transform (coroutines): move any misplaced entry-block allocas to the start of the entry block before lowering

transform (coroutines): fix memory corruption for tail calls that reference stack allocations
This change fixes a bug in which `alloca` memory lifetimes would not extend past the suspend of an asynchronous tail call. This would typically manifest as memory corruption, and could happen with or without normal suspending calls within the function.
2026-07-29 16:18:41 +00:00 · 2021-09-21 20:08:30 +02:00 · 2021-09-21 20:08:30 +02:00 · 2021-09-21 18:05:21 +02:00 · 2021-09-17 22:22:27 +02:00 · 2021-09-16 20:13:04 +02:00
471 changed files with 16486 additions and 5866 deletions
@@ -157,7 +157,7 @@ commands:
      - llvm-source-linux
      - restore_cache:
          keys:
-            - llvm-build-11-linux-v3-assert
+            - llvm-build-11-linux-v4-assert
      - run:
          name: "Build LLVM"
          command: |
@@ -176,7 +176,7 @@ commands:
              find llvm-build -name CMakeFiles -prune -exec rm -r '{}' \;
            fi
      - save_cache:
-          key: llvm-build-11-linux-v3-assert
+          key: llvm-build-11-linux-v4-assert
          paths:
            llvm-build
      - run: make ASSERT=1
@@ -225,7 +225,7 @@ commands:
      - llvm-source-linux
      - restore_cache:
          keys:
-            - llvm-build-11-linux-v3-noassert
+            - llvm-build-11-linux-v4-noassert
      - run:
          name: "Build LLVM"
          command: |
@@ -244,7 +244,7 @@ commands:
              find llvm-build -name CMakeFiles -prune -exec rm -r '{}' \;
            fi
      - save_cache:
-          key: llvm-build-11-linux-v3-noassert
+          key: llvm-build-11-linux-v4-noassert
          paths:
            llvm-build
      - build-wasi-libc
@@ -286,24 +286,24 @@ commands:
      - run:
          name: "Install dependencies"
          command: |
-            curl https://dl.google.com/go/go1.16.darwin-amd64.tar.gz -o go1.16.darwin-amd64.tar.gz
-            sudo tar -C /usr/local -xzf go1.16.darwin-amd64.tar.gz
+            curl https://dl.google.com/go/go1.17.darwin-amd64.tar.gz -o go1.17.darwin-amd64.tar.gz
+            sudo tar -C /usr/local -xzf go1.17.darwin-amd64.tar.gz
            ln -s /usr/local/go/bin/go /usr/local/bin/go
            HOMEBREW_NO_AUTO_UPDATE=1 brew install qemu
      - install-xtensa-toolchain:
          variant: "macos"
      - restore_cache:
          keys:
-            - go-cache-macos-v2-{{ checksum "go.mod" }}-{{ .Environment.CIRCLE_PREVIOUS_BUILD_NUM }}
-            - go-cache-macos-v2-{{ checksum "go.mod" }}
+            - go-cache-macos-v3-{{ checksum "go.mod" }}-{{ .Environment.CIRCLE_PREVIOUS_BUILD_NUM }}
+            - go-cache-macos-v3-{{ checksum "go.mod" }}
      - restore_cache:
          keys:
-            - llvm-source-11-macos-v2
+            - llvm-source-11-macos-v3
      - run:
          name: "Fetch LLVM source"
          command: make llvm-source
      - save_cache:
-          key: llvm-source-11-macos-v2
+          key: llvm-source-11-macos-v3
          paths:
            - llvm-project/clang/lib/Headers
            - llvm-project/clang/include
@@ -311,7 +311,7 @@ commands:
            - llvm-project/llvm/include
      - restore_cache:
          keys:
-            - llvm-build-11-macos-v3
+            - llvm-build-11-macos-v5
      - run:
          name: "Build LLVM"
          command: |
@@ -327,17 +327,17 @@ commands:
              find llvm-build -name CMakeFiles -prune -exec rm -r '{}' \;
            fi
      - save_cache:
-          key: llvm-build-11-macos-v3
+          key: llvm-build-11-macos-v5
          paths:
            llvm-build
      - restore_cache:
          keys:
-            - wasi-libc-sysroot-macos-v3
+            - wasi-libc-sysroot-macos-v4
      - run:
          name: "Build wasi-libc"
          command: make wasi-libc
      - save_cache:
-          key: wasi-libc-sysroot-macos-v3
+          key: wasi-libc-sysroot-macos-v4
          paths:
            - lib/wasi-libc/sysroot
      - run:
@@ -359,30 +359,18 @@ commands:
            tinygo version
      - run: make smoketest AVR=0
      - save_cache:
-          key: go-cache-macos-v2-{{ checksum "go.mod" }}-{{ .Environment.CIRCLE_BUILD_NUM }}
+          key: go-cache-macos-v3-{{ checksum "go.mod" }}-{{ .Environment.CIRCLE_BUILD_NUM }}
          paths:
            - ~/.cache/go-build
            - /go/pkg/mod

 jobs:
-  test-llvm10-go113:
-    docker:
-      - image: circleci/golang:1.13-buster
-    steps:
-      - test-linux:
-          llvm: "10"
-  test-llvm10-go114:
-    docker:
-      - image: circleci/golang:1.14-buster
-    steps:
-      - test-linux:
-          llvm: "10"
-  test-llvm11-go115:
+  test-llvm10-go115:
    docker:
      - image: circleci/golang:1.15-buster
    steps:
      - test-linux:
-          llvm: "11"
+          llvm: "10"
  test-llvm11-go116:
    docker:
      - image: circleci/golang:1.16-buster
@@ -391,17 +379,17 @@ jobs:
          llvm: "11"
  assert-test-linux:
    docker:
-      - image: circleci/golang:1.14-stretch
+      - image: circleci/golang:1.17-stretch
    steps:
      - assert-test-linux
  build-linux:
    docker:
-      - image: circleci/golang:1.14-stretch
+      - image: circleci/golang:1.17-stretch
    steps:
      - build-linux
  build-macos:
    macos:
-      xcode: "10.1.0"
+      xcode: "11.1.0" # macOS 10.14
    steps:
      - build-macos

@@ -410,9 +398,7 @@ jobs:
 workflows:
  test-all:
    jobs:
-      - test-llvm10-go113
-      - test-llvm10-go114
-      - test-llvm11-go115
+      - test-llvm10-go115
      - test-llvm11-go116
      - build-linux
      - build-macos
@@ -0,0 +1,3 @@
+build/
+llvm-*/
+
@@ -0,0 +1,45 @@
+name: CI for tinygo-dev docker container
+on:
+  push:
+    branches: [ dev, fix-docker-llvm-build ]
+
+jobs:
+  push_to_registry:
+    name: Push Docker image to GHCR/Docker Hub
+    runs-on: ubuntu-latest
+    permissions:
+      packages: write
+      contents: read
+    steps:
+      - name: Check out the repo
+        uses: actions/checkout@v2
+        with:
+          submodules: recursive
+      - name: Docker meta
+        id: meta
+        uses: docker/metadata-action@v3
+        with:
+          images: |
+            tinygo/tinygo-dev
+            ghcr.io/${{ github.repository }}/tinygo-dev
+          tags: |
+            type=sha,format=long
+            type=raw,value=latest
+      - name: Log in to Docker Hub
+        uses: docker/login-action@v1
+        with:
+          username: ${{ secrets.DOCKER_HUB_USERNAME }}
+          password: ${{ secrets.DOCKER_HUB_ACCESS_TOKEN }}
+      - name: Log in to Github Container Registry
+        uses: docker/login-action@v1
+        with:
+          registry: ghcr.io
+          username: ${{ github.actor }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+      - name: Build and push
+        uses: docker/build-push-action@v2
+        with:
+          context: .
+          push: true
+          tags: ${{ steps.meta.outputs.tags }}
+          labels: ${{ steps.meta.outputs.labels }}
@@ -16,6 +16,8 @@ src/device/stm32/*.go
 src/device/stm32/*.s
 src/device/kendryte/*.go
 src/device/kendryte/*.s
+src/device/rp/*.go
+src/device/rp/*.s
 vendor
 llvm-build
 llvm-project
@@ -23,7 +23,7 @@ different guide:
 LLVM, Clang and LLD are quite light on dependencies, requiring only standard
 build tools to be built. Go is of course necessary to build TinyGo itself.

-  * Go (1.13+)
+  * Go (1.15+)
  * Standard build tools (gcc/clang)
  * git
  * CMake
@@ -1,3 +1,261 @@
+0.20.0
+---
+
+* **command line**
+  - add support for Go 1.17
+  - improve Go version detection
+  - add support for the Black Magic Probe (BMP)
+  - add a flag for creating cpu profiles
+* **compiler**
+  - `builder:` list libraries at the end of the linker command
+  - `builder:` strip debug information at link time instead of at compile time
+  - `builder:` add missing error check for `ioutil.TempFile()`
+  - `builder:` simplify running of jobs
+  - `compiler:` move LLVM math builtin support into the compiler
+  - `compiler:` move math aliases from the runtime to the compiler
+  - `compiler:` add aliases for many hashing packages
+  - `compiler:` add `*ssa.MakeSlice` bounds tests
+  - `compiler:` fix max possible slice
+  - `compiler:` add support for new language features of Go 1.17
+  - `compiler:` fix equally named structs in different scopes
+  - `compiler:` avoid zero-sized alloca in channel operations
+  - `interp:` don't ignore array indices for untyped objects
+  - `interp:` keep reverted package initializers in order
+  - `interp:` fix bug in compiler-time/run-time package initializers
+  - `loader:` fix panic in CGo files with syntax errors
+  - `transform:` improve GC stack slot pass to work around a bug
+* **standard library**
+  - `crypto/rand`: switch to `arc4random_buf`
+  - `math:` fix `math.Max` and `math.Min`
+  - `math/big`: fix undefined symbols error
+  - `net:` add MAC address implementation
+  - `os:` implement `os.Executable`
+  - `os:` add `SEEK_SET`, `SEEK_CUR`, and `SEEK_END`
+  - `reflect:` add StructField.IsExported method
+  - `runtime:` reset heapptr to heapStart after preinit()
+  - `runtime:` add `subsections_via_symbols` to assembly files on darwin
+  - `testing:` add subset implementation of Benchmark
+  - `testing:` test testing package using `tinygo test`
+  - `testing:` add support for the `-test.v` flag
+* **targets**
+  - `386:` bump minimum requirement to the Pentium 4
+  - `arm:` switch to Thumb instruction set on ARM
+  - `atsamd:` fix copy-paste error for atsamd21/51 calibTrim block
+  - `baremetal`,`wasm`: support command line params and environment variables
+  - `cortexm:` fix stack overflow because of unaligned stacks
+  - `esp32c3:` add support for the ESP32-C3 from Espressif
+  - `nrf52840:` fix ram size
+  - `nxpmk66f18:` fix a suspicious bitwise operation
+  - `rp2040:` add SPI support
+  - `rp2040:` add I2C support
+  - `rp2040:` add PWM implementation
+  - `rp2040:` add openocd configuration
+  - `stm32:` add support for PortMask* functions for WS2812 support
+  - `unix:` fix time base for time.Now()
+  - `unix:` check for mmap error and act accordingly
+  - `wasm:` override dlmalloc heap implementation from wasi-libc
+  - `wasm:` align heap to 16 bytes
+  - `wasm:` add support for the crypto/rand package
+* **boards**
+  - add `DefaultUART` to adafruit boards
+  - `arduino-mkrwifi1010:` add board definition for Arduino MKR WiFi 1010
+  - `arduino-mkrwifi1010:` fix pin definition of `NINA_RESETN`
+  - `feather-nrf52:` fix pin definition of uart
+  - `feather-rp2040:` add pin name definition
+  - `gameboy-advance:` fix ROM header
+  - `mdbt50qrx-uf2:` add Raytac MDBT50Q-RX Dongle with TinyUF2
+  - `nano-rp2040:` define `NINA_SPI` and fix wifinina pins
+  - `teensy40:` enable hardware UART reconfiguration, fix receive watermark interrupt
+
+
+0.19.0
+---
+
+* **command line**
+  - don't consider compile-only tests as failing
+  - add -test flag for `tinygo list`
+  - escape commands while printing them with the -x flag
+  - make flash-command portable and safer to use
+  - use `extended-remote` instead of `remote` in GDB
+  - detect specific serial port IDs based on USB vid/pid
+  - add a flag to the command line to select the serial implementation
+* **compiler**
+  - `cgo`: improve constant parser
+  - `compiler`: support chained interrupt handlers
+  - `compiler`: add support for running a builtin in a goroutine
+  - `compiler`: do not emit nil checks for loading closure variables
+  - `compiler`: skip context parameter when starting regular goroutine
+  - `compiler`: refactor method names
+  - `compiler`: add function and global section pragmas
+  - `compiler`: implement `syscall.rawSyscallNoError` in inline assembly
+  - `interp`: ignore inline assembly in markExternal
+  - `interp`: fix a bug in pointer cast workaround
+  - `loader`: fix testing a main package
+* **standard library**
+  - `crypto/rand`: replace this package with a TinyGo version
+  - `machine`: make USBCDC global a pointer
+  - `machine`: make UART objects pointer receivers
+  - `machine`: define Serial as the default output
+  - `net`: add initial support for net.IP
+  - `net`: add more net compatibility
+  - `os`: add stub for os.ReadDir
+  - `os`: add FileMode constants from Go 1.16
+  - `os`: add stubs required for net/http
+  - `os`: implement process related functions
+  - `reflect`: implement AppendSlice
+  - `reflect`: add stubs required for net/http
+  - `runtime`: make task.Data a 64-bit integer to avoid overflow
+  - `runtime`: expose memory stats
+  - `sync`: implement NewCond
+  - `syscall`: fix int type in libc version
+* **targets**
+  - `cortexm`: do not disable interrupts on abort
+  - `cortexm`: bump default stack size to 2048 bytes
+  - `nrf`: avoid heap allocation in waitForEvent
+  - `nrf`: don't trigger a heap allocation in SPI.Transfer
+  - `nrf52840`: add support for flashing with the BOSSA tool
+  - `rp2040`: add support for GPIO input
+  - `rp2040`: add basic support for ADC
+  - `rp2040`: gpio and adc pin definitions
+  - `rp2040`: implement UART
+  - `rp2040`: patch elf to checksum 2nd stage boot
+  - `stm32`: add PWM for most chips
+  - `stm32`: add support for pin interrupts
+  - `stm32f103`: add support for PinInputPullup / PinInputPulldown
+  - `wasi`: remove wasm build tag
+* **boards**
+  - `feather-rp2040`: add support for this board
+  - `feather-nrf52840-sense`: add board definition for this board
+  - `pca10059`: support flashing from Windows
+  - `nano-rp2040`: add this board
+  - `nano-33-ble`: add support for this board
+  - `pico`: add the Raspberry Pi Pico board with the new RP2040 chip
+  - `qtpy`: add pin for neopixels
+  - all: add definition for ws2812 for supported boards
+
+
+0.18.0
+---
+
+* **command line**
+  - drop support for Go 1.11 and 1.12
+  - throw an error when no target is specified on Windows
+  - improve error messages in `getDefaultPort()`, support for multiple ports
+  - remove `-cflags` and `-ldflags` flags
+  - implement `-ldflags="-X ..."`
+  - add `-print-allocs` flag that lets you print all heap allocations
+  - openocd commands in tinygo command line
+  - add `-llvm-features` parameter
+  - match `go test` output
+  - discover USB ports only, this will ignore f.ex. bluetooth
+  - use physicmal path instead of cached GOROOT in function getGoroot
+  - add goroot for snap installs
+* **compiler**
+  - `builder`: add support for `-opt=0`
+  - `builder`, `compiler`: compile and cache packages in parallel
+  - `builder`: run interp per package
+  - `builder`: cache C and assembly file outputs
+  - `builder`: add support for `-x` flag to print commands
+  - `builder`: add optsize attribute while building the package
+  - `builder`: run function passes per package
+  - `builder`: hard code Clang compiler
+  - `compiler`: do not use `llvm.GlobalContext()`
+  - `compiler`: remove SimpleDCE pass
+  - `compiler`: do not emit nil checks for `*ssa.Alloc` instructions
+  - `compiler`: merge `runtime.typecodeID` and runtime.typeInInterface
+  - `compiler`: do not check for impossible type asserts
+  - `compiler`: fix use of global context: `llvm.Int32Type()`
+  - `compiler`: add interface IR test
+  - `compiler`: fix lack of method name in interface matching
+  - `compiler`: fix "fragment covers entire variable" bug
+  - `compiler`: optimize string literals and globals
+  - `compiler`: decouple func lowering from interface type codes
+  - `compiler`: add function attributes to some runtime calls
+  - `compiler`: improve position information in error messages
+  - `cgo`: add support for CFLAGS in .c files
+  - `interp`: support GEP on fixed (MMIO) addresses
+  - `interp`: handle `(reflect.Type).Elem()`
+  - `interp`: add support for runtime.interfaceMethod
+  - `interp`: make toLLVMValue return an error instead of panicking
+  - `interp`: add support for switch statement
+  - `interp`: fix phi instruction
+  - `interp`: remove map support
+  - `interp`: support extractvalue/insertvalue with multiple operands
+  - `transform`: optimize string comparisons against ""
+  - `transform`: optimize `reflect.Type` `Implements()` method
+  - `transform`: fix bug in interface lowering when signatures are renamed
+  - `transform`: don't rely on struct name of `runtime.typecodeID`
+  - `transform`: use IPSCCP pass instead of the constant propagation pass
+  - `transform`: fix func lowering assertion failure
+  - `transform`: do not lower zero-sized alloc to alloca
+  - `transform`: split interface and reflect lowering
+* **standard library**
+  - `runtime`: add dummy debug package
+  - `machine`: fix data shift/mask in newUSBSetup
+  - `machine`: make `machine.I2C0` and similar objects pointers
+  - `machine`: unify usbcdc code
+  - `machine`: refactor PWM support
+  - `machine`: avoid heap allocations in USB code
+  - `reflect`: let `reflect.Type` be of interface type
+  - `reflect`: implement a number of stub functions
+  - `reflect`: check for access in the `Interface` method call
+  - `reflect`: fix `AssignableTo` and `Implements` methods
+  - `reflect`: implement `Value.CanAddr`
+  - `reflect`: implement `Sizeof` and `Alignof` for func values
+  - `reflect`: implement `New` function
+  - `runtime`: implement command line arguments in hosted environments
+  - `runtime`: implement environment variables for Linux
+  - `runtime`: improve timers on nrf, and samd chips
+* **targets**
+  - all: use -Qunused-arguments only for assembly files
+  - `atmega1280`: add PWM support
+  - `attiny`: remove dummy UART
+  - `atsamd21`: improve SPI
+  - `atsamd51`: fix PWM support in atsamd51p20
+  - `atsamd5x`: improve SPI
+  - `atsamd51`, `atsame5x`: unify samd51 and same5x
+  - `atsamd51`, `atsamd21`: fix `ADC.Get()` value at 8bit and 10bit
+  - `atsame5x`: add support for CAN
+  - `avr`: remove I2C stubs from attiny support
+  - `cortexm`: check for `arm-none-eabi-gdb` and `gdb-multiarch` commands
+  - `cortexm`: add `__isr_vector` symbol
+  - `cortexm`: disable FPU on Cortex-M4
+  - `cortexm`: clean up Cortex-M target files
+  - `fe310`: fix SPI read
+  - `gameboy-advance`: Fix RGBA color interpretation
+  - `nrf52833`: add PWM support
+  - `stm32l0`: use unified UART logic
+  - `stm32`: move f103 (bluepill) to common i2c code
+  - `stm32`: separate altfunc selection for UART Tx/Rx
+  - `stm32`: i2c implementation for F7, L5 and L4 MCUs
+  - `stm32`: make SPI CLK fast to fix data issue
+  - `stm32`: support SPI on L4 series
+  - `unix`: avoid possible heap allocation with `-opt=0`
+  - `unix`: use conservative GC by default
+  - `unix`: use the tasks scheduler instead of coroutines
+  - `wasi`: upgrade WASI version to wasi_snapshot_preview1
+  - `wasi`: darwin: support basic file io based on libc
+  - `wasm`: only export explicitly exported functions
+  - `wasm`: use WASI ABI for exit function
+  - `wasm`: scan globals conservatively
+* **boards**
+  - `arduino-mega1280`: add support for the Arduino Mega 1280
+  - `arduino-nano-new`: Add Arduino Nano w/ New Bootloader target
+  - `atsame54-xpro`: add initial support this board
+  - `feather-m4-can`: add initial support for this board
+  - `grandcentral-m4`: add board support for Adafruit Grand Central M4 (SAMD51)
+  - `lgt92`: update to new UART structure
+  - `microbit`: remove LED constant
+  - `microbit-v2`: add support for S113 SoftDevice
+  - `nucleol432`: add support for this board
+  - `nucleo-l031k6`: add this board
+  - `pca10059`: initial support for this board
+  - `qtpy`: fix msd-volume-name
+  - `qtpy`: fix i2c setting
+  - `teensy40`: move txBuffer allocation to UART declaration
+  - `teensy40`: add UART0 as alias for UART1
+
+
 0.17.0

 ---
@@ -1,8 +1,8 @@
-# TinyGo base stage installs the most recent Go 1.15.x, LLVM 11 and the TinyGo compiler itself.
-FROM golang:1.15 AS tinygo-base
+# TinyGo base stage installs the most recent Go 1.17.x, LLVM 11 and the TinyGo compiler itself.
+FROM golang:1.17 AS tinygo-base

 RUN wget -O- https://apt.llvm.org/llvm-snapshot.gpg.key| apt-key add - && \
-    echo "deb http://apt.llvm.org/buster/ llvm-toolchain-buster-11 main" >> /etc/apt/sources.list && \
+    echo "deb http://apt.llvm.org/bullseye/ llvm-toolchain-bullseye-11 main" >> /etc/apt/sources.list && \
    apt-get update && \
    apt-get install -y llvm-11-dev libclang-11-dev lld-11 git

@@ -36,7 +36,7 @@ else
    LLVM_OPTION += '-DLLVM_ENABLE_ASSERTIONS=OFF'
 endif

-.PHONY: all tinygo test $(LLVM_BUILDDIR) llvm-source clean fmt gen-device gen-device-nrf gen-device-nxp gen-device-avr
+.PHONY: all tinygo test $(LLVM_BUILDDIR) llvm-source clean fmt gen-device gen-device-nrf gen-device-nxp gen-device-avr gen-device-rp

 LLVM_COMPONENTS = all-targets analysis asmparser asmprinter bitreader bitwriter codegen core coroutines coverage debuginfodwarf executionengine frontendopenmp instrumentation interpreter ipo irreader linker lto mc mcjit objcarcopts option profiledata scalaropts support target

@@ -71,7 +71,7 @@ else
 endif

 # Libraries that should be linked in for the statically linked Clang.
-CLANG_LIB_NAMES = clangAnalysis clangARCMigrate clangAST clangASTMatchers clangBasic clangCodeGen clangCrossTU clangDriver clangDynamicASTMatchers clangEdit clangFormat clangFrontend clangFrontendTool clangHandleCXX clangHandleLLVM clangIndex clangLex clangParse clangRewrite clangRewriteFrontend clangSema clangSerialization clangStaticAnalyzerCheckers clangStaticAnalyzerCore clangStaticAnalyzerFrontend clangTooling clangToolingASTDiff clangToolingCore clangToolingInclusions
+CLANG_LIB_NAMES = clangAnalysis clangAST clangASTMatchers clangBasic clangCodeGen clangCrossTU clangDriver clangDynamicASTMatchers clangEdit clangFormat clangFrontend clangFrontendTool clangHandleCXX clangHandleLLVM clangIndex clangLex clangParse clangRewrite clangRewriteFrontend clangSema clangSerialization clangTooling clangToolingASTDiff clangToolingCore clangToolingInclusions
 CLANG_LIBS = $(START_GROUP) $(addprefix -l,$(CLANG_LIB_NAMES)) $(END_GROUP) -lstdc++

 # Libraries that should be linked in for the statically linked LLD.
@@ -100,14 +100,14 @@ endif
 clean:
 	@rm -rf build

-FMT_PATHS = ./*.go builder cgo compiler interp loader src/device/arm src/examples src/machine src/os src/reflect src/runtime src/sync src/syscall src/internal/reflectlite transform
+FMT_PATHS = ./*.go builder cgo compiler interp loader src/device/arm src/examples src/machine src/os src/reflect src/runtime src/sync src/syscall src/testing src/internal/reflectlite transform
 fmt:
 	@gofmt -l -w $(FMT_PATHS)
 fmt-check:
 	@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-esp gen-device-nrf gen-device-sam gen-device-sifive 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-kendryte gen-device-nxp gen-device-rp
 ifneq ($(STM32), 0)
 gen-device: gen-device-stm32
 endif
@@ -124,6 +124,7 @@ build/gen-device-svd: ./tools/gen-device-svd/*.go

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

 gen-device-nrf: build/gen-device-svd
@@ -150,6 +151,9 @@ gen-device-stm32: build/gen-device-svd
 	./build/gen-device-svd -source=https://github.com/tinygo-org/stm32-svd lib/stm32-svd/svd src/device/stm32/
 	GO111MODULE=off $(GO) fmt ./src/device/stm32

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

 # Get LLVM sources.
 $(LLVM_PROJECTDIR)/llvm:
@@ -159,7 +163,7 @@ llvm-source: $(LLVM_PROJECTDIR)/llvm
 # Configure LLVM.
 TINYGO_SOURCE_DIR=$(shell pwd)
 $(LLVM_BUILDDIR)/build.ninja: llvm-source
-	mkdir -p $(LLVM_BUILDDIR); cd $(LLVM_BUILDDIR); cmake -G Ninja $(TINYGO_SOURCE_DIR)/$(LLVM_PROJECTDIR)/llvm "-DLLVM_TARGETS_TO_BUILD=X86;ARM;AArch64;RISCV;WebAssembly" "-DLLVM_EXPERIMENTAL_TARGETS_TO_BUILD=AVR;Xtensa" -DCMAKE_BUILD_TYPE=Release -DLIBCLANG_BUILD_STATIC=ON -DLLVM_ENABLE_TERMINFO=OFF -DLLVM_ENABLE_ZLIB=OFF -DLLVM_ENABLE_LIBEDIT=OFF -DLLVM_ENABLE_Z3_SOLVER=OFF -DLLVM_ENABLE_OCAMLDOC=OFF -DLLVM_ENABLE_LIBXML2=OFF -DLLVM_ENABLE_PROJECTS="clang;lld" -DLLVM_TOOL_CLANG_TOOLS_EXTRA_BUILD=OFF $(LLVM_OPTION)
+	mkdir -p $(LLVM_BUILDDIR); cd $(LLVM_BUILDDIR); cmake -G Ninja $(TINYGO_SOURCE_DIR)/$(LLVM_PROJECTDIR)/llvm "-DLLVM_TARGETS_TO_BUILD=X86;ARM;AArch64;RISCV;WebAssembly" "-DLLVM_EXPERIMENTAL_TARGETS_TO_BUILD=AVR;Xtensa" -DCMAKE_BUILD_TYPE=Release -DLIBCLANG_BUILD_STATIC=ON -DLLVM_ENABLE_TERMINFO=OFF -DLLVM_ENABLE_ZLIB=OFF -DLLVM_ENABLE_LIBEDIT=OFF -DLLVM_ENABLE_Z3_SOLVER=OFF -DLLVM_ENABLE_OCAMLDOC=OFF -DLLVM_ENABLE_LIBXML2=OFF -DLLVM_ENABLE_PROJECTS="clang;lld" -DLLVM_TOOL_CLANG_TOOLS_EXTRA_BUILD=OFF -DCLANG_ENABLE_STATIC_ANALYZER=OFF -DCLANG_ENABLE_ARCMT=OFF $(LLVM_OPTION)

 # Build LLVM.
 $(LLVM_BUILDDIR): $(LLVM_BUILDDIR)/build.ninja
@@ -182,25 +186,33 @@ tinygo:
 test: wasi-libc
 	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GO) test -v -buildmode exe -tags byollvm ./builder ./cgo ./compileopts ./compiler ./interp ./transform .

+TEST_PACKAGES = \
+	container/heap \
+	container/list \
+	container/ring \
+	crypto/des \
+	crypto/md5 \
+	crypto/sha1 \
+	crypto/sha256 \
+	crypto/sha512 \
+	encoding \
+	encoding/ascii85 \
+	encoding/base32 \
+	encoding/hex \
+	hash/adler32 \
+	hash/fnv \
+	hash/crc64 \
+	math \
+	math/cmplx \
+	testing \
+	text/scanner \
+	unicode/utf8 \
+
 # Test known-working standard library packages.
-# TODO: do this in one command, parallelize, and only show failing tests (no
-# implied -v flag).
+# TODO: parallelize, and only show failing tests (no implied -v flag).
 .PHONY: tinygo-test
 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
+	$(TINYGO) test $(TEST_PACKAGES)

 .PHONY: smoketest
 smoketest:
@@ -224,6 +236,8 @@ smoketest:
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=pca10040            examples/mcp3008
 	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pca10040            examples/memstats
+	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=microbit            examples/microbit-blink
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=pca10040            examples/pininterrupt
@@ -328,9 +342,11 @@ smoketest:
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=feather-nrf52840  	examples/blinky1
 	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=feather-nrf52840-sense examples/blinky1
+	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=itsybitsy-nrf52840  examples/blinky1
 	@$(MD5SUM) test.hex
-	$(TINYGO) build -size short -o test.hex -target=qtpy  				examples/blinky1
+	$(TINYGO) build -size short -o test.hex -target=qtpy                examples/serial
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=teensy40            examples/blinky1
 	@$(MD5SUM) test.hex
@@ -340,8 +356,24 @@ smoketest:
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=atsame54-xpro       examples/blinky1
 	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=atsame54-xpro       examples/can
+	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=feather-m4-can      examples/blinky1
 	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=feather-m4-can      examples/caninterrupt
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=arduino-nano33      examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=arduino-mkrwifi1010 examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pico                examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=nano-33-ble         examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=nano-rp2040         examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=feather-rp2040 		examples/blinky1
+	@$(MD5SUM) test.hex
 	# test pwm
 	$(TINYGO) build -size short -o test.hex -target=itsybitsy-m0        examples/pwm
 	@$(MD5SUM) test.hex
@@ -372,6 +404,8 @@ ifneq ($(STM32), 0)
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=stm32f4disco-1      examples/blinky1
 	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=stm32f4disco-1      examples/pwm
+	@$(MD5SUM) test.hex
 endif
 ifneq ($(AVR), 0)
 	$(TINYGO) build -size short -o test.hex -target=atmega1284p         examples/serial
@@ -399,6 +433,8 @@ ifneq ($(XTENSA), 0)
 	$(TINYGO) build -size short -o test.bin -target=nodemcu             examples/blinky1
 	@$(MD5SUM) test.bin
 endif
+	$(TINYGO) build -size short -o test.bin -target=esp32c3           	examples/serial
+	@$(MD5SUM) test.bin
 	$(TINYGO) build -size short -o test.hex -target=hifive1b            examples/blinky1
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=hifive1-qemu        examples/serial
@@ -412,10 +448,15 @@ endif
 	@$(MD5SUM) test.hex
 	$(TINYGO) build -size short -o test.hex -target=pca10040 -opt=1     examples/blinky1
 	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pca10040 -serial=none examples/echo
+	@$(MD5SUM) test.hex
 	$(TINYGO) build             -o test.nro -target=nintendoswitch      examples/serial
 	@$(MD5SUM) test.nro
 	$(TINYGO) build -size short -o test.hex -target=pca10040 -opt=0     ./testdata/stdlib.go
 	@$(MD5SUM) test.hex
+ifneq ($(OS),Windows_NT)
+	$(TINYGO) build -o test.elf -gc=leaking -scheduler=none examples/serial
+endif


 wasmtest:
@@ -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 57 microcontroller boards are currently supported:
+The following 70 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)
@@ -52,7 +52,10 @@ The following 57 microcontroller boards are currently supported:
 * [Adafruit Feather M4](https://www.adafruit.com/product/3857)
 * [Adafruit Feather M4 CAN](https://www.adafruit.com/product/4759)
 * [Adafruit Feather nRF52840 Express](https://www.adafruit.com/product/4062)
+* [Adafruit Feather nRF52840 Sense](https://www.adafruit.com/product/4516)
+* [Adafruit Feather RP2040](https://www.adafruit.com/product/4884)
 * [Adafruit Feather STM32F405 Express](https://www.adafruit.com/product/4382)
+* [Adafruit Grand Central M4](https://www.adafruit.com/product/4064)
 * [Adafruit ItsyBitsy M0](https://www.adafruit.com/product/3727)
 * [Adafruit ItsyBitsy M4](https://www.adafruit.com/product/3800)
 * [Adafruit ItsyBitsy nRF52840](https://www.adafruit.com/product/4481)
@@ -63,10 +66,15 @@ The following 57 microcontroller boards are currently supported:
 * [Adafruit PyPortal](https://www.adafruit.com/product/4116)
 * [Adafruit QT Py](https://www.adafruit.com/product/4600)
 * [Adafruit Trinket M0](https://www.adafruit.com/product/3500)
+* [Arduino Mega 1280](https://www.arduino.cc/en/Main/arduinoBoardMega/)
 * [Arduino Mega 2560](https://store.arduino.cc/arduino-mega-2560-rev3)
 * [Arduino MKR1000](https://store.arduino.cc/arduino-mkr1000-wifi)
+* [Arduino MKR WiFi 1010](https://store.arduino.cc/usa/mkr-wifi-1010)
 * [Arduino Nano](https://store.arduino.cc/arduino-nano)
-* [Arduino Nano33 IoT](https://store.arduino.cc/nano-33-iot)
+* [Arduino Nano 33 BLE](https://store.arduino.cc/nano-33-ble)
+* [Arduino Nano 33 BLE Sense](https://store.arduino.cc/nano-33-ble-sense)
+* [Arduino Nano 33 IoT](https://store.arduino.cc/nano-33-iot)
+* [Arduino Nano RP2040 Connect](https://store.arduino.cc/nano-rp2040-connect)
 * [Arduino Uno](https://store.arduino.cc/arduino-uno-rev3)
 * [Arduino Zero](https://store.arduino.cc/usa/arduino-zero)
 * [BBC micro:bit](https://microbit.org/)
@@ -84,6 +92,7 @@ The following 57 microcontroller boards are currently supported:
 * [Nordic Semiconductor PCA10031](https://www.nordicsemi.com/eng/Products/nRF51-Dongle)
 * [Nordic Semiconductor PCA10040](https://www.nordicsemi.com/eng/Products/Bluetooth-low-energy/nRF52-DK)
 * [Nordic Semiconductor PCA10056](https://www.nordicsemi.com/Software-and-Tools/Development-Kits/nRF52840-DK)
+* [Nordic Semiconductor pca10059](https://www.nordicsemi.com/Software-and-tools/Development-Kits/nRF52840-Dongle)
 * [Particle Argon](https://docs.particle.io/datasheets/wi-fi/argon-datasheet/)
 * [Particle Boron](https://docs.particle.io/datasheets/cellular/boron-datasheet/)
 * [Particle Xenon](https://docs.particle.io/datasheets/discontinued/xenon-datasheet/)
@@ -92,12 +101,16 @@ The following 57 microcontroller boards are currently supported:
 * [PJRC Teensy 3.6](https://www.pjrc.com/store/teensy36.html)
 * [PJRC Teensy 4.0](https://www.pjrc.com/store/teensy40.html)
 * [ProductivityOpen P1AM-100](https://facts-engineering.github.io/modules/P1AM-100/P1AM-100.html)
+* [Raspberry Pi Pico](https://www.raspberrypi.org/products/raspberry-pi-pico/)
+* [Raytac MDBT50Q-RX Dongle (with TinyUF2 bootloader)](https://www.adafruit.com/product/5199)
 * [Seeed Wio Terminal](https://www.seeedstudio.com/Wio-Terminal-p-4509.html)
 * [Seeed Seeeduino XIAO](https://www.seeedstudio.com/Seeeduino-XIAO-Arduino-Microcontroller-SAMD21-Cortex-M0+-p-4426.html)
 * [Seeed Sipeed MAix BiT](https://www.seeedstudio.com/Sipeed-MAix-BiT-for-RISC-V-AI-IoT-p-2872.html)
 * [SiFIve HiFive1](https://www.sifive.com/boards/hifive1)
 * [ST Micro "Nucleo" F103RB](https://www.st.com/en/evaluation-tools/nucleo-f103rb.html)
 * [ST Micro "Nucleo" F722ZE](https://www.st.com/en/evaluation-tools/nucleo-f722ze.html)
+* [ST Micro "Nucleo" L031K6](https://www.st.com/ja/evaluation-tools/nucleo-l031k6.html)
+* [ST Micro "Nucleo" L432KC](https://www.st.com/ja/evaluation-tools/nucleo-l432kc.html)
 * [ST Micro "Nucleo" L552ZE](https://www.st.com/en/evaluation-tools/nucleo-l552ze-q.html)
 * [ST Micro STM32F103XX "Bluepill"](https://stm32-base.org/boards/STM32F103C8T6-Blue-Pill)
 * [ST Micro STM32F407 "Discovery"](https://www.st.com/en/evaluation-tools/stm32f4discovery.html)
@@ -12,7 +12,7 @@ jobs:
  steps:
    - task: GoTool@0
      inputs:
-        version: '1.16'
+        version: '1.17'
    - checkout: self
      fetchDepth: 1
    - task: Cache@2
@@ -28,7 +28,7 @@ jobs:
    - task: CacheBeta@0
      displayName: Cache LLVM build
      inputs:
-        key: llvm-build-11-windows-v4
+        key: llvm-build-11-windows-v5
        path: llvm-build
    - task: Bash@3
      displayName: Build LLVM
@@ -12,7 +12,9 @@ import (
 	"errors"
 	"fmt"
 	"go/types"
+	"hash/crc32"
 	"io/ioutil"
+	"math/bits"
 	"os"
 	"path/filepath"
 	"runtime"
@@ -39,6 +41,11 @@ type BuildResult struct {
 	// 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
+
+	// ImportPath is the import path of the main package. This is useful for
+	// correctly printing test results: the import path isn't always the same as
+	// the path listed on the command line.
+	ImportPath string
 }

