main: version 0.7.1

This release contains a small fix for some atsamd21-based boards to make
them usable on macOS: it now allows to specify the serial port using the
-port flag.

.circleci/config.yml

@@ -40,57 +40,34 @@ commands:
             sudo tar -C /usr/local -xf node-v10.15.1-linux-x64.tar.xz
             sudo ln -s /usr/local/node-v10.15.1-linux-x64/bin/node /usr/bin/node
             rm node-v10.15.1-linux-x64.tar.xz
-  dep:
+  llvm-source-linux:
     steps:
+      - restore_cache:
+          keys:
+            - llvm-source-8-v5
       - run:
-          name: "Install Go dependencies"
-          command: |
-            curl https://raw.githubusercontent.com/golang/dep/master/install.sh | sh
-            dep ensure --vendor-only
-  smoketest:
-    steps:
-      - smoketest-no-avr
-      - run: tinygo build -size short -o test.elf -target=arduino             examples/blinky1
-      - run: tinygo build -size short -o test.elf -target=digispark           examples/blinky1
-  smoketest-no-avr:
-    steps:
-      - run: tinygo build -size short -o test.elf -target=pca10040            examples/blinky1
-      - run: tinygo build -size short -o test.elf -target=pca10040            examples/blinky2
-      - run: tinygo build             -o blinky2                              examples/blinky2 # TODO: re-enable -size flag with MachO support
-      - run: tinygo build -size short -o test.elf -target=pca10040            examples/test
-      - run: tinygo build -size short -o test.elf -target=microbit            examples/echo
-      - run: tinygo build -size short -o test.elf -target=nrf52840-mdk        examples/blinky1
-      - run: tinygo build -size short -o test.elf -target=pca10031            examples/blinky1
-      - run: tinygo build -size short -o test.elf -target=bluepill            examples/blinky1
-      - run: tinygo build -size short -o test.elf -target=reelboard           examples/blinky1
-      - run: tinygo build -size short -o test.elf -target=reelboard           examples/blinky2
-      - run: tinygo build -size short -o test.elf -target=pca10056            examples/blinky1
-      - run: tinygo build -size short -o test.elf -target=pca10056            examples/blinky2
-      - run: tinygo build -size short -o test.elf -target=itsybitsy-m0        examples/blinky1
-      - run: tinygo build -size short -o test.elf -target=circuitplay-express examples/blinky1
-      - run: tinygo build -size short -o test.elf -target=stm32f4disco        examples/blinky1
-      - run: tinygo build -size short -o test.elf -target=stm32f4disco        examples/blinky2
-      - run: tinygo build             -o wasm.wasm -target=wasm               examples/wasm/export
-      - run: tinygo build             -o wasm.wasm -target=wasm               examples/wasm/main
+          name: "Fetch LLVM source"
+          command: make llvm-source
+      - save_cache:
+          key: llvm-source-8-v5
+          paths:
+            - llvm-project
   test-linux:
-    parameters:
-      llvm:
-        type: string
     steps:
       - checkout
       - submodules
       - apt-dependencies:
-          llvm: <<parameters.llvm>>
+          llvm: "-8"
       - install-node
       - restore_cache:
           keys:
             - go-cache-{{ checksum "Gopkg.lock" }}-{{ .Environment.CIRCLE_PREVIOUS_BUILD_NUM }}
             - go-cache-{{ checksum "Gopkg.lock" }}
-      - dep
+      - llvm-source-linux
       - run: go install .
       - run: go test -v
       - run: make gen-device -j4
-      - smoketest
+      - run: make smoketest RISCV=0
       - save_cache:
           key: go-cache-{{ checksum "Gopkg.lock" }}-{{ .Environment.CIRCLE_BUILD_NUM }}
           paths:
@@ -105,7 +82,6 @@ commands:
           name: "Install apt dependencies"
           command: |
             sudo apt-get install \
-                libtinfo-dev \
                 python3 \
                 gcc-arm-linux-gnueabihf \
                 binutils-arm-none-eabi \
@@ -121,19 +97,10 @@ commands:
           keys:
             - go-cache-{{ checksum "Gopkg.lock" }}-{{ .Environment.CIRCLE_PREVIOUS_BUILD_NUM }}
             - go-cache-{{ checksum "Gopkg.lock" }}
+      - llvm-source-linux
       - restore_cache:
           keys:
-            - llvm-source-8-v2
-      - run:
-          name: "Fetch LLVM source"
-          command: make llvm-source
-      - save_cache:
-          key: llvm-source-8-v2
-          paths:
-            - llvm
-      - restore_cache:
-          keys:
-            - llvm-build-8-v2
+            - llvm-build-8-linux-v7
       - run:
           name: "Build LLVM"
           command: |
@@ -151,7 +118,7 @@ commands:
               make llvm-build
             fi
       - save_cache:
-          key: llvm-build-8-v2
+          key: llvm-build-8-linux-v7
           paths:
             llvm-build
       - run:
@@ -160,7 +127,6 @@ commands:
             ln -s $PWD/llvm-build/bin/clang-8 /go/bin/clang-8
             ln -s $PWD/llvm-build/bin/ld.lld  /go/bin/ld.lld-8
             ln -s $PWD/llvm-build/bin/wasm-ld /go/bin/wasm-ld-8
-      - dep
       - run:
           name: "Test TinyGo"
           command: make test
@@ -183,7 +149,12 @@ commands:
             tar -C ~/lib -xf /tmp/tinygo.linux-amd64.tar.gz
             ln -s ~/lib/tinygo/bin/tinygo /go/bin/tinygo
             tinygo version
-      - smoketest
+      - run:
+          name: "Download SiFive GNU toolchain"
+          command: |
+            curl -O https://static.dev.sifive.com/dev-tools/riscv64-unknown-elf-gcc-8.2.0-2019.05.3-x86_64-linux-ubuntu14.tar.gz
+            sudo tar -C /usr/local --strip-components=1 -xf riscv64-unknown-elf-gcc-8.2.0-2019.05.3-x86_64-linux-ubuntu14.tar.gz
+      - run: make smoketest
   build-macos:
     steps:
       - checkout
@@ -191,20 +162,23 @@ commands:
       - run:
           name: "Install dependencies"
           command: |
-            HOMEBREW_NO_AUTO_UPDATE=1 brew install go dep qemu
+            curl https://dl.google.com/go/go1.12.5.darwin-amd64.tar.gz -o go1.12.5.darwin-amd64.tar.gz
+            sudo tar -C /usr/local -xzf go1.12.5.darwin-amd64.tar.gz
+            ln -s /usr/local/go/bin/go /usr/local/bin/go
+            HOMEBREW_NO_AUTO_UPDATE=1 brew install qemu
       - restore_cache:
           keys:
-            - llvm-source-8-macos-v2
+            - llvm-source-8-macos-v5
       - run:
           name: "Fetch LLVM source"
           command: make llvm-source
       - save_cache:
-          key: llvm-source-8-macos-v2
+          key: llvm-source-8-macos-v5
           paths:
-            - llvm
+            - llvm-project
       - restore_cache:
           keys:
-            - llvm-build-8-macos-v2
+            - llvm-build-8-macos-v6
       - run:
           name: "Build LLVM"
           command: |
@@ -216,16 +190,13 @@ commands:
               make llvm-build
             fi
       - save_cache:
-          key: llvm-build-8-macos-v2
+          key: llvm-build-8-macos-v6
           paths:
             llvm-build
       - run:
           name: "Create LLVM symlinks"
           command: |
             ln -s $PWD/llvm-build/bin/clang-8 /usr/local/bin/clang-8
-      - run:
-          name: "Install Go dependencies"
-          command: dep ensure --vendor-only
       - run:
           name: "Test TinyGo"
           command: make test
@@ -243,34 +214,33 @@ commands:
             tar -C /usr/local/opt -xf /tmp/tinygo.darwin-amd64.tar.gz
             ln -s /usr/local/opt/tinygo/bin/tinygo /usr/local/bin/tinygo
             tinygo version
-      - smoketest-no-avr
+      - run:
+          name: "Download SiFive GNU toolchain"
+          command: |
+            curl -O https://static.dev.sifive.com/dev-tools/riscv64-unknown-elf-gcc-8.2.0-2019.05.3-x86_64-apple-darwin.tar.gz
+            sudo tar -C /usr/local --strip-components=1 -xf riscv64-unknown-elf-gcc-8.2.0-2019.05.3-x86_64-apple-darwin.tar.gz
+      - run: make smoketest AVR=0
 
 
 jobs:
   test-llvm8-go111:
     docker:
-      - image: circleci/golang:1.11
-    working_directory: /go/src/github.com/tinygo-org/tinygo
+      - image: circleci/golang:1.11-stretch
     steps:
-      - test-linux:
-          llvm: "-8"
+      - test-linux
   test-llvm8-go112:
     docker:
-      - image: circleci/golang:1.12
-    working_directory: /go/src/github.com/tinygo-org/tinygo
+      - image: circleci/golang:1.12-stretch
     steps:
-      - test-linux:
-          llvm: "-8"
+      - test-linux
   build-linux:
     docker:
-      - image: circleci/golang:1.12
-    working_directory: /go/src/github.com/tinygo-org/tinygo
+      - image: circleci/golang:1.12-stretch
     steps:
       - build-linux
   build-macos:
     macos:
       xcode: "10.1.0"
-    working_directory: ~/go/src/github.com/tinygo-org/tinygo
     steps:
       - build-macos
 

.gitignore

@@ -5,8 +5,12 @@ src/device/avr/*.ld
 src/device/avr/*.s
 src/device/nrf/*.go
 src/device/nrf/*.s
-src/device/stm32/*.go
-src/device/stm32/*.s
 src/device/sam/*.go
 src/device/sam/*.s
+src/device/sifive/*.go
+src/device/sifive/*.s
+src/device/stm32/*.go
+src/device/stm32/*.s
 vendor
+llvm
+llvm-build

.gitmodules

@@ -9,7 +9,7 @@
 	url = https://github.com/avr-rust/avr-mcu.git
 [submodule "lib/cmsis-svd"]
 	path = lib/cmsis-svd
-	url = https://github.com/posborne/cmsis-svd
+	url = https://github.com/tinygo-org/cmsis-svd
 [submodule "lib/compiler-rt"]
 	path = lib/compiler-rt
 	url = https://github.com/llvm-mirror/compiler-rt.git

BUILDING.md

@@ -27,8 +27,8 @@ on a different system like Mac.
 
 ## Download the source
 
-The first step is to download the TinyGo sources. Then, inside the directory,
-perform these steps:
+The first step is to download the TinyGo sources (use `--recursive` if you clone
+the git repository). Then, inside the directory, perform these steps:
 
     dep ensure -vendor-only # download Go dependencies
     make llvm-source        # download LLVM

CHANGELOG.md

@@ -1,3 +1,99 @@
+0.7.1
+---
+* **targets**
+  - `atsamd21`: add support for the `-port` flag in the flash subcommand
+
+0.7.0
+---
+* **command line**
+  - try more locations to find Clang built-in headers
+  - add support for `tinygo test`
+  - build current directory if no package is specified
+  - support custom .json target spec with `-target` flag
+  - use zversion.go to detect version of GOROOT version
+  - make initial heap size configurable for some targets (currently WebAssembly
+    only)
+* **cgo**
+  - add support for bitfields using generated getters and setters
+  - add support for anonymous structs
+* **compiler**
+  - show an error instead of panicking on duplicate function definitions
+  - allow packages like github.com/tinygo-org/tinygo/src/\* by aliasing it
+  - remove `//go:volatile` support  
+    It has been replaced with the runtime/volatile package.
+  - allow poiners in map keys
+  - support non-constant syscall numbers
+  - implement non-blocking selects
+  - add support for the `-tags` flag
+  - add support for `string` to `[]rune` conversion
+  - implement a portable conservative garbage collector (with support for wasm)
+  - add the `//go:noinline` pragma
+* **standard library**
+  - `os`: add `os.Exit` and `syscall.Exit`
+  - `os`: add several stubs
+  - `runtime`: fix heap corruption in conservative GC
+  - `runtime`: add support for math intrinsics where supported, massively
+    speeding up some benchmarks
+  - `testing`: add basic support for testing
+* **targets**
+  - add support for a generic target that calls `__tinygo_*` functions for
+    peripheral access
+  - `arduino-nano33`: add support for this board
+  - `hifive1`: add support for this RISC-V board
+  - `reelboard`: add e-paper pins
+  - `reelboard`: add `PowerSupplyActive` to enable voltage for on-board devices
+  - `wasm`: put the stack at the start of linear memory, to detect stack
+    overflows
+
+0.6.0
+---
+* **command line**
+  - some portability improvements
+  - make `$GOROOT` more robust and configurable
+  - check for Clang at the Homebrew install location as fallback
+* **compiler driver**
+  - support multiple variations of LLVM commands, for non-Debian distributions
+* **compiler**
+  - improve code quality in multiple ways
+  - make panic configurable, adding trap on panic
+  - refactor many internal parts of the compiler
+  - print all errors encountered during compilation
+  - implement calling function values of a named type
+  - implement returning values from blocking functions
+  - allow larger-than-int values to be sent across a channel
+  - implement complex arithmetic
+  - improve hashmap support
+  - add debuginfo for function arguments
+  - insert nil checks on stores (increasing code size)
+  - implement volatile operations as compiler builtins
+  - add `//go:inline` pragma
+  - add build tags for the Go stdlib version
+* **cgo**
+  - implement `char`, `enum` and `void*` types
+  - support `#include` for builtin headers
+  - improve typedef/struct/enum support
+  - only include symbols that are necessary, for broader support
+  - mark external function args as `nocapture`
+  - implement support for some `#define` constants
+  - implement support for multiple CGo files in a single package
+- **standard library**
+  - `machine`: remove microbit matrix (moved to drivers repository)
+  - `machine`: refactor pins to use `Pin` type instead of `GPIO`
+  - `runtime`: print more interface types on panic, including `error`
+* **targets**
+  - `arm`: print an error on HardFault (including stack overflows)
+  - `atsamd21`: fix a bug in the ADC peripheral
+  - `atsamd21`: add support for I2S
+  - `feather-m0`: add support for this board
+  - `nrf51`: fix a bug in I2C
+  - `stm32f103xx`: fix a bug in I2C
+  - `syscall`: implement `Exit` on unix
+  - `trinket-m0`: add support for this board
+  - `wasm`: make _main_ example smaller
+  - `wasm`: don't cache wasm file in the server, for ease of debugging
+  - `wasm`: work around bug #41508 that caused a deadlock while linking
+  - `wasm`: add support for `js.FuncOf`
+
 0.5.0
 ---
 - **compiler driver**

CONTRIBUTING.md

@@ -16,12 +16,16 @@ Please open a Github issue with your problem, and we will be happy to assist.
 
 We probably have not implemented it yet. Please take a look at our [Roadmap](https://github.com/tinygo-org/tinygo/wiki/Roadmap). Your pull request adding the functionality to TinyGo would be greatly appreciated.
 
+Please open a Github issue. We want to help, and also make sure that there is no duplications of efforts. Sometimes what you need is already being worked on by someone else.
+
 A long tail of small (and large) language features haven't been implemented yet. In almost all cases, the compiler will show a `todo:` error from `compiler/compiler.go` when you try to use it. You can try implementing it, or open a bug report with a small code sample that fails to compile.
 
 ### Some specific hardware you want to use does not appear to be in TinyGo
 
 As above, we probably have not implemented it yet. Your contribution adding the hardware support to TinyGo would be greatly appreciated.
 
+Please start by opening a Github issue. We want to help you to help us to help you.
+
 Lots of targets/boards are still unsupported. Adding an architecture often requires a few compiler changes, but if the architecture is supported you can try implementing support for a new chip or board in `src/runtime`. For details, see [this wiki entry on adding archs/chips/boards](https://github.com/tinygo-org/tinygo/wiki/Adding-a-new-board).
 
 Microcontrollers have lots of peripherals (I2C, SPI, ADC, etc.) and many don't have an implementation yet in the `machine` package. Adding support for new peripherals is very useful.

Dockerfile

@@ -10,8 +10,11 @@ RUN wget -O- https://raw.githubusercontent.com/golang/dep/master/install.sh | sh
 
 COPY . /go/src/github.com/tinygo-org/tinygo
 
+# remove submodules directories and re-init them to fix any hard-coded paths
+# after copying the tinygo directory in the previous step.
 RUN cd /go/src/github.com/tinygo-org/tinygo/ && \
-    git submodule update --init
+    rm -rf ./lib/* && \
+    git submodule update --init --recursive --force
 
 RUN cd /go/src/github.com/tinygo-org/tinygo/ && \
     dep ensure --vendor-only && \
@@ -43,7 +46,7 @@ RUN cd /go/src/github.com/tinygo-org/tinygo/ && \
     apt-get update && \
     apt-get install -y apt-utils python3 make binutils-avr gcc-avr avr-libc && \
     make gen-device-avr && \
-    apt-get remove -y python3 make && \
+    apt-get remove -y python3 && \
     apt-get autoremove -y && \
     apt-get clean
 
@@ -61,7 +64,7 @@ RUN cd /go/src/github.com/tinygo-org/tinygo/ && \
     apt-get update && \
     apt-get install -y apt-utils python3 make clang-8 && \
     make gen-device-nrf && make gen-device-stm32 && \
-    apt-get remove -y python3 make && \
+    apt-get remove -y python3 && \
     apt-get autoremove -y && \
     apt-get clean
 
@@ -76,7 +79,7 @@ RUN cd /go/src/github.com/tinygo-org/tinygo/ && \
     apt-get update && \
     apt-get install -y apt-utils python3 make clang-8 binutils-avr gcc-avr avr-libc && \
     make gen-device && \
-    apt-get remove -y python3 make && \
+    apt-get remove -y python3 && \
     apt-get autoremove -y && \
     apt-get clean
 

LICENSE

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

Makefile

@@ -3,12 +3,12 @@
 all: tinygo
 tinygo: build/tinygo
 
-.PHONY: all tinygo build/tinygo test llvm-build llvm-source clean fmt gen-device gen-device-nrf gen-device-avr
-
 # Default build and source directories, as created by `make llvm-build`.
 LLVM_BUILDDIR ?= llvm-build
-CLANG_SRC ?= llvm/tools/clang
-LLD_SRC ?= llvm/tools/lld
+CLANG_SRC ?= llvm-project/clang
+LLD_SRC ?= llvm-project/lld
+
+.PHONY: all tinygo build/tinygo test $(LLVM_BUILDDIR) llvm-source clean fmt gen-device gen-device-nrf gen-device-avr
 
 LLVM_COMPONENTS = all-targets analysis asmparser asmprinter bitreader bitwriter codegen core coroutines debuginfodwarf executionengine instrumentation interpreter ipo irreader linker lto mc mcjit objcarcopts option profiledata scalaropts support target
 
@@ -18,27 +18,30 @@ ifeq ($(UNAME_S),Linux)
     END_GROUP = -Wl,--end-group
 endif
 
-CLANG_LIBS = $(START_GROUP) $(abspath $(LLVM_BUILDDIR))/lib/libclang.a -lclangAnalysis -lclangARCMigrate -lclangAST -lclangASTMatchers -lclangBasic -lclangCodeGen -lclangCrossTU -lclangDriver -lclangDynamicASTMatchers -lclangEdit -lclangFormat -lclangFrontend -lclangFrontendTool -lclangHandleCXX -lclangHandleLLVM -lclangIndex -lclangLex -lclangParse -lclangRewrite -lclangRewriteFrontend -lclangSema -lclangSerialization -lclangStaticAnalyzerCheckers -lclangStaticAnalyzerCore -lclangStaticAnalyzerFrontend -lclangTooling -lclangToolingASTDiff -lclangToolingCore -lclangToolingInclusions -lclangToolingRefactor $(END_GROUP) -lstdc++
+CLANG_LIBS = $(START_GROUP) $(abspath $(LLVM_BUILDDIR))/lib/libclang.a -lclangAnalysis -lclangARCMigrate -lclangAST -lclangASTMatchers -lclangBasic -lclangCodeGen -lclangCrossTU -lclangDriver -lclangDynamicASTMatchers -lclangEdit -lclangFormat -lclangFrontend -lclangFrontendTool -lclangHandleCXX -lclangHandleLLVM -lclangIndex -lclangLex -lclangParse -lclangRewrite -lclangRewriteFrontend -lclangSema -lclangSerialization -lclangStaticAnalyzerCheckers -lclangStaticAnalyzerCore -lclangStaticAnalyzerFrontend -lclangTooling -lclangToolingASTDiff -lclangToolingCore -lclangToolingInclusions $(END_GROUP) -lstdc++
 
 LLD_LIBS = $(START_GROUP) -llldCOFF -llldCommon -llldCore -llldDriver -llldELF -llldMachO -llldMinGW -llldReaderWriter -llldWasm -llldYAML $(END_GROUP)
 
 
 # For static linking.
-CGO_CPPFLAGS=$(shell $(LLVM_BUILDDIR)/bin/llvm-config --cppflags) -I$(abspath $(CLANG_SRC))/include -I$(abspath $(LLD_SRC))/include
-CGO_CXXFLAGS=-std=c++11
-CGO_LDFLAGS=-L$(LLVM_BUILDDIR)/lib $(CLANG_LIBS) $(LLD_LIBS) $(shell $(LLVM_BUILDDIR)/bin/llvm-config --ldflags --libs --system-libs $(LLVM_COMPONENTS))
+ifneq ("$(wildcard $(LLVM_BUILDDIR)/bin/llvm-config)","")
+    CGO_CPPFLAGS=$(shell $(LLVM_BUILDDIR)/bin/llvm-config --cppflags) -I$(abspath $(CLANG_SRC))/include -I$(abspath $(LLD_SRC))/include
+    CGO_CXXFLAGS=-std=c++11
+    CGO_LDFLAGS=-L$(LLVM_BUILDDIR)/lib $(CLANG_LIBS) $(LLD_LIBS) $(shell $(LLVM_BUILDDIR)/bin/llvm-config --ldflags --libs --system-libs $(LLVM_COMPONENTS))
+endif
 
 
 clean:
 	@rm -rf build
 
-FMT_PATHS = ./*.go compiler interp ir loader src/device/arm src/examples src/machine src/os src/reflect src/runtime src/sync src/syscall
+FMT_PATHS = ./*.go cgo compiler interp ir loader src/device/arm src/examples src/machine src/os src/reflect src/runtime src/sync src/syscall
 fmt:
 	@gofmt -l -w $(FMT_PATHS)
 fmt-check:
 	@unformatted=$$(gofmt -l $(FMT_PATHS)); [ -z "$$unformatted" ] && exit 0; echo "Unformatted:"; for fn in $$unformatted; do echo "  $$fn"; done; exit 1
 
-gen-device: gen-device-avr gen-device-nrf gen-device-sam gen-device-stm32
+
+gen-device: gen-device-avr gen-device-nrf gen-device-sam gen-device-sifive gen-device-stm32
 
 gen-device-avr:
 	./tools/gen-device-avr.py lib/avr/packs/atmega src/device/avr/
@@ -53,46 +56,96 @@ gen-device-sam:
 	./tools/gen-device-svd.py lib/cmsis-svd/data/Atmel/ src/device/sam/ --source=https://github.com/posborne/cmsis-svd/tree/master/data/Atmel
 	go fmt ./src/device/sam
 
+gen-device-sifive:
+	./tools/gen-device-svd.py lib/cmsis-svd/data/SiFive-Community/ src/device/sifive/ --source=https://github.com/AdaCore/svd2ada/tree/master/CMSIS-SVD/SiFive-Community
+	go fmt ./src/device/sifive
+
 gen-device-stm32:
 	./tools/gen-device-svd.py lib/cmsis-svd/data/STMicro/ src/device/stm32/ --source=https://github.com/posborne/cmsis-svd/tree/master/data/STMicro
 	go fmt ./src/device/stm32
 
 
 # Get LLVM sources.
-llvm/README.txt:
-	git clone -b release_80 https://github.com/llvm-mirror/llvm.git llvm
-llvm/tools/clang/README.txt:
-	git clone -b release_80 https://github.com/llvm-mirror/clang.git llvm/tools/clang
-llvm/tools/lld/README.md:
-	git clone -b release_80 https://github.com/llvm-mirror/lld.git llvm/tools/lld
-llvm-source: llvm/README.txt llvm/tools/clang/README.txt llvm/tools/lld/README.md
+llvm-project/README.md:
+	git clone -b release/8.x https://github.com/llvm/llvm-project
+llvm-source: llvm-project/README.md
 
 # Configure LLVM.
-llvm-build/build.ninja: llvm-source
-	mkdir -p llvm-build; cd llvm-build; cmake -G Ninja ../llvm "-DLLVM_TARGETS_TO_BUILD=X86;ARM;AArch64;WebAssembly" "-DLLVM_EXPERIMENTAL_TARGETS_TO_BUILD=AVR" -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_ASSERTIONS=OFF -DLIBCLANG_BUILD_STATIC=ON
+TINYGO_SOURCE_DIR=$(shell pwd)
+$(LLVM_BUILDDIR)/build.ninja: llvm-source
+	mkdir -p $(LLVM_BUILDDIR); cd $(LLVM_BUILDDIR); cmake -G Ninja $(TINYGO_SOURCE_DIR)/llvm-project/llvm "-DLLVM_TARGETS_TO_BUILD=X86;ARM;AArch64;WebAssembly" "-DLLVM_EXPERIMENTAL_TARGETS_TO_BUILD=AVR;RISCV" -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_ASSERTIONS=OFF -DLIBCLANG_BUILD_STATIC=ON -DLLVM_ENABLE_TERMINFO=OFF -DLLVM_ENABLE_ZLIB=OFF -DLLVM_ENABLE_PROJECTS="clang;lld" -DLLVM_TOOL_CLANG_TOOLS_EXTRA_BUILD=OFF
 
 # Build LLVM.
-llvm-build: llvm-build/build.ninja
-	cd llvm-build; ninja
+$(LLVM_BUILDDIR): $(LLVM_BUILDDIR)/build.ninja
+	cd $(LLVM_BUILDDIR); ninja
 
 
 # Build the Go compiler.
 build/tinygo:
-	@if [ ! -f llvm-build/bin/llvm-config ]; then echo "Fetch and build LLVM first by running:\n  make llvm-source\n  make llvm-build"; exit 1; fi
+	@if [ ! -f "$(LLVM_BUILDDIR)/bin/llvm-config" ]; then echo "Fetch and build LLVM first by running:"; echo "  make llvm-source"; echo "  make $(LLVM_BUILDDIR)"; exit 1; fi
 	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" go build -o build/tinygo -tags byollvm .
 
 test:
 	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" go test -v -tags byollvm .
 
+tinygo-test:
+	cd tests/tinygotest && tinygo test
+
+.PHONY: smoketest
+smoketest:
+	# test all examples
+	tinygo build -size short -o test.elf -target=pca10040            examples/blinky1
+	tinygo build -size short -o test.elf -target=pca10040            examples/adc
+	tinygo build -size short -o test.elf -target=pca10040            examples/blinkm
+	tinygo build -size short -o test.elf -target=pca10040            examples/blinky2
+	tinygo build -size short -o test.elf -target=pca10040            examples/button
+	tinygo build -size short -o test.elf -target=pca10040            examples/button2
+	tinygo build -size short -o test.elf -target=pca10040            examples/echo
+	tinygo build -size short -o test.elf -target=circuitplay-express examples/i2s
+	tinygo build -size short -o test.elf -target=pca10040            examples/mcp3008
+	tinygo build -size short -o test.elf -target=microbit            examples/microbit-blink
+	tinygo build -size short -o test.elf -target=pca10040            examples/pwm
+	tinygo build -size short -o test.elf -target=pca10040            examples/serial
+	tinygo build -size short -o test.elf -target=pca10040            examples/test
+	# test all targets/boards
+	tinygo build             -o test.wasm -tags=pca10040             examples/blinky2
+	tinygo build -size short -o test.elf -target=microbit            examples/echo
+	tinygo build -size short -o test.elf -target=nrf52840-mdk        examples/blinky1
+	tinygo build -size short -o test.elf -target=pca10031            examples/blinky1
+	tinygo build -size short -o test.elf -target=bluepill            examples/blinky1
+	tinygo build -size short -o test.elf -target=reelboard           examples/blinky1
+	tinygo build -size short -o test.elf -target=reelboard           examples/blinky2
+	tinygo build -size short -o test.elf -target=pca10056            examples/blinky1
+	tinygo build -size short -o test.elf -target=pca10056            examples/blinky2
+	tinygo build -size short -o test.elf -target=itsybitsy-m0        examples/blinky1
+	tinygo build -size short -o test.elf -target=feather-m0          examples/blinky1
+	tinygo build -size short -o test.elf -target=trinket-m0          examples/blinky1
+	tinygo build -size short -o test.elf -target=circuitplay-express examples/blinky1
+	tinygo build -size short -o test.elf -target=stm32f4disco        examples/blinky1
+	tinygo build -size short -o test.elf -target=stm32f4disco        examples/blinky2
+	tinygo build -size short -o test.elf -target=circuitplay-express examples/i2s
+ifneq ($(AVR), 0)
+	tinygo build -size short -o test.elf -target=arduino             examples/blinky1
+	tinygo build -size short -o test.elf -target=digispark           examples/blinky1
+endif
+ifneq ($(RISCV), 0)
+	tinygo build -size short -o test.elf -target=hifive1b            examples/blinky1
+endif
+	tinygo build             -o wasm.wasm -target=wasm               examples/wasm/export
+	tinygo build             -o wasm.wasm -target=wasm               examples/wasm/main
+
 release: build/tinygo gen-device
 	@mkdir -p build/release/tinygo/bin
+	@mkdir -p build/release/tinygo/lib/clang/include
 	@mkdir -p build/release/tinygo/lib/CMSIS/CMSIS
 	@mkdir -p build/release/tinygo/lib/compiler-rt/lib
 	@mkdir -p build/release/tinygo/lib/nrfx
 	@mkdir -p build/release/tinygo/pkg/armv6m-none-eabi
 	@mkdir -p build/release/tinygo/pkg/armv7m-none-eabi
 	@mkdir -p build/release/tinygo/pkg/armv7em-none-eabi
+	@echo copying source files
 	@cp -p  build/tinygo                 build/release/tinygo/bin
+	@cp -p $(abspath $(CLANG_SRC))/lib/Headers/*.h build/release/tinygo/lib/clang/include
 	@cp -rp lib/CMSIS/CMSIS/Include      build/release/tinygo/lib/CMSIS/CMSIS
 	@cp -rp lib/CMSIS/README.md          build/release/tinygo/lib/CMSIS
 	@cp -rp lib/compiler-rt/lib/builtins build/release/tinygo/lib/compiler-rt/lib

README.md

@@ -17,8 +17,8 @@ import (
 )
 
 func main() {
-    led := machine.GPIO{machine.LED}
-    led.Configure(machine.GPIOConfig{Mode: machine.GPIO_OUTPUT})
+    led := machine.LED
+    led.Configure(machine.PinConfig{Mode: machine.PinOutput})
     for {
         led.Low()
         time.Sleep(time.Millisecond * 1000)
@@ -43,13 +43,17 @@ See the [getting started instructions](https://tinygo.org/getting-started/) for
 
 You can compile TinyGo programs for microcontrollers, WebAssembly and Linux.
 
-The following microcontroller boards are currently supported:
+The following 15 microcontroller boards are currently supported:
 
 * [Adafruit Circuit Playground Express](https://www.adafruit.com/product/3333)
+* [Adafruit Feather M0](https://www.adafruit.com/product/2772)
 * [Adafruit ItsyBitsy M0](https://www.adafruit.com/product/3727)
+* [Adafruit Trinket M0](https://www.adafruit.com/product/3500)
+* [Arduino Nano33 IoT](https://store.arduino.cc/nano-33-iot)
 * [Arduino Uno](https://store.arduino.cc/arduino-uno-rev3)
-* [BBC:Microbit](https://microbit.org/)
+* [BBC micro:bit](https://microbit.org/)
 * [ST Micro STM32F103XX "Bluepill"](http://wiki.stm32duino.com/index.php?title=Blue_Pill)
+* [ST Micro STM32F407 "Discovery"](https://www.st.com/en/evaluation-tools/stm32f4discovery.html)
 * [Digispark](http://digistump.com/products/1)
 * [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)

