interp: support GEP on fixed (MMIO) addresses

GetElementPtr would not work on values that weren't pointers. Because fixed addresses (often used in memory-mapped I/O) are integers rather than pointers in interp, it would return an error. This resulted in the teensy40 target not compiling correctly since the interp package rewrite. This commit should fix that.
interp: fix alignment of untyped globals
2026-07-27 07:08:42 +00:00 · 2021-03-05 21:24:28 +01:00 · 2020-12-27 11:21:35 +01:00 · 2020-12-22 15:54:23 +01:00 · 2020-12-22 15:54:23 +01:00 · 2020-12-22 15:54:23 +01:00
297 changed files with 16556 additions and 4351 deletions
@@ -28,6 +28,7 @@ commands:
                qemu-user \
                gcc-avr \
                avr-libc
+            sudo apt-get install --no-install-recommends libc6-dev-i386 lib32gcc-8-dev
  install-node:
    steps:
      - run:
@@ -44,28 +45,47 @@ commands:
          command: |
            wget https://dl.google.com/linux/direct/google-chrome-stable_current_amd64.deb
            sudo apt install ./google-chrome-stable_current_amd64.deb
+  install-wasmtime:
+    steps:
+      - run:
+          name: "Install wasmtime"
+          command: |
+            curl https://wasmtime.dev/install.sh -sSf | bash
+            sudo ln -s ~/.wasmtime/bin/wasmtime /usr/local/bin/wasmtime
+  install-xtensa-toolchain:
+    parameters:
+      variant:
+        type: string
+    steps:
+      - run:
+          name: "Install Xtensa toolchain"
+          command: |
+            curl -L https://github.com/espressif/crosstool-NG/releases/download/esp-2020r2/xtensa-esp32-elf-gcc8_2_0-esp-2020r2-<<parameters.variant>>.tar.gz -o xtensa-esp32-elf-gcc8_2_0-esp-2020r2-<<parameters.variant>>.tar.gz
+            sudo tar -C /usr/local -xf xtensa-esp32-elf-gcc8_2_0-esp-2020r2-<<parameters.variant>>.tar.gz
+            sudo ln -s /usr/local/xtensa-esp32-elf/bin/xtensa-esp32-elf-ld /usr/local/bin/xtensa-esp32-elf-ld
+            rm xtensa-esp32-elf-gcc8_2_0-esp-2020r2-<<parameters.variant>>.tar.gz
  llvm-source-linux:
    steps:
      - restore_cache:
          keys:
-            - llvm-source-10-v0
+            - llvm-source-11-v1
      - run:
          name: "Fetch LLVM source"
          command: make llvm-source
      - save_cache:
-          key: llvm-source-10-v0
+          key: llvm-source-11-v1
          paths:
            - llvm-project
  build-wasi-libc:
    steps:
      - restore_cache:
          keys:
-            - wasi-libc-sysroot-v2
+            - wasi-libc-sysroot-v3
      - run:
          name: "Build wasi-libc"
          command: make wasi-libc
      - save_cache:
-          key: wasi-libc-sysroot-v2
+          key: wasi-libc-sysroot-v3
          paths:
            - lib/wasi-libc/sysroot
  test-linux:
@@ -79,6 +99,7 @@ commands:
          llvm: "<<parameters.llvm>>"
      - install-node
      - install-chrome
+      - install-wasmtime
      - restore_cache:
          keys:
            - go-cache-v2-{{ checksum "go.mod" }}-{{ .Environment.CIRCLE_PREVIOUS_BUILD_NUM }}
@@ -87,15 +108,16 @@ commands:
      - run: go install -tags=llvm<<parameters.llvm>> .
      - restore_cache:
          keys:
-            - wasi-libc-sysroot-systemclang-v1
+            - wasi-libc-sysroot-systemclang-v2
      - run: make wasi-libc
      - save_cache:
-          key: wasi-libc-sysroot-systemclang-v1
+          key: wasi-libc-sysroot-systemclang-v2
          paths:
            - lib/wasi-libc/sysroot
      - run: go test -v -tags=llvm<<parameters.llvm>> ./cgo ./compileopts ./interp ./transform .
      - run: make gen-device -j4
-      - run: make smoketest
+      - 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 }}
@@ -112,7 +134,6 @@ commands:
          command: |
            sudo apt-get install \
                gcc-arm-linux-gnueabihf \
-                binutils-arm-none-eabi \
                libc6-dev-armel-cross \
                gcc-aarch64-linux-gnu \
                libc6-dev-arm64-cross \
@@ -120,7 +141,11 @@ commands:
                qemu-user \
                gcc-avr \
                avr-libc
+            sudo apt-get install --no-install-recommends libc6-dev-i386 lib32gcc-6-dev
      - install-node
+      - install-wasmtime
+      - install-xtensa-toolchain:
+          variant: "linux-amd64"
      - restore_cache:
          keys:
            - go-cache-v2-{{ checksum "go.mod" }}-{{ .Environment.CIRCLE_PREVIOUS_BUILD_NUM }}
@@ -128,17 +153,14 @@ commands:
      - llvm-source-linux
      - restore_cache:
          keys:
-            - llvm-build-10-linux-v0-assert
+            - llvm-build-11-linux-v1-assert
      - run:
          name: "Build LLVM"
          command: |
            if [ ! -f llvm-build/lib/liblldELF.a ]
            then
              # install dependencies
-              sudo apt-get install cmake clang ninja-build
-              # make build faster
-              export CC=clang
-              export CXX=clang++
+              sudo apt-get install cmake ninja-build
              # hack ninja to use less jobs
              echo -e '#!/bin/sh\n/usr/bin/ninja -j3 "$@"' > /go/bin/ninja
              chmod +x /go/bin/ninja
@@ -146,7 +168,7 @@ commands:
              make ASSERT=1 llvm-build
            fi
      - save_cache:
-          key: llvm-build-10-linux-v0-assert
+          key: llvm-build-11-linux-v1-assert
          paths:
            llvm-build
      - run: make ASSERT=1
@@ -170,7 +192,6 @@ commands:
          command: |
            sudo apt-get install \
                gcc-arm-linux-gnueabihf \
-                binutils-arm-none-eabi \
                libc6-dev-armel-cross \
                gcc-aarch64-linux-gnu \
                libc6-dev-arm64-cross \
@@ -178,7 +199,11 @@ commands:
                qemu-user \
                gcc-avr \
                avr-libc
+            sudo apt-get install --no-install-recommends libc6-dev-i386 lib32gcc-6-dev
      - install-node
+      - install-wasmtime
+      - install-xtensa-toolchain:
+          variant: "linux-amd64"
      - restore_cache:
          keys:
            - go-cache-v2-{{ checksum "go.mod" }}-{{ .Environment.CIRCLE_PREVIOUS_BUILD_NUM }}
@@ -186,17 +211,14 @@ commands:
      - llvm-source-linux
      - restore_cache:
          keys:
-            - llvm-build-10-linux-v0
+            - llvm-build-11-linux-v1-noassert
      - run:
          name: "Build LLVM"
          command: |
            if [ ! -f llvm-build/lib/liblldELF.a ]
            then
              # install dependencies
-              sudo apt-get install cmake clang ninja-build
-              # make build faster
-              export CC=clang
-              export CXX=clang++
+              sudo apt-get install cmake ninja-build
              # hack ninja to use less jobs
              echo -e '#!/bin/sh\n/usr/bin/ninja -j3 "$@"' > /go/bin/ninja
              chmod +x /go/bin/ninja
@@ -204,7 +226,7 @@ commands:
              make llvm-build
            fi
      - save_cache:
-          key: llvm-build-10-linux-v0
+          key: llvm-build-11-linux-v1-noassert
          paths:
            llvm-build
      - build-wasi-libc
@@ -246,27 +268,29 @@ commands:
      - run:
          name: "Install dependencies"
          command: |
-            curl https://dl.google.com/go/go1.14.darwin-amd64.tar.gz -o go1.14.darwin-amd64.tar.gz
-            sudo tar -C /usr/local -xzf go1.14.darwin-amd64.tar.gz
+            curl https://dl.google.com/go/go1.15.5.darwin-amd64.tar.gz -o go1.15.5.darwin-amd64.tar.gz
+            sudo tar -C /usr/local -xzf go1.15.5.darwin-amd64.tar.gz
            ln -s /usr/local/go/bin/go /usr/local/bin/go
            HOMEBREW_NO_AUTO_UPDATE=1 brew install qemu
+      - install-xtensa-toolchain:
+          variant: "macos"
      - restore_cache:
          keys:
            - go-cache-macos-v2-{{ checksum "go.mod" }}-{{ .Environment.CIRCLE_PREVIOUS_BUILD_NUM }}
            - go-cache-macos-v2-{{ checksum "go.mod" }}
      - restore_cache:
          keys:
-            - llvm-source-10-macos-v0
+            - llvm-source-11-macos-v1
      - run:
          name: "Fetch LLVM source"
          command: make llvm-source
      - save_cache:
-          key: llvm-source-10-macos-v0
+          key: llvm-source-11-macos-v1
          paths:
            - llvm-project
      - restore_cache:
          keys:
-            - llvm-build-10-macos-v0
+            - llvm-build-11-macos-v1
      - run:
          name: "Build LLVM"
          command: |
@@ -278,17 +302,17 @@ commands:
              make llvm-build
            fi
      - save_cache:
-          key: llvm-build-10-macos-v0
+          key: llvm-build-11-macos-v1
          paths:
            llvm-build
      - restore_cache:
          keys:
-            - wasi-libc-sysroot-macos-v1
+            - wasi-libc-sysroot-macos-v2
      - run:
          name: "Build wasi-libc"
          command: make wasi-libc
      - save_cache:
-          key: wasi-libc-sysroot-macos-v1
+          key: wasi-libc-sysroot-macos-v2
          paths:
            - lib/wasi-libc/sysroot
      - run:
@@ -314,53 +338,6 @@ commands:
          paths:
            - ~/.cache/go-build
            - /go/pkg/mod
-  arch-release:
-    steps:
-      - run:
-          name: Install dependencies
-          command: pacman -Sy --noconfirm openssh git pacman-contrib binutils
-      - run:
-          name: Create TinyGo user
-          command: useradd -m tinygo
-      - run:
-          name: Become TinyGo user
-          command: su tinygo
-      - run:
-          name: Start SSH Agent
-          command: eval $(ssh-agent -s)
-      - run:
-          name: Add ARCH_RELEASE_SSH_PRIVATE_KEY identity
-          command: echo "${ARCH_RELEASE_SSH_PRIVATE_KEY}" | tr -d '\r' | ssh-add -
-      - run:
-          name: Create SSH directory
-          command: mkdir -p ~/.ssh && chmod 700 ~/.ssh
-      - run:
-          name: Add aur.archlinux.org to known hosts
-          command: ssh-keyscan aur.archlinux.org >> ~/.ssh/known_hosts
-      - run:
-          name: Clone tinygo-bin repo
-          command: git clone ssh://aur@aur.archlinux.org/tinygo-bin.git ~/tinygo-bin
-      - run:
-          name: Update package version
-          command: sed -i -E "s/(pkgver=)(.*)$/\1${CIRCLE_TAG}/" ~/tinygo-bin/PKGBUILD
-      - run:
-          name: Update file checksums
-          command: cd ~/tinygo-bin && updpkgsums
-      - run:
-          name: Update .SRCINFO
-          command: cd ~/tinygo-bin && makepkg --printsrcinfo > .SRCINFO
-      # Commit the update
-      - run:
-          name: Set git commit config
-          command: |
-            git config --global user.email "tinygo-bot@tinygo.org" &&
-            git config --global user.name "TinyGo Release Bot"
-      - run:
-          name: Commit and push changes
-          command: |
-            cd ~/tinygo-bin &&
-            git commit -a -m "Update tinygo-bin to v${CIRCLE_TAG}" &&
-            git push origin master

 jobs:
  test-llvm9-go111:
@@ -387,6 +364,12 @@ jobs:
    steps:
      - test-linux:
          llvm: "10"
+  test-llvm11-go115:
+    docker:
+      - image: circleci/golang:1.15-buster
+    steps:
+      - test-linux:
+          llvm: "11"
  assert-test-linux:
    docker:
      - image: circleci/golang:1.14-stretch
@@ -402,12 +385,6 @@ jobs:
      xcode: "10.1.0"
    steps:
      - build-macos
-  arch-release:
-    docker:
-      - image: archlinux:latest
-    steps:
-      - arch-release
-



@@ -418,15 +395,7 @@ workflows:
      - test-llvm10-go112
      - test-llvm10-go113
      - test-llvm10-go114
+      - test-llvm11-go115
      - build-linux
      - build-macos
      - assert-test-linux
-  release:
-    jobs:
-      - arch-release:
-          filters:
-            branches:
-              ignore: /.*/
-            tags:
-              # Runs on every semver release
-              only: /v[0-9]+(\.[0-9]+)*(-.*)*/
@@ -3,6 +3,7 @@ docs/_build
 src/device/avr/*.go
 src/device/avr/*.ld
 src/device/avr/*.s
+src/device/esp/*.go
 src/device/nrf/*.go
 src/device/nrf/*.s
 src/device/nxp/*.go
@@ -1,3 +1,137 @@
+0.16.0
+---
+
+* **command-line**
+  - add initial support for LLVM 11
+  - make lib64 clang include path check more robust
+  - `build`: improve support for GOARCH=386 and add tests
+  - `gdb`: add support for qemu-user targets
+  - `test`: support non-host tests
+  - `test`: add support for -c and -o flags
+  - `test`: implement some benchmark stubs
+* **compiler**
+  - `builder`: improve detection of clang on Fedora
+  - `compiler`: fix floating point comparison bugs
+  - `compiler`: implement negate for complex numbers
+  - `loader`: fix linkname in test binaries
+  - `transform`: add missing return pointer restore for regular coroutine tail
+    calls
+* **standard library**
+  - `machine`: switch default frequency to 4MHz
+  - `machine`: clarify caller's responsibility in `SetInterrupt`
+  - `os`: add `LookupEnv()` stub
+  - `reflect`: implement `Swapper`
+  - `runtime`: fix UTF-8 decoding
+  - `runtime`: gc: use raw stack access whenever possible
+  - `runtime`: use dedicated printfloat32
+  - `runtime`: allow ranging over a nil map
+  - `runtime`: avoid device/nxp dependency in HardFault handler
+  - `testing`: implement dummy Helper method
+  - `testing`: add Run method
+* **targets**
+  - `arm64`: add support for SVCall intrinsic
+  - `atsamd51`: avoid panic when configuring SPI with SDI=NoPin
+  - `avr`: properly support the `.rodata` section
+  - `esp8266`: implement `Pin.Get` function
+  - `nintendoswitch`: fix crash when printing long lines (> 120)
+  - `nintendoswitch`: add env parser and removed unused stuff
+  - `nrf`: add I2C error checking
+  - `nrf`: give more flexibility in picking SPI speeds
+  - `nrf`: fix nrf52832 flash size
+  - `stm32f103`: support wakeups from interrupts
+  - `stm32f405`: add SPI support
+  - `stm32f405`: add I2C support
+  - `wasi`: add support for this target
+  - `wasi`: use 'generic' ABI by default
+  - `wasi`: remove --no-threads flag from wasm-ld
+  - `wasm`: add instanceof support for WebAssembly
+  - `wasm`: use fixed length buffer for putchar
+* **boards**
+  - `d1mini`: add this ESP8266 based board
+  - `esp32`: use board definitions instead of chip names
+  - `qtpy`: add board definition for Adafruit QTPy
+  - `teensy40`: add this board
+
+0.15.0
+---
+
+* **command-line**
+  - add cached GOROOT to info subcommand
+  - embed git-hash in tinygo-dev executable
+  - implement tinygo targets to list usable targets
+  - use simpler file copy instead of file renaming to avoid issues on nrf52840 UF2 bootloaders
+  - use ToSlash() to specify program path
+  - support flashing esp32/esp8266 directly from tinygo
+  - when flashing call PortReset only on other than openocd
+* **compiler**
+  - `compileopts`: add support for custom binary formats
+  - `compiler`: improve display of goroutine wrappers
+  - `interp`: don't panic in the Store method
+  - `interp`: replace some panics with error messages
+  - `interp`: show error line in first line of the traceback
+  - `loader`: be more robust when creating the cached GOROOT
+  - `loader`: rewrite/refactor much of the code to use go list directly
+  - `loader`: use ioutil.TempDir to create a temporary directory
+  - `stacksize`: deal with DW_CFA_advance_loc1
+* **standard library**
+  - `runtime`: use waitForEvents when appropriate
+* **wasm**
+  - `wasm`: Remove --no-threads from wasm-ld calls.
+  - `wasm`: update wasi-libc dependency
+* **targets**
+  - `arduino-mega2560`: fix flashing on Windows
+  - `arm`: automatically determine stack sizes
+  - `arm64`: make dynamic loader structs and constants private
+  - `avr`: configure emulator in board files
+  - `cortexm`: fix stack size calculation with interrupts
+  - `flash`: add openocd settings to atsamd21 / atsamd51
+  - `flash`: add openocd settings to nrf5
+  - `microbit`: reelboard: flash using OpenOCD when needed
+  - `nintendoswitch`: Add dynamic loader for runtime loading PIE sections
+  - `nintendoswitch`: fix import cycle on dynamic_arm64.go
+  - `nintendoswitch`: Fix invalid memory read / write in print calls
+  - `nintendoswitch`: simplified assembly code
+  - `nintendoswitch`: support outputting .nro files directly
+* **boards**
+  - `arduino-zero`: Adding support for the Arduino Zero (#1365)
+  - `atsamd2x`: fix BAUD value
+  - `atsamd5x`: fix BAUD value
+  - `bluepill`: Enable stm32's USART2 for the board and map it to UART1 tinygo's device
+  - `device/atsamd51x`: add all remaining bitfield values for PCHCTRLm Mapping
+  - `esp32`: add libgcc ROM functions to linker script
+  - `esp32`: add SPI support
+  - `esp32`: add support for basic GPIO
+  - `esp32`: add support for the Espressif ESP32 chip
+  - `esp32`: configure the I/O matrix for GPIO pins
+  - `esp32`: export machine.PortMask* for bitbanging implementations
+  - `esp8266`: add support for this chip
+  - `machine/atsamd51x,runtime/atsamd51x`: fixes needed for full support for all PWM pins. Also adds some useful constants to clarify peripheral clock usage
+  - `machine/itsybitsy-nrf52840`: add support for Adafruit Itsybitsy nrf52840 (#1243)
+  - `machine/stm32f4`: refactor common code and add new build tag stm32f4 (#1332)
+  - `nrf`: add SoftDevice support for the Circuit Playground Bluefruit
+  - `nrf`: call sd_app_evt_wait when the SoftDevice is enabled
+  - `nrf52840`: add build tags for SoftDevice support
+  - `nrf52840`: use higher priority for USB-CDC code
+  - `runtime/atsamd51x`: use PCHCTRL_GCLK_SERCOMX_SLOW for setting clocks on all SERCOM ports
+  - `stm32f405`: add basic UART handler
+  - `stm32f405`: add STM32F405 machine/runtime, and new board/target feather-stm32f405
+* **build**
+  - `all`: run test binaries in the correct directory
+  - `build`: Fix arch release job
+  - `ci`: run `tinygo test` for known-working packages
+  - `ci`: set git-fetch-depth to 1
+  - `docker`: fix the problem with the wasm build (#1357)
+  - `Makefile`: check whether submodules have been downloaded in some common cases
+* **docs**
+  - add ESP32, ESP8266, and Adafruit Feather STM32F405 to list of supported boards
+
+0.14.1
+---
+* **command-line**
+  - support for Go 1.15
+* **compiler**
+  - loader:  work around Windows symlink limitation
+
 0.14.0
 ---
 * **command-line**
@@ -15,4 +15,4 @@ Ayke van Laethem <aykevanlaethem@gmail.com>
 Daniel Esteban <conejo@conejo.me>
 Loon, LLC.
 Ron Evans <ron@hybridgroup.com>
-Jaden Weiss <jaden@jadendw.dev>
+Nia Weiss <niaow1234@gmail.com>
@@ -1,5 +1,5 @@
-# TinyGo base stage installs Go 1.14, LLVM 10 and the TinyGo compiler itself.
-FROM golang:1.14 AS tinygo-base
+# TinyGo base stage installs the most recent Go 1.15.x, LLVM 10 and the TinyGo compiler itself.
+FROM golang:1.15 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-10 main" >> /etc/apt/sources.list && \
@@ -27,7 +27,10 @@ COPY --from=tinygo-base /go/bin/tinygo /go/bin/tinygo
 COPY --from=tinygo-base /tinygo/src /tinygo/src
 COPY --from=tinygo-base /tinygo/targets /tinygo/targets

-RUN apt-get install -y libllvm10 lld-10
+RUN cd /tinygo/ && \
+    apt-get update && \
+    apt-get install -y make clang-10 libllvm10 lld-10 && \
+    make wasi-libc

 # tinygo-avr stage installs the needed dependencies to compile TinyGo programs for AVR microcontrollers.
 FROM tinygo-base AS tinygo-avr
@@ -9,25 +9,10 @@ CLANG_SRC ?= $(LLVM_PROJECTDIR)/clang
 LLD_SRC ?= $(LLVM_PROJECTDIR)/lld

 # Try to autodetect LLVM build tools.
-ifneq (, $(shell command -v llvm-build/bin/clang 2> /dev/null))
-    CLANG ?= $(abspath llvm-build/bin/clang)
-else
-    CLANG ?= clang-10
-endif
-ifneq (, $(shell command -v llvm-build/bin/llvm-ar 2> /dev/null))
-    LLVM_AR ?= $(abspath llvm-build/bin/llvm-ar)
-else ifneq (, $(shell command -v llvm-ar-10 2> /dev/null))
-    LLVM_AR ?= llvm-ar-10
-else
-    LLVM_AR ?= llvm-ar
-endif
-ifneq (, $(shell command -v llvm-build/bin/llvm-nm 2> /dev/null))
-    LLVM_NM ?= $(abspath llvm-build/bin/llvm-nm)
-else ifneq (, $(shell command -v llvm-nm-10 2> /dev/null))
-    LLVM_NM ?= llvm-nm-10
-else
-    LLVM_NM ?= llvm-nm
-endif
+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))

 # Go binary and GOROOT to select
 GO ?= go
@@ -37,7 +22,7 @@ export GOROOT = $(shell $(GO) env GOROOT)
 MD5SUM = md5sum

 # tinygo binary for tests
-TINYGO ?= tinygo
+TINYGO ?= $(word 1,$(call detect,tinygo)$(call detect,build/tinygo))

 # Use CCACHE for LLVM if possible
 ifneq (, $(shell command -v ccache 2> /dev/null))
@@ -63,23 +48,13 @@ ifeq ($(OS),Windows_NT)
    # LLVM compiled using MinGW on Windows appears to have problems with threads.
    # Without this flag, linking results in errors like these:
    #     libLLVMSupport.a(Threading.cpp.obj):Threading.cpp:(.text+0x55): undefined reference to `std::thread::hardware_concurrency()'
-    LLVM_OPTION += -DLLVM_ENABLE_THREADS=OFF
+    LLVM_OPTION += -DLLVM_ENABLE_THREADS=OFF -DLLVM_ENABLE_PIC=OFF

+    CGO_CPPFLAGS += -DCINDEX_NO_EXPORTS
    CGO_LDFLAGS += -static -static-libgcc -static-libstdc++
    CGO_LDFLAGS_EXTRA += -lversion

-    # Build libclang manually because the CMake-based build system on Windows
-    # doesn't allow building libclang as a static library.
-    LIBCLANG_PATH = $(abspath build/libclang-custom.a)
-    LIBCLANG_FILES = $(abspath $(wildcard $(LLVM_BUILDDIR)/tools/clang/tools/libclang/CMakeFiles/libclang.dir/*.cpp.obj))
-
-    # Add the libclang dependency to the tinygo binary target.
-tinygo: $(LIBCLANG_PATH)
-test: $(LIBCLANG_PATH)
-    # Build libclang.
-$(LIBCLANG_PATH): $(LIBCLANG_FILES)
-	@mkdir -p build
-	ar rcs $(LIBCLANG_PATH) $^
+    LIBCLANG_PATH = $(abspath $(LLVM_BUILDDIR))/lib/liblibclang.a

 else ifeq ($(shell uname -s),Darwin)
    MD5SUM = md5
@@ -102,9 +77,9 @@ LLD_LIBS = $(START_GROUP) -llldCOFF -llldCommon -llldCore -llldDriver -llldELF -

 # For static linking.
 ifneq ("$(wildcard $(LLVM_BUILDDIR)/bin/llvm-config*)","")
-    CGO_CPPFLAGS=$(shell $(LLVM_BUILDDIR)/bin/llvm-config --cppflags) -I$(abspath $(LLVM_BUILDDIR))/tools/clang/include -I$(abspath $(CLANG_SRC))/include -I$(abspath $(LLD_SRC))/include
+    CGO_CPPFLAGS+=$(shell $(LLVM_BUILDDIR)/bin/llvm-config --cppflags) -I$(abspath $(LLVM_BUILDDIR))/tools/clang/include -I$(abspath $(CLANG_SRC))/include -I$(abspath $(LLD_SRC))/include
    CGO_CXXFLAGS=-std=c++14
-    CGO_LDFLAGS+=$(LIBCLANG_PATH) -std=c++14 -L$(abspath $(LLVM_BUILDDIR)/lib) $(CLANG_LIBS) $(LLD_LIBS) $(shell $(LLVM_BUILDDIR)/bin/llvm-config --ldflags --libs --system-libs $(LLVM_COMPONENTS)) -lstdc++ $(CGO_LDFLAGS_EXTRA)
+    CGO_LDFLAGS+=$(LIBCLANG_PATH) -L$(abspath $(LLVM_BUILDDIR)/lib) $(CLANG_LIBS) $(LLD_LIBS) $(shell $(LLVM_BUILDDIR)/bin/llvm-config --ldflags --libs --system-libs $(LLVM_COMPONENTS)) -lstdc++ $(CGO_LDFLAGS_EXTRA)
 endif


@@ -118,9 +93,10 @@ fmt-check:
 	@unformatted=$$(gofmt -l $(FMT_PATHS)); [ -z "$$unformatted" ] && exit 0; echo "Unformatted:"; for fn in $$unformatted; do echo "  $$fn"; done; exit 1


-gen-device: gen-device-avr gen-device-nrf gen-device-sam gen-device-sifive gen-device-stm32 gen-device-kendryte gen-device-nxp
+gen-device: gen-device-avr gen-device-esp gen-device-nrf gen-device-sam gen-device-sifive gen-device-stm32 gen-device-kendryte gen-device-nxp

 gen-device-avr:
+	@if [ ! -e lib/avr/README.md ]; then echo "Submodules have not been downloaded. Please download them using:\n  git submodule update --init"; exit 1; fi
 	$(GO) build -o ./build/gen-device-avr ./tools/gen-device-avr/
 	./build/gen-device-avr lib/avr/packs/atmega src/device/avr/
 	./build/gen-device-avr lib/avr/packs/tiny src/device/avr/
@@ -129,6 +105,10 @@ gen-device-avr:
 build/gen-device-svd: ./tools/gen-device-svd/*.go
 	$(GO) build -o $@ ./tools/gen-device-svd/

+gen-device-esp: build/gen-device-svd
+	./build/gen-device-svd -source=https://github.com/posborne/cmsis-svd/tree/master/data/Espressif-Community -interrupts=software lib/cmsis-svd/data/Espressif-Community/ src/device/esp/
+	GO111MODULE=off $(GO) fmt ./src/device/esp
+
 gen-device-nrf: build/gen-device-svd
 	./build/gen-device-svd -source=https://github.com/NordicSemiconductor/nrfx/tree/master/mdk lib/nrfx/mdk/ src/device/nrf/
 	GO111MODULE=off $(GO) fmt ./src/device/nrf
@@ -156,13 +136,13 @@ gen-device-stm32: build/gen-device-svd

 # Get LLVM sources.
 $(LLVM_PROJECTDIR)/README.md:
-	git clone -b release/10.x --depth=1 https://github.com/llvm/llvm-project $(LLVM_PROJECTDIR)
+	git clone -b xtensa_release_11.0.0 --depth=1 https://github.com/tinygo-org/llvm-project $(LLVM_PROJECTDIR)
 llvm-source: $(LLVM_PROJECTDIR)/README.md

 # 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" -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_PROJECTS="clang;lld" -DLLVM_TOOL_CLANG_TOOLS_EXTRA_BUILD=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_PROJECTS="clang;lld" -DLLVM_TOOL_CLANG_TOOLS_EXTRA_BUILD=OFF $(LLVM_OPTION)

 # Build LLVM.
 $(LLVM_BUILDDIR): $(LLVM_BUILDDIR)/build.ninja
@@ -173,19 +153,37 @@ $(LLVM_BUILDDIR): $(LLVM_BUILDDIR)/build.ninja
 .PHONY: wasi-libc
 wasi-libc: lib/wasi-libc/sysroot/lib/wasm32-wasi/libc.a
 lib/wasi-libc/sysroot/lib/wasm32-wasi/libc.a:
+	@if [ ! -e lib/wasi-libc/Makefile ]; then echo "Submodules have not been downloaded. Please download them using:\n  git submodule update --init"; exit 1; fi
 	cd lib/wasi-libc && make -j4 WASM_CC=$(CLANG) WASM_AR=$(LLVM_AR) WASM_NM=$(LLVM_NM)


 # Build the Go compiler.
 tinygo:
 	@if [ ! -f "$(LLVM_BUILDDIR)/bin/llvm-config" ]; then echo "Fetch and build LLVM first by running:"; echo "  make llvm-source"; echo "  make $(LLVM_BUILDDIR)"; exit 1; fi
-	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GO) build -o build/tinygo$(EXE) -tags byollvm .
+	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GO) build -buildmode exe -o build/tinygo$(EXE) -tags byollvm -ldflags="-X main.gitSha1=`git rev-parse --short HEAD`" .

 test: wasi-libc
-	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GO) test -v -tags byollvm ./cgo ./compileopts ./interp ./transform .
+	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GO) test -v -buildmode exe -tags byollvm ./cgo ./compileopts ./interp ./transform .

+# Test known-working standard library packages.
+# TODO: do this in one command, parallelize, and only show failing tests (no
+# implied -v flag).
+.PHONY: tinygo-test
 tinygo-test:
-	cd tests/tinygotest && tinygo test
+	$(TINYGO) test container/heap
+	$(TINYGO) test container/list
+	$(TINYGO) test container/ring
+	$(TINYGO) test crypto/des
+	$(TINYGO) test encoding/ascii85
+	$(TINYGO) test encoding/base32
+	$(TINYGO) test encoding/hex
+	$(TINYGO) test hash/adler32
+	$(TINYGO) test hash/fnv
+	$(TINYGO) test hash/crc64
+	$(TINYGO) test math
+	$(TINYGO) test math/cmplx
+	$(TINYGO) test text/scanner
+	$(TINYGO) test unicode/utf8

 .PHONY: smoketest
 smoketest:
@@ -222,8 +220,6 @@ smoketest:
 	# test simulated boards on play.tinygo.org
 	$(TINYGO) build             -o test.wasm -tags=arduino              examples/blinky1
 	@$(MD5SUM) test.wasm
-	$(TINYGO) build             -o test.wasm -tags=hifive1-qemu         examples/serial
-	@$(MD5SUM) test.wasm
 	$(TINYGO) build             -o test.wasm -tags=hifive1b             examples/blinky1
 	@$(MD5SUM) test.wasm
 	$(TINYGO) build             -o test.wasm -tags=reelboard            examples/blinky1
@@ -267,13 +263,15 @@ smoketest:
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=stm32f4disco        examples/blinky2
 	@$(MD5SUM) test.hex
-	$(TINYGO) build -size short -o test.hex -target=stm32f4disco-1       examples/blinky1
+	$(TINYGO) build -size short -o test.hex -target=stm32f4disco-1      examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=feather-stm32f405   examples/blinky1
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=circuitplay-bluefruit examples/blinky1
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=circuitplay-express examples/i2s
 	@$(MD5SUM) test.hex
-	$(TINYGO) build -size short -o test.hex -target=clue_alpha          examples/blinky1
+	$(TINYGO) build -size short -o test.hex -target=clue-alpha          examples/blinky1
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.gba -target=gameboy-advance     examples/gba-display
 	@$(MD5SUM) test.gba
@@ -315,6 +313,25 @@ smoketest:
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=feather-nrf52840  	examples/blinky1
 	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=itsybitsy-nrf52840  examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=qtpy  				examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=teensy40            examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=teensy36            examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=nucleo-f722ze       examples/blinky1
+	@$(MD5SUM) test.hex
+	# test pwm
+	$(TINYGO) build -size short -o test.hex -target=itsybitsy-m0        examples/pwm
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=itsybitsy-m4        examples/pwm
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=feather-m4          examples/pwm
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pyportal            examples/pwm
+	@$(MD5SUM) test.hex
 ifneq ($(AVR), 0)
 	$(TINYGO) build -size short -o test.hex -target=atmega1284p         examples/serial
 	@$(MD5SUM) test.hex
@@ -330,9 +347,17 @@ ifneq ($(AVR), 0)
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=digispark -gc=leaking examples/blinky1
 	@$(MD5SUM) test.hex
+endif
+ifneq ($(XTENSA), 0)
+	$(TINYGO) build -size short -o test.bin -target=esp32-mini32      	examples/blinky1
+	@$(MD5SUM) test.bin
+	$(TINYGO) build -size short -o test.bin -target=nodemcu             examples/blinky1
+	@$(MD5SUM) test.bin
 endif
 	$(TINYGO) build -size short -o test.hex -target=hifive1b            examples/blinky1
 	@$(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
 	$(TINYGO) build             -o wasm.wasm -target=wasm               examples/wasm/export
@@ -342,8 +367,8 @@ endif
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=pca10040 -opt=1     examples/blinky1
 	@$(MD5SUM) test.hex
-	$(TINYGO) build             -o test.elf -target=nintendoswitch      examples/serial
-	@$(MD5SUM) test.elf
+	$(TINYGO) build             -o test.nro -target=nintendoswitch      examples/serial
+	@$(MD5SUM) test.nro

 wasmtest:
 	$(GO) test ./tests/wasm
@@ -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 39 microcontroller boards/devices are currently supported:
+The following 44 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)
@@ -51,8 +51,10 @@ The following 39 microcontroller boards/devices are currently supported:
 * [Adafruit Feather M0](https://www.adafruit.com/product/2772)
 * [Adafruit Feather M4](https://www.adafruit.com/product/3857)
 * [Adafruit Feather nRF52840 Express](https://www.adafruit.com/product/4062)
+* [Adafruit Feather STM32F405 Express](https://www.adafruit.com/product/4382)
 * [Adafruit ItsyBitsy M0](https://www.adafruit.com/product/3727)
 * [Adafruit ItsyBitsy M4](https://www.adafruit.com/product/3800)
+* [Adafruit ItsyBitsy nRF52840](https://www.adafruit.com/product/4481)
 * [Adafruit Metro M4 Express Airlift](https://www.adafruit.com/product/4000)
 * [Adafruit PyBadge](https://www.adafruit.com/product/4200)
 * [Adafruit PyGamer](https://www.adafruit.com/product/4242)
@@ -62,8 +64,11 @@ The following 39 microcontroller boards/devices are currently supported:
 * [Arduino Nano](https://store.arduino.cc/arduino-nano)
 * [Arduino Nano33 IoT](https://store.arduino.cc/nano-33-iot)
 * [Arduino Uno](https://store.arduino.cc/arduino-uno-rev3)
+* [Arduino Zero](https://store.arduino.cc/usa/arduino-zero)
 * [BBC micro:bit](https://microbit.org/)
 * [Digispark](http://digistump.com/products/1)
+* [ESP32](https://www.espressif.com/en/products/socs/esp32)
+* [ESP8266](https://www.espressif.com/en/products/socs/esp8266)
 * [Game Boy Advance](https://en.wikipedia.org/wiki/Game_Boy_Advance)
 * [Makerdiary nRF52840-MDK](https://wiki.makerdiary.com/nrf52840-mdk/)
 * [Nintendo Switch](https://www.nintendo.com/switch/)
@@ -143,6 +148,6 @@ The original reasoning was: if [Python](https://micropython.org/) can run on mic

 This project is licensed under the BSD 3-clause license, just like the [Go project](https://golang.org/LICENSE) itself.

-Some code has been copied from the LLVM project and is therefore licensed under [a variant of the Apache 2.0 license](http://releases.llvm.org/10.0.0/LICENSE.TXT). This has been clearly indicated in the header of these files.
+Some code has been copied from the LLVM project and is therefore licensed under [a variant of the Apache 2.0 license](http://releases.llvm.org/11.0.0/LICENSE.TXT). This has been clearly indicated in the header of these files.

 Some code has been copied and/or ported from Paul Stoffregen's Teensy libraries and is therefore licensed under PJRC's license. This has been clearly indicated in the header of these files.
@@ -12,22 +12,27 @@ jobs:
  steps:
    - task: GoTool@0
      inputs:
-        version: '1.14.1'
+        version: '1.15'
    - checkout: self
-    - task: CacheBeta@0
+      fetchDepth: 1
+    - task: Cache@2
      displayName: Cache LLVM source
      inputs:
-        key: llvm-source-10-windows-v0
+        key: llvm-source-11-windows-v1
        path: llvm-project
    - task: Bash@3
      displayName: Download LLVM source
      inputs:
        targetType: inline
-        script: make llvm-source
+        script: |
+          make llvm-source
+          # Workaround for bad symlinks:
+          # https://github.com/microsoft/azure-pipelines-tasks/issues/13418
+          rm -f llvm-project/libcxx/test/std/input.output/filesystems/Inputs/static_test_env/bad_symlink
    - task: CacheBeta@0
      displayName: Cache LLVM build
      inputs:
-        key: llvm-build-10-windows-v0
+        key: llvm-build-11-windows-v3
        path: llvm-build
    - task: Bash@3
      displayName: Build LLVM
@@ -36,7 +41,11 @@ jobs:
        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
@@ -47,7 +56,7 @@ jobs:
    - task: CacheBeta@0
      displayName: Cache wasi-libc sysroot
      inputs:
-        key: wasi-libc-sysroot-v2
+        key: wasi-libc-sysroot-v3
        path: lib/wasi-libc/sysroot
    - task: Bash@3
      displayName: Build wasi-libc
@@ -80,4 +89,4 @@ jobs:
        script: |
          export PATH="$PATH:./llvm-build/bin:/c/Program Files/qemu"
          unset GOROOT
-          make smoketest TINYGO=build/tinygo AVR=0
+          make smoketest TINYGO=build/tinygo AVR=0 XTENSA=0
@@ -5,6 +5,7 @@ package builder

 import (
 	"debug/elf"
+	"encoding/binary"
 	"errors"
 	"fmt"
 	"io/ioutil"
@@ -23,22 +24,35 @@ import (
 	"tinygo.org/x/go-llvm"
 )

+// BuildResult is the output of a build. This includes the binary itself and
+// some other metadata that is obtained while building the binary.
+type BuildResult struct {
+	// A path to the output binary. It will be removed after Build returns, so
+	// if it should be kept it must be copied or moved away.
+	Binary string
+
+	// The directory of the main package. This is useful for testing as the test
+	// binary must be run in the directory of the tested package.
+	MainDir string
+}
+
 // Build performs a single package to executable Go build. It takes in a package
 // name, an output path, and set of compile options and from that it manages the
 // whole compilation process.
 //
 // The error value may be of type *MultiError. Callers will likely want to check
 // for this case and print such errors individually.
-func Build(pkgName, outpath string, config *compileopts.Config, action func(string) error) error {
+func Build(pkgName, outpath string, config *compileopts.Config, action func(BuildResult) error) error {
 	// Compile Go code to IR.
 	machine, err := compiler.NewTargetMachine(config)
 	if err != nil {
 		return err
 	}
-	mod, extraFiles, extraLDFlags, errs := compiler.Compile(pkgName, machine, config)
+	buildOutput, errs := compiler.Compile(pkgName, machine, config)
 	if errs != nil {
 		return newMultiError(errs)
 	}
+	mod := buildOutput.Mod

 	if config.Options.PrintIR {
 		fmt.Println("; Generated LLVM IR:")
@@ -65,7 +79,7 @@ func Build(pkgName, outpath string, config *compileopts.Config, action func(stri
 	// keep functions interoperable, pass int64 types as pointers to
 	// stack-allocated values.
 	// Use -wasm-abi=generic to disable this behaviour.
-	if config.Options.WasmAbi == "js" && strings.HasPrefix(config.Triple(), "wasm") {
+	if config.WasmAbi() == "js" {
 		err := transform.ExternalInt64AsPtr(mod)
 		if err != nil {
 			return err
@@ -103,16 +117,22 @@ func Build(pkgName, outpath string, config *compileopts.Config, action func(stri
 		return errors.New("verification failure after LLVM optimization passes")
 	}

-	// On the AVR, pointers can point either to flash or to RAM, but we don't
-	// know. As a temporary fix, load all global variables in RAM.
-	// In the future, there should be a compiler pass that determines which
-	// pointers are flash and which are in RAM so that pointers can have a
-	// correct address space parameter (address space 1 is for flash).
-	if strings.HasPrefix(config.Triple(), "avr") {
-		transform.NonConstGlobals(mod)
-		if err := llvm.VerifyModule(mod, llvm.PrintMessageAction); err != nil {
-			return errors.New("verification error after making all globals non-constant on AVR")
-		}
+	// LLVM 11 by default tries to emit tail calls (even with the target feature
+	// disabled) unless it is explicitly disabled with a function attribute.
+	// This is a problem, as it tries to emit them and prints an error when it
+	// can't with this feature disabled.
+	// Because as of september 2020 tail calls are not yet widely supported,
+	// they need to be disabled until they are widely supported (at which point
+	// the +tail-call target feautre can be set).
+	if strings.HasPrefix(config.Triple(), "wasm") {
+		transform.DisableTailCalls(mod)
+	}
+
+	// Make sure stack sizes are loaded from a separate section so they can be
+	// modified after linking.
+	var stackSizeLoads []string
+	if config.AutomaticStackSize() {
+		stackSizeLoads = transform.CreateStackSizeLoads(mod, config)
 	}

 	// Generate output.
@@ -188,7 +208,7 @@ func Build(pkgName, outpath string, config *compileopts.Config, action func(stri
 		}

 		// Compile C files in packages.
-		for i, file := range extraFiles {
+		for i, file := range buildOutput.ExtraFiles {
 			outpath := filepath.Join(dir, "pkg"+strconv.Itoa(i)+"-"+filepath.Base(file)+".o")
 			err := runCCompiler(config.Target.Compiler, append(config.CFlags(), "-c", "-o", outpath, file)...)
 			if err != nil {
@@ -197,8 +217,8 @@ func Build(pkgName, outpath string, config *compileopts.Config, action func(stri
 			ldflags = append(ldflags, outpath)
 		}

-		if len(extraLDFlags) > 0 {
-			ldflags = append(ldflags, extraLDFlags...)
+		if len(buildOutput.ExtraLDFlags) > 0 {
+			ldflags = append(ldflags, buildOutput.ExtraLDFlags...)
 		}

 		// Link the object files together.
@@ -207,6 +227,26 @@ func Build(pkgName, outpath string, config *compileopts.Config, action func(stri
 			return &commandError{"failed to link", executable, err}
 		}

+		var calculatedStacks []string
+		var stackSizes map[string]functionStackSize
+		if config.Options.PrintStacks || config.AutomaticStackSize() {
+			// Try to determine stack sizes at compile time.
+			// Don't do this by default as it usually doesn't work on
+			// unsupported architectures.
+			calculatedStacks, stackSizes, err = determineStackSizes(mod, executable)
+			if err != nil {
+				return err
+			}
+		}
+		if config.AutomaticStackSize() {
+			// Modify the .tinygo_stacksizes section that contains a stack size
+			// for each goroutine.
+			err = modifyStackSizes(executable, stackSizeLoads, stackSizes)
+			if err != nil {
+				return fmt.Errorf("could not modify stack sizes: %w", err)
+			}
+		}
+
 		if config.Options.PrintSizes == "short" || config.Options.PrintSizes == "full" {
 			sizes, err := loadProgramSize(executable)
 			if err != nil {
@@ -228,28 +268,250 @@ func Build(pkgName, outpath string, config *compileopts.Config, action func(stri

 		// Print goroutine stack sizes, as far as possible.
 		if config.Options.PrintStacks {
-			printStacks(mod, executable)
+			printStacks(calculatedStacks, stackSizes)
 		}

 		// Get an Intel .hex file or .bin file from the .elf file.
-		if outext == ".hex" || outext == ".bin" || outext == ".gba" {
+		outputBinaryFormat := config.BinaryFormat(outext)
+		switch outputBinaryFormat {
+		case "elf":
+			// do nothing, file is already in ELF format
+		case "hex", "bin":
+			// Extract raw binary, either encoding it as a hex file or as a raw
+			// firmware file.
 			tmppath = filepath.Join(dir, "main"+outext)
-			err := objcopy(executable, tmppath)
+			err := objcopy(executable, tmppath, outputBinaryFormat)
 			if err != nil {
 				return err
 			}
-		} else if outext == ".uf2" {
+		case "uf2":
 			// Get UF2 from the .elf file.
 			tmppath = filepath.Join(dir, "main"+outext)
 			err := convertELFFileToUF2File(executable, tmppath, config.Target.UF2FamilyID)
 			if err != nil {
 				return err
 			}
+		case "esp32", "esp8266":
+			// Special format for the ESP family of chips (parsed by the ROM
+			// bootloader).
+			tmppath = filepath.Join(dir, "main"+outext)
+			err := makeESPFirmareImage(executable, tmppath, outputBinaryFormat)
+			if err != nil {
+				return err
+			}
+		default:
+			return fmt.Errorf("unknown output binary format: %s", outputBinaryFormat)
 		}
-		return action(tmppath)
+		return action(BuildResult{
+			Binary:  tmppath,
+			MainDir: buildOutput.MainDir,
+		})
 	}
 }

+// functionStackSizes keeps stack size information about a single function
+// (usually a goroutine).
+type functionStackSize struct {
+	humanName        string
+	stackSize        uint64
+	stackSizeType    stacksize.SizeType
+	missingStackSize *stacksize.CallNode
+}
+
+// determineStackSizes tries to determine the stack sizes of all started
+// goroutines and of the reset vector. The LLVM module is necessary to find
+// functions that call a function pointer.
+func determineStackSizes(mod llvm.Module, executable string) ([]string, map[string]functionStackSize, error) {
+	var callsIndirectFunction []string
+	gowrappers := []string{}
+	gowrapperNames := make(map[string]string)
+	for fn := mod.FirstFunction(); !fn.IsNil(); fn = llvm.NextFunction(fn) {
+		// Determine which functions call a function pointer.
+		for bb := fn.FirstBasicBlock(); !bb.IsNil(); bb = llvm.NextBasicBlock(bb) {
+			for inst := bb.FirstInstruction(); !inst.IsNil(); inst = llvm.NextInstruction(inst) {
+				if inst.IsACallInst().IsNil() {
+					continue
+				}
+				if callee := inst.CalledValue(); callee.IsAFunction().IsNil() && callee.IsAInlineAsm().IsNil() {
+					callsIndirectFunction = append(callsIndirectFunction, fn.Name())
+				}
+			}
+		}
+
+		// Get a list of "go wrappers", small wrapper functions that decode
+		// parameters when starting a new goroutine.
+		attr := fn.GetStringAttributeAtIndex(-1, "tinygo-gowrapper")
+		if !attr.IsNil() {
+			gowrappers = append(gowrappers, fn.Name())
+			gowrapperNames[fn.Name()] = attr.GetStringValue()
+		}
+	}
+	sort.Strings(gowrappers)
+
+	// Load the ELF binary.
+	f, err := elf.Open(executable)
+	if err != nil {
+		return nil, nil, fmt.Errorf("could not load executable for stack size analysis: %w", err)
+	}
+	defer f.Close()
+
+	// Determine the frame size of each function (if available) and the callgraph.
+	functions, err := stacksize.CallGraph(f, callsIndirectFunction)
+	if err != nil {
+		return nil, nil, fmt.Errorf("could not parse executable for stack size analysis: %w", err)
+	}
+
+	// Goroutines need to be started and finished and take up some stack space
+	// that way. This can be measured by measuing the stack size of
+	// tinygo_startTask.
+	if numFuncs := len(functions["tinygo_startTask"]); numFuncs != 1 {
+		return nil, nil, fmt.Errorf("expected exactly one definition of tinygo_startTask, got %d", numFuncs)
+	}
+	baseStackSize, baseStackSizeType, baseStackSizeFailedAt := functions["tinygo_startTask"][0].StackSize()
+
+	sizes := make(map[string]functionStackSize)
+
+	// Add the reset handler function, for convenience. The reset handler runs
+	// startup code and the scheduler. The listed stack size is not the full
+	// stack size: interrupts are not counted.
+	var resetFunction string
+	switch f.Machine {
+	case elf.EM_ARM:
+		// Note: all interrupts happen on this stack so the real size is bigger.
+		resetFunction = "Reset_Handler"
+	}
+	if resetFunction != "" {
+		funcs := functions[resetFunction]
+		if len(funcs) != 1 {
+			return nil, nil, fmt.Errorf("expected exactly one definition of %s in the callgraph, found %d", resetFunction, len(funcs))
+		}
+		stackSize, stackSizeType, missingStackSize := funcs[0].StackSize()
+		sizes[resetFunction] = functionStackSize{
+			stackSize:        stackSize,
+			stackSizeType:    stackSizeType,
+			missingStackSize: missingStackSize,
+			humanName:        resetFunction,
+		}
+	}
+
+	// Add all goroutine wrapper functions.
+	for _, name := range gowrappers {
+		funcs := functions[name]
+		if len(funcs) != 1 {
+			return nil, nil, fmt.Errorf("expected exactly one definition of %s in the callgraph, found %d", name, len(funcs))
+		}
+		humanName := gowrapperNames[name]
+		if humanName == "" {
+			humanName = name // fallback
+		}
+		stackSize, stackSizeType, missingStackSize := funcs[0].StackSize()
+		if baseStackSizeType != stacksize.Bounded {
+			// It was not possible to determine the stack size at compile time
+			// because tinygo_startTask does not have a fixed stack size. This
+			// can happen when using -opt=1.
+			stackSizeType = baseStackSizeType
+			missingStackSize = baseStackSizeFailedAt
+		} else if stackSize < baseStackSize {
+			// This goroutine has a very small stack, but still needs to fit all
+			// registers to start and suspend the goroutine. Otherwise a stack
+			// overflow will occur even before the goroutine is started.
+			stackSize = baseStackSize
+		}
+		sizes[name] = functionStackSize{
+			stackSize:        stackSize,
+			stackSizeType:    stackSizeType,
+			missingStackSize: missingStackSize,
+			humanName:        humanName,
+		}
+	}
+
+	if resetFunction != "" {
+		return append([]string{resetFunction}, gowrappers...), sizes, nil
+	}
+	return gowrappers, sizes, nil
+}
+
+// modifyStackSizes modifies the .tinygo_stacksizes section with the updated
+// stack size information. Before this modification, all stack sizes in the
+// section assume the default stack size (which is relatively big).
+func modifyStackSizes(executable string, stackSizeLoads []string, stackSizes map[string]functionStackSize) error {
+	fp, err := os.OpenFile(executable, os.O_RDWR, 0)
+	if err != nil {
+		return err
+	}
+	defer fp.Close()
+
+	elfFile, err := elf.NewFile(fp)
+	if err != nil {
+		return err
+	}
+
+	section := elfFile.Section(".tinygo_stacksizes")
+	if section == nil {
+		return errors.New("could not find .tinygo_stacksizes section")
+	}
+
+	if section.Size != section.FileSize {
+		// Sanity check.
+		return fmt.Errorf("expected .tinygo_stacksizes to have identical size and file size, got %d and %d", section.Size, section.FileSize)
+	}
+
+	// Read all goroutine stack sizes.
+	data := make([]byte, section.Size)
+	_, err = fp.ReadAt(data, int64(section.Offset))
+	if err != nil {
+		return err
+	}
+
+	if len(stackSizeLoads)*4 != len(data) {
+		// Note: while AVR should use 2 byte stack sizes, even 64-bit platforms
+		// should probably stick to 4 byte stack sizes as a larger than 4GB
+		// stack doesn't make much sense.
+		return errors.New("expected 4 byte stack sizes")
+	}
+
+	// Modify goroutine stack sizes with a compile-time known worst case stack
+	// size.
+	for i, name := range stackSizeLoads {
+		fn, ok := stackSizes[name]
+		if !ok {
+			return fmt.Errorf("could not find symbol %s in ELF file", name)
+		}
+		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 elfFile.Machine {
+			case elf.EM_ARM:
+				// 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
+				// (MSP).
+				// Some background:
+				// https://interrupt.memfault.com/blog/cortex-m-rtos-context-switching
+				stackSize += 32
+			}
+
+			// Finally write the stack size to the binary.
+			binary.LittleEndian.PutUint32(data[i*4:], stackSize)
+		}
+	}
+
+	// Write back the modified stack sizes.
+	_, err = fp.WriteAt(data, int64(section.Offset))
+	if err != nil {
+		return err
+	}
+
+	return nil
+}
+
 // printStacks prints the maximum stack depth for functions that are started as
 // goroutines. Stack sizes cannot always be determined statically, in particular
 // recursive functions and functions that call interface methods or function
@@ -260,78 +522,22 @@ func Build(pkgName, outpath string, config *compileopts.Config, action func(stri
 //
 //     function                         stack usage (in bytes)
 //     Reset_Handler                    316
-//     .Lexamples/blinky2.led1          92
-//     .Lruntime.run$1                  300
-func printStacks(mod llvm.Module, executable string) {
-	// Determine which functions call a function pointer.
-	var callsIndirectFunction []string
-	for fn := mod.FirstFunction(); !fn.IsNil(); fn = llvm.NextFunction(fn) {
-		for bb := fn.FirstBasicBlock(); !bb.IsNil(); bb = llvm.NextBasicBlock(bb) {
-			for inst := bb.FirstInstruction(); !inst.IsNil(); inst = llvm.NextInstruction(inst) {
-				if inst.IsACallInst().IsNil() {
-					continue
-				}
-				if callee := inst.CalledValue(); callee.IsAFunction().IsNil() && callee.IsAInlineAsm().IsNil() {
-					callsIndirectFunction = append(callsIndirectFunction, fn.Name())
-				}
-			}
-		}
-	}
-
-	// Load the ELF binary.
-	f, err := elf.Open(executable)
-	if err != nil {
-		fmt.Fprintln(os.Stderr, "could not load executable for stack size analysis:", err)
-		return
-	}
-	defer f.Close()
-
-	// Determine the frame size of each function (if available) and the callgraph.
-	functions, err := stacksize.CallGraph(f, callsIndirectFunction)
-	if err != nil {
-		fmt.Fprintln(os.Stderr, "could not parse executable for stack size analysis:", err)
-		return
-	}
-
-	// Get a list of "go wrappers", small wrapper functions that decode
-	// parameters when starting a new goroutine.
-	var gowrappers []string
-	for name := range functions {
-		if strings.HasSuffix(name, "$gowrapper") {
-			gowrappers = append(gowrappers, name)
-		}
-	}
-	sort.Strings(gowrappers)
-
-	switch f.Machine {
-	case elf.EM_ARM:
-		// Add the reset handler, which runs startup code and is the
-		// interrupt/scheduler stack with -scheduler=tasks.
-		// Note that because interrupts happen on this stack, the stack needed
-		// by just the Reset_Handler is not enough. Stacks needed by interrupt
-		// handlers should also be taken into account.
-		gowrappers = append([]string{"Reset_Handler"}, gowrappers...)
-	}
-
+//     examples/blinky2.led1            92
+//     runtime.run$1                    300
+func printStacks(calculatedStacks []string, stackSizes map[string]functionStackSize) {
 	// Print the sizes of all stacks.
 	fmt.Printf("%-32s %s\n", "function", "stack usage (in bytes)")
-	for _, name := range gowrappers {
-		for _, fn := range functions[name] {
-			stackSize, stackSizeType, missingStackSize := fn.StackSize()
-			strippedName := name
-			if strings.HasSuffix(name, "$gowrapper") {
-				strippedName = name[:len(name)-len("$gowrapper")]
-			}
-			switch stackSizeType {
-			case stacksize.Bounded:
-				fmt.Printf("%-32s %d\n", strippedName, stackSize)
-			case stacksize.Unknown:
-				fmt.Printf("%-32s unknown, %s does not have stack frame information\n", strippedName, missingStackSize)
-			case stacksize.Recursive:
-				fmt.Printf("%-32s recursive, %s may call itself\n", strippedName, missingStackSize)
-			case stacksize.IndirectCall:
-				fmt.Printf("%-32s unknown, %s calls a function pointer\n", strippedName, missingStackSize)
-			}
+	for _, name := range calculatedStacks {
+		fn := stackSizes[name]
+		switch fn.stackSizeType {
+		case stacksize.Bounded:
+			fmt.Printf("%-32s %d\n", fn.humanName, fn.stackSize)
+		case stacksize.Unknown:
+			fmt.Printf("%-32s unknown, %s does not have stack frame information\n", fn.humanName, fn.missingStackSize)
+		case stacksize.Recursive:
+			fmt.Printf("%-32s recursive, %s may call itself\n", fn.humanName, fn.missingStackSize)
+		case stacksize.IndirectCall:
+			fmt.Printf("%-32s unknown, %s calls a function pointer\n", fn.humanName, fn.missingStackSize)
 		}
 	}
 }
@@ -101,7 +101,7 @@ bool AssemblerInvocation::CreateFromArgs(AssemblerInvocation &Opts,

  // Target Options
  Opts.Triple = llvm::Triple::normalize(Args.getLastArgValue(OPT_triple));
-  Opts.CPU = Args.getLastArgValue(OPT_target_cpu);
+  Opts.CPU = std::string(Args.getLastArgValue(OPT_target_cpu));
  Opts.Features = Args.getAllArgValues(OPT_target_feature);

  // Use the default target triple if unspecified.
@@ -132,13 +132,19 @@ bool AssemblerInvocation::CreateFromArgs(AssemblerInvocation &Opts,

  Opts.RelaxELFRelocations = Args.hasArg(OPT_mrelax_relocations);
  Opts.DwarfVersion = getLastArgIntValue(Args, OPT_dwarf_version_EQ, 2, Diags);
-  Opts.DwarfDebugFlags = Args.getLastArgValue(OPT_dwarf_debug_flags);
-  Opts.DwarfDebugProducer = Args.getLastArgValue(OPT_dwarf_debug_producer);
-  Opts.DebugCompilationDir = Args.getLastArgValue(OPT_fdebug_compilation_dir);
-  Opts.MainFileName = Args.getLastArgValue(OPT_main_file_name);
+  Opts.DwarfDebugFlags =
+      std::string(Args.getLastArgValue(OPT_dwarf_debug_flags));
+  Opts.DwarfDebugProducer =
+      std::string(Args.getLastArgValue(OPT_dwarf_debug_producer));
+  Opts.DebugCompilationDir =
+      std::string(Args.getLastArgValue(OPT_fdebug_compilation_dir));
+  Opts.MainFileName = std::string(Args.getLastArgValue(OPT_main_file_name));

-  for (const auto &Arg : Args.getAllArgValues(OPT_fdebug_prefix_map_EQ))
-    Opts.DebugPrefixMap.insert(StringRef(Arg).split('='));
+  for (const auto &Arg : Args.getAllArgValues(OPT_fdebug_prefix_map_EQ)) {
+    auto Split = StringRef(Arg).split('=');
+    Opts.DebugPrefixMap.insert(
+        {std::string(Split.first), std::string(Split.second)});
+  }

  // Frontend Options
  if (Args.hasArg(OPT_INPUT)) {
@@ -154,8 +160,9 @@ bool AssemblerInvocation::CreateFromArgs(AssemblerInvocation &Opts,
    }
  }
  Opts.LLVMArgs = Args.getAllArgValues(OPT_mllvm);
-  Opts.OutputPath = Args.getLastArgValue(OPT_o);
-  Opts.SplitDwarfOutput = Args.getLastArgValue(OPT_split_dwarf_output);
+  Opts.OutputPath = std::string(Args.getLastArgValue(OPT_o));
+  Opts.SplitDwarfOutput =
+      std::string(Args.getLastArgValue(OPT_split_dwarf_output));
  if (Arg *A = Args.getLastArg(OPT_filetype)) {
    StringRef Name = A->getValue();
    unsigned OutputType = StringSwitch<unsigned>(Name)
@@ -183,8 +190,9 @@ bool AssemblerInvocation::CreateFromArgs(AssemblerInvocation &Opts,
  Opts.NoExecStack = Args.hasArg(OPT_mno_exec_stack);
  Opts.FatalWarnings = Args.hasArg(OPT_massembler_fatal_warnings);
  Opts.NoWarn = Args.hasArg(OPT_massembler_no_warn);
-  Opts.RelocationModel = Args.getLastArgValue(OPT_mrelocation_model, "pic");
-  Opts.TargetABI = Args.getLastArgValue(OPT_target_abi);
+  Opts.RelocationModel =
+      std::string(Args.getLastArgValue(OPT_mrelocation_model, "pic"));
+  Opts.TargetABI = std::string(Args.getLastArgValue(OPT_target_abi));
  Opts.IncrementalLinkerCompatible =
      Args.hasArg(OPT_mincremental_linker_compatible);
  Opts.SymbolDefs = Args.getAllArgValues(OPT_defsym);
@@ -314,12 +322,7 @@ bool ExecuteAssembler(AssemblerInvocation &Opts, DiagnosticsEngine &Diags) {
                            SrcMgr.getMemoryBuffer(BufferIndex)->getBuffer());

  // Build up the feature string from the target feature list.
-  std::string FS;
-  if (!Opts.Features.empty()) {
-    FS = Opts.Features[0];
-    for (unsigned i = 1, e = Opts.Features.size(); i != e; ++i)
-      FS += "," + Opts.Features[i];
-  }
+  std::string FS = llvm::join(Opts.Features, ",");

  std::unique_ptr<MCStreamer> Str;

@@ -383,7 +386,7 @@ bool ExecuteAssembler(AssemblerInvocation &Opts, DiagnosticsEngine &Diags) {
    MCSection *AsmLabel = Ctx.getMachOSection(
        "__LLVM", "__asm", MachO::S_REGULAR, 4, SectionKind::getReadOnly());
    Str.get()->SwitchSection(AsmLabel);
-    Str.get()->EmitZeros(1);
+    Str.get()->emitZeros(1);
  }

  // Assembly to object compilation should leverage assembly info.
@@ -11,6 +11,7 @@
 #include <clang/FrontendTool/Utils.h>
 #include <llvm/ADT/IntrusiveRefCntPtr.h>
 #include <llvm/Option/Option.h>
+#include <llvm/Support/Host.h>

 using namespace llvm;
 using namespace clang;
@@ -25,8 +25,8 @@ 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 < 11 || minor > 14 {
-		return nil, fmt.Errorf("requires go version 1.11, 1.12, 1.13, or 1.14, got go%d.%d", major, minor)
+	if major != 1 || minor < 11 || minor > 15 {
+		return nil, fmt.Errorf("requires go version 1.11 through 1.15, got go%d.%d", major, minor)
 	}
 	clangHeaderPath := getClangHeaderPath(goenv.Get("TINYGOROOT"))
 	return &compileopts.Config{
@@ -6,6 +6,9 @@ import (
 	"os/exec"
 	"path/filepath"
 	"sort"
+	"strings"
+
+	"tinygo.org/x/go-llvm"
 )

 // getClangHeaderPath returns the path to the built-in Clang headers. It tries
@@ -26,6 +29,7 @@ func getClangHeaderPath(TINYGOROOT string) string {

 	// It looks like we are built with a system-installed LLVM. Do a last
 	// attempt: try to use Clang headers relative to the clang binary.
+	llvmMajor := strings.Split(llvm.Version, ".")[0]
 	for _, cmdName := range commands["clang"] {
 		binpath, err := exec.LookPath(cmdName)
 		if err == nil {
@@ -40,22 +44,38 @@ func getClangHeaderPath(TINYGOROOT string) string {
 			// Example executable:
 			//     /usr/lib/llvm-9/bin/clang
 			// Example include path:
-			//     /usr/lib/llvm-9/lib/clang/9.0.1/include/
+			//     /usr/lib/llvm-9/lib64/clang/9.0.1/include/
 			llvmRoot := filepath.Dir(filepath.Dir(binpath))
-			clangVersionRoot := filepath.Join(llvmRoot, "lib", "clang")
-			dirs, err := ioutil.ReadDir(clangVersionRoot)
-			if err != nil {
+			clangVersionRoot := filepath.Join(llvmRoot, "lib64", "clang")
+			dirs64, err64 := ioutil.ReadDir(clangVersionRoot)
+			// Example include path:
+			//     /usr/lib/llvm-9/lib/clang/9.0.1/include/
+			clangVersionRoot = filepath.Join(llvmRoot, "lib", "clang")
+			dirs32, err32 := ioutil.ReadDir(clangVersionRoot)
+			if err64 != nil && err32 != nil {
 				// Unexpected.
 				continue
 			}
-			dirnames := make([]string, len(dirs))
-			for i, d := range dirs {
-				dirnames[i] = d.Name()
+			dirnames := make([]string, len(dirs64)+len(dirs32))
+			dirCount := 0
+			for _, d := range dirs32 {
+				name := d.Name()
+				if name == llvmMajor || strings.HasPrefix(name, llvmMajor+".") {
+					dirnames[dirCount] = filepath.Join(llvmRoot, "lib", "clang", name)
+					dirCount++
+				}
+			}
+			for _, d := range dirs64 {
+				name := d.Name()
+				if name == llvmMajor || strings.HasPrefix(name, llvmMajor+".") {
+					dirnames[dirCount] = filepath.Join(llvmRoot, "lib64", "clang", name)
+					dirCount++
+				}
 			}
 			sort.Strings(dirnames)
 			// Check for the highest version first.
-			for i := len(dirnames) - 1; i >= 0; i-- {
-				path := filepath.Join(clangVersionRoot, dirnames[i], "include")
+			for i := dirCount - 1; i >= 0; i-- {
+				path := filepath.Join(dirnames[i], "include")
 				_, err := os.Stat(filepath.Join(path, "stdint.h"))
 				if err == nil {
 					return path
@@ -0,0 +1,153 @@
+package builder
+
+// This file implements support for writing ESP image files. These image files
+// are read by the ROM bootloader so have to be in a particular format.
+//
+// In the future, it may be necessary to implement support for other image
+// formats, such as the ESP8266 image formats (again, used by the ROM bootloader
+// to load the firmware).
+
+import (
+	"bytes"
+	"crypto/sha256"
+	"debug/elf"
+	"encoding/binary"
+	"fmt"
+	"io/ioutil"
+	"sort"
+)
+
+type espImageSegment struct {
+	addr uint32
+	data []byte
+}
+
+// makeESPFirmare converts an input ELF file to an image file for an ESP32 or
+// ESP8266 chip. This is a special purpose image format just for the ESP chip
+// family, and is parsed by the on-chip mask ROM bootloader.
+//
+// The following documentation has been used:
+// https://github.com/espressif/esptool/wiki/Firmware-Image-Format
+// https://github.com/espressif/esp-idf/blob/8fbb63c2a701c22ccf4ce249f43aded73e134a34/components/bootloader_support/include/esp_image_format.h#L58
+// https://github.com/espressif/esptool/blob/master/esptool.py
+func makeESPFirmareImage(infile, outfile, format string) error {
+	inf, err := elf.Open(infile)
+	if err != nil {
+		return err
+	}
+	defer inf.Close()
+
+	// Load all segments to be written to the image. These are actually ELF
+	// sections, not true ELF segments (similar to how esptool does it).
+	var segments []*espImageSegment
+	for _, section := range inf.Sections {
+		if section.Type != elf.SHT_PROGBITS || section.Size == 0 || section.Flags&elf.SHF_ALLOC == 0 {
+			continue
+		}
+		data, err := section.Data()
+		if err != nil {
+			return fmt.Errorf("failed to read section data: %w", err)
+		}
+		for len(data)%4 != 0 {
+			// Align segment to 4 bytes.
+			data = append(data, 0)
+		}
+		if uint64(uint32(section.Addr)) != section.Addr {
+			return fmt.Errorf("section address too big: 0x%x", section.Addr)
+		}
+		segments = append(segments, &espImageSegment{
+			addr: uint32(section.Addr),
+			data: data,
+		})
+	}
+
+	// Sort the segments by address. This is what esptool does too.
+	sort.SliceStable(segments, func(i, j int) bool { return segments[i].addr < segments[j].addr })
+
+	// Calculate checksum over the segment data. This is used in the image
+	// footer.
+	checksum := uint8(0xef)
+	for _, segment := range segments {
+		for _, b := range segment.data {
+			checksum ^= b
+		}
+	}
+
+	// Write first to an in-memory buffer, primarily so that we can easily
+	// calculate a hash over the entire image.
+	// An added benefit is that we don't need to check for errors all the time.
+	outf := &bytes.Buffer{}
+
+	// Image header.
+	switch format {
+	case "esp32":
+		// Header format:
+		// https://github.com/espressif/esp-idf/blob/8fbb63c2/components/bootloader_support/include/esp_image_format.h#L58
+		binary.Write(outf, binary.LittleEndian, struct {
+			magic          uint8
+			segment_count  uint8
+			spi_mode       uint8
+			spi_speed_size uint8
+			entry_addr     uint32
+			wp_pin         uint8
+			spi_pin_drv    [3]uint8
+			reserved       [11]uint8
+			hash_appended  bool
+		}{
+			magic:          0xE9,
+			segment_count:  byte(len(segments)),
+			spi_mode:       0, // irrelevant, replaced by esptool when flashing
+			spi_speed_size: 0, // spi_speed, spi_size: replaced by esptool when flashing
+			entry_addr:     uint32(inf.Entry),
+			wp_pin:         0xEE, // disable WP pin
+			hash_appended:  true, // add a SHA256 hash
+		})
+	case "esp8266":
+		// Header format:
+		// https://github.com/espressif/esptool/wiki/Firmware-Image-Format
+		// Basically a truncated version of the ESP32 header.
+		binary.Write(outf, binary.LittleEndian, struct {
+			magic          uint8
+			segment_count  uint8
+			spi_mode       uint8
+			spi_speed_size uint8
+			entry_addr     uint32
+		}{
+			magic:          0xE9,
+			segment_count:  byte(len(segments)),
+			spi_mode:       0,    // irrelevant, replaced by esptool when flashing
+			spi_speed_size: 0x20, // spi_speed, spi_size: replaced by esptool when flashing
+			entry_addr:     uint32(inf.Entry),
+		})
+	default:
+		return fmt.Errorf("builder: unknown binary format %#v, expected esp32 or esp8266", format)
+	}
+
+	// Write all segments to the image.
+	// https://github.com/espressif/esptool/wiki/Firmware-Image-Format#segment
+	for _, segment := range segments {
+		binary.Write(outf, binary.LittleEndian, struct {
+			addr   uint32
+			length uint32
+		}{
+			addr:   segment.addr,
+			length: uint32(len(segment.data)),
+		})
+		outf.Write(segment.data)
+	}
+
+	// Footer, including checksum.
+	// The entire image size must be a multiple of 16, so pad the image to one
+	// byte less than that before writing the checksum.
+	outf.Write(make([]byte, 15-outf.Len()%16))
+	outf.WriteByte(checksum)
+
+	if format == "esp32" {
+		// SHA256 hash (to protect against image corruption, not for security).
+		hash := sha256.Sum256(outf.Bytes())
+		outf.Write(hash[:])
+	}
+
+	// Write the image to the output file.
+	return ioutil.WriteFile(outfile, outf.Bytes(), 0666)
+}
@@ -4,12 +4,15 @@ import (
 	"debug/elf"
 	"io/ioutil"
 	"os"
-	"path/filepath"
 	"sort"