 // packageAction is the struct that is serialized to JSON and hashed, to work as
@@ -93,7 +100,8 @@ func Build(pkgName, outpath string, config *compileopts.Config, action func(Buil
 		AutomaticStackSize: config.AutomaticStackSize(),
 		DefaultStackSize:   config.Target.DefaultStackSize,
 		NeedsStackObjects:  config.NeedsStackObjects(),
-		Debug:              config.Debug(),
+		Debug:              true,
+		LLVMFeatures:       config.LLVMFeatures(),
 	}

 	// Load the target machine, which is the LLVM object that contains all
@@ -116,11 +124,6 @@ func Build(pkgName, outpath string, config *compileopts.Config, action func(Buil
 		return err
 	}

-	// The slice of jobs that orchestrates most of the build.
-	// This is somewhat like an in-memory Makefile with each job being a
-	// Makefile target.
-	var jobs []*compileJob
-
 	// Create the *ssa.Program. This does not yet build the entire SSA of the
 	// program so it's pretty fast and doesn't need to be parallelized.
 	program := lprogram.LoadSSA()
@@ -303,7 +306,6 @@ func Build(pkgName, outpath string, config *compileopts.Config, action func(Buil
 				return os.Rename(f.Name(), bitcodePath)
 			},
 		}
-		jobs = append(jobs, job)
 		packageJobs = append(packageJobs, job)
 	}

@@ -394,14 +396,13 @@ func Build(pkgName, outpath string, config *compileopts.Config, action func(Buil
 			return nil
 		},
 	}
-	jobs = append(jobs, programJob)

 	// Check whether we only need to create an object file.
 	// If so, we don't need to link anything and will be finished quickly.
 	outext := filepath.Ext(outpath)
 	if outext == ".o" || outext == ".bc" || outext == ".ll" {
 		// Run jobs to produce the LLVM module.
-		err := runJobs(jobs)
+		err := runJobs(programJob)
 		if err != nil {
 			return err
 		}
@@ -442,7 +443,6 @@ func Build(pkgName, outpath string, config *compileopts.Config, action func(Buil
 			return ioutil.WriteFile(objfile, llvmBuf.Bytes(), 0666)
 		},
 	}
-	jobs = append(jobs, outputObjectFileJob)

 	// Prepare link command.
 	linkerDependencies := []*compileJob{outputObjectFileJob}
@@ -457,38 +457,13 @@ func Build(pkgName, outpath string, config *compileopts.Config, action func(Buil
 		if err != nil {
 			return err
 		}
-		jobs = append(jobs, job.dependencies...)
-		jobs = append(jobs, job)
 		linkerDependencies = append(linkerDependencies, job)
 	}

-	// Add libc dependency if needed.
-	root := goenv.Get("TINYGOROOT")
-	switch config.Target.Libc {
-	case "picolibc":
-		job, err := Picolibc.load(config.Triple(), config.CPU(), dir)
-		if err != nil {
-			return err
-		}
-		// The library needs to be compiled (cache miss).
-		jobs = append(jobs, job.dependencies...)
-		jobs = append(jobs, job)
-		linkerDependencies = append(linkerDependencies, job)
-	case "wasi-libc":
-		path := filepath.Join(root, "lib/wasi-libc/sysroot/lib/wasm32-wasi/libc.a")
-		if _, err := os.Stat(path); os.IsNotExist(err) {
-			return errors.New("could not find wasi-libc, perhaps you need to run `make wasi-libc`?")
-		}
-		ldflags = append(ldflags, path)
-	case "":
-		// no library specified, so nothing to do
-	default:
-		return fmt.Errorf("unknown libc: %s", config.Target.Libc)
-	}
-
 	// Add jobs to compile extra files. These files are in C or assembly and
 	// contain things like the interrupt vector table and low level operations
 	// such as stack switching.
+	root := goenv.Get("TINYGOROOT")
 	for _, path := range config.ExtraFiles() {
 		abspath := filepath.Join(root, path)
 		job := &compileJob{
@@ -499,7 +474,6 @@ func Build(pkgName, outpath string, config *compileopts.Config, action func(Buil
 				return err
 			},
 		}
-		jobs = append(jobs, job)
 		linkerDependencies = append(linkerDependencies, job)
 	}

@@ -518,7 +492,6 @@ func Build(pkgName, outpath string, config *compileopts.Config, action func(Buil
 					return err
 				},
 			}
-			jobs = append(jobs, job)
 			linkerDependencies = append(linkerDependencies, job)
 		}
 	}
@@ -529,9 +502,63 @@ func Build(pkgName, outpath string, config *compileopts.Config, action func(Buil
 		ldflags = append(ldflags, lprogram.LDFlags...)
 	}

+	// Add libc dependency if needed.
+	switch config.Target.Libc {
+	case "picolibc":
+		job, err := Picolibc.load(config.Triple(), config.CPU(), dir)
+		if err != nil {
+			return err
+		}
+		linkerDependencies = append(linkerDependencies, job)
+	case "wasi-libc":
+		path := filepath.Join(root, "lib/wasi-libc/sysroot/lib/wasm32-wasi/libc.a")
+		if _, err := os.Stat(path); os.IsNotExist(err) {
+			return errors.New("could not find wasi-libc, perhaps you need to run `make wasi-libc`?")
+		}
+		job := dummyCompileJob(path)
+		linkerDependencies = append(linkerDependencies, job)
+	case "":
+		// no library specified, so nothing to do
+	default:
+		return fmt.Errorf("unknown libc: %s", config.Target.Libc)
+	}
+
+	// Strip debug information with -no-debug.
+	if !config.Debug() {
+		for _, tag := range config.BuildTags() {
+			if tag == "baremetal" {
+				// Don't use -no-debug on baremetal targets. It makes no sense:
+				// the debug information isn't flashed to the device anyway.
+				return fmt.Errorf("stripping debug information is unnecessary for baremetal targets")
+			}
+		}
+		if config.Target.Linker == "wasm-ld" {
+			// Don't just strip debug information, also compress relocations
+			// while we're at it. Relocations can only be compressed when debug
+			// information is stripped.
+			ldflags = append(ldflags, "--strip-debug", "--compress-relocations")
+		} else {
+			switch config.GOOS() {
+			case "linux":
+				// Either real linux or an embedded system (like AVR) that
+				// pretends to be Linux. It's a ELF linker wrapped by GCC in any
+				// case.
+				ldflags = append(ldflags, "-Wl,--strip-debug")
+			case "darwin":
+				// MacOS (darwin) doesn't have a linker flag to strip debug
+				// information. Apple expects you to use the strip command
+				// instead.
+				return errors.New("cannot remove debug information: MacOS doesn't suppor this linker flag")
+			default:
+				// Other OSes may have different flags.
+				return errors.New("cannot remove debug information: unknown OS: " + config.GOOS())
+			}
+		}
+	}
+
 	// Create a linker job, which links all object files together and does some
 	// extra stuff that can only be done after linking.
-	jobs = append(jobs, &compileJob{
+	linkJob := &compileJob{
 		description:  "link",
 		dependencies: linkerDependencies,
 		run: func(job *compileJob) error {
@@ -541,8 +568,8 @@ func Build(pkgName, outpath string, config *compileopts.Config, action func(Buil
 				}
 				ldflags = append(ldflags, dependency.result)
 			}
-			if config.Options.PrintCommands {
-				fmt.Printf("%s %s\n", config.Target.Linker, strings.Join(ldflags, " "))
+			if config.Options.PrintCommands != nil {
+				config.Options.PrintCommands(config.Target.Linker, ldflags...)
 			}
 			err = link(config.Target.Linker, ldflags...)
 			if err != nil {
@@ -560,6 +587,8 @@ func Build(pkgName, outpath string, config *compileopts.Config, action func(Buil
 					return err
 				}
 			}
+
+			// Apply ELF patches
 			if config.AutomaticStackSize() {
 				// Modify the .tinygo_stacksizes section that contains a stack size
 				// for each goroutine.
@@ -568,6 +597,13 @@ func Build(pkgName, outpath string, config *compileopts.Config, action func(Buil
 					return fmt.Errorf("could not modify stack sizes: %w", err)
 				}
 			}
+			if config.RP2040BootPatch() {
+				// Patch the second stage bootloader CRC into the .boot2 section
+				err = patchRP2040BootCRC(executable)
+				if err != nil {
+					return fmt.Errorf("could not patch RP2040 second stage boot loader: %w", err)
+				}
+			}

 			if config.Options.PrintSizes == "short" || config.Options.PrintSizes == "full" {
 				sizes, err := loadProgramSize(executable)
@@ -595,12 +631,12 @@ func Build(pkgName, outpath string, config *compileopts.Config, action func(Buil

 			return nil
 		},
-	})
+	}

 	// Run all jobs to compile and link the program.
 	// Do this now (instead of after elf-to-hex and similar conversions) as it
 	// is simpler and cannot be parallelized.
-	err = runJobs(jobs)
+	err = runJobs(linkJob)
 	if err != nil {
 		return err
 	}
@@ -625,7 +661,7 @@ func Build(pkgName, outpath string, config *compileopts.Config, action func(Buil
 		if err != nil {
 			return err
 		}
-	case "esp32", "esp8266":
+	case "esp32", "esp32c3", "esp8266":
 		// Special format for the ESP family of chips (parsed by the ROM
 		// bootloader).
 		tmppath = filepath.Join(dir, "main"+outext)
@@ -633,12 +669,25 @@ func Build(pkgName, outpath string, config *compileopts.Config, action func(Buil
 		if err != nil {
 			return err
 		}
+	case "nrf-dfu":
+		// special format for nrfutil for Nordic chips
+		tmphexpath := filepath.Join(dir, "main.hex")
+		err := objcopy(executable, tmphexpath, "hex")
+		if err != nil {
+			return err
+		}
+		tmppath = filepath.Join(dir, "main"+outext)
+		err = makeDFUFirmwareImage(config.Options, tmphexpath, tmppath)
+		if err != nil {
+			return err
+		}
 	default:
 		return fmt.Errorf("unknown output binary format: %s", outputBinaryFormat)
 	}
 	return action(BuildResult{
-		Binary:  tmppath,
-		MainDir: lprogram.MainPkg().Dir,
+		Binary:     tmppath,
+		MainDir:    lprogram.MainPkg().Dir,
+		ImportPath: lprogram.MainPkg().ImportPath,
 	})
 }

@@ -901,30 +950,7 @@ func determineStackSizes(mod llvm.Module, executable string) ([]string, map[stri
 // 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))
+	data, fileHeader, err := getElfSectionData(executable, ".tinygo_stacksizes")
 	if err != nil {
 		return err
 	}
@@ -946,15 +972,19 @@ func modifyStackSizes(executable string, stackSizeLoads []string, stackSizes map
 		if fn.stackSizeType == stacksize.Bounded {
 			stackSize := uint32(fn.stackSize)

-			// Adding 4 for the stack canary. Even though the size may be
-			// automatically determined, stack overflow checking is still
-			// important as the stack size cannot be determined for all
-			// goroutines.
-			stackSize += 4
-
 			// Add stack size used by interrupts.
-			switch elfFile.Machine {
+			switch fileHeader.Machine {
 			case elf.EM_ARM:
+				if stackSize%8 != 0 {
+					// If the stack isn't a multiple of 8, it means the leaf
+					// function with the biggest stack depth doesn't have an aligned
+					// stack. If the STKALIGN flag is set (which it is by default)
+					// the interrupt controller will forcibly align the stack before
+					// storing in-use registers. This will thus overwrite one word
+					// past the end of the stack (off-by-one).
+					stackSize += 4
+				}
+
 				// On Cortex-M (assumed here), this stack size is 8 words or 32
 				// bytes. This is only to store the registers that the interrupt
 				// may modify, the interrupt will switch to the interrupt stack
@@ -962,6 +992,14 @@ func modifyStackSizes(executable string, stackSizeLoads []string, stackSizes map
 				// Some background:
 				// https://interrupt.memfault.com/blog/cortex-m-rtos-context-switching
 				stackSize += 32
+
+				// Adding 4 for the stack canary, and another 4 to keep the
+				// stack aligned. Even though the size may be automatically
+				// determined, stack overflow checking is still important as the
+				// stack size cannot be determined for all goroutines.
+				stackSize += 8
+			default:
+				return fmt.Errorf("unknown architecture: %s", fileHeader.Machine.String())
 			}

 			// Finally write the stack size to the binary.
@@ -969,13 +1007,7 @@ func modifyStackSizes(executable string, stackSizeLoads []string, stackSizes map
 		}
 	}

-	// Write back the modified stack sizes.
-	_, err = fp.WriteAt(data, int64(section.Offset))
-	if err != nil {
-		return err
-	}
-
-	return nil
+	return replaceElfSection(executable, ".tinygo_stacksizes", data)
 }

 // printStacks prints the maximum stack depth for functions that are started as
@@ -1007,3 +1039,41 @@ func printStacks(calculatedStacks []string, stackSizes map[string]functionStackS
 		}
 	}
 }
+
+// RP2040 second stage bootloader CRC32 calculation
+//
+// Spec: https://datasheets.raspberrypi.org/rp2040/rp2040-datasheet.pdf
+// Section: 2.8.1.3.1. Checksum
+func patchRP2040BootCRC(executable string) error {
+	bytes, _, err := getElfSectionData(executable, ".boot2")
+	if err != nil {
+		return err
+	}
+
+	if len(bytes) != 256 {
+		return fmt.Errorf("rp2040 .boot2 section must be exactly 256 bytes")
+	}
+
+	// From the 'official' RP2040 checksum script:
+	//
+	//  Our bootrom CRC32 is slightly bass-ackward but it's
+	//  best to work around for now (FIXME)
+	//  100% worth it to save two Thumb instructions
+	revBytes := make([]byte, len(bytes))
+	for i := range bytes {
+		revBytes[i] = bits.Reverse8(bytes[i])
+	}
+
+	// crc32.Update does an initial negate and negates the
+	// result, so to meet RP2040 spec, pass 0x0 as initial
+	// hash and negate returned value.
+	//
+	// Note: checksum is over 252 bytes (256 - 4)
+	hash := bits.Reverse32(crc32.Update(0x0, crc32.IEEETable, revBytes[:252]) ^ 0xFFFFFFFF)
+
+	// Write the CRC to the end of the bootloader.
+	binary.LittleEndian.PutUint32(bytes[252:], hash)
+
+	// Update the .boot2 section to included the CRC
+	return replaceElfSection(executable, ".boot2", bytes)
+}
@@ -56,7 +56,7 @@ import (
 //   depfile but without invalidating its name. For this reason, the depfile is
 //   written on each new compilation (even when it seems unnecessary). However, it
 //   could in rare cases lead to a stale file fetched from the cache.
-func compileAndCacheCFile(abspath, tmpdir string, cflags []string, printCommands bool) (string, error) {
+func compileAndCacheCFile(abspath, tmpdir string, cflags []string, printCommands func(string, ...string)) (string, error) {
 	// Hash input file.
 	fileHash, err := hashFile(abspath)
 	if err != nil {
@@ -128,8 +128,8 @@ func compileAndCacheCFile(abspath, tmpdir string, cflags []string, printCommands
 		// flags (for the assembler) is a compiler error.
 		flags = append(flags, "-Qunused-arguments")
 	}
-	if printCommands {
-		fmt.Printf("clang %s\n", strings.Join(flags, " "))
+	if printCommands != nil {
+		printCommands("clang", flags...)
 	}
 	err = runCCompiler(flags...)
 	if err != nil {
@@ -155,6 +155,10 @@ func compileAndCacheCFile(abspath, tmpdir string, cflags []string, printCommands

 	// Write dependencies file.
 	f, err := ioutil.TempFile(filepath.Dir(depfileCachePath), depfileName)
+	if err != nil {
+		return "", err
+	}
+
 	buf, err = json.MarshalIndent(dependencySlice, "", "\t")
 	if err != nil {
 		panic(err) // shouldn't happen
@@ -17,18 +17,28 @@ func NewConfig(options *compileopts.Options) (*compileopts.Config, error) {
 	if err != nil {
 		return nil, err
 	}
+
+	if options.OpenOCDCommands != nil {
+		// Override the OpenOCDCommands from the target spec if specified on
+		// the command-line
+		spec.OpenOCDCommands = options.OpenOCDCommands
+	}
+
 	goroot := goenv.Get("GOROOT")
 	if goroot == "" {
 		return nil, errors.New("cannot locate $GOROOT, please set it manually")
 	}
+
 	major, minor, err := goenv.GetGorootVersion(goroot)
 	if err != nil {
 		return nil, fmt.Errorf("could not read version from GOROOT (%v): %v", goroot, err)
 	}
-	if major != 1 || minor < 13 || minor > 16 {
-		return nil, fmt.Errorf("requires go version 1.13 through 1.16, got go%d.%d", major, minor)
+	if major != 1 || minor < 15 || minor > 17 {
+		return nil, fmt.Errorf("requires go version 1.15 through 1.17, got go%d.%d", major, minor)
 	}
+
 	clangHeaderPath := getClangHeaderPath(goenv.Get("TINYGOROOT"))
+
 	return &compileopts.Config{
 		Options:        options,
 		Target:         spec,
@@ -0,0 +1,57 @@
+package builder
+
+import (
+	"debug/elf"
+	"fmt"
+	"os"
+)
+
+func getElfSectionData(executable string, sectionName string) ([]byte, elf.FileHeader, error) {
+	elfFile, err := elf.Open(executable)
+	if err != nil {
+		return nil, elf.FileHeader{}, err
+	}
+	defer elfFile.Close()
+
+	section := elfFile.Section(sectionName)
+	if section == nil {
+		return nil, elf.FileHeader{}, fmt.Errorf("could not find %s section", sectionName)
+	}
+
+	data, err := section.Data()
+
+	return data, elfFile.FileHeader, err
+}
+
+func replaceElfSection(executable string, sectionName string, data []byte) error {
+	fp, err := os.OpenFile(executable, os.O_RDWR, 0)
+	if err != nil {
+		return err
+	}
+	defer fp.Close()
+
+	elfFile, err := elf.Open(executable)
+	if err != nil {
+		return err
+	}
+	defer elfFile.Close()
+
+	section := elfFile.Section(sectionName)
+	if section == nil {
+		return fmt.Errorf("could not find %s section", sectionName)
+	}
+
+	// Implicitly check for compressed sections
+	if section.Size != section.FileSize {
+		return fmt.Errorf("expected section %s to have identical size and file size, got %d and %d", sectionName, section.Size, section.FileSize)
+	}
+
+	// Only permit complete replacement of section
+	if section.Size != uint64(len(data)) {
+		return fmt.Errorf("expected section %s to have size %d, was actually %d", sectionName, len(data), section.Size)
+	}
+
+	// Write the replacement section data
+	_, err = fp.WriteAt(data, int64(section.Offset))
+	return err
+}
@@ -78,11 +78,21 @@ func makeESPFirmareImage(infile, outfile, format string) error {
 	// An added benefit is that we don't need to check for errors all the time.
 	outf := &bytes.Buffer{}

+	// Chip IDs. Source:
+	// https://github.com/espressif/esp-idf/blob/v4.3/components/bootloader_support/include/esp_app_format.h#L22
+	chip_id := map[string]uint16{
+		"esp32":   0x0000,
+		"esp32c3": 0x0005,
+	}[format]
+
 	// Image header.
 	switch format {
-	case "esp32":
+	case "esp32", "esp32c3":
 		// Header format:
-		// https://github.com/espressif/esp-idf/blob/8fbb63c2/components/bootloader_support/include/esp_image_format.h#L58
+		// https://github.com/espressif/esp-idf/blob/v4.3/components/bootloader_support/include/esp_app_format.h#L71
+		// Note: not adding a SHA256 hash as the binary is modified by
+		// esptool.py while flashing and therefore the hash won't be valid
+		// anymore.
 		binary.Write(outf, binary.LittleEndian, struct {
 			magic          uint8
 			segment_count  uint8
@@ -91,15 +101,18 @@ func makeESPFirmareImage(infile, outfile, format string) error {
 			entry_addr     uint32
 			wp_pin         uint8
 			spi_pin_drv    [3]uint8
-			reserved       [11]uint8
+			chip_id        uint16
+			min_chip_rev   uint8
+			reserved       [8]uint8
 			hash_appended  bool
 		}{
 			magic:          0xE9,
 			segment_count:  byte(len(segments)),
-			spi_mode:       0, // irrelevant, replaced by esptool when flashing
-			spi_speed_size: 0, // spi_speed, spi_size: replaced by esptool when flashing
+			spi_mode:       2,    // ESP_IMAGE_SPI_MODE_DIO
+			spi_speed_size: 0x1f, // ESP_IMAGE_SPI_SPEED_80M, ESP_IMAGE_FLASH_SIZE_2MB
 			entry_addr:     uint32(inf.Entry),
 			wp_pin:         0xEE, // disable WP pin
+			chip_id:        chip_id,
 			hash_appended:  true, // add a SHA256 hash
 		})
 	case "esp8266":
@@ -142,7 +155,7 @@ func makeESPFirmareImage(infile, outfile, format string) error {
 	outf.Write(make([]byte, 15-outf.Len()%16))
 	outf.WriteByte(checksum)

-	if format == "esp32" {
+	if format != "esp8266" {
 		// SHA256 hash (to protect against image corruption, not for security).
 		hash := sha256.Sum256(outf.Bytes())
 		outf.Write(hash[:])
@@ -65,12 +65,29 @@ func (job *compileJob) readyToRun() bool {
 	return true
 }

-// runJobs runs all the jobs indicated in the jobs slice and returns the error
-// of the first job that fails to run.
-// It runs all jobs in the order of the slice, as long as all dependencies have
-// already run. Therefore, if some jobs are preferred to run before others, they
-// should be ordered as such in this slice.
-func runJobs(jobs []*compileJob) error {
+// runJobs runs the indicated job and all its dependencies. For every job, all
+// the dependencies are run first. It returns the error of the first job that
+// fails.
+// It runs all jobs in the order of the dependencies slice, depth-first.
+// Therefore, if some jobs are preferred to run before others, they should be
+// ordered as such in the job dependencies.
+func runJobs(job *compileJob) error {
+	// Create a slice of jobs to run, where all dependencies are run in order.
+	jobs := []*compileJob{}
+	addedJobs := map[*compileJob]struct{}{}
+	var addJobs func(*compileJob)
+	addJobs = func(job *compileJob) {
+		if _, ok := addedJobs[job]; ok {
+			return
+		}
+		for _, dep := range job.dependencies {
+			addJobs(dep)
+		}
+		jobs = append(jobs, job)
+		addedJobs[job] = struct{}{}
+	}
+	addJobs(job)
+
 	// Create channels to communicate with the workers.
 	doneChan := make(chan *compileJob)
 	workerChan := make(chan *compileJob)
@@ -46,15 +46,14 @@ func (l *Library) Load(target, tmpdir string) (path string, err error) {
 	if err != nil {
 		return "", err
 	}
-	jobs := append([]*compileJob{job}, job.dependencies...)
-	err = runJobs(jobs)
+	err = runJobs(job)
 	return job.result, err
 }

 // load returns a compile job to build this library file for the given target
 // and CPU. It may return a dummy compileJob if the library build is already
-// cached. The path is stored as job.result but is only valid if the job and
-// job.dependencies have been run.
+// cached. The path is stored as job.result but is only valid after the job has
+// been run.
 // The provided tmpdir will be used to store intermediary files and possibly the
 // output archive file, it is expected to be removed after use.
 func (l *Library) load(target, cpu, tmpdir string) (job *compileJob, err error) {
@@ -0,0 +1,27 @@
+package builder
+
+import (
+	"fmt"
+	"io/ioutil"
+	"os/exec"
+
+	"github.com/tinygo-org/tinygo/compileopts"
+)
+
+// https://infocenter.nordicsemi.com/index.jsp?topic=%2Fug_nrfutil%2FUG%2Fnrfutil%2Fnrfutil_intro.html
+
+func makeDFUFirmwareImage(options *compileopts.Options, infile, outfile string) error {
+	cmdLine := []string{"nrfutil", "pkg", "generate", "--hw-version", "52", "--sd-req", "0x0", "--debug-mode", "--application", infile, outfile}
+
+	if options.PrintCommands != nil {
+		options.PrintCommands(cmdLine[0], cmdLine[1:]...)
+	}
+
+	cmd := exec.Command(cmdLine[0], cmdLine[1:]...)
+	cmd.Stdout = ioutil.Discard
+	err := cmd.Run()
+	if err != nil {
+		return fmt.Errorf("could not run nrfutil pkg generate: %w", err)
+	}
+	return nil
+}
@@ -12,7 +12,6 @@ import (
 	"go/types"
 	"io/ioutil"
 	"path/filepath"
-	"regexp"
 	"runtime"
 	"strings"
 	"testing"
@@ -25,14 +24,6 @@ var flagUpdate = flag.Bool("update", false, "Update images based on test output.
 // platforms and Go versions.
 func normalizeResult(result string) string {
 	actual := strings.ReplaceAll(result, "\r\n", "\n")
-
-	// Make sure all functions are wrapped, even those that would otherwise be
-	// single-line functions. This is necessary because Go 1.14 changed the way
-	// such functions are wrapped and it's important to have consistent test
-	// results.
-	re := regexp.MustCompile(`func \((.+)\)( .*?) +{ (.+) }`)
-	actual = re.ReplaceAllString(actual, "func ($1)$2 {\n\t$3\n}")
-
 	return actual
 }

@@ -11,105 +11,184 @@ import (
 	"strings"
 )

+var (
+	prefixParseFns map[token.Token]func(*tokenizer) (ast.Expr, *scanner.Error)
+	precedences    = map[token.Token]int{
+		token.ADD: precedenceAdd,
+		token.SUB: precedenceAdd,
+		token.MUL: precedenceMul,
+		token.QUO: precedenceMul,
+		token.REM: precedenceMul,
+	}
+)
+
+const (
+	precedenceLowest = iota + 1
+	precedenceAdd
+	precedenceMul
+	precedencePrefix
+)
+
+func init() {
+	// This must be done in an init function to avoid an initialization order
+	// failure.
+	prefixParseFns = map[token.Token]func(*tokenizer) (ast.Expr, *scanner.Error){
+		token.IDENT:  parseIdent,
+		token.INT:    parseBasicLit,
+		token.FLOAT:  parseBasicLit,
+		token.STRING: parseBasicLit,
+		token.CHAR:   parseBasicLit,
+		token.LPAREN: parseParenExpr,
+		token.SUB:    parseUnaryExpr,
+	}
+}
+
 // parseConst parses the given string as a C constant.
 func parseConst(pos token.Pos, fset *token.FileSet, value string) (ast.Expr, *scanner.Error) {
 	t := newTokenizer(pos, fset, value)
-	expr, err := parseConstExpr(t)
-	if t.token != token.EOF {
+	expr, err := parseConstExpr(t, precedenceLowest)
+	t.Next()
+	if t.curToken != token.EOF {
 		return nil, &scanner.Error{
-			Pos: t.fset.Position(t.pos),
-			Msg: "unexpected token " + t.token.String(),
+			Pos: t.fset.Position(t.curPos),
+			Msg: "unexpected token " + t.curToken.String() + ", expected end of expression",
 		}
 	}
 	return expr, err
 }

 // parseConstExpr parses a stream of C tokens to a Go expression.
-func parseConstExpr(t *tokenizer) (ast.Expr, *scanner.Error) {
-	switch t.token {
-	case token.LPAREN:
-		lparen := t.pos
-		t.Next()
-		x, err := parseConstExpr(t)
-		if err != nil {
-			return nil, err
-		}
-		if t.token != token.RPAREN {
-			return nil, unexpectedToken(t, token.RPAREN)
-		}
-		expr := &ast.ParenExpr{
-			Lparen: lparen,
-			X:      x,
-			Rparen: t.pos,
-		}
-		t.Next()
-		return expr, nil
-	case token.INT, token.FLOAT, token.STRING, token.CHAR:
-		expr := &ast.BasicLit{
-			ValuePos: t.pos,
-			Kind:     t.token,
-			Value:    t.value,
-		}
-		t.Next()
-		return expr, nil
-	case token.IDENT:
-		expr := &ast.Ident{
-			NamePos: t.pos,
-			Name:    "C." + t.value,
-		}
-		t.Next()
-		return expr, nil
-	case token.EOF:
+func parseConstExpr(t *tokenizer, precedence int) (ast.Expr, *scanner.Error) {
+	if t.curToken == token.EOF {
 		return nil, &scanner.Error{
-			Pos: t.fset.Position(t.pos),
+			Pos: t.fset.Position(t.curPos),
 			Msg: "empty constant",
 		}
-	default:
+	}
+	prefix := prefixParseFns[t.curToken]
+	if prefix == nil {
 		return nil, &scanner.Error{
-			Pos: t.fset.Position(t.pos),
-			Msg: fmt.Sprintf("unexpected token %s", t.token),
+			Pos: t.fset.Position(t.curPos),
+			Msg: fmt.Sprintf("unexpected token %s", t.curToken),
 		}
 	}
+	leftExpr, err := prefix(t)
+
+	for t.peekToken != token.EOF && precedence < precedences[t.peekToken] {
+		switch t.peekToken {
+		case token.ADD, token.SUB, token.MUL, token.QUO, token.REM:
+			t.Next()
+			leftExpr, err = parseBinaryExpr(t, leftExpr)
+		}
+	}
+
+	return leftExpr, err
+}
+
+func parseIdent(t *tokenizer) (ast.Expr, *scanner.Error) {
+	return &ast.Ident{
+		NamePos: t.curPos,
+		Name:    "C." + t.curValue,
+	}, nil
+}
+
+func parseBasicLit(t *tokenizer) (ast.Expr, *scanner.Error) {
+	return &ast.BasicLit{
+		ValuePos: t.curPos,
+		Kind:     t.curToken,
+		Value:    t.curValue,
+	}, nil
+}
+
+func parseParenExpr(t *tokenizer) (ast.Expr, *scanner.Error) {
+	lparen := t.curPos
+	t.Next()
+	x, err := parseConstExpr(t, precedenceLowest)
+	if err != nil {
+		return nil, err
+	}
+	t.Next()
+	if t.curToken != token.RPAREN {
+		return nil, unexpectedToken(t, token.RPAREN)
+	}
+	expr := &ast.ParenExpr{
+		Lparen: lparen,
+		X:      x,
+		Rparen: t.curPos,
+	}
+	return expr, nil
+}
+
+func parseBinaryExpr(t *tokenizer, left ast.Expr) (ast.Expr, *scanner.Error) {
+	expression := &ast.BinaryExpr{
+		X:     left,
+		Op:    t.curToken,
+		OpPos: t.curPos,
+	}
+	precedence := precedences[t.curToken]
+	t.Next()
+	right, err := parseConstExpr(t, precedence)
+	expression.Y = right
+	return expression, err
+}
+
+func parseUnaryExpr(t *tokenizer) (ast.Expr, *scanner.Error) {
+	expression := &ast.UnaryExpr{
+		OpPos: t.curPos,
+		Op:    t.curToken,
+	}
+	t.Next()
+	x, err := parseConstExpr(t, precedencePrefix)
+	expression.X = x
+	return expression, err
 }