buildcache.go

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

builtins.go

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

cgo/cgo.go

@@ -0,0 +1,997 @@
+// Package cgo implements CGo by modifying a loaded AST. It does this by parsing
+// the `import "C"` statements found in the source code with libclang and
+// generating stub function and global declarations.
+//
+// There are a few advantages to modifying the AST directly instead of doing CGo
+// as a preprocessing step, with the main advantage being that debug information
+// is kept intact as much as possible.
+package cgo
+
+// This file extracts the `import "C"` statement from the source and modifies
+// the AST for CGo. It does not use libclang directly: see libclang.go for the C
+// source file parsing.
+
+import (
+	"go/ast"
+	"go/token"
+	"sort"
+	"strconv"
+	"strings"
+
+	"golang.org/x/tools/go/ast/astutil"
+)
+
+// cgoPackage holds all CGo-related information of a package.
+type cgoPackage struct {
+	generated       *ast.File
+	generatedPos    token.Pos
+	errors          []error
+	dir             string
+	fset            *token.FileSet
+	tokenFiles      map[string]*token.File
+	missingSymbols  map[string]struct{}
+	constants       map[string]constantInfo
+	functions       map[string]*functionInfo
+	globals         map[string]globalInfo
+	typedefs        map[string]*typedefInfo
+	elaboratedTypes map[string]*elaboratedTypeInfo
+	enums           map[string]enumInfo
+	anonStructNum   int
+}
+
+// constantInfo stores some information about a CGo constant found by libclang
+// and declared in the Go AST.
+type constantInfo struct {
+	expr *ast.BasicLit
+	pos  token.Pos
+}
+
+// functionInfo stores some information about a CGo function found by libclang
+// and declared in the AST.
+type functionInfo struct {
+	args    []paramInfo
+	results *ast.FieldList
+	pos     token.Pos
+}
+
+// paramInfo is a parameter of a CGo function (see functionInfo).
+type paramInfo struct {
+	name     string
+	typeExpr ast.Expr
+}
+
+// typedefInfo contains information about a single typedef in C.
+type typedefInfo struct {
+	typeExpr ast.Expr
+	pos      token.Pos
+}
+
+// elaboratedTypeInfo contains some information about an elaborated type
+// (struct, union) found in the C AST.
+type elaboratedTypeInfo struct {
+	typeExpr  *ast.StructType
+	pos       token.Pos
+	bitfields []bitfieldInfo
+}
+
+// bitfieldInfo contains information about a single bitfield in a struct. It
+// keeps information about the start, end, and the special (renamed) base field
+// of this bitfield.
+type bitfieldInfo struct {
+	field    *ast.Field
+	name     string
+	pos      token.Pos
+	startBit int64
+	endBit   int64 // may be 0 meaning "until the end of the field"
+}
+
+// enumInfo contains information about an enum in the C.
+type enumInfo struct {
+	typeExpr ast.Expr
+	pos      token.Pos
+}
+
+// globalInfo contains information about a declared global variable in C.
+type globalInfo struct {
+	typeExpr ast.Expr
+	pos      token.Pos
+}
+
+// cgoAliases list type aliases between Go and C, for types that are equivalent
+// in both languages. See addTypeAliases.
+var cgoAliases = map[string]string{
+	"C.int8_t":    "int8",
+	"C.int16_t":   "int16",
+	"C.int32_t":   "int32",
+	"C.int64_t":   "int64",
+	"C.uint8_t":   "uint8",
+	"C.uint16_t":  "uint16",
+	"C.uint32_t":  "uint32",
+	"C.uint64_t":  "uint64",
+	"C.uintptr_t": "uintptr",
+}
+
+// builtinAliases are handled specially because they only exist on the Go side
+// of CGo, not on the CGo side (they're prefixed with "_Cgo_" there).
+var builtinAliases = map[string]struct{}{
+	"char":      struct{}{},
+	"schar":     struct{}{},
+	"uchar":     struct{}{},
+	"short":     struct{}{},
+	"ushort":    struct{}{},
+	"int":       struct{}{},
+	"uint":      struct{}{},
+	"long":      struct{}{},
+	"ulong":     struct{}{},
+	"longlong":  struct{}{},
+	"ulonglong": struct{}{},
+}
+
+// cgoTypes lists some C types with ambiguous sizes that must be retrieved
+// somehow from C. This is done by adding some typedefs to get the size of each
+// type.
+const cgoTypes = `
+typedef char                _Cgo_char;
+typedef signed char         _Cgo_schar;
+typedef unsigned char       _Cgo_uchar;
+typedef short               _Cgo_short;
+typedef unsigned short      _Cgo_ushort;
+typedef int                 _Cgo_int;
+typedef unsigned int        _Cgo_uint;
+typedef long                _Cgo_long;
+typedef unsigned long       _Cgo_ulong;
+typedef long long           _Cgo_longlong;
+typedef unsigned long long  _Cgo_ulonglong;
+`
+
+// Process extracts `import "C"` statements from the AST, parses the comment
+// with libclang, and modifies the AST to use this information. It returns a
+// newly created *ast.File that should be added to the list of to-be-parsed
+// files. If there is one or more error, it returns these in the []error slice
+// but still modifies the AST.
+func Process(files []*ast.File, dir string, fset *token.FileSet, cflags []string) (*ast.File, []error) {
+	p := &cgoPackage{
+		dir:             dir,
+		fset:            fset,
+		tokenFiles:      map[string]*token.File{},
+		missingSymbols:  map[string]struct{}{},
+		constants:       map[string]constantInfo{},
+		functions:       map[string]*functionInfo{},
+		globals:         map[string]globalInfo{},
+		typedefs:        map[string]*typedefInfo{},
+		elaboratedTypes: map[string]*elaboratedTypeInfo{},
+		enums:           map[string]enumInfo{},
+	}
+
+	// Add a new location for the following file.
+	generatedTokenPos := p.fset.AddFile(dir+"/!cgo.go", -1, 0)
+	generatedTokenPos.SetLines([]int{0})
+	p.generatedPos = generatedTokenPos.Pos(0)
+
+	// Construct a new in-memory AST for CGo declarations of this package.
+	unsafeImport := &ast.ImportSpec{
+		Path: &ast.BasicLit{
+			ValuePos: p.generatedPos,
+			Kind:     token.STRING,
+			Value:    "\"unsafe\"",
+		},
+		EndPos: p.generatedPos,
+	}
+	p.generated = &ast.File{
+		Package: p.generatedPos,
+		Name: &ast.Ident{
+			NamePos: p.generatedPos,
+			Name:    files[0].Name.Name,
+		},
+		Decls: []ast.Decl{
+			&ast.GenDecl{
+				TokPos: p.generatedPos,
+				Tok:    token.IMPORT,
+				Specs: []ast.Spec{
+					unsafeImport,
+				},
+			},
+		},
+		Imports: []*ast.ImportSpec{unsafeImport},
+	}
+
+	// Find all C.* symbols.
+	for _, f := range files {
+		astutil.Apply(f, p.findMissingCGoNames, nil)
+	}
+	for name := range builtinAliases {
+		p.missingSymbols["_Cgo_"+name] = struct{}{}
+	}
+
+	// Find `import "C"` statements in the file.
+	for _, f := range files {
+		for i := 0; i < len(f.Decls); i++ {
+			decl := f.Decls[i]
+			genDecl, ok := decl.(*ast.GenDecl)
+			if !ok {
+				continue
+			}
+			if len(genDecl.Specs) != 1 {
+				continue
+			}
+			spec, ok := genDecl.Specs[0].(*ast.ImportSpec)
+			if !ok {
+				continue
+			}
+			path, err := strconv.Unquote(spec.Path.Value)
+			if err != nil {
+				panic("could not parse import path: " + err.Error())
+			}
+			if path != "C" {
+				continue
+			}
+			cgoComment := genDecl.Doc.Text()
+
+			pos := genDecl.Pos()
+			if genDecl.Doc != nil {
+				pos = genDecl.Doc.Pos()
+			}
+			position := fset.PositionFor(pos, true)
+			p.parseFragment(cgoComment+cgoTypes, cflags, position.Filename, position.Line)
+
+			// Remove this import declaration.
+			f.Decls = append(f.Decls[:i], f.Decls[i+1:]...)
+			i--
+		}
+
+		// Print the AST, for debugging.
+		//ast.Print(fset, f)
+	}
+
+	// Declare functions found by libclang.
+	p.addFuncDecls()
+
+	// Declare stub function pointer values found by libclang.
+	p.addFuncPtrDecls()
+
+	// Declare globals found by libclang.
+	p.addConstDecls()
+
+	// Declare globals found by libclang.
+	p.addVarDecls()
+
+	// Forward C types to Go types (like C.uint32_t -> uint32).
+	p.addTypeAliases()
+
+	// Add type declarations for C types, declared using typedef in C.
+	p.addTypedefs()
+
+	// Add elaborated types for C structs and unions.
+	p.addElaboratedTypes()
+
+	// Add enum types and enum constants for C enums.
+	p.addEnumTypes()
+
+	// Patch the AST to use the declared types and functions.
+	for _, f := range files {
+		astutil.Apply(f, p.walker, nil)
+	}
+
+	// Print the newly generated in-memory AST, for debugging.
+	//ast.Print(fset, p.generated)
+
+	return p.generated, p.errors
+}
+
+// addFuncDecls adds the C function declarations found by libclang in the
+// comment above the `import "C"` statement.
+func (p *cgoPackage) addFuncDecls() {
+	names := make([]string, 0, len(p.functions))
+	for name := range p.functions {
+		names = append(names, name)
+	}
+	sort.Strings(names)
+	for _, name := range names {
+		fn := p.functions[name]
+		obj := &ast.Object{
+			Kind: ast.Fun,
+			Name: "C." + name,
+		}
+		args := make([]*ast.Field, len(fn.args))
+		decl := &ast.FuncDecl{
+			Name: &ast.Ident{
+				NamePos: fn.pos,
+				Name:    "C." + name,
+				Obj:     obj,
+			},
+			Type: &ast.FuncType{
+				Func: fn.pos,
+				Params: &ast.FieldList{
+					Opening: fn.pos,
+					List:    args,
+					Closing: fn.pos,
+				},
+				Results: fn.results,
+			},
+		}
+		obj.Decl = decl
+		for i, arg := range fn.args {
+			args[i] = &ast.Field{
+				Names: []*ast.Ident{
+					&ast.Ident{
+						NamePos: fn.pos,
+						Name:    arg.name,
+						Obj: &ast.Object{
+							Kind: ast.Var,
+							Name: arg.name,
+							Decl: decl,
+						},
+					},
+				},
+				Type: arg.typeExpr,
+			}
+		}
+		p.generated.Decls = append(p.generated.Decls, decl)
+	}
+}
+
+// addFuncPtrDecls creates stub declarations of function pointer values. These
+// values will later be replaced with the real values in the compiler.
+// It adds code like the following to the AST:
+//
+//     var (
+//         C.add unsafe.Pointer
+//         C.mul unsafe.Pointer
+//         // ...
+//     )
+func (p *cgoPackage) addFuncPtrDecls() {
+	if len(p.functions) == 0 {
+		return
+	}
+	gen := &ast.GenDecl{
+		TokPos: token.NoPos,
+		Tok:    token.VAR,
+		Lparen: token.NoPos,
+		Rparen: token.NoPos,
+	}
+	names := make([]string, 0, len(p.functions))
+	for name := range p.functions {
+		names = append(names, name)
+	}
+	sort.Strings(names)
+	for _, name := range names {
+		fn := p.functions[name]
+		obj := &ast.Object{
+			Kind: ast.Typ,
+			Name: "C." + name + "$funcaddr",
+		}
+		valueSpec := &ast.ValueSpec{
+			Names: []*ast.Ident{&ast.Ident{
+				NamePos: fn.pos,
+				Name:    "C." + name + "$funcaddr",
+				Obj:     obj,
+			}},
+			Type: &ast.SelectorExpr{
+				X: &ast.Ident{
+					NamePos: fn.pos,
+					Name:    "unsafe",
+				},
+				Sel: &ast.Ident{
+					NamePos: fn.pos,
+					Name:    "Pointer",
+				},
+			},
+		}
+		obj.Decl = valueSpec
+		gen.Specs = append(gen.Specs, valueSpec)
+	}
+	p.generated.Decls = append(p.generated.Decls, gen)
+}
+
+// addConstDecls declares external C constants in the Go source.
+// It adds code like the following to the AST:
+//
+//     const (
+//         C.CONST_INT = 5
+//         C.CONST_FLOAT = 5.8
+//         // ...
+//     )
+func (p *cgoPackage) addConstDecls() {
+	if len(p.constants) == 0 {
+		return
+	}
+	gen := &ast.GenDecl{
+		TokPos: token.NoPos,
+		Tok:    token.CONST,
+		Lparen: token.NoPos,
+		Rparen: token.NoPos,
+	}
+	names := make([]string, 0, len(p.constants))
+	for name := range p.constants {
+		names = append(names, name)
+	}
+	sort.Strings(names)
+	for _, name := range names {
+		constVal := p.constants[name]
+		obj := &ast.Object{
+			Kind: ast.Con,
+			Name: "C." + name,
+		}
+		valueSpec := &ast.ValueSpec{
+			Names: []*ast.Ident{&ast.Ident{
+				NamePos: constVal.pos,
+				Name:    "C." + name,
+				Obj:     obj,
+			}},
+			Values: []ast.Expr{constVal.expr},
+		}
+		obj.Decl = valueSpec
+		gen.Specs = append(gen.Specs, valueSpec)
+	}
+	p.generated.Decls = append(p.generated.Decls, gen)
+}
+
+// addVarDecls declares external C globals in the Go source.
+// It adds code like the following to the AST:
+//
+//     var (
+//         C.globalInt  int
+//         C.globalBool bool
+//         // ...
+//     )
+func (p *cgoPackage) addVarDecls() {
+	if len(p.globals) == 0 {
+		return
+	}
+	gen := &ast.GenDecl{
+		TokPos: token.NoPos,
+		Tok:    token.VAR,
+		Lparen: token.NoPos,
+		Rparen: token.NoPos,
+	}
+	names := make([]string, 0, len(p.globals))
+	for name := range p.globals {
+		names = append(names, name)
+	}
+	sort.Strings(names)
+	for _, name := range names {
+		global := p.globals[name]
+		obj := &ast.Object{
+			Kind: ast.Var,
+			Name: "C." + name,
+		}
+		valueSpec := &ast.ValueSpec{
+			Names: []*ast.Ident{&ast.Ident{
+				NamePos: global.pos,
+				Name:    "C." + name,
+				Obj:     obj,
+			}},
+			Type: global.typeExpr,
+		}
+		obj.Decl = valueSpec
+		gen.Specs = append(gen.Specs, valueSpec)
+	}
+	p.generated.Decls = append(p.generated.Decls, gen)
+}
+
+// addTypeAliases aliases some built-in Go types with their equivalent C types.
+// It adds code like the following to the AST:
+//
+//     type (
+//         C.int8_t  = int8
+//         C.int16_t = int16
+//         // ...
+//     )
+func (p *cgoPackage) addTypeAliases() {
+	aliasKeys := make([]string, 0, len(cgoAliases))
+	for key := range cgoAliases {
+		aliasKeys = append(aliasKeys, key)
+	}
+	sort.Strings(aliasKeys)
+	gen := &ast.GenDecl{
+		TokPos: token.NoPos,
+		Tok:    token.TYPE,
+		Lparen: token.NoPos,
+		Rparen: token.NoPos,
+	}
+	for _, typeName := range aliasKeys {
+		goTypeName := cgoAliases[typeName]
+		obj := &ast.Object{
+			Kind: ast.Typ,
+			Name: typeName,
+		}
+		typeSpec := &ast.TypeSpec{
+			Name: &ast.Ident{
+				NamePos: token.NoPos,
+				Name:    typeName,
+				Obj:     obj,
+			},
+			Assign: p.generatedPos,
+			Type: &ast.Ident{
+				NamePos: token.NoPos,
+				Name:    goTypeName,
+			},
+		}
+		obj.Decl = typeSpec
+		gen.Specs = append(gen.Specs, typeSpec)
+	}
+	p.generated.Decls = append(p.generated.Decls, gen)
+}
+
+func (p *cgoPackage) addTypedefs() {
+	if len(p.typedefs) == 0 {
+		return
+	}
+	gen := &ast.GenDecl{
+		TokPos: token.NoPos,
+		Tok:    token.TYPE,
+	}
+	names := make([]string, 0, len(p.typedefs))
+	for name := range p.typedefs {
+		names = append(names, name)
+	}
+	sort.Strings(names)
+	for _, name := range names {
+		typedef := p.typedefs[name]
+		typeName := "C." + name
+		isAlias := true
+		if strings.HasPrefix(name, "_Cgo_") {
+			typeName = "C." + name[len("_Cgo_"):]
+			isAlias = false // C.short etc. should not be aliased to the equivalent Go type (not portable)
+		}
+		if _, ok := cgoAliases[typeName]; ok {
+			// This is a type that also exists in Go (defined in stdint.h).
+			continue
+		}
+		obj := &ast.Object{
+			Kind: ast.Typ,
+			Name: typeName,
+		}
+		typeSpec := &ast.TypeSpec{
+			Name: &ast.Ident{
+				NamePos: typedef.pos,
+				Name:    typeName,
+				Obj:     obj,
+			},
+			Type: typedef.typeExpr,
+		}
+		if isAlias {
+			typeSpec.Assign = typedef.pos
+		}
+		obj.Decl = typeSpec
+		gen.Specs = append(gen.Specs, typeSpec)
+	}
+	p.generated.Decls = append(p.generated.Decls, gen)
+}
+
+// addElaboratedTypes adds C elaborated types as aliases. These are the "struct
+// foo" or "union foo" types, often used in a typedef.
+//
+// See also:
+// https://en.cppreference.com/w/cpp/language/elaborated_type_specifier
+func (p *cgoPackage) addElaboratedTypes() {
+	if len(p.elaboratedTypes) == 0 {
+		return
+	}
+	gen := &ast.GenDecl{
+		TokPos: token.NoPos,
+		Tok:    token.TYPE,
+	}
+	names := make([]string, 0, len(p.elaboratedTypes))
+	for name := range p.elaboratedTypes {
+		names = append(names, name)
+	}
+	sort.Strings(names)
+	for _, name := range names {
+		typ := p.elaboratedTypes[name]
+		typeName := "C." + name
+		obj := &ast.Object{
+			Kind: ast.Typ,
+			Name: typeName,
+		}
+		typeSpec := &ast.TypeSpec{
+			Name: &ast.Ident{
+				NamePos: typ.pos,
+				Name:    typeName,
+				Obj:     obj,
+			},
+			Type: typ.typeExpr,
+		}
+		obj.Decl = typeSpec
+		gen.Specs = append(gen.Specs, typeSpec)
+		// If this struct has bitfields, create getters for them.
+		for _, bitfield := range typ.bitfields {
+			p.createBitfieldGetter(bitfield, typeName)
+			p.createBitfieldSetter(bitfield, typeName)
+		}
+	}
+	p.generated.Decls = append(p.generated.Decls, gen)
+}
+
+// createBitfieldGetter creates a bitfield getter function like the following:
+//
+//     func (s *C.struct_foo) bitfield_b() byte {
+//         return (s.__bitfield_1 >> 5) & 0x1
+//     }
+func (p *cgoPackage) createBitfieldGetter(bitfield bitfieldInfo, typeName string) {
+	// The value to return from the getter.
+	// Not complete: this is just an expression to get the complete field.
+	var result ast.Expr = &ast.SelectorExpr{
+		X: &ast.Ident{
+			NamePos: bitfield.pos,
+			Name:    "s",
+			Obj:     nil,
+		},
+		Sel: &ast.Ident{
+			NamePos: bitfield.pos,
+			Name:    bitfield.field.Names[0].Name,
+		},
+	}
+	if bitfield.startBit != 0 {
+		// Shift to the right by .startBit so that fields that come before are
+		// shifted off.
+		result = &ast.BinaryExpr{
+			X:     result,
+			OpPos: bitfield.pos,
+			Op:    token.SHR,
+			Y: &ast.BasicLit{
+				ValuePos: bitfield.pos,
+				Kind:     token.INT,
+				Value:    strconv.FormatInt(bitfield.startBit, 10),
+			},
+		}
+	}
+	if bitfield.endBit != 0 {
+		// Mask off the high bits so that fields that come after this field are
+		// masked off.
+		and := (uint64(1) << uint64(bitfield.endBit-bitfield.startBit)) - 1
+		result = &ast.BinaryExpr{
+			X:     result,
+			OpPos: bitfield.pos,
+			Op:    token.AND,
+			Y: &ast.BasicLit{
+				ValuePos: bitfield.pos,
+				Kind:     token.INT,
+				Value:    "0x" + strconv.FormatUint(and, 16),
+			},
+		}
+	}
+
+	// Create the getter function.
+	getter := &ast.FuncDecl{
+		Recv: &ast.FieldList{
+			Opening: bitfield.pos,
+			List: []*ast.Field{
+				&ast.Field{
+					Names: []*ast.Ident{
+						&ast.Ident{
+							NamePos: bitfield.pos,
+							Name:    "s",
+							Obj: &ast.Object{
+								Kind: ast.Var,
+								Name: "s",
+								Decl: nil,
+							},
+						},
+					},
+					Type: &ast.StarExpr{
+						Star: bitfield.pos,
+						X: &ast.Ident{
+							NamePos: bitfield.pos,
+							Name:    typeName,
+							Obj:     nil,
+						},
+					},
+				},
+			},
+			Closing: bitfield.pos,
+		},
+		Name: &ast.Ident{
+			NamePos: bitfield.pos,
+			Name:    "bitfield_" + bitfield.name,
+		},
+		Type: &ast.FuncType{
+			Func: bitfield.pos,
+			Params: &ast.FieldList{
+				Opening: bitfield.pos,
+				Closing: bitfield.pos,
+			},
+			Results: &ast.FieldList{
+				List: []*ast.Field{
+					&ast.Field{
+						Type: bitfield.field.Type,
+					},
+				},
+			},
+		},
+		Body: &ast.BlockStmt{
+			Lbrace: bitfield.pos,
+			List: []ast.Stmt{
+				&ast.ReturnStmt{
+					Return: bitfield.pos,
+					Results: []ast.Expr{
+						result,
+					},
+				},
+			},
+			Rbrace: bitfield.pos,
+		},
+	}
+	p.generated.Decls = append(p.generated.Decls, getter)
+}
+
+// createBitfieldSetter creates a bitfield setter function like the following:
+//
+//     func (s *C.struct_foo) set_bitfield_b(value byte) {
+//         s.__bitfield_1 = s.__bitfield_1 ^ 0x60 | ((value & 1) << 5)
+//     }
+//
+// Or the following:
+//
+//     func (s *C.struct_foo) set_bitfield_c(value byte) {
+//         s.__bitfield_1 = s.__bitfield_1 & 0x3f | (value << 6)
+//     }
+func (p *cgoPackage) createBitfieldSetter(bitfield bitfieldInfo, typeName string) {
+	// The full field with all bitfields.
+	var field ast.Expr = &ast.SelectorExpr{
+		X: &ast.Ident{
+			NamePos: bitfield.pos,
+			Name:    "s",
+			Obj:     nil,
+		},
+		Sel: &ast.Ident{
+			NamePos: bitfield.pos,
+			Name:    bitfield.field.Names[0].Name,
+		},
+	}
+	// The value to insert into the field.
+	var valueToInsert ast.Expr = &ast.Ident{
+		NamePos: bitfield.pos,
+		Name:    "value",
+	}
+
+	if bitfield.endBit != 0 {
+		// Make sure the value is in range with a mask.
+		valueToInsert = &ast.BinaryExpr{
+			X:     valueToInsert,
+			OpPos: bitfield.pos,
+			Op:    token.AND,
+			Y: &ast.BasicLit{
+				ValuePos: bitfield.pos,
+				Kind:     token.INT,
+				Value:    "0x" + strconv.FormatUint((uint64(1)<<uint64(bitfield.endBit-bitfield.startBit))-1, 16),
+			},
+		}
+		// Create a mask for the AND NOT operation.
+		mask := ((uint64(1) << uint64(bitfield.endBit-bitfield.startBit)) - 1) << uint64(bitfield.startBit)
+		// Zero the bits in the field that will soon be inserted.
+		field = &ast.BinaryExpr{
+			X:     field,
+			OpPos: bitfield.pos,
+			Op:    token.AND_NOT,
+			Y: &ast.BasicLit{
+				ValuePos: bitfield.pos,
+				Kind:     token.INT,
+				Value:    "0x" + strconv.FormatUint(mask, 16),
+			},
+		}
+	} else { // bitfield.endBit == 0
+		// We don't know exactly how many high bits should be zeroed. So we do
+		// something different: keep the low bits with a mask and OR the new
+		// value with it.
+		mask := (uint64(1) << uint64(bitfield.startBit)) - 1
+		// Extract the lower bits.
+		field = &ast.BinaryExpr{
+			X:     field,
+			OpPos: bitfield.pos,
+			Op:    token.AND,
+			Y: &ast.BasicLit{
+				ValuePos: bitfield.pos,
+				Kind:     token.INT,
+				Value:    "0x" + strconv.FormatUint(mask, 16),
+			},
+		}
+	}
+
+	// Bitwise OR with the new value (after the new value has been shifted).
+	field = &ast.BinaryExpr{
+		X:     field,
+		OpPos: bitfield.pos,
+		Op:    token.OR,
+		Y: &ast.BinaryExpr{
+			X:     valueToInsert,
+			OpPos: bitfield.pos,
+			Op:    token.SHL,
+			Y: &ast.BasicLit{
+				ValuePos: bitfield.pos,
+				Kind:     token.INT,
+				Value:    strconv.FormatInt(bitfield.startBit, 10),
+			},
+		},
+	}
+
+	// Create the setter function.
+	setter := &ast.FuncDecl{
+		Recv: &ast.FieldList{
+			Opening: bitfield.pos,
+			List: []*ast.Field{
+				&ast.Field{
+					Names: []*ast.Ident{
+						&ast.Ident{
+							NamePos: bitfield.pos,
+							Name:    "s",
+							Obj: &ast.Object{
+								Kind: ast.Var,
+								Name: "s",
+								Decl: nil,
+							},
+						},
+					},
+					Type: &ast.StarExpr{
+						Star: bitfield.pos,
+						X: &ast.Ident{
+							NamePos: bitfield.pos,
+							Name:    typeName,
+							Obj:     nil,
+						},
+					},
+				},
+			},
+			Closing: bitfield.pos,
+		},
+		Name: &ast.Ident{
+			NamePos: bitfield.pos,
+			Name:    "set_bitfield_" + bitfield.name,
+		},
+		Type: &ast.FuncType{
+			Func: bitfield.pos,
+			Params: &ast.FieldList{
+				Opening: bitfield.pos,
+				List: []*ast.Field{
+					&ast.Field{
+						Names: []*ast.Ident{
+							&ast.Ident{
+								NamePos: bitfield.pos,
+								Name:    "value",
+								Obj:     nil,
+							},
+						},
+						Type: bitfield.field.Type,
+					},
+				},
+				Closing: bitfield.pos,
+			},
+		},
+		Body: &ast.BlockStmt{
+			Lbrace: bitfield.pos,
+			List: []ast.Stmt{
+				&ast.AssignStmt{
+					Lhs: []ast.Expr{
+						&ast.SelectorExpr{
+							X: &ast.Ident{
+								NamePos: bitfield.pos,
+								Name:    "s",
+								Obj:     nil,
+							},
+							Sel: &ast.Ident{
+								NamePos: bitfield.pos,
+								Name:    bitfield.field.Names[0].Name,
+							},
+						},
+					},
+					TokPos: bitfield.pos,
+					Tok:    token.ASSIGN,
+					Rhs: []ast.Expr{
+						field,
+					},
+				},
+			},
+			Rbrace: bitfield.pos,
+		},
+	}
+	p.generated.Decls = append(p.generated.Decls, setter)
+}
+
+// addEnumTypes adds C enums to the AST. For example, the following C code:
+//
+//     enum option {
+//         optionA,
+//         optionB = 5,
+//     };
+//
+// is translated to the following Go code equivalent:
+//
+//     type C.enum_option int32
+//
+// The constants are treated just like macros so are inserted into the AST by
+// addConstDecls.
+// See also: https://en.cppreference.com/w/c/language/enum
+func (p *cgoPackage) addEnumTypes() {
+	if len(p.enums) == 0 {
+		return
+	}
+	gen := &ast.GenDecl{
+		TokPos: token.NoPos,
+		Tok:    token.TYPE,
+	}
+	names := make([]string, 0, len(p.enums))
+	for name := range p.enums {
+		names = append(names, name)
+	}
+	sort.Strings(names)
+	for _, name := range names {
+		typ := p.enums[name]
+		typeName := "C.enum_" + name
+		obj := &ast.Object{
+			Kind: ast.Typ,
+			Name: typeName,
+		}
+		typeSpec := &ast.TypeSpec{
+			Name: &ast.Ident{
+				NamePos: typ.pos,
+				Name:    typeName,
+				Obj:     obj,
+			},
+			Type: typ.typeExpr,
+		}
+		obj.Decl = typeSpec
+		gen.Specs = append(gen.Specs, typeSpec)
+	}
+	p.generated.Decls = append(p.generated.Decls, gen)
+}
+
+// findMissingCGoNames traverses the AST and finds all C.something names. Only
+// these symbols are extracted from the parsed C AST and converted to the Go
+// equivalent.
+func (p *cgoPackage) findMissingCGoNames(cursor *astutil.Cursor) bool {
+	switch node := cursor.Node().(type) {
+	case *ast.SelectorExpr:
+		x, ok := node.X.(*ast.Ident)
+		if !ok {
+			return true
+		}
+		if x.Name == "C" {
+			name := node.Sel.Name
+			if _, ok := builtinAliases[name]; ok {
+				name = "_Cgo_" + name
+			}
+			p.missingSymbols[name] = struct{}{}
+		}
+	}
+	return true
+}
+
+// walker replaces all "C".<something> expressions to literal "C.<something>"
+// expressions. Such expressions are impossible to write in Go (a dot cannot be
+// used in the middle of a name) so in practice all C identifiers live in a
+// separate namespace (no _Cgo_ hacks like in gc).
+func (p *cgoPackage) walker(cursor *astutil.Cursor) bool {
+	switch node := cursor.Node().(type) {
+	case *ast.CallExpr:
+		fun, ok := node.Fun.(*ast.SelectorExpr)
+		if !ok {
+			return true
+		}
+		x, ok := fun.X.(*ast.Ident)
+		if !ok {
+			return true
+		}
+		if _, ok := p.functions[fun.Sel.Name]; ok && x.Name == "C" {
+			node.Fun = &ast.Ident{
+				NamePos: x.NamePos,
+				Name:    "C." + fun.Sel.Name,
+			}
+		}
+	case *ast.SelectorExpr:
+		x, ok := node.X.(*ast.Ident)
+		if !ok {
+			return true
+		}
+		if x.Name == "C" {
+			name := "C." + node.Sel.Name
+			if _, ok := p.functions[node.Sel.Name]; ok {
+				name += "$funcaddr"
+			}
+			cursor.Replace(&ast.Ident{
+				NamePos: x.NamePos,
+				Name:    name,
+			})
+		}
+	}
+	return true
+}

cgo/libclang.go

@@ -0,0 +1,763 @@
+package cgo
+
+// This file parses a fragment of C with libclang and stores the result for AST
+// modification. It does not touch the AST itself.
+
+import (
+	"fmt"
+	"go/ast"
+	"go/scanner"
+	"go/token"
+	"path/filepath"
+	"strconv"
+	"strings"
+	"unsafe"
+)
+
+/*
+#include <clang-c/Index.h> // if this fails, install libclang-8-dev
+#include <stdlib.h>
+#include <stdint.h>
+
+// This struct should be ABI-compatible on all platforms (uintptr_t has the same
+// alignment etc. as void*) but does not include void* pointers that are not
+// always real pointers.
+// The Go garbage collector assumes that all non-nil pointer-typed integers are
+// actually pointers. This is not always true, as data[1] often contains 0x1,
+// which is clearly not a valid pointer. Usually the GC won't catch this issue,
+// but occasionally it will leading to a crash with a vague error message.
+typedef struct {
+	enum CXCursorKind kind;
+	int xdata;
+	uintptr_t data[3];
+} GoCXCursor;
+
+// Forwarding functions. They are implemented in libclang_stubs.c and forward to
+// the real functions without doing anything else, thus they are entirely
+// compatible with the versions without tinygo_ prefix. The only difference is
+// the CXCursor type, which has been replaced with GoCXCursor.
+GoCXCursor tinygo_clang_getTranslationUnitCursor(CXTranslationUnit tu);
+unsigned tinygo_clang_visitChildren(GoCXCursor parent, CXCursorVisitor visitor, CXClientData client_data);
+CXString tinygo_clang_getCursorSpelling(GoCXCursor c);
+enum CXCursorKind tinygo_clang_getCursorKind(GoCXCursor c);
+CXType tinygo_clang_getCursorType(GoCXCursor c);
+GoCXCursor tinygo_clang_getTypeDeclaration(CXType t);
+CXType tinygo_clang_getTypedefDeclUnderlyingType(GoCXCursor c);
+CXType tinygo_clang_getCursorResultType(GoCXCursor c);
+int tinygo_clang_Cursor_getNumArguments(GoCXCursor c);
+GoCXCursor tinygo_clang_Cursor_getArgument(GoCXCursor c, unsigned i);
+CXSourceLocation tinygo_clang_getCursorLocation(GoCXCursor c);
+CXSourceRange tinygo_clang_getCursorExtent(GoCXCursor c);
+CXTranslationUnit tinygo_clang_Cursor_getTranslationUnit(GoCXCursor c);
+long long tinygo_clang_getEnumConstantDeclValue(GoCXCursor c);
+CXType tinygo_clang_getEnumDeclIntegerType(GoCXCursor c);
+unsigned tinygo_clang_Cursor_isBitField(GoCXCursor c);
+
+int tinygo_clang_globals_visitor(GoCXCursor c, GoCXCursor parent, CXClientData client_data);
+int tinygo_clang_struct_visitor(GoCXCursor c, GoCXCursor parent, CXClientData client_data);
+int tinygo_clang_enum_visitor(GoCXCursor c, GoCXCursor parent, CXClientData client_data);
+*/
+import "C"
+
+// storedRefs stores references to types, used for clang_visitChildren.
+var storedRefs refMap
+
+var diagnosticSeverity = [...]string{
+	C.CXDiagnostic_Ignored: "ignored",
+	C.CXDiagnostic_Note:    "note",
+	C.CXDiagnostic_Warning: "warning",
+	C.CXDiagnostic_Error:   "error",
+	C.CXDiagnostic_Fatal:   "fatal",
+}
+
+func (p *cgoPackage) parseFragment(fragment string, cflags []string, posFilename string, posLine int) {
+	index := C.clang_createIndex(0, 0)
+	defer C.clang_disposeIndex(index)
+
+	// pretend to be a .c file
+	filenameC := C.CString(posFilename + "!cgo.c")
+	defer C.free(unsafe.Pointer(filenameC))
+
+	// fix up error locations
+	fragment = fmt.Sprintf("# %d %#v\n", posLine+1, posFilename) + fragment
+
+	fragmentC := C.CString(fragment)
+	defer C.free(unsafe.Pointer(fragmentC))
+
+	unsavedFile := C.struct_CXUnsavedFile{
+		Filename: filenameC,
+		Length:   C.ulong(len(fragment)),
+		Contents: fragmentC,
+	}
+
+	// convert Go slice of strings to C array of strings.
+	cmdargsC := C.malloc(C.size_t(len(cflags)) * C.size_t(unsafe.Sizeof(uintptr(0))))
+	defer C.free(cmdargsC)
+	cmdargs := (*[1 << 16]*C.char)(cmdargsC)
+	for i, cflag := range cflags {
+		s := C.CString(cflag)
+		cmdargs[i] = s
+		defer C.free(unsafe.Pointer(s))
+	}
+
+	var unit C.CXTranslationUnit
+	errCode := C.clang_parseTranslationUnit2(
+		index,
+		filenameC,
+		(**C.char)(cmdargsC), C.int(len(cflags)), // command line args
+		&unsavedFile, 1, // unsaved files
+		C.CXTranslationUnit_DetailedPreprocessingRecord,
+		&unit)
+	if errCode != 0 {
+		panic("loader: failed to parse source with libclang")
+	}
+	defer C.clang_disposeTranslationUnit(unit)
+
+	if numDiagnostics := int(C.clang_getNumDiagnostics(unit)); numDiagnostics != 0 {
+		addDiagnostic := func(diagnostic C.CXDiagnostic) {
+			spelling := getString(C.clang_getDiagnosticSpelling(diagnostic))
+			severity := diagnosticSeverity[C.clang_getDiagnosticSeverity(diagnostic)]
+			location := C.clang_getDiagnosticLocation(diagnostic)
+			var libclangFilename C.CXString
+			var line C.unsigned
+			var column C.unsigned
+			C.clang_getPresumedLocation(location, &libclangFilename, &line, &column)
+			filename := getString(libclangFilename)
+			if filepath.IsAbs(filename) {
+				// Relative paths for readability, like other Go parser errors.
+				relpath, err := filepath.Rel(p.dir, filename)
+				if err == nil {
+					filename = relpath
+				}
+			}
+			p.errors = append(p.errors, &scanner.Error{
+				Pos: token.Position{
+					Filename: filename,
+					Offset:   0, // not provided by clang_getPresumedLocation
+					Line:     int(line),
+					Column:   int(column),
+				},
+				Msg: severity + ": " + spelling,
+			})
+		}
+		for i := 0; i < numDiagnostics; i++ {
+			diagnostic := C.clang_getDiagnostic(unit, C.uint(i))
+			addDiagnostic(diagnostic)
+
+			// Child diagnostics (like notes on redefinitions).
+			diagnostics := C.clang_getChildDiagnostics(diagnostic)
+			for j := 0; j < int(C.clang_getNumDiagnosticsInSet(diagnostics)); j++ {
+				addDiagnostic(C.clang_getDiagnosticInSet(diagnostics, C.uint(j)))
+			}
+		}
+		return
+	}
+
+	ref := storedRefs.Put(p)
+	defer storedRefs.Remove(ref)
+	cursor := C.tinygo_clang_getTranslationUnitCursor(unit)
+	C.tinygo_clang_visitChildren(cursor, C.CXCursorVisitor(C.tinygo_clang_globals_visitor), C.CXClientData(ref))
+}
+
+//export tinygo_clang_globals_visitor
+func tinygo_clang_globals_visitor(c, parent C.GoCXCursor, client_data C.CXClientData) C.int {
+	p := storedRefs.Get(unsafe.Pointer(client_data)).(*cgoPackage)
+	kind := C.tinygo_clang_getCursorKind(c)
+	pos := p.getCursorPosition(c)
+	switch kind {
+	case C.CXCursor_FunctionDecl:
+		name := getString(C.tinygo_clang_getCursorSpelling(c))
+		if _, required := p.missingSymbols[name]; !required {
+			return C.CXChildVisit_Continue
+		}
+		cursorType := C.tinygo_clang_getCursorType(c)
+		if C.clang_isFunctionTypeVariadic(cursorType) != 0 {
+			return C.CXChildVisit_Continue // not supported
+		}
+		numArgs := int(C.tinygo_clang_Cursor_getNumArguments(c))
+		fn := &functionInfo{
+			pos: pos,
+		}
+		p.functions[name] = fn
+		for i := 0; i < numArgs; i++ {
+			arg := C.tinygo_clang_Cursor_getArgument(c, C.uint(i))
+			argName := getString(C.tinygo_clang_getCursorSpelling(arg))
+			argType := C.clang_getArgType(cursorType, C.uint(i))
+			if argName == "" {
+				argName = "$" + strconv.Itoa(i)
+			}
+			fn.args = append(fn.args, paramInfo{
+				name:     argName,
+				typeExpr: p.makeASTType(argType, pos),
+			})
+		}
+		resultType := C.tinygo_clang_getCursorResultType(c)
+		if resultType.kind != C.CXType_Void {
+			fn.results = &ast.FieldList{
+				List: []*ast.Field{
+					&ast.Field{
+						Type: p.makeASTType(resultType, pos),
+					},
+				},
+			}
+		}
+	case C.CXCursor_StructDecl:
+		typ := C.tinygo_clang_getCursorType(c)
+		name := getString(C.tinygo_clang_getCursorSpelling(c))
+		if _, required := p.missingSymbols["struct_"+name]; !required {
+			return C.CXChildVisit_Continue
+		}
+		p.makeASTType(typ, pos)
+	case C.CXCursor_TypedefDecl:
+		typedefType := C.tinygo_clang_getCursorType(c)
+		name := getString(C.clang_getTypedefName(typedefType))
+		if _, required := p.missingSymbols[name]; !required {
+			return C.CXChildVisit_Continue
+		}
+		p.makeASTType(typedefType, pos)
+	case C.CXCursor_VarDecl:
+		name := getString(C.tinygo_clang_getCursorSpelling(c))
+		if _, required := p.missingSymbols[name]; !required {
+			return C.CXChildVisit_Continue
+		}
+		cursorType := C.tinygo_clang_getCursorType(c)
+		p.globals[name] = globalInfo{
+			typeExpr: p.makeASTType(cursorType, pos),
+			pos:      pos,
+		}
+	case C.CXCursor_MacroDefinition:
+		name := getString(C.tinygo_clang_getCursorSpelling(c))
+		if _, required := p.missingSymbols[name]; !required {
+			return C.CXChildVisit_Continue
+		}
+		sourceRange := C.tinygo_clang_getCursorExtent(c)
+		start := C.clang_getRangeStart(sourceRange)
+		end := C.clang_getRangeEnd(sourceRange)
+		var file, endFile C.CXFile
+		var startOffset, endOffset C.unsigned
+		C.clang_getExpansionLocation(start, &file, nil, nil, &startOffset)
+		if file == nil {
+			panic("could not find file where macro is defined")
+		}
+		C.clang_getExpansionLocation(end, &endFile, nil, nil, &endOffset)
+		if file != endFile {
+			panic("expected start and end location of a #define to be in the same file")
+		}
+		if startOffset > endOffset {
+			panic("startOffset > endOffset")
+		}
+
+		// read file contents and extract the relevant byte range
+		tu := C.tinygo_clang_Cursor_getTranslationUnit(c)
+		var size C.size_t
+		sourcePtr := C.clang_getFileContents(tu, file, &size)
+		if endOffset >= C.uint(size) {
+			panic("endOffset lies after end of file")
+		}
+		source := string(((*[1 << 28]byte)(unsafe.Pointer(sourcePtr)))[startOffset:endOffset:endOffset])
+		if !strings.HasPrefix(source, name) {
+			panic(fmt.Sprintf("expected #define value to start with %#v, got %#v", name, source))
+		}
+		value := strings.TrimSpace(source[len(name):])
+		for len(value) != 0 && value[0] == '(' && value[len(value)-1] == ')' {
+			value = strings.TrimSpace(value[1 : len(value)-1])
+		}
+		if len(value) == 0 {
+			// Pretend it doesn't exist at all.
+			return C.CXChildVisit_Continue
+		}
+		// For information about integer literals:
+		// https://en.cppreference.com/w/cpp/language/integer_literal
+		if value[0] == '"' {
+			// string constant
+			p.constants[name] = constantInfo{&ast.BasicLit{pos, token.STRING, value}, pos}
+			return C.CXChildVisit_Continue
+		}
+		if value[0] == '\'' {
+			// char constant
+			p.constants[name] = constantInfo{&ast.BasicLit{pos, token.CHAR, value}, pos}
+			return C.CXChildVisit_Continue
+		}
+		// assume it's a number (int or float)
+		value = strings.Replace(value, "'", "", -1) // remove ' chars
+		value = strings.TrimRight(value, "lu")      // remove llu suffixes etc.
+		// find the first non-number
+		nonnum := byte(0)
+		for i := 0; i < len(value); i++ {
+			if value[i] < '0' || value[i] > '9' {
+				nonnum = value[i]
+				break
+			}
+		}
+		// determine number type based on the first non-number
+		switch nonnum {
+		case 0:
+			// no non-number found, must be an integer
+			p.constants[name] = constantInfo{&ast.BasicLit{pos, token.INT, value}, pos}
+		case 'x', 'X':
+			// hex integer constant
+			// TODO: may also be a floating point number per C++17.
+			p.constants[name] = constantInfo{&ast.BasicLit{pos, token.INT, value}, pos}
+		case '.', 'e':
+			// float constant
+			value = strings.TrimRight(value, "fFlL")
+			p.constants[name] = constantInfo{&ast.BasicLit{pos, token.FLOAT, value}, pos}
+		default:
+			// unknown type, ignore
+		}
+	}
+	return C.CXChildVisit_Continue
+}
+
+func getString(clangString C.CXString) (s string) {
+	rawString := C.clang_getCString(clangString)
+	s = C.GoString(rawString)
+	C.clang_disposeString(clangString)
+	return
+}
+
+// getCursorPosition returns a usable token.Pos from a libclang cursor. If the
+// file for this cursor has not been seen before, it is read from libclang
+// (which already has the file in memory) and added to the token.FileSet.
+func (p *cgoPackage) getCursorPosition(cursor C.GoCXCursor) token.Pos {
+	location := C.tinygo_clang_getCursorLocation(cursor)
+	var file C.CXFile
+	var line C.unsigned
+	var column C.unsigned
+	var offset C.unsigned
+	C.clang_getExpansionLocation(location, &file, &line, &column, &offset)
+	if line == 0 || file == nil {
+		// Invalid token.
+		return token.NoPos
+	}
+	filename := getString(C.clang_getFileName(file))
+	if _, ok := p.tokenFiles[filename]; !ok {
+		// File has not been seen before in this package, add line information
+		// now by reading the file from libclang.
+		tu := C.tinygo_clang_Cursor_getTranslationUnit(cursor)
+		var size C.size_t
+		sourcePtr := C.clang_getFileContents(tu, file, &size)
+		source := ((*[1 << 28]byte)(unsafe.Pointer(sourcePtr)))[:size:size]
+		lines := []int{0}
+		for i := 0; i < len(source)-1; i++ {
+			if source[i] == '\n' {
+				lines = append(lines, i+1)
+			}
+		}
+		f := p.fset.AddFile(filename, -1, int(size))
+		f.SetLines(lines)
+		p.tokenFiles[filename] = f
+	}
+	return p.tokenFiles[filename].Pos(int(offset))
+}
+
+// makeASTType return the ast.Expr for the given libclang type. In other words,
+// it converts a libclang type to a type in the Go AST.
+func (p *cgoPackage) makeASTType(typ C.CXType, pos token.Pos) ast.Expr {
+	var typeName string
+	switch typ.kind {
+	case C.CXType_Char_S, C.CXType_Char_U:
+		typeName = "C.char"
+	case C.CXType_SChar:
+		typeName = "C.schar"
+	case C.CXType_UChar:
+		typeName = "C.uchar"
+	case C.CXType_Short:
+		typeName = "C.short"
+	case C.CXType_UShort:
+		typeName = "C.ushort"
+	case C.CXType_Int:
+		typeName = "C.int"
+	case C.CXType_UInt:
+		typeName = "C.uint"
+	case C.CXType_Long:
+		typeName = "C.long"
+	case C.CXType_ULong:
+		typeName = "C.ulong"
+	case C.CXType_LongLong:
+		typeName = "C.longlong"
+	case C.CXType_ULongLong:
+		typeName = "C.ulonglong"
+	case C.CXType_Bool:
+		typeName = "bool"
+	case C.CXType_Float, C.CXType_Double, C.CXType_LongDouble:
+		switch C.clang_Type_getSizeOf(typ) {
+		case 4:
+			typeName = "float32"
+		case 8:
+			typeName = "float64"
+		default:
+			// Don't do anything, rely on the fallback code to show a somewhat
+			// sensible error message like "undeclared name: C.long double".
+		}
+	case C.CXType_Complex:
+		switch C.clang_Type_getSizeOf(typ) {
+		case 8:
+			typeName = "complex64"
+		case 16:
+			typeName = "complex128"
+		}
+	case C.CXType_Pointer:
+		pointeeType := C.clang_getPointeeType(typ)
+		if pointeeType.kind == C.CXType_Void {
+			// void* type is translated to Go as unsafe.Pointer
+			return &ast.SelectorExpr{
+				X: &ast.Ident{
+					NamePos: pos,
+					Name:    "unsafe",
+				},
+				Sel: &ast.Ident{
+					NamePos: pos,
+					Name:    "Pointer",
+				},
+			}
+		}
+		return &ast.StarExpr{
+			Star: pos,
+			X:    p.makeASTType(pointeeType, pos),
+		}
+	case C.CXType_ConstantArray:
+		return &ast.ArrayType{
+			Lbrack: pos,
+			Len: &ast.BasicLit{
+				ValuePos: pos,
+				Kind:     token.INT,
+				Value:    strconv.FormatInt(int64(C.clang_getArraySize(typ)), 10),
+			},
+			Elt: p.makeASTType(C.clang_getElementType(typ), pos),
+		}
+	case C.CXType_FunctionProto:
+		// Be compatible with gc, which uses the *[0]byte type for function
+		// pointer types.
+		// Return type [0]byte because this is a function type, not a pointer to
+		// this function type.
+		return &ast.ArrayType{
+			Lbrack: pos,
+			Len: &ast.BasicLit{
+				ValuePos: pos,
+				Kind:     token.INT,
+				Value:    "0",
+			},
+			Elt: &ast.Ident{
+				NamePos: pos,
+				Name:    "byte",
+			},
+		}
+	case C.CXType_Typedef:
+		name := getString(C.clang_getTypedefName(typ))
+		if _, ok := p.typedefs[name]; !ok {
+			p.typedefs[name] = nil // don't recurse
+			c := C.tinygo_clang_getTypeDeclaration(typ)
+			underlyingType := C.tinygo_clang_getTypedefDeclUnderlyingType(c)
+			expr := p.makeASTType(underlyingType, pos)
+			if strings.HasPrefix(name, "_Cgo_") {
+				expr := expr.(*ast.Ident)
+				typeSize := C.clang_Type_getSizeOf(underlyingType)
+				switch expr.Name {
+				case "C.char":
+					if typeSize != 1 {
+						// This happens for some very special purpose architectures
+						// (DSPs etc.) that are not currently targeted.
+						// https://www.embecosm.com/2017/04/18/non-8-bit-char-support-in-clang-and-llvm/
+						panic("unknown char width")
+					}
+					switch underlyingType.kind {
+					case C.CXType_Char_S:
+						expr.Name = "int8"
+					case C.CXType_Char_U:
+						expr.Name = "uint8"
+					}
+				case "C.schar", "C.short", "C.int", "C.long", "C.longlong":
+					switch typeSize {
+					case 1:
+						expr.Name = "int8"
+					case 2:
+						expr.Name = "int16"
+					case 4:
+						expr.Name = "int32"
+					case 8:
+						expr.Name = "int64"
+					}
+				case "C.uchar", "C.ushort", "C.uint", "C.ulong", "C.ulonglong":
+					switch typeSize {
+					case 1:
+						expr.Name = "uint8"
+					case 2:
+						expr.Name = "uint16"
+					case 4:
+						expr.Name = "uint32"
+					case 8:
+						expr.Name = "uint64"
+					}
+				}
+			}
+			p.typedefs[name] = &typedefInfo{
+				typeExpr: expr,
+				pos:      pos,
+			}
+		}
+		return &ast.Ident{
+			NamePos: pos,
+			Name:    "C." + name,
+		}
+	case C.CXType_Elaborated:
+		underlying := C.clang_Type_getNamedType(typ)
+		switch underlying.kind {
+		case C.CXType_Record:
+			return p.makeASTType(underlying, pos)
+		case C.CXType_Enum:
+			return p.makeASTType(underlying, pos)
+		default:
+			panic("unknown elaborated type")
+		}
+	case C.CXType_Record:
+		cursor := C.tinygo_clang_getTypeDeclaration(typ)
+		name := getString(C.tinygo_clang_getCursorSpelling(cursor))
+		if name == "" {
+			// Anonymous record, probably inside a typedef.
+			typeExpr, bitfieldList := p.makeASTRecordType(cursor, pos)
+			if bitfieldList != nil {
+				// This struct has bitfields, so we have to declare it as a
+				// named type (for bitfield getters/setters to work).
+				p.anonStructNum++
+				cgoName := "struct_" + strconv.Itoa(p.anonStructNum)
+				p.elaboratedTypes[cgoName] = &elaboratedTypeInfo{
+					typeExpr:  typeExpr,
+					pos:       pos,
+					bitfields: bitfieldList,
+				}
+				return &ast.Ident{
+					NamePos: pos,
+					Name:    "C." + cgoName,
+				}
+			}
+			return typeExpr
+		} else {
+			var cgoName string
+			switch C.tinygo_clang_getCursorKind(cursor) {
+			case C.CXCursor_StructDecl:
+				cgoName = "struct_" + name
+			case C.CXCursor_UnionDecl:
+				cgoName = "union_" + name
+			default:
+				panic("unknown record declaration")
+			}
+			if _, ok := p.elaboratedTypes[cgoName]; !ok {
+				p.elaboratedTypes[cgoName] = nil // predeclare (to avoid endless recursion)
+				typeExpr, bitfieldList := p.makeASTRecordType(cursor, pos)
+				p.elaboratedTypes[cgoName] = &elaboratedTypeInfo{
+					typeExpr:  typeExpr,
+					pos:       pos,
+					bitfields: bitfieldList,
+				}
+			}
+			return &ast.Ident{
+				NamePos: pos,
+				Name:    "C." + cgoName,
+			}
+		}
+	case C.CXType_Enum:
+		cursor := C.tinygo_clang_getTypeDeclaration(typ)
+		name := getString(C.tinygo_clang_getCursorSpelling(cursor))
+		underlying := C.tinygo_clang_getEnumDeclIntegerType(cursor)
+		if name == "" {
+			// anonymous enum
+			ref := storedRefs.Put(p)
+			defer storedRefs.Remove(ref)
+			C.tinygo_clang_visitChildren(cursor, C.CXCursorVisitor(C.tinygo_clang_enum_visitor), C.CXClientData(ref))
+			return p.makeASTType(underlying, pos)
+		} else {
+			// named enum
+			if _, ok := p.enums[name]; !ok {
+				ref := storedRefs.Put(p)
+				defer storedRefs.Remove(ref)
+				C.tinygo_clang_visitChildren(cursor, C.CXCursorVisitor(C.tinygo_clang_enum_visitor), C.CXClientData(ref))
+				p.enums[name] = enumInfo{
+					typeExpr: p.makeASTType(underlying, pos),
+					pos:      pos,
+				}
+			}
+			return &ast.Ident{
+				NamePos: pos,
+				Name:    "C.enum_" + name,
+			}
+		}
+	}
+	if typeName == "" {
+		// Report this as an error.
+		spelling := getString(C.clang_getTypeSpelling(typ))
+		p.errors = append(p.errors, scanner.Error{
+			Pos: p.fset.PositionFor(pos, true),
+			Msg: fmt.Sprintf("unknown C type: %v (libclang type kind %d)", spelling, typ.kind),
+		})
+		// Fallback, probably incorrect but at least the error points to an odd
+		// type name.
+		typeName = "C." + spelling
+	}
+	return &ast.Ident{
+		NamePos: pos,
+		Name:    typeName,
+	}
+}
+
+// makeASTRecordType parses a C record (struct or union) and translates it into
+// a Go struct type. Unions are implemented by setting the first field to a
+// zero-lengt "C union" field, which cannot be written in Go directly.
+func (p *cgoPackage) makeASTRecordType(cursor C.GoCXCursor, pos token.Pos) (*ast.StructType, []bitfieldInfo) {
+	fieldList := &ast.FieldList{
+		Opening: pos,
+		Closing: pos,
+	}
+	var bitfieldList []bitfieldInfo
+	inBitfield := false
+	bitfieldNum := 0
+	ref := storedRefs.Put(struct {
+		fieldList    *ast.FieldList
+		pkg          *cgoPackage
+		inBitfield   *bool
+		bitfieldNum  *int
+		bitfieldList *[]bitfieldInfo
+	}{fieldList, p, &inBitfield, &bitfieldNum, &bitfieldList})
+	defer storedRefs.Remove(ref)
+	C.tinygo_clang_visitChildren(cursor, C.CXCursorVisitor(C.tinygo_clang_struct_visitor), C.CXClientData(ref))
+	switch C.tinygo_clang_getCursorKind(cursor) {
+	case C.CXCursor_StructDecl:
+		return &ast.StructType{
+			Struct: pos,
+			Fields: fieldList,
+		}, bitfieldList
+	case C.CXCursor_UnionDecl:
+		if bitfieldList != nil {
+			// This is valid C... but please don't do this.
+			p.errors = append(p.errors, scanner.Error{
+				Pos: p.fset.PositionFor(pos, true),
+				Msg: fmt.Sprintf("bitfield in a union is not supported"),
+			})
+		}
+		if len(fieldList.List) > 1 {
+			// Insert a special field at the front (of zero width) as a
+			// marker that this is struct is actually a union. This is done
+			// by giving the field a name that cannot be expressed directly
+			// in Go.
+			// Other parts of the compiler look at the first element in a
+			// struct (of size > 2) to know whether this is a union.
+			// Note that we don't have to insert it for single-element
+			// unions as they're basically equivalent to a struct.
+			unionMarker := &ast.Field{
+				Type: &ast.StructType{
+					Struct: pos,
+				},
+			}
+			unionMarker.Names = []*ast.Ident{
+				&ast.Ident{
+					NamePos: pos,
+					Name:    "C union",
+					Obj: &ast.Object{
+						Kind: ast.Var,
+						Name: "C union",
+						Decl: unionMarker,
+					},
+				},
+			}
+			fieldList.List = append([]*ast.Field{unionMarker}, fieldList.List...)
+		}
+		return &ast.StructType{
+			Struct: pos,
+			Fields: fieldList,
+		}, bitfieldList
+	default:
+		panic("unknown record declaration")
+	}
+}
+
+//export tinygo_clang_struct_visitor
+func tinygo_clang_struct_visitor(c, parent C.GoCXCursor, client_data C.CXClientData) C.int {
+	passed := storedRefs.Get(unsafe.Pointer(client_data)).(struct {
+		fieldList    *ast.FieldList
+		pkg          *cgoPackage
+		inBitfield   *bool
+		bitfieldNum  *int
+		bitfieldList *[]bitfieldInfo
+	})
+	fieldList := passed.fieldList
+	p := passed.pkg
+	inBitfield := passed.inBitfield
+	bitfieldNum := passed.bitfieldNum
+	bitfieldList := passed.bitfieldList
+	if C.tinygo_clang_getCursorKind(c) != C.CXCursor_FieldDecl {
+		panic("expected field inside cursor")
+	}
+	name := getString(C.tinygo_clang_getCursorSpelling(c))
+	if name == "" {
+		// Assume this is a bitfield of 0 bits.
+		// Warning: this is not necessarily true!
+		return C.CXChildVisit_Continue
+	}
+	typ := C.tinygo_clang_getCursorType(c)
+	pos := p.getCursorPosition(c)
+	field := &ast.Field{
+		Type: p.makeASTType(typ, p.getCursorPosition(c)),
+	}
+	offsetof := int64(C.clang_Type_getOffsetOf(C.tinygo_clang_getCursorType(parent), C.CString(name)))
+	alignOf := int64(C.clang_Type_getAlignOf(typ) * 8)
+	bitfieldOffset := offsetof % alignOf
+	if bitfieldOffset != 0 {
+		if C.tinygo_clang_Cursor_isBitField(c) != 1 {
+			panic("expected a bitfield")
+		}
+		if !*inBitfield {
+			*bitfieldNum++
+		}
+		bitfieldName := "__bitfield_" + strconv.Itoa(*bitfieldNum)
+		prevField := fieldList.List[len(fieldList.List)-1]
+		if !*inBitfield {
+			// The previous element also was a bitfield, but wasn't noticed
+			// then. Add it now.
+			*inBitfield = true
+			*bitfieldList = append(*bitfieldList, bitfieldInfo{
+				field:    prevField,
+				name:     prevField.Names[0].Name,
+				startBit: 0,
+				pos:      prevField.Names[0].NamePos,
+			})
+			prevField.Names[0].Name = bitfieldName
+			prevField.Names[0].Obj.Name = bitfieldName
+		}
+		prevBitfield := &(*bitfieldList)[len(*bitfieldList)-1]
+		prevBitfield.endBit = bitfieldOffset
+		*bitfieldList = append(*bitfieldList, bitfieldInfo{
+			field:    prevField,
+			name:     name,
+			startBit: bitfieldOffset,
+			pos:      pos,
+		})
+		return C.CXChildVisit_Continue
+	}
+	*inBitfield = false
+	field.Names = []*ast.Ident{
+		&ast.Ident{
+			NamePos: pos,
+			Name:    name,
+			Obj: &ast.Object{
+				Kind: ast.Var,
+				Name: name,
+				Decl: field,
+			},
+		},
+	}
+	fieldList.List = append(fieldList.List, field)
+	return C.CXChildVisit_Continue
+}
+
+//export tinygo_clang_enum_visitor
+func tinygo_clang_enum_visitor(c, parent C.GoCXCursor, client_data C.CXClientData) C.int {
+	p := storedRefs.Get(unsafe.Pointer(client_data)).(*cgoPackage)
+	name := getString(C.tinygo_clang_getCursorSpelling(c))
+	pos := p.getCursorPosition(c)
+	value := C.tinygo_clang_getEnumConstantDeclValue(c)
+	p.constants[name] = constantInfo{
+		expr: &ast.BasicLit{pos, token.INT, strconv.FormatInt(int64(value), 10)},
+		pos:  pos,
+	}
+	return C.CXChildVisit_Continue
+}

cgo/libclang_config.go

@@ -1,6 +1,6 @@
 // +build !byollvm
 
-package loader
+package cgo
 
 /*
 #cgo linux  CFLAGS: -I/usr/lib/llvm-8/include

cgo/libclang_stubs.c

@@ -44,3 +44,27 @@ int tinygo_clang_Cursor_getNumArguments(CXCursor c) {
 CXCursor tinygo_clang_Cursor_getArgument(CXCursor c, unsigned i) {
 	return clang_Cursor_getArgument(c, i);
 }
+
+CXSourceLocation tinygo_clang_getCursorLocation(CXCursor c) {
+	return clang_getCursorLocation(c);
+}
+
+CXSourceRange tinygo_clang_getCursorExtent(CXCursor c) {
+	return clang_getCursorExtent(c);
+}
+
+CXTranslationUnit tinygo_clang_Cursor_getTranslationUnit(CXCursor c) {
+	return clang_Cursor_getTranslationUnit(c);
+}
+
+long long tinygo_clang_getEnumConstantDeclValue(CXCursor c) {
+	return clang_getEnumConstantDeclValue(c);
+}
+
+CXType tinygo_clang_getEnumDeclIntegerType(CXCursor c) {
+	return clang_getEnumDeclIntegerType(c);
+}
+
+unsigned tinygo_clang_Cursor_isBitField(CXCursor c) {
+	return clang_Cursor_isBitField(c);
+}

cgo/sync.go

@@ -1,4 +1,4 @@
-package loader
+package cgo
 
 import (
 	"sync"
@@ -8,17 +8,17 @@ import (
 // #include <stdlib.h>
 import "C"
 
