSignatureID(sig)
 		funcPtr = b.createRuntimeCall("getFuncPtr", []llvm.Value{funcValue, sigGlobal}, "")
@@ -96,8 +109,7 @@ func (b *builder) decodeFuncValue(funcValue llvm.Value, sig *types.Signature) (f
 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)
+		return c.ctx.StructType([]llvm.Type{c.i8ptrType, c.rawVoidFuncType}, false)
 	case "switch":
 		return c.getLLVMRuntimeType("funcValue")
 	default:
@@ -79,7 +79,15 @@ func (b *builder) createGo(instr *ssa.Go) {
 		}
 		b.createBuiltin(argTypes, argValues, builtin.Name(), instr.Pos())
 		return
-	} else if !instr.Call.IsInvoke() {
+	} else if instr.Call.IsInvoke() {
+		// This is a method call on an interface value.
+		itf := b.getValue(instr.Call.Value)
+		itfTypeCode := b.CreateExtractValue(itf, 0, "")
+		itfValue := b.CreateExtractValue(itf, 1, "")
+		funcPtr = b.getInvokeFunction(&instr.Call)
+		params = append([]llvm.Value{itfValue}, params...) // start with receiver
+		params = append(params, itfTypeCode)               // end with typecode
+	} else {
 		// This is a function pointer.
 		// At the moment, two extra params are passed to the newly started
 		// goroutine:
@@ -92,22 +100,19 @@ func (b *builder) createGo(instr *ssa.Go) {
 		switch b.Scheduler {
 		case "none", "coroutines":
 			// There are no additional parameters needed for the goroutine start operation.
-		case "tasks":
+		case "tasks", "asyncify":
 			// 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":
+	case "none", "tasks", "asyncify":
 		callee = b.createGoroutineStartWrapper(funcPtr, prefix, hasContext, instr.Pos())
 		if b.AutomaticStackSize {
 			// The stack size is not known until after linking. Call a dummy
@@ -119,7 +124,7 @@ func (b *builder) createGo(instr *ssa.Go) {
 		} 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 {
+			if (b.Scheduler == "tasks" || b.Scheduler == "asyncify") && b.DefaultStackSize == 0 {
 				b.addError(instr.Pos(), "default stack size for goroutines is not set")
 			}
 			stackSize = llvm.ConstInt(b.uintptrType, b.DefaultStackSize, false)
@@ -165,6 +170,11 @@ func (c *compilerContext) createGoroutineStartWrapper(fn llvm.Value, prefix stri
 	builder := c.ctx.NewBuilder()
 	defer builder.Dispose()

+	var deadlock llvm.Value
+	if c.Scheduler == "asyncify" {
+		deadlock = c.getFunction(c.program.ImportedPackage("runtime").Members["deadlock"].(*ssa.Function))
+	}
+
 	if !fn.IsAFunction().IsNil() {
 		// See whether this wrapper has already been created. If so, return it.
 		name := fn.Name()
@@ -176,6 +186,7 @@ func (c *compilerContext) createGoroutineStartWrapper(fn llvm.Value, prefix stri
 		// Create the wrapper.
 		wrapperType := llvm.FunctionType(c.ctx.VoidType(), []llvm.Type{c.i8ptrType}, false)
 		wrapper = llvm.AddFunction(c.mod, name+"$gowrapper", wrapperType)
+		c.addStandardAttributes(wrapper)
 		wrapper.SetLinkage(llvm.LinkOnceODRLinkage)
 		wrapper.SetUnnamedAddr(true)
 		wrapper.AddAttributeAtIndex(-1, c.ctx.CreateStringAttribute("tinygo-gowrapper", name))
@@ -219,6 +230,12 @@ func (c *compilerContext) createGoroutineStartWrapper(fn llvm.Value, prefix stri
 		// Create the call.
 		builder.CreateCall(fn, params, "")

+		if c.Scheduler == "asyncify" {
+			builder.CreateCall(deadlock, []llvm.Value{
+				llvm.Undef(c.i8ptrType), llvm.Undef(c.i8ptrType),
+			}, "")
+		}
+
 	} else {
 		// For a function pointer like this:
 		//
@@ -240,6 +257,7 @@ func (c *compilerContext) createGoroutineStartWrapper(fn llvm.Value, prefix stri
 		// Create the wrapper.
 		wrapperType := llvm.FunctionType(c.ctx.VoidType(), []llvm.Type{c.i8ptrType}, false)
 		wrapper = llvm.AddFunction(c.mod, prefix+".gowrapper", wrapperType)
+		c.addStandardAttributes(wrapper)
 		wrapper.SetLinkage(llvm.LinkOnceODRLinkage)
 		wrapper.SetUnnamedAddr(true)
 		wrapper.AddAttributeAtIndex(-1, c.ctx.CreateStringAttribute("tinygo-gowrapper", ""))
@@ -285,11 +303,22 @@ func (c *compilerContext) createGoroutineStartWrapper(fn llvm.Value, prefix stri

 		// Create the call.
 		builder.CreateCall(fnPtr, params, "")
+
+		if c.Scheduler == "asyncify" {
+			builder.CreateCall(deadlock, []llvm.Value{
+				llvm.Undef(c.i8ptrType), llvm.Undef(c.i8ptrType),
+			}, "")
+		}
 	}

-	// Finish the function. Every basic block must end in a terminator, and
-	// because goroutines never return a value we can simply return void.
-	builder.CreateRetVoid()
+	if c.Scheduler == "asyncify" {
+		// The goroutine was terminated via deadlock.
+		builder.CreateUnreachable()
+	} else {
+		// Finish the function. Every basic block must end in a terminator, and
+		// because goroutines never return a value we can simply return void.
+		builder.CreateRetVoid()
+	}

 	// Return a ptrtoint of the wrapper, not the function itself.
 	return builder.CreatePtrToInt(wrapper, c.uintptrType, "")
@@ -72,7 +72,7 @@ func (b *builder) createInlineAsmFull(instr *ssa.CallCommon) (llvm.Value, error)
 	args := []llvm.Value{}
 	constraints := []string{}
 	hasOutput := false
-	asmString = regexp.MustCompile("\\{\\}").ReplaceAllStringFunc(asmString, func(s string) string {
+	asmString = regexp.MustCompile(`\{\}`).ReplaceAllStringFunc(asmString, func(s string) string {
 		hasOutput = true
 		return "$0"
 	})
@@ -80,7 +80,7 @@ func (b *builder) createInlineAsmFull(instr *ssa.CallCommon) (llvm.Value, error)
 		constraints = append(constraints, "=&r")
 		registerNumbers[""] = 0
 	}
-	asmString = regexp.MustCompile("\\{[a-zA-Z]+\\}").ReplaceAllStringFunc(asmString, func(s string) string {
+	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]
@@ -47,6 +47,7 @@ func (c *compilerContext) getTypeCode(typ types.Type) llvm.Value {
 		var length int64
 		var methodSet llvm.Value
 		var ptrTo llvm.Value
+		var typeAssert llvm.Value
 		switch typ := typ.(type) {
 		case *types.Named:
 			references = c.getTypeCode(typ.Underlying())
@@ -69,6 +70,9 @@ func (c *compilerContext) getTypeCode(typ types.Type) llvm.Value {
 		}
 		if _, ok := typ.Underlying().(*types.Interface); !ok {
 			methodSet = c.getTypeMethodSet(typ)
+		} else {
+			typeAssert = c.getInterfaceImplementsFunc(typ)
+			typeAssert = llvm.ConstPtrToInt(typeAssert, c.uintptrType)
 		}
 		if _, ok := typ.Underlying().(*types.Pointer); !ok {
 			ptrTo = c.getTypeCode(types.NewPointer(typ))
@@ -87,6 +91,9 @@ func (c *compilerContext) getTypeCode(typ types.Type) llvm.Value {
 		if !ptrTo.IsNil() {
 			globalValue = llvm.ConstInsertValue(globalValue, ptrTo, []uint32{3})
 		}
+		if !typeAssert.IsNil() {
+			globalValue = llvm.ConstInsertValue(globalValue, typeAssert, []uint32{4})
+		}
 		global.SetInitializer(globalValue)
 		global.SetLinkage(llvm.LinkOnceODRLinkage)
 		global.SetGlobalConstant(true)
@@ -193,7 +200,11 @@ func getTypeCodeName(t types.Type) string {
 	case *types.Interface:
 		methods := make([]string, t.NumMethods())
 		for i := 0; i < t.NumMethods(); i++ {
-			methods[i] = t.Method(i).Name() + ":" + getTypeCodeName(t.Method(i).Type())
+			name := t.Method(i).Name()
+			if !token.IsExported(name) {
+				name = t.Method(i).Pkg().Path() + "." + name
+			}
+			methods[i] = name + ":" + getTypeCodeName(t.Method(i).Type())
 		}
 		return "interface:" + "{" + strings.Join(methods, ",") + "}"
 	case *types.Map:
@@ -306,10 +317,9 @@ func (c *compilerContext) getInterfaceMethodSet(typ types.Type) llvm.Value {
 	return llvm.ConstGEP(global, []llvm.Value{zero, zero})
 }

-// getMethodSignature returns a global variable which is a reference to an
-// external *i8 indicating the indicating the signature of this method. It is
-// used during the interface lowering pass.
-func (c *compilerContext) getMethodSignature(method *types.Func) llvm.Value {
+// getMethodSignatureName returns a unique name (that can be used as the name of
+// a global) for the given method.
+func (c *compilerContext) getMethodSignatureName(method *types.Func) string {
 	signature := methodSignature(method)
 	var globalName string
 	if token.IsExported(method.Name()) {
@@ -317,6 +327,14 @@ func (c *compilerContext) getMethodSignature(method *types.Func) llvm.Value {
 	} else {
 		globalName = method.Type().(*types.Signature).Recv().Pkg().Path() + ".$methods." + signature
 	}
+	return globalName
+}
+
+// getMethodSignature returns a global variable which is a reference to an
+// external *i8 indicating the indicating the signature of this method. It is
+// used during the interface lowering pass.
+func (c *compilerContext) getMethodSignature(method *types.Func) llvm.Value {
+	globalName := c.getMethodSignatureName(method)
 	signatureGlobal := c.mod.NamedGlobal(globalName)
 	if signatureGlobal.IsNil() {
 		// TODO: put something useful in these globals, such as the method
@@ -345,15 +363,16 @@ func (b *builder) createTypeAssert(expr *ssa.TypeAssert) llvm.Value {
 	commaOk := llvm.Value{}
 	if _, ok := expr.AssertedType.Underlying().(*types.Interface); ok {
 		// Type assert on interface type.
-		// This pseudo call will be lowered in the interface lowering pass to a
-		// real call which checks whether the provided typecode is any of the
-		// concrete types that implements this interface.
+		// This is a call to an interface type assert function.
+		// The interface lowering pass will define this function by filling it
+		// with a type switch over all concrete types that implement this
+		// interface, and returning whether it's one of the matched types.
 		// This is very different from how interface asserts are implemented in
 		// the main Go compiler, where the runtime checks whether the type
 		// implements each method of the interface. See:
 		// https://research.swtch.com/interfaces
-		methodSet := b.getInterfaceMethodSet(expr.AssertedType)
-		commaOk = b.createRuntimeCall("interfaceImplements", []llvm.Value{actualTypeNum, methodSet}, "")
+		fn := b.getInterfaceImplementsFunc(expr.AssertedType)
+		commaOk = b.CreateCall(fn, []llvm.Value{actualTypeNum}, "")

 	} else {
 		globalName := "reflect/types.typeid:" + getTypeCodeName(expr.AssertedType)
@@ -420,39 +439,52 @@ func (b *builder) createTypeAssert(expr *ssa.TypeAssert) llvm.Value {
 	}
 }

-// 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,
-		b.getInterfaceMethodSet(instr.Value.Type()),
-		b.getMethodSignature(instr.Method),
+// getMethodsString returns a string to be used in the "tinygo-methods" string
+// attribute for interface functions.
+func (c *compilerContext) getMethodsString(itf *types.Interface) string {
+	methods := make([]string, itf.NumMethods())
+	for i := range methods {
+		methods[i] = c.getMethodSignatureName(itf.Method(i))
 	}
-	fn := b.createRuntimeCall("interfaceMethod", values, "invoke.func")
-	return b.CreateIntToPtr(fn, llvmFnType, "invoke.func.cast")
+	return strings.Join(methods, "; ")
 }

-// 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, b.getValue(arg))
+// getInterfaceImplementsfunc returns a declared function that works as a type
+// switch. The interface lowering pass will define this function.
+func (c *compilerContext) getInterfaceImplementsFunc(assertedType types.Type) llvm.Value {
+	fnName := getTypeCodeName(assertedType.Underlying()) + ".$typeassert"
+	llvmFn := c.mod.NamedFunction(fnName)
+	if llvmFn.IsNil() {
+		llvmFnType := llvm.FunctionType(c.ctx.Int1Type(), []llvm.Type{c.uintptrType}, false)
+		llvmFn = llvm.AddFunction(c.mod, fnName, llvmFnType)
+		c.addStandardDeclaredAttributes(llvmFn)
+		methods := c.getMethodsString(assertedType.Underlying().(*types.Interface))
+		llvmFn.AddFunctionAttr(c.ctx.CreateStringAttribute("tinygo-methods", methods))
 	}
-	// Add the context parameter. An interface call never takes a context but we
-	// have to supply the parameter anyway.
-	args = append(args, llvm.Undef(b.i8ptrType))
-	// Add the parent goroutine handle.
-	args = append(args, llvm.Undef(b.i8ptrType))
+	return llvmFn
+}

-	return fnCast, args
+// getInvokeFunction returns the thunk to call the given interface method. The
+// thunk is declared, not defined: it will be defined by the interface lowering
+// pass.
+func (c *compilerContext) getInvokeFunction(instr *ssa.CallCommon) llvm.Value {
+	fnName := getTypeCodeName(instr.Value.Type().Underlying()) + "." + instr.Method.Name() + "$invoke"
+	llvmFn := c.mod.NamedFunction(fnName)
+	if llvmFn.IsNil() {
+		sig := instr.Method.Type().(*types.Signature)
+		var paramTuple []*types.Var
+		for i := 0; i < sig.Params().Len(); i++ {
+			paramTuple = append(paramTuple, sig.Params().At(i))
+		}
+		paramTuple = append(paramTuple, types.NewVar(token.NoPos, nil, "$typecode", types.Typ[types.Uintptr]))
+		llvmFnType := c.getRawFuncType(types.NewSignature(sig.Recv(), types.NewTuple(paramTuple...), sig.Results(), false)).ElementType()
+		llvmFn = llvm.AddFunction(c.mod, fnName, llvmFnType)
+		c.addStandardDeclaredAttributes(llvmFn)
+		llvmFn.AddFunctionAttr(c.ctx.CreateStringAttribute("tinygo-invoke", c.getMethodSignatureName(instr.Method)))
+		methods := c.getMethodsString(instr.Value.Type().Underlying().(*types.Interface))
+		llvmFn.AddFunctionAttr(c.ctx.CreateStringAttribute("tinygo-methods", methods))
+	}
+	return llvmFn
 }

 // getInterfaceInvokeWrapper returns a wrapper for the given method so it can be
@@ -489,6 +521,7 @@ func (c *compilerContext) getInterfaceInvokeWrapper(fn *ssa.Function, llvmFn llv
 	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.addStandardAttributes(wrapper)
 	wrapper.LastParam().SetName("parentHandle")

 	wrapper.SetLinkage(llvm.LinkOnceODRLinkage)
@@ -36,6 +36,8 @@ func (b *builder) createInterruptGlobal(instr *ssa.CallCommon) (llvm.Value, erro
 		// Fall back to a generic error.
 		return llvm.Value{}, b.makeError(instr.Pos(), "interrupt function must be constant")
 	}
+	funcRawPtr, funcContext := b.decodeFuncValue(funcValue, nil)
+	funcPtr := llvm.ConstPtrToInt(funcRawPtr, b.uintptrType)

 	// Create a new global of type runtime/interrupt.handle. Globals of this
 	// type are lowered in the interrupt lowering pass.
@@ -47,8 +49,9 @@ func (b *builder) createInterruptGlobal(instr *ssa.CallCommon) (llvm.Value, erro
 	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})
+	initializer = llvm.ConstInsertValue(initializer, funcContext, []uint32{0})
+	initializer = llvm.ConstInsertValue(initializer, funcPtr, []uint32{1})
+	initializer = llvm.ConstInsertValue(initializer, llvm.ConstInt(b.intType, uint64(id.Int64()), true), []uint32{2, 0})
 	global.SetInitializer(initializer)

 	// Add debug info to the interrupt global.
@@ -48,3 +48,57 @@ func (b *builder) createMemoryZeroCall(args []ssa.Value) (llvm.Value, error) {
 	b.CreateCall(llvmFn, params, "")
 	return llvm.Value{}, nil
 }
+
+var mathToLLVMMapping = map[string]string{
+	"math.Sqrt":  "llvm.sqrt.f64",
+	"math.Floor": "llvm.floor.f64",
+	"math.Ceil":  "llvm.ceil.f64",
+	"math.Trunc": "llvm.trunc.f64",
+}
+
+// createMathOp tries to lower the given call as a LLVM math intrinsic, if
+// possible. It returns the call result if possible, and a boolean whether it
+// succeeded. If it doesn't succeed, the architecture doesn't support the given
+// intrinsic.
+func (b *builder) createMathOp(call *ssa.CallCommon) (llvm.Value, bool) {
+	// Check whether this intrinsic is supported on the given GOARCH.
+	// If it is unsupported, this can have two reasons:
+	//
+	//  1. LLVM can expand the intrinsic inline (using float instructions), but
+	//     the result doesn't pass the tests of the math package.
+	//  2. LLVM cannot expand the intrinsic inline, will therefore lower it as a
+	//     libm function call, but the libm function call also fails the math
+	//     package tests.
+	//
+	// Whatever the implementation, it must pass the tests in the math package
+	// so unfortunately only the below intrinsic+architecture combinations are
+	// supported.
+	name := call.StaticCallee().RelString(nil)
+	switch name {
+	case "math.Ceil", "math.Floor", "math.Trunc":
+		if b.GOARCH != "wasm" && b.GOARCH != "arm64" {
+			return llvm.Value{}, false
+		}
+	case "math.Sqrt":
+		if b.GOARCH != "wasm" && b.GOARCH != "amd64" && b.GOARCH != "386" {
+			return llvm.Value{}, false
+		}
+	default:
+		return llvm.Value{}, false // only the above functions are supported.
+	}
+
+	llvmFn := b.mod.NamedFunction(mathToLLVMMapping[name])
+	if llvmFn.IsNil() {
+		// The intrinsic doesn't exist yet, so declare it.
+		// At the moment, all supported intrinsics have the form "double
+		// foo(double %x)" so we can hardcode the signature here.
+		llvmType := llvm.FunctionType(b.ctx.DoubleType(), []llvm.Type{b.ctx.DoubleType()}, false)
+		llvmFn = llvm.AddFunction(b.mod, mathToLLVMMapping[name], llvmType)
+	}
+	// Create a call to the intrinsic.
+	args := make([]llvm.Value, len(call.Args))
+	for i, arg := range call.Args {
+		args[i] = b.getValue(arg)
+	}
+	return b.CreateCall(llvmFn, args, ""), true
+}
@@ -1,6 +1,11 @@
 package compiler

 import (
+	"fmt"
+	"go/token"
+	"go/types"
+	"math/big"
+
 	"github.com/tinygo-org/tinygo/compiler/llvmutil"
 	"tinygo.org/x/go-llvm"
 )
@@ -8,21 +13,6 @@ import (
 // This file contains helper functions for LLVM that are not exposed in the Go
 // bindings.

-// Return a list of values (actually, instructions) where this value is used as
-// an operand.
-func getUses(value llvm.Value) []llvm.Value {
-	if value.IsNil() {
-		return nil
-	}
-	var uses []llvm.Value
-	use := value.FirstUse()
-	for !use.IsNil() {
-		uses = append(uses, use.User())
-		use = use.NextUse()
-	}
-	return uses
-}
-
 // createTemporaryAlloca creates a new alloca in the entry block and adds
 // lifetime start information in the IR signalling that the alloca won't be used
 // before this point.
@@ -44,7 +34,7 @@ func (b *builder) emitLifetimeEnd(ptr, size llvm.Value) {
 // 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)
+	return llvmutil.EmitPointerPack(b.Builder, b.mod, b.pkg.Path(), b.NeedsStackObjects, values)
 }