 	"github.com/marcinbor85/gohex"
 )

+// maxPadBytes is the maximum allowed bytes to be padded in a rom extraction
+// this value is currently defined by Nintendo Switch Page Alignment (4096 bytes)
+const maxPadBytes = 4095
+
 // objcopyError is an error returned by functions that act like objcopy.
 type objcopyError struct {
 	Op  string
@@ -58,7 +61,7 @@ func extractROM(path string) (uint64, []byte, error) {

 	progs := make(progSlice, 0, 2)
 	for _, prog := range f.Progs {
-		if prog.Type != elf.PT_LOAD || prog.Filesz == 0 {
+		if prog.Type != elf.PT_LOAD || prog.Filesz == 0 || prog.Off == 0 {
 			continue
 		}
 		progs = append(progs, prog)
@@ -70,8 +73,19 @@ func extractROM(path string) (uint64, []byte, error) {

 	var rom []byte
 	for _, prog := range progs {
+		romEnd := progs[0].Paddr + uint64(len(rom))
+		if prog.Paddr > romEnd && prog.Paddr < romEnd+16 {
+			// Sometimes, the linker seems to insert a bit of padding between
+			// segments. Simply zero-fill these parts.
+			rom = append(rom, make([]byte, prog.Paddr-romEnd)...)
+		}
 		if prog.Paddr != progs[0].Paddr+uint64(len(rom)) {
-			return 0, nil, objcopyError{"ROM segments are non-contiguous: " + path, nil}
+			diff := prog.Paddr - (progs[0].Paddr + uint64(len(rom)))
+			if diff > maxPadBytes {
+				return 0, nil, objcopyError{"ROM segments are non-contiguous: " + path, nil}
+			}
+			// Pad the difference
+			rom = append(rom, make([]byte, diff)...)
 		}
 		data, err := ioutil.ReadAll(prog.Open())
 		if err != nil {
@@ -93,7 +107,7 @@ func extractROM(path string) (uint64, []byte, error) {

 // objcopy converts an ELF file to a different (simpler) output file format:
 // .bin or .hex. It extracts only the .text section.
-func objcopy(infile, outfile string) error {
+func objcopy(infile, outfile, binaryFormat string) error {
 	f, err := os.OpenFile(outfile, os.O_RDWR|os.O_CREATE|os.O_TRUNC, 0666)
 	if err != nil {
 		return err
@@ -107,23 +121,20 @@ func objcopy(infile, outfile string) error {
 	}

 	// Write to the file, in the correct format.
-	switch filepath.Ext(outfile) {
-	case ".gba":
-		// The address is not stored in a .gba file.
-		_, err := f.Write(data)
-		return err
-	case ".bin":
-		// The address is not stored in a .bin file (therefore you
-		// should use .hex files in most cases).
-		_, err := f.Write(data)
-		return err
-	case ".hex":
+	switch binaryFormat {
+	case "hex":
+		// Intel hex file, includes the firmware start address.
 		mem := gohex.NewMemory()
 		err := mem.AddBinary(uint32(addr), data)
 		if err != nil {
 			return objcopyError{"failed to create .hex file", err}
 		}
 		return mem.DumpIntelHex(f, 16)
+	case "bin":
+		// The start address is not stored in raw firmware files (therefore you
+		// should use .hex files in most cases).
+		_, err := f.Write(data)
+		return err
 	default:
 		panic("unreachable")
 	}
@@ -1,5 +1,5 @@
 // +build !byollvm
-// +build !llvm9
+// +build !llvm9,!llvm11

 package cgo

@@ -0,0 +1,14 @@
+// +build !byollvm
+// +build llvm11
+
+package cgo
+
+/*
+#cgo linux   CFLAGS:  -I/usr/lib/llvm-11/include
+#cgo darwin  CFLAGS:  -I/usr/local/opt/llvm@11/include
+#cgo freebsd CFLAGS:  -I/usr/local/llvm11/include
+#cgo linux   LDFLAGS: -L/usr/lib/llvm-11/lib -lclang
+#cgo darwin  LDFLAGS: -L/usr/local/opt/llvm@11/lib -lclang -lffi
+#cgo freebsd LDFLAGS: -L/usr/local/llvm11/lib -lclang
+*/
+import "C"
@@ -118,12 +118,12 @@ func (c *Config) NeedsStackObjects() bool {
 	switch c.GC() {
 	case "conservative", "extalloc":
 		for _, tag := range c.BuildTags() {
-			if tag == "baremetal" {
-				return false
+			if tag == "wasm" {
+				return true
 			}
 		}

-		return true
+		return false
 	default:
 		return false
 	}
@@ -164,6 +164,16 @@ func (c *Config) PanicStrategy() string {
 	return c.Options.PanicStrategy
 }

+// AutomaticStackSize returns whether goroutine stack sizes should be determined
+// automatically at compile time, if possible. If it is false, no attempt is
+// made.
+func (c *Config) AutomaticStackSize() bool {
+	if c.Target.AutoStackSize != nil && c.Scheduler() == "tasks" {
+		return *c.Target.AutoStackSize
+	}
+	return false
+}
+
 // CFlags returns the flags to pass to the C compiler. This is necessary for CGo
 // preprocessing.
 func (c *Config) CFlags() []string {
@@ -229,6 +239,31 @@ func (c *Config) Debug() bool {
 	return c.Options.Debug
 }

+// BinaryFormat returns an appropriate binary format, based on the file
+// extension and the configured binary format in the target JSON file.
+func (c *Config) BinaryFormat(ext string) string {
+	switch ext {
+	case ".bin", ".gba", ".nro":
+		// The simplest format possible: dump everything in a raw binary file.
+		if c.Target.BinaryFormat != "" {
+			return c.Target.BinaryFormat
+		}
+		return "bin"
+	case ".hex":
+		// Similar to bin, but includes the start address and is thus usually a
+		// better format.
+		return "hex"
+	case ".uf2":
+		// Special purpose firmware format, mainly used on Adafruit boards.
+		// More information:
+		// https://github.com/Microsoft/uf2
+		return "uf2"
+	default:
+		// Use the ELF format for unrecognized file formats.
+		return "elf"
+	}
+}
+
 // Programmer returns the flash method and OpenOCD interface name given a
 // particular configuration. It may either be all configured in the target JSON
 // file or be modified using the -programmmer command-line option.
@@ -294,6 +329,15 @@ func (c *Config) RelocationModel() string {
 	return "static"
 }

+// WasmAbi returns the WASM ABI which is specified in the target JSON file, and
+// the value is overridden by `-wasm-abi` flag if it is provided
+func (c *Config) WasmAbi() string {
+	if c.Options.WasmAbi != "" {
+		return c.Options.WasmAbi
+	}
+	return c.Target.WasmAbi
+}
+
 type TestConfig struct {
 	CompileTestBinary bool
 	// TODO: Filter the test functions to run, include verbose flag, etc
@@ -8,6 +8,7 @@ import (
 	"io"
 	"os"
 	"path/filepath"
+	"reflect"
 	"runtime"
 	"strings"

@@ -33,11 +34,13 @@ type TargetSpec struct {
 	Linker           string   `json:"linker"`
 	RTLib            string   `json:"rtlib"` // compiler runtime library (libgcc, compiler-rt)
 	Libc             string   `json:"libc"`
+	AutoStackSize    *bool    `json:"automatic-stack-size"` // Determine stack size automatically at compile time.
+	DefaultStackSize uint64   `json:"default-stack-size"`   // Default stack size if the size couldn't be determined at compile time.
 	CFlags           []string `json:"cflags"`
 	LDFlags          []string `json:"ldflags"`
 	LinkerScript     string   `json:"linkerscript"`
 	ExtraFiles       []string `json:"extra-files"`
-	Emulator         []string `json:"emulator"`
+	Emulator         []string `json:"emulator" override:"copy"` // inherited Emulator must not be append
 	FlashCommand     string   `json:"flash-command"`
 	GDB              string   `json:"gdb"`
 	PortReset        string   `json:"flash-1200-bps-reset"`
@@ -45,99 +48,56 @@ type TargetSpec struct {
 	FlashVolume      string   `json:"msd-volume-name"`
 	FlashFilename    string   `json:"msd-firmware-name"`
 	UF2FamilyID      string   `json:"uf2-family-id"`
+	BinaryFormat     string   `json:"binary-format"`
 	OpenOCDInterface string   `json:"openocd-interface"`
 	OpenOCDTarget    string   `json:"openocd-target"`
 	OpenOCDTransport string   `json:"openocd-transport"`
 	JLinkDevice      string   `json:"jlink-device"`
 	CodeModel        string   `json:"code-model"`
 	RelocationModel  string   `json:"relocation-model"`
+	WasmAbi          string   `json:"wasm-abi"`
 }

-// copyProperties copies all properties that are set in spec2 into itself.
-func (spec *TargetSpec) copyProperties(spec2 *TargetSpec) {
-	// TODO: simplify this using reflection? Inherits and BuildTags are special
-	// cases, but the rest can simply be copied if set.
-	spec.Inherits = append(spec.Inherits, spec2.Inherits...)
-	if spec2.Triple != "" {
-		spec.Triple = spec2.Triple
-	}
-	if spec2.CPU != "" {
-		spec.CPU = spec2.CPU
-	}
-	spec.Features = append(spec.Features, spec2.Features...)
-	if spec2.GOOS != "" {
-		spec.GOOS = spec2.GOOS
-	}
-	if spec2.GOARCH != "" {
-		spec.GOARCH = spec2.GOARCH
-	}
-	spec.BuildTags = append(spec.BuildTags, spec2.BuildTags...)
-	if spec2.GC != "" {
-		spec.GC = spec2.GC
-	}
-	if spec2.Scheduler != "" {
-		spec.Scheduler = spec2.Scheduler
-	}
-	if spec2.Compiler != "" {
-		spec.Compiler = spec2.Compiler
-	}
-	if spec2.Linker != "" {
-		spec.Linker = spec2.Linker
-	}
-	if spec2.RTLib != "" {
-		spec.RTLib = spec2.RTLib
-	}
-	if spec2.Libc != "" {
-		spec.Libc = spec2.Libc
-	}
-	spec.CFlags = append(spec.CFlags, spec2.CFlags...)
-	spec.LDFlags = append(spec.LDFlags, spec2.LDFlags...)
-	if spec2.LinkerScript != "" {
-		spec.LinkerScript = spec2.LinkerScript
-	}
-	spec.ExtraFiles = append(spec.ExtraFiles, spec2.ExtraFiles...)
-	if len(spec2.Emulator) != 0 {
-		spec.Emulator = spec2.Emulator
-	}
-	if spec2.FlashCommand != "" {
-		spec.FlashCommand = spec2.FlashCommand
-	}
-	if spec2.GDB != "" {
-		spec.GDB = spec2.GDB
-	}
-	if spec2.PortReset != "" {
-		spec.PortReset = spec2.PortReset
-	}
-	if spec2.FlashMethod != "" {
-		spec.FlashMethod = spec2.FlashMethod
-	}
-	if spec2.FlashVolume != "" {
-		spec.FlashVolume = spec2.FlashVolume
-	}
-	if spec2.FlashFilename != "" {
-		spec.FlashFilename = spec2.FlashFilename
-	}
-	if spec2.UF2FamilyID != "" {
-		spec.UF2FamilyID = spec2.UF2FamilyID
-	}
-	if spec2.OpenOCDInterface != "" {
-		spec.OpenOCDInterface = spec2.OpenOCDInterface
-	}
-	if spec2.OpenOCDTarget != "" {
-		spec.OpenOCDTarget = spec2.OpenOCDTarget
-	}
-	if spec2.OpenOCDTransport != "" {
-		spec.OpenOCDTransport = spec2.OpenOCDTransport
-	}
-	if spec2.JLinkDevice != "" {
-		spec.JLinkDevice = spec2.JLinkDevice
-	}
-	if spec2.CodeModel != "" {
-		spec.CodeModel = spec2.CodeModel
-	}
+// overrideProperties overrides all properties that are set in child into itself using reflection.
+func (spec *TargetSpec) overrideProperties(child *TargetSpec) {
+	specType := reflect.TypeOf(spec).Elem()
+	specValue := reflect.ValueOf(spec).Elem()
+	childValue := reflect.ValueOf(child).Elem()

-	if spec2.RelocationModel != "" {
-		spec.RelocationModel = spec2.RelocationModel
+	for i := 0; i < specType.NumField(); i++ {
+		field := specType.Field(i)
+		src := childValue.Field(i)
+		dst := specValue.Field(i)
+
+		switch kind := field.Type.Kind(); kind {
+		case reflect.String: // for strings, just copy the field of child to spec if not empty
+			if src.Len() > 0 {
+				dst.Set(src)
+			}
+		case reflect.Uint, reflect.Uint32, reflect.Uint64: // for Uint, copy if not zero
+			if src.Uint() != 0 {
+				dst.Set(src)
+			}
+		case reflect.Ptr: // for pointers, copy if not nil
+			if !src.IsNil() {
+				dst.Set(src)
+			}
+		case reflect.Slice: // for slices...
+			if src.Len() > 0 { // ... if not empty ...
+				switch tag := field.Tag.Get("override"); tag {
+				case "copy":
+					// copy the field of child to spec
+					dst.Set(src)
+				case "append", "":
+					// or append the field of child to spec
+					dst.Set(reflect.AppendSlice(src, dst))
+				default:
+					panic("override mode must be 'copy' or 'append' (default). I don't know how to '" + tag + "'.")
+				}
+			}
+		default:
+			panic("unknown field type : " + kind.String())
+		}
 	}
 }

@@ -186,11 +146,11 @@ func (spec *TargetSpec) resolveInherits() error {
 		if err != nil {
 			return err
 		}
-		newSpec.copyProperties(subtarget)
+		newSpec.overrideProperties(subtarget)
 	}

 	// When all properties are loaded, make sure they are properly inherited.
-	newSpec.copyProperties(spec)
+	newSpec.overrideProperties(spec)
 	*spec = *newSpec

 	return nil
@@ -256,6 +216,7 @@ func LoadTarget(target string) (*TargetSpec, error) {
 		}
 		goarch := map[string]string{ // map from LLVM arch to Go arch
 			"i386":    "386",
+			"i686":    "386",
 			"x86_64":  "amd64",
 			"aarch64": "arm64",
 			"armv7":   "arm",
@@ -271,39 +232,40 @@ func defaultTarget(goos, goarch, triple string) (*TargetSpec, error) {
 	// No target spec available. Use the default one, useful on most systems
 	// with a regular OS.
 	spec := TargetSpec{
-		Triple:      triple,
-		GOOS:        goos,
-		GOARCH:      goarch,
-		BuildTags:   []string{goos, goarch},
-		Compiler:    "clang",
-		Linker:      "cc",
-		CFlags:      []string{"--target=" + triple},
-		GDB:         "gdb",
-		PortReset:   "false",
-		FlashMethod: "native",
+		Triple:    triple,
+		GOOS:      goos,
+		GOARCH:    goarch,
+		BuildTags: []string{goos, goarch},
+		Compiler:  "clang",
+		Linker:    "cc",
+		CFlags:    []string{"--target=" + triple},
+		GDB:       "gdb",
+		PortReset: "false",
 	}
 	if goos == "darwin" {
 		spec.LDFlags = append(spec.LDFlags, "-Wl,-dead_strip")
 	} else {
 		spec.LDFlags = append(spec.LDFlags, "-no-pie", "-Wl,--gc-sections") // WARNING: clang < 5.0 requires -nopie
 	}
+	if goarch != "wasm" {
+		spec.ExtraFiles = append(spec.ExtraFiles, "src/runtime/gc_"+goarch+".S")
+	}
 	if goarch != runtime.GOARCH {
 		// Some educated guesses as to how to invoke helper programs.
+		spec.GDB = "gdb-multiarch"
 		if goarch == "arm" && goos == "linux" {
 			spec.CFlags = append(spec.CFlags, "--sysroot=/usr/arm-linux-gnueabihf")
 			spec.Linker = "arm-linux-gnueabihf-gcc"
-			spec.GDB = "arm-linux-gnueabihf-gdb"
 			spec.Emulator = []string{"qemu-arm", "-L", "/usr/arm-linux-gnueabihf"}
 		}
 		if goarch == "arm64" && goos == "linux" {
 			spec.CFlags = append(spec.CFlags, "--sysroot=/usr/aarch64-linux-gnu")
 			spec.Linker = "aarch64-linux-gnu-gcc"
-			spec.GDB = "aarch64-linux-gnu-gdb"
 			spec.Emulator = []string{"qemu-aarch64", "-L", "/usr/aarch64-linux-gnu"}
 		}
-		if goarch == "386" {
-			spec.CFlags = []string{"-m32"}
-			spec.LDFlags = []string{"-m32"}
+		if goarch == "386" && runtime.GOARCH == "amd64" {
+			spec.CFlags = append(spec.CFlags, "-m32")
+			spec.LDFlags = append(spec.LDFlags, "-m32")
 		}
 	}
 	return &spec, nil
@@ -1,6 +1,9 @@
 package compileopts

-import "testing"
+import (
+	"reflect"
+	"testing"
+)

 func TestLoadTarget(t *testing.T) {
 	_, err := LoadTarget("arduino")
@@ -17,3 +20,67 @@ func TestLoadTarget(t *testing.T) {
 		t.Error("LoadTarget failed for wrong reason:", err)
 	}
 }
+
+func TestOverrideProperties(t *testing.T) {
+	baseAutoStackSize := true
+	base := &TargetSpec{
+		GOOS:             "baseGoos",
+		CPU:              "baseCpu",
+		Features:         []string{"bf1", "bf2"},
+		BuildTags:        []string{"bt1", "bt2"},
+		Emulator:         []string{"be1", "be2"},
+		DefaultStackSize: 42,
+		AutoStackSize:    &baseAutoStackSize,
+	}
+	childAutoStackSize := false
+	child := &TargetSpec{
+		GOOS:             "",
+		CPU:              "chlidCpu",
+		Features:         []string{"cf1", "cf2"},
+		Emulator:         []string{"ce1", "ce2"},
+		AutoStackSize:    &childAutoStackSize,
+		DefaultStackSize: 64,
+	}
+
+	base.overrideProperties(child)
+
+	if base.GOOS != "baseGoos" {
+		t.Errorf("Overriding failed : got %v", base.GOOS)
+	}
+	if base.CPU != "chlidCpu" {
+		t.Errorf("Overriding failed : got %v", base.CPU)
+	}
+	if !reflect.DeepEqual(base.Features, []string{"cf1", "cf2", "bf1", "bf2"}) {
+		t.Errorf("Overriding failed : got %v", base.Features)
+	}
+	if !reflect.DeepEqual(base.BuildTags, []string{"bt1", "bt2"}) {
+		t.Errorf("Overriding failed : got %v", base.BuildTags)
+	}
+	if !reflect.DeepEqual(base.Emulator, []string{"ce1", "ce2"}) {
+		t.Errorf("Overriding failed : got %v", base.Emulator)
+	}
+	if *base.AutoStackSize != false {
+		t.Errorf("Overriding failed : got %v", base.AutoStackSize)
+	}
+	if base.DefaultStackSize != 64 {
+		t.Errorf("Overriding failed : got %v", base.DefaultStackSize)
+	}
+
+	baseAutoStackSize = true
+	base = &TargetSpec{
+		AutoStackSize:    &baseAutoStackSize,
+		DefaultStackSize: 42,
+	}
+	child = &TargetSpec{
+		AutoStackSize:    nil,
+		DefaultStackSize: 0,
+	}
+	base.overrideProperties(child)
+	if *base.AutoStackSize != true {
+		t.Errorf("Overriding failed : got %v", base.AutoStackSize)
+	}
+	if base.DefaultStackSize != 42 {
+		t.Errorf("Overriding failed : got %v", base.DefaultStackSize)
+	}
+
+}
@@ -5,18 +5,15 @@ import (
 	"errors"
 	"fmt"
 	"go/ast"
-	"go/build"
 	"go/constant"
 	"go/token"
 	"go/types"
-	"os"
 	"path/filepath"
 	"strconv"
 	"strings"

 	"github.com/tinygo-org/tinygo/compileopts"
 	"github.com/tinygo-org/tinygo/compiler/llvmutil"
-	"github.com/tinygo-org/tinygo/goenv"
 	"github.com/tinygo-org/tinygo/ir"
 	"github.com/tinygo-org/tinygo/loader"
 	"golang.org/x/tools/go/ssa"
@@ -130,6 +127,28 @@ func NewTargetMachine(config *compileopts.Config) (llvm.TargetMachine, error) {
 	return machine, nil
 }

+// CompilerOutput is returned from the Compile() call. It contains the compile
+// output and information necessary to continue to compile and link the program.
+type CompilerOutput struct {
+	// The LLVM module that contains the compiled but not optimized LLVM module
+	// for all the Go code in the program.
+	Mod llvm.Module
+
+	// ExtraFiles is a list of C source files included in packages that should
+	// be built and linked together with the main executable to form one
+	// program. They can be used from CGo, for example.
+	ExtraFiles []string
+
+	// ExtraLDFlags are linker flags obtained during CGo processing. These flags
+	// must be passed to the linker which links the entire executable.
+	ExtraLDFlags []string
+
+	// MainDir is the absolute directory path to the directory of the main
+	// package. This is useful for testing: tests must be run in the package
+	// directory that is being tested.
+	MainDir string
+}
+
 // Compile the given package path or .go file path. Return an error when this
 // fails (in any stage). If successful it returns the LLVM module and a list of
 // extra C files to be compiled. If not, one or more errors will be returned.
@@ -138,7 +157,7 @@ func NewTargetMachine(config *compileopts.Config) (llvm.TargetMachine, error) {
 // violation. Eventually, this Compile function should only compile a single
 // package and not the whole program, and loading of the program (including CGo
 // processing) should be moved outside the compiler package.
-func Compile(pkgName string, machine llvm.TargetMachine, config *compileopts.Config) (mod llvm.Module, extrafiles []string, extraldflags []string, errors []error) {
+func Compile(pkgName string, machine llvm.TargetMachine, config *compileopts.Config) (output CompilerOutput, errors []error) {
 	c := &compilerContext{
 		Config:     config,
 		difiles:    make(map[string]llvm.Metadata),
@@ -154,6 +173,7 @@ func Compile(pkgName string, machine llvm.TargetMachine, config *compileopts.Con
 	if c.Debug() {
 		c.dibuilder = llvm.NewDIBuilder(c.mod)
 	}
+	output.Mod = c.mod

 	c.uintptrType = c.ctx.IntType(c.targetData.PointerSize() * 8)
 	if c.targetData.PointerSize() <= 4 {
@@ -172,55 +192,29 @@ func Compile(pkgName string, machine llvm.TargetMachine, config *compileopts.Con
 	c.funcPtrAddrSpace = dummyFunc.Type().PointerAddressSpace()
 	dummyFunc.EraseFromParentAsFunction()

-	wd, err := os.Getwd()
+	lprogram, err := loader.Load(c.Config, []string{pkgName}, c.ClangHeaders, types.Config{
+		Sizes: &stdSizes{
+			IntSize:  int64(c.targetData.TypeAllocSize(c.intType)),
+			PtrSize:  int64(c.targetData.PointerSize()),
+			MaxAlign: int64(c.targetData.PrefTypeAlignment(c.i8ptrType)),
+		}})
 	if err != nil {
-		return c.mod, nil, nil, []error{err}
-	}
-	goroot, err := loader.GetCachedGoroot(c.Config)
-	if err != nil {
-		return c.mod, nil, nil, []error{err}
-	}
-	lprogram := &loader.Program{
-		Build: &build.Context{
-			GOARCH:      c.GOARCH(),
-			GOOS:        c.GOOS(),
-			GOROOT:      goroot,
-			GOPATH:      goenv.Get("GOPATH"),
-			CgoEnabled:  c.CgoEnabled(),
-			UseAllFiles: false,
-			Compiler:    "gc", // must be one of the recognized compilers
-			BuildTags:   c.BuildTags(),
-		},
-		Tests: c.TestConfig.CompileTestBinary,
-		TypeChecker: types.Config{
-			Sizes: &stdSizes{
-				IntSize:  int64(c.targetData.TypeAllocSize(c.intType)),
-				PtrSize:  int64(c.targetData.PointerSize()),
-				MaxAlign: int64(c.targetData.PrefTypeAlignment(c.i8ptrType)),
-			},
-		},
-		Dir:          wd,
-		TINYGOROOT:   goenv.Get("TINYGOROOT"),
-		CFlags:       c.CFlags(),
-		ClangHeaders: c.ClangHeaders,
-	}
-
-	err = lprogram.Load(pkgName)
-	if err != nil {
-		return c.mod, nil, nil, []error{err}
+		return output, []error{err}
 	}

 	err = lprogram.Parse()
 	if err != nil {
-		return c.mod, nil, nil, []error{err}
+		return output, []error{err}
 	}
+	output.ExtraLDFlags = lprogram.LDFlags
+	output.MainDir = lprogram.MainPkg().Dir

 	c.ir = ir.NewProgram(lprogram)

 	// Run a simple dead code elimination pass.
 	err = c.ir.SimpleDCE()
 	if err != nil {
-		return c.mod, nil, nil, []error{err}
+		return output, []error{err}
 	}

 	// Initialize debug information.
@@ -359,17 +353,13 @@ func Compile(pkgName string, machine llvm.TargetMachine, config *compileopts.Con
 	}

 	// Gather the list of (C) file paths that should be included in the build.
-	var extraFiles []string
 	for _, pkg := range c.ir.LoaderProgram.Sorted() {
-		for _, file := range pkg.OtherFiles {
-			switch strings.ToLower(filepath.Ext(file)) {
-			case ".c":
-				extraFiles = append(extraFiles, file)
-			}
+		for _, filename := range pkg.CFiles {
+			output.ExtraFiles = append(output.ExtraFiles, filepath.Join(pkg.Dir, filename))
 		}
 	}

-	return c.mod, extraFiles, lprogram.LDFlags, c.diagnostics
+	return output, c.diagnostics
 }

 // getLLVMRuntimeType obtains a named type from the runtime package and returns
@@ -1346,12 +1336,14 @@ 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 == "device.Asm" || name == "device/arm.Asm" || name == "device/avr.Asm" || name == "device/riscv.Asm":
+		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/avr.AsmFull" || name == "device/riscv.AsmFull":
+		case name == "device.AsmFull" || name == "device/arm.AsmFull" || name == "device/arm64.AsmFull" || name == "device/avr.AsmFull" || name == "device/riscv.AsmFull":
 			return b.createInlineAsmFull(instr)
 		case strings.HasPrefix(name, "device/arm.SVCall"):
 			return b.emitSVCall(instr.Args)
+		case strings.HasPrefix(name, "device/arm64.SVCall"):
+			return b.emitSV64Call(instr.Args)
 		case strings.HasPrefix(name, "(device/riscv.CSR)."):
 			return b.emitCSROperation(instr)
 		case strings.HasPrefix(name, "syscall.Syscall"):
@@ -2054,17 +2046,17 @@ func (b *builder) createBinOp(op token.Token, typ, ytyp types.Type, x, y llvm.Va
 			case token.QUO: // /
 				return b.CreateFDiv(x, y, ""), nil
 			case token.EQL: // ==
-				return b.CreateFCmp(llvm.FloatUEQ, x, y, ""), nil
+				return b.CreateFCmp(llvm.FloatOEQ, x, y, ""), nil
 			case token.NEQ: // !=
 				return b.CreateFCmp(llvm.FloatUNE, x, y, ""), nil
 			case token.LSS: // <
-				return b.CreateFCmp(llvm.FloatULT, x, y, ""), nil
+				return b.CreateFCmp(llvm.FloatOLT, x, y, ""), nil
 			case token.LEQ: // <=
-				return b.CreateFCmp(llvm.FloatULE, x, y, ""), nil
+				return b.CreateFCmp(llvm.FloatOLE, x, y, ""), nil
 			case token.GTR: // >
-				return b.CreateFCmp(llvm.FloatUGT, x, y, ""), nil
+				return b.CreateFCmp(llvm.FloatOGT, x, y, ""), nil
 			case token.GEQ: // >=
-				return b.CreateFCmp(llvm.FloatUGE, x, y, ""), nil
+				return b.CreateFCmp(llvm.FloatOGE, x, y, ""), nil
 			default:
 				panic("binop on float: " + op.String())
 			}
@@ -2588,7 +2580,17 @@ func (b *builder) createUnOp(unop *ssa.UnOp) (llvm.Value, error) {
 			if typ.Info()&types.IsInteger != 0 {
 				return b.CreateSub(llvm.ConstInt(x.Type(), 0, false), x, ""), nil
 			} else if typ.Info()&types.IsFloat != 0 {
-				return b.CreateFSub(llvm.ConstFloat(x.Type(), 0.0), x, ""), nil
+				return b.CreateFNeg(x, ""), nil
+			} else if typ.Info()&types.IsComplex != 0 {
+				// Negate both components of the complex number.
+				r := b.CreateExtractValue(x, 0, "r")
+				i := b.CreateExtractValue(x, 1, "i")
+				r = b.CreateFNeg(r, "")
+				i = b.CreateFNeg(i, "")
+				cplx := llvm.Undef(x.Type())
+				cplx = b.CreateInsertValue(cplx, r, 0, "")
+				cplx = b.CreateInsertValue(cplx, i, 1, "")
+				return cplx, nil
 			} else {
 				return llvm.Value{}, b.makeError(unop.Pos(), "todo: unknown basic type for negate: "+typ.String())
 			}
@@ -19,16 +19,30 @@ import (
 // Because a go statement doesn't return anything, return undef.
 func (b *builder) createGoInstruction(funcPtr llvm.Value, params []llvm.Value, prefix string, pos token.Pos) llvm.Value {
 	paramBundle := b.emitPointerPack(params)
-	var callee llvm.Value
+	var callee, stackSize llvm.Value
 	switch b.Scheduler() {
 	case "none", "tasks":
 		callee = b.createGoroutineStartWrapper(funcPtr, prefix, pos)
+		if b.AutomaticStackSize() {
+			// The stack size is not known until after linking. Call a dummy
+			// function that will be replaced with a load from a special ELF
+			// section that contains the stack size (and is modified after
+			// linking).
+			stackSize = b.createCall(b.mod.NamedFunction("internal/task.getGoroutineStackSize"), []llvm.Value{callee, llvm.Undef(b.i8ptrType), llvm.Undef(b.i8ptrType)}, "stacksize")
+		} else {
+			// The stack size is fixed at compile time. By emitting it here as a
+			// constant, it can be optimized.
+			stackSize = llvm.ConstInt(b.uintptrType, b.Target.DefaultStackSize, false)
+		}
 	case "coroutines":
 		callee = b.CreatePtrToInt(funcPtr, b.uintptrType, "")
+		// There is no goroutine stack size: coroutines are used instead of
+		// stacks.
+		stackSize = llvm.Undef(b.uintptrType)
 	default:
 		panic("unreachable")
 	}
-	b.createCall(b.mod.NamedFunction("internal/task.start"), []llvm.Value{callee, paramBundle, llvm.Undef(b.i8ptrType), llvm.ConstPointerNull(b.i8ptrType)}, "")
+	b.createCall(b.mod.NamedFunction("internal/task.start"), []llvm.Value{callee, paramBundle, stackSize, llvm.Undef(b.i8ptrType), llvm.ConstPointerNull(b.i8ptrType)}, "")
 	return llvm.Undef(funcPtr.Type().ElementType().ReturnType())
 }

@@ -67,8 +81,9 @@ 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)
-		wrapper.SetLinkage(llvm.PrivateLinkage)
+		wrapper.SetLinkage(llvm.InternalLinkage)
 		wrapper.SetUnnamedAddr(true)
+		wrapper.AddAttributeAtIndex(-1, c.ctx.CreateStringAttribute("tinygo-gowrapper", name))
 		entry := c.ctx.AddBasicBlock(wrapper, "entry")
 		builder.SetInsertPointAtEnd(entry)

@@ -125,6 +140,7 @@ func (c *compilerContext) createGoroutineStartWrapper(fn llvm.Value, prefix stri
 		wrapper = llvm.AddFunction(c.mod, prefix+".gowrapper", wrapperType)
 		wrapper.SetLinkage(llvm.InternalLinkage)
 		wrapper.SetUnnamedAddr(true)
+		wrapper.AddAttributeAtIndex(-1, c.ctx.CreateStringAttribute("tinygo-gowrapper", ""))
 		entry := c.ctx.AddBasicBlock(wrapper, "entry")
 		builder.SetInsertPointAtEnd(entry)

@@ -163,6 +163,44 @@ func (b *builder) emitSVCall(args []ssa.Value) (llvm.Value, error) {
 	return b.CreateCall(target, llvmArgs, ""), nil
 }

+// This is a compiler builtin which emits an inline SVCall instruction. It can
+// be one of:
+//
+//     func SVCall0(num uintptr) uintptr
+//     func SVCall1(num uintptr, a1 interface{}) uintptr
+//     func SVCall2(num uintptr, a1, a2 interface{}) uintptr
+//     func SVCall3(num uintptr, a1, a2, a3 interface{}) uintptr
+//     func SVCall4(num uintptr, a1, a2, a3, a4 interface{}) uintptr
+//
+// The num parameter must be a constant. All other parameters may be any scalar
+// value supported by LLVM inline assembly.
+// Same as emitSVCall but for AArch64
+func (b *builder) emitSV64Call(args []ssa.Value) (llvm.Value, error) {
+	num, _ := constant.Uint64Val(args[0].(*ssa.Const).Value)
+	llvmArgs := []llvm.Value{}
+	argTypes := []llvm.Type{}
+	asm := "svc #" + strconv.FormatUint(num, 10)
+	constraints := "={x0}"
+	for i, arg := range args[1:] {
+		arg = arg.(*ssa.MakeInterface).X
+		if i == 0 {
+			constraints += ",0"
+		} else {
+			constraints += ",{x" + strconv.Itoa(i) + "}"
+		}
+		llvmValue := b.getValue(arg)
+		llvmArgs = append(llvmArgs, llvmValue)
+		argTypes = append(argTypes, llvmValue.Type())
+	}
+	// Implement the ARM64 calling convention by marking x1-x7 as
+	// clobbered. x0 is used as an output register so doesn't have to be
+	// marked as clobbered.
+	constraints += ",~{x1},~{x2},~{x3},~{x4},~{x5},~{x6},~{x7}"
+	fnType := llvm.FunctionType(b.uintptrType, argTypes, false)
+	target := llvm.InlineAsm(fnType, asm, constraints, true, false, 0)
+	return b.CreateCall(target, llvmArgs, ""), nil
+}
+
 // This is a compiler builtin which emits CSR instructions. It can be one of:
 //
 //     func (csr CSR) Get() uintptr
@@ -465,9 +465,7 @@ func (c *compilerContext) getInterfaceInvokeWrapper(f *ir.Function) llvm.Value {
 	paramTypes := append([]llvm.Type{c.i8ptrType}, fnType.ParamTypes()[len(expandedReceiverType):]...)
 	wrapFnType := llvm.FunctionType(fnType.ReturnType(), paramTypes, false)
 	wrapper = llvm.AddFunction(c.mod, wrapperName, wrapFnType)
-	if f.LLVMFn.LastParam().Name() == "parentHandle" {
-		wrapper.LastParam().SetName("parentHandle")
-	}
+	wrapper.LastParam().SetName("parentHandle")

 	wrapper.SetLinkage(llvm.InternalLinkage)
 	wrapper.SetUnnamedAddr(true)
@@ -26,7 +26,6 @@ func EmitPointerPack(builder llvm.Builder, mod llvm.Module, config *compileopts.
 	packedType := ctx.StructType(valueTypes, false)

 	// Allocate memory for the packed data.
-	var packedAlloc, packedHeapAlloc llvm.Value
 	size := targetData.TypeAllocSize(packedType)
 	if size == 0 {
 		return llvm.ConstPointerNull(i8ptrType)
@@ -39,9 +38,39 @@ func EmitPointerPack(builder llvm.Builder, mod llvm.Module, config *compileopts.
 			// Try to keep this cast in SSA form.
 			return builder.CreateIntToPtr(values[0], i8ptrType, "pack.int")
 		}
+
 		// Because packedType is a struct and we have to cast it to a *i8, store
-		// it in an alloca first for bitcasting (store+bitcast+load).
-		packedAlloc, _, _ = CreateTemporaryAlloca(builder, mod, packedType, "")
+		// it in a *i8 alloca first and load the *i8 value from there. This is
+		// effectively a bitcast.
+		packedAlloc, _, _ := CreateTemporaryAlloca(builder, mod, i8ptrType, "")
+
+		if size < targetData.TypeAllocSize(i8ptrType) {
+			// The alloca is bigger than the value that will be stored in it.
+			// To avoid having some bits undefined, zero the alloca first.
+			// Hopefully this will get optimized away.
+			builder.CreateStore(llvm.ConstNull(i8ptrType), packedAlloc)
+		}
+
+		// Store all values in the alloca.
+		packedAllocCast := builder.CreateBitCast(packedAlloc, llvm.PointerType(packedType, 0), "")
+		for i, value := range values {
+			indices := []llvm.Value{
+				llvm.ConstInt(ctx.Int32Type(), 0, false),
+				llvm.ConstInt(ctx.Int32Type(), uint64(i), false),
+			}
+			gep := builder.CreateInBoundsGEP(packedAllocCast, indices, "")
+			builder.CreateStore(value, gep)
+		}
+
+		// Load value (the *i8) from the alloca.
+		result := builder.CreateLoad(packedAlloc, "")
+
+		// End the lifetime of the alloca, to help the optimizer.
+		packedPtr := builder.CreateBitCast(packedAlloc, i8ptrType, "")
+		packedSize := llvm.ConstInt(ctx.Int64Type(), targetData.TypeAllocSize(packedAlloc.Type()), false)
+		EmitLifetimeEnd(builder, mod, packedPtr, packedSize)
+
+		return result
 	} else {
 		// Check if the values are all constants.
 		constant := true
@@ -67,7 +96,7 @@ func EmitPointerPack(builder llvm.Builder, mod llvm.Module, config *compileopts.
 		// Packed data is bigger than a pointer, so allocate it on the heap.
 		sizeValue := llvm.ConstInt(uintptrType, size, false)
 		alloc := mod.NamedFunction("runtime.alloc")
-		packedHeapAlloc = builder.CreateCall(alloc, []llvm.Value{
+		packedHeapAlloc := builder.CreateCall(alloc, []llvm.Value{
 			sizeValue,
 			llvm.Undef(i8ptrType),            // unused context parameter
 			llvm.ConstPointerNull(i8ptrType), // coroutine handle
@@ -80,28 +109,19 @@ func EmitPointerPack(builder llvm.Builder, mod llvm.Module, config *compileopts.
 				llvm.ConstPointerNull(i8ptrType), // coroutine handle
 			}, "")
 		}
-		packedAlloc = builder.CreateBitCast(packedHeapAlloc, llvm.PointerType(packedType, 0), "")
-	}
-	// Store all values in the alloca or heap pointer.
-	for i, value := range values {
-		indices := []llvm.Value{
-			llvm.ConstInt(ctx.Int32Type(), 0, false),
-			llvm.ConstInt(ctx.Int32Type(), uint64(i), false),
-		}
-		gep := builder.CreateInBoundsGEP(packedAlloc, indices, "")
-		builder.CreateStore(value, gep)
-	}
+		packedAlloc := builder.CreateBitCast(packedHeapAlloc, llvm.PointerType(packedType, 0), "")

-	if packedHeapAlloc.IsNil() {
-		// Load value (as *i8) from the alloca.
-		packedAlloc = builder.CreateBitCast(packedAlloc, llvm.PointerType(i8ptrType, 0), "")
-		result := builder.CreateLoad(packedAlloc, "")
-		packedPtr := builder.CreateBitCast(packedAlloc, i8ptrType, "")
-		packedSize := llvm.ConstInt(ctx.Int64Type(), targetData.TypeAllocSize(packedAlloc.Type()), false)
-		EmitLifetimeEnd(builder, mod, packedPtr, packedSize)
-		return result
-	} else {
-		// Get the original heap allocation pointer, which already is an *i8.
+		// Store all values in the heap pointer.
+		for i, value := range values {
+			indices := []llvm.Value{
+				llvm.ConstInt(ctx.Int32Type(), 0, false),
+				llvm.ConstInt(ctx.Int32Type(), uint64(i), false),
+			}
+			gep := builder.CreateInBoundsGEP(packedAlloc, indices, "")
+			builder.CreateStore(value, gep)
+		}
+
+		// Return the original heap allocation pointer, which already is an *i8.
 		return packedHeapAlloc
 	}
 }
@@ -4,11 +4,11 @@ go 1.11

 require (
 	github.com/blakesmith/ar v0.0.0-20150311145944-8bd4349a67f2
-	github.com/chromedp/cdproto v0.0.0-20200116234248-4da64dd111ac
-	github.com/chromedp/chromedp v0.5.3
+	github.com/chromedp/cdproto v0.0.0-20200709115526-d1f6fc58448b
+	github.com/chromedp/chromedp v0.5.4-0.20200303084119-2bb39134ab9e
 	github.com/google/shlex v0.0.0-20181106134648-c34317bd91bf
 	github.com/marcinbor85/gohex v0.0.0-20200531091804-343a4b548892
 	go.bug.st/serial v1.0.0
 	golang.org/x/tools v0.0.0-20200216192241-b320d3a0f5a2
-	tinygo.org/x/go-llvm v0.0.0-20200503225853-345b2947b59d
+	tinygo.org/x/go-llvm v0.0.0-20201104183921-570e7a6841d9
 )
@@ -1,9 +1,10 @@
 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 v0.0.0-20200116234248-4da64dd111ac h1:T7V5BXqnYd55Hj/g5uhDYumg9Fp3rMTS6bykYtTIFX4=
 github.com/chromedp/cdproto v0.0.0-20200116234248-4da64dd111ac/go.mod h1:PfAWWKJqjlGFYJEidUM6aVIWPr0EpobeyVWEEmplX7g=
-github.com/chromedp/chromedp v0.5.3 h1:F9LafxmYpsQhWQBdCs+6Sret1zzeeFyHS5LkRF//Ffg=
-github.com/chromedp/chromedp v0.5.3/go.mod h1:YLdPtndaHQ4rCpSpBG+IPpy9JvX0VD+7aaLxYgYj28w=
+github.com/chromedp/cdproto v0.0.0-20200709115526-d1f6fc58448b h1:LF+GRwyzxrO3MUzPvejv+yBup0lNG+/QdIRrkxOPseA=
+github.com/chromedp/cdproto v0.0.0-20200709115526-d1f6fc58448b/go.mod h1:E6LPWRdIJc11h/di5p0rwvRmUYbhGpBEH7ZbPfzDIOE=
+github.com/chromedp/chromedp v0.5.4-0.20200303084119-2bb39134ab9e h1:Hv0JVyHhbIXb9NiYQe4NsrfgrSofAp0q2FnhhJOXgi8=
+github.com/chromedp/chromedp v0.5.4-0.20200303084119-2bb39134ab9e/go.mod h1:vmQMRHFZrY3T+Jv51T0n87OK/i6bK+5P9a+Fg5jPwgQ=
 github.com/creack/goselect v0.1.1 h1:tiSSgKE1eJtxs1h/VgGQWuXUP0YS4CDIFMp6vaI1ls0=
 github.com/creack/goselect v0.1.1/go.mod h1:a/NhLweNvqIYMuxcMOuWY516Cimucms3DglDzQP3hKY=
 github.com/davecgh/go-spew v1.1.0/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
@@ -11,14 +12,15 @@ github.com/gobwas/httphead v0.0.0-20180130184737-2c6c146eadee h1:s+21KNqlpePfkah
 github.com/gobwas/httphead v0.0.0-20180130184737-2c6c146eadee/go.mod h1:L0fX3K22YWvt/FAX9NnzrNzcI4wNYi9Yku4O0LKYflo=
 github.com/gobwas/pool v0.2.0 h1:QEmUOlnSjWtnpRGHF3SauEiOsy82Cup83Vf2LcMlnc8=
 github.com/gobwas/pool v0.2.0/go.mod h1:q8bcK0KcYlCgd9e7WYLm9LpyS+YeLd8JVDW6WezmKEw=
-github.com/gobwas/ws v1.0.2 h1:CoAavW/wd/kulfZmSIBt6p24n4j7tHgNVCjsfHVNUbo=
-github.com/gobwas/ws v1.0.2/go.mod h1:szmBTxLgaFppYjEmNtny/v3w89xOydFnnZMcgRRu/EM=
+github.com/gobwas/ws v1.0.3 h1:ZOigqf7iBxkA4jdQ3am7ATzdlOFp9YzA6NmuvEEZc9g=
+github.com/gobwas/ws v1.0.3/go.mod h1:szmBTxLgaFppYjEmNtny/v3w89xOydFnnZMcgRRu/EM=
 github.com/google/shlex v0.0.0-20181106134648-c34317bd91bf h1:7+FW5aGwISbqUtkfmIpZJGRgNFg2ioYPvFaUxdqpDsg=
 github.com/google/shlex v0.0.0-20181106134648-c34317bd91bf/go.mod h1:RpwtwJQFrIEPstU94h88MWPXP2ektJZ8cZ0YntAmXiE=
 github.com/knq/sysutil v0.0.0-20191005231841-15668db23d08 h1:V0an7KRw92wmJysvFvtqtKMAPmvS5O0jtB0nYo6t+gs=
 github.com/knq/sysutil v0.0.0-20191005231841-15668db23d08/go.mod h1:dFWs1zEqDjFtnBXsd1vPOZaLsESovai349994nHx3e0=
-github.com/mailru/easyjson v0.7.0 h1:aizVhC/NAAcKWb+5QsU1iNOZb4Yws5UO2I+aIprQITM=
 github.com/mailru/easyjson v0.7.0/go.mod h1:KAzv3t3aY1NaHWoQz1+4F1ccyAH66Jk7yos7ldAVICs=
+github.com/mailru/easyjson v0.7.1 h1:mdxE1MF9o53iCb2Ghj1VfWvh7ZOwHpnVG/xwXrV90U8=
+github.com/mailru/easyjson v0.7.1/go.mod h1:KAzv3t3aY1NaHWoQz1+4F1ccyAH66Jk7yos7ldAVICs=
 github.com/marcinbor85/gohex v0.0.0-20200531091804-343a4b548892 h1:6J+qramlHVLmiBOgRiBOnQkno8uprqG6YFFQTt6uYIw=
 github.com/marcinbor85/gohex v0.0.0-20200531091804-343a4b548892/go.mod h1:Pb6XcsXyropB9LNHhnqaknG/vEwYztLkQzVCHv8sQ3M=
 github.com/pmezard/go-difflib v1.0.0/go.mod h1:iKH77koFhYxTK1pcRnkKkqfTogsbg7gZNVY4sRDYZ/4=
@@ -46,5 +48,5 @@ golang.org/x/xerrors v0.0.0-20191011141410-1b5146add898 h1:/atklqdjdhuosWIl6AIbO
 golang.org/x/xerrors v0.0.0-20191011141410-1b5146add898/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 v0.0.0-20200503225853-345b2947b59d h1:hcX7vpB067GWM/EH4sGGOti0PMgIx+0bbZwUXctOIvE=
-tinygo.org/x/go-llvm v0.0.0-20200503225853-345b2947b59d/go.mod h1:fv1F0BSNpxMfCL0zF3M4OPFbgYHnhtB6ST0HvUtu/LE=
+tinygo.org/x/go-llvm v0.0.0-20201104183921-570e7a6841d9 h1:l2kTQOhqEoeDTK3ckUnwReOQwMPUmURMIdjJbeAuDT4=
+tinygo.org/x/go-llvm v0.0.0-20201104183921-570e7a6841d9/go.mod h1:fv1F0BSNpxMfCL0zF3M4OPFbgYHnhtB6ST0HvUtu/LE=
@@ -12,7 +12,7 @@ import (

 // Version of TinyGo.
 // Update this value before release of new version of software.
-const Version = "0.14.0"
+const Version = "0.17.0-dev"

 // GetGorootVersion returns the major and minor version for a given GOROOT path.
 // If the goroot cannot be determined, (0, 0) is returned.
@@ -6,50 +6,81 @@ possible and only run unknown expressions (e.g. external calls) at runtime. This
 is in practice a partial evaluator of the `runtime.initAll` function, which
 calls each package initializer.

-It works by directly interpreting LLVM IR:
+This package is a rewrite of a previous partial evaluator that worked
+directly on LLVM IR and used the module and LLVM constants as intermediate
+values. This newer version instead uses a mostly Go intermediate form. It
+compiles functions and extracts relevant data first (compiler.go), then
+executes those functions (interpreter.go) in a memory space that can be
+rolled back per function (memory.go). This means that it is not necessary to
+scan functions to see whether they can be run at compile time, which was very
+error prone. Instead it just tries to execute everything and if it hits
+something it cannot interpret (such as a store to memory-mapped I/O) it rolls
+back the execution of that function and runs the function at runtime instead.
+All in all, this design provides several benefits:

-  * Almost all operations work directly on constants, and are implemented using
-    the llvm.Const* set of functions that are evaluated directly.
-  * External function calls and some other operations (inline assembly, volatile
-    load, volatile store) are seen as having limited side effects. Limited in
-    the sense that it is known at compile time which globals it affects, which
-    then are marked 'dirty' (meaning, further operations on it must be done at
-    runtime). These operations are emitted directly in the `runtime.initAll`
-    function. Return values are also considered 'dirty'.
-  * Such 'dirty' objects and local values must be executed at runtime instead of
-    at compile time. This dirtyness propagates further through the IR, for
-    example storing a dirty local value to a global also makes the global dirty,
-    meaning that the global may not be read or written at compile time as it's
-    contents at that point during interpretation is unknown.
-  * There are some heuristics in place to avoid doing too much with dirty
-    values. For example, a branch based on a dirty local marks the whole
-    function itself as having side effect (as if it is an external function).
-    However, all globals it touches are still taken into account and when a call
-    is inserted in `runtime.initAll`, all globals it references are also marked
-    dirty.
-  * Heap allocation (`runtime.alloc`) is emulated by creating new objects. The
-    value in the allocation is the initializer of the global, the zero value is
-    the zero initializer.
-  * Stack allocation (`alloca`) is often emulated using a fake alloca object,
-    until the address of the alloca is taken in which case it is also created as
-    a real `alloca` in `runtime.initAll` and marked dirty. This may be necessary
-    when calling an external function with the given alloca as paramter.
+  * Much better error handling. By being able to revert to runtime execution
+    without the need for scanning functions, this version is able to
+    automatically work around many bugs in the previous implementation.
+  * More correct memory model. This is not inherent to the new design, but the
+    new design also made the memory model easier to reason about.
+  * Faster execution of initialization code. While it is not much faster for
+    normal interpretation (maybe 25% or so) due to the compilation overhead,
+    it should be a whole lot faster for loops as it doesn't have to call into
+    LLVM (via CGo) for every operation.
+
+As mentioned, this partial evaulator comes in three parts: a compiler, an
+interpreter, and a memory manager.
+
+## Compiler
+
+The main task of the compiler is that it extracts all necessary data from
+every instruction in a function so that when this instruction is interpreted,
+no additional CGo calls are necessary. This is not currently done for all
+instructions (`runtime.alloc` is a notable exception), but at least it does
+so for the vast majority of instructions.
+
+## Interpreter
+
+The interpreter runs an instruction just like it would if it were executed
+'for real'. The vast majority of instructions can be executed at compile
+time. As indicated above, some instructions need to be executed at runtime
+instead.
+
+## Memory
+
+Memory is represented as objects (the `object` type) that contains data that
+will eventually be stored in a global and values (the `value` interface) that
+can be worked with while running the interpreter. Values therefore are only
+used locally and are always passed by value (just like most LLVM constants)
+while objects represent the backing storage (like LLVM globals). Some values
+are pointer values, and point to an object.
+
+Importantly, this partial evaluator can roll back the execution of a
+function. This is implemented by creating a new memory view per function
+activation, which makes sure that any change to a global (such as a store
+instruction) is stored in the memory view. It creates a copy of the object
+and stores that in the memory view to be modified. Once the function has
+executed successfully, all these modified objects are then copied into the
+parent function, up to the root function invocation which (on successful
+execution) writes the values back into the LLVM module. This way, function
+invocations can be rolled back without leaving a trace.
+
+Pointer values point to memory objects, but not to a particular memory
+object. Every memory object is given an index, and pointers use that index to
+look up the current active object for the pointer to load from or to copy
+when storing to it.
+
+Rolling back a function should roll back everyting, including the few
+instructions emitted at runtime. This is done by treating instructions much
+like memory objects and removing the created instructions when necessary.

 ## Why is this necessary?

 A partial evaluator is hard to get right, so why go through all the trouble of
 writing one?

-The main reason is that the previous attempt wasn't complete and wasn't sound.
-It simply tried to evaluate Go SSA directly, which was good but more difficult
-than necessary. An IR based interpreter needs to understand fewer instructions
-as the LLVM IR simply has less (complex) instructions than Go SSA. Also, LLVM
-provides some useful tools like easily getting all uses of a function or global,
-which Go SSA does not provide.
-
-But why is it necessary at all? The answer is that globals with initializers are
-much easier to optimize by LLVM than initialization code. Also, there are a few
-other benefits:
+The answer is that globals with initializers are much easier to optimize by
+LLVM than initialization code. Also, there are a few other benefits:

  * Dead globals are trivial to optimize away.
  * Constant globals are easier to detect. Remember that Go does not have global
@@ -60,5 +91,29 @@ other benefits:
  * Constants are much more efficent on microcontrollers, as they can be
    allocated in flash instead of RAM.

+The Go SSA package does not create constant initializers for globals.
+Instead, it emits initialization functions, so if you write the following:
+
+```go
+var foo = []byte{1, 2, 3, 4}
+```
+
+It would generate something like this:
+
+```go
+var foo []byte
+
+func init() {
+    foo = make([]byte, 4)
+    foo[0] = 1
+    foo[1] = 2
+    foo[2] = 3
+    foo[3] = 4
+}
+```
+
+This is of course hugely wasteful, it's much better to create `foo` as a
+global array instead of initializing it at runtime.
+
 For more details, see [this section of the
 documentation](https://tinygo.org/compiler-internals/differences-from-go/).
@@ -0,0 +1,410 @@
+package interp
+
+// This file compiles the LLVM IR to a form that's easy to efficiently
+// interpret.
+
+import (
+	"strings"
+
+	"tinygo.org/x/go-llvm"
+)
+
+// A function is a compiled LLVM function, which means that interpreting it
+// avoids most CGo calls necessary. This is done in a separate step so the
+// result can be cached.
+// Functions are in SSA form, just like the LLVM version if it. The first block
+// (blocks[0]) is the entry block.
+type function struct {
+	llvmFn llvm.Value
+	name   string       // precalculated llvmFn.Name()
+	params []llvm.Value // precalculated llvmFn.Params()
+	blocks []*basicBlock
+	locals map[llvm.Value]int
+}
+
+// basicBlock represents a LLVM basic block and contains a slice of
+// instructions. The last instruction must be a terminator instruction.
+type basicBlock struct {
+	instructions []instruction
+}
+
+// instruction is a precompiled LLVM IR instruction. The operands can be either
+// an already known value (such as literalValue or pointerValue) but can also be
+// the special localValue, which means that the value is a function parameter or
+// is produced by another instruction in the function. In that case, the
+// interpreter will replace the operand with that local value.
+type instruction struct {
+	opcode     llvm.Opcode
+	localIndex int
+	operands   []value
+	llvmInst   llvm.Value
+	name       string
+}
+
+// String returns a nice human-readable version of this instruction.
+func (inst *instruction) String() string {
+	operands := make([]string, len(inst.operands))
+	for i, op := range inst.operands {
+		operands[i] = op.String()
+	}
+
+	name := instructionNameMap[inst.opcode]
+	if name == "" {
+		name = "<unknown op>"
+	}
+	return name + " " + strings.Join(operands, " ")
+}
+
+// compileFunction compiles a given LLVM function to an easier to interpret
+// version of the function. As far as possible, all operands are preprocessed so
+// that the interpreter doesn't have to call into LLVM.
+func (r *runner) compileFunction(llvmFn llvm.Value) *function {
+	fn := &function{
+		llvmFn: llvmFn,
+		name:   llvmFn.Name(),
+		params: llvmFn.Params(),
+		locals: make(map[llvm.Value]int),
+	}
+	if llvmFn.IsDeclaration() {
+		// Nothing to do.
+		return fn
+	}
+
+	for i, param := range fn.params {
+		fn.locals[param] = i
+	}
+
+	// Make a map of all the blocks, to quickly find the block number for a
+	// given branch instruction.
+	blockIndices := make(map[llvm.Value]int)
+	for llvmBB := llvmFn.FirstBasicBlock(); !llvmBB.IsNil(); llvmBB = llvm.NextBasicBlock(llvmBB) {
+		index := len(blockIndices)
+		blockIndices[llvmBB.AsValue()] = index
+	}
+
+	// Compile every block.
+	for llvmBB := llvmFn.FirstBasicBlock(); !llvmBB.IsNil(); llvmBB = llvm.NextBasicBlock(llvmBB) {
+		bb := &basicBlock{}
+		fn.blocks = append(fn.blocks, bb)
+
+		// Compile every instruction in the block.
+		for llvmInst := llvmBB.FirstInstruction(); !llvmInst.IsNil(); llvmInst = llvm.NextInstruction(llvmInst) {
+			// Create instruction skeleton.
+			opcode := llvmInst.InstructionOpcode()
+			inst := instruction{
+				opcode:     opcode,
+				localIndex: len(fn.locals),
+				llvmInst:   llvmInst,
+			}
+			fn.locals[llvmInst] = len(fn.locals)
+
+			// Add operands specific for this instruction.
+			switch opcode {
+			case llvm.Ret:
+				// Return instruction, which can either be a `ret void` (no
+				// return value) or return a value.
+				numOperands := llvmInst.OperandsCount()
+				if numOperands != 0 {
+					inst.operands = []value{
+						r.getValue(llvmInst.Operand(0)),
+					}
+				}
+			case llvm.Br:
+				// Branch instruction. Can be either a conditional branch (with
+				// 3 operands) or unconditional branch (with just one basic
+				// block operand).
+				numOperands := llvmInst.OperandsCount()
+				switch numOperands {
+				case 3:
+					// Conditional jump to one of two blocks. Comparable to an
+					// if/else in procedural languages.
+					thenBB := llvmInst.Operand(2)
+					elseBB := llvmInst.Operand(1)
+					inst.operands = []value{
+						r.getValue(llvmInst.Operand(0)),
+						literalValue{uint32(blockIndices[thenBB])},
+						literalValue{uint32(blockIndices[elseBB])},
+					}
+				case 1:
+					// Unconditional jump to a target basic block. Comparable to
+					// a jump in C and Go.
+					jumpBB := llvmInst.Operand(0)
+					inst.operands = []value{
+						literalValue{uint32(blockIndices[jumpBB])},
+					}
+				default:
+					panic("unknown number of operands")
+				}
+			case llvm.PHI:
+				inst.name = llvmInst.Name()
+				incomingCount := inst.llvmInst.IncomingCount()
+				for i := 0; i < incomingCount; i++ {
+					incomingBB := inst.llvmInst.IncomingBlock(i)
+					incomingValue := inst.llvmInst.IncomingValue(i)
+					inst.operands = append(inst.operands,
+						literalValue{uint32(blockIndices[incomingBB.AsValue()])},
+						r.getValue(incomingValue),
+					)
+				}
+			case llvm.Select:
+				// Select is a special instruction that is much like a ternary
+				// operator. It produces operand 1 or 2 based on the boolean
+				// that is operand 0.
+				inst.name = llvmInst.Name()
+				inst.operands = []value{
+					r.getValue(llvmInst.Operand(0)),
+					r.getValue(llvmInst.Operand(1)),
+					r.getValue(llvmInst.Operand(2)),
+				}
+			case llvm.Call:
+				// Call is a regular function call but could also be a runtime
+				// intrinsic. Some runtime intrinsics are treated specially by
+				// the interpreter, such as runtime.alloc. We don't
+				// differentiate between them here because these calls may also
+				// need to be run at runtime, in which case they should all be
+				// created in the same way.
+				llvmCalledValue := llvmInst.CalledValue()
+				if !llvmCalledValue.IsAFunction().IsNil() {
+					name := llvmCalledValue.Name()
+					if name == "llvm.dbg.value" || strings.HasPrefix(name, "llvm.lifetime.") {
+						// These intrinsics should not be interpreted, they are not
+						// relevant to the execution of this function.
+						continue
+					}
+				}
+				inst.name = llvmInst.Name()
+				numOperands := llvmInst.OperandsCount()
+				inst.operands = append(inst.operands, r.getValue(llvmCalledValue))
+				for i := 0; i < numOperands-1; i++ {
+					inst.operands = append(inst.operands, r.getValue(llvmInst.Operand(i)))
+				}
+			case llvm.Load:
+				// Load instruction. The interpreter will load from the
+				// appropriate memory view.
+				// Also provide the memory size to be loaded, which is necessary
+				// with a lack of type information.
+				inst.name = llvmInst.Name()
+				inst.operands = []value{
+					r.getValue(llvmInst.Operand(0)),
+					literalValue{r.targetData.TypeAllocSize(llvmInst.Type())},
+				}
+			case llvm.Store:
+				// Store instruction. The interpreter will create a new object
+				// in the memory view of the function invocation and store to
+				// that, to make it possible to roll back this store.
+				inst.operands = []value{
+					r.getValue(llvmInst.Operand(0)),
+					r.getValue(llvmInst.Operand(1)),
+				}
+			case llvm.Alloca:
+				// Alloca allocates stack space for local variables.
+				numElements := r.getValue(inst.llvmInst.Operand(0)).(literalValue).value.(uint32)
+				elementSize := r.targetData.TypeAllocSize(inst.llvmInst.Type().ElementType())
+				inst.operands = []value{
+					literalValue{elementSize * uint64(numElements)},
+				}
+			case llvm.GetElementPtr:
+				// GetElementPtr does pointer arithmetic.
+				inst.name = llvmInst.Name()
+				ptr := llvmInst.Operand(0)
+				n := llvmInst.OperandsCount()
+				elementType := ptr.Type().ElementType()
+				// gep: [source ptr, dest value size, pairs of indices...]
+				inst.operands = []value{
+					r.getValue(ptr),
+					literalValue{r.targetData.TypeAllocSize(llvmInst.Type().ElementType())},
+					r.getValue(llvmInst.Operand(1)),
+					literalValue{r.targetData.TypeAllocSize(elementType)},
+				}
+				for i := 2; i < n; i++ {
+					operand := r.getValue(llvmInst.Operand(i))
+					if elementType.TypeKind() == llvm.StructTypeKind {
+						index := operand.(literalValue).value.(uint32)
+						elementOffset := r.targetData.ElementOffset(elementType, int(index))
+						// Encode operands in a special way. The elementOffset
+						// is just the offset in bytes. The elementSize is a
+						// negative number (when cast to a int64) by flipping
+						// all the bits. This allows the interpreter to detect
+						// this is a struct field and that it should not
+						// multiply it with the elementOffset to get the offset.
+						// It is important for the interpreter to know the
+						// struct field index for when the GEP must be done at
+						// runtime.
+						inst.operands = append(inst.operands, literalValue{elementOffset}, literalValue{^uint64(index)})
+						elementType = elementType.StructElementTypes()[index]
+					} else {
+						elementType = elementType.ElementType()
+						elementSize := r.targetData.TypeAllocSize(elementType)
+						elementSizeOperand := literalValue{elementSize}
+						// Add operand * elementSizeOperand bytes to the pointer.
+						inst.operands = append(inst.operands, operand, elementSizeOperand)
+					}
+				}
+			case llvm.BitCast, llvm.IntToPtr, llvm.PtrToInt:
+				// Bitcasts are ususally used to cast a pointer from one type to
+				// another leaving the pointer itself intact.
+				inst.name = llvmInst.Name()
+				inst.operands = []value{
+					r.getValue(llvmInst.Operand(0)),
+				}
+			case llvm.ExtractValue:
+				inst.name = llvmInst.Name()
+				agg := llvmInst.Operand(0)
+				var offset uint64
+				indexingType := agg.Type()
+				for _, index := range inst.llvmInst.Indices() {
+					switch indexingType.TypeKind() {
+					case llvm.StructTypeKind:
+						offset += r.targetData.ElementOffset(indexingType, int(index))
+						indexingType = indexingType.StructElementTypes()[index]
+					default: // ArrayTypeKind
+						indexingType = indexingType.ElementType()
+						elementSize := r.targetData.TypeAllocSize(indexingType)
+						offset += elementSize * uint64(index)
+					}
+				}
+				size := r.targetData.TypeAllocSize(inst.llvmInst.Type())
+				// extractvalue [agg, byteOffset, byteSize]
+				inst.operands = []value{
+					r.getValue(agg),
+					literalValue{offset},
+					literalValue{size},
+				}
+			case llvm.InsertValue:
+				inst.name = llvmInst.Name()
+				agg := llvmInst.Operand(0)
+				var offset uint64
+				indexingType := agg.Type()
+				for _, index := range inst.llvmInst.Indices() {
+					switch indexingType.TypeKind() {
+					case llvm.StructTypeKind:
+						offset += r.targetData.ElementOffset(indexingType, int(index))
+						indexingType = indexingType.StructElementTypes()[index]
+					default: // ArrayTypeKind
+						indexingType = indexingType.ElementType()
+						elementSize := r.targetData.TypeAllocSize(indexingType)
+						offset += elementSize * uint64(index)
+					}
+				}
+				// insertvalue [agg, elt, byteOffset]
+				inst.operands = []value{
+					r.getValue(agg),
+					r.getValue(llvmInst.Operand(1)),
+					literalValue{offset},
+				}
+			case llvm.ICmp:
+				inst.name = llvmInst.Name()
+				inst.operands = []value{
+					r.getValue(llvmInst.Operand(0)),
+					r.getValue(llvmInst.Operand(1)),
+					literalValue{uint8(llvmInst.IntPredicate())},
+				}
+			case llvm.FCmp:
+				inst.name = llvmInst.Name()
+				inst.operands = []value{
+					r.getValue(llvmInst.Operand(0)),
+					r.getValue(llvmInst.Operand(1)),
+					literalValue{uint8(llvmInst.FloatPredicate())},
+				}
+			case llvm.Add, llvm.Sub, llvm.Mul, llvm.UDiv, llvm.SDiv, llvm.URem, llvm.SRem, llvm.Shl, llvm.LShr, llvm.AShr, llvm.And, llvm.Or, llvm.Xor:
+				// Integer binary operations.
+				inst.name = llvmInst.Name()
+				inst.operands = []value{
+					r.getValue(llvmInst.Operand(0)),
+					r.getValue(llvmInst.Operand(1)),
+				}
+			case llvm.SExt, llvm.ZExt, llvm.Trunc:
+				// Extend or shrink an integer size.
+				// No sign extension going on so easy to do.
+				// zext: [value, bitwidth]
+				// trunc: [value, bitwidth]
+				inst.name = llvmInst.Name()
+				inst.operands = []value{
+					r.getValue(llvmInst.Operand(0)),
+					literalValue{uint64(llvmInst.Type().IntTypeWidth())},
+				}
+			case llvm.SIToFP, llvm.UIToFP:
+				// Convert an integer to a floating point instruction.
+				// opcode: [value, bitwidth]
+				inst.name = llvmInst.Name()
+				inst.operands = []value{
+					r.getValue(llvmInst.Operand(0)),
+					literalValue{uint64(r.targetData.TypeAllocSize(llvmInst.Type()) * 8)},
+				}
+			default:
+				// Unknown instruction, which is already set in inst.opcode so
+				// is detectable.
+				// This error is handled when actually trying to interpret this
+				// instruction (to not trigger on code that won't be executed).
+			}
+			bb.instructions = append(bb.instructions, inst)
+		}
+	}
+	return fn
+}
+
+// instructionNameMap maps from instruction opcodes to instruction names. This
+// can be useful for debug logging.
+var instructionNameMap = [...]string{
+	llvm.Ret:         "ret",
+	llvm.Br:          "br",
+	llvm.Switch:      "switch",
+	llvm.IndirectBr:  "indirectbr",
+	llvm.Invoke:      "invoke",
+	llvm.Unreachable: "unreachable",
+
+	// Standard Binary Operators
+	llvm.Add:  "add",
+	llvm.FAdd: "fadd",
+	llvm.Sub:  "sub",
+	llvm.FSub: "fsub",
+	llvm.Mul:  "mul",
+	llvm.FMul: "fmul",
+	llvm.UDiv: "udiv",
+	llvm.SDiv: "sdiv",
+	llvm.FDiv: "fdiv",
+	llvm.URem: "urem",
+	llvm.SRem: "srem",
+	llvm.FRem: "frem",
+
+	// Logical Operators
+	llvm.Shl:  "shl",
+	llvm.LShr: "lshr",
+	llvm.AShr: "ashr",
+	llvm.And:  "and",
+	llvm.Or:   "or",
+	llvm.Xor:  "xor",
+
+	// Memory Operators
+	llvm.Alloca:        "alloca",
+	llvm.Load:          "load",
+	llvm.Store:         "store",
+	llvm.GetElementPtr: "getelementptr",
+
+	// Cast Operators
+	llvm.Trunc:    "trunc",
+	llvm.ZExt:     "zext",
+	llvm.SExt:     "sext",
+	llvm.FPToUI:   "fptoui",
+	llvm.FPToSI:   "fptosi",
+	llvm.UIToFP:   "uitofp",
+	llvm.SIToFP:   "sitofp",
+	llvm.FPTrunc:  "fptrunc",
+	llvm.FPExt:    "fpext",
+	llvm.PtrToInt: "ptrtoint",
+	llvm.IntToPtr: "inttoptr",
+	llvm.BitCast:  "bitcast",
+
+	// Other Operators
+	llvm.ICmp:           "icmp",
+	llvm.FCmp:           "fcmp",
+	llvm.PHI:            "phi",
+	llvm.Call:           "call",
+	llvm.Select:         "select",
+	llvm.VAArg:          "vaarg",
+	llvm.ExtractElement: "extractelement",
+	llvm.InsertElement:  "insertelement",
+	llvm.ShuffleVector:  "shufflevector",
+	llvm.ExtractValue:   "extractvalue",
+	llvm.InsertValue:    "insertvalue",
+}
@@ -11,15 +11,17 @@ import (
 	"tinygo.org/x/go-llvm"
 )