-// RefMap is a convenient way to store opaque references that can be passed to
+// refMap is a convenient way to store opaque references that can be passed to
 // C. It is useful if an API uses function pointers and you cannot pass a Go
 // pointer but only a C pointer.
-type RefMap struct {
+type refMap struct {
 	refs map[unsafe.Pointer]interface{}
 	lock sync.Mutex
 }
 
 // Put stores a value in the map. It can later be retrieved using Get. It must
 // be removed using Remove to avoid memory leaks.
-func (m *RefMap) Put(v interface{}) unsafe.Pointer {
+func (m *refMap) Put(v interface{}) unsafe.Pointer {
 	m.lock.Lock()
 	defer m.lock.Unlock()
 	if m.refs == nil {
@@ -31,14 +31,14 @@ func (m *RefMap) Put(v interface{}) unsafe.Pointer {
 
 // Get returns a stored value previously inserted with Put. Use the same
 // reference as you got from Put.
-func (m *RefMap) Get(ref unsafe.Pointer) interface{} {
+func (m *refMap) Get(ref unsafe.Pointer) interface{} {
 	m.lock.Lock()
 	defer m.lock.Unlock()
 	return m.refs[ref]
 }
 
 // Remove deletes a single reference from the map.
-func (m *RefMap) Remove(ref unsafe.Pointer) {
+func (m *refMap) Remove(ref unsafe.Pointer) {
 	m.lock.Lock()
 	defer m.lock.Unlock()
 	delete(m.refs, ref)

commands.go

@@ -1,10 +1,45 @@
-// +build !darwin
-
 package main
 
-// commands used by the compilation process might have different file names on Linux than those used on macOS.
-var commands = map[string]string{
-	"clang":   "clang-8",
-	"ld.lld":  "ld.lld-8",
-	"wasm-ld": "wasm-ld-8",
+import (
+	"errors"
+	"os"
+	"os/exec"
+	"runtime"
+	"strings"
+)
+
+// Commands used by the compilation process might have different file names
+// across operating systems and distributions.
+var commands = map[string][]string{
+	"clang":   {"clang-8"},
+	"ld.lld":  {"ld.lld-8", "ld.lld"},
+	"wasm-ld": {"wasm-ld-8", "wasm-ld"},
+}
+
+func init() {
+	// Add the path to a Homebrew-installed LLVM 8 for ease of use (no need to
+	// manually set $PATH).
+	if runtime.GOOS == "darwin" {
+		commands["clang"] = append(commands["clang"], "/usr/local/opt/llvm/bin/clang-8")
+		commands["ld.lld"] = append(commands["ld.lld"], "/usr/local/opt/llvm/bin/ld.lld")
+		commands["wasm-ld"] = append(commands["wasm-ld"], "/usr/local/opt/llvm/bin/wasm-ld")
+	}
+}
+
+func execCommand(cmdNames []string, args ...string) error {
+	for _, cmdName := range cmdNames {
+		cmd := exec.Command(cmdName, args...)
+		cmd.Stdout = os.Stdout
+		cmd.Stderr = os.Stderr
+		err := cmd.Run()
+		if err != nil {
+			if err, ok := err.(*exec.Error); ok && err.Err == exec.ErrNotFound {
+				// this command was not found, try the next
+				continue
+			}
+			return err
+		}
+		return nil
+	}
+	return errors.New("none of these commands were found in your $PATH: " + strings.Join(cmdNames, " "))
 }

commands_macos.go

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

compiler/asserts.go

@@ -42,7 +42,7 @@ func (c *Compiler) emitLookupBoundsCheck(frame *Frame, arrayLen, index llvm.Valu
 
 	// Fail: this is a nil pointer, exit with a panic.
 	c.builder.SetInsertPointAtEnd(faultBlock)
-	c.createRuntimeCall("lookuppanic", nil, "")
+	c.createRuntimeCall("lookupPanic", nil, "")
 	c.builder.CreateUnreachable()
 
 	// Ok: this is a valid pointer.
@@ -103,7 +103,7 @@ func (c *Compiler) emitSliceBoundsCheck(frame *Frame, capacity, low, high llvm.V
 
 	// Fail: this is a nil pointer, exit with a panic.
 	c.builder.SetInsertPointAtEnd(faultBlock)
-	c.createRuntimeCall("slicepanic", nil, "")
+	c.createRuntimeCall("slicePanic", nil, "")
 	c.builder.CreateUnreachable()
 
 	// Ok: this is a valid pointer.
@@ -114,6 +114,11 @@ func (c *Compiler) emitSliceBoundsCheck(frame *Frame, capacity, low, high llvm.V
 // has no effect in well-behaved programs, but makes sure no uncaught nil
 // pointer dereferences exist in valid Go code.
 func (c *Compiler) emitNilCheck(frame *Frame, ptr llvm.Value, blockPrefix string) {
+	// Check whether we need to emit this check at all.
+	if !ptr.IsAGlobalValue().IsNil() {
+		return
+	}
+
 	// Check whether this is a nil pointer.
 	faultBlock := c.ctx.AddBasicBlock(frame.fn.LLVMFn, blockPrefix+".nil")
 	nextBlock := c.ctx.AddBasicBlock(frame.fn.LLVMFn, blockPrefix+".next")
@@ -141,7 +146,7 @@ func (c *Compiler) emitNilCheck(frame *Frame, ptr llvm.Value, blockPrefix string
 
 	// Fail: this is a nil pointer, exit with a panic.
 	c.builder.SetInsertPointAtEnd(faultBlock)
-	c.createRuntimeCall("nilpanic", nil, "")
+	c.createRuntimeCall("nilPanic", nil, "")
 	c.builder.CreateUnreachable()
 
 	// Ok: this is a valid pointer.

compiler/calls.go

@@ -38,7 +38,7 @@ func (c *Compiler) createCall(fn llvm.Value, args []llvm.Value, name string) llv
 }
 
 // Expand an argument type to a list that can be used in a function call
-// paramter list.
+// parameter list.
 func (c *Compiler) expandFormalParamType(t llvm.Type) []llvm.Type {
 	switch t.TypeKind() {
 	case llvm.StructTypeKind:
@@ -55,6 +55,25 @@ func (c *Compiler) expandFormalParamType(t llvm.Type) []llvm.Type {
 	}
 }
 
+// Expand an argument type to a list of offsets from the start of the object.
+// Used together with expandFormalParam to get the offset of each value from the
+// start of the non-expanded value.
+func (c *Compiler) expandFormalParamOffsets(t llvm.Type) []uint64 {
+	switch t.TypeKind() {
+	case llvm.StructTypeKind:
+		fields := c.flattenAggregateTypeOffsets(t)
+		if len(fields) <= MaxFieldsPerParam {
+			return fields
+		} else {
+			// failed to lower
+			return []uint64{0}
+		}
+	default:
+		// TODO: split small arrays
+		return []uint64{0}
+	}
+}
+
 // Equivalent of expandFormalParamType for parameter values.
 func (c *Compiler) expandFormalParam(v llvm.Value) []llvm.Value {
 	switch v.Type().TypeKind() {
@@ -92,6 +111,27 @@ func (c *Compiler) flattenAggregateType(t llvm.Type) []llvm.Type {
 	}
 }
 
+// Return the offsets from the start of the object if this object type were
+// flattened like in flattenAggregate. Used together with flattenAggregate to
+// know the start indices of each value in the non-flattened object.
+func (c *Compiler) flattenAggregateTypeOffsets(t llvm.Type) []uint64 {
+	switch t.TypeKind() {
+	case llvm.StructTypeKind:
+		fields := make([]uint64, 0, t.StructElementTypesCount())
+		for fieldIndex, field := range t.StructElementTypes() {
+			suboffsets := c.flattenAggregateTypeOffsets(field)
+			offset := c.targetData.ElementOffset(t, fieldIndex)
+			for i := range suboffsets {
+				suboffsets[i] += offset
+			}
+			fields = append(fields, suboffsets...)
+		}
+		return fields
+	default:
+		return []uint64{0}
+	}
+}
+
 // Break down a struct into its elementary types for argument passing. The value
 // equivalent of flattenAggregateType
 func (c *Compiler) flattenAggregate(v llvm.Value) []llvm.Value {
@@ -123,10 +163,7 @@ func (c *Compiler) collapseFormalParamInternal(t llvm.Type, fields []llvm.Value)
 	switch t.TypeKind() {
 	case llvm.StructTypeKind:
 		if len(c.flattenAggregateType(t)) <= MaxFieldsPerParam {
-			value, err := c.getZeroValue(t)
-			if err != nil {
-				panic("could not get zero value of struct: " + err.Error())
-			}
+			value := c.getZeroValue(t)
 			for i, subtyp := range t.StructElementTypes() {
 				structField, remaining := c.collapseFormalParamInternal(subtyp, fields)
 				fields = remaining

compiler/channel.go

@@ -4,6 +4,7 @@ package compiler
 // or pseudo-operations that are lowered during goroutine lowering.
 
 import (
+	"fmt"
 	"go/types"
 
 	"golang.org/x/tools/go/ssa"
@@ -12,86 +13,226 @@ import (
 
 // emitMakeChan returns a new channel value for the given channel type.
 func (c *Compiler) emitMakeChan(expr *ssa.MakeChan) (llvm.Value, error) {
-	valueType, err := c.getLLVMType(expr.Type().(*types.Chan).Elem())
-	if err != nil {
-		return llvm.Value{}, err
-	}
-	if c.targetData.TypeAllocSize(valueType) > c.targetData.TypeAllocSize(c.intType) {
-		// Values bigger than int overflow the data part of the coroutine.
-		// TODO: make the coroutine data part big enough to hold these bigger
-		// values.
-		return llvm.Value{}, c.makeError(expr.Pos(), "todo: channel with values bigger than int")
-	}
-	chanType := c.mod.GetTypeByName("runtime.channel")
-	size := c.targetData.TypeAllocSize(chanType)
+	chanType := c.getLLVMType(expr.Type())
+	size := c.targetData.TypeAllocSize(chanType.ElementType())
 	sizeValue := llvm.ConstInt(c.uintptrType, size, false)
 	ptr := c.createRuntimeCall("alloc", []llvm.Value{sizeValue}, "chan.alloc")
-	ptr = c.builder.CreateBitCast(ptr, llvm.PointerType(chanType, 0), "chan")
+	ptr = c.builder.CreateBitCast(ptr, chanType, "chan")
+	// Set the elementSize field
+	elementSizePtr := c.builder.CreateGEP(ptr, []llvm.Value{
+		llvm.ConstInt(c.ctx.Int32Type(), 0, false),
+		llvm.ConstInt(c.ctx.Int32Type(), 0, false),
+	}, "")
+	elementSize := c.targetData.TypeAllocSize(c.getLLVMType(expr.Type().(*types.Chan).Elem()))
+	if elementSize > 0xffff {
+		return ptr, c.makeError(expr.Pos(), fmt.Sprintf("element size is %d bytes, which is bigger than the maximum of %d bytes", elementSize, 0xffff))
+	}
+	elementSizeValue := llvm.ConstInt(c.ctx.Int16Type(), elementSize, false)
+	c.builder.CreateStore(elementSizeValue, elementSizePtr)
 	return ptr, nil
 }
 
 // emitChanSend emits a pseudo chan send operation. It is lowered to the actual
 // channel send operation during goroutine lowering.
-func (c *Compiler) emitChanSend(frame *Frame, instr *ssa.Send) error {
-	valueType, err := c.getLLVMType(instr.Chan.Type().(*types.Chan).Elem())
-	if err != nil {
-		return err
-	}
-	ch, err := c.parseExpr(frame, instr.Chan)
-	if err != nil {
-		return err
-	}
-	chanValue, err := c.parseExpr(frame, instr.X)
-	if err != nil {
-		return err
-	}
-	valueSize := llvm.ConstInt(c.uintptrType, c.targetData.TypeAllocSize(chanValue.Type()), false)
-	valueAlloca := c.builder.CreateAlloca(valueType, "chan.value")
+func (c *Compiler) emitChanSend(frame *Frame, instr *ssa.Send) {
+	ch := c.getValue(frame, instr.Chan)
+	chanValue := c.getValue(frame, instr.X)
+
+	// store value-to-send
+	valueType := c.getLLVMType(instr.X.Type())
+	valueAlloca, valueAllocaCast, valueAllocaSize := c.createTemporaryAlloca(valueType, "chan.value")
 	c.builder.CreateStore(chanValue, valueAlloca)
-	valueAllocaCast := c.builder.CreateBitCast(valueAlloca, c.i8ptrType, "chan.value.i8ptr")
-	c.createRuntimeCall("chanSendStub", []llvm.Value{llvm.Undef(c.i8ptrType), ch, valueAllocaCast, valueSize}, "")
-	return nil
+
+	// Do the send.
+	coroutine := c.createRuntimeCall("getCoroutine", nil, "")
+	c.createRuntimeCall("chanSend", []llvm.Value{coroutine, ch, valueAllocaCast}, "")
+
+	// End the lifetime of the alloca.
+	// This also works around a bug in CoroSplit, at least in LLVM 8:
+	// https://bugs.llvm.org/show_bug.cgi?id=41742
+	c.emitLifetimeEnd(valueAllocaCast, valueAllocaSize)
 }
 
 // emitChanRecv emits a pseudo chan receive operation. It is lowered to the
 // actual channel receive operation during goroutine lowering.
-func (c *Compiler) emitChanRecv(frame *Frame, unop *ssa.UnOp) (llvm.Value, error) {
-	valueType, err := c.getLLVMType(unop.X.Type().(*types.Chan).Elem())
-	if err != nil {
-		return llvm.Value{}, err
-	}
-	valueSize := llvm.ConstInt(c.uintptrType, c.targetData.TypeAllocSize(valueType), false)
-	ch, err := c.parseExpr(frame, unop.X)
-	if err != nil {
-		return llvm.Value{}, err
-	}
-	valueAlloca := c.builder.CreateAlloca(valueType, "chan.value")
-	valueAllocaCast := c.builder.CreateBitCast(valueAlloca, c.i8ptrType, "chan.value.i8ptr")
-	valueOk := c.builder.CreateAlloca(c.ctx.Int1Type(), "chan.comma-ok.alloca")
-	c.createRuntimeCall("chanRecvStub", []llvm.Value{llvm.Undef(c.i8ptrType), ch, valueAllocaCast, valueOk, valueSize}, "")
+func (c *Compiler) emitChanRecv(frame *Frame, unop *ssa.UnOp) llvm.Value {
+	valueType := c.getLLVMType(unop.X.Type().(*types.Chan).Elem())
+	ch := c.getValue(frame, unop.X)
+
+	// Allocate memory to receive into.
+	valueAlloca, valueAllocaCast, valueAllocaSize := c.createTemporaryAlloca(valueType, "chan.value")
+
+	// Do the receive.
+	coroutine := c.createRuntimeCall("getCoroutine", nil, "")
+	c.createRuntimeCall("chanRecv", []llvm.Value{coroutine, ch, valueAllocaCast}, "")
 	received := c.builder.CreateLoad(valueAlloca, "chan.received")
+	c.emitLifetimeEnd(valueAllocaCast, valueAllocaSize)
+
 	if unop.CommaOk {
-		commaOk := c.builder.CreateLoad(valueOk, "chan.comma-ok")
+		commaOk := c.createRuntimeCall("getTaskPromiseData", []llvm.Value{coroutine}, "chan.commaOk.wide")
+		commaOk = c.builder.CreateTrunc(commaOk, c.ctx.Int1Type(), "chan.commaOk")
 		tuple := llvm.Undef(c.ctx.StructType([]llvm.Type{valueType, c.ctx.Int1Type()}, false))
 		tuple = c.builder.CreateInsertValue(tuple, received, 0, "")
 		tuple = c.builder.CreateInsertValue(tuple, commaOk, 1, "")
-		return tuple, nil
+		return tuple
 	} else {
-		return received, nil
+		return received
 	}
 }
 
 // emitChanClose closes the given channel.
-func (c *Compiler) emitChanClose(frame *Frame, param ssa.Value) error {
-	valueType, err := c.getLLVMType(param.Type().(*types.Chan).Elem())
-	valueSize := llvm.ConstInt(c.uintptrType, c.targetData.TypeAllocSize(valueType), false)
-	if err != nil {
-		return err
-	}
-	ch, err := c.parseExpr(frame, param)
-	if err != nil {
-		return err
-	}
-	c.createRuntimeCall("chanClose", []llvm.Value{ch, valueSize}, "")
-	return nil
+func (c *Compiler) emitChanClose(frame *Frame, param ssa.Value) {
+	ch := c.getValue(frame, param)
+	c.createRuntimeCall("chanClose", []llvm.Value{ch}, "")
+}
+
+// emitSelect emits all IR necessary for a select statements. That's a
+// non-trivial amount of code because select is very complex to implement.
+func (c *Compiler) emitSelect(frame *Frame, expr *ssa.Select) llvm.Value {
+	if len(expr.States) == 0 {
+		// Shortcuts for some simple selects.
+		llvmType := c.getLLVMType(expr.Type())
+		if expr.Blocking {
+			// Blocks forever:
+			//     select {}
+			c.createRuntimeCall("deadlockStub", nil, "")
+			return llvm.Undef(llvmType)
+		} else {
+			// No-op:
+			//     select {
+			//     default:
+			//     }
+			retval := llvm.Undef(llvmType)
+			retval = c.builder.CreateInsertValue(retval, llvm.ConstInt(c.intType, 0xffffffffffffffff, true), 0, "")
+			return retval // {-1, false}
+		}
+	}
+
+	// This code create a (stack-allocated) slice containing all the select
+	// cases and then calls runtime.chanSelect to perform the actual select
+	// statement.
+	// Simple selects (blocking and with just one case) are already transformed
+	// into regular chan operations during SSA construction so we don't have to
+	// optimize such small selects.
+
+	// Go through all the cases. Create the selectStates slice and and
+	// determine the receive buffer size and alignment.
+	recvbufSize := uint64(0)
+	recvbufAlign := 0
+	hasReceives := false
+	var selectStates []llvm.Value
+	chanSelectStateType := c.getLLVMRuntimeType("chanSelectState")
+	for _, state := range expr.States {
+		ch := c.getValue(frame, state.Chan)
+		selectState := c.getZeroValue(chanSelectStateType)
+		selectState = c.builder.CreateInsertValue(selectState, ch, 0, "")
+		switch state.Dir {
+		case types.RecvOnly:
+			// Make sure the receive buffer is big enough and has the correct alignment.
+			llvmType := c.getLLVMType(state.Chan.Type().(*types.Chan).Elem())
+			if size := c.targetData.TypeAllocSize(llvmType); size > recvbufSize {
+				recvbufSize = size
+			}
+			if align := c.targetData.ABITypeAlignment(llvmType); align > recvbufAlign {
+				recvbufAlign = align
+			}
+			hasReceives = true
+		case types.SendOnly:
+			// Store this value in an alloca and put a pointer to this alloca
+			// in the send state.
+			sendValue := c.getValue(frame, state.Send)
+			alloca := c.createEntryBlockAlloca(sendValue.Type(), "select.send.value")
+			c.builder.CreateStore(sendValue, alloca)
+			ptr := c.builder.CreateBitCast(alloca, c.i8ptrType, "")
+			selectState = c.builder.CreateInsertValue(selectState, ptr, 1, "")
+		default:
+			panic("unreachable")
+		}
+		selectStates = append(selectStates, selectState)
+	}
+
+	// Create a receive buffer, where the received value will be stored.
+	recvbuf := llvm.Undef(c.i8ptrType)
+	if hasReceives {
+		allocaType := llvm.ArrayType(c.ctx.Int8Type(), int(recvbufSize))
+		recvbufAlloca := c.builder.CreateAlloca(allocaType, "select.recvbuf.alloca")
+		recvbufAlloca.SetAlignment(recvbufAlign)
+		recvbuf = c.builder.CreateGEP(recvbufAlloca, []llvm.Value{
+			llvm.ConstInt(c.ctx.Int32Type(), 0, false),
+			llvm.ConstInt(c.ctx.Int32Type(), 0, false),
+		}, "select.recvbuf")
+	}
+
+	// Create the states slice (allocated on the stack).
+	statesAllocaType := llvm.ArrayType(chanSelectStateType, len(selectStates))
+	statesAlloca := c.builder.CreateAlloca(statesAllocaType, "select.states.alloca")
+	for i, state := range selectStates {
+		// Set each slice element to the appropriate channel.
+		gep := c.builder.CreateGEP(statesAlloca, []llvm.Value{
+			llvm.ConstInt(c.ctx.Int32Type(), 0, false),
+			llvm.ConstInt(c.ctx.Int32Type(), uint64(i), false),
+		}, "")
+		c.builder.CreateStore(state, gep)
+	}
+	statesPtr := c.builder.CreateGEP(statesAlloca, []llvm.Value{
+		llvm.ConstInt(c.ctx.Int32Type(), 0, false),
+		llvm.ConstInt(c.ctx.Int32Type(), 0, false),
+	}, "select.states")
+	statesLen := llvm.ConstInt(c.uintptrType, uint64(len(selectStates)), false)
+
+	// Convert the 'blocking' flag on this select into a LLVM value.
+	blockingInt := uint64(0)
+	if expr.Blocking {
+		blockingInt = 1
+	}
+	blockingValue := llvm.ConstInt(c.ctx.Int1Type(), blockingInt, false)
+
+	// Do the select in the runtime.
+	results := c.createRuntimeCall("chanSelect", []llvm.Value{
+		recvbuf,
+		statesPtr, statesLen, statesLen, // []chanSelectState
+		blockingValue,
+	}, "")
+
+	// The result value does not include all the possible received values,
+	// because we can't load them in advance. Instead, the *ssa.Extract
+	// instruction will treat a *ssa.Select specially and load it there inline.
+	// Store the receive alloca in a sidetable until we hit this extract
+	// instruction.
+	if frame.selectRecvBuf == nil {
+		frame.selectRecvBuf = make(map[*ssa.Select]llvm.Value)
+	}
+	frame.selectRecvBuf[expr] = recvbuf
+
+	return results
+}
+
+// getChanSelectResult returns the special values from a *ssa.Extract expression
+// when extracting a value from a select statement (*ssa.Select). Because
+// *ssa.Select cannot load all values in advance, it does this later in the
+// *ssa.Extract expression.
+func (c *Compiler) getChanSelectResult(frame *Frame, expr *ssa.Extract) llvm.Value {
+	if expr.Index == 0 {
+		// index
+		value := c.getValue(frame, expr.Tuple)
+		index := c.builder.CreateExtractValue(value, expr.Index, "")
+		if index.Type().IntTypeWidth() < c.intType.IntTypeWidth() {
+			index = c.builder.CreateSExt(index, c.intType, "")
+		}
+		return index
+	} else if expr.Index == 1 {
+		// comma-ok
+		value := c.getValue(frame, expr.Tuple)
+		return c.builder.CreateExtractValue(value, expr.Index, "")
+	} else {
+		// Select statements are (index, ok, ...) where ... is a number of
+		// received values, depending on how many receive statements there
+		// are. They are all combined into one alloca (because only one
+		// receive can proceed at a time) so we'll get that alloca, bitcast
+		// it to the correct type, and dereference it.
+		recvbuf := frame.selectRecvBuf[expr.Tuple.(*ssa.Select)]
+		typ := llvm.PointerType(c.getLLVMType(expr.Type()), 0)
+		ptr := c.builder.CreateBitCast(recvbuf, typ, "")
+		return c.builder.CreateLoad(ptr, "")
+	}
 }

compiler/defer.go

@@ -29,14 +29,14 @@ func (c *Compiler) deferInitFunc(frame *Frame) {
 	frame.deferClosureFuncs = make(map[*ir.Function]int)
 
 	// Create defer list pointer.
-	deferType := llvm.PointerType(c.mod.GetTypeByName("runtime._defer"), 0)
+	deferType := llvm.PointerType(c.getLLVMRuntimeType("_defer"), 0)
 	frame.deferPtr = c.builder.CreateAlloca(deferType, "deferPtr")
 	c.builder.CreateStore(llvm.ConstPointerNull(deferType), frame.deferPtr)
 }
 
 // emitDefer emits a single defer instruction, to be run when this function
 // returns.
-func (c *Compiler) emitDefer(frame *Frame, instr *ssa.Defer) error {
+func (c *Compiler) emitDefer(frame *Frame, instr *ssa.Defer) {
 	// The pointer to the previous defer struct, which we will replace to
 	// make a linked list.
 	next := c.builder.CreateLoad(frame.deferPtr, "defer.next")
@@ -56,18 +56,12 @@ func (c *Compiler) emitDefer(frame *Frame, instr *ssa.Defer) error {
 
 		// Collect all values to be put in the struct (starting with
 		// runtime._defer fields, followed by the call parameters).
-		itf, err := c.parseExpr(frame, instr.Call.Value) // interface
-		if err != nil {
-			return err
-		}
+		itf := c.getValue(frame, instr.Call.Value) // interface
 		receiverValue := c.builder.CreateExtractValue(itf, 1, "invoke.func.receiver")
 		values = []llvm.Value{callback, next, receiverValue}
 		valueTypes = append(valueTypes, c.i8ptrType)
 		for _, arg := range instr.Call.Args {
-			val, err := c.parseExpr(frame, arg)
-			if err != nil {
-				return err
-			}
+			val := c.getValue(frame, arg)
 			values = append(values, val)
 			valueTypes = append(valueTypes, val.Type())
 		}
@@ -86,10 +80,7 @@ func (c *Compiler) emitDefer(frame *Frame, instr *ssa.Defer) error {
 		// runtime._defer fields).
 		values = []llvm.Value{callback, next}
 		for _, param := range instr.Call.Args {
-			llvmParam, err := c.parseExpr(frame, param)
-			if err != nil {
-				return err
-			}
+			llvmParam := c.getValue(frame, param)
 			values = append(values, llvmParam)
 			valueTypes = append(valueTypes, llvmParam.Type())
 		}
@@ -101,10 +92,7 @@ func (c *Compiler) emitDefer(frame *Frame, instr *ssa.Defer) error {
 		// pointer.
 		// TODO: ignore this closure entirely and put pointers to the free
 		// variables directly in the defer struct, avoiding a memory allocation.
-		closure, err := c.parseExpr(frame, instr.Call.Value)
-		if err != nil {
-			return err
-		}
+		closure := c.getValue(frame, instr.Call.Value)
 		context := c.builder.CreateExtractValue(closure, 0, "")
 
 		// Get the callback number.
@@ -120,10 +108,7 @@ func (c *Compiler) emitDefer(frame *Frame, instr *ssa.Defer) error {
 		// context pointer).
 		values = []llvm.Value{callback, next}
 		for _, param := range instr.Call.Args {
-			llvmParam, err := c.parseExpr(frame, param)
-			if err != nil {
-				return err
-			}
+			llvmParam := c.getValue(frame, param)
 			values = append(values, llvmParam)
 			valueTypes = append(valueTypes, llvmParam.Type())
 		}
@@ -131,15 +116,13 @@ func (c *Compiler) emitDefer(frame *Frame, instr *ssa.Defer) error {
 		valueTypes = append(valueTypes, context.Type())
 
 	} else {
-		return c.makeError(instr.Pos(), "todo: defer on uncommon function call type")
+		c.addError(instr.Pos(), "todo: defer on uncommon function call type")
+		return
 	}
 
 	// Make a struct out of the collected values to put in the defer frame.
 	deferFrameType := c.ctx.StructType(valueTypes, false)
-	deferFrame, err := c.getZeroValue(deferFrameType)
-	if err != nil {
-		return err
-	}
+	deferFrame := c.getZeroValue(deferFrameType)
 	for i, value := range values {
 		deferFrame = c.builder.CreateInsertValue(deferFrame, value, i, "")
 	}
@@ -147,15 +130,17 @@ func (c *Compiler) emitDefer(frame *Frame, instr *ssa.Defer) error {
 	// Put this struct in an alloca.
 	alloca := c.builder.CreateAlloca(deferFrameType, "defer.alloca")
 	c.builder.CreateStore(deferFrame, alloca)
+	if c.needsStackObjects() {
+		c.trackPointer(alloca)
+	}
 
 	// Push it on top of the linked list by replacing deferPtr.
 	allocaCast := c.builder.CreateBitCast(alloca, next.Type(), "defer.alloca.cast")
 	c.builder.CreateStore(allocaCast, frame.deferPtr)
-	return nil
 }
 
 // emitRunDefers emits code to run all deferred functions.
-func (c *Compiler) emitRunDefers(frame *Frame) error {
+func (c *Compiler) emitRunDefers(frame *Frame) {
 	// Add a loop like the following:
 	//     for stack != nil {
 	//         _stack := stack
@@ -190,13 +175,13 @@ func (c *Compiler) emitRunDefers(frame *Frame) error {
 	//     stack = stack.next
 	//     switch stack.callback {
 	c.builder.SetInsertPointAtEnd(loop)
-	nextStackGEP := c.builder.CreateGEP(deferData, []llvm.Value{
+	nextStackGEP := c.builder.CreateInBoundsGEP(deferData, []llvm.Value{
 		llvm.ConstInt(c.ctx.Int32Type(), 0, false),
 		llvm.ConstInt(c.ctx.Int32Type(), 1, false), // .next field
 	}, "stack.next.gep")
 	nextStack := c.builder.CreateLoad(nextStackGEP, "stack.next")
 	c.builder.CreateStore(nextStack, frame.deferPtr)
-	gep := c.builder.CreateGEP(deferData, []llvm.Value{
+	gep := c.builder.CreateInBoundsGEP(deferData, []llvm.Value{
 		llvm.ConstInt(c.ctx.Int32Type(), 0, false),
 		llvm.ConstInt(c.ctx.Int32Type(), 0, false), // .callback field
 	}, "callback.gep")
@@ -218,13 +203,9 @@ func (c *Compiler) emitRunDefers(frame *Frame) error {
 			}
 
 			// Get the real defer struct type and cast to it.
-			valueTypes := []llvm.Type{c.uintptrType, llvm.PointerType(c.mod.GetTypeByName("runtime._defer"), 0), c.i8ptrType}
+			valueTypes := []llvm.Type{c.uintptrType, llvm.PointerType(c.getLLVMRuntimeType("_defer"), 0), c.i8ptrType}
 			for _, arg := range callback.Args {
-				llvmType, err := c.getLLVMType(arg.Type())
-				if err != nil {
-					return err
-				}
-				valueTypes = append(valueTypes, llvmType)
+				valueTypes = append(valueTypes, c.getLLVMType(arg.Type()))
 			}
 			deferFrameType := c.ctx.StructType(valueTypes, false)
 			deferFramePtr := c.builder.CreateBitCast(deferData, llvm.PointerType(deferFrameType, 0), "deferFrame")
@@ -233,7 +214,7 @@ func (c *Compiler) emitRunDefers(frame *Frame) error {
 			forwardParams := []llvm.Value{}
 			zero := llvm.ConstInt(c.ctx.Int32Type(), 0, false)
 			for i := 2; i < len(valueTypes); i++ {
-				gep := c.builder.CreateGEP(deferFramePtr, []llvm.Value{zero, llvm.ConstInt(c.ctx.Int32Type(), uint64(i), false)}, "gep")
+				gep := c.builder.CreateInBoundsGEP(deferFramePtr, []llvm.Value{zero, llvm.ConstInt(c.ctx.Int32Type(), uint64(i), false)}, "gep")
 				forwardParam := c.builder.CreateLoad(gep, "param")
 				forwardParams = append(forwardParams, forwardParam)
 			}
@@ -246,23 +227,16 @@ func (c *Compiler) emitRunDefers(frame *Frame) error {
 			// Parent coroutine handle.
 			forwardParams = append(forwardParams, llvm.Undef(c.i8ptrType))
 
-			fnPtr, _, err := c.getInvokeCall(frame, callback)
-			if err != nil {
-				return err
-			}
+			fnPtr, _ := c.getInvokeCall(frame, callback)
 			c.createCall(fnPtr, forwardParams, "")
 
 		case *ir.Function:
 			// Direct call.
 
 			// Get the real defer struct type and cast to it.
-			valueTypes := []llvm.Type{c.uintptrType, llvm.PointerType(c.mod.GetTypeByName("runtime._defer"), 0)}
+			valueTypes := []llvm.Type{c.uintptrType, llvm.PointerType(c.getLLVMRuntimeType("_defer"), 0)}
 			for _, param := range callback.Params {
-				llvmType, err := c.getLLVMType(param.Type())
-				if err != nil {
-					return err
-				}
-				valueTypes = append(valueTypes, llvmType)
+				valueTypes = append(valueTypes, c.getLLVMType(param.Type()))
 			}
 			deferFrameType := c.ctx.StructType(valueTypes, false)
 			deferFramePtr := c.builder.CreateBitCast(deferData, llvm.PointerType(deferFrameType, 0), "deferFrame")
@@ -271,7 +245,7 @@ func (c *Compiler) emitRunDefers(frame *Frame) error {
 			forwardParams := []llvm.Value{}
 			zero := llvm.ConstInt(c.ctx.Int32Type(), 0, false)
 			for i := range callback.Params {
-				gep := c.builder.CreateGEP(deferFramePtr, []llvm.Value{zero, llvm.ConstInt(c.ctx.Int32Type(), uint64(i+2), false)}, "gep")
+				gep := c.builder.CreateInBoundsGEP(deferFramePtr, []llvm.Value{zero, llvm.ConstInt(c.ctx.Int32Type(), uint64(i+2), false)}, "gep")
 				forwardParam := c.builder.CreateLoad(gep, "param")
 				forwardParams = append(forwardParams, forwardParam)
 			}
@@ -289,14 +263,10 @@ func (c *Compiler) emitRunDefers(frame *Frame) error {
 		case *ssa.MakeClosure:
 			// Get the real defer struct type and cast to it.
 			fn := c.ir.GetFunction(callback.Fn.(*ssa.Function))
-			valueTypes := []llvm.Type{c.uintptrType, llvm.PointerType(c.mod.GetTypeByName("runtime._defer"), 0)}
+			valueTypes := []llvm.Type{c.uintptrType, llvm.PointerType(c.getLLVMRuntimeType("_defer"), 0)}
 			params := fn.Signature.Params()
 			for i := 0; i < params.Len(); i++ {
-				llvmType, err := c.getLLVMType(params.At(i).Type())
-				if err != nil {
-					return err
-				}
-				valueTypes = append(valueTypes, llvmType)
+				valueTypes = append(valueTypes, c.getLLVMType(params.At(i).Type()))
 			}
 			valueTypes = append(valueTypes, c.i8ptrType) // closure
 			deferFrameType := c.ctx.StructType(valueTypes, false)
@@ -306,7 +276,7 @@ func (c *Compiler) emitRunDefers(frame *Frame) error {
 			forwardParams := []llvm.Value{}
 			zero := llvm.ConstInt(c.ctx.Int32Type(), 0, false)
 			for i := 2; i < len(valueTypes); i++ {
-				gep := c.builder.CreateGEP(deferFramePtr, []llvm.Value{zero, llvm.ConstInt(c.ctx.Int32Type(), uint64(i), false)}, "")
+				gep := c.builder.CreateInBoundsGEP(deferFramePtr, []llvm.Value{zero, llvm.ConstInt(c.ctx.Int32Type(), uint64(i), false)}, "")
 				forwardParam := c.builder.CreateLoad(gep, "param")
 				forwardParams = append(forwardParams, forwardParam)
 			}
@@ -334,5 +304,4 @@ func (c *Compiler) emitRunDefers(frame *Frame) error {
 
 	// End of loop.
 	c.builder.SetInsertPointAtEnd(end)
-	return nil
 }

compiler/errors.go

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

compiler/func-lowering.go

@@ -51,7 +51,7 @@ func (c *Compiler) LowerFuncValues() {
 	}
 
 	// Find all func values used in the program with their signatures.
-	funcValueWithSignaturePtr := llvm.PointerType(c.mod.GetTypeByName("runtime.funcValueWithSignature"), 0)
+	funcValueWithSignaturePtr := llvm.PointerType(c.getLLVMRuntimeType("funcValueWithSignature"), 0)
 	signatures := map[string]*funcSignatureInfo{}
 	for global := c.mod.FirstGlobal(); !global.IsNil(); global = llvm.NextGlobal(global) {
 		if global.Type() != funcValueWithSignaturePtr {
@@ -154,17 +154,17 @@ func (c *Compiler) LowerFuncValues() {
 				// What we'll do is transform the following:
 				//     rawPtr := runtime.getFuncPtr(fn)
 				//     if func.rawPtr == nil {
-				//         runtime.nilpanic()
+				//         runtime.nilPanic()
 				//     }
 				//     result := func.rawPtr(...args, func.context)
 				// into this:
 				//     if false {
-				//         runtime.nilpanic()
+				//         runtime.nilPanic()
 				//     }
 				//     var result // Phi
 				//     switch fn.id {
 				//     case 0:
-				//         runtime.nilpanic()
+				//         runtime.nilPanic()
 				//     case 1:
 				//         result = call first implementation...
 				//     case 2:
@@ -222,7 +222,7 @@ func (c *Compiler) LowerFuncValues() {
 				// The 0 case, which is actually a nil check.
 				nilBlock := llvm.InsertBasicBlock(nextBlock, "func.nil")
 				c.builder.SetInsertPointAtEnd(nilBlock)
-				c.createRuntimeCall("nilpanic", nil, "")
+				c.createRuntimeCall("nilPanic", nil, "")
 				c.builder.CreateUnreachable()
 				sw.AddCase(llvm.ConstInt(c.uintptrType, 0, false), nilBlock)
 

compiler/func.go

@@ -41,7 +41,7 @@ func (c *Compiler) funcImplementation() funcValueImplementation {
 
 // createFuncValue creates a function value from a raw function pointer with no
 // context.
-func (c *Compiler) createFuncValue(funcPtr, context llvm.Value, sig *types.Signature) (llvm.Value, error) {
+func (c *Compiler) createFuncValue(funcPtr, context llvm.Value, sig *types.Signature) llvm.Value {
 	var funcValueScalar llvm.Value
 	switch c.funcImplementation() {
 	case funcValueDoubleword:
@@ -52,7 +52,7 @@ func (c *Compiler) createFuncValue(funcPtr, context llvm.Value, sig *types.Signa
 		funcValueWithSignatureGlobalName := funcPtr.Name() + "$withSignature"
 		funcValueWithSignatureGlobal := c.mod.NamedGlobal(funcValueWithSignatureGlobalName)
 		if funcValueWithSignatureGlobal.IsNil() {
-			funcValueWithSignatureType := c.mod.GetTypeByName("runtime.funcValueWithSignature")
+			funcValueWithSignatureType := c.getLLVMRuntimeType("funcValueWithSignature")
 			funcValueWithSignature := llvm.ConstNamedStruct(funcValueWithSignatureType, []llvm.Value{
 				llvm.ConstPtrToInt(funcPtr, c.uintptrType),
 				sigGlobal,
@@ -66,14 +66,11 @@ func (c *Compiler) createFuncValue(funcPtr, context llvm.Value, sig *types.Signa
 	default:
 		panic("unimplemented func value variant")
 	}
-	funcValueType, err := c.getFuncType(sig)
-	if err != nil {
-		return llvm.Value{}, err
-	}
+	funcValueType := c.getFuncType(sig)
 	funcValue := llvm.Undef(funcValueType)
 	funcValue = c.builder.CreateInsertValue(funcValue, context, 0, "")
 	funcValue = c.builder.CreateInsertValue(funcValue, funcValueScalar, 1, "")
-	return funcValue, nil
+	return funcValue
 }
 
 // getFuncSignature returns a global for identification of a particular function
@@ -106,16 +103,13 @@ func (c *Compiler) extractFuncContext(funcValue llvm.Value) llvm.Value {
 
 // decodeFuncValue extracts the context and the function pointer from this func
 // value. This may be an expensive operation.
-func (c *Compiler) decodeFuncValue(funcValue llvm.Value, sig *types.Signature) (funcPtr, context llvm.Value, err error) {
+func (c *Compiler) decodeFuncValue(funcValue llvm.Value, sig *types.Signature) (funcPtr, context llvm.Value) {
 	context = c.builder.CreateExtractValue(funcValue, 0, "")
 	switch c.funcImplementation() {
 	case funcValueDoubleword:
 		funcPtr = c.builder.CreateExtractValue(funcValue, 1, "")
 	case funcValueSwitch:
-		llvmSig, err := c.getRawFuncType(sig)
-		if err != nil {
-			return llvm.Value{}, llvm.Value{}, err
-		}
+		llvmSig := c.getRawFuncType(sig)
 		sigGlobal := c.getFuncSignature(sig)
 		funcPtr = c.createRuntimeCall("getFuncPtr", []llvm.Value{funcValue, sigGlobal}, "")
 		funcPtr = c.builder.CreateIntToPtr(funcPtr, llvmSig, "")
@@ -126,25 +120,21 @@ func (c *Compiler) decodeFuncValue(funcValue llvm.Value, sig *types.Signature) (
 }
 
 // getFuncType returns the type of a func value given a signature.
-func (c *Compiler) getFuncType(typ *types.Signature) (llvm.Type, error) {
+func (c *Compiler) getFuncType(typ *types.Signature) llvm.Type {
 	switch c.funcImplementation() {
 	case funcValueDoubleword:
-		rawPtr, err := c.getRawFuncType(typ)
-		if err != nil {
-			return llvm.Type{}, err
-		}
-		return c.ctx.StructType([]llvm.Type{c.i8ptrType, rawPtr}, false), nil
+		rawPtr := c.getRawFuncType(typ)
+		return c.ctx.StructType([]llvm.Type{c.i8ptrType, rawPtr}, false)
 	case funcValueSwitch:
-		return c.mod.GetTypeByName("runtime.funcValue"), nil
+		return c.getLLVMRuntimeType("funcValue")
 	default:
 		panic("unimplemented func value variant")
 	}
 }
 