 // emitPointerUnpack extracts a list of values packed using emitPointerPack.
@@ -67,3 +57,199 @@ func (c *compilerContext) makeGlobalArray(buf []byte, name string, elementType l
 	global.SetInitializer(value)
 	return global
 }
+
+// createObjectLayout returns a LLVM value (of type i8*) that describes where
+// there are pointers in the type t. If all the data fits in a word, it is
+// returned as a word. Otherwise it will store the data in a global.
+//
+// The value contains two pieces of information: the length of the object and
+// which words contain a pointer (indicated by setting the given bit to 1). For
+// arrays, only the element is stored. This works because the GC knows the
+// object size and can therefore know how this value is repeated in the object.
+func (c *compilerContext) createObjectLayout(t llvm.Type, pos token.Pos) llvm.Value {
+	// Use the element type for arrays. This works even for nested arrays.
+	for {
+		kind := t.TypeKind()
+		if kind == llvm.ArrayTypeKind {
+			t = t.ElementType()
+			continue
+		}
+		if kind == llvm.StructTypeKind {
+			fields := t.StructElementTypes()
+			if len(fields) == 1 {
+				t = fields[0]
+				continue
+			}
+		}
+		break
+	}
+
+	// Do a few checks to see whether we need to generate any object layout
+	// information at all.
+	objectSizeBytes := c.targetData.TypeAllocSize(t)
+	pointerSize := c.targetData.TypeAllocSize(c.i8ptrType)
+	pointerAlignment := c.targetData.PrefTypeAlignment(c.i8ptrType)
+	if objectSizeBytes < pointerSize {
+		// Too small to contain a pointer.
+		layout := (uint64(1) << 1) | 1
+		return llvm.ConstIntToPtr(llvm.ConstInt(c.uintptrType, layout, false), c.i8ptrType)
+	}
+	bitmap := c.getPointerBitmap(t, pos)
+	if bitmap.BitLen() == 0 {
+		// There are no pointers in this type, so we can simplify the layout.
+		// TODO: this can be done in many other cases, e.g. when allocating an
+		// array (like [4][]byte, which repeats a slice 4 times).
+		layout := (uint64(1) << 1) | 1
+		return llvm.ConstIntToPtr(llvm.ConstInt(c.uintptrType, layout, false), c.i8ptrType)
+	}
+	if objectSizeBytes%uint64(pointerAlignment) != 0 {
+		// This shouldn't happen except for packed structs, which aren't
+		// currently used.
+		c.addError(pos, "internal error: unexpected object size for object with pointer field")
+		return llvm.ConstNull(c.i8ptrType)
+	}
+	objectSizeWords := objectSizeBytes / uint64(pointerAlignment)
+
+	pointerBits := pointerSize * 8
+	var sizeFieldBits uint64
+	switch pointerBits {
+	case 16:
+		sizeFieldBits = 4
+	case 32:
+		sizeFieldBits = 5
+	case 64:
+		sizeFieldBits = 6
+	default:
+		panic("unknown pointer size")
+	}
+	layoutFieldBits := pointerBits - 1 - sizeFieldBits
+
+	// Try to emit the value as an inline integer. This is possible in most
+	// cases.
+	if objectSizeWords < layoutFieldBits {
+		// If it can be stored directly in the pointer value, do so.
+		// The runtime knows that if the least significant bit of the pointer is
+		// set, the pointer contains the value itself.
+		layout := bitmap.Uint64()<<(sizeFieldBits+1) | (objectSizeWords << 1) | 1
+		return llvm.ConstIntToPtr(llvm.ConstInt(c.uintptrType, layout, false), c.i8ptrType)
+	}
+
+	// Unfortunately, the object layout is too big to fit in a pointer-sized
+	// integer. Store it in a global instead.
+
+	// Try first whether the global already exists. All objects with a
+	// particular name have the same type, so this is possible.
+	globalName := "runtime/gc.layout:" + fmt.Sprintf("%d-%0*x", objectSizeWords, (objectSizeWords+15)/16, bitmap)
+	global := c.mod.NamedGlobal(globalName)
+	if !global.IsNil() {
+		return llvm.ConstBitCast(global, c.i8ptrType)
+	}
+
+	// Create the global initializer.
+	bitmapBytes := make([]byte, int(objectSizeWords+7)/8)
+	copy(bitmapBytes, bitmap.Bytes())
+	var bitmapByteValues []llvm.Value
+	for _, b := range bitmapBytes {
+		bitmapByteValues = append(bitmapByteValues, llvm.ConstInt(c.ctx.Int8Type(), uint64(b), false))
+	}
+	initializer := c.ctx.ConstStruct([]llvm.Value{
+		llvm.ConstInt(c.uintptrType, objectSizeWords, false),
+		llvm.ConstArray(c.ctx.Int8Type(), bitmapByteValues),
+	}, false)
+
+	global = llvm.AddGlobal(c.mod, initializer.Type(), globalName)
+	global.SetInitializer(initializer)
+	global.SetUnnamedAddr(true)
+	global.SetGlobalConstant(true)
+	global.SetLinkage(llvm.LinkOnceODRLinkage)
+	if c.targetData.PrefTypeAlignment(c.uintptrType) < 2 {
+		// AVR doesn't have alignment by default.
+		global.SetAlignment(2)
+	}
+	if c.Debug && pos != token.NoPos {
+		// Creating a fake global so that the value can be inspected in GDB.
+		// For example, the layout for strings.stringFinder (as of Go version
+		// 1.15) has the following type according to GDB:
+		//   type = struct {
+		//       uintptr numBits;
+		//       uint8 data[33];
+		//   }
+		// ...that's sort of a mixed C/Go type, but it is readable. More
+		// importantly, these object layout globals can be read and printed by
+		// GDB which may be useful for debugging.
+		position := c.program.Fset.Position(pos)
+		diglobal := c.dibuilder.CreateGlobalVariableExpression(c.difiles[position.Filename], llvm.DIGlobalVariableExpression{
+			Name: globalName,
+			File: c.getDIFile(position.Filename),
+			Line: position.Line,
+			Type: c.getDIType(types.NewStruct([]*types.Var{
+				types.NewVar(pos, nil, "numBits", types.Typ[types.Uintptr]),
+				types.NewVar(pos, nil, "data", types.NewArray(types.Typ[types.Byte], int64(len(bitmapByteValues)))),
+			}, nil)),
+			LocalToUnit: false,
+			Expr:        c.dibuilder.CreateExpression(nil),
+		})
+		global.AddMetadata(0, diglobal)
+	}
+
+	return llvm.ConstBitCast(global, c.i8ptrType)
+}
+
+// getPointerBitmap scans the given LLVM type for pointers and sets bits in a
+// bigint at the word offset that contains a pointer. This scan is recursive.
+func (c *compilerContext) getPointerBitmap(typ llvm.Type, pos token.Pos) *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)
+		if typ.StructName() == "runtime.funcValue" {
+			// Hack: the type runtime.funcValue contains an 'id' field which is
+			// of type uintptr, but before the LowerFuncValues pass it actually
+			// contains a pointer (ptrtoint) to a global. This trips up the
+			// interp package. Therefore, make the id field a pointer for now.
+			typ = c.ctx.StructType([]llvm.Type{c.i8ptrType, c.i8ptrType}, false)
+		}
+		for i, subtyp := range typ.StructElementTypes() {
+			subptrs := c.getPointerBitmap(subtyp, pos)
+			if subptrs.BitLen() == 0 {
+				continue
+			}
+			offset := c.targetData.ElementOffset(typ, i)
+			if offset%uint64(alignment) != 0 {
+				// This error will let the compilation fail, but by continuing
+				// the error can still easily be shown.
+				c.addError(pos, "internal error: allocated struct contains unaligned pointer")
+				continue
+			}
+			subptrs.Lsh(subptrs, uint(offset)/uint(alignment))
+			ptrs.Or(ptrs, subptrs)
+		}
+		return ptrs
+	case llvm.ArrayTypeKind:
+		subtyp := typ.ElementType()
+		subptrs := c.getPointerBitmap(subtyp, pos)
+		ptrs := big.NewInt(0)
+		if subptrs.BitLen() == 0 {
+			return ptrs
+		}
+		elementSize := c.targetData.TypeAllocSize(subtyp)
+		if elementSize%uint64(alignment) != 0 {
+			// This error will let the compilation fail (but continues so that
+			// other errors can be shown).
+			c.addError(pos, "internal error: allocated array contains unaligned pointer")
+			return ptrs
+		}
+		for i := 0; i < typ.ArrayLength(); i++ {
+			ptrs.Lsh(ptrs, uint(elementSize)/uint(alignment))
+			ptrs.Or(ptrs, subptrs)
+		}
+		return ptrs
+	default:
+		// Should not happen.
+		panic("unknown LLVM type")
+	}
+}
@@ -12,7 +12,7 @@ import (
 // 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 {
+func EmitPointerPack(builder llvm.Builder, mod llvm.Module, prefix string, needsStackObjects bool, values []llvm.Value) llvm.Value {
 	ctx := mod.Context()
 	targetData := llvm.NewTargetData(mod.DataLayout())
 	i8ptrType := llvm.PointerType(mod.Context().Int8Type(), 0)
@@ -83,12 +83,11 @@ func EmitPointerPack(builder llvm.Builder, mod llvm.Module, needsStackObjects bo
 		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+"$pack")
+			global := llvm.AddGlobal(mod, packedType, prefix+"$pack")
 			global.SetInitializer(ctx.ConstStruct(values, false))
 			global.SetGlobalConstant(true)
 			global.SetUnnamedAddr(true)
-			global.SetLinkage(llvm.PrivateLinkage)
+			global.SetLinkage(llvm.InternalLinkage)
 			return llvm.ConstBitCast(global, i8ptrType)
 		}

@@ -97,6 +96,7 @@ func EmitPointerPack(builder llvm.Builder, mod llvm.Module, needsStackObjects bo
 		alloc := mod.NamedFunction("runtime.alloc")
 		packedHeapAlloc := builder.CreateCall(alloc, []llvm.Value{
 			sizeValue,
+			llvm.ConstNull(i8ptrType),
 			llvm.Undef(i8ptrType),            // unused context parameter
 			llvm.ConstPointerNull(i8ptrType), // coroutine handle
 		}, "")
@@ -26,6 +26,7 @@ type functionInfo struct {
 	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
+	interrupt  bool       // go:interrupt
 	nobounds   bool       // go:nobounds
 	variadic   bool       // go:variadic (CGo only)
 	inline     inlineType // go:inline
@@ -108,6 +109,7 @@ func (c *compilerContext) getFunction(fn *ssa.Function) llvm.Value {
 			llvmFn.AddFunctionAttr(attr)
 		}
 	}
+	c.addStandardDeclaredAttributes(llvmFn)

 	dereferenceableOrNullKind := llvm.AttributeKindID("dereferenceable_or_null")
 	for i, info := range paramInfos {
@@ -250,6 +252,10 @@ func (info *functionInfo) parsePragmas(f *ssa.Function) {

 				importName = parts[1]
 				info.exported = true
+			case "//go:interrupt":
+				if hasUnsafeImport(f.Pkg.Pkg) {
+					info.interrupt = true
+				}
 			case "//go:wasm-module":
 				// Alternative comment for setting the import module.
 				if len(parts) != 2 {
@@ -323,6 +329,53 @@ func getParams(sig *types.Signature) []*types.Var {
 	return params
 }

+// addStandardDeclaredAttributes adds attributes that are set for any function,
+// whether declared or defined.
+func (c *compilerContext) addStandardDeclaredAttributes(llvmFn llvm.Value) {
+	if c.SizeLevel >= 1 {
+		// Set the "optsize" attribute to make slightly smaller binaries at the
+		// cost of minimal performance loss (-Os in Clang).
+		kind := llvm.AttributeKindID("optsize")
+		attr := c.ctx.CreateEnumAttribute(kind, 0)
+		llvmFn.AddFunctionAttr(attr)
+	}
+	if c.SizeLevel >= 2 {
+		// Set the "minsize" attribute to reduce code size even further,
+		// regardless of performance loss (-Oz in Clang).
+		kind := llvm.AttributeKindID("minsize")
+		attr := c.ctx.CreateEnumAttribute(kind, 0)
+		llvmFn.AddFunctionAttr(attr)
+	}
+	if c.CPU != "" {
+		llvmFn.AddFunctionAttr(c.ctx.CreateStringAttribute("target-cpu", c.CPU))
+	}
+	if c.Features != "" {
+		llvmFn.AddFunctionAttr(c.ctx.CreateStringAttribute("target-features", c.Features))
+	}
+}
+
+// addStandardDefinedAttributes adds the set of attributes that are added to
+// every function defined by TinyGo (even thunks/wrappers), possibly depending
+// on the architecture. It does not set attributes only set for declared
+// functions, use addStandardDeclaredAttributes for this.
+func (c *compilerContext) addStandardDefinedAttributes(llvmFn llvm.Value) {
+	// TinyGo does not currently raise exceptions, so set the 'nounwind' flag.
+	// This behavior matches Clang when compiling C source files.
+	// It reduces binary size on Linux a little bit on non-x86_64 targets by
+	// eliminating exception tables for these functions.
+	llvmFn.AddFunctionAttr(c.ctx.CreateEnumAttribute(llvm.AttributeKindID("nounwind"), 0))
+	if strings.Split(c.Triple, "-")[0] == "x86_64" {
+		// Required by the ABI.
+		llvmFn.AddFunctionAttr(c.ctx.CreateEnumAttribute(llvm.AttributeKindID("uwtable"), 0))
+	}
+}
+
+// addStandardAttribute adds all attributes added to defined functions.
+func (c *compilerContext) addStandardAttributes(llvmFn llvm.Value) {
+	c.addStandardDeclaredAttributes(llvmFn)
+	c.addStandardDefinedAttributes(llvmFn)
+}
+
 // 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).
@@ -156,12 +156,12 @@ func (b *builder) createRawSyscall(call *ssa.CallCommon) (llvm.Value, error) {
 // 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":
+		syscallResult, err := b.createRawSyscall(call)
+		if err != nil {
+			return syscallResult, err
+		}
 		// Return values: r0, r1 uintptr, err Errno
 		// Pseudocode:
 		//     var err uintptr
@@ -180,6 +180,10 @@ func (b *builder) createSyscall(call *ssa.CallCommon) (llvm.Value, error) {
 		retval = b.CreateInsertValue(retval, errResult, 2, "")
 		return retval, nil
 	case "darwin":
+		syscallResult, err := b.createRawSyscall(call)
+		if err != nil {
+			return syscallResult, err
+		}
 		// Return values: r0, r1 uintptr, err Errno
 		// Pseudocode:
 		//     var err uintptr
@@ -195,6 +199,57 @@ func (b *builder) createSyscall(call *ssa.CallCommon) (llvm.Value, error) {
 		retval = b.CreateInsertValue(retval, zero, 1, "")
 		retval = b.CreateInsertValue(retval, errResult, 2, "")
 		return retval, nil
+	case "windows":
+		// On Windows, syscall.Syscall* is basically just a function pointer
+		// call. This is complicated in gc because of stack switching and the
+		// different ABI, but easy in TinyGo: just call the function pointer.
+		// The signature looks like this:
+		//   func Syscall(trap, nargs, a1, a2, a3 uintptr) (r1, r2 uintptr, err Errno)
+
+		// Prepare input values.
+		var paramTypes []llvm.Type
+		var params []llvm.Value
+		for _, val := range call.Args[2:] {
+			param := b.getValue(val)
+			params = append(params, param)
+			paramTypes = append(paramTypes, param.Type())
+		}
+		llvmType := llvm.FunctionType(b.uintptrType, paramTypes, false)
+		fn := b.getValue(call.Args[0])
+		fnPtr := b.CreateIntToPtr(fn, llvm.PointerType(llvmType, 0), "")
+
+		// Prepare some functions that will be called later.
+		setLastError := b.mod.NamedFunction("SetLastError")
+		if setLastError.IsNil() {
+			llvmType := llvm.FunctionType(b.ctx.VoidType(), []llvm.Type{b.ctx.Int32Type()}, false)
+			setLastError = llvm.AddFunction(b.mod, "SetLastError", llvmType)
+		}
+		getLastError := b.mod.NamedFunction("GetLastError")
+		if getLastError.IsNil() {
+			llvmType := llvm.FunctionType(b.ctx.Int32Type(), nil, false)
+			getLastError = llvm.AddFunction(b.mod, "GetLastError", llvmType)
+		}
+
+		// Now do the actual call. Pseudocode:
+		//     SetLastError(0)
+		//     r1 = trap(a1, a2, a3, ...)
+		//     err = uintptr(GetLastError())
+		//     return r1, 0, err
+		// Note that SetLastError/GetLastError could be replaced with direct
+		// access to the thread control block, which is probably smaller and
+		// faster. The Go runtime does this in assembly.
+		b.CreateCall(setLastError, []llvm.Value{llvm.ConstNull(b.ctx.Int32Type())}, "")
+		syscallResult := b.CreateCall(fnPtr, params, "")
+		errResult := b.CreateCall(getLastError, nil, "err")
+		if b.uintptrType != b.ctx.Int32Type() {
+			errResult = b.CreateZExt(errResult, b.uintptrType, "err.uintptr")
+		}
+
+		// Return r1, 0, err
+		retval := llvm.ConstNull(b.ctx.StructType([]llvm.Type{b.uintptrType, b.uintptrType, b.uintptrType}, false))
+		retval = b.CreateInsertValue(retval, syscallResult, 0, "")
+		retval = b.CreateInsertValue(retval, errResult, 2, "")
+		return retval, nil
 	default:
 		return llvm.Value{}, b.makeError(call.Pos(), "unknown GOOS/GOARCH for syscall: "+b.GOOS+"/"+b.GOARCH)
 	}
@@ -10,6 +10,22 @@ func equalInt(x, y int) bool {
 	return x == y
 }

+func divInt(x, y int) int {
+	return x / y
+}
+
+func divUint(x, y uint) uint {
+	return x / y
+}
+
+func remInt(x, y int) int {
+	return x % y
+}
+
+func remUint(x, y uint) uint {
+	return x % y
+}
+
 func floatEQ(x, y float32) bool {
 	return x == y
 }
@@ -55,3 +71,18 @@ func complexMul(x, y complex64) complex64 {
 }

 // TODO: complexDiv (requires runtime call)
+
+// A type 'kv' also exists in function foo. Test that these two types don't
+// conflict with each other.
+type kv struct {
+	v float32
+}
+
+func foo(a *kv) {
+	// Define a new 'kv' type.
+	type kv struct {
+		v byte
+	}
+	// Use this type.
+	func(b *kv) {}(nil)
+}
@@ -1,74 +1,159 @@
 ; ModuleID = 'basic.go'
 source_filename = "basic.go"
 target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
-target triple = "wasm32--wasi"
+target triple = "wasm32-unknown-wasi"

-declare noalias nonnull i8* @runtime.alloc(i32, i8*, i8*)
+%main.kv = type { float }
+%main.kv.0 = type { i8 }

-define hidden void @main.init(i8* %context, i8* %parentHandle) unnamed_addr {
+declare noalias nonnull i8* @runtime.alloc(i32, i8*, i8*, i8*)
+
+; Function Attrs: nounwind
+define hidden void @main.init(i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  ret void
 }

-define hidden i32 @main.addInt(i32 %x, i32 %y, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden i32 @main.addInt(i32 %x, i32 %y, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  %0 = add i32 %x, %y
  ret i32 %0
 }

-define hidden i1 @main.equalInt(i32 %x, i32 %y, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden i1 @main.equalInt(i32 %x, i32 %y, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 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 {
+; Function Attrs: nounwind
+define hidden i32 @main.divInt(i32 %x, i32 %y, i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  %0 = icmp eq i32 %y, 0
+  br i1 %0, label %divbyzero.throw, label %divbyzero.next
+
+divbyzero.throw:                                  ; preds = %entry
+  call void @runtime.divideByZeroPanic(i8* undef, i8* null) #0
+  unreachable
+
+divbyzero.next:                                   ; preds = %entry
+  %1 = icmp eq i32 %y, -1
+  %2 = icmp eq i32 %x, -2147483648
+  %3 = and i1 %1, %2
+  %4 = select i1 %3, i32 1, i32 %y
+  %5 = sdiv i32 %x, %4
+  ret i32 %5
+}
+
+declare void @runtime.divideByZeroPanic(i8*, i8*)
+
+; Function Attrs: nounwind
+define hidden i32 @main.divUint(i32 %x, i32 %y, i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  %0 = icmp eq i32 %y, 0
+  br i1 %0, label %divbyzero.throw, label %divbyzero.next
+
+divbyzero.throw:                                  ; preds = %entry
+  call void @runtime.divideByZeroPanic(i8* undef, i8* null) #0
+  unreachable
+
+divbyzero.next:                                   ; preds = %entry
+  %1 = udiv i32 %x, %y
+  ret i32 %1
+}
+
+; Function Attrs: nounwind
+define hidden i32 @main.remInt(i32 %x, i32 %y, i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  %0 = icmp eq i32 %y, 0
+  br i1 %0, label %divbyzero.throw, label %divbyzero.next
+
+divbyzero.throw:                                  ; preds = %entry
+  call void @runtime.divideByZeroPanic(i8* undef, i8* null) #0
+  unreachable
+
+divbyzero.next:                                   ; preds = %entry
+  %1 = icmp eq i32 %y, -1
+  %2 = icmp eq i32 %x, -2147483648
+  %3 = and i1 %1, %2
+  %4 = select i1 %3, i32 1, i32 %y
+  %5 = srem i32 %x, %4
+  ret i32 %5
+}
+
+; Function Attrs: nounwind
+define hidden i32 @main.remUint(i32 %x, i32 %y, i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  %0 = icmp eq i32 %y, 0
+  br i1 %0, label %divbyzero.throw, label %divbyzero.next
+
+divbyzero.throw:                                  ; preds = %entry
+  call void @runtime.divideByZeroPanic(i8* undef, i8* null) #0
+  unreachable
+
+divbyzero.next:                                   ; preds = %entry
+  %1 = urem i32 %x, %y
+  ret i32 %1
+}
+
+; Function Attrs: nounwind
+define hidden i1 @main.floatEQ(float %x, float %y, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 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 {
+; Function Attrs: nounwind
+define hidden i1 @main.floatNE(float %x, float %y, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 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 {
+; Function Attrs: nounwind
+define hidden i1 @main.floatLower(float %x, float %y, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 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 {
+; Function Attrs: nounwind
+define hidden i1 @main.floatLowerEqual(float %x, float %y, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 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 {
+; Function Attrs: nounwind
+define hidden i1 @main.floatGreater(float %x, float %y, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 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 {
+; Function Attrs: nounwind
+define hidden i1 @main.floatGreaterEqual(float %x, float %y, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 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 {
+; Function Attrs: nounwind
+define hidden float @main.complexReal(float %x.r, float %x.i, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  ret float %x.r
 }

-define hidden float @main.complexImag(float %x.r, float %x.i, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden float @main.complexImag(float %x.r, float %x.i, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 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 {
+; Function Attrs: nounwind
+define hidden { float, float } @main.complexAdd(float %x.r, float %x.i, float %y.r, float %y.i, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  %0 = fadd float %x.r, %y.r
  %1 = fadd float %x.i, %y.i
@@ -77,7 +162,8 @@ entry:
  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 {
+; Function Attrs: nounwind
+define hidden { float, float } @main.complexSub(float %x.r, float %x.i, float %y.r, float %y.i, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  %0 = fsub float %x.r, %y.r
  %1 = fsub float %x.i, %y.i
@@ -86,7 +172,8 @@ entry:
  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 {
+; Function Attrs: nounwind
+define hidden { float, float } @main.complexMul(float %x.r, float %x.i, float %y.r, float %y.i, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  %0 = fmul float %x.r, %y.r
  %1 = fmul float %x.i, %y.i
@@ -98,3 +185,18 @@ entry:
  %7 = insertvalue { float, float } %6, float %5, 1
  ret { float, float } %7
 }
+
+; Function Attrs: nounwind
+define hidden void @main.foo(%main.kv* dereferenceable_or_null(4) %a, i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  call void @"main.foo$1"(%main.kv.0* null, i8* undef, i8* undef)
+  ret void
+}
+
+; Function Attrs: nounwind
+define hidden void @"main.foo$1"(%main.kv.0* dereferenceable_or_null(1) %b, i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  ret void
+}
+
+attributes #0 = { nounwind }
@@ -0,0 +1,25 @@
+package main
+
+func chanIntSend(ch chan int) {
+	ch <- 3
+}
+
+func chanIntRecv(ch chan int) {
+	<-ch
+}
+
+func chanZeroSend(ch chan struct{}) {
+	ch <- struct{}{}
+}
+
+func chanZeroRecv(ch chan struct{}) {
+	<-ch
+}
+
+func selectZeroRecv(ch1 chan int, ch2 chan struct{}) {
+	select {
+	case ch1 <- 1:
+	case <-ch2:
+	default:
+	}
+}
@@ -0,0 +1,122 @@
+; ModuleID = 'channel.go'
+source_filename = "channel.go"
+target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
+target triple = "wasm32-unknown-wasi"
+
+%runtime.channel = type { i32, i32, i8, %runtime.channelBlockedList*, i32, i32, i32, i8* }
+%runtime.channelBlockedList = type { %runtime.channelBlockedList*, %"internal/task.Task"*, %runtime.chanSelectState*, { %runtime.channelBlockedList*, i32, i32 } }
+%"internal/task.Task" = type { %"internal/task.Task"*, i8*, i64, %"internal/task.state" }
+%"internal/task.state" = type { i32, i8*, %"internal/task.stackState", i1 }
+%"internal/task.stackState" = type { i32, i32 }
+%runtime.chanSelectState = type { %runtime.channel*, i8* }
+
+declare noalias nonnull i8* @runtime.alloc(i32, i8*, i8*, i8*)
+
+; Function Attrs: nounwind
+define hidden void @main.init(i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  ret void
+}
+
+; Function Attrs: nounwind
+define hidden void @main.chanIntSend(%runtime.channel* dereferenceable_or_null(32) %ch, i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  %chan.blockedList = alloca %runtime.channelBlockedList, align 8
+  %chan.value = alloca i32, align 4
+  %chan.value.bitcast = bitcast i32* %chan.value to i8*
+  call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull %chan.value.bitcast)
+  store i32 3, i32* %chan.value, align 4
+  %chan.blockedList.bitcast = bitcast %runtime.channelBlockedList* %chan.blockedList to i8*
+  call void @llvm.lifetime.start.p0i8(i64 24, i8* nonnull %chan.blockedList.bitcast)
+  call void @runtime.chanSend(%runtime.channel* %ch, i8* nonnull %chan.value.bitcast, %runtime.channelBlockedList* nonnull %chan.blockedList, i8* undef, i8* null) #0
+  call void @llvm.lifetime.end.p0i8(i64 24, i8* nonnull %chan.blockedList.bitcast)
+  call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull %chan.value.bitcast)
+  ret void
+}
+
+; Function Attrs: argmemonly nounwind willreturn
+declare void @llvm.lifetime.start.p0i8(i64 immarg, i8* nocapture) #1
+
+declare void @runtime.chanSend(%runtime.channel* dereferenceable_or_null(32), i8*, %runtime.channelBlockedList* dereferenceable_or_null(24), i8*, i8*)
+
+; Function Attrs: argmemonly nounwind willreturn
+declare void @llvm.lifetime.end.p0i8(i64 immarg, i8* nocapture) #1
+
+; Function Attrs: nounwind
+define hidden void @main.chanIntRecv(%runtime.channel* dereferenceable_or_null(32) %ch, i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  %chan.blockedList = alloca %runtime.channelBlockedList, align 8
+  %chan.value = alloca i32, align 4
+  %chan.value.bitcast = bitcast i32* %chan.value to i8*
+  call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull %chan.value.bitcast)
+  %chan.blockedList.bitcast = bitcast %runtime.channelBlockedList* %chan.blockedList to i8*
+  call void @llvm.lifetime.start.p0i8(i64 24, i8* nonnull %chan.blockedList.bitcast)
+  %0 = call i1 @runtime.chanRecv(%runtime.channel* %ch, i8* nonnull %chan.value.bitcast, %runtime.channelBlockedList* nonnull %chan.blockedList, i8* undef, i8* null) #0
+  call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull %chan.value.bitcast)
+  call void @llvm.lifetime.end.p0i8(i64 24, i8* nonnull %chan.blockedList.bitcast)
+  ret void
+}
+
+declare i1 @runtime.chanRecv(%runtime.channel* dereferenceable_or_null(32), i8*, %runtime.channelBlockedList* dereferenceable_or_null(24), i8*, i8*)
+
+; Function Attrs: nounwind
+define hidden void @main.chanZeroSend(%runtime.channel* dereferenceable_or_null(32) %ch, i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  %chan.blockedList = alloca %runtime.channelBlockedList, align 8
+  %chan.blockedList.bitcast = bitcast %runtime.channelBlockedList* %chan.blockedList to i8*
+  call void @llvm.lifetime.start.p0i8(i64 24, i8* nonnull %chan.blockedList.bitcast)
+  call void @runtime.chanSend(%runtime.channel* %ch, i8* null, %runtime.channelBlockedList* nonnull %chan.blockedList, i8* undef, i8* null) #0
+  call void @llvm.lifetime.end.p0i8(i64 24, i8* nonnull %chan.blockedList.bitcast)
+  ret void
+}
+
+; Function Attrs: nounwind
+define hidden void @main.chanZeroRecv(%runtime.channel* dereferenceable_or_null(32) %ch, i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  %chan.blockedList = alloca %runtime.channelBlockedList, align 8
+  %chan.blockedList.bitcast = bitcast %runtime.channelBlockedList* %chan.blockedList to i8*
+  call void @llvm.lifetime.start.p0i8(i64 24, i8* nonnull %chan.blockedList.bitcast)
+  %0 = call i1 @runtime.chanRecv(%runtime.channel* %ch, i8* null, %runtime.channelBlockedList* nonnull %chan.blockedList, i8* undef, i8* null) #0
+  call void @llvm.lifetime.end.p0i8(i64 24, i8* nonnull %chan.blockedList.bitcast)
+  ret void
+}
+
+; Function Attrs: nounwind
+define hidden void @main.selectZeroRecv(%runtime.channel* dereferenceable_or_null(32) %ch1, %runtime.channel* dereferenceable_or_null(32) %ch2, i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  %select.states.alloca = alloca [2 x %runtime.chanSelectState], align 8
+  %select.send.value = alloca i32, align 4
+  store i32 1, i32* %select.send.value, align 4
+  %select.states.alloca.bitcast = bitcast [2 x %runtime.chanSelectState]* %select.states.alloca to i8*
+  call void @llvm.lifetime.start.p0i8(i64 16, i8* nonnull %select.states.alloca.bitcast)
+  %.repack = getelementptr inbounds [2 x %runtime.chanSelectState], [2 x %runtime.chanSelectState]* %select.states.alloca, i32 0, i32 0, i32 0
+  store %runtime.channel* %ch1, %runtime.channel** %.repack, align 8
+  %.repack1 = getelementptr inbounds [2 x %runtime.chanSelectState], [2 x %runtime.chanSelectState]* %select.states.alloca, i32 0, i32 0, i32 1
+  %0 = bitcast i8** %.repack1 to i32**
+  store i32* %select.send.value, i32** %0, align 4
+  %.repack3 = getelementptr inbounds [2 x %runtime.chanSelectState], [2 x %runtime.chanSelectState]* %select.states.alloca, i32 0, i32 1, i32 0
+  store %runtime.channel* %ch2, %runtime.channel** %.repack3, align 8
+  %.repack4 = getelementptr inbounds [2 x %runtime.chanSelectState], [2 x %runtime.chanSelectState]* %select.states.alloca, i32 0, i32 1, i32 1
+  store i8* null, i8** %.repack4, align 4
+  %select.states = getelementptr inbounds [2 x %runtime.chanSelectState], [2 x %runtime.chanSelectState]* %select.states.alloca, i32 0, i32 0
+  %select.result = call { i32, i1 } @runtime.tryChanSelect(i8* undef, %runtime.chanSelectState* nonnull %select.states, i32 2, i32 2, i8* undef, i8* null) #0
+  call void @llvm.lifetime.end.p0i8(i64 16, i8* nonnull %select.states.alloca.bitcast)
+  %1 = extractvalue { i32, i1 } %select.result, 0
+  %2 = icmp eq i32 %1, 0
+  br i1 %2, label %select.done, label %select.next
+
+select.done:                                      ; preds = %select.body, %select.next, %entry
+  ret void
+
+select.next:                                      ; preds = %entry
+  %3 = icmp eq i32 %1, 1
+  br i1 %3, label %select.body, label %select.done
+
+select.body:                                      ; preds = %select.next
+  br label %select.done
+}
+
+declare { i32, i1 } @runtime.tryChanSelect(i8*, %runtime.chanSelectState*, i32, i32, i8*, i8*)
+
+attributes #0 = { nounwind }
+attributes #1 = { argmemonly nounwind willreturn }
@@ -1,16 +1,18 @@
 ; ModuleID = 'float.go'
 source_filename = "float.go"
 target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
-target triple = "wasm32--wasi"
+target triple = "wasm32-unknown-wasi"