 // unexpectedToken returns an error of the form "unexpected token FOO, expected
 // BAR".
 func unexpectedToken(t *tokenizer, expected token.Token) *scanner.Error {
 	return &scanner.Error{
-		Pos: t.fset.Position(t.pos),
-		Msg: fmt.Sprintf("unexpected token %s, expected %s", t.token, expected),
+		Pos: t.fset.Position(t.curPos),
+		Msg: fmt.Sprintf("unexpected token %s, expected %s", t.curToken, expected),
 	}
 }

 // tokenizer reads C source code and converts it to Go tokens.
 type tokenizer struct {
-	pos   token.Pos
-	fset  *token.FileSet
-	token token.Token
-	value string
-	buf   string
+	curPos, peekPos     token.Pos
+	fset                *token.FileSet
+	curToken, peekToken token.Token
+	curValue, peekValue string
+	buf                 string
 }

 // newTokenizer initializes a new tokenizer, positioned at the first token in
 // the string.
 func newTokenizer(start token.Pos, fset *token.FileSet, buf string) *tokenizer {
 	t := &tokenizer{
-		pos:   start,
-		fset:  fset,
-		buf:   buf,
-		token: token.ILLEGAL,
+		peekPos:   start,
+		fset:      fset,
+		buf:       buf,
+		peekToken: token.ILLEGAL,
 	}
-	t.Next() // Parse the first token.
+	// Parse the first two tokens (cur and peek).
+	t.Next()
+	t.Next()
 	return t
 }

 // Next consumes the next token in the stream. There is no return value, read
 // the next token from the pos, token and value properties.
 func (t *tokenizer) Next() {
-	t.pos += token.Pos(len(t.value))
+	// The previous peek is now the current token.
+	t.curPos = t.peekPos
+	t.curToken = t.peekToken
+	t.curValue = t.peekValue
+
+	// Parse the next peek token.
+	t.peekPos += token.Pos(len(t.curValue))
 	for {
 		if len(t.buf) == 0 {
-			t.token = token.EOF
+			t.peekToken = token.EOF
 			return
 		}
 		c := t.buf[0]
@@ -118,17 +197,28 @@ func (t *tokenizer) Next() {
 			// Skip whitespace.
 			// Based on this source, not sure whether it represents C whitespace:
 			// https://en.cppreference.com/w/cpp/string/byte/isspace
-			t.pos++
+			t.peekPos++
 			t.buf = t.buf[1:]
-		case c == '(' || c == ')':
+		case c == '(' || c == ')' || c == '+' || c == '-' || c == '*' || c == '/' || c == '%':
 			// Single-character tokens.
+			// TODO: ++ (increment) and -- (decrement) operators.
 			switch c {
 			case '(':
-				t.token = token.LPAREN
+				t.peekToken = token.LPAREN
 			case ')':
-				t.token = token.RPAREN
+				t.peekToken = token.RPAREN
+			case '+':
+				t.peekToken = token.ADD
+			case '-':
+				t.peekToken = token.SUB
+			case '*':
+				t.peekToken = token.MUL
+			case '/':
+				t.peekToken = token.QUO
+			case '%':
+				t.peekToken = token.REM
 			}
-			t.value = t.buf[:1]
+			t.peekValue = t.buf[:1]
 			t.buf = t.buf[1:]
 			return
 		case c >= '0' && c <= '9':
@@ -146,17 +236,17 @@ func (t *tokenizer) Next() {
 					break
 				}
 			}
-			t.value = t.buf[:tokenLen]
+			t.peekValue = t.buf[:tokenLen]
 			t.buf = t.buf[tokenLen:]
 			if hasDot {
 				// Integer constants are more complicated than this but this is
 				// a close approximation.
 				// https://en.cppreference.com/w/cpp/language/integer_literal
-				t.token = token.FLOAT
-				t.value = strings.TrimRight(t.value, "f")
+				t.peekToken = token.FLOAT
+				t.peekValue = strings.TrimRight(t.peekValue, "f")
 			} else {
-				t.token = token.INT
-				t.value = strings.TrimRight(t.value, "uUlL")
+				t.peekToken = token.INT
+				t.peekValue = strings.TrimRight(t.peekValue, "uUlL")
 			}
 			return
 		case c >= 'A' && c <= 'Z' || c >= 'a' && c <= 'z' || c == '_':
@@ -170,9 +260,9 @@ func (t *tokenizer) Next() {
 					break
 				}
 			}
-			t.value = t.buf[:tokenLen]
+			t.peekValue = t.buf[:tokenLen]
 			t.buf = t.buf[tokenLen:]
-			t.token = token.IDENT
+			t.peekToken = token.IDENT
 			return
 		case c == '"':
 			// String constant. Find the first '"' character that is not
@@ -188,8 +278,8 @@ func (t *tokenizer) Next() {
 					escape = c == '\\'
 				}
 			}
-			t.token = token.STRING
-			t.value = t.buf[:tokenLen]
+			t.peekToken = token.STRING
+			t.peekValue = t.buf[:tokenLen]
 			t.buf = t.buf[tokenLen:]
 			return
 		case c == '\'':
@@ -206,12 +296,12 @@ func (t *tokenizer) Next() {
 					escape = c == '\\'
 				}
 			}
-			t.token = token.CHAR
-			t.value = t.buf[:tokenLen]
+			t.peekToken = token.CHAR
+			t.peekValue = t.buf[:tokenLen]
 			t.buf = t.buf[tokenLen:]
 			return
 		default:
-			t.token = token.ILLEGAL
+			t.peekToken = token.ILLEGAL
 			return
 		}
 	}
@@ -18,7 +18,7 @@ func TestParseConst(t *testing.T) {
 		{`(5)`, `(5)`},
 		{`(((5)))`, `(5)`},
 		{`)`, `error: 1:1: unexpected token )`},
-		{`5)`, `error: 1:2: unexpected token )`},
+		{`5)`, `error: 1:2: unexpected token ), expected end of expression`},
 		{"  \t)", `error: 1:4: unexpected token )`},
 		{`5.8f`, `5.8`},
 		{`foo`, `C.foo`},
@@ -30,7 +30,24 @@ func TestParseConst(t *testing.T) {
 		{`'a'`, `'a'`},
 		{`0b10`, `0b10`},
 		{`0x1234_5678`, `0x1234_5678`},
-		{`5 5`, `error: 1:3: unexpected token INT`}, // test for a bugfix
+		{`5 5`, `error: 1:3: unexpected token INT, expected end of expression`}, // test for a bugfix
+		// Binary operators.
+		{`5+5`, `5 + 5`},
+		{`5-5`, `5 - 5`},
+		{`5*5`, `5 * 5`},
+		{`5/5`, `5 / 5`},
+		{`5%5`, `5 % 5`},
+		{`(5/5)`, `(5 / 5)`},
+		{`1 - 2`, `1 - 2`},
+		{`1 - 2 + 3`, `1 - 2 + 3`},
+		{`1 - 2 * 3`, `1 - 2*3`},
+		{`(1 - 2) * 3`, `(1 - 2) * 3`},
+		{`1 * 2 - 3`, `1*2 - 3`},
+		{`1 * (2 - 3)`, `1 * (2 - 3)`},
+		// Unary operators.
+		{`-5`, `-5`},
+		{`-5-2`, `-5 - 2`},
+		{`5 - - 2`, `5 - -2`},
 	} {
 		fset := token.NewFileSet()
 		startPos := fset.AddFile("", -1, 1000).Pos(0)
@@ -1,7 +1,7 @@
 // CGo errors:
 //     testdata/errors.go:4:2: warning: some warning
 //     testdata/errors.go:11:9: error: unknown type name 'someType'
-//     testdata/errors.go:13:23: unexpected token )
+//     testdata/errors.go:13:23: unexpected token ), expected end of expression

 // Type checking errors after CGo processing:
 //     testdata/errors.go:102: cannot use 2 << 10 (untyped int constant 2048) as uint8 value in variable declaration (overflows)
@@ -67,9 +67,7 @@ type C.union3_t = C.union_1
 type C.union_nested_t = C.union_3
 type C.unionarray_t = struct{ arr [10]C.uchar }

-func (s *C.struct_4) bitfield_a() C.uchar {
-	return s.__bitfield_1 & 0x1f
-}
+func (s *C.struct_4) bitfield_a() C.uchar { return s.__bitfield_1 & 0x1f }
 func (s *C.struct_4) set_bitfield_a(value C.uchar) {
 	s.__bitfield_1 = s.__bitfield_1&^0x1f | value&0x1f<<0
 }
@@ -105,15 +103,9 @@ type C.struct_type1 struct {
 }
 type C.struct_type2 struct{ _type C.int }

-func (union *C.union_1) unionfield_i() *C.int {
-	return (*C.int)(unsafe.Pointer(&union.$union))
-}
-func (union *C.union_1) unionfield_d() *float64 {
-	return (*float64)(unsafe.Pointer(&union.$union))
-}
-func (union *C.union_1) unionfield_s() *C.short {
-	return (*C.short)(unsafe.Pointer(&union.$union))
-}
+func (union *C.union_1) unionfield_i() *C.int   { return (*C.int)(unsafe.Pointer(&union.$union)) }
+func (union *C.union_1) unionfield_d() *float64 { return (*float64)(unsafe.Pointer(&union.$union)) }
+func (union *C.union_1) unionfield_s() *C.short { return (*C.short)(unsafe.Pointer(&union.$union)) }

 type C.union_1 struct{ $union uint64 }

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

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

-func (union *C.union_union2d) unionfield_i() *C.int {
-	return (*C.int)(unsafe.Pointer(&union.$union))
-}
+func (union *C.union_union2d) unionfield_i() *C.int { return (*C.int)(unsafe.Pointer(&union.$union)) }
 func (union *C.union_union2d) unionfield_d() *[2]float64 {
 	return (*[2]float64)(unsafe.Pointer(&union.$union))
 }
@@ -55,7 +55,7 @@ func (c *Config) GOARCH() string {

 // BuildTags returns the complete list of build tags used during this build.
 func (c *Config) BuildTags() []string {
-	tags := append(c.Target.BuildTags, []string{"tinygo", "gc." + c.GC(), "scheduler." + c.Scheduler()}...)
+	tags := append(c.Target.BuildTags, []string{"tinygo", "math_big_pure_go", "gc." + c.GC(), "scheduler." + c.Scheduler(), "serial." + c.Serial()}...)
 	for i := 1; i <= c.GoMinorVersion; i++ {
 		tags = append(tags, fmt.Sprintf("go1.%d", i))
 	}
@@ -80,12 +80,7 @@ func (c *Config) GC() string {
 	if c.Target.GC != "" {
 		return c.Target.GC
 	}
-	for _, tag := range c.Target.BuildTags {
-		if tag == "baremetal" || tag == "wasm" {
-			return "conservative"
-		}
-	}
-	return "extalloc"
+	return "conservative"
 }

 // NeedsStackObjects returns true if the compiler should insert stack objects
@@ -94,7 +89,7 @@ func (c *Config) NeedsStackObjects() bool {
 	switch c.GC() {
 	case "conservative", "extalloc":
 		for _, tag := range c.BuildTags() {
-			if tag == "wasm" {
+			if tag == "tinygo.wasm" {
 				return true
 			}
 		}
@@ -118,6 +113,18 @@ func (c *Config) Scheduler() string {
 	return "coroutines"
 }

+// Serial returns the serial implementation for this build configuration: uart,
+// usb (meaning USB-CDC), or none.
+func (c *Config) Serial() string {
+	if c.Options.Serial != "" {
+		return c.Options.Serial
+	}
+	if c.Target.Serial != "" {
+		return c.Target.Serial
+	}
+	return "none"
+}
+
 // OptLevels returns the optimization level (0-2), size level (0-2), and inliner
 // threshold as used in the LLVM optimization pipeline.
 func (c *Config) OptLevels() (optLevel, sizeLevel int, inlinerThreshold uint) {
@@ -181,6 +188,15 @@ func (c *Config) AutomaticStackSize() bool {
 	return false
 }

+// RP2040BootPatch returns whether the RP2040 boot patch should be applied that
+// calculates and patches in the checksum for the 2nd stage bootloader.
+func (c *Config) RP2040BootPatch() bool {
+	if c.Target.RP2040BootPatch != nil {
+		return *c.Target.RP2040BootPatch
+	}
+	return false
+}
+
 // CFlags returns the flags to pass to the C compiler. This is necessary for CGo
 // preprocessing.
 func (c *Config) CFlags() []string {
@@ -193,9 +209,8 @@ func (c *Config) CFlags() []string {
 		cflags = append(cflags, "-nostdlibinc", "-Xclang", "-internal-isystem", "-Xclang", filepath.Join(root, "lib", "picolibc", "newlib", "libc", "include"))
 		cflags = append(cflags, "-I"+filepath.Join(root, "lib/picolibc-include"))
 	}
-	if c.Debug() {
-		cflags = append(cflags, "-g")
-	}
+	// Always emit debug information. It is optionally stripped at link time.
+	cflags = append(cflags, "-g")
 	return cflags
 }

@@ -234,8 +249,9 @@ func (c *Config) VerifyIR() bool {
 	return c.Options.VerifyIR
 }

-// Debug returns whether to add debug symbols to the IR, for debugging with GDB
-// and similar.
+// Debug returns whether debug (DWARF) information should be retained by the
+// linker. By default, debug information is retained but it can be removed with
+// the -no-debug flag.
 func (c *Config) Debug() bool {
 	return c.Options.Debug
 }
@@ -259,6 +275,11 @@ func (c *Config) BinaryFormat(ext string) string {
 		// More information:
 		// https://github.com/Microsoft/uf2
 		return "uf2"
+	case ".zip":
+		if c.Target.BinaryFormat != "" {
+			return c.Target.BinaryFormat
+		}
+		return "zip"
 	default:
 		// Use the ELF format for unrecognized file formats.
 		return "elf"
@@ -276,6 +297,9 @@ func (c *Config) Programmer() (method, openocdInterface string) {
 	case "openocd", "msd", "command":
 		// The -programmer flag only specifies the flash method.
 		return c.Options.Programmer, c.Target.OpenOCDInterface
+	case "bmp":
+		// The -programmer flag only specifies the flash method.
+		return c.Options.Programmer, ""
 	default:
 		// The -programmer flag specifies something else, assume it specifies
 		// the OpenOCD interface name.
@@ -342,6 +366,10 @@ func (c *Config) WasmAbi() string {
 	return c.Target.WasmAbi
 }

+func (c *Config) LLVMFeatures() string {
+	return c.Options.LLVMFeatures
+}
+
 type TestConfig struct {
 	CompileTestBinary bool
 	// TODO: Filter the test functions to run, include verbose flag, etc
@@ -9,6 +9,7 @@ import (
 var (
 	validGCOptions            = []string{"none", "leaking", "extalloc", "conservative"}
 	validSchedulerOptions     = []string{"none", "tasks", "coroutines"}
+	validSerialOptions        = []string{"none", "uart", "usb"}
 	validPrintSizeOptions     = []string{"none", "short", "full"}
 	validPanicStrategyOptions = []string{"print", "trap"}
 	validOptOptions           = []string{"none", "0", "1", "2", "s", "z"}
@@ -17,24 +18,27 @@ var (
 // Options contains extra options to give to the compiler. These options are
 // usually passed from the command line.
 type Options struct {
-	Target        string
-	Opt           string
-	GC            string
-	PanicStrategy string
-	Scheduler     string
-	PrintIR       bool
-	DumpSSA       bool
-	VerifyIR      bool
-	PrintCommands bool
-	Debug         bool
-	PrintSizes    string
-	PrintAllocs   *regexp.Regexp // regexp string
-	PrintStacks   bool
-	Tags          string
-	WasmAbi       string
-	GlobalValues  map[string]map[string]string // map[pkgpath]map[varname]value
-	TestConfig    TestConfig
-	Programmer    string
+	Target          string
+	Opt             string
+	GC              string
+	PanicStrategy   string
+	Scheduler       string
+	Serial          string
+	PrintIR         bool
+	DumpSSA         bool
+	VerifyIR        bool
+	PrintCommands   func(cmd string, args ...string)
+	Debug           bool
+	PrintSizes      string
+	PrintAllocs     *regexp.Regexp // regexp string
+	PrintStacks     bool
+	Tags            string
+	WasmAbi         string
+	GlobalValues    map[string]map[string]string // map[pkgpath]map[varname]value
+	TestConfig      TestConfig
+	Programmer      string
+	OpenOCDCommands []string
+	LLVMFeatures    string
 }

 // Verify performs a validation on the given options, raising an error if options are not valid.
@@ -57,6 +61,15 @@ func (o *Options) Verify() error {
 		}
 	}

+	if o.Serial != "" {
+		valid := isInArray(validSerialOptions, o.Serial)
+		if !valid {
+			return fmt.Errorf(`invalid serial option '%s': valid values are %s`,
+				o.Serial,
+				strings.Join(validSerialOptions, ", "))
+		}
+	}
+
 	if o.PrintSizes != "" {
 		valid := isInArray(validPrintSizeOptions, o.PrintSizes)
 		if !valid {
@@ -31,6 +31,7 @@ type TargetSpec struct {
 	BuildTags        []string `json:"build-tags"`
 	GC               string   `json:"gc"`
 	Scheduler        string   `json:"scheduler"`
+	Serial           string   `json:"serial"` // which serial output to use (uart, usb, none)
 	Linker           string   `json:"linker"`
 	RTLib            string   `json:"rtlib"` // compiler runtime library (libgcc, compiler-rt)
 	Libc             string   `json:"libc"`
@@ -40,10 +41,12 @@ type TargetSpec struct {
 	LDFlags          []string `json:"ldflags"`
 	LinkerScript     string   `json:"linkerscript"`
 	ExtraFiles       []string `json:"extra-files"`
+	RP2040BootPatch  *bool    `json:"rp2040-boot-patch"`        // Patch RP2040 2nd stage bootloader checksum
 	Emulator         []string `json:"emulator" override:"copy"` // inherited Emulator must not be append
 	FlashCommand     string   `json:"flash-command"`
 	GDB              []string `json:"gdb"`
 	PortReset        string   `json:"flash-1200-bps-reset"`
+	SerialPort       []string `json:"serial-port"` // serial port IDs in the form "acm:vid:pid" or "usb:vid:pid"
 	FlashMethod      string   `json:"flash-method"`
 	FlashVolume      string   `json:"msd-volume-name"`
 	FlashFilename    string   `json:"msd-firmware-name"`
@@ -169,6 +172,7 @@ func LoadTarget(target string) (*TargetSpec, error) {
 			"386":   "i386",
 			"amd64": "x86_64",
 			"arm64": "aarch64",
+			"arm":   "thumbv7",
 		}[goarch]
 		if llvmarch == "" {
 			llvmarch = goarch
@@ -239,22 +243,29 @@ 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},
-		Linker:    "cc",
-		CFlags:    []string{"--target=" + triple},
-		GDB:       []string{"gdb"},
-		PortReset: "false",
+		Triple:           triple,
+		GOOS:             goos,
+		GOARCH:           goarch,
+		BuildTags:        []string{goos, goarch},
+		Scheduler:        "tasks",
+		Linker:           "cc",
+		DefaultStackSize: 1024 * 64, // 64kB
+		CFlags:           []string{"--target=" + triple},
+		GDB:              []string{"gdb"},
+		PortReset:        "false",
+	}
+	if goarch == "386" {
+		spec.CPU = "pentium4"
 	}
 	if goos == "darwin" {
+		spec.CFlags = append(spec.CFlags, "-isysroot", "/Library/Developer/CommandLineTools/SDKs/MacOSX.sdk")
 		spec.LDFlags = append(spec.LDFlags, "-Wl,-dead_strip")
 	} else {
 		spec.LDFlags = append(spec.LDFlags, "-no-pie", "-Wl,--gc-sections") // WARNING: clang < 5.0 requires -nopie
 	}
 	if goarch != "wasm" {
 		spec.ExtraFiles = append(spec.ExtraFiles, "src/runtime/gc_"+goarch+".S")
+		spec.ExtraFiles = append(spec.ExtraFiles, "src/internal/task/task_stack_"+goarch+".S")
 	}
 	if goarch != runtime.GOARCH {
 		// Some educated guesses as to how to invoke helper programs.
@@ -0,0 +1,99 @@
+package compiler
+
+// This file defines alias functions for functions that are normally defined in
+// Go assembly.
+//
+// The Go toolchain defines many performance critical functions in assembly
+// instead of plain Go. This is a problem for TinyGo as it currently (as of
+// august 2021) is not able to compile these assembly files and even if it
+// could, it would not be able to make use of them for many targets that are
+// supported by TinyGo (baremetal RISC-V, AVR, etc). Therefore, many of these
+// functions are aliased to their generic Go implementation.
+// This results in slower than possible implementations, but at least they are
+// usable.
+
+import "tinygo.org/x/go-llvm"
+
+var stdlibAliases = map[string]string{
+	// crypto packages
+	"crypto/md5.block":         "crypto/md5.blockGeneric",
+	"crypto/sha1.block":        "crypto/sha1.blockGeneric",
+	"crypto/sha1.blockAMD64":   "crypto/sha1.blockGeneric",
+	"crypto/sha256.block":      "crypto/sha256.blockGeneric",
+	"crypto/sha512.blockAMD64": "crypto/sha512.blockGeneric",
+
+	// math package
+	"math.Asin":      "math.asin",
+	"math.Asinh":     "math.asinh",
+	"math.Acos":      "math.acos",
+	"math.Acosh":     "math.acosh",
+	"math.Atan":      "math.atan",
+	"math.Atanh":     "math.atanh",
+	"math.Atan2":     "math.atan2",
+	"math.Cbrt":      "math.cbrt",
+	"math.Ceil":      "math.ceil",
+	"math.archCeil":  "math.ceil",
+	"math.Cos":       "math.cos",
+	"math.Cosh":      "math.cosh",
+	"math.Erf":       "math.erf",
+	"math.Erfc":      "math.erfc",
+	"math.Exp":       "math.exp",
+	"math.archExp":   "math.exp",
+	"math.Expm1":     "math.expm1",
+	"math.Exp2":      "math.exp2",
+	"math.archExp2":  "math.exp2",
+	"math.Floor":     "math.floor",
+	"math.archFloor": "math.floor",
+	"math.Frexp":     "math.frexp",
+	"math.Hypot":     "math.hypot",
+	"math.archHypot": "math.hypot",
+	"math.Ldexp":     "math.ldexp",
+	"math.Log":       "math.log",
+	"math.archLog":   "math.log",
+	"math.Log1p":     "math.log1p",
+	"math.Log10":     "math.log10",
+	"math.Log2":      "math.log2",
+	"math.Max":       "math.max",
+	"math.archMax":   "math.max",
+	"math.Min":       "math.min",
+	"math.archMin":   "math.min",
+	"math.Mod":       "math.mod",
+	"math.Modf":      "math.modf",
+	"math.archModf":  "math.modf",
+	"math.Pow":       "math.pow",
+	"math.Remainder": "math.remainder",
+	"math.Sin":       "math.sin",
+	"math.Sinh":      "math.sinh",
+	"math.Sqrt":      "math.sqrt",
+	"math.archSqrt":  "math.sqrt",
+	"math.Tan":       "math.tan",
+	"math.Tanh":      "math.tanh",
+	"math.Trunc":     "math.trunc",
+	"math.archTrunc": "math.trunc",
+}
+
+// createAlias implements the function (in the builder) as a call to the alias
+// function.
+func (b *builder) createAlias(alias llvm.Value) {
+	if b.Debug {
+		if b.fn.Syntax() != nil {
+			// Create debug info file if present.
+			b.difunc = b.attachDebugInfo(b.fn)
+		}
+		pos := b.program.Fset.Position(b.fn.Pos())
+		b.SetCurrentDebugLocation(uint(pos.Line), uint(pos.Column), b.difunc, llvm.Metadata{})
+	}
+	entryBlock := llvm.AddBasicBlock(b.llvmFn, "entry")
+	b.SetInsertPointAtEnd(entryBlock)
+	if b.llvmFn.Type() != alias.Type() {
+		b.addError(b.fn.Pos(), "alias function should have the same type as aliasee "+alias.Name())
+		b.CreateUnreachable()
+		return
+	}
+	result := b.CreateCall(alias, b.llvmFn.Params(), "")
+	if result.Type().TypeKind() == llvm.VoidTypeKind {
+		b.CreateRetVoid()
+	} else {
+		b.CreateRet(result)
+	}
+}
@@ -101,6 +101,55 @@ func (b *builder) createSliceBoundsCheck(capacity, low, high, max llvm.Value, lo
 	b.createRuntimeAssert(outOfBounds, "slice", "slicePanic")
 }

+// createSliceToArrayPointerCheck adds a check for slice-to-array pointer
+// conversions. This conversion was added in Go 1.17. For details, see:
+// https://tip.golang.org/ref/spec#Conversions_from_slice_to_array_pointer
+func (b *builder) createSliceToArrayPointerCheck(sliceLen llvm.Value, arrayLen int64) {
+	// From the spec:
+	// > If the length of the slice is less than the length of the array, a
+	// > run-time panic occurs.
+	arrayLenValue := llvm.ConstInt(b.uintptrType, uint64(arrayLen), false)
+	isLess := b.CreateICmp(llvm.IntULT, sliceLen, arrayLenValue, "")
+	b.createRuntimeAssert(isLess, "slicetoarray", "sliceToArrayPointerPanic")
+}
+
+// createUnsafeSliceCheck inserts a runtime check used for unsafe.Slice. This
+// function must panic if the ptr/len parameters are invalid.
+func (b *builder) createUnsafeSliceCheck(ptr, len llvm.Value, lenType *types.Basic) {
+	// From the documentation of unsafe.Slice:
+	//   > At run time, if len is negative, or if ptr is nil and len is not
+	//   > zero, a run-time panic occurs.
+	// However, in practice, it is also necessary to check that the length is
+	// not too big that a GEP wouldn't be possible without wrapping the pointer.
+	// These two checks (non-negative and not too big) can be merged into one
+	// using an unsiged greater than.
+
+	// Make sure the len value is at least as big as a uintptr.
+	if len.Type().IntTypeWidth() < b.uintptrType.IntTypeWidth() {
+		if lenType.Info()&types.IsUnsigned != 0 {
+			len = b.CreateZExt(len, b.uintptrType, "")
+		} else {
+			len = b.CreateSExt(len, b.uintptrType, "")
+		}
+	}
+
+	// Determine the maximum slice size, and therefore the maximum value of the
+	// len parameter.
+	maxSize := b.maxSliceSize(ptr.Type().ElementType())
+	maxSizeValue := llvm.ConstInt(len.Type(), maxSize, false)
+
+	// Do the check. By using unsigned greater than for the length check, signed
+	// negative values are also checked (which are very large numbers when
+	// interpreted as signed values).
+	zero := llvm.ConstInt(len.Type(), 0, false)
+	lenOutOfBounds := b.CreateICmp(llvm.IntUGT, len, maxSizeValue, "")
+	ptrIsNil := b.CreateICmp(llvm.IntEQ, ptr, llvm.ConstNull(ptr.Type()), "")
+	lenIsNotZero := b.CreateICmp(llvm.IntNE, len, zero, "")
+	assert := b.CreateAnd(ptrIsNil, lenIsNotZero, "")
+	assert = b.CreateOr(assert, lenOutOfBounds, "")
+	b.createRuntimeAssert(assert, "unsafe.Slice", "unsafeSlicePanic")
+}
+
 // createChanBoundsCheck creates a bounds check before creating a new channel to
 // check that the value is not too big for runtime.chanMake.
 func (b *builder) createChanBoundsCheck(elementSize uint64, bufSize llvm.Value, bufSizeType *types.Basic, pos token.Pos) {
@@ -162,6 +211,10 @@ func (b *builder) createNilCheck(inst ssa.Value, ptr llvm.Value, blockPrefix str
 	case *ssa.Alloc:
 		// An alloc is never nil.
 		return
+	case *ssa.FreeVar:
+		// A free variable is allocated in a parent function and is thus never
+		// nil.
+		return
 	case *ssa.IndexAddr:
 		// This pointer is the result of an index operation into a slice or
 		// array. Such slices/arrays are already bounds checked so the pointer
@@ -32,8 +32,15 @@ func (b *builder) createChanSend(instr *ssa.Send) {

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

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

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

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

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

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

 	if unop.CommaOk {
 		tuple := llvm.Undef(b.ctx.StructType([]llvm.Type{valueType, b.ctx.Int1Type()}, false))
@@ -116,7 +137,6 @@ func (b *builder) createSelect(expr *ssa.Select) llvm.Value {
 	// determine the receive buffer size and alignment.
 	recvbufSize := uint64(0)
 	recvbufAlign := 0
-	hasReceives := false
 	var selectStates []llvm.Value
 	chanSelectStateType := b.getLLVMRuntimeType("chanSelectState")
 	for _, state := range expr.States {
@@ -133,7 +153,6 @@ func (b *builder) createSelect(expr *ssa.Select) llvm.Value {
 			if align := b.targetData.ABITypeAlignment(llvmType); align > recvbufAlign {
 				recvbufAlign = align
 			}
-			hasReceives = true
 		case types.SendOnly:
 			// Store this value in an alloca and put a pointer to this alloca
 			// in the send state.
@@ -150,7 +169,7 @@ func (b *builder) createSelect(expr *ssa.Select) llvm.Value {

 	// Create a receive buffer, where the received value will be stored.
 	recvbuf := llvm.Undef(b.i8ptrType)
-	if hasReceives {
+	if recvbufSize != 0 {
 		allocaType := llvm.ArrayType(b.ctx.Int8Type(), int(recvbufSize))
 		recvbufAlloca, _, _ := b.createTemporaryAlloca(allocaType, "select.recvbuf.alloca")
 		recvbufAlloca.SetAlignment(recvbufAlign)
@@ -23,7 +23,7 @@ import (
 // Version of the compiler pacakge. Must be incremented each time the compiler
 // package changes in a way that affects the generated LLVM module.
 // This version is independent of the TinyGo version number.
-const Version = 9 // last change: implement reflect.New()
+const Version = 19 // last change: fix channel ops with zero values

 func init() {
 	llvm.InitializeAllTargets()
@@ -59,6 +59,7 @@ type Config struct {
 	DefaultStackSize   uint64
 	NeedsStackObjects  bool
 	Debug              bool // Whether to emit debug information in the LLVM module.
+	LLVMFeatures       string
 }

 // compilerContext contains function-independent data that should still be
@@ -73,6 +74,7 @@ type compilerContext struct {
 	cu               llvm.Metadata
 	difiles          map[string]llvm.Metadata
 	ditypes          map[types.Type]llvm.Metadata
+	llvmTypes        map[types.Type]llvm.Type
 	machine          llvm.TargetMachine
 	targetData       llvm.TargetData
 	intType          llvm.Type
@@ -93,6 +95,7 @@ func newCompilerContext(moduleName string, machine llvm.TargetMachine, config *C
 		DumpSSA:     dumpSSA,
 		difiles:     make(map[string]llvm.Metadata),
 		ditypes:     make(map[types.Type]llvm.Metadata),
+		llvmTypes:   make(map[types.Type]llvm.Type),
 		machine:     machine,
 		targetData:  machine.CreateTargetData(),
 		astComments: map[string]*ast.CommentGroup{},
@@ -185,7 +188,12 @@ func NewTargetMachine(config *Config) (llvm.TargetMachine, error) {
 	if err != nil {
 		return llvm.TargetMachine{}, err
 	}
-	features := strings.Join(config.Features, ",")
+
+	feat := config.Features
+	if len(config.LLVMFeatures) > 0 {
+		feat = append(feat, config.LLVMFeatures)
+	}
+	features := strings.Join(feat, ",")

 	var codeModel llvm.CodeModel
 	var relocationModel llvm.RelocMode
@@ -309,10 +317,23 @@ func (c *compilerContext) getLLVMRuntimeType(name string) llvm.Type {
 	return c.getLLVMType(typ)
 }

-// getLLVMType creates and returns a LLVM type for a Go type. In the case of
-// named struct types (or Go types implemented as named LLVM structs such as
-// strings) it also creates it first if necessary.
+// getLLVMType returns a LLVM type for a Go type. It doesn't recreate already
+// created types. This is somewhat important for performance, but especially
+// important for named struct types (which should only be created once).
 func (c *compilerContext) getLLVMType(goType types.Type) llvm.Type {
+	// Try to load the LLVM type from the cache.
+	if t, ok := c.llvmTypes[goType]; ok {
+		return t
+	}
+	// Not already created, so adding this type to the cache.
+	llvmType := c.makeLLVMType(goType)
+	c.llvmTypes[goType] = llvmType
+	return llvmType
+}
+
+// makeLLVMType creates a LLVM type for a Go type. Don't call this, use
+// getLLVMType instead.
+func (c *compilerContext) makeLLVMType(goType types.Type) llvm.Type {
 	switch typ := goType.(type) {
 	case *types.Array:
 		elemType := c.getLLVMType(typ.Elem())
@@ -361,12 +382,10 @@ func (c *compilerContext) getLLVMType(goType types.Type) llvm.Type {
 			// LLVM. This is because it is otherwise impossible to create
 			// self-referencing types such as linked lists.
 			llvmName := typ.Obj().Pkg().Path() + "." + typ.Obj().Name()
-			llvmType := c.mod.GetTypeByName(llvmName)
-			if llvmType.IsNil() {
-				llvmType = c.ctx.StructCreateNamed(llvmName)
-				underlying := c.getLLVMType(st)
-				llvmType.StructSetBody(underlying.StructElementTypes(), false)
-			}
+			llvmType := c.ctx.StructCreateNamed(llvmName)
+			c.llvmTypes[goType] = llvmType // avoid infinite recursion
+			underlying := c.getLLVMType(st)
+			llvmType.StructSetBody(underlying.StructElementTypes(), false)
 			return llvmType
 		}
 		return c.getLLVMType(typ.Underlying())
@@ -709,11 +728,11 @@ func (c *compilerContext) createPackage(irbuilder llvm.Builder, pkg *ssa.Package
 		member := pkg.Members[name]
 		switch member := member.(type) {
 		case *ssa.Function:
+			// Create the function definition.
+			b := newBuilder(c, irbuilder, member)
 			if member.Blocks == nil {
 				continue // external function
 			}
-			// Create the function definition.
-			b := newBuilder(c, irbuilder, member)
 			b.createFunction()
 		case *ssa.Type:
 			if types.IsInterface(member.Type()) {
@@ -752,10 +771,36 @@ func (c *compilerContext) createPackage(irbuilder llvm.Builder, pkg *ssa.Package
 		case *ssa.Global:
 			// Global variable.
 			info := c.getGlobalInfo(member)
+			global := c.getGlobal(member)
 			if !info.extern {
-				global := c.getGlobal(member)
 				global.SetInitializer(llvm.ConstNull(global.Type().ElementType()))
 				global.SetVisibility(llvm.HiddenVisibility)
+				if info.section != "" {
+					global.SetSection(info.section)
+				}
+			}
+		}
+	}
+
+	// Add forwarding functions for functions that would otherwise be
+	// implemented in assembly.
+	for _, name := range members {
+		member := pkg.Members[name]
+		switch member := member.(type) {
+		case *ssa.Function:
+			if member.Blocks != nil {
+				continue // external function
+			}
+			info := c.getFunctionInfo(member)
+			if aliasName, ok := stdlibAliases[info.linkName]; ok {
+				alias := c.mod.NamedFunction(aliasName)
+				if alias.IsNil() {
+					// Shouldn't happen, but perhaps best to just ignore.
+					// The error will be a link error, if there is an error.
+					continue
+				}
+				b := newBuilder(c, irbuilder, member)
+				b.createAlias(alias)
 			}
 		}
 	}
@@ -781,6 +826,9 @@ func (b *builder) createFunction() {
 		b.llvmFn.SetVisibility(llvm.HiddenVisibility)
 		b.llvmFn.SetUnnamedAddr(true)
 	}
+	if b.info.section != "" {
+		b.llvmFn.SetSection(b.info.section)
+	}
 	if b.info.exported && strings.HasPrefix(b.Triple, "wasm") {
 		// Set the exported name. This is necessary for WebAssembly because
 		// otherwise the function is not exported.
@@ -964,11 +1012,61 @@ func (b *builder) createFunction() {
 	}
 }