 // getRawFuncType returns a LLVM function pointer type for a given signature.
-func (c *Compiler) getRawFuncType(typ *types.Signature) (llvm.Type, error) {
+func (c *Compiler) getRawFuncType(typ *types.Signature) llvm.Type {
 	// Get the return type.
-	var err error
 	var returnType llvm.Type
 	switch typ.Results().Len() {
 	case 0:
@@ -152,21 +142,14 @@ func (c *Compiler) getRawFuncType(typ *types.Signature) (llvm.Type, error) {
 		returnType = c.ctx.VoidType()
 	case 1:
 		// Just one return value.
-		returnType, err = c.getLLVMType(typ.Results().At(0).Type())
-		if err != nil {
-			return llvm.Type{}, err
-		}
+		returnType = c.getLLVMType(typ.Results().At(0).Type())
 	default:
 		// Multiple return values. Put them together in a struct.
 		// This appears to be the common way to handle multiple return values in
 		// LLVM.
 		members := make([]llvm.Type, typ.Results().Len())
 		for i := 0; i < typ.Results().Len(); i++ {
-			returnType, err := c.getLLVMType(typ.Results().At(i).Type())
-			if err != nil {
-				return llvm.Type{}, err
-			}
-			members[i] = returnType
+			members[i] = c.getLLVMType(typ.Results().At(i).Type())
 		}
 		returnType = c.ctx.StructType(members, false)
 	}
@@ -174,10 +157,7 @@ func (c *Compiler) getRawFuncType(typ *types.Signature) (llvm.Type, error) {
 	// Get the parameter types.
 	var paramTypes []llvm.Type
 	if typ.Recv() != nil {
-		recv, err := c.getLLVMType(typ.Recv().Type())
-		if err != nil {
-			return llvm.Type{}, err
-		}
+		recv := c.getLLVMType(typ.Recv().Type())
 		if recv.StructName() == "runtime._interface" {
 			// This is a call on an interface, not a concrete type.
 			// The receiver is not an interface, but a i8* type.
@@ -186,10 +166,7 @@ func (c *Compiler) getRawFuncType(typ *types.Signature) (llvm.Type, error) {
 		paramTypes = append(paramTypes, c.expandFormalParamType(recv)...)
 	}
 	for i := 0; i < typ.Params().Len(); i++ {
-		subType, err := c.getLLVMType(typ.Params().At(i).Type())
-		if err != nil {
-			return llvm.Type{}, err
-		}
+		subType := c.getLLVMType(typ.Params().At(i).Type())
 		paramTypes = append(paramTypes, c.expandFormalParamType(subType)...)
 	}
 	// All functions take these parameters at the end.
@@ -197,7 +174,7 @@ func (c *Compiler) getRawFuncType(typ *types.Signature) (llvm.Type, error) {
 	paramTypes = append(paramTypes, c.i8ptrType) // parent coroutine
 
 	// Make a func type out of the signature.
-	return llvm.PointerType(llvm.FunctionType(returnType, paramTypes, false), c.funcPtrAddrSpace), nil
+	return llvm.PointerType(llvm.FunctionType(returnType, paramTypes, false), c.funcPtrAddrSpace)
 }
 
 // parseMakeClosure makes a function value (with context) from the given
@@ -209,57 +186,17 @@ func (c *Compiler) parseMakeClosure(frame *Frame, expr *ssa.MakeClosure) (llvm.V
 	f := c.ir.GetFunction(expr.Fn.(*ssa.Function))
 
 	// Collect all bound variables.
-	boundVars := make([]llvm.Value, 0, len(expr.Bindings))
-	boundVarTypes := make([]llvm.Type, 0, len(expr.Bindings))
-	for _, binding := range expr.Bindings {
+	boundVars := make([]llvm.Value, len(expr.Bindings))
+	for i, binding := range expr.Bindings {
 		// The context stores the bound variables.
-		llvmBoundVar, err := c.parseExpr(frame, binding)
-		if err != nil {
-			return llvm.Value{}, err
-		}
-		boundVars = append(boundVars, llvmBoundVar)
-		boundVarTypes = append(boundVarTypes, llvmBoundVar.Type())
-	}
-	contextType := c.ctx.StructType(boundVarTypes, false)
-
-	// Allocate memory for the context.
-	contextAlloc := llvm.Value{}
-	contextHeapAlloc := llvm.Value{}
-	if c.targetData.TypeAllocSize(contextType) <= c.targetData.TypeAllocSize(c.i8ptrType) {
-		// Context fits in a pointer - e.g. when it is a pointer. Store it
-		// directly in the stack after a convert.
-		// Because contextType is a struct and we have to cast it to a *i8,
-		// store it in an alloca first for bitcasting (store+bitcast+load).
-		contextAlloc = c.builder.CreateAlloca(contextType, "")
-	} else {
-		// Context is bigger than a pointer, so allocate it on the heap.
-		size := c.targetData.TypeAllocSize(contextType)
-		sizeValue := llvm.ConstInt(c.uintptrType, size, false)
-		contextHeapAlloc = c.createRuntimeCall("alloc", []llvm.Value{sizeValue}, "")
-		contextAlloc = c.builder.CreateBitCast(contextHeapAlloc, llvm.PointerType(contextType, 0), "")
+		llvmBoundVar := c.getValue(frame, binding)
+		boundVars[i] = llvmBoundVar
 	}
 
-	// Store all bound variables in the alloca or heap pointer.
-	for i, boundVar := range boundVars {
-		indices := []llvm.Value{
-			llvm.ConstInt(c.ctx.Int32Type(), 0, false),
-			llvm.ConstInt(c.ctx.Int32Type(), uint64(i), false),
-		}
-		gep := c.builder.CreateInBoundsGEP(contextAlloc, indices, "")
-		c.builder.CreateStore(boundVar, gep)
-	}
-
-	context := llvm.Value{}
-	if c.targetData.TypeAllocSize(contextType) <= c.targetData.TypeAllocSize(c.i8ptrType) {
-		// Load value (as *i8) from the alloca.
-		contextAlloc = c.builder.CreateBitCast(contextAlloc, llvm.PointerType(c.i8ptrType, 0), "")
-		context = c.builder.CreateLoad(contextAlloc, "")
-	} else {
-		// Get the original heap allocation pointer, which already is an
-		// *i8.
-		context = contextHeapAlloc
-	}
+	// Store the bound variables in a single object, allocating it on the heap
+	// if necessary.
+	context := c.emitPointerPack(boundVars)
 
 	// Create the closure.
-	return c.createFuncValue(f.LLVMFn, context, f.Signature)
+	return c.createFuncValue(f.LLVMFn, context, f.Signature), nil
 }

compiler/gc.go

@@ -0,0 +1,471 @@
+package compiler
+
+// This file provides IR transformations necessary for precise and portable
+// garbage collectors.
+
+import (
+	"go/token"
+	"math/big"
+
+	"golang.org/x/tools/go/ssa"
+	"tinygo.org/x/go-llvm"
+)
+
+// needsStackObjects returns true if the compiler should insert stack objects
+// that can be traced by the garbage collector.
+func (c *Compiler) needsStackObjects() bool {
+	if c.selectGC() != "conservative" {
+		return false
+	}
+	for _, tag := range c.BuildTags {
+		if tag == "cortexm" || tag == "tinygo.riscv" {
+			return false
+		}
+	}
+
+	return true
+}
+
+// trackExpr inserts pointer tracking intrinsics for the GC if the expression is
+// one of the expressions that need this.
+func (c *Compiler) trackExpr(frame *Frame, expr ssa.Value, value llvm.Value) {
+	// There are uses of this expression, Make sure the pointers
+	// are tracked during GC.
+	switch expr := expr.(type) {
+	case *ssa.Alloc, *ssa.MakeChan, *ssa.MakeMap:
+		// These values are always of pointer type in IR.
+		c.trackPointer(value)
+	case *ssa.Call, *ssa.Convert, *ssa.MakeClosure, *ssa.MakeInterface, *ssa.MakeSlice, *ssa.Next:
+		if !value.IsNil() {
+			c.trackValue(value)
+		}
+	case *ssa.Select:
+		if alloca, ok := frame.selectRecvBuf[expr]; ok {
+			if alloca.IsAUndefValue().IsNil() {
+				c.trackPointer(alloca)
+			}
+		}
+	case *ssa.UnOp:
+		switch expr.Op {
+		case token.MUL:
+			// Pointer dereference.
+			c.trackValue(value)
+		case token.ARROW:
+			// Channel receive operator.
+			// It's not necessary to look at commaOk here, because in that
+			// case it's just an aggregate and trackValue will extract the
+			// pointer in there (if there is one).
+			c.trackValue(value)
+		}
+	}
+}
+
+// trackValue locates pointers in a value (possibly an aggregate) and tracks the
+// individual pointers
+func (c *Compiler) trackValue(value llvm.Value) {
+	typ := value.Type()
+	switch typ.TypeKind() {
+	case llvm.PointerTypeKind:
+		c.trackPointer(value)
+	case llvm.StructTypeKind:
+		if !typeHasPointers(typ) {
+			return
+		}
+		numElements := typ.StructElementTypesCount()
+		for i := 0; i < numElements; i++ {
+			subValue := c.builder.CreateExtractValue(value, i, "")
+			c.trackValue(subValue)
+		}
+	case llvm.ArrayTypeKind:
+		if !typeHasPointers(typ) {
+			return
+		}
+		numElements := typ.ArrayLength()
+		for i := 0; i < numElements; i++ {
+			subValue := c.builder.CreateExtractValue(value, i, "")
+			c.trackValue(subValue)
+		}
+	}
+}
+
+// trackPointer creates a call to runtime.trackPointer, bitcasting the poitner
+// first if needed. The input value must be of LLVM pointer type.
+func (c *Compiler) trackPointer(value llvm.Value) {
+	if value.Type() != c.i8ptrType {
+		value = c.builder.CreateBitCast(value, c.i8ptrType, "")
+	}
+	c.createRuntimeCall("trackPointer", []llvm.Value{value}, "")
+}
+
+// typeHasPointers returns whether this type is a pointer or contains pointers.
+// If the type is an aggregate type, it will check whether there is a pointer
+// inside.
+func typeHasPointers(t llvm.Type) bool {
+	switch t.TypeKind() {
+	case llvm.PointerTypeKind:
+		return true
+	case llvm.StructTypeKind:
+		for _, subType := range t.StructElementTypes() {
+			if typeHasPointers(subType) {
+				return true
+			}
+		}
+		return false
+	case llvm.ArrayTypeKind:
+		if typeHasPointers(t.ElementType()) {
+			return true
+		}
+		return false
+	default:
+		return false
+	}
+}
+
+// makeGCStackSlots converts all calls to runtime.trackPointer to explicit
+// stores to stack slots that are scannable by the GC.
+func (c *Compiler) makeGCStackSlots() bool {
+	// Check whether there are allocations at all.
+	alloc := c.mod.NamedFunction("runtime.alloc")
+	if alloc.IsNil() {
+		// Nothing to. Make sure all remaining bits and pieces for stack
+		// chains are neutralized.
+		for _, call := range getUses(c.mod.NamedFunction("runtime.trackPointer")) {
+			call.EraseFromParentAsInstruction()
+		}
+		stackChainStart := c.mod.NamedGlobal("runtime.stackChainStart")
+		if !stackChainStart.IsNil() {
+			stackChainStart.SetInitializer(c.getZeroValue(stackChainStart.Type().ElementType()))
+			stackChainStart.SetGlobalConstant(true)
+		}
+	}
+
+	trackPointer := c.mod.NamedFunction("runtime.trackPointer")
+	if trackPointer.IsNil() || trackPointer.FirstUse().IsNil() {
+		return false // nothing to do
+	}
+
+	// Look at *all* functions to see whether they are free of function pointer
+	// calls.
+	// This takes less than 5ms for ~100kB of WebAssembly but would perhaps be
+	// faster when written in C++ (to avoid the CGo overhead).
+	funcsWithFPCall := map[llvm.Value]struct{}{}
+	n := 0
+	for fn := c.mod.FirstFunction(); !fn.IsNil(); fn = llvm.NextFunction(fn) {
+		n++
+		if _, ok := funcsWithFPCall[fn]; ok {
+			continue // already found
+		}
+		done := false
+		for bb := fn.FirstBasicBlock(); !bb.IsNil() && !done; bb = llvm.NextBasicBlock(bb) {
+			for call := bb.FirstInstruction(); !call.IsNil() && !done; call = llvm.NextInstruction(call) {
+				if call.IsACallInst().IsNil() {
+					continue // only looking at calls
+				}
+				called := call.CalledValue()
+				if !called.IsAFunction().IsNil() {
+					continue // only looking for function pointers
+				}
+				funcsWithFPCall[fn] = struct{}{}
+				markParentFunctions(funcsWithFPCall, fn)
+				done = true
+			}
+		}
+	}
+
+	// Determine which functions need stack objects. Many leaf functions don't
+	// need it: it only causes overhead for them.
+	// Actually, in one test it was only able to eliminate stack object from 12%
+	// of functions that had a call to runtime.trackPointer (8 out of 68
+	// functions), so this optimization is not as big as it may seem.
+	allocatingFunctions := map[llvm.Value]struct{}{} // set of allocating functions
+
+	// Work from runtime.alloc and trace all parents to check which functions do
+	// a heap allocation (and thus which functions do not).
+	markParentFunctions(allocatingFunctions, alloc)
+
+	// Also trace all functions that call a function pointer.
+	for fn := range funcsWithFPCall {
+		// Assume that functions that call a function pointer do a heap
+		// allocation as a conservative guess because the called function might
+		// do a heap allocation.
+		allocatingFunctions[fn] = struct{}{}
+		markParentFunctions(allocatingFunctions, fn)
+	}
+
+	// Collect some variables used below in the loop.
+	stackChainStart := c.mod.NamedGlobal("runtime.stackChainStart")
+	if stackChainStart.IsNil() {
+		panic("stack chain start not found!")
+	}
+	stackChainStartType := stackChainStart.Type().ElementType()
+	stackChainStart.SetInitializer(c.getZeroValue(stackChainStartType))
+
+	// Iterate until runtime.trackPointer has no uses left.
+	for use := trackPointer.FirstUse(); !use.IsNil(); use = trackPointer.FirstUse() {
+		// Pick the first use of runtime.trackPointer.
+		call := use.User()
+		if call.IsACallInst().IsNil() {
+			panic("expected runtime.trackPointer use to be a call")
+		}
+
+		// Pick the parent function.
+		fn := call.InstructionParent().Parent()
+
+		if _, ok := allocatingFunctions[fn]; !ok {
+			// This function nor any of the functions it calls (recursively)
+			// allocate anything from the heap, so it will not trigger a garbage
+			// collection cycle. Thus, it does not need to track local pointer
+			// values.
+			// This is a useful optimization but not as big as you might guess,
+			// as described above (it avoids stack objects for ~12% of
+			// functions).
+			call.EraseFromParentAsInstruction()
+			continue
+		}
+
+		// Find all calls to runtime.trackPointer in this function.
+		var calls []llvm.Value
+		var returns []llvm.Value
+		for bb := fn.FirstBasicBlock(); !bb.IsNil(); bb = llvm.NextBasicBlock(bb) {
+			for inst := bb.FirstInstruction(); !inst.IsNil(); inst = llvm.NextInstruction(inst) {
+				switch inst.InstructionOpcode() {
+				case llvm.Call:
+					if inst.CalledValue() == trackPointer {
+						calls = append(calls, inst)
+					}
+				case llvm.Ret:
+					returns = append(returns, inst)
+				}
+			}
+		}
+
+		// Determine what to do with each call.
+		var allocas, pointers []llvm.Value
+		for _, call := range calls {
+			ptr := call.Operand(0)
+			call.EraseFromParentAsInstruction()
+			if ptr.IsAInstruction().IsNil() {
+				continue
+			}
+
+			// Some trivial optimizations.
+			if ptr.IsAInstruction().IsNil() {
+				continue
+			}
+			switch ptr.InstructionOpcode() {
+			case llvm.PHI, llvm.GetElementPtr:
+				// These values do not create new values: the values already
+				// existed locally in this function so must have been tracked
+				// already.
+				continue
+			case llvm.ExtractValue, llvm.BitCast:
+				// These instructions do not create new values, but their
+				// original value may not be tracked. So keep tracking them for
+				// now.
+				// With more analysis, it should be possible to optimize a
+				// significant chunk of these away.
+			case llvm.Call, llvm.Load, llvm.IntToPtr:
+				// These create new values so must be stored locally. But
+				// perhaps some of these can be fused when they actually refer
+				// to the same value.
+			default:
+				// Ambiguous. These instructions are uncommon, but perhaps could
+				// be optimized if needed.
+			}
+
+			if !ptr.IsAAllocaInst().IsNil() {
+				if typeHasPointers(ptr.Type().ElementType()) {
+					allocas = append(allocas, ptr)
+				}
+			} else {
+				pointers = append(pointers, ptr)
+			}
+		}
+
+		if len(allocas) == 0 && len(pointers) == 0 {
+			// This function does not need to keep track of stack pointers.
+			continue
+		}
+
+		// Determine the type of the required stack slot.
+		fields := []llvm.Type{
+			stackChainStartType, // Pointer to parent frame.
+			c.uintptrType,       // Number of elements in this frame.
+		}
+		for _, alloca := range allocas {
+			fields = append(fields, alloca.Type().ElementType())
+		}
+		for _, ptr := range pointers {
+			fields = append(fields, ptr.Type())
+		}
+		stackObjectType := c.ctx.StructType(fields, false)
+
+		// Create the stack object at the function entry.
+		c.builder.SetInsertPointBefore(fn.EntryBasicBlock().FirstInstruction())
+		stackObject := c.builder.CreateAlloca(stackObjectType, "gc.stackobject")
+		initialStackObject := c.getZeroValue(stackObjectType)
+		numSlots := (c.targetData.TypeAllocSize(stackObjectType) - c.targetData.TypeAllocSize(c.i8ptrType)*2) / uint64(c.targetData.ABITypeAlignment(c.uintptrType))
+		numSlotsValue := llvm.ConstInt(c.uintptrType, numSlots, false)
+		initialStackObject = llvm.ConstInsertValue(initialStackObject, numSlotsValue, []uint32{1})
+		c.builder.CreateStore(initialStackObject, stackObject)
+
+		// Update stack start.
+		parent := c.builder.CreateLoad(stackChainStart, "")
+		gep := c.builder.CreateGEP(stackObject, []llvm.Value{
+			llvm.ConstInt(c.ctx.Int32Type(), 0, false),
+			llvm.ConstInt(c.ctx.Int32Type(), 0, false),
+		}, "")
+		c.builder.CreateStore(parent, gep)
+		stackObjectCast := c.builder.CreateBitCast(stackObject, stackChainStartType, "")
+		c.builder.CreateStore(stackObjectCast, stackChainStart)
+
+		// Replace all independent allocas with GEPs in the stack object.
+		for i, alloca := range allocas {
+			gep := c.builder.CreateGEP(stackObject, []llvm.Value{
+				llvm.ConstInt(c.ctx.Int32Type(), 0, false),
+				llvm.ConstInt(c.ctx.Int32Type(), uint64(2+i), false),
+			}, "")
+			alloca.ReplaceAllUsesWith(gep)
+			alloca.EraseFromParentAsInstruction()
+		}
+
+		// Do a store to the stack object after each new pointer that is created.
+		for i, ptr := range pointers {
+			c.builder.SetInsertPointBefore(llvm.NextInstruction(ptr))
+			gep := c.builder.CreateGEP(stackObject, []llvm.Value{
+				llvm.ConstInt(c.ctx.Int32Type(), 0, false),
+				llvm.ConstInt(c.ctx.Int32Type(), uint64(2+len(allocas)+i), false),
+			}, "")
+			c.builder.CreateStore(ptr, gep)
+		}
+
+		// Make sure this stack object is popped from the linked list of stack
+		// objects at return.
+		for _, ret := range returns {
+			c.builder.SetInsertPointBefore(ret)
+			c.builder.CreateStore(parent, stackChainStart)
+		}
+	}
+
+	return true
+}
+
+func (c *Compiler) addGlobalsBitmap() bool {
+	if c.mod.NamedGlobal("runtime.trackedGlobalsStart").IsNil() {
+		return false // nothing to do: no GC in use
+	}
+
+	var trackedGlobals []llvm.Value
+	var trackedGlobalTypes []llvm.Type
+	for global := c.mod.FirstGlobal(); !global.IsNil(); global = llvm.NextGlobal(global) {
+		if global.IsDeclaration() {
+			continue
+		}
+		typ := global.Type().ElementType()
+		ptrs := c.getPointerBitmap(typ, global.Name())
+		if ptrs.BitLen() == 0 {
+			continue
+		}
+		trackedGlobals = append(trackedGlobals, global)
+		trackedGlobalTypes = append(trackedGlobalTypes, typ)
+	}
+
+	globalsBundleType := c.ctx.StructType(trackedGlobalTypes, false)
+	globalsBundle := llvm.AddGlobal(c.mod, globalsBundleType, "tinygo.trackedGlobals")
+	globalsBundle.SetLinkage(llvm.InternalLinkage)
+	globalsBundle.SetUnnamedAddr(true)
+	initializer := llvm.Undef(globalsBundleType)
+	for i, global := range trackedGlobals {
+		initializer = llvm.ConstInsertValue(initializer, global.Initializer(), []uint32{uint32(i)})
+		gep := llvm.ConstGEP(globalsBundle, []llvm.Value{
+			llvm.ConstInt(c.ctx.Int32Type(), 0, false),
+			llvm.ConstInt(c.ctx.Int32Type(), uint64(i), false),
+		})
+		global.ReplaceAllUsesWith(gep)
+		global.EraseFromParentAsGlobal()
+	}
+	globalsBundle.SetInitializer(initializer)
+
+	trackedGlobalsStart := llvm.ConstPtrToInt(globalsBundle, c.uintptrType)
+	c.mod.NamedGlobal("runtime.trackedGlobalsStart").SetInitializer(trackedGlobalsStart)
+
+	alignment := c.targetData.PrefTypeAlignment(c.i8ptrType)
+	trackedGlobalsLength := llvm.ConstInt(c.uintptrType, c.targetData.TypeAllocSize(globalsBundleType)/uint64(alignment), false)
+	c.mod.NamedGlobal("runtime.trackedGlobalsLength").SetInitializer(trackedGlobalsLength)
+
+	bitmapBytes := c.getPointerBitmap(globalsBundleType, "globals bundle").Bytes()
+	bitmapValues := make([]llvm.Value, len(bitmapBytes))
+	for i, b := range bitmapBytes {
+		bitmapValues[len(bitmapBytes)-i-1] = llvm.ConstInt(c.ctx.Int8Type(), uint64(b), false)
+	}
+	bitmapArray := llvm.ConstArray(llvm.ArrayType(c.ctx.Int8Type(), len(bitmapBytes)), bitmapValues)
+	bitmapNew := llvm.AddGlobal(c.mod, bitmapArray.Type(), "runtime.trackedGlobalsBitmap.tmp")
+	bitmapOld := c.mod.NamedGlobal("runtime.trackedGlobalsBitmap")
+	bitmapOld.ReplaceAllUsesWith(bitmapNew)
+	bitmapNew.SetInitializer(bitmapArray)
+	bitmapNew.SetName("runtime.trackedGlobalsBitmap")
+
+	return true // the IR was changed
+}
+
+func (c *Compiler) getPointerBitmap(typ llvm.Type, name string) *big.Int {
+	alignment := c.targetData.PrefTypeAlignment(c.i8ptrType)
+	switch typ.TypeKind() {
+	case llvm.IntegerTypeKind, llvm.FloatTypeKind, llvm.DoubleTypeKind:
+		return big.NewInt(0)
+	case llvm.PointerTypeKind:
+		return big.NewInt(1)
+	case llvm.StructTypeKind:
+		ptrs := big.NewInt(0)
+		for i, subtyp := range typ.StructElementTypes() {
+			subptrs := c.getPointerBitmap(subtyp, name)
+			if subptrs.BitLen() == 0 {
+				continue
+			}
+			offset := c.targetData.ElementOffset(typ, i)
+			if offset%uint64(alignment) != 0 {
+				panic("precise GC: global contains unaligned pointer: " + name)
+			}
+			subptrs.Lsh(subptrs, uint(offset)/uint(alignment))
+			ptrs.Or(ptrs, subptrs)
+		}
+		return ptrs
+	case llvm.ArrayTypeKind:
+		subtyp := typ.ElementType()
+		subptrs := c.getPointerBitmap(subtyp, name)
+		ptrs := big.NewInt(0)
+		if subptrs.BitLen() == 0 {
+			return ptrs
+		}
+		elementSize := c.targetData.TypeAllocSize(subtyp)
+		for i := 0; i < typ.ArrayLength(); i++ {
+			ptrs.Lsh(ptrs, uint(elementSize)/uint(alignment))
+			ptrs.Or(ptrs, subptrs)
+		}
+		return ptrs
+	default:
+		panic("unknown type kind of global: " + name)
+	}
+}
+
+// markParentFunctions traverses all parent function calls (recursively) and
+// adds them to the set of marked functions. It only considers function calls:
+// any other uses of such a function is ignored.
+func markParentFunctions(marked map[llvm.Value]struct{}, fn llvm.Value) {
+	worklist := []llvm.Value{fn}
+	for len(worklist) != 0 {
+		fn := worklist[len(worklist)-1]
+		worklist = worklist[:len(worklist)-1]
+		for _, use := range getUses(fn) {
+			if use.IsACallInst().IsNil() || use.CalledValue() != fn {
+				// Not the parent function.
+				continue
+			}
+			parent := use.InstructionParent().Parent()
+			if _, ok := marked[parent]; !ok {
+				marked[parent] = struct{}{}
+				worklist = append(worklist, parent)
+			}
+		}
+	}
+}

compiler/goroutine-lowering.go

@@ -10,8 +10,8 @@ package compiler
 //         go foo()
 //         time.Sleep(2 * time.Second)
 //         println("some other operation")
-//         bar()
-//         println("done")
+//         i := bar()
+//         println("done", *i)
 //     }
 //
 //     func foo() {
@@ -21,9 +21,10 @@ package compiler
 //         }
 //     }
 //
-//     func bar() {
+//     func bar() *int {
 //         time.Sleep(time.Second)
 //         println("blocking operation completed)
+//         return new(int)
 //     }
 //
 // It is transformed by the IR generator in compiler.go into the following
@@ -34,8 +35,8 @@ package compiler
 //         fn()
 //         time.Sleep(2 * time.Second)
 //         println("some other operation")
-//         bar() // imagine an 'await' keyword in front of this call
-//         println("done")
+//         i := bar() // imagine an 'await' keyword in front of this call
+//         println("done", *i)
 //     }
 //
 //     func foo() {
@@ -45,9 +46,10 @@ package compiler
 //         }
 //     }
 //
-//     func bar() {
+//     func bar() *int {
 //         time.Sleep(time.Second)
 //         println("blocking operation completed)
+//         return new(int)
 //     }
 //
 // The pass in this file transforms this code even further, to the following
@@ -59,9 +61,11 @@ package compiler
 //         runtime.sleepTask(hdl, 2 * time.Second) // ask the scheduler to re-activate this coroutine at the right time
 //         llvm.suspend(hdl)                       // suspend point
 //         println("some other operation")
+//         var i *int                              // allocate space on the stack for the return value
+//         runtime.setTaskPromisePtr(hdl, &i)      // store return value alloca in our coroutine promise
 //         bar(hdl)                                // await, pass a continuation (hdl) to bar
 //         llvm.suspend(hdl)                       // suspend point, wait for the callee to re-activate
-//         println("done")
+//         println("done", *i)
 //         runtime.activateTask(parent)            // re-activate the parent (nop, there is no parent)
 //     }
 //
@@ -142,10 +146,9 @@ func (c *Compiler) LowerGoroutines() error {
 	realMain.SetLinkage(llvm.InternalLinkage)
 	c.mod.NamedFunction("runtime.alloc").SetLinkage(llvm.InternalLinkage)
 	c.mod.NamedFunction("runtime.free").SetLinkage(llvm.InternalLinkage)
-	c.mod.NamedFunction("runtime.chanSend").SetLinkage(llvm.InternalLinkage)
-	c.mod.NamedFunction("runtime.chanRecv").SetLinkage(llvm.InternalLinkage)
 	c.mod.NamedFunction("runtime.sleepTask").SetLinkage(llvm.InternalLinkage)
-	c.mod.NamedFunction("runtime.activateTask").SetLinkage(llvm.InternalLinkage)
+	c.mod.NamedFunction("runtime.setTaskPromisePtr").SetLinkage(llvm.InternalLinkage)
+	c.mod.NamedFunction("runtime.getTaskPromisePtr").SetLinkage(llvm.InternalLinkage)
 	c.mod.NamedFunction("runtime.scheduler").SetLinkage(llvm.InternalLinkage)
 
 	return nil
@@ -174,13 +177,13 @@ func (c *Compiler) markAsyncFunctions() (needsScheduler bool, err error) {
 	if !deadlockStub.IsNil() {
 		worklist = append(worklist, deadlockStub)
 	}
-	chanSendStub := c.mod.NamedFunction("runtime.chanSendStub")
-	if !chanSendStub.IsNil() {
-		worklist = append(worklist, chanSendStub)
+	chanSend := c.mod.NamedFunction("runtime.chanSend")
+	if !chanSend.IsNil() {
+		worklist = append(worklist, chanSend)
 	}
-	chanRecvStub := c.mod.NamedFunction("runtime.chanRecvStub")
-	if !chanRecvStub.IsNil() {
-		worklist = append(worklist, chanRecvStub)
+	chanRecv := c.mod.NamedFunction("runtime.chanRecv")
+	if !chanRecv.IsNil() {
+		worklist = append(worklist, chanRecv)
 	}
 
 	if len(worklist) == 0 {
@@ -278,9 +281,6 @@ func (c *Compiler) markAsyncFunctions() (needsScheduler bool, err error) {
 	coroBeginType := llvm.FunctionType(c.i8ptrType, []llvm.Type{c.ctx.TokenType(), c.i8ptrType}, false)
 	coroBeginFunc := llvm.AddFunction(c.mod, "llvm.coro.begin", coroBeginType)
 
-	coroPromiseType := llvm.FunctionType(c.i8ptrType, []llvm.Type{c.i8ptrType, c.ctx.Int32Type(), c.ctx.Int1Type()}, false)
-	coroPromiseFunc := llvm.AddFunction(c.mod, "llvm.coro.promise", coroPromiseType)
-
 	coroSuspendType := llvm.FunctionType(c.ctx.Int8Type(), []llvm.Type{c.ctx.TokenType(), c.ctx.Int1Type()}, false)
 	coroSuspendFunc := llvm.AddFunction(c.mod, "llvm.coro.suspend", coroSuspendType)
 
@@ -292,7 +292,7 @@ func (c *Compiler) markAsyncFunctions() (needsScheduler bool, err error) {
 
 	// Transform all async functions into coroutines.
 	for _, f := range asyncList {
-		if f == sleep || f == deadlockStub || f == chanSendStub || f == chanRecvStub {
+		if f == sleep || f == deadlockStub || f == chanSend || f == chanRecv {
 			continue
 		}
 
@@ -309,7 +309,7 @@ func (c *Compiler) markAsyncFunctions() (needsScheduler bool, err error) {
 			for inst := bb.FirstInstruction(); !inst.IsNil(); inst = llvm.NextInstruction(inst) {
 				if !inst.IsACallInst().IsNil() {
 					callee := inst.CalledValue()
-					if _, ok := asyncFuncs[callee]; !ok || callee == sleep || callee == deadlockStub || callee == chanSendStub || callee == chanRecvStub {
+					if _, ok := asyncFuncs[callee]; !ok || callee == sleep || callee == deadlockStub || callee == chanSend || callee == chanRecv {
 						continue
 					}
 					asyncCalls = append(asyncCalls, inst)
@@ -321,7 +321,7 @@ func (c *Compiler) markAsyncFunctions() (needsScheduler bool, err error) {
 
 		// Coroutine setup.
 		c.builder.SetInsertPointBefore(f.EntryBasicBlock().FirstInstruction())
-		taskState := c.builder.CreateAlloca(c.mod.GetTypeByName("runtime.taskState"), "task.state")
+		taskState := c.builder.CreateAlloca(c.getLLVMRuntimeType("taskState"), "task.state")
 		stateI8 := c.builder.CreateBitCast(taskState, c.i8ptrType, "task.state.i8")
 		id := c.builder.CreateCall(coroIdFunc, []llvm.Value{
 			llvm.ConstInt(c.ctx.Int32Type(), 0, false),
@@ -336,6 +336,9 @@ func (c *Compiler) markAsyncFunctions() (needsScheduler bool, err error) {
 			size = c.builder.CreateZExt(size, c.uintptrType, "task.size.uintptr")
 		}
 		data := c.createRuntimeCall("alloc", []llvm.Value{size}, "task.data")
+		if c.needsStackObjects() {
+			c.trackPointer(data)
+		}
 		frame.taskHandle = c.builder.CreateCall(coroBeginFunc, []llvm.Value{id, data}, "task.handle")
 
 		// Modify async calls so this function suspends right after the child
@@ -347,10 +350,18 @@ func (c *Compiler) markAsyncFunctions() (needsScheduler bool, err error) {
 			// Split this basic block.
 			await := c.splitBasicBlock(inst, llvm.NextBasicBlock(c.builder.GetInsertBlock()), "task.await")
 
-			// Set task state to TASK_STATE_CALL.
-			c.builder.SetInsertPointAtEnd(inst.InstructionParent())
+			// Allocate space for the return value.
+			var retvalAlloca llvm.Value
+			if inst.Type().TypeKind() != llvm.VoidTypeKind {
+				c.builder.SetInsertPointBefore(inst.InstructionParent().Parent().EntryBasicBlock().FirstInstruction())
+				retvalAlloca = c.builder.CreateAlloca(inst.Type(), "coro.retvalAlloca")
+				c.builder.SetInsertPointBefore(inst)
+				data := c.builder.CreateBitCast(retvalAlloca, c.i8ptrType, "")
+				c.createRuntimeCall("setTaskPromisePtr", []llvm.Value{frame.taskHandle, data}, "")
+			}
 
 			// Suspend.
+			c.builder.SetInsertPointAtEnd(inst.InstructionParent())
 			continuePoint := c.builder.CreateCall(coroSuspendFunc, []llvm.Value{
 				llvm.ConstNull(c.ctx.TokenType()),
 				llvm.ConstInt(c.ctx.Int1Type(), 0, false),
@@ -358,44 +369,76 @@ func (c *Compiler) markAsyncFunctions() (needsScheduler bool, err error) {
 			sw := c.builder.CreateSwitch(continuePoint, frame.suspendBlock, 2)
 			sw.AddCase(llvm.ConstInt(c.ctx.Int8Type(), 0, false), await)
 			sw.AddCase(llvm.ConstInt(c.ctx.Int8Type(), 1, false), frame.cleanupBlock)
+
+			if inst.Type().TypeKind() != llvm.VoidTypeKind {
+				// Load the return value from the alloca. The callee has
+				// written the return value to it.
+				c.builder.SetInsertPointBefore(await.FirstInstruction())
+				retval := c.builder.CreateLoad(retvalAlloca, "coro.retval")
+				inst.ReplaceAllUsesWith(retval)
+			}
 		}
 
 		// Replace return instructions with suspend points that should
 		// reactivate the parent coroutine.
 		for _, inst := range returns {
-			if inst.OperandsCount() == 0 {
-				// These properties were added by the functionattrs pass.
-				// Remove them, because now we start using the parameter.
-				// https://llvm.org/docs/Passes.html#functionattrs-deduce-function-attributes
-				for _, kind := range []string{"nocapture", "readnone"} {
-					kindID := llvm.AttributeKindID(kind)
-					f.RemoveEnumAttributeAtIndex(f.ParamsCount(), kindID)
-				}
-
-				// Reactivate the parent coroutine. This adds it back to
-				// the run queue, so it is started again by the
-				// scheduler when possible (possibly right after the
-				// following suspend).
-				c.builder.SetInsertPointBefore(inst)
-
-				parentHandle := f.LastParam()
-				c.createRuntimeCall("activateTask", []llvm.Value{parentHandle}, "")
-
-				// Suspend this coroutine.
-				// It would look like this is unnecessary, but if this
-				// suspend point is left out, it leads to undefined
-				// behavior somehow (with the unreachable instruction).
-				continuePoint := c.builder.CreateCall(coroSuspendFunc, []llvm.Value{
-					llvm.ConstNull(c.ctx.TokenType()),
-					llvm.ConstInt(c.ctx.Int1Type(), 1, false),
-				}, "ret")
-				sw := c.builder.CreateSwitch(continuePoint, frame.suspendBlock, 2)
-				sw.AddCase(llvm.ConstInt(c.ctx.Int8Type(), 0, false), frame.unreachableBlock)
-				sw.AddCase(llvm.ConstInt(c.ctx.Int8Type(), 1, false), frame.cleanupBlock)
-				inst.EraseFromParentAsInstruction()
-			} else {
-				panic("todo: return value from coroutine")
+			// These properties were added by the functionattrs pass. Remove
+			// them, because now we start using the parameter.
+			// https://llvm.org/docs/Passes.html#functionattrs-deduce-function-attributes
+			for _, kind := range []string{"nocapture", "readnone"} {
+				kindID := llvm.AttributeKindID(kind)
+				f.RemoveEnumAttributeAtIndex(f.ParamsCount(), kindID)
 			}
+
+			c.builder.SetInsertPointBefore(inst)
+
+			var parentHandle llvm.Value
+			if f.Linkage() == llvm.ExternalLinkage {
+				// Exported function.
+				// Note that getTaskPromisePtr will panic if it is called with
+				// a nil pointer, so blocking exported functions that try to
+				// return anything will not work.
+				parentHandle = llvm.ConstPointerNull(c.i8ptrType)
+			} else {
+				parentHandle = f.LastParam()
+				if parentHandle.IsNil() || parentHandle.Name() != "parentHandle" {
+					// sanity check
+					panic("trying to make exported function async")
+				}
+			}
+
+			// Store return values.
+			switch inst.OperandsCount() {
+			case 0:
+				// Nothing to return.
+			case 1:
+				// Return this value by writing to the pointer stored in the
+				// parent handle. The parent coroutine has made an alloca that
+				// we can write to to store our return value.
+				returnValuePtr := c.createRuntimeCall("getTaskPromisePtr", []llvm.Value{parentHandle}, "coro.parentData")
+				alloca := c.builder.CreateBitCast(returnValuePtr, llvm.PointerType(inst.Operand(0).Type(), 0), "coro.parentAlloca")
+				c.builder.CreateStore(inst.Operand(0), alloca)
+			default:
+				panic("unreachable")
+			}
+
+			// Reactivate the parent coroutine. This adds it back to the run
+			// queue, so it is started again by the scheduler when possible
+			// (possibly right after the following suspend).
+			c.createRuntimeCall("activateTask", []llvm.Value{parentHandle}, "")
+
+			// Suspend this coroutine.
+			// It would look like this is unnecessary, but if this
+			// suspend point is left out, it leads to undefined
+			// behavior somehow (with the unreachable instruction).
+			continuePoint := c.builder.CreateCall(coroSuspendFunc, []llvm.Value{
+				llvm.ConstNull(c.ctx.TokenType()),
+				llvm.ConstInt(c.ctx.Int1Type(), 0, false),
+			}, "ret")
+			sw := c.builder.CreateSwitch(continuePoint, frame.suspendBlock, 2)
+			sw.AddCase(llvm.ConstInt(c.ctx.Int8Type(), 0, false), frame.unreachableBlock)
+			sw.AddCase(llvm.ConstInt(c.ctx.Int8Type(), 1, false), frame.cleanupBlock)
+			inst.EraseFromParentAsInstruction()
 		}
 
 		// Coroutine cleanup. Free resources associated with this coroutine.
@@ -420,6 +463,14 @@ func (c *Compiler) markAsyncFunctions() (needsScheduler bool, err error) {
 		c.builder.CreateUnreachable()
 	}
 
+	// Replace calls to runtime.getCoroutineCall with the coroutine of this
+	// frame.
+	for _, getCoroutineCall := range getUses(c.mod.NamedFunction("runtime.getCoroutine")) {
+		frame := asyncFuncs[getCoroutineCall.InstructionParent().Parent()]
+		getCoroutineCall.ReplaceAllUsesWith(frame.taskHandle)
+		getCoroutineCall.EraseFromParentAsInstruction()
+	}
+
 	// Transform calls to time.Sleep() into coroutine suspend points.
 	for _, sleepCall := range getUses(sleep) {
 		// sleepCall must be a call instruction.
@@ -453,7 +504,7 @@ func (c *Compiler) markAsyncFunctions() (needsScheduler bool, err error) {
 		c.builder.SetInsertPointBefore(deadlockCall)
 		continuePoint := c.builder.CreateCall(coroSuspendFunc, []llvm.Value{
 			llvm.ConstNull(c.ctx.TokenType()),
-			llvm.ConstInt(c.ctx.Int1Type(), 1, false), // final suspend
+			llvm.ConstInt(c.ctx.Int1Type(), 0, false),
 		}, "")
 		c.splitBasicBlock(deadlockCall, llvm.NextBasicBlock(c.builder.GetInsertBlock()), "task.wakeup.dead")
 		c.builder.SetInsertPointBefore(deadlockCall)
@@ -463,37 +514,11 @@ func (c *Compiler) markAsyncFunctions() (needsScheduler bool, err error) {
 		deadlockCall.EraseFromParentAsInstruction()
 	}
 
-	// Transform calls to runtime.chanSendStub into channel send operations.
-	for _, sendOp := range getUses(chanSendStub) {
+	// Transform calls to runtime.chanSend into channel send operations.
+	for _, sendOp := range getUses(chanSend) {
 		// sendOp must be a call instruction.
 		frame := asyncFuncs[sendOp.InstructionParent().Parent()]
 
-		// Send the value over the channel, or block.
-		sendOp.SetOperand(0, frame.taskHandle)
-		sendOp.SetOperand(sendOp.OperandsCount()-1, c.mod.NamedFunction("runtime.chanSend"))
-
-		// Use taskState.data to store the value to send:
-		//     *(*valueType)(&coroutine.promise().data) = valueToSend
-		//     runtime.chanSend(coroutine, ch)
-		bitcast := sendOp.Operand(2)
-		valueAlloca := bitcast.Operand(0)
-		c.builder.SetInsertPointBefore(valueAlloca)
-		promiseType := c.mod.GetTypeByName("runtime.taskState")
-		promiseRaw := c.builder.CreateCall(coroPromiseFunc, []llvm.Value{
-			frame.taskHandle,
-			llvm.ConstInt(c.ctx.Int32Type(), uint64(c.targetData.PrefTypeAlignment(promiseType)), false),
-			llvm.ConstInt(c.ctx.Int1Type(), 0, false),
-		}, "task.promise.raw")
-		promise := c.builder.CreateBitCast(promiseRaw, llvm.PointerType(promiseType, 0), "task.promise")
-		dataPtr := c.builder.CreateGEP(promise, []llvm.Value{
-			llvm.ConstInt(c.ctx.Int32Type(), 0, false),
-			llvm.ConstInt(c.ctx.Int32Type(), 2, false),
-		}, "task.promise.data")
-		sendOp.SetOperand(2, llvm.Undef(c.i8ptrType))
-		valueAlloca.ReplaceAllUsesWith(c.builder.CreateBitCast(dataPtr, valueAlloca.Type(), ""))
-		bitcast.EraseFromParentAsInstruction()
-		valueAlloca.EraseFromParentAsInstruction()
-
 		// Yield to scheduler.
 		c.builder.SetInsertPointBefore(llvm.NextInstruction(sendOp))
 		continuePoint := c.builder.CreateCall(coroSuspendFunc, []llvm.Value{
@@ -506,21 +531,11 @@ func (c *Compiler) markAsyncFunctions() (needsScheduler bool, err error) {
 		sw.AddCase(llvm.ConstInt(c.ctx.Int8Type(), 1, false), frame.cleanupBlock)
 	}
 
-	// Transform calls to runtime.chanRecvStub into channel receive operations.
-	for _, recvOp := range getUses(chanRecvStub) {
+	// Transform calls to runtime.chanRecv into channel receive operations.
+	for _, recvOp := range getUses(chanRecv) {
 		// recvOp must be a call instruction.
 		frame := asyncFuncs[recvOp.InstructionParent().Parent()]
 