-declare noalias nonnull i8* @runtime.alloc(i32, i8*, i8*)
+declare noalias nonnull i8* @runtime.alloc(i32, i8*, i8*, i8*)

-define hidden void @main.init(i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden void @main.init(i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  ret void
 }

-define hidden i32 @main.f32tou32(float %v, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden i32 @main.f32tou32(float %v, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  %positive = fcmp oge float %v, 0.000000e+00
  %withinmax = fcmp ole float %v, 0x41EFFFFFC0000000
@@ -21,22 +23,26 @@ entry:
  ret i32 %0
 }

-define hidden float @main.maxu32f(i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden float @main.maxu32f(i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  ret float 0x41F0000000000000
 }

-define hidden i32 @main.maxu32tof32(i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden i32 @main.maxu32tof32(i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  ret i32 -1
 }

-define hidden { i32, i32, i32, i32 } @main.inftoi32(i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden { i32, i32, i32, i32 } @main.inftoi32(i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  ret { i32, i32, i32, i32 } { i32 -1, i32 0, i32 2147483647, i32 -2147483648 }
 }

-define hidden i32 @main.u32tof32tou32(i32 %v, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden i32 @main.u32tof32tou32(i32 %v, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  %0 = uitofp i32 %v to float
  %withinmax = fcmp ole float %0, 0x41EFFFFFC0000000
@@ -45,7 +51,8 @@ entry:
  ret i32 %1
 }

-define hidden float @main.f32tou32tof32(float %v, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden float @main.f32tou32tof32(float %v, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  %positive = fcmp oge float %v, 0.000000e+00
  %withinmax = fcmp ole float %v, 0x41EFFFFFC0000000
@@ -57,7 +64,8 @@ entry:
  ret float %1
 }

-define hidden i8 @main.f32tou8(float %v, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden i8 @main.f32tou8(float %v, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  %positive = fcmp oge float %v, 0.000000e+00
  %withinmax = fcmp ole float %v, 2.550000e+02
@@ -68,7 +76,8 @@ entry:
  ret i8 %0
 }

-define hidden i8 @main.f32toi8(float %v, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden i8 @main.f32toi8(float %v, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  %abovemin = fcmp oge float %v, -1.280000e+02
  %belowmax = fcmp ole float %v, 1.270000e+02
@@ -80,3 +89,5 @@ entry:
  %0 = select i1 %inbounds, i8 %normal, i8 %remapped
  ret i8 %0
 }
+
+attributes #0 = { nounwind }
@@ -1,33 +1,35 @@
 ; ModuleID = 'func.go'
 source_filename = "func.go"
 target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
-target triple = "wasm32--wasi"
+target triple = "wasm32-unknown-wasi"

 %runtime.funcValueWithSignature = type { i32, i8* }

@"reflect/types.funcid:func:{basic:int}{}" = external constant i8
@"main.someFunc$withSignature" = linkonce_odr constant %runtime.funcValueWithSignature { i32 ptrtoint (void (i32, i8*, i8*)* @main.someFunc to i32), i8* @"reflect/types.funcid:func:{basic:int}{}" }

-declare noalias nonnull i8* @runtime.alloc(i32, i8*, i8*)
+declare noalias nonnull i8* @runtime.alloc(i32, i8*, i8*, i8*)

-define hidden void @main.init(i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden void @main.init(i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  ret void
 }

-define hidden void @main.foo(i8* %callback.context, i32 %callback.funcptr, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden void @main.foo(i8* %callback.context, i32 %callback.funcptr, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
-  %0 = call i32 @runtime.getFuncPtr(i8* %callback.context, i32 %callback.funcptr, i8* nonnull @"reflect/types.funcid:func:{basic:int}{}", i8* undef, i8* null)
+  %0 = call i32 @runtime.getFuncPtr(i8* %callback.context, i32 %callback.funcptr, i8* nonnull @"reflect/types.funcid:func:{basic:int}{}", i8* undef, i8* null) #0
  %1 = icmp eq i32 %0, 0
  br i1 %1, label %fpcall.throw, label %fpcall.next

 fpcall.throw:                                     ; preds = %entry
-  call void @runtime.nilPanic(i8* undef, i8* null)
+  call void @runtime.nilPanic(i8* undef, i8* null) #0
  unreachable

 fpcall.next:                                      ; preds = %entry
  %2 = inttoptr i32 %0 to void (i32, i8*, i8*)*
-  call void %2(i32 3, i8* %callback.context, i8* undef)
+  call void %2(i32 3, i8* %callback.context, i8* undef) #0
  ret void
 }

@@ -35,13 +37,17 @@ declare i32 @runtime.getFuncPtr(i8*, i32, i8* dereferenceable_or_null(1), i8*, i

 declare void @runtime.nilPanic(i8*, i8*)

-define hidden void @main.bar(i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden void @main.bar(i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  call void @main.foo(i8* undef, i32 ptrtoint (%runtime.funcValueWithSignature* @"main.someFunc$withSignature" to i32), i8* undef, i8* undef)
  ret void
 }

-define hidden void @main.someFunc(i32 %arg0, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden void @main.someFunc(i32 %arg0, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  ret void
 }
+
+attributes #0 = { nounwind }
@@ -0,0 +1,72 @@
+package main
+
+var (
+	scalar1 *byte
+	scalar2 *int32
+	scalar3 *int64
+	scalar4 *float32
+
+	array1 *[3]byte
+	array2 *[71]byte
+	array3 *[3]*byte
+
+	struct1 *struct{}
+	struct2 *struct {
+		x int
+		y int
+	}
+	struct3 *struct {
+		x *byte
+		y [60]uintptr
+		z *byte
+	}
+
+	slice1 []byte
+	slice2 []*int
+	slice3 [][]byte
+)
+
+func newScalar() {
+	scalar1 = new(byte)
+	scalar2 = new(int32)
+	scalar3 = new(int64)
+	scalar4 = new(float32)
+}
+
+func newArray() {
+	array1 = new([3]byte)
+	array2 = new([71]byte)
+	array3 = new([3]*byte)
+}
+
+func newStruct() {
+	struct1 = new(struct{})
+	struct2 = new(struct {
+		x int
+		y int
+	})
+	struct3 = new(struct {
+		x *byte
+		y [60]uintptr
+		z *byte
+	})
+}
+
+func newFuncValue() *func() {
+	// On some platforms that use runtime.funcValue ("switch" style) function
+	// values, a func value is allocated as having two pointer words while the
+	// struct looks like {unsafe.Pointer; uintptr}. This is so that the interp
+	// package won't get confused, see getPointerBitmap in compiler/llvm.go for
+	// details.
+	return new(func())
+}
+
+func makeSlice() {
+	slice1 = make([]byte, 5)
+	slice2 = make([]*int, 5)
+	slice3 = make([][]byte, 5)
+}
+
+func makeInterface(v complex128) interface{} {
+	return v // always stored in an allocation
+}
@@ -0,0 +1,112 @@
+; ModuleID = 'gc.go'
+source_filename = "gc.go"
+target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
+target triple = "wasm32-unknown-wasi"
+
+%runtime.typecodeID = type { %runtime.typecodeID*, i32, %runtime.interfaceMethodInfo*, %runtime.typecodeID*, i32 }
+%runtime.interfaceMethodInfo = type { i8*, i32 }
+%runtime._interface = type { i32, i8* }
+
+@main.scalar1 = hidden global i8* null, align 4
+@main.scalar2 = hidden global i32* null, align 4
+@main.scalar3 = hidden global i64* null, align 4
+@main.scalar4 = hidden global float* null, align 4
+@main.array1 = hidden global [3 x i8]* null, align 4
+@main.array2 = hidden global [71 x i8]* null, align 4
+@main.array3 = hidden global [3 x i8*]* null, align 4
+@main.struct1 = hidden global {}* null, align 4
+@main.struct2 = hidden global { i32, i32 }* null, align 4
+@main.struct3 = hidden global { i8*, [60 x i32], i8* }* null, align 4
+@main.slice1 = hidden global { i8*, i32, i32 } zeroinitializer, align 8
+@main.slice2 = hidden global { i32**, i32, i32 } zeroinitializer, align 8
+@main.slice3 = hidden global { { i8*, i32, i32 }*, i32, i32 } zeroinitializer, align 8
+@"runtime/gc.layout:62-2000000000000001" = linkonce_odr unnamed_addr constant { i32, [8 x i8] } { i32 62, [8 x i8] c" \00\00\00\00\00\00\01" }
+@"reflect/types.type:basic:complex128" = linkonce_odr constant %runtime.typecodeID { %runtime.typecodeID* null, i32 0, %runtime.interfaceMethodInfo* null, %runtime.typecodeID* @"reflect/types.type:pointer:basic:complex128", i32 0 }
+@"reflect/types.type:pointer:basic:complex128" = linkonce_odr constant %runtime.typecodeID { %runtime.typecodeID* @"reflect/types.type:basic:complex128", i32 0, %runtime.interfaceMethodInfo* null, %runtime.typecodeID* null, i32 0 }
+
+declare noalias nonnull i8* @runtime.alloc(i32, i8*, i8*, i8*)
+
+; Function Attrs: nounwind
+define hidden void @main.init(i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  ret void
+}
+
+; Function Attrs: nounwind
+define hidden void @main.newScalar(i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  %new = call i8* @runtime.alloc(i32 1, i8* nonnull inttoptr (i32 3 to i8*), i8* undef, i8* null) #0
+  store i8* %new, i8** @main.scalar1, align 4
+  %new1 = call i8* @runtime.alloc(i32 4, i8* nonnull inttoptr (i32 3 to i8*), i8* undef, i8* null) #0
+  store i8* %new1, i8** bitcast (i32** @main.scalar2 to i8**), align 4
+  %new2 = call i8* @runtime.alloc(i32 8, i8* nonnull inttoptr (i32 3 to i8*), i8* undef, i8* null) #0
+  store i8* %new2, i8** bitcast (i64** @main.scalar3 to i8**), align 4
+  %new3 = call i8* @runtime.alloc(i32 4, i8* nonnull inttoptr (i32 3 to i8*), i8* undef, i8* null) #0
+  store i8* %new3, i8** bitcast (float** @main.scalar4 to i8**), align 4
+  ret void
+}
+
+; Function Attrs: nounwind
+define hidden void @main.newArray(i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  %new = call i8* @runtime.alloc(i32 3, i8* nonnull inttoptr (i32 3 to i8*), i8* undef, i8* null) #0
+  store i8* %new, i8** bitcast ([3 x i8]** @main.array1 to i8**), align 4
+  %new1 = call i8* @runtime.alloc(i32 71, i8* nonnull inttoptr (i32 3 to i8*), i8* undef, i8* null) #0
+  store i8* %new1, i8** bitcast ([71 x i8]** @main.array2 to i8**), align 4
+  %new2 = call i8* @runtime.alloc(i32 12, i8* nonnull inttoptr (i32 67 to i8*), i8* undef, i8* null) #0
+  store i8* %new2, i8** bitcast ([3 x i8*]** @main.array3 to i8**), align 4
+  ret void
+}
+
+; Function Attrs: nounwind
+define hidden void @main.newStruct(i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  %new = call i8* @runtime.alloc(i32 0, i8* nonnull inttoptr (i32 3 to i8*), i8* undef, i8* null) #0
+  store i8* %new, i8** bitcast ({}** @main.struct1 to i8**), align 4
+  %new1 = call i8* @runtime.alloc(i32 8, i8* nonnull inttoptr (i32 3 to i8*), i8* undef, i8* null) #0
+  store i8* %new1, i8** bitcast ({ i32, i32 }** @main.struct2 to i8**), align 4
+  %new2 = call i8* @runtime.alloc(i32 248, i8* bitcast ({ i32, [8 x i8] }* @"runtime/gc.layout:62-2000000000000001" to i8*), i8* undef, i8* null) #0
+  store i8* %new2, i8** bitcast ({ i8*, [60 x i32], i8* }** @main.struct3 to i8**), align 4
+  ret void
+}
+
+; Function Attrs: nounwind
+define hidden { i8*, void ()* }* @main.newFuncValue(i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  %new = call i8* @runtime.alloc(i32 8, i8* nonnull inttoptr (i32 197 to i8*), i8* undef, i8* null) #0
+  %0 = bitcast i8* %new to { i8*, void ()* }*
+  ret { i8*, void ()* }* %0
+}
+
+; Function Attrs: nounwind
+define hidden void @main.makeSlice(i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  %makeslice = call i8* @runtime.alloc(i32 5, i8* nonnull inttoptr (i32 3 to i8*), i8* undef, i8* null) #0
+  store i8* %makeslice, i8** getelementptr inbounds ({ i8*, i32, i32 }, { i8*, i32, i32 }* @main.slice1, i32 0, i32 0), align 8
+  store i32 5, i32* getelementptr inbounds ({ i8*, i32, i32 }, { i8*, i32, i32 }* @main.slice1, i32 0, i32 1), align 4
+  store i32 5, i32* getelementptr inbounds ({ i8*, i32, i32 }, { i8*, i32, i32 }* @main.slice1, i32 0, i32 2), align 8
+  %makeslice1 = call i8* @runtime.alloc(i32 20, i8* nonnull inttoptr (i32 67 to i8*), i8* undef, i8* null) #0
+  store i8* %makeslice1, i8** bitcast ({ i32**, i32, i32 }* @main.slice2 to i8**), align 8
+  store i32 5, i32* getelementptr inbounds ({ i32**, i32, i32 }, { i32**, i32, i32 }* @main.slice2, i32 0, i32 1), align 4
+  store i32 5, i32* getelementptr inbounds ({ i32**, i32, i32 }, { i32**, i32, i32 }* @main.slice2, i32 0, i32 2), align 8
+  %makeslice3 = call i8* @runtime.alloc(i32 60, i8* nonnull inttoptr (i32 71 to i8*), i8* undef, i8* null) #0
+  store i8* %makeslice3, i8** bitcast ({ { i8*, i32, i32 }*, i32, i32 }* @main.slice3 to i8**), align 8
+  store i32 5, i32* getelementptr inbounds ({ { i8*, i32, i32 }*, i32, i32 }, { { i8*, i32, i32 }*, i32, i32 }* @main.slice3, i32 0, i32 1), align 4
+  store i32 5, i32* getelementptr inbounds ({ { i8*, i32, i32 }*, i32, i32 }, { { i8*, i32, i32 }*, i32, i32 }* @main.slice3, i32 0, i32 2), align 8
+  ret void
+}
+
+; Function Attrs: nounwind
+define hidden %runtime._interface @main.makeInterface(double %v.r, double %v.i, i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  %0 = call i8* @runtime.alloc(i32 16, i8* null, i8* undef, i8* null) #0
+  %.repack = bitcast i8* %0 to double*
+  store double %v.r, double* %.repack, align 8
+  %.repack1 = getelementptr inbounds i8, i8* %0, i32 8
+  %1 = bitcast i8* %.repack1 to double*
+  store double %v.i, double* %1, align 8
+  %2 = insertvalue %runtime._interface { i32 ptrtoint (%runtime.typecodeID* @"reflect/types.type:basic:complex128" to i32), i8* undef }, i8* %0, 1
+  ret %runtime._interface %2
+}
+
+attributes #0 = { nounwind }
@@ -0,0 +1,41 @@
+package main
+
+// Test changes to the language introduced in Go 1.17.
+// For details, see: https://tip.golang.org/doc/go1.17#language
+// These tests should be merged into the regular slice tests once Go 1.17 is the
+// minimun Go version for TinyGo.
+
+import "unsafe"
+
+func Add32(p unsafe.Pointer, len int) unsafe.Pointer {
+	return unsafe.Add(p, len)
+}
+
+func Add64(p unsafe.Pointer, len int64) unsafe.Pointer {
+	return unsafe.Add(p, len)
+}
+
+func SliceToArray(s []int) *[4]int {
+	return (*[4]int)(s)
+}
+
+func SliceToArrayConst() *[4]int {
+	s := make([]int, 6)
+	return (*[4]int)(s)
+}
+
+func SliceInt(ptr *int, len int) []int {
+	return unsafe.Slice(ptr, len)
+}
+
+func SliceUint16(ptr *byte, len uint16) []byte {
+	return unsafe.Slice(ptr, len)
+}
+
+func SliceUint64(ptr *int, len uint64) []int {
+	return unsafe.Slice(ptr, len)
+}
+
+func SliceInt64(ptr *int, len int64) []int {
+	return unsafe.Slice(ptr, len)
+}
@@ -0,0 +1,147 @@
+; ModuleID = 'go1.17.go'
+source_filename = "go1.17.go"
+target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
+target triple = "wasm32-unknown-wasi"
+
+declare noalias nonnull i8* @runtime.alloc(i32, i8*, i8*, i8*)
+
+; Function Attrs: nounwind
+define hidden void @main.init(i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  ret void
+}
+
+; Function Attrs: nounwind
+define hidden i8* @main.Add32(i8* %p, i32 %len, i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  %0 = getelementptr i8, i8* %p, i32 %len
+  ret i8* %0
+}
+
+; Function Attrs: nounwind
+define hidden i8* @main.Add64(i8* %p, i64 %len, i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  %0 = trunc i64 %len to i32
+  %1 = getelementptr i8, i8* %p, i32 %0
+  ret i8* %1
+}
+
+; Function Attrs: nounwind
+define hidden [4 x i32]* @main.SliceToArray(i32* %s.data, i32 %s.len, i32 %s.cap, i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  %0 = icmp ult i32 %s.len, 4
+  br i1 %0, label %slicetoarray.throw, label %slicetoarray.next
+
+slicetoarray.throw:                               ; preds = %entry
+  call void @runtime.sliceToArrayPointerPanic(i8* undef, i8* null) #0
+  unreachable
+
+slicetoarray.next:                                ; preds = %entry
+  %1 = bitcast i32* %s.data to [4 x i32]*
+  ret [4 x i32]* %1
+}
+
+declare void @runtime.sliceToArrayPointerPanic(i8*, i8*)
+
+; Function Attrs: nounwind
+define hidden [4 x i32]* @main.SliceToArrayConst(i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  %makeslice = call i8* @runtime.alloc(i32 24, i8* nonnull inttoptr (i32 3 to i8*), i8* undef, i8* null) #0
+  br i1 false, label %slicetoarray.throw, label %slicetoarray.next
+
+slicetoarray.throw:                               ; preds = %entry
+  unreachable
+
+slicetoarray.next:                                ; preds = %entry
+  %0 = bitcast i8* %makeslice to [4 x i32]*
+  ret [4 x i32]* %0
+}
+
+; Function Attrs: nounwind
+define hidden { i32*, i32, i32 } @main.SliceInt(i32* dereferenceable_or_null(4) %ptr, i32 %len, i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  %0 = icmp ugt i32 %len, 1073741823
+  %1 = icmp eq i32* %ptr, null
+  %2 = icmp ne i32 %len, 0
+  %3 = and i1 %1, %2
+  %4 = or i1 %3, %0
+  br i1 %4, label %unsafe.Slice.throw, label %unsafe.Slice.next
+
+unsafe.Slice.throw:                               ; preds = %entry
+  call void @runtime.unsafeSlicePanic(i8* undef, i8* null) #0
+  unreachable
+
+unsafe.Slice.next:                                ; preds = %entry
+  %5 = insertvalue { i32*, i32, i32 } undef, i32* %ptr, 0
+  %6 = insertvalue { i32*, i32, i32 } %5, i32 %len, 1
+  %7 = insertvalue { i32*, i32, i32 } %6, i32 %len, 2
+  ret { i32*, i32, i32 } %7
+}
+
+declare void @runtime.unsafeSlicePanic(i8*, i8*)
+
+; Function Attrs: nounwind
+define hidden { i8*, i32, i32 } @main.SliceUint16(i8* dereferenceable_or_null(1) %ptr, i16 %len, i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  %0 = icmp eq i8* %ptr, null
+  %1 = icmp ne i16 %len, 0
+  %2 = and i1 %0, %1
+  br i1 %2, label %unsafe.Slice.throw, label %unsafe.Slice.next
+
+unsafe.Slice.throw:                               ; preds = %entry
+  call void @runtime.unsafeSlicePanic(i8* undef, i8* null) #0
+  unreachable
+
+unsafe.Slice.next:                                ; preds = %entry
+  %3 = zext i16 %len to i32
+  %4 = insertvalue { i8*, i32, i32 } undef, i8* %ptr, 0
+  %5 = insertvalue { i8*, i32, i32 } %4, i32 %3, 1
+  %6 = insertvalue { i8*, i32, i32 } %5, i32 %3, 2
+  ret { i8*, i32, i32 } %6
+}
+
+; Function Attrs: nounwind
+define hidden { i32*, i32, i32 } @main.SliceUint64(i32* dereferenceable_or_null(4) %ptr, i64 %len, i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  %0 = icmp ugt i64 %len, 1073741823
+  %1 = icmp eq i32* %ptr, null
+  %2 = icmp ne i64 %len, 0
+  %3 = and i1 %1, %2
+  %4 = or i1 %3, %0
+  br i1 %4, label %unsafe.Slice.throw, label %unsafe.Slice.next
+
+unsafe.Slice.throw:                               ; preds = %entry
+  call void @runtime.unsafeSlicePanic(i8* undef, i8* null) #0
+  unreachable
+
+unsafe.Slice.next:                                ; preds = %entry
+  %5 = trunc i64 %len to i32
+  %6 = insertvalue { i32*, i32, i32 } undef, i32* %ptr, 0
+  %7 = insertvalue { i32*, i32, i32 } %6, i32 %5, 1
+  %8 = insertvalue { i32*, i32, i32 } %7, i32 %5, 2
+  ret { i32*, i32, i32 } %8
+}
+
+; Function Attrs: nounwind
+define hidden { i32*, i32, i32 } @main.SliceInt64(i32* dereferenceable_or_null(4) %ptr, i64 %len, i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  %0 = icmp ugt i64 %len, 1073741823
+  %1 = icmp eq i32* %ptr, null
+  %2 = icmp ne i64 %len, 0
+  %3 = and i1 %1, %2
+  %4 = or i1 %3, %0
+  br i1 %4, label %unsafe.Slice.throw, label %unsafe.Slice.next
+
+unsafe.Slice.throw:                               ; preds = %entry
+  call void @runtime.unsafeSlicePanic(i8* undef, i8* null) #0
+  unreachable
+
+unsafe.Slice.next:                                ; preds = %entry
+  %5 = trunc i64 %len to i32
+  %6 = insertvalue { i32*, i32, i32 } undef, i32* %ptr, 0
+  %7 = insertvalue { i32*, i32, i32 } %6, i32 %5, 1
+  %8 = insertvalue { i32*, i32, i32 } %7, i32 %5, 2
+  ret { i32*, i32, i32 } %8
+}
+
+attributes #0 = { nounwind }
@@ -1,7 +1,7 @@
 ; ModuleID = 'goroutine.go'
 source_filename = "goroutine.go"
 target datalayout = "e-m:e-p:32:32-Fi8-i64:64-v128:64:128-a:0:32-n32-S64"
-target triple = "armv7m-none-eabi"
+target triple = "thumbv7m-unknown-unknown-eabi"