+// posser is an interface that's implemented by both ssa.Value and
+// ssa.Instruction. It is implemented by everything that has a Pos() method,
+// which is all that getPos() needs.
+type posser interface {
+	Pos() token.Pos
+}
+
+// getPos returns position information for a ssa.Value or ssa.Instruction.
+//
+// Not all instructions have position information, especially when they're
+// implicit (such as implicit casts or implicit returns at the end of a
+// function). In these cases, it makes sense to try a bit harder to guess what
+// the position really should be.
+func getPos(val posser) token.Pos {
+	pos := val.Pos()
+	if pos != token.NoPos {
+		// Easy: position is known.
+		return pos
+	}
+
+	// No position information is known.
+	switch val := val.(type) {
+	case *ssa.MakeInterface:
+		return getPos(val.X)
+	case *ssa.MakeClosure:
+		return val.Fn.(*ssa.Function).Pos()
+	case *ssa.Return:
+		syntax := val.Parent().Syntax()
+		if syntax != nil {
+			// non-synthetic
+			return syntax.End()
+		}
+		return token.NoPos
+	case *ssa.FieldAddr:
+		return getPos(val.X)
+	case *ssa.IndexAddr:
+		return getPos(val.X)
+	case *ssa.Slice:
+		return getPos(val.X)
+	case *ssa.Store:
+		return getPos(val.Addr)
+	case *ssa.Extract:
+		return getPos(val.Tuple)
+	default:
+		// This is reachable, for example with *ssa.Const, *ssa.If, and
+		// *ssa.Jump. They might be implemented in some way in the future.
+		return token.NoPos
+	}
+}
+
 // createInstruction builds the LLVM IR equivalent instructions for the
 // particular Go SSA instruction.
 func (b *builder) createInstruction(instr ssa.Instruction) {
 	if b.Debug {
-		pos := b.program.Fset.Position(instr.Pos())
+		pos := b.program.Fset.Position(getPos(instr))
 		b.SetCurrentDebugLocation(uint(pos.Line), uint(pos.Column), b.difunc, llvm.Metadata{})
 	}

@@ -994,52 +1092,8 @@ func (b *builder) createInstruction(instr ssa.Instruction) {
 	case *ssa.Defer:
 		b.createDefer(instr)
 	case *ssa.Go:
-		// Get all function parameters to pass to the goroutine.
-		var params []llvm.Value
-		for _, param := range instr.Call.Args {
-			params = append(params, b.getValue(param))
-		}
-
 		// Start a new goroutine.
-		if callee := instr.Call.StaticCallee(); callee != nil {
-			// Static callee is known. This makes it easier to start a new
-			// goroutine.
-			var context llvm.Value
-			switch value := instr.Call.Value.(type) {
-			case *ssa.Function:
-				// Goroutine call is regular function call. No context is necessary.
-				context = llvm.Undef(b.i8ptrType)
-			case *ssa.MakeClosure:
-				// A goroutine call on a func value, but the callee is trivial to find. For
-				// example: immediately applied functions.
-				funcValue := b.getValue(value)
-				context = b.extractFuncContext(funcValue)
-			default:
-				panic("StaticCallee returned an unexpected value")
-			}
-			params = append(params, context) // context parameter
-			b.createGoInstruction(b.getFunction(callee), params, "", callee.Pos())
-		} else if !instr.Call.IsInvoke() {
-			// This is a function pointer.
-			// At the moment, two extra params are passed to the newly started
-			// goroutine:
-			//   * The function context, for closures.
-			//   * The function pointer (for tasks).
-			funcPtr, context := b.decodeFuncValue(b.getValue(instr.Call.Value), instr.Call.Value.Type().Underlying().(*types.Signature))
-			params = append(params, context) // context parameter
-			switch b.Scheduler {
-			case "none", "coroutines":
-				// There are no additional parameters needed for the goroutine start operation.
-			case "tasks":
-				// Add the function pointer as a parameter to start the goroutine.
-				params = append(params, funcPtr)
-			default:
-				panic("unknown scheduler type")
-			}
-			b.createGoInstruction(funcPtr, params, b.fn.RelString(nil), instr.Pos())
-		} else {
-			b.addError(instr.Pos(), "todo: go on interface call")
-		}
+		b.createGo(instr)
 	case *ssa.If:
 		cond := b.getValue(instr.Cond)
 		block := instr.Block()
@@ -1252,6 +1306,38 @@ func (b *builder) createBuiltin(argTypes []types.Type, argValues []llvm.Value, c
 	case "ssa:wrapnilchk":
 		// TODO: do an actual nil check?
 		return argValues[0], nil
+
+	// Builtins from the unsafe package.
+	case "Add": // unsafe.Add
+		// This is basically just a GEP operation.
+		// Note: the pointer is always of type *i8.
+		ptr := argValues[0]
+		len := argValues[1]
+		return b.CreateGEP(ptr, []llvm.Value{len}, ""), nil
+	case "Slice": // unsafe.Slice
+		// This creates a slice from a pointer and a length.
+		// Note that the exception mentioned in the documentation (if the
+		// pointer and length are nil, the slice is also nil) is trivially
+		// already the case.
+		ptr := argValues[0]
+		len := argValues[1]
+		slice := llvm.Undef(b.ctx.StructType([]llvm.Type{
+			ptr.Type(),
+			b.uintptrType,
+			b.uintptrType,
+		}, false))
+		b.createUnsafeSliceCheck(ptr, len, argTypes[1].Underlying().(*types.Basic))
+		if len.Type().IntTypeWidth() < b.uintptrType.IntTypeWidth() {
+			// Too small, zero-extend len.
+			len = b.CreateZExt(len, b.uintptrType, "")
+		} else if len.Type().IntTypeWidth() > b.uintptrType.IntTypeWidth() {
+			// Too big, truncate len.
+			len = b.CreateTrunc(len, b.uintptrType, "")
+		}
+		slice = b.CreateInsertValue(slice, ptr, 0, "")
+		slice = b.CreateInsertValue(slice, len, 1, "")
+		slice = b.CreateInsertValue(slice, len, 2, "")
+		return slice, nil
 	default:
 		return llvm.Value{}, b.makeError(pos, "todo: builtin: "+callName)
 	}
@@ -1280,6 +1366,11 @@ func (b *builder) createFunctionCall(instr *ssa.CallCommon) (llvm.Value, error)
 			return b.createMemoryCopyCall(fn, instr.Args)
 		case name == "runtime.memzero":
 			return b.createMemoryZeroCall(instr.Args)
+		case name == "math.Ceil" || name == "math.Floor" || name == "math.Sqrt" || name == "math.Trunc":
+			result, ok := b.createMathOp(instr)
+			if ok {
+				return result, nil
+			}
 		case name == "device.Asm" || name == "device/arm.Asm" || name == "device/arm64.Asm" || name == "device/avr.Asm" || name == "device/riscv.Asm":
 			return b.createInlineAsm(instr.Args)
 		case name == "device.AsmFull" || name == "device/arm.AsmFull" || name == "device/arm64.AsmFull" || name == "device/avr.AsmFull" || name == "device/riscv.AsmFull":
@@ -1292,6 +1383,8 @@ func (b *builder) createFunctionCall(instr *ssa.CallCommon) (llvm.Value, error)
 			return b.emitCSROperation(instr)
 		case strings.HasPrefix(name, "syscall.Syscall"):
 			return b.createSyscall(instr)
+		case strings.HasPrefix(name, "syscall.rawSyscallNoError"):
+			return b.createRawSyscallNoError(instr)
 		case strings.HasPrefix(name, "runtime/volatile.Load"):
 			return b.createVolatileLoad(instr)
 		case strings.HasPrefix(name, "runtime/volatile.Store"):
@@ -1391,6 +1484,28 @@ func (b *builder) getValue(expr ssa.Value) llvm.Value {
 	}
 }

+// maxSliceSize determines the maximum size a slice of the given element type
+// can be.
+func (c *compilerContext) maxSliceSize(elementType llvm.Type) uint64 {
+	// Calculate ^uintptr(0), which is the max value that fits in uintptr.
+	maxPointerValue := llvm.ConstNot(llvm.ConstInt(c.uintptrType, 0, false)).ZExtValue()
+	// Calculate (^uint(0))/2, which is the max value that fits in an int.
+	maxIntegerValue := llvm.ConstNot(llvm.ConstInt(c.intType, 0, false)).ZExtValue() / 2
+
+	// Determine the maximum allowed size for a slice. The biggest possible
+	// pointer (starting from 0) would be maxPointerValue*sizeof(elementType) so
+	// divide by the element type to get the real maximum size.
+	maxSize := maxPointerValue / c.targetData.TypeAllocSize(elementType)
+
+	// len(slice) is an int. Make sure the length remains small enough to fit in
+	// an int.
+	if maxSize > maxIntegerValue {
+		maxSize = maxIntegerValue
+	}
+
+	return maxSize
+}
+
 // createExpr translates a Go SSA expression to LLVM IR. This can be zero, one,
 // or multiple LLVM IR instructions and/or runtime calls.
 func (b *builder) createExpr(expr ssa.Value) (llvm.Value, error) {
@@ -1604,10 +1719,8 @@ func (b *builder) createExpr(expr ssa.Value) (llvm.Value, error) {
 		elemSize := b.targetData.TypeAllocSize(llvmElemType)
 		elemSizeValue := llvm.ConstInt(b.uintptrType, elemSize, false)

-		// Calculate (^uintptr(0)) >> 1, which is the max value that fits in
-		// uintptr if uintptr were signed.
-		maxSize := llvm.ConstLShr(llvm.ConstNot(llvm.ConstInt(b.uintptrType, 0, false)), llvm.ConstInt(b.uintptrType, 1, false))
-		if elemSize > maxSize.ZExtValue() {
+		maxSize := b.maxSliceSize(llvmElemType)
+		if elemSize > maxSize {
 			// This seems to be checked by the typechecker already, but let's
 			// check it again just to be sure.
 			return llvm.Value{}, b.makeError(expr.Pos(), fmt.Sprintf("slice element type is too big (%v bytes)", elemSize))
@@ -1616,7 +1729,8 @@ func (b *builder) createExpr(expr ssa.Value) (llvm.Value, error) {
 		// Bounds checking.
 		lenType := expr.Len.Type().Underlying().(*types.Basic)
 		capType := expr.Cap.Type().Underlying().(*types.Basic)
-		b.createSliceBoundsCheck(maxSize, sliceLen, sliceCap, sliceCap, lenType, capType, capType)
+		maxSizeValue := llvm.ConstInt(b.uintptrType, maxSize, false)
+		b.createSliceBoundsCheck(maxSizeValue, sliceLen, sliceCap, sliceCap, lenType, capType, capType)

 		// Allocate the backing array.
 		sliceCapCast, err := b.createConvert(expr.Cap.Type(), types.Typ[types.Uintptr], sliceCap, expr.Pos())
@@ -1859,6 +1973,17 @@ func (b *builder) createExpr(expr ssa.Value) (llvm.Value, error) {
 		default:
 			return llvm.Value{}, b.makeError(expr.Pos(), "unknown slice type: "+typ.String())
 		}
+	case *ssa.SliceToArrayPointer:
+		// Conversion from a slice to an array pointer, as the name clearly
+		// says. This requires a runtime check to make sure the slice is at
+		// least as big as the array.
+		slice := b.getValue(expr.X)
+		sliceLen := b.CreateExtractValue(slice, 1, "")
+		arrayLen := expr.Type().Underlying().(*types.Pointer).Elem().Underlying().(*types.Array).Len()
+		b.createSliceToArrayPointerCheck(sliceLen, arrayLen)
+		ptr := b.CreateExtractValue(slice, 0, "")
+		ptr = b.CreateBitCast(ptr, b.getLLVMType(expr.Type()), "")
+		return ptr, nil
 	case *ssa.TypeAssert:
 		return b.createTypeAssert(expr), nil
 	case *ssa.UnOp:
@@ -9,6 +9,7 @@ import (
 	"testing"

 	"github.com/tinygo-org/tinygo/compileopts"
+	"github.com/tinygo-org/tinygo/goenv"
 	"github.com/tinygo-org/tinygo/loader"
 	"tinygo.org/x/go-llvm"
 )
@@ -16,9 +17,16 @@ import (
 // Pass -update to go test to update the output of the test files.
 var flagUpdate = flag.Bool("update", false, "update tests based on test output")

+type testCase struct {
+	file   string
+	target string
+}
+
 // Basic tests for the compiler. Build some Go files and compare the output with
 // the expected LLVM IR for regression testing.
 func TestCompiler(t *testing.T) {
+	t.Parallel()
+
 	// Check LLVM version.
 	llvmMajor, err := strconv.Atoi(strings.SplitN(llvm.Version, ".", 2)[0])
 	if err != nil {
@@ -32,43 +40,64 @@ func TestCompiler(t *testing.T) {
 		t.Skip("compiler tests require LLVM 11 or above, got LLVM ", llvm.Version)
 	}

-	target, err := compileopts.LoadTarget("i686--linux")
-	if err != nil {
-		t.Fatal("failed to load target:", err)
-	}
-	config := &compileopts.Config{
-		Options: &compileopts.Options{},
-		Target:  target,
-	}
-	compilerConfig := &Config{
-		Triple:             config.Triple(),
-		GOOS:               config.GOOS(),
-		GOARCH:             config.GOARCH(),
-		CodeModel:          config.CodeModel(),
-		RelocationModel:    config.RelocationModel(),
-		Scheduler:          config.Scheduler(),
-		FuncImplementation: config.FuncImplementation(),
-		AutomaticStackSize: config.AutomaticStackSize(),
-	}
-	machine, err := NewTargetMachine(compilerConfig)
-	if err != nil {
-		t.Fatal("failed to create target machine:", err)
+	tests := []testCase{
+		{"basic.go", ""},
+		{"pointer.go", ""},
+		{"slice.go", ""},
+		{"string.go", ""},
+		{"float.go", ""},
+		{"interface.go", ""},
+		{"func.go", ""},
+		{"pragma.go", ""},
+		{"goroutine.go", "wasm"},
+		{"goroutine.go", "cortex-m-qemu"},
+		{"channel.go", ""},
+		{"intrinsics.go", "cortex-m-qemu"},
+		{"intrinsics.go", "wasm"},
 	}

-	tests := []string{
-		"basic.go",
-		"pointer.go",
-		"slice.go",
-		"string.go",
-		"float.go",
-		"interface.go",
-		"func.go",
+	_, minor, err := goenv.GetGorootVersion(goenv.Get("GOROOT"))
+	if err != nil {
+		t.Fatal("could not read Go version:", err)
+	}
+	if minor >= 17 {
+		tests = append(tests, testCase{"go1.17.go", ""})
 	}

-	for _, testCase := range tests {
-		t.Run(testCase, func(t *testing.T) {
+	for _, tc := range tests {
+		name := tc.file
+		targetString := "wasm"
+		if tc.target != "" {
+			targetString = tc.target
+			name = tc.file + "-" + tc.target
+		}
+
+		t.Run(name, func(t *testing.T) {
+			target, err := compileopts.LoadTarget(targetString)
+			if err != nil {
+				t.Fatal("failed to load target:", err)
+			}
+			config := &compileopts.Config{
+				Options: &compileopts.Options{},
+				Target:  target,
+			}
+			compilerConfig := &Config{
+				Triple:             config.Triple(),
+				GOOS:               config.GOOS(),
+				GOARCH:             config.GOARCH(),
+				CodeModel:          config.CodeModel(),
+				RelocationModel:    config.RelocationModel(),
+				Scheduler:          config.Scheduler(),
+				FuncImplementation: config.FuncImplementation(),
+				AutomaticStackSize: config.AutomaticStackSize(),
+			}
+			machine, err := NewTargetMachine(compilerConfig)
+			if err != nil {
+				t.Fatal("failed to create target machine:", err)
+			}
+
 			// Load entire program AST into memory.
-			lprogram, err := loader.Load(config, []string{"./testdata/" + testCase}, config.ClangHeaders, types.Config{
+			lprogram, err := loader.Load(config, []string{"./testdata/" + tc.file}, config.ClangHeaders, types.Config{
 				Sizes: Sizes(machine),
 			})
 			if err != nil {
@@ -76,13 +105,13 @@ func TestCompiler(t *testing.T) {
 			}
 			err = lprogram.Parse()
 			if err != nil {
-				t.Fatalf("could not parse test case %s: %s", testCase, err)
+				t.Fatalf("could not parse test case %s: %s", tc.file, err)
 			}

 			// Compile AST to IR.
 			program := lprogram.LoadSSA()
 			pkg := lprogram.MainPkg()
-			mod, errs := CompilePackage(testCase, pkg, program.Package(pkg.Pkg), machine, compilerConfig, false)
+			mod, errs := CompilePackage(tc.file, pkg, program.Package(pkg.Pkg), machine, compilerConfig, false)
 			if errs != nil {
 				for _, err := range errs {
 					t.Error(err)
@@ -105,18 +134,22 @@ func TestCompiler(t *testing.T) {
 			}
 			funcPasses.FinalizeFunc()

-			outfile := "./testdata/" + testCase[:len(testCase)-3] + ".ll"
+			outFilePrefix := tc.file[:len(tc.file)-3]
+			if tc.target != "" {
+				outFilePrefix += "-" + tc.target
+			}
+			outPath := "./testdata/" + outFilePrefix + ".ll"

 			// Update test if needed. Do not check the result.
 			if *flagUpdate {
-				err := ioutil.WriteFile(outfile, []byte(mod.String()), 0666)
+				err := ioutil.WriteFile(outPath, []byte(mod.String()), 0666)
 				if err != nil {
 					t.Error("failed to write updated output file:", err)
 				}
 				return
 			}

-			expected, err := ioutil.ReadFile(outfile)
+			expected, err := ioutil.ReadFile(outPath)
 			if err != nil {
 				t.Fatal("failed to read golden file:", err)
 			}
@@ -5,25 +5,110 @@ package compiler

 import (
 	"go/token"
+	"go/types"

 	"github.com/tinygo-org/tinygo/compiler/llvmutil"
 	"golang.org/x/tools/go/ssa"
 	"tinygo.org/x/go-llvm"
 )

-// createGoInstruction starts a new goroutine with the provided function pointer
-// and parameters.
-// In general, you should pass all regular parameters plus the context parameter.
-// There is one exception: the task-based scheduler needs to have the function
-// pointer passed in as a parameter too in addition to the context.
-//
-// 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 {
+// createGo emits code to start a new goroutine.
+func (b *builder) createGo(instr *ssa.Go) {
+	// Get all function parameters to pass to the goroutine.
+	var params []llvm.Value
+	for _, param := range instr.Call.Args {
+		params = append(params, b.getValue(param))
+	}
+
+	var prefix string
+	var funcPtr llvm.Value
+	hasContext := false
+	if callee := instr.Call.StaticCallee(); callee != nil {
+		// Static callee is known. This makes it easier to start a new
+		// goroutine.
+		var context llvm.Value
+		switch value := instr.Call.Value.(type) {
+		case *ssa.Function:
+			// Goroutine call is regular function call. No context is necessary.
+			if b.Scheduler == "coroutines" {
+				// The context parameter is assumed to be always present in the
+				// coroutines scheduler.
+				context = llvm.Undef(b.i8ptrType)
+			}
+		case *ssa.MakeClosure:
+			// A goroutine call on a func value, but the callee is trivial to find. For
+			// example: immediately applied functions.
+			funcValue := b.getValue(value)
+			context = b.extractFuncContext(funcValue)
+		default:
+			panic("StaticCallee returned an unexpected value")
+		}
+		if !context.IsNil() {
+			params = append(params, context) // context parameter
+			hasContext = true
+		}
+		funcPtr = b.getFunction(callee)
+	} else if builtin, ok := instr.Call.Value.(*ssa.Builtin); ok {
+		// We cheat. None of the builtins do any long or blocking operation, so
+		// we might as well run these builtins right away without the program
+		// noticing the difference.
+		// Possible exceptions:
+		//   - copy: this is a possibly long operation, but not a blocking
+		//     operation. Semantically it makes no difference to run it right
+		//     away (not in a goroutine). However, in practice it makes no sense
+		//     to run copy in a goroutine as there is no way to (safely) know
+		//     when it is finished.
+		//   - panic: the error message would appear in the parent goroutine.
+		//     But because `go panic("err")` would halt the program anyway
+		//     (there is no recover), panicking right away would give the same
+		//     behavior as creating a goroutine, switching the scheduler to that
+		//     goroutine, and panicking there. So this optimization seems
+		//     correct.
+		//   - recover: because it runs in a new goroutine, it is never a
+		//     deferred function. Thus this is a no-op.
+		if builtin.Name() == "recover" {
+			// This is a no-op, even in a deferred function:
+			//   go recover()
+			return
+		}
+		var argTypes []types.Type
+		var argValues []llvm.Value
+		for _, arg := range instr.Call.Args {
+			argTypes = append(argTypes, arg.Type())
+			argValues = append(argValues, b.getValue(arg))
+		}
+		b.createBuiltin(argTypes, argValues, builtin.Name(), instr.Pos())
+		return
+	} else if !instr.Call.IsInvoke() {
+		// This is a function pointer.
+		// At the moment, two extra params are passed to the newly started
+		// goroutine:
+		//   * The function context, for closures.
+		//   * The function pointer (for tasks).
+		var context llvm.Value
+		funcPtr, context = b.decodeFuncValue(b.getValue(instr.Call.Value), instr.Call.Value.Type().Underlying().(*types.Signature))
+		params = append(params, context) // context parameter
+		hasContext = true
+		switch b.Scheduler {
+		case "none", "coroutines":
+			// There are no additional parameters needed for the goroutine start operation.
+		case "tasks":
+			// Add the function pointer as a parameter to start the goroutine.
+			params = append(params, funcPtr)
+		default:
+			panic("unknown scheduler type")
+		}
+		prefix = b.fn.RelString(nil)
+	} else {
+		b.addError(instr.Pos(), "todo: go on interface call")
+		return
+	}
+
 	paramBundle := b.emitPointerPack(params)
 	var callee, stackSize llvm.Value
 	switch b.Scheduler {
 	case "none", "tasks":
-		callee = b.createGoroutineStartWrapper(funcPtr, prefix, pos)
+		callee = b.createGoroutineStartWrapper(funcPtr, prefix, hasContext, instr.Pos())
 		if b.AutomaticStackSize {
 			// The stack size is not known until after linking. Call a dummy
 			// function that will be replaced with a load from a special ELF
@@ -34,6 +119,9 @@ func (b *builder) createGoInstruction(funcPtr llvm.Value, params []llvm.Value, p
 		} else {
 			// The stack size is fixed at compile time. By emitting it here as a
 			// constant, it can be optimized.
+			if b.Scheduler == "tasks" && b.DefaultStackSize == 0 {
+				b.addError(instr.Pos(), "default stack size for goroutines is not set")
+			}
 			stackSize = llvm.ConstInt(b.uintptrType, b.DefaultStackSize, false)
 		}
 	case "coroutines":
@@ -46,7 +134,6 @@ func (b *builder) createGoInstruction(funcPtr llvm.Value, params []llvm.Value, p
 	}
 	start := b.getFunction(b.program.ImportedPackage("internal/task").Members["start"].(*ssa.Function))
 	b.createCall(start, []llvm.Value{callee, paramBundle, stackSize, llvm.Undef(b.i8ptrType), llvm.ConstPointerNull(b.i8ptrType)}, "")
-	return llvm.Undef(funcPtr.Type().ElementType().ReturnType())
 }

 // createGoroutineStartWrapper creates a wrapper for the task-based
@@ -67,7 +154,12 @@ func (b *builder) createGoInstruction(funcPtr llvm.Value, params []llvm.Value, p
 // allows a single (pointer) argument to the newly started goroutine. Also, it
 // ignores the return value because newly started goroutines do not have a
 // return value.
-func (c *compilerContext) createGoroutineStartWrapper(fn llvm.Value, prefix string, pos token.Pos) llvm.Value {
+//
+// The hasContext parameter indicates whether the context parameter (the second
+// to last parameter of the function) is used for this wrapper. If hasContext is
+// false, the parameter bundle is assumed to have no context parameter and undef
+// is passed instead.
+func (c *compilerContext) createGoroutineStartWrapper(fn llvm.Value, prefix string, hasContext bool, pos token.Pos) llvm.Value {
 	var wrapper llvm.Value

 	builder := c.ctx.NewBuilder()
@@ -114,8 +206,15 @@ func (c *compilerContext) createGoroutineStartWrapper(fn llvm.Value, prefix stri

 		// Create the list of params for the call.
 		paramTypes := fn.Type().ElementType().ParamTypes()
-		params := llvmutil.EmitPointerUnpack(builder, c.mod, wrapper.Param(0), paramTypes[:len(paramTypes)-1])
-		params = append(params, llvm.Undef(c.i8ptrType))
+		paramTypes = paramTypes[:len(paramTypes)-1] // strip parentHandle parameter
+		if !hasContext {
+			paramTypes = paramTypes[:len(paramTypes)-1] // strip context parameter
+		}
+		params := llvmutil.EmitPointerUnpack(builder, c.mod, wrapper.Param(0), paramTypes)
+		if !hasContext {
+			params = append(params, llvm.Undef(c.i8ptrType)) // add dummy context parameter
+		}
+		params = append(params, llvm.Undef(c.i8ptrType)) // add dummy parentHandle parameter

 		// Create the call.
 		builder.CreateCall(fn, params, "")
@@ -177,8 +276,11 @@ func (c *compilerContext) createGoroutineStartWrapper(fn llvm.Value, prefix stri
 		// Get the function pointer.
 		fnPtr := params[len(params)-1]

-		// Ignore the last param, which isn't used anymore.
-		// TODO: avoid this extra "parent handle" parameter in most functions.
+		// The last parameter in the packed object has somewhat of a dual role.
+		// Inside the parameter bundle it's the function pointer, stored right
+		// after the context pointer. But in the IR call instruction, it's the
+		// parentHandle function that's always undef outside of the coroutines
+		// scheduler. Thus, make the parameter undef here.
 		params[len(params)-1] = llvm.Undef(c.i8ptrType)

 		// Create the call.
@@ -311,10 +311,21 @@ func (c *compilerContext) getInterfaceMethodSet(typ types.Type) llvm.Value {
 // used during the interface lowering pass.
 func (c *compilerContext) getMethodSignature(method *types.Func) llvm.Value {
 	signature := methodSignature(method)
-	signatureGlobal := c.mod.NamedGlobal("func " + signature)
+	var globalName string
+	if token.IsExported(method.Name()) {
+		globalName = "reflect/methods." + signature
+	} else {
+		globalName = method.Type().(*types.Signature).Recv().Pkg().Path() + ".$methods." + signature
+	}
+	signatureGlobal := c.mod.NamedGlobal(globalName)
 	if signatureGlobal.IsNil() {
-		signatureGlobal = llvm.AddGlobal(c.mod, c.ctx.Int8Type(), "func "+signature)
+		// TODO: put something useful in these globals, such as the method
+		// signature. Useful to one day implement reflect.Value.Method(n).
+		signatureGlobal = llvm.AddGlobal(c.mod, c.ctx.Int8Type(), globalName)
+		signatureGlobal.SetInitializer(llvm.ConstInt(c.ctx.Int8Type(), 0, false))
+		signatureGlobal.SetLinkage(llvm.LinkOnceODRLinkage)
 		signatureGlobal.SetGlobalConstant(true)
+		signatureGlobal.SetAlignment(1)
 	}
 	return signatureGlobal
 }
@@ -41,10 +41,7 @@ func (b *builder) createInterruptGlobal(instr *ssa.CallCommon) (llvm.Value, erro
 	// type are lowered in the interrupt lowering pass.
 	globalType := b.program.ImportedPackage("runtime/interrupt").Type("handle").Type()
 	globalLLVMType := b.getLLVMType(globalType)
-	globalName := "runtime/interrupt.$interrupt" + strconv.FormatInt(id.Int64(), 10)
-	if global := b.mod.NamedGlobal(globalName); !global.IsNil() {
-		return llvm.Value{}, b.makeError(instr.Pos(), "interrupt redeclared in this program")
-	}
+	globalName := b.fn.Package().Pkg.Path() + "$interrupt" + strconv.FormatInt(id.Int64(), 10)
 	global := llvm.AddGlobal(b.mod, globalLLVMType, globalName)
 	global.SetVisibility(llvm.HiddenVisibility)
 	global.SetGlobalConstant(true)
@@ -48,3 +48,57 @@ func (b *builder) createMemoryZeroCall(args []ssa.Value) (llvm.Value, error) {
 	b.CreateCall(llvmFn, params, "")
 	return llvm.Value{}, nil
 }
+
+var mathToLLVMMapping = map[string]string{
+	"math.Sqrt":  "llvm.sqrt.f64",
+	"math.Floor": "llvm.floor.f64",
+	"math.Ceil":  "llvm.ceil.f64",
+	"math.Trunc": "llvm.trunc.f64",
+}
+
+// createMathOp tries to lower the given call as a LLVM math intrinsic, if
+// possible. It returns the call result if possible, and a boolean whether it
+// succeeded. If it doesn't succeed, the architecture doesn't support the given
+// intrinsic.
+func (b *builder) createMathOp(call *ssa.CallCommon) (llvm.Value, bool) {
+	// Check whether this intrinsic is supported on the given GOARCH.
+	// If it is unsupported, this can have two reasons:
+	//
+	//  1. LLVM can expand the intrinsic inline (using float instructions), but
+	//     the result doesn't pass the tests of the math package.
+	//  2. LLVM cannot expand the intrinsic inline, will therefore lower it as a
+	//     libm function call, but the libm function call also fails the math
+	//     package tests.
+	//
+	// Whatever the implementation, it must pass the tests in the math package
+	// so unfortunately only the below intrinsic+architecture combinations are
+	// supported.
+	name := call.StaticCallee().RelString(nil)
+	switch name {
+	case "math.Ceil", "math.Floor", "math.Trunc":
+		if b.GOARCH != "wasm" && b.GOARCH != "arm64" {
+			return llvm.Value{}, false
+		}
+	case "math.Sqrt":
+		if b.GOARCH != "wasm" && b.GOARCH != "amd64" && b.GOARCH != "386" {
+			return llvm.Value{}, false
+		}
+	default:
+		return llvm.Value{}, false // only the above functions are supported.
+	}
+
+	llvmFn := b.mod.NamedFunction(mathToLLVMMapping[name])
+	if llvmFn.IsNil() {
+		// The intrinsic doesn't exist yet, so declare it.
+		// At the moment, all supported intrinsics have the form "double
+		// foo(double %x)" so we can hardcode the signature here.
+		llvmType := llvm.FunctionType(b.ctx.DoubleType(), []llvm.Type{b.ctx.DoubleType()}, false)
+		llvmFn = llvm.AddFunction(b.mod, mathToLLVMMapping[name], llvmType)
+	}
+	// Create a call to the intrinsic.
+	args := make([]llvm.Value, len(call.Args))
+	for i, arg := range call.Args {
+		args[i] = b.getValue(arg)
+	}
+	return b.CreateCall(llvmFn, args, ""), true
+}
@@ -24,6 +24,7 @@ type functionInfo struct {
 	module     string     // go:wasm-module
 	importName string     // go:linkname, go:export - The name the developer assigns
 	linkName   string     // go:linkname, go:export - The name that we map for the particular module -> importName
+	section    string     // go:section - object file section name
 	exported   bool       // go:export, CGo
 	nobounds   bool       // go:nobounds
 	variadic   bool       // go:variadic (CGo only)
@@ -270,6 +271,10 @@ func (info *functionInfo) parsePragmas(f *ssa.Function) {
 				if hasUnsafeImport(f.Pkg.Pkg) {
 					info.linkName = parts[2]
 				}
+			case "//go:section":
+				if len(parts) == 2 && hasUnsafeImport(f.Pkg.Pkg) {
+					info.section = parts[1]
+				}
 			case "//go:nobounds":
 				// Skip bounds checking in this function. Useful for some
 				// runtime functions.
@@ -325,6 +330,7 @@ type globalInfo struct {
 	linkName string // go:extern
 	extern   bool   // go:extern
 	align    int    // go:align
+	section  string // go:section
 }

 // loadASTComments loads comments on globals from the AST, for use later in the
@@ -438,6 +444,10 @@ func (info *globalInfo) parsePragmas(doc *ast.CommentGroup) {
 			if err == nil {
 				info.align = align
 			}
+		case "//go:section":
+			if len(parts) == 2 {
+				info.section = parts[1]
+			}
 		}
 	}
 }
@@ -10,11 +10,11 @@ import (
 	"tinygo.org/x/go-llvm"
 )