-		bitcast := recvOp.Operand(2)
-		commaOk := recvOp.Operand(3)
-		valueAlloca := bitcast.Operand(0)
-
-		// Receive the value over the channel, or block.
-		recvOp.SetOperand(0, frame.taskHandle)
-		recvOp.SetOperand(recvOp.OperandsCount()-1, c.mod.NamedFunction("runtime.chanRecv"))
-		recvOp.SetOperand(2, llvm.Undef(c.i8ptrType))
-		bitcast.EraseFromParentAsInstruction()
-
 		// Yield to scheduler.
 		c.builder.SetInsertPointBefore(llvm.NextInstruction(recvOp))
 		continuePoint := c.builder.CreateCall(coroSuspendFunc, []llvm.Value{
@@ -532,32 +547,6 @@ func (c *Compiler) markAsyncFunctions() (needsScheduler bool, err error) {
 		c.builder.SetInsertPointAtEnd(recvOp.InstructionParent())
 		sw.AddCase(llvm.ConstInt(c.ctx.Int8Type(), 0, false), wakeup)
 		sw.AddCase(llvm.ConstInt(c.ctx.Int8Type(), 1, false), frame.cleanupBlock)
-
-		// The value to receive is stored in taskState.data:
-		//     runtime.chanRecv(coroutine, ch)
-		//     promise := coroutine.promise()
-		//     valueReceived := *(*valueType)(&promise.data)
-		//     ok := promise.commaOk
-		c.builder.SetInsertPointBefore(wakeup.FirstInstruction())
-		promiseType := c.mod.GetTypeByName("runtime.taskState")
-		promiseRaw := c.builder.CreateCall(coroPromiseFunc, []llvm.Value{
-			frame.taskHandle,
-			llvm.ConstInt(c.ctx.Int32Type(), uint64(c.targetData.PrefTypeAlignment(promiseType)), false),
-			llvm.ConstInt(c.ctx.Int1Type(), 0, false),
-		}, "task.promise.raw")
-		promise := c.builder.CreateBitCast(promiseRaw, llvm.PointerType(promiseType, 0), "task.promise")
-		dataPtr := c.builder.CreateGEP(promise, []llvm.Value{
-			llvm.ConstInt(c.ctx.Int32Type(), 0, false),
-			llvm.ConstInt(c.ctx.Int32Type(), 2, false),
-		}, "task.promise.data")
-		valueAlloca.ReplaceAllUsesWith(c.builder.CreateBitCast(dataPtr, valueAlloca.Type(), ""))
-		valueAlloca.EraseFromParentAsInstruction()
-		commaOkPtr := c.builder.CreateGEP(promise, []llvm.Value{
-			llvm.ConstInt(c.ctx.Int32Type(), 0, false),
-			llvm.ConstInt(c.ctx.Int32Type(), 1, false),
-		}, "task.promise.comma-ok")
-		commaOk.ReplaceAllUsesWith(commaOkPtr)
-		recvOp.SetOperand(3, llvm.Undef(commaOk.Type()))
 	}
 
 	return true, c.lowerMakeGoroutineCalls()

compiler/inlineasm.go

@@ -18,10 +18,19 @@ import (
 //     func ReadRegister(name string) uintptr
 //
 // The register name must be a constant, for example "sp".
-func (c *Compiler) emitReadRegister(args []ssa.Value) (llvm.Value, error) {
+func (c *Compiler) emitReadRegister(name string, args []ssa.Value) (llvm.Value, error) {
 	fnType := llvm.FunctionType(c.uintptrType, []llvm.Type{}, false)
 	regname := constant.StringVal(args[0].(*ssa.Const).Value)
-	target := llvm.InlineAsm(fnType, "mov $0, "+regname, "=r", false, false, 0)
+	var asm string
+	switch name {
+	case "device/arm.ReadRegister":
+		asm = "mov $0, " + regname
+	case "device/riscv.ReadRegister":
+		asm = "mv $0, " + regname
+	default:
+		panic("unknown architecture")
+	}
+	target := llvm.InlineAsm(fnType, asm, "=r", false, false, 0)
 	return c.builder.CreateCall(target, nil, ""), nil
 }
 
@@ -68,11 +77,7 @@ func (c *Compiler) emitAsmFull(frame *Frame, instr *ssa.CallCommon) (llvm.Value,
 			}
 			key := constant.StringVal(r.Key.(*ssa.Const).Value)
 			//println("value:", r.Value.(*ssa.MakeInterface).X.String())
-			value, err := c.parseExpr(frame, r.Value.(*ssa.MakeInterface).X)
-			if err != nil {
-				return llvm.Value{}, err
-			}
-			registers[key] = value
+			registers[key] = c.getValue(frame, r.Value.(*ssa.MakeInterface).X)
 		case *ssa.Call:
 			if r.Common() == instr {
 				break
@@ -145,10 +150,7 @@ func (c *Compiler) emitSVCall(frame *Frame, args []ssa.Value) (llvm.Value, error
 		} else {
 			constraints += ",{r" + strconv.Itoa(i) + "}"
 		}
-		llvmValue, err := c.parseExpr(frame, arg)
-		if err != nil {
-			return llvm.Value{}, err
-		}
+		llvmValue := c.getValue(frame, arg)
 		llvmArgs = append(llvmArgs, llvmValue)
 		argTypes = append(argTypes, llvmValue.Type())
 	}

compiler/interface-lowering.go

@@ -163,8 +163,8 @@ func (c *Compiler) LowerInterfaces() {
 // run runs the pass itself.
 func (p *lowerInterfacesPass) run() {
 	// Collect all type codes.
-	typecodeIDPtr := llvm.PointerType(p.mod.GetTypeByName("runtime.typecodeID"), 0)
-	typeInInterfacePtr := llvm.PointerType(p.mod.GetTypeByName("runtime.typeInInterface"), 0)
+	typecodeIDPtr := llvm.PointerType(p.getLLVMRuntimeType("typecodeID"), 0)
+	typeInInterfacePtr := llvm.PointerType(p.getLLVMRuntimeType("typeInInterface"), 0)
 	var typesInInterfaces []llvm.Value
 	for global := p.mod.FirstGlobal(); !global.IsNil(); global = llvm.NextGlobal(global) {
 		switch global.Type() {
@@ -527,11 +527,40 @@ func (p *lowerInterfacesPass) replaceInvokeWithCall(use llvm.Value, typ *typeInf
 	}
 	inttoptr := inttoptrs[0]
 	function := typ.getMethod(signature).function
-	if inttoptr.Type() != function.Type() {
-		p.builder.SetInsertPointBefore(use)
-		function = p.builder.CreateBitCast(function, inttoptr.Type(), "")
+	if inttoptr.Type() == function.Type() {
+		// Easy case: the types are the same. Simply replace the inttoptr
+		// result (which is directly called) with the actual function.
+		inttoptr.ReplaceAllUsesWith(function)
+	} else {
+		// Harder case: the type is not actually the same. Go through each call
+		// (of which there should be only one), extract the receiver params for
+		// this call and replace the call with a direct call to the target
+		// function.
+		for _, call := range getUses(inttoptr) {
+			if call.IsACallInst().IsNil() || call.CalledValue() != inttoptr {
+				panic("expected the inttoptr to be called as a method, this is not a method call")
+			}
+			operands := make([]llvm.Value, call.OperandsCount()-1)
+			for i := range operands {
+				operands[i] = call.Operand(i)
+			}
+			paramTypes := function.Type().ElementType().ParamTypes()
+			receiverParamTypes := paramTypes[:len(paramTypes)-(len(operands)-1)]
+			methodParamTypes := paramTypes[len(paramTypes)-(len(operands)-1):]
+			for i, methodParamType := range methodParamTypes {
+				if methodParamType != operands[i+1].Type() {
+					panic("expected method call param type and function param type to be the same")
+				}
+			}
+			p.builder.SetInsertPointBefore(call)
+			receiverParams := p.emitPointerUnpack(operands[0], receiverParamTypes)
+			result := p.builder.CreateCall(function, append(receiverParams, operands[1:]...), "")
+			if result.Type().TypeKind() != llvm.VoidTypeKind {
+				call.ReplaceAllUsesWith(result)
+			}
+			call.EraseFromParentAsInstruction()
+		}
 	}
-	inttoptr.ReplaceAllUsesWith(function)
 	inttoptr.EraseFromParentAsInstruction()
 	use.EraseFromParentAsInstruction()
 }

compiler/interface.go

@@ -22,56 +22,23 @@ import (
 // value field.
 //
 // An interface value is a {typecode, value} tuple, or {i16, i8*} to be exact.
-func (c *Compiler) parseMakeInterface(val llvm.Value, typ types.Type, pos token.Pos) (llvm.Value, error) {
-	var itfValue llvm.Value
-	size := c.targetData.TypeAllocSize(val.Type())
-	if size > c.targetData.TypeAllocSize(c.i8ptrType) {
-		// Allocate on the heap and put a pointer in the interface.
-		// TODO: escape analysis.
-		sizeValue := llvm.ConstInt(c.uintptrType, size, false)
-		alloc := c.createRuntimeCall("alloc", []llvm.Value{sizeValue}, "makeinterface.alloc")
-		itfValueCast := c.builder.CreateBitCast(alloc, llvm.PointerType(val.Type(), 0), "makeinterface.cast.value")
-		c.builder.CreateStore(val, itfValueCast)
-		itfValue = c.builder.CreateBitCast(itfValueCast, c.i8ptrType, "makeinterface.cast.i8ptr")
-	} else if size == 0 {
-		itfValue = llvm.ConstPointerNull(c.i8ptrType)
-	} else {
-		// Directly place the value in the interface.
-		switch val.Type().TypeKind() {
-		case llvm.IntegerTypeKind:
-			itfValue = c.builder.CreateIntToPtr(val, c.i8ptrType, "makeinterface.cast.int")
-		case llvm.PointerTypeKind:
-			itfValue = c.builder.CreateBitCast(val, c.i8ptrType, "makeinterface.cast.ptr")
-		case llvm.StructTypeKind, llvm.FloatTypeKind, llvm.DoubleTypeKind:
-			// A bitcast would be useful here, but bitcast doesn't allow
-			// aggregate types. So we'll bitcast it using an alloca.
-			// Hopefully this will get optimized away.
-			mem := c.builder.CreateAlloca(c.i8ptrType, "makeinterface.cast.struct")
-			memStructPtr := c.builder.CreateBitCast(mem, llvm.PointerType(val.Type(), 0), "makeinterface.cast.struct.cast")
-			c.builder.CreateStore(val, memStructPtr)
-			itfValue = c.builder.CreateLoad(mem, "makeinterface.cast.load")
-		default:
-			return llvm.Value{}, c.makeError(pos, "todo: makeinterface: cast small type to i8*")
-		}
-	}
+func (c *Compiler) parseMakeInterface(val llvm.Value, typ types.Type, pos token.Pos) llvm.Value {
+	itfValue := c.emitPointerPack([]llvm.Value{val})
 	itfTypeCodeGlobal := c.getTypeCode(typ)
-	itfMethodSetGlobal, err := c.getTypeMethodSet(typ)
-	if err != nil {
-		return llvm.Value{}, nil
-	}
+	itfMethodSetGlobal := c.getTypeMethodSet(typ)
 	itfConcreteTypeGlobal := c.mod.NamedGlobal("typeInInterface:" + itfTypeCodeGlobal.Name())
 	if itfConcreteTypeGlobal.IsNil() {
-		typeInInterface := c.mod.GetTypeByName("runtime.typeInInterface")
+		typeInInterface := c.getLLVMRuntimeType("typeInInterface")
 		itfConcreteTypeGlobal = llvm.AddGlobal(c.mod, typeInInterface, "typeInInterface:"+itfTypeCodeGlobal.Name())
 		itfConcreteTypeGlobal.SetInitializer(llvm.ConstNamedStruct(typeInInterface, []llvm.Value{itfTypeCodeGlobal, itfMethodSetGlobal}))
 		itfConcreteTypeGlobal.SetGlobalConstant(true)
 		itfConcreteTypeGlobal.SetLinkage(llvm.PrivateLinkage)
 	}
 	itfTypeCode := c.builder.CreatePtrToInt(itfConcreteTypeGlobal, c.uintptrType, "")
-	itf := llvm.Undef(c.mod.GetTypeByName("runtime._interface"))
+	itf := llvm.Undef(c.getLLVMRuntimeType("_interface"))
 	itf = c.builder.CreateInsertValue(itf, itfTypeCode, 0, "")
 	itf = c.builder.CreateInsertValue(itf, itfValue, 1, "")
-	return itf, nil
+	return itf
 }
 
 // getTypeCode returns a reference to a type code.
@@ -81,7 +48,7 @@ func (c *Compiler) getTypeCode(typ types.Type) llvm.Value {
 	globalName := "type:" + getTypeCodeName(typ)
 	global := c.mod.NamedGlobal(globalName)
 	if global.IsNil() {
-		global = llvm.AddGlobal(c.mod, c.mod.GetTypeByName("runtime.typecodeID"), globalName)
+		global = llvm.AddGlobal(c.mod, c.getLLVMRuntimeType("typecodeID"), globalName)
 		global.SetGlobalConstant(true)
 	}
 	return global
@@ -185,22 +152,22 @@ func getTypeCodeName(t types.Type) string {
 
 // getTypeMethodSet returns a reference (GEP) to a global method set. This
 // method set should be unreferenced after the interface lowering pass.
-func (c *Compiler) getTypeMethodSet(typ types.Type) (llvm.Value, error) {
+func (c *Compiler) getTypeMethodSet(typ types.Type) llvm.Value {
 	global := c.mod.NamedGlobal(typ.String() + "$methodset")
 	zero := llvm.ConstInt(c.ctx.Int32Type(), 0, false)
 	if !global.IsNil() {
 		// the method set already exists
-		return llvm.ConstGEP(global, []llvm.Value{zero, zero}), nil
+		return llvm.ConstGEP(global, []llvm.Value{zero, zero})
 	}
 
 	ms := c.ir.Program.MethodSets.MethodSet(typ)
 	if ms.Len() == 0 {
 		// no methods, so can leave that one out
-		return llvm.ConstPointerNull(llvm.PointerType(c.mod.GetTypeByName("runtime.interfaceMethodInfo"), 0)), nil
+		return llvm.ConstPointerNull(llvm.PointerType(c.getLLVMRuntimeType("interfaceMethodInfo"), 0))
 	}
 
 	methods := make([]llvm.Value, ms.Len())
-	interfaceMethodInfoType := c.mod.GetTypeByName("runtime.interfaceMethodInfo")
+	interfaceMethodInfoType := c.getLLVMRuntimeType("interfaceMethodInfo")
 	for i := 0; i < ms.Len(); i++ {
 		method := ms.At(i)
 		signatureGlobal := c.getMethodSignature(method.Obj().(*types.Func))
@@ -209,10 +176,7 @@ func (c *Compiler) getTypeMethodSet(typ types.Type) (llvm.Value, error) {
 			// compiler error, so panic
 			panic("cannot find function: " + f.LinkName())
 		}
-		fn, err := c.getInterfaceInvokeWrapper(f)
-		if err != nil {
-			return llvm.Value{}, err
-		}
+		fn := c.getInterfaceInvokeWrapper(f)
 		methodInfo := llvm.ConstNamedStruct(interfaceMethodInfoType, []llvm.Value{
 			signatureGlobal,
 			llvm.ConstPtrToInt(fn, c.uintptrType),
@@ -225,7 +189,7 @@ func (c *Compiler) getTypeMethodSet(typ types.Type) (llvm.Value, error) {
 	global.SetInitializer(value)
 	global.SetGlobalConstant(true)
 	global.SetLinkage(llvm.PrivateLinkage)
-	return llvm.ConstGEP(global, []llvm.Value{zero, zero}), nil
+	return llvm.ConstGEP(global, []llvm.Value{zero, zero})
 }
 
 // getInterfaceMethodSet returns a global variable with the method set of the
@@ -274,19 +238,9 @@ func (c *Compiler) getMethodSignature(method *types.Func) llvm.Value {
 //
 // Type asserts on concrete types are trivial: just compare type numbers. Type
 // asserts on interfaces are more difficult, see the comments in the function.
-func (c *Compiler) parseTypeAssert(frame *Frame, expr *ssa.TypeAssert) (llvm.Value, error) {
-	itf, err := c.parseExpr(frame, expr.X)
-	if err != nil {
-		return llvm.Value{}, err
-	}
-	assertedType, err := c.getLLVMType(expr.AssertedType)
-	if err != nil {
-		return llvm.Value{}, err
-	}
-	valueNil, err := c.getZeroValue(assertedType)
-	if err != nil {
-		return llvm.Value{}, err
-	}
+func (c *Compiler) parseTypeAssert(frame *Frame, expr *ssa.TypeAssert) llvm.Value {
+	itf := c.getValue(frame, expr.X)
+	assertedType := c.getLLVMType(expr.AssertedType)
 
 	actualTypeNum := c.builder.CreateExtractValue(itf, 0, "interface.type")
 	commaOk := llvm.Value{}
@@ -339,67 +293,35 @@ func (c *Compiler) parseTypeAssert(frame *Frame, expr *ssa.TypeAssert) (llvm.Val
 		// Type assert on concrete type. Extract the underlying type from
 		// the interface (but only after checking it matches).
 		valuePtr := c.builder.CreateExtractValue(itf, 1, "typeassert.value.ptr")
-		size := c.targetData.TypeAllocSize(assertedType)
-		if size > c.targetData.TypeAllocSize(c.i8ptrType) {
-			// Value was stored in an allocated buffer, load it from there.
-			valuePtrCast := c.builder.CreateBitCast(valuePtr, llvm.PointerType(assertedType, 0), "")
-			valueOk = c.builder.CreateLoad(valuePtrCast, "typeassert.value.ok")
-		} else if size == 0 {
-			valueOk, err = c.getZeroValue(assertedType)
-			if err != nil {
-				return llvm.Value{}, err
-			}
-		} else {
-			// Value was stored directly in the interface.
-			switch assertedType.TypeKind() {
-			case llvm.IntegerTypeKind:
-				valueOk = c.builder.CreatePtrToInt(valuePtr, assertedType, "typeassert.value.ok")
-			case llvm.PointerTypeKind:
-				valueOk = c.builder.CreateBitCast(valuePtr, assertedType, "typeassert.value.ok")
-			default: // struct, float, etc.
-				// A bitcast would be useful here, but bitcast doesn't allow
-				// aggregate types. So we'll bitcast it using an alloca.
-				// Hopefully this will get optimized away.
-				mem := c.builder.CreateAlloca(c.i8ptrType, "")
-				c.builder.CreateStore(valuePtr, mem)
-				memCast := c.builder.CreateBitCast(mem, llvm.PointerType(assertedType, 0), "")
-				valueOk = c.builder.CreateLoad(memCast, "typeassert.value.ok")
-			}
-		}
+		valueOk = c.emitPointerUnpack(valuePtr, []llvm.Type{assertedType})[0]
 	}
 	c.builder.CreateBr(nextBlock)
 
 	// Continue after the if statement.
 	c.builder.SetInsertPointAtEnd(nextBlock)
 	phi := c.builder.CreatePHI(assertedType, "typeassert.value")
-	phi.AddIncoming([]llvm.Value{valueNil, valueOk}, []llvm.BasicBlock{prevBlock, okBlock})
+	phi.AddIncoming([]llvm.Value{c.getZeroValue(assertedType), valueOk}, []llvm.BasicBlock{prevBlock, okBlock})
 
 	if expr.CommaOk {
 		tuple := c.ctx.ConstStruct([]llvm.Value{llvm.Undef(assertedType), llvm.Undef(c.ctx.Int1Type())}, false) // create empty tuple
 		tuple = c.builder.CreateInsertValue(tuple, phi, 0, "")                                                  // insert value
 		tuple = c.builder.CreateInsertValue(tuple, commaOk, 1, "")                                              // insert 'comma ok' boolean
-		return tuple, nil
+		return tuple
 	} else {
 		// This is kind of dirty as the branch above becomes mostly useless,
 		// but hopefully this gets optimized away.
 		c.createRuntimeCall("interfaceTypeAssert", []llvm.Value{commaOk}, "")
-		return phi, nil
+		return phi
 	}
 }
 
 // getInvokeCall creates and returns the function pointer and parameters of an
 // interface call. It can be used in a call or defer instruction.
-func (c *Compiler) getInvokeCall(frame *Frame, instr *ssa.CallCommon) (llvm.Value, []llvm.Value, error) {
+func (c *Compiler) getInvokeCall(frame *Frame, instr *ssa.CallCommon) (llvm.Value, []llvm.Value) {
 	// Call an interface method with dynamic dispatch.
-	itf, err := c.parseExpr(frame, instr.Value) // interface
-	if err != nil {
-		return llvm.Value{}, nil, err
-	}
+	itf := c.getValue(frame, instr.Value) // interface
 
-	llvmFnType, err := c.getRawFuncType(instr.Method.Type().(*types.Signature))
-	if err != nil {
-		return llvm.Value{}, nil, err
-	}
+	llvmFnType := c.getRawFuncType(instr.Method.Type().(*types.Signature))
 
 	typecode := c.builder.CreateExtractValue(itf, 0, "invoke.typecode")
 	values := []llvm.Value{
@@ -413,11 +335,7 @@ func (c *Compiler) getInvokeCall(frame *Frame, instr *ssa.CallCommon) (llvm.Valu
 
 	args := []llvm.Value{receiverValue}
 	for _, arg := range instr.Args {
-		val, err := c.parseExpr(frame, arg)
-		if err != nil {
-			return llvm.Value{}, nil, err
-		}
-		args = append(args, val)
+		args = append(args, c.getValue(frame, arg))
 	}
 	// Add the context parameter. An interface call never takes a context but we
 	// have to supply the parameter anyway.
@@ -425,7 +343,7 @@ func (c *Compiler) getInvokeCall(frame *Frame, instr *ssa.CallCommon) (llvm.Valu
 	// Add the parent goroutine handle.
 	args = append(args, llvm.Undef(c.i8ptrType))
 
-	return fnCast, args, nil
+	return fnCast, args
 }
 
 // interfaceInvokeWrapper keeps some state between getInterfaceInvokeWrapper and
@@ -441,19 +359,16 @@ type interfaceInvokeWrapper struct {
 // the underlying value, dereferences it, and calls the real method. This
 // wrapper is only needed when the interface value actually doesn't fit in a
 // pointer and a pointer to the value must be created.
-func (c *Compiler) getInterfaceInvokeWrapper(f *ir.Function) (llvm.Value, error) {
+func (c *Compiler) getInterfaceInvokeWrapper(f *ir.Function) llvm.Value {
 	wrapperName := f.LinkName() + "$invoke"
 	wrapper := c.mod.NamedFunction(wrapperName)
 	if !wrapper.IsNil() {
 		// Wrapper already created. Return it directly.
-		return wrapper, nil
+		return wrapper
 	}
 
 	// Get the expanded receiver type.
-	receiverType, err := c.getLLVMType(f.Params[0].Type())
-	if err != nil {
-		return llvm.Value{}, err
-	}
+	receiverType := c.getLLVMType(f.Params[0].Type())
 	expandedReceiverType := c.expandFormalParamType(receiverType)
 
 	// Does this method even need any wrapping?
@@ -462,7 +377,7 @@ func (c *Compiler) getInterfaceInvokeWrapper(f *ir.Function) (llvm.Value, error)
 		// Casting a function signature to a different signature and calling it
 		// with a receiver pointer bitcasted to *i8 (as done in calls on an
 		// interface) is hopefully a safe (defined) operation.
-		return f.LLVMFn, nil
+		return f.LLVMFn
 	}
 
 	// create wrapper function
@@ -475,12 +390,12 @@ func (c *Compiler) getInterfaceInvokeWrapper(f *ir.Function) (llvm.Value, error)
 		wrapper:      wrapper,
 		receiverType: receiverType,
 	})
-	return wrapper, nil
+	return wrapper
 }
 
 // createInterfaceInvokeWrapper finishes the work of getInterfaceInvokeWrapper,
 // see that function for details.
-func (c *Compiler) createInterfaceInvokeWrapper(state interfaceInvokeWrapper) error {
+func (c *Compiler) createInterfaceInvokeWrapper(state interfaceInvokeWrapper) {
 	wrapper := state.wrapper
 	fn := state.fn
 	receiverType := state.receiverType
@@ -490,10 +405,7 @@ func (c *Compiler) createInterfaceInvokeWrapper(state interfaceInvokeWrapper) er
 	// add debug info if needed
 	if c.Debug {
 		pos := c.ir.Program.Fset.Position(fn.Pos())
-		difunc, err := c.attachDebugInfoRaw(fn, wrapper, "$invoke", pos.Filename, pos.Line)
-		if err != nil {
-			return err
-		}
+		difunc := c.attachDebugInfoRaw(fn, wrapper, "$invoke", pos.Filename, pos.Line)
 		c.builder.SetCurrentDebugLocation(uint(pos.Line), uint(pos.Column), difunc, llvm.Metadata{})
 	}
 
@@ -501,28 +413,7 @@ func (c *Compiler) createInterfaceInvokeWrapper(state interfaceInvokeWrapper) er
 	block := c.ctx.AddBasicBlock(wrapper, "entry")
 	c.builder.SetInsertPointAtEnd(block)
 
-	var receiverPtr llvm.Value
-	if c.targetData.TypeAllocSize(receiverType) > c.targetData.TypeAllocSize(c.i8ptrType) {
-		// The receiver is passed in using a pointer. We have to load it here
-		// and pass it by value to the real function.
-
-		// Load the underlying value.
-		receiverPtrType := llvm.PointerType(receiverType, 0)
-		receiverPtr = c.builder.CreateBitCast(wrapper.Param(0), receiverPtrType, "receiver.ptr")
-	} else {
-		// The value is stored in the interface, but it is of type struct which
-		// is expanded to multiple parameters (e.g. {i8, i8}). So we have to
-		// receive the struct as parameter, expand it, and pass it on to the
-		// real function.
-
-		// Cast the passed-in i8* to the struct value (using an alloca) and
-		// extract its values.
-		alloca := c.builder.CreateAlloca(c.i8ptrType, "receiver.alloca")
-		c.builder.CreateStore(wrapper.Param(0), alloca)
-		receiverPtr = c.builder.CreateBitCast(alloca, llvm.PointerType(receiverType, 0), "receiver.ptr")
-	}
-
-	receiverValue := c.builder.CreateLoad(receiverPtr, "receiver")
+	receiverValue := c.emitPointerUnpack(wrapper.Param(0), []llvm.Type{receiverType})[0]
 	params := append(c.expandFormalParam(receiverValue), wrapper.Params()[1:]...)
 	if fn.LLVMFn.Type().ElementType().ReturnType().TypeKind() == llvm.VoidTypeKind {
 		c.builder.CreateCall(fn.LLVMFn, params, "")
@@ -531,6 +422,4 @@ func (c *Compiler) createInterfaceInvokeWrapper(state interfaceInvokeWrapper) er
 		ret := c.builder.CreateCall(fn.LLVMFn, params, "ret")
 		c.builder.CreateRet(ret)
 	}
-
-	return nil
 }

compiler/llvm.go

@@ -22,6 +22,65 @@ func getUses(value llvm.Value) []llvm.Value {
 	return uses
 }
 
+// createEntryBlockAlloca creates a new alloca in the entry block, even though
+// the IR builder is located elsewhere. It assumes that the insert point is
+// at the end of the current block.
+func (c *Compiler) createEntryBlockAlloca(t llvm.Type, name string) llvm.Value {
+	currentBlock := c.builder.GetInsertBlock()
+	entryBlock := currentBlock.Parent().EntryBasicBlock()
+	if entryBlock.FirstInstruction().IsNil() {
+		c.builder.SetInsertPointAtEnd(entryBlock)
+	} else {
+		c.builder.SetInsertPointBefore(entryBlock.FirstInstruction())
+	}
+	alloca := c.builder.CreateAlloca(t, name)
+	c.builder.SetInsertPointAtEnd(currentBlock)
+	return alloca
+}
+
+// createTemporaryAlloca creates a new alloca in the entry block and adds
+// lifetime start infromation in the IR signalling that the alloca won't be used
+// before this point.
+//
+// This is useful for creating temporary allocas for intrinsics. Don't forget to
+// end the lifetime using emitLifetimeEnd after you're done with it.
+func (c *Compiler) createTemporaryAlloca(t llvm.Type, name string) (alloca, bitcast, size llvm.Value) {
+	alloca = c.createEntryBlockAlloca(t, name)
+	bitcast = c.builder.CreateBitCast(alloca, c.i8ptrType, name+".bitcast")
+	size = llvm.ConstInt(c.ctx.Int64Type(), c.targetData.TypeAllocSize(t), false)
+	c.builder.CreateCall(c.getLifetimeStartFunc(), []llvm.Value{size, bitcast}, "")
+	return
+}
+
+// emitLifetimeEnd signals the end of an (alloca) lifetime by calling the
+// llvm.lifetime.end intrinsic. It is commonly used together with
+// createTemporaryAlloca.
+func (c *Compiler) emitLifetimeEnd(ptr, size llvm.Value) {
+	c.builder.CreateCall(c.getLifetimeEndFunc(), []llvm.Value{size, ptr}, "")
+}
+
+// getLifetimeStartFunc returns the llvm.lifetime.start intrinsic and creates it
+// first if it doesn't exist yet.
+func (c *Compiler) getLifetimeStartFunc() llvm.Value {
+	fn := c.mod.NamedFunction("llvm.lifetime.start.p0i8")
+	if fn.IsNil() {
+		fnType := llvm.FunctionType(c.ctx.VoidType(), []llvm.Type{c.ctx.Int64Type(), c.i8ptrType}, false)
+		fn = llvm.AddFunction(c.mod, "llvm.lifetime.start.p0i8", fnType)
+	}
+	return fn
+}
+
+// getLifetimeEndFunc returns the llvm.lifetime.end intrinsic and creates it
+// first if it doesn't exist yet.
+func (c *Compiler) getLifetimeEndFunc() llvm.Value {
+	fn := c.mod.NamedFunction("llvm.lifetime.end.p0i8")
+	if fn.IsNil() {
+		fnType := llvm.FunctionType(c.ctx.VoidType(), []llvm.Type{c.ctx.Int64Type(), c.i8ptrType}, false)
+		fn = llvm.AddFunction(c.mod, "llvm.lifetime.end.p0i8", fnType)
+	}
+	return fn
+}
+
 // splitBasicBlock splits a LLVM basic block into two parts. All instructions
 // after afterInst are moved into a new basic block (created right after the
 // current one) with the given name.

compiler/map.go

@@ -10,12 +10,14 @@ import (
 )
 
 func (c *Compiler) emitMapLookup(keyType, valueType types.Type, m, key llvm.Value, commaOk bool, pos token.Pos) (llvm.Value, error) {
-	llvmValueType, err := c.getLLVMType(valueType)
-	if err != nil {
-		return llvm.Value{}, err
-	}
-	mapValueAlloca := c.builder.CreateAlloca(llvmValueType, "hashmap.value")
-	mapValuePtr := c.builder.CreateBitCast(mapValueAlloca, c.i8ptrType, "hashmap.valueptr")
+	llvmValueType := c.getLLVMType(valueType)
+
+	// Allocate the memory for the resulting type. Do not zero this memory: it
+	// will be zeroed by the hashmap get implementation if the key is not
+	// present in the map.
+	mapValueAlloca, mapValuePtr, mapValueSize := c.createTemporaryAlloca(llvmValueType, "hashmap.value")
+
+	// Do the lookup. How it is done depends on the key type.
 	var commaOkValue llvm.Value
 	if t, ok := keyType.(*types.Basic); ok && t.Info()&types.IsString != 0 {
 		// key is a string
@@ -23,15 +25,24 @@ func (c *Compiler) emitMapLookup(keyType, valueType types.Type, m, key llvm.Valu
 		commaOkValue = c.createRuntimeCall("hashmapStringGet", params, "")
 	} else if hashmapIsBinaryKey(keyType) {
 		// key can be compared with runtime.memequal
-		keyAlloca := c.builder.CreateAlloca(key.Type(), "hashmap.key")
-		c.builder.CreateStore(key, keyAlloca)
-		keyPtr := c.builder.CreateBitCast(keyAlloca, c.i8ptrType, "hashmap.keyptr")
-		params := []llvm.Value{m, keyPtr, mapValuePtr}
+		// Store the key in an alloca, in the entry block to avoid dynamic stack
+		// growth.
+		mapKeyAlloca, mapKeyPtr, mapKeySize := c.createTemporaryAlloca(key.Type(), "hashmap.key")
+		c.builder.CreateStore(key, mapKeyAlloca)
+		// Fetch the value from the hashmap.
+		params := []llvm.Value{m, mapKeyPtr, mapValuePtr}
 		commaOkValue = c.createRuntimeCall("hashmapBinaryGet", params, "")
+		c.emitLifetimeEnd(mapKeyPtr, mapKeySize)
 	} else {
-		return llvm.Value{}, c.makeError(pos, "only strings, bools, ints or structs of bools/ints are supported as map keys, but got: "+keyType.String())
+		// Not trivially comparable using memcmp.
+		return llvm.Value{}, c.makeError(pos, "only strings, bools, ints, pointers or structs of bools/ints are supported as map keys, but got: "+keyType.String())
 	}
+
+	// Load the resulting value from the hashmap. The value is set to the zero
+	// value if the key doesn't exist in the hashmap.
 	mapValue := c.builder.CreateLoad(mapValueAlloca, "")
+	c.emitLifetimeEnd(mapValuePtr, mapValueSize)
+
 	if commaOk {
 		tuple := llvm.Undef(c.ctx.StructType([]llvm.Type{llvmValueType, c.ctx.Int1Type()}, false))
 		tuple = c.builder.CreateInsertValue(tuple, mapValue, 0, "")
@@ -42,27 +53,25 @@ func (c *Compiler) emitMapLookup(keyType, valueType types.Type, m, key llvm.Valu
 	}
 }
 
-func (c *Compiler) emitMapUpdate(keyType types.Type, m, key, value llvm.Value, pos token.Pos) error {
-	valueAlloca := c.builder.CreateAlloca(value.Type(), "hashmap.value")
+func (c *Compiler) emitMapUpdate(keyType types.Type, m, key, value llvm.Value, pos token.Pos) {
+	valueAlloca, valuePtr, valueSize := c.createTemporaryAlloca(value.Type(), "hashmap.value")
 	c.builder.CreateStore(value, valueAlloca)
-	valuePtr := c.builder.CreateBitCast(valueAlloca, c.i8ptrType, "hashmap.valueptr")
 	keyType = keyType.Underlying()
 	if t, ok := keyType.(*types.Basic); ok && t.Info()&types.IsString != 0 {
 		// key is a string
 		params := []llvm.Value{m, key, valuePtr}
 		c.createRuntimeCall("hashmapStringSet", params, "")
-		return nil
 	} else if hashmapIsBinaryKey(keyType) {
 		// key can be compared with runtime.memequal
-		keyAlloca := c.builder.CreateAlloca(key.Type(), "hashmap.key")
+		keyAlloca, keyPtr, keySize := c.createTemporaryAlloca(key.Type(), "hashmap.key")
 		c.builder.CreateStore(key, keyAlloca)
-		keyPtr := c.builder.CreateBitCast(keyAlloca, c.i8ptrType, "hashmap.keyptr")
 		params := []llvm.Value{m, keyPtr, valuePtr}
 		c.createRuntimeCall("hashmapBinarySet", params, "")
-		return nil
+		c.emitLifetimeEnd(keyPtr, keySize)
 	} else {
-		return c.makeError(pos, "only strings, bools, ints or structs of bools/ints are supported as map keys, but got: "+keyType.String())
+		c.addError(pos, "only strings, bools, ints, pointers or structs of bools/ints are supported as map keys, but got: "+keyType.String())
 	}
+	c.emitLifetimeEnd(valuePtr, valueSize)
 }
 
 func (c *Compiler) emitMapDelete(keyType types.Type, m, key llvm.Value, pos token.Pos) error {
@@ -73,14 +82,14 @@ func (c *Compiler) emitMapDelete(keyType types.Type, m, key llvm.Value, pos toke
 		c.createRuntimeCall("hashmapStringDelete", params, "")
 		return nil
 	} else if hashmapIsBinaryKey(keyType) {
-		keyAlloca := c.builder.CreateAlloca(key.Type(), "hashmap.key")
+		keyAlloca, keyPtr, keySize := c.createTemporaryAlloca(key.Type(), "hashmap.key")
 		c.builder.CreateStore(key, keyAlloca)
-		keyPtr := c.builder.CreateBitCast(keyAlloca, c.i8ptrType, "hashmap.keyptr")
 		params := []llvm.Value{m, keyPtr}
 		c.createRuntimeCall("hashmapBinaryDelete", params, "")
+		c.emitLifetimeEnd(keyPtr, keySize)
 		return nil
 	} else {
-		return c.makeError(pos, "only strings, bools, ints or structs of bools/ints are supported as map keys, but got: "+keyType.String())
+		return c.makeError(pos, "only strings, bools, ints, pointers or structs of bools/ints are supported as map keys, but got: "+keyType.String())
 	}
 }
 
@@ -112,6 +121,8 @@ func hashmapIsBinaryKey(keyType types.Type) bool {
 	switch keyType := keyType.(type) {
 	case *types.Basic:
 		return keyType.Info()&(types.IsBoolean|types.IsInteger) != 0
+	case *types.Pointer:
+		return true
 	case *types.Struct:
 		for i := 0; i < keyType.NumFields(); i++ {
 			fieldType := keyType.Field(i).Type().Underlying()

compiler/optimizer.go

@@ -18,6 +18,10 @@ func (c *Compiler) Optimize(optLevel, sizeLevel int, inlinerThreshold uint) erro
 	}
 	builder.AddCoroutinePassesToExtensionPoints()
 
+	if c.PanicStrategy == "trap" {
+		c.replacePanicsWithTrap() // -panic=trap
+	}
+
 	// Run function passes for each function.
 	funcPasses := llvm.NewFunctionPassManagerForModule(c.mod)
 	defer funcPasses.Dispose()
@@ -110,9 +114,36 @@ func (c *Compiler) Optimize(optLevel, sizeLevel int, inlinerThreshold uint) erro
 	builder.Populate(modPasses)
 	modPasses.Run(c.mod)
 
+	hasGCPass := c.addGlobalsBitmap()
+	hasGCPass = c.makeGCStackSlots() || hasGCPass
+	if hasGCPass {
+		if err := c.Verify(); err != nil {
+			return errors.New("GC pass caused a verification failure")
+		}
+	}
+
 	return nil
 }
 
+// Replace panic calls with calls to llvm.trap, to reduce code size. This is the
+// -panic=trap intrinsic.
+func (c *Compiler) replacePanicsWithTrap() {
+	trap := c.mod.NamedFunction("llvm.trap")
+	for _, name := range []string{"runtime._panic", "runtime.runtimePanic"} {
+		fn := c.mod.NamedFunction(name)
+		if fn.IsNil() {
+			continue
+		}
+		for _, use := range getUses(fn) {
+			if use.IsACallInst().IsNil() || use.CalledValue() != fn {
+				panic("expected use of a panic function to be a call")
+			}
+			c.builder.SetInsertPointBefore(use)
+			c.builder.CreateCall(trap, nil, "")
+		}
+	}
+}
+
 // Eliminate created but not used maps.
 //
 // In the future, this should statically allocate created but never modified
@@ -249,7 +280,7 @@ func (c *Compiler) OptimizeAllocs() {
 			sizeInWords := (size + uint64(alignment) - 1) / uint64(alignment)
 			allocaType := llvm.ArrayType(c.ctx.IntType(alignment*8), int(sizeInWords))
 			alloca := c.builder.CreateAlloca(allocaType, "stackalloc.alloca")
-			zero, _ := c.getZeroValue(alloca.Type().ElementType())
+			zero := c.getZeroValue(alloca.Type().ElementType())
 			c.builder.CreateStore(zero, alloca)
 			stackalloc := c.builder.CreateBitCast(alloca, bitcast.Type(), "stackalloc")
 			bitcast.ReplaceAllUsesWith(stackalloc)
@@ -284,18 +315,6 @@ func (c *Compiler) doesEscape(value llvm.Value) bool {
 				return true
 			}
 		} else if use.IsACallInst() != nilValue {
-			// Call only escapes when the (pointer) parameter is not marked
-			// "nocapture". This flag means that the parameter does not escape
-			// the give function.
-			if use.CalledValue().IsAFunction() != nilValue {
-				if use.CalledValue().IsDeclaration() {
-					// Kind of dirty: assume external functions don't let
-					// pointers escape.
-					// TODO: introduce //go:noescape that sets the 'nocapture'
-					// flag on each input parameter.
-					continue
-				}
-			}
 			if !c.hasFlag(use, value, "nocapture") {
 				return true
 			}

compiler/sizes.go

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

compiler/symbol.go

@@ -0,0 +1,107 @@
+package compiler
+
+// This file manages symbols, that is, functions and globals. It reads their
+// pragmas, determines the link name, etc.
+
+import (
+	"go/ast"
+	"go/token"
+	"go/types"
+	"strings"
+
+	"github.com/tinygo-org/tinygo/loader"
+	"golang.org/x/tools/go/ssa"
+	"tinygo.org/x/go-llvm"
+)
+
+// globalInfo contains some information about a specific global. By default,
+// linkName is equal to .RelString(nil) on a global and extern is false, but for
+// some symbols this is different (due to //go:extern for example).
+type globalInfo struct {
+	linkName string // go:extern
+	extern   bool   // go:extern
+}
+
+// loadASTComments loads comments on globals from the AST, for use later in the
+// program. In particular, they are required for //go:extern pragmas on globals.
+func (c *Compiler) loadASTComments(lprogram *loader.Program) {
+	c.astComments = map[string]*ast.CommentGroup{}
+	for _, pkgInfo := range lprogram.Sorted() {
+		for _, file := range pkgInfo.Files {
+			for _, decl := range file.Decls {
+				switch decl := decl.(type) {
+				case *ast.GenDecl:
+					switch decl.Tok {
+					case token.VAR:
+						if len(decl.Specs) != 1 {
+							continue
+						}
+						for _, spec := range decl.Specs {
+							switch spec := spec.(type) {
+							case *ast.ValueSpec: // decl.Tok == token.VAR
+								for _, name := range spec.Names {
+									id := pkgInfo.Pkg.Path() + "." + name.Name
+									c.astComments[id] = decl.Doc
+								}
+							}
+						}
+					}
+				}
+			}
+		}
+	}
+}
+
+// getGlobal returns a LLVM IR global value for a Go SSA global. It is added to
+// the LLVM IR if it has not been added already.
+func (c *Compiler) getGlobal(g *ssa.Global) llvm.Value {
+	info := c.getGlobalInfo(g)
+	llvmGlobal := c.mod.NamedGlobal(info.linkName)
+	if llvmGlobal.IsNil() {
+		llvmType := c.getLLVMType(g.Type().(*types.Pointer).Elem())
+		llvmGlobal = llvm.AddGlobal(c.mod, llvmType, info.linkName)
+		if !info.extern {
+			llvmGlobal.SetInitializer(c.getZeroValue(llvmType))
+			llvmGlobal.SetLinkage(llvm.InternalLinkage)
+		}
+	}
+	return llvmGlobal
+}
+
+// getGlobalInfo returns some information about a specific global.
+func (c *Compiler) getGlobalInfo(g *ssa.Global) globalInfo {
+	info := globalInfo{}
+	if strings.HasPrefix(g.Name(), "C.") {
+		// Created by CGo: such a name cannot be created by regular C code.
+		info.linkName = g.Name()[2:]
+		info.extern = true
+	} else {
+		// Pick the default linkName.
+		info.linkName = g.RelString(nil)
+		// Check for //go: pragmas, which may change the link name (among
+		// others).
+		doc := c.astComments[info.linkName]
+		if doc != nil {
+			info.parsePragmas(doc)
+		}
+	}
+	return info
+}
+
+// Parse //go: pragma comments from the source. In particular, it parses the
+// //go:extern pragma on globals.
+func (info *globalInfo) parsePragmas(doc *ast.CommentGroup) {
+	for _, comment := range doc.List {
+		if !strings.HasPrefix(comment.Text, "//go:") {
+			continue
+		}
+		parts := strings.Fields(comment.Text)
+		switch parts[0] {
+		case "//go:extern":
+			info.extern = true
+			if len(parts) == 2 {
+				info.linkName = parts[1]
+			}
+		}
+	}
+}

compiler/syscall.go

@@ -4,7 +4,6 @@ package compiler
 // compiler builtins.
 
 import (
-	"go/constant"
 	"strconv"
 