 %runtime.channel = type { i32, i32, i8, %runtime.channelBlockedList*, i32, i32, i32, i8* }
 %runtime.channelBlockedList = type { %runtime.channelBlockedList*, %"internal/task.Task"*, %runtime.chanSelectState*, { %runtime.channelBlockedList*, i32, i32 } }
@@ -9,26 +9,31 @@ target triple = "armv7m-none-eabi"
 %"internal/task.state" = type { i32, i32* }
 %runtime.chanSelectState = type { %runtime.channel*, i8* }

-declare noalias nonnull i8* @runtime.alloc(i32, i8*, i8*)
+@"main$string" = internal unnamed_addr constant [4 x i8] c"test", align 1

-define hidden void @main.init(i8* %context, i8* %parentHandle) unnamed_addr {
+declare noalias nonnull i8* @runtime.alloc(i32, i8*, i8*, i8*)
+
+; Function Attrs: nounwind
+define hidden void @main.init(i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  ret void
 }

-define hidden void @main.regularFunctionGoroutine(i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden void @main.regularFunctionGoroutine(i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
-  %stacksize = call i32 @"internal/task.getGoroutineStackSize"(i32 ptrtoint (void (i8*)* @"main.regularFunction$gowrapper" to i32), i8* undef, i8* undef)
-  call void @"internal/task.start"(i32 ptrtoint (void (i8*)* @"main.regularFunction$gowrapper" to i32), i8* nonnull inttoptr (i32 5 to i8*), i32 %stacksize, i8* undef, i8* null)
+  %stacksize = call i32 @"internal/task.getGoroutineStackSize"(i32 ptrtoint (void (i8*)* @"main.regularFunction$gowrapper" to i32), i8* undef, i8* undef) #0
+  call void @"internal/task.start"(i32 ptrtoint (void (i8*)* @"main.regularFunction$gowrapper" to i32), i8* nonnull inttoptr (i32 5 to i8*), i32 %stacksize, i8* undef, i8* null) #0
  ret void
 }

 declare void @main.regularFunction(i32, i8*, i8*)

-define linkonce_odr void @"main.regularFunction$gowrapper"(i8* %0) unnamed_addr #0 {
+; Function Attrs: nounwind
+define linkonce_odr void @"main.regularFunction$gowrapper"(i8* %0) unnamed_addr #1 {
 entry:
  %unpack.int = ptrtoint i8* %0 to i32
-  call void @main.regularFunction(i32 %unpack.int, i8* undef, i8* undef)
+  call void @main.regularFunction(i32 %unpack.int, i8* undef, i8* undef) #0
  ret void
 }

@@ -36,51 +41,57 @@ declare i32 @"internal/task.getGoroutineStackSize"(i32, i8*, i8*)

 declare void @"internal/task.start"(i32, i8*, i32, i8*, i8*)

-define hidden void @main.inlineFunctionGoroutine(i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden void @main.inlineFunctionGoroutine(i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
-  %stacksize = call i32 @"internal/task.getGoroutineStackSize"(i32 ptrtoint (void (i8*)* @"main.inlineFunctionGoroutine$1$gowrapper" to i32), i8* undef, i8* undef)
-  call void @"internal/task.start"(i32 ptrtoint (void (i8*)* @"main.inlineFunctionGoroutine$1$gowrapper" to i32), i8* nonnull inttoptr (i32 5 to i8*), i32 %stacksize, i8* undef, i8* null)
+  %stacksize = call i32 @"internal/task.getGoroutineStackSize"(i32 ptrtoint (void (i8*)* @"main.inlineFunctionGoroutine$1$gowrapper" to i32), i8* undef, i8* undef) #0
+  call void @"internal/task.start"(i32 ptrtoint (void (i8*)* @"main.inlineFunctionGoroutine$1$gowrapper" to i32), i8* nonnull inttoptr (i32 5 to i8*), i32 %stacksize, i8* undef, i8* null) #0
  ret void
 }

-define hidden void @"main.inlineFunctionGoroutine$1"(i32 %x, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden void @"main.inlineFunctionGoroutine$1"(i32 %x, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  ret void
 }

-define linkonce_odr void @"main.inlineFunctionGoroutine$1$gowrapper"(i8* %0) unnamed_addr #1 {
+; Function Attrs: nounwind
+define linkonce_odr void @"main.inlineFunctionGoroutine$1$gowrapper"(i8* %0) unnamed_addr #2 {
 entry:
  %unpack.int = ptrtoint i8* %0 to i32
  call void @"main.inlineFunctionGoroutine$1"(i32 %unpack.int, i8* undef, i8* undef)
  ret void
 }

-define hidden void @main.closureFunctionGoroutine(i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden void @main.closureFunctionGoroutine(i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
-  %n = call i8* @runtime.alloc(i32 4, i8* undef, i8* null)
+  %n = call i8* @runtime.alloc(i32 4, i8* nonnull inttoptr (i32 3 to i8*), i8* undef, i8* null) #0
  %0 = bitcast i8* %n to i32*
  store i32 3, i32* %0, align 4
-  %1 = call i8* @runtime.alloc(i32 8, i8* undef, i8* null)
+  %1 = call i8* @runtime.alloc(i32 8, i8* null, i8* undef, i8* null) #0
  %2 = bitcast i8* %1 to i32*
  store i32 5, i32* %2, align 4
  %3 = getelementptr inbounds i8, i8* %1, i32 4
  %4 = bitcast i8* %3 to i8**
  store i8* %n, i8** %4, align 4
-  %stacksize = call i32 @"internal/task.getGoroutineStackSize"(i32 ptrtoint (void (i8*)* @"main.closureFunctionGoroutine$1$gowrapper" to i32), i8* undef, i8* undef)
-  call void @"internal/task.start"(i32 ptrtoint (void (i8*)* @"main.closureFunctionGoroutine$1$gowrapper" to i32), i8* nonnull %1, i32 %stacksize, i8* undef, i8* null)
+  %stacksize = call i32 @"internal/task.getGoroutineStackSize"(i32 ptrtoint (void (i8*)* @"main.closureFunctionGoroutine$1$gowrapper" to i32), i8* undef, i8* undef) #0
+  call void @"internal/task.start"(i32 ptrtoint (void (i8*)* @"main.closureFunctionGoroutine$1$gowrapper" to i32), i8* nonnull %1, i32 %stacksize, i8* undef, i8* null) #0
  %5 = load i32, i32* %0, align 4
-  call void @runtime.printint32(i32 %5, i8* undef, i8* null)
+  call void @runtime.printint32(i32 %5, i8* undef, i8* null) #0
  ret void
 }

-define hidden void @"main.closureFunctionGoroutine$1"(i32 %x, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden void @"main.closureFunctionGoroutine$1"(i32 %x, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  %unpack.ptr = bitcast i8* %context to i32*
  store i32 7, i32* %unpack.ptr, align 4
  ret void
 }

-define linkonce_odr void @"main.closureFunctionGoroutine$1$gowrapper"(i8* %0) unnamed_addr #2 {
+; Function Attrs: nounwind
+define linkonce_odr void @"main.closureFunctionGoroutine$1$gowrapper"(i8* %0) unnamed_addr #3 {
 entry:
  %1 = bitcast i8* %0 to i32*
  %2 = load i32, i32* %1, align 4
@@ -93,23 +104,25 @@ entry:

 declare void @runtime.printint32(i32, i8*, i8*)

-define hidden void @main.funcGoroutine(i8* %fn.context, void (i32, i8*, i8*)* %fn.funcptr, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden void @main.funcGoroutine(i8* %fn.context, void ()* %fn.funcptr, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
-  %0 = call i8* @runtime.alloc(i32 12, i8* undef, i8* null)
+  %0 = call i8* @runtime.alloc(i32 12, i8* null, i8* undef, i8* null) #0
  %1 = bitcast i8* %0 to i32*
  store i32 5, i32* %1, align 4
  %2 = getelementptr inbounds i8, i8* %0, i32 4
  %3 = bitcast i8* %2 to i8**
  store i8* %fn.context, i8** %3, align 4
  %4 = getelementptr inbounds i8, i8* %0, i32 8
-  %5 = bitcast i8* %4 to void (i32, i8*, i8*)**
-  store void (i32, i8*, i8*)* %fn.funcptr, void (i32, i8*, i8*)** %5, align 4
-  %stacksize = call i32 @"internal/task.getGoroutineStackSize"(i32 ptrtoint (void (i8*)* @main.funcGoroutine.gowrapper to i32), i8* undef, i8* undef)
-  call void @"internal/task.start"(i32 ptrtoint (void (i8*)* @main.funcGoroutine.gowrapper to i32), i8* nonnull %0, i32 %stacksize, i8* undef, i8* null)
+  %5 = bitcast i8* %4 to void ()**
+  store void ()* %fn.funcptr, void ()** %5, align 4
+  %stacksize = call i32 @"internal/task.getGoroutineStackSize"(i32 ptrtoint (void (i8*)* @main.funcGoroutine.gowrapper to i32), i8* undef, i8* undef) #0
+  call void @"internal/task.start"(i32 ptrtoint (void (i8*)* @main.funcGoroutine.gowrapper to i32), i8* nonnull %0, i32 %stacksize, i8* undef, i8* null) #0
  ret void
 }

-define linkonce_odr void @main.funcGoroutine.gowrapper(i8* %0) unnamed_addr #3 {
+; Function Attrs: nounwind
+define linkonce_odr void @main.funcGoroutine.gowrapper(i8* %0) unnamed_addr #4 {
 entry:
  %1 = bitcast i8* %0 to i32*
  %2 = load i32, i32* %1, align 4
@@ -119,32 +132,78 @@ entry:
  %6 = getelementptr inbounds i8, i8* %0, i32 8
  %7 = bitcast i8* %6 to void (i32, i8*, i8*)**
  %8 = load void (i32, i8*, i8*)*, void (i32, i8*, i8*)** %7, align 4
-  call void %8(i32 %2, i8* %5, i8* undef)
+  call void %8(i32 %2, i8* %5, i8* undef) #0
  ret void
 }

-define hidden void @main.recoverBuiltinGoroutine(i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden void @main.recoverBuiltinGoroutine(i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  ret void
 }

-define hidden void @main.copyBuiltinGoroutine(i8* %dst.data, i32 %dst.len, i32 %dst.cap, i8* %src.data, i32 %src.len, i32 %src.cap, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden void @main.copyBuiltinGoroutine(i8* %dst.data, i32 %dst.len, i32 %dst.cap, i8* %src.data, i32 %src.len, i32 %src.cap, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
-  %copy.n = call i32 @runtime.sliceCopy(i8* %dst.data, i8* %src.data, i32 %dst.len, i32 %src.len, i32 1, i8* undef, i8* null)
+  %copy.n = call i32 @runtime.sliceCopy(i8* %dst.data, i8* %src.data, i32 %dst.len, i32 %src.len, i32 1, i8* undef, i8* null) #0
  ret void
 }

 declare i32 @runtime.sliceCopy(i8* nocapture writeonly, i8* nocapture readonly, i32, i32, i32, i8*, i8*)

-define hidden void @main.closeBuiltinGoroutine(%runtime.channel* dereferenceable_or_null(32) %ch, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden void @main.closeBuiltinGoroutine(%runtime.channel* dereferenceable_or_null(32) %ch, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
-  call void @runtime.chanClose(%runtime.channel* %ch, i8* undef, i8* null)
+  call void @runtime.chanClose(%runtime.channel* %ch, i8* undef, i8* null) #0
  ret void
 }

 declare void @runtime.chanClose(%runtime.channel* dereferenceable_or_null(32), i8*, i8*)

-attributes #0 = { "tinygo-gowrapper"="main.regularFunction" }
-attributes #1 = { "tinygo-gowrapper"="main.inlineFunctionGoroutine$1" }
-attributes #2 = { "tinygo-gowrapper"="main.closureFunctionGoroutine$1" }
-attributes #3 = { "tinygo-gowrapper" }
+; Function Attrs: nounwind
+define hidden void @main.startInterfaceMethod(i32 %itf.typecode, i8* %itf.value, i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  %0 = call i8* @runtime.alloc(i32 16, i8* null, i8* undef, i8* null) #0
+  %1 = bitcast i8* %0 to i8**
+  store i8* %itf.value, i8** %1, align 4
+  %2 = getelementptr inbounds i8, i8* %0, i32 4
+  %.repack = bitcast i8* %2 to i8**
+  store i8* getelementptr inbounds ([4 x i8], [4 x i8]* @"main$string", i32 0, i32 0), i8** %.repack, align 4
+  %.repack1 = getelementptr inbounds i8, i8* %0, i32 8
+  %3 = bitcast i8* %.repack1 to i32*
+  store i32 4, i32* %3, align 4
+  %4 = getelementptr inbounds i8, i8* %0, i32 12
+  %5 = bitcast i8* %4 to i32*
+  store i32 %itf.typecode, i32* %5, align 4
+  %stacksize = call i32 @"internal/task.getGoroutineStackSize"(i32 ptrtoint (void (i8*)* @"interface:{Print:func:{basic:string}{}}.Print$invoke$gowrapper" to i32), i8* undef, i8* undef) #0
+  call void @"internal/task.start"(i32 ptrtoint (void (i8*)* @"interface:{Print:func:{basic:string}{}}.Print$invoke$gowrapper" to i32), i8* nonnull %0, i32 %stacksize, i8* undef, i8* null) #0
+  ret void
+}
+
+declare void @"interface:{Print:func:{basic:string}{}}.Print$invoke"(i8*, i8*, i32, i32, i8*, i8*) #5
+
+; Function Attrs: nounwind
+define linkonce_odr void @"interface:{Print:func:{basic:string}{}}.Print$invoke$gowrapper"(i8* %0) unnamed_addr #6 {
+entry:
+  %1 = bitcast i8* %0 to i8**
+  %2 = load i8*, i8** %1, align 4
+  %3 = getelementptr inbounds i8, i8* %0, i32 4
+  %4 = bitcast i8* %3 to i8**
+  %5 = load i8*, i8** %4, align 4
+  %6 = getelementptr inbounds i8, i8* %0, i32 8
+  %7 = bitcast i8* %6 to i32*
+  %8 = load i32, i32* %7, align 4
+  %9 = getelementptr inbounds i8, i8* %0, i32 12
+  %10 = bitcast i8* %9 to i32*
+  %11 = load i32, i32* %10, align 4
+  call void @"interface:{Print:func:{basic:string}{}}.Print$invoke"(i8* %2, i8* %5, i32 %8, i32 %11, i8* undef, i8* undef) #0
+  ret void
+}
+
+attributes #0 = { nounwind }
+attributes #1 = { nounwind "tinygo-gowrapper"="main.regularFunction" }
+attributes #2 = { nounwind "tinygo-gowrapper"="main.inlineFunctionGoroutine$1" }
+attributes #3 = { nounwind "tinygo-gowrapper"="main.closureFunctionGoroutine$1" }
+attributes #4 = { nounwind "tinygo-gowrapper" }
+attributes #5 = { "tinygo-invoke"="reflect/methods.Print(string)" "tinygo-methods"="reflect/methods.Print(string)" }
+attributes #6 = { nounwind "tinygo-gowrapper"="interface:{Print:func:{basic:string}{}}.Print$invoke" }
@@ -0,0 +1,210 @@
+; ModuleID = 'goroutine.go'
+source_filename = "goroutine.go"
+target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
+target triple = "wasm32-unknown-wasi"
+
+%runtime.channel = type { i32, i32, i8, %runtime.channelBlockedList*, i32, i32, i32, i8* }
+%runtime.channelBlockedList = type { %runtime.channelBlockedList*, %"internal/task.Task"*, %runtime.chanSelectState*, { %runtime.channelBlockedList*, i32, i32 } }
+%"internal/task.Task" = type { %"internal/task.Task"*, i8*, i64, %"internal/task.state" }
+%"internal/task.state" = type { i32, i8*, %"internal/task.stackState", i1 }
+%"internal/task.stackState" = type { i32, i32 }
+%runtime.chanSelectState = type { %runtime.channel*, i8* }
+
+@"main$string" = internal unnamed_addr constant [4 x i8] c"test", align 1
+
+declare noalias nonnull i8* @runtime.alloc(i32, i8*, i8*, i8*)
+
+; Function Attrs: nounwind
+define hidden void @main.init(i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  ret void
+}
+
+; Function Attrs: nounwind
+define hidden void @main.regularFunctionGoroutine(i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  call void @"internal/task.start"(i32 ptrtoint (void (i8*)* @"main.regularFunction$gowrapper" to i32), i8* nonnull inttoptr (i32 5 to i8*), i32 8192, i8* undef, i8* null) #0
+  ret void
+}
+
+declare void @main.regularFunction(i32, i8*, i8*)
+
+declare void @runtime.deadlock(i8*, i8*)
+
+; Function Attrs: nounwind
+define linkonce_odr void @"main.regularFunction$gowrapper"(i8* %0) unnamed_addr #1 {
+entry:
+  %unpack.int = ptrtoint i8* %0 to i32
+  call void @main.regularFunction(i32 %unpack.int, i8* undef, i8* undef) #0
+  call void @runtime.deadlock(i8* undef, i8* undef) #0
+  unreachable
+}
+
+declare void @"internal/task.start"(i32, i8*, i32, i8*, i8*)
+
+; Function Attrs: nounwind
+define hidden void @main.inlineFunctionGoroutine(i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  call void @"internal/task.start"(i32 ptrtoint (void (i8*)* @"main.inlineFunctionGoroutine$1$gowrapper" to i32), i8* nonnull inttoptr (i32 5 to i8*), i32 8192, i8* undef, i8* null) #0
+  ret void
+}
+
+; Function Attrs: nounwind
+define hidden void @"main.inlineFunctionGoroutine$1"(i32 %x, i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  ret void
+}
+
+; Function Attrs: nounwind
+define linkonce_odr void @"main.inlineFunctionGoroutine$1$gowrapper"(i8* %0) unnamed_addr #2 {
+entry:
+  %unpack.int = ptrtoint i8* %0 to i32
+  call void @"main.inlineFunctionGoroutine$1"(i32 %unpack.int, i8* undef, i8* undef)
+  call void @runtime.deadlock(i8* undef, i8* undef) #0
+  unreachable
+}
+
+; Function Attrs: nounwind
+define hidden void @main.closureFunctionGoroutine(i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  %n = call i8* @runtime.alloc(i32 4, i8* nonnull inttoptr (i32 3 to i8*), i8* undef, i8* null) #0
+  %0 = bitcast i8* %n to i32*
+  store i32 3, i32* %0, align 4
+  %1 = call i8* @runtime.alloc(i32 8, i8* null, i8* undef, i8* null) #0
+  %2 = bitcast i8* %1 to i32*
+  store i32 5, i32* %2, align 4
+  %3 = getelementptr inbounds i8, i8* %1, i32 4
+  %4 = bitcast i8* %3 to i8**
+  store i8* %n, i8** %4, align 4
+  call void @"internal/task.start"(i32 ptrtoint (void (i8*)* @"main.closureFunctionGoroutine$1$gowrapper" to i32), i8* nonnull %1, i32 8192, i8* undef, i8* null) #0
+  %5 = load i32, i32* %0, align 4
+  call void @runtime.printint32(i32 %5, i8* undef, i8* null) #0
+  ret void
+}
+
+; Function Attrs: nounwind
+define hidden void @"main.closureFunctionGoroutine$1"(i32 %x, i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  %unpack.ptr = bitcast i8* %context to i32*
+  store i32 7, i32* %unpack.ptr, align 4
+  ret void
+}
+
+; Function Attrs: nounwind
+define linkonce_odr void @"main.closureFunctionGoroutine$1$gowrapper"(i8* %0) unnamed_addr #3 {
+entry:
+  %1 = bitcast i8* %0 to i32*
+  %2 = load i32, i32* %1, align 4
+  %3 = getelementptr inbounds i8, i8* %0, i32 4
+  %4 = bitcast i8* %3 to i8**
+  %5 = load i8*, i8** %4, align 4
+  call void @"main.closureFunctionGoroutine$1"(i32 %2, i8* %5, i8* undef)
+  call void @runtime.deadlock(i8* undef, i8* undef) #0
+  unreachable
+}
+
+declare void @runtime.printint32(i32, i8*, i8*)
+
+; Function Attrs: nounwind
+define hidden void @main.funcGoroutine(i8* %fn.context, void ()* %fn.funcptr, i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  %0 = call i8* @runtime.alloc(i32 12, i8* null, i8* undef, i8* null) #0
+  %1 = bitcast i8* %0 to i32*
+  store i32 5, i32* %1, align 4
+  %2 = getelementptr inbounds i8, i8* %0, i32 4
+  %3 = bitcast i8* %2 to i8**
+  store i8* %fn.context, i8** %3, align 4
+  %4 = getelementptr inbounds i8, i8* %0, i32 8
+  %5 = bitcast i8* %4 to void ()**
+  store void ()* %fn.funcptr, void ()** %5, align 4
+  call void @"internal/task.start"(i32 ptrtoint (void (i8*)* @main.funcGoroutine.gowrapper to i32), i8* nonnull %0, i32 8192, i8* undef, i8* null) #0
+  ret void
+}
+
+; Function Attrs: nounwind
+define linkonce_odr void @main.funcGoroutine.gowrapper(i8* %0) unnamed_addr #4 {
+entry:
+  %1 = bitcast i8* %0 to i32*
+  %2 = load i32, i32* %1, align 4
+  %3 = getelementptr inbounds i8, i8* %0, i32 4
+  %4 = bitcast i8* %3 to i8**
+  %5 = load i8*, i8** %4, align 4
+  %6 = getelementptr inbounds i8, i8* %0, i32 8
+  %7 = bitcast i8* %6 to void (i32, i8*, i8*)**
+  %8 = load void (i32, i8*, i8*)*, void (i32, i8*, i8*)** %7, align 4
+  call void %8(i32 %2, i8* %5, i8* undef) #0
+  call void @runtime.deadlock(i8* undef, i8* undef) #0
+  unreachable
+}
+
+; Function Attrs: nounwind
+define hidden void @main.recoverBuiltinGoroutine(i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  ret void
+}
+
+; Function Attrs: nounwind
+define hidden void @main.copyBuiltinGoroutine(i8* %dst.data, i32 %dst.len, i32 %dst.cap, i8* %src.data, i32 %src.len, i32 %src.cap, i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  %copy.n = call i32 @runtime.sliceCopy(i8* %dst.data, i8* %src.data, i32 %dst.len, i32 %src.len, i32 1, i8* undef, i8* null) #0
+  ret void
+}
+
+declare i32 @runtime.sliceCopy(i8* nocapture writeonly, i8* nocapture readonly, i32, i32, i32, i8*, i8*)
+
+; Function Attrs: nounwind
+define hidden void @main.closeBuiltinGoroutine(%runtime.channel* dereferenceable_or_null(32) %ch, i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  call void @runtime.chanClose(%runtime.channel* %ch, i8* undef, i8* null) #0
+  ret void
+}
+
+declare void @runtime.chanClose(%runtime.channel* dereferenceable_or_null(32), i8*, i8*)
+
+; Function Attrs: nounwind
+define hidden void @main.startInterfaceMethod(i32 %itf.typecode, i8* %itf.value, i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  %0 = call i8* @runtime.alloc(i32 16, i8* null, i8* undef, i8* null) #0
+  %1 = bitcast i8* %0 to i8**
+  store i8* %itf.value, i8** %1, align 4
+  %2 = getelementptr inbounds i8, i8* %0, i32 4
+  %.repack = bitcast i8* %2 to i8**
+  store i8* getelementptr inbounds ([4 x i8], [4 x i8]* @"main$string", i32 0, i32 0), i8** %.repack, align 4
+  %.repack1 = getelementptr inbounds i8, i8* %0, i32 8
+  %3 = bitcast i8* %.repack1 to i32*
+  store i32 4, i32* %3, align 4
+  %4 = getelementptr inbounds i8, i8* %0, i32 12
+  %5 = bitcast i8* %4 to i32*
+  store i32 %itf.typecode, i32* %5, align 4
+  call void @"internal/task.start"(i32 ptrtoint (void (i8*)* @"interface:{Print:func:{basic:string}{}}.Print$invoke$gowrapper" to i32), i8* nonnull %0, i32 8192, i8* undef, i8* null) #0
+  ret void
+}
+
+declare void @"interface:{Print:func:{basic:string}{}}.Print$invoke"(i8*, i8*, i32, i32, i8*, i8*) #5
+
+; Function Attrs: nounwind
+define linkonce_odr void @"interface:{Print:func:{basic:string}{}}.Print$invoke$gowrapper"(i8* %0) unnamed_addr #6 {
+entry:
+  %1 = bitcast i8* %0 to i8**
+  %2 = load i8*, i8** %1, align 4
+  %3 = getelementptr inbounds i8, i8* %0, i32 4
+  %4 = bitcast i8* %3 to i8**
+  %5 = load i8*, i8** %4, align 4
+  %6 = getelementptr inbounds i8, i8* %0, i32 8
+  %7 = bitcast i8* %6 to i32*
+  %8 = load i32, i32* %7, align 4
+  %9 = getelementptr inbounds i8, i8* %0, i32 12
+  %10 = bitcast i8* %9 to i32*
+  %11 = load i32, i32* %10, align 4
+  call void @"interface:{Print:func:{basic:string}{}}.Print$invoke"(i8* %2, i8* %5, i32 %8, i32 %11, i8* undef, i8* undef) #0
+  call void @runtime.deadlock(i8* undef, i8* undef) #0
+  unreachable
+}
+
+attributes #0 = { nounwind }
+attributes #1 = { nounwind "tinygo-gowrapper"="main.regularFunction" }
+attributes #2 = { nounwind "tinygo-gowrapper"="main.inlineFunctionGoroutine$1" }
+attributes #3 = { nounwind "tinygo-gowrapper"="main.closureFunctionGoroutine$1" }
+attributes #4 = { nounwind "tinygo-gowrapper" }
+attributes #5 = { "tinygo-invoke"="reflect/methods.Print(string)" "tinygo-methods"="reflect/methods.Print(string)" }
+attributes #6 = { nounwind "tinygo-gowrapper"="interface:{Print:func:{basic:string}{}}.Print$invoke" }
@@ -1,7 +1,7 @@
 ; ModuleID = 'goroutine.go'
 source_filename = "goroutine.go"
 target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
-target triple = "wasm32--wasi"
+target triple = "wasm32-unknown-wasi"