-// createSyscall emits an inline system call instruction, depending on the
-// target OS/arch.
-func (b *builder) createSyscall(call *ssa.CallCommon) (llvm.Value, error) {
+// createRawSyscall creates a system call with the provided system call number
+// and returns the result as a single integer (the system call result). The
+// result is not further interpreted.
+func (b *builder) createRawSyscall(call *ssa.CallCommon) (llvm.Value, error) {
 	num := b.getValue(call.Args[0])
-	var syscallResult llvm.Value
 	switch {
 	case b.GOARCH == "amd64":
 		if b.GOOS == "darwin" {
@@ -57,7 +57,7 @@ func (b *builder) createSyscall(call *ssa.CallCommon) (llvm.Value, error) {
 		constraints += ",~{rcx},~{r11}"
 		fnType := llvm.FunctionType(b.uintptrType, argTypes, false)
 		target := llvm.InlineAsm(fnType, "syscall", constraints, true, false, llvm.InlineAsmDialectIntel)
-		syscallResult = b.CreateCall(target, args, "")
+		return b.CreateCall(target, args, ""), nil
 	case b.GOARCH == "386" && b.GOOS == "linux":
 		// Sources:
 		//   syscall(2) man page
@@ -83,7 +83,7 @@ func (b *builder) createSyscall(call *ssa.CallCommon) (llvm.Value, error) {
 		}
 		fnType := llvm.FunctionType(b.uintptrType, argTypes, false)
 		target := llvm.InlineAsm(fnType, "int 0x80", constraints, true, false, llvm.InlineAsmDialectIntel)
-		syscallResult = b.CreateCall(target, args, "")
+		return b.CreateCall(target, args, ""), nil
 	case b.GOARCH == "arm" && b.GOOS == "linux":
 		// Implement the EABI system call convention for Linux.
 		// Source: syscall(2) man page.
@@ -115,7 +115,7 @@ func (b *builder) createSyscall(call *ssa.CallCommon) (llvm.Value, error) {
 		}
 		fnType := llvm.FunctionType(b.uintptrType, argTypes, false)
 		target := llvm.InlineAsm(fnType, "svc #0", constraints, true, false, 0)
-		syscallResult = b.CreateCall(target, args, "")
+		return b.CreateCall(target, args, ""), nil
 	case b.GOARCH == "arm64" && b.GOOS == "linux":
 		// Source: syscall(2) man page.
 		args := []llvm.Value{}
@@ -147,10 +147,19 @@ func (b *builder) createSyscall(call *ssa.CallCommon) (llvm.Value, error) {
 		constraints += ",~{x16},~{x17}" // scratch registers
 		fnType := llvm.FunctionType(b.uintptrType, argTypes, false)
 		target := llvm.InlineAsm(fnType, "svc #0", constraints, true, false, 0)
-		syscallResult = b.CreateCall(target, args, "")
+		return b.CreateCall(target, args, ""), nil
 	default:
 		return llvm.Value{}, b.makeError(call.Pos(), "unknown GOOS/GOARCH for syscall: "+b.GOOS+"/"+b.GOARCH)
 	}
+}
+
+// createSyscall emits instructions for the syscall.Syscall* family of
+// functions, depending on the target OS/arch.
+func (b *builder) createSyscall(call *ssa.CallCommon) (llvm.Value, error) {
+	syscallResult, err := b.createRawSyscall(call)
+	if err != nil {
+		return syscallResult, err
+	}
 	switch b.GOOS {
 	case "linux", "freebsd":
 		// Return values: r0, r1 uintptr, err Errno
@@ -190,3 +199,16 @@ func (b *builder) createSyscall(call *ssa.CallCommon) (llvm.Value, error) {
 		return llvm.Value{}, b.makeError(call.Pos(), "unknown GOOS/GOARCH for syscall: "+b.GOOS+"/"+b.GOARCH)
 	}
 }
+
+// createRawSyscallNoError emits instructions for the Linux-specific
+// syscall.rawSyscallNoError function.
+func (b *builder) createRawSyscallNoError(call *ssa.CallCommon) (llvm.Value, error) {
+	syscallResult, err := b.createRawSyscall(call)
+	if err != nil {
+		return syscallResult, err
+	}
+	retval := llvm.ConstNull(b.ctx.StructType([]llvm.Type{b.uintptrType, b.uintptrType}, false))
+	retval = b.CreateInsertValue(retval, syscallResult, 0, "")
+	retval = b.CreateInsertValue(retval, llvm.ConstInt(b.uintptrType, 0, false), 1, "")
+	return retval, nil
+}
@@ -55,3 +55,18 @@ func complexMul(x, y complex64) complex64 {
 }

 // TODO: complexDiv (requires runtime call)
+
+// A type 'kv' also exists in function foo. Test that these two types don't
+// conflict with each other.
+type kv struct {
+	v float32
+}
+
+func foo(a *kv) {
+	// Define a new 'kv' type.
+	type kv struct {
+		v byte
+	}
+	// Use this type.
+	func(b *kv) {}(nil)
+}
@@ -1,7 +1,10 @@
 ; ModuleID = 'basic.go'
 source_filename = "basic.go"
-target datalayout = "e-m:e-p:32:32-p270:32:32-p271:32:32-p272:64:64-f64:32:64-f80:32-n8:16:32-S128"
-target triple = "i686--linux"
+target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
+target triple = "wasm32--wasi"
+
+%main.kv = type { float }
+%main.kv.0 = type { i8 }

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

@@ -98,3 +101,14 @@ entry:
  %7 = insertvalue { float, float } %6, float %5, 1
  ret { float, float } %7
 }
+
+define hidden void @main.foo(%main.kv* dereferenceable_or_null(4) %a, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  call void @"main.foo$1"(%main.kv.0* null, i8* undef, i8* undef)
+  ret void
+}
+
+define hidden void @"main.foo$1"(%main.kv.0* dereferenceable_or_null(1) %b, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  ret void
+}
@@ -0,0 +1,25 @@
+package main
+
+func chanIntSend(ch chan int) {
+	ch <- 3
+}
+
+func chanIntRecv(ch chan int) {
+	<-ch
+}
+
+func chanZeroSend(ch chan struct{}) {
+	ch <- struct{}{}
+}
+
+func chanZeroRecv(ch chan struct{}) {
+	<-ch
+}
+
+func selectZeroRecv(ch1 chan int, ch2 chan struct{}) {
+	select {
+	case ch1 <- 1:
+	case <-ch2:
+	default:
+	}
+}
@@ -0,0 +1,114 @@
+; ModuleID = 'channel.go'
+source_filename = "channel.go"
+target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
+target triple = "wasm32--wasi"
+
+%runtime.channel = type { i32, i32, i8, %runtime.channelBlockedList*, i32, i32, i32, i8* }
+%runtime.channelBlockedList = type { %runtime.channelBlockedList*, %"internal/task.Task"*, %runtime.chanSelectState*, { %runtime.channelBlockedList*, i32, i32 } }
+%"internal/task.Task" = type { %"internal/task.Task"*, i8*, i64, %"internal/task.state" }
+%"internal/task.state" = type { i8* }
+%runtime.chanSelectState = type { %runtime.channel*, i8* }
+
+declare noalias nonnull i8* @runtime.alloc(i32, i8*, i8*)
+
+define hidden void @main.init(i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  ret void
+}
+
+define hidden void @main.chanIntSend(%runtime.channel* dereferenceable_or_null(32) %ch, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %chan.blockedList = alloca %runtime.channelBlockedList, align 8
+  %chan.value = alloca i32, align 4
+  %chan.value.bitcast = bitcast i32* %chan.value to i8*
+  call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull %chan.value.bitcast)
+  store i32 3, i32* %chan.value, align 4
+  %chan.blockedList.bitcast = bitcast %runtime.channelBlockedList* %chan.blockedList to i8*
+  call void @llvm.lifetime.start.p0i8(i64 24, i8* nonnull %chan.blockedList.bitcast)
+  call void @runtime.chanSend(%runtime.channel* %ch, i8* nonnull %chan.value.bitcast, %runtime.channelBlockedList* nonnull %chan.blockedList, i8* undef, i8* null)
+  call void @llvm.lifetime.end.p0i8(i64 24, i8* nonnull %chan.blockedList.bitcast)
+  call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull %chan.value.bitcast)
+  ret void
+}
+
+; Function Attrs: argmemonly nounwind willreturn
+declare void @llvm.lifetime.start.p0i8(i64 immarg, i8* nocapture) #0
+
+declare void @runtime.chanSend(%runtime.channel* dereferenceable_or_null(32), i8*, %runtime.channelBlockedList* dereferenceable_or_null(24), i8*, i8*)
+
+; Function Attrs: argmemonly nounwind willreturn
+declare void @llvm.lifetime.end.p0i8(i64 immarg, i8* nocapture) #0
+
+define hidden void @main.chanIntRecv(%runtime.channel* dereferenceable_or_null(32) %ch, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %chan.blockedList = alloca %runtime.channelBlockedList, align 8
+  %chan.value = alloca i32, align 4
+  %chan.value.bitcast = bitcast i32* %chan.value to i8*
+  call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull %chan.value.bitcast)
+  %chan.blockedList.bitcast = bitcast %runtime.channelBlockedList* %chan.blockedList to i8*
+  call void @llvm.lifetime.start.p0i8(i64 24, i8* nonnull %chan.blockedList.bitcast)
+  %0 = call i1 @runtime.chanRecv(%runtime.channel* %ch, i8* nonnull %chan.value.bitcast, %runtime.channelBlockedList* nonnull %chan.blockedList, i8* undef, i8* null)
+  call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull %chan.value.bitcast)
+  call void @llvm.lifetime.end.p0i8(i64 24, i8* nonnull %chan.blockedList.bitcast)
+  ret void
+}
+
+declare i1 @runtime.chanRecv(%runtime.channel* dereferenceable_or_null(32), i8*, %runtime.channelBlockedList* dereferenceable_or_null(24), i8*, i8*)
+
+define hidden void @main.chanZeroSend(%runtime.channel* dereferenceable_or_null(32) %ch, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %chan.blockedList = alloca %runtime.channelBlockedList, align 8
+  %chan.blockedList.bitcast = bitcast %runtime.channelBlockedList* %chan.blockedList to i8*
+  call void @llvm.lifetime.start.p0i8(i64 24, i8* nonnull %chan.blockedList.bitcast)
+  call void @runtime.chanSend(%runtime.channel* %ch, i8* null, %runtime.channelBlockedList* nonnull %chan.blockedList, i8* undef, i8* null)
+  call void @llvm.lifetime.end.p0i8(i64 24, i8* nonnull %chan.blockedList.bitcast)
+  ret void
+}
+
+define hidden void @main.chanZeroRecv(%runtime.channel* dereferenceable_or_null(32) %ch, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %chan.blockedList = alloca %runtime.channelBlockedList, align 8
+  %chan.blockedList.bitcast = bitcast %runtime.channelBlockedList* %chan.blockedList to i8*
+  call void @llvm.lifetime.start.p0i8(i64 24, i8* nonnull %chan.blockedList.bitcast)
+  %0 = call i1 @runtime.chanRecv(%runtime.channel* %ch, i8* null, %runtime.channelBlockedList* nonnull %chan.blockedList, i8* undef, i8* null)
+  call void @llvm.lifetime.end.p0i8(i64 24, i8* nonnull %chan.blockedList.bitcast)
+  ret void
+}
+
+define hidden void @main.selectZeroRecv(%runtime.channel* dereferenceable_or_null(32) %ch1, %runtime.channel* dereferenceable_or_null(32) %ch2, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %select.states.alloca = alloca [2 x %runtime.chanSelectState], align 8
+  %select.send.value = alloca i32, align 4
+  store i32 1, i32* %select.send.value, align 4
+  %select.states.alloca.bitcast = bitcast [2 x %runtime.chanSelectState]* %select.states.alloca to i8*
+  call void @llvm.lifetime.start.p0i8(i64 16, i8* nonnull %select.states.alloca.bitcast)
+  %.repack = getelementptr inbounds [2 x %runtime.chanSelectState], [2 x %runtime.chanSelectState]* %select.states.alloca, i32 0, i32 0, i32 0
+  store %runtime.channel* %ch1, %runtime.channel** %.repack, align 8
+  %.repack1 = getelementptr inbounds [2 x %runtime.chanSelectState], [2 x %runtime.chanSelectState]* %select.states.alloca, i32 0, i32 0, i32 1
+  %0 = bitcast i8** %.repack1 to i32**
+  store i32* %select.send.value, i32** %0, align 4
+  %.repack3 = getelementptr inbounds [2 x %runtime.chanSelectState], [2 x %runtime.chanSelectState]* %select.states.alloca, i32 0, i32 1, i32 0
+  store %runtime.channel* %ch2, %runtime.channel** %.repack3, align 8
+  %.repack4 = getelementptr inbounds [2 x %runtime.chanSelectState], [2 x %runtime.chanSelectState]* %select.states.alloca, i32 0, i32 1, i32 1
+  store i8* null, i8** %.repack4, align 4
+  %select.states = getelementptr inbounds [2 x %runtime.chanSelectState], [2 x %runtime.chanSelectState]* %select.states.alloca, i32 0, i32 0
+  %select.result = call { i32, i1 } @runtime.tryChanSelect(i8* undef, %runtime.chanSelectState* nonnull %select.states, i32 2, i32 2, i8* undef, i8* null)
+  call void @llvm.lifetime.end.p0i8(i64 16, i8* nonnull %select.states.alloca.bitcast)
+  %1 = extractvalue { i32, i1 } %select.result, 0
+  %2 = icmp eq i32 %1, 0
+  br i1 %2, label %select.done, label %select.next
+
+select.done:                                      ; preds = %select.body, %select.next, %entry
+  ret void
+
+select.next:                                      ; preds = %entry
+  %3 = icmp eq i32 %1, 1
+  br i1 %3, label %select.body, label %select.done
+
+select.body:                                      ; preds = %select.next
+  br label %select.done
+}
+
+declare { i32, i1 } @runtime.tryChanSelect(i8*, %runtime.chanSelectState*, i32, i32, i8*, i8*)
+
+attributes #0 = { argmemonly nounwind willreturn }
@@ -1,7 +1,7 @@
 ; ModuleID = 'float.go'
 source_filename = "float.go"
-target datalayout = "e-m:e-p:32:32-p270:32:32-p271:32:32-p272:64:64-f64:32:64-f80:32-n8:16:32-S128"
-target triple = "i686--linux"
+target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
+target triple = "wasm32--wasi"

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

@@ -1,7 +1,7 @@
 ; ModuleID = 'func.go'
 source_filename = "func.go"
-target datalayout = "e-m:e-p:32:32-p270:32:32-p271:32:32-p272:64:64-f64:32:64-f80:32-n8:16:32-S128"
-target triple = "i686--linux"
+target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
+target triple = "wasm32--wasi"

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

@@ -0,0 +1,41 @@
+package main
+
+// Test changes to the language introduced in Go 1.17.
+// For details, see: https://tip.golang.org/doc/go1.17#language
+// These tests should be merged into the regular slice tests once Go 1.17 is the
+// minimun Go version for TinyGo.
+
+import "unsafe"
+
+func Add32(p unsafe.Pointer, len int) unsafe.Pointer {
+	return unsafe.Add(p, len)
+}
+
+func Add64(p unsafe.Pointer, len int64) unsafe.Pointer {
+	return unsafe.Add(p, len)
+}
+
+func SliceToArray(s []int) *[4]int {
+	return (*[4]int)(s)
+}
+
+func SliceToArrayConst() *[4]int {
+	s := make([]int, 6)
+	return (*[4]int)(s)
+}
+
+func SliceInt(ptr *int, len int) []int {
+	return unsafe.Slice(ptr, len)
+}
+
+func SliceUint16(ptr *byte, len uint16) []byte {
+	return unsafe.Slice(ptr, len)
+}
+
+func SliceUint64(ptr *int, len uint64) []int {
+	return unsafe.Slice(ptr, len)
+}
+
+func SliceInt64(ptr *int, len int64) []int {
+	return unsafe.Slice(ptr, len)
+}
@@ -0,0 +1,136 @@
+; ModuleID = 'go1.17.go'
+source_filename = "go1.17.go"
+target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
+target triple = "wasm32--wasi"
+
+declare noalias nonnull i8* @runtime.alloc(i32, i8*, i8*)
+
+define hidden void @main.init(i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  ret void
+}
+
+define hidden i8* @main.Add32(i8* %p, i32 %len, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %0 = getelementptr i8, i8* %p, i32 %len
+  ret i8* %0
+}
+
+define hidden i8* @main.Add64(i8* %p, i64 %len, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %0 = trunc i64 %len to i32
+  %1 = getelementptr i8, i8* %p, i32 %0
+  ret i8* %1
+}
+
+define hidden [4 x i32]* @main.SliceToArray(i32* %s.data, i32 %s.len, i32 %s.cap, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %0 = icmp ult i32 %s.len, 4
+  br i1 %0, label %slicetoarray.throw, label %slicetoarray.next
+
+slicetoarray.throw:                               ; preds = %entry
+  call void @runtime.sliceToArrayPointerPanic(i8* undef, i8* null)
+  unreachable
+
+slicetoarray.next:                                ; preds = %entry
+  %1 = bitcast i32* %s.data to [4 x i32]*
+  ret [4 x i32]* %1
+}
+
+declare void @runtime.sliceToArrayPointerPanic(i8*, i8*)
+
+define hidden [4 x i32]* @main.SliceToArrayConst(i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %makeslice = call i8* @runtime.alloc(i32 24, i8* undef, i8* null)
+  br i1 false, label %slicetoarray.throw, label %slicetoarray.next
+
+slicetoarray.throw:                               ; preds = %entry
+  unreachable
+
+slicetoarray.next:                                ; preds = %entry
+  %0 = bitcast i8* %makeslice to [4 x i32]*
+  ret [4 x i32]* %0
+}
+
+define hidden { i32*, i32, i32 } @main.SliceInt(i32* dereferenceable_or_null(4) %ptr, i32 %len, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %0 = icmp ugt i32 %len, 1073741823
+  %1 = icmp eq i32* %ptr, null
+  %2 = icmp ne i32 %len, 0
+  %3 = and i1 %1, %2
+  %4 = or i1 %3, %0
+  br i1 %4, label %unsafe.Slice.throw, label %unsafe.Slice.next
+
+unsafe.Slice.throw:                               ; preds = %entry
+  call void @runtime.unsafeSlicePanic(i8* undef, i8* null)
+  unreachable
+
+unsafe.Slice.next:                                ; preds = %entry
+  %5 = insertvalue { i32*, i32, i32 } undef, i32* %ptr, 0
+  %6 = insertvalue { i32*, i32, i32 } %5, i32 %len, 1
+  %7 = insertvalue { i32*, i32, i32 } %6, i32 %len, 2
+  ret { i32*, i32, i32 } %7
+}
+
+declare void @runtime.unsafeSlicePanic(i8*, i8*)
+
+define hidden { i8*, i32, i32 } @main.SliceUint16(i8* dereferenceable_or_null(1) %ptr, i16 %len, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %0 = icmp eq i8* %ptr, null
+  %1 = icmp ne i16 %len, 0
+  %2 = and i1 %0, %1
+  br i1 %2, label %unsafe.Slice.throw, label %unsafe.Slice.next
+
+unsafe.Slice.throw:                               ; preds = %entry
+  call void @runtime.unsafeSlicePanic(i8* undef, i8* null)
+  unreachable
+
+unsafe.Slice.next:                                ; preds = %entry
+  %3 = zext i16 %len to i32
+  %4 = insertvalue { i8*, i32, i32 } undef, i8* %ptr, 0
+  %5 = insertvalue { i8*, i32, i32 } %4, i32 %3, 1
+  %6 = insertvalue { i8*, i32, i32 } %5, i32 %3, 2
+  ret { i8*, i32, i32 } %6
+}
+
+define hidden { i32*, i32, i32 } @main.SliceUint64(i32* dereferenceable_or_null(4) %ptr, i64 %len, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %0 = icmp ugt i64 %len, 1073741823
+  %1 = icmp eq i32* %ptr, null
+  %2 = icmp ne i64 %len, 0
+  %3 = and i1 %1, %2
+  %4 = or i1 %3, %0
+  br i1 %4, label %unsafe.Slice.throw, label %unsafe.Slice.next
+
+unsafe.Slice.throw:                               ; preds = %entry
+  call void @runtime.unsafeSlicePanic(i8* undef, i8* null)
+  unreachable
+
+unsafe.Slice.next:                                ; preds = %entry
+  %5 = trunc i64 %len to i32
+  %6 = insertvalue { i32*, i32, i32 } undef, i32* %ptr, 0
+  %7 = insertvalue { i32*, i32, i32 } %6, i32 %5, 1
+  %8 = insertvalue { i32*, i32, i32 } %7, i32 %5, 2
+  ret { i32*, i32, i32 } %8
+}
+
+define hidden { i32*, i32, i32 } @main.SliceInt64(i32* dereferenceable_or_null(4) %ptr, i64 %len, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %0 = icmp ugt i64 %len, 1073741823
+  %1 = icmp eq i32* %ptr, null
+  %2 = icmp ne i64 %len, 0
+  %3 = and i1 %1, %2
+  %4 = or i1 %3, %0
+  br i1 %4, label %unsafe.Slice.throw, label %unsafe.Slice.next
+
+unsafe.Slice.throw:                               ; preds = %entry
+  call void @runtime.unsafeSlicePanic(i8* undef, i8* null)
+  unreachable
+
+unsafe.Slice.next:                                ; preds = %entry
+  %5 = trunc i64 %len to i32
+  %6 = insertvalue { i32*, i32, i32 } undef, i32* %ptr, 0
+  %7 = insertvalue { i32*, i32, i32 } %6, i32 %5, 1
+  %8 = insertvalue { i32*, i32, i32 } %7, i32 %5, 2
+  ret { i32*, i32, i32 } %8
+}
@@ -0,0 +1,150 @@
+; ModuleID = 'goroutine.go'
+source_filename = "goroutine.go"
+target datalayout = "e-m:e-p:32:32-Fi8-i64:64-v128:64:128-a:0:32-n32-S64"
+target triple = "armv7m-none-eabi"
+
+%runtime.channel = type { i32, i32, i8, %runtime.channelBlockedList*, i32, i32, i32, i8* }
+%runtime.channelBlockedList = type { %runtime.channelBlockedList*, %"internal/task.Task"*, %runtime.chanSelectState*, { %runtime.channelBlockedList*, i32, i32 } }
+%"internal/task.Task" = type { %"internal/task.Task"*, i8*, i64, %"internal/task.state" }
+%"internal/task.state" = type { i32, i32* }
+%runtime.chanSelectState = type { %runtime.channel*, i8* }
+
+declare noalias nonnull i8* @runtime.alloc(i32, i8*, i8*)
+
+define hidden void @main.init(i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  ret void
+}
+
+define hidden void @main.regularFunctionGoroutine(i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %stacksize = call i32 @"internal/task.getGoroutineStackSize"(i32 ptrtoint (void (i8*)* @"main.regularFunction$gowrapper" to i32), i8* undef, i8* undef)
+  call void @"internal/task.start"(i32 ptrtoint (void (i8*)* @"main.regularFunction$gowrapper" to i32), i8* nonnull inttoptr (i32 5 to i8*), i32 %stacksize, i8* undef, i8* null)
+  ret void
+}
+
+declare void @main.regularFunction(i32, i8*, i8*)
+
+define linkonce_odr void @"main.regularFunction$gowrapper"(i8* %0) unnamed_addr #0 {
+entry:
+  %unpack.int = ptrtoint i8* %0 to i32
+  call void @main.regularFunction(i32 %unpack.int, i8* undef, i8* undef)
+  ret void
+}
+
+declare i32 @"internal/task.getGoroutineStackSize"(i32, i8*, i8*)
+
+declare void @"internal/task.start"(i32, i8*, i32, i8*, i8*)
+
+define hidden void @main.inlineFunctionGoroutine(i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %stacksize = call i32 @"internal/task.getGoroutineStackSize"(i32 ptrtoint (void (i8*)* @"main.inlineFunctionGoroutine$1$gowrapper" to i32), i8* undef, i8* undef)
+  call void @"internal/task.start"(i32 ptrtoint (void (i8*)* @"main.inlineFunctionGoroutine$1$gowrapper" to i32), i8* nonnull inttoptr (i32 5 to i8*), i32 %stacksize, i8* undef, i8* null)
+  ret void
+}
+
+define hidden void @"main.inlineFunctionGoroutine$1"(i32 %x, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  ret void
+}
+
+define linkonce_odr void @"main.inlineFunctionGoroutine$1$gowrapper"(i8* %0) unnamed_addr #1 {
+entry:
+  %unpack.int = ptrtoint i8* %0 to i32
+  call void @"main.inlineFunctionGoroutine$1"(i32 %unpack.int, i8* undef, i8* undef)
+  ret void
+}
+
+define hidden void @main.closureFunctionGoroutine(i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %n = call i8* @runtime.alloc(i32 4, i8* undef, i8* null)
+  %0 = bitcast i8* %n to i32*
+  store i32 3, i32* %0, align 4
+  %1 = call i8* @runtime.alloc(i32 8, i8* undef, i8* null)
+  %2 = bitcast i8* %1 to i32*
+  store i32 5, i32* %2, align 4
+  %3 = getelementptr inbounds i8, i8* %1, i32 4
+  %4 = bitcast i8* %3 to i8**
+  store i8* %n, i8** %4, align 4
+  %stacksize = call i32 @"internal/task.getGoroutineStackSize"(i32 ptrtoint (void (i8*)* @"main.closureFunctionGoroutine$1$gowrapper" to i32), i8* undef, i8* undef)
+  call void @"internal/task.start"(i32 ptrtoint (void (i8*)* @"main.closureFunctionGoroutine$1$gowrapper" to i32), i8* nonnull %1, i32 %stacksize, i8* undef, i8* null)
+  %5 = load i32, i32* %0, align 4
+  call void @runtime.printint32(i32 %5, i8* undef, i8* null)
+  ret void
+}
+
+define hidden void @"main.closureFunctionGoroutine$1"(i32 %x, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %unpack.ptr = bitcast i8* %context to i32*
+  store i32 7, i32* %unpack.ptr, align 4
+  ret void
+}
+
+define linkonce_odr void @"main.closureFunctionGoroutine$1$gowrapper"(i8* %0) unnamed_addr #2 {
+entry:
+  %1 = bitcast i8* %0 to i32*
+  %2 = load i32, i32* %1, align 4
+  %3 = getelementptr inbounds i8, i8* %0, i32 4
+  %4 = bitcast i8* %3 to i8**
+  %5 = load i8*, i8** %4, align 4
+  call void @"main.closureFunctionGoroutine$1"(i32 %2, i8* %5, i8* undef)
+  ret void
+}
+
+declare void @runtime.printint32(i32, i8*, i8*)
+
+define hidden void @main.funcGoroutine(i8* %fn.context, void (i32, i8*, i8*)* %fn.funcptr, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %0 = call i8* @runtime.alloc(i32 12, i8* undef, i8* null)
+  %1 = bitcast i8* %0 to i32*
+  store i32 5, i32* %1, align 4
+  %2 = getelementptr inbounds i8, i8* %0, i32 4
+  %3 = bitcast i8* %2 to i8**
+  store i8* %fn.context, i8** %3, align 4
+  %4 = getelementptr inbounds i8, i8* %0, i32 8
+  %5 = bitcast i8* %4 to void (i32, i8*, i8*)**
+  store void (i32, i8*, i8*)* %fn.funcptr, void (i32, i8*, i8*)** %5, align 4
+  %stacksize = call i32 @"internal/task.getGoroutineStackSize"(i32 ptrtoint (void (i8*)* @main.funcGoroutine.gowrapper to i32), i8* undef, i8* undef)
+  call void @"internal/task.start"(i32 ptrtoint (void (i8*)* @main.funcGoroutine.gowrapper to i32), i8* nonnull %0, i32 %stacksize, i8* undef, i8* null)
+  ret void
+}
+
+define linkonce_odr void @main.funcGoroutine.gowrapper(i8* %0) unnamed_addr #3 {
+entry:
+  %1 = bitcast i8* %0 to i32*
+  %2 = load i32, i32* %1, align 4
+  %3 = getelementptr inbounds i8, i8* %0, i32 4
+  %4 = bitcast i8* %3 to i8**
+  %5 = load i8*, i8** %4, align 4
+  %6 = getelementptr inbounds i8, i8* %0, i32 8
+  %7 = bitcast i8* %6 to void (i32, i8*, i8*)**
+  %8 = load void (i32, i8*, i8*)*, void (i32, i8*, i8*)** %7, align 4
+  call void %8(i32 %2, i8* %5, i8* undef)
+  ret void
+}
+
+define hidden void @main.recoverBuiltinGoroutine(i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  ret void
+}
+
+define hidden void @main.copyBuiltinGoroutine(i8* %dst.data, i32 %dst.len, i32 %dst.cap, i8* %src.data, i32 %src.len, i32 %src.cap, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %copy.n = call i32 @runtime.sliceCopy(i8* %dst.data, i8* %src.data, i32 %dst.len, i32 %src.len, i32 1, i8* undef, i8* null)
+  ret void
+}
+
+declare i32 @runtime.sliceCopy(i8* nocapture writeonly, i8* nocapture readonly, i32, i32, i32, i8*, i8*)
+
+define hidden void @main.closeBuiltinGoroutine(%runtime.channel* dereferenceable_or_null(32) %ch, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  call void @runtime.chanClose(%runtime.channel* %ch, i8* undef, i8* null)
+  ret void
+}
+
+declare void @runtime.chanClose(%runtime.channel* dereferenceable_or_null(32), i8*, i8*)
+
+attributes #0 = { "tinygo-gowrapper"="main.regularFunction" }
+attributes #1 = { "tinygo-gowrapper"="main.inlineFunctionGoroutine$1" }
+attributes #2 = { "tinygo-gowrapper"="main.closureFunctionGoroutine$1" }
+attributes #3 = { "tinygo-gowrapper" }
@@ -0,0 +1,106 @@
+; ModuleID = 'goroutine.go'
+source_filename = "goroutine.go"
+target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
+target triple = "wasm32--wasi"
+
+%runtime.funcValueWithSignature = type { i32, i8* }
+%runtime.channel = type { i32, i32, i8, %runtime.channelBlockedList*, i32, i32, i32, i8* }
+%runtime.channelBlockedList = type { %runtime.channelBlockedList*, %"internal/task.Task"*, %runtime.chanSelectState*, { %runtime.channelBlockedList*, i32, i32 } }
+%"internal/task.Task" = type { %"internal/task.Task"*, i8*, i64, %"internal/task.state" }
+%"internal/task.state" = type { i8* }
+%runtime.chanSelectState = type { %runtime.channel*, i8* }
+
+@"main.regularFunctionGoroutine$pack" = private unnamed_addr constant { i32, i8* } { i32 5, i8* undef }
+@"main.inlineFunctionGoroutine$pack" = private unnamed_addr constant { i32, i8* } { i32 5, i8* undef }
+@"reflect/types.funcid:func:{basic:int}{}" = external constant i8
+@"main.closureFunctionGoroutine$1$withSignature" = linkonce_odr constant %runtime.funcValueWithSignature { i32 ptrtoint (void (i32, i8*, i8*)* @"main.closureFunctionGoroutine$1" to i32), i8* @"reflect/types.funcid:func:{basic:int}{}" }
+
+declare noalias nonnull i8* @runtime.alloc(i32, i8*, i8*)
+
+define hidden void @main.init(i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  ret void
+}
+
+define hidden void @main.regularFunctionGoroutine(i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  call void @"internal/task.start"(i32 ptrtoint (void (i32, i8*, i8*)* @main.regularFunction to i32), i8* bitcast ({ i32, i8* }* @"main.regularFunctionGoroutine$pack" to i8*), i32 undef, i8* undef, i8* null)
+  ret void
+}
+
+declare void @main.regularFunction(i32, i8*, i8*)
+
+declare void @"internal/task.start"(i32, i8*, i32, i8*, i8*)
+
+define hidden void @main.inlineFunctionGoroutine(i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  call void @"internal/task.start"(i32 ptrtoint (void (i32, i8*, i8*)* @"main.inlineFunctionGoroutine$1" to i32), i8* bitcast ({ i32, i8* }* @"main.inlineFunctionGoroutine$pack" to i8*), i32 undef, i8* undef, i8* null)
+  ret void
+}
+
+define hidden void @"main.inlineFunctionGoroutine$1"(i32 %x, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  ret void
+}
+
+define hidden void @main.closureFunctionGoroutine(i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %n = call i8* @runtime.alloc(i32 4, i8* undef, i8* null)
+  %0 = bitcast i8* %n to i32*
+  store i32 3, i32* %0, align 4
+  %1 = call i8* @runtime.alloc(i32 8, i8* undef, i8* null)
+  %2 = bitcast i8* %1 to i32*
+  store i32 5, i32* %2, align 4
+  %3 = getelementptr inbounds i8, i8* %1, i32 4
+  %4 = bitcast i8* %3 to i8**
+  store i8* %n, i8** %4, align 4
+  call void @"internal/task.start"(i32 ptrtoint (void (i32, i8*, i8*)* @"main.closureFunctionGoroutine$1" to i32), i8* nonnull %1, i32 undef, i8* undef, i8* null)
+  %5 = load i32, i32* %0, align 4
+  call void @runtime.printint32(i32 %5, i8* undef, i8* null)
+  ret void
+}
+
+define hidden void @"main.closureFunctionGoroutine$1"(i32 %x, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %unpack.ptr = bitcast i8* %context to i32*
+  store i32 7, i32* %unpack.ptr, align 4
+  ret void
+}
+
+declare void @runtime.printint32(i32, i8*, i8*)
+
+define hidden void @main.funcGoroutine(i8* %fn.context, i32 %fn.funcptr, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %0 = call i32 @runtime.getFuncPtr(i8* %fn.context, i32 %fn.funcptr, i8* nonnull @"reflect/types.funcid:func:{basic:int}{}", i8* undef, i8* null)
+  %1 = call i8* @runtime.alloc(i32 8, i8* undef, i8* null)
+  %2 = bitcast i8* %1 to i32*
+  store i32 5, i32* %2, align 4
+  %3 = getelementptr inbounds i8, i8* %1, i32 4
+  %4 = bitcast i8* %3 to i8**
+  store i8* %fn.context, i8** %4, align 4
+  call void @"internal/task.start"(i32 %0, i8* nonnull %1, i32 undef, i8* undef, i8* null)
+  ret void
+}
+
+declare i32 @runtime.getFuncPtr(i8*, i32, i8* dereferenceable_or_null(1), i8*, i8*)
+
+define hidden void @main.recoverBuiltinGoroutine(i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  ret void
+}
+
+define hidden void @main.copyBuiltinGoroutine(i8* %dst.data, i32 %dst.len, i32 %dst.cap, i8* %src.data, i32 %src.len, i32 %src.cap, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %copy.n = call i32 @runtime.sliceCopy(i8* %dst.data, i8* %src.data, i32 %dst.len, i32 %src.len, i32 1, i8* undef, i8* null)
+  ret void
+}
+
+declare i32 @runtime.sliceCopy(i8* nocapture writeonly, i8* nocapture readonly, i32, i32, i32, i8*, i8*)
+
+define hidden void @main.closeBuiltinGoroutine(%runtime.channel* dereferenceable_or_null(32) %ch, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  call void @runtime.chanClose(%runtime.channel* %ch, i8* undef, i8* null)
+  ret void
+}
+
+declare void @runtime.chanClose(%runtime.channel* dereferenceable_or_null(32), i8*, i8*)
@@ -0,0 +1,43 @@
+package main
+
+func regularFunctionGoroutine() {
+	go regularFunction(5)
+}
+
+func inlineFunctionGoroutine() {
+	go func(x int) {
+	}(5)
+}
+
+func closureFunctionGoroutine() {
+	n := 3
+	go func(x int) {
+		n = 7
+	}(5)
+	print(n) // note: this is racy (but good enough for this test)
+}
+
+func funcGoroutine(fn func(x int)) {
+	go fn(5)
+}
+
+func recoverBuiltinGoroutine() {
+	// This is a no-op.
+	go recover()
+}
+
+func copyBuiltinGoroutine(dst, src []byte) {
+	// This is not run in a goroutine. While this copy operation can indeed take
+	// some time (if there is a lot of data to copy), there is no race-free way
+	// to make use of the result so it's unlikely applications will make use of
+	// it. And doing it this way should be just within the Go specification.
+	go copy(dst, src)
+}
+
+func closeBuiltinGoroutine(ch chan int) {
+	// This builtin is executed directly, not in a goroutine.
+	// The observed behavior is the same.
+	go close(ch)
+}
+
+func regularFunction(x int)
@@ -1,7 +1,7 @@
 ; ModuleID = 'interface.go'
 source_filename = "interface.go"
-target datalayout = "e-m:e-p:32:32-p270:32:32-p271:32:32-p272:64:64-f64:32:64-f80:32-n8:16:32-S128"
-target triple = "i686--linux"
+target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
+target triple = "wasm32--wasi"