-// errUnreachable is returned when an unreachable instruction is executed. This
-// error should not be visible outside of the interp package.
-var errUnreachable = &Error{Err: errors.New("interp: unreachable executed")}
+// These errors are expected during normal execution and can be recovered from
+// by running the affected function at runtime instead of compile time.
+var (
+	errIntegerAsPointer       = errors.New("interp: trying to use an integer as a pointer (memory-mapped I/O?)")
+	errUnsupportedInst        = errors.New("interp: unsupported instruction")
+	errUnsupportedRuntimeInst = errors.New("interp: unsupported instruction (to be emitted at runtime)")
+	errMapAlreadyCreated      = errors.New("interp: map already created")
+)

-// unsupportedInstructionError returns a new "unsupported instruction" error for
-// the given instruction. It includes source location information, when
-// available.
-func (e *evalPackage) unsupportedInstructionError(inst llvm.Value) *Error {
-	return e.errorAt(inst, errors.New("interp: unsupported instruction"))
+func isRecoverableError(err error) bool {
+	return err == errIntegerAsPointer || err == errUnsupportedInst || err == errUnsupportedRuntimeInst || err == errMapAlreadyCreated
 }

 // ErrorLine is one line in a traceback. The position may be missing.
@@ -46,11 +48,13 @@ func (e *Error) Error() string {
 // errorAt returns an error value for the currently interpreted package at the
 // location of the instruction. The location information may not be complete as
 // it depends on debug information in the IR.
-func (e *evalPackage) errorAt(inst llvm.Value, err error) *Error {
+func (r *runner) errorAt(inst instruction, err error) *Error {
+	pos := getPosition(inst.llvmInst)
 	return &Error{
-		ImportPath: e.packagePath,
-		Pos:        getPosition(inst),
+		ImportPath: r.pkgName,
+		Pos:        pos,
 		Err:        err,
+		Traceback:  []ErrorLine{{pos, inst.llvmInst}},
 	}
 }

@@ -1,680 +0,0 @@
-package interp
-
-// This file implements the core interpretation routines, interpreting single
-// functions.
-
-import (
-	"errors"
-	"strings"
-
-	"tinygo.org/x/go-llvm"
-)
-
-type frame struct {
-	*evalPackage
-	fn     llvm.Value
-	locals map[llvm.Value]Value
-}
-
-// evalBasicBlock evaluates a single basic block, returning the return value (if
-// ending with a ret instruction), a list of outgoing basic blocks (if not
-// ending with a ret instruction), or an error on failure.
-// Most of it works at compile time. Some calls get translated into calls to be
-// executed at runtime: calls to functions with side effects, external calls,
-// and operations on the result of such instructions.
-func (fr *frame) evalBasicBlock(bb, incoming llvm.BasicBlock, indent string) (retval Value, outgoing []llvm.Value, err *Error) {
-	for inst := bb.FirstInstruction(); !inst.IsNil(); inst = llvm.NextInstruction(inst) {
-		if fr.Debug {
-			print(indent)
-			inst.Dump()
-			println()
-		}
-		switch {
-		case !inst.IsABinaryOperator().IsNil():
-			lhs := fr.getLocal(inst.Operand(0)).(*LocalValue).Underlying
-			rhs := fr.getLocal(inst.Operand(1)).(*LocalValue).Underlying
-
-			switch inst.InstructionOpcode() {
-			// Standard binary operators
-			case llvm.Add:
-				fr.locals[inst] = &LocalValue{fr.Eval, fr.builder.CreateAdd(lhs, rhs, "")}
-			case llvm.FAdd:
-				fr.locals[inst] = &LocalValue{fr.Eval, fr.builder.CreateFAdd(lhs, rhs, "")}
-			case llvm.Sub:
-				fr.locals[inst] = &LocalValue{fr.Eval, fr.builder.CreateSub(lhs, rhs, "")}
-			case llvm.FSub:
-				fr.locals[inst] = &LocalValue{fr.Eval, fr.builder.CreateFSub(lhs, rhs, "")}
-			case llvm.Mul:
-				fr.locals[inst] = &LocalValue{fr.Eval, fr.builder.CreateMul(lhs, rhs, "")}
-			case llvm.FMul:
-				fr.locals[inst] = &LocalValue{fr.Eval, fr.builder.CreateFMul(lhs, rhs, "")}
-			case llvm.UDiv:
-				fr.locals[inst] = &LocalValue{fr.Eval, fr.builder.CreateUDiv(lhs, rhs, "")}
-			case llvm.SDiv:
-				fr.locals[inst] = &LocalValue{fr.Eval, fr.builder.CreateSDiv(lhs, rhs, "")}
-			case llvm.FDiv:
-				fr.locals[inst] = &LocalValue{fr.Eval, fr.builder.CreateFDiv(lhs, rhs, "")}
-			case llvm.URem:
-				fr.locals[inst] = &LocalValue{fr.Eval, fr.builder.CreateURem(lhs, rhs, "")}
-			case llvm.SRem:
-				fr.locals[inst] = &LocalValue{fr.Eval, fr.builder.CreateSRem(lhs, rhs, "")}
-			case llvm.FRem:
-				fr.locals[inst] = &LocalValue{fr.Eval, fr.builder.CreateFRem(lhs, rhs, "")}
-
-			// Logical operators
-			case llvm.Shl:
-				fr.locals[inst] = &LocalValue{fr.Eval, fr.builder.CreateShl(lhs, rhs, "")}
-			case llvm.LShr:
-				fr.locals[inst] = &LocalValue{fr.Eval, fr.builder.CreateLShr(lhs, rhs, "")}
-			case llvm.AShr:
-				fr.locals[inst] = &LocalValue{fr.Eval, fr.builder.CreateAShr(lhs, rhs, "")}
-			case llvm.And:
-				fr.locals[inst] = &LocalValue{fr.Eval, fr.builder.CreateAnd(lhs, rhs, "")}
-			case llvm.Or:
-				fr.locals[inst] = &LocalValue{fr.Eval, fr.builder.CreateOr(lhs, rhs, "")}
-			case llvm.Xor:
-				fr.locals[inst] = &LocalValue{fr.Eval, fr.builder.CreateXor(lhs, rhs, "")}
-
-			default:
-				return nil, nil, fr.unsupportedInstructionError(inst)
-			}
-
-		// Memory operators
-		case !inst.IsAAllocaInst().IsNil():
-			allocType := inst.Type().ElementType()
-			alloca := llvm.AddGlobal(fr.Mod, allocType, fr.packagePath+"$alloca")
-			alloca.SetInitializer(llvm.ConstNull(allocType))
-			alloca.SetLinkage(llvm.InternalLinkage)
-			fr.locals[inst] = &LocalValue{
-				Underlying: alloca,
-				Eval:       fr.Eval,
-			}
-		case !inst.IsALoadInst().IsNil():
-			operand := fr.getLocal(inst.Operand(0)).(*LocalValue)
-			var value llvm.Value
-			if !operand.IsConstant() || inst.IsVolatile() || (!operand.Underlying.IsAConstantExpr().IsNil() && operand.Underlying.Opcode() == llvm.BitCast) {
-				value = fr.builder.CreateLoad(operand.Value(), inst.Name())
-			} else {
-				value = operand.Load()
-			}
-			if value.Type() != inst.Type() {
-				return nil, nil, fr.errorAt(inst, errors.New("interp: load: type does not match"))
-			}
-			fr.locals[inst] = fr.getValue(value)
-		case !inst.IsAStoreInst().IsNil():
-			value := fr.getLocal(inst.Operand(0))
-			ptr := fr.getLocal(inst.Operand(1))
-			if inst.IsVolatile() {
-				fr.builder.CreateStore(value.Value(), ptr.Value())
-			} else {
-				ptr.Store(value.Value())
-			}
-		case !inst.IsAGetElementPtrInst().IsNil():
-			value := fr.getLocal(inst.Operand(0))
-			llvmIndices := make([]llvm.Value, inst.OperandsCount()-1)
-			for i := range llvmIndices {
-				llvmIndices[i] = inst.Operand(i + 1)
-			}
-			indices := make([]uint32, len(llvmIndices))
-			for i, llvmIndex := range llvmIndices {
-				operand := fr.getLocal(llvmIndex)
-				if !operand.IsConstant() {
-					// Not a constant operation.
-					// This should be detected by the scanner, but isn't at the
-					// moment.
-					return nil, nil, fr.errorAt(inst, errors.New("todo: non-const gep"))
-				}
-				indices[i] = uint32(operand.Value().ZExtValue())
-			}
-			result, err := value.GetElementPtr(indices)
-			if err != nil {
-				return nil, nil, fr.errorAt(inst, err)
-			}
-			if result.Type() != inst.Type() {
-				return nil, nil, fr.errorAt(inst, errors.New("interp: gep: type does not match"))
-			}
-			fr.locals[inst] = result
-
-		// Cast operators
-		case !inst.IsATruncInst().IsNil():
-			value := fr.getLocal(inst.Operand(0))
-			fr.locals[inst] = &LocalValue{fr.Eval, fr.builder.CreateTrunc(value.(*LocalValue).Value(), inst.Type(), "")}
-		case !inst.IsAZExtInst().IsNil():
-			value := fr.getLocal(inst.Operand(0))
-			fr.locals[inst] = &LocalValue{fr.Eval, fr.builder.CreateZExt(value.(*LocalValue).Value(), inst.Type(), "")}
-		case !inst.IsASExtInst().IsNil():
-			value := fr.getLocal(inst.Operand(0))
-			fr.locals[inst] = &LocalValue{fr.Eval, fr.builder.CreateSExt(value.(*LocalValue).Value(), inst.Type(), "")}
-		case !inst.IsAFPToUIInst().IsNil():
-			value := fr.getLocal(inst.Operand(0))
-			fr.locals[inst] = &LocalValue{fr.Eval, fr.builder.CreateFPToUI(value.(*LocalValue).Value(), inst.Type(), "")}
-		case !inst.IsAFPToSIInst().IsNil():
-			value := fr.getLocal(inst.Operand(0))
-			fr.locals[inst] = &LocalValue{fr.Eval, fr.builder.CreateFPToSI(value.(*LocalValue).Value(), inst.Type(), "")}
-		case !inst.IsAUIToFPInst().IsNil():
-			value := fr.getLocal(inst.Operand(0))
-			fr.locals[inst] = &LocalValue{fr.Eval, fr.builder.CreateUIToFP(value.(*LocalValue).Value(), inst.Type(), "")}
-		case !inst.IsASIToFPInst().IsNil():
-			value := fr.getLocal(inst.Operand(0))
-			fr.locals[inst] = &LocalValue{fr.Eval, fr.builder.CreateSIToFP(value.(*LocalValue).Value(), inst.Type(), "")}
-		case !inst.IsAFPTruncInst().IsNil():
-			value := fr.getLocal(inst.Operand(0))
-			fr.locals[inst] = &LocalValue{fr.Eval, fr.builder.CreateFPTrunc(value.(*LocalValue).Value(), inst.Type(), "")}
-		case !inst.IsAFPExtInst().IsNil():
-			value := fr.getLocal(inst.Operand(0))
-			fr.locals[inst] = &LocalValue{fr.Eval, fr.builder.CreateFPExt(value.(*LocalValue).Value(), inst.Type(), "")}
-		case !inst.IsAPtrToIntInst().IsNil():
-			value := fr.getLocal(inst.Operand(0))
-			fr.locals[inst] = &LocalValue{fr.Eval, fr.builder.CreatePtrToInt(value.Value(), inst.Type(), "")}
-		case !inst.IsABitCastInst().IsNil() && inst.Type().TypeKind() == llvm.PointerTypeKind:
-			operand := inst.Operand(0)
-			if !operand.IsACallInst().IsNil() {
-				fn := operand.CalledValue()
-				if !fn.IsAFunction().IsNil() && fn.Name() == "runtime.alloc" {
-					continue // special case: bitcast of alloc
-				}
-			}
-			if _, ok := fr.getLocal(operand).(*MapValue); ok {
-				// Special case for runtime.trackPointer calls.
-				// Note: this might not be entirely sound in some rare cases
-				// where the map is stored in a dirty global.
-				uses := getUses(inst)
-				if len(uses) == 1 {
-					use := uses[0]
-					if !use.IsACallInst().IsNil() && !use.CalledValue().IsAFunction().IsNil() && use.CalledValue().Name() == "runtime.trackPointer" {
-						continue
-					}
-				}
-				// It is not possible in Go to bitcast a map value to a pointer.
-				return nil, nil, fr.errorAt(inst, errors.New("unimplemented: bitcast of map"))
-			}
-			value := fr.getLocal(operand).(*LocalValue)
-			fr.locals[inst] = &LocalValue{fr.Eval, fr.builder.CreateBitCast(value.Value(), inst.Type(), "")}
-
-		// Other operators
-		case !inst.IsAICmpInst().IsNil():
-			lhs := fr.getLocal(inst.Operand(0)).(*LocalValue).Underlying
-			rhs := fr.getLocal(inst.Operand(1)).(*LocalValue).Underlying
-			predicate := inst.IntPredicate()
-			if predicate == llvm.IntEQ {
-				var lhsZero, rhsZero bool
-				var ok1, ok2 bool
-				if lhs.Type().TypeKind() == llvm.PointerTypeKind {
-					// Unfortunately, the const propagation in the IR builder
-					// doesn't handle pointer compares of inttoptr values. So we
-					// implement it manually here.
-					lhsZero, ok1 = isPointerNil(lhs)
-					rhsZero, ok2 = isPointerNil(rhs)
-				}
-				if lhs.Type().TypeKind() == llvm.IntegerTypeKind {
-					lhsZero, ok1 = isZero(lhs)
-					rhsZero, ok2 = isZero(rhs)
-				}
-				if ok1 && ok2 {
-					if lhsZero && rhsZero {
-						// Both are zero, so this icmp is always evaluated to true.
-						fr.locals[inst] = &LocalValue{fr.Eval, llvm.ConstInt(fr.Mod.Context().Int1Type(), 1, false)}
-						continue
-					}
-					if lhsZero != rhsZero {
-						// Only one of them is zero, so this comparison must return false.
-						fr.locals[inst] = &LocalValue{fr.Eval, llvm.ConstInt(fr.Mod.Context().Int1Type(), 0, false)}
-						continue
-					}
-				}
-			}
-			fr.locals[inst] = &LocalValue{fr.Eval, fr.builder.CreateICmp(predicate, lhs, rhs, "")}
-		case !inst.IsAFCmpInst().IsNil():
-			lhs := fr.getLocal(inst.Operand(0)).(*LocalValue).Underlying
-			rhs := fr.getLocal(inst.Operand(1)).(*LocalValue).Underlying
-			predicate := inst.FloatPredicate()
-			fr.locals[inst] = &LocalValue{fr.Eval, fr.builder.CreateFCmp(predicate, lhs, rhs, "")}
-		case !inst.IsAPHINode().IsNil():
-			for i := 0; i < inst.IncomingCount(); i++ {
-				if inst.IncomingBlock(i) == incoming {
-					fr.locals[inst] = fr.getLocal(inst.IncomingValue(i))
-				}
-			}
-		case !inst.IsACallInst().IsNil():
-			callee := inst.CalledValue()
-			switch {
-			case callee.Name() == "runtime.alloc":
-				// heap allocation
-				users := getUses(inst)
-				var resultInst = inst
-				if len(users) == 1 && !users[0].IsABitCastInst().IsNil() {
-					// happens when allocating something other than i8*
-					resultInst = users[0]
-				}
-				size := fr.getLocal(inst.Operand(0)).(*LocalValue).Underlying.ZExtValue()
-				allocType := resultInst.Type().ElementType()
-				typeSize := fr.TargetData.TypeAllocSize(allocType)
-				elementCount := 1
-				if size != typeSize {
-					// allocate an array
-					if size%typeSize != 0 {
-						return nil, nil, fr.unsupportedInstructionError(inst)
-					}
-					elementCount = int(size / typeSize)
-					allocType = llvm.ArrayType(allocType, elementCount)
-				}
-				alloc := llvm.AddGlobal(fr.Mod, allocType, fr.packagePath+"$alloc")
-				alloc.SetInitializer(llvm.ConstNull(allocType))
-				alloc.SetLinkage(llvm.InternalLinkage)
-				result := &LocalValue{
-					Underlying: alloc,
-					Eval:       fr.Eval,
-				}
-				if elementCount == 1 {
-					fr.locals[resultInst] = result
-				} else {
-					result, err := result.GetElementPtr([]uint32{0, 0})
-					if err != nil {
-						return nil, nil, fr.errorAt(inst, err)
-					}
-					fr.locals[resultInst] = result
-				}
-			case callee.Name() == "runtime.hashmapMake":
-				// create a map
-				keySize := inst.Operand(0).ZExtValue()
-				valueSize := inst.Operand(1).ZExtValue()
-				fr.locals[inst] = &MapValue{
-					Eval:      fr.Eval,
-					PkgName:   fr.packagePath,
-					KeySize:   int(keySize),
-					ValueSize: int(valueSize),
-				}
-			case callee.Name() == "runtime.hashmapStringSet":
-				// set a string key in the map
-				keyBuf := fr.getLocal(inst.Operand(1)).(*LocalValue)
-				keyLen := fr.getLocal(inst.Operand(2)).(*LocalValue)
-				valPtr := fr.getLocal(inst.Operand(3)).(*LocalValue)
-				m, ok := fr.getLocal(inst.Operand(0)).(*MapValue)
-				if !ok || !keyBuf.IsConstant() || !keyLen.IsConstant() || !valPtr.IsConstant() {
-					// The mapassign operation could not be done at compile
-					// time. Do it at runtime instead.
-					m := fr.getLocal(inst.Operand(0)).Value()
-					fr.markDirty(m)
-					llvmParams := []llvm.Value{
-						m,                                    // *runtime.hashmap
-						fr.getLocal(inst.Operand(1)).Value(), // key.ptr
-						fr.getLocal(inst.Operand(2)).Value(), // key.len
-						fr.getLocal(inst.Operand(3)).Value(), // value (unsafe.Pointer)
-						fr.getLocal(inst.Operand(4)).Value(), // context
-						fr.getLocal(inst.Operand(5)).Value(), // parentHandle
-					}
-					fr.builder.CreateCall(callee, llvmParams, "")
-					continue
-				}
-				// "key" is a Go string value, which in the TinyGo calling convention is split up
-				// into separate pointer and length parameters.
-				m.PutString(keyBuf, keyLen, valPtr)
-			case callee.Name() == "runtime.hashmapBinarySet":
-				// set a binary (int etc.) key in the map
-				keyBuf := fr.getLocal(inst.Operand(1)).(*LocalValue)
-				valPtr := fr.getLocal(inst.Operand(2)).(*LocalValue)
-				m, ok := fr.getLocal(inst.Operand(0)).(*MapValue)
-				if !ok || !keyBuf.IsConstant() || !valPtr.IsConstant() {
-					// The mapassign operation could not be done at compile
-					// time. Do it at runtime instead.
-					m := fr.getLocal(inst.Operand(0)).Value()
-					fr.markDirty(m)
-					llvmParams := []llvm.Value{
-						m,                                    // *runtime.hashmap
-						fr.getLocal(inst.Operand(1)).Value(), // key
-						fr.getLocal(inst.Operand(2)).Value(), // value
-						fr.getLocal(inst.Operand(3)).Value(), // context
-						fr.getLocal(inst.Operand(4)).Value(), // parentHandle
-					}
-					fr.builder.CreateCall(callee, llvmParams, "")
-					continue
-				}
-				m.PutBinary(keyBuf, valPtr)
-			case callee.Name() == "runtime.stringConcat":
-				// adding two strings together
-				buf1Ptr := fr.getLocal(inst.Operand(0))
-				buf1Len := fr.getLocal(inst.Operand(1))
-				buf2Ptr := fr.getLocal(inst.Operand(2))
-				buf2Len := fr.getLocal(inst.Operand(3))
-				buf1 := getStringBytes(buf1Ptr, buf1Len.Value())
-				buf2 := getStringBytes(buf2Ptr, buf2Len.Value())
-				result := []byte(string(buf1) + string(buf2))
-				vals := make([]llvm.Value, len(result))
-				for i := range vals {
-					vals[i] = llvm.ConstInt(fr.Mod.Context().Int8Type(), uint64(result[i]), false)
-				}
-				globalType := llvm.ArrayType(fr.Mod.Context().Int8Type(), len(result))
-				globalValue := llvm.ConstArray(fr.Mod.Context().Int8Type(), vals)
-				global := llvm.AddGlobal(fr.Mod, globalType, fr.packagePath+"$stringconcat")
-				global.SetInitializer(globalValue)
-				global.SetLinkage(llvm.InternalLinkage)
-				global.SetGlobalConstant(true)
-				global.SetUnnamedAddr(true)
-				stringType := fr.Mod.GetTypeByName("runtime._string")
-				retPtr := llvm.ConstGEP(global, getLLVMIndices(fr.Mod.Context().Int32Type(), []uint32{0, 0}))
-				retLen := llvm.ConstInt(stringType.StructElementTypes()[1], uint64(len(result)), false)
-				ret := llvm.ConstNull(stringType)
-				ret = llvm.ConstInsertValue(ret, retPtr, []uint32{0})
-				ret = llvm.ConstInsertValue(ret, retLen, []uint32{1})
-				fr.locals[inst] = &LocalValue{fr.Eval, ret}
-			case callee.Name() == "runtime.sliceCopy":
-				elementSize := fr.getLocal(inst.Operand(4)).(*LocalValue).Value().ZExtValue()
-				dstArray := fr.getLocal(inst.Operand(0)).(*LocalValue).stripPointerCasts()
-				srcArray := fr.getLocal(inst.Operand(1)).(*LocalValue).stripPointerCasts()
-				dstLen := fr.getLocal(inst.Operand(2)).(*LocalValue)
-				srcLen := fr.getLocal(inst.Operand(3)).(*LocalValue)
-				if elementSize != 1 && dstArray.Type().ElementType().TypeKind() == llvm.ArrayTypeKind && srcArray.Type().ElementType().TypeKind() == llvm.ArrayTypeKind {
-					// Slice data pointers are created by adding a global array
-					// and getting the address of the first element using a GEP.
-					// However, before the compiler can pass it to
-					// runtime.sliceCopy, it has to perform a bitcast to a *i8,
-					// to make it a unsafe.Pointer. Now, when the IR builder
-					// sees a bitcast of a GEP with zero indices, it will make
-					// a bitcast of the original array instead of the GEP,
-					// which breaks our assumptions.
-					// Re-add this GEP, in the hope that it it is then of the correct type...
-					dstArrayValue, err := dstArray.GetElementPtr([]uint32{0, 0})
-					if err != nil {
-						return nil, nil, fr.errorAt(inst, err)
-					}
-					dstArray = dstArrayValue.(*LocalValue)
-					srcArrayValue, err := srcArray.GetElementPtr([]uint32{0, 0})
-					if err != nil {
-						return nil, nil, fr.errorAt(inst, err)
-					}
-					srcArray = srcArrayValue.(*LocalValue)
-				}
-				if fr.Eval.TargetData.TypeAllocSize(dstArray.Type().ElementType()) != elementSize {
-					return nil, nil, fr.errorAt(inst, errors.New("interp: slice dst element size does not match pointer type"))
-				}
-				if fr.Eval.TargetData.TypeAllocSize(srcArray.Type().ElementType()) != elementSize {
-					return nil, nil, fr.errorAt(inst, errors.New("interp: slice src element size does not match pointer type"))
-				}
-				if dstArray.Type() != srcArray.Type() {
-					return nil, nil, fr.errorAt(inst, errors.New("interp: slice element types don't match"))
-				}
-				length := dstLen.Value().SExtValue()
-				if srcLength := srcLen.Value().SExtValue(); srcLength < length {
-					length = srcLength
-				}
-				if length < 0 {
-					return nil, nil, fr.errorAt(inst, errors.New("interp: trying to copy a slice with negative length?"))
-				}
-				for i := int64(0); i < length; i++ {
-					var err error
-					// *dst = *src
-					dstArray.Store(srcArray.Load())
-					// dst++
-					dstArrayValue, err := dstArray.GetElementPtr([]uint32{1})
-					if err != nil {
-						return nil, nil, fr.errorAt(inst, err)
-					}
-					dstArray = dstArrayValue.(*LocalValue)
-					// src++
-					srcArrayValue, err := srcArray.GetElementPtr([]uint32{1})
-					if err != nil {
-						return nil, nil, fr.errorAt(inst, err)
-					}
-					srcArray = srcArrayValue.(*LocalValue)
-				}
-			case callee.Name() == "runtime.stringToBytes":
-				// convert a string to a []byte
-				bufPtr := fr.getLocal(inst.Operand(0))
-				bufLen := fr.getLocal(inst.Operand(1))
-				result := getStringBytes(bufPtr, bufLen.Value())
-				vals := make([]llvm.Value, len(result))
-				for i := range vals {
-					vals[i] = llvm.ConstInt(fr.Mod.Context().Int8Type(), uint64(result[i]), false)
-				}
-				globalType := llvm.ArrayType(fr.Mod.Context().Int8Type(), len(result))
-				globalValue := llvm.ConstArray(fr.Mod.Context().Int8Type(), vals)
-				global := llvm.AddGlobal(fr.Mod, globalType, fr.packagePath+"$bytes")
-				global.SetInitializer(globalValue)
-				global.SetLinkage(llvm.InternalLinkage)
-				global.SetGlobalConstant(true)
-				global.SetUnnamedAddr(true)
-				sliceType := inst.Type()
-				retPtr := llvm.ConstGEP(global, getLLVMIndices(fr.Mod.Context().Int32Type(), []uint32{0, 0}))
-				retLen := llvm.ConstInt(sliceType.StructElementTypes()[1], uint64(len(result)), false)
-				ret := llvm.ConstNull(sliceType)
-				ret = llvm.ConstInsertValue(ret, retPtr, []uint32{0}) // ptr
-				ret = llvm.ConstInsertValue(ret, retLen, []uint32{1}) // len
-				ret = llvm.ConstInsertValue(ret, retLen, []uint32{2}) // cap
-				fr.locals[inst] = &LocalValue{fr.Eval, ret}
-			case callee.Name() == "runtime.typeAssert":
-				actualTypeInt := fr.getLocal(inst.Operand(0)).(*LocalValue).Underlying
-				assertedType := fr.getLocal(inst.Operand(1)).(*LocalValue).Underlying
-				if actualTypeInt.IsAConstantExpr().IsNil() || actualTypeInt.Opcode() != llvm.PtrToInt {
-					return nil, nil, fr.errorAt(inst, errors.New("interp: expected typecode in runtime.typeAssert to be a ptrtoint"))
-				}
-				actualType := actualTypeInt.Operand(0)
-				if actualType.IsAConstant().IsNil() || assertedType.IsAConstant().IsNil() {
-					return nil, nil, fr.errorAt(inst, errors.New("interp: unimplemented: type assert with non-constant interface value"))
-				}
-				assertOk := uint64(0)
-				if llvm.ConstExtractValue(actualType.Initializer(), []uint32{0}) == assertedType {
-					assertOk = 1
-				}
-				fr.locals[inst] = &LocalValue{fr.Eval, llvm.ConstInt(fr.Mod.Context().Int1Type(), assertOk, false)}
-			case callee.Name() == "runtime.interfaceImplements":
-				typecode := fr.getLocal(inst.Operand(0)).(*LocalValue).Underlying
-				interfaceMethodSet := fr.getLocal(inst.Operand(1)).(*LocalValue).Underlying
-				if typecode.IsAConstantExpr().IsNil() || typecode.Opcode() != llvm.PtrToInt {
-					return nil, nil, fr.errorAt(inst, errors.New("interp: expected typecode to be a ptrtoint"))
-				}
-				typecode = typecode.Operand(0)
-				if interfaceMethodSet.IsAConstantExpr().IsNil() || interfaceMethodSet.Opcode() != llvm.GetElementPtr {
-					return nil, nil, fr.errorAt(inst, errors.New("interp: expected method set in runtime.interfaceImplements to be a constant gep"))
-				}
-				interfaceMethodSet = interfaceMethodSet.Operand(0).Initializer()
-				methodSet := llvm.ConstExtractValue(typecode.Initializer(), []uint32{1})
-				if methodSet.IsAConstantExpr().IsNil() || methodSet.Opcode() != llvm.GetElementPtr {
-					return nil, nil, fr.errorAt(inst, errors.New("interp: expected method set to be a constant gep"))
-				}
-				methodSet = methodSet.Operand(0).Initializer()
-
-				// Make a set of all the methods on the concrete type, for
-				// easier checking in the next step.
-				definedMethods := map[string]struct{}{}
-				for i := 0; i < methodSet.Type().ArrayLength(); i++ {
-					methodInfo := llvm.ConstExtractValue(methodSet, []uint32{uint32(i)})
-					name := llvm.ConstExtractValue(methodInfo, []uint32{0}).Name()
-					definedMethods[name] = struct{}{}
-				}
-				// Check whether all interface methods are also in the list
-				// of defined methods calculated above.
-				implements := uint64(1) // i1 true
-				for i := 0; i < interfaceMethodSet.Type().ArrayLength(); i++ {
-					name := llvm.ConstExtractValue(interfaceMethodSet, []uint32{uint32(i)}).Name()
-					if _, ok := definedMethods[name]; !ok {
-						// There is a method on the interface that is not
-						// implemented by the type.
-						implements = 0 // i1 false
-						break
-					}
-				}
-				fr.locals[inst] = &LocalValue{fr.Eval, llvm.ConstInt(fr.Mod.Context().Int1Type(), implements, false)}
-			case callee.Name() == "runtime.nanotime":
-				fr.locals[inst] = &LocalValue{fr.Eval, llvm.ConstInt(fr.Mod.Context().Int64Type(), 0, false)}
-			case callee.Name() == "llvm.dbg.value":
-				// do nothing
-			case strings.HasPrefix(callee.Name(), "llvm.lifetime."):
-				// do nothing
-			case callee.Name() == "runtime.trackPointer":
-				// do nothing
-			case strings.HasPrefix(callee.Name(), "runtime.print") || callee.Name() == "runtime._panic":
-				// This are all print instructions, which necessarily have side
-				// effects but no results.
-				// TODO: print an error when executing runtime._panic (with the
-				// exact error message it would print at runtime).
-				var params []llvm.Value
-				for i := 0; i < inst.OperandsCount()-1; i++ {
-					operand := fr.getLocal(inst.Operand(i)).Value()
-					fr.markDirty(operand)
-					params = append(params, operand)
-				}
-				// TODO: accurate debug info, including call chain
-				fr.builder.CreateCall(callee, params, inst.Name())
-			case !callee.IsAFunction().IsNil() && callee.IsDeclaration():
-				// external functions
-				var params []llvm.Value
-				for i := 0; i < inst.OperandsCount()-1; i++ {
-					operand := fr.getLocal(inst.Operand(i)).Value()
-					fr.markDirty(operand)
-					params = append(params, operand)
-				}
-				// TODO: accurate debug info, including call chain
-				result := fr.builder.CreateCall(callee, params, inst.Name())
-				if inst.Type().TypeKind() != llvm.VoidTypeKind {
-					fr.markDirty(result)
-					fr.locals[inst] = &LocalValue{fr.Eval, result}
-				}
-			case !callee.IsAFunction().IsNil():
-				// regular function
-				var params []Value
-				dirtyParams := false
-				for i := 0; i < inst.OperandsCount()-1; i++ {
-					local := fr.getLocal(inst.Operand(i))
-					if !local.IsConstant() {
-						dirtyParams = true
-					}
-					params = append(params, local)
-				}
-				var ret Value
-				scanResult, err := fr.hasSideEffects(callee)
-				if err != nil {
-					return nil, nil, err
-				}
-				if scanResult.severity == sideEffectLimited || dirtyParams && scanResult.severity != sideEffectAll {
-					// Side effect is bounded. This means the operation invokes
-					// side effects (like calling an external function) but it
-					// is known at compile time which side effects it invokes.
-					// This means the function can be called at runtime and the
-					// affected globals can be marked dirty at compile time.
-					llvmParams := make([]llvm.Value, len(params))
-					for i, param := range params {
-						llvmParams[i] = param.Value()
-					}
-					result := fr.builder.CreateCall(callee, llvmParams, inst.Name())
-					ret = &LocalValue{fr.Eval, result}
-					// mark all mentioned globals as dirty
-					for global := range scanResult.mentionsGlobals {
-						fr.markDirty(global)
-					}
-				} else {
-					// Side effect is one of:
-					//   * None: no side effects, can be fully interpreted at
-					//     compile time.
-					//   * Unbounded: cannot call at runtime so we'll try to
-					//     interpret anyway and hope for the best.
-					ret, err = fr.function(callee, params, indent+"    ")
-					if err != nil {
-						// Record this function call in the backtrace.
-						err.Traceback = append(err.Traceback, ErrorLine{
-							Pos:  getPosition(inst),
-							Inst: inst,
-						})
-						return nil, nil, err
-					}
-				}
-				if inst.Type().TypeKind() != llvm.VoidTypeKind {
-					fr.locals[inst] = ret
-				}
-			default:
-				// function pointers, etc.
-				return nil, nil, fr.unsupportedInstructionError(inst)
-			}
-		case !inst.IsAExtractValueInst().IsNil():
-			agg := fr.getLocal(inst.Operand(0)).(*LocalValue) // must be constant
-			indices := inst.Indices()
-			if agg.Underlying.IsConstant() {
-				newValue := llvm.ConstExtractValue(agg.Underlying, indices)
-				fr.locals[inst] = fr.getValue(newValue)
-			} else {
-				if len(indices) != 1 {
-					return nil, nil, fr.errorAt(inst, errors.New("interp: cannot handle extractvalue with not exactly 1 index"))
-				}
-				fr.locals[inst] = &LocalValue{fr.Eval, fr.builder.CreateExtractValue(agg.Underlying, int(indices[0]), inst.Name())}
-			}
-		case !inst.IsAInsertValueInst().IsNil():
-			agg := fr.getLocal(inst.Operand(0)).(*LocalValue) // must be constant
-			val := fr.getLocal(inst.Operand(1))
-			indices := inst.Indices()
-			if agg.IsConstant() && val.IsConstant() {
-				newValue := llvm.ConstInsertValue(agg.Underlying, val.Value(), indices)
-				fr.locals[inst] = &LocalValue{fr.Eval, newValue}
-			} else {
-				if len(indices) != 1 {
-					return nil, nil, fr.errorAt(inst, errors.New("interp: cannot handle insertvalue with not exactly 1 index"))
-				}
-				fr.locals[inst] = &LocalValue{fr.Eval, fr.builder.CreateInsertValue(agg.Underlying, val.Value(), int(indices[0]), inst.Name())}
-			}
-		case !inst.IsASelectInst().IsNil():
-			// var result T
-			// if cond {
-			//   result = x
-			// } else {
-			//   result = y
-			// }
-			// return result
-			cond := fr.getLocal(inst.Operand(0)).(*LocalValue).Underlying
-			x := fr.getLocal(inst.Operand(1)).(*LocalValue).Underlying
-			y := fr.getLocal(inst.Operand(2)).(*LocalValue).Underlying
-			fr.locals[inst] = &LocalValue{fr.Eval, fr.builder.CreateSelect(cond, x, y, "")}
-
-		case !inst.IsAReturnInst().IsNil() && inst.OperandsCount() == 0:
-			return nil, nil, nil // ret void
-		case !inst.IsAReturnInst().IsNil() && inst.OperandsCount() == 1:
-			return fr.getLocal(inst.Operand(0)), nil, nil
-		case !inst.IsABranchInst().IsNil() && inst.OperandsCount() == 3:
-			// conditional branch (if/then/else)
-			cond := fr.getLocal(inst.Operand(0)).Value()
-			if cond.Type() != fr.Mod.Context().Int1Type() {
-				return nil, nil, fr.errorAt(inst, errors.New("expected an i1 in a branch instruction"))
-			}
-			thenBB := inst.Operand(1)
-			elseBB := inst.Operand(2)
-			if !cond.IsAInstruction().IsNil() {
-				return nil, nil, fr.errorAt(inst, errors.New("interp: branch on a non-constant"))
-			}
-			if !cond.IsAConstantExpr().IsNil() {
-				// This may happen when the instruction builder could not
-				// const-fold some instructions.
-				return nil, nil, fr.errorAt(inst, errors.New("interp: branch on a non-const-propagated constant expression"))
-			}
-			switch cond {
-			case llvm.ConstInt(fr.Mod.Context().Int1Type(), 0, false): // false
-				return nil, []llvm.Value{thenBB}, nil // then
-			case llvm.ConstInt(fr.Mod.Context().Int1Type(), 1, false): // true
-				return nil, []llvm.Value{elseBB}, nil // else
-			default:
-				return nil, nil, fr.errorAt(inst, errors.New("branch was not true or false"))
-			}
-		case !inst.IsABranchInst().IsNil() && inst.OperandsCount() == 1:
-			// unconditional branch (goto)
-			return nil, []llvm.Value{inst.Operand(0)}, nil
-		case !inst.IsAUnreachableInst().IsNil():
-			// Unreachable was reached (e.g. after a call to panic()).
-			// Report this as an error, as it is not supposed to happen.
-			// This is a sentinel error value.
-			return nil, nil, errUnreachable
-
-		default:
-			return nil, nil, fr.unsupportedInstructionError(inst)
-		}
-	}
-
-	panic("interp: reached end of basic block without terminator")
-}
-
-// Get the Value for an operand, which is a constant value of some sort.
-func (fr *frame) getLocal(v llvm.Value) Value {
-	if ret, ok := fr.locals[v]; ok {
-		return ret
-	} else if value := fr.getValue(v); value != nil {
-		return value
-	} else {
-		// This should not happen under normal circumstances.
-		panic("cannot find value")
-	}
-}
@@ -1,59 +1,77 @@
-// Package interp interprets Go package initializers as much as possible. This
-// avoid running them at runtime, improving code size and making other
-// optimizations possible.
+// Package interp is a partial evaluator of code run at package init time. See
+// the README in this package for details.
 package interp

-// This file provides the overarching Eval object with associated (utility)
-// methods.
-
 import (
+	"fmt"
+	"os"
 	"strings"
+	"time"

 	"tinygo.org/x/go-llvm"
 )

-type Eval struct {
-	Mod             llvm.Module
-	TargetData      llvm.TargetData
-	Debug           bool
-	builder         llvm.Builder
-	dirtyGlobals    map[llvm.Value]struct{}
-	sideEffectFuncs map[llvm.Value]*sideEffectResult // cache of side effect scan results
+// Enable extra checks, which should be disabled by default.
+// This may help track down bugs by adding a few more sanity checks.
+const checks = true
+
+// runner contains all state related to one interp run.
+type runner struct {
+	mod           llvm.Module
+	targetData    llvm.TargetData
+	builder       llvm.Builder
+	pointerSize   uint32                   // cached pointer size from the TargetData
+	i8ptrType     llvm.Type                // often used type so created in advance
+	maxAlign      int                      // maximum alignment of an object, alignment of runtime.alloc() result
+	debug         bool                     // log debug messages
+	pkgName       string                   // package name of the currently executing package
+	functionCache map[llvm.Value]*function // cache of compiled functions
+	objects       []object                 // slice of objects in memory
+	globals       map[llvm.Value]int       // map from global to index in objects slice
+	start         time.Time
+	callsExecuted uint64
 }

-// evalPackage encapsulates the Eval type for just a single package. The Eval
-// type keeps state across the whole program, the evalPackage type keeps extra
-// state for the currently interpreted package.
-type evalPackage struct {
-	*Eval
-	packagePath string
-}
-
-// Run evaluates the function with the given name and then eliminates all
-// callers.
+// Run evaluates runtime.initAll function as much as possible at compile time.
+// Set debug to true if it should print output while running.
 func Run(mod llvm.Module, debug bool) error {
-	if debug {
-		println("\ncompile-time evaluation:")
+	r := runner{
+		mod:           mod,
+		targetData:    llvm.NewTargetData(mod.DataLayout()),
+		debug:         debug,
+		functionCache: make(map[llvm.Value]*function),
+		objects:       []object{{}},
+		globals:       make(map[llvm.Value]int),
+		start:         time.Now(),
 	}
+	r.pointerSize = uint32(r.targetData.PointerSize())
+	r.i8ptrType = llvm.PointerType(mod.Context().Int8Type(), 0)
+	r.maxAlign = r.targetData.PrefTypeAlignment(r.i8ptrType) // assume pointers are maximally aligned (this is not always the case)

-	name := "runtime.initAll"
-	e := &Eval{
-		Mod:          mod,
-		TargetData:   llvm.NewTargetData(mod.DataLayout()),
-		Debug:        debug,
-		dirtyGlobals: map[llvm.Value]struct{}{},
-	}
-	e.builder = mod.Context().NewBuilder()
-
-	initAll := mod.NamedFunction(name)
+	initAll := mod.NamedFunction("runtime.initAll")
 	bb := initAll.EntryBasicBlock()
+
+	// Create a builder, to insert instructions that could not be evaluated at
+	// compile time.
+	r.builder = mod.Context().NewBuilder()
+	defer r.builder.Dispose()
+
 	// Create a dummy alloca in the entry block that we can set the insert point
 	// to. This is necessary because otherwise we might be removing the
 	// instruction (init call) that we are removing after successful
 	// interpretation.
-	e.builder.SetInsertPointBefore(bb.FirstInstruction())
-	dummy := e.builder.CreateAlloca(e.Mod.Context().Int8Type(), "dummy")
-	e.builder.SetInsertPointBefore(dummy)
+	r.builder.SetInsertPointBefore(bb.FirstInstruction())
+	dummy := r.builder.CreateAlloca(r.mod.Context().Int8Type(), "dummy")
+	r.builder.SetInsertPointBefore(dummy)
+	defer dummy.EraseFromParentAsInstruction()
+
+	// Get a list if init calls. A runtime.initAll might look something like this:
+	// func initAll() {
+	//     unsafe.init()
+	//     machine.init()
+	//     runtime.init()
+	// }
+	// This function gets a list of these call instructions.
 	var initCalls []llvm.Value
 	for inst := bb.FirstInstruction(); !inst.IsNil(); inst = llvm.NextInstruction(inst) {
 		if inst == dummy {
@@ -63,99 +81,67 @@ func Run(mod llvm.Module, debug bool) error {
 			break // ret void
 		}
 		if inst.IsACallInst().IsNil() || inst.CalledValue().IsAFunction().IsNil() {
-			return errorAt(inst, "interp: expected all instructions in "+name+" to be direct calls")
+			return errorAt(inst, "interp: expected all instructions in "+initAll.Name()+" to be direct calls")
 		}
 		initCalls = append(initCalls, inst)
 	}

-	// Do this in a separate step to avoid corrupting the iterator above.
-	undefPtr := llvm.Undef(llvm.PointerType(mod.Context().Int8Type(), 0))
+	// Run initializers for each package. Once the package initializer is
+	// finished, the call to the package initializer can be removed.
 	for _, call := range initCalls {
 		initName := call.CalledValue().Name()
 		if !strings.HasSuffix(initName, ".init") {
-			return errorAt(call, "interp: expected all instructions in "+name+" to be *.init() calls")
+			return errorAt(call, "interp: expected all instructions in "+initAll.Name()+" to be *.init() calls")
 		}
-		pkgName := initName[:len(initName)-5]
+		r.pkgName = initName[:len(initName)-len(".init")]
 		fn := call.CalledValue()
+		if r.debug {
+			fmt.Fprintln(os.Stderr, "call:", fn.Name())
+		}
+		_, mem, callErr := r.run(r.getFunction(fn), nil, nil, "    ")
+		if callErr != nil {
+			if isRecoverableError(callErr.Err) {
+				if r.debug {
+					fmt.Fprintln(os.Stderr, "not interpreting", r.pkgName, "because of error:", callErr.Error())
+				}
+				mem.revert()
+				continue
+			}
+			return callErr
+		}
 		call.EraseFromParentAsInstruction()
-		evalPkg := evalPackage{
-			Eval:        e,
-			packagePath: pkgName,
+		for index, obj := range mem.objects {
+			r.objects[index] = obj
 		}
-		_, err := evalPkg.function(fn, []Value{&LocalValue{e, undefPtr}, &LocalValue{e, undefPtr}}, "")
-		if err == errUnreachable {
-			break
+	}
+	r.pkgName = ""
+
+	// Update all global variables in the LLVM module.
+	mem := memoryView{r: &r}
+	for _, obj := range r.objects {
+		if obj.llvmGlobal.IsNil() {
+			continue
 		}
-		if err != nil {
-			return err
+		if obj.buffer == nil {
+			continue
 		}
+		initializer := obj.buffer.toLLVMValue(obj.llvmGlobal.Type().ElementType(), &mem)
+		if checks && initializer.Type() != obj.llvmGlobal.Type().ElementType() {
+			panic("initializer type mismatch")
+		}
+		obj.llvmGlobal.SetInitializer(initializer)
 	}

 	return nil
 }

-// function interprets the given function. The params are the function params
-// and the indent is the string indentation to use when dumping all interpreted
-// instructions.
-func (e *evalPackage) function(fn llvm.Value, params []Value, indent string) (Value, *Error) {
-	fr := frame{
-		evalPackage: e,
-		fn:          fn,
-		locals:      make(map[llvm.Value]Value),
-	}
-	for i, param := range fn.Params() {
-		fr.locals[param] = params[i]
-	}
-
-	bb := fn.EntryBasicBlock()
-	var lastBB llvm.BasicBlock
-	for {
-		retval, outgoing, err := fr.evalBasicBlock(bb, lastBB, indent)
-		if outgoing == nil {
-			// returned something (a value or void, or an error)
-			return retval, err
-		}
-		if len(outgoing) > 1 {
-			panic("unimplemented: multiple outgoing blocks")
-		}
-		next := outgoing[0]
-		if next.IsABasicBlock().IsNil() {
-			panic("did not switch to a basic block")
-		}
-		lastBB = bb
-		bb = next.AsBasicBlock()
-	}
-}
-
-// getValue determines what kind of LLVM value it gets and returns the
-// appropriate Value type.
-func (e *Eval) getValue(v llvm.Value) Value {
-	return &LocalValue{e, v}
-}
-
-// markDirty marks the passed-in LLVM value dirty, recursively. For example,
-// when it encounters a constant GEP on a global, it marks the global dirty.
-func (e *Eval) markDirty(v llvm.Value) {
-	if !v.IsAGlobalVariable().IsNil() {
-		if v.IsGlobalConstant() {
-			return
-		}
-		if _, ok := e.dirtyGlobals[v]; !ok {
-			e.dirtyGlobals[v] = struct{}{}
-			e.sideEffectFuncs = nil // re-calculate all side effects
-		}
-	} else if v.IsConstant() {
-		if v.OperandsCount() >= 2 && !v.Operand(0).IsAGlobalVariable().IsNil() {
-			// looks like a constant getelementptr of a global.
-			// TODO: find a way to make sure it really is: v.Opcode() returns 0.
-			e.markDirty(v.Operand(0))
-			return
-		}
-		return // nothing to mark
-	} else if !v.IsAGetElementPtrInst().IsNil() {
-		panic("interp: todo: GEP")
-	} else {
-		// Not constant and not a global or GEP so doesn't have to be marked
-		// non-constant.
+// getFunction returns the compiled version of the given LLVM function. It
+// compiles the function if necessary and caches the result.
+func (r *runner) getFunction(llvmFn llvm.Value) *function {
+	if fn, ok := r.functionCache[llvmFn]; ok {
+		return fn
 	}
+	fn := r.compileFunction(llvmFn)
+	r.functionCache[llvmFn] = fn
+	return fn
 }
@@ -3,6 +3,7 @@ package interp
 import (
 	"io/ioutil"
 	"os"
+	"regexp"
 	"strings"
 	"testing"

@@ -41,9 +42,29 @@ func runTest(t *testing.T, pathPrefix string) {
 	// Perform the transform.
 	err = Run(mod, false)
 	if err != nil {
+		if err, match := err.(*Error); match {
+			println(err.Error())
+			if !err.Inst.IsNil() {
+				err.Inst.Dump()
+				println()
+			}
+			if len(err.Traceback) > 0 {
+				println("\ntraceback:")
+				for _, line := range err.Traceback {
+					println(line.Pos.String() + ":")
+					line.Inst.Dump()
+					println()
+				}
+			}
+		}
 		t.Fatal(err)
 	}