 %runtime.funcValueWithSignature = type { i32, i8* }
 %runtime.channel = type { i32, i32, i8, %runtime.channelBlockedList*, i32, i32, i32, i8* }
@@ -10,21 +10,24 @@ target triple = "wasm32--wasi"
 %"internal/task.state" = type { i8* }
 %runtime.chanSelectState = type { %runtime.channel*, i8* }

-@"main.regularFunctionGoroutine$pack" = private unnamed_addr constant { i32, i8* } { i32 5, i8* undef }
-@"main.inlineFunctionGoroutine$pack" = private unnamed_addr constant { i32, i8* } { i32 5, i8* undef }
+@"main$pack" = internal unnamed_addr constant { i32, i8* } { i32 5, i8* undef }
+@"main$pack.1" = internal unnamed_addr constant { i32, i8* } { i32 5, i8* undef }
@"reflect/types.funcid:func:{basic:int}{}" = external constant i8
@"main.closureFunctionGoroutine$1$withSignature" = linkonce_odr constant %runtime.funcValueWithSignature { i32 ptrtoint (void (i32, i8*, i8*)* @"main.closureFunctionGoroutine$1" to i32), i8* @"reflect/types.funcid:func:{basic:int}{}" }
+@"main$string" = internal unnamed_addr constant [4 x i8] c"test", align 1

-declare noalias nonnull i8* @runtime.alloc(i32, i8*, i8*)
+declare noalias nonnull i8* @runtime.alloc(i32, i8*, i8*, i8*)

-define hidden void @main.init(i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden void @main.init(i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  ret void
 }

-define hidden void @main.regularFunctionGoroutine(i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden void @main.regularFunctionGoroutine(i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
-  call void @"internal/task.start"(i32 ptrtoint (void (i32, i8*, i8*)* @main.regularFunction to i32), i8* bitcast ({ i32, i8* }* @"main.regularFunctionGoroutine$pack" to i8*), i32 undef, i8* undef, i8* null)
+  call void @"internal/task.start"(i32 ptrtoint (void (i32, i8*, i8*)* @main.regularFunction to i32), i8* bitcast ({ i32, i8* }* @"main$pack" to i8*), i32 undef, i8* undef, i8* null) #0
  ret void
 }

@@ -32,35 +35,39 @@ declare void @main.regularFunction(i32, i8*, i8*)

 declare void @"internal/task.start"(i32, i8*, i32, i8*, i8*)

-define hidden void @main.inlineFunctionGoroutine(i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden void @main.inlineFunctionGoroutine(i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
-  call void @"internal/task.start"(i32 ptrtoint (void (i32, i8*, i8*)* @"main.inlineFunctionGoroutine$1" to i32), i8* bitcast ({ i32, i8* }* @"main.inlineFunctionGoroutine$pack" to i8*), i32 undef, i8* undef, i8* null)
+  call void @"internal/task.start"(i32 ptrtoint (void (i32, i8*, i8*)* @"main.inlineFunctionGoroutine$1" to i32), i8* bitcast ({ i32, i8* }* @"main$pack.1" to i8*), i32 undef, i8* undef, i8* null) #0
  ret void
 }

-define hidden void @"main.inlineFunctionGoroutine$1"(i32 %x, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden void @"main.inlineFunctionGoroutine$1"(i32 %x, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  ret void
 }

-define hidden void @main.closureFunctionGoroutine(i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden void @main.closureFunctionGoroutine(i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
-  %n = call i8* @runtime.alloc(i32 4, i8* undef, i8* null)
+  %n = call i8* @runtime.alloc(i32 4, i8* nonnull inttoptr (i32 3 to i8*), i8* undef, i8* null) #0
  %0 = bitcast i8* %n to i32*
  store i32 3, i32* %0, align 4
-  %1 = call i8* @runtime.alloc(i32 8, i8* undef, i8* null)
+  %1 = call i8* @runtime.alloc(i32 8, i8* null, i8* undef, i8* null) #0
  %2 = bitcast i8* %1 to i32*
  store i32 5, i32* %2, align 4
  %3 = getelementptr inbounds i8, i8* %1, i32 4
  %4 = bitcast i8* %3 to i8**
  store i8* %n, i8** %4, align 4
-  call void @"internal/task.start"(i32 ptrtoint (void (i32, i8*, i8*)* @"main.closureFunctionGoroutine$1" to i32), i8* nonnull %1, i32 undef, i8* undef, i8* null)
+  call void @"internal/task.start"(i32 ptrtoint (void (i32, i8*, i8*)* @"main.closureFunctionGoroutine$1" to i32), i8* nonnull %1, i32 undef, i8* undef, i8* null) #0
  %5 = load i32, i32* %0, align 4
-  call void @runtime.printint32(i32 %5, i8* undef, i8* null)
+  call void @runtime.printint32(i32 %5, i8* undef, i8* null) #0
  ret void
 }

-define hidden void @"main.closureFunctionGoroutine$1"(i32 %x, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden void @"main.closureFunctionGoroutine$1"(i32 %x, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  %unpack.ptr = bitcast i8* %context to i32*
  store i32 7, i32* %unpack.ptr, align 4
@@ -69,38 +76,66 @@ entry:

 declare void @runtime.printint32(i32, i8*, i8*)

-define hidden void @main.funcGoroutine(i8* %fn.context, i32 %fn.funcptr, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden void @main.funcGoroutine(i8* %fn.context, i32 %fn.funcptr, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
-  %0 = call i32 @runtime.getFuncPtr(i8* %fn.context, i32 %fn.funcptr, i8* nonnull @"reflect/types.funcid:func:{basic:int}{}", i8* undef, i8* null)
-  %1 = call i8* @runtime.alloc(i32 8, i8* undef, i8* null)
+  %0 = call i32 @runtime.getFuncPtr(i8* %fn.context, i32 %fn.funcptr, i8* nonnull @"reflect/types.funcid:func:{basic:int}{}", i8* undef, i8* null) #0
+  %1 = call i8* @runtime.alloc(i32 8, i8* null, i8* undef, i8* null) #0
  %2 = bitcast i8* %1 to i32*
  store i32 5, i32* %2, align 4
  %3 = getelementptr inbounds i8, i8* %1, i32 4
  %4 = bitcast i8* %3 to i8**
  store i8* %fn.context, i8** %4, align 4
-  call void @"internal/task.start"(i32 %0, i8* nonnull %1, i32 undef, i8* undef, i8* null)
+  call void @"internal/task.start"(i32 %0, i8* nonnull %1, i32 undef, i8* undef, i8* null) #0
  ret void
 }

 declare i32 @runtime.getFuncPtr(i8*, i32, i8* dereferenceable_or_null(1), i8*, i8*)

-define hidden void @main.recoverBuiltinGoroutine(i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden void @main.recoverBuiltinGoroutine(i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  ret void
 }

-define hidden void @main.copyBuiltinGoroutine(i8* %dst.data, i32 %dst.len, i32 %dst.cap, i8* %src.data, i32 %src.len, i32 %src.cap, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden void @main.copyBuiltinGoroutine(i8* %dst.data, i32 %dst.len, i32 %dst.cap, i8* %src.data, i32 %src.len, i32 %src.cap, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
-  %copy.n = call i32 @runtime.sliceCopy(i8* %dst.data, i8* %src.data, i32 %dst.len, i32 %src.len, i32 1, i8* undef, i8* null)
+  %copy.n = call i32 @runtime.sliceCopy(i8* %dst.data, i8* %src.data, i32 %dst.len, i32 %src.len, i32 1, i8* undef, i8* null) #0
  ret void
 }

 declare i32 @runtime.sliceCopy(i8* nocapture writeonly, i8* nocapture readonly, i32, i32, i32, i8*, i8*)

-define hidden void @main.closeBuiltinGoroutine(%runtime.channel* dereferenceable_or_null(32) %ch, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden void @main.closeBuiltinGoroutine(%runtime.channel* dereferenceable_or_null(32) %ch, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
-  call void @runtime.chanClose(%runtime.channel* %ch, i8* undef, i8* null)
+  call void @runtime.chanClose(%runtime.channel* %ch, i8* undef, i8* null) #0
  ret void
 }

 declare void @runtime.chanClose(%runtime.channel* dereferenceable_or_null(32), i8*, i8*)
+
+; Function Attrs: nounwind
+define hidden void @main.startInterfaceMethod(i32 %itf.typecode, i8* %itf.value, i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  %0 = call i8* @runtime.alloc(i32 16, i8* null, i8* undef, i8* null) #0
+  %1 = bitcast i8* %0 to i8**
+  store i8* %itf.value, i8** %1, align 4
+  %2 = getelementptr inbounds i8, i8* %0, i32 4
+  %.repack = bitcast i8* %2 to i8**
+  store i8* getelementptr inbounds ([4 x i8], [4 x i8]* @"main$string", i32 0, i32 0), i8** %.repack, align 4
+  %.repack1 = getelementptr inbounds i8, i8* %0, i32 8
+  %3 = bitcast i8* %.repack1 to i32*
+  store i32 4, i32* %3, align 4
+  %4 = getelementptr inbounds i8, i8* %0, i32 12
+  %5 = bitcast i8* %4 to i32*
+  store i32 %itf.typecode, i32* %5, align 4
+  call void @"internal/task.start"(i32 ptrtoint (void (i8*, i8*, i32, i32, i8*, i8*)* @"interface:{Print:func:{basic:string}{}}.Print$invoke" to i32), i8* nonnull %0, i32 undef, i8* undef, i8* null) #0
+  ret void
+}
+
+declare void @"interface:{Print:func:{basic:string}{}}.Print$invoke"(i8*, i8*, i32, i32, i8*, i8*) #1
+
+attributes #0 = { nounwind }
+attributes #1 = { "tinygo-invoke"="reflect/methods.Print(string)" "tinygo-methods"="reflect/methods.Print(string)" }
@@ -41,3 +41,11 @@ func closeBuiltinGoroutine(ch chan int) {
 }

 func regularFunction(x int)
+
+type simpleInterface interface {
+	Print(string)
+}
+
+func startInterfaceMethod(itf simpleInterface) {
+	go itf.Print("test")
+}
@@ -49,6 +49,15 @@ func isStringer(itf interface{}) bool {
 	return ok
 }

+type fooInterface interface {
+	String() string
+	foo(int) byte
+}
+
+func callFooMethod(itf fooInterface) uint8 {
+	return itf.foo(3)
+}
+
 func callErrorMethod(itf error) string {
 	return itf.Error()
 }
@@ -1,58 +1,67 @@
 ; ModuleID = 'interface.go'
 source_filename = "interface.go"
 target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
-target triple = "wasm32--wasi"
+target triple = "wasm32-unknown-wasi"

-%runtime.typecodeID = type { %runtime.typecodeID*, i32, %runtime.interfaceMethodInfo*, %runtime.typecodeID* }
+%runtime.typecodeID = type { %runtime.typecodeID*, i32, %runtime.interfaceMethodInfo*, %runtime.typecodeID*, i32 }
 %runtime.interfaceMethodInfo = type { i8*, i32 }
 %runtime._interface = type { i32, i8* }
 %runtime._string = type { i8*, i32 }

-@"reflect/types.type:basic:int" = linkonce_odr constant %runtime.typecodeID { %runtime.typecodeID* null, i32 0, %runtime.interfaceMethodInfo* null, %runtime.typecodeID* @"reflect/types.type:pointer:basic:int" }
-@"reflect/types.type:pointer:basic:int" = linkonce_odr constant %runtime.typecodeID { %runtime.typecodeID* @"reflect/types.type:basic:int", i32 0, %runtime.interfaceMethodInfo* null, %runtime.typecodeID* null }
-@"reflect/types.type:pointer:named:error" = linkonce_odr constant %runtime.typecodeID { %runtime.typecodeID* @"reflect/types.type:named:error", i32 0, %runtime.interfaceMethodInfo* null, %runtime.typecodeID* null }
-@"reflect/types.type:named:error" = linkonce_odr constant %runtime.typecodeID { %runtime.typecodeID* @"reflect/types.type:interface:{Error:func:{}{basic:string}}", i32 0, %runtime.interfaceMethodInfo* null, %runtime.typecodeID* @"reflect/types.type:pointer:named:error" }
-@"reflect/types.type:interface:{Error:func:{}{basic:string}}" = linkonce_odr constant %runtime.typecodeID { %runtime.typecodeID* bitcast ([1 x i8*]* @"reflect/types.interface:interface{Error() string}$interface" to %runtime.typecodeID*), i32 0, %runtime.interfaceMethodInfo* null, %runtime.typecodeID* @"reflect/types.type:pointer:interface:{Error:func:{}{basic:string}}" }
+@"reflect/types.type:basic:int" = linkonce_odr constant %runtime.typecodeID { %runtime.typecodeID* null, i32 0, %runtime.interfaceMethodInfo* null, %runtime.typecodeID* @"reflect/types.type:pointer:basic:int", i32 0 }
+@"reflect/types.type:pointer:basic:int" = linkonce_odr constant %runtime.typecodeID { %runtime.typecodeID* @"reflect/types.type:basic:int", i32 0, %runtime.interfaceMethodInfo* null, %runtime.typecodeID* null, i32 0 }
+@"reflect/types.type:pointer:named:error" = linkonce_odr constant %runtime.typecodeID { %runtime.typecodeID* @"reflect/types.type:named:error", i32 0, %runtime.interfaceMethodInfo* null, %runtime.typecodeID* null, i32 0 }
+@"reflect/types.type:named:error" = linkonce_odr constant %runtime.typecodeID { %runtime.typecodeID* @"reflect/types.type:interface:{Error:func:{}{basic:string}}", i32 0, %runtime.interfaceMethodInfo* null, %runtime.typecodeID* @"reflect/types.type:pointer:named:error", i32 ptrtoint (i1 (i32)* @"interface:{Error:func:{}{basic:string}}.$typeassert" to i32) }
+@"reflect/types.type:interface:{Error:func:{}{basic:string}}" = linkonce_odr constant %runtime.typecodeID { %runtime.typecodeID* bitcast ([1 x i8*]* @"reflect/types.interface:interface{Error() string}$interface" to %runtime.typecodeID*), i32 0, %runtime.interfaceMethodInfo* null, %runtime.typecodeID* @"reflect/types.type:pointer:interface:{Error:func:{}{basic:string}}", i32 ptrtoint (i1 (i32)* @"interface:{Error:func:{}{basic:string}}.$typeassert" to i32) }
@"reflect/methods.Error() string" = linkonce_odr constant i8 0, align 1
@"reflect/types.interface:interface{Error() string}$interface" = linkonce_odr constant [1 x i8*] [i8* @"reflect/methods.Error() string"]
-@"reflect/types.type:pointer:interface:{Error:func:{}{basic:string}}" = linkonce_odr constant %runtime.typecodeID { %runtime.typecodeID* @"reflect/types.type:interface:{Error:func:{}{basic:string}}", i32 0, %runtime.interfaceMethodInfo* null, %runtime.typecodeID* null }
-@"reflect/types.type:pointer:interface:{String:func:{}{basic:string}}" = linkonce_odr constant %runtime.typecodeID { %runtime.typecodeID* @"reflect/types.type:interface:{String:func:{}{basic:string}}", i32 0, %runtime.interfaceMethodInfo* null, %runtime.typecodeID* null }
-@"reflect/types.type:interface:{String:func:{}{basic:string}}" = linkonce_odr constant %runtime.typecodeID { %runtime.typecodeID* bitcast ([1 x i8*]* @"reflect/types.interface:interface{String() string}$interface" to %runtime.typecodeID*), i32 0, %runtime.interfaceMethodInfo* null, %runtime.typecodeID* @"reflect/types.type:pointer:interface:{String:func:{}{basic:string}}" }
+@"reflect/types.type:pointer:interface:{Error:func:{}{basic:string}}" = linkonce_odr constant %runtime.typecodeID { %runtime.typecodeID* @"reflect/types.type:interface:{Error:func:{}{basic:string}}", i32 0, %runtime.interfaceMethodInfo* null, %runtime.typecodeID* null, i32 0 }
+@"reflect/types.type:pointer:interface:{String:func:{}{basic:string}}" = linkonce_odr constant %runtime.typecodeID { %runtime.typecodeID* @"reflect/types.type:interface:{String:func:{}{basic:string}}", i32 0, %runtime.interfaceMethodInfo* null, %runtime.typecodeID* null, i32 0 }
+@"reflect/types.type:interface:{String:func:{}{basic:string}}" = linkonce_odr constant %runtime.typecodeID { %runtime.typecodeID* bitcast ([1 x i8*]* @"reflect/types.interface:interface{String() string}$interface" to %runtime.typecodeID*), i32 0, %runtime.interfaceMethodInfo* null, %runtime.typecodeID* @"reflect/types.type:pointer:interface:{String:func:{}{basic:string}}", i32 ptrtoint (i1 (i32)* @"interface:{String:func:{}{basic:string}}.$typeassert" to i32) }
@"reflect/methods.String() string" = linkonce_odr constant i8 0, align 1
@"reflect/types.interface:interface{String() string}$interface" = linkonce_odr constant [1 x i8*] [i8* @"reflect/methods.String() string"]
@"reflect/types.typeid:basic:int" = external constant i8
-@"error$interface" = linkonce_odr constant [1 x i8*] [i8* @"reflect/methods.Error() string"]

-declare noalias nonnull i8* @runtime.alloc(i32, i8*, i8*)
+declare noalias nonnull i8* @runtime.alloc(i32, i8*, i8*, i8*)

-define hidden void @main.init(i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden void @main.init(i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  ret void
 }

-define hidden %runtime._interface @main.simpleType(i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden %runtime._interface @main.simpleType(i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  ret %runtime._interface { i32 ptrtoint (%runtime.typecodeID* @"reflect/types.type:basic:int" to i32), i8* null }
 }

-define hidden %runtime._interface @main.pointerType(i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden %runtime._interface @main.pointerType(i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  ret %runtime._interface { i32 ptrtoint (%runtime.typecodeID* @"reflect/types.type:pointer:basic:int" to i32), i8* null }
 }

-define hidden %runtime._interface @main.interfaceType(i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden %runtime._interface @main.interfaceType(i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  ret %runtime._interface { i32 ptrtoint (%runtime.typecodeID* @"reflect/types.type:pointer:named:error" to i32), i8* null }
 }

-define hidden %runtime._interface @main.anonymousInterfaceType(i8* %context, i8* %parentHandle) unnamed_addr {
+declare i1 @"interface:{Error:func:{}{basic:string}}.$typeassert"(i32) #1
+
+; Function Attrs: nounwind
+define hidden %runtime._interface @main.anonymousInterfaceType(i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  ret %runtime._interface { i32 ptrtoint (%runtime.typecodeID* @"reflect/types.type:pointer:interface:{String:func:{}{basic:string}}" to i32), i8* null }
 }

-define hidden i1 @main.isInt(i32 %itf.typecode, i8* %itf.value, i8* %context, i8* %parentHandle) unnamed_addr {
+declare i1 @"interface:{String:func:{}{basic:string}}.$typeassert"(i32) #2
+
+; Function Attrs: nounwind
+define hidden i1 @main.isInt(i32 %itf.typecode, i8* %itf.value, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
-  %typecode = call i1 @runtime.typeAssert(i32 %itf.typecode, i8* nonnull @"reflect/types.typeid:basic:int", i8* undef, i8* null)
+  %typecode = call i1 @runtime.typeAssert(i32 %itf.typecode, i8* nonnull @"reflect/types.typeid:basic:int", i8* undef, i8* null) #0
  br i1 %typecode, label %typeassert.ok, label %typeassert.next

 typeassert.ok:                                    ; preds = %entry
@@ -64,9 +73,10 @@ typeassert.next:                                  ; preds = %typeassert.ok, %ent

 declare i1 @runtime.typeAssert(i32, i8* dereferenceable_or_null(1), i8*, i8*)