 %runtime.typecodeID = type { %runtime.typecodeID*, i32, %runtime.interfaceMethodInfo*, %runtime.typecodeID* }
 %runtime.interfaceMethodInfo = type { i8*, i32 }
@@ -13,15 +13,15 @@ target triple = "i686--linux"
@"reflect/types.type:pointer:named:error" = linkonce_odr constant %runtime.typecodeID { %runtime.typecodeID* @"reflect/types.type:named:error", i32 0, %runtime.interfaceMethodInfo* null, %runtime.typecodeID* null }
@"reflect/types.type:named:error" = linkonce_odr constant %runtime.typecodeID { %runtime.typecodeID* @"reflect/types.type:interface:{Error:func:{}{basic:string}}", i32 0, %runtime.interfaceMethodInfo* null, %runtime.typecodeID* @"reflect/types.type:pointer:named:error" }
@"reflect/types.type:interface:{Error:func:{}{basic:string}}" = linkonce_odr constant %runtime.typecodeID { %runtime.typecodeID* bitcast ([1 x i8*]* @"reflect/types.interface:interface{Error() string}$interface" to %runtime.typecodeID*), i32 0, %runtime.interfaceMethodInfo* null, %runtime.typecodeID* @"reflect/types.type:pointer:interface:{Error:func:{}{basic:string}}" }
-@"func Error() string" = external constant i8
-@"reflect/types.interface:interface{Error() string}$interface" = linkonce_odr constant [1 x i8*] [i8* @"func Error() string"]
+@"reflect/methods.Error() string" = linkonce_odr constant i8 0, align 1
+@"reflect/types.interface:interface{Error() string}$interface" = linkonce_odr constant [1 x i8*] [i8* @"reflect/methods.Error() string"]
@"reflect/types.type:pointer:interface:{Error:func:{}{basic:string}}" = linkonce_odr constant %runtime.typecodeID { %runtime.typecodeID* @"reflect/types.type:interface:{Error:func:{}{basic:string}}", i32 0, %runtime.interfaceMethodInfo* null, %runtime.typecodeID* null }
@"reflect/types.type:pointer:interface:{String:func:{}{basic:string}}" = linkonce_odr constant %runtime.typecodeID { %runtime.typecodeID* @"reflect/types.type:interface:{String:func:{}{basic:string}}", i32 0, %runtime.interfaceMethodInfo* null, %runtime.typecodeID* null }
@"reflect/types.type:interface:{String:func:{}{basic:string}}" = linkonce_odr constant %runtime.typecodeID { %runtime.typecodeID* bitcast ([1 x i8*]* @"reflect/types.interface:interface{String() string}$interface" to %runtime.typecodeID*), i32 0, %runtime.interfaceMethodInfo* null, %runtime.typecodeID* @"reflect/types.type:pointer:interface:{String:func:{}{basic:string}}" }
-@"func String() string" = external constant i8
-@"reflect/types.interface:interface{String() string}$interface" = linkonce_odr constant [1 x i8*] [i8* @"func String() string"]
+@"reflect/methods.String() string" = linkonce_odr constant i8 0, align 1
+@"reflect/types.interface:interface{String() string}$interface" = linkonce_odr constant [1 x i8*] [i8* @"reflect/methods.String() string"]
@"reflect/types.typeid:basic:int" = external constant i8
-@"error$interface" = linkonce_odr constant [1 x i8*] [i8* @"func Error() string"]
+@"error$interface" = linkonce_odr constant [1 x i8*] [i8* @"reflect/methods.Error() string"]

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

@@ -92,7 +92,7 @@ typeassert.next:                                  ; preds = %typeassert.ok, %ent

 define hidden %runtime._string @main.callErrorMethod(i32 %itf.typecode, i8* %itf.value, i8* %context, i8* %parentHandle) unnamed_addr {
 entry:
-  %invoke.func = call i32 @runtime.interfaceMethod(i32 %itf.typecode, i8** getelementptr inbounds ([1 x i8*], [1 x i8*]* @"error$interface", i32 0, i32 0), i8* nonnull @"func Error() string", i8* undef, i8* null)
+  %invoke.func = call i32 @runtime.interfaceMethod(i32 %itf.typecode, i8** getelementptr inbounds ([1 x i8*], [1 x i8*]* @"error$interface", i32 0, i32 0), i8* nonnull @"reflect/methods.Error() string", i8* undef, i8* null)
  %invoke.func.cast = inttoptr i32 %invoke.func to %runtime._string (i8*, i8*, i8*)*
  %0 = call %runtime._string %invoke.func.cast(i8* %itf.value, i8* undef, i8* undef)
  ret %runtime._string %0
@@ -0,0 +1,27 @@
+; ModuleID = 'intrinsics.go'
+source_filename = "intrinsics.go"
+target datalayout = "e-m:e-p:32:32-Fi8-i64:64-v128:64:128-a:0:32-n32-S64"
+target triple = "armv7m-none-eabi"
+
+declare noalias nonnull i8* @runtime.alloc(i32, i8*, i8*)
+
+define hidden void @main.init(i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  ret void
+}
+
+define hidden double @main.mySqrt(double %x, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %0 = call double @math.Sqrt(double %x, i8* undef, i8* undef)
+  ret double %0
+}
+
+declare double @math.Sqrt(double, i8*, i8*)
+
+define hidden double @main.myTrunc(double %x, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %0 = call double @math.Trunc(double %x, i8* undef, i8* undef)
+  ret double %0
+}
+
+declare double @math.Trunc(double, i8*, i8*)
@@ -0,0 +1,31 @@
+; ModuleID = 'intrinsics.go'
+source_filename = "intrinsics.go"
+target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
+target triple = "wasm32--wasi"
+
+declare noalias nonnull i8* @runtime.alloc(i32, i8*, i8*)
+
+define hidden void @main.init(i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  ret void
+}
+
+define hidden double @main.mySqrt(double %x, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %0 = call double @llvm.sqrt.f64(double %x)
+  ret double %0
+}
+
+; Function Attrs: nounwind readnone speculatable willreturn
+declare double @llvm.sqrt.f64(double) #0
+
+define hidden double @main.myTrunc(double %x, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %0 = call double @llvm.trunc.f64(double %x)
+  ret double %0
+}
+
+; Function Attrs: nounwind readnone speculatable willreturn
+declare double @llvm.trunc.f64(double) #0
+
+attributes #0 = { nounwind readnone speculatable willreturn }
@@ -0,0 +1,14 @@
+package main
+
+// Test how intrinsics are lowered: either as regular calls to the math
+// functions or as LLVM builtins (such as llvm.sqrt.f64).
+
+import "math"
+
+func mySqrt(x float64) float64 {
+	return math.Sqrt(x)
+}
+
+func myTrunc(x float64) float64 {
+	return math.Trunc(x)
+}
@@ -1,7 +1,7 @@
 ; ModuleID = 'pointer.go'
 source_filename = "pointer.go"
-target datalayout = "e-m:e-p:32:32-p270:32:32-p271:32:32-p272:64:64-f64:32:64-f80:32-n8:16:32-S128"
-target triple = "i686--linux"
+target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
+target triple = "wasm32--wasi"

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

@@ -0,0 +1,66 @@
+package main
+
+import _ "unsafe"
+
+// Creates an external global with name extern_global.
+//go:extern extern_global
+var externGlobal [0]byte
+
+// Creates a
+//go:align 32
+var alignedGlobal [4]uint32
+
+// Test conflicting pragmas (the last one counts).
+//go:align 64
+//go:align 16
+var alignedGlobal16 [4]uint32
+
+// Test exported functions.
+//export extern_func
+func externFunc() {
+}
+
+// Define a function in a different package using go:linkname.
+//go:linkname withLinkageName1 somepkg.someFunction1
+func withLinkageName1() {
+}
+
+// Import a function from a different package using go:linkname.
+//go:linkname withLinkageName2 somepkg.someFunction2
+func withLinkageName2()
+
+// Function has an 'inline hint', similar to the inline keyword in C.
+//go:inline
+func inlineFunc() {
+}
+
+// Function should never be inlined, equivalent to GCC
+// __attribute__((noinline)).
+//go:noinline
+func noinlineFunc() {
+}
+
+// This function should have the specified section.
+//go:section .special_function_section
+func functionInSection() {
+}
+
+//export exportedFunctionInSection
+//go:section .special_function_section
+func exportedFunctionInSection() {
+}
+
+// This function should not: it's only a declaration and not a definition.
+//go:section .special_function_section
+func undefinedFunctionNotInSection()
+
+//go:section .special_global_section
+var globalInSection uint32
+
+//go:section .special_global_section
+//go:extern undefinedGlobalNotInSection
+var undefinedGlobalNotInSection uint32
+
+//go:align 1024
+//go:section .global_section
+var multipleGlobalPragmas uint32
@@ -0,0 +1,59 @@
+; ModuleID = 'pragma.go'
+source_filename = "pragma.go"
+target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
+target triple = "wasm32--wasi"
+
+@extern_global = external global [0 x i8], align 1
+@main.alignedGlobal = hidden global [4 x i32] zeroinitializer, align 32
+@main.alignedGlobal16 = hidden global [4 x i32] zeroinitializer, align 16
+@main.globalInSection = hidden global i32 0, section ".special_global_section", align 4
+@undefinedGlobalNotInSection = external global i32, align 4
+@main.multipleGlobalPragmas = hidden global i32 0, section ".global_section", align 1024
+
+declare noalias nonnull i8* @runtime.alloc(i32, i8*, i8*)
+
+define hidden void @main.init(i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  ret void
+}
+
+define void @extern_func() #0 {
+entry:
+  ret void
+}
+
+define hidden void @somepkg.someFunction1(i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  ret void
+}
+
+declare void @somepkg.someFunction2(i8*, i8*)
+
+; Function Attrs: inlinehint
+define hidden void @main.inlineFunc(i8* %context, i8* %parentHandle) unnamed_addr #1 {
+entry:
+  ret void
+}
+
+; Function Attrs: noinline
+define hidden void @main.noinlineFunc(i8* %context, i8* %parentHandle) unnamed_addr #2 {
+entry:
+  ret void
+}
+
+define hidden void @main.functionInSection(i8* %context, i8* %parentHandle) unnamed_addr section ".special_function_section" {
+entry:
+  ret void
+}
+
+define void @exportedFunctionInSection() #3 section ".special_function_section" {
+entry:
+  ret void
+}
+
+declare void @main.undefinedFunctionNotInSection(i8*, i8*)
+
+attributes #0 = { "wasm-export-name"="extern_func" }
+attributes #1 = { inlinehint }
+attributes #2 = { noinline }
+attributes #3 = { "wasm-export-name"="exportedFunctionInSection" }
@@ -23,3 +23,21 @@ func sliceAppendSlice(ints, added []int) []int {
 func sliceCopy(dst, src []int) int {
 	return copy(dst, src)
 }
+
+// Test bounds checking in *ssa.MakeSlice instruction.
+
+func makeByteSlice(len int) []byte {
+	return make([]byte, len)
+}
+
+func makeInt16Slice(len int) []int16 {
+	return make([]int16, len)
+}
+
+func makeArraySlice(len int) [][3]byte {
+	return make([][3]byte, len) // slice with element size of 3
+}
+
+func makeInt32Slice(len int) []int32 {
+	return make([]int32, len)
+}
@@ -1,7 +1,7 @@
 ; ModuleID = 'slice.go'
 source_filename = "slice.go"
-target datalayout = "e-m:e-p:32:32-p270:32:32-p271:32:32-p272:64:64-f64:32:64-f80:32-n8:16:32-S128"
-target triple = "i686--linux"
+target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
+target triple = "wasm32--wasi"

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

@@ -86,3 +86,79 @@ entry:
 }

 declare i32 @runtime.sliceCopy(i8* nocapture writeonly, i8* nocapture readonly, i32, i32, i32, i8*, i8*)
+
+define hidden { i8*, i32, i32 } @main.makeByteSlice(i32 %len, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %slice.maxcap = icmp slt i32 %len, 0
+  br i1 %slice.maxcap, label %slice.throw, label %slice.next
+
+slice.throw:                                      ; preds = %entry
+  call void @runtime.slicePanic(i8* undef, i8* null)
+  unreachable
+
+slice.next:                                       ; preds = %entry
+  %makeslice.buf = call i8* @runtime.alloc(i32 %len, i8* undef, i8* null)
+  %0 = insertvalue { i8*, i32, i32 } undef, i8* %makeslice.buf, 0
+  %1 = insertvalue { i8*, i32, i32 } %0, i32 %len, 1
+  %2 = insertvalue { i8*, i32, i32 } %1, i32 %len, 2
+  ret { i8*, i32, i32 } %2
+}
+
+declare void @runtime.slicePanic(i8*, i8*)
+
+define hidden { i16*, i32, i32 } @main.makeInt16Slice(i32 %len, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %slice.maxcap = icmp slt i32 %len, 0
+  br i1 %slice.maxcap, label %slice.throw, label %slice.next
+
+slice.throw:                                      ; preds = %entry
+  call void @runtime.slicePanic(i8* undef, i8* null)
+  unreachable
+
+slice.next:                                       ; preds = %entry
+  %makeslice.cap = shl i32 %len, 1
+  %makeslice.buf = call i8* @runtime.alloc(i32 %makeslice.cap, i8* undef, i8* null)
+  %makeslice.array = bitcast i8* %makeslice.buf to i16*
+  %0 = insertvalue { i16*, i32, i32 } undef, i16* %makeslice.array, 0
+  %1 = insertvalue { i16*, i32, i32 } %0, i32 %len, 1
+  %2 = insertvalue { i16*, i32, i32 } %1, i32 %len, 2
+  ret { i16*, i32, i32 } %2
+}
+
+define hidden { [3 x i8]*, i32, i32 } @main.makeArraySlice(i32 %len, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %slice.maxcap = icmp ugt i32 %len, 1431655765
+  br i1 %slice.maxcap, label %slice.throw, label %slice.next
+
+slice.throw:                                      ; preds = %entry
+  call void @runtime.slicePanic(i8* undef, i8* null)
+  unreachable
+
+slice.next:                                       ; preds = %entry
+  %makeslice.cap = mul i32 %len, 3
+  %makeslice.buf = call i8* @runtime.alloc(i32 %makeslice.cap, i8* undef, i8* null)
+  %makeslice.array = bitcast i8* %makeslice.buf to [3 x i8]*
+  %0 = insertvalue { [3 x i8]*, i32, i32 } undef, [3 x i8]* %makeslice.array, 0
+  %1 = insertvalue { [3 x i8]*, i32, i32 } %0, i32 %len, 1
+  %2 = insertvalue { [3 x i8]*, i32, i32 } %1, i32 %len, 2
+  ret { [3 x i8]*, i32, i32 } %2
+}
+
+define hidden { i32*, i32, i32 } @main.makeInt32Slice(i32 %len, i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  %slice.maxcap = icmp ugt i32 %len, 1073741823
+  br i1 %slice.maxcap, label %slice.throw, label %slice.next
+
+slice.throw:                                      ; preds = %entry
+  call void @runtime.slicePanic(i8* undef, i8* null)
+  unreachable
+
+slice.next:                                       ; preds = %entry
+  %makeslice.cap = shl i32 %len, 2
+  %makeslice.buf = call i8* @runtime.alloc(i32 %makeslice.cap, i8* undef, i8* null)
+  %makeslice.array = bitcast i8* %makeslice.buf to i32*
+  %0 = insertvalue { i32*, i32, i32 } undef, i32* %makeslice.array, 0
+  %1 = insertvalue { i32*, i32, i32 } %0, i32 %len, 1
+  %2 = insertvalue { i32*, i32, i32 } %1, i32 %len, 2
+  ret { i32*, i32, i32 } %2
+}
@@ -1,7 +1,7 @@
 ; ModuleID = 'string.go'
 source_filename = "string.go"
-target datalayout = "e-m:e-p:32:32-p270:32:32-p271:32:32-p272:64:64-f64:32:64-f80:32-n8:16:32-S128"
-target triple = "i686--linux"
+target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
+target triple = "wasm32--wasi"

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

@@ -1,6 +1,6 @@
 module github.com/tinygo-org/tinygo

-go 1.13
+go 1.15

 require (
 	github.com/blakesmith/ar v0.0.0-20150311145944-8bd4349a67f2
@@ -10,7 +10,7 @@ require (
 	github.com/marcinbor85/gohex v0.0.0-20200531091804-343a4b548892
 	github.com/mattn/go-colorable v0.1.8
 	go.bug.st/serial v1.1.2
-	golang.org/x/sys v0.0.0-20210113181707-4bcb84eeeb78
-	golang.org/x/tools v0.0.0-20200216192241-b320d3a0f5a2
+	golang.org/x/sys v0.0.0-20210510120138-977fb7262007
+	golang.org/x/tools v0.1.6-0.20210813165731-45389f592fe9
 	tinygo.org/x/go-llvm v0.0.0-20210325115028-e7b85195e81c
 )
@@ -30,6 +30,8 @@ github.com/mattn/go-isatty v0.0.12/go.mod h1:cbi8OIDigv2wuxKPP5vlRcQ1OAZbq2CE4Ky
 github.com/pmezard/go-difflib v1.0.0/go.mod h1:iKH77koFhYxTK1pcRnkKkqfTogsbg7gZNVY4sRDYZ/4=
 github.com/stretchr/objx v0.1.0/go.mod h1:HFkY916IF+rwdDfMAkV7OtwuqBVzrE8GR6GFx+wExME=
 github.com/stretchr/testify v1.4.0/go.mod h1:j7eGeouHqKxXV5pUuKE4zz7dFj8WfuZ+81PSLYec5m4=
+github.com/yuin/goldmark v1.3.5 h1:dPmz1Snjq0kmkz159iL7S6WzdahUTHnHB5M56WFVifs=
+github.com/yuin/goldmark v1.3.5/go.mod h1:mwnBkeHKe2W/ZEtQ+71ViKU8L12m81fl3OWwC1Zlc8k=
 go.bug.st/serial v1.0.0 h1:ogEPzrllCsnG00EqKRjeYvPRsO7NJW6DqykzkdD6E/k=
 go.bug.st/serial v1.0.0/go.mod h1:rpXPISGjuNjPTRTcMlxi9lN6LoIPxd1ixVjBd8aSk/Q=
 go.bug.st/serial v1.1.2 h1:6xDpbta8KJ+VLRTeM8ghhxXRMLE/Lr8h9iDKwydarAY=
@@ -38,23 +40,32 @@ golang.org/x/crypto v0.0.0-20190308221718-c2843e01d9a2/go.mod h1:djNgcEr1/C05ACk
 golang.org/x/crypto v0.0.0-20191011191535-87dc89f01550/go.mod h1:yigFU9vqHzYiE8UmvKecakEJjdnWj3jj499lnFckfCI=
 golang.org/x/mod v0.1.1-0.20191105210325-c90efee705ee h1:WG0RUwxtNT4qqaXX3DPA8zHFNm/D9xaBpxzHt1WcA/E=
 golang.org/x/mod v0.1.1-0.20191105210325-c90efee705ee/go.mod h1:QqPTAvyqsEbceGzBzNggFXnrqF1CaUcvgkdR5Ot7KZg=
+golang.org/x/mod v0.4.2/go.mod h1:s0Qsj1ACt9ePp/hMypM3fl4fZqREWJwdYDEqhRiZZUA=
 golang.org/x/net v0.0.0-20190404232315-eb5bcb51f2a3/go.mod h1:t9HGtf8HONx5eT2rtn7q6eTqICYqUVnKs3thJo3Qplg=
 golang.org/x/net v0.0.0-20190620200207-3b0461eec859/go.mod h1:z5CRVTTTmAJ677TzLLGU+0bjPO0LkuOLi4/5GtJWs/s=
+golang.org/x/net v0.0.0-20210405180319-a5a99cb37ef4/go.mod h1:p54w0d4576C0XHj96bSt6lcn1PtDYWL6XObtHCRCNQM=
 golang.org/x/sync v0.0.0-20190423024810-112230192c58/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
+golang.org/x/sync v0.0.0-20210220032951-036812b2e83c/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
 golang.org/x/sys v0.0.0-20190215142949-d0b11bdaac8a/go.mod h1:STP8DvDyc/dI5b8T5hshtkjS+E42TnysNCUPdjciGhY=
 golang.org/x/sys v0.0.0-20190412213103-97732733099d/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
-golang.org/x/sys v0.0.0-20191128015809-6d18c012aee9 h1:ZBzSG/7F4eNKz2L3GE9o300RX0Az1Bw5HF7PDraD+qU=
-golang.org/x/sys v0.0.0-20191128015809-6d18c012aee9/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.0.0-20200116001909-b77594299b42/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.0.0-20200223170610-d5e6a3e2c0ae/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.0.0-20200909081042-eff7692f9009/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
-golang.org/x/sys v0.0.0-20210113181707-4bcb84eeeb78 h1:nVuTkr9L6Bq62qpUqKo/RnZCFfzDBL0bYo6w9OJUqZY=
+golang.org/x/sys v0.0.0-20201119102817-f84b799fce68/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.0.0-20210113181707-4bcb84eeeb78/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
+golang.org/x/sys v0.0.0-20210330210617-4fbd30eecc44/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
+golang.org/x/sys v0.0.0-20210510120138-977fb7262007 h1:gG67DSER+11cZvqIMb8S8bt0vZtiN6xWYARwirrOSfE=
+golang.org/x/sys v0.0.0-20210510120138-977fb7262007/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
+golang.org/x/term v0.0.0-20201126162022-7de9c90e9dd1/go.mod h1:bj7SfCRtBDWHUb9snDiAeCFNEtKQo2Wmx5Cou7ajbmo=
 golang.org/x/text v0.3.0/go.mod h1:NqM8EUOU14njkJ3fqMW+pc6Ldnwhi/IjpwHt7yyuwOQ=
-golang.org/x/tools v0.0.0-20200216192241-b320d3a0f5a2 h1:0sfSpGSa544Fwnbot3Oxq/U6SXqjty6Jy/3wRhVS7ig=
-golang.org/x/tools v0.0.0-20200216192241-b320d3a0f5a2/go.mod h1:TB2adYChydJhpapKDTa4BR/hXlZSLoq2Wpct/0txZ28=
-golang.org/x/xerrors v0.0.0-20191011141410-1b5146add898 h1:/atklqdjdhuosWIl6AIbOeHJjicWYPqR9bpxqxYG2pA=
+golang.org/x/text v0.3.3/go.mod h1:5Zoc/QRtKVWzQhOtBMvqHzDpF6irO9z98xDceosuGiQ=
+golang.org/x/tools v0.0.0-20180917221912-90fa682c2a6e/go.mod h1:n7NCudcB/nEzxVGmLbDWY5pfWTLqBcC2KZ6jyYvM4mQ=
+golang.org/x/tools v0.0.0-20191119224855-298f0cb1881e/go.mod h1:b+2E5dAYhXwXZwtnZ6UAqBI28+e2cm9otk0dWdXHAEo=
+golang.org/x/tools v0.1.6-0.20210813165731-45389f592fe9 h1:nvvuMxmx1q0gfRki3T0hjG8EwAcVCs91oWAXvyt4zhI=
+golang.org/x/tools v0.1.6-0.20210813165731-45389f592fe9/go.mod h1:o0xws9oXOQQZyjljx8fwUC0k7L1pTE6eaCbjGeHmOkk=
+golang.org/x/xerrors v0.0.0-20190717185122-a985d3407aa7/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
 golang.org/x/xerrors v0.0.0-20191011141410-1b5146add898/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
+golang.org/x/xerrors v0.0.0-20200804184101-5ec99f83aff1/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
 gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
 gopkg.in/yaml.v2 v2.2.2/go.mod h1:hI93XBmqTisBFMUTm0b8Fm+jr3Dg1NNxqwp+5A1VGuI=
 tinygo.org/x/go-llvm v0.0.0-20210308112806-9ef958b6bed4 h1:CMUHxVTb+UuUePuMf8vkWjZ3gTp9BBK91KrgOCwoNHs=
@@ -12,7 +12,7 @@ import (

 // Version of TinyGo.
 // Update this value before release of new version of software.
-const Version = "0.18.0-dev"
+const Version = "0.20.0"

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

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

 		return string(matches[1]), nil

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

 // Enable extra checks, which should be disabled by default.
 // This may help track down bugs by adding a few more sanity checks.
@@ -110,17 +110,26 @@ func Run(mod llvm.Module, debug bool) error {
 			fmt.Fprintln(os.Stderr, "call:", fn.Name())
 		}
 		_, mem, callErr := r.run(r.getFunction(fn), nil, nil, "    ")
+		call.EraseFromParentAsInstruction()
 		if callErr != nil {
 			if isRecoverableError(callErr.Err) {
 				if r.debug {
 					fmt.Fprintln(os.Stderr, "not interpreting", r.pkgName, "because of error:", callErr.Error())
 				}
+				// Remove instructions that were created as part of interpreting
+				// the package.
 				mem.revert()
+				// Create a call to the package initializer (which was
+				// previously deleted).
+				i8undef := llvm.Undef(r.i8ptrType)
+				r.builder.CreateCall(fn, []llvm.Value{i8undef, i8undef}, "")
+				// Make sure that any globals touched by the package
+				// initializer, won't be accessed by later package initializers.
+				r.markExternalLoad(fn)
 				continue
 			}
 			return callErr
 		}
-		call.EraseFromParentAsInstruction()
 		for index, obj := range mem.objects {
 			r.objects[index] = obj
 		}
@@ -129,7 +138,7 @@ func Run(mod llvm.Module, debug bool) error {

 	// Update all global variables in the LLVM module.
 	mem := memoryView{r: r}
-	for _, obj := range r.objects {
+	for i, obj := range r.objects {
 		if obj.llvmGlobal.IsNil() {
 			continue
 		}
@@ -159,6 +168,12 @@ func Run(mod llvm.Module, debug bool) error {
 			name := obj.llvmGlobal.Name()
 			obj.llvmGlobal.EraseFromParentAsGlobal()
 			newGlobal.SetName(name)
+
+			// Update interp-internal references.
+			delete(r.globals, obj.llvmGlobal)
+			obj.llvmGlobal = newGlobal
+			r.globals[newGlobal] = i
+			r.objects[i] = obj
 			continue
 		}
 		if err != nil {
@@ -264,3 +279,19 @@ func (r *runner) getFunction(llvmFn llvm.Value) *function {
 	r.functionCache[llvmFn] = fn
 	return fn
 }
+
+// markExternalLoad marks the given llvmValue as being loaded externally. This
+// is primarily used to mark package initializers that could not be run at
+// compile time. As an example, a package initialize might store to a global
+// variable. Another package initializer might read from the same global
+// variable. By marking this function as being run at runtime, that load
+// instruction will need to be run at runtime instead of at compile time.
+func (r *runner) markExternalLoad(llvmValue llvm.Value) {
+	mem := memoryView{r: r}
+	mem.markExternalLoad(llvmValue)
+	for index, obj := range mem.objects {
+		if obj.marked > r.objects[index].marked {
+			r.objects[index].marked = obj.marked
+		}
+	}
+}
@@ -17,6 +17,7 @@ func TestInterp(t *testing.T) {
 		"slice-copy",
 		"consteval",
 		"interface",
+		"revert",
 	} {
 		name := name // make tc local to this closure
 		t.Run(name, func(t *testing.T) {
@@ -5,6 +5,7 @@ import (
 	"fmt"
 	"math"
 	"os"
+	"strconv"
 	"strings"
 	"time"

@@ -589,18 +590,14 @@ func (r *runner) run(fn *function, params []value, parentMem *memoryView, indent
 			// GetElementPtr does pointer arithmetic, changing the offset of the
 			// pointer into the underlying object.
 			var offset uint64
-			var gepOperands []uint64
 			for i := 2; i < len(operands); i += 2 {
 				index := operands[i].Uint()
 				elementSize := operands[i+1].Uint()
 				if int64(elementSize) < 0 {
 					// This is a struct field.
-					// The field number is encoded by flipping all the bits.
-					gepOperands = append(gepOperands, ^elementSize)
 					offset += index
 				} else {
 					// This is a normal GEP, probably an array index.
-					gepOperands = append(gepOperands, index)
 					offset += elementSize * index
 				}
 			}
@@ -930,16 +927,31 @@ func (r *runner) runAtRuntime(fn *function, inst instruction, locals []value, me
 		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")
+		// Note: the Go LLVM API doesn't support multiple indices, so simulate
+		// this operation with some extra extractvalue instructions. Hopefully
+		// this is optimized to a single instruction.
+		agg := operands[0]
+		for i := 0; i < len(indices)-1; i++ {
+			agg = r.builder.CreateExtractValue(agg, int(indices[i]), inst.name+".agg")
 		}
-		result = r.builder.CreateExtractValue(operands[0], int(indices[0]), inst.name)
+		result = r.builder.CreateExtractValue(agg, int(indices[len(indices)-1]), inst.name)
 	case llvm.InsertValue:
 		indices := inst.llvmInst.Indices()
-		if len(indices) != 1 {
-			panic("expected exactly one index")
+		// Similar to extractvalue, we're working around a limitation in the Go
+		// LLVM API here by splitting the insertvalue into multiple instructions
+		// if there is more than one operand.
+		agg := operands[0]
+		aggregates := []llvm.Value{agg}
+		for i := 0; i < len(indices)-1; i++ {
+			agg = r.builder.CreateExtractValue(agg, int(indices[i]), inst.name+".agg"+strconv.Itoa(i))
+			aggregates = append(aggregates, agg)
 		}
-		result = r.builder.CreateInsertValue(operands[0], operands[1], int(indices[0]), inst.name)
+		result = operands[1]
+		for i := len(indices) - 1; i >= 0; i-- {
+			agg := aggregates[i]
+			result = r.builder.CreateInsertValue(agg, result, int(indices[i]), inst.name+".insertvalue"+strconv.Itoa(i))
+		}
+
 	case llvm.Add:
 		result = r.builder.CreateAdd(operands[0], operands[1], inst.name)
 	case llvm.Sub:
@@ -178,6 +178,9 @@ func (mv *memoryView) markExternal(llvmValue llvm.Value, mark uint8) {
 		default:
 			panic("interp: unknown constant expression")
 		}
+	} else if !llvmValue.IsAInlineAsm().IsNil() {
+		// Inline assembly can modify globals but only exported globals. Let's
+		// hope the author knows what they're doing.
 	} else {
 		llvmType := llvmValue.Type()
 		switch llvmType.TypeKind() {
@@ -569,6 +572,17 @@ func (v pointerValue) toLLVMValue(llvmType llvm.Type, mem *memoryView) (llvm.Val
 	}

 	if llvmType.IsNil() {
+		if v.offset() != 0 {
+			// If there is an offset, make sure to use a GEP to index into the
+			// pointer. Because there is no expected type, we use whatever is
+			// most convenient: an *i8 type. It is trivial to index byte-wise.
+			if llvmValue.Type() != mem.r.i8ptrType {
+				llvmValue = llvm.ConstBitCast(llvmValue, mem.r.i8ptrType)
+			}
+			llvmValue = llvm.ConstInBoundsGEP(llvmValue, []llvm.Value{
+				llvm.ConstInt(llvmValue.Type().Context().Int32Type(), uint64(v.offset()), false),
+			})
+		}
 		return llvmValue, nil
 	}

@@ -8,6 +8,7 @@ target triple = "x86_64--linux"
@main.exportedValue = global [1 x i16*] [i16* @main.exposedValue1]
@main.exposedValue1 = global i16 0
@main.exposedValue2 = global i16 0
+@main.insertedValue = global {i8, i32, {float, {i64, i16}}} zeroinitializer

 declare void @runtime.printint64(i64) unnamed_addr

@@ -65,12 +66,22 @@ entry:
  call void @modifyExternal(i32* bitcast (void ()* @willModifyGlobal to i32*))
  store i16 7, i16* @main.exposedValue2

+  ; Test that inline assembly is ignored.
+  call void @modifyExternal(i32* bitcast (void ()* @hasInlineAsm to i32*))
+
  ; Test switch statement.
  %switch1 = call i64 @testSwitch(i64 1) ; 1 returns 6
  %switch2 = call i64 @testSwitch(i64 9) ; 9 returns the default value -1
  call void @runtime.printint64(i64 %switch1)
  call void @runtime.printint64(i64 %switch2)