 	"golang.org/x/tools/go/ssa"
@@ -14,7 +13,7 @@ import (
 // emitSyscall emits an inline system call instruction, depending on the target
 // OS/arch.
 func (c *Compiler) emitSyscall(frame *Frame, call *ssa.CallCommon) (llvm.Value, error) {
-	num, _ := constant.Uint64Val(call.Args[0].(*ssa.Const).Value)
+	num := c.getValue(frame, call.Args[0])
 	var syscallResult llvm.Value
 	switch {
 	case c.GOARCH == "amd64":
@@ -29,12 +28,12 @@ func (c *Compiler) emitSyscall(frame *Frame, call *ssa.CallCommon) (llvm.Value,
 			// > All system classes enter the kernel via the syscall instruction.
 			//
 			// Source: https://opensource.apple.com/source/xnu/xnu-792.13.8/osfmk/mach/i386/syscall_sw.h
-			num += 0x2000000
+			num = c.builder.CreateOr(num, llvm.ConstInt(c.uintptrType, 0x2000000, false), "")
 		}
 		// Sources:
 		//   https://stackoverflow.com/a/2538212
 		//   https://en.wikibooks.org/wiki/X86_Assembly/Interfacing_with_Linux#syscall
-		args := []llvm.Value{llvm.ConstInt(c.uintptrType, num, false)}
+		args := []llvm.Value{num}
 		argTypes := []llvm.Type{c.uintptrType}
 		// Constraints will look something like:
 		//   "={rax},0,{rdi},{rsi},{rdx},{r10},{r8},{r9},~{rcx},~{r11}"
@@ -51,10 +50,7 @@ func (c *Compiler) emitSyscall(frame *Frame, call *ssa.CallCommon) (llvm.Value,
 				"{r12}",
 				"{r13}",
 			}[i]
-			llvmValue, err := c.parseExpr(frame, arg)
-			if err != nil {
-				return llvm.Value{}, err
-			}
+			llvmValue := c.getValue(frame, arg)
 			args = append(args, llvmValue)
 			argTypes = append(argTypes, llvmValue.Type())
 		}
@@ -80,14 +76,11 @@ func (c *Compiler) emitSyscall(frame *Frame, call *ssa.CallCommon) (llvm.Value,
 				"{r5}",
 				"{r6}",
 			}[i]
-			llvmValue, err := c.parseExpr(frame, arg)
-			if err != nil {
-				return llvm.Value{}, err
-			}
+			llvmValue := c.getValue(frame, arg)
 			args = append(args, llvmValue)
 			argTypes = append(argTypes, llvmValue.Type())
 		}
-		args = append(args, llvm.ConstInt(c.uintptrType, num, false))
+		args = append(args, num)
 		argTypes = append(argTypes, c.uintptrType)
 		constraints += ",{r7}" // syscall number
 		for i := len(call.Args) - 1; i < 4; i++ {
@@ -113,14 +106,11 @@ func (c *Compiler) emitSyscall(frame *Frame, call *ssa.CallCommon) (llvm.Value,
 				"{x4}",
 				"{x5}",
 			}[i]
-			llvmValue, err := c.parseExpr(frame, arg)
-			if err != nil {
-				return llvm.Value{}, err
-			}
+			llvmValue := c.getValue(frame, arg)
 			args = append(args, llvmValue)
 			argTypes = append(argTypes, llvmValue.Type())
 		}
-		args = append(args, llvm.ConstInt(c.uintptrType, num, false))
+		args = append(args, num)
 		argTypes = append(argTypes, c.uintptrType)
 		constraints += ",{x8}" // syscall number
 		for i := len(call.Args) - 1; i < 8; i++ {

compiler/volatile.go

@@ -0,0 +1,26 @@
+package compiler
+
+// This file implements volatile loads/stores in runtime/volatile.LoadT and
+// runtime/volatile.StoreT as compiler builtins.
+
+import (
+	"golang.org/x/tools/go/ssa"
+	"tinygo.org/x/go-llvm"
+)
+
+func (c *Compiler) emitVolatileLoad(frame *Frame, instr *ssa.CallCommon) (llvm.Value, error) {
+	addr := c.getValue(frame, instr.Args[0])
+	c.emitNilCheck(frame, addr, "deref")
+	val := c.builder.CreateLoad(addr, "")
+	val.SetVolatile(true)
+	return val, nil
+}
+
+func (c *Compiler) emitVolatileStore(frame *Frame, instr *ssa.CallCommon) (llvm.Value, error) {
+	addr := c.getValue(frame, instr.Args[0])
+	val := c.getValue(frame, instr.Args[1])
+	c.emitNilCheck(frame, addr, "deref")
+	store := c.builder.CreateStore(val, addr)
+	store.SetVolatile(true)
+	return llvm.Value{}, nil
+}

compiler/wordpack.go

@@ -0,0 +1,120 @@
+package compiler
+
+// This file contains utility functions to pack and unpack sets of values. It
+// can take in a list of values and tries to store it efficiently in the pointer
+// itself if possible and legal.
+
+import (
+	"tinygo.org/x/go-llvm"
+)
+
+// emitPointerPack packs the list of values into a single pointer value using
+// bitcasts, or else allocates a value on the heap if it cannot be packed in the
+// pointer value directly. It returns the pointer with the packed data.
+func (c *Compiler) emitPointerPack(values []llvm.Value) llvm.Value {
+	valueTypes := make([]llvm.Type, len(values))
+	for i, value := range values {
+		valueTypes[i] = value.Type()
+	}
+	packedType := c.ctx.StructType(valueTypes, false)
+
+	// Allocate memory for the packed data.
+	var packedAlloc, packedHeapAlloc llvm.Value
+	size := c.targetData.TypeAllocSize(packedType)
+	if size == 0 {
+		return llvm.ConstPointerNull(c.i8ptrType)
+	} else if len(values) == 1 && values[0].Type().TypeKind() == llvm.PointerTypeKind {
+		return c.builder.CreateBitCast(values[0], c.i8ptrType, "pack.ptr")
+	} else if size <= c.targetData.TypeAllocSize(c.i8ptrType) {
+		// Packed data fits in a pointer, so store it directly inside the
+		// pointer.
+		if len(values) == 1 && values[0].Type().TypeKind() == llvm.IntegerTypeKind {
+			// Try to keep this cast in SSA form.
+			return c.builder.CreateIntToPtr(values[0], c.i8ptrType, "pack.int")
+		}
+		// Because packedType is a struct and we have to cast it to a *i8, store
+		// it in an alloca first for bitcasting (store+bitcast+load).
+		packedAlloc, _, _ = c.createTemporaryAlloca(packedType, "")
+	} else {
+		// Packed data is bigger than a pointer, so allocate it on the heap.
+		sizeValue := llvm.ConstInt(c.uintptrType, size, false)
+		packedHeapAlloc = c.createRuntimeCall("alloc", []llvm.Value{sizeValue}, "")
+		if c.needsStackObjects() {
+			c.trackPointer(packedHeapAlloc)
+		}
+		packedAlloc = c.builder.CreateBitCast(packedHeapAlloc, llvm.PointerType(packedType, 0), "")
+	}
+	// Store all values in the alloca or heap pointer.
+	for i, value := range values {
+		indices := []llvm.Value{
+			llvm.ConstInt(c.ctx.Int32Type(), 0, false),
+			llvm.ConstInt(c.ctx.Int32Type(), uint64(i), false),
+		}
+		gep := c.builder.CreateInBoundsGEP(packedAlloc, indices, "")
+		c.builder.CreateStore(value, gep)
+	}
+
+	if packedHeapAlloc.IsNil() {
+		// Load value (as *i8) from the alloca.
+		packedAlloc = c.builder.CreateBitCast(packedAlloc, llvm.PointerType(c.i8ptrType, 0), "")
+		result := c.builder.CreateLoad(packedAlloc, "")
+		packedPtr := c.builder.CreateBitCast(packedAlloc, c.i8ptrType, "")
+		packedSize := llvm.ConstInt(c.ctx.Int64Type(), c.targetData.TypeAllocSize(packedAlloc.Type()), false)
+		c.emitLifetimeEnd(packedPtr, packedSize)
+		return result
+	} else {
+		// Get the original heap allocation pointer, which already is an *i8.
+		return packedHeapAlloc
+	}
+}
+
+// emitPointerUnpack extracts a list of values packed using emitPointerPack.
+func (c *Compiler) emitPointerUnpack(ptr llvm.Value, valueTypes []llvm.Type) []llvm.Value {
+	packedType := c.ctx.StructType(valueTypes, false)
+
+	// Get a correctly-typed pointer to the packed data.
+	var packedAlloc, packedRawAlloc llvm.Value
+	size := c.targetData.TypeAllocSize(packedType)
+	if size == 0 {
+		// No data to unpack.
+	} else if len(valueTypes) == 1 && valueTypes[0].TypeKind() == llvm.PointerTypeKind {
+		// A single pointer is always stored directly.
+		return []llvm.Value{c.builder.CreateBitCast(ptr, valueTypes[0], "unpack.ptr")}
+	} else if size <= c.targetData.TypeAllocSize(c.i8ptrType) {
+		// Packed data stored directly in pointer.
+		if len(valueTypes) == 1 && valueTypes[0].TypeKind() == llvm.IntegerTypeKind {
+			// Keep this cast in SSA form.
+			return []llvm.Value{c.builder.CreatePtrToInt(ptr, valueTypes[0], "unpack.int")}
+		}
+		// Fallback: load it using an alloca.
+		packedRawAlloc, _, _ = c.createTemporaryAlloca(llvm.PointerType(c.i8ptrType, 0), "unpack.raw.alloc")
+		packedRawValue := c.builder.CreateBitCast(ptr, llvm.PointerType(c.i8ptrType, 0), "unpack.raw.value")
+		c.builder.CreateStore(packedRawValue, packedRawAlloc)
+		packedAlloc = c.builder.CreateBitCast(packedRawAlloc, llvm.PointerType(packedType, 0), "unpack.alloc")
+	} else {
+		// Packed data stored on the heap. Bitcast the passed-in pointer to the
+		// correct pointer type.
+		packedAlloc = c.builder.CreateBitCast(ptr, llvm.PointerType(packedType, 0), "unpack.raw.ptr")
+	}
+	// Load each value from the packed data.
+	values := make([]llvm.Value, len(valueTypes))
+	for i, valueType := range valueTypes {
+		if c.targetData.TypeAllocSize(valueType) == 0 {
+			// This value has length zero, so there's nothing to load.
+			values[i] = c.getZeroValue(valueType)
+			continue
+		}
+		indices := []llvm.Value{
+			llvm.ConstInt(c.ctx.Int32Type(), 0, false),
+			llvm.ConstInt(c.ctx.Int32Type(), uint64(i), false),
+		}
+		gep := c.builder.CreateInBoundsGEP(packedAlloc, indices, "")
+		values[i] = c.builder.CreateLoad(gep, "")
+	}
+	if !packedRawAlloc.IsNil() {
+		allocPtr := c.builder.CreateBitCast(packedRawAlloc, c.i8ptrType, "")
+		allocSize := llvm.ConstInt(c.ctx.Int64Type(), c.targetData.TypeAllocSize(c.uintptrType), false)
+		c.emitLifetimeEnd(allocPtr, allocSize)
+	}
+	return values
+}

go.mod

@@ -0,0 +1,10 @@
+module github.com/tinygo-org/tinygo
+
+go 1.11
+
+require (
+	github.com/blakesmith/ar v0.0.0-20150311145944-8bd4349a67f2
+	github.com/marcinbor85/gohex v0.0.0-20180128172054-7a43cd876e46
+	golang.org/x/tools v0.0.0-20190227180812-8dcc6e70cdef
+	tinygo.org/x/go-llvm v0.0.0-20190224120431-7707ae5d1261
+)

go.sum

@@ -0,0 +1,14 @@
+github.com/blakesmith/ar v0.0.0-20150311145944-8bd4349a67f2 h1:oMCHnXa6CCCafdPDbMh/lWRhRByN0VFLvv+g+ayx1SI=
+github.com/blakesmith/ar v0.0.0-20150311145944-8bd4349a67f2/go.mod h1:PkYb9DJNAwrSvRx5DYA+gUcOIgTGVMNkfSCbZM8cWpI=
+github.com/golang/protobuf v1.2.0/go.mod h1:6lQm79b+lXiMfvg/cZm0SGofjICqVBUtrP5yJMmIC1U=
+github.com/marcinbor85/gohex v0.0.0-20180128172054-7a43cd876e46 h1:wXG2bA8fO7Vv7lLk2PihFMTqmbT173Tje39oKzQ50Mo=
+github.com/marcinbor85/gohex v0.0.0-20180128172054-7a43cd876e46/go.mod h1:Pb6XcsXyropB9LNHhnqaknG/vEwYztLkQzVCHv8sQ3M=
+golang.org/x/net v0.0.0-20180724234803-3673e40ba225/go.mod h1:mL1N/T3taQHkDXs73rZJwtUhF3w3ftmwwsq0BUmARs4=
+golang.org/x/net v0.0.0-20190213061140-3a22650c66bd/go.mod h1:mL1N/T3taQHkDXs73rZJwtUhF3w3ftmwwsq0BUmARs4=
+golang.org/x/sync v0.0.0-20181108010431-42b317875d0f/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
+golang.org/x/text v0.3.0/go.mod h1:NqM8EUOU14njkJ3fqMW+pc6Ldnwhi/IjpwHt7yyuwOQ=
+golang.org/x/tools v0.0.0-20190227180812-8dcc6e70cdef h1:ymc9FeDom3RIEA3coKokSllBB1hRcMT0tZ1W3Jf9Ids=
+golang.org/x/tools v0.0.0-20190227180812-8dcc6e70cdef/go.mod h1:9Yl7xja0Znq3iFh3HoIrodX9oNMXvdceNzlUR8zjMvY=
+google.golang.org/appengine v1.4.0/go.mod h1:xpcJRLb0r/rnEns0DIKYYv+WjYCduHsrkT7/EB5XEv4=
+tinygo.org/x/go-llvm v0.0.0-20190224120431-7707ae5d1261 h1:rJS2Hga39YAnm7DE4qrPm6Dr/67EOojL0XPzvbEeBiw=
+tinygo.org/x/go-llvm v0.0.0-20190224120431-7707ae5d1261/go.mod h1:fv1F0BSNpxMfCL0zF3M4OPFbgYHnhtB6ST0HvUtu/LE=

interp/frame.go

@@ -176,6 +176,20 @@ func (fr *frame) evalBasicBlock(bb, incoming llvm.BasicBlock, indent string) (re
 					continue // special case: bitcast of alloc
 				}
 			}
+			if _, ok := fr.getLocal(operand).(*MapValue); ok {
+				// Special case for runtime.trackPointer calls.
+				// Note: this might not be entirely sound in some rare cases
+				// where the map is stored in a dirty global.
+				uses := getUses(inst)
+				if len(uses) == 1 {
+					use := uses[0]
+					if !use.IsACallInst().IsNil() && !use.CalledValue().IsAFunction().IsNil() && use.CalledValue().Name() == "runtime.trackPointer" {
+						continue
+					}
+				}
+				// It is not possible in Go to bitcast a map value to a pointer.
+				panic("unimplemented: bitcast of map")
+			}
 			value := fr.getLocal(operand).(*LocalValue)
 			fr.locals[inst] = &LocalValue{fr.Eval, fr.builder.CreateBitCast(value.Value(), inst.Type(), "")}
 
@@ -184,6 +198,25 @@ func (fr *frame) evalBasicBlock(bb, incoming llvm.BasicBlock, indent string) (re
 			lhs := fr.getLocal(inst.Operand(0)).(*LocalValue).Underlying
 			rhs := fr.getLocal(inst.Operand(1)).(*LocalValue).Underlying
 			predicate := inst.IntPredicate()
+			if predicate == llvm.IntEQ && lhs.Type().TypeKind() == llvm.PointerTypeKind {
+				// Unfortunately, the const propagation in the IR builder
+				// doesn't handle pointer compares of inttoptr values. So we
+				// implement it manually here.
+				lhsNil, ok1 := isPointerNil(lhs)
+				rhsNil, ok2 := isPointerNil(rhs)
+				if ok1 && ok2 {
+					if lhsNil && rhsNil {
+						// Both are nil, so this icmp is always evaluated to true.
+						fr.locals[inst] = &LocalValue{fr.Eval, llvm.ConstInt(fr.Mod.Context().Int1Type(), 1, false)}
+						continue
+					}
+					if lhsNil != rhsNil {
+						// Only one of them is nil, so this comparison must return false.
+						fr.locals[inst] = &LocalValue{fr.Eval, llvm.ConstInt(fr.Mod.Context().Int1Type(), 0, false)}
+						continue
+					}
+				}
+			}
 			fr.locals[inst] = &LocalValue{fr.Eval, fr.builder.CreateICmp(predicate, lhs, rhs, "")}
 		case !inst.IsAFCmpInst().IsNil():
 			lhs := fr.getLocal(inst.Operand(0)).(*LocalValue).Underlying
@@ -347,6 +380,10 @@ func (fr *frame) evalBasicBlock(bb, incoming llvm.BasicBlock, indent string) (re
 				fr.locals[inst] = &LocalValue{fr.Eval, llvm.ConstInt(fr.Mod.Context().Int1Type(), implements, false)}
 			case callee.Name() == "runtime.nanotime":
 				fr.locals[inst] = &LocalValue{fr.Eval, llvm.ConstInt(fr.Mod.Context().Int64Type(), 0, false)}
+			case callee.Name() == "llvm.dbg.value":
+				// do nothing
+			case callee.Name() == "runtime.trackPointer":
+				// do nothing
 			case strings.HasPrefix(callee.Name(), "runtime.print") || callee.Name() == "runtime._panic":
 				// This are all print instructions, which necessarily have side
 				// effects but no results.
@@ -459,17 +496,21 @@ func (fr *frame) evalBasicBlock(bb, incoming llvm.BasicBlock, indent string) (re
 			}
 			thenBB := inst.Operand(1)
 			elseBB := inst.Operand(2)
-			if !cond.IsConstant() {
+			if !cond.IsAInstruction().IsNil() {
 				return nil, nil, errors.New("interp: branch on a non-constant")
-			} else {
-				switch cond.ZExtValue() {
-				case 0: // false
-					return nil, []llvm.Value{thenBB}, nil // then
-				case 1: // true
-					return nil, []llvm.Value{elseBB}, nil // else
-				default:
-					panic("branch was not true or false")
-				}
+			}
+			if !cond.IsAConstantExpr().IsNil() {
+				// This may happen when the instruction builder could not
+				// const-fold some instructions.
+				return nil, nil, errors.New("interp: branch on a non-const-propagated constant expression")
+			}
+			switch cond {
+			case llvm.ConstInt(fr.Mod.Context().Int1Type(), 0, false): // false
+				return nil, []llvm.Value{thenBB}, nil // then
+			case llvm.ConstInt(fr.Mod.Context().Int1Type(), 1, false): // true
+				return nil, []llvm.Value{elseBB}, nil // else
+			default:
+				panic("branch was not true or false")
 			}
 		case !inst.IsABranchInst().IsNil() && inst.OperandsCount() == 1:
 			// unconditional branch (goto)

interp/interp.go

@@ -18,7 +18,6 @@ type Eval struct {
 	TargetData      llvm.TargetData
 	Debug           bool
 	builder         llvm.Builder
-	dibuilder       *llvm.DIBuilder
 	dirtyGlobals    map[llvm.Value]struct{}
 	sideEffectFuncs map[llvm.Value]*sideEffectResult // cache of side effect scan results
 }
@@ -38,7 +37,6 @@ func Run(mod llvm.Module, targetData llvm.TargetData, debug bool) error {
 		dirtyGlobals: map[llvm.Value]struct{}{},
 	}
 	e.builder = mod.Context().NewBuilder()
-	e.dibuilder = llvm.NewDIBuilder(mod)
 
 	initAll := mod.NamedFunction(name)
 	bb := initAll.EntryBasicBlock()
@@ -49,7 +47,6 @@ func Run(mod llvm.Module, targetData llvm.TargetData, debug bool) error {
 	e.builder.SetInsertPointBefore(bb.FirstInstruction())
 	dummy := e.builder.CreateAlloca(e.Mod.Context().Int8Type(), "dummy")
 	e.builder.SetInsertPointBefore(dummy)
-	e.builder.SetInstDebugLocation(bb.FirstInstruction())
 	var initCalls []llvm.Value
 	for inst := bb.FirstInstruction(); !inst.IsNil(); inst = llvm.NextInstruction(inst) {
 		if inst == dummy {

interp/scan.go

@@ -35,6 +35,10 @@ func (e *Eval) hasSideEffects(fn llvm.Value) *sideEffectResult {
 		return &sideEffectResult{severity: sideEffectLimited}
 	case "runtime.interfaceImplements":
 		return &sideEffectResult{severity: sideEffectNone}
+	case "runtime.trackPointer":
+		return &sideEffectResult{severity: sideEffectNone}
+	case "llvm.dbg.value":
+		return &sideEffectResult{severity: sideEffectNone}
 	}
 	if e.sideEffectFuncs == nil {
 		e.sideEffectFuncs = make(map[llvm.Value]*sideEffectResult)
@@ -107,7 +111,16 @@ func (e *Eval) hasSideEffects(fn llvm.Value) *sideEffectResult {
 				default:
 					panic("unreachable")
 				}
-			case llvm.Load, llvm.Store:
+			case llvm.Load:
+				if inst.IsVolatile() {
+					result.updateSeverity(sideEffectLimited)
+				}
+				if _, ok := e.dirtyGlobals[inst.Operand(0)]; ok {
+					if e.hasLocalSideEffects(dirtyLocals, inst) {
+						result.updateSeverity(sideEffectLimited)
+					}
+				}
+			case llvm.Store:
 				if inst.IsVolatile() {
 					result.updateSeverity(sideEffectLimited)
 				}

interp/utils.go

@@ -94,3 +94,29 @@ func isScalar(t llvm.Type) bool {
 		return false
 	}
 }
+
+// isPointerNil returns whether this is a nil pointer or not. The ok value
+// indicates whether the result is certain: if it is false the result boolean is
+// not valid.
+func isPointerNil(v llvm.Value) (result bool, ok bool) {
+	if !v.IsAConstantExpr().IsNil() {
+		switch v.Opcode() {
+		case llvm.IntToPtr:
+			// Whether a constant inttoptr is nil is easy to
+			// determine.
+			operand := v.Operand(0)
+			if operand.IsConstant() {
+				return operand.ZExtValue() == 0, true
+			}
+		case llvm.BitCast, llvm.GetElementPtr:
+			// These const instructions are just a kind of wrappers for the
+			// underlying pointer.
+			return isPointerNil(v.Operand(0))
+		}
+	}
+	if !v.IsAConstantPointerNull().IsNil() {
+		// A constant pointer null is always null, of course.
+		return true, true
+	}
+	return false, false // not valid
+}

ir/ir.go

@@ -2,7 +2,6 @@ package ir
 
 import (
 	"go/ast"
-	"go/token"
 	"go/types"
 	"sort"
 	"strings"
@@ -23,43 +22,18 @@ type Program struct {
 	mainPkg       *ssa.Package
 	Functions     []*Function
 	functionMap   map[*ssa.Function]*Function
-	Globals       []*Global
-	globalMap     map[*ssa.Global]*Global
-	comments      map[string]*ast.CommentGroup
-	NamedTypes    []*NamedType
 }
 
 // Function or method.
 type Function struct {
 	*ssa.Function
 	LLVMFn    llvm.Value
-	linkName  string // go:linkname, go:export, go:interrupt
-	exported  bool   // go:export
-	nobounds  bool   // go:nobounds
-	flag      bool   // used by dead code elimination
-	interrupt bool   // go:interrupt
-}
-
-// Global variable, possibly constant.
-type Global struct {
-	*ssa.Global
-	program    *Program
-	LLVMGlobal llvm.Value
-	linkName   string // go:extern
-	extern     bool   // go:extern
-}
-
-// Type with a name and possibly methods.
-type NamedType struct {
-	*ssa.Type
-	LLVMType llvm.Type
-}
-
-// Type that is at some point put in an interface.
-type TypeWithMethods struct {
-	t       types.Type
-	Num     int
-	Methods map[string]*types.Selection
+	linkName  string     // go:linkname, go:export, go:interrupt
+	exported  bool       // go:export
+	nobounds  bool       // go:nobounds
+	flag      bool       // used by dead code elimination
+	interrupt bool       // go:interrupt
+	inline    InlineType // go:inline
 }
 
 // Interface type that is at some point used in a type assert (to check whether
@@ -69,37 +43,27 @@ type Interface struct {
 	Type *types.Interface
 }
 
-// Create and intialize a new *Program from a *ssa.Program.
-func NewProgram(lprogram *loader.Program, mainPath string) *Program {
-	comments := map[string]*ast.CommentGroup{}
-	for _, pkgInfo := range lprogram.Sorted() {
-		for _, file := range pkgInfo.Files {
-			for _, decl := range file.Decls {
-				switch decl := decl.(type) {
-				case *ast.GenDecl:
-					switch decl.Tok {
-					case token.TYPE, token.VAR:
-						if len(decl.Specs) != 1 {
-							continue
-						}
-						for _, spec := range decl.Specs {
-							switch spec := spec.(type) {
-							case *ast.TypeSpec: // decl.Tok == token.TYPE
-								id := pkgInfo.Pkg.Path() + "." + spec.Name.Name
-								comments[id] = decl.Doc
-							case *ast.ValueSpec: // decl.Tok == token.VAR
-								for _, name := range spec.Names {
-									id := pkgInfo.Pkg.Path() + "." + name.Name
-									comments[id] = decl.Doc
-								}
-							}
-						}
-					}
-				}
-			}
-		}
-	}
+type InlineType int
 
+// How much to inline.
+const (
+	// Default behavior. The compiler decides for itself whether any given
+	// function will be inlined. Whether any function is inlined depends on the
+	// optimization level.
+	InlineDefault InlineType = iota
+
+	// Inline hint, just like the C inline keyword (signalled using
+	// //go:inline). The compiler will be more likely to inline this function,
+	// but it is not a guarantee.
+	InlineHint
+
+	// Don't inline, just like the GCC noinline attribute. Signalled using
+	// //go:noinline.
+	InlineNone
+)
+
+// Create and initialize a new *Program from a *ssa.Program.
+func NewProgram(lprogram *loader.Program, mainPath string) *Program {
 	program := lprogram.LoadSSA()
 	program.Build()
 
@@ -171,8 +135,6 @@ func NewProgram(lprogram *loader.Program, mainPath string) *Program {
 		LoaderProgram: lprogram,
 		mainPkg:       mainPkg,
 		functionMap:   make(map[*ssa.Function]*Function),
-		globalMap:     make(map[*ssa.Global]*Global),
-		comments:      comments,
 	}
 
 	for _, pkg := range packageList {
@@ -197,8 +159,6 @@ func (p *Program) AddPackage(pkg *ssa.Package) {
 		case *ssa.Function:
 			p.addFunction(member)
 		case *ssa.Type:
-			t := &NamedType{Type: member}
-			p.NamedTypes = append(p.NamedTypes, t)
 			methods := getAllMethods(pkg.Prog, member.Type())
 			if !types.IsInterface(member.Type()) {
 				// named type
@@ -207,13 +167,7 @@ func (p *Program) AddPackage(pkg *ssa.Package) {
 				}
 			}
 		case *ssa.Global:
-			g := &Global{program: p, Global: member}
-			doc := p.comments[g.RelString(nil)]
-			if doc != nil {
-				g.parsePragmas(doc)
-			}
-			p.Globals = append(p.Globals, g)
-			p.globalMap[member] = g
+			// Ignore. Globals are not handled here.
 		case *ssa.NamedConst:
 			// Ignore: these are already resolved.
 		default:
@@ -247,10 +201,6 @@ func (p *Program) GetFunction(ssaFn *ssa.Function) *Function {
 	return p.functionMap[ssaFn]
 }
 
-func (p *Program) GetGlobal(ssaGlobal *ssa.Global) *Global {
-	return p.globalMap[ssaGlobal]
-}
-
 func (p *Program) MainPkg() *ssa.Package {
 	return p.mainPkg
 }
@@ -279,6 +229,10 @@ func (f *Function) parsePragmas() {
 				}
 				f.linkName = parts[1]
 				f.exported = true
+			case "//go:inline":
+				f.inline = InlineHint
+			case "//go:noinline":
+				f.inline = InlineNone
 			case "//go:interrupt":
 				if len(parts) != 2 {
 					continue
@@ -332,6 +286,11 @@ func (f *Function) IsInterrupt() bool {
 	return f.interrupt
 }
 
+// Return the inline directive of this function.
+func (f *Function) Inline() InlineType {
+	return f.inline
+}
+
 // Return the link name for this function.
 func (f *Function) LinkName() string {
 	if f.linkName != "" {
@@ -366,72 +325,6 @@ func (f *Function) CName() string {
 	return ""
 }
 
-// Parse //go: pragma comments from the source.
-func (g *Global) parsePragmas(doc *ast.CommentGroup) {
-	for _, comment := range doc.List {
-		if !strings.HasPrefix(comment.Text, "//go:") {
-			continue
-		}
-		parts := strings.Fields(comment.Text)
-		switch parts[0] {
-		case "//go:extern":
-			g.extern = true
-			if len(parts) == 2 {
-				g.linkName = parts[1]
-			}
-		}
-	}
-}
-
-// Return the link name for this global.
-func (g *Global) LinkName() string {
-	if g.linkName != "" {
-		return g.linkName
-	}
-	if name := g.CName(); name != "" {
-		return name
-	}
-	return g.RelString(nil)
-}
-
-func (g *Global) IsExtern() bool {
-	return g.extern || g.CName() != ""
-}
-
-// Return the name of the C global if this is a CGo wrapper. Otherwise, return a
-// zero-length string.
-func (g *Global) CName() string {
-	name := g.Name()
-	if strings.HasPrefix(name, "C.") {
-		// created by ../loader/cgo.go
-		return name[2:]
-	}
-	return ""
-}
-
-// Return true if this named type is annotated with the //go:volatile pragma,
-// for volatile loads and stores.
-func (p *Program) IsVolatile(t types.Type) bool {
-	if t, ok := t.(*types.Named); !ok {
-		return false
-	} else {
-		if t.Obj().Pkg() == nil {
-			return false
-		}
-		id := t.Obj().Pkg().Path() + "." + t.Obj().Name()
-		doc := p.comments[id]
-		if doc == nil {
-			return false
-		}
-		for _, line := range doc.List {
-			if strings.TrimSpace(line.Text) == "//go:volatile" {
-				return true
-			}
-		}
-		return false
-	}
-}
-
 // Get all methods of a type.
 func getAllMethods(prog *ssa.Program, typ types.Type) []*types.Selection {
 	ms := prog.MethodSets.MethodSet(typ)

linker-builtin.go

@@ -24,7 +24,7 @@ import "C"
 // This version uses the built-in linker when trying to use lld.
 func Link(linker string, flags ...string) error {
 	switch linker {
-	case "ld.lld", commands["ld.lld"]:
+	case "ld.lld":
 		flags = append([]string{"tinygo:" + linker}, flags...)
 		var cflag *C.char
 		buf := C.calloc(C.size_t(len(flags)), C.size_t(unsafe.Sizeof(cflag)))
@@ -39,7 +39,7 @@ func Link(linker string, flags ...string) error {
 			return errors.New("failed to link using built-in ld.lld")
 		}
 		return nil
-	case "wasm-ld", commands["wasm-ld"]:
+	case "wasm-ld":
 		flags = append([]string{"tinygo:" + linker}, flags...)
 		var cflag *C.char
 		buf := C.calloc(C.size_t(len(flags)), C.size_t(unsafe.Sizeof(cflag)))
@@ -57,9 +57,13 @@ func Link(linker string, flags ...string) error {
 		return nil
 	default:
 		// Fall back to external command.
+		if cmdNames, ok := commands[linker]; ok {
+			return execCommand(cmdNames, flags...)
+		}
 		cmd := exec.Command(linker, flags...)
 		cmd.Stdout = os.Stdout
 		cmd.Stderr = os.Stderr
+		cmd.Dir = sourceDir()
 		return cmd.Run()
 	}
 }

linker-external.go

@@ -14,8 +14,12 @@ import (
 //
 // This version always runs the linker as an external command.
 func Link(linker string, flags ...string) error {
+	if cmdNames, ok := commands[linker]; ok {
+		return execCommand(cmdNames, flags...)
+	}
 	cmd := exec.Command(linker, flags...)
 	cmd.Stdout = os.Stdout
 	cmd.Stderr = os.Stderr
+	cmd.Dir = sourceDir()
 	return cmd.Run()
 }

loader/cgo.go

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

loader/libclang.go

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

loader/loader.go

@@ -1,6 +1,7 @@
 package loader
 
 import (
+	"bytes"
 	"errors"
 	"go/ast"
 	"go/build"
@@ -10,19 +11,26 @@ import (
 	"os"
 	"path/filepath"
 	"sort"
+	"strings"
+	"text/template"
+
+	"github.com/tinygo-org/tinygo/cgo"
 )
 
 // Program holds all packages and some metadata about the program as a whole.
 type Program struct {
-	Build         *build.Context
-	OverlayBuild  *build.Context
-	ShouldOverlay func(path string) bool
-	Packages      map[string]*Package
-	sorted        []*Package
-	fset          *token.FileSet
-	TypeChecker   types.Config
-	Dir           string // current working directory (for error reporting)
-	CFlags        []string
+	mainPkg      string
+	Build        *build.Context
+	OverlayBuild *build.Context
+	OverlayPath  func(path string) string
+	Packages     map[string]*Package
+	sorted       []*Package
+	fset         *token.FileSet
+	TypeChecker  types.Config
+	Dir          string // current working directory (for error reporting)
+	TINYGOROOT   string // root of the TinyGo installation or root of the source code
+	CFlags       []string
+	ClangHeaders string
 }
 
 // Package holds a loaded package, its imports, and its parsed files.
@@ -45,8 +53,9 @@ func (p *Program) Import(path, srcDir string) (*Package, error) {
 
 	// Load this package.
 	ctx := p.Build
-	if p.ShouldOverlay(path) {
+	if newPath := p.OverlayPath(path); newPath != "" {
 		ctx = p.OverlayBuild
+		path = newPath
 	}
 	buildPkg, err := ctx.Import(path, srcDir, build.ImportComment)
 	if err != nil {
@@ -59,6 +68,11 @@ func (p *Program) Import(path, srcDir string) (*Package, error) {
 	p.sorted = nil // invalidate the sorted order of packages
 	pkg := p.newPackage(buildPkg)
 	p.Packages[buildPkg.ImportPath] = pkg
+
+	if p.mainPkg == "" {
+		p.mainPkg = buildPkg.ImportPath
+	}
+
 	return pkg, nil
 }
 
@@ -88,6 +102,11 @@ func (p *Program) ImportFile(path string) (*Package, error) {
 	p.sorted = nil // invalidate the sorted order of packages
 	pkg := p.newPackage(buildPkg)
 	p.Packages[buildPkg.ImportPath] = pkg
+
+	if p.mainPkg == "" {
+		p.mainPkg = buildPkg.ImportPath
+	}
+
 	return pkg, nil
 }
 
@@ -166,10 +185,12 @@ func (p *Program) sort() {
 // The returned error may be an Errors error, which contains a list of errors.
 //
 // Idempotent.
-func (p *Program) Parse() error {
+func (p *Program) Parse(compileTestBinary bool) error {
+	includeTests := compileTestBinary
+
 	// Load all imports
 	for _, pkg := range p.Sorted() {
-		err := pkg.importRecursively()
+		err := pkg.importRecursively(includeTests)
 		if err != nil {
 			if err, ok := err.(*ImportCycleError); ok {
 				if pkg.ImportPath != err.Packages[0] {
@@ -182,7 +203,14 @@ func (p *Program) Parse() error {
 
 	// Parse all packages.
 	for _, pkg := range p.Sorted() {
-		err := pkg.Parse()
+		err := pkg.Parse(includeTests)
+		if err != nil {
+			return err
+		}
+	}
+
+	if compileTestBinary {
+		err := p.SwapTestMain()
 		if err != nil {
 			return err
 		}
@@ -199,6 +227,83 @@ func (p *Program) Parse() error {
 	return nil
 }
 
+func (p *Program) SwapTestMain() error {
+	var tests []string
+
+	isTestFunc := func(f *ast.FuncDecl) bool {
+		// TODO: improve signature check
+		if strings.HasPrefix(f.Name.Name, "Test") && f.Name.Name != "TestMain" {
+			return true
+		}
+		return false
+	}
+	mainPkg := p.Packages[p.mainPkg]
+	for _, f := range mainPkg.Files {
+		for i, d := range f.Decls {
+			switch v := d.(type) {
+			case *ast.FuncDecl:
+				if isTestFunc(v) {
+					tests = append(tests, v.Name.Name)
+				}
+				if v.Name.Name == "main" {
+					// Remove main
+					if len(f.Decls) == 1 {
+						f.Decls = make([]ast.Decl, 0)
+					} else {
+						f.Decls[i] = f.Decls[len(f.Decls)-1]
+						f.Decls = f.Decls[:len(f.Decls)-1]
+					}
+				}
+			}
+		}
+	}
+
+	// TODO: Check if they defined a TestMain and call it instead of testing.TestMain
+	const mainBody = `package main
+
+import (
+	"testing"
+)
+
+func main () {
+	m := &testing.M{
+		Tests: []testing.TestToCall{
+{{range .TestFunctions}}
+			{Name: "{{.}}", Func: {{.}}},
+{{end}}
+		},
+	}
+
+	testing.TestMain(m)
+}
+`
+	tmpl := template.Must(template.New("testmain").Parse(mainBody))
+	b := bytes.Buffer{}
+	tmplData := struct {
+		TestFunctions []string
+	}{
+		TestFunctions: tests,
+	}
+
+	err := tmpl.Execute(&b, tmplData)
+	if err != nil {
+		return err
+	}
+	path := filepath.Join(p.mainPkg, "$testmain.go")
+
+	if p.fset == nil {
+		p.fset = token.NewFileSet()
+	}
+
+	newMain, err := parser.ParseFile(p.fset, path, b.Bytes(), parser.AllErrors)
+	if err != nil {
+		return err
+	}
+	mainPkg.Files = append(mainPkg.Files, newMain)
+
+	return nil
+}
+
 // parseFile is a wrapper around parser.ParseFile.
 func (p *Program) parseFile(path string, mode parser.Mode) (*ast.File, error) {
 	if p.fset == nil {
@@ -223,7 +328,7 @@ func (p *Program) parseFile(path string, mode parser.Mode) (*ast.File, error) {
 // Parse parses and typechecks this package.
 //
 // Idempotent.
-func (p *Package) Parse() error {
+func (p *Package) Parse(includeTests bool) error {
 	if len(p.Files) != 0 {
 		return nil
 	}
@@ -237,7 +342,7 @@ func (p *Package) Parse() error {
 		return nil
 	}
 
-	files, err := p.parseFiles()
+	files, err := p.parseFiles(includeTests)
 	if err != nil {
 		return err
 	}
@@ -276,11 +381,21 @@ func (p *Package) Check() error {
 }
 
 // parseFiles parses the loaded list of files and returns this list.
-func (p *Package) parseFiles() ([]*ast.File, error) {
+func (p *Package) parseFiles(includeTests bool) ([]*ast.File, error) {
 	// TODO: do this concurrently.
 	var files []*ast.File
 	var fileErrs []error
-	for _, file := range p.GoFiles {
+
+	var gofiles []string
+	if includeTests {
+		gofiles = make([]string, 0, len(p.GoFiles)+len(p.TestGoFiles))
+		gofiles = append(gofiles, p.GoFiles...)
+		gofiles = append(gofiles, p.TestGoFiles...)
+	} else {
+		gofiles = p.GoFiles
+	}
+
+	for _, file := range gofiles {
 		f, err := p.parseFile(filepath.Join(p.Package.Dir, file), parser.ParseComments)
 		if err != nil {
 			fileErrs = append(fileErrs, err)
@@ -299,16 +414,23 @@ func (p *Package) parseFiles() ([]*ast.File, error) {
 			fileErrs = append(fileErrs, err)
 			continue
 		}
-		errs := p.processCgo(path, f, append(p.CFlags, "-I"+p.Package.Dir))
+		files = append(files, f)
+	}
+	if len(p.CgoFiles) != 0 {
+		cflags := append(p.CFlags, "-I"+p.Package.Dir)
+		if p.ClangHeaders != "" {
+			cflags = append(cflags, "-I"+p.ClangHeaders)
+		}
+		generated, errs := cgo.Process(files, p.Program.Dir, p.fset, cflags)
 		if errs != nil {
 			fileErrs = append(fileErrs, errs...)
-			continue
 		}
-		files = append(files, f)
+		files = append(files, generated)
 	}
 	if len(fileErrs) != 0 {
 		return nil, Errors{p, fileErrs}
 	}
+
 	return files, nil
 }
 
@@ -329,11 +451,17 @@ func (p *Package) Import(to string) (*types.Package, error) {
 // importRecursively() on the imported packages as well.
 //
 // Idempotent.
-func (p *Package) importRecursively() error {
+func (p *Package) importRecursively(includeTests bool) error {
 	p.Importing = true
-	for _, to := range p.Package.Imports {
+
+	imports := p.Package.Imports
+	if includeTests {
+		imports = append(imports, p.Package.TestImports...)
+	}
+
+	for _, to := range imports {
 		if to == "C" {
-			// Do Cgo processing in a later stage.
+			// Do CGo processing in a later stage.
 			continue
 		}
 		if _, ok := p.Imports[to]; ok {
@@ -349,7 +477,7 @@ func (p *Package) importRecursively() error {
 		if importedPkg.Importing {
 			return &ImportCycleError{[]string{p.ImportPath, importedPkg.ImportPath}, p.ImportPos[to]}
 		}
-		err = importedPkg.importRecursively()
+		err = importedPkg.importRecursively(false)
 		if err != nil {
 			if err, ok := err.(*ImportCycleError); ok {
 				err.Packages = append([]string{p.ImportPath}, err.Packages...)

main.go

@@ -33,16 +33,30 @@ func (e *commandError) Error() string {
 	return e.Msg + " " + e.File + ": " + e.Err.Error()
 }
 
+// multiError is a list of multiple errors (actually: diagnostics) returned
+// during LLVM IR generation.
+type multiError struct {
+	Errs []error
+}
+
+func (e *multiError) Error() string {
+	return e.Errs[0].Error()
+}
+
 type BuildConfig struct {
-	opt        string
-	gc         string
-	printIR    bool
-	dumpSSA    bool
-	debug      bool
-	printSizes string
-	cFlags     []string
-	ldFlags    []string
-	wasmAbi    string
+	opt           string
+	gc            string
+	panicStrategy string
+	printIR       bool
+	dumpSSA       bool
+	debug         bool
+	printSizes    string
+	cFlags        []string
+	ldFlags       []string
+	tags          string
+	wasmAbi       string
+	heapSize      int64
+	testConfig    compiler.TestConfig
 }
 
 // Helper function for Compiler object.
@@ -51,23 +65,56 @@ func Compile(pkgName, outpath string, spec *TargetSpec, config *BuildConfig, act
 		config.gc = spec.GC
 	}
 
-	// Append command line passed CFlags and LDFlags
-	spec.CFlags = append(spec.CFlags, config.cFlags...)
-	spec.LDFlags = append(spec.LDFlags, config.ldFlags...)
+	root := sourceDir()
 
+	// Merge and adjust CFlags.
+	cflags := append([]string{}, config.cFlags...)
+	for _, flag := range spec.CFlags {
+		cflags = append(cflags, strings.Replace(flag, "{root}", root, -1))
+	}
+
+	// Merge and adjust LDFlags.
+	ldflags := append([]string{}, config.ldFlags...)
+	for _, flag := range spec.LDFlags {
+		ldflags = append(ldflags, strings.Replace(flag, "{root}", root, -1))
+	}
+
+	goroot := getGoroot()
+	if goroot == "" {
+		return errors.New("cannot locate $GOROOT, please set it manually")
+	}
+	tags := spec.BuildTags
+	major, minor, err := getGorootVersion(goroot)
+	if err != nil {
+		return fmt.Errorf("could not read version from GOROOT (%v): %v", goroot, err)
+	}
+	if major != 1 {
+		return fmt.Errorf("expected major version 1, got go%d.%d", major, minor)
+	}
+	for i := 1; i <= minor; i++ {
+		tags = append(tags, fmt.Sprintf("go1.%d", i))
+	}
+	if extraTags := strings.Fields(config.tags); len(extraTags) != 0 {
+		tags = append(tags, extraTags...)
+	}
 	compilerConfig := compiler.Config{
-		Triple:    spec.Triple,
-		CPU:       spec.CPU,
-		GOOS:      spec.GOOS,
-		GOARCH:    spec.GOARCH,
-		GC:        config.gc,
-		CFlags:    spec.CFlags,
-		LDFlags:   spec.LDFlags,
-		Debug:     config.debug,
-		DumpSSA:   config.dumpSSA,
-		RootDir:   sourceDir(),
-		GOPATH:    getGopath(),
-		BuildTags: spec.BuildTags,
+		Triple:        spec.Triple,
+		CPU:           spec.CPU,
+		Features:      spec.Features,
+		GOOS:          spec.GOOS,
+		GOARCH:        spec.GOARCH,
+		GC:            config.gc,
+		PanicStrategy: config.panicStrategy,
+		CFlags:        cflags,
+		LDFlags:       ldflags,
+		ClangHeaders:  getClangHeaderPath(root),
+		Debug:         config.debug,
+		DumpSSA:       config.dumpSSA,
+		TINYGOROOT:    root,
+		GOROOT:        goroot,
+		GOPATH:        getGopath(),
+		BuildTags:     tags,
+		TestConfig:    config.testConfig,
 	}
 	c, err := compiler.NewCompiler(pkgName, compilerConfig)
 	if err != nil {
@@ -75,12 +122,15 @@ func Compile(pkgName, outpath string, spec *TargetSpec, config *BuildConfig, act
 	}
 
 	// Compile Go code to IR.
-	err = c.Compile(pkgName)
-	if err != nil {
-		return err
+	errs := c.Compile(pkgName)
+	if len(errs) != 0 {
+		if len(errs) == 1 {
+			return errs[0]
+		}
+		return &multiError{errs}
 	}
 	if config.printIR {
-		fmt.Println("Generated LLVM IR:")
+		fmt.Println("; Generated LLVM IR:")
 		fmt.Println(c.IR())
 	}
 	if err := c.Verify(); err != nil {
@@ -191,19 +241,26 @@ func Compile(pkgName, outpath string, spec *TargetSpec, config *BuildConfig, act
 		// Prepare link command.
 		executable := filepath.Join(dir, "main")
 		tmppath := executable // final file
-		ldflags := append(spec.LDFlags, "-o", executable, objfile, "-L", sourceDir())
+		ldflags = append(ldflags, "-o", executable, objfile, "-L", root)
 		if spec.RTLib == "compiler-rt" {
 			ldflags = append(ldflags, librt)
 		}
+		if spec.GOARCH == "wasm" {
+			// Round heap size to next multiple of 65536 (the WebAssembly page
+			// size).
+			heapSize := (config.heapSize + (65536 - 1)) &^ (65536 - 1)
+			ldflags = append(ldflags, "--initial-memory="+strconv.FormatInt(heapSize, 10))
+		}
 
 		// Compile extra files.
 		for i, path := range spec.ExtraFiles {
+			abspath := filepath.Join(root, path)
 			outpath := filepath.Join(dir, "extra-"+strconv.Itoa(i)+"-"+filepath.Base(path)+".o")
-			cmd := exec.Command(spec.Compiler, append(spec.CFlags, "-c", "-o", outpath, path)...)
-			cmd.Stdout = os.Stdout
-			cmd.Stderr = os.Stderr
-			cmd.Dir = sourceDir()
-			err := cmd.Run()
+			cmdNames := []string{spec.Compiler}
+			if names, ok := commands[spec.Compiler]; ok {
+				cmdNames = names
+			}
+			err := execCommand(cmdNames, append(cflags, "-c", "-o", outpath, abspath)...)
 			if err != nil {
 				return &commandError{"failed to build", path, err}
 			}
@@ -215,11 +272,11 @@ func Compile(pkgName, outpath string, spec *TargetSpec, config *BuildConfig, act
 			for _, file := range pkg.CFiles {
 				path := filepath.Join(pkg.Package.Dir, file)
 				outpath := filepath.Join(dir, "pkg"+strconv.Itoa(i)+"-"+file+".o")
-				cmd := exec.Command(spec.Compiler, append(spec.CFlags, "-c", "-o", outpath, path)...)
-				cmd.Stdout = os.Stdout
-				cmd.Stderr = os.Stderr
-				cmd.Dir = sourceDir()
-				err := cmd.Run()
+				cmdNames := []string{spec.Compiler}
+				if names, ok := commands[spec.Compiler]; ok {
+					cmdNames = names
+				}
+				err := execCommand(cmdNames, append(cflags, "-c", "-o", outpath, path)...)
 				if err != nil {
 					return &commandError{"failed to build", path, err}
 				}
@@ -228,11 +285,7 @@ func Compile(pkgName, outpath string, spec *TargetSpec, config *BuildConfig, act
 		}
 