-define hidden i1 @main.isError(i32 %itf.typecode, i8* %itf.value, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden i1 @main.isError(i32 %itf.typecode, i8* %itf.value, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
-  %0 = call i1 @runtime.interfaceImplements(i32 %itf.typecode, i8** getelementptr inbounds ([1 x i8*], [1 x i8*]* @"error$interface", i32 0, i32 0), i8* undef, i8* null)
+  %0 = call i1 @"interface:{Error:func:{}{basic:string}}.$typeassert"(i32 %itf.typecode) #0
  br i1 %0, label %typeassert.ok, label %typeassert.next

 typeassert.ok:                                    ; preds = %entry
@@ -76,11 +86,10 @@ typeassert.next:                                  ; preds = %typeassert.ok, %ent
  ret i1 %0
 }

-declare i1 @runtime.interfaceImplements(i32, i8** dereferenceable_or_null(4), i8*, i8*)
-
-define hidden i1 @main.isStringer(i32 %itf.typecode, i8* %itf.value, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden i1 @main.isStringer(i32 %itf.typecode, i8* %itf.value, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
-  %0 = call i1 @runtime.interfaceImplements(i32 %itf.typecode, i8** getelementptr inbounds ([1 x i8*], [1 x i8*]* @"reflect/types.interface:interface{String() string}$interface", i32 0, i32 0), i8* undef, i8* null)
+  %0 = call i1 @"interface:{String:func:{}{basic:string}}.$typeassert"(i32 %itf.typecode) #0
  br i1 %0, label %typeassert.ok, label %typeassert.next

 typeassert.ok:                                    ; preds = %entry
@@ -90,12 +99,26 @@ typeassert.next:                                  ; preds = %typeassert.ok, %ent
  ret i1 %0
 }

-define hidden %runtime._string @main.callErrorMethod(i32 %itf.typecode, i8* %itf.value, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden i8 @main.callFooMethod(i32 %itf.typecode, i8* %itf.value, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
-  %invoke.func = call i32 @runtime.interfaceMethod(i32 %itf.typecode, i8** getelementptr inbounds ([1 x i8*], [1 x i8*]* @"error$interface", i32 0, i32 0), i8* nonnull @"reflect/methods.Error() string", i8* undef, i8* null)
-  %invoke.func.cast = inttoptr i32 %invoke.func to %runtime._string (i8*, i8*, i8*)*
-  %0 = call %runtime._string %invoke.func.cast(i8* %itf.value, i8* undef, i8* undef)
+  %0 = call i8 @"interface:{String:func:{}{basic:string},main.foo:func:{basic:int}{basic:uint8}}.foo$invoke"(i8* %itf.value, i32 3, i32 %itf.typecode, i8* undef, i8* undef) #0
+  ret i8 %0
+}
+
+declare i8 @"interface:{String:func:{}{basic:string},main.foo:func:{basic:int}{basic:uint8}}.foo$invoke"(i8*, i32, i32, i8*, i8*) #3
+
+; Function Attrs: nounwind
+define hidden %runtime._string @main.callErrorMethod(i32 %itf.typecode, i8* %itf.value, i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  %0 = call %runtime._string @"interface:{Error:func:{}{basic:string}}.Error$invoke"(i8* %itf.value, i32 %itf.typecode, i8* undef, i8* undef) #0
  ret %runtime._string %0
 }

-declare i32 @runtime.interfaceMethod(i32, i8** dereferenceable_or_null(4), i8* dereferenceable_or_null(1), i8*, i8*)
+declare %runtime._string @"interface:{Error:func:{}{basic:string}}.Error$invoke"(i8*, i32, i8*, i8*) #4
+
+attributes #0 = { nounwind }
+attributes #1 = { "tinygo-methods"="reflect/methods.Error() string" }
+attributes #2 = { "tinygo-methods"="reflect/methods.String() string" }
+attributes #3 = { "tinygo-invoke"="main.$methods.foo(int) byte" "tinygo-methods"="reflect/methods.String() string; main.$methods.foo(int) byte" }
+attributes #4 = { "tinygo-invoke"="reflect/methods.Error() string" "tinygo-methods"="reflect/methods.Error() string" }
@@ -0,0 +1,32 @@
+; ModuleID = 'intrinsics.go'
+source_filename = "intrinsics.go"
+target datalayout = "e-m:e-p:32:32-Fi8-i64:64-v128:64:128-a:0:32-n32-S64"
+target triple = "thumbv7m-unknown-unknown-eabi"
+
+declare noalias nonnull i8* @runtime.alloc(i32, i8*, i8*, i8*)
+
+; Function Attrs: nounwind
+define hidden void @main.init(i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  ret void
+}
+
+; Function Attrs: nounwind
+define hidden double @main.mySqrt(double %x, i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  %0 = call double @math.Sqrt(double %x, i8* undef, i8* undef) #0
+  ret double %0
+}
+
+declare double @math.Sqrt(double, i8*, i8*)
+
+; Function Attrs: nounwind
+define hidden double @main.myTrunc(double %x, i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  %0 = call double @math.Trunc(double %x, i8* undef, i8* undef) #0
+  ret double %0
+}
+
+declare double @math.Trunc(double, i8*, i8*)
+
+attributes #0 = { nounwind }
@@ -0,0 +1,35 @@
+; ModuleID = 'intrinsics.go'
+source_filename = "intrinsics.go"
+target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
+target triple = "wasm32-unknown-wasi"
+
+declare noalias nonnull i8* @runtime.alloc(i32, i8*, i8*, i8*)
+
+; Function Attrs: nounwind
+define hidden void @main.init(i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  ret void
+}
+
+; Function Attrs: nounwind
+define hidden double @main.mySqrt(double %x, i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  %0 = call double @llvm.sqrt.f64(double %x)
+  ret double %0
+}
+
+; Function Attrs: nounwind readnone speculatable willreturn
+declare double @llvm.sqrt.f64(double) #1
+
+; Function Attrs: nounwind
+define hidden double @main.myTrunc(double %x, i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  %0 = call double @llvm.trunc.f64(double %x)
+  ret double %0
+}
+
+; Function Attrs: nounwind readnone speculatable willreturn
+declare double @llvm.trunc.f64(double) #1
+
+attributes #0 = { nounwind }
+attributes #1 = { nounwind readnone speculatable willreturn }
@@ -0,0 +1,14 @@
+package main
+
+// Test how intrinsics are lowered: either as regular calls to the math
+// functions or as LLVM builtins (such as llvm.sqrt.f64).
+
+import "math"
+
+func mySqrt(x float64) float64 {
+	return math.Sqrt(x)
+}
+
+func myTrunc(x float64) float64 {
+	return math.Trunc(x)
+}
@@ -1,51 +1,61 @@
 ; ModuleID = 'pointer.go'
 source_filename = "pointer.go"
 target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
-target triple = "wasm32--wasi"
+target triple = "wasm32-unknown-wasi"

-declare noalias nonnull i8* @runtime.alloc(i32, i8*, i8*)
+declare noalias nonnull i8* @runtime.alloc(i32, i8*, i8*, i8*)

-define hidden void @main.init(i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden void @main.init(i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  ret void
 }

-define hidden [0 x i32] @main.pointerDerefZero([0 x i32]* %x, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden [0 x i32] @main.pointerDerefZero([0 x i32]* %x, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  ret [0 x i32] zeroinitializer
 }

-define hidden i32* @main.pointerCastFromUnsafe(i8* %x, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden i32* @main.pointerCastFromUnsafe(i8* %x, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  %0 = bitcast i8* %x to i32*
  ret i32* %0
 }

-define hidden i8* @main.pointerCastToUnsafe(i32* dereferenceable_or_null(4) %x, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden i8* @main.pointerCastToUnsafe(i32* dereferenceable_or_null(4) %x, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  %0 = bitcast i32* %x to i8*
  ret i8* %0
 }

-define hidden i8* @main.pointerCastToUnsafeNoop(i8* dereferenceable_or_null(1) %x, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden i8* @main.pointerCastToUnsafeNoop(i8* dereferenceable_or_null(1) %x, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  ret i8* %x
 }

-define hidden i8* @main.pointerUnsafeGEPFixedOffset(i8* dereferenceable_or_null(1) %ptr, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden i8* @main.pointerUnsafeGEPFixedOffset(i8* dereferenceable_or_null(1) %ptr, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  %0 = getelementptr inbounds i8, i8* %ptr, i32 10
  ret i8* %0
 }

-define hidden i8* @main.pointerUnsafeGEPByteOffset(i8* dereferenceable_or_null(1) %ptr, i32 %offset, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden i8* @main.pointerUnsafeGEPByteOffset(i8* dereferenceable_or_null(1) %ptr, i32 %offset, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  %0 = getelementptr inbounds i8, i8* %ptr, i32 %offset
  ret i8* %0
 }

-define hidden i32* @main.pointerUnsafeGEPIntOffset(i32* dereferenceable_or_null(4) %ptr, i32 %offset, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden i32* @main.pointerUnsafeGEPIntOffset(i32* dereferenceable_or_null(4) %ptr, i32 %offset, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  %0 = getelementptr i32, i32* %ptr, i32 %offset
  ret i32* %0
 }
+
+attributes #0 = { nounwind }
@@ -1,7 +1,7 @@
 ; ModuleID = 'pragma.go'
 source_filename = "pragma.go"
 target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
-target triple = "wasm32--wasi"
+target triple = "wasm32-unknown-wasi"

@extern_global = external global [0 x i8], align 1
@main.alignedGlobal = hidden global [4 x i32] zeroinitializer, align 32
@@ -10,50 +10,56 @@ target triple = "wasm32--wasi"
@undefinedGlobalNotInSection = external global i32, align 4
@main.multipleGlobalPragmas = hidden global i32 0, section ".global_section", align 1024

-declare noalias nonnull i8* @runtime.alloc(i32, i8*, i8*)
+declare noalias nonnull i8* @runtime.alloc(i32, i8*, i8*, i8*)

-define hidden void @main.init(i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden void @main.init(i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  ret void
 }

-define void @extern_func() #0 {
+; Function Attrs: nounwind
+define void @extern_func() #1 {
 entry:
  ret void
 }

-define hidden void @somepkg.someFunction1(i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden void @somepkg.someFunction1(i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  ret void
 }

 declare void @somepkg.someFunction2(i8*, i8*)

-; Function Attrs: inlinehint
-define hidden void @main.inlineFunc(i8* %context, i8* %parentHandle) unnamed_addr #1 {
+; Function Attrs: inlinehint nounwind
+define hidden void @main.inlineFunc(i8* %context, i8* %parentHandle) unnamed_addr #2 {
 entry:
  ret void
 }

-; Function Attrs: noinline
-define hidden void @main.noinlineFunc(i8* %context, i8* %parentHandle) unnamed_addr #2 {
+; Function Attrs: noinline nounwind
+define hidden void @main.noinlineFunc(i8* %context, i8* %parentHandle) unnamed_addr #3 {
 entry:
  ret void
 }

-define hidden void @main.functionInSection(i8* %context, i8* %parentHandle) unnamed_addr section ".special_function_section" {
+; Function Attrs: nounwind
+define hidden void @main.functionInSection(i8* %context, i8* %parentHandle) unnamed_addr #0 section ".special_function_section" {
 entry:
  ret void
 }

-define void @exportedFunctionInSection() #3 section ".special_function_section" {
+; Function Attrs: nounwind
+define void @exportedFunctionInSection() #4 section ".special_function_section" {
 entry:
  ret void
 }

 declare void @main.undefinedFunctionNotInSection(i8*, i8*)

-attributes #0 = { "wasm-export-name"="extern_func" }
-attributes #1 = { inlinehint }
-attributes #2 = { noinline }
-attributes #3 = { "wasm-export-name"="exportedFunctionInSection" }
+attributes #0 = { nounwind }
+attributes #1 = { nounwind "wasm-export-name"="extern_func" }
+attributes #2 = { inlinehint nounwind }
+attributes #3 = { noinline nounwind }
+attributes #4 = { nounwind "wasm-export-name"="exportedFunctionInSection" }
@@ -23,3 +23,21 @@ func sliceAppendSlice(ints, added []int) []int {
 func sliceCopy(dst, src []int) int {
 	return copy(dst, src)
 }
+
+// Test bounds checking in *ssa.MakeSlice instruction.
+
+func makeByteSlice(len int) []byte {
+	return make([]byte, len)
+}
+
+func makeInt16Slice(len int) []int16 {
+	return make([]int16, len)
+}
+
+func makeArraySlice(len int) [][3]byte {
+	return make([][3]byte, len) // slice with element size of 3
+}
+
+func makeInt32Slice(len int) []int32 {
+	return make([]int32, len)
+}
@@ -1,32 +1,36 @@
 ; ModuleID = 'slice.go'
 source_filename = "slice.go"
 target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
-target triple = "wasm32--wasi"
+target triple = "wasm32-unknown-wasi"

-declare noalias nonnull i8* @runtime.alloc(i32, i8*, i8*)
+declare noalias nonnull i8* @runtime.alloc(i32, i8*, i8*, i8*)

-define hidden void @main.init(i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden void @main.init(i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  ret void
 }

-define hidden i32 @main.sliceLen(i32* %ints.data, i32 %ints.len, i32 %ints.cap, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden i32 @main.sliceLen(i32* %ints.data, i32 %ints.len, i32 %ints.cap, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  ret i32 %ints.len
 }

-define hidden i32 @main.sliceCap(i32* %ints.data, i32 %ints.len, i32 %ints.cap, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden i32 @main.sliceCap(i32* %ints.data, i32 %ints.len, i32 %ints.cap, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  ret i32 %ints.cap
 }

-define hidden i32 @main.sliceElement(i32* %ints.data, i32 %ints.len, i32 %ints.cap, i32 %index, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden i32 @main.sliceElement(i32* %ints.data, i32 %ints.len, i32 %ints.cap, i32 %index, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  %.not = icmp ult i32 %index, %ints.len
  br i1 %.not, label %lookup.next, label %lookup.throw

 lookup.throw:                                     ; preds = %entry
-  call void @runtime.lookupPanic(i8* undef, i8* null)
+  call void @runtime.lookupPanic(i8* undef, i8* null) #0
  unreachable

 lookup.next:                                      ; preds = %entry
@@ -37,9 +41,10 @@ lookup.next:                                      ; preds = %entry

 declare void @runtime.lookupPanic(i8*, i8*)

-define hidden { i32*, i32, i32 } @main.sliceAppendValues(i32* %ints.data, i32 %ints.len, i32 %ints.cap, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden { i32*, i32, i32 } @main.sliceAppendValues(i32* %ints.data, i32 %ints.len, i32 %ints.cap, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
-  %varargs = call i8* @runtime.alloc(i32 12, i8* undef, i8* null)
+  %varargs = call i8* @runtime.alloc(i32 12, i8* nonnull inttoptr (i32 3 to i8*), i8* undef, i8* null) #0
  %0 = bitcast i8* %varargs to i32*
  store i32 1, i32* %0, align 4
  %1 = getelementptr inbounds i8, i8* %varargs, i32 4
@@ -49,7 +54,7 @@ entry:
  %4 = bitcast i8* %3 to i32*
  store i32 3, i32* %4, align 4
  %append.srcPtr = bitcast i32* %ints.data to i8*
-  %append.new = call { i8*, i32, i32 } @runtime.sliceAppend(i8* %append.srcPtr, i8* nonnull %varargs, i32 %ints.len, i32 %ints.cap, i32 3, i32 4, i8* undef, i8* null)
+  %append.new = call { i8*, i32, i32 } @runtime.sliceAppend(i8* %append.srcPtr, i8* nonnull %varargs, i32 %ints.len, i32 %ints.cap, i32 3, i32 4, i8* undef, i8* null) #0
  %append.newPtr = extractvalue { i8*, i32, i32 } %append.new, 0
  %append.newBuf = bitcast i8* %append.newPtr to i32*
  %append.newLen = extractvalue { i8*, i32, i32 } %append.new, 1
@@ -62,11 +67,12 @@ entry:

 declare { i8*, i32, i32 } @runtime.sliceAppend(i8*, i8* nocapture readonly, i32, i32, i32, i32, i8*, i8*)

-define hidden { i32*, i32, i32 } @main.sliceAppendSlice(i32* %ints.data, i32 %ints.len, i32 %ints.cap, i32* %added.data, i32 %added.len, i32 %added.cap, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden { i32*, i32, i32 } @main.sliceAppendSlice(i32* %ints.data, i32 %ints.len, i32 %ints.cap, i32* %added.data, i32 %added.len, i32 %added.cap, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  %append.srcPtr = bitcast i32* %ints.data to i8*
  %append.srcPtr1 = bitcast i32* %added.data to i8*
-  %append.new = call { i8*, i32, i32 } @runtime.sliceAppend(i8* %append.srcPtr, i8* %append.srcPtr1, i32 %ints.len, i32 %ints.cap, i32 %added.len, i32 4, i8* undef, i8* null)
+  %append.new = call { i8*, i32, i32 } @runtime.sliceAppend(i8* %append.srcPtr, i8* %append.srcPtr1, i32 %ints.len, i32 %ints.cap, i32 %added.len, i32 4, i8* undef, i8* null) #0
  %append.newPtr = extractvalue { i8*, i32, i32 } %append.new, 0
  %append.newBuf = bitcast i8* %append.newPtr to i32*
  %append.newLen = extractvalue { i8*, i32, i32 } %append.new, 1
@@ -77,12 +83,95 @@ entry:
  ret { i32*, i32, i32 } %2
 }

-define hidden i32 @main.sliceCopy(i32* %dst.data, i32 %dst.len, i32 %dst.cap, i32* %src.data, i32 %src.len, i32 %src.cap, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden i32 @main.sliceCopy(i32* %dst.data, i32 %dst.len, i32 %dst.cap, i32* %src.data, i32 %src.len, i32 %src.cap, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  %copy.dstPtr = bitcast i32* %dst.data to i8*
  %copy.srcPtr = bitcast i32* %src.data to i8*
-  %copy.n = call i32 @runtime.sliceCopy(i8* %copy.dstPtr, i8* %copy.srcPtr, i32 %dst.len, i32 %src.len, i32 4, i8* undef, i8* null)
+  %copy.n = call i32 @runtime.sliceCopy(i8* %copy.dstPtr, i8* %copy.srcPtr, i32 %dst.len, i32 %src.len, i32 4, i8* undef, i8* null) #0
  ret i32 %copy.n
 }

 declare i32 @runtime.sliceCopy(i8* nocapture writeonly, i8* nocapture readonly, i32, i32, i32, i8*, i8*)
+
+; Function Attrs: nounwind
+define hidden { i8*, i32, i32 } @main.makeByteSlice(i32 %len, i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  %slice.maxcap = icmp slt i32 %len, 0
+  br i1 %slice.maxcap, label %slice.throw, label %slice.next
+
+slice.throw:                                      ; preds = %entry
+  call void @runtime.slicePanic(i8* undef, i8* null) #0
+  unreachable
+
+slice.next:                                       ; preds = %entry
+  %makeslice.buf = call i8* @runtime.alloc(i32 %len, i8* nonnull inttoptr (i32 3 to i8*), i8* undef, i8* null) #0
+  %0 = insertvalue { i8*, i32, i32 } undef, i8* %makeslice.buf, 0
+  %1 = insertvalue { i8*, i32, i32 } %0, i32 %len, 1
+  %2 = insertvalue { i8*, i32, i32 } %1, i32 %len, 2
+  ret { i8*, i32, i32 } %2
+}
+
+declare void @runtime.slicePanic(i8*, i8*)
+
+; Function Attrs: nounwind
+define hidden { i16*, i32, i32 } @main.makeInt16Slice(i32 %len, i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  %slice.maxcap = icmp slt i32 %len, 0
+  br i1 %slice.maxcap, label %slice.throw, label %slice.next
+
+slice.throw:                                      ; preds = %entry
+  call void @runtime.slicePanic(i8* undef, i8* null) #0
+  unreachable
+
+slice.next:                                       ; preds = %entry
+  %makeslice.cap = shl i32 %len, 1
+  %makeslice.buf = call i8* @runtime.alloc(i32 %makeslice.cap, i8* nonnull inttoptr (i32 3 to i8*), i8* undef, i8* null) #0
+  %makeslice.array = bitcast i8* %makeslice.buf to i16*
+  %0 = insertvalue { i16*, i32, i32 } undef, i16* %makeslice.array, 0
+  %1 = insertvalue { i16*, i32, i32 } %0, i32 %len, 1
+  %2 = insertvalue { i16*, i32, i32 } %1, i32 %len, 2
+  ret { i16*, i32, i32 } %2
+}
+
+; Function Attrs: nounwind
+define hidden { [3 x i8]*, i32, i32 } @main.makeArraySlice(i32 %len, i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  %slice.maxcap = icmp ugt i32 %len, 1431655765
+  br i1 %slice.maxcap, label %slice.throw, label %slice.next
+
+slice.throw:                                      ; preds = %entry
+  call void @runtime.slicePanic(i8* undef, i8* null) #0
+  unreachable
+
+slice.next:                                       ; preds = %entry
+  %makeslice.cap = mul i32 %len, 3
+  %makeslice.buf = call i8* @runtime.alloc(i32 %makeslice.cap, i8* nonnull inttoptr (i32 3 to i8*), i8* undef, i8* null) #0
+  %makeslice.array = bitcast i8* %makeslice.buf to [3 x i8]*
+  %0 = insertvalue { [3 x i8]*, i32, i32 } undef, [3 x i8]* %makeslice.array, 0
+  %1 = insertvalue { [3 x i8]*, i32, i32 } %0, i32 %len, 1
+  %2 = insertvalue { [3 x i8]*, i32, i32 } %1, i32 %len, 2
+  ret { [3 x i8]*, i32, i32 } %2
+}
+
+; Function Attrs: nounwind
+define hidden { i32*, i32, i32 } @main.makeInt32Slice(i32 %len, i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  %slice.maxcap = icmp ugt i32 %len, 1073741823
+  br i1 %slice.maxcap, label %slice.throw, label %slice.next
+
+slice.throw:                                      ; preds = %entry
+  call void @runtime.slicePanic(i8* undef, i8* null) #0
+  unreachable
+
+slice.next:                                       ; preds = %entry
+  %makeslice.cap = shl i32 %len, 2
+  %makeslice.buf = call i8* @runtime.alloc(i32 %makeslice.cap, i8* nonnull inttoptr (i32 3 to i8*), i8* undef, i8* null) #0
+  %makeslice.array = bitcast i8* %makeslice.buf to i32*
+  %0 = insertvalue { i32*, i32, i32 } undef, i32* %makeslice.array, 0
+  %1 = insertvalue { i32*, i32, i32 } %0, i32 %len, 1
+  %2 = insertvalue { i32*, i32, i32 } %1, i32 %len, 2
+  ret { i32*, i32, i32 } %2
+}
+
+attributes #0 = { nounwind }
@@ -27,3 +27,8 @@ func stringCompareUnequal(s1, s2 string) bool {
 func stringCompareLarger(s1, s2 string) bool {
 	return s1 > s2
 }
+
+func stringLookup(s string, x uint8) byte {
+	// Test that x is correctly extended to an uint before comparison.
+	return s[x]
+}
@@ -1,41 +1,46 @@
 ; ModuleID = 'string.go'
 source_filename = "string.go"
 target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
-target triple = "wasm32--wasi"
+target triple = "wasm32-unknown-wasi"

 %runtime._string = type { i8*, i32 }

-@"main.someString$string" = internal unnamed_addr constant [3 x i8] c"foo", align 1
+@"main$string" = internal unnamed_addr constant [3 x i8] c"foo", align 1

-declare noalias nonnull i8* @runtime.alloc(i32, i8*, i8*)
+declare noalias nonnull i8* @runtime.alloc(i32, i8*, i8*, i8*)

-define hidden void @main.init(i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden void @main.init(i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  ret void
 }

-define hidden %runtime._string @main.someString(i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden %runtime._string @main.someString(i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
-  ret %runtime._string { i8* getelementptr inbounds ([3 x i8], [3 x i8]* @"main.someString$string", i32 0, i32 0), i32 3 }
+  ret %runtime._string { i8* getelementptr inbounds ([3 x i8], [3 x i8]* @"main$string", i32 0, i32 0), i32 3 }
 }

-define hidden %runtime._string @main.zeroLengthString(i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden %runtime._string @main.zeroLengthString(i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  ret %runtime._string zeroinitializer
 }

-define hidden i32 @main.stringLen(i8* %s.data, i32 %s.len, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden i32 @main.stringLen(i8* %s.data, i32 %s.len, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  ret i32 %s.len
 }