+  ; Test extractvalue/insertvalue with multiple operands.
+  %agg = call {i8, i32, {float, {i64, i16}}} @nestedStruct()
+  %elt = extractvalue {i8, i32, {float, {i64, i16}}} %agg, 2, 1, 0
+  call void @runtime.printint64(i64 %elt)
+  %agg2 = insertvalue {i8, i32, {float, {i64, i16}}} %agg, i64 5, 2, 1, 0
+  store {i8, i32, {float, {i64, i16}}} %agg2, {i8, i32, {float, {i64, i16}}}* @main.insertedValue
+
  ret void
 }

@@ -94,6 +105,15 @@ entry:
  ret void
 }

+; Inline assembly should be ignored in the interp package. While it is possible
+; to modify other globals that way, usually that's not the case and there is no
+; real way to check.
+define void @hasInlineAsm() {
+entry:
+  call void asm sideeffect "", ""()
+  ret void
+}
+
 define i64 @testSwitch(i64 %val) {
 entry:
  ; Test switch statement.
@@ -112,3 +132,5 @@ two:
 otherwise:
  ret i64 -1
 }
+
+declare {i8, i32, {float, {i64, i16}}} @nestedStruct()
@@ -7,6 +7,7 @@ target triple = "x86_64--linux"
@main.exportedValue = global [1 x i16*] [i16* @main.exposedValue1]
@main.exposedValue1 = global i16 0
@main.exposedValue2 = local_unnamed_addr global i16 0
+@main.insertedValue = local_unnamed_addr global { i8, i32, { float, { i64, i16 } } } zeroinitializer

 declare void @runtime.printint64(i64) unnamed_addr

@@ -25,8 +26,20 @@ entry:
  store i16 5, i16* @main.exposedValue1
  call void @modifyExternal(i32* bitcast (void ()* @willModifyGlobal to i32*))
  store i16 7, i16* @main.exposedValue2
+  call void @modifyExternal(i32* bitcast (void ()* @hasInlineAsm to i32*))
  call void @runtime.printint64(i64 6)
  call void @runtime.printint64(i64 -1)
+  %agg = call { i8, i32, { float, { i64, i16 } } } @nestedStruct()
+  %elt.agg = extractvalue { i8, i32, { float, { i64, i16 } } } %agg, 2
+  %elt.agg1 = extractvalue { float, { i64, i16 } } %elt.agg, 1
+  %elt = extractvalue { i64, i16 } %elt.agg1, 0
+  call void @runtime.printint64(i64 %elt)
+  %agg2.agg0 = extractvalue { i8, i32, { float, { i64, i16 } } } %agg, 2
+  %agg2.agg1 = extractvalue { float, { i64, i16 } } %agg2.agg0, 1
+  %agg2.insertvalue2 = insertvalue { i64, i16 } %agg2.agg1, i64 5, 0
+  %agg2.insertvalue1 = insertvalue { float, { i64, i16 } } %agg2.agg0, { i64, i16 } %agg2.insertvalue2, 1
+  %agg2.insertvalue0 = insertvalue { i8, i32, { float, { i64, i16 } } } %agg, { float, { i64, i16 } } %agg2.insertvalue1, 2
+  store { i8, i32, { float, { i64, i16 } } } %agg2.insertvalue0, { i8, i32, { float, { i64, i16 } } }* @main.insertedValue
  ret void
 }

@@ -47,6 +60,12 @@ entry:
  ret void
 }

+define void @hasInlineAsm() {
+entry:
+  call void asm sideeffect "", ""()
+  ret void
+}
+
 define i64 @testSwitch(i64 %val) local_unnamed_addr {
 entry:
  switch i64 %val, label %otherwise [
@@ -67,3 +86,5 @@ two:                                              ; preds = %entry
 otherwise:                                        ; preds = %entry
  ret i64 -1
 }
+
+declare { i8, i32, { float, { i64, i16 } } } @nestedStruct() local_unnamed_addr
@@ -0,0 +1,32 @@
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64--linux"
+
+declare void @externalCall(i64)
+
+@foo.knownAtRuntime = global i64 0
+@bar.knownAtRuntime = global i64 0
+
+define void @runtime.initAll() unnamed_addr {
+entry:
+  call void @foo.init(i8* undef, i8* undef)
+  call void @bar.init(i8* undef, i8* undef)
+  call void @main.init(i8* undef, i8* undef)
+  ret void
+}
+
+define internal void @foo.init(i8* %context, i8* %parentHandle) unnamed_addr {
+  store i64 5, i64* @foo.knownAtRuntime
+  unreachable ; this triggers a revert of @foo.init.
+}
+
+define internal void @bar.init(i8* %context, i8* %parentHandle) unnamed_addr {
+  %val = load i64, i64* @foo.knownAtRuntime
+  store i64 %val, i64* @bar.knownAtRuntime
+  ret void
+}
+
+define internal void @main.init(i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+  call void @externalCall(i64 3)
+  ret void
+}
@@ -0,0 +1,21 @@
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64--linux"
+
+@foo.knownAtRuntime = local_unnamed_addr global i64 0
+@bar.knownAtRuntime = local_unnamed_addr global i64 0
+
+declare void @externalCall(i64) local_unnamed_addr
+
+define void @runtime.initAll() unnamed_addr {
+entry:
+  call fastcc void @foo.init(i8* undef, i8* undef)
+  %val = load i64, i64* @foo.knownAtRuntime, align 8
+  store i64 %val, i64* @bar.knownAtRuntime, align 8
+  call void @externalCall(i64 3)
+  ret void
+}
+
+define internal fastcc void @foo.init(i8* %context, i8* %parentHandle) unnamed_addr {
+  store i64 5, i64* @foo.knownAtRuntime, align 8
+  unreachable
+}
@@ -23,3 +23,13 @@ type Error struct {
 func (e Error) Error() string {
 	return e.Err.Error()
 }
+
+// Error returned when loading a *Program for a test binary but no test files
+// are present.
+type NoTestFilesError struct {
+	ImportPath string
+}
+
+func (e NoTestFilesError) Error() string {
+	return "no test files"
+}
@@ -2,6 +2,14 @@ package loader

 // This file constructs a new temporary GOROOT directory by merging both the
 // standard Go GOROOT and the GOROOT from TinyGo using symlinks.
+//
+// The goal is to replace specific packages from Go with a TinyGo version. It's
+// never a partial replacement, either a package is fully replaced or it is not.
+// This is important because if we did allow to merge packages (e.g. by adding
+// files to a package), it would lead to a dependency on implementation details
+// with all the maintenance burden that results in. Only allowing to replace
+// packages as a whole avoids this as packages are already designed to have a
+// public (backwards-compatible) API.

 import (
 	"crypto/sha512"
@@ -139,6 +147,7 @@ func mergeDirectory(goroot, tinygoroot, tmpgoroot, importPath string, overrides
 		if err != nil {
 			return err
 		}
+		hasTinyGoFiles := false
 		for _, e := range tinygoEntries {
 			if e.IsDir() {
 				// A directory, so merge this thing.
@@ -154,6 +163,7 @@ func mergeDirectory(goroot, tinygoroot, tmpgoroot, importPath string, overrides
 				if err != nil {
 					return err
 				}
+				hasTinyGoFiles = true
 			}
 		}

@@ -164,21 +174,30 @@ func mergeDirectory(goroot, tinygoroot, tmpgoroot, importPath string, overrides
 			return err
 		}
 		for _, e := range gorootEntries {
-			if !e.IsDir() {
-				// Don't merge in files from Go. Otherwise we'd end up with a
-				// weird syscall package with files from both roots.
-				continue
-			}
-			if _, ok := overrides[path.Join(importPath, e.Name())+"/"]; ok {
-				// Already included above, so don't bother trying to create this
-				// symlink.
-				continue
-			}
-			newname := filepath.Join(tmpgoroot, "src", importPath, e.Name())
-			oldname := filepath.Join(goroot, "src", importPath, e.Name())
-			err := symlink(oldname, newname)
-			if err != nil {
-				return err
+			if e.IsDir() {
+				if _, ok := overrides[path.Join(importPath, e.Name())+"/"]; ok {
+					// Already included above, so don't bother trying to create this
+					// symlink.
+					continue
+				}
+				newname := filepath.Join(tmpgoroot, "src", importPath, e.Name())
+				oldname := filepath.Join(goroot, "src", importPath, e.Name())
+				err := symlink(oldname, newname)
+				if err != nil {
+					return err
+				}
+			} else {
+				// Only merge files from Go if TinyGo does not have any files.
+				// Otherwise we'd end up with a weird mix from both Go
+				// implementations.
+				if !hasTinyGoFiles {
+					newname := filepath.Join(tmpgoroot, "src", importPath, e.Name())
+					oldname := filepath.Join(goroot, "src", importPath, e.Name())
+					err := symlink(oldname, newname)
+					if err != nil {
+						return err
+					}
+				}
 			}
 		}
 	}
@@ -189,7 +208,7 @@ 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" || tag == "wasi" {
+		if tag == "baremetal" || tag == "darwin" || tag == "nintendoswitch" || tag == "tinygo.wasm" {
 			return true
 		}
 	}
@@ -201,6 +220,8 @@ func needsSyscallPackage(buildTags []string) bool {
 func pathsToOverride(needsSyscallPackage bool) map[string]bool {
 	paths := map[string]bool{
 		"/":                     true,
+		"crypto/":               true,
+		"crypto/rand/":          false,
 		"device/":               false,
 		"examples/":             false,
 		"internal/":             true,
@@ -208,6 +229,7 @@ func pathsToOverride(needsSyscallPackage bool) map[string]bool {
 		"internal/reflectlite/": false,
 		"internal/task/":        false,
 		"machine/":              false,
+		"net/":                  true,
 		"os/":                   true,
 		"reflect/":              false,
 		"runtime/":              false,
@@ -210,6 +210,12 @@ func Load(config *compileopts.Config, inputPkgs []string, clangHeaders string, t
 		p.Packages[pkg.ImportPath] = pkg
 	}

+	if config.TestConfig.CompileTestBinary && !strings.HasSuffix(p.sorted[len(p.sorted)-1].ImportPath, ".test") {
+		// Trying to compile a test binary but there are no test files in this
+		// package.
+		return p, NoTestFilesError{p.sorted[len(p.sorted)-1].ImportPath}
+	}
+
 	return p, nil
 }

@@ -337,12 +343,11 @@ func (p *Package) Check() error {
 	checker.Importer = p

 	packageName := p.ImportPath
-	if p.Name == "main" {
-		// The main package normally has a different import path, such as
-		// "command-line-arguments" or "./testdata/cgo". Therefore, use the name
-		// "main" in such a case: this package isn't imported from anywhere.
-		// This is safe as it isn't possible to import a package with the name
-		// "main".
+	if p == p.program.MainPkg() {
+		if p.Name != "main" {
+			// Sanity check. Should not ever trigger.
+			panic("expected main package to have name 'main'")
+		}
 		packageName = "main"
 	}
 	typesPkg, err := checker.Check(packageName, p.program.fset, p.Files, &p.info)
@@ -381,7 +386,11 @@ func (p *Package) parseFiles() ([]*ast.File, error) {
 	}

 	// Do CGo processing.
-	if len(p.CgoFiles) != 0 {
+	// This is done when there are any CgoFiles at all. In that case, len(files)
+	// should be non-zero. However, if len(GoFiles) == 0 and len(CgoFiles) == 1
+	// and there is a syntax error in a CGo file, len(files) may be 0. Don't try
+	// to call cgo.Process in that case as it will only cause issues.
+	if len(p.CgoFiles) != 0 && len(files) != 0 {
 		var initialCFlags []string
 		initialCFlags = append(initialCFlags, p.program.config.CFlags()...)
 		initialCFlags = append(initialCFlags, "-I"+p.Dir)
@@ -15,6 +15,8 @@ import (
 	"path/filepath"
 	"regexp"
 	"runtime"
+	"runtime/pprof"
+	"strconv"
 	"strings"
 	"sync/atomic"
 	"time"
@@ -95,12 +97,31 @@ func copyFile(src, dst string) error {

 // executeCommand is a simple wrapper to exec.Cmd
 func executeCommand(options *compileopts.Options, name string, arg ...string) *exec.Cmd {
-	if options.PrintCommands {
-		fmt.Printf("%s %s\n", name, strings.Join(arg, " "))
+	if options.PrintCommands != nil {
+		options.PrintCommands(name, arg...)
 	}
 	return exec.Command(name, arg...)
 }

+// printCommand prints a command to stdout while formatting it like a real
+// command (escaping characters etc). The resulting command should be easy to
+// run directly in a shell, although it is not guaranteed to be a safe shell
+// escape. That's not a problem as the primary use case is printing the command,
+// not running it.
+func printCommand(cmd string, args ...string) {
+	command := append([]string{cmd}, args...)
+	for i, arg := range command {
+		// Source: https://www.oreilly.com/library/view/learning-the-bash/1565923472/ch01s09.html
+		const specialChars = "~`#$&*()\\|[]{};'\"<>?! "
+		if strings.ContainsAny(arg, specialChars) {
+			// See: https://stackoverflow.com/questions/15783701/which-characters-need-to-be-escaped-when-using-bash
+			arg = "'" + strings.ReplaceAll(arg, `'`, `'\''`) + "'"
+			command[i] = arg
+		}
+	}
+	fmt.Fprintln(os.Stderr, strings.Join(command, " "))
+}
+
 // Build compiles and links the given package and writes it to outpath.
 func Build(pkgName, outpath string, options *compileopts.Options) error {
 	config, err := builder.NewConfig(options)
@@ -136,15 +157,17 @@ func Build(pkgName, outpath string, options *compileopts.Options) error {
 	})
 }

-// Test runs the tests in the given package.
-func Test(pkgName string, options *compileopts.Options, testCompileOnly bool, outpath string) error {
+// Test runs the tests in the given package. Returns whether the test passed and
+// possibly an error if the test failed to run.
+func Test(pkgName string, options *compileopts.Options, testCompileOnly, testVerbose bool, outpath string) (bool, error) {
 	options.TestConfig.CompileTestBinary = true
 	config, err := builder.NewConfig(options)
 	if err != nil {
-		return err
+		return false, err
 	}

-	return builder.Build(pkgName, outpath, config, func(result builder.BuildResult) error {
+	passed := true
+	err = builder.Build(pkgName, outpath, config, func(result builder.BuildResult) error {
 		if testCompileOnly || outpath != "" {
 			// Write test binary to the specified file name.
 			if outpath == "" {
@@ -158,48 +181,83 @@ func Test(pkgName string, options *compileopts.Options, testCompileOnly bool, ou
 			// Do not run the test.
 			return nil
 		}
-		if len(config.Target.Emulator) == 0 {
-			// Run directly.
-			cmd := executeCommand(config.Options, 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 {
-					os.Exit(err.ExitCode())
-				}
-				return &commandError{"failed to run compiled binary", result.Binary, err}
-			}
-			return nil
+
+		// Run the test.
+		start := time.Now()
+		var err error
+		passed, err = runPackageTest(config, result, testVerbose)
+		if err != nil {
+			return err
+		}
+		duration := time.Since(start)
+
+		// Print the result.
+		importPath := strings.TrimSuffix(result.ImportPath, ".test")
+		if passed {
+			fmt.Printf("ok  \t%s\t%.3fs\n", importPath, duration.Seconds())
 		} else {
-			// Run in an emulator.
-			args := append(config.Target.Emulator[1:], result.Binary)
-			cmd := executeCommand(config.Options, 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}
-				}
+			fmt.Printf("FAIL\t%s\t%.3fs\n", importPath, duration.Seconds())
+		}
+		return nil
+	})
+	if err, ok := err.(loader.NoTestFilesError); ok {
+		fmt.Printf("?   \t%s\t[no test files]\n", err.ImportPath)
+		// Pretend the test passed - it at least didn't fail.
+		return true, nil
+	}
+	return passed, err
+}
+
+// runPackageTest runs a test binary that was previously built. The return
+// values are whether the test passed and any errors encountered while trying to
+// run the binary.
+func runPackageTest(config *compileopts.Config, result builder.BuildResult, testVerbose bool) (bool, error) {
+	if len(config.Target.Emulator) == 0 {
+		// Run directly.
+		var flags []string
+		if testVerbose {
+			flags = append(flags, "-test.v")
+		}
+		cmd := executeCommand(config.Options, result.Binary, flags...)
+		cmd.Stdout = os.Stdout
+		cmd.Stderr = os.Stderr
+		cmd.Dir = result.MainDir
+		err := cmd.Run()
+		if err != nil {
+			if _, ok := err.(*exec.ExitError); ok {
+				// Binary exited with a non-zero exit code, which means the test
+				// failed.
+				return false, nil
 			}
-			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 false, &commandError{"failed to run compiled binary", result.Binary, err}
+		}
+		return true, nil
+	} else {
+		// Run in an emulator.
+		// TODO: pass the -test.v flag if needed.
+		args := append(config.Target.Emulator[1:], result.Binary)
+		cmd := executeCommand(config.Options, 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 false, &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 true, nil
+		} else {
+			// Test failed, either by ending with the word "FAIL" or with a
+			// panic of some sort.
+			return false, nil
+		}
+	}
 }

 // Flash builds and flashes the built binary to the given serial port.
@@ -224,6 +282,8 @@ func Flash(pkgName, port string, options *compileopts.Options) error {
 			fileExt = ".bin"
 		case strings.Contains(config.Target.FlashCommand, "{uf2}"):
 			fileExt = ".uf2"
+		case strings.Contains(config.Target.FlashCommand, "{zip}"):
+			fileExt = ".zip"
 		default:
 			return errors.New("invalid target file - did you forget the {hex} token in the 'flash-command' section?")
 		}
@@ -234,6 +294,8 @@ func Flash(pkgName, port string, options *compileopts.Options) error {
 		fileExt = filepath.Ext(config.Target.FlashFilename)
 	case "openocd":
 		fileExt = ".hex"
+	case "bmp":
+		fileExt = ".elf"
 	case "native":
 		return errors.New("unknown flash method \"native\" - did you miss a -target flag?")
 	default:
@@ -243,7 +305,7 @@ func Flash(pkgName, port string, options *compileopts.Options) 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" && flashMethod != "openocd" {
-			port, err := getDefaultPort(strings.FieldsFunc(port, func(c rune) bool { return c == ',' }))
+			port, err := getDefaultPort(port, config.Target.SerialPort)
 			if err != nil {
 				return err
 			}
@@ -261,36 +323,36 @@ func Flash(pkgName, port string, options *compileopts.Options) error {
 		case "", "command":
 			// Create the command.
 			flashCmd := config.Target.FlashCommand
-			fileToken := "{" + fileExt[1:] + "}"
-			flashCmd = strings.ReplaceAll(flashCmd, fileToken, result.Binary)
+			flashCmdList, err := shlex.Split(flashCmd)
+			if err != nil {
+				return fmt.Errorf("could not parse flash command %#v: %w", flashCmd, err)
+			}

 			if strings.Contains(flashCmd, "{port}") {
 				var err error
-				port, err = getDefaultPort(strings.FieldsFunc(port, func(c rune) bool { return c == ',' }))
+				port, err = getDefaultPort(port, config.Target.SerialPort)
 				if err != nil {
 					return err
 				}
 			}

-			flashCmd = strings.ReplaceAll(flashCmd, "{port}", port)
-
-			// Execute the command.
-			var cmd *exec.Cmd
-			switch runtime.GOOS {
-			case "windows":
-				command := strings.Split(flashCmd, " ")
-				if len(command) < 2 {
-					return errors.New("invalid flash command")
-				}
-				cmd = executeCommand(config.Options, command[0], command[1:]...)
-			default:
-				cmd = executeCommand(config.Options, "/bin/sh", "-c", flashCmd)
+			// Fill in fields in the command template.
+			fileToken := "{" + fileExt[1:] + "}"
+			for i, arg := range flashCmdList {
+				arg = strings.ReplaceAll(arg, fileToken, result.Binary)
+				arg = strings.ReplaceAll(arg, "{port}", port)
+				flashCmdList[i] = arg
 			}

+			// Execute the command.
+			if len(flashCmdList) < 2 {
+				return fmt.Errorf("invalid flash command: %#v", flashCmd)
+			}
+			cmd := executeCommand(config.Options, flashCmdList[0], flashCmdList[1:]...)
 			cmd.Stdout = os.Stdout
 			cmd.Stderr = os.Stderr
 			cmd.Dir = goenv.Get("TINYGOROOT")
-			err := cmd.Run()
+			err = cmd.Run()
 			if err != nil {
 				return &commandError{"failed to flash", result.Binary, err}
 			}
@@ -326,6 +388,25 @@ func Flash(pkgName, port string, options *compileopts.Options) error {
 				return &commandError{"failed to flash", result.Binary, err}
 			}
 			return nil
+		case "bmp":
+			gdb, err := config.Target.LookupGDB()
+			if err != nil {
+				return err
+			}
+			var bmpGDBPort string
+			bmpGDBPort, _, err = getBMPPorts()
+			if err != nil {
+				return err
+			}
+			args := []string{"-ex", "target extended-remote " + bmpGDBPort, "-ex", "monitor swdp_scan", "-ex", "attach 1", "-ex", "load", filepath.ToSlash(result.Binary)}
+			cmd := executeCommand(config.Options, gdb, args...)
+			cmd.Stdout = os.Stdout
+			cmd.Stderr = os.Stderr
+			err = cmd.Run()
+			if err != nil {
+				return &commandError{"failed to flash", result.Binary, err}
+			}
+			return nil
 		default:
 			return fmt.Errorf("unknown flash method: %s", flashMethod)
 		}
@@ -380,8 +461,15 @@ func FlashGDB(pkgName string, ocdOutput bool, options *compileopts.Options) erro
 		switch gdbInterface {
 		case "native":
 			// Run GDB directly.
+		case "bmp":
+			var bmpGDBPort string
+			bmpGDBPort, _, err = getBMPPorts()
+			if err != nil {
+				return err
+			}
+			gdbCommands = append(gdbCommands, "target extended-remote "+bmpGDBPort, "monitor swdp_scan", "compare-sections", "attach 1", "load")
 		case "openocd":
-			gdbCommands = append(gdbCommands, "target remote :3333", "monitor halt", "load", "monitor reset halt")
+			gdbCommands = append(gdbCommands, "target extended-remote :3333", "monitor halt", "load", "monitor reset halt")

 			// We need a separate debugging daemon for on-chip debugging.
 			args, err := config.OpenOCDConfiguration()
@@ -400,7 +488,7 @@ func FlashGDB(pkgName string, ocdOutput bool, options *compileopts.Options) erro
 				daemon.Stderr = w
 			}
 		case "jlink":
-			gdbCommands = append(gdbCommands, "target remote :2331", "load", "monitor reset halt")
+			gdbCommands = append(gdbCommands, "target extended-remote :2331", "load", "monitor reset halt")

 			// We need a separate debugging daemon for on-chip debugging.
 			daemon = executeCommand(config.Options, "JLinkGDBServer", "-device", config.Target.JLinkDevice)
@@ -415,7 +503,7 @@ func FlashGDB(pkgName string, ocdOutput bool, options *compileopts.Options) erro
 				daemon.Stderr = w
 			}
 		case "qemu":
-			gdbCommands = append(gdbCommands, "target remote :1234")
+			gdbCommands = append(gdbCommands, "target extended-remote :1234")

 			// Run in an emulator.
 			args := append(config.Target.Emulator[1:], result.Binary, "-s", "-S")
@@ -423,7 +511,7 @@ func FlashGDB(pkgName string, ocdOutput bool, options *compileopts.Options) erro
 			daemon.Stdout = os.Stdout
 			daemon.Stderr = os.Stderr
 		case "qemu-user":
-			gdbCommands = append(gdbCommands, "target remote :1234")
+			gdbCommands = append(gdbCommands, "target extended-remote :1234")

 			// Run in an emulator.
 			args := append(config.Target.Emulator[1:], "-g", "1234", result.Binary)
@@ -431,7 +519,7 @@ func FlashGDB(pkgName string, ocdOutput bool, options *compileopts.Options) erro
 			daemon.Stdout = os.Stdout
 			daemon.Stderr = os.Stderr
 		case "mgba":
-			gdbCommands = append(gdbCommands, "target remote :2345")
+			gdbCommands = append(gdbCommands, "target extended-remote :2345")

 			// Run in an emulator.
 			args := append(config.Target.Emulator[1:], result.Binary, "-g")
@@ -439,7 +527,7 @@ func FlashGDB(pkgName string, ocdOutput bool, options *compileopts.Options) erro
 			daemon.Stdout = os.Stdout
 			daemon.Stderr = os.Stderr
 		case "simavr":
-			gdbCommands = append(gdbCommands, "target remote :1234")
+			gdbCommands = append(gdbCommands, "target extended-remote :1234")

 			// Run in an emulator.
 			args := append(config.Target.Emulator[1:], "-g", result.Binary)
@@ -665,7 +753,8 @@ func windowsFindUSBDrive(volume string, options *compileopts.Options) (string, e
 }

 // getDefaultPort returns the default serial port depending on the operating system.
-func getDefaultPort(portCandidates []string) (port string, err error) {
+func getDefaultPort(portFlag string, usbInterfaces []string) (port string, err error) {
+	portCandidates := strings.FieldsFunc(portFlag, func(c rune) bool { return c == ',' })
 	if len(portCandidates) == 1 {
 		return portCandidates[0], nil
 	}
@@ -681,11 +770,70 @@ func getDefaultPort(portCandidates []string) (port string, err error) {
 			return "", err
 		}

-		for _, p := range portsList {
-			ports = append(ports, p.Name)
+		var preferredPortIDs [][2]uint16
+		for _, s := range usbInterfaces {
+			parts := strings.Split(s, ":")
+			if len(parts) != 3 || (parts[0] != "acm" && parts[0] == "usb") {
+				// acm and usb are the two types of serial ports recognized
+				// under Linux (ttyACM*, ttyUSB*). Other operating systems don't
+				// generally make this distinction. If this is not one of the
+				// given USB devices, don't try to parse the USB IDs.
+				continue
+			}
+			vid, err := strconv.ParseUint(parts[1], 16, 16)
+			if err != nil {
+				return "", fmt.Errorf("could not parse USB vendor ID %q: %w", parts[1], err)
+			}
+			pid, err := strconv.ParseUint(parts[2], 16, 16)
+			if err != nil {
+				return "", fmt.Errorf("could not parse USB product ID %q: %w", parts[1], err)
+			}
+			preferredPortIDs = append(preferredPortIDs, [2]uint16{uint16(vid), uint16(pid)})
 		}

-		if ports == nil || len(ports) == 0 {
+		var primaryPorts []string   // ports picked from preferred USB VID/PID
+		var secondaryPorts []string // other ports (as a fallback)
+		for _, p := range portsList {
+			if !p.IsUSB {
+				continue
+			}
+			if p.VID != "" && p.PID != "" {
+				foundPort := false
+				vid, vidErr := strconv.ParseUint(p.VID, 16, 16)
+				pid, pidErr := strconv.ParseUint(p.PID, 16, 16)
+				if vidErr == nil && pidErr == nil {
+					for _, id := range preferredPortIDs {
+						if uint16(vid) == id[0] && uint16(pid) == id[1] {
+							primaryPorts = append(primaryPorts, p.Name)
+							foundPort = true
+							continue
+						}
+					}
+				}
+				if foundPort {
+					continue
+				}
+			}
+
+			secondaryPorts = append(secondaryPorts, p.Name)
+		}
+		if len(primaryPorts) == 1 {
+			// There is exactly one match in the set of preferred ports. Use
+			// this port, even if there may be others available. This allows
+			// flashing a specific board even if there are multiple available.
+			return primaryPorts[0], nil
+		} else if len(primaryPorts) > 1 {
+			// There are multiple preferred ports, probably because more than
+			// one device of the same type are connected (e.g. two Arduino
+			// Unos).
+			ports = primaryPorts
+		} else {
+			// No preferred ports found. Fall back to other serial ports
+			// available in the system.
+			ports = secondaryPorts
+		}
+
+		if len(ports) == 0 {
 			// fallback
 			switch runtime.GOOS {
 			case "darwin":
@@ -727,6 +875,35 @@ func getDefaultPort(portCandidates []string) (port string, err error) {
 	return "", errors.New("port you specified '" + strings.Join(portCandidates, ",") + "' does not exist, available ports are " + strings.Join(ports, ", "))
 }

+// getBMPPorts returns BlackMagicProbe's serial ports if any
+func getBMPPorts() (gdbPort, uartPort string, err error) {
+	var portsList []*enumerator.PortDetails
+	portsList, err = enumerator.GetDetailedPortsList()
+	if err != nil {
+		return "", "", err
+	}
+	var ports []string
+	for _, p := range portsList {
+		if !p.IsUSB {
+			continue
+		}
+		if p.VID != "" && p.PID != "" {
+			vid, vidErr := strconv.ParseUint(p.VID, 16, 16)
+			pid, pidErr := strconv.ParseUint(p.PID, 16, 16)
+			if vidErr == nil && pidErr == nil && vid == 0x1d50 && pid == 0x6018 {
+				ports = append(ports, p.Name)
+			}
+		}
+	}
+	if len(ports) == 2 {
+		return ports[0], ports[1], nil
+	} else if len(ports) == 0 {
+		return "", "", errors.New("no BMP detected")
+	} else {
+		return "", "", fmt.Errorf("expected 2 BMP serial ports, found %d - did you perhaps connect more than one BMP?", len(ports))
+	}
+}
+
 func usage() {
 	fmt.Fprintln(os.Stderr, "TinyGo is a Go compiler for small places.")
 	fmt.Fprintln(os.Stderr, "version:", goenv.Version)
@@ -895,6 +1072,7 @@ func main() {
 	gc := flag.String("gc", "", "garbage collector to use (none, leaking, extalloc, conservative)")
 	panicStrategy := flag.String("panic", "print", "panic strategy (print, trap)")
 	scheduler := flag.String("scheduler", "", "which scheduler to use (none, coroutines, tasks)")
+	serial := flag.String("serial", "", "which serial output to use (none, uart, usb)")
 	printIR := flag.Bool("printir", false, "print LLVM IR")
 	dumpSSA := flag.Bool("dumpssa", false, "dump internal Go SSA")
 	verifyIR := flag.Bool("verifyir", false, "run extra verification steps on LLVM IR")
@@ -904,25 +1082,30 @@ func main() {
 	printStacks := flag.Bool("print-stacks", false, "print stack sizes of goroutines")
 	printAllocsString := flag.String("print-allocs", "", "regular expression of functions for which heap allocations should be printed")
 	printCommands := flag.Bool("x", false, "Print commands")
-	nodebug := flag.Bool("no-debug", false, "disable DWARF debug symbol generation")
+	nodebug := flag.Bool("no-debug", false, "strip debug information")
+	ocdCommandsString := flag.String("ocd-commands", "", "OpenOCD commands, overriding target spec (can specify multiple separated by commas)")
 	ocdOutput := flag.Bool("ocd-output", false, "print OCD daemon output during debug")
 	port := flag.String("port", "", "flash port (can specify multiple candidates separated by commas)")
 	programmer := flag.String("programmer", "", "which hardware programmer to use")
 	ldflags := flag.String("ldflags", "", "Go link tool compatible ldflags")
 	wasmAbi := flag.String("wasm-abi", "", "WebAssembly ABI conventions: js (no i64 params) or generic")
+	llvmFeatures := flag.String("llvm-features", "", "comma separated LLVM features to enable")
+	cpuprofile := flag.String("cpuprofile", "", "cpuprofile output")

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

 	// Early command processing, before commands are interpreted by the Go flag
@@ -943,6 +1126,7 @@ func main() {
 		fmt.Fprintln(os.Stderr, err)
 		os.Exit(1)
 	}
+
 	var printAllocs *regexp.Regexp
 	if *printAllocsString != "" {
 		printAllocs, err = regexp.Compile(*printAllocsString)
@@ -951,24 +1135,35 @@ func main() {
 			os.Exit(1)
 		}
 	}
+
+	var ocdCommands []string
+	if *ocdCommandsString != "" {
+		ocdCommands = strings.Split(*ocdCommandsString, ",")
+	}
+
 	options := &compileopts.Options{
-		Target:        *target,
-		Opt:           *opt,
-		GC:            *gc,
-		PanicStrategy: *panicStrategy,
-		Scheduler:     *scheduler,
-		PrintIR:       *printIR,
-		DumpSSA:       *dumpSSA,
-		VerifyIR:      *verifyIR,
-		Debug:         !*nodebug,
-		PrintSizes:    *printSize,
-		PrintStacks:   *printStacks,
-		PrintAllocs:   printAllocs,
-		PrintCommands: *printCommands,
-		Tags:          *tags,
-		GlobalValues:  globalVarValues,
-		WasmAbi:       *wasmAbi,
-		Programmer:    *programmer,
+		Target:          *target,
+		Opt:             *opt,
+		GC:              *gc,
+		PanicStrategy:   *panicStrategy,
+		Scheduler:       *scheduler,
+		Serial:          *serial,
+		PrintIR:         *printIR,
+		DumpSSA:         *dumpSSA,
+		VerifyIR:        *verifyIR,
+		Debug:           !*nodebug,
+		PrintSizes:      *printSize,
+		PrintStacks:     *printStacks,
+		PrintAllocs:     printAllocs,
+		Tags:            *tags,
+		GlobalValues:    globalVarValues,
+		WasmAbi:         *wasmAbi,
+		Programmer:      *programmer,
+		OpenOCDCommands: ocdCommands,
+		LLVMFeatures:    *llvmFeatures,
+	}
+	if *printCommands {
+		options.PrintCommands = printCommand
 	}

 	os.Setenv("CC", "clang -target="+*target)
@@ -980,6 +1175,20 @@ func main() {
 		os.Exit(1)
 	}