 		// Link the object files together.
-		if linker, ok := commands[spec.Linker]; ok {
-			err = Link(linker, ldflags...)
-		} else {
-			err = Link(spec.Linker, ldflags...)
-		}
+		err = Link(spec.Linker, ldflags...)
 		if err != nil {
 			return &commandError{"failed to link", executable, err}
 		}
@@ -309,6 +362,33 @@ func Build(pkgName, outpath, target string, config *BuildConfig) error {
 	})
 }
 
+func Test(pkgName, target string, config *BuildConfig) error {
+	spec, err := LoadTarget(target)
+	if err != nil {
+		return err
+	}
+
+	spec.BuildTags = append(spec.BuildTags, "test")
+	config.testConfig.CompileTestBinary = true
+	return Compile(pkgName, ".elf", spec, config, func(tmppath string) error {
+		cmd := exec.Command(tmppath)
+		cmd.Stdout = os.Stdout
+		cmd.Stderr = os.Stderr
+		err := cmd.Run()
+		if err != nil {
+			// Propagate the exit code
+			if err, ok := err.(*exec.ExitError); ok {
+				if status, ok := err.Sys().(syscall.WaitStatus); ok {
+					os.Exit(status.ExitStatus())
+				}
+				os.Exit(1)
+			}
+			return &commandError{"failed to run compiled binary", tmppath, err}
+		}
+		return nil
+	})
+}
+
 func Flash(pkgName, target, port string, config *BuildConfig) error {
 	spec, err := LoadTarget(target)
 	if err != nil {
@@ -467,6 +547,30 @@ func Run(pkgName, target string, config *BuildConfig) error {
 	})
 }
 
+// parseSize converts a human-readable size (with k/m/g suffix) into a plain
+// number.
+func parseSize(s string) (int64, error) {
+	s = strings.ToLower(strings.TrimSpace(s))
+	if len(s) == 0 {
+		return 0, errors.New("no size provided")
+	}
+	multiply := int64(1)
+	switch s[len(s)-1] {
+	case 'k':
+		multiply = 1 << 10
+	case 'm':
+		multiply = 1 << 20
+	case 'g':
+		multiply = 1 << 30
+	}
+	if multiply != 1 {
+		s = s[:len(s)-1]
+	}
+	n, err := strconv.ParseInt(s, 0, 64)
+	n *= multiply
+	return n, err
+}
+
 func usage() {
 	fmt.Fprintln(os.Stderr, "TinyGo is a Go compiler for small places.")
 	fmt.Fprintln(os.Stderr, "version:", version)
@@ -474,6 +578,7 @@ func usage() {
 	fmt.Fprintln(os.Stderr, "\ncommands:")
 	fmt.Fprintln(os.Stderr, "  build: compile packages and dependencies")
 	fmt.Fprintln(os.Stderr, "  run:   compile and run immediately")
+	fmt.Fprintln(os.Stderr, "  test:  test packages")
 	fmt.Fprintln(os.Stderr, "  flash: compile and flash to the device")
 	fmt.Fprintln(os.Stderr, "  gdb:   run/flash and immediately enter GDB")
 	fmt.Fprintln(os.Stderr, "  clean: empty cache directory ("+cacheDir()+")")
@@ -484,19 +589,24 @@ func usage() {
 
 func handleCompilerError(err error) {
 	if err != nil {
-		if errUnsupported, ok := err.(*interp.Unsupported); ok {
+		switch err := err.(type) {
+		case *interp.Unsupported:
 			// hit an unknown/unsupported instruction
 			fmt.Fprintln(os.Stderr, "unsupported instruction during init evaluation:")
-			errUnsupported.Inst.Dump()
+			err.Inst.Dump()
 			fmt.Fprintln(os.Stderr)
-		} else if errCompiler, ok := err.(types.Error); ok {
-			fmt.Fprintln(os.Stderr, errCompiler)
-		} else if errLoader, ok := err.(loader.Errors); ok {
-			fmt.Fprintln(os.Stderr, "#", errLoader.Pkg.ImportPath)
-			for _, err := range errLoader.Errs {
+		case types.Error:
+			fmt.Fprintln(os.Stderr, err)
+		case loader.Errors:
+			fmt.Fprintln(os.Stderr, "#", err.Pkg.ImportPath)
+			for _, err := range err.Errs {
 				fmt.Fprintln(os.Stderr, err)
 			}
-		} else {
+		case *multiError:
+			for _, err := range err.Errs {
+				fmt.Fprintln(os.Stderr, err)
+			}
+		default:
 			fmt.Fprintln(os.Stderr, "error:", err)
 		}
 		os.Exit(1)
@@ -506,10 +616,12 @@ func handleCompilerError(err error) {
 func main() {
 	outpath := flag.String("o", "", "output filename")
 	opt := flag.String("opt", "z", "optimization level: 0, 1, 2, s, z")
-	gc := flag.String("gc", "", "garbage collector to use (none, dumb, marksweep)")
+	gc := flag.String("gc", "", "garbage collector to use (none, leaking, conservative)")
+	panicStrategy := flag.String("panic", "print", "panic strategy (abort, trap)")
 	printIR := flag.Bool("printir", false, "print LLVM IR")
 	dumpSSA := flag.Bool("dumpssa", false, "dump internal Go SSA")
-	target := flag.String("target", "", "LLVM target")
+	tags := flag.String("tags", "", "a space-separated list of extra build tags")
+	target := flag.String("target", "", "LLVM target | .json file with TargetSpec")
 	printSize := flag.String("size", "", "print sizes (none, short, full)")
 	nodebug := flag.Bool("no-debug", false, "disable DWARF debug symbol generation")
 	ocdOutput := flag.Bool("ocd-output", false, "print OCD daemon output during debug")
@@ -517,6 +629,7 @@ func main() {
 	cFlags := flag.String("cflags", "", "additional cflags for compiler")
 	ldFlags := flag.String("ldflags", "", "additional ldflags for linker")
 	wasmAbi := flag.String("wasm-abi", "js", "WebAssembly ABI conventions: js (no i64 params) or generic")
+	heapSize := flag.String("heap-size", "1M", "default heap size in bytes (only supported by WebAssembly)")
 
 	if len(os.Args) < 2 {
 		fmt.Fprintln(os.Stderr, "No command-line arguments supplied.")
@@ -527,13 +640,15 @@ func main() {
 
 	flag.CommandLine.Parse(os.Args[2:])
 	config := &BuildConfig{
-		opt:        *opt,
-		gc:         *gc,
-		printIR:    *printIR,
-		dumpSSA:    *dumpSSA,
-		debug:      !*nodebug,
-		printSizes: *printSize,
-		wasmAbi:    *wasmAbi,
+		opt:           *opt,
+		gc:            *gc,
+		panicStrategy: *panicStrategy,
+		printIR:       *printIR,
+		dumpSSA:       *dumpSSA,
+		debug:         !*nodebug,
+		printSizes:    *printSize,
+		tags:          *tags,
+		wasmAbi:       *wasmAbi,
 	}
 
 	if *cFlags != "" {
@@ -544,6 +659,19 @@ func main() {
 		config.ldFlags = strings.Split(*ldFlags, " ")
 	}
 
+	if *panicStrategy != "print" && *panicStrategy != "trap" {
+		fmt.Fprintln(os.Stderr, "Panic strategy must be either print or trap.")
+		usage()
+		os.Exit(1)
+	}
+
+	var err error
+	if config.heapSize, err = parseSize(*heapSize); err != nil {
+		fmt.Fprintln(os.Stderr, "Could not read heap size:", *heapSize)
+		usage()
+		os.Exit(1)
+	}
+
 	os.Setenv("CC", "clang -target="+*target)
 
 	switch command {
@@ -553,8 +681,11 @@ func main() {
 			usage()
 			os.Exit(1)
 		}
-		if flag.NArg() != 1 {
-			fmt.Fprintln(os.Stderr, "No package specified.")
+		pkgName := "."
+		if flag.NArg() == 1 {
+			pkgName = flag.Arg(0)
+		} else if flag.NArg() > 1 {
+			fmt.Fprintln(os.Stderr, "build only accepts a single positional argument: package name, but multiple were specified")
 			usage()
 			os.Exit(1)
 		}
@@ -562,7 +693,7 @@ func main() {
 		if target == "" && filepath.Ext(*outpath) == ".wasm" {
 			target = "wasm"
 		}
-		err := Build(flag.Arg(0), *outpath, target, config)
+		err := Build(pkgName, *outpath, target, config)
 		handleCompilerError(err)
 	case "build-builtins":
 		// Note: this command is only meant to be used while making a release!
@@ -604,6 +735,17 @@ func main() {
 		}
 		err := Run(flag.Arg(0), *target, config)
 		handleCompilerError(err)
+	case "test":
+		pkgName := "."
+		if flag.NArg() == 1 {
+			pkgName = 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()
+			os.Exit(1)
+		}
+		err := Test(pkgName, *target, config)
+		handleCompilerError(err)
 	case "clean":
 		// remove cache directory
 		dir := cacheDir()

main_test.go

@@ -13,17 +13,19 @@ import (
 	"runtime"
 	"sort"
 	"testing"
+
+	"github.com/tinygo-org/tinygo/loader"
 )
 
 const TESTDATA = "testdata"
 
 func TestCompiler(t *testing.T) {
-	matches, err := filepath.Glob(TESTDATA + "/*.go")
+	matches, err := filepath.Glob(filepath.Join(TESTDATA, "*.go"))
 	if err != nil {
 		t.Fatal("could not read test files:", err)
 	}
 
-	dirMatches, err := filepath.Glob(TESTDATA + "/*/main.go")
+	dirMatches, err := filepath.Glob(filepath.Join(TESTDATA, "*", "main.go"))
 	if err != nil {
 		t.Fatal("could not read test packages:", err)
 	}
@@ -64,7 +66,7 @@ func TestCompiler(t *testing.T) {
 	if runtime.GOOS == "linux" {
 		t.Log("running tests for linux/arm...")
 		for _, path := range matches {
-			if path == "testdata/cgo/" {
+			if path == filepath.Join("testdata", "cgo")+string(filepath.Separator) {
 				continue // TODO: improve CGo
 			}
 			t.Run(path, func(t *testing.T) {
@@ -74,7 +76,7 @@ func TestCompiler(t *testing.T) {
 
 		t.Log("running tests for linux/arm64...")
 		for _, path := range matches {
-			if path == "testdata/cgo/" {
+			if path == filepath.Join("testdata", "cgo")+string(filepath.Separator) {
 				continue // TODO: improve CGo
 			}
 			t.Run(path, func(t *testing.T) {
@@ -84,7 +86,7 @@ func TestCompiler(t *testing.T) {
 
 		t.Log("running tests for WebAssembly...")
 		for _, path := range matches {
-			if path == "testdata/gc.go" {
+			if path == filepath.Join("testdata", "gc.go") {
 				continue // known to fail
 			}
 			t.Run(path, func(t *testing.T) {
@@ -97,7 +99,7 @@ func TestCompiler(t *testing.T) {
 func runTest(path, tmpdir string, target string, t *testing.T) {
 	// Get the expected output for this test.
 	txtpath := path[:len(path)-3] + ".txt"
-	if path[len(path)-1] == '/' {
+	if path[len(path)-1] == os.PathSeparator {
 		txtpath = path + "out.txt"
 	}
 	f, err := os.Open(txtpath)
@@ -121,7 +123,13 @@ func runTest(path, tmpdir string, target string, t *testing.T) {
 	binary := filepath.Join(tmpdir, "test")
 	err = Build("./"+path, binary, target, config)
 	if err != nil {
-		t.Log("failed to build:", err)
+		if errLoader, ok := err.(loader.Errors); ok {
+			for _, err := range errLoader.Errs {
+				t.Log("failed to build:", err)
+			}
+		} else {
+			t.Log("failed to build:", err)
+		}
 		t.Fail()
 		return
 	}

src/device/arm/arm.go

@@ -30,6 +30,7 @@
 package arm
 
 import (
+	"runtime/volatile"
 	"unsafe"
 )
 
@@ -69,9 +70,6 @@ func SVCall3(num uintptr, a1, a2, a3 interface{}) uintptr
 // Run the following system call (SVCall) with 4 arguments.
 func SVCall4(num uintptr, a1, a2, a3, a4 interface{}) uintptr
 
-//go:volatile
-type RegValue uint32
-
 const (
 	SCS_BASE  = 0xE000E000
 	NVIC_BASE = SCS_BASE + 0x0100
@@ -82,24 +80,24 @@ const (
 // Source:
 // http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.dui0553a/CIHIGCIF.html
 type NVIC_Type struct {
-	ISER [8]RegValue // Interrupt Set-enable Registers
+	ISER [8]volatile.Register32 // Interrupt Set-enable Registers
 	_    [24]uint32
-	ICER [8]RegValue // Interrupt Clear-enable Registers
+	ICER [8]volatile.Register32 // Interrupt Clear-enable Registers
 	_    [24]uint32
-	ISPR [8]RegValue // Interrupt Set-pending Registers
+	ISPR [8]volatile.Register32 // Interrupt Set-pending Registers
 	_    [24]uint32
-	ICPR [8]RegValue // Interrupt Clear-pending Registers
+	ICPR [8]volatile.Register32 // Interrupt Clear-pending Registers
 	_    [24]uint32
-	IABR [8]RegValue // Interrupt Active Bit Registers
+	IABR [8]volatile.Register32 // Interrupt Active Bit Registers
 	_    [56]uint32
-	IPR  [60]RegValue // Interrupt Priority Registers
+	IPR  [60]volatile.Register32 // Interrupt Priority Registers
 }
 
 var NVIC = (*NVIC_Type)(unsafe.Pointer(uintptr(NVIC_BASE)))
 
 // Enable the given interrupt number.
 func EnableIRQ(irq uint32) {
-	NVIC.ISER[irq>>5] = 1 << (irq & 0x1F)
+	NVIC.ISER[irq>>5].Set(1 << (irq & 0x1F))
 }
 
 // Set the priority of the given interrupt number.
@@ -116,7 +114,7 @@ func SetPriority(irq uint32, priority uint32) {
 	regpos := irq % 4
 	mask := uint32(0xff) << (regpos * 8) // bits to clear
 	priority = priority << (regpos * 8)  // bits to set
-	NVIC.IPR[regnum] = RegValue((uint32(NVIC.IPR[regnum]) &^ mask) | priority)
+	NVIC.IPR[regnum].Set((uint32(NVIC.IPR[regnum].Get()) &^ mask) | priority)
 }
 
 // DisableInterrupts disables all interrupts, and returns the old state.

src/device/arm/cortexm.s

@@ -0,0 +1,21 @@
+.syntax unified
+
+.section .text.HardFault_Handler
+.global  HardFault_Handler
+.type    HardFault_Handler, %function
+HardFault_Handler:
+    // Put the old stack pointer in the first argument, for easy debugging. This
+    // is especially useful on Cortex-M0, which supports far fewer debug
+    // facilities.
+    mov r0, sp
+
+    // Load the default stack pointer from address 0 so that we can call normal
+    // functions again that expect a working stack. However, it will corrupt the
+    // old stack so the function below must not attempt to recover from this
+    // fault.
+    movs r3, #0
+    ldr r3, [r3]
+    mov sp, r3
+
+    // Continue handling this error in Go.
+    bl handleHardFault

src/device/riscv/riscv.go

@@ -0,0 +1,10 @@
+package riscv
+
+// Run the given assembly code. The code will be marked as having side effects,
+// as it doesn't produce output and thus would normally be eliminated by the
+// optimizer.
+func Asm(asm string)
+
+// ReadRegister returns the contents of the specified register. The register
+// must be a processor register, reachable with the "mov" instruction.
+func ReadRegister(name string) uintptr

src/device/riscv/start.S

@@ -0,0 +1,13 @@
+.section .init
+.global _start
+.type _start,@function
+
+_start:
+    // Workaround for missing support of the la pseudo-instruction in Clang 8:
+    // https://reviews.llvm.org/D55325
+    lui sp,      %hi(_stack_top)
+    addi sp, sp, %lo(_stack_top)
+    // see https://gnu-mcu-eclipse.github.io/arch/riscv/programmer/#the-gp-global-pointer-register
+    lui gp,      %hi(__global_pointer$)
+    addi gp, gp, %lo(__global_pointer$)
+    call main

src/examples/adc/adc.go

@@ -11,8 +11,8 @@ import (
 func main() {
 	machine.InitADC()
 
-	led := machine.GPIO{machine.LED}
-	led.Configure(machine.GPIOConfig{Mode: machine.GPIO_OUTPUT})
+	led := machine.LED
+	led.Configure(machine.PinConfig{Mode: machine.PinOutput})
 
 	sensor := machine.ADC{machine.ADC2}
 	sensor.Configure()

src/examples/blinky1/blinky1.go

@@ -8,8 +8,8 @@ import (
 )
 
 func main() {
-	led := machine.GPIO{machine.LED}
-	led.Configure(machine.GPIOConfig{Mode: machine.GPIO_OUTPUT})
+	led := machine.LED
+	led.Configure(machine.PinConfig{Mode: machine.PinOutput})
 	for {
 		led.Low()
 		time.Sleep(time.Millisecond * 500)

src/examples/blinky2/blinky2.go

@@ -16,8 +16,8 @@ func main() {
 }
 
 func led1() {
-	led := machine.GPIO{machine.LED}
-	led.Configure(machine.GPIOConfig{Mode: machine.GPIO_OUTPUT})
+	led := machine.LED1
+	led.Configure(machine.PinConfig{Mode: machine.PinOutput})
 	for {
 		println("+")
 		led.Low()
@@ -30,8 +30,8 @@ func led1() {
 }
 
 func led2() {
-	led := machine.GPIO{machine.LED2}
-	led.Configure(machine.GPIOConfig{Mode: machine.GPIO_OUTPUT})
+	led := machine.LED2
+	led.Configure(machine.PinConfig{Mode: machine.PinOutput})
 	for {
 		println("  +")
 		led.Low()

src/examples/button/button.go

@@ -7,14 +7,14 @@ import (
 
 // This example assumes that the button is connected to pin 8. Change the value
 // below to use a different pin.
-const buttonPin = 8
+const (
+	led    = machine.LED
+	button = machine.Pin(8)
+)
 
 func main() {
-	led := machine.GPIO{machine.LED}
-	led.Configure(machine.GPIOConfig{Mode: machine.GPIO_OUTPUT})
-
-	button := machine.GPIO{buttonPin}
-	button.Configure(machine.GPIOConfig{Mode: machine.GPIO_INPUT})
+	led.Configure(machine.PinConfig{Mode: machine.PinOutput})
+	button.Configure(machine.PinConfig{Mode: machine.PinInput})
 
 	for {
 		if button.Get() {

src/examples/button2/button2.go

@@ -8,29 +8,29 @@ import (
 // This example assumes that you are using the pca10040 board
 
 func main() {
-	led1 := machine.GPIO{machine.LED1}
-	led1.Configure(machine.GPIOConfig{Mode: machine.GPIO_OUTPUT})
+	led1 := machine.LED1
+	led1.Configure(machine.PinConfig{Mode: machine.PinOutput})
 
-	led2 := machine.GPIO{machine.LED2}
-	led2.Configure(machine.GPIOConfig{Mode: machine.GPIO_OUTPUT})
+	led2 := machine.LED2
+	led2.Configure(machine.PinConfig{Mode: machine.PinOutput})
 
-	led3 := machine.GPIO{machine.LED3}
-	led3.Configure(machine.GPIOConfig{Mode: machine.GPIO_OUTPUT})
+	led3 := machine.LED3
+	led3.Configure(machine.PinConfig{Mode: machine.PinOutput})
 
-	led4 := machine.GPIO{machine.LED4}
-	led4.Configure(machine.GPIOConfig{Mode: machine.GPIO_OUTPUT})
+	led4 := machine.LED4
+	led4.Configure(machine.PinConfig{Mode: machine.PinOutput})
 
-	button1 := machine.GPIO{machine.BUTTON1}
-	button1.Configure(machine.GPIOConfig{Mode: machine.GPIO_INPUT_PULLUP})
+	button1 := machine.BUTTON1
+	button1.Configure(machine.PinConfig{Mode: machine.PinInputPullup})
 
-	button2 := machine.GPIO{machine.BUTTON2}
-	button2.Configure(machine.GPIOConfig{Mode: machine.GPIO_INPUT_PULLUP})
+	button2 := machine.BUTTON2
+	button2.Configure(machine.PinConfig{Mode: machine.PinInputPullup})
 
-	button3 := machine.GPIO{machine.BUTTON3}
-	button3.Configure(machine.GPIOConfig{Mode: machine.GPIO_INPUT_PULLUP})
+	button3 := machine.BUTTON3
+	button3.Configure(machine.PinConfig{Mode: machine.PinInputPullup})
 
-	button4 := machine.GPIO{machine.BUTTON4}
-	button4.Configure(machine.GPIOConfig{Mode: machine.GPIO_INPUT_PULLUP})
+	button4 := machine.BUTTON4
+	button4.Configure(machine.PinConfig{Mode: machine.PinInputPullup})
 
 	for {
 		led1.Set(button1.Get())

src/examples/echo/echo.go

@@ -9,9 +9,9 @@ import (
 
 // change these to test a different UART or pins if available
 var (
-	uart       = machine.UART0
-	tx   uint8 = machine.UART_TX_PIN
-	rx   uint8 = machine.UART_RX_PIN
+	uart = machine.UART0
+	tx   = machine.UART_TX_PIN
+	rx   = machine.UART_RX_PIN
 )
 
 func main() {

src/examples/i2s/i2s.go

@@ -0,0 +1,25 @@
+// Example using the i2s hardware interface on the Adafruit Circuit Playground Express
+// to read data from the onboard MEMS microphone.
+//
+package main
+
+import (
+	"machine"
+)
+
+func main() {
+	machine.I2S0.Configure(machine.I2SConfig{
+		Mode:        machine.I2SModePDM,
+		ClockSource: machine.I2SClockSourceExternal,
+		Stereo:      true,
+	})
+
+	data := make([]uint32, 64)
+
+	for {
+		// get the next group of samples
+		machine.I2S0.Read(data)
+
+		println("data", data[0], data[1], data[2], data[4], "...")
+	}
+}

src/examples/mcp3008/mcp3008.go

@@ -8,19 +8,17 @@ import (
 	"time"
 )
 
-// CS_PIN is the pin used for Chip Select (CS). Change to whatever is in use on your board.
-const CS_PIN = 3
+// cs is the pin used for Chip Select (CS). Change to whatever is in use on your board.
+const cs = machine.Pin(3)
 
 var (
 	tx          []byte
 	rx          []byte
 	val, result uint16
-	cs          machine.GPIO
 )
 
 func main() {
-	cs = machine.GPIO{CS_PIN}
-	cs.Configure(machine.GPIOConfig{Mode: machine.GPIO_OUTPUT})
+	cs.Configure(machine.PinConfig{Mode: machine.PinOutput})
 
 	machine.SPI0.Configure(machine.SPIConfig{
 		Frequency: 4000000,

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

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

src/examples/wasm/Makefile

@@ -1,10 +1,15 @@
 export: clean wasm_exec
-	tinygo build -o ./html/wasm.wasm -target wasm ./export/wasm.go
+	tinygo build -o ./html/wasm.wasm -target wasm -no-debug ./export/wasm.go
 	cp ./export/wasm.js ./html/
 	cp ./export/index.html ./html/
 
+callback: clean wasm_exec
+	tinygo build -o ./html/wasm.wasm -target wasm ./callback/wasm.go
+	cp ./callback/wasm.js ./html/
+	cp ./callback/index.html ./html/
+
 main: clean wasm_exec
-	tinygo build -o ./html/wasm.wasm -target wasm ./main/main.go
+	tinygo build -o ./html/wasm.wasm -target wasm -no-debug ./main/main.go
 	cp ./main/index.html ./html/
 
 wasm_exec:

src/examples/wasm/README.md

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

src/examples/wasm/callback/index.html

@@ -0,0 +1,19 @@
+<!DOCTYPE html>
+
+<html>
+
+<head>
+	<meta charset="utf-8" />
+	<title>Go WebAssembly</title>
+	<meta name="viewport" content="width=device-width, initial-scale=1" />
+	<script src="wasm_exec.js" defer></script>
+	<script src="wasm.js" defer></script>
+</head>
+
+<body>
+	<h1>WebAssembly</h1>
+	<p>Add two numbers, using WebAssembly:</p>
+	<input type="number" id="a" value="0" /> + <input type="number" id="b" value="0" /> = <input type="number" id="result" readonly />
+</body>
+
+</html>

src/examples/wasm/callback/wasm.go

@@ -0,0 +1,27 @@
+package main
+
+import (
+	"strconv"
+	"syscall/js"
+)
+
+var a, b int
+
+func main() {
+	document := js.Global().Get("document")
+	document.Call("getElementById", "a").Set("oninput", updater(&a))
+	document.Call("getElementById", "b").Set("oninput", updater(&b))
+	update()
+}
+
+func updater(n *int) js.Func {
+	return js.FuncOf(func(this js.Value, args []js.Value) interface{} {
+		*n, _ = strconv.Atoi(this.Get("value").String())
+		update()
+		return nil
+	})
+}
+
+func update() {
+	js.Global().Get("document").Call("getElementById", "result").Set("value", a+b)
+}

src/examples/wasm/callback/wasm.js

@@ -0,0 +1,26 @@
+'use strict';
+
+const WASM_URL = 'wasm.wasm';
+
+var wasm;
+
+function init() {
+  const go = new Go();
+  if ('instantiateStreaming' in WebAssembly) {
+    WebAssembly.instantiateStreaming(fetch(WASM_URL), go.importObject).then(function (obj) {
+      wasm = obj.instance;
+      go.run(wasm);
+    })
+  } else {
+    fetch(WASM_URL).then(resp =>
+      resp.arrayBuffer()
+    ).then(bytes =>
+      WebAssembly.instantiate(bytes, go.importObject).then(function (obj) {
+        wasm = obj.instance;
+        go.run(wasm);
+      })
+    )
+  }
+}
+
+init();

src/examples/wasm/main/main.go

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

src/examples/wasm/server.go

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

src/machine/board_arduino.go

@@ -1,24 +1,24 @@
-// +build avr,arduino
+// +build arduino
 
 package machine
 
 const CPU_FREQUENCY = 16000000
 
 // LED on the Arduino
-const LED = 13
+const LED Pin = 13
 
 // ADC on the Arduino
 const (
-	ADC0 = 0
-	ADC1 = 1
-	ADC2 = 2
-	ADC3 = 3
-	ADC4 = 4 // Used by TWI for SDA
-	ADC5 = 5 // Used by TWI for SCL
+	ADC0 Pin = 0
+	ADC1 Pin = 1
+	ADC2 Pin = 2
+	ADC3 Pin = 3
+	ADC4 Pin = 4 // Used by TWI for SDA
+	ADC5 Pin = 5 // Used by TWI for SCL
 )
 
 // UART pins
 const (
-	UART_TX_PIN = 1
-	UART_RX_PIN = 0
+	UART_TX_PIN Pin = 1
+	UART_RX_PIN Pin = 0
 )

src/machine/board_arduino_nano33.go

@@ -0,0 +1,138 @@
+// +build sam,atsamd21,arduino_nano33
+
+// This contains the pin mappings for the Arduino Nano33 IoT board.
+//
+// For more information, see: https://store.arduino.cc/nano-33-iot
+//
+package machine
+
+import "device/sam"
+
+// GPIO Pins
+const (
+	RX0 Pin = PB23 // UART2 RX
+	TX1 Pin = PB22 // UART2 TX
+
+	D2 Pin = PB10 // PWM available
+	D3 Pin = PB11 // PWM available
+	D4 Pin = PA07
+	D5 Pin = PA05 // PWM available
+	D6 Pin = PA04 // PWM available
+	D7 Pin = PA06
+
+	D8  Pin = PA18
+	D9  Pin = PA20 // PWM available
+	D10 Pin = PA21 // PWM available
+	D11 Pin = PA16 // PWM available
+	D12 Pin = PA19 // PWM available
+
+	D13 Pin = PA17
+)
+
+// Analog pins
+const (
+	A0 Pin = PA02 // ADC/AIN[0]
+	A1 Pin = PB02 // ADC/AIN[10]
+	A2 Pin = PA11 // ADC/AIN[19]
+	A3 Pin = PA10 // ADC/AIN[18]
+	A4 Pin = PB08 // ADC/AIN[2], SCL:  SERCOM2/PAD[1]
+	A5 Pin = PB09 // ADC/AIN[3], SDA:  SERCOM2/PAD[1]
+	A6 Pin = PA09 // ADC/AIN[17]
+	A7 Pin = PB03 // ADC/AIN[11]
+)
+
+const (
+	LED = D13
+)
+
+// NINA-W102 Pins
+
+const (
+	NINA_MOSI   Pin = PA12
+	NINA_MISO   Pin = PA13
+	NINA_CS     Pin = PA14
+	NINA_SCK    Pin = PA15
+	NINA_GPIO0  Pin = PA27
+	NINA_RESETN Pin = PA08
+	NINA_ACK    Pin = PA28
+)
+
+// UART0 aka USBCDC pins
+const (
+	USBCDC_DM_PIN Pin = PA24
+	USBCDC_DP_PIN Pin = PA25
+)
+
+// UART1 on the Arduino Nano 33 connects to the onboard NINA-W102 WiFi chip.
+var (
+	UART1 = UART{Bus: sam.SERCOM5_USART,
+		Buffer: NewRingBuffer(),
+		Mode:   PinSERCOMAlt,
+		IRQVal: sam.IRQ_SERCOM5,
+	}
+)
+
+// UART1 pins
+const (
+	UART_TX_PIN Pin = PA22
+	UART_RX_PIN Pin = PA23
+)
+
+//go:export SERCOM5_IRQHandler
+func handleUART1() {
+	defaultUART1Handler()
+}
+
+// UART2 on the Arduino Nano 33 connects to the normal TX/RX pins.
+var (
+	UART2 = UART{Bus: sam.SERCOM3_USART,
+		Buffer: NewRingBuffer(),
+		Mode:   PinSERCOMAlt,
+		IRQVal: sam.IRQ_SERCOM3,
+	}
+)
+
+//go:export SERCOM3_IRQHandler
+func handleUART2() {
+	// should reset IRQ
+	UART2.Receive(byte((UART2.Bus.DATA.Get() & 0xFF)))
+	UART2.Bus.INTFLAG.SetBits(sam.SERCOM_USART_INTFLAG_RXC)
+}
+
+// I2C pins
+const (
+	SDA_PIN Pin = A4 // SDA: SERCOM4/PAD[1]
+	SCL_PIN Pin = A5 // SCL: SERCOM4/PAD[1]
+)
+
+// I2C on the Arduino Nano 33.
+var (
+	I2C0 = I2C{Bus: sam.SERCOM4_I2CM,
+		SDA:     SDA_PIN,
+		SCL:     SCL_PIN,
+		PinMode: PinSERCOMAlt}
+)
+
+// SPI pins
+const (
+	SPI0_SCK_PIN  Pin = PB11 // SCK: SERCOM4/PAD[3]
+	SPI0_MOSI_PIN Pin = PB10 // MOSI: SERCOM4/PAD[2]
+	SPI0_MISO_PIN Pin = PA12 // MISO: SERCOM4/PAD[0]
+)
+
+// SPI on the Arduino Nano 33.
+var (
+	SPI0 = SPI{Bus: sam.SERCOM1_SPI}
+)
+
+// I2S pins
+const (
+	I2S_SCK_PIN Pin = PA10
+	I2S_SD_PIN  Pin = PA08
+	I2S_WS_PIN      = NoPin // TODO: figure out what this is on Arduino Nano 33.
+)
+
+// I2S on the Arduino Nano 33.
+var (
+	I2S0 = I2S{Bus: sam.I2S}
+)

src/machine/board_bluepill.go

@@ -1,4 +1,4 @@
-// +build stm32,bluepill
+// +build bluepill
 
 package machine
 

src/machine/board_circuitplay_express.go

@@ -17,7 +17,7 @@ const (
 	D8  = PB23
 	D9  = PA06
 	D10 = PA07
-	D11 = 0xff // does not seem to exist
+	D11 = NoPin // does not seem to exist
 	D12 = PA02
 	D13 = PA17 // PWM available
 )
@@ -65,6 +65,20 @@ const (
 	UART_RX_PIN = PB09 // PORTB
 )
 
+// UART1 on the Circuit Playground Express.
+var (
+	UART1 = UART{Bus: sam.SERCOM1_USART,
+		Buffer: NewRingBuffer(),
+		Mode:   PinSERCOM,
+		IRQVal: sam.IRQ_SERCOM1,
+	}
+)
+
+//go:export SERCOM1_IRQHandler
+func handleUART1() {
+	defaultUART1Handler()
+}
+
 // I2C pins
 const (
 	SDA_PIN = PB02 // I2C0 external
@@ -80,12 +94,12 @@ var (
 	I2C0 = I2C{Bus: sam.SERCOM5_I2CM,
 		SDA:     SDA_PIN,
 		SCL:     SCL_PIN,
-		PinMode: GPIO_SERCOM}
+		PinMode: PinSERCOM}
 	// internal device
 	I2C1 = I2C{Bus: sam.SERCOM1_I2CM,
 		SDA:     SDA1_PIN,
 		SCL:     SCL1_PIN,
-		PinMode: GPIO_SERCOM_ALT}
+		PinMode: PinSERCOMAlt}
 )
 
 // SPI pins (internal flash)
@@ -99,3 +113,15 @@ const (
 var (
 	SPI0 = SPI{Bus: sam.SERCOM3_SPI}
 )
+
+// I2S pins
+const (
+	I2S_SCK_PIN = PA10
+	I2S_SD_PIN  = PA08
+	I2S_WS_PIN  = NoPin // no WS, instead uses SCK to sync
+)
+
+// I2S on the Circuit Playground Express.
+var (
+	I2S0 = I2S{Bus: sam.I2S}
+)

src/machine/board_digispark.go

@@ -1,7 +1,7 @@
-// +build attiny85,digispark
+// +build digispark
 
 package machine
 
 const (
-	LED = 1
+	LED Pin = 1
 )

src/machine/board_fe310.go

@@ -0,0 +1,38 @@
+// +build hifive1b
+
+package machine
+
+const (
+	P00 Pin = 0
+	P01 Pin = 1
+	P02 Pin = 2
+	P03 Pin = 3
+	P04 Pin = 4
+	P05 Pin = 5
+	P06 Pin = 6
+	P07 Pin = 7
+	P08 Pin = 8
+	P09 Pin = 9
+	P10 Pin = 10
+	P11 Pin = 11
+	P12 Pin = 12
+	P13 Pin = 13
+	P14 Pin = 14
+	P15 Pin = 15
+	P16 Pin = 16
+	P17 Pin = 17
+	P18 Pin = 18
+	P19 Pin = 19
+	P20 Pin = 20
+	P21 Pin = 21
+	P22 Pin = 22
+	P23 Pin = 23
+	P24 Pin = 24
+	P25 Pin = 25
+	P26 Pin = 26
+	P27 Pin = 27
+	P28 Pin = 28
+	P29 Pin = 29
+	P30 Pin = 30
+	P31 Pin = 31
+)

src/machine/board_feather-m0.go

@@ -0,0 +1,96 @@
+// +build sam,atsamd21,feather_m0
+
+package machine
+
+import "device/sam"
+
+// GPIO Pins
+const (
+	D0  = PA11  // UART0 RX
+	D1  = PA10  // UART0 TX
+	D2  = NoPin // does not seem to exist
+	D3  = PA09
+	D4  = PA08
+	D5  = PA15  // PWM available
+	D6  = PA20  // PWM available
+	D7  = NoPin // does not seem to exist
+	D8  = PA06
+	D9  = PA07 // PWM available
+	D10 = PA18 // can be used for PWM or UART1 TX
+	D11 = PA16 // can be used for PWM or UART1 RX
+	D12 = PA19 // PWM available
+	D13 = PA17 // PWM available
+)
+
+// Analog pins
+const (
+	A0 = PA02 // ADC/AIN[0]
+	A1 = PB08 // ADC/AIN[2]
+	A2 = PB09 // ADC/AIN[3]
+	A3 = PA04 // ADC/AIN[4]
+	A4 = PA05 // ADC/AIN[5]
+	A5 = PB02 // ADC/AIN[10]
+)
+
+const (
+	LED = D13
+)
+
+// UART0 aka USBCDC pins
+const (
+	USBCDC_DM_PIN = PA24
+	USBCDC_DP_PIN = PA25
+)
+
+// UART1 pins
+const (
+	UART_TX_PIN = D10
+	UART_RX_PIN = D11
+)
+
+// UART1 on the Feather M0.
+var (
+	UART1 = UART{Bus: sam.SERCOM1_USART,
+		Buffer: NewRingBuffer(),
+		Mode:   PinSERCOM,
+		IRQVal: sam.IRQ_SERCOM1,
+	}
+)
+
+//go:export SERCOM1_IRQHandler
+func handleUART1() {
+	defaultUART1Handler()
+}
+
+// I2C pins
+const (
+	SDA_PIN = PA22 // SDA: SERCOM3/PAD[0]
+	SCL_PIN = PA23 // SCL: SERCOM3/PAD[1]
+)
+
+// I2C on the Feather M0.
+var (
+	I2C0 = I2C{Bus: sam.SERCOM3_I2CM,
+		SDA:     SDA_PIN,
+		SCL:     SCL_PIN,
+		PinMode: PinSERCOM}
+)
+
+// SPI pins
+const (
+	SPI0_SCK_PIN  = PB11 // SCK: SERCOM4/PAD[3]
+	SPI0_MOSI_PIN = PB10 // MOSI: SERCOM4/PAD[2]
+	SPI0_MISO_PIN = PA12 // MISO: SERCOM4/PAD[0]
+)
+
+// SPI on the Feather M0.
+var (
+	SPI0 = SPI{Bus: sam.SERCOM4_SPI}
+)
+
+// I2S pins
+const (
+	I2S_SCK_PIN = PA10
+	I2S_SD_PIN  = PA08
+	I2S_WS_PIN  = NoPin // TODO: figure out what this is on Feather M0.
+)

src/machine/board_hifive1b.go

@@ -0,0 +1,19 @@
+// +build hifive1b
+
+package machine
+
+const (
+	LED       = LED1
+	LED1      = LED_RED
+	LED2      = LED_GREEN
+	LED3      = LED_BLUE
+	LED_RED   = P22
+	LED_GREEN = P19
+	LED_BLUE  = P21
+)
+
+const (
+	// TODO: figure out the pin numbers for these.
+	UART_TX_PIN = NoPin
+	UART_RX_PIN = NoPin
+)

src/machine/board_itsybitsy-m0.go

@@ -48,6 +48,20 @@ const (
 	UART_RX_PIN = D11
 )
 
+// UART1 on the ItsyBitsy M0.
+var (
+	UART1 = UART{Bus: sam.SERCOM1_USART,
+		Buffer: NewRingBuffer(),
+		Mode:   PinSERCOM,
+		IRQVal: sam.IRQ_SERCOM1,
+	}
+)
+
+//go:export SERCOM1_IRQHandler
+func handleUART1() {
+	defaultUART1Handler()
+}
+
 // I2C pins
 const (
 	SDA_PIN = PA22 // SDA: SERCOM3/PAD[0]
@@ -59,7 +73,7 @@ var (
 	I2C0 = I2C{Bus: sam.SERCOM3_I2CM,
 		SDA:     SDA_PIN,
 		SCL:     SCL_PIN,
-		PinMode: GPIO_SERCOM}
+		PinMode: PinSERCOM}
 )
 
 // SPI pins
@@ -73,3 +87,15 @@ const (
 var (
 	SPI0 = SPI{Bus: sam.SERCOM4_SPI}
 )
+
+// I2S pins
+const (
+	I2S_SCK_PIN = PA10
+	I2S_SD_PIN  = PA08
+	I2S_WS_PIN  = NoPin // TODO: figure out what this is on ItsyBitsy M0.
+)
+
+// I2S on the ItsyBitsy M0.
+var (
+	I2S0 = I2S{Bus: sam.I2S}
+)

src/machine/board_microbit.go

@@ -1,143 +1,76 @@
-// +build nrf51,microbit
+// +build microbit
 
 package machine
 
-import (
-	"device/nrf"
-	"errors"
-)
-
 // The micro:bit does not have a 32kHz crystal on board.
 const HasLowFrequencyCrystal = false
 
 // Buttons on the micro:bit (A and B)
 const (
-	BUTTON  = BUTTONA
-	BUTTONA = 17
-	BUTTONB = 26
+	BUTTON  Pin = BUTTONA
+	BUTTONA Pin = 17
+	BUTTONB Pin = 26
 )
 
 // UART pins
 const (
-	UART_TX_PIN = 24
-	UART_RX_PIN = 25
+	UART_TX_PIN Pin = 24
+	UART_RX_PIN Pin = 25
 )
 
 // ADC pins
 const (
-	ADC0 = 3 // P0 on the board
-	ADC1 = 2 // P1 on the board
-	ADC2 = 1 // P2 on the board
+	ADC0 Pin = 3 // P0 on the board
+	ADC1 Pin = 2 // P1 on the board
+	ADC2 Pin = 1 // P2 on the board
 )
 
 // I2C pins
 const (
-	SDA_PIN = 30 // P20 on the board
-	SCL_PIN = 0  // P19 on the board
+	SDA_PIN Pin = 30 // P20 on the board
+	SCL_PIN Pin = 0  // P19 on the board
 )
 
 // SPI pins
 const (
-	SPI0_SCK_PIN  = 23 // P13 on the board
-	SPI0_MOSI_PIN = 21 // P15 on the board
-	SPI0_MISO_PIN = 22 // P14 on the board
+	SPI0_SCK_PIN  Pin = 23 // P13 on the board
+	SPI0_MOSI_PIN Pin = 21 // P15 on the board
+	SPI0_MISO_PIN Pin = 22 // P14 on the board
 )
 
 // GPIO/Analog pins
 const (
-	P0  = 3
-	P1  = 2
-	P2  = 1
-	P3  = 4
-	P4  = 5
-	P5  = 17
-	P6  = 12
-	P7  = 11
-	P8  = 18
-	P9  = 10
-	P10 = 6
-	P11 = 26
-	P12 = 20
-	P13 = 23
-	P14 = 22
-	P15 = 21
-	P16 = 16
+	P0  Pin = 3
+	P1  Pin = 2
+	P2  Pin = 1
+	P3  Pin = 4
+	P4  Pin = 5
+	P5  Pin = 17
+	P6  Pin = 12
+	P7  Pin = 11
+	P8  Pin = 18
+	P9  Pin = 10
+	P10 Pin = 6
+	P11 Pin = 26
+	P12 Pin = 20
+	P13 Pin = 23
+	P14 Pin = 22
+	P15 Pin = 21
+	P16 Pin = 16
 )
 