-define hidden i8 @main.stringIndex(i8* %s.data, i32 %s.len, i32 %index, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden i8 @main.stringIndex(i8* %s.data, i32 %s.len, i32 %index, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
  %.not = icmp ult i32 %index, %s.len
  br i1 %.not, label %lookup.next, label %lookup.throw

 lookup.throw:                                     ; preds = %entry
-  call void @runtime.lookupPanic(i8* undef, i8* null)
+  call void @runtime.lookupPanic(i8* undef, i8* null) #0
  unreachable

 lookup.next:                                      ; preds = %entry
@@ -46,26 +51,48 @@ lookup.next:                                      ; preds = %entry

 declare void @runtime.lookupPanic(i8*, i8*)

-define hidden i1 @main.stringCompareEqual(i8* %s1.data, i32 %s1.len, i8* %s2.data, i32 %s2.len, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden i1 @main.stringCompareEqual(i8* %s1.data, i32 %s1.len, i8* %s2.data, i32 %s2.len, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
-  %0 = call i1 @runtime.stringEqual(i8* %s1.data, i32 %s1.len, i8* %s2.data, i32 %s2.len, i8* undef, i8* null)
+  %0 = call i1 @runtime.stringEqual(i8* %s1.data, i32 %s1.len, i8* %s2.data, i32 %s2.len, i8* undef, i8* null) #0
  ret i1 %0
 }

 declare i1 @runtime.stringEqual(i8*, i32, i8*, i32, i8*, i8*)

-define hidden i1 @main.stringCompareUnequal(i8* %s1.data, i32 %s1.len, i8* %s2.data, i32 %s2.len, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden i1 @main.stringCompareUnequal(i8* %s1.data, i32 %s1.len, i8* %s2.data, i32 %s2.len, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
-  %0 = call i1 @runtime.stringEqual(i8* %s1.data, i32 %s1.len, i8* %s2.data, i32 %s2.len, i8* undef, i8* null)
+  %0 = call i1 @runtime.stringEqual(i8* %s1.data, i32 %s1.len, i8* %s2.data, i32 %s2.len, i8* undef, i8* null) #0
  %1 = xor i1 %0, true
  ret i1 %1
 }

-define hidden i1 @main.stringCompareLarger(i8* %s1.data, i32 %s1.len, i8* %s2.data, i32 %s2.len, i8* %context, i8* %parentHandle) unnamed_addr {
+; Function Attrs: nounwind
+define hidden i1 @main.stringCompareLarger(i8* %s1.data, i32 %s1.len, i8* %s2.data, i32 %s2.len, i8* %context, i8* %parentHandle) unnamed_addr #0 {
 entry:
-  %0 = call i1 @runtime.stringLess(i8* %s1.data, i32 %s1.len, i8* %s2.data, i32 %s2.len, i8* undef, i8* null)
+  %0 = call i1 @runtime.stringLess(i8* %s1.data, i32 %s1.len, i8* %s2.data, i32 %s2.len, i8* undef, i8* null) #0
  %1 = xor i1 %0, true
  ret i1 %1
 }

 declare i1 @runtime.stringLess(i8*, i32, i8*, i32, i8*, i8*)
+
+; Function Attrs: nounwind
+define hidden i8 @main.stringLookup(i8* %s.data, i32 %s.len, i8 %x, i8* %context, i8* %parentHandle) unnamed_addr #0 {
+entry:
+  %0 = zext i8 %x to i32
+  %.not = icmp ult i32 %0, %s.len
+  br i1 %.not, label %lookup.next, label %lookup.throw
+
+lookup.throw:                                     ; preds = %entry
+  call void @runtime.lookupPanic(i8* undef, i8* null) #0
+  unreachable
+
+lookup.next:                                      ; preds = %entry
+  %1 = getelementptr inbounds i8, i8* %s.data, i32 %0
+  %2 = load i8, i8* %1, align 1
+  ret i8 %2
+}
+
+attributes #0 = { nounwind }
@@ -1,16 +1,17 @@
 module github.com/tinygo-org/tinygo

-go 1.13
+go 1.15

 require (
+	github.com/aykevl/go-wasm v0.0.2-0.20211030161413-11881cb9032d // indirect
 	github.com/blakesmith/ar v0.0.0-20150311145944-8bd4349a67f2
 	github.com/chromedp/cdproto v0.0.0-20210113043257-dabd2f2e7693
 	github.com/chromedp/chromedp v0.6.4
 	github.com/google/shlex v0.0.0-20181106134648-c34317bd91bf
 	github.com/marcinbor85/gohex v0.0.0-20200531091804-343a4b548892
 	github.com/mattn/go-colorable v0.1.8
-	go.bug.st/serial v1.1.2
-	golang.org/x/sys v0.0.0-20210113181707-4bcb84eeeb78
-	golang.org/x/tools v0.0.0-20200216192241-b320d3a0f5a2
+	go.bug.st/serial v1.1.3
+	golang.org/x/sys v0.0.0-20210510120138-977fb7262007
+	golang.org/x/tools v0.1.6-0.20210813165731-45389f592fe9
 	tinygo.org/x/go-llvm v0.0.0-20210325115028-e7b85195e81c
 )
@@ -1,3 +1,5 @@
+github.com/aykevl/go-wasm v0.0.2-0.20211030161413-11881cb9032d h1:JeuI5/546naK5hpOIX+Lq5xE8rvt7uwiTp6iL+pLQgk=
+github.com/aykevl/go-wasm v0.0.2-0.20211030161413-11881cb9032d/go.mod h1:7sXyiaA0WtSogCu67R2252fQpVmJMh9JWJ9ddtGkpWw=
 github.com/blakesmith/ar v0.0.0-20150311145944-8bd4349a67f2 h1:oMCHnXa6CCCafdPDbMh/lWRhRByN0VFLvv+g+ayx1SI=
 github.com/blakesmith/ar v0.0.0-20150311145944-8bd4349a67f2/go.mod h1:PkYb9DJNAwrSvRx5DYA+gUcOIgTGVMNkfSCbZM8cWpI=
 github.com/chromedp/cdproto v0.0.0-20210113043257-dabd2f2e7693 h1:11eq/RkpaotwdF6b1TRMcdgQUPNmyFEJOB7zLvh0O/Y=
@@ -6,8 +8,8 @@ github.com/chromedp/chromedp v0.6.4 h1:Gx7ZkRyrSVmbbDDja/ieNgNGJIvElroPOyeqYQGVD
 github.com/chromedp/chromedp v0.6.4/go.mod h1:vodUdJf5dF/b8n0UBJv6NeM/QK28RjP3j+eM7fq4+84=
 github.com/chromedp/sysutil v1.0.0 h1:+ZxhTpfpZlmchB58ih/LBHX52ky7w2VhQVKQMucy3Ic=
 github.com/chromedp/sysutil v1.0.0/go.mod h1:kgWmDdq8fTzXYcKIBqIYvRRTnYb9aNS9moAV0xufSww=
-github.com/creack/goselect v0.1.1 h1:tiSSgKE1eJtxs1h/VgGQWuXUP0YS4CDIFMp6vaI1ls0=
-github.com/creack/goselect v0.1.1/go.mod h1:a/NhLweNvqIYMuxcMOuWY516Cimucms3DglDzQP3hKY=
+github.com/creack/goselect v0.1.2 h1:2DNy14+JPjRBgPzAd1thbQp4BSIihxcBf0IXhQXDRa0=
+github.com/creack/goselect v0.1.2/go.mod h1:a/NhLweNvqIYMuxcMOuWY516Cimucms3DglDzQP3hKY=
 github.com/davecgh/go-spew v1.1.0/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
 github.com/gobwas/httphead v0.1.0 h1:exrUm0f4YX0L7EBwZHuCF4GDp8aJfVeBrlLQrs6NqWU=
 github.com/gobwas/httphead v0.1.0/go.mod h1:O/RXo79gxV8G+RqlR/otEwx4Q36zl9rqC5u12GKvMCM=
@@ -30,31 +32,40 @@ github.com/mattn/go-isatty v0.0.12/go.mod h1:cbi8OIDigv2wuxKPP5vlRcQ1OAZbq2CE4Ky
 github.com/pmezard/go-difflib v1.0.0/go.mod h1:iKH77koFhYxTK1pcRnkKkqfTogsbg7gZNVY4sRDYZ/4=
 github.com/stretchr/objx v0.1.0/go.mod h1:HFkY916IF+rwdDfMAkV7OtwuqBVzrE8GR6GFx+wExME=
 github.com/stretchr/testify v1.4.0/go.mod h1:j7eGeouHqKxXV5pUuKE4zz7dFj8WfuZ+81PSLYec5m4=
-go.bug.st/serial v1.0.0 h1:ogEPzrllCsnG00EqKRjeYvPRsO7NJW6DqykzkdD6E/k=
-go.bug.st/serial v1.0.0/go.mod h1:rpXPISGjuNjPTRTcMlxi9lN6LoIPxd1ixVjBd8aSk/Q=
-go.bug.st/serial v1.1.2 h1:6xDpbta8KJ+VLRTeM8ghhxXRMLE/Lr8h9iDKwydarAY=
-go.bug.st/serial v1.1.2/go.mod h1:VmYBeyJWp5BnJ0tw2NUJHZdJTGl2ecBGABHlzRK1knY=
+github.com/yuin/goldmark v1.3.5 h1:dPmz1Snjq0kmkz159iL7S6WzdahUTHnHB5M56WFVifs=
+github.com/yuin/goldmark v1.3.5/go.mod h1:mwnBkeHKe2W/ZEtQ+71ViKU8L12m81fl3OWwC1Zlc8k=
+go.bug.st/serial v1.1.3 h1:YEBxJa9pKS9Wdg46B/jiaKbvvbUrjhZZZITfJHEJhaE=
+go.bug.st/serial v1.1.3/go.mod h1:8TT7u/SwwNIpJ8QaG4s+HTjFt9ReXs2cdOU7ZEk50Dk=
 golang.org/x/crypto v0.0.0-20190308221718-c2843e01d9a2/go.mod h1:djNgcEr1/C05ACkg1iLfiJU5Ep61QUkGW8qpdssI0+w=
 golang.org/x/crypto v0.0.0-20191011191535-87dc89f01550/go.mod h1:yigFU9vqHzYiE8UmvKecakEJjdnWj3jj499lnFckfCI=
 golang.org/x/mod v0.1.1-0.20191105210325-c90efee705ee h1:WG0RUwxtNT4qqaXX3DPA8zHFNm/D9xaBpxzHt1WcA/E=
 golang.org/x/mod v0.1.1-0.20191105210325-c90efee705ee/go.mod h1:QqPTAvyqsEbceGzBzNggFXnrqF1CaUcvgkdR5Ot7KZg=
+golang.org/x/mod v0.4.2/go.mod h1:s0Qsj1ACt9ePp/hMypM3fl4fZqREWJwdYDEqhRiZZUA=
 golang.org/x/net v0.0.0-20190404232315-eb5bcb51f2a3/go.mod h1:t9HGtf8HONx5eT2rtn7q6eTqICYqUVnKs3thJo3Qplg=
 golang.org/x/net v0.0.0-20190620200207-3b0461eec859/go.mod h1:z5CRVTTTmAJ677TzLLGU+0bjPO0LkuOLi4/5GtJWs/s=
+golang.org/x/net v0.0.0-20210405180319-a5a99cb37ef4/go.mod h1:p54w0d4576C0XHj96bSt6lcn1PtDYWL6XObtHCRCNQM=
 golang.org/x/sync v0.0.0-20190423024810-112230192c58/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
+golang.org/x/sync v0.0.0-20210220032951-036812b2e83c/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
 golang.org/x/sys v0.0.0-20190215142949-d0b11bdaac8a/go.mod h1:STP8DvDyc/dI5b8T5hshtkjS+E42TnysNCUPdjciGhY=
 golang.org/x/sys v0.0.0-20190412213103-97732733099d/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
-golang.org/x/sys v0.0.0-20191128015809-6d18c012aee9 h1:ZBzSG/7F4eNKz2L3GE9o300RX0Az1Bw5HF7PDraD+qU=
-golang.org/x/sys v0.0.0-20191128015809-6d18c012aee9/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.0.0-20200116001909-b77594299b42/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.0.0-20200223170610-d5e6a3e2c0ae/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.0.0-20200909081042-eff7692f9009/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
-golang.org/x/sys v0.0.0-20210113181707-4bcb84eeeb78 h1:nVuTkr9L6Bq62qpUqKo/RnZCFfzDBL0bYo6w9OJUqZY=
+golang.org/x/sys v0.0.0-20201119102817-f84b799fce68/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.0.0-20210113181707-4bcb84eeeb78/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
+golang.org/x/sys v0.0.0-20210330210617-4fbd30eecc44/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
+golang.org/x/sys v0.0.0-20210510120138-977fb7262007 h1:gG67DSER+11cZvqIMb8S8bt0vZtiN6xWYARwirrOSfE=
+golang.org/x/sys v0.0.0-20210510120138-977fb7262007/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
+golang.org/x/term v0.0.0-20201126162022-7de9c90e9dd1/go.mod h1:bj7SfCRtBDWHUb9snDiAeCFNEtKQo2Wmx5Cou7ajbmo=
 golang.org/x/text v0.3.0/go.mod h1:NqM8EUOU14njkJ3fqMW+pc6Ldnwhi/IjpwHt7yyuwOQ=
-golang.org/x/tools v0.0.0-20200216192241-b320d3a0f5a2 h1:0sfSpGSa544Fwnbot3Oxq/U6SXqjty6Jy/3wRhVS7ig=
-golang.org/x/tools v0.0.0-20200216192241-b320d3a0f5a2/go.mod h1:TB2adYChydJhpapKDTa4BR/hXlZSLoq2Wpct/0txZ28=
-golang.org/x/xerrors v0.0.0-20191011141410-1b5146add898 h1:/atklqdjdhuosWIl6AIbOeHJjicWYPqR9bpxqxYG2pA=
+golang.org/x/text v0.3.3/go.mod h1:5Zoc/QRtKVWzQhOtBMvqHzDpF6irO9z98xDceosuGiQ=
+golang.org/x/tools v0.0.0-20180917221912-90fa682c2a6e/go.mod h1:n7NCudcB/nEzxVGmLbDWY5pfWTLqBcC2KZ6jyYvM4mQ=
+golang.org/x/tools v0.0.0-20191119224855-298f0cb1881e/go.mod h1:b+2E5dAYhXwXZwtnZ6UAqBI28+e2cm9otk0dWdXHAEo=
+golang.org/x/tools v0.1.6-0.20210813165731-45389f592fe9 h1:nvvuMxmx1q0gfRki3T0hjG8EwAcVCs91oWAXvyt4zhI=
+golang.org/x/tools v0.1.6-0.20210813165731-45389f592fe9/go.mod h1:o0xws9oXOQQZyjljx8fwUC0k7L1pTE6eaCbjGeHmOkk=
+golang.org/x/xerrors v0.0.0-20190717185122-a985d3407aa7/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
 golang.org/x/xerrors v0.0.0-20191011141410-1b5146add898/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
+golang.org/x/xerrors v0.0.0-20200804184101-5ec99f83aff1/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
 gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
 gopkg.in/yaml.v2 v2.2.2/go.mod h1:hI93XBmqTisBFMUTm0b8Fm+jr3Dg1NNxqwp+5A1VGuI=
 tinygo.org/x/go-llvm v0.0.0-20210308112806-9ef958b6bed4 h1:CMUHxVTb+UuUePuMf8vkWjZ3gTp9BBK91KrgOCwoNHs=
@@ -3,12 +3,15 @@
 package goenv

 import (
+	"bytes"
+	"errors"
 	"fmt"
 	"os"
 	"os/exec"
 	"os/user"
 	"path/filepath"
 	"runtime"
+	"strings"
 )

 // Keys is a slice of all available environment variable keys.
@@ -22,6 +25,12 @@ var Keys = []string{
 	"TINYGOROOT",
 }

+func init() {
+	if Get("GOARCH") == "arm" {
+		Keys = append(Keys, "GOARM")
+	}
+}
+
 // TINYGOROOT is the path to the final location for checking tinygo files. If
 // unset (by a -X ldflag), then sourceDir() will fallback to the original build
 // directory.
@@ -41,6 +50,20 @@ func Get(name string) string {
 			return dir
 		}
 		return runtime.GOARCH
+	case "GOARM":
+		if goarm := os.Getenv("GOARM"); goarm != "" {
+			return goarm
+		}
+		if goos := Get("GOOS"); goos == "windows" || goos == "android" {
+			// Assume Windows and Android are running on modern CPU cores.
+			// This matches upstream Go.
+			return "7"
+		}
+		// Default to ARMv6 on other devices.
+		// The difference between ARMv5 and ARMv6 is big, much bigger than the
+		// difference between ARMv6 and ARMv7. ARMv6 binaries are much smaller,
+		// especially when floating point instructions are involved.
+		return "6"
 	case "GOROOT":
 		return getGoroot()
 	case "GOPATH":
@@ -67,11 +90,98 @@ func Get(name string) string {
 		return "1"
 	case "TINYGOROOT":
 		return sourceDir()
+	case "WASMOPT":
+		if path := os.Getenv("WASMOPT"); path != "" {
+			err := wasmOptCheckVersion(path)
+			if err != nil {
+				fmt.Fprintf(os.Stderr, "cannot use %q as wasm-opt (from WASMOPT environment variable): %s", path, err.Error())
+				os.Exit(1)
+			}
+
+			return path
+		}
+
+		return findWasmOpt()
 	default:
 		return ""
 	}
 }

+// Find wasm-opt, or exit with an error.
+func findWasmOpt() string {
+	tinygoroot := sourceDir()
+	searchPaths := []string{
+		tinygoroot + "/bin/wasm-opt",
+		tinygoroot + "/build/wasm-opt",
+	}
+
+	var paths []string
+	for _, path := range searchPaths {
+		if runtime.GOOS == "windows" {
+			path += ".exe"
+		}
+
+		_, err := os.Stat(path)
+		if err != nil && os.IsNotExist(err) {
+			continue
+		}
+
+		paths = append(paths, path)
+	}
+
+	if path, err := exec.LookPath("wasm-opt"); err == nil {
+		paths = append(paths, path)
+	}
+
+	if len(paths) == 0 {
+		fmt.Fprintln(os.Stderr, "error: could not find wasm-opt, set the WASMOPT environment variable to override")
+		os.Exit(1)
+	}
+
+	errs := make([]error, len(paths))
+	for i, path := range paths {
+		err := wasmOptCheckVersion(path)
+		if err == nil {
+			return path
+		}
+
+		errs[i] = err
+	}
+	fmt.Fprintln(os.Stderr, "no usable wasm-opt found, update or run \"make binaryen\"")
+	for i, path := range paths {
+		fmt.Fprintf(os.Stderr, "\t%s: %s\n", path, errs[i].Error())
+	}
+	os.Exit(1)
+	panic("unreachable")
+}
+
+// wasmOptCheckVersion checks if a copy of wasm-opt is usable.
+func wasmOptCheckVersion(path string) error {
+	cmd := exec.Command(path, "--version")
+	var buf bytes.Buffer
+	cmd.Stdout = &buf
+	cmd.Stderr = os.Stderr
+	err := cmd.Run()
+	if err != nil {
+		return err
+	}
+
+	str := buf.String()
+	if strings.Contains(str, "(") {
+		// The git tag may be placed in parentheses after the main version string.
+		str = strings.Split(str, "(")[0]
+	}
+
+	str = strings.TrimSpace(str)
+	var ver uint
+	_, err = fmt.Sscanf(str, "wasm-opt version %d", &ver)
+	if err != nil || ver < 102 {
+		return errors.New("incompatible wasm-opt (need 102 or newer)")
+	}
+
+	return nil
+}
+
 // Return the TINYGOROOT, or exit with an error.
 func sourceDir() string {
 	// Use $TINYGOROOT as root, if available.
@@ -12,7 +12,7 @@ import (

 // Version of TinyGo.
 // Update this value before release of new version of software.
-const Version = "0.19.0"
+const Version = "0.21.0-dev"

 // GetGorootVersion returns the major and minor version for a given GOROOT path.
 // If the goroot cannot be determined, (0, 0) is returned.
@@ -48,7 +48,10 @@ func GetGorootVersion(goroot string) (major, minor int, err error) {
 // toolchain for the given GOROOT path. It is usually of the form `go1.x.y` but
 // can have some variations (for beta releases, for example).
 func GorootVersionString(goroot string) (string, error) {
-	if data, err := ioutil.ReadFile(filepath.Join(
+	if data, err := ioutil.ReadFile(filepath.Join(goroot, "VERSION")); err == nil {
+		return string(data), nil
+
+	} else if data, err := ioutil.ReadFile(filepath.Join(
 		goroot, "src", "runtime", "internal", "sys", "zversion.go")); err == nil {

 		r := regexp.MustCompile("const TheVersion = `(.*)`")
@@ -59,9 +62,6 @@ func GorootVersionString(goroot string) (string, error) {

 		return string(matches[1]), nil

-	} else if data, err := ioutil.ReadFile(filepath.Join(goroot, "VERSION")); err == nil {
-		return string(data), nil
-
 	} else {
 		return "", err
 	}
@@ -57,6 +57,7 @@ func (r *runner) errorAt(inst instruction, err error) *Error {
 	pos := getPosition(inst.llvmInst)
 	return &Error{
 		ImportPath: r.pkgName,
+		Inst:       inst.llvmInst,
 		Pos:        pos,
 		Err:        err,
 		Traceback:  []ErrorLine{{pos, inst.llvmInst}},
@@ -15,7 +15,7 @@ import (
 // package is changed in a way that affects the output so that cached package
 // builds will be invalidated.
 // This version is independent of the TinyGo version number.
-const Version = 1
+const Version = 2 // last change: fix GEP on untyped pointers

 // Enable extra checks, which should be disabled by default.
 // This may help track down bugs by adding a few more sanity checks.
@@ -110,17 +110,26 @@ func Run(mod llvm.Module, debug bool) error {
 			fmt.Fprintln(os.Stderr, "call:", fn.Name())
 		}
 		_, mem, callErr := r.run(r.getFunction(fn), nil, nil, "    ")
+		call.EraseFromParentAsInstruction()
 		if callErr != nil {
 			if isRecoverableError(callErr.Err) {
 				if r.debug {
 					fmt.Fprintln(os.Stderr, "not interpreting", r.pkgName, "because of error:", callErr.Error())
 				}
+				// Remove instructions that were created as part of interpreting
+				// the package.
 				mem.revert()
+				// Create a call to the package initializer (which was
+				// previously deleted).
+				i8undef := llvm.Undef(r.i8ptrType)
+				r.builder.CreateCall(fn, []llvm.Value{i8undef, i8undef}, "")
+				// Make sure that any globals touched by the package
+				// initializer, won't be accessed by later package initializers.
+				r.markExternalLoad(fn)
 				continue
 			}
 			return callErr
 		}
-		call.EraseFromParentAsInstruction()
 		for index, obj := range mem.objects {
 			r.objects[index] = obj
 		}
@@ -136,6 +145,9 @@ func Run(mod llvm.Module, debug bool) error {
 		if obj.buffer == nil {
 			continue
 		}
+		if obj.constant {
+			continue // constant buffers can't have been modified
+		}
 		initializer, err := obj.buffer.toLLVMValue(obj.llvmGlobal.Type().ElementType(), &mem)
 		if err == errInvalidPtrToIntSize {
 			// This can happen when a previous interp run did not have the
@@ -239,6 +251,9 @@ func RunFunc(fn llvm.Value, debug bool) error {
 		if obj.buffer == nil {
 			continue
 		}
+		if obj.constant {
+			continue // constant, so can't have been modified
+		}
 		initializer, err := obj.buffer.toLLVMValue(obj.llvmGlobal.Type().ElementType(), &mem)
 		if err != nil {
 			return err
@@ -270,3 +285,19 @@ func (r *runner) getFunction(llvmFn llvm.Value) *function {
 	r.functionCache[llvmFn] = fn
 	return fn
 }
+
+// markExternalLoad marks the given llvmValue as being loaded externally. This
+// is primarily used to mark package initializers that could not be run at
+// compile time. As an example, a package initialize might store to a global
+// variable. Another package initializer might read from the same global
+// variable. By marking this function as being run at runtime, that load
+// instruction will need to be run at runtime instead of at compile time.
+func (r *runner) markExternalLoad(llvmValue llvm.Value) {
+	mem := memoryView{r: r}
+	mem.markExternalLoad(llvmValue)
+	for index, obj := range mem.objects {
+		if obj.marked > r.objects[index].marked {
+			r.objects[index].marked = obj.marked
+		}
+	}
+}
@@ -3,7 +3,6 @@ package interp
 import (
 	"io/ioutil"
 	"os"
-	"regexp"
 	"strings"
 	"testing"