+	if *cpuprofile != "" {
+		f, err := os.Create(*cpuprofile)
+		if err != nil {
+			fmt.Fprintln(os.Stderr, "could not create CPU profile: ", err)
+			os.Exit(1)
+		}
+		defer f.Close()
+		if err := pprof.StartCPUProfile(f); err != nil {
+			fmt.Fprintln(os.Stderr, "could not start CPU profile: ", err)
+			os.Exit(1)
+		}
+		defer pprof.StopCPUProfile()
+	}
+
 	switch command {
 	case "build":
 		if outpath == "" {
@@ -1061,16 +1270,26 @@ func main() {
 		err := Run(pkgName, options)
 		handleCompilerError(err)
 	case "test":
-		pkgName := "."
-		if flag.NArg() == 1 {
-			pkgName = filepath.ToSlash(flag.Arg(0))
-		} else if flag.NArg() > 1 {
-			fmt.Fprintln(os.Stderr, "test only accepts a single positional argument: package name, but multiple were specified")
-			usage()
+		var pkgNames []string
+		for i := 0; i < flag.NArg(); i++ {
+			pkgNames = append(pkgNames, filepath.ToSlash(flag.Arg(i)))
+		}
+		if len(pkgNames) == 0 {
+			pkgNames = []string{"."}
+		}
+		allTestsPassed := true
+		for _, pkgName := range pkgNames {
+			// TODO: parallelize building the test binaries
+			passed, err := Test(pkgName, options, *testCompileOnlyFlag, *testVerboseFlag, outpath)
+			handleCompilerError(err)
+			if !passed {
+				allTestsPassed = false
+			}
+		}
+		if !allTestsPassed {
+			fmt.Println("FAIL")
 			os.Exit(1)
 		}
-		err := Test(pkgName, options, *testCompileOnlyFlag, outpath)
-		handleCompilerError(err)
 	case "targets":
 		dir := filepath.Join(goenv.Get("TINYGOROOT"), "targets")
 		entries, err := ioutil.ReadDir(dir)
@@ -1145,6 +1364,9 @@ func main() {
 		if *flagDeps {
 			extraArgs = append(extraArgs, "-deps")
 		}
+		if *flagTest {
+			extraArgs = append(extraArgs, "-test")
+		}
 		cmd, err := loader.List(config, extraArgs, flag.Args())
 		if err != nil {
 			fmt.Fprintln(os.Stderr, "failed to run `go list`:", err)
@@ -35,9 +35,9 @@ func TestCompiler(t *testing.T) {
 		"calls.go",
 		"cgo/",
 		"channel.go",
-		"coroutines.go",
 		"float.go",
 		"gc.go",
+		"goroutines.go",
 		"init.go",
 		"init_multi.go",
 		"interface.go",
@@ -51,9 +51,18 @@ func TestCompiler(t *testing.T) {
 		"stdlib.go",
 		"string.go",
 		"structs.go",
+		"testing.go",
 		"zeroalloc.go",
 	}

+	_, minor, err := goenv.GetGorootVersion(goenv.Get("GOROOT"))
+	if err != nil {
+		t.Fatal("could not read version from GOROOT:", err)
+	}
+	if minor >= 17 {
+		tests = append(tests, "go1.17.go")
+	}
+
 	if *testTarget != "" {
 		// This makes it possible to run one specific test (instead of all),
 		// which is especially useful to quickly check whether some changes
@@ -94,21 +103,9 @@ func TestCompiler(t *testing.T) {
 		t.Run("ARM64Linux", func(t *testing.T) {
 			runPlatTests("aarch64--linux-gnu", tests, t)
 		})
-		goVersion, err := goenv.GorootVersionString(goenv.Get("GOROOT"))
-		if err != nil {
-			t.Error("could not get Go version:", err)
-			return
-		}
-		minorVersion := strings.Split(goVersion, ".")[1]
-		if minorVersion != "13" {
-			// WebAssembly tests fail on Go 1.13, so skip them there. Versions
-			// below that are also not supported but still seem to pass, so
-			// include them in the tests for now.
-			t.Run("WebAssembly", func(t *testing.T) {
-				runPlatTests("wasm", tests, t)
-			})
-		}
-
+		t.Run("WebAssembly", func(t *testing.T) {
+			runPlatTests("wasm", tests, t)
+		})
 		t.Run("WASI", func(t *testing.T) {
 			runPlatTests("wasi", tests, t)
 		})
@@ -125,7 +122,7 @@ func TestCompiler(t *testing.T) {
 		// Test with few optimizations enabled (no inlining, etc).
 		t.Run("opt=1", func(t *testing.T) {
 			t.Parallel()
-			runTestWithConfig("stdlib.go", "", t, &compileopts.Options{
+			runTestWithConfig("stdlib.go", "", t, compileopts.Options{
 				Opt: "1",
 			}, nil, nil)
 		})
@@ -134,15 +131,14 @@ func TestCompiler(t *testing.T) {
 		// TODO: fix this for stdlib.go, which currently fails.
 		t.Run("opt=0", func(t *testing.T) {
 			t.Parallel()
-			runTestWithConfig("print.go", "", t, &compileopts.Options{
+			runTestWithConfig("print.go", "", t, compileopts.Options{
 				Opt: "0",
 			}, nil, nil)
 		})

 		t.Run("ldflags", func(t *testing.T) {
 			t.Parallel()
-			runTestWithConfig("ldflags.go", "", t, &compileopts.Options{
-				Opt: "z",
+			runTestWithConfig("ldflags.go", "", t, compileopts.Options{
 				GlobalValues: map[string]map[string]string{
 					"main": {
 						"someGlobal": "foobar",
@@ -163,14 +159,20 @@ func runPlatTests(target string, tests []string, t *testing.T) {
 			runTest(name, target, t, nil, nil)
 		})
 	}
-	if target == "wasi" || target == "" {
+	t.Run("env.go", func(t *testing.T) {
+		t.Parallel()
+		runTest("env.go", target, t, []string{"first", "second"}, []string{"ENV1=VALUE1", "ENV2=VALUE2"})
+	})
+	if target == "" || target == "wasi" {
 		t.Run("filesystem.go", func(t *testing.T) {
 			t.Parallel()
 			runTest("filesystem.go", target, t, nil, nil)
 		})
-		t.Run("env.go", func(t *testing.T) {
+	}
+	if target == "" || target == "wasi" || target == "wasm" {
+		t.Run("rand.go", func(t *testing.T) {
 			t.Parallel()
-			runTest("env.go", target, t, []string{"first", "second"}, []string{"ENV1=VALUE1", "ENV2=VALUE2"})
+			runTest("rand.go", target, t, nil, nil)
 		})
 	}
 }
@@ -188,20 +190,20 @@ func runBuild(src, out string, opts *compileopts.Options) error {
 }

 func runTest(name, target string, t *testing.T, cmdArgs, environmentVars []string) {
-	options := &compileopts.Options{
-		Target:     target,
-		Opt:        "z",
-		PrintIR:    false,
-		DumpSSA:    false,
-		VerifyIR:   true,
-		Debug:      true,
-		PrintSizes: "",
-		WasmAbi:    "",
+	options := compileopts.Options{
+		Target: target,
 	}
 	runTestWithConfig(name, target, t, options, cmdArgs, environmentVars)
 }

-func runTestWithConfig(name, target string, t *testing.T, options *compileopts.Options, cmdArgs, environmentVars []string) {
+func runTestWithConfig(name, target string, t *testing.T, options compileopts.Options, cmdArgs, environmentVars []string) {
+	// Set default config.
+	options.Debug = true
+	options.VerifyIR = true
+	if options.Opt == "" {
+		options.Opt = "z"
+	}
+
 	// Get the expected output for this test.
 	// Note: not using filepath.Join as it strips the path separator at the end
 	// of the path.
@@ -217,57 +219,78 @@ func runTestWithConfig(name, target string, t *testing.T, options *compileopts.O
 	}

 	// Create a temporary directory for test output files.
-	tmpdir, err := ioutil.TempDir("", "tinygo-test")
+	tmpdir := t.TempDir()
+
+	// Determine whether we're on a system that supports environment variables
+	// and command line parameters (operating systems, WASI) or not (baremetal,
+	// WebAssembly in the browser). If we're on a system without an environment,
+	// we need to pass command line arguments and environment variables through
+	// global variables (built into the binary directly) instead of the
+	// conventional way.
+	spec, err := compileopts.LoadTarget(target)
 	if err != nil {
-		t.Fatal("could not create temporary directory:", err)
+		t.Fatal("failed to load target spec:", err)
 	}
-	defer func() {
-		rerr := os.RemoveAll(tmpdir)
-		if rerr != nil {
-			t.Errorf("failed to remove temporary directory %q: %s", tmpdir, rerr.Error())
+	needsEnvInVars := spec.GOOS == "js"
+	for _, tag := range spec.BuildTags {
+		if tag == "baremetal" {
+			needsEnvInVars = true
 		}
-	}()
+	}
+	if needsEnvInVars {
+		runtimeGlobals := make(map[string]string)
+		if len(cmdArgs) != 0 {
+			runtimeGlobals["osArgs"] = strings.Join(cmdArgs, "\x00")
+		}
+		if len(environmentVars) != 0 {
+			runtimeGlobals["osEnv"] = strings.Join(environmentVars, "\x00")
+		}
+		if len(runtimeGlobals) != 0 {
+			// This sets the global variables like they would be set with
+			// `-ldflags="-X=runtime.osArgs=first\x00second`.
+			// The runtime package has two variables (osArgs and osEnv) that are
+			// both strings, from which the parameters and environment variables
+			// are read.
+			options.GlobalValues = map[string]map[string]string{
+				"runtime": runtimeGlobals,
+			}
+		}
+	}

 	// Build the test binary.
 	binary := filepath.Join(tmpdir, "test")
-	err = runBuild("./"+path, binary, options)
+	err = runBuild("./"+path, binary, &options)
 	if err != nil {
 		printCompilerError(t.Log, err)
 		t.Fail()
 		return
 	}

-	// Run the test.
-	runComplete := make(chan struct{})
+	// Create the test command, taking care of emulators etc.
 	var cmd *exec.Cmd
-	ranTooLong := false
-	if target == "" {
+	if len(spec.Emulator) == 0 {
 		cmd = exec.Command(binary)
-		cmd.Env = append(cmd.Env, environmentVars...)
+	} else {
+		args := append(spec.Emulator[1:], binary)
+		cmd = exec.Command(spec.Emulator[0], args...)
+	}
+	if len(spec.Emulator) != 0 && spec.Emulator[0] == "wasmtime" {
+		// Allow reading from the current directory.
+		cmd.Args = append(cmd.Args, "--dir=.")
+		for _, v := range environmentVars {
+			cmd.Args = append(cmd.Args, "--env", v)
+		}
 		cmd.Args = append(cmd.Args, cmdArgs...)
 	} else {
-		spec, err := compileopts.LoadTarget(target)
-		if err != nil {
-			t.Fatal("failed to load target spec:", err)
-		}
-		if len(spec.Emulator) == 0 {
-			cmd = exec.Command(binary)
-		} else {
-			args := append(spec.Emulator[1:], binary)
-			cmd = exec.Command(spec.Emulator[0], args...)
-		}
-
-		if len(spec.Emulator) != 0 && spec.Emulator[0] == "wasmtime" {
-			// Allow reading from the current directory.
-			cmd.Args = append(cmd.Args, "--dir=.")
-			for _, v := range environmentVars {
-				cmd.Args = append(cmd.Args, "--env", v)
-			}
-			cmd.Args = append(cmd.Args, cmdArgs...)
-		} else {
+		if !needsEnvInVars {
+			cmd.Args = append(cmd.Args, cmdArgs...) // works on qemu-aarch64 etc
 			cmd.Env = append(cmd.Env, environmentVars...)
 		}
 	}
+
+	// Run the test.
+	runComplete := make(chan struct{})
+	ranTooLong := false
 	stdout := &bytes.Buffer{}
 	cmd.Stdout = stdout
 	cmd.Stderr = os.Stderr
@@ -0,0 +1,19 @@
+// Copyright 2010 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Package rand implements a cryptographically secure
+// random number generator.
+package rand
+
+import "io"
+
+// Reader is a global, shared instance of a cryptographically
+// secure random number generator.
+var Reader io.Reader
+
+// Read is a helper function that calls Reader.Read using io.ReadFull.
+// On return, n == len(b) if and only if err == nil.
+func Read(b []byte) (n int, err error) {
+	return io.ReadFull(Reader, b)
+}
@@ -0,0 +1,30 @@
+// +build darwin freebsd tinygo.wasm
+
+// This implementation of crypto/rand uses the arc4random_buf function
+// (available on both MacOS and WASI) to generate random numbers.
+//
+// Note: arc4random_buf (unlike what the name suggets) does not use the insecure
+// RC4 cipher. Instead, it uses a high-quality cipher, varying by the libc
+// implementation.
+
+package rand
+
+import "unsafe"
+
+func init() {
+	Reader = &reader{}
+}
+
+type reader struct {
+}
+
+func (r *reader) Read(b []byte) (n int, err error) {
+	if len(b) != 0 {
+		libc_arc4random_buf(unsafe.Pointer(&b[0]), uint(len(b)))
+	}
+	return len(b), nil
+}
+
+// void arc4random_buf(void *buf, size_t buflen);
+//export arc4random_buf
+func libc_arc4random_buf(buf unsafe.Pointer, buflen uint)
@@ -0,0 +1,38 @@
+// +build linux,!baremetal,!wasi
+
+// This implementation of crypto/rand uses the /dev/urandom pseudo-file to
+// generate random numbers.
+// TODO: convert to the getentropy or getrandom libc function on Linux once it
+// is more widely supported.
+
+package rand
+
+import (
+	"syscall"
+)
+
+func init() {
+	Reader = &reader{}
+}
+
+type reader struct {
+	fd int
+}
+
+func (r *reader) Read(b []byte) (n int, err error) {
+	if len(b) == 0 {
+		return
+	}
+
+	// Open /dev/urandom first if needed.
+	if r.fd == 0 {
+		fd, err := syscall.Open("/dev/urandom", syscall.O_RDONLY, 0)
+		if err != nil {
+			return 0, err
+		}
+		r.fd = fd
+	}
+
+	// Read from the file.
+	return syscall.Read(r.fd, b)
+}
@@ -0,0 +1,143 @@
+// Copyright 2011 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package rand
+
+import (
+	"errors"
+	"io"
+	"math/big"
+)
+
+// smallPrimes is a list of small, prime numbers that allows us to rapidly
+// exclude some fraction of composite candidates when searching for a random
+// prime. This list is truncated at the point where smallPrimesProduct exceeds
+// a uint64. It does not include two because we ensure that the candidates are
+// odd by construction.
+var smallPrimes = []uint8{
+	3, 5, 7, 11, 13, 17, 19, 23, 29, 31, 37, 41, 43, 47, 53,
+}
+
+// smallPrimesProduct is the product of the values in smallPrimes and allows us
+// to reduce a candidate prime by this number and then determine whether it's
+// coprime to all the elements of smallPrimes without further big.Int
+// operations.
+var smallPrimesProduct = new(big.Int).SetUint64(16294579238595022365)
+
+// Prime returns a number, p, of the given size, such that p is prime
+// with high probability.
+// Prime will return error for any error returned by rand.Read or if bits < 2.
+func Prime(rand io.Reader, bits int) (p *big.Int, err error) {
+	if bits < 2 {
+		err = errors.New("crypto/rand: prime size must be at least 2-bit")
+		return
+	}
+
+	b := uint(bits % 8)
+	if b == 0 {
+		b = 8
+	}
+
+	bytes := make([]byte, (bits+7)/8)
+	p = new(big.Int)
+
+	bigMod := new(big.Int)
+
+	for {
+		_, err = io.ReadFull(rand, bytes)
+		if err != nil {
+			return nil, err
+		}
+
+		// Clear bits in the first byte to make sure the candidate has a size <= bits.
+		bytes[0] &= uint8(int(1<<b) - 1)
+		// Don't let the value be too small, i.e, set the most significant two bits.
+		// Setting the top two bits, rather than just the top bit,
+		// means that when two of these values are multiplied together,
+		// the result isn't ever one bit short.
+		if b >= 2 {
+			bytes[0] |= 3 << (b - 2)
+		} else {
+			// Here b==1, because b cannot be zero.
+			bytes[0] |= 1
+			if len(bytes) > 1 {
+				bytes[1] |= 0x80
+			}
+		}
+		// Make the value odd since an even number this large certainly isn't prime.
+		bytes[len(bytes)-1] |= 1
+
+		p.SetBytes(bytes)
+
+		// Calculate the value mod the product of smallPrimes. If it's
+		// a multiple of any of these primes we add two until it isn't.
+		// The probability of overflowing is minimal and can be ignored
+		// because we still perform Miller-Rabin tests on the result.
+		bigMod.Mod(p, smallPrimesProduct)
+		mod := bigMod.Uint64()
+
+	NextDelta:
+		for delta := uint64(0); delta < 1<<20; delta += 2 {
+			m := mod + delta
+			for _, prime := range smallPrimes {
+				if m%uint64(prime) == 0 && (bits > 6 || m != uint64(prime)) {
+					continue NextDelta
+				}
+			}
+
+			if delta > 0 {
+				bigMod.SetUint64(delta)
+				p.Add(p, bigMod)
+			}
+			break
+		}
+
+		// There is a tiny possibility that, by adding delta, we caused
+		// the number to be one bit too long. Thus we check BitLen
+		// here.
+		if p.ProbablyPrime(20) && p.BitLen() == bits {
+			return
+		}
+	}
+}
+
+// Int returns a uniform random value in [0, max). It panics if max <= 0.
+func Int(rand io.Reader, max *big.Int) (n *big.Int, err error) {
+	if max.Sign() <= 0 {
+		panic("crypto/rand: argument to Int is <= 0")
+	}
+	n = new(big.Int)
+	n.Sub(max, n.SetUint64(1))
+	// bitLen is the maximum bit length needed to encode a value < max.
+	bitLen := n.BitLen()
+	if bitLen == 0 {
+		// the only valid result is 0
+		return
+	}
+	// k is the maximum byte length needed to encode a value < max.
+	k := (bitLen + 7) / 8
+	// b is the number of bits in the most significant byte of max-1.
+	b := uint(bitLen % 8)
+	if b == 0 {
+		b = 8
+	}
+
+	bytes := make([]byte, k)
+
+	for {
+		_, err = io.ReadFull(rand, bytes)
+		if err != nil {
+			return nil, err
+		}
+
+		// Clear bits in the first byte to increase the probability
+		// that the candidate is < max.
+		bytes[0] &= uint8(int(1<<b) - 1)
+
+		n.SetBytes(bytes)
+		if n.Cmp(max) < 0 {
+			return
+		}
+	}
+}
@@ -26,12 +26,21 @@ type SCB_Type struct {
 	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
+	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
+	AFSR  volatile.Register32    // 0xD3C: Auxiliary Fault Status Register
+	PFR   [2]volatile.Register32 // 0xD40: Processor Feature Register
+	DFR   volatile.Register32    // 0xD48: Debug Feature Register
+	ADR   volatile.Register32    // 0xD4C: Auxiliary Feature Register
+	MMFR  [4]volatile.Register32 // 0xD50: Memory Model Feature Register
+	ISAR  [5]volatile.Register32 // 0xD60: Instruction Set Attributes Register
+	_     [5]uint32              // reserved
+	CPACR volatile.Register32    // 0xD88: Coprocessor Access Control Register
+
 }

 var SCB = (*SCB_Type)(unsafe.Pointer(uintptr(SCB_BASE)))
@@ -0,0 +1,49 @@
+// This is a very minimal bootloader for the ESP32-C3. It only initializes the
+// flash and then continues with the generic RISC-V initialization code, which
+// in turn will call runtime.main.
+// It is written in assembly (and not in a higher level language) to make sure
+// it is entirely loaded into IRAM and doesn't accidentally call functions
+// stored in IROM.
+//
+// For reference, here is a nice introduction into RISC-V assembly:
+// https://www.imperialviolet.org/2016/12/31/riscv.html
+
+.section .init
+.global call_start_cpu0
+.type call_start_cpu0,@function
+call_start_cpu0:
+    // At this point:
+    // - The ROM bootloader is finished and has jumped to here.
+    // - We're running from IRAM: both IRAM and DRAM segments have been loaded
+    //   by the ROM bootloader.
+    // - We have a usable stack (but not the one we would like to use).
+    // - No flash mappings (MMU) are set up yet.
+
+    // Reset MMU, see bootloader_reset_mmu in the ESP-IDF.
+    call Cache_Suspend_ICache
+    mv s0, a0 // autoload value
+    call Cache_Invalidate_ICache_All
+    call Cache_MMU_Init
+
+    // Set up DROM from flash.
+    // Somehow, this also sets up IROM from flash. Not sure why, but it avoids
+    // the need for another such call.
+    // C equivalent:
+    //   Cache_Dbus_MMU_Set(MMU_ACCESS_FLASH, 0x3C00_0000, 0, 64, 128, 0)
+    li a0, 0              // ext_ram: MMU_ACCESS_FLASH
+    li a1, 0x3C000000     // vaddr: address in the data bus
+    li a2, 0              // paddr: physical address in the flash chip
+    li a3, 64             // psize: always 64 (kilobytes)
+    li a4, 128            // num: pages to be set (8192K / 64K = 128)
+    li a5, 0              // fixed
+    call Cache_Dbus_MMU_Set
+
+    // Enable the flash cache.
+    mv a0, s0 // restore autoload value from Cache_Suspend_ICache call
+    call Cache_Resume_ICache
+
+    // Jump to generic RISC-V initialization, which initializes the stack
+    // pointer and globals register. It should not return.
+    // (It appears that the linker relaxes this jump and instead inserts the
+    // _start function right after here).
+    j _start
@@ -2,6 +2,7 @@
 // Type definitions, fields, and constants associated with the MPU peripheral
 // of the NXP MIMXRT1062.

+//go:build nxp && mimxrt1062
 // +build nxp,mimxrt1062

 package nxp
@@ -216,6 +217,12 @@ func enableIcache(enable bool) {
 	}
 }

+var (
+	dcacheCcsidr volatile.Register32
+	dcacheSets   volatile.Register32
+	dcacheWays   volatile.Register32
+)
+
 func enableDcache(enable bool) {
 	if enable != SystemControl.CCR.HasBits(SCB_CCR_DC_Msk) {
 		if enable {
@@ -244,11 +251,6 @@ func enableDcache(enable bool) {
        isb 0xF
      `, nil)
 		} else {
-			var (
-				ccsidr volatile.Register32
-				sets   volatile.Register32
-				ways   volatile.Register32
-			)
 			SystemControl.CSSELR.Set(0)
 			arm.AsmFull(`
        dsb 0xF
@@ -257,17 +259,17 @@ func enableDcache(enable bool) {
 			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 {
+			dcacheCcsidr.Set(SystemControl.CCSIDR.Get())
+			dcacheSets.Set((dcacheCcsidr.Get() & SCB_CCSIDR_NUMSETS_Msk) >> SCB_CCSIDR_NUMSETS_Pos)
+			for dcacheSets.Get() != 0 {
+				dcacheWays.Set((dcacheCcsidr.Get() & SCB_CCSIDR_ASSOCIATIVITY_Msk) >> SCB_CCSIDR_ASSOCIATIVITY_Pos)
+				for dcacheWays.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)
+						((dcacheSets.Get() << SCB_DCCISW_SET_Pos) & SCB_DCCISW_SET_Msk) |
+							((dcacheWays.Get() << SCB_DCCISW_WAY_Pos) & SCB_DCCISW_WAY_Msk))
+					dcacheWays.Set(dcacheWays.Get() - 1)
 				}
-				sets.Set(sets.Get() - 1)
+				dcacheSets.Set(dcacheSets.Get() - 1)
 			}
 			arm.AsmFull(`
        dsb 0xF
@@ -0,0 +1,15 @@
+// +build feather_m4_can
+
+package main
+
+import (
+	"machine"
+)
+
+func init() {
+	// power on the CAN Transceiver
+	// https://learn.adafruit.com/adafruit-feather-m4-can-express/pinouts#can-bus-3078990-8
+	boost_en := machine.BOOST_EN
+	boost_en.Configure(machine.PinConfig{Mode: machine.PinOutput})
+	boost_en.High()
+}
@@ -0,0 +1,53 @@
+package main
+
+import (
+	"fmt"
+	"machine"
+	"time"
+)
+
+func main() {
+	can1 := machine.CAN1
+	can1.Configure(machine.CANConfig{
+		TransferRate:   machine.CANTransferRate500kbps,
+		TransferRateFD: machine.CANTransferRate1000kbps,
+		Rx:             machine.CAN1_RX,
+		Tx:             machine.CAN1_TX,
+		Standby:        machine.CAN1_STANDBY,
+	})
+
+	can0 := machine.CAN0
+	can0.Configure(machine.CANConfig{
+		TransferRate:   machine.CANTransferRate500kbps,
+		TransferRateFD: machine.CANTransferRate1000kbps,
+		Rx:             machine.CAN0_RX,
+		Tx:             machine.CAN0_TX,
+		Standby:        machine.NoPin,
+	})
+
+	rxMsg := machine.CANRxBufferElement{}
+
+	for {
+		can1.Tx(0x123, []byte{0x01, 0x23, 0x45, 0x67, 0x89, 0xAB, 0xCD, 0xEF}, false, false)
+		can1.Tx(0x789, []byte{0x02, 0x24, 0x46, 0x67, 0x89, 0xAB, 0xCD, 0xEF, 0x01, 0x23, 0x45, 0x67, 0x89, 0xAB, 0xCD, 0xEF}, true, false)
+		time.Sleep(time.Millisecond * 1000)
+
+		sz0 := can0.RxFifoSize()
+		if sz0 > 0 {
+			fmt.Printf("CAN0 %d\r\n", sz0)
+			for i := 0; i < sz0; i++ {
+				can0.RxRaw(&rxMsg)
+				fmt.Printf("-> %08X %X %#v\r\n", rxMsg.ID, rxMsg.DLC, rxMsg.Data())
+			}
+		}
+
+		sz1 := can1.RxFifoSize()
+		if sz1 > 0 {
+			fmt.Printf("CAN1 %d\r\n", sz1)
+			for i := 0; i < sz1; i++ {
+				can1.RxRaw(&rxMsg)
+				fmt.Printf("-> %08X %X %#v\r\n", rxMsg.ID, rxMsg.DLC, rxMsg.Data())
+			}
+		}
+	}
+}
@@ -0,0 +1,15 @@
+// +build feather_m4_can
+
+package main
+
+import (
+	"machine"
+)
+
+func init() {
+	// power on the CAN Transceiver
+	// https://learn.adafruit.com/adafruit-feather-m4-can-express/pinouts#can-bus-3078990-8
+	boost_en := machine.BOOST_EN
+	boost_en.Configure(machine.PinConfig{Mode: machine.PinOutput})
+	boost_en.High()
+}
@@ -0,0 +1,75 @@
+package main
+
+import (
+	"device/sam"
+	"fmt"
+	"machine"
+	"time"
+)
+
+type canMsg struct {
+	ch   byte
+	id   uint32
+	dlc  byte
+	data []byte
+}
+
+func main() {
+	ch := make(chan canMsg, 10)
+	go func() {
+		for {
+			select {
+			case m := <-ch:
+				fmt.Printf("%d %03X %X ", m.ch, m.id, m.dlc)
+				for _, d := range m.data {
+					fmt.Printf("%02X ", d)
+				}
+				fmt.Printf("\r\n")
+			}
+
+		}
+	}()
+
+	can1 := machine.CAN1
+	can1.Configure(machine.CANConfig{
+		TransferRate:   machine.CANTransferRate500kbps,
+		TransferRateFD: machine.CANTransferRate1000kbps,
+		Rx:             machine.CAN1_RX,
+		Tx:             machine.CAN1_TX,
+		Standby:        machine.CAN1_STANDBY,
+	})
+	// RF0NE : Rx FIFO 0 New Message Interrupt Enable
+	can1.SetInterrupt(sam.CAN_IE_RF0NE, func(*machine.CAN) {
+		rxMsg := machine.CANRxBufferElement{}
+		can1.RxRaw(&rxMsg)
+		m := canMsg{ch: 1, id: rxMsg.ID, dlc: rxMsg.DLC, data: rxMsg.Data()}
+		select {
+		case ch <- m:
+		}
+	})
+
+	can0 := machine.CAN0
+	can0.Configure(machine.CANConfig{
+		TransferRate:   machine.CANTransferRate500kbps,
+		TransferRateFD: machine.CANTransferRate1000kbps,
+		Rx:             machine.CAN0_RX,
+		Tx:             machine.CAN0_TX,
+		Standby:        machine.NoPin,
+	})
+	// RF0NE : Rx FIFO 0 New Message Interrupt Enable
+	can0.SetInterrupt(sam.CAN_IE_RF0NE, func(*machine.CAN) {
+		rxMsg := machine.CANRxBufferElement{}
+		can0.RxRaw(&rxMsg)
+		m := canMsg{ch: 2, id: rxMsg.ID, dlc: rxMsg.DLC, data: rxMsg.Data()}
+		select {
+		case ch <- m:
+		}
+	})
+
+	for {
+		can0.Tx(0x123, []byte{0x01, 0x23, 0x45, 0x67, 0x89, 0xAB, 0xCD, 0xEF}, false, false)
+		time.Sleep(time.Millisecond * 500)
+		can1.Tx(0x456, []byte{0xAA, 0xBB, 0xCC}, false, false)
+		time.Sleep(time.Millisecond * 1000)
+	}
+}
@@ -9,7 +9,7 @@ import (

 // change these to test a different UART or pins if available
 var (
-	uart = machine.UART0
+	uart = machine.Serial
 	tx   = machine.UART_TX_PIN
 	rx   = machine.UART_RX_PIN
 )
@@ -0,0 +1,32 @@
+package main
+
+import (
+	"math/rand"
+	"runtime"
+	"time"
+)
+
+func main() {
+
+	ms := runtime.MemStats{}
+
+	for {
+		escapesToHeap()
+		runtime.ReadMemStats(&ms)
+		println("Heap before GC. Used: ", ms.HeapInuse, " Free: ", ms.HeapIdle, " Meta: ", ms.GCSys)
+		runtime.GC()
+		runtime.ReadMemStats(&ms)
+		println("Heap after  GC. Used: ", ms.HeapInuse, " Free: ", ms.HeapIdle, " Meta: ", ms.GCSys)
+		time.Sleep(5 * time.Second)
+	}
+
+}
+
+func escapesToHeap() {
+	n := rand.Intn(100)
+	println("Doing ", n, " iterations")
+	for i := 0; i < n; i++ {
+		s := make([]byte, i)
+		_ = append(s, 42)
+	}
+}
@@ -0,0 +1,10 @@
+// +build stm32
+
+package main
+
+import "machine"
+
+const (
+	buttonMode      = machine.PinInputPulldown
+	buttonPinChange = machine.PinRising | machine.PinFalling
+)
@@ -0,0 +1,11 @@
+// +build bluepill
+
+package main
+
+import "machine"
+
+var (
+	pwm  = &machine.TIM2
+	pinA = machine.PA0
+	pinB = machine.PA1
+)
@@ -0,0 +1,11 @@
+// +build stm32f7
+
+package main
+
+import "machine"
+
+var (
+	pwm  = &machine.TIM1
+	pinA = machine.PA8
+	pinB = machine.PA9
+)
@@ -0,0 +1,11 @@
+// +build stm32l0
+
+package main
+
+import "machine"
+
+var (
+	pwm  = &machine.TIM2
+	pinA = machine.PA0
+	pinB = machine.PB3
+)
@@ -0,0 +1,11 @@
+// +build stm32l4
+
+package main
+
+import "machine"
+
+var (
+	pwm  = &machine.TIM2
+	pinA = machine.PA0
+	pinB = machine.PB3
+)
@@ -0,0 +1,11 @@
+// +build stm32l5
+
+package main
+
+import "machine"
+
+var (
+	pwm  = &machine.TIM1
+	pinA = machine.PA8
+	pinB = machine.PA9
+)
@@ -0,0 +1,11 @@
+// +build pico
+
+package main
+
+import "machine"
+
+var (
+	pwm  = machine.PWM4 // Pin 25 (LED on pico) corresponds to PWM4.
+	pinA = machine.LED
+	pinB = machine.GPIO24
+)
@@ -0,0 +1,13 @@
+// +build stm32f4disco
+
+package main
+
+import "machine"
+
+var (
+	// These pins correspond to LEDs on the discovery
+	// board
+	pwm  = &machine.TIM4
+	pinA = machine.PD12
+	pinB = machine.PD13
+)
@@ -13,7 +13,7 @@ type Task struct {
 	Ptr unsafe.Pointer

 	// Data is a field which can be used for storing state information.
-	Data uint
+	Data uint64

 	// state is the underlying running state of the task.
 	state state
@@ -0,0 +1,58 @@
+.section .text.tinygo_startTask
+.global  tinygo_startTask
+.type    tinygo_startTask, %function
+tinygo_startTask:
+    .cfi_startproc
+    // Small assembly stub for starting a goroutine. This is already run on the
+    // new stack, with the callee-saved registers already loaded.
+    // Most importantly, EBX contain the pc of the to-be-started function and
+    // ESI contain the only argument it is given. Multiple arguments are packed
+    // into one by storing them in a new allocation.
+
+    // Indicate to the unwinder that there is nothing to unwind, this is the
+    // root frame. It avoids bogus extra frames in GDB.
+    .cfi_undefined eip
+
+    // Set the first argument of the goroutine start wrapper, which contains all
+    // the arguments.
+    pushl %esi
+
+    // Branch to the "goroutine start" function.
+    calll *%ebx
+
+    // Rebalance the stack (to undo the above push).
+    addl $4, %esp
+
+    // After return, exit this goroutine. This is a tail call.
+    jmp tinygo_pause
+    .cfi_endproc
+
+.global tinygo_swapTask
+.type tinygo_swapTask, %function
+tinygo_swapTask:
+    // This function gets the following parameters:
+    movl 4(%esp), %eax // newStack uintptr
+    movl 8(%esp), %ecx // oldStack *uintptr
+    // More information on the calling convention:
+    // https://wiki.osdev.org/System_V_ABI#i386
+
+    // Save all callee-saved registers:
+    pushl %ebp
+    pushl %edi
+    pushl %esi
+    pushl %ebx
+
+    // Save the current stack pointer in oldStack.
+    movl %esp, (%ecx)
+
+    // Switch to the new stack pointer.
+    movl %eax, %esp
+
+    // Load saved register from the new stack.
+    popl %ebx
+    popl %esi
+    popl %edi
+    popl %ebp
+
+    // Return into the new task, as if tinygo_swapTask was a regular call.
+    ret
@@ -0,0 +1,62 @@
+// +build scheduler.tasks,386
+
+package task
+
+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_386.S that relies on the exact
+// layout of this struct.
+type calleeSavedRegs struct {
+	ebx uintptr
+	esi uintptr
+	edi uintptr
+	ebp uintptr
+
+	pc uintptr
+
+	// Pad this struct so that tasks start on a 16-byte aligned stack.
+	_ [3]uintptr
+}
+
+// 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_386.S). This assembly code calls a function
+	// (passed in EBX) with a single argument (passed in ESI). After the
+	// function returns, it calls Pause().
+	r.pc = uintptr(unsafe.Pointer(&startTask))
+
+	// Pass the function to call in EBX.
+	// 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.ebx = fn
+
+	// Pass the pointer to the arguments struct in ESI.
+	r.esi = 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
+}
--- a/Show More
+++ b/Show More