+	// To be sure, verify that the module is still valid.
+	if llvm.VerifyModule(mod, llvm.PrintMessageAction) != nil {
+		t.FailNow()
+	}
+
 	// Run some cleanup passes to get easy-to-read outputs.
 	pm := llvm.NewPassManager()
 	defer pm.Dispose()
@@ -66,6 +87,8 @@ 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.).
@@ -75,8 +98,18 @@ func fuzzyEqualIR(s1, s2 string) bool {
 	if len(lines1) != len(lines2) {
 		return false
 	}
-	for i, line := range lines1 {
-		if line != lines2[i] {
+	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
 		}
 	}
@@ -0,0 +1,932 @@
+package interp
+
+import (
+	"errors"
+	"fmt"
+	"math"
+	"os"
+	"strings"
+	"time"
+
+	"tinygo.org/x/go-llvm"
+)
+
+func (r *runner) run(fn *function, params []value, parentMem *memoryView, indent string) (value, memoryView, *Error) {
+	mem := memoryView{r: r, parent: parentMem}
+	locals := make([]value, len(fn.locals))
+	r.callsExecuted++
+
+	if time.Since(r.start) > time.Minute {
+		// Running for more than a minute. This should never happen.
+		return nil, mem, r.errorAt(fn.blocks[0].instructions[0], fmt.Errorf("interp: running for more than a minute, timing out (executed calls: %d)", r.callsExecuted))
+	}
+
+	// Parameters are considered a kind of local values.
+	for i, param := range params {
+		locals[i] = param
+	}
+
+	// Start with the first basic block and the first instruction.
+	// Branch instructions may modify both bb and instIndex when branching.
+	bb := fn.blocks[0]
+	currentBB := 0
+	lastBB := -1 // last basic block is undefined, only defined after a branch
+	var operands []value
+	for instIndex := 0; instIndex < len(bb.instructions); instIndex++ {
+		inst := bb.instructions[instIndex]
+		operands = operands[:0]
+		isRuntimeInst := false
+		if inst.opcode != llvm.PHI {
+			for _, v := range inst.operands {
+				if v, ok := v.(localValue); ok {
+					if localVal := locals[fn.locals[v.value]]; localVal == nil {
+						return nil, mem, r.errorAt(inst, errors.New("interp: local not defined"))
+					} else {
+						operands = append(operands, localVal)
+						if _, ok := localVal.(localValue); ok {
+							isRuntimeInst = true
+						}
+						continue
+					}
+				}
+				operands = append(operands, v)
+			}
+		}
+		if isRuntimeInst {
+			err := r.runAtRuntime(fn, inst, locals, &mem, indent)
+			if err != nil {
+				return nil, mem, err
+			}
+			continue
+		}
+		switch inst.opcode {
+		case llvm.Ret:
+			if len(operands) != 0 {
+				if r.debug {
+					fmt.Fprintln(os.Stderr, indent+"ret", operands[0])
+				}
+				// Return instruction has a value to return.
+				return operands[0], mem, nil
+			}
+			if r.debug {
+				fmt.Fprintln(os.Stderr, indent+"ret")
+			}
+			// Return instruction doesn't return anything, it's just 'ret void'.
+			return nil, mem, nil
+		case llvm.Br:
+			switch len(operands) {
+			case 1:
+				// Unconditional branch: [nextBB]
+				lastBB = currentBB
+				currentBB = int(operands[0].(literalValue).value.(uint32))
+				bb = fn.blocks[currentBB]
+				instIndex = -1 // start at 0 the next cycle
+				if r.debug {
+					fmt.Fprintln(os.Stderr, indent+"br", operands, "->", currentBB)
+				}
+			case 3:
+				// Conditional branch: [cond, thenBB, elseBB]
+				lastBB = currentBB
+				switch operands[0].Uint() {
+				case 1: // true -> thenBB
+					currentBB = int(operands[1].(literalValue).value.(uint32))
+				case 0: // false -> elseBB
+					currentBB = int(operands[2].(literalValue).value.(uint32))
+				default:
+					panic("bool should be 0 or 1")
+				}
+				if r.debug {
+					fmt.Fprintln(os.Stderr, indent+"br", operands, "->", currentBB)
+				}
+				bb = fn.blocks[currentBB]
+				instIndex = -1 // start at 0 the next cycle
+			default:
+				panic("unknown operands length")
+			}
+			break // continue with next block
+		case llvm.PHI:
+			var result value
+			for i := 0; i < len(inst.operands); i += 2 {
+				if int(inst.operands[i].(literalValue).value.(uint32)) == lastBB {
+					incoming := inst.operands[i+1]
+					if local, ok := incoming.(localValue); ok {
+						result = locals[fn.locals[local.value]]
+					} else {
+						result = incoming
+					}
+					break
+				}
+			}
+			if r.debug {
+				fmt.Fprintln(os.Stderr, indent+"phi", inst.operands, "->", result)
+			}
+			if result == nil {
+				panic("could not find PHI input")
+			}
+			locals[inst.localIndex] = result
+		case llvm.Select:
+			// Select is much like a ternary operator: it picks a result from
+			// the second and third operand based on the boolean first operand.
+			var result value
+			switch operands[0].Uint() {
+			case 1:
+				result = operands[1]
+			case 0:
+				result = operands[2]
+			default:
+				panic("boolean must be 0 or 1")
+			}
+			locals[inst.localIndex] = result
+			if r.debug {
+				fmt.Fprintln(os.Stderr, indent+"select", operands, "->", result)
+			}
+		case llvm.Call:
+			// A call instruction can either be a regular call or a runtime intrinsic.
+			fnPtr, err := operands[0].asPointer(r)
+			if err != nil {
+				return nil, mem, r.errorAt(inst, err)
+			}
+			callFn := r.getFunction(fnPtr.llvmValue(&mem))
+			switch {
+			case callFn.name == "runtime.trackPointer":
+				// Allocas and such are created as globals, so don't need a
+				// runtime.trackPointer.
+				// Unless the object is allocated at runtime for example, in
+				// which case this call won't even get to this point but will
+				// already be emitted in initAll.
+				continue
+			case callFn.name == "(reflect.Type).Elem" || strings.HasPrefix(callFn.name, "runtime.print") || callFn.name == "runtime._panic" || callFn.name == "runtime.hashmapGet":
+				// These functions should be run at runtime. Specifically:
+				//   * (reflect.Type).Elem is a special function. It should
+				//     eventually be interpreted, but fall back to a runtime call
+				//     for now.
+				//   * Print and panic functions are best emitted directly without
+				//     interpreting them, otherwise we get a ton of putchar (etc.)
+				//     calls.
+				//   * runtime.hashmapGet tries to access the map value directly.
+				//     This is not possible as the map value is treated as a special
+				//     kind of object in this package.
+				err := r.runAtRuntime(fn, inst, locals, &mem, indent)
+				if err != nil {
+					return nil, mem, err
+				}
+			case callFn.name == "runtime.nanotime" && r.pkgName == "time":
+				// The time package contains a call to runtime.nanotime.
+				// This appears to be to work around a limitation in Windows
+				// Server 2008:
+				//   > Monotonic times are reported as offsets from startNano.
+				//   > We initialize startNano to runtimeNano() - 1 so that on systems where
+				//   > monotonic time resolution is fairly low (e.g. Windows 2008
+				//   > which appears to have a default resolution of 15ms),
+				//   > we avoid ever reporting a monotonic time of 0.
+				//   > (Callers may want to use 0 as "time not set".)
+				// Simply let runtime.nanotime return 0 in this case, which
+				// should be fine and avoids a call to runtime.nanotime. It
+				// means that monotonic time in the time package is counted from
+				// time.Time{}.Sub(1), which should be fine.
+				locals[inst.localIndex] = literalValue{uint64(0)}
+			case callFn.name == "runtime.alloc":
+				// Allocate heap memory. At compile time, this is instead done
+				// by creating a global variable.
+
+				// Get the requested memory size to be allocated.
+				size := operands[1].Uint()
+
+				// Create the object.
+				alloc := object{
+					globalName: r.pkgName + "$alloc",
+					buffer:     newRawValue(uint32(size)),
+					size:       uint32(size),
+				}
+				index := len(r.objects)
+				r.objects = append(r.objects, alloc)
+
+				// And create a pointer to this object, for working with it (so
+				// that stores to it copy it, etc).
+				ptr := newPointerValue(r, index, 0)
+				if r.debug {
+					fmt.Fprintln(os.Stderr, indent+"runtime.alloc:", size, "->", ptr)
+				}
+				locals[inst.localIndex] = ptr
+			case callFn.name == "runtime.sliceCopy":
+				// sliceCopy implements the built-in copy function for slices.
+				// It is implemented here so that it can be used even if the
+				// runtime implementation is not available. Doing it this way
+				// may also be faster.
+				// Code:
+				// func sliceCopy(dst, src unsafe.Pointer, dstLen, srcLen uintptr, elemSize uintptr) int {
+				//     n := srcLen
+				//     if n > dstLen {
+				//         n = dstLen
+				//     }
+				//     memmove(dst, src, n*elemSize)
+				//     return int(n)
+				// }
+				dstLen := operands[3].Uint()
+				srcLen := operands[4].Uint()
+				elemSize := operands[5].Uint()
+				n := srcLen
+				if n > dstLen {
+					n = dstLen
+				}
+				if r.debug {
+					fmt.Fprintln(os.Stderr, indent+"copy:", operands[1], operands[2], n)
+				}
+				if n != 0 {
+					// Only try to copy bytes when there are any bytes to copy.
+					// This is not just an optimization. If one of the slices
+					// (or both) are nil, the asPointer method call will fail
+					// even though copying a nil slice is allowed.
+					dst, err := operands[1].asPointer(r)
+					if err != nil {
+						return nil, mem, r.errorAt(inst, err)
+					}
+					src, err := operands[2].asPointer(r)
+					if err != nil {
+						return nil, mem, r.errorAt(inst, err)
+					}
+					nBytes := uint32(n * elemSize)
+					dstObj := mem.getWritable(dst.index())
+					dstBuf := dstObj.buffer.asRawValue(r)
+					srcBuf := mem.get(src.index()).buffer.asRawValue(r)
+					copy(dstBuf.buf[dst.offset():dst.offset()+nBytes], srcBuf.buf[src.offset():])
+					dstObj.buffer = dstBuf
+					mem.put(dst.index(), dstObj)
+				}
+				switch inst.llvmInst.Type().IntTypeWidth() {
+				case 16:
+					locals[inst.localIndex] = literalValue{uint16(n)}
+				case 32:
+					locals[inst.localIndex] = literalValue{uint32(n)}
+				case 64:
+					locals[inst.localIndex] = literalValue{uint64(n)}
+				default:
+					panic("unknown integer type width")
+				}
+			case strings.HasPrefix(callFn.name, "llvm.memcpy.p0i8.p0i8.") || strings.HasPrefix(callFn.name, "llvm.memmove.p0i8.p0i8."):
+				// Copy a block of memory from one pointer to another.
+				dst, err := operands[1].asPointer(r)
+				if err != nil {
+					return nil, mem, r.errorAt(inst, err)
+				}
+				src, err := operands[2].asPointer(r)
+				if err != nil {
+					return nil, mem, r.errorAt(inst, err)
+				}
+				nBytes := uint32(operands[3].Uint())
+				dstObj := mem.getWritable(dst.index())
+				dstBuf := dstObj.buffer.asRawValue(r)
+				srcBuf := mem.get(src.index()).buffer.asRawValue(r)
+				copy(dstBuf.buf[dst.offset():dst.offset()+nBytes], srcBuf.buf[src.offset():])
+				dstObj.buffer = dstBuf
+				mem.put(dst.index(), dstObj)
+			case callFn.name == "runtime.typeAssert":
+				// This function must be implemented manually as it is normally
+				// implemented by the interface lowering pass.
+				if r.debug {
+					fmt.Fprintln(os.Stderr, indent+"typeassert:", operands[1:])
+				}
+				typeInInterfacePtr, err := operands[1].asPointer(r)
+				if err != nil {
+					return nil, mem, r.errorAt(inst, err)
+				}
+				actualType, err := mem.load(typeInInterfacePtr, r.pointerSize).asPointer(r)
+				if err != nil {
+					return nil, mem, r.errorAt(inst, err)
+				}
+				assertedType, err := operands[2].asPointer(r)
+				if err != nil {
+					return nil, mem, r.errorAt(inst, err)
+				}
+				result := assertedType.asRawValue(r).equal(actualType.asRawValue(r))
+				if result {
+					locals[inst.localIndex] = literalValue{uint8(1)}
+				} else {
+					locals[inst.localIndex] = literalValue{uint8(0)}
+				}
+			case callFn.name == "runtime.interfaceImplements":
+				if r.debug {
+					fmt.Fprintln(os.Stderr, indent+"interface assert:", operands[1:])
+				}
+
+				// Load various values for the interface implements check below.
+				typeInInterfacePtr, err := operands[1].asPointer(r)
+				if err != nil {
+					return nil, mem, r.errorAt(inst, err)
+				}
+				methodSetPtr, err := mem.load(typeInInterfacePtr.addOffset(r.pointerSize), r.pointerSize).asPointer(r)
+				if err != nil {
+					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()
+
+				// Make a set of all the methods on the concrete type, for
+				// easier checking in the next step.
+				concreteTypeMethods := map[string]struct{}{}
+				for i := 0; i < methodSet.Type().ArrayLength(); i++ {
+					methodInfo := llvm.ConstExtractValue(methodSet, []uint32{uint32(i)})
+					name := llvm.ConstExtractValue(methodInfo, []uint32{0}).Name()
+					concreteTypeMethods[name] = struct{}{}
+				}
+
+				// Check whether all interface methods are also in the list
+				// 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()
+					if _, ok := concreteTypeMethods[name]; !ok {
+						// There is a method on the interface that is not
+						// implemented by the type. The assertion will fail.
+						assertOk = 0 // i1 false
+						break
+					}
+				}
+				// If assertOk is still 1, the assertion succeeded.
+				locals[inst.localIndex] = literalValue{assertOk}
+			case callFn.name == "runtime.hashmapMake":
+				// Create a new map.
+				hashmapPointerType := inst.llvmInst.Type()
+				keySize := uint32(operands[1].Uint())
+				valueSize := uint32(operands[2].Uint())
+				m := newMapValue(r, hashmapPointerType, keySize, valueSize)
+				alloc := object{
+					llvmType:   hashmapPointerType,
+					globalName: r.pkgName + "$map",
+					buffer:     m,
+					size:       m.len(r),
+				}
+				index := len(r.objects)
+				r.objects = append(r.objects, alloc)
+
+				// Create a pointer to this map. Maps are reference types, so
+				// are implemented as pointers.
+				ptr := newPointerValue(r, index, 0)
+				if r.debug {
+					fmt.Fprintln(os.Stderr, indent+"runtime.hashmapMake:", keySize, valueSize, "->", ptr)
+				}
+				locals[inst.localIndex] = ptr
+			case callFn.name == "runtime.hashmapBinarySet":
+				// Do a mapassign operation with a binary key (that is, without
+				// a string key).
+				if r.debug {
+					fmt.Fprintln(os.Stderr, indent+"runtime.hashmapBinarySet:", operands[1:])
+				}
+				mapPtr, err := operands[1].asPointer(r)
+				if err != nil {
+					return nil, mem, r.errorAt(inst, err)
+				}
+				m := mem.getWritable(mapPtr.index()).buffer.(*mapValue)
+				keyPtr, err := operands[2].asPointer(r)
+				if err != nil {
+					return nil, mem, r.errorAt(inst, err)
+				}
+				valuePtr, err := operands[3].asPointer(r)
+				if err != nil {
+					return nil, mem, r.errorAt(inst, err)
+				}
+				err = m.putBinary(&mem, keyPtr, valuePtr)
+				if err != nil {
+					return nil, mem, r.errorAt(inst, err)
+				}
+			case callFn.name == "runtime.hashmapStringSet":
+				// Do a mapassign operation with a string key.
+				if r.debug {
+					fmt.Fprintln(os.Stderr, indent+"runtime.hashmapBinarySet:", operands[1:])
+				}
+				mapPtr, err := operands[1].asPointer(r)
+				if err != nil {
+					return nil, mem, r.errorAt(inst, err)
+				}
+				m := mem.getWritable(mapPtr.index()).buffer.(*mapValue)
+				stringPtr, err := operands[2].asPointer(r)
+				if err != nil {
+					return nil, mem, r.errorAt(inst, err)
+				}
+				stringLen := operands[3].Uint()
+				valuePtr, err := operands[4].asPointer(r)
+				if err != nil {
+					return nil, mem, r.errorAt(inst, err)
+				}
+				err = m.putString(&mem, stringPtr, stringLen, valuePtr)
+				if err != nil {
+					return nil, mem, r.errorAt(inst, err)
+				}
+			default:
+				if len(callFn.blocks) == 0 {
+					// Call to a function declaration without a definition
+					// available.
+					err := r.runAtRuntime(fn, inst, locals, &mem, indent)
+					if err != nil {
+						return nil, mem, err
+					}
+					continue
+				}
+				// Call a function with a definition available. Run it as usual,
+				// possibly trying to recover from it if it failed to execute.
+				if r.debug {
+					argStrings := make([]string, len(operands)-1)
+					for i := range argStrings {
+						argStrings[i] = operands[i+1].String()
+					}
+					fmt.Fprintln(os.Stderr, indent+"call:", callFn.name+"("+strings.Join(argStrings, ", ")+")")
+				}
+				retval, callMem, callErr := r.run(callFn, operands[1:], &mem, indent+"    ")
+				if callErr != nil {
+					if isRecoverableError(callErr.Err) {
+						// This error can be recovered by doing the call at
+						// runtime instead of at compile time. But we need to
+						// revert any changes made by the call first.
+						if r.debug {
+							fmt.Fprintln(os.Stderr, indent+"!! revert because of error:", callErr.Err)
+						}
+						callMem.revert()
+						err := r.runAtRuntime(fn, inst, locals, &mem, indent)
+						if err != nil {
+							return nil, mem, err
+						}
+						continue
+					}
+					// Add to the traceback, so that error handling code can see
+					// how this function got called.
+					callErr.Traceback = append(callErr.Traceback, ErrorLine{
+						Pos:  getPosition(inst.llvmInst),
+						Inst: inst.llvmInst,
+					})
+					return nil, mem, callErr
+				}
+				locals[inst.localIndex] = retval
+				mem.extend(callMem)
+			}
+		case llvm.Load:
+			// Load instruction, loading some data from the topmost memory view.
+			ptr, err := operands[0].asPointer(r)
+			if err != nil {
+				return nil, mem, r.errorAt(inst, err)
+			}
+			size := operands[1].(literalValue).value.(uint64)
+			if mem.hasExternalStore(ptr) {
+				// If there could be an external store (for example, because a
+				// pointer to the object was passed to a function that could not
+				// be interpreted at compile time) then the load must be done at
+				// runtime.
+				err := r.runAtRuntime(fn, inst, locals, &mem, indent)
+				if err != nil {
+					return nil, mem, err
+				}
+				continue
+			}
+			result := mem.load(ptr, uint32(size))
+			if r.debug {
+				fmt.Fprintln(os.Stderr, indent+"load:", ptr, "->", result)
+			}
+			locals[inst.localIndex] = result
+		case llvm.Store:
+			// Store instruction. Create a new object in the memory view and
+			// store to that, to make it possible to roll back this store.
+			ptr, err := operands[1].asPointer(r)
+			if err != nil {
+				return nil, mem, r.errorAt(inst, err)
+			}
+			if mem.hasExternalLoadOrStore(ptr) {
+				err := r.runAtRuntime(fn, inst, locals, &mem, indent)
+				if err != nil {
+					return nil, mem, err
+				}
+				continue
+			}
+			val := operands[0]
+			if r.debug {
+				fmt.Fprintln(os.Stderr, indent+"store:", val, ptr)
+			}
+			mem.store(val, ptr)
+		case llvm.Alloca:
+			// Alloca normally allocates some stack memory. In the interpreter,
+			// it allocates a global instead.
+			// This can likely be optimized, as all it really needs is an alloca
+			// in the initAll function and creating a global is wasteful for
+			// this purpose.
+
+			// Create the new object.
+			size := operands[0].(literalValue).value.(uint64)
+			alloca := object{
+				llvmType:   inst.llvmInst.Type(),
+				globalName: r.pkgName + "$alloca",
+				buffer:     newRawValue(uint32(size)),
+				size:       uint32(size),
+			}
+			index := len(r.objects)
+			r.objects = append(r.objects, alloca)
+
+			// Create a pointer to this object (an alloca produces a pointer).
+			ptr := newPointerValue(r, index, 0)
+			if r.debug {
+				fmt.Fprintln(os.Stderr, indent+"alloca:", operands, "->", ptr)
+			}
+			locals[inst.localIndex] = ptr
+		case llvm.GetElementPtr:
+			// 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
+				}
+			}
+			ptr, err := operands[0].asPointer(r)
+			if err != nil {
+				if err != errIntegerAsPointer {
+					return nil, mem, r.errorAt(inst, err)
+				}
+				// GEP on fixed pointer value (for example, memory-mapped I/O).
+				ptrValue := operands[0].Uint() + offset
+				switch operands[0].len(r) {
+				case 8:
+					locals[inst.localIndex] = literalValue{uint64(ptrValue)}
+				case 4:
+					locals[inst.localIndex] = literalValue{uint32(ptrValue)}
+				case 2:
+					locals[inst.localIndex] = literalValue{uint16(ptrValue)}
+				default:
+					panic("pointer operand is not of a known pointer size")
+				}
+				continue
+			}
+			ptr = ptr.addOffset(uint32(offset))
+			locals[inst.localIndex] = ptr
+			if r.debug {
+				fmt.Fprintln(os.Stderr, indent+"gep:", operands, "->", ptr)
+			}
+		case llvm.BitCast, llvm.IntToPtr, llvm.PtrToInt:
+			// Various bitcast-like instructions that all keep the same bits
+			// while changing the LLVM type.
+			// Because interp doesn't preserve the type, these operations are
+			// identity operations.
+			if r.debug {
+				fmt.Fprintln(os.Stderr, indent+instructionNameMap[inst.opcode]+":", operands[0])
+			}
+			locals[inst.localIndex] = operands[0]
+		case llvm.ExtractValue:
+			agg := operands[0].asRawValue(r)
+			offset := operands[1].(literalValue).value.(uint64)
+			size := operands[2].(literalValue).value.(uint64)
+			elt := rawValue{
+				buf: agg.buf[offset : offset+size],
+			}
+			if r.debug {
+				fmt.Fprintln(os.Stderr, indent+"extractvalue:", operands, "->", elt)
+			}
+			locals[inst.localIndex] = elt
+		case llvm.InsertValue:
+			agg := operands[0].asRawValue(r)
+			elt := operands[1].asRawValue(r)
+			offset := int(operands[2].(literalValue).value.(uint64))
+			newagg := newRawValue(uint32(len(agg.buf)))
+			copy(newagg.buf, agg.buf)
+			copy(newagg.buf[offset:], elt.buf)
+			if r.debug {
+				fmt.Fprintln(os.Stderr, indent+"insertvalue:", operands, "->", newagg)
+			}
+			locals[inst.localIndex] = newagg
+		case llvm.ICmp:
+			predicate := llvm.IntPredicate(operands[2].(literalValue).value.(uint8))
+			var result bool
+			lhs := operands[0]
+			rhs := operands[1]
+			switch predicate {
+			case llvm.IntEQ, llvm.IntNE:
+				lhsPointer, lhsErr := lhs.asPointer(r)
+				rhsPointer, rhsErr := rhs.asPointer(r)
+				if (lhsErr == nil) != (rhsErr == nil) {
+					// Fast path: only one is a pointer, so they can't be equal.
+					result = false
+				} else if lhsErr == nil {
+					// Both must be nil, so both are pointers.
+					// Compare them directly.
+					result = lhsPointer.equal(rhsPointer)
+				} else {
+					// Fall back to generic comparison.
+					result = lhs.asRawValue(r).equal(rhs.asRawValue(r))
+				}
+				if predicate == llvm.IntNE {
+					result = !result
+				}
+			case llvm.IntUGT:
+				result = lhs.Uint() > rhs.Uint()
+			case llvm.IntUGE:
+				result = lhs.Uint() >= rhs.Uint()
+			case llvm.IntULT:
+				result = lhs.Uint() < rhs.Uint()
+			case llvm.IntULE:
+				result = lhs.Uint() <= rhs.Uint()
+			case llvm.IntSGT:
+				result = lhs.Int() > rhs.Int()
+			case llvm.IntSGE:
+				result = lhs.Int() >= rhs.Int()
+			case llvm.IntSLT:
+				result = lhs.Int() < rhs.Int()
+			case llvm.IntSLE:
+				result = lhs.Int() <= rhs.Int()
+			default:
+				return nil, mem, r.errorAt(inst, errors.New("interp: unsupported icmp"))
+			}
+			if result {
+				locals[inst.localIndex] = literalValue{uint8(1)}
+			} else {
+				locals[inst.localIndex] = literalValue{uint8(0)}
+			}
+			if r.debug {
+				fmt.Fprintln(os.Stderr, indent+"icmp:", operands[0], intPredicateString(predicate), operands[1], "->", result)
+			}
+		case llvm.FCmp:
+			predicate := llvm.FloatPredicate(operands[2].(literalValue).value.(uint8))
+			var result bool
+			var lhs, rhs float64
+			switch operands[0].len(r) {
+			case 8:
+				lhs = math.Float64frombits(operands[0].Uint())
+				rhs = math.Float64frombits(operands[1].Uint())
+			case 4:
+				lhs = float64(math.Float32frombits(uint32(operands[0].Uint())))
+				rhs = float64(math.Float32frombits(uint32(operands[1].Uint())))
+			default:
+				panic("unknown float type")
+			}
+			switch predicate {
+			case llvm.FloatOEQ:
+				result = lhs == rhs
+			case llvm.FloatUNE:
+				result = lhs != rhs
+			case llvm.FloatOGT:
+				result = lhs > rhs
+			case llvm.FloatOGE:
+				result = lhs >= rhs
+			case llvm.FloatOLT:
+				result = lhs < rhs
+			case llvm.FloatOLE:
+				result = lhs <= rhs
+			default:
+				return nil, mem, r.errorAt(inst, errors.New("interp: unsupported fcmp"))
+			}
+			if result {
+				locals[inst.localIndex] = literalValue{uint8(1)}
+			} else {
+				locals[inst.localIndex] = literalValue{uint8(0)}
+			}
+			if r.debug {
+				fmt.Fprintln(os.Stderr, indent+"fcmp:", operands[0], predicate, operands[1], "->", result)
+			}
+		case llvm.Add, llvm.Sub, llvm.Mul, llvm.UDiv, llvm.SDiv, llvm.URem, llvm.SRem, llvm.Shl, llvm.LShr, llvm.AShr, llvm.And, llvm.Or, llvm.Xor:
+			// Integer binary operations.
+			lhs := operands[0]
+			rhs := operands[1]
+			lhsPtr, err := lhs.asPointer(r)
+			if err == nil {
+				// The lhs is a pointer. This sometimes happens for particular
+				// pointer tricks.
+				switch inst.opcode {
+				case llvm.Add:
+					// This likely means this is part of a
+					// unsafe.Pointer(uintptr(ptr) + offset) pattern.
+					lhsPtr = lhsPtr.addOffset(uint32(rhs.Uint()))
+					locals[inst.localIndex] = lhsPtr
+					continue
+				case llvm.Xor:
+					if rhs.Uint() == 0 {
+						// Special workaround for strings.noescape, see
+						// src/strings/builder.go in the Go source tree. This is
+						// the identity operator, so we can return the input.
+						locals[inst.localIndex] = lhs
+						continue
+					}
+				default:
+					// Catch-all for weird operations that should just be done
+					// at runtime.
+					err := r.runAtRuntime(fn, inst, locals, &mem, indent)
+					if err != nil {
+						return nil, mem, err
+					}
+					continue
+				}
+			}
+			var result uint64
+			switch inst.opcode {
+			case llvm.Add:
+				result = lhs.Uint() + rhs.Uint()
+			case llvm.Sub:
+				result = lhs.Uint() - rhs.Uint()
+			case llvm.Mul:
+				result = lhs.Uint() * rhs.Uint()
+			case llvm.UDiv:
+				result = lhs.Uint() / rhs.Uint()
+			case llvm.SDiv:
+				result = uint64(lhs.Int() / rhs.Int())
+			case llvm.URem:
+				result = lhs.Uint() % rhs.Uint()
+			case llvm.SRem:
+				result = uint64(lhs.Int() % rhs.Int())
+			case llvm.Shl:
+				result = lhs.Uint() << rhs.Uint()
+			case llvm.LShr:
+				result = lhs.Uint() >> rhs.Uint()
+			case llvm.AShr:
+				result = uint64(lhs.Int() >> rhs.Uint())
+			case llvm.And:
+				result = lhs.Uint() & rhs.Uint()
+			case llvm.Or:
+				result = lhs.Uint() | rhs.Uint()
+			case llvm.Xor:
+				result = lhs.Uint() ^ rhs.Uint()
+			default:
+				panic("unreachable")
+			}
+			switch lhs.len(r) {
+			case 8:
+				locals[inst.localIndex] = literalValue{result}
+			case 4:
+				locals[inst.localIndex] = literalValue{uint32(result)}
+			case 2:
+				locals[inst.localIndex] = literalValue{uint16(result)}
+			case 1:
+				locals[inst.localIndex] = literalValue{uint8(result)}
+			default:
+				panic("unknown integer size")
+			}
+			if r.debug {
+				fmt.Fprintln(os.Stderr, indent+instructionNameMap[inst.opcode]+":", lhs, rhs, "->", result)
+			}
+		case llvm.SExt, llvm.ZExt, llvm.Trunc:
+			// Change the size of an integer to a larger or smaller bit width.
+			// We make use of the fact that the Uint() function already
+			// zero-extends the value and that Int() already sign-extends the
+			// value, so we only need to truncate it to the appropriate bit
+			// width. This means we can implement sext, zext and trunc in the
+			// same way, by first {zero,sign}extending all the way up to uint64
+			// and then truncating it as necessary.
+			var value uint64
+			if inst.opcode == llvm.SExt {
+				value = uint64(operands[0].Int())
+			} else {
+				value = operands[0].Uint()
+			}
+			bitwidth := operands[1].Uint()
+			if r.debug {
+				fmt.Fprintln(os.Stderr, indent+instructionNameMap[inst.opcode]+":", value, bitwidth)
+			}
+			switch bitwidth {
+			case 64:
+				locals[inst.localIndex] = literalValue{value}
+			case 32:
+				locals[inst.localIndex] = literalValue{uint32(value)}
+			case 16:
+				locals[inst.localIndex] = literalValue{uint16(value)}
+			case 8:
+				locals[inst.localIndex] = literalValue{uint8(value)}
+			default:
+				panic("unknown integer size in sext/zext/trunc")
+			}
+		case llvm.SIToFP, llvm.UIToFP:
+			var value float64
+			switch inst.opcode {
+			case llvm.SIToFP:
+				value = float64(operands[0].Int())
+			case llvm.UIToFP:
+				value = float64(operands[0].Uint())
+			}
+			bitwidth := operands[1].Uint()
+			if r.debug {
+				fmt.Fprintln(os.Stderr, indent+instructionNameMap[inst.opcode]+":", value, bitwidth)
+			}
+			switch bitwidth {
+			case 64:
+				locals[inst.localIndex] = literalValue{math.Float64bits(value)}
+			case 32:
+				locals[inst.localIndex] = literalValue{math.Float32bits(float32(value))}
+			default:
+				panic("unknown integer size in sitofp/uitofp")
+			}
+		default:
+			if r.debug {
+				fmt.Fprintln(os.Stderr, indent+inst.String())
+			}
+			return nil, mem, r.errorAt(inst, errUnsupportedInst)
+		}
+	}
+	return nil, mem, r.errorAt(bb.instructions[len(bb.instructions)-1], errors.New("interp: reached end of basic block without terminator"))
+}
+
+func (r *runner) runAtRuntime(fn *function, inst instruction, locals []value, mem *memoryView, indent string) *Error {
+	numOperands := inst.llvmInst.OperandsCount()
+	operands := make([]llvm.Value, numOperands)
+	for i := 0; i < numOperands; i++ {
+		operand := inst.llvmInst.Operand(i)
+		if !operand.IsAInstruction().IsNil() || !operand.IsAArgument().IsNil() {
+			operand = locals[fn.locals[operand]].toLLVMValue(operand.Type(), mem)
+		}
+		operands[i] = operand
+	}
+	if r.debug {
+		fmt.Fprintln(os.Stderr, indent+inst.String())
+	}
+	var result llvm.Value
+	switch inst.opcode {
+	case llvm.Call:
+		llvmFn := operands[len(operands)-1]
+		args := operands[:len(operands)-1]
+		for _, arg := range args {
+			if arg.Type().TypeKind() == llvm.PointerTypeKind {
+				mem.markExternalStore(arg)
+			}
+		}
+		result = r.builder.CreateCall(llvmFn, args, inst.name)
+	case llvm.Load:
+		mem.markExternalLoad(operands[0])
+		result = r.builder.CreateLoad(operands[0], inst.name)
+		if inst.llvmInst.IsVolatile() {
+			result.SetVolatile(true)
+		}
+	case llvm.Store:
+		mem.markExternalStore(operands[1])
+		result = r.builder.CreateStore(operands[0], operands[1])
+		if inst.llvmInst.IsVolatile() {
+			result.SetVolatile(true)
+		}
+	case llvm.BitCast:
+		result = r.builder.CreateBitCast(operands[0], inst.llvmInst.Type(), inst.name)
+	case llvm.ExtractValue:
+		indices := inst.llvmInst.Indices()
+		if len(indices) != 1 {
+			panic("expected exactly one index")
+		}
+		result = r.builder.CreateExtractValue(operands[0], int(indices[0]), inst.name)
+	case llvm.InsertValue:
+		indices := inst.llvmInst.Indices()
+		if len(indices) != 1 {
+			panic("expected exactly one index")
+		}
+		result = r.builder.CreateInsertValue(operands[0], operands[1], int(indices[0]), inst.name)
+	case llvm.Add:
+		result = r.builder.CreateAdd(operands[0], operands[1], inst.name)
+	case llvm.Sub:
+		result = r.builder.CreateSub(operands[0], operands[1], inst.name)
+	case llvm.Mul:
+		result = r.builder.CreateMul(operands[0], operands[1], inst.name)
+	case llvm.UDiv:
+		result = r.builder.CreateUDiv(operands[0], operands[1], inst.name)
+	case llvm.SDiv:
+		result = r.builder.CreateSDiv(operands[0], operands[1], inst.name)
+	case llvm.URem:
+		result = r.builder.CreateURem(operands[0], operands[1], inst.name)
+	case llvm.SRem:
+		result = r.builder.CreateSRem(operands[0], operands[1], inst.name)
+	case llvm.ZExt:
+		result = r.builder.CreateZExt(operands[0], inst.llvmInst.Type(), inst.name)
+	default:
+		return r.errorAt(inst, errUnsupportedRuntimeInst)
+	}
+	locals[inst.localIndex] = localValue{result}
+	mem.instructions = append(mem.instructions, result)
+	return nil
+}
+
+func intPredicateString(predicate llvm.IntPredicate) string {
+	switch predicate {
+	case llvm.IntEQ:
+		return "eq"
+	case llvm.IntNE:
+		return "ne"
+	case llvm.IntUGT:
+		return "ugt"
+	case llvm.IntUGE:
+		return "uge"
+	case llvm.IntULT:
+		return "ult"
+	case llvm.IntULE:
+		return "ule"
+	case llvm.IntSGT:
+		return "sgt"
+	case llvm.IntSGE:
+		return "sge"
+	case llvm.IntSLT:
+		return "slt"
+	case llvm.IntSLE:
+		return "sle"
+	default:
+		return "cmp?"
+	}
+}
@@ -1,259 +0,0 @@
-package interp
-
-import (
-	"errors"
-	"strings"
-
-	"tinygo.org/x/go-llvm"
-)
-
-type sideEffectSeverity int
-
-func (severity sideEffectSeverity) String() string {
-	switch severity {
-	case sideEffectInProgress:
-		return "in progress"
-	case sideEffectNone:
-		return "none"
-	case sideEffectLimited:
-		return "limited"
-	case sideEffectAll:
-		return "all"
-	default:
-		return "unknown"
-	}
-}
-
-const (
-	sideEffectInProgress sideEffectSeverity = iota // computing side effects is in progress (for recursive functions)
-	sideEffectNone                                 // no side effects at all (pure)
-	sideEffectLimited                              // has side effects, but the effects are known
-	sideEffectAll                                  // has unknown side effects
-)
-
-// sideEffectResult contains the scan results after scanning a function for side
-// effects (recursively).
-type sideEffectResult struct {
-	severity        sideEffectSeverity
-	mentionsGlobals map[llvm.Value]struct{}
-}
-
-// hasSideEffects scans this function and all descendants, recursively. It
-// returns whether this function has side effects and if it does, which globals
-// it mentions anywhere in this function or any called functions.
-func (e *evalPackage) hasSideEffects(fn llvm.Value) (*sideEffectResult, *Error) {
-	name := fn.Name()
-	switch {
-	case name == "runtime.alloc":
-		// Cannot be scanned but can be interpreted.
-		return &sideEffectResult{severity: sideEffectNone}, nil
-	case name == "runtime.nanotime":
-		// Fixed value at compile time.
-		return &sideEffectResult{severity: sideEffectNone}, nil
-	case name == "runtime._panic":
-		return &sideEffectResult{severity: sideEffectLimited}, nil
-	case name == "runtime.typeAssert":
-		return &sideEffectResult{severity: sideEffectNone}, nil
-	case name == "runtime.interfaceImplements":
-		return &sideEffectResult{severity: sideEffectNone}, nil
-	case name == "runtime.sliceCopy":
-		return &sideEffectResult{severity: sideEffectNone}, nil
-	case name == "runtime.trackPointer":
-		return &sideEffectResult{severity: sideEffectNone}, nil
-	case name == "llvm.dbg.value":
-		return &sideEffectResult{severity: sideEffectNone}, nil
-	case name == "(*sync/atomic.Value).Load" || name == "(*sync/atomic.Value).Store":
-		// These functions do some unsafe pointer loading/storing but are
-		// otherwise safe.
-		return &sideEffectResult{severity: sideEffectLimited}, nil
-	case strings.HasPrefix(name, "llvm.lifetime."):
-		return &sideEffectResult{severity: sideEffectNone}, nil
-	}
-	if fn.IsDeclaration() {
-		return &sideEffectResult{severity: sideEffectLimited}, nil
-	}
-	if e.sideEffectFuncs == nil {
-		e.sideEffectFuncs = make(map[llvm.Value]*sideEffectResult)
-	}
-	if se, ok := e.sideEffectFuncs[fn]; ok {
-		return se, nil
-	}
-	result := &sideEffectResult{
-		severity:        sideEffectInProgress,
-		mentionsGlobals: map[llvm.Value]struct{}{},
-	}
-	e.sideEffectFuncs[fn] = result
-	dirtyLocals := map[llvm.Value]struct{}{}
-	for bb := fn.EntryBasicBlock(); !bb.IsNil(); bb = llvm.NextBasicBlock(bb) {
-		for inst := bb.FirstInstruction(); !inst.IsNil(); inst = llvm.NextInstruction(inst) {
-			if inst.IsAInstruction().IsNil() {
-				// Should not happen in valid IR.
-				panic("not an instruction")
-			}
-
-			// Check for any globals mentioned anywhere in the function. Assume
-			// any mentioned globals may be read from or written to when
-			// executed, thus must be marked dirty with a call.
-			for i := 0; i < inst.OperandsCount(); i++ {
-				operand := inst.Operand(i)
-				if !operand.IsAGlobalVariable().IsNil() {
-					result.mentionsGlobals[operand] = struct{}{}
-				}
-			}
-
-			switch inst.InstructionOpcode() {
-			case llvm.IndirectBr, llvm.Invoke:
-				// Not emitted by the compiler.
-				return nil, e.errorAt(inst, errors.New("unknown instructions"))
-			case llvm.Call:
-				child := inst.CalledValue()
-				if !child.IsAInlineAsm().IsNil() {
-					// Inline assembly. This most likely has side effects.
-					// Assume they're only limited side effects, similar to
-					// external function calls.
-					result.updateSeverity(sideEffectLimited)
-					continue
-				}
-				if child.IsAFunction().IsNil() {
-					// Indirect call?
-					// In any case, we can't know anything here about what it
-					// affects exactly so mark this function as invoking all
-					// possible side effects.
-					result.updateSeverity(sideEffectAll)
-					continue
-				}
-				if child.IsDeclaration() {
-					// External function call. Assume only limited side effects
-					// (no affected globals, etc.).
-					switch child.Name() {
-					case "runtime.alloc":
-						continue
-					case "runtime.typeAssert":
-						continue // implemented in interp
-					case "runtime.interfaceImplements":
-						continue // implemented in interp
-					}
-					if e.hasLocalSideEffects(dirtyLocals, inst) {
-						result.updateSeverity(sideEffectLimited)
-					}
-					continue
-				}
-				childSideEffects, err := e.hasSideEffects(child)
-				if err != nil {
-					return nil, err
-				}
-				switch childSideEffects.severity {
-				case sideEffectInProgress, sideEffectNone:
-					// no side effects or recursive function - continue scanning
-				case sideEffectLimited:
-					// The return value may be problematic.
-					if e.hasLocalSideEffects(dirtyLocals, inst) {
-						result.updateSeverity(sideEffectLimited)
-					}
-				case sideEffectAll:
-					result.updateSeverity(sideEffectAll)
-				default:
-					panic("unreachable")
-				}
-			case llvm.Load:
-				if inst.IsVolatile() {
-					result.updateSeverity(sideEffectLimited)
-				}
-				if _, ok := e.dirtyGlobals[inst.Operand(0)]; ok {
-					if e.hasLocalSideEffects(dirtyLocals, inst) {
-						result.updateSeverity(sideEffectLimited)
-					}
-				}
-			case llvm.Store:
-				if inst.IsVolatile() {
-					result.updateSeverity(sideEffectLimited)
-				}
-			case llvm.IntToPtr:
-				// Pointer casts are not yet supported.
-				result.updateSeverity(sideEffectLimited)
-			default:
-				// Ignore most instructions.
-				// Check this list for completeness:
-				// https://godoc.org/github.com/llvm-mirror/llvm/bindings/go/llvm#Opcode
-			}
-		}
-	}
-
-	if result.severity == sideEffectInProgress {
-		// No side effect was reported for this function.
-		result.severity = sideEffectNone
-	}
-	return result, nil
-}
-
-// hasLocalSideEffects checks whether the given instruction flows into a branch
-// or return instruction, in which case the whole function must be marked as
-// having side effects and be called at runtime.
-func (e *Eval) hasLocalSideEffects(dirtyLocals map[llvm.Value]struct{}, inst llvm.Value) bool {
-	if _, ok := dirtyLocals[inst]; ok {
-		// It is already known that this local is dirty.
-		return true
-	}
-
-	for use := inst.FirstUse(); !use.IsNil(); use = use.NextUse() {
-		user := use.User()
-		if user.IsAInstruction().IsNil() {
-			// Should not happen in valid IR.
-			panic("user not an instruction")
-		}
-		switch user.InstructionOpcode() {
-		case llvm.Br, llvm.Switch:
-			// A branch on a dirty value makes this function dirty: it cannot be
-			// interpreted at compile time so has to be run at runtime. It is
-			// marked as having side effects for this reason.
-			return true
-		case llvm.Ret:
-			// This function returns a dirty value so it is itself marked as
-			// dirty to make sure it is called at runtime.
-			return true
-		case llvm.Store:
-			ptr := user.Operand(1)
-			if !ptr.IsAGlobalVariable().IsNil() {
-				// Store to a global variable.
-				// Already handled in (*Eval).hasSideEffects.
-				continue
-			}
-			// This store might affect all kinds of values. While it is
-			// certainly possible to traverse through all of them, the easiest
-			// option right now is to just assume the worst and say that this
-			// function has side effects.
-			// TODO: traverse through all stores and mark all relevant allocas /
-			// globals dirty.
-			return true
-		default:
-			// All instructions that take 0 or more operands (1 or more if it
-			// was a use) and produce a result.
-			// For a list:
-			// https://godoc.org/github.com/llvm-mirror/llvm/bindings/go/llvm#Opcode
-			dirtyLocals[user] = struct{}{}
-			if e.hasLocalSideEffects(dirtyLocals, user) {
-				return true
-			}
-		}
-	}
-
-	// No side effects found.
-	return false
-}
-
-// updateSeverity sets r.severity to the max of r.severity and severity,
-// conservatively assuming the worst severity.
-func (r *sideEffectResult) updateSeverity(severity sideEffectSeverity) {
-	if severity > r.severity {
-		r.severity = severity
-	}
-}
-
-// updateSeverity updates the severity with the severity of the child severity,
-// like in a function call. This means it also copies the mentioned globals.
-func (r *sideEffectResult) update(child *sideEffectResult) {
-	r.updateSeverity(child.severity)
-	for global := range child.mentionsGlobals {
-		r.mentionsGlobals[global] = struct{}{}
-	}
-}
@@ -1,95 +0,0 @@
-package interp
-
-import (
-	"os"
-	"sort"
-	"testing"
-
-	"tinygo.org/x/go-llvm"
-)
-
-var scanTestTable = []struct {
-	name            string
-	severity        sideEffectSeverity
-	mentionsGlobals []string
-}{
-	{"returnsConst", sideEffectNone, nil},
-	{"returnsArg", sideEffectNone, nil},
-	{"externalCallOnly", sideEffectNone, nil},
-	{"externalCallAndReturn", sideEffectLimited, nil},
-	{"externalCallBranch", sideEffectLimited, nil},
-	{"readCleanGlobal", sideEffectNone, []string{"cleanGlobalInt"}},
-	{"readDirtyGlobal", sideEffectLimited, []string{"dirtyGlobalInt"}},
-	{"callFunctionPointer", sideEffectAll, []string{"functionPointer"}},
-	{"getDirtyPointer", sideEffectLimited, nil},
-	{"storeToPointer", sideEffectLimited, nil},
-	{"callTypeAssert", sideEffectNone, nil},
-	{"callInterfaceImplements", sideEffectNone, nil},
-}
-
-func TestScan(t *testing.T) {
-	t.Parallel()
-
-	// Read the input IR.
-	path := "testdata/scan.ll"
-	ctx := llvm.NewContext()
-	buf, err := llvm.NewMemoryBufferFromFile(path)
-	os.Stat(path) // make sure this file is tracked by `go test` caching
-	if err != nil {
-		t.Fatalf("could not read file %s: %v", path, err)
-	}
-	mod, err := ctx.ParseIR(buf)
-	if err != nil {
-		t.Fatalf("could not load module:\n%v", err)
-	}
-
-	// Check all to-be-tested functions.
-	for _, tc := range scanTestTable {
-		// Create an eval object, for testing.
-		e := &Eval{
-			Mod:          mod,
-			TargetData:   llvm.NewTargetData(mod.DataLayout()),
-			dirtyGlobals: map[llvm.Value]struct{}{},
-		}
-
-		// Mark some globals dirty, for testing.
-		e.markDirty(mod.NamedGlobal("dirtyGlobalInt"))
-
-		// Scan for side effects.
-		fn := mod.NamedFunction(tc.name)
-		if fn.IsNil() {
-			t.Errorf("scan test: could not find tested function %s in the IR", tc.name)
-			continue
-		}
-		evalPkg := &evalPackage{e, "testdata"}
-		result, err := evalPkg.hasSideEffects(fn)
-		if err != nil {
-			t.Errorf("scan test: failed to scan %s for side effects: %v", fn.Name(), err)
-		}
-
-		// Check whether the result is what we expect.
-		if result.severity != tc.severity {
-			t.Errorf("scan test: function %s should have severity %s but it has %s", tc.name, tc.severity, result.severity)
-		}
-
-		// Check whether the mentioned globals match with what we'd expect.
-		mentionsGlobalNames := make([]string, 0, len(result.mentionsGlobals))
-		for global := range result.mentionsGlobals {
-			mentionsGlobalNames = append(mentionsGlobalNames, global.Name())
-		}
-		sort.Strings(mentionsGlobalNames)
-		globalsMismatch := false
-		if len(result.mentionsGlobals) != len(tc.mentionsGlobals) {
-			globalsMismatch = true
-		} else {
-			for i, globalName := range mentionsGlobalNames {
-				if tc.mentionsGlobals[i] != globalName {
-					globalsMismatch = true
-				}
-			}
-		}
-		if globalsMismatch {
-			t.Errorf("scan test: expected %s to mention globals %v, but it mentions globals %v", tc.name, tc.mentionsGlobals, mentionsGlobalNames)
-		}
-	}
-}
@@ -4,6 +4,10 @@ target triple = "x86_64--linux"
@main.v1 = internal global i64 0
@main.nonConst1 = global [4 x i64] zeroinitializer
@main.nonConst2 = global i64 0
+@main.someArray = global [8 x {i16, i32}] zeroinitializer
+@main.exportedValue = global [1 x i16*] [i16* @main.exposedValue1]
+@main.exposedValue1 = global i16 0
+@main.exposedValue2 = global i16 0

 declare void @runtime.printint64(i64) unnamed_addr

@@ -47,6 +51,20 @@ entry:
  %value2 = load i64, i64* %gep2
  store i64 %value2, i64* @main.nonConst2

+  ; Test that the following GEP works:
+  ;     var someArray
+  ;     modifyExternal(&someArray[3].field1)
+  %gep3 = getelementptr [8 x {i16, i32}], [8 x {i16, i32}]* @main.someArray, i32 0, i32 3, i32 1
+  call void @modifyExternal(i32* %gep3)
+
+  ; Test that marking a value as external also marks all referenced values.
+  call void @modifyExternal(i32* bitcast ([1 x i16*]* @main.exportedValue to i32*))
+  store i16 5, i16* @main.exposedValue1
+
+  ; Test that this even propagates through functions.
+  call void @modifyExternal(i32* bitcast (void ()* @willModifyGlobal to i32*))
+  store i16 7, i16* @main.exposedValue2
+
  ret void
 }

@@ -58,3 +76,14 @@ entry:
 }

 declare i64 @someValue()
+
+declare void @modifyExternal(i32*)
+
+; This function will modify an external value. By passing this function as a
+; function pointer to an external function, @main.exposedValue2 should be
+; marked as external.
+define void @willModifyGlobal() {
+entry:
+  store i16 8, i16* @main.exposedValue2
+  ret void
+}
@@ -3,6 +3,10 @@ target triple = "x86_64--linux"

@main.nonConst1 = local_unnamed_addr global [4 x i64] zeroinitializer
@main.nonConst2 = local_unnamed_addr global i64 0
+@main.someArray = global [8 x { i16, i32 }] zeroinitializer
+@main.exportedValue = global [1 x i16*] [i16* @main.exposedValue1]
+@main.exposedValue1 = global i16 0
+@main.exposedValue2 = local_unnamed_addr global i16 0

 declare void @runtime.printint64(i64) unnamed_addr

@@ -16,6 +20,11 @@ entry:
  store i64 %value1, i64* getelementptr inbounds ([4 x i64], [4 x i64]* @main.nonConst1, i32 0, i32 0)
  %value2 = load i64, i64* getelementptr inbounds ([4 x i64], [4 x i64]* @main.nonConst1, i32 0, i32 0)
  store i64 %value2, i64* @main.nonConst2
+  call void @modifyExternal(i32* getelementptr inbounds ([8 x { i16, i32 }], [8 x { i16, i32 }]* @main.someArray, i32 0, i32 3, i32 1))
+  call void @modifyExternal(i32* bitcast ([1 x i16*]* @main.exportedValue to i32*))
+  store i16 5, i16* @main.exposedValue1
+  call void @modifyExternal(i32* bitcast (void ()* @willModifyGlobal to i32*))
+  store i16 7, i16* @main.exposedValue2
  ret void
 }

@@ -27,3 +36,11 @@ entry:
 }

 declare i64 @someValue() local_unnamed_addr
+
+declare void @modifyExternal(i32*) local_unnamed_addr
+
+define void @willModifyGlobal() {
+entry:
+  store i16 8, i16* @main.exposedValue2
+  ret void
+}
@@ -48,8 +48,7 @@ entry:
 define internal void @main.testNonConstantBinarySet() {
  %hashmap.key = alloca i8
  %hashmap.value = alloca i8
-  ; Create hashmap from global. This breaks the normal hashmapBinarySet
-  ; optimization, to test the fallback.
+  ; Create hashmap from global.
  %map.new = call %runtime.hashmap* @runtime.hashmapMake(i8 1, i8 1, i32 1, i8* undef, i8* null)
  store %runtime.hashmap* %map.new, %runtime.hashmap** @main.binaryMap
  %map = load %runtime.hashmap*, %runtime.hashmap** @main.binaryMap
@@ -64,8 +63,7 @@ define internal void @main.testNonConstantBinarySet() {
 ; operations (with string keys).
 define internal void @main.testNonConstantStringSet() {
  %hashmap.value = alloca i8
-  ; Create hashmap from global. This breaks the normal hashmapStringSet
-  ; optimization, to test the fallback.
+  ; Create hashmap from global.
  %map.new = call %runtime.hashmap* @runtime.hashmapMake(i8 8, i8 1, i32 1, i8* undef, i8* null)
  store %runtime.hashmap* %map.new, %runtime.hashmap** @main.stringMap
  %map = load %runtime.hashmap*, %runtime.hashmap** @main.stringMap
@@ -2,27 +2,19 @@ target datalayout = "e-m:e-p:32:32-Fi8-i64:64-v128:64:128-a:0:32-n32-S64"
 target triple = "armv6m-none-eabi"

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

@main.m = local_unnamed_addr global %runtime.hashmap* @"main$map"
-@main.binaryMap = local_unnamed_addr global %runtime.hashmap* @"main$map.4"
-@main.stringMap = local_unnamed_addr global %runtime.hashmap* @"main$map.6"
+@main.binaryMap = local_unnamed_addr global %runtime.hashmap* @"main$map.1"
+@main.stringMap = local_unnamed_addr global %runtime.hashmap* @"main$map.3"
@main.init.string = internal unnamed_addr constant [7 x i8] c"CONNECT"
-@"main$mapbucket" = internal unnamed_addr global { [8 x i8], i8*, [8 x i8], [8 x %runtime._string] } { [8 x i8] c"\04\00\00\00\00\00\00\00", i8* null, [8 x i8] c"\01\00\00\00\00\00\00\00", [8 x %runtime._string] [%runtime._string { i8* getelementptr inbounds ([7 x i8], [7 x i8]* @main.init.string, i32 0, i32 0), i32 7 }, %runtime._string zeroinitializer, %runtime._string zeroinitializer, %runtime._string zeroinitializer, %runtime._string zeroinitializer, %runtime._string zeroinitializer, %runtime._string zeroinitializer, %runtime._string zeroinitializer] }
-@"main$map" = internal unnamed_addr global %runtime.hashmap { %runtime.hashmap* null, i8* getelementptr inbounds ({ [8 x i8], i8*, [8 x i8], [8 x %runtime._string] }, { [8 x i8], i8*, [8 x i8], [8 x %runtime._string] }* @"main$mapbucket", i32 0, i32 0, i32 0), i32 1, i8 1, i8 8, i8 0 }
-@"main$alloca.2" = internal global i8 1
-@"main$alloca.3" = internal global i8 2
-@"main$map.4" = internal unnamed_addr global %runtime.hashmap { %runtime.hashmap* null, i8* null, i32 0, i8 1, i8 1, i8 0 }
-@"main$alloca.5" = internal global i8 2
-@"main$map.6" = internal unnamed_addr global %runtime.hashmap { %runtime.hashmap* null, i8* null, i32 0, i8 8, i8 1, i8 0 }
-
-declare void @runtime.hashmapBinarySet(%runtime.hashmap*, i8*, i8*, i8*, i8*) local_unnamed_addr
-
-declare void @runtime.hashmapStringSet(%runtime.hashmap*, i8*, i32, i8*, i8*, i8*) local_unnamed_addr
+@"main$map" = internal global %runtime.hashmap { %runtime.hashmap* null, i8* getelementptr inbounds ({ [8 x i8], i8*, { i8, [7 x i8] }, { { [7 x i8]*, [4 x i8] }, [56 x i8] } }, { [8 x i8], i8*, { i8, [7 x i8] }, { { [7 x i8]*, [4 x i8] }, [56 x i8] } }* @"main$mapbucket", i32 0, i32 0, i32 0), i32 1, i8 1, i8 8, i8 0 }
+@"main$mapbucket" = internal unnamed_addr global { [8 x i8], i8*, { i8, [7 x i8] }, { { [7 x i8]*, [4 x i8] }, [56 x i8] } } { [8 x i8] c"\04\00\00\00\00\00\00\00", i8* null, { i8, [7 x i8] } { i8 1, [7 x i8] zeroinitializer }, { { [7 x i8]*, [4 x i8] }, [56 x i8] } { { [7 x i8]*, [4 x i8] } { [7 x i8]* @main.init.string, [4 x i8] c"\07\00\00\00" }, [56 x i8] zeroinitializer } }
+@"main$map.1" = internal global %runtime.hashmap { %runtime.hashmap* null, i8* getelementptr inbounds ({ [8 x i8], i8*, { i8, [7 x i8] }, { i8, [7 x i8] } }, { [8 x i8], i8*, { i8, [7 x i8] }, { i8, [7 x i8] } }* @"main$mapbucket.2", i32 0, i32 0, i32 0), i32 1, i8 1, i8 1, i8 0 }
+@"main$mapbucket.2" = internal unnamed_addr global { [8 x i8], i8*, { i8, [7 x i8] }, { i8, [7 x i8] } } { [8 x i8] c"\04\00\00\00\00\00\00\00", i8* null, { i8, [7 x i8] } { i8 1, [7 x i8] zeroinitializer }, { i8, [7 x i8] } { i8 2, [7 x i8] zeroinitializer } }
+@"main$map.3" = internal global %runtime.hashmap { %runtime.hashmap* null, i8* getelementptr inbounds ({ [8 x i8], i8*, { { [7 x i8]*, [4 x i8] }, [56 x i8] }, { i8, [7 x i8] } }, { [8 x i8], i8*, { { [7 x i8]*, [4 x i8] }, [56 x i8] }, { i8, [7 x i8] } }* @"main$mapbucket.4", i32 0, i32 0, i32 0), i32 1, i8 8, i8 1, i8 0 }
+@"main$mapbucket.4" = internal unnamed_addr global { [8 x i8], i8*, { { [7 x i8]*, [4 x i8] }, [56 x i8] }, { i8, [7 x i8] } } { [8 x i8] c"x\00\00\00\00\00\00\00", i8* null, { { [7 x i8]*, [4 x i8] }, [56 x i8] } { { [7 x i8]*, [4 x i8] } { [7 x i8]* @main.init.string, [4 x i8] c"\07\00\00\00" }, [56 x i8] zeroinitializer }, { i8, [7 x i8] } { i8 2, [7 x i8] zeroinitializer } }

 define void @runtime.initAll() unnamed_addr {
 entry:
-  call void @runtime.hashmapBinarySet(%runtime.hashmap* @"main$map.4", i8* @"main$alloca.2", i8* @"main$alloca.3", i8* undef, i8* null)
-  call void @runtime.hashmapStringSet(%runtime.hashmap* @"main$map.6", i8* getelementptr inbounds ([7 x i8], [7 x i8]* @main.init.string, i32 0, i32 0), i32 7, i8* @"main$alloca.5", i8* undef, i8* null)
  ret void
 }
@@ -1,78 +0,0 @@
-target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
-target triple = "x86_64--linux"
-
-%runtime.typecodeID = type { %runtime.typecodeID*, i64 }
-
-declare i1 @runtime.typeAssert(i64, %runtime.typecodeID*, i8*, i8*)
-declare i1 @runtime.interfaceImplements(i64, i8**)
-
-define i64 @returnsConst() {
-  ret i64 0
-}
-
-define i64 @returnsArg(i64 %arg) {
-  ret i64 %arg
-}
-
-declare i64 @externalCall()
-
-define i64 @externalCallOnly() {
-  %result = call i64 @externalCall()
-  ret i64 0
-}
-
-define i64 @externalCallAndReturn() {
-  %result = call i64 @externalCall()
-  ret i64 %result
-}
-
-define i64 @externalCallBranch() {
-  %result = call i64 @externalCall()
-  %zero = icmp eq i64 %result, 0
-  br i1 %zero, label %if.then, label %if.done
-
-if.then:
-  ret i64 2
-
-if.done:
-  ret i64 4
-}
-
-@cleanGlobalInt = global i64 5
-define i64 @readCleanGlobal() {
-  %global = load i64, i64* @cleanGlobalInt
-  ret i64 %global
-}
-
-@dirtyGlobalInt = global i64 5
-define i64 @readDirtyGlobal() {
-  %global = load i64, i64* @dirtyGlobalInt
-  ret i64 %global
-}
-
-declare i64* @getDirtyPointer()
-
-define void @storeToPointer() {
-  %ptr = call i64* @getDirtyPointer()
-  store i64 3, i64* %ptr
-  ret void
-}
-
-@functionPointer = global i64()* null
-define i64 @callFunctionPointer() {
-  %fp = load i64()*, i64()** @functionPointer
-  %result = call i64 %fp()
-  ret i64 %result
-}
-
-define i1 @callTypeAssert() {
-  ; Note: parameters are not realistic.
-  %ok = call i1 @runtime.typeAssert(i64 0, %runtime.typecodeID* null, i8* undef, i8* null)
-  ret i1 %ok
-}
-
-define i1 @callInterfaceImplements() {
-  ; Note: parameters are not realistic.
-  %ok = call i1 @runtime.interfaceImplements(i64 0, i8** null)
-  ret i1 %ok
-}
@@ -1,6 +1,8 @@
 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
 target triple = "x86_64--linux"

+@"main$alloc.1" = internal unnamed_addr constant [6 x i8] c"\05\00{\00\00\04"
+
 declare void @runtime.printuint8(i8) local_unnamed_addr

 declare void @runtime.printint16(i16) local_unnamed_addr
@@ -15,6 +17,7 @@ entry:
  call void @runtime.printuint8(i8 3)
  call void @runtime.printuint8(i8 3)
  call void @runtime.printint16(i16 5)
-  call void @runtime.printint16(i16 5)
+  %int16SliceDst.val = load i16, i16* bitcast ([6 x i8]* @"main$alloc.1" to i16*)
+  call void @runtime.printint16(i16 %int16SliceDst.val)
  ret void
 }
@@ -1,121 +0,0 @@
-package interp
-
-import (
-	"tinygo.org/x/go-llvm"
-)
-
-// Return a list of values (actually, instructions) where this value is used as
-// an operand.
-func getUses(value llvm.Value) []llvm.Value {
-	var uses []llvm.Value
-	use := value.FirstUse()
-	for !use.IsNil() {
-		uses = append(uses, use.User())
-		use = use.NextUse()
-	}
-	return uses
-}
-
-// getStringBytes loads the byte slice of a Go string represented as a
-// {ptr, len} pair.
-func getStringBytes(strPtr Value, strLen llvm.Value) []byte {
-	if !strLen.IsConstant() {
-		panic("getStringBytes with a non-constant length")
-	}
-	buf := make([]byte, strLen.ZExtValue())
-	for i := range buf {
-		gep, err := strPtr.GetElementPtr([]uint32{uint32(i)})
-		if err != nil {
-			panic(err) // TODO
-		}
-		c := gep.Load()
-		buf[i] = byte(c.ZExtValue())
-	}
-	return buf
-}
-
-// getLLVMIndices converts an []uint32 into an []llvm.Value, for use in
-// llvm.ConstGEP.
-func getLLVMIndices(int32Type llvm.Type, indices []uint32) []llvm.Value {
-	llvmIndices := make([]llvm.Value, len(indices))
-	for i, index := range indices {
-		llvmIndices[i] = llvm.ConstInt(int32Type, uint64(index), false)
-	}
-	return llvmIndices
-}
-
-// Return true if this type is a scalar value (integer or floating point), false
-// otherwise.
-func isScalar(t llvm.Type) bool {
-	switch t.TypeKind() {
-	case llvm.IntegerTypeKind, llvm.FloatTypeKind, llvm.DoubleTypeKind:
-		return true
-	default:
-		return false
-	}
-}
-
-// isPointerNil returns whether this is a nil pointer or not. The ok value
-// indicates whether the result is certain: if it is false the result boolean is
-// not valid.
-func isPointerNil(v llvm.Value) (result bool, ok bool) {
-	if !v.IsAConstantExpr().IsNil() {
-		switch v.Opcode() {
-		case llvm.IntToPtr:
-			// Whether a constant inttoptr is nil is easy to
-			// determine.
-			result, ok = isZero(v.Operand(0))
-			if ok {
-				return
-			}
-		case llvm.BitCast, llvm.GetElementPtr:
-			// These const instructions are just a kind of wrappers for the
-			// underlying pointer.
-			return isPointerNil(v.Operand(0))
-		}
-	}
-	if !v.IsAConstantPointerNull().IsNil() {
-		// A constant pointer null is always null, of course.
-		return true, true
-	}
-	if !v.IsAGlobalValue().IsNil() {
-		// A global value is never null.
-		return false, true
-	}
-	return false, false // not valid
-}
-
-// isZero returns whether the value in v is the integer zero, and whether that
-// can be known right now.
-func isZero(v llvm.Value) (result bool, ok bool) {
-	if !v.IsAConstantExpr().IsNil() {
-		switch v.Opcode() {
-		case llvm.PtrToInt:
-			return isPointerNil(v.Operand(0))
-		}
-	}
-	if !v.IsAConstantInt().IsNil() {
-		val := v.ZExtValue()
-		return val == 0, true
-	}
-	return false, false // not valid
-}
-
-// unwrap returns the underlying value, with GEPs removed. This can be useful to
-// get the underlying global of a GEP pointer.
-func unwrap(value llvm.Value) llvm.Value {
-	for {
-		if !value.IsAConstantExpr().IsNil() {
-			switch value.Opcode() {
-			case llvm.GetElementPtr:
-				value = value.Operand(0)
-				continue
-			}
-		} else if !value.IsAGetElementPtrInst().IsNil() {
-			value = value.Operand(0)
-			continue
-		}
-		break
-	}
-	return value
-}
@@ -1,419 +0,0 @@
-package interp
-
-// This file provides a litte bit of abstraction around LLVM values.
-
-import (
-	"errors"
-	"strconv"
-
-	"tinygo.org/x/go-llvm"
-)
-
-// A Value is a LLVM value with some extra methods attached for easier
-// interpretation.
-type Value interface {
-	Value() llvm.Value                     // returns a LLVM value
-	Type() llvm.Type                       // equal to Value().Type()
-	IsConstant() bool                      // returns true if this value is a constant value
-	Load() llvm.Value                      // dereference a pointer
-	Store(llvm.Value)                      // store to a pointer
-	GetElementPtr([]uint32) (Value, error) // returns an interior pointer
-	String() string                        // string representation, for debugging
-}
-
-// A type that simply wraps a LLVM constant value.
-type LocalValue struct {
-	Eval       *Eval
-	Underlying llvm.Value
-}
-
-// Value implements Value by returning the constant value itself.
-func (v *LocalValue) Value() llvm.Value {
-	return v.Underlying
-}
-
-func (v *LocalValue) Type() llvm.Type {
-	return v.Underlying.Type()
-}
-
-func (v *LocalValue) IsConstant() bool {
-	if _, ok := v.Eval.dirtyGlobals[unwrap(v.Underlying)]; ok {
-		return false
-	}
-	return v.Underlying.IsConstant()
-}
-
-// Load loads a constant value if this is a constant pointer.
-func (v *LocalValue) Load() llvm.Value {
-	if !v.Underlying.IsAGlobalVariable().IsNil() {
-		return v.Underlying.Initializer()
-	}
-	switch v.Underlying.Opcode() {
-	case llvm.GetElementPtr:
-		indices := v.getConstGEPIndices()
-		if indices[0] != 0 {
-			panic("invalid GEP")
-		}
-		global := v.Eval.getValue(v.Underlying.Operand(0))
-		agg := global.Load()
-		return llvm.ConstExtractValue(agg, indices[1:])
-	case llvm.BitCast:
-		panic("interp: load from a bitcast")
-	default:
-		panic("interp: load from a constant")
-	}
-}
-
-// Store stores to the underlying value if the value type is a pointer type,
-// otherwise it panics.
-func (v *LocalValue) Store(value llvm.Value) {
-	if !v.Underlying.IsAGlobalVariable().IsNil() {
-		if !value.IsConstant() {
-			v.MarkDirty()
-			v.Eval.builder.CreateStore(value, v.Underlying)
-		} else {
-			v.Underlying.SetInitializer(value)
-		}
-		return
-	}
-	if !value.IsConstant() {
-		v.MarkDirty()
-		v.Eval.builder.CreateStore(value, v.Underlying)
-		return
-	}
-	switch v.Underlying.Opcode() {
-	case llvm.GetElementPtr:
-		indices := v.getConstGEPIndices()
-		if indices[0] != 0 {
-			panic("invalid GEP")
-		}
-		global := &LocalValue{v.Eval, v.Underlying.Operand(0)}
-		agg := global.Load()
-		agg = llvm.ConstInsertValue(agg, value, indices[1:])
-		global.Store(agg)
-		return
-	default:
-		panic("interp: store on a constant")
-	}
-}
-
-// GetElementPtr returns a GEP when the underlying value is of pointer type.
-func (v *LocalValue) GetElementPtr(indices []uint32) (Value, error) {
-	if !v.Underlying.IsAGlobalVariable().IsNil() {
-		int32Type := v.Underlying.Type().Context().Int32Type()
-		gep := llvm.ConstGEP(v.Underlying, getLLVMIndices(int32Type, indices))
-		return &LocalValue{v.Eval, gep}, nil
-	}
-	if !v.Underlying.IsAConstantExpr().IsNil() {
-		switch v.Underlying.Opcode() {
-		case llvm.GetElementPtr, llvm.IntToPtr, llvm.BitCast:
-			int32Type := v.Underlying.Type().Context().Int32Type()
-			llvmIndices := getLLVMIndices(int32Type, indices)
-			return &LocalValue{v.Eval, llvm.ConstGEP(v.Underlying, llvmIndices)}, nil
-		}
-	}
-	return nil, errors.New("interp: unknown GEP")
-}
-
-// stripPointerCasts removes all const bitcasts from pointer values, if there
-// are any.
-func (v *LocalValue) stripPointerCasts() *LocalValue {
-	value := v.Underlying
-	for {
-		if !value.IsAConstantExpr().IsNil() {
-			switch value.Opcode() {
-			case llvm.BitCast:
-				value = value.Operand(0)
-				continue
-			}
-		}
-		return &LocalValue{
-			Eval:       v.Eval,
-			Underlying: value,
-		}
-	}
-}
-
-func (v *LocalValue) String() string {
-	isConstant := "false"
-	if v.IsConstant() {
-		isConstant = "true"
-	}
-	return "&LocalValue{Type: " + v.Type().String() + ", IsConstant: " + isConstant + "}"
-}
-
-// getConstGEPIndices returns indices of this constant GEP, if this is a GEP
-// instruction. If it is not, the behavior is undefined.
-func (v *LocalValue) getConstGEPIndices() []uint32 {
-	indices := make([]uint32, v.Underlying.OperandsCount()-1)
-	for i := range indices {
-		operand := v.Underlying.Operand(i + 1)
-		indices[i] = uint32(operand.ZExtValue())
-	}
-	return indices
-}
-
-// MarkDirty marks this global as dirty, meaning that every load from and store
-// to this global (from now on) must be performed at runtime.
-func (v *LocalValue) MarkDirty() {
-	underlying := unwrap(v.Underlying)
-	if underlying.IsAGlobalVariable().IsNil() {
-		panic("trying to mark a non-global as dirty")
-	}
-	if !v.IsConstant() {
-		return // already dirty
-	}
-	v.Eval.dirtyGlobals[underlying] = struct{}{}
-}
-
-// MapValue implements a Go map which is created at compile time and stored as a
-// global variable.
-type MapValue struct {
-	Eval       *Eval
-	PkgName    string
-	Underlying llvm.Value
-	Keys       []Value
-	Values     []Value
-	KeySize    int
-	ValueSize  int
-	KeyType    llvm.Type
-	ValueType  llvm.Type
-}
-
-func (v *MapValue) newBucket() llvm.Value {
-	ctx := v.Eval.Mod.Context()
-	i8ptrType := llvm.PointerType(ctx.Int8Type(), 0)
-	bucketType := ctx.StructType([]llvm.Type{
-		llvm.ArrayType(ctx.Int8Type(), 8), // tophash
-		i8ptrType,                         // next bucket
-		llvm.ArrayType(v.KeyType, 8),      // key type
-		llvm.ArrayType(v.ValueType, 8),    // value type
-	}, false)
-	bucketValue := llvm.ConstNull(bucketType)
-	bucket := llvm.AddGlobal(v.Eval.Mod, bucketType, v.PkgName+"$mapbucket")
-	bucket.SetInitializer(bucketValue)
-	bucket.SetLinkage(llvm.InternalLinkage)
-	bucket.SetUnnamedAddr(true)
-	return bucket
-}
-
-// Value returns a global variable which is a pointer to the actual hashmap.
-func (v *MapValue) Value() llvm.Value {
-	if !v.Underlying.IsNil() {
-		return v.Underlying
-	}
-
-	ctx := v.Eval.Mod.Context()
-	i8ptrType := llvm.PointerType(ctx.Int8Type(), 0)
-
-	var firstBucketGlobal llvm.Value
-	if len(v.Keys) == 0 {
-		// there are no buckets
-		firstBucketGlobal = llvm.ConstPointerNull(i8ptrType)
-	} else {
-		// create initial bucket
-		firstBucketGlobal = v.newBucket()
-	}
-
-	// Insert each key/value pair in the hashmap.
-	bucketGlobal := firstBucketGlobal
-	for i, key := range v.Keys {
-		var keyBuf []byte
-		llvmKey := key.Value()
-		llvmValue := v.Values[i].Value()
-		if key.Type().TypeKind() == llvm.StructTypeKind && key.Type().StructName() == "runtime._string" {
-			keyPtr := llvm.ConstExtractValue(llvmKey, []uint32{0})
-			keyLen := llvm.ConstExtractValue(llvmKey, []uint32{1})
-			keyPtrVal := v.Eval.getValue(keyPtr)
-			keyBuf = getStringBytes(keyPtrVal, keyLen)
-		} else if key.Type().TypeKind() == llvm.IntegerTypeKind {
-			keyBuf = make([]byte, v.Eval.TargetData.TypeAllocSize(key.Type()))
-			n := key.Value().ZExtValue()
-			for i := range keyBuf {
-				keyBuf[i] = byte(n)
-				n >>= 8
-			}
-		} else if key.Type().TypeKind() == llvm.ArrayTypeKind &&
-			key.Type().ElementType().TypeKind() == llvm.IntegerTypeKind &&
-			key.Type().ElementType().IntTypeWidth() == 8 {
-			keyBuf = make([]byte, v.Eval.TargetData.TypeAllocSize(key.Type()))
-			for i := range keyBuf {
-				keyBuf[i] = byte(llvm.ConstExtractValue(llvmKey, []uint32{uint32(i)}).ZExtValue())
-			}
-		} else {
-			panic("interp: map key type not implemented: " + key.Type().String())
-		}
-		hash := v.hash(keyBuf)
-
-		if i%8 == 0 && i != 0 {
-			// Bucket is full, create a new one.
-			newBucketGlobal := v.newBucket()
-			zero := llvm.ConstInt(ctx.Int32Type(), 0, false)
-			newBucketPtr := llvm.ConstInBoundsGEP(newBucketGlobal, []llvm.Value{zero})
-			newBucketPtrCast := llvm.ConstBitCast(newBucketPtr, i8ptrType)
-			// insert pointer into old bucket
-			bucket := bucketGlobal.Initializer()
-			bucket = llvm.ConstInsertValue(bucket, newBucketPtrCast, []uint32{1})
-			bucketGlobal.SetInitializer(bucket)
-			// switch to next bucket
-			bucketGlobal = newBucketGlobal
-		}
-
-		tophashValue := llvm.ConstInt(ctx.Int8Type(), uint64(v.topHash(hash)), false)
-		bucket := bucketGlobal.Initializer()
-		bucket = llvm.ConstInsertValue(bucket, tophashValue, []uint32{0, uint32(i % 8)})
-		bucket = llvm.ConstInsertValue(bucket, llvmKey, []uint32{2, uint32(i % 8)})
-		bucket = llvm.ConstInsertValue(bucket, llvmValue, []uint32{3, uint32(i % 8)})
-		bucketGlobal.SetInitializer(bucket)
-	}
-
-	// Create the hashmap itself.
-	zero := llvm.ConstInt(ctx.Int32Type(), 0, false)
-	bucketPtr := llvm.ConstInBoundsGEP(firstBucketGlobal, []llvm.Value{zero})
-	hashmapType := v.Type()
-	hashmap := llvm.ConstNamedStruct(hashmapType, []llvm.Value{
-		llvm.ConstPointerNull(llvm.PointerType(hashmapType, 0)),                        // next
-		llvm.ConstBitCast(bucketPtr, i8ptrType),                                        // buckets
-		llvm.ConstInt(hashmapType.StructElementTypes()[2], uint64(len(v.Keys)), false), // count
-		llvm.ConstInt(ctx.Int8Type(), uint64(v.KeySize), false),                        // keySize
-		llvm.ConstInt(ctx.Int8Type(), uint64(v.ValueSize), false),                      // valueSize
-		llvm.ConstInt(ctx.Int8Type(), 0, false),                                        // bucketBits
-	})
-
-	// Create a pointer to this hashmap.
-	hashmapPtr := llvm.AddGlobal(v.Eval.Mod, hashmap.Type(), v.PkgName+"$map")
-	hashmapPtr.SetInitializer(hashmap)
-	hashmapPtr.SetLinkage(llvm.InternalLinkage)
-	hashmapPtr.SetUnnamedAddr(true)
-	v.Underlying = llvm.ConstInBoundsGEP(hashmapPtr, []llvm.Value{zero})
-	return v.Underlying
-}
-
-// Type returns type runtime.hashmap, which is the actual hashmap type.
-func (v *MapValue) Type() llvm.Type {
-	return v.Eval.Mod.GetTypeByName("runtime.hashmap")
-}
-
-func (v *MapValue) IsConstant() bool {
-	return true // TODO: dirty maps
-}
-
-// Load panics: maps are of reference type so cannot be dereferenced.
-func (v *MapValue) Load() llvm.Value {
-	panic("interp: load from a map")
-}
-
-// Store panics: maps are of reference type so cannot be stored to.
-func (v *MapValue) Store(value llvm.Value) {
-	panic("interp: store on a map")
-}
-
-// GetElementPtr panics: maps are of reference type so their (interior)
-// addresses cannot be calculated.
-func (v *MapValue) GetElementPtr(indices []uint32) (Value, error) {
-	return nil, errors.New("interp: GEP on a map")
-}
-
-// PutString does a map assign operation, assuming that the map is of type
-// map[string]T.
-func (v *MapValue) PutString(keyBuf, keyLen, valPtr *LocalValue) {
-	if !v.Underlying.IsNil() {
-		panic("map already created")
-	}
-
-	if valPtr.Underlying.Opcode() == llvm.BitCast {
-		valPtr = &LocalValue{v.Eval, valPtr.Underlying.Operand(0)}
-	}
-	value := valPtr.Load()
-	if v.ValueType.IsNil() {
-		v.ValueType = value.Type()
-		if int(v.Eval.TargetData.TypeAllocSize(v.ValueType)) != v.ValueSize {
-			panic("interp: map store value type has the wrong size")
-		}
-	} else {
-		if value.Type() != v.ValueType {
-			panic("interp: map store value type is inconsistent")
-		}
-	}
-
-	keyType := v.Eval.Mod.GetTypeByName("runtime._string")
-	v.KeyType = keyType
-	key := llvm.ConstNull(keyType)
-	key = llvm.ConstInsertValue(key, keyBuf.Value(), []uint32{0})
-	key = llvm.ConstInsertValue(key, keyLen.Value(), []uint32{1})
-
-	// TODO: avoid duplicate keys
-	v.Keys = append(v.Keys, &LocalValue{v.Eval, key})
-	v.Values = append(v.Values, &LocalValue{v.Eval, value})
-}
-
-// PutBinary does a map assign operation.
-func (v *MapValue) PutBinary(keyPtr, valPtr *LocalValue) {
-	if !v.Underlying.IsNil() {
-		panic("map already created")
-	}
-
-	if valPtr.Underlying.Opcode() == llvm.BitCast {
-		valPtr = &LocalValue{v.Eval, valPtr.Underlying.Operand(0)}
-	}
-	value := valPtr.Load()
-	if v.ValueType.IsNil() {
-		v.ValueType = value.Type()
-		if int(v.Eval.TargetData.TypeAllocSize(v.ValueType)) != v.ValueSize {
-			panic("interp: map store value type has the wrong size")
-		}
-	} else {
-		if value.Type() != v.ValueType {
-			panic("interp: map store value type is inconsistent")
-		}
-	}
-
-	if !keyPtr.Underlying.IsAConstantExpr().IsNil() {
-		if keyPtr.Underlying.Opcode() == llvm.BitCast {
-			keyPtr = &LocalValue{v.Eval, keyPtr.Underlying.Operand(0)}
-		} else if keyPtr.Underlying.Opcode() == llvm.GetElementPtr {
-			keyPtr = &LocalValue{v.Eval, keyPtr.Underlying.Operand(0)}
-		}
-	}
-	key := keyPtr.Load()
-	if v.KeyType.IsNil() {
-		v.KeyType = key.Type()
-		if int(v.Eval.TargetData.TypeAllocSize(v.KeyType)) != v.KeySize {
-			panic("interp: map store key type has the wrong size")
-		}
-	} else {
-		if key.Type() != v.KeyType {
-			panic("interp: map store key type is inconsistent")
-		}
-	}
-
-	// TODO: avoid duplicate keys
-	v.Keys = append(v.Keys, &LocalValue{v.Eval, key})
-	v.Values = append(v.Values, &LocalValue{v.Eval, value})
-}
-
-// Get FNV-1a hash of this string.
-//
-// https://en.wikipedia.org/wiki/Fowler%E2%80%93Noll%E2%80%93Vo_hash_function#FNV-1a_hash
-func (v *MapValue) hash(data []byte) uint32 {
-	var result uint32 = 2166136261 // FNV offset basis
-	for _, c := range data {
-		result ^= uint32(c)
-		result *= 16777619 // FNV prime
-	}
-	return result
-}
-
-// Get the topmost 8 bits of the hash, without using a special value (like 0).
-func (v *MapValue) topHash(hash uint32) uint8 {
-	tophash := uint8(hash >> 24)
-	if tophash < 1 {
-		// 0 means empty slot, so make it bigger.
-		tophash += 1
-	}
-	return tophash
-}
-
-func (v *MapValue) String() string {
-	return "&MapValue{KeySize: " + strconv.Itoa(v.KeySize) + ", ValueSize: " + strconv.Itoa(v.ValueSize) + "}"
-}
@@ -67,7 +67,7 @@ func NewProgram(lprogram *loader.Program) *Program {
 	program := lprogram.LoadSSA()
 	program.Build()

-	mainPkg := program.ImportedPackage(lprogram.MainPkg.PkgPath)
+	mainPkg := program.ImportedPackage(lprogram.MainPkg().ImportPath)
 	if mainPkg == nil {
 		panic("could not find main package")
 	}
@@ -79,7 +79,7 @@ func NewProgram(lprogram *loader.Program) *Program {
 	}

 	for _, pkg := range lprogram.Sorted() {
-		p.AddPackage(program.ImportedPackage(pkg.PkgPath))
+		p.AddPackage(program.ImportedPackage(pkg.ImportPath))
 	}

 	return p
@@ -1,5 +1,7 @@
 package loader

+import "go/scanner"
+
 // Errors contains a list of parser errors or a list of typechecker errors for
 // the given package.
 type Errors struct {
@@ -10,3 +12,14 @@ type Errors struct {
 func (e Errors) Error() string {
 	return "could not compile: " + e.Errs[0].Error()
 }
+
+// Error is a regular error but with an added import stack. This is especially
+// useful for debugging import cycle errors.
+type Error struct {
+	ImportStack []string
+	Err         scanner.Error
+}
+
+func (e Error) Error() string {
+	return e.Err.Error()
+}
@@ -8,19 +8,21 @@ import (
 	"encoding/hex"
 	"errors"
 	"fmt"
+	"io"
 	"io/ioutil"
-	"math/rand"
 	"os"
 	"os/exec"
 	"path"
 	"path/filepath"
 	"runtime"
-	"strconv"
+	"sync"

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

+var gorootCreateMutex sync.Mutex
+
 // GetCachedGoroot creates a new GOROOT by merging both the standard GOROOT and
 // the GOROOT from TinyGo using lots of symbolic links.
 func GetCachedGoroot(config *compileopts.Config) (string, error) {
@@ -49,16 +51,29 @@ func GetCachedGoroot(config *compileopts.Config) (string, error) {
 	fmt.Fprintln(hash, tinygoroot)
 	gorootsHash := hash.Sum(nil)
 	gorootsHashHex := hex.EncodeToString(gorootsHash[:])
-	cachedgoroot := filepath.Join(goenv.Get("GOCACHE"), "goroot-"+version+"-"+gorootsHashHex)
+	cachedgorootName := "goroot-" + version + "-" + gorootsHashHex
+	cachedgoroot := filepath.Join(goenv.Get("GOCACHE"), cachedgorootName)
 	if needsSyscallPackage(config.BuildTags()) {
 		cachedgoroot += "-syscall"
 	}

+	// Do not try to create the cached GOROOT in parallel, that's only a waste
+	// of I/O bandwidth and thus speed. Instead, use a mutex to make sure only
+	// one goroutine does it at a time.
+	// This is not a way to ensure atomicity (a different TinyGo invocation
+	// could be creating the same directory), but instead a way to avoid
+	// creating it many times in parallel when running tests in parallel.
+	gorootCreateMutex.Lock()
+	defer gorootCreateMutex.Unlock()
+
 	if _, err := os.Stat(cachedgoroot); err == nil {
 		return cachedgoroot, nil
 	}
-	tmpgoroot := cachedgoroot + ".tmp" + strconv.Itoa(rand.Int())
-	err = os.MkdirAll(tmpgoroot, 0777)
+	err = os.MkdirAll(goenv.Get("GOCACHE"), 0777)
+	if err != nil {
+		return "", err
+	}
+	tmpgoroot, err := ioutil.TempDir(goenv.Get("GOCACHE"), cachedgorootName+".tmp")
 	if err != nil {
 		return "", err
 	}
@@ -85,6 +100,15 @@ func GetCachedGoroot(config *compileopts.Config) (string, error) {
 			// deleted by the defer above.
 			return cachedgoroot, nil
 		}
+		if runtime.GOOS == "windows" && os.IsPermission(err) {
+			// On Windows, a rename with a destination directory that already
+			// exists does not result in an IsExist error, but rather in an
+			// access denied error. To be sure, check for this case by checking
+			// whether the target directory exists.
+			if _, err := os.Stat(cachedgoroot); err == nil {
+				return cachedgoroot, nil
+			}
+		}
 		return "", err
 	}
 	return cachedgoroot, nil
@@ -165,7 +189,7 @@ func mergeDirectory(goroot, tinygoroot, tmpgoroot, importPath string, overrides
 // with the TinyGo version. This is the case on some targets.
 func needsSyscallPackage(buildTags []string) bool {
 	for _, tag := range buildTags {
-		if tag == "baremetal" || tag == "darwin" || tag == "nintendoswitch" {
+		if tag == "baremetal" || tag == "darwin" || tag == "nintendoswitch" || tag == "wasi" {
 			return true
 		}
 	}
@@ -188,7 +212,7 @@ func pathsToOverride(needsSyscallPackage bool) map[string]bool {
 		"reflect/":              false,
 		"runtime/":              false,
 		"sync/":                 true,
-		"testing/":              false,
+		"testing/":              true,
 	}
 	if needsSyscallPackage {
 		paths["syscall/"] = true // include syscall/js
@@ -228,10 +252,27 @@ func symlink(oldname, newname string) error {
 				return symlinkErr
 			}
 		} else {
-			// Make a hard link.
+			// Try making a hard link.
 			err := os.Link(oldname, newname)
 			if err != nil {
-				return symlinkErr
+				// Making a hardlink failed. Try copying the file as a last
+				// fallback.
+				inf, err := os.Open(oldname)
+				if err != nil {
+					return err
+				}
+				defer inf.Close()
+				outf, err := os.Create(newname)
+				if err != nil {
+					return err
+				}
+				defer outf.Close()
+				_, err = io.Copy(outf, inf)
+				if err != nil {
+					os.Remove(newname)
+					return err
+				}
+				// File was copied.
 			}
 		}
 		return nil // success
@@ -0,0 +1,30 @@
+package loader
+
+import (
+	"os"
+	"os/exec"
+	"strings"
+
+	"github.com/tinygo-org/tinygo/compileopts"
+)
+
+// List returns a ready-to-run *exec.Cmd for running the `go list` command with
+// the configuration used for TinyGo.
+func List(config *compileopts.Config, extraArgs, pkgs []string) (*exec.Cmd, error) {
+	goroot, err := GetCachedGoroot(config)
+	if err != nil {
+		return nil, err
+	}
+	args := append([]string{"list"}, extraArgs...)
+	if len(config.BuildTags()) != 0 {
+		args = append(args, "-tags", strings.Join(config.BuildTags(), " "))
+	}
+	args = append(args, pkgs...)
+	cgoEnabled := "0"
+	if config.CgoEnabled() {
+		cgoEnabled = "1"
+	}
+	cmd := exec.Command("go", args...)
+	cmd.Env = append(os.Environ(), "GOROOT="+goroot, "GOOS="+config.GOOS(), "GOARCH="+config.GOARCH(), "CGO_ENABLED="+cgoEnabled)
+	return cmd, nil
+}
@@ -2,165 +2,212 @@ package loader

 import (
 	"bytes"
+	"encoding/json"
 	"errors"
 	"fmt"
 	"go/ast"
-	"go/build"
 	"go/parser"
 	"go/scanner"
 	"go/token"
 	"go/types"
+	"io"
 	"os"
+	"os/exec"
 	"path/filepath"
-	"sort"
 	"strconv"
 	"strings"
-	"text/template"
+	"syscall"