@@ -17,6 +16,8 @@ func TestInterp(t *testing.T) {
 		"slice-copy",
 		"consteval",
 		"interface",
+		"revert",
+		"alloc",
 	} {
 		name := name // make tc local to this closure
 		t.Run(name, func(t *testing.T) {
@@ -87,8 +88,6 @@ func runTest(t *testing.T, pathPrefix string) {
 	}
 }

-var alignRegexp = regexp.MustCompile(", align [0-9]+$")
-
 // fuzzyEqualIR returns true if the two LLVM IR strings passed in are roughly
 // equal. That means, only relevant lines are compared (excluding comments
 // etc.).
@@ -100,15 +99,6 @@ func fuzzyEqualIR(s1, s2 string) bool {
 	}
 	for i, line1 := range lines1 {
 		line2 := lines2[i]
-		match1 := alignRegexp.MatchString(line1)
-		match2 := alignRegexp.MatchString(line2)
-		if match1 != match2 {
-			// Only one of the lines has the align keyword. Remove it.
-			// This is a change to make the test work in both LLVM 10 and LLVM
-			// 11 (LLVM 11 appears to automatically add alignment everywhere).
-			line1 = alignRegexp.ReplaceAllString(line1, "")
-			line2 = alignRegexp.ReplaceAllString(line2, "")
-		}
 		if line1 != line2 {
 			return false
 		}
@@ -234,11 +234,15 @@ func (r *runner) run(fn *function, params []value, parentMem *memoryView, indent
 				// Get the requested memory size to be allocated.
 				size := operands[1].Uint()

+				// Get the object layout, if it is available.
+				llvmLayoutType := r.getLLVMTypeFromLayout(operands[2])
+
 				// Create the object.
 				alloc := object{
-					globalName: r.pkgName + "$alloc",
-					buffer:     newRawValue(uint32(size)),
-					size:       uint32(size),
+					globalName:     r.pkgName + "$alloc",
+					llvmLayoutType: llvmLayoutType,
+					buffer:         newRawValue(uint32(size)),
+					size:           uint32(size),
 				}
 				index := len(r.objects)
 				r.objects = append(r.objects, alloc)
@@ -378,7 +382,7 @@ func (r *runner) run(fn *function, params []value, parentMem *memoryView, indent
 				} else {
 					locals[inst.localIndex] = literalValue{uint8(0)}
 				}
-			case callFn.name == "runtime.interfaceImplements":
+			case strings.HasSuffix(callFn.name, ".$typeassert"):
 				if r.debug {
 					fmt.Fprintln(os.Stderr, indent+"interface assert:", operands[1:])
 				}
@@ -393,11 +397,9 @@ func (r *runner) run(fn *function, params []value, parentMem *memoryView, indent
 					return nil, mem, r.errorAt(inst, err)
 				}
 				methodSet := mem.get(methodSetPtr.index()).llvmGlobal.Initializer()
-				interfaceMethodSetPtr, err := operands[2].asPointer(r)
-				if err != nil {
-					return nil, mem, r.errorAt(inst, err)
-				}
-				interfaceMethodSet := mem.get(interfaceMethodSetPtr.index()).llvmGlobal.Initializer()
+				llvmFn := inst.llvmInst.CalledValue()
+				methodSetAttr := llvmFn.GetStringAttributeAtIndex(-1, "tinygo-methods")
+				methodSetString := methodSetAttr.GetStringValue()

 				// Make a set of all the methods on the concrete type, for
 				// easier checking in the next step.
@@ -412,8 +414,7 @@ func (r *runner) run(fn *function, params []value, parentMem *memoryView, indent
 				// of defined methods calculated above. This is the interface
 				// assert itself.
 				assertOk := uint8(1) // i1 true
-				for i := 0; i < interfaceMethodSet.Type().ArrayLength(); i++ {
-					name := llvm.ConstExtractValue(interfaceMethodSet, []uint32{uint32(i)}).Name()
+				for _, name := range strings.Split(methodSetString, "; ") {
 					if _, ok := concreteTypeMethods[name]; !ok {
 						// There is a method on the interface that is not
 						// implemented by the type. The assertion will fail.
@@ -423,20 +424,20 @@ func (r *runner) run(fn *function, params []value, parentMem *memoryView, indent
 				}
 				// If assertOk is still 1, the assertion succeeded.
 				locals[inst.localIndex] = literalValue{assertOk}
-			case callFn.name == "runtime.interfaceMethod":
-				// This builtin returns the function (which may be a thunk) to
-				// invoke a method on an interface. It does not call the method.
+			case strings.HasSuffix(callFn.name, "$invoke"):
+				// This thunk is the interface method dispatcher: it is called
+				// with all regular parameters and a type code. It will then
+				// call the concrete method for it.
 				if r.debug {
-					fmt.Fprintln(os.Stderr, indent+"interface method:", operands[1:])
+					fmt.Fprintln(os.Stderr, indent+"invoke method:", operands[1:])
 				}

-				// Load the first param, which is the type code (ptrtoint of the
-				// type code global).
-				typecodeIDPtrToInt, err := operands[1].toLLVMValue(inst.llvmInst.Operand(0).Type(), &mem)
+				// Load the type code of the interface value.
+				typecodeIDBitCast, err := operands[len(operands)-3].toLLVMValue(inst.llvmInst.Operand(len(operands)-4).Type(), &mem)
 				if err != nil {
 					return nil, mem, r.errorAt(inst, err)
 				}
-				typecodeID := typecodeIDPtrToInt.Operand(0).Initializer()
+				typecodeID := typecodeIDBitCast.Operand(0).Initializer()

 				// Load the method set, which is part of the typecodeID object.
 				methodSet := llvm.ConstExtractValue(typecodeID, []uint32{2}).Operand(0).Initializer()
@@ -444,7 +445,10 @@ func (r *runner) run(fn *function, params []value, parentMem *memoryView, indent
 				// We don't need to load the interface method set.

 				// Load the signature of the to-be-called function.
-				signature := inst.llvmInst.Operand(2)
+				llvmFn := inst.llvmInst.CalledValue()
+				invokeAttr := llvmFn.GetStringAttributeAtIndex(-1, "tinygo-invoke")
+				invokeName := invokeAttr.GetStringValue()
+				signature := r.mod.NamedGlobal(invokeName)

 				// Iterate through all methods, looking for the one method that
 				// should be returned.
@@ -457,9 +461,13 @@ func (r *runner) run(fn *function, params []value, parentMem *memoryView, indent
 					}
 				}
 				if method.IsNil() {
-					return nil, mem, r.errorAt(inst, errors.New("could not find method: "+signature.Name()))
+					return nil, mem, r.errorAt(inst, errors.New("could not find method: "+invokeName))
 				}
-				locals[inst.localIndex] = r.getValue(method)
+
+				// Change the to-be-called function to the underlying method to
+				// be called and fall through to the default case.
+				callFn = r.getFunction(method)
+				fallthrough
 			default:
 				if len(callFn.blocks) == 0 {
 					// Call to a function declaration without a definition
@@ -590,18 +598,14 @@ func (r *runner) run(fn *function, params []value, parentMem *memoryView, indent
 			// GetElementPtr does pointer arithmetic, changing the offset of the
 			// pointer into the underlying object.
 			var offset uint64
-			var gepOperands []uint64
 			for i := 2; i < len(operands); i += 2 {
 				index := operands[i].Uint()
 				elementSize := operands[i+1].Uint()
 				if int64(elementSize) < 0 {
 					// This is a struct field.
-					// The field number is encoded by flipping all the bits.
-					gepOperands = append(gepOperands, ^elementSize)
 					offset += index
 				} else {
 					// This is a normal GEP, probably an array index.
-					gepOperands = append(gepOperands, index)
 					offset += elementSize * index
 				}
 			}
@@ -937,6 +941,7 @@ func (r *runner) runAtRuntime(fn *function, inst instruction, locals []value, me
 		agg := operands[0]
 		for i := 0; i < len(indices)-1; i++ {
 			agg = r.builder.CreateExtractValue(agg, int(indices[i]), inst.name+".agg")
+			mem.instructions = append(mem.instructions, agg)
 		}
 		result = r.builder.CreateExtractValue(agg, int(indices[len(indices)-1]), inst.name)
 	case llvm.InsertValue:
@@ -949,11 +954,15 @@ func (r *runner) runAtRuntime(fn *function, inst instruction, locals []value, me
 		for i := 0; i < len(indices)-1; i++ {
 			agg = r.builder.CreateExtractValue(agg, int(indices[i]), inst.name+".agg"+strconv.Itoa(i))
 			aggregates = append(aggregates, agg)
+			mem.instructions = append(mem.instructions, agg)
 		}
 		result = operands[1]
 		for i := len(indices) - 1; i >= 0; i-- {
 			agg := aggregates[i]
 			result = r.builder.CreateInsertValue(agg, result, int(indices[i]), inst.name+".insertvalue"+strconv.Itoa(i))
+			if i != 0 { // don't add last result to mem.instructions as it will be done at the end already
+				mem.instructions = append(mem.instructions, result)
+			}
 		}

 	case llvm.Add:
@@ -18,6 +18,7 @@ import (
 	"encoding/binary"
 	"errors"
 	"math"
+	"math/big"
 	"strconv"
 	"strings"

@@ -27,17 +28,21 @@ import (
 // An object is a memory buffer that may be an already existing global or a
 // global created with runtime.alloc or the alloca instruction. If llvmGlobal is
 // set, that's the global for this object, otherwise it needs to be created (if
-// it is still reachable when the package initializer returns).
+// it is still reachable when the package initializer returns). The
+// llvmLayoutType is not necessarily a complete type: it may need to be
+// repeated (for example, for a slice value).
 //
 // Objects are copied in a memory view when they are stored to, to provide the
 // ability to roll back interpreting a function.
 type object struct {
-	llvmGlobal llvm.Value
-	llvmType   llvm.Type // must match llvmGlobal.Type() if both are set, may be unset if llvmGlobal is set
-	globalName string    // name, if not yet created (not guaranteed to be the final name)
-	buffer     value     // buffer with value as given by interp, nil if external
-	size       uint32    // must match buffer.len(), if available
-	marked     uint8     // 0 means unmarked, 1 means external read, 2 means external write
+	llvmGlobal     llvm.Value
+	llvmType       llvm.Type // must match llvmGlobal.Type() if both are set, may be unset if llvmGlobal is set
+	llvmLayoutType llvm.Type // LLVM type based on runtime.alloc layout parameter, if available
+	globalName     string    // name, if not yet created (not guaranteed to be the final name)
+	buffer         value     // buffer with value as given by interp, nil if external
+	size           uint32    // must match buffer.len(), if available
+	constant       bool      // true if this is a constant global
+	marked         uint8     // 0 means unmarked, 1 means external read, 2 means external write
 }

 // clone() returns a cloned version of this object, for when an object needs to
@@ -219,7 +224,7 @@ func (mv *memoryView) hasExternalLoadOrStore(v pointerValue) bool {
 // possible for the interpreter to read from the object.
 func (mv *memoryView) hasExternalStore(v pointerValue) bool {
 	obj := mv.get(v.index())
-	return obj.marked >= 2
+	return obj.marked >= 2 && !obj.constant
 }

 // get returns an object that can only be read from, as it may return an object
@@ -259,6 +264,9 @@ func (mv *memoryView) put(index uint32, obj object) {
 	if checks && mv.get(index).buffer.len(mv.r) != obj.buffer.len(mv.r) {
 		panic("put() with a differently-sized object")
 	}
+	if checks && obj.constant {
+		panic("interp: store to a constant")
+	}
 	mv.objects[index] = obj
 }

@@ -522,6 +530,18 @@ func (v pointerValue) llvmValue(mem *memoryView) llvm.Value {
 // bitcast. The llvm.Type parameter is optional, if omitted the pointer type may
 // be different than expected.
 func (v pointerValue) toLLVMValue(llvmType llvm.Type, mem *memoryView) (llvm.Value, error) {
+	// If a particular LLVM type is requested, cast to it.
+	if !llvmType.IsNil() && llvmType.TypeKind() != llvm.PointerTypeKind {
+		// The LLVM value has (or should have) the same bytes once compiled, but
+		// does not have the right LLVM type. This can happen for example when
+		// storing to a struct with a single pointer field: this pointer may
+		// then become the value even though the pointer should be wrapped in a
+		// struct.
+		// This can be worked around by simply converting to a raw value,
+		// rawValue knows how to create such structs.
+		return v.asRawValue(mem.r).toLLVMValue(llvmType, mem)
+	}
+
 	// Obtain the llvmValue, creating it if it doesn't exist yet.
 	llvmValue := v.llvmValue(mem)
 	if llvmValue.IsNil() {
@@ -529,7 +549,7 @@ func (v pointerValue) toLLVMValue(llvmType llvm.Type, mem *memoryView) (llvm.Val
 		// runtime.alloc.
 		// First allocate a new global for this object.
 		obj := mem.get(v.index())
-		if obj.llvmType.IsNil() {
+		if obj.llvmType.IsNil() && obj.llvmLayoutType.IsNil() {
 			// Create an initializer without knowing the global type.
 			// This is probably the result of a runtime.alloc call.
 			initializer, err := obj.buffer.asRawValue(mem.r).rawLLVMValue(mem)
@@ -543,7 +563,23 @@ func (v pointerValue) toLLVMValue(llvmType llvm.Type, mem *memoryView) (llvm.Val
 			obj.llvmGlobal = llvmValue
 			mem.put(v.index(), obj)
 		} else {
-			globalType := obj.llvmType.ElementType()
+			// The global type is known, or at least its structure.
+			var globalType llvm.Type
+			if !obj.llvmType.IsNil() {
+				// The exact type is known.
+				globalType = obj.llvmType.ElementType()
+			} else { // !obj.llvmLayoutType.IsNil()
+				// The exact type isn't known, but the object layout is known.
+				globalType = obj.llvmLayoutType
+				// The layout may not span the full size of the global because
+				// of repetition. One example would be make([]string, 5) which
+				// would be 10 words in size but the layout would only be two
+				// words (for the string type).
+				typeSize := mem.r.targetData.TypeAllocSize(globalType)
+				if typeSize != uint64(obj.size) {
+					globalType = llvm.ArrayType(globalType, int(uint64(obj.size)/typeSize))
+				}
+			}
 			if checks && mem.r.targetData.TypeAllocSize(globalType) != uint64(obj.size) {
 				panic("size of the globalType isn't the same as the object size")
 			}
@@ -562,6 +598,11 @@ func (v pointerValue) toLLVMValue(llvmType llvm.Type, mem *memoryView) (llvm.Val
 				return llvm.Value{}, errors.New("interp: allocated value does not match allocated type")
 			}
 			llvmValue.SetInitializer(initializer)
+			if obj.llvmType.IsNil() {
+				// The exact type isn't known (only the layout), so use the
+				// alignment that would normally be expected from runtime.alloc.
+				llvmValue.SetAlignment(mem.r.maxAlign)
+			}
 		}

 		// It should be included in r.globals because otherwise markExternal
@@ -571,76 +612,24 @@ func (v pointerValue) toLLVMValue(llvmType llvm.Type, mem *memoryView) (llvm.Val
 		llvmValue.SetLinkage(llvm.InternalLinkage)
 	}

-	if llvmType.IsNil() {
-		return llvmValue, nil
-	}
-
-	if llvmType.TypeKind() != llvm.PointerTypeKind {
-		// The LLVM value has (or should have) the same bytes once compiled, but
-		// does not have the right LLVM type. This can happen for example when
-		// storing to a struct with a single pointer field: this pointer may
-		// then become the value even though the pointer should be wrapped in a
-		// struct.
-		// This can be worked around by simply converting to a raw value,
-		// rawValue knows how to create such structs.
-		if v.offset() != 0 {
-			return llvm.Value{}, errors.New("interp: offset set without known pointer type")
+	if v.offset() != 0 {
+		// If there is an offset, make sure to use a GEP to index into the
+		// pointer.
+		// Cast to an i8* first (if needed) for easy indexing.
+		if llvmValue.Type() != mem.r.i8ptrType {
+			llvmValue = llvm.ConstBitCast(llvmValue, mem.r.i8ptrType)
 		}
-		return v.asRawValue(mem.r).toLLVMValue(llvmType, mem)
+		llvmValue = llvm.ConstInBoundsGEP(llvmValue, []llvm.Value{
+			llvm.ConstInt(llvmValue.Type().Context().Int32Type(), uint64(v.offset()), false),
+		})
 	}

-	requestedType := llvmType
-	objectElementType := llvmValue.Type()
-	if requestedType == objectElementType {
-		if v.offset() != 0 {
-			// This should never happen, if offset is non-zero, the types
-			// shouldn't match.
-			return llvm.Value{}, errors.New("interp: offset set while there is no way to convert the type")
-		}
-		return llvmValue, nil
+	// If a particular LLVM pointer type is requested, cast to it.
+	if !llvmType.IsNil() && llvmType != llvmValue.Type() {
+		llvmValue = llvm.ConstBitCast(llvmValue, llvmType)
 	}

-	if v.offset() == 0 {
-		// Offset is zero, so we can just bitcast to get a correct pointer.
-		return llvm.ConstBitCast(llvmValue, llvmType), nil
-	}
-
-	// We need to make a constant GEP for pointer arithmetic.
-	int32Type := llvmType.Context().Int32Type()
-	indices := []llvm.Value{llvm.ConstInt(int32Type, 0, false)}
-	requestedType = requestedType.ElementType()
-	objectElementType = objectElementType.ElementType()
-	offset := int64(v.offset())
-	for offset > 0 {
-		switch objectElementType.TypeKind() {
-		case llvm.ArrayTypeKind:
-			elementType := objectElementType.ElementType()
-			elementSize := mem.r.targetData.TypeAllocSize(elementType)
-			elementIndex := uint64(offset) / elementSize
-			indices = append(indices, llvm.ConstInt(int32Type, elementIndex, false))
-			offset -= int64(elementIndex * elementSize)
-			objectElementType = elementType
-		case llvm.StructTypeKind:
-			element := mem.r.targetData.ElementContainingOffset(objectElementType, uint64(offset))
-			indices = append(indices, llvm.ConstInt(int32Type, uint64(element), false))
-			offset -= int64(mem.r.targetData.ElementOffset(objectElementType, element))
-			objectElementType = objectElementType.StructElementTypes()[element]
-		default:
-			return llvm.Value{}, errors.New("interp: pointer index with something other than a struct or array?")
-		}
-	}
-	if offset < 0 {
-		return llvm.Value{}, errors.New("interp: offset has somehow gone negative, this should be impossible")
-	}
-
-	// Finally do the gep, using the above computed indices.
-	// If it still doesn't match te requested type, it's possible to bitcast (as
-	// the bits of the pointer are now correct, just not the type).
-	gep := llvm.ConstInBoundsGEP(llvmValue, indices)
-	if gep.Type() != llvmType {
-		return llvm.ConstBitCast(gep, llvmType), nil
-	}
-	return gep, nil
+	return llvmValue, nil
 }

 // rawValue is a raw memory buffer that can store either pointers or regular
@@ -974,12 +963,9 @@ func (v *rawValue) set(llvmValue llvm.Value, r *runner) {
 		case llvm.GetElementPtr:
 			ptr := llvmValue.Operand(0)
 			index := llvmValue.Operand(1)
-			if checks && index.IsAConstantInt().IsNil() || index.ZExtValue() != 0 {
-				panic("expected first index of const gep to be i32 0")
-			}
 			numOperands := llvmValue.OperandsCount()
 			elementType := ptr.Type().ElementType()
-			totalOffset := uint64(0)
+			totalOffset := r.targetData.TypeAllocSize(elementType) * index.ZExtValue()
 			for i := 2; i < numOperands; i++ {
 				indexValue := llvmValue.Operand(i)
 				if checks && indexValue.IsAConstantInt().IsNil() {
@@ -1156,6 +1142,7 @@ func (r *runner) getValue(llvmValue llvm.Value) value {
 				obj.size = uint32(r.targetData.TypeAllocSize(llvmValue.Type().ElementType()))
 				if initializer := llvmValue.Initializer(); !initializer.IsNil() {
 					obj.buffer = r.getValue(initializer)
+					obj.constant = llvmValue.IsGlobalConstant()
 				}
 			} else if !llvmValue.IsAFunction().IsNil() {
 				// OK
@@ -1198,3 +1185,108 @@ func (r *runner) getValue(llvmValue llvm.Value) value {
 		panic("unknown value")
 	}
 }
+
+// readObjectLayout reads the object layout as it is stored by the compiler. It
+// returns the size in the number of words and the bitmap.
+func (r *runner) readObjectLayout(layoutValue value) (uint64, *big.Int) {
+	pointerSize := layoutValue.len(r)
+	if checks && uint64(pointerSize) != r.targetData.TypeAllocSize(r.i8ptrType) {
+		panic("inconsistent pointer size")
+	}
+
+	// The object layout can be stored in a global variable, directly as an
+	// integer value, or can be nil.
+	ptr, err := layoutValue.asPointer(r)
+	if err == errIntegerAsPointer {
+		// It's an integer, which means it's a small object or unknown.
+		layout := layoutValue.Uint()
+		if layout == 0 {
+			// Nil pointer, which means the layout is unknown.
+			return 0, nil
+		}
+		if layout%2 != 1 {
+			// Sanity check: the least significant bit must be set. This is how
+			// the runtime can separate pointers from integers.
+			panic("unexpected layout")
+		}
+
+		// Determine format of bitfields in the integer.
+		pointerBits := uint64(pointerSize * 8)
+		var sizeFieldBits uint64
+		switch pointerBits {
+		case 16:
+			sizeFieldBits = 4
+		case 32:
+			sizeFieldBits = 5
+		case 64:
+			sizeFieldBits = 6
+		default:
+			panic("unknown pointer size")
+		}
+
+		// Extract fields.
+		objectSizeWords := (layout >> 1) & (1<<sizeFieldBits - 1)
+		bitmap := new(big.Int).SetUint64(layout >> (1 + sizeFieldBits))
+		return objectSizeWords, bitmap
+	}
+
+	// Read the object size in words and the bitmap from the global.
+	buf := r.objects[ptr.index()].buffer.(rawValue)
+	objectSizeWords := rawValue{buf: buf.buf[:r.pointerSize]}.Uint()
+	rawByteValues := buf.buf[r.pointerSize:]
+	rawBytes := make([]byte, len(rawByteValues))
+	for i, v := range rawByteValues {
+		if uint64(byte(v)) != v {
+			panic("found pointer in data array?") // sanity check
+		}
+		rawBytes[i] = byte(v)
+	}
+	bitmap := new(big.Int).SetBytes(rawBytes)
+	return objectSizeWords, bitmap
+}
+
+// getLLVMTypeFromLayout returns the 'layout type', which is an approximation of
+// the real type. Pointers are in the correct location but the actual object may
+// have some additional repetition, for example in the buffer of a slice.
+func (r *runner) getLLVMTypeFromLayout(layoutValue value) llvm.Type {
+	objectSizeWords, bitmap := r.readObjectLayout(layoutValue)
+	if bitmap == nil {
+		// No information available.
+		return llvm.Type{}
+	}
+
+	if bitmap.BitLen() == 0 {
+		// There are no pointers in this object, so treat this as a raw byte
+		// buffer. This is important because objects without pointers may have
+		// lower alignment.
+		return r.mod.Context().Int8Type()
+	}
+
+	// Create the LLVM type.
+	pointerSize := layoutValue.len(r)
+	pointerAlignment := r.targetData.PrefTypeAlignment(r.i8ptrType)
+	var fields []llvm.Type
+	for i := 0; i < int(objectSizeWords); {
+		if bitmap.Bit(i) != 0 {
+			// Pointer field.
+			fields = append(fields, r.i8ptrType)
+			i += int(pointerSize / uint32(pointerAlignment))
+		} else {
+			// Byte/word field.
+			fields = append(fields, r.mod.Context().IntType(pointerAlignment*8))
+			i += 1
+		}
+	}
+	var llvmLayoutType llvm.Type
+	if len(fields) == 1 {
+		llvmLayoutType = fields[0]
+	} else {
+		llvmLayoutType = r.mod.Context().StructType(fields, false)
+	}
+
+	objectSizeBytes := objectSizeWords * uint64(pointerAlignment)
+	if checks && r.targetData.TypeAllocSize(llvmLayoutType) != objectSizeBytes {
+		panic("unexpected size") // sanity check
+	}
+	return llvmLayoutType
+}
--- a/Show More
+++ b/Show More