 // LED matrix pins
 const (
-	LED_COL_1 = 4
-	LED_COL_2 = 5
-	LED_COL_3 = 6
-	LED_COL_4 = 7
-	LED_COL_5 = 8
-	LED_COL_6 = 9
-	LED_COL_7 = 10
-	LED_COL_8 = 11
-	LED_COL_9 = 12
-	LED_ROW_1 = 13
-	LED_ROW_2 = 14
-	LED_ROW_3 = 15
+	LED_COL_1 Pin = 4
+	LED_COL_2 Pin = 5
+	LED_COL_3 Pin = 6
+	LED_COL_4 Pin = 7
+	LED_COL_5 Pin = 8
+	LED_COL_6 Pin = 9
+	LED_COL_7 Pin = 10
+	LED_COL_8 Pin = 11
+	LED_COL_9 Pin = 12
+	LED_ROW_1 Pin = 13
+	LED_ROW_2 Pin = 14
+	LED_ROW_3 Pin = 15
 )
-
-// matrixSettings has the legs of the LED grid in the form {row, column} for each LED position.
-var matrixSettings = [5][5][2]uint8{
-	{{LED_ROW_1, LED_COL_1}, {LED_ROW_2, LED_COL_4}, {LED_ROW_1, LED_COL_2}, {LED_ROW_2, LED_COL_5}, {LED_ROW_1, LED_COL_3}},
-	{{LED_ROW_3, LED_COL_4}, {LED_ROW_3, LED_COL_5}, {LED_ROW_3, LED_COL_6}, {LED_ROW_3, LED_COL_7}, {LED_ROW_3, LED_COL_8}},
-	{{LED_ROW_2, LED_COL_2}, {LED_ROW_1, LED_COL_9}, {LED_ROW_2, LED_COL_3}, {LED_ROW_3, LED_COL_9}, {LED_ROW_2, LED_COL_1}},
-	{{LED_ROW_1, LED_COL_8}, {LED_ROW_1, LED_COL_7}, {LED_ROW_1, LED_COL_6}, {LED_ROW_1, LED_COL_5}, {LED_ROW_1, LED_COL_4}},
-	{{LED_ROW_3, LED_COL_3}, {LED_ROW_2, LED_COL_7}, {LED_ROW_3, LED_COL_1}, {LED_ROW_2, LED_COL_6}, {LED_ROW_3, LED_COL_2}}}
-
-// InitLEDMatrix initializes the LED matrix, by setting all of the row/col pins to output
-// then calling ClearLEDMatrix.
-func InitLEDMatrix() {
-	set := 0
-	for i := LED_COL_1; i <= LED_ROW_3; i++ {
-		set |= 1 << uint8(i)
-	}
-	nrf.GPIO.DIRSET = nrf.RegValue(set)
-	ClearLEDMatrix()
-}
-
-// ClearLEDMatrix clears the entire LED matrix.
-func ClearLEDMatrix() {
-	set := 0
-	for i := LED_COL_1; i <= LED_COL_9; i++ {
-		set |= 1 << uint8(i)
-	}
-	nrf.GPIO.OUTSET = nrf.RegValue(set)
-	nrf.GPIO.OUTCLR = (1 << LED_ROW_1) | (1 << LED_ROW_2) | (1 << LED_ROW_3)
-}
-
-// SetLEDMatrix turns on a single LED on the LED matrix.
-// Currently limited to a single LED at a time, it will clear the matrix before setting it.
-func SetLEDMatrix(x, y uint8) error {
-	if x > 4 || y > 4 {
-		return errors.New("Invalid LED matrix row or column")
-	}
-
-	// Clear matrix
-	ClearLEDMatrix()
-
-	nrf.GPIO.OUTSET = (1 << matrixSettings[y][x][0])
-	nrf.GPIO.OUTCLR = (1 << matrixSettings[y][x][1])
-
-	return nil
-}
-
-// SetEntireLEDMatrixOn turns on all of the LEDs on the LED matrix.
-func SetEntireLEDMatrixOn() error {
-	set := 0
-	for i := LED_ROW_1; i <= LED_ROW_3; i++ {
-		set |= 1 << uint8(i)
-	}
-	nrf.GPIO.OUTSET = nrf.RegValue(set)
-
-	set = 0
-	for i := LED_COL_1; i <= LED_COL_9; i++ {
-		set |= 1 << uint8(i)
-	}
-	nrf.GPIO.OUTCLR = nrf.RegValue(set)
-
-	return nil
-}

src/machine/board_nrf52840-mdk.go

@@ -6,27 +6,27 @@ const HasLowFrequencyCrystal = true
 
 // LEDs on the nrf52840-mdk (nRF52840 dev board)
 const (
-	LED       = LED_GREEN
-	LED_GREEN = 22
-	LED_RED   = 23
-	LED_BLUE  = 24
+	LED       Pin = LED_GREEN
+	LED_GREEN Pin = 22
+	LED_RED   Pin = 23
+	LED_BLUE  Pin = 24
 )
 
 // UART pins
 const (
-	UART_TX_PIN = 20
-	UART_RX_PIN = 19
+	UART_TX_PIN Pin = 20
+	UART_RX_PIN Pin = 19
 )
 
 // I2C pins (unused)
 const (
-	SDA_PIN = 0xff
-	SCL_PIN = 0xff
+	SDA_PIN = NoPin
+	SCL_PIN = NoPin
 )
 
 // SPI pins (unused)
 const (
-	SPI0_SCK_PIN  = 0
-	SPI0_MOSI_PIN = 0
-	SPI0_MISO_PIN = 0
+	SPI0_SCK_PIN  = NoPin
+	SPI0_MOSI_PIN = NoPin
+	SPI0_MISO_PIN = NoPin
 )

src/machine/board_pca10031.go

@@ -1,4 +1,4 @@
-// +build nrf51,pca10031
+// +build pca10031
 
 // pca10031 is a nrf51 based dongle, intended for use in wireless applications.
 //
@@ -10,30 +10,30 @@ const HasLowFrequencyCrystal = true
 
 // LED on the pca10031
 const (
-	LED       = LED_RED
-	LED1      = LED_RED
-	LED2      = LED_GREEN
-	LED3      = LED_BLUE
-	LED_RED   = 21
-	LED_GREEN = 22
-	LED_BLUE  = 23
+	LED       Pin = LED_RED
+	LED1      Pin = LED_RED
+	LED2      Pin = LED_GREEN
+	LED3      Pin = LED_BLUE
+	LED_RED   Pin = 21
+	LED_GREEN Pin = 22
+	LED_BLUE  Pin = 23
 )
 
 // UART pins
 const (
-	UART_TX_PIN = 9
-	UART_RX_PIN = 11
+	UART_TX_PIN Pin = 9
+	UART_RX_PIN Pin = 11
 )
 
 // I2C pins (disabled)
 const (
-	SDA_PIN = 0xff
-	SCL_PIN = 0xff
+	SDA_PIN = NoPin
+	SCL_PIN = NoPin
 )
 
 // SPI pins (unused)
 const (
-	SPI0_SCK_PIN  = 0
-	SPI0_MOSI_PIN = 0
-	SPI0_MISO_PIN = 0
+	SPI0_SCK_PIN  = NoPin
+	SPI0_MOSI_PIN = NoPin
+	SPI0_MISO_PIN = NoPin
 )

src/machine/board_pca10040.go

@@ -1,4 +1,4 @@
-// +build nrf,pca10040
+// +build pca10040
 
 package machine
 
@@ -7,47 +7,47 @@ const HasLowFrequencyCrystal = true
 
 // LEDs on the PCA10040 (nRF52832 dev board)
 const (
-	LED  = LED1
-	LED1 = 17
-	LED2 = 18
-	LED3 = 19
-	LED4 = 20
+	LED  Pin = LED1
+	LED1 Pin = 17
+	LED2 Pin = 18
+	LED3 Pin = 19
+	LED4 Pin = 20
 )
 
 // Buttons on the PCA10040 (nRF52832 dev board)
 const (
-	BUTTON  = BUTTON1
-	BUTTON1 = 13
-	BUTTON2 = 14
-	BUTTON3 = 15
-	BUTTON4 = 16
+	BUTTON  Pin = BUTTON1
+	BUTTON1 Pin = 13
+	BUTTON2 Pin = 14
+	BUTTON3 Pin = 15
+	BUTTON4 Pin = 16
 )
 
 // UART pins for NRF52840-DK
 const (
-	UART_TX_PIN = 6
-	UART_RX_PIN = 8
+	UART_TX_PIN Pin = 6
+	UART_RX_PIN Pin = 8
 )
 
 // ADC pins
 const (
-	ADC0 = 3
-	ADC1 = 4
-	ADC2 = 28
-	ADC3 = 29
-	ADC4 = 30
-	ADC5 = 31
+	ADC0 Pin = 3
+	ADC1 Pin = 4
+	ADC2 Pin = 28
+	ADC3 Pin = 29
+	ADC4 Pin = 30
+	ADC5 Pin = 31
 )
 
 // I2C pins
 const (
-	SDA_PIN = 26
-	SCL_PIN = 27
+	SDA_PIN Pin = 26
+	SCL_PIN Pin = 27
 )
 
 // SPI pins
 const (
-	SPI0_SCK_PIN  = 25
-	SPI0_MOSI_PIN = 23
-	SPI0_MISO_PIN = 24
+	SPI0_SCK_PIN  Pin = 25
+	SPI0_MOSI_PIN Pin = 23
+	SPI0_MISO_PIN Pin = 24
 )

src/machine/board_pca10056.go

@@ -6,47 +6,47 @@ const HasLowFrequencyCrystal = true
 
 // LEDs on the pca10056
 const (
-	LED  = LED1
-	LED1 = 13
-	LED2 = 14
-	LED3 = 15
-	LED4 = 16
+	LED  Pin = LED1
+	LED1 Pin = 13
+	LED2 Pin = 14
+	LED3 Pin = 15
+	LED4 Pin = 16
 )
 
 // Buttons on the pca10056
 const (
-	BUTTON  = BUTTON1
-	BUTTON1 = 11
-	BUTTON2 = 12
-	BUTTON3 = 24
-	BUTTON4 = 25
+	BUTTON  Pin = BUTTON1
+	BUTTON1 Pin = 11
+	BUTTON2 Pin = 12
+	BUTTON3 Pin = 24
+	BUTTON4 Pin = 25
 )
 
 // UART pins
 const (
-	UART_TX_PIN = 6
-	UART_RX_PIN = 8
+	UART_TX_PIN Pin = 6
+	UART_RX_PIN Pin = 8
 )
 
 // ADC pins
 const (
-	ADC0 = 3
-	ADC1 = 4
-	ADC2 = 28
-	ADC3 = 29
-	ADC4 = 30
-	ADC5 = 31
+	ADC0 Pin = 3
+	ADC1 Pin = 4
+	ADC2 Pin = 28
+	ADC3 Pin = 29
+	ADC4 Pin = 30
+	ADC5 Pin = 31
 )
 
 // I2C pins
 const (
-	SDA_PIN = 26 // P0.26
-	SCL_PIN = 27 // P0.27
+	SDA_PIN Pin = 26 // P0.26
+	SCL_PIN Pin = 27 // P0.27
 )
 
 // SPI pins
 const (
-	SPI0_SCK_PIN  = 47 // P1.15
-	SPI0_MOSI_PIN = 45 // P1.13
-	SPI0_MISO_PIN = 46 // P1.14
+	SPI0_SCK_PIN  Pin = 47 // P1.15
+	SPI0_MOSI_PIN Pin = 45 // P1.13
+	SPI0_MISO_PIN Pin = 46 // P1.14
 )

src/machine/board_reelboard.go

@@ -4,39 +4,58 @@ package machine
 
 const HasLowFrequencyCrystal = true
 
-// LEDs on the reel board
+// Pins on the reel board
 const (
-	LED        = LED1
-	LED1       = LED_YELLOW
-	LED2       = LED_RED
-	LED3       = LED_GREEN
-	LED4       = LED_BLUE
-	LED_RED    = 11
-	LED_GREEN  = 12
-	LED_BLUE   = 41
-	LED_YELLOW = 13
+	LED              Pin = LED1
+	LED1             Pin = LED_YELLOW
+	LED2             Pin = LED_RED
+	LED3             Pin = LED_GREEN
+	LED4             Pin = LED_BLUE
+	LED_RED          Pin = 11
+	LED_GREEN        Pin = 12
+	LED_BLUE         Pin = 41
+	LED_YELLOW       Pin = 13
+	EPD_BUSY_PIN     Pin = 14
+	EPD_RESET_PIN    Pin = 15
+	EPD_DC_PIN       Pin = 16
+	EPD_CS_PIN       Pin = 17
+	EPD_SCK_PIN      Pin = 19
+	EPD_MOSI_PIN     Pin = 20
+	POWER_SUPPLY_PIN Pin = 32
 )
 
 // User "a" button on the reel board
 const (
-	BUTTON = 7
+	BUTTON Pin = 7
 )
 
 // UART pins
 const (
-	UART_TX_PIN = 6
-	UART_RX_PIN = 8
+	UART_TX_PIN Pin = 6
+	UART_RX_PIN Pin = 8
 )
 
 // I2C pins
 const (
-	SDA_PIN = 26
-	SCL_PIN = 27
+	SDA_PIN Pin = 26
+	SCL_PIN Pin = 27
 )
 
 // SPI pins
 const (
-	SPI0_SCK_PIN  = 47
-	SPI0_MOSI_PIN = 45
-	SPI0_MISO_PIN = 46
+	SPI0_SCK_PIN  Pin = 47
+	SPI0_MOSI_PIN Pin = 45
+	SPI0_MISO_PIN Pin = 46
 )
+
+// PowerSupplyActive enables the supply voltages for nRF52840 and peripherals (true) or only for nRF52840 (false)
+// This controls the TPS610981 boost converter. You must turn the power supply active in order to use the EPD and
+// other onboard peripherals.
+func PowerSupplyActive(active bool) {
+	POWER_SUPPLY_PIN.Configure(PinConfig{Mode: PinOutput})
+	if active {
+		POWER_SUPPLY_PIN.High()
+	} else {
+		POWER_SUPPLY_PIN.Low()
+	}
+}

src/machine/board_stm32.go

@@ -0,0 +1,16 @@
+// +build bluepill stm32f4disco
+
+package machine
+
+// Peripheral abstraction layer for the stm32.
+
+const (
+	portA Pin = iota * 16
+	portB
+	portC
+	portD
+	portE
+	portF
+	portG
+	portH
+)

src/machine/board_stm32f4disco.go

@@ -1,4 +1,4 @@
-// +build stm32,stm32f4disco
+// +build stm32f4disco
 
 package machine
 

src/machine/board_trinket.go

@@ -0,0 +1,87 @@
+// +build sam,atsamd21,trinket_m0
+
+package machine
+
+import "device/sam"
+
+// GPIO Pins
+const (
+	D0  = PA08 // PWM available
+	D1  = PA02
+	D2  = PA09 // PWM available
+	D3  = PA07 // PWM available / UART0 RX
+	D4  = PA06 // PWM available / UART0 TX
+	D13 = PA10 // LED
+)
+
+// Analog pins
+const (
+	A0 = D1
+	A1 = D2
+	A2 = D0
+	A3 = D3
+	A4 = D4
+)
+
+const (
+	LED = D13
+)
+
+// UART0 aka USBCDC pins
+const (
+	USBCDC_DM_PIN = PA24
+	USBCDC_DP_PIN = PA25
+)
+
+// UART1 pins
+const (
+	UART_TX_PIN = D4
+	UART_RX_PIN = D3
+)
+
+// UART1 on the Trinket M0.
+var (
+	UART1 = UART{Bus: sam.SERCOM1_USART,
+		Buffer: NewRingBuffer(),
+		Mode:   PinSERCOM,
+		IRQVal: sam.IRQ_SERCOM1,
+	}
+)
+
+//go:export SERCOM1_IRQHandler
+func handleUART1() {
+	defaultUART1Handler()
+}
+
+// SPI pins
+const (
+	SPI0_SCK_PIN  = D3
+	SPI0_MOSI_PIN = D4
+	SPI0_MISO_PIN = D2
+)
+
+// SPI on the Trinket M0.
+var (
+	SPI0 = SPI{Bus: sam.SERCOM0_SPI}
+)
+
+// I2C pins
+const (
+	SDA_PIN = D0 // SDA
+	SCL_PIN = D2 // SCL
+)
+
+// I2C on the Trinket M0.
+var (
+	I2C0 = I2C{Bus: sam.SERCOM2_I2CM,
+		SDA:     SDA_PIN,
+		SCL:     SCL_PIN,
+		PinMode: PinSERCOMAlt}
+)
+
+// I2S pins
+const (
+	I2S_SCK_PIN = PA10
+	I2S_SD_PIN  = PA08
+	I2S_WS_PIN  = NoPin // TODO: figure out what this is on Trinket M0.
+)

src/machine/buffer.go

@@ -1,9 +1,10 @@
 package machine
 
-const bufferSize = 128
+import (
+	"runtime/volatile"
+)
 
-//go:volatile
-type volatileByte byte
+const bufferSize = 128
 
 // RingBuffer is ring buffer implementation inspired by post at
 // https://www.embeddedrelated.com/showthread/comp.arch.embedded/77084-1.php
@@ -12,9 +13,9 @@ type volatileByte byte
 // members of a struct are not compiled correctly by TinyGo.
 // See https://github.com/tinygo-org/tinygo/issues/151 for details.
 type RingBuffer struct {
-	rxbuffer [bufferSize]volatileByte
-	head     volatileByte
-	tail     volatileByte
+	rxbuffer [bufferSize]volatile.Register8
+	head     volatile.Register8
+	tail     volatile.Register8
 }
 
 // NewRingBuffer returns a new ring buffer.
@@ -24,15 +25,15 @@ func NewRingBuffer() *RingBuffer {
 
 // Used returns how many bytes in buffer have been used.
 func (rb *RingBuffer) Used() uint8 {
-	return uint8(rb.head - rb.tail)
+	return uint8(rb.head.Get() - rb.tail.Get())
 }
 
 // Put stores a byte in the buffer. If the buffer is already
 // full, the method will return false.
 func (rb *RingBuffer) Put(val byte) bool {
 	if rb.Used() != bufferSize {
-		rb.head++
-		rb.rxbuffer[rb.head%bufferSize] = volatileByte(val)
+		rb.head.Set(rb.head.Get() + 1)
+		rb.rxbuffer[rb.head.Get()%bufferSize].Set(val)
 		return true
 	}
 	return false
@@ -42,8 +43,8 @@ func (rb *RingBuffer) Put(val byte) bool {
 // the method will return a false as the second value.
 func (rb *RingBuffer) Get() (byte, bool) {
 	if rb.Used() != 0 {
-		rb.tail++
-		return byte(rb.rxbuffer[rb.tail%bufferSize]), true
+		rb.tail.Set(rb.tail.Get() + 1)
+		return rb.rxbuffer[rb.tail.Get()%bufferSize].Get(), true
 	}
 	return 0, false
 }

src/machine/i2c.go

@@ -1,4 +1,4 @@
-// +build avr nrf sam stm32f103xx
+// +build !stm32f4disco,!hifive1b
 
 package machine
 

src/machine/i2s.go

@@ -0,0 +1,54 @@
+// +build sam
+
+// This is the definition for I2S bus functions.
+// Actual implementations if available for any given hardware
+// are to be found in its the board definition.
+//
+// For more info about I2S, see: https://en.wikipedia.org/wiki/I%C2%B2S
+//
+
+package machine
+
+type I2SMode uint8
+type I2SStandard uint8
+type I2SClockSource uint8
+type I2SDataFormat uint8
+
+const (
+	I2SModeMaster I2SMode = iota
+	I2SModeSlave
+	I2SModePDM
+)
+
+const (
+	I2StandardPhilips I2SStandard = iota
+	I2SStandardMSB
+	I2SStandardLSB
+)
+
+const (
+	I2SClockSourceInternal I2SClockSource = iota
+	I2SClockSourceExternal
+)
+
+const (
+	I2SDataFormatDefault I2SDataFormat = 0
+	I2SDataFormat8bit                  = 8
+	I2SDataFormat16bit                 = 16
+	I2SDataFormat24bit                 = 24
+	I2SDataFormat32bit                 = 32
+)
+
+// All fields are optional and may not be required or used on a particular platform.
+type I2SConfig struct {
+	SCK               Pin
+	WS                Pin
+	SD                Pin
+	Mode              I2SMode
+	Standard          I2SStandard
+	ClockSource       I2SClockSource
+	DataFormat        I2SDataFormat
+	AudioFrequency    uint32
+	MasterClockOutput bool
+	Stereo            bool
+}

src/machine/machine.go

@@ -1,25 +1,36 @@
 package machine
 
-type GPIOConfig struct {
-	Mode GPIOMode
+type PinConfig struct {
+	Mode PinMode
 }
 
-type GPIO struct {
-	Pin uint8
-}
+// Pin is a single pin on a chip, which may be connected to other hardware
+// devices. It can either be used directly as GPIO pin or it can be used in
+// other peripherals like ADC, I2C, etc.
+type Pin int8
 
-func (p GPIO) High() {
+// NoPin explicitly indicates "not a pin". Use this pin if you want to leave one
+// of the pins in a peripheral unconfigured (if supported by the hardware).
+const NoPin = Pin(-1)
+
+// High sets this GPIO pin to high, assuming it has been configured as an output
+// pin. It is hardware dependent (and often undefined) what happens if you set a
+// pin to high that is not configured as an output pin.
+func (p Pin) High() {
 	p.Set(true)
 }
 
-func (p GPIO) Low() {
+// Low sets this GPIO pin to low, assuming it has been configured as an output
+// pin. It is hardware dependent (and often undefined) what happens if you set a
+// pin to low that is not configured as an output pin.
+func (p Pin) Low() {
 	p.Set(false)
 }
 
 type PWM struct {
-	Pin uint8
+	Pin Pin
 }
 
 type ADC struct {
-	Pin uint8
+	Pin Pin
 }

src/machine/machine_atmega.go

@@ -4,105 +4,106 @@ package machine
 
 import (
 	"device/avr"
+	"runtime/volatile"
 )
 
 // Configure sets the pin to input or output.
-func (p GPIO) Configure(config GPIOConfig) {
-	if config.Mode == GPIO_OUTPUT { // set output bit
-		if p.Pin < 8 {
-			*avr.DDRD |= 1 << p.Pin
+func (p Pin) Configure(config PinConfig) {
+	if config.Mode == PinOutput { // set output bit
+		if p < 8 {
+			avr.DDRD.SetBits(1 << uint8(p))
 		} else {
-			*avr.DDRB |= 1 << (p.Pin - 8)
+			avr.DDRB.SetBits(1 << uint8(p-8))
 		}
 	} else { // configure input: clear output bit
-		if p.Pin < 8 {
-			*avr.DDRD &^= 1 << p.Pin
+		if p < 8 {
+			avr.DDRD.ClearBits(1 << uint8(p))
 		} else {
-			*avr.DDRB &^= 1 << (p.Pin - 8)
+			avr.DDRB.ClearBits(1 << uint8(p-8))
 		}
 	}
 }
 
 // Get returns the current value of a GPIO pin.
-func (p GPIO) Get() bool {
-	if p.Pin < 8 {
-		val := *avr.PIND & (1 << p.Pin)
+func (p Pin) Get() bool {
+	if p < 8 {
+		val := avr.PIND.Get() & (1 << uint8(p))
 		return (val > 0)
 	} else {
-		val := *avr.PINB & (1 << (p.Pin - 8))
+		val := avr.PINB.Get() & (1 << uint8(p-8))
 		return (val > 0)
 	}
 }
 
-func (p GPIO) getPortMask() (*avr.RegValue, uint8) {
-	if p.Pin < 8 {
-		return avr.PORTD, 1 << p.Pin
+func (p Pin) getPortMask() (*volatile.Register8, uint8) {
+	if p < 8 {
+		return avr.PORTD, 1 << uint8(p)
 	} else {
-		return avr.PORTB, 1 << (p.Pin - 8)
+		return avr.PORTB, 1 << uint8(p-8)
 	}
 }
 
 // InitPWM initializes the registers needed for PWM.
 func InitPWM() {
 	// use waveform generation
-	*avr.TCCR0A |= avr.TCCR0A_WGM00
+	avr.TCCR0A.SetBits(avr.TCCR0A_WGM00)
 
 	// set timer 0 prescale factor to 64
-	*avr.TCCR0B |= avr.TCCR0B_CS01 | avr.TCCR0B_CS00
+	avr.TCCR0B.SetBits(avr.TCCR0B_CS01 | avr.TCCR0B_CS00)
 
 	// set timer 1 prescale factor to 64
-	*avr.TCCR1B |= avr.TCCR1B_CS11
+	avr.TCCR1B.SetBits(avr.TCCR1B_CS11)
 
 	// put timer 1 in 8-bit phase correct pwm mode
-	*avr.TCCR1A |= avr.TCCR1A_WGM10
+	avr.TCCR1A.SetBits(avr.TCCR1A_WGM10)
 
 	// set timer 2 prescale factor to 64
-	*avr.TCCR2B |= avr.TCCR2B_CS22
+	avr.TCCR2B.SetBits(avr.TCCR2B_CS22)
 
 	// configure timer 2 for phase correct pwm (8-bit)
-	*avr.TCCR2A |= avr.TCCR2A_WGM20
+	avr.TCCR2A.SetBits(avr.TCCR2A_WGM20)
 }
 
 // Configure configures a PWM pin for output.
 func (pwm PWM) Configure() {
 	if pwm.Pin < 8 {
-		*avr.DDRD |= 1 << pwm.Pin
+		avr.DDRD.SetBits(1 << uint8(pwm.Pin))
 	} else {
-		*avr.DDRB |= 1 << (pwm.Pin - 8)
+		avr.DDRB.SetBits(1 << uint8(pwm.Pin-8))
 	}
 }
 
 // Set turns on the duty cycle for a PWM pin using the provided value. On the AVR this is normally a
 // 8-bit value ranging from 0 to 255.
 func (pwm PWM) Set(value uint16) {
-	value8 := value >> 8
+	value8 := uint8(value >> 8)
 	switch pwm.Pin {
 	case 3:
 		// connect pwm to pin on timer 2, channel B
-		*avr.TCCR2A |= avr.TCCR2A_COM2B1
-		*avr.OCR2B = avr.RegValue(value8) // set pwm duty
+		avr.TCCR2A.SetBits(avr.TCCR2A_COM2B1)
+		avr.OCR2B.Set(value8) // set pwm duty
 	case 5:
 		// connect pwm to pin on timer 0, channel B
-		*avr.TCCR0A |= avr.TCCR0A_COM0B1
-		*avr.OCR0B = avr.RegValue(value8) // set pwm duty
+		avr.TCCR0A.SetBits(avr.TCCR0A_COM0B1)
+		avr.OCR0B.Set(value8) // set pwm duty
 	case 6:
 		// connect pwm to pin on timer 0, channel A
-		*avr.TCCR0A |= avr.TCCR0A_COM0A1
-		*avr.OCR0A = avr.RegValue(value8) // set pwm duty
+		avr.TCCR0A.SetBits(avr.TCCR0A_COM0A1)
+		avr.OCR0A.Set(value8) // set pwm duty
 	case 9:
 		// connect pwm to pin on timer 1, channel A
-		*avr.TCCR1A |= avr.TCCR1A_COM1A1
+		avr.TCCR1A.SetBits(avr.TCCR1A_COM1A1)
 		// this is a 16-bit value, but we only currently allow the low order bits to be set
-		*avr.OCR1AL = avr.RegValue(value8) // set pwm duty
+		avr.OCR1AL.Set(value8) // set pwm duty
 	case 10:
 		// connect pwm to pin on timer 1, channel B
-		*avr.TCCR1A |= avr.TCCR1A_COM1B1
+		avr.TCCR1A.SetBits(avr.TCCR1A_COM1B1)
 		// this is a 16-bit value, but we only currently allow the low order bits to be set
-		*avr.OCR1BL = avr.RegValue(value8) // set pwm duty
+		avr.OCR1BL.Set(value8) // set pwm duty
 	case 11:
 		// connect pwm to pin on timer 2, channel A
-		*avr.TCCR2A |= avr.TCCR2A_COM2A1
-		*avr.OCR2A = avr.RegValue(value8) // set pwm duty
+		avr.TCCR2A.SetBits(avr.TCCR2A_COM2A1)
+		avr.OCR2A.Set(value8) // set pwm duty
 	default:
 		panic("Invalid PWM pin")
 	}
@@ -121,19 +122,19 @@ func (i2c I2C) Configure(config I2CConfig) {
 	}
 
 	// Activate internal pullups for twi.
-	*avr.PORTC |= (avr.DIDR0_ADC4D | avr.DIDR0_ADC5D)
+	avr.PORTC.SetBits((avr.DIDR0_ADC4D | avr.DIDR0_ADC5D))
 
 	// Initialize twi prescaler and bit rate.
-	*avr.TWSR |= (avr.TWSR_TWPS0 | avr.TWSR_TWPS1)
+	avr.TWSR.SetBits((avr.TWSR_TWPS0 | avr.TWSR_TWPS1))
 
 	// twi bit rate formula from atmega128 manual pg. 204:
 	// SCL Frequency = CPU Clock Frequency / (16 + (2 * TWBR))
 	// NOTE: TWBR should be 10 or higher for master mode.
 	// It is 72 for a 16mhz board with 100kHz TWI
-	*avr.TWBR = avr.RegValue(((CPU_FREQUENCY / config.Frequency) - 16) / 2)
+	avr.TWBR.Set(uint8(((CPU_FREQUENCY / config.Frequency) - 16) / 2))
 
 	// Enable twi module.
-	*avr.TWCR = avr.TWCR_TWEN
+	avr.TWCR.Set(avr.TWCR_TWEN)
 }
 
 // Tx does a single I2C transaction at the specified address.
@@ -162,10 +163,10 @@ func (i2c I2C) Tx(addr uint16, w, r []byte) error {
 // start starts an I2C communication session.
 func (i2c I2C) start(address uint8, write bool) {
 	// Clear TWI interrupt flag, put start condition on SDA, and enable TWI.
-	*avr.TWCR = (avr.TWCR_TWINT | avr.TWCR_TWSTA | avr.TWCR_TWEN)
+	avr.TWCR.Set((avr.TWCR_TWINT | avr.TWCR_TWSTA | avr.TWCR_TWEN))
 
 	// Wait till start condition is transmitted.
-	for (*avr.TWCR & avr.TWCR_TWINT) == 0 {
+	for !avr.TWCR.HasBits(avr.TWCR_TWINT) {
 	}
 
 	// Write 7-bit shifted peripheral address.
@@ -179,36 +180,36 @@ func (i2c I2C) start(address uint8, write bool) {
 // stop ends an I2C communication session.
 func (i2c I2C) stop() {
 	// Send stop condition.
-	*avr.TWCR = (avr.TWCR_TWEN | avr.TWCR_TWINT | avr.TWCR_TWSTO)
+	avr.TWCR.Set(avr.TWCR_TWEN | avr.TWCR_TWINT | avr.TWCR_TWSTO)
 
 	// Wait for stop condition to be executed on bus.
-	for (*avr.TWCR & avr.TWCR_TWSTO) == 0 {
+	for !avr.TWCR.HasBits(avr.TWCR_TWSTO) {
 	}
 }
 
 // writeByte writes a single byte to the I2C bus.
 func (i2c I2C) writeByte(data byte) {
 	// Write data to register.
-	*avr.TWDR = avr.RegValue(data)
+	avr.TWDR.Set(data)
 
 	// Clear TWI interrupt flag and enable TWI.
-	*avr.TWCR = (avr.TWCR_TWEN | avr.TWCR_TWINT)
+	avr.TWCR.Set(avr.TWCR_TWEN | avr.TWCR_TWINT)
 
 	// Wait till data is transmitted.
-	for (*avr.TWCR & avr.TWCR_TWINT) == 0 {
+	for !avr.TWCR.HasBits(avr.TWCR_TWINT) {
 	}
 }
 
 // readByte reads a single byte from the I2C bus.
 func (i2c I2C) readByte() byte {
 	// Clear TWI interrupt flag and enable TWI.
-	*avr.TWCR = (avr.TWCR_TWEN | avr.TWCR_TWINT | avr.TWCR_TWEA)
+	avr.TWCR.Set(avr.TWCR_TWEN | avr.TWCR_TWINT | avr.TWCR_TWEA)
 
 	// Wait till read request is transmitted.
-	for (*avr.TWCR & avr.TWCR_TWINT) == 0 {
+	for !avr.TWCR.HasBits(avr.TWCR_TWINT) {
 	}
 
-	return byte(*avr.TWDR)
+	return byte(avr.TWDR.Get())
 }
 
 // UART on the AVR.
@@ -226,32 +227,32 @@ func (uart UART) Configure(config UARTConfig) {
 	// https://www.microchip.com/webdoc/AVRLibcReferenceManual/FAQ_1faq_wrong_baud_rate.html
 	// ((F_CPU + UART_BAUD_RATE * 8L) / (UART_BAUD_RATE * 16L) - 1)
 	ps := ((CPU_FREQUENCY+config.BaudRate*8)/(config.BaudRate*16) - 1)
-	*avr.UBRR0H = avr.RegValue(ps >> 8)
-	*avr.UBRR0L = avr.RegValue(ps & 0xff)
+	avr.UBRR0H.Set(uint8(ps >> 8))
+	avr.UBRR0L.Set(uint8(ps & 0xff))
 
 	// enable RX, TX and RX interrupt
-	*avr.UCSR0B = avr.UCSR0B_RXEN0 | avr.UCSR0B_TXEN0 | avr.UCSR0B_RXCIE0
+	avr.UCSR0B.Set(avr.UCSR0B_RXEN0 | avr.UCSR0B_TXEN0 | avr.UCSR0B_RXCIE0)
 
 	// 8-bits data
-	*avr.UCSR0C = avr.UCSR0C_UCSZ01 | avr.UCSR0C_UCSZ00
+	avr.UCSR0C.Set(avr.UCSR0C_UCSZ01 | avr.UCSR0C_UCSZ00)
 }
 
 // WriteByte writes a byte of data to the UART.
 func (uart UART) WriteByte(c byte) error {
 	// Wait until UART buffer is not busy.
-	for (*avr.UCSR0A & avr.UCSR0A_UDRE0) == 0 {
+	for !avr.UCSR0A.HasBits(avr.UCSR0A_UDRE0) {
 	}
-	*avr.UDR0 = avr.RegValue(c) // send char
+	avr.UDR0.Set(c) // send char
 	return nil
 }
 
 //go:interrupt USART_RX_vect
 func handleUSART_RX() {
 	// Read register to clear it.
-	data := *avr.UDR0
+	data := avr.UDR0.Get()
 
 	// Ensure no error.
-	if (*avr.UCSR0A & (avr.UCSR0A_FE0 | avr.UCSR0A_DOR0 | avr.UCSR0A_UPE0)) == 0 {
+	if !avr.UCSR0A.HasBits(avr.UCSR0A_FE0 | avr.UCSR0A_DOR0 | avr.UCSR0A_UPE0) {
 		// Put data from UDR register into buffer.
 		UART0.Receive(byte(data))
 	}

src/machine/machine_atsamd21e18.go

@@ -0,0 +1,334 @@
+// +build sam,atsamd21,atsamd21e18
+
+// Peripheral abstraction layer for the atsamd21.
+//
+// Datasheet:
+// http://ww1.microchip.com/downloads/en/DeviceDoc/SAMD21-Family-DataSheet-DS40001882D.pdf
+//
+package machine
+
+import (
+	"device/sam"
+)
+
+// Return the register and mask to enable a given GPIO pin. This can be used to
+// implement bit-banged drivers.
+func (p Pin) PortMaskSet() (*uint32, uint32) {
+	return &sam.PORT.OUTSET0.Reg, 1 << uint8(p)
+}
+
+// Return the register and mask to disable a given port. This can be used to
+// implement bit-banged drivers.
+func (p Pin) PortMaskClear() (*uint32, uint32) {
+	return &sam.PORT.OUTCLR0.Reg, 1 << uint8(p)
+}
+
+// Set the pin to high or low.
+// Warning: only use this on an output pin!
+func (p Pin) Set(high bool) {
+	if high {
+		sam.PORT.OUTSET0.Set(1 << uint8(p))
+	} else {
+		sam.PORT.OUTCLR0.Set(1 << uint8(p))
+	}
+}
+
+// Get returns the current value of a GPIO pin.
+func (p Pin) Get() bool {
+	return (sam.PORT.IN0.Get()>>uint8(p))&1 > 0
+}
+
+// Configure this pin with the given configuration.
+func (p Pin) Configure(config PinConfig) {
+	switch config.Mode {
+	case PinOutput:
+		sam.PORT.DIRSET0.Set(1 << uint8(p))
+		// output is also set to input enable so pin can read back its own value
+		p.setPinCfg(sam.PORT_PINCFG0_INEN)
+
+	case PinInput:
+		sam.PORT.DIRCLR0.Set(1 << uint8(p))
+		p.setPinCfg(sam.PORT_PINCFG0_INEN)
+
+	case PinInputPulldown:
+		sam.PORT.DIRCLR0.Set(1 << uint8(p))
+		sam.PORT.OUTCLR0.Set(1 << uint8(p))
+		p.setPinCfg(sam.PORT_PINCFG0_INEN | sam.PORT_PINCFG0_PULLEN)
+
+	case PinInputPullup:
+		sam.PORT.DIRCLR0.Set(1 << uint8(p))
+		sam.PORT.OUTSET0.Set(1 << uint8(p))
+		p.setPinCfg(sam.PORT_PINCFG0_INEN | sam.PORT_PINCFG0_PULLEN)
+
+	case PinSERCOM:
+		if uint8(p)&1 > 0 {
+			// odd pin, so save the even pins
+			val := p.getPMux() & sam.PORT_PMUX0_PMUXE_Msk
+			p.setPMux(val | (uint8(PinSERCOM) << sam.PORT_PMUX0_PMUXO_Pos))
+		} else {
+			// even pin, so save the odd pins
+			val := p.getPMux() & sam.PORT_PMUX0_PMUXO_Msk
+			p.setPMux(val | (uint8(PinSERCOM) << sam.PORT_PMUX0_PMUXE_Pos))
+		}
+		// enable port config
+		p.setPinCfg(sam.PORT_PINCFG0_PMUXEN | sam.PORT_PINCFG0_DRVSTR | sam.PORT_PINCFG0_INEN)
+
+	case PinSERCOMAlt:
+		if uint8(p)&1 > 0 {
+			// odd pin, so save the even pins
+			val := p.getPMux() & sam.PORT_PMUX0_PMUXE_Msk
+			p.setPMux(val | (uint8(PinSERCOMAlt) << sam.PORT_PMUX0_PMUXO_Pos))
+		} else {
+			// even pin, so save the odd pins
+			val := p.getPMux() & sam.PORT_PMUX0_PMUXO_Msk
+			p.setPMux(val | (uint8(PinSERCOMAlt) << sam.PORT_PMUX0_PMUXE_Pos))
+		}
+		// enable port config
+		p.setPinCfg(sam.PORT_PINCFG0_PMUXEN | sam.PORT_PINCFG0_DRVSTR)
+
+	case PinCom:
+		if uint8(p)&1 > 0 {
+			// odd pin, so save the even pins
+			val := p.getPMux() & sam.PORT_PMUX0_PMUXE_Msk
+			p.setPMux(val | (uint8(PinCom) << sam.PORT_PMUX0_PMUXO_Pos))
+		} else {
+			// even pin, so save the odd pins
+			val := p.getPMux() & sam.PORT_PMUX0_PMUXO_Msk
+			p.setPMux(val | (uint8(PinCom) << sam.PORT_PMUX0_PMUXE_Pos))
+		}
+		// enable port config
+		p.setPinCfg(sam.PORT_PINCFG0_PMUXEN)
+	case PinAnalog:
+		if uint8(p)&1 > 0 {
+			// odd pin, so save the even pins
+			val := p.getPMux() & sam.PORT_PMUX0_PMUXE_Msk
+			p.setPMux(val | (uint8(PinAnalog) << sam.PORT_PMUX0_PMUXO_Pos))
+		} else {
+			// even pin, so save the odd pins
+			val := p.getPMux() & sam.PORT_PMUX0_PMUXO_Msk
+			p.setPMux(val | (uint8(PinAnalog) << sam.PORT_PMUX0_PMUXE_Pos))
+		}
+		// enable port config
+		p.setPinCfg(sam.PORT_PINCFG0_PMUXEN | sam.PORT_PINCFG0_DRVSTR)
+	}
+}
+
+// getPMux returns the value for the correct PMUX register for this pin.
+func (p Pin) getPMux() uint8 {
+	switch p >> 1 {
+	case 0:
+		return sam.PORT.PMUX0_0.Get()
+	case 1:
+		return sam.PORT.PMUX0_1.Get()
+	case 2:
+		return sam.PORT.PMUX0_2.Get()
+	case 3:
+		return sam.PORT.PMUX0_3.Get()
+	case 4:
+		return sam.PORT.PMUX0_4.Get()
+	case 5:
+		return sam.PORT.PMUX0_5.Get()
+	case 6:
+		return sam.PORT.PMUX0_6.Get()
+	case 7:
+		return sam.PORT.PMUX0_7.Get()
+	case 8:
+		return sam.PORT.PMUX0_8.Get()
+	case 9:
+		return sam.PORT.PMUX0_9.Get()
+	case 10:
+		return sam.PORT.PMUX0_10.Get()
+	case 11:
+		return sam.PORT.PMUX0_11.Get()
+	case 12:
+		return sam.PORT.PMUX0_12.Get()
+	case 13:
+		return sam.PORT.PMUX0_13.Get()
+	case 14:
+		return sam.PORT.PMUX0_14.Get()
+	case 15:
+		return sam.PORT.PMUX0_15.Get()
+	default:
+		return 0
+	}
+}
+
+// setPMux sets the value for the correct PMUX register for this pin.
+func (p Pin) setPMux(val uint8) {
+	switch p >> 1 {
+	case 0:
+		sam.PORT.PMUX0_0.Set(val)
+	case 1:
+		sam.PORT.PMUX0_1.Set(val)
+	case 2:
+		sam.PORT.PMUX0_2.Set(val)
+	case 3:
+		sam.PORT.PMUX0_3.Set(val)
+	case 4:
+		sam.PORT.PMUX0_4.Set(val)
+	case 5:
+		sam.PORT.PMUX0_5.Set(val)
+	case 6:
+		sam.PORT.PMUX0_6.Set(val)
+	case 7:
+		sam.PORT.PMUX0_7.Set(val)
+	case 8:
+		sam.PORT.PMUX0_8.Set(val)
+	case 9:
+		sam.PORT.PMUX0_9.Set(val)
+	case 10:
+		sam.PORT.PMUX0_10.Set(val)
+	case 11:
+		sam.PORT.PMUX0_11.Set(val)
+	case 12:
+		sam.PORT.PMUX0_12.Set(val)
+	case 13:
+		sam.PORT.PMUX0_13.Set(val)
+	case 14:
+		sam.PORT.PMUX0_14.Set(val)
+	case 15:
+		sam.PORT.PMUX0_15.Set(val)
+	}
+}
+
+// getPinCfg returns the value for the correct PINCFG register for this pin.
+func (p Pin) getPinCfg() uint8 {
+	switch p {
+	case 0:
+		return sam.PORT.PINCFG0_0.Get()
+	case 1:
+		return sam.PORT.PINCFG0_1.Get()
+	case 2:
+		return sam.PORT.PINCFG0_2.Get()
+	case 3:
+		return sam.PORT.PINCFG0_3.Get()
+	case 4:
+		return sam.PORT.PINCFG0_4.Get()
+	case 5:
+		return sam.PORT.PINCFG0_5.Get()
+	case 6:
+		return sam.PORT.PINCFG0_6.Get()
+	case 7:
+		return sam.PORT.PINCFG0_7.Get()
+	case 8:
+		return sam.PORT.PINCFG0_8.Get()
+	case 9:
+		return sam.PORT.PINCFG0_9.Get()
+	case 10:
+		return sam.PORT.PINCFG0_10.Get()
+	case 11:
+		return sam.PORT.PINCFG0_11.Get()
+	case 12:
+		return sam.PORT.PINCFG0_12.Get()
+	case 13:
+		return sam.PORT.PINCFG0_13.Get()
+	case 14:
+		return sam.PORT.PINCFG0_14.Get()
+	case 15:
+		return sam.PORT.PINCFG0_15.Get()
+	case 16:
+		return sam.PORT.PINCFG0_16.Get()
+	case 17:
+		return sam.PORT.PINCFG0_17.Get()
+	case 18:
+		return sam.PORT.PINCFG0_18.Get()
+	case 19:
+		return sam.PORT.PINCFG0_19.Get()
+	case 20:
+		return sam.PORT.PINCFG0_20.Get()
+	case 21:
+		return sam.PORT.PINCFG0_21.Get()
+	case 22:
+		return sam.PORT.PINCFG0_22.Get()
+	case 23:
+		return sam.PORT.PINCFG0_23.Get()
+	case 24:
+		return sam.PORT.PINCFG0_24.Get()
+	case 25:
+		return sam.PORT.PINCFG0_25.Get()
+	case 26:
+		return sam.PORT.PINCFG0_26.Get()
+	case 27:
+		return sam.PORT.PINCFG0_27.Get()
+	case 28:
+		return sam.PORT.PINCFG0_28.Get()
+	case 29:
+		return sam.PORT.PINCFG0_29.Get()
+	case 30:
+		return sam.PORT.PINCFG0_30.Get()
+	case 31:
+		return sam.PORT.PINCFG0_31.Get()
+	default:
+		return 0
+	}
+}
+
+// setPinCfg sets the value for the correct PINCFG register for this pin.
+func (p Pin) setPinCfg(val uint8) {
+	switch p {
+	case 0:
+		sam.PORT.PINCFG0_0.Set(val)
+	case 1:
+		sam.PORT.PINCFG0_1.Set(val)
+	case 2:
+		sam.PORT.PINCFG0_2.Set(val)
+	case 3:
+		sam.PORT.PINCFG0_3.Set(val)
+	case 4:
+		sam.PORT.PINCFG0_4.Set(val)
+	case 5:
+		sam.PORT.PINCFG0_5.Set(val)
+	case 6:
+		sam.PORT.PINCFG0_6.Set(val)
+	case 7:
+		sam.PORT.PINCFG0_7.Set(val)
+	case 8:
+		sam.PORT.PINCFG0_8.Set(val)
+	case 9:
+		sam.PORT.PINCFG0_9.Set(val)
+	case 10:
+		sam.PORT.PINCFG0_10.Set(val)
+	case 11:
+		sam.PORT.PINCFG0_11.Set(val)
+	case 12:
+		sam.PORT.PINCFG0_12.Set(val)
+	case 13:
+		sam.PORT.PINCFG0_13.Set(val)
+	case 14:
+		sam.PORT.PINCFG0_14.Set(val)
+	case 15:
+		sam.PORT.PINCFG0_15.Set(val)
+	case 16:
+		sam.PORT.PINCFG0_16.Set(val)
+	case 17:
+		sam.PORT.PINCFG0_17.Set(val)
+	case 18:
+		sam.PORT.PINCFG0_18.Set(val)
+	case 19:
+		sam.PORT.PINCFG0_19.Set(val)
+	case 20:
+		sam.PORT.PINCFG0_20.Set(val)
+	case 21:
+		sam.PORT.PINCFG0_21.Set(val)
+	case 22:
+		sam.PORT.PINCFG0_22.Set(val)
+	case 23:
+		sam.PORT.PINCFG0_23.Set(val)
+	case 24:
+		sam.PORT.PINCFG0_24.Set(val)
+	case 25:
+		sam.PORT.PINCFG0_25.Set(val)
+	case 26:
+		sam.PORT.PINCFG0_26.Set(val)
+	case 27:
+		sam.PORT.PINCFG0_27.Set(val)
+	case 28:
+		sam.PORT.PINCFG0_28.Set(val)
+	case 29:
+		sam.PORT.PINCFG0_29.Set(val)
+	case 30:
+		sam.PORT.PINCFG0_30.Set(val)
+	case 31:
+		sam.PORT.PINCFG0_31.Set(val)
+	}
+}