 	"github.com/tinygo-org/tinygo/cgo"
+	"github.com/tinygo-org/tinygo/compileopts"
 	"github.com/tinygo-org/tinygo/goenv"
-	"golang.org/x/tools/go/packages"
 )

 // Program holds all packages and some metadata about the program as a whole.
 type Program struct {
-	Build        *build.Context
-	Tests        bool
-	Packages     map[string]*Package
-	MainPkg      *Package
-	sorted       []*Package
-	fset         *token.FileSet
-	TypeChecker  types.Config
-	Dir          string // current working directory (for error reporting)
-	TINYGOROOT   string // root of the TinyGo installation or root of the source code
-	CFlags       []string
-	LDFlags      []string
-	ClangHeaders string
+	config       *compileopts.Config
+	clangHeaders string
+	typeChecker  types.Config
+	goroot       string // synthetic GOROOT
+	workingDir   string
+
+	Packages map[string]*Package
+	sorted   []*Package
+	fset     *token.FileSet
+
+	// Information obtained during parsing.
+	LDFlags []string
+}
+
+// PackageJSON is a subset of the JSON struct returned from `go list`.
+type PackageJSON struct {
+	Dir        string
+	ImportPath string
+	ForTest    string
+
+	// Source files
+	GoFiles  []string
+	CgoFiles []string
+	CFiles   []string
+
+	// Dependency information
+	Imports []string
+
+	// Error information
+	Error *struct {
+		ImportStack []string
+		Pos         string
+		Err         string
+	}
 }

 // Package holds a loaded package, its imports, and its parsed files.
 type Package struct {
-	*Program
-	*packages.Package
-	Files []*ast.File
-	Pkg   *types.Package
-	types.Info
+	PackageJSON
+
+	program *Program
+	Files   []*ast.File
+	Pkg     *types.Package
+	info    types.Info
 }

 // Load loads the given package with all dependencies (including the runtime
 // package). Call .Parse() afterwards to parse all Go files (including CGo
 // processing, if necessary).
-func (p *Program) Load(importPath string) error {
-	if p.Packages == nil {
-		p.Packages = make(map[string]*Package)
+func Load(config *compileopts.Config, inputPkgs []string, clangHeaders string, typeChecker types.Config) (*Program, error) {
+	goroot, err := GetCachedGoroot(config)
+	if err != nil {
+		return nil, err
+	}
+	wd, err := os.Getwd()
+	if err != nil {
+		return nil, err
+	}
+	p := &Program{
+		config:       config,
+		clangHeaders: clangHeaders,
+		typeChecker:  typeChecker,
+		goroot:       goroot,
+		workingDir:   wd,
+		Packages:     make(map[string]*Package),
+		fset:         token.NewFileSet(),
 	}

-	err := p.loadPackage(importPath)
+	// List the dependencies of this package, in raw JSON format.
+	extraArgs := []string{"-json", "-deps"}
+	if config.TestConfig.CompileTestBinary {
+		extraArgs = append(extraArgs, "-test")
+	}
+	cmd, err := List(config, extraArgs, inputPkgs)
 	if err != nil {
-		return err
+		return nil, err
 	}
-	p.MainPkg = p.sorted[len(p.sorted)-1]
-	if _, ok := p.Packages["runtime"]; !ok {
-		// The runtime package wasn't loaded. Although `go list -deps` seems to
-		// return the full dependency list, there is no way to get those
-		// packages from the go/packages package. Therefore load the runtime
-		// manually and add it to the list of to-be-compiled packages
-		// (duplicates are already filtered).
-		return p.loadPackage("runtime")
+	buf := &bytes.Buffer{}
+	cmd.Stdout = buf
+	cmd.Stderr = os.Stderr
+	err = cmd.Run()
+	if err != nil {
+		if exitErr, ok := err.(*exec.ExitError); ok {
+			if status, ok := exitErr.Sys().(syscall.WaitStatus); ok {
+				os.Exit(status.ExitStatus())
+			}
+			os.Exit(1)
+		}
+		return nil, fmt.Errorf("failed to run `go list`: %s", err)
 	}
-	return nil
-}

-func (p *Program) loadPackage(importPath string) error {
-	cgoEnabled := "0"
-	if p.Build.CgoEnabled {
-		cgoEnabled = "1"
-	}
-	pkgs, err := packages.Load(&packages.Config{
-		Mode:       packages.NeedName | packages.NeedFiles | packages.NeedImports | packages.NeedDeps,
-		Env:        append(os.Environ(), "GOROOT="+p.Build.GOROOT, "GOOS="+p.Build.GOOS, "GOARCH="+p.Build.GOARCH, "CGO_ENABLED="+cgoEnabled),
-		BuildFlags: []string{"-tags", strings.Join(p.Build.BuildTags, " ")},
-		Tests:      p.Tests,
-	}, importPath)
-	if err != nil {
-		return err
-	}
-	var pkg *packages.Package
-	if p.Tests {
-		// We need the second package. Quoting from the docs:
-		// > For example, when using the go command, loading "fmt" with Tests=true
-		// > returns four packages, with IDs "fmt" (the standard package),
-		// > "fmt [fmt.test]" (the package as compiled for the test),
-		// > "fmt_test" (the test functions from source files in package fmt_test),
-		// > and "fmt.test" (the test binary).
-		pkg = pkgs[1]
-	} else {
-		if len(pkgs) != 1 {
-			return fmt.Errorf("expected exactly one package while importing %s, got %d", importPath, len(pkgs))
+	// Parse the returned json from `go list`.
+	decoder := json.NewDecoder(buf)
+	for {
+		pkg := &Package{
+			program: p,
+			info: types.Info{
+				Types:      make(map[ast.Expr]types.TypeAndValue),
+				Defs:       make(map[*ast.Ident]types.Object),
+				Uses:       make(map[*ast.Ident]types.Object),
+				Implicits:  make(map[ast.Node]types.Object),
+				Scopes:     make(map[ast.Node]*types.Scope),
+				Selections: make(map[*ast.SelectorExpr]*types.Selection),
+			},
 		}
-		pkg = pkgs[0]
-	}
-	var importError *Errors
-	var addPackages func(pkg *packages.Package)
-	addPackages = func(pkg *packages.Package) {
-		if _, ok := p.Packages[pkg.PkgPath]; ok {
-			return
+		err := decoder.Decode(&pkg.PackageJSON)
+		if err != nil {
+			if err == io.EOF {
+				break
+			}
+			return nil, err
 		}
-		pkg2 := p.newPackage(pkg)
-		p.Packages[pkg.PkgPath] = pkg2
-		if len(pkg.Errors) != 0 {
-			if importError != nil {
-				// There was another error reported already. Do not report
-				// errors from multiple packages at once.
-				return
-			}
-			importError = &Errors{
-				Pkg: pkg2,
-			}
-			for _, err := range pkg.Errors {
-				pos := token.Position{}
-				fields := strings.Split(err.Pos, ":")
-				if len(fields) >= 2 {
-					// There is some file/line/column information.
-					if n, err := strconv.Atoi(fields[len(fields)-2]); err == nil {
-						// Format: filename.go:line:colum
-						pos.Filename = strings.Join(fields[:len(fields)-2], ":")
-						pos.Line = n
-						pos.Column, _ = strconv.Atoi(fields[len(fields)-1])
-					} else {
-						// Format: filename.go:line
-						pos.Filename = strings.Join(fields[:len(fields)-1], ":")
-						pos.Line, _ = strconv.Atoi(fields[len(fields)-1])
-					}
-					pos.Filename = p.getOriginalPath(pos.Filename)
+		if pkg.Error != nil {
+			// There was an error while importing (for example, a circular
+			// dependency).
+			pos := token.Position{}
+			fields := strings.Split(pkg.Error.Pos, ":")
+			if len(fields) >= 2 {
+				// There is some file/line/column information.
+				if n, err := strconv.Atoi(fields[len(fields)-2]); err == nil {
+					// Format: filename.go:line:colum
+					pos.Filename = strings.Join(fields[:len(fields)-2], ":")
+					pos.Line = n
+					pos.Column, _ = strconv.Atoi(fields[len(fields)-1])
+				} else {
+					// Format: filename.go:line
+					pos.Filename = strings.Join(fields[:len(fields)-1], ":")
+					pos.Line, _ = strconv.Atoi(fields[len(fields)-1])
 				}
-				importError.Errs = append(importError.Errs, scanner.Error{
-					Pos: pos,
-					Msg: err.Msg,
-				})
+				pos.Filename = p.getOriginalPath(pos.Filename)
 			}
-			return
+			err := scanner.Error{
+				Pos: pos,
+				Msg: pkg.Error.Err,
+			}
+			if len(pkg.Error.ImportStack) != 0 {
+				return nil, Error{
+					ImportStack: pkg.Error.ImportStack,
+					Err:         err,
+				}
+			}
+			return nil, err
 		}
-
-		// Get the list of imports (sorted alphabetically).
-		names := make([]string, 0, len(pkg.Imports))
-		for name := range pkg.Imports {
-			names = append(names, name)
+		if config.TestConfig.CompileTestBinary {
+			// When creating a test binary, `go list` will list two or three
+			// packages used for testing the package. The first is the original
+			// package as if it were built normally, the second is the same
+			// package but with the *_test.go files included. A possible third
+			// may be included for _test packages (such as math_test), used to
+			// test the external API with no access to internal functions.
+			// All packages that are necessary for testing (including the to be
+			// tested package with *_test.go files, but excluding the original
+			// unmodified package) have a suffix added to the import path, for
+			// example the math package has import path "math [math.test]" and
+			// test dependencies such as fmt will have an import path of the
+			// form "fmt [math.test]".
+			// The code below removes this suffix, and if this results in a
+			// duplicate (which happens with the to-be-tested package without
+			// *.test.go files) the previous package is removed from the list of
+			// packages included in this build.
+			// This is necessary because the change in import paths results in
+			// breakage to //go:linkname. Additionally, the duplicated package
+			// slows down the build and so is best removed.
+			if pkg.ForTest != "" && strings.HasSuffix(pkg.ImportPath, " ["+pkg.ForTest+".test]") {
+				newImportPath := pkg.ImportPath[:len(pkg.ImportPath)-len(" ["+pkg.ForTest+".test]")]
+				if _, ok := p.Packages[newImportPath]; ok {
+					// Delete the previous package (that this package overrides).
+					delete(p.Packages, newImportPath)
+					for i, pkg := range p.sorted {
+						if pkg.ImportPath == newImportPath {
+							p.sorted = append(p.sorted[:i], p.sorted[i+1:]...) // remove element from slice
+							break
+						}
+					}
+				}
+				pkg.ImportPath = newImportPath
+			}
 		}
-		sort.Strings(names)
-
-		// Add all the imports.
-		for _, name := range names {
-			addPackages(pkg.Imports[name])
-		}
-
-		p.sorted = append(p.sorted, pkg2)
+		p.sorted = append(p.sorted, pkg)
+		p.Packages[pkg.ImportPath] = pkg
 	}
-	addPackages(pkg)
-	if importError != nil {
-		return *importError
-	}
-	return nil
+
+	return p, nil
 }

 // getOriginalPath looks whether this path is in the generated GOROOT and if so,
@@ -168,23 +215,23 @@ func (p *Program) loadPackage(importPath string) error {
 // the input path is returned.
 func (p *Program) getOriginalPath(path string) string {
 	originalPath := path
-	if strings.HasPrefix(path, p.Build.GOROOT+string(filepath.Separator)) {
+	if strings.HasPrefix(path, p.goroot+string(filepath.Separator)) {
 		// If this file is part of the synthetic GOROOT, try to infer the
 		// original path.
-		relpath := path[len(filepath.Join(p.Build.GOROOT, "src"))+1:]
+		relpath := path[len(filepath.Join(p.goroot, "src"))+1:]
 		realgorootPath := filepath.Join(goenv.Get("GOROOT"), "src", relpath)
 		if _, err := os.Stat(realgorootPath); err == nil {
 			originalPath = realgorootPath
 		}
 		maybeInTinyGoRoot := false
-		for prefix := range pathsToOverride(needsSyscallPackage(p.Build.BuildTags)) {
+		for prefix := range pathsToOverride(needsSyscallPackage(p.config.BuildTags())) {
 			if !strings.HasPrefix(relpath, prefix) {
 				continue
 			}
 			maybeInTinyGoRoot = true
 		}
 		if maybeInTinyGoRoot {
-			tinygoPath := filepath.Join(p.TINYGOROOT, "src", relpath)
+			tinygoPath := filepath.Join(goenv.Get("TINYGOROOT"), "src", relpath)
 			if _, err := os.Stat(tinygoPath); err == nil {
 				originalPath = tinygoPath
 			}
@@ -193,28 +240,18 @@ func (p *Program) getOriginalPath(path string) string {
 	return originalPath
 }

-// newPackage instantiates a new *Package object with initialized members.
-func (p *Program) newPackage(pkg *packages.Package) *Package {
-	return &Package{
-		Program: p,
-		Package: pkg,
-		Info: types.Info{
-			Types:      make(map[ast.Expr]types.TypeAndValue),
-			Defs:       make(map[*ast.Ident]types.Object),
-			Uses:       make(map[*ast.Ident]types.Object),
-			Implicits:  make(map[ast.Node]types.Object),
-			Scopes:     make(map[ast.Node]*types.Scope),
-			Selections: make(map[*ast.SelectorExpr]*types.Selection),
-		},
-	}
-}
-
 // Sorted returns a list of all packages, sorted in a way that no packages come
 // before the packages they depend upon.
 func (p *Program) Sorted() []*Package {
 	return p.sorted
 }

+// MainPkg returns the last package in the Sorted() slice. This is the main
+// package of the program.
+func (p *Program) MainPkg() *Package {
+	return p.sorted[len(p.sorted)-1]
+}
+
 // Parse parses all packages and typechecks them.
 //
 // The returned error may be an Errors error, which contains a list of errors.
@@ -222,22 +259,16 @@ func (p *Program) Sorted() []*Package {
 // Idempotent.
 func (p *Program) Parse() error {
 	// Parse all packages.
-	for _, pkg := range p.Sorted() {
+	// TODO: do this in parallel.
+	for _, pkg := range p.sorted {
 		err := pkg.Parse()
 		if err != nil {
 			return err
 		}
 	}

-	if p.Tests {
-		err := p.swapTestMain()
-		if err != nil {
-			return err
-		}
-	}
-
 	// Typecheck all packages.
-	for _, pkg := range p.Sorted() {
+	for _, pkg := range p.sorted {
 		err := pkg.Check()
 		if err != nil {
 			return err
@@ -247,82 +278,6 @@ func (p *Program) Parse() error {
 	return nil
 }

-func (p *Program) swapTestMain() error {
-	var tests []string
-
-	isTestFunc := func(f *ast.FuncDecl) bool {
-		// TODO: improve signature check
-		if strings.HasPrefix(f.Name.Name, "Test") && f.Name.Name != "TestMain" {
-			return true
-		}
-		return false
-	}
-	for _, f := range p.MainPkg.Files {
-		for i, d := range f.Decls {
-			switch v := d.(type) {
-			case *ast.FuncDecl:
-				if isTestFunc(v) {
-					tests = append(tests, v.Name.Name)
-				}
-				if v.Name.Name == "main" {
-					// Remove main
-					if len(f.Decls) == 1 {
-						f.Decls = make([]ast.Decl, 0)
-					} else {
-						f.Decls[i] = f.Decls[len(f.Decls)-1]
-						f.Decls = f.Decls[:len(f.Decls)-1]
-					}
-				}
-			}
-		}
-	}
-
-	// TODO: Check if they defined a TestMain and call it instead of testing.TestMain
-	const mainBody = `package main
-
-import (
-	"testing"
-)
-
-func main () {
-	m := &testing.M{
-		Tests: []testing.TestToCall{
-{{range .TestFunctions}}
-			{Name: "{{.}}", Func: {{.}}},
-{{end}}
-		},
-	}
-
-	testing.TestMain(m)
-}
-`
-	tmpl := template.Must(template.New("testmain").Parse(mainBody))
-	b := bytes.Buffer{}
-	tmplData := struct {
-		TestFunctions []string
-	}{
-		TestFunctions: tests,
-	}
-
-	err := tmpl.Execute(&b, tmplData)
-	if err != nil {
-		return err
-	}
-	path := filepath.Join(p.MainPkg.Dir, "$testmain.go")
-
-	if p.fset == nil {
-		p.fset = token.NewFileSet()
-	}
-
-	newMain, err := parser.ParseFile(p.fset, path, b.Bytes(), parser.AllErrors)
-	if err != nil {
-		return err
-	}
-	p.MainPkg.Files = append(p.MainPkg.Files, newMain)
-
-	return nil
-}
-
 // parseFile is a wrapper around parser.ParseFile.
 func (p *Program) parseFile(path string, mode parser.Mode) (*ast.File, error) {
 	if p.fset == nil {
@@ -342,14 +297,13 @@ func (p *Program) parseFile(path string, mode parser.Mode) (*ast.File, error) {
 // Idempotent.
 func (p *Package) Parse() error {
 	if len(p.Files) != 0 {
-		return nil
+		return nil // nothing to do (?)
 	}

 	// Load the AST.
-	// TODO: do this in parallel.
-	if p.PkgPath == "unsafe" {
-		// Special case for the unsafe package. Don't even bother loading
-		// the files.
+	if p.ImportPath == "unsafe" {
+		// Special case for the unsafe package, which is defined internally by
+		// the types package.
 		p.Pkg = types.Unsafe
 		return nil
 	}
@@ -369,11 +323,11 @@ func (p *Package) Parse() error {
 // Idempotent.
 func (p *Package) Check() error {
 	if p.Pkg != nil {
-		return nil
+		return nil // already typechecked
 	}

 	var typeErrors []error
-	checker := p.TypeChecker
+	checker := p.program.typeChecker // make a copy, because it will be modified
 	checker.Error = func(err error) {
 		typeErrors = append(typeErrors, err)
 	}
@@ -381,7 +335,7 @@ func (p *Package) Check() error {
 	// Do typechecking of the package.
 	checker.Importer = p

-	typesPkg, err := checker.Check(p.PkgPath, p.fset, p.Files, &p.Info)
+	typesPkg, err := checker.Check(p.ImportPath, p.program.fset, p.Files, &p.info)
 	if err != nil {
 		if err, ok := err.(Errors); ok {
 			return err
@@ -394,40 +348,46 @@ func (p *Package) Check() error {

 // parseFiles parses the loaded list of files and returns this list.
 func (p *Package) parseFiles() ([]*ast.File, error) {
-	// TODO: do this concurrently.
 	var files []*ast.File
 	var fileErrs []error

-	var cgoFiles []*ast.File
-	for _, file := range p.GoFiles {
-		f, err := p.parseFile(file, parser.ParseComments)
+	// Parse all files (incuding CgoFiles).
+	parseFile := func(file string) {
+		if !filepath.IsAbs(file) {
+			file = filepath.Join(p.Dir, file)
+		}
+		f, err := p.program.parseFile(file, parser.ParseComments)
 		if err != nil {
 			fileErrs = append(fileErrs, err)
-			continue
-		}
-		if err != nil {
-			fileErrs = append(fileErrs, err)
-			continue
-		}
-		for _, importSpec := range f.Imports {
-			if importSpec.Path.Value == `"C"` {
-				cgoFiles = append(cgoFiles, f)
-			}
+			return
 		}
 		files = append(files, f)
 	}
-	if len(cgoFiles) != 0 {
-		cflags := append(p.CFlags, "-I"+filepath.Dir(p.GoFiles[0]))
-		if p.ClangHeaders != "" {
-			cflags = append(cflags, "-Xclang", "-internal-isystem", "-Xclang", p.ClangHeaders)
+	for _, file := range p.GoFiles {
+		parseFile(file)
+	}
+	for _, file := range p.CgoFiles {
+		parseFile(file)
+	}
+
+	// Do CGo processing.
+	if len(p.CgoFiles) != 0 {
+		var cflags []string
+		cflags = append(cflags, p.program.config.CFlags()...)
+		cflags = append(cflags, "-I"+p.Dir)
+		if p.program.clangHeaders != "" {
+			cflags = append(cflags, "-Xclang", "-internal-isystem", "-Xclang", p.program.clangHeaders)
 		}
-		generated, ldflags, errs := cgo.Process(files, p.Program.Dir, p.fset, cflags)
+		generated, ldflags, errs := cgo.Process(files, p.program.workingDir, p.program.fset, cflags)
 		if errs != nil {
 			fileErrs = append(fileErrs, errs...)
 		}
 		files = append(files, generated)
-		p.LDFlags = append(p.LDFlags, ldflags...)
+		p.program.LDFlags = append(p.program.LDFlags, ldflags...)
 	}
+
+	// Only return an error after CGo processing, so that errors in parsing and
+	// CGo can be reported together.
 	if len(fileErrs) != 0 {
 		return nil, Errors{p, fileErrs}
 	}
@@ -441,8 +401,8 @@ func (p *Package) Import(to string) (*types.Package, error) {
 	if to == "unsafe" {
 		return types.Unsafe, nil
 	}
-	if _, ok := p.Imports[to]; ok {
-		return p.Packages[p.Imports[to].PkgPath].Pkg, nil
+	if imported, ok := p.program.Packages[to]; ok {
+		return imported.Pkg, nil
 	} else {
 		return nil, errors.New("package not imported: " + to)
 	}
@@ -10,8 +10,8 @@ import (
 func (p *Program) LoadSSA() *ssa.Program {
 	prog := ssa.NewProgram(p.fset, ssa.SanityCheckFunctions|ssa.BareInits|ssa.GlobalDebug)

-	for _, pkg := range p.Sorted() {
-		prog.CreatePackage(pkg.Pkg, pkg.Files, &pkg.Info, true)
+	for _, pkg := range p.sorted {
+		prog.CreatePackage(pkg.Pkg, pkg.Files, &pkg.info, true)
 	}

 	return prog
@@ -8,6 +8,7 @@ import (
 	"go/scanner"
 	"go/types"
 	"io"
+	"io/ioutil"
 	"os"
 	"os/exec"
 	"os/signal"
@@ -28,6 +29,12 @@ import (
 	"go.bug.st/serial"
 )

+var (
+	// This variable is set at build time using -ldflags parameters.
+	// See: https://stackoverflow.com/a/11355611
+	gitSha1 string
+)
+
 // commandError is an error type to wrap os/exec.Command errors. This provides
 // some more information regarding what went wrong while running a command.
 type commandError struct {
@@ -58,33 +65,28 @@ func moveFile(src, dst string) error {
 	return os.Remove(src)
 }

-// copyFile copies the given file from src to dst. It copies first to a .tmp
-// file which is then moved over a possibly already existing file at the
-// destination.
+// 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)
+	source, err := os.Open(src)
 	if err != nil {
 		return err
 	}
-	defer inf.Close()
-	outpath := dst + ".tmp"
-	outf, err := os.Create(outpath)
+	defer source.Close()
+
+	st, err := source.Stat()
 	if err != nil {
 		return err
 	}

-	_, err = io.Copy(outf, inf)
-	if err != nil {
-		os.Remove(outpath)
-		return err
-	}
-
-	err = outf.Close()
+	destination, err := os.OpenFile(dst, os.O_RDWR|os.O_CREATE|os.O_TRUNC, st.Mode())
 	if err != nil {
 		return err
 	}
+	defer destination.Close()

-	return os.Rename(dst+".tmp", dst)
+	_, err = io.Copy(destination, source)
+	return err
 }

 // Build compiles and links the given package and writes it to outpath.
@@ -94,10 +96,10 @@ func Build(pkgName, outpath string, options *compileopts.Options) error {
 		return err
 	}

-	return builder.Build(pkgName, outpath, config, func(tmppath string) error {
-		if err := os.Rename(tmppath, outpath); err != nil {
+	return builder.Build(pkgName, outpath, config, func(result builder.BuildResult) error {
+		if err := os.Rename(result.Binary, outpath); err != nil {
 			// Moving failed. Do a file copy.
-			inf, err := os.Open(tmppath)
+			inf, err := os.Open(result.Binary)
 			if err != nil {
 				return err
 			}
@@ -123,35 +125,71 @@ func Build(pkgName, outpath string, options *compileopts.Options) error {
 }

 // Test runs the tests in the given package.
-func Test(pkgName string, options *compileopts.Options) error {
+func Test(pkgName string, options *compileopts.Options, testCompileOnly bool, outpath string) error {
 	options.TestConfig.CompileTestBinary = true
 	config, err := builder.NewConfig(options)
 	if err != nil {
 		return err
 	}

-	// Add test build tag. This is incorrect: `go test` only looks at the
-	// _test.go file suffix but does not add the test build tag in the process.
-	// However, it's a simple fix right now.
-	// For details: https://github.com/golang/go/issues/21360
-	config.Target.BuildTags = append(config.Target.BuildTags, "test")
-
-	return builder.Build(pkgName, ".elf", config, func(tmppath string) error {
-		cmd := exec.Command(tmppath)
-		cmd.Stdout = os.Stdout
-		cmd.Stderr = os.Stderr
-		err := cmd.Run()
-		if err != nil {
-			// Propagate the exit code
-			if err, ok := err.(*exec.ExitError); ok {
-				if status, ok := err.Sys().(syscall.WaitStatus); ok {
-					os.Exit(status.ExitStatus())
-				}
-				os.Exit(1)
+	return builder.Build(pkgName, outpath, config, func(result builder.BuildResult) error {
+		if testCompileOnly || outpath != "" {
+			// Write test binary to the specified file name.
+			if outpath == "" {
+				// No -o path was given, so create one now.
+				// This matches the behavior of go test.
+				outpath = filepath.Base(result.MainDir) + ".test"
+			}
+			copyFile(result.Binary, outpath)
+		}
+		if testCompileOnly {
+			// Do not run the test.
+			return nil
+		}
+		if len(config.Target.Emulator) == 0 {
+			// Run directly.
+			cmd := exec.Command(result.Binary)
+			cmd.Stdout = os.Stdout
+			cmd.Stderr = os.Stderr
+			cmd.Dir = result.MainDir
+			err := cmd.Run()
+			if err != nil {
+				// Propagate the exit code
+				if err, ok := err.(*exec.ExitError); ok {
+					if status, ok := err.Sys().(syscall.WaitStatus); ok {
+						os.Exit(status.ExitStatus())
+					}
+					os.Exit(1)
+				}
+				return &commandError{"failed to run compiled binary", result.Binary, err}
+			}
+			return nil
+		} else {
+			// Run in an emulator.
+			args := append(config.Target.Emulator[1:], result.Binary)
+			cmd := exec.Command(config.Target.Emulator[0], args...)
+			buf := &bytes.Buffer{}
+			w := io.MultiWriter(os.Stdout, buf)
+			cmd.Stdout = w
+			cmd.Stderr = os.Stderr
+			err := cmd.Run()
+			if err != nil {
+				if err, ok := err.(*exec.ExitError); !ok || !err.Exited() {
+					// Workaround for QEMU which always exits with an error.
+					return &commandError{"failed to run emulator with", result.Binary, err}
+				}
+			}
+			testOutput := string(buf.Bytes())
+			if testOutput == "PASS\n" || strings.HasSuffix(testOutput, "\nPASS\n") {
+				// Test passed.
+				return nil
+			} else {
+				// Test failed, either by ending with the word "FAIL" or with a
+				// panic of some sort.
+				os.Exit(1)
+				return nil // unreachable
 			}
-			return &commandError{"failed to run compiled binary", tmppath, err}
 		}
-		return nil
 	})
 }

@@ -193,9 +231,9 @@ func Flash(pkgName, port string, options *compileopts.Options) error {
 		return errors.New("unknown flash method: " + flashMethod)
 	}

-	return builder.Build(pkgName, fileExt, config, func(tmppath string) error {
+	return builder.Build(pkgName, fileExt, config, func(result builder.BuildResult) error {
 		// do we need port reset to put MCU into bootloader mode?
-		if config.Target.PortReset == "true" {
+		if config.Target.PortReset == "true" && flashMethod != "openocd" {
 			if port == "" {
 				var err error
 				port, err = getDefaultPort()
@@ -206,7 +244,7 @@ func Flash(pkgName, port string, options *compileopts.Options) error {

 			err := touchSerialPortAt1200bps(port)
 			if err != nil {
-				return &commandError{"failed to reset port", tmppath, err}
+				return &commandError{"failed to reset port", result.Binary, err}
 			}
 			// give the target MCU a chance to restart into bootloader
 			time.Sleep(3 * time.Second)
@@ -218,7 +256,7 @@ func Flash(pkgName, port string, options *compileopts.Options) error {
 			// Create the command.
 			flashCmd := config.Target.FlashCommand
 			fileToken := "{" + fileExt[1:] + "}"
-			flashCmd = strings.Replace(flashCmd, fileToken, tmppath, -1)
+			flashCmd = strings.Replace(flashCmd, fileToken, result.Binary, -1)

 			if port == "" && strings.Contains(flashCmd, "{port}") {
 				var err error
@@ -248,21 +286,21 @@ func Flash(pkgName, port string, options *compileopts.Options) error {
 			cmd.Dir = goenv.Get("TINYGOROOT")
 			err := cmd.Run()
 			if err != nil {
-				return &commandError{"failed to flash", tmppath, err}
+				return &commandError{"failed to flash", result.Binary, err}
 			}
 			return nil
 		case "msd":
 			switch fileExt {
 			case ".uf2":
-				err := flashUF2UsingMSD(config.Target.FlashVolume, tmppath)
+				err := flashUF2UsingMSD(config.Target.FlashVolume, result.Binary)
 				if err != nil {
-					return &commandError{"failed to flash", tmppath, err}
+					return &commandError{"failed to flash", result.Binary, err}
 				}
 				return nil
 			case ".hex":
-				err := flashHexUsingMSD(config.Target.FlashVolume, tmppath)
+				err := flashHexUsingMSD(config.Target.FlashVolume, result.Binary)
 				if err != nil {
-					return &commandError{"failed to flash", tmppath, err}
+					return &commandError{"failed to flash", result.Binary, err}
 				}
 				return nil
 			default:
@@ -273,13 +311,13 @@ func Flash(pkgName, port string, options *compileopts.Options) error {
 			if err != nil {
 				return err
 			}
-			args = append(args, "-c", "program "+tmppath+" reset exit")
+			args = append(args, "-c", "program "+filepath.ToSlash(result.Binary)+" reset exit")
 			cmd := exec.Command("openocd", args...)
 			cmd.Stdout = os.Stdout
 			cmd.Stderr = os.Stderr
 			err = cmd.Run()
 			if err != nil {
-				return &commandError{"failed to flash", tmppath, err}
+				return &commandError{"failed to flash", result.Binary, err}
 			}
 			return nil
 		default:
@@ -304,22 +342,28 @@ func FlashGDB(pkgName string, ocdOutput bool, options *compileopts.Options) erro
 		return errors.New("gdb not configured in the target specification")
 	}

-	return builder.Build(pkgName, "", config, func(tmppath string) error {
+	return builder.Build(pkgName, "", config, func(result builder.BuildResult) error {
 		// Find a good way to run GDB.
 		gdbInterface, openocdInterface := config.Programmer()
 		switch gdbInterface {
 		case "msd", "command", "":
 			if len(config.Target.Emulator) != 0 {
-				// Assume QEMU as an emulator.
 				if config.Target.Emulator[0] == "mgba" {
 					gdbInterface = "mgba"
-				} else {
+				} else if config.Target.Emulator[0] == "simavr" {
+					gdbInterface = "simavr"
+				} else if strings.HasPrefix(config.Target.Emulator[0], "qemu-system-") {
 					gdbInterface = "qemu"
+				} else {
+					// Assume QEMU as an emulator.
+					gdbInterface = "qemu-user"
 				}
 			} else if openocdInterface != "" && config.Target.OpenOCDTarget != "" {
 				gdbInterface = "openocd"
 			} else if config.Target.JLinkDevice != "" {
 				gdbInterface = "jlink"
+			} else {
+				gdbInterface = "native"
 			}
 		}

@@ -367,7 +411,15 @@ func FlashGDB(pkgName string, ocdOutput bool, options *compileopts.Options) erro
 			gdbCommands = append(gdbCommands, "target remote :1234")

 			// Run in an emulator.
-			args := append(config.Target.Emulator[1:], tmppath, "-s", "-S")
+			args := append(config.Target.Emulator[1:], result.Binary, "-s", "-S")
+			daemon = exec.Command(config.Target.Emulator[0], args...)
+			daemon.Stdout = os.Stdout
+			daemon.Stderr = os.Stderr
+		case "qemu-user":
+			gdbCommands = append(gdbCommands, "target remote :1234")
+
+			// Run in an emulator.
+			args := append(config.Target.Emulator[1:], "-g", "1234", result.Binary)
 			daemon = exec.Command(config.Target.Emulator[0], args...)
 			daemon.Stdout = os.Stdout
 			daemon.Stderr = os.Stderr
@@ -375,7 +427,15 @@ func FlashGDB(pkgName string, ocdOutput bool, options *compileopts.Options) erro
 			gdbCommands = append(gdbCommands, "target remote :2345")

 			// Run in an emulator.
-			args := append(config.Target.Emulator[1:], tmppath, "-g")
+			args := append(config.Target.Emulator[1:], result.Binary, "-g")
+			daemon = exec.Command(config.Target.Emulator[0], args...)
+			daemon.Stdout = os.Stdout
+			daemon.Stderr = os.Stderr
+		case "simavr":
+			gdbCommands = append(gdbCommands, "target remote :1234")
+
+			// Run in an emulator.
+			args := append(config.Target.Emulator[1:], "-g", result.Binary)
 			daemon = exec.Command(config.Target.Emulator[0], args...)
 			daemon.Stdout = os.Stdout
 			daemon.Stderr = os.Stderr
@@ -413,7 +473,7 @@ func FlashGDB(pkgName string, ocdOutput bool, options *compileopts.Options) erro
 		// Construct and execute a gdb command.
 		// By default: gdb -ex run <binary>
 		// Exit GDB with Ctrl-D.
-		params := []string{tmppath}
+		params := []string{result.Binary}
 		for _, cmd := range gdbCommands {
 			params = append(params, "-ex", cmd)
 		}
@@ -423,7 +483,7 @@ func FlashGDB(pkgName string, ocdOutput bool, options *compileopts.Options) erro
 		cmd.Stderr = os.Stderr
 		err := cmd.Run()
 		if err != nil {
-			return &commandError{"failed to run gdb with", tmppath, err}
+			return &commandError{"failed to run gdb with", result.Binary, err}
 		}
 		return nil
 	})
@@ -439,10 +499,10 @@ func Run(pkgName string, options *compileopts.Options) error {
 		return err
 	}

-	return builder.Build(pkgName, ".elf", config, func(tmppath string) error {
+	return builder.Build(pkgName, ".elf", config, func(result builder.BuildResult) error {
 		if len(config.Target.Emulator) == 0 {
 			// Run directly.
-			cmd := exec.Command(tmppath)
+			cmd := exec.Command(result.Binary)
 			cmd.Stdout = os.Stdout
 			cmd.Stderr = os.Stderr
 			err := cmd.Run()
@@ -451,12 +511,12 @@ func Run(pkgName string, options *compileopts.Options) error {
 					// Workaround for QEMU which always exits with an error.
 					return nil
 				}
-				return &commandError{"failed to run compiled binary", tmppath, err}
+				return &commandError{"failed to run compiled binary", result.Binary, err}
 			}
 			return nil
 		} else {
 			// Run in an emulator.
-			args := append(config.Target.Emulator[1:], tmppath)
+			args := append(config.Target.Emulator[1:], result.Binary)
 			cmd := exec.Command(config.Target.Emulator[0], args...)
 			cmd.Stdout = os.Stdout
 			cmd.Stderr = os.Stderr
@@ -466,7 +526,7 @@ func Run(pkgName string, options *compileopts.Options) error {
 					// Workaround for QEMU which always exits with an error.
 					return nil
 				}
-				return &commandError{"failed to run emulator with", tmppath, err}
+				return &commandError{"failed to run emulator with", result.Binary, err}
 			}
 			return nil
 		}
@@ -652,36 +712,6 @@ func getDefaultPort() (port string, err error) {
 	return d[0], nil
 }

-// runGoList runs the `go list` command but using the configuration used for
-// TinyGo.
-func runGoList(config *compileopts.Config, flagJSON, flagDeps bool, pkgs []string) error {
-	goroot, err := loader.GetCachedGoroot(config)
-	if err != nil {
-		return err
-	}
-	args := []string{"list"}
-	if flagJSON {
-		args = append(args, "-json")
-	}
-	if flagDeps {
-		args = append(args, "-deps")
-	}
-	if len(config.BuildTags()) != 0 {
-		args = append(args, "-tags", strings.Join(config.BuildTags(), " "))
-	}
-	args = append(args, pkgs...)
-	cgoEnabled := "0"
-	if config.CgoEnabled() {
-		cgoEnabled = "1"
-	}
-	cmd := exec.Command("go", args...)
-	cmd.Stdout = os.Stdout
-	cmd.Stderr = os.Stderr
-	cmd.Env = append(os.Environ(), "GOROOT="+goroot, "GOOS="+config.GOOS(), "GOARCH="+config.GOARCH(), "CGO_ENABLED="+cgoEnabled)
-	cmd.Run()
-	return nil
-}
-
 func usage() {
 	fmt.Fprintln(os.Stderr, "TinyGo is a Go compiler for small places.")
 	fmt.Fprintln(os.Stderr, "version:", goenv.Version)
@@ -755,10 +785,16 @@ func printCompilerError(logln func(...interface{}), err error) {
 			}
 		}
 	case loader.Errors:
-		logln("#", err.Pkg.PkgPath)
+		logln("#", err.Pkg.ImportPath)
 		for _, err := range err.Errs {
 			printCompilerError(logln, err)
 		}
+	case loader.Error:
+		logln(err.Err.Error())
+		logln("package", err.ImportStack[0])
+		for _, pkgPath := range err.ImportStack[1:] {
+			logln("\timports", pkgPath)
+		}
 	case *builder.MultiError:
 		for _, err := range err.Errs {
 			printCompilerError(logln, err)
@@ -785,7 +821,6 @@ func main() {
 	}
 	command := os.Args[1]

-	outpath := flag.String("o", "", "output filename")
 	opt := flag.String("opt", "z", "optimization level: 0, 1, 2, s, z")
 	gc := flag.String("gc", "", "garbage collector to use (none, leaking, extalloc, conservative)")
 	panicStrategy := flag.String("panic", "print", "panic strategy (print, trap)")
@@ -803,14 +838,22 @@ func main() {
 	programmer := flag.String("programmer", "", "which hardware programmer to use")
 	cFlags := flag.String("cflags", "", "additional cflags for compiler")
 	ldFlags := flag.String("ldflags", "", "additional ldflags for linker")
-	wasmAbi := flag.String("wasm-abi", "js", "WebAssembly ABI conventions: js (no i64 params) or generic")
+	wasmAbi := flag.String("wasm-abi", "", "WebAssembly ABI conventions: js (no i64 params) or generic")
 	heapSize := flag.String("heap-size", "1M", "default heap size in bytes (only supported by WebAssembly)")

 	var flagJSON, flagDeps *bool
-	if command == "list" {
+	if command == "help" || command == "list" {
 		flagJSON = flag.Bool("json", false, "print data in JSON format")
 		flagDeps = flag.Bool("deps", false, "")
 	}
+	var outpath string
+	if command == "help" || command == "build" || command == "build-library" || command == "test" {
+		flag.StringVar(&outpath, "o", "", "output filename")
+	}
+	var testCompileOnlyFlag *bool
+	if command == "help" || command == "test" {
+		testCompileOnlyFlag = flag.Bool("c", false, "compile the test binary but do not run it")
+	}

 	// Early command processing, before commands are interpreted by the Go flag
 	// library.
@@ -868,7 +911,7 @@ func main() {

 	switch command {
 	case "build":
-		if *outpath == "" {
+		if outpath == "" {
 			fmt.Fprintln(os.Stderr, "No output filename supplied (-o).")
 			usage()
 			os.Exit(1)
@@ -881,15 +924,15 @@ func main() {
 			usage()
 			os.Exit(1)
 		}
-		if options.Target == "" && filepath.Ext(*outpath) == ".wasm" {
+		if options.Target == "" && filepath.Ext(outpath) == ".wasm" {
 			options.Target = "wasm"
 		}

-		err := Build(pkgName, *outpath, options)
+		err := Build(pkgName, outpath, options)
 		handleCompilerError(err)
 	case "build-library":
 		// Note: this command is only meant to be used while making a release!
-		if *outpath == "" {
+		if outpath == "" {
 			fmt.Fprintln(os.Stderr, "No output filename supplied (-o).")
 			usage()
 			os.Exit(1)
@@ -914,13 +957,8 @@ func main() {
 		}
 		path, err := lib.Load(*target)
 		handleCompilerError(err)
-		copyFile(path, *outpath)
+		copyFile(path, outpath)
 	case "flash", "gdb":
-		if *outpath != "" {
-			fmt.Fprintln(os.Stderr, "Output cannot be specified with the flash command.")
-			usage()
-			os.Exit(1)
-		}
 		pkgName := filepath.ToSlash(flag.Arg(0))
 		if command == "flash" {
 			err := Flash(pkgName, *port, options)
@@ -952,8 +990,37 @@ func main() {
 			usage()
 			os.Exit(1)
 		}
-		err := Test(pkgName, options)
+		err := Test(pkgName, options, *testCompileOnlyFlag, outpath)
 		handleCompilerError(err)
+	case "targets":
+		dir := filepath.Join(goenv.Get("TINYGOROOT"), "targets")
+		entries, err := ioutil.ReadDir(dir)
+		if err != nil {
+			fmt.Fprintln(os.Stderr, "could not list targets:", err)
+			os.Exit(1)
+			return
+		}
+		for _, entry := range entries {
+			if !entry.Mode().IsRegular() || !strings.HasSuffix(entry.Name(), ".json") {
+				// Only inspect JSON files.
+				continue
+			}
+			path := filepath.Join(dir, entry.Name())
+			spec, err := compileopts.LoadTarget(path)
+			if err != nil {
+				fmt.Fprintln(os.Stderr, "could not list target:", err)
+				os.Exit(1)
+				return
+			}
+			if spec.FlashMethod == "" && spec.FlashCommand == "" && spec.Emulator == nil {
+				// This doesn't look like a regular target file, but rather like
+				// a parent target (such as targets/cortex-m.json).
+				continue
+			}
+			name := entry.Name()
+			name = name[:len(name)-5]
+			fmt.Println(name)
+		}
 	case "info":
 		if flag.NArg() == 1 {
 			options.Target = flag.Arg(0)
@@ -973,12 +1040,18 @@ func main() {
 			fmt.Fprintln(os.Stderr, err)
 			os.Exit(1)
 		}
+		cachedGOROOT, err := loader.GetCachedGoroot(config)
+		if err != nil {
+			fmt.Fprintln(os.Stderr, err)
+			os.Exit(1)
+		}
 		fmt.Printf("LLVM triple:       %s\n", config.Triple())
 		fmt.Printf("GOOS:              %s\n", config.GOOS())
 		fmt.Printf("GOARCH:            %s\n", config.GOARCH())
 		fmt.Printf("build tags:        %s\n", strings.Join(config.BuildTags(), " "))
 		fmt.Printf("garbage collector: %s\n", config.GC())
 		fmt.Printf("scheduler:         %s\n", config.Scheduler())
+		fmt.Printf("cached GOROOT:     %s\n", cachedGOROOT)
 	case "list":
 		config, err := builder.NewConfig(options)
 		if err != nil {
@@ -986,11 +1059,31 @@ func main() {
 			usage()
 			os.Exit(1)
 		}
-		err = runGoList(config, *flagJSON, *flagDeps, flag.Args())
+		var extraArgs []string
+		if *flagJSON {
+			extraArgs = append(extraArgs, "-json")
+		}
+		if *flagDeps {
+			extraArgs = append(extraArgs, "-deps")
+		}
+		cmd, err := loader.List(config, extraArgs, flag.Args())
 		if err != nil {
 			fmt.Fprintln(os.Stderr, "failed to run `go list`:", err)
 			os.Exit(1)
 		}
+		cmd.Stdout = os.Stdout
+		cmd.Stderr = os.Stderr
+		err = cmd.Run()
+		if err != nil {
+			if exitErr, ok := err.(*exec.ExitError); ok {
+				if status, ok := exitErr.Sys().(syscall.WaitStatus); ok {
+					os.Exit(status.ExitStatus())
+				}
+				os.Exit(1)
+			}
+			fmt.Fprintln(os.Stderr, "failed to run `go list`:", err)
+			os.Exit(1)
+		}
 	case "clean":
 		// remove cache directory
 		err := os.RemoveAll(goenv.Get("GOCACHE"))
@@ -1005,7 +1098,11 @@ func main() {
 		if s, err := goenv.GorootVersionString(goenv.Get("GOROOT")); err == nil {
 			goversion = s
 		}
-		fmt.Printf("tinygo version %s %s/%s (using go version %s and LLVM version %s)\n", goenv.Version, runtime.GOOS, runtime.GOARCH, goversion, llvm.Version)
+		version := goenv.Version
+		if strings.HasSuffix(goenv.Version, "-dev") && gitSha1 != "" {
+			version += "-" + gitSha1
+		}
+		fmt.Printf("tinygo version %s %s/%s (using go version %s and LLVM version %s)\n", version, runtime.GOOS, runtime.GOARCH, goversion, llvm.Version)
 	case "env":
 		if flag.NArg() == 0 {
 			// Show all environment variables.
@@ -78,6 +78,9 @@ func TestCompiler(t *testing.T) {
 	}

 	if runtime.GOOS == "linux" {
+		t.Run("X86Linux", func(t *testing.T) {
+			runPlatTests("i386--linux-gnu", matches, t)
+		})
 		t.Run("ARMLinux", func(t *testing.T) {
 			runPlatTests("arm--linux-gnueabihf", matches, t)
 		})
@@ -98,6 +101,10 @@ func TestCompiler(t *testing.T) {
 				runPlatTests("wasm", matches, t)
 			})
 		}
+
+		t.Run("WASI", func(t *testing.T) {
+			runPlatTests("wasi", matches, t)
+		})
 	}
 }

@@ -109,7 +116,6 @@ func runPlatTests(target string, matches []string, t *testing.T) {

 		t.Run(filepath.Base(path), func(t *testing.T) {
 			t.Parallel()
-
 			runTest(path, target, t)
 		})
 	}
@@ -157,10 +163,11 @@ func runTest(path, target string, t *testing.T) {
 		PrintIR:    false,
 		DumpSSA:    false,
 		VerifyIR:   true,
-		Debug:      false,
+		Debug:      true,
 		PrintSizes: "",
-		WasmAbi:    "js",
+		WasmAbi:    "",
 	}
+
 	binary := filepath.Join(tmpdir, "test")
 	err = runBuild("./"+path, binary, config)
 	if err != nil {
@@ -181,10 +188,11 @@ func runTest(path, target string, t *testing.T) {
 			t.Fatal("failed to load target spec:", err)
 		}
 		if len(spec.Emulator) == 0 {
-			t.Fatal("no emulator available for target:", target)
+			cmd = exec.Command(binary)
+		} else {
+			args := append(spec.Emulator[1:], binary)
+			cmd = exec.Command(spec.Emulator[0], args...)
 		}
-		args := append(spec.Emulator[1:], binary)
-		cmd = exec.Command(spec.Emulator[0], args...)
 	}
 	stdout := &bytes.Buffer{}
 	cmd.Stdout = stdout
@@ -76,32 +76,8 @@ const (
 	SCS_BASE  = 0xE000E000
 	SYST_BASE = SCS_BASE + 0x0010
 	NVIC_BASE = SCS_BASE + 0x0100
-	SCB_BASE  = SCS_BASE + 0x0D00
 )

-const (
-	SCB_AIRCR_VECTKEY_Pos     = 16
-	SCB_AIRCR_SYSRESETREQ_Pos = 2
-	SCB_AIRCR_SYSRESETREQ_Msk = 1 << SCB_AIRCR_SYSRESETREQ_Pos
-)
-
-// System Control Block (SCB)
-//
-// SCB_Type provides the definitions for the System Control Block Registers.
-type SCB_Type struct {
-	CPUID volatile.Register32    // CPUID Base Register
-	ICSR  volatile.Register32    // Interrupt Control and State Register
-	VTOR  volatile.Register32    // Vector Table Offset Register
-	AIRCR volatile.Register32    // Application Interrupt and Reset Control Register
-	SCR   volatile.Register32    // System Control Register
-	CCR   volatile.Register32    // Configuration Control Register
-	_     volatile.Register32    // RESERVED1;
-	SHP   [2]volatile.Register32 // System Handlers Priority Registers. [0] is RESERVED
-	SHCSR volatile.Register32    // System Handler Control and State Register
-}
-
-var SCB = (*SCB_Type)(unsafe.Pointer(uintptr(SCB_BASE)))
-
 // Nested Vectored Interrupt Controller (NVIC).
 //
 // Source:
@@ -213,17 +189,6 @@ func EnableInterrupts(mask uintptr) {
 	})
 }

-// SystemReset performs a hard system reset.
-func SystemReset() {
-	// SCB->AIRCR  = ((0x5FA << SCB_AIRCR_VECTKEY_Pos)      |
-	//              SCB_AIRCR_SYSRESETREQ_Msk);
-	SCB.AIRCR.Set((0x5FA << SCB_AIRCR_VECTKEY_Pos) | SCB_AIRCR_SYSRESETREQ_Msk)
-
-	for {
-		Asm("wfi")
-	}
-}
-
 // Set up the system timer to generate periodic tick events.
 // This will cause SysTick_Handler to fire once per tick.
 // The cyclecount parameter is a counter value which can range from 0 to
@@ -1,9 +1,11 @@
 .syntax unified
+.cfi_sections .debug_frame

 .section .text.HardFault_Handler
 .global  HardFault_Handler
 .type    HardFault_Handler, %function
 HardFault_Handler:
+    .cfi_startproc
    // Put the old stack pointer in the first argument, for easy debugging. This
    // is especially useful on Cortex-M0, which supports far fewer debug
    // facilities.
@@ -19,6 +21,7 @@ HardFault_Handler:

    // Continue handling this error in Go.
    bl handleHardFault
+    .cfi_endproc
 .size HardFault_Handler, .-HardFault_Handler

 // This is a convenience function for semihosting support.
@@ -27,5 +30,8 @@ HardFault_Handler:
 .global  SemihostingCall
 .type    SemihostingCall, %function
 SemihostingCall:
+    .cfi_startproc
    bkpt 0xab
    bx   lr
+    .cfi_endproc
+.size SemihostingCall, .-SemihostingCall
@@ -0,0 +1,431 @@
+// Hand created file. DO NOT DELETE.
+// Cortex-M System Control Block-related definitions.
+
+// +build cortexm
+
+package arm
+
+import (
+	"runtime/volatile"
+	"unsafe"
+)
+
+const SCB_BASE = SCS_BASE + 0x0D00
+
+// System Control Block (SCB)
+//
+// SCB_Type provides the definitions for the System Control Block Registers.
+type SCB_Type struct {
+	CPUID volatile.Register32 // 0xD00: CPUID Base Register
+	ICSR  volatile.Register32 // 0xD04: Interrupt Control and State Register
+	VTOR  volatile.Register32 // 0xD08: Vector Table Offset Register
+	AIRCR volatile.Register32 // 0xD0C: Application Interrupt and Reset Control Register
+	SCR   volatile.Register32 // 0xD10: System Control Register
+	CCR   volatile.Register32 // 0xD14: Configuration and Control Register
+	SHPR1 volatile.Register32 // 0xD18: System Handler Priority Register 1 (Cortex-M3/M33/M4/M7 only)
+	SHPR2 volatile.Register32 // 0xD1C: System Handler Priority Register 2
+	SHPR3 volatile.Register32 // 0xD20: System Handler Priority Register 3
+	// the following are only applicable for Cortex-M3/M33/M4/M7
+	SHCSR volatile.Register32 // 0xD24: System Handler Control and State Register
+	CFSR  volatile.Register32 // 0xD28: Configurable Fault Status Register
+	HFSR  volatile.Register32 // 0xD2C: HardFault Status Register
+	DFSR  volatile.Register32 // 0xD30: Debug Fault Status Register
+	MMFAR volatile.Register32 // 0xD34: MemManage Fault Address Register
+	BFAR  volatile.Register32 // 0xD38: BusFault Address Register
+}
+
+var SCB = (*SCB_Type)(unsafe.Pointer(uintptr(SCB_BASE)))
+
+// SystemReset performs a hard system reset.
+func SystemReset() {
+	SCB.AIRCR.Set((0x5FA << SCB_AIRCR_VECTKEY_Pos) | SCB_AIRCR_SYSRESETREQ_Msk)
+	for {
+		Asm("wfi")
+	}
+}
+
+const (
+	// CPUID: CPUID Base Register
+	SCB_CPUID_REVISION_Pos     = 0x0        // Position of REVISION field.
+	SCB_CPUID_REVISION_Msk     = 0xf        // Bit mask of REVISION field.
+	SCB_CPUID_PARTNO_Pos       = 0x4        // Position of PARTNO field.
+	SCB_CPUID_PARTNO_Msk       = 0xfff0     // Bit mask of PARTNO field.
+	SCB_CPUID_ARCHITECTURE_Pos = 0x10       // Position of ARCHITECTURE field.
+	SCB_CPUID_ARCHITECTURE_Msk = 0xf0000    // Bit mask of ARCHITECTURE field.
+	SCB_CPUID_VARIANT_Pos      = 0x14       // Position of VARIANT field.
+	SCB_CPUID_VARIANT_Msk      = 0xf00000   // Bit mask of VARIANT field.
+	SCB_CPUID_IMPLEMENTER_Pos  = 0x18       // Position of IMPLEMENTER field.
+	SCB_CPUID_IMPLEMENTER_Msk  = 0xff000000 // Bit mask of IMPLEMENTER field.
+
+	// ICSR: Interrupt Control and State Register
+	SCB_ICSR_VECTACTIVE_Pos          = 0x0        // Position of VECTACTIVE field.
+	SCB_ICSR_VECTACTIVE_Msk          = 0x1ff      // Bit mask of VECTACTIVE field.
+	SCB_ICSR_RETTOBASE_Pos           = 0xb        // Position of RETTOBASE field.
+	SCB_ICSR_RETTOBASE_Msk           = 0x800      // Bit mask of RETTOBASE field.
+	SCB_ICSR_RETTOBASE               = 0x800      // Bit RETTOBASE.
+	SCB_ICSR_RETTOBASE_RETTOBASE_0   = 0x0        // there are preempted active exceptions to execute
+	SCB_ICSR_RETTOBASE_RETTOBASE_1   = 0x1        // there are no active exceptions, or the currently-executing exception is the only active exception
+	SCB_ICSR_VECTPENDING_Pos         = 0xc        // Position of VECTPENDING field.
+	SCB_ICSR_VECTPENDING_Msk         = 0x1ff000   // Bit mask of VECTPENDING field.
+	SCB_ICSR_ISRPENDING_Pos          = 0x16       // Position of ISRPENDING field.
+	SCB_ICSR_ISRPENDING_Msk          = 0x400000   // Bit mask of ISRPENDING field.
+	SCB_ICSR_ISRPENDING              = 0x400000   // Bit ISRPENDING.
+	SCB_ICSR_ISRPENDING_ISRPENDING_0 = 0x0        // No external interrupt pending.
+	SCB_ICSR_ISRPENDING_ISRPENDING_1 = 0x1        // External interrupt pending.
+	SCB_ICSR_PENDSTCLR_Pos           = 0x19       // Position of PENDSTCLR field.
+	SCB_ICSR_PENDSTCLR_Msk           = 0x2000000  // Bit mask of PENDSTCLR field.
+	SCB_ICSR_PENDSTCLR               = 0x2000000  // Bit PENDSTCLR.
+	SCB_ICSR_PENDSTCLR_PENDSTCLR_0   = 0x0        // no effect
+	SCB_ICSR_PENDSTCLR_PENDSTCLR_1   = 0x1        // removes the pending state from the SysTick exception
+	SCB_ICSR_PENDSTSET_Pos           = 0x1a       // Position of PENDSTSET field.
+	SCB_ICSR_PENDSTSET_Msk           = 0x4000000  // Bit mask of PENDSTSET field.
+	SCB_ICSR_PENDSTSET               = 0x4000000  // Bit PENDSTSET.
+	SCB_ICSR_PENDSTSET_PENDSTSET_0   = 0x0        // write: no effect; read: SysTick exception is not pending
+	SCB_ICSR_PENDSTSET_PENDSTSET_1   = 0x1        // write: changes SysTick exception state to pending; read: SysTick exception is pending
+	SCB_ICSR_PENDSVCLR_Pos           = 0x1b       // Position of PENDSVCLR field.
+	SCB_ICSR_PENDSVCLR_Msk           = 0x8000000  // Bit mask of PENDSVCLR field.
+	SCB_ICSR_PENDSVCLR               = 0x8000000  // Bit PENDSVCLR.
+	SCB_ICSR_PENDSVCLR_PENDSVCLR_0   = 0x0        // no effect
+	SCB_ICSR_PENDSVCLR_PENDSVCLR_1   = 0x1        // removes the pending state from the PendSV exception
+	SCB_ICSR_PENDSVSET_Pos           = 0x1c       // Position of PENDSVSET field.
+	SCB_ICSR_PENDSVSET_Msk           = 0x10000000 // Bit mask of PENDSVSET field.
+	SCB_ICSR_PENDSVSET               = 0x10000000 // Bit PENDSVSET.
+	SCB_ICSR_PENDSVSET_PENDSVSET_0   = 0x0        // write: no effect; read: PendSV exception is not pending
+	SCB_ICSR_PENDSVSET_PENDSVSET_1   = 0x1        // write: changes PendSV exception state to pending; read: PendSV exception is pending
+	SCB_ICSR_NMIPENDSET_Pos          = 0x1f       // Position of NMIPENDSET field.
+	SCB_ICSR_NMIPENDSET_Msk          = 0x80000000 // Bit mask of NMIPENDSET field.
+	SCB_ICSR_NMIPENDSET              = 0x80000000 // Bit NMIPENDSET.
+	SCB_ICSR_NMIPENDSET_NMIPENDSET_0 = 0x0        // write: no effect; read: NMI exception is not pending
+	SCB_ICSR_NMIPENDSET_NMIPENDSET_1 = 0x1        // write: changes NMI exception state to pending; read: NMI exception is pending
+
+	// VTOR: Vector Table Offset Register
+	SCB_VTOR_TBLOFF_Pos = 0x7        // Position of TBLOFF field.
+	SCB_VTOR_TBLOFF_Msk = 0xffffff80 // Bit mask of TBLOFF field.
+
+	// AIRCR: Application Interrupt and Reset Control Register
+	SCB_AIRCR_VECTRESET_Pos                 = 0x0        // Position of VECTRESET field.
+	SCB_AIRCR_VECTRESET_Msk                 = 0x1        // Bit mask of VECTRESET field.
+	SCB_AIRCR_VECTRESET                     = 0x1        // Bit VECTRESET.
+	SCB_AIRCR_VECTRESET_VECTRESET_0         = 0x0        // No change
+	SCB_AIRCR_VECTRESET_VECTRESET_1         = 0x1        // Causes a local system reset
+	SCB_AIRCR_VECTCLRACTIVE_Pos             = 0x1        // Position of VECTCLRACTIVE field.
+	SCB_AIRCR_VECTCLRACTIVE_Msk             = 0x2        // Bit mask of VECTCLRACTIVE field.
+	SCB_AIRCR_VECTCLRACTIVE                 = 0x2        // Bit VECTCLRACTIVE.
+	SCB_AIRCR_VECTCLRACTIVE_VECTCLRACTIVE_0 = 0x0        // No change
+	SCB_AIRCR_VECTCLRACTIVE_VECTCLRACTIVE_1 = 0x1        // Clears all active state information for fixed and configurable exceptions
+	SCB_AIRCR_SYSRESETREQ_Pos               = 0x2        // Position of SYSRESETREQ field.
+	SCB_AIRCR_SYSRESETREQ_Msk               = 0x4        // Bit mask of SYSRESETREQ field.
+	SCB_AIRCR_SYSRESETREQ                   = 0x4        // Bit SYSRESETREQ.
+	SCB_AIRCR_SYSRESETREQ_SYSRESETREQ_0     = 0x0        // no system reset request
+	SCB_AIRCR_SYSRESETREQ_SYSRESETREQ_1     = 0x1        // asserts a signal to the outer system that requests a reset
+	SCB_AIRCR_PRIGROUP_Pos                  = 0x8        // Position of PRIGROUP field.
+	SCB_AIRCR_PRIGROUP_Msk                  = 0x700      // Bit mask of PRIGROUP field.
+	SCB_AIRCR_ENDIANNESS_Pos                = 0xf        // Position of ENDIANNESS field.
+	SCB_AIRCR_ENDIANNESS_Msk                = 0x8000     // Bit mask of ENDIANNESS field.
+	SCB_AIRCR_ENDIANNESS                    = 0x8000     // Bit ENDIANNESS.
+	SCB_AIRCR_ENDIANNESS_ENDIANNESS_0       = 0x0        // Little-endian
+	SCB_AIRCR_ENDIANNESS_ENDIANNESS_1       = 0x1        // Big-endian
+	SCB_AIRCR_VECTKEY_Pos                   = 0x10       // Position of VECTKEY field.
+	SCB_AIRCR_VECTKEY_Msk                   = 0xffff0000 // Bit mask of VECTKEY field.
+
+	// SCR: System Control Register
+	SCB_SCR_SLEEPONEXIT_Pos           = 0x1  // Position of SLEEPONEXIT field.
+	SCB_SCR_SLEEPONEXIT_Msk           = 0x2  // Bit mask of SLEEPONEXIT field.
+	SCB_SCR_SLEEPONEXIT               = 0x2  // Bit SLEEPONEXIT.
+	SCB_SCR_SLEEPONEXIT_SLEEPONEXIT_0 = 0x0  // o not sleep when returning to Thread mode
+	SCB_SCR_SLEEPONEXIT_SLEEPONEXIT_1 = 0x1  // enter sleep, or deep sleep, on return from an ISR
+	SCB_SCR_SLEEPDEEP_Pos             = 0x2  // Position of SLEEPDEEP field.
+	SCB_SCR_SLEEPDEEP_Msk             = 0x4  // Bit mask of SLEEPDEEP field.
+	SCB_SCR_SLEEPDEEP                 = 0x4  // Bit SLEEPDEEP.
+	SCB_SCR_SLEEPDEEP_SLEEPDEEP_0     = 0x0  // sleep
+	SCB_SCR_SLEEPDEEP_SLEEPDEEP_1     = 0x1  // deep sleep
+	SCB_SCR_SEVONPEND_Pos             = 0x4  // Position of SEVONPEND field.
+	SCB_SCR_SEVONPEND_Msk             = 0x10 // Bit mask of SEVONPEND field.
+	SCB_SCR_SEVONPEND                 = 0x10 // Bit SEVONPEND.
+	SCB_SCR_SEVONPEND_SEVONPEND_0     = 0x0  // only enabled interrupts or events can wakeup the processor, disabled interrupts are excluded
+	SCB_SCR_SEVONPEND_SEVONPEND_1     = 0x1  // enabled events and all interrupts, including disabled interrupts, can wakeup the processor
+
+	// CCR: Configuration and Control Register
+	SCB_CCR_NONBASETHRDENA_Pos              = 0x0     // Position of NONBASETHRDENA field.
+	SCB_CCR_NONBASETHRDENA_Msk              = 0x1     // Bit mask of NONBASETHRDENA field.
+	SCB_CCR_NONBASETHRDENA                  = 0x1     // Bit NONBASETHRDENA.
+	SCB_CCR_NONBASETHRDENA_NONBASETHRDENA_0 = 0x0     // processor can enter Thread mode only when no exception is active
+	SCB_CCR_NONBASETHRDENA_NONBASETHRDENA_1 = 0x1     // processor can enter Thread mode from any level under the control of an EXC_RETURN value
+	SCB_CCR_USERSETMPEND_Pos                = 0x1     // Position of USERSETMPEND field.
+	SCB_CCR_USERSETMPEND_Msk                = 0x2     // Bit mask of USERSETMPEND field.
+	SCB_CCR_USERSETMPEND                    = 0x2     // Bit USERSETMPEND.
+	SCB_CCR_USERSETMPEND_USERSETMPEND_0     = 0x0     // disable
+	SCB_CCR_USERSETMPEND_USERSETMPEND_1     = 0x1     // enable
+	SCB_CCR_UNALIGN_TRP_Pos                 = 0x3     // Position of UNALIGN_TRP field.
+	SCB_CCR_UNALIGN_TRP_Msk                 = 0x8     // Bit mask of UNALIGN_TRP field.
+	SCB_CCR_UNALIGN_TRP                     = 0x8     // Bit UNALIGN_TRP.
+	SCB_CCR_UNALIGN_TRP_UNALIGN_TRP_0       = 0x0     // do not trap unaligned halfword and word accesses
+	SCB_CCR_UNALIGN_TRP_UNALIGN_TRP_1       = 0x1     // trap unaligned halfword and word accesses
+	SCB_CCR_DIV_0_TRP_Pos                   = 0x4     // Position of DIV_0_TRP field.
+	SCB_CCR_DIV_0_TRP_Msk                   = 0x10    // Bit mask of DIV_0_TRP field.
+	SCB_CCR_DIV_0_TRP                       = 0x10    // Bit DIV_0_TRP.
+	SCB_CCR_DIV_0_TRP_DIV_0_TRP_0           = 0x0     // do not trap divide by 0
+	SCB_CCR_DIV_0_TRP_DIV_0_TRP_1           = 0x1     // trap divide by 0
+	SCB_CCR_BFHFNMIGN_Pos                   = 0x8     // Position of BFHFNMIGN field.
+	SCB_CCR_BFHFNMIGN_Msk                   = 0x100   // Bit mask of BFHFNMIGN field.
+	SCB_CCR_BFHFNMIGN                       = 0x100   // Bit BFHFNMIGN.
+	SCB_CCR_BFHFNMIGN_BFHFNMIGN_0           = 0x0     // data bus faults caused by load and store instructions cause a lock-up
+	SCB_CCR_BFHFNMIGN_BFHFNMIGN_1           = 0x1     // handlers running at priority -1 and -2 ignore data bus faults caused by load and store instructions
+	SCB_CCR_STKALIGN_Pos                    = 0x9     // Position of STKALIGN field.
+	SCB_CCR_STKALIGN_Msk                    = 0x200   // Bit mask of STKALIGN field.
+	SCB_CCR_STKALIGN                        = 0x200   // Bit STKALIGN.
+	SCB_CCR_STKALIGN_STKALIGN_0             = 0x0     // 4-byte aligned
+	SCB_CCR_STKALIGN_STKALIGN_1             = 0x1     // 8-byte aligned
+	SCB_CCR_DC_Pos                          = 0x10    // Position of DC field.
+	SCB_CCR_DC_Msk                          = 0x10000 // Bit mask of DC field.
+	SCB_CCR_DC                              = 0x10000 // Bit DC.
+	SCB_CCR_DC_DC_0                         = 0x0     // L1 data cache disabled
+	SCB_CCR_DC_DC_1                         = 0x1     // L1 data cache enabled
+	SCB_CCR_IC_Pos                          = 0x11    // Position of IC field.
+	SCB_CCR_IC_Msk                          = 0x20000 // Bit mask of IC field.
+	SCB_CCR_IC                              = 0x20000 // Bit IC.
+	SCB_CCR_IC_IC_0                         = 0x0     // L1 instruction cache disabled
+	SCB_CCR_IC_IC_1                         = 0x1     // L1 instruction cache enabled
+	SCB_CCR_BP_Pos                          = 0x12    // Position of BP field.
+	SCB_CCR_BP_Msk                          = 0x40000 // Bit mask of BP field.
+	SCB_CCR_BP                              = 0x40000 // Bit BP.
+
+	// SHPR1: System Handler Priority Register 1
+	SCB_SHPR1_PRI_4_Pos = 0x0      // Position of PRI_4 field.
+	SCB_SHPR1_PRI_4_Msk = 0xff     // Bit mask of PRI_4 field.
+	SCB_SHPR1_PRI_5_Pos = 0x8      // Position of PRI_5 field.
+	SCB_SHPR1_PRI_5_Msk = 0xff00   // Bit mask of PRI_5 field.
+	SCB_SHPR1_PRI_6_Pos = 0x10     // Position of PRI_6 field.
+	SCB_SHPR1_PRI_6_Msk = 0xff0000 // Bit mask of PRI_6 field.
+
+	// SHPR2: System Handler Priority Register 2
+	SCB_SHPR2_PRI_11_Pos = 0x18       // Position of PRI_11 field.
+	SCB_SHPR2_PRI_11_Msk = 0xff000000 // Bit mask of PRI_11 field.
+
+	// SHPR3: System Handler Priority Register 3
+	SCB_SHPR3_PRI_14_Pos = 0x10       // Position of PRI_14 field.
+	SCB_SHPR3_PRI_14_Msk = 0xff0000   // Bit mask of PRI_14 field.
+	SCB_SHPR3_PRI_15_Pos = 0x18       // Position of PRI_15 field.
+	SCB_SHPR3_PRI_15_Msk = 0xff000000 // Bit mask of PRI_15 field.
+
+	// SHCSR: System Handler Control and State Register
+	SCB_SHCSR_MEMFAULTACT_Pos                 = 0x0     // Position of MEMFAULTACT field.
+	SCB_SHCSR_MEMFAULTACT_Msk                 = 0x1     // Bit mask of MEMFAULTACT field.
+	SCB_SHCSR_MEMFAULTACT                     = 0x1     // Bit MEMFAULTACT.
+	SCB_SHCSR_MEMFAULTACT_MEMFAULTACT_0       = 0x0     // exception is not active
+	SCB_SHCSR_MEMFAULTACT_MEMFAULTACT_1       = 0x1     // exception is active
+	SCB_SHCSR_BUSFAULTACT_Pos                 = 0x1     // Position of BUSFAULTACT field.
+	SCB_SHCSR_BUSFAULTACT_Msk                 = 0x2     // Bit mask of BUSFAULTACT field.
+	SCB_SHCSR_BUSFAULTACT                     = 0x2     // Bit BUSFAULTACT.
+	SCB_SHCSR_BUSFAULTACT_BUSFAULTACT_0       = 0x0     // exception is not active
+	SCB_SHCSR_BUSFAULTACT_BUSFAULTACT_1       = 0x1     // exception is active
+	SCB_SHCSR_USGFAULTACT_Pos                 = 0x3     // Position of USGFAULTACT field.
+	SCB_SHCSR_USGFAULTACT_Msk                 = 0x8     // Bit mask of USGFAULTACT field.
+	SCB_SHCSR_USGFAULTACT                     = 0x8     // Bit USGFAULTACT.
+	SCB_SHCSR_USGFAULTACT_USGFAULTACT_0       = 0x0     // exception is not active
+	SCB_SHCSR_USGFAULTACT_USGFAULTACT_1       = 0x1     // exception is active
+	SCB_SHCSR_SVCALLACT_Pos                   = 0x7     // Position of SVCALLACT field.
+	SCB_SHCSR_SVCALLACT_Msk                   = 0x80    // Bit mask of SVCALLACT field.
+	SCB_SHCSR_SVCALLACT                       = 0x80    // Bit SVCALLACT.
+	SCB_SHCSR_SVCALLACT_SVCALLACT_0           = 0x0     // exception is not active
+	SCB_SHCSR_SVCALLACT_SVCALLACT_1           = 0x1     // exception is active
+	SCB_SHCSR_MONITORACT_Pos                  = 0x8     // Position of MONITORACT field.
+	SCB_SHCSR_MONITORACT_Msk                  = 0x100   // Bit mask of MONITORACT field.
+	SCB_SHCSR_MONITORACT                      = 0x100   // Bit MONITORACT.
+	SCB_SHCSR_MONITORACT_MONITORACT_0         = 0x0     // exception is not active
+	SCB_SHCSR_MONITORACT_MONITORACT_1         = 0x1     // exception is active
+	SCB_SHCSR_PENDSVACT_Pos                   = 0xa     // Position of PENDSVACT field.
+	SCB_SHCSR_PENDSVACT_Msk                   = 0x400   // Bit mask of PENDSVACT field.
+	SCB_SHCSR_PENDSVACT                       = 0x400   // Bit PENDSVACT.
+	SCB_SHCSR_PENDSVACT_PENDSVACT_0           = 0x0     // exception is not active
+	SCB_SHCSR_PENDSVACT_PENDSVACT_1           = 0x1     // exception is active
+	SCB_SHCSR_SYSTICKACT_Pos                  = 0xb     // Position of SYSTICKACT field.
+	SCB_SHCSR_SYSTICKACT_Msk                  = 0x800   // Bit mask of SYSTICKACT field.
+	SCB_SHCSR_SYSTICKACT                      = 0x800   // Bit SYSTICKACT.
+	SCB_SHCSR_SYSTICKACT_SYSTICKACT_0         = 0x0     // exception is not active
+	SCB_SHCSR_SYSTICKACT_SYSTICKACT_1         = 0x1     // exception is active
+	SCB_SHCSR_USGFAULTPENDED_Pos              = 0xc     // Position of USGFAULTPENDED field.
+	SCB_SHCSR_USGFAULTPENDED_Msk              = 0x1000  // Bit mask of USGFAULTPENDED field.
+	SCB_SHCSR_USGFAULTPENDED                  = 0x1000  // Bit USGFAULTPENDED.
+	SCB_SHCSR_USGFAULTPENDED_USGFAULTPENDED_0 = 0x0     // exception is not pending
+	SCB_SHCSR_USGFAULTPENDED_USGFAULTPENDED_1 = 0x1     // exception is pending
+	SCB_SHCSR_MEMFAULTPENDED_Pos              = 0xd     // Position of MEMFAULTPENDED field.
+	SCB_SHCSR_MEMFAULTPENDED_Msk              = 0x2000  // Bit mask of MEMFAULTPENDED field.
+	SCB_SHCSR_MEMFAULTPENDED                  = 0x2000  // Bit MEMFAULTPENDED.
+	SCB_SHCSR_MEMFAULTPENDED_MEMFAULTPENDED_0 = 0x0     // exception is not pending
+	SCB_SHCSR_MEMFAULTPENDED_MEMFAULTPENDED_1 = 0x1     // exception is pending
+	SCB_SHCSR_BUSFAULTPENDED_Pos              = 0xe     // Position of BUSFAULTPENDED field.
+	SCB_SHCSR_BUSFAULTPENDED_Msk              = 0x4000  // Bit mask of BUSFAULTPENDED field.
+	SCB_SHCSR_BUSFAULTPENDED                  = 0x4000  // Bit BUSFAULTPENDED.
+	SCB_SHCSR_BUSFAULTPENDED_BUSFAULTPENDED_0 = 0x0     // exception is not pending
+	SCB_SHCSR_BUSFAULTPENDED_BUSFAULTPENDED_1 = 0x1     // exception is pending
+	SCB_SHCSR_SVCALLPENDED_Pos                = 0xf     // Position of SVCALLPENDED field.
+	SCB_SHCSR_SVCALLPENDED_Msk                = 0x8000  // Bit mask of SVCALLPENDED field.
+	SCB_SHCSR_SVCALLPENDED                    = 0x8000  // Bit SVCALLPENDED.
+	SCB_SHCSR_SVCALLPENDED_SVCALLPENDED_0     = 0x0     // exception is not pending
+	SCB_SHCSR_SVCALLPENDED_SVCALLPENDED_1     = 0x1     // exception is pending
+	SCB_SHCSR_MEMFAULTENA_Pos                 = 0x10    // Position of MEMFAULTENA field.
+	SCB_SHCSR_MEMFAULTENA_Msk                 = 0x10000 // Bit mask of MEMFAULTENA field.
+	SCB_SHCSR_MEMFAULTENA                     = 0x10000 // Bit MEMFAULTENA.
+	SCB_SHCSR_MEMFAULTENA_MEMFAULTENA_0       = 0x0     // disable the exception
+	SCB_SHCSR_MEMFAULTENA_MEMFAULTENA_1       = 0x1     // enable the exception
+	SCB_SHCSR_BUSFAULTENA_Pos                 = 0x11    // Position of BUSFAULTENA field.
+	SCB_SHCSR_BUSFAULTENA_Msk                 = 0x20000 // Bit mask of BUSFAULTENA field.
+	SCB_SHCSR_BUSFAULTENA                     = 0x20000 // Bit BUSFAULTENA.
+	SCB_SHCSR_BUSFAULTENA_BUSFAULTENA_0       = 0x0     // disable the exception
+	SCB_SHCSR_BUSFAULTENA_BUSFAULTENA_1       = 0x1     // enable the exception
+	SCB_SHCSR_USGFAULTENA_Pos                 = 0x12    // Position of USGFAULTENA field.
+	SCB_SHCSR_USGFAULTENA_Msk                 = 0x40000 // Bit mask of USGFAULTENA field.
+	SCB_SHCSR_USGFAULTENA                     = 0x40000 // Bit USGFAULTENA.
+	SCB_SHCSR_USGFAULTENA_USGFAULTENA_0       = 0x0     // disable the exception
+	SCB_SHCSR_USGFAULTENA_USGFAULTENA_1       = 0x1     // enable the exception
+
+	// CFSR: Configurable Fault Status Register
+	SCB_CFSR_IACCVIOL_Pos              = 0x0       // Position of IACCVIOL field.
+	SCB_CFSR_IACCVIOL_Msk              = 0x1       // Bit mask of IACCVIOL field.
+	SCB_CFSR_IACCVIOL                  = 0x1       // Bit IACCVIOL.
+	SCB_CFSR_IACCVIOL_IACCVIOL_0       = 0x0       // no instruction access violation fault
+	SCB_CFSR_IACCVIOL_IACCVIOL_1       = 0x1       // the processor attempted an instruction fetch from a location that does not permit execution
+	SCB_CFSR_DACCVIOL_Pos              = 0x1       // Position of DACCVIOL field.
+	SCB_CFSR_DACCVIOL_Msk              = 0x2       // Bit mask of DACCVIOL field.
+	SCB_CFSR_DACCVIOL                  = 0x2       // Bit DACCVIOL.
+	SCB_CFSR_DACCVIOL_DACCVIOL_0       = 0x0       // no data access violation fault
+	SCB_CFSR_DACCVIOL_DACCVIOL_1       = 0x1       // the processor attempted a load or store at a location that does not permit the operation
+	SCB_CFSR_MUNSTKERR_Pos             = 0x3       // Position of MUNSTKERR field.
+	SCB_CFSR_MUNSTKERR_Msk             = 0x8       // Bit mask of MUNSTKERR field.
+	SCB_CFSR_MUNSTKERR                 = 0x8       // Bit MUNSTKERR.
+	SCB_CFSR_MUNSTKERR_MUNSTKERR_0     = 0x0       // no unstacking fault
+	SCB_CFSR_MUNSTKERR_MUNSTKERR_1     = 0x1       // unstack for an exception return has caused one or more access violations
+	SCB_CFSR_MSTKERR_Pos               = 0x4       // Position of MSTKERR field.
+	SCB_CFSR_MSTKERR_Msk               = 0x10      // Bit mask of MSTKERR field.
+	SCB_CFSR_MSTKERR                   = 0x10      // Bit MSTKERR.
+	SCB_CFSR_MSTKERR_MSTKERR_0         = 0x0       // no stacking fault
+	SCB_CFSR_MSTKERR_MSTKERR_1         = 0x1       // stacking for an exception entry has caused one or more access violations
+	SCB_CFSR_MLSPERR_Pos               = 0x5       // Position of MLSPERR field.
+	SCB_CFSR_MLSPERR_Msk               = 0x20      // Bit mask of MLSPERR field.
+	SCB_CFSR_MLSPERR                   = 0x20      // Bit MLSPERR.
+	SCB_CFSR_MLSPERR_MLSPERR_0         = 0x0       // No MemManage fault occurred during floating-point lazy state preservation
+	SCB_CFSR_MLSPERR_MLSPERR_1         = 0x1       // A MemManage fault occurred during floating-point lazy state preservation
+	SCB_CFSR_MMARVALID_Pos             = 0x7       // Position of MMARVALID field.
+	SCB_CFSR_MMARVALID_Msk             = 0x80      // Bit mask of MMARVALID field.
+	SCB_CFSR_MMARVALID                 = 0x80      // Bit MMARVALID.
+	SCB_CFSR_MMARVALID_MMARVALID_0     = 0x0       // value in MMAR is not a valid fault address
+	SCB_CFSR_MMARVALID_MMARVALID_1     = 0x1       // MMAR holds a valid fault address
+	SCB_CFSR_IBUSERR_Pos               = 0x8       // Position of IBUSERR field.
+	SCB_CFSR_IBUSERR_Msk               = 0x100     // Bit mask of IBUSERR field.
+	SCB_CFSR_IBUSERR                   = 0x100     // Bit IBUSERR.
+	SCB_CFSR_IBUSERR_IBUSERR_0         = 0x0       // no instruction bus error
+	SCB_CFSR_IBUSERR_IBUSERR_1         = 0x1       // instruction bus error
+	SCB_CFSR_PRECISERR_Pos             = 0x9       // Position of PRECISERR field.
+	SCB_CFSR_PRECISERR_Msk             = 0x200     // Bit mask of PRECISERR field.
+	SCB_CFSR_PRECISERR                 = 0x200     // Bit PRECISERR.
+	SCB_CFSR_PRECISERR_PRECISERR_0     = 0x0       // no precise data bus error
+	SCB_CFSR_PRECISERR_PRECISERR_1     = 0x1       // a data bus error has occurred, and the PC value stacked for the exception return points to the instruction that caused the fault
+	SCB_CFSR_IMPRECISERR_Pos           = 0xa       // Position of IMPRECISERR field.
+	SCB_CFSR_IMPRECISERR_Msk           = 0x400     // Bit mask of IMPRECISERR field.
+	SCB_CFSR_IMPRECISERR               = 0x400     // Bit IMPRECISERR.
+	SCB_CFSR_IMPRECISERR_IMPRECISERR_0 = 0x0       // no imprecise data bus error
+	SCB_CFSR_IMPRECISERR_IMPRECISERR_1 = 0x1       // a data bus error has occurred, but the return address in the stack frame is not related to the instruction that caused the error
+	SCB_CFSR_UNSTKERR_Pos              = 0xb       // Position of UNSTKERR field.
+	SCB_CFSR_UNSTKERR_Msk              = 0x800     // Bit mask of UNSTKERR field.
+	SCB_CFSR_UNSTKERR                  = 0x800     // Bit UNSTKERR.
+	SCB_CFSR_UNSTKERR_UNSTKERR_0       = 0x0       // no unstacking fault
+	SCB_CFSR_UNSTKERR_UNSTKERR_1       = 0x1       // unstack for an exception return has caused one or more BusFaults
+	SCB_CFSR_STKERR_Pos                = 0xc       // Position of STKERR field.
+	SCB_CFSR_STKERR_Msk                = 0x1000    // Bit mask of STKERR field.
+	SCB_CFSR_STKERR                    = 0x1000    // Bit STKERR.
+	SCB_CFSR_STKERR_STKERR_0           = 0x0       // no stacking fault
+	SCB_CFSR_STKERR_STKERR_1           = 0x1       // stacking for an exception entry has caused one or more BusFaults
+	SCB_CFSR_LSPERR_Pos                = 0xd       // Position of LSPERR field.
+	SCB_CFSR_LSPERR_Msk                = 0x2000    // Bit mask of LSPERR field.
+	SCB_CFSR_LSPERR                    = 0x2000    // Bit LSPERR.
+	SCB_CFSR_LSPERR_LSPERR_0           = 0x0       // No bus fault occurred during floating-point lazy state preservation
+	SCB_CFSR_LSPERR_LSPERR_1           = 0x1       // A bus fault occurred during floating-point lazy state preservation
+	SCB_CFSR_BFARVALID_Pos             = 0xf       // Position of BFARVALID field.
+	SCB_CFSR_BFARVALID_Msk             = 0x8000    // Bit mask of BFARVALID field.
+	SCB_CFSR_BFARVALID                 = 0x8000    // Bit BFARVALID.
+	SCB_CFSR_BFARVALID_BFARVALID_0     = 0x0       // value in BFAR is not a valid fault address
+	SCB_CFSR_BFARVALID_BFARVALID_1     = 0x1       // BFAR holds a valid fault address
+	SCB_CFSR_UNDEFINSTR_Pos            = 0x10      // Position of UNDEFINSTR field.
+	SCB_CFSR_UNDEFINSTR_Msk            = 0x10000   // Bit mask of UNDEFINSTR field.
+	SCB_CFSR_UNDEFINSTR                = 0x10000   // Bit UNDEFINSTR.
+	SCB_CFSR_UNDEFINSTR_UNDEFINSTR_0   = 0x0       // no undefined instruction UsageFault
+	SCB_CFSR_UNDEFINSTR_UNDEFINSTR_1   = 0x1       // the processor has attempted to execute an undefined instruction
+	SCB_CFSR_INVSTATE_Pos              = 0x11      // Position of INVSTATE field.
+	SCB_CFSR_INVSTATE_Msk              = 0x20000   // Bit mask of INVSTATE field.
+	SCB_CFSR_INVSTATE                  = 0x20000   // Bit INVSTATE.
+	SCB_CFSR_INVSTATE_INVSTATE_0       = 0x0       // no invalid state UsageFault
+	SCB_CFSR_INVSTATE_INVSTATE_1       = 0x1       // the processor has attempted to execute an instruction that makes illegal use of the EPSR
+	SCB_CFSR_INVPC_Pos                 = 0x12      // Position of INVPC field.
+	SCB_CFSR_INVPC_Msk                 = 0x40000   // Bit mask of INVPC field.
+	SCB_CFSR_INVPC                     = 0x40000   // Bit INVPC.
+	SCB_CFSR_INVPC_INVPC_0             = 0x0       // no invalid PC load UsageFault
+	SCB_CFSR_INVPC_INVPC_1             = 0x1       // the processor has attempted an illegal load of EXC_RETURN to the PC
+	SCB_CFSR_NOCP_Pos                  = 0x13      // Position of NOCP field.
+	SCB_CFSR_NOCP_Msk                  = 0x80000   // Bit mask of NOCP field.
+	SCB_CFSR_NOCP                      = 0x80000   // Bit NOCP.
+	SCB_CFSR_NOCP_NOCP_0               = 0x0       // no UsageFault caused by attempting to access a coprocessor
+	SCB_CFSR_NOCP_NOCP_1               = 0x1       // the processor has attempted to access a coprocessor
+	SCB_CFSR_UNALIGNED_Pos             = 0x18      // Position of UNALIGNED field.
+	SCB_CFSR_UNALIGNED_Msk             = 0x1000000 // Bit mask of UNALIGNED field.
+	SCB_CFSR_UNALIGNED                 = 0x1000000 // Bit UNALIGNED.
+	SCB_CFSR_UNALIGNED_UNALIGNED_0     = 0x0       // no unaligned access fault, or unaligned access trapping not enabled
+	SCB_CFSR_UNALIGNED_UNALIGNED_1     = 0x1       // the processor has made an unaligned memory access
+	SCB_CFSR_DIVBYZERO_Pos             = 0x19      // Position of DIVBYZERO field.
+	SCB_CFSR_DIVBYZERO_Msk             = 0x2000000 // Bit mask of DIVBYZERO field.
+	SCB_CFSR_DIVBYZERO                 = 0x2000000 // Bit DIVBYZERO.
+	SCB_CFSR_DIVBYZERO_DIVBYZERO_0     = 0x0       // no divide by zero fault, or divide by zero trapping not enabled
+	SCB_CFSR_DIVBYZERO_DIVBYZERO_1     = 0x1       // the processor has executed an SDIV or UDIV instruction with a divisor of 0
+
+	// HFSR: HardFault Status register
+	SCB_HFSR_VECTTBL_Pos         = 0x1        // Position of VECTTBL field.
+	SCB_HFSR_VECTTBL_Msk         = 0x2        // Bit mask of VECTTBL field.
+	SCB_HFSR_VECTTBL             = 0x2        // Bit VECTTBL.
+	SCB_HFSR_VECTTBL_VECTTBL_0   = 0x0        // no BusFault on vector table read
+	SCB_HFSR_VECTTBL_VECTTBL_1   = 0x1        // BusFault on vector table read
+	SCB_HFSR_FORCED_Pos          = 0x1e       // Position of FORCED field.
+	SCB_HFSR_FORCED_Msk          = 0x40000000 // Bit mask of FORCED field.
+	SCB_HFSR_FORCED              = 0x40000000 // Bit FORCED.
+	SCB_HFSR_FORCED_FORCED_0     = 0x0        // no forced HardFault
+	SCB_HFSR_FORCED_FORCED_1     = 0x1        // forced HardFault
+	SCB_HFSR_DEBUGEVT_Pos        = 0x1f       // Position of DEBUGEVT field.
+	SCB_HFSR_DEBUGEVT_Msk        = 0x80000000 // Bit mask of DEBUGEVT field.
+	SCB_HFSR_DEBUGEVT            = 0x80000000 // Bit DEBUGEVT.
+	SCB_HFSR_DEBUGEVT_DEBUGEVT_0 = 0x0        // No Debug event has occurred.
+	SCB_HFSR_DEBUGEVT_DEBUGEVT_1 = 0x1        // Debug event has occurred. The Debug Fault Status Register has been updated.
+
+	// DFSR: Debug Fault Status Register
+	SCB_DFSR_HALTED_Pos          = 0x0  // Position of HALTED field.
+	SCB_DFSR_HALTED_Msk          = 0x1  // Bit mask of HALTED field.
+	SCB_DFSR_HALTED              = 0x1  // Bit HALTED.
+	SCB_DFSR_HALTED_HALTED_0     = 0x0  // No active halt request debug event
+	SCB_DFSR_HALTED_HALTED_1     = 0x1  // Halt request debug event active
+	SCB_DFSR_BKPT_Pos            = 0x1  // Position of BKPT field.
+	SCB_DFSR_BKPT_Msk            = 0x2  // Bit mask of BKPT field.
+	SCB_DFSR_BKPT                = 0x2  // Bit BKPT.
+	SCB_DFSR_BKPT_BKPT_0         = 0x0  // No current breakpoint debug event
+	SCB_DFSR_BKPT_BKPT_1         = 0x1  // At least one current breakpoint debug event
+	SCB_DFSR_DWTTRAP_Pos         = 0x2  // Position of DWTTRAP field.
+	SCB_DFSR_DWTTRAP_Msk         = 0x4  // Bit mask of DWTTRAP field.
+	SCB_DFSR_DWTTRAP             = 0x4  // Bit DWTTRAP.
+	SCB_DFSR_DWTTRAP_DWTTRAP_0   = 0x0  // No current debug events generated by the DWT
+	SCB_DFSR_DWTTRAP_DWTTRAP_1   = 0x1  // At least one current debug event generated by the DWT
+	SCB_DFSR_VCATCH_Pos          = 0x3  // Position of VCATCH field.
+	SCB_DFSR_VCATCH_Msk          = 0x8  // Bit mask of VCATCH field.
+	SCB_DFSR_VCATCH              = 0x8  // Bit VCATCH.
+	SCB_DFSR_VCATCH_VCATCH_0     = 0x0  // No Vector catch triggered
+	SCB_DFSR_VCATCH_VCATCH_1     = 0x1  // Vector catch triggered
+	SCB_DFSR_EXTERNAL_Pos        = 0x4  // Position of EXTERNAL field.
+	SCB_DFSR_EXTERNAL_Msk        = 0x10 // Bit mask of EXTERNAL field.
+	SCB_DFSR_EXTERNAL            = 0x10 // Bit EXTERNAL.
+	SCB_DFSR_EXTERNAL_EXTERNAL_0 = 0x0  // No external debug request debug event
+	SCB_DFSR_EXTERNAL_EXTERNAL_1 = 0x1  // External debug request debug event
+
+	// MMFAR: MemManage Fault Address Register
+	SCB_MMFAR_ADDRESS_Pos = 0x0        // Position of ADDRESS field.
+	SCB_MMFAR_ADDRESS_Msk = 0xffffffff // Bit mask of ADDRESS field.
+
+	// BFAR: BusFault Address Register
+	SCB_BFAR_ADDRESS_Pos = 0x0        // Position of ADDRESS field.
+	SCB_BFAR_ADDRESS_Msk = 0xffffffff // Bit mask of ADDRESS field.
+)
@@ -0,0 +1,36 @@
+package arm64
+
+// Run the given assembly code. The code will be marked as having side effects,
+// as it doesn't produce output and thus would normally be eliminated by the
+// optimizer.
+func Asm(asm string)
+
+// Run the given inline assembly. The code will be marked as having side
+// effects, as it would otherwise be optimized away. The inline assembly string
+// recognizes template values in the form {name}, like so:
+//
+//     arm.AsmFull(
+//         "str {value}, {result}",
+//         map[string]interface{}{
+//             "value":  1
+//             "result": &dest,
+//         })
+//
+// You can use {} in the asm string (which expands to a register) to set the
+// return value.
+func AsmFull(asm string, regs map[string]interface{}) uintptr
+
+// Run the following system call (SVCall) with 0 arguments.
+func SVCall0(num uintptr) uintptr
+
+// Run the following system call (SVCall) with 1 argument.
+func SVCall1(num uintptr, a1 interface{}) uintptr
+
+// Run the following system call (SVCall) with 2 arguments.
+func SVCall2(num uintptr, a1, a2 interface{}) uintptr
+
+// Run the following system call (SVCall) with 3 arguments.
+func SVCall3(num uintptr, a1, a2, a3 interface{}) uintptr
+
+// Run the following system call (SVCall) with 4 arguments.
+func SVCall4(num uintptr, a1, a2, a3, a4 interface{}) uintptr
@@ -0,0 +1,57 @@
+
+// The following definitions were copied from:
+// esp-idf/components/xtensa/include/xtensa/corebits.h
+#define PS_WOE_MASK                0x00040000
+#define PS_OWB_MASK                0x00000F00
+#define PS_CALLINC_MASK            0x00030000
+#define PS_WOE                     PS_WOE_MASK
+
+// Only calling it call_start_cpu0 for consistency with ESP-IDF.
+.section .text.call_start_cpu0
+1:
+    .long _stack_top
+.global call_start_cpu0
+call_start_cpu0:
+    // We need to set the stack pointer to a different value. This is somewhat
+    // complicated in the Xtensa architecture. The code below is a modified
+    // version of the following code:
+    // https://github.com/espressif/esp-idf/blob/c77c4ccf/components/xtensa/include/xt_instr_macros.h#L47
+
+    // Disable WOE.
+    rsr.ps a2
+    movi a3, ~(PS_WOE_MASK)
+    and a2, a2, a3
+    wsr.ps a2
+    rsync
+
+    // Set WINDOWSTART to 1 << WINDOWBASE.
+    rsr.windowbase  a2
+    ssl  a2
+    movi a2, 1
+    sll  a2, a2
+    wsr.windowstart a2
+    rsync
+
+    // Load new stack pointer.
+    l32r sp, 1b
+
+    // Re-enable WOE.
+    rsr.ps a2
+    movi a3, PS_WOE
+    or a2, a2, a3
+    wsr.ps a2
+    rsync
+
+    // Enable the FPU (coprocessor 0 so the lowest bit).
+    movi a2, 1
+    wsr.cpenable a2
+    rsync
+
+    // Jump to the runtime start function written in Go.
+    call4 main
+
+.section .text.tinygo_scanCurrentStack
+.global tinygo_scanCurrentStack
+tinygo_scanCurrentStack:
+    // TODO: save callee saved registers on the stack
+    j tinygo_scanstack
@@ -0,0 +1,6 @@
+
+.section .text.tinygo_scanCurrentStack
+.global tinygo_scanCurrentStack
+tinygo_scanCurrentStack:
+    // TODO: save callee saved registers on the stack
+    j tinygo_scanstack
@@ -0,0 +1,29 @@
+// Hand created file. DO NOT DELETE.
+// Hardfault aliases for definitions that have inconsistent naming (which are
+// auto-generated by gen-device-svd.go) among devices in package nxp.
+
+// +build nxp,!mimxrt1062
+
+package nxp
+
+const (
+	HardFault_CFSR_IACCVIOL    = SystemControl_CFSR_IACCVIOL
+	HardFault_CFSR_DACCVIOL    = SystemControl_CFSR_DACCVIOL
+	HardFault_CFSR_MUNSTKERR   = SystemControl_CFSR_MUNSTKERR
+	HardFault_CFSR_MSTKERR     = SystemControl_CFSR_MSTKERR
+	HardFault_CFSR_MLSPERR     = SystemControl_CFSR_MLSPERR
+	HardFault_CFSR_IBUSERR     = SystemControl_CFSR_IBUSERR
+	HardFault_CFSR_PRECISERR   = SystemControl_CFSR_PRECISERR
+	HardFault_CFSR_IMPRECISERR = SystemControl_CFSR_IMPRECISERR
+	HardFault_CFSR_UNSTKERR    = SystemControl_CFSR_UNSTKERR
+	HardFault_CFSR_STKERR      = SystemControl_CFSR_STKERR
+	HardFault_CFSR_LSPERR      = SystemControl_CFSR_LSPERR
+	HardFault_CFSR_UNDEFINSTR  = SystemControl_CFSR_UNDEFINSTR
+	HardFault_CFSR_INVSTATE    = SystemControl_CFSR_INVSTATE
+	HardFault_CFSR_INVPC       = SystemControl_CFSR_INVPC
+	HardFault_CFSR_NOCP        = SystemControl_CFSR_NOCP
+	HardFault_CFSR_UNALIGNED   = SystemControl_CFSR_UNALIGNED
+	HardFault_CFSR_DIVBYZERO   = SystemControl_CFSR_DIVBYZERO
+	HardFault_CFSR_MMARVALID   = SystemControl_CFSR_MMARVALID
+	HardFault_CFSR_BFARVALID   = SystemControl_CFSR_BFARVALID
+)
@@ -0,0 +1,529 @@
+// Hand created file. DO NOT DELETE.
+// Type definitions, fields, and constants associated with various clocks and
+// peripherals of the NXP MIMXRT1062.
+
+// +build nxp,mimxrt1062
+
+package nxp
+
+import (
+	"runtime/volatile"
+	"unsafe"
+)
+
+// Clock represents an individual peripheral clock that may be enabled/disabled
+// at runtime. Clocks also have a method `Mux` for selecting the clock source
+// and a method `Div` for selecting the hardware divisor. Note that many
+// peripherals have an independent prescalar configuration applied to the output
+// of this divisor.
+type (
+	Clock     uint32
+	ClockMode uint8
+)
+
+// Enable activates or deactivates the clock gate of receiver Clock c.
+func (c Clock) Enable(enable bool) {
+	if enable {
+		c.setGate(clockNeededRunWait)
+	} else {
+		c.setGate(clockNotNeeded)
+	}
+}
+
+// Mux selects a clock source for the mux of the receiver Clock c.
+func (c Clock) Mux(mux uint32) { c.setCcm(mux) }
+
+// Div configures the prescalar divisor of the receiver Clock c.
+func (c Clock) Div(div uint32) { c.setCcm(div) }
+
+const (
+	ClockModeRun  ClockMode = 0 // Remain in run mode
+	ClockModeWait ClockMode = 1 // Transfer to wait mode
+	ClockModeStop ClockMode = 2 // Transfer to stop mode
+)
+
+// Set configures the run mode of the MCU.
+func (m ClockMode) Set() {
+	CCM.CLPCR.Set((CCM.CLPCR.Get() & ^uint32(CCM_CLPCR_LPM_Msk)) |
+		((uint32(m) << CCM_CLPCR_LPM_Pos) & CCM_CLPCR_LPM_Msk))
+}
+
+// Named oscillators
+const (
+	ClockCpu         Clock = 0x0  // CPU clock
+	ClockAhb         Clock = 0x1  // AHB clock
+	ClockSemc        Clock = 0x2  // SEMC clock
+	ClockIpg         Clock = 0x3  // IPG clock
+	ClockPer         Clock = 0x4  // PER clock
+	ClockOsc         Clock = 0x5  // OSC clock selected by PMU_LOWPWR_CTRL[OSC_SEL]
+	ClockRtc         Clock = 0x6  // RTC clock (RTCCLK)
+	ClockArmPll      Clock = 0x7  // ARMPLLCLK
+	ClockUsb1Pll     Clock = 0x8  // USB1PLLCLK
+	ClockUsb1PllPfd0 Clock = 0x9  // USB1PLLPDF0CLK
+	ClockUsb1PllPfd1 Clock = 0xA  // USB1PLLPFD1CLK
+	ClockUsb1PllPfd2 Clock = 0xB  // USB1PLLPFD2CLK
+	ClockUsb1PllPfd3 Clock = 0xC  // USB1PLLPFD3CLK
+	ClockUsb2Pll     Clock = 0xD  // USB2PLLCLK
+	ClockSysPll      Clock = 0xE  // SYSPLLCLK
+	ClockSysPllPfd0  Clock = 0xF  // SYSPLLPDF0CLK
+	ClockSysPllPfd1  Clock = 0x10 // SYSPLLPFD1CLK
+	ClockSysPllPfd2  Clock = 0x11 // SYSPLLPFD2CLK
+	ClockSysPllPfd3  Clock = 0x12 // SYSPLLPFD3CLK
+	ClockEnetPll0    Clock = 0x13 // Enet PLLCLK ref_enetpll0
+	ClockEnetPll1    Clock = 0x14 // Enet PLLCLK ref_enetpll1
+	ClockEnetPll2    Clock = 0x15 // Enet PLLCLK ref_enetpll2
+	ClockAudioPll    Clock = 0x16 // Audio PLLCLK
+	ClockVideoPll    Clock = 0x17 // Video PLLCLK
+)
+
+// Named clocks of integrated peripherals
+const (
+	ClockIpAipsTz1     Clock = (0 << 8) | CCM_CCGR0_CG0_Pos  // CCGR0, CG0
+	ClockIpAipsTz2     Clock = (0 << 8) | CCM_CCGR0_CG1_Pos  // CCGR0, CG1
+	ClockIpMqs         Clock = (0 << 8) | CCM_CCGR0_CG2_Pos  // CCGR0, CG2
+	ClockIpFlexSpiExsc Clock = (0 << 8) | CCM_CCGR0_CG3_Pos  // CCGR0, CG3
+	ClockIpSimMMain    Clock = (0 << 8) | CCM_CCGR0_CG4_Pos  // CCGR0, CG4
+	ClockIpDcp         Clock = (0 << 8) | CCM_CCGR0_CG5_Pos  // CCGR0, CG5
+	ClockIpLpuart3     Clock = (0 << 8) | CCM_CCGR0_CG6_Pos  // CCGR0, CG6
+	ClockIpCan1        Clock = (0 << 8) | CCM_CCGR0_CG7_Pos  // CCGR0, CG7
+	ClockIpCan1S       Clock = (0 << 8) | CCM_CCGR0_CG8_Pos  // CCGR0, CG8
+	ClockIpCan2        Clock = (0 << 8) | CCM_CCGR0_CG9_Pos  // CCGR0, CG9
+	ClockIpCan2S       Clock = (0 << 8) | CCM_CCGR0_CG10_Pos // CCGR0, CG10
+	ClockIpTrace       Clock = (0 << 8) | CCM_CCGR0_CG11_Pos // CCGR0, CG11
+	ClockIpGpt2        Clock = (0 << 8) | CCM_CCGR0_CG12_Pos // CCGR0, CG12
+	ClockIpGpt2S       Clock = (0 << 8) | CCM_CCGR0_CG13_Pos // CCGR0, CG13
+	ClockIpLpuart2     Clock = (0 << 8) | CCM_CCGR0_CG14_Pos // CCGR0, CG14
+	ClockIpGpio2       Clock = (0 << 8) | CCM_CCGR0_CG15_Pos // CCGR0, CG15
+
+	ClockIpLpspi1   Clock = (1 << 8) | CCM_CCGR1_CG0_Pos  // CCGR1, CG0
+	ClockIpLpspi2   Clock = (1 << 8) | CCM_CCGR1_CG1_Pos  // CCGR1, CG1
+	ClockIpLpspi3   Clock = (1 << 8) | CCM_CCGR1_CG2_Pos  // CCGR1, CG2
+	ClockIpLpspi4   Clock = (1 << 8) | CCM_CCGR1_CG3_Pos  // CCGR1, CG3
+	ClockIpAdc2     Clock = (1 << 8) | CCM_CCGR1_CG4_Pos  // CCGR1, CG4
+	ClockIpEnet     Clock = (1 << 8) | CCM_CCGR1_CG5_Pos  // CCGR1, CG5
+	ClockIpPit      Clock = (1 << 8) | CCM_CCGR1_CG6_Pos  // CCGR1, CG6
+	ClockIpAoi2     Clock = (1 << 8) | CCM_CCGR1_CG7_Pos  // CCGR1, CG7
+	ClockIpAdc1     Clock = (1 << 8) | CCM_CCGR1_CG8_Pos  // CCGR1, CG8
+	ClockIpSemcExsc Clock = (1 << 8) | CCM_CCGR1_CG9_Pos  // CCGR1, CG9
+	ClockIpGpt1     Clock = (1 << 8) | CCM_CCGR1_CG10_Pos // CCGR1, CG10
+	ClockIpGpt1S    Clock = (1 << 8) | CCM_CCGR1_CG11_Pos // CCGR1, CG11
+	ClockIpLpuart4  Clock = (1 << 8) | CCM_CCGR1_CG12_Pos // CCGR1, CG12
+	ClockIpGpio1    Clock = (1 << 8) | CCM_CCGR1_CG13_Pos // CCGR1, CG13
+	ClockIpCsu      Clock = (1 << 8) | CCM_CCGR1_CG14_Pos // CCGR1, CG14
+	ClockIpGpio5    Clock = (1 << 8) | CCM_CCGR1_CG15_Pos // CCGR1, CG15
+
+	ClockIpOcramExsc  Clock = (2 << 8) | CCM_CCGR2_CG0_Pos  // CCGR2, CG0
+	ClockIpCsi        Clock = (2 << 8) | CCM_CCGR2_CG1_Pos  // CCGR2, CG1
+	ClockIpIomuxcSnvs Clock = (2 << 8) | CCM_CCGR2_CG2_Pos  // CCGR2, CG2
+	ClockIpLpi2c1     Clock = (2 << 8) | CCM_CCGR2_CG3_Pos  // CCGR2, CG3
+	ClockIpLpi2c2     Clock = (2 << 8) | CCM_CCGR2_CG4_Pos  // CCGR2, CG4
+	ClockIpLpi2c3     Clock = (2 << 8) | CCM_CCGR2_CG5_Pos  // CCGR2, CG5
+	ClockIpOcotp      Clock = (2 << 8) | CCM_CCGR2_CG6_Pos  // CCGR2, CG6
+	ClockIpXbar3      Clock = (2 << 8) | CCM_CCGR2_CG7_Pos  // CCGR2, CG7
+	ClockIpIpmux1     Clock = (2 << 8) | CCM_CCGR2_CG8_Pos  // CCGR2, CG8
+	ClockIpIpmux2     Clock = (2 << 8) | CCM_CCGR2_CG9_Pos  // CCGR2, CG9
+	ClockIpIpmux3     Clock = (2 << 8) | CCM_CCGR2_CG10_Pos // CCGR2, CG10
+	ClockIpXbar1      Clock = (2 << 8) | CCM_CCGR2_CG11_Pos // CCGR2, CG11
+	ClockIpXbar2      Clock = (2 << 8) | CCM_CCGR2_CG12_Pos // CCGR2, CG12
+	ClockIpGpio3      Clock = (2 << 8) | CCM_CCGR2_CG13_Pos // CCGR2, CG13
+	ClockIpLcd        Clock = (2 << 8) | CCM_CCGR2_CG14_Pos // CCGR2, CG14
+	ClockIpPxp        Clock = (2 << 8) | CCM_CCGR2_CG15_Pos // CCGR2, CG15
+
+	ClockIpFlexio2       Clock = (3 << 8) | CCM_CCGR3_CG0_Pos  // CCGR3, CG0
+	ClockIpLpuart5       Clock = (3 << 8) | CCM_CCGR3_CG1_Pos  // CCGR3, CG1
+	ClockIpSemc          Clock = (3 << 8) | CCM_CCGR3_CG2_Pos  // CCGR3, CG2
+	ClockIpLpuart6       Clock = (3 << 8) | CCM_CCGR3_CG3_Pos  // CCGR3, CG3
+	ClockIpAoi1          Clock = (3 << 8) | CCM_CCGR3_CG4_Pos  // CCGR3, CG4
+	ClockIpLcdPixel      Clock = (3 << 8) | CCM_CCGR3_CG5_Pos  // CCGR3, CG5
+	ClockIpGpio4         Clock = (3 << 8) | CCM_CCGR3_CG6_Pos  // CCGR3, CG6
+	ClockIpEwm0          Clock = (3 << 8) | CCM_CCGR3_CG7_Pos  // CCGR3, CG7
+	ClockIpWdog1         Clock = (3 << 8) | CCM_CCGR3_CG8_Pos  // CCGR3, CG8
+	ClockIpFlexRam       Clock = (3 << 8) | CCM_CCGR3_CG9_Pos  // CCGR3, CG9
+	ClockIpAcmp1         Clock = (3 << 8) | CCM_CCGR3_CG10_Pos // CCGR3, CG10
+	ClockIpAcmp2         Clock = (3 << 8) | CCM_CCGR3_CG11_Pos // CCGR3, CG11
+	ClockIpAcmp3         Clock = (3 << 8) | CCM_CCGR3_CG12_Pos // CCGR3, CG12
+	ClockIpAcmp4         Clock = (3 << 8) | CCM_CCGR3_CG13_Pos // CCGR3, CG13
+	ClockIpOcram         Clock = (3 << 8) | CCM_CCGR3_CG14_Pos // CCGR3, CG14
+	ClockIpIomuxcSnvsGpr Clock = (3 << 8) | CCM_CCGR3_CG15_Pos // CCGR3, CG15
+
+	ClockIpIomuxc    Clock = (4 << 8) | CCM_CCGR4_CG1_Pos  // CCGR4, CG1
+	ClockIpIomuxcGpr Clock = (4 << 8) | CCM_CCGR4_CG2_Pos  // CCGR4, CG2
+	ClockIpBee       Clock = (4 << 8) | CCM_CCGR4_CG3_Pos  // CCGR4, CG3
+	ClockIpSimM7     Clock = (4 << 8) | CCM_CCGR4_CG4_Pos  // CCGR4, CG4
+	ClockIpTsc       Clock = (4 << 8) | CCM_CCGR4_CG5_Pos  // CCGR4, CG5
+	ClockIpSimM      Clock = (4 << 8) | CCM_CCGR4_CG6_Pos  // CCGR4, CG6
+	ClockIpSimEms    Clock = (4 << 8) | CCM_CCGR4_CG7_Pos  // CCGR4, CG7
+	ClockIpPwm1      Clock = (4 << 8) | CCM_CCGR4_CG8_Pos  // CCGR4, CG8
+	ClockIpPwm2      Clock = (4 << 8) | CCM_CCGR4_CG9_Pos  // CCGR4, CG9
+	ClockIpPwm3      Clock = (4 << 8) | CCM_CCGR4_CG10_Pos // CCGR4, CG10
+	ClockIpPwm4      Clock = (4 << 8) | CCM_CCGR4_CG11_Pos // CCGR4, CG11
+	ClockIpEnc1      Clock = (4 << 8) | CCM_CCGR4_CG12_Pos // CCGR4, CG12
+	ClockIpEnc2      Clock = (4 << 8) | CCM_CCGR4_CG13_Pos // CCGR4, CG13
+	ClockIpEnc3      Clock = (4 << 8) | CCM_CCGR4_CG14_Pos // CCGR4, CG14
+	ClockIpEnc4      Clock = (4 << 8) | CCM_CCGR4_CG15_Pos // CCGR4, CG15
+
+	ClockIpRom     Clock = (5 << 8) | CCM_CCGR5_CG0_Pos  // CCGR5, CG0
+	ClockIpFlexio1 Clock = (5 << 8) | CCM_CCGR5_CG1_Pos  // CCGR5, CG1
+	ClockIpWdog3   Clock = (5 << 8) | CCM_CCGR5_CG2_Pos  // CCGR5, CG2
+	ClockIpDma     Clock = (5 << 8) | CCM_CCGR5_CG3_Pos  // CCGR5, CG3
+	ClockIpKpp     Clock = (5 << 8) | CCM_CCGR5_CG4_Pos  // CCGR5, CG4
+	ClockIpWdog2   Clock = (5 << 8) | CCM_CCGR5_CG5_Pos  // CCGR5, CG5
+	ClockIpAipsTz4 Clock = (5 << 8) | CCM_CCGR5_CG6_Pos  // CCGR5, CG6
+	ClockIpSpdif   Clock = (5 << 8) | CCM_CCGR5_CG7_Pos  // CCGR5, CG7
+	ClockIpSimMain Clock = (5 << 8) | CCM_CCGR5_CG8_Pos  // CCGR5, CG8
+	ClockIpSai1    Clock = (5 << 8) | CCM_CCGR5_CG9_Pos  // CCGR5, CG9
+	ClockIpSai2    Clock = (5 << 8) | CCM_CCGR5_CG10_Pos // CCGR5, CG10
+	ClockIpSai3    Clock = (5 << 8) | CCM_CCGR5_CG11_Pos // CCGR5, CG11
+	ClockIpLpuart1 Clock = (5 << 8) | CCM_CCGR5_CG12_Pos // CCGR5, CG12
+	ClockIpLpuart7 Clock = (5 << 8) | CCM_CCGR5_CG13_Pos // CCGR5, CG13
+	ClockIpSnvsHp  Clock = (5 << 8) | CCM_CCGR5_CG14_Pos // CCGR5, CG14
+	ClockIpSnvsLp  Clock = (5 << 8) | CCM_CCGR5_CG15_Pos // CCGR5, CG15
+
+	ClockIpUsbOh3  Clock = (6 << 8) | CCM_CCGR6_CG0_Pos  // CCGR6, CG0
+	ClockIpUsdhc1  Clock = (6 << 8) | CCM_CCGR6_CG1_Pos  // CCGR6, CG1
+	ClockIpUsdhc2  Clock = (6 << 8) | CCM_CCGR6_CG2_Pos  // CCGR6, CG2
+	ClockIpDcdc    Clock = (6 << 8) | CCM_CCGR6_CG3_Pos  // CCGR6, CG3
+	ClockIpIpmux4  Clock = (6 << 8) | CCM_CCGR6_CG4_Pos  // CCGR6, CG4
+	ClockIpFlexSpi Clock = (6 << 8) | CCM_CCGR6_CG5_Pos  // CCGR6, CG5
+	ClockIpTrng    Clock = (6 << 8) | CCM_CCGR6_CG6_Pos  // CCGR6, CG6
+	ClockIpLpuart8 Clock = (6 << 8) | CCM_CCGR6_CG7_Pos  // CCGR6, CG7
+	ClockIpTimer4  Clock = (6 << 8) | CCM_CCGR6_CG8_Pos  // CCGR6, CG8
+	ClockIpAipsTz3 Clock = (6 << 8) | CCM_CCGR6_CG9_Pos  // CCGR6, CG9
+	ClockIpSimPer  Clock = (6 << 8) | CCM_CCGR6_CG10_Pos // CCGR6, CG10
+	ClockIpAnadig  Clock = (6 << 8) | CCM_CCGR6_CG11_Pos // CCGR6, CG11
+	ClockIpLpi2c4  Clock = (6 << 8) | CCM_CCGR6_CG12_Pos // CCGR6, CG12
+	ClockIpTimer1  Clock = (6 << 8) | CCM_CCGR6_CG13_Pos // CCGR6, CG13
+	ClockIpTimer2  Clock = (6 << 8) | CCM_CCGR6_CG14_Pos // CCGR6, CG14
+	ClockIpTimer3  Clock = (6 << 8) | CCM_CCGR6_CG15_Pos // CCGR6, CG15
+
+	ClockIpEnet2    Clock = (7 << 8) | CCM_CCGR7_CG0_Pos // CCGR7, CG0
+	ClockIpFlexSpi2 Clock = (7 << 8) | CCM_CCGR7_CG1_Pos // CCGR7, CG1
+	ClockIpAxbsL    Clock = (7 << 8) | CCM_CCGR7_CG2_Pos // CCGR7, CG2
+	ClockIpCan3     Clock = (7 << 8) | CCM_CCGR7_CG3_Pos // CCGR7, CG3
+	ClockIpCan3S    Clock = (7 << 8) | CCM_CCGR7_CG4_Pos // CCGR7, CG4
+	ClockIpAipsLite Clock = (7 << 8) | CCM_CCGR7_CG5_Pos // CCGR7, CG5
+	ClockIpFlexio3  Clock = (7 << 8) | CCM_CCGR7_CG6_Pos // CCGR7, CG6
+)
+
+// PLL name
+const (
+	ClockPllArm     Clock = ((offPllArm & 0xFFF) << 16) | CCM_ANALOG_PLL_ARM_ENABLE_Pos            // PLL ARM
+	ClockPllSys     Clock = ((offPllSys & 0xFFF) << 16) | CCM_ANALOG_PLL_SYS_ENABLE_Pos            // PLL SYS
+	ClockPllUsb1    Clock = ((offPllUsb1 & 0xFFF) << 16) | CCM_ANALOG_PLL_USB1_ENABLE_Pos          // PLL USB1
+	ClockPllAudio   Clock = ((offPllAudio & 0xFFF) << 16) | CCM_ANALOG_PLL_AUDIO_ENABLE_Pos        // PLL Audio
+	ClockPllVideo   Clock = ((offPllVideo & 0xFFF) << 16) | CCM_ANALOG_PLL_VIDEO_ENABLE_Pos        // PLL Video
+	ClockPllEnet    Clock = ((offPllEnet & 0xFFF) << 16) | CCM_ANALOG_PLL_ENET_ENABLE_Pos          // PLL Enet0
+	ClockPllEnet2   Clock = ((offPllEnet & 0xFFF) << 16) | CCM_ANALOG_PLL_ENET_ENET2_REF_EN_Pos    // PLL Enet1
+	ClockPllEnet25M Clock = ((offPllEnet & 0xFFF) << 16) | CCM_ANALOG_PLL_ENET_ENET_25M_REF_EN_Pos // PLL Enet2
+	ClockPllUsb2    Clock = ((offPllUsb2 & 0xFFF) << 16) | CCM_ANALOG_PLL_USB2_ENABLE_Pos          // PLL USB2
+)
+
+// PLL PFD name
+const (
+	ClockPfd0 Clock = 0 // PLL PFD0
+	ClockPfd1 Clock = 1 // PLL PFD1
+	ClockPfd2 Clock = 2 // PLL PFD2
+	ClockPfd3 Clock = 3 // PLL PFD3
+)
+
+// Named clock muxes of integrated peripherals
+const (
+	MuxIpPll3Sw     Clock = (offCCSR & 0xFF) | (CCM_CCSR_PLL3_SW_CLK_SEL_Pos << 8) | (((CCM_CCSR_PLL3_SW_CLK_SEL_Msk >> CCM_CCSR_PLL3_SW_CLK_SEL_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)                          // pll3_sw_clk mux name
+	MuxIpPeriph     Clock = (offCBCDR & 0xFF) | (CCM_CBCDR_PERIPH_CLK_SEL_Pos << 8) | (((CCM_CBCDR_PERIPH_CLK_SEL_Msk >> CCM_CBCDR_PERIPH_CLK_SEL_Pos) & 0x1FFF) << 13) | (CCM_CDHIPR_PERIPH_CLK_SEL_BUSY_Pos << 26) // periph mux name
+	MuxIpSemcAlt    Clock = (offCBCDR & 0xFF) | (CCM_CBCDR_SEMC_ALT_CLK_SEL_Pos << 8) | (((CCM_CBCDR_SEMC_ALT_CLK_SEL_Msk >> CCM_CBCDR_SEMC_ALT_CLK_SEL_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)                   // semc mux name
+	MuxIpSemc       Clock = (offCBCDR & 0xFF) | (CCM_CBCDR_SEMC_CLK_SEL_Pos << 8) | (((CCM_CBCDR_SEMC_CLK_SEL_Msk >> CCM_CBCDR_SEMC_CLK_SEL_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)                               // semc mux name
+	MuxIpPrePeriph  Clock = (offCBCMR & 0xFF) | (CCM_CBCMR_PRE_PERIPH_CLK_SEL_Pos << 8) | (((CCM_CBCMR_PRE_PERIPH_CLK_SEL_Msk >> CCM_CBCMR_PRE_PERIPH_CLK_SEL_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)             // pre-periph mux name
+	MuxIpTrace      Clock = (offCBCMR & 0xFF) | (CCM_CBCMR_TRACE_CLK_SEL_Pos << 8) | (((CCM_CBCMR_TRACE_CLK_SEL_Msk >> CCM_CBCMR_TRACE_CLK_SEL_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)                            // trace mux name
+	MuxIpPeriphClk2 Clock = (offCBCMR & 0xFF) | (CCM_CBCMR_PERIPH_CLK2_SEL_Pos << 8) | (((CCM_CBCMR_PERIPH_CLK2_SEL_Msk >> CCM_CBCMR_PERIPH_CLK2_SEL_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)                      // periph clock2 mux name
+	MuxIpFlexSpi2   Clock = (offCBCMR & 0xFF) | (CCM_CBCMR_FLEXSPI2_CLK_SEL_Pos << 8) | (((CCM_CBCMR_FLEXSPI2_CLK_SEL_Msk >> CCM_CBCMR_FLEXSPI2_CLK_SEL_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)                   // flexspi2 mux name
+	MuxIpLpspi      Clock = (offCBCMR & 0xFF) | (CCM_CBCMR_LPSPI_CLK_SEL_Pos << 8) | (((CCM_CBCMR_LPSPI_CLK_SEL_Msk >> CCM_CBCMR_LPSPI_CLK_SEL_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)                            // lpspi mux name
+	MuxIpFlexSpi    Clock = (offCSCMR1 & 0xFF) | (CCM_CSCMR1_FLEXSPI_CLK_SEL_Pos << 8) | (((CCM_CSCMR1_FLEXSPI_CLK_SEL_Msk >> CCM_CSCMR1_FLEXSPI_CLK_SEL_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)                  // flexspi mux name
+	MuxIpUsdhc2     Clock = (offCSCMR1 & 0xFF) | (CCM_CSCMR1_USDHC2_CLK_SEL_Pos << 8) | (((CCM_CSCMR1_USDHC2_CLK_SEL_Msk >> CCM_CSCMR1_USDHC2_CLK_SEL_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)                     // usdhc2 mux name
+	MuxIpUsdhc1     Clock = (offCSCMR1 & 0xFF) | (CCM_CSCMR1_USDHC1_CLK_SEL_Pos << 8) | (((CCM_CSCMR1_USDHC1_CLK_SEL_Msk >> CCM_CSCMR1_USDHC1_CLK_SEL_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)                     // usdhc1 mux name
+	MuxIpSai3       Clock = (offCSCMR1 & 0xFF) | (CCM_CSCMR1_SAI3_CLK_SEL_Pos << 8) | (((CCM_CSCMR1_SAI3_CLK_SEL_Msk >> CCM_CSCMR1_SAI3_CLK_SEL_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)                           // sai3 mux name
+	MuxIpSai2       Clock = (offCSCMR1 & 0xFF) | (CCM_CSCMR1_SAI2_CLK_SEL_Pos << 8) | (((CCM_CSCMR1_SAI2_CLK_SEL_Msk >> CCM_CSCMR1_SAI2_CLK_SEL_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)                           // sai2 mux name
+	MuxIpSai1       Clock = (offCSCMR1 & 0xFF) | (CCM_CSCMR1_SAI1_CLK_SEL_Pos << 8) | (((CCM_CSCMR1_SAI1_CLK_SEL_Msk >> CCM_CSCMR1_SAI1_CLK_SEL_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)                           // sai1 mux name
+	MuxIpPerclk     Clock = (offCSCMR1 & 0xFF) | (CCM_CSCMR1_PERCLK_CLK_SEL_Pos << 8) | (((CCM_CSCMR1_PERCLK_CLK_SEL_Msk >> CCM_CSCMR1_PERCLK_CLK_SEL_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)                     // perclk mux name
+	MuxIpFlexio2    Clock = (offCSCMR2 & 0xFF) | (CCM_CSCMR2_FLEXIO2_CLK_SEL_Pos << 8) | (((CCM_CSCMR2_FLEXIO2_CLK_SEL_Msk >> CCM_CSCMR2_FLEXIO2_CLK_SEL_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)                  // flexio2 mux name
+	MuxIpCan        Clock = (offCSCMR2 & 0xFF) | (CCM_CSCMR2_CAN_CLK_SEL_Pos << 8) | (((CCM_CSCMR2_CAN_CLK_SEL_Msk >> CCM_CSCMR2_CAN_CLK_SEL_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)                              // can mux name
+	MuxIpUart       Clock = (offCSCDR1 & 0xFF) | (CCM_CSCDR1_UART_CLK_SEL_Pos << 8) | (((CCM_CSCDR1_UART_CLK_SEL_Msk >> CCM_CSCDR1_UART_CLK_SEL_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)                           // uart mux name
+	MuxIpSpdif      Clock = (offCDCDR & 0xFF) | (CCM_CDCDR_SPDIF0_CLK_SEL_Pos << 8) | (((CCM_CDCDR_SPDIF0_CLK_SEL_Msk >> CCM_CDCDR_SPDIF0_CLK_SEL_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)                         // spdif mux name
+	MuxIpFlexio1    Clock = (offCDCDR & 0xFF) | (CCM_CDCDR_FLEXIO1_CLK_SEL_Pos << 8) | (((CCM_CDCDR_FLEXIO1_CLK_SEL_Msk >> CCM_CDCDR_FLEXIO1_CLK_SEL_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)                      // flexio1 mux name
+	MuxIpLpi2c      Clock = (offCSCDR2 & 0xFF) | (CCM_CSCDR2_LPI2C_CLK_SEL_Pos << 8) | (((CCM_CSCDR2_LPI2C_CLK_SEL_Msk >> CCM_CSCDR2_LPI2C_CLK_SEL_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)                        // lpi2c mux name
+	MuxIpLcdifPre   Clock = (offCSCDR2 & 0xFF) | (CCM_CSCDR2_LCDIF_PRE_CLK_SEL_Pos << 8) | (((CCM_CSCDR2_LCDIF_PRE_CLK_SEL_Msk >> CCM_CSCDR2_LCDIF_PRE_CLK_SEL_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)            // lcdif pre mux name
+	MuxIpCsi        Clock = (offCSCDR3 & 0xFF) | (CCM_CSCDR3_CSI_CLK_SEL_Pos << 8) | (((CCM_CSCDR3_CSI_CLK_SEL_Msk >> CCM_CSCDR3_CSI_CLK_SEL_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)                              // csi mux name
+)
+
+// Named hardware clock divisors of integrated peripherals
+const (
+	DivIpArm        Clock = (offCACRR & 0xFF) | (CCM_CACRR_ARM_PODF_Pos << 8) | (((CCM_CACRR_ARM_PODF_Msk >> CCM_CACRR_ARM_PODF_Pos) & 0x1FFF) << 13) | (CCM_CDHIPR_ARM_PODF_BUSY_Pos << 26)           // core div name
+	DivIpPeriphClk2 Clock = (offCBCDR & 0xFF) | (CCM_CBCDR_PERIPH_CLK2_PODF_Pos << 8) | (((CCM_CBCDR_PERIPH_CLK2_PODF_Msk >> CCM_CBCDR_PERIPH_CLK2_PODF_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)     // periph clock2 div name
+	DivIpSemc       Clock = (offCBCDR & 0xFF) | (CCM_CBCDR_SEMC_PODF_Pos << 8) | (((CCM_CBCDR_SEMC_PODF_Msk >> CCM_CBCDR_SEMC_PODF_Pos) & 0x1FFF) << 13) | (CCM_CDHIPR_SEMC_PODF_BUSY_Pos << 26)       // semc div name
+	DivIpAhb        Clock = (offCBCDR & 0xFF) | (CCM_CBCDR_AHB_PODF_Pos << 8) | (((CCM_CBCDR_AHB_PODF_Msk >> CCM_CBCDR_AHB_PODF_Pos) & 0x1FFF) << 13) | (CCM_CDHIPR_AHB_PODF_BUSY_Pos << 26)           // ahb div name
+	DivIpIpg        Clock = (offCBCDR & 0xFF) | (CCM_CBCDR_IPG_PODF_Pos << 8) | (((CCM_CBCDR_IPG_PODF_Msk >> CCM_CBCDR_IPG_PODF_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)                             // ipg div name
+	DivIpFlexSpi2   Clock = (offCBCMR & 0xFF) | (CCM_CBCMR_FLEXSPI2_PODF_Pos << 8) | (((CCM_CBCMR_FLEXSPI2_PODF_Msk >> CCM_CBCMR_FLEXSPI2_PODF_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)              // flexspi2 div name
+	DivIpLpspi      Clock = (offCBCMR & 0xFF) | (CCM_CBCMR_LPSPI_PODF_Pos << 8) | (((CCM_CBCMR_LPSPI_PODF_Msk >> CCM_CBCMR_LPSPI_PODF_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)                       // lpspi div name
+	DivIpLcdif      Clock = (offCBCMR & 0xFF) | (CCM_CBCMR_LCDIF_PODF_Pos << 8) | (((CCM_CBCMR_LCDIF_PODF_Msk >> CCM_CBCMR_LCDIF_PODF_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)                       // lcdif div name
+	DivIpFlexSpi    Clock = (offCSCMR1 & 0xFF) | (CCM_CSCMR1_FLEXSPI_PODF_Pos << 8) | (((CCM_CSCMR1_FLEXSPI_PODF_Msk >> CCM_CSCMR1_FLEXSPI_PODF_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)             // flexspi div name
+	DivIpPerclk     Clock = (offCSCMR1 & 0xFF) | (CCM_CSCMR1_PERCLK_PODF_Pos << 8) | (((CCM_CSCMR1_PERCLK_PODF_Msk >> CCM_CSCMR1_PERCLK_PODF_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)                // perclk div name
+	DivIpCan        Clock = (offCSCMR2 & 0xFF) | (CCM_CSCMR2_CAN_CLK_PODF_Pos << 8) | (((CCM_CSCMR2_CAN_CLK_PODF_Msk >> CCM_CSCMR2_CAN_CLK_PODF_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)             // can div name
+	DivIpTrace      Clock = (offCSCDR1 & 0xFF) | (CCM_CSCDR1_TRACE_PODF_Pos << 8) | (((CCM_CSCDR1_TRACE_PODF_Msk >> CCM_CSCDR1_TRACE_PODF_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)                   // trace div name
+	DivIpUsdhc2     Clock = (offCSCDR1 & 0xFF) | (CCM_CSCDR1_USDHC2_PODF_Pos << 8) | (((CCM_CSCDR1_USDHC2_PODF_Msk >> CCM_CSCDR1_USDHC2_PODF_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)                // usdhc2 div name
+	DivIpUsdhc1     Clock = (offCSCDR1 & 0xFF) | (CCM_CSCDR1_USDHC1_PODF_Pos << 8) | (((CCM_CSCDR1_USDHC1_PODF_Msk >> CCM_CSCDR1_USDHC1_PODF_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)                // usdhc1 div name
+	DivIpUart       Clock = (offCSCDR1 & 0xFF) | (CCM_CSCDR1_UART_CLK_PODF_Pos << 8) | (((CCM_CSCDR1_UART_CLK_PODF_Msk >> CCM_CSCDR1_UART_CLK_PODF_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)          // uart div name
+	DivIpFlexio2    Clock = (offCS1CDR & 0xFF) | (CCM_CS1CDR_FLEXIO2_CLK_PODF_Pos << 8) | (((CCM_CS1CDR_FLEXIO2_CLK_PODF_Msk >> CCM_CS1CDR_FLEXIO2_CLK_PODF_Pos) & 0x1FFF) << 13) | (noBusyWait << 26) // flexio2 pre div name
+	DivIpSai3Pre    Clock = (offCS1CDR & 0xFF) | (CCM_CS1CDR_SAI3_CLK_PRED_Pos << 8) | (((CCM_CS1CDR_SAI3_CLK_PRED_Msk >> CCM_CS1CDR_SAI3_CLK_PRED_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)          // sai3 pre div name
+	DivIpSai3       Clock = (offCS1CDR & 0xFF) | (CCM_CS1CDR_SAI3_CLK_PODF_Pos << 8) | (((CCM_CS1CDR_SAI3_CLK_PODF_Msk >> CCM_CS1CDR_SAI3_CLK_PODF_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)          // sai3 div name
+	DivIpFlexio2Pre Clock = (offCS1CDR & 0xFF) | (CCM_CS1CDR_FLEXIO2_CLK_PRED_Pos << 8) | (((CCM_CS1CDR_FLEXIO2_CLK_PRED_Msk >> CCM_CS1CDR_FLEXIO2_CLK_PRED_Pos) & 0x1FFF) << 13) | (noBusyWait << 26) // sai3 pre div name
+	DivIpSai1Pre    Clock = (offCS1CDR & 0xFF) | (CCM_CS1CDR_SAI1_CLK_PRED_Pos << 8) | (((CCM_CS1CDR_SAI1_CLK_PRED_Msk >> CCM_CS1CDR_SAI1_CLK_PRED_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)          // sai1 pre div name
+	DivIpSai1       Clock = (offCS1CDR & 0xFF) | (CCM_CS1CDR_SAI1_CLK_PODF_Pos << 8) | (((CCM_CS1CDR_SAI1_CLK_PODF_Msk >> CCM_CS1CDR_SAI1_CLK_PODF_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)          // sai1 div name
+	DivIpSai2Pre    Clock = (offCS2CDR & 0xFF) | (CCM_CS2CDR_SAI2_CLK_PRED_Pos << 8) | (((CCM_CS2CDR_SAI2_CLK_PRED_Msk >> CCM_CS2CDR_SAI2_CLK_PRED_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)          // sai2 pre div name
+	DivIpSai2       Clock = (offCS2CDR & 0xFF) | (CCM_CS2CDR_SAI2_CLK_PODF_Pos << 8) | (((CCM_CS2CDR_SAI2_CLK_PODF_Msk >> CCM_CS2CDR_SAI2_CLK_PODF_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)          // sai2 div name
+	DivIpSpdif0Pre  Clock = (offCDCDR & 0xFF) | (CCM_CDCDR_SPDIF0_CLK_PRED_Pos << 8) | (((CCM_CDCDR_SPDIF0_CLK_PRED_Msk >> CCM_CDCDR_SPDIF0_CLK_PRED_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)        // spdif pre div name
+	DivIpSpdif0     Clock = (offCDCDR & 0xFF) | (CCM_CDCDR_SPDIF0_CLK_PODF_Pos << 8) | (((CCM_CDCDR_SPDIF0_CLK_PODF_Msk >> CCM_CDCDR_SPDIF0_CLK_PODF_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)        // spdif div name
+	DivIpFlexio1Pre Clock = (offCDCDR & 0xFF) | (CCM_CDCDR_FLEXIO1_CLK_PRED_Pos << 8) | (((CCM_CDCDR_FLEXIO1_CLK_PRED_Msk >> CCM_CDCDR_FLEXIO1_CLK_PRED_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)     // flexio1 pre div name
+	DivIpFlexio1    Clock = (offCDCDR & 0xFF) | (CCM_CDCDR_FLEXIO1_CLK_PODF_Pos << 8) | (((CCM_CDCDR_FLEXIO1_CLK_PODF_Msk >> CCM_CDCDR_FLEXIO1_CLK_PODF_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)     // flexio1 div name
+	DivIpLpi2c      Clock = (offCSCDR2 & 0xFF) | (CCM_CSCDR2_LPI2C_CLK_PODF_Pos << 8) | (((CCM_CSCDR2_LPI2C_CLK_PODF_Msk >> CCM_CSCDR2_LPI2C_CLK_PODF_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)       // lpi2c div name
+	DivIpLcdifPre   Clock = (offCSCDR2 & 0xFF) | (CCM_CSCDR2_LCDIF_PRED_Pos << 8) | (((CCM_CSCDR2_LCDIF_PRED_Msk >> CCM_CSCDR2_LCDIF_PRED_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)                   // lcdif pre div name
+	DivIpCsi        Clock = (offCSCDR3 & 0xFF) | (CCM_CSCDR3_CSI_PODF_Pos << 8) | (((CCM_CSCDR3_CSI_PODF_Msk >> CCM_CSCDR3_CSI_PODF_Pos) & 0x1FFF) << 13) | (noBusyWait << 26)                         // csi div name
+)
+
+// Selected clock offsets
+const (
+	offCCSR   = 0x0C
+	offCBCDR  = 0x14
+	offCBCMR  = 0x18
+	offCSCMR1 = 0x1C
+	offCSCMR2 = 0x20
+	offCSCDR1 = 0x24
+	offCDCDR  = 0x30
+	offCSCDR2 = 0x38
+	offCSCDR3 = 0x3C
+	offCACRR  = 0x10
+	offCS1CDR = 0x28
+	offCS2CDR = 0x2C
+
+	offPllArm   = 0x00
+	offPllSys   = 0x30
+	offPllUsb1  = 0x10
+	offPllAudio = 0x70
+	offPllVideo = 0xA0
+	offPllEnet  = 0xE0
+	offPllUsb2  = 0x20
+
+	noBusyWait = 0x20
+)
+
+// analog PLL definition
+const (
+	pllBypassPos       = 16
+	pllBypassClkSrcMsk = 0xC000
+	pllBypassClkSrcPos = 14
+)
+
+// PLL clock source, bypass cloco source also
+const (
+	pllSrc24M   = 0 // Pll clock source 24M
+	pllSrcClkPN = 1 // Pll clock source CLK1_P and CLK1_N
+)
+
+const (
+	clockNotNeeded     uint32 = 0 // Clock is off during all modes
+	clockNeededRun     uint32 = 1 // Clock is on in run mode, but off in WAIT and STOP modes
+	clockNeededRunWait uint32 = 3 // Clock is on during all modes, except STOP mode
+)
+
+// getGate returns the CCM clock gating register for the receiver clk.
+func (clk Clock) getGate() *volatile.Register32 {
+	switch clk >> 8 {
+	case 0:
+		return &CCM.CCGR0
+	case 1:
+		return &CCM.CCGR1
+	case 2:
+		return &CCM.CCGR2
+	case 3:
+		return &CCM.CCGR3
+	case 4:
+		return &CCM.CCGR4
+	case 5:
+		return &CCM.CCGR5
+	case 6:
+		return &CCM.CCGR6
+	case 7:
+		return &CCM.CCGR7
+	default:
+		panic("nxp: invalid clock")
+	}
+}
+
+// setGate enables or disables the receiver clk using its gating register.
+func (clk Clock) setGate(value uint32) {
+	reg := clk.getGate()
+	shift := clk & 0x1F
+	reg.Set((reg.Get() & ^(3 << shift)) | (value << shift))
+}
+
+func (clk Clock) setCcm(value uint32) {
+	const ccmBase = 0x400fc000
+	reg := (*volatile.Register32)(unsafe.Pointer(uintptr(ccmBase + (uint32(clk) & 0xFF))))
+	msk := ((uint32(clk) >> 13) & 0x1FFF) << ((uint32(clk) >> 8) & 0x1F)
+	pos := (uint32(clk) >> 8) & 0x1F
+	bsy := (uint32(clk) >> 26) & 0x3F
+	reg.Set((reg.Get() & ^uint32(msk)) | ((value << pos) & msk))
+	if bsy < noBusyWait {
+		for CCM.CDHIPR.HasBits(1 << bsy) {
+		}
+	}
+}
+
+func setSysPfd(value ...uint32) {
+	for i, val := range value {
+		pfd528 := CCM_ANALOG.PFD_528.Get() &
+			^((CCM_ANALOG_PFD_528_PFD0_CLKGATE_Msk | CCM_ANALOG_PFD_528_PFD0_FRAC_Msk) << (8 * uint32(i)))
+		frac := (val << CCM_ANALOG_PFD_528_PFD0_FRAC_Pos) & CCM_ANALOG_PFD_528_PFD0_FRAC_Msk
+		// disable the clock output first
+		CCM_ANALOG.PFD_528.Set(pfd528 | (CCM_ANALOG_PFD_528_PFD0_CLKGATE_Msk << (8 * uint32(i))))
+		// set the new value and enable output
+		CCM_ANALOG.PFD_528.Set(pfd528 | (frac << (8 * uint32(i))))
+	}
+}
+
+func setUsb1Pfd(value ...uint32) {
+	for i, val := range value {
+		pfd480 := CCM_ANALOG.PFD_480.Get() &
+			^((CCM_ANALOG_PFD_480_PFD0_CLKGATE_Msk | CCM_ANALOG_PFD_480_PFD0_FRAC_Msk) << (8 * uint32(i)))
+		frac := (val << CCM_ANALOG_PFD_480_PFD0_FRAC_Pos) & CCM_ANALOG_PFD_480_PFD0_FRAC_Msk
+		// disable the clock output first
+		CCM_ANALOG.PFD_480.Set(pfd480 | (CCM_ANALOG_PFD_480_PFD0_CLKGATE_Msk << (8 * uint32(i))))
+		// set the new value and enable output
+		CCM_ANALOG.PFD_480.Set(pfd480 | (frac << (8 * uint32(i))))
+	}
+}
+
+// PLL configuration for ARM
+type ClockConfigArmPll struct {
+	LoopDivider uint32 // PLL loop divider. Valid range for divider value: 54-108. Fout=Fin*LoopDivider/2.
+	Src         uint8  // Pll clock source, reference _clock_pll_clk_src
+}
+
+func (cfg ClockConfigArmPll) Configure() {
+
+	// bypass PLL first
+	src := (uint32(cfg.Src) << CCM_ANALOG_PLL_ARM_BYPASS_CLK_SRC_Pos) & CCM_ANALOG_PLL_ARM_BYPASS_CLK_SRC_Msk
+	CCM_ANALOG.PLL_ARM.Set(
+		(CCM_ANALOG.PLL_ARM.Get() & ^uint32(CCM_ANALOG_PLL_ARM_BYPASS_CLK_SRC_Msk)) |
+			CCM_ANALOG_PLL_ARM_BYPASS_Msk | src)
+
+	sel := (cfg.LoopDivider << CCM_ANALOG_PLL_ARM_DIV_SELECT_Pos) & CCM_ANALOG_PLL_ARM_DIV_SELECT_Msk
+	CCM_ANALOG.PLL_ARM.Set(
+		(CCM_ANALOG.PLL_ARM.Get() & ^uint32(CCM_ANALOG_PLL_ARM_DIV_SELECT_Msk|CCM_ANALOG_PLL_ARM_POWERDOWN_Msk)) |
+			CCM_ANALOG_PLL_ARM_ENABLE_Msk | sel)
+
+	for !CCM_ANALOG.PLL_ARM.HasBits(CCM_ANALOG_PLL_ARM_LOCK_Msk) {
+	}
+
+	// disable bypass
+	CCM_ANALOG.PLL_ARM.ClearBits(CCM_ANALOG_PLL_ARM_BYPASS_Msk)
+}
+
+// PLL configuration for System
+type ClockConfigSysPll struct {
+	LoopDivider uint8  // PLL loop divider. Intended to be 1 (528M): 0 - Fout=Fref*20, 1 - Fout=Fref*22
+	Numerator   uint32 // 30 bit Numerator of fractional loop divider.
+	Denominator uint32 // 30 bit Denominator of fractional loop divider
+	Src         uint8  // Pll clock source, reference _clock_pll_clk_src
+	SsStop      uint16 // Stop value to get frequency change.
+	SsEnable    uint8  // Enable spread spectrum modulation
+	SsStep      uint16 // Step value to get frequency change step.
+}
+
+func (cfg ClockConfigSysPll) Configure(pfd ...uint32) {
+
+	// bypass PLL first
+	src := (uint32(cfg.Src) << CCM_ANALOG_PLL_SYS_BYPASS_CLK_SRC_Pos) & CCM_ANALOG_PLL_SYS_BYPASS_CLK_SRC_Msk
+	CCM_ANALOG.PLL_SYS.Set(
+		(CCM_ANALOG.PLL_SYS.Get() & ^uint32(CCM_ANALOG_PLL_SYS_BYPASS_CLK_SRC_Msk)) |
+			CCM_ANALOG_PLL_SYS_BYPASS_Msk | src)
+
+	sel := (uint32(cfg.LoopDivider) << CCM_ANALOG_PLL_SYS_DIV_SELECT_Pos) & CCM_ANALOG_PLL_SYS_DIV_SELECT_Msk
+	CCM_ANALOG.PLL_SYS.Set(
+		(CCM_ANALOG.PLL_SYS.Get() & ^uint32(CCM_ANALOG_PLL_SYS_DIV_SELECT_Msk|CCM_ANALOG_PLL_SYS_POWERDOWN_Msk)) |
+			CCM_ANALOG_PLL_SYS_ENABLE_Msk | sel)
+
+	// initialize the fractional mode
+	CCM_ANALOG.PLL_SYS_NUM.Set((cfg.Numerator << CCM_ANALOG_PLL_SYS_NUM_A_Pos) & CCM_ANALOG_PLL_SYS_NUM_A_Msk)
+	CCM_ANALOG.PLL_SYS_DENOM.Set((cfg.Denominator << CCM_ANALOG_PLL_SYS_DENOM_B_Pos) & CCM_ANALOG_PLL_SYS_DENOM_B_Msk)
+
+	// initialize the spread spectrum mode
+	inc := (uint32(cfg.SsStep) << CCM_ANALOG_PLL_SYS_SS_STEP_Pos) & CCM_ANALOG_PLL_SYS_SS_STEP_Msk
+	enb := (uint32(cfg.SsEnable) << CCM_ANALOG_PLL_SYS_SS_ENABLE_Pos) & CCM_ANALOG_PLL_SYS_SS_ENABLE_Msk
+	stp := (uint32(cfg.SsStop) << CCM_ANALOG_PLL_SYS_SS_STOP_Pos) & CCM_ANALOG_PLL_SYS_SS_STOP_Msk
+	CCM_ANALOG.PLL_SYS_SS.Set(inc | enb | stp)
+
+	for !CCM_ANALOG.PLL_SYS.HasBits(CCM_ANALOG_PLL_SYS_LOCK_Msk) {
+	}
+
+	// disable bypass
+	CCM_ANALOG.PLL_SYS.ClearBits(CCM_ANALOG_PLL_SYS_BYPASS_Msk)
+
+	// update PFDs after update
+	setSysPfd(pfd...)
+}
+
+// PLL configuration for USB
+type ClockConfigUsbPll struct {
+	Instance    uint8 // USB PLL number (1 or 2)
+	LoopDivider uint8 // PLL loop divider: 0 - Fout=Fref*20, 1 - Fout=Fref*22
+	Src         uint8 // Pll clock source, reference _clock_pll_clk_src
+}
+
+func (cfg ClockConfigUsbPll) Configure(pfd ...uint32) {
+
+	switch cfg.Instance {
+	case 1:
+
+		// bypass PLL first
+		src := (uint32(cfg.Src) << CCM_ANALOG_PLL_USB1_BYPASS_CLK_SRC_Pos) & CCM_ANALOG_PLL_USB1_BYPASS_CLK_SRC_Msk
+		CCM_ANALOG.PLL_USB1.Set(
+			(CCM_ANALOG.PLL_USB1.Get() & ^uint32(CCM_ANALOG_PLL_USB1_BYPASS_CLK_SRC_Msk)) |
+				CCM_ANALOG_PLL_USB1_BYPASS_Msk | src)
+
+		sel := uint32((cfg.LoopDivider << CCM_ANALOG_PLL_USB1_DIV_SELECT_Pos) & CCM_ANALOG_PLL_USB1_DIV_SELECT_Msk)
+		CCM_ANALOG.PLL_USB1_SET.Set(
+			(CCM_ANALOG.PLL_USB1.Get() & ^uint32(CCM_ANALOG_PLL_USB1_DIV_SELECT_Msk)) |
+				CCM_ANALOG_PLL_USB1_ENABLE_Msk | CCM_ANALOG_PLL_USB1_POWER_Msk |
+				CCM_ANALOG_PLL_USB1_EN_USB_CLKS_Msk | sel)
+
+		for !CCM_ANALOG.PLL_USB1.HasBits(CCM_ANALOG_PLL_USB1_LOCK_Msk) {
+		}
+
+		// disable bypass
+		CCM_ANALOG.PLL_USB1_CLR.Set(CCM_ANALOG_PLL_USB1_BYPASS_Msk)
+
+		// update PFDs after update
+		setUsb1Pfd(pfd...)
+
+	case 2:
+		// bypass PLL first
+		src := (uint32(cfg.Src) << CCM_ANALOG_PLL_USB2_BYPASS_CLK_SRC_Pos) & CCM_ANALOG_PLL_USB2_BYPASS_CLK_SRC_Msk
+		CCM_ANALOG.PLL_USB2.Set(
+			(CCM_ANALOG.PLL_USB2.Get() & ^uint32(CCM_ANALOG_PLL_USB2_BYPASS_CLK_SRC_Msk)) |
+				CCM_ANALOG_PLL_USB2_BYPASS_Msk | src)
+
+		sel := uint32((cfg.LoopDivider << CCM_ANALOG_PLL_USB2_DIV_SELECT_Pos) & CCM_ANALOG_PLL_USB2_DIV_SELECT_Msk)
+		CCM_ANALOG.PLL_USB2.Set(
+			(CCM_ANALOG.PLL_USB2.Get() & ^uint32(CCM_ANALOG_PLL_USB2_DIV_SELECT_Msk)) |
+				CCM_ANALOG_PLL_USB2_ENABLE_Msk | CCM_ANALOG_PLL_USB2_POWER_Msk |
+				CCM_ANALOG_PLL_USB2_EN_USB_CLKS_Msk | sel)
+
+		for !CCM_ANALOG.PLL_USB2.HasBits(CCM_ANALOG_PLL_USB2_LOCK_Msk) {
+		}
+
+		// disable bypass
+		CCM_ANALOG.PLL_USB2.ClearBits(CCM_ANALOG_PLL_USB2_BYPASS_Msk)
+
+	default:
+		panic("nxp: invalid USB PLL")
+	}
+}
@@ -0,0 +1,29 @@
+// Hand created file. DO NOT DELETE.
+// Hardfault aliases for definitions that have inconsistent naming (which are
+// auto-generated by gen-device-svd.go) among devices in package nxp.
+
+// +build nxp,mimxrt1062
+
+package nxp
+
+const (
+	HardFault_CFSR_IACCVIOL    = SCB_CFSR_IACCVIOL
+	HardFault_CFSR_DACCVIOL    = SCB_CFSR_DACCVIOL
+	HardFault_CFSR_MUNSTKERR   = SCB_CFSR_MUNSTKERR
+	HardFault_CFSR_MSTKERR     = SCB_CFSR_MSTKERR
+	HardFault_CFSR_MLSPERR     = SCB_CFSR_MLSPERR
+	HardFault_CFSR_IBUSERR     = SCB_CFSR_IBUSERR
+	HardFault_CFSR_PRECISERR   = SCB_CFSR_PRECISERR
+	HardFault_CFSR_IMPRECISERR = SCB_CFSR_IMPRECISERR
+	HardFault_CFSR_UNSTKERR    = SCB_CFSR_UNSTKERR
+	HardFault_CFSR_STKERR      = SCB_CFSR_STKERR
+	HardFault_CFSR_LSPERR      = SCB_CFSR_LSPERR
+	HardFault_CFSR_UNDEFINSTR  = SCB_CFSR_UNDEFINSTR
+	HardFault_CFSR_INVSTATE    = SCB_CFSR_INVSTATE
+	HardFault_CFSR_INVPC       = SCB_CFSR_INVPC
+	HardFault_CFSR_NOCP        = SCB_CFSR_NOCP
+	HardFault_CFSR_UNALIGNED   = SCB_CFSR_UNALIGNED
+	HardFault_CFSR_DIVBYZERO   = SCB_CFSR_DIVBYZERO
+	HardFault_CFSR_MMARVALID   = SCB_CFSR_MMARVALID
+	HardFault_CFSR_BFARVALID   = SCB_CFSR_BFARVALID
+)
@@ -0,0 +1,278 @@
+// Hand created file. DO NOT DELETE.
+// Type definitions, fields, and constants associated with the MPU peripheral
+// of the NXP MIMXRT1062.
+
+// +build nxp,mimxrt1062
+
+package nxp
+
+import (
+	"device/arm"
+	"runtime/volatile"
+	"unsafe"
+)
+
+type MPU_Type struct {
+	TYPE    volatile.Register32 // 0x000 (R/ ) - MPU Type Register
+	CTRL    volatile.Register32 // 0x004 (R/W) - MPU Control Register
+	RNR     volatile.Register32 // 0x008 (R/W) - MPU Region RNRber Register
+	RBAR    volatile.Register32 // 0x00C (R/W) - MPU Region Base Address Register
+	RASR    volatile.Register32 // 0x010 (R/W) - MPU Region Attribute and Size Register
+	RBAR_A1 volatile.Register32 // 0x014 (R/W) - MPU Alias 1 Region Base Address Register
+	RASR_A1 volatile.Register32 // 0x018 (R/W) - MPU Alias 1 Region Attribute and Size Register
+	RBAR_A2 volatile.Register32 // 0x01C (R/W) - MPU Alias 2 Region Base Address Register
+	RASR_A2 volatile.Register32 // 0x020 (R/W) - MPU Alias 2 Region Attribute and Size Register
+	RBAR_A3 volatile.Register32 // 0x024 (R/W) - MPU Alias 3 Region Base Address Register
+	RASR_A3 volatile.Register32 // 0x028 (R/W) - MPU Alias 3 Region Attribute and Size Register
+}
+
+var MPU = (*MPU_Type)(unsafe.Pointer(uintptr(0xe000ed90)))
+
+type (
+	RegionSize  uint32
+	AccessPerms uint32
+	Extension   uint32
+)
+
+// MPU Control Register Definitions
+const (
+	MPU_CTRL_PRIVDEFENA_Pos = 2                            // MPU CTRL: PRIVDEFENA Position
+	MPU_CTRL_PRIVDEFENA_Msk = 1 << MPU_CTRL_PRIVDEFENA_Pos // MPU CTRL: PRIVDEFENA Mask
+	MPU_CTRL_HFNMIENA_Pos   = 1                            // MPU CTRL: HFNMIENA Position
+	MPU_CTRL_HFNMIENA_Msk   = 1 << MPU_CTRL_HFNMIENA_Pos   // MPU CTRL: HFNMIENA Mask
+	MPU_CTRL_ENABLE_Pos     = 0                            // MPU CTRL: ENABLE Position
+	MPU_CTRL_ENABLE_Msk     = 1                            // MPU CTRL: ENABLE Mask
+)
+
+// MPU Region Base Address Register Definitions
+const (
+	MPU_RBAR_ADDR_Pos   = 5                              // MPU RBAR: ADDR Position
+	MPU_RBAR_ADDR_Msk   = 0x7FFFFFF << MPU_RBAR_ADDR_Pos // MPU RBAR: ADDR Mask
+	MPU_RBAR_VALID_Pos  = 4                              // MPU RBAR: VALID Position
+	MPU_RBAR_VALID_Msk  = 1 << MPU_RBAR_VALID_Pos        // MPU RBAR: VALID Mask
+	MPU_RBAR_REGION_Pos = 0                              // MPU RBAR: REGION Position
+	MPU_RBAR_REGION_Msk = 0xF                            // MPU RBAR: REGION Mask
+)
+
+// MPU Region Attribute and Size Register Definitions
+const (
+	MPU_RASR_ATTRS_Pos  = 16                           // MPU RASR: MPU Region Attribute field Position
+	MPU_RASR_ATTRS_Msk  = 0xFFFF << MPU_RASR_ATTRS_Pos // MPU RASR: MPU Region Attribute field Mask
+	MPU_RASR_XN_Pos     = 28                           // MPU RASR: ATTRS.XN Position
+	MPU_RASR_XN_Msk     = 1 << MPU_RASR_XN_Pos         // MPU RASR: ATTRS.XN Mask
+	MPU_RASR_AP_Pos     = 24                           // MPU RASR: ATTRS.AP Position
+	MPU_RASR_AP_Msk     = 0x7 << MPU_RASR_AP_Pos       // MPU RASR: ATTRS.AP Mask
+	MPU_RASR_TEX_Pos    = 19                           // MPU RASR: ATTRS.TEX Position
+	MPU_RASR_TEX_Msk    = 0x7 << MPU_RASR_TEX_Pos      // MPU RASR: ATTRS.TEX Mask
+	MPU_RASR_S_Pos      = 18                           // MPU RASR: ATTRS.S Position
+	MPU_RASR_S_Msk      = 1 << MPU_RASR_S_Pos          // MPU RASR: ATTRS.S Mask
+	MPU_RASR_C_Pos      = 17                           // MPU RASR: ATTRS.C Position
+	MPU_RASR_C_Msk      = 1 << MPU_RASR_C_Pos          // MPU RASR: ATTRS.C Mask
+	MPU_RASR_B_Pos      = 16                           // MPU RASR: ATTRS.B Position
+	MPU_RASR_B_Msk      = 1 << MPU_RASR_B_Pos          // MPU RASR: ATTRS.B Mask
+	MPU_RASR_SRD_Pos    = 8                            // MPU RASR: Sub-Region Disable Position
+	MPU_RASR_SRD_Msk    = 0xFF << MPU_RASR_SRD_Pos     // MPU RASR: Sub-Region Disable Mask
+	MPU_RASR_SIZE_Pos   = 1                            // MPU RASR: Region Size Field Position
+	MPU_RASR_SIZE_Msk   = 0x1F << MPU_RASR_SIZE_Pos    // MPU RASR: Region Size Field Mask
+	MPU_RASR_ENABLE_Pos = 0                            // MPU RASR: Region enable bit Position
+	MPU_RASR_ENABLE_Msk = 1                            // MPU RASR: Region enable bit Disable Mask
+)
+
+const (
+	SCB_DCISW_WAY_Pos = 30                         // SCB DCISW: Way Position
+	SCB_DCISW_WAY_Msk = 3 << SCB_DCISW_WAY_Pos     // SCB DCISW: Way Mask
+	SCB_DCISW_SET_Pos = 5                          // SCB DCISW: Set Position
+	SCB_DCISW_SET_Msk = 0x1FF << SCB_DCISW_SET_Pos // SCB DCISW: Set Mask
+)
+
+const (
+	SCB_DCCISW_WAY_Pos = 30                          // SCB DCCISW: Way Position
+	SCB_DCCISW_WAY_Msk = 3 << SCB_DCCISW_WAY_Pos     // SCB DCCISW: Way Mask
+	SCB_DCCISW_SET_Pos = 5                           // SCB DCCISW: Set Position
+	SCB_DCCISW_SET_Msk = 0x1FF << SCB_DCCISW_SET_Pos // SCB DCCISW: Set Mask
+)
+
+const (
+	RGNSZ_32B   RegionSize = 0x04 // MPU Region Size 32 Bytes
+	RGNSZ_64B   RegionSize = 0x05 // MPU Region Size 64 Bytes
+	RGNSZ_128B  RegionSize = 0x06 // MPU Region Size 128 Bytes
+	RGNSZ_256B  RegionSize = 0x07 // MPU Region Size 256 Bytes
+	RGNSZ_512B  RegionSize = 0x08 // MPU Region Size 512 Bytes
+	RGNSZ_1KB   RegionSize = 0x09 // MPU Region Size 1 KByte
+	RGNSZ_2KB   RegionSize = 0x0A // MPU Region Size 2 KBytes
+	RGNSZ_4KB   RegionSize = 0x0B // MPU Region Size 4 KBytes
+	RGNSZ_8KB   RegionSize = 0x0C // MPU Region Size 8 KBytes
+	RGNSZ_16KB  RegionSize = 0x0D // MPU Region Size 16 KBytes
+	RGNSZ_32KB  RegionSize = 0x0E // MPU Region Size 32 KBytes
+	RGNSZ_64KB  RegionSize = 0x0F // MPU Region Size 64 KBytes
+	RGNSZ_128KB RegionSize = 0x10 // MPU Region Size 128 KBytes
+	RGNSZ_256KB RegionSize = 0x11 // MPU Region Size 256 KBytes
+	RGNSZ_512KB RegionSize = 0x12 // MPU Region Size 512 KBytes
+	RGNSZ_1MB   RegionSize = 0x13 // MPU Region Size 1 MByte
+	RGNSZ_2MB   RegionSize = 0x14 // MPU Region Size 2 MBytes
+	RGNSZ_4MB   RegionSize = 0x15 // MPU Region Size 4 MBytes
+	RGNSZ_8MB   RegionSize = 0x16 // MPU Region Size 8 MBytes
+	RGNSZ_16MB  RegionSize = 0x17 // MPU Region Size 16 MBytes
+	RGNSZ_32MB  RegionSize = 0x18 // MPU Region Size 32 MBytes
+	RGNSZ_64MB  RegionSize = 0x19 // MPU Region Size 64 MBytes
+	RGNSZ_128MB RegionSize = 0x1A // MPU Region Size 128 MBytes
+	RGNSZ_256MB RegionSize = 0x1B // MPU Region Size 256 MBytes
+	RGNSZ_512MB RegionSize = 0x1C // MPU Region Size 512 MBytes
+	RGNSZ_1GB   RegionSize = 0x1D // MPU Region Size 1 GByte
+	RGNSZ_2GB   RegionSize = 0x1E // MPU Region Size 2 GBytes
+	RGNSZ_4GB   RegionSize = 0x1F // MPU Region Size 4 GBytes
+)
+
+const (
+	PERM_NONE AccessPerms = 0 // MPU Access Permission no access
+	PERM_PRIV AccessPerms = 1 // MPU Access Permission privileged access only
+	PERM_URO  AccessPerms = 2 // MPU Access Permission unprivileged access read-only
+	PERM_FULL AccessPerms = 3 // MPU Access Permission full access
+	PERM_PRO  AccessPerms = 5 // MPU Access Permission privileged access read-only
+	PERM_RO   AccessPerms = 6 // MPU Access Permission read-only access
+)
+
+const (
+	EXTN_NORMAL Extension = 0
+	EXTN_DEVICE Extension = 2
+)
+
+func (mpu *MPU_Type) Enable(enable bool) {
+	if enable {
+		mpu.CTRL.Set(MPU_CTRL_PRIVDEFENA_Msk | MPU_CTRL_ENABLE_Msk)
+		SystemControl.SHCSR.SetBits(SCB_SHCSR_MEMFAULTENA_Msk)
+		arm.AsmFull(`
+      dsb 0xF
+      isb 0xF
+    `, nil)
+		enableDcache(true)
+		enableIcache(true)
+	} else {
+		enableIcache(false)
+		enableDcache(false)
+		arm.AsmFull(`
+      dmb 0xF
+    `, nil)
+		SystemControl.SHCSR.ClearBits(SCB_SHCSR_MEMFAULTENA_Msk)
+		mpu.CTRL.ClearBits(MPU_CTRL_ENABLE_Msk)
+	}
+}
+
+// MPU Region Base Address Register value
+func (mpu *MPU_Type) SetRBAR(region uint32, baseAddress uint32) {
+	mpu.RBAR.Set((baseAddress & MPU_RBAR_ADDR_Msk) |
+		(region & MPU_RBAR_REGION_Msk) | MPU_RBAR_VALID_Msk)
+}
+
+// MPU Region Attribute and Size Register value
+func (mpu *MPU_Type) SetRASR(size RegionSize, access AccessPerms, ext Extension, exec, share, cache, buffer, disable bool) {
+	boolBit := func(b bool) uint32 {
+		if b {
+			return 1
+		}
+		return 0
+	}
+	attr := ((uint32(ext) << MPU_RASR_TEX_Pos) & MPU_RASR_TEX_Msk) |
+		((boolBit(share) << MPU_RASR_S_Pos) & MPU_RASR_S_Msk) |
+		((boolBit(cache) << MPU_RASR_C_Pos) & MPU_RASR_C_Msk) |
+		((boolBit(buffer) << MPU_RASR_B_Pos) & MPU_RASR_B_Msk)
+	mpu.RASR.Set(((boolBit(!exec) << MPU_RASR_XN_Pos) & MPU_RASR_XN_Msk) |
+		((uint32(access) << MPU_RASR_AP_Pos) & MPU_RASR_AP_Msk) |
+		(attr & (MPU_RASR_TEX_Msk | MPU_RASR_S_Msk | MPU_RASR_C_Msk | MPU_RASR_B_Msk)) |
+		((boolBit(disable) << MPU_RASR_SRD_Pos) & MPU_RASR_SRD_Msk) |
+		((uint32(size) << MPU_RASR_SIZE_Pos) & MPU_RASR_SIZE_Msk) |
+		MPU_RASR_ENABLE_Msk)
+}
+
+func enableIcache(enable bool) {
+	if enable != SystemControl.CCR.HasBits(SCB_CCR_IC_Msk) {
+		if enable {
+			arm.AsmFull(`
+        dsb 0xF
+        isb 0xF
+      `, nil)
+			SystemControl.ICIALLU.Set(0)
+			arm.AsmFull(`
+        dsb 0xF
+        isb 0xF
+      `, nil)
+			SystemControl.CCR.SetBits(SCB_CCR_IC_Msk)
+			arm.AsmFull(`
+        dsb 0xF
+        isb 0xF
+      `, nil)
+		} else {
+			arm.AsmFull(`
+        dsb 0xF
+        isb 0xF
+      `, nil)
+			SystemControl.CCR.ClearBits(SCB_CCR_IC_Msk)
+			SystemControl.ICIALLU.Set(0)
+			arm.AsmFull(`
+        dsb 0xF
+        isb 0xF
+      `, nil)
+		}
+	}
+}
+
+func enableDcache(enable bool) {
+	if enable != SystemControl.CCR.HasBits(SCB_CCR_DC_Msk) {
+		if enable {
+			SystemControl.CSSELR.Set(0)
+			arm.AsmFull(`
+        dsb 0xF
+      `, nil)
+			ccsidr := SystemControl.CCSIDR.Get()
+			sets := (ccsidr & SCB_CCSIDR_NUMSETS_Msk) >> SCB_CCSIDR_NUMSETS_Pos
+			for sets != 0 {
+				ways := (ccsidr & SCB_CCSIDR_ASSOCIATIVITY_Msk) >> SCB_CCSIDR_ASSOCIATIVITY_Pos
+				for ways != 0 {
+					SystemControl.DCISW.Set(
+						((sets << SCB_DCISW_SET_Pos) & SCB_DCISW_SET_Msk) |
+							((ways << SCB_DCISW_WAY_Pos) & SCB_DCISW_WAY_Msk))
+					ways--
+				}
+				sets--
+			}
+			arm.AsmFull(`
+        dsb 0xF
+      `, nil)
+			SystemControl.CCR.SetBits(SCB_CCR_DC_Msk)
+			arm.AsmFull(`
+        dsb 0xF
+        isb 0xF
+      `, nil)
+		} else {
+			var (
+				ccsidr volatile.Register32
+				sets   volatile.Register32
+				ways   volatile.Register32
+			)
+			SystemControl.CSSELR.Set(0)
+			arm.AsmFull(`
+        dsb 0xF
+      `, nil)
+			SystemControl.CCR.ClearBits(SCB_CCR_DC_Msk)
+			arm.AsmFull(`
+        dsb 0xF
+      `, nil)
+			ccsidr.Set(SystemControl.CCSIDR.Get())
+			sets.Set((ccsidr.Get() & SCB_CCSIDR_NUMSETS_Msk) >> SCB_CCSIDR_NUMSETS_Pos)
+			for sets.Get() != 0 {
+				ways.Set((ccsidr.Get() & SCB_CCSIDR_ASSOCIATIVITY_Msk) >> SCB_CCSIDR_ASSOCIATIVITY_Pos)
+				for ways.Get() != 0 {
+					SystemControl.DCCISW.Set(
+						((sets.Get() << SCB_DCCISW_SET_Pos) & SCB_DCCISW_SET_Msk) |
+							((ways.Get() << SCB_DCCISW_WAY_Pos) & SCB_DCCISW_WAY_Msk))
+					ways.Set(ways.Get() - 1)
+				}
+				sets.Set(sets.Get() - 1)
+			}
+			arm.AsmFull(`
+        dsb 0xF
+        isb 0xF
+      `, nil)
+		}
+	}
+}
@@ -0,0 +1,71 @@
+// Hand created file. DO NOT DELETE.
+// atsamd51x bitfield definitions that are not auto-generated by gen-device-svd.go
+
+// +build sam,atsamd51
+
+// These are the supported pchctrl function numberings on the atsamd51x
+// See http://ww1.microchip.com/downloads/en/DeviceDoc/SAM_D5xE5x_Family_Data_Sheet_DS60001507F.pdf
+// table 14-9
+
+package sam
+
+const (
+	PCHCTRL_GCLK_OSCCTRL_DFLL48     = 0 // DFLL48 input clock source
+	PCHCTRL_GCLK_OSCCTRL_FDPLL0     = 1 // Reference clock for FDPLL0
+	PCHCTRL_GCLK_OSCCTRL_FDPLL1     = 2 // Reference clock for FDPLL1
+	PCHCTRL_GCLK_OSCCTRL_FDPLL0_32K = 3 // FDPLL0 = 3 // 32KHz clock for internal lock timer
+	PCHCTRL_GCLK_OSCCTRL_FDPLL1_32K = 3 // FDPLL1 = 3 // 32KHz clock for internal lock timer
+	PCHCTRL_GCLK_SDHC0_SLOW         = 3 // SDHC0 = 3 // Slow
+	PCHCTRL_GCLK_SDHC1_SLOW         = 3 // SDHC1 = 3 // Slow
+	PCHCTRL_GCLK_SERCOMX_SLOW       = 3 // GCLK_SERCOM[0..7]_SLOW = 3
+	PCHCTRL_GCLK_EIC                = 4
+	PCHCTRL_GCLK_FREQM_MSR          = 5 // FREQM Measure
+	PCHCTRL_GCLK_FREQM_REF          = 6 // FREQM Reference
+	PCHCTRL_GCLK_SERCOM0_CORE       = 7 // SERCOM0 Core
+	PCHCTRL_GCLK_SERCOM1_CORE       = 8 // SERCOM1 Core
+	PCHCTRL_GCLK_TC0                = 9
+	PCHCTRL_GCLK_TC1                = 9  // TC0, TC1
+	PCHCTRL_GCLK_USB                = 10 // USB
+	PCHCTRL_GCLK_EVSYS0             = 11
+	PCHCTRL_GCLK_EVSYS1             = 12
+	PCHCTRL_GCLK_EVSYS2             = 13
+	PCHCTRL_GCLK_EVSYS3             = 14
+	PCHCTRL_GCLK_EVSYS4             = 15
+	PCHCTRL_GCLK_EVSYS5             = 16
+	PCHCTRL_GCLK_EVSYS6             = 17
+	PCHCTRL_GCLK_EVSYS7             = 18
+	PCHCTRL_GCLK_EVSYS8             = 19
+	PCHCTRL_GCLK_EVSYS9             = 20
+	PCHCTRL_GCLK_EVSYS10            = 21
+	PCHCTRL_GCLK_EVSYS11            = 22
+	PCHCTRL_GCLK_SERCOM2_CORE       = 23 // SERCOM2 Core
+	PCHCTRL_GCLK_SERCOM3_CORE       = 24 // SERCOM3 Core
+	PCHCTRL_GCLK_TCC0               = 25
+	PCHCTRL_GCLK_TCC1               = 25 // TCC0, TCC1
+	PCHCTRL_GCLK_TC2                = 26
+	PCHCTRL_GCLK_TC3                = 26 // TC2, TC3
+	PCHCTRL_GCLK_CAN0               = 27 // CAN0
+	PCHCTRL_GCLK_CAN1               = 28 // CAN1
+	PCHCTRL_GCLK_TCC2               = 29
+	PCHCTRL_GCLK_TCC3               = 29 // TCC2, TCC3
+	PCHCTRL_GCLK_TC4                = 30
+	PCHCTRL_GCLK_TC5                = 30 // TC4, TC5
+	PCHCTRL_GCLK_PDEC               = 31 // PDEC
+	PCHCTRL_GCLK_AC                 = 32 // AC
+	PCHCTRL_GCLK_CCL                = 33 // CCL
+	PCHCTRL_GCLK_SERCOM4_CORE       = 34 // SERCOM4 Core
+	PCHCTRL_GCLK_SERCOM5_CORE       = 35 // SERCOM5 Core
+	PCHCTRL_GCLK_SERCOM6_CORE       = 36 // SERCOM6 Core
+	PCHCTRL_GCLK_SERCOM7_CORE       = 37 // SERCOM7 Core
+	PCHCTRL_GCLK_TCC4               = 38 // TCC4
+	PCHCTRL_GCLK_TC6                = 39
+	PCHCTRL_GCLK_TC7                = 39 // TC6, TC7
+	PCHCTRL_GCLK_ADC0               = 40 // ADC0
+	PCHCTRL_GCLK_ADC1               = 41 // ADC1
+	PCHCTRL_GCLK_DAC                = 42 // DAC
+	PCHCTRL_GCLK_I2S0               = 43
+	PCHCTRL_GCLK_I2S1               = 44
+	PCHCTRL_GCLK_SDHC0              = 45 // SDHC0
+	PCHCTRL_GCLK_SDHC1              = 46 // SDHC1
+	PCHCTRL_GCLK_CM4_TRACE          = 47 // CM4 Trace
+)
@@ -1,7 +1,9 @@
-// These are the supported alternate function numberings on the stm32f407
-// +build stm32,stm32f407
+// Hand created file. DO NOT DELETE.
+// STM32FXXX (except stm32f1xx) bitfield definitions that are not
+//  auto-generated by gen-device-svd.go
+// +build stm32f4

-// Alternate function settings on the stm32f4xx series
+// Alternate function settings on the stm32f4 series

 package stm32

@@ -8,7 +8,7 @@ import (
 // This example assumes that an RGB LED is connected to pins 3, 5 and 6 on an Arduino.
 // Change the values below to use different pins.
 const (
-	redPin   = machine.D3
+	redPin   = machine.D4
 	greenPin = machine.D5
 	bluePin  = machine.D6
 )
@@ -124,3 +124,130 @@ func IndexString(str, sub string) int {
 	}
 	return -1
 }
+
+// Copyright 2020 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// The following code has been copied from the Go 1.15 release tree.
+
+// PrimeRK is the prime base used in Rabin-Karp algorithm.
+const PrimeRK = 16777619
+
+// HashStrBytes returns the hash and the appropriate multiplicative
+// factor for use in Rabin-Karp algorithm.
+func HashStrBytes(sep []byte) (uint32, uint32) {
+	hash := uint32(0)
+	for i := 0; i < len(sep); i++ {
+		hash = hash*PrimeRK + uint32(sep[i])
+	}
+	var pow, sq uint32 = 1, PrimeRK
+	for i := len(sep); i > 0; i >>= 1 {
+		if i&1 != 0 {
+			pow *= sq
+		}
+		sq *= sq
+	}
+	return hash, pow
+}
+
+// HashStr returns the hash and the appropriate multiplicative
+// factor for use in Rabin-Karp algorithm.
+func HashStr(sep string) (uint32, uint32) {
+	hash := uint32(0)
+	for i := 0; i < len(sep); i++ {
+		hash = hash*PrimeRK + uint32(sep[i])
+	}
+	var pow, sq uint32 = 1, PrimeRK
+	for i := len(sep); i > 0; i >>= 1 {
+		if i&1 != 0 {
+			pow *= sq
+		}
+		sq *= sq
+	}
+	return hash, pow
+}
+
+// HashStrRevBytes returns the hash of the reverse of sep and the
+// appropriate multiplicative factor for use in Rabin-Karp algorithm.
+func HashStrRevBytes(sep []byte) (uint32, uint32) {
+	hash := uint32(0)
+	for i := len(sep) - 1; i >= 0; i-- {
+		hash = hash*PrimeRK + uint32(sep[i])
+	}
+	var pow, sq uint32 = 1, PrimeRK
+	for i := len(sep); i > 0; i >>= 1 {
+		if i&1 != 0 {
+			pow *= sq
+		}
+		sq *= sq
+	}
+	return hash, pow
+}
+
+// HashStrRev returns the hash of the reverse of sep and the
+// appropriate multiplicative factor for use in Rabin-Karp algorithm.
+func HashStrRev(sep string) (uint32, uint32) {
+	hash := uint32(0)
+	for i := len(sep) - 1; i >= 0; i-- {
+		hash = hash*PrimeRK + uint32(sep[i])
+	}
+	var pow, sq uint32 = 1, PrimeRK
+	for i := len(sep); i > 0; i >>= 1 {
+		if i&1 != 0 {
+			pow *= sq
+		}
+		sq *= sq
+	}
+	return hash, pow
+}
+
+// IndexRabinKarpBytes uses the Rabin-Karp search algorithm to return the index of the
+// first occurence of substr in s, or -1 if not present.
+func IndexRabinKarpBytes(s, sep []byte) int {
+	// Rabin-Karp search
+	hashsep, pow := HashStrBytes(sep)
+	n := len(sep)
+	var h uint32
+	for i := 0; i < n; i++ {
+		h = h*PrimeRK + uint32(s[i])
+	}
+	if h == hashsep && Equal(s[:n], sep) {
+		return 0
+	}
+	for i := n; i < len(s); {
+		h *= PrimeRK
+		h += uint32(s[i])
+		h -= pow * uint32(s[i-n])
+		i++
+		if h == hashsep && Equal(s[i-n:i], sep) {
+			return i - n
+		}
+	}
+	return -1
+}
+
+// IndexRabinKarp uses the Rabin-Karp search algorithm to return the index of the
+// first occurence of substr in s, or -1 if not present.
+func IndexRabinKarp(s, substr string) int {
+	// Rabin-Karp search
+	hashss, pow := HashStr(substr)
+	n := len(substr)
+	var h uint32
+	for i := 0; i < n; i++ {
+		h = h*PrimeRK + uint32(s[i])
+	}
+	if h == hashss && s[:n] == substr {
+		return 0
+	}
+	for i := n; i < len(s); {
+		h *= PrimeRK
+		h += uint32(s[i])
+		h -= pow * uint32(s[i-n])
+		i++
+		if h == hashss && s[i-n:i] == substr {
+			return i - n
+		}
+	}
+	return -1
+}
@@ -18,3 +18,8 @@ type Task struct {
 	// state is the underlying running state of the task.
 	state state
 }
+
+// getGoroutineStackSize is a compiler intrinsic that returns the stack size for
+// the given function and falls back to the default stack size. It is replaced
+// with a load from a special section just before codegen.
+func getGoroutineStackSize(fn uintptr) uintptr
@@ -67,7 +67,7 @@ func createTask() *Task {
 // start invokes a function in a new goroutine. Calls to this are inserted by the compiler.
 // The created goroutine starts running immediately.
 // This is implemented inside the compiler.
-func start(fn uintptr, args unsafe.Pointer)
+func start(fn uintptr, args unsafe.Pointer, stackSize uintptr)

 // Current returns the current active task.
 // This is implemented inside the compiler.
@@ -17,7 +17,7 @@ func Current() *Task {
 }

 //go:noinline
-func start(fn uintptr, args unsafe.Pointer) {
+func start(fn uintptr, args unsafe.Pointer, stackSize uintptr) {
 	// The compiler will error if this is reachable.
 	runtimePanic("scheduler is disabled")
 }
@@ -45,6 +45,11 @@ func Pause() {
 	currentTask.state.pause()
 }

+//export tinygo_pause
+func pause() {
+	Pause()
+}
+
 // Resume the task until it pauses or completes.
 // This may only be called from the scheduler.
 func (t *Task) Resume() {
@@ -54,22 +59,44 @@ func (t *Task) Resume() {
 }

 // initialize the state and prepare to call the specified function with the specified argument bundle.
-func (s *state) initialize(fn uintptr, args unsafe.Pointer) {
+func (s *state) initialize(fn uintptr, args unsafe.Pointer, stackSize uintptr) {
 	// Create a stack.
 	stack := make([]uintptr, stackSize/unsafe.Sizeof(uintptr(0)))

+	// Set up the stack canary, a random number that should be checked when
+	// switching from the task back to the scheduler. The stack canary pointer
+	// points to the first word of the stack. If it has changed between now and
+	// the next stack switch, there was a stack overflow.
+	s.canaryPtr = &stack[0]
+	*s.canaryPtr = stackCanary
+
+	// Get a pointer to the top of the stack, where the initial register values
+	// are stored. They will be popped off the stack on the first stack switch
+	// to the goroutine, and will start running tinygo_startTask (this setup
+	// happens in archInit).
+	r := (*calleeSavedRegs)(unsafe.Pointer(&stack[uintptr(len(stack))-(unsafe.Sizeof(calleeSavedRegs{})/unsafe.Sizeof(uintptr(0)))]))
+
 	// Invoke architecture-specific initialization.
-	s.archInit(stack, fn, args)
+	s.archInit(r, fn, args)
 }

+//export tinygo_swapTask
+func swapTask(oldStack uintptr, newStack *uintptr)
+
+// startTask is a small wrapper function that sets up the first (and only)
+// argument to the new goroutine and makes sure it is exited when the goroutine
+// finishes.
+//go:extern tinygo_startTask
+var startTask [0]uint8
+
 //go:linkname runqueuePushBack runtime.runqueuePushBack
 func runqueuePushBack(*Task)

 // start creates and starts a new goroutine with the given function and arguments.
 // The new goroutine is scheduled to run later.
-func start(fn uintptr, args unsafe.Pointer) {
+func start(fn uintptr, args unsafe.Pointer, stackSize uintptr) {
 	t := &Task{}
-	t.state.initialize(fn, args)
+	t.state.initialize(fn, args, stackSize)
 	runqueuePushBack(t)
 }

@@ -0,0 +1,99 @@
+.section .bss.tinygo_systemStack
+.global tinygo_systemStack
+.type   tinygo_systemStack, %object
+tinygo_systemStack:
+    .short 0
+
+.section .text.tinygo_startTask
+.global  tinygo_startTask
+.type    tinygo_startTask, %function
+tinygo_startTask:
+    // Small assembly stub for starting a goroutine. This is already run on the
+    // new stack, with the callee-saved registers already loaded.
+    // Most importantly, r2r3 contain the pc of the to-be-started function and
+    // r4r5 contain the only argument it is given. Multiple arguments are packed
+    // into one by storing them in a new allocation.
+
+    // Set the first argument of the goroutine start wrapper, which contains all
+    // the arguments.
+    movw  r24, r4
+
+    // Branch to the "goroutine start" function. Note that the Z register is
+    // call-clobbered, so does not need to be restored after use.
+    movw  Z, r2
+    icall
+
+    // After return, exit this goroutine. This is a tail call.
+#if __AVR_ARCH__ == 2 || __AVR_ARCH__ == 25
+    // Small memory devices (≤8kB flash) that do not have the long call
+    // instruction availble will need to use rcall instead.
+    // Note that they will probably not be able to run more than the main
+    // goroutine anyway, but this file is compiled for all AVRs so it needs to
+    // compile at least.
+    rcall  tinygo_pause
+#else
+    // Other devices can (and must) use the regular call instruction.
+    call tinygo_pause
+#endif
+
+.global tinygo_swapTask
+.type tinygo_swapTask, %function
+tinygo_swapTask:
+    // This function gets the following parameters:
+    // r24:r25 = newStack uintptr
+    // r22:r23 = oldStack *uintptr
+
+    // Save all call-saved registers:
+    // https://gcc.gnu.org/wiki/avr-gcc#Call-Saved_Registers
+    push r29 // Y
+    push r28 // Y
+    push r17
+    push r16
+    push r15
+    push r14
+    push r13
+    push r12
+    push r11
+    push r10
+    push r9
+    push r8
+    push r7
+    push r6
+    push r5
+    push r4
+    push r3
+    push r2
+
+    // Save the current stack pointer in oldStack.
+    in  r2, 0x3d; SPL
+    in  r3, 0x3e; SPH
+    movw Y, r22
+    std Y+0, r2
+    std Y+1, r3
+
+    // Switch to the new stack pointer.
+    out  0x3d, r24; SPL
+    out  0x3e, r25; SPH
+
+    // Load saved register from the new stack.
+    pop r2
+    pop r3
+    pop r4
+    pop r5
+    pop r6
+    pop r7
+    pop r8
+    pop r9
+    pop r10
+    pop r11
+    pop r12
+    pop r13
+    pop r14
+    pop r15
+    pop r16
+    pop r17
+    pop r28 // Y
+    pop r29 // Y
+
+    // Return into the new task, as if tinygo_swapTask was a regular call.
+    ret
@@ -4,11 +4,12 @@ package task

 import "unsafe"

-const stackSize = 256
+//go:extern tinygo_systemStack
+var systemStack uintptr

 // calleeSavedRegs is the list of registers that must be saved and restored when
-// switching between tasks. Also see scheduler_avr.S that relies on the
-// exact layout of this struct.
+// switching between tasks. Also see task_stack_avr.S that relies on the exact
+// layout of this struct.
 //
 // https://gcc.gnu.org/wiki/avr-gcc#Call-Saved_Registers
 type calleeSavedRegs struct {
@@ -25,34 +26,15 @@ type calleeSavedRegs struct {
 	pc uintptr
 }

-// registers gets a pointer to the registers stored at the top of the stack.
-func (s *state) registers() *calleeSavedRegs {
-	return (*calleeSavedRegs)(unsafe.Pointer(s.sp + 1))
-}
-
-// startTask is a small wrapper function that sets up the first (and only)
-// argument to the new goroutine and makes sure it is exited when the goroutine
-// finishes.
-//go:extern tinygo_startTask
-var startTask [0]uint8
-
 // archInit runs architecture-specific setup for the goroutine startup.
 // Note: adding //go:noinline to work around an AVR backend bug.
 //go:noinline
-func (s *state) archInit(stack []uintptr, fn uintptr, args unsafe.Pointer) {
-	// Set up the stack canary, a random number that should be checked when
-	// switching from the task back to the scheduler. The stack canary pointer
-	// points to the first word of the stack. If it has changed between now and
-	// the next stack switch, there was a stack overflow.
-	s.canaryPtr = &stack[0]
-	*s.canaryPtr = stackCanary
-
+func (s *state) archInit(r *calleeSavedRegs, fn uintptr, args unsafe.Pointer) {
 	// Store the initial sp for the startTask function (implemented in assembly).
-	s.sp = uintptr(unsafe.Pointer(&stack[uintptr(len(stack))-(unsafe.Sizeof(calleeSavedRegs{})/unsafe.Sizeof(uintptr(0)))])) - 1
+	s.sp = uintptr(unsafe.Pointer(r)) - 1

 	// Initialize the registers.
 	// These will be popped off of the stack on the first resume of the goroutine.
-	r := s.registers()

 	// Start the function at tinygo_startTask.
 	startTask := uintptr(unsafe.Pointer(&startTask))
@@ -69,20 +51,17 @@ func (s *state) archInit(stack []uintptr, fn uintptr, args unsafe.Pointer) {
 }

 func (s *state) resume() {
-	switchToTask(s.sp)
+	swapTask(s.sp, &systemStack)
 }

-//export tinygo_switchToTask
-func switchToTask(uintptr)
-
-//export tinygo_switchToScheduler
-func switchToScheduler(*uintptr)
-
 func (s *state) pause() {
-	switchToScheduler(&s.sp)
+	newStack := systemStack
+	systemStack = 0
+	swapTask(newStack, &s.sp)
 }

-//export tinygo_pause
-func pause() {
-	Pause()
+// SystemStack returns the system stack pointer when called from a task stack.
+// When called from the system stack, it returns 0.
+func SystemStack() uintptr {
+	return systemStack
 }
@@ -30,22 +30,7 @@ tinygo_startTask:
    .cfi_endproc
 .size tinygo_startTask, .-tinygo_startTask

-.section .text.tinygo_getSystemStackPointer
-.global  tinygo_getSystemStackPointer
-.type    tinygo_getSystemStackPointer, %function
-tinygo_getSystemStackPointer:
-    .cfi_startproc
-    // The system stack pointer is always stored in the MSP register.
-    mrs r0, MSP
-    bx lr
-    .cfi_endproc
-.size tinygo_getSystemStackPointer, .-tinygo_getSystemStackPointer
-
-
-// switchToScheduler and switchToTask are also in the same section, to make sure
-// relative branches work.
-.section .text.tinygo_swapTask
-
+.section .text.tinygo_switchToScheduler
 .global  tinygo_switchToScheduler
 .type    tinygo_switchToScheduler, %function
 tinygo_switchToScheduler:
@@ -62,6 +47,7 @@ tinygo_switchToScheduler:
    .cfi_endproc
 .size tinygo_switchToScheduler, .-tinygo_switchToScheduler

+.section .text.tinygo_switchToTask
 .global  tinygo_switchToTask
 .type    tinygo_switchToTask, %function
 tinygo_switchToTask:
@@ -72,11 +58,11 @@ tinygo_switchToTask:
    // and then we can invoke swapTask.
    msr PSP, r0

-    // Continue executing in the swapTask function, which swaps the stack
-    // pointer.
+    b.n tinygo_swapTask
    .cfi_endproc
 .size tinygo_switchToTask, .-tinygo_switchToTask

+.section .text.tinygo_swapTask
 .global  tinygo_swapTask
 .type    tinygo_swapTask, %function
 tinygo_swapTask:
@@ -139,34 +125,3 @@ tinygo_swapTask:
    #endif
    .cfi_endproc
 .size tinygo_swapTask, .-tinygo_swapTask
-
-.section .text.tinygo_scanCurrentStack
-.global  tinygo_scanCurrentStack
-.type    tinygo_scanCurrentStack, %function
-tinygo_scanCurrentStack:
-    .cfi_startproc
-    // Save callee-saved registers onto the stack.
-    #if defined(__thumb2__)
-    push {r4-r11, lr}
-    .cfi_def_cfa_offset 9*4
-    #else
-    mov r0, r8
-    mov r1, r9
-    mov r2, r10
-    mov r3, r11
-    push {r0-r3, lr}
-    .cfi_def_cfa_offset 5*4
-    push {r4-r7}
-    .cfi_def_cfa_offset 4*4
-    #endif
-
-    // Scan the stack.
-    mov r0, sp
-    bl tinygo_scanstack
-
-    // Restore stack state and return.
-    add sp, #32
-    .cfi_def_cfa_offset 1*4
-    pop {pc}
-    .cfi_endproc
-.size tinygo_scanCurrentStack, .-tinygo_scanCurrentStack
@@ -2,12 +2,13 @@

 package task

-import "unsafe"
-
-const stackSize = 1024
+import (
+	"device/arm"
+	"unsafe"
+)

 // calleeSavedRegs is the list of registers that must be saved and restored when
-// switching between tasks. Also see scheduler_cortexm.S that relies on the
+// switching between tasks. Also see task_stack_cortexm.S that relies on the
 // exact layout of this struct.
 type calleeSavedRegs struct {
 	r4  uintptr
@@ -22,34 +23,15 @@ type calleeSavedRegs struct {
 	pc uintptr
 }

-// registers gets a pointer to the registers stored at the top of the stack.
-func (s *state) registers() *calleeSavedRegs {
-	return (*calleeSavedRegs)(unsafe.Pointer(s.sp))
-}
-
-// startTask is a small wrapper function that sets up the first (and only)
-// argument to the new goroutine and makes sure it is exited when the goroutine
-// finishes.
-//go:extern tinygo_startTask
-var startTask [0]uint8
-
 // archInit runs architecture-specific setup for the goroutine startup.
-func (s *state) archInit(stack []uintptr, fn uintptr, args unsafe.Pointer) {
-	// Set up the stack canary, a random number that should be checked when
-	// switching from the task back to the scheduler. The stack canary pointer
-	// points to the first word of the stack. If it has changed between now and
-	// the next stack switch, there was a stack overflow.
-	s.canaryPtr = &stack[0]
-	*s.canaryPtr = stackCanary
-
+func (s *state) archInit(r *calleeSavedRegs, fn uintptr, args unsafe.Pointer) {
 	// Store the initial sp for the startTask function (implemented in assembly).
-	s.sp = uintptr(unsafe.Pointer(&stack[uintptr(len(stack))-(unsafe.Sizeof(calleeSavedRegs{})/unsafe.Sizeof(uintptr(0)))]))
+	s.sp = uintptr(unsafe.Pointer(r))

 	// Initialize the registers.
 	// These will be popped off of the stack on the first resume of the goroutine.
-	r := s.registers()

-	// Start the function at tinygo_startTask (defined in src/runtime/scheduler_cortexm.S).
+	// Start the function at tinygo_startTask (defined in src/internal/task/task_stack_cortexm.S).
 	// This assembly code calls a function (passed in r4) with a single argument (passed in r5).
 	// After the function returns, it calls Pause().
 	r.pc = uintptr(unsafe.Pointer(&startTask))
@@ -77,7 +59,8 @@ func (s *state) pause() {
 	switchToScheduler(&s.sp)
 }

-//export tinygo_pause
-func pause() {
-	Pause()
+// SystemStack returns the system stack pointer. On Cortex-M, it is always
+// available.
+func SystemStack() uintptr {
+	return arm.AsmFull("mrs {}, MSP", nil)
 }
@@ -0,0 +1,86 @@
+.section .text.tinygo_startTask,"ax",@progbits
+.global  tinygo_startTask
+.type    tinygo_startTask, %function
+tinygo_startTask:
+    // Small assembly stub for starting a goroutine. This already runs on the
+    // new stack, control reaches this function after returning from the initial
+    // tinygo_swapTask below (the retw.n instruction).
+    //
+    // The stack was set up in such a way that it looks as if this function was
+    // paused using tinygo_swapTask by setting up the parent register window and
+    // return pointer as a call4 instruction - except such a call never took
+    // place. Instead, the stack pointer is switched to the new stack after all
+    // live-but-invisible registers have been flushed to the stack. This means
+    // that all registers as present in tinygo_swapTask are moved four up (a2 in
+    // tinygo_swapTask is a6 in this function). We don't use any of those
+    // registers however. Instead, the retw.n instruction will load them through
+    // an underflow exception from the stack which means we get a0-a3 as defined
+    // in task_stack_esp32.go.
+
+    // Branch to the "goroutine start" function. The first (and only) parameter
+    // is stored in a2, but has to be moved to a6 to make it appear as a2 in the
+    // goroutine start function (due to changing the register window by four
+    // with callx4).
+    mov.n a6, a2
+    callx4 a3
+
+    // After return, exit this goroutine. This call never returns.
+    call4  tinygo_pause
+
+.section .text.tinygo_swapTask,"ax",@progbits
+.global tinygo_swapTask
+.type tinygo_swapTask, %function
+tinygo_swapTask:
+    // This function gets the following parameters:
+    // a2 = newStack uintptr
+    // a3 = oldStack *uintptr
+
+    // Reserve 32 bytes on the stack. It really needs to be 32 bytes, with 16
+    // extra at the bottom to adhere to the ABI.
+    entry sp, 32
+
+    // Disable interrupts while flushing registers. This is necessary because
+    // interrupts might want to use the stack pointer (at a2) which will be some
+    // arbitrary register while registers are flushed.
+    rsil a4, 3 // XCHAL_EXCM_LEVEL
+
+    // Flush all unsaved registers to the stack.
+    // This trick has been borrowed from the Zephyr project:
+    // https://github.com/zephyrproject-rtos/zephyr/blob/d79b003758/arch/xtensa/include/xtensa-asm2-s.h#L17
+    and a12, a12, a12
+    rotw 3
+    and a12, a12, a12
+    rotw 3
+    and a12, a12, a12
+    rotw 3
+    and a12, a12, a12
+    rotw 3
+    and a12, a12, a12
+    rotw 4
+
+    // Restore interrupts.
+    wsr.ps a4
+
+    // At this point, the following is true:
+    //     WindowStart == 1 << WindowBase
+    // Therefore, we don't need to do this manually.
+    // It also means that the stack pointer can now be safely modified.
+
+    // Save a0, which stores the return address and the parent register window
+    // in the upper two bits.
+    s32i.n a0, sp, 0
+
+    // Save the current stack pointer in oldStack.
+    s32i.n  sp, a3, 0
+
+    // Switch to the new stack pointer (newStack).
+    mov.n   sp, a2
+
+    // Load a0, which is the previous return addres from before the previous
+    // switch or the constructed return address to tinygo_startTask. This
+    // register also stores the parent register window.
+    l32i.n a0, sp, 0
+
+    // Return into the new stack. This instruction will trigger a window
+    // underflow, reloading the saved registers from the stack.
+    retw.n
@@ -0,0 +1,76 @@
+// +build scheduler.tasks,esp32
+
+package task
+
+// The windowed ABI (used on the ESP32) is as follows:
+//   a0:    return address (link register)
+//   a1:    stack pointer (must be 16-byte aligned)
+//   a2-a7: incoming arguments
+//   a7:    stack frame pointer (optional, normally unused in TinyGo)
+// Sources:
+//   http://cholla.mmto.org/esp8266/xtensa.html
+//   https://0x04.net/~mwk/doc/xtensa.pdf
+
+import (
+	"unsafe"
+)
+
+var systemStack uintptr
+
+// calleeSavedRegs is the list of registers that must be saved and restored when
+// switching between tasks. Also see task_stack_esp8266.S that relies on the
+// exact layout of this struct.
+type calleeSavedRegs struct {
+	// Registers in the register window of tinygo_startTask.
+	a0 uintptr
+	a1 uintptr
+	a2 uintptr
+	a3 uintptr
+
+	// Locals that can be used by tinygo_swapTask.
+	// The first field is the a0 loaded in tinygo_swapTask, the rest is unused.
+	locals [4]uintptr
+}
+
+// archInit runs architecture-specific setup for the goroutine startup.
+func (s *state) archInit(r *calleeSavedRegs, fn uintptr, args unsafe.Pointer) {
+	// Store the stack pointer for the tinygo_swapTask function (implemented in
+	// assembly). It needs to point to the locals field instead of a0 so that
+	// the retw.n at the end of tinygo_swapTask will return into
+	// tinygo_startTask with a0-a3 loaded (using the register window mechanism).
+	s.sp = uintptr(unsafe.Pointer(&r.locals[0]))
+
+	// Start the goroutine at tinygo_startTask (defined in
+	// src/internal/task/task_stack_esp32.S). The topmost two bits are not part
+	// of the address but instead store the register window of the caller.
+	// In this case there is no caller, instead we set up the return address as
+	// if tinygo_startTask called tinygo_swapTask with a call4 instruction.
+	r.locals[0] = uintptr(unsafe.Pointer(&startTask))&^(3<<30) | (1 << 30)
+
+	// Set up the stack pointer inside tinygo_startTask.
+	// Unlike most calling conventions, the windowed ABI actually saves the
+	// stack pointer on the stack to make register windowing work.
+	r.a1 = uintptr(unsafe.Pointer(r)) + 32
+
+	// Store the function pointer and the (only) parameter on the stack in a
+	// location that will be reloaded into registers when doing the
+	// pseudo-return to tinygo_startTask using the register window mechanism.
+	r.a3 = fn
+	r.a2 = uintptr(args)
+}
+
+func (s *state) resume() {
+	swapTask(s.sp, &systemStack)
+}
+
+func (s *state) pause() {
+	newStack := systemStack
+	systemStack = 0
+	swapTask(newStack, &s.sp)
+}
+
+// SystemStack returns the system stack pointer when called from a task stack.
+// When called from the system stack, it returns 0.
+func SystemStack() uintptr {
+	return systemStack
+}
@@ -0,0 +1,54 @@
+.section .text.tinygo_startTask,"ax",@progbits
+.global  tinygo_startTask
+.type    tinygo_startTask, %function
+tinygo_startTask:
+    // Small assembly stub for starting a goroutine. This is already run on the
+    // new stack, with the callee-saved registers already loaded.
+    // Most importantly, r4 contains the pc of the to-be-started function and r5
+    // contains the only argument it is given. Multiple arguments are packed
+    // into one by storing them in a new allocation.
+
+    // Set the first argument of the goroutine start wrapper, which contains all
+    // the arguments.
+    mov.n  a2, a13
+
+    // Branch to the "goroutine start" function.
+    callx0 a12
+
+    // After return, exit this goroutine. This is a tail call.
+    call0  tinygo_pause
+.size tinygo_startTask, .-tinygo_startTask
+
+.global tinygo_swapTask
+.type tinygo_swapTask, %function
+tinygo_swapTask:
+    // This function gets the following parameters:
+    //   a2 = newStack uintptr
+    //   a3 = oldStack *uintptr
+    // Note:
+    //   a0 is the return address
+    //   a1 is the stack pointer (sp)
+
+    // Save all callee-saved registers:
+    addi    sp,  sp, -20
+    s32i.n  a12, sp, 0
+    s32i.n  a13, sp, 4
+    s32i.n  a14, sp, 8
+    s32i.n  a15, sp, 12
+    s32i.n  a0,  sp, 16
+
+    // Save the current stack pointer in oldStack.
+    s32i.n  sp, a3, 0
+
+    // Switch to the new stack pointer.
+    mov.n   sp, a2
+
+    // Load state from new task and branch to the previous position in the
+    // program.
+    l32i.n  a12, sp, 0
+    l32i.n  a13, sp, 4
+    l32i.n  a14, sp, 8
+    l32i.n  a15, sp, 12
+    l32i.n  a0,  sp, 16
+    addi    sp,  sp, 20
+    ret.n
@@ -0,0 +1,70 @@
+// +build scheduler.tasks,esp8266
+
+package task
+
+// Stack switch implementation for the ESP8266, which does not use the windowed
+// ABI of Xtensa. Registers are assigned as follows:
+//   a0:      return address (link register)
+//   a1:      stack pointer (must be 16-byte aligned)
+//   a2-a7:   incoming arguments
+//   a8:      static chain (unused)
+//   a12-a15: callee-saved
+//   a15:     stack frame pointer (optional, unused)
+// Sources:
+// http://cholla.mmto.org/esp8266/xtensa.html
+// https://0x04.net/~mwk/doc/xtensa.pdf
+
+import "unsafe"
+
+var systemStack uintptr
+
+// calleeSavedRegs is the list of registers that must be saved and restored when
+// switching between tasks. Also see task_stack_esp8266.S that relies on the
+// exact layout of this struct.
+type calleeSavedRegs struct {
+	a12 uintptr
+	a13 uintptr
+	a14 uintptr
+	a15 uintptr
+
+	pc uintptr // also link register or r0
+}
+
+// archInit runs architecture-specific setup for the goroutine startup.
+func (s *state) archInit(r *calleeSavedRegs, fn uintptr, args unsafe.Pointer) {
+	// Store the initial sp for the startTask function (implemented in assembly).
+	s.sp = uintptr(unsafe.Pointer(r))
+
+	// Initialize the registers.
+	// These will be popped off of the stack on the first resume of the goroutine.
+
+	// Start the function at tinygo_startTask (defined in
+	// src/internal/task/task_stack_esp8266.S).
+	// This assembly code calls a function (passed in a12) with a single argument
+	// (passed in a13). After the function returns, it calls Pause().
+	r.pc = uintptr(unsafe.Pointer(&startTask))
+
+	// Pass the function to call in a12.
+	// This function is a compiler-generated wrapper which loads arguments out of a struct pointer.
+	// See createGoroutineStartWrapper (defined in compiler/goroutine.go) for more information.
+	r.a12 = fn
+
+	// Pass the pointer to the arguments struct in a13.
+	r.a13 = uintptr(args)
+}
+
+func (s *state) resume() {
+	swapTask(s.sp, &systemStack)
+}
+
+func (s *state) pause() {
+	newStack := systemStack
+	systemStack = 0
+	swapTask(newStack, &s.sp)
+}
+
+// SystemStack returns the system stack pointer when called from a task stack.
+// When called from the system stack, it returns 0.
+func SystemStack() uintptr {
+	return systemStack
+}
@@ -0,0 +1,91 @@
+// +build arduino_mkr1000
+
+// This contains the pin mappings for the Arduino MKR1000 board.
+//
+// For more information, see: https://store.arduino.cc/usa/arduino-mkr1000-with-headers-mounted
+//
+package machine
+
+// used to reset into bootloader
+const RESET_MAGIC_VALUE = 0x07738135
+
+// GPIO Pins
+const (
+	RX0 Pin = PB23 // UART2 RX
+	TX1 Pin = PB22 // UART2 TX
+
+	D0 Pin = PA22 // PWM available
+	D1 Pin = PA23 // PWM available
+	D2 Pin = PA10 // PWM available
+	D3 Pin = PA11 // PWM available
+	D4 Pin = PB10 // PWM available
+	D5 Pin = PB11 // PWM available
+
+	D6  Pin = PA20 // PWM available
+	D7  Pin = PA21 // PWM available
+	D8  Pin = PA16 // PWM available
+	D9  Pin = PA17
+	D10 Pin = PA19 // PWM available
+	D11 Pin = PA08 // SDA
+	D12 Pin = PA09 // PWM available, SCL
+	D13 Pin = PB23 // RX
+	D14 Pin = PB22 // TX
+)
+
+// Analog pins
+const (
+	A0 Pin = PA02 // ADC0/AIN[0]
+	A1 Pin = PB02 // AIN[10]
+	A2 Pin = PB03 // AIN[11]
+	A3 Pin = PA04 // AIN[04]
+	A4 Pin = PA05 // AIN[05]
+	A5 Pin = PA06 // AIN[06]
+	A6 Pin = PA07 // AIN[07]
+)
+
+const (
+	LED = D6
+)
+
+// UART0 aka USBCDC pins
+const (
+	USBCDC_DM_PIN Pin = PA24
+	USBCDC_DP_PIN Pin = PA25
+)
+
+// UART1 pins
+const (
+	UART_TX_PIN Pin = PB22
+	UART_RX_PIN Pin = PB23
+)
+
+// I2C pins
+const (
+	SDA_PIN Pin = D11 // SDA
+	SCL_PIN Pin = D12 // SCL
+)
+
+// SPI pins
+const (
+	SPI0_SCK_PIN Pin = D9  // SCK: S1
+	SPI0_SDO_PIN Pin = D8  // SDO: S1
+	SPI0_SDI_PIN Pin = D10 // SDI: S1
+)
+
+// I2S pins
+const (
+	I2S_SCK_PIN Pin = PA10
+	I2S_SD_PIN  Pin = PA07
+	I2S_WS_PIN      = NoPin // TODO: figure out what this is on Arduino Nano 33.
+)
+
+// USB CDC identifiers
+const (
+	usb_STRING_PRODUCT      = "Arduino MKR1000"
+	usb_STRING_MANUFACTURER = "Arduino"
+)
+
+var (
+	usb_VID uint16 = 0x2341
+	usb_PID uint16 = 0x804e
+)
@@ -0,0 +1,101 @@
+// +build sam,atsamd21,arduino_zero
+
+package machine
+
+// used to reset into bootloader
+const RESET_MAGIC_VALUE = 0x07738135
+
+// GPIO Pins - Digital Low
+const (
+	D0 = PA11 // RX
+	D1 = PA10 // TX
+	D2 = PA14
+	D3 = PA09 // PWM available
+	D4 = PA08 // PWM available
+	D5 = PA15 // PWM available
+	D6 = PA20 // PWM available
+	D7 = PA21
+)
+
+// GPIO Pins - Digital High
+const (
+	D8  = PA06 // PWM available
+	D9  = PA07 // PWM available
+	D10 = PA18 // PWM available
+	D11 = PA16 // PWM available
+	D12 = PA19 // PWM available
+	D13 = PA17 // PWM available
+)
+
+// LEDs on the Arduino Zero
+const (
+	LED      = LED1
+	LED1 Pin = D13
+	LED2 Pin = PA27 // TX LED
+	LED3 Pin = PB03 // RX LED
+)
+
+// ADC pins
+const (
+	AREF Pin = PA03
+	ADC0 Pin = PA02
+	ADC1 Pin = PB08
+	ADC2 Pin = PB09
+	ADC3 Pin = PA04
+	ADC4 Pin = PA05
+	ADC5 Pin = PB02
+)
+
+// SPI pins - EDBG connected
+const (
+	SPI0_SDO_PIN Pin = PA16 // MOSI: SERCOM1/PAD[0]
+	SPI0_SDI_PIN Pin = PA19 // MISO: SERCOM1/PAD[2]
+	SPI0_SCK_PIN Pin = PA17 // SCK:  SERCOM1/PAD[3]
+)
+
+// SPI pins (Legacy ICSP)
+const (
+	SPI1_SDO_PIN Pin = PB10 // MOSI: SERCOM4/PAD[2] - Pin 4
+	SPI1_SDI_PIN Pin = PA12 // MISO: SERCOM4/PAD[0] - Pin 1
+	SPI1_SCK_PIN Pin = PB11 // SCK:  SERCOM4/PAD[3] - Pin 3
+)
+
+// I2C pins - EDBG connected
+const (
+	SDA_PIN Pin = PA22 // SDA: SERCOM3/PAD[0] - Pin 20
+	SCL_PIN Pin = PA23 // SCL: SERCOM3/PAD[1] - Pin 21
+)
+
+// I2S pins - might not be exposed
+const (
+	I2S_SCK_PIN Pin = PA10
+	I2S_SD_PIN  Pin = PA07
+	I2S_WS_PIN  Pin = PA11
+)
+
+// UART0 pins - EDBG connected
+const (
+	UART_RX_PIN Pin = D0
+	UART_TX_PIN Pin = D1
+)
+
+// 'native' USB port pins
+const (
+	USBCDC_DM_PIN Pin = PA24
+	USBCDC_DP_PIN Pin = PA25
+)
+
+// USB CDC identifiers
+const (
+	usb_STRING_PRODUCT      = "Arduino Zero"
+	usb_STRING_MANUFACTURER = "Arduino LLC"
+
+	usb_VID uint16 = 0x2341
+	usb_PID uint16 = 0x804d
+)
+
+// 32.768 KHz Crystal
+const (
+	XIN32  Pin = PA00
+	XOUT32 Pin = PA01
+)
@@ -65,11 +65,15 @@ var (
 		Buffer: NewRingBuffer(),
 		Bus:    stm32.USART1,
 	}
-	UART1 = &UART0
+	UART1 = UART{
+		Buffer: NewRingBuffer(),
+		Bus:    stm32.USART2,
+	}
 )

 func init() {
 	UART0.Interrupt = interrupt.New(stm32.IRQ_USART1, UART0.handleInterrupt)
+	UART1.Interrupt = interrupt.New(stm32.IRQ_USART2, UART1.handleInterrupt)
 }

 // SPI pins
@@ -2,7 +2,7 @@

 package machine

-const HasLowFrequencyCrystal = true
+const HasLowFrequencyCrystal = false

 // GPIO Pins
 const (
@@ -0,0 +1,88 @@
+// +build esp32_coreboard_v2
+
+package machine
+
+// Built-in LED on some ESP32 boards.
+const LED = IO2
+
+const (
+	CLK  Pin = 6
+	CMD  Pin = 11
+	IO0  Pin = 0
+	IO1  Pin = 1
+	IO10 Pin = 10
+	IO16 Pin = 16
+	IO17 Pin = 17
+	IO18 Pin = 18
+	IO19 Pin = 19
+	IO2  Pin = 2
+	IO21 Pin = 21
+	IO22 Pin = 22
+	IO23 Pin = 23
+	IO25 Pin = 25
+	IO26 Pin = 26
+	IO27 Pin = 27
+	IO3  Pin = 3
+	IO32 Pin = 32
+	IO33 Pin = 33
+	IO34 Pin = 34
+	IO35 Pin = 35
+	IO36 Pin = 36
+	IO39 Pin = 39
+	IO4  Pin = 4
+	IO5  Pin = 5
+	IO9  Pin = 9
+	RXD  Pin = 3
+	SD0  Pin = 7
+	SD1  Pin = 8
+	SD2  Pin = 9
+	SD3  Pin = 10
+	SVN  Pin = 39
+	SVP  Pin = 36
+	TCK  Pin = 13
+	TD0  Pin = 15
+	TDI  Pin = 12
+	TMS  Pin = 14
+	TXD  Pin = 1
+)
+
+// SPI pins
+const (
+	SPI0_SCK_PIN = IO18
+	SPI0_SDO_PIN = IO23
+	SPI0_SDI_PIN = IO19
+	SPI0_CS0_PIN = IO5
+)
+
+// I2C pins
+const (
+	SDA_PIN = IO21
+	SCL_PIN = IO22
+)
+
+// ADC pins
+const (
+	ADC0 Pin = IO34
+	ADC1 Pin = IO35
+	ADC2 Pin = IO36
+	ADC3 Pin = IO39
+)
+
+// UART0 pins
+const (
+	UART_TX_PIN = IO1
+	UART_RX_PIN = IO3
+)
+
+// UART1 pins
+const (
+	UART1_TX_PIN = IO9
+	UART1_RX_PIN = IO10
+)
+
+// PWM pins
+const (
+	PWM0_PIN Pin = IO2
+	PWM1_PIN Pin = IO0
+	PWM2_PIN Pin = IO4
+)
@@ -0,0 +1,245 @@
+// +build feather_stm32f405
+
+package machine
+
+import (
+	"device/stm32"
+	"runtime/interrupt"
+)
+
+const (
+	NUM_DIGITAL_IO_PINS = 39
+	NUM_ANALOG_IO_PINS  = 7
+)
+
+// Digital pins
+const (
+	// Arduino pin = MCU port pin // primary functions (alternate functions)
+	D0  = PB11 // USART3 RX, PWM TIM2_CH4 (I2C2 SDA)
+	D1  = PB10 // USART3 TX, PWM TIM2_CH3 (I2C2 SCL, I2S2 BCK)
+	D2  = PB3  // GPIO, SPI3 FLASH SCK
+	D3  = PB4  // GPIO, SPI3 FLASH MISO
+	D4  = PB5  // GPIO, SPI3 FLASH MOSI
+	D5  = PC7  // GPIO, PWM TIM3_CH2 (USART6 RX, I2S3 MCK)
+	D6  = PC6  // GPIO, PWM TIM3_CH1 (USART6 TX, I2S2 MCK)
+	D7  = PA15 // GPIO, SPI3 FLASH CS
+	D8  = PC0  // GPIO, Neopixel
+	D9  = PB8  // GPIO, PWM TIM4_CH3 (CAN1 RX, I2C1 SCL)
+	D10 = PB9  // GPIO, PWM TIM4_CH4 (CAN1 TX, I2C1 SDA, I2S2 WSL)
+	D11 = PC3  // GPIO (I2S2 SD, SPI2 MOSI)
+	D12 = PC2  // GPIO (I2S2ext SD, SPI2 MISO)
+	D13 = PC1  // GPIO, Builtin LED
+	D14 = PB7  // I2C1 SDA, PWM TIM4_CH2 (USART1 RX)
+	D15 = PB6  // I2C1 SCL, PWM TIM4_CH1 (USART1 TX, CAN2 TX)
+	D16 = PA4  // A0 (DAC OUT1)
+	D17 = PA5  // A1 (DAC OUT2, SPI1 SCK)
+	D18 = PA6  // A2, PWM TIM3_CH1 (SPI1 MISO)
+	D19 = PA7  // A3, PWM TIM3_CH2 (SPI1 MOSI)
+	D20 = PC4  // A4
+	D21 = PC5  // A5
+	D22 = PA3  // A6
+	D23 = PB13 // SPI2 SCK, PWM TIM1_CH1N (I2S2 BCK, CAN2 TX)
+	D24 = PB14 // SPI2 MISO, PWM TIM1_CH2N (I2S2ext SD)
+	D25 = PB15 // SPI2 MOSI, PWM TIM1_CH3N (I2S2 SD)
+	D26 = PC8  // SDIO
+	D27 = PC9  // SDIO
+	D28 = PC10 // SDIO
+	D29 = PC11 // SDIO
+	D30 = PC12 // SDIO
+	D31 = PD2  // SDIO
+	D32 = PB12 // SD Detect
+	D33 = PC14 // OSC32
+	D34 = PC15 // OSC32
+	D35 = PA11 // USB D+
+	D36 = PA12 // USB D-
+	D37 = PA13 // SWDIO
+	D38 = PA14 // SWCLK
+)
+
+// Analog pins
+const (
+	A0 = D16 // ADC12 IN4
+	A1 = D17 // ADC12 IN5
+	A2 = D18 // ADC12 IN6
+	A3 = D19 // ADC12 IN7
+	A4 = D20 // ADC12 IN14
+	A5 = D21 // ADC12 IN15
+	A6 = D22 // VBAT
+)
+
+func init() {
+	initLED()
+	initUART()
+	initSPI()
+	initI2C()
+}
+
+// -- LEDs ---------------------------------------------------------------------
+
+const (
+	NUM_BOARD_LED      = 1
+	NUM_BOARD_NEOPIXEL = 1
+
+	LED_RED      = D13
+	LED_NEOPIXEL = D8
+	LED_BUILTIN  = LED_RED
+	LED          = LED_BUILTIN
+)
+
+func initLED() {}
+
+// -- UART ---------------------------------------------------------------------
+
+const (
+	// #===========#==========#==============#============#=======#=======#
+	// | Interface | Hardware |  Bus(Freq)   | RX/TX Pins | AltFn | Alias |
+	// #===========#==========#==============#============#=======#=======#
+	// |   UART1   |  USART3  | APB1(42 MHz) |   D0/D1    |   7   |   ~   |
+	// |   UART2   |  USART6  | APB2(84 MHz) |   D5/D6    |   8   |   ~   |
+	// |   UART3   |  USART1  | APB2(84 MHz) |  D14/D15   |   7   |   ~   |
+	// | --------- | -------- | ------------ | ---------- | ----- | ----- |
+	// |   UART0   |  USART3  | APB1(42 MHz) |   D0/D1    |   7   | UART1 |
+	// #===========#==========#==============#============#=======#=======#
+	NUM_UART_INTERFACES = 3
+
+	UART1_RX_PIN = D0 // UART1 = hardware: USART3
+	UART1_TX_PIN = D1 //
+
+	UART2_RX_PIN = D5 // UART2 = hardware: USART6
+	UART2_TX_PIN = D6 //
+
+	UART3_RX_PIN = D14 // UART3 = hardware: USART1
+	UART3_TX_PIN = D15 //
+
+	UART0_RX_PIN = UART1_RX_PIN // UART0 = alias: UART1
+	UART0_TX_PIN = UART1_TX_PIN //
+
+	UART_RX_PIN = UART0_RX_PIN // default/primary UART pins
+	UART_TX_PIN = UART0_TX_PIN //
+)
+
+var (
+	UART1 = UART{
+		Buffer:          NewRingBuffer(),
+		Bus:             stm32.USART3,
+		AltFuncSelector: stm32.AF7_USART1_2_3,
+	}
+	UART2 = UART{
+		Buffer:          NewRingBuffer(),
+		Bus:             stm32.USART6,
+		AltFuncSelector: stm32.AF8_USART4_5_6,
+	}
+	UART3 = UART{
+		Buffer:          NewRingBuffer(),
+		Bus:             stm32.USART1,
+		AltFuncSelector: stm32.AF7_USART1_2_3,
+	}
+	UART0 = UART1
+)
+
+func initUART() {
+	UART1.Interrupt = interrupt.New(stm32.IRQ_USART3, UART1.handleInterrupt)
+	UART2.Interrupt = interrupt.New(stm32.IRQ_USART6, UART2.handleInterrupt)
+	UART3.Interrupt = interrupt.New(stm32.IRQ_USART1, UART3.handleInterrupt)
+}
+
+// -- SPI ----------------------------------------------------------------------
+
+const (
+	// #===========#==========#==============#==================#=======#=======#
+	// | Interface | Hardware |  Bus(Freq)   | SCK/SDI/SDO Pins | AltFn | Alias |
+	// #===========#==========#==============#==================#=======#=======#
+	// |   SPI1    |   SPI2   | APB1(42 MHz) |    D23/D24/D25   |   5   |   ~   |
+	// |   SPI2    |   SPI3   | APB1(42 MHz) |     D2/D3/D4     |   6   |   ~   |
+	// |   SPI3    |   SPI1   | APB2(84 MHz) |    D17/D18/D19   |   5   |   ~   |
+	// | --------- | -------- | ------------ | ---------------- | ----- | ----- |
+	// |   SPI0    |   SPI2   | APB1(42 MHz) |    D23/D24/D25   |   5   | SPI1  |
+	// #===========#==========#==============#==================#=======#=======#
+	NUM_SPI_INTERFACES = 3
+
+	SPI1_SCK_PIN = D23 //
+	SPI1_SDI_PIN = D24 // SPI1 = hardware: SPI2
+	SPI1_SDO_PIN = D25 //
+
+	SPI2_SCK_PIN = D2 //
+	SPI2_SDI_PIN = D3 // SPI2 = hardware: SPI3
+	SPI2_SDO_PIN = D4 //
+
+	SPI3_SCK_PIN = D17 //
+	SPI3_SDI_PIN = D18 // SPI3 = hardware: SPI1
+	SPI3_SDO_PIN = D19 //
+
+	SPI0_SCK_PIN = SPI1_SCK_PIN //
+	SPI0_SDI_PIN = SPI1_SDI_PIN // SPI0 = alias: SPI1
+	SPI0_SDO_PIN = SPI1_SDO_PIN //
+
+	SPI_SCK_PIN = SPI0_SCK_PIN //
+	SPI_SDI_PIN = SPI0_SDI_PIN // default/primary SPI pins
+	SPI_SDO_PIN = SPI0_SDO_PIN //
+)
+
+var (
+	SPI1 = SPI{
+		Bus:             stm32.SPI2,
+		AltFuncSelector: stm32.AF5_SPI1_SPI2,
+	}
+	SPI2 = SPI{
+		Bus:             stm32.SPI3,
+		AltFuncSelector: stm32.AF6_SPI3,
+	}
+	SPI3 = SPI{
+		Bus:             stm32.SPI1,
+		AltFuncSelector: stm32.AF5_SPI1_SPI2,
+	}
+	SPI0 = SPI1
+)
+
+func initSPI() {}
+
+// -- I2C ----------------------------------------------------------------------
+
+const (
+	// #===========#==========#==============#==============#=======#=======#
+	// | Interface | Hardware |  Bus(Freq)   | SDA/SCL Pins | AltFn | Alias |
+	// #===========#==========#==============#==============#=======#=======#
+	// |   I2C1    |   I2C1   | APB1(42 MHz) |   D14/D15    |   4   |   ~   |
+	// |   I2C2    |   I2C2   | APB1(42 MHz) |    D0/D1     |   4   |   ~   |
+	// |   I2C3    |   I2C1   | APB1(42 MHz) |    D9/D10    |   4   |   ~   |
+	// | --------- | -------- | ------------ | ------------ | ----- | ----- |
+	// |   I2C0    |   I2C1   | APB1(42 MHz) |   D14/D15    |   4   | I2C1  |
+	// #===========#==========#==============#==============#=======#=======#
+	NUM_I2C_INTERFACES = 3
+
+	I2C1_SDA_PIN = D14 // I2C1 = hardware: I2C1
+	I2C1_SCL_PIN = D15 //
+
+	I2C2_SDA_PIN = D0 // I2C2 = hardware: I2C2
+	I2C2_SCL_PIN = D1 //
+
+	I2C3_SDA_PIN = D9  // I2C3 = hardware: I2C1
+	I2C3_SCL_PIN = D10 //   (interface duplicated on second pair of pins)
+
+	I2C0_SDA_PIN = I2C1_SDA_PIN // I2C0 = alias: I2C1
+	I2C0_SCL_PIN = I2C1_SCL_PIN //
+
+	I2C_SDA_PIN = I2C0_SDA_PIN // default/primary I2C pins
+	I2C_SCL_PIN = I2C0_SCL_PIN //
+)
+
+var (
+	I2C1 = I2C{
+		Bus:             stm32.I2C1,
+		AltFuncSelector: stm32.AF4_I2C1_2_3,
+	}
+	I2C2 = I2C{
+		Bus:             stm32.I2C2,
+		AltFuncSelector: stm32.AF4_I2C1_2_3,
+	}
+	I2C3 = I2C{
+		Bus:             stm32.I2C1,
+		AltFuncSelector: stm32.AF4_I2C1_2_3,
+	}
+	I2C0 = I2C1
+)
+
+func initI2C() {}
@@ -0,0 +1,100 @@
+// +build itsybitsy_nrf52840
+
+package machine
+
+const HasLowFrequencyCrystal = true
+
+// GPIO Pins
+const (
+	D0  = P0_25 // UART TX
+	D1  = P0_24 // UART RX
+	D2  = P1_02
+	D3  = P0_06 // LED1
+	D4  = P0_29 // Button
+	D5  = P0_27
+	D6  = P1_09 // DotStar Clock
+	D7  = P1_08
+	D8  = P0_08 // DotStar Data
+	D9  = P0_07
+	D10 = P0_05
+	D11 = P0_26
+	D12 = P0_11
+	D13 = P0_12
+	D14 = P0_04 // A0
+	D15 = P0_30 // A1
+	D16 = P0_28 // A2
+	D17 = P0_31 // A3
+	D18 = P0_02 // A4
+	D19 = P0_03 // A5
+	D20 = P0_05 // A6
+	D21 = P0_16 // I2C SDA
+	D22 = P0_14 // I2C SCL
+	D23 = P0_20 // SPI SDI
+	D24 = P0_15 // SPI SDO
+	D25 = P0_13 // SPI SCK
+	D26 = P0_19 // QSPI SCK
+	D27 = P0_23 // QSPI CS
+	D28 = P0_21 // QSPI Data 0
+	D29 = P0_22 // QSPI Data 1
+	D30 = P1_00 // QSPI Data 2
+	D31 = P0_17 // QSPI Data 3
+)
+
+// Analog Pins
+const (
+	A0 = D14
+	A1 = D15
+	A2 = D16
+	A3 = D17
+	A4 = D18
+	A5 = D19
+	A6 = D20
+)
+
+const (
+	LED    = D3
+	LED1   = LED
+	BUTTON = D4
+
+	QSPI_SCK   = D26
+	QSPI_CS    = D27
+	QSPI_DATA0 = D28
+	QSPI_DATA1 = D29
+	QSPI_DATA2 = D30
+	QSPI_DATA3 = D31
+)
+
+// UART0 pins (logical UART1)
+const (
+	UART_RX_PIN = D0
+	UART_TX_PIN = D1
+)
+
+// UART0 is the USB device
+var (
+	UART0 = USB
+)
+
+// I2C pins
+const (
+	SDA_PIN = D21 // I2C0 external
+	SCL_PIN = D22 // I2C0 external
+)
+
+// SPI pins
+const (
+	SPI0_SCK_PIN = D25
+	SPI0_SDO_PIN = D24
+	SPI0_SDI_PIN = D23
+)
+
+// USB CDC identifiers
+const (
+	usb_STRING_PRODUCT      = "Adafruit ItsyBitsy nRF52840 Express"
+	usb_STRING_MANUFACTURER = "Adafruit"
+)
+
+var (
+	usb_VID uint16 = 0x239A
+	usb_PID uint16 = 0x8051
+)
@@ -0,0 +1,178 @@
+// +build matrixportal_m4
+
+package machine
+
+// used to reset into bootloader
+const RESET_MAGIC_VALUE = 0xF01669EF
+
+// Digital pins
+const (
+	//    Pin   // Function          SERCOM  PWM  Interrupt
+	//    ----  // ----------------  ------  ---  ---------
+	D0  = PA01  // UART RX            1[1]   PWM  EXTI1
+	D1  = PA00  // UART TX            1[0]   PWM  EXTI0
+	D2  = PB22  // Button "Up"                    EXTI6
+	D3  = PB23  // Button "Down"                  EXTI7
+	D4  = PA23  // NeoPixel                       EXTI7
+	D5  = PB31  // I2C SDA            5[1]        EXTI15
+	D6  = PB30  // I2C SCL            5[0]        EXTI14
+	D7  = PB00  // HUB75 R1                       EXTI0
+	D8  = PB01  // HUB75 G1                       EXTI1
+	D9  = PB02  // HUB75 B1                       EXTI2
+	D10 = PB03  // HUB75 R2                       EXTI3
+	D11 = PB04  // HUB75 G2                       EXTI4
+	D12 = PB05  // HUB75 B2                       EXTI5
+	D13 = PA14  // LED                       PWM  EXTI14
+	D14 = PB06  // HUB75 CLK                      EXTI6
+	D15 = PB14  // HUB75 LAT                      EXTI14
+	D16 = PB12  // HUB75 OE                       EXTI12
+	D17 = PB07  // HUB75 ADDR A                   EXTI7
+	D18 = PB08  // HUB75 ADDR B                   EXTI8
+	D19 = PB09  // HUB75 ADDR C                   EXTI9
+	D20 = PB15  // HUB75 ADDR D                   EXTI15
+	D21 = PB13  // HUB75 ADDR E                   EXTI13
+	D22 = PA02  // ADC (A0)                       EXTI2
+	D23 = PA05  // ADC (A1)                       EXTI5
+	D24 = PA04  // ADC (A2)                  PWM  EXTI4
+	D25 = PA06  // ADC (A3)                  PWM  EXTI6
+	D26 = PA07  // ADC (A4)                       EXTI7
+	D27 = PA12  // ESP32 UART RX      4[1]   PWM  EXTI12
+	D28 = PA13  // ESP32 UART TX      4[0]   PWM  EXTI13
+	D29 = PA20  // ESP32 GPIO0               PWM  EXTI4
+	D30 = PA21  // ESP32 Reset               PWM  EXTI5
+	D31 = PA22  // ESP32 Busy                PWM  EXTI6
+	D32 = PA18  // ESP32 RTS                 PWM  EXTI2
+	D33 = PB17  // ESP32 SPI CS              PWM  EXTI1
+	D34 = PA16  // ESP32 SPI SCK      3[1]   PWM  EXTI0
+	D35 = PA17  // ESP32 SPI SDI      3[0]   PWM  EXTI1
+	D36 = PA19  // ESP32 SPI SDO      1[3]   PWM  EXTI3
+	D37 = NoPin // USB Host enable
+	D38 = PA24  // USB DM
+	D39 = PA27  // USB DP
+	D40 = PA03  // DAC/VREFP
+	D41 = PB10  // Flash QSPI SCK
+	D42 = PB11  // Flash QSPI CS
+	D43 = PA08  // Flash QSPI I00
+	D44 = PA09  // Flash QSPI IO1
+	D45 = PA10  // Flash QSPI IO2
+	D46 = PA11  // Flash QSPI IO3
+	D47 = PA27  // LIS3DH IRQ                     EXTI11
+	D48 = PA05  // SPI SCK            0[1]        EXTI5
+	D49 = PA04  // SPI SDO            0[0]   PWM  EXTI4
+	D50 = PA07  // SPI SDI            0[3]        EXTI7
+)
+
+// Analog pins
+const (
+	A0 = PA02 // ADC Channel 0
+	A1 = PA05 // ADC Channel 5
+	A2 = PA04 // ADC Channel 4
+	A3 = PA06 // ADC Channel 6
+	A4 = PA07 // ADC Channel 7
+)
+
+// LED pins
+const (
+	LED      = D13
+	NEOPIXEL = D4
+)
+
+// Button pins
+const (
+	BUTTON_UP   = D2
+	BUTTON_DOWN = D3
+)
+
+// UART pins
+const (
+	UART1_RX_PIN = D0 // SERCOM1[1]
+	UART1_TX_PIN = D1 // SERCOM1[0]
+
+	UART2_RX_PIN = D27 // SERCOM4[1] (ESP32 RX)
+	UART2_TX_PIN = D28 // SERCOM4[0] (ESP32 TX)
+
+	UART_RX_PIN = UART1_RX_PIN
+	UART_TX_PIN = UART1_TX_PIN
+)
+
+// SPI pins
+const (
+	SPI0_SCK_PIN = D34 // SERCOM3[1] (ESP32 SCK)
+	SPI0_SDO_PIN = D36 // SERCOM1[3] (ESP32 SDO)
+	SPI0_SDI_PIN = D35 // SERCOM3[0] (ESP32 SDI)
+
+	SPI1_SCK_PIN = D48 // SERCOM0[1]
+	SPI1_SDO_PIN = D49 // SERCOM0[0]
+	SPI1_SDI_PIN = D50 // SERCOM0[3]
+
+	SPI_SCK_PIN = SPI0_SCK_PIN
+	SPI_SDO_PIN = SPI0_SDO_PIN
+	SPI_SDI_PIN = SPI0_SDI_PIN
+)
+
+// I2C pins
+const (
+	I2C0_SDA_PIN = D5 // SERCOM5[1]
+	I2C0_SCL_PIN = D6 // SERCOM5[0]
+
+	I2C_SDA_PIN = I2C0_SDA_PIN
+	I2C_SCL_PIN = I2C0_SCL_PIN
+
+	SDA_PIN = I2C_SDA_PIN // awkward naming required by machine_atsamd51.go
+	SCL_PIN = I2C_SCL_PIN //
+)
+
+// ESP32 pins
+const (
+	NINA_ACK    = D31
+	NINA_GPIO0  = D29
+	NINA_RESETN = D30
+
+	NINA_RX  = UART2_RX_PIN
+	NINA_TX  = UART2_TX_PIN
+	NINA_RTS = D32
+
+	NINA_CS  = D33
+	NINA_SDO = SPI0_SDO_PIN
+	NINA_SDI = SPI0_SDI_PIN
+	NINA_SCK = SPI0_SCK_PIN
+)
+
+// HUB75 pins
+const (
+	HUB75_R1 = D7
+	HUB75_G1 = D8
+	HUB75_B1 = D9
+	HUB75_R2 = D10
+	HUB75_G2 = D11
+	HUB75_B2 = D12
+
+	HUB75_CLK    = D14
+	HUB75_LAT    = D15
+	HUB75_OE     = D16
+	HUB75_ADDR_A = D17
+	HUB75_ADDR_B = D18
+	HUB75_ADDR_C = D19
+	HUB75_ADDR_D = D20
+	HUB75_ADDR_E = D21
+)
+
+// USB CDC pins (UART0)
+const (
+	USBCDC_DM_PIN = D38
+	USBCDC_DP_PIN = D39
+
+	UART0_RX_PIN = USBCDC_DM_PIN
+	UART0_TX_PIN = USBCDC_DP_PIN
+)
+
+// USB CDC identifiers
+const (
+	usb_STRING_PRODUCT      = "Matrix Portal M4"
+	usb_STRING_MANUFACTURER = "Adafruit Industries"
+)
+
+var (
+	usb_VID uint16 = 0x239A
+	usb_PID uint16 = 0x80C9
+)
@@ -0,0 +1,50 @@
+// +build sam,atsamd51,matrixportal_m4
+
+package machine
+
+import (
+	"device/sam"
+	"runtime/interrupt"
+)
+
+// UART on the MatrixPortal M4
+var (
+	UART1 = UART{
+		Buffer: NewRingBuffer(),
+		Bus:    sam.SERCOM1_USART_INT,
+		SERCOM: 1,
+	}
+
+	UART2 = UART{
+		Buffer: NewRingBuffer(),
+		Bus:    sam.SERCOM4_USART_INT,
+		SERCOM: 4,
+	}
+)
+
+func init() {
+	UART1.Interrupt = interrupt.New(sam.IRQ_SERCOM1_1, UART1.handleInterrupt)
+	UART2.Interrupt = interrupt.New(sam.IRQ_SERCOM4_1, UART2.handleInterrupt)
+}
+
+// I2C on the MatrixPortal M4
+var (
+	I2C0 = I2C{
+		Bus:    sam.SERCOM5_I2CM,
+		SERCOM: 5,
+	}
+)
+
+// SPI on the MatrixPortal M4
+var (
+	SPI0 = SPI{
+		Bus:    sam.SERCOM3_SPIM,
+		SERCOM: 3, // BUG: SDO on SERCOM1!
+	}
+	NINA_SPI = SPI0
+
+	SPI1 = SPI{
+		Bus:    sam.SERCOM0_SPIM,
+		SERCOM: 0,
+	}
+)
@@ -0,0 +1,35 @@
+// +build nodemcu
+
+// Pinout for the NodeMCU dev kit.
+
+package machine
+
+// GPIO pins on the NodeMCU board.
+const (
+	D0 Pin = 16
+	D1 Pin = 5
+	D2 Pin = 4
+	D3 Pin = 0
+	D4 Pin = 2
+	D5 Pin = 14
+	D6 Pin = 12
+	D7 Pin = 13
+	D8 Pin = 15
+)
+
+// Onboard blue LED (on the AI-Thinker module).
+const LED = D4
+
+// SPI pins
+const (
+	SPI0_SCK_PIN = D5
+	SPI0_SDO_PIN = D7
+	SPI0_SDI_PIN = D6
+	SPI0_CS0_PIN = D8
+)
+
+// I2C pins
+const (
+	SDA_PIN = D2
+	SCL_PIN = D1
+)
@@ -0,0 +1,58 @@
+// +build nucleof722ze
+
+package machine
+
+import (
+	"device/stm32"
+	"runtime/interrupt"
+)
+
+const (
+	LED         = LED_BUILTIN
+	LED_BUILTIN = LED_GREEN
+	LED_GREEN   = PB0
+	LED_BLUE    = PB7
+	LED_RED     = PB14
+)
+
+const (
+	BUTTON      = BUTTON_USER
+	BUTTON_USER = PC13
+)
+
+// UART pins
+const (
+	// PD8 and PD9 are connected to the ST-Link Virtual Com Port (VCP)
+	UART_TX_PIN = PD8
+	UART_RX_PIN = PD9
+	UART_ALT_FN = 7 // GPIO_AF7_UART3
+)
+
+var (
+	// USART3 is the hardware serial port connected to the onboard ST-LINK
+	// debugger to be exposed as virtual COM port over USB on Nucleo boards.
+	// Both UART0 and UART1 refer to USART2.
+	UART0 = UART{
+		Buffer:          NewRingBuffer(),
+		Bus:             stm32.USART3,
+		AltFuncSelector: UART_ALT_FN,
+	}
+	UART1 = &UART0
+)
+
+func init() {
+	UART0.Interrupt = interrupt.New(stm32.IRQ_USART3, UART0.handleInterrupt)
+}
+
+// SPI pins
+const (
+	SPI0_SCK_PIN = PA5
+	SPI0_SDI_PIN = PA6
+	SPI0_SDO_PIN = PA7
+)
+
+// I2C pins
+const (
+	SCL_PIN = PB6
+	SDA_PIN = PB7
+)
@@ -0,0 +1,118 @@
+// +build sam,atsamd21,qtpy
+
+package machine
+
+import (
+	"device/sam"
+	"runtime/interrupt"
+)
+
+// used to reset into bootloader
+const RESET_MAGIC_VALUE = 0xf01669ef
+
+// GPIO Pins
+const (
+	D0  = PA02 // PWM available
+	D1  = PA03
+	D2  = PA04 // PWM available
+	D3  = PA05 // PWM available
+	D4  = PA16 // PWM available
+	D5  = PA17 // PWM available
+	D6  = PA06
+	D7  = PA07
+	D8  = PA11
+	D9  = PA09
+	D10 = PA10
+	D11 = PA18
+	D12 = PA15
+	D13 = PA27
+	D14 = PA23
+	D15 = PA19
+	D16 = PA22
+	D17 = PA08
+)
+
+// Analog pins
+const (
+	A0 = D1
+	A1 = D1
+	A2 = D2
+	A3 = D3
+	A4 = D4
+)
+
+const (
+	LED = D13
+)
+
+// UART0 aka USBCDC pins
+const (
+	USBCDC_DM_PIN = PA24
+	USBCDC_DP_PIN = PA25
+)
+
+// UART1 pins
+const (
+	UART_TX_PIN = D6
+	UART_RX_PIN = D7
+)
+
+// UART1 on the QT Py M0.
+var (
+	UART1 = UART{
+		Buffer: NewRingBuffer(),
+		Bus:    sam.SERCOM0_USART,
+		SERCOM: 0,
+	}
+)
+
+func init() {
+	UART1.Interrupt = interrupt.New(sam.IRQ_SERCOM0, UART1.handleInterrupt)
+}
+
+// SPI pins
+const (
+	SPI0_SCK_PIN = D8
+	SPI0_SDO_PIN = D10
+	SPI0_SDI_PIN = D9
+)
+
+// SPI on the QT Py M0.
+var (
+	SPI0 = SPI{
+		Bus:    sam.SERCOM0_SPI,
+		SERCOM: 0,
+	}
+)
+
+// I2C pins
+const (
+	SDA_PIN = D4 // SDA
+	SCL_PIN = D5 // SCL
+)
+
+// I2C on the QT Py M0.
+var (
+	I2C0 = I2C{
+		Bus:    sam.SERCOM2_I2CM,
+		SERCOM: 2,
+	}
+)
+
+// I2S pins
+const (
+	I2S_SCK_PIN = PA10
+	I2S_SD_PIN  = PA08
+	I2S_WS_PIN  = NoPin // TODO: figure out what this is on QT Py M0.
+)
+
+// USB CDC identifiers
+const (
+	usb_STRING_PRODUCT      = "Adafruit QTPy M0"
+	usb_STRING_MANUFACTURER = "Adafruit"
+)
+
+var (
+	usb_VID uint16 = 0x239A
+	usb_PID uint16 = 0x80CB
+)
@@ -1,16 +1,5 @@
-// +build bluepill nucleof103rb stm32f4disco
+// +build bluepill nucleof103rb stm32f4

 package machine

 // Peripheral abstraction layer for the stm32.
-
-const (
-	portA Pin = iota * 16
-	portB
-	portC
-	portD
-	portE
-	portF
-	portG
-	portH
-)
@@ -7,96 +7,6 @@ import (
 	"runtime/interrupt"
 )

-const (
-	PA0  = portA + 0
-	PA1  = portA + 1
-	PA2  = portA + 2
-	PA3  = portA + 3
-	PA4  = portA + 4
-	PA5  = portA + 5
-	PA6  = portA + 6
-	PA7  = portA + 7
-	PA8  = portA + 8
-	PA9  = portA + 9
-	PA10 = portA + 10
-	PA11 = portA + 11
-	PA12 = portA + 12
-	PA13 = portA + 13
-	PA14 = portA + 14
-	PA15 = portA + 15
-
-	PB0  = portB + 0
-	PB1  = portB + 1
-	PB2  = portB + 2
-	PB3  = portB + 3
-	PB4  = portB + 4
-	PB5  = portB + 5
-	PB6  = portB + 6
-	PB7  = portB + 7
-	PB8  = portB + 8
-	PB9  = portB + 9
-	PB10 = portB + 10
-	PB11 = portB + 11
-	PB12 = portB + 12
-	PB13 = portB + 13
-	PB14 = portB + 14
-	PB15 = portB + 15
-
-	PC0  = portC + 0
-	PC1  = portC + 1
-	PC2  = portC + 2
-	PC3  = portC + 3
-	PC4  = portC + 4
-	PC5  = portC + 5
-	PC6  = portC + 6
-	PC7  = portC + 7
-	PC8  = portC + 8
-	PC9  = portC + 9
-	PC10 = portC + 10
-	PC11 = portC + 11
-	PC12 = portC + 12
-	PC13 = portC + 13
-	PC14 = portC + 14
-	PC15 = portC + 15
-
-	PD0  = portD + 0
-	PD1  = portD + 1
-	PD2  = portD + 2
-	PD3  = portD + 3
-	PD4  = portD + 4
-	PD5  = portD + 5
-	PD6  = portD + 6
-	PD7  = portD + 7
-	PD8  = portD + 8
-	PD9  = portD + 9
-	PD10 = portD + 10
-	PD11 = portD + 11
-	PD12 = portD + 12
-	PD13 = portD + 13
-	PD14 = portD + 14
-	PD15 = portD + 15
-
-	PE0  = portE + 0
-	PE1  = portE + 1
-	PE2  = portE + 2
-	PE3  = portE + 3
-	PE4  = portE + 4
-	PE5  = portE + 5
-	PE6  = portE + 6
-	PE7  = portE + 7
-	PE8  = portE + 8
-	PE9  = portE + 9
-	PE10 = portE + 10
-	PE11 = portE + 11
-	PE12 = portE + 12
-	PE13 = portE + 13
-	PE14 = portE + 14
-	PE15 = portE + 15
-
-	PH0 = portH + 0
-	PH1 = portH + 1
-)
-
 const (
 	LED         = LED_BUILTIN
 	LED1        = LED_GREEN
@@ -0,0 +1,262 @@
+// +build teensy40
+
+package machine
+
+import (
+	"device/nxp"
+	"runtime/interrupt"
+)
+
+// Digital pins
+const (
+	//  = Pin  // [Pad]:       Alt Func 0       Alt Func 1       Alt Func 2       Alt Func 3      Alt Func 4            Alt Func 5  Alt Func 6            Alt Func 7             Alt Func 8             Alt Func 9
+	//  = ----    -----------  ---------------  ---------------  ---------------  --------------  --------------------  ----------  --------------------  ---------------------  ---------------------  ----------------
+	D0  = PA3  // [AD_B0_03]:  FLEXCAN2_RX      XBAR1_INOUT17    LPUART6_RX       USB_OTG1_OC     FLEXPWM1_PWMX01       GPIO1_IO03  REF_CLK_24M           LPSPI3_PCS0             ~                      ~
+	D1  = PA2  // [AD_B0_02]:  FLEXCAN2_TX      XBAR1_INOUT16    LPUART6_TX       USB_OTG1_PWR    FLEXPWM1_PWMX00       GPIO1_IO02  LPI2C1_HREQ           LPSPI3_SDI              ~                      ~
+	D2  = PD4  // [EMC_04]:    SEMC_DATA04      FLEXPWM4_PWMA02  SAI2_TX_DATA     XBAR1_INOUT06   FLEXIO1_FLEXIO04      GPIO4_IO04   ~                     ~                      ~                      ~
+	D3  = PD5  // [EMC_05]:    SEMC_DATA05      FLEXPWM4_PWMB02  SAI2_TX_SYNC     XBAR1_INOUT07   FLEXIO1_FLEXIO05      GPIO4_IO05   ~                     ~                      ~                      ~
+	D4  = PD6  // [EMC_06]:    SEMC_DATA06      FLEXPWM2_PWMA00  SAI2_TX_BCLK     XBAR1_INOUT08   FLEXIO1_FLEXIO06      GPIO4_IO06   ~                     ~                      ~                      ~
+	D5  = PD8  // [EMC_08]:    SEMC_DM00        FLEXPWM2_PWMA01  SAI2_RX_DATA     XBAR1_INOUT17   FLEXIO1_FLEXIO08      GPIO4_IO08   ~                     ~                      ~                      ~
+	D6  = PB10 // [B0_10]:     LCD_DATA06       QTIMER4_TIMER1   FLEXPWM2_PWMA02  SAI1_TX_DATA03  FLEXIO2_FLEXIO10      GPIO2_IO10  SRC_BOOT_CFG06        ENET2_CRS               ~                      ~
+	D7  = PB17 // [B1_01]:     LCD_DATA13       XBAR1_INOUT15    LPUART4_RX       SAI1_TX_DATA00  FLEXIO2_FLEXIO17      GPIO2_IO17  FLEXPWM1_PWMB03       ENET2_RDATA00          FLEXIO3_FLEXIO17        ~
+	D8  = PB16 // [B1_00]:     LCD_DATA12       XBAR1_INOUT14    LPUART4_TX       SAI1_RX_DATA00  FLEXIO2_FLEXIO16      GPIO2_IO16  FLEXPWM1_PWMA03       ENET2_RX_ER            FLEXIO3_FLEXIO16        ~
+	D9  = PB11 // [B0_11]:     LCD_DATA07       QTIMER4_TIMER2   FLEXPWM2_PWMB02  SAI1_TX_DATA02  FLEXIO2_FLEXIO11      GPIO2_IO11  SRC_BOOT_CFG07        ENET2_COL               ~                      ~
+	D10 = PB0  // [B0_00]:     LCD_CLK          QTIMER1_TIMER0   MQS_RIGHT        LPSPI4_PCS0     FLEXIO2_FLEXIO00      GPIO2_IO00  SEMC_CSX01            ENET2_MDC               ~                      ~
+	D11 = PB2  // [B0_02]:     LCD_HSYNC        QTIMER1_TIMER2   FLEXCAN1_TX      LPSPI4_SDO      FLEXIO2_FLEXIO02      GPIO2_IO02  SEMC_CSX03            ENET2_1588_EVENT0_OUT   ~                      ~
+	D12 = PB1  // [B0_01]:     LCD_ENABLE       QTIMER1_TIMER1   MQS_LEFT         LPSPI4_SDI      FLEXIO2_FLEXIO01      GPIO2_IO01  SEMC_CSX02            ENET2_MDIO              ~                      ~
+	D13 = PB3  // [B0_03]:     LCD_VSYNC        QTIMER2_TIMER0   FLEXCAN1_RX      LPSPI4_SCK      FLEXIO2_FLEXIO03      GPIO2_IO03  WDOG2_RESET_B_DEB     ENET2_1588_EVENT0_IN    ~                      ~
+	D14 = PA18 // [AD_B1_02]:  USB_OTG1_ID      QTIMER3_TIMER2   LPUART2_TX       SPDIF_OUT       ENET_1588_EVENT2_OUT  GPIO1_IO18  USDHC1_CD_B           KPP_ROW06              GPT2_CLK               FLEXIO3_FLEXIO02
+	D15 = PA19 // [AD_B1_03]:  USB_OTG1_OC      QTIMER3_TIMER3   LPUART2_RX       SPDIF_IN        ENET_1588_EVENT2_IN   GPIO1_IO19  USDHC2_CD_B           KPP_COL06              GPT2_CAPTURE1          FLEXIO3_FLEXIO03
+	D16 = PA23 // [AD_B1_07]:  FLEXSPIB_DATA00  LPI2C3_SCL       LPUART3_RX       SPDIF_EXT_CLK   CSI_HSYNC             GPIO1_IO23  USDHC2_DATA3          KPP_COL04              GPT2_COMPARE3          FLEXIO3_FLEXIO07
+	D17 = PA22 // [AD_B1_06]:  FLEXSPIB_DATA01  LPI2C3_SDA       LPUART3_TX       SPDIF_LOCK      CSI_VSYNC             GPIO1_IO22  USDHC2_DATA2          KPP_ROW04              GPT2_COMPARE2          FLEXIO3_FLEXIO06
+	D18 = PA17 // [AD_B1_01]:  USB_OTG1_PWR     QTIMER3_TIMER1   LPUART2_RTS_B    LPI2C1_SDA      CCM_PMIC_READY        GPIO1_IO17  USDHC1_VSELECT        KPP_COL07              ENET2_1588_EVENT0_IN   FLEXIO3_FLEXIO01
+	D19 = PA16 // [AD_B1_00]:  USB_OTG2_ID      QTIMER3_TIMER0   LPUART2_CTS_B    LPI2C1_SCL      WDOG1_B               GPIO1_IO16  USDHC1_WP             KPP_ROW07              ENET2_1588_EVENT0_OUT  FLEXIO3_FLEXIO00
+	D20 = PA26 // [AD_B1_10]:  FLEXSPIA_DATA03  WDOG1_B          LPUART8_TX       SAI1_RX_SYNC    CSI_DATA07            GPIO1_IO26  USDHC2_WP             KPP_ROW02              ENET2_1588_EVENT1_OUT  FLEXIO3_FLEXIO10
+	D21 = PA27 // [AD_B1_11]:  FLEXSPIA_DATA02  EWM_OUT_B        LPUART8_RX       SAI1_RX_BCLK    CSI_DATA06            GPIO1_IO27  USDHC2_RESET_B        KPP_COL02              ENET2_1588_EVENT1_IN   FLEXIO3_FLEXIO11
+	D22 = PA24 // [AD_B1_08]:  FLEXSPIA_SS1_B   FLEXPWM4_PWMA00  FLEXCAN1_TX      CCM_PMIC_READY  CSI_DATA09            GPIO1_IO24  USDHC2_CMD            KPP_ROW03              FLEXIO3_FLEXIO08        ~
+	D23 = PA25 // [AD_B1_09]:  FLEXSPIA_DQS     FLEXPWM4_PWMA01  FLEXCAN1_RX      SAI1_MCLK       CSI_DATA08            GPIO1_IO25  USDHC2_CLK            KPP_COL03              FLEXIO3_FLEXIO09        ~
+	D24 = PA12 // [AD_B0_12]:  LPI2C4_SCL       CCM_PMIC_READY   LPUART1_TX       WDOG2_WDOG_B    FLEXPWM1_PWMX02       GPIO1_IO12  ENET_1588_EVENT1_OUT  NMI_GLUE_NMI            ~                      ~
+	D25 = PA13 // [AD_B0_13]:  LPI2C4_SDA       GPT1_CLK         LPUART1_RX       EWM_OUT_B       FLEXPWM1_PWMX03       GPIO1_IO13  ENET_1588_EVENT1_IN   REF_CLK_24M             ~                      ~
+	D26 = PA30 // [AD_B1_14]:  FLEXSPIA_SCLK    ACMP_OUT02       LPSPI3_SDO       SAI1_TX_BCLK    CSI_DATA03            GPIO1_IO30  USDHC2_DATA6          KPP_ROW00              ENET2_1588_EVENT3_OUT  FLEXIO3_FLEXIO14
+	D27 = PA31 // [AD_B1_15]:  FLEXSPIA_SS0_B   ACMP_OUT03       LPSPI3_SCK       SAI1_TX_SYNC    CSI_DATA02            GPIO1_IO31  USDHC2_DATA7          KPP_COL00              ENET2_1588_EVENT3_IN   FLEXIO3_FLEXIO15
+	D28 = PC18 // [EMC_32]:    SEMC_DATA10      FLEXPWM3_PWMB01  LPUART7_RX       CCM_PMIC_RDY    CSI_DATA21            GPIO3_IO18  ENET2_TX_EN            ~                      ~                      ~
+	D29 = PD31 // [EMC_31]:    SEMC_DATA09      FLEXPWM3_PWMA01  LPUART7_TX       LPSPI1_PCS1     CSI_DATA22            GPIO4_IO31  ENET2_TDATA01          ~                      ~                      ~
+	D30 = PC23 // [EMC_37]:    SEMC_DATA15      XBAR1_IN23       GPT1_COMPARE3    SAI3_MCLK       CSI_DATA16            GPIO3_IO23  USDHC2_WP             ENET2_RX_EN            FLEXCAN3_RX             ~
+	D31 = PC22 // [EMC_36]:    SEMC_DATA14      XBAR1_IN22       GPT1_COMPARE2    SAI3_TX_DATA    CSI_DATA17            GPIO3_IO22  USDHC1_WP             ENET2_RDATA01          FLEXCAN3_TX             ~
+	D32 = PB12 // [B0_12]:     LCD_DATA08       XBAR1_INOUT10    ARM_TRACE_CLK    SAI1_TX_DATA01  FLEXIO2_FLEXIO12      GPIO2_IO12  SRC_BOOT_CFG08        ENET2_TDATA00           ~                      ~
+	D33 = PD7  // [EMC_07]:    SEMC_DATA07      FLEXPWM2_PWMB00  SAI2_MCLK        XBAR1_INOUT09   FLEXIO1_FLEXIO07      GPIO4_IO07   ~                     ~                      ~                      ~
+	D34 = PC15 // [SD_B0_03]:  USDHC1_DATA1     FLEXPWM1_PWMB01  LPUART8_RTS_B    XBAR1_INOUT07   LPSPI1_SDI            GPIO3_IO15  ENET2_RDATA00         SEMC_CLK6               ~                      ~
+	D35 = PC14 // [SD_B0_02]:  USDHC1_DATA0     FLEXPWM1_PWMA01  LPUART8_CTS_B    XBAR1_INOUT06   LPSPI1_SDO            GPIO3_IO14  ENET2_RX_ER           SEMC_CLK5               ~                      ~
+	D36 = PC13 // [SD_B0_01]:  USDHC1_CLK       FLEXPWM1_PWMB00  LPI2C3_SDA       XBAR1_INOUT05   LPSPI1_PCS0           GPIO3_IO13  FLEXSPIB_SS1_B        ENET2_TX_CLK           ENET2_REF_CLK2          ~
+	D37 = PC12 // [SD_B0_00]:  USDHC1_CMD       FLEXPWM1_PWMA00  LPI2C3_SCL       XBAR1_INOUT04   LPSPI1_SCK            GPIO3_IO12  FLEXSPIA_SS1_B        ENET2_TX_EN            SEMC_DQS4               ~
+	D38 = PC17 // [SD_B0_05]:  USDHC1_DATA3     FLEXPWM1_PWMB02  LPUART8_RX       XBAR1_INOUT09   FLEXSPIB_DQS          GPIO3_IO17  CCM_CLKO2             ENET2_RX_EN             ~                      ~
+	D39 = PC16 // [SD_B0_04]:  USDHC1_DATA2     FLEXPWM1_PWMA02  LPUART8_TX       XBAR1_INOUT08   FLEXSPIB_SS0_B        GPIO3_IO16  CCM_CLKO1             ENET2_RDATA01           ~                      ~
+)
+
+// Analog pins
+const (
+	//  = Pin  // Dig | [Pad]      {ADC1/ADC2}
+	A0  = PA18 // D14 | [AD_B1_02] {  7 / 7  }
+	A1  = PA19 // D15 | [AD_B1_03] {  8 / 8  }
+	A2  = PA23 // D16 | [AD_B1_07] { 12 / 12 }
+	A3  = PA22 // D17 | [AD_B1_06] { 11 / 11 }
+	A4  = PA17 // D18 | [AD_B1_01] {  6 / 6  }
+	A5  = PA16 // D19 | [AD_B1_00] {  5 / 5  }
+	A6  = PA26 // D20 | [AD_B1_10] { 15 / 15 }
+	A7  = PA27 // D21 | [AD_B1_11] {  0 / 0  }
+	A8  = PA24 // D22 | [AD_B1_08] { 13 / 13 }
+	A9  = PA25 // D23 | [AD_B1_09] { 14 / 14 }
+	A10 = PA12 // D24 | [AD_B0_12] {  1 / -  }
+	A11 = PA13 // D25 | [AD_B0_13] {  2 / -  }
+	A12 = PA30 // D26 | [AD_B1_14] {  - / 3  }
+	A13 = PA31 // D27 | [AD_B1_15] {  - / 4  }
+)
+
+// Default peripheral pins
+const (
+	LED = D13
+
+	UART_RX_PIN = UART1_RX_PIN // D0
+	UART_TX_PIN = UART1_TX_PIN // D1
+
+	SPI_SDI_PIN = SPI1_SDI_PIN // D12
+	SPI_SDO_PIN = SPI1_SDO_PIN // D11
+	SPI_SCK_PIN = SPI1_SCK_PIN // D13
+	SPI_CS_PIN  = SPI1_CS_PIN  // D10
+
+	I2C_SDA_PIN = I2C1_SDA_PIN // D18/A4
+	I2C_SCL_PIN = I2C1_SCL_PIN // D19/A5
+)
+
+func init() {
+	// register any interrupt handlers for this board's peripherals
+	UART1.Interrupt = interrupt.New(nxp.IRQ_LPUART6, UART1.handleInterrupt)
+	UART2.Interrupt = interrupt.New(nxp.IRQ_LPUART4, UART2.handleInterrupt)
+	UART3.Interrupt = interrupt.New(nxp.IRQ_LPUART2, UART3.handleInterrupt)
+	UART4.Interrupt = interrupt.New(nxp.IRQ_LPUART3, UART4.handleInterrupt)
+	UART5.Interrupt = interrupt.New(nxp.IRQ_LPUART8, UART5.handleInterrupt)
+	UART6.Interrupt = interrupt.New(nxp.IRQ_LPUART1, UART6.handleInterrupt)
+	UART7.Interrupt = interrupt.New(nxp.IRQ_LPUART7, UART7.handleInterrupt)
+}
+
+// #=====================================================#
+// |                        UART                         |
+// #===========#===========#=============#===============#
+// | Interface | Hardware  | Clock(Freq) |  RX/TX  : Alt |
+// #===========#===========#=============#=========-=====#
+// |   UART1   |  LPUART6  | OSC(24 MHz) |  D0/D1  : 2/2 |
+// |   UART2   |  LPUART4  | OSC(24 MHz) |  D7/D8  : 2/2 |
+// |   UART3   |  LPUART2  | OSC(24 MHz) | D15/D14 : 2/2 |
+// |   UART4   |  LPUART3  | OSC(24 MHz) | D16/D17 : 2/2 |
+// |   UART5   |  LPUART8  | OSC(24 MHz) | D21/D20 : 2/2 |
+// |   UART6   |  LPUART1  | OSC(24 MHz) | D25/D24 : 2/2 |
+// |   UART7   |  LPUART7  | OSC(24 MHz) | D28/D29 : 2/2 |
+// #===========#===========#=============#=========-=====#
+const (
+	UART1_RX_PIN = D0
+	UART1_TX_PIN = D1
+
+	UART2_RX_PIN = D7
+	UART2_TX_PIN = D8
+
+	UART3_RX_PIN = D15
+	UART3_TX_PIN = D14
+
+	UART4_RX_PIN = D16
+	UART4_TX_PIN = D17
+
+	UART5_RX_PIN = D21
+	UART5_TX_PIN = D20
+
+	UART6_RX_PIN = D25
+	UART6_TX_PIN = D24
+
+	UART7_RX_PIN = D28
+	UART7_TX_PIN = D29
+)
+
+var (
+	UART1 = UART{
+		Buffer: NewRingBuffer(),
+		Bus:    nxp.LPUART6,
+		muxRX: muxSelect{ // D0 (PA3 [AD_B0_03])
+			mux: nxp.IOMUXC_LPUART6_RX_SELECT_INPUT_DAISY_GPIO_AD_B0_03_ALT2,
+			sel: &nxp.IOMUXC.LPUART6_RX_SELECT_INPUT,
+		},
+		muxTX: muxSelect{ // D1 (PA2 [AD_B0_02])
+			mux: nxp.IOMUXC_LPUART6_TX_SELECT_INPUT_DAISY_GPIO_AD_B0_02_ALT2,
+			sel: &nxp.IOMUXC.LPUART6_TX_SELECT_INPUT,
+		},
+	}
+	UART2 = UART{
+		Buffer: NewRingBuffer(),
+		Bus:    nxp.LPUART4,
+		muxRX: muxSelect{ // D7 (PB17 [B1_01])
+			mux: nxp.IOMUXC_LPUART4_RX_SELECT_INPUT_DAISY_GPIO_B1_01_ALT2,
+			sel: &nxp.IOMUXC.LPUART4_RX_SELECT_INPUT,
+		},
+		muxTX: muxSelect{ // D8 (PB16 [B1_00])
+			mux: nxp.IOMUXC_LPUART4_TX_SELECT_INPUT_DAISY_GPIO_B1_00_ALT2,
+			sel: &nxp.IOMUXC.LPUART4_TX_SELECT_INPUT,
+		},
+	}
+	UART3 = UART{
+		Buffer: NewRingBuffer(),
+		Bus:    nxp.LPUART2,
+		muxRX: muxSelect{ // D15 (PA19 [AD_B1_03])
+			mux: nxp.IOMUXC_LPUART2_RX_SELECT_INPUT_DAISY_GPIO_AD_B1_03_ALT2,
+			sel: &nxp.IOMUXC.LPUART2_RX_SELECT_INPUT,
+		},
+		muxTX: muxSelect{ // D14 (PA18 [AD_B1_02])
+			mux: nxp.IOMUXC_LPUART2_TX_SELECT_INPUT_DAISY_GPIO_AD_B1_02_ALT2,
+			sel: &nxp.IOMUXC.LPUART2_TX_SELECT_INPUT,
+		},
+	}
+	UART4 = UART{
+		Buffer: NewRingBuffer(),
+		Bus:    nxp.LPUART3,
+		muxRX: muxSelect{ // D16 (PA23 [AD_B1_07])
+			mux: nxp.IOMUXC_LPUART3_RX_SELECT_INPUT_DAISY_GPIO_AD_B1_07_ALT2,
+			sel: &nxp.IOMUXC.LPUART3_RX_SELECT_INPUT,
+		},
+		muxTX: muxSelect{ // D17 (PA22 [AD_B1_06])
+			mux: nxp.IOMUXC_LPUART3_TX_SELECT_INPUT_DAISY_GPIO_AD_B1_06_ALT2,
+			sel: &nxp.IOMUXC.LPUART3_TX_SELECT_INPUT,
+		},
+	}
+	UART5 = UART{
+		Buffer: NewRingBuffer(),
+		Bus:    nxp.LPUART8,
+		muxRX: muxSelect{ // D21 (PA27 [AD_B1_11])
+			mux: nxp.IOMUXC_LPUART8_RX_SELECT_INPUT_DAISY_GPIO_AD_B1_11_ALT2,
+			sel: &nxp.IOMUXC.LPUART8_RX_SELECT_INPUT,
+		},
+		muxTX: muxSelect{ // D20 (PA26 [AD_B1_10])
+			mux: nxp.IOMUXC_LPUART8_TX_SELECT_INPUT_DAISY_GPIO_AD_B1_10_ALT2,
+			sel: &nxp.IOMUXC.LPUART8_TX_SELECT_INPUT,
+		},
+	}
+	UART6 = UART{
+		Buffer: NewRingBuffer(),
+		Bus:    nxp.LPUART1,
+		// LPUART1 not connected via IOMUXC
+		//   RX: D24 (PA12 [AD_B0_12])
+		//   TX: D25 (PA13 [AD_B0_13])
+	}
+	UART7 = UART{
+		Buffer: NewRingBuffer(),
+		Bus:    nxp.LPUART7,
+		muxRX: muxSelect{ // D28 (PC18 [EMC_32])
+			mux: nxp.IOMUXC_LPUART7_RX_SELECT_INPUT_DAISY_GPIO_EMC_32_ALT2,
+			sel: &nxp.IOMUXC.LPUART7_RX_SELECT_INPUT,
+		},
+		muxTX: muxSelect{ // D29 (PD31 [EMC_31])
+			mux: nxp.IOMUXC_LPUART7_TX_SELECT_INPUT_DAISY_GPIO_EMC_31_ALT2,
+			sel: &nxp.IOMUXC.LPUART7_TX_SELECT_INPUT,
+		},
+	}
+)
+
+// #===========#==========#===============#===========================#
+// | Interface | Hardware |  Clock(Freq)  | SDI/SDO/SCK/CS  :   Alt   |
+// #===========#==========#===============#=================-=========#
+// |   SPI1    |  LPSPI4  | PLL2(132 MHz) | D12/D11/D13/D10 : 3/3/3/3 |
+// |   SPI2    |  LPSPI3  | PLL2(132 MHz) |  D1/D26/D27/D0  : 7/2/2/7 |
+// |   SPI3    |  LPSPI1  | PLL2(132 MHz) | D34/D35/D37/D36 : 4/4/4/4 |
+// #===========#==========#===============#=================-=========#
+const (
+	SPI1_SDI_PIN = D12
+	SPI1_SDO_PIN = D11
+	SPI1_SCK_PIN = D13
+	SPI1_CS_PIN  = D10
+
+	SPI2_SDI_PIN = D1
+	SPI2_SDO_PIN = D26
+	SPI2_SCK_PIN = D27
+	SPI2_CS_PIN  = D0
+
+	SPI3_SDI_PIN = D34
+	SPI3_SDO_PIN = D35
+	SPI3_SCK_PIN = D37
+	SPI3_CS_PIN  = D36
+)
+
+// #====================================================#
+// |                         I2C                        |
+// #===========#==========#=============#===============#
+// | Interface | Hardware | Clock(Freq) | SDA/SCL : Alt |
+// #===========#==========#=============#=========-=====#
+// |   I2C1    |  LPI2C1  | OSC(24 MHz) | D18/D19 : 3/3 |
+// |   I2C2    |  LPI2C3  | OSC(24 MHz) | D17/D16 : 1/1 |
+// |   I2C3    |  LPI2C4  | OSC(24 MHz) | D25/D24 : 0/0 |
+// #===========#==========#=============#=========-=====#
+const (
+	I2C1_SDA_PIN = D18
+	I2C1_SCL_PIN = D19
+
+	I2C2_SDA_PIN = D17
+	I2C2_SCL_PIN = D16
+
+	I2C3_SDA_PIN = D25
+	I2C3_SCL_PIN = D24
+)
@@ -1,4 +1,4 @@
-// +build avr nrf sam stm32,!stm32f4disco fe310 k210
+// +build avr nrf sam stm32,!stm32f407,!stm32f7x2 fe310 k210

 package machine

@@ -136,6 +136,11 @@ var pinPadMapping = [32]byte{
 // found" (indicated by returning ok=false). The pad number is returned to
 // calculate the DOPO/DIPO bitfields of the various serial peripherals.
 func findPinPadMapping(sercom uint8, pin Pin) (pinMode PinMode, pad uint32, ok bool) {
+	if int(pin)/2 >= len(pinPadMapping) {
+		// This is probably NoPin, for which no mapping is available.
+		return
+	}
+
 	nibbles := pinPadMapping[pin/2]
 	upper := nibbles >> 4
 	lower := nibbles & 0xf
@@ -165,7 +170,8 @@ func findPinPadMapping(sercom uint8, pin Pin) (pinMode PinMode, pad uint32, ok b
 }

 // SetInterrupt sets an interrupt to be executed when a particular pin changes
-// state.
+// state. The pin should already be configured as an input, including a pull up
+// or down if no external pull is provided.
 //
 // This call will replace a previously set callback on this pin. You can pass a
 // nil func to unset the pin change interrupt. If you do so, the change
@@ -1188,7 +1194,10 @@ func (spi SPI) Configure(config SPIConfig) error {
 	}

 	// Set synch speed for SPI
-	baudRate := (CPUFrequency() / (2 * config.Frequency)) - 1
+	baudRate := CPUFrequency() / (2 * config.Frequency)
+	if baudRate > 0 {
+		baudRate--
+	}
 	spi.Bus.BAUD.Set(uint8(baudRate))

 	// Enable SPI port.
@@ -322,6 +322,11 @@ var pinPadMapping = [64]uint16{
 // (indicated by returning ok=false). The pad number is returned to calculate
 // the DOPO/DIPO bitfields of the various serial peripherals.
 func findPinPadMapping(sercom uint8, pin Pin) (pinMode PinMode, pad uint32, ok bool) {
+	if int(pin)/2 >= len(pinPadMapping) {
+		// This is probably NoPin, for which no mapping is available.
+		return
+	}
+
 	bytes := pinPadMapping[pin/2]
 	upper := byte(bytes >> 8)
 	lower := byte(bytes & 0xff)
@@ -353,7 +358,8 @@ func findPinPadMapping(sercom uint8, pin Pin) (pinMode PinMode, pad uint32, ok b
 }

 // SetInterrupt sets an interrupt to be executed when a particular pin changes
-// state.
+// state. The pin should already be configured as an input, including a pull up
+// or down if no external pull is provided.
 //
 // This call will replace a previously set callback on this pin. You can pass a
 // nil func to unset the pin change interrupt. If you do so, the change
@@ -1331,12 +1337,13 @@ func (spi SPI) Configure(config SPIConfig) error {

 	// Determine the input pinout (for SDI).
 	var dataInPinout uint32
-	SDIPinMode, SDIPad, ok := findPinPadMapping(spi.SERCOM, config.SDI)
+	var SDIPinMode PinMode
 	if config.SDI != NoPin {
+		var ok bool
+		SDIPinMode, dataInPinout, ok = findPinPadMapping(spi.SERCOM, config.SDI)
 		if !ok {
 			return ErrInvalidInputPin
 		}
-		dataInPinout = SDIPad // mapped directly
 	}

 	// Determine the output pinout (for SDO/SCK).
@@ -1417,6 +1424,9 @@ func (spi SPI) Configure(config SPIConfig) error {

 	// Set synch speed for SPI
 	baudRate := SERCOM_FREQ_REF / (2 * config.Frequency)
+	if baudRate > 0 {
+		baudRate--
+	}
 	spi.Bus.BAUD.Set(uint8(baudRate))

 	// Enable SPI port.
@@ -1580,14 +1590,14 @@ func (pwm PWM) setPinCfg(val uint8) {
 // setChannel sets the value for the correct channel for PWM on this pin.
 func (pwm PWM) setChannel(timer *sam.TCC_Type, val uint32) {
 	switch pwm.Pin {
-	case PA16:
-		timer.CC[0].Set(val)
-	case PA17:
-		timer.CC[1].Set(val)
 	case PA14:
 		timer.CC[0].Set(val)
 	case PA15:
 		timer.CC[1].Set(val)
+	case PA16:
+		timer.CC[0].Set(val)
+	case PA17:
+		timer.CC[1].Set(val)
 	case PA18:
 		timer.CC[2].Set(val)
 	case PA19:
@@ -1596,10 +1606,22 @@ func (pwm PWM) setChannel(timer *sam.TCC_Type, val uint32) {
 		timer.CC[0].Set(val)
 	case PA21:
 		timer.CC[1].Set(val)
-	case PA23:
-		timer.CC[3].Set(val)
 	case PA22:
 		timer.CC[2].Set(val)
+	case PA23:
+		timer.CC[3].Set(val)
+	case PB12:
+		timer.CC[0].Set(val)
+	case PB13:
+		timer.CC[1].Set(val)
+	case PB14:
+		timer.CC[0].Set(val)
+	case PB15:
+		timer.CC[1].Set(val)
+	case PB16:
+		timer.CC[4].Set(val)
+	case PB17:
+		timer.CC[5].Set(val)
 	case PB31:
 		timer.CC[1].Set(val)
 	default:
@@ -1610,14 +1632,14 @@ func (pwm PWM) setChannel(timer *sam.TCC_Type, val uint32) {
 // setChannelBuffer sets the value for the correct channel buffer for PWM on this pin
 func (pwm PWM) setChannelBuffer(timer *sam.TCC_Type, val uint32) {
 	switch pwm.Pin {
-	case PA16:
-		timer.CCBUF[0].Set(val)
-	case PA17:
-		timer.CCBUF[1].Set(val)
 	case PA14:
 		timer.CCBUF[0].Set(val)
 	case PA15:
 		timer.CCBUF[1].Set(val)
+	case PA16:
+		timer.CCBUF[0].Set(val)
+	case PA17:
+		timer.CCBUF[1].Set(val)
 	case PA18:
 		timer.CCBUF[2].Set(val)
 	case PA19:
@@ -1626,10 +1648,22 @@ func (pwm PWM) setChannelBuffer(timer *sam.TCC_Type, val uint32) {
 		timer.CCBUF[0].Set(val)
 	case PA21:
 		timer.CCBUF[1].Set(val)
-	case PA23:
-		timer.CCBUF[3].Set(val)
 	case PA22:
 		timer.CCBUF[2].Set(val)
+	case PA23:
+		timer.CCBUF[3].Set(val)
+	case PB12:
+		timer.CCBUF[0].Set(val)
+	case PB13:
+		timer.CCBUF[1].Set(val)
+	case PB14:
+		timer.CCBUF[0].Set(val)
+	case PB15:
+		timer.CCBUF[1].Set(val)
+	case PB16:
+		timer.CCBUF[4].Set(val)
+	case PB17:
+		timer.CCBUF[5].Set(val)
 	case PB31:
 		timer.CCBUF[1].Set(val)
 	default:
@@ -1640,14 +1674,14 @@ func (pwm PWM) setChannelBuffer(timer *sam.TCC_Type, val uint32) {
 // getMux returns the pin mode mux to be used for PWM on this pin.
 func (pwm PWM) getMux() PinMode {
 	switch pwm.Pin {
-	case PA16:
-		return PinPWMF
-	case PA17:
-		return PinPWMF
 	case PA14:
 		return PinPWMF
 	case PA15:
 		return PinPWMF
+	case PA16:
+		return PinPWMF
+	case PA17:
+		return PinPWMF
 	case PA18:
 		return PinPWMF
 	case PA19:
@@ -1656,9 +1690,21 @@ func (pwm PWM) getMux() PinMode {
 		return PinPWMG
 	case PA21:
 		return PinPWMG
+	case PA22:
+		return PinPWMG
 	case PA23:
 		return PinPWMG
-	case PA22:
+	case PB12:
+		return PinPWMF
+	case PB13:
+		return PinPWMF
+	case PB14:
+		return PinPWMF
+	case PB15:
+		return PinPWMF
+	case PB16:
+		return PinPWMG
+	case PB17:
 		return PinPWMG
 	case PB31:
 		return PinPWMF
--- a/Show More
+++ b/